-rBE

INTERNATJONAL SERrEs
OF
MONOGRAPBS ON I'HYSICS

GENERAL EDITQRS

J. BIRMAN S, F. EDWARDS R. H. FRJEND

C. H. LLEWELLYN SMITH M. REES D. SHERRINGTON
G. VEJ'lEZIANO
---------------·~--"
 .:s'"' ........
? ...,
I .

~ ltJ ;r
Quantum Field Theory and
Criticai Phenomena
Th1rd Edition

JEAN ZINN-JUSTIN
Commissariat à l'Energie A wmique
Direc1ion des Sciences de la mariêre
Gif sur Yvette, France

CLARENDON PRESS • OXFORD

1996

__________________ ,
 Oxford Unil'er.rily Pre.r~, WiilJon Stl~el, Oxfard OX2 6DP
Oxfard N••· l'ork
Alhens Aurkiand &mgkok Bombay
Colcut1a CaptJ Tc1171 DaresSalaam Delhi
Flortru:f! Hrmg Kmrg /.stonbul K.árachi
KualaLllmpur Madras Madríd Melbour-ne
Me:cico Ciry Naircbi Paris Singapore
Taipei Tokyo Toromo
and tlS;wdated componie-s in
Berfin fbadnn

Oxjord is o lrade mark ofOxford UniverSÍI)' Pre"

Pu!J/ished in 1he Unir.d S1o1es b1·

Oxjord Untversity Press lnr.. Net+' -York

10 J. Zinn-Justin. /996

Ali rights re.rerved. No pari ofrhis publicmion may be

teprodur~d. stored in a reuievaJ sysunr, or tr(Jnsmiued, in any
form or by any means, withoui the prior permiuion í11 writing of Oxford
UniversitJ' Press. Within the UK, e:uep1ir>rrs are allowe.tl in respecr ojany
/air dealing fot the purpose ofreseo.rch or priWJte s1udy, ot critici.Jtn or
revie~). ru permitred u,der the Cop.vright, Designs and Pau:nu A CI, 1988, or
in the ca.re ofr~ptogtaphic reprrxiucrion ín accordance with the rerms of
licence:r issued by Jlre CtJp.vrighr Liu-n.rirrg Age1rcy. E~quir~s conrerning
repro.thtcrian autside lho se U'riJ'JS .an4 in other cor.muies shcx;ld be sem lo
the Rights Deparlmem, Oxford Universít)l Pri!J"S? at lhe address .wbore.

Thúhoak is so/d subjecr to 1M <oodítfon tlrat 11 slwl/ool,

by wQy oj trade or otherwise. bt"lenr. te· solei. kintd ollt, or otherwis.e
c-irculatedwi!houtthe publísher's priorconsent i11 an.y form ofbindi~g
or c:ot.oer orher thah lhat in wlrich it is puh!UJred and ~~ithout a similar
rondition incl:uding thi.r conditüm being i.mpos~
on the .tub.Y~ent purchas2r.

A <Oialogue rewrdfor 1his book i$ availabie from the Brilish Librar)'

Librory of Congress Cato/ogi"l! in Pubiication Data

( Da1o a•ai/able)

ISB€-851882-Xj
TJ-pe.ser by the a~t.thar usín.g LA T ~x
Printed in Grear 8ritain by
Bookcraft ( Bath) Lld
Midsomer Norton. A 11011

- - - - - - - - - - . -..- -..
Preface

Tbe last thirty years have witne.ssed the Sp€Ctacula.:r progress oi quantum iield theory
(QIT). Originally intruduced to descrihe quantum electrudynamios (QED), Q:IT hM
hecome the frarnework for the díscussion of all fundamental ínteractions except gravicy.
Much more surprisíngly, ít ha.s also provided the framework for the understanding of
second order phase tra.nsitions in statístical rnechanics. In fact, as will hopefully becorne
clear in this work, QFT is the natural framework for the discussion of rnost systerns in
which an infinite number of degret$ of freedom are coupled.
Therefore, although seve~al good textbooks about Q:FT have already been publiflhed,
I thought that it might not be completely worthless to present a work in which the
oommon aspects of partícle physics and the theory of criticai phenomena are systemal-
íca.lly emphasízed. This option explains some of the choices made in tbe presentation.
A formulation in terms of path ll.lld functional integrais has been adoptcd to study the
properties of QFT. Less hnportant, tbe spac..-time metríc has been chosen euclidean, as
is natural for statistícal mechanics and convenient in general for perturbative calculations
even in particle physícs. The language of partition and correl&.tion functions has been
used even in applications of field tbeory to particle physics. Renormalízation and renor-
malization group ptoperties have been systematically discussed, whereas little space has
heen devotcd to scattering theory. Only formal aspects of QED have been considered
since excellent textbooks already cover thifl subject.
The idea of renormalímble quantum field theoríes first appeared empirkally in Quan-
tum el<l<:trodynamics. QED, as well as ali more complete field theories describing parti-
ele physics, is plagued by a serious diseose. In a straightforward calculation ali physícal
quantities are infinite, due to the short dista.nce singularities of the theory. Thls situ-
ation ha.s to be contra.sted with what happens in Classical or non-relativistic Q:uantum
Mechanics; there the replacernent of macrosropic by point-like objects leads in general
to no mathematical inconsistencies and is often a very good approximatíon: the absence
of thls property would indeed have mede progress in physics quite dífficult. This can
be summarized by sayíng that in the latter theoriffi phenomena of very different scales
approximately decouple.
A strange remedy to tbis disease has been found empirically: one artifidally modilies
tbe thoory at short distance (in a way whlch in general leeds to unphysical short distance
properties), at a scale charact-.rized by a short distance cut-off. lnspired by methods
of condensed matter physics, one then re-exp"'"""" ali physical quantities in terms of a
small number of physical constants, like the physical masses a.nd charges, instead of the
original pa.rarneters of tbe lagranglan. After this change of parametrization the cut-
off is removed, and somewbat miraculously all otber physícal qnantitíes have a finite
limit, when the theory iE so-called renormalizable. Moreover this limit is independent
of the precise forro of the short distance modification. Applied to QED this strategy
led to pr.,dictions of extraordina.ry accuracy. Therefore it wa.s natural to try also to

------------------··
vi Prcjaet

construct a renonnalizable field theoty for all other intera.ctions. Thís led to another
great achleveinent: a model for ali three strong, weak and electrornagnetíc interactíong.
The so-called Standard Model hw; no"' successfull~ confronted ail exrerírnental data, for
more than twenty years.
As the consequence of these truly remarkable results, renormalizabílity was then slowly
pramoted to a kind of edditionallaw of nature. ln particular. once the standard model of
weak, electromagnetic and strong interactions was estabiished, much effort was devoted to
cast gravity in tbe same frarnework. Despite ma.ny ingenious attempts, no renormaliznble
form of quantum gravity has been found yet.
Note that .very early it was realized, as a mathernatical curiosity, that in m!1.5sless
renorrnalizable theories a renormalization group could be a.ssociated with transformation
properties under space dilatations. Only late:r was it realiwd that this property could
be used to discuss the short distance structure of some physical processes. The basic
ídea was to introduce a set of scale-dependent coupling constants. In all)lmplotica.lly free
lield theories, these effective coupliogs becoroe sroall at lerge euclideaJJ momenta and
therefore perturbation theory, improved by renormalization group, can be used. Only tbe
theory ofstrong interactíons, a.n SU(3) gauge theory, shares this property. However, most
of tbe field tbeories proposed to describe strong, electromagnetic and weak interactions
are not asymptatically free.
More genera.lly it was suggested by Weinbetg that the existence of UV fuced points, i.e.
the existem;e of limits for the effective short dlstance couplings, was a necessary condition
for the comistency of a lield theory. Of coun;e the exlstence of other non-trivjaJ lixed
points cannot be established in the framework of perturbation theory. However many
numerical símulations of field theories on the lattice, whíclJ al]ow for nOIJ-perturbative
exp)oratíons, have failed to discover ncm-trivi.J lixed points. Therefore it seems that
the StandRrd Model, which describes so accurate!y Partícle Physics at present scale, is
not consistent on all scales and has to be modífied at short dlstance. Thís ls a second
indicatlon that maybe the property of renorrnalizabi!ity has a ditferent origin.
Somewhat surprisingly QFT has also bec<Jme llll essential tool for tbe undenltandíng
of some critica! phenomena: second order phase transitions in condensed matter physics.
Near the critical temperature cooperative phenomena generate a large scale, the corre-
lation length, although the fundamental ínteractions are short range, and the large scale
properties of the system become independem of most of the details of the rnicroscopic
structure. First attempts to explain these propertíes were based on classical ide&: a
descriptíon in terms on!y of macr05copic degrees of freedom adapted to the scale of !arge
distance physics. Such a description naturally emerges in simple approximatíons like
meBJl lield throry. The corr<!Sponding general ideas were surmnarízed in Landau's theory
of crít.ical phenomena. Unfortunately it becrune slowly clear that the extremaly univer-
sal predictiollB of such a theory were in cot.flict with numerical calculations of criticai
exponents, experimental data and finally exact resulta in two dímensiotJs. Ali these data
stíll supported the concept of universality in the sense that broa.d classes of systems h ave
índeed the same large distance properti.,., but, unlike in mean field throry, these prop-
erties depend on a small number of qua!i!.ative features lilre dimension of space, number
of components of the order parameter, syrnmetries.... Actually an analysis of leeding
correctíons to mean field theory indeed reveals that, at least in !ow space dimensíans,
degrees of freedom associated with shorter dístances never completely decouple.
To e;q>!Bln this remarkable situation, i.e. that Jarge distaJJce properties of second order
phase are to a larg" extent short distance ínsensitive, although the degrees of freedom
on ali 5Cales seem to be coupled, Wilson, partially inspired by some prior attempts

------------····--"'
 Preface vii

of Kadanoff, íntroduced the renormalization group idea: starting from a mlcroscopic

hamiltonian one integrates out the degrees of freedom corresponding to short dista.nce
l!uctuations o.nd generates a scale-dependent effectiVE" hamiltonlan. Universality reiies
then upon the existence of lR fixed points in hamiltonian space. One of the spectacular
implkations wa.s that the universal properties of a large class of criticai pheoomena could
be accurately predicted by the same lield theory methods which had been invented for
partide pb)"'ics. The appearanoe of renormali:r.able field theories wa.s there related to the
fixed point structure, and the property that the effective theory relevant for long distance
physics depended only on a srnall number of parameters.
Predíctions obtained frcm a renormalization group a.nalysis of simple lield theories
2
like the (tJP) field theory he.ve been successfully oompated to experiments as well as
numerical data from lattice models. The same field theory methods have been shown to
describe vastly different physical systerns at critícality, lilre fenomagnets, liquid-vapour,
binary mixtures, superfluid helium and, even more surprisingly, statistical propertíes of
polymers.
lf quantum field theory h as led to an understanding of the concept of universalíty and
al]owed the calculation of many univen>al physícal quantities, conversely critical phenom-
ena have shed a new light on the myst.erious role of renormalizatíon Md renormalizable
field theories in particle physícs.
For rnany~ars the renormalization procedure has been oonsidered as an ad hoc method
introduced only in order to calculate physical quantities in perturbation theory in term:s
of a small number of pa.rameters. One sometimes even tried tQ put the blame on the per-
turbative treatment of field tbeory, a.nd one tríed to hide the renormalization procedure
as much as possible (llke the BPHZ effort). Aft.er ali it worked and still worka.
However, directly inspired from the theory of criticai phenomena, another interpre-
tation, orlginally also propored by Wilson, has gained strength over the years. New
physícs should be expected at very short distances (maybe at the Planck's sca!e). At this
scale the familiar notion of renormallzable local field theories probably toses ite me!Wing.
However possible oon-local elfecte are limited to this short scale (the equivalent of the
rondition uf short range forces in statistical systems). In addition dynBIIlical effects, of
a nature whit:h at present can only be guessed, generate long dístance physics assod-
ated with the appea;a.nce of almost massless partides (rompared to the P!anck mass for
example ali known particles rue essentially massless). Nevertheless, because degrees of
freedom on ali scales remaín coupled, the short distance cut-off can never be elimínated
from the theory and this explains the impossibility of oonstrueting a finite hamiltonian
formalism. Fortunately, some renormalization group is also at work here, in sucb a way
that observa.tions at present energies can be sccurately described in terrns or ao effec-
tive long distance renormali:r.ahle field theory. Large dlstance physics is short distance
insensitive but in a way that is thus mucb more subtle than in classical physics.
Tbe classíficn.tion of interru:tions in terms o( mnonnalization group properties becomes
relevant. The neco=ity of a UV fixed point disappe!YS, because Green's funetíons no
longer need to be well defined on ali serdes, and marginal or itrelevant interaetions have
to be considered. Weak and electromagnetic interaetions are probably mruginal and, the
free field theory being an infrared fixed point, thei:r strength is proportional to an inverse
power of the logarithm of the cut-off: Green's functíons are ronsistent on ali scales only for
varili<hing coupling constants (the trimalit!l problem). Tbis wou!d e:xplaín why they can
be described by a renormalizable tbeory with small (altbough much largertban in gravlty)
roupling constants. For strong interaetions the situation seerns to be more romplicated.
One must imag1ne that the elfective interaction first decreases at shortet" distances and
viíi Prefo.ce

then increa.ses aga.in because the free field theoty is an ultraviolet fixed point. Finally
gravitation is presumably írrelevant in the sense of criticai phenomena which means that
it is non-renorrnalizable and therefore very weak because, for dimensional reasons, its
strength i.s proportíonal to a power of the short distance cut-ofL
Hcrwever, in rontrsst to criticai phenomena in which a control paramete-, like the
temperature, can be adjusted to make the conelation length large, in particle physics the
existence of mass]ess pa.rticles has to be explained from general properties of the unknown
fundamental theoty. This i.s the famous hierorchy problem. Spontaneous breaking of a
continuous symmetty, gauge principie and chiral invariance are the known mechanisms
which generate massiess pa.rticles. Supersymmetry can be helpful to deal with scalar
bosons. At present the set of general conditioll5 to be imposed on any fundamental tbeoty,
i.e. in the language of criticai phenomena the complete description of the universality class
of particle physics, has not been fonnulated. Thls ís one o[ the remaining fundamental
proh!ems of quantum field theory. An intriguing question among many is the following:
if a throry contains a light V>!Ctor boson, does the effective theory automa.tlcally take tbe
form of a Higgs model?
On the othei hand, since the large distance physia is short dist.ance insensiti....,, the
real nature of fundamental ínteiactions may remain elusiV<! in the foreseeable future, in
the same way that a precise knowledge of the criticai exponent:;; of the líquid-vapour
pha&! transitions gives v.:ry little information about the molecular interactions in water.
Tbís work, which does not claim to shed any ligbt on these difficult problems, sim-
ply tries to describe particle physia and criticai phenomena in statlstical mechanics in
a unified framework. lt can be roughly divided into four parts. Cb&pters 1-12 deal
with general field tbemy, functional integrais and functional methods. An introduc-
tion to renormalization theory is provided on the simple example of the t/>4 field throry
and renormalization group (Callan-Symanzik) equations are derived. In Chapters !3-22
renormalization properties of theories with symmetries 11re studied and specific apPlica-
tíons to partide physics are emphasized. Chapters 23-36 are mainly devoted to criticai
phenomena. A brief introduction to lattice gauge theories is included aiJd asymptotic
freedom in four dlmeusions (at large momentum, a problem releva.nt to Partide Physics)
is discussed (Chapter 33, 34). Chapters 36-43 describe the role of instantons in quan-
tum mechanics and fie!d theory, the app!ication of instanton calculus to the analysi.s of
large arder behaviour of perturbation theory IIIId the problem of summntion of the per-
turbative expansion. Fína.Uy Chapter 44 contaíns so!utions or hints of solutlons for the
e;<ercises whích are proposed at the end of severa! chapters.
I am perlectly aware that tbís work is largely incomplete. My ignorance or lack of
understanding of many important tapics is of course mootly responsible for this weakness.
However I also believe that a complete survey of quantum field tbeory a.nd its applícations
is now beyond the scopeof a single pbysicist and can only be produced by a more collective
effort.
Thís work incorporates notes of lectnres delivered in many summer schools mo.st no.
tably Cargese 1973, Bonn 1974 and Les Houches 1982, as well a.; notes prepared for grad-
uate rourses in Princeton, Louvain-la-NeuV<!, Berlin, Cambridge (Harvard) Bnd Paris.

Aeknow]edgements
lt is impoosible to list ali the physicists from whom I have benefited in my already long
career and whose inf!uence can therefore be felt in one form or another in thls work.
However I wish to specially thank my ma.;ters M. Froissa.rt and D. Bessís who guided
my Jirst steps in physics, E. Brézin and J.C. Le Guíllou who have co!laborated with me

---------------·~--"
 Prefact. i:r

for more than fifteen years and wíthout whom obviously this work would never haw
been produced. I also thínk with a deep emotion of the late B.W. Lee. The year I spent
working with him at Stony-Brook wa.s one of the most exciting of my life os a physlcist.
Wilson,s renormalízatlon group idcas are of c.outse the main source of inspiration for
this work. S. Coleman and the late K. Symanzik played a very important role in my
understanding of severa! aspects of physics through thelr articles and lecture notes as
well as in private discussions. T.D. Lee and C.N. Yang have consistently honoured me
with their friendship and hospítality in their ínstítutions. Their deep remarks have been
predous to me. Finally ali my colleagues of the Saday thoory group, wíth whom I have
had so many discussíons, have dírectly infiuenced thís work. Let me specíally mentíon
here C. Itzyk!ión whose untímely death stíll shocks me while I am writing these línes.
Severa! collea.gues agreed to read part of the manuscrípt befure publícation and I have
benefited from their criticísms, remarks and wisrlom, in particular E. Br&ín who dire.::tly
collabora.ted in the lirst chapters, R. Stora, C. Bervillier, O. Napoly, A.N. Vasil'ev,
P. Zinn-Justín and J.-B. Zuber.
Finally the ma.ny lectures I have a.ttended in L<.s Houches dur:ing nine summers have
provided me with additional inspiration.
S. Zaffanella eyped a pa.rt of the manuscript with cere and competence. M. Porneuf
tra.cked with persistence the innumerab]e misprints of the m=uscript.
Ali deserve my deepest gratitude.

3'd edition, Saclay, 25 October 1995

__________________ ,
X Prr.jace

General References
Severa! textbooks or reviews have been a direct source of inspiration or cornplement
naturally the present work (or both):
N.N. Bogoliubov and D.V. Shirkov, lntroduetíon to the Theory of Quanti.zed Fíelds
(luterscience, New York !959);
J. Bjorken and S. Drell, Relativi.stíc Quantum Mechanics, and Relalimstic Quanlum
Fields (McGraw-Hill, New York !964, 1965);
S. Weinberg, Grallitation and Cosmology (John Wiley & Soru;, New York 1972); The
Quantum Theory of Fields (Cambridge Utúversity Press, Cambridge 1995);
E. S. Abers and B.W. Lee, Phys. Rep. 9C (1973) 141;
K.G. Wllson and J.B. Kogut, Phys. &p. I2C (1974) 75;
R. Balian and J. Zinn-Justin eds. Methods in Field Theory, Proceedings ofLe; Houch""
Sununer Schooll971i (North-Hollo.nd, Anurt.erdam 1976);
Phase 1ronsitions and Criticai Phenamena, vol. 6, C. Domb and M.S. Creen eds.
(Academic Press, London 1976}, in particular tbe review Pield Tkoorehcai Appmach to
Critú:al Phenomena by E. Bn!zin, J.C. Le Guiliou and J. Ziun-Justin;
S.K. Ma, Modem Theory of Criticai Phenomena (Benjamin, Reeding, MA 1976);
C. Itzykson and J.-B. Zuber, Quantum F~ld Thwry (McGraw-Híll, New York !980);
L.D. Feddeev and A.A. Slavnov, Gauge Pields: IntrodtJction to Quantum Field Theory
(Benjamin, Reading, MA 1980);
P. Rarnond, Field Theory: A Modem Primer (Benjamin, Reeding, MA 1980);
T.D. Lee, Particle Physicl. and Jntrnduction To Field Theory (Ha.rwood Academic,
New York 1981 );
R. Stora and J.-B. Zuber eds. Recent Adoona:s in Field Theory and Statistical Me-
chanic8, Proceedíngs of Le; Houches Summer School 1982, (Elsevier, Amsterdam 1984);
S. Colemo.n, Aspects of Symrnetry (Cambridge Universíty Press, Cambridge 1985);
G. Pa.risi, Statistical Field ThW111 (Addison-Wesley, New York !988);
A.M. Polyakov, Gauge ~lds and Strings (Harwood Academíc, New York 1988};
L.S. Brown, Quantum Field Theary (Cambridge Utúversity Press, Cambridge 1992).
A discussion of the coMtruction of 6e!d theories from a more rigorous point of view can
be found in
R.F. Streater and A.S. Wightman, PCT, Spin & Statistics and Ali That (Benjamin,
New York 1964};
G. Gfunm and A. Jaffe, Quantum Physics: A FUnctional Integral Poínt of Víew,
Springer Verlag (Berlin J98J);
See also the Criticai Phenomena, Ran!Wm Systems, GatJge TMories, Proceedings of Les
Houches Summer School 1984, K. Osterwaider and R. Stora eds. (Elsevi.Er, Amsterdam
1986).

--------------···--"
Contents

1 Algebralc Preliminaries
1.1 The Gaussian Integral 1
1.2 Perturba.tion Theory 3
1.3 Comple:x Structures 4
1.4 Integral Representation Clf Constnúnts 4
1.5 Algebraic Functional Thchniques 5
1.6 Gra.s:smann Algebras. Differential Forms 6
1.7 Differentiation in Grassma.nn Algebras 7
1.8 Jntegration in Grtl.llSmanu Algebras . . 9
1.9 Gaussían Integrais with Grassmann Variables 12
Bibliographical NCltes . . . . . . . . . . 16
Exercises ................• 16
2 Euclídean Path lnteg...,ls in Ouantum Mechanics 17
2.1 Path Integrais: The General Idea . . . . . . 18
2.2 Path Integral Representa.tion of the Evolution Operaror 19
2.3 Explicit Evaluation of a Simple Path Integral: The Harmonic
Oscillator in an Externai Force . . 21
2.4 Perturbed Harmonic Oscillaror . . • . . 25
2.5 Partitlon Functioo. Correlation Functions 27
2.6 Evolution in Real Time. The S·matrix 30
2.7 Applica.tion: The Eikonal Approximatíon 34
Bibliographical Notes 36
Exercises . . . . . . . . . . . . 36
A2 Perturbation Theory. The Two..point Function 38
A2.1 Perturbation Theory in the Operator Formalism 38
A2.2 The Two-Point Function: Spectral Representation 39
3 Path Integrais in Quantum Mechanics: GeneralizatiorJS 40
3.1 General Harniltonians: Phnse Space Path Integral 4()
3.2 Ha.míltonia.ru; Quadratic in Momentum Variables . . 43
3.3 The Spectrum of the 0(2} Synunetric Rigid Rntator 47
3.4 The Spectrum of the O( N) Symmetric Rigíd Rotator 48
3.5 Path Integral and Holomorphi<: Representation 51
3.6 Fermioníc Path Integral 56
Bibliographica.l Notes . . . . . . . . . . . . . . 59
Exercises ................. . 60
A3 Ouanti>.ation of Spin Degrees of Freedom, Topological Actions 62
A3.1 General Symplectic Forms a.nd Quantization . . . . . . . . 62

------------------·· ll'td
 Contem.<

A3.2 Spín Dynamícs and Quantization . . 63

A3.3 The Magnetic Monopole 66
4 Stochastic Differential Equations: langevin, Fokker-Piancl< Equations 67
4.1 The Langevin Equation 67
4.2 The Fokker-Planck Equation 68
4.3 Equilibrium Distribution. Correlation Functíons 70
4.4 A Special Case: The Purdy Díssipative Langevin Equation 73
4.5 A Simple Example: The Linear Langevín Equation 74
4.6 Patb lnr.egral Representation 75
4. 7 Stochastic Processes on Riemannian Manifolds 7í
4.8 Discreti•ed Langevin Equation 80
Bibliographical Notes . . . . . . 84
Exercises 84
A4 Markov's Stochastic Processes: A Few Remarks 85
A4.! Tbe Spectrum of the Transition Matrix 85
A4.2 Detailed Bal!l.llce . . . . . . . 87
A4.3 Continuum Time Limit 88
A4.4 ConstTuction of a Stochastic Process with a Giwn Equilibrium Distribution 89
5 Functional Integrais in Field Theory . . . . . . 90
5.1 Fllnctional Integrais. Correlation Functions 90
5.2 Perturbative Expansion of Fllnctional Integrais 93
5.3 Wick's Theorern and Feynman Dia.grlllllS 95
5.4 Algebraic Identities and Functional Integrais 98
5.5 Quantum Fleld Theories with Fermions 104
Biblíographical Notes . . . . . . 107
Exercises . . . . . . . . . . 107
A5 Eudidean Dirac Fermions and 'Y Matrices 108
AS.! Diroc 1 Matrices . . 108
A5.2 An Explícit Constructíon 109
A5.3 Transformation of Spínors 110
A5.4 The Matrix 1s 111
A S..!i Reflections 112
A5.6 Other Symmetries 112
A5.7 Tbe Examp!e of Dimension Four 114
A5.8 Trace of Products of "/ Ma.trices 115
A5.9 The Fien Transformation 117
6 Generating Functionals of Correlation Funo::tions. Loopwõse Expansion 118
6.1 Generating Fllnctionals of Connected Correlation Functions:
Cluster Properties 118
6.2 Proper Vertices . . . . . . . . 122
6.3 Momentum Representatíon 126
6.4 Semiclassical or Loopwise Expansion 127
6.5 Legendre Transformation and 1-lrreducíbility 133
6.6 Pbysicallnterpretation of the IPI Functional !35
Bibliographical Notes . . . . . . . . . 137
Exercises . . . . . , . . . 137
Aó Hig'- Orders in the loop Expansion 138
7 Real-time Quantum Field Theory and S-Matrix !40
7,1 ScaJ..., Free Field Theory: Holomorphic Representation 140

-----------·~--"
 Contents xiii

7.2 The S-Malrix 144

7.3 S-Matrix and Correlation Functians 145
7.4 Field Renarmu.lization 150
7.5 h1assive Dirac Fermions . . ~ . . 153
7.6 Nan-Relativistic Limit ofQuantum Field Theory 156
Bib!iographical Notes . . . . . . . . . . . 160
A7 • . . . . . • . . . ......•.•. 161
A 7.1 Cluster Properties of Correlation Functions IGl
A 7.2 Connection between Spin and Statístícs . . Hl2
8 Divergences in Penurbation Theory, Power Counting 164
8.1 The Problem of Divergences and Renormalization: The .p3 Field Theory
at One-Loop Order . . . . . . . . . . . . . . . 164
8.2 Divergences in Perturbation Tbeory: General AnalysiE 168
8.3 Classlfication of Renormalizable Field Tbeories 172
8.4 Power Counting for Operator Insertions 174
Bibliographical Notes . . . . . . . . . . . 176
9 Regularization Methods . . . . . . . . . 177
9.1 Cut-Ofi and Pauli- Villars's Reguhu-ization 177
9.2 Lattice Regularízation 180
9.3 Dimensional Regula.rization 183
Bibliographical Notes . . . . . . 188
Exercises . . . . . . . . . . . 189
A9 Schwinger's Proper Time Regularization 190
lO lntroduction to Ren()fmalization Theory. Renormalization Group Equations 193
10.1 PDW<lr Countíng. Renormalized Action 193
10.2 One-Loop Divergen<:es . . . . • . • . . . . . . . . . . . . . . 196
10.3 Dívergences Beyond One-Loop: Skeleton Diagrams . . . . . . . . . 199
I 0.4 Bare and Renormalized Correlatlon Functions, Operator r/>' llJBertions 202
10.5 Callan-Symanzík Equ.,tions 204
10.6 lnductive Proof of Renormalizability 206
10.7 The (~ 2 1/!2) Correlation Function 211
I 0.8 Tbe RenormaliZI'd Action 212
10.9 The Massless Theory . . . . . . . 213
10.10 Homogeneous Renorrnali:t,atioo Group (RG) Equations in a Massive Theory 217
I 0.11 Covariance of RG Functions 218
Bib!iographical Notes 219
Exercíses 220
AlO 221
AlO. I La.rge Momenturn Mode lntegration and General RG Equations 221
A10.2 Super-renormalizable Field Theories: The Normai-Ordered Product 223
AJ0.3 B&ckgtound Fie!d Method and the S-Matrix . . . 226
11 Dimensional Regularization, Minimal Subtraction: Calculation of
RG Functions
1LI Bare and Renormalized Actions: Renormalization Group (RG) Functíons
l 1.2 Dimensional Regularization: Tbe Forrn of Renormalization
Constants . . . . . . . 231
11.3 Minimal Subtra.ctlon Scheme 232
11 A The Massless Theory 234

________________
11.5 A Few Technko.l Remarks
,
235
xiv Contents

1L6 Twu-Loop Calculation of Renormalízation Constants and RG Functíon~

in the ,p4 Fie!d Theory . . . . . - . . . - 236
11.7 Generalization to Severa.! Component Fields 242
11.8 R.:uorma.l.izatkm and RG Functions at One-Loop in a Theory
with Bosons and Fermions 245
Bíbliographka.l Notes - - - . . . - - - - . - . 249
Exerdses . . . . . . . - - . . - - - - . . - . 249
12 Renormalilation of Composite Operaton;. Short Distance Expansion 251
12.1 R.:norma.lh.ation of Operator lnsertions . . . . . . . 251
12.2 Equations of Motion . . . . . . . . . . . 255
12.3 Short Distance Expansion (SDE) of Operator Products 259
!2.4 Large Momentum Expansion of the SDE Coefficients 263
12.5 Callan-Symanzik Equation for the Fírst Coefficient of the SDE 264
12.6 SOE beyond Leading Order . . . . . . . . 265
12.7 SDE of Product.s of Arbitrary Loca.! Operato:rs . . 266
12.8 Light Cone Expansion (LCE) of Operator Products 267
Bibliographical Notes . . . 268
Exerci.ses . . . . . . . . . . . . 268
13 Symmetries and RenormaUzation 269
13.1 Prelimínary R.:marks 269
13.2 Lillea:rly R.:a.lízed Global Symmetries 271
13.3 Linea:r Symmetry Breaking 274
13.4 Spontaneous Symmetry Breakíng . . 278
13.5 Quadratic Symmetry Breaking , . . 280
13.6 Application to Strong Interaction Physics: Chira.l Symmetry 285
Bibliograpbical Notes . . . 293
Exerdses . . . . . . . . . . • • . 294
Al3 Currents and Noether's Theorem 295
A13.1 Real Time C!assícal Field Theory 295
Al3.2 Euclidean Field Theory . . . . 296
A13.3 The Energy Momentum Tensor 297
Al3.4 Energy Momentum Tensor and Euclidean Field Theory 299
Al3.5 DUatation and Confonnal lnvariance 300
14 The Non-linear a-Model: An Example of Non-linearly Realized Symmetries 302
14.1 The Non-Linea:r tr-Model: Delinition 302
14.2 Perturbation Theory. Power Counting 304
14.3 ReguÍarizatíon . . . . . 306
14.4 lnfrared (IR) Divergences 308
14.5 WT ldentities 309
14.6 R.:norma.lizatíon 312
14.7 Solution of the WT ldentitíes 315
14.8 Renorma.lization of Compooite Operaton; 317
14.9 The Linearized Representation :ns
Bibliographical Notes . • . . . . . . , . . 319
Exerci.ses . . . . . . . . . . . . .. . . 320
15 Models on Homogeneous Spaces in Two Dimensions 321
15.1 Construction of the Model . . . . . . . . 321
15.2 WT ldentities and Renorma.lization in Linear Coordinates 325
15.3 Renorma.lízation in Arbitmry Coordinates, BRS Symmetry 328-

- - - - - - - - - - - - - - - - - -..
 Content.

15.4 Symmetric Spaces: Delinition 332

15.5 The Classical Action. Comervation Laws 333
15.6 Quantum Theory: Perturbation Theory and RG Functíons 336
15.7 Generalizations 341
Bibliogra.phical Notes . . . . 342
Exercises 343
Al5 Homogeneous Spaces: A Few Algebraic Properties 344
A15.1 Pure Gauge. Maurer...Cartan Equations . . . 344
A15.2 Metric and Curvature in Homogenoous Spaces 344
A15.3 Explícit Express!ons for the Metric 346
A15.4 Symmetric Spa.ces: CIIISsHication . M7
16 Slavnov-Taylor and BRS Symmetry. Stochastic Field Equations 352
16.1 Slavnov-Taylor (ST) Symmetry 352
16.2 Effective Action and BRS Symmetry 354
16.3 Grassmann Coordinates, Gradient Equations 356
16.4 BRS Symmetry and Compatibility Condition, Group Manifolds 358
16.5 Stochastic Equations . . . . . . . . . . . . . . 359
16.6 Consequooces of BRS Symmetry: General Discussion 362
16.7 Extension of Grassmannian Symmetries 364
16..8 Application: StoclJBstic Field Equations . . . . . . 365
Bibliographical Notes . . . . . . . . . . . . . . . . 369
l7 ReRQrrnalization and Stochastic Field Equations. Supersymmetry 370
17.1 Langevin and Fokker-Planck Equations . 370
I 7.2 Time-Dependent Correlation F\mctions 372
17.3 ST Synrrnetry and Equitibríum Correlatíon ~unctions 374
17.4 Renormalizatíon and BRS Symmetry . . . . . . 376
17.5 Tbe Purely Dissipative Langevin Equation 377
17.6 Supen;ymmetry and Equilibrium Corre!ation Functíons 381
17.7 Stochastic Quantizatíon of Two-Dimensional Chiral Models 382
17.8 Tbe Langevin Equation on Riemannia.o Manífolds 385
17.9 Stochastic Quanti~ation: the Exa.mple of Gauge Theories 388
17.10 Supersymmetry in Four DimeDSionfl 389
Bibliographical Nates . . . . . . . . . . . . . . . . . 396
A17 Renormalization Constants and RG Functions at Two..Loops:
Supersymmetric Perturbation Theory 398
2
Al7.1 The ( </>2 ) Field Theory 398
A17.2 Perturbative Calculation of q.,(g} 399
Al7.3 The Non-Linear a-Model 401
18 Abelian Gauge Theories 404
18.1 The Massive Vector Field 404
18.2 Action with Fermion Matter 408
!8.3 Massl€5s Vector Field: Abellan Gauge Symmetry 409
!8.4 Canonical Quantization and Gauge Invaríance 4!2
18.5 Perturbation Thoory, Regularization 417
18.6 WT ldentities, R.enonnalizntion 421
18.7 Gauge Dependence . . . 423
!8.8 The RG Equ"tions 426
18.9 The One-Loop ll-Function . . 427
!8.10 ·The Abelían Hlggs Mechanism 430

- - - - - - - - - - - - - - - - - -..
 Conter.ts

18.11 Quantization of the Higgs Mechanism , . . 431

18.12 Pbysical Observables. Unitarity of the S-Matríx 434
Bibliographical Notes 435
Exerciscs . . . . . . . . . . 436
AlB 437
A18.l Vacuum Energy and Casimit Elfect 437
A18.2 Gauge Dependenoe 440
19 Non-Abelian Gauge Theories: lntroduction 441
]9.1 Goometrical Conrttuction 441
19.2 The Gauge lnvariant Action . . . . . 444
19.3 Hamiltonian Formalism. Quantization 444
19.4 The Faddeev-Popov Quantization, BRS Symmetry 448
19.5 Perturbation Tbeory, Regu!arization 450
19.6 The Non-Abelían Higgs Mechanism 452
Bibliographical Notes . . . . . . . 456
Al9 Massive Yang-Mills Fields 457
20 The Standard Model- Anomalies 459
20.1 Tbe Standard Model of Weak-Electrowagnetic lnteractions 459
20.2 Quantum Chromodynamics (QCD) 466
20.3 The Abelian Anomaly 471
20.4 Non-Abelian Anomaly 478
20.5 Physical Consequences 480
20.6 Wess-Zurnino Consistency Condítioru; 481
Bibliographlcal Notes 482
21 Renormalization of Gauge Theories: General Fonnalism 483
21.1 Notations and Group Structure 483
21.2 Quantization 485
21.3 BRS Symmetry . . . . . . 486
21.4 WT ldentities 488
21.5 Renormalization: General Considerations 489
21.6 Tbe Renormalized Action 492
21.7 Gauge lndependence 497
Bíbliograpbical Notes . . . . . 49S
Exercises . . • . . . . . . . 499
22 Classical and Quantum Gravity. Tensors and Riemannian Manifolds 500
22.1 Change of Coordinates. Tenson; . . , , . . , , 500
22.2 Patallel Transport: Connection, Covariant Derivative 503
22.3 The Metric Tensor &06
22.4 The Curvature (Riemann) Tensor 507
22.5 Covariant Volume Element 510
22.6 Fermions, the Vielbein 511
22.7 Classical Gravity. Equations of Motion 513
22.8 Quantization in tbe Temporal Gauge: Pure Gravity 516
Bibtiograpbical Notes . . . . . . . . . . . . . . • , 519
Exercises . . . . . . . . 520
A22 Matrix Models in lhe Large N Limit and 2D Gravity 522
A22.1 Quantum 2D Euclidean Gravity 522
A22.2 Tbe One-Matrix Model . , . . . . . 523
A22.3 The Metbod of Orthogonal Polynomials 526

--------------.-~-- ..
 Crmtent.; xvii

23 Criticai Phenomena: General Considerations 529

23.1 Phase Ttansitions aod Transfer Matrix 531
23.2 The lnlinite Ttansverse Size Limit: )slng-Like Systems 533
23.3 Order Parameter and Cluster Properties 536
23.4 Stochastic Processes and Phare Ttansitíons 538
23.5 Continuous Symmetries 539
Biblíographical Notes . . . . . . . . . . . 540
A23 Ouem:hed Averages 541
24 Mean Field Theory f.,. Ferromagneti<: Systems 544
24.1 lsing-like Ferromagnetic Systerns 544
24.2 Mea.n Field Expansion 546
24.3 Mean Field Appraximation 547
24.4 Dniversaiity within the Mean Field Approximation 549
24.5 Beyond the Mellll Field Approximation 554
24.6 Power Counting and the Role of Dimension Four 555
24.7 Tticritical Points 55 i
Bibliographical Notes . . . . . . . . . . . 558
Exerdses . . . . . . . . . . . . 559
A24 Mean Field Theory: General Fotmalism 560
A24.1 Mean Field Theory . . . . . . . 560
A24.2 Mean Field Expansion . . . . . . • 563
A24.3 High Tempemture Series and Mean Field 564
A24.4 Low Thrnperature and Mean Field 564
25 General Renormalization Group Analysls. The Criticai Theory
near Dimension Four . . . . . . . . . . 566
25.1 The Abstract R.€normaiization Croup: General Formulation 567
25.2 The Gaussian Fixed Point . . • . . . . . . . 570
25.3. The Criticai Theory near Dimension Fonr: The Effective Field Thoory 572
25.4 R.€normalization Group Equations for the Criticai Thoory:
The E:-Expansion • . . . . . . . . . . . . . 573
25.5 Solution of the Rc Equations . . . . . . . . . 576
25.6 Criticai Cottelation Functions with ,P2 (x) Jnsertions 579
Bibliographical Notes . . . . . . . . . . . 582
Exercises . . . . . . . . . . . . . . . . . . 583
26 S<:aling Behaviour in the Criticai Oomain 584
26.1 Strong Sca!ing above T~: The Renormaiized Theory 585
26.2 Expansion around the Criticai Theory 589
26.3 Scaiing Laws above Te 589
26.4 Correlation Functions with rjJ 2 lnsertions 591
:16.5 Scaling Laws in a Magnetic Field and Below Te 593
26.6 The N-Vector Model . . . . . . . . . . , 596
26.7 Asymptotic ExpMsion of the 2-Point F\mction in the Criticai Domain 602
Bibliographical Notes . . . . . . . . . . . 604
ExerciEes . . . . . . . . . . . . . . . . . . 604
A26 The S<:aling Behaviour of the Specllic Heat for o =O 606
27 Correction$ to Staling Behaviour . . . . , . . . . 608
27.1 Corrections to Scaling: Generic Dimensions 608
27.2 Logarithmic Corrections at the Upper-Critical Dimensíon 610
27.3 Irrelevant Operators and the Question of Universality 613

- - - - - - - - - - - - - - - - - -..
xviii Content.!

27.4 Corrections Coming from lrrelevant Operators. lmptoved Adion 615

27.5 Applicatioo: Unírudal Systems with Strong Dipolar Forces 617
Bib!íographical Notes 622
Exercíses , . . . . . . . . 623
28 Calculation of Uniwersal Quantities 624
28.1 The [-Expansion . . . . . . 624
28.2 The Perturbative Expansion at Fixed Dimension 632
28.3 The Series Summation . . . . . . . . 634
28.4 Numerical Estirnates of Criticai Exponents 636
28.5 Comparison wíth Lattice Model Estimat.es 638
28.6 CriticE>I Exponents frorn Experirnents 638
28.7 Amplitude R.atíos 640
BibliographicE>I Notes . . . . . . . . . . 641
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . 64 3
2
A2B Non- Magnetic Statistical Models and the ( 4>2 ) Field Theory 644
A28.! Statistícs of SeU-Repellíng Chains, Approximations 644
A28.2 Liquid~ Vapour Phase Transitlon IUld Field Theory 649
2
29 The (4> 2 ) Field Theory in the Large N limit 652
29.1 lntroduction . 652
29.2 The La:rge N Limit 653
2!f.3 The 1/N Expansion 659
2!f.4 Generalizations 664
Bibliographí<:al Notes , . 665
Exercises . . . . . . . 665
30 Ferromagnetic Order at Low Temperature: the Non-Linear a-Model 667
2
:>0.1 The [4>2 (x)j Field Theory at Low Temperature , . . . . . 668
30.2 IR Divergen~, Spontaneous Symmetry Breaking and the Role
of Dimensíon Two . . . . . . . . . . 669
30.3 The Non-Linear u-Model: RG Equatlons 670
30.4 Discussion of the RG Equations 672
30.5 Resulta beyond One-Loop 676
30.6 The Dimension Two 678
30.7 The Large N Limit 679
30.8 Generalizations 660
Bibliographical Notes . . . 681
AJO Th., Lin.,arized Gross~Neveu Model 682
A30.1 RG Equations near Four Dimensions 1383
A30.2 Large N Limit . . . . . . . . . . 684
A30.3 The Groas-Neveu Model near Two Dimensions 1387
31 Two-Dlmensional Models and Bosonization Method 689
31.1 The Free Massless Scalar Field . 689
31.2 The Free Massless Dirac Fermion 693
3U\ The Sine-Gordon Mode! . . 699
31.4 The Schwinger Model . . . 701
31.5 The Massive Thirring Model 704
31.6 A Two-Fermion Model 708
Bibliographical Notes , 711
Exercises 711
A31 A Few Additional Results 713

----------------~-- ..
 Contents

A31.1 The Schwinger Mode! • 713

A3!.2 The SU(N) Thírríng Model 714
A31.3 Solitons in the Sine-Gotdon model 717
32 The 0(2) 20 Classio:al Spin Model 718
32.1 The Spin Correlation Functions 719
32.2 Correlatíon Functions in a Field 721
32.3 The Coulomb Gas in 2 Dimensions 722
32.4 0(2') Non-Linear u-Model and Coulomb Gas 726
32.5 The Criticai 2-poínt Function in the 0(2) Model 727
32.6 The Generalízed Thirríng Model 729
Bib!iographícal Notes . . . . . . . . 729
A32 Non-relativistic Fermions in one dimension 730
33 Critkal Properties oF Gauge Theories 731
33.1 Gauge lnvariance on the Lattice 731
33.2 The Pure Gauge Theory 733
33.3 Wilson's Loop and Conlinement 735
33A Mean Fíeld Theory 741
Bibliographícal Notes . . 744
Exercises 745
A33 Gauge Theory and Confinement in Two Dimensions 746
34 large Momentum Behaviour in Field Theory , . . . . . 748
2 2
34.1 The (4> ) Field Theory: Large Momentum Behaviour and Triviality 748
4
34.2 The General</> Field Theory 754
34.3 Theories with Scalar Bosons and Fermions . . . 755
34.4 Gauge Theories . . . . . . . . . . . . . . 757
34.5 Applicatíons: The Theory of Strong lntera.ct;ions 760
Bibliographícal Nota . . . . . . . . 762
35 Critio:al Dynami.::s . . . . . , . . . , , . . 764
35.1 RG Equatíon.s near 4 Dimensíon.s: The Purely Dissipative Case 765
35.2 RG Equations near 2 Dimensíons: The Purely Dissipa.tive Case 769
35.3 Conserved Order Parameter (Model B) . . . . . . • . 770
35.4 Relaxa.tional Model with Energy Conservation (Model C) 771
35.5 A Non-Relaxational Model (Model E) , , . . . . . 774
Bibliogra.phical Nota . . . . . . , . . . . . . . . . 777
36 Field Theory in ;o Finite Geometry: Finite Size Scaling 778
36.1 Renormalization Group in Fíníte Geometries . 778
36.2 Perturbatiou Theory in Fínite Geometríes . . . 782
36.3 The </>4 Field Theory: The Periodíc Hypercube . 783
36.4 The </>4 Field Theory: The Cylindrical Geometry 789
36.5 Fínite Size Effects in the Non-Linear u-Model 794
36.6 Fínite Size Effects and Dynamics 801
Bibliographical Nota . . . . . . . . . . 806
A36 ..... . 807
A36.1 Discrete Syrnmetries and Finite Síze Effects 807
A36.2 Perturbation Theory in a Finite Volume . . 811
37 lnstantons in Quantum Mechaníc:s: The Anharmonic Osdllator 812
37.1 The Anharmonic Oscillator for Negative Coupling: Preliminary
Consideration.s . 812
37.2 A Simple Integral
______________ ,
814
 ContenL<

37.3 Quantum Mec.hanics: The Sadd!e Poínt . 816

3 7 A ln.stanton Contribution a.t Leading Order 817
Biblíogra.phical Notes 822
E:xercises . . . . . . . . . . 822
A37 The Jacobian, Constrainu 823
A37.1 The jacobian 823
A37.2 Con.straint Equations . . 824
38 lnstantons in Quantum Mechania: Generalization 825
38.1 The Instanton Contribution . . . . . . . 825
38.2 Cakulation of the Determinam: The Shifting Method 827
38.3 The Large {3 Limit 832
Bibliographkal Notes . . . . . . . . . . . . 833
Exercises . . . . . . . . . . . . . . . . . 833
AJB The WKB Method: Miscellaneous Remarks- 834
A38.1 The Classical Equations of Motion 834
A38.2 The WKB Method . . . . . . . . 835
A3S.3 The Average Actlon in Path Integrais 838
39 Unstable Vacua in Field Theor}l 840
39.1 The q,• Field Theory for Negative Couplíng 840
39.2 General Potentials: lnstanton Contributioru; 846
39.3 The Decay of the False Vacuum: Cosmological lnterpretation 848
Bibliogra.phlcal Notes . . 849
Exercise:s . . . . . . . . . . ... 850
A39 Sobolev lnequ.alíties 851
40 Degenerate Ciassícal Mínima and lnstantons 854
40.1 The Doubie-Well Potential 854
40.2 The Periodic Cosine Potential 856
40.3 lnstantons in Sta.ble Boson Fíeld Tbeories: General Rema.rks 859
40.4 lnstantons in CP(N -1) Models . . . . 861
40.5 lnstanton5 in the SU (2) Gauge Theory 863
40.6 Jnstantons and Stochastic Differentíal Equations 865
Bíbliographical N<>tes . . . • . . . • . 868
A40 ... - .. · · • · · 870
A40.1 'I'race Formula for Periodíc Potentials . 870
41 Perturbation Theory at Large Drden and lnstantons.
The Summation Problem 871
41.1 Qua.ntum Mechanics 871
41.2 Scnlar Field Theory 874
41.3 Field Thoories wíth Fermions 875
41.4 Divergent Series, Borel Summability 880
41.5 Large Order Behaviour and Borel Summability 882
41.6 Practical Summation Methods 883
Bibliogra.phical Notes . • . . . . . . . . 886
A41 ..•............ 888
A4l.l Large Order Behaviour for Simple lntegrals 888
A41.2 Non-Loopwise Expansions . 888
A41.3 Linear Differential Approxima.nts . . . . . 889
42 lnstantons: The q, 4 Fíeld Theory in Dirnensíon Four 890

____________________
42.1 The EuclideWl Equation of MQtion. The lnstanton AcUon 890
,.
 Contents .:z:xt

42.2 The lnstanton Contribution a.t Leading Order 891

42.3 The Coupling ConstBJlt Renormaliza.tion 895
42.4 The lmagínary Pan of the n,Point Flmction . 897
42.5 Instanton Contribution to Large Otder Behaviour 898
42.6 UV and IR Contributions to Large Order Behaviour 899
42.7 The Ma&'5ive Theory 902
Bibliographical Notes 902
Exercises . . . . . . . . 903
A42 . . . . . . . . . . 904
A42.1 InsLantons and RG Equations 904
A42.2 Conforma! Imrariance . . . . 90E>
43 Multi-lnstantons in Quantum Mechanics 907
43.1 The Doubl~Well Potential 9{)8
43.2 The Periodic Cosine Potentia.l 916
43.3 General Potentials with Degenerate Mínima. 919
43.4 The O( v) Symmetric Anharmonic Oscillator 922
43.5 A General Conjecture 924
Bibliographical Nol.ffi 925
Exercíses . . . . . . . . 925
A43 . . . . . . . . • • 926
A43.1 Multi-instantons: The Determína.nt 926
A43.2 The lnstanton Interaction . . . . 927
A43.3 Multi-instBnt.ons from Constraints 929
A43.4 A Simple Exarnple of Non-Borel Summa.bility 929
A43.5 Multi-Inrtantons a.nd WKB Approximation 931
44 Exercises: Some Solutions . . . . . . . . . 934
44 Cha.pter 1: Algebraic Prelimina.ties 934
44 Cha.pter 2: Eudidean Patb Integrais in Quantum MecbB.Ilics 9$
44 Chapter 3: Pa.th integrais in Quantum Mechanics: Gener!>liza.tion 940
44 Chapter 4: Stochastic Düferential Equations • . . . • . 944
44 Chaptet 5: Func:tional Integrais in Field Theory 946
44 Cbaptec 6: Generating Punctions of Correla.tion Function.s 948
44 Chapter 9: Regularization Methods . . . . . . . . . . 950
44 Cha.pter lO: lntroduc:tion to Ren!Jrrnalization Theory 952
44 Chaptet 11: Dimensional Regularization a.nd Mínima! Subtraction 952
44 Cbapter 12: Renormalízation of Composite Operators . . . . 955
44 Chapter 13: Synunetries and Renormalization . . . . . . . . . 959
44 Chapter 14: The Non-Linear a-Model: An Example of Non-Linearly
Realized Symmetries . . . . . . . . . . . . . . . . . . . 962
44 Cba.pter 15: Models on Homogeneous Spa.ces in Two Dimensions 967
44 Cbapter 18: Abelian Gauge Theories . . . . . . . . 968
44 Chapter 21: Renormalization of Gauge Tbeories. General Formalism 973
44 Chapter 22: Classical and Quantum Gravity . . . 976
44 Chapter 24: Mea.n Field Theory . . . . . . . . . 979
44 Chapter 25: General Renonnalization Group Analy•is 982
44 Chapter 26: Scaling Laws in tbe Critica! Domain . . 983
44 Chapter 28: Calculation of Universal Quantities . . 984
2
44 Chapter 29: Tbe ( .p2) Field Theory in tbe Large N Limit 985

__________________
44 Chapter 31: Two-dimensional Mode!s and Bosonization Method
,
986
 Contents

44 Chapter 33: Critica! Properties of Gauge Theories 990

44 Chapter 37: lnstantons in Quantum Mechania; . 990
44 Chapter 38: lnstantolll5 in Quantum Mechanics: Generalízatíon 992
44 Chapter 39: Unstable Va.cua in Fíeld Theory . . . . 996
44 Chapter 42: (lll5tantons: The .P 4 Field Theory in Dimension Four 997
44 Chapter 43: Mu!ti·instantons in Quantum M€Chanics 998
lnde:x 1001

--------------···--"
1 ALGEBRAIC PRELIMINARIES

It is somewhat unusual to begin a physics textbook with algebraic identities, which are ;n
general hidden in appendices. However, our discussion of perturbative aspects of qua.n-
tum mechanics and quantum field theory will entirely be based on path or functional
integrais and more general!y functional techniques. Tberefore a rea.der not alrea.dy famil-
iar with these concepts may fmd it difficult to follow the algebraic manipulations whkh
enter in the derivation of many results. Moreover we want to indicate by such a choice
that the vaxious technical difficulties which we shall meet, wíll in general be directly
confronted tllther than ca.refully hídden.
Therefore in this first chapter we recali a few algebraic identiti"" about gaussian inte-
grais. We also recall the concept of functional dillerentiation and the algebraic definition
of the determínant of an operator.
We then define Blld discuss e. íew properties of dilferentie.tion e.nd int.egratic>n in a
Grassmann, i.e. antisymmetric algebra. In pe.rticula:r we calcule.te gaussian "fermionic"
integrais. Throughout the cha.pter all expressions axe given.for a finite but arbitre.ry
number of variabies, becauoe the focus is mainly on algebraic properties. However the
generalization to a.n inlinite number of variables will be easy, as will be discussed in the
following chapters.
Note that in this chapter, as well as in this whole work, s-ummation over repeatm
índices will alway• be meo.nt (except íf explicitly stated otherwise).

1.1 The Gaussian Integral

In this section we bríefly revíew a few algebraíc properties of gaussian lntegrale in the
case of a fmite number of integration variables.
A general gaussian integral has the forro:

in which A is a symmetríc matrix with eigenve.lues .>., satisfying

To calculate I one first looks for the minimum of the qua.dratic form:

=0.

Tbe solutíon is:

Xt = (A- 1 )i}b, ~ (1.2)
(summatíon over j being meant as stated a.bove) and one sets:

:tj (A- 1 );;b; +Y•· (1.3)

----------------··
2 A/gebroic Preliminaries 1.1

The integral becomes:

1 ~ exp [ !b, (A- 1 ).; b;] j (fl,dy;)exp ( -h;A;;Yi). (1.4)

The last lntegml can be calculated by changing variables, setting
(A'n),, !I;= !I;,
12
tbe eígenvalues A! o[ A'' being chosen sucb that -1t/4 :S Arg>.!/ 2 :S +or/4. Tbe
2

integral over the y;'s ís then straightforward and one obtalns:

2
1 (A, h)= (2,..)n/ (detA)-I/ exp
2 [.tl -ibi(~-l),,b;]. (1.5)

By differentiating this last expression with respect to the variables b;, it is then possible
to calculate the a.verage of a.ny polynomíal with a gaussían weight:

(x.,x., ... x,,) N j (TI dx;) "'••"'•• ... "'••

1.
exp (-.E tx,A;,x;),
1J=l
(1.6)

in wbicb the normalization N is cbasen in such a way tbat (1) = 1:

N- 1 =l(A,O)
Indeed from expression (1.1) one derives:

a~. I (A, b) =f m,dx;}xtexP (-!x;A;1 Xj + b;x;). (1.7)

Repeated differentiation with r""]J<'Ct to b then leads to the identity:

(x.,x., -· .x •• l = (21t)-..n (detAJl/2 [a:•. a:.. ... a:./ (A, b>]L=O,

or repladng the integrs.l I (A, b) by its explidt form (1.5):

(xk,·· .x.,) {a: --·a:A:, Ir,

exp [t ?b, (A- ),;b'] }I
i,j,:;d
1
b=.O
(1.8)

Wiclt's theorem. This identity leads to Wick's theorem. In the r.b.s. of equation (1.8)
each time a differentis.J operator acts on the exponentis.l it generates a factor b. Anather
differential operatar h as to act on this factor, otherwise the corresponding contrihution
vanishes when we set b = O. We condu de that the average of the praduct :r:k, ... "'l• witb
the gaussia.n Wf!íght exp (- ~x,A;;Zj) is obtalned iu the following way: one considen. ali
possible pairings of tbe indices k" ... , k, (t must thus be even). To each pair kpkq one
associates the matrix element (A- 1)k.t, of the matrix A -1. Then:

(Xkt ••. x*, ) = "

~ ' A-'
A:Pl k,::~: ~.. A-'
k,.l-1 k,.l . (1.9}
.Jl pot:I!;Íble t;t&iring5
of (lrt···k.t)
Equatlon {1.9) which expr"'lllell Wici<'s theorem is, in the form adapted to Quantum
Mechanics or Field Theory, the basis of perturbative calculations.
Remark. The gaussia.n int-egral has another remarkable property: if we integrate the
exponential of a. quadratlc forro over a snbset ofvariables, the result is still the exponential
of a quadratic form. This structw:al $tabíllty is related to some of the properties of the
harmonic oscillator which will be discussed in Chapter 2.

---------------·~--"'
1.2 .41gebmic Preliminaries 3

1.2 Perturbation Theory

We now want to calculate the integral:

I= j fi dx, exp (- t
1=1 tJ=l
i:z:,A,,xj- ,\V (x)) , (1.10)

in which V(x) is a polynomial in the variab!es X; and ), is a parameter. We c&n expand

the integra.nd in a formal power series in .\:

(1.11)

Using equatíons {1.6,1.8) we can formally rewrite (1.10):

I {•xp [ -ÀV (~)] exp [ t ~b;(A- 1 ),, b3]} I

t,J-1 b=O
. (1.12)

We can also directly calculate each term in the expansíon usi.ng Wick's thoorem (1.9).
Steepest-descent In the case of contour integral<; in the complex dom,.;n, one sometimes
uses a method, steepest-descent, which reduces theil' evaluation to gaussían integrals. Let
us eonsider the integral:

1 f fJ dx; exp [ -~S(x1, ... ,xn)]. (1.13)

In the limit ,\~O, the integral is dominated by saddle points {xí}:

ôS
i:lx,
(xí,x~, ... ,x~) =O. ( 1.14)

To calculate tbe contributíon of the leading saddle point x<, we change variables, setting:
x =x• + y,/À. (1.15)
We then expand S(x) in powers of >. (and thus y):

1 s (X~o ... ,x,.)=,

> ] s (Xe) +-1 -a2Ô
s ( c)
. . X Yi!lj
A A 21 Z X • 1 1

"" A"''- a• s
+L lc:::..!
k'
.
I
a .
.l:lt•••
íh
lJ,;
(x•) 31•• .. · y,, · (1.16)

The cha.nge of variables is such that the term quadratic in y is independent of >.. The
integral bocomes:

1 = ..-s<x•J;>. f •=Ifi dx, exp L.!. _a_2_s_ (x") Yi!li -

l 2! 8XjÔXj
R(y)] {1.17)

â"S
R(y) L"" ),k/2-l
-:;----;;-- (x<) y,, ... Y;. . (Ll8)
k,_,
We then expand the íntegrand in powers of ,/À: At each order we have to calculate the
average of a polynomial with a gauss.ian weight.
__________________ ,
 Algebroic Prelimínaries 1.4

1.3 Complex Structures

We shall often meet complex structures: we have 2n integratiou variables {x;} and {y;},
i 1, ... , n, and the integrand is invariant under a simult.aneo.,.; irlentícal rotation in ali
(x;,!l;) planes. 1t is then natural to introduce formal complex variables and z, which, z,
for normalization purp05eS, we define by:

z, = (x; + iy,) j,J2 , ;;;, ~ (x; - iy;) ;v0.. (1.19)

Note however that z, and i, are independent integmtion 11ariahle• anrl only formally
complex conj uga.tes since x, and !I< could themselves be complex.
The generic gaussían integral now becomes:

I (A; b, b} = f (ii t.=l

dzidZ,) exp (- .t
t,]=l
Z;A;;z; + t
t.'=')
(b;z; + b;i<)] , (1.20)

in whkh A is a complex matrix with non-vanish.íng determinant.

As before, to calculate tbis integral we first eliminate tbe terms linear in Z; and E, by
a shíft of variables, setting:

(1.21)

Tbe resulting gaussian integral can be calcula.ted either by returning to the "realp vari-
a.bles (1.19) or by a change of varia.bles like A.;v; v;. We obtaín

I (A;b,b) = (2í1r)" (detA)- 1 exp ft

lt.J=l
b; (A- 1 ) 1, b;]. (1.22)

By systematically differentia.ting wíth respect to b; a.nd bj, one establíshes Wick's tbeorem
for averages with the gaussia.n weigbt exp (- z,A;,%; ). Only monomíals with equal number
of factors z a.nd 2 have a non va.nisbing average:

~ A;·l , A:-J.P,_tSc
1 . • • . A:·• · . (1.23)
~ :JP1lt Jp,..t..,_
a.U pennut.atí.ons
p o{ {.;. •••• ,j.}

1.4 Integral Representation of Constraints

We sball oft.en use a simple identity about Dirsc lí-functions. By dell.nitioo:

/fi'1=1
dy;Ó (y,) 1. (1.24)

If we cha.nge variables:
y;=/;(x) (1.25)

------------··~--.,
1.5 Aigebroic Pre.líminaries 5

and assume that equation (1.25) defines a unique set of functions x,(y) for IYI sma.ll
enough, then we obtain the identity:

j ~TI dx;ó lfi[x)]} J(x)

lt--1
= 1, (1.26)

in whích J (x) ís the jacobinn of the change of variables (L25):

(1.27)

1dentity (1.26) hos a stnúghtforward generaJi.,ation: assume that we want to calculate

a function a(x) for x solution of the equatíon f{x) "' O, i.e. for x x (y =O), without
solving the equatíon explicítly. We can then use the identíty:

(1.28)

This ídentity, as well as the identities about gaussian integrais, has the ínterestíng prop-
erty that they can be easily generalired to an infinite number of vntíables.

1.5 Algebraíc Functional Tochniques

1.5.1 Punctional differentiation
In the di.scussíon of algebra.ic properties of correlation functions the concept of genemting
functionals will be very uaeful. Let f(x) be a function of a variable x, we shall consider
objects of the form:

F(f) (1.29)

in which pln) (:r,, ... , "'") is a symmetrk function of its arguments. We aha.ll also need
the concept of funetional derivative 6/6! (:r). It il; defined by the properties that it
satisiies the usual algebraic ruks of a.ny dífferentíal operator:

(1.30)

and in addition:
ó
óf(yJ'(x) = 6(:r y). (1.31)

This irnplies for example:

- - - - - - - - - - - - - - - - - -..
6 Al_qebrnic Prdiminari<s LG

J.5.í! Determino.nt.s of opero.to"

Often we shall have to cakulate determinants of operators represented by some kem•l
M(:r., y) which, after some transformations, can be cast ínto the form ó(x- y) + K(x, y).
Provided the traces of ali powers of K exist, the following identity, valid for any matrix
M,
lndetM"" trlnM, (1.33)

expanded in powers of the kernel K:

!ndet[l+K]= jdxK(x,x) ~~dx,dx,K(x.,xo)K(x,.x,)+

(-ljn+l
+- -0 - f dx, · · ·d:Z:n K (:r:t.x2) K (x,,x,) · · · K (xmxd + · · , (1.34)

will oft.en be useful.

1.6 G:rassmann Algebras. Differential Forms

We shall also de!>! with theorie; conta.ining fermions. Since f€Tmion fie!d correlation
functions (ar Green's functíons) are antlóymmetric with respect to tbe exchange of two
w:guments, the construction of generating functionals requires the introductiou of o.nti-
commuting classical functions, and thus Grassrnann vaxia.bles.
Gmssrnann o.lgebm. We only consider Grassmann nlgebrn.s over R or C (real or com-
plex). A Grassmann algebra 21 is a.n algebra constructed from a set of generators 6; and
their ant.icommuting products:

o, oi+ e, e,= o 'rli,j. (1.35)

Note th"t as a consequence:

(i) ali elements in a Grassma.nn algebra are lirst degree polynomials in each generntor;
(ii) if the number n of generators is finite, the algebra forros a finite dimeru;ional vector
spa.ce on R or C of dimension 2fl.
';!I ÍS aJso a graded aJgebra in tbe sense tha.t to any IDODOIDÍ!>J fl; 1 fi;, .. , il;. we C!lll
associa.Ul an integer p counting the number of generators in the prúduct.
Finally let us note that elements of lll are invertible if and only if their expansíon
as a sum of products of generators contairu; a term of degree zero which is invertible.
For exarnple the element 1 + (I is invertible, and has 1 fi as inverse; however fi is not
inverti ble.
Gmssmannian parity. On the a.lgebra '! we can implement " simple automorphism
which is a reflectkm P defined by:

P(IJ;) -8;. (1.36)

Then on a monomial of degree p, P acts like:

P(fi,, ... fl;,) =(-l)PIJ;, .. 6,,. (1.37)

The reflection P divides the a.lgebra 21. in twn eigenspaces 211 containing the even or odd
elements
(1.38)

- - - - - - - - - - - - - - . - . - - -..
1.7 A/gebraic Prelíminanes 7

In particular 21+ is a subalgebra, the subalgebra of comnmting elements.

Dífferential forms. An app!ication of Grassmann algebras is the representation of
differentia! forros. The !anguage of dilferential forms will not be used often in tbis work.
However iL is interesting to here recall one concept, the exterior deriva.tive of forms,
whose generalizatíon will appear in the context of BRS symrnetry (see Chapter 16). L<lt
us consider tota!ly antisymmetric tensons (}"•·····~' ( :z:), functions of n commuting variables
x". Associating n Gra.ssmann vari.ables 9" with x", we can write the corresponding 1-form
rl:
n = n.,. ...,,.. (x)il"' ... 9"' , (1.39)
where l :S n otherwise the form vanishes.
One ca.n define a dífferential operator d acting on forms:
a (1.40)
d <i' 8" Õ:z:l' •
We note that if rl is a 1-form, drl is a I + 1-form (see Cbapter 22 for dets.ils). One
immediately verifies that d is nilpotent:

d2 = 0"~8.~ =o (1.41)
Õxl' ar" '
beeause the product {JI'IJ" is antlsyrnmetric in J.l...., v.
We aJso recall that a form f! whlch satisfies drl Ois cal!ed closed and a form rl which
can be written f!= dfl' is called exo.cL The property (1.41) implies that any exact form
is closed.
Note that it is customary to write in the case of forros the generators of the algebra.
dx" instead of 9" and to then use the A notatíon for the product to show that it is
antisymmetr ic.

1.7 Differentiation in Grassmann Algebras

lt is then useful to define ditrerentiation in Grassmann algebras. A ns.ive definition would

be inconsi.stent due to the non-commutative cha.racter of the algebra. The problem can
be solved in the following way: Considered as functlons of a generator 8;, all elements A
of 21 can be wrítten
A A, +I!;A2 ,
after some commutations, where A 1 a.nd A 2 do not dE>pend on O,. Then by de!inition
ôA
ao, =A,. (1.42)

Note that the dillerential operator a;oo, is nilp<>lent (ôfa9,) 2 =O, like the form dilfer-
entíatíon (see eqnation (1.41)).
Remark. The equation ( 1.42) defines a left..differentintion in the seruse that the action
of a;oo, consists in bringing O; on the left in a monomial and suppressing it. Similarly
a right-differentíation could have defined by commuting 6; to the right.
Chain role. It is easy to vetify that chain rule applies to Grassmann differentiation. ]f
o-(0) belongs to 21- and :t(6') belong5 to 2(+ we can write:
a 8o-8f axâf
Fef(a-,x) ao ao-+ ao ãX · (1.43 )
------------------··
 ., ., · - - e - - - - - - - - - - - - - - -
Algebroic PreliminaTies Li

For the second term in the r.h.s. the arder between factors matters.
Formal construction. To show the consistency of the definition (1.42) and exhibit some
properties, we now more generally define differentiation in a Grassmann algebra by some
form.U rules, similar but slightly different lrom commutative a.!gebras. A Gtassmann
dífferentiel opera.tor D (also c<illed an anti-derlvati<m) acting on 21 is defined by the two
properties:
(i} lt is a linear mapping of 21, considered as a voctor space, into ítself
D(>. 1 A 1 +-\2A2) -\ 1 D(Ai)+.>.,D(A2) for.\i>.l.,ERorC. (1.44)
(ii) lt satísfies the condition
D(A 1 A;): P(Ad D(A2) + D (At) A,. (1.45)
The unusual form of equation ( 1.45) compa.red to the differentiation rule for commuting
variables is required if we WMt D to a.ntkornmute with P
DP + PD = o, (1.46)
which means tha.t the ima.ge of m± by D belonw; to 21'~'.
Note that if A belongs to m+ and F(x) is an ordínary function af real or complex
variables:
D IF(.4)]- D(A)F'(A) for A Em+. (1.47)
Note finally tha.t the forro dífferentiation ( 1.40) shares ali these propertie5, but a.ct.s on
different Vlll'iables.
Anticommutation relatían.. A short calculation shows that i f D and D' are two oper-
a.tors satisfying conditioru (1.44,1.45), then the antioommutator I'>:
I'> = DD' + D'D, (1.48)
is a usual differentiaJ operator:
t:. (Àl At +>.,A,) =>. 1 t:. (A,}+ .l.2t:. (Ao),
(1.49)
{ I'> (A,A,) =I'> (At) A,+ A,t:. (A2).
F'nrtbermore:
t:.P=Pt:.. (1.50)
These properties, which are the consequence of the addition of relation (l .46) to tbe
delinitions (1.44.,1.45 ), allow to extend the notion of Lie a.lgebra and are directly relevant
to tbe discussion of supersymmetries.
A hasis. Since a dilferential opera.tor •atlsfies conditíons (1.44,1.45), it is completely
defined by its action ou the generators e,.
In addition any differential opera.tor left..
multiplied by a.n clement of m+ still s!'tisfies (1.44,1.45). We conclude, therefore, that
a.ny dilferential operator ca.n be expanded on a ba.sis of operators â/OB; delined by:

_!!_e.1 = 61 (1.51}
aei ' '

with left. coefficients in m+. One eosily verifies tha.t tbe differential operators ôfôB;
coincide with the operators defined by equation (1.42). The nilpotent differential o~
erators 8/ 89;, together with the genera.tors 8; consideted as operatots a.cting on !li by
left..multiplíca.tion, satisfy the anticommutation relations:
â à é! â ô â
11;6;+9;9, o, --+--~0, 6;-ô +-6; 6íj, (1.52)
ao, àll; ae; ao, fi; ao,
and thus form a Ciifford algehta.
., .,--~~------------

1.8 Algebmic Frdtminarie•

1.8 Integration in Grassmann Algébrru;

We now also define integration over Grassmann va.ríables which we denote by the integral
symhol. v.re define the integration to be "" operation identical to differentiation

f d6;A: ~,A, \IA. E 21

The use of an integral or deriVHtive symbol is thus just a matter of oon""nience.

Gene m/ properties. Let us now show that this operation sa.tisfies the formal propertie>
we expect from a definite integraL Quite genero.lly we associate to a given differentia
opetator D an operator I which has the followíng defining propertíes. lt is a lineru
operator a.cting on 9:

which satisfies the three properties:

1D O, (1.55:
DI O, (1.56"
and
D(A} O= l(BA) = l(B)A. (1.57
ln addition it changes the gradmg in the same way as a differentio.l operator:

PI+IP=O.

Let us explain the conditions (1.55-1.57): condition (1.55) expresses that in tbe absena
of bounda.ry terrns the integral of a total derivative vanishes; conditíon (1.56) expresse
that if we integrate over a. va.riable, the r"-'ult no longer depends on tbis va.riable; fi.no.IIJ
condition (1.57) bnplies tbat a foctor whose derivative vanisbes can be ta.ken out of tb•
integral.
ln the <:ase of Grassmann o.lgebrB!i, if D is nilpotent, D itself satisfies o.li condítions
Tbe differential opera.to:rs 8/BB, 1m> indeed nilpotent.

1.8.1 Change of variableli in a Grassm1111n integro/

Let us consider the integral I:
1= f d{!j(O), (1.58

and perform the (necessarlly) a.ffine change of varíables:

(1.59

in which parity conservation implies that a E m+ and b E 21-. Tbe element a must b·
invettible, i.e. its term of degree zero in the Grassmann variables must be different fron
zero. Then using definition (1.53) we find:

f dli /(0) =a- 1 JdO' f(a6

1
+ b). (1.60
10 Algebmi<: Prelimínanes L8
1
We lind a very import>mt property of Gra.ssmann integrais: the jacobian is a- , "'hile
in the ca.se of oommuting vnriables we would have found a. This difference reftects the
identity of differentiation and íntegrntion for Grassmann variab!es.
Generolizatíon. More generally let us show that a change of variables

e,= e,(O'),
for which the matrix â(!,fEJf!j has an invertible pa.rt of degree zero, leads to a ja;:,obia.n
which is the inverse of the determinant of ae,;oo;.:
de, ... diJ" = d/f, ... de~J{I1'), (1.61)

with:
;-I= det ao, (1.62)
aer
Note thnt the determinant is we!l-delined hecau5e ali elements of the matrix Be;f ae;
belong tom+.
We a.gAin start from the identity between differentíation and integratíon:

f d8 1 ... d6n/ (8)

.
=IT ;;/ (8),
in which the product in l.h.s. is ordered. We then assume that f is a function of variable
e: and therefore, usíng chain role {1.43):

II~.tce)
i •

We now factoríze the element& 88j.féJ9; which comroute. The differentia! opera.tors a;ae:
anticommute (see equations (1.52)) and are thus ali proportiona! to the product ordere
from 1 to n. A sign is generated whlch is the signature of the permutation it. ]2, ... , j,
We then recognize the determina.nt of the mattix 8FJ',Jé18;:

The ídentity between differentiatíon 3Ild integration then irnmediately leads to equatio·
(1.61,1.62).
Remark. A straightforward verification of equation (1.61) is provided by the followiJ
e>:ample:
f ao, ... de, o, ... a, .
1f we make the linear changE of vatiables:

the result relies u pon t.he identity:

O., .. • 81 =e;, .. a; deta

 1.8 Algebro.ic Preliminaries

1.8.2 Mi:ud change of var-iables

ln thís work we shall meet integrais involving both commuting and anticorumuting WJ.ri
ables (bosons and fermíon.s). It is sometimes useful to perform mixed changes o f varíables
Denoting by e,(!' and x,x' the anticommuting and commutíng variables respectively
we set ( respectíng paríty):
e,= e, (x',li') E 2L(O'). (1.63
We introduce the matrix M of partia! derivatives

M (~ ~)'
with
âiJ; tJ6
Cto.=-{}'' D;; = &8:.
..:. J
lt is convenient to chaDge variabies in two steps:
(i) one first passes from (O,x) to (IJ,:r'). Thi• step generates tbe jacabian J 1 :

J1 detôx.l
ox~ 9
det(A-BD-'C). (1.64

(ii} One then goes from (8, :c') to (6', :r'). The second step just gives, as expiain..
above, the jacobian h:
(detD)- 1 . (1.65
The complete j~obian J, also called the beTUinian of the matrix of derivatives, is thru
J= D(x,(l) J 1 J 2 =BerM=:det(A-BD- 1 C)(detD)- 1 (L6f
- D(x',8')
For the jacobian to be not singular, the matrices A and D have to be invertible (an
therefore their contributions of degree zero in IJ').
í; 1mce of mixed matrices. In the case of the integration over ordinary commutin
d variables, if we perform a change of varinblcs inlinitesimally close to the ídentity:
"'• = "'~ + ~r.(x'),
then from identity (1.33), we see that the jaoobian has the form

J = det âx.,1 + ~ tr ôaf~ + O(t:2) =l + "aâf~ +O (t:2).

Bxf. xb :rt:i
Let us now consider the mixed C8Sie:
ns
B; = llj + "'Pi(:r', IJ'). (1.6~

Setting then
M = 1 HMJ+ O (e
2
), M1 = ( ~~ ~:) ,

we lind, as a consequence of identity (1.66):

&f. ârp,
J-l+t:{trA 1 trA 1 -trD 1 = - - - . (L6l
âx. &9;
To maintain the connection between jaoobian and trace, we are therefore led to defir
the trace of a mixed matrix, denoted by Str, as the trace of the commuting block mim
the trace o[ the anticommuting block:
StrM1 trA, -trD,. (1.6!
1.9 Alge~raic Preliminaries 13

Since the product f!;IJJ is antisymmetric in (ij), the matrix a;; can be chosen anti5ym-
metric
(1.74)
Expálldlng the exponential ln a pm••er series we ob:<erve that only the tenn of order n
which contains ali products of degree 2n in (J gives a non-zero contributiún:

J(a) (1.75)

In the expansion of the prodnct only the terms contu.lning a permutation of Ot ... t12n do
not vanish. Ordering then ali terms to foctorize the product 81IJ2 ... l12n we find:

1
J(a) = 2nn! (1.76)
})Ermutations P
IJf {i) ...1><~ .. }

in which ~(P) = ±l is the signature ofthe permutation. The quantity In the r.h.s. called
the pfaffian of the antisymmetric matrix a

I(a) Pf(ll). (1.77)

Grassmann integral technique; allow to prove a we!l-known algebru.lc identity

(1.78)

Let us indeed calculate I'(a):

(1.79)

We change variables, setting:

(1.80)

Tbe j!lCObia.n il; (-1)n. Also:

e,ei + IJ:IJj = fi•77J - iiJ'I•, (1.81)

d'72n .•. d'ltdi):!n ... diJt ; (-1)n' n d'ltdi).. (1.82)

Uslng the antisymmetry of the matrix a we find:

(1.83)

The Dime ~ -ftmction. In Grassma.nn algebra.. the role of the Dirac 6-functiou is played
by the f\lnction;
6(8) ~ 9. (1.84)

----------------··
12 Algebraic Prelimmaries !.9

l.IJ Gaussian Integrais with Gras.smann Variables

In this work we sha.ll maínly dea.l with Gra.ssmann algebras in which the generators can
be separated into two conjugated sets. We then denote by 8; and ê,, ' 1, ... , n, thesc
genera.tors. In many cases one can define a complex conjugation in the a.lgebra which
e,
exchanges and é;.
As in the case of commuting variables, we now calculate gaussian integrais, with the
same motivation: we shall in general try to reduce any integral to a formal finit€ or
infinite sum of gaussian integrais.
Let us consider first:

I(a) (L70)

According to the rules of Grassmann íntegration, the result is simply the coefficient of the
product BnBn ... Õ,6, in the €Xpansíon of the integrand. The integrand can be rewrítten
(no implicit summation over repeated índices):
n n
TI TI <1+ ii,a,;e; l .
i=lj=l

Expanding the product. we see that the ternts which give non-zero contribution to the
integral are oí th~ form

2: a..;.lln-t, ... , . .. a,;,Õ,.IIi•.. ÔtBir·

permut.ationll!l
{;, ..,;.)

Commuting the genera.tors to put them in the standard arder /l.,ll,. ... Btlll we find a sign
whích is the signa.ture of the permutation, and recognize the coefficlent as the determinant
of a;;:
J(a) = deta. (1.71)
The result ls the invetse of the one obta.ined with complex commuting variables. This will
eventua.lly lead in pertur bation theory to a sign (-1 )L in the front of Feynman diagram,;
with L fermions loops. The calculation "b""' is maínly a verification since we could have
changed variables
o.;;ll; e:, (L 12)
and used the form (1.61) of the jacoblan (nó summation over repeate<l índices)

J(a) deta j dflídlit ... de',.dtin exp (t, ii,e;)

- det a I lli=l
dlljdê, (1 + ti,e;) deta.

Rema-r*. We can a.lso more generally ca.lc:ulate a gaussian integral of the form:

J(a) = I d6 2•..• d9od9 1 exp (~ f: (J.ave;) .

t,J=l
(1.73)

- - - - - - - - - - - - - - - -..
14 Algebraic Preliminaries 1.9

Indeed
f d06f(O) /{0),

in which /(0) mea.ns th" constant part ofthe a!line function /(9). In some applications it
is useful to use an integrw representation of 6(6) similar to the usual Fourier tra.nsform:

6(ry) 'I j dO eliTJ, (LB5)

where 'l js an additional Grassmann variable.

Generol gaussian integT"!I-ls. We now consider another copy o{ the GraSsmann algebra
lll whose generators wi!l be deno~ed by 'li and ij;. Following the strategy of Section Ll,
we first calculate the integral:

(1.86)

in which the integrand is an elemento( the direct sum of the two Grassma.nn algebras.
The calculation as before relies on a cha.nge of variables:

(l.B7)

and leads to the result:

(1.88)

lfwe denote by (} averages with respect to the ga.ussian weight of equation (1.86), with
our defmition of ZG:
{}
- _ zG = deta (e,), (1.89)
8 'li
{}
-ZG deta(-9;). (1.90)
ÔTii
Note the sign in equation (1.90). Expressions (1.89,1.90) are the basís of perturbation
thoory. If we want to calculate the integral:

(1.91)

we can formally expand in a power series of V a.nd integrate term by terrn. We then find:

(1.92)
1.9 Algebroic Prelimmaries

Wíck's theorem. Following the same lines, we can derive Wick's theorem for fermionl
Defining no"''

..ii..,o;. exp (.t a;/!;fl;),

l,J=l
(1.9~
we can write

or using equation (1.88):

(iJ,,ojJi;,B10 ••• 8;JI;.) = (-lt

x{_!!_-4- ... _!!_~exp[-

ÔTh, &f}J, %.
tilt(a- ),,'1•]}1
Ôf/•• •.:~~1
1
(1.9~
'1-~,o

After explicit dilferentiation (which is the same as integration) we obtain:

(ii,,e,, ... íJ,•.fJ;.) = deta;;:.
c(P) (a- 1);p,i, (a- 1);,.,;, ... (a- 1);P.•• , (1.9!
permuUtiO!l!;
P oi [j, ...j.)

in which o:( P) is the signat ure of the permutation P. This result is analogous to ü
expression ( 1.23) obtained in the case of the integral over complex commuting varia.bl1
except for the sign.
Remarlt. Another property emphasizes the 8llalogy betweeo eomplex and G rassman
variables. lt is possible to define a scala.r product between entire functions by:
dzd% --
(l,g) = Í!f e-H f(z} g(z), (1.9~
/ 2
where the integral is defined in Section 1.3. Normalizable entire fuoctlons are then vectO'
in a Hilbert space. From the point o f view of Quantum Mechanics, the coefficients of tl
Taylor series expansion represent the components of the wave function f(z) on differe1
occupation number states (see Section 3.5).
!f we lmpose that complex conjugation exchanges the generators 8 and we can' e,
the same way introduce a scalar product between "analytic" Grassmann functíollS, í.
functions of (J only:
(f,g) = j d8dÕ e-s~ !(8) g(O). (1.91

A function !(8) is automatically an affine function of 8, thw;

!(8) =a+ b6, with {a, b} E C 2 - {O, O} ~ (!,/) = aa + bb >O.

Analytic Grassmann functions of one variable form a complex vector space of dime:
sion two. This is in direct correspondence with the Fermi-Dirac statistícs: the fermic
occupation number can only be O or 1.
16 Algebmic Preliminaríes 1.9

Bibliographical Notes
lntegration over antícommutíng variables was introduced in
F.A. Berezin, The Method of Seamd Quantitation (Aca.demic Press, New York 1966).

Exercises
E:ceTrise 1.1
Prove, using the method of fermion integration

(1.99)

E:rerr:ise 1. 2
Shaw that the condition that P is a homomorphism of algebr a is a necessary condition
for the oonsistency of the delinition of a dífferentiation operation in non-commutative
algebras by calculating D(ABC) in two different ways.

E:rercise l.S
Consider a gr11.ded algebra 1;!1 on C wíth one generator x saHsfying xn O. Construct a
diff.,rential operator D linear, decreasing the degree by one unit, and consistent witb this
structure.

E:ceTCise 1.4
Consider a graded algebra on C with generators x; such that

xf :0, (1.100)

and satisfying the commutation relation

for j >i,

where c is a complex number.

Show tbat a necessary and suffi.cient condition for the property {1.100) to be índepen-
dent on the basis in the vector space of genentors is é" = 1 witb c f l. In other words
express the condition

'lia; f {0,0, ... ,O}.

2 EUCllDEAN PATH INTEGRALS IN QUANTUM MECHANICS

In most of this work we sha.ll study Quantuw Fíeld Theory in its euclidean formulation.
This means that we shall discuss matrix elements of the quanturn statistical operator
e-BH (H is the hamiltonian and f3 the invei'Sé temperature) rather than the quantum
evolution operator .,-<Jit.
The former operator, whose traceis the quanturn partítion function ztm:

(2.1)

de&Iibes "evolution" in imaginruy time, a.nd in this :oeo.se most of the algebraic properties
which will be deriv.d, also apply to the reaJ time evolution operator, explicit expressions
being obtained by analytic continuation fJ - it.
Let us note, however, a few specific properties. The statistical operator provides a tool
to determine the structure of the quantum mechanical gmund state. For example if H
is bounded from below, the ground state energy Eo is given by:

Eo = lim (-!lntre-PH). (2.2}

P~= fJ
lf the ground state is in a.ddition unique and iso]ated, e-IJH projects, for fJ large, onto
the ground state vector lO):

(2.3)

Our basic to ois to study first Quantum Mechanics and then Quantum Field Theory will
be path íntegrats and functíonal 1nteg1'Cl/.9. We shall see that the path integral formulatíon
of quantum mechanics is wellsuited to tbe study of systems with an arbitrary number of
degrees of freedom. lt allows therefore a smooth transition between quantum mechanics
and quantum fie!d theory. One advantage of the euclidean formulation is tbat it is
generally easier to define properly the path integral representing the operator e-PH (the
Feynman-Kac formula) than e-iH<.
The euclidean functional integral ofteo leads to a sirnple and intuitl~ understanding of
the structure of the ground state of systems witb an infinite number of degrees of freedom.
In particular it gives a natural interpretation to barrier penetratíon elfects in the semi-
dassical approximation. Finally it emphasi7.1!S the deep ronnection between Qua.ntum
Field Theory and Statistical Mechanícs of criticai systems and phase transitions.
The main disa.dvantage of the euclidean presentation of quantum mechanics is that
classical expressions have a somewhat unusual form because time is imaginaty. We
shall spea.k of euclidean action and euclídean ]agrangían. In addition the calculation of
scattering amplitudes then involves an an&lytic rontinuatíon.
In this chapter we derive the path integral representation of the t~volutíon opetator
only for hamiltonians of the simple form p 2 + V(q). We then explicitly c&lculate the path
integral C<Jrrespondíng to a harmoníc oscillator ín a time-dependept e;dernal force. This
result can be used to red uce the e\'aluati<>n of path integrais in the case of analytic poten-
tillls to perturbation theory. Comparíng one-dímensionaJ classical statístic&l mechanics
16 Eudidean Path lntegmls in Quantum Mechanics 2.1

and quantum sta.tistical mechanics of the partide, we motivate the introduction of cor-
relation funetions and discuss their quantum mochanical meaning. ~'inally, retumíng to
real time, we construct a path integral repretOentation for the S-matruc. For illustration
purpose we use this representation in the case of the Coulomb potential to derive the
eikonal approximation.

2.1 Path Integrais: The General ldea

We considet a quantum mechanícal hamiltonian H (pA; t) in which f; and íj are the usual
momentum and position opera.tors:

lr!,p] = ili, (2A)

a.nd t the time. We wa.nt to construct a path integral representation for the rnatrix
elements in configuration spa.ce of the evolution operator in imaginary time U (t",t').
Since we have allowed tíme-dependent hamiltonians the operator U has no longer the
simple forro e-flH. \\e tbus start from Schrõdinger's equatíon:

au
11 ôt (t,t') =-H (t}U (t,t'), u (t', t'} = 1. (2.5)

In most of the exiUllples we shall encounter, U (t, t 1) ,.iiJ be de!ined only for t ::> t'.
The solution of equation (2.5) has the semi·gl"Oup propetey:

(2.6)

The proof is símp]e. One considers the operator U(t, t t} defined by:

{ u (t,tt} = u (t,t!)
~(t,tt) = U(t,t2}U(t2,tt) for t 2: t, ,
for t ::S tz.

Ü (t,t 1) is continuous and satisfies both conditions (2.5) (for t' =ti). It ís thus identical
to U(t, t1).
This property a.llows to write U (t", t') as a product of evolution operator~ on sm&l.l
time ínterv&ls:
n
U(t",t') TI Ujt'+mE,t'+(m-1}.:1, 1U' := t'' - t 1 • (2.7)
m=l

The product (2.7) is tíme-ordered according to the rnle of equation (2.6).

If the hamíltonian R ;., local in cnnfigurotion space, which means that its matrix
elements, in the base in which ii is diàgonal, (q 1 IH (t)l q2 ) have a. support restricted to
q1 = rn, we are able to derive from identity (2.7) a. path integral representation for the
matrix elements {q"IU (t", t'}l q').
First we write identity (2. 7}, taking matrix elements:

{q" IU(t",t'}lq') f TI
n-t

k~l
dqk
n
TI (q<IU(tk,tH)!q<-J)'
k~l
(2.8)
2.2 Euclidean Path Integrnls in Quantum Mechanf.cs 19

with the conventíons

tk = t' + kt:. 9o == q', 9n = q".
We then take the large n, small E limit. and thus reduce the evaluation of expression (2.8)
to the cakulatioo of the matrix elernents {q lU (t + <:, t)l q') in the small time interval
limit. In this limit, as a consequence of the locality of the hamíltonian, on)y the ma.trix
elements with 1'1 q'l small will contribute significa.ntly to expression (2.8). Note that
the operator U (t +E, t) is the equivalent of the tra.nsfer matrix in classical statistical
mecltanics (see Chapter 23).
In thú; éhapter we evaluate expression (2.8) for a special ciass of hamiltonians. General
hamíltoniaru; will be discussed in Chapt.er 3.

2.2 Path Integral Representation of tbe Evolution Operator

Hamiltonians of the form

(2.9)
(p,q now being d-dirnensional vectors) as well as all hamiltonians polynomial in p, are
local. Let us write the Schrõdinger equation for the evolution operator corresponding to
the hamiltonia.n (2.9):

(2.10)

with the boundary condition:

(q lU (t', f'}l q') = .stdl(q q').

When the potential V vanishes, (qiU (!,t')lq') is obtained by Fourier transforma.tion:

(q IV (t, t')l q'} =

f2 ddp
( 1rh)<~ exp
[1i (i(q q') · p- (t- t') p2)]
~
2
m

m
= ( 21rli(t- t•)
)d/2exp [-~~ m(q t')q')") ·
1
2(t-
(2.11)

1b solve equation (2.10) in the smal] E:= t- t' limit, it is con,"enient to set

• = t- t'' (q lU (t, t')l q') = exp {-a (q, q'; I, õ) /li)

Tbe functíon a (q, q'; t,t) is given at order e: by:

( q')2 d
a(q,q';t,<) =m~ + '2hlne+K +m 1(q,q';t) +0(•2), (2.12)

al(q,q';t)= [d.XY(q'+.l.(q-q'),t}. (2.13)

The constant K i• determined by compa.ring the limít • __,O with á (q- q'). This Jeads
to:
20 Euclidecm Path lntegrol.s tn Quantum MechanlC~ 2.2

Fundamental remark. We note that the singular terms in the f-expansion are
inde!l"endent of the potential. They imply tha.t the support of the matri.x element
(q iU (t, t')i q') Í5 lq'- ql - Jê, whích is typical of the brownían motíon (for details
see Cha!}ter 4). For lq' - ql ~ ..fi we have

!7i(q,q';t) = V((q +q')/2,t) +O(e) = V(q,t) +0 (<'1 2 ) ,

if tbe !}Otential is differentiable. Hence replaclng in the e><pansion (2.12} a 1 by for instance
V(q, t) modifie!; !7 by terms of order c 312 which, llS we shall see below, are negligible.
More generally for the three terms to be equivalent, the potential has to be at lea.st
continuous, other potentials requíre a spedal a.nalysís.
From equation (2.8) we thus derive:

m dnf2J"-l
(q" !U (t",t')l q') = ~~"" c1r.li.<) TI ddqk exp [-S(q,<)/lí}, (2.15)
1:--l

with:
2
S(q,t) ~ [m(q•-~H) HV(q•,t•)]. (2.16)

We observe that higher orders in e in (:!.12) give vauishing contributions in the smali
E, TI€ Iixed, limit. Let us now introduce a function q(t) which interpolates in time the
variables q;,:
q. ""q(t.).
The integral over the variabJ., q~
is now equivalent to the íntegration over the points of
the P"iecewise linear path represented in figure 2.1.

q(t)

~
. . '

t,.._l t"

Fig. 2.1 A path rontríbuting to integro.! (2.15).

In terms cf the function q(t), the function S(q,E) has the formallimit:

lim S (q•,<) = S(q) <=

c-o
1'"
t'
dt [~rru.'J.~(t] + V(q(t), t)}. (2.17}

The functkma.l S(q) is the eudídean a.ction, the integral of the "euclidea.n lagra.ngian"
a.ssocia.ted with the original hamiltonian.
2.3 Euclidean Pa.th lnte~ in Q.uantum Mechanícs 21

We wríte the formal Jimit of expressíon ( 2.15) a.s:

(q"!U(t",t')]q') 1 q(t")-q"

q(t;)=ql
[dq(t)]exp[-S(q)/h], (2.18)

and call ít path integro! sínce the r .h.s. involves a summation over ali paths satisfying
the prescribed boundary conditions, with a weight exp [-S/Ii].
The expression (2.18) immediately shows that paths which minimize the action (2.17),
i.e. paths c]ose to classical eudídean paths, which satisfy:

6S
oq;(t)

give the largest contríbutions to the path integral. This observation is at the bMis of
semicla55ical approximatíons.
Discussion. Note that in actíon (2.17) the two terms play quite different roles. The
kinetic term J dt 4? selects paths regular enough, í.e., as one ca.u see from expressíon
(2.16), three for which [q(t + õ)- q(t)]~ f<: remains finite when f: goes to zero (which in
particular implies that the paths are continuous). The kinetic: term thus determines the
functional space over which to integrate and is es5ential to the very existence of the path
integral. The relevant paths are typic:al of the brownian motion (see Chapter 4 ). The
potential weights the paths accordíng to the value of q(t) at ea.ch time, and determines
the physical properties of the theory.
Note that we shaU always write the integratíon meesure [dq(t)] with brackets to dís-
tínguísh path integrais from ordínary integrais.
Note also tha.t in the symbol [dq(t)J ís buried an ínfinite nonnalization foctor N:

N= (~)dn/2.
2trlié:
Therefore in explicít calculations we shall &lways normalize the result by díviding the
path integral hy a reference path integral for which the result is exactly known ( the free
motion V= O for example).
Generolization. The generalization of the prevíous constructíon to sever&l particles is
straightforward and leads to a path íntegml involving the corresponding eudidean action.

2.3 Explicit Evaluation of a Simple Path lntegral: The Harmonic Oscillator

in an Externai Force

At this point in the disc:ussion it may not be very clear bow useful the concept of pa.th
integral is. So let us give a simple example which shOW5 that even in quantum m&
chanics, sometimes, the path integral representation símplifies ccmsidera.bly &lgebraíc
manipulatíons. We consider the hamiltonian of a harmonic oscille.tor ooupled linearly to
a tím&dependent extemal force:
I 2 ~ qf (t)
H = - p2 + lrru.lq
2 (2.19)
2m
The corresponding eudidean action Sa(q) is:

(2.20)
22 Euclidean Path Integrnls in Quantum Mechanics 2.3

which is a quadratic form in q(t). When this expression is inserted ínto equation (2.18).
ít le&ds to a ga ussia.n integral which can be caiculated following the method explained in
S.,Ction 1.1. \>.ie first el"tminate the linear term in cxpression (2.20), changing varíables:

q (t) = q, (t) + r (t), (2.21)

in which q, is the solution of tbe classical equation of motion:

witb Q0 (t'} = g', Qc (t") = q"- (2.22)

Tbe new patb r(t) then satlsfies the boundary conditions:

r (t') =r (t'') O (2.23)

As a consequence of equations (2.22), the action (2.20) can be written e.s the sum of two
terrns:
SG(q) = S {q,) + So (r), (2.24)

in wbich S (q,) is the classical action on the trajectory and So(r) ís simply:

(2.25)

Thus the integral over r(t) no longer depends on f(t) and q',q". We now first calculat<!
explicitly S (q,) and then evaluate tbe remaining p.ath integral.
The classica! o.ction. Owing to the linearity of equatíon (2.22) we can wtite the solution
q< as tbe sum of two terms, one depending on the boundary conditiolll> q', q", the otber
on f(t). Setting /3 t''- 11 and

we find

x 1 (t) = q' sinh [w (t" - t)j + q" sinb !w (t - t')], (2.26)

x 2 (t) = _!._
wm
lt"
tt
[9(t- u)sinh(w(t"- tl) sinh{w(u -t'))

+8(u- t) sinh(w(t" u)) sinb(w(t- t'))} /(u)du, (2.27)

8{v) being the usual Heaviside step function.

To calculate S (qc) it is useful to note:

~md use the equation of motion (2.22). One tben finds:

1 (2.26)
wm sinbw/3 (h+ h+ h)'
2.3 Euclidean Path lntograls in Quantum Mechanics 23

(~1 )
2

h= [({'+q"2)coshwjj'-2q'q"].

I t"
h -mwf t'
du [q"sinhw(u t.')+q'sínhw(t" u)J/(u), (2.29)

!3 = [ " dudvB(v-u)sinbw(t"- v)sinhw(t' -u)f(u)f(v).

The path integrfll. We still want to cal.culate the gaussi8ll integral. over r (t) because it
yields a w-dependent normlllization factor

(2.30)

with r (t') r (t'') O.

The boundary conditíons imp)y f r2 dt = -I rrdt, and therefore the quadratic fonn
in the exponential. can be rewritten:

with:
A (tt, t2) = (- (d~:) 2 + w2 ) li (t 1 - t2).

The gaussian integral is proportional to the determinant of the hermitia.n differentilll

operator A (! 1 , t 2 ) to the power -1/2 (we omit the multiplico.tive factor m/ fi which is
w-independent). We caJJ calculate the detennin8llt by diagonalizing the operator. We
have to solve the equation:

(2.31}

in the space of functions satisfying the conditions:

rn (t') =r,. (t'') =O.

The normalized eigenfunctions are thus:

,-,.(t) J2/Psin[rrn(t-t')/Ji'], n>O, (2.32)

and the corresponding eigenvalues >.,. are:

(2.33)

2
The determinant of the opera.tor [- (d/dt) + w 2] is the product of eigenvalues:
24 Euclidean Path lntegrol8 in Quantum Mechanics 2.3

Evaluat.ing explicitly the r.h.s. we get an expression for the path integral.
A difikulty rernains: the ínfiníte product in t.he r.h.s. of equation (2.34) ís dívergent.
However the ratio:

(2.35)

ís fmite.
Thus the e.xpression for the evolutíon operator is:

(q" jU (t", t')! q') = Af'J s1n

.wh/3w/3 exp 1-S (qc) /h] (2.36)

The new norrnalization constr.nt N' is independent of w r.nd c:an be fixed by comparíng
to the free case w =O (see equa.tíon (2.11)):

!~o (q"jU (t", t')l g') =

J~o
c:h/3) 112
exp [- ';p (q'
2
q")2],

obtained directly by solving Schrodinger's equa.tíon.

In this limit the classical action S (q,) becomes:

S(qc) = ~ (q' 11 2
Q ) •

Together with equation (2.36) this determines the constant Af'. The complete result is:
(q''jU (t", t')jq') = (q"jU0 {t",t')jq'}

x exp ( k["L (u] f (u) 1

du + ,. [ "
2
K (u, v) f (u) f(v)dudv] ,(2.37)

with the definitíons:

~
112
(q'' !U0 (t",t')i q') = [ ]
2'1Th smbwtJ

x exp {
2
hs:w/3 [(q"' + q't2) coshw{j- 2q'q")}, (2.38)
1
L(t) = ...,..-h [q" sinhw (t- t') + q' sinhw (t"- !)] , (2.39)
sm w{J

K(t1,t2)= 111WS:nbw(JIII(h tt)sinbw(t''-tz)sínhw(tl t')

+8 (t 1 - t2) sinhw (t"- h) sinhw (t2 t')]. (2.40)

Remark. An equivalent "nd useful presenta.tion of the evaluation of the r(t) path
integn.l is the followíng: We expand, in the Hilbert space 1:, 2 , the functíon r(t) on the
ortbonormal basis formed by the fnnctíons rfl (t]:

r(t) = L c,rn(t).
n>O
2.4 Euclidean Patil Integrais in Qt.antum Mechanics 25

The functional measure of integration is then

[dr(t)] =TI de,, (2.41)

and the quadratic form Sn (r(t)) becomes in the variables c,:

So (r(t)) = m'"'
2ii L.. ÀnCn.
2
n>O

Tbe gausslan integration over the variables c, yíelds a result consistent with equation
(2.34) up to an w-independent infinite normàlization fa.ctor. The functional measure
can thus also be defined as the integration over the coefficients of the expansion of the
function describing the path over an orthonorm&l basis ín .C.' as in equatíon (2.41 ). This
alternative definítíon will be useful for the ínstanton càlculations in Chapter 37.
The p<>rtition function. From the equations (2.37,2.38) we can càlcula.te the partition
functíon. Ta.king the trace of U (fJ/2, ~{3/2) we find (we set li= 1):

tr U(~, -ü = trUo (~,~~) exp [ 2~ t:>udut. (v -u)f{v) f (u}], (2.42)

with:
trUo(f3/2, -!3/2) = .,-fiw/2 [1- .,-p...r l
a.nd
1
ll.(t) = 2w sinb{w{3/'2) cosh(w(f3/2 -ltl)). (2.43}

The function A (t) is the So]ution of the equation:

with periodic boundary conditions t.(fJ/2) t. (-fJ/2).

2.4 Perturbed Harmonic Osciltator

Let us consider the hamíltonian H:

1
H= 2mp2 + !mw2q2 + \'J{q), (2.44)

in which we impose to \'J(q) only to be expa.ndable in powers of q:

The corresponding evolution operator is gíven by (in this section we set h= 1):

(2.4:>)
26 Eudidecm Path lnteyrols in Quantum Meehanics 2.4

To evaluate exptession (2.45), we âpply to the expression (2.37) the genern.lization of

identitíes (1.10,1.12). In Sedion 2.3 we have cakulated the evolution opera.tor Uo cor-
re;ponding to action (2.20):

(q"IUo (t", t')l q') = j [dq)exp f-[" (~mq 2 + ~mwV qj(t)) dt). (2.46)

lf we differentiate with respect to f(t) we obtain:

of ~t ) (q"IUo (t'',t'JI q') =f [dq]q (tt)exp(-So (q, f)].

1 (2.47)

Therefore by differentiating p times with resped Lo /(t) we can genera.te any product
q(t,) ... q(t,):

fi 61 ~t
J=1 J
) (q" lUo (t", t')l q') =f [dq] fi q (tj)exp 1-Sc(q.f)J.
,=1
(2.48)

More generally a. differential opera.tor F(lJjlJf (t)) generates in the r.h.s. the functional
F (q (t)).
Applied Lo exp [- J,;"
'Vj (q (t)) dt), these identities lead to:

(q"[U (t", t')l q'} "' { exp [- [" dt Ví ( li;(t))] (q"IUc (t'', t')l r/)} 1/=0. (2.49)

Using then the result of equa.tion (2.37) we obta.in:

(q" [U (t" >t')l rl) (q" w~ (t"' 1')1 q')

x exp [-['' dtVí (

6 /(t})) exp [{L, f)+~ (J,KJ)jlf""', (2.50)
(expres:sion (2.37) has been rewritten in symbolic notation).
lf Vr(q) ís a polynomíal we can expand (2.:i0) in powers of Vr(q} a.nd calcule.te system-
a.tically the.successive terms in lhê m<pansion using Wick's theorem (1.9). This ís tbe
basis of perturbation theory.
Example:
Ví(q) = .l.q'.
At fu-st order in À:

(q" \U (t",t')l q') = (q" [Uo(t",t')lti'}

x [1 À{' (6/(t)y dt.) exp [(L,f} +i (/,Kf)}L
0
+O (ÃZ). (2.51)

Differentiating 4 times with respect to f and setting f to :~~ero, we pkk out the terms of
degree 4 in f in the expansion of

exp [(L, f)+~ (f,Kf)].

 Euc!idean Palh [nlegrnls in Quantum Mechanics 27

Note that

In the expansion of the exponential the terms of degree 4 in f are;

Combining these various expre!!Sion.s we find:

(2.52)

The classical !imit and perturbation theory. A perturbation expa.nsion can be generated
for any decomposition of the potential in the form of the sum of a qurukatic term àlld
a remainder a.s in equation (2.44). However if we want the perturbative expansion to be
assodated with a formal expa.nsion in powen~ of li, then we see from expression (2.18)
that we h ave to look for mínima of the a.ction, i.e. for mínima of the potential. Callíng
Qo a mínímum of the potential we then wríte:

and this is the startíng point of a form of perturbation theory whích leads to the loopwise
expansion. Problems associated with a possible degenera.cy of the classical minimum will
be exBmined in Chapter 40.

2.5 Partition Function. Correlntion Functions

In the last pa.rt of the chapter we restrict ourselves, for simplicíty reasons, to tíme-
independent potentials.
The partitioo function. Let us fin;t note that the quantum partition function Z({3)
tre-ll 8 has a path integral repr.,..,ntation whlch is immediately deduced from the ~ep­
resentation of the evolution operawr

Z({3) ""'tre-CH := tr U(.6,0) =f dq (q] U(iJ,O) lq) =f (dq(t))exp[-S{q)), (!2.~3)

where the paths now satisfy periodic boundary conditíons; q(O) q({3), and we integra.te
over ali V&lues of q(O).
The larye i3 limit. One c~ also set the boundary conditions at t "" ±{3/2. Since the
action is time translation invariant the resu]ts are the sa.me. Howev.er in the large {3 limit
(relevant for the ground state energy) in the first case we are led to integrate over paths
on the positíve realline with the boundary condltion q(O) =O, whlle in the 5econd case
we obtain an explicitly time tra.nslation invariant formalism on the whole realline. The
IBtter formalism is clearly simpler.
!8 Euclidean Path lntegmls in Quantum Mechanics 2.5

!!.5.1 Quantum and classical stati.<ltícal mechanics

The time-discretized form of the path integr.J has an interpretation in terms of one-
dimensional classical stati.sticf>l mech!lnics on the lattice. Let us indero consider the
partítion function Z

Zn(T) =f TI dp(qo) exp[-E(q,)(T]. (2.54)

k=J

where q; characterizes the configuration on the site i of a 1D lattiee. n is the size of

the lattice, T the temper!lture, dp(q) the distribution of the q variable and E(Q<) the
configumtion energy whích we choose of the spedal form

E(q) =L:" V (q; 2

qi-1) ' J >o. (2.55)
•=I
Finally we assume periodic boundary conditions: q,. = q0 • We then recognize a.n expres--
sion very similar to expression (2.15). Let us show that the oontínuum limitE--> O of the
quantum expression corresponds here to the low temperature limit, large volume limit.
lt is oonvenient to set

The simplest exa.mple corresponds to a function v(q) with a. uníqu~ minimum a.t g =O
where ít is regular
v(q) = ~v 2 g 2 +O (q3 ) .
For T small the lea.ding configura.tíon is obta.ined by minimizing the energy {2.54) and
corresponds to ali q; 's equal. Then for n large only the neighbourhood of the conliguration
which is a maximum of the distribution, i.e. g; =O, contributes to the partition function.
Wefind

Zn(T) ~f IT dq* exp (-E(q;)), (2.56)

k-1
with
Ê(q) = ~ t [J
i:.l
(g, q;-1)
2
/T + v2q?] (2.57)

Aft.er the rescaling q; ,__. q,T 1 i•n-i/2 we recognize tha.t Zn(T) is all$0 the qua.ntum par-
títion function Z(tl) of an harmonic osdllator, where (3 = nv'T, and the hamiltonian H
ís
(2.58)

2. 5.1! Con-elatinn junctions

The analogy between classical and quantum sta.tistical mechanics suggests to consider
the integra.nd in path integr.J (2.18) as a probability distribution. In particular we can
use it to calculat.e averages, like the q(t) correlation functions:
2.5 Eudídean po.th Integrais in Quantum Mechanics 29

where we ba~ assumed periodic boundary conditions, q(O} = q(,1). The normalization
Z(.l'l) is fixed by requiring that the average {1} is 1. We recogni.ze the quantum partitioiJ
function (2.53), Z(jj) = tte-IIH.
V.'ben the euclidean adion is not real, averages of the form (2.59) no longer have
a probabilistic interpretation. For reasons whicb ,.·e shall explain below, they still are
useful quantities to consider, and we shall still call them correlation functions.
Operator mterpn:tatíon. The expression (2.59) has ll.Il operator interpretation. Lét us
assume that we have ordered times such that:

(2.60)

We decompose the interval (0,8) into n + 1 subintervals (D,t 1 ), (t 1 ,t 2 ), ••. ,(t~,t'l). The
total a.ction is the sum of the correspanding contributions:

S(q) ='f t
i-1 t,_l
(~mq 2 + V(q)] dt, with to O, t,.+ 1 ,1. (2.61)

We then rewrite integral (2.59) with the use of the identity:

fi
":1
q (t,] = f fi
t-1
dqdJ i'l (t;) q;] q;.

The path integral then fa.ctorizeo; into a product of pa.th integrais corresponding to the
different subintervals. Returnlng to the very definitínn of the path integral (equations
(2.17,2.18)), we see that the numerator In expression (2.59) is exa.ctly (recalling the
ordEring (2.6))

lntroducing Q(t) the Heisenberg representation of the operator q:

(2.62)

(for t real and positive e' 8 does not necessarily exist and then the definition is somewha.t
formal) we can wríte:

(2.63)

In the l.h.s. of this equation the time-ordering (2.60) is assumed.

If we set all times equal we simply find the static statistical average of a. power of
the coordinate q in finite temperature quantum mechanics. More generally, after ana-
lytic oontinuation to real times t;, (but nnt in ,1) we find the time-dependent q uantum
oorrela.tion functions.
lf we intraduce a time-nrdering operation T, which to a set of time-dependent opera.tars
A 1 (t 1 ), ... ,A1 (t!) assodates the time-ordered prnduct (T-product) of these operators, for
example for I = 2:
30 Eucl.dean Path lntegrobl in Quantum Mechanic• 2.6

we can rev.'Ti"te expression (2.63) írrespectíve now of the arder between the times t1, ... , t~:

(2.64)

Final!~· íf H has a unique isolated ground state lO), then equation {2.2) shows that:

(2.65)

These time-ordered products are the analytic continuation in imaglnary time of the
tíme-ordered products introduced in the real time formu)ation of quantum field theory.
After a.ne.lytíc continuation, they thus generate Green's functions from which one can for
instllllce calculate scattering amplitudes (see Section 7.3). However, physkally reasonable
tbeories from the point of view of Green's function•, do uot alwa.vs correspond to real
euclidean actions and then their analytic rontinua.tions are correlati<"ln functions only in
a formal sense.
Remark. We have assumed ín taking the large {3 limit that the ground state is unique.
The existence of phase transitioru, as will be discussed ín Chapter 23, ís related to a
póssible ground state degeneracy.
Perturllation throry. We have shown in Sectíon 2.4 how to calcul ate perturbatively the
path integral for any hamiltonian of the form p2 /2m + V ( q) in tenns of path integral
( 2.46). The argument can ímmediately be generalízed for the corresponding córrelatíon
functions.

2.6 Evolution in Real Time. The S-matrix

To calculate matríx elements of the evolution operator in real time we can proceed by
analytic continuation replacíng {3 by 13 e'"' in ali expressions and rotating in the campl""
{3 plane from 'P = O to 'P 1r J2 in the positive dírection.
The evolution operator U (t'', t') then satisfies Schrõdlnger's equatíon:

i/i: (t, t') =H (t) U (t,t'), (2.66)

which, for the example of a hamiltonian of the form

leads to:
(q" W (t'', t')lq'} = J. •(•")-•"
oít'l~•'
[dq (t)J exp [tA(q)fli]. (2.67)

The function A(q) ís now the usual dasskal action, integral of the lagrangían:

A(q) = f'" dt [~mÇ 2 - V(q,t)].

},,
(2.68)

Expression (2.67) relates in a beautiful way classica.l a.nd qua.ntum mechanics. In the
quantum mechanical evolution all paths contribute but they are weíghted with the com-
plex weíght eiA/ h. Therefore path.s dose to extrema of the actíon, í.e. dassical paths,
2.6 Euclídean Path Integmt.. in Quantum Mechanics 31

give the largest contributions to the path integral. ln particular if the V<!-lue of the da.~­
sical action on classical paths is large cornpared to 11, regions in phase space dose to the
dassical paths dominate the path integral.
Th• S-matru. The scattering S-matrix is obtained by cornparing the quantum ev<:>-
lution to the evolutíon in the absence of the potentíal. More precisely the S-matrix can
be defined as the limit of the evo!ution operator in the ínteraction representation. For a
time--independent potential:

S = lim e<Hot"fhe-iH(t''-t']ft.e-iHot'fh, (2.69)

t.'--oo
tn--roo

in which H 0 is the free hamiltonian correspondíng to H. For exarnplê in Rd:

We as..ume that the potentíal decrea5es fast enough for large lql for the S-matrix to exist.
Note that in momentum representation the relation (2.69) between the S-matrix and
the evolutíon operator takes the form:

(P"IS!p') lim e•E''t" fh (p"IU(t", t')lp') e-•E't' /li' (2.70)

t~--oc
t''-+oc-

whêre
E' E(p'), E"= E(p"), E(p} = p 2 j2m.
However the límit has to be understood mathematically in the sense of dístributíons (one
should use test functíons, j.e. wave pa.ckets).
We finally recai! that the S-matrix in momentum representation i5 in general parame-
trized in terms of the Sçattering matri:x T as:

(p" ISI p') = 6d (p" p') 2i7r6 (E"- E') (p" ITI p'} (2.71)

2.6.1 Path integro! and S-matru in perturbation theory

Let us show first how the path integral for the evolution operator can be expanded in
power$ of the potential. The path integral formalism then organizes the perturbative
expansion in the same way as the operator formalism recalled in Section A2.L
We thus consider the harniltonian H

H = ;:f + V(:z:), (2.72)

(settíng for convenienc:e m = h = 1 ). The dassical actíons corresponding to the ft~

hamiltonían and to H are

.A(.:r) J.t" [~:t>(t)- V(x(t)])dt.

I'
(2.73)

We want to calculate the evolutíon operator in a power series of the potential V(.:r) and
then obtain the S-matrix. We now assume that tbe potential has a Fourier representation
 Euclidean Path Integrnls •n Quantum Mechanic> 2.6

~he path integral (2.67), expande<! in powers of V, then takes the form

(:r" lU (t", t')l x') -1 r(t;~}-~''

=!t'l=•'
ldx(t)j exp liA(x))
1
= (*")~x" ldx(t)] ei.Ao(~) L c-;)l [1'" V(x(t))dt] (2.i5)
fz(t') .... :r.' l i. t'

[f we now introduce the Fourier representation (2.74) in the path integral we see that we
>ave reduced perturbation theory to the calculation of gaussían integrais. The /'h term
becomes

(x"lul1l(t",t')lx') (~?' fgv(k;)dk; {'I} &r;

X 1=(<")=~"1dx(t)Jexpí
.r-(tl),.,.:rl
l! {"'ft,
:t2(t)+ Í:k;x(rJ)]. (2.76)
)

The path integral gíves a physical interpretatlon to the terms in the perturbative expan-
sion: The leading patb oontributíng to the J1h order is a succession of stra.ight lmes, where
at times T!, • • , TI the momentum changes by amounts k 1 , •.• , k1. One then integrates
the corresponding pha5e factor over ali times and ali momenta weighted by V( k ).
The term of order zerO in V ylelds b(p" p'), let us then calculate explícitly the first
arder. We can write the evolution operator as the products of two free evolutions ftom
t' to .,. and from r to t". We then FOurier transform with respect to x' and ';r". We find

(p''IUl1l(t", t')IP') ==-i f dk V(k) i. t"

M

X f dx(1')e-••"'(t"-•l/2 (p"ix(T)}e'kz(r) (x(T)\P')e-•P''(r·t')/>. (2.77)

The integration over the variable x(T)' ~'ie]ds 6(k + p' p 11 ). We then find for the first
order oontrib.ution sn l to the S-matrix

( P"ISOl IP') "' t ,--oc

lim -iV(p"- p') 1 t"

ti
dT eh(p''->'')JZ.
t' 1 -+oo

lt is clear tbat this expression has no limít as a functíon but as a dístribution we obtain
a ó-function of energy conservatíon:

(2.78)
2.6 Euclidean Pafh integrais in Quantum Mechanics 33

2.6.!2 Path integro/ and tile S-matrix in lhe semiclassicallimit

The representa.tion (2.57) of the evolution operawr (eads to a path integral reptesentation
for ma.ttix elements ar the scattering S-mauix whích ÍS particularly ,n suited to study
the semíclassical limit.
Let us calculate the matrix elements of the S-matrix between t-..'0 wave packets:

(.P2 :s1 w,) = ,.~':'= j <lq'dq" ( .P2I"'n•'"1•1 q") (q" 1u (t", t')l q'J ( q' 1.,-mo,· 1"1'~''•) .
t'F-+00
(2.711)
lntrodudng the two wave functions ~ 1 (P) a.nd ;j,., (P) in momentum representation a.s-
sociated with the vectors IV'!) and IV-•2) we ca.n define:

,pl(q,t) = (qleno</hl>~-·•) = j ~~:~d~l{p)exp [í(p·q-t:~) !h], (2.80)

and a similar expression for 1/!2.

When t becomes large the phase fo.ctor in expression (2.80) bas large osdllations, and
the integral is dominated by the ste.tíonary points of the phase:

:p (p·q-1:~) o = q t-!;;. (2.81)

Integral (2.60) Í5 thus equivalent to:

_
,p, (q,t)l•l:~ 1/ll (p) (2")d/2
1 (m/i)df2 (ir. p• )
lif exp 4sgnt+it2mli ' (2.82)

with
m
P=Tq.
We then change variables in integre.! (2. 79), setting:

(2.83)

and obtaín:

/dp'dp"~:i (p'') ~~ (p') exp [-lii (t''p2m'"' -

2
\>h ISJ,p,) 0< lim t'p' )]
tt--'=:10 2m
fU-+(X}

l< (t"~ JU(t",t')lt'~). (2.64)

In this equation we can now introduce the path integral representation (2.67) of the
evolution operator:

t"p"
( m
lU (t", t')l t'p') = iq(t")~t"p"/m ldq (L)] exp [-lii A (q)l ,
m q(t.l)=t'p'/m

We conclude that the S-inatrix is obtaíned by calcula.ting the path integral with classical
sca.tteríng bounde.ry conditions, i.e. summing over pa.ths solutions at large time of thE'
free equation of motion. In particular if we kncrw hcrw to solve the ciMSical equations of
motion with such boundary conditions we can calculate the evolution operator and thus
the S-rnatrix for li srnall. This leads to semiclassical approximations of the S-matrix. A
calculation in thia spirít ia presenteei belcrw.
34 Eudidean Path Integml.s in Quanto.m Mechanics 2.7

2. 7 Application: The Eikonal Approximation

From the path integral representation of the 5-matrix, it ís easy to derive a well-known
approximati(m for the scattering amplitude. valid in the high energy, low momentum
transfer regime: the eikonal approximation.
In the absence of a potentia.i the evolution operator is given by a gaussian path im.egral
which we can calculate s.s usual by first solving the classical e<j\tation of motion. The
solution which satisfies the boundary conditions implied by representation (2.67) and
corresponds to the free hamiltonian:

is:
q(t ) =q' + (q" t t'
-q') t""t'" (2.85)

'II"anslatíng the integration varíables q (!) by the dassical solution [2.85) we then have
to calculate a normalízatíon integral which we can obta.in by comparing wíth the exa.ct
result (h l ):

u " 1 ,
(q jU (t , t )I q) =
(
Í1r
m
(t" t'}
)d/2 eKp [·m(q"-q'J
• (t'' t!}
2
]
. (2.86)
2 2

The idea of the eíkonal apprOxímatíon is that in the momentum regime:

p'=p-k/2, p"=p+k/2,
the dominant contributions come ftom classical trajectories which themselves can be
approxlmated by the stra.ight lines of the free motion (2.85). The calculatíon of the
evolution Dperator is then straíghtforward:

(p + k/21U (t", t')jp- k/2) ~ j dsd:r: exp[-i (p · s + k · x) +iA (s,x)], (2.87}

in which the classkal action is now:

im s 2
A(s,x}= 2t"-t'
ri"
ijt' dtV x
( s t - t' )
2+~S . (2.88}

The normalization in equation (2.87) is determined by comparing wíth the result (2.86)
for the free motion.
Taking the large time !imit, we find that the integral over s is dominated by the sa.ddle
paínt:
s == (t''- t'}pfm. {2.89)
We assume that the pDtential decreases fast enough for the integral in (2.88) to have a
large time límit. Once this limit is taken we can make an arbitrary shíft on t he integratíon
variable t. llis ha.s the effect on the argument of the potential V that we can translate
the vector x by an arbitrary vector proportional to p. We can choose thís vector ín such
a way that V depends only on tbe component b of x orthogonal to p:

b= x -p{x·p/p 7 ). (2.90)
2.7 Euclidean Po.th lntegr<W in Quantum Mechanics 35

The integra.! over the component of x along p can then be performed a.nd implies: p ·k O
which just expresses energy conservatjon.
We then obtain:

{p+ k/2IU(t 11 ,t')IP -k/2) ""N(p) f d'Hbe-•k·bexp [-i/_:"" dtV (~ + b)].

(2.91)
wíth:
N (p)- exp [; (!" -t') :~] (2.92)

Equation (2.91) yíelds, after Fourier transformation, the matrix elements of the transition
uperator T in momentum representation (defmed by equation (2.71)):

{p + k/21TI p- kf2) ~i ~I f~::;~ e-ik-1> {exp [-i L:<x> dt V(~+ b)] l}.
(2.93)
The Coulomb potentiaL Let us use the eikonal appraximation to evaluate the scattering
amplitude for e. 1/q Coulom!rlike potential. The integral over the potential has in this
ease no lnfinite time limit beca.use the potential decreases too slowly. lt is necessary to
integrate over a finite time ínterval. We write the potential:
Q
V(q) = jql' (2.94)
then:

1•" (E.+ b) ~ (-t't" 4~ 2

dt V In (rm ) • (2.95c}
,, m p ml?
The appearance of this inftnite phase has the following meaning: since the Coulomb
potential decreases too slowly at la.rge dmtane«, the c)assical trajectory approaclles too
,]owly the free motion whlch, in our definition of the S-matrix, has been t&ken as the
reference motion. In the Coulomb potential only the cross-sections are well-defined, not
lhe amplitudes.
Factorizing the infinite phase, we can cnmplete the calculatíon of the scattering am-
plitude. lntegrating nver the vector h we obtain:

(
p+ ~ ITip- ~) "='
2 2
Í7f(d-l)/>
(2•l
E. exp [-i~ In
m p
(-t't" p
4 2 )]
m2
x r[! (d 1) e] (~)!e+tJ-dJ;•J
r (B) 4 {2.96)
with:
O= -íomfp. (2.97)
In three dimension~ the expression (2.96) is identical to the exnct result. It contains also,
lor a < O, the oorrect Coulomb boundstate energies En which are given by the poles of
the scattering amp]it ude:
d-1
(} = -2- +n => En
E.
2m (2.98}

'l'he eikonal approximation has a relativistic generalization, and again yields remarkably
111teresting expressions for the energy of bound states in quantum elect.rodynamics. lt
corresponds there to an approxima.te summation of ladder and cr<>SS<!d ladder Feynman
<liagrams.
!6 Euclidean Path lntegrols in Quantum Mechamcs 2.7

Bibliographical Notes
The formal analogy between time and inverse temperature has been fin•t noticed in
F. Bloch, ZeiL f. Phys. 74 (1932) 295.
For the formulation of qua.ntum mechanics in t.erms of path integrais, bask pnpers are
P.A.M. Dírac, Physik. Z. Sawjetunion 3 (1933) 64; R.P. Feynman, Reu. Mod. Phys.
367 (194.8) 20; Phys. Rev. 80 (1950) 440.
An expanded version is found in
R.P. Feynman and A.R. Hibbs, Quantum Mecha.nics cmd Path Integrais (McGraw-Hil\,
New York 1965).
The perturbative equivalence of the operator formalism and gaussian integration is shown
in
N.N. Bogoliubov and D.V. Shirkov, Jntrodudion to ThMry of Quantized Fields (lnter-
science, New York, London, 1959) ch. 7.
The approach through functional derívatives and the idea of réducing any harniltonian
problem to one in an ext-ernai source is due to J. Schwínger. See for example
J. Schwingér, Phys. Rev. 82 (1951) 914; Proc. Natl. Arod. Sei. USA 37 (1951) 452.
For a disa~ssion of the use of path integrais in statístical mechanics see for example
J. Ginibre in Statiotirol Mechanics and Qua.ntum Fíeld Them1J, Les Houches 1970, C.
De Witt and R. Stora eds. (Gordon and Breach, New York 1971).
For the Kii.llen-Lehma.nn representation see
G. Kii.!len, Quantum Electrodynamics (Springer-Verlag, Berlin 1972).
Severa! sections of this chapter have been inspire<! by
E.S. Abers a.nd B.W. Lee, Phys. Rep. 9C (1973) 1; L.D. Faddeev in Methods in Field
Thwry, Les Houcheti 1975, R. Balian and J. Zinn-Justin eds. {North-Holland, Amster-
dam 1976).
Tbe argument of Subseetion 2.6.2 originates from
W. Tobocman, Nuovo Cimento 3 (1956) 1213.
The calculation of the S-matrix in tenns of functional integrais wíth presc::ribed field
asymptotic behaviour was proposed in
R.P. Feynman, Acta Phys. Polon. 24 (1963) 697; Magic without Magic: J.A. Wheeler,
J. Klauder ed. (FTeeman, San Francisco 1972).
For a dis<:ussion of the eikonal approximation in field theory see for example
H.D.I. Abarbanel, Cargese Lectures in Physics vo!. 5, D. Bessis ed. (Gordon and
Breach, New York 1972).
The interaction representation is discussed in:
F.J. Dyson, Phys. Rev. 75 (1949) 486.

E:xercises
E:xercí.$e l!.I
2.1.1. Combíning equations (2.42,2.43) and (2.50), calculat.e the energy leve]!; En of
the hamiltonian H:
H ~P2 + !<J2
+ >.q4,
to first order in >., from the expression of the partition function Z:
00

z "'tre-PH L:e·I1Eft.
n;;;;;::O
2.7 Eudidean Path Jntegml.s iTl Q!Lantum Mechanics 37

2.1.2. Eva!uate the contribution of order >. 2 to En·

2.1.3. Calculate t.o order À the energy leveis for n. potential of the form:

for N = 6,8.

Exerci.se 2.2
Compare the exact expression (2.38) of the evolution operator of the harmonic oscillator
(set fOr conveníence h= m = w = ] ) with the appmximate expression obtained by using
equation (2.14) and calculnting explidtly tbe integrais in (2.15). The analysís is simplest
when V(q) in equa.tíon (2.14) is replaced by a.n equh-alent symmetríc function of q and
q', for example V !(q + q')/2]. Discuss the accura.cy for different equivalent fundions of
q and q' .

.&en:íse 2. Y
Use expressíon (2.84) to obtain a semidassical apprmtímatlon for the S-matrix of the
one dimensional potential V(q) = À{ c<rsl} q. Compare with exact results (to be found
for example in R.G. Newton, Scattering Theory af Waves and Particle• {McGraw-Hill,
New York 1966)) for the forward a.nd backward S-rna.trix elements, S+ a.nd S_ at energy
E= k2 /2:
r(d+ n)r(n+ 1- d) S = sínordr(d+o}r(a+l d)
8+ = r(l + a)r(a) ' - sin'll"a r(l + o)r(o)

wlth
d= w v"l-8À) a=-ik, k>O.
38 Euclidean Path Integruls m Quantum Meclianics A2.J

APPENDIX 2
PERTURBATION THEORY. THE TWO-POINT FUNCTION

A2.1 Perturbatlon Theory in the Operator Formalislll

For completeness and to illustrate the differences with the patb integral formulation,
let us here recai! the basiE of perturbation theory in the operator formalism of quantum
mechanlcs. Tó calculate the S-matríx for example, we need an expression for the opere.tor
n(t) (see equa.tion (2.69)):
n(t) = eiHot e-rHt, (A2.1)
in wbich H 0 ís the unperturbed harnlltonian and:
V =H Ho
the perturbation.
The operator íl(t) satislies the eque.tion:
fl(t) = -iV.(t)íl(!), (A2.2)
wíth the boundary conditíon:
íl(O) 1.
The operator Ví (t) is the perturbation in the interaction representation:
Vi(t) = e>Hot V e-iH0 t. (A2.3)
lt ís easy to verify that the solution of equation (A2.2) ca.n be formally written:

!1(1) "" f: (-it f

n=O
dt, dt2 ... dt,. vl (tn) \-) (t,_,) ... Ví (t,) \-)(h)' (A2.4)

the domain of integration in the r.h.s. being:

o :S t, $ t, :S .•. :::; t,. $ t.
lf we now introduce the time-ordered product, as defined in Sect.íon 2.5, of the product
of perturbation terms which o.ppears in the l.b.s., the product becomes symmetric in the
time arguments. \Ve can therefore symmetrize the domain of íntegration provided we
divide by a counting factor n!:

(A2.5)

This e:xpression can be formally rewritten:

íl(t) T [exp (-i 1' dt' Vt (!'))] (A2.6)

We can in particular apply these results to a harniltonian H perturbed by a term linear

in q. We can also write a path integral representatíon for the corresponding pll.rtition
function. Comparing the expansíon of the path ínt.egrel in powers of the perturbation
with expressíon ( A2.5), we recover the relation between correlatíon functions and T-
products establíshed in Sectíon 2.5,
A2.2 Euclidean Path lntegrols in Quantum Mechanic8 39

A2.2 The Two-Point Function: Spectral Repr<>sentation

The 2-point correla.tion function pla)'lS a special role in quantum field theory and st.a-
tistical mechanics. Let us here derive an important property. 1Ve assume t.hat the
hamiltonian H is hermitía.n, bounded from below, and, to símplify the notation, has a
discrete spoctrum. In the basis in which H is diagonal the 2-point correlation function:
z!2l(t) (q(O)q(t)) = iOI q .,-ltiH q .,ltiH lO),
can then be written:
z12l(t) =L l(o 141 n)l2 e-( •• -.,)1,,. (A2.7)
n~O

Tbe quantities In) and En are respectively the eigenfunctions and eigenvalues of H. The
important observation is that, as a consequence of the hermitícity of H. the eigenvalues
are real, and the sum of exponentlals in the r.h.s. has po.sítíve coefficients. The Fourier
transform zm (w) of the 2-point function has the representation:

z(2)(w) = /dtz12l(t) e""t = 211' I(O 1<11 0)12 tl(w) + 2 L (ón- <o) !(O !til nJ12 (A2.8)
n>O !w2
+(e,.- eo) ]
Two properties of the Fourier transform of the 2-point function follow: except for a
possible distributive pa.rt a.t w = O, it is an analytic function of w2 with poles only on the
r!al negativa axls. ln addition the residues o[ the poles are ali positive; it follows that
zl~l(w) cannot decrealle faster than I/w2 for w2 large. Actually we ca.n be more precise.
Let us calcula.te the the limit when t - 0+ of the derivative of Z( 2l(t)

lím _dd Z( 2 l(t) =: (OI <i iri, H] lO}

t-0+ t
Since the l.h.s. is real we can replace tbe operator in the r.h.s. by its hermitian pa.rt:

lim _dd z( 2 l(t)

t-O+ t
= ~ (OI]oí lq, H]] 10).
For a hamiltonian quadratic in the momentum variable H = 2
l_,.p2 +O(p), we can evalua.te
the oommutators and obtain

I<J!<i,HJl = m
We can a.pply this result to the expression (A2. 7) and find

f-=
m
L n?_O
I(O l<ll n)l 2 (<'n- <o)-

We conclude

'J'hís useful result is not surprisíng. The behaviour for w large is related to short time
~volution, and we he.ve seen that the most singular part of the evolution operator is then
related to the free hamiltonian.
Finally when the spectrum of H has a continuous part, the sum in (A;?.S) is replaced
h~· an integral, the poles are replaced by a cut with a positive discontinuity and the
· condusions are the same. The rela.tivistic generalization of representa.tion ( A2.8) is called
thc Kíillen-Lehmann representation.
PATH INTEGRALS IN QUANTUM MECHANICS:
GENERALIZATIONS

ln Chapter 2 we hAve constructed a path integral representa.tion of the operator e-~H ín

the case of hamiltonians of the form p2 + V(q). In this chapter we extend the construc-
tion to hamiltonians which are more general functions of phase sp""e variables. We thus
lirst define general path integrais on phase space. An important example is provided by
hamiltonians general quadratic functions of rnomentum variables, a situation we analyze
more thoroughly. Suc:h hamiltonians arise in the qua.ntization o( the motion on Rieman-
nian manifolds. We illustrate the Ana.lysis by the quantizàtlon of the free motion on the
sphere SN-I·
A parametrization of phase space in terms of complex variables leads to the holo-
morphic representation of quantum mechanics. The constrnction of the corresponding
path integral provídes a useful introduction to the construction of the path integral with
fermionic degrees of freedom a.nd to the functional representation of the S-matrix in
qua.ntum field theory whic:h will be described in Chapter 7.
In the appendix we also discuss path integra.! quantization of syst.ems for which an ac-
tion genera.tiug tbe classical equation of motion ca.nnot be globally defined. Tbis problem
arises when phase sp,.,e, as in the quantization of spin degrees of freedom, or coordi-
nate spa.ce, as in the example o( tbe magnetic monopole, have non-trivial topological
properties.

3.1 General Hamlltonians: Phase Space Path Integral

We now consíder a general local quantum hamíltonian obtaí.ned by some quántization rule
from a cla.ssical bamiltonían H(p, q,t). lf we know the quautum hamiltonian explicitly
and if it is no more than quadratic in p, we can generalize tbe strategy of Section 2.2 and
calculate the imaginary time evolution operator, i.e. the quantum statístical opera.tor, for
small time intervals, by solving the Sc:hrõdinger equa.tioo. A ctually often the problem has
a dillerent formulation: we are given a classical hamiltonian whic:h we want to qua.ntize
while preserving it.s symmetries. Let us in a.!l ea.ses establish a formal path integral
representa.tion of the evolution operator.
We again start from expression (2.8}:

(q"IU(t",t')lq') =!'fi dq~ ll

k=l k=l
(q• lU (t•,t•- 1)lqk- 1), (3.1)

with the conventions

t'' :..t' +ne,
9n = q".
Taking the laxge n, small <: limit, we again reduce the calculation of the evolution operator
to the ev&.lua.tion of the matrix elements (q iU (t + t:, t)l q') in tbe small time interval
limlt. In tlús limit, as a consequence of the locality of the hamiltonían, only the matrix
elements with lq- q'l small will contribute signifi.cantly to expression {3.1). To calculate
the evolutíon opera.tor U (t, t - €) we soive formally the Schrõdínger equation:
u (t,t- <}=I- ~H lP,q,t) +o(~), (3.2)
3.1 Path lntegrals in Quantum Mechanics: Genemlizations 41

in whích fi is one quantum hamíltonían which ha.s H as the classícallimit. We then takt>
the ma.trb: elements of thís equation. The matrix elements of the identity yield a Dirac
6-funcüon for which we use the Fourier representation:

ó (q _ q') =f~ e'P(o-o')/A.

2rrh

S!nce the hamíltonian H is local, we can replace q by its average value q•• which tends
towards the v-alue q in tbe limit q = q'. Then ÍI becomes a function of only one operator
p and no commutation problems remain. The quanti:tat!an problem is now the following:
the knowledge of the classical hamiltonian H does not determine the value of q•• for
q' -1- q a.nd instead yields only H (fi, q••. = q) at li O. To proceed further, a choice of
quantization is thus required. We here choose, as a.n example, the Wigner quantization
rule, which ha.s at least the property to associate with a real classícal hamiltonian a
hermitia.n quantum hamlltonian:

q•• =! (q + q') "'* (q ikl q') = j 2':;'1'1 e•P(q-o'}/ft H (p, ~ (q + q'), t). (3.3)

Only in the case of hamí)tonia.ns o f the form p 2 +V (q) will the final result be independent
of thi• choice of quantization, as has been dis<:ussed in Section 2.2. We sha.ll also find
out !ater that the situa.tion in field tbeory is more favoumble, in a sense, since for local
hamiltonians commuta.tors of conjuga.te va.riables 3re divergent (equatíon (5.5)} and have
to be removed by renormalization. Dimensional regula.rizatíon wíll be spedally useful in
tbis respect (see Section 9.3).
Exponentlating, we can rewrite the matrb: elements of equation (3.2} as:

Equation (3.1) then takes the form:

(q" iU (t", t')l q') = ~~m00 f TI d~:ct;'k

k-1
li (q., - q") exp [-S, (p, q) Jn], (3.5)

in which S, (p, q) is gíven by:

(3.6)
k=l

1f we introduce a trajectory in phase space {p(t),q{t)} which interpola.tes between the

discrete set of times:

we ca.n take the formal continuum time límit e,; _, O, n - oo and write a path integral ín
phase space:

q(t")-q"
\q" lU (t", t')l q')
J
q(t•)-q'
[dp(t)dq(t)J exp [-S(p, q) /li], (3.7)
42 Path Jntegmls in Quanlum Mechanics: Genemlization.s 3.1

in which the measure in phase space is normalize<! with respect to 21rli a.nd S (p, q) is the
imaginary time classícal action wrítten in the hamiltonian formulation;

S = 1'"
I'
dt [-ip(t)q(t) +H (p(t),q(t), t)J. (3.8)

Note that in this formallímit ali traces of the specíal choice of quantization (3.3) have
disappeared.
Expression {3.7) ls specially aesthetical since ít ínvolvcs only the invariant Liouville
measure on pha.s<l space and the classica.l action. In particular it ls formally invariant
under canonícal transformations: transformations in phase spa.ce preserving the Poisson
brackets

with
(p,q) ,_{F (p,q) ,Q (p,q))'

{p,q} {P,Q).
i
The .!pllce o/ integmtíon. Problems Mise when one tries to characterize in general the
space oftrajectories in pha.se spacewhich oontribute to the path integral (3.7). The term
II
which connects different time steps in expression (3.6) is now ipk (qk- Qk-1)· lt leads
to oscillations in the integral (3.5) which suppress trajectories not regular enough. The
typical scale of the difference (qk - qk- 1 ) for oontributing trajectories is given by the
typical values of Pk. For e:xample if as in Section 2.2 the hamiltonia.n is quadmtic in p,
the relevant values of P• in integral (3.5) are of order 1/ ,fi and one therefore finds the.t
(q• q._t) is of order ,;E:, a result which is consistent with the analysis of Section 2.2.
In the same way we can rearrange expression (3.6) to transform the term ipk (qt- qt- 1 )
into ÍQt (pk - Pt- 1) ( "integratlon by parts~). To find the regularity condítions imposed to
I
iI
the function p (t) we have to know the relevant values of Qt in integral (3.5). Consldering
again the example of a hamiltonian r?+ V(q) we see that íf for ex&mple V (q) groW'S for q
la:rge like q 2 ~'~ tbe televant values of qk are such that ~q 2 ~'~ is of arder 1, and therefore the
düference (pt Pk-1) should be of order E 112N. For a gener!U hamiltonia.n the d.iscUBSion
obviously becomes rather involved.
Remar/c. The canonical invariance ca.n be true only for a very restricted class of
transformations. We will show el.sewhere (Appendix A3B.l) that in the case of a one
dimensional, one degree of freedom hamiltonian H one can always find a canonical trans-
formation which maps H onto a. free hamiltonian

One would then naively conclude that semiclassical approximation.s are always exact. lt
is very easy to produce counter-examples. The discrete form (3.6) shows one origin of
the difficulty. A variable Pk is associated to a pair (qk, qk-1 ).
The conclusion is that the path integral in phase space is more difficult to ha.ndle in
practice than the integral in coordinate representation defined previously ar1d thus has
found fewer applications up to now. For any non-star~dard example one has to return to
the discret.e form (3.5) and make a specíal a.nalysis.
Again the extension to severa! degrees of freedom is straightforward. In terms of
the classical action and the Liouville measure, the path integral representation of the
evolution operator h as the sa.me form as in equation (3. 7).
3.2 Path Integro!. in Quantum Mech'!nics: Cenerolizations 43

Real time formulation. In the real títne fonnulation, which leads to the eYolution
operator (e-•<H/n when t.be hatniltoniáll is time independent) the representation (3.7)
ha.s to be replaced by

(q"IU(t",t')lq'} =f (dp(t)dq(t)]exp [*A(p,q)]. (3.9)

The <mclidea.n action S in tbe path integral ha.s been replaced by A(p, q), the clas5ical
action in the ha.miltonian formalism:

A (p, q) = l' [p(t}q(t} H (p, q, t)) dt. (3.10)

E""n in this general situatíon the quantum mechanical evolution is obtained by summing
over ali paths weighted with the complex weight é-"1 ". Therefore paths dose to extrema
of the action, i.e. clas5ícal paths, still give the largest contributiorn; to the path integral.

3.2 Hamiltonisns Qusdratlc in Momentum Varisbles

We have explained the difficulties one enoounters when one tries to define a path integral
in pha.se space. To show that ""pression (3.8) has, however, at least .orne heuristic value,
let us examine the ca:>e of general hamlltonians quadratic in momenta.
The consistency. Let us first verify that in the case of hamiltonians

rP
H""' 2m+ V(q),

we recover previous results after integraticn over p(t) in expression (3.7).

The classical action is:

S L,..
dt
[
~ipq + :m +
:1 ]
V(q) . (3.11)

1n expression (3.7) the integral over the rnomentum varia.bles p(t) ís then gaussian. Fol-
lowing the strategy explained in Chapter 1, we change variables

p(t) = im;j(t) + r(t). (3.12)

The action becomes:

(3.13)

The path integral thus factorizes ínto an integral over r(t):

N (t', t") f idr(t)J exp (-~{'r;~) dt] , (3.14)

44 Path Integreis in Quantum Mechanics: Genemlízations

which does not depend on the potentíal V ( q) and only yields a normalization factor }!
functíon of t' and t", and a.n integral over q{t):

•(•")~.-·
(q"[U(t".t')lq')=/1!1 .
q(t'\~q·
[dq(t)}exp
[ 1
--lij t"
,.
]
dt(im•l+V(g)) , (3.15)

The factor N ca.n be calculated from expression (~.5);

N= (~)"'~ (3.16)
2:trhé

We have therefore verified explicitly that expression (3.7) ís consistent with expression
(2.18).
The dectromagnetic hamiltvnian. Let us consider the electromagnetk hamiltonia.n:

1 2
H= m[p+eA(q)J +V(q). (3.17)
2
Tb elíminate the terms linear in p in the action we now change variables

p(t) = imq(t)- eA (q(t)) + r(t). (3.18)

After integration over r(t), we obtain an integral over the path q(t) with the actíon:

S{q) = J.'" dt [~mq2 +ieA (q) · q- V (q)],

t'
(3.19)

in which we reoognize the eu clidean clllSSícal act.íon. Let us verify that the pa.th integral
is consistent with ga.uge inVàriance: If we add a gradient term to A(q)

A(q}- A(q) + VA(q), (3.20)

the variation of the lagrangian is a tóta.l derivative and the variation of the action 6S is:

óS = ie!A(q") A(q')]. (3.21)

The matriX elements of the evolution operator therefore are multiplied by a phase fa.ctor
exp [ -ie(J\ (q")- A (q'})/li], as they should.
The quantum hamiltonian fl, which corresponds to the classical hamiltonian (3.17),
is completely chara.cterize<l by the condition of hermitícity:

(3.22)

A direct calculatíon of ·the evolution operator for small time intervals, startin,g directly
from the Schrõdinger equation and usíng hamiltonian (3.22), again yields expression
{3.] 9). However the path integral is consistent with the (hermitian) choice of quantizatíon
( 3.22) for H only i f we respect the symmetry t --+ -t, S - S• in its calculation, Thís
is dil'ectly related to the li{O) 1/2 assignment (8(!) being the step function). We shall
agaín meet this ptoblem in Sections 3.5,3.6 and in Chapters 4 and 38.
3.2 Path lntegrals in Quantum Mecl1anics: Genemlizabons 45
The general =•e. A general hamíltonian quadratic in p ean be derived from a general
lagra.ngian quadratic in the velodties. Because, in all examples we shall meet, the quan-
tízation problem "'ill initially be formulated in ternlll of a lagrangian, let us now assume
t.hat we want to quantize the real time lagrnngian L:(q,q)

L:(q, q) ~q''gap(q)q 19 - tj"h,.(q)- v(q), (3.23)

where 9atl{q) ma positive matrix.

The corresponding classical hamiltonian is obtained by Legendre transfonnation. The
conjugate momenta are:

We then obtain the ha.miltonian

H(p, q) = Pcr<i" -l(q,q)

= ~(Per+ ha(q)) g"ll(q) (pe + hp(q)) + v(q), (3.24)

where we ha.ve used the traditional notation g"ll for the matrix inverse of g,fJ:

because most interesting exarnples correspond to qua.ntum mechanics on Ríemannian

manifolds, as we shall brielly discuss in Section 4. 7 (for details see Chapter 22). The
tensor g,._s(q) is then the metric tensor. Field theoretícal generalizations will be studied
in Chapters 14, 15.
Aga.in the integration over p(t) il; gaussian and can be performed. However a dif!iculty
avpears wjth the evaluation of the determinant resulting from the ga.ussian integration
and we therefore perform the p integratlon on the discretizad form (3.5) or aetually (3.4).
Let us rewrlte the expression (3.4) for hamiltonian (3.24):

(qjU(t,t -E)[q') =f 2~/iexp{i u(q- q')"p.. -! (Pa + hcr (-'I:!f))

X g"fi ( ~) (Pfi + hll ( .<I:!f)) v ( ~) D· (3.25)

The integration yields:

{qJU {t,t- E)[q'} = [211'/ie detg (~)r· exp [-S (q,q'; t) /li],

with:

s;~ ~ [(q"- q"')/E] 9<>/l ( ~) [(q~- q'~l/•]

+i [ (q"- q'")je} h., (~)+v (-'I:!f), (3.26)

and we have denote by g the matrix with matrix elements 9,.p.

4tl Path. Integmls in Quantum Mechanir:s: Generolizat•ons 3.2

Returning to expression (3.5) and taking the formal continuum limit we verify that the
exponential factor aga.in yields the euclidean action S. integral o( the classícnllagrangian
(3.23) in which we have made the time imagi:nary:

S(q) 1'"
t'
dt [~<Í 0 9oii(q)qp +íq"h., (q) +v(q)]. (3.27)

Thls is not surprising since, in order to integrate over p., we have solved the dassical
equation of motíon for p.,.
However, ín contrast with the previous two cBOOS, the integration genera.tes a non-trivial
q(t) dependent normalization factor N(q):

(q"JU(t"t'))q') j !dq(t))N[q)exp[-S(q)/li], (3.28}

with:
N(q) <:::0 (2,.~)-n/l exp [~ tlndetg ( ~)], (3.29)
·~1

and thus yields a divergent quantum mechanical correction to the classical action (it has
no factor l/1i):

N(q) -exp [_.!_

2e
1'"
t'
lndetg!q(t)] dt), (3.30)

or equivalently using the identity (1.33);

lndetg = trlng,

N(q)- eXp [~ { ' tdng!q(t)]dt]. (3.31)

A formal calcu)ation, starting from expression (3.7), yields a similar resu)t with 1/t
rep!B.ced by 6(0) (6 being the Dirac ~-function). Thls difficulty is directly related to the
problem ai ordering of operatorn. lf one performs a small 1i (semiclassical) exp!Uillion of
the path integral, one finds a divergent quantum rorrection (see Chapters 14,15). This
divergence is canceled by the leadíng contribution ooming from (3.31 ). However the
temaining llnite part is ill-defined in the formal continuous time limit. lt is necessary
to use the discretized form (3.5), which reflects a choice of quantization, to calculate it.
Another direct way of understanding this ambiguity isto notice that, since in expression
(3.30) the difference ]g ~ q1 ] is generica!ly of order ,fi, a repla.rement of g ((q + q') /2) by
any symmetric function of q and q' whích has the same q q' limit changes in general
this quantity at order .:. The modification of N(q) theu generates a finite quantum
mechanical correction to tbe classical action, typkal of a commutation of momentum
and position operators. Some more details about this problem can be found in Appendix
A38.2.
Let us final..ly note that this problem worsens i[ the classical h!>miltonian is a polynomial
of higher degree in p.
Remark. In the case in which 9a.P is the metric tensor on a RlemanDian manifold, the
fa.ctor (3.29) formo.lly reconstructs the covariant measure on the manifold {see Section
22.5).
3.3 Path Integrals m Quantum MechrmiCJJ: Genemlizations 47

3.3 The Spectrum of the 0(2) Symmetric Rlgid Rotator

To illustrate the discussion of general hamiltonians quadratic in momentum va.ríables we

uow calcula.te the spectrum of the O(N) rigid rotator from the path integral represen-
tation (thís model is also the one-dimensiooal O(N) non-linear a model, see Cbapter
14). We examine first the N = 2 case which is simpler and can be treated exactly. The
0(2) rotator provides actually an e:xample o{ the peculia:rltíes of the path integral when
position space has non-trivial topological properties. Iodred in angular coordinates the
hamiltonian of the 0(2) rotator is very simple:

(3.32)

and would be a free hamiltonian íf 8 were not an angular variable. The corresponding
patb integral for the matrix elements of the operator e-I>H is:

(3.33)

Let us show how the cyclic character of the variable modifies the evaluation of the path
integral. ~ usual we first solve the classical equation of motion. However, since 0' and
8" are a.ngles, w€ have to consider ali trajectoríes which go from 8' to ()'' mod (2or ). We
then shift O(t} by tbe solution:

t
ll(t) = e'+ {3 (li"- 9' + 2..-n) + u(t)' wíthn E Z. (3.34)

The path integral (3.33) then becomes a sum of cootributions:

(3.35)

The normaliza.tion N (/3) is given by a path íntey;Yal wbich is independent of lJ', ()'' and
n. Since thE> intey;Yation over u(t) sums fluctuatiOJI5 around the clsssical trajectory, it is
expected that the angular character of u(t) is irrelevant and therefore:

/11({3) = v'2i/i3. (3.36)

The E>Xpression (3.35) can be rewritten using Poisson's formula: Being a periodic function
af ()'1 ~ 8 1, it ba.s a Fourier series expansion;

+o<>
(8"\e-.lill\8') = Í:: eíi(O"-B')e-il~•, (3.37)
1=-00

whose coeffidents are giVI!n by:

(3.38)
48 Path Integmls in Quantum Mechanic.s: Generolúations 3.4

lnverting sum and integratíon, and changíng variabies, {I+ 2mr ,_. 8, one then finds:

exp(-PEI) = ~ t: i- (i!O+ ~~)]

d6 exp (3.39)

The gaussian int.egra.tion finally yields:

(3.40)

which is indeed the exact result. lt has been possible to perform an exact calculation
because the O (2) group is abelia.n. The discussion of the general O(N) group ís more
involved as we shall see now.

3.4 The Spectrum of the O(N) Symmetric Rigid Rotator

The harniltonian of the O(N) rigid rotator has the form:

(3.41)

in which the vector L, the angular mornentum, representa the set of generators of the
Lie a.lgebra of the O(N) group. To write H explicitly we have to choo.<;e a particulEil
parameb-izatíon of the sphere. Then the harníltonian has the form:

(3.42)

in whích 9>i (q) is the metrk tensor on the sphere (see also Sect;ion 4.7): For example
if we parametríze loca!ly the sphere by a vector in RN of components ( q,yl - ql) the
metric tensor rea.ds:
(3.43}
Acrording to the discussion of Section 3.2, the corresponding path integral representat.ion
of e-IJH then is:

(3.44)

in which g ( q) is the determinant of the matrix Yií. The contribution to the measure
coming from the gaussian integration over the momenta p; has forrnally generated the
invariant measure on the sphere. We can then rewrite the path integral {3.44) in terrns
of a vector r in R"" of unit length 1:

(3.45)

Let us cal! li the angle between r' and r":

cosli r'·r", O~II~'IT- (3.46)

3.4 Path Integrols in Quantum Mechanics: Genemlizatioll$ 49

We now íntroduce a matrix R(t) whích acts on r(t) and rotates r' onto r" in the plane
(r', r") in a time /3. lts restríction w the twa-dímensional (r', r'') space has the form:

cos (flt/(J) sin(flt//3)]

[ - sin(flt/ /3) cos(flt//3) · (3.47)

lt is the identity in the subspace orthogonal to the (r', r") plane.

We then change variables, setting:

r(t) = R(t)p(t). (3.48)

Let us call u and 11 the two components o( p in the (r', r") plane, u being the compenent
along r' and ~ the component in the orthogonal subspace. With this notation we find:

1
p(f3)~··
(r" le- 8 HI r')= [dp(t)ó (1 - p 2 (t))) exp [-S (p)), (3.49)
p(O)=r'

with:

(3.50)

We then use:
(3.51)
to '""'Tite S(p):

S( P ) = -192
-
2(3
+ -2llafld t
0
(·2
PT - -fP Pi
{F
, ·2 ·2
+ ti + 1! + 29{.
-(J VU - ti. Vl) . (3.52)

In contrast with the abelian case where the calculation was exact, we can here petform
only a small f3 expa.nsion. We can therefore neglect oontributions exponentially small in
13-1. At leading order only smell fluctuations around the classical solution are relevant.
We thus eliminate the variable u from the action {3.52) by using equation (3.51 );

(3.53)

and expa.nd the action in powers of PT and v. The leading order result is given by the
gaussian approxímation and requires only the quadratic term.s. In particular the term
linear in 8 in (3.52) is irrelevant at thisorder, the integral over v(t} is thus independent of6
and can be absorbed ínto the normalization. The componente of PT become independent
variables, and the integral over PT is the (N - 2)th power of the integral over one
component. Since each component satislres the boundary condition

p;(O) = p;(/3} O,

we can expand the functions p,(t) on the basis:

Pi(tJ = L: Pin sín (mrt/ (J) (3.54)

n>O
50 Path Jntegrols in Quantum Mecho:nics: Generoiizations 3.4

The ga.u.ssian integral over the va.ria.bles p,., then yields:

{3.55)

The normaliza.tion constant K is índepend~nt of 8. The infinite product can be calculated:

(3.56)

and therefore:
(3.57)

To extract the eigenvalues oí H, we have to project this expressíon over the orthogonal
polynomials Pf (cos9) associa.ted with the O(N) group:

i~ dB sinN- 2 8 .Ff (cosO)~~ (c.,.&)= 1rÓf1·. (3.58)

which are the Gegenbauer polynomials ct- >1

2 2
. For {3 sma.JJ we need only the ~ma.ll e
expansion of these polynomials:

P,N (cosBJ = ~N (1) (~-~~t: ~)2le2 + o(o•)) (3.59)

If we assume tha.t to eacll value of l corresponds only one eigenvalue E! of H then we

can write:

e-IJI>, = K([j) fo" d9 P,N (cosO) (8 sin 9)(N- li 2 2 e-•'lí 2 1ll

= e-llEo (I- tl(i + N- 2){J +O (/3 2)), (3.60)

and therefore:
(3.61)
Since E 1 is independent of (3 we can deduce from this calcuhttion the exact result, up to
an additive constant Eo. Therefore if we calcula.te the correction of order of (3 we shall
find a w.nlshing result.
Concerning this calculation a comment is now in order: we have explained in Sectíon
3.2 that the path integrais (3.44.,3.45) are ill-defined because the measure gíves formally
divergent contríbutions. We have stated that these divergences are ca.ncélled by diver-
gences in perturbation theory. However, as a consequence, the resulting expressions are
a.mbiguous and these a.mbiguities reftect the problem of operator ordering in the quan-
tization of a classica.J hamiltonian. Still we have here obtained some non-trjvial result.s.
The rellSl>n is that at every stage of the calculation we have assumed O( N) ínvariance.
This chooses implicitly among a.JJ possible disct'etízat.ions of the path integra.! a subdass
which corresponds to a O(N) invariant quantised hamiltonian. We shall see !ater, when
díscussing the non-linear a model, that such a hamiltonia.n is fully determíned up to an
additive constam. The a.mbiguities of the calculation are thus entirely contained in Eo·
3.5 Path. Integrais in Quantum Mechanics: Genemlizations 51

3.5 Path Integral and Holomorphic Representation

We now derive a.nother path integral representation of the evolutíon operator. This rep-
resentation leads to interesting expressions for pertrubed hMmonic oscíllatots, provides
a useful introduction to the fermion path integral, a.nd in quantum lield theory to the
construction of the S-matrix and the study of the non-relativistic limit in Chapter 7.
The idea is to BSSOciate clBSsical {compiex) variables !, z to creation and annihilation
operators a t, a rather than to position and momentum operators.

3.5.1 The Jwlomorphic representation

To motivate the construction we lirst consid<>r the ha.nnonic oscillator

(3.62)

We then define the annihilation and creation operators a, at as usual by

p- iwq = -i../2,;;a, (3.63)

In terlll5 of a, a t the hamiltonian ta.kes the standard form

with w > O. We omit in what follOWli the constant energy shift !w.
We introduce a complex variable .ii and represent at,a by the operators z and 8f8Z:
respectively, acting on functions of z, which have tbe same commutatíón relations. The
hamiltonian H 0 is then represented by

_8
H 0 =wz z. (3.64)
0
ll is clear that this complex parametrization of phase space is particularly well adapted
to the descdption of the harmonic osclllator. In particular the eigenfunctions are simply

~ 8 -n =wnf".
wza.;; z

The action of the quantum statistica.l operator U(t} =..,-Hot on functions of z is:
1
U(t)f(z} = f(e-"' z). (3.65)

The corresponding kernel, however, hiiS the form of a 6-function

a representation which is not specially oonveníent for tbe study of perturbed harmonic
oscillators. We therefore introduce a mixed representation of operators by perlorming a
formal generalized Laplace transformation on the right:

O(z,z) = (zl Olz} = j dz' (zl Olz') e•'•. (3.66)

52 Path lntegmls in Quantum Mechamcs: Genero.lizat.ons 3,5

A somewhat similar representation in the case of phase space variables is a mi.xed com·
dinate momentum representation (q IOI p) which is obtained by a Fourier transformatíon
on the right argument:
(ql O IP) =f dq' e''~"~' /h (ql O iq').

With the definition (3.ô6) the identity is represented by {ziz} e'' and the matrix
elements of the hamiltonian (3.64) and the operata U(l) a.re respectívely
(zl H o lz) = wize'', (21 U(t) !z) e'"·~· 13.67}
More generally to the normal-ordered product of creation and annihílatíon operators
alma" is a5SOCiated the kemel

at~a"- .f" (:,J" ..-+ i"'z" eh

Let us now verlfy that the action of an operator O with kernel 0(1, z) on a vector f(f)
is given by

By systematic commuta.tions any operator ca.n be written a.s a tine& rombínation of

normal-ordered operators. lt ís thus sufficient to verify tbe identity for nonnal-ordered
operators:

f d.i'dz' e••' z"'z'" .,-f'•' /(i') =.f"

2tlf
f dt'dz' z'" e-f'•' !(i'+ f)
2tlf
= :;m f di'dz' z'" e-•'•'
2i:lf
Li .!.ztlf(l)(~)
l1 • ,
Tbe result then relies on the orthogonality oonditions

z"m
f dzdz
~e
-h fl;
:I =n. m'l,

which can be easily verified, for example by cha.nging variables, parametrizing the com-
pl<lx plane in terms of modulus and a.rgument, 2 = pe; 6 , i = pe- 16 • We then findas
requíred

f
dz'dz'
--..
2•~r
' = z~ ( -::
,---e>•' 2mz ' n e- •••• f(ii!)
âz
/(f). a)"
The ortbogone.lity relation is a. specia.l example of the scalar product defined in Section
1.9, equation (1.97):
dzdz --
(f,g) =
f
~e-u f(z) g(z).

We ha.ve heen naturaily led to the structure of Hilbert space of analytic functions.
lt follows tha.t the kernel associated with the product of the two operators is given by
d "d"''
fz ilf- {f 1021 z'') e-•"•" (z" IOt! z') (i I020d z').
Finally a few remarks:
2
(3.68)

(i) Th an operatot O "''hich ba.s the matrix elements Omn in the harmonic oscillator
ba.sis is 8550Ciated the kernel Lm,n O..,n(i"' ;Jm!J(z"f;Ín!).
(ii) Hermitian conjugation is replaced by formal complex conjugation
o,_,oi '* O(z,z)>-+O(z,z).
Clearly witb this definition H 0 is hermitian.
3.5 Path Integrais in Quantum Mechanics: Generoluations 53

S. S.!! Quantum statistical opemtor: path integral representation

We now use this formalism to establish a pnth integral representation of th" qua.ntum
statistical operato~ first in the case of the harmonic oscillator and then for more general
hamiltoníans. We expect the path integral to be rela.ted to the form (3.7) by a sim ple
change of variables of the form (3.63) (quantum operaton; belng replaced by classical
variables) but th" bounda.ry conditions and boundary terms requlli! a. more detailed
analysís.
We fin;t súlve Schrõdinger's equation for t =all:

(z IU(t)l z') = [1- twi :I+ O (t2 )] e"' = e"'(l-"") +0 (t2). (3.69)

We then use the group property in the fonn (3.68) to writ.e the evolution opera.tor at
finite time:
(3.70)

with:

Z..Z..-! + W~ t ZkZk-I] , (3.71)

where E (!
11
- t fn is the time step and the bound ary conditions are:
1
)
·-·
(3.72)
In the formal!Mge n limit we obtain a. path integral representa.tíon for the matrix elements
of U(t 11 ,t1 ):

U(z'',z';t",t') = j [dz(ii!i(t)] exp (-S(z(t),z(t))],

S(z(t), z{t)) = -z(t'')z(t") + 1',," dt z(t) [i:(t) + wz(t)J,

with the boundary conditions z(t") = 2", z(t') = z'. The symmetry ofthe act\on between
initial a.nd final times, which is not explicit, can be verified by a.n integra.tion by parts oi
the term zz.
This expression generalizes ímmediately to a system línearly coupled to externa] sources
](t) a.nd j(t)
H ](l)a j(t)at. (3.73)
The corresponding action S (z, z) is:

S(z(t), z(t)) = -z(t")z(t") + 1'"

t'
dt {z(t) [i(t) + wz(t))- ](t)z(t)- z(t)j(t)}. (3.74)

We leave the explicit çalculation of the pa.th integre.! as a.n exercise. For the hamlltonian
H+ !w,
H being the hamíltonian (3.73), the result is
U(z",z'; t", t') e-"'llf2 exp[z"z' e-"IJ +EG,i)J,

E(j,j) = [ " dt [z-•e-w(t"-r) j(t) + z 1 ew(t'-t) }(tJ]

+ J.',, " dtd'T j(t)O(t- r) e-"'(t-T) j('T), (3.75)

54 Path Jntegrols in Quantum Meclumics: Genernlizatioru 3.5

with fJ t" t' _

Taking the trace we also obta.in the partition function

(3.76)

(3.77)

The propagator ó(t) is the solution of the differeotial equation

Ó.(i) + wó(t) = 15(1), (3.78)

wíth periodic boundary conditions on the interval (-iJ/2, JJ/2].

For a general qu~mtum hamiltonian severo.l strategies are available, for example:
{i) We fust write the phase spa.ce path integro.l (equation (3.7)) and then cbange
variab!es
p- iwq -iv'2Wz, (3.79)
Then

S(z(t),z(t)) = -i{t")z(i") + l' dt [>(t)i(t) + H(iyÇfi(f- z), (i+ z)/v'2w)].

(3.80)
(ii) Another strategy isto write the hamlltonian in terms of creation and annihilation
operators, to normo.l-order it, i.e. to rommute ali creation operators to the left, and then
replace operators by the corresponding dassical variables.
The results are obviously different, a ref!ection again of the problem of operator or-
dering and quantisation of classical harniltonians. Moreover, we leave as an exercise to
verify that if we use the path integral to calculate perturbatively energy leveis tbe quan-
tity t.(O) will appear, as a result of self-contractions of vertices generated by the product.s
of the non-commutlng operators a,at. Since ó(O) involves 9(0) ít is clearly ill-defined,
and even simple hamiltonians of the form -Pl-+ V(q) lead to difficulties. For tbe first path
integro.l thl: solution to the problem oeems to set 9(0) = !
wbile in the second case it
seerns necessary to set B(O+) ~O (and thus 0(0-) = 1 8(0+) = 1).
Perturbed hannonic cscillator. Let us finally consider the example of an harrnonic
oscillatot to which we add a perturbatíon of the form \1 (q}.

H = ~P2 + t,,?q2 + \'I(q).

The perturbative expa.nsion in \1then can be written in a form analogous to equation
(2.50), i.e. as functional derivativas a.cting on tbe path integral representing an harrnonic
oscjllator perturbed by a source term coupled linearly to g. We note that q(l) whích is
proportiono.l to i+ z can be generated by a real source j(t). Thus the expressions (3.75)
a.nd (3.77) can be ~ymmetrized in time. Re!i<:aling a.lso j(t) ,_. j(t)f.J2W, one finds

U (z", z'; t"' t')"" exp [-\1 (.sj~t))) Uo(j; z",z'; t", t')t~o' (3.81)
3.5 Path. Integrols in Quantum MechflnÍcs: Generolizations 55

with

ln the same way equatiun (3.77) catJ be tewritten

trVG(j;t",t') = 2 sinh~w/3/ 2) exp (!f dtdrj(t)A(t r)j(r)], (3.83)

A(t) = coshw(JJ/2 ltiJ (3.84)

2wsinh(w/3/2) '

in agreement with equation (2.43}.

9.5.9 Real time evoltdum

Let us rewrite these expressions for real time evolution. The path integral representation
for the evolution operator is fotmally obta.íned by the rotation t ,_. it. Then

U(z",z';t",t') f [d•(;i:f(t)) exp[íA(z(t},z(t))],

t"
A(z(t), :o(t)) -if(t")z(t") +
J.
t•
dt [iz(t}i(t) h(i(t), z(t))],

with the boundary conditions z{t") =

z", z(t'} = z'.
From the evolution operator we can derive the corresponding matrix S-matrix. This
formalism, however, is mainly useful if the asymprotic states are eigensto.tes of the ha.r-
monic oscillator, a situation which we encounter in quantum field theory. Defining the
S-matm by expression (2.69), where H 0 is the h8I11iltonian (3.62), we find

d -trd rt d-'d I
- ) =
S( z,z .
11m
,;~___
t,íf-+- f -z- -z- -z- zE -•"•" e -•'•' e;..,"/Z exp (-zz 11 e" - " " )
2i11' 2i1r

x U(i',z';t",t')e-"" 1 '1'exp(z'ze-'"'~').

Using the equation (3.65), or integrating directly we find:

S(z,z) = lim e'"''"I2 U(.fe;..,'",ze·'"'t';t",t')e-õ...>t'l 2 (3.85}

1'------
t"-+uc

As in the coordinate representation (Section 2.6), the configurations in the path integral
which contribute to the S-matrix are for large time asymptotic to the solutions of the
classical equation of motíon. For the harmoníc oscillator H o tlris mearu:

z(t") ~ ze'"''"
t'"'-+IXl
56 Path lntegrols m Quantum Mechanics: Genernluatíons 3.6

In quantum mechanics the main applica.tion is tbe evaluation of transitions between

eígenstates of the harmoníc oscillator índuced by a tim<"dependent perturbation whicb
vanishes for large positive and negative times. As an example let us apply the result
{3.85) to the hami)tonian {3.73), where we assume that j(t) vanishes at t __, ±oo. Af1.er
a short calculation we lind

S(i,z) exp

(3.86)

The coei!icients o f the expansion of S(z, z) in powers of z and z yield the matrix element~
s ....., of the tra.nsition between the corresponding eígenetates of the harmonic oscillator,
dueto a time dependenl force coupled linearly lo the coordinate a.nd momentum:
... 1711 z'"
S(z, z) = 2:: S,.,.--,.=;
vm!vnl
r..
m,n

3.6 Fermionic Path Integral

The derivatioo of a path integral representatíon for a fermion hamíltonian follows closely
the rnethod <>f the preceding section, the main dílferençe being that complex varíables
are repl.aced by Grassmann variables.
One-fermion hamiltonian. 1be one-fermíoo problem ís a rather trivial exercise on
2 x 2 matríces, however the generalization oí the final path integral to many fermions is
straíghtforward.
Let us consider a harniltonian for one fermí<m:

(3.87)

in whích a and af are the usual a.nnihílatíon and creaticm operators for fernúons:

a2 at2 o~ (3.88)

whose a.ntícommutatíon relations are a rellection of the Paulí principie.

We want to construct a path integral representatíon for the QUantum statistícal opera-
tor U(t) =.,-tE. We have shown in Sectíon 1.7 how to define a dilferentiatíon operation
in a Grassmann algebra 2L Equations (1.52) exhlbit a rea!izatlon of the commutation
relation.s (3.88} as multiplication and differemiation operations on 21. We can therefore
represent the hamíltonian operator H as an operator a.cting on functions delined on '.11.
i.e. functions of Grassmann variables:
a
H =wc~, (3.89)

Functions oi C farm a vector space of dimension two: a çonstant ha.s fermion number
zero~ the function C has fermion nmnber one.
We here have a.n. additional rea.son to introduce a mbr.:ed representation, compared t.o
the commuting case: the matrix e)ements oi the identity are anticommuting element'
:1.{1 Path IntegraL, in Quantum M echanics: Genero/izatiun.< 5i

c i<" c). Vve therefore introduce an additional Grassma.nn varia.ble c', and perform a
<; rn.ssmann Fourier tra.nsformation on the r .h.s. va:riable ê. Then the identity is repre-
,..,c•nted b:y e~ã and the hamiltonian H hY -wêcP-t:c:. ln the mixed representation the
ho·rnel for the product U2 Ut of two apera~ars is given by:

(3.90)

Tlois result relies on the properties of the scalar product defined in Section 1.9 (equation
t I. 98)), on functions of c

(f,g) j droc e·•! f( c) g(c).

Path integml. Let us first solve Schrõdinger's equation for t small:

(3.91)

Wp t.hen use the group property in the form (3.90) to write the evolution operator at
lluite time:
n-1
{c" iU(t",t')ic )
1
J!..."!xo/ fl dckdck exp 1-S(êhck)], (3.92)
•-t
w!th:

S (C>, Ck) = [I:

k~l
Co (C. Co-t) (3.93)

(t" - t')Jn, a.nd the definitíons:

(3.94)

l11 t.he formal large n límít we obtain a. path integral teptesentation for the mattix elemem.•
nl U(t):
•(•"):'"
(c" iU (tu, !1 )1 c') = f !dc(t)dê(t)]expj-S (ê,c)], (3.95)
Jc(t')'!!!!!e'

with:
S (ê,c) -1',." dtc(t) Jé(t) + wc(t)] + ê(t") c(t 0 ) . (3.96)

Thís expression generalizes ímmedíately to a system coupled linearly to externai fennionic

"""rces fj(t) nnd 1) (t), i.e. sources which are generators of a Grassmann algebra nnd in
udtlition a.nticommute with a and a.t;

H= wata + ij(t)a- at'l(t). (3.97)

Tlw corréSponding action S (~,c) is:

S(ê,c)
1'"
t'
dt {ê(t) [é(t) + wc(t)J + fj(t)c(t) +c(t)l)(t)} +c (t") c (t"), (3.98)
 Pn.th ]ntegrols in Quantum Mechanics: Genenlliza.tions 3.6
58

where now the elements c(t), c(t), 71(!) nnd ij(t) are ali independent generators of a Grass-
mann algebra.
Calculation af tl<e gaWJsian integraL Since the integral (3.95} with action (3.98) is
gaussian, it can be calculated exactly. The saddle point equation obtained by varying
c(t) yields, takíng in to a.ccount boundary canditions:

é+wc+!)=O ....:> c(t) .,-w(t-t')c' J.'e-w(t-u)'l(u)du. {3.99)

t'

Simíla.rly tbe va.riatíon of c(t) yields:

t"
c-wc-ii O,=<> c(t)""e"'(t-t")cl'-1 e"'(t-~)ij(u)du. (:UOO)

Translating ê[t) and c(t) by the solutions of tbe saddle point equations we then obta.in:

(<!'lU (t", t')lc'} = N (t', !")exp [-So (c'', c'; i!, 11)], (3.101)

with:

So =c'' c' e-w(t"-t') -1. t"

dt lij (t) é e-w(t-t') +C"'l(t)e"'(t-t")}
+ J.
t•
t"
11 (t2- t1) ij (!2)11 (t,)e_,.,(,,-t, l dt1dt2. (3.102)

The normaliz&tion factor N can be calculated by setting ij "" 'I O and comparing to
the ditect solution of the Schrõdinger equa.tion written in the mixed repreoontation. One
findsN = 1.
As in the bosonic case, correlation functions are abta.lned by dífferentíating expressions
(3.101,3.10'1) with respect to 'I and ij.
The partitiun functWn. The partition function tr U ({jf2, -(J/2) can be obtaíned from
expression (3.101). With our conventions the traceis given by:

trU ({J(2, -(3/2) j de" de' e-<"<' (c" iU (13/2, -{J/2)1 c'), (3.103)

and tberefore:

with:
(3.105)

Note that, in contr!!St to the bosonk case (equatíons (2.43,3.78)) the iunctíon ó(t) is the
solution o( the differential equatiou:

A+ wt. = 6(t),
3.6 Palh lntegmls in Quantum Mech.onícs: Genero/i::ations 59

with antiperiodic boundary conditiom:

t. (p/2) = -t::. H3J2)
The solution with pcriodic boundary conditions appears instead in the calculation of
tr (-l)F .,-~H, in which F is the fermion number. whicb is obtuined by integrating ex-
pression (3.101) with e<"<'.
Genemwatirm. The method can be immedíately generalized to several species of
fermions and a hamiltonian H:
H= h(u!,aJ) wíth aJa}+ aja! a;a; +a;ai =O and a) a;+ a;a] = Óij, (3.106)

because relations (1.52) realize the anticommutation reiations of the operators {a;,a;}.
Let us assume that H has been normai-Qrdered, which means that, with the help of the
commutatíon relations. ali operators a! hav« been pushed at the left of ali operators a j
in ali rn(>nomíall; contributing to h(ai,a). Then eqUation (3.91) generalizes under the
form:
(3.107)

At ftnite time a path integral representation follows:

•(t")~•"
(ê" lU (t'', t')l c') "" f (dc(t)dc(t}] exp [-S (t:, c)], (3.108)
Je(t')=o'
with:
S (t:, c) - ('" dt {é;(t)é;(t)- h [e(t), -c (t)J}. (3.109)
},,
We have seen in Section 1.9 how to calculate gaussian integrais and average of poly-
nomials. We can here use the same methods to expand the path integral (3.109) in
perturbation theory. However again problems due to operator ordering arise. In partitu-
lo.r, "" in the case of the holomorphic path integral of the commuting case, perturbative
calculations involve 11(0). The conect ansatz, consistent wíth our construction, a.gain
seems to be to set é!( O+) = O.

Bibliographical Notes

Most of the references of Chapter 2 are still relevam here, The construction of path
integrall; in phase space for general hamiltonian systems can be found in
R.P. Feynman, Phys. Rev. 84 (1951) 108 (Appendix B); C. Garrod, Rev. Mod. Phys.
38 {1966) 483.
The generalization to constraíned systems is given in
L.D. Faddeev, Theor, Math., Phys, 1 (1969) 3.
An early discussion of tbe problem of quantization of velocity dependent potentials in
tbe operator formalism can be found in
T.D. L.... and C,N. Yang, Phys. Reu. 128 (1962) 88&.
For m"re details ahout the holomorphk representation see
L. D. Faddeev in MethodJ in Field Thea'1}, Les Houches 1975, R. Balian and J. Zinn-
Justin eds. (North-H"IIand, Amsterdam 1976).
The fermionlc path integral is diseussed in
F.A. Berezin, The Methad of Second Quantirotion (Academic Preas, New York 1966),
60 Fath Integrais in Quantum Mechanics: Genera1izat:íon..< 36

Exerdses
E:xercíse S.J
Calcul<>te tbe energy leveis of the O( v) lm'E.rlant hamílwnían

in which p and q are vectors ín JR", at order >.. Note that since levcls ;ue degenerate at
leadíng order, it Í5 not suffident to calculate tr e- fi H as in the " = 1 case (exercise 2.1.1).

Ex.ercise S.2
Assume for the evolution Op<lrat.or a path-íntegral representa.tion of the form

U(q",q')"" j\dq(t)N(t)Je~p Uj dtL:(q,q)].

The torm in the ~onential ís then fixed to be the lagrangian by demanding the correct
classical limit. This leaves the norrna.líz:~.tíon factor undeterrníned. Show by evaluatíng
the evolution operator for t small that N(t) ís fixed to have the furm of Section 3.2 by
unit.arity. Note tbat one h as to uo;e the property tbat for t small only short traje<:torle;
contribute to the path-integral.

Eurcise !J.:J
Dlscuss, ín the free case, the change of varíables q f(q') in the pa.th integral (2.18).
Jt ís suggesU<I to return to discrete forms {2.15,2.16) and then to compare with the
expressíons (3.25,3.26).

Exercise !q
3.4.1. Usíng expression (3.19), calculate the evolutíon operator for a for A(q) = ~B xq,
B constant, i.e. a constant magnetic field, and V(q) = O.
3.4.2. Add to the prevíous action an harmonic potential term V( q)
calcu}ate agnln the evolution operator.

E:xercise S. 5
Following the ídeas expla.íned ín Sectíon 3.3, calcula.te the real time evolutíon operator
e-óHT in the case of the hamíltonia.n:

Note that the re;ult obta.ined by adding ali the classical traje<:tories is íncorrect and that
a phase factor in front of half of the contributions is required. This difficulty reflects the
singular nature of the function V(q).
3.6 Path Integrais in Quantum Mechanícs: Generolizations 61

Exercise 3,6
3.6.1. As suggested in Section 3.5 evaluate thepath integyal (3.74) and derive the result
(3.75).
3.6.2. Verify the unitaríty of the S-matrh: (3.86).
3.6.3. Use the holomorphíc representatíon to c.alculate the order !. correction to the
energy leveis of the hamiltcnían

Perform the calculatíon for the two different path integral representatíons, obtained either
by directly replacing the cla:ssícal phase space variables p, q by the complex variables z, .i,
or bywriting the hamlltonían in terms of creation and aonihilatíon operators, commuting
ali creation operators te the left, and then only replacing quantum by classical varinbles.

Exen:i3e 9. 1
Consider, in the holomorphic representation of quantum mechanics, the state5 with wave
function ,P(z) = e 01 , where o ls a.n arbitrary complex number (also called coherent
state5). Calculate the average of the position and momentum opera.tors q,p and then
the díspersions ó.q ((q2 ) (q) 2 ) 112 , ó.p ((p2 ) - (p) 2 ) 112 • Show that tbe states ,pare
minimal dispersíon states: ó.qD,p = !.
62 PatA integrais ifl Quantum Mechani<!s: Generolil:atiom A3,J

APPENDIX 3
QUANTIZATION OF SPIN DEGREES OF FREEDOM, TOPOlOGICAl ACTIONS

ln Section 3.3 we have evaluated the path integral in an example where coordin~te
space has non-trivíal topo\ogical properties. We now want to discuss two other examples
where topology plays an essential role, in the sense that an action generating the dassical
equations of motion cannot be globally defined. We first quantize in the path integral
formalísm angular momentum operators in a fixed representation. One of tbe pe<:uliarity
of this system is that phase space itself has non-trivial topological propertíes. The second
example iE provided by the rnagnetic rnonopole which glves a non-trivial topological
structure to coordinate space. In both exarnples one speaks about topological a.ction. The
property that a topological action cannot be globally defined leads t<> the quantization of
its amplitude, a property specific to quantum mechanics. lndeed, in classical mechanics,
a multiplication of the action by a <onstant does not modify the equatíons of motion.
Note that In this sectíon, to sirnplify notations, we use a real time forrnalism.

A3.1 General Symplectic Forms and Quantízatíon

We fimt discuss the general exampie of quantization in the case in wbich phase space has
non·tri'lial topological propertíes. We then consider the exarnple of spin dynrunics,
In the ha.miltonian formulation of classical mechanlcs the action bas the forrn (3.8):

S = [ ' dt Jp(t}4(t) H {p(t),q(t),t)]. (A3.1)

We note that the term f ptj dt represents the area in phase spa.ce between the dassical
trajectory C and the axls p = O. lt ca.n be rewritten:

(A3.2)

in which BD, tbe boundary <>f the domain D, oontains the dassical trajectory C and
a !ixed reference cunre. H we now pa:rametri"" phase space dlfferently, introdudng new
coordinates u,, the r.h.s. oftbe equatíon becomes:

In the I!Ulgua.ge of forrns w w"~ du<>. 1\ dull is a 2-form (see Sectíon 1.6) which by
constructíon is obtained by differentiating a 1-form, here:

w;::::dw'1 (A3.3)
dq
w' =p(u),.a du 0 w,du,, (A3.4)

a.nd called the syw.plectic form.

Since the operator d acting on differential forms ís nílpot.-nt, the forrn w iE clased, i.e.
satisfies:
dw=O. (A3.5)
A3.2 Pa.th Integro.& in Quantum Medwnics: Generalizatwns 63

E:rample. ln Section 3.5 we have discussed the holomorphie representation of the har-
monic osdllator. \Ve: have introduced a complex parametrlzation of phase space (equation
(3.79))
p- iq -ihz
ln term.s of z, z the symplectk forrn becomes:

dpi\ dQ

Previous consideràtions immediately generalize to several degrees of freedom. Let u 0

be 2n variables pnrametrizing a phase space for n degrees of freedom. The action S in
the hamiltonilin formulatíon can be written:

S"" f "'- f dtH (u(t),t), (A3.6)

JD lan
where w, a symplectic form, is a dosed Z.form:

dw=O.

This latter conditlon is a sufficient condition for the equations of rnotion to depend only
'"' the hounda.ry éJD of the domain D but not on the interior, as was obvious for the
inítial sction (A3.1). They then take the form:

(A3.7)

Locally equa.tion (A3.5) can be integrated in the sàme way as dp 11 dq ca.n be integrated
into pdq. However, if phàSe space has non-trivial topological properties, it cannot always
be integrated globally, i.e. the symplectíc form is not exact. Tbis phenomenon has
peculiar consequences in quantum mecha.nics sinoe the path integral involves the actíon
explicitly in the form e' 8 f•. Fbr the path integral to makes sense, this pha.se factor must
be independent of the choíce of the action S. Let us examine this problem in the example
of the qua.ntlzation of spin degrees of freedom.

A3.2 Spln Dynamics and Quantization

A3.2.J Cla.ssícal spin dyna.mics
Let us c:onslder a vector S in three dimensions of fixed lengtb s:

S 2 = s2 •

Tbe simplest dynamics we can write for S is:

dS
H><S, (A:l.B)
dt
in which H is a constant vector. Tbis equation is first order in time and involves two
iudependent varíables corresponding to a poínt on the sphere. These variables ca.n be
64 Path Integro!, m Quantum Medwnicsc Genemlizatioru A3.2

considered "" a couple positíon and conjugat€ momentum. Phase space ís therefore
i•omorphic to the sphere s2. Th<' hamihonian is simply:

(A3.9)

More precísely it can be ~rified that i! we parametrize S as:

S = s(sín Ocos~t.sin 6sinll', cos 0),

then the actíon S,

S =f jscosilij>-1i']dt, (A3.10)

generates equation (A3.8). We see in thjs expression that coa 9 and 11' play the role of
conjugate variables. However two rematks are in order: First we have int..grated the
symplectíc 2-form but, for this purpose we ba..., been forced to use a parametrization of
the sphere which is singular at 6 O and ". When the traject.ory contains the north or
the south pole of the sphere the integ>-al is not defined. The 2-form cannot be integrated
globally ínto a 1-form because the integral J dcpdc058, whlch is the area on the sphere,
is delined only (mod47r).
In classíc:al mechanics, of course, these properties are irrelevant since only equations
of rnotion are pbysical.
Note the symplectic form has other useful reprcsentations:

dcosB 1\ dip = ~s- 3 c;; 1 S,dS; A dSk 2idz; 1\ dz;,

where z, is a twe>-component complex vector of length 1, corresponding to the isamor-

phism be-nS, a.nd the symmetric space CP1 (see Chapter 15):

where u is the set of Pauli matrie<"S. In a special glluge the vector z can also be wrítten:

z, = é"'/ 2 coo(fJ /2),

'2 = e-'"'' 2 sin(B/2) .
•4S..2.2 Qu~nti.zatíon of .!pin deyrres of frredom
L<ot us consíder the path integral representation of the corresponding evolution operator in
quantum mechanjcs. The action itself now appears expllcttly and the problem discussed
in the preceding section becomes relevant. The path integral exists only if the integrand
eiS/h ís defined. Since the total area is defined only (mod 41f) the integ>-and must be
invarlant under such a cha.nge. This implies that 41f s must be a multiple of 21f li: The
patameter sfli is quMtized and Cll.ll tal<e only half-integer values. This a is a generic
quantum mechanlcal property of topologirol contributions to the action, i.e. contributions
which are not globally defined: their amplitude is quanti.zed. The magnetic monopole of
Section A3.3 provides another example of such a situation.
In the parametrization (A3.10) the action can be written (we now set 1i = 1);

(A3.11)
A3.2 Path lntegrnl! in Quantum Meclutnícs: Generalizations 65

expression which differ. from (A3.10) only by 8 total derivatíve: by choosing I 1/2.0
for s half-ínteger ~ integer respectively one renders eiS regular near fJ O, 1r.
To relate this action to the usual operator formulation of the angular momentum
relation.s iet us lirst note that clussically we have:

After quantization S, become.s P..- the conjugate momentum of the angular varíables op:

It can be verilied that the qua.ntum operator S± can be wrítten:

Using then:

we lind the relation between the angular momentum j and the parameter s:

j + 1/2. (AS.12)

In particular, since • is quantized, we tecover a property of quantum mechanics, quanti-

zation of spin.
If we call m the eigenvalues of S, we observe that in the <p configuration spac<! the
corresponding elgenvectors are of the form e'"'" and the projector K on the basis is:

m,...j
K (op",<p') = L e•m(..,'-.,") sin lU+ 1/2) (rp' rp")J
sin [(IP' -'f'") /21
(A3.13)
"1"""-J

We can compare this expression with the short-time path integral representatíon whích
leads to:
K (rp",<p') <X Lj• dp.., eiP.,(IP'-.,"+ln•),
n -•

. ,
"' sm [s ('P - 'P ) L. ( '
" I "' :zt -on
" + 2mr).
n 'f' 'f'

The sum over n has to be regularízed but the fa.ctor sín s (<,o' -'f'") is consístent with
(A3.13) and the identilication (A3.12).
A final word of caution: Although the path integral quBJitízation of spín varíables is
quite useful to study the classical limit (which is also the limit s --> oo), ambiguities due
to operator ordering leads to djfficulties in explicit calculatioll8.
 Pat.h Jnt"9rn!s in Quanium Mechamcs: Generolizations A3.3

A3.3 The Magnetic Monopole

Electromagnetism provides another example of the situation encountered in Section

A3.2.L The contributíon of the vector potential to the eudidean action is of the folm
(see (3.19)) ic J A(x) · dx, Le. the integral ofl-form. However the only physical quantity
which appears in the dassical equations of motion is the electromagnetic tensor. The
contribution to the action can be, as above, more generally written as the integral oi a
2-form (involving the magnetic field) pr<Nided th\s forro ís closed:

Agaín íf thís forro is not exact the vectOJ potential cannot be globally delined. This is
precisely what happens when a magnetíc field is generated by a magnetíc monopole.
The formal dualícy symmetry between magnetic OJld electric fields in Maxwell's equa-
tions has led Dirac to speculate about the existence of yet undis~overed isolated m"gnetíc
charges, magnetíc equivalent to electríc charges. An ísolated ma.gnetic charge g cre,tes
a magnetic lield B of the form:

B=g~ r
r",
The lield B also ís " singular solutíon to the ftee statk Maxwell's equation. It has an
infinite energy and hence, in thio form, it is irrelevant to physics. Hawever in Híggs
models finite energy solutions (solitons) can be found which coincide at large distan<:e
wíth magnetic monopoles.
The integral of the magnetic field over a. closed surface containing the magnetíc charge
is 4rrg, as one immediately verífies by usíng polar coordínates {r, 6, 'I'} with the monopo)e
as origin:
j F,jdX; 1\ dxj = g j ~X r dcos6d<p =411g
2

H the vector potentíal could be globally defined the integral wauld obviously vanish.
More directly if we try to calculate the corresponding V<'Ctor potential we lind in a family
of gauges

where n is a consta.nt unit vector. We observe that lhe potential ís singular along the
line of dírection n pessing througb the origin. This line o f singularíties ís unphysical and
can be dísplaced but not removed.
Again this property has no ciBSsícal oonsequences. However in quantum mechanics
since the action can only be defined (mod 4i7rey) the wejght factor e- 5 ís only de!ined if:

4neg =O (mod 21r) => 2eg = integer.

We recognize Dirac's quantization condition.

Note that when this conditio11 is fulfiUed, parralel transport (see Subsection 18.3.1) is
globally defined in lll'.3 ,
4 STOCHASTIC DIFf'ERENTIAL EQUATIONS: LANGEVIN,
FOKKER-PLANCK EQUATIONS

In Chapters 2,3 we have shOWn bow the con,.,pt of path jntegra.ls naturally aríses in the
calculation of the matríx elements of the quantum mechanical operator e-OH when the
ha.miltonían H is local. Let us note that, in the particular case of hamiltonians quadratic
and even in momenta, the integrand of the path integral defines a posítive measure
and theref<Jre ha.s a probabilistic ínterpretation. In thls chapter we díscuss Langevín
equatíons, i.e. stochastic dilferentia) equations related to diffusion processes or brownían
motion. From the Langevin equation we derive the Fokker-Planck (FP) equation for
probabílity distribution of the stochastic variables. The FP equation has the form of a
Schródinger equation in ímaginary time, C>f the typ€ we have studied in previous chapters
(the corresponding hamiltonían ís however in general non~hermitían). We then show that
averaged observables can also be ca!culated from path integrais, whose integrando define
automatically positíw mea.sures. In some cases, líke brownian motion on Riemannían
manifolds, difficulties appear in the proper definítion of stochastic equatiom;, quite similar
to the qu!llltization problem in qu!Ultum mechanícs. Time discretization allows to study
the problem.
This chapter will serve as an introduction to Chapters 17 and 35 in which stoehastic
quanlization and criticai dynamics will be discussed.

4.1 The Langevln Equatlon

We call the Langevin equation a lirst order in time stocba.stic dilferential equation of the
form:
{4.1)

in which q(t) ís a point in Rd, f; (q) a [unctiou of q and v(t) a set of stochastic functions
called hereafte<" the "noisc". The noise can be defined by a functional probability distri-
bution [dp (v)J. In what foUows, we sha!l specialize to a gaussian noíse with a probability
distribution of the form

[dp(v)] [dv]exp [- 2~/ vf(t)dt], (4.2)

This particular form of the gaussian noise, called gaussian white noíse, is related to
Markov's processes (see Appendix A4). The const~mt !1 charo.cterízes the width of the
noise distrlbution. Note that by a rescaling of time t ,__, nt {and a redefinition of the
noíse variable) it can be transferred in front of the drivíng term f- f /fl.
Alternatívely the gaussian noise can be characterized by its 1~ and 2-point correlation
functions:
(v;(t)} =0, (4.3)
as can be wrified immediately using the results of Chapter 2. Equation ( 4.1) is not the
most general stochBStic first order dilferential equstion, but we want to postpone the
discus.sion of th" general form, beca use it involves additional problems, until Section 4. 7.
6f Stocho.stic Difjerential Equation> 4,2

Giv€n the Vlllue of q(t) at inítíal time to, q(to) Qa, the Langevín equation generate•
a tíme-dependent probability distrihutian P (q, t) for thé stochastic vector q (t) whícb
forrnally can be written as:

p (q, t) = ú] 6 [g;(t) q;J) (4.4)

ln equation (4.4) as in equation ( 4..3) brackets indicate average over the noise. The vector
q is th€ argurnent of P (q, t) and has no relation with the function q(t). If O {q) is an
arbitrary function, definition (4.4) tben implíes:

f P(q,i)O(q)dq= (O(q(t))) (4.5)

4.2 Tbe Fokker-Planck Equation

Let us show that equations (4.1,4.2) irnply a differential equation for P(q,t). Dífferen-
tiating with respect to time equation (4.4) and using equation (4.1) we obtain:

Ê'(q,i} ([-!f, (q(t)) + v1(t)} ôq~(t) Hq(t)- q)). (4.6)

In this equatíon tbe syrnmetry of the é-function a!lows us to replace the differentíatíon
ôjôq;(t) by -8/ôq; which can be taken out of tbe average. Then using the Ó·function
we can replace q(t) by q in /; ( q(t)):

ô
P(q,t) éiq; [V; (q) P (q, t) {v;(t)6 (q(t)- q)}], (4.7)

Equation (4.7), which is índependent of the noíse dístrlbutíon, ímplies the conservation
of probability since lntegrating over -..·hole q spaoe we find:

f dgF(q,t) =o. (4.8)

However, to be a.b!e to establísh a partia! dífferential equation for P(q, t), we need a noise
distributíon of precisely the form (4.2), í.e. local in time and ga.ussian.
A wejul aigebmic identity. Let F (~o~) be an arbitrary functional of v(t). Then:

Note that:

v, (r)exp [- 2~ f v (t)dt] "'-n6v,6(r} exp (- 2~ f v (tjdt].

2 2
(4.10)

We can therefore íntegrate patb integral (4.9) by parts:

f 1
\dv] F(v) V; (T) exp [- !l.
2
f 2
v (t)dt] = !l. f \dvjexp [ ~ 2~ f 2
v (t)dt] óv,.S(r)F(v).
(4,11)
4.2 Stocha.stil; Di[ferentioJ. Eguations 69

This identíty can also be written:

(F(v)v,(-r)) =fijfJF(v))). (4.12)

\liv;(r
A similar identity can be proven for any gaussian distribution by the same argumenL
For example, for:

!dp(v)] ldv]exp [-~/ dtdt'v,(t)Ki.i(t,t')v.;(t')],

one finds:
(4.13)

Applicatíon. Using ídentity (4.12) and chain rule, since q(t) ls a functional of v(t)
through the Langevin equatíon, we find:

(vi(t)ol (q(t)- q)) =

1 ó
n \ 6 ~-'i{t) l! (q(l)- q)
) 1óq,(tl a
n \ óv,(t) &q•(t) li (q(t) - q)
). (4.14)

Aga.in of&qk(t) ca.n be reploced by -fJ(Bqt and taken out of the average. We stlll have
to clllculate the funct.ional deriva.tive 6gk(t)f6vj(t). Let us therefore formally integrate
the La.ngevín equation (4.1):

q;(t) = q;(to) ! f'/; (q (t'')) dt" +

fto
t
JkJ
v; (t") dt, (4.15)

Differentia.tíng witb respe<:t to "i (t1) we ftnd:

Óq;(t)
(4.16)
6vj (t')

The function 6(t) is the usual step function. The Langevin equation ís causal, which
implíes that q (t") depends on v (t') only for t" > t'. Thís property has been used to
restrict the integration range to t" > t' in the r.h.s. of equatíon (4.16).
Actually we need only l!q, (t) jbvi (t') for t = t'. Ftom equation (4.16) we see immedi-
ately that this limit is ill-defined because we find 6 (O} in the r .h.s. Thls is a dísease o f
our deriva.tíon, not of the equatíon, The difficulty is rela.ted with a well-known property
of the Langevín equatíon (4,1): d (q 2 (t)) fdt ís well-deftned but 2 (q(t) · q(t)) ls not. Due
to the singular nature of the noise 2-point rorrelation functjon, time differentiation and
averaglng are two operations whícb do not commute. We shall illustrate this polnt !ater
ou a particular example (see equation (4.53)). To get around this difficulty it is possible
eíther to first consíder the time discretised Langevin equation (see Sertion 4.8) or to
reploce tlle noise 2-poínt function by a regulsrized functíon:

(v,(t)v; (t')) =é,; n '1 (t- t'). (4.17)

in whicb 'l(t) is a functíon peakOO around t O, even by definition be.:ause the 2-poínt
functioo is symmetric in the exchange t .... t' a.nd satisfies:
+O<>
l-00 'l(t)dt = 1. (4.18)
70 Stochastic Differential E'[Uations 4.3

Then, using identities (4.12-4.15), we find that. we ínstead have to calculate

dt 11) (t- t') 6q.(t) .

f lwi (t')

The quantity 8 (O) is then replaced by:

j-oo+oo '7(1-t')&(l- t')dl j" 1

= _.., l)(t')dt' = ~' (4.19)

because IJ(I) is even.

Retunúng now t.o equation (4.16) and taking for IJ(!) the ~-function limit we see that
the first term of the r .h.s. goes t.o zero when t goes to t 1 and therefore:

(4.20)

Equation (4. 7) tha>!fore becomes:

. l & [ 8P ]
P(q,t) ="2aq, n&q, +f,(q)P. (4.21)

Thís ís the Fokker-Pianck equation ossociated with La.ngevin equation (4.1) Md noise
(4.2). Taking ínt.o account the boundary condítion q(to} = Qo, we observe that the
equation (4.21) ís identicEII to the Schrõdinger equation for the matrix elements of the
imaginary time evolution operat.or of a generally non-hermiti8ll hamilt.onian H:

H ! (ni>2 + if> ·r (<il) , (4.22)

P(q,t) (ql e-(t-to)H lqo) . (4.23)

We have therefore estabHshed tbe formal relatíon between st.ochastic dilferential equations
and euclidean quantum mechanícs. All observables which can be calculated from the
Langllvín equatiou by averagíng over the noise, can be recovered by methods of quantum
mechanics using the Fokker-Planck (FP) hamilt.onian.

4.3 Equilibrium Distt"ihution. Correlation Functions

Methods of quantum mechanics can now be used t.o study severa! interesting questions,
tbe li.rst beíng the exísteoce of an equílibrium dístribution. We have only to remember
that the FP hamilt.onian i.s not hermltían and therefore its left and right eigenvect.ors do
not coincide. Note that the arguments whích follow c&n be generalized for each conoected
component if the ~tem is not ergodic and the configllration space ís decomposed int.o
disconnected components, a sítuation we will face in Quantum Field Theory.
Tbe equilibrium dístribution Pn (q) il; the la.rge time limlt, if it exists, of P (q,!):

p~ (q) = • ..!!~= p (q, t)' (4.24)

Tbls implícs that ~ (q) is a time-independent solutíon of the FP equation (4.21 ), i.e"
ríght eigenvector ar H wlth eigenvalue O, that it is positive and normalízable

f P 0 (q)dq < oo. (4.25)

4.3 Stochastic Dífferentíal _f;quations 71

\Ve ?o"' note that conservation of probability has imposed the spet:ial form of the FP
hamtltomao (4.22), 1.e. the íactotizati<>n on the left of the differentíal operators âj&qi.
Therefore a coru;tant is a candidate fo• being 8 left eigenvectm with eigenvalue zero of
the hamiltonian H. lntrodudng a br~t and ket notation (not to be confused with the
notaticn mean:ing .avera.ge over the noi~ v) we ca.n wrlte-

(O]H =O, with (O]q) =1.

With the same notation we can write

H]O) O, with (q]O) = Po(q),

The eonditlon (4.25) thus ís the condition that the vector ]O) has a finíte nonn, a necessary
candition for the eigenvalue zero to really belong to the spectrum

(0]0) "' j P (q) dq < oo

0

Since one can show that the eigenvalues of H have a non-negative real part, the vector
!O) then ls the grouud state of the ba.miltonian.
If no such solution exists, ali eigenvalues of H have sttictly posítive real parts and
lhe langeYin equation (4.1) has runawa;y solutions ín the sense tbat tbe probability of
finding q( t) inside a bal) of arbitrary but finite radius goes to zero at large time:
(i) algebraícally if the spectrum of H has a continuous part extending down to the
ori.gín (thís is tbe cBBe with the brownian motiOD, for example),
(ii) exponentíally witb a rate whícb is the iuvel'5e of the real part of the ground state
t'Ígenvalue otherwise. This rate ís called the relaxatíon time. (For more details see
Appendix A4).
Finally let us note tha.t tbe sole knowledge of the equílibrium distribution:

(4.26)

does not determine the funr.tíons /;(q) uniquely. Indeed, by demanding that .A.J(q) is a
Lím<>-índependent solution of the FP equation, we only obtain the condítion:

(é!;- B;E){fW;E- /;) O, (4.27)

setting:
(4.28)
be rewritten as a. current consen'IJ.tion eqUation:

Õ;V;(q) =o. (4.29)

Gorrelation functians. Up to now we haye discussed only equal time averages. However
La.ngeYin equa.tíon can also be used to calculate correlation functions_ of observables
dilferent times, like z<n)(t, t2, ... , In):

(4,30)
72 Stochastic Differential Equation• 4.3

Assuming that the boundary condítions are q(to) = Qo and ordering times !o :S t 1 S
t,. we can use the ddinition of the FP hamiltonia.n to ,..,.,..,.ite ztnl as:
t 2 <:S ... ::;

(4.31)

with Q(t) being tbe opera.tor q ín the Heísenberg representatlon

Q(t) eH'qe-H'.

We recognize a forro similar to the representa.tion (2.63) of correlation functions in quan-

tum mecbanjcs. lntroducing the tíme-ordering operat.ion T acting on quantum operators
....., can then write, irrespective now of the order between time., (see equation (2.64))

(4.32)

When àJl equílibrium distribution exists, in the limit t0 ~ -oo (ooundazy conditions ín
fru past) the correlatíon functions converge towards equilibríum cc.,.,...,lation function.s:

(4.33)

We recogni>Al the vaeuum ar ground state expectation value (2.65) of time-ordered prod-
ucts of operators íntroduced in &ection 2.5.
Time evolution of obserr;ables. Goíng from the FP equatíon to expressions in terms
of time-dependent operaton, we have passed from a Schrõdinger picture to a Heísenberg
picture. In the Heisenberg picture one directly wriws equations for the evolution of tim,..
dependent aperators. Let us give an application here. In terms of Heísenberg operatorn
the average af an observable O(q) at time t can be wrítten

(O(q(t))). =(OI O[Q(t)) eHto lqo).

We can write an evolution equation for the operator O[Q(t)] eHto:

For the matrix elements O(q, t) of the operator

O{q, t) ..., (O!OIQ(t)J eHto iq),

this translates into the partia! dHierential equation

.
O(q, t) = 1[ô
'2 n&q- /(q) ]a aqO(q, t). (4.34)

Then the averages (O(q(t))). are simply obtained by integratíng O(q,t) with the initial
dístribution at time t 0 , here

(O(q(t))). = f dq O(q, t)~(q- q0 ). (4.35)

4.4 Stochastic Dijjerential. Eqv.atiuns 73

4.4 A Spedal Case: The Purely Dissipative Langevin Equation

In one special case the harníltonian (4.22) ís equivalent to a hermítia.n hamHtonian, when
f; (q) i• a gradient:
J, (q) = na,E(qJ (4.36)
This corresponds to the purely dissipative Lallgevin equation:
g; = -~í16;E (q) + v,(t). (4.37)
Then setting:
I I
P(q,t) = e-:;Eiq) (qjU(t,to}lqo)e2E(qoJ, (4.38)
we transform equn.tion {4.21) into:

~ (qiU(t,to)l Go) = -~ (- ~; + ~6;E) ( 6~, + ~6;E) {qiU (t,to)l Go). (4.39)

The hn.miltonia.n H transforms into a hamiltonía.n H:

H ~A)A;;;;;~[f> 2 +~(VE{q)) 2 -i.6.E(q)J, (4.4D)

in which A Ls the opern.tor:

A(q,ôfôq);;;V.+!VE ip+~V'E, (4.41)

n.nd U{t, to)= e-1<-<o)H is the corresponding emlution operator in imagina.ry time.
We see that the hamíltonia.n H Ls positíve. Moreover lf the wave function e-E(q)/> ís
uonna!izable:
(010) f
dq .,-E(q) < 00,

it corresponds to the eigenvalue zero:

AIO) o ~ HIO) =o,
and thus is the ground state of H. At large times the operator e-(<-to)H projects onto
its ground sta.te. The interpr<>t.ation of this result in terms of the probabi!Uy distribution
P (q,t) iS then:
,~':!, p (q, t) = .,-E(q)/2 eE(qo)/~(qiO)(Oiqo) = e-E(q).

The distribution P (q, t} converg""' at large time towards the equilibrium distribution
e-E('IJ.
lf the waV<> function e-E(q)n is not normalizable instead, then. exísts no equilibrium
distribution, Langevin equat.km (4.1) has only run.away solutions,
Remark. The case of the purely disslpa.tive Langevin equation (4.37) corresponds to
detailed baln.nce for discrete procesres (see Appendix A4.2). The drlft force f, is then
called conservative. In the absence of noise the Langevin equation reduces to a gl"adient
llow:
ç,(t) = -!na,E(q(t)J (4.42}
Taking the scalar product with the vector q and integrating we obtain:

f' <l(t')dt' = -ãn [E (q(t)) E (q(to))J. (4.43)

'··
Therefore, In tbe absence of noise, E(q (t)) Ls a. monotonous decreasing functíon of time.
 S!ochastic Differential Equai;Wn:; 4.5
74

4.5 A Simple Exmnple: The Linear Langevln Equation

Let us consider the equation

q= -wq + v(t), (4.44)

in which v(t) i• the gaussian noise of equation {4.3), with !l =L This equation provides
an example of the special case discussed in Section 4.4:

After the transformation ( 4.38), the a.ssociated FP hamiltonian takes the form:

(4.45)

Tbe eígenva.l ues <n of H are:

<, (n+ ~)jwj- ~w, n ?_0. (4.46)

see immediately tbat if w is positive, <o vanishes. At the same time

\ \1e e~wq' ís normal-
izable ond is thus the equilibrium distrlbution,
H w is negatiw:
w<O, (o=-wt
the lowest eígenvalue is positive, e-wo' ís not normali?.able and for large time the distri-
bution P(q, t) becomes (taking to O for convenienre):

(4.47)

This expressíon sbows that the prob!lbi!ity of finding q at a finíte distam:e from the orígín
decreases exponentilllly at a rate r "' 1f jw].
Fínally the special c!lSe w = O corresponds to the simple brownian motion, the spectrum
of the FP hamiltoniBn is continuous and covers !0, +oo], and the probability of remaining
at a 6níte distance from the orlgjn decreases algebraically !lS lf.,/t.
Let us r.ecover these results by solving the Langevin equation ( 4.44) explicitly:

q(t) = 'lll e-""+ lo' e-"'( H') v (t') dt'. (4.48)

Averaging the equation over the noise we find:

(4.49)

This expression already shows that for w < O no equUíbrium ís reached. Moreover for
q0 #O, (q(t)} grows exponentíally with time.
Since q(t) islinearly relate<! to v(t), it has a gausslan dístribution which is characterized
by (q (t)) and (q2 (t)). Ftom (4.48) we obtain:

( jq(t) (q(t))J 2 ) = ~ (1 - e-
2
w') (4.50)
SSJ&&UhLJJJPl A.k
4.6 75

Tbe gau.ssian distribution P(q, t) of q(t), determlned by its two fu'St moments, is tbus
given by:

l-1/2 exp [ w -w<J2] •

P(q,t)
I;:; (1- e
7r _,...,,
) {l _e ...,.) (q lltJC (4.51)

This expression isequivnlent to eo<pression (2.37) and conta.ins the asymptotic form (4.47).
To illustra.te the difficulty encountered in the deri..,..tion of the FP equation let us
calculate the q(t) 2-paint function for 11o = 0:

(4.52)

Differentiating witb respect to t 1 , we find:

(4.53)

The limit t1 _, h ís clearly Ul-defined while if we take thls llmlt in expressíon ( 4.52) fust
and then differeotiate we find:

Tbe sa.me result is obtained frorn equation (4.53) by setting e(O) = 4, i.e. taking tbe half
sum of the derivn.tive from above and below.
RemarA:, lt is also eaBy to solve a more geneml Langevin equation of the fonn:

q -wq + g{t) + v(t),

in whicb g(t) is a given function of time. The srune argurneots as above would have led
to a formula for P (q, t) equivalent ta expression (2.37) with the idP.ntification:

/(t) = m [wg{t) g(t)].

We leave thís calculation as an exercise.

4.6 Path Integral Representation

Applying the method described iD Chapter 2 ta equation (4. 21 ), we can derive path
integral representations for the probability distribution P (q, t) and averaged observables,
Let us however give here a more direct derivatioo baseei on Langevin equation (4.1) aod
the generalízation of identlty (1.26) ta ao infinite number of variables.
We calculate the generating functiooal Z{j) of correlation functions

Z(.j) = ( exp j dt j(t)q(t)) • . (4.54)

We use the boundary condition:

q(to) =qo.

'
,.;,...,.t_.,·,.~~~~~.:.<tJ,~·..,.;;'IDe.~--K~'i4.<'~.•,;,,,••"""'-~~0iif1~..__~~---~-e~~
76 Stocha.stic Differential Equations 4.6

lmposíJJg then Langevin equatíon (4.1) by a product of 6-functíons, we wríte ZU) as:

ZU) = f [dv(t)] exp [-

2
1
n f v 2(t)dt] j
q{to)=<oo
[dq{t)] exp f dtj(t)q(t)

x det M TI tl [v;(!) q,(t) -~fi (q(t)}j (4.55)

'·'
According to the identity (1.26) the operator M i.s the functi<mal derivative of Langevín
equation (4.1) with respect to q(t)

d
M;; (t, r)= dttl (t (4.56)

In expres•ion (4.55) the integral over the noise is trivial and we obtain

ZUJ = 1
q(t(l)-qo
[dq(t)JdetM exp {-fdt
2
[(<H !f(q)) /2ll-j(t)q(t)J}. (4.57)

Tlu. detenninant. We now eva.luate the determinant of the operator M. It can be

factorized into:
(4.58)

The first factor is an i:rrelevant consta.nt. The operator M iB obtalned by differentiatiJJg

with respect to q(t) the integrated form (4.15} of the Langevin equation:

(4.59]

in whlch B(t) is again the step function.

Let uo set:
(4.60)

and use identity (1.33):

d~ exptrln,
.to expand âet M:

det M exp{ O(O} f tr u(t)dt ! f O(t1 - t,) (J (t2 -h) tr [v (t 1] u (!2)] dt,dt~ + · ·

+ n f 9(!1 t2)···8(tn -t!)tr[v(t,]·"V(tn)]dtt .. ·dtn+ ··}·(4.61)

Due to tbe product of 6'-functions ali terms vanil;b but the first one:

det M exp [e (O) f trv(t) dt] . (4.62)

We aga.in find a result involving the ill-defined quantity 6'(0). Semiclassícal evaluations
of path integrals generate simila.rly ambiguous expressions (oee Section 38.2). To glve
4.7 Stochastic Differential. Equations 77

a meaníng to expression (4.62) one can díscretize time ín Langevín equation (4.1) (see
Sedion 4.8). After taking the contínuum limlt one finds:

The path integral. We thu• obtain the path integral representation

ZU) f.
q(l.o)=qo
ídq{t)] exp f-S(q)/fl
l
+f dt j(t)q(t)] , (4.63)

with the definítíon:

(4.64)

Similarly one derives a representation for P (q,!):

J.
q(t)~q
P(q,t)'"' (dq(r)]exp [-S(q(r})/il], (4.65)
q('o)~ ...

It is easy to verify tbat the same patb integral representation is obtained when the
metbod explained in Chapter 2 is applied to tbe Schrõdinger-like equation (4.21).
lt follows in particular ftom the path integral representation ( 4.65) that, a.s expected,
a solution of equation (4.21) wbich is a positive dístrihution P(q0 ,t0 ) at inítial time to
rema.ins positive at any !ater time.
Perturbcti"e ezpansian. The parameter n here plays tbe role of li in qu!Wtum mechan-
ics. It orders naturally perturb..tion theory. We then note tbat, at leading arder, the
classical action reduces to:

(4.66)

Thís means that in the particular example of La.ngevin equation (4.37), perturbation the-
ory ha.; to be expanded ..round one ofthe extrema of E(q). However not ali extrema are
equivalent once one tak!'S into account the first quantum correction dueto the additional
term n&,J.(q), as the example of Section 4.5 reveals. Expression (4.46) indeed shows
that the second term ll.E( q) lifts the degeneracy between mínima and maxima ..ud that
only minima are good startíng points for perturbation theory.

4.7 Sto<ohastic Processes on Rlemanoi!lll Manifolds

We now consider the most general markovian Langevin equation wbich ha.s the form:

q'(t) = -!f'(q(t)) + e~(q(t)h(t), (4.67)

in which v.(t) is agaln a gaussian white noise:

(v.(t)) =O, (v0 (t)Vb (1 1)) = 6 (t t') 6ob. (4.68)

78 Stochastic DitJerential Equatians 4.7

The linearity in the noise is a eonsequence of the marlwvian charw::ter of the noise' distrí~
bution which implies that the noise correlation functions are proportional to 6~functions
in time (see Section 4.8).
In contrast with the LMgevln equatíon (4.1), equation (4.67) is somewha~ ambiguous,
the difficulty being of exactly the sam" nature as the problern of ordering of operators
a.rising in the quantizatíon of a harnlltonian of the form (3.24).
The problem can as usual be best understood by discretizíng time (see Section 4.8 for
details). Let us cal! e the tíme step and replace equation (4.67) by:

q' (t +c) q'(t)- ~é f' (q(t)) +e~ [i (q(t) + q(t + ~))Jv.(t), (4.69)

with for dimensional reasons:

(4.70)

Equations (4.69,4.70) show that, in contrast with usual differential equations, the varia~
tion q' (t +c)- q'(t) ís of arder ,fi, which iE typical of the brownian motion. Hence ii in
tbe llisctetized equation (4.69) we replace for example e~(q(t)) by e! [1!q(t) + q(t +e))],
quantities wbich are [orma.lly indistínguishable in the rontinuum limit, we change the
equation at order e. InsteAd, if we ma.ke the SIUIIe substitution in f(q), the equatiou is
modified by terms of arder e 3 12 wbích are negligible.
In what follows we adopt the symmetric Stratanovich convention (4.69), althougb it
is unna.tura.l from tbe practícal point of view, becawe it bas simpler transformation
properties under a change of variables Blld therefore of coordínates on a manifold.
The Foitker-Planck equation. Adapting the method of Section 4.2 one derives a Fokker-
Planck equation for the time--dependent distribution P (q, t).
Without giving ail details let us just indicate the main st.epS. Prom the Langevin
equation ( 4.67) we immediately obtain:

. 8 l
P(q,t) = oq' !J'(q)P e~(q) (v.(t)6 (q(t)- q))j, (4.71)

Using the same regularization for the noise 2~p0int function we can calculate the second
tenn in the r.b.s.:
(v.(t)ó(q(t)- q)) (4.72)

The FP equation tben takl:s the form:

.
p (q,t) 2lof a(.).]
8qi le~(q) âqj &,.(q)P + f'(q)P (4.73)

The only examples we shall meet in this work correspond to Langevín equatioll5 on
IUemannian manifolds.
Brownian motion on a Riemannicm manifolti Covruiance under repfll'ametrization of
the manifold implies that the FP equatíon for the free brownian motion on a manifold
should have the form (for more detsils see Cha.pter 22):

(4..74}
4.7 Stochastic D'!Jerential. Equ.ations 79

in whicb 9i;(q) \s the metric tensor on th., llUl.Difold, 9 the determinant of g,;(q):

g"' detg;;' (4.75)

a.nd D(q, t) a scalar density, i.e. a density- normalized with tbe covariant measure ,jYdq.
1t is therefore related to P(q, t) by:

,jYD(q, t) = P{q, t). (4.76)

If the manífold \s compact, a constant scalar density D is abvjous)y a static solution to

equation (4.74) and therefare corresponds to the equilibrium distribution.
Comparing equations (4.73) and (4.74) we condude that:

(4.77)

which means th!Lt the matrix e~, if it is a square matrix, ls the inverse af the vielbein
(Section 22.6). We see th8t by constructíon the t.h.s. ls a positive symmetric ma.tríx.
Furthermare we observe tha.t the free brownian motion does not correspond in general
to f Osínce, 8S a short calculatíon shows, equations (4.73) and (4.74) are identicalonly
if:
J'(q) = -e~(q)V;C:,(q), (4.78)
in which V \s the covariant derivative on the manifold. The equation shows that f does
not depend only on the geometry o{ the manifold but a.lso on t.he choice of the vielbein.
Let us finally note that equa.tíon (4. 73) ls the most general second order stochastic
differential equation: the operator 8/ôq' ls implied by total probability conso,rvation, the
positivity of t.he coefficíent of 8 2 / 8q'ôqi by the condition that the solution remains a
positive dlstribution a.t ali times as will become clear once we construct. the path integt"al
representation.
Path inteyrnl representation. From now on we restrict ourselves, for simplicity, to the
case in which the matríx e~(q) i!! a.n invertible square ma.tríx. Equa.tion (4.67) can then
be rewritten
=
vQ(t) [e- 1].,(ql(q'{t) + ~/;(q)). (4.79)
We ha.ve first to calculate the determinant of 6v.j/iq':

óv.(t) = [e- 1 ] ·(q(t))Mi (4.80)

bq' (1") OJ I '

with
M"!"'{htft+~~: +oi~,l•- 1 lw[l+!ltq)J}<~ct 1"). (4.81)

The determinant ca.n thus be written:

detM ~ detMfl [dete(q(t))J- 1 , (4.82)

with, "" a consequence of ídentity ( 4.62)

-
detMocexp [1! (18! ô [ 'l (•t
1

2 dt 28q,+••eq•"-hl9+1.1
i t))1 . 1 (4.83)
80 Stocha.stic Differential Equations 4.B

In expression (4J!2) we again finda divergent fO>CWr analogo\IS to the normalizatlon factor
(3.29). This factor can only be defined by a limíting procedure, for example we can US€
the discretízed Langevin equation (4.69). In both cases the appearance of this divergent
factor is related to the pmblem oi orderíng of operaton; for the quantum mechanics or
FP hamiltonian as already discussed in Se<;tion 3.2.
Co\lecting aU factors we obtain a path integral representatíon for P(q, t):

q(t)=•
P(q,t) =
J.q(lo)=!IO
ldq(-r)J fl [dete(q(-r)W' exp \-S(q)],
T
(4.84)

wíth the definítion:

S(q) ~i dT [ (q' + !J1(q}) [e- 1].;(q)(e- 1 ]aj(q) (q.i + Ul(q))

~~:;~e~ (~;[e- 1 )bt) (<i 1 + 4ft(q})]. (4.85)

In the c~~Se of a Langevin equation on a. Riemannian manifold the determinant in expres-

sion (4.84), as a consequence of relation (4. 77), is giveu by:

where g is the metric tensor. 1t therefore forrnally reconstructs the covería.nt measure on
the manifold {see Section 22.5).

4.8 Diseretized Langevin Equation

We now discuss a discretízed, markovian, time translation invariant Langevin equation in

the sma.ll step size limit a.nd in the simple case of a gaussian noise. This is a. useful exercise
in the sense that it cl.orifies a few subtle points which have appeared in the discussíon of
the continuum equation. Let us call e tbe time step. The Langevin equation, discrete
analogue of eqoation (4.1), ls then:

q;(t +é)= Ç;(t) !Efs(q(t)) + V;(t). (4.85)

Note tha.t the equation can be considered a.s a discretized form of the continuum equa.tion
only if the function f(q)Jiql remains bounded for lq\ la.rge. Otherwise it ls necessary to
imprCI\Ie it by a.dding higher order terms in ~-
Tbe noise distribution ls factorized Md a.t a given time, for ea.ch degree of freedom,
tbe distribution is:
dp(v) dv exp ( .? ) (4.87)
v'2,.-eü - 2Eü ·
Equivaleotly tbe ga.ussia.n distribution can be defined the 1- and 2-point correlation func-
tions:
(v;(t)) =O, (v;(t)v; (t'}) = E: 6wóiJ .n (4.88)
The dependence of tbe noi.se on é, v o: .f'., is typical of the btownia.n motion and it wUI
be justified helow that it yields the correct continuum limit.
4.8 Stucha.stíc Differential. Equatíom 81

From equations (4.86,4.87) ít is ea.sy to derive the rfucrete analogue of the FP equation.
Equation (4.86) implíes a rel<~tion between the tim~dependent distributions P(q,t) at
consecutivt: times:

P(q,t+E) (6(q(t+e) q))"=(ó(q(t)-~ef(q(t))+v(t) q))"· (4.89)

The average over the noi.«€ v 4; restrícted to the nois<> at time t, the average ovet anterior
times beíng performed by the integration over q(t). Moreover the noise v(t) and q(t) at
equal time are uncorrelated, as a coru;equence of the causallty of the Llillgevin equation.
The q(t) distribution by definition is P(q, t) and the nois<> distribution is gi~ by (4.87).
Thus:
P (q,t +<)=I dq' P(q', t)dp(v).5 (q' !ef(q') +" q) (4.90)

The integral over the noise js trivial Md can be performed even for a. general noise
distríbution (we leave as an exercise to study the continuum limít in the latter case).
With the gaussían noise (4.87) we obtain

P(q,t+e):. (2.,.n)-df 2 I dq'P(q',t}exp [-~(oi- !ef(q1) - q) 2 ] • (4.91)

Tbis equatlon has the general form:

P(q,t +õ) =I dq 1 p(q,q')P(q',t). (4.92)

The kernel p(q,q') whlch acts on P(q,t) is the equivalent of the evo)ution operator in a
small títne interval as given by (3.25). It plays exactly the role of a transfer matríx in
a on~imenslonal statistical system. A few rernarks concerning equatioll$ of the forro
(4.92) CBJl be found in Appendix A4.
T1re CD11tinuum limit. To derive the continuum FP equ&tion it is conveníent to start
from equation (4.90) and to fitst integrate aver q' using tbe 6-function. Tb!.. leads to a
jacobian .J whicb we expand at first otder in< (using lndet =trln):

In the same way we to expa.nd P(q',t) in powers of E, remembering that " is of order
.fi:
2
P(q',f)=P(q,t)+(!<f,(q) v; )
8P 1 8 P +0 ( E3 12 ) .
Bq, +;II;V;aq,Bq;

Averaging then over v and collecting ali terms we find

equation whlch ls equivalent to equation (4.21) in the small t: limit.

Note that in equation (4.86) only the terms up to order t: are relevant in the continuutn
limit. In pa.rtlculac it is possible ( although not converuent for practicaJ use) to rep)a<:e
J(q(t)), by f!(q(t) + q(t + t:))/2] since this leads to a. modification of arder il1 2 in the
82 Swchasti<: Differential Equa.tion> 4.8

equatíon. This last form Í5 the most convenient, howt'Ver, wben change of cootdina.tes
are involved beca use ít has simpler tra.nsformation ptopertíes.
Let us finally also note that equation (4.91) iterated in time immediately lea.ds to a
discretized form of the path integral representatíon of P(q, t). lt is then possíbla to study
the sma.ll time step limít by =panding in powers of E up to or der E. The ímportant point
is tha.t the typical va.lues of q q' are of order <: 1/ 2 • Then

q' ~d(q')~q q' -q

2..
2E
(q'- ltõf(q')-
2
q) 1

We need the terrns of arder <. In the last term which already íf of order E we can rep}ilCe
(q!- q;)(qj- Q;) by its leading arder e.verage f!E6 1,. We then finda discretized form of
the action (4.64).
Generolízation. A more general fotm of the discretízed Langevin equatíon ~:

q,(t +c)= q;(t)- %Ji (q(t)) +e;.. (q(t))v.(t) + d;.~(q{t)) v (t)vr,(t),

0 (4.93)

with tbe gaussie.n noíse still delined by:

(v4 {t)}=0, (4.94)

Equatíons (4.93) are the discrete e.nalogues of equations (4.67,4.68).

Note that we haw only expa.nded the r.b.s. of the Langevin equa.tíon up to seoond
arder in v(t). The rea.sons are the following:
(i) the noise 2-point function must be proportional to E for dimensional reasons; the
oonslstency of thiz choice wül be checl<ed in the derivation of the FP equatlon;
(U) since v,.(t) is of arder .ft, it ls sufficient to consider terniS at most quadratic in tbe
noise in the L~mgevin equation, bigher order terms give contríbutions of order at JeBSt
E3 / 2 and therefore negligible in the srnall ê linút.

From the Langevin equation (4.93) we ca.n derive by the method indicated above an
equation for the probahility distríbutlon P(q,t). Writing equation (4.93) as:

qi (t +E) = q;(t)- F; {q(t), v(t)), (4.95)

we obtain:
P(q,t +E)=- f dq' P (q',t)(6[q- q' + F(q', v)]) • . (4.96)

We now average over the noise at time t. Since the function F (q,v) is of order .fi., ít is
necessary to expàlld the r.h.s. <mly up to second ordet in F. This leads to:

P(q,t+•)- P(q,t)

~ Jdq' P(q',t) ô~; (F,(q,v)6(q-q')+~a~/i{q,v)F;(q,v)6(q q')). (4.97)

The dílferential operawrs ajéJq; can be taken out of tbe average e.nd facwrlzed in front
of the integrei. Th" integTal over q1 is then trivial. Replacing the function F (q, v) by
4.8 83

its exp!ícit expression taken from equation (4.93), we cau perí01 m the nolse average and
take the e = O limit. We obta.in the FP equation:

a
ijip (q,t) =
8
21 &q, [ 8 ]
íl&q, (e,.e,.P) +(f< - 2!1d; •• ) P . (4.98)

This re•ult. calls for two simple observations:

(í) The term quudratic in the noise in Langevin equation (4.93) can be replaced by its
average over v. (t). In the continuous time limit only equations linear in the nolse need
to be considere<!.
(li) In Section 4.7 we have considered the continuum L3llgevin equation as a limlt of a
discrete equation (equation (4.69)) v.'ith a different argument in e; 0 : we have taken the
symmetric form ~ [q(t) + q(t +c)] (Stratanovich convention). Ir we then expande,. for
c small we get:

fi
e;.. H [q(t) + q (t H)l} e;. [q(t)] + ~e;b [q(t)]vb(t)-. e,. [q(t)) + 0(.-:).
8 q,
(4.99)

This oorresponds to an equation (4.93) (ltô calculus) with a spec!fic form of the term
quudratic in the noise. Replacing in equation (4.93) the function d..b by:

1 {j
d..,. (q)=;e1.{q) -[} e.,. (q), (4.100)
q;

we reco,.,.- FP equation {4.73).

Finally from the Langevin equation which we now write without loss of generality in
Lhe small ~ limit:

q, (t +E) q;(t) - ict. (q(t)) +e,. (q(t}) v0(t), (4.101)

we can directly derive a patb integral representation in discretized form for P (q, t}.
We start from equation (4.96), use the explicit form (4.93) of the Langevin equation and
integnte over the noise, exactly as we baYe done in the oontinuum case, with the measure
(4.87). Assuming that the matrix e;0 is irrvertible, we then obt&.in tbe equation:

P(q,tH) =f dq'p(q,q')P(q',t),
with the transition kernel p ( q, q') now given by:

{4.102)

in wbich we bave set:

d q <Í + ~e:f (q'). (4.103)

This equation directly leuds to tbe discreti~ed form o f the pa.th integral ( 4.84).
84 Stochastic Dijjerential Equatiom 4.8

Bíbliographical Notes
A hístorica.l r<"ferenc:e is
P. Langevin, Comptes Rendus Acad. Sei. Paris 146 (1908) 530.
For an introduction into the subject of stocbastic procf!SSe.s in ph)'lSiCS see:
N.G. Van Kampen, Stocha~;tic Processe. in Physics and Chemislry (Nortb-Holland,
Amsterdaro 1981).
A selection of early papers can be found ín
Seiected Papers on Noise and Stochastic Pro=ses, N. Wax ed. (Dover, New York
1954).
Matbematícal contributioru; are contained in
New Stochastic Methods in Physics, C. De Witt-Morette and K.D. Elworthy eds., Phys.
Rep. 77 (19Bl) 121.
Examples of matbematical teJCtbooks are
I.L Gihman and A.V. Skorobod, Stochastic Differential Equation5 (Springer Verlag,
Berlin 1972); N. lkeda and S. Watanabe, Stochastic Differential Equations and Diffu·
síon Processea (Nortb-Holland, Amsterdam lg81);
K.D. Elworthy, Stochastic Differential Equations on Manifoids, Landon Math. Soe.
ucttJres Note11 10 (Ca.mbridge University Press, Cambridge 1982).
For an early reference about tbe a.pplication of path integra.ls ta brownian motion and
statistical mechanics aee for example
S.G. Brush, Rev. Mod. Phys. 33 (1961) 79.
Additional comments on numerical simulations and stocbastic proc"""es as discussed in
Appendix A4 cau be found in
S.K. Ma, Statiatical Mechanics, Ch. 22 (World Scientific, Singapore 1985); G. Paris\,
Statistical fuld Theory, Ch. 19,20 (Addison-Wesley, New York 1988) .
Exercises
E:rerciso 4.1
We wa.nt ta study the Langevin equation:
± ~ - !ax"(t) + v(l)
in which v(t) is tbe gaussian noise c011$idered in equation (4.3) a.nd a is a parameter.
4.1.1. Establisb coupled di!ferentia.l equations for the moments (xk(t)) using tbe me-
thod which leads to the Fokker-Planck equation. Deduce tbe pollllible form of the mo-
ments at equilihrium.
4. 1.2. Wtite tbe Fokker-Planck equation and study the existence of tbe equilibrium
dístribution.
4.1.3. What can be said if the pararneter a and :z:(!) are complex and v(t) is still the
real noise (4.2). Begin with n 1; for n > 2 this ís a research problem.

Exemse ~ .2
The goal of tbis exercise is to compare the properties of the ordinary La.ngevin equa-
tion with a second order stochastic differential equation. Consider tbus the dynamica.l
equation

ai + b:i: + c:r = v(t),

in wbich a, b and c are constants, a.nd v(t) still a gaussian wbite noise. Determine the
equilibrium distribution, the relaxation time....
 A4.1 8tochastic Difjerentia/ Equatitmli 85

APPENDIX 4
MARKOV'S STDCHASTIC PROCESSES: A FEW REMARKS

We here recall a few simple properties of Markov's processes because it is useful to have
them in mind when discussing Langevín or Fokker-Pianck equations.
We first assume that we have a finite number of díscrete states labelled by an index
a., a = 1 ... A. The stochastic process ís defined In terms of a matrix Pob which gíves the
trnnsition probability at any given time from state b to a:

p•• ;::: o. (A4.1)

(A4.2)

In terms of the matrix Pab we can write the evoh.ttion equation for tbe probability Pn(a)
to be at time n in state a:
Pn+J (a) = 2>••Pn
b
(b). (A4.3)

Equatíon (A4.2) expr=ses the collllervation of probabilities. Indeed summing equation

(A4.3) over the index a, as a consequence of the equation (A4.2), we find:

LPn+l (a)= LP.,(b) .

•
Equation (A4.2) also implies that the wctor "•'

Va = 1 Va,

is a left eigenvector of the matrix Pob with eigenvalue 1:

We cnll P( a) a.n equilibrium probability distríbution if it iB a right eigenvector of p,.b

with eigenvalue 1:
P(a) = LPobP(b), (A4.4)
b

i.e. a stationary solution of equation (A4.3).

í' A4.1 The Spectrum of the Transition Matrix

r*'' U.t us show that the largest eigenvalu"" of the matrix Pob have modulus L Let v. be an
~ eigenvector of P•• and ). the corresponding elgenvalm>:

2:: Pu~>llb = >.v• . (A4.S)

b

Note first that summation over a using equation (A4.2) yields:

86 Stochastic Díjjerential Equations A4.l

which implies either >. 1, or L V 0 = O.

Comparing the modulus of both sides of equation (A4.5) we a.lso find:

l:P•b lvbl?.: I:XIIv.l, (A4.6)

b

and thu.s using again equa.tion (A4.2):

(A4.7)

Equality is possible only íf ínequality (A4.6) is an equality for ail va.lues of a:

(A4.8)

which means that ]va\ is an equilibrium dístribution.

Let us now 8SSUtne that not ali cornponents of the vector v. can be made positive. We
then decompose the set of integers !1, A] into a union of subsets l; such tha.t within a
subset the components of "• ha"" a.ll the SBme pha.se and consíder tbe v..ctor w'
such
that:
(w'). "• for a E 1; ,
{ (w'). =0 otherwise. (A4.9)

Equality in equation (A4.7) implies th&t for a given index a, Pob ca.n only be di!Jerent
from zero for b belonging to a uoique subset I.. The con'"'n.e is also true. The ma.trlx
Pab describes transítíons ftom one subset to another.
(i) lf Pai> has non-zero ma.tri.x elements between a subset and itself:

Pab '/.O for a,be 1,,

then >. "' 1 and the space of sta.t.es is disconnected since sta.rtíng from the statcs belonging
to J, one always remains in 1;.
(ii) The m.a.trlx Pob is a permutation matrix between the subset.s which are organized
in va.rious clusters. The cotresponding eígenva.lues are roots of the identíty. If the space
of st..tes is connected, the matrix Pub corresponds to a cyclíc perroutation between ali
the su bsets !;. Thís is ele arly a highly non-generic case.
Connectillity assumption. Ftom now on we assume tb ..t the space of states ís conne.:ted,
i .e. that sta.rtlng from any state a there is a non•zero probability to reacll any other state
b. This means that there exists a set of states c1 ,.,,, Cr such that the product:

We assume in addition tbat we are not in the case of the cyclic pemmtation.
It then follows that >. "' 1 is the unique Jargest eigenvalue a.Dd the corresponding
eigensp.a.ce ha.s dimension 1 (if the dimeruion were larger than 1, one couid forro a linear
combination of eigenvectors witb components with non-positíve ratios).
Furthermore tbe corresponding elgenvector ha.s only strictly positive oomponents. ln-
deed acting Tf'peatedly with P~• on a ve<:tor wítb real non-negative companents, we
always obtain a.fter a finite number of lteratiom, as a consequence of the connectivity
A4.2 Stocha..Cic DifJerenbaJ EquatlrJm 87

sssumptíon, a veetor witb only strictly positiw components. If we act on ail equilibrium
distribution the result follows.
Therefore, as a comequence of the previous assumptions, any distribution converges at
time infinity towards the unique equilibrium distribution P (a) which can be parametrized
as:
P(a) =e-EI<>l with E (a) real. (A4.10)

Infinite num.ber of state•. If we let the number of states increase to infinity an important
new phenomenon may occur: the components oi the norrnnfued equilibrium distribution:

LP(a)=l,

m.ay ali go to zero.

In this case even with the previous assumptions there is no equilibrium distrlbution.
Thiz wíll be a.n important \ssue in realistic cases.
We shall also encounter another situation in the study of critica.! phenomena In Chap-
ters 23-35. In the case of ferromagnetic systelll5 tbe states correspond to configuratioru;
of spins on a lattice: as long as the spin system is in a linite volume, the space of states
remains connected, but the space becomes disconnected in the in.finite volume llmit. In
a linite volume the equilibrium state is unique but not In an inlinite volume.

A4,2 Deta.iled Balance

One is sometimes confronted with the following problem: one has chosen a príori an
equilibrium distribution P( a) and one wants to corllltruct a stochastic process wbicb
converges towards tbis distribution. This can be achieved for example by imposing on
the ma.tríx p.~ the condítion:

P•bp (b) = pr..P (a) for ali paira (a, b). (A4.ll}

This condition is called detailed balance. lt is a local condition involving only the states
a and b,
Let us show tha.t P(a) is an equilibrium distribut.ion:

P'(a} LP••P(b) = I:Pr..P{a) = P(a}. (A4.12)

b

In what follows we can assume without loss of generality that the space of states is
connected and that ali probabilities P( a} are thus strictly positive.
Let us then set in equation (A4.3)

P.,(a) y'P(a).P,.(a). (A4.13)

'fhen:
Pn+l (a)- L P (a)-l/Z PobP(b) 1~Pn (b),
1
(A4.14)
b

E;quation (A4.11) implies that the matrix JJ'.b:

P~• = p (a)-112 P••p (b}l/2, (A4.15)

86 Stochastic Ditferential E<JUatíons A4.3

is symmetric. lts spectrum is real. The distribution P(a) will be an equilibrium distrl-
bution except if Pab hos -I as eigenvalue. This means, according to prcvious discussian
that the space of states is dívided into two equal subsets J+ and J_ which the matríx Pab
exchanges.
1t is easy to see that the correspanding left eigenvector h as components +I for one
subset and -I for the others. This left eigenvenor has to be orthogonal to the right
eígenvector P (a).
This implies the condítion

2: P(a) 2: P(a).
12El+ aEL

This ís a non-generíc situation which is likely to be realize<! only íf there exists some
discrete symmetry in the BP"-"" of states.

A4.3 Continuum Time Lim;t

In various circumst.ances, in particulnr wben the matrix p4 b is clooe to the identity:

(A4.16)

equation (A4.3) ca.n be approximated by a dífferential equation of tbe form:

(A4.17)

The rnatríx Qob hos the properties

ll.lld LQab=O, (A4.18)

wlúch ensure that if initially P(to, o) is a probability dístTibution it remains one at ali
!ater times.
Results obtained in tbe discrete time case c11.11 be recovered by a.nalogous methods.
One first shows tbat the eigenvalucs of q0 b have a. positive real part. The eigenvalue
equation
(A4.19)

implies:
L<Jabt!b =(A -q•• )v.,
bf'o

and thus:
L Q.b lv~l ~ J>. q•• llvol· (A4.20)
b#•
A summation over a finally yields:

(A4.21)
A4.4 89
Stoch.Mtú: Differenli41 Equatitms

At least one term in this sum must be non-positive:

3a: q.h + i>.- q•• l ::; o. (A4.2:Z)
Since as a consequence of coudítions (A4.18) q_ is non-positíV<", the real pa:rt of >. must
also be non-positlve. Furthermore íf Re (>.) v!Ulishes, ). vanishes. This result shows that
equation (A4.17) wíth conditions (A4.18) is slightly less general that the discrete equation
(A4.3). Fínally if the space of states is connected by 9ab• the equílihrium YeCtor is uníque
and has strictly p<l5itive components.
Detailed balance. Deta.íled balance now implies:

M -~ (A4.23)
Qbo - P(b).
Therefore the matrbt q':
q~ 0 P(a)-11 2 q.oP(b)l/2, (A4.24)
is symmetric. It ha.s real, negative, eigenvalues. The defilútíon (A4.24) ímplies that the
matrkes q a.nd q' have the same eigenvalues, Thus 9ob a.lso has real negative eigenvalues.
This case is the díscrete analogue of the case studíed in Sectíon 4.4.
A4.4 Construction of a Stochastic Process with a Given Equmbrium Distri-
bution

The~e are many ways to construct stcchastic processes whích converge towa.rds a given
equílibrium distribution. Firnt one ho.s to construct the matríx Pab· Let us just give the
example of discrete set of states a.nd use detailed balance.
One possibílity is to connect ali initial states to the same number r of final states and
then take:
P<>h=;
l
iCP(a)?:P(b) andafb
(A4.25)
1 P(a)
Pao ; P(b) if P(a) < P(b).
These conditions imply detailed balance. In a.ddition as required:

Another ideais to take:

Pbl>
....
1-2:Pob>O. {A4.26)

(A4.27)
with
9.! llo.E{0,1},
and p adjusted such that for any state b
p2:P(a)9ab<l.
o#b
Dependíng on the structure of the space of states there are many other methods.
Once the matrix Pab ís gíven, it is easy in the discrete case to construct the rorre-
spondíng stochastic process, i.e. to describe a motion in the space of states such that
asymptotically at la.rge time the prohabi!ity of being at state a is just P (a). At fixed
initial sta.te b, to the ma.trix elements Pab corresponds a partition of the interval [0, 1].
By drawing a randam numher with uniform probability on (O, 1J, one can select the final
state a with probability Pab·
5 FUNCTIONAL INTEGRALS IN FIELD THEORY

In this chapter we begin our study ar local quantum field theory, the relativistic gener-
alization of quantum mecha.nics. We discuss the exam pie of a neutra! self-coupled scala.r
field ,P(x,t) which depends on time anda space coordina.te x belonging to Rd-i, We
ooostruct a functiortal integral representation of the imaginary time evolution O<' quan-
tum statístical operator, natural exteru;ion of tbe path integral o! quantum mechanics.
The functional integral defines a functional measure to which correspond field correlation
functioDS. By adding an externalsourre to the action we obtain an integral representation
of their generating functional.
As in quantum mechanics, ~ first calculate the gausoian integral, whlch corresponds
to a free field theory. Then, considering a general action wíth a.n interaction expandable
in powers of the field, we expresa the corresponding functional integral in terrns of a series
o! gaussiaD integrais, and establísh Wick's theorem. Thi.s provides an alternative, per-
turbative and a.lgebraic definition of the functional integral. We show, because thi.s is no
Jonger obvious, that thm definition i.s consistent with the usual manlpulations performed
on integrais like integration by parte and change of variables. We exhlbit severa! alge-
bra.ic properties of functional integrais and derive the Dy=n-Schwinger field equations
o( motion. We also define the fuDctional 6-function. We finally generalize these notíons
to spin 1/2 íermion fields.
In the AppendD< AS we recall a few properties of spin I /2 fennion tbeories, of the
spinorial representations of the O(d) rotation g:~oup and the a.lgebra of '7 matrices.

5.1 Functionallntegrals. Correlation Funetions

We consider a dassical lagrangian density l(.P) for "scalar field <1> of the form:

l(,P} = ~~2 (.:r,!)

2
HV4>(x,t}) - V{<!>[x,t)), (5.1)

in whicb V(</>} is a function expandable in powers of rf>, a polynomial in the simplest

examples !ih:

The pa.rameter m \s called the mass because for g = O it repr....ents the pbysical masa of
the particle a.ssociated with the field rf>.
The lagrangian density {5.1) is the simplest exaDlple of a lagranglan density having
tbe following properties:
(i) It is local in space and time because it depends only on the field .P(.:r, t) and íts
partial derivatives (and not on product of fields at different points). Tbls property,
locality, plays, as we shall see, a central role in mOBt of tbis work.
(ii) lt is invariant under space and time tra.nslations since space and time do not appear
explicitly in expre!16ion (5.1).
(ill) lt is relotivmtic invariant, i.e. invaria.nt under the pseudo-mthogonal group O(d-
1, I) acting linearly on x and t.
(iv) As we shall verify below, for a suitable da.ss of potentiahl V(4>), it leads to a
hermitian quantum hruniltonian bounded from below.
5.1 Functiona.! lntegraJs in Fie!d Theof'li 91

5.1.1 QuantizatiDfl. F\mctional integro/

'Ib quantize the theory we first have to cak.ulate the hamiltoni11n density corre"ponding
to the lagrangian density (5.1 ). The (agrangillll and the hamiltonian are related by a
Legendre transformation blvolving J, a.nd 1r(:r) oonjugate mornentum of 1/>:

1{ (1r,<f;) + qJ,, rp) 1r(:r)J, (:r, t) o. (5.2)

The relation between .,. and J, is obtained by expressing that the l.h.s. of the equa.tion is
statíonary with respect to vatíatinns of J, at ,. and <f; fixed, or equivalently with respect
to variation.s of 1r at J, and <f; fixed:
f)[. • [J1t
1r(x) = - .-
01/J(x,t)
= ,P(:r,f) = "--( )'
vn X
(5.3)

We emphru;i"I.e the concept ofLegendre transformation because it also plays an important

role in the theory of renormaliza.tion through the introduction of the generatíng functional
of proper vertices (see Section 6.2 and Chapter 10) a.nd in statistical rnechanics (it relates
ftee energy a.nd thermodynamic potential for example).
The total hamiltonian is the integTal of the hamiltonian density:

(5.4)

The coordinates q; of quantum mechanlcs are here replooed by the field ,P(x). The
transition from quantum mechanics to field theory can be undeJ-stood in much the same
wa.y e.s the transltion betwoon the discretized a.ction (2.16) and the oontinuous time
version {2.17).
To quanti'l.e the hamlltonian H we can start from the basic commutatíon relations
between operators:
[~r(x),<f;(x')] = ~ód-1 (:r :r')' (5.5)
I

and then develop a quantum theory UBing the standard methods of quantum mechanics.
We instead work immedia.tely with the formalism of functíonal integrais, whicb generalize
the path integrais introduced in Cbapter 2. We verify in Sect.ion 7.6 that this quantum
field theory leads to a qnant.nm theory of partícles in lhe non-relativistic limit.
lAJt us write explicitly the hamiltonio.n densíty corresponding to the lagrangían (5.1 ):

(5.6)

This hamíltonian lw.s an important property: it is quadratic in the momentum >r( X). We
can therefore follaw the method explained in Section 2.2 and obtain a functional integral
representation for the evolution operator (in imaginary time) U (t2, ti):

{1/>2 lU {t2,t.)] oj>1) j [d.P (x,t)]exp [-S (.PJ], (5.7)

with the boundary conditions ,P (x, t,) I/J1 (x), rf>(x, t2) = 1/>2 (x).
lt iB important to note tbat states in the configuratíou representation are defined by
the descriptiou of a complete da.ssical fieid, rf> 1 (:r] or rf>2(:r) in expression (5.7), which
corresponds to an lnfinite number of usual variables.
92 Fund•onal Integrais in Field Theory 5.1

As we know from the analysis of Sectíon 2.2, S(</>) is the classical euclidean action,
0btained by íntegrating the analytic continuation to imaginwo· time of lagrangian density
{5.1):
(5.8)

One advantage of the imagínary time formulatíon, obvious in expression (5.8), is that
time and space now play an equivalent role (up Lo possible boundary mndítions). In
particular the original non-cornpact O (d 1, 1) pseudo-orthogonal syrnmeuy, is replaced
by the compact O(d) orthogonal symmetry.

5.1.2 Regularüalion
We have already seen that path integral!; reduced to their formal definition in continuouo
time are sometimes ill-defined and that in sm:h a case it is necessary to complete the
formal definition with a limiting procedure based on a time dis<:retized version. We ~h ali
eventua!Iy find out that this problem is even more severe in field tbeories. This will
be the subject of many chapters. Let us here note only that we may define functional
integral (5. 7) as the limit of an integral in which both time and spa.ce are discretized.
Let us introduce a hypercuhic lattice zd, with lattioe sp,.,ing a, and u>e e.s dynamical
variables the values of the field on ali lattice sites. We can replace t.he field derivatives
a,..p by:
</>(r)], (5.9)
in whir:h p. and r rt!fer both to space and Ume and n,. is the unit vector in the p. dírection.
The regularized euclidean action S. (<f;) is then:
8 0 (</J) ad L {HV',.</>(r)[ 2 + V(çll(r))}, (5.10)
rE6L!t 1 ~,j

in whír:h L(t 1 ,t,) is t.he subset of 'l,d such the time coordinate belongll to [t,t,].
In expression {5.10) it is clear agaln that the potential and the gradient squared (ki-
netic) term play diffeent rOles: Tbe potentíal term V (</>) is ultra-local, it weight;s the
functional integnl accórding to the values (I[ the lield at e&r:h point independently; the
gradient squared term !nstead suppresses the lields whícb are toa singular in the small
lattioe spacing llmit a - O, those for which tbe quantity:
\1/>(r+an~)- ,P(r)i 2 ad-O
diverges.
However, tbis condition becomes weake wben the dimension increases. lt implieo
continu;ty only for d < 2, i.e. quantum mechanics. This feature has deep conoequence<.
As we shall discuss !ater, for d ? 2 the continuum limit does not exist in general.
Statistical mechanics and the theory of phase transitions provide a natural framework to
discuss this problem. Indeed the díscretized action can be considered as the configuration
enegy of a lattice model in cla.ssical statistical mechanics. The infinite f3 t2- C1 {zero
temperature) limit corresponds to the infinite volume of the classical model. Only if tbe
model has a oontínuous phase transition, cana continuum limit be defined. However,
even then, tllis continuum limit exists only when one parameter of tbe theory, for example
m 1 tbe second deriva.tive of V (I/>) at its minimum, is taken close to wrne special val.ue. In
the corresponding statístical model the ternperature is dose to the <:ritical teroperature.
Various aspects of tbis problem will be discussed in Chapters 8-10 and in tbe chapters
devoted to critica! phenomena (Chapters 23-36).
5.2 Functional Inlegmls in Field 1'/laJry 93

5.] .9 Correllllion functóons, genemting funetíonal

The regularizatíon pmpm;ed in Subsection 5.1.2 •haws that the zero temperat ure quan-
tum lield theory can be considered as a formal continuum Jimit of a d dimensíonallattice
model in dassical statistical mechanics. Lattice variables correlation functions have a
formallimit which we therefore also call if>-lield cottelation functions. This denomlnation
is consistent wíth the property that the lotegnwd of functional integral (5.7) defines a
positive measute:

(5.11)

in which Z(B) is the quantum partition function tre-~H_ Equal-time correlatíon func-
tions are also the static correla.tlon functions of the finite temperature quantum field
theory. To obtaín the time-dependent quantum correlation functions one bas to ptoceed
by analytic oontinuation tcrwards real time in the argument of r/> but not in {3.
In the large 13-limit these finite temperatute correlation functions become the ground
state (usually called uacuum in quantum field theory) expectation values of the quaotum
field operator time-ordered products, as we have shown in Section 2.5. After analytic
continuation towards real time, they yield the Green's functions from which S-matrix
elements can ha calculated.
Usiog the property:

(5.12)

where J(x) is an externai field or source, we V<erify that q,..field correlation functions can
be obtained from a genemting funct;ional Z (J)

Z(J) =f [dl/li exp [-S(.p) +f ddx J(x).P(x)] , (5.13)

by functional dilferentiation:

(5.14)

5.2 Perturbative Expansion of Functional Integrais

ln Section 2.4 we have shown how to calculate a pa.th integral for a hamiltonian of the
form tp 2 + !w 2
l + l-'J(q) as a seties expansion in powers of l-'J(q), fur any function
H(q) expaodable in powers of q. The result was based on the calculation of a reference
gaussian integral (in Chapter 2 the harmonic oscillntor). We hen! uS<! the same method
for functional integrais.
In the remaíoing part of this chapter we work directly at zero temperature in infinít.e
Rd-l space, and therefore lo infinite Rd euclidean space: correlation functions therefore
cotrespond to vacuum expectation values. As shown io Section 5.1, the <!uclidean action
ís then fully O(d} invariant and therefore we no longer distinguish space and time; x
be!ongs to Rd.
94 FuncticnaJ Integrais in Fie!d Themy 5.2

Furthermore, although ali results derive<! in this ch&pter wi!l be íllustrated "~th ex-
amplas corresponding to an actíon of the form (5.8), these results are more general and
tbis exp]ains the abstract notations used below.
Tho gaussian integraL Let us con:sider a general gaussian functional integral:

(5.!5)

where we bave used the symbolic notations;

We assume tbat tbe kernel K is symmetric and positive. ln expressíon {5.15} a normal-
ization i• implied, we have dtoaen Z 0 (0) = 1.
In the example of a free massive thoory with mB>;S m, which eorresponds in expression
(5.8) to the choice

the kemel K is a local operator:

(5.16)

2
where we denote by 8 the laplacian io d dimensioos.
Let us introduce ti, the írrverse of K:

(5.17)

lo tbe example (5.16) ti takes the form

1
ti(x,y)=--4
f d e>P·(,.-y)
d P~· (5.18)
(2") ., +m

To calculate the functional integral (5.15) we simp)y translate rf> by t;.J and find after
integrationi
Zo(J) = exp ( !J t!.J) , (5.19)

with again the convention:

From the identitics derived in Subsection 5.1.3, we see that Zo(J) ís a.lro the generating
functíonal of correlatlon functions corresponding to a general actíon quadratic in the field
rf>. The inverse kernel ti is thus the 2-point function of the quadratic tbeory or Jree fidd
the<Jf11
t;. x -
( 'y)-
ó2Zo(J)
óJ(x)~J(y)
I
J-o ·
!..3 1'\mctionr>l lntegrals in Field Theory 95

Perturbation thwry. We shall verify in the ooming chapters that interacting theories
with a propagator of the form (5.18) have large momentum or short distance dívergences.
Therefore in what follows we assume either that the field theory has been regularize<! by
replacing continuum space hy a lattice as in Subsectíon 5.1.2 or the propagator t. has
been replaced by a more complícated funct.íon wh.ich ensures the convergence of ali tetms
in the perturba.tlve expa.nsion (see Chapter 9).
Let us now oonsider a more general euclidean actíon of tbe form:

(5.20)

Using the property (5.12)

"
-.-é·~=
6J(x)
4>(x) é·'f.
'
we can express the functional integml

Z(J) =f ld.Plexp 1-S(rp) + J · <t>J, (5.21)

In terms of Zo(J) "'"

Z(J) exp [-VI (.sJ~x))] Zo(J) = exp [ -Vr (.sJ~x))] exp (paJ), (5.22)

expression analogous to (2.50). Note that in this chapter we use the oonventíon that a.
differential operator like 6jóJ acts on ali fectors placed on ítG right.
We can then combine identities (5.22,5.14} to calculate a.ll </>-field oorrelatíon functions
... a formal series in powen; of the interaction term Vr: (</>).
We see that perturbation theory invnlves a. basic ingredient: the 2-point function t. of
the quadratic theory {equation (5.17)) which we ca.ll the propogator.

r..s Wick's Theorem and Feynman Diagrams

When the interaction terms are loca~ for example integrais of polynomials of the field
.~(x), it is possible to give a. graphical representation of each term of the perturbative
expo.nsion. Indeed a contríbution to the n-point correlation function is a gaussían average
form:

( 4>(xt) ···<P (xn) f ddy 1 </>"'(!I!) f ddy, <!>""(V>) ... f ddy* <!>"' (!lk)) I>

c [ 6J é(Xt) ... 6J (xn)

6 f d d !ll ( 61 6(y.) )"' .. • f d d!ik ( óJ 6(Vk) )p' I
exp bJ t.J) ] IJ,.,'
which the < >o means average wíth respect to tbe gau&Siao mea.sure.
Section 1.1 we have already dlscussed such expressions. A straightforward general-
ar equations (1.6-1.9) yields Wick's theorem in field theory:

( TI<t>(z.))
1 Ct
= [iit;J~z.;)exp(!JM))l
t=l J=O

L l1(z;,z;,) ... l1(.•••·->'z,.,). (5.23)

ali po58ible
~ngr;of
{1,2,....2•]
!)6 5.3

For exarnple, for t 2 one finds:

rnl1=1 bJ ~z;) exp(~JL\.J)1\

J-0
Á (z,,z,)Á(z,,z.) +Á (ZJo2a)A(zz,"4)

+ Á(z,,.::,) Á l•2, za)

Graphlcally each term in the sum can be represented by a set of contra.ctions correspond-
ing to the particular pair\ng chosen:

ú] ~ (z,)) =
0
~ (zz)~ (z,) + 2 terms.

Any perturbatíve contributí<m ís a sum of products of propr.gators integTated over the

points corresponding to intera.ctíon t.erms. We can give a. graphlcal tepresentatíon of
each product: a propagator is represented by a segment joining the two points which
appear as arguments; moreover any point which is common to more than one segment
corresponds to e.n argument which has to be integra.ted over.
Example
Vi{~) !!_
41
fd"x ~ (x).4

The twn-point function has the expansion:

I 92 92 92
(<f;(x,) ,P{x 2)) =(a)- 2y(b) + 4 {c)+ 4 (d) + 6 (e)+ O (l)

(a) is the propr.gator:

(a): Xt------------X2

(b), the Feynman diBgram whlch appearn at arder g, i.s represented in figure 5.1:

[b)
Fíg. 5.1 2-polnt function at order g.

and (c) ...{e) are represented in figure 5.2:

(c)
8 (d)

Fig. 5.2 ContribuüoDS of order g' to the 2-po::=;:*

{ej

,,:~
5.3 Functianal Integmi.J; in Field The.ory 97

Let us expla.in for exa.mple in detail the weight 1/6 in &ont of diagram (e). We have
to calculate the gaussian average of:

and we apply Wick's the<>rem.

First 4> (:r 1 ) ce.n be associated with e.ny ,p Jield of the interaction terms, there are 8
choices and one interaction term is distinguíshed. Then r/o (:r2 ) must be contracted with
a lield of the remaining interaction term: 4 choices. The 3 remaíning fields of the first
interactíon term can finally be pa.ired with any permutation of the ftelds of the second
one: 3! equivalent possibilities. Multíplying ali factor5 one finds:

~-1 -x8x4x3!
2 {4!)2 6
Let us also note that the factor 1/6 in &ont of the diagra.m can be shown to have
an interpretation as 1/3!, the combínatorial factor in the denomínator reflecting the
permute.tioo symmetry between the three !ines joining the twn vertices. There exist"
~rnatic expressions giving the ""'igbt factor of such terms, called Feynman diagra.ms,
in tenns of the symmetry g:roup of the graph.
A useful practical remark is the fullowing: the sum of all wcight factor5 at a given
arder can be calcnlated from the "zero dimellilional" field theory obtained by sUpprf$$ing
the arguments of the field and all derivatives and integ:ration in the action because the
propagator can tben be normalized t() 1. For exrunple in the case of the 4>4 field considere<!
bere the action beromes:

and the two-point function is given by:

z<•J- fdq,rp•exp{-S(q,}j -1 g 2 2 { ')

- fdq,exp{-S{<f>)] - 2 + 39 +0 g '

in which tbe expressions correspond to ordinary o~variable integrais.

The sum mles are sa&fied. For exa.mple a.t arder 1i:

Let us now give the expansion of the 4-point function:

(I} (x.) <f> (x2) r/o (xs) 4> (:r 4)) = l(a)n (a)34 + 2 te!1Il8] ~ I(ah2 (b)s,

+5 ter=) g (f)+ g2 { (ah2 [~ ({c) 34 + {d)34 ) + ~(e) 34 ] + 5 terms}

~
+4 [(b),2 (b)a•
i [(g) +
+ 2terms] + "2 3 terms] +
~
'21(h) + 2 temm] +O (i).

The new diagra.ms (f), (g), (h) are represented In figore 5.3. The notation for exe.mple
(a) 12 means the diagra.m (a), contributing to the 2-point functi<>n, with a.rguments :r 1
9S Functional Integrais in Field Theory 5.4

::x:: (I)
Fig. 5.3
X> x=x (g)
New Feynman diagramo contributíng
(h)
w the 4-point fWlctíon.

and x 2 • Finaily the additional terms are obtained by exchanging tbe externa! arguments
te rest.nre the permutation symmetry of the 4-point function.
Note that the graphs which ínvo!ve the 2-pomt functions are disconnected i.e. factcrize
into a. product of functions depending on disjoint subsets of variabi€S. We shall discuss
this question in more details in the next cha.pter.
On the other hand we have omitted systernatioally disconnected día.gra.ms in whkh
one factor had no externa! a.rguments. M shall become clearer in the next chapter these
diagrams are cancelled by the perturbative expansion of Z (J =O) in expression (5.14).
The diagrams contributing to Z(O) are called va.cuum diagz-mns.
A final rema.rk: local ínteraction terms may aloo involve derivatiws of the field .p(:r).
Then in expression (5.23) derivatives of the propr.gator appear. The repr€S€ntation in
tenru; of the Feynman diagrams given above is no longer faithful since it does not indka.te
where the derivatíves are. A more fa.ithfu\ representation can b" obtained by splitting
póints at the vertices and putting arrows on lines.

5.4 Algebralc ldentities and Functional Integrais

F\mction.al ínt"!!mls: perturootí,. definition. Th avoid the difficulties related to tbe pre-
cise definítion of the functional integral, we can tal<e expression ( 5.22) as a proper, al-
tllough perturbative, definltíon of the functional integrnl. In particular we may then
avoid tbe reíerence to a discretized spa.ce and time. However with this new definition,
we have to prove that tbe usual trans[ormations performed on standard integrais lead
t.n ident.ities which are also true with definition (5.22). This is one of the purposes of
this section. At the same time we shall derive useful identities satisfied by correlation
functions like Dy&m-Schwinger equations of motion. Finally, through the formal ma.nlp-
ulatíons invo\ved in tbese derivations, we shall beoome more fanúliar wíth the algebraic
properties bf the perturbative expa.nsion.
We therefore now assume that tbe functíonal integral (5.21) with action (5.20):

Z{J)= /ldcP]..:xp[~!t/lKtf; VJ(if>)+J if>), (5.24)

is really defined by expressíon (5.22):

Z(J) cxp [-H (/iJ~x))] exp (!JilJ).

We assume of course that ali terms in the expa.nsion of this expression in powers of VJ
exist and {>l'e 6nits (which, e.s we shall see !ater, ímplies some condítions for the kernel K
and the intera.ction term 1-i(<Pll· From the functional integral representa.tion (5.21) we
now heuristically derive identities satisfied by the genera.ting functional Z(J) and then
prove algehrokally tha.t there identitíes aloo follow frarn the definition (5.22).
'
Funct•onal lntegmls in Field Throry 99

Ii 5.4.1 Integrotion i>y parts and equation oj motion

The integral of a total derivativt< vanishes. Let us use this remark
an equatíon [or Z(J):
to derive heurístically

f [d,PJ _!_) ex:p(-S +1 ·<;\)=O.

6</J(x
Let us write this equation more explidtly:
(5.25)

f !doi>] [1 (x) - _!_§___]

Ór/J(:r)
exp ( -S + J ·1>) =O. (5.26)

As a consequence of identity (5.12) this equatíon can be tra.nsformed ínto:

[ó!fx} (6~)- J(x)] Z(J) =O, (5.27)

which for tbe explicit form (5.20) of S (<I>) yields:

J{:r)] Z(J) =O. (5.28)

Eqnation (5.27) or (5.28) is the equa.tion of motion or Dyron-Schwinger equation of the

quantum field theory. lt ls eqnivalent to an infinite set of relatlons hetM~en correlation
functions obtaineà hy expanding in powers of the source J(x). These equations, in
tum, can he so)ved perturhatively, and, as we now show, determine correlation functions
completely.
Solu!'ion of the equatíon of motion. Conversely the solntion of equation (5.27) is the
functional int;egral (5.21). To prove this ststement let us write Z(J}, the solution of
equation (5.27), as a generalized Laplaoe transform:

(5.29)

Equation (5.27) implles:

(5.30)

Using equation (5.26) but with S replaced hy S we transform this eqnation into:

Therefore:
S(,P} S(~)o=const .. (5.32)

The constant is fixeà by the nonnalization of Z(J).

Example. Let us again consider the euchdean sction:
100 Hmctiona.! Integrais in Field Theory 5.4

Equatíon (5.28) in tbís case reads:

Let us wríte the lirst equations obtained by expanding in powers of J(x). Dífferentíatíng
once witb respect to J(y) before setting J to zero we find:

Differentiating three times we obta.ín

(-i::t; + m 2
) (</>(:z:)</>{11! }q,(y,).j>(y,)) + t (1>(:z:)'~'ÚI1)1>(y,)1>(!13))
= .l"(:r y!) (1>{y,).j>(y3 )) + 2 terms.

More generally these equations relate the 2n, 2n + 2 and 2n + 4- poínt functions. They
ca.n be solved by expanding ali functíoDS in powers of g a.nd tbis ieads to perturba.tion
theory. One could imagine sohring these equations ín a non-perturbative way, but thís
requires truncatíng the inlinite set. Examples of such approxímations can be found in
tbe exercises of Cbapter 9.

5.4.1< Direet algebraíc proof oj lhe equati<m Q/ moti<m

Let us now show by purely a.lgebraic tra.nsformations tbat Z (J) defined by equation
(5.22) indeed satisfies the equation of motion (5.28). lt is here con\'ellient to introduce a
notation: when 6f1JJ appears as an arguroent wewill denote it by DJ:

We want thus to prove algebraically

The proof is based on the commutation relation:

(5.33)

which can be easily verified by expa.nding F in powers of éf6J. If we take for F(</>)

F{1>) = exp[-VJ(4>)],

we obtaín:
(J(x),exp(-VI(D..T))J"' ô!~)(DJ)exp(-l-í(DJ)). (5.34)

Using this equation in expression (5.22} we obta.ín:

J(:z:)Z(J) /\1 )(DJ)Z(J) + exp í-VI (DJ)]J(:z:)exp (lJ~J).

u<f>(:Z:
(5.35)
5.4 R.nctional lntegrnls in F:ield Theo111 101

Equation (5,28) then relie:; on the identíty:

exp [-Yt (DJ)]J(x)exp (Vt.J) =f dyK (x,y) 6 / exp [-V, (DJ)]exp {pt.J) ·
(y) (5.36)
which follows from:

f dy K (x,y) /(y) exp

6
(Vt>J) - J(x) exp (!Jt.J), (5.37)

since by definition of t.

f dyK(x,y)t.(y,z)- 6(x-z).

5.4 .9 The infiníte.simal chrmge of vllrinblell

Tb prove various identities about correlation functions in the case of field theories pos-
sessing symmetries, we often shall use the metbod of the Infinitesimal change of va:rio.bles.
We set:
tfl(x)- x(x) +cF(x;x), (5.38)
in wbich F(x; X) is a functional of x:

F(x;x) "" l f d)l! ... dlinX(!IJ) ... x(!ln)F{n) (:r:;y., ... ,!In),
=L--; (5.39)
n.
I

and E an infinitesimal parameter.

l.et us lirst calcula.te the va.riation of the a.ction S (tfl):

(5.40)

At the same time the chMge of va:ria.bles (5.38) in the functlonal integral generates a
ja.cobian D:
6,P(x) [ ) liF(x;x)]
D = det éx(y) = det 6(x y H õx(ll) . (5.41)

As a consequence of identity ( 1.33):

det(l+.,M) l+ctrM+O{c~)

a.nd thus:
(5.42)

it follOWE:

Z(J) f
f[dxl(l+E dx ~~~;~))
6
[1-e f dxli~~)F(x;x)
H f dx J(x)F (x; x)] exp 1-S (x) + J · xl +O (E2). (5.43)
102 Funetianal Integrais in Field Theory

Collecting a.ll t.erms of arder c we find the identity:

r
tt
o. (5.44)
..'·
Algebrnic proof. To prove algebraica.lly this identity, we start from tbe equation of
motion (5.27):
o,
and act on the left witb the dífferentia.l operator J dx F (x; ó/61). We tben use commu-
tation relation (5.33) under the fonn:

6F
\F(y;DJ),J(xJ] = éx(x)(y;DJ).

Equation (5.44} follows immediately.

Note that we have dealt in this subsection with an infinitesimal cltange of variables at
first arder ín e, a.lthougb tbe algebraic proof can be extended to ali orden; in E.

5..j.,j ln1!ariance oJ the funél.irmal integro! with nspect to lhe cluríel! of the quadrotíc part
L€t u.s decompose the kernel K in expressi011 (5.20) in to a sum of two terms:

with K 1 >O.

We want to prove algebraically that:

e.xp(-VJ(DJ) !D;K2 D;] exp (!JK,-' J)

=JV(K,, K2)exp H-i(DJ))exp [!J(K, + K,)- 1 ; ) ' (5.45)

where )I{ is índependent of the source J.

Sínce the operator exp [-V. (6/61)] can be Iactorized on both sides of equation (5.45),
it is sufficient to pr~ the identity for V1 O. Let us calculate the quB.Dtity

f dy K, (x,!l) bJ~y) exp (-~DJK2DJ) exp (!JK!'J)

=exp (-~DJK2 D;) J(:r)exp (pKj 1
J).

We again commute J to bring it to the left. The r.h.s. then becomes:

Comparing the initia.l and final expressions, we find that the functional Z 0 (J):

satis6es the equation;

[! dy (K 1 + K2)(:r, y) ~J~y) - J(:r]] Zo(J) =O. (5.46)

5.4 Rmctional]ntegmls in FieJd Thwry 103

lntegrating the equa.tion we thus find:

proving identity (5.45). After some edditional algebra one vermes that N2 det{l +
K2K) 1 ).
In the same way one can show that a part of the wurce term can be treated as an
interaction, without changlnr; Z ( J). The result follows from the identity:

€XP [! d:rL(:r} óJ~:r:}] F(J) ~ F(J +L).

Of course all these identities make sense only if both side5 exist.

5.,(.5 The functional Dírac 6-function

Let us consíder a multi-component field q,,(x), i ""1, .•. ,N, satisfyíng a constraínt:

F(x,<;!J)=O. (5.47)

We assume that it is possihle to solve the constraint and calcula.te one oomponent, for
example </>N(x) as a formal power series in the remaining components.
We then define the functíonal Dira.c 6-functlon by:

6 [F]= f [dA( o:)] exp [! d:d(z)F (x, <;!>}] . (5.48)

This ís a generalized Fourier representation. We have on purpose omitted the fartar i

in tbe representation beeause the form of the contour of integration in A(x) is irrelevant
for the algebre.ic CoDSideta.tions which follow. Also for reasoJIS explained before we have
omitted WJ)' normalization factor.
Let us now show that representation (5.48) has the properties expet'ted from a ~­
function. Let us CO•tsider a functional integral in which we wiah to restrict the iutegration
lield <;!>to the manifold F (x,.p) =O;

{5.49)

We use the representation (5.48) to write Zp as;

N
Zp(J) =f [dÃJTI [d</>;jexp[-S(<J>) +À F(<J>) + J,q>;] (5.50)

·-·
We now assume tbat the quedratic part in the fields (</>;.À) in tbe total actíon is non-
singular in such a way that a.fter adding a source term for ,\(x) we ca.n use the equivalent
of equation (5.22) to define algebraically integral (5.50) (it is suffident in particular for
the quadra.tic part of S( 4>) to he non-sínr;ular). Then if we add a term proportional to
F(:r,t/>) to S(<J>):
S(tf>)- S(t/>) +f IL(x)F(:r,t/>)dx, (5.51)
104 FuTldional lnte,qroi, in Field Theory 5.&

we can compensate this change b)' a clumge of variables:

À( :r) - À(x) + p. (:r). (5.52)

We have actually proven invariante by change of va.riables only for the infinitesimal
change of va.riables, although for sirnple translations llke in (5.52) it is easy to genetaliz~
the result to finite cbanges.
Let us liually note that with the lield À( :r) is associated a sirnple equation of rnotion.
Since:
S(rf>)] =0,

we obta.ín:
F (x; ó~) Zp(J) =O. (5.53)

This equation expresses the constraiut F = Oon the generating functíonal of correlation
functions Z(J).
E:w.mple$. In Soction 14.9 a functional ó-function in the forro (5.48) will be used to
express the non-Jinear a model in terms of 11 N-vector lield rf> satisfying an O(N) inYariant
constraint <J}(:r) 1.
ln Chapter 18, a cov.uiant gauge will be mentioned, which consists in imposíng to the
gauge field A~ tbe condition 8,.A,. = O. Aga.ín tbe representatiou {5.48) is usefuJ. The
equation (5.53) then becomes U:,!JZf!JJ,.(:r) =O, where J,. ú; the gauge field source.

1>.5 Quantum Field Tbeories with Fermions

Although we shall mostly distuss boson !ield theories, occasíonally ""' shall be led to
consider theories with fermíons, when the fermiooic nature of lields plays an essentíal
role. ln Section 3.6 we already have disct=ed the path integral rep,.,.,ntation of the
evolution operator for a system of fermions in quantum mechanics. Here w.; generalize
the formalism to quantum field theory a.nd present the basis of perturbation theory from
the point of view of functional methods.
A deep consequence of locality, hermitkity of the hamiltonian and relati"istic inv.ui-
ance is the spin-statistics connoctiou: bosom must have ínteger spin, i.e. transform under
representatiOilS of the O(d) group, while fermions must have half-integer spin, i.e. traJ:&
form under spinorial representat.ions of O(d). We comment ín Soctious 7.5, A7.2 on this
relation which can be proven rigorously with a great deal of genera.lity. We o.lso show in
Section 7.6 that the fermions we íntroduce below, lead in the non-relativistic limit to a
theory of fermion quantum particles.
Since the spin structure leads to additíonal technical complicatious, we recai! in AI>-
pendix A5 the essential properties o{ Dirae fermions in tbe euclidean representatlon and
expla.ín our notations ll.lld c:onventiom.
The gaussian integro.!: perturbation th.eory. We first consider a ganssiM integral with
externa.l sources:

Zo (1), '1) = j [d>t•(:r)d,P(:rJ] exp [f ddx (,P(:t) (;i!+ m) ,P(x) + i](:r)t,(l(:r) + ,);(x)IJ(:r))) ,
(5.54)
5.5 105
Fbnctional lntegrnls in Field Thoory

in which >b, 11·. '1 and Tj are elements of a Grassmann algebra and the fermion integralion
rules have been defined in Section 1.8. This functional integral can be calculated by
shifting variables:
w(:r) + (;J + m)- 1 1J(:t) '1/J'(:r) (5.55)
{ ~{:r)+fi(:r)(~+m)- 1 =·Jl(x).

Normalízing the functional integral (5.54) by Z(O,O) l, we obtain:

(5.56)

in whkh the fermion propagator ~F is given by:

~F(!I,X) (5.57)

We""" that on mass-shell (p' -m~), m~ip becomes h projector on a space of dimension
2ia/ll- 1. This rellects tbe property tbnt physica.l massive fermiou states can be dassified
according to the spinorial representation of the O(d- 1) subgroup oí O(d) which leaves
the momentum p invarlant.
The fermion 2-point correlation function in a free theory is then:

(5.58)

lt is now possible to give a perturbative expansion for a functional integral of the form:

Z (íj,fl) =f (d,P(:r)d~(.r)] exp[/ d".r (~{x)(;J + m).p(x)+ V (,p, ..b)

H(x)w(:r) + ~(x)"l(x))]. (5.59)

The functiona.l Z (íj, '1) is the gener-ating functiona.l of 1/J, ~ field correlation functions
since:

TI-TI
n é n Ó
-_-Z(f!.ii)
<=IÓTJ(X;) i-l Ófi(Y;) '7"'~-il
I Z (O) [H((~ (x1) .. -~(x.,) ,P(yl) ... ,p (y.,))j.

BecaU!>e the sources are Grassmann variables, the correlation functions are antisymmetric
in their a.rgument.s, and this is consistent with Fermi-Dira.c statistics.
Generalízation of equation (1.92) then yields the identity:

(5.60)

This result leads to the pertur bative expansion of a field tbeory with self-ínteracting
fermions. The corresponding Wick's theorem is a simple generalízation of equation (1.96).
1()6 Functionul Jntegrol.o in Field Theory r
Theories with basom and /ermioru. Most field theories with fermions contaín both
ferroíons and borons. A typical example is:

2 ('i, TJ, J) =f [d.pd~d4>] exp{ -S (.jj, 1/!, <!>}

+f ddx [iJ(x).p(x} + ~(x)17{x) + J(x),P(x)] }• (5.61)

in which 1/!, ~. 1), ij are CrBSStrumn variables and </> and J usual commuting variabies,
and in which the action S (~. 1/!, q,} h as the form:

S =f d"x { -,j;(x) li~+ M + gf#(x)] ,P(:t) + ~ (B~</>(z)) 2

+ ~m2 .P2 (x) +V (,P(x))}.
(5.62)
ln equation (5.61) Z (ii, 1), J) is the generating functional of both <f> field nnd .p, .jj lield
correlation functions.
Perturbation theory is generated by:

Z (i'j, 1), J) exp [-f d•x (V (.'íJ~xJ + g éJ~z) 611~x) ófí~x))] Zo (ij, '1, J), (5.63)

in whirh Zo (ij, '1, J) is the product of free fermion and free boson functionals.

AI! algebraic transformations we have perfonned on expression (5.22) can easily be gen-
eralized to the represente.tion (5.63), in pertkula.r field eque.tions can be derived or in-
finitesimal change oí variables justified in perturbation theory. Morrover a functional
6-function can be delined for femúons.
ln the exa.rnple (5 .62) the integral over fermions is gaussie.n. Therefore we may integrate
over fennion fields to generate e.n addítional non-Iocal ~field interactíon

/fd.p d~J exp f ddx [~(x) (/1 + M + g<f>(x)) ,P(x) + ij(x).p(x) + .)i(x)T)(x)]
e< exp [-SF(,P, '!, ií)J, (5.65)

with:

The expBDSion of Sp(,P,O,O) in pCJWers of 1/>, generate!l a set of one f~rmion loop Feynman
diagrams (see Section 5.3). A similar integral uver boson lields would have generated
a contribution of the form + tr In. Hence, compareci to boson ioops, fermion lOóps are
affect.ed by an additional minu.s sign.
5.5 Functionallntegrnls in F.idd TheoT'!J 107

Bib!iographical Note><
Covariant perturbation v.•as developed in;
S. Thmonoga, Pro.q. Theor. Phys. 1 (1945) 27; J. Schwinger, Phys. Rev. 74 (1 948) 1439;
75 (1949) 651; 76 (1949) 700; R.P. Feynman, Phys. R.ev. 76 (1949) 769; F.J. Dyson,
Phy•. Rev. 15 (1949) 486, 1736.
For Wick's theorem see:
G.C. Wick, Phy5. Rev. 80 ()950} 268.
The discussion of the perturbative definition of the functional integral ís taken from
A.A. Slavnov, Theor. Math. Phys. 22 (1975) 123; J. Zinn-Justin in :rronds in Elemen-
tary Particle Phy&ics (Lectures Notes in Physics S7} Bonn 1974, H. RDllnik and K.
Dietz ed•. (Springer-Verla.g, Berlin 1975).
The field equations of motion appear in
F.J. Dyson, Phys. Rev. 75 (1949) 1735; J. Schwinger, Proc. Natl. Acad. Sei. USA 37
(1 951) 452, 455.
ln Appendix AS we present an elementary introductíon to spin 1/2 fields. For more
deta.ils see
Y. Choquet-Bruhat and C. DeWitt-Morette, Analysis, Manifo!d5 and Physics: Part II
(North-Holland, Amsterdam 1989),
About spin in 4 dimensioru; see also
P. Moussa and R. Stora in Methods in Subnuclear Physics, M. Nikolic ed. (Gordon o.nd
Breach New Yark 1966).
P. Moussa in Particie Physús, Les Houches 1971, C. De Witt and C. ltzykron eds.
(Gordon and Breach, New York 1973).

Exer-cises
E:rerr:i&t 5.1
5.1.1. Write explicitly the field equation ofmotion (5.28) for the generating functional
of correlation functions in the esse of the action

5.1.2. Deduce the equa.tions for cattelatíon functions obtained by differentiating zero,
one, two and three times with respect to the souree.
5.1.3. Use these equations to calculate the 1 and 3-point functions at arder g and the
2-point function at order g 2 •

Ezerr:i&e 5.2
5.2.1. Establísh the quantum equations of motion in the case of a.ction (5.62) in the
absenre of Ql-field self-intetactions.
5.2.2. Derive from tbese equations a set of relations between correlatíon functions and
use them to determine the avera.ge of ,P(x) at order g and tbe fermion and boson 2·point
functions at order g'.
IDE Functional lntcgrols in Field Theory A5.l

APPENDIX 5
EUCLIOEAN OIRAC FERMIONS ANO -y MATRICES

ln this appendix we assume a minímal knowledge of the Dirac equation and the ftee
fermion field action. These questions &e well discussed in standard textoooks of Particl<
Phvsics. We want to briefly describe instead the {ormallsm of euclidean Dirac fermions,
i.e:the continu,.tion to imagína.ry time o[ the thoory of spin 1/2 fielda. As we have noticed
in Cho.pter 5, in this continuation the pseudo-orthogonal group O(d-1,1} transforms ínto
the orthogonal group O(d), d being the euclideo.n spa;:e dimension. Therefore eudídean
fermiolllS transform under the spinorial representat;;on o[ O(d).
The appendix is organized as follows: we define the algebra of Dirac -y matríces a.nd
exhlbit matrices whicb realize the algebra; we describe the transformations of spinors.
We discuss the symmetries of the free fermion action. A section is devoted to the specíal
example of dimension four. We linally show how to calculate traces of products of -y·
matrices, quo.ntíties which a.ppear in Feynman diagrarns with fermions, a.nd define the
Fierz transformation.

Aá.l Dirac 1 Matrices

Space of even dimen.oions d. In this appendix the dimension d of spo.ce is assumed to be

even except when sta.ted explidtly othenvise. Let -,~, I' 1 .•• d, be a set of d matríces
satisfyi.ng the a.nticommuting rela.tions:
(A5.1)
in which 1 is the unit ma.trix.
These matrices are the genera.tors of a Clifford algebra similar to tbe algebta of op-
erators acting on Grassma.nn algebras. The oper~tors {li;+ 8/89,), in the notations of
Section 1.7, satisfy exactly relations (A5.1). It folloW!õ from relations (A5.9 that the -y
matrices generate a.n algebra which, as a vector space, ha.s a dimension 2 . Therefore
they cannot be represented a.s ma.trires of dimension smaller tha.n 2d/'. In Section A5.2
- give a.n inductive construdioo (d....., d +2) of hermitian ma.trices satisfying (A5.1). In
the algebra one elemeot plays a special role, the product of ali -y ma.trices. The matrix
'J's:
(A5.2)
enticommutes, beca.u"" d is even, with ali other 'Y m"trices and

'!~ 1.
lo calcula.tions involvíng '! matrices, it is not always necessa.ry to distinguish ')'s frorn
other -y matríc:es. ldentifying thus 'YS with 'ld+l, we ha.w:
i,j J, ... ,d,d+l. (A5.3)
We shall use greek letters I' v ... to indicate that we exclude the value d + 1 for the index.
Space Qj odd dim0118Íons. lo most of what follows,- limit ourselves to spaces of even
dimensíons. However equation { A5.3) shows that in odd dimensions we can represent
the -y matrices by taking the -y matrices of dimension d- 1, t<> which we add IS·
Note however that in this cMe, in contra.st to the even case, the -y matrices are not all
algebraically independent.
A5.2 Functional lntegrals in Fíeld Theory 109

A5.2 An Explicit Construction

IL is sometimes useful to haV<! an exp!icit realízation of the algebra of 1 matrices.

For d = 2 the standrud Pauli matrices realize the algebra:

1'!(d=2) -= .,., -- ( o1 01 ) (A5.4a)

.._,íd-2) -
<S = 13
(d=2) -
=era
( 1
O -1
o) (A5.4b}

The three matrices are hermitian:

1'1 "'J.
The matrices "'l and "f• are symmetric, and "12 is antísymmetric:
1'2 = - T/'2'

where we have denoted hy T"'

the transposed of the matrix 1·
To construct the 1 matrices for higher even dimensions we then proceed by ind uctlon,
setting:

(d+2) (dJ ( o 1<

{4)) (A5.5)
1; = "' 181 !'; = (d)
1; o '
{dH) ( 0 -il 0 )
l'd+2 = 112 181 l.t = íld o (A5.6)

io whích ld is the unit matríx in zó-/2 dimensions.

As a. consequence 1~d+?) has the form:

(d+2) = ..,(d+2) -
I's - ld+3 -
~
ua""
"" 1 -
d -
( ld
o (A5.7)

The matrices -rJd+ 2) are the tensor products of the matrices -yJdl a.nd ld by the matrices
o;. A Etraightforward calculation shows tbat if the matrice151!d) satisfy relations (A5.3},
the 1!"+
2
) matrices satisfy the same r~lations.
By inspection we """that the 1 matrices are all bennitian. In addltion:

T..,(d+Z) _ ( 0
r; - T (d)
-r.
Therefore if 'l'}dl is symmetríc or antisymmetric, ,.jd<- 2) h"" the same property. The
matrix 1~~~ ) is antisymmetríc, and -r1d+ ) which is also 1'~~~ l is symmetric. It follows
2 2 2

immediately that, in tbis representatíon, a.ll 1' matrices witb odd index are symmetric,
ali matrices with even index are antisymmetric:

(A5.8)
110 Pun<tional Integro/& in FieM Theory A5.3

A5.3 Transformation of Spinors

Let us consider the d(d- 1)/2 hermitian traceless matrices a P• ( with for example I' < v):

(A5.9)

A short calculation show<; that, as a consequence of relations (A5.1), the matrices "'"""
satisfy the commutation relations:

(Aó.lO)

We reoognize that the matrices ia,../2 satisfy the commutation T<llations of the gene r ator<;
of the Lie algebra of the group O(d) which in the delining representation can be cho<;en
as d x d antisymmetric matrices T~":

(A5.11)

Let us now consider the &ee ml.'ISSive fermion action SD { .jj,1jl):

(A5.12)

continuation to ímaginary time of the standard a.ction for spin 1/2 fields.
Let us express that it is in"ariant under SO ( d) transformations which on the spinors
,p and !fj read:
A-;;~ {R) 'Í'J1 (lU) ,
(A5.13)
!fjp (lU) AfJa (R).
The matrix Ris orthogonal wíth determinant 1 and can be para.metrized in terms of an
antisymmetric ma.trix 9:
(Aó.14)
The invariance of the a.ctíon implies that the transformation matrix A should satisfy:

A!.R,.A- 1 ""'--,". (A5.15)

Expa.nding R~. at first arder in 6 and taking A close to the identity we find:

A solution o( this equation is:

and therefore:
(A5.16)
Since the matrix a • is· hermitian and traceless, A is a unitary matrix: A helongs to a
subgroup of SU (2df.,) called the spin group. 'Ib illustrate the geometrícal meaning of this
result, let us give a símple example. From the deftnítion (Aó.9) we see that a,..
(I" i- v)
bas the property:
(A5.17)
 A5.4 Functional Integrais in Field Theory lll

(no summatíon over 11 and v beíng implied). lf we take an antisymmetric matrlx of o,..
tne form
(A5.18)
t.he corroopondíng orthogonal matrix Ris a rotation of angle e in the {/1, v) plane. The
llSSOciated transformatíon matrix J\ is then
J\ cos{fl/2) +ú'""sin(B/2). (A5.19)
í.e. a matrlx belonging to a U(l) subgroup of SU(2) and corresponding to a rotation of
angle ll/2.
From this example we understand tha.t the spin group defined by equation (A5.16) is
not isomorplúc to SO( d), since J\ and - J\ correspond to the S!illle element of SO (d).
Gíven the rnatríx R.,., the equatíon (A5.15) does not determine a unique matrix A,
ínstead it determines R~v as a functíon of A:
R.,. tr (A- 1 ')',.A1'v) / tr 1.
The spinorial representation is a representation only up to a sign. For exsmple the group
S0(3) is associated with SU(2), S0(4) with SU(2) x SU(2). As an abuse of languag"
we shall neverth.,Iess use the expression spinorial representation.

f A5.4 The Matrix -rs

:ic Equation (A&.l9) sbows that 17,..
is an element ofthe group, associated with a rotation
~ of angle .,. in the (lw) plane. Tberefore íf the dimension of spaae is even

'
d
t-il~~, II ')',.- u.,u34 .. . (7d-1,d
,JJ.=l

is also a group element corresponding to the symmetry x ,__. -x.

The free action (A5.12) ís invariant under the tr!UJ6forrnation:
.P'(:r} = 'ls.P (~:r), ..j/(x) = i/J (-:z:}-rs. (A5.20)
In additíon -rs wrnrnutes W'ith ali transformations of the spin group of SO (d):
!A,"fsl O. (A5.21)
Therefore this representation of SO (d) ís not irreducible. Jntroducing the projectors
(l ± -rs) /2, we can reduce the representatlon by defining two spínors, chíral oomponents
of >/>=
(A5.22)
and correspondíngly ~±(x):
tÍi±(x) = ..j;{:r)! (1 ± ')'s), (AS.23)
often denoted by >/>R(:r), >/>L(x), tiin(x), ..j;L(:z:) for rigbt and left components.
However, with two of these spínors it is possible to construct only either a massless
tbeory since:

S(>,L,>I>+) =-f dd:z:~-(x)(i!+m)>l>+(:z:) =-f dd:z:~_(:r)ii>P+(x). {A5.24)

'", a IDB88 tenn alone i f we take another pair like {..j; _, .p_}. To conStmct B.Il action
lor a ma.ssive propagatíng fermion we need four spinors. In expression ( A5 .24) we have
lntroduced tbe tradítional notation {I to represent the matrix &,.-r,..
112 Functional lntegrols in Fíel.d Theory Aó.6

A5.5 Refiections

Let us first concentrate on spaces of even dímensians. To obtain the iull orthogona.l
group we have still to represent reflections, which correspond to change the sign oi One
component of tbe position vector x, and which are orthogona.l matrices with determínant
-L Let us ca.ll P the matrix which cha.ngffi the sign of 1 : :r
Px =i wíth i: {~~=-:r,, (.45.25)
;:,. x,., I";< 1

To compensate in action (A5.12) the effect of this transformation which changes llt ín
-ll1 we have to find a matrix which anticommutes wíth 1'! and commutes with ali other
1'• matrices. We can tak.e 1's"r•'
P: 1/JP (x) = -rs"ft.P (x), (A5.26)

Note that 151'! antioommutes with "fS· In perticular:

1'srd(l+"'fsl1't1s=!(t 1'S)· (Af>.27}

Therefare a reflection exchanges right 311d left components. The spinorial representation
of O(d) ís írreducible. The action {A5.24), in oontrest with the actíon (A5.12), is oot
invaríe.nt under reftectíon.
Odd dimensions. In odd dimensions the ma.trix -1 ha.s detenn.inant -1. Therefore the
whole group O(d) can be generated by addíng -1 to SO(d). Since -1 commutes with
a.ll matrices, O(d) is just the produet SO(d) x Z2. For the spinors we can, for example,
represent total apare refiection hy:

We then note tha.t a ferrnion mass term is nat invaria:nt under reflectian o.nd violates
parity conservation in odd dime~~Sions.

A5.6 Otber Symwetries

If we sta:rt from an action in real time, it may have a set of discrete symmetries: her·
mitícity, ·parity, time reversal a.nd charge conjugation which determine the free action
as well as the couplíng to other fields. After Wick •s rotation, the symmetries whlch in-
volv<l a complex conjugation are no longer directly Eymmetries of the euclidean sction:
hermiticity is lost and time reversa.l has e.nother natural definition which makes it ín-
distinguishable fram the refloctíon as defined in Section A5.5. We therefore gíve below
some euclidean equivalent of charge and hermitie.n canjugations in even dimensions.

A5.6.1 Ch.arge conju9a.1íon

Let us set:
(A5.28)

As a function of the new :fields >/J' and .fi,•, the o.ctiou (A5.12) now reads:

So = j ddx,il(.r) (-c-' ~C+ m) ,P' (z). (A5.29)

 Functional lntegrai.s in Field Throf'V 113

If we "-'Bnt to recover tbe original form (A5.12) we have to find a matrix C such that:

(A5.30)

In the representation of Sectíon A5.2, we can take for d even:

C== (ls)d/> TI '".

ell .. orld

Example
d= 2: C = a.scr1 = io-2 t

d 4: C= 7!13
Let us also consíder a<:tion (A5.24). The transforma.tion (A5.28) leads to

{1(1 + 7s)- -c-' T(l + 7s) ~c.

We have chosen 75 aymmetric. ]f the dimension d is o( the form:

d=2 (mod4),

then:
c- 1 (1+ T7s)C (l-7s),
and therefore the a.ction ( A5.24) is invariant. lf the dimension ls a multiple of 4 then
instearl:
/1 (1 + ')'S) ,_, /1 (1 - 75) ,
and charge conjuga.tion is not a symmetry. How.m<r reflection multíplied by charge con·
jugation is a symmetry. FinaJiy equation (A5.30) justilies the name charge conjugation:
if .p and ;j, are charged fields with charges 'fe, and íf we caJI A,. (:r) the electromagnetic
field coupled to the charge of o{; and ~. the action has the form:

After charge conjugation, a.s a consequence of equation (A5.30) the aign of the charge e
bllS changed.

A5.6.2 Hennitüm conjugation

Let us oonsider the effe<:t of a complex coi\Íugation on actíon (A5.12):

We now make the transformation:

',!l(:r} 'íí'!:r)ls, ;j,(:r) = -7s.P'(:r). (A5.31)

Sil>ce the ma.trices ,,. are hermitian

114 1'\Lnctional lntegrols in Field Theory A5.7

and theNfore:

The determinant resulting from t.he integYal over !/! and 1/> in the functíonal integral il;
real. With eoch eigenvalue of the operator {I + m is assodated an elgenvalue which is ÍL.<
complex conjugate.
Note that complex conjugatinn fnllowed by transiorrnation (A5.31) ;, not a symmetry
of action ( A5.24)

However, the product of a hermitian conjuga.tion by a reHection, ls a symmetry of the

actíon. This product of tra.nsfonnations is the analogue of hermitian conjugation in real
time. We ull the corresponding symmetry rejlec!ion hermíticity.
These different symmetries can be used systematically to classify the couphngs of fii:
fennions to other 6elds. ~ii
i~
A5.6.9 Contin'IWUS intemal symmetries !f
lf we assign a fermion number +1 to >J• and -1 to 1/> we see that action (A5.12) conserves i
the fermion number. To fermion number conservation corresponds a U(l) invariance of
the oction:
>Ps{x) =e" ,p(x), ijí8 (x) = e-••
,P(x). (A5.32)
A massless free theory in ..ven dimensíons possesses an ímportant ..dditional U(l) sym-
metry ca.Iled chiml symm<lry:

(A5.33)

This shows a deep difference b - n boson and fermion fields. ln contrast with hosons,
the property for ferminns to b., massless can be enforced by a symmetry of the oction.

A5. 7 The Exarop\., of Dimension Four

l
Sincc the dimension four plays a special role let us briefly recai! what the previous
results become in the case of the spinorial representation of S0(4) and 0(4). With the
conventions of Section A5.2 the 4 x 4 1 matrices take the form

1• =
o
( ilz
-íb)
o •

(u, are the three Pauli matri=J and IS 15

')'; = ( 1~o
The matrices a"" then become:
a;;=••J•(a~ ~.) fori,j,k:S3,

a,4 = ( ~ _:.) fori S 3.

A5.8 FUnctiana! lntegral!! in Field Theory 115

\Ve reçognize in the matrices

the generators ofthe group SU{2) x SU(2). Tl,e pro}ectors ~(1 ±1•) decompose a Dira.c
spinor into the sum oftwo vectors transforming as t.he (1/2, O) and (0, I/2) representations
,( the group {Weylspinors). Note that a reflectlon exchanges the two vecl.ors (as expected
sir,ce the répresentation ís then no longer reducible). In terms of Weyl spinors the
mnstructíon of ínvariants with respect to the spinor group reduces to consideratíons
ahout SU(2). A useful remark in this context, is that the representatíon and its complex
""njugated are equivalent since

u• "">Uo2 'I U E SU(2),

(,,ee also Section A5.6) a.nd thus if cp and x are two SU(2) spinors the combination

1/'o (o2),1l X.ll = -Í<n.ll'foX~,

where <,.fi is the antisymmetríc tensor (< 12 -- 1), is a SU(2) invariant.

AS.8 Trace of Products of "r Matrices

rerturbatiw calculations involving spin 1/2 fermions often require the calculation of
traces of products of 1 matríces, whích we therefore explain in detaiL
Multiplying equation (A5.3) by "'i
(but not summing over the index i) and taking the
trace we find:
tr1; = 6;1 tr11 •
Taking i 'f j ""' lind a result which is obvious from the explicit realization given in
l;ection A5.2:
tr·n= o. {A5.34)
lu the ce.lculations which follow, the results always will be propartional to the trace of
I lo€ unit matríx. Let us define:
trl = N. (A5.35)

Product o/ even numbers of 1 matríces. To calculate the trace of the product of an

"ven number of 1 matríces, tr"(;, •. ·"~••·, we successively commute ·n,. through ali other
'Y matrices ,.,,, ... ,,,,._, using e.nticommutation relations (A5.3). We then generate
11 ,;et of traces or (2n 2) 1 matrices. At each commutation the sign changes. After
rtll commutations, as a consequence of the cyclic property of t.he trace, we get back the
original expressíon with a minus sign. As a consequence we find:

tr"1'1.1 · • '1'th =c61.,,s2... tr (l'i2 • • ·1'i2"'_t}- c!ii2t2...

x tr ( '"Yit1'13 · • '"'Yt.a... -t) + · · · + ó..:~.-1 t~"" tr {"'Yt, • • · it:3.... -2) • (A5.36)

We therefore prove by induction Wick's theorem for the trace of a product of an even
rmmber of 1 matríces:

tr "r•• ... 'Y.,. ~ N L

all pcJSsib~e
c (P) 6;,., ,, ... 6,,__ ,'"'·, (A5.37)
paí.dnp of
(1.2•... ,2n)
116 Functional lntegml$ in Fie!d TheoT")I A5.8

in which E {P) is the siguature of the permutation P.

The same result can be obtained by calculating

( fl., + ,; )
v ft
(o,, + a: )···(o.,. + a:
t:; \2n
)
Nll
e,-::::.0
.

Odd numl.>er of 1 mll.trices. Let us consider the qunntity tr 1•, .. -1<,.-,. Three cases
may arise:
(i} ]'\;ot all 1 matrices appear in the product. For exBmple 1 1 is absent. We ean then
write:

We eM now a.ntieommute one matrbt 1l with all other 1 matrices. Since their number
is odd we generate a rninus sign and as a consequence of the cydic property of the tr"""
we reconstitute the original product. Therefore the tra.ce vaoishes. In particular:

(.45.38)

(ií)lf some -r rnatrices appear an even number of times, we Cflll use the antícommuting
relatíons to cnncel them pairwise and we return to the pi:evious case.
{íii) Ali -y matrices appear in the product, and ali a.n odd number of times. This
eorrespond• in the odd dimension case to take ali rnatrices 1,., and in tbe even case to
ta.ke ali rnatrices -r,. and -rs-
Antirommuting 1 matrices we can relate the trace to the calculation of tr71 •• ·1•·
in the odd case, which is not determine<! by ant.icommutation relations (A5.1). lt is
necessary to use relation (.45.2) between 1s = 1d+l and the other 1,. matrices. We find:

tr-y,., .. -1,, = N i(d-!)!

2
<,., ... ,.,. (A5.39)
l
In the even case from re!ation (A5.2) follows ínstead: l
(A5.40) l
~
in whích <,., ... ,., is the completely antisymmetric tensor normalired by:
(A5.41)
··~···· = 1.
We shal! see \ater that relation ( A5.40) whicb depends explicitly on the number of space
dimensions has deep consequences. In particl.llar dimensional regu!arization does not
preserve this re\a.tion and this is the source of pos:síble chiral anornalies in field theories
whích are chiral invariant at the classical. levei.
A5.9 Functionallnt~ in Field Theory 1]7

A5.9 The Fierz Transforrnatíon

Within the a.lgebra of 1 matrices ít is possíble to define a ba.sis of :ri hermítíW1 matrices
orthogona.J by the tru.ce. Let us call these matrkeS r A. Any ferrnion 2-poínt correlation
function can then be expanded on such basls. A 4-poínt fermion correlation function can
be expanded on a basis formed by the tensor products of theo;e matrices. However in this
case one has lirst to separate the 4 fermion fields in two sets oi two fields a.nd there are
three ways of doing ít. A connectíon between these dífferent bases can be found through
• Fierz transformation. Let us express that any 2dfo x 2°12 X matrix can be expanded
on the baois of r matrices:
x.~ = N-' trxrAr~"' (A5.42)
in which N is the trace of 1. Let us now choose a matrix X of the form:

(A5.43)

ldentity (A5.42) becomes:

(A5.44}

By expandíng the product r8rA rc on the basis of r matrices we obta.ln the <:lecompo-
sítion of any element of one basís onto another.
Exampies
{i) For d 2 a basís Í5 1, lp a.nd 1's· We Jeave a.s an exercise t.o verify that the set
1 ® 1, 1',. ® 1" and 1s ® -rs transforms into ítself with a matrix Mo:

M2 =-
1(1 1o I)
2
2 -2 .
1 -1 1

As expected the square of the matrix M, ís the identity.

=
(ii) For d 4 a ba.sis ís:
11 'YJ.5l "'TS) i1'S1'Pt Up.v
We leave as an exercise t.o verify that the set:

lrii.DSform.. into itself with a matrix M4 of square 1:

1
-
4
(! _\
1 -1
4 2
-4
1
-4
1
2
-1
-2 t)·
6 o 6 o -2
GENERATING FUNCTIONALS OF CORRELATION FUNCTIONS.
LOOPWISE EXPANSION

In Chapter 5 we have introduced the generating functional of correlation functioJJE Z(J)

which ís nlso, from the point of view of statistical physics, the pa.rtitíon functíon in
the presence of a.n external source {ar lield) J(x). We have shown how to caleula.te
it in a formal power S€ries of the interaction. We now introduce and discu.ss two nevo·
generating functíonals, W(J) the generating iunctíonal of connected oorrelation functions
(th" free en"'gy of statístical physics), a.nd it~ Leg.,ndre transform f(<p), the generating
functional of proper verti~ (nlso sometimes called efiective potential a.nd whicb ís the
thernmdyruuní.cal potential of statistical physics). Connected correlation functions, as a
consequence of locality, have duster properties. To proper vertices contribute only one-
line irTeducible Feynman d\agTams (diagrnms which cannot he disconnected by cutting
only one Jine), and thus in the Particle Physics sense one-particle írreducible {lPI). We
therefore often call proper vertices lPI correlatíon functions.
We explain how to caleulate tbe<;e various functionals in a reorganized perturbative
expanllion, called loopwise expansion.
Tbe functional r(cp) plays an important role in the renonnalization of local quantum
lield theory. In Sectian 6.6 we give a quantum mechanical interpretation for f(<p). We
relate it to the pa.rtition function at fu<ed lield time average. Thls relo.tion explains that
f(cp) also a.ppears in the discussion of symmetry breakíng.
ln the appendix we calculate the generating functíonal of ~loop Feynman diugrruns.
6.1 Generating functíonals of Connected Correlatlon l'Unctions: Cluster
Properties

In Chapter 5 we have defined field correlation functions assodated with an a.ction 8(4>).
The n-point 4>-field correlation functíon z!n) (x, ... , x.,) is given by:

with
Jor 1
= jtM>Je- 514-l.
We have shown in Subsedion 5.1.3 (equatíon (5.14)) that the functional Z(J):

Z(J) "".N f [dq,}exp (-S(</>) +f dxJ(x)\V(x)], (6.'2)

is the generating functional of correla.tion functions:

z!n) (:tt,····:tn) (6.3)

Thi.s relation can also be written:

Z(J)=fkJdxt
"=O
... dx,Z(n)(XJ,· ,:rn)J(xl) ... J(:r:n). {6.4)
 Genemting FUnctíona/3 of Cor:relation Functíons 119

ln Section 5.3 we have noticed that some Feynman diagrarns contributing to correlation
lunctions were disconnected in the sense of graphs. They could then be factorized into a
product of the form:
F 1 (x., ... , xp) F2 (x,.. 1 , .•• , x,.),
where the two dí.sjoint set of ru:guments are not empty.
Let us define a new generating hmctíonal W(J), ana.logous to the free energy of sta-
\Ístical mechanics,
W(J) =In Z(J). (6.5)
We introduce the notation

(6.6)

which implies

(6.7)

We want to prove that only connected Feynm11.11 diagra.ms contribute to W(J).

Let us first wl'ite explicítly a few relations, as implied by the definition (6.5), between
correlation functlons and tbe functions w<"l:

zOl(x) = Wl 1l(x),
z<'l(x~ox~) = W< 2 l(x,,x2) + W(l)(x!)W(t)(xo),
z!3 l(x,,X2,x3) = W! 3 l(x,:r2,Xs) + W(ll(x!)W!'l(x2,xa) + w<tl(:r,)Wl2l(xa, X!)
+ w< 1l(x3 2) + w{ll(x 1 )W(Il(x2 )W< l(x3 ).
)Wl2 l(:q,:r 1

Note that the r.h.s. involves ali possible products of connected functions with coefficíent
1.

6.1.1 A proof of t:onneclivity

We now give a purely algebraic proof of the connectivity property whicb relies on the
linearity property: A linear combination of connected functioru is still connected.
In Sectíon 5.2 we have calculated the generating functional Z ( J) for a free lie!d theory.
We have fonnd (equation (5.19))

Z(J) = exp [i j ddxddyJ(x)tl.(x -y)J(y}].

Expanding ín powen; of J we immediately verify tbat, exoept the :l.point correlation

function, ali correie.tíon functioru; are diaoonnected. The functiona.l W(J) instead is

Ali correle.tio:ru; function vanish, except the 2-point function •thicb is connected.
120 Genemting runctionals oj Correlatíon runctions 6.1

We now assume that for some action S(rp) we have shown that W(J) generates con·
nected correlation functions and we add a sma.lllocal perturbe.tíon to the actíon

(derivative couplingsleave the ,.rgument unchanged). Then

exp W,(J) o: f [doj>]e-8•(~)+J,. = exp [-~f dx CJ~x)) N] exp W(J).

lt follows
W,(J) = W(J) - < .-W(J) f dx ( éJ~x)) N eW(J) +0 (<2) .

Exa.mining tbe contributíon of arder E we see t.he.t it is a linear combine.tlon of products

of derlvatívcs of W ( J) witb respect to a source at a unique point :r. For example for
N=3

Therefore if W(J) is conneeted, ali additional terrns are also connected. The corre-
sponding diagra.ms are connected diagrama contributíng to W ( J) nttacl>ed to the point
x.
Since W( J) is connected for a general free field theory, it follaws, after iotegration ovet
the rorresponding coupling const!Ult, that it remains connected for any interactíon.
We thus cal! t.hi! functions W(n) ronnected correle.tion ftmctiollS. To insíst on the
connected character we sometimes use the ootation

(6.8}

wbere the symbol <>c means oonnected pe.rt of the corresponding correlatíon functioo.

6.1.2 Cónnected co,...lation functiom anà duster properties

For simplicity, w.- now coiiSider only local euclidean ections S{oi>) functlons of a scalar
field Ç>(:t) and its derivatives, and invaria.nt under space trlUISlations. Tbis mea.ns that
the lagrangian den.-ity does not depend oo spaçe explicitly but only througb the fieid (Bii
in the example (li.l )).
We bave already observed in quantum mecha.nics tbat tbe partltion functíon tre-~ 11
in tbe large volume limit, í.e. for {3 -+ oo, bas an exponential behaviour; more precísely
{equation (2.2)):

in wbich Eo iE the ground state energy. Moreover the convergence towards the límit is
exponential when the ground state is isolated (equation (2.3) ).
Let os here generali..e the quil.lltum mechonical result to local field theories. The
díscussion which follows is highly intuitive and tries only to rnotivate results which cWJ
G.l Generating FUnctionals af C0 rrdation FUnctions 12]

be proven rígorously. Peeuliarities found in massless theories or due to ground state

tlegeneracy will be ignored here and discussed on)y )ater.
The functíonal Z(J) i>; a partítion function in presence of an externai source J(x). Let
us take for source J(x) the sum of two terms li(x) and J2(x),
(6.9)
in which J1 (x) and J2(x) have dísjoint supports consisting of two domain.s !11 and n, of
illlge vo)um., V, a.nd V2

o,
o,
\'<'e also assume that J 1(x) a.nd J,(x) fluctuate around ao arbitrarily small but non·
vanishing const!illt. Then the locnlity of S (li>) implies tbat we can write:

S(Ç>) fdx}Ç>= f dx[.C(Ç>(x)) J 1 (x),P(x)J+ f dx[.C(Ç>(x})-J2 (x)<f>(x)]

l:en 1 l::r.'in~
+ 1
~n,un,
dx .C (t/J (x)) + contributíous from boundari"". {6.10)

Let us th<'n write Z(J) as:

(6.JJ)
with the definitíona:

z, (11) = 1en. +f dx J (x)Ç>(x)],

[dçf>(x)l exp [-S(Ç>) 1

z, (12) = len, fdÇ>(x)] exp [-S(Ç>) +f d:d (x)Ç>(x)] . 2

Both functionals Z 1 and Z 2 are norma.Ji:red to 1 for 1 1 = O or J 2 = O mspectively. The

functional Z 12 is defined by equatíon (6.ll). lt.s dependence in 1t and 1 2 comes entirely
from the existenoe of boundary terms in equation {6.10). Wben we scale up and n, n,,
these boumlary terms grow Jike surf&ces while the two ftrst terms in equation (6.10) grow
Uke the volumes V1 and V2 . Therefore In Z 12 (J1 , 1 2 ) becomes asym.ptotic:ally negligihle
· campared to In Z 1 (Jt) and In Z 2 (J,) when V1 and V2 _, oo.
To express thi>< property it is natural to introduce the functíona.l W(J) = In Z(J)
(equation (6.5)) which then satisftes:
W (11 + 12 ) ""'W1 (1,) + W, (h)+ negligible. {6.12)
In particular if S{q,) JÇ> ís tra.nslation invariant (which ímpli"" that J is a constant),
W( J) is extensive, í.e. proportíonal to the total volume. This property generalizes prop-
erty (2.2).
Cluster properlies. After havíng taken lhe lnftnite volume limit we can expand W ( J 1 +
J2) in powers of J, anã 1 2:

W(lt +h)= f:~

n=l
t (")f
p=O p
dxl ... dx.dyP+l ... dynll'(n) (x,, ... ,xp,Yp+J, ... ,y,)

x 1, (xt) ... 1, (xp) J, (Y.+l) .. ·lo (y,.), (6.13)

122 Generating Functionals oj CoTTelation Funct1ons 6.2

with x, E flt, 1/) E fh.

Property (6.12) ímplies that ali terms with p #O or p # n are negligible for \'1 aod
\!2 large. Consideríng expression (6.13) we see that this imp\ies that the lunctions W'"l
must decrease rapidly enough when the two non-empty sets oi points {x 1 •.• x,) and
{Yp+ 1 ... Yn} are largely separated:

W(n) (Xt···Xp,!/p+l···Yn)- 0 when ,!'1i~P l.r; ~yj[- DC. (6.14)

J..:Op-i-l···n

Thls property, whíc.h we have bere describW very qualitatively, is called the duster
property, and ís "cha.ra.ct.eristic property of the connected correlation lunctions generated
by the functional W(J).
Feynman diagroms. We have seen that a Feynman díagram which ís disconnected ín
the sense of graphs can be !ectorized into a product of the fonn:

Ft (xt,. · · ,x,)F, (Yl,• · · ,yq) ·

lt is clear that we can separate the set {:r,) and the 8<ôt {y;) in" translation invariant
theory, in a way whic.h leaves suc.h a product invariant: A disconnecle<l díagram cannot
satisfy the duster propeTty. We recaver the property that the Feynman diagrams which
contribute to the perturbative expa.nsion of W(J) are ali connected. Furthermore it can
be verilied that, in a lield thoory conta.ining on!y massive fields, connected Feynman
diagra.rns decrease exponentially, when points are sepMated, with a mínimal rate which
is the inverse of the srnallest mass in the theory. This property ís a CDnseQuence of the
exponential decrea.se of the propa.gator (see Appendix A7.1 for details).
Remar*. The relation betw<!en action a.nd generatíng functional, wbich hi>S the form
of a Laplace tramfonnation, can formally be invetted ·

.,-s(~l =J!dJ]exp [w(J)- j dxJ(x)Q>(x)], (6.15)

where one integram over imagioary sources J(x). A truncat.ed loopwL<;e expansion (see
Section 6.4} of the functional integral then yields approximate non-iinear equations for
correlation function.s. lt ís actually conveníent to íntroduce in the r.h.s. the generating
functional of proper vertices delined in next soction.

6.2 Proper VeTtices

We now introduce a new generating functional called the generating functianal of propet
•ertices. lt is obtaíned by applying a Legendre transfonnation to the generating func-
tional of connected oorrelation functíons W(J).
Legtndre tronsfomw.ti<m. We have a!ready defined the Legeodre tra.ns!ormation I>S
the transformation re)ating the lagra.ngian and the hamiltonian. We sball see !ater that
in the context of statistícal mec.hanics a.nd phase tra.nsitions it ís natural to consider
the Legendre transfonn of the free energy W(J). Here proper vertices are introduced
because they hav<! special properties from the point of view of perturbation theory as we
demonstrate in Sections 6.4 and 6.5. The generating functional of proper verti= r (cp),
in which cp(x) is a classical fiekl argument of r, is obtained from W(J) by:

f(.,o)+ W(J) J dxJ(x}tp(x) O, (6.16)

6.2 Genemting FtmctionaLs of Correlation F\r.nctians 123

;o(x) being related to J(x) by

(6.17)

This equatíon expresses that the J.h.s. of equation (6.16) is stationary wíth respect. to
variations of J(x) at r.p fixed.
Differentiating (6.16) with respect to ~t(x) we obtain:

6 ;~x)- J(x) + Jdy !~~~~ b}(y) [w (J)- f dxJ(x)10(x)] O,

a.nd therefore takíng into account equation (6.17)

bf
J(x) (6.18)

Equa.tion (6.18) shows that the Legendre tnmsforma.tion is involutive. In particular

equa.tion (6.17) implies tha.t

\!>(X)

i.e. that ;o(z) in o.ero souree becomes the expeetation value of the field tJ>. Conversely the
equation (6.18) then lmpli.es tha.t the expectation value of lf>(x) is an extremum of r('P)·
Finally let us note tha.t if W(J) depends on some a.dditional argument a then:

arl awl
8a ~ + 7);; J +
f dy
âJ(y)l 6
~ ~ M[y) lW(J)
r (6.19)

Taking lnto account equation (6.17) we ohtain the important resuJt:

(6.20)

Note tha.t we have deríved this result for one externai variable but it obviously applíes
~tlso
for an externai field or source.
E:r:pan.rion o/r('f!). 1f we se!. J =O we find from equation (6.17)

op(x) = wo>(x) {lf>(:t)}.

lnverting the relation (6.17) we can thus expand the source J(x) as a series in powers of
x(x):
x(x)"' <P(:t)- w(ll(x). (6.21)

We see that x(x) is related to the correlation functions of the field

~(:t) IJ>(:t)- {tJ>(:t)) (6.22)

which has zero expectation value.

124 Genemting Ftmctiana/s o[ Correlatwn FuncLions 6.2

The first terms of the expansion of equation (6.1 í) then are

x(x) =f à:r, wm (:r,:r,) J (:ri)+~ f d:~:,dx, w' l {x,x,x2) J (:ri) J

3 (:rz) +
(6.23)
We intrndure the inven;e of the connect.eà 2-point functíon:

f dzS(x,z)W''l(z,y) ::d(x y). (6.24)

We can then ...rtt,e the expansion of l(x):

J(x) =f dx, S (:t, >:1) X (:r1) - ~ f dy 1dy 2dy3dx 1dx 2 S (x, !IJ)

X S (xt, Yt) S (x2, !t>) W(J) (1/!,lt.!, YJ) X (xl) X (xz) + · ·· . (6.25)

Note that this expansion can be naturally expressed in terms of ro-cal!ed amputated
correlation functíons W~b_,

In terms of Feymnan diagrams thís mea.ns that propagators and contributions to the
2-poínt functíon ou externallíncs are omitted.
We can now use equatíon (6.18) to calculate r {<,0)· Setting:

f:-\
I n.
/dzt ... dzn fÍ"l (xl,···, x~) X (zt) ... X (z,.), (6.27)

where the f(n) are associated with the fiel.:! ~. we find:

f1 1l(x)-= O,
f(Z) (ZI,X2) S (Zt,X2) = [W1 1 lj-l (:r~oz 2 ),
f{J) (:rt,Xz,ZJ) -W~p. (xi> Z2,:r3),
r<•l (z1,_:r 2,:.: 3,o: 4) = -W~. (:r 1,:r2,z3,x4)

+f dy dz W~!/p. (:tt,Xo,Y) W(>) (y,z) WJ;:fp, (z, xa, >:4) + 2 terrru;

The inverse relatíons are even more useful, because, as we show in Section 6.5, r(IP) ha.
simpler properties thBll W(J)

wl'l (z 1 ,x2 ) = (rl 2lj-1 (:r~,x 2 ),

w,l!?p,(x.,:r,,x3) = -f( 3 l{:~:l,:r2,x3).
W.!!i". (:r~ox2,:r3,x•) = -r(•J (xl,x,,xs,x•)
+ j dydzfC 3 l (x 1,x2 ,y)W' 2' (y,z)f('l (z,xa,x 4) + 2 terms
6.2 Generoting Functionals of Correlo.tian Functíons 125

Let us give a graphical representatíon of the first eqnations. We define:

Fig. 6.!

The correlatiou functíoru w(3) and w••l can then b~ represented as shown in figures
6.2 and 6.3 respectively.

Fíg. 6.2 The 3-point function wl•l.

+2 terms.

Fig. 6.3 The 4-polnt function w(•l.

lt follows frorn the set of relations batW""n connected correlation functions Md proper
vertíces that Feynman diagra.ms which contribute to proper vertices appear in the ex-
ransion of eonnected functions with tbe opposite sign except io the case of the 2-poínt
function. lndeed let us set:

r< 2l (o;,y) = K(x,y) + E(x,y),

in which K (x, y) is just the ioverse of the propagator and E (x, y), called the me.ss op-
crator, contains all perturbative corrections. The connected 2-point function w<2 l (x,y)
then takes the form of a geometricl>l series:

iu which 6. (x, y) ís tbe propagator. In thís cm;e tbe lirst term in perturbation theory,
the propagator, ís special from the point of view of Feynman graph expansion, since it
hru; the same sign in both functions.
126

expressíon;

we immediately verify:
Generot;ny Punclionals of Correlatíon F\lnctions

Conv.x-ity. lt follows from the definí~ion of connected corre!atíon functions that the
2-point function W\ 2 l(z, y) is the kernel of a posítíve operator. Writing its exp!icít

Wf2 l(x,y)= ([<p(x)

f
(<p(x))]iop(y)

dxdyJ(x)J(y)W\ 2 l(x,y)
(<p(y))J),

~O.
5.3

'
í

I
Since r\•J (:r, 11) is the inverse of wm( '"· y) in the sense of kernels, it is a.lso the kernel of
a pos-itive operator,
f dxdy<p(x)ff 7 l(x,y):p(y)? O. (6.28)

For the same reasons, 62 W(J)flíJ(x}6J(y) and 62 r(:p)f6.p(:r)óo;l>(y) whích are the 2-
point functions in an externa.! source, are positive operators. In particular, in tbe case
of constant sources, W(J) and r(;p), both divided hy the total spa.ce volume, are convex
functions of tbe sources. We shall recall this property when we examine the physics of
spontaneous symmetry breaking and meet functions r('P) which are not obvlously convex
(Eee Section 6.6).

6.3 Momentum Representation

In tlús work we shall mainly study theories jnvariant under space tra.nslations. Corre-
lation functions then depend ooly on differences of space arguments. For example the
connected 2. point functlon has the forro:

lt is thus natursl to introduce tbe Fourier transforms of rorrelation functíons:

(6.29)

More generally the translation invo.rja.noe of w<nl (X!.~ •. ' . , x.) is equlvalent to IDC>-
mentum C<!nservatíon of its Fourier transforrn w<nl (p1 , ••. ,p,. ). We set:

(2r.)d6 (tPi)
1."""'1
w<nl(Pt>···,Pn)= f dxl···d:r.,W\"l(x,, ... ,x.,)exp (itx;P;).
3'01:'1
(6.30)
Similarly we siso introduce the Fourier transforms f(n) (p 1, •.. , p,) of proper ""rtices:

(2r.)dli (i:P•)
•=1
f(n) (ph ... ,p.) =f dx1 ... dx.,r(n) (:r 1, ... ,x,)exp (i t:t;P;).
j!"!'-l
(6.31)
Note that the Fourier transform ís delined in sucb a way that íf we set in the generating
functiona.l f!<P(x)]:
(6.32}
6.4 Generoting Fhnctitmals oj Cc>rrelatíon FuncLions l2í

then the l.h.s. of equatíon (6.31) is the coellicient of \Õ [p.) ... \Ô [p.,).
The explídt expressians of Section 6.2 (for example equatíon (6.26)) then show that
amputation and Legendre transforrnation become in momentum representation purely
a.lgebraic opetations in the sense that they ínvolve no momentum íntegratíon because
tbe two-point functíon is diagonal in this representation:

and a.lso:

Thís remark will be useful for the discUs5íon of divergences in perturbatíon theory.

6.4 Semiclassical or Loopwise Expansion

For reasons whicb will become clear at the end of this section, it is useful to reorganize
perturbation theory by grouping some classes of Feynma.n diagrams. For this purpose we
shall perform a semld8BSical expansion of the functional integral, Le. 8 formal expansion
in powers of li whkh we therefore a.gain set in front of the classícal action and the source
term (the normalization Z(O) e I is implicit):

(6.33)

For h small, we see tbat we can calculate the integral by the steepest descent method.

6..p Loop aptJn.rion at kadin9 arder

The saddle point equation is:

6S
.5r,D(z) l<i>,(J)] =J(x). (6.34)

Substituting the solution ,P,(J) in to the classica.l a.ction we obtain Z( J) at leading order

1
lnZ(J) -JnZo(J) li (-S(,P.) +J1i,) (6.35)

When fi is explicit, it is conveníent to normalíz.e W(J) by:

W(J) =h lnZ(J). (6.36)

128 Genemting Functiana.ls of Gorrelation Functions 6A

Then at leadíng arder:

W(J) = H'o( J) '"' -5 (</!,) + lrPc, (6.37)

Pe-rturbation theof'!J. Ordína.ry penurbation tbeory is recovered hy expanding the ·.

solutíon </J,(J) in powers of J.
Let us write S(<P) as the sum of a free part and interaction tenns:
f
~;

S(ri>)=~tiíKHV(c/J),
wjth:
V(Qi) O(.p3 ) for,P--+0.
Equa.tion (6.34) then takes the form:
.IV
K,P, + ó,P, =J. (6.38)

1t can be itera.tive!y solve<! as:

w
6.1- 6.- (6.J) + ... (6.39)
ó.p,
If for example
V (</J) = ~
4..
/dx</>4 (x),
1/J, has the Feynmnn diagrams expa.nsion:
J

rp,=---J- 3! f~J + .
J
Fíg. 6.4
\Ve observe that only tree, Le. wjthout loops, diagra.rm a.re generated. Substituting
the expansion into equation (6.37) W>l note that the perturbatíve expansion of Wo (J) in
power; of J al.so contalllli only connected tree Feynman die.gra.rns. The functional Wo(J)
ís the generatíng functíonal of connected tr"" diagra.rns.
Legendre tron.sformation. Let us now perlorm the Legendre transform!ltion:
6W0
'l'(x) = óJ (x) = ,P,(x) +
f éifl, (1/) 6
dy óJ(x) 6,P,(y) {J,P, - S (</J,)] •

Using the saddle point equation (6.34) we find:

rp(o:) </l,(x). (6.40)
Replacing 4>, by rp in equation (6.37) we can write;
Wo(J) + S ('I') J<P O, 16.4.1)
and therefore by definition (equation (6.16));
fo(<p)=S(<p)· (6.42)
At leading order I'('I') is identical to the cl"""ical action. The action and tbe generating
functional of connected tree diagrams are rela.ted by a Legendre transformation. From
the point of vi...w of Feynm!Ul diagrams, S ('I') contains only the vertices of the throry.
GA Genernting F'IJnctwnaL' of Cor-rolation Functwm 129

6,4. 2 Order h corrections

To calculate the order h corrections we have to evaluate the gaussían integral obtained
hy expanding around the saddle poínt. Setting:

<f;= ~c(J) + v'nx, (6.43)

and expanding the action in powers of 1t we find:

2
1 <P<+~ fd:r1d:r2
Óc/J (XI
6
;: ( )I
t/> X2 ~-~,
x(x!)x(:r•)+O(n
3
1').
The functíonal integral at this order becomes:

and therefore:
628 ] -1/2
Z(J) oc Zr,(J) [det 6q,, (:rt)ótl>o(:r2) (6.44)

Finally we nonnalíze Z(J) by tbe conditíon Z(O) = l.

The connected generating functional W(J) at tbis order ís then:

W(J) = W0 (J) + f!W!(J) +O (1>2 ) , (6.45)

with

-~2 ~-- 6
----1
28
W 1 (J) = [trln {6.46)
óq,e (:r!) 6,P, (:rz) J

Let us again take tbe exàmpli! of the q,• potential to iiJustrate thís result:

Therefore:

trln 6•s
6q) (:rJ) Ó\f> (:r2)
I-
J
trln ó'S
Óq) (:rJ) 61,!> (:r2)
IJ=O

trln [6 (x 1 :r2) + ~~ (x,,:r,) ,t.;(:r•J].

Let us then expand W 1 ( J) in powers o f q,,:

W1(J) f
-H~ dx 1 ~(:r~,:r!)\f>;(xl)
f f d:~:, d:r2 ~ (x 1 ,x 2 ) q,~ (x2 ) ~ (:r2 , x 1) r/J~ (x 1 ) + .. j.
ll is ímportant to note that the trace operation ·h as generated a set of one-loop Feynman
diagrams.
130 Generottng Functíona.ls of Correlation Functíons 6.4

Fig. 6.5

To recover perturbation theory, we ha.w still to expand <I> c( J) in powers of J. A typical

contributíon to W1 { J} has the representation of ngure 6.5.
Let us now perforrn the Legendre transformatíon

(6.47)

By definition of the Legendre transformation the r.h.s. is statíonary with respect to

wria.tions of J(x) at 'l'(x) fu<ed. Therefore a rorrectíon of arder li to tbe relatian between
J(:z:) and <p(x) will produce a clmnge of order li2 to the r.h.s. of equation (6.47).
Hence at order li we can stiU replace J(x) by the solution of the equa.tion

<p(x) = 1/Jc (J;x)

We therefore conclude:

r l'l'l s l'Pl + ll.I', ('f'l +o (li2 ) , (6.48)

(6.49)

S(,P) = ~Ç>K,P+ f V (ci>(x))dx, (6.50)

the expa.nsion of r, ('f') in pawets of v takes the form:

1 "" (-1(+1
ft!10)=2L n
f dxt ... dx,V"('I'(x 1 ))6.(x1 ,x 2 }
n-1
x V"(IO(x,))L>. (x2,xa) . . V"(IO(xn))6 (:tn,X!}, (6.51)

whích in tenro; of Feynma.n diagrams has the representation of & sum of one-loop día.grams
(see figure 6.6).

V'O'x,
xaV"
yu :.tJ •

Vu Zn
•

Fig. 6.6
6.4 Genem!íng Punctionals r!/ Co.....-elatíon Functíons 131

The remarkable property of the Feynman grapb expansíon of the functional r 1 ( ;p) at
this order ís that all diagrams are one line irreducible ( l-particle irreducible ín the lan-
guage ofParticle Physics), i.e. they cannot be dísconnected by cutting only one line. The
functional r 1 (;p) ís tbe generating functional of one tine irre.ducible one-loop Feynman
diagrarru;.

li.J.9 Ferrnions
We have discussed tb" Legendre tra.nsform only in the case of a boson field theory,
bawever the extensioo to fermions is straightforward. Let .p and ;j, be two fermion fields
and S(;f,, 1/J) the corresponding local action. lt is clear that W('l, fi) = In Z('1, jj) is stil!
the generating functíonal of connected correlation functíons (we have called r; and '1 the
sources for .p and ;j,). If, following the conventions of Section 5.5, we write tbe souroe
terrru; in the functional integral f/111 + ;J,'I, then we define the Legendre tran5form of W
by

r(;j,, !/!} + W('J,fl) =f dx [ii(x).p(x) + ;j,(x)'l(x)j (6.52u)

6W - .SW
1/J(x) = ~1J(x) ' 1/J(x) =- 6l'}(x) • (6.52b}

Tbe equations (6.52b) are equiva.lent to

õr ór
T](:r) = é~(x) ' ij(:r) = - ô.p(x) ·

With theoe conventíons one oosily verilies that r(;j,, ..p) = S(~, .P) a.t tree order. Ali the
otber algebraic properties derive<! for bosons generalize to tbe fermion case. \Ve here
recai! bowever that a gaussian integration over fermions yields a determinant instead of
tbe inverse of a determinllJlt for a complex scalar. This implies that fermion loops are
affected by a additional mínus slgn compa.red to boson loops. Let us, as an example, give
tbe one- Joop -ults corresponding to the boson-ferrnion action

S(.P,-ii>,!/1)""' f d0 x [-~(x){!ll + A(.Pl).P{:r) + Sa(.P)]. (6.53)

We bave to eva.luate the functional integral:

Z(J, ii, 'I) = f!drf>d~d.pJ exp [ -S(.p, ~' 1/J) +f d0 x [J(:r)</J(x) + 7j(:z:)!/J(x) + ~(x)'l(x)]]
(6.54)
We first a.gain look for the solutions r/>., ;J>., .Pc of the field equations. After shifting the
fields we then have to calculate a gaussían functional integral over boson and fermíon
fields. The result can be deduced from the expressions ( 1. 64--1.66) fortbe determ!nant of
a mixed matríx involving baoons and fermions, or can be obtained directly by integrating
over fermions first, and tben """' bosons.
After a short r::alcu)ation we find for connected correlation functwns

W(J,f/,11) = -S(.f>c,Ofic, 1/Jd +f ddx [J(x)ofJe(:r:) + f/(x)i/Jc(x) + ~.(x)'l(x)]

62 Sa _ IJ 2 A - óA ~1 6A ]
- ! tr In [ ó,P.(:r)ó4Jc(Y) - i/Jc 6,P.(:z:).S.p.(y) .Pc + 21/lc o,P.(x) [jil + A(,Pc)] o,Pe(Y) 1/Jc
+ trln [li'+ A(.f>c)J,
132 Generu.ting Functionals of Carrdatwn FunctioTl5 6.4

where the fie!ds tf>,, ijJ,, '1/J, a.re solut>ons of the field equat>ons

liSB - 6.4
6.p,(x) 11>< 6<;1>,(x) .p,
- J(x) = 0' (6.55a)

!!lJ + A(tf>,)] rJ•,(x} + 1J(X) = 0, (6.55b)

ttb) [11 + A(<,t>,)J +i)( r)= O, (6.55c)

(by conwmtion the operator â in (6.55c) acts on the left. with a minus sígn).
A Legendre transforma.tion then yield• the l Pl functiona.l at ane-loap arder:

r(.,.,.J,,'I/J) S(<p,i/J,.P)- trln(il + A(lf))

1 { 62 Ss • [ éA I" )]-l
+ 2 trln O!p(x)6<p(y) + '1/J 2/i!p(x) "'+ A{rp

6 ..1.4 Loopwíse upansicm at higher orders

We now expand the functionaJ integral

with S(Ql) af the fonn

S(Ql) = ~<t>KH V(Ql),
to ali arders in perturbat.ian theory. Let us determine the power of h in front a cmmet:ted
Feynman diagra.m contributing to W(J}. The propaga.tor is lii).. Any vertex generated
by V (<p) yjelds a power h-1 • In the sa.me way at the end of ali externaJ !ines is attached
a la.ctor J whích also yíelds "factor h- 1 • Calling I the number af intérnallínes of the di-
agrarn (propagators which joio two vertioes), E the number of exte.rnalli.nes (propagatoT
joining a vertex to a source J), V the number of vertices, we find the power

the last fa.ctor 1i oorning from our normaliza.tion oi W(J) (equatían (6.36)).
Note tbat the same result is obtalned for one !íne irreducible Feynman diagrams (i.e.
a.s we prove in next section contríbutions to r (.p)) because tbe facto r li. coming from the
source·cancels tbe f...::tor coming from the externaJ propagator.
Let us now sbow that. the power of h that we have lound counts the number af loops
of a diagram. The number of loops is defined in the following way: if by cutting a Une
of a connected diagram -y we obtain a new connected diagram -r' then:

# loops -y # loops -y1 + l .

Ffom this definitíon follows a ralation betwren the numher of loops L, the number af
internaJ !ines I and tbe number of vertices V:

L=! V+l (6.57)

We prove this relation in the following way: We see tbat eaclt time we can remove
an internai line without disconnecting the diagram we decrease I by 1 and L by 1.
G.5 Generoting Functionals of Corrdation Functions 133

Event.ually we get a tree diagra.m, i.e. a diagrem in wbich no line can be removed without.
disconnening the diagram. We then h ave to show:

1-V+l=-0.

Ftom a tre<> diagram we can remove systematictilly a 'rertex at the boundaty with the
line connecting it to the diagram until we obt.ain the simplest diagram, composed of a
line joíning two vertices, which satisfies the equation.
We have thus shown that the expansion in powenJ of h ís a reordering of perturbation
tbeory a.ccording to the number of loops of the Feynman diagrams.
The number of loops is alro the number of lndependent int..rnal íntensities in tbe
correspondíng electric circuit, the current being conserved at each vertex, the intensítíes
ftowíng ínto tbe díagre.m being fixed. Thls can be proven by showing tha.t equatíon (6.57)
is again satisfied. Indeed the number of L of independem lntensities is equal to the total
number of intensities I mlnus the number of conservation equations (V- 1) \because one
equation gives the total consel"'mtíon of tbe current). This remw:k wi)l eventually allow
115 to relate the number of loops to tbe number of índependent momentum integratlon
variables.

6.5 Legendre 'Ira:nsformation and l·h-reducibility

We h ave shown that the two first ordeno of the )oop expansion of r(ip} contain only one-
llne irtedncible Feynman díagrams. We now suspect that at arder hl. we shall obta.in
the generating functional of all L-loop diagrams. By realizing that proper vertices a.nd
connected correlation functions are oillected by global comhinatorial factors ídentical
respectively to those of vertices In the action and connected tree diagrams, it 1s possible
to give a general proof that l'(.p) 15 the gener&ting functional of one line irredudble
Feynman diagrwns: one takes r(IP} as the action, one calculates at Jeooing in h, and
r = W (J) as the generating functional of the correspondíng connected tree diagrams.
The result then foll!JW!l from the <:onsiderations of the heginnlng of the section {equation
(6.41)). Haw<Mlr we here give a more powerful and completely algebraic proof.
To prove that proper wrtices are given in perturbation theory by a sum of one line
or one particle irteducible (IPI} Feynma.n diagrams, we directly use the definition and
prove that by cutting one Jine iJJ ali possible way:; in a diagram contributing to r('Pl.
the díagrrun remains eonnccted.
Let us oonsider the a.ction:

(6.58)

The symbol E repre:sents a small pararneter ín which we expand at first arder. The
corresponding propaga.tor !'J... (:z:, y) ls:

f é:..,(x,z)[K(z,y)+eJdz =ó(:z:-y),

é:.., (x,y) é:.. {:z:, !/)- E'](X)'J(!I) + 0 (~), {6.59)

with the definition:
'l(x)= j é:..{:z:,z)dz.
134 Generoting Fundianals oj Gorrelation Functíons

If we ncrw expand a Feynman diagram wíth the new propagator À, (x, y) in E, we obtaiu
at first order a sum of terms in whích in ali possibie ways a propagator À (x, y) hrus beeu
replaced by the product -11 (x) '1(!1}· Since ín this product the dependence in x and )I
is fa.ctorized, this mea.ns topologica.lly that in the Feynman diagram the correspondíu~
line has been cut. A necessary and tiUfficiellt condition for a diagram to be lPI ís thnt
ali terms at order E are connected.
Higher orders in E can be used to study irredudbility with respect to cutting 2, 3,...
I ines.
Let us ncrw ca.lculate the part.ition function Z, (J) at first order in <:

Z,(J)= fldo~>J(l ifd:z:dy,P(x)oi>(!l))exp[-S(oi>)+J,P]+0(• 2),

and therefore:

Z,(J) = [1 ~f dxdy óJ~x)óJ~y) + 0(•')1 Z(J).

The geuerating functional W, (J):
W,(J) In Z, ( J},
is then g'iren by:

W,(J) W(J) ~ {(f dx /,%J +f dx dy 6J (~~~(x)} +O (< 2

).

The first arder in < of W, (J) con~ a contribution of the forro:

[j dxó~~J
wlüch iB disconnected, as expected.
Let us now perform the Legendre transl'ormation. We use identíty (6.20) under
form:

a.nd therefore:

r,{~)= r(ip) + ~ { [! dxip(X)r +f d:z:dy [ 6 \0(~~:(y)] -l} + 0 {~ 2)

2
We see that the first order in € contaíns two terms, the terrn [J dx <p(x)] , wlüch wc
added to the a.ction, !Lnd a seoond t.erm which is the connected propagator in
of an cxt.ema.l field. Since the variation ofi' (10) is conneet.ed, f(op) is indeed one
in the language of particle physics, one partlcle irreducib!e. In the chapters whích
we sha.ll use indifferently the terms of prúper vertices or lPI correlation functioos.
finally that in Section 6.2 we have shawn that
W!::/p. = -r(n) + redudble terms (n > 2),
because the difference contains only Jower oorrelation functlons related by
The diagrams contributing to r(n) thus differ from the lPl amputated díagra.rru; oi
only by a sign. In the same way the diagrams contributíng to r(2) beyond tree 1~""1
up to a change of sign, the amputated diagrams of the mass operator.
 Generating Functional. of Corr-elation Functions 135

ll.l; Physkal lnterpretation of the lPl Functional

In ~<·hat follows we explicitly requir" time trans!atíon invariance; the quantum mechanical
luuniltonian H is time-independent.

L. 6.1 A quantum mechanica! inteT]JTCtation

W<· first give a quantum mechanical interpretation of r( I") when the fiald I{J, argument of
r' ís time-independent. Let us therefore distingui.sh in the d dimensionalspa.ce lRd a time
d"ection, and now call t, x the time and space coordinates respeetively, We assume that
•'~>ries in a finite interva.l lO, P') and impose periodíc boundary oonditions on the fi.,!ds in
time direction. Mor<lQVer we restrict ournelves in what follows to time-independent
J(x,t):
J(x,t) J(:z:).
functional Z( J) is then the partition functlon tre-llH for a soorce dependent hamii-
H{,P,J) (equation (5.7)):

H(,P,J)-=H(,P)- f d'Hx,P(x}J(:tJ (6.65)

tbe large [J lirnit, the operator e~fJH projects onto the ground state of the hamiltonía.n.
Eo(J) the ground state energy of H (J), we obtaín:
W(J) ,.:::;-,., -[JE 0(J). {tL6!l)

definition E o( J) is the expectation value of the hamiltonian H (4>, J) in its ground

Denoting by (·) J the average of an operator in the groundstate of H(,P, J) we can
the rela.tion:
W(J)...., -{j(H(if>,J))J...., -[J(H(,P))J -f dd-'zJ(:x) (,P(x)); . (6.67)

time translation invwiance it is eosy to verify that W(J)f[J is tbe Legendre trans-
of r(<p)/{3 when <p(.z) is .. time-índependent lield:

r( \O)=- W(J) I
+ [J ~- 1 xJ(:r)IO(:t), (6.68)

16W(J)
(6.69)
73
6J(:z:).
the functional integral tepresentation of the partítíon funclion J (x) is a source for
dt if>(:r,t). The functional derivative of W(J) with respect to J(:r) is the 1-point
· functíon of this quantity, in a source. Due to time translation invariance
is time-independem. In the large [J limit l"(:r) thus becomes the expeetation
of tbe field operator ,P(:r) in the ground state of H(J} (equation {5.11)):
'P("') (,P(x)) J • (6.70)
equatioru (6.67,6.70) one tben fmds:
r('Pl f3(H)J. (6.71}
lPI functional r(I{J), restricted to time-independent sourceo;, is proportional to the
llip<>et~mo•n value of the ínitial hamíltonian in the ground state of H(.;fJ,J). As equation
sbows, it is expressed in tenns ofthe ""peetation value 10(x) oftbe field ,P(x) in the
state. This result can be u:oed to establli;h variational upperbounds on tbe ground
enesgy of the hamiltonian.
136 Genemting Functionals af Carrelation Functwns G.li

6.6.!2 The lPI functional and the free energy at fixed field time a"emg•
Let us now present a related but slightly different interpretation. \Ve calculate t.hr·
partition function with the same periodic boundary condition in time but restricted t<>
fields satisfying
I rfJ
\P(:r) fj f o dt Q>(x, t). (6.72)

Note that this ímplies trivially 'I'( X) = (<I>( :r, t)). We cal! -fl9(10) the corresponding frc•·
energy
(6.73)

We have written the free energy in the form -/39('1') because we know that in the larg<'
fllimit the free energy iH proportional to {3.
Thom the free energy corresponding to the surn over all field configurations in presenc•
of a tíme-independent source J(:r) is given by

(6.74)

For {3 large the functional integral can be calculated by steepest descent. The saddle
point equation ÍH
b9
J(:r) = 6op(:r}. (6.75)

When the equation has several solutions one has to take the stable solution which yields
the largest contribution to the free energy. Then

W(J) ~ -{J!/(rp) + fl f d:t J(:r)rp(:t). (6.76)

After l.egendre transformation one finds

where aga.in f(~p) is the JPI functiona.l restticted to time-independent fields. Note how·
ever that Ç(.p} has in general no reasons to be convex. One may find field configuration•
such.that the operator
~·a(l"l
6'P(:r)51"(!1)
is not positive. On the other ha.nd because r[ op) is the result of a steepest desrent
calculation, it may coincide with iJÇ(rp) only in regions of field space where the operatm
ís poo;iti..,. In general in perturbatíon theory one ca.lcu]ates a quantity which, restricted
to time-independent fields, ooincides with g rather than r. Thís explains an apparent
paradox; In tbe sewral phase region one often pretends discussing the mínima of f(rp).
i.e. the minima of a quantity which has convexíty properties and can have only one
minimum. Really one discusses the propertie!l of Ç.
Let finnlly note that in the framework of statistica.l mechanics, the Legendr" trHDSform
f(.p) is also the therrnodynarnical potential, a quantity wbich plays a central role in th<·
discussion of criticai phenomena.
 6.6 Generating Functianal.< a} Correlation FUnctioll..! !3i

Bibliographical Notes
The idea of usíng generating functionals with sources was originated in:
J. Scbwinger, Proc. Natl. Acad. Sei. USA 37 (1951) 452,455.
The Legendre tmnsform of connected conelation functions ha:; been díscussed ín the
statistícal mechanics context in:
C. De Dominíds, J. Math. Phys. 4 (1963) 255; C. De Dominicis and
P.C. Martin, J. Math. Phy•. 5 (1954) !4, 51;
and introduced in field theory in:
C. Jona-Laslnio, Nuovo Cimento 34 (1964) l790.
See also:
J. Goldstone, A. Saiam and S. Weinberg, Phys. Rev. 127 (1962) 955.
A proof of the one-line irreducibility can be found in:
S. Coleman, Laws af Rad:ronic Matter (discussion of lecture 3), Erice 1973, A. Zichichi
ed. (Academic Preas, New York 1975).
The physical i:oterpretation of the IPI functional is tllken frorn:
K. Syma.nzik, Camm.un. Math. Phys. 16 (1970) 48; see also S. Coleman, refernnce
above.
lt has been rerognized that the loop expansion l5 also an expansion in powers of li in:
Y. Narnbu, Phys. LeU. 26B (1966) 626.
The one-loop functíanal U; calculated by functional methods in:
B.W. Lee and J. Zinn.Justin, Phys. Rev. D5 (1972} 3121 Appendix B; R. Jackiw, Phys.
Rev. D9 (1974) 1686.

Exen::ises
Exercise 6. I
Write explicitly the field equation of motion (5.28) for the various generating functíonals
in tbe exarnple of the action:

and deduce from the equatíons tbe one-loop correction to the one, two and three l Pl
correlation functíons.

ExeTCise 6.2
Write the field equation of motion satisfied by the generating functionals of correlation
functions in the exrunple of the action:

Recaver the form of the IPI functional at one-]oop arder.

Exerci> e 6. 3
Obtaín the quantum equations of motion satísfied by the IPI functional in the theory
with bosons and fermions described by the action {5.62).
 AG.!I
Generottng Functionals of Corn.;lation FunctionJ
138

APPENDIX 6
HIGJ-IER ORDERS IN THE LOOP EXPANSION

Let us show hov.- successíve terms in the loop expansion can be calculated, using tlw
method of Section 6.4:
(A5.11

The saddle point ,p, (J) ís solution of the equation:

~s (A5.2)
81/>(z) !<Pc] :::-. J (z).

Setting: (A6.3)
rJ>(x) <l>c(z) + .Jiix(x),
we can exp!>Dd S (rJ>) in powers of x:

(A6A)

(A6.5)

(A6G)

Let us íllustrate these remarb by an explicít two-loop calculatíon. Expanding the intcr-
actíon terms Sl3) and S(4 ) and ínt.egrating term by term we obtain:
1 1 @S
lnZ(J) K + ;;i-5(1/>c)+J ·.Pc]- 2trln~
+n{ -~f dx1 ... (x~o X2,X3,Z4) L!. (:q,x2) (x3,:t4)
dx4 5!<) L!,

+f d!/3 _5(3] (ZJ.X2,x3)

dx1 .• , l/2, !/3) (XJ>X2)
S(J) (l/J, a
C>.

x C> (!Ih y ) L!. (x3, y,)

2
+~L!. (x1, !li) C> (xo, lh) L!. (zs, !la)}} +O (1i2) (A6.7)
 A6.0 Generoting Functiona.ls aj Correlati= Funotions 139

Tloe constnnt K depends on the normalization of the functional integral. The expres-
•ion for W(J) lilnZ follows immediately. We haVe now to perform the Legendre
lmnsformation:

r (<P) + W(J) J(x).p(x)dx, =f 6W

.p(x) = 6J(:r) . (A6.8)

A~ we have already noticed we need 'l'(x) only up to arder li because expression (A6.8)
!~ stationary in &p(:r):

li -6- ( trln -
'l'{x) = </>c ( J; x) - - IPS -) + O (li2 ) (A6.9)
2/íJ(:r) lí<jJ,Ií</Jc
The arder li correction c:an be rewrltten:
é l 6 5
2

6J(x) tr n 6</J,Õ</J, =
f 61jl, (y)
dy liJ(x) tt
[ lí3S
6<jJ,(y)6<fJ,61jJ,
( /i•s
6</>,áQ>.
)-1] ·
Using then equations (A6.2,A6.9) and definition (A6.5), we can express <f>, in terms of 'I''

.p(x) = rp, (J,x) ~f S!3l (Y.Y~>Y•;'I') 6. (x,y;&p) 6. (Yl.Y2;'1')dydy, dy2 +O (li2).

(A6.10)
We still need J(:r) a.t arder fi. Using equations (A6.2) and (A6.10) we find:

J(:r) 6;fx) +~f s<sl (:t,!lhY2;'1'} 6.(!1t,!/2;'1'}d!ltdll:! +O (112). (Aô.ll)

..Equation (A6.B) tben yields r (IP} at two-loop arder. As expected the reducíble part in
cxpre;sian (A6.7) cancels (figure 6. 7) and we obtoún:

Fig. 6. 7
o-o
The miucible part at two-loop order.

li. 62 S
r(~p) ""S (\P) + "2 trln liopfl.p

+ 112 uf s<<) (Xt,X2,Xa,x4; \P) 6. (xt,X2; \P) 6. (z3,;r4; 'I') dxld:r,dz3d.r4

- T:i f s<•J (:z:1,.r2, XJ;<p) S(a) (Yt.!l2d/3õ'l') 6. (:z:t, !li; <P) 6. (xz, Yz; &p)

6. (:z:3, y3; <p) d:r1 dx, dx 3 dy 1 dy2 d!/3] +O {li3). (A6.12)

Pigure 6.8 gives a diagrammatic represeotation of the tw~loop terms in tbe equation.

Fig. 6.8 The twt>-loop contributions to r(ip).

7 REAL·TIME QUANTUM FIELO THEORY AND 5-MATRIX

Up to oow ""' have discussed only euclidean lield theory and corrdation functions. To
calculate S-matrix e!ements, quantities relevant to Partide Physics, it is necessary to
return to a real time formulation of quantum 6eld theory. We díscuss in this chapter
scalar bo:rons and massive Dimc fermions. For scalar bosom; a natural formalism is
provided by the holomorphic representatíon. A simple generalization of the path integral
of Section 3.5 then leads to a functional integral representation of the S-matrix. For
Dirac fennioru;, we combine the considerations of Section 3.6 (after continuation to real
time} with the discussion of Section 5.5. Technical detaíls concerning the symmetries
of the fermion sction and the properties of ")'-matrices can be found in Appendix A5.
Relevant for tltis chapter are also the considerations of Section 2 .6.
Tho\lgh quantum field theory appears as a natural formal generalization of quantum
mecbanics, it ís not quite obvious how quantum mecbanics m recovered in the non-
relativistic limit. Indeed one could have also thought about d~loping a formalism of
relativistk quantum particles. The observation of physical elfects d ue to tbe electro-
magnetic field, whlch predates quantum mecba.nics, makes this alternative posaibility
less attrsctive since it would not have unified fields and part.icles. We therefore brielly
indicate how quantum mechanics formally emerges in the low-energy, low-rnornentum of
a m....,ive quantum field theory, in the forrn of a quantum rnany- body theory.
As a warning we also have to stress that severa! quant.ities we meet in this chapter are
really infinite, or at leMt inlinir... in high enough dimensions. Therefore the discussion
will be rather formal. However the divergenre problems will be carefully studied !ater,
in the coming chapters. A regularization by o space lattice would have rendered most
quantities meaningful but complicated the arguments.

7.1 Scalar Free Field Theory: Holmnorphic Representation

The starting point of the construction;, free 6eld theory. The free lield action Ao(4>) for
a scalar field 4> in real time can be written

(7.1)

Because the action ;, quadratic in 4>(t, x), the field 4> can be considered as a collection
of hannonic oscillators. The dilferent harmonic osdllators decouple in the momentum
basis. We therefore perform a Fourier transformation over space variables a.nd set:

In this chapter we denote the time and space components of a d-momentum vector p as
{Po,p}, Po being the energy and p the momentum.
The free action becomes
 7.1 Real-time Quantum Píeld 1'/oELn-y and S-Matrn ]41

The COI\Íugate momentum fi of J, is then

The correspondíng hamiltonian density 1t is

v:ith w(p) = .,jp2 + m 2 • The canonical colllllJutation relatioru; take the form

We now introduce creation and annihilation operatora A I, A related to fi, ;j, by

J,(fi) = 2w~) [AI(-P)+A!P)], (7.2a)

ÍÍ(ií) = ~(Z>r)a-l (AI (ií)- A(-fi)] , (7.2b)

which have the commutation ralations

Note that tbe transformation (7.2) is non-local in coordinate space.

The free hamiltonian H then reads

H( AI ,A)= f d"- 1 ;7'/t !(2lf)d-t f ct•- 1pAI(ji)A(fi) + Et;,

wbere E 0 is a const8Jlt obtained by commutation, whlcb we discuss below. We ha...,
chosen a normalizatíon of operaton; such tha.t tbe hamiltonian h"" the form

~
H(AI,A) (7.3)

~ where we have introduced the O(d 1,1) CO'Vllriant measure dpj2w(p)

i
•
wltb the notation ( corrvenient but inconsístent witb our euclidean notation}

f P2 = PÕ- ii 2 , 6+(p 2 2
m ) = 6(p
2 2
m )11(p.,).

f We define the ground sta.te JO) ~ also called tbe vacuum (zero pa.rticle state), by the
~ property that it ís annihilated by ali operat.ors A(jí)

(OjO) = 1, A(p)IO) =a.

The vacuum energy. The quantity Eo ls thus tbe vacuum or ground state energy
HIO) EoiO).
142 Real-time Quantum Field Theory and S-Matrix i.l

Jt ís formally proportional to [A(p),A 1(iill and thus to ld- 1 (0). Th give a precise meanin~
to Eo, it is necessary to quantize in 11 large bax of linear size L and to modify the theor\'
at short distance or at large momenta so that the Fourier modes are cut-off at som<•
momentum scale A (a space lattíce would provide such a cut-off). The Fourier variabl""
f> are then quantized
p= 21rn/L,
and the vacuum energy becomes

For L llll'ge sums can be repla.ced by integrais and dn = Ld-l /(2")d-ldp. The spa.ce
volume f&etorizes, showing as expected that the energy is an extensjve quantity,

(7.4)

but the energy density is cut-<lff dependent. The large momentum divergence of the
vacuum energy iB not relevant here because in this non-gravitational theory the ha.mil-
tonían can be shifted by a constam in such a way tha.t the vacuum has rero energy.
Note, however, that lf the vacuum energy itoolf is not a physical observable, a varíatíon
(ímposerl for exa.mple by a change in boundary conditions) of th<! va.cuum energy may
be (see Appendíx Al8.1).
We can tben construct a Hilbert space by acting repeatedly witb creation operaton;
At on tbe vacuum. The vectors

span the space of n-particle states. The n-partide state energy is the sum of the energies
of eoch particle: lndeed from

we obtain

(H Eo)At(il,)AI(f>2} ... A1(pn)IO} L w(f>i)AI(ji 1}Af(i>2) ... At(pn) ID).

i=l,n

The direct sum of ali n-particle spaces iB called the Fock spa.ce.
Finally the free hantiltonian commutes with the particle number operator N

(N,H]=O,

a property in general no longer true in presence "f interactions.

Note that if one tben wrítes a general one-particle state IW)
 Ren.l-tím~ Quantum Fidtl Thepry and S-Matriz 143

11"' &calar product take5 the 1elativistic covarlant form

A ~eneral n-particle vector Jo1>) then has the form

(7.5)

,P(pt, P2, ... , fJn) ís a totally symmetríc wave function, in momentum representa..

Two-point function. Note that the 4>-field 2-point function, expresse<! as the expecta-
value of a tíme-ordered product of two fields (see for example S.O.:tion 2.5), is given

,il)~(O,p')) [O) = (OI ~(P) e-<HI!I J(ii'JJIO) (21r)t-d ~(p)5d-1 (p + fJ') e-"-'UlJI•I.
(7.6}
Fourier transformation over time one finds

the i• term in the denominator indicates th8t we have to add 8 small posítive
part. The real time 2-point function is 8 distribution in the mathematical
boundary value of an analytic function

_,
n-w~+u
j
''"}
- '( 2 2'""}}
. =2orap0 -wll' +•·pp ~-w~
.:J
l
"'"')'

Halomoryhic reprearntation. We now introduce the corresponding complex classical

A"(t,p),A(t,p), related to the field (i, by the clasoical form of the relations (7.2).
holomorphlc form of the free actíon then is

dd-1·
Ao(A',A) = (21t)d-l
f dt 2w(p~ [iA'(t,p)éJ,A(t,p) -w(p)A'(t,p)A(t,p)j.

evo)utlon operator is given by a functional integral of the form

U = jfdA"(t,fi)dA(t,p)j expiAo(A", A).

144 RtW.l-tímt Quantum Fteld Theory and S-MGtrix 7.2

7.2 The S-Matr-ix

Srottering by an exlernal S<mrce. Let us first consider the S-matríx corresponding to th<·
addition to the f:ree action of a source term f dtdx J(t, x),P{t, x). After Fourier transfor-
mation
J{t, x) =f eif" Í(t,p)dp,
thís leads to the action A
d'
A( A' ,A}= Ao{A',A) + (21f)d-l
f dt2wrp) i(t, -P) [A'(-p,t) + A(p,t)).

The cakulation closely follows the lin<:S of Subsectíon 3.5.3. The result ís

So(l,A' ,A)= exp [(2:rr)d-l f~~) K(p)),

K(p) = A'(p)A(ii) +i f dti(t, -ti) [A'(-p)e""(Silt +A{p)e-""(p)tl

- ~ j dtdT J(t, -fi)e-<1!-rlw(p) Í(T,p).

As explained in Subsection 3.5.3, the expa.nsíon of this functional in powers of A • and A

then leads to the S-mstríx elements.
It is octually more wnveníent to write the result in terms of the Fourier componenl.t
of Í{t,j))

Then

lnS0 (J,A' ,A)= {21f)d-l f~~) A'(p)A(p} + i(2:rr)d f dP<JPo J(-;po, -ft)

x ó(p~- w (ii)} [A'(-ji)8(-Po) + A(p)81Po))

2

- f
~(2:•f dP<JPoÍ(-;po,-p)p~ w•;(p) +íJ!Po.fi).

In the coofficient o f the term quaàratíc in 1 we recogni:ze the free 2-point functian (7. 7).
Generol representation. An interaction tenn V1(.p) ean then be added to the free
where .P has to he expressed in terms of A', A,

Using the functional expressions for the perturbatíve expansion proven in Seçtíon 5.2
find the fonn of the S-matrlx ror the inte.raction theory
 Roa/-time Q-uantum Fíeld Theory and S-Matrú 145

S-matriJ.: thus ha.s " Fe~'llman diagrarn expan.sion with internai propagators A given
tl1< quadratic term in J in (7.8):

(7.11)

llmtarity. Let us note thàl with om conventions the unitaríly of the S-matriJ.: takes
I Ul)ClÍonal form

fJdA'(p)dA"'(p)jS'(A"',A}S(A"',.4)exp [-{21r)d-l f~~) A"'(il)A'(il)]

= exp [(2nf-l f 2~~) A"(p)A(p)], (7.12)

'·lh"'cti.Ssi<Pn. We have constructed our b!lSÍl! of states from tbe eigensta.tes of the unper-
bamlltonia.n. More generally we can tal<e any similar harmoníc oscillator basis,
price of adding to the ínteraction quadratic terms in lhe field. Actually, and thís
become clearer when we discus.; lhe structure of lhe ground state in field theory, if
l.fike an arbítrary ba.sis, in general ali eigenstates of the interacting harniltonian wíll
'orlhogonalto ali vectors of the bru;is (a property specific to systerns wilh an infinite
of degrees of freed<:>m, see Chapter 23). We have therefore to tal<e as vacuum
true ground state of the complete he.miltonían, a state we can only conslruct

hamiltoni!lll of a massive theory has a unique, translation inva.ríant,

excited state (in the case of severa! fields this can be generali7.ed to ali
M!tJt~t!k.ct'ion secters). The physjcal mru;s m (or inveTEe correlation length in the sta-
is defined as the energy of this state. We define tbe zero momentum
as creating this stete. The general creation operator AI (p) is obtained
lhe zero momentum operator, i.e. performing a O(d- I, 1) tra.nsformation,
a on.,..partide stllte of momentum p and energy .Jm2 + p2. Additíonal eigell-
an energy at lea.st equal to 2m.
conditions implicitly define a reference free theory with action Ao, and ensures
the a.symptotic states at large times. Opetationally, in a perturbative
calculate scattering amplitudes one has te perform a ma.s• renorma.lization,
the physical mass ru; a parameter of the perturbatlve expansíon, by inverting
between the physícal mass as defined by the pole of the 2-poínt function,
coefficient of 1/? as it appears in the action.
that the functional integral hru; te be normalized by lhe condítíon S(O, O) =
means that we divide by a faclor related to difference in energles between the
11.11d the unperturbed ground state.

and Cottelation Functions

1»1 compare the explicit form of the S-matrix as derived from the bolomorphic rep-
with a direct evaluation of correlation functions in the real time forma!ism:
146 Real-time Quantum Field Theory and S -Matrix 7.3

Unlike the euciidean functional integral, the functional integral for the real time cvolution
operat.or has convergence problems, wbich "'A"C soh,.e b.:- con~idering it as obt.ained from
the euclidear1 integral by analytic continuation. Such consideratíons lea.d to the i< rui•·
for real time Feynma.n d\agramti.
The real time propo.gator. To the íree action (7.1) formally corresponds the propagatm

which is singular on the mass-shell p 2 = m 2 . Let us give a rnea.ning to this form by

constructing the real time theory by ao analytic continuation of the eudidean theor:-•.
The real time theory is obtained frorn the euclídean theory by an analytic continu-
ation on tjme variables t .- t e'6 where (} varies between O ( the euclidean thoory) and
1t f2, (the Mínkowsky theory). For Feynmllll diagrarns in momenturn representatíon, th<·
corresponding rotation of the energy variable Po is Po ....., PIJ e-••. However we canno1
simply replace !'5 by - p~ in Feynman diagràffis beca use they then become singular 011
the msss-shell PÕ fj2 + m 2 • lf we follow the rotation we find that must give to f'Õ a
small positive imaginary part

Tbe imaginasy term also ensures the convergence of tbe gaussian functional integral. W.·
note that this propugator is identkal to the internai propagator (7.11) which appear"
in the Feynman graph expansion of the S·matrix. Then in real time expression (5.22)
(Section 5.2) becomcs

(7.13)

This expressíon can be rewritten

We then ei<pand

Comparing this expression with expression (7.8) we see that the main formal differenc•
between expr<>.ssions (7.13) and (7.10) comes from the term linear in J:

m 2 ) [A"(-p)O{ -Po) +A(p)8(Po)], (7.1f,,

We therefore conclude that S-matrix elements can be obtained from real time correlati<>l•
functions by first multíplying them by the product of externai invcrse propagators, awl
 jlea.l-time Quantum Fiel,i J'heory and S-Matrix 147

'"" restricting the externa! momenta to tht~ mass-shell p 2 m 2 • This does not imply
1•.• 1 the result vanishes. lndeed correiatíon f 110ctíons have poles on the mass-shell. Th~
"""I an.;wer ís proportíonal to the mass-shell amputated correlatíon fm,ctions.
Nnnar/c. Thc relat.ion between correiation functions and S~matrix elements shows that
l1•' lltatrLx elements as defined here have disc~nnected contríbutions. The new func::ti:onal

T(A",A) = ilnS(A" ,A). (7.16)

,,,st.e!ld the generatíng functional of conneeted scan.ering amplitudes.
' 'mssin9 symmetT)J- We see that only a linear oombinatíon of A" and A appcars, with
·I· ut.ícal coeffident, up to thc sign of the energy. The sigo of Po specífies the incoming
,,.j outgoing particles. This has decp coru;equences, specilic to telatí\'Ístic quantum field
1..-.wy: in d > 2 dimensione, the '>ame analytic [unctioru; lead to scattering amplitudes
' ·lilferent pbysical processes, a property known as crossing symmetry.

; I S -matrix and field ruymptotic conditions

llv· functíonal repre!lenta.tion (7.10) can be r"written in a different way. We use the
,. ... ,.tonal ídentity

e><p [-!{h·)d f dp](-p)p ~2 +i<](p)l

= f[d<i>] exp [i(2n)d f ddp [ !~( -p)(p 2 m2 )~(p) + J( -p)if>(p)l} .
,_, ,,,dudng thi.s identity in [7.10) and applying the functional derivatives we lind:

S(A",A) =T(A',A) f!dtf;jexp[i~(<f,) -il'J(cf>+'l')],

,P(p) = 6(p2 - m 2 ) jA' (-p)li[-Po) + A(jl)li!Po)], (7.17)

,11, I representíng the identity

1 ; ... •:xpression (7.17), up to the prefactor T, ú; a functional integral in presence of a

"l•!;tound lield '+'· Moreover the dassical field 'i' is restrict.ed to be a general solution
: 1 he free field equation
[p' -m2),P(p).., O.
2
1,,. ,olution is non-vanishing only on th.. mass Jlyperbolnid P m 2 • The parametríza-
.. ,. (7.17) of the solutlon reliects the property fbat the ma.ss hyperboloid has two dis-
'"'""'ted components, depending on the sign of the energy Po·
\1.: cat\ thus rewrite the equation (7.17)

S(A',A) I(A",A) f[dq,)exp[Ü(cf>+'P))

' lu • functional integral differs from the vacuum ~plitud~ only ~y tbe boundary condi·
'"·· ...hich are Iree field boundary condítions. This r-e.ult 15 COnslatent with the analysís
148 Real-time Quantum Fíe!d Theofll and S -Matrí:r 7.3

of Subsection 2.6.2. We h ave shown thut in quMtum m~hanics S-matrbc elements can
be cakulated frorn the path integra.! representation of the evolution operator. by ínt<'
grating over paths whích satísry prescribed classica.l scatteríng bounda.ry cond itíons, i. e
which correspond to asymptol.ic free classica.l motion. In particular the startíng poínt
of the semíclassíca.l expansion is a classical scattering trajectory. The arguments can
be genera.lized to quantum field theory with masslve particles (to ensure proper duster
properties and thus the existence of an S-m,.tríx).
Let us now return to the discussion of the end of Section 7.2. We ha"" argued that.
in contrast to simple quantum mechanlcs wlth scattering states, the asymptotic states
of the complete 6eld thoory are not gí""n by leadíng arder perturbation theory. On~
must therefore take for op not a solutíon to the classícal equation or motion, but instead
a. solution to
éS(r,o) , ,
0 (7.18)
clr,o(x)
which contains tbe information »bout the scattering states of the complete theory. For
instance the ínteract.ioru modify the masoes of particles. The solutíon of the classica.l
equation Euffices only for a the leading order ca.lculation (the tree apprrndmatíon). At
higher orders, os we have already diseussed, the renormalized mass has to be introduced.
RemarA:. Considerations based on asymptotic field boundary conditions, or the more
direct considerations of Section 7.1 lead to the sarne perturb,.tive S-matrix. However
the preceding considerations generalize to the scattering of so!iwru, i.e. states obtained
by expa.nding the functional integral a.round finite energy solutíons of the classical field
e<JUations. ln this case the S-matrix of scliton scattering ls obtaíned by expanding the
functional integra.! a.round classica.l soliton seattering sclutions.

7.3.!'. The 4>3 example

Let us illustrate the analysis by calculating a 4.point seatteríng Amplitude in the simple
,P3 field theory, in the tree approximation. We consi<ler the action

V•le íntroduce the Fourier transform of the field V'·

(7.19)

In t.he Fourier representatíon the field equation takes the forro

(7.;!0)

This equation can be solved as a series in the coupling constant g starting from a free
solution. The free solution lp0 can be written

.pg(p) = 6(r} m 2 ) (A'(-ji)O(-po) + A(p)IJ(Po)].

Let us calculate the classical solution at order g

ip(p} \Õo(p) + _21 ~

P -m
j ddq.Po(q)ipo(-p- q} +O (g').
 Rrol-!ime Quantum Fidd Theory and S-Matri:x 14.9

llr looking for penurbative solutions of the field equation we have explicitly ""cluded
ocnt.tering states corresponding to bound states or sollmns (see the remlil"k at the end of
llw preceding subsection).
Using equatíon (7.20) w.. can rewrite T in the tree approximation:

We then repla.ce 'P by its expansion in pcrwers of g. The term of order g, whích would
tlcscribe one rf; partíde decaying ínto P>'o, vaníshes by energy conservation. Th<' next
wrm of ord., sf ha.s the form

Tfte ronnerled 4-particle scattering amplitude is then obtained by differentiatíng with

r...pect to ,Po{p). The .esult is the product of a factor which contains the momentum

~md an amplltude

1'1te term we have calculated contains in principie eJso the decay of one ps:rticle ínto
three but again thís process va.nishes by energy cunservation.
Hígher orders in g yíelds 5,6 ..• partícle scattering amplitudes.

?.S.9 S -rrw.trú: anel JPJ geroeroting functional

Dy comparing the general form of the S-matrix with correlation functíons we have seen
1.ha1 S-matrix elements BJ"e relat<!d to real time analytic continuation (t >-< it,P!) ,_. tP!))
oi amputated correlatíon functions on the mass-sbell

~·hereJ ís deduced from equatíon (7.15), and rp is gtven by equation (7.17).

Let us now show that there IS also a direct relatíon to the analytic contínuation of the
~eneratíng functional of proper vertícffi f( I")- We start from the eudidea.n form

eW(J) f [d~i .,-A(~)+N. (7.21)

We then substitute r(;p), the Legendre transform of W(J):

W(J) + q,.,) =f dtddxJ(t,x)',O(t,x), ',O{t,x)

óW
l!J(t,x)

.-r('I')+J., = f ldr/J) e-A(<Il+N. (7.22)

150 Real-time Quantum Field The<Jry and S-Matri:r 7A

Using:
J(t,x)
õr
we can write equalion (7 .22):

e-r('l') = fid,Pjexp [-A(,P) +f dxdt(<l> '1') 6 'P~~x)], (7.23)

or eqrnvalently translating 1/>(t,x):

e-r(")= f!drfo]exp [-A(ti>+<P) +f dtdxrp(l,x) 610~;,,1 ]. (7.24}

We now ta.ke tbe limit of a vanishíng source J:

~=0. (7.25}
6<P(t,x)

Of course tbis equatíon ha!l propagating type soluti<ms only sfter continuation to reoJ
time. We tben observe tbat f(cp) coincides witb íT(cp), when <p(x) is r<!Stricted to th~
solutions of equation (7.18).

7.4 Field Renormalization

To describe predS<lly how the S-matrix can be calculated in a general thecry of scallll
particles, we still have to dis<:uss the field renormalization. This can more o:onveuiently ·
done iu the euclidean form. We define the field 4>(t,:r) in such a way that ít has zero·
expectation value. This can always be a.dlíeved by a coiJSt.ant shift 4>(t,x) - ,P(t,x)-'
(011>10). We then consíder the 2-poínt function correlatÍ(Jn function W( 2l. It follows from
the analysis of Section 2.5 that it is equal to the vacuum average value of the time-nrderod
product of two fields

(7.26)

We now adapt the arguments of Sectíon A2.2 to the kínematical properties of a

invaríant field theory. We introduce a complete set of eigenvectol'5 li') of H, wíth
eigenvalues ",. and obtain

The theory being tra~JSlati<>u invaríant, we know that the answer is proportíonnl
+fi'). ln this limit, since .f,<-p) = .;it (p), the Lh.s. is a sum of positive
6d- 1 (ji
(a point we have already discussed in Section A2.2). We thus find

where p is pos.it.ive messure. Let us now calculate the Fourier transform over time
 Real-time Quantttm Field Thecry and S-Matrix 151

O and replac<> pg by p 2 without changing the

P(JJ), proportional to p(j.<,O), is a positive measure. This [ortn is called the Kãllen-
(KL) representation. Since by definition the physical mass rn is the lowest
· eigell6taw above the ground state, the domain of integmtion is /.1 ?_ rn. Morwver
O the state is isolated. Therefore the measure has an isolated <5-function and then
nultw•uo•us pa.rt start.ing at the threshold for scatwring states (in the símple sca.lar field
:!m, or 3m íf the parity of the number of part.icles ís conserved}

1'(1') Zó(!J- m) + p'(IJ.), 1

p (p) O for Jl <2m. (7.27)

let us Fourier transform with respect ta the energy variable PO

IIOW return to the definition (7.26), assume t > O, talre the derivative with respect to
bJld take the limít t O. We find

éd-'(P+ jl'} .i wt2l(t,jl)l ~(21r)d-l (OI ~(jl)(H- Eo)J>(fr'J lO). (7.29)

ât lt~o~
product (H Eo)J> in the r.h.s. ca.n be replaced by the commutator [H, J;j. Then
lhe reality of tbe l.h.s. we can t.a.ke the hermitian part af tbe operator. Thus

ód-l(jl +i/) ~wt2l(t,p)l -W"ld-t (OI [~(ii),{H,:j,(i/}JliO).

t-o+

n.ction h'"' the form (7.9) tbe commutator is prop<~rtionnl to the conjugated mo-
Then

~WCll(t,J'))I -t.
ât •-o..
now the Lb.s. from the representatíon (7.28), we obtaln

that, except ín a free lield tbeory, Z, the resid ue of the pole at p> = -m 2 ,
sma.ller th8ll 1
0< Z< 1. (7.30)
ltl<iult has severa! implica.tions, ane being related ta the S-matrix. Let us evaluate
1, fJ) for t la.rge. Tbe staw o f lowest energy, with momentum f!, gives the leading
152 Real-time Quantum Fídd Thenry and S-Matrix IA

contríbution. From the KL representation (7.28) and the decomposition (7.27) we then
learn

If we compare this result with the contributiou of a normalize<! one-particle eigenstate

of the hamiltonian H (see equatíon (7.6) ),
ll,p'}:. {jd-l(ji +f/) ~ e-l<lv'i''+m',
(Z1f)d-l (l,ple-I•HH-Eo)
2
we observe that the field q, has only a component v'Z
on the one-partide states. Another
way to formulate the sa.me answer is to verífy that in real time the Heisenberg íield
has • free íield large time behavíour with an amplitude v'Z
on normalize<! creatíon or
annihilation operaton;. Let us write the real-time 2-poínt function:
{OI T(~(t,p)~(O,f/)l to) = (OI ~(ji) e-•HI<I.j;(p')Jio)

=(Z1f)t-d6•-'(ii+p')/dp. ~e-•l•lv'P'+~·.
2 ' +i'
It can be verified that the la.rge time behaviour is related to the leading síngularity of the
measure p. Sínce p(p.) is the sum of a õ-íunction and a contínuous function (for d ;:= 2),
we obtain
(Di,P(ii} e-<Hi<l .j;(p')liO} = Z(21f }1-d_I__ {jd-1 (p + p') e-'"'(llH•I +0(1/t).
1•1-<>o 2w(ji)
We condu de that for large time the field ,P(t, p) tends in weak sens<l (not in op.,rator
sense, but in ali average va.lues) towa.rds

~(t,j!) 1<1-<»
- v'Z 2wl(ji) [A 1(-fi)+ A(ii)] ,
where AI, A are the properly normalize<! creation and Mnihilation operaton; of the one-
particle states. The const81Jt v'Z i.s the field renormalization oonstant.
Nof'Ttlalúed S-matrú elements. To calculate properly normalízed S-matrix elements
we can calculate with the action A( cl>v'Z}. Alternatively if we krep the initia.l field we
have to renormalize the matrix elements.
We e><:press the correlatíon functions in terms of euclídean amputated functions

wln) (p., ... , Pn) [n

•-l
w,<') (p,)] w!::J•. (p,, ...• Pnl ,
IL!ld call -m~ tbe pole in the varíable Pi of W, (p,}:
w?) (p;) - •• for Pl' - -mi, (7.31)
1if + m?
Le. the m; 's are the physlcal masses of the scalar particles. Then the coefficient 7(n) of
7 ín the expanslon ín powers of rp is given by:

T(n) (ph ... ,Pn) -TI v'Zi w~. (p~,

t=-1
... ,Pn) lp~=-mr . (7.32}

The factots y'Zi in thi.s equation correspond to a finite renorma.lízation of the field wch
that the residue oftbe 2·point funct.ion (equatioo (7.31)) oo the physical pole Pf ~ -m~
is 1. They ensur.., that the matrix elements T(") satisfy the unita.rity relati0115 witb. tbe
proper oorJUSlizatioo.
.'"' Real-time Quantum Field The.ory and S-Matrix 153

7.5 Massive Dirac Fermioru;

Fír<1 let us note that with our conventions, for a finite nutnber of fermions, for a normal-
'''dered (á.ll <:reatíon operators at on the left of all annihilation operators a) hamiltonian
nf the form
ll=h(a1 ,a),
tl1e evolution operator in real time is given by the path integral

U j ldc(!)dc(t)] exp i.A(e, c] (7.33a)

.A( c, c)= j dt {;ea(t]ta(t)- h (ê(t),-c(t)]}, (7.33b)

real time continuation of equa.tion (3.109} llf Section 3.6.

We now study massive Dirac fermions, iiJ even dimensío~, since otherwise the fermion
miiSs breaks spoce reflection lnva.ria.nce. In Appendix A5 we have verified the spin O(d)
rmd ot.her symmetri<lll of tbe fermion eudidean action. In rélll time the O(d) symmetry
hfX'.omes a O(d -1,1) symmetry. Here we will verify real time hermitícity of the harnil-
tonian, structure of fermion states in the free field theory, and construct the S-matrix.

1.5.1 Tlu? jree massitH? Dínu: action

Again we first discuss the free action, expressed in terms of Grassrna.nn anticommuting
spinor fields -i[,, 1/J:

"·here we have denote<! by ')'o the matrix 1'd assodated with the time varia.ble.
'Ib identífy the action witb an action resultíng from a ha.miltonian formalism we set

Then from the construction of Sectíon 3.6, a.nd generalizing expressíon (7.33), we identífy
the corresponding hamíltonian density

where -if,(x), 1&(x) are now operators with the corresponding anticommutation relations

To díagonalize tbe hamiltonia.n, we proceed by Fourier transformation, setting

We find the total hamilt.onia.n H

154 Renl-time Quantum Fidd Theory and s,~
and the a.ntícommutation relations (from now on we write onJy the non-trivial part) l
l
I~
We see tbat, in contrast with the scal& case, due to the spin structure the hruniltonian
is not totally diagonalized, but the diagonalízation has been reduced to a símple matri.x
problem. Let us denote by h(jj) the matrix

h(jj) 1o (i1' ·fi+ m}, (7.34)

Then we verify
h2 (p} = "}(jj), w(ji) =
We thus find that the two possible eígenvalues of h are ±w(p). The two corresponding
subspaces have equal dimensíons since

Thernfore if a •pinor u{P) is a.n eigenvector with eigenvn.lue w(jj), 1o1su{P) ( 1\:115 is
associated with time refiection) is a.n eigomV>!Ctor wíth eigenvn.lue -w(jj).
Let us íntroduce the two orthogonal, hermitian projectors P± on tbe positive a.nd
negat~ energy sect.or:

P± Í [J ± h(p)fw(p}]. ~ P+ + p_ J, P;, P±. (7.35)

We note that if we a&<;Ociate crnation operators to ,pt and annihila.tíon operators to ,P,
then ""' create states with negatíve energy. This símply means that we have improperly
chosen the reference state which should be the ground state. As in the <;ealll.f case, and
in contr..st to what one could have naively guessed, we have to deoompose both ,P and
.pt in a sum of creation and annihilation operators, in such a wa.y that creatorG always
create states with positive energy.
Vole thus set

[2w(Ji)ll/2 Tcp_ ~(ti), [2w(jj)jl12p+~(jj) (7.36a)

[2w(p)J'I 2 ,,i} (p)P+, [2w(fill'ni],t [p)P_ c· . (7.36b)

where C is a unitary matrix, ele= 1. Conversely

i)',[p) .j~(p) (C' A~ (f>) + A+ (fi)) (7.37a)

i)',t(jl) .j~(p) (A!..{P)TC+A+(f>)). (7.37/;)

Anticommutation relations follow (we write only the non-trivial part)

{[A+ lo (jj), !-4~J~(jj')} 2w(jj)[P+l.. e(21f)I-d6d-I (p- p'),

{[A-]a(jj),{A!.,jp(ft')} ~ 2w(jj)[Cf Tp_CJaa(21r}l-dód-l(jj fi').
The first commutator can be rewritten

Pu = w(p).
 155

We recognize in the parenthesis tbe covariant form o[ Lhe projectOr. .

The two set of antico!llmutat.Wn relations are dilferent except if we can find a matrP<
C such that
ctTp_c=P+
Sínce C is unítary this equation reduces to

In Appendix A5.6 we have constructed a charge conjugation roatrix C which has lhe
property (equation (A5.30))
c-l T1,.c = --y.,.
This is exactly the matrix we need here. In these variables, taking into account the
unitarity of Lhe conjugation matrix C, the free hamiltonian reads

lf we now normal-order the hamiltonian by a.ntioomrnuting A~ and A_, we find

The ground state (vacuurn) energy Eo(Dirac) is negative, and proportional to the free
scalar vacuum energy (7.4):

Eo(Dirac) _z41> Eo(scalar).

Note therefore that by adding 2d/ 2 scalar bosons of the same mass m to one Dírac fermion,
one ca.n construct a theory wíth zero \'I!CUllm energy. One can show tbat this boson.
ferrnion free theory has then a special type of fermioníc symmetry called supersymmetry.
The final form of the hamiltonian, together with the final form of the comrnulation
relations, shows that lhe Dirac field carríes two partides transforming under the funda-
mental representation ofthe O(d -I) spin group (spin 1/2 particles), related by charge
conjugatíon (in the case of charged particles they will have opposite charge).
Another rema.rk: the last step depends crucíally on Lhe anticommuting character of
fermions. Tbis is the N)flection of the connection between syin and statístics, a specífic
property of local relativístic quantum field theory.
We are now in a situation quite similar to the scalar case. Startíng from the ground
state !O) which is annihílated by ali operatars A± {P)

we CIW. construct a basis for n-particle states by acting with creation. operators on the
vacuum. A general n-particle state is obtained by íntroducing a lotally antisyrnmetric
wave function.
The particle number operators for both particles commute wíth tbe hámiltonian
156 Rroi-timc Qua.ntum Field Tlteory and S-Mo.tri:r
1
S -matri:r in an interocting theory. The expression of the S-matríx in an interactínr
theory then follows from u simple extensíon of the method explained in the scalar case.
One has to be only careful of the signs. lt is straightforward to veril'y tha.t the S-matrix
is given b~-
(7.38)

i.e. again a functionaJ in~egral in a background field, where the classkaJ a.nticommuting
fields are solutions ta the free field equations, which can be parametrized in the form
(7.37),

X(p) = p_c• A~(p) + P+A+(jl), Xt{ji) = A~(fi)P_ + A+(fi) TCP.;., X= x11o,

x,
where however the quantities x, At, A are no longer field <:>perators, but the correspond-
ing a.nticommuting varia.bles. A constant Z is here also requíred, to obtain S-matrix
elements with the proper normalization.
In the sarne notation the unitarity of the S,matrix ta.kes the form:

f[dA'(fi)dA 1'(fi)]S"(At',A)S(AI',A)exp [{2n-)d-l f ~ip) ,~1 A!'(fi)A~(ji)]

= exp [(2n-)"-l f ~fp) ,~1 A!tfi)A,(fi)]. (7.39)

7.6 Non-Relativistic Limit of Quantum Field Theory

Since a general discussion of tbe non-telativistic limit of qua.ntum field theory would b<'
somewhat involved, we here consider only two examples whích i!lustra.te the main point:
The low-energy limit of quantum lield theory is many-body quantum mechanics, and
leads to & formalism naturally adapted to the statistical mechanícs of quantum particles.

7.6.1 Scalar fields

We first considera massive scala.r field theory with a q,~ t)'P€ interal!tiou. To discuss the
non-relativistic limit it is necessary to employ the real time formalism. The real-time
evolution operator is given by • functional int-egral of the farm

u f \d<i>] e:xp i.A(q,)

A(<i>) =f dtdo: [~(a,.p)' t ('V,q,)2 ~m2.p>- ~g<P4]. (7.40)

At least for a coupling wea.k enough, the int.egral is dominated by fields satisl'ying the
free field equatíon
(8{- V!+ m 2 ) ,P(t,o:) ""O.
In the non-relativistic limit the space va.riation is small cumpared to the time variation.
1f tbe space v&Iiatíons are comp\etely neglected, the solutions to the field equation ar<·
fJi Real-time Quantum Field Theory and S-Matri:r: 157

Mmply ql(t.x) ox é""'. lt is thus natural to íntroduce the holomorphi< representatíon of

lwlds, taking as the unperturbed harrnoníc oscillator .Au(ql):

Ao(~) j dtd:r [!(/'1tf/!) (t,:r)- ~m 2 ç1 2 (t,x)j.

0

llenotíng by A"(t,x),A(t,x) the wmple:x fields, ln terms of which the field çl{t,:r) and
ll~ conjugate momentum rea.d

ql(t,:r) {2m)-li2(A"(t,:r) + A(t,x)),

ll(t,x) iy'm/2(A"(t,x)- A(t,x)),
wr find thc actíon

To separate the fast time frequencies we then ch<UJge variables A(t,x)- e-•~• A(t,x),
A"(t,x) .-.. e""1 A"(t,x), where the new A, A" are slowly varying in time rompared to
the fact.ors e'"''. After this roange of va:riables the monomia.ls of the form A• A*' become
multiplied by a factor e•~!•-•l•. For T # s the corresponding time integrais give small
oonttibutions dne to the rapid time oscillations. Hence at leading order the only surviving
t~rms are those whicb have an equal number of A and A" factors. The non-relativistic
~>ttion is tben

A( A, A")= f dtdx (iA'&,A-

2~ V,A"V,A- 16~lA" A" AA). (7.41)

We recognize a real time action written in terms of complex variables, of the form (3.74).
Up to an infiníte energy shíft (the vacuum energy), it corresponds to a quantum hamil-
lonian of the form (Section 3.5)

H= f dx ( 2~ V.A'V.A + 1!. 2 A1AIAA), (7.42)

w}iere A I, A now are crea.tion and e.nnihilation operatom whlch satísfy the cornrnutatlon
relatíollll:

{AI(x),A'(y)) = [A(x),A(y)) =O, jA(x),AI(y)] = b(r y).

The hamiltonían (7 .42) has been written In normal ordered form in such a way that the
corresponding vacuum energy vanishes.
We can construct a Hilbert space hy actlng repeatedly with creation operators on a
rcference state (zero particle st.ate) ]O) which is annihilated by aJI A(x). The vectors

Al(x!)A'{xz) ... A 1(xn) lO},

•pan the space of n--particle states. The important property is th!lt the hamiltonian
(7.42) then commutes with the partícle number N, a property whicb, in quanturn field
lhoory, in general is shared only by frw field hamiltonillllS:

N =f d:~:AI(x)A(r), [N,HJ O.
158 Real-time Quantum Field Theary and S-Matn:r 7.6

lndeed when all momenta are small compared to mru;ses, pair creatíon of particles from
the vacuum ís na longer possible.
Acting on the n-particle vector IW).

where T{•(x 1 ,.. x.,) ís a tota.lly symmetric wave function, the ha.míltonian leads to the
Schrõdinger equation:

This ís a n·particle harniltonian with two-body 6-function repulsive forces.

This analysis shows that the low-energy. non-relativístic límit of quantum field theory
is many-body quantum mechanics.
Remark. A hamiltonian with a 6-function potential is only well defined in one dimen-
sion (d = 2 in the ínítíal field theory). In higber dimensions it leads to divergences. But
the initial 6eld tbeory has also divergences.

1.6.2 Fennions
A similar analysis applies to the case of Iermions and the result ls similar, as we now
show. Since a ferrnion mass breaks space reftection symmetry in odd dimensions we
restrict ourselves below to even dimensions.
Tbe starting poínt again ís the non-relativístic limít of field equatíons. Let us take the
example of a self-inrerecting fermion field in real time '

(7.43)

Due to the spin structure and the linearity in 'Vr of the 6eld equations, extracting the
non-relativístic lindt requires slightly more work than in the $Calar case.
However, we can avoid the cakulat.ion, by extra.polating the results of the analysis
of the scalar case. We transform the lagra.ngian into a hamütonian written in terms of
creation and annihilation opero.tors. In the kinematic part we expand the one-pa.rtlde
energy
.Jr+m2 m+ft2 /2m+O(m- 3 )
and translate the one-particle energy by the mass. In the interactian terms we take
the non-relativistic limit, neglectíng ali momentum-dependent rontributions relative to
the mMS. The transformation between fields ;j,, r/J and creation, lLilnihilatíon operators
A f, A then becomes local, The projectors P" defined by equatíon (7.35) berome in the
non-relatívistic:: limit
Po1 = ~(1 ±"lo)·
The limitíng fermion creation and annihilation quantum operators AI, A are now defined
by

Tcp_.j;(x), A+(x) P+;j,(x)

.j;f(x)P+, A-(x) ~t(x)P_c•
7,6 Real-time Quantum Field Theory and S-Matm 159

For 1 mattices we choose_ a ba.is in which 10 is diagonru and restrict below t~e spínor
•nd1ces to the non-v<>mshmg tompollents of A,.,. The antícommutatíon relatlons then
I ake the simple fonn

{A,.,(z), A,•!J(ll)} =O,

Finallr we neglect ali interaction tenns with an unequal number of Al's and A's.
Using the relation
,P(z);b(x) =L
AZ{z)Ac(z),

and horrowing the result of Subsection 7.5.1, we find the non-relativmic hamiltonian, up
to an infinite energy shift,

Thi• action describes a many-body theory of two fermions of opposíte charge and with
.<pin, the spin p!aying the role of an internal quantum number decoupled frorn space-time.
One then proceeds in analogy with the commuting case. One again vermes that the
,un-relativistíc tbeory conserves the number of partides and therefore sectors witb dif-
h·rent partícle number derouple

1"" summation over the spinor index cr and tbe particle index ±). The vectors

A!,n, (z!)A), 0 ,(x,) .. . AL.,~(xN) 10),

wln:re lO) ia the refetence state annihilated b.r ali aperators A(z), span the space of the
11- pa.rticle states. A general n-particle vettor I'~>) can he written

14•) J dx1 .dxn 1Óo 1a 1 •.,o,,...,.<>. (x,, .. , ,xn)A1,a, (:c!}~ 20,(x2) ... A!.a. (x.,) 10),
"'h<'re lf>,,.,,.,n....... a.(xl, ... , x,) is a totally antísymmetric wave function.
l n tbe n-partícle sectQr the Schrôdinger equation reads

Tlw fermions interact through a two-body 6(x) functíon potential wbích can bere be re-
pubive or attractive. The spin and particle numbets act only through the Pauli principie
,..l,ich dictates the possíble symmetries of the wa.ve function tf;.
A fina.! remark: in two dimensions the fermion theory we have cansidered here, is equiv-
•h•t to the well-known massive Thírring model (see Chapt.er 31). Tbe non-re)atívístic
l""it, the .5-functíon model, can be C)Ca.ctly so!ved by the Bethe ansatz, i.e. a complete
.. ~ uf wave-functíons is provided by a superpasítian of a finite number af plane waves
In ""~.h of the n! sectars carrespondlng to ali possíble ordering of particle pasitions. lts
lt·lnüvistic generalízatian ís also integrable, because partícle production does not arise.
160 Real-time Quantum FieJd Theorg and S-Matrix 7.6

Bibllographical Notes
Two importam references for this chapt.er are
F.A. Berezin, The MEth.od of Second Quanti.tation (Aca.demíc Press, New York 1956):
L. D. Fa.ddeev in Method. tn Field Thcorg, Les Houches 1975, R. Balian and J. Zinn-
Justin eds. (North-Holland, Arru;terdam 1976).
Besides the work of C. ltzykson and J.B. Zuber, quoted in the introduction, another
general reference is here
J.D. Bjorken and S.D. Drell, Relativistic Quantum Mech.anics, (McGraw-Hill, New-
York 1964).
The non-reiativist.ic limít ís considered in
L.L. Foldy and S.A. Wouthuysen, Phys. Rev. 78 (1950) 29.
See also
LS. Brown, Quantum Field Theory, (Cambridge University Press, Cambridge 1992).
For a díscussion of the PCT theorem and the spin-statistics connectíon, starting from
first principies see
R.F. Strea.ter and A.S. Wíghtman, PCT, Spin 1:1 Statishcs and Ali That (Benjamin,
New York 1964).
A7.1 Real-time Quantum Field Thf'.IJTY a.nd S-Matri:s:
161

APPENOIX 7

A 7.1 Cluster Properties of Correlalít>n Functions

Let us briefty describe the cluster properties of conner::ted Feynman diagrams in a ma.s-
sive field theory. We restrict ourselves to a theory with <me mru;sive !lCIJ.)ar lie!d, but
the generalization i.s straightforward. Sinee calculations a.re involved, we retum in this
nppendíx to the euclidean formalism.

A 1.1.1 Dei:lly of Connected CorTelatíon F\mction.s

The propagator CIW then be written (equatíon (5.18)):

ó (x,y) --
1 f e'~>·<• -~l
d"p---· (A7.1}
(2··l r+m
2

To evaluate the behaviour of ó for lx yllarge, we rewrite the integral:

ó (x y) = - 1 -
' (211")4
fd p lor+oo dte'•·lr-~) e-•(p'+m').
4
(A7.2)

The gaussian integration over the momentum p can then be performed:

1rdf2
ó (x, y) = - -0
(2'1r)
f dt
td/> exp
[
-tm 2 ]
-(x-y)
4t
2] . (A7.3)

The laxge sepa.ration behaviour of .0. is given by the method of steepest descent. The
sa.ddle point is:
t=lx-YI. (A7.4)
2m
The gaussi!Ul integral OVtt fluctuations around the sa.ddle point finally yields:
(d-l)/2
ó(x y} ~ 2_ __m__ e-mlz-•1, (A7.5)
2m (
211 l:r- Yl )
Using this asymptotic estimate, it is not dlfficult to verify the following result; if in a
conne<:ted diagram we separate two sets of points by a distante I, then at Jarge I the
diagrarn detreases as exp( -nml). In this expression n is the sroallest number of !ines
ít is necessary to cut in order to disconnect the díagram, the two sets of points being
attached to different connected components (see figure 7.1).
!62 &o.l-time Qu.antum Field Thwry and S-Mo.tm A7.Z

In a massless theory ínstead {m ~ O) the de<:ay ís algebrak when the propagator exíst•
(thís irnplíes, in perturbt>tion theory, d > 2).

A 7.1.2 Threshald effects

The precise large distance behavíour of diagratnE is related to tbe strength of the lea.ding
singularity in momentum space. For example if, in momenturn space, a contribution to
the 2-poínt function has an algebraic singularity

a generalization of the previous ca.tcula.tíon yíelds a large dista.nce behaviour

]f"" now considera lPI diagrarn with n internallines, it yields for,. large a contúbution

This in turn corresponds to a singularity

o ~[(n-l)d-n-1].

In particular for d > 1 the singularity softens when n íncrcases. The 2-particle thresb-
old yields the strongest síngularity (p2 + 4m2 }(d-S)/2 • The nature of the singularity is
ímportant for the large time behavíour of correlation functions in real time: The leading
large time behaviour ís then relate<! to the leading singularíty in the energy variahle (a
property of the Fooríer transformation). Therefore if we consider the 2-point function,
its Iarge time bebaviour iB given by the one-particle pole, then the next to leading term
is related to tbe 2-particie threshold ....

A7.2 Connection between Spln and Statistícs

We have notíced in Section 7.5.1 that spin 1/2 fields rouid only quantized as fermions.
lt can be proven more generally that as a consequence of locality, urútarity of quantum
evolution and relativistic invariance bosons mnst have integer spin whlle fenníons rnnst
have half integer spin. Let us julrt illustrate this point here.
We bave shown in Section A2.2 that, as a coiJBe:juenee of the hermiticity of the hamil-
tonian the 2-point functíon ha5 a spectral representation in terms of a positive rneasure.
We have translated this result in the rela.tivistic kinematics in Sectlon 7.4. Ali possible
intermediate states contribute with the same sign. This result can easíly be genera.fu.ed
to the discontinuity in the phyoical domain of the diagonal scatteríng amplitudes. Let us
shaw tbat the sign of fermion loops irnplies a relation between spín and lrtatistics. The
contrihution to a scalar boson inverse propagator of a loop of scalar bosons of mass m
ha.s the form (see equation (6.49) and followings}

(A7.6)
A7.2 &.nl-tim< Qua.,tum Field Thm'iJ and S-Matri:J: 163

wbere the specific fa.ctor in front of tbe integral (but not the sigo) corre5j)onds to tbe
action
S{qí) = j ddx [í (8.<1>) 2
+ !m2 tll 2 + ~g,p3].
A conveníent integral representatíon of the diagram is given at the end of Section 11.5
(equation (11.75)). If""' take tbe discontlnuity of the function for s -r>' >4m2 ~
find

Thís discontinuíty bas a pasitive imaginary part, "•hlch is indeed const.f.ent v.•ith the
•pectral representation of WC2 l(p).
If we instead consider the contribution coming from scalax fermions, the tr In genera.ting
one-loop diagramB gets a rninus sign, and clearly the contribution has tbe opposite sign.
Let us instead consider the contribution of spin 1/2 fermions:

We now use the identity

The two terrns inside the hrackets cancel a denomina.tor and thus y\eld a constant (in
general divergent, see Chapter B) result, which has no discontinuíty:

(A7.7)

In the region of pbysical emission:

the prefactor (p" +4m2 ), which rellects tbe spin structute, ís negative and compensares
the negative slgn due to the Iermion loop.
8 DIVERGENCES IN PERTURBATION THEORY, POWER
CDUNTING

A local field theory is characterized by the absence of a "sma.ll" fundamental length: the
action depends only on prod\lct.s of fields and their derlvatíves at the =e point. In
perturbatlon theory the propagator has a simple power law behaviour at short distances
and the interactíon vertices are constants or differentíal operators act.ing on 8-functíons.
We have explained in previous chapters how physical quantities can be calculated
as power serie:s in the various interactíons. As we now show, perturbative calculations
are affected by dívergences due to severe short distance singularíties. Aiter Fourier
transformation, these divergences take the form of integrais diverging at large momenta:
one speaks also of UV singularities. These divergences are peculiar to Quantum Field
Tbeory: in contrast witb Cla.ssical Mecha.n.ics and Quantum Mechanics, it is imp05Sible
to define in a straightforward way a quantum field theory of point-llke objects.
ln Chapters 8-ll we shall expl.a.in how to deal w:ith tlús problem, at leust in the sense
of formal perturbation tbeory. The study of criticai phenomena by renormalization group
metbods will tben give us a clue about tbe possible origin of these divergences. In this
chapter - begín tbe discussion by analyzing the nature of divetgences. We explain
the problem first on the example of the .p3-field theory at one-loop order. We then
systematlca!ly chs.rs.cteri:oe divergem:es in a large class of local field t.hoories. We !inally
discuss the divergences of correlation functions involving also composíte operators.
To analyoe the divergences of Feynman diagrarns it is convenient to work in tbe mo-
mentum representation because it is simpler to dÍBCU>113 divergences of integrais at large
momenta rather than singuls.rities of distributions in space representation. Moreover the
explicit expressions of Section 6.3 show that amputation and Legendre transfonnation
become in momentum representation purely algebraic operatíons in the sense that they
involve no momentum íntegration because the propagator is diagonal in this represeo-
tation. In particular the divergences of connected correlation function.s can be simply
deduced from tbe divergences of proper vertices. Since a short analysis reveals that the
structure of divergenr:eo: of proper vertices is simpler,- examine in what follows tbe di-
vergences of Feynman diagrarns contribtiting to the generatíng furtetional r(<p), i.e. IPI
Feynman diagrarns.

8.1 The Problem of Dlvergem:es and Renormalizatiom The ,p3 Field Theory
at One-Loop Order

We first explain the problem of divergences ofFeynman díagramJS on the símple ex&mple
of a scalar .p3 field theory with a.n action:

S(~} j d".1: [~ (EJ,..P/ + ~m2 4>2 + ~g .p'] . (8.1)

This theory is unphysical because tbe potential is not bounded trom below. However
it has a well-delined perturbative expansion wbere this non-perturbatíve pathology does
not show up (see Chapter 39}. Moreover, when g is purely ímaginary, ;t makes sense
beyond perturbation theory and describes the universal properties of the Yang-Lee edge
singularity of the Isíng model.
8.1 Díveryences in Perturbation Theor!l 165

8.1. 1 Perturbatíon theory "t one-loop order

fue order. At the tree levei the IPI functional is identícal to S(if>) and thus:

rt•l (x, !1) ( -!>. + m~) 6 (x -!I).

The Fourier transfonn of the l PI 2-point function is then:

(8.2)

Also:
f( 3) (pl' P2, -pl P2) = g
{ (8.3)
r(n) (p~. .. ,Pn)"' o for n > 3.

One-loop ord.,., U•ing equation (6.49), we can now calculate the one-loop contribution
rd'Pl=
(8.4)

If we expand thls expression in powe,.. of 'P we obtain the one-loop oontributions to r!n).
After Fourier trrmsformatíon we find

-(n) (n- 1)!

rlloop"'---2-(-g)
, f ddq 1 l
(211)dq2+m2(q+p,};+m2 ...

x--------~~~--~--- (8.5)

quantíty represente:! by the Feynman diagram of figure 8.1.

Fig. B.l One-loop lPl diagrmn.

On this cxpression we see that the one-loop contríbution to the n-point function, which
for large momentum q behaves líke f<» q- 2•d•q, thus diverges for

d 2:: 2n.

Thereiore it is only for d = 1 (quantum mechanics) that ail correlation functions are
6nite. For d 2 the 1-point functíon, which hll!l no momentum dependence, diverges:

(8.6)
166 Di11ergencu m Perturbation Theory 8.1

When d íncrease:s more correlation functions diverge. For d = 5 the 1, 2 and 3-point
functions diverge:

(8.7)

,8.8)

lf we imagine that we cut the momentum integral by integrating in a sphere !a' <' '· then
the l and 2-point functians diverge like powers of A, while f(3) diverges log; nically.
To explicitly calculate the contríbutions which diverge With A, we can perfc Taylor
series expansion of the integrand ín the externai momenta. lt is easy to verde hat the
coefficients of the tenns of global degree k in the momenta are given by integ: .s which
diverge only for d ~ k + 2T>. Therefore the divergent part of a one-locp contr: ;,tíon to
the n-point function ls a polynomial of degree d 2n. For d ~ 6 we línd:

(8.9)

We recai! that the surface of the sphere S5 iE ... 3 .

We see that the dimension 6 is special in the following sense: the 1PI ' •rrelation
functíons which diverge are a.ll those wbich are already non-trívial at the tr< ' levei (a
term linear ín rjJ can be added to action (8.1) by translatlng q, by a. constant). \lloroover
the divergent terms bave the same momentum dependence as the tree levei com ibutions.
On the contrary, for d c 8 tbe 4-point function, whicb vanishes a.t the trb levei, is
also divergent.

8.1.2 Empirical removal af díveryence.s at the one-loop levei

For d ~ 6 diruensions, the divergent parts of the one-loop correlation functions h. ve the
structure of the original action. For example in d = 6 dimensicms tbe diverger: ..,a.rt
f"li•. ('I') oi the one-loop lPl functlonal is:

(8.10)

The functions Gi (A) are derived from equa.tions (8.9) a.nd are therefore defined only up to
additive finite pa.rts. For !ater purpooe (the minírnal subtraction scheme) ít is convenient
to give a ca.nonical definition of the divergent part of a Feynman diagra.m 813 the surn of
8.1 Dit~ergences in Perlurbation Theo'1! 167

the divergent ter~ru; in the asymptotic expa.nBion in a dltnensionless parameter. We here

choose A/m. Then:

ai(A) 1 (11 m A 4 A)4 2 2

27 ,.~ T- - 2 - +m ln;;:; ,
1 A
a>(A) 327 ~ lnm' (8.11)
(A)= - -3 (- A +2m2 in~)
2
1
a3
2'" 2 m '
1 A
a. (A)= 26,.aln;n.

If we now add to the original action S(~). the quantíty 6St(,P):

tiS, (,P) -I d6 z [g b (A) ,p (z) + ~g2 b, (A) (&.,.p)

1
2

+~lba (A) rP2 (z) + ~g 3 b4 (A) tjl 3 (z)) , (8.12)

in whlch tbe coeffi.cients b; are such that tbe differences b; - a; have finite large cut.off
limit5 but are otherwise arbitrary, the new action

S1 (,P) S(t/1) + 6S,(,P},

dilfers from the original actlon by its parametrization, hut involves the sa.me monomials
of the lield and therefore depends on the sa.me number of parameters.

S!(.P) I d6 z [~ (1- g'e.:,(A)) (8.,1JJ)2 + 4{m2 g

2
ba (A)) ,p 2

+~ (9- g 3b4 (A)) ,P3 - gb1 (A) 1/J] .

Tbe new IPI functional r (I"} at one-loop order ís then:

r (rp) s, (I")+ ri (op) +o (two loops)

S (I")+ r 1 ('P) + éS1 ('P) +O (two loops),
where r 1 {rp) ls the som oí one-loop diagrams calculated only with S(tjJ). Therefore r ('P)
now has a limit at one-loop order wben the cut-off becomes infmite.
Thls ís tbe method we would like to generalbe to arbitrary orders in the ioopwise
expansion and for more general lield tbeories.
Note that for d ;::: 8 this method fails. lndeed to gEmera.te a.t tbe tree levei a term
proportional to the divergence of tbe 4-point function for example, we ha.ve to a.dd a
4
,P term to the a.ction. We show below that such an intera.ction induces in turn worse
divergences whlcb cannot be reproduced. Power counting will show tbat it is not poso;ihle
10 add terrns to the action in •uch a way that the form of the tree levej terms a.nd the
one-loop divergences become similar.
lnterpntatian of perturbative divergenCI!.'!. To circumvent the dlfficulties associated
with the dívergences of perturbation theory, we shall try to implement the following idea
168 Divergences in Perturbation Thwry 8.2

whlcb is at the basis of renormali•ation theory: we introduce a Ja.rge momentum cut-olf

in the thoory, or equivalently modify the iield tbeory at short dlstance, as we have dane
here to characterlze the divergenres of Feynman diagrams, (the physical reality of sucb
a cut-olf is at this stage irreleva.nt}. We then investigate the po95ibility of cboosing the
original parameters of the theory as functions of the cut-off in sueh a way that correlatíon
functions have a finite la.rge cut-off limit. When sucb a limit exists, we show that it is
independent of the cut-off procedure (under some general conditions). We shall call
tbe corresponding local field theories renormalizable (ar super-renormalizable if some
para.meters in the interaction are cut-off independent). Renormllllzable theories are
short distance insensítive in the sense th!lt even if a la.rge ma.ss or a microscopic scale
in space provide a true physical cut-off, their long dista.nce or low momentum properties
cii.Il be described, without explícit knowledge of the short distam:e structure, in terms of B
small number of ejjective scale- dependent pammeter8. We already emphasize these ideas
here because they motivate all the technical analysis w hicb follows and are at the basis
of one major application of renormalization theory: the renormalization group analysis
of criticai phenomena (see Chapters 23-36).

8.2 Divergenc:es in Perturbation Theory: General Analysis

Dimension of fields. We first introdure the notion of canonical, or engineering dimension

of fields. The dimension [1>] of a field </J is related to the la;rge momentum behaviour of
the r/1 propagator. We explicitly MSume that the propagator {l, (P) is O (d) covariant in
d dimensions, or at ]eMt has a uniform large momentum behaviaur of the form:

(8.13)

in which C 1 is a strictly positive constBllt. Other cases requíre a special analysis.

We then define the dimension [9] of Ql(:r) by:

[.Pl=Hd u). (8.14)

Exa.mp/es. In the scala.r theory taken M an example in Section 8.!:

1
ó(>..p) ~ ~, for>.---. co ~ [,P] = ~(d 2}. (8.15)

For the fennions considered in Section 5.5 the ptop~ator in Fourier space reads:

and thus
(8.Hi)

The definition (8.14) is sucb that, in tbe &imple case of scalar and spin 1/2 fields, it
coinddes with the natmal mMS dimension of the field as deduced from the quadrat.ie
part of the action by dimensional analysis. Assigning a dimension +1 to momenta and

[p] lm] 1,
8.2 Dít~ergenc.es in Penurbaticm Theory 169

correspondingly a dimension -1 to length and space coordinates:

[x] = -1 =} [:,] "'+1,

a.nd expressing that the action is dimensionless, we indeed find:

[j ddx(ô~IJ)(:r)}2] =0 '* -d+2+2[rt>J~O,

(JddxJ,(:z;)iiJ;!J(x)J=o 9 -d+1+2{~)=0.
This is no longer true in general for higher spin fields. For exarnple the free action for a
massive vector field A,. is:

(8.17)

The propaga.tor LI.•• (p) is then:

(8.18)

and thereiore
d
(8.19)
2'
while just by inspection of the a.ctíon one wou1d have conduded tbat A, has the dimension
of a scalar fie!d. The reasons can be found in the appearance of negative powers of m 2
in the propagator: the quadra.tic forrn in the action is not invertible for m O. On the
pole of the propo,gator r? = -m 2 the numerator of ll. •• is a projector orthogonal to p,.
The same phenomenon will also oecur with hígher spin fields. A spin s massive field
propage.tor has the form:
(8.20)

in which ~. (p) 1s a polynomial of degree 2s in p, which is a projoctor on ''mass-she!J",

=
i.e. for r? -ml. Tbe dirnension of the corresponding iield rP• is:

(8.21)

Equation (8.21) generalizes equations (8.15,8.16,8.19).

DimensiVTI of verti.ces. The interaction terrn in tbe action is a linear combination of
vertices which are space integrais of a product of fields and their derivatives at tbe sa.me
points. Let us write a vertex V(if>} symbolically:
170 Divergenct!s in Perturootion Theary 8.2

where the k derivative!l act in an unspedned way on the fields </>;. We cal! these elementary
interaction terms vertíces because they are reprerented by vertices in Feynman diagra.rns.
We now define the dimension 6[V] of the vertex V(<!>) by:
.
h[V) = -d+ k+ Ln.;[</>;J, (8.Z2)
i=l

in which k is the tot.al number of differential operatcrs ll.Bd n; the number of fields r{>;
appearing in the vertex V.
Superficial degree of diveryence of a Feynman diagrom. As explained before, we con-
síder only IPI día,grams. Let us first write the vertex V(of>) in terms of the Fourier
transforms :f;. {p} of the fields q., (z) (still in syrnbolic notation):

V(Çl)- f ddpl · · .ddPn,+n•+···+n,lid (p, + P'> + ··· + Pn1 +···+n,)

X Pk~l (pt) · · • ~,(p,.,+···+n,) •

Each vertex of thís type brings a ó-function o[ momentum conserva.tion. In addition thís
vertex adds the product of k momenta in the numerator of a Feynman dia,gram. Jf we now
count the number of independent moment& ~r which to integrate in a diagram, taking
into account rnomentum conservation at vertices, we just find the number of loops L.
This follows directly from one of the definitions of the number of loops in a dia,gram gíven
in Section 6.4. Finally a díagrarn containl; ]; internal lineo correspondíng to propagators
Ll.; joining t.he different vertic<S.
Therefore if ali integration momenta in a dia,gram -r are scaled by a factor À , for ),
large the dlagram is scaled by a factor ), 6(-,) with:

(8.23)

in which v., is the nurnber of vertices of type a with k., derivatives. The number 6{-r)
ís called the superficial degree of divergence of the dia,grarn 'Y· For a. one-loop diagram
regularize<! with a. momentum cut-off, it charocteriz"" the divergence of the diagram as
a power of the cut-off.
More generally if ó("!) is positive a regularize<! diagram diverges at least like A~hl.
If 6(1) -= O it diverges at least like a power of !nA. !f ó("/) is negative the diagram is
superlicia!ly convergent, which meBJlS tho.t divergences can come only from suhdia.grams.
Let us verify equation (8.23) in the case of the q,3 lield theory in d = 5 dimell5ions.
The three divergent one-loop diagrams are shown in figure 8.2.

-o -o-Fig. 8.2
3
Divergent one-loop diagroms in " t/> field theory.
8.2 Divef'!}ences in Perturbqtion Theory 171 (
(
Since o 2 and k O, expnession (8.23) yields

6('Y) = 6 2/.
c
c
For I= 1, 2,3 respectively we find the values 4, 2, O in agreement with equauons (8.9). (
For I > 3 the diagrams are tonvergent.
Figure 8.3 exhihits a superlicially convergent díagram with dívergent subdíagrams in c
the same fieid theory: (
(

c
(
c
(
{
Fig. 8.3 Dívergont .ubdiagram. (
(
The superfidal degree of divergente is -2, the diagram is superficially convergent, but
(
the subdiagram ínsíde the dotted box is divergent.
A different e:g>ressíon of Lhe superficial degree of diveryen"". Various topological re!a-
c
tions on graphs allow to wríte <'>('Y) in different forms. (
Consider a díagram 'Y contributing to a lPI correlation function with E, (for externa!
line) fields <I><· Then if we cal! nj the number of fie!ds ,P, at a vertex o belongíng t.o the (
diagram, we have the relation:
c
(
{8.24)
(
(
The interpretation of this fottnula is simplé: every iuternal line connects lwo vertices
while an externa! line is only attached to ooe vertex. (
Figure 8.4 gives an example. (
(
(

E,=4, I.=4 (
E-..=1,1,=7 (
'V=3t nt=2, n2=l
{
v= 2, n1:;:;-0, •t>= 3
v=3, n 1 = 2, n,=2 (
(
Flg. 8.4 (
(
lf we combine equation (8.24) wíth the relatían (6.57) writteo unde.- the form:
(

(8.25)
172 Diveryenres in PerturOOtíon Theo'f"IJ 8.3

we can elimina.te L and I; in 6('y). We then obtain

6(-y) =d- LE.{,P;J + L"a6(V,) (8.26)

i

in which 6 (Va) is the dimension of the vertex a (equation (8.22))

and [ql,] the dimension of ,P; (equation (8.14)):

Equa.tion (8.26) directly leads to a classification of renormalizable theories.

8.3 C!assification o f Renormaliznble Fle!d Theories

We now assume that the program out!ined in Section 8.1 can be realize<!. This point
of view will be justified in Chapters 9 and 10. With thls assumptíon we ca.n classify
renormalizable field theories. In what follows we restrict ourselves to the most frequent
situation: l<kl :? O for ali fields. Otber cases require a special a.nalysis.
Non·renormalizable theories. !f at least one vertex V has a positive dimension ó (V),
then by considering dia,grams wíth an increasing number v of vertíces of this type we
can make the degree of divergen~ arbitrarily large and this for any IPI correlation
function. A field theory with such a vertex is not renormalhable because in order to
cancel divergences one would have to add an inlinite number of new ínteroctíons to the
action, and the final theory would depend on an iufinite number of pararneters.
Super·renormalizab!e theories. When only a finíte number of Feynma.n diagrams are
superfidally dívergent we call the field theory super-renorma.lizab}e. This happens when
the dimensions of ali verticrs are strictly negative.
Eo:ample. In the ,p• field theory in d = 3 dirnensions

Tbe superfidally dívergent díagrams rue lísted in figure 8.5.

Q Fíg. 8.5 Superfidally dirergent diagrams in

e
1/4.
8.3 173

R•normali•able theories. These theories are characterized by the property that at

least one verte:< ha.s dímension zero, and no vertex ha.s a. p05ítíve dímension. Then an
infinite number of diagtams havt: a positiw superficial degree of divergence, bowever the
me.rimal degree of divergence at E. lixed does not íncrease with the number of loops,
a.nd is independent of the number of insertions of the vertíces of dimension zero.
In addition íf ali dimensions of lie!ds (.p;J are strictly positive, only a finlte number of
correlation functions are superfidally divergent.
Such theories are calied renormalizable by power counting.
If at least one field has dimensioo zero, the situatíon ís more complicated: the degree
of divergence is bounded, however an ínlinit.e number of wrrela.tion functions are superfi-
cia.lly divergent. Generieally this lea.ds to field theorles dependíng on an infiníte number
of parameters although, in contrast to the c""" of non-renorma.lizable theories, only a
subclass of ali pooslble interutions is generated by renormalization.
In addition in some cases symmetries can relate ali these parameters so that only a
finite nnmber are really independent. We sba.ll meet such a situation in Chapters 14, 15
when we discuss models delined on homogeneous spaces like the non-línear u-model
Let us examine conventional field theori<ls oontaining fields of spin s with diwensions
(<t>,J given by equation (8.21 }:

The condition 141,) > O is se.tisfied for d ::>: 2 except for B = O, d = 2, c""" which rnust be
examined sepa.re.tely.
Ali vertioeo Bhould satisfy:

(8.27)

This condition bounds Ir, l.he number of derivatives, n. the number of fields of spin s at
the vertex, s the spin a.nd the dimension d.
For Ir • = O, the oondition (8.27} implies for ali vertices:
2d
n$d-2·
The corresponding renorma.lizable field tbeories are:
1>3 in d = 6 diwensions,
.p4 in d = 4 dimensions,
q,~. "'~ in d - 3 dimensions.
Note that any polynomial in ,P iB super-renormalizable in 2 dimensions.
=
If we allow in addition 2 derivatives, k 2, the only solutíon is d = 2, but then I.P) = O.
Let us now consider vertices wlth one spín 1/2 fermion pair #, and n scalar fields.
Condition (8.27) then reads:

Renormalizable field theories are:

..p1ijq, in d = 4 dhnensions ,
!ÍJ>M} in d 3 dimelll3Íons.
174 Divergences in Perturbation Theory 8.4

In additíon P(tp)iJI/!, in which P(</>) is a polynomial in c/>, ls super-renormalizable in 2

dímensioDS.
Finally tbe vertex ( ~~l1s renormalizable in 2 dimensioll5,
The vertices P(q,) (,&1/!)l and P(,P}~IN are also of dimension zero in 2 dimensions but
again the dimension of q, vanishes.
For spin 1 vecwr fields, general O( d) invariance leaves only dimension 2 M a possibility.
The only candidate with only fermions ls the vertex:

wbich ÍE renormallzable in Z dimensioll5. In addition the vertices ,Pí'Jp,PA,. liJld q? A!,

wbich appear in gauge theories, are dimensionless. However, they aga.in lead to a non-
trivial renormalization problem because scalar ficlàs are dimensionless ([1>] =O).
No hígher spin ficld leads tQ renormalizable theoríes.
Note however that spin 1 wctor fields associated with ga.uge symmetries do not enter
in the previo\1.\i classilica.tion because their propagator has in some gauges the hehaviour
of a scalar field propagator (for a dlscussion of geuge theories see Chepters 18-21).
Finally let us note that no physically acceptable, from the point of view of Particle
Physica, and renormalizable theory exlsts above dimension 4. lt ls not known whether
this property is logically connected with the foct that sp""'e time has just four dimensjons,
or ls a mere coincídence.

8.4 Power Counting for Operator lnsertions

Up to now we bave analy~ed only the divergences of the field cortelation functioos. How-
ever we shall also need in verious pl""'es correlation functions involving local polynomisls
of the field, called hexeafter composite fields or for h:ístorical reasons composite operotom
( this terminology comes frorn the operator formulatJon of quantum field theory ).
Typical ex8lllples are:

O (4> (:t)) ""q,2 (x), q,• (x), [ô~lj> (x)l2 - .•

One inoertion of operawr CJ(,P) yields the correlation functions:

Sucb correlation functions can in principie heohta.ined from the field correlation functions ;1
by lettíng various points coincide. However this procedure rorresponds in momentum ~
space to additional integrations, and therefore genera.tes new divergences. 'li
We have to ana.lyze the IPI correlation fuoctions witb operator iDSertíon, frorn the ,~
point of view of power counting. However we can use the results of previous sectíon in ·~
the following way. Let us again cal! S(,P) the euclídean action and let us add a source -~
(which ls a space-dependent coup)ing constant) for the operotor O (<f> (x)): j
s.(<f>) = S(q,) +f d 4
xg (x) o(!/> (x)) (8.28) !
To this new action s. corresponds a generatíng functional Z (J,g): j
Z(J,g)= J!di>!exp[-s.(,P)+ j J(x),P(x)ddx}. (8.29) ~
8.4 ]75

The correlation functions with one operator O (1>) insertion can be obt.ained from the
generating functional 6Zf6g (.r) taken at g = 0:

More generally successive differentiations with respect to g (.r) yield generating function-
als of correlation functions witb several operotor insertíons.
Performing tbe l.egendre transformation of W (J,g) = ln Z (J,g) with respect to J (x)
we obtain tbe IPI functionru r ('t',g). The genera.ting functionru of IPI correlation func-
tions witb one O (if>(Y)) insertion, rt'l, is then:

-ó'--)-
g {Y
I -
ó'r('t',g) - ~
L, 2_I
g~o n~l n
f dxl ... dxn't'(.r,) ... 'P(x,.) r {n) .
o (Y,X!,"'" ,.r,.).

The corresponding Feynman diagrarns have tbe •tructuxe shown on figure 8.6.

Fig. 8.6

After Fburier transformation they are just ordinary diagrams "Wlth one additiouru ver-
tex 0(1>), except that a.n additiona.l momentum enters the díagmm at the vertex sa tbat
· momentum is not conserved. However if we scrue rullntegration momenta by a factor À,
In tbe larg<o À limit ali externai mamenta are negligible. Therefore the pow..r counting
of 1Pl corre)ation functions with one operator insertion ís the sa.me as "Wlth one vertex
· lnsertion. We assign to an operator O (li>) the dimension [0]:

(8.31)

which k is tbe number of derivatiV<>S in the operator, and nt the number of fields of
Lype i. This detinition díffers from the definítiou of the dirnension of the corresponding
vertex (equatiov (8.22)) by d.
)f we tben ;ru;ert the operators o, (.:r1) ••• Or (x. ), the superficial degree of diver~
of IPI d iagram "' ís:

I 6 [O, ... o.]= 6~ + [O,f+ ... +[O.]

7 -rd J (8.32)

which 6, is tfle degree of divergence of the diagnm wlth only the "true" vertices.
176 Divergences in Pe.rturbaiíon TIU!ory 8.4

For example for d 4 one insertion of ~m (.r) in the r/> lield theory yields
4

(8.33)

In the same field theory the IPI n-point function with I <1>2 insertions:

has a degree of divergence 6:

é=4-n-2l. (8.34)

The new divergent correlation functíons are shown in figure 8.7 .

. .o ·····O···· ······Q
Fig. 8.7

Bibliogrnphical Notes
For an early reference on power counting see
N.N. Bogoliubov and D.V. Shirkov, Jntrrxfuction to lhe Theory af Quantized Fields
(Interscieuce, N ew York 1959).
For the Yang-Lee edge singularity see
C.N, Yang and T.D. Lee, Phys. Rev. 87 (1952) 404; T.D. Lee and C.N. Yang, Phys.
Rev. 87 (1952) 410; M.E. Fisher, Phy3. Rev. Lett. 40 (1978) 1610.
'9 REGULARIZATION METHODS

The divergences that we haw rnet a.nd annlyred in Chapter 8 show that the tield theories
we wa.nted to construct <=not be defined by a straigbtforwa.rd perturbative expansion.
Our strategy will thus be tbe following: we sball modify the field tbeory at Ja.rge mo-
mentum, shott dmance or otberwise in sucb a way tbat the new Feynman dlagrams
become well-defined finite qua.ntities, and sucb that when one contrai pararneter reacbes
some limit (for example the cut.off is sent to infinity), we formally recover the original
perturbation theory. This procedure is called regularization. lt will allow us to isolate
Well-defined divergent parta of diagrams a.nd denl with them with renormalization as
will be explained in Chapter 10. Many regularizstion rnethods have been introduced
ln tbe literature, but we shnll only describe tbree of thern: Pauli- Vülars, lattice a.nd
dimensional regularizations, which bave different advantages a.nd shortcornings. In nny
particular application two critería will guide our choice: (i) in some theories, symmetries
play a crucial role a.nd it is essential to find a regularization wbich preserves the sym~
metry; (li) if we wa.nt to explidtly calculate Feymna.n diagrarns, we shall look for the
regulruization which leads to the simplest practical calculations.

0.1 Cut-Off and Pauli-Villars's Regularization

The simplest idea is to modlfy the propagator in such a way that it decreases faster at
large momentum. For example we can replace

(9.1)

·: omd choose the degree n to make ali àiagrams convergent. The pararneter A is tbe cut-
olf. In the large cut-off limit the original propagator iB recovered. This is tbe spírit of
Pauli- Vlllars's regnla.rization.
Note that such a propagator ca.nnot be derived from a hermitia.n hamiltonian. lndeed
; tbe hetmitícity of the hamiltonian Jeads to the representation (A2.8) for the 2-point
function. lf the propagator is, as above, a rational fraction, it must be a sllm of poles
·, with positive residues and thus cannot decr"""" faster than 1/r.
More general modífications are possib!e. Schwinger's proper time method (Appendix
-A9) suggest.s:

(
1 )
p2+m' ,....
1""
v
dtp{tA2}e-•{P'+m'), (9.2)

i In which p (t) satisfies the condition

lim p(t) = 1, (9.3)
•-oo
.,lllld decreases fast enough when t goes to zero.
-~ · An exnmple is:
p(t) =li(! 1), (9.4)
178 Regulari>ation MethOO.. 9.1

e(t) being the step functinn, which yields

(9.5)

Howe\•er the regulari~ation has to satisfy one important condition: the regularized prop-
agator should remain a smooth function of the momentum p. lndeed singularities in the
momentum representatíon generate, after Fourier transformation, contributions to the
the la.rge diste.nce behaviour of the propagator, and we want to modify the theory only
at short distance.
As the example (9.5) shows, it is possible to find in this more general class, propagators
whích have no unpbysjcal singula.rities; however there is another price to pay, the thoory
is no Jonger stríctly local and there eJcists no corresponding hamiltonian formali!;m.
For fe;mions the same method ís applícabie. We can replace

(9.6)

Regulator fields. Let us note that Pauli-Villars's regularization has another, often
equivaleot, formulation based on the lntroduction of a set of reguiator fields.
To regularize the action S (</>):

(9.7)

we introduce a new set of fields <fi•. k 1, ... , n, over which we integrate in the functional
integral, and comider the modified action 8,.8 . (11>,4>•):

With action (9.8) any internai</> propagator iB replaced after regul~ation by tbe sum
of the </> propage.tor and &11 the </>k propagators -Ck/ (p 2 + Ml) • To improve the large
momentum behaviour of Feynman diagrarns we impose constre.ints to the parameters
Ck. Cancel!ation of the 1/r term in the expansion at large momentnm of the sum of
propagators requires immediately:

{9.9)

This implies that at lea.st one of the Ck i!; pooitive and therefare to this field </>• cannot
correspond a particle in the sense of Partide Physics: the actíon cannot be generated by
a hermitian hamiltonian, for tbe same reason as the regularize<! pr<>page.tor (9.1 ).
On expression (9.8) the large cut-off limit i.s a limit in whích ali masses Mk becoroe
infinite at fixed relative ratios. ·
 179

It is posslble to use a fuuctional argument to sbow that regularizations (9.1) and (9.8)
are equívalent. In the integral

f [d.P] Ir [d<Pt]
I
exp [-8""'. (t/l, </>k)], (9.10)

we lirst cbange variables:

(9.11)

The regularized a.ction then beromes:

01
S, (4>',4>•l fd"x[~(çl' ~4>t)(-ll+m2 )(<~>'-~çlk)
-t ~ 1 k
,P.(-ll + Mf} <fl>+ V(</>')] .

The integral over the fields ç\k is now gaussian and ca.n be expllcitly P<'rformed. An
ekmlentary calculation leads as expected to:

f [d,P)ex.p [-f ddx (!.pS! 2

lH V(.Pl) 1, (9.12)

(-ll +m 2 )-
1
L Ct(-ll + Ml}-
1
• (9.13)
k

ÍB possible to cboose the ooefficients ck in .such 8 way that:

[ -ll~m2 Lk
Ct
-ll + M2
J-t = (-ll
2 (-ll+Mi)
+m) IJ (-m' + M2)' (9.14.)
k • k

the ç\ propagator becomes analogous to propagator (9.1).

,;_"'' 11 ·~·· inspire<:! regu!arizations have one important advantage, they ca.n be
sucb a way that alargo number of symmetries of the inítial action are preserved.
in all models which already in quantum me<::ha.nícs have divergences due to
ordering problema (see Section 3.2) a class of Feynma.n diagra.n>s ca.nnot ba
by tlúB method. This indud..,, as "M> shall J<OO Inter, models wbich have
(see Chapter 14) or gauge symmetries (Cho.pter 19).
show, hawever, tho.t these methods work for simple scalar lield theories. We
that the regularized pl'opagator llA (p) behaves Iike:

1
ll,\ (p)- 2;·
p

(8.23) gives the degree of divergenoe of a regularize<:! diagr&m 7:

Ó (7) = dL- 2nl +L k0 (9.15)

. V 0 •
180 Regulo.rization Methods 9.2

U6ing topological relation (6.57):

we can rewrite equa.tion (9.15)

lt is thus neceBSacy to choose

2n > d a.nd 2n > sup k,.,

Q

sínce both L and v,. ca.n increa.se indefinitely.

For example tbe choice

regularizes ali diagrams.

Examples. (i) In the q,a field theory in 6 dimenslons, 2n 8 renders ali díag:rams finite.
(U) In the ~4 field theory in 4 dimensions, 2n ~ 6 suflices.

9.2 Lattice Regularization

We shall meet examples in which Pauli-Villars's regularization does not work: li.eld the<:>-
ries in which the action has a delinite goometrical character like models on bomogenoous
spaces {ror example the non-linear a-model) or gauge theories. In these tbeoríes the
forros of the propa.gator and of tbe interactian !.erms are not independent. When the
propa.gator is regularized, new more 6ingular interactions have to be added to the action
to preserve some symmetry and, as we sball show on examples, some one-loop diagrams
cannot be regularize<!. Other regularization methods are needed. In most ca.ses lat-
tice regularization, which we have introduced in Subsection 5.1.2 to define precisely the
functional wtegral, can be used. The advantages are the following:
{i) Lattie<l regularízation is the only regularization which has a meaning outside per-
turbation tbeory that we know. For instance the regularize<! functíonal integral can be
calculatoo by numerical methods, like stochastic methods (Monte Carlo calculations) or
strong coupling el<p!Ulsians.
(li) Jt preserves most global or local syrometries with the exception of the space O(d)
symmetry which is replaced by a hypercubic symmetry (but this is not too serious as will
be argued iD Chapters 24 and 25) a.nd fermion chirality, which turns out to be a
problem.
The main disadvanta,ge is that it leads to very complicated perturbatiw calculations.
Exampk of a scalar field theory. Let us exo.mine the effect of the regularizatíon on the · .
action:
 11.2 Regularúatíon M~fhod8 181

Let a be tbe lattíce spacíng. Tbe derivative Ôp</> can be replaced by the finíte dil'ference:

l
8.,1> ..... V',</>= ã[</>(x +an,.) -<f>(x)),

"" being the unit W!ctor in the iJ dlrection.

The action S(</>) bocomes

The arguments o f the Fourier transform J(p) of the field are now cyclic varinbles p.:

1>(pJ c:t L luoZd

<f>(x)e-'P·>< ... <f>(x) =f ddptfr(p)ép·x, {9.17)

with 1,. ,d .

. .The domain of varia.tion of momenta. is often called the first Bríllouin zone.
In Fourier components the quadratic part of the a.ction is:

.
ad:L! (Cv' .,q,J2 + m2.p•J = (2•/ f ddp!<fr(pJ sm (pJ if>< -pJ,

In wbicb the function sl~l (p) is:

sC>l (p) m2 +f. I(e"'"·~

p=l
-1) 12

correspondíng propa.gator 6~ (p) is thus:

(9.18)

momentum represent.a.tion the Feynrnan diagrams are now periodic functions of all
with period 21r{a. In the small Iattice spacing limit we recover the usual

t.~ (p)-1 m2+1J2 -;\ :La•p! + O{p:) (9.19)

..
symmetry implies O(d) •ymmetry at orde:r p 2 .
theories. Let us briefiy indícate a fuw pr0hlems arising when fermíons are
in the a.ction. \>,'e consider the free fermion action:
182 Regulnri>ation Meth~ 9.2

If we want to regularize this action by a lattice and preserve paríty (see Sectíons A5.5,
A5.6) we can replace a,,P(x) by

Then after Fourier transformaticm the action beçomes:

(9.20)

and the fennion propagatot i$:

(9.21)

The problem with this propagator is that the equations relevant to tbe small lattice
spadng !imit:
sín(ap.,) O

h ave each two solutions "• O and p, = 1f f a withln one períod, i.e. witbin the Brillouin
:rone. Therefore the propagator (9.21) propagates 2d fermions. To remove this degeneracy
it is possible to add to the regularize<! action a.n additional se alar term 6S involving second
derivatives:

óS (.ti, .P)"" ~L [2.ti(x)\ll(x)- .ti(x+an,)<Ji(x)- .ti(:r}.p(x +an,.J]. (9.22)

z,p

After Fourier transformation the new fermion propagator reads:

with:

Therefore we have broken the degeneracy between the different states. For each oom·
ponent Pp which takes the value 1f f a the mass is increased by M. By choosíng M of
or<kr l/a we elimlnate the spurious states in the oontinuum limit. This is the recipe of
Wilson's fermions.
However a serious problem arises if one wants to construct a theory with messless
fermions and chiral symmetry (equation (A5.33))

Then both th., ma.ss term and the term (9.22) are excluded.
9.3 183

Of course one cou!d think modifyíng the fermion propagator by adding terPlB conned-
ing fE'l'mions separa.ted by more than one lattice spa.cing. Tbe propagator would then
have the fonn:

in which JP (p) is a smooth function (sit~gularities introduce violation oflocality), periodic

with period 21r /a, which vaníshes linearly for IPI small

In one dimensíon it is easy to understand that this modification does oat solve the
problem: if f (p) is pe.riodic a.nd cootinuous, it has to Vl!.llish !ineMiy an even number of
times in each perlod. Thls a.rgument ca.n be gene.ralized to a.ny dimension. This doubling
of the number of fermioo degrees of freedom is related to the problem of a.nomalies (oee
Chapter 20).
Since the most naive form of the propags.tor yields 2d fcnnion states, one trles in
practical calculatíom to reduce this number to a smaller multiple of two. The idea of
staggered fermions introduced by Kogut and Susskiod is often used: first by modifyíng
the action one is able to decrease the multip!ication íactor from 2• to 'P/2 with respect
to form {9.20). Then the rema.ining degeneracy is interpreted as the reflection of an
internai eymmetry SU('PI 2 } of the act.ion. The discussiou is slightly involved and will
not be given here.

9.S Dimeru;iona.l Regularization

In contrast with lattice regularization, dimens!ona.l regulari~Míon seems to have no mean-

ing outs.ide perturbation thoory since it involveo; continus.tion of Feyrunan diHgra.rns in
the parameter d (d is the space dirnelJBion) to a.rbitrary complex values. However it is
the regularization which leads in geneJ'al to the Súnplest perturhative calculations.
It is not applicable when some essential pr0perty of the lield theory is specific to the
lnltial dimension. An example is provided by vlolations of parity symmet.r.Y involving
the oomplete a.ntisymmetric tensor E~ 1 ...,.., for example through the re!Btion between "(s
(called -ys in 4 dimensions) and the otha:"' ms.trices (see Subsection 9.3.3):

Before discussing the dimensional continuatiou of an arbitrary Feynman diHgram, let UB

explain the idea on a particular exs.rnp!e.
A simp!e e:r:ample. Let UB lirst ronsider the integral I

I= f d4 qf(q
2
). (9.23)

lnvaria.noe under rotations allows us to integrate over angular variables:

2..-dl• f d d-1
1 ( q') · (9.24}
l=r(d/2) o qq
 Regularization Method.. 9.3
184

lf f(:r:) is a.n analytic function Md íf there exists a domain in tbe d complex plane
for wbich the integral converges, tnen expression (9.24) gives the continuation to any
dimeusion of integral (9.23).
H~ if we consider o.n iotegral involving several rnomenta, we need a more system-
atic rnethod.
We shall therefore define an integral in d dimensions by a set of conditions which, íf d
is an integer, lead to the usual integral.

9.3.1 Defining properties of d dimenrional integrais

Let us define the dimensional continuation of integrais by the three conditions:

(i) I d4 pF (p+ q) = J ddpF(p) transla.tion

(ii) I d"pF (Àp) = p.j-d I ddpF (p) dílatation
(ili) I ddpdd' q f (p) g (q) = I ddp f (P) I dd' q g (q) factorization

Let us show tbat these simple rules pravide a dimensional continuation to Feynman
diagratDS.
From property (ill) we derive:

(9.25)

We now write any scalar propagator D. (p) as:

D. (P) =lof"" dtp(t)e-P' '. (9.26)

For example:
p (t} = .,-tm• "* D. (p) (p2 + m') -1
Using this representation and the property tha.t any momentum on a line is a linear
combination of loop momenta and ~rnal momenta, we can write a scalar Feynma.n
diagrarn r witb coiiStant vertices under the form:

I L [ L
I~(p)=(2n)-L.t ][ldt,p,(t•)}]ddqt exp -~qt·cu·Mtt'(t;)

-2~qt · kt(p,t;) -S(p,t,}]. (9.27)

Properties {i), (ii} and equatíon {9.25} allow us to integrate uver ali loop momenta q,
and we find:

I., (P) (411')-Ld/ 2 ! ÍI

r=1
dt; p; (t;) {detM)-d/ exp
2
[tk, (M- )u·
l
1
kt•- S(p,t,)] .
(9.28)
In this expressíon the dependence in d is completely explicit, and therefore continuation
in d (a.t g<meric momenta if the theory is massless) can be ochieved.
 9.3 Regu/arization M etfwd.. 185

Examp!e: One-)oop contribution to the 2-point function in the ma.ssless (m '"" O) <fi"
field theory

'Ird/2 [""' dt 1dt2 ( t1 to •)

"' (21r)d Jo {t 1 + t 2)df2 exp - t1 + t2 p ·

For p i O, the integral converge; fot 2 < d < 4.

We can complete the ca.lculation by setting

lt ts, t, = (1 t) s.

The integral becomes:

We integrate over s;

r., (::;d (p•)fdt2l-2r (z·- ÜJ.' dt[t(l- t)J!<~t2l-2'

and therefore finally:

I = "d/2 r (2- d/2) r~ ({d/2) -I) ( ')!d/2)-•- (9.29)

., (21t)d r{d- 2) p

Thia expression has a pole at d = 2 corresponding to IR {law momentum) singularities

because the theory is ma.ssless and poles at d = 4, 6, ... wbkh clearly are consequences of
the UV (large momentum) divergences of tbe Feynman diagram.
It is interesting to explain on thia example the interpla.y between dimenBiona.l contin-
uation and cut-off regularization. li we regularize the propaga.tor, for exa.mple by the
· method of Pauli-Villar~, the Feynman diagram I, beoomes a regular function of cl for
d :> 2 up to some even integer larger than 4.
In the neighbourhood of d = 4, it has tbe form:

lf, at cl lixed d < 4, we sent the cut-off to infinity we obtain the oontinuation of the initial
diagram with a pole at d 4. If at cut-off lixoo we take the limit d = 4, we get a finite
result in whích !nA has replaced tbe pole in (d- 4).
186 Regtdarizatian Meth.<HJ, 9.3

lmportant remark
The property (ü) implies:

(9.30)

for the:se integrais which exist for no value of d.

A lirst important consequence of this property is the following: if we consider models
in which the hamiltonian density conta.ins terms which a.re products nf the lield ,P(x) and
its conjugated momentum 1r(x) taken, dueto locality, at the same point o:, the questíon
of operator otdering arises. However equation ( 5.5) shows that the commutator of these
oper ators ís of the forro:

and therefore vanishes identically in dimensional regularhation for any dimensinn d >
1, i.e. except in quantum mechanics. The precise quanthation rule a.rui the arder of
operators are irrelevant in dimensional regularization. This fee.ture will be used in the
quantization of models délined on homogeneous manifolds whose haroiltonians have the
generic form (3.24) {,_ Chapters 14, 15).
Let us point out IUIOther consequence. The result:

!~ O,

can be interpreted as a cancelation between an UV and IR dive.rgence:

However this property of dimensional continuatíon has oue dangerous ccmsequence: in

a field theory involving ma.ssless fields, having for exaro pie a propagator l/p2 , IR diver-
gences appear iu 2 dimensíons. If this thoory is renormalízable ín 2 dimensions, it has
also UV divergences. In such a case UV and IR singularíties get rnixed. Therefore to be
able to identify poles roming from the la.rge momentum region, it is then necessary to
introduce an IR cut-off, for example by giving a mass to the field.
Finally since the property (9.30) of dimensional regularization is at lirst sight somewhat
strange, let us cbeck ire consístency (guaranteed a priori by the consistency of the defi.ning
rules) on an explicit example:
I

which can be calcutated in two ways: first one notes

r~l] = f r+l
ddp .
1

Thls yields:
1= _,.df'2J' (1 - d/2)
Second, one uses the transformation (9.26) on the initial expression. lt is eaBy to verlfy
that one obtains the same result.
9.3 187

9. 9. 2 Continuation of tensorial structures

Up to now we have considere<! diagrams corresponding to scalar fields. Any dia.gram
which ie not a scalar cBll be expaoded on a set of tensors with sca)a.r coefficients. For
example:
(9.31)

The scalar diagrams contributing to A (p2 ) and B (p2 ) can be obtained by taking the
traoe and the scalar product with Pp:

(9.32)

We btJve re<luce<l the problem to tbe calculation of integraU. of the form (9.27) witb
additíonal factors polynomial in momenta. Tbe integration over momenta. can then a.lso
be performed to yield the continuation in d.

9.$.9 Fermions
For fermions belonging to the spinorial representation of O(d) the strategy is the same.
The spin problem can be re<luce<l to the calcula.tion of traoes of 1' ma.trices. Therefore
only an additional prescrlption for the trace of the unít matrix is needed. There is no
natural continuation since odd and even dimenslons behave differently. Hawever we have
shown in Appendix A5.8 that no algebrnic ma.n.ipulntion depends on the explicit value of
the trace. Thus a.ny smooth continuation In the neighbourhood of the releva.nt dimension
will be satisfactory. A convenient choice which we sha.Jlnlways ndopt isto take the trace
consta.nt. In even dimension as longas only 1'~ me.trices are ínvolved no other problem
arises. However if the día.gtams involve "fs and if it beoomes necessa.ry to use the identity

then no dimensional continuation which preserves all properties of '")'s exists. As we h ave
seen before, )att!ce regularízation i!; equally impossible in this case. This difliculty is the
source of chírol anomalíes.
Since we have to calculate traces, one possibility is to define -rs by its expression in
tenns o{ -y,. matricea in the inltial dimension; for example if we start from 4 dirnensions
=
we define 1'5 'YS as the product:

4!1'5 = -(.......,. 1',., ••• -y,., •

It is then easy to verify that, with thls definition, 1'5 anticommutes wlth ali other 1',.
matríces only in 4 dimensions. Iffor example we stert from dimension n (n even} 3nd
evaluate the product 1'•1'S1'• in d dimensions, we find:

1'•1'S1'v (d- 2n)1's.

Anticommuting properties of the 1'S would have le<l to a fa.c:tor -d instead of d 2n in

the r.h.s ..
188 Regulari.zatian Methods 9.3

9.3.4 Dimensional reyulari.zatíon a.nd UV diveryence.o;

The principie of diroension regularization is the following; we define ali diagr3.1I15 by
analytic continuation in the dimension pa.rameter d as expla.ined before. When the di-
mension d approaches the i11itial dimension, Feynman día.gra.rru; become singular as a
consequence of the original UV dívergences. The singular contributíons can be isola.ted
by performing a Laurent expansion of the diagram. For example expression {9.29) is the
value of a Feynman dia.gram of the massless 4!' field theory whích is renormaliza.ble in
d 4. The Laurent expa.nsion ís:
1 1
1., ""Nd [ --+-
4 -d 2
As we bave ímplicitly done a.bove, we shall in general include a fa.ctor
21rdf2
Nd---..,.---
- (21r)dr(d/2)'

product of the surfa.ce of the sphere Sd-i by (21<)- 0, in the <lefinition of the loopwj,;e
expansion pa.rwneter, beca.use ít is generated naturally by each loop integration. As
we have already shown in an ex3mple, powers of !nA (A being the cnt-<>ff) which would
appear in a cut-offregula:rization in the large A limit, are replaced by powers of 1/ (d- 4).
However as a coJJBequence of identity (9.30), no divergent contribution equivalent to a
power of A çan appear.

Bibliographical Notes
For the Pauli-Villars's regularization see
W. Pauli and F. Villars, Rev. Mod. Phys. 21 (1949) 434.
For the proper time method see
J. Schwinger, Phys. Rev. 82 (1951) 664.
The consístency of the lattice regularization is proven (except for theories with chíral
fermions) in
T. R.eísz, Commun. Math. Phys. 117 (1988) 79, 639.
The doublíng phenomenon for lattice fermions has been proven quite generally by
H.B. Nielsen and M. Ninorniya, NucL Phys. B185 (1981) 20.
Wilson's solution to the fermion doubling problem is described in
K.G. Wilson in New Ph.enomena in Subnuclear Physics, Erice 1975,
A. Zlchichi ed. (Plenum, New York 1977).
St.aggered fermions have been proposed in
T. Ba.nks, L. Susskind and J. Kogut, Ph~. Rev. D13 (1976) 1043.
Dimensional regulariza.tion has been lntroduced by:
J. Ashmore, Lett. Nuauo Cimento 4 (1972) 289; G. 't Hooft and M. Veltman, Nucl.
Phys. B44 (1972) 189; C.G. Bo!lini and J.J. Giambiaggi, Phys. Lett. 40B (1972) 566,
Nuovo Oimento 12B (1972).
Its use in prohlems with chiral twomaliéS has been proposed in
D.A. Akyelllllpong and R. Delbourgo, Nuovo Cimento 17A (1973} 578.
For an early review see
G. Leibbra.ndt, Rw. Mod. Phy3. 47 (1975) 849.
For other schemes see also
E.R. Speer in Renorma!iJ.atitm Theory, Erice 1975, G. Velo and A.S. Wightma.n eds.
(D. R.eidel, Dordrecht, Holland 1976).
9.3 Regulari.zation Me t/wlb 189

Exercises
Exerciae 9.1
We oonsider tbe example of the O(N) invaria.nt action:

"'bere <f> is a N -component vector.

9.).1. Write the field equation of motíon (see Chapter 5) satisfied by the generating
functíonal of correlation functions.
9.1 .2. Deduce a relation between the two and four-point fundions.
9.1.3. As an approldmation, assume now that tbe two correlation functions correspond
to an unknown quadratic action (i.e. to a gaussian functional measure) and calculale the
two-point fundion.

F.xercise 9.2
Repeat the whole analr.;is (equation of motion and self-consistent fre<l ection approxi-
mation) for the so-.:alled GTDBS--Neveu model (GN) discussed in Appendix A30, which
is a model of N massless self-interacting Ditac fermions {V.', 1/1'}. The U(N) symmetric
actioo is:

The GN model ha.s in eYEn dimensions a discrete chiral symmetry:

1f; ,_. 'rst/>, ifr ...... -ihs,

which forbids a fermion m8S!l term.
190 ~tion Melhods A9.0

APPENDIX 9
SCHWINGER'S PROPER TIME REGULARIZATION

In Sectíon 9.1 we have alluded to Schwinger's proper time regularization. Let lll! show
here that it leads to a compact representetion of a regularize<! one-loop contrihution to
the g<merating functional of proper vertíces r('P)· In Section 6.4 we have shown that
r iloop( 'P) is given by:

(A9.1)

Tbe quantity ó2 Sfó<P (x1 ) Ó!p (:r 2 } is a local operator. For exo.mple if S (</>) is:

S (<I>) j ddx [ !(â,.Q)) 2

+ !m 2 1J)2 +g ~] , (A9.2}

it takes the fonn of a simple quantum mecha.nical hamiltonian H:

H;;;
Ó!p
IPS
(x 1 ) 6<p (x,)
[-Do+ m' + (N ll 2)1</>N-2 (:rt)] 6 (xt - xo). {A9.3)

Let us also define H o:

Ho = (;18 j (A9.4)
- Ó<pÓ<p ,=O
Using then the general ídentity:

tr(lnH lnHo) -fo""~tr(e-• 8 -e-•Ho), (A9.5)

we rewrite ruooo( <p) as:

rlloop(<P) =-21 1"' ttr

o dt (e-tH -e-lH•) . (A9.6)

Expression (A9.6) ia the representation of the cme--loop functional as an lntegtaii'JVer

Schwinger's proper time. Large momentum divergences appear as divergences at t = O.
If we multiply the lntegra.nd by a cuttlng factor p (tA2 ) we recover the regularization of
equatíon (9.2).
Expression (A9.6), once regularize<!, provides a compact representation ofthe one--loop
functional which o!ten leads to a rapíd evaluation of its divergent part. The general idea
is to write Schrõdinger's equation for tbe matrix eJements (:.: le-•H I
:r'} and use it to
deri"" their sma.ll t expa.n.sion. This is a problem we bave already encountered ln Section
2.2. 'lb determine the trace, we only need the coefficients of the expansion ln the Iimit
1
X=x .
Let us for illustration purpooo expand up to order t" in the cose of action (A9.2). We
first solve an auxiliary problem. We consider the Schrôdinger equation:

(A9.7}
A9.0 Regularization Mothod.s 191

We set:
(:z: je-'HI x') e-•(r,•';<). (A9.B)
The Schrõdinger equation then takes the form:

(A9.9)

We expand u for ! sma.ll:

1 2 d
o= 4t (x- :r') + 2ln411't + At + Bt 2 + Ct 3 +O (t'), (A9.10)

and obta.in equations for the coefficients A, B and C

A+(:r-:r'),.a.. A V(x},
2B+(:r-x'),.a,,B l!.A, (A9.11)
(
3C + (x -x') 10 8~C = l!.B- (8,.A) 2 •

The equation requires the calculation of A up to order (:r- x') 2 :

{
A=~(! (x- x'J) + ~ (x- :r'),. (x x'),ô,.ôvV + O(lx- x'l) 3 ,

B = GóV(:r)+O(jx-:r'IJ, (A9.12)
1 2
C= -12 (8,. V(x)) + óB +O (i:r x'l).

We do not need l!.B at this order becaUBe if disappea.rs in the trace. We calculate now
the regularired expression:

r .e-1 :toop (lfJ) = 1

--
2 •
!."" t
dt
~ tr (e-<H -e -tH0 ) • (A9.13)

To compare with other regularizations we can eventually set & = I/A2 •

Since we are here only interested in divet'gences, we only calculate contlibutions coming
the lower bound:
1
~
-2{4.-) J. !!,, /d":t{[m
E t1
2
- A (:r:,x)] t+!
2
W (:r, :r)- B(x,:t) -m4 ] t 2
~ (A3 (x,x}- 6AB + 6C- m6 ) t 8 } + O(t 4 ). (A9.14)

(A9.15)
192 Regularizatíon M ethods A9.0

0
When d is an """" integer, !: /0
has to be replaced by In (1/E). This expressíon gives ali
divergences for d $ 6.
For example let us apply thís result to the rj>3 field theory in 6 dimemíons:

f11!oop('l')"" 27~3 f d6x { ~4 g<p(x) ~· [g'<p' (x) + grn\o(x)]

+~In~ 2
[o 5 op3 (x) + 3lm2 <p (x) + 3gm\o(x) + ~ (ô,<p(x)) 2]}. (A9.16)

This express.íon should be compared with expressions (8.10,8.1l). Because the regula.r-
ízations are different, only the logarithmic divergent tenns coincide.
We can also oonsíder the <1>4 field theory in 4 dimensions:

rtt'"" = 32~ 2 { ~· g f 4
d x<p (x)
2
-In~ f 4
d x [ ~'1'4 (x) +gm2 op2{z)]}. (A9.17}

An ídentical expression will be recovered in Section 10.2, equatton (10.8). Both in equa-
tions (A9.16) and (A9.17), we have defined the divergent part of r (<p} as the surn of the
divergent terrns in the asymptotíc expansion for A/m Jarge. Finally we can also apply
equation (A9.15} to d = 2. If "í(4>(x)) is the interaction term, we find:

f't\:,.,P = 2:..1n
4.r
~f d2z Ví'('l'(x)).
m
(A9.18)
Althougb in those exa.rnples the results eM be easily recovered from the Feynm.an
graph expansion, in more complicated cases, in which synunetries play an essential role,
this metbod to evaluate divergenoes of one-loop dia.g:rams can be quite useful.
<-function regularization. A variant of the previous xegularizatíon method is to reploce
expressíon ( A9. 6) by:
1
r"'s (m) = - - -- - f"" dtt"- 1 tr(e-tH -e-<Ho) (A9.19)
1 lcop .. zr(l + u} lo '
and to take, after analytic continuation in G, the limít a O in the spirlt of the dimen-
sional regulru:!2ation.
Let us consider for example again the ljl4 field theory in four dlmensions and calculate
r 1 loop pe< unit volume for a constant fieJd op:

y
1 r••&
1 loop ('I')
___1__
2r(l +<r)
['" dtta-1
lo
f d'p [e-t(p'+m'+oiP'/Z)
(2w)•
-(cp O)j.
Tbe integration aver the momentum p is simple Md yields:
.!.r••• (m)
1 ["" dt ta-3 f.,-<(m'+n•' /2) -(•o OJ]
V I loop r 32lt2r(1 + u) lo l r •

The integration over t can then also be performed

V1 r'"~< ( )
1 loop 'I' 3211'20'(1
1<r)(2 G) [Cm , + gcp2;z)<•-•l - ('I'= 0 l] •

Expanding this expression for u small and keepiDg only the divergent and linite parts,
we finally obtain:

('I')= - -- (m' + 1!..,•) [-~+In (m2 + !!."'•)- ~] - (<p

1 2
.!..rreg
V 1 1oop 64or2 2 O' 2 2
O).
The coefficient of the divergeot part differs by a factor 2 from the one obtained in di-
mensional regulariul.tion. This is due to the choice of the normalization of ".
10 INTRODUCTION TO RENORMALIZATION THEORY.
RENORMALIZATION GROUP EQUATIONS

We shall not enter into an extensive and general discussion of renonnalization theory,
but instead present the essential steps of tbe proof of the renormalizability of a. simple
sçala.r lield theory: the tjJ4 field theory in d = 4 dlmensíons. Huwe""r, ali the fundamen-
tal dífliculties of renonnalization theory are already present in thís particular exrunple
a.nd it will eventua!ly become dear that the extension to other thoories ís not difficult.
We ha"" folluwed the elegant presenta.tion of Callan (Les Houches 1975} which allows
renormalizability and renormalization group (Callan-Syma.nzik) equations to be proved
at once. ThiE presentation ís specially suited to our general purpose since a la.rge pa.rt
of thís work is devoted to applications of renonnalization group (RO). Moreo""', item-
phasi?.eS already at this teclmícallevel the equivalence between renormalizabilíty and the
existence of a renormalization group.
One drawback of our proof o( renormalizability ís thst it directly applies only to massive
theories and the existence of a masskm theory requires a specific discussion. Section 10.9
is devoted to this problem. A di1ferent form (homogeneous) of RG equations foUows.
Finally Section 10.11 oontain.s a few remarks about the cova.riance of RG fuoctions.
ln the appendíx we briefiy outline another method which more clea.nly separa.tes tbe
small and large momentum regjon a.nd had been employed to gíve another proof of
renormalizability. lt relies on a partia! integration of large momentum modes. Finally
W<! discu.,; some super-renormalizable thoories a.nd the background field method.

10.1 Power Counting. Renormalized Action

We consider the local action for a scalar field ,P(x) in four dimensions:

S(,P) = j d x ( !a,.,pa_..,p + !m ,P + itg,p•).

4 2 2 (10.1)

In Section 8.3 we ha.ve shown that the </> 4 vertex ha.s dimension zero in four dimensloru; and
this actíon i• thus renormalizable from the puwer counting point of view: the superficial
degree of divergence of correlatíon functions is independent of the order In perturbation
theory.
Power C<>Unting. For a proper vertex with n externai lín.., (IPI n-poínt correlatíon
function} the degree of divergence ó ia:

li=4-n.
2
We alsú consider insertion ofthe q) (x) operstor. The degree of divergenceof the funetíon
r(l,nJ,

rll,n) (!/!, ... ,l/1; X1, · •• ,x,) 2-l (qí2 (!11) • • · f/! 2 (Y!) lj> (xl) · · .<J>(z,)) 1pl,
is (equation (8.34)):
6=4-n-2L
194 lntroduction to Renormali2atian Theory 10.1

Q Fig. 10.1
e
The ,P..field 2-point function: I= O, n = 2, 6 = 2.

Fig. 10.7 The ,P..ficld 4-point function: ! =O, n 4, li O.

····C:X
Fig. 10.3
····c=x=x
The (</J'N) correlation function: I= 1, n
·-<ú
2, 6 O.

····O ····CXJ
Fíg. 10.4 The (4>2) average whlch is a coMta.nt: I= 1, n
····<ill
O, 6 = 2.

····O··- ·-·~··- ··-<TI>··

Fig. 10.5 The (~?<J; 2 ) correlation function: I= 1, n O, IJ =O.

In figures 10.1-5 are Jisted the superficiaDy divergent functions !IIId the correspondiz\g
first few diagrams.
The diagra.ms witb n O never arise a.s subdia.grams since a subdia.gram ha.s necessarily
W.l lntroduction to &normalí:z:atiQfl Theory 195

externai 1/> !ines. Their renormalization can be díscussed separately.

To give a meaning to perturbation theory we Dow replace the action S( !/>) by a reg-
ularized oction S,. (r;!>), usíng one of the regulariza.tion schemes presente<! in Chapter 9.
In what follows we explicitly refer to a cut-off because it ís more in the spirít of tbis
wor*, and some arguments would have to be slightly modilied in the case of dimensional
regula.riza.tion.
Dimensional analysis. In wha.t foUows we systematically use mass dimensional analysjs.
Let us rescale in ection (10.1), after regula.rization, mass and cut-off by a factot
p and
cale ui ate the generating functional W (p, J) of correlation functíons of q, (x / p):

m, A,_. pm, pA, f d4 xr;t>(x)J(x),..... f d 4 xr;t>(xjp)J(x).

We then make in tbe functíonal integral the change of variables on the fie!d:

r;b(x) = P<P' (px).

The source term heoolll€S:

After the change of variables x' px, the action (1 0.1) resumes its inltlal fonn (the
durnmy varigbJes <f/ and x' being renamed ifJ and x). Since the source term ís now
multíplíed by a f.o.ctor p, th" generating functional W (p, J) satísfies:

W(p,J) W(Jp).

For tbe L<>g<>ndre tra.nsform r ('P) this implie;s:

lf w.. now cxpa.nd r (ip) in powers of I" we ftnd:

r'"' (:r,fp, ... ,x,.fp, pm,pA) "'p 3

"f(n) (x~>,., ,:r,., m, A),

After Fouri<>r transformation and fectorí•ation of the é-function of momentum conserva-

tion the relation becomes:

rtn) (pp,, ... ,pp,.,pm, pA) = p•-nr{n} (p 1 , ••• ,p.., m, A).

We find " m""" dimens!on 4 -n which is also the dimension in the sense of power counting
of r(n}. The argument g<>neralizes to f(l,n).
The renorma!ized acti<>n: counterterm•. We now considerthe regularized action SA(<fl).
We introduce two renorma!ized parnmeters m, and g,. We then want to show tha.t it is
possible to rescale the field r;t>:

RIId to choose the bare parnmeters m and 9 as functions of rn,, y, and A ín such a
way that ali tf>r correlatíon functions have a finite limit, arder by order in the loopwise
 Jntrod11Ctie>ll to Renonnalizaticn Tkeory 10.2
196
expansion, when the cut-off A becoroes infinite at m, and g, fixed. The field .;,, is called
the cenonntillzed fi.eld.
More precisely let us intrnduce the notion of renmma!Ued action S, (Ql,) which i8 the
original action expressed in tenns of renormalized variables:

SA(if>) S,(<P,) jctd:r[l<f>r(m~-6)~</>r+~g,if,~

(10.2)

The notation (m;- LI.},.. i8 there to remind that the action is regularized. For Pauli-
Villars's regularization it means:
t,3
(m~ a, A• + ....

The identity between the renormalized action (10.2) and the regularized action (10.1),
called the bare o.ction, is expresse<! by the set of rel!ltions betweeu renormalized end bare
quantities:
.;. = z'l'q,,,

{ 2
9 = u.z.;z',
m = (m; +6m2 ) fZ.
Z and z.fZ2 are respectively called the fi.eld amplitude a.nd coupling constant renormal-
iza.tion constants, 6m2 cha.raderizes the ma.ss renormalization.
In action (10.2) we have explicitly sepa.rated a. tn!e levei action:

from the set of coun!ertcm18 pa.ra.metrized in terms of the renorma.liza.ti<Jn constants

6m2 1 z.,and Z which are formal series in g,:

6m2 = 0.1 (A) g, + a2(A) g: + · · ·

Zg = 1 +b1 (A)g, +b:.(A}l +··
{
Z 1 +ct(A)g,+c•(A)g~ +··· ·

We want to prove that the coefficients a,. (A), bn (A} and c, (A) can be chosen in such a
way that ali correlation functions ha.ve a finite large A limit, <Jrder by order in g,.

10.2 One-Loop Divergences

We have a.lready shown m Section 8.1 that the t/J~ (the lower index is the spoce dinoen,ion)
fi.eld theory could be renormalized at one-loop order. Let us here repeat the a.rgument
in the """" of the t/!1 field theory. ·
First we note that if we rescale the field ~,:
10.2 Introdti.ction to Renormal?at:ion Theory 197

the dependence m g, is factorized in front of the tree levei action:

In the q,• fie!d theory, the Joopwise expansíon is an expa.nsion in powers of g, at ( ,;g;q,,}
fixed.
We now write the generating functional of proper vertices r ('P) at one-loop order:

At the tree leve) the counterterms by definition do not contribute and therefore:

At one-loop order the oontributions ccming from the tree level actlon are (equation
(6.49)):

Expo.nding in powers of cp; we find two divergent diagrarns w!Uch are reprerented in figure
10.6.

Q Fig. 10.6 On<>-loop divergent diagra.mB.

· Using the regularize<! propag&tor .,-p'f11' (p2 + m~)- 1 we can caloulate the divergent

(i) n = 2: the coefficient (a) of g,cp;:

1
4(a.)
1
4
f 4
d q .-•'1 '
11
1 (A 2 A)
(2,.)'(q2+m;)=32-.r' 2-m~lnm; +O(Ao).

n = 4: the coefficient (b) of g~cp~:

divergent part fi'• ('P) of r, ('P) in the absence of counterterms is thUE:

J1i• 3
;,.2 f d 4x [ ( A
2
2
- m~ In~) g (cp,(z))z- ~In~. g 2 (cp,(x)) 4 ] •• (10.8)
198 Jntroduction !<J Renormalúation Thw.y 10.2

Note the absence of a term proportional to f d 4 x (él.,<p, ) This is a pecullaríty of the ,p4
2

fiald theory at one-loop order.

Let us now ta.ke into acwunt the oounterterms at one-loop order:

! (Z -1) (8.,<P.l + ~6m2 f/>~ + ~g, (Zg -1) q,:lonoloop

= ic1 (A)g,(8.,q,.)
2
+ ~a1 (A)g,,P~ + ~b 1 (A)g~.p~.

At leadmg order tbese counterterms contribute additively to r (l"r)· We can choose them
to elinJinate the divergen<:es and tberefore renormalize the ,P4 field theory at one-loop
order. Of course the condition of finitenes.s of correlation functions determmes only the
divergent part of the counterterms. lt is pOSBible to add for example to a1 , b1 and
c1 any a.rbitrary finite coru;tant. Tbe difference between the IPI correlation functions
corresponding to two dlfferent choices is of the form of the tree levei functions, j.e. a
coru;tant for the 4- poínt function in Fourier space and a first degree polynomial in y,
p being the momentum, for the 2-point function. In this chapter ít ís convenient to
determine the renormalization coru;tantB by a set of renonnalization conditioru; imposed.
to the renormalízed lPI correlation functíoru;:

which in Fourier space r"ad:

(10.9)

We have of course chosen conditions consístent wíth the tree apprruúmation. Since we
have three renormalization constants, by ímposing three conditions tG the superli.cià.lly
dívergent correlation functions, we determiue them completdy. The renormalization
coru;tants !l.l"e then given by:

(10.10)
 lntro<Ú.lclion to &rwrmul-4ation Theory; 199

~
(c)
p (d)
Fíg. 10.7

10.3 Divergences Beyond One-Loop: Skeleton Diagrams

At any arder in perturbatlon theoJ)' the 2-point function is quadratically divergent and
the 4-point function logarithmically divergent as powo;r counting shows. However at
higher orden; a new difficulty, which we ha'"' already mentioned in Section 8.2, arises:
superlicially convergent díagrams have diver~nt subdiagrams. Let us ta.Jre the example
of the 6-point function: at one-Joop arder it is given by a convergent diagram, but at
~loops the diagrams shown in figure 10.7 appear.
We recognize inside tbe dashed boxes divergent subdiagrams. However we identify
them with one-loop divergences of the 2-point function (c) and the 4-point function (d),
for which counterterllllS have already been provlded. lndeed at this arder a diagraw
(c') appears in whích the one-loop counterterm for the 2-point function is inserted on a
propagator and another one (d') in which the vertex of the tree levei action is replaced
by the one-loop counterterm of the 4-point function (figure 10.8).

(c')
Fig. 10.8
A (d')
TW<>-loop contributions from one-loop counterterm.s.

We want to show that this Í5 generally true, i.e. that counterterms which render the
dívergent functions linite, at higher orders also ca.ncel the dlvergeilCe of subdiagrams of
i B<lpeml:J&Jtly convergent functions.
purpose ""' íntroduce the notíon of skeleton diagrnm: a skeleton diagram
oonvergent diagram, i.e. it ÍB superlicíally convergent, and has no di""rgent

the one-loop diagrarnB of the 2n-point functions, n > 2, ali a:re skeleton
10.9).
superficially convergent diagram, can then be obtained from a slreleton
by replacing ali vertices by r(•) Wld al1 propagators by (f(2 ))- 1 BJid expanding
powers of the coupling ccmtant g,.
For exa.mple tbe diagrams (c) and (d) are generated by the expansion of the dressed
diagrnm of figure 10.10.
important property !s the followíng: íf in B dressed slreleton diagram, f(4} Wid f(2)
replaced by the renormalized functions r} l and r~ l the dreszed 51releton diagram
4 2
200 lntroduction to Renonnali.zatian T}u)ory 10.3

Fig. 10.9

Fig, 10.10

is fmite. This is a direct consequence of the following; bounds on the large momentum
behaviour:

!r;2J (>.p) I:S >! X power of ln ;\, )

!r;•) (>.p,)l :$ power of In>., at any finite order for ;\ __. oo. (10.11)

!t; l (>.q; Àp., ÀPl) I:S P""""' of ln ).,

1 2
•

Wedo not derive the&> hounds here butafew comments can befound at the end ofSection
10.6. These bounds for the large momentum beba.viour of the. va.rious renormallzed
functions, which are valid for arbitrary momenta, differ from the tree levei behli'/iour ·
only by powefl! of logarithms (at any finite order in the loop expansion). Therefore
power counting arguments still apply and superficíally convergent diagrama are then
coll'rergent.
Note that similar estimates exiBt for superfidally convergent functions but are then
valid only for generic momenta (""" Sections 12.3-12.5).

Fig. 10.11 Dlwrgent contribution to rtt.•).

Tbe bounds (10.11) together with the skeleton expanslon completely reduce the proh-
Jem of renormallzBt\on of superfidally con"""gent proper ,.,..e;íces to tbe renoll'll&li2oation
of the divergent proper vertices. Tbe a.rgument a.loo applies to tbe proper \'ertices rtl,n) '
with 1/12 insertion. Let us for exarnple consider tbe diagra.m of figure 10.11, wbich con-
tributes to the superficially convergent function r-11 •41: lt bas divefl"llt subdiagrams and
10.3 lntroduction to Renormaliwtion Theory 201

is generated from a skeleton diagra.m by repledng propaga.tors by (r<•l)- 1, vertices by

f'{') and the q,2 .p,P ""rtex by rU,>l as sbown in figure 10.12.

Fig. 10.12

In the next section we tberefore examine the renormali:r.ation of r<1l, r<•l nnd r< 1•2l.
The díagranu; contributing to these functions are superficially divergent but ha"" also
divergent subdiagrarns cotresponding to the divergence of tbe same functions a.t lower
orders. Figure 10.13 provldes an exarnple.

Fig. 10.13 Tbr.,..loop rontribution to the 2-p<>lnt functkm.

However another problem arisel3, the problem of overlapping divergences. Let us con-
tiider for exa.mple the diagram of figwe 10.14.

Fíg. 10.14 Thre&-loop contribution to the 4-p<>int functíon.

Figure 10.15 sbows the set of divergent subdiagra.ms. The three subdiagrams have a
part. Tbe concept of insertion of divergent diagra.ms of lower arder is therefore
longer well-defined. Tbú; is the problem of so-called overlapping dívergences.

·~ ··n··
Fíg. 10.15
t~
OverlappÍI\g di_,.geDt subdiagr.....,.
. 'Y.. .
202 lntroduction to Renormaliwtion Theory 10.4

In the next section we develop a specific technique to deal with this problem, based
ou a diiferentiation of the diagrams witb respect to tbe mass.

10.4 Bare and Renormalized Con-elation Functions, Operator <P 2 lnsertions

Before we begín the discussion of the renormalitation of the superficially divergent proper
vertices a few remarl<s are in arder.
Bare and renorm.alized correlation functivns. By definition the bare oorrelation fnnc-
tions are th~e calculated witb actioo (10.1) properly regula.ri2ed. Too renormalli.ed
correlation fnnctions are calculated with action (10.2). The relation between ba.re and
renormalized fields:

implies immediately the relation:

w(J!v'Z) =Wr(J), (10.12)

in whích W (J) and W, (J} are tesP<'ctively the generating fnnctioniili; of conoected bare
and renormalized conelation fnnctions. Sine<l in the Legendre transformation J and <p
are dual, it follows immediately:

(10.13)

a télation whích on tbe proper vertlces trMslates into:

(10.14)

Opemtor q,2 insertíons. Actnally we need also the bate and renormalli.ed q, 2 insertions.
We therefore add to the bare action (10.1) a souroe term! J K(x)rp 2 (x}dx:

S(rp,K) ~ S(rfJ) +!f dxK(:r)ifJ (x). 2

(10.15)

We now ooruider the fnnctional irrt.egral:

Z(J,K) f [dif!)exp [-s(q,,K) + J dzJ(x),P(x)]. (10.16)

Fnnctional differentiatíon witb respect to K (:r) generates insertions of the operator

-!<fl
2 (:r) (Sectloo 8.4). In the sameway ifwe ooruíder the action S,(,P.,,K):

in which Z 2 is a new reoormollzation oonsta.nt, the fnnctional integral:

z, (J,K)-= j [d.p,.)exp [-s, (,P.,,K) + j d:r J(:r)if!,(x)), (10.18).

10.4 Introductiun to Renarml>l'::ation Theory 203

genera.tes the renormalízed correlation functioM with -~.p 2 insertioos (we temporarily
normalize Z (J, K) to Z (O, K) 1 to eliminate the pure .p2 (x) correlation functions).
The relation between renormalized and ba.re functionals is then:

Wr(J,K) W (Jf./Z,XZz/Z). (10.19)

After Legendre tra.Il5form&tion tbis relation become5:

(10.20)

or iu terms of proper vertices:

(10.21)

Since we now ha.ve a new superficially divergent function r(I,Z) and a new renormalization
constant, we ímpose an additional renormaliz.ation condition:

r;t·•l (q"' O;p, = 1>2 =O)= 1, (10.22)

wlüch &gain is consistent with the tree approximation.

Let oc; here note that if tbe source Xis a constant then, according to equation (10.15),
it generates just a shift of m2 , the ba.re mass squa.red:

(10.23}

Therefore in this case differentiation with respect to X is equivalent to differentiation

wíth respect to m 2 • On the other hand if X(:z:) is a oonsta.nt, its Fourier trllDSform
is proportlona.l to 6 (p), which means that it generates correlation functions with ~
2

insertions at zero momentum:

{10.24)

This equation hss a di&grammatic interpretatiou: the diagrams contributing to the r.h.s.
are obtained from the diagrams contrlbuting to rl!.nJ by doubling a prop&gator in aU
_ ways (up to a sign). In~ 10.16 we give the example of r<4l and r< 1•4l.

Fig. 10.16
 ·
204 lntroductwn to RenormaiizatiDn Throf"'J 10.5

By differentiflting with respect to the bare ma.ss, we are able to relate superficlally
divergent oorrelation functions to functions whicb have a skeleton expansion. F\Jrther-
more, at a given number of loops a function whícb has a skeleton expansion is only
expresse.:! in terms of divergent functkms whicb have at lea.st orn: loop less. We see there
a mechanism to prove renormalizabi!ity by induction. However we want to insert ínto
the skeleton expansion renormaliMd proper vertices. This introduces some a.dditional
diflicult.ies whicb we will disrover om:e we have tra.nsforroed equation (10.24) into a.n
equation for the renormallzed proper vertices.

10,5 CaJlan--Symanzik Equations

The starting point of our analysi< is equatíon (10.24) which shows that by differentíating
with respect to the bare mass we improve the large momentum behaviour of FeynmlUl
diagrams. We first translate equation (10.24) into an equation for renormalized functions.
Let us rewrite it under the form:

m, ~~
8~nr g,h
f'(l,n) = (m. _!_I m2) ~f'(l,n) (m,. _!_I m2)
{)m, n,A 8m
=
8m, •·A
f'(l+!.n) (O, .•• ).

(10.25)
We can then use rela.tion (10.21), replace bare by renormalized functions in equation
(10.25) and apply che.in rule to transform differentiation with respect tom, at g aod A
fixed, into düferentiation at g, and A fixed.
We now introduce a set of definitions, and immediately take into account dimensional
aoalysis:

m, -/-] g, = fJ (g,, mA,), (10.25)

l1lr ··"
m,:!_jVl'"r g,l\
lnZ(g,mA,) (m,C:, r
8
+JJ 8 )tnZ=J1(g,,m ,).
Yr 11
(10.27)

l1lr a':,., L,. In (Z2(Z) = (m, a':n. + fJ a~,) In (Z~/Z) = '12 (g,. ~) '(10.28}
Z Zi
1
( m, a: lu.A m2 ) = m';t; (g, ~·) . (10.29)

In terms o f the düferentíal operator Dcs:

;;~i,n)(q), ... ,qz,Pt,··-,Pn

Dcs.L·; . ) =mra
2 ( lnr)r-(l+l,nl(Q
9nJl r . )
,qt, ... ,qt,Pl1•~·,Pn.
(10.30)
Equatlon (10.30), in the infinite cut-<~ff Umit, then yields an equation for the renormal-
ízed proper vertices, fin;t derived by Callan and Syma=ik, called therefore the Calla.n-
Symanzik (CS) equation, which will piBy a central role in the part of this work devoted
to phase transitíons ( Chapters 2&-36).
10.5 205

To prove renonnalizability, we shall prave inductively on the number ofloops both the
exist.ence of the CS equation end the finiteness of oorrelation fuw:tions.
Note that tbe CS equation in the form (10.30) expresses only that we bave rescaled the
correlation functions a.nd made an arbitrary cha.nge of parametrization. Tb be able tD
prove that the renormalized correlation functions have a finite cut-ofr liwit it is necessary
tD determine the renormalization constants and therefore to impose on equatlon (10.30)
the consequences of renormalization conditioru; (10.9) and (10.22):
(i) n 2,1 =0
At zero momentum we obta.in:

The function (1 is re!ated to 11:

(1 2 -fi. (10.31)

!f we then differentia.te wíth respect to momentum squared we find:

(10.32)

(ii) .. =4,! o
At zero momentum we obta.in:

(10.33)

(ili) n 2, I
Aga.in at zero momentum we get:

-'1- TI'J = m~ (2-11) r~··" (0,0;0,0). (10.34)

We bave rela.ted ali coeflicients of tbe partia! dilferential equa.tion {10.30} to values of
proper vertices at >ero momentum. From tbese relations ít foUows tha.t if we cBD show
that the renorme.li>ed proper vertioes bowe a. limit when the cut-oJf becom"" infinite, the
functions fJ , 'I 1111d TI'J will alsc have a liwit.
Note also that if we know the coeflicients of the CS equatioDB, we ca.n calculate the
renormalization consta.nts from equatlons {10.2&-10.29).
Prnlíminary "'marks
f'(l,2) (O;p, -p) at arder g does not depend on p:

f'<1.2) (O· I'- p) = 1- _!_g_

' 2 (21r) 4
f (p2
d4p
+ m~)~
+O (!l).

(10.21,10.22) thEn imply:

z_, z. = l
1 g,
+ -2 (2 )4
·j 4
d p
2 +o
( 2)
g, '
1f (p2 + m~)A
f'~1,2)(0;p,-p) = 1 +0 (gt).
206 lntroduction to Renormalization Theory 10.6

Fig. 10.17

Fig. 10.18

We condude that the exparu;ion of '1 (g, ), which ean be calculated from equation (10.32),
begins at order g~.
(ii) The first diagTam contributmg to fll.•J is of order g'; (see figure 10.17). lt then
foll<>W~> from equation (10.33) and the previous rema.rk tbat the expansion of tbe function
{3 begins at order g'{. Thus the operator {Jfi/IJg, whiro appears in the CS equation is of
g,.
(iií) The function f< 2 -2 l has a first contríbution of order g, (see figure 10.18). Equation
(10.34) show.; that f12 also begins at order g,.
Cl1L'Jter properties and analytictty at low momentum. ln Section 8.1 we bave uzed the
regularity of the one-loop diagTams nea:r zero rnomentum to show tbat the divergent
contríbutlons a.re polynomíals in tbe momentum variahles. Tbis is more generally true:
in a massive theory (witb a mass m :5 A), oonn~ and lPI correlation functions are
analytic functíons around p = O as can be 5eell on the expression of regularized Feyrnnan
diagnuns. This property, whicb will again be needed in the indnctive proof, implíes
cluster properties: connected correlation functions decrease exponentially in space for
large sepa.rations of the arguments (for detllils see Appendix A7.1).

10.6 lnductlve Proof of RenornnalbabUity

We have sbown in Section 10.2 that we could render tbe tbeory one-loop finite. We now
assume we have sbown that the correlativn functions defined by equations (10.21) and
renormallzation condítions (10.9,10.22) ha.ve a finite límit up to loop order L, at m, and
g.. fixed, when the t:ut-olf A becomes íofuúte. 'I'his means that fFl, f~'l, r!
1 1
·' ha...., .
an infinite cut-off Jirnit up to arder g~, g~+ 1 and 9:-
respectívely. Mó we have shown in ,.
Section lO.:i, from equations (10.21,10.9,10.22) follow the CS equations (10.30) and the
relatiollll (10.31-10.34).
We now use CS equation (10.30) in tbe form:

(10.35) -
10.6 Introduction to Renarmtdj.zatíon Thet>ry 207

and show tbat tbe r.hJ!. i.s finite at (L+ 1) loop order. We note that r<l,n) ill the r.h.s. is
only needed at loop order L because its coelficient is oi order g,. For r! 1+t,n) two cases
a;rille: either it h as a skeleton expll.llSion and is therefore finit.e at loop order L+ 1, or the
CS equation hss to be iterated. However before discW!Sing the <orrelat;on functions in
detalllet us examine tbe coeflicient functíons.

10.6.1 Coeffir:ients of the. CS e.quatíon

(i) Because Bf'(t,•l (ar? is of arder g~, equation (10.32) then impliffl that 'f/ is finite up
to order gf.
· (ii) The funct\on f0. 4J ís superficially ronvergent. lt bas tberefore askeleton expansion.
The first dressed skel..ton diagram rontributing to f!t, 4 J is shown in figure 10.19.

H the functions fVl, 4

f~ ) and
A tP·'l
(L+ I) loop order, which means up to order
Fig. 10.19

are finlte up to L loops,

gl+•.
r!'·
4
l is ânite up to
Equation (10.33) then sbows that
the combination {3- 2g,'f/ is finite up to order g, H, Since 11 is finite up to order gf, tbis
ímplies that {3 ls finlte up to order gf+>.
(iii) The function f\ 2 •2> is n.lso auperl:icinlly' convergent. It has a skeleton expansion.
The first dressed skeleton is shown in figure 10.'20.

Fig. 10.20

2 21
function f! · i.s ther~fore also finite up to (L + 1) loop order which means up to
(L + 1) in g,. Since r< 2·2l ís of OJ'der g., at order gf+> tbe sum '1 + '11:1 caJculated
equation (10.34) invol""" only 11 at arder gf a.nd N2•2 ) at order gf+ 1 and thus is
The function 11:1 ís then :6nite up to order gf,
We now pr!Ml t.hat the functíons f'~2J, f'f') and N'·2 ) are, wit.h the stated """umptioDII,
· up to (L+ 1) loop arder.
208 Introdvction to &normalization Theory 10.6

10.6.2 The (1$>411$>41) wrrelo.tion function (l =O, n = 4)

Let us conslder the coefficient af arder gf-+2 in equatian (10.35):

m.~
ôm,.
[r<•}J L+2
T

Since (Jôfôg, is of order g, !llld '7 of arder g~ for g, srnall, in the ftrst term af the r.h.s.
we need r!'J only up to arder gf+' which is tinite by assumption. Let w; now separa.te
in r~'l the Ieading terrn g, a.nd a rernainder of order g~. For the terms of arder g; and
lrlgher we need TJ only up to order L !llld (3 up to arder L+ 1 whlch are ftnJte. The Jea.ding
terrn in r~·) then ínvolves the combination:

[(-f)&~, + 211) g.] L+2 = [-(3 + 2'79r)u2 •

nnd we have shown above that fJ- 2g,f1 is finite up to arder g:-+ 2•
f'!
Finally 1•4 ) is finíte up to order gf+'. In eddítion its expansion in powen. of g,
begins only at order r!,. Therefore the factor (2 - '7) ís only needed up to' arder gf. The
conclusian is that m,af>S•l fãm, is finite at (L+ 1) loop order.
Perturbative integrotion of CS equations. By integrating equation (10.35) we now wa.nt
to show that f~4 l iteslf is finite at (L+ 1) loop arder. The function r~'l is dimeiJSianless,
i.e. ínvariant in a dila.tation

(p,m,,A) >-+ (pp,pm.,pA),

(in wbat follows we denote by p oll four rnomenta Pt···P4)·

Therefo.re we can write:

(10.36)

wíth:

A-t'JO
(R.. ~.fh) <
lim tt•J
mr 'T7lr
00.

{ j( 4l (o.;,.9•) m, ~g,=O.
Let us now integrate equation (10.36) between m, andA (A ca.n be replaced by e.ny scale
proportíonal to A):

Renormaliza.tion conditíons (10.9) together with the regula.rity at low momentum ímply:
 10.6 Jn!roduc!ion to Renorma!i•ation Theory 209

The integral h as a large cut-olf limit if we restrict the domain of integration to [m., 14],
« 14 «
rn, A:
.
hrn 1~ ~f
A-CJ:)
dp (<l
m~r P
(P-,-,g,
P P
A ) <co.
pC:J\

We bave still to examine the values of p of order A. Then Afp is of order I, whüe pfp is
sruall. The integral depeneis on tbe s:mall momentum behaviour of the ~point correlatíon
function. Again we use the property that in a massive theory (the mass is p ~A) the
- correlation functions are analytic around p "" O.
Since in addition f( 4 l vanishes at p O:

the remaining integral is then bounded by:

We wnclude that f'~ 4 ) has a large cut-off llmít at (L+ I) loop order gíven by:

r;•) (p, m,, g,) = g, ! "" ~11•> (L,u.) .

I p pm,
(10.37)

both sidcs being expanded up to g~+>. We now repeat the argumomt for f'!'·"l.

10. 6. 9 The { .p 24>4'} con-eiatwn functíon (l 1, n = 2)

consider the terrn of arder gf+ 1 in equatlon (10.35):

~ [r<'·•l]
m'ômr r L+l
=

Tbe first terrn in the r.h.s. involves rP· 2

) up to order L since {38f8g, is of order g., and
. (3 up to order (L+ 1). Both are finite. The second term again involves
2
f1'·
l up to arder
• L since ('1 + '1:1) ÍB of order g., lllld ('1 + '1:1) at order {L + 1) whkh ís finite (altbough 'I
N
imd '1:1 separately are not). The Iast term invoives 2 •2 l up to order (L+ 1), and 11 up
· t.o arder L sínce N • l is of order g,. We conclude that m,ôf~ • ) fôrn, is finite at (L+ 1)
22 1 2

loop arder.
- The function f! l,>l is aJso dimensionless. I ts valoo at zero momentum is fixed by the
l:""•mnalliz•LtkJ,n oondítion (10.22), therefore:

m ~f'll,
'âm. r
2 l(O·OO)
'
O
,

BJlalysis ís then tbe 51111le 38 rorN•> anct we wnc!ude that rP·•J

bas a llnite !imit
_!!Ut-off at loop arder (L+ 11 We can now use equation (1 0.32) ond argument
Since r(!,>) ís finíte up to order g, +I, 'l is finite up to the same order gf+'.
210 lntn>duction to Renarmalüaticn Theory

JO.ff...l The (#J) rorre!ation functíon (! O, n 2)

The term of arder gf-+ 1 in equation (10.3á) is:

m,....!!_
âm,
[r~''] L+l = [(-(3_!!_
âg,
+ Tj) f! 2 '] L+l + m~ [!2 -'I) i'!'·2'] L+l •

In the first term of the r.b.s. we need only n•l

up to arder !l
since it is multiplied
by terms of arder g,. We nave al.so sh!JWII abo-re tbat 11 is finite up to order g~H and
fJ is finite at thls arder by argument (ü). For the second term we ha.ve just shawn
above that the two fa.ctors are finite up to order g~+'. Let us consider the quantity
r! 2
' (p)- m~- p 2 . Renormalizatíon cooditions imply tllat it vanishes as (p )' for IPI
2

small. It has mBBB dimension 2. We set:

m, !. [r!2J (p)- m~- p'] = m~t<•> (!,, ;, ,g,),

witb:
(.!....~,g.)

l
lim /(2) <ao
A-ao fflt mr

!(') ( ~.' ;,.g,) = o_ ({p2 ) 2 ) for IPI - o.

Let us then integrate equation (10.39):

f(2)
r r~ 2
V'tmr • .9B A)- mr- -
2
P -- 1!.2 [f'lr2 l (E.
A' 1t9rt 1)- 1 -
2 2
P JA ]

- I" pdpf(>) (!!..~,y.).

1.... p p
2
The integrated term in the r.h.s. decreases like (p2) /A 4 for A large and tberefore goes
to zero.
For the same reBB<m f1 2l (pjp,Afp,g,) is of order ljp4 , for p ~A, and therefore the
inlinite cu~fllimit can be taken in the integral.
This concludes the induction. The a.dvantage of this method is that at tbe reme
we ha.ve deri..OO the CS equatioos:
a a
[m, âm, + ,B(g,) âg, ~II(Ur) -lrn(Pr)] f~!,n) (q;;p;)
[2 -11 (9r)] m~f~l+l,n) (O,qj;P<) {10.41) :';.

10.6,5 Tlle larye momentum beh.avi.our of superficially divefl1ent correlation functians

We see from this derivation that we can al.so use induction to estimate the la.rge momen:··:·
tum behaviour of corre)ation functions. lf we assume the bounds (10.11) at loop arder'··
L, then r<•.•J and r(2,2) are given by convergent integrais "t Ioop order L+ 1. It is not
too difficult to bound their large momentum behaviour hy powers of logarithms.
Let us now again consider the extunple of the 4-point functions. Onoe we ha.ve estab-
lished representation (10.37), the induction hypothesis tellB us that the function f(4) w"
the r.b.s. is bound by:

lt!•l (.\p/11lr,9r)l < oonst. (ln.\l{LJ for À--> oo.

L±±S._, :XJJ!
 Introduc!ion to Renorma~ization Theory 211

We then decompose the integral over p intú the suw of two terms:

1.
I
00
dp 1,.,
p
(..2.
pm,
9.) 1~r= dp 11•1 (~.g.)
=
pm,p
+f.~ dp 11•1 (~.g.) _
pm. l P
The first integral can be bound by a oonstant. The second integral can he bound using
the Jw-ge momentum behaviour of j14):

{ d: jf!•J (!,9•) I< ronst. t ~ [tn G)]~(L) _ (lnÀ)'(L)+I.

The argument is the sa.me far r< 1•2 l. This last bound can then be used to bound f'! 2l.

10.7 The (q,2q,2} Correlation Function

have seen that the (t/J 2t/J2 ) proper vertex I'f2 ,0) bllS §Uperficial degree of divergence
zero. Actually even in free field theory (g, = O) it is divergent (see figure 10.21 ). An
!ldditional renormali2ation is needed. We impose the renorma.lization condition:

····c=:>···· Fig. 10.21

(lo.42)
obviously consistent with the tree levei appraximation. We expect the relation between
bare and renormalized proper vertex to be now:
fF.ol (p} ( Z,J Z)2 [:f!2,oJ (p)- :f!2,o) (o)]. (10.43)

Dlffer·enlcial;in~r with re:spect to m, at g and A fixed we theu obta.in:

[ m, a: + ,B(g,} 8~, - 2'1:1 (g,}] r;2,0) (p;m,g,)

m~ !2- TJ (g,)j [f! •n) (p,-p,O;m.,g,)- Í'~3•0l (O,O,O;m,,g,)]
3
(Io.44)
derlvation of this equation follows the sa.me lines os in previous cases. It uses the
that r<J,O) has a skeletún expan§ion and that ths L loop order of r(•.o) is of
Finally in the integration with rffipect tú m, one to.kes intú account tbe
enc>rmlnlization ooudition (10.42} and notes that f'< 2 •0l is dimensionless in mass units.
It is possible then tQ summa.rize all CS eque.tions by:

[m. a!.. + ,B{g,) a~. - ~T)(g,) -l'l:l(gr)] t; 1

·") (q;;p,;m,, !Ir}

= rn; (2- 'I (g,)) f'!l+l,n) (O,qm>i; m,g,) + 6noó12B (g,), (10.45)

-rn~(2 TJ(g,))fl3•0l (O,O,O;m,g,) B(g,), (10.46)

the l.h.s. is dimensionless.
212 lntroduction to Renonnalization Theory 10.8

10.8 The Renormalized Action

The proC>f of the renonnalisabillty of arbitta.ry massive local field thoories is a rather sim-
pie genm-alization of the proof wo: have given above for the !{!• field theory in 4 dimension&.
We tberefore only explain the rcsults.
We now consider an arbitta.ry field theory renmmalizable by power counting. We
assume that we have added to the tree levei a.ction SA(r/>) ali counterterma needed to
render the tbeory finite up to loop order L. lf we then perform a loC>pwise expansion of
the generating functional of pro per vertices r ('P):

"" r1 (op),
r {'Pl = s~o (!f'}+ L
l=l

the functíonals r1 ('I') have a finite limit for l ~L when the cut-off hecomes in!inite, and
the diagrams contrihuting to rL+l (<P) have no divergent subdiagnuus.
I'f+ 1 (ip), the divergent part of rL+I ('I'), is therefore a generaJJocal functionalJinear
combination of all vertices of non-pooitive canonícal dimensions (except if symmetries
forbid some terms). By adding to the renormalized action the counterterms -I'f~ 1 (;p)
we render the theory finite at {L+ 1) loop order. The counterterrns are of cour.m only
de!ined up to a.n arbítta.ry finite part, linear comblna.tion of the sa.me vertices which ·
appear in r1i+,. lt is sometimes convenient, in arder to fix this arbitrariness, to impo""
specific renormalization conditíons, "" we bave dane in previous sections. The resulting
renormolized action is then also a geneml local functional of lhe jields, linear combination "
of all vertices of non-positive canonical dimension.
Let us show in the case of the t/!' field theory in 4 dimensions th"t this sta.tement
indeed summarlzes the result derived in Section 10.6. The divergent part ofr (I") at loop
order L, after renormalization up to arder L 1, should have the form:

(10.47)

because the vertex 4>2 have dimension -2, and the vertices ( 8~1{!) 2 and 1/14 dimei!Bion O.
No odd paw<>rs of tjJ appear hecause the tree leve! action is symmetric in rJ>,... -if>.
Let us now take another example, a field theory containing a boson field rp, a.nd fermion
fields '1/J, 1[1:

S(r/>,o/1,11>)= jd•x[~(8,.rJ>) 2 +!m2 1{! 2 1{>(~+M).P g1f>lfrtjJ]. (10.48)

We have shawn in Section 8.2 that the dimensions of fields in this theory are:

[tjlj =I, [!fr) = [1f>] = 3/2,

and that the theocy is renormali.,able by power counting. Using power couating we can
=ite the most general divergent term:

-rdiv j d 4 x[f6Z~ (8,.;p) 2

+!6m2 .,~- 6Z.poiii.P
-liMoiio/1- góZ9 iin/J.p+ ~DÃ<'P4 + ~líÀ3'P3 +<l'op]. (UJ.49)
 Jntroduction to Renonnalization Theory 213

Ftom these expreo;sions we <ee that ternu; not present in the tree levei action (10.48) have
been generated, proportíonal to f 4>'d4x, f q\8 d 4x and f q\d 4 x. We say tha.t although
action (10.48) ís renormalísa.ble, in contrast to the q\ 4 lield tbeory it ís not multíplicatively
renormal.iza.ble, in the sense that not ali ooeflicients of tbe renorrnalbed action can be
obtained by a rescaling of those of the tree levei oction. To avoid thís difficulty and to
always make the theory multíplkatively renormalízahle, ~ sball in general try to include
in the tree levei oction ali terms which we expect renormalization to generate.

The M!156less Theory

· We have given a derivatíon of the renormalizabílity of a field theory wbich applíes only
to move field thoories, eince we have used in an essential manner the mass insertion
opera.tion to de.::rease the degree of divergence of Feynma.n diagriiDlS.
We now wa.nt to justify tbe existence of renormalized massless field thoories. The
, correlation functions of a massless theory ca.n be obtained by rescaling, at cut-off and
ltlOmenta lixed, the mass
p __, oo.
At fixed cut-off the límit exists, as will be discussed extensively in Cbpters 24-27, in
dimensions larger than ar equal to the dimension in which the tbeory is exoctly renor-
provided this dimensíon is larger than 2 (because the propagator is 1/~).
the set of arguments of the conelation function in momentum representa-
be non-exceptional, í.e. ali non-trivial subsete o[ momenta. should have a
non-,oanillbing sum.
conditions are met in the g,~ field theory (at non-exceptional momenta) and we
examine what happens when the cut-olf is removed. We sball use the CS equatio:ns
with Weínberg's theorem. From now on we omit the symbol - indicatíng a
tnws,fotlrn when there is oo ambiguity. The argumenta x, y, z will indicate space
and k,p, q momenta, argumenta of the Fourier transform.

Úl'TJ• momentum behaviour and massless theory

us write the CS equatio:ns:
8 â
["" &m, + {3 <u.) og,
; "''eJn,belrg's theorem states that if we scale ali momenta p; ,.... pp;, tbe la.rge p behaviour
iotnun-t:xoeptiorLO.l momenta, at any finíte order of perturbation theory, ís given by the
dimension up to powerB of logaritluns. Thus:

['~l,n) ~ p•-n-2!
(pp;.) _, X power of lnp,

['~l+l,n) (O;pp;) ~
-oo..,.S,p
4 -n-'ll X power of lnp.

thetefote In tbe asymptotic eJq)8JlSion of ['~") (PP•) for p large, we eliminate ali terms
_IUI>Iea.din.g by a power p-2 up to powers of In p, we find a set of IPI correlatíon functions
which satisfy a homogenoous CS equation:

a a
["" âm, + f3 (g,) ag, (10.50)
214 lntrodÚctWn to Renormalitation Theory 10.9

However we know from dimensional analysis

(10.51)

therefore scaling aJl momenta by a fa.ctor p, is equivalent, up to e global factor, to scale

the mass by a fa.ctor p- 1 . The solutions of equatioll (10.50) are thus the correletion
functions of the ma.ssless ,P4 field theory.
Perturbatíve solu!íon of the homogenwWI CS t<l'!'ations. lt !s octually interesting to
study the structure of oorrelatíon functiona implied by equa.tion (10.50). Let us just
here consider, for illustration purpose, the 2-point functíon r~:;i..(p). lt is convenient to
introduce the function ((o,)

ln({g,) =
r·
Jo
dg'7J(g')
{3(g') .

Note that since both 1J and {3 are of order g~ this function hii.S a perturbative expansion.
We then set

The function A satisfies

[m. t!, +f3(g,) ~,] A(g,pfm,) =O. (10.52)

We then expand A and {3(g,) in powers of g, setting

00

A(g.,p/m,) = 1 +LO~ a.. (p/m,), {3 (g, >

"" !3no~ .
2: (lo.53)
2

lntroducing these expansions in equatian (10.52) we find for n ~ 2:

n-1
-za~(z)+ L toom(.z).B..-m+l =0. (10.54)
m.=l

The equation n = 1 !s special:

Let us examine n = 2:

From this example we understand the general structure: a,.(z) is a polynomial of degree
(n- 1) in lnz:

<ln(z) "= Pn-l(luz), (10.55)

n-1
~_,(x) =L mPm-l(:r)ti'n-m+l· (10.56)
m=l

====="""'==~~-~
 Introduction to Renorma.I_ization Theory 215

Note in particular that the new infurmatiun specific tu tbe urder n is cllaracterized by
two constanta [3., whkh enter in the coefficient of In z a.nd C, which is the integration
constMt (t.o which one sbould add the ooeflicients of 71(9,) which appear in the functíon
(C!Irll· Moroover the term of highest degree in P. is entirely determined by one-loop
results, tbe next term by one Md two-loop and so on. Fina.lly tbe r)~,!..(p) is entirely
determined by the functions f3(g,) and fl(!h} and for exa.mple ['~~./.. (l,g,) Jm~ wbich is
a third function of g,.
It also !ollowB from these equations tha.t rf,.l... (p) ha.s .. limit for p = 0:

(10.57)

confirmi.ng, as expected, that tbe theory is massless. However ite derívatíve arí~2..Jôp'
· . has no rero momentum límít. It ís easy tu verify tbat no other correlation functíon bll5
a. zero momentum limit either.
We have here constructed a massless theory by scaling a mass.ive theory and shown
that the corresponding 1PI functions satisfy a homogeneous CS equation, called also
renormalization group (RG) equation. Let us show that such an equation can also be
derived directly from tbe assumption of the existence of a renormalí?.ed ma.5Sless field
theory.

10.9.2 RG equatiow in a masal.ess field theory

Renormalitatían ronditions. We haV€ •hown tha.t the renormalimd massl""" q,4 field the-
ory exist.s. If we want to directly determine renormalization consta.nts by renormalization
conditíoll.B 1 we haw to impose them at Don-exceptíonal momenta~ in particular ~'e can-
not use zero momentum except for the 2-point functíon as we ha.ve indicat.ed above. We
. therefore introduce a mass sca!e p and impooe:

r~•J (rl =o)= o, (10.58)

;;,rFI (p 2
p2) = 1' (10.59)

f'!4l[p, idJ,) 9r. (1D.60)

in which the 8, form a. set of arbitrary non-exceptíonal numerical vectors.

Tbe bare correlatíon functions in a ma.ssless theory depend only on tbe cut-o!f and
.... •···--·~··---· sínce tbe bare m8S5 parameter is fixed by imposing that tbe renorma!i?.ed
m, vanishes. The renormalimd correla.tion functíons depeoo on the arbitra.ry scale
a.nd rnomenta. They are related for A large by:

(10.61)

While in a massive theory the value of the renormalized 2-point functian

r.ero momentum depends on tbe renorma.lizatíon condítions, in a massless theory the
2-point function and the renormalízed function, independently of its normalization,
at zexo momentum ae equatíon {10.61) shaws.
Note also that when m, vanish03, the bare mass squared m 2 begins witb a term of arder
g,: The first of equatíons (10.10} then shOWB that m 2, the square of the bare mass, ís
~egative for g, small. For traditiona.l reasons, we shall ofteo denote by m 2 the coefficíent
216 lntroducti011 to Renormalúatíon Theary

of ~2 in the bare o.ction but tbe reader sbould be awa.re that this does not imply that
this coefficient is posltive.
Reno<malization group el]Wltúms. Tbe bare theory of course is completely indepen-
dent of tbe para.meter p whicb bas just beea introduced to fix the nonna.lization of the
renormalized functions:

Therefore lf we differentiate equation (10.61) with respect to ~ at A and g fixed, we find

using cbain rule:
a - a
[I' a~ + ~{y,) a9 ,
with the definition.s:

- ap.
fl(y,J = ~-'a9' I
g,A
• ij(g,J "': ~nzj .
I' g,A

A priori {la.nd ii oould also depend of the dimen.sionlesB ratio A/p, but slnce tbey can be
expressed in terms of the renorma.li2ed correlation functions themselves, they mnst baV<:
a large cut-olf limit.
Equation (10.62} is ana.logous to equa.tion (J 0.50). Tbey differ only in the definítion
of g, and a finite lield amplitude renormlilization.
Both sets of equations will be an essentie.l tool for the analysis of the la.rg:e momentum
behaViour af correla.tían functions. In addítion equatlon (1 0.62) will be used in the third
p:.rt of this work to discuss the small momentum bebaviour of m=less theories.
In the massJess theory we can also define renorma.lized carrelation functions with 4>2
iDSertions:

The rp 2 renormalizatian constant Z2 can be fixed by a. renorma.lization oondition of tbe

form:

AE, before, the (r/J 2 ~ 2 ) correlation function needs a.n additiona.l renormallzation,
ca.n be fixed by a reoormalization condition a.t non-zero momentum. ..
From equa.tion (!0.64) RG equations ca.n be daived by düferentiating with respect to ·
Jl· Introducing the RG function ih (g,):

ih (g,) = p.:I' In Zzl ,

g,A

we obtain:

n- ) O,
2'1(Jir

equatian valid exC<Jpt for n =O, I 5 2 where the r.h.s. is a function of g, a.< in
(10.45).
10.10 lntroductio!l lo Renarmaliza!ion Theory 217

10.10 Homogeneous Renormalization Group (RG) Equatimw In a Massi ....

Theory

Once the renormall:red conelation funct!ons d1·"l of the massless theory have been con-
structed, a natural question arises: since a constant sourre term for q,• a.t zero momentum
genera.tes a m1113B shift (equa.tion (10.23)}, is it posaible to express the correla.tion func-
ti.ons of a massive theory in terms of the correla.tion functions with q,2 insertions of the
massless theory?
An immediate difficulty arises: it is easy to verlfy that inaertions at rero momentum
in a. massless thoory are IR dívergent. However, because the resulting theory is massive,
this difficulty can be circumvented by first using a. non-consta.nt source for q,2 , then
performing a parti ai summa.tion of tbe 2-point function and fin..Uy ta.king the consta.nt
sóuroe limit. After summation the propaga.tor beromes ma.ssive a.nd the limit is no Jonger
IR divergeot.
Let us therefore consider the renormallied action witb a 60Ul"Ce K(:r) for renormall2ed
</1 2- insertions:

{10.68)

rorrelation function in presence of the source K(x) has the expansion:

r(n) (pi,Plo ... p,;K) =-L ~TI f dq,dq, ... dqiK (qi)K (q,} ... K(ql)
l=O
X r~!,n) (ql, q2, ... ,ql;Pt.Pl ... - ,p,.)' (10.69)

which K is the Fourier transform of the source.

Let us a.pply the differential operator D: ·

8 • 8
D "'óP-;;- +IJ(g,) ;;-, (10.70)
up ug,

equation (10.69) and UBe tbe RG equations (10.67). Noting the.t:

. ó
f dqK(q) k(q) , I,
6
{10.71)

·~.;: + {i(g,) :U. - iii (g,)- Jr.z(g,) f dq f< (q) ó;(q)] r~"> (K;p 1, ... ,p,.) o.
(10.72)
the summa.tion of tbe propagator we set:

K(x) =m;, (10.73)

equation (lO. 72) becomes:

- ( ) 2 a ] r<n) { ~
'12 g, m,âm~ r \l'lo•··,PniP,g,,m,) :0. (10.74)
218 lntrndttction to Renorma!uation Theory 10.11

We have therefore derived a new RG equation for a massive theory which differs from
the original CS equations in various r'"'pects:
(i) Correlatíon functions depend on two mass pa.ro.meters while we know that only oue
is necessa.ry. However, in this pa.rametrlzation the massless limit of correlstion functioill!
is directly ohtained by setting m, = O aod no asymptotic expansion at large mom!!Dta is
needed.
(ií) In contrsst to the CS equations, these RG equationB are homogeneous. We shall
exploit this property !ater when we sol11e the RG equatioiJS.

10.11 Covarlance of RG Functions

In the example of the massless r/>4 we have been naturally led to consider two düferent
renormalízation schemes. lt is thus necessary to understa.nd how renormalization group
functíons of different schemes are r1liated. Both theories diffet by a tedelinitiou of the
coupling constant and a linlte field amplitude renorma~W!.tion. Let us call g and g the
renormalized coupling constants in the two schemes. Comparing the two renormallzation
group equations, we obtain:
ô - ô
{J(g) ôg ={J(g) ôg'

and therefore using chain rule:

ôg •
{J(g) ôg ={J(g). (10.75)

In a rf>4 like field theory, the function {3 (g) has the expansion:

(10.76)

Expa.nding fi in terms of g:
g= g + "tl92 + o (g3) '
and w;ing equation (10.76), after a short calculation we lind:

~ (g) = /h92 + /Jaiia + O (g•) .

The ftrat two terms in the expa.nsion of the {3-function ~ universal a.nd ali others are for-
mall.y arbitrary. One ahO'IIld not conclude from thi$ result that tM phy$Ícál con.o~
derived from tM form of tM RG {3-function are also to a la']Je extent <>rbítmry. Only
change of Vll.riables g ._. g, whkh are regular mappings, are allowed. This implies in.
particular that ôgfôg in equation (10.75) must remain strictly positive. Therefore the
sign and zeros of tbe ,B-functíon are properties of the theory. Equation (1 O. 75) shOW8
also tbat if the {3-function vanishes with a finite slope the slope is scbem.-independent.
The difficulty a.rises from the fact tha.t we do not know in general whicl:t re:JJLorJna.Jliza.tion.
scl:teme leads to regular functions of the coupling constant. Our intuition is that
definitions as induced by momentum or minimal subtrsction haw the most chance
sat.isfy this crlterion.
In the same way if we c ali ( (ii) the edditional fmlte field amplitude renormalí....tion
we find: .
ii(fi) "''1 (!1) +~(!i)~ In( (fi).
 10.11 Jntroductian to RenOf'll!alization Theory 219

Since the field amplitude renormalization appears a.t order g 2 , In Ç(.9) is of order § 2 a.nd
therefore the rnodification to fi of order g3 • The ooofficient of order g 2 is universal. The
value of 'l at a zero of {}(g) is also universal.
Severol ooupling con.stants. We shall meet act.ions depending on severa! lields and
coupling constants g;. The transformation law under a cha.nge of parametrization of the
coupling space of the RG {3-funct.ions becomes then

- âg
{}; (§) = âg; {}j (g), (10.80)

where the mapping shonld satisfy det(Tii ;;r âg;/ âg1 ) lo O. Then the existence of zeros
of the [J...func::tion Í5 uniwrsal and at a. zero g•:

a.nd the eigenvalues of Ôf3i/8gj at a zero thus are para.rnetrization independent.

Bibliographlcal Notes
The method we have edopted hereto prOVE renonnalizability ofthe .p• field theory follows
the presruttation of
C.G. Callan in MefhodJ; in Field Theory, L€1> Houches 1975, R- Balian a.nd J. Zínn-
Jua.tin eds. (North-Holla.nd, Amsterdam 1976).
The idea. to combine proof of renormalizabi!ity and RG equatioos is due to
A$. Blaer and K. Young, NucL Phys. 883 (1974) 493.
. . systematic discussion of renormallzation theory in a di.Jferent language as well as
historical references can be found in
W. Zimmermann, l.edures on Eiemenlary Particies and Quantum Pieid Theory, Bron-
deis 1!170, S. Deser, M. Grtsaru and H. Pendleton eds. {MIT Press, Carnbridge 1970);
Ann. of Phys. (NY) 77 (1973) 536.
the early work see for example
_ F.J. Dyoon, Phys. Rev. 75 (1949) 1736; N.N. Bogoliubov a.nd O.S. Parasiuk, Acta Malh.
: • · 97 (1957) 227; and the textbook N.N. Bogoliubov and D.V. Shirkov, Introduction to
' · lhe Theory of Quantized F'ieúl.s, (lnterscience, New York 1959).
asymptotic behaviour of renormalized integrais was atudied in
Weinberg, Phys. Re!J. 118 (1960) 838; K. Hepp, Commun. Math. Phys. 2 (1966) 30!.
ditfenmt Mpects of tbiB problem in parametric space see ai=
.. T. Appelquist, Ann. af Phy~. {NY) 54 (1969) 27; M.C. Bergere and J.-B. Zuber, Com-
.· mun. Malh. Ploys. 35 {1974) 113; M. Bergere and Y.M.P. LILID, Commun. Math. Phys.
39 (1974) 1.
referen<:eS about RG equations a.re
· E.C.G. Stueckelberg and A. Peterman, Helv. Phys. Acta 26 (1953) 499; M. Gel!-Mann
a.nd F.E. Low, Phys. Rev. 95 (1954) 1300.
also N.N. Bogoliubov and D.V. Shirkov as quoted above and
K.G. Wilson, Phys. Rev. 179 (1969) 1499.
their modern form the RG or CS equations bave been derived independentiy by:
C.G. Callan, Phys. Re!J, D2 (1970) 1541; K. Syma.nzik, Commun. Malh. Phys. 18
(1970) 227.
220 Introducti<>n to Renormalization Theory 10.11

A pedagogical presentation has been gi~n in

S. Colemllll, Dilatationll, Erice Ledure3 1971 reprinted in Aspects of Symmetry (Caro-
bridge University Press, Ca.mbridge 1985).
Ma.ssless theoties have been discussed in
K. Sym!Ulzik, Commun. Math. Phys. 7 (1973) 34.
Hornogeneous RG equations ha~ been proposed in
S. Weinberg, Phys. Rev. DB (1973) 3497; G. 't Hooft, Nt.~cl. Phys. B61 (1973) 455;
J. Zinn-Justin, Caryese Lectu= 1973unpubllsbed, incorporated in the review a.rticle:
E. Brézin, J.C. Le Guillou and J. Zinn-Justin, Field Theoretirol Approach to Critirol
Phenomena in Phase 1hmsitions and Criticai Phenomena vol. 6, C. Damb and M.S.
Green eds. (Academic Press, New York 1976).
See also
J.C. Coltins, Nucl. Phys. BBO (1974) 341.
In the appendix we brieRy outline a diflerent very interestíng approach which combines
renormalization and renormalization group see
J. Polchinski, Nucl. Phys. B231 (1984) 269.
F ore more recent applicatíons of the method see for example
M. Bonini, M. D'Attanasio and G. Marchesini, Nucf. Phys. B409 (1993) 441, B444
(1995) 602,
and references therein.
The use of the 60lution of the classical lield equation as argument of the generatjng
functional or background lield method (see Appendix Al0.3) was suggested in:
B.S. DeWitt in Relatillity, Groups and Topology, Les Houches 1963, C. DeWitt !llld B.
DeWitt eds. (Gordon and Breach, New York 1964); Phys. Rev. 162 [1967} ll95; G. 't
Hooft, Nucl. Phys. B62 (1973) 444.
The generating functional for the S-matrix was also discussed in
B. Zumino in Theory and Phenormmology in Particle Physics, Erice 1968, A. Zichichi
ed. (Academic Press, New York 1958).

Exercises
E:urcise 10.1
Calculate rl;;!.., as de!ined in Subsection 10.9.1, in an apprwdrnation in which only the
term ofhighest degree in ln(pfm,) ls kept at each order (the leading log apptoximation).
Discuss the result.
AlO. I lntroduction to Renoro~alization Theory 221

APPENDIX 10

AlO.l Large Momentum Mode lntegration and General RG Equations

We here btiefiy describe a dírect approach to renormalízation nnd renormalization group,

closer to Wi150n 's ideas, which was first propo.OO by Pokhinskí. This approa.ch leeds
to another proof of renorma!Izabilíty. The physical motivation for such 11- method wi!l
become more apparent wben we discuss RG and criticai phenomena starting in Cbapter
25. Tbe idea is to study the cut-off dependence of bare correlation functions by int<>-
grating systematico.lly on the large momentum rnodes of the field. In such a way one
constructs a renormalizatíon group is the sp"-<'e of ali local interactions, expressing the
equívalence between a variation of the cut-olf and of the coefficients of tbe ínteraction
terms, equivalence valid only for long dist!mce, low momentum (in the cut-off scale)
physícs.

A 10.1.1 A simple ~uivo.lence

In Section 9.1 we have shown by a gaussian intagration that the two actions 5(4>):

8(4>) ! f d:rrbf>- 1 4>+ V(4>), (AJO.l)

and S(rbl,<Í'>):

S (.P 1 , q,,) = ~f dx (4> 1 6! 1 4>1 + .Po.ó.2'4>•) +V (4>1 + .P,), (Al0.2)

wíth:
{Al0.3)
generate tbe same perturbation theory.
Let us now use this equivalenre in the .Jimit in which tbe propagator 62 goes to
zero. Tben only small values of '1>2 rontribute to the partition function. El<pnnding the
interaction for '1>2 smo.ll:

V (rbt + 4>>) V (4'tl + f 6V (1/>i) I

dx bq,{:r) Ql,(x) + 2
f lj2V
dxdy l!!fJ(:r)li!fJ(y) Ql,(x).P,(y) + · · · ,

we can integrate over ç\2 to ohtain tbe leadlng otder correction.

f [Mo] exp {-(~f dx.P,f>i'rb• + V(,P + <Po)- V{<J;)]} ~1

+ 21 f d:rdy.ó.,(x,y)
ró.P(:r)
óV óV 8 V
2
ó,P(y)- 1!4>{x)ó,P(y)
]
+ ... · (AlOA)
Taking the logarithm of expression (AlOA} we can rewrlte the partition function as a
functiollal integral over a field q, (the index 1 is no longer usefnl) witb "-'1 effective action
S'(çl):

S'(</l)= ~f d:r.p6j 14J+V(4>)+! f dxdyf>,(x,y) L'irb(:;~(y) .s!~)ó:~)] + --.

(A10.5)
We ha"" thilll established tbat the a.ctions (AlO.l) and (A10.5) lead to the same pa.rtition
runction, up to a trivial renormalization.
We show below that this equivalence can be u.OO to partially integrate out tbe large
momenta in a !ield theory with a cut-off A.
 222 In!roduction to Renormalization Theory AlO.J

A1U.1.2 Lnrge momentum mode partial int"!J'Ution and RG equalions

Let us take for propagator ll a mliSSless propagator of the form:

(A10.6)

in whích the functíon C(t) ís smooth, goes to 1 for t small and decreases faster than aoy
power for t large (oee Section 9.1). We then take for ll 1 the propagator conesponding
to ao infinitesimal resce.ling of the cut-off A:

C (k 2J1. 2 (1 + cr)Z)
(A10.7)
k2
At leading order for a small ll2 is then:

(A10.8)

V{e see that the propagator ll2 h as no pole at k = O. Moreover if we choose a function
C(t) which is very dose to 1 for t sme.!l:

then ll2 is only large for k of arder A. The integration over f/>2 thus corresponds to an
integration over the large mornentum modes of the field 1/J.
The equive.lence between actions (A10.1) and (A10.5), whlch is the starting point of a
renormalization group, cao be written:

d ,
A dA V(4', A) = z
I d d [ ézv éV éV ]
d x d y D(x, y) óp(x)6rj>(y) - 6,P(x) étf>(Y) ' (Al0.9)

or aft.er Fourier transformation:

(AJO. lO)

To study the ex.lstence offixed points vre sta.rt with agíven interaction Vo(rP) ata scale Ao
and use equation (Al0.9) to calcu)ate the effective interoction V(p,ll) ata scale li«~.
A fixed point is defined by the property that V(rj>, 1.), after asuitable rescaling of t/1, goes
to alimit.
~t us call y(nJ (p,P2,•• .,p,.) the coefficients of V(,P,A) in an expansion in powers
of ,p(p), the Fourier transform of tne field. Equation (AlO.lO) can then be written in
component form (assuming translation invariaoce ):

d - (n)
A dA V
I
(pt.P2, ... ,p,)"' 2
I ddk
( 1r)d D(k}V
2
- (nH) (p 1,P2, ... ,p,.,k, -k )

- !DCPo) I:: y(l+t) (p;, ... ,PiJ ,Po) y(n-Ht) (P<Ht• ... ,Pin• -Po) lAlO.ll)
I

in whích the momentum Po is determined by momentum conservation and the set I E

{í1 , i~, ... , ít} runs over e.ll distinct subsets of {1,2, ... , n}.

======·=·~·-····
Al0.2 lntroduction to &noTTllclizalion Tht!(}ry 223

We see in these equations that even ifwe start with apure grp• ínteradion, at scale A
we obtain a general local interaction because ali functions í/(n) are coupled. However, in
the spirít of the perturbative methods used so fa.r, it is possible to solve equatíon (Al0.9)
as an expansion in the coupling constnnt g with the ansatz that the terms of V(rp, A)
quadratk and quartíc in rp are of arder g and the general term of degree 2n is of arder
n-l.
9
Carrr.!ation functions. To generate co,.reiatíon functions we have to add a source to
the interadion V(,P}:
V(</>)>-> V{4>) f dxJ(x}<fJ(x).

However, equation (Al0.5) then sbows that S'(</1) hecomes in g€11eral a complicated
functional of the source J(x). A solution to this problem js the foUowing: one take.< a
source whose Fourier transform i(k) vanísbes for k2 <:: A2 together with a propagator
1).2 which propagates only momenta such th!Lt k 2 <:: A2 . This implies that C'(t) vanishes
identlcally for I $ 1 (unfortunat.,ly such cut-olf functions are inconvenient for practical
calculations). Then j dx J(x)rp,(x) does not contribute in integral (Al0.4) and S'(if;)
S(.p) does not depend on J(x).
We note however that the RG transformation is then such that the correlatíon func-
tions corre;ponding to the action S(<f>) and S'{.P) are only identical when aU momentn
are smaller than the cut-olf. The differences between correlation functions are smooth
functiorn; of momenta and thWl decay at large distances in space faster than any power.

A10.2 Snper-renonnalizable Field Theories: The NormaJ-Ordered Product

In most of this work we discuss strictly renonnalizahle field theories. Let us however malte
a few simple comments o.bout the stiper-renormalizable case. We fust take the example
of tbe <f14 field thoory in tbree dimensions. We then examine the special propertíes of two
dimensional field thoories.
The <f;4 thoory in 9 dimensíons. AB derived in Chapter 8, the </; 4 field thoory in 3
dimensions has only three superficially dlwrgent diagraiJJB, ali contributing to the 2-
point function, which are shown in figure 10.22.

Q (") (h) (c)

F"IJ!. 10.22 The three di-gent dlagra.ms In the .,;t theory.
224 Jntroduclion to Ren01maliw.tian Theory A10.2

By edding a countecterm of tbe form:

~ [at (A)g + <:12{A)g'j 1>2 ,

we render the first two diagram.s finite.
The di3grarn (c) is also superficially divergent, however it is clear that the counterterm
whicb renders diagrarn (a) finíte, renorma.liMS also díagram (c). Therefore the couotert-
erm.s whicb renorma.lize diagrams (a) and (b) render the whole tboory finite.
Actually tbe divergence (a), whlcb rorresponds to a self-contraction of the vertex (>4 ,
CBII be elimlnated a priori by repladng the ~ (> by a nonnal-on!ered verte% : q, :
4 4

(Al0.12)

in wbicb the average {t/12 (:~;)} is calculated in a free field theory with a mass /L whicb ma.y
or may not be equa.l to the renorma.llzed m""' m:

(</>2(:~;)} P
l
(2•/
f (q2
d3q
+ /L2)A • (Al0.13)

The denomination normal ordering comes úom tbe operator la.nguage. The quantity
: t/14 (:z;) : is such that
(:<J>•(:~;) :) o
(A10.14)
{ (:t/J4 (:~;) t/J(!IJ)tfl(!t2))=0,
in whicb aga.in the averagea &e ca.lculated witb tbe a.ction S,.(ç;);

f
S~(,P) = 4 dJx [tV4J)' + ,u>.p•]. (A10.15)

Tbe averages calcul&ted with wother mass a.re then linite.

Of course we still have t.o add a counterterm for diagram {b }.
Super-renarmali.table smlar field theories in two dimenJlions. An actian S(t/J) of the
form
(Al0.16}

in which V(</l) is an arbitra.ry furrction of .P(z) (but not of it.s derivatives) is super-
renormalizable. H V ( t/J) is a polynomial, only a finite number of dia.gram.s are superficially
divergent. However it is a peculiarity of dimension 2 tba.t the field is dimensionless and
therefore the ínteractíon V(,P) may have an ínfinite sem expansion in powers oí t/J. Then,
a.lthougb the tbe01:y is super-renorma.lizable, one finds an inlinite numbet of superfidally
divergent diagrams. However ali divergences come from self-contr&Ctions of the verlex
(see figure 10.23), therefore tbe operation: V(<J>): removes aU divergences.

Fig. 10.23
Al0.2 Jntmduction to Renannplization Theory 225

To obtain an explicit expression for: V(,P): let us fust conSider the interesting example
of the íntera.ction'
(Al0.17)
The average of V(Ç) in presence of a source term ca.n then be calculated explicitly:

f [d..p(y)]exp {-f dy [ (! (\79'>}2 + 1-'292) - (J(y) + .M(x ~y))t/l(v)l}

exp[~ j dydy'J(y)A (y,y')J (y'J + ,\ J dyA (x,y)J (y) + p 2 A{x,x)](Al0.18)

in whlch C. (x, y} ls the free propagator with mass 1-'· The normal ordering operation has
to suppress the term corning from self-contraction.s whích i.s proportional to A (x,x). It
is thus clea:r that the normal ordered interact.ion is:

(Al0.19}

in whích we have used:

(Al0.20)

We then write an arbitrary interaction term as a Laplace transform:

(Al0.21)

The normal ordering is a linear operatlon. Thus:

:V (op) :=f dp(À): é«z):

We can then use the result (AlO.l!}) for tbe exponential interaction and obtain:

(Al0.22)

or more explicitly:

: V(,P) := V{QI) + I~ <;n~~ (t/12)" ( :</>) 2n] V(t/J}.

The eodstence of the Laplace transform of a given interaction is irrelevant in this argnment
since the final identity is purely algebraic.
726 lntroduction to Renorma.lization Theory Al0.3

Al0.3 Background Field Method and the S-Matrlx

Tlu; motivation. We have examined in Chapter 10 the renormalization of a fieid theory,

regarding the correlation functions of the field as fundamental pbysical objeds. However
ín some cases ali local functiori~>ls of the field whícb have a non-trivial linear part in tb
are equívalent. Two ímportant exa.mples can be given:
(i) Some madels are defined on Riemannian manifolds and the fields ,P;(x) correspond to
a system of coardinates on the manífold. In some situations only quantities intrinsic to
the manifotd are physical.
(ü) In Partide Physics, S-matríx elements are calculated from connected correlatíon
functions by taking the residues af the potes of the externa! propagator {see Section
7.3). Let us consider for slmplícity the case oi only one species of field rj>, which we
haw defined in such a way tbat it has a vanishing expectatíon va.lue. Then in terms of
mas&shell amputated correlation functions:

where -m2 is the location of the pale ín tbe variable p2 of W1 2 l(p):

w(2>(v) - _z_ (A10.23)

p'+m2
the matrix element S(nl is given by:

s(n) (p1, .. ,pn)"' Z" 12 W!::,lp. (p,,. ·. ,Pn} !.•=-~' (A10.24)

The factor ,fZ in tbis eq uation corresponds to a finite renormaliza.tion of the field sucb
t.hat the residue of the 2-point function (equation (AJ0.23)) on the physical pole p 2 =
-m2 becomes 1 (see Section 7.4).
S-m!ltrix and field representation. We now compare the S-ma.trix obtaíned from the
4> field correla.tíon functions to the S-matrix derived frorn the correlation functions of a
dífferent field q/(x) related to q,(x) by:

q/{x) (A10.25)

We assume, when necessary, that the theory bas been regularized in such a way tha.t the
new <:orrelatíon functions exist.
Using rela.tion (A10.25) we ca.n reexpress the 4>' correlation Iunctioil& in terms of the
tjJ <;orrelation functíons.
The expansion of tbe q,' propa.gator shuws immedia.tely that tbe </; and l/l1 propagators
have poles at the same position (see ligure 10.24).

tb'---.--4>'-4>_._1>+ ~ + ~
+ ..
Al0.3 lntroduction to Renomudization Theory 227

Fig. 10.25

The contributions tn n-poínt fum::tions which have poles on the externai !ines then
hav.o the form shown in figure 10.25.
In the mass sheU limit (p~ =- -m 7 ), th01 ,P' and r/J <:orrels.tion fun<:tions become propor·
tional. The S-matrix elerne!lts are idelltical. Here again ali fields relate<! by transforma-
tion (Al0.23) are equívalent.
In such situations not ali parameters of the theory are physical. For example the field
amplitude renorrnalization is obviously unphysical. The same physical theory may have
renormalizable and non-nmormalízahle realizations.
Using the hackground field method <me can avoid the calculation of unphysical quan-
tities (however to prove renormalizabillty the study of correlation functions ca.nnot be
avoided).
The ba.ckground field method. Let us considera field theory with an a.ction S(tf):

ew(J) =f ldr/J]e-S!~)+J<I (Al0.26)

Let us introduce immediately the Legendre transform of W(J):

.,-r(.,)+J, f ldtf]e-S!*)+J.O. (Al0.27)

Using:
J(:t)
6r
DI!'( X)'
we can write equation (A 10.27):

.-r(,c) =f !dr/J]exp [-S(r/J) +f dx(r/J(x)- ,o(z)) .~:~J, (Al0.28)

or equivalently translating </J(x):

e-l'(op) =f (dr/J]exp {-S(</1+ rp) +f dxÇ>(x) 6!~J (A10.29)

Let us now assume tha.t the equa.tion:

o, (A10.30)

l.as a non·trivial rolution 'Pd:t) which a.t lea.ding order in perturbation thoory is a. solution
\1'~ \ x) oí the dassical equatlon of motion:
0

....!!__ [ (O)] - Q (Al0.31)

6'!'(x) 'Pc - ·
228 Introduclion to Renormalí.zation Throry A10.3

Then equatk>n (A 10.29) becomes:

(AI0.32)

The quantity r (I" c} is obviously independent of the representation of the field </>, and
contains therefore only physical informations in the seru;e delined at the beginning of this
section. Let us íntroduce renormalized qua.ntlties:

r r ('Po) =-In I [d<f>j exp [-So (</> + 'Pc) + oounterterms),

in which S0 (</>) is the tree levei a.ction. The so!ution 'Pc of:

is expanded around the solution of rp1°l (x) of

In real time, as shown in Subsection 7.3.3, the background field method yields the 8-
matríx. We shall provide other examples of such calculations in the coming chapters.
11 DIMENSIONAL REGULARIZATION, MINIMAL SUBTRACTION:
CALCULATION OF RG FUNCTIONS

Dimtmsiona.l regularization is one of the most powerful regulo.rization teclmiques, a.nd

the most convenient for practical perturbative ca.lcul..tiODB, when applica.ble. We have
discussed its impleme.ntntion in Section 9.3. We wRDt to here show how it leads naturally
to the concept of renormalization by minima.l subtraction. We first discuss tbe form of
renormalhation constnnts and renormalizatíon group functions iJ(g), 'l(g), fl>(g) wben
dimeDBiona.l regulariza.tion is used, nnd then show that tbis form is specia.lly símple in
the minirnal su btraction scheme. We perform explidt ca.lculations at twu-loop order in
tbe t/J 4 field theory. The ca.lculations in this theory are sirnple a.nd usefuJ for tbe tbeory
of Criticai Phenomena. Moreover the results Gllll be compared with those obta.ined in
previous chapters. We then generalize to a severa! component four field interaction.
Fina.lly, to give an exnmple of a theory involving fermioos, w ca.lcuhte tbe RG functions
at one-loop order in a theory containing fermions ínteracting tbrough a Yukaw...like
interac:tion with a scalar hoson: the linearised Grosa--Neveu model.
11.1 Bare and Renormalized Actions: Renormalization Group (RG) Func-
tions

We again discusa the example of the r/>4 6eld theory with the be:re action:

S(~) =f d":.c [~ (a~t/Jo) 2 + ~~~ + ;ijgorf>g]. (11.1)

When dimensional regu!a.rization is used, tbe only dimensional parameters are the bare
parameters {m 0 , go}: In d = 4- e dimension the ba.re couplíng constant hBS a dimension
[go]
(9o]~d-4lt/JoJ=d-2(d-2)=4 d=e. (11.2)
It is then convenierrt to introduce, besides tbe renmmallzed mBSll m, a dirneru~ionless
renormalized couplíng constant g, in such a way that the renormallzed action cen be
wrltten:

S,(~) =f dd:.c {iZ(g) (8~4>) 2 + ~m2 Zm(9)t/J2 + ;ijm•-•uzg(g)t/>4]. (1L3)

· · Indeed w:ith this parametrization the renormali.tation constanta Z(g), Z,.(g) and Z9 (g),
are dimell5ionless, and thus depend on the only dimensionless parameter availnble, the
coupling coiiBtRDt g. In particular the rnass is multíplicatively renorma.lizable.
!rmce the renormalized action is obtained from the bare action by the field ;rescaling
=
rf>o rf>./Z, the bare and renormalized parameters are ra.lated by:
m~ =m2 Zm(9)/Z(g), (11.4)
=
go gm•-dz9(g) /Z 2(g). (11.5)

Remember that in the dimensio~al regular'.zntion scheme renonnalization corultants de-

pend on all additiona.l hidden parameter, the dimension d.
 Dimensiona.l Regularízatían, Minima.l Subtmctian lU
230

From the relation (11.4) we can now cslcula.te the RG 1>-function, tbe coefficient of tbe
CS equ ation. Setting
(11.6}

and dilferentiating equation (11.5) w!th respect tQ m a.t go fured, we find (equa.tion
(10.26}):
O= (4 -d)G(g)+ tJ(g)
a G(g),
89
or:
(11.7)

Using the notation of Section 10.4, we ca.ll Z~(g) the renormalization constant associated
with tbe renormaliz.ation of ifJ 2 • From equa.tions (10.27 ,10.28) we then derive:

d (11.8)
1/(Y) "'iJ(g)dglnZ(g),
d
'72 (g) ~ iJ(g)dg In [Z2(g) I Z(g) l, (lU)

lllld finally from equations (11.4) and (10.29):

(11.10)

Thís equation shows that Zm/ Z2 is a finite function of g, =d tberefure Z.., is not a new
renormalízation constant.
Tbe renormalizability of the t/>4 field theory iD 4 dimension.s implies tbat the renormal-
ized con-elation functions and therefore also the RG functions iJ(g), l)(g), 17 2 (g) and u(g)
have a finíte limit when the devia.tion

<=4 d

from tbe dimension 4 goes to O. Since

tbe function {3(g) can thus be written:

in which ali the functloos {3., (<) have regulRr Taylor series expa.nsion atE: O.

{J.. (e) {J.. (O) +E{3~ (0) + · ·· ·

11.2 Dimensional Reg'UÚI.roation, Mínima/ Subtroction 231

11.2 Dimensional Regularization: Tbe Form of Renormalization

Constants

ConverBely let us now derive the form of G(g) from tbe knowledge of fJ(g ):

G'(g)
g G(g)

lf we expand the r.h.s. in powers of g, we observe that, at a fixed order in g, the most
singular terminE comes from the term of arder g 2 in fJ(g);

Integratíng the expansíon term by term we get:

G(g) = g +]; g" [( 1-'12(0))"-l + less singular terms ]

-E-

The coefficient G,.(E} of the expansion of G(g) in powers of g;

00

G(g) g+ Lg"G,.(E),

has thus a Laurent series expansion in<!, for ~ small, of the form:

13;'- 1 (0) G,. 2-n

--;,;=;-+ .~- 2 + .. ·+G,.p+G,., 1 e+··· ·

liniteness of f!(g) and Tf2(9) leadE to simllar condusions for Z(g) and Z2(g) which
be written:
oo (n)( )
Z(g) 1+~~ + regular terms in E,
L E"
I

"' (n) )
Z2 (g) : 1 + L l
02
E"
(g + regular tenns in E,

o~n) O(g").
cooclude tbat at order L in the loopwise expansion the divergent part of r(cp) is
ool'\'llll,IIJÍI!.I of degree L in 1/E.
 Dimensional Reyularization, Mínimal Subtroction 11.3
232

11.3 Minimal Subtraction Scheme

Although the rninimal subtraction idea can he used ín a.ny regula.ri~atíon scheme (see for
example equation (8.11)), it is specially useful, in particular for criticai phenomena, in
dimensional regularizatíon. Renormalization constant.s are determined in the following
way: insteed of imposing renorma.lization conditions to divergent lPI correlation func-
tions, one just subtracts, at each order in the loop expansion, the singular part of Laurent
expansion In f;. Let U8 call rr·(<p) the dive<gent pa.rt of the genera.ting functional of
proper vertíces, renormalized up to L 1 loops:

(11.12)

We add as a counterterm to the action just -rt• ('P) as defined above.

Examp!e. Let us again consider the ,P4 field throry. At one-loop <>rder only two
diagra.ms oontributing to r( 2 l and f( 4J are divergent:

_1_
{27t)d
I r_<!_:'L
+m'
m._, __ 1_r (1- ~) - -~ +o (1)
(4>r)dj2 2 - 87T2€ '
(11.13)

_1_1
(21f)d (p2
ddp
+ m2J'
=-1-
s,•.., +
01
( ).
(ll.l4)

The one.-loop divergent part of r ('P) 1s tben:

__1_
32"2"
[m'g I \?2(:r)d":r + .!.g2m•
4
I \?•(x)ddx] • (11.15)

Thís expression should be compared to equations (A9.l7) a.nd (10.8). The functions Z,
Zm, Z9 at this order ate then:
(11.16)

(11.17)

(1Ll8)

The calcule.tion of the function {1/J',Prp) at one-loop also leads to:

z, (11.19)

The RG functions are then:

3 (11.20)
{J(g) '""-Eg + 1611"292 + 0 (g3)'
(1L21)
'1(9) O(g"),
112(9) "" -~
16?t·
+o (!l) . (11.22)
11.3 Dimensional Regulamation, Minimal Subtroction 233

We note that at this arder Z., = Z,.

Since the mirüma.l subtraction scheme eliminates
any possible finíte renorma.lization and Z2/ Z., is firute, thís relation rema.íns true to ali
orders.
Let us explore more generr.Uy the consequences of thiB choice for the RG functions.
Writing for ínstance:

G(g)- g +f G;~g),
I
G,.(g) =O (gn+l), {11.23)

we can calcula te fJ(g ):

Since G~(g) ia of order g", we can expand the denominator:

Tbis expression ca.n be rewritten:

l)(g) = -t::g- G1 (g) + gG~ (g) + ~ b,.(g) •

~c"
I

The finiteness of /1(9) then implies:

fJ(g) = -eg + gG~(g)- G1(9), (11.24)

b.,(g) O 'Vn. (11.25)

The functions fJ(g) and G,.(g), n ';;> 2, are uniquely determined by the function G1(9),
i.e. the coeffident.s of 1/E in the divergentes.
Similar a.rguments apply for the other RG functioos and renormalization constants.
In the expansion of equation (11.12) the whole new L-loop lnformation about diwr-
.genc"" ís conta.iued in l'L,J (rp). AU other functions are determíned by the counterterrns
of previous orders.
In additíon the RG functioll!! fJ(g), q(g) and '1/2(9) have a. very simple dependente in e .
.For fJ(g) it ís given by equation (11.24). Let us ca.lculate 1) Md "12· The renormalization
CO!IEtants Z and Z2 now have the f0f111;

Z(g) = 1+ L= a (n){}9 ,
E" é•l(g) =O (g"+ 1 ), {1L2ô)
l

(11.27)

relation (11.8) and the forro (11.24) of /J(g) we obtain:

q(g) = 1-óg +g~{g) G,(g)J U1ga< l(g)+ O(~)].

1
 Dimensirmal Reg1daril.ation, Minimal Subtroctíon 11.4
234

Since ry(g) has a finlte limit it is given by:

(11.28)

Simihaly for f/2(9) we find:

'12(9)

Finally the explicit dependence of the renormalization constMts in E: can be obtained by

calculating them frorn fJ, ry a.nd '72· For exa.mple:

G(g) ""9 exp (-e [ [-E:9' ~ b (9') +e~'] dg')'

in which we ha.ve set
fJ(g,e) ~ -<g + b(g).

ll.4 The Massles.s Theory

4
Dimensional regulariza.tion eM also be used to define the ~IU>BSiess <f1 field tbeory. Unlike,
hawever, the ma.ssive tbeory, due to small momentum (IR) divergences tbe regularized
theory only exlsts in an infinitesimal neigbbourhood crf the dimemion four. Thls prob-
lem will be discussed at length in the Chapters 25--26 devoted to criticai phenomena.
Then since the mJ>SSJess theory ll; tenorma.lizable, the mínima.! subtraction scheme is also
applicable.
The hare action cnn be written

Note tbe absence of a bare ma.ss term. Indeed if tbe propagator is massless no mass can
be generated because there no dimensional parameter, besides the coupling constant: All
diagra.ms contributing to the two-point function have a power-law behaviour given by
simple dimensional considerations:

f'(2)(p) = p2 +L Cn(õ)l/!:p2-n<,
..~2

(remember that E ll; infinitesimal).

To define s renormalized theory it is necessary to íntroduce a mass scale IJ which
take care of the dímension of the <f14 coupling consta.nt:

(11.33)

'Ib the action corresponds a set oi relations between bare and renormalized conelation:
functions: ·
zn12(g)r~n) (p,;iJ,g) ""r(n) (p;;9o)' (11.34)'
with:
(U.35}
11.5 235

DiHerentiation of equation (ll.34) with respect to I' at g0 fixed, yields tb.e RG equations
which express that rorrelation functions depend on I' and g only through the combination
!ltJ(p, g):
f) f)
[ I' 8jJ + i3 (g) 8g (11.36)

with for i3 and 71 expressions simil!ll' to equations (11.7,11.8):

fj(g) ""-(4 d)(dlnG[g))-l (11.37)

dg '

As we ha.ve diseussed in Section 10.10, it is then possible to define a tn,..,:;ive theor:y by

addíng to the renorrna.lized action a mass term of the fonn

Note tha.t since in the mínima.! subtraction scheme the renorma.lization constants are
uníquely defined and índependent on tbe r a tio m/ p one concludes by setting m = p that
thEJ renormaliza.tion constantE in the m"""less theor:y and the massive tbeor:y of Section
11.1 are the same.

11.5 A Few Technical Remarks

Feynman parometers. Before presentíng explicit ca.lculations of Feynma.n diagrams a.t

two.loop order, let ua derive a sirnple identity wbich is often useful. One starts from:

(11.38)

(11.39)

t; B'Ui 1 (11.40)

"
u.;2;0,
L"'
i=l
1'

n n
I1 <"'·>-o'= r r"t + <>2 + ... + a,.J II (r (a•>>_,
í=l

(11.41)
236 Dime11Sional Regulari.wtion, Minimal Subtroction 11.6

mt,
If the quantities n~o .. -,an correspond to propagators, a.; =" p~ + the integral over
momenta can then be explicitly performed. At one-loup arder only one integral is needed:

_1_1 d"p - 1 f""tndtjdd .,-t..'-t_ r(n+1-d/2)

(2~r)d (p2 + 1)n+l - (21ll r(n + 1) lo p - (4lr)dl2 r(n + l).
(11.42)
Note that representatlon (9.28) Jelllis to au expresaion similar to (11.41). When p(t)
ís of the fonn .,-•=', tbe argument of the exponentíal ;, linear in the varlabléS f; and
after the change of variables (11.40), the integral over the bomogenrous variable s can
be performed.
In what fo!lows we also need the integral

(11.43)

The modif!eà mínima! subtmction 8<J.eme. In the calculRtion of Feynm= diagtams, a

foctor (Nd) :
Nd area of the sphere Sd-1 2 {11.4<1)
(z,)d (4" )d12 r (d/2)'
L being tbe number of loops of the diagram, is generated naturaHy. It ;, tberefore
convenient to rescale the loopwise expansion para.meter to suppress thiB factor. In the
4>4 field theory for example two choices are possible. Either one multiplies eoch Feynman
diagram by a factor (N4/Nd/'·, then the normalizations of field and coupling ronstant in
4 dimensions are only modified by the change in the renormalization constants. Or une
oompletely absorbs the factor Nd in a coupling constant and field redelinition.

11.6 Two-Loop Calculation of Renormali:r.atlon Constants and RG Functions

in tbe tP• Field Theory

As on exercise, we now explicit!y calculate the tw<rloop renormaliz.ation constants in

the 4;4 lield theory, and in the minimal subtraction scheme. We work with the massive
theory though the calculations are somewhat easier in the massless theory. Note that
for hígher order calculatíons even more sophisticated methods can be used, where the
masses in diagtams are chosen iB the fo!lowing way: Only diagtarns to which all dívergent
subdiagrams have been subtracted are cc>nsidered. The remainíng global divergem:e h;
then independent of momenta and interna! rnasses. One sets to zero as many masses and
momenta as posslble, as long as 011<! does not enoounter IR dívergences, ronsístently in
the diagram and the suhtraeted subdiagrams.

11.6.1 The diagmms

In Section 5.3 we bave already given the diagtams contributing to the 2-point function
at thh; arder. A short ca.lculation gives also r!•l,

{11.45)

(1!.46)
!1.6 Dimensional Reuulariwtion, Minimal Subtmction 237

Q (a)
e (c)
Fíg. 11.1 Díagrmn.s contributing to r( 2 ).

(d) (e)
Fig. 11.2 Diagrarns contrlbuting to r(•l.

To determine all renormalization constants we tbus need tbe diagrams displayed in figure;
ll.l and 11.2.
The diagram (g) contributing to r!•l corresponds to tbe insertion of tbe one-loop ri 2 )
diagram into the one-loop r<•l díagram. When the one-loop díagrams are renom~alízed,
it is a.utomatica.lly fin.ite. At this arder only the diagra.w (c) ba.s a diV<>rgeoce which is
momentum dependent:

(c) r;(')( ) - _1_

p,m -
(27r)
u
f (qr + m')(q?
0
d qtd"q2
+ m2 J [ (p + QJ + q,) > + m.2 J
(11.47)

Th renormalize the 2-poínt function we need (a), (h), E(O,m) and ôE(p, m)/8p2 for
p =O. Let us first set: ·

(a) (11.48)

(11.49)

{11.50)

The díagra.m (b) as well as the d!agrams (d) ... (f) at zero momentum can be expressed
in terms of the integrais 1t a.nd !, hy the identíties:

(d); (27r)d
1 f (.r+ dàq 8
mZ)2 = - ôm' (a.) (1 - d/2) md-4ft,
(11.51)
(b) = {a)(d) (1 - d/2) m2d- 6 If,

(e) a {(d)) 2 (1 df2) 2 mu-sll,

finally:

(f)
238 Dimensional Regularization, Minimal Subtroction 11.6

The inteyrnl ]j. The integral is a specíal example of (11.42). As expla.ined in Section
1u;, we extract a factor Nd (equation (11.44)) for each loop. Setting then:

d=4-e,

-find
(11.53)

Calculat'ion of I2. 'Ib deriw an expresslon hoth for I2 and I 3 we consider E! 2 l(p, m):

E!') (p, m 1)
1 f ddq]ddq,
(11.54)
(21T) d
2
(q~ + 1) (q~ + 1) [ (p + q, + q,) 2 + 1]

Using the method expllrined in Section 9.3 we write:

(11.55)

with:

A (g,; 92,P) = (t, + t3) q~ + (t2 + L3) qª + 2taql <h+ 2t,p (q1 + q,) + tsp2 + (t 1 + !2 + L3).
(11.56)
Integrating over q1 and q2 we find (equation (9.28)}:

(11.57)

in whlch A is the determinant a.ssociated with the quadratic form:

(11.58)

We only need the two fust terms of the expansion in powers of p2. Let us first calculate
l2 which is more complicated:

I,= r:!'l(o 1) = _1_

' (41T)d •
J."" dlt dt2 dta e-<••+••+<•l.
[),d/2
(11.59)

As expla.ined in Sectíon 11.5""' now cbange variables, nsttíng:

tl stu,
t,=st(l u), (1L60)
{ !3 = s(1-t).

The íntegr al hecomes

(11.61)
11.6 Dimemwnal Regularuation, Minimal Subtroction 239

The integral """' s ca.n be performed. The rema.ining integra.! J, from which we wa.nt to
extract the divergent a.nd the finite parts, has dívergences corning from t O and t = 1,
u=Oandu=l.
J 1' du dt t'-412 ]1 - t +tu (1 - u)]-•1•. {11.62)

'Ib separate dive:rgences, we subtract and add 1 to each factor in the product:

J = 1' dudt { 1+ (tl-d/2 -1) + [11- t + tu{l-u)j-d;o 1]

+(tl-d/ 2 1) [<1-t+tu(l-u))-4/ 2 -1]}. (11.63)

'Ibe last tenn is finite and we ca.n set d = 4. The otbets ca.n be integrated immedíately
over t:

J u(1

{11.64)

'Ibe last integra.! in the expression yields:

A 1ast piece hos to he worked out:

u)]l-d/2 1}
1] +O(e).

lot du {
1-u(l-u) [u(l-u)]
1-d 2
I -I
}
"'lat du[u(1 (11.65)

now use the identity (11.43) and collect ali contributions:

1=2 (11.66)
€

2
1 (€/2) 4 (}
rc~l = 'i+o € •
(11.67)

obta.in the final result

{11.68)
 Dimensianal Regulllrúation, Minimal Subtmction 11.6

Tbio leads for I, to:

(11.69)

Using:
_(1-e/'il +O(e:)
r 2 (d/2) r (3-d) E(1- !) '
{11.70)

we deduce the singular pa.rt of integral l2:

!2"' -N1 ~ 2 (1+ ~) + 0(1). (11.71)

The jntegrnll3. The ca.lculation of Js relies on tbe same method but is simpler. Ia is
the coefficíent of p 2 fn expresslon (11.57):

I =~Ei 2 l( )1 =--~-1""dt,dt,dt3ttt2t3.,-l••+••+<•l
3 ar p,
1 l>=• (4"1" o .o.l+.df2 .

After the change of val'iables (11.60) and the integration over s we lind:

h~--1-I'(e) tdtdu (1 t)u(1-u) s· (11.72)

(41r)d Jo \1-t+ut(l-u)J

A short calcula.tion then leeds to:

(11.13)

The 4 -point function at one-loop c>rder. As a final exercise let us also calcula.te ""J'licitly
the finite p!ll't of the one-loop 4-poínt function which is given by the integra.ll4 (p):

I4(p) =
m•-d
--d
J l ddq
. (11.74)
(27<) (q2+m') (p+d+m2}

The calculation ha.s been already performed in Sectíon 9.3 for the casem O. The511.1Ile
method leads to:
] l~2
/.
2 ] d/2-2
14 (p) = Nã~ (1- df2) sin :d/2 0
dt t (1 t) + I
(11.75)

Expanding now fur e sma.ll we obta.in:

1 1-
14 (p)=Nd [ &-
2
11
2 0
1
dtln ( m
2
p 2 t(l-t)+l ) +0{&).
] (11.76) f·

Thking the finite pa.rt and performlng explicitly the la.st integral we find:

(
I.(p) Nd)
e •-"
= _..]__ [~Jp' +4.m
s"• 2 p 2
2
In ...JíTt-;r;:;j2+ H
.,Jr +4m2 - v?
~12 ·
ll.6 241

11.6.!! Renormalízation constants and RG /undWna

We now introduce the renormalization conste.nts, i.e. replace in expressions (11.45) a.nd
(11.46) the propagator (p2 + m 2 ) -• by:

and tbe coupling constant g by:

g>->gZg.

We then expand the renormolízation oonste.nts in powers of the roupling constant g and
adjust the coeflidents to cancel all divergences.
Renormalizatian constants. Integral / 3 immediately yields Z:

(11.78)

Expreosíng that rl 4 l is liníte, we obtain Z9 :

Z9 3g
I+ Na2e + Nd2 ( «92 - 4:f
3) g
2
+O ( g") · (11.79)

· Flnally the propagator at zero momentum determines Z 2 :

(IL80)

RG functions. Equatlons (11.7-11.9) tben yield the RG functions:

(11.81)

(11.82)

(11.83)

(ll.84)

Field amplitude renormalization: Pasitívíty of the RG functíon. Note that for g small,
in the perturbative domain, the field renormalization (11. 78) satisfies Z < 1. This
general property in unltary theories implied by the "Spectral representation o[ the
functlon (see Section 7.4). It implíes 'l(g) >O for g small.
242 Dím.-nsional Regularization, Minimnl Subtmction 11.7

11.7 Generalization to Severa] Component Fields

We now generalize previous calculations to a renonnalrned action of the form:

(11.85)

We here consider only the massless theory a.nd cal! 1t the renorrnalíza.tion scale. We
agaln use dimensional regularizatkm and minimal subtraction. This explwns the absence
of mass counterterms. We call g;;kl the renormalized coupliog oonstant, which is a
symmetríc tenoor in its four índices. The qua.ntíty G;;kl which indudes coupling constant
renormaliza.tíon has the fonn:
(11.86)
To calculate the generalization of the renormalization constant Z;, we use the renormal-
izatjon of the insertíon of tbe operator !1/>;(x)q};(x).
The ca.lculation of the renonna.lli.atíon constants reduces to the caloulation of weight
fr.ct.oT5, the values of the divergent parts of tbe diagrams being the same as in Sectlon
11.6. .

11. 7. I Diagroms and reno1'171<11izatwn cvnstants

The 2-point function has at two-loop arder the expansion:

r)? Z;;p2 - ~G;k!~G,klm (c') + O(c?). (11.87}

The 4-point function is at the same arder:

rj;l1 G;;>r ~ (G;;mnG,.,..,(d') + 2 terms) +i (G;;,.,.G~,.,.,G,.~r (e')+ 2 terms)

+ ~ cc,,~,c ...... c ..,.,r (f') + 5 terma) + O(G 4 ) ' (11.88)

The edditíonal terms restare the permutation symmetry of the 4-point funeti.on in its
four índices. The quantities (c'), ... , (f') are, in the massless thoory, the corresponding
of the diagrama of figures 11.1,2 (this acoounts for the primes}.
4
Let us finally calcu!a.te the (rp;(x)t/l;(:t)q}• (!ll) <l>r (y2)) correlation function ru::i:
{1 2)
riÚI = 21 (óuó;1 + 6;r6;~)- 2Gijkl
1 1
(d) + 4G,,,.,.G,.,.tl (e
1 ')

+ ~ (G~>mnG;rmn + GJkmnGilmn) (f') +O (G3). (11.89)

As we h ave argued in Sectíon 11.4, the divergent parts of the massive diagrams of Sectíon
11.6 and the massless tbeory are the same. Expanding tbe renorma.lli.ation constantE in
powers of g;;•r. and using tbe values of the divergent parts of diagrams (c), (d) and (f)
of figures 11.1,2, we obtain:
Nd NJ
9Ukl + 2€ {g;jm,.gmnkl + 2 terrns) + 4E2 (g,;mnf!mnJH19POkl + 2 terms)

+ :~ (1- D (g;;rnn!Jmpqkg•pql + 5 terms) +O (g 4 ), (11.90)

N2
tí;; ~9iklm9iklm +O (g") · (11.91)
11.7 Dimensional Regtdarization, A(inimal SubtroclÍQn 24&

From equations (11.90,11.91) we deduce the expansjon of t.he bare coupling constant
which we paratnetrize as 1J'1<ikl:

(11.92)

Nd N2
)ijkl 9ijkl + 2~ {g;;,."9mnld + 2 terms) + <k~ (g;;mn9mnpq9pqkl + 2 tertns)
+ <k 2
E) (g;;mn9mpqk9npql + 5 ter=)
NJ ( 1 - 2
N2
+ ~ (9ijkm9mnpq9npq! + 3 terms) +O (g4). (11.93)

Calling zgl,
the renonnaJization consta.nt of the operator ~,P; (x) </>;{x), we derive frorn
equatíon tll.89)'

!>l
Z;j,kl = 2I (li;16;1 + ~üli;•) + Nd
2
NJ
E Yijkl + ~ 2 9ijmnf}mnkl
N2 (
+ &~ E)
1 - 2 (9tkmn9jlmn + 9jlm>n9ílmn} + O (g ).
3 (11.94)

we n~ the matrix (JJJ, whiclt expresses the renormalized operator in terms of

bare fields:
d,~.t. z!•J (z-t/2) (z-•1•) .
tJ;mn 1'nk n.!
(li.%)

(2)
G;,kl = z,j,kl
(2) !!L ,
+ 192~: (Yimnv9kmnp6il + 3 terms) +o (g ) .
3
(11.9{))

additíonaJ tenns restare the symmetry of exchange ( i - j) and (k +-> !) and'of the
($Írs(ij) - (kl).

7.1l Renonnalization gmup equatíons

us first derive the RC equations for a. multicomponent massless fie!d theory. The
be~ bare and renormalízed oorrelation functíons now take8 t.he fonn:

(p.,u,!') = (z'' 2
), .
1]1
(z''') . .... (z't'),.. ,... r~~.;. ...;ft (P.,Yo,A),
•:~J:a
(11.97)

. B stands for baxe, g fur Yíikl and 90 for tJ.<"{ijkl·

we diJfereotíate equation (11.97) with respect to I' at 9o and A fixed, we obtain the
equatfon:

(11.98}

(11.99)
 Dimensional Regalari.zation, Mínimal Subtrnction 11.7
244

(11.100)

Similarly to renormalize conelation functions w:íth ~~;(:r)~j(:r) insertions we have to

multiply each insertion by the matrix cg:.,. This leads to the RG equation:

(11.101)

with the definition:

(11.102)

Renormalization group functions. From the expressions (11.91-11.96} we detive the

expanslon of the RG functions at two-loop arder:

Nd
·'""
\~=-,NJ
f3i;kl ~€9ijkl +2 (g;lm~9~nkl + 2 terms1..i;J4 (g;;mn9mp~;kilnpq! + 5 terms)
+ ~i (!h/km9mnpq!Jnpq! + 3 terms) +O (g4 ) , (11.103)

'flij = NJ
?Ail<k!m9jklm +O (9 3) , (1U04)

(2)
'1ij.ki
Nd
= -29ijk!
NJ
+S (IJ;kmniljlmn + 9jkmn9i!mn)

NJ (9;mnpilkmnp6J! + 3 terms) +O (9 3) . (11.105)

96
These expressiom will be used in Chapter 26 in their general form. Let us here spreialize -"
to the ( <J:12 ) 2 fi.eld theory with O( N) symmetry. We then have to substituteo :

g;;J<~ = 3g (ó;;6nl + 6;•6;1 + 6u.ê1•). (11.106)

A short calculatíon Jeads to:

~(g) = Nd{3(g) -<g + ~ (N +8){/- (3N1~ 14) g' +O (!í•), (11.107)

+ 2) 9+
( .) - (N72
'19-
-> O(-")
9'
(ll.lOBh

in the notation of equtr.tion (ll.B4).

The matrix 11li.~ becomes a linear combinatíon of the identity matrix I and the pro--,.
1
jector P: ·
.1
lij,kl "' 2 (li;kÓjt + 6.1o;k),
P··
..,,."I= ..!_6;-Ókl
N '
 lUl Dimensional RRyularuation, Mimmal Subtrnction 245

One finds:
. -~
1
1)( ) - -(NP + 21) ~ + I(N + 10)1 +4NPI ~ + 0(§3 ). {11.110)

The eígenvalue corresponding to a mass Íll5ertion r/>2 is given by the eigenvalue +1 oí P:

(11.111)

The otheJ" eigenvalue related to the eigenvalue O of P corresponds to the insertion of a

mass term hrealcing the O ( N) synnnetry:

~,(2)
., -ãg + --7-2-g
(N + 10) _2 O (_3)
+ g • (11.112)

11.1! Renormalízatlon and RG Functions at One-Loop in a Theory with Bosons

and Fermions

We now consider a simple field theory involving one acalar field r/>(:z:) a.nd N massleas
Dírac fermion fields .P'(x) and ~'(:z:), coupled through a Yukawa type t/ijJ,p interaction.
The actíon is

(I 1.113)

h has an obvious U (N} symmetry and in addition a discrete chiral symmetry:

.,S(:z:)--,P(:z:), (11.114)

which forbids a fermion ma.ss term and odd powers of q, as r!> self-interaction. The physics
of this model, spontaneous chlral symmetry breaking and fermion mass generation, will
be discussed in Appendix A30.
We haVll already shown tha.t such a field theory is renormall.,.able in 4 dimensions by
pawer counting. 'Ib calculate renormalization constants we use dímelllSional regu!ariza-
Uon and minimal subtraction. We render the renormallzed coupling conlltiúlts dfmen-
sionless in d dimensions by setting:

(11.115)

:in which >.o; represents symbo!ically the bare coupling constants {g~, >. 0} and À,: the
renormalized coupling constants (g2,>.).
·, . We ha.ve given in Section 6.4 (equation (6.56)) the IPI functional r (V', .p, i/J) at on~
loop order for an action which contains tbe action (II-113), N'"" 1, as a special example .
. Here we ohtaín:

r, loop -Ntrln [1 + ur'l"(:r)) + 1trln (1 + !-' (-t:.. + m2)- 1 1"2(x)

1
+2g 2 (-tl.+m2 )- i/J(:z:) (~+YI"(:r:W 1 tJ>(:~:)). (11.116)
246 Dimensional Rtgtolori:ation, Minímal Snbtmction 11.8

Expa:nding in powers of <P, t/1 a:nd ij,, ""' get a.U lPI correlation functions at one-loop
order. We consider anly the divergent functions and omit ali contributions involving
on)y the q,• vertex wbich bave a.lrea.dy been evaluated in Section 11.3:

Figure 11.3 displays the corresponding Feynma.n diagnllns.

Pt ..

P<···
·D
.... ~.

•••• ,.

Fig. 11.3 Boson fennion diagrMJS (the fermion:; and bosons conespond to continuous and
dolted lin~ r~ectively).

11.8.1 Explicit rolculatio!IJl af lhe 2-point fu.nctiom;

AP. an exerclse let us first ca.Iculate completely tbe contríbutions to the two.-polnt func--
tions.
FennionB. We have to compute the integralft (p):

Rotation ínvariance implie;; that J 1 (p) is a linear combínation of tbe form A(p) +iB(p)p.
To obtain A and B we can take the trace of the integrand multiplied suocessively by
and p. Applying the roles axplained in Appendix A5.8 we immediately soo that A =
and thus no fermion rn85$ is generated.
 Dimensional Rey}Uiaritation, ,Yinimal Sv.btracticn 247

Using the met.hod of Feynman parameters as expla.ined in Sectíon 11.5 we transform

(11.121) into:

J,(p) = _1_ f' dxj ddqitf (11.l22)

(21f)" lo (.r((p-q) 2 +m2 }+(1 x)q2 J2 '

We then shift q, q- xp ,_. q, and íntegra.te over q. We find

_ ·/<2 d/2)
1I (p) - . (41f)df2 lof
1
d ,.,2-1 rm• + (l
XX
(11.123}

Expa.nding a.t first order in E 4 d and integra.ting over x we finally obta.in

(11.124)

Bosrm.s. We have to ca.lcula.te the integral h(p):

(11.125)

Iz(P) (11.126)

q> +p·q =t [q'+'(p+q)•]- ~p·,

can decompose J2(p) into a sum of simple integrais:

(11.127)

first integral vanishes in dimensional ~Ja.ri>.a.tion and the second is the massless
of a.n integral already calcnla.ted in Section 11.6. We find:

m(;i)]. (11.l28)

Other dioeryerw;.s. Refl01'malízation group funclions

one-loop diagrlliiiS can be calcula.ted by similar methods. We bere ...alua.te only the
pa.rt. Taking into account the results of Section 11.3 we find:

{ lft/14>)1Pl,l loop, dlv. -~ ~d' (IU29)

{4$N>bPI,ll<><>p, dlv. = ( -~),2 + 24Ng•) ~d • (11.130)

 Dimensional Regvlo.rú:ation, Mínima! Subtroction 11.8
24S

Renomwlú:ation constants. We first replace action (11.113) by the renormalized actíon:

S, (ifJ, 1/>, tjJ) ~f d•x {- Z~ [t:p(x) · (1!1 +g0 z!12 rj~(x)) ;p(x}]
+ ~z,. [ca•.P(x)} + m~rJhrl] + ~>-ozlr;.•(x}}.
2 (11.131)

We call !11(), g0 , .l. 0 the bare parameters, m, g, .l. the renorma.lized parameters, and """
equation {1 Lll:>) to expand the bare roup!ing constants in terms of g a.nd .l..
Substitutíng into the Feynman diagrams a.nd ideotifying the divergent parts we obtain
equations for the renorma.lízation const.ants:

i,l ( Z,p + ~9 2 : • ) = finite O (2 loops),

m -•1
2 Z z'l' g 3 Nd
Uo~,.-E
finite O (2 loops),

Z4> (p2 + mS) + ( -~m2 + 2Np2 g2 ) ~d = finite O (2 loops},

m-'.l.oZ~+ (-~)..2 +24Ng4 ) ~d linite0(21oops).

I t follows that:

z., = N•
1 - 19 2 € + O (2 loops) , (1U32)
2
Nd
Z~ = 1 2Ng2t + 0(2 loops), (11.133)

(2N+3)
9o = m'12g ( 1 + --- - g 2 E
Nd) + 0(2 loops), (1Ll34)
2

.l.o = m' [), + G>. 2+ 4N.l.g

4
24Ng ) ~d] + 0(2 (oops),
2
-
(11.135)

~= m 2[1+ (~ +2Nl) ~d] + 0(2 loops). (11.136)

RG-functions. When bare and renorma.lized coupling constant• are related by an

equation of tbe form (11.11"), the coupling constant RG functíons /1;(>.) are gíven by:

(11.137)

and thus using equation (11.115):

m!Àot m' (</; (>.) + fJ1 {>.) !{;) · (11.138)

A short ca.lculation then yields:

{J•• = -tg2 + Nd(2N + 3)g' + · · · , (11.139a)

-e.l. + Nd (~-\ + 4N.l.g - 24Ng•) + ·

{JJo. =
2 2 (1l.l39b)
 11.8 Dimensional Regu!arization, Mínima! Subtraction 249

In the mínima! subtraction scheme:

where TJ(l)(J,) is the coefficient of 1/t: in the correspondíng renormalization canstant.

Then:

'1<1> = 2NNdg1 , {11.140)

TI<~>= ~Ndg ,
2
{11.141)
'72 = -~NdÀ 2NNdg 2 • (11.142)

Again note that the functions IJ<i>, !Jo are positlve for small couplings, a result consistent
wíth the general property Z.p, Z 0 < l (see Sectíon 7.4).

Bibliographical Notes

Tapo)ogical properties and parametric representatíons of Feynma.n diagrams are studied

in
N. Nakanishi, Omph Theory and Fe!fllman lntegrals (Gordon and Breach, New York
1971).
Feynma.n diagrams and dimensional regularization are discussed in
G. 't Hooft and M. Veltman, in Particle lntemction.o at Very High Eneryies, D. Speiser,
F. Halzen and J. Weyers eds. (Plenum Press, New York 1973) voL 4B.
The properties of RG functions in the mlnimal subtraction scheme is dlscussed in
G. 't Hooft, Nud. Phys. B61 (1973) 455.
A good presentation can be found In
D.J. Grosa ín Methods in Field Thenry, LesHouches 1975, R. Ba.lian and J. Zinn-Justín
et:L.. (North-Holland, Amsterdam 1\l76).
For more systematíc techníque<> to calcula.te RG functions at higher orders see:
A.A. V1adimirov, Teor. Mat. Fiz. 43 (1980) 210.

Exercis.es
E:J:en::ise 11.1
Characterize the form of renorma.lízation constants in a mínima.! subtraction scheme
with cut-off regularization, the parameter heing A/m in which m is a m..ss scale (see
Subsection 8.1.1).

E:rerr::ise 11.2
Pseudoscalar Yttkawa interaction. Consider the action

S (~.-r/-,{,) j ddx{ -{,(x) [I"+ M + (ns<l>(x)lf(x) + t (ô,.~(:r)) 2

+ ~m 2 4>2 (:r) + ~>.,P4 (:r] }. (1LI43)

' In which ~ is a pseudoscalar boson field and -rj., {, Dirac fermioll8.

250 Dimensional Regularizatioll, Minimal Subtmction 11.8

The factor i in front of "Ts is imposed by invariance under rellection hermiticlty, i.e.
euclidean herrnítian conjugation followed by spa.ce reflection (see Appendix A5.6) which
implies:

In addition the a.ctíon is reflectíon symmetric (,_ Appendix Af>.5):

~(x)..., li' (i)1'1, ,P(x) ..... -,P{i),

with i= (-x1,x2,··· ,xd),

The transformation of <Jl(:r) is imposed by the factor -rs in front ofthe qnp,P ínteraction
and justifies the denomination pseud~a.r for the field <P (x).
Cakulations can be dane using dimensional regularization and minima.l subtraction
becaw;e it is never necessary to evaluate any trace of ptoducts of "TS by an even number
of /;. matrices.
11.2.1. Calculate the ~..;, 2-point functíon at one-loop arder. It is suggested to expa.nd
on the basis {1,,1} and to use identíties of the form

11.2,2. Calculate the dívergent part of the rt> 2-point function at one-loop arder.
11.2.3. Writing the renormalised action:

S,(<P,,P.~) = J ddx{-Z,; [.j;(:r) (~+Mo+ irsBoZ!12 ,P(x)) l,{l(x)J

+ ~Z,; [(ôi',P(x)) 2 + m~<P2 (x}J + ~>.oZ;<P4 (x) }. (11.144)

calculate ali renormalization consta.nts and RG functions. One rnay introduce a renor-
mallzation scale 1-' and define the ,8-functions by

(11.145)

Compare with the resultB of Section 11.8 for N = 1.

11.2.4. Relate couplíng and fiekl renonnallzations of models (1 L 113) and (1U43) by
perforiiiing in action (11.143), restricted toM O, a chíral rotation on the fields ,p,;jJ
{see Appendix A5Jl).

E.xerciSe 11.9
Explain how the inductíve proof of renormalizability of the <P4 fie!d theory given in
Chapter 10 has to be modífied in the case of dimensional regularization. Note that it
is !lSSential to take ínto account the additional term of arder g, (g, ís the dimensiocless
renormalized coupling constant) in the CS-,8-function
12 RENORMALIZATION OF COMPOSITE OPERATORS. SHORT
DISTANCE EXPANSION

In this cha.pter we discuss two rela.ted prohlems: The renorma.Iiza.tion of oomposíte oper-
ators, Le. loca.! polynomials of the field, and the short distante expansion (SOE) of tb..
product of operators. The relation is easy to understand: we can consider the insertíon
of a product of opera.tors A(x)B(y)"" a regularization by poíot splitting of the operator
Ali(:r: + y}]B{!(x + y)J. Therefore in the limit x ..... y""' e:xpect the product to become
dominated by a linear combinatíon of the local operators whích appea.r in the renorma.I.
ization of the product AB, with singular coefficíents functions of (:r y) replacing the
usual cut-oH dependent renormalization constants.
We first discuss the renormallzation of eompooite operators in general from the point
of víew of pawer counting. We use the relations between bare and renorma.Iized operators
to establlsh a set of CS "<!Uations for the insertíon of operators of dimem;ion four in the
~Pl field theory. We then show that in a given field throry there exist linear relations
hetween aperators due to the oquation of motions and relations derived in Section 5.4.
In the second part ofthis chapter, devoted to the SOE, we first establish the existence
of a SOE for the product of two basic fields, and discuss the SDE at leading arder in the
,p~ lield theory.
We then p8Ss from short distance behavíour to lruge momentum behavíour a.nd derive
a CS equation for the coefficient of the expansion at leeding order, We finally briefly
discllSS the generalization of this analysis to the S DE beyond leeding arder, to the SDE of
arbitra.ry operators, and to the light cone expansion which appears when one studies the
Iarge momentum behaviour of real time correlation functions [in contrast to euclidea.n
imsgina.ry time) ,

· 12.1 Renormalization of Operator lnsertions

We have shawn in Section 10.6 how to renormali:w insertions of the operator r/> 2 • We
conld have comidered other vertices like ,p•, (â~q,l ... They all generate new divergences
which have to be eliminated by addltional renormallzations.
In Section 8.4 we have explained bow to calculate the superficial degree of divergence
.;{ the insertion of a local operator O (x, ,P) by adding 8 source g(x) for tbis operator In
action:
j
S(q,) ,__,. S(,P) + dx O (:r,,P) g(x).

this choice of sign, differentiation of W (J, g) with respect to g(:r) generatffi in-
of -0 {:x, ,P) in connected correlation functions. However since .lT j6g(x) =
j6g(x), 61' jfJg(x) corresponds to tbe iDBertion of O [x, f>) in proper vertices.
. All 8 convention we assign a canonical dimension Jg(x)] to the souroe g(x), opposite to
tbe dimension ofthe vertex assoda.ted to O(,p) and thus related to the dimeliBion JO (4>))
the operator O( q,) by:
!.9) = d JO(,p)).
· Witb tbls convention the COUS"<inences of power counting a.nd renonnalisation theory
a simple formu!ation: a.fter the field theory has been renormallzed up to (L- 1)
252 Renormaliz~>tion of Camposite Opemtors

loops, the div.,:gent pa.rt 1' of r((!>, g) f1 loop order,is the most' gen~ra!Íoéai
átL
tional of .p(x) and g(:r) allowed by power eounting, i. e. th• moat general linear c"o•"rnb•inati<J,n'>
of aU ve.-tiC!'.' in (l>(x) and g(x) v! non-pasitivo dimensians. Note that this in ~ ... ;,.;,1;.~
implíes that an operator of a glvén dimension inserted one time in general mlxes
renormalization v.ith all operators of equal or.lower dimension. It is thus natural to
the renormalization of oper.-tors of ineieásing dimeilsion.
Let us thus first consider again the insertion of ,P 2 (x) in a ç,•. field thooey in 4 dbneilsi<lllS:
which W, have already diBcussed in. Chaptér lO> ;

J2.1.1:The'(l>2 (x}i;serl?.on:. . , ,-
We llil~ the ooriventío~ oi Chapt~r i ofói 'thi; b,;.,:.. ~d ren<>r~ized ·
. . . -~(1/J);=. :/·. J~.·; (i. (b:.p).~+ 1;,;~. . .;.~;· ...:'~.,4!·-.l.g(/.....
. ;.... : .. 2. ,, . . 2
>•~.J'·.·..'•
:. ·.

Jd•x (f~ (8p~) \i(~ -r 6;~)~+ ~~,z~~:J>

S, ((/>.)
2
In the ri>4 field theocy in 4 dbnensions th<!op(,..á~or 4>2 (x))ias
; [.P"] =2. '·
Let us Cai! !(:<) the so,;rce for. (1> 2 (:~:). ~~ dim;msion !t]
-· ' - ' . ' . . : - .~ '

lt(x)] =:i.

f 2
t{x)(l> (x)d z
4
~hich has dime~ío~ zero
f e(z)d4 x which has also dimension ~o
.f t(x)d 4 x which has dim..:Osion . 2.

The reoormalizéd aétion then hBB the form:

5, (9,; tj S, (~) +!Z2/ d 4~~(~}(1>~(:~:) +/d•i[h~2 (,;) :tbt(x)],

The last tw~ terias contríbúte on]y to the '=uwn ~plltud..;, Eipr"""ion
a set of reÍations hetwoon bare and. rénormalised generating functionals:

f,(J,t) = z(J!VZ,t Z.,f z) exp [~ ~~~x'(!a~ + ÚÜ •

For t~ eonnected funct!O:ns this gi~es:
W(J!·dt z;; z)
Áfti!t lÁ!i~~e):fáiiSfo~~tíbn wlth ~~;;,à ~ J
r,(cp,t) r (..,v'Z,tz,;z) +f d x(!at 4 2
+'bt);
Th<pariding in powen; of t and <p, one recoven; the relation5 betwoon bare an<i ,;.,,;v,.,....,.,u.,
proper .e:rtiC<!S proven in Cbapter 10.
 253

1_.2 1'he tf>3 ínserticn __

~ us n~ consider the example ofthe 1/>3 insertio:O:. The operator .p3(x) '"'" dünension
~,;thê cmres>Jonding .souree t(:r) bas_dímellSÍon 1. T~ r<;normtilized actial'-.S,\4>, t)thllll
foriri:_- • , ::·..-·.-: ·..-.'".: .-- -.<-c-_. \;: :_.i:_.,._'i- ....... ,__ , -

;- s,(;,t)·~:~:C~) ~ [~4~ F~) (~~3~{;)+-~~~z) +, ~(~)) __

.:· .+e(?,.t.(~)); ;_t;(:) ~d~~Cxl·~· e)+ lt.3 (xWx).+ ~·(z)r

ex·pr<ISSion;úl~:ul~t.tz;'~~{3-~~r~:;~~~~ ~~(1)n:ixe3"~~r ren~;ma~;~~i;n

the d"::'blií i:ns.lrtionoí(!> g'?""'"atese. c~untei"tef?lproportíon<;ltô 1j>2. ~-, ~
· _· ·__ the ~quiwlent ofrelation5 (12.4::l2.él) 'WehaYe th~ref()re to ~~(iüy_
SoUTrn ror
~ 2,{x). }Ve IÓ~ it to- the reader).<)\vl-ite the feDOrÍnaiJze(f â,;,tion
• Jor ,p a:n:d '-4l: and thé relátiônsK~il rênoromuíéd á.nd bare. :ProJ>É>r
3
· _t!Íe discuS!lÍon of the CS eqlw.tiorui of ccrrrelatiOil functions Vr"Íth ·
>Se:m<llll:W<U_ SectÍóij 12:2. in OrdeJ; ~ be able,to iÜCOtp<)Tàte infotmafiOU WmÍIIÍ!; from
IJ<J;Wl_ eq1JM~On ,_ . . . . ..· : :,,:, L: "· . _. ;_. , ,: .·. .- ·-· .._
.. ,· ·:·:,c: '_,;:~; .-.. ' :::~.-~ ~.:.> .;·<<:.:

Theie.··
 254 Renmma!ization of Composite Operntors

in which A and B are 2 x 2 matrice:s. We ha.ve used foi the renormalization of r/; 2 the
notation of Section 10.4.
CS equations. Froi:n equation (12.9) we cari derive a. CS equatíon foi- (rt:\-·
hei<! some ca.re is requited. The CS operntion ínvolve:s qi 2 ínsertions a.ild the produét
exaÍnple (.P4 (%)), {.p2(y)), ilÍserted in a correlation function is not finite: since the ".···-··~ ···"'·•
for .p4 has diinension zero iuid tlH> sourcé fur qi 2 dim<mSion 2 the product of ~ by
sourres ha.s díÍDension 4: Figure 12.1 give5 the first di\'ergent díagra.m. · ·. ·

f, ((.p4(x)), (.p'(y)).J;
. ín'which C4 is anew renormaÍG.;.iion constant. Identity(12.10) isóclytrue es'an.
in. a.n n-poínt correlatiori functíon, n f. O.-
After Fourier transformiition arid li:Jr an insertion óf qi2 at zero molnentum;
(12.10) becomes: · · · ·· · .
' ;: •,c

[(õ. (p)). ( 61 (Ol)J. _·. (~4 (rl), (õ,(oJ), +c. (é\ (pl)~.
A sí.Ínilar equation holds !Ór tl:ie operator Oa. · • .. ·. . . • . ..
Let us now apply tbe CS óperàtor, rri,f!//Jin, àt g ..WdA'Ifxed, on equat.iõn (12
A new set oi RG fum:~~llS is generated invohing t~e ~atrix /liJ'c

;.'·.~c •

As ;; consequence of ~Jation (t:i.Ú),b"' eleme~bí of thé. matrÍx q;i are J1o(lici~.

the cut-olf becomes · lnfullte: ;J'3i Ãnd ~Ú· Thclr di..:•ergent part Ca.nCéu.· tlie ·
<.ÕmiiÍg frcim· ~he insertíori o(IJ)~ as reyruented by ,;qua.t.iÓii (12.11): Defioíng thênô
' ·;- .,.. ·· ' · · .' . · • 7:r .·'• · · ·i.··· :
 Renorrnalízation of Compos;te. Opercitors 255

.. -
'ébmplet<es.the diàciÍSÚôn ~ftbe inserÚÓnollé tlmé~f tlie opera~s'oi dbn..nsion 4
·_ It te\'eiili; the general fea.tures of the Íllsertiori of ány other operator
c!ínlensíoJo.· ;_-_ -r·~~:·~!~"~7v; ..tr. ~:;:·.~- :~··:.<! ..,~-~--, .;:; .·-

·~ :-"•>.: ~:é::~:::~ ~_!\~:.::~~~r~:;~.;~·tt·~:-~·;:I/~:; ·;. ·. ,.:..··:·~·~';~,5}:~;'4-·:'··::~-.~.··~~\>~>: ··>·:,) ;_.:·~:~~~<'

(~"ivllrt =UP4 ("')); (~,4 (ÍIÍ), +iíJi::; '!)) I;.Q4; (Ó~(~{i))J; :;"
-_>:~r-~:{:·:~-;~~/~-:<~~ -· .~> . _..:·:. :;_-,~~-1- . <-:;_ :-~~}~~:~~:·:-;<:; ;.-~,.\;-\:,~.,-, .õ~-~ .. -_--~-·-i .::
·+,~~ó(:t-y)Ei8,.01 (<i\(:t:J)r+fiZ$(:tiY)F,Oj(Ç>(x)),, (1'2.16)
....... -:~.~::;~:~:;L'-<-~~=.._--:-. .· --"~, .' ·-._r_::''.-·"· -. . ·
E• and F4 ·a.re neW- renorirializatlon ool.stant8. -A simllar eqú..tion is valid
Again_ equa.tídri (12.16) ;_is válid only as ao inaertíoiL .: ·: .

countlng arguments, based on the dimensionor operators and sources, teii us quite
that íf O (,P(x)) is anoperator oC ctwoincal dimension D:
··~ ;;~;;:tiiv:'7<"i;,Zf)J:k~:;;, .· _; ,.
(12.17)

· - - -·:_·~·>·à·-~;,· ,·;,nrJI tafrl:• tro'n. ~\;;.~ ~~í!íl~:i:, "'""

•. are intt~pendent' ;!," a"gi\'eri field throry,'bedi:use tbe"eqizaúiln ofmotlon and
éntitil'l5 discussed in seétíon 5.4 imply relatlai!S betWeen opéra.tnrs.
2f>6 Renormnlization of Composite Operotors 12.2

3
12.2.1 lnsert:wn of lhe 4> opemtar
Let us agaín consider the example of tbe ,P3 (;r) operator in the framework of the 4>4 field
tneory.
We therefore consider the action S (</!, t, u):

S (<,6, t) j d :r [ ~ (lJ.,<,6)
4 2
+ ~m2 rfJ2 + ~9</>4 + ~t(:z:)4>2 + ~u(x)<,1>3] • (12.19)

Differentiation with respect to u(:z:) and t(:z:) generates -;b4> and insertions. 3
-!4> 2

A regularization is assumed. The simplest equation of motlon is obtained from the

identity:

/!dQi} [à~~) J(;r)] exp(-S +NJ o.

Since we have introduced sourres for and <1;0 t/>3 ,
we can use them to express the terms
r/> 2 and .p 3 in õS /54> as functional derivatives witb tespect to t(:r) and u(x). We then
obtain:
2 f,Z bZ õZ
[(-6 +m )" +t(:r)J õJ(x) -9óu(x) -u(:z:) ét(:z:) J(x)Z (.1,u,t) (12.20)
This yields an equation for W ]n Z:
éW óW 6W
[( -li+m
2
),., +t(:z:)]õJ(x) -!l6u(:r) u(:r) ót (x) = J(:z:). (12.21)

The Legendre traru;formation is straightforward. We remember tbat:

óW õr õW 6r
óu(:r) - óu(x)' M(x) - ét(x)'
and find:
õr ór .sr (12.22)
[(-l1+m2)" +t(x)]ll'(x)+96u(x) +u(x)6t(x)- "'O.

To explore a lirst consequence of thís rela.tinn, ]et us set t(x) = u(x) = 0:

6r
6u(x) •~<=O
r.,.' I (12.23)

r 4>' ís the generating functíonal of IPI 9-field correlation functions witb one ~ 4> inser-
3

t:ion.
(12.24)

Tbis relation shows that up to explicit subtractlons affecting only the (4lif;) correlation
function, the insertion of q, 3 is equivalent to the insertíon of .p itse]f.
Tbe diagrammatic interpreta.tion of equation {12.24) is simple: the insertion of r/>3 is
indistinguishable from the addition of a. ,P4 vertex wíth one of the !ines a.tr.ached to the
vertex being an externalline (see figure 12.2).

Fíg. 12.2 4>3 ÍJJSertion.

12.2 Rtlnonnalization oj Compoeite Operotors 257

4
Of course diagrams without a 9 vertex cannot be gen.,rated and this expla.ins tbe
subtractions.
L...t us now introduce the generating functional of renormali>.ec! proper vertices r,(o;t>):

r, (op) =r (r.ovz) ,
and thus:
6r (,.,rz)
rz .s..,(x) '
We use this relation in equation (12.24):

(12.25)

The r.h.s. is finite in the infinite cut-off limit. We now introduce the ,P4 field theory
renorm&Jization constants as defined by equations {12.1,12.2):

(12.26)

This relation shows iliat ali -P-field correlation functiom; with one inrertion of the opera-
tor g,Z9 tf>~(x) are finite except the (1/?.p) correlation function which needs two addítional
subtractions. \Ve can determine the corresponding renormalization constants by impos-
ing:

(12.27)

Equatíon (12.26} expanded in powers of tp tben yields aplicitly:

r r(n+l) ( q,pl,···)Pn. ) _
-9r (r(n)) r (9,Pt,·••,Pn
4/~
. ) +l)ol
< (
P 2
+m ') · (12.28)

Note that with this definítion (r~~), satísfies the renormalization condition:

(r~~). (o; o, o, o)= 1. (12.29)

Equation (12.22) also conta.ins information about multiple insertions of .p3. For example
aft.er some alg"btaic manipulations one finds:

2 .5 2r
4?'(<>)~'(•>) + g6 (xl ~'(z,) + (-Ll + m ) Ó (XJ - Xo) "' b'P (X!) Ó'P (:t~)
2
g [ :r o) r
(12.30)
Tbe equation rel&tes two in.sertions of </; 3 to two ínsertions of ,P, again with •~!oé;c.otío:J
2
terms which now involve the insertion of ,P •
258 Renormalization of Composíte Operotors 12.2

12.2.2 Other relo.tions: renormalízation of operolon of dimensian 4

We have shown in Section 5.4 that more general equatlons could be obtained by perform-
ing infinitesimal changes of variables. We can use them t.o establisb relations between
operators. For example in the change of variables
</J'(:r:) = ql(:r:)+ E(:r:)ql(:r:),
the variation of the action (12.1) in the presence of a oource i>l:

6[5(</.>)- Jql] = õ(:r:) [<i>(:r:) (-D. + m 2) h 4>(x) + ~<1> 4 (:r:) J(:r)4>(x)].

From the invariance of the functional integral follows:

J( 6W (12.31)
x) 6J(x)'

The discussion of the large cut-off limit of this equation, or the corresponding one ob-
tained after Legendre transformation:
g .sr
3/
r ~(·)(-l'>+m'J,>I>(•) + 4>'(•) = <p(:z:) 6op (:r:)' (12.32)
2
is more delicate than in the case of e<Jilation (12.24): tbe operat.or 4>(:r) ( -D. + m ),.,, 4>(:z:)
which in Pauli-Vi!lars regularization is:

contains operators of canonical dimensions larger than 4 divided by powers of the cut-
off. We sha.ll discuss in Chapter 27 the problem of irrelevant operat.ors wblch is directly
related with tlle Jarge cut-off limít of such operators. Let us here only state the resu!t: in
the !arge cut-off Jimit the operator <i> ( -D. + m 2) A </J is equivalent to a linear combination
of ali operr.tors of dimensions 4 and 2.
Equation (12.32) implies after renormr.lízatíon an identity satisfied by the operat.ors
(0;(4>)), as defined in equations (12.8,12.9):

L• C; (g,) (r:;}), (q;p, ... ,p..) L rln) (p, ...• Pm + q, ... ,p,).
n
(12.33)
t=l m.=l

If we ímpose some renormalizr.tion condítions to define explicitly the insertions of

operators 0;, in the spirit of Chapter 10;
2
( r<>l) (O· O O)- m r-'
Ot ~" ' '

(rg!), (q;p~.P7) = -p,. P7 +0 (P4),

(r~n. (o;o,o,o.o) =o. (12.34)

(rg;), {q;PhP2) 0 (P4),

(rg;), {o;o,o,o,o) = 1,
tben we can calculate the coefficients C; (g,) only from (r~"?), and its derivtltives at zero
momentum.
12.3 Renormaiízation of Comporite Opernt<>r• 259

12.3 Short Distance Expansion {SDE) of Operator Products

Severa! chapters will be devoted to the discu~sion of the large or small momentum be.
haviour of correlation functions. Our essential tool wíll be the CS or RG equations.
However these equations are directly useful only for non-exreptional momenta. For Ja.rge
momenta we shall be able to characterize the behaviour of

províded no subset of mornenta p; h as a va.nishíng sum:

L Pn.!O V!t'(1,2, ... n).

iEl;i~

For exceptional momenta a new tool is needed: the shott distance expansion (SDE) of
product of operators. In this soction we shall mainly díscuss the SDE of the product of two
lields at leeding arder, but we shall indícate haw the method we use can be generalízed to
more complicated cases. lt is entirely bBSed on the theory of renormalization of composite
operators we have just presented.
Definitwn. We consider the lPI cotrelation function:

r(n+1) (x +~,X ~,))!, • • ,!in)= (ifi (x+ ~)<f> (X-~) ,P(y!) •· '</J(Yn)) !PJ ,
(12.35}
in which ali arguments are lixed, except th!' vector v which goes to zero. We want to
study the v -> O limit. In a theory which is sufficient!y regularized, llll expansion in
powers of v can be obtained by expanding the product of fields <P (:t + v/2) 4> (x- v/2)
in powers of v:

in which o,., ...,.,.(</>) is a local operator quadratic in <P with 2n derivatíves.

lf we naw insert expansion (12.36) into a correlation function, even after field renor-
malization ali terrns of the expansion In general diverge when the cut-off is removed. As
we have extensively discussed in Soction 12.1, the va.rious composite operators appearing
in expansíon (12.36) have to be renormalized. Therefore we have to expand each bare
operator on a ba.sis of renormalised operators. In terms of renorma.lited operators O;' (4>)
expansion ( 12.36) then takes the form:

(12.37)

in which, at cut-uff A Jixed, the coefficients C~ (v, A) are regular even functions of the
vector v. An O;'(x) receives contributions from bare operators of equal or higher dimen-
•ions. We conclude that for lvlsmall the coeffident functions C, (v, A) behave lilre:

{12.38)
260 Renormalization of CamptJsite Operators 12.3

\'lhen the cut-off becomeE infinite, the coefficients of the expansion of C,. in powen; of ti,
being renormalízation constants, diverge, and if Co has a limit, the limiting function i•
singular at v = O. The coefficients are functions of the cut-olf which for v O diverge in
a way predicted by power counting. Since v is smali but non-vanishing, the coefficients
will grow wíth the cut--off until Ais of arder lflvl. In this range ali contributions to a
given coefficient are of the same arder, up to powers of logarithms because powers of v
compensate the powets of A. Therefore, at least in perturbation theory, the ordering of
operators consequence of equation (12.3B) will survive, the operators of Jowest dimen-
sions will dominate expansion (12.37) for jvl small and the behaviour of the coefficients
Co.(v, A) is given by equatíon (12.38) up to powers of In v.
The expansion (12.37) is the short distance expansion (SDE) of the product of two
operators rp.

12.$.1 SDE at lead•ng ord~r

F\Jrther insight is gained into the structure of the SDE by rea.li•ing tha.t the product.
<f;, (x + v/2) <f>, (x -v/2) can be <:onsid~red as a regulad2ation by point splitting of the
composite operator (rfl2 (:z:J),. Let us discuss this point in detail in the framework ofthe
<P1 field theory._ We then expect:

(12.39)

The singularities of C 1 (v) for lvl sma.ll should be dírectly related to the divergentes of
the renormalization constant Z, which renders q,;(x) finite.
In what fo!lows it. wi!l tberefore be con~nient to treat the product:

as one composite operator depending on the point x and the parameter v, in particular
fiom the poínt of view of connectivity and one-particle irredudbility. To make thís
explicit we shall introduce a notation for this operator: ~ I<P, (x + vj2) rp, v/2)}. (:r-
By definitíon we have:

! ( l\)\, (X + n (X -
rp, ~) l \)\, (yi) ... 1/J, (y,.))
! ("'· (X+ ~) .p, (X- D"'·(!li) ... \)\, (y,)). (12.40)

However the relation between connected correlation functions is different:

( [rflr ( x+ ~)</;,(:r 1) j.p, (yJ) · .. <Pn (Yn) ),

! (<P, (x +~)<f;, (x- ~) </>r(Yt) .•. .p, (Ynl)e
+~L (<P.(:r+~)<Pr(Y;,) ... </;,(!1;,)) 0
JUJ={1, ... ,n)

X ( ,P, (:r- D,P, (Y;,) · .. 1/Jr (YJ.-.))e , (12.41)

in which I =d J are all non empty partitíons of (I,···, n).

 !2.3 Renormalil:ation of Composi.te Operators 261

Rather than writing explicitly the corresponding relations between proper vertices, we
give the relation in terms of generating functionals. Calling Z ("' + v/2,x- v/2; J) the
generating functional of correlation functions with the operator insertion:

(12.42}

we find for connected correlation functions with obvious notation:

W (x + !:_2' X - !!..2' J) : z-l (J) Z (x + !:2' X-!:_.2' J)

I li 2 W(J) I 6W oW
=
2 6J 2 6J ( 2u)+-2{jJ { 2") fJJ ( 2),(12.43)
(v) X+ X X+ :l:-
0

and finally for the generating functional of proper ""rtices:

r (x + !:2 'X- _2V; <p) !<p (x + ~) <p (x - !!.) + ~ [ ó'r ] -I

2 2 2 2 6<p(x+i)6<p(x-~)
(12.44)
Note that this equation ís similar to equation (6.64) which we have used to prove the
irreducibility of r (\I')·
To now ensure the limit (12.39) we determine C 1 (v) by imposing that the insertion of
the operat.or ~C! 1 (v) {4>(x + v/2)tf>(x- v/2)] in the 2-point function satisfies for any v
the renormalization condition ( 10.22). Defining:
2
Cjl(v)jdxdy,dy., ei~H•,.•+'o•"' 6 f,{x+v/2,r-vj2;<p)l
li'{) (Jii) 6'{) (y.,) 11'-o
4
= (21<) 6 (p + ql + q,) r;•.•l (v;p; q], q,). (12.45)

we impoSE:
r;"2l (u; O; O, O) = I. (12.46)
q2 O in equation (12.45), we derive:

62 r, (v/2, -u/2;11')
6rj> (O) 6rj>{0)
I
<p=O '
(12.47)

which r{! (q) is the Fourier transform of ll'(x).

By differentiatíng equation (12.44) twice with respect to '{J, we can relate the r.h.s. of
.. equation (12.47) to the '{J-field proper vertices:

Ct(v) =I _1
2
f -
d-
4

(2n/ '
1•
k e-•k• [w!2 l(k) r!•l (k -k 1 o O)
' ' ' '
(12.48)

The equation ca.n also be rewritten:

(12.49)
262 Renormaluation of Composite Opemtors 12.3

·we have introduced a mixed connected correlation functions, ~ (P) being the Fourier
transform of the field if;(x), and used renormalí.,..tion conditions (10.9).
The roeilkient C1 (v) is defined in such a way that the renormalized operator if>,(x +
vj2),P,(x uf2)C( 1 (v) then converges towa.rds the operator {.P~(x)), correctly normal-
ized.
Let us fur example write the consequence for lhe 4--point functian:

c,(v)r!'·,l
'
'-·q~
'""' '
q,) ~
lvi-O
1- 12 f d4k
(2n)•
.,-•kv wi2J
'
(E-
2
k) w<2J (~2 + k)
r

x r~4l (~ + k, ~ - k, q~, q2) • (12.50)

The neglected conttibutions are of arder v 2 (In u 2) P at lilllf finite order in perturbation
theory.
Renormalization theory te!ls us that equatíon (12.39) is valid as long as the replace-
ment of the operat.or product by ,P2 (x) does not generate new renormalizations. 'Ve can
therefore use equation (12.3!1) in a.ll .;l>(x) and .;62 (o:) correlation functions except
2
(i) the 2-point function (<P (x + vj2) .P (x- v/2)) which leads te {.P (x));
2 2
{ii) tbe 4-point function (I/> (x + vj2) 4> (x 11/2) r,/1 (yl} which leads to the ,P 2-poínt
function (1;2 (x) ,P2 (y)).
Bath require an additiona.l additive r<morma.lhation. The stra.tegy in such CBSlS isto
first apply the CS di.fferentia.l operator on the correlation to generate additiona.l f ;t.~(x)dx
insertions until rela.tion (12.39) can be used. As a consequenoe the SDE is modilied by
contributions which are solutions of the homogeneous CS equations.

12.3.2 One-Joop rolculntion o/ the lroding coefficient af the SDE

Equatíon (12.48) ca.n be used to ca.lculate the coefficient functíon Ct(11) in perturba.tion
theory. At one-loop order it is suffident to repla.ce correlation functions by their tree
levei values:
g d"k f
.-ikv
C,(v) = 1 - 2 1611'2 (k' + m') 2 +O (g ) •
2
(12.51]

1t is clear from thls expressíon that, as expected, C,(v) has at on.,..loop arder the iorm:
2 (12.52)
Ct(V)- Aln{lvim) +B +O (v ).

A n often usefu] idea to extract an asymptotic expansion of this form is to ca.lculate the
Mellin transform iJ (a) of the function:

(12.53)

in which v is the length lvl of the vector v. The exp8ll5ion (12.52) then implies:

A 1 (12.54.)
p(a) - al +(A lnm+B) I>+ O (1) for a_, O.

(12.55}
 12.4 Renvrmalúation of Compo~ite Opemtor5 263

As 115ual we rewrite the integral as:

/(a)= J.duvo-l (""' tdi~e~•(>'+m')~ilrv. (12.56)
o Jo Ifur2
lntegrating owr k, v a.nd t, in this order, we finally obtaín;
/(a)=.!
8 m
(2)o
r2 (1 +2 af2).
a
(12.57)

An expansion for a - Oyields the coeffidents A and B:

C1(v) ]~fG[-ln(";')+.P(l)] +0(g7), (12.58)
ín which .p (z) is the logarithmic derivative of the function r (z).
At this order t.he bare parameters can be replaced by renormalízed parameters.
12.4 Large Momentum Expansion of the SDE Coefficients

To the behaviour of the product of fields </J (:r+ vfZ) </J (x- v/2) at short distance;,; asso-
ciated, after Fourier tnmsformation, the behaviour at large relative momentum k of the
product ~ (P/2 k) ~{p/2 + k). Howevet oome infonnation is lost in the transformation.
The large momentum behaviour is only sensitive to the singular part of the short distance
llehaviour. For instnnce the constant tenns in the asymptotic expa.nsion of C 1 (v), yield,
after FourieJC tr!UlSformation, terms proportional to 64 (.1:) which do not contribute to the
la.rge moroentum behaviour. At the same time tbe algebraic structure is for the same
reason somewhat simpli!i.ed.
Let us ta.ke the example of equation (12.50). We introduce the Fourier transform C 1 (k)
of C 1 (v) (as stated before, we omit the tilde sign indicating Fourier transformation when,
dueto the change of arguments, :r, y, z, v to k,p, q, there is no nmbiguity):

Ct(k) f e''"C1 (v)d 4 v. (12.59)

After Fourier transformation, in the large k limit, equation (12.50) yields:

rl'l (~ + k, ~- k,q 1 ,q2 ) ~ -2C1(k) [rj2l(k)r r;1. 2l (p;q 1, q2 ), (12.60)

in wbich the neglected tenns are of order (lnk)" Jk at any finite order in perturbation
2

theory.
More geuerally, due to momentum canservation, the dísconnected contributions in
eQuation {12.43) coming from 6W f6J (x + vj2) 8W f6J (x vj2) do not contribute to
thelarge momentum behaviour.
ln addítion if we expand the r.h.s. of equation (12.44) in powers of rp, we obtaín
two types of contributions: one term wbich becomes IPI after amputo.tion of the lines
·. i:orresponding to the fields ci> (:r+ vf2) and <1> (x- v/2), and tbe other tem>E which remain
 Renormalization of Compo5ite Operators 12.5
264
Due to momentum conservation, in reducible terms the propagator which connects the
proper vertices carries a momentum of arder k for k Iarge. The corresponding contribu·
tjons are thus suppresood by a factor l/k2 (up to powers of In k) and can be neglected at.
leeding arder.
At leeding order we can therefore write fnr 11 > 0:

r~•+ 2l (~ +k.~ k,qJ.···•qn) ~ -2C1 (k) [r~2l(k)]'rp.•l (p;q., ... ,qn)·

2 2 1•1-~ l (12.61)
2
As expla:ined in Section 12.3, this equation generalizes to </> insertions provided either n
is positive ar I > 1:

r;1·•+>l (v,, ... .P!: ~ + k, ~ - k,q 1 , •.• ,q.) ~ -2Cl (k) [r~'l(k)
x
r
r!1+l.•l (p, ... , Pt.v; Q~> ... , q.). (12.62)

12.5 Callan-Symanzik Equation for the First Coefficient of the SDE

In this section all quantities are assumed to be renormalized and we omit the subscript
r.
Let us firS1; note by comparing equations (12.48) and (12.61) that in momentum r•P·
resontation the relevant function ís rt•l (k,-k,O,ú) which we sball cal! B(k) in what
follows

Equation (12.61) becomes:

Similarly:

r (l.n+•l (o·1.?. ~
'2 + k '2
2
k ,Q~o ... ,qn ) ~ B(k)r! ·"l (O ,p;q,, ... ,Qn ) ·

Let us íntroduce a notation for the CS differential operator:

2
We write the CS equations for rtn+ ):

[d- t (n + 2) 'l(gJ] f!n+ 2l = m 2 a (g)r(l.n+>J (O; ... ).

We now use equations (12.64} and {12.65) in the large jk\lirnít:
12.6 Renormlllization of Composite Operator• 265

Using the CS equation for rt 1·">:

dr(l,n)"' [~'1(9) + '12(9)] rti,n) + m'u(g)rt2,n) (O,.,.)' (12.69)

we finally obtain an equa.tion for B(k):

ld- f/(9) + 'l'l(g)jB(k) ~O. (12.70)

We can also write an equa.tion for c,(k), using relation (12.63):

(12.71)

We can compare thís equa.tíon with equa.tions (10.27,10.213) which imply:

(12.72)

This indeed shows that C 1 {kfm,fl) plays the stlJ1le role as Zi 1 (Afm,g).

l2.6 SDE beyond Leading Order

ln thls section also all qua.ntíties except the operators are aasumed to be renonnalized.
The product </J(x + vf2) <!>(x- v/2) is not only a regularization of </J2 (x) by poínt split-
tbog but also of operators of higher dimenslons which can be obt.ained by differentíation.
At next orMr, which means taking into occount in expansion (12.37) ali operators of
diPoensioilS 2 and 4, the SDE of <P (x + v/2) ,P (x- vf2) is an expansion of a regularize<!
operator aí dimension 4 on a bMis of renormalized operators of dlmensions 2 and
discussed in previous sectlaD$. lndeed l!!t us differentiate expansion (12.37) twice
respect to v:

[<~> (x + ~) â,&~p (x ~) +<~>(x ~) a,o.<P (:.:+i)- o,<!> (x + ~) Ôv</J{x 7 ~)

-8,4> (x- ~) a.q. (:t+ ~)J L:a,a.C.,(v)O~ (</>(x)). (12.73)

product in the r.h.s. can be considered as a form regularized by point splitting of

)8,8.</>(x) B,,P(x)8.rj>(x)) which is a linear combination of operators of dímen-
and spins O and 2. In Section 12.1 we have discussed the renormaliza.tion of
of dimension 4. and spin zero. The operators of spin 2 introduce two new lin-
independent opera.tors which can be chosen to be the traceless part of â,p{x)ô.</> (x)
â,â., (rj>2 (z)). R.otatiOU invariance in space tells 11S that operators OÍ different SpÍn
not mix under renormaliza.tion. Therefore in a.ddition to rela.tions (12.9) we ha.ve:
o~ {</>(r)) ; Ôpr)>{x)âv<P (x) - !li,. (â,P(x)) 2
= zr;• {05 (<l>(x))), B 5 (iJ,â. !6,.. ~) (<i>2(xl),. (12.74)

can impose the renormallzation conditions:

rg! (q;p1,P2l"" -WP2 + P'íz:il- ili,.p,p,) +o (p4 ) . (12.75)

rg; (O; o, o, o, o) o.
266 Renormo.!izo.t:ion af Compoaite Operotors 12.7

We can now use exactly the sa.me strategy as for the leading term in the SDE. We
insert expamion (12. 73) truncated by omitting aU operators of dimension larger than
4 which correspond to vanishing çoefficient.s, in the 2 and 4-point correlation functions.
We go over to the 1Pl correlation functions in the way explained in Section 12.3, i.e.
ronsidering the product of fields in (12. 73) as a oomposite opetator. We finally use the
renormalízatíon conditíons (12.34) and (12. 75) to determine ali the roellicients ô~8" Co (v)
in the truncated SDE.
Note that we have lost no informatíon by dífferentiating twíce sínce only C 1(v), which
hns been already determíned, hns a constant part when 1"1 goes to zero. In thís limit
the coefficients 8.,ô"Ccr(t1) have singularities in v which are similar to the divergences in
A/m of the renormalization constants which appear in the expansion of operators O~ on
a basis of dimension 4 renonnalized operators. To obtain the asymptotic behaviour of
the coefficients C.,(v) we then have to establish RG equations for them, by introducing
the SDE in tbe CS equations for IPI correlation functions. 'We have to worry about tbe
SDE expan.sion in presen<.'e of a ,P 2 (x) insertíon. We have to use the analogue of equation
(12.10), the difference being that the new renormalization constant wbic:h appears in
front of the contact term now is a function of v.
We conclude that the coefficients of the SDE satisfy RG equations which are formally
identical to the relations between the renormalízation constants and the finite RG funo-
tions which arise in the CS equations for the operators of dimension 4 and spins O and
2, ín complete analogy with tbe oorrespondence between equations (12.11) and (12.72).

12.7 SDE of Products of Arbitrary Local Operators

For general local operators A and B we expect:

A (X+ -"2) B (~~ _v2) "c"

L., AB (v)O,a( X }, (12.76)

in which at any finite arder in perturbation theory:

C_4 8 (..\v)- ,),1°"1-JAJ-IBJ for ..\ ..... O, (12.77)

up to powers of In>. and the O~(x) forma complete basis of local operators.
Let us take the example of:

The product .p2 (x + t•/2) 4> 2 (x v/2} is a regularized form of q,• (x), however .p 4 (x) b}'
renormalization mixes with ali operators of dimension 4 and 4>2 (x). P<>wer counting tells
us that among these operators ~jo2 (x) has the most dívergent coefficient. Tberefore at
leading order:

(12.78)

in which the coefficient can be determined hy using the renormalization condition for
(.P,),:
 12.8 Renormalízation of Composíte Operotors 261

a.nd expressing tbat equation (12. 78) should be exact when inserted in the 2-point func·
tion at the subtraetion point where ali momenta vanish.
It is then easy to derive RG equations for this new coefficient by inserting the relation
in the cs equatíoru; for the proper vertices rtl,n)'
We do not wish to go ínto further detai! sínce the discusslon is very technical. The most
important idea t.o keep in mínd is the complete parallelism between the SDE of operator
products and the renormalization equatíons of the corresponding composite opera.tors.

12.8 Light Cone Expansion (LCE) of Operator Products

After állalytic continuatíon to real time, the length of the vect.or squared x 2 may va.nish,
while the vect.or x,. remairlS finite. In such a sítuation tbe relevnnt cxpnnsion for a
product of fie!ds is no longer the SDE but instead the light cone expansion (LCE).
lt is necessary to c!assify ai! operators not only according to tbeír canonical dimensíons,
but also their spin s which characterizes their trBllsformation propertíes under space
rotatíons. The LCE tll.kcs thc form:

tP (x + DtjJ (x ~) =LC~
,,.
(v 2
) ~1 • ,.,(u)q.·: .• ,(x}. (12.79)

The polynomial P;, ...... (v) is a homogeneous, traceless for s > O, polynomial of the
vect.or v, and the operators O!f...,.. form a complete basis of local operators.
For example:
P;1 ,J.t~(v) = 2
Vp. 1 V,u 2 - Jv 6~,,~.~.~
When u~ goes to zero with v~ finite, the polynomials P' have a finite limit and there(ore
C! (u2 ) contaln the whole non-trivial dependence in u2 • The Malysis
tite coefficients
Riready perfurmed for the SDE can be extended lLDd shows that in perturbation throry
c~ (if) behaves like:

(12.80)

'l'herefure the importàllt parameter is no longer the dimension of the operator o··~ but
n quantity called tbe twist -r:
T = (O'•"] S. (12.81)

The operators of Jowest twíst dominate the LCE of product of opera.tors.

In expansion (12.79) the lowest twist is 2 which is the twist of ,P2 (x). Each new factor
· .j>(z} increases the twist by one unit, while edditional derivatives eíther increa.se the twist
·. br leave it unchanged.
Therefore the most general operator of twlst 2 has the form:

Ql(:r)(â,., ... 8,., ... a,..- traces)q\(x),

:·or ís a combination of derivatíves of a twíst 2 operators.

Operators of twist 2 and spin s, sjnce they are the operators of Iowest dimension for a
· giwn spin, renormalize among themselves. Using previous consideratiol18 about the SOE
I~ is easy t.o write RG equa.tions for tbe corresponding coeffidents C~ (if).
 Renormcúization of Composite Opemtors 12.8
268

Bibliographícal Notes
The renormalization of compooite operators is discussed by:
W. Zímmermann in Lectures in Eiementary Particles and Quantum Field Theory al-
reedy quoted in Chapter 10.
Tbe short distance expansion was postulated in:
L.P. Kedanoff, Phys. &t!. Lett. 23 (!969) 14:>0; KG. Wilson, Phys. Rev. 179 (1969)
1499, D3 (1971) 1818.
RG equations for the leàding coef!icients of the SDE are established in:
K Symanzik, Commun, Math. Phys. 23 (1971) 49; C.G. Callan, Phys. Rev. D5 (1972)
3202.
A thorough investigation is carried out in:
W. Zimmermann, Ann. Phys. (NY) 77 (1973) 536 and 570.
See a1so
C.G. Callan in Meth~ in Fie!d Theory, already quoted in Chapter 9.
The extension of Wilson's work to the light cone expansion is found in
R.A. Brandt and G. Preparata, Nud. Phys. B27 (1971) 541.

Exercioes
Exercise 11?.1
12.1.1. Study the renorma!ization of the q,3 field theory in six dimensions,

and derive the renormalized equation of motion.

12.1.2. Establísh the corresponding Callan-Symanzik equatíons.

Exercise 12.2 ··"'·

In the <P~ field theory, write explidtly the relation (12.33) in the case of dimensional -:;::
regule.rízation and mínima! subtroction. Obtain two other equations by differentiating
the functional integral with respect to the coupling constant and tbe mass, and determine
the RG functions of CS equation (12.15). _,:
 SYMMETRIES ANO RENORMALIZATION

Up to now we have discussed the problem of renormalízation froro the sole point of view
of power counting. In Chapters 13-21 we shall discuss the consequences for renorma!U.a-
tíon of symmetries of the action. Indeed when the tree levei action has some symmetry
properties, 11\'e expect that the renormalized action will not be of the most general form
allowed by pure power cnuntíng arguments but will instead keep some trace of the ini-
tial symmetry. Technica.lly this means that, as a consequence of the symmetry, the
divergences generated in pertmbation thoory are not of generic form, alld therefore the
renormalizatlon constants are not ali independent. In this chapter we deal only wlth
linearly rtalízed continuous 8)Jmmetries correspondíng to compact Lie groups because
they leed to very interesting formal properties; consequences of d!screte symmetries can
ttlso be studied with, howewr, somewhat different methods (soe the remark at the end
of Section 13.5). Also we consider beltJW only infinitesimal group transfotmations and
t.herefore t.opological properties of groups will play no role.
Our general strategy will be as follows:
(i) We fust introduce a regularization whích preserves the SYIDffietry.
(ii} We then prove identities, generally cal!ed Ward-Takahasbi (WT) identitíes, come-
quentes of the symrnetry of the action and satisfied by the generating functional of IPI
correlation functions.
(iií) These identities irnply relntions betw.e€n the divergences of correlation functions
. nnd thus between the counterterms which render the theory finíte. From these relatioDE
we derive the generic form of the counterterms. Such an analysis is based on a loopwlse
m:pansion of petturbation theory.
(iv) We finally read off the properties of the renormalized action.
More generally non-trivcial identities survive when terms in the action induce a •oft
of symmetry. We specifica.lly consider the examples of linear symmetry break-
the important limiting case of spontaneous symmetry breaking, and quadratic
breaking.
in Section 13.6 we apply the formal.ism to the pbysical exa.mple of chiral sym-
breaking in low-energy effective models of hedroru;.
In the appendix we outline the relation between WT identíties and current conserva-
We derive the energy-momentum tensor and dilatation current.

entering the discussion, we describe our notations and oonventions for group and
algebras. A few additional algebraic rcmarks conrerning the reprcsentation of the
algebra in terrns of dilferential operetors are also usefuJ.

consider continuous symmetries corresponding to various Lie groups and algebras.

context we adopt the following set of conventions except if explidtly sta.ted other·
for continuous symmetries we only consider compact Lie groups, Therefore we can
270 Symmetries and Renonnalization 13.1

represent the generators of the corresponding Lie algebras as N x N real antisymmetric

matrices tn. The trace of two antisymmetric matrices defines a se alar product. We can
use it to normalize the matrices by:

(13.1)

With thís convention the Lie algebra skucture constants !arn:

(13.2)

are completely antisymmetríc in the three índices. Tbe basis of the Lie algebra ís fixed up
to an orthogonal tra.nsformation. In the special case o f unitary groups, we also oometimes
represent the generators by hermitian or anti-hermitian matrices (this will be a matter
of convenience) and then normalize them by

tr tnt~ = Né,p,

(in the hermitian case).

As a ooru;equence, as in the orthogonal case, tbe structure constants defined by:

a.nd thus:
f,~ 1 = ~,1mtr (t"t~f'),
2
are completely antísyrnmetric.

1S.l.fl Líe algebro and diJjerential operatoro

Let us first establish a simple group theoretical property which will become increasingly
useful when we discuss the renormaliza.tion of symmetries in more complicated situations.
Let </J; be a vec~or transforming linellily under a representation V(G) of aLie group G:

(13.3)

Let us write also the corresponding infinitesimal group transformations. In terms of the
matrices tij representing the generators of the Líe algebra, the variation 6</1 of </1 under
an infinitesimal transforma~íon (13.3) takes the form:

(13.4)

in whích w, are the infinitesimal pa.rameters of the transformation.

The va:riation o f a function S (4>) under such a transformation is then:

In particular an invarlant function S(</1) satisfies:

, as (13.5}
t;j<PJ iJrp; = 0 '
13.2 271

The differential operators ~~:

(13.6)

are thus the generators of the Lie a\gebra of the group G realioed as differential operators
acting on fuoctíons of 4J;. lt is easy to verily by direct calculation, UBin,g tbe commuta.tion
relations (13.2) of the generators tf1:

{13.7)

Note linally tbat equation (13.5) ~an be considered as a system of differential equalions
for S (,P).
(13.8)
Quite generally the commutators of first arder differential operators are again lirst Gr-
der differential operators. Therefore if a function is solution of a system of lirst order
partial differential equations described io terrns of opera.tors ~fr. it is also solutíon of
ali eqtiatiollS corresponding to opaa.tors belonging to the Lie algebra generated by ~a­
The system (13.B) ls sa.id to be compatible if no new independent equation iB obtained
from the commutators ~~fr. ~lll· This condition is verílied if a.ll oommutators are linear
combinations of tbe operators ~o, i.e. if the ~a forro a basis of tbe Lie algebra tbey
genera.te. Therefore the Lie algebra. commutation relatlons (13. 7) are the compatibi!ity
conditioJJB of the linear system (13.8).
We sha.ll be conc<rrned with cases in which </> is a field depending ou a.n a.dditional
va.riable :r, S, the action, is a functional of q,, and ~a has the typical form:

but the analysis is the same.

13.2 Linearly Rea.lized Global Symmetries

Definition. We ca.ll global symmetry a syrnJII<ltry which corresponds to a transformation

of the fields whose parameters are space-independent. More precisely let 4>i(x) be a set
of liélds transforming linearly under a repre<;entation 'D ( G) of a compact Lie group G:

4{(x) = D,i (g) <i>J("')· (13.9)

The transformation (13.9) is global if the group element g do"" not depend on the space
\-ariable x. Sometimes the expression rigid symmetry is also u.OO to avoid confwdons
with "global" in the sense of global topological properties of tbe syrnmetry group. In
what follows we explore the consequences of invariance only under infinitesimal group
tnwsformations. In tbe notations of Section 13.1.2, W1! ca.n WTÍte the variation sq, of <I>
under transformation (13.9) as:

6,P,(x) = t~IJ>i(x)wa , (13.10)

in which w., are the space-independent paramet.ers of the transformation.

A d85Sical action S ( </>) inva.riant under such a ttBIJsformation then satisfi<lS:

f dxt~f/l1 (x) 6:,~x) O. (13.ll)

272 Symmetries and Renonna.lization 13.2

Regtdarization. In the case oflinearly realized global symmetries, it is always possible

to find a regulooioation which preserves the symmetry of the action. For purely boson
field theories, we can use dimensional, lattice or Pauli-Villars's type regularizations. In
the ]atter case, we modify the propa.gator by adding to the tree action S ( q,) quadratic
im=iant terms involving hlgher arder derivatíves (Section 9.1 ):

(13.12)

in which Ais the cut-off. By introdudng enough terms it is always possible to render the
theory finite. The regula.rization terms are obviously symmetric since they are invariant
under arbitrary orthogonal transformations.
In the case af massless chiral fermions, i! the transformation Iaw involves the matrlx
75:

the substit ution

~(x)9l.P(x) ,_. ~(x)9l (1- o 1 ~ + · ) .p(x),

preserves the chiral symmetry.

Derívation of WT identitie<. Let us naw consider the generating functional of corre-
lation functions Z(J) corresponding to the symmetric a.ction S(r!>):

Z(J) =f [d,P] exp [-S{rp) +f dx J; (x) ,P;(x)]. (13.!3)

We have shown in Seetion 5.4 that general identities satisfied by the generatíng functional
Z(J), Iike the equation o! motion, can be obtained by expr<:ssing that the functional
integral is invariant in an infinitesima.l change o! variables. We here use thís observatian
to derive the consequences o! equation (13.11) for Z(J). We perform a change of variables
o[ the íorm of a transformation (ll.lO), setting:

(13.14)

in the functional integral (13.13). As a consequence of the symmetry as expressed by

equation (13.11), the action S (</>) and therefore the regularized action S/1. (</I) are left
invarlant by the transformation (13.14). The measure of integration [dr/>;] in the func-
tional integral (13.13) is the fiat euclidean measure which is invariant in an orthogona.l
transformation (more generally in any unimodular transformation, i.e. corresponding to
matrices of determinant 1). The on!y vl'lriation comes from the source term. This implies:

O= 6Z(J) =f [dQl']ó ]source term]exp [ ~S (ql') +f dxJ;(x)r,6:(:r)]

The variation af the source term is:

6 [source term] = f dx J;(x)t?,<Pj(x)w.,.

13.2 Symmetries and Renormalitatúm 273

Tb.is leads to the equation:

O= w, f f[dq,j dx J,(x)t~<,/lj (z) exp [ -S {'/>) + f dx Jk (x) <Í'k(z)] . (13.15)

We now have repla.ced the notat.ion <P: by <,/1; since <fJ is a dummy íntegration va.riab.Je.
Equation (13.15), being valid for any set of parameters w,,
cao be rewritten for eacb
component a. Finally the identity:

f [d.p] <Ji;(x) exp [ -S (<Ji) +f dy Jk (y) <Í'k (y)]

é f [d<fl) exp [ -S (<Jl) +f dy J• (y) q..{y)] , (13.16)

al!ows us to rewrite equation (13.15) as a.n identity for the functional Z(J):

f " '
6Z(J) =0
d:tt·3 J;(x)--
6J1 (x) ·
(13.17)

From equation (13.17) we immediately derive an equation for the generating funetional
W(J) =In Z(J) of connected correlation functioJJB:

(13.18)

Expanding equation (13.18) in a power series of the source J(x), we obtain identities
between the conneeted correlation functions whicb describe the ph)'llícal implications o(
the symmetry of the action.
However for renormalization purp""" it is more convenient to derive o.n equation for
the IPI functional r ('1'). We therefore perfotm a Legendre transformation:

=f
l
r( li')+ W{J) dxJ;(x)ll'; (x),
(13.!9)
éW
ll'<(x) hJ,(x)'

and immediat.ely obtain the equatlon:

(13.20)

whicb, expanded in powers of <p, yields WT identities for proper vertices. The equation
ímplies t.hat the regularised functional r (li') is invaxiaot undex the trans!ormation (13.10).
RenormaluatioTL We IJOW perform a loopwise expansion of r (11'):
00

r(<p) = E r, (li') gl. (13.21)

I =O

The parameter g is any coupling constant pla)ing the role of 11. and introduced to order
the loopwise expansion. Sin~ equatio~ (13.20) ;,; linear ín f'( 'I') and independent of g,
ali functionals r, (li') also satisfy equation (13.20).
274 S!!"'metries and Renarmalizatian 13.3

The functional f o ('F) is just the action S (rp) and satislies by assumption equa.tion
(13.20). The regularized one-loop functional r 1 (I") satislies (13.20):
~ .sr,('l'l
f d:r t,j\O<(x) Ó\Oj (x) O. (13.22)

We can now perform a.n asymptotic expansion of r,

{I") in terms of the regula.rizing
parameter (large cut-off expansion or lfê in dimensional regularizat.ion for example).
Because equation {13.22) is valid for any va.lue of the regularizing parameter, it is va.lid
for each term in the expansion and thus for the sum of the divergent contributioru;
ryiv (I"):
1
6Ti
" (<p)
f a
dxt;;'Pi(x) Ótp;(:r) ~O. (13.23)

General renormaliza.tion theory tells us that ri'• ('I') is a general local functional of the
lields restricted only by power counting; eqnntion (13.23) tells us in eddítion that it is
symmetric. Adding -r~i• ('P} to the actíon render.; the theory one-loop linite. The one-
lonp renormalized action is still symmetric and therefore the new two-loop functional
f 2 ('P) stiU satisfies equation (13.20). After one-loop renonnali7.ation r 2 ('P) has only
local divergences which also satisfy equation (13.23) and ali arguments can be repeated.
It ia clear tbat the arguments extend t.o ali orders.
The conclusion is that the renormalized action S, is the most general local functional
of the lield q,,(x) compatible with power counting and invariant under the translormation
(13.10).
Each reader familiar with perturbative calculations will of course have realized that
thls is a pedantic derivation of a. rather obvious resu]t. However, since the sarne strategy,
suitably adapted, a.llows us to dlscuss much more general situations, we believe that it
has been useful to expose it lirst in a case in which it can be easily understood.
Note fmally that we have renormalized using aminimal subtraction scheme. Additional
finite renormalizations which are consistent with the symmetry can still be performed.

13.3 Linear Syrnmetry Breaking

For some applications (see for example Sections 13.6, 26.5) it is useful to consider the
following situation: the action S (c/>) is the sum of asymmetric part S.ym (cJ>), Le. irrvariant
under the transformation (13.9), and a term breaking the symmetry linear in the fields
1/J;(:r):
s (cJ>) = s.,m (cJ>)- f c;.;{!;(x)dx' (13.24)

in which c is a constant vector.

An example of such a situation ls provided by the action:

S[cJ>] =f d:< [! (8~1/1) 2 + ~m'c/>2 + ~9 W) 2

c' cJ>] (13.25)

in which ,P(x) is a N-componentvector. The action S (I/I) is the sum of anO (N) inva.riMt
part and a linear symmetry breaking term.
The perturbe.tive expansíon correspondingtn action (13.25) is obtained by the following
method: one fitst looks for a classical minimum of the action which corresponds to a
constant field vo satisfy:ing:
o, (13.26)
13.3 Symmetrm and ReMrmalization 215

with tbe condition:

in the mat:rix seru;e.

In the example (13.25) tio; satisfies the equation:

(13.27)

!f the action h as severa! mínima, one is in general instructed to choose the absoiute min-
imum of the potential but this is irrelevant from the point of view of formal perturbatíon
theory. The quantity v 0 is, at the tree arder, the expectation value (vacuum expectation
value in the partide phys.ícs language) of the field q,.
One then translates the iield q, setting:

q,(x) vo + x(x). (13.28)

After translation, the action no Jonger contuins a linear term and the perturbative calcu-
lation proceeds in the standard manner. However the examp]e (13.25) shows that after
tráDSlation tbe mass term is no longer symmetric and a non-symmetric x3 interoction
has been generated. Conelation functions will no Jonger be symmetric and the form of
the UV divergences from the point of vjew of the symmetry is a pnori unknown. Jt is
thus irnportant to understand whether the structure of the renormalized action reflects
in some woy the structure of the actlon (1:'..24).
The answer here follows ftom a. very simple argument. With om•ious nota.tíon ""' h ave:

Z (J) = z.,... (J +c), (13.29)

and thus:
W(J) W,ym(J+c). (13.30)
Equation (13.18) then in particular implies:

(13.31)

Expanding in powers of J, (x), we obtain a set of rnlations (\\'T idenbties) between

connected correlatlon functions which can be most conveniently expressed in momentum
representation:

c,t:jWJ~.:.~l.k. (O, p,, ... ,p,.)

(13.32)
r= I

Let us now perform the Legendre transformation:

J
I
r(<p) +W (J) dxJ,(x)\Oi(x),
(13.33)
6W 6W">'m (J +c)
\Oi( I)= 6J,(x) ~ .U,(x) '
276 Symmetries and Renonnalization 13.3

Let us compare these relatkms with those of th~ symmetric case:

f
l
f.oym (Ü + W,ym (J) dx J;(x)Ç;(x),
(13.34)
~ ( ) = IÍWsym (J) .
'x 6J;(x)

Replacing J;(x) by J; (x) +c, in relations (13.34) we obta.in:

f + e;)<p;(x),

l
r,ym ('f')+ w.,m. (J +c) dx (J;(x)
(13.35)
·( ) _ óW,ym (J + c)
<p, x - JJ,(x) '

and therefore comparing {13.33) with (13.35):

(13.36)

This identity shows that the divergences of the functíonals f(~.p) and foym (tp) are iden-
tícal. lf we therefore replace the regulàrized symmetric action by the renormalized sym-
metric oction, the theory is finite for any value of c;. This casually is expressed by saying
that the linear breaking term ls not renormalí:.ed.
Of course to obtain the IPI correlation functions of q,., we h ave to translate <p by the
4> lield expectation value setting (see Sectíon 6.2):

'f;(:z:) = "' + X;(:z:), (13.37)

with:
éf
0 'Pi(x)
I 0 ""'* 6f,ym (v)
IÍ<p,(x) '
(13.38)
'ó'>(rl=•;

and tPr (v) f6rp;órp1 ~ o.

The IPI correlation functions are then the coefficients of the expansion of f {~.p) ín
pmvers of X· In th.e tree approximation one recovers Vt vo.-.
The WT identities for f(~.p) are immediately deduced from the identity (13.20) for
rljloym:

f dxt~ [ 6:.~x) + e.J 'P;(x) =o, (13.:>9)

which after the translation (13.37) be<:omes:

(13.40)

Application. Let us show that this identíty leads to some non-trivial relatians between
the IPI correlatíon functions. Setting X :c O we obtBin:

{13.41)
13.3 Symmdries anti Renannalization 277

which shows the breaking vector c and the expectation value v a.re left imwiant by the
snme subgroup of G. In the example of the O(N) symrnetry, equation (13.41) implíes
that the vector v is pmportíonal to the vector c.
Differentiating once wíth respect to X> (li) and settíng then x equal to zero, we relate
the 1 and 2- point functions:

(13.42)

with
2
r D(>l(x,y l = ó r(x+t1) 1
) .
ÓX;(x)6Xj (li x=O
Let us introduce the Fourier transform of the 2-poínt functíon:

.-ip(•-~) f!2) (p) '

f ddp
(2n/ "
(13.43)

In terms of f'\>) (P) equatíon (13.42) becomes:

(13.44)

This equatíon determines the goometrical structure of the zero mornentum propagator
in the presence of the linear symmetry breaking terrn.
In the exarnple (13.25), the ídentity (13.44) yields the value of the propagator of the
components of the field orthogonal to the vector c, at zero momentum:
c
f'!ll (O}
"
Equation (13.44) ís the last equatíon whlch involves c; explicitly. The terms of higher
degree in x are functions only of the expectation value v;. By identifying the coefficient of
degree (n + 1) ín x, one obtains a relation between the Fourier transform ofthe (n + 1)-
point function fln+l} with one momentum set to zero and the n-point function r<n);

(13.45)

For example this equation for n 2 reads:

!f we choose to renormalize by lixing the value of the primitively divergent corre!ation

functions at some given point in momentum space, then the set of WT identities irnplies
relations between the different pa.rameters. Apart from the vectot v, the non-symmetric
theory dependa on as many índependent parameters as the symmetric theory. In the
2
oxample of the O (N) symmetrk (4>') field theory in four dimensiono, it is possible
to impose one arbitrary renorrnalization condition to rW
11 (p;) and two conditions to
ri~ (p). All others are given by the WT identities (13.45) used for n = 1 to 4.
278 Symmetries and Renonnalizatian 13.4

13.4 Spontaneous Symmetry Breaking

Spontaneous symmetry breakíng {SSB) is a possible límit of linear symmetry breaking

when the breaking parameter goes to zero. As we discuss below, in this Hmit the action
becomes symmetric, but depending on the va)ues of other pa.ramet<lrs (in our exa.rnples
the coefficient of the r/;~ operator), the physícs may or may not be symmetric. Many
physical models in particle physics are bàSed on the concept of SSB. The reason is that
the mechanism of SSB allows one to construct models with broken symmetries which
depend on a.s many parameters as the symmetric models. The appearance of massless
partides is in general the most cha:racterístic feature of such models (in the absence of
gauge symmetries ).
In statistical mechanícs alro many phase transitions are related to SSB, as we shall
extensively discuss in the Chapters 23-3() devoted to criticai phenomena. We shall de-
termine when, from the functional integral point of view, we should expand perturbation
theory a:round one o f the degenerate rninima of the c]asskal potential, situation of spon-
taneous symmetry breaking (SSB) and degenerncy of the quantum mechanical ground
state, and when ít is necessary to sum over the contríbutions of ali mínima, in which
case quantum fluctuations restore the symmetry broken at the classical leve), the ground
state being unique. Below we assume that the sítuation of SSB ís realized.
Classica.l analysis. Let us Jirst consider the example of the O( NJ symmetric modeJ and
discuss the expectation value of the Jield at the tree leve(. The action per unit volume
for a constant field </> is:

V(</>) S(</>)/volume (13.46)

As Jong as c does not vanish, ít ís possible to pass continuously from a situation in which
the parameter m 2 is posítive to a situation in which m 2 is negative (wmember that m
is no longer the physical mass) without encountering any singularity. For instance the
expectation value v is at c fixed, a regular function of m 2 at m 2 O. If instead c
vanishes, the expectation value v vnnishes identically for m 2 > O !md takes a non-trivial
value ror m 2 < o such that:
Jvl= (13.47)
as can be easily understood by drawing the potentíal for both cases (see figure 13.1).

V(~) V(ql)

Fig. 13.1 Section of the qlcpotential.

r• Symmetries o.nd Reno!:1lla!Uation 279
13.4

In the latter case the dassical mínhnnm of the potential i.s degenerate. Startíng fwm
!· a gíven minimum, it ís possible to describe ali other mínima by acting on the vector v
with the >ymmetry group G. In the O(N) example the surface af mínima ís a sphere
with a radíus given by equation (13.47).
Assuming a sítuation of SSB, we construct a perturbation theory around one mínimum
v which is, at leadíng order, the field expectation va.lue. We thus shift the field

f/>(x) =v+ x(x).

The x-field mass matrix is obtained by calculatíng the second derivatíves of the potential
at the minímum. Using equation (13.47) we find

This matrix has N 1 zero eigenvalues corre>ponding to eigenvectúrs orthogonal to v.

Thú; Í5 not surprísing since the potential ís Rat along a group orbit. The physical conse-
quence is that spontaneous breaking of a continuous symmetry implíes the appearance
ofm!ll>s!ess (Goldstone) modes: from the poínt of view of partícle physics massless scalar
partícles called Goldstone bosons.
Let us now examine a more general sítuation. We assume that a G-syrnmetric action
has degenerate miníma. We cal! v the minímum chosen to expand perturbation theory,
and thus the field expectation va.lue at !eading arder. We introduce the subgroup H of
G, little group (stabi!izer) of the vedar v, i.e. the subgroup of G which leaves the vector
v inV!lJíant. By definítion the p generators of the Lie algebra, C( H) of H satisfy:

.C(H) :

We denote by .C(G/H) the vector space (it ú; not an algebra!) generated by the comple-
rnentary set in the Lie algebra .C(G) of G. It is chr.racterized by:

.c(G/H) : L t;jvJw<> = O '* wa O for ali a .

<»1'

For a > p, the vectors (v~), = tljv; thus a:re linea.rly índependent. We then pa.rametríze
the field 4> in the forrn of a group element a.cting on a vector:

4>(x) ~ exp (2::

<>~P
l
0
{
0
(x)) (v+ p(x)) =v+ {"(x)t"v + p(x) + · .. ,

in which p(z) hos components on!y in the subspace orthogonal to ali vectors t"v. In the
O(N) example p ha.s only one component along v. This parametrizat.ion is such that
the mapping of fields {p(x),{"(x)} - f/>(x)- v can be inverte<! for small fields. This
property ensures that if the l!uctuatioliB of the field 4> around it. expectation va.lue are
in some sense small, perturbation tbeory ú; at least qualltatively sensíble.
Inserting this parametrization into the actlop. we note the fol!awing: the contributions
to the action which are derívative-free depend only on p(x) becaure tbey are G-invariant.
The dependenoe in the fields { 0 (x) ú; enti:rely contained in the terms wíth derivatíves,
therefore these fields are massless. We conc)ude that spontaneous breaking of symmetry
280 Symmetries and Renormalization 1:'1.5

of agroup G to asubgroup H, the group which Jeaves the field expectation value ínvariant,
yields a number of massless Goldstone modes (bosons) equal to the nomber of generators
of G which do not belong to H. This result is valíd ín the classical approximatíon. Let
us now generalize ít to the fui! quantum theory.
WT identitíes and spontaneous symmetry breaking. To connect contínuous!y the two
pbases, symmetric and with SSB, without encountering any singularity we start from the
sítua.tion m 2 > 0 1 c O, one gives to c a non-trivial value, perform the continuaUon trom
m 2 > O to m 2 < O, and agaín take tbe vanishing c limit. We then assume the existence
of non-trivial solutions to the equatíon

(13.48)

In Section 6.6 we have explained hnw the existem:€ of wlutions to this equation is <:on-
slstent with the convexity of the function r( v) (equation (6.28)}.
Sínce the WT identitíes (13.40) hold for any value of the parameters and we have made
an analytic continuation, we still have in the m 2 < O, c =O limit:

6r(x+ v)
f u
dx t,; <~x,(x) (x3 + vJ) o, (13.49)

the dírection of v, being fixed by equation (13.41 ).

Go/dstone modes. L«t us here discuss only one impartant consequence of the WT
identitíes. Equatíon (13.44} becomes:

(13.50)

To explain the significance of this equatlon, inspired by the classícal analysis, we introduce
the subgroup H of G, little group (stabilizer) of the vector v. Sínce for a > p, the vectors
(u 0 ); tí';v; are linearly índependent, the equation (13.50) implies that the matrix f 1;(0)
has as many elgenvectors with eigenvalue zero as they are generators in L (G /H). The
carresponding components of the field are Nambu--Goldstone modes assod11ted with the
spontaneous breaking of the G-symmetry, confirming the classical 11nalysís.

13.5 Quadratic Symmetry Breaking

A syrnmetry may be broken by terms of higher canonícal dimensions. We sha!l examine

in detail only the cose of the qui>dmtic symmetry breaking. We shall then brielly indicate
what happens with breaking terms of even hlgher dimensions. Let us consider the actíon:

(13.51)

in which llij is a symmetric traceless constant rnatrix. We can assume /Jij traceless
without )o55 of generality since a term proportíonal to the unít matrix can always be
absorbed into the symmetric actlon S,ym (4>). In action (13.51) interaction terms are
symmetríc but the mass t.errns break the cymmetry.
We can try to again derive WT identities by performing an infinite5imal change of
varíablo;s in the functional integral:

tPi ,p; + t':;w,t/>j .

13.5 Symmetries and RenormaCizatíon 281

The variation of the integrand now comes both from the source term and the breaking
term:

Ex.pressing as usual that the result of the functional integral is not modified by a change
of variables we obtain the equation:

Using equation (13.51) and replacing factors of the forro rp;(:r) by b /lJJ;(x) wederive an
equation for Z(J):

(13.52)

Two features distinguish this equation from the equa.tion of the synunetric case (JL;; 0):
(i) It involves functional second derivatives with respect to the sources. The corre-
2
sponding equation for W(J) then also involves a term of the form ó2Wj {6J(:r)) • ]f we
now try to perform a Legendre transformation, we have to introduce the qua.ntity:

{13.53)

the inverse being understood in the sense of kernels. The WT identities ta.ke a very
comp]jcated form.
(ii) lt is no longer a relation really ~n 4>-field rorrelation functíons because tbe
two Iunct\onal der\vatives are taken at tbe sarne point; \t insteed also involves insertion.s
of the composite operator ,P.(x).P;(x).
The dilficulties we encounter have several orlgíns. One ca.n be directly understood by
forma!ly expa.nding the corre!ation functions in a power series of the symmetry breaking
term. This generates a sum of symmetrie correlation functions with multiple insertíons
of the operator ~I'<! f dx ,P.(x)<PJ(x) {see figure 13.:2).

··.

.•.. Fig. 13.2

These inoertions, as we have already extensive!y discussed, may generate new diver-
gences which h a"" to be taken care of. This situation has to be contraste<! with the linear
case, in which only fiakl correlation functions are generated by such an expa.nsion,
Another difficu]ty stems from the fact that an infinitesimal group tra.nsformation gen-
erates a new quadratic term linearly independent of the original one, as can be seen on
282 Symmetriu and Renormalization 13.5

equation (13.51). Irwe want to write WT ídentities for renorma.lired quantltíes, we have
to also renormalize the operator f dx p.,1tjkt/>; (x) l>k(x) and examine how it transforrns
under the group. This may generate new quadratic operators and ali the operations
have to be repeated. We therefore have to adapt our strategy to this new sítuatíon. We
shall use a slightly different method dírectly inspíred from the ll.lla.lysis of the equation
of motion of Section 12.2.
The general method. The method we now introduce is fairly general and will allow us
to discuss rnany cases of renormalization with symmetries, as the subsequent cbapters
devoted to dilferent type of symmetries will show.
The idea ís to perform an infinitesimal group transformation, to collect ali new linearly
independent composite operators generated in thls way and to add source terms for them
in the action. Let us call C., (t/>) such operators. In ou r example:

Since we have now added a source for C, in the a.ction, we have to worry sbout the elfect
of ll.l1 infinitesimal trllllllformatíon on C., (<f>):

Two cases may arise: either 6C., (t/>) ls a linear c.ombination of Q);(x) and C 0 (..P) and
we proceed deriving WT identities, or new independent operators are generated and we
again add sources for them in the action.
We repeat the procedure as long as necessary, It i>; easy to verify that under these
conditions the WT identitíes will always be lirst arder dífferentíal equatiollS for the
generating functional Z considered as a functional of ali sourres. Severa! exampk.s will
illustrate this point.
Applicatian. In the example of quadratic symmetry breaking, it i>; clear tha.t a.ny
Infinitesimal transformation made on a linear combina.tion of operators of the form
<l>i(x)Q)i(x) generates another linear combination of these sarne operators. We therefore
a.t once consider the total action S (,P, K):

S(t/>,K} Ssym (<f>)+~ j dxK;;(x)t/>i(x)tf>;(x), (13.54)

with:
Kii(x)=O.
Actually we need only insertiollS at zero momentum and we could in principie restrict
ourselves to comtant sources Ki;. However it il; not more dillicult to use spa<:e-dependent
sources. Furthermore. in the c..se in which the symmetric theory is massless, zero mo-
mentum insertions could Iead to IR divergences which we avoid in this way. We now
coru;ider the generatíng functiona.l Z (J, K):

(13.55)
13.5 Symmetties and Renqrmnlization 283

An infinitesimal change of variable (13.10) Jeads to:

O= f [d,P]6 0 [-S(<J>,K) +f d:cJ;,P;] exp [ -S(,P,K) +f dzJ,.p,], (1:1.56)

with:

é., [s(.p,K) f dxJ,,P;] =f dxtij<P {x) [K; (z)4>• (x)- J;(x)].

1 0 (13.57)

As before the product </>i(z)ifl•(x) appea.rs. However we are now abte to express ít in
6joK.1(z) instead of ti' fDJi(x)óf(JJ•(x).
terlll6 of

Rff'rlark We define derivatíves '1\ith respect to cornplicated objects (here the symmetric
traceless matrix K;;) in tbe following way; ]et F{K) be a function (or functional) ofK;
we calcu!ate the variation of F (K) at first arder wben K varies of a quantity 6K,

(13.58)

The derivative is then defined in the sense of diff.,.ential goometry: it is the linear operator
acting on 6K;5(x) in the r.h.s. of equation (13.58).
In the example we are considering here, dífferentiation with respect to Kij generates
the trru:el"'-"S part of tbe product r/>;1>J· Sin~ equatíon (13.57) involves only the traceless
part of this same product, the definition (13.58) a.Jtows us to rewrite equation (13.56):

f dx{ [tijK;.(x) + t?kK;;(x)j tK.~ (x)- J;(x)tij ,1J,6(xJz {l,K) O. (13.59)

Because equation ( 13.59) is a Mt arder differential equation, an identical equ a.tion holds
for W(J,K).
The U.gendre transforrnation is only performed with respect to the •aurce J;(x), be-
cause the reducibility corresponds only to externa! q, Unes.

r {<p,K) + W (J,K) f dxl;(x)<p,(x),

6W(J,K)
~·

The sources K,;(x) do not participate in the U.gendre transformation and have to be
considere<! as externa.l parameters. This then immediately implies (equation (6.20)):

6W
6K;j(X) J
I (13.60)

This relation provides an a.dditional justificatíon for the rnethod we have proposed. With
sources for the composite operators, the U.gendre transformatlon becomcs strníghtfor-
ward. The WT identity for r ('I', K) obtained by U.gendre transformation from (13.59)
is then:

(13.61)
284 Symmetries o.nd Renonnalization 13.5

This equa.tion ha.s a very simple interpreta.tion: r (cp, K) is invariant under the double
transformation:
Scp;(x) = tijwa<,Oj(x),
(13.62)
{ 6Kt;(x) ""'Wa [tj,K;~:(x) + t~;K<;(x)] .

By performing a transformation both on </> and K, we have rendered the breàking term
f J dx K;; (x) .P,(x).P,(x) group invariant, lt is then almost obvious that the symmetry
of S {<;\, K) under the equiwlent of tbe transformations (13.62) implies the WT identitíes
(13.61).
Using the arguments gíven for the case K = O, we can shoW that if tbe regularized
functíonal r (cp, K) satisfies {13,61), tbe renormalized functional r(<p,K) and tbe renor-
malized action will satisfy the same ídentity: the renonnalízed a.ction S, (q,, K) is the
most general local functiona! af <f> and K compatib!e with power counting and is invari-
ant under tbe group transformation ( ll.62).
For a .;.4 -!ike field theory in four dimeusions the action is the integral of a local functinn
of dimension four, the field <P has dimension [.P] = 1 and the dimensíon [Kl of the source
K is two because it ís coupled to an operator of dimension 2 (see Sectíon 12.1)

11'] = 1, IKJ = 2. IS{4l,K)l=4.

The r€normali.ed a.ctioo S, (</>,K) has thus the general form:

S, (tfl,K) IS•ym]r (q'>) +~f dz K;; (x) A;; (rp(x))

+ ~b;;,kl f dxKü(x)Ku(x), (13.63)

in whích A;j (</>) ís a local derivati....,.free polynomial of dimension 2, and b;j,kl is a set af
constants.
Constraints on A;; (</>) and b,;,kl are obtained by expressing the invaríance of the renor-
malízed a.ctioo under transformation (1:>.62). In the símplESt case A;; (oi>) has the forrn:

(13.64)

in whicb N ís the number of components of q,.

Dependíng on the representation content of </>;(x), ít can be a linear combination
of severa! dimension 2 opero.tors and also conta.ins a contribution linear in ifl. The tenn
quadratic in K,; does not depend on ,P and can be factorbed in front of tbe functlonal in-
tegral. lt gives an additive contribution to W {J, K). lt therefore additively renormallzes
the correlation functíon li 2W(J O) /6K;;(x)6Kkt (Y)IK-o which has the diagrammatíc
form shown on figure 13.3.

Fig. 13.3
13.6 S!!"'rnelriu and Renermolization 285

The renorma!ized a.ttion for the case of quadratic symmetry breaking is then ohtained
by setting K;i (:r:) constant
K;;(x) l'ij.
We see from the analysis of this more complex situation that the initial form of the
action is not always completely preserved. In particular a breaking term o f a given
dimension may generate new breaking terms of lower dimensions. This happens here
when A;j (4>) hBS a term linear in tj>. The renormalized theory may therefore depend
on more pa.rameters than one would have naively anticípated when adding the breaking
term.
Fína.lly let m note that, as in the CBS€ of the linear symmetry breaking, one can use
the WT identities (13.61) to constrain the renormalization conditions. Howeve:r it is then
necessary to consider together ali the super:lidally divergent correlation functions botb
of the field ;!>;(:z:} and the composite operator 4.>; (x) 4>i(:z:), and write ali relations derived
from equation (13.61) by expanding in 'I'( :r:) and K (x), in whicb they appear.
Breakin!J terms oj hígheT canoniwl dímensions. The prevíous analysis can be easíly
generalized to breaking terms of higher cnnonical dimensions. One can verify tbat in
a 4>4 -like theory in four dimensions, a cubic breaking term, since it is coupffld to a
source of canonical dímensíon 1, generates in general, by renormalization, breaking terms
quadratic and linear in tf,. Breaking terms whicb are of canonical dimensíon lower tban
the symmetric interaction are called •oft.
Fina.lly ít makes little sense in the context of a renorma.lizahle q,~ theory to speak of
a bteaking term of dimension 4. Indeed sucb a term js coupled to a source of dimension
zero. The renormalized action will contain an infinite series in the source. It is easy to
verify that then all traces of tbe initial symmetry are lost a !'ter renormalization.
Discrete "'l"'metries. Discrete symmetríes do not leed to 'NT identitíes and most of
the previous analysis does not apply. However it is easy to prove that when the tree
action is symmetric, the renormalízed octi~n remains symmetric. lf we add to the action
a linear symmetry brea.ldng term, we can expa.nd the new correlation functions in power
series of tbe breaking parameter. The coefficients are symmetríc correlatíon fupctjons.
Therefore it remains true that the counterterms which render the symmetric theory finite
renormalizes tbe theory with linear symmetry breaking. For breaking terms of higher
degree the same stcategy ce.n be applied. However we have then to renormalíze sym-
metric correlation functions with operator insertions. This question has been examined
in Section 12.1. We arld sources for th~ operators in the actiou, re=rmalize accotding
to power counting and use the discrete symmetry to constraín the polynomial in the
sources and the field. One qualitative aspect of the previous analysis survives: symmetry
breaking terms of a given dimension can only generate additional breaking tenns of equal
or lowe:r dimensions~

13.6 Application to Strong lntere.ction Physics: Chiral Symmetry

One of the most striking feature of Strcmg Interactions in low energy Particle Physics is
the observation ofapproximate spontaneously broken SU(N) X SU(N) chiral symmetries,
whicb ma.ni.fest themselv"'' in particular in the sm.all m"""'-'S of the pseudoscalar mesons.
In particular the .. meson is specially light, an indication that the breaking of the SU (2) X
SU(2) symmetry is quite small. With our present understanding this property í.s a
consequence of the small tnasses of the u and d quarks (sce for example Section A5.6)
and the vector-líke couplíng of quarks to gluons. The mass of the s quark and thus the
285 Symmetries and Renarmo.!ízation 13.6

explicit breaking of the SU(3) x SU(3) symmetry are larger as can be seen from the
masses of the K and 1J pseudoscalar mesons.
Note that, according to our previous ana!ysisl since a fermion mass opera.tor in a
renormalízabie field theory in 4 dimensions has dimension 3, the concept of a symmetry
broken by fermion mass terms ís indeed meaningful. However tbe search for analytic
rnetbods to derive )QW energy properties Of hadrons from A b:>sÍC theory Of QUatk and
gluons has up to now proven e!usive. Lattíce gauge theories (see Cbapter 33) have been
studied through intensive use of computers, but have yielded only partia) results. Finally
tbe dynamics of quarks has not been ful!y incorporated yet.
Therefore we here explain instead how one can construct phenomenological low energy
theories based on observab!e partídes like protons, neutrons, .,.-mesons... In such tbeories
tbe chiral syrnmetry ís explicitly broken by linear terms in some scala:r fields, whlch
have the transformation properties of fermion lll!LS6 terms, and which togetber with the
pseudosca]ars transform under reprerentations ofthe chíral group. We therefore f~>Ce the
situation we bave díscussed at some length ín Section 13.6.

19.6.1 The chim! symmetf'y: general structure

Let ns first consider tbe actíon for N free mru;sless fermions in even dimensions:

(13.65)

lt hns a chiral U (N) x U (N) syrnmetry torresponding to the tramdormatians:

.p' ~ [1+ ~ (1 H1s) (U l)].p, (13.66)

<b' .;p [1+! (1- ~1s)(U'- 1}], (13.67)

in wbich U is a N x N unitary matrix and the two values < = ± 1 correspond to t be two
U (N) groups.
We now couple the fermions to scalar bosons forming a complex N x N matrix M(x).
It is easy to verify that the interaction term:

-g j d"x,Ji, [~ (1 +1sl Mi; + ~ (1 -ls)M};] o/1 1, (13.68)

is invariant under the transformations (13.66,13.67), íf the matríx M transforms líke:

(13.69)

in which the two índices ± of the matrices U correspond to the two values of c, The total
a.ction satislies also rel!ection hermiticity as defined in A ppendix A5.6. It can be made
ínvruiant under a "P""" reflection P (Appendíx A5.5) by impooing to M to transform
like:
(13.70)
in whlch i ís obtaíned from :r by changing the sign of one component. Therefore the
matrix :E (M + Mt) J,/2 represents a set of scalar fields and n- (M- Mf) j,fi a
set of pseudoocalar fields. Finally under a charge conjugation C, M transforms líke:

Me M' íf d/2 is odd, and Me= TM ií d/2 is even. (13.71)

 13.6 Svmmetries and Renonnolization 287

Jt remains to oonstruct a symmetric action for the boscm fields. The octíon:

{13.72)

in which V (<p) is a.n arbitrary function of traces of powers of <p, is symmetric under
V (N) x V (N) tmnsformations. If in addition we add a term proportional to detM +
detMl, we reduce the symmetryto SU (N)xSV (N)x V (I) (thefactor V(l) corresponds
to the baryonic charge).
,.' Finally the moo general symm~try breaking term linear in the búson fields, e<>nsistent
with the discrete symmetries (13.70) and (13.71), is:

Ss(M) (13.73)

in which C is a hermltian matrix:

c ct.
Th the tra.nsformations (13.66,13.67) and (13.69) correspond two currents (for more de-
tails see Appendix Al3.1). It is convenient to consider the vector current V ~(x) which
ls assoclated wíth the Líe algebra of the diagonal subgroup U (N} of U (N) x V (N)
(U+ = U_) which conservoo patíty:

(13.74)

and the axial current A,.(x) associated wíth the complementary set of generators in the
Lie algebra, í.e. .C (V (N) x V (N) JU (N)):

-iA;(x) = -~ /SI,.'I/> + trto { [i:\,Mt, MJ+

0
+ (8,.M,Mij+}, (13.75)

The + índex means that the expressíon between brackets is an anticommuta.tor.

!f tbe matrix C is proportíonal to the identíty, the chiral symmetry is broken, but the
diagonal symmetry remains and the vector current is conserved. The axial current ís
conserved only if C vanishes:

a..v~
0
(x) = -itr{!t",Cj :E}' ( 13. 76)

a.. A~(x) tr {lt'',C)+ IT}. (13.77)

19.6.2 A special case: the linear a-model

The case N = 2 is of special physical interest. Previous analysis leads to a theory wíth 8
real bason fields. However the group SV (2) (but not the group U(2)} has the property
tha.t a representatíon and its complex conjugate are equivalent:

in whieh T2 is the usual Pauli matrix (we denote in this section the Pa.ulí matrices Ti
rather than a,, as in Appeudix A5.2, to elíminate possible confusion with the trad itional
notation for fields). Therefore M and T2M•r2 have the same transíormation law. We
288 Symmetries and Renmmalization 13.6

can reduce the representation and therefore parametrize the matrix M in terms of two
fie!ds a(,;) and 1r(:r) under the forro {see also equation (A5.30)):

(13.78)

The group SU(2) x SU(2) is the covering group of 0(4) which is also the symmetry
group of the bosonic part of the actíon. If we break the 0(4) S}'mmetry by a term linear
in the boson fields, we distínguish one direction in t,he 4-dimensional space and therefore
reduce the 0(4) symmetry to a residual 0(3) 5ymmetry. We can asoume without loss of
generality that the linear bree.king term is proportíonal to a(x). Tbe action can then be
~~ I
S = j ddz{ ~.j,!~ + g(a + i-)sT ·1T)].p + ~ ((ô~a) 2
+ (81'n)1
) +V (a 2 1
+ 1r ) - cn }, ~
(13.79)
with:
V(p) lm2P+}_.\p2
2 4! . (13.80)

Actíon (13.79) ha.s an exact SU(2) x U{l) symmetry to which corresponds the conserva-
tion of the vector current, and imp!ements the íd€a of partially conserved a:ria.I currenl
(PCAC) for SU(2). In the standard normalízation, which differs by a factor 2 from
definilion (13.75) (see equations (13.88,13.99)):

(13.81)

Fínally from equation (1:>.70) ít follows that a(:r) is a scalar field and 1T(:r) a pseudoscalar
fteld. In d 4 dimensions a and "O correspond to neutra! mesons, whíle the combínations:

conespond to charged mesons as charge conjugation shows.

19. fi. 9 Tree levei analysiõ

Bosanic sector. We discuss the pa.ttern of symmetry breaking at the e!assical leve!.
Furthermore w• consíder only the case N = 2 because ít is the símplest and pbysically
the most ímportant. In the absence of fermions we have just the q,4 field theory with
0(4) symmetry. Equatíon (13.27) gives the relatíon between the expectation va!ue v of
tbe field a and the symmetry breaking parameter c at the tree levei:

v(m 2 + Àv2 /6) =c. (13.82)

Setting:
a(,;) v+ s(x),
in action (13.79), W€ resd olf the masses of the 'Ir anda particles at the tree orde-r:

(13.83)

The hypothesis which accounts for the success of PCAC phenomenology ís that the
explicit symmetry breaking term is sma.ll and oue ís dose a situatíon of SSB. For the
13.6 SJ!ffimeiries and Renormalizatíon 289

model (13. 79) this means in partícular that m~ is sm.a.ll compared to m 0 • With this
hypothesi5 it is pos5ible to predkt some general feat.ures of the low energy ,_, scattering.
Introducing the standacd variables
2 2 2
s= (1'!+1'2) , t= (Pl+Pa) , U= (Pl+P•) , (13.84)

we can write the connected amputated ?r-lield 4-poínt function at this arder:

(13.85)

We have used the relations ( 13.83) to elimina tem and >.. The physical scattering ampli-
tude is obtained by settíng ali mo:menta on the :ma.ss shell: pf = -m;.
The expectation value v is experirnentally accessible fn>rn the wea.k r.-rneoon decay a.s
a consequence of rela.tion (13.81) and is denoted traditionally f •. Sinoe m~ is supposed
to be large compared to m~, the expression (13.85) makes qua.ntitative prediction.s for
s, t, u of arder m!, i.e. at low energy. Values corresponding to infinite O"-ma.ss are often
quoted. Although the 11'--11' scattering amplitude of course cannot be mea.sured directly,
indirect methods provide an experimental confirmatíon of the resultíng pattern.
Fennion sector. In the unbroken phase, the mass of the fermion vanishes. lt is the
boson fie!d expectation value which gives a mass to the fermions. However thís mass
cannot be predicted because the Yukawa coupling constant g, in the model, is arbitrary.
However some information about the value of g can be obtaíned fro:m experiment: At
this arder, the parameter 9 can be identilied with the roupling constant g.NN which
governs the long range part of the N-N po~entia.l dueto " exchange. We then have the
relation between physical quantitieo:

(13.86}

Tbis relation is the tree levei approximation of the Go!dberger-Treiman relation and
a.grees semi-quantitatively with experiment since:
mN 939.
91fNN 13.6. T.::eru=lo .. (13.87)

Then a.ll parameters but m. are fixed. The low energy 1r-N scattering amplitude for
cxample can be calculated. A definite predlction ean be made on]y for m,. infinite; it
agrees very well witb experimental observations.
Beyond the tree approrimation. Since the field theory model is renorma.lizable, it
ís possible to calculate Ioop corrections. Then severa.l problems arise. First there is
" questlon of principie. As we shall argue !ater, the .p• lield theory, as well as the
l.heory (13.79) wlth fermions, is most líke!y inronsistent in 4 dimension.s for non-va.nishing
coupling (see Chapters 25-29,34). More precísely, a.lthough the theory is renormalizable
in perturbation thoory, it is lmpossíble to send the cut-off A to infinity: The model
malres sense at a mass scale ,. only for renvrmalizc.ct couy:lr.gs whlch are bounded by
•·unst.jln(A/1'): thís ís the trimality prol>lem. Therefore the additlon ofloop corrections
290 Symmetries and Re!Wrrnalízatíon 13.6

is meaningful only if the momento. and the <X>Uplíng constants are small enough (in a
correlated way as stated above). A Landau "ghost" will typícally be a rnanlfestation of
this problem. Stí!l the loop corrections may be useful to improve the tree levei amplitudes
from the poínt of view o f unitarity at low energy.
r
J
Second from tbe computational point of view severa! difficulties are encountered.
(i) Loop corrections become large at moderai:€ energíes. Fbr examplE in,._,. scattering
one encounters the p resonance. Tben ít becornes necessary to apply a summation method
to the perturbatíon series. Calculatíons have been performed using the method of Padé
approximants.
(li) Since the a-partícle is heavier than 2m~, ít can decay lnto 2 pions and is therefore
unstab!e. In the true 11-11 scatteríng amplitude, it leads to síngularitíes in the second sheet
of the unitarity cut in the complex s-plane. However, at any finíte order in perturhation
theory, the síngularities assodated with the <7-partícle are on the real a.xis since the width
of the particle is a non-perturbatíve effect. Fits of experimental data seem to ímpose a
rather small <7 mass. Therefore loop corrections are affected by unphysícal síngularíties
even at rather low energy. Thís problem oí perturbatíve treatment of fields correspondíng
to unstable partícles has neoer been solved in a completely satisfactory way. One possible
idea is to make a ~ystematk large mo expansion, but the validity of the expansion is
then limited to energies smaller than 4m~, Le. very low energies,
(iií) Finally perturbative corrections to the nucieon mass are large, and thís also ad-
versely affects the positíon o f singu laríties in scatteríng amplitudes involving fermions.
Therefore, although a lot of effort has gone ínto the study of the model (!3.79), only
limited results have been obtained, beyond the simple predictíons which rely on the
geometry of the model and are therefore mostly contaíned in WT ídentitíes as we explain
below.

13.6.4 Ward-Takahashi ídentities

We have described the difficultíes one encounten when one tries to deri"" consequences
from a phenomP.nologic a! chiral action. However some re!ations are valid beyond pertur-
ba.tion theory: th~ WT ídentitíes which are direct consequenees of the broken symmetry.
Unfortunately the equatíon (13.32) shows that the WT identities always ínvolve cotrela-
tion functions with one 11-field a.t zero momentum. Therefore they would lead to relatíons
between observables on!y if the .--meson were mas$less, i.e. if the symrnetry were spon-
taneously broken. In reality it is necessary to extrapolate from zero momentum to pion
mass-shell. This extrapolatíon is mode]-dependent, and the results are only reliable if
the predictions at zero pion mass are already in qualitative agreement wíth experíment.
Let us again first discuss the purely bosonic sector.

Boson sector. The ínteresting part of the WT identities corresponds to the transfor-
mations:
6,..(x) ""-wa(x), 6a(x) w·1r(x}. (13.88)

Ca!Hng J (:r) the source for the 7f-field and H (x) the source for the a-field, we can
write the WT identities for the generating functíonal of connected correlation functions
W(J,H):
13.6 Symmetries and Renormalization 291

It ls convenient to íntroduce some addítíonRI notation to t~ke into account the residual
0(3) symmetry. Let us set (
(
w,jllPl = ó,;D.(p),
w<2l(pJ D.(p), c
(
wgl (PhP2;P3) = 6;;D.(p,)D.(p,)D.(p,)C (PI,P3;PJ),
• (4)
[ Wijkl (P!.Po.P3,P•)l - wnp- 6;j6kiA(pJ,P2,fJJ,p<) + ó,~8.;1A (p~,Ps,P3,P•)
(
(

+ 8;t6k;A (PJ.P4,P3,P3},
(
in which our conventions are that índices correspond of course to 71"-lields, and in mixed
71"-<Y correlation functions the arguments of the 1r-fields are placed first. (
By differentíating one time with respect to J.;, and setting the sources to zero, we E
obtaín the equivalent of equation (13.44):
c
t1 = (rr) = cD,{O) =cjp 2
, (13.90) (

where we have denoted by p 2 the value of the inverse of the ,. propagator at -.ero mo-- c
mentum which is now different from the pion mass squared m;. c
Differentiating once with respect to J_; and H, we obtain: (

óüW< 2
l(p) 1
~V,~l (p) = cw,~' (O,p;-p), c
(
and thus using equation (13.90): c
D; 1 (p)- D; 1 (p) = vC(O,p; -p). (13.91) (

Settíng p = O, and ca!ling now m~ the value !)f the ínverse q propagator at zero momen-
c
tum, we get in particular: (
m~ = vC(O,O;O). (13.92) (
Finally dlfferentlatíng three times with respect to J we obtain a relation between 3 and (
4-point correlation functions: (

(13.93) (
(
lt follows (
(15.94)
(
First setting p~ ..., -m! we derive frc.>m this equation Adler's oonsistency condition
(
(13.95) c
Moreover settíng fJJ =O a.nd elimínating the function C between (13.91) and (13.94)'"" (
find (
v 2 A(O,p,O,-p} = D;; 1(p) [D.(p)D; 1{p) -1]. (13.96) (
The fust term in the r.h.s. has a double zero a.t the pion mass. Usíng this property we (
recover Wein berg's relation
c
v' :!., (A(O,p,O, -p) + D; (pl)l ,.:--ma =O.
uy-
1
{13.97) (
(
292 SyrnmetrieJ and Renormalization 13.6

These equations yield model- and parameter-independent constr aínts on the 7r-11 scat-
tering amplitude, which unfortunately is slight.ly off-shell because at least one of the 11
momenta vanishes. One verlfies immediately that the funetion A in the tree approxíma-
tion (13.85) satisfies both condítions (13.95,13.97).
Another condition for the 1r-11 scattering amplitude is obtained for example by setting
ali momenta to zero in (13.96). Setting m~ D;;- 1 (0) one finds:
v2 A(O, O, O, O) = 1>2 (1> 2 /m~ - I) . (13.98)
This equation, however, involves an independent free parameter m •. Again one verifies
that expression ( 13.85) satisfies equation ( !3.98) in tbe tree approximation.
Fermion sectar. The infinitesimal transformations of the fermion fields, correspondíng
to equations (13.88) are:
(13.99)
We then call tbe sources for the fermion fields ij aod 11 . The genera.ting functional
lV(1J, ij, J, H) of connected correla.tíon functíons then satísfies the foUowing WT identil:)·:

f {2i[
dx
ó ó]
il(X)/sT llij(x) -1/(:r}"rsrli'l(x)
fj
-J(x) 6H(x)

+(H(x) +c) õJ~x)} W(1/,ii,J,H) =O. (13.!00)

The ídentitíes relevant for ph)'sics correspond to H O. The simplest a.nd most farnous
identity is obtaíned by differentia.ting wíth respect to Tf rmd i) and settíng all sources to
•ero. It is actuaily most conveniently written in terms of IPI functions:
{13.101)

The index + in the r.h.s. means anticommutator in the space of 1 matrices. We h"""
explicitly taken ínto necount the fact that the fermíon propagator ís proportional to the
identity in the group índices. This relation between the inverse nuclean propagator aod
the TI'NN vertex generalizes relatíon (13.86). lt has a physical interpretation in terms o[
the weak current under the name of Goldberger-Treiman relatíon, The r.h.s. is known
"'hen the nudeons are on mass-shell. The Lh.s. can be approxímately related to the
nucleon weak ,8-decay which involves the matrlx element of the aJ<Íal current at zero
momentum between nucloon states: since the pion ha.s the quantum numbers of the
divergence of the axial current, os can be :>een on equation (!3.81), one contríbution to
tbis matrix element has the pion pole. In the stríct chiral límlt with zero ma.ss píoos,
tbis would be the only contribution. One assumes that sínce the pion mass is small, the
chiral llmit ÍB a good apptoximation. The relation then becomes in tra.ditional notation:

Gil ~ OwNN.b_· (13.102)

Gv mN
Replacíng by experimental numbers one finds 1.22 for the l.h.s. ll.lld 1.36 for the r.h.s., a
notable lmprovement over the tree approximation ( 13.87).
More gener aliy, one can set H to zero, differentiate once with respe<:t to 'I and fi, and
an arbitrary number of times with respect to J and finally set ali momenta On mass-shell.
One then obtains model-independent re!ations (generalizing equation (13.95)) involving
amplitudes for the emission of one unpbysical pion a.t zero momentum. To determine
completely the low energy 11-N scattering amplitude, ít is however nece55axy to introduce
also the N Na vertex and the result then depends on tbe a-wass. The predktions for the
-.-N scattering lengths in the infinite cr-ma.ss limit agree weU with experimental results.
13.6 Symm€trie8 and Renonnalization 293

Bibliographícal Notes
For early deríva.tions of the Nambu-Goldstone theorem see
Y. l';ambu, Phys. Rev. útt. 4 (1960) 380; J. Goldstone, Nu.ovo Cimento 19 (1961)
155; Y. Na.mbu and G, Jona-Lasirrio, Phys. &v. 122 (1961) 345; 124 (1961) 246; J.
Goldstone, A. Salarn and S. Weinberg, Phys. Rev. 127 (1962) 965; S. Bludman a.nd A.
Klein, Phys. &v. 131 (1963) 2364; G. Jona-Lasinío, N11.ovo Cimento 34 (1964) 1790.
For a. general survey
G.S. Gura.lník, C.R. Hagen and T.W. Kíbble, Advances in Physícs, R.E. Marshak and
R. Cool eds. (Interscience, New York 1968).
The Proceedings of the Ca:rgese Summer Institute 1970 contaíns severa! contríbutions
devoted to breakíng of symmetries, WT identitíes and renormAii2Btioru;:
Gargêse Lectl!res i11 Physies vol. 5, D. Bessis ed. (Gordon and Breoch, New York 1971 ).
Among them let us quote more specifically:
K. Symanzik, who gíves a general discussíon of renormalization of symmetry breaking
terms; A.N. Vasil'ev (in French) who shows how WT ídentities can be derived from
the functíonal integral representation of correla.tion functions; B, W. Lee who discusses
Chira.l Dynamics.
These authors have been a direct source of inspiration for tbe presentation we h ave given
here, which is an update of
J, Zinn-Justin in Trends in Elementary Particle Physies (Lectures Notes in Physics
31), Bonn 1974, H. Rollnik and K. Dietz ed$. (Springer-Verlag, Berlin 1975).
See also the lectures of Colemon
S. Coleman in Laws of Hadmnic Matter, Erice 1973, A. Zichichi ed. (Academíc Press,
New York 1975).
The linear u-model h"" been proposed in
M. Gell-Mann and M. Lévy, Nuovo Cimento 16 (1960) 705.
Its properties have been studied in
B.W. Lee, Nucl. Phys. B9 (1969) 649; J.-1.. Gervals and B.W. Lee, NucL Phys. Bl2
(1969) 627; J.L. Basdevant and B.W. Lee, Phys. R€v. D2 (1970) 1680.
Thís work is reviewed in
B.W. Lee, Ghiral Dyna.mics (Gordon and Breach, New York 1972).
For a review of tbe applicatíon of Padé apprcxímants to the summation of pertur hation
t.hoory for phenomenologíra.l lagrangians see for example
J. Zinn-J ustín, Phys. Rep. I C {1970) 55.
SU(2) X SU(2) symmetry breaking has been .<fiscussed in
S. Weinberg, Phys. Rev. 166 (1968) 1568.
Chiral Dynamics ls dírectly connected to Current Algebra. Among the large number of
nrtides and text books let us just quote:
S. Adler and R. Dashen, Current Algebms and Applications t& Particle Physics (Ben-
jamin, New York, 1968); S. Weinberg in Leetures on Elementary Partiele Physics and
Quantum Field Theory, Brandeis 1970, S. Deser, M. Grisaru and H. Pendleton eds.
(MIT Press, Cambridge 1970); S.B. Treiman, R. Jockiw and D.J. Gross, LecttJ.res on
Gummt Algebm and its Applicotiom (Prinr-eton University Press, Princeton I 972);
C. Itzykson and J.-B. Zuber's teJ<t book as quoted in the introductíon; M.L. Goldberger
and S.B. Treiman, Phys. Rev.llO {1958) 1178; S.L. Adler, Phys. Rev. 139B (1965) 1638;
l40B (1965) B736; W.l. Weisberger, Phy~. Rev. 143 (1966) 1302; S. Weinberg, Phys.
Rev. Let!. 17 (1966) 616;
SU{3) x SU(3) symmetry breaking is discussed in
294 SymmetrieJ and Renonnalization 13.6

M. Geli-Mann, R.J. Oakes and B. Renner, Phys. Rev. 175 (1968) 2195.
For Lhe ímproved energy-momentum tensor (Appendix Al3.3} see:
C.G. Calla.n, S. Co!eman and R. Jackiw, Ann. Phys. (NY) 59 (1970) 42.
lt has been noticed that at zero of the RG 13-function both dila.tation and conformai
inva..-ianctl are restored in
B. Schroer, Lett. Nuow Cimento 2 {1971) 867.
Conformal invariance ~>nd its spectaculnr consequences in two dimensions are thoroughly
discussed in the Proceedings of Les Houches Summer lnstitute 1988 Fiellis, Strings and
Criticai Phenomena, E. Brézin a.nd J. Zinn-Justin eds. (Elsevier, Amsterdam 1989). See
in particular the lectures by
J. Cardy, Con.Jurmal ln.variance and Statistical Mechanics; P. Ginsparg, Applied Con-
formai Field Themy.

E:x:ercises
A useful exercise isto verlfy that the SU(2) x SU{2) linear rr-model with linear symmetry
breaking can be renormalized at one-loop order consistently with WT identities. Some
help ca.n be found in the fol!owing references
B.W. Lee, Chírn! Dynamics, Gordon and Brea.ch, New-York 1972;
J.L. Basdevant and B.W. Lee, Phys. Rev. 02 (1970) 1680;
and the proct!edings of the Cargese sumroer institute 1970, D. Bessis ed. Cargese Lectures
in Physics vol. 5, Gordon and Brea.ch, New-YQrk 1972.
We here propose instead to study the structure of symmetry breaking, at the classical
levei, for an effective SU(3) x SU(3) invariant renorma.lizable lield theory. The first
re[erence is also relevant for this ~.J<ercíse. We consider as fundamental fields the scalar
and pseudoscala.r (rr,K,'),•J') bound states o[ the three u,d,s quarks. They can be
described by a 3 x 3 oomplex matríx M which transforms under SU(3) x SU(3) as:
M(x)- UJM(x)U2, U,,U2 E SU(3).
Exercise 13.1
Write the most general renormalí~able SU(3) x SU(3) invaría.nt scalar a.ction, in rour
dimensions, a.nd díscuss the general pattern of spontaneous symmetry breaking.

Exen:ise 13.2
Jdentify the Goldstone modes in the dífferent cases first by purely group tbeory consid-
eratíons and then by determiníng the rnass matríx.

Exercise 13. 3
Sponta.neous symmetry breaking can lead to a realistíc residual symmetry but not of
course to realistic masses. In particular the explicit SU(3) symmetry breaking is rathet
la.rge in the scale of physics we want to deocribe. The explicit breaking of SU(2) x SU(2)
ú; smal!er but responsible for the pion mass. Therefore to make the mo de! more realístic
we add a linear brea.king term:
Ss S.,.m. trC (M + Mt),
where C i!; a diagonal rnatrix with diagonal elements c, c, c'. The model then depends
on 6 parameters. Determine analytically the mass spectrum. ls it possible to adjust the
patameters in such e way that the pseudoocalar deca.y consta.nts and the ma.sses agree
in a reasonable way with the experimental values? This last exercise requires a particle
data table a.nd a. handpoeket calculator.
A13.1 Symmetríes and Renormalízation 295

APPENDIX 13
CURRENTS AND NOETHER'S THEOREM

A13.1 Real Time Classical Field Theory

In this sect.lon, in order to discuss properties of the telativistic classical equations of

rnotion, we shall adopt the cova.rlant notation of real time fie!d theory with the metric o f
sign.ature (- .:.. · · · + ). Summa.tion over successlve upper and lower índices will be implied
(see Chapter 22).
Ifthe lagra.ngian density I:.(l/!,8,.41) depends only em the field ,P(x) and its deriva.tives
o,.,P(x), the classícal equation of motion obtained by va.rying the actíon S:

S (9) =j ddx I:.(.P(x), â,.\l>(x)), (Al3.I)

âl. 8L
8
~ â[8,.,P(x)] - 8!J>(x)
0
' (A13.2)

(in this notation ,P(x) IUid â,.!J>(x) are considered as independent variables).
lfwe perform on ,P(z) a space.dependent group transformation whích we parametrized
by a field A(x):

as a consequence of the equation nf motion, the action is also stationary wíth respect. to
va.riations o f A(x) a.t <J! fured:

ar. (Al3.3)
aA(x) =O.

We define a cummt J~(x), functional of !J>(x), by:

J"{x)= - - -
a[8,.A(x)]
ar. IA(x):O '
(Al3.4)

in whicb we ha.ve assumed that A(x) = O corresponds in the group to the identity. By
oonstruction currents are direct.ly assodated with the generators of the Lie algebra. of the
symmetry group.
We ea.n then rewrite equation (A13.3) as:

ar.
a" J''(x)=-- (AI3.5)
8A(x)'
which ia Noether's theorern.
lf in addition the ]agrangian ls in\'&I'iant under spac~independent group transforma·
tions, ôl.j8A vaníshes and thus the current J,. is conserved:

(Al3.6)

we take the example of a lagrangian density of the form:

L= f;ô,.#J"rfl•- V [1/>(x)], (AI3.7)

296 Symmetries and Renonnalization Al3.2

(in real time covaríant notation) and if the infinitesimal group tra.nsformations are:

(ill3.8)
the current J:(:x) is gíven by:

(A13.9)

In classical mechanics the space integral of the time-component af the current is a charge
Q"' {t Xd):
(A13.10)

By dilferentiatíng with respect to t and using the current conservt>t.ion equation (Al3.6)
one linds:

ftf!' (t) o.

The charges Q" (t) are constants of the classical motion.

Al3.2 Eudidean Field Theory

We have already exarnined the consequences of symmetries for field theories and derived
WT identities. These identities can also be deríved in the operator formalísm of quantum
mechanics and in this case currents and charges, considered as quantum operators, play
an important role, In our formulation, currents will appear either in the coupling at
leading order of matter to gauge fields (see Ch..pters 18,19) or as polynomials in the
fields (operators in the sense of Chapter 12) satisfying some identities which we shall
derive and therefore havíng special renormalizr.tion properties.
Let us therefore consider the geueratíng functíonal Z(J):

Z( 1) =f ld~l exp ( -S (QI) +f l;(x).P;(x) dxJ, {Al3.1l)

in which the actíon is inva.riant under group transformations whose infinitesimal form ís
given hy equation (A13.8) when A (x) is a constant.
In what follows dimen•ional regularization is assumed.
We now perform a change of variables in integral (A13.11) of the form of a transfor·
matíon (13.10). We define the eudidean current J~(x) by equatian (A13.4) in terms of
the euclidean action density. lf S (4>) is symmetric, the va.riation of the action reads:

(A13.12)

ldentifying the coefficient of A''(x), we obtain:

f [drbl[B,,J;:(x) l,(:r)t~<l>;(x)] exp [ -S(if>) +f l;{x)•Mo:)dx] =O. (A13.13)

This identity can be rewritten:

Al3.3 Symrnetries and Renormalization 297

ZJ;:(z) is the generating functíonal of correlation functions with a J;:

(x) opera.tor inser-
tíon.
The same equation ís valid for connected mrrelation functions. After Legendre trans-
formation we find:
'"fJ;:(z) = - b\l';(x)
u,. bf t,j\1']
o (X) • ( Al3.15 )

Equations (1113.14,1113.15) are the anruogues for correlation functions of the current
conservation equation (Al3.6). Integrated over ali space, they yield, not surprisingly,
equations (13.17-13.20), i.e. the WT identities of the symmetry.
Prom the point ofview ofrenorm..Jization, equation (Al3.1&) tells us that the insertion
of a,.JJ: (x) in a renormalízed correlation function is finite.
In a simple renorm..Jiza.ble q\ 4 -like field theory, covariance then lmplies tbat the same
must be true for the current J;?(x). This result is non-trivial since from expression
(1113.9) we see that J;:(x) is an operator of dimension 3. A further consequence is that
the insertion of a conserved current in a correlatíon function does not modify the form
of the RG equations.

A13.3 The Energy Momentum Tensor

If the action is translation invariant, the substitution qí(x) - ifJ (x + <), in whích c is
a constant, leaves the action ínvariant. In the spi:rit of Section Al3.l, let us perform a
spo.ce-dependent translation, which coincides of course wíth a generru change of variables
(ooe ..Jso Section 22.1). We thus substitute in the action qí(x) - ,P(x + c:(x)). lf .P(x)
satisfies the equation of motíon, the variation of the action (A13.1) at first arder in €
vaníshes. In the suhstítution the derivatives transform like

To c..Jculate the varia.tion we then charige variables x + E y. Translation invariance

implíes that the action density depends on x through the field </! but not explidtly.
Therefore the only new effect is to change the measure of integration

lf we now compare the new action with the initioJ one (A13.1) we see that the mod-
ilícaUons come only from the derivatives and the integration measure (V is a dummy
integration variable). Collecting the terms of order E and integrating by parts, we obtaín
t.he following identity:
8~T,','(x)=0, (1113.16)
In which the eneryy momentum tensor Tt'(x) ís defined by:

(1113.17)

For exam pie for the lagrangian (1113. 7) T"" has the form:

(1113.18)

In which gl'" is the usuoJ Minkawski metric t.ensor.

 Symmetries and Renorrnnliwtion Al3.3
298

To the tensor T~v(x) cortespond constants of the classícal motion P~ which are just
energy and momentum, obtained by integrating the time components (with respect to
one index) of T,..
on space:
(A13.19)

with x. :;;;t
(Al3.20)

We note that a space-time-dej>endent change of variables on x~ is an arbitrary change of

coordinates. This explains that the tensor T~. appears in the coupling of ma.tt.er field to
the metric tensor in General Relativity (for deta.ils see Chapter 22 and the corresponding
references).
A!so any current a.'5socíated \\'ith an additional space-time symmetry of the action can
be related to T,. •.
For instance the O (d 1, 1) pseudo-orthogonal transformations whose infinitesimal
form is:
!Jx" = A~(x):r'', (Al3.2l)

correspond to the choice:

(Al3.22)

The corresponding currents M ""• are then:

(Al3.23)

Dilatation invariante. Let us, as an example, again consider the ca.<e of the rp4 theory
in 4 dímensions:
(Al3.24)

In the absence of the mass term, t.he action is scale-invaríant, i.e. invaria.nt in the substi-
tution:
(A13.25)

For what concerns the varie.tion of the argument, dilatation corresponds to take E" as:

(AJ3.26)

We expect thus the di!atation current S" to involve :r;vTt. A short calcula.tion leads t.o:

(Al3.27)

In the presence of a mass term, the current S~' {x) is not conserved. Instead:

(A13.28)

Let us introduce the tensor T~v(x) :

(Al3.29)
I Al3.4 Symmetnes and Re'!armalization

The tensor T1,u can be nsed as energy momentmn tensor ínstead of T~v: ít is a polynomial
in the field, symrnetric as a tensor, and satisfies the conservation equation:
a~ft "'o.
29!}

(Al3.30)
In terms of Tt' (x) equation (A13.27} then reads:
S~'(x) = x"Tf (x), (Al3.31)
and tbe divergence of tbe dilatation current is:

a"s" =r;. (A13.32)

In dilatation.jnvariant theories, the trace of the "improved" energy momentum tensor
Tt vaníshes.
A13.4 Energy Momentum Tensor and Euclidean Field Theory

Performing the infinitesimal change of variables:

,P(x) = ,P'(x + e(x)), (Al3.33)
in the functíonal integral, one can easily derive WT identitíes for the insertion of the
energy momentum tensor (also called the stress tensor). The variation of the actíon with
a source is:

6 [f J,Pdz-S(,P)] =~.(r)[J(x)8• .;&(r) +â,.T,..(x)l· (Al3.34)

lt follows that:
~ZT,.(•) + J(x)íf! ó~~) 0. (Al3.35)

lntegrating this identity over space yields

I!Z
f dxJ(x)8. 6J(x) =0,

which expresses the tra.nslatíon ínvariance of correlation functions.

(A13.36)

After Legendre transformatíon, one finds:

o (Al3.37)

Again we conclude tha.t th.e insertion of the operator B,.T,..(x) in a renormalí'Z.fld correla-
tion function ís finite. However this does not imply tbat the insertion of T,...
itseU is finíte.
In the .p:
field theory for example, T~. is of dimension 4. The quantíty {1! •• 8 2 - a.a.) ql
iB also a symmetric tensor of dimensíon 4 whose divergence va.nishes. Therefore it can
appear as an addítive counterterm in the renormalizatíon of T,..:
(T.... J, r•• + A(ó••a' a,.a.) {4>'),. (AI3.3S)
Note tbat the renormalized energy momentum tensor has antomatically a non-vanishing
trace, and it can no longer be improved since the coefficient A is divergent. The dilata-
tion current is not conserved but this should have been expected since it is impossible
to regularize the tbeory without breaking tbe classical dilatation invarlance, elther by
íntroducing a cut-off, ar by changing the dimensíon. lt is nevertheless possible to derive
WT identitles involving the divergence of the dilatatíoo current. By integrating them
( over space, one just obtains the CS equations der:ived in Cbapter 10.
300 Symmetries and Renormalization Al3.5

Al3.5 Dilatation and Conformai Invariance

Let us now consider a general euclidean action S invariant under translation, rota.tion
Md dila.tation. We perform the infinitesimal change of variables:
(Al3.39)

Tra.nslation invariance implies that the varia.tion of the action involves only the partia!
derivatives of ,,.(:z:):
6S =f ddxT,..(x)él,.tu(x). (A13.40)

Rotation invariance implies that óS vanishes for:

(Al3.41)

in which A,.. is an arbitrary antisymmetríc rn"trix. Therefore the integral of the stress
tensor must be symJne.ttit:
f ddx (T,..- T.,.) == O.

Dilatation invariance corresponds to:

(Al3.42)

and implies the vaníshíng of the integral of the trace of the stress tensor:

f 4
d xT,.,. O.

For the simplest class of theories, like scalar field theories with BJI a.ction S(t/>) depending
only on the field q,(x) and its first partia! derivatives, the two integral conditions imply
the existence of a sylnmetric, tra.celess stress-energy tensor:

It then fo!lows tha.t the va.riation of the action also vaníshes for a.ny function <,. which
satisfies::
(A13.43)
where d is the dimensíon of eudídean space. The group of transformations which satísfy
eqne.tion (Al3.43) is larger tban the prodnct o(transformations which we ha.ve considered
so far: it is the whole canformal group. lndeed let us calculate the varia.tíon of a line
element of the form:
(ds) 2 g(:r)dx,.dx.,, (A13.44)

which corresponds to a conforma!ly fiat metric.

Wefind:
(A13.45)

We now see that the equation (Al3.43) is the necessary ond sufficient condítion for
the líne element to retain the form (A13.44). By defmition the tra.nsforma.tions which
preserve the form of the rnetric (A13.44) are conformai transforrnatíons.
Al3.5 Symmetries and Renonnalization 301

Fram equatíon (A!3.43) follows:

(A!3.46)

For d > 2 the equation implíes that ali second derivatives of 8-e: vanish. Returning then to
equation (Al3.43) one shows easily that ali thlrd derivatives of ~, also ''anísh. Solutions
of degree Ocorrespond to tran•lations. Solutions of degree l correspond to rotntions and
dilatations. The additional so]utions of equation (A 13.43) are second degree polynomials
ofthe form:
(Al3.47)
They correspond to special conformai transformations. The integrated form of these
tnmsformations is:
2
x' _ xJL+a.ux (A13.48)
" - 1 +2a·:z:+a 2 x 2 •
The conformai group is ísomorphlc to SO(d + 1, 1). lmposmg conformai invnr\nnce to
correlation functions determines in particular 2 aml 3--point functions.
In dimension d 2 the situation ís complet.ely different: t)N, set of equations (Al3.43}
are just the Cauchy conditions and express the well-known property that a.ll analytic
tranoformations are conformai. The conformai group has an infinite number oi genera~
tors. The consequences are much mote striking and lead to the classilication of a whole
class of conforma! ínvariant field theories.
Of course we have seen in Chapter 10 that the scale invariance of the clossical thoory
is btoken at the quantum levei. However, a:; we shall discuss in Chapter 25, there exist
situatjons in which the RG P-function vanishes, at leMt for some values of the coupling
constants. Then both the dilatation invariance and therefore the conformai inva.ria.nce
are restored.
Remark. The conditíon that the action should depend only on the field and its lirst
derivatives can be illustrated by a simple counter-exnmple. Consider the free action S( .P)

8(4>) = j d•x(â'.p(:r)) 2
.

The propngator in Fourier space ls lfp4 • The theory ís obviously translation, rotation
and scaling invariant. However it ís easy to verify that it is not wnformal invariant.
14 THE NON-LINEAR a-MOOEL: AN EXAMPLE OF NON-LINEARLY
REALIZED SYMMETRIES

We naw consider models possessing global symmetries non-linea.rly realize<! on the fields.
This means in particular that under an infinitesimal group transform~>tion, the variatían
of thc field is a non-Jinear function of the field. Since such models have non-trivial
geometrícal properties, we first extensively discuss a particular ~ample, the non-linear
u-model, a model wíth an O(N) symmetry, the field heing a N-vector of fixed length.
A tree levei analysis reveals that in the non-linear u-model the O(N) symmetry is
automatically realize<! in the phase of spontaneous symmetry breaking, unlike a .p• like
theory with the same symmetry: The action describes the interactions of on N l
massless fields, the Goldstone modes.
Pawer counting shows lhat this model is renormalizable in two dimensions. Therefore
the field is dímensionless and we face a problem already mentioned in Se.::tion 8.3: the
degroo of divergence of Feynman diagrams is bounded, although an inlinite number of
counterterms are generated because ali correlation functíons are divergent. We prove in
this chapter that, due to the special geometrícal properties of the model, the coeffidents
of all counterterrns can be calculated as a functlon of two of them so that the renormalized
theory depends only on a finite numher of parameters.
Note that since the lields are massless in two dimensions IR divergence< appear in the
perturbative expansion. An IR regulator is thus required.
In Section 14.8 we díscuss the renormalization of <:omposíte operators. Finally in Sec.
tion 14.9 we indica.te how the results can he recovered from another, linear, representation
of the model where the condition that the lield is a vector of lixed length ís enforced by
a Lagrange multíplíer.
In Chapter 15 we shall show how the arguments geneta.lize to ai! models defined on
homogeneous spaces. Actuel!y the O(N) non-linear a model belongs to a class of models
constructed on special homogeneous spa.ces, the symmetric spaces, which as Rlemrumían
manifolds, admít a uníque metric. We shall study them in more detail both as classical
and quantum field thooríes.

14.1 The Non-Linear a-Model: Definitjon

We consider a.n N-vector fieid <t>(x) which satisfies an O(N) inva.ria.nt constraint:

,P 2 (x) =I. (14.1)

Tbe fie!d .p can also be identified with an element of a. homogenoous (symmetríc) Bpa.ce
O(N)/O(N -1): Jet g(x) be an element of O(N) depending on the coordinate x, D (g(x))
the oorr.,ponding matrix in the fundamental representation, and u a lixed N-vector on
which D(G) acts:
u (1,0, ... ,0) (14.2)

The field .p can then be parametrized as:

.P(x) = D (g(x)) u, (14.3)

14.1 The Non-Linenr u-Modei 303

since the column vectors of an orthogonal matríx are by definition normalized to length
1.
The little group of u ( or stabi!izer), i.e. the subgroup of O(N) wbich leaves u invariant,
is O(N -1). Therefore if one multiplies on the right. g(x) by any element ofO{N -1), the
líttle group of u, the r.h.s. of equation (14.3) is left unchanged: relation (14.3) exhibits
tbe isomorphism between the COS<>t (homogeneous) space O(N)/O(N -1) and th<> sphere
SN-1.
To be able to generate the perturbative expansion of the corresponding !ield theory, it
is necessary to parametrize the fie!d in terms of independent vatiables. A pararnetrizatíon
of the sphere (14.1) convenient for renormalir.ation purpose is:

.P(=t) = {u(x},11'(:z:)}, (14.4)

in which 1r(x) is a (N- 1)-component fie!d. The field u(:t) is then a function of n(x)
through equa.tion {14.1). The equation can be solved locally, for example if a(x) is
positiv<>:
(14.5)
The consequences of the singularity of this parwnetrization will be discussed )ater.
The stabílizeJ" group O(N - I) acts linearly on 1r(x). We ca.n decompose the set
of generators of the Lie algebra of O(N} into the set of generators of lhe Lie algebra of
O(N -1) and the complementary set. To thís complementary set correspond infinitesimal
transforma.tions of the form

(14.6)

in which w,are constants, infinitesimal parameters of the transformation. The trans-

formation la.w of the <7-field then is a consequence of tbe transformation (14.6) of the
1r-fie!d:
6a(x) (14.7)

The a.ction and functional integraL The most general O( N) symmetríc action contaín-
ing at most two derivativos is, up to a multipllcative constant:

(14.8)

Indeed, dueto the constraint (14.1), any symmetric derivative-free term reduces to a
constant (and .p · ôl'.p va.nish.,.).
In terms of the field 1r(x), the action (14.8) can be cast into the following geometrical
form:
S (,.) =! j ddxG;i (1r) â,.>r;Ô~'"i, (14.9)

in which G;j ( ") is a metric tensor on the sphere:

(14.10)

In the form (14.9) the action is covariant under a reparametriza.tion of the sphere.
304 The Non-Linear a-Model 14.2

We have reen in Section 3.2 that the quantization of actions of the form (14.9) intro-
duces an additional ill-defined (sinre infinite) determinant. From equations (3.27--S.30)
we derive a rep.resentAtion of the generating functional of correlation functions:

Z (J) =f [(1 d1r(:r)

.,..2(:z;)) 112
] exp [-~g (s (1r) f dd:r..J(x) ·1r (:r))), (14.11)

in whích g is the coupling constant of the quantum field theory.

We had noticed that in the case of R.iemánnian manifolds the determinant ís needed
for geometrical reasons: here indeed it allows us to write the functional measure of
integrati<>n <Ner the field 1r(x) in an O(N) invariant way.
This formal interpretation of the determinant, however, doos not elíminate the diffi-
culty. We have to find a method to deal with thís in!inite contribution to the a.ctíon:

(14.12)

As we bave explained in Section 3.2, this difficulty is directly related to the problem of
operator ordering which a.ppears ín the qua.ntí,.,.tion.

14.2 Perturbation Theory. Puwer Counting

The mínima of the classical action (14.8) are given by:

la~.p(xJI = O ""} ,P(x) = const..

Dueto tbe O(N) invaríance, the action hàs a continuous set of degenerate a.nd equiva}ent
mínima. To expa.nd pertur bation theory we h ave to choose one minimum. lf we calculate
only averages of O(N) ínvatient correlation functions with action (14.8) a.nd the O (N)
ínvnriant measure of integral (14.11), ali mínima give exactly the same contr!bution.
Summing over ali mínima yie!ds a factor, the volume of the sphere SN-1, which dísap-
pea.rs in the normalization of the functíonal integral. However for non-O( N) invariant
correla.tion functions a. summ,.tion over ali mínima is equivalent to an average over the
O(N) group. Therefore it would seem that ali n<>n-vanishing correlation functions are
O(N) invariant. This problem is directly related to the question of spontaneous symme-
try breaking 8Jld breaking o f ergodicity in the ordered phase. In thís chapter we shall rely
on rather superficial and formal arguments, postponing a deeper analysis to Chapters
23-32 in whích the theory of pha:>e transitions is discussed. In the parametrization (14.4)
we choose the mlnimum 'II'(X) = O or .p(x) u.

1,4.2.1 Perlurbation theory

In the functíonal integral (14.11 J we ha.ve introduced a couplíng constant .g wbich plays
the formal role of l\. and therefore orders perturhation tbeory. For g small, the fields 1r (:r)
whích conttibute to the functional integral then must be sucb that:

and since we expa.nd around 'li'( X) = O, the field itself must satisfy:

(14.13)
14.2 The Non-Linear o-Model 305

Values af ?T(x} of arder 1 give exponentially small contributions to the functional integral
( of arder exp (-const./g)) which are neg!igible at any finite arder af perturbatioD theory.
This has two consequences: the restrictions imposed by the parametrjzation (14.5)
(o(:r) >O} are irrelevant in perturbation theory and in addition, in the functional integral,
we can freely integrate over 1r(x) from +oo to -oo, disrega.rding the constraint:

I?T(x)l ~ 1.

Perturbation theory then again relies on the evaluo.tíon of simple gaussia.n integrais.
We can now discuss formal perturbation thoory, setting temporarily aside the question
af UV ar IR (low momentum) divergences. We rewJ>ite the functianal integral (14.11):

.C( ... ,J) -1 [ (8".,-) ,

2g
{,...a
+- -P-
••f] + 2
1-'lf2
-
1 d
~ b (O} In (1 1
~.,. 2(z))- -J(x) ·1r(x). (14.14)
g

Note that the measure term has no 1/g fa.ctor and sta.rts contributing an]y at one-laap
arder. Since ,.. is of arder /Y, it is convenient to rescale the fie1d:

After this rescaling the a.ction deru;ity .C becomes

Expressian (14.15) shows that the intera<;tion term in the oction, once expanded i..; powen;
of g, generates an infinite number of dilferent vertkes wítb arbitrary even powers of.,.
and two derivatives. Still it is eMy to verify tbat at any finite order in pertu~ba.tion
tbeory and for a given correlation function, only a finite number af vertices <:ontribute.
Formally the meMure term yields additional vertices withaut derivatives.
The propagator tl.;; (p) of the 1r-field is:

(14.18)

At the tree levei the .,._field is mass]ess. Retuming to the analysis of Section 13.4, we
understand that, at leading order in pertUrbation theory, the noD·linear <T-model a.uto-
matical!y realizes the O( N) symmetry in the phase of spontaneous symmetry breaking,
the 1r-field rorresponding to the Galdstone modes. The massive partner of tbe 1r-field
in the linear realization, the o oomponent, h as been eliminated by the constraint (14.1 ).
This canstrnint can be formally obtained by sending the <T mass to infinity in order to
freeze the lluctuations of the <r-field.
Let us note that these properties are independent of the spedal choice {14.4) of
parametrizatíon of <l>(x).
306 The Nan-Linenr u-Model 14.3
1r2.2 Power counting
The general analysis has been given in Chapter 8. The form of the prnpagator shoWl!
that the dimensinn [7r] of the -.r-lield is:

[7r] = ~ (d- 2). (14.17)

Therefore the dimension of a ""rtex contaíning 2n -.r-lields is:

(14.18)

As a coru;equence:
(i) fnr d < 2 the theory is supeNenonnalizable by power counting;
(ii) for d = 2 it ís just renorrnalizable;
(iH) for d > 2 the thoory is not renorrnalizable.
We therefore study tbe model in dimension two. We have already mentioned a pecu-
liarity of thís case: altbough the theory ls renormalizable by power counting, MY local
monomial in the field containing at most two derivatlves and an arbitrary power of 1r can
a prúJri appear as a counterterrn. The symmetry O( N 1), whkh is linearly realize<:!,
only restricts the counterternlS to be of the general form:

However, at the tree levei, the non-linear O( N) syrnmetry impli"" that, up to a normal-
ízation constant, the actiou ís unique. Th understand the structure of the tllilory after
renormalízation, we ha"" to imrestígate the irnplícations of the non-línear O(N) symme-
try on the fonn of the divergences in perturbàtion thoory. We bave lirst to exbibit a
regula.rlzatíon scheme whicb preserves the O(N) symmetry and then derive a set of WT
identlties which express tbe consequences of the gymrnetry for correlation functíons.

14.3 Regu)àrization

In constructing a regularized version of the non·linear a-model, one hBS to be careful

to pr.,.,rve the O( N) symmetry. This is less trivial than in tbe linear case since, as a
consequence of the syrnmetry, the interoction terms in the action (14.9) are relo.ted to
the quadratic part. A simple method ís to start from the descriptíon of the model in
terma of the ~field because the action (14.8) formally is a free fteld action.

1,(.!!.1 Perturltative regu!ariza!ions

Pauii-Villars's Regularization. We replace the action S(r/>) hy SA(o;b):

(14.19)

Expressíng then o;b(x) in terrns of 7r(x), we di.scover that the large momentum behaviour
of the propagator has ímproved, but at the sa.me time new more singular interactíons
have been generated. I f the propag~>tor behaves líke:

6(p) ex 1/p•,
 14.3 The Non.-Línear a·Model 307

then the most singular interaction has s derivatives. Using equation (8.23) whích gives
the superficial degree of divergem:e of a diagram "f:

and elimina.ting the number of internallines I through the topological relatíon (8.25):

we find:
6(1·) (2-s)L+s+L(k.,-s)ua, (14.20)

in which we recall that L is the number of loops and k,. the number of derivatives at
vertex a.
As stated above, the worst caseis k,. = s:

6{"r):S(2-s)L+s.

As a consequence for L ~ 2 it is sufficient to take s ~ 6 to regularize ali diagrama.

However the one-loop diagrams have a behaviour independent of s, and thus cannot be
reguhttized. This property is not independent of the othi3T limitation of P"uli-Víl!ars's
regularization, that it c..nnot regularize the divergent measure term. Indeed we show
!ater that the one-loop divergenres generated by tbe interaction are needed to cancel the
divergences coming úom the measure.
Pauli-Villars's regu!arization is mainly useful in the study of non-linear models coupled
to chiral fermions.
Dimensional regularization. Dimensional regularization preser""" the O( N) symmetry
of the adion. FUrthermo<e, as a consequence of the formal rule:

the measure term can be ignored, anel therefore perturbation tbeory has no la.rge IIKlmen-
. : tum divergences for d < 2. Owing to his teclmica! simplicity, thís is the regulw:i~ation we
•ha!l in general use for practical calculations. One theoretica! drawback is that the role
·. of the measure is hidden. Therefore for the theoretical discussion of the renorma.lization
of tbe non-lineM a-model we shall consider both dimensional ..ud lattice regulari•ations.

1./.9.2 Lattice ~!ariza!ion and stat;,tical mechanica

. 'Ib construct a lattice regularized version of the non-linear u-model, which preserves the
O( N) symmetry, it ís again convenient to start from the descríption of the. mo de! ín
terms of the ,p-lield. We then replace derh-ati~ by finite dilferences, i, e. in the notation
· of Section 9.2:
1
a,..p(x) 1--+ 'i1 ~.P(x) = ã [.P (x + an~) - .p(:r)]'
·1n which x now belongs to a hypercubic lattice of lattice spacing n, and n,. is the unit
in the 11 direction.
308 The Non~Linen.r u~Model 14.4

Finally, w implement condition (14.1), we integtate over ,P(x} wlth the invariant mea-
sure on the sphere. The regularized functional integral has the fotm:

Z (l) = f IJ •5(1?(x)
;z€ó.Z 111
1) dql(x) exp [- ;s (,P,J)1, (14.21)

with:
S(,P,J) = ~L[VI'cf>(x)J 2 LJ(x)· .p(x). (14.22)
:t,,JJ ~

Using the parametrization (14.4), we ca.n express the lattice fie]d .p(x) in terms of :11'(x).
We obtain a ragularized form of functional integral (14.11). In particular the measure
term now generates well-defmed interactions:

In expression ( 14.21) we recognize the pa.rtition function of a cla.ssical spin lattice model
with a nearest-neighbour ferromagnetic interaction and in presence of an externa! mag-
netic field J (x). The coupling constaot g plays the role of the temperature. The critica!
properties (in the ..,nse of phase transitions) of tbis model will be discussed in Cbap-
ters 23-32in particular in Chapter 30 we will use the perturbative expansion of the lield
theory model to study ferromagnetic arder at low temperature. Expression (14.21} not
only provides a regularization of perturbation theory but also allows the use o! various
non-perturbatlve methods w study the non-linear u~model. Moroover it ís the only reg-
ularization which allows a discussíon of the role of the measure in perturbation thoory
(see Sections 14.5,14.6).

14.4 Infrared (IR} Divergences

Since in a massless thoory the propagator beha.ves like lfp2 , perturbatlon theory ls
divergent at low momentum (IR) in dimension two, the dimension in whkh the nnn-
lineat u-mndel is renormalizable.
Th generat.e a well-delined perturbation theory, it is neeessary w introduce an IR cut-
off. Since the absence of mass is a consequence of the spontaneous breaking of the O(N)
eymmetry, it is necessary, to give a ma.ss to the 71:-lield, w break the symmetry explicitly.
We can for example introduce an explidt mass term. HOW€"'et the study of syrnmetry
breaking mechanisms in Chapter 13 suggests a more convenient method which colll3ísts
in adding to the action (14.9) a consta.nt source h for the 0'-field (a magnetic field in the
statistícfll model of int.eracting spins)

S(w) >--+ 8(:11') h f cr(x)d"x, h >0. (14.23)

A lirst consequence of this modification ís that the minimum of the oction is no longer
degenerate. Instead we now have to maximize the source term and thís implies ,. O at
the minimum.
Second, if we expfllld cr in powers of 1r:

(14.24)
14.5 The Non-Linear r1·Model 309

and oollect the quadratic terms in the actioo, we liod the new 1r-field propagator Ll.ü(p)

g
Ll.;j (p) = li;;p2 +h' (14.25)

The linear " terrn has thus generated a mass h 112 for the ,-.field. At the same time it
has generated new deriva.tive-free interactíons.
Let us recai! that in the case of the linea:rly realized symrnetries, the brea.king term hr1
íslinear in an independ,nt field and therefore, as we'have shown in Chapter 13, genera.tes
no new renormalí:zation consta.nt.

Finite volume. We shall diS<:uss in Chapter 36 a.nother IR regularizotion scheme, which

does not break the O(N) symmetry. lt is based on the property that a symmetry ca.nnot
be spootaneously broken in a tini te volume and therefore no Goldstone mode is generated.
T.clmically, if we put the system into a hypercube of linear si.:re L, the momenta after
Fourier transforrnation are quantized. This solves the IR problem because integrals
are repii!Ced by diS<:rete sums. In the special case of periodic boundary conditions the
rnomenta have the form:
211
p=-yn,

lu this case some care is needed to handle properly the zero momentum mode p O
which seems to still lead to divergences. Howewr it has to be eliminated in [a\'OUT of a.n
integral over a coru;tant unit vector which represents the sum over ali degenerate mínima.
The other moment.a do not lead to IR divergences and can be treated perturbatively.
IR Finiteness of O(N) Inooriant Correlatíon Prmctions in 2 Dímensions. Let os here
mention briefly without proof an inleresting result whose signilicance wíll be discussed
!ater with the physics (in the sen:se of statistícal mechanics) of lhe non-linear c·model.
We have explained lhat in 2 dimensions correlation functions are IR divergent and we
have introduced an IR cut-off in lhe form of a sourre term giving a mass term for the
1r-field. At the same time, this additional term, brea.king the O(N) symmetry, lifts the
degeneracy of the classical minimum of the action and therefore elirninates a potential
difficulty witb penurbation theory: is it necessary ·to ta.ke into account ali degenerate
mínima of the action or can one choose one of them? However as we noted in Section
14.2 this quest.ion is irreleva.nt for O( N) inva.ria.nt correlation functions which therefore
play a special role. Actually it bas beeo oonjectnred by Elitzur and proven by David
order by order in perturbation theory, that O(N) inva.riant correlation functions have in
2 dimcnsions a finite IR límít, i.e. for example a limit when in the notation (14.23) the
breaklng pa:rameter h goes to zero, or any IR -cut-off is removed (see also Chapter 30).

14.5 WT ldentities

Havíug regulaflzed the theory both at short and loug distances (large and sml'dl mornenta)
we can now derive a set of wr identities expressing the consequence of the O(N) sym-
metry for oorrelation functions. We discuss only the part of the O{N) synunetry which
acts non-linearly, the consequences of the linear O( N - 1} symmetry have been díscussed
in Chapter 13 and now are obvíous. We thus cousider infinitesimal tra.nsformations of
t!le form:
(14.26)
310 The Non-Linf!4r u-Motú:l 14.5

The O(N) symmetric part of the actíon and the measure are left invaríant by such a
transformation. Only the source term and tbe brea.kmg term are affected:

A new compooite opera.tor a(x) = (1- :~t 2 (x))

112
is generated by the varia.tion of the
sonrce term. Following the general stra.tegy explained already in the case oi quadra.tic
symmetry breaking ternJS, we introduce a source for this operator. Then since the vari-
ation under a transformation (14.26) of this new operator as well as the variation of the
brealdng term are both proportíonal to 1r itself, no additional operator is needed. Let
us note that this would not have been tbe case with a breaking term proportional for
examp\e to J ,-2 (x)ddz. Let us cal! H(:z:) the source for a, a.nd Z (J,H) the generatíng

functíonal:

Z(J,H)=f[ -/1dw(x)
- n (x)
]exp{![-S(rr,H)+fddxJ(x)·7r(x)]}•
2 g
(14.27)

wíth:
S(n,H) =B(n) f ddxH{x)a(x). (14.28)

The generatíng functional of w-field correlatíon functions with the linear hu breakíng
term is given by:
(J) z
Z(J,H)IH(•)~h

Performing tbe infinitesimal transformatíon (14.26), we obtain tbe equation:

O'"' f [v'ld~ 1f1r

2
ddx p · h1r + H(x)óaJ exp [~ ( -S (1r ,H)+ f ddz J(x) · n(x)) 1,
or expllcitly:

O= f f ddx[a(x)J(x)- H(x)7t(:t)]exp [; ( -S(7t,H) +f ddxJ ·?r)]

(14.29)
We now replace 1r(x) by g (~/oJ(x)) and a(x) by g (6j6H(x)). Equation (14.29) then
becomes:
(14.30)

Thís first arder linear diHerentíal equa.tíon ma.kes no reference t,o the nau-linear char&cter
of the transformation (14.26). It is identícal to the equatíon one obtains in the case
of a linearly reafued O{N) symmetry, wheo J(x) and H(x} Me the sources for the
independent fie\ds n(:r) Bnd u(x).
lt is clear that W (J,H) = glnZ(J,H) setisfies the same equatíon:

f d4 o: [J(x) ;(x)
6
-H(x) 6J~x)} W(J,H) =0. (14.31)
14.E> The N<m·Linwr: o;-Model 311

We now perform a Legendre tra.nsformation. In contrMt with the linear case, homver,
u(x) is here a function of the ,.(x) field, and therefore the Legendre transformatíon
epplies only to the souroe J(x) and not to H(x):
6W
,.(x) (14.32)
6J(x)
We agaín use identity (6.20) sínoe H(x) is a set of externai parameters:

(14.33)

The Legendre transform of €<JUation ( 14.31) is therefore:

(14.34)

Thís is the ba.sic equation from whích we shal! derive the propertíes of the renormalízed
theory.
Remark. Equa.tion (14.34) ís quadratic in r.
This ís an essential differenoe with the
case of linea.rly rea.llzed symmetries. Gauge theories will provi de another example sharing
this property. Actually ít ís easy to show that if one uses the strategy explained above,
i.e. adds sources for ali new composite operators generated in the group transformation,
then the WT identitíes derived for the 1PI functional are at rnrn;t quadratic in r.
The !"f:riprocal properly. lt is easy to verify that the ínitial actíon satísfies equa.tion
(14.:>4), either directly, or by performing a loopwise expansion of "'luation (14.34) and
remembering that:
r(1r,H) ~ S(tt,H} + O(g).
Let us conversely assume that the oction S obeys the equation

(14.35)

We then perform an infinitesimal change of var:iables in the functional integral (14.27) of

the form:
6S(1r',H)
'lf
1
+ 61r with lhr liH{o:) w. (14.36)

In the dimensional regula.rízation scheme we can omít the measure term and the jacobian
because S ís local (for other regularizations see the remark below).
The variations of the action and the souroe term are (we omit the primes on the dummy
variable 7r):

6{exp [~ (-s(,.,H)+ f J(x)·11'(x)ddx)]}

= ; · [H(x) óJ~x) J(:r) 5;(x)] exp [~ ( -S (11',H) +f J (x). ,.(x)ddx)] . (14.37)

Therefore equation (14.35) by itself implies equation (14.30) for Z (J, H) and thus equa-
tion (14.34) for r (1r ,H).
312 The Nan-Linear u-Model 14.6

The mea.rure. The argument given above ís va)id only for dimensional regularization
where the me,..ure term va.nishes ídentically. We now extend it to the case of lattice
regularization. lndeed let us show that the invariant measure for the transfo,-mations
(14.36) ís

. 6S ]-'
TI d1r(:r) [ 6H(x)
First the change of variables (14.36) generates a jacobían J
(14.38)

The varíatíon of the measure term then is

6S ]-' [ 6S ]-'{ 1 [ 68 ]-' f12S }

[ 6H(x) ~ 6H(x} - õH{:r) ,s.,.,(x)6H(x) h,(x) '

which, using the explicit form (14.36), exactly cancels the jacobian. Note finally the
initial mea.sure !d,-(1 ,- 2 r 1 /~ ha.s tbe form (14.38).

14.6 Renorma.lization

Before explaíning the technkal detaíls, let us d""crib" the vaxious steps of the proof of
the renorrnalizability of the model.
Fírst the stability of equation (14.35) under renormalizatíon will be established: thís
means that íf the action at the tree levei satisfies equation (14.35), then it U; possíble
to renormalíze the theory in such a. wa.y that the renormali>:ed action stiil sa.tisfies equa-
tíon (14.35). Then equation {14.35) will be solved. It is important to realize that the
equation does not exphcitly refer to the transformation law (14.26}. This explains why
the explicit form of t!Je transforrnation law is rnodilied by the renorrnalizati<m although
the geometrical structure doe!; not change. lndeed solving equatíon ( 14.35) with the
constraints coming frorn power counting, one linds that only two renorrnalization con-
stants are needed, a coupling constant and a field renormalization. After renonnalization
the mo dei is still O( N) invariant for h O but the field <f> now belongs to a sphere of
renorrnalízed radius:

Renormalization. We assume that the theory has been regula.rized. We make a loop-
wU;e expansion, i.e. as the explícit form of the action sbows, an expansion in powers of g
of correla.tion functions:
00

r(,-,H) L:rnon.
n..O

We then insert this expansion into equatíon (14.34). The functíonal fo is simply the
original action and satisfies by itself equation (14.34). The one-loop functional rl satís-
fies:
(14.39)
14.6 The Non-Línear u-Model 313

This is a linear partia! dilferential equation for [' 1 which can be written sywbolically:

(14.40)

with:
+ éro _6_]
J d~x [~-6-
é~r,(x)
6H (.r) éH(x) 61r,(x) ·
(14.41)

Using the property that f' 0 satislies the equation (14.34), it is ea.sy to verify that 11;
is a generator of O(N) JO(N 1) acting on functionals of 'lf and H. In particular the
cornmutators [11;, 11;] are generators of the subgroup O(N -l). Applied to an O(N -1)·
invariant functional they vanish, which ahows that the system ( 14.39) i.s int.egrable (see
also Section 13.1.2).
We now examine the large cut-off behaviour (ar tbe behaviour when d approaches 2).
Equation (14.39) is satislied for ali values of the regu!arizing parameter. We conclude
that the divergent part rt· of r, defined in any núnimal subtraction scheme, also satisfies
equation (14.39):
=
11,[)''• o .

By adding to the tree levei a<:tion S(1r, H), -gi"f'• ('Ir, H) we render the theory finite at
one--loop order. Actually it ís ne<:essary to also add higher order ter ma to construct the
one--loop renormalized action S1(1r ,H):

=
81 ('lf,H) = S('lf,H) -gr'f'• (1<, H)+ L:>"6S~N) (1r,H).
2

These terms do not contrlbute to the one--Ioop order which is now finite, an<i are chosen
in such a way that 8 1 (1r,H) satisfiea the non-linear equation (14.35) exa.ctly. lndeed at
arder O, equa.tion (14.35) is verified since S (,.,H) satisfies it. At order 1 equatlon (14.35)
imp!ies equation (14.39} for f1;• (,.-,H) which is also satislied. Higher otder equations
determine the higher arder terms .5Sj"l.
Let us now show that this argument can be generalized to ali ordem. We write equation
(14.34) symbolically:
(14.42)

Performing a loopwise expansion of r, we wríte the term of order n (n >O) in equation

(14.42}:
" * r,._p =o.
L:rp (14..43}
~o .

Oor induction hypothesís ís that it has been possible to construct an action S,._, (n, H)
wbich satislies equation (14.35) exactly, and sucb that r 1, ...,rn- 1 have been rendered
finite. Then as we have shown above the generating functione.l r (1r, H) renonnalized up
to order (n- 1) alro satis6es equation (14.34), which implies equation (1,1.43). Let us
rewrite this equation in the furm:

,._)

ru•r.. +r... ro= L:rp•r,._,. (14.44)

.-1
314 The Non-Linenr u-Me>del 14.6

The indu.:otion hypothesis implies that the r.h.s. is tinite. The divergent part of the
equation satisfi~ therefore
{14.45)

The form of the equation is independent ofn. We then define Sn, the renormlillzed action
a.t order n, by
""
Sn =Sn-1 - g"r;!iv +L r/65~) • (14.46)
n+l

Jt follows

8, • 8, K = (Sn-l - g"r,!i•) • (8,_, - g"r;!1•) - K + 0 (g"+!)

-g" (Sn-1 * r;!1' + r,!i• * Sn-1) +O (g"+ ) •
1
(14.47)

Since at this arder in the r.h.s., Bn-1 Call be repla.ced by ro, i.e. S(7r,H), equation
(14.45} then implies:
(14.48)

Hence 8, satisfies equation (14.35) at order n. As in the case n = 1 we then choose the
hígher order terms 68~) in such a way that Sn sa.tisfies equation (14.35) identical)y.
This completes the derivation. The renormalized IPI functionlil satisfies equation
(14.34) while the complete renormalized a.ction satisfies equation (14.35).
To determine the form of the renormalized octian, we now bave to solve equation
(14.35) taking into account ]ocE>Iity and power rounting.

The measure. In tbis discussion we ba'"' never mentioned the role of the meosure. ln
the next section we sbal! verify tbat the field transformation ba.s been renormf>lized, and
tlrus the measure must have changed a.ccordingly; our discussíon was really only valid for
dimensionrJ regula.rization where the measure terms vanish identicrJ)y. We now extend
tbe derivation to the CMe of lattice regula.rization.
We bave shown in Section 14.5 that the invariant meosure for tbe transformations
(14.36) is
oS ]
TI. d... (x) [ /;H(:z:) -l

We note that the vertices generated by the mea.sure term are nat multiplied by a factor
1/íJ in contrast wíth those coming from the cla.ssicrJ action. As a conseqnence tbey
lilways cantribute at the next arder in cmnpariS<Jn to the vertices comlng from tbe actjon.
For instance at one-loop arder tbey contribute under tbe form of tbeir tree approximation.
Therefare u:, once r n is rendered finite, we modify tbe measure term by a divergent term
of ocder g", to take into account the n <h order renormrJization of the tield, this will
r r
affect n+l• n+ 2 , •.. , wbich are not yetrenormf>lized, and leave r,.
unchanged. It is
therefore possible to introduce tbe field renonnalization tbe measure without changing
the arguments given above about renormf>li:l;ation.
ActurJJy the measure term cancels at next arder tbe potential quadratic divergences
which could appear according to power counting. lt prevent.s at the same time the
appearance of an induced mass terrn.
14.7 The Non-Linear u-Model 315

14.7 Solution of the WT ldentities

14. 7.1 Globo.! solutian

We have to solve equation (14.35):

Powex counting tells us that the dimension [11'1 of 1f and [H] of H are in twO dimensions:

The action density has dimension 2, the action has therefore the general form:

(14.49)

a term quadratic in H having at least dimension 4. Moreover the coefficient B(1r) is

dimensionless and thus a derivative-free function of >r(x} while A( >r) has dímension 2
and cont;rins terms with ai most two derivatives.
Let us write the coefficient of H(x) in equ11tion (14.35):

(14.50)

Since B ("') is d.,-;vative free andO (N- 1)-invariant, the solution of equation (14.50) ill
simp]y:
(14.51)
This shows that B (1r) is the renormalized a-field and Z ls a first renormafuation con-
stant. Let us now write the part of the equation which is independent of H(x):

O. (14.52)

This equation tells us that A (1r) is invariant under an infinitesimal transformatíon of the
form:
ó1r(x} = w B (1r},
or solving equation (14.51):

(14.53)

This is the renormalized form of the non-linea.r part of the O(N) txansfonna.tions. Equ&-
tions (14.51) and (14.52) show that the renonnalized functíonal A (1r) is O (N)'invaríant,
but the radius of the sphere h115 been renormalized. The renormalized action ran there-
lore be written:
316 The Non-Linr:ar a-Model 14.7

with the relation

(14.55)
We have proven a rather remarkable result: although the interactíon in the non-línear a-
model is non•polynomia.l, the themy ca.n be renorma.lized with only two renormalization
constants. Let us also stress the striking similarity with the línea.r O(N)-invariant (4>2)
2
field theory, the only differences come from the absence of the renormaliz.ation of (,p2)
2
coup!lng which of course has no equivnlent in the non-linear theory, and the atructure of
the mass renormalization. In particular no spontaneous ma.ss is genera.ted. Fina.lly let
us note that by giving a mass to the 1r-lield through 11 term of the forrn f a(I)d"x, we
have introduced no new renormalizàtion canstant. This wauld not have been thE' Cll.Se
for a term of the form f 1r 2 (x)ddx for example.

14. 1.!! Linearized WT identities

In our discussion of the renormalization of the model we have admitted without proof
that it is always passible to solve in a local way the equations determíning the additíonal
hígher order terrns 6S}."l which make Sn exactly O(N) invariant, This is a point whích
can here be easily justitied.
Exp!icit •olution of the linEJJrized WT identities. To solve explicitly the equations
(14.39) or (14.45) satisfied by the divergent part of r, we limt discuss the general solutíon
of an equatíon of the form:

(14.56)

in which CJ (7T, H) is an arbitrary O(N 1) symmetric local operator and S is the bare
action:
S= j ddx {! [<â,.n) 2
+ (8.") 2] - H(x)<r{x)}, (14.57)

112
with o(x) .., (1- n 2 (x)) • Equation (14.56) reads explícitly:

(14.58)

in which we have delined:

1
a(x) = a( :c) [H(I) + M(x)]. (14.59)

The operator cr(x) ís an affine function of H{x). Let us choose in equation (14.58) a. new
set ofvariables and consider O(tt,H) as afunctionalÓ(1r,a). A straightforward but
careful calculation then leads to a new equation:

)6Ô ( .... ,<>)

f
d
d xo-(x brr;(I) =O, (14.60}

which shows thet the operator 6 (:>r, a) is O(N) symmetric at cr(x) fu<ed.
This result has two applications: it completes our proof and, a.s we discuss belaw, it
yields the renormalized form of a general O(N) invariant operator.
14.8 The Non-Línea.r a-Model 317

From power counting we know that Ji\1v has dimensíon 2. It has therefore terms of
dBgree O and 1 in cr. Accordíng to the previous result ít has the form:

(14.61)

The first term can be aboorbed into a couplíng oonstant renormalhation:

(14.62)

We DOW calculate the variation of the action when the ra.dius of the sphere i5 renormalized.
The ~ation oa
of O'(:r) is:

lt follows:
b {j ddx [! !B,.a(x)) 2
H(:r)O'(x)J} ~b"Z j o{x) ddx. (14.63)

Therefore the second tenn can be absorbed in a field renormalízation:

(14.64}

Thíz completes alt.ogether our proof of tbe renormalizatíon of the non-liDear u-rnodel.
Nevertheless for completeness we shall outline in Section 14.9 a different derivation, whose
basic idea is to return to a linear formulation of the symmetry.

14.8 Renorrnalization of Cornposite Operators

G•n•ml O{N) invariant opemtors. l..et us show that by solving equation (14.56) in
its most general fonn, we have also obtaínad the structure of renormalizad O(N) in-
variant operators of arbitrary dimension, inserted once in ,P(x) correlation functions.
lndeed to generate such insertioos we ca.n add to the action a source term of the form
f ddx K(x)O (7r(x)). Since O {7r) ill O(N) inva.riant, equation (14.35) ho!ds forthe com-
plete a.ction which includes thil< new term. An expan.sion of equation (14.35) at first
order in K (:r) lea.ds to equation (14.56) with S(,..,H) replacad by the renormalized ac-
tion. The renormalized operator CJ (7r, H) is thus the most general local functional ofthe
renorma!izad fields 1r(x) and CT(:r), of dimensioD (OJ and O(N) invariant at a(x) fixed.
Renorrnalizatian af dimenaianless opemtors and parametrization of the sph.ere. Dimen-
~ionless op€tators are derivative-free local functions of the 7r{:r) lield. A simple extension
or previoU& arguments shows that they should be classified according to irredudble rep-
resentations o f tbe O(N) group. Th each different irraducible representation is assodated
~ new renormalization constant. For example a mass term f ddx ....~(x) is a compoDent
tbe symmetric trnceless tensor

introd uces an additional renormalízation constant. lt now becomes clea.r why we

chosen the particular parametrization of the sphere in terms of the ,.. field. Let us
318 The Non-Linear a-Model 14.9

choose another parametrízation (r€6pecting the O{N- 1) symmetry) in terms of ne!ds

l':l(x) related to :rr(x) by:

l':l(x),.... ?t(x)f (:rr 2 ).

We can rewrite this expansion in terms ofthe (i, O... O) components of the tensors tran&-
forming under irreducíble representations of O( N). These components have the form
:rr(x)P(' (a(x)), in which Pl are byperspherical polynomials: I
Each non-vanishing coeflicient Ct introduces an independent renormaliza.tion constant. In
particular in the generic case in wh;ch ali coefficients cr are present, the renormalization
of O(x) corresponds to an arbitrary change of parametrization:
lI
l':l{x) 8r(x)Z [B?(x)]. !

14.9 The Linearized Representation

For a.ll models on homogeneous spaces, it is always possible to express the relations
between the components of the fie!d in the linear representation by introducing a set
of Lagrange muliipliers. In the càSe of the non-linear a-model this is specially easy
since only one O(N) invariant addítional neld, which we denote by a(x), is required to
implement the constraint (14.1):

The path integral representation (14.27) of the generating functional Z (J) can be rewrit-
ten:
(14.65)

with:
(14.66)

Note that here J(x) is a N-component source for the N-component lie!d <,f>(x).
We choose an extremum of the potential as the starting point for perturbation theory

a(x) O, <r(x) ;;;; .P1(x) = 1, Q'>;(x) =O foc i> 1. (14.67)

The propagator of the (N- 1) remaining components 1r(:r) is still proportional to 1/TÍ'-
The component rr(x) ís coupled to o(x). At the tree levei the connected 2-point functions
are:
(aa) (rra))
( (arr} (a-a) c (p) (~ g~) (14.6B)

In this framework, the model presents a spontaneously broken linearly realized O(N)
symi'IHltry. The consequences of such a situation have already boon analyzed in Section
13.4. Power counting tells us that the dimensions of the fields in dimension two are'

!<1>]=0, [a] 2
14.9 The Non-Linecw 11.-Model 319

The general form of the renormaliud actíon S, (o;!>, a} follows:

S, (r/>, a} ~f ddx{Bl (o;!>2 (xl) lâ,.o;l>(x)J 2 + ~ (o;b2 (x)) (o;!>· ô,.,p)?

+ Ba (4>2 (xl) a(x) B4 (o;!>2 (x)) }. (14.69)

in which B 1 , B 2 , Ba and B• are arbitra.ry functions, and therefore an infinite number of

renormalization constants are needed.
However if we are only interested in the <J>-field correlation functions (and not the
a-field), we ca.n explícitly integrate over the a-field. This fixes the value oi o;!>2 (x) to a
solution of the equatíon:
Ba (4>2 (x)) =O. (14.70)
This equation has a wlution order by arder in perturbation theory:

(14.71)
After integration, BJ, B2 and Ba become pure constants and 4> ·&,.o;!> vaníshes identicaiJy.
We therefore recover the results of previous :;ertion. The main disadvantage of this
forrnulatlon is that it introduces an infinite number of renormalization constants as an
intermediate step, and that its generalizatíon to other homogeneoUB spa.ces iB not aes-
thetically appealing.
Finally the non-linear formulation emphasires the connection with other geometric
models like gauge Lheories. On the other hand the linear formulation clarifies the dis-
cussíon of multiple insertions of general O(N) ínvaríant operators. In particular we
understand that the insertions of operators of dimension larger than two wil! eventually
genemte terms of degree larger than one in n(x). The integration over a then no longer
implies the stríct constraint (14.71).

Bibliographlcal Notes
For ea.rly references on the a-model see:
M. Gell-Mann and M. Lévy, Nuovo Cimento 16 (1960) 705; S. Weinberg, Phys. Rev.
166 [1968) 1568; J, Schwinger, Phy•. Reu. 167 (1968) 1432; W.A. Bardeen and B. W.
Lee, Nuclear and Parlicle P/J.ysia, Montreal 1957, B. Margolís and C.S. Lam eds.
(Benjamin, New York 1969); K. Meetz, J. Math. Phys. 10 (1969) 589.
The quantízation was diBcussed in:
I.S. Gersteín, R. Jacldw, B. W. Lee and S. Weinberg, Phys. Rev. 03 (1971) 2486; J.
Honerkrunp and K. Meetz, Phys. Rev. 03 (1971) 1996; J. Honerkamp, Nucl. Phy•. B36
(1972) 130; A.A. S!avnov and L.D. Faddeev, Theor. Math. Phys. 8 (1971) 843.
Pauli- Villa:rs's regularization for chiral models has been proposed by
A.A. Slavnov, Nud. Phys. 831 (1971) 301.
Tlle renormalization group properties were discussed in
A.M. Polyakov, Phys. Lett. 598 (1975) 79; E. Bré-zin and J. Zínn-Justin, Phys. Rev.
Lett. 36 (1976) 691; Phys. Rev. B14 (1976) 3110; W.A. 8a.rdeen, B.W. Lee and R.E.
Shrock, Phys. Rev. 014 (1975) ll85.
'l'he renormali•ability in two dímensions was proven in
E. 8rézin, J.C. Le Guillou and J. Zinn-Justin, Phvs. Rev. 014 [1976) 2615; 814 (1976)
4976.
The Elitzur conjecture has been proven in
F. David, Commun. Math. Phvs. 81 (1981) 149.
 The Non-Línear a-Moáel 14.9
320

Exercises
Exercise 14.1
Verify explidtly that when the action S(11, H) is O(N- 1) invarla.nt a.nd satisfies thc
relation (14.35) then the differentia.l oper.at.ors ~i

éS 6 6S ó ]
f
d [
~. = d X Ó1!; (x) óH(x) + óH(x) oori(x) '

and
~ij =f ddx [or;(x) 6,.:(x) - or;(x) 6,:(xJ,

form a representatlon of the Lie algebra of O( N).

Exercise 14. f!
We consider the partition functíon of the 0(4) ~ SU(Z) x SU(2) non-línear a-model

(14.72)

with the .action

{14.73)

2
lt what follows it will be conveníent to set c = p. f.
We propose to study this model at one-loop arder, to verify in particular that it can
be renorma.lizad with only two renormaliza.tion constsnt..s Z, z,
as proven in Chapter 14.
It will be convenient to set d = 2 + ' a.nd to define a dimensionless couplíng const.a!lt
g; 2
r'r(I - ê/2JJJ'(4,r•' 9.
Dimensional regula.rization will everywhere be used.
14.2.1. Establish the Feynma.n rules. For the calculations we have in mínd, only tht•
propagator, the 4-point and 6-poínt vertíces a.re required.
14.2.2. Write the renormalízed oction.
14.2.3. One-ioop order rolculation•. Calculate the va.cuum expectation value (<1) a.nd
determine the field renorma.lization constant. Calculale the inver""' two-point function
r!Jl [Wi 2l);, 1 and determine the coupling constant renorroalizatíon.
14.2.4. Calculate the connected amputated four-point function WJ:/•. at one-loop. In
what follows we adopt the convention that ali momenta enter so that Pl +P2 +Pa +P• = O.
Verífy that the anorwer is finite in two dimensions.
14.2.5. Wríte and verify the consequences of wr identities which involve only the
1--4-point IPI correlation functions, using equation (14.34).
15 MODELS ON HOMOGENEOUS SPACES IN TWO DIMENSIONS

In this chapter we describe the formal properties and discuss the renormalization of
a dass of geometrical rnodels: Models based on homogeneous spa10es. Homogeneous
spaces a:re associated with non-linear realizations of group representations and these
models are natural generalizations of the non-línear a--model considered in Cbapter 14.
They can be studied in dilferent pararnetrizations corresponding to different choices of
coordinates when these spaces are considera.d as Riemannian manifolds. However in
contrast with arbitrary manífolds, there exist natural ways to embed these maniioJds
in fiat euclídean spaces, spaces in which the symmetry group acts linea:rly. This is the
system of coordinates that we have used in the discussion of the non-linea:r a-model
and again use in the first. part of this chapter because Lhe renormali:<ation prope.-ties
are simpler and the physical interpretation of correlation functions more direct. We
then also exa:rníne some properties of these models in a generic parametrízation. The
renotmalization prohlern is solved by the introduction of a symmetry (generally called
BRS symmetry) with a.nticomrnuting (Gra:;smann) parameters which !ater will play an
essential role in the renormalization of gauge theories.
In a second part of the chapter we study the more sped!ic propertíes of mode!s cor-
responding to a spedal cl= of homogeneous spaces: symmetric spaces. The non-linear
a-model of Chapter 14 pro•ddcs the simplcst example. These models are characterized
by the uniqueness of the metr:ic and thus of the classical action. The classical equations
of motion genera.te, in two dimensions, an infinite numbér of non-loca! conservation laws.
We calculate RG functíons at one-loop arder, and find the first examplcs of asymptotí-
cally free theories.
The chapter ends with a few comments about more general models based on non-
compact groUP" and arbitrary ffiemannian manifolds. The appendix oontains a few
additional remarks about rnetric and curvature in homogeneous spaces. We also bríelly
describe classical familles of symmetric spaces.
Note fmally that in the description of these models, two different formalisms and sets
of notation can be employed, depending whether one wMts to emphasíze the group
•tructure or the Riemannian manifold point of view.

15.1 Construction of the Model

Notat.Wn. Let us wnsider a lield ~;(x) transforming under a linear representation of a

~[roup G. We use the notation of Chapter 13 for the infinitesimal group transformation,
denoting by t" the generators of the Lie algebra C( G) in some matrix representation,
. ~nd by w., the infinitesimal group parameters of the transformation:

ór/J;(x) = tíjw.,if>;(x). (15.1)

Since we con5ider only cornpact groups, we ca.n assume that the representation ís orthog-
mml, and that the generators of C.(G) are N x N antísymmetric matrices.
322 Madels on Homogenmw Spo.= in Two Dimensiona 15.1

15.1.1 Spontaneous symmetry broakin9: Goldston• mod.s

Gold.otone modes. Rather than giving a purely mathematícal constructíon, we shall try
to motívate the study of these ruodels by some physícal arguments. Our start;íng poínt
is, as in Chapter 13, a G-invariant action S(<J>) of the form:

(15.2)

We now assume that at the cla.ssícallevel we are in the situation of spontaneous symmetry
breaking (SSB), i.e. Y(<J>) has non-G-ínvarlant global mínima. We dístinguish one of
them, q,c, around which we expand perturbation theory, and ca!l H the isotropy or
líttle group {the stabilher) of q,c. We have shown in Section 13.4 that under these
circumstanc€5 a number of field components of <J>{x), which correspond to the Goldstone
modes of the broken symmetry, are massless. lf we are only int.eresta.d in the long distance
behavíour (IR limit) of correlation functíons, or equivalently íf the mass of the massive
fields is sent to infinity (the low temperature limit ofthe corresponding statistical model),
the lluctuatíons of the massíve fields in the functional integral can be neglected.
In the limit the remaíning massless component.s of the field <J>(x) can be entirely
parametriza.d in terms of a matrbt R(g(x)) of the representation of G, acting on the
vector <J>c
g(x) E G. (15.3)
Note, however, that ifwe multiply g(x) on the right by an element h(x) of H, since 4>j
ís left ínva.riant by the group H, .f>;(x) is not modilied:

,P,(x) = R.;(g(x)h(x))~] = R.k(g(x))Ro;(h(x))Q'ij R.;(g(x)).P~.

This shows that .p,(x) ís ...,ally a function of the elements of the coset spoce GfH. Let
us then divide the set of generators of the Lie algebra L:(G) into the set of generators
belonging to the Lie algebra .C(H), {t"}, cr > I, ond the complementary set {t"}, 1 :0::
a :-:: !, whích is such that:

The matrbc R can be canonically parametrized in t.erms of fietds ~ (x) as:

(15.4)

15.1.2 Golástone mode ejjet:tive actíon

Since Y(<J>) ís derivative-froe a.nd group-invariant, it is independent of g(x) and thus yíeids
an irrelevant constant contribution to the a.ctíon whích can be omítted. The action S (ti>)
can then be written:

(15.5)

By expanding action (15.5) for y(x) dose to the identity we ""rify indeed that aU re-
maining fields are mass)ess.
 15.1 Modek< on Homogeneous Spaa~ in Two Dimeruúm• 323

Notation. lt will a.lso be convernent to use B bra. and ket notation to indíca.te vectors,
denoting by lO) the vector <!>'. The equation (15.3) then ta.kes the form:

11/J(x)) .... R (g(x)) lO), (15.6)

in which lof>(x)) is a notatíon for the lield Q!;(x). With this notation the dassical actíon
S(<f>) can be rewritten in tenru; of a current R -'a,.R:

(15.7)

Equation of motion. Since the field can be entirely pararnetrized in term.s of group ele-
ments, the classical equations of motion are a consequence of the G-current conservation.
A straightforward calcu!ation yields the form of the current J~:

15.1.9 Metric and action in general cooniínates

In this su bsection we use a notation and conventions adapted to Riemannian geometry
(see Chapter 22 for deta.ils).
We denote by v>' an arbitrary set of coordinates on the manifo!d (coset space) GfH.
The action (15.5) can then be written in the form:

(15.8)

where 9ii iB a metric on G /H considered as a. Riemannian manifold. We know from tbe

dis<::ussion of Subsection 15.1.1 tha.t the matrix R when a.r:ting on the vector lO) can be
parametrized in the form (15.4). The varia.bles {" in equation (15.4) thus provide 8ll
cxarnple of a set of coordinates on the coset space Gf H.
Comparing the expressions (15.8) and (15.7), we can relate the metric to other geo-
metrical objects. The current R-•a,.R can be written

R -'a" R = R~l -;:;-'"•'P

8R,. i (x ) • (15.9)
Vlf''

In what follows él; with roman indices meam; derivative wíth respect to lf'' and à,. with
greek índices means derivative wíth respect 1.0 x,_
Any matrix of the form R- 1 8R (a.
pure gauge} belongs to the Lie alg..bra. of G. We can thetefore expand on tbe basis of
· generator.s (more details can be found in Appendix Al5.1):

éiR-
R- 1 8<pi 1
• = Lu(
= R- àR i 'P
)t 0 • (15.10)

R- 1 él,.R= Lr (v>)taà•v>'(:t). (15.11)

expression of the metric tenBor in terms of Lf(<p) follows:

(15.12)
324 Mode/.ç on Homogencous SptJJ:Es in Two DimemJions 15,1

wíth the definitíon

(15.13)
The latin letters (n, lt) indicate that the indioes run only over values corre<;ponding to
generators belonging to J:. (G /H).
One verifies that the quantities Lf are independent of the representation (Section
Al5.1). Therefore the dependence in the choices of the dassical vector lO) and the
particular representation of the group G is entirely contained in the positíve matrix p..~.
The form of the roatrix;,; only restricted by the symmetry under the subgroup H:

for t•, t• E C(G /H), t> E f.( H) and therefore:

(15.14)

The ínterpretation of this equation is simple: the Ve<:tors t• JO) transform under an (in
general reducible) representation of H which has the matrices J.-,b as generators.
The symmetric matrix J.l. commutes with ali generators í' ((í') •• ~ [a,•) of J:. (H):

[p,í'] =o.
As a consequence, the number of parameters on which P,0 o depends, is the number of
different H ·ínvaria.nt scalar products one can form with the irreducible components of
the vector t"IO) (Schur's lemma).
This result exhausts ali consequences of the G-symmetry.

15.1.4 Quantízation and perturbation theof'l/ near dímensíon 2

So far we have only examined the classkal theory. Tb quanti:oe it, we can start from
the euclidean actioo (15.5), deriVB a quantum mecbanical hamiltonian and use canonical
quantiza.tion. lt is faster to begin with action (15.2) and freeze the massive degrees of
freedom as expla.íned above. The result is the same and yields a functional representation:

(15.15)

The measure of the .,..;ntegration is the G-invariant measure induced by the ftat 4> mea-
sure, and is also the restriction to the coset space Gf H of the Haar group measure of
G.
From the p<>int of view of power counting the theory is clear]y renormalízable in two
dimensions for any parametrízation since the .,..6etd has dímension !(d- 2) and the
ínteraction terms contain two derivatives and arbitrary powem of the .p-field.
Tb prove the structural stability of oction (15.5) we derive in the next sections a set
of WT identities corresponding to the non-linearly realized symmetry under the group
G. Dimensional or ]attice regularízation respecting the G-symmetry is implíed in what
fo!lows. The choice of a group-invariant regularízation conesponds alro to a choice of
quantization of the classical hamiltonian consistent with the symmetry.
15.2 M odels on Homogeneo!J.S Spo.res in Two Dimensions 325

15.2 WT ldentities and Renormalizlition in Linear Coordinates

For renonnalízation problems it;,; convenient to chúose a specíal parametrization, which

consists in embedding the homogeneous space into an euclidean space, &li we have ex-
plaiued in the precading chapter. Let us describe thi. parametrization first.
Remark. Note that the vector space V spanned by the family of vectors of the form
(15.3) may have lower dimension th3Jl the original space of representation to which .p
belongs .if this representation ;,; redudb!e. The space V is still a space oi representation
for the group G. In what follows we restrict .p to its components in V.

15.2.1 Linear ClJordinates

Let us consider the vector spare V spanned by the wctors of forro (15.3), as well as the
subspace V of dimension I spannad by the vectors t:j<f>~. We choose an orthogonal basís
1

in V such that the vectors of V' ha"" on!y the first I componente non-vanishing. We then
dístinguísh the I first components of the ""ctor .p of equation (15.3), callíng them "~(x),
and the others called a-;(x):

<J>(x) = { 1r0 (x), 1 :S .a~ l ; {15.16)

a,(x), I< t

The fields a;(x) and 1r0 (x) are functions ofthe fields {.[r). lfwe expand equation (15.4)
Ín pOWI!TS O f {:

<,D;(x)
'
!f>'t + L{,(x)tfiq'>j +O (ç2), (15.17)
•-I
we see, comparing expressions (15.3) and (,15.16), that ..-.(x) and Ç.(r) are canonically
related:
(15.1S)
i.e. this relation can be ínverted to express the fields {. as functions of the fields "•· On
l.he other hand the fields a,(x) are of the Íúrm:

nnd can therefore be calculated in terms of the fields "•· The fields 1Ta(x) Bnd cr;(x)
tr:>nsform under dílferent linear representations of the group H. However the fields
"•(x) transform under a non-línear representation of the group G sínce the generators
jt"} mix the fields 1ra(x) and the a;(x) which are functions of '!Ta:

(15.19)

Note that since the cr; are functions of ,-., the transformation laws:

(15.20)

llte now consequences of equation (15.19) and the functíonal form of the u; {11').
326 Modeis on Homogeneous Spact'-• in Two Dimensitms 15.2

15.2.2 Correlation junctions, WT identitie8, renonnali.tation

We now consíder the generating functionnl of correlation functions

(15.21)

the integrand being expressed in terms of these specinl coordínates. Sources have been
added for all components of <1> in V for the reasons whicb have already been explalned in
the preceding chapter: when we try to derive WT identitles expressing the oonsequenc""
of the symmetry for çorrelation functions, the composite na operatcus appea:rs in the
variatíon of the ,.field. Moreover, since ali 11'-fields are massless, we have to break the
G-symmetry explicitly to pruvlde an IR cut-oiL This can be achievad by expanding
perturbation theory at fi.-..:ed constant values of the " sources.
Let us now derive the WT ídentities correspond!ng to the non-linearly realiza:! sym-
metry under the group G Md show how they imply the structural stability of action
(15.5).
WT identitie8. We perform an infinitesimal transformation (15.19) in the functional
integral (15.21). Since we have introduced sources for ali components of </>, the cor-
responding WT identity for Z ( J) and W (J) ;, identícal to the identlty obtained for
linearly realized symmetries:
,óZ(J)
f dx l;(x)t,, éJ;(:z:)
=O, (15.22)

and thus:
f d:d,(x)tf; 6 ~:x) =O. (15.23)

Again the differeoce appears in the Legendre t,ransformation. Let us now cal! J.(:r),
1 S aS I, the sources for the fields ?Ta, and H,, I <i 'f n, the sourc"" for the composite
fie!ds "•· The Legendre transform has to be performed only on 10 :

r(,-,H) + W (J,H) =f dna(x)Ja(x), (15.24)

5W 6r (15.25)
"a(:>:) óJ.(x) # J.(x) 6" 0 (:>:) ·

Since, as explained severa! times,

lir éW
6H;(x) - hH;(x)'
identity (15.23) ímplies for the lPI functional r:
f dx { é~~x) [t~•"•(:r)- t~i 6 :~x)] + H;(x) [tll,,.o(x)- tij 6 ;~,J} =O. (15.26)

Renonnalization. This identity has a. quadratic form as in the case of the !J· model.
The arguments of Section 14.6 apply and prove the sta.bility of the identity under renor-
malízation. The sa.me identity is thus fullil!ed by the renormalized action S,:

f ru:{ ti::r·x) [t~rr.(:z:)- t;:; 6 !~Cx)] +H;(:>:) [ti•"'•(x)- t<, 6 ~:{x)]} =O. (15.27)
 15.2 M<>d.<ls on Hamagenwus Spaces ín Twu Dimensions 327

Power countíng tells us that S, (1r,H) and H 0 (x) have dimension 2. Therefore agaín S,
is linear in the source H 0 (r). Let U5 wríte 8, as:

S, (1r, H) -f u,. (1r) H,(x)dx +E, (1r), (15.28)

in which the functions u,; (-rr) are derivatíve-free andE, has dimension 2, i.e. has at most
two derivatives. The term linear in H in equation (15.27) yields:

(15.29)

These partia.! differentía.l equations for the function u.; (1r) imply that if the fields "•
have the transformation law:

(15.30)

then as a consequence:
(15.31)
'fhus the field <f> with component (7r0 ,rr,,) transfurms under a linear representation of
the group G. Let us now write the second equation obta.ined by setting H equa.l to zero

(15.32)

'J'nis equatíon teUs us that E, {1r) is the most general functiona.l of "(r) with two deriva-
liws ínvarier.t under the group G.
The action can thus be constructed startíng from the most general G-invariant action
with two derivatives written in terms of the field <f>(x}, and then eliminating the fields
o,,(x). Also we have shown in Subsectíon 15.1.3 that tbe action, expressed in terrns of
t.he metric tensor, can be parametrí•ed in terms of lhe matrix 1-' (equation (15.13)). We
here find that even if at the tree leve) we begín utith a special choice of tbe matrix 1-'
'· solution of equation (15.14), we obtain after renormalization the most general solution
of this equation.

us bere make a brief comment about the solutions of equation (15.29). For renor-
rnalizo•til>n purpose, we are looking for yeneric solutions "•( ,.) expa.ndable in powers of

ar.= s, + Sfl1ra1 + isr1ifl1 71'o,l1fa., + ... , (15.33)

wlüch S;, Sj' ... are consta.nts.
The equation of order zero in ,. is:

(15.34)

I" belongs to the Lie a.lgebra L (H), % VllJlisbes sínce " a.nd a belong to different
ft![.ores;en1;ati.ons and therefore:

t~S; = O for ali t" E C. (H). (15.35)

328 Models on Homogeneous Spaas in Two Dimensio,.. 15.3

As a coDEequence the generíc vector Si is the most general vector havíng H as a.n isotropy
group. Note that sínce S; düfers from 1/Je only a.t on~loop order, the isotropy group
(stabilizer) of S; ca.nnot be larger than H.
For thesame reason the generators of J:.(Gj H) are such that !J:;S, spans the "subspace.
Therefore equation (15.34) implies:

(15.36)

The scalar product of the vectors S; with ali vectors of a complete bnsis is known, there-
fore ali S; are determined. ln particular if the " subspace contains no vector invariant
under the action of the group H, the r.h.s. of equation (15.:16) and therefore also ali
vectors S; vanish. This property holds for symmetric spaces.
Higher order equations determine the ooefficients of monmniats of incrensing degree in
rr. lf we assume that we know (]i [rr) at arder ,-k, then the equation for the coefficient of
,..k+J takes the form:

s:'b, ... b•... t~,sj = known quantities.

The coefficients S:~, ...b, are completely determined since cons.ídered as vectors s!• .. ·••
their sca.lar products with aU vectors of a complete besis are given.
The conclusíon is that the functíons o-;(11) depend on as many pa.rameters a.s the
number of independent vectors S; whlrh have H as an isotropy group. The renormalized
action is then the most general "free" rnassless action in the linear 4> varia.bles. We have
therefore enumerated ali the renormalizatíon constants of the mode(.

15.3 Renormali'llation in Arbitrary Coordinates, BRS Symmetry

We have shown in Section 15.2 how a special choice of parametriza.tion leads to a rathet
simple díscussíon of the symm~try properties a.nd renormalization of correlation func·
tions.
We have índicatad in Cha.pter 14, in the case of the non-linear o-model, that wíth
another choice we would have been forced to introduce a number, generically inlinite, of
additional renormalization constants, correspondíng to a renormalization of the param<>"
tri2ation of the manifold. The field is no longer multiplkatively renormallzed, the bare
field becollling a function of the renormalized field.
However, as we have indicated in Appendix Al0.3, in some C8!5!':5 the only physically
relevant quantities are those whích are parametrization-independent (related to geometri·
cal properties of the manifold or to the S-matrix). lt is therefore useful to aho investigate
models on homogenoous spaces in an arbitrary parametrization to more clearly exhibit
the parametrization dependence. Moreover we shall see on the m<ample of the calculation
of the one--loop íJ-function that some parametrizations are more convenlent for practical
calculations.
In this section we derive WT identities expressing the group symmetry in an arbitrary
parametrization and show that they are stable under renormalization. The general strat-
egy, j.e. to add to the action aources for a oet of composjte operators which is
infinitesimal group transformatíons on the fie!d, is only sultable if the minimum num·
ber of dilferenL operators is finíte. Then the renorma.lization of the thoory follows from
a rather straightforward genera!izatlon of the arguments given in the first part of
chapter. However, for a generíc parametrization an infinite number of composite
15.3 Models on Homogen= Space& in Two Dimensions 329

is required, and this strategy is no longer useful. We therefore int.roduce a new method
which, in some generalized form, will also be relevant to the renormaliza.tion of gauge
theories and which is based on infinitesimal group transformations with anticommuting
parameters.

15.3.1 Infinitesimal !J"'UUP tro.nsformations

We coosider a non-linear realizatíon of the representation o[ a group G acting on a field
<P'(z). We write the infinitesimal group transformations corresponding to parameters w"
in the form:
6w<P'(x) = D~ (<p(x)) WQ. (15.37)
We assume that the functions .v;,(<p) are smootb, i.e. infinitely differentia.ble, at '{> O.
We write the action in the form (15.8),

(15.38)

lf the action S ( <p) IS invaríant under group transformatíons, it :>atísfies:

(15.39)

This equation implies an identity for the metric tensor g,1 (,o):

aD~
D' a9ik
" a'P' + 9ik arpJ + g,, an~
a,• = o. (15.40)

'Jb solve this equation is equivalent to find ali possible metric tensors on a given homoge-
. neous space, compatible with the group structure. The Appendix Al5.2 contains some
detalls about the nature and the structure o/ this equaticn.
The functions D~('l') satiofy identities which can be ohtainad either by direct calcula-
tion or by noting that the differential operators Óa:
. â
D~(tp) â<p', (15.41)

on functions uf '~'• form themselves a representation of the Lie algebra. (see also
S...:tion 13.1.2) and therefore have commutation relations of the form:

(15.42)

that these commutation relations a.re compa.tibility conditíons for the equation
coru;ídered as a set of linear differentlal equations for S('l')· Calcnlating explicitly
rommutator in terms of the functions D~('P) we obta.in:

(15.43)

is the form of the commutation relations of tbe Lie algebra which is useful in the
of non-Jinear representations. Moreover in Chapter 21 we shall encounter
are formally identícal a.nd play an essential role in the discussion of the
'iln•>rl1~alization of gauge theories.
330 Modds un HomogeneoWJ Spaas in Two Dimensions 15.3

J 5.3.2 WT ídentities
We assume that we have introduced a group-ínvaríant regu!a.rizatíon and consider the
generating functiona.l Z(J,K):

Z(J,K)'"' f [dp(<p)jexp [-S(<p} +f dx (K{x)A (<,o(x)) + J;(.x},o'(x))]. (15.44)

The measure [dp(<,o)] is the group-ínvariant measure. Note that we have added to the
action sources not only for the field 'l''(x) but also for a local dcrívative-free funclíon of
l"(x), A(~?(x)). The function A('l') has tosatísfy only one condition: it h as to begin "'ith
a term of order rp 2 such tha.t, for K(x) constant, masses are generated for Bll components
of <p, which ca.n serve as IR regulators.
To derive consequences of the non-linear symmetry (15.39), we intmduce a set of
anticommuting constants c~ and l., i.e. ali belongíng to an antisymmetTic algebra., and
perform a transformation;

so'{x) ..... 'I'' (x) + W~('P(x})C" (15.45)

The a.ction and the measure are ínvariant. The varíation of the source terms in expression
(15.44) is:

6 {f dx [J,(:r);:i(x) + K(x)A ('l'(x))]} =i: f dx [J,(z) + K(x) ~:i1x)] D~(<p)C"'.

(15.46)
This varía.tion involves two composite operators: D~ (I"( X)) C"', because the transforma-
tion is non-linear, a.nd (6A/6'1'') D~('I')C". ln accordance with our general strategy, W<>
introduce for them two anticommuting sources h;(x) and L(x). Note that if we assign a
fermion number +1 to C" and -1 to A;(x) and L(x), this fermion number ís conserved.
We then calculate the varia.tíon of these operators under the transformatíon ( 15.45):

(15.47)

Sínce C"' and c~' a.nticommute, we can antisymmetrize the coefficient of C"' C~. Using
then the commutation relations (15.43) we find:

(15.48)

It is possible to cancel thís variation by performing a simultaneoiiS transformation OI>

the sources ca
of the form:
(15.49)

The geometrícal orígin of these transformations will be explained in Section 16.4 when
we discuss BRS symmetry.
Let ns now calculate the variation of (6Afli'l'') D~(I")C 0 under the combined trans-
formation (15.45,15.49):

(15.50)
15.3 Models cn Romogeneou.s Spares .in Two Dímen.sions 331

Tha first term vanishes because the coellicient of C"Cii is symmetric in (a,/3), and the
second term vaníshes by construction. The algebra ;,; thus closed because the combined
transformatíon (15.45,15.49) is nilpotent. We obtain a.n identity for:

Z (J,K, C, A, L)= f !dp(<,o))exp[ -S(<p) +f dz (J,<p + K A(<p))1

+f dx(A;+L::.)n~(vo)C"], (15.51)

whích, according to the previous ana.lysi.s, ha.s the form:

Z(J,K,C +6C,A,L) = [1 +t: f dx (J;(:r)M:(x) +K(x) 6L~x)) l Z(J,K,C,A,L),

(15.52)
and therefore:

o. (1!'>.53)

A similar ídentity for W = In Z foUows. After Legendre transformation with respect to

J, we then obtaín thf' wr identíties for r:

f dx ( óvo'
~~r~ ~r) 1 ,. ~·,~-
SA, + K bL + ,JIJ,C C éJC"- O. (15.54)

15.3.3 The renormalized action

Arguments, which by now should be familiar to the reader, allow one to show that equa-
tion (15.54) is stable under renorroa.lization and also satislied by tbe e!fective renormalized
action S,:

fd (6S, tiS, + KbS,) +

X IJ<pi 6A, 6L
.!J"
2 ~7
c/3
c
~ éJS, _
aca - o. (15.55)

To solve the equation, we not(: tha.t, as a consequence of fermion number conservation,

A, L and C can appear only in tbe combinations (AC) and (LC). In two dimensíons
lhe sources K, (AC) and (LC) have dimensíon 2, and therefore S, is a linear function of
t.hese sources with derivative-free coefficients:

S, (<,o, K, C, A, L)= S,(<p)- f dx (KA,(<,o) + A,D:"(o;)Ca + LMa(<,o)Ca]. (15.56)

Let"" first write the equation obtained by identifying the coefficient of A;(:r) in equatíon
(15.55):
&D:,~
ârpi ·~
C"'C/3- 1 r Cilc'D' =O
2 (J, '"' •
(15.57)

The ant.isymmetric coefficient of C"'CB must vani.;h:

(15.58)
332 Models on Homogenoous Space• in Thw Dimensíons 15.4

The set of functions n;.,(<P) are thus associated with a non-linear representation of the
group G.
We now identify the ro<:ffident of K:

M,D, (
<P, ,., l(J)C
o
- M.,C
a
{15.59)
6

and the coeflicíeot of L:

(15.60)

The latter equatíon is already implied by the two equations (15.58,15.59). We conclude
tha.t A,(<p) ís in general an arbitrary function of '+'·
Finalty the last equatíon, independent of the dilferent sour<::es, ís:

o. (15.61)

The renormalízed action is ínvariant under the non-linear tra.nsforrnatíons of the group
G generated by D;a('r")·
Let us note that very often one choooos a H symmetríc parametrízation for homoge-
neous spaces G /H. This ímposes addítional """trictions upon the renormall~ad forro of
D:f(IP) and ímplies that S,(ip) ís H-syrnrnetric.
Remarks. The r<:Su(ts obtained by this method are less detailed than th08é obtained in
the case of the linear parametrízatlon. The method bBSed on addíng oources for composite
operators should be used when applicable.
We do not need to again discuss thoroughly the solutíons of the WT identíties which
we have reduced to equations {15.58,15.61}. The latter equation implies a renormalized
form of equatíon (15.40), whicb ís an equation for the renormalized rnetric tensor.

15.4 Symmetric Spacest Deflnition

Syrnrnetric spaces are special homogeneous spaces such that the syrnmet.ry group G
possesses an ínvolutive automorphism, and the subgroup H ís the subgroup of ínva.riant
elements. Consideríng the case in whicb G is oornpact, we show in Appendix Al5.4.2
that H is then a maximal proper subgroup. Equi valently a paríty can be as&igned to the
generators of the Lie algebra and tbe Lie algebra C (H) is the algebra of even elements.
More deto.ils can be [ound in Appendix Al5.4.
Field theory rnodels in two dímensions, constructad on symmetric spaces, haV€ special
properties both on the classicallevel and after quantization, which we examine belaw.
The non-linear a-model providcs one of the simplest examples. In particular once
parametrization of the manifold is chosen, the eudidean a.ction is unique up to a oon·
stant multiplicative factor. This refiects the uniqueness of the rnetric on the manifold
compatible with the group structure. The parity of generators of the Lie algebra leads
to a parity assígnment for the fields, +1 for the composite 11-field, -1 for the ,.field.
The coset space G JH w hen it is oymrnetric ca.n be constructed from a group manifold
in the followíng way; We consider the elements g of a group 0' of the fonn (see
A15.4 for deta.íls}
g
 15.5 Models on Hornageneous Spare.o in Two Dimensions 333

where !Jo,!/1,!12 are elements of I!'> wh:icl> satisfy some conditíons. We can then distinguL'Ih
two cases:
(i) g0 1 and !lt. !/2 are two indep<mdent elements of G. The automorphism is g1 <--+
g2 • We recognize the eoset space I!'> x 15/15, the automorphism exchanging the two
components of I!'> x e.. As manlfolds they are identical to the mrmifold of the group
15. The correspondíng field theory models are called principal chírol models. They are
related to the chiral models studied in Section 13.6.
(ií) The element go is a fixed element different from the identity and satisfies

€=±1,
in which the star operntion is &n involutive automorphism of the group 15 which may
be trivial (typically for unitary groups it caJJ also be the complex con.iugation). The
elements g1 and g, are related by

which ímplies that the elements g have thé form:

9 =(!-Ir Yof • (15.62)

where the elements f vary freely in the group e.. As a consequence:

g•g =a. (15.63)

Then G ~ l!l, thé automorphism of G ís g ~ gõ 1g•9o and H is thus the subgroup of

elements whlch satisfy h g0 1h•9o.
Note that below we extcnd by contínuity the • operatíon to the Líe algebra.

15.5 The Classical Actíon. Conservetibn Laws

In the case of symmetric Spa.ces the classícal action S can alw~·s be written in a simple
. geometrical form, sínce symmetríc spaces can be realized in the goup manifold ítse!f:

(15.64)

·In which g(x) belongs to some ma.tríx representation of 1!'>. The different symmetric
are characterized by the group e. and the constra.ints imposed on g(x).
actlon can be rewritten in terms of the associat.d current A,. whích belongs to
Lie algebra C(15):
(15.65)
the Janguage of gauge theories (see Chaptere 18,19) A,. ís apure gauge.
The action then takes the form

S(g) 12
-- f .
d:z:t.A ... (15.66}

the action (15.64) corresponds a classical equation of motlon. In ali cases a general
of g( x) can be Wl'itten:

g(x) >--+ (1- E• (x))g(x) (l + E(x)), (15.67)

334 Models on Homogeneous Spares in Two Di.mensíon.s 15.5

in which ~(o:) and ~·(x) belong to L (l!l), andare either independent in the case of chiral
models or otherwise related by the star automorphism in arder to preserve the condition
(15.63):
g•g =si. (15.68)

The correspondíng variation of A~ is

where we have íntroduced the couaríant derívative associated witb the current A,., whích
octs ou an element w of the Lie algebra. as:

(15.69)

The variation of the actíon tben takes the form:

ln the case of chira.l models we can restrict ourselves for example to c• = O. In the other
ca.ses, the rela.tion (15.68) implies

In both cases we thus find the equation of motion

{15.70)

which expresse<; the conservation of the current relat.ed to the 0-symmetry •

The non-local wnserved currents. Equations {15.65,15.70) imply, in two dirnensíons,
the existence of an infinite nurnber of uon-local conserved currents.
Let us now define a covariant derivntive n,. by:

(15.71)

We note that the explicit form cf the covariant deriva tive Í5 dilferent ftom the expression
(15.69): Indeed, as also discussed in Section 22.2, thís form depends on the representa.tion.
We o:onsider the linear partia! dífferential equa.tions for rnatrices x:

n,.x(x) u,.a.x(o:}, (15.72)

in wbich X is a functíon of x and the speetral paramet.er 1<.

Let us show that the linear system ( 15. 72) is compa.tible. Setting:

(15.73)

we have to ca.lculate the eornmutator

(15.74}
 15.5 M odels on Homogeflf!DIJS Spo.ces. in TW<J Dimeruions 335

In two dimensions, since 11 "! v, in the last term of the r .h.s. only p "" 11 and u 11 give
a non-vani.olhíng contribution:

(15.75)
The equation of moiíon (15. 70) implles:

(15.76)
The oornmutator [D •, Dv [ is the curvature F,.. associated wíth the connection or gauge
lield A,. (these concepts will be díscu..OO in Cha.pters 19,22):

(15.77)
Equatíon (15.65) ímplies that A• is apure gauge (for more details see Cha.ptem 18,19}.
The operator D,., acting on oome matrix X(x) can then be wrítten:

(15.78)
Thus tbe product DI'D.,
D~'D,.X = g- 1 âl'ô.(gX),
wsymmetric in the exchange (11,11) and the curvature vanishes
(15.80)
The oonclusion is:
[Ll.,.,Ll..,) O, (15.81)
IUldthe system (15. 72) is cornpatib!e.
We can now define the current Ji< (x, 11:):

(15.82)
As a ronsequence ofequations (15.72), tbe current is conserved:

a,.;r,. o. (15.83)
oolution of equation (15.72) has an expansion in powers of"' of the form:

""
X (x, 11:) = 1 +L Xn(x)"-"· (15.84)
n=l

corresponding expansion of J,. then genera.te.; an infinite number of conserved cur-

.J,. (x, ~~:) A,.(x) '"' J~(x)1C".

+L (15.85)
n=l
interesting question, which we will not to investigate here, then ls whether these
t'O~•serva.tion laws survive quanti2.ation. Let us also mention that the corresponding
conoorvatíon laws lead to the factorizatíon of the S-matrix wbich can then
oompletely determined. Vve refer the interested reader to the literature for more
336 Modds an Homogeneous Spaces in Two Dímensions 15.6

15.6 Quanturo Theory: Perturbation Tbeory and RG Functions

To quantize the theory we consider the functional in~;egral Z U):

(15.86)

in which f(x) is a group element either identical to g(z) for \5 x 15/0 or such that g(x)
ís rela~;ed to f(x) by equatíon (15.62):

g= W'f g,r,
and df is the de Haar measure for the group 0. The parameter ). !s the coupling constant.
We can then parametrize f(x) in a form analogous to (15.4), in terms of índependent
field variables ( 0 (:t), components on the generators helonging to CGJH. We can then ex-
pand perturbatíon theory around a finite value of the sourcej (x) to pravide perturbation
theory with ao IR cut-off.
Actually for practícal calculations it is convenient to return to the field representation
(15.4,15.6):

I.P(x)) R(Ç)jO) (15.87)

valid for general homogeneous spa.ces. With these notations the generating functional
Z(J) reads:

To calculate in perturbation theory we expand the action in powers of ( 0 (x). The goomet-
rical po.rt of the Feynman diagra.m calculation then involves the evaluation of averages
of the form (OI t•• t"' ... t•• lO}.

15.6.1 RG functions at one-loop order

We shall eventually discuss the properties of th""" models from the pcint of view of
the renormalization group. Thetefore we here calculate the renormalizatlon constants
and the renormalization group functions a.t one-loop arder. At the sa.me time these
calcu!ations will illustrate some of the previous considerations.
Preliminary remarks. We normalize the vector jO):

{O lO)= 1.

To evaluate the one-loop diagrams, we need also a few tensors:

This relation is a co=equence (up to the normalization) of the property that in the
of symmetric spaces the vectors t"IO) form an irreducible representation of the
H.
1!">.6 Mvdel& on HomogeneoWJ S'[N>ceB ôn Two DímenM.ons 337

In the sa.me way because the structure constants are antisymmetric and the vector !O)
is the unique vector having H for the stabllizer group:

t"t"jO} -= - 1< jO) . (15.91)

In equation (15.91} and in the equations which follow, repeated índices means summation
over indioes running from I to I, values which correspond to the generators of G /H.
Combining equatlom (15.89-15.91) we find the value of p.:

t"t"IO) -ljO). (15.92)

We need two tensors with three indic"". Since the generatora are repr<JSented by a.ntl-
symmetric matrices:

A few commutations a.nd the properties of cymmetric spa<:es lead to:

(15.93)

6
A different useful qua.ntity is obtained hy replacing t by a generator 7fJ of H:

(15.94)

We have still to evaluate three tel150ts with four índices. Combining cornmutation rela-
tions and previous relations """ find:

(OI t•t•thtb lO) = 12 , (15.95)

(OI t"t 6t•t& lO) = !? - 2: t,.b}-,.b, {15.96)
-,>1
(OI t"t6t 6t"ID} (OI t"t6 t"t 6 lO)

Using the antisynunetry of the structure consta.nts and the special properties of symmet-
ric spaces it is easy to vetifY:

I
LL/oabfoa., !õwC, (15.97)
o>ica=l

=
whicb C ls tbe Casimir of the group G, Summing over b b' we find an e:xpressíon
the serond term in the r.h.s. of equation (15.96). Therefore:

(15.98}

Renormalítatian vonstants a! one-wop, To obtain the two on~laop renormalizatian

demand that the functlonal W {J} is <>ne-loop linlte when calculated with
toiiStlUltS, """
renormallzed action S,:
338 Model.s on HomogenEous Spaces in Two Dimensions 15.6

for the special constant source:

IJ(x)) ""'m2 ]0}. (15.100)

Dimensional regularization will be used so that the measure tenn can be omitted. The
action has to be expanded up to order { 4 (.x). Using:

(15.101)

which for symmetric spaces lcads to:

(15.102)

we can "'rite the renormalized oction up to arder ( 4 :

s, (()"' ~f ddx{ ;~ [~ {â"() + ~ ~ {f t:.â~ed]

2
1 ••

- z; 112m 2 (1 ~e+~ (Ojt•t•tctd IO)M~{,(d)} +o (ç"). (15.103)

A short calculation yields W (m 2 ) up to order .\:

W (m2) = Z(; 112 ~ +~f ddg ln (I+~: z; 1 1 2 z~)

2
,\ 2 l (I- C) (
-m
8
f __+
q2
ddq_ )
m2
2

+o (>·2) (15.104)

Setting d 2+é, we ma.ke a Laurent expansion for E small and define the renormalization
constantE by mínima! subtraction:

(15.105)

(15.106)

The coupling constant RG fnnction fi(.\) and the field RG function IJ()I) are then:

(15.107)

(15.108)

In contrast with field theories like .\,P!, the sign of the Jeading term of the fl-function,
in the dimension in which the theory is just renormali2.able (<" = O), is negative. The
physical significance (a.symptotic freedom) ofthis remaxkable property will be d.is<:US5ed .
in ChapteJ' 30.
 15.6 Models on Homageneous S'fK'ce~ in Ttoo Dimensions 33>1

15.5.2 One-loop {J-junctíon and background field melhod

lf we Wl'illt only to calculate the coellicients of the perturbative expa.nsion of the /3-
function, we can shorten the ca.lculatíon by usiiJg tbe ba.ckground fiald met.hod whose
principies h ave been explalned in Appendix A10.3. As an exerci.se and to show some
adva.ntages of tbis rnethod, we agaln perfonn tbe calculation at one-Joop arder. Fbr t.his
purpose we evaluate the vaeuum amplitude Z(6) Blld tbe free energy W(9) in a finite
volume witb non-trivial boundary conditions {for more deta.ih! see Chapter 36): in (d- 1)
dimensions the coordinates x,. vary in the interva.J:

and we impose periodic boundary conditions on the field:

We have called ;: the IBSt coordinate, tbe imaginary time coordinate, whlcb varies in the
interval:
o~ z :5. L,
and for wbich we impose lixed "twisted" boundary conditions:

I.P9 (z: = o,x)) =!O}, {15.109)

in which 9 is a linear combination of generators belongíng to .C(G JH):

8= L:t•B., t• E C(G/H). (15.110)

· As a consequence, momenta io Fourier spa.ce are quantized:

p,. =
27l'
L.1 n,., n,. E Z <1-1
, mEZ.

Tbe Ja.rge L llmit wi!l be taken before tbe large L .L limit.

Th dea.J with the Jongitudina.J boundary conditions it is convenient to set:

i.Ps (~,x)) = e•B/L l1>(z,x)}. (15.111)

new field tben satisfi..s:

!q..(O,x)) = !q..{L,x)) ~]O}, (15.112)

in the pararnetrization:

l<fl (z,x)) = exp !t"{., (z, x)]IO}, (15.113)

{.,(O,x) = {.{L,x) =O. (15.114.)

renonnalized action then reads (8.: 8/8z) :
 Models on HomogeneoUB Spaces in Two Dímension.< 15.6
340

The calculation of the one-loop contribution involves only the expansion of S, (() up to
arder e. Using the relations (15.90) and (15.93):

(DI t•t't< lO) = O,

we find:

{&.<fi ,a. <fi)= (&.€)

2
+o (çJ), (15.116)

f d.. {&.r/118\4>) ~f a,e.e.dz +o (ç 3

) =o (e) , (15.117)

(<t>IB 1<P)
2
= -6.6..+ }Mb (o l[t•, [t',B 2
JJI o)+ o (e). (15.118)

Let. us evaluate th<l lost term:

(15.119)

The action in the gaussian appróXlmation is:

The free energy as a function of o. is at one-loop arder:

(15.121)

where ll..1 is the laplacian in d- 1 dimensions. To compute the renormalization constant

Z;., it is sufficient to expand up to recond arder in 9:

L1-'&.ll. 1
W (6) =- 2L».Z~ + zL2 v••.,.eced L r.+ v'· (15.122)
p~:::::.mw/L
p=2.-n/L_j_

We have shown that in a symmetric space (equation (15.97)):

in whkh C is the Casimir of the group G. .}

Since we want to eva.luate the UV divergences of the one-loop surn for d = 2 +e:, we ·:;~
can replace tbe sum by an integral: ,__.
'~~~
(15.123) .·~

lt follows that Z>. in the mínima! suhtraction scheme is:

~)~

z~ = 1 +.E_ A+ o (A'),
4.-E
(15.124) 1
in agreement with the result (15.105) of Subsection 15.6.1.
:~
'}~~
15.7 Mode~ an Homoyeneou• Spaces. in Two Dimeruions 341

15.7 Generalizations

Non-compact homoyeneous spaces. Up to now we ha.ve restricted the discussion to ho-

mogeneous spa.ces based on compa.ct groups. However most of the arguments can be
generalized in a stra.ightforwmd manner to a ciBSS of homogeneous spaces G /H in whkh
G is non--wmpact. Obvious conditíons are that the metríc 9ij should be posítive, a.nd the
group G unimodular to preserve the G invarlance of the functional measur-e. For example
we ean take G semi-simple. The positivity of the metric implies that H is oompact and
even the ma.rimal compact subgroup of G.
Some care has to be taken to define properly the functional integral since the volume
of the group manífold is in general infinite so tbat appropriate boundary conditions have
to be impoood.
Simple examples of moàels belonging to this class are provided by analytic con--
tinuations of oompact syrnmetric spaces replacing formally in the orthogonal repre-
sentation the generators t" in GfH by it". For example, after this tr!UISformation,
the cornpact Gra.ssmannian manifold O (N + M) /0 (M) x O(N) becornes the manifold
O (M,N) jO (M) x O (N) ln whlch O (M,N) ís the pseudo-orthogonal group leaving the
metric witb M + signs a.Dd N signs invaria.nt.
The perturbative expa.nsion of the correlation functions of these models is obtalned,
up to global slgns, by changing the sign of the ooupling constant in the compact models.
This changes in particular the sign of the one Joop JJ-function in two dimensions, and
has thus important consequences from· the renormalization group point of view.
A rbitrary Riemannian manífold.s. A much wider dass of mo deis h as been studied
hy Ftiedan. One considera an arbitrary smootb. (infinítely differenti&b!e) Riemannia.n
ma.nifokl with a smooth positive definite metric g,3 (<p) and one takes the classical action
S('l'):
(15.125)

The generating functional of correlatíon functions Z(J) is then given b)''

Z(J)= J!dP('I')]exp[-s(l")+ jdzJ,(:r)'l''(:r)], (15.126)

ln which dp( rp) has to be a smooth, strlctly posjtive, covariant measure on the manifold,
for example {see Chapter 22) ·
dp(<p) .Jgd<p,

whlch y.ls the determinant of the metríc tensor.

As we have discu.ssed in Sections 3.2 a.nd 4. 7, in one dirnension the functional inte-
(15.126) is associated wlth a hami!tonian of the fonn of a Laplace operator on the
with rnetric !Jij a.nd ís related to brownian motíon or dilfusi<m pro= on the

dimensions, the.action (15.125) corresponds to theoríes renormalizable by pow-er

which generalizes theoríes on homogeneous spaces. However since the symme-
are lost, and in particular the Goldstone theorem which forces fields to
mBSS!ess, important differences appear after renormalízation:
(i) in general the space of all possible metrics is inlinite dimensional;
342 Model.s on HomOfjenecus Sp<1ces in Two Dímeruions 15.7

(íi) with a given metdc can be a.ssociated an infinite numb~r of different cova.rlant mea-
sure terms since one can construct an infinite number of scalars like the scalo.r curvature
(see Chapter 22), and a covaríant mea.sure multiplied by a sca!ar is still covariant.
As a consequence, in contra.st to homogeneous cases, derívative-free terms will be
generated in the renormalization and it is only possible to maintain the forro ( 15.125) o f
the action by adj usting an infinite number or paro.rneters.
In addition the renormalhed metric will be generically the most general metric on the
manifold. In other words the renormalized o.<:tion is th~ most general 1U:tion allowed by
power counting arguments.
Consíderations based on oova.tiance (Chapter 22) nevertheless a.re useful since tbey
símplify perturbative calculations in the general situation by restricting the form of the
countert.enns at a gíven arder in the Joop expansion.
For example the equivalent of the coupling constant RG function ís a. functiona.l of the
rnetric {3 (9ij ). lt hes the covariance of the metric teru;or. At one-loop it can only involve
first e.nd second derivatives of the metric and therefore R,; the Ricci tensor e.nd Rg;i in
which R is the scalar curvature. By inspection it is possible to eliminate Rg;j a.nd the
constant in front of Fl.,j can be obt.a.íned from a particular model. Friedan has in this way
obtained the first two terms. We have expressed them in terms of g'i the im·erse of the
metric &ensor g,,, beca use it naturally orders perturbation theory. In 2 +e dímensioru;:

(15.127)

in which R'j is obta.íned from the Riccí tell50r by raísing the índices with g'i and R{,1.,
is the curvature tenscr.

Biblíographical Note«
The general group theoretícal problem of non-liuear rea.líza.tions is discussed in
S. Coleman, J. Wess and B. Zumín.o, Phy•. Rev. 177 (1969) 2239.
See a1so
S. Gasiorowicz and D. Geffin, Reu. Mod. Phys. 41 (1969) 531.
Quantization of chiral models is described in
L.D. Faddeev and A.A. S!avnov, TheoT. Mo.lh. Phys. 8 (1971) 843.
Classkal conservation laws and RG functions in symmetric spaces have been discrn;sed
in
V.E. Zakharov and A.V. Mikhailov, Zh. E"'P· TMr. Fiz. 74 (1978) 1953, Commun.
Math. Phys. 74 (1980) 21; V.L. Galo end A.M. Perelomov, Phys. Lett. 79B (1978) 112;
H. Eichenherr, Nucl. Phys. BJ46 (1978) 215; A. D' Adde, P. di Vecchia and M. Lüscher,
Nucl. Phys. B146 (1978} 63; E. Witten, Nw;l. Phys. B149 (1979) 285; H. Eichenherr
and M. Forger, Nud. Phys. B155 (1979) 381; M. Lüscher, Nucl. Phys. B135 (1978) 1;
E. Brézin, C. Itzykson, J. Zinn-Just.in and J.-B. Zuber, Phys. Lett. 82B (19711) 442;
H.J. de Vega, Phys. útt. 87B (1979) 233.
For a recent discu.ssion of quantum conservation Jaws and S-matríx in symmetríc spaces
see
E. Abdalla, M.C.B. Abdallaa.nd M. Forger, Nucl. Phys. B297 (1988) 374 and references
therein.
Add.ltional results concerning non-loca.l currents can be found in
D. Bernard, Commun. Malh. Phys. 137 (1991) 191; D. Bernard lllld A. LeCiair, Com-
mun. Mo.th.. Phys. 142 (1991) 99.
15.7 Models on Homogeneaus Spaces. in Two Dimensions 343

Dilferent properties are described and RG functions calculated in

R.D. Pisarskí, Phys. Rev. D20 (1979) 3358; E. Brézin, S. Hikarni and J. Zinn-JuStin,
Nucl. Phys. Bl65 (1980) 528.
For the application of these models to the meta.l-insulator transition of independent
electrons in a random potential see
F. Wegner, Z. Phyt. B35 (1979) 207; L. Schãfer and F. Wegner, Z. Phys. B38 (1980)
113; S. Hikarni, Prog. Theor. Phys. Suppl. 84 (1985) 120.
Calculation of tbe RG tl-functions up to three and four loops for a da.ss of models can
be found in:
S. Hikami, Phys. Lett. 98B (1981) 208; W. Bernreuther and F.J. Wegner, Phys. Ret1.
Lett. 57 (1986) 1383; F. Wegner, Nucl. Phys. 6316 {1989) 663.
For a discussion of general two-dimensíonal models on Riemanllian manifolds see
D.H. Friedan, Ann. Phys. (NYJ 163 (1985) 318.

Exercises
Exereise 15.1
Show by explícit calculation tbat tbe differential operators

(15.128)

where the functions a; are defined in equation (15,19), satisfy the commuta.tion relations
(15.42).

Exerrise 15.1!
Verify then by direct calculation the covariance of equatíons (15.43) in a reparametriza-
tion of the hnmngeneous space.

Exerciae 15.3
Prove that the tensor gi1 (A15.14) satisfies the relation (A15.7) and is therefore a syro-
metric inverse metric tensor.

Verify explidtly in the ca.se ofthe 0(4) group that the tensor (AJ5.14) is the inverse nf
the metric on the sphere Sa.
 Models an Homagenwus SfXICU in Two Dímensions A!5.2

APPENDIX 15
HOMOGENEOUS SPACES: A FEW AlGEBRAIC PROPERTIES

This app€ndix first describes same additional properties of homogeneaus spoc"" when
considered as Riema.nnian maniro!ds. lt assumes some minimal familiarity with the
elements of differential geometry presented in Chapter 22. The second pa.rt is devoted
to some elements af cl&ssification of symmetrk spoces.

A15.1 Pure Gauge. Maurer-Cartan Equatlons

Mu!tiplying R( I") by a group elernent Ry on the left we see that R- 1 il;R transformslike

R- 1 il;R'-' R; 1R- 1il,RR,.

This implies that the matrices R -'ô,R transform like elements of the adjoint represen·
tation of the group G. They can therefore be expanded on the generators t" and we set.
(equation (15.10)):
R- 1o,R= Lf(tp)t" (Al5.1)
lt is ea.sy to verify more dírectly by pa.rametrizing the Lh.s. as
R et"<",

e
that the expansíon of Lf in powers of involves only commutators of generators of i:(G)
and therefore L f (<p) depends on the parametriza.tion of the group elementll, but not on
the representa.tion to which R(<p) belongs.
The definition (A15.1) implie:s that the quantities Li(<p) are the components of a. V<J<:tor
belonging to the adjoint repres<ontation and have the ,.,• dependence of a pute gauge (see
Chaptero 19, 20 for details). We have shown in Section 15.5 that the corresponding
curvature vanishes (equation (15.80)). In terms of the components Lf one finds

81 (R- 1 81R) o1 (R- 1 il;R) = t• [8,Lj- 81 Lf]

(R- 1 ô;R,R- 1 ô;R] L;' LjJ;!Jf'.
We thus find
(Al5.2)
These relations, which rucpress that the curvature corresponding to the gauge connection
Lf vanishes, are known as the Maurer-Ca.rtan equations.
A15.2 Metric and Curvature in Homogeneous Spaces

A group transformation acting on the coordinates <pi can a.lso be considered as a repara-
metmation of the manifold. The infinit...imal form is given by equa.tion (15.37):

(Al5.3)

The generator ~o of L:{G), as defined bY equation (15.41), then characterizes the corre-
sponding infinitesimal variation or scalars

(AUi.4)
 Al5.2 Models on Homogeneous Spaces in Two Dimensíurn; 345

More generally equation (22.9) defines its action on ali tellllOrs on the homogeneous space.
For veetors V;(rp) it yields

(Aló.5)

Thi.s tr=formation law can be verified by a short calculation in tbe case of the gauge
fi.ld L:{rp), defined by equation (15.10). As explained in Soction 22.1 à,. V. is a. vettor,
BS expected.
For general tensors the re:sult is

With this definítion à, obeys the usual rule of differentiation for products of tensors.
The invariance of the metric, as expressed by equation (15.40), then takes the simple
form
à.,gjk =O ç:. V;D;" + VjD;, =O, {Al5.7)
where the 8€Cond equation follows from equation (22.110).
Cansistency with parollel tronsport. The metríc tensor defines uniquely a torsion- free
parallel transport on the manifo!d. If the infinitesimal change ofvariables (A15.3) leaves
the metric invariant, it leaves invariant ali quantities funetion only of the metric. With
the definition (Al5.6) we can write

à,.Rt,i o, (Al5.8)
à.,R;j o, (Al5.9)
à,;R O. (A15.10)

Th" Christolfel connaction is aiso invariant, but since it is not a. t..nsar the action of à.,
takes the inhomogeneous form (22.33);

à.,r;k- .:I;D~Dk a,D~r}k + ô,D~rj 1 + 8;D~rjk + éljélkD~ o.

The compatibility between paral!el transport and symmetry can then be expressed by
the commutation relation
(A15.11)
Th prove this relation, it is sufficient to verify it on scalars and vec-t.ors, it tben follows
frorn forrning tensor products. For scala.rs it is a.n immediate consequence of the definition
· {Al5.6). For veetors one finds

[à.,,Vi]V; = ykc,..,f'{, =0.

The equ!ltion (A15.9) implies that R;; is an acceptable metric t..nsar and the equation
(Al5.10) that the scalar curvature is a oonstant in homngeneoru; spaces. More generally
"11 symmetric t..nsors with two índices conBtructed from the curvature tensor satisfy the
. equivalent of eqnatlon (A15.9) and are of tbe fonn of a metric tensor. Since in the case
of homogeneous spaces, as we bave shown, the set of metrics fo1111 a ftnitl! dimensional
~r space, only a finitO< number of these II<WlOrs are Iinearly independent.
345 Models on Homogeneous Sp=es in Two Dimensíons Al5.3

A final remark: expression (15.12) for the rnetric show~ that the quantíties LH<t>l play
€55entially the role of a vielbein in the case of homogeneous spaces, tbe only difference
being the constant internai metric Jl.ab·
Remark. In Chapter 15 the coordinates 'I'' are tbemselves fields depending on variables
x~. Then â,.'F'[z} belongs to the spoce tangent to the manífold at poínt cp'(x), and thus
transforms like a vector. Using the metric tensor, we have then constructed scalars like
the a.ction density of (15.8). Moreover in this situatíon, another cavariant derivative Dp
can be defined, which ínvolveo the connection ( 22.20):

(A15.12)

On functions of <p(x) only, this definition is redundant, since D~ can be rewritten in

terrns of the covariant derivative, but it i:s useful when applied to derivatives of the field
'I'· Generalizatíon to higher arder tensors ís stràlghtfotward: one contracts the free index
with a.rp'.
The definítion (A15.12) allm>,•s us to write the classical equation of motion correspond-
íng to w:tion (15.125) in covariant form:

(AI5.13)

Al5.3 Explidt Expressions for the Metric

Let finally give several more explícit expressions for the metric.

AJ5.3.1 Metric tensor and tronsformo.tianlaw

Let us show that, for a general homogeneous space, we can lind !Lil inverse metric tensor
g''of the form
(A15.14)

w here m<>~ is a constant symmetric non-singular matríx. The tensor h as to satisfy the
equívalent of equations (15.40) or (Al5.7):

We replace g•i by the forro ( A15.14)

and use tbe Lie algebra commutation relations (15.43). We then obtain

m.S7 J!,s ( iY,.D~ + D~~) o. (A15.15)

Exchanging -y .... ó in one of the terms, we see that the equation is satislied if ma!l is
solutíon to the numerical equatíon
AI5.4 Mode!s on Homogeneous Space.s in Two Dimenoions 347

In a basis in which the generators !~ are orthogonal by the trace, the structure constants
are antisymmetric and m"~ = m6af> is the solution to the last equation.
In the case of symmetric spaces expressions simplify. The vector !6 10) transforms
under an irreducible representation of H, the matrix ~ defined by equation (15.13):

(Al5.16)

is diagonal
/Jab JJÓM ·
This provide.s another proof of the uniqueness of the metric.
The tenoor g'1 (<p) (equation (AJ5.14)) is a possible bwerse metric tensor. In the case of
symmetric spaces the unique metric, in a basis in which the generatcrs ta are orthogonal
by the trace, is thus,
(Al5.17)
up te the normalization.

AJS.9.2 Manifolás embedded in euclídean &pace

lfwe know an embedding of a Rlemannian manifold !lJt in euclidean spa.ce, we can describe
it with constrained eudidean coordinates, as we have dane in the case of homogeneous
spaces in Section 15.2.
Let (o', !P') be such a set of euclidean coordinates. We assume that locally the o• ca.n
be expressed as functions of the independent coordinates !P':

(Al5.18)

The metric tensor in this representation is obtained from

(A15.19)

and therefore:
Y•i(!P) ti,,+ a,"'aj,.•. (AI5.20)
A short calculation shows that the connectíon ha.s the simple form:

(Alf>.21}

and the curvature tensor is given by:

(Al5.22)

in which N is the dímensíon of !lJl and tr g- 1 the trace of the inverse of the metric:

(Al5.23)

Al5.4 Symmetric Spaces: Cla.sslfication

ln this appendix we examine a few símple properties of symmetríc spaces and provide
wme elements of classi.fication~
348 Models on Homogeneous Spaces in Two Dimen.rions A15.4

.415.4.1 Definitwn
Let us consider a semi~simple compact Lie group G and assume that we have constructed
a non-trivia.I involutive automorphism of G, which to an element g of G associates an
elementg:
(g,g2) ~ üt!l>' with g ~ g.

We consider the coset space G I H obtained by taking for H the subgroup of invariant
elements under the automorphism:
(Al5.25)

The automorphism can be extended to the Lie algebta C (G). It then becomes a refiection
a.nd each clement of C { G) can be deeomposed into a sum of an even and a.n odd element.
Even element.s belong by definition to C. (H) and the genemtors of L.( H) are denoted by
r". The genera.tors of .C(G) not belonging to C.(H) (we denote the corresponding vector
space C (G I H ) ) can be chosen odd. We denote them by t•. We choose the Lie a.lgebra
structure constants fi;k comp!etely antisymmetrk. We then have the rules:

t"EL(G/H);
(Al5.26)
r"E f:.(H).

lt follows:

I•", .,.a] f,fJ..,r'", (Al5.27)

[r", t.~] f,r,,f, (Al5.28)
[t•, t"] = t ••,,.-r. (AI5.29)

Note tha~ in the ca.se of a compact group only the last set (A15.29) of commutation
relations is characteristic of a symmetric space, since (Al5.28) is then a consequence of
(Al5.27) and the antísymmetry of f~~<·
Prelímínary n:marks
(i) \Ve wíll consider symmetric spaces derived from non-simple groups G, bvt we want.
to exdude the possibility that:

G G 1 xG,,
H= H 1 x H, with H 1 C G 1 ,H, c G2, (Al5.30)

because in this case the coset space decomposes into two independent spa.ces G 1 f H 1 and
G2/ H,. In particular tbis exdudes the trivial situation G, ;;; H, s.nd this property will
be used in wbat fol!ows.
(ii) All genera.tors of f:.( H) ca.n be obt.alned a.o; linear combinations of commut.a.tors of
generaton; of t:.(G/H). lndeed assume a generator 76 co.rmot be obtained. We ca.n then
rearrange the generators in such a. way that iaM O. It fol!ows that •~ commutes with
L.(GjH) snd thus with ali genera.tors of L(H) wbicb can be obtained as commutators
(Al5.29). We are exactly in the situation we just excluded.
A15.4 M ode!s on Harnageneous Spo.ces in Two Dirnensíons 349

A15.,f.2 A Ba:~ic pro]l"rty

The purpose of this appendix is not to present a complete description of the mathematical
pmperties of symmetríc spaces. However a fuw of these pmpertíes are diractly relevant
to the problem of renormalizatíon and can be derived by elementar)' wethods. A very
important property is the fo!lowing:
If a homogeneous space Gf H is symmetric, H is a maximal proper subgroup of G.
To prove this a.ssertion Jet us assume that there exists a subgroup G' of G which
contains H, and exhibit a contradiction:

Note first that G' I H is then also a symmetric space.

1ft"" belongs to .C (G) -.C (G') and t>~> belongs to C{G1) -.C (H}, then [t"", t'b] belongs
to .C(H) from (A15.29). However the relations (A15.2B) and the antisymmetry of the
structure constants imply that such a commuta.tor vanishes. Sinoe L(H) is obtained
from the commutato<s of generators in C(G' f H), agaín we find that the generators t~
commute with L:{G'), and thus
G=G'xGtlt
in whicb H is a subgroup of G', sítuaúon we have excluded.
Therefore the maximality of H has been derived. Several other lmportant propertles
of symmetric spaces follow.
A few consequences
(i) The maxímalíty of H has one very important consequence: the generators {t•}
form a real irreducíble representatJon of the group H.
To derive this result, let us e.ssume the converse, i. e. that the generators { t"} can be
divided in two representatlons of H, L 1 and 1:2 :
t" E Ct,;. [t~,T 0 j E L 1 ,
t• E C2 => lt",Tuj E L2.

Since as a consequence of (Al5.29) the commutator of two elements of C 1 belongs to

C( H), C. (H) Ell C 1 forms a subalgebra of C. (G} and H is not maximal. The converse ís
ubvious: lf H is not maximal the representation is reducible.
{ii) Let us assume that we h:we constructed the space G1H in the manner described in
Sub:;ection !5.1.1. From the previous result.s we conclude that the field <f>(x) of equatjon

L
(15.4):

<i\;(x) = [exp (t.€ 0 (x)t") IJ>'j,

l>elongs to an irreducible representation of G. Also simple consideratiollS show that since

t.he {t"} form a.n irreducible representation of H, the vector 1/>' is unique in the following
:rense: given H there exists a unique ve<:tor <f>' Jn the representa.tion which has H as Jittle
group (stabilízer).
Therefore to each symmetric space is assodated a unique cla.ssical model. The quantum
model will be defined in terms of a unique coupling constant, and the perturbative
c>.lJansion is equally unique. The general a.rguments on homogeneous spaces given before
l.ell us tha.t two renormalization constants are suffident to renormalize the mode!.
We now describe symmetric spaces correspondíng to orthogonal and unitary groups
G.
350 Models on Hom01Jeneow Spaces in Two Dimensions Al5.4

AJs.,;.s Th• principal chiral mode!s

We first examine the case in which C is not simple and fBCtcrizes intc C \!> 1 x 152 •
Since we have exduded the situation (A15.30), the automorphism miiSt map elements of
\!> 1 into 1!1 2 and vice V€rsa. This implies 1!1 1 = 1!5 2 and the autcrnorphisrn is:

(g,,g,) =(!12,!lt) '

The subgr<~up H is then given by elements of the form

The symmetric spBCe l!l x l!l/l!l is isomorplúc to the group space l!l ítself. A canonkal
1
reall2ation isto consider gTOup elements of the form g! 92·

AJ$.4.4 Simple groups

We now assume that the group G ís simple. We decompooe the automorphism into tbe
product of an inner automorphism and a remaining irreducible involutive automorphisrn:
-1 •
Yo 9 9o,

with:
g;g; (g19z)'.
(g')' =g.
The condition g ~ g then has the form:

The element !lÔYO oommutes wjth ali elements of the gr<:>up. This implies that g(igo
belongs to the centre of the group:

1 being the unit matrix in the delining representation, and), a pha.se factor which reduces
to ,\ = ±1 for orthogonal graups.
The subgroup H is deflned by the invariant group elements h:

A realization of G/H in the group space will be given by group elements of the form

s (r')' 901 'I f E C. (Al5.31)

Indced if we multiply f on the left by an elrunent of H, 9 ís not modilied.

Note that these gTOup elements satisfy:

g*g ÀL (A15.32)

We can now classify symmetric spaces ccrrespondlng to orthogonal and unitary groups:
Al5.4 Mude!s on Homogeneous Space• in Two Dimen.rions 351

Orthogonal groups. The group G ís O {N), the sta.r automorphísm is the identity and
À= ±l.
(i) À= +1
We take !lo diagonal without Ioss of generality. It has only ±1 as eigenvalues. If
it possesses p eigenvalues +1 and N p eigenvalues ~1, the subgroup H is clearly
O{p) X O(N -p).
The syrnmetric space O ( N) f O (p) x O ( N - p} is ca!led a real Grassrnannían mani-
fold. The O(N} non-linear d-model corresponds to p I.
(ü)), = -1
This implies that N must be even: N 2N'. Without loss of generality we can choose
9o of the form:
o
90 = [ -lN'
a.nd the subgroup which commutes with 9o is isomorphíc to the unitary group U (N').
Unita111 !)1'0Ups. The group G is U(N).
(i) The star automorphism is tbe identity. The phase À ís irrelevant. Taking >.= 1, one
sees that go h as only ± 1 as eigenvalues. Jf it has p eigenvalues + 1 and N- p eigenvalues
-1, the subgroup H is U (p) x U (N- p).
The symmetrlc spaces U (N) fU {p) x U (N- p) are complex Grassmannian maní-
folds. The case p 1 corresponds to the complex projective spac<l C PN -l·
(ii) The star automorphísm is the complex conjugation. The condition BÓBo = À1
then implies >. = ±l. lf we take À + 1, we can diagona!ize 9r> by an orthogonal
transformation and then set it equal to one by a diagonal unitary tr11nsfonnation. Since
the elernents h of H then satisty:
h h·,
the subgroup H is the orthogona.l subgroup' O(N) of U(N).
!f we take À = -1, we ngain see that we must have taken N even

N 2N'.

We can then cl10ose go of the form

1"'o '] .
subgroup H is delined by the elements h which satiscy:

1N']h=h•[
o -1,..,.
0 lN']
o '
is by de6nition the symplectic subgroup Sp(N) of U(N).
16 SLAVNOV-TAYLOR ANO BRS SYMMETRY. STOCHASTIC FIEL O
EQUATIONS

In Section 1&.3 we have naturally been led to introduce a transformation depending

on anticommuting para.metets, to prove tbe geometrical stability of homogenoous spaces
under renormalization. Tbere is a set of topics, stochastic field equations, gauge theories,
in which similar tra.nsformations are met. These new problems have one common feature:
They ali involve a col15ttaint equation to which, by a set of formal transformations, is
assoçiated an effective action. This action has an anticommuting type symmetry which
has no geometrical origin but rnerely is a consequence of these transformations.
In this chapter we first discuss thls mathematical structure from a rather ahstract
point of view. We explain the appear8llce of Slavnov-Taylor symmetry, which is a con-
ventional non·lineax symmetry, in the integral reprewnta.tion of canstraint equations. We
then show haw it leads to a symmetry with anticommutíng parameters first discovered in
quantized gauge theories by Becchi, Rouet and Stora and hereafter called therefore BRS
symmetty. lts generat<JT has a vanishing square, and generalizes exterior differentiation.
We show that this symmetry is remarkably stable against a number of algebraic defor-
mations. In some cases it can be expressed in compact form by introducing Grassmann
coordinates. We show how BRS symmetry can encode the rompa.tibility conditions of
a system of linear first arder differential equations. We exhibit the special forro BRS
symmetry takes when the constraint equations apply to group manilolds. Finally we
indicate hmv, as a generalization of BRS symruetry, a richer structure called supersym-
metry emerges. Thraughout most of this chapter we use a notation adapt.ed to a finite
number of degrees of freedom. However, the extension of ali algebraic identities to field
theories is straightforward and in the last section we discuss the role of BRS synunetry
in this latter context.

16.1 Slavnov-Taylor (ST) Symmetry

Let <p0 be a set of dynamical variables satisfying a system of equations:

where the functions Eu (<p) axe smootb, 8Jld E. Ea('P) is a one-to-one map in some
neighbourhood of Ea O which can be inverted in 'Pa <p"(E). Thill implles in partk.
u]ar that the equation (16.1) has a unique solution '1'':. We then col15ider some function
F{'P) and we want to give a formal expression for F (<p,), without solving equation (16.1)
explicitly. We can then use equation ( 1.28) and write'

F(tp.) =f {I]dE~~(E.,.)}F(<p(EJ)
=f{fl d<p"6IE,(tp)] }.7{<P)F(<P).
"
with:
 16.1 Slavnot~-Taylor aná BRS Symmetry 353

We h ave chosen E., (\O) such tha.t det E is positiVe.

Notation. lt will be convenient throughout this chapter to use the nota.tion

An invariant measure. The measure dp(ip):

dp(\0), .J(<p) n dtp", (16.4)

has a simple but ímportant induced property. The mea.sure TI" dE" is the inva.ria.nt
measure for the group of translatíollll E, .... E.,+ v.,. It follows tha.t dp{tp} is the
invariant measure for the translatíon group non-linea.rly realísed on the new coordinates
'Pa (provided "" is srnall enough):

<p" ,_.. 'P'" with E, (lO') "" En('Pl· (16.5)

With the corresponding inva.riance is associated, in gauge theories, the Slavnov-Taylor

. symm..try.
The infinitesimal form of the tra.nsfonnation la.w can be written more explícitly:

(Hl.6)

We shall see !ater that these somewhat straightforward rema.rks lead, when applied to
lield theories, to very useful identities.
Reciproca! praperty. Conversely let us cha.racterize the general form of ncn-!inea.r rep..
re;ent.ations of the translation group. We write an infinitesimal group transformation:

(16.7)

which the matrix M.,tJ(\0) has to be determíned. Following the strategy explained in
15.3, we ímpose to the dilferential operators t.0 :

(16.8)

form a representation of the Lie a!gebra of the translation group, i.e. commute:

(16.9)

(16.10)
354 Slaunov-Taylor and BRS Symmelry 16.2

This implíes (for a simply connected <p-manifold, an implicit assumption througbout the
whole chapter) that Map has the forro:
(16.11)

Let us now characterize the invariant meWlUre .J(JP)d'l' for these non-línear transfor-
mations. The varia.tion of .J (<p) has to cancel th.- jacobian ooming from the cha.nge of
variables corresponding to the transformation (16.7):
(16.12)

This yields a system of partia! differential equations for tbe function .J(<p):
Bo ln.J = !M-'pPa~M~o. (16.13)

The equa.tion (16.10) is, as expe<:ted, an integrability condition for thís system.
use it to rewrite equation (16.13):

ao 1n.J =- Bo lndetM,
which has the solutí<m:
.7 = const. det M .
We sha.ll derive similar identities !ater in this chapter.

16.2 Effective Action and BRS Symmetry

In Field Theory the non-línear and as we sha!l see !ater non-loca.l character of trans-
formations {16.5) and (16.6) leads to many technkal difficulties. Remarkahly enougb
the infinitesimal transforma.tions (16.6) can he repiaced by a linear anticommuting type
transformation at the price of introducing additional variables.
Let us age.in sta.rt from ídentity (16.2) ond first replace the 6-function by its Fourier
representation:

The À-íntegration runs along tbe imaginary axil;. As a consequence of the rules of fermion
integration we can also WTite the detennina.nt as an integral over Gra.ssmann variables
c" and ê 0 :
detE =f fl(dc0 de")exp (C"E0 pd').
n

Expression ( 16.2) then becomes:

F (<p,) = .N f I) (d<p 0 dc"de"d>."') F(<p) exp [-S{<p, c, E, J.)l,

in which .N is a constant normalization factor and S ('I', e, c,>.) the qnantity:

S(<P,c,c,>.) .\0 Eo('l') -C"Eap('f')d'.
In field theory problems we shall ca.ll S the effective action associated with the equa.
tion (16.1). Quite surprísingly the function(al) S bas a new type of symmetry, which
actually ia a consequencE! of the inVRriance of tbe measure (16.4) under the group .
transformations (16.6), and whirn we explain below.
16.2 Slavnov-Taylor and BRS Symmetry 355

lfi.!U BRB symmetf!l

The BRS symmetry, lirst disoovered in the quantization of gauge theories by Becchi,
Rouet and Stora (BRS) (see Chapters 19 and 21 ), is a fermionic symmetry in the sense
that lt transforms bosons into fermions and vice versa. The parameter of tbe transfor-
mation ís a Grassmann variable, an anticommuting constant E. The variations of the
various dymunic varíables are:

{ Ó<pu =[rf'' fie" =0,

(16.20)
6C" ~[Àa, 6), 0 =0,

tP o, Eé' +Cu€ =O, W+Uf' o.

The transformation ís obviously nilpatenl of vanishing square: óz O.
The BRS transformation can be represented by a Gra.ssmann differential operator "D,
when acting on functions of {<p,c,c,J.}:

(16.21)

The nilpotency of the BRS transformation is then expressed by the identity;

(16.22)

Tloe differentíal operator "D is a simple generaliz,tion of the exterior differentiation of

rlifferential forms (see also Sections 1.6 and 22.1).
The equation (16.22} implies that all quantíties of the form "DQ ( <p,c,c, >. }, quantities
we cal! BRS exact, are BRS invariaJJt. We irnmediately verify that the function S defuled
by equation (16.19) ís BRS exact:

S = "D JC" E,(<p)j. (16.23)

follows that S is BRS invarianL

"DS =o. (16.24)

property plays a.n importa.nt role in particulax in the discussion o! the renormalim-
of gauge theoriea.
mea.ning and implications of the BRS symmetry will be discussed in the coming

Remarlc. Analogous expresslons can be derived for a ISY'ltern of Gras5lllann equations

Grassrnann variablea. The Lagrange multipller À is then a Grassrnann va.riable.
properties of change of variables in a Grassmann integral imply that the determinant
is repl~ed by the inverse of a determinant. Therefore the &uxiliary variabl"-" which
to introduce to repres<mt the determinant are now commuting variables. lt can be
however, that the final integrand ís sti!l BRS symmetric and many cOnBiderations
to the BRS symrnetry apply to both the commuting and ant.icommuting cases.
shallnot discuss this question any fUrther here becnuse we sha!l encounter no equation
Lhis type in this work.
356 Slavnov~ Taylor and BRS Symmetry 16.3

16.11.2 Relation between ST and BRS symmetry

A simple way of understandipg the relatíon between transformations (16.6) and (15.2(})
is to note that:
(16.25)

Therefore if we fa.ctorio:e the integral over c &nd c, we can rewrite inside the integral:

(16.26)

BRS transfonn&tions oorrespond to:

(16.27)

A form of thls argument can be used to prove that as long as only averages of functiol'S
of 'P are concerned, the consequences of ST or BRS symmetry are the sarne. The BR.S
symmetry extends the tranBformations to functions of <P, c, i': and À. This exteru;ion is
usefu! for two reasons:
(i) The transformations (16.20) are linear, while the transformation (H5.6) is non-linear.
We have already seen in the case of the non-línear a-rnodel that non-linear tro.nsfor-
mations cou!d be linea:rized at the price of introdudng auxiliary lields.
(ii) More important, in lield tbeory transformations (16.20) will be local, in contrast
to transforma.tion (16.6).
These two properties greatly simpiify the discussion of renormalization of va.rious field
thmries.

16.3 Grassmann Coordinates, Gradient Equations

A more compa.ct representation of BR.S transforma.tions is obtained by introducing a

Gro.ssmann coordinate é and then the followíng two functions of é:

4>" (U) = <p" + Oc", (16.28)

With this notation the transformations·(J6.20) simply become a tra.nslation of 9:

J 64>" (O)= l:; =ti>" (il H) -4>0 (é),

l óCo (é) =~a~; C'(~+<")- Õ" (O).

(16.29)

In particular the BRS operator 'D is represented by aI ôÕ:

a
'D .... ãB.
Let us note the identíty:
16.3 Slavruw-Taylor and BR$ Symmetcy 357

We recai! that the integration over 6 selects the coefficl<!nt of é in the integra.nd. Therefore
S (ip,c, c,>.) can be rewrítten as an integral over 6:

s =f déta (8) E,. (4> (6)). (16.31)

In thls expression the BRS -symmetry is manifest: the integrand does not depend ex-
plicitly on 6. Another consequence follows: in Chapter 1, when we have delined fermion
integration, we have emphasized that íntegration and differentiation in the case of Grass-
mann variabies were identical operations. Thus:

S(ip,c,c,>.) = ~ {C (8)E,..(<,t>(6))].
0
(16.32)

We reco""r equation (16.23) in a different notation.

In eddition S has the property that it involves only a fermionic combination of the
!orm fu. In a field theory langua.ge, if we assign a fermlon charge + 1 to c !Uld -1 to
é, S conserves the fermionic charge. This implies that in a representation in terms of
the functíons (16.28), as in e<JUation (16.31), each integration over 9 is associated with a
!actor é {e) .
0

Gmdient eguati<Jn8. The two Grassmann variables é' and C' that we have introduced
in the preceding sectie>n play in general totally different roles. There is however an
exceptional sítuation in which a symmetry is established betwoon them: it is when the
matrix E,..~ is syunnetric:

(16.33)

Hence in this case Ea('l'} i.s itself a gradient;,there exists some function A(<p) such that:

(16.34)

" 1'his symm.etry between c and c dearly generates an additi<mal independent BRS sym-
nwtry. It is thus natural to intmduce two Gra5smann variables 8 and B, and a function
!/>" (11, 9):
4>"' (6,8) 'I'"+ Üc" + iNJ + iie>.a. (16.35)
In terms of t/> the expressie>n (16.31) quite generally reads:

S(<i>)-"" f 8
d8diio : 11"'E.,(<t>(B,õ)J. (16.36)

the function E('l'} has the particular form (16.34), ít io possible to integrate by
over 9 and the function S(t,l>) then ta.kes tbe remarkable form:

S(<,t>) =f d9dBA[<,t>(6,6)] = VVA(<p). (16.37)

two symmetríes, which correspond lo independent translations of Õ and 8, are here

and we have d<snoted by f> the generator of the second BRS symmetry.
358 Slavnov-Taylor and BRS Symmetry 16.4

16.4 BRS Symmetry and Compatibility Condition, Group Mànifolds

The pllrt of the BR.S operator whiclJ acts on the variables '{'0 is identical to the diJferen-
tiation of forrns. The Gra.ssma.nn varlables however plny a dilferent role. In the case of
dilferential forms they are externai vnriables, introduced for convenience to exhibít the
antisymmetry of the correspooding tensors. Here instead the c"'s are addítional dynam-
ical vnriables. In particular it may be conveníent, as we sbow below, to change variables
and to set

It is of course the second equation which is cha.racteristic o[ the dilference. lf we now

rewrite the tra.nsforma.tion (16.20) in these new varia.bles it takes a more complicated
form {omitting the primes on the new variables)

(16.38)

where the fundions D~(rp) and fê~(;o), which is antisymmetric in (3 ..... ")", can be ex-
pressed in terms of F and V. Note that the BR.S generator V also now takes a more
ccmplicated form but the equation (16.23) is still true.
More generally we can consider sit ua.tions in which physícal quMtities dep<md on some
variables A', themseJves functions of the cp". Then BRS transformations will have the
form
{16.39)
Let us then express the condition that the BR.S operator V is nilpotent, V 2 = O, directly
in terms of D and f. We identify the terms cubic and quadratic in the Grassmann
va.riables r? and write thaL the properly antisymmetrized coefliei€nts vanish

{f!p !',! + D~B;Jp, tfJ, =O, (16.40)

D~ô;Dh D{AD~ t!JlD~, (16.41)

where the global subscript a(J")" in (16.40) means antisymmetrized in the tbree indices. In
equation (16.41) we immediately recognize the equation (15.43) of Section 15.3, i.e. the
commutation relation between generators of aLie algebra in a non-linear representation.
Moreover if fji7 is independent of A, equation (16.40) ís simply the Jacobi identity for
the structure constants o[ the Lie algebra..
Compatíbilily conditíons. The extension we find here can be understood in the following
wa.y. Let us consider the set of first order partial diff.,.ential e<JUations

ó.,S(A) =0, (16.42)

where the 6., are tbe dilfereutial operators

As explained in Section.s 13.1.2 a.nd 15.3 the system (16.42) is called compatible if t.he
equations l6a,6/JIS =O are linear combinations of t.he initial e<juatíons (16.42). Then

(16.43)
 16.5 Slatmo!J-Taylor anil BR!j Symmetry 359

We have encountered before only examples where the structure constant. JJfJ were A-
independent (i.e. associated with Líe algebraa), but this is not the general cltuation.
The eQuation (16.43) itself has an integrability oondítíon, tbe Jaeobi identitics since
the l.h.s. is a commutator. A short calculation yie!ds the oondition (16.40). Therefore the
nilpotency of the BRS operator encodes the compatihility of the linear system (16.42).
Note final!y that the eQUation (16.42) is eQuivalent to VS(A) O. If it ôs has non-trivial
solutions these solutions cannot be cast into the form S(A) VS'(A, c), i.e. are not BRS
exact.
BRS symmetTil and group manifolds. Group manifolds provida a simple example of
such a situa.tion. !f the varia.bles 'Pa parEliiietrize a group element g(<p) in some matrix
representatíon, it ís convenient to rewrite BRS transformations on g(<p) directly. This
can be most easily done by noting that witb r/J (O) (delined by eQU!Itíon (16.28)) we can
ElSSOcíate some group element g (O), on which BRS transformations acrording to equation
(16.:29) read:
(15.44)

However, sínce g (6) is a group element, it is natural to parametrize ít unde~: tbe form:

Q(O) = exp (tic) 9 = (1 + éc) g, {!6.45)

in which c is now a Grassmann matrix belonging to tbe Lie algehra of the group. ln
component form the transformation {!6.44) then becomes

óg=tcg, (16.46)
éc = tc2 • (16.47)

lntroducing matrices t, generators of the Lie algebra and parametrizíng c as

(16.48)

we can rewrite equation (16.47) as:

(16.4.9)

In this forro we recognize the transformation (15.49). Moreover the tra.nsformatíon

(16.46) applied t.o the matríx R of the representation (15.6) leads to the tra.nsforma-
tion (15.45) for the fie!d <p(x). We have thus found a geometrical interpretation to the
tr!lllsfonnations used in the derivation of WT identities for homogeneous spa.ces in an
Mhitrary system of coordinates.

16.5 Stochastic EquatioDS

Let ue now assume tbat equation (16.1) depends on a set of stocbastic variables v 0 with
· probability distribution dp(11):
E., {'I', V)= o. (16.50)
.. The solution 1(' 0 of tbe equation becomes a stochastic variahle. Quantities of interest are
now average values of functions of <p:

(16.51)
 Sla~~nov- Taylor and ERS Symmetry 16.5
360
After the set of tra.nsformations described in Section 16.2, this equa.tion becomes:

(16.52)

with S given by equation (16.19):

S = À0 Ea (.p,v)- é' Ea~ (.p,v)C. {16.53)

We ha.ve shown tha.t S ha.s a. BRS symmetry. The irnportant remark is that the function
E(.p,c,c,.l-) obta.ined after noise aveTaging:

exp!-E(.p,c,é,.l-)] =- j dp(v)exp!-S(\O,c,é,).,v)J, (16.54)

is still BRS symmetríc, although it no longer has the simpie form (16.53), i.e. o. function
linear in), and tr.. Using the definition (16.21) of the BRS generator, we can write the
BRS syrnmetry as: (Hl.55)
VE G.

16. 5.1 Stocha..,tic equations linear in the noise: A •imple example

In the coming chapters exarnples of stocllastic equations of the simpie form
(16.56)

will be met. 1ntrodutíng the Laplace transform of the measure dp(v),

.,w(>) =f dp(v) e"~"~'

we obta.in for the function E defined by (16.54):

This is a minimal modification of expression {16.19).

Remarl:s.
(i) After integration over tbe noise, the expression of the function I: in the notation oi
Grassmann coordinates is in general rather complicated. However, in the case of equation
(16.56) with ga.ussia.n noise the additional term w().).: ~w 0 11Àa )l.tl is represented in the
notation (16.35) by

(ii) In the latter case it also possible to integrate explicitly over the ). variables.
resulting integra.nd correspond to
16.5 361
Slatn1<111-Taylor anti B~ Sllfllnletrl/

The BRS tra.nsformatíon of ~ is now as exp<>eted non-linea.r

6snsC" qw- 1 ]".BE11 .

We note that in this form tbe BRS tra.nsformation of c still j,; nilp<>tent but no longer of
vani•hing square. Therefore the Jatter property oi tha BRS transformation may not be
apparent in ali formulations and may re<juire the introduction of aome auxíliary variab)es.
A sufficient conditwn. We now want to characterize the mOBt general BRS invariant
function of the form

ldentifying the coefficient of &? and c in equation (16.55) we obtain two e<JUations:

c"i!'ôaM~..,c' O, (16.60a)
caô,..E->."M,..t'('P,À)d' O. (16.60b)

The equation (16.60a) implies that the antieyrnmetric part of tbe coefficient of r!'c>
vanishes
(16.61)
ond therefore the matrix M has the form (again assumíng the rp-manifold to be simply
connected):
(16.62)

The e<JUation (16.61) is the integrability condition of e<Juation (l6.60b} which now can
be soived:

(16.63)

The complete solution thus ís

(16.64)

ú; BRS exact:
(16.65)
particular expressian (16.58) is the mOBt general BRS symmetric expression of the
(16.59) with M,..(! independent of >..

The general linear a~se

olightly more general form will often be met:

{15.66)
 Slaunot>-Taylor and BRS Symmetry 16.6
362

Let us now consider the special example of gaussian stochastic variables v with probability
distrib1>tion:
(15.67)

After integra.tion over v a.nd some algebra. we find that E can be written:

E(<p,c,ê,À) = V~(<p,c,ê,Ã),
~(op,c,c,À) =<f' E,- !C" (w.,~.>.ll w~~ •..,.cilc'),

wíth the definitions:

E differs from S by the addition of a function quadratic in both À and ec.

Conversely one may ask the question: wha.t is the most general forro of a function E
qua.dr~>tic in {À, ec} and BR.S symmetric:

VE""O.

The answer, as we shall discuss in next sectíon, is that E i.s BR.S exact:

E=V~, (16.68)

with: (16.69)

16.6 Consequences of BRS Symmetry: General Discussion

The property that the most general BR.S invariant quantity is BR.S exact (in t.he case o(
simpiy connected manifoids), as discussed in Subsection 16.5.2, is a simple consequence
of a similar property for dífferential forros. However because this property is important,
let ns discuss it in the present context. Let us agaín introduce nn a.nticommuting variabl<>
eand tbe functions 4>(6) and C (11). The equation (16.54) can then be rewritten:
exp [E (q,,C)] =f dp(v)exp [-f dÕC"' (õ) E,. (4>(6} ,v)).
Expanding the exponential in the integra.nd in power series and integrating term by term,
we obtaín for E (.P, é) a. formal expansíon:

E (q,,é) f~
l .
j d/1, ... dll,. é"'• (9,) ... é"'n (9,.p.:t>.,.n [.P (91), ... , lt (9,.)].
Conversely we observe that the expansion bas the most general form aliowed by "'""''w,;;c;;~
tion invariance in 9 (BR.S symmetry) and fernrlon number conservation.
Since the BRS operator V corresponds to a translatíon of the coordinat.e 9, on
function of é (8;) and 1t (8;), as those which appear in the r.h.s. of expression (16.71),
acts as: n (!
v Leu·
i=l t
16.6 Slavnov- Taylor arni Bfl:S Symmetry 363

The identity between differentíation and integration for Grassmann variables then lewis
to:
V f diJ, .•. dOn = ~f dô,d/h ... dÕ,. = f diJ1 ... dÕn (16.73)

Hence the expression (16. 71) can be rewritten:

~ (IJ",c,c,>.) =DE (IJ",c,c,.l.), (16.74)

wíth:

I: <<P. c,;;, J.}

From this a.na!ysis we draw two conclusíons:

(i) Even after the most general integratíon áVet the noise the functíon r: remains BR5
exact and thus, since 'D2 0 BRS invariant,
1

(i i) Conversely, in the case of simply connected manifolds, any BR5 symmetric function
ís BR5 exa.ct:
v~ o,~~ vÊ. (16.76)

A n important consequence. To appreriate the signilicance of thís result, let us consider

tbe following integral Z:

(16.77)

nnd calculate the variation of tiZ of Z induced by an infinitesimal variation 6Eo(IP) of

the function Ea('P)·
'Th this variation corresponds a variation 6!: of r: whicb, occording to our previous
result, can be written:
(16.78)

Therefore 6 Z h as the form:

6Z =f If (d'f'0 dC"dc0 dÀ0 )D[6E]exp(-~). (16.79)

Since V is a differential operator we can integrate by parts:

6Z f f} (d<p"dE"dcDd).n) liÊD~exp( -E). (16.80)

then the BRS symmetry condition DE "' O we condude:

6Z o. (16.81)

or course this result is not surprísing since by construction z is a constant índependent

of E 0 ('f')· However it teUs us that equation (16.76) implíes by itself this independence
. without any further assumption about the form of E.
This result bw; impartant implications in quB.Iltízed gauge theories.
364 Slavnov- Taylor a rui BRS Symmetry 16.7

16.7 Extension of Grassrnannian Syrnmetries

A trivial extension of BRS transformation can be constructed by considering a function

of severa! Gra.ssmann variables f (li1, .•• , 8n) and performing independent translations on
eacb variable. The generators of these translations are the diiferential operators â/89;.
The equivalent of the usual commutation relations between Lie algebra generators are
replaced by antic:ommutation relations:

a a a a
~-+--
aa, aej aei &li; o. (15.82)

Translatíons on Grassmann variables form the equivalent of usual abelian Lie groups;
non-trivial extensions, "non-abelianft, will cor.espond to non-vanishin.g anticommutators.
First a few general rema.rks are in order:
(i) lf D is the unique generator of this generaliwd structure called "supergroup" then
exponentiation irnplies

in which e1 and .:2 are Grassmann Vllliables. Expanding both sióes we obtain the condi-
tlon:
(16.83)
whkh is indeed satisfied by the generators of translatlons. lf this conditian i>; not satisfied
the supergroup has two generat.ors D and D2 in which D2 is a cornmuting differential
operator.
(ii) More generally the anticommutators of generators are even elements of the Grass-
rnann algebra. Thcy form an or<linary Lie algebra. If we cal! D;, La the anticommutíng
and ccmmuting elements respectívely, the general structure of aLie superalgebra is

(iii) Triplets of generators thus s.atisfy a mixed Jacobj identity of the form:

[D;, (D;D• +D.D,]] + cyclic permutations (ijk) =O. (16.84)

Many aspects of the theory of Lie groupõ and algebra.s can be extended to supergroupõ
and super or graded Lie algebras. We shall not discuss them here because this goes
beyond the scope of this work. We shall construct only a sírnple example.
Super3J1111metry and •pace translation. We want to realize our algebra under the fonn
of diiferential operators. From the prevrous considerations we condu de that we need at
least two Grassmann variables (i and Ô and a usual variable which we <:ali t.
Differential operatore, odd elements of the corresponding GrMsmann algebra, have the
form

If we impose the condition (16.82) we find two non-trivial combinations:

(16.85)
16.B 365
Slavrwv-7hylor and BJ?.S Svmmetf'Y

in which a and ã are two ordinary arbitrary consta.nts. It fo!lows'

- -
DD+ DD = (ã+a)at.
a (16.86)

The Lie su bgToup is the group of trí1.111ilations on the variable t. Tbi;; i;; a realiza.tio o of
quantum mecha.nical supersymmetry as will be diseussed in next chapter.
This structure can immediately be generalized to d-<limensional spa.ce:
a
0 <> -_ {}fJ" ~
+ a,piJ~ lJx,.,
a -
D .. ~ {}(!,
a + 9pa/Ja
- -~ ôx,.,
a (16.87}
in which x,.
corresponds to the d rommuting variables, 8<> and li, to a.nticommuting
va.riables and a~/J' ã~~ are constants. Then:

l
DaDp+ DpD, =O,
D.,Dp +D~Da =o, (16.88)
- - (-~ ,. ) a
D,D~ +Do Da = a.,/J + a"/J ax,..

Age.in the Lie subgTOUp is a product of tra.nslation groups. Note that to D, D eorrespond
two other supen;ymmetry generators D', D':
D' = _!!_ _,. 9 ô D~ {}
" a
IJpllll,-a, (16.89)
o ao.. ao{J {J ax.' ao" x,.
wbich anticommute with D, D.

16.8 Application: Stocbastic Field Equations

Let us now consider a. stochastic field eqnation whlch we write formally as:
Eo (IP(x)Í 110 (x), (16.90)
ín whlch a corresponds to the various components of the field rp"(x) and v.,(x) is a
field, hereafter called the noise, for whicb a probability disttibution is provided. This
probabillty distribution is generally given in terms of an action and in most exa.mples
this action will be quadratic in 110 (x) and thus the noíse gaussian.
We shall be conoerned with local stochastic field equations and we shall find out from
the example of tbe Langevin equation that the integration aver the noise leads to la.rge
momentum divergences. lt is therefoN; necessnry to understand their properties frorn
the point of víew of renormalizntion and renormaliza.tion group. lt is tben convenient to
associate witb these equat;ons a local effectiw action becau"" renormali~ation of actions
is much better understood than renormalization of equations. At the same time it is
possible to explicitly perform the integration aver the noise. This naturally leads to the
geometrical structure we have studied up to now in this chapter, and in particular to
actions with BRS symmetries.
Starting from a.n equation as general as equation (Jli.90), we shali derive results whlch
apply to severa! problems which have the same formal structure, for example Ea('l') can
torrespond to a classical field equation (of the form (16.34)) far a spín 'I' and v0 to a
random m~tic field, or to a Langevin equation, as will be discussed in Chapter 17,
and x then basto be 1.mdetstood o f the colloction o f spoce a.nd time Vàrlables. In additíon
as shall see in Chapters 19,21, a similar structure emerges in the quantization of gauge
lheories.
366 Sln.unov-Taylor and BRS Symmetry 16.8

J 6.8.1 The effectíve action

We write the noíse probahility distribution [dp(v)] as:

[dp(v)] -= [dv] e:xp [-a(v)]. (16.91)

The most useful example corresponds to a(ll) quadratic in v,.(x):

a(v) = ~f d:r dy v,.(x)[!1-lj"~' (x, y) 11~ (y), (16.92)

in which the fl,.p{x, y) is a local symmetric operator.

We want to calculate the generating functional Z(J) of ,_fMl)d correlation functions
which is the average over the ooise of:

{16.93)

lt ís clear that the discussion of preceding t;<>ctions can be extended to the problem of
stachastic field equations. Let us rapidly again review the argumento. Using identity
(16.2) we write Z(J) under the form:

(16.94)

with:
6Eo. ('P(x)) (16.95)
Ea/l(:r,y) ,5'1'/l(y) .
lu writing the identity (16.94) we have implicitly assumed that equr.tion (16.90) has
a uuique solution. In field theory this condition may be a oource of serious difficulties
because it requires an investiga.tiou of the uniqueness of the solution beyond perturbation
theory. The Langevin equation we consider in Chapter 17, is a first order differeotial
equation in time and thus this condition ia always sa.tisfied.
We could now integrate over the noise•. lnstead, followíng the method explained in
Section I 6.2, to more clearly exhibit the general algebraic structure, we introduce an
integral representatlon for equatíon (16.90):

6 {Ea('P}- v.,) = f [dA"(:r)J eotp {f dx )."(x) (v.. (x)- E., (rp(.x))J}. (16.96)

The integration over t.he noise v involv"" the generating functlonal w(.l.) of connected
v-field correlation functions:

exp [w(.\)J = f [dp(v)[exp [f dxÀ" (x) v.. (x)] (16.97)

.~
Finally we rewrite det E as an integral aver furmion fields c<' and r!'; ...~
::l
detE f [de"(x)l[dc"'(x)jexp [! d:tdye"(x)E.,p(x,y)Jl(y)). (16.98)~
16.8 367

The functlonal representatíon (16.94} of Z( J} then becomes:

Z(J} =f !d<pj[dê)[dcJ!dÃ]exp [-S(<p,E,c,À) +f d:rJa(:r)op"(:t)] (16.99)

in which S(<p,ê,c,J.) is the effectíve action:

S -w(.\) +f dx J.«(x)Ea(\P) f d:rdyê"(:r)Eall (:r,y)Jl(y}. (16.100)

This expression is, in more expUcit nota.tions adapted to field theory, the expressíon
(HU9). lt therefore has a simp!e BRS symmetry.
The transformations of the fieids can b.. written in terms of a.n anticommuting constant
tas:
(16.101)
The action (16.100) can also be expressed in terms of the fields:
tj,O (x,õ) = 0
<p (:r) + Ôé'(:r), C"' (x,9) = C"(:z) + ÔJ.0 (:z;),
as has been dane In Section 16.3, to render the BRS obviam. In some cases, as we have
shown, lt is also uaeful to combine ali ftelds Íl:l one superfield:
~a (.:r, õ, o) = ~" (:r, õ) +co (x,ê) o. (16.102)
In the case of gaussian noise (16.92), the effective oction S(ifl) then reads:

S(q,) = f dOdií {f dxlf89Ea

aq,a f {}rf/' {)q,fi }
[~(:r,ê,O)J + d:rdyaoíla.B ôiJ . (16.103)

16. 8.2 BRS syrnmetry and WT identities

To the BRS symmetry corresp<mds a set of WT identities which can be derived by
intraducing sources for ali ftelds nnd perlorrning in the funct!onal integral a change of
va.riables of forro (16.101):
Z(J,fi,'l,l) f ldiPJidê] [dc]{dJ.Jexp[-S(<p,c,c,.\}

(16.104)

Since the transformations are linear in the fields, the consequences are símp!e, the gen-
era.ting functional of connected rorrelation functions satisfies:

f 6
dx (Ja(:r} 6ij:(x) 'la(x) Dl., (x)) W (J, ij,'7,l) =O, (16.105)
the IPI generating functional r(op,ê,c,>.) l.s BRS symmetric, i.e. is also invariant
transfonnation (16.101 ). ThiB property is expressed by the equation:

f dx [c"(:r) ~;~:r) + ÀD(x) 6~~x)] =O. (16.106)

the effective actlon has the form of a local action renormalizable by pawer counting, and
a BRS invariant regularization can be found, then equation (16.106) implies that the
needed to render the theoty finite are BRS symmetric. Equation (16.106) is
und<>r renonnalization Bnd the renormalized effective actlon S, is BRS symmetric,
also satisfies:

/
d:r
6'{P
(é"
tiS, +À"óS,) =O.
6C"
(16.107)
368 Slaooov-Tayior and BRS Symmetry

16.8.3 So/u!ions o/ WT identities

Let us assume that we can show, using power counting arguments and fermion number
conservatlon, that S, has the general forro:

S, =-f dxdyE,il(<p)C"é' +E(<p,À). (16.108)

Tt then follows fmm the analysis of Subsection 16.!>.1 tbat E;,l'(<p) and E(<p,À) have
respectlvely the form:

E;,tl
aE;(x)
= éJ<p/l(y)'
E(<p,>.)"' f dxÀ"(x)E~(<p)-w,(À). (16.109)

Thus we ha--e shown that the form (16.100) is preserved by the renormalization. In the
parametrization (16.102) these properties have a simple interpretation, they correspond
to trans!f1tion symmetry on the coordinate Õ.

/6.8.4 An examp!e: lhe rondam fieCd Ising model

lt can be shown that the long distance prOpertiel of an lsing model in a random magnetic
field can be descdbed by a. stocha.stíc field equation of the fonn:

(16.110)

in which the field <p(x) represents the Ising spin Md h(x), the magnetic field, has a
gaussian dístríbution:

(h(x)h(x')) = a6 (x- x'), (h(x)} O.

The equa.tíon (l6.ll0) ís an exarnple of a gradíent equation of the form (16.34) and
therefore leeds to at Jeast two BRE symmetries. Introducing the superfield if> (r, e, li) we
can write the effective a.ction S(.P), after avera.ging over the magnetic field,

(16.112)

withc
A(1>) =f ddx [~ (8,.</>)2 + !m2 1>2 + ft<t>•].
The remarkable property of a.ction ( 16.112) is that it is invaria.nt by a large group
tra.nsformations: traru;Jatíons of 9 a.nd i1 as a.nticipated but also "rotations" which leave
the line element:
(ds) 2 = dx,.dx,. + adedÕ
inva.ria.nt. In addition to transforma.tions internal to the {8, í:í} a.nd x,. spaces one findJ
the two infinitesimal transformations:

B=0 1 +oa 11 x~, x~-26 a~,

1
xj.l.
ij Õ' + aã""'", "" = x~- 2ã,.ll'.
16.8 Sla.vnav-Taylor and BRS Symmetry 369

The vectoro a,. and ã~ are anticommuting elements of the Grassmann algebra. lf we cal!
D ~ and ÕI' the corresponding generatoro, we find the anticommutation relations:

(16.117)

(16.118)

(16.119)

We recognize in the r .h.s. generators af intemal transformatíons.

Using this symmetry it is possible to prove a property of <iimensíonal reduction for
action (16.112) by a vllriant of the method of Section 17.6. One shows that tbe corre-
i&ti<m functions calculated from action (16.112) in d dimensions are the same a,; those
calcu!ated with A(IO(x)) in d- 2 dimensions. This result maps the random lield Jsing
model in d dimensions onto the pure Ising model in d-2 dimensions. It has unfortunately
co11SeQuencés which contrBdict pbyskal intuition. A re85<ln can be found in the starting
point: equation (16.110) in the region of interest, m 2 small, has certainly for some fields
a large number of solutions and therefore the whole method is not a.pplicable without
modífications. A simi!ll.l' probiem, called Gribov's ambiguity, llrises in gauge theories.

8ibliographical Notes
The ST and BR.S symmetríes have been introduced in the çontext of gauge theories by
A .A. Siavnov, Theor. Math. Phys. 10 (1972) 99; J.C. Taylor, NucL Phys. 833 (1971)
436; C. Becchi, A. R.ouet and R. Stcra, Commun. Math. Phys. 42 (1975) 127.
· The present formulntion is ta.ken from
J. Zinn-Justin, Nucl. Phys. B275IFS18] (1986) 135.
discussion of Section 16.1 is lnspired by
B.W. Lee and J. Zinn-Justin, Phys. Rev. D7 (1973) 1049 (Appendbc).
Supersymmetry has been introduced In Field Theory context by
J. Wess and B. Zumino, Nucl. Phys. 870 (1974) 39.
Anticommuting coordinates were used in
A. Salam and J. Strathdee, Phys. útt. 51B (1974) 353; S. Ferrara, 8. Zumíno and J.
Wess, Phys. Lett. 5lB (1974) 239.
~·Nnmtenms textbooks have been devoted te supersymmetry see for exa.mple
J. Wess and J. Bagger, Supersymmetry and Supergravity (Princeton University Press,
Princeton 1983); S. Ferrara, Supersymmet.., and Supergrnmty (Lecture Notes in Physícs
2a8), (Springer-Veriag, Berlin 1984); B.S. DeWitt, Supermanifolds (Cambridge Univer-
sity Press, Cambridge 1984).
articles are reprinted in
SUJll!rs•vrnme;tro, S. Ferrara ed. (Nort!J..Holland and World Sclentüic 1987).
' s,,pe,tsJrmmetry and the dimensional reducúon in the random-li<!ld lsing model are
in
• Parisi and N. Sourlas, Phys. Rev. útt. 43 (1979) 744; J. Cardy, Ph~s. últ. 1258
983) 470; A. Klein il!ld J. Fern!llldo-Perez, Phys. Lett. 1258 (1983) 473.
consequences of dímensiona.l reducti<>n contradict pbysical intultíon
. lmry a.nd S.K. Ma, Phys. iletJ. LeU 35 (1976)
are disproved in
J. lmbrie, Commun. Math.. Phys. 98 {1985) 145; J. Bricmont andA. Kupíainen, Phys.
Reu. útt. 59 (Hl87) 1829.
17 RENORMALIZATION ANO STOCHASTIC FIELD EQUATIONS.
SUPERSYMMETRY

Stocha.stic dynamic equations like the La.ngevin or Fokker-Planck equations, extensions

to Field Theory of the equatíons introduoed in Chapter 4, have been proposed to de-
scribe the dynamics of critíca.l phenomena (see Chapter 35), oras an altemative method
of qua.ntization which cou!d be useful in cases where ordina.ry methods lea.d to difficult
problems like gauge theories. In both cases divergen.::es appear in perturbative calcule..
tions and it is necessa.ry to understa.nd how these equations renormame.
For many problems related to perturbation theory and renormalization, it is more
convenient to work with an effective action (often called in this conte:xt dynamlcal action)
associated with the Langevin equatlon rather than with the equation directly, because
then standard methods o f quantum fieid theory become applicable. The method has been
explained in Section 16.8. We have shown quite genera.lly that this dynamical action has
a BRS symmetry. We shall use thís symmetry and its ronsequences in the form of WT
identities to prove that under some general conditions the structure of the Langevin
equation is stable under renormalizatíon.
A particular c!ass of Langevin equatJons ís of special ínterest: purely dissípative eque..
tíons wjth gaussian noise, in which the dnft force derives from a stetic actíon and is thus
conservative (the corresponding stochastic processes satisfy the detailed balance condi-
tion). Then the dyoamical action has a second fermionic symmetry which, combined
with the fust one, provídes an example of quantum mechanical supersymmetry.
For completeness in Soction 17.9 we discuss the poculiarities of the stochastic qu8Jlti-
zatíon of gauge theories.
The formalism presented here may seem ratber heavy wben applied to simple models,
bllt it enables us to exhibit generru structures with many different appiications.
Finally in Soction 17.10 we briefly review how the concept of supersymmetry, suita.bly
generalized, can be used to construet theories potentially relevant for four-dimensional
partic!e physics. The new technical feature ís tbe combination of spin and supersymmetry.
The appendíx is devoted to the calculation at twt>-loop arder of some dynamical RG
functious.

17.1 Langevin and Fokker-Pianck Equatkms

To keep notations as simple as possible we expose tbe problem on the example of

component scalar field lt'(x, t). The Jield depends in additlon to the time variable t on
euclidean variables "w lts time evolution is governed by " first arder in time stochastic
differential equation, straightforward generallzation of the Langevin equation intro<lu<:ed
in Chapter 4:
<,i>(z, t) ""-~!1L [l"(x,t)j + v{z, t),
wbere L [!f'(x)] is a local functional of 'I'(X) and n-l a constant introduced for OOOLveiiÍerLce_
to provide a time scale. The noise fie.ld v(x, t) ís a stochastic field which we assume to
hav.:: a gaussia:n looal distribution ldp(u)]'

[dp(v)l [ctvlexp [-f 2

ddxdtv (x,t)/2!1],
17.1 Renonnalúation and Stochasüc Field EquationB 371

whícb can also be chara.cterized by its l and 2-poínt functíons:

(v(x, t)) = O, (v(x,t)v(x',t')) Oó(x x')ó(t t'). (17.3)

We discm;s the probiem of renormali•ation on this simple example though ma.ey results
a.pply to more general eque.tions and gau.ssian distributions of noise (see for e:xample
Sections 17.7,17.8).
The Foklcer-Planck equation. Given the noise (17.2), and some initial conditioru; for
the field <P{ x, t), Langevin equation ( 17 .] ) generates a tim ...dependent field disttibution
P (t, tp(x)):
P (t,cp(x)) = (ó (<p(x,t)- <P (x)J}v. (17.4)
The derivation o f Section 4.2 can be immooiately generalized and yieids a Fokker-Pla.nck
equation, tbe field theory form of equation (4.21):

(17.5)

where 1-iFP, the Fokker-Planck hamiltonian, is gíven by

1-iFP (<p, {J:) -~f ddx 6'{J~X) [6'{J~Z} + L(cp(x))] (17.6)

The diesipc.nve Langemn equation. In Sectlon 17.5 we discuss in detail a special case,
the purely dissipative Langevin equation which ccrresponds to the choice:

L [<p(x)[ (17.7)

generali.,;s the exa.mple studioo in Section 4.4. The fundional .A (I") is a static
{til:ne-in•dej>enldent) euclidean action for the scalar field 'f', for CJ~.a.mple:

p = .,-A/2 p,

transforms the hamilwnian (17.6) into a hennitian positive hroniitonian. This

allows to prove the convergence at time +oo of P towards the static distribution
the latter is normalizable). It is at the basis of the idea of stocbaatic
QU11ntízatíon, the Langevin equation being then only a devíce to generate tbe statíc
dist.-ibu•tion A.
Chapter 4 we have a.lrea.dy discussed these equat.íons for a fwíte number of degrees
All lll'guments base<:! on purely algebraic identities can be genere.lized to
theories. We here díscuss new problema which arise because, as we shall show,
Langevin equation in general requires renormalízations. In the c:a.se (17.7) when
action A(<P) is renormali•.able, we expect that the renotmali.ations of the static
together with a time scale reJJormalizatíon, render the Langevin equation finite.
this problem, we have to set up a formalism more directly 8Jllenahle to the
methods of quantum field theory. This ca.n be done by constructing a functional
representation of the time-dependent l"'field correlation functions in terms of an
local action, whicb in this framework it is natmaJ to also call dynamic action.
 Renormalizatian and Stochastic Field Equatíons 17.2
37'1

17.2 Time-Dep<:lndent Correlation Functions

E!Jectiue or dynamicol action. The genere.ting functional Z(J) of dynrunic correlation

functions of the field <p(.r, t} solution of equation (17.1), is gíven hy the noíse avera.ge:

Z(J) = ( exp (f ddxdt J(x,t)<p(:t, t)])",

=f [dv]exp [~f ddxdt (~-!"2 (x,t)- J(x,t)<p(x,t))]. (17.8)

To impose equation (17.1), following the method explained in Section 4.6, we insert tbe
identity:
f [d<p] det M 8 [<,1> + (!1/2)L (op) v]= 1, (17.9)

where M is the differentíal op<lrator:

(17.10)

into expression (17 .8). We then find:

~)]
2f! .
(17.11)

The 6~function ca.n be used to integrate over the noise "'

(17.12) ;~

Fbr a system with a discrete set of degrees of freedom (a d = O dimensional ar a lattice · -~

regularized field thoory), the determinant can be calculated, using the identity: ~

detM (( exp(trln) [1 + (~ ~~~]r1

As a consequence of the ca.usality of the Le.ngevin equation the inverse of the opera.tor ~
(ôj&t.) 6 (t t') is the kernel/1 (t- t') (ll(t) is tbe Heaviside step function). In an expan- -~
sion in po-n; of !1 ali terms thus va.nish when one ta.kes tbe trace, except the first one - ~.-
which yields:
detM <x exp{e{O)~ f dtd"x ;~~~::?t=J.
6
(17.14) l'I'

As we baveseen in Seetion 4.6 the symmetry (t- t') ofthe noise 2--point function imposes J
11 (O) ~ 1 j'l. The final expression then formally ree.ds: -~
J
Z(J) =f ld<,o]exp [~s(<p}+ f d xdtJ(x,t)<p{x,t)],
4

S('P )~2_fdd dt(·<p

2
~~fdtd 4 x 6 L[<,o(z,t)]l
2f! X + !!L(
2 <p )) 4 Ótp (x', f) ,.,_,. •
·.,;·
 17.2 373

When the force in the Langevin equation derives from a static action ( equation ( 17 ·7)) •
the term linear in ,p in the adion S is a total time derivative and contributes only to
boundary ter=.
Note that the dyna1nica! action (17.16) is th<! generalization to field thoory of action
(4.64), and can thus be directly obta.íned by wrlting the solution P(rp, t) of the Fokker-
Pianck équation (17 .5) as a functionaJ integral.
Divergence.s and the problem of renormalizatícm. In dimension d > O, expression
(17.16) is ill-defined when L(<p} is a local functional because the contribution of tbe
determinant is formally proportional to ód (0):

dd:r 6L{'I'(~,t)]l
f 6<p(:r ,t) 1'~z
e< ó"(O).

We bave already encountered a similar situ&tion in the discussion of homogeneous spa.ces

in Chapter.; 14, 15 where also a divergem mea.sure term appea.red. The determinant ha.s
thus .to be regularized. Aga.ín we have two choices:
(I) With dimensional regula.rization terms !ike 6d(O) vanish and therefore the determi-
nant ca.n be oompletely omitted. With this convention expression (17.15) ca.n be used in
practical perturbative calculatians.
(ií) It is, however, interesting to keep this <livergent term in some regularized form, as
we shall do in the discussion of the renormallzation of the dyna.mic theory, ín order to be
able to understand its geometrical signilica.nce. This can be achieved by using a lattice
regularization.
Let us show, however, that even with dimensional regularízation a non-trivial renor-
malization problem arises. We take the example of a dissipative Langevin equation
(equation (17.7)) with the four-dimensional action Á ('I')'

Ã('i') =f d :r [! (8'"'1")
4 2
+ ~m2 '1'2 + (!1/41)'1' 4 ] . (17.17)

. From the dynamical a.ction (17.16) we calcu!ate the propa.gator .ó. of the <p(x,t) lield.
· After Fimrier transformation, "" a function of k and w the variables associated with
and time respectively, the propagatot reeds:

(17.18)

the standard power counting analysis, I>S presented in Chapter 8, it is generally as-
that the propa.gator is, in momentum representation and when ali arguments
large, a homogeneous function. This is not, at fimt sight, the case hera. How·
propagator has a homogeneous a.symptotic behaviour if we scale the frequency
momentum squared, or equivalently the time t as a distanoe squa:red (the brownian
behaviour). Then the canonícal (engíueering) dimensions {'l'l ofthe field 'I' in the
and dynamic thoories coincide:

(17.19)

the interllCtlons coming from (l!Afóv:l have dimensions 4 and 6:

(17.20)
374 Renormalí:ation and Stochastic Field Equatíon.s 17.3

General renormalizatlon theory only teUs us that the renormalized action íS the most.
general local functional of canonical dimension 6, i.e. containing ali vertíces of non-
posítive dimensíons. Such a fuuctional in general depends on more parameters than
the bare action (17.16); in particular it ís not a square of a local functional and can
no longer be derived from a Langevin equation by the algebraic transformations (17.8-
17.16). To prove that the Langevin equation can be renormalized we have thus to find
identitíes satisfied by correlation functions, which imp!y relatíons hetween counterterms
and ensure that the structure of action (17.16) is pr~rved by the rennrmalízatíon.

17.3 ST Symmetry and Equílibrium Correlation Functions

Langevin equations belong to the general dass of stochastic equ&tions discussed in Sectíon
16.8. We shall exploit more systema.tically this remark in next settion. However, let
us alrea.dy note that the ST syrnrnetry discu.ssed in Section 16.1 implíes identities for
corre!ation functions. In tbe first p&rt of this seçtion we consider only the example
(11.7). It folkrws from the identity (17.9) that the deterrninant of M:

' '
M(:r,t;:r,t) n J~A) I'
(:r,t[ ( at+2órpó<p
ô x,t ') , (17.21)

generates an invaríant measure for a set of non-lineax transformations which transla.te

v(x, t) by a function p.{:r, t) independent of <p. For p.(:r, !) iufi.nitesimal the variation of <p

f
is:
Ó<p(x, t) dt' d:r' M(-IJ (:r, t;x', t'; 'P) fJ (x', t'). (17.22)

ln an infinitesimal change of variables of the forrn of a transformation (17 .22) the varia·
tioru; of the action and the souroo term in the functional integral (17 .12) a.re:

{j 1
{ 211
f (
d:rdt rj;+ 2nM)
6<p
1
n
} ~ 1 f [d:rdt .P(x,t) +2 u JtJ(:r,t)
n Ó<p(x,t)
!J [f dxdtJ<p] =f d:rdtdx dt'J(:r,t)M-
1 1
(x',t';x,t)l-'(:r',t').

Expressing as usual the invariance of the functiona.l integral under a. change of V&riables
we obtain the identíty:

1 {) ~
[ OFiM(x,t)
óA ( {jJ
+ 21 6<p(x,t) 6 )] Z(J) = f dx'dt' J(x',t')ZM (x',t';x,t), (17.23)

with the defmition:

ZM (x,t;x',t';J) M (-lJ (:r' , t'·:r ~) Z(J) )

' ' t·'éJ (17.24)

or equivalently:

{ ~8t + ~~
2 [li<p(x, t}]
(_j_)}
óJ
ZM (x,t;x',t';J) = 6(x-x')6(t t')Z(J) (17.25)
17.3 Renonnal:ization and Stochastic Field EC[!Wtion.s 375

Equation (17 .23), which can of course be derive<! directly from the Langevin equa.tion,
impiies that the large time l'unit of the equal-time correlation fuuctíons satisfies the field
equa.tions of the static theory.
Indeed let us take a special source:

(17.26)

and set t = 1' in equation (17.23). Then equation (17.23) inv<>Ives only Zu (x', T;x, r; J).
lf we expa.nd ZM in powers of n, again due to the 8 (t} function only the first term
survíves and ZM becomes:

ZM (x',T;x,r;J) ló(x-x')Z(J), (8(0) !) . (17.27)

We have aga.in used a consequence of the causal cha.rocter of the Langevin equatíon. Also
ít can be checked in perturba.tion theory that Z { J) ha.s a limit and that for a Langevin
equation of t.YJl'l (17. 7) the term:

(8(0)"' u'
vanishes at la:rge time, so that the !imlting functional Z (J) satisfies as a. coru;equence of
cquations {17.23} and (17.27):

M ( 6)
(:r) 5J z
Õ!p (J) J (x) Z (J), (17.28)

whlch lmplies

Z(J) = f[d<p(x)]exp [-A(<p)+ f dxJ(x}~p(x)} (17.29)

From these considerations we conclude that some form of equa.tíons (17.23) and (17.24)
must be used in the proof of the renormalizability of the dynamic theory, i.e. m the proof
that after renormalization the structure of the Langevin equation is preserved and that
l.he equal-time correlation functions have in the large time llmít the correlation functions
nf the renormalized sta.tic theory corresponding to A (<t')·
Remark. We can also derive a dilferent equation directly for the generating functional
Z (1) of equal-time. correlation functions, by using the Fokker-Plauck equation (17.5):

(17.30)

tly definition:

Z {J,t) =f (d<p(x)]P(cp,t}exp [f dd:r J (:z:) Jtl(x)] (17.31)

Tberefore:
376 Renormali•ation and Stochastic Field Equations 17.4

lntegrating then by parts, we obtaln an equation for Z (J,t):

(17.33)

The existence of a.n equilibrium distribution implieo tbat t.he time derivative of Z (J,t)
vanishes for lnrge time. Then the limiting generatíng functional of correlation funciions
satisfies the static equation:

(17.34)

By constructíon, the discussíon o f t he solutíons of this equatinn and of the static solutions
of the Fokker-Pianck equation are directly related.

17.4 Renormaliz.ation and BRS Symmetry

The Langevin equatíon is a special example of the stoch8!ltíc equations we h ave discussed
in Section 16.6. lt is a first order differentíal equation in time and thus has a uníque
solution: The methods oi Chapter 16 can be applied and ali algebrak results remain
valid. We naw considera particular example of Langevin equation (17.1):

<}!(:r, t) + ~n [-~tp + Ltnt. (tp)] = ~.<(x, t),

in which Ltnt. (<p) ís a polynomial in <{', and "(x, t) the gaussion noise with distribution
(17.2). For notat.ional simpHcity we have written the equation for a one-component lield
but ali arguments irnmediately generalize to a lield 'P with severa! romponents. The
dynamical action corresponding to equation (17.35) is then:

where M is the operator (17-10):

M =!._-1M+ lflâL\nt.[tp(x, t)J

Gt 2 ~ Ô<p -

In this form the associated dynamical action (17.36) is BRS syrnmetric. We have seen
in tbe CllSe of the Langevin equation the determinant det M can be !ormally calcula.ted
such a way that the final dynam.lcal action ís in dírect correspondence with the rmu<.,,-,"·Jill
Planck equation. However, to exhibit more clearly the general algebraic structure, ít
useful to kolep it in the form of a fermion integral.
Powe:r countíng and renonnalization. Note tha.t below the canonical dimension of
quantíty Q is denoted by [Q]. As discussed in Section 17.2, from the point of view
pawer oounting, nne has to assign to frequencies w the canonical dimemion of momentum~.
squa.red, Le. 2 and thus to time the dimensíon -2. The dynamical a.ction density {17.36);&
tben bas dímension d + 2 since the integration measure here is dt ddx. The dimansion8_7,
[>.], [\0], [c] and [C] of the fields follow:
[.i] = !(d + 2), [<P] = !(d 2), lc] + [ê] = d.
17.5 Renormalization and Stoclumic Fíeld Equatíons 377

The theory is thus renormallzable by power counting in dlmension d if the dimeru;íon of

Línt.(<P} is:
[L;nd<P)] = !(d+2),
The dimension of M is then IM] "' [Lint. (<P)]-! (d 2) 2. lf the dimension d satislies:

2 (lc] + [c]) 2d > d + 2 ,

i.e. if the dimension d -of spoce is larger than 2, the renorma.liz.,d action S, is at most
quadrntic in the fenn1on fields.
Moreover if the dimensíon d is larger than 2 tbe sum of dimensioru;[c] +!c) +[.l.J satisfies:

[cl + [c] + [)I[ "' ~ d + 1 > d + 2,

and therefore M,, the renormalized coefficient of éc, cannot depend on À.
The results of Section 16.8 {which follow from BRS symmetry) tben apply and the
structure of tbe dynamical action (17.36) is preserved by the renormalization. lt follows
tha.t it ca.n be derived from a renormalized La.ngevin equa.tion.
We ha.ve thus shown tha.t if the canonical dimension of L;nt.(<P) satisfies [L;.. .] :S
!<d + 2) and the dimens.íon d is IBrger than 2, the Langevín equation (17.35) with
the gaussian noíse (17.2) can be renormalized. Moreover sínce the renormalized action
remsíns quadratic in )1, the renormali~ed noíse is still ga.ussian. Note that this result can
be immedia.tely generalized to severa! component fields.
The analysis can also be extended to more complicated ce.ses, for example to models
with diHerent noise 2-point functions (see Chapter 35).
In the particular case (17.7) more precise results can be obtained which....., shall de-
scribe below (for a on~component field equatlon the drift fon:e in equation (17.35) can
of course always be written as the vari.o.tion of an action as in (17.7)).

17.5 The Purely Díssipative Langevin Equatíon

We ha.ve shown in Subsection 16.3 that when, in symbolic notations, the equa.tion has
the form 6Af6<p = 11, the dynamical action S (4!} has two índependent BRS symmetries
wbich, in the superfield notation (16.102), correspond to translations of 9 and Õ. The
Langevin equa.tíon in the spocíal case (17.7):

<P(:t,t)"" -~ btp~:,t) + v{:~:,t), (17.39)

sha.res this property. The force term ís a gradíent but not the time derivative <(J.
~cs.~mewlltatsurprisingly, howewr, it is poosible to find a.slightly modified BR5 tra.nsforrna-
leaves the dynamical action (17 .36) invariant a.nd which combines non-trivially
the first one to yield the simplest example of supel'll}"IIlmetry.

. the superfield notation (16.102), the action (17.36) corresponding to the Langevín
(17.39) with the noise (17.2) can be rewritten:

(17.40)
378 Renonnalization and Stochastic Field Equatiom 17.5

with the delinítions:

-
0
â a
D= - - 9 - .
a (17.41}
"" âO' ao ât
2
For conveníence we have rescaled O and é by a factor (2/fl.f .
D and D satisfy the anticommutation relations:
- - a (17.42)
DD+DD=-i§i·

Let us tben introduce two other GrBSSmano-type differential opetators D', Õ':

D' (17-43)

Both anticommute with D and D and satisfy

D'ÍY +Õ'D' = ~.
8t
(17.44)

The two pairs D, D and D', D' provide tbe símplest exarnple of generators of supereym-
metry. We already know that D' (a translaticm of ii} is tbe generator of a BRS symmetry
of the action (17.40). Let us verify that Õ' is the generator of en addítiono.l eyrnmetry.
If we perform a variation of q, of the form:

which in component form reads:

5rp = cc, bc O,
{ 6c=-(À ,P)E, 8>. b:'
we observe that the variation of the action densíty is a total derivative: This ís obvious
for A because ít does not explicitly depend on t and {/. For tbe remaining term the
additional property that D' anticommutes with D and D has to be used

The action is thus supersymmetríc. The operatorB D and D can be considered as covaríant
deriYII.tives from the point of view of super!lymmetry.
This supersymmetry is directly coJlllected with the property that the corresponding
Fokker-Planck hamiltonian (17 .5) is then equivalent to a positive hamiltonian of the form
(4.40).
R•marb.
(i) It is possible to emphasize the symmetric role played by IJ and ii by performing
the substitution t - t + !/!8. We lind it more conwnient to remain with the original
variables but this is mainly e. matter of taste.
(ii) Consldering the fermions as real dynamic variables, we can write the hamiltonian
asoocíated with the supersymmetric action in boson-fermion space. We find that
funetíonal integral describes both the L!lllgevin equation and íts tírne-reversed furm.
 17.5 Renonnalúation and St~tic Field Equations 379

17.5.1! Wanf-Takah«shi (WT) iàentities

The BRS eymmetry has a sirnple consequence, couelatioo functions are invariant un-
der a tra.ruslatíon nf the coordinate U. The tr&.nsforlllation (17.45) has !' slightly more
complicated forrn. It leads for the generating functional W(J) to the equation:

f -(-ãJ + -ôJ)
dx dt dO d8 0-;:;:;
v• ae ( 6W } = O.
[,J x,t,8,(}

Connected correlation functions w<n> (x., t,, 9,, e,) and proper vertices rrnl (Pi,w;,U;, 6,)
thus satisfy the WT identitiell:

vw<n) (x,, t,, ií,, e,) =o' (17.47)

with
-V:~
~((] -Ô)
âB +B<at .
k~l k •

After Fourier trnnsformatíon over spsce and time the operator V takes the form

(17.48)

Examp/e: a 2-point junction. Let us explore the irnplications of WT identíties for

a 2-point function. As relations (17.43,17.44) show, supersymmetry implí.., tr8llslation
invariance on time and ii. We can therefore write 8llY 2-point function W! 2l:

The WT íd ..ntity (17.47) then implies:

aA
2B{t) = &j· (17.50)

. Tbe WT identíty does not detennine the function C. Au addítional constraint comes
frorn causality whose role we have already emphasi:zed. For the 2-point function causality
u,
' irnplies that tbe ooefficient of (1, vanishes for 11 < t2 and tbe coefficient of e~el for
·t 2 < t 1. The IMt function ís thus determined, up a possíhle distributíon localízed at
· !1 = !2. We find:
8A
2C{t} = -€(t)&i. (17.51)

<(I) is the sign o( t, and therefore:

(17.52)
380 Renormalization and Stocha.<tic Field Equations 17.5

17.5.9 Renormalization of the dissipati1!e Langevín equation

In the spedal case of tbe supersymmetric dynamical action (17.40) a compa.rison between
the two explicit quadratic ter!IlS in ,P of the action yields the relatíon between dimensions:

[t] JeJ - [íi] o 9 JdtJ + [d8] + (d&J = o. (17.53}

(We recall tbat oince integration and differcntiation over anticommuting variables are
equiwlent operatíons, the dimension of de is -]61].)
Therefore the term proportional to A (I/>) in the action hBS the same canonical dimen·
sion as in the statíc ease: the power counting is thus the sa.me and the dyna.mic tbeory
is always renorma!izable in the same space dimension d as the static theory. Note that
equation (17.53) also implíes
2[4>] = d + [t],
equation which relates the dimensions of field and time.
Let us then write in superfield notatíon the most general form of the renormalized
action S, cnnsístent with the results derived in Sectíon 17.4:

(17.54}

and impose the constraints coming from the supersymmetry of the ba.re action. The
transformation ( 17.45) is linearly represented on the fields ,P and therefore the renormal-
ize<l action remains symmetric. Performing the transfonnation (17.45) and expressing
the inva.riance of the action we obtain two equations. First identifying the coeflicient of
(8<f,j8fJ)(â4>Jõt) we find:
Z'= Z. (17.55)
The BeCOnd equation comes from the variation ofthe last term of expression (17.&4):

Thus L (4>) is a total derivative and can be written:

ó(A,)
L(4>) ~-

An íntegration by parts over 9 of the last term in equation (17.54) finally yields t.M
supersymmetric form of the renormalized action:

After renormalízation the drift force in the Langevin equation is t.hus still of the form of.
the variation of a.n action.
Remnrk. Because the Fokker-Planck equntion ha.s static (tíme-iudependent)
which are not of the form (17.7), it is easy to construct bare Langevin equatlons
generate o.n equilibríum dlstribution chara.ct.erize<l by a local sta.tic action, for whkh
dynamical action is not supersymmetric. Direct proofs tbat the renormalized equilíbrium
distribution still corresponds to a local static action have only be given in special cases.
17.6 Renarmali.zation anel Stoch0$tic Fidd Equotions 3B1

17.6 Sup.,rsymmetry and Equilibrium Correlation Functions

The supersymmetry of action (17.40), leads to a direct algebralc proof that the eq~al­
time <f>-lield correlation functions converge o,t Jarge times towards the static correlat10n
functions corresponding to the action A(;p). 'Ib simplify notations we shall consider
the action (17.40), but the generalization to actions of the type examined in Sections
17.7,17.8 is straightforward. Let us assume that the initial conditions in the Langevin
equatiotl are given at time t' a.nd let us calculate the equal-tíme oorrela.tion functions a.t
time t". The source J associated with the field ifJ then has the special form:

J (x,t,e,o) = J{x)ó (t t") ó (8) li (O) (17.59)

We consíder the s-dependent dynanrical acticm:

(17.60)

For s =
O we cecover the dynarnícal action for time.dependent correlation functions
relevant when ali times are in the interval [t', t"l and for s = 1, because the sourre has
the forro (17.59), the action generates the static c<>rrelation functions. lf we díHerentiate
the connected oorrelation functions calculated with the action S (fi'>,•) with respect tos,
we generate the insertion of the operator:

- ['' d6dtidt [~ j ddx~: (:-e~;)+ A(</Jl] +A[</> (t",ii =e= o)].

. (17.61)
Let us first neglect the breaking of supereymmetry due to the boundary conditíon at
t'. Our díscu•sion will be stríctly correct only in the case in which we are already at
equilibrium, í.e. t' = -oo. In this case the insertíon of the opera.tor âS/Bs in an equal-
time correlation function generates a two time correlation function. In SubsecUon 17.5.2
we have determined the most general two time connected correlation functions satisfying
requirements of supersymrnetry and ca.usality. Using t" > t and IJ" t.í" = O, we can =
rewrite equatíon (17.52):

(O(t)O (t")) ( + - ô)
l f:IOãi A (t - t") = A (t + OB- t"). (17.62)

cll!l therefore replace fi'> (t, iJ, O) by ;p (t + 89) in the operator insertlon. We tben note:

(~ lli)t(t+IÍii) O,

tberefore the in.sertlon of the fuot term in the r.h.s. of equa.tion (17J:il) immediately
We remain with the insertion of the operator R (çl):

,.
R(</>)=-1'" dtd6'dÕA('1'(1+ÕO)j +A[;p(t")]. (17.63)
382 Renormalizatian and Stochasiic Field Eq~~ations 17.7

The integration C>Ver the va.ríables ii and 9 can be performed and generates the time
derívative of the action. The last time integration is then immedía.te. The contribution
coming from the upper bound of the integral ca.ncels the second term in the r.b .•. and
we remain with:
R(</>) =AI'I'(t')]. (17.64)
Since w(t') ís lixed, the ínsertion equals the correlation function itself multiplied by
a factor independent of the correlation function. It thus corresponds to a change in
the free energy or vacuum amplitude. We have therefore shown that the connected
correlatíon functions are independent of s: the equal-tirne conne<:ted correlation functions
are ídentícal to the stati<: correlatkm functions.
Let us now discuss the effect of brea.king of supersJ•mmetry d11e to the bounda.ry
condition at t = t'. Th prove ínvariance of the action under transformation (17 .46)
one ha.s to integrate by parts. Furthermore, since <p (t') is fixed, one cfl.!11lot perform the
=
transforrnation for t t' and must therefore multiply the va.riations of the lields in (17 .46)
by some functíon of time which is 1 everywhere except close to t '. The result is tha.t
a. supersymmetry transformation generates the insertion of an operator function of the
fields taken at t t'. Dueto cluster properties, colillected correlation functions involving
the ínsertion of such opera.tors vanísh in the large time separation limit, lt'' t'l -+ oo.
This impJies that the equal-tíroe correlation functions converge at large time towards
tbe static equilibrium correlation functions. In explkit calculations W<> shall always set
the initial condition.s in the Langevin equation at t 1 -oo. Then, at any time, the
equal-time rr!ield correlatlon functions are time-independent a.nd therefore equal to the
static corre!a.tion functions.

17.7 Stochastk Quantizatlou of Two-Dimensional Chiral Models

lt follows from the analysis of Section 17.4 that two-dimensional scalar field theories,
which are special in the static case because the field is dimensionl=, a.lso have spedal
dynamic properties. In particular power counting allows quartíc terms in the auxiliary
fermion lields.
We have shown in Chapter 15 that two-dimensional models which depend on e linite
number of parameters and are strictly renormalizable are related to coset spaces G f H in
which H is a subgroup of the group G. Of particular ínterest are the models delined on
symmetric spaces discussed in Sectíons 15.4-15.6. We here describe Langevjn equa.tions
for G x GjG chiral mode!s, G being a símp!e oompact group. Note tha.t some expressions
explicitly refer to unitary groups b11t the generalization to other groups is straightforwa.rd.
A Langeoin e~~uation for chiral fields. In chiral models the field g(x) varies in some
representation of a Lie group G. V.re ha.ve shown in Section 15.4 that a sta.tic actíon A
wíth only two derivativas has the form:

A=-~ j ~X trj!(x), (17.65)

where j ~ (x) ís the G-current given by:

j~ (x) =g- 1
(x)8~g(x), (17.66)

(it belongs to the Lie algebra of G) a.nd f3 a coupling constant which plays the role of
the lnverse temperature in stat istical physics. The classical equa.tion of motíon (equatíon
(15.70)) expresses the current conservation 8.J,Ax) =O.
 17,7 Renormaluation and Stochastic Field Eguation.s 383

lt is possible to write a group covariant Langt?Viu equation wbích leads to an equilibrlum

distribution corresponding to the static action [17.65):

fo(x,t) = (0/3/2)8~j~(:z:,t) + y- 1 (x,t}v(:r,t)- vl(x,t)g(x,t), (17.67)

wbere tbe current j 0 (:r, t) is defined by:

(17.68)

(below lhe index O will alw.,ys refer to time). Tbe noise v(:r, t) is a general complex
m"trix with gaussian probability distribution [dp(v)J:

[dp(v)] = [dv]exp [-
2~ j ddxdt tr(vl(x, t)v(x,t))) (17.69}

Note that v( :r, t) belongs to the linear representation ofG and not only to the Lie algebra:
ít hE>.S more degrees of freedom than the field.
The dynamirol action. To construct the dynamical action we follow the same steps as
ín preceding sectkms. Howev<:r, since we have to integrate over g(x, t) with the group
ínvariant mea.sure, the expressions are slightly modified as we expect from tbe """lysis
of Section 16.4. Similar expressions will agaín appear in the quantization of non-abelian
gauge theories with gauge group G.
Let us introduce the covariant derivatives \1 "'\lo ll&SOdal.ed with the currents j.,,j~
(for more details see Sectíon 19.1)

v,. a,.+!J,.,·J Vo= a,+ [i~. ·I· (17.70)

In an infinitesimal spaoe-tíme-dependent group transfúrmation:

g(:r,t) >--+ g(x,t) (I +l(x,t)), (17.71)

lhe currents transform like:

io ,_. io + \lof. (17.72)

Therefore the analogue of the operator M introduced in equation (17 .37) is now delined
acting on a fteld t as:

(lT.73)

nn,trcodudr•g fermion lields ê and c a.s well as a La.grange multíplier À, ali belonging to the
of G, we can write in the normalization of equation (17.40) the dynamical
S resulting from t.he integration over the noise:

c (\lo- !0/38,. V,.) c].

(17.74)
384 Renormalizaticm nnd Siochastic Fidd Equatíons 17.7

Note the appea.rance of a quartlc fermion terrn induced by the dependence of the noise
term on tbe field g(x, t) in equation (17.67). As we expect from the analysis of Section
16.4 the set of BRS transformatioru; whlch leave the action inva.riant is now:

6g lgc, óc = -lc2 , (17.75}

bc=i!>., 6>. =0. {17.76)

The BRS tra.nsformation of the current j" induced by equa.tion (17.75) ca.n then be
wrítten:
(17.77)
The BRS transformatioru; (17. 75-17.77) are exactly those which will aga.in appear in the
qua.ntizatíon of non-abelian ga.uge tbeories in Cha.pter 19. The gauge field associated
witb the group G ha.s the transformation law (17.77).
We now show that the Langevin equation ( 17 .67) is the natural generaliza.tion of equ a.
tion (17.39). We introduce the superlield Ç, element ofthe group G,

Ç = g (1 + Õc) (1 + 1'.8) {1 + ê9>.). (17.78)

The set of equa.tions (17.75,17.76) is then equivalent to

(17.79)

From Ç we derive a set of associated currents JP, Jn, Ja, J9:

Jo = ç-'atç,
{17.80)
J§ = ç-' :uç·

The action (17.74) can then be rewritten in a form analogous to (17.40):

(17.81)

In this form it Í5 clear that S is left inva.riant by the transformation (17 .45) and is thu•
supereymmetric. This supersymmetry implíes that equal-time g(:r, t)-field correlation
functions converge at time +oo towa.rds the correlation functions of the static actíon
(17.65), iustifying the choice of the Langevín equation (17.67). We ahall show in next
sect.ion tha.t equation {17 .67) is a specia.l example of a family of Langevin equatlons
corresponding to two.dimensional modeffi defined on Riemannian manifolds which contain
a.1l models defmed on homogeneous spac.es.
The form of tbe renorrna.lized action is tben dictated by the structure of symrnetrk
space G X G/G a.nd supersymmetry (equation (17.45)): two renorrnalization consta.nts
are needed, the usual coupling coru;ta.nt renorrnalization of the sta.tic theory and Bgain
the time scàle renorrnalízation. In addition, in general the pa.ra.metrization of the group
elements is also renorrna!ized.
17.8 Renonnaiuation and S!ochastic Fleld Equ4tions 385

17.8 The Langevin Equatlon on Riemannian Manifolds

We have already seen in Section 4. 7 that it Js imposs!ble to wTite a La.ngevin equation

which only depends on the geometry on the manifold. We therefore construct a Langevin
equation in the follawing way: we ronsider smooth manifolds embedded in an euclidean
space and defined by a set of equations constraining the eudidean coordinates X<>'

E' (x.,) =o. (17.82)

To simplify calculations, we assume tbat we have solved locally these equations and
e:xpressed some components t7 8 of x., as functions of a set of independent components
r,p"':
(17.83)
To construct a Langevin equatíon on tbe manifold, we start from a Langevin equation
of type (17.39) in the embedding euclidean space:

Xa(x,t) (17.84)

in which v.,(x, t) is a ge.ussian noise:

(v.,(x, t)v~ (x', !'))."' !M (t ~ t') 6 (x ~ x') é.,p. (17 .85)

We see immediately that, as in the case of chiral fields, we are naturally Ied to introduce
a noise with more degrees of freedom than the ne)d.
We now project ::i:o(x,t) onto the spru:e tangent to the manifold at point x., (x,t). For
this purpose let us introduce some notations. On the manifold the va.riations 6x., of the
field X<> are constraíned by:
o. (17.86)

We introduce an orthogonal ba5is e~ for the variatioru; of xo:

aE•
- e~=O, (17.87)
8 Xo
with the orthogona.lity condition.s:

(11.88)

terms of the coordinates 'P', equation ( 17.87) ea.n be rewritten:

(17.89)

therefore equation (17.88) hecomes, after elimination ofthe components e~,

(17.90)

we ha.ve introduced the metric tensor 9ii on the manífold (see Appendix A15.3.2):

(17.91)
 Rencnnali:l:ation and Stor:hastic Fie!d Equa.tiona 17.8
386
Equatíon (17.90) shows that the matrix e! is the inverse of the vielbein (Section 22.6).
We can now construct from equatíon (17.84) a Langevin equation on the manifold

xD(.x,t)
D/J(f!óA)
-2 6xtJ +
"•"•. "fJ •
(17.92)

Let us write tbill equation in terms of tbe independent components rp'

(17.93)

We tben observe that:

6A (<p,U)
Stpi , , (I")~=
+ u;a óA " (ip,U
u;A . (I"),
) (17.94)

in which in the r.h.s. B;A means total deri:vative of A with respect to rpÍ •
.AJ; a coiiE<lquen<:e of equation (17 .90) the inverse metric tensor g''
is gíven by:
(17.95)

Equation (17 .93) can therefore be rewritten:

,pi = -!flg'1ôfA+g'itjvD,
wíth the definítion:

Since the noise "~ is multiplied by a fundion of rp, we have to specify the meaning of the
product. We choose the covaria.nt definition as in Section 4. 7.
The qua.ntíties tf satisfy:
tftj = 9ij'

and
a,t:;t: 9kzr:; ,
in wbich r!; is the usual Chrilltoffel symbol (equation (22.41)). This equation can be
rewritten in covariant form:
vitJtr =o,
in which V; is the covariant derivati-re on the manifo1d.
Dynamicai action. lt is convenient to rewrite equation (17 .96) as:
u.;riJ + ~na,A-trv~ o.
Introducing a La.grange multiplier >.' and fermion fields c' 8.Il.d c!, we cá.n write the
corresponding dynamk:al action S before integration over tbe noise as the sum of tvio
contributions s~ and SI: . . .

So =f dxdt ( 2~v.,v,. ijtivD>.' + ~c'â;t~civ,.), (1

s1 f ru:dt~ [>.; (!l;;rpi +~&,A) t 9,1ci- ta.u<;riJd'- ~c<a,a;Aci), (17

17.8 Renonnalõzation and Stochastic field Equatio,.. 387

and:
S=So+S1 • (17.104)
Integration over t.he noise leads after a short calculation to:

s= S1 + j dxdtÕ (-Á'g,;A; +2A' 9.,fi1;;ic" +ê'eii!'daiai 9k1). (17.105)

Introdudng now the stiperfield qh

q,' 'I'' +Õc' +i!II+Õ6A', (17.100)
we éan rewrite the action:

S(<fJ)=
f _[j
dtdede dx
2
0869
aq,• .·
ij(l/>l
(aq,;
86
(17.107)

This supersyriJrnetric form then ímpUes thst st equilibrium the equal-time field configu-
ratioru are weighted by the rnessure [dp (10)):' ·· ·
(17.108)
This wa.s of course tbe field distribution we bad in mind when we wrote equatíon (17.96).
Note that in contrast to Langevin equation (17.95), the dynamical action (17.107) and
therefore the oorrelation functions at equilibritmi depeiid only on
the geometry of the
mailifold. In the case of homogeneous space:S the original action A (X) ís símply: ·

A(x) = 2
1 f (B,.xn)2 dx, (17.109}

which, as a functional of ip, becomes on the ma.nifold:

A (\O,a (<P)) li
2 ' '
O;; ('P.) a,.10'8"cpidx. (17.110}

lt is then eaey to verify that expression (17.107) co~tÍl.ins expression (17.81) as a 5pecíal
<;a.se. .
Note finally that t.he functíonal integral representation for the generatíJJg (unctíonal
(J) can also be rewrltten introducing the superfield X in euclidean coordínates:
(17.1ll)

Z(J)= j<d<P)exp[-sc<P)+ jdxdtdlldÍ.ÍJ;!f>']

= J!dXJ6[E•(x)Jelip(~r.(~)+jdxdtdedõJ,,p'), (17.112)

fw: 'E (X) the expression:

'E (X] =f de dÕdt [f d.: Õa:e,. ( 8:0" /'~") +A (X)] . (17.113)

expressíons emphasize the símilarity between the properties of the rtatic and the
theory.
case of homogeneous spaces the renormalization properties of the Langevin
then follow from the general analysis of Chapter 15 and supersymmetry.
388 Renormali.zatíon a~td St.xhastic Fie!d Equativm 17.9

17.9 SLochastic Quantizatiom the Example of Gauge Theories

We have mentioned that, rus suggested by Parisi, stochastic dynarnic equations can be
used to quantme fie!d theories when non·trivia.l quantizatíon problems arise. The time
vaTíable íu the Langevin equation is then a jictitions additíonal variable •ince only the
equilíbrium dist.ribution ís physical (it can be interpreted as the romputer time of nu-
merical simulations where stochastic methods are uoed to generate tield confignratíons).
As an anticípation to Chapters 18--19, where gauge theories will be extensívely discussed,
let us brieliy review the application of this idea to gauge thoories. Tbe ba.ckground of
Chapter 18 wi!l thus b" assumed in tbis section. The problem of gauge theoríes is that
one field degree of freedom, which corresponds to gauge transformations, is redundant
a.nd the conventiona.l quantization method has to be a.dapted to this peculiar situation.
The stochastic <JUantization idea is to write a gauge invariant Langevin equatian. Let
us consider the abelian case wíthout matter field. We use as a. àrift force the functíonal
derívative of the classical gauge invariant actíon. The gauge tield A~ then satislies

(17.114)

where
F~. a~A. -a. A~,
and N"(x, t) is a gaussian noise with measure

(17.115)

]f we make a time-independent gauge transformation

the equation is clearly invariant. However it is not invariant under time-dependent traus-
formations and thus providas a dynamlcs to ali components of the gauge field. Let us
solve the equation in Fourier space

Ap(t, k) (~~'" k~k.jk2 ) e- 0 "''1 2 [A.(O, k) + l e0 k'l' f2 N.(t', k)dt'J

+ (k"k,jk 2 ) [A.(O,k) +f N.(t',k)dt'}.

We immediately see that the component of A., a.long lc~ follows a brownian motion and
thus does not equilibrate. Let ue indeed calculate the equal time 2-point function, ofter
avera.ging over the noise. Calling AL and AT the component of A,. respectively along
and perpendicular to k,. we find

(A T(t, k) A T(t,- k )) = k2
1 +0 (e -m'•) ,
(AL(t, k)AL(t, -k)) = AL(O, k)AL(O, -k) +fit,

We conclude that, due to gauge imwiance, the Langevin equation (17.114) does noi
genera.te an equilibrium distribution, however the ga.uge in"3ria.nt functions hav., a large
 389
17.10 Renormali.zation and St.ocJu.stjc Field EquatítmS
time limit. Since only t.he latter functions have a physical mea.nin.g the quantitation
problem has been solved.
lt is dear that the sarne condusion will be rea.ched lf the gauge field is interacting with
matter, 1111d in the non-abelian case.
The next problems are the formal rele.tions with the standa.rd quantization procedure
by gauge lixing as descríbed in Chapters 18,19 and the prublem of power counting and
renormalization. Let us write tbe gauge field propagator obtaíned from the eJfective
dyllAIIlical action. In Fourier space one finds

Ll.~.(w,k)

The l/w2 singularity relle<:ts tbe absence of equilibrium distribution whícb now takes the
form of IR divergenc:es. lt is ne<:essary to work in a finite time íntef'\•al.
We see also that the loDgitudinal propagata< does not deçrease at large momentum for
lixed w. An analogcms problem will appear in the quantization wítb non-covariant gauges:
the theory ís DOt renormalizable by power counting. One solution to thls problem is to
zu:ld to the Langevin equation a non-conservative drift force of the form D.,V(A) "•here
o,
Dso is the oovariant derivative, and V (A) a linear function of A for example A.,. It is
easy to verify that sucb a term does not contribute to the evolutíon eqnations (equation
(4.34)) for equal-time gauge invaria.nt correlation function, and thus these functions are
not modified. However with this Lerm the Langevín equa.tion is no longer gauge invariant,
an equilibrium distribution is generated a.nd with a suitable covariant choice of D., V(AJ
the theory is renormalizable by power counting. It can al.so be shown that it is equivalent
to the theories obtained by canonical quantí~a.tion procedure, but with a non-local gauge
fixing term.
The drawback is that a. k>t of the ae5thetíc appeal of the original formu!ation bM
been lost, and the proof of renormalizabilíty becomes as complicated as in canonícal
<1ua.ntizatíon. It of course remains the hope that this method willlead to the quantízation
uf systems where no alternativa method has yet been found.

17.10 Supersymmetry in Four Dimensiono;

In this seçtion we present a very bríef díscussion of supersymmetry in four dimensions,

since supersymmetry has been proposed as a principie to oolve the so-called hiemrr:hy
problem in pa.rtíde physics by relating the masses of sca!Br particles (like Higgs fields) to
those of fermions which can be prote<:ted agaínst ~large" mass renorma!ization by chiral
')'mmetry (see Subsectíon 20.1.4). Exact supersymmetry would also solve the problem of
Vl<llíshíng or at least unnatural small value of the cosmological constant (see Subsection
22,7.2). However experímentally the, ""yet undíscovered, supersymmetric partners of
known particles have to be much heavier. Supersymmetry is necessarily broken a.nd
the problem of the oosmological constant only slight!y improved. Finally gauging of
•upersymmetry naturally leads to unification with gravity, OO<:ause the commutators of
supersymmetry currents involve the energy-momentum tensor.
The new feature of supersymmetry in higher dimensions, compared to the quantum
mechanícal case, is the combínation of supersymmetry with spin since fermions have
'.lll>ll-int;eg,er spirus, and thís lezu:ls to modificatións we want to discuss here.
introduce supersymmetry we lmmediately use the superlield notation. The algebra
genera.tors has been brielly described in Se<:tion 16.7. If we combine scalar boson and
390 Renomlaliz<ltíon and Stoehastic Field Equations 17.10

spin 1/2 fermion fields into a superfield we get sornethlng like

<t>(x, Õ) A(x) + êw(x) + ·. · ,

and thus Grassmann ooordinates mU5t be a.ssociated with the splnor índices. We imme-
diately find serious restrictions coming from power counbllg if we want to construct a
renormalizable lield theory. In four dimensions these restrict.ions somewhat destrcy the
beautiful simplicity of the 1-D case. In any space dimensíon, the dimension [§} of the
coordinates ê is [9] {A] [,P] = -l/2. If we give to 9 Dirac spinor indices tbe cor-
responding Grassmann algebra has 8 different generators B,, 8Q. Therefore the volume
elements Da d9"d8, has dimension 4 and tbus only a constant action is allowed. To cir-
curnvent tbis difficulty one uses the special properties of the spinor group correspoodlng
to 80(4) which is SU(2) x SU(2). First the spinorial representation can be reduced into
left-handed and right-handed components (Weyl spinors, see Appendix A5. 7). Moreover
the representations of SU(2) are self-conjugated. lt is thus suffident to associate Grru&
mann coordinates with the two Weyl spinors. We only need four Grassrnann coordinat,.;
lia, é,. In thís form the Crassm&nn volume element has only dimension 2 and a.n action
bilinear In the ocalar superfield can be written: such an sction term is called a 0-term.
However this leads only to a fre<l field action. As we shall explain, to construct interac-
·tion terms, one has to consider cubic polynomíals of only one kínd of chiral superfteld
(left or right-handed). Such a polynomlal will only depend on two variables 8 or 8, and
thus can be integrated with tbe V<:>lume elements d9,d8• (dÕ1d1iz respeetively) whlch have
dimension 1. The corresponding cantributíon is called a F -term.
To write the generators of supersymmetry (see Section 16.7) it is convenient to in-
troduce a notation for the 2-.spinor traosformlng under the conjugated representation of
SU(2)
(17.ll6)
wbere •atJ is the antisymmetric tensor with •a 1. We also define the four matrices u,.,
where for I" "' 1, 2, 3 u,. a:re just the usual Pau li matriCffi and 0'4 = -il. Two relatíous
are useful in what fo!lows:

Then under a SU(2) x SU (2) transforrnation

11;, = U/!8~, Ó0 ' : v;a{JfJ, u,x~ = Vu"x,,Ul, U, V E SU(2).

SU(2) inva:riant quantities are then for example 8"8,., ii"éa ar ii"u"/!9p.
The supersymmetry generators

(J7.117a)

D" (17.117b)

belong to tbe representations (0,1/2} and (1/2,0} of SU(2) x SU(2) respectively. They
bave the anticommutation telations

{D,D} ~ {JS,n) =o. (17.118)

 17.10 Renormali:ation and St.ochastíc Field Equ4tions 391

Note therefore tha.t the supersymmetry current will have spin 3/2. This implles in par-
ticular that the spontaneous breaking of supersymmetry generates spln l/2 Goldstone
ferrnions. This ls an unwanted feature from the point of view of Particle Physics. Al-
though massless spin 1/2 fermions do exist (neutrinoo) they do not obey the consequences
of the conespondiog WT identities. lt is actually a problem of the supersymmetry as-
sumption that none of the partners of knOWJ.l pa.rtides in supermultiplets haVI! been
found yet. Note linally that since the commutator of supersymmetries is the generator
of translations any attempt to gauge supersymmetry {which would provide a solution
to the unwanted Coldstone fermion) will result in a theory which conta.ins gravitation.
Therefore the problem of spnntaneous supersymmetry breaking does not seem to have a
:;olution outside of supergravity, and thus within the framework of renormalizable 6eld
theories.
As we have already indicated in Section 16.7, to D,fl correspond two operators LI., li
which also satisfy the anticommutation relations or supersymmetry generators, antioom-
mute with them and thus pla.y the role of covariant derivatives:

(17.119)

(17.120)

I 7.10.1 Scalar chiroi superfieJ~

Beca.use li antirommutes with D, D the fields which sa.tisfy

ii,.q, (>:,8, 8) =o' {17.121)

·· form a space of representation for the supersymmetry generators. The general solutiou
. of equa.tion (17.121) can be written

,P(x,e,õ) =ql(y,9),
tbe new space coordinate !I is

11~ x. + !ir a ..~op. (17.122)

that y also satlsfies the equation

D"y,.(x,ll,U) =0. (17.123)

on functions of y,O, the geneta.tors D, fl then take the forro:

D" = _!!_,
ÔBa

that the variable y plays a role similar to time in the l·D case. A scalar right-
superfield can be expanded on the ll., basis. Since cr takes only two values the
general expression has the fotm:

(17.124)
 Renormali.zation and Stochastic Field Equations 17.10
392
A aJJd E are tw<J complex scalar fields; A as well as 4> itself have dimension one a.nd thus
E has diroension twn. A renonnalizahle action CaJI be at most quadratic in E and, as in
the 1-D case, E does not propagate a.nd CaJJ be elimín&ted from the action by using the
correspondíng equation of motion.
The action of the supersymmetry generator ij"D., -t).,D" in component form then is:

óA l"f,.!J/", (17.125a)
ó,P8 ftrr ~~â~A -'I~ E, (17.125b)
6E =-rtrr.~a• .p~ (17.125c)

For !ater purpose it is useful to a.lso expand </J in terrns of the space coordinate x:

<P A(x) +1/-•"(x)&,. + ~â,.A(x)ê"aJii~ + fE(x)ll"8.,

- ~0°rrl'~â"'\b~(:r)~8 1 + ft,â2 A(x)O"fJ0 {j$íjfJ. (17.126)

ln the same way left-handed chiralsuperfields can be defined which satisfy A"4> O a.nd
thus depend on é and a space variable fi
(17.127)

The corresponding chlral Jield Cllll be written:

(17.128)

The action of the supersymmetry generators then is:

6A• ~r,"' i/;., (17.129a)

(17.129b)
fJi/;13 ct,.pn"â"A• +IJiJE•,
6E' = -o,.·ifPrr.~T1a. (17.129c)

The expansíon o[ ~ in terms of the space va.riable x reads:

"r[,= A"(x) + .j;"(x)9., !&~A'(x)Õ"ct~~91l + ~E'(:r)B"Õ.,

+ ~IJ,.w"(r)rr..~etlil'i>i., + f.B2 A(x)O"O,õflõ11 .

The free action. We c:an now construct a free actiOn

Co= r/>(ii,9).p(y,O),
f d4 r fl dÕ.,dO, Ln
o
f d4 x (â,.A'â~A- if,"rr~~&.w13 - E• E).

The supersymmetry o[ the )agrangian LJJ is here obvious since it does not depend
plicltly on B, Õ and Xw lf we have severa.\ superfields we simply add the ,..,,.,.,.nmuii•••
contributíons.
17.10 Renarma.lizatian and Stoclw.stic Field Equations 393

lntemction temu. We here write the F-term contrlbutions fo:r severa! superlields q,,
(we need at leBSt two to wnstruct a Dirac fermion). The most general reuormalizable
lagrangian density has the form:

(17.132)

lntegrating over O,, O, Wé obtain

Adding the kinetic term (17.131), (17.133) and its conjugated left-handed contribution

l.F(4>) =c; ;f,;+ !Mij;j!,.j,; + f,g;jk~oif>;;J,k,

f ide2dÕ1lF =C: E;+ M;j (A;Ej- !~f,ji,;) + !Y:ik (AjAjEZ- ,lif'.j;",Ak) .(17.134)
we obtaín a physica.lly gensible supersymmetríc lagrangian. A useful exercise isto verify,
nsing the expliclt expressions (17.125,17.129), that the resulting lagrangian C va.ties in a
supersymmetry transformation by a total derivative and the action thus ís invariant.
We can now integrate over E,E• since the integral is gaussian. This is equivalent to
use the corresponding field equatíons:

(l7.135a)

(17.1351>)

'l'he lagrangian can then be cBSt into the form

l\ lorm reminiscent of expression {17.16).

to now we h ave explained how to construct a supersyrnmetric action containlng only

and spin I/2 fennion fields. For a realistic theory of particles, -.ector fields are
equired. Tberefore from now on we again assume some of the concepts wbich will
int:rodluc•ed only in Chapters 18,19.
consider a general real superfield, wbich we cal! a vector superfield. We can
it in the form:

v= B + x"O., + xn@" + !C0"9., + !c• i? e"+ i2- 112 vpé"u ..~o11

+ !8" (>." ~a.~8pXi!) ({TO.,+! (>.tl + !õpX 0 (.r)a,.~) BpfNJ.,
t
+ (K + Vi B(.r)) e"e..8PfJfl, (r 7.136}
394 Renormalizatíon and Stochastic Fíeld Equations 17.10

in which B, K !ll'e real fields. The reason for such a parametrization will beco me clear
be!ow. Let us fust note that in this f<>rrn the vector superfield contains four real scalar
fields, four Weyl spínors and a vector field. However, supersymmetry naturally Jeads to
a generalized form of gauge oymmetry. We oonsider two scalar superfields ;p, 4> of the
form (17.124,17.127). A generalized abelian supersymmetric gauge transformation then
is delined by the tra.nslation:
(17.137)
In component form thi5 leads to:

B.-B+A+A·, X>->X+.P,
C,_ C+ E, V,. - V,.- 2~1/ 2 i8,. (A- A•),
),,_.,)., K.-K.

With <>11r parametrization >. and K are gauge invaríant, while v., transforms like a usual
abelian gauge field.
In a theory which h as this kind of gauge inva:riance, tbe vector superfield can be sím-
plified, the fields B, C,)( being eliminated by a gauge transforrnatíon. This characterizes
the non-supersymmetric Wes•-Zumino (WZ) gauge. The vector field then takes the form

(17.138)

The physical degtees of freedom, "" we shall see, reduce to a massless vector field v,. and
a massless spin 1/2 fermion À, since the field K does not propagat.e.
lt is important to note that, in this gauge, powers of the vector field have the property:

V" Oforn~3. (17.139)

Supersymmetric curuatu..., tensor. The vector superfield ís not chiral and therefore, a,,
in the 1-D case, the operators ii, 6 can be used to construct other superlields. Note that.
quo.ntities of the form óll i5.~ W and ~J..B l>p W, wbere W is an arbitrary superfield, or<:
right and left-handed chirallie!ds respectívely, beca use the product of three operators ó
or li. Vl>nishes:
i5. 0 .li.B[:,!JW ÓaÓ.B!J.~W =O 'V W.
In particulax tbe quantities

(17.140)

are chiral and ga.uge invariant (in the sense (17.137)) since for example

l~/'6[Jb."(</> + ;p) = ii~iiflt,,c•q, = -u.,3fJ.,i5.IJ,p =O,

(we ha...., uzed the antioommutation re)ation (17.120) ll.lld t.he chirality conditlons) and
thus generalize tbe gauge field curvature. Tt> calculat.e them it is convenient to exp...,_.
V in t.erms of the variables y and if:

V= i2- 1 1 2 V,.(y)Ô"o"~e~ + ~Õ"5.a8~'~8[J + i>.il(y)8pÕ"iia

+ ~ [K(y} iâ.,V,.(y)j 8"8a91lõ.8,
;:r JI 2 V,.(ii)Õ"o .. ~e.B + ~B"J. .. allefJ + i.>l'(ii)llfJÕ"ii.,
+ ~ [K(ii) +iâ,.V.(i])JB"BuiJI'õo.
 17.10 Renormali.eation anti Stoduutic FWd Equations 395

We then find:

F" -Ãn(y) i2- 312 9~ ("ta,.) ~F,..(y) -K(yW" + ~8,).t'(y)uJIT'9.,,

Ãa(!l)- 12- 12 (a,.at) ~F".(Y)Õp + K(y)Õ, + ~cr,.~8,.Ãp(y)Õ"Õ,.
3
F"

with the usual notation F"" o,.

A. - B.A,. We observe that indeed F a.nd F depend
only on the gauge inva.riaJlt combinatíons F,.., .l., K. A check of these expressions is
provided by the simple relat.ion (derived from the definitions (17.140) after a few com·
mutations)

The field F" is chita) and has dirnension 3/2. Therefore C. F -iFn F,. is candidate
to contribute to the kinetic term. We then find

(17.141)

where we have used the identity

lf we add the conjugated contribution coming from - ~pn Fa we obtain the supersym-
metric free gauge action S which can be Written

(17.142)

This action could also have been obtalned by .noting tbat the v<lCtor superfield hUB di-
tnension zero. A term of the forro V IJ.l:. 88 V has dimen6ion two and is thus candidate
to be a D term.
Note finally that it is also possible to give a mass to vector superfield by adding the D
contribution of v•.
Gauge invariant interoctions. A charged scalar superfield transforms under global
IJ(l) as t/J >--> e'•" t/J. However if we want to introduce space-dependent U(1) tra.nsforma-
t.ions consistent with supersymmetry we cannot simply replace A by A(x). We have to
illtroduce the const.ant A by a scalar chíral superfield:

rp(y, 8)....., e'"'(y,B) q,(y,O) with ÍicxA = O, (17.143a)

~(i), Õ) ,_. e-••Aiil.9J ij;{y, Õ) with ·8"Ã"'"O. (l7.143b)

Wc see irnmediately that if the charges and couplings are such that the interwotjon term
(17.132) is ínvarla.nt under global U(l) transformations, it is also invariallt under the
',, trensforrnations (17.143). However the free term (17.131) is not i.nvariant since

;j,q, ,_, ixf>é<IA-h).

:n,. render the kinetic tenn gauge invariant we replace it by

{.D"' íf,e•V t/J,
396 Renormalization and Stochastic Field Equatíons 17.10

where the vect.or superfield V must tran.sforrn like

V~--+ V -i(A- Ã},

transformation indeed consistent with (17.137).

At first sight it would seem that such a theory cannot be renormalizable, but at least
in the WZ gauge the expansion of the exponential reduces to three first terms (property
(1'1'.139)) and contains no term of dimension larger than four.
We have thus presented ali the ingredientE necessary to construct n supersymrnetric
version of QED.
Non-abelian gauge the.oríes. The extension to non-abelian gauge theories is not difficult.
We denote by ! 0 the matrices generat.ors ofthe gauge group G. The gauge transformation
ís now parametrized by a set of scalar superfields A, and can be written

A gauge invariant legrangian is

where the gauge field transforms like

The generalized form of the curva.ture F" is

{17.144)

which is chiral and indeed transformo like

(17.145)

Finally the contribution to the action takes the form tr F" F.,..

Bibliographical Notes
In the ma.in part of this chapter we have follawed
J. Zinn-Justin, Nuc/. Phys. B275 [FS18] (1986) 135.
The relation between supersymrnetry and dissipative Langevin or Fokker-Planck equa·
tíons have been shown in
E. Witten, Nuc!. Phy3. Bl88 (1981) 513; M.V. Feígel'man and A.M. Tsvelik, 8011.
Phys.-JETP 56 (1982) 823; H. Na.kazat.o, M. Namiki, L Ohba and K. Okano,
Thear. Phys. 70 (198:>) 298; E. Egorian and S. Ka!itzin, Phys. Lett. 129B (1983)
Our construction of the effective action provides an exa.mple of the so-called Nicolai
H. Nicolai, Phys. Lett. 89B (1980) 341; Nud. Phys. B176 (198ú) 419.
The characterization of the equilibrium distribution by supersymmetry arguments is
adapted from .
R Kirschner, Phys. Lett. 139B (1984) 180; E. Gozzi, Phys. Lett. 143B (1984) 183.
The non-linear a-model hllS been considered in
R. Batm:h, H.K. Jansoen and Y. Yamazaki, Z. Phys. B37 (1980) 163.
17.10 Renormalizaüan and Stochastie Fíeld Efll'l'I.Wns 397

Applica.tion of stochastic quantízation to gauge theories is discussed in

G. Pa.risi a.nd Y.S. Wu, Acad. Sei. Sin. 24 (1981) 483; D. Zwanziger, Nu.cL Phys. Bl92
(1981) 259; B209 (1982) 336; J. Zinn-Justin and D. Zwan•iger, Nucl. Phys. B295IFS21]
(1987) 297;
The corresponding RG fi-funetion is g\ven by:
K. Okano, NacL Phys. B289 (1987) 109.
See also the review
P.H. Damgi!Atd &nd H. Hüffel, Phys. Rep. 152 (1987) 227.
In Section 17.10 we have fo!lowed
J. Wess and J. Bagger, Supersymmetry and Supergmt>ity (Princeton Univen;ity Press,
Princeton ]983).
398 Renormalization and Stocho.stic Field Equo.tions Al7.1

APPENDIX l7
RENORMALIZATION CONSTANTS AND RG FUNCTIONS AT
TWO.LOOPS: SUPERSYMMETRIC PERTURBATION THEORY

From the form of the renormaUzed dynamical actíon RG equatioru; can be derived. In the
case of the purely díssípative Langevin equation (17.1,17.7) only one new RG function
appeBJS, BSsocíated wíth the time scale renormalization. For illustration purpose we here
crmsider the examples ofthe rfJ 4 field thoory and the O(N) symmetric non-linear cr-model
and calculate the correspondíng RG functions at two--loop order.
Pertt.rbative calculatíons. In perturbative calculations of dynamic quantitics the for·
malism of Section 17.2 can be used, dimensional regularizatíon eliminatíng the determi-
nant. In the particular case of the di:>Sipative Langevin equatíon, it is also possible to use
the method of super-díagra.Jru3, trea.ting the Grassma.nn coordinates on the same footíng
as the usual cnmmuting coordinates. In this way the supersymmetry is explidt at aJl
steps. Moreover dynamit and statíc perturbation theories become remarkably similar,
the topology a.nd weight factors of Feynman diagrams being the same. lt is also COllVe·
nient to take the boundary condition in the La.ngevm equation at time -oo so that the
system is at equilibriurn at a.ny finite time, and therefore time transla.tion invarianre is
socured.

Al7.1 The (4>2) 2 Field Theory

The dynamirol oction. The bare dynamical action is:

(A17.1)

witb:
A(<t>o) f d&x [~ (8,.4>o) 2 + ~m~.pl + ~go (.P~) 2 }. (A17.2)

As we ha.ve shown in Chapter 17, the renormalized action is then:

(A17.3)

The RG equations. 'Ib renormalize action (Al7.1) we have introduced, in addition to

the static renormalization constants, a renormalization of the parameter íl:

ílo ílZfZ..,,

where Z is the field renormali:z,ation constant. The CS operator of Section 10.5 thent
takes tbe form:
â â â ...
Dcs = m âm + /J(g) ôg + fl.,(g)íl fJí! - 211(9), (Al7.5) i

where we ha"" defined

m~~ lníL
dm "",n"
 Al7.2 Renonnalization u.nd Stochw;j.ic Fíeíd Equations 399

In the case of dimensional regularization, equation (Al7.6) becomes:

d
'lw(9) = {l(g)dg In ( Zw/ Z) (AI7.7)

As usual the solutions of RG equations have to be combined wítb dimensional analysis.

One here finds:

(A17.B)

where we have introduced the vect.or notation 8 =(0,8).

WT identitie8. Let us also recai! the WT identíties (17.47,17.48) corresponding to
transformation {17.45) satísfied by the proper vertices after Fourier transformation:

(Al7.!l)

These identíties provide cbecks in pert.urbative calcutations.

. A17.2 Perturbative Calculation of fl..,(g)

To the action (Al7.1} corresponds a propagator ll.{k,w,ll,81):

(AI7.l0)

have omitted the fact.or 6;.j corresponding to group índices. For practical calcu!atíons
is actually more conveníent to use a rnixed representation for the propagator, Fourier
Orã.nd"~r.n....-1 on space but not in time:

(8 + 6'')' (-r)]}

(Al7.11)

whicb t (-r)ls the sign of -r = t - t'. Note that, in agreement with the analysis of
17 .5.2, ~ can be rewritten:

t.(k,-r,8,8') {l+!{Õ-Õ')[9+1J'-(9 /11)<(-r)J!}

x k2 ~ m• exp [-~ (k' + m') l-ri] .

form follows directly from the supersymmetry and the causalíty of the Langevin
(see also equations (35.15--35.18)).
400 Renmmalization and Stocha...tíc Field Equat:ions A17.2

From the supersymmetric form (Al7.1) of the dyna.mical action we deduce that the
dynamic and the static theory have very similar perturbativeexpansion.s and differ ma.inly
by the form of the propagator. For N = 1the ínteractíon vertex v<•l has the form:
y(•l=m'~62 (11t-82)P(lll 8a)~ (11 1 -8.). 2
(Al7.12)
4.
We have already calculated all static renorrnalizàtion constants to two-loop order in
Sections 1Lfi,ll.7. To calculate the new renormaliz&.tíon constant at leading order (i.e.
two-looj} order) we need only the 2-point function. Using the expressjons of Section 11. 7,
wefind:
rl2l = -~ [2- íw (e- il') (9 +O')]+ (k 2 + m 2) 62 (8 8')

(A17.13)
ín which the three diagrams D~, D2 ,D3, are given by:

D1 =6 (9
2 8' 1
) (21r)d
f r+m~'
ddp (Al7.14)

Do D1 f ddp dtd 2 6" .-n(p'+m')ltl __

(2?1)•
l_
p2 +m2

X { p2: mZ + ~o (Õ"- .9) [8"- (! E (t) (8'1 +ri')]}, (Al7.15)

_1_
(21r)2d
f dt e'"'' f d"v I
ddp .-n•(P·)I<il' TIJ
2 >-1

X { 1 + ~ 8 {pi} 6 (íJ- Ô') lO- 8' - é (t)(O + !!')]} , (Al7.16)

with tbe definitions:

P• - (k + P1 +Pz) , (Al7.17)
3
s (P;) =L (p~ + m 2
). (A17.l8)

The dia.gra.ms Dt a.nd Dz (after integration over t) are identical to the die.grams of the
static theory. Only D, contains a new dynruníc divergence contríbuting to the renor•
mali:mtion of fl. Actually Da can be wrítten as the sum of two terms, D~ 1 ) praportional
to 62 (8- 8') and which conto.in& the stetíc two-loop divergence a.nd a.nother one Di2l
wbich after an integration hy parts owr t can be written:

D(2) = [1-ljw (ô- ô') (8 + 6')] fdt eówt f ddp1 ddP2 e-O•(p;)!tlf2 n3 ~-1- •
3 2
(271 )d {211)d i= I P~ + m• .;
(Al7.19) >;,
The divergent part of the integral is a constant which ca.n be calcu!ated at w O a.nd ·:t"': =
k O. lntegrating over t, P2 and p 1 successively, one finally obtaíns: ,.,. ·
N +2 3
Z.., NJlin(4/3)""""24e+O(g ),
and therefore, using the results of Section 11.7,
N+2 2 (61n(4/3)
flw(g) = N 02 "72"'"g I) +0 (•)
g •
A17.3 &normalizatúm ond Stochastic Fíeld Equc.Wftu 401

AlT.3 The Non-Linear u-Model

Let us now consider the case of the O(N) non-linea.r u-rnodel (as definad in Chapter 14),
still with the supersymmetric dynarnics, to illustrate the díscussion ofSections 17.7,17.8.
The dynarnical bare action can be written:

él.Po.(ô~-e 8
<Po)+A(<t>o)],
1 2
S(.Po)= jdod.9dtfjd•xn (A17.22)
Uo t o ao ao ôt
Mth:
(Al7.23)

Note tha.t for practica.l reasons we have adopted normalizations which dílfer frorn those
of Sectioru; 17.7,17.8.
The renormalized theory is defined by the substitutions:

<Po = vlz{a(x,t), n(x,t)}, (Al7.24)

with then:
u(x,t) ,fz-•-n•, (A17.25)
a.nd:
uo=J..-•uZ9 , fl.ouo=fl.g!J·•z;z.,, h<Jvlz!uo hiJ'Jg. (A17.26)
We have called Jl. the renorrna.lization scale a.nd :set o: d. 2.
With dimemional regllla.rization, the dynamic RG function 17w(9) is then defined hy:

7Jw(g) = IJdd l
JJ. •• .n.
lnfl. = /1(g) dd In
g
(Z.,ZZg). (A17.27)

Dímension al analysis yjelds the relation:

(A17.28)

Perlurbation theory. The propagator is the same as in equation (A17.10) up to the

Mrrnalization:

_,n [r- ~iw(&- õ-) (1'1 +9') + ~ (k2 + h)6 2 (li- li'J]
(Al7.29)
UIJ w2 +!f (P + h) 2

calculate the 2-point function at tw(}-)oop arder we need the ,.• and the ,.6 vertires:

= ~<1 2 (8, -B2)ó 2 (ll 3 - e,)õ.,,,6,.,.

{A [-2 + í{wt + w,) (.9, -.9,) (8, + 9a)] + ( (p, + 1>2)

2
+h) li2 {81 -83)}, (A17.30)
= L6.,,
16g
2
• ó;.;.ó;,,,6 (8, IJ,) 62 (8.- e.) ó~ (83- Bs] ó0 (83- Bs)

{~ [-2+i(w1 +w2) (.91 Ba) (81 +9•)) +6'(6, Ba) ((p1 +1'1) 2 +h)}· (A17.31)
402 Renonnalization and Stor:hastic Field EquatioTI.> Al7.3

>················< o _O
Fi.g. 17.1 One--loop diagrams: the duttod llnes do not corr.,;pond ta propaga.ton; but are used
only to represent faíthfully the flow of group indiC<JS.

2-ptJint function at one-loop onler. At one-loop arder, the 2-point function then has
the expansion:

The coefficients whích correspond to the dia.grams of figure 17.1 have the following
expressions:

(A17.33)

(Al7.34)

(A17.35)

We recover the one-loop static results:

and in a.ddítion we find:

z., z;z, +0( 92).
This equation together with equation (A17.27) implies the ahsenoe of dynamks renor- ·
malization at this order:
=
1),.,(9) o (y') .

Two-loop calculation. We now have to calculate the one-loop diagra.ms with the
malization constants expa.nded at one-loop order, and the various two-loop
gen&nted by two 4-point vertioes and one 6-point vertex. Since the static reilor:ma.lizl•-·
tion constants ha.ve already been calcula.ted, to calcula.te 'lw(9) a.t two-loop order we
= =
only r(>) for vanishing arguments: k O, w O, 9 = 6' = O.
 Al7.3 Renormalízatíon anel St~c Field Equation, 403

o
-(-41 2 +1/'rr) - HN- 1)(412 - 32J)

o
2(-4/ 2 ) (N- 1)(-412 ) - ( -81 2 + 16J)
Fíg, 17.2 Two-Joop diagrOJ1lS: fllithful representation and contribution.

Figure 17.2 lists ali contribut.ing two-loop diagrams with their natural weight factms.
The integral J is given by:

(A17.38)

r<2l = - 4~., [I+ gZg (hz.;z)''2 I+ CN 2- s) I2u' Ji

- 4 ( N - 2) J g 1 - N ~ 3 Ndl g 2 + O (93 )] • (A17.39)

;·Ex:pat!dillg ali temu;, we find finally the expression of Z., at two-loop order:

g ( N- 1) g
1
N-
Z "" 1 + - + - - - - - - - I n - - - +O {g 3 ) .
., 2u :JE2 (2'11')2 2c 3 (2'11')2
24 g
7
(Al7.40)

expansion of lj.,(g) at order g 2 follows:

2
~(g) (N- 2) (1 -ln(4/3J)~ +O (g"). (AlT.4l)
(2.,.)
18 ABELIAN GAUGE THEORIES

With this chapter we begin the study of a new elas; of geometrical models which h ave
dominated the studies of fundamental intera.ctions for more than a century: gauge the-
ories. They are characterized by new physical properties and new technical difficulties.
Gauge theories can be considered in some sense as geometrícal generalizations of tbe
principal chiral rnodel. In particular the pure abelian gauge lield leads to a free lield
theory, as the pure abelian chíral modet
We devote first a whole cha.pter to a simple and physically important example, the
abelian gauge theory, whose physical reali~atíon is Quantum Electtodynamics (QED).
However, since many excellent textbooi<E deal extensively wíth QED, we wiil mainly
concentrate on the formal aspects of abellan gauge theories.
The set-up of this chapter is the following: We begin with elernentary considerations
about the massíve ve<:tor lleld in perturbation theory. We show that coupling to matter
field leads to lield theories \\'hich are renormalíza.ble in four dirnensions only if the vedar
field is coupled to a conserved current. ln the latter case the rna.ssless vector limit can
be defiped. The corresponding field theories are gauge imwiant, We then discuss the
specilic propertíes of gauge invaria.nt theories and mentlon the IR problem o[ physical
ohservables. We quantize gauge theories sta.rting directly from lirst principies. The
formal equivalence between different gauges is estahlished.
In Section 18.5 regularization methods a.re presented whích allow to overcorne the
new difficulties one encounters in gauge theories. The ~belian ga.uge eymmetry, btoken
by gauge fixing terrns, then leads to a set of WT identities which are used to prove the
renorma.hzabi!íty of the theory. The gauge dependence of correlation function• in a set of
covariant gauges io determined. Renormalization group equatíons foUow and we calculate
the RG tl-function at lea.ding order.
Finally, as an introduction to the next cha.pter, we ana)yze the abe!ian Higgs mecha-
nism. The appendix contains the deriva.tion of the Ca.simir effect and some more details
about gauge dependence.

18.1 The Massive Vector Field

The quantization of the free ma.ssive vector !ield does not immedia.tely follow from the
quantiza.tion o[ the scalar !ield a.nd thus l'llqUires a short àiscussion. In tbe fust of this
section we work in real time with the metric { +, -,-, · · ·} where th<l 6rst component is
the time component. Space-time is denoted by {x0 ;;;; t,:z:;}, z; E JRd-l, and the vector
fi"\d A,. {Ao, A;}. Finally time derivative is indic.a.ted by

The local O(l,d -1) inva.riant classical a.ction for a free massive vector field then can
be written:

(18.1)
)8.1 405
Allelian Gauge 7'h,eories
with
F,..(t,x) a,..A~(t,x) B.A,.(t,:z:). {18.2)
One may wonder ab\>ut this peculio.r fonn of the deriVB.tive term but is easy to verify that
the additional we could think adding, 8,.A,.B" A•, dependíng ofíts sign eitbercorresJlonds
to a Ao field witb a negative metric, or to an unhounded potential.
Quantization. The action (18.1) has one peculiar feature: the time cowponeot Ao of
lhe vector field has no conjugate momentum since the action does not depend on the
time derivative Âo. Actunlly the a.ction ca.n be rewritten

A( A)
f dt dà - I :z: [1;A;(t,:r}
• (•"ii - OlÔ;Ôj
_ m~ ) .
Ai(t, 1 2A;7(x)- ;jFij
:z:)- 2m 1 2(t,:z: )

+! ( (-8f + m2 )A 0 (t,x) + 8;À;(t,x)) (-8f + m 2 J- 1

x ((-ai +m2 )Ao{t,x) +8,Â,(t,:z:J)], (18.3)

where roman índices denote spooe components, and ôi is the space laplacian.
Theref~ Ao should not be considered as a dynamical degree of freedom and the
corresponding field equation

M
óAo(t,x) 2 +m2) Ao(t,:z:) +&;A;(t,:z:)
~ { -ih · =O, (18.4)

is a constraint equation which can be use(! to eliminate Ao from tbe action. This feature
rdlects the property that a massive vector field has only d-I physical degrees of freedom.
The reduced lagraogían density then takes the forro

C(A;)
1 .
2A,(t,z) (
li;;- &f8;8;
-m2 ) · I l 2
A;{t,:z:J- m 1 A,2 (t,:z:)- 4F;;(t,x).
2 (18.5)

After Legendre transfonnation we obtain the hamiltonian den&ity

(18.6)

wh.ere we have denoted by E, (because it becomes the electric field in the massless limit)
the momentum conjugated to A;

The differential operator -8,8; belng non-negative, the hamiltonían is posit.ive. The
<Jitantization procedure from now on is standard. It leads ro an euclidea.n functional
integral in which appears the euclidean reduced legraogian. Thls lagrangia.n hWl the
11npleasant properties that it is non-local and not O(d) space-time symmetric. We note,
however, that the dependence in Ao of the action (18.3) is quadratic. Therefore, we can
J•roceed in the following way: We substitute in the functional integral representation of
the partition function the initial euclidean lagrangian. We then perform the gaussinn in-
tegral over the time component. As w., know this is equivalent to solve the correspondmg
"quatiou of mot.ion, and we thus recover the reduced lagrangian. Finally the determi-
nnnt resulting from the integnl.tion is field independent {see Appendix AI9 for the more
406 Abelian Gauge Theories 18.1

interesting non-abelian c-). This pr"""" that i! we ha.d ignored allthese consideratjons
about quantization and immediately used the euclidean action,

(18.7)

with:
(18.B)

we would have obtai.ned the correct result (note that in the continuation to imaginary
time we have set Ao = iAd)·
CorrelatiDn .{lLnctíons. The generating functional Z( J) of A~-1ie!d correlation functions
ís then glven by;

(18.9)

Integyating over ali fields we obtain the generating functional W (J) of connected func-
tlons:
(18.10)

in which .:l~.(k) is the vector field propagator in mornentum space:

(18.11)

We verify that on the mass-shell, k2 = -m2 , the numerator is a projector transverse to

the vector k, propagating (d- 1} components belonging to the vector reprE>~entation of
the O( d 1) subgroup of O( d) which Jeaves k inVlll'iant.
As we have explained in Section 8.5, lle!d theories involving massive vcctor fields
coupled to matter are renormalizable only in dimen.sions d :<; 2, whlch iB ba.d for 4-
dimensional physics. An obviously related disease (because the propagator is a homo-
geneous iunction of k and m) is tbe impossibility to pass continuously to the massless
límít. However we note that ifthe BOilrce J,.(x) has the form of a conserved current:

{18.12)

then expression (18.10) becomes:

W(J) 21 f dd kJ,.(k)
- l
/;: 2 +m -
2 J,.(-k). (1s.l3r1;:

ThiE means that the propa.gator can be reploced by """ I (k 2 + m') whicb behaves uki.
the propa.gator of a scalar pa.rtícle. ln this case both problems of large moroentun{
behaviour and massless limit are solve<!. One ca.n now wonder why we have not used af
once such a propagator: the reason is that it propagates, in a.ddition to a vector field;"
a scalar particle with negative metric (like the regulator fields of Section 9.1). This .~··
better illustrated by a shart calculation. ><.
18.1 Abelitm Ga,;ge TluJorie8 407

More general prop<>.gators, interpretation. Let us add to the ..etion (18.7), a tenn of
the form of a regulat.or field actioil:

(18.14)

In tbe absence of a source for x, the generating functional

(18.15)

is proportional to the functional (18.9).

We now change va:riahles in the integral (18.15):

(18.16)

This cha.nge leaves F,.. invariant. lf the source satisftes the conservation equation (18.12),
the source term ís not modified. Only the vecLor fteld mass term is affected:

1 2 2 J 2 12 I J )2
2m A,.= 2"' A,. + mA.a,.x + 2 (81'x , (18.17)

and therefore:

(18.18)

We now integra.te over x(x) and finda new action S( (A~):

s{ (A~)= s (A~)+-~ j (a,.A~) 2 ddx, (18.19)

Ç p?jm,_ {18.20)
is easy to calculate the correspondiog propagator {~d,..,(k):

{18.21)

e
ali finíte values of the propagator behaves at large morneoturn like a scalar field
)TOJ>agatc>r. By varying { from O to +oo we describe a set of gauges: { =O corresponda
gauge, { = I is Feynman's gauge. For { = oo (the unitaf11 gauge) we recover
original propagator of the vector field.

6,.. k,.k.
[~~I... (/c ) =-/c, +({-1)~­ (18.22)
{k"J

e
for values of I 1' the terrn proportional to 1/lc4 may geoerate IR divergences
interactíng theories for dírnensions d 5 4.
408 Abelían Gauge Thwrie$ 18.2

18.2 Action with Fermlon Matter

We conclude from the previous a.nalysis that vector fi.elds coupled to conserved currents,
and thus associated with continuous symmetdes, are candidates for the oonstruction of
theories renormalizable in four dimensions. Let us now construct the simplest example.
We start from a free action for a ma.ssive fermion:

(18.23}

and want to add an O(d) invariant couplíng to a. vector field. The only íntera.ction term
which, from the power countlng point of view, has a. chance to be renormalízable is
proportional to

As we h!lve alroody noted in Soction A5.6, the frre Dirac fermion action (18.23) has a
U (1) symmetry associat.ed with the conserv.>tion of the fermioníc charge:

,P(x) =e"' •.{/(:::), (18.24)

This property rernains true aft.er the addítion of the interaction term. To this symmetry
corresponds a. current whose expression is obta.ined by calculatíng the variation of the
a.ction under a spa.ce-dependent group translormation {see Appendix A13.1). lf Ais
space-dependent, the va.ria.tion of the action is:

(18.25)

and thus the correspondíng conserved current J,.(:z:) is:

Therefore the int.eraction t.erm haG exactly the forro ;:,{ a vector field linearly coupled to
tbe conserved current J"(:z:). The complete action then is

(18.26)

The parameter e is tbe current-vector field coupling constant.

The tran«formations o[ Sectiou 18.1, which have led to the propagatOr (18.21 ), rely on
a change ofvariables in the functional integral ofthe form (18.16):

(18.27)

We verify that tbe induced varia.tion of SF(A,., ~. l/1) ca.n be ca.nceiled by a clla.nge of
the fermion va:riables of the form oi the trMsformation {18.24) with a space dependent
function A(x ). Therefore the algebraic tr!"lllsformations which allow to pass from a unitary
non-renormaliz.able action to a.n action non-u:nítary but renormali~a.ble bY power counting
are justified.
18.3 Abdian Gauge Theoríes 409

This result is Ya!id for the at:tion. Eventually we shall have to dlscuss the problem of
correlation functions. However before we want to examine the special properties of the
massless vector field theory.
Higher spins One may wonder why tbe strategy whlch bas led to a renormaliza.ble
thoory of vector pa.rticles does not work for bigher spins. Lei us take the example of the
symmetric traceless rank 2 tensor. Jt must be coupled to a conserved current which is
also a ra.nk 2 tensor. Only the energy-momentum tell50r T,..has the required property
(see Section A13.3). But Tp• has at least dimension four for d = 4 end once coupled to
a field of dimension at least one genenúes a non-renormalizable interat:tion. A similar
argument applies to a spin 3/2 field coupled to tbe supersymmetry current.

18.3 Massless V<:ctor Field: Abelian Gauge Symmetr-y

The action (18.26) in the massless vector field limit m {) has a remarkable property: it
has, at the classicallevel, a local (i.e. with a space-dependent parameter) U{l) symmetry,
called also U(l) gauge symmet'1J- The vect.or field then is called a gauge field. This
symmetry h as a geometrkal interpretation which we discuss now.

IB.:U Gauge symmet111

Tbe invariance of the fennion part of the at:tion can he seen to be a consequenoe of the
replacement of the derivative {}~ of the free fermion theory by the covariant derivative
â~ + ieA", which allows the transformatíon of tbe gauge field to Cllllcel the terrn conúng
from the derivative. We are thns reminded of the concepts of covariant derivative, affine
connectíon, curvature and parallel transport íntroduccd for Riemannian manííolds (see
Chapter 22). In particulu we note the símílarity witb the rotations of the local frame
ronsidered in Section 22.6. The dífference is that the vectors which here. are parallel-
transported do not helong to the spa.ce tang'entíal to the manifold (whlch is fiat), but are
vectors for Lie gtoup tr!lJlBformations.
Let us now indícate the correspondence:
(i) 1j>(x) and >tí{x) are vectors for U (l) transformatíons.
(íi) A,.(x} is the afline connection. The connectíon is expected to have three índices,
one, here callcd !J-, which refers to the manifold, and two which refer to the voup space.
the group U{l) corresponds to multiplication by complex numbers, tbey are here

D,. =8,. +ieA,.. (18.28)

(iv) It fe>llows that the curvature tell>lor is ieFl',:

(18.29)

sínce the group is abellan, it is easy to write down expllcitly the parallel trans-
U(C), which is an element of tbe U(l) gtoup, in terms of a line integral:

U(C) exp[-iefcA,.(s)ds,.]. (18.30)

410 Abdwn Gauge Theories 18.3

One immediately verlfies that the transformation of U(C) indueed by a gauge transfor-
mation (18.27) of A,., has indeed the expeded form:

U(C) U'(C) exp [i fc a,. A (s) ds,.] = U'(C) exp {i(A(y) A(x))j, (18.31)

where x and y are the end-points of C. By ttyin,g to build a renormalizable theory

for a vector in 4 dimensíons we have been naturelly led to íntroduce a new geometri-
cal structure, "" abelian gauge theory, which Í.'l just the quantum version of MaxweU's
electromagnetism.
Quantizatwn of charge. I f we introduce addítional charged fields, we have to assign
them charges which chara.cteríze theír transforrnatíon properties under U(l ). A deHcate
question here arises. Since the U(l) group has the same Lie algebra as the group of
translatioru>, propert.ie.s dependlng only on ínlinitesirnal gtoup transformations do not
require a quantization of charge. In particular in perturbation theory WT identities will
be true even if the charges are not ratíonally related llJld the necessity of a quantízation
of charges will therefore never appear. The conventíonal wisdom is that QED does
not imply charge quantization. Note bowever that the only known non- perturbative
regularization, based on a lattim approxiwation, involves group elements in the form of
pa:rallel transporters and therefore requires charge quantization (see Section 18.5).
The question remains open.
Charged scalar fields. With the geometrical ideas of parallel transport in rnind, it
ís easy to construct an gauge action for a charged scalar lield. We start from a U(l)
invariant action SB(<I>):

(18.32)

in which the field r/>( :r) is complex IUld replace the derivative ô,. by a cova.riant derivative.
The explicit form of the covariant derivative depends on the charge we aBSign to the field
t/>(x). Jf we assume that r/> eouples to A,. with a coupling consto.nt ee we find

Ss (rt>,A,.) = j d•o: [ID,.'i\1 2 +V (14>1')], (18.33)

D,. =a,.+ ieaA,..

Note that power counting allows three independent interaction terms, two linear and one
quadratic in the vector lield. The gauge symmetry relates them.

18.3.2 The mass/.e.5s vectur field /ímit

Geometrical and physical oonsiderations single out the tbeory with a massless vector
field. It is the only one which is exactly gauge invariant and it describes the physics of
QED because the photon is, at least to a very good approximatian, massless. We eould
therefore ha.ve restricted our discw;sion to the massleos c08e, as we shall do in the non-
abelian case. However, oonsidering the ma.ssless theory as a limit of the massive tbeory
provides us with a simple resolution to severa! diíficnlties.
First we h ave a.lready seen in Sectíao 18.1 that we could not WTite at once a propagator
for a m6Ssless vector field. In an interacting theory this diffi<:ulty persista, and if the
18.3 A belian C auge 'fheories 411

theory is gauge invruiant it is not only a disease of perturbation theory. Indored the gauge
symmetry implies that the actlon does not depend on one of the dyna.mical variables
wbich i'l related to 11 gauge transformation. In particular the functíonal integral is ill-
delined bec3U5e it i'l infinite by a factor which is the volume of the gauge uoup (this
statement ha.s a precise meaning only in the framework of lattice ;egularization).
This difliculty reflects the property that in electrodynarnics only the electromagnetic
tew;or F,_v is physical, and the vector field A,_ has been introduced as a ma.themstical
devioe that enable us to write a local cm'Bliant action which generates the classical
equations of motion. The vector lield A,_ contains redundant degrees of freedom and i'l
not 'Uniquely determined by the eqnations of motion. We show in Section 18.4 how the
gauge field action can be qua.ntized starting from lirst principies. The procedure is less
stra.ightforward and lea.ds to non-covaria.nt gauges, problems whose analysís we wanted
to postpone.
By giving a mass to the ge.uge field A,, we make ali field compon~nts dynamícal.
The symmetry properties of the action allow the algebraic manipulatioru; indica.ted in
Sectíon 18.1 (provided the action ís used to calculate only gauge invaríant observables,
see Section 18.7). We can eventually take the zero mass limit. As we have shown, in the
proce$8, the gauge-dependent part of the gauge field has acquired a dynamics: we have
"fixed~ tbe gauge.
Fínally the mass of the vector field providas the theory with a natural IR cut-off which
somewhat simplifies the a.ne.lysis. For this reason in what foUows we mostly work with
the massive theory, the gauge invariant theory 3ppearing as a limiting case.

18.9.9 Physical observabl~s and IR problems

In the limit in whicb the mass of the vector field goes to zero, the gauge ínvaria.nt limit,
correlation functions calculated in Feynman's gauge have a finite limit. In Section 18.7
we derive a relation (equatíon (18.94)) whlch exhibits the gauge dependence of the bare
2-point function. The corresponding re!ations for other correlation functions ca.n be
obta.ined hy expanding expre,.,;on (18.93). They imply that bare correlation functions
then have a limit in ali gauges. The renonne.lized correlation functions will then also
have a límit provided one has been careful not to choose IR divergent renormalization
constants (as it is usual in massless theories].
The important question is however to understand whether phy$ícal, i.e. gauge indepen-
dent observables are IR finite. The e.verages of gauge invariant ap<lrators have certainly
límits. On the rontxary we do not expect scattering amplitudes to be IR finlte. Actually
we have shown in Section 2. 7, using the eikonal apprnximation, that even in quantum
mechanics IR divergem:es appear in the phase of the scattering amplitude as a conse-
quence of the long distance behaviour of the Coulomb force. These divergenees survive
in the relativístic theory. However it can be shown that a.dditional relativístic IR diver·
gences appeac which only cancel if one adds to the sc3ttering amplitude, the amplitude
for producing any number of additionallow momentum gauge fields. This iB physically
l!l'ceptable because ga.uge fidds which have momenta smaller thtw the uncertainty in the
momenta me...urements cannot be detected. Since this question is somewhat teclmical,
we refer to the literature where detailed dlscussio!lB of this question can be found.
412 Abelian Gauge Thearies 18.4

18.4 Canon!cal Quantization and Gauge lnvariance

Although we have been able to construct a gauge ínvariant theory as a limit of a theory
of a massive vector field coupled to a conserved current it is useful to contemplate the
difficulties one encounten; when one tries to quantize a gauge theory starting from first
principies. Moreover, as we shall show in Appendix AI9, in the case of non-abelian gauge
symmetríes the massless limít is not continuous. 'Ve therefore show how, starting directly
from tbe classical field equations of a gauge in,=iant theory, it ís possible tu re<:over the
functional integral representation of tbe generating functionnl of correlation functions.
The problem can be solved by severa! dífferent strategies and we shall present two
of them, corresponding to so-call~.d Coulomb 's gauge and temporol gauge. We again
consider the simple action

S (AI', J,) = S (A~)- f ddx J"(x)A 10 (x), (18.34)

S (A,.) ~f ddx F};.(x), (18.35)

of a gauge field coupled to a conserve<! current.

18.4.1 Coulomb's gauge

We first proceed as in the massive ca.se and eliminate the field time component from the
action. Taking into account current conservation (note J0 = íJd), we then obtaín the
integral of a reduced lagrangia.n density:

C(A,) =
1 . ( a,a
âf1) •
2 A,(t,x) 6,1 - A1 (t,x)

In contrast with the massive case the actlon depends only on (6; 1 - 81â1 (<Ji) A;(t,x).
After Fourier transformation this impli<>s that the action does not depend on the com-
ponent of Ã;(t, k) along k, the space component of the momentum k. We recover tbe
well- known property that a massless vector field has oniy d- 2 physíca.l components. We
thus expand the vector Ã;(t, k) on a tr>msverse basis ef(k), calling Êa the corresponding
d - 2 components:

and J 0 the conesponding sources. The lagrangia.n density in these varíables beoomes

d-2
C (B.) I; [â,.B.(t, x)â" B.(t,x) + J.(t,x)Ba(t,x)l.
a-1

The quantizatíon is now straightforwa.rd. We ~entually obtain a functional integral over

the fields B •. The conesponding action, once expressed in terms of the initíal current
J,., is however non-local. We can reintroduce the components A; of tbe gauge
provided we multiply the íntegrand by b(â,A;). We finally retum to an integral owr
18.4 Abelian Gouge Tlm>ries 413

the time component, "" in the massive case. The final result ís the euclidean generating
functional in Caulamb 's gauge

Zeoul (J) ~f [dA~(x}ó (B,A,(x))]exp [-S(A) +f ddxJ10 (x)A~(%)]. {18.36)

Coulomb's gauge, in th<> abelian ca.;e, hM a nice physical interpretation: only physical
degrees of freedom propaga.te, but it lea.ds to non-covariant calculations, a.nd thls is a
=
serious drawback. In particular let us write the ga.uge field propagator (k {A:d,kl}}

(
w~J(kJ={,
k.l
w,\;J(I<J o, (18.37)

w'~l(k)
'3
=...!...
k2 (6··-
•J
k,k1
ki ) .

\Ve see that the time component does not decrea.se in the large time direction. Therefore,
with this propagator the theory is not expliçitly renormalizable by power counting.
We sti]J have t.o prove that this gauge is equivalent to the cmoaria.nt gauges introduced
in Section 18.1, but let us postpone this point and before discuss another quantization
scheme.

18.,4.!! The temparol gauge

In the non-abelian ca.se the quantizntion in Coulomb's gauge is a comp!icated problem
tUid therefore we now explaln another, more easi!y generalizable, method.
Let us write the Jield equations in real time t which correspond t.o the action (18.34):

(18.38)

The extension of tbe arguments which follow to a theory contalning matter fields coupled
in a gauge invariant manner is stra.ightforward.
The method relies on the observation that the gauge tra.nsformed of any 5oluüon of
equation (18.38) is again a solution. We thus can describe the set of ali solutionS by
restrícting the gauge field to a gauge section, considerlng for example on!y the solutions
5atisfying:
Ao(t,x) O, (18.39)
which Ao is the time component of the lield A ... Let us then rewrite r,quation (18.38),
·..,oru·atin" time a.nd space components, and taking into account the conditíon (18.39):

ô;Â;(t, :r)= J 0 (t,:r), (18.40)

Ã;(t,%)- ô;F;;(t,:r) ~ J;(t,:r), (18.41)

which the índices i !llld j run only from 1 to d-I. The equation (18.41) is simply the
equation thnt can be derived from the classical lagra.ngian density C;

C( A,)= !A;- V;}+ J;A;. (18.42)

conjugated momentum E;(t, :r) of the lield A;(t,%) is the electric field:

E;(t,%) À;(t,x). (18.43)

414 Abeli~tn Gauge Theories 18.4

The expression of the ba.lniltonian density follows:

H(E(x),A(x)) = !EJ(x) + iF,3(x)- J;(x)A,(x). (18.44)
The partítion function Z(Ji) then is:

Z(J;) =f !dA;]exp [-S(A;)+ f ddxJ;(x)A;(x)], (18.4&)

in which the eudidean action S(A;) is the covariant action (18.35), in which A.r- O has
heen set.
We stíll have to implement the constraint (18.40) which ís Gauss's law. After quan·
tization, ít becomes a constnúnt on the physica!ly acceptable states IP(A). Sínce the
conjugated momenta Ei are represented by dífferentíal operators -ié/6A, the condítíon
(18.40) then takes the form:
1 6
iâ; M;(x) iV(A) Jo(t, x)i'(A). (18.46)
We rerogníze in the Lh.s. of equation (18.46) the generator of time-independent gauge
tr!UlSformatíons of the field A(x) acting on w. In the absence of an externa! source
(Jo(t,x) = 0), the physica! states must be gauge invaríant. Thjg condition ís consjgtent
with quantum mechanical evolution because in the gauge (18.39) tbe theory has still an
invariance under time-independent gauge transt'ormations.
For a general externai source, the condition (18.46) gives us the state tra.nsformation
law. Consjgtency with quantum meclianical evolution then requires the commutation
a,
of the operator E, - Jo witb the hamlltonian. A short ca!cula.tion sbows that this
commutation is implied by the current conservation (see exercíse 18.1 ).
Let us fina.lly exhibit, for !ater purpose, a state satisfying the conditiDn {18.46) in the
case of two opposite statíc charges:
J 0 (t, x) = e !â(x x2) 6(x- XJ )], J, (t, x) =O. (18.47)
We consider the state:
IJ>(A) = exp [-ie t A,(s)ds,], (18.48)

in which C is a.n arbitrary path joining XJ to "2· We indeed verify:

I tJ I b
ia' óA;(x) IP(A)"' i 6A(x) iV(A V,A)IA;o
e l6(x- x 2 ) 8(x xi)JIP(A ),
a result that is consístent wíth equations (18.46,18.47).
Note that the representation {18.48) has tbe form of a parallel transporter correspond-
ing to time-independent gauge transformations. Thi< representa.tion, as well as its non-
abelian genera.li?;ation, will be useful in Chapter 33, in the discussion of the confinement
prob!em.
The propagator in lhe temporo! gauJie. UI: us calculate the gauge field propogator in
the temporal gauge (in euclidean form):
w},'l
• k2I (é ij -
~) + 1
kl k;kj
'ijkf' (18.49}

in which k..l is the "space" part of k, í.e. its projection on Rd-l. This propogator, as
in the case of Coulomb's glluge, has a large momentum behaviour which jg not uniform
and thuo, in contrase with the covariant gauges, leads to theories whlch are nc;t explidtly
renormalízab!e in four dimensions.
18.4 Abelitm Gauge Thearies 415

18.4.3 Equivalence with muariant quantization

We ha.ve obtained dilfernnt fniJctional iiJtegral repre:;entr.tions of the same theory. We
would like to show that, a.t lea.st for gauge invariant oboervables, they are formally equiv-
alent to the O( d) covariant representatlons we have diBcuosed in Section 18.1. We here
show the equivalence between the temporal gauge and a class of gauges characterized by
a gauge condition of the form

n~ (ô) A"(:r) = h(x). (18.50)

The vector n" is a constant or a differentíal operator. We shall eventually either set
the externai field h(:r) to zero to enforce the stríct gauge condition n,.. (8) A~(x) O or
integrate over it with a gaussian weíght ín order to generate actíons of the farm {18.19).
ThiB oovers ali the examp!es met so far. The arguments easíly generalize to otber gauges.
First let us rewrite the expression (18.45) as an integral over a d-component vector
field:

Z(J) f [dA~Jll. 6 (Ad(x)) exp{- f dd:c [1F!.(x) J"(x)A.(x)]}. (18.51)

We then use the identíty:

f [dll(:c)j I] 6 [n~(8)(ô~ll(:c) + A~(x)) h(x)] = const., (18.52)

inside expression (18.51 ):

Z n:: f [dA,.dAJl] 5(Ad)5 [n~(.8)(éV(:c) +A.( :c)) h(er)]

x exp [-s (A)+ f ddxJ,..(x)A_.(x)J. (18.53)

IVe then perform a change of va.riables of the form of a gauge transformation:

A .. (x) "'A~(x) - a.A(:c).

Sínre we h ave assumed that the current J~ is conserved, only the li-functions are modified:

6 (Ad) ó [n" (â) (8_.1\(:c) + A.(x)) - h(x)] - 6 (A" A){; [n. (ô) A,.(x) h(x)J.

The integra.tion over A can again be perfarmed:

(18.54)

lll!d therefore:
416 Abelian Gauge Theories 18.4

Sinre the result by corustruction does not depend on h(x) we can either set h(x) to zero
or integtate over h(x) with for example the gaussian measure dp(h):

(18.56)

We then obt!Ún

(18.57)

with
S••••• (A) S (A)+~ f d0x In., (â) A,.(x))
2
• {18.58)

Speciali?.Jng to n., (,- 112 8" we see that we have in particular demonstrated the equiv-
alence between the tempornl gauge Ad = O and the covariant gauges (18.19). !f instead
we choose nd O and n.1 = â_t, and set h= O we find Coulomb'a gauge.
The propagator. To the action (18.58) corresponds the gauge lield propagator:

{18.59)

Remark. The scdct gauge condition is recoveltld in the limit In~ I -+ oo, which exists
for the propagator but not for the action. To write explicitly the !imitíng action one has
to introduce a Lagrange multiplier l(x) which imposes the gauge condition:

s..... (A)= 8 (A)+ f ddx l(x)n.,(8)A.,(x).

18.-i.4 lnterpretation; the Faddeev-Popov quantization

The result we have obtaine.d has the following ínterpretation which can be only íustílied in
the lattice approximation on a finíte lattíce (""" Section 18.5). The problem of the gauge
inV<Oriant theory is that locality requires an actian with redunda.nt decees of freedom
or equivalently that the local gauge invariant action does not provide a dynamics te
the gauge degrees of freedom. We therefore supply them with a stochastic dynamics
the sense of Chapter 16. Let us write the gauge field Ai' in terms of a gauge field
projection of A., on some gauge section, i.e. satísfying some gauge condition, anda
transformation A:
A., :B~'+a,.A.

We assume that this decomposítion is unique. Gauge invariance tells us that the gauge :;;
action depends only on B I' and specifies its dynamics. · ··
To A(:~:) we ímpOGe for example:

in which h(:~:) is a stochll5tic field for whích a probability dmtributlon Íl3 given.
18.& Abelian Gau,ge Thearies 417

We now impose this equation in the functional integral. Normally we expect to get
the determinant of Lhe functional derivative of equation (18.62) witb respect to the fie]d
A{x). However here this operator is just 8 2 and therefore the determinant is a constant
which disappears in the oortnalization of the functiona.I integral,
The functional integral in the presence of sources for gauge invariant operators {poly-
nomials in the fields that are invariant in gauge transformations) only becomes;

Z f [dB,.dA[6 [82 A(x) + a,.B"(:r:)- h{x)j exp[-S(B)]. (18.63)

The functional measure [dB,.dA] is the decomposition of the lla.t measure [dA,.) into a
product of mea.sures on B,. and A. The action S( B) is the gauge invariant action S(A) in
which equatian (I 8.61) has been used. We now recognize tha.t we can rewrite the whole
expressíon in ter ms a f A-":

Z= j [dA,.]b[éi,.A,.(x)- h(x))exp[-S{A)[- (18.64)

Moreover since the result of the functiona.l integratíon does not depend on the dyna.mics
of A(x}, the result does not depend on the field h(:r:) either a.nd we can íntegrate over
h(x) with for exa.mple the gaussin.o mea.sure dp(h).

18.5 Perturbation Theory, Regularlzatlon

Ftom now on we will only consider the CCI\.,.,.iant gauges of Section 18.1. With the
prapagator (18.21) pawer counting is the sa.me as for a scala.r field. We can therefore
construct interacting thearies renonnalizal:ile for dimensione d $ 4. We shall specially
consider the dimension 4. Since WT identities play an essential role in gauge theories,
we have first to find gauge invariant regularizations.
Dimensionai regularízatíon. We h ave defined dimensional regula.rizatíon in Section 9.3.
Thís regularization is weU 5Uíted to perturbative calculatioru; in QED. Exa.mples will be
in Section 18.9. lt is inappllcable only in the cHSe of gauge theories with chlral
dueto the "ls problem.

lU Pauli- Ví/lars 's regularization.

this chapter the special prablems generated by chiral fermions will not be met, and
will be perfectly happy witb dimensional regularization. However for !ater purposes
a.n instructive exercise to try to 1l5e Pauli-Villars's regula.rization.
Because Fw is gauge invarlant and the action for the scalM X(x) of equation (18.14)
árbitra.ry, we can a.rbitrarily modify the gauge fiald propagator and therefore to make
118 oonvergent as we want. However the situa.tion is quite different for charged matter
Since we are anly allowed to use the covariant derivative, we get for example for
fermion action regularized expressions of the form:

(18.65)
418 Abelian Gattge Theoríes 18.5

..::!•-0-A:....
Fig. 18.1 One-loop contribution to the A. 2·point function.

Note this method of regulariza.tion, unllke dimensional regularization, also applies to

massleos or cbiral fermions.
In momentum space the propaga.tor now decreases indeed as ~c->•-l. However, by
trying to improve the behavíour of the propagator, we have at the same time introduced
more singular interactions. Expanding "0~•+1 in powers of Ae we obtain in particular a
term proportional to ~ 2 ' $_. Therefore the action contains a '1/1$-tjl vertex with 2s deriva.
tives. If we now consider a one-loop di.agram with only fennions in the loop and rptj!J_
vertices o{ this kind (figure 18.1 }, we see immediately that the power counting is indepen-
dent of s because there are exactly as many fermion propagators as vertices and therefore
the additional powers o{ momentum in the numerator cancel the additional powen; in
the denominator.
This elfect is howll'Ver special to one-loop matter diagrams. A several-loop diagrarn is
formed by fermion loopsjoined by gauge field !ines (figure 18.2).

'
-o==o~~
/
Flg. 18.2 Exrunple of a se..,.all<><>p diagram.

Since the gauge field propagator can be made arbitrarlly convergent, severa! loop dia-
grams can be mede superficially convergent. In tbe case of scalar matter diagrarns, scalar
self-interaction vertices can be added, but then the number of matter propagators ex-
ceeds the number of gauge field vertices and again the diagrams ca.n be made superficially
convergent.
We condu de therefore that the only remaining divergent diagrarns are one-loop matter
field diagrams whicb can be generated by the determinant coming from the gau88ian
integration over matter field in an externai gauge field. This analysís signals a difliculty
in constructing in general a regularized gauge invarillllt expression for the determinant
of operators of the form f!+ m +i e$. or D!
in the contlnuum and at fixed dimensiono
(see Section 20.3).
In the case of massive fermions the one-loop diagrarns can finally be regularized hy
adding to the action a s.et of boson (with spin) and fermion regul&.tor fields with suitable
masses (but of the arder of the cut-oH):

;p (M + 0) ,P ,__.L ;j,, (MF,< + "0) !/>; + <PÍ (Ma.• + "0) cp;.

i=l

Subsection 31.2.3 will provide an explicit example.

 18.5 Abeiian Cauge Theories 419

Let us, however, here discuss in 50me detail Scbwinger's proper time regularization
scbeme (see Appendix A9}:
Let llS assume tbat the dímension of space is even. Then:

det (I + M + ie..4) = det -rs (I + M + i e..() 'fs = det ( -1 +M i e;!{) .

It follows that we can write, at least for e small enaugh:

det(ií} + M + ie.if) = [det (M2 - 0 2 )] 112 (18.66)

lt is ea.sier to deal with tbe operat.or M 2 -l!l 2 than l!J + M because it is positive. Usíng
the identity:

in which the quantities <J ~· h ave been defined by equatíon ( A5.9), we can rewrite the
operator in the r.h.s.:

2
M -0 2 (IB.67)

which shows that it differs from the scalar operator only by a spin term proportional to
(]""''The regularized form of the d!'terminant is then given by:

lndet [Wl + M)(iíJ + M)- ]I 1

=-1 tr 1~ -[V
dt (t)- V. (t)J, (18.68)
reg. 2 -« t

with the definitions:

V (t) exp [-t ( M Dz - ~a,..F,..)] ,

2
.- (18.69)
V. (t) exp [-t (M 2 -&')]. (18.70)

In both the boson and fermion cases, in a gauge tra.nsformation of the form:

(18.71)

the ma.trix element.; (!!IV (t)l x) become:

(JJIVI x) ,__. .-••Aiul (!I lVI x) é•Aí•l • (18.72)

The tra.ce of the integra.nd is therefore by itself gauge invariant.

The general strategy in the case of Pau li-Villars's regularization then is the following:
we fust regularize tbe one-loop diagrams a.nd show that we can renormalize them by
.. gauge invariant counterterms, we then insert the renorma.lized diagrama in tbe general
díagra.nu; regularized by Pauli-Villa.rs's method.
In the case of the sealar determinant in whlch the spin term is absent, equation (3.19)
lea.ds directly to a path integral representatiol! of V(t). It then shows that the matrix
element5 (!I IV (t)l x} are gaussian averages of the pasallel transporters transporting from
1: to y. In the spin case, one would have to introduce a.dditional fermionic degrees of
!teedom beca.use the corresponding bamiltonian has a matrix form.
420 Abelian Gauge Thearies 18.5

18.5.2 Laltíce regularizatian

To undetstrmd how to construci a lattice regularization of a gauge theory, it is essen-
tial to keep in mínd the geometrical interpretation of the gauge field as a oonnection.
Since on the lattice poínts are splít, the gauge lield has to be replaced by link variables
correspondiDg to parallel transport along línks of the lattice (see Chapters 22, 33).
Gauge invariant interactíon terms on tm lattice then have the fonn:

,j, (:r+ an,.) ~~U (x + an,.,x) ,p(x), (18.73)

in which U is given by equation (18.30), the curve C beíng the link joining the points x
to :r+ an,. on the lo.ttice. We have called n,. the unit vector in 1J direction and a the
lattice spacing.
The link variable U;J, linking site i to j is an element of the U(1) gmup, whicb can
therefore be parametri~ in terms of an angle 8;;, and is such that:

(18.74)

We show in Section 22.4 that the curvature teDSor is associated with parallel transport
around a closed curve. Thís suggests tbat we should take as a regularized form of f dx F~v
the product of iink variables on a dosed curve on the lattice, the simplest being a square
on a hypercubíc lattlce. Such a product is dear\y gauge invariant. We then obtain the
well-known plaquetU action, each square forming a plequette (for details see Chapter
33):
L U(i,j) u U,k) U(k,l) u (l, i) (18.75)
aU p1&Quet.tes

We have denoted symbolically by i, j, Ir:,! four site. formíng a square on the lattice.
The typical gauge invariant lattice action corresponding to the continuum action of a
ga.uge field coupled to fermions then has the form:

s (u, .jj,./J) = {j L u,iuJI<u>~uu +"L t/Jn.;U.,V!i +L M,j,,,p,. (18.76)

pli~.qué"r,.t.e:.- Unka sit.es

We h ave denoted by i, j ,... the lattice sitos, {j and "' are the coupling constants.
action (18.76) ís ínva.riant under independent U(l) t.romsformations on eaeh lattioe site.
Th..., transformatíons are the lattí«l equívalents of the gauge transfonnations of the
contínuum theory. The mensure of íntegratíon over the gauge varlables is the group
ínvaríant measure, i. e. the flat mea.sure d6;j. Note that on the lattíce and in a finite
volume the ga.uge invariant action leads to a well--defined partition function because tbe -
U(l) gt"oup is compact: the volume oí the gauge group is (2~r) 0 if n is the number of
lattice sítes. However in the contínuum limit the compact cbaracter of the group Íl! lost.
lt is therefore n~ to fix the gauge on the Jattice in order to be able t.o construc'"'"'
a regularlzed parturbatian theory. Since we shall devote Chapter 33 to lattice gauge
theories, we postPone the discussion of thls problem. Finally it ís possible to add to the
gauge invariant lattice a.ction ~ term o[ the fonn Ecos 6;; to give a mass to the
field.
18.6 421
Abelitm Gauge ~·

18.6 WT Identities, Renormalb;ation

In this sect!on, for convenienete, we assume dimensional regularizatlon. In gauge theories

WT identitíes play an essential role because it is necessary to mamtain the gauge sym-
metry in order to ensure that the theory which is not explicitly unitary, ís equivalent to
a unitary thoory, at least for the "physical" observables, l.e. gauge invariant observables
(íncluding S-matrix elemento;).
Theír derívatíon is rather straightforward. Let us take the example of the action
(18.26) and cal! J~(x), fl(x) and fi(x) the sources for the fields A,.(x), !)í(x) and tfl(x)
respectively. We make infinitesimal gauge transformations (18.24,18.27) on the artíon in
presence of sources. The t.erms whicb are not invariant are the A,. mass t.erm, the gauge
fixing t.erm and the sources:

6 [S source terms] =-;f dx A(x) { G&- m

2
) a,.A,.{x) + a~J,.(x)

+ie [ii(x)!,b(x) t)/(x)1J(x)] }· (IB.77)

This leads, following the usual arguments, to an equatían for the generating functianals
Z (J, ij, IJ) and thus W (J, ij, IJ) which has the form:

{(m 2
~&)a~,íJ:(x) ie[ij(x)óij~x) IJ(x)éi)~X)]}W(J,ij,fl) ô,.J,.{x).
(18.78)
This equation is equivalem to a set of identities for correlation functions. Let us lin;t
write the equatíon for the gauge field 2-poínt function whicb is speciol After Fourier
transformation ít has t he form:

k W(2)(k) ={-"-·~·
2 2 (18.79)
" I"' k +l;m

Correlatíon functions with only matter lields satU;fy:

(f +m2) kpW!2n+l) (k;p!.- ••,Pn;q,, ... ,q,)

=e L [w(2 n) {p, .. . ,p; + k, ... ,pn; q, ... ,Qn)

•

in which k is the gauge field momentum, and p; and q; the momenta of T/1 and ;j, fields
respectively. The presence of additíonal externai gauge fields does not modífy the idBIJ-
titiel!.
Equation (18.78) iB a linear first arder partia! dUferentiaJ equation. It is tberefore
&traightforward to transfonn it into an equation for the generating functional of proper
•···vertices r (A~,~' !,b):

(18.81)
 Abelían Gauge Theories 18.6
422

lt can be verified that the equations (18.78) and (18.81) ha.,e the same contentas the
quantum equatíons of motion of the x-lield of Section 18.1.
Equatíon (18.81) has a simple ínterpretation: ít tells Uo tha.t r has the forro:

r r ,m. + 2 lj[••
m A~(x) + ~1 (8,. A•]
,.) dx, (18.82)

in which r sym. is' gauge ínvaríant.

Renonnalization. We perfonn a loopwíse expansion of the functíonal f. Because equa-
tíon (18.81) is linear, the tree levei appro:ximation satisfies the inhomogeneous equatíon
while ali hígher order terms satisfy the homogeneous equation. Callíng r1
the 1-loop
contribution to r we have for [ > 0:

Therefore the generating functional fz of 1-loop proper vertices is gauge invariant. The
singular part of the Laurent exp~sion in t: = 4 d of f1 is a!so gauge invariant, which
means that the divergent part r~>v is gauge invaria.nt.
The condusion is that the action can be completely renormalized by adding gauge
invariant counterterms. As in the case of the linear symrnetry breaking in Section 13 .3,
one <:an say that the terms which break the gauge invariance, the g"uge lield m,.;
term and the gauge fix:ing terrn, a.re not renormalized sínce they are not modified by
counterte-rms.
The full renotma.lízed actíon can then be written:

S, (A~,{>, w) f ddx [iz_.F!'" + !m2A!+* (â~A 1l

- z~{l(tf + M + óM + ie.J:J w]. (18.83)

where z.. is the gauge lield, Z>J> and bM are the 1/J field and mass renormalization con-
sta.nts. We note that there ís no specia.l renorma.lization <:onstant for the charge e. lndeed
if we introduce the bare fields •!P, {>0 , A~,

(18.84}

we see that we rnust set

(18.85)

In other words tbe tra.nsformation la.w ha.s not been renormn.lized:

8,. + te.JA~ = â~ + íeA,.. (18.86)

Gauge invnriance relates the renormalízation of the charge a.nd the gauge lield.
 18.7 Abdian Gauge Th<wriu 423

18.7 Gauge Dependence

1b understa.nd the physíca.l properties of the theory, it ís necessary to characterlze the

gauge dependence, bere the depeJ~dence on the parameter { whích appears in the gauge
fixlng term, of rorrelatíon functions.
Let us therefore ca.lculate the variation of correlation functions when we add a term
proportjonal f (éi"A") dx to the renormalízed action.
2

The fermion ll-point functicm. Vok first consider the fermion 2·point function. Using
twice the wr identity (18.80) we obtain a relation which, in Fourier apace, reads:

(18.87}

1'wo rernarks are now in arder; in perturbation theory only the second term in the r.h.s.
ís divergent. At the same time only the second term has a pole. Therefore on the fermion
mass--shell (p2 ~ - M'), the only effect of the change of the gauge fixíng term is a fermion
6eld r€normalizatíon: The physical mass M is gauge independent (a gauge dependence
of the mllSS would have b..en led to a double pole at r? -M 2 ). Ftom equation (18.87)
we deduce the edditional renormalization needed when cha.nging {:

(18.88)

and therefore integr ating aver Ç:

(18.89)

Geneml wrrelation functiun•. We could use the wr identities in the same way to
exhíbit the {-dependenceofall correlation functions. However, since tbe identities (18.78)
legitímate at the perturbative levei the set ofma.nipulations ofSection 18.1, we shall use
the latter to prove a general identity. Let us add to the action S (A, .;i, w) a new tertn
depending on a sealar 6eld x:

x, we see that only tbe, irrelevant, normalizaticm of the partition function

~lntegnLtir•g over
cha.nged. We now shift x(x):

I
x(x) ,_. x{x) + {m a,.A,.(x). (18.90}

this change of variables the action reado:

e; f dxx~(x)+m f
2
A,.(x)8,.x(x)
424 A be!ian Gauge Theorieo 18.7

The last term cancels the gauge fixing term, and the second term in the r.h.s. ca.n be
elimínated by 11 gauge transformation:

(18.92)

The functione.l integral representa.tion of Z (J, i), '1) then takes the form;

Z(J,IJ,IJ) j[dA.d,j;d.p]exp[-s(A,,fr,.p) ~m2 jdxA~]Jidx)

x exp {j dx [~ (B~x)2 + Hm x 2 2- xã,Jp(m + ijei•x/m w+ ,fre-><:./mf/J}, (18.93)

in which S (A, 'fi, ?f) is the gauge invari~>nt a.ction. Expanding in P"'!"rs of fi a.nd lj, we
can intevate over x- This calculation for example in the case of the 1)>1/J 2-point function
yields the ratio of bare functions corresponding to two different values of {:

(18.94)

lntrodudng the fermion field renormalization constants, we get the relation between
renormalized correlntlon functions;

(18.95)

In particular with the choiee (18.89) for Z.p (0, the relation becomes finite:

(18.96)

Equation (18.96) has additionlll c:onsequences: no other ~-dependent renormalization

been needed. lt follows that the renorme.lization constant z.
and thus, from equation
(18.85), aJso Z..t are gauge índependent. In the renormaiized action (1B.83) on!y isz.,
gauge dependent. '
Other .p-6eld correlation functíons ha"" a similar structure. The ~ dependence,
comes only from the x integration, fa.ctorizeo. lf we now consider correlation
wíth externai gauge fields, the previous analysls stíU applies provided the gauge
transverse, i.e. the source J~ satisfies apJ, = 0: Transverse externai gauge lields do
introduce any new gauge depend""ce.
Note that we can obta.in the gauge dependence of the full gauge lield corre!ation
tions by integrating over x after making a translation of a term proportional to
However identities (18.78) diroctly relate correlation functions with externai
na! gauge fields (in FouríeT space Al'(k) proportionai to k,.) to fermion field
functions.
18.7 425

Unitarity. Since on the mass-shell the propagator of the ga.uge iield 1s a tra.nsverse
projector:

(
S
""+ (ç- 1) k2 k•k•
+ ~m2
) qk2 + m 2 )
k. = Jc2 + em2 k,. mo for k
2
+m
2 o,
we have obtalned the gauge dependence of n!l matrix elements. In the mass-shell limit,
tbe e-dependence of the singular part of correlation functions beoornes simp]y .. mul-
t.iplicative renormalization (see Section A18.2). Therefore the properly normnlized S-
IHatrix elements are gauge independent. Being gauge independent, they cannot have the
{-dependent singularíties which we have introduced with the x-field to ma.ke the thoory
renormalizable. The full S-matrix, in tbe subspa.ce of physicnl states, is thus unitary.
Note that by using WT ídentities one can nlso directly prove that the contribution of
x-field cMcels in the interrnediate ste.tes in the unitarity relations.
Gav.ge ínvaríant operators. We have examined the gauge dependence of S-matrix
elements. From the point of v:iew of correlation functions, the only gauge independent
quantities are the averages of products of gaug€ ínvariant operators, i.e.locnl polynomials
in the field invariant under the transformations (18.24,18.27).
The simplest such operators are F," which select the transversa part of the gauge field,
~.(x).p(x) or more generally .,b(x]rA.P(x) in which the matrix r A is any element of the
nlgebra of"' matrices. Equation (18.95) shows explidtly the mechanism which makes the
rorrelation functions of ,P ge.uge dependent while ..j,(x),P(x) is gauge independent. When
in the product ,P(x),P(y), y approaches x the additionnl gauge dependent renormnli:t.ation
needed to make the product ,J;(x),p(x) finita cancels the gauge dependent part of the
lcrmion field renormnlizatíon.
To study the renormalization properties of gauge invariant operators one has to add to
action oources for tbem. The forro of WT identities is not modified. The D.fguments
18.6 are still valíd: the oounterterms are gauge invariant. This proves that
invariant operators mix under renormalization only with gauge invariant operators
lawer or equal canonicnl dimensions.
Non-gav.ye in11ariant correlation functiom in lhe unitary yauge. With the originnl
(18,7) for the gauge fields, ali eorrelation functions are "physicnl", but the theory
renormnlizable. However we have been able to construct some correlation functioru;
theory, the gauge invariant correlation functions, which have a IMge cut-off limit.
e
{18.94). in prosence of a cut-olf, leads in the large limit to an explanation for
surprising property:

W(2) (x !I) = exp

oo '
[-__!__f
(2•/ dd k I - eik{~-Yl] W.o(>) (x !I).
m>k2 ,
(18.97)

111 # O, the dominant term in the large cut-off limit in the exponential is:

e>
(2,f
f ctdk
m2P -e
2 sd Ad-•
(2·•l -;:;;:;-· {18.98)

in the physícnl representatioo ali noo-gauge invaria.rtt oorrelation functions van-

The explanatíon is the following: nlthougb the masa term breaks gauge invariance,
brea.king is not sufficient to prevent fluctuatiollJl coming from the gauge degreer of
to suppress these correlation functions.
426 Abeltan Gauge Theoríe• 18.8

18.8 The RG Equations

In this sectíon we derive RG equa.tíons in the case of the &ction (18.26), which corre-
sponds to massive QED with fermions, displaying the dependence of RG functions on the
g~>uge fixing para.meter {. Let f(l,n) be the lPl correlat.ion fundions corresponding to I
ga.uge fields, and n fermion paírs t/! and 1[,. The relatíon between b&re and renormalized
correla.tion functions is:

(18.99}

in which 1-1 is the renormaliza.tíon scale, and we have called a the loop expansion pa.ram-
eter:
(18.100)

Differentiating equation (18.99) with respect to 11- at bare p&rameters fixM., we get the
RG equations:

a a a a a
[~-'ail- +iJ(a) âa +tl(aHã( +'lm(a)mam +'IM(a)M ôM
-~TIA (a)- n').p (o)] f(l,n) (p,,q,;_u,a,{, m,M) = 0. (18.101)

Equation (18.85) relates ZA a.nd Z,, the gauge field and coupling consta.nt renormaliza.-
tion constants,

Therefore:
(18.102)

Moreover we have shown in Sectíon 18.6 that the parameters m and ~ are not renorma.l-
ízed, It follows that:

Finally in Section 18.7, we have shown that the renormalíz&tion constant ZA ca.n be
chosen to be independent of { (the minimaJ subtraction scheme satisfies this requirement).
Equations (18.102,18.103) then imply a. set of relations between RG functíons:

iJ(a) ""'!A(a),
tl(a) = -tJ[a),
llm(o) ~ iJ(o)/2.

In addition iJ(a) is independent of {. The functíon '71>1 ca.n also be chosen independent ofk
{, only the feunion field renormalizatíon is necessarily gauge dependent. ActuaJly from :,; '
equation (18.89) it is even possible to determine the gauge dependence of 'lo~>· A short·;-
caJculation leads ín the minima.l subtraction scheme to: ·
 Abelian Gauge Thwries 427
18.9

18.9 The One-Loop ~-Function

Fennion contrilrutwn. We nuw calculate the {i'-functlon at one-loop order in the case of
the action ( 18.26) whicb describes the interaction o f a gauge field with cha.rged ferrnlons.
We first evaluat.e the gauge field renormalization consta.nt a.nd then deduce the coupllng
constant renormalization from relatíon {18.85): DimeJJSional regularization will be used
in the calculation.
The one-loop contribution to the genera.ting functional of proper v<Jrtices eoming from
the fermion int.egration is:
(18.108)

Dífferentia.ting twice wíth respect to A,., we obtain the one-Ioop contribution to the
renormalized IPI 2-point functíon:

r (2]• (p) =
"
zA {'u,..p2 - P,.Pv
) 1
+ -p,_.p, +e
e
2 f d k tr [1',. (~ + iM) lv {~- p + iM}]
4

(2.-)
d [ 2
(k' + M 2 ) (p k) + M 2
] •

(18.109)
We verify immediately that, as expected, the one-Joop contribution is transveTSe and
e
therefore is not renormalired. Setting:

r~J (p)"' [z.. + e2 N4trlE{p)] (p 2 6,..v -P,.Pv) + p,.p,j{, (18.110)

in which Nd is the usual geometrie factor ratio of the surface of (d 1) sphere divided
by (2" J", we lind:

E (p) (18.111)

Using:
k2 k·p ~[(k2 +M 2 )+(k-p) 2 +M']-M 2 -jp', (18.112)

we can rewrite E (p):

1
:E (p) - - - -
- (d 1)r
f ddk { {2 d)
-
S4 k'+M
!P (d- 2)- 2M
2
(k2+M2)[(k
2
- -2 + --""'-....;_....-''---::------;-
}
2

p) +M2] ,
(18.113)

(18.114)

(18.115)
428 Abelian Gauge Theories 18.9

In particular the divergent part of E (p) is:

E(p)=2_+0(1). (18.116)
3e

This determines the A,. lield renorrnalization ZA and thus, from equation (18.85), a.lso
Z,:

(18.117)

(18.ll8)

Repla.cing Nd and tr 1 by theír values for d 4 (N4 = 1/8.,. 2 , tr 1 4), ~ obtain the
corresponding )9-function at one-loop order:

{}(a) = _, [ dIn~=Z.)] _,
ScaJ.ar boson contribution. We now consider the actíon

(18.120)

Again we calculate the one-loop contribution to the gauge lield propagator. Since there
are now twn interaction terms of the fonn q,2 A;
and q,2 A,., two diagrams contribute a.t
one-loop arder (see figure 18.3).

Fi!;. !8.3
Q 0--
Ch..,-ged bo..<;ons: tbe two onc-loop diagrams contributing to the A" 2-point functíon.

The two terms can a.lso be obtained from the expansion of:

{18.121)

One linds:

2! ddk (2k-p),{2k -p).

e (2")d (P +M2) [(k p}
2
+M'].
(18.122)
We again verify that the result ís tra.nsven;e. Settlng:
 18.9 Abelian Gauge Theories 429

and takíng the trace of expressim1 (18.122), we obtain;

p 2 (d-l)E(p) (18.124)

and therefore:

,; (d 1)~ (p) = 2(d- 2)1 s.d"k_l_

2
k + M'
+f ddk
Sd (k2
(r} +4M2)
+ Af2) [(k- p) 2 + M 21
(18.125)
The identíty already used in the fermíon case leads to;

(18.!26)

ln partirnlar the divergent part of E (p) is;

(18.127)

The last part of the calculation is the same &S in the ferrnion case. The difference between
fermions and booons comes only from the trace of the identity in the spoce of -r matrices
which yields an additional factor 4 in the rontribution of fermions to the 13-function.
Calling np and na the number of chàlged fermions and scàlar bosons respectively, we
finally obtain;
(18.1:18)

The sign of the {3-function. Note that in the domain of validity of expansion (18.117)
(e2 small), ZA sati.sfies:
ZA :S 1.
This property a.lso holds in the 1>4 -field theory. The field renormalization constant z.,
satisfies (equation (11.78)) z., : :;1. &th are consequences of the Kii.llen-Lehmann
rcpresentation for the 2-point function (see Section 7.4). In the case of the gauge field
the property is true because ZA i.s related to the transverse pa.rt of the 2-point function
to which unphysical states do not rontríbute.
Since ZAZ• = 1 we see that the sign of the {3-function in four dimensiono is }ized for
. t> small enough.

F'u"1/'S theorern. For more general perturbative calcula.tions the followlng observation
useful. Charge conjugat.ion implies that correlation functions without matter field a.nd
nn odd number of gauge fields vanish. The proof is based on charge eonjugation. Let us
· consider the contribution to the effective gauge field action, det (iJ + ie$. + M), which is
430 Abelian Gauge Theories 18.10

genere.ted by the integration over the fermion fields, and use the property of tbe charg••
conjugation matrix C introduced in Section A5.6:

det(9J + ieJ + M) ~ det [T(~ +i e$.+ M)j = det c- I [T(9J + ie$. + M)) C
~det(?l ie$.+M).

Therefore the interaction between ge.uge fieids generated by the ferrnions is even in A 1,.
Note iu particular the implication for Feynman diagrams: fermion loops with an odd
number of external gauge fields can be omitted.

18.10 The Abelian Higgs Mechani•m

As an introduction to the next chapter we now discuss the theory of a gauge field coupled
to the charged scalar fie!d <t>(x) in an unusual phase. We start from the actíon (18.33).
4
Renormalízabi!íty implies that the scalar field se!Hnteraction is of the fq\1 type:

(18.129)

The field q\(:r) ís complex and the covariant derívative D~ is defined by equation (18.28).
At the dassical levei, íf the gauge symmetry is exact, and the U(1) symmetry is not.
spontaneously broken, the gauge lield is massless and the scalar lield q\ has two real
components with equal mass.
However we know that another situation is possible in which the U(l) symmetry ;.,
spontaneously broken, and ç\(x) has an expectation value v, which for convenience wr·
assume real (we shall comment !ater on the signíficance of (q\) which is not ga.uge ínvari-
a.nt). Let us examine in some detail, at tbe classicallevel, the physics of such a situation.
We haw met an analogous problem in Section 13.4 and we have concluded that SSB of
continuou& symrnetries Jeads to the appearance of a massless state, a Goldstone partidv.
At the classi~al levei this can most easily be seen by parametrizing the lield ,p:

{18.130)

and verifying that the a.ction depends only on a.o,tbe lield O being therefore massless.
However if we use the same parnmetrization in the gauge theory, we observe that th11
li-field can be completely e!i.minated from the action by a gauge tran:sformation:

(18.131)

After this transformation the action S (A., q\) indeed reads:

S(A,.,p)"" f ddx [~F.:v+ ~(élpp) 2 + ~e2 A!(p+v) 2 + ~M2 (1>+u) 2 +!.i (P+V)j].
(18.132)
At the tree levei the spectrum of the tbf!Qry is thus composed of a massive vector lielrl
of mass e2"V2 and a massi""" scalar lield ca.lled the Higgs li~ld:

m 2 (p) = ~gt1 •
2
(18.133)
 l8.11 Abelian Gauge Theories 431

As a consequence of gauge inva:riance, no Goldstone field has been generated. This ís a.

111ost remarkab!e property, which is also at the basis of the Meissner elfect in supercon-
ductivity. It is induced by the long range forces (in the non-re!1ltivistic case the Coulomb
force) generated by the ma.ssless gauge field.
Note that the total number of physical degrees of freedom has not cha.nged between
•he symmetric phase a.nd the spontaneously broken phase since one degree of freedom of
1.he scalar field has been transferred to the vector field.
From the technical point of view, the surprising property of 1lttion (18.132) is tha.t
alt.hough it does not correspond to a massive vector fi"ld coupled to a conserved current,
wc nevertheless expect the physical observables to be calculable via a renormalizahle
!icld theory, while aga.in in the so-called nnitary gauge (the representation (18.132)), the
• heory expressed in terms of physical fields is not renormalizable. Tbere exists however
u relation between the massiw vector lield and the Higgs modei: íf we take the formal
uon-linear o-model limit of the action (18.129), Le. a. limit in which the bare mass of
1l•e Higgs field becomes infinite at lixed expectation value v, we recover the action (I B.7)
with the identificatíon m =eu.
To calculate gauge invarianl observables and S·mntrix elements, we have of course to
r•·turn to an action in whích the gauge has been fixed through a term proportional to
2
(il1,AI') • This corresponds to give a dynamics to the pha.se field B(x) which plays the
role of the A(x) field of Section 18.1. We now examine the structnre of the theory in this
n~presentation.

As a final remark let os 1l!so re<:all for later purposes that the mechanism of spontaneous
•.nnmetry breaking can also be used to give a mass to fermions in a. theory with chiral
•rmmetry (Sectíon !3.6).

18.11 Quantizatlon of the Higgs Mechanism

W<: 110W start from the action

S(A.-f;) =f ddx [~F;v + ~ (õ~<A•)2 + ~m'A~ + D,,PD,.p• + M <fJ<P' + ~ (t/J</!')2],

2

(18.134)
In wllich a mass has been given to lhe vector lield to provide an IR cnt-olf.
We assume that tf> hoo a real expectation vnlue v at the dassica.l leve!. We introduce
t 1;., real a.nd imaginary parts o f </J !llld set:

tf>(x) = ~[v+ rp (x) + ix(x)]- (18.135)

TI"' quadratk part S2 of the action then is :

(18.136)

""' that a~A, is coupled to the Goldstone lieid X· The corresponding propagatore
 A belian Gauge Theoríes 18.11
432

are:

(18.137)

and
(18.!38)

Tbe spectrum of tbe theocy consists in 3 physical states and the usual state witb negatíve
norm coniíng from the regulator. We see that in the ahsence of a ma.ss term for the voctor
field in the action (!8.134), the theory is potentially IR divergent in 4 dimensions. On the
other band with the ma.ss tffm the gauge symmetry is broken and the x-tield corresponds
really to a Goldstone mode. Even in the physical gauge a ma.ssless field is present and
coupled.

18.11.1 WT identities and renormalízation

It follows from the combined analysis of Chapter 13 and Sectíon 18.6 that after renor·
malization the oorrelation functioru; satisfy the equivalent of WT identities {18.78) and ·.
(18.S1 )- As a consequence the dependence of conelation functions on the pa.rameter C
can be determined as in Section 18.7. In particular only corre!ation functions of gauge
invariant operatorS and 8-matrix elements a.re gauge independent.
The explicit form of the WT for correlation functíons ís now rather complicated. Let
us here write on)y the identítíes corresponding to th~ (A,., x) two-point proper verticc;,
Calling v the expectation vnlue of the renorma!ized <p-field, we obtain by dilferentiati~~&
equation (18.81) with respect to A,.:
(18.13!1)

Differentiating then with respect to x we find:

k,.rt~ - íe11 r~~ = o.
Let us pa.rametrize the díffe:rent functions as:

r~'J = (l (k~) õ,.v- b (k 2 ) k.kv,

r~ ~ = ie11 c (k7) k~<,
7

r~~= tt(k
7
)

At the tree levei the values of a, b, c and d are:

a(k2) = e2v2 + m2 + k2, b(k2 ) 1~1/~.

{ c (k 2
) 1' d (k2 ) = k2 •

Let us now express the identity (18.140):

2
d (k 2 ) == k2 c (k )
18.11 Abdian Gav.ge Theanes 433

The identity (18.139) then lea.ds to:

a (k 2 ) - k 2 b (k2 ) - í?v 2 c (k2 ) k' /f;+ m 2 • (18.144)

In the k = O limit, this equation in particular implies:

a(O)- e2 v2 c:(O) = m 2 • (18.145}

Let us calculate the correspondíng connected correlation functíons:

(2) _ 1 ( k.,k") ~k~kv

W,. - a b,.w Jil + 1:,2 (k' + f;m')'
w<•J = - f;ievk, (18.146)
"x l<'(k'+f;m•j'
woJ- .2...
xx- ck"'

quantization method we have used above leads to massless fields and thus IR diver-
gences, even though the physical theory contained only massive fields. By the deverer
choice of a gauge which explícitly break.s the global U(l) synunetry af the action (and
. therefote gets rid of the GoldstoDe modes), it ís possible to circumvent this difliculty. In
. the notation of the actian (18.134) let us impose the condition:

(18.147)

lll which À ia a constant which will be adjusted eventually. The ímportant new feature is
that the operator M, functional derívatíve of equation (18.147) with respect to h, now
· on the dynamical lields:

(!8.148)

the associated determinant det M is no Jonger a. constant. This is tbe source of

of a kind we ha.w already encountered in Chapters 16 and 17: we have to
fermion fields to wríte det M in local form, and then use the induced BRS
to show that renormall111ation preserves the form of the action. We postpone
untíl Chapter 21, and here give only a discusslon at tree ordeT.
we íntegrate 0\ler h(x) wíth the distribution (18.56) and use the gauge
of the iniiíal action

s."'" (B,.,t/l) = s.TM (A,.,q,e-••A). {18.149)

then in the functlonal integral:

.pe-ó•A""' tfl',
obtain the elfectlve action s.lf
s.,. (A,.,;p,ê,C) j d"x {~F!,+~ (li,.A,. +Ãevlmo;/1) 2
+ ID,4>f

2
+M 2 1<t>l + ~gltfl4 1- C(a2 + >.evRe,P) C}, (18.150)
434 Abelirm Gauge Theoril!.~ 18.12

in which C and C are two scalar fermion fields which genera.te the determinant det M. As
we h1>ve explained on sn exarnple in Section A7.2, sca.lar fermions cannot be interprete<:!
as physicai partícles but are of a no.ture similar to Pauli-Villars regu!ator fields. We now
use the p&ametrization ( 18.135) and choose >.:

(18.151}

This is of course the va)ue only at tree ordex. The propagators then are:

w<2l-~+ (Ç-l)kp,k.
~v - k2 + e 2 v 2 (I<'+ e2v2) (k2 + {e 2v 2)'
l
(18.152)

w(z)
CC p + {e2v2"
The advantages of this gauge (introduced by 't Hooft) are that by construction there is
no A x propagator and that all unphysical lie!ds are messive and have tbe same mass
~e 2 vf. It suffices to prOVe ga.uge independence of physical observables to show that the
pole at 1<2 = -Çe 2v 2 cancels. The price to pay here is the more complicated form of WT
ídentities whicb now are mixed with BRS symmetry. We shal.l examine this question in
Chapter 21 in detail.

18.12 Physícal Observab]es. Unitarity of the S-Matrix

The unphysicai pole at k 2 -~m2 can be shown to cancel in physicai observables (gnuge
invariant Operators, S-mntrix) eíther tbrough a gauge dependence anaiysjs as we have
done in Section 18.7, or directly by usíng the whole set of WT ídentities and showing
explicitly that the pole coming from wJ~l ca.ncels the contríbution comíng from w~;>
in the intermediate state in unitarity re)atíons. As the expressions (18.146) show, the
resldues of the pole are related and tberefore one understands that a ca.ncellation is
possib!e. The proof is not very difficu!t but tedious and we refer to tbe literature.
In the limít m O, we expect also the pole at k 2 O to ca.ncel in physical observab)es.
According to relatíon (18.145), for P-+ O the difiereot propaga.tors behave like:

Again a direct argument based on WT identities for connected corre!ation functions

and unitarity re!ations ailows to prove that in the m 2 = O limit the )(-field derouples
from physical observables. Here we do not have an aiternative proof based upon gauge
dependence. However we shail construct one in a more general context by using a different
gauge.
Gauge invariant opemtors. When the action (18.134) is used, only a.verages of gauge
invariant operators are physícai correlation functions. The simplest examples are F ...
 18.12 Abelian Gauge Tfteories 435

a(x) =(v+ p(x)) "' ~t/>(x)t/>.(x). lo particular the pa.rameter vis really the square root
2

of (o} On the other hand, the action (18.132) can be rewritten in terlllS ofthis field a(x).
One attractive feature of this representation is that the measure of integration in the
functional integral is just the flat measure [da(x)j. Fl.trthermore all oorrelation functions
of the tra.nsverse part of the vector field and the sca.lar field correspond directly in thls
physkal representation to gauge invariant correlation functlons of the renormalizable
representation. However an inspectioo of the action written in terms of the Ap(x) a.nd
cr(x) fields does not provide a direct explanation for the finiteness, after the introduction
of finite number of rounterterms, of the correlation functions of F,.,_ a.nd á.

Bibliographlcal Notes
Ma.ny early artides about QED are reprinted in
J. Schwinger ed., Quantum Electrndynamics (Dover, New York 1958).
There exists of cour"" a vast literature devoted to QED. Among the classical textbooks
let us quote
J. Bjorken a.nd S. Drell, Relativistíc Quantum Mechanics, and Relativistíc Q-uantum
Pields (McGraw-Hill, New York 1964., 1965); AJ. Akhiezer and V.B. Berestetskíi,
Quantum Electrodynamic. (Interscience, New York 1965); C. ltzykson and J.-8. Zuber,
Quantum Pield Theory (McGraw-Hill, New York 1980).
and referenccs therein.
For infrared divergences ore for example
D.R. Yennie, S.C. Frautschi a.nd H. Suura, Ann. Phys. (NY) 13 (1961) 379; T. Ki-
noshita., J. Ma.th. Phys. 3 (1962) 650; T.D. Lee a.nd M. Nauenberg, Phys. Rev. 133B
(1964) 1549.
: WT identities are derived in
J.C. Ward, Phys. Rev. 78 (1950) 182; Y. Thkahashi, Nuovo Cimento 6 (1957) 371.
Quantum Etectrodynamics stíll provides the most accurate tests of Quantum Field The-
.. óry. For exwnple the ca.lculation of the eighth arder contribution to the electron anoma-
lous magnetic moment has recently been completed:
T. Kinoshita and W.B. Líndquist, Phys. Rev. 042 (1990) 636.
the muon magnetic moment, which is more sensitive to Strong and Weak lntera.ction

T. Kinoshita, B. Nizic and Y. Okamoto, Phys. Rev. 041 (1990) 593; T. Kinoshita,
Phys. R€1!. D47 (1993) 5013.
abelía.n Higgs mecha.nísm is discussed in
B.W. Lee, Phys. Rev. D5 (19n) 823.
general da.ss of gauges is constructed in
't Hooft, Nucl. Phys. B35 (1971)167.
statistical physics analogue of the a.belia.n Higgs mechanism ís superconductivity.
the theocy of superronductivity see
P.G. de Germes, SupeTCOnductivity of Metais and Alloys (Benjamin, New York 1966);
D.R. Tilley and J. Tilley, SupeTjiuidily and Superr:Dnductivity (Adam Hilger, Bdstol
1986}.
of vacuum fl uctuations are discussed in
CMimir, Proc. Kon. Ned. Akad. Wetenschap,. B51 (1946) 793.
436 A be/ian Gauge Theories 18.12

Exercises
Exercise 18.1
In the nota.tion of subsection )8.4.2 verify explicltly the compatíbility of Gauss's law with
quo.ntum mechaníca.l evolution in tbe case of a non-vanishlng exWJ1al current.

Exercise 18.2
For the one-IDOp ca.lculations wbich follow dimensional regularization will be used. For
the tTace of the unit matríx in the "r matrix space one can take the 4D vruue: tr 1 = 4.
Establish the Feynman rules. Calculate the fennion 2-point function at one-loop arder,
in the Feynma.n gauge, usíng the action (18.26). Infer the fermion ma.ss and fennion field
RG functions.

Mercise 18.!1
We now consider the QED actian (18.26), with a mass term for the photon and quantized
in a general covariant gauge

(18.154)

18.3.1. Write explicitly the WT identities .involvíng the A 2-poínt function and the
ii;.PA :!-point function both for the IPI and connected correla.tion functions a.nd verify
them a.t the tree levei.
18.3.2. Cakulate tbe inversa fermion 2-point function r< 2l(p) at one-loop arder, í.e.
reduce it to a linear combination of scalar "bubble" integrais of the form

It is convenieot to also introduce the in~ra.l I (11.)

18.3.3. Calculate in the bare theory the one-loop correction to the fermion mass. WhaL
can be •aid of the gauge dependence of the result. Explain.
18.3.4. Characterize the divergences and calculate the corresponding renormalization
constents. It is convenient to set d = 4 e and

What can be said of the ga.uge dependence of ZM and Z.p? Explain.

18.3.5. Write the regula.rized A~ii;.P vertex (1PI) at one-loop arder a.nd check the
ldentity relatlng it to the fermion two-poínt function.
Al8.1 Ahdian Gn:uge Throries 437

APPENOIX 18

Al8.1 Vacuum Energy and Casimir Effect

Let us here discuss the free vector field vacuum energy. Tbe vacuum energy is a gauge
independent quantíty hut bocause the formelism depends on the gauge this is not always
obvious. We then apply the result to the Casimir elfect.

A18.1.1 The free e/ectromagnetic field: Vacuum energy

The free massive vector field. Let us first consider the massive case. One way to calculote
the ground sl.ate energy is to evaluate the free energy in a large euclidean volume. The
result ís proportional to a space-time volume factor. Divíding by the volume we obt,.;n
the vacuum energy density, the energy per unit space volume:

c/space volume -lnZjspace-time volume.

In the inítial unitary gauge one immediately finds, in Fourier variables,

Up to an irrelevant constant we find (d- 1) times the vacuum energy of a free scalar
field of mass m, a result which is not surprising since the massive voctor field has d 1
degrees of freedom. The result can be verified directly by diagonalizíng the hamlltonían
(18.6).
Let us however repeat the calculation wíth the a.ction S( of equatíon (18.19). We find
instead
-lnZ = }(d l}trln(k2 + m 2 ) + !trln(k2 + {m2 )/Ç.
This gauge-dependent result for the va.cuum energy is dearly incorrect. The reason can
be sirnply undersl.ood. In the varíous nlgebralc manipulations whi<:h have led to the
nction (18.19) we have omitted field·independent normnlization factors. Thi.s ís jusl.ified
for correlation functíons but not for the vacuum energy. The additional term originates
from the x field we have added to the theory in equatíon (18.14]. Normalizing oorrectly
the x integral cancels the addítional unwanted factor.
The ma.ssle•s gauge fie!d. We have seen that correlation functions have a smooth
massless limít. Thís ís nat the ca:ae for the vacuum energy. Indeed in the ma.ssless
limit one degree of freedom of the vector field, the longitudinal mode, decouples from
.the tbeory hut st.ill contributes to the wcuum energy. Let us thus evaluate the energy
and then discuss the gauge dependence.
In the temporal gauge we can use the hamiltonian (18.44) (ín a va.nishing source). The
l,amiltonían can be partially dia.gonalí.zed by a Fourier transformation (see Section 7.1)

(AI8.2)

pisa space momentum. We immediately see that we have d-2 harmonic oscillators
state energy li>l in the field dírection perpendicul"' to tbe vector fJ, and a free
mechanical hamiltonian in the p díroction. However Gauss's Jaw implies that
f>. E(p) anníhilates ali physical states and therefore does not contribute to the energy.
438 A bdian Go.uye TheDrie• A18.1

We conclude that the vacuum energy is d- 2 times the vacuum energy of a free massless
scalar particle.
Covariant gauge. Let us insteed eva.luate the energy in a covariant gauge, using the
corresponding pa.rtition function. Naivcly we find

lnZ=!trln{k 2 6~·-k~kv(I l/ÇJl=!dtrlnk 2 ~trlnÇ.

llp to constant we !ind the energy of d massless states instead of d- 2. The discussion of
the massive example give "" a.n ídea of the problem. We bave omitted field independent
facton; ío algebraic transformations. F()l' insta.nce in the Faddrev-Popov quantW.ation of
Sectíon 18.4.4, if we wa.nt the multiplication by the integral over A( o:) to be an identity,
we bave to multiply the integral (18.63) by det( -&2). Such a factor cancela a cornplex
ar two real massless scalar booons. It reduces d to d 2. In non-abe!ian gauge theori""
"ghost fields" autornatic ally produce the right book-keeping.

A I 8.1.2 Casímir effect

ln Chapter 36 we di.cuss quantum field theory ín a finite volume or more general)y in
restricted geometries. However the slmplest exa.mple of fmite size effects ís the Ca.simír
effect, l.e. the attractive force between two paral!el perfectly conducting plates ín tbe
vacuum, dtJe to the change in vacuum energy produced by a cbange in boundary condi-
tions. At leading order, ali charged pa.rticles can be omitted beca use only mMSiess fields
have a significant contribution at large plate separation. Hence the problem is reduced
to a calcuJatíon of the change of v6Cuum energy of the free electromagnetic field due the
boundaty conditions. The oonducting plates impose to the electric field to be perpendic-
ular to the p)atês. lt is eB.'ly to verify tha.t this condítion is satisfied if the vector field A~
ítself vanishes on the plates. Caliing L the distance between the platês, .z: = O a.nd z = L
the p)ate positions, we thus integrate over fields whlch have the Fourier repr.....,ntation

A~(XJ.,•)- f dd-lPJ. L e1 P.l.·x.~. sin(:II'Z/L)Ã(pJ.,n),

n~l
(Al8.3)

where XJ. ..,.., the space-time coordinates in the remaining directions. Since ai! compo-
nenl.s of A~ satisfy the sarne boundary conditíons and since tbe field vacuum energy is
(d- 2), i.e. the number o( field degrees of freedom, tim"" the free sealar vacuurn energy,
we now solve the problem for a free ma.ssless scalar field.
The fru massless •calar theory. We consider tbe action for a scalar field ql(o:):

where the pbysícal situation of interest is d = 4. We Msurne that in one di:rection the
field satisfy the boundary conditions

qj(;q,z = 0) = q)(x.L, t =L) 0.

For the other directions we first BSS\lme periodic bound..,.y con<litions with finite size L .L,
with LJ. » L. Iotegratíng over the field we know tbat the vocuum energy E: is sirnply
given by
Lj_E: = LL 12 In (4,.L' J. + --y:.-
2
n
2 2 2
1r n )
. (A18.4)
n!!l n..1._ J..
A18.1 Abelian Gauge ~
439

Of course thiB expression is only tneaningful in the presence of a UV cut-off. Sínce

l ""~e-•'-
t:t E"----+0
In(€:•) +li'( I)+ O(E),

we replace it by the regularized f<mn

where A I/a plays the role of a UV cut-off. We now ta.ke the large LJ.. limit. We can
repÍace sums by integra.ls:

The nJ.. integral is gaussian. We then obtaln the energy per unit area

(Al8.5)

We evaluate the sum, using identity {Al6.26):

L 1 L
L eXp (->c2n2t/L2) = - - - -2 +-L exp(-L2n'ft}.
n~l 2 .,/ift ,Jit n?;l

The first term yields a contribution proportiona.l to LA 0 which is the vacuum energy Co
in tbe absence of boundaries. The second term yíelds a L-independent surface energy
of order A d-I due to the boundaries. Fína.lly the remaining terms which are cut-off
independent but depend on L give the interesting contributíon. After integration over t
it takes t.he form
(Al8.6)

(Al8.7)

4 one finds
A( 4 ) = 1440' "'
Casimir effect. The resultíng force between plat.es is therefore attractive. To pa.ss from
Lhe sce.la.r result to the electromagnetic result we hove to ta.ke into account the d- 2 = 2
degrees of freedom of the gauge fie!d. For t.he energy and force F per unít a.rea we find,
restoring the physica.l units,

1 dE ,.• fie
(E- 4) /Li= const. =:. F=~TT-=---
LJ.. dL 240L4'

qua.ntum relativistic effect is V<Jry sma.ll but measurable. lt is remarkable beca use,
though electromagnetic tn na.ture, it is independent of the va.lue of the electric charge.
440 Abelian Gauge Theoríes Al8.2

AI8.2 Gauge Dependence

To characterize the gauge dependence of a general correlation function with only matter
fields (or with transverse gauge field) one has to integrate exp!kit!y over the fie!d X-
Settíng:

wJ'"1 (x., .. ' ,>:,;y., .. ','!In)= w.;;n) (x"' .. , x,;yl,.' ',y,.)

X UJ'n) (X!,- .. ,x,; !Ih··· ,!/n),

we find:

where <> means gaussian lntegration over the x·field as deiined by €<JUation (1&.93).
The integral can be ca.lculated (for details see for example Sectíon 31.1) and yields:

(A18.9)

A( being the x-lield propagator:

(A18.10)

As indicated in Section 1&.7, ít is convenient. to take the ratío of correlatíon functions

e.
corre.spondíng to two finite values of The ratio between for exoonple the generic and e
Landau's gauge (e= O) is then still given by an expreesíon of the form (AJ8.9) but with
the propagator (A18.10) rep!aced by:

In expression (Al8.9) only the diagonal terrm; i= j wíth vanishíng arguments yield UV
divergences. One recognizes immediately the di_..,rgent factor (Z,.(e)/Z~(O))" !18 given
by equation (18.89):

Yj))- t: K( (x;- YJ)]}

(Al8.11)
lf, after renorma.lization, we expand the ratio between correlation functions in powen;
of e2 we note that ali terms but the first one correspond to a matter field correlatlon
function in which at least two externa.! línes have been joined by a x-fie!d propagator.
Therefore when we go on mass--shell these terms do not have the corresponding poles a.nd
thus do not contribute to the S-ma.trix, as indicated in Section 18.7.
19 NON-ABEliAN GAUGE THEORIES: INTRODUCTION

In Cba.pter 18 we have described the structure and the formal propert.iés of abelian
gauge théOries whích p~ov:ide a framework for the discussion of Quantum Electrodynam-
lcs. However, to be ab!e to describe other fundamental interactions, Weak a.nd Strong
Interactíons, it is necessary to generalize the concept o[ gauge t.heories to non-abelian
groups. In t.his chapter we therefore constroct a lie!d theory invariant under !oro~ i.e.
space-<lependent, transformatíons of a general compact Lie group G. !nspired by the
abelian example of Chapter 18, we immediately introduce the geometrical concept of
paral)el transport, 8 concept discussed more extensively in Chapter 22 in the examp!e
of Riemannian manifolds. Ali the required mathematical quantities t.hen appeat quite
na.turally. We quantize gauge theories and study wrne of the formal prOpertjes of the
quanturn theory lilre the BRS symmetry, We show how perturbation theory can be regu-
larized, a somewhat non-trivial problem. Wé fmally discuss general aspects of the Higgs
rnechanism.
In the appendbc we quantize rnMSive non-aooli8n gauge fields and briefly explain the
non-renorm8!izability of the lield theory.

19-1 Geometrical Conslruction

Let us consider a scalar field 1>(x) trnnsformíng under a linear representation of the
group G. Denoting by g a matrix belonging to the repre5elltation, we write the fie!d
tra.nsformatíon:
q,'(x) =: g1>(x). (19.1)
We want to construct a model wb:ich has a IDeal G-symmetry, i.e. a syrnmetry which is
defined by the transformation (19.1) in wbich the matrbc g depends on the point z. lf
we consider only invarinnt products of fields taken at the sarne point, globa.l invariance
imp!íes loca.l invariance. However if we consider invariant functioru; of fields and their
derivatives, or more generally products of lields taken at different points this is no longer
true. An ana.Jogous prob!em arises in tbe etudy of RiemaoníWJ manifolds. The analogue
of local group transforrnations are there reparametrizations of RlemanniM manifolds
&nd we refer to Ch8pter 22 for 8 more detailed geometrica.J and algebraic discussion .
.:. 'lb solve tbe problem it is necessary to introduce para.Jlel transporters U{C) whicb are
· curv~>-dependent elements of the group G, in tbe sarne matrix representation. If C ís
. 8 curve joining point y to X, and g(z) a space-<lependent group element, we write the
transformatíon of U (C):
U'(C) ""g(z)U(C)g- 1 (y). (19.2)
ll then follows that the vector 1>u defined by:

1>u = U(C).p(y), (19.3)

Lransforms by g(z) ínetead of g(y).

In the limít of an infinitesimal differentiable curve:

(19.4)
442 Non-A belian Gau.qe Theories: lntroduction 19.1

we can pa.rametrize U(C) in terms of the connection A"(:r) which is a vector for d-space
transformations, and a matrix beionging to the Lie algebra of G:

U(C) = 1 + A~(x)dx. +o (~dx,.ll). (lll.!i)

The transformation law for A,.(x) is obtained by expanding equation (19.2) at first arder
in dx,.:
(19.6)

Fram this expression we learn that, from the point of view of global transformation•
(g{x) constant), the field A,.(x), which ís usually calied the gauge field ar rang-Mills
field, transforms by the adjoint representatíon of the group G.
Co~ariant derivatiue. 'Jb the coonection A.(x) is assodated a covarlant deriv&.ive,
wbose explícít forrn depends on the tensor on which ít ís acting. 'lb obtain íts expresslon
when actíng on q,(x) we consider ín equation (19.3} the límit (19.4) of an infinitesimal
curve. Equation (19.3) becomes:

if!u (1 + A,(x)d.:,) (t/>(:r) + 8"q\(x)dx,.) + o(lldx,.ll)

(1 + dx.,D.,) ,P(.:), (19.7)

where we have introduced tbe couariant derivative D,:

(19.8)

D,. is both a dífferentíal operator acting on space va.ria.bles and a matrix. The ídentity:

shows that D~, the transform of D,. under the gauge transformatíon (19.6), is:

(19.10)

as expect.ed. In equa.tions (19.9,19.10) the products have to be understood as products

of dífferentíal and multiplicative operators.
Infinitesimal gauge tmMformations. Settíng:

g(x) =1 +w(x) +o (llwll), (19.11)

in which w(x) belongs to the Lie algebra "f G, we derive from equation (19.6) the form
of the infinitesimal gauge transfarmation of the field A,:

-6A, (x) a,w + IA,., w] D,w. (19.12)

In equatíon (19.8) we have given the form of the covaríant derivative corresponding to
the initial representation of the group, i.e. when acting on 4!í(x). Equation (19.12) yields
the form of the covariant derivative in the adjoint representation. lt is eal;J to verify
that:
(19.13)
 19.1 Non-Abe/ian Ga"!Je Theories: lntroduclion 443

in whicb A~ is given by equation (19.6) a.nd w;' by:

w'{x) = g(x)w(x)g- 1 [.r). (19.14)

Curoature tensor. From the fmm of tbe covaria.nt derivative we ca.n deriw an expression
for the curvature tensor F ~v· genera.lization of the electromagnetic tensor of QED:

(19.15)

which is aga.in an element of the Lie algebra of G, and which transforms, as a consequence
oi equation (19.10), as:
(19.16)
We recognize the action of the adjoint representa.tion o f the group. As we show in Cha.pter
22 in a different context, lhe curvature tensor is assocíated with pe.rallel transport a.long
a closed curv<J.
Expressíons in component form. ln rnany ca.ses ít is useful to write previous expressions
in component forrn. Let us call t 4 the generators of the Lie algebra, and introduce the (
romponents A:(x) of A"(x): (
A,.(x) A:(x)t4 • (19.17)
The covariant derivative (19.8) then reads:

(19.18)

Equation (19.12) involves the structure consta.nts !abc of tbe Lie a.lgebra,

(19.19)

This equation yields also the representation of the covariant derivative in the case of
adjoint representation.
Fínally the curva.ture tensor can be written:

(19.20}

WJd therefore:
F;~(x) = â,.A~(z) 8uA~(x) + !b~A~(x)Ae(z). (19.21)
. This last. expression is independeut of the group rep.--nl.al.ion.
Forms and gauge jields. It iS sometimes oonvenient to use the lat>guage of differential
forms and to associate with gauge tields 1-furrns, and 2-funns witlt tbe curvaCnre tensor:

(19.22,1

· where the dlfferentia.l operator d act.ing on rorms has been delined in Section 1.6: d ,.
. _dx,.B,..
444 Non-Abeli.an Gauge Theories: Jntmductiml 19.3

19.2 The Gauge Invariant Action

The simplest gauge invariant action S(AI') function ofthe g;al!gelield A,., and general-
ization of the action (18.35) of the abelian case, has the fon1''

S{A,.) - 21
4e
f d d xtrF 1•,_(x)· (19.23)

Note that we h ave not added a mass term for tbe gauge field eS in Section 18.1. We indeed
show in Appendix A 19 that in the non·abe!ian case the ze:rO rnass limit is singular. We
have also changed the norma]ization of the gauge field, in su,;h a wa.y that all geometríeal
quantities become independent of the gauge coupling constl>Dt. The sign in front of the
action takes into occount that, wíth our delinitíon, the matrbt F,.~ is anti-hermitian or
antísymmetric.
Two remarks are immediately in order:
(i) As in the abeUan case, tbe action does not provide a d:ynamics to all components
of the gauge tield. An additíonal dynamícs has to be províded to the degrees of freedom
whlch correspond to gauge trBJISformations.
(ii) In contrast with the a.belian case, because the gauge iield transforms non-triviaUy
under the group as equa.tion (19.6) shows (the gauge tield is "charged"), the curvatura
tenror F•" is not gauge invariant, and thllS not direct]y a pbysical observable. The
action (19.23) is no longe:r a free field aclion; the gauge field has self-ínteractions and
even the spectrum of the pure gauge action ís non-perturl1ative (some analytic results
can be obtained in dimensian two). We indicate in Cbaptef 33 how lattice gauge theory
provida. a framework for non-perturbative investigatíons. 'the difference between the
abelían and the non-abelian case ís remíniscent of the nau-linear chiral madel.
Matter field.s. It is now easy to add to the action gauge i~Vl!.riant terms for the matter
tields. For fermions they will have the form:

and for the boson fields:

in which V(~) is a group invariant functíon of the scalar field </J.

19.3 Hamlltonian Fbrmalism. Quantization

We tirst show that non-abelian gauge theories can be quonti.ed, using a simple hamil-
touian formalism, as we have done in the abelian case in &.::tíon 18.4. In tbis way we
construct a theory whlch, at ]east at the formal levei, is unitary because it corresponds
to a hermítian hamiltonían. AI! expressions in this sectiOn will be written for a pure
gauge theory, because the extension of ali arguments to a geberal gauge invariant tbeory
ís straightforward.
C!assico.l field equ.ations. We tirst consider real time fie!d tneory, we denote by t : "'" =
ixd time and the corresponding lield component by Ao " -;.A.,. We use the notation
19.3 Non-Abelian Gauge Th.eoriu: Introduction 445

Q to ~epresent the time derivatlve of Q. Space compone:nts will carry rom~ índices
(A;, x;). To the continuation to real time of action (19.23) conesponds a clas;;lca.l field
equation;
(19.24)

in which the explicit form of D,. is gíven by equa.tion (19.12). The equation {19.24)
do"" not lead to a standard quantization because, as we have a.lready discussed in the
abelian case, the action does not depend on Á{,, the time derivative of Ao. Thus again
Ao is not a dynamical variabk and C!!.ll be elimina.ted from the a.ction by solving th•
conesponding field equation. The Ao field equation is a constraint equation. However in
the ab5ence of a mass term, M in the abe!ian case the reduced action does not depend on
all space components of the gauge field. Only the combination [6,; D;(D:i)- 1D;} À;
appears (Di is th• covariant spare lap!acian). But in contrast with the abelian case the
projector acting on A; depends on the field itse!f, and therefore the procedure which
!ed to Coulomb's gauge here does not "'Drk, at least in its simplest form. Therefore we
ímmediately discuss the quantizatíon in the temporal gauge.
As in the abelia.n case we first note that íf A.(t, x) ís a solution of the equation (19.24),
any gauge transform of A"(t,x) is also a solutiou. 'Ib describe ali solutíons we can thus
make a gauge section, i.e. coosider a section in the space oí all gauge fields which intersects
once n.ll gauge orbits. We then represent ali solutíons by an element of the S<Xtíon and
a gauge transformation. One choice of gauge condition is specia.lly well-suited to the
construction of a hamiltonían formalism,

1\,(t,:r) =O, (19.25)

which defines the tempornl gauge.

Let us wríte the euclidean eq uation ( 19.24), separatíng the space and time components:

DoFOI< D1Fu = O,
(19.26)
{ D1F1o O.

The índices k, i vary only from 1 to d - 1 and rorrespond to space components. In the
gauge (19.25) the equatíons simplify and become:

(19.27)
(19.28)

with:
(19.29)
The equation (19.27) can be directly derived from the origínallagrangian in which the
tonditíon (19.25) has been used:

l:(A~) -tr f d"- 1x [À~(t,x)- 4~2 F'i1 {t,x)]. (19.30)

Expression ( 19.30) defines a conventíonal lagrangi~m. for the space eomponents of the
gauge field: Ek is the conjugated momentum of Ak and the corresponding bamiltonian
La.kes the form:
7-i(E,A),.,. (19.31)
446 Non-Abelian Gauge TheoriES: Jntroduction 19.3

However the only relevant solutions of the Hamilton equations are those which satisfy
the constraint equation {19.28), non-abe)ian generalizatíon of Gauss's law. Jt is easy
to verify that equation (19.28) is compatible with the classical motion, i.e. tbat the
Poisson brackets of the equation witb the hamiltonian vanish. The reason is easy to
understand: the gauge rondition (19.25) is left invariant by time-índependent gauge
transformations. Therefore time-independent gauge transformations forro a symmetry
group of the lagrangian (19.30) and tbus of the hamíltonian (19.31). The quantities
D1E1 are the generators, in the sense of Poísson brackets, of the symmetry group. These
considerations immediately generalize to the quantum hamíltonian, tbe operators D1Er
are the generators of a symmetry group ar the hami!tonian. The spare of admissible
physical states W(A) is restricted by the quantum generalization of Ga.uss's law:
b
~E,W{A) D'6A,(x) W (A)= O. {19.32)

This equation implies that physical wave functions are ga.uge invariant, i.e. tha.t physical
states belong to the invaría.nt sector of the symmetry group, space which is left invariant
by quantum evolution.
Quantization in the tempoml gauge, as in the abelian case, then follow:s completeiy
conventionallines. Returning to the eudidron jormalism, we conclude that the partition
function can be written:

Z j \dA,] ó (Ad) exp [ 4~ 2 j dax tr F!.(x)] . (19.3:>)

Remacks. The theory we have constructed is not exp!icitly space time covariant and
this Jeads to serious dífficulties as we bave already empbasized in the abelian case (see
Section 18.4). In the temporal gauge the theory it is not renormalízable by simple power
counting. lndeed if we examine the propagator in this gauge

(19.34)

in which k.< is the "spa«" part of k, we observe that it does not decrease at kd fixed for
large spatial momenta.
To solve these various problems we now show that gauge invariant observables can
equivalently be calculated from another quantum a.ction which leads to a theory whích
Íll explicitly covariant and renormalizable by power counting.

Elf'lívalence with a covariant gauge. We would like to work witb a covariant gauge
constraining a,A" rather than with the temporal gauge (19.33). We first define thc
functional L> (A,., 11):

L> (A,., 11) J

[dgl6 [ã,.A!(x) v(x)] , {19.35)

in which g denotes a space-dependent group element of G and At the gauge transform of

A,.. The ficld v(x), whicb belongs to the Lie algebra of G, up to this point is arbitrary.
We then multiply the integrand in expression (19.33) by the r.h.s. of equation (19.35).
Equation (19.35) implies that the partition function has not changed. We inverse the
order of integrations between A,. and g, and set:
(19.36)
19.3 Non-Abelian Gauge Theoriell: Intrnduction 447

The a.ction (19.23) is gauge invariant. The functiona.l 6 (A~, v), being an sverage over
ali gauge transfonnations is a.lso gauge invariant. Changíng g into g-l and dropping the
' in A', we then obtain:

We now integrate over the gau,ges g(x). The result of the integral j ldg]6 (A~) is gauge
invariant. We can thus ca.lculate it only for fields satisfying the condition Ad ~ O. We
then have to evaluate the deterrnina.nt ooming from the variation of A~ at g 1. We
find

Hawever since Ad ~ O the covariant derivutive Dd is equal to the ordinary derivative 8d

and the corresponding determinant ís a constant índependent of A •.
We see in expression (19.37) tha.t we need the functioJJal 6 (A", v) only for fields
satisfying a,.A,. v. This me3lll< in equation (19.35) that the 6-function determines
g L The functional 6 is thus the determinant of the operator obtained as the response
to an infinitesimal gauge transformation o f the gauge cond ition:

(19.38)

with
Még -a,.D,. (A)óg. [19.39)
In this form the theory ís no longer locnl. We thus rewrite detM as a.n integrnl over
scala.r fermions C and é, the Feddeev-Popov "ghosts", transforming under the a.djoint
representation of G. We repla.ce the .5-functio!l by its Fourier representation and cal! Jt. the
corresponding integration variable. This leads to the functional integral representation

Z =f (dA,.dCdéd>.] exp [-S (A,.,C, é,>., v)], (19.40)

where the quantum action S (A,.,C,é,>.,v) is given by

S (A,., C, é, >.,v) =f ddx tr [~ 4~~F!v + >.(z)lâ,.A,.(x)- v(xJI + C(x)â,.D,.C(x)].

(19.41)
Sínce the partition function is independent of 11 we can now averag€ over the field v(x)
with a gaussian distríbution {dp(v)]:

(19.42)

We here assume for sirnplícity that the representation of G il; orthogonal, and therefore
the generat<>rs of the Lie algebra are antisymmetric matrices. The av«raged partition
f11nciíon Z then reads:

Z = f [dA,.dCdCd>.] exp [ -s (A,., C, C,>.}], (19.43)

448 Non-Abelian Gauge Thwries: lntroduction 19.4

wíth:

S (A.,, C, C, .X) ""f d"x tr [- 4: F!. + Ç~\~(:r} + .X(:r)IJ~A.J:r) + ê(:r)8"D C(x)).

7 0

(19.44)
Except in the limít in whích t; vanisbes, it is also possible to íntegrate over .X(x) to find
a newquantum action S (A.,,C,ê):

(19.45)

Tbis forro of tl>e quantum a.ction is spectally suited for perturba tive calculations, although
geometrical properties ofthe action are more apparent on expression (19.44), in particnla.r
the EXpected BRS symmetry of the a.ction.
We bave therefore established the formal equivalenre between tbe two expressions
{19.45) and (19.33) oi the partítion functíon. The fonnal equivalence with other gauges
can be proven by a similar rnethod. The obvious drawback of the COVOJ"iant gange, which
leads to a cova:riant, local and ren<>rmalizabl<! theory, is the Iael< of explicit positivity and
thus mútarity. In particular Faddeev-Popov fermions do not obey to the spin--statistics
connection.

19.4 The Faddeev-Popov Quantization, BRS Symmetry

We now present a more heuristic form of the quantization procedure, whích lead• to the
same results, and which can be justified in the framework of lattice regularization in a
finite volume. Tlús derivation has one advantage: it provldes a simple explanation of tbe
appea.ranr.e of a BRS symmetry.

19.4.1 Faddeetr- Popov quantizatíon

As in the case of abelian gauge theories, the naive functional integral corre:;pondíng to
the gauge invariant action (19.23) is not defined because the a.ction does oot depend on
the degJees o{ freedom associated with gauge transformations. Following the method
íntroduced in Section 18.4, we give them a stochastic dynamics. Let us assume that we
can find a gauge seetion in the spa.ce of gauge fields which intersect once and only once
ali gauge orbits. Tbis is certain\y possible for small fields and therefore in perturba.tion
theory (see tbe remark at the end of the section concerning this problem beyond pertur~
bation theory). We mil B,. the gange field belonging to the section and para.metrize an
arbitrary gauge field A,. as:

(19.46)

The action (19.23) depends only on B,. and takes cOJ"e ofits dynamics. The dynamiC3 of
g(:r:) will be given \>y a stochastíc equation. A convenient oovariant choice ís:

(19.47)

in which v(x) is a stochostic field.

To justify the manipulations whích follow, we have to assume that the equation (19.47)
ha.s, at v(:r) and B~(:r) fixed, a unique solution for g(:r), rondition which again is met
19.4 N on-AlMlian Gauge Theories: Jfltro®ction 449

for sma.ll fields. We can then w-rite the partition functioo cor=ponding to the action
(19.23) and gauge condition (19.47) as:

(19.48)

where (dg dBp] is a symbolic notation for the factorhed measure induced by the llat
measure (dA 11 ]. Tbe matrtx .\ belongs to the Lie algebra a.nd imposes the condition
(19.47). The operator M comes from the varlation of equation (19.47) with respect to
g(x) (soo Chapter 16 for a general discussion). Fram equatiun (19.12) it follows that M
is the oper~or (19.39). Finally, since the adiao (19.23) is gauge invariaot, the whole
quantum action can be entirely expn.-.ssed in terma of A:

(19.49)

with:

S (A,., C, C,.\, v) j dd:t tt [- 4~2 F!. +.\(:r) (8,.A,.(x) v(x)] + C(x)ôi'D,.C(x))

(19.50)
We now a.vera.ge over the noise v(x) in the o;pecial caae of the gaussiWI distribution [dp(v)J
of equation (19.42). The a.veraged partition function Z then takes the form (19.43), where
the action S (A,., C, C,.\) is exactly the action (19.44).
Remark. As noted by Gribov, in oontrast with the abelía.n case, depending on the
value o! tbe gauge field B,.(x), tbe gauge condition (19.47) does not always have a unique
solution in g(x). Gribov's ambíguity is related to the property tbat tbere exist fields for
which the opera.tOr a,..D,.(A) has zero eigertvalm'5. This implies that the representa.tion
(19.43) is not meaningful beyond perturbation theory. The same arnbiguity has been
shown to a.rise for a la.rge cla.ss of gauge oonditions.

19. .;.2 BRS symmet'Y

One ooosequence of the quantization procedure is that the action (19.45) no longer bas a
ga.uge symmetry. On the other hand we expect from the e.nalysis of Chapter 16 applied to
lhe formulati<m of the quantization procedure a.s explained in tlle preceding su bsection,
that the action now h as a BRS symmetry, conrequence of the stochastic dynDJnics gíV»n
to the degrees of freedom of tlle gauge group vartables. Th obta.in the fonn of the BRS
tra.nsformation•, it is convenient to eeparate the ga.uge group degrees of freedom which
induce the BRS symmetry from the other degrees of freedom of the ga.uge field which
pla.y no role. We thus a.gain use tbe pEU"runetrlzation (19.46). From the analysls ofSection
15.4 we know the fnrm of the BRS tre.nsformations in the case of group ma.nífolds:

6~(x) = tC(x)g(x), 6C(x) = ~(x),

{ óC(:r) (19.51)
lÃ(x), .l>.(:r) o.

The field B,..(x) hB5 a dynamics provided by the ga.uge action and is not alfected by the
BRS transformation:
~19.52)
450 Non-Abelian Gauge Theories: lntrnductímt 19.5

Let us then calcula te the efiect of a BRS transformation on the field A,. We find:

(19.53)

Tliis result is not surprising: the transforrnations (19.51) correspond to an infinitesi-

mal change in the gauge group degrees of freedom a.nd thu• to an infinitesim5l gauge
transformation for A,.. We recognize also in equation (19.53) the transformation law
of the current associated with gmup elements as discussed in Section 17.7 (equations
( 17.75-17. 77)).
We show in Chapter 21, in a more general framework, that WT ídentíties associated
wíth the BRS eymrnetry (19.51,19.53) imply the structural stability of the quantum
action (19.44) under renormalization.

19.5 Perturbation Theory, Regularbation

Compared with the abelian case, the new features of the action (19.45) are tbe presence
of gauge field self-interactions and ghost terms. Let us first rewrite the diHerent teriilll of
the gauge action in component form to establish conventions and normalizations. The
gauge actíon will be written:

(19.54)

where the curvature tensor i1l given by equation (19.21) a.nd the tr&e of the unit matrix
ha. be€n swallowed into a redefinitíon of the coupling constant. In the rovarient gauge
of Section 19.3, the vector field propaga.tor ís {equation (18.22))'

(19.55)

ln four dimensions, as in the abellan case, the gauge field ha.s dimension 1. The ghost
lield aetion takes the form:

(19.56)

The ghost fields thus have a. simple li.b/p2 propagator a.nd canonical dimensjon 1 in
four dimensions. The interaction terms have ali dimension 4 and therefore the theory
is renormalizable by pawer oounting in four dimensions. The power counting for matter
fields is of course the sarne as ín the ahelian case. Let us write explicitly the new vertices
of the non-abelian theory. The gauge field 3-point function at leading order is:

(19.57)

The gauge 4-point function is given by:

lr( 4 )J:~~. = ií1 !feabfecd (15.,,6.. é,.aÓvp) + !oac]obd {6,...6,.a- 6,..6.,)

+foadfech (lipphvn- 6,..6._.)]. (19.58}
19.5 Nó11-Abelian Gauge Theoriçs: lntmduction 451

Ali terms are obtaíned from the firnt by exchangíng the indices to make the correlation
function totally symmetric. Finally there is a gflost gauge field vertex:

(19.59}

Notice that in the generic case both ghost fields play a different role and no syrnmetlj'
excha.nges thern. In a graphic representation ofFeynrnan diagrarns ghoot propagator Unes
generally carry an arrow. A special case is { O, Landau's gauge, in which, because
â"A" vanishes, the vertex can be antisyrnmetrized.
Infmred diuergences. Tn the covariant ga.uge, and in the absence of a Híggs mechanism
which provides a ma.ss for the gauge fields, only the gauge { 1 (the Feynman gauge)
leads to a th<lOry which is obviously IR finite. In oontrast to the abelían case, it is
impossible to gíve an e:x:plicit mass to the gauge field and to then constmct a theory
which is both unita.ry and renormalizable (for more deta.ils see Appendi.x A19). On
t.he other ha.nd we want eventually to prove the gauge independence of the theory and
t.herefore we must be able to define it for more than one gauge. One way to introduce an
IR regulator is to consider the theory in a. finite volume. This necessitates a discussion
of finite volume effects wbich wíll be given in Chapter 35.
Matter fields. The ooupling to matter fields differs from the abelian case only by some
goometrical fa.ctors corresponding to group índices. For example tbe roupling to fermion.s
generated by the covariant derivative (111.18) is simply j'~trj·
&gulari.zation. The problem of regularization in non-abelian gauge theories has many
featureo in comrnon with the abelian case, as well as with the non-llnear u-model. Three
regulatizatíon metbods are of particular interest: dimensional regularlzation, whích is the
most convenient for pra.ctical calculations and works in the absence of chiral fermions;
lattice regulariza.tion, which also leads to non-perturbative calculations (see Chapter 33
for details) and which a.gwn fa.ils for chítal fermions; and finally Pauli-Villars's type
regularizations which work in general except that, as we ha.ve already seen in prevíous
~c'Ometrical models, they do not regularize ali one-loop cií.;zc .· 'US. Let us discu.ss briefly
Lhe latter method. The regularized gauge a.ction takeo the form.

in whích P is a polynomial of arbitrary degree. In the same way the gauge function
il1, A, is changed in to:
(19.61}
In which Q is a polynomial of same degree as P. As a consequence both the gauge field
;>ropagator and the ghost propagator can be made arbitrarily convergent. However, as
In the abelian case, the cavariant derivatives generate new interactions whích are more
•ingular. lt is easy to verify that the power counting of one-loop diagrams is unchanged
while higher order diagrams can be made oonvergent by taking the degrees of P and Q
!urge enough. For massiVe fermions it is however stil! possible as, in the abelian case, to
,.dJ a ..,t of regulator fields, ma.ssive fermions and bosons with spin.
In the case of chiral fermions this is impassible, but the problem of the compatíbility
l><•lween the gauge symmetry and tbe quantum corrections is reduced to an explicit
v..rification of the WT identítíes for the one-loop díagrams. Note tha.t the preservation
452 Non-Abe!ian Gauge Thearies: Iniroductian 19.6

of ga.uge symmetry is necessa.ry for tbe CllliCella.tion of unphyBicsl states in physical

amplitudes, and thus essential to the physical releva.nce of the quantum lield theory.
WT identíties anel renor=alizatíon. From the BRS symmetry corresponding to the
transformations (19.51-1 9.53) follow wr identities. Their form ls somewhat complicated
and we postpone the discussion to Chapter 21, where we shall derive the general form
of the renorm&lízed action. Let us here glve the result only in the eJ<ample of t.he pure
gauge action in the rovariant gauge. We can assume that the gauge group G ís simple.
Then the renorma.lized form of the action (19.45) ls given by the substitution:

A",_. z~l2 A,.,

(19.62)
êc-zcêC.
This result h as a simple interpretation: the gauge structure (19.45)ls preserved and the
2
coefficient of (&JLA,.) is unrenormaliz.ed exactly M in the "belian case. However, unlike
the a.beli..u case, the tra.nsformatíon law of tbe gauge field and, more generally the form
of the covariant derivatíve, are moditied by the ga.uge tield renormalization.

19.6 Tbe Non-Abelian Higgs Mechanism

We have already discussed the Higgs mechanism in the abeli..u case. The b1<Sic idee. is the
same in the general case: the spont..ueous bree.king of a global symmetry associated with
a gauge invariance leads tomasses for gauge lields without generating mru;sless Goldstone
modes. Simply, beca.use the group strudure is richer, more complicated situations may
arise. Let us therefore coll5ider a da.ssical ga.uge invaria.nt action for a gauge field and a
scalar bo.son </J transforming under an orthogonal representation of the symmetry group:

S (A,., </J) f d•x [- 4~ 2 tr F~. (x) +i (D~</J) · (D.,</J) +V (<f;)J {19.63)

We assume that the symmetric potential V(</J) has non-symmetric mínima.

sence of a gauge symmetry, this is the situation which we have already analyzed in Sect!on
13.4. Since the spectrum in the classicallimit depends on the group strudure and the
represent,tion rontent of the tie!d </J we shall here consider only two set of examples.

1g.6.1 Simple Lie groups

Let us first assume that G, the symmetry group of the action, is simple and is tbus
also the gauge group. Mo~ we assume for simplicity thnt the field </J belongs tQ
an irreducible representation. We cal! H the sta.bilize.- of the vector v, the minitnum
of the potential. We separate the generators ta of G in the mBtrix representa.tion into
two subsets a :<=; p corresponding to the Lie algebrB i:(H) of the subgroup H, and
complement.ary oet .qG f H). We p..rametrize the scalar field <f;(x) by:

</J(x} ~[I:t"lf'(x)] (v +p(:z:)),

a>p

in which the vectors p a.nd {tiv•} span twc orthogonal subspoces. The transformation:

<f;(x) ,__.., {11"(x), p(x)},

19.6 Non·Abelian Gauge Theories: lntroduction 453

Íllsuch that the new fields p(x) ....,d 9"(x) can be e:xpanded in JX191"r'5 of .p(z) -v. In the
absence of gauge lields, we have used tbe representation (19.64) to filiow tbat the lields
8"(x) couespoud to ma..sless Go]dstone modes índUCed hy tbe spoutaneous hreaking of
the G-symmetry.
Equation (19.64) can also be viewed BS alocai group transformation relating the two
fields <I> and p +v. If we perform on tbe field A,. a gauge transformation af tbe form
(19.6) with:

g(x) exp [2..:: t"B"(:c)] , (19.66)

a>p

we completely elimina.te the fields B" from the aetion. lu faet we have fixed (at least
part.iaily) the gauge. lf we now examine the scala.r field contribution to the aetion, we
'""' that for p =O it reduces to a mass term for the gauge field:

(19.67)

with tbe mass matrix C! af<:

(19.68)

The matrix a.,p is positive and has a rank equal to the number of generators of L(G j H)
which is also the uumber of fields fP, i. e. tbe would-be Goldstoue bosons. We conclude
that tbe spontaneous breakiug of the G-symmetry generatcs no Goldstone bosons but in-
stead giveB masses tú ali gauge lields except th03e which are MS<Jciated with the unbroken
subgroup H. In the case in whích the symmetry is completely broken, ali components of
tue gauge field acquire a msss.
!f one considers directJy the d"-""ical action obt.ained a.fter the gauge tra.usformation
associated with group element (19.66), the set of massive vectúr fields can be quantized
in a completely standa.rd wa.y. AF, in tbe B~lian case, the quantized theory is however
not renormaliz.able.

19.6.2 The G x G •ymmetry

. Another possibílity is thnt the symmetry group o f the action is the dítect product of the
group G by a.nother group G'. We here consider only the simplest example where
symmetry group is G x G a.nd G is simple. We assume that the scalar boson field q.
a matrix transforming under G x G by:

(19.69)

which g1 and g 2 are two elements of G in a matrix representntion. We further assume

one minimum of the potential is proportional to the unit matrix q. = vi in such a
that the subgroup H is isomorphic tn G with elements of the form (g, g). We recall
the coset space G/H is then a synunetric space (soe ApPendix A15.4). As Bbove the
be Goldstone bosons oorrespond to gauge tramformations and can thll8 be <'limi-
from the action. In this eJ<ample all components of the gauge field acquire a mass
the symmetry corresponding to the gauge field ÍS complet.ely broken. Furthermore
romponents of the gauge field have the same mas$ mA:

(19.70)
454 Non-Abelian Gauge Theories: lntroduction 19.6

The action obto.lned after the gauge transfonnation specified hy equation (19.66) contains
only physica.l degrees of freedom a.nd the quantization of ali vector fields is strà.lghtfor-
ward. From the poínt of view of the ínitial theory the gauge ha.s been completely fixed,
hence the name of 1tnita'Y gauge. We have constructed an a.ction for massive vector fields
transforming under the adjoint representation of a symmetry group G. The correspond-
ing field theory is, however, not renorma.lízable by power countíng. The difference with
the nwssive vector field tbeocy we exB.llline in Appendix Al9 ís that the suitable additi<>n
of some scalar fields makes thls theory equiva.lent, at least for the physícal observa.bles,
with a renormalizable theory with additional unphysical degrees of freedom.
Remarlc. li """' formally take the non-llnear model limit, i.e. send the masses of ali
rernaining scalar fields towards infiníty at v fixed, we obtain an actíon for a self-interacting
massive vect.or field (see Appendix Al 9).
The SU(2) example. Let us discuss more specifica.liy the irnportant example of the
SU(2) group because it can be considered as a simpli6ed version of the Standard Model
of week-electromagnetic interactions which will be descrihed in Section 2ü. L We take for
scalar field </>a 2 x 2 compleoc ma.tri:x transforming under the (1/2,1/2) representation of
SU(2) x SU(2) (see also Section 13.6). The simplest action then reads:

S(A,.,</>) =f d x [ 4~2 F!.+tr(D,.</>) 1 D.</>+atr</J</Jt +~(tr.p.J>I) ],

4 2
(19.71)

We have represented the gauge field as a 3-component vect.or A,.. The curvature tensor
reads:
F,.v = lJ,.A.- &.A,.- A• x Av, (19.72)
and the covariant derivative octs like:

(19.73)

The T matrices are identical to i:he u matrices defined in Section AS.l. We choose
the expectatíon value of </> proportionà.l to tbe unit matrix. It is then conveníent to
parametrize </> under the fonn (see Sectíon 13.6):

(19.74)

in whlch q and 71" are real lields. Let us wríte the form of an infinitesimal gauge tra.ns-
formation in this representation;

6A,. = lJ,.w A,. x w,

6u = !w·.,.-,
{
671" -juw + ~w x 71".
The scalar field a.ction in these variables ís given hy:

SKOiar f d4 x [~ (a,.u ~.,...A,.)'+~ (a..... +faA,. !A,. x,.)'+V (o•+or 2)].

(19.76)
As we have already discussed in Section 13.6, if the potential V has degenerate cl!L'lsical
mini ma, the field </> hM a non-zero expectation value. We may by oonvention decide
the expectation value corresponds to the compon.mt u:
{u) =t!.
 19.6 Nan-Abdian GafJ.!l• Thoories: lmroductíon 455

Then the symmetry SU(2) x SU(2) is broken down to the diagonal SU{2) group. ln
the absence of gauge fields, the 7r-field becomes a mas:sless Goldstone boson. Here the
>r-field can be eliminated by a gauge transforma.tion, in such a w3y that the total action
written in the unitary ga.uge becomes:

S (A~, o) = j ddx [ 4~ 2 F!,+~ (8pa)2 + ~"" 2 A! + ~.,.2 + ~o•] (19.77)

This is a 0(3}-inva.ria.nt a.ction in whích t.he vector field is " 0(3) vector and the sca.lru-
field a 0(3) scala.r. The gauge field mass m11 is gíven at tree arder in terrns of the a-field
expectation value tJ by:
mA evf2. (19.78)

19.6.3 Gauge fixing of th.e Híggs mode! in a covariant gauge

lfwe consider the contribution (19.76) to the action, we see that when the a-tield ha.s an
expectation value, a. term of the form &,.11: ·A~ is generated whích introduces a mi.xing
between the would-be Goldstone boson 11: and the longitudinal part of the vector field.
This is a feature which we have already encountered in the abelian C!>lie (Section 11:1.10).
· As suggested by 't Hooft, it is possib!e to use the gauge function to elimina.te such a
term. In the SU (2) example we ca.n take M gauge function F:

(19.79)

After a gaussian integration, the corresponding contribution to the a.ction is:

(19.80)

(19.81)

Sgboot f ddx [aPC · (fJ,.C- A,. x C)+ fC (o-C+ 1r x C)J. (19.82)

leading order the term f) I',.· A,. is eliminated by tbe choice:

.\ =e'v. (19.83)

gauge has two advanta.ges: it decouples the gauge field from the would-be Goldstone
and therefore simplifies the propagaton;; by explicitly breaking the global SU(2) x
J(~!)-•>vn1rnetry, it genera.tes a mass for the '11"-field which is no longer a Goldstone boson.
gauge the propagators (equations (18.152)) have no poles at zero momentum and
IR problems are enoountered:
2
w<•l- e'6,.., e ({ -l)k,.k.
""- k'+m~ + (k2 +m3,)(k2+(m;.)'
1
w~~ (19.84)
k' +(m;,_'
w<•l - 1
cc- k2+{m~'
458 Nan-Abelian Gauge Theories: Introduction Al9.0

is that, due to 5pace gauge invaria.nce, Gauss's law commutes with the h!lJlliltonian.
Therefore if it is sa.tisfie.d at one time it ís sa.tísfted at alllater times. An integration over
Ao(t,x) will be infinitely redunda.nt, as we immediately verify since this procedure Jee.d~
to the initíal oction and thus to an undeftned functional integral.
The massle.ss limit. Provided ~ only consider gauge invariant correlation functions,
we can introduce a gauge condition a.nd follow ali the algebraic steps of Section 19.4. The
only modifica.tion, which ls induce.d by the non-gauge invariance of the rnass term, is thal
the field g, a.ssocíated with the gauge transformations, rema.íns coupled. lt is easy to
verify that the resulting action is the sum of terms dueto the gauge fixing procedure anti
a gau~e invariant part obt.a.íne.d from the action (Al 9.4) by t.he substitution A" - A~
(equat10n (19.6)):

(AI9.&)

with:
D" fJ,+A,(x). (Al9.6)
We thus ftnd:

(Al9.7)

We t<lCOgnize a G x GfG non-lineM a-model (see Section 15.4) in which one of the G
components of the symmetry group has been gauged. ln contrast with the a.belian CI>.'IC
the scala.r field has a self-coupling a.nd is coupled to the gauge field.
Tbe non-linear a-model actíon is not renormalizable for dímensions d > 2 (see Chapter
14). 1f we assume that the theory has a cut-<:>lf A, we see that the g-fteld Jluctuation&
will not be very much da.mped because m « A; perturbation theory is not pa:rti<:ub.rly
reliable. Moreover the zem gauge fteld mass limit appean; as a. strong coupling limit,
therefore we do not expect the scalar field to decouple (in perturbe.tion theory we
get IR divergence<).
We also note tha.t the complete action can be consídered as the limit of a Higgs modcl
action in which the bare mass of the Higgs field h as been sent to infinity. Calling r/l the
scalar field we can view action {Al9.7) as the farmallimit, when g goes to zero, of:

Recalling the equivalence between the non-linear a model and the </>4 Jield
Chapter 30), we me.y specula1e that, beyond perturbntion tbeory, the actions
and (Al9.8) lee.d to the same renorrnalized oorrelation functions.
20 THE STANDARD MODEL. ANOMALIES

In Chapter 19 we have discussed the structure and the forma.! properties of non-abelian
ga.uge theories. We now apply this forma.lism to the description of oome genera.l properties
of the Stendard Model of Weak, Electromagnetic and Strong Interactions. In partículas
we ca.lculate the RG 1'1-function ofthe strong sector (Quantum Chromodynamics or QCD)
and verífy the property of asymptotic freedom.
The v.'8ak--1llectromagnetic theory (w.e.m. ), thanks to the sma.llnes« of t.he coupling
constant at low energy, has now been tested quite systematically, in particular in e+ e-
rolliders, and when ra.diative corrections n:re taken into account, provides an accurate
description of all experimenta.! results. The situatíon for QCD is somewhat different.
High-energy so-called inclusive results can be ptedicted, as a consequence of the prop-
erty of large momentum esymptotic freedom (see Chapter 34). How=er low energy
properties cennot be derived from perturbation theory, the elfective coupling being too
large. Therefore evidence for the validity of the quark confinement scheme relies on
non-perturbative, numeríca.l, investígations of lattice gauge theories (see Chapter 33).
In :some cases, when gauge tields are coupled to axial currents, the V.'T identities
which are necessary to prove the consistency of gauge theories are not satisfied beyond
the classical leve!. There al'<l spoilad by anomalies. Therefore the second part of this
cbapt<.r is devotad to the discus.sion of this pbysically ímportant problem. Results are
· lllustrated by some physica.l consequences like the 1ro decay and the solution to the U (l)
problem.

20.1 The Standard Model of Weak-Electromagnetic lnteractions

·. Let us now brielly descríbe the Standard Model of weak and electromagnetic interactions
· it provides the main physica.l application of the non-abelían Higg.s mechanism. We
fin;t restríct the díscussion to one generation and two flavours, and indicate at the
it can be generalized to severa! generations. For deta.iled phenomenological
apj>ücati,ons of the modcl the rea.der is referred to the literature. The g•mge group of
is SU(2) x U(l). The form of tbe actiou for the Higg.s field sector can be
from the actíon considered in the previous oection by gauging a U(l) subgroup
rema.ining non-gauge SU(2) symmetry. The scalar field is thus a SU(2)-doublet.
pattern of symmeiry breaking is the same as befoie but the consequence is now
an unbroken U(I) gauge symmetry remains whàse g<lnerator is a linear combination
the origina.! U(1) generator and one ofthe SU(2)generators, and which corresponds
electromagnetic interactíons. All fermions are either singlets or doublets. Since the
group is a product of two groups, the model depends on two independent gauge
and therefore weak and electromagnetk interactions are combi11ed rather than

(20.1}
456 Non-Abeli<>n Gm>9< Theories: Introdurtion 19.6

in which mA iB the mass of A. in the tree approximation (equa.tion (19.78)). Furthennore

ali unphysica.l states have a mass wlúch expl.icitly depends on the gauge parnmeter (.
Unitarity. This property can be used to prove unita:rity of tbe physical S-matrix: Tbe
S-matrix satisfies a generaiized unitarity relation in which in the intermediate states
one must include ali partícles both physical and unphysical. By showing that the S-
matrix does not depend on the gauge, one proves sirnultaneously that the contrlbutions
of unphysícal states cancels in the interrnedíate states and thus the S-rnatrix is unitary
in the pbysical subspace. A general proof of this kind will he given in Chapter 21.

Bibliographical Notes
Non-abelian gauge fields were lirSt discussed In
C.N. Yang and R.L. Mills, Phys. Rev. 96 (1954) 11!1; R. Utiyama, Phys. Rev. 101
(1956) 1597; M. Gell-Mann and S. Gla.show, Ann. Phys. (NY) 15 (1961) 437.
Original articles on the quantization of gauge theories include
R.P. Feynms.n, Acta Phy~. Pvlon. 24 (1963) 697; B.S. OeWítt, Phys. Rev. 162 (1967)
1195, 1239; L.O. Faddeev and V.N. Popov, Phys. Lett. 25B (1967) 29; S. Mandelstam,
Phys. Rev. 175 (1968} 1580; M. Veltman, Nucl. Phys. B2l (1970) 288; G. 't Hooft,
Nucl. Phys. B33 (1971) 173;
The axial gauge was pmposed in
R.L. Arnowitt and S.L Fickler, Phys. Rev. 127 (1962) 1821.
General linear gauges were proposed in
G. 't Hooft, Nuc~ Phys. 835 (1971) 167;
and further discussed in
K. Fujikawa, B.W. Lee and AJ. Sanda, Phys. Rev. 06 (1972) 2923; B.W. Lee and
J. Zinn-J ustin, Phys. Rtv. D7 ( 1973) 1049.
The quantization of constmined systems is discussed in
L.O. Faddeev, Theor. Math. Phys. 1 (1969) 3.
The BRB syrnmetry of the quantized action was discovered hy
C. Becchi, A. Rouet and R. Stora, Comm. Math. Phys. 42 (1975) 127.
Tbe firSt discussion of the ambiguity of gauge conditions iB due to
V.N. Gribuv, NucL Phys. B139 (1978) 1.
The Higgs mecbanism was introduced in
P.W. Higgs, Phys. LetL 12 (1964) 132; Phys. Rev. 145 (1966) 1156; F. Englert and R.
Brout, Phys. Rev. Lett. 13 (1964) 321; G.S. Guralnlk, C.R. Hagen and T.W.B. ·
Phys. Rev. Lett. 13 (1964) 585; T.W.B. Kibble, Phys. Rev. 155 (1967) 1554.
lt was cast into the fra:mework of renormaliza.ble field theories in
G. 't Hooft, NucL Phya. B35 (1971) 167.
Tbe corresponding wr identities were derived in general form in
B.W. Lee and J. Zinn-Justin, Phys. Rev. 05 (1972) 3137,
a.n article whose Appendix C ha.s directly inspired Appendix A19.
A selection of original articles ha.s been reprinted in
Gauge Thetlry of W<!ák and Electrnmagnetic Interoction.s, C.H. Lní ed. (World
tific, Singapore 1981).
An early review ís
E.S. Abers and B.W. Lee, Phys. Rep. 9C (1973) I.
Finally among the textbooks sre for exarnple the works of Itzykson and Zuber,
Faddeev a.nd SlavnoV, quoted in the ptefare.
 A19.0 lVon-Abelian Gauge Theoriea: ln!roductíon 457

APPENDIX 19
MASSIVE YANG-MILLS FIELDS

Foc complet"'less, we briefly explain why, in contrast with the a.belian case, ít is not
possible to construct a renonnaliza.ble field theory in which a mBSS is given to the gauge
field hy directly ad ding a ma.ss term to the action. We therefore consider the real time
la.grangian density:

C(A,.) (A19.1)

The first problem we meet is quantization. We know that A 0 , the time cornponent of
·: tbe gauge fietd, has no conjugated mornentum and thus is not a dynamical varia.ble.
lt ca.n be eliminated, using the corresponding field equa.tion. lt is octually algebraically
ronvenient to Jirst determine the hamiltonia.n, and use the equa.tion of motion afierwards.
The hamiltonian is bere obtained by perfortning a Legendre transformation only on space
oomponents. The conjugated momenta E, are ( roman índices mean space components)

ar. 1
E,- BÀ; =-êlFo;.
an integration by pa.rts the hamiltonian density 'H. can be written

'H. - tr E; À, - C
1 {'F,
ê7 1 2A2)]
4 ij +2m i • (Al9.2)

a familiar property of the Legendre transformation we ohtain the Ao equation of

0= _!!f_ (Al9.3)
&Ao
equa.tion determines Ao which can be eliminated from the hamiltonian. We then
the functional integral representation of the evolutíon operator in tertnS of the
hamiltonian, integrating over {E;, A.}. We note however that in tlús expression
hamiltoDian can be replaced by the initial hamiltonian (Al9.2) provldad one
also over A 0 . lndeed the integral over Ao is gaussian: The integration thus sets
the solution of the field equation (Al9.3) a.nd yields a constant determínant beca.use
coefficient of AÕ in expression (Al9.2) is field-independent. Finally, integrating with
initial hamiltonia.n over E, at A!' fixed, we recover the initial action. Thus the
function can be exprcssed in tenns of tbe simple euclidean oction

S(A") = -~ j ddxtr [~F!,(x) + ~m2 A!], (A19.4)

~ flat field integration measure. In contram with the massless gauge-invariant situ-
this functional integral is well-defined.
ID the massless case one can fol!ow the sa.rne strategy. The Ao field equation
~ ...........~.-er, ::m Ionger ~"'7~--..,.~!~~ ~4_0 but instead yields Gauss's law. One may then
why one does not impose Gauss's law by íntegratíng <1Vf!f Ao as above. The reason
 The Standarr:t Model. Anomalies 20.1
460
in which F~" is the curvature correspondlng to the SU(2) component and the ga.uge field
A~ with the conventional norrnallzation for thls problem:
(20.2)

and B~" the curvature corresponding to the U(1) component and the ga.uge field B~:
(20.3)

As in Subsection 19.6.2, the scalar fie!d </> can be written as a 2 x 2 complex ma.trix
formíng a pair of SU(2) doublets (equation (19.74)), but which has in addition a. charge
g'/2 for U(l) transformations. The scalar field action ia then:

S=o1. f d 4 x [tr (ô"</>t- ~ig'B,.T3</>l + ~ig<P'A" · T}

(&p</J+ 1i91 B~<Ta</>- !ig</JA0 ·1") +V(</>)],
in wbich </> is a complex ma.trix of the form (19.74) and the potential V htlS the SWie
formas in action (19.71). However in this case it is more coJJVenient to pa.ra.rnetriu </>in
terms of a complex vector <p forrning a SU(2) doublet. The scala.r action then tnkes the
forro:

s..,..~. ~f d•, {[a."'t ~i"'t (g' B" + 9A" . .-)](a.. "'+ ~i (g'B~ +gA,. . .-) cp}
+V (cp)} (20.4)

with
(20.5)

2lJ.l. J The Higgs mech.anism

We now assume tha.t the potential is such tha.t at tree order the field <p has a non-zero
expectatíon value proportional to the vector (0, 1):

(20.6) ::
~.'

\Vitb this {orm of the expectation va.Iue, the subgroup of SU(2) x U{l) which conesponds ~::
to the transformations:
{20.7) ~~

leaves the expectation value of <p unchanged, a.nd therefore, 35 a.nnounced, SU(2) x U(l)
is broken dovm to U(l). Reple.cing <p by íts expectation value in action (20.4) we read
off the mose terms for thE gauge fielrls at tree order:

ltl2 f d':r [ (g'B,.- gA~3lf +g2 ~A~'l + iA}?'ln.

It follows that the two components AV·21 hsve the common mass mw:
mw = ~gv.
20.1 The Standarn ModeL Anomalies 461

The linear combination g' B.,- gAi:) is aJso ma8Sive while the ortbogonal oombination
remaíns massless and thus represents the pboton. Let us define the weak a.ngle Dw:

g'fg = tanew. (20.10)

Tbe photon field A,. then corresponds to:

whíle the massive field z" ts:

(20.12)

mz !v v'gl + g" . (20.13)

The components A~ • 21 IU"e coupled to A. and correspond to a cha.rged vector field which
1

is usually written in complex notatíon:

(20.14)

From the coupling of the charged vector bosons with the pboton, M! derive the relatíon
between the electrk charge e and the couplíng constants g and g':

e= gg'fy'g2 + gt'l = !}Sinliw = g' em tíw (20.15)

. 20.1.2 Coupling ta leptons

We here consider only the elactron a.nd the corresponding neutrino sínce the couplings
to other 1eptons (J.<, v,. a.ud T, vT) have exactly the same structure. The neutrino and
the left-handed pa.rt of tbe electron are grouped in a left-handed doub]et L of SU(2):

(20.16)

where fL =~(I - '15) .-' while the right-handed pa.rt of the electron R ~ (1 + 'l>l .,-
" SU(2) singlet. In addition L and R have dilferent charges with respect to U(1).
lepton contribution to the action is:

ig' B~) R+ ift~ (ô~ ~ig' B,. + ~igAP ·..-)L). {20.17)

this model, sínce the left-handed and right-banded components of the fermion fields
treated diJferently, the breaking of the parity symmetry is explicit. Note that the
symmetry forbids a mass term for the electron fie!d. On tbe other hand a coupling
the scalar field a.nd the fermions is a.llowed by the symmetry:

(20.18)
462 The Standará Modd. Anomalies 20.1

!f we replace the <p-field by its expecta.tion value, we see that the spontaneous breaking of
the SU(2) x U(l) symmetry generates the ele<:tron mass m, whicll is therefore talculable,
but, in the absence of new dynamical principie, in terrns of another arbítrary para.meter,
the Yukawa couplíng constant G,:

ml'l ;;vG~. (20.19)

Note that this implies that the Yukawa. coupling of leptons ís proportional to theii mM.'<
and thercfore the perturbatíve approxlmatíon becomes worse for heavíer leptons.
The coupling constant G (the Fermi consta.nt), characteristic o[ the strengtb of weak
interactions, ís defined in terms of an effective low energy current-<:urrent a.nd thus four-
fermion interaction;
G
.j'l f t
d 4 x J• ( x)J.(x). (20.2ü)

The contributíon to the tbarged current J" coming from the electron nnd the neutrino
has the form:
(20.21)
The relation betwren G &nd the coupling constnnts g and g' is obtajned by taking the
large W-mBSS limit of the tree levei electron neutrino scattering amplitude. The re;uli
can be obtaínad by integra.ting <Mlr the vector fields A~1 ' 2 l, taking only into account the
màSS term and neglecting the ldnetic part. The corresponding part of the action is:

(20.22)

Completing squares we immedia.tely obtaln the result of the lntegration:

1 [ - )2] Y~1
2v 2 (.úy,.nL) + Jrr,:roL
2 (- (20.23)

Comparing with delinition (20.20} we conclude:

G/>/2 Ij2v2 = 92 j8m?... (20.24) ~

The phenomenologica.l Fermi model of low energy cllarged wea.k intera.ctions determine!{~
ali para.mcters of the Standa.rd Model but two, for example 8w and the Higgs lield sel1-:~~
coupling À which ha.ve to be deduced from additiona.l experimental results. The recerii:;~
direct measurements of the W and Z masses, for eltll.mple, determine the parameter 9w;;~
Note that a.t leading order the masses can be rewritten: 1;!
I
::;;
i.e.
112
,-a ) 38
Mw

:::f
(-
v2o- =--GeV
sinliw '
and
Mz= Mw.
oosliw
20.1 The S!andard Model• .;lnomalies 463

The present experimental values are mw 80.3 ± 0.2 GeV, mz ::= 91.17 ± 0.02 GeV,
from which one dedures sin2 llw 0.225:1::0.004, a value which ís aJs,;, in good a.greement
with estimates coming from the couplings to neutra! currents {o.n even better a.greement
is obtained when radiative conectlons are ta.keo into oocount). The pa.ra.meter À ma.inly
determines the Higgs ma.ss, o.nd since tbe Higgs partícle bas not yet been observe<!, its
value is nnknown. Let ns fina.lly write the coupling of the charged vector bosons W* to
e-11,. We introduce the delinition (20.14) into equa.tion (20.17) o.nd obt.ain:

~L-r" (T 1 A~'J + T 2 A~l) L= 2 ~ [ti1. -rpeL + eL "YpVL) cw: + w;)

+(i\. 'YpCL ilL'Ypvt.) (W:- w;)]
9
= .,12 (-"L 'YpfL w+ - 'Yp"L w-)
p + fL p • (20.27)

Using definitions (20.1(}-20.12) we can lina.lly write the couplings of fermions to the
neutra! vector fields A,. and Z"'

(20.28)

20.1.3 Gov.pling to ha<1mm

lt is now easy to add interoction terms for qua:rka in the model. ln what follows we
first consider only one generation corresponding tO 2 flavoun;. We cal! u and d the
torresponding quark.s. Each quark ha.s a colour quantum number and forms a SU(3)
ltiplet (see next section). The kft components of the quark.s forma SU(2) doublet QL.

(20,29)

right-ha.nded components form SU(2) singlets. The coupling to quarks can be written:

S,u~ko ~ -f d 4
x [QL(,íil + ~ig'Yt.$ + tig.( ·-r) QL
+Q!R (,í! + !ig'Ymil) QIR + Q2R (lil + tíg'Y2RJ) Q2R]· (20.30)

T3 the eigenvalne of 1'3 the generator of SU (2), Y the U ( 1) charge and q the
charge, we derive from relations (20.11,20.12,20.15) for each fermion:

q = ~(Ta+Y). (20.31)

charge q 1 af the lir:;t component of the fermion being given, a set of colllltraints on
U(l} charges YL, Yí.2R fol!ow:

llJ.=!(YL+l),
(/2=~(YL-l),
(20.32)
q, ~Ym,
(/2 = tl-2R•
464 TM Standaro Model. Anomalies 20.1

We note that a.ll U(l) charges are determined bythe electric charge ofthe first component
of the ferroion doublet. Let us vedfy these relations in the case of the leptons:

(20.33)

The proton being a uud state and the neutron a udd state, the charges of tbe u and d
qua.rks a.re 2/3 and -1/3 respectively. Relations (20.31) then iroply:

(20.34)

Note that the charges of the quarks are compatible wíth the SU(2) doublet assignment.
We shall verify in Seetion 20.5 that the SU(3) triplet structure of quarks leads to thc
cancellation of the possible anomaly due to the chiral coupling of gauge fields to fermions
in ea.ch generation and therefore securas the consistency of the gauge theory of weak
electromagnetic interactions.
Gouplings ta the Higgs field and quark musses. AI; in the case of leptons dírect quark
mass tenns are forbidden by the SU (2) synunetry. The quark mallses are induced by
the coupling to the Higgs scala.r field and the spontBIIeoUS symmetry breaking. The ·"'
SU(2) x U(l) invariBIJt Yuka.wa oouplings are:

(20.35) ·':;

which can therefore provide masses for the two quarks. This is at least the situation foro:i'·
one generation. However three generations have been discovered (see table 20.1), the last :fi
quark, the "top", wh06e ma.ss had been predicted from the analysis of LEP data, h!!S ·.•;.
recently been coofirm,;d at Fermilab. Note moreover that from the Z decays one infm·:J'~
that the number of generations with "light" neutrinos (i.e. with a mass below 45 GeV) -~Jjg
is exac:tly three. ;<':ri
In the interactions (20.35) the spinors which appear on the right Blld the left then need~
not be the sarne. Therefore when we replace the sca.lar field <p by its expectation valuec;r,;
we obtain a general non-diagonal mass matrb< of the form: . ·~
.'fi;

Q~RMooQ~L + QfLM1 0 Q~R• (20.36)i

for the quarks of charge 2(3, a.nd a similar one for the charge -1/3 quarks. Perfonniot~
independent unitary transformatioru; UR.L On the ríght and Jeft qua.rk oomponents Ít Ui~
poosible to replace the matrix M by a real diagonal matrb< M: .. O;~
·-~

U1MUL = M, (20.37~~
ln this representatíon the quarks are mass e;g.,nstates. However the weak interoctio,.,.
no longer haw the simple form (20.30) because the unitary t.ransformations on the quar~;. ·
components Q 1L and Q 2L are in general different. It is custoroaty to put the blame ont.ó
the charge -1/3 quarks. The mismatcli ís exprem>ed in terms o! a S x 3 unita.ry matr • ·
(because there ru-e three generations at present), the Kobayashi-Maskawa. matrix (KM~~-'
which relates the quark mass eigeru;ta.tes d, s and b to the qua.rks appe;uing m the weoll
interadions: ·:
42 j woak in<. Ua~ [Q;] rt>o.. ••••""'· •
l (20.aSt
20.1 The Standard ModeL f4nomalies 465

Thble 20.1
The t.hree genEmlio-n.s., qua.r.b o.nd úpton.a.

charge -1/3 qu...-ks chu:rge -1 leptons neutrinoo

d, m = 5 to 15 Me V e, m =0.511 MeV v~, JJ1
< 7 eV
s, m 100 to 300 MeV IJ, m ~ 105.6 MeV "•• m < 0.27 Me V
h, m = 4.1 to 4.5 GeV T, m = ).777 GeV v., m < 31 MeV

Witb only two generations (d and s) it ís possible to cast tbe matrix into the form:

U _ ( cooO, sinO•)
c~ -sin .c e cosBc '
ín whích IJ. is the celebrated Cabíbbo angle, after unobserva.ble changes of tbe relati""
phases between the quarks. In the presence of the third b quark, one can v.:rify that the
. 3 x 3 KM matrix car> be parametrlzed in terms of tbree rotatíon angles and one phase
parameter. This phase seems to be responsible for tbe observed weak CP violation in
lawn decay.

J!O.LI The prob!em o! elementary scalar field.

The Higgs model has scala.r fields as an essentía.l ingradient. This is the SOlUce of severa!
J>tOhlems. The HiggS field is responsible for the masses of ali pa.rticles, but in the Standard
· Model these masses a.re ali given in terms of free parameters. For example arhitrary
Yukawa couplings determine the fermíons massas. Of course this could be expected
: lrom an effective low energy theory. Still puzzling is the diversity oi these muplings.
lf the couplings were "naturol", i.e. of arder unity, ali fermion massas would be in the
100 GeV range, as tbe W and Z masses or the Higgs expectation m\ue. This is
nbviously not the case, exeept for the top whích hM now been discovered wíth a mass in
the range 160-200 GeV. Another mnybe even more fundamental problem ís related to the
field rnass renormalizatíon. Generically the sca.lar bare and renorma.lized masses
to be of the order af the "cut-off" which gives the scale of uew physics. Jt
by a fine tuning of the ocalar bare mass that ane can reduce the renormalized
we shall extensively discuss in the cbapters devoted to criticai phenomena, this
possible in statistical mechanícs where the bare mass represents the temperatura
system. It is tbus possible to adjnst it to líe very dose to the criticai temperature
the correlatíon length (the inverse physical mass in the partide language) diverges.
in pa.rticle physics the pa.rameters are gíven a.nd it is unnatura.l for the ocalar
mass to lie acddentally dose to such a criticai mine. One ís thus led to conjecture
the scale of new physics CEIDilOt be too high (1 to 10 'ThV?). Moreover at tbís scale
problem mnst be cured. Two type of scbemes have been proposed so far:
Higgs bason is a bound state of a new type of fermions. Thís requir"" a spocífic
hopefu!ly not involving new sealars aga.in. Models in which the forces are a.gain
gauge ínteraetions bave bcen proposed e.nd fali under the na.me of technicolour
section). No completely convindng model h as been constructed so far.
The Higgs boson remains associàted witli. a fundamental field, but the mass renor-
problem is solved with tbe help of :rupersymmetry. Since ferrníons, due to
sym;c-~•:-;·, c...n be ;,atllrally mBSSiess, tbe idea ís to use supersymmetry to relate
466 The Standard MotkL Anomali"s 20.2

them to scalars. In such models the scalar mMs renormalization grows only logarlth·
mically with the cut-off (it would be absent in the absence of supersymmetry breaklng)
and thus the problem is much les:; severe even if the cut--off is of the order of the Planck
mass. The main difficulty with tlús approa.ch ís that none of the existing particles can be
grouped into sUpérmultiplets. The superpartners of ali the exlsting pa.rticles have still to
be found. Moroover tlJe mechanism of spontaneous supersynunetry brenkíng is not fully
understood.

20.2 Quantum Chromodynamics (QCD)

We now concentrate on tbe so-called strong interactíons wllich bind the quarks together
to produce the observed hadrons, neglecting completely the wenk a.nd electromagnetíc in-
teractions which we described in the preceding section. The model of Strong lnteractions,
as it stands at present, consists in a set of quarks characteri:~.ed by a flavour quanturn
number, relevant for Weak lnteractions, which are also triplets of a SU(3) gauge sym-
metry, the colour, rea!ized in the symmetrk phase, and which therefore interact through
the corresponding gauge fields (called gluons):

(20.39)

The most importlll'lt physical arguments in favour of such ll model a:re:

i) Quarks behave almost líke free particles at short distanceo, "" is seen in deep inelastic
scattering experiments or as indicated by the spectrum of bound states of heavy quar)<s.
We shall calculate below the RG J)-function and show thllt a pure non-abelian gauge
thoory is asymptotically free ( AF) at large momentum in four dimensions (like the non-
línear a-model in two dimensions). In Chapter 34 we shall prove that this property :~
survives the inclusion of a limited nurnber of fermions a.nd furthermore that only non· :~
abeli an gauge theories have this property. · ~j
ií) No free quarks have ever been observed at la.rge distance. This is consistent with ~
the simplest pict~;re in which the J)-function (which, due to AF, is negatíve at small -
coupling) remains negative for ali couplings in such a way t.hat the elfective coupling -~
constant grows without bounds at large distances. Numerical simulations support this .;'f
conjecture, called the confinement hypothesís (see Chapter 33). ·{i
20. 2.1 The R G eqooticms in the rovariant gauge -~
We first discuss the gauge dependence of the RG equations and functions of pure gauge ~~
theoríes in the oovariant gauge (19.45), í.e. t.he dependence on the parameter {. A short~
discussion of the abelian case can be found in Sectíon 18.8. We call ZA the gauge fielcÍ't!
::~...~.... _ ~, z. "· : : . : - · _".h< ~. ,.... '"'"'"' I
a in Strong lntera.ctions is in general denoted as to distinguish it from it.s QED anal~~?­
ln this section no confusion is po!!Sible. lt will be shown in Subsection 21.6.2 that for,
such gauges, as in the abelian case, the gauge lixíng term iB not renormalized. Therefore;

f.a = {ZA (Z9 . ( 20.40) ..

20.2 The Sto.ndord Model. ..Anomalies 467

Calling 1J. the renonnalízation oca.le, the RG equations for the gauge field n-point function
read:

a +J3{a,Ç} élaa + r5(a,e}Çae-

[1J ÔIJ. a 2'7A(a,{)
n ]
I'{")(l',a,Ç) ~ 0, (20.41)

where
ó(a,~) =C 1 IJ: I
P eto.!o fixed
{
"'j."l(a,eJ- '7A(a,Ç). (20.42)

We prove in Section 21.7 that the bare correlation functions o f gauge invariant operators
are gauge independent. Tlris in particular implies tbat they are independent of {o. The
same property applies to the renorma.lization constants needed to render these correlation
functions finite. It iB thus possible to construct renormalized correlation functions which
are also { 0 independent. Let us cal) I' such a correla.tion function. lt satislies:

~~ no, coL-oH fixOO r=(~+ p(a,{) !) I'(JJ,a,Ç) =O,

'3. 'I>
(20.43)

with:
p(o,() = ~ai . (20.44)
e a-o, cut--oft' fixed
r also satisfies a RG equation which we assume to be homogeneous:

(20.45)

Usiog equation (20.43) to elimínaxe â/â{, we obtain a new RG equation for r:

o, (20.46)

(20.47)
Wrlting then the cornpatibílíty condition between the two linear equations (20.43) and
(20.46) we obtain two equations:

(20.48)

(20.49)

first equatíon expresses tha.t, a.s expeeted, the multiplicative renormalization of ris
lndlep•md'ent of ~0 • The second equation shows that the ?.eroes of jj are gauge independent.
J)i:lferem;iat.jng the equatíon with respect to a one finds a.Jso that the slope of {3 at it.s
is ga.uge independent. Finally let us show tha.t in a mínima! subtrsction scheme
p vanisbes. In dimensional regulariza.t.ion the relation between ao and a
the form:

{20.50)
468 The Standani Mudei. Anomalies 20.2

The importa.nt point is that the tenn without pole in t in the. expansion of Z0 is ~
índependent. Using the delinition (20.44) of p we then find:

1
BZI 1
O=p ( 1+--"-+--· ) +a (âZ
-- 9 1
-+0 ( -;;
1 )) . (20.51)
aa€ {}(. < €"

Therefore the e>!pa.nsion of p for c small hM only singular contributíons. Since p ís of

course fmite, ali singular contributíons must cancel nnd thus p vanishes identícally. Jt.
follows that in the minimal subtraction scheme the fl-.function and 'fiJ' are independent
of ~-

20.2.2 The RG {J-function at one-loop order

We shall now calculate the RG {J-function at leading arder in a gauge theory correspond-
ing to a simple group G, to verífy asyrnptotic freedom. No simple explnnation ha.s yet
bffin proposed which allows one to understand the sign without performing the explicít
calculation.
The calcu!ation can be dane by various methods, for example we could use the back-
ground field method 1'15 we have done in the case of the models on homogeneous spaces in
Section 15.6.2. Here ínstead we calcula te direct!y in a conventional manner the IJ-function
from the renormalization of the gauge ooupling const!lllt as defined by the fennion-gauge
field vertex. We need thus t.he dívergent pruts of tbe gauge field and ferrnion 2-poínt
functions a.nd the fermion gauge field 3-point functiou.
The gauge fie/d 2-poínt function. Four dia,grarns contribute to the 2-point function
corresponding to the gauge field loops, the Faddeev-Popov ghost loop and the fennion
Joops (see figure 20.1).

--0-Q (a) (b) (c)

Fíg. 20.1 The go.uge field propagator (dotted lines repre.ent ghosts).

The díagrarn (b) corresponding to the self-contn>ction of the gauge 4-poínt vertex ·'~
vanishes in dimensional regularization. The fermion loop contribution ( d) has already i~
been calculated in Section 18.9 up to a sirnple geometrical factor. Only 2 diagramo ·:!'
remain. For tbe vertices and the propa,gator we use the normalizatiollll given in Section J
19.5. We use the Feynma.n gauge and dimensional regulazization. Diagram {a) ha.s tbe ;~
expression .;~
.~
(20.52) ;<j
~!
~ 1
N,..{k,q) ó,.. (5k 2 +2k·q+ 2q2 ) + kl'k.(d-6)+(q"k. +q"k,.)(2d- 3)
+~·~-~ ~~~~
;i
20.2 The Standn.rd ModeL . An<>maliu 469

'Ib calculate the diagrams, we project the integra.Dd over lJ~· and k,.k. (see equations
(9.31 ,9.32)), and use repeatedly the identity:

Zk·q = (k+q) 2 k 2 -q'.

Weset:
/.cO./bcd = C(G)ó.o (20.54)
A short calculation yields the divergent part'

(a}div 6 C(G) (19k2 6 (20.55)

•• 12 ...

Díagram (c) is given hy

(20.56)

The divergent part i.s

(20.57)

Note that both dívergent contributions are not :;epa.rately transverse. By adding the two
contributions we get the divergent part of the 2·point funotic.n in the absenoe of fermions,
which ís now transverse as expected:

(20.58)

CalUng Z A and Z9 the renormali.sation consrants of the gauge field and the coupling
constant g 2 we ohtain the r<>latlon at on.,..Joop order:

(20.59)

U5Íilg result {18.117), we can add the fermion contribution and fina.lly find:

zA
Zg
= 1+ (~c(G)
3
-~T(R))
3
L
8,.-2E'
{20.60)

where the fermions belong to the representation R and T(R) is the tr""" of the square
··. of the generators in ttus representation:

(20.61)

Fig. 20.2
~--~--~
One-!oop contributioo to the fermion 2-point function.
 The Standard Modet Anomalies 20.2
470

The fermíon 2-point funct.ion. One diagram contributes to the ferrnion 2-point function
rl2l (see figure 20.2) wbich diJiers from its QED counterpart only by a geometrical factor:

(20.62)

(for simplidty we have given the same mass m to ali fermions since it does not affect tbe
result). Since we need only the field renormalízation we can project the integrand over
j{. The following ídentity is useful:

Calculating the divergent part of the integral, we obta.in the fermion field renormalization
Zp at one--loop:
Zp =1 (20J)3)

with:
(20.64)

The gcmge field fermion veTtex. Two diagrams contribute at one-loop arder (see figure
20.3), the first has a QED counterpart, the second being characteristic of a non-abelian
theory.

-<1 (b)
Fig. 20.3 The gauge field-lennion vertex at one-loop.

(20.65)

To calculate the divergent part of the integral we multiply by -,~ and take the trace. We
use also the identíty:

We then lind:
(a)di•. = (C(R) -1C(G))t"'Y~s~
,.. € ,
(20.66)

(20.67)
(b)

with (equation (19.57)):

(20.68)
Vuvo(k, q, r) "" (r- q)" 6vp + (k r). 6p~ + (q k)"6~".
 471
20.3 The Standnrd Moáel. ,Anomlllí""

The divergent part is:

(20.69)

lt follows that:
{20.70)

and therefore {equatian {20.63)):

g•
ZA = 1- 2G{G} g,.2, (20.71)

Using finally the reoult (20.60) we <Jbtain Zg:

11 4
-T(R) ) - ~ (20.72)
-C(G)-
( 3 3 s...•r.
The P'-functíon at one-loop order follows:

2
p(g) = -~
(dln (g
dg•
2
Z,)1- 1
11 4
- [ -G(G)--T(R)
3 3
] -+O(g)
g
8.r2
6
4

.
(20.73)

In t he case of the SU (N) group with NF fermions in the fundamental representation the
values of C(G) and T(R) are:

C(G)=N,

and tbereiore: 4
a(g2) = _ (11 N
jJ
2NF)
""'3" g +o (g6)
a,.• (20.74)
3
The theory is asymptotica!ly free, i.e. the !9-function is negative for srnall coupling for:

Nv < UN/2, (20.75)

which, in the case of SU(3}, means et moot 16 flavours. If this condition is met, g Ois
a UV fixed poínt (for details see Chapter 34).

20.3 The Abelian Anomaly

We have pointed out sereral times that none of the standa:rd regularizatíon methods can
deal with one-loop díagrarns in the case of chiral gauge fields. Let us now show that
inde<rl one can find exarnples in which it is impossible to satiBfy WT identities. These
exrunples are physically ímportant in two CIW'lS, the electromagnetic deca.y of the "o
meson, a.nd the theory of weak electroma.gnetic interactions. We first discuss the abelian
case in kmr dimensions and then the general non-abelian case.
The only case in which anomalies may occur in gauge tbeories corresponds to one-loop
diagrams in\/Olving fermions ooupled to cbiral ga.uge lields. This reduces the problem to
the calculation of the determinant resultíng from the fermíon integration. In the abelian
case we have only t<J deal with

D = det (/i!+ ie$. + hslt).

472 The Standard ModeL Anomalies 20.3

We can find regularizations which preserve ga.uge invariance assocíated with tbe vector
field A,. and, since the fermions are massless, invaria.nce under space-independent chlral
transformatíons. We ha.ve to examine the behaviour of the determínant under chir o.!
gauge tra.nsformations. We have only to consider the divergent contributions. In four
dimensions they correspond to the terms of degrees 2,3 and 4 in the gauge fields.
Since the axial field B~ is a gauge lield for chiral transformations, a related problem
is the behaviour of the functional integral:

(20.76)

in which S ( 1/J, ;j,) is the QED oc:tion for massless f<lrmion fields:

(20.77)

under a spac~.>-dependent chiral transformation:

(20.78)

The correspanding current J~ (x) defined in terms of the variation of the action is:

(20.79)

Sínce the determinant of ei~• 0 is 1 we wou\d formally conclude that Z (A~):

Z (A,.) det (/I+ íe.j() =

detlJI, (20.80}

ls invariant and therefore that the current J;(:c) is eonserved. This is a conclusion we
would like to cheek by ~plidt calcu!ation. We have to examine the expectation value
of â,..J;(:c) in the case of the action (20.77). Let us note that after the transformations
(20.78), Z (Ap) takes the form:

(20.81)

For any regularization which is consistent with the hermiticity of 1• wc find that

is invariant. Therefore we expect that an anomaly can only appear in the ímaginiU}' part
oflnZ.
The operat.or &pJ!(x) has dimenskm 4 and s!nce a possible anomaly ís a large mo-
mentum, i.e. local effect, ( &,.J!(:c)) can only be a local function of Ap of dimension 4.
In addition parity implies that it is proportional to the completely antisymmetric tensor
<pupo· Thís determines (â~J!(x}) up to multiplicative consta.nt,
20.3 The Standard Mode!. Anomalia 473

,.,, ..-Pt,J.!

k.~ •• -
0 (a.)
....

'P:hv

Fig. 20.4 Anome.lous diograms.

To find the constant it is thus sufficient to calculate the coefficient of degree 2 in the
expansion of (â.>.J~(xl} in powers of A. Let us define the 3-point function:

(20.82)

r<3 l is the sum of the two corresponding Feynman diagrams of figure 20.4.
We note that r(') is formally a complete!y symrnetric function of the three sets of
externa! arguments. lt has a lineM divergence. Since the fermions are massless, the
finite part of r<•l is ambiguous by a linear combination of the vectors Pl and P2 whích,
for symmetry reasons, has the form:

(20.83)

Note that such a term breaks the symmetry between the three externa! arguments. In
addition the WT identity which fol!ows froJJ;! gauge inva.riance:

(20.84)

fixes alw oompletely the coefficient K. Therefore, gauge invariant regularitations yield a
finite result a.nd by calculating k~[~~. (k;p 1 ,p,) we can check the current conseiVation.
In particular a symrnetrk definítion of rm
in the three externa! argument.s is either com-
pletely divergence free or brea.ks both gauge invaria.nce and axial current conservation.

20.3.1 Formal verificatum of cun-enl conservation

To understand from the point of view of Feynman diagt'ams where the current conset-
vation oomes from, let us compute the cnntribution to the divergence comíng from the
diagram (a) (figure 20.4):

(a),_. {2:)• tr[1'sli j d4q(fl+ll)-11'"(f/-p2)-I1•11'-']. (20.85)

lf the integral 1\'ef€ convergent we would proceed in the following way: We would first
transform the e:xpressioo, using the identity:

~<!l+fT = t-i!i+W'·
1
(20.86)
474 The Standard Model. Anomolies 20.3

For the contrlbution coming from the second tenn in the r.h.s. we then use the cyc!íc
property of the trace and the anticommt~tation of r/ and ')'s to cancel the vropagator r/- 1 .
lt follow• that:

(20.87)

The contribution comíng from diagram (b) is obtained by intercllanging (p 1 , p) and

(p2, 11). It is not too difficult to verify by shifting va.riables in the integrais that tbe
contributions coming from (a) and (b) cancel (the twQ terms in the r.h.s. of equation
{20.87) have to be shlfted differently).
However since tbe integrais diverge all these manipulations are not neces.sa.rUy legiti-
mate. If nn anomaly is generated, since it is a short distance effect, it must come from
a local term, i.e. here from a linear term in momenta of the form (20.83). We could
at this point do the calculat.inn using one of the ,...nous gauge iuvariant regula.ri2ations,
for example Pau li- Villa.rs's regula.rization or dimensional regula.rization with I'> being
defined as in dinrension 4 and thus no longer anticommuting with other ')' matrjces (see
Subsection 9.3.3). However, it is more convenient t.o use a regularization whícb respects
the symmetty between external a.rguments but breaks gauge invariance, replacing in the
fermion propagator:
w-~ ~ tril-' .,-•q'.

By calculatíng k~ri~. (k;p 1 ,p2 ) we shall check both current conservation and gauge
inva.rianoo.

20. 3, 2 The calcu!ation

In order to indicat.e how the method generalizes to arbitrary even dímensjons it ís con-
venient to consider directly the contrihution c< 2l(k) of arder A 2 to (kÀJ~(kJ):

c! 2l(k) == e 2 f d 4p1 d 4P2 AP (p 1) A, (p,} f (~:~. e-•)(q+O}'+(o-1'2)'+•')

x tr [/'S~ {r/+~)-' ')'p (r/- Ji•l_, -r.r']· (20.88)

We now use identity ( 20.86) to separate this contribution into the suw of two terms:

In tbe fust contribution we interchange (p, p) and (P>. ")· We then find:

c<2l(k) = e2 f d 4p 1 d 4 p,A,. Ú>J) A., (p,) f (~;J4 e-•[(o+kJ'+<')

tr { 15 [e-•(q-p,)' (fi+ ll)- 1 ')'v (r/~ Ji,)- 1 ')'~- e-•(o-...l' r''l',. (t'- Ji·)- 1 ')'.]} (20.90)

Finally stiii in the first term we change q into q + p 1 • Using the cyclic property of the
trace and the anticommutation properties of ")'!, we then '"*' that the twQ terms wou)d
 20.3 The Standard ModeL. Anomaiiea 475

cancel in the absence of regulators. Now instead we find a non-vanishing •um due to the
exponential factors which are dlffetent:

çl>)(k) e2 f 4 4
d pl d P2 Ap {p1 ) A. IP2) f (::). e-•[(q-,.)'+o']

x tr [1'5fi 1-yp (rf- ,12 )- 1 -y.J (e-•(o+pl)' _ e-•(o+t)') , (20.91)

After evaluation of the trace (notite our convention (A5.2) for -y5 ) this sum berome5c

ç!>)(k) 4e
2
f d 4p 1 d 4 P2A~ (piJ A. (P2) f ~~;)4 e-•[(q-,.,)'+o'j
X fp.llp~ 7'"2p'lo 2 (e-E(I'J+Pl)l ~ e-e(q+k}2), (20.92)
q2 (q- 1'2)

The relevant va.lue5 of q are of arder ,-t/>, We can therdore replace lhe integrand in
the expression by:

(20.93)

The identity:
f d qq~qfjf (q
4 2
) = ~Óafj f d4 qq2 f (q2 ),
transforms the integral into:

(20.94)

Since:

we finally obta.in:

(20.95)

From the definition (20.82) it follows that:

(20.96)

· This non-van.i.shlng result teUs us that any definltion of the determina.nt detl<} breaks
· a.t least current conservation or gauge invaríance. Since gauge invariance is essentia.l t,o
the construction of QED we chooBe of course to break current conservation. Exchanging
argumenta, we obtain the value of p 1 ,.r}.~., (k;PhP2):

(20.97)
 'The Standard Model. AnomalifS 20.3
476

If instead we had usad a gauge invariant regularizatlon, the result for r(•J would have dif-
ferad by e. term of the form (20.83) and Pt,ri~. (k;pJ,Pz) wou!d have vanished. Imposing
3
gauge invaríance we can find the additional contribution 6f( ):
óf~~. (k;p1,1'2) K{).~up (p,- pz)p. (20.98)

Since:
(20.99)

welind:
K = e2 / (61r2 ) (20.100)

This gives an additional contribution to k~ri~. (k;p~oPz):

(a) ez
k).óf ÀI'V (k;p), Pz) = J1r' <l'),paJ1JpPl6, (20.101)

Ther.,fore ín a QED-like ge.uge invariant field theory with massless fermions we fmd a
non-vanishing chiral anomaly:

k,f~~. (k;p, ,Pz) u~:2 '= ·~·••Plp/>l• .

2
:) (20.102)

Note that the anomaly is ge.uge ínvariant as expected.

Equation (20.102) can be rewritten after Fourier transformation as an axial current
non-ronservation equation:
(20.103)

ín whicb F,.,. is the electromagnetic tensor. Note that slnce global chiral symmetry is
not broken, the integral owr the whole space of the anomalous term must vanish. This
condition is indeed verified since the anomaly can immediately be written as a total
derivative:
E~vpoFJlvFpa = 4E,.upa8~ (t:_."'P,..,A~âpAc) _ (20.104)

Equation (20.103) also implies:

(20.105)

Note that since the chiral group is e.beliM, this equation which has been derived at fust
order in 9(x) is actually exact.
In the form we have presented it the calculation generalizes witllout difficulty to hlgher
even dimensions 2n. Note simply that the permutation (P~o I') +-+ (pz, v) in the first term
of equation (20.89) is rephu:ed by a cyclic permutation. The result for the anomaly is:

(20.106)

Chirnl gaug• thenrjJ. Current oonservation becomes a WT identlty in a gauge theory.

The anomaly prevents the construction of a theory which would have both an abelian
gauge vect.or and wdal symmetry. Furthermore the analysis of the B B B correlation
function is identical and yields e.lso an anomaly which cannot be removed slnce the
correlation function is by definition symmetric in its three arguments. This prevents the
construction of a pure wdal gauge symmetry.
20.3 The Standarri Modef.. Anomalies 477

20.3.9 Genemlization to non-abelian vector gauge theoríes

We still consider a unique axial curtent but now ínside a non-abelian gauge theory. The
fermíon fields transform non-trivially under a gauge group G andA,. is the corresponding
ga\lge fie!d. The action is:

(20.107)

with:
(20.108)
No new calculation is needed; the result is complet.ely fixed by the previous calcula.tion,
dimensionai a.nalysis a.nd gauge invariance:

(20.109}

in whlch F'"" is now the corresponding curvature. Again this e:xpression •hould be a total
derivative. lndeed:

(20.110)

20.9.4 Anomaly and eigenvalues ofthe Dimc opemtor

Let us assume tbat the spectr\lm of VI, the Dirac operator in a non-a.belian gauge field
(equation (20.108)), is discrete and let us call dn and \On(x) tlle corresponding eigenvalues
and eigenvect.ors:
(20.lll}
For a unita.ry or orthogonal group the Dírac operat.or is antihermitian; therefore the
eígenvalues are írnaginary and the eigenvectorn orthogonal. In addítion they satisfy:

(:WJ12)

Therefore either d, is diHerent from zero, ancl1's4"n is an eígenvector of VI with eígenvalue

-d,, or dq van~. In the !atter case the eigen~ors of VI can be cl10sen eigenvectors
of defmed chírality, i.e. with eigenvalue ± of 15:

Let us now calculate the variatíon lin at first order of tbe eigenvalues d, +m of the
operator VI + m in a chíral transformation (20. 78):

VI+ m,......... VI+ m + i)'sO(x)(:jZ) + m) + i(VI + mhs9(x).

Perturbation thoory tells us:

én =i f d x;o~(x)
4 + +{VI+
h•O(:r)(VI m) m)'J'59(x)j'Pn(x),

2i (d,+ f d xO(x)rp~(:rho;on(x),
m) 4
(20.113)
478 The Standard Model. Anomalies 20.4

where we h ave assumed that the eigenvectors are orthonormal.

lf we now take the límit of con.stant 9 we find in the r.h.s. the scala:r product of
eigenvectors of 121 which vanishes except when à, = O. Calllng n+ and n_ the numbers
of e.igenvalue zero of chira!ity + and - respectively, we can theu write:

lndet[(e;-,,e(JI)+m)eh• 9 )(J?.I+m)- 1 J I](l+ d,.õ~m)

2i(n+-n-)8. (20.114)

Taking the limit m = O we can relate the r.h.s. to the space integral of the chiral anomaly
(20.109):
(20.115)

This result has severa! consequenres:

(i) Equation (20.110) shows that the integral of the o.nomaly over the whole space
depends only on the behaviour at large dístances of the curvature tensor F~·. The
property that, in addition, the integral is quantized shows that the form of the anomaly
is related to topological properties of the gauge field since the integral cannot change
when the gauge field is varied continuously.
(ii) We shall see in Section 40.5 that gauge field configurations exist for whích the r.h.s.
of equation (20.115) does not vanish, for exrunple instantons. We have shown above that
if massless fermions are coupled to such gauge fidds the detenninaot resulting from the
fermion integration necessaríly vanishes. Thffi has some physical implications which are
examined in Sections 20.5 and 40.5.
(iií) Note also thllt the l.h.s. of equation (20.114) is ob,iously l when 1f mr. Also
one realizes that the r .h.s. is the exact result although it has only boon derived for e
small, due to the group properties of the abelie.n clliral group. Therefore we shou!d bave
expected that the integral of the anomaly yields an even integer.
(iv) One IDight be surprísed by the result that detCI is not invariant under global chiral
transformations. However we havejust established that when the integral of the anomaly
does not vanish, detlZI VMishes. This explains that to giV':l a meaning to the r.h.s. of
equation (20.114) we have boon forced to introduce a mass term. In tbe limit m--. O,
we calculate the detl!) in the subspace orthogonal to eigenV':lctors with zero eigenvalue.
When n+ I n_ t!riB subspace ;, not clüral invariant.

20.4 Non-Abelian Anomaly

:w.4.1 General axial current
Let us first consider the problem of the conservation of a general axial current in the case
of a non-abelian vector g11uge theory. We assume that the fermion actlon has a G >< G
chiral symmetry and discuss the conservation of the corresponding axial current. The
generators of the gauge group rnay or may not be related to the diagonal subgroup G of
G x G which correspond to vector currents. We cal I t 0 the generators of G. The current
then has the form:
(20.116)
When the gauge group i;; connected with the chita! group, the current conservatíon
equation involves the gauge covnriant derivative:

(2ü.ll7)
 20.4 The Standnnl Motiel. Anomalies 479

In the calculation of Lhe contrlbutíon to the !Ulomaly quadratic in tbe gauge tields the
only modifica.tion in the prevíous arguments is the appenrance of a dilferent geometrical
factor. Then the complete fonn of the anomaly is dictated by gauge covaria.oce. One
finds:
D).Jf"(x) = - 16;,z'~'"P" trt"F,.,F,.,. (20.[18)
In particular if the gauge group ls disconnected from the chiral group the anomaly is
proportíonal to trt0 and therefore only different from :z.ero for the abelian factors of G.
!!OA.2 Obstruction to gauge ínvariance
We now want to consider a non-abelian gauge field coupled to left or right-banded
fermi<Jns:
(20.11!/)

a(I -1s) is treated in the same way). We can construct a consistent gauge theory only
if the partitíon function
Z (A,.)= f [drJ,>d,b) exp (-S (r/>, ,b)) (20.120)
is gauge invaria.nt.
lf we introduce the generators t 0 of the gauge group in the fermion representation, we
can write the corresponding cunent J,. "'"
J;(x) = -rP!
(l +1~)1,.! 0 \b. (20.121)
The invariance of Z (A~) under an infinitesimal gauge tra.nsformation agaln leads to a
covariant conserwLtion equation for the current:
(D,.Jp} =O.
The calculation of the term of degree two in the ga.uge field of the a.nomaly is straíght-
forward: the regulnrization adopted for the calculatíon in Subooction 20.3.2 is also suited
to the present case since the current-gauge tield 5-point function is symmetríc in the
externai sr guments. The group structure is reflected by a simple geometrical factor. The
global fact.or can be taken from the abelia.n calculation. lt díffers from result (20.95) by
!
a [actor I/2 which comes from the projector (1 + ')'s). The general fonn of the term
of degr"" 3 in the gauge lield can also be easily found, but the calculation of the global
factor is somewhat longer. We shall argue in next section that it can be obtained from
conslstency conditions. The complete expression reads:

(D,.J"(x))" - 24i"2 8,.'""""tr[t" (A.âpA.,+4A.ApA,)J (20.122)

!
lf the projector t(l + 1'>) is replaced by (1 ~ 1'•) the sígn of the anomaly changes .
. Unless this term \'llnishes identica!ly there is an obstruction to the oonstruction of the
g~uge theory. It is easy to verify, taking ínto account the antil;ymmetry of the, tensor,
that the group (setor is:
(20.123)
· For a unitary representation the generators t" are, with ou r conwntions, antihermitia.n.
-~ Therefore the coefficients dal'h fite putely imagínary:
d",p, = fr tr [t" (t~t' + t't")J" -da/3-,. (20.124)
,!:"For ali real (the t" antisymmetric) or "pseudo-real" (to= -S Ttos-1) representations
these coefficients vanish. It fo\lows that the only non-abelian groups wbich can lead to
·''llllomalies in four dimensions are: SU(N) for N ~ 3, S0(6) a.nd Es.
480 The Standard Model. Anomalíes 20.5

20.5 Physical Consequences

Electromagnetú: "o decay. We have eXamíned in Section 13.6 a phenomenological model

for low energy Strong lnteractions bosed on a Jinearly broken SU(2) x SU(2) symmet.ry.
The cmrent non-cunsen~atíon equation for the axial current J~ wos at lee.díng order
(equatian (13.81)):
(20.125)
We here concentra.te on the third component J! of the current which corresponds in the
r.h.s. to the neutral píon 1ro lield. If we introduce electromagnetic interactions in the
model, this relation between the divergence of the axial current and the "o field ailows
us to calculate the electromagnetíc decay rate of the neutra.! pion when the momentum
k of the píon goes to zero. In the absence of anoma.lies, the av..rage of relation (20.125)
multiplied by nvo gauge fields yields ímmediately that the decay rat.e vanishes for lc O in
contradiction with reasonable smoothness assumptions and experimental results. Thking
iDStead into account anoma.ly equation (2Q.103) and a factor 1/2 which cvmes from a
different normalization (13.99) of chiral transformations we find:

(20.126)

Taking a.ga.in the average of this equation with two photon fields and golng to the limlt
k O to e!iminate the l.h.s., one now abtains a non-\'!llllshíng decay amplitude for
an unphysica.l "o at zerO total momentum. Tn the a-model at leading order one can
ex:trapolat.e to lc2 -m;. The result is in excellent Bgre€ment witb experiment. The
thooretica.l rat.e r is given by:

while r••• = (7.37 ± 1.5)eV. A similar ""timate was lirst derived by Steinberger from
direct Feynman graph calculation, before the relation to anoma.lies had been discovered.
Jllote that a similar theoretica.l estimate is obtained in the quo.rk model with massless
quarks for three colours.
The solv.tion of lhe U (I) problem. In à theory of Strang Interactions in which the
quarks are massless nnd interact through a colour gauge group, the actíon has a chiral
U(NF) x U(Np) symmetry, in which Np is the number of flavours. The spontaneous
breaklng of tbe chiral group to its diagonal subgroup U(Np) let us expect N~ Gold-
stone b06!'1ns associated with the axial cuuents. From the previa"" ana.lysis we know
that the axial current corresponding to the U (1) abelian subgroup has an a.ooma.ly. Of
course the WT identlties which imply that the existence of Goldstone bosons corre:spond
to constant group trBllSformations and tberefore involve only the space integral of the
divergence of the cuxrent. Since the anoma.ly is a total derivativa one might have ei"
pected the integral LO vanish. Howevel' W<l shall see in Seetion 40.5 that non-abelian
gauge theories e.dmit instanton solutions which give a periodk structuré to the vacuum.
These instanton solutions correspond LO gauge configurations which approach non-trivl&l
pure gauges at infinlty and give the set of discrete non-vanishing values one expects from
equation (20.115) to the space integral of the anomaly (20.109). This indicates (but no
sat.isfactory calculatíon of the instanton contribution has been performed) that for omall,
bur. non-\"Mishing, quark masses tbe U(l) axial current is fu from being con>Jerved and
20.6 481
The Sto.ndartl ModeL Ano:rmalies

therefore no light would-be Goidstone hoson is generated. This observation resolves .a

long standing puzzle since experimentally no corrcsponding tight pseudoscalar booon "'
found (for details see Section 40.5).
Weak-electromagnetic interactions. The cancellation of the anomaly discussed in Sec-
tion 20.4.2 imposes constraints on the modei of w .e.m. interactions. In tbe Stand8ld
Model, for exa.mple, the anomalous contribution of the leptons has to cancel the quark
contribution. Let us show tha.t within the Stand8ld Model this cancallation occurs within
each generation provided that for each llavour quarks can exil;t in three states. In the
w.e.m. group SU(2) x U(l), SU(2) alone is a safe group. Therefore the problems come
from the U(1) factor. We expect a priori two conditions coming from the vertlces with
one U(l) and two SU(2) gauge fields and with three U(l) ga.uge fields. Actually we shall
.ee that both condítíoru; are equivalent. Ifwe consider two SU(2) and one U1l) gauge
fields, only SU(2) doublets contribute and equation (20.122) leads to the condition:

L YL trT"-r-" =O,
ali -doubiet111

in which YL ís the U(l) charge (see Section 20.1). This condition reduces to:

L
ali double1os
YL=O. (20.127)

The three U (1} gauge field vertex yields the condltion:

L Y[- L YJ=O,
ld't-h!U':Ided puu. J'igtrt..-handed parta

because the cootributions to the anomaly of right-handed and lcft-handed couplings h ave
opposite signs. However in the Standard Model the left and right charges are related
(equation (20.32)). Summing the charges of <me doub]et and the corresponding singlets,
we obt&in:
L 3
(YL+1) +(YL 1}
3
2Y[=O,
aU doubl~l5

a condition identical to equation (20.127).

In one generation the lepton doublet has YL = -1 and the quark YL = 1/3. Therefore
a cancellation requires that the qu8lks exist in three states. These states are provided
by the oolour quontum numb.,-.

20.6 Wess-Zumino Consistency Conditions

In Subsection 20.4.2 we have calculated the part of the anomaly which is quadratic in
the gauge field and asserted th.àt the rema>ning part could be obtained from geometrical
_arguments. Let us now explain the idea. The anomaly is the variation of a functional
· under an infinitesimal gauge transformation. As we argued in Section 15.3, this implies
compatibility conditiom, which here are constraints on the form of the anomaly. One
oomenient way to write these constraints is to expr= the nilpotency oí BRS trans-
formatlons (for details see Chapter 21}. We perform on the gauge field A,. the BRS
.trartsfoJrmation (19.r.J):
óA,.{x) = D,.C(x)ê'. (20.128)
482 The Standard ModeL Anomalú<;, 20.6

The variation of In Z (A") takes the fonn:

ci]nZ(A~) =-f d :t (J,.(:r))D.,C(:t}L

4
(20.129)
lf we write the anomaly equation:
(D.J,.(:t)) .A(A,.,:t), (20.130)
equation (20.129} takes the form:

élnZ (A.,)= f d 4:t.A(A,,x)C(:r)i". (20.131)

Since .AC(:t) is a BRS variation it is invariant under BRS transformations (19.53,20.128),

i.e. transforming the gauge field by (20.128) and the renníon C(x) by:
6C(:r) = -tC2(:t). (20.132)
Conversely in arder to find the form of possible anornalies one has to Iook for local
functionals of A,. and C linear in C and BRS invariant. One verifies that tbis condition
determines the term cubic in A in the r.h.s. of equation (20.122).
Bibliographical Notes

The Standard Model was proposed in its present form by

S. Weinberg Phys. Rev. Letl. 19 (1967) 1264; A. Saiam, Elementary Parlil!le Theory,
N. Svarthohn ed. (Aimqvist and Wiksell, Stockholm 1968).
Earlier attempts include
S.L. Glashow, Nucl. Phys. 22 (1961) 579.
1b explain the suppression of strangencss changing neutra! currents, a fourth quark (c
for charm) was conjectured in
S.L. Glashow, J. Illopoulos and L. Maiani, Phys. &l!. D2 (1970) 1285.
A selection of pape:rs is col!ected in
Gauge Theory of Weo.k and Eledromagnetíc lntemctions, C.H. Lai ed. (World Scien-
tific, Singa.pore 1981 ).
Most references given in Chnpter 19 are still relevant hete. See also
E.S. Abers and B. W. Lee, Phys. &p. 9C (1973) 1; J.C. Taylor, Gavge
Weo.k ln!eroctions (Cambrldge Univ..síty Press, Cambridge 1976).
For the KM matrix see
M. Kobayashi and T. Maskawa, Prog. Theor. Ph!/8. 49 (1973) 652.
Among the texthooks see for example
T.D. Lee, Particle Physics and Introduction to Fíeld Thwrv (Harwood Acadernic,
York 1981).
The problem of anomalies is discU$Sed ui
J.S. Bell and R. Jacklw, Nuovo Cimento A60 (1969) 47; S.L. Adler, Phys. Rev.
(1969) 2426; D.J. Groos and R. Jackiw, Phys. Rev. D6 (1972) 477; H. Georgi and
Glashow, Phvs. Rev. D6 (1972) 429; C. Bouchiat, J. Iliopoulos and Ph. Meyer,
Lett. 38B (1972) 519; E. Witten, Phys. Lett. U7B (1982) 324.
See also the lectures
M. Pesltin, Rreent Advances in Field Theory and Statistical Mechanics, Les
1982, R. Stora =d J.-B. Zuber eds. (North-Holland, Amsterdam 19S4),
and the volume
S.B. 'Ireirnan, R. Jackiw, B. Zumino andE. Witten, Currrnt Algebm and All,om•nllti
(World Sdentific, Singapore 1985) and references tberein.
 RENORMALIZATION OF GAUGE THEORIES: GENERAL
FORMALISM

'Ib díscuss the renormalization of gaug<> theories in the non-abelían case in its full gener-
e.lity, ít is ne<::ess>UY to use a rather abstract formalism, which a.llows one to understand
the algebra.ic structure of the renormali>a.tion procedure without beíng overwhelmed by
the notational complexity. There ls however a prioe to pay: the translation of the general
identities which then appear into usua.I and more concrete not!>tions becomea a oon-trivia.I
exercise.
Ou r stràtegy is the fol!owíng. We quantíze the theory hy províding a dynarniC'l for the
degrees of Ereedom correaponding to gaug<> transfonnations. We fmd, as =pected, that
the quanti.Zed actíon has a BRS symmetry. We then derive the conesponding set of WT
identities for correlation functions. We use these identities to determine the most genera.I
form of the renormallzed action.
Physícal quantities in the theory should be independent of the dynamics of the gauge
group degrees of freedom. Following the ana.Iysis of previous chapters, we first prove
that the va.cuum amplitude does not depend on thls dynarnics, i.e. is gauge independent.
We argue that this property remains true if we add to the origina.I action sourcoo for
gauge inwriant operators of canoniCAl dimensíon low enough, so that the total action
· remains renonna.lizahle. As a consequence, correlation fonctions of these gauge invariant
operators are independent of the gauge futing procedure, a.nd have therefore an intrin•ic
meiWing. A Bimílar property holds for ali gauge invaria.nt operators but the discussion ls
more involved and will not be gíven. Fínally, when a S-matrix can be defined, S-matrix
elements are aJso gauge independent.

21.1 Notations and Group Structure

Nota.tion.s. In what follows, we restrict ourselves to boson fields, the generalization to

fermions being straightforward. All gauge fields and other se alar boson fields are com-
into one vector denoted by A', in which the index i sta.nds for space variables x,
index p, and group índices a, b:

gauge trensformation corresponding to the compact Lie group G i$ written:

liA; D~(A)w", (21.1)

w" are the infinitesimal parameters o! the group transformation. As we have

equation (21.1 ), we reserve greek índices for the adjoínt representntion of the Lie
Summation over repeated indices ls always meant. lt includes integratíon over
The index a aJso includes group índices and space variables. In more
form, equation (21.1) stands for:

6A:(x) =I dy [B;ó(x- y)6: + f~A~(x)5(x -y)j w~(y),

Ó<p0 ' (x) = I dy ó(x y)[tb·]~.\o<' (z)w~· (y),
(21.2)
484 Renmma!ízation of Gauge TheorieJ 21.1

wbere the matrices tb' form a (in general reducible) representation of the Lie algebra of
G and /~ are the corresponding structure constants.
Finally L( A) is the gauge invariant action and thus sàtisfies:

:~;D~(A) =0, (21.3)

A group identíty. We ha"" explalned in Section 15.3 that, as a consequence of the

group structure, quite generally the functionàls D~(A) satisfy equations which can be
reg..rded as compatíbility ronditions for l;he system (21.3) considered as a set of differ-
ential equatíons for C( A):

(21.4)

These equations are formally ídentícal to equations (15.43) of Section 15.3 and (16.41)
of Section 16.4.
Since here the functional D;,(A) is only alline, it is not difficult to verify the identities
(2L4} by a direct calculation, using the commutation relations of the Lie algebra of G.
Note however that because D~(A) is a differential operator, it is necessary to càrefully
keep trock of the ti-functions. In particular since the índices a and i include space coor-
dinates, the structure constants J~p which are proportional to the numerical structure
constants fí:e of the Líe algebra, have a non-trjvial dependence on space variables:

(21.5)

Due to the distríbution ch..racter of the various quantities it i!) actually conveníent to
reWTite tbe identity (21.4) with the heJp of two functions Wf : wf{:t) and W~ : w2(:t)
and the opera to r
u
•(w) w" D~(
; A) óA'
ó (21.6)

The identity (2L4) takes the form

[!:.(w 1 ), l:.(w2 )] = l:.(w 1z), w"{2 = !7,flwfw~, (21.7}

Let us for ínstance indicate here how this identity can be reoovered for the gauge field
part. We wríte !:.(w) in explidt notation:

We tben calculate explicitly the oommutator

Using the antisymmetry of ft, and tbe Jocobi ídentity aue indeed obtains equatíon (21. 7)
in BXp!ícit form, wíth wf2 (x)"' J~wtlx)wHx).
2L2 Renormalization of Ga"?" Theories 485

21.2 Quantization

We h ave shown in Chapters 18 and 19 that the quantization problem can be reformulated
in the following way, which leads to the same results as canoni<:al quanti7-r.tion: Because
the action is gauge invariant, tlw degrees of freedom a.srociated with gauge trlUlSforma-
tíons have no dynruni~. 'Ib quantize tlwse degrees of freedom and to give a roeaníng to
the functional integral in the continuum, it is necessary to provide one. This dynamics
is sornewha.t ar bitrary a.nd we shall have eventually to prove that ''physica.l results" (we
shall expla.in !ater what we mean hy physical results) do not depend on its choice. Gauge
invariance defines classes in the field space, corresponding to fields and ali their gauge
transforms, i.e. orbits of the gauge group. Let us choos,. in some way one representative
B' for each class, i. e. a gauge section, and parametrize fi general Jield A' a.s;

(21.8)

in which Bg denotes the gauge transform of B by thc gauge tmnsformfition corresponding

to the group element g. Since C( A) is gauge invariant,

C(A) C(B),

and r. (B) provides a dynamics for the field B'. For the group element g which is a fie!d
sínce it depends on space we introduce fi stochastíc qua.ntization in the sense of Chapters
16 a.nd 17:
(21.9)
We assume that, at B fu:ed, the set of equa.tions (21.11) determines g uníquely (at least
for B' small so tha.t the e<jl.la.tion can be so]ved perturbatively) "" a. function of v. Tbe
lield v has a distribution dp(v} which we do not need specifying more precisely at thls
stage, but which eventually will be chosen local and gaussían. We denote the llat and
thus gauge invariant measure [dA'] written in terms of B a.nd g by {dBJ [dg] {gauge
ínvariance implíes that the measure em be factorized).
'Th avoid modifying the dynarnics of B' we must multiply the B' integrand by a quantity
independent of B'. We thus use the identity:

f {dg].5{F,(B 9 ) -v0 ]detM 1, (21.10)

with;
_ (íJF) 6F, ( ) ; ( ) (2Lll}
MafJ = íJg ali= 6A' Bg Do B" .
We can then write the generating functional of correlation functíons Z(J):

Z{J) = f jdp (v)] f fdBj[dg]6 [F, (B9 ) - ..,,] det M exp [-C (B} + J;B!)· (21.12}

Since C (B) is gauge ínvariant we can repla.ce B by B 9 • The íntegra.nd then depends only
on the argument B~ A 1 and we can rewrite the generatiDg functional (21.12) as:

Z(J) =f (dp(v)J f {dA)detM (A)ó[F,(A) vu].exp (-i:(A) + l;A']. (2Ll3)

486 Renormolizatíon of Gauge Theorie 8 '2LJ

A!; in Chapter 16 we impose the constraint (21.9) by integrating over an auxiliary field
).o and introduce fermions "ghost" fields to represent the determinant:

Z(J) f jdp(v)J f (dAd).dCdC] exp [-C(A) + >. (vo

0
Fo(A)) +é" MapCf;l
(21.14)

We perfonn the integration over V a, settíng

(21.15)

The final form of the generating functional of the A-field correlation functioru is then:

Z(J) =f jdAjjd>.J [dCdéJ exp [~S (A,C,é,A) + J A'], 1 ('21.16)

with the deftnítions:

(21.17)

21.3 BRS Symmetry

Let us fiTSt note that in the representation (21.13) the functional integral for J O ís
invariant under gauge transformations whkh trarulate Va (see Section 16.1)
M' = D~jM-ljai1,Ujj, (21.18)

and thus
(2Ll9)
This Slavnov-Taylor symmetry WM an essentie.l toa) in the initial proof of the renormal-
ízability of non-a.belian gauge theoríes. However the non-local character of this tranBfor-
mation exple.lns the oomplexity of such a derivation.
lt follows from the genera!Me.lysisof Chapter 16 tha.t the 6ne.l action (21.17) has a BRS
symmetry. Because the field which is stochMtically quantized is the gauge group, the BRS
transforrnation for the field A' has the form of a gauge trBDSformatiou (equation (17.77)).
We have vertfied in Chapter 19 that, since we have provided a dynamics to a group
element, the BRS eymmetry is similar in form to the symmetry of the dynami-c act.ion
of chíral models as described in Sections 16.4,17.7 (equa.tíons (16.46,16.47) and (17.75-
17.77)). Also, in these more abstract notations, the basic equation (21.4) ís formally
identical to the commutation relations in the case of non-linea.r representation of groups
of Section 15.3, and to the relations {16.41) of Section 16.4. The corresponding BRS
tra.nsformations fur the A and C 6elds thus are gíven by the equations (16.39). As
we know from the general analysis of Chapter 16, the tra.nsformatíons of >. and c
are
independent of the dynamics. We conolude that the BRS transformations have the form

(21.20a)

á>."= o.
21.3 Renornullúaticm of Gaupe Theoríes 487

By rewritíng these transformatíons in more explicit notations, we ímmedia.tely verífy that

they are identical 1:0 the BRS transformations derived in Cha.pter 19. For further use we
set:
(21.21)

The BRS opemtar. Th express the BRS symmetry, ít is also useful to introduce the
antícommuting dilíerential operator Vo:

(21.22)

The essential property of the BRS operator is that it has a vanisbing square:

2
!Vo] =O, (21.23)

as we know from general arguments and can again check explicitly here. Hence, apart
from contributions which are gauge inva.riant, BRS symmetric ternn; include contributions
whlch Are BRS exact, i.e. of the form Vo4> (A, C,ê,>.). We verify that the action (21.17)
has exactly such a decomposition:

S ,C( A}+ Vo4>, (21.24)

4> ê" [F.. (A) â~.. l dsw(sÀ)].

We are not surpriaed because this property has been proven quite generally in Chapter
16 (see equation (16.58)). The quantized O<:tion is therefore BRS symrnetric:

t::\,S(A,C,é,>.) =O. (21.25)

It had been generally oonjectured that any local soluti<>n of equation ( 21.25) can be
written in the form {21.24). After sev"ral partia! resulta, a proof now exists. Such
a property plays, as we shall see !ater, a fundamental role in the proof of the gauge
lndependence of physical quantities. We shall show below that this property is preserved
the renormali7.ation procedure.

Remark. In ali ..xamples we oonsider here the function w(.À) defined by equation (21.15)
quadntic in À and thus the corresponding gaussia.n integral can be performed. Aft.er
clnte!!:ration the new acti<>n ia still BRB symmetric, the va.riation of ê" taking now the

6C" = roa~ F 11 (A),

being a constant matrix. The BRS operator is still nilpoteot but no longer ha.s a
square. Therefore the property (21.23) is not sha.md hy all realizations of BRS
Jr•msform••tio•ns, and may require the introduction of additional auxiliary varíables. The
which follow can be proven in ali formulatioDS.
 Renormalizatíon o[ Gall!Je Theories 21.4
4B8

21.4 WT Identities

To study the consequence of the BRS symmetry (21.20), ít is necessary to intrnduce

souroes fnr ali !relds and all nperstors generated by the tre.nsformation:

Z (J,.,,ij,l,K,L) =f [dAJid>.j [dGdé) exp [-S (A,C,é,>.,K,L)

+J;A' + 7i.,C" + C"Tfa +la>.") , (21.26)

witb:

S(A,C,C,;.,K,L) C(A) w{À)+>."Fa(A)-C"Ma~C~ K,D'~~L"fíJ~Cfl(!',

(21.27)
Under a BRS tre.nsformation, only the souroe terms for A, C and C vary:
(21.28)

Thís implies e.n equation for Z:

(21.29)

The generatíng functional W of connected correlation functions satislies the same equa-
tion~

(1,6~, +ij,Ó1o +'loli~JW=O. (21.30)

We then perform a Lagendre trans[ormation;

r (A,C,é,>.,K,L) + W (J, 11 ,ii,l,K,L) J,A' + i).,C" + é"'YJ., + t">.", {21.31)

with; lir
6r
1;= 6A'' r;,.= liC"' {21.32)
6r 6r
{ 1,.= !J:ã'
'lo= 6C"'

Equations (21.31 ,21.32) imply as usual for the sources coupled to composite fields:

(21.33)

Tberefnre tbe generating functional of proper vertices r satísfies:

,;r 6r lir br o. (21.34)
M' tiK; + liC" tiL,.
As in the case of the non-linear a-me>del, because tbe transfonnation laws inve>lves ccm-
posite nperators, the WT identity satisfied by r talres tbe form of a quedratic equation.
We thru; expect. that the transformation laws will in general be renormali.OO. As implied
21.5 489
Renonnaluntiun of Gaug.e Thecri.u

by equation (21.34). the original a.ctíon satísiies equation (21.34), IIS eM be easlly wrified
by a direct calculation:

(21.35)

Basic vroperty. Let us show that it is sufficient to assume tbat the action satislies
equation (21.35) to derive equation (21.34).
Equation (21.35) implies:

f[ dA) (d.l.} (dCdC] [ liS E_

X
óA' éK,
.!.§._!!_
+ éC~ éL.,
exp (-S + l;A' +fio C"+ ê"ry0 +lo>.")= O. (21.36)

We now integrate by parts ova: A', C" and C" using:

liS -S é -S liS -s li -S
bC" e = -óC" e 'ic<>" = -Wo e

Equation (21.36) then becomes:

f - (
[dA] [dGdC) [d.l.J
2
li S
tiA'óK; + 6C
é2 S
0 liL
0
éS oS
+ l; 6K; + íj, 6La
x exp ( -S + sources) =O. (21.37)

At the regularized levei 62 SjliA'6K; and !J 2 S/óC"liLa, which are pToportional to traces
of matrices of the representation of the Lie algebra, vanísh because the group is compact.
Then equation (21.37) can be rewritten;

f [dA] [dGdC][d>.] ( J, li~, + íio 01 0

; ry 0 À
0
) exp ( -S + sources) O. (2Ü8)

This equation directly leads to the wr identíties (21.34).

The quantities 62 SfóA'éK; and 62 SjliC"'liL" still vanish a.fter renorma.liza.tion if some
group structure ís preserved or quite generally if dimensional regularization has been
used.

21.5 Renormalization: General Considerations

21.5.1 Renonnalil:ation
The methad explained below is dosely related to the method presented in the case of
the non-line11r cr-model (see Ch11pter 14}. We assume that the local actíon (21.27) has
!Jeen regu!11rized in a way compatible with gauge invariance. Perturbation theory then
oxhibits UV divergences which have to be removed by adding cmmterterms to the a.ction.
The identities (21. 34) imply relatlcms among divergences. We use them to prove that
equation (21.35) ís stable under nmarmalizatiou. In next section we then solve equation
{21.35) to find the most general form of the renormalízcd action. As usual our analysis
· l~ ba.sed on a laapwíse expansion of the regularizcd functional r, the first tenn being the
unrenormalized action S:
(21.39)
490 Rmtmnalizalwn of C auge Theories 21.5

To simplify nota.tíons, we introduce an auxilia.ry field /lR and add to the action (21.17)
and to the IPI functional r the combination -~>R),O. The new functlonal then satisfies:

ór =->.o, (21.40)

and the equations (21.34) and (21.35) become homogeneous quadratic equations which
can be written syrnbolically:

(21.41)
(21.42)

The general arguments presented ln Section 14.6 now prove that these equatíons are
stable under renormalizatjon. We here recall only the rnain steps.
Assumlng as an índuction hypothesis that we have bt!en able to construct a renormal-
ized action St-J which satiBfies equatíon (21.41) and reuders r linite e.t l-lloop arder,
we write the consequences of equation (21.42) at loop arder f:

t-1
S•rt+ft•S= I::rm*ft-m· (21.43)
m-1

The r.h.s. of the equation is linite by induction, so that the divergent pa:rt o( r 1 th•m
sati&fies:
(21.44)
We recall that rd1
• is the sum of the divergent terms in the asyrnptotic expansion of r in
terms of the regularizing para.meter. Equation (21.44) sbows that by deftning the i-loop
renormalized BCtion s, by:

St = St- 1 - r~ 1 • + hlgher orders, (21.45)

it is possible to render r !-laop finite with " renormalized action stíll satisfying equation
(21.41).

21.5.2 Solution of WT ídentities: genernl considerations

It now remo.ins to solve equations (21.44) and (21.41) using power CO>Jnting argurnent.s
to find the forro of the counterterms a.nd af the renormalized action. lt is however
useful to lirst exhíbit severa! properties of equation (21.44). In explicit form equatian
(21.44) reads:
(21.46)
where Vis the dífferential operator:

(21.47)
·éj

Quite remarkably the basic equation (21.41) implies t!w.t Vis nilpotent as we now show.
It is convenient to introduce some notation. We denote below by 6; the set of ali anti- ;:\li
-i
commuting fields K;, ca, co and x; ali comrnuting tields A'. L~, P.~· As we see on thc ·:JJ
21.5 491

explicit form (21.35), the fundamental equation for the a.ction S (and tbns r) then takes
the forrn:
(21.48)

and the operator iJ then reads:

{21.49)

The niipotency of the dífferentioJ <>pemtor iJ. Let us calculate iJ2. Since iJ is of
antkommuting type only the terrns generated by the non-commutation of õSjõO, and
âSjfJ:r:, with the differential operators 8fâ9, and ô/âx; survive:

(21.50)

We notice that this expression can be rewritten:

(21.51)

as a cons<:>quence of equation (21.48). The operator iJ, which play:s an essential role in
solving WT identities, agaln is a BRS operator with vanishing square.
Canonical invariance of the ji.mdamentoJ equation. Equation (21.48) has properties
analogous to the symplectic form dp A dq of dassical mechanics, it is invariant under
canonical transformatioru;. Let us malte the change of variables (O, :r:)- (9',:&1 ):

x, : , (9,x'), (21.52)

I Ô<p ( I
(21.53)
9,=- , 9,z),
8:&;

in which .p(:r:',ll) ís a.n a.nticommuting cype function. We first eliminate x; in equation

(21.48) using equation (21.52)

(21.54)

We then eliminate 9; llsing (21.53). We verify that we recover equation (21.48) in the
new va.rlables
(21.55)

InfinilesimoJ tron.<formation. Let us write these transformations in infinitesimal form:

(21.56)
492 Renormalization of Gauge TheonM . 21.6

calculate S(B',x'):

(21.57)

anà therefure (using definitíon (21.49)):

, , EhpâS
S(• ,x)- S(B,:r.) =-o: âxl 88, (21.58)

We thus find that o.ny infinitesimal a.ddit.ion to S of a BRS exact term can obtained by
a canonical transformation actíng on S.
Let us verlf:y that the effect of the quantizatíon procedure can be understood as such a
transformation. In ou r original problem the dependence on IJ.o, which is an artificial vari-
able, canoot change. This ímposes the dependence on IJ.o of the function I{J in equations
{21.52,21.53)
I{J(A,C,C,Ã,K,L,J1) ê"p.. + r,õ(A,C,C,)..,K,L).
lt íollows that the general change of variables is equivalent to a change induced by the
function .p on the restricted set {A, C, K, L J with in a.ddition the translation

(21.59)

One verifies that the gauge invuiant action with the K and L source terms, and the
tenormalized quantized actíon (21.63) are related by such a transformation with

21.6 The Renormalized Action

We have shown that the renormalized action ís the most general functional solutíon of
equation (21.35) local in the fields and sources, and consistent with power counting. We
now look for the most general solution of equation (21.35) in 4 dímensíons.

2!.6.1 General gauges

We have to solve equation {21.35) to.king into account power counting, symmetries and
locality. A!so we note that in the original action only the product CC ·appears. This
lea.ds to ghost number conservation. lf""' assign a ghost number +1 to C and ~ 1 to ã,
then K, has aghost number -1, and L,. ~2.
Power counting. In four dimensions the lagrangian density has d\mensioo four:

íL(A)J 4.

We choose the gauge lixing terrn in such a way that the field A has the minimal dimension,
i.e. 1 and F( A} dimension 2 (we haV>O exhíbited such gayges in Chapters 18 and 19),

[AI= 1, [F(A)]=2.
21.6 Renormalizatian of Ga:uge Tbeories 493

We impose [À] 2, by choosíng w(>.) sucb À has a. ronstant propa.gator;

(21.61)

The other dimensions follaw. Sínce D~{A) hM a term linear in A and ll. ronstant part
with one derivative it has dimension 1. The operator M,.0 then bas diroension 2. This
implies tha.t
[c]+ [CJ 2.
By convention we can cho01;e [CJ O, [Cl 2. This implies;

[K] = 3, [L) = 4.
Let US SUIDIIUJJ'Íze:

[.C]=4, [A)•d, [CJ=O, [C]=2, [>.]=2, [KJ 3, [L] 4. (21.62)

We immadiately conclude that, since K and L have dimensions 3 and 4 r~pectively, tbe
renonnalized oction can contain at most terms linear in K a.nd L. Simi!arly, since À has
dimension 2, the renormalízed action ís at most a polynomial of degree 2 in >., and the
coefficient of À" À~ ís a constrmt matríx.
The solution. Th solve equa.tion (21.35) we now pa.ra.metrize tbe solution in a. wa.y
reminiscent of the initial oction (21.27), hut ít should he kept in mind thl>t the pa.rameters
which appea.r a.re renormali&ed a.nd in general different from those pMametrízing the
action (21.27). The subscript "renormalized" is always impliad, a.nd is omitted only for
notational simplidty,
Because the part involving é and À is simple we expand S in powers of é and >.. Sínce
C has dimenslon 2, S Í8 a polynornial of degree 2 in C

s = s< 0l + s<ll + s<2l.

Similarly, since >. has dimension 2, S{0 ) ís a polynomial of d"gree 2 in >. and s{.1l of degree
!. We thus set

s<OJ = -~a,~À"À" + >."F,(A) + S(A,C,K,L), {21.63a)

sOl= g,p~À°C~C~ + Õ" M,.fJ(A)C 8 (21.63!>)
s< 2> = 9otJ.-.6C"éfJCJTc6 (21.63c).

We have immediately taken into account the dimensions of K and L: They can appear
only in s<0 l.
We first deal with s<•l ta.ken for >. O, i.e. S. The functional S satisfies by itself a
quadratic equa.tion consequeuce of equation (21.35):

o. (21.64)

Power rounting implies that S h; a.n affine function o( K and L, and ghost number
conservation implies that only the combinations KC and LcP ca.n appear. We then set:
494 Renornwlization of Gauge Theories 21.6

Power counting teUs us in additíon that ~~~ h a.s dimension O and is thus field independent.
D~ ha.s dirnension 1 and ca.n therefore only be B.I1 affine function of A.
The terms linear ín L and K of equation (21.54) yield respedively:

f~,Jt/l'C c' =
6
o. (21.65)

!f~tln;) co cn o. (21.66)

The first equation t.ells us that the constants f~7 satisfy a Jacobí identity sinoe the prod-
uct C"'C6c• ls antlsymmetric in (-r, 6,.-). They are therefore structure constao~.<; of aLie
algebra. lt remains to show tha.t the Lie algebra hos not <hanged. This is straightfor-
wa.rd when the gauge conditíon does not break the global symmetry. Otberwlse when the
original algebra is S<Jmí-símple this follows from a continuíty argument. Finally in the
general case one can still use gauge índependenoe of physical obse....,.bles a.s derived in
Section 21.7. The structure constants are thus línearly related to the structure constB.Ilts
appearing in the original action (21.27).
Equation (21.65) is also the integrability condition for equation {21.66} whir.h implies
the commutation relations:

(21.67)

Therefore the whole group structure is recovered. The last equation reads

(21.68)

This equation implies that C( A) is gauge invariant.

Let us return to equation (21.35) whi<h according to t.he preceding analysis now reads

. tiS
VS;;;:; D~(A)C" M; o. (21.69)

The nCJCt step oonsists in expanding the equatíon (21.69) in powers of é".
we find

Equation (21.64) impli"" that the À-independent term ca.ncels in the r .h.s..
term linear in À contributes. We thus find

The operator >.atJf6C" bas a vanishíng square. The general solution sol is the sum
a particular solutilln and an ex""t term À 0 6~fhC
0
 21.6 Renormalization Df Gauge Th.eariu 495

where the consta.nt tensor g"0 ., is antísymmetric in the two fiJSt indices:
9afh = -gpo., • (21.73)
The second term in equation (2 L 72) is new in the sense that it was not prcsent in the
bare action.
We now write the equation relating s< 2l a.nd s(l):
4lS12l ésül bSU)
>."~-
6C"
=-D'" (A)C"-- !J" C~ C"--
6A, + tl"r liC" .2
(21.74)

Again only tbe exact tenn, proportional to 9afh• survives in the r.h.s .. Moreover there is
a unique 50]ution of dimension 4 because a term proportional to >. is no longer ali<JWW.
Wefind
{21.75)
Again this tenn was not present in the bare a.ction. The last. equation for Sl2 l is a.uto-
matically satlsfied.
We conclude that the renormalized a.ct.ion h"" a. form similar to the tree levei w:tion
(21.27) except for an additional BRS invaríant tenn 1: 4 (>.,C, C):
.C4 (>.,C, C) = gal1., ( -~C"é 11 fi. C~ C'+ J."êlic>) (21.76)
whlch ls BRS exact
(21.77)
This term has exactly the form given in equation (21.58) and ca.n be associated with a
shift L,- La+ ho.,aCIIC'>.
The qtUtrtíc gh.ost !erm. A comment now is in order: since in general the renormalízed
action ls quartíc in the ghost temJB, in contrast to the original ection, the direct interpre-
tation of the ghost íntegra! as representing a <;leterminant In local form ls lost. HOW(ever
the following ~ult can be proven: íf one adds to the gauge function F., a term linear in
ao auxiliary field tra.nsfonning non-trivíally under tbe gauge groop, tben the integration
over this auxílie.ry field with an appropriate gaus.sian weight generates the quartic ghoot
terms in their most general form.
This property is expected from the general analysis of Chapter 16.
Renormal4ation of gav.ge invariant opemtors. Th generate correlation functions with
operator insertions one can add sources for them in the actíon. If the dímension of tbe
gauge inva:ria:JJt operatore is a.t most four the new a.ction is stíll renorma.lizable. The
, genera.l analysis is not modifiad; the only difference is that some coupling constants
are now spa.ce dependent. In the C8Be of operators of higher dimensions, the action with
iE, no longe r renormalizable. It is stiU possible to renorma.líze ít at any finite arder
introducing enough renormalízation constants. The determlnatíon of the general form
the renormalized action, i.e. the solution of equation (21.35) is a non-trivial prt>blem
require:< the use of sophistícated cohomological teclullques. In the case of compact
with semi-símple Líe algebras, the most general rolution of equation (21.46)
sum of gauge invariant terms and BRS el<act. contributions, i.e. of tbe form W.
first conj""tured ha.s now been rigourously proven. The part amcerning é, À
but the difficultie:< come from tbe set {A, C, K, L}. Note that tbe fonn of
re•ao•wali:red operators, when inserted in fie!d correlation functions, depends an the
gauge. Only the r.verages of produets of gauge invariant operators, or the ma.trlx
betw""" pbysical states, as we show in Section 21.7, ue gauge independent.
496 Renormalization of Gauge Theories 21.6

!!1.6.2 Linenr {lllUges

For a special class of ga.uges, the previous a.ne.lysis can be símplilied. This class is
cba.cacterized by tbe property that the ge.uge fixing function F~(A) is linear in the field
A, ra.tber than quadratic as in tbe most general renormalizable case:

F,(A) = F,.,A'. (21.78)

ln this case the opera.tor F,.(A) ís in general stíll of dímension two, but its correlation
functíons now are dimctly related to the correlation functions of tbe A' field and therefore
introduce no new independent renormalize.tion.
To derive the consequenc"" of equatíon (21.78), we use the .\-field equation of motion.
Let us aga.in explicitly pa.cam..trize w( A) as:

w(À) = ~aat:~À" À~.

Then the ),.field equation of motion which relíes an the identity:

j [d.\j 6~a [exp (-S + sources)] =O,

reads:
6 5
( -aa~ él~ + Pai óJ, o.
or in terms of the functional W In Z:

Since F(A) is linear in A, the À-lield equatíon of motlon is a lirst arder differential
equation and can be easíly Legendre transformed. The generating functional of proper
vertices r thus satislies:
:~ ~ -aaaÀP +FaiA'.
Equation (21.82) is satisfied by the action S and dea:rly is stab!e under renormalization.
lt implies that the quadratic and linear parts in À of the action are unrenormalized.
In particular no term of the form 9afhéflc> can be generated. The action rernains!i"l
quadratic in the ghoot fields. The renormalized actíon takes the simple form

(V being defi.ned by equation (21.69)} where in addition as stated above aa/J =d F ai

unrenonnalized.
Remar/c. ID the case of linear gauges anotber equation c!Ul be used to show tbat
geuge function is unrenormalized, the é ghost equatioo of motion. The equatíon
6S , tJ 65
IJ(]a =Fa;DpC ""Fai bK;,

implies
21.7 Renormalization of Ga~ge Thwries 497

and thus, after Legendre tra.nsforma.tion:

(21.86)

Equa.tion (21.84) is thus stable onder renorma.liza.tion.

We mention this property here for the following reasons: in the CBSe of linear ga.uges,
the introdoction of the ,\-field is not alwayg useful, except for strict gauge conditions
(generalízed La.ndau gauges). Ir we íntegrate over J.." a.nd set 1., to zero, then equation
(21.82) disappea.rs, while equation (21.86) rema.ins a.nd cao be used to ahow that the
gauge fixing term F(A) is not renorrnalized.

21.7 Gauge lndependence

Physical observables should be gauge-independ<mt. On the other ha.nd general correlation

functions are not. We show here that averages of gauge ínva:riant operators and S-
matrbt elements are unaffected by infinitesimal chiiJlges of gauges. Thís establishes gauge
índependence at least for gauges which can be continuously connected.
Gu.uge invanant vpemtors. To generate correlation functions of gauge invariant oper-
ators, we add source terros for them to the actíon. The actíon with these sources is still
gauge invariant (equation (21.3)) and the effective action BRS symmetric. Assumíng a
gauge invariant regularization, let us now examine how the vacuurn amplitude is alfected
by an infinitesimal change of gauge 6Fa, before renormalization.
Since the non-gauge inwriant part of the action has the general form Do <I> in which
D 0 has been defined by equatíon (21.22) (see equation (21.24)), the varia.tion éS of the
action takes the form IJS = Do(6<l>}, and thus the variation of the vnéuum amplitude ís:

(21.87)

The operator 1)0 is a dífferential operator and therefore we can íntegrate by parts. Using
/li
again the property that the traces J:;~ and óV:, A' vanish we obtain:

(21.88)

and this expression vanishes as a consequenoe of the BRS symmetry.

We have therefore shown the bare correlation functíoDS of gauge ínvaríant operators are
gauge iudependent, at least within a class of gauges whích can be continuous]y connected.
There exists therefore a renonnalization proced ure which produces gauge independent
tenormalized corre!ation functions of these opera.tors.
These correlation functious cont&in the complete information about the physical prop-
of the gauge theory.
S-matrix elements. We now want to study the gauge indepen<len~ of the perturbatíve
p>-mawtJI., when it exista. We first calculate the variation of renorma.lized oorrelatíon
I' fnn"'•'n••~ uuder an
infinitesimal change of gauge. We assume that we bave renormalízed
in a given gauge, but not yet e!imina.ted tbe regulo.rizing parameter.
then proceed as above. The varlation of the action in an infinitesimal change of
has exactly the form D5<l> (V being now the renormalized BRS operator) considered
498 Renarmaltzaíion of C auge Theories 21.7

····~
Fig. 21.1

in the preceding subse<:tion. We can still integrate by parts, but the resultíng integrand
does not vanísb identically becau.e we have introduced sources for non gauge invaria.nt
fields:

6Z(J} =-f [dAdCdCdJ.) ó<I>J;D~(A)Caexp(-S + J,A;). (21.89)

'W'hen we calcula.te correlatlon functíons we obtain a sum of contributions in which one

A; field has been replaced by 6/t>J;D~(A)C"', which ís a linear combination ofcompco;it.e
operators. When we go to tbe mass-shell, after amputation, we get a non-va.níshing
contrlbution only if there is a pole in esch externai momentum squared. For a cornpCISÍte
operator tbis happens only íf the line \s one particle reducíble. Then ou the mass-shell
we get a contribution proportional to tbe mlltríx element of the field itself (see figure
21.1 ). Thls argument was pr.aented for the first time in Appendíx A 10.3. The final
result ís that an infinitesimal change of gauge renorrnalizes multiplicatively the S-mlltríx
elements. Tbís corresponds to a field amplitude renormallzation. Tberefore tbe S-mo>trix,
properly normalized, is gaug.,.independent.

Bihliographical Notes
Renormalization of gauge tbeori.,. is díacussed in
G. 't Hooft, Nuc/. Phys. B33 {1971) 173; B35 (1971) 167; A.A. Slavnov, Theor. Malh.
Phys. 10 (1972) 99; J.C. Taylor, Nucl Phys. B33 (1971) 436; B.W. Lee and J. Zinn-
Justin, Phys. Rev. D5 (1972) 31Zl, 3137, 3155; D7 (1973) 1049; G. 't Hooft a.nd M.
Veltman, Nucl. Phys. BSO (1972) 318; D.A. Ross and J.C. Thylor, Nud. Phy8. B51
(1973) 125; B.W. Lee, Phys. Lett. 46B (1973) 214; Phys. Rev. D9 (1974) 933.
The anticommuting type symmetry of the quantized octíon is exhibited in
C. Becchí, A. Rouet and R. Stora, Comrn. Math. Phys. 42 (1975) 127,
where it is UBed to renonnalíze the abeliM Híggs rnodel. In
H. Kluberg-Stern and J.-B. Zuber, Phys. Rev. D12 (1975} 467,
quadratic WT ídentítíes generated by BRS syrnmetry appear and and
of a clllSS of gauge ínva:ríant operators is discussed.
The general proof, based on BRS symmetry, of renonnalizability ín an arbitrary
presented here, is given in tbe proceedings of the Bonn summer school 1974,
J. Zinn-Jnstin in 11-ends in Elementary Partícle Physics (wtures Notes in Phvsicl';~
37), H. Rollnlk and K. Dietz eds. (Springer-Verlag, Berlin 1975). For details"""
J, Zínn-Justln in Proc. iJf the J!tth Schoai of Theoretical Physics, Karpaa
Uniwrsíta.tis Wratislaviensis 368; B. W. }..,., in Methods in Fie!d TI.eory, L<:s no!lrJl""'"""
1975, R. Balian and J. Zinn-Justin eds. (Nortlí-Holland, Amsterda.m 1976).
An alte:matíve proof based on BRS symmetry and tbe BPHZ formalism can be found
C. Becchi, A. Rouet and R. Stora. Ann. Phys. (NY) 98 (1976) 287.
 21.7 Refl.armalization of Ga9ge Theories 499

R.enormalization of be.ckground field gaugeo; is discussed in

H. Kluberg-Stern and J.-B. Zuber, Phys. Rev. Dl2 (1975) 482, 3159.
A general study of gauge invariant operators can be found in
S.D. Joglekar and B.W. Lee, Ann. Phys. (NY) 97 (1976) 160.
The complete proof of the conjectured result is found in
G. Barnich, and M. HenneaUJ<, Phys. Rev. Lett. 72 (1994) 1588; G. Bamich, F. BriLildt
and M. Henneaux, Camm. Mr.th. Phys. (1995} to appear.
Attempts to quantize in quadratic gauges include
G. 't Hooft and M. Veltman, Nucl. Phys. B50 (1972) 318; S.D. Joglekar, Phys. Rev.
DlO (1974) 4095.
Consequences of BRS symmetry in the operator formalism are exhihited in
T. Kugo ànd I. Ojima, Phys. Lett. B73 (1978} 459, Nud. Phys. B144 (1978) 234.
The problem of non-linear gauges has been discussed in its full generality in
J. Zinn-Justín, Nucl. Phys. B246 (1984) 246.

These exerdses try to illustrate the methods whlch have been used in the quantizatíon
of non-abelían gauge theories. However, one should be warned that the abelian example
has some símplifying features which are not sharcd by the general oon-~>belían ca.se.
We want to study the quantiza.tion ~>nd renormaliza.tion of the free abelían gauge field

(21.90)

in the oovaríant gauge

G(A) âA,. + ieA! =O. (21.91)

21.1.1. Write the quanti:z.cd a.ctíon. Note that after quo.otization tbe actioo is no longer
free.
21.1.2. Verify the BRS symmetry of the quantized action.
21.1.3. Derive the corresponding WT identíties.
21.1.4. Show tha.t the addítíonal terms in the action generatcd by tbe quantiza.tion
,... orO<:eaure are themsel-ves the BRS variation of a local qua.ntity, i.e; tha.t tbey can be
W(A,>.,ê,C) .

. 21.2.2. Derive from the WT iderrtíties the general form of tbe renorma.lized action.
21.2.3. Wha.t can be said about the stability under renorma.lization of the gauge coJJ-
G{A) 07 lntegrate with the renormalíZ<l<i artion aver the >.-field •
. 21.2.4. Study the vari~>tion of tbe gauge field 2-point function w!~ under an infinitesi-
change 6e of the bate pa.rameter e ín the qWUlti:z.cd action. lt ls suggestcd to use the
of the exerci"" 21.1.4, i.e. expres5 the varie.tion of w,i!l in tenna ofthe varíe.tion
4>. An integr~>tion by parts will then lead to tbe result.
22 CLASSICAL ANO QUANTUM GRAVITY. TENSORS ANO
RIEMANNIAN MANIFOLOS

Thls ehapter has two purposes, to present the few elements of differentíal geornetry which
are required in different places in this volume a.nd to provide, for compteteness, a short
introduction to the problem of quantization of gravit~··
We fust briefty recai! a few concepts relllted to repararnetrizàtion (more accurately dif-
feomorphism) of R.íemannia.n manifolds. We introduce tbe notions of parallel transport,
a.ffine connoction, cnrva.ture, in analogy wíth gauge theories as discussed in Chapters
19-21. To define fermlons on Riemannían manifolds additional mathcmutical objects
are required, the vielbein and the spin connection. We then construct Einstein's action
for classícal gravity (General Relath<ity) a.nd derive tbe equa.tion of motion. In the last
section we finally study tbe formal aspects of the qua.nt.ization of thc theory of gravity,
following the !ines of the quantization of non-ohelian gauge theories of Chapter 19.
Because tbe theory of quantum gravity is not renormalizable in four dimensions (even
ín ít5 ertended fonns like supersymmetric gravity), the gl!lleral prejudice at present time
is that gnwíty is the low energy, large distance, remnaDt of a more complete tbeory whlcb
probably no Ionger has the fonn of a quantum field theory (strings?): In the language
of criticai phenomena, gravity belongs to tbe class of irrelevant interactions (due to the
presenre of the massless gra.viton tbe situation can be compared with the ínteractíon
of Goldstone modes in the low temperature phase). The scale of thls new physiCll is
indicated by Planck's mass, i.e. iJ; of the order of Joc/CN ~ 1019 GeV, where GN is
Newton's gravitational constant.
The a.ppendix is devoted to a short introduction to 20 qu81ltum gravíty, approacbed
from the point of view of matrix models in the large N limit. ·
The levei of mathema.tical rigour will be low and the notation old-fashioned. For
instance we sball write mcxst expressíons in t.erms of !ocoi coordinaíes, ignoring, beca.use it
is not essential for our pUrpose, that ~ral sets of overlappíng coordinates (eharts) with
transition functions are in general required to fully describe a manifold. The language
of fiber bundles will be avoided. The reader interested in more detailli is referred to tbe
literature.
The conventíon of summation over repeated lower a.nd upper índices will always be
uzed, except when the metríc is explicitly eudldean.

22.1 Change of Coordinates. Tensors

Let rp', i = 1, ... , N, be a set of coordinates whieh pararnetrize a manifold 9Jt and let us
consíder the reparametrizatíon:
'f''="''
(yl), (22.1)
in t.enns of coordinates •rl, the functions 'P' ('P') being differentiable.
We set:
d<p' = T} ('!'') d<p'; ~ T} ('I'')= :;: := B;'P'· (22.2)

As a matrix, Tj is an element of the definíng representation of the general linear group

GL{N) (general iovertible ma.trices).
22.1 501

We now consider a set of fields delined on !lJl and classify them accordíng to their
transformation ptoperties in a reparametrizatiou of !lJl.
Fields S (<p) which transform by a simple substitution:
(22.3)

are called scalars.

Quantities like drp', whích belong to tbe spáee tangent to the mllllifold at point <p',
define the transformation law of vector lields V'(<p) (called a.lso coutravaríant vector~):
V' ('f') ljV'i ('f''). (22.4)
Vector fields W;('P) of the dual spaee ""'chara.cterized by the ptoperty that the scalar
product v< (tp) W; {'f') is o. scalar (they are alsa called covaria.nt vectors). Thü; írnplies
the transformation la.w:
(22.5)
in which ( r-1 ): is the ínverse of the transpceed of the matrix TJ (
and tberefore belongs
to another representation of CL(N)). An example is provided by partia! derivatives of a
scalar. Indeed differentiating equation (22.3) we find
Ô<{li .
a,s"' -8'P'. =(r-')'• Bjs·, (22.6)

which shows that a,s tra.nsforms like a dual "<'Ctor.

We see that the transformation of vectors is delined in tet'ms of the substitution tp ~ <p1
a.nd a linear transformation. More generally we can classífy ali quantities with respect
to their tra.nsformation properties under the linear group. General n·tensors v;;;:·~~ ••
transform like the tensor product. of p vectors and n- p dual vectors:

(22.7)

For what follows it is useful to introduce a notatíon for the element T of the abstract
linear group GL(N). Then

TV'., ... ;_. ( ') = TÍ' ... T~· (T- 1 )i•"' ... (T- 1 ) 1" v:' ;, ...h { ').
ip•l···'" rp Jl J' t,..,.., t.. jp+1·--,.. ep

This represent.ntion can in general be mduced accordíng to the representations of the

permutation group acting on lower or upper índices. Moreover summing a n-tensor over
a pa.ir of upper .and lower índices yields a (n- 2)-tensor. We call this operation taking
the covariant trace. Using then the property that the tensor b~, wbere li is the Kronecker
6, is a.n invariant tensor we can render tensors traceless.
Infiníte5imal change of caonlinates. Often in this work we consider iofini!.esimal trans-
formations. Let us then consider the chMge of coordinates:

(22.8)
Md let us denote by 6, V ~ TV' V the va:riatíon at lirst arder ín ' of ali tensors. We
find

(22.9)
502 Classi.rol and Quantum Gravity 22.1

By ooru;tructíon é, V, being the dHierenoe between two tensors, is a tensor. Applied to a

tensor product 6, satisfies

Finally note tbe commutation relation

{é,.'~ .. ] .. 6,. ' with .; = c\ BA ~aj(\. (22.10)

DifJerential farms. A class of tensors ís oft.en encountered which h ave special proper-
ties: totaJiy antisymmetric tensor. or forms. They can be conveniently represented by
rontracting índices witb the generàl.ors e• of a Gra.ssmann algebra (see also Section 1.6}:

n 811 B'2 •. 8\.,. niti;: ... iw. •

Examples are provided by gauge theories where thevector potential AI' can be considered
as a 1-fonn and the curvature F,...
(the electromagnetíc tensor in QED) as a 2--forw (see
Chapters 18,19).
Let us now define " dílferentia.tion operation denoted below by d,

d =rP.a&p,. (22.11)

lt is clear that d is nilpotent because the /P•s aoticommute

d~ = 9Í8i~·a. =o.
With this notation the transformatíon (22.7) can be written
1 1
f);(I"}= d!p ; d!p1i, .. • dl"'"'(l!,;, ... in (I"').
The operator d h as an importa.nt property: a.ctíng on an n-form it generates an n+ 1-forrn,
í.e. a n + 1 antisymmetric tensor

Let us verify that th., antisymmetríc qu!Ultity in the r.b.s. is indeed a tensor. Using
d 2 =Owe find

= d'P''d~"'' ... d~";.,d'P''~;n:,,, ..... (l"'),

which proves the tensor character.
In tbis notatíon the 2 form associated with the curvature tensor F,..v can be related to
the g,.uge lield one form by
F 2(dA+A 2 J.
A form f); which satisfies dn O is called a closed form. A form f); which can be written
f); df);' ís eroct. Since d is nilpotent any exact form is closed.
Conversely a typical problem in dilferential forms isto know wbether a. dü<led form is
exact. The correspooding partia! differential equations can always be integrated locally
J
but depending on tbe topology o f the x-spa.ce sometimes no global solution can be found.
Such a situation has been encountered in the quantization of spin va.riables (Appeodix
A3.2.1).
Finally the property that by differentiation one generates new tensors is not general
as we shall argue in the next sectlon.
22.2 Classical and Q.. antum Graui.ty
503

22.2 Parallel 'I\-ansport: Connection, Covariant Derivative

We havc soon tltat the derivatives a,S of a .calar form a tensor. However the derivative
of a vector is not a tensor as one eaoily verifies. Therefore a new concept has to be
introduced, which will allaw to write covariant quantities depending on products of ve<:tor
fields at different points and thus also, ns a limiting case, on derivatíves. A similar
problem has alteady been discussed in the çontext of gauge theories in Chapters 18--21.

22.2.1 Pamlle/ tmnsporl

We consider two points, 'P(t),\0(2) on the manifold and a.n oriented, continuous, almost
everywhere differentiabie curve C joiníng them. To this curve we assocíate a linear
mapping U(C) from the tangent space at 'l'ol to the tangent "P""" at 'P(2)· We cal! this
operation parallel transport from 'f'(l) to 'l'(l)· Let v•
('l'o)) be .. veetor belongíng t<l the
space tangent at point 'f'Ol· lt is transformed into VJ
(22.13)

which now belongs to the spa.ce tangent at the point <P1 2 )·

We demand to U(C) ro be a differentiable functionalofthe curve C, which in addition
satísfy the conditions:
If a curve C 1 goes from <Pp) to I"(>) and a curve C2 from 'P<•l ro '1'(3) we impose:

(22.14)

The definition (22.14) implies that to a curve reduced to a point corresponds the group
identity:
U(C 1 point) = 1 "'* Uj (C= 1 point) = 6}. (22.15)
lf C is a curve going from 'l'!•l to 'f'!>)> we cal! c-• the sarne curve hut oriented frmn
'f'(2 ) ro 'f'O)· We then impose:

(22.16)

The transformation laws for other tensors follow from the following rules: A scalar is
invariant in a parallel transport. Demanding the invariance of the scalar product then
determines the form of parallel transport for dual vectors

!f W<! denote by U (C) the abstract linear transformation then the tensor product V ® W
transform like
U(C) (V® W) = U(C)V ® U(C)W. (22.17)
We can then more generally define parallel trarn;port for a.ny tensor, by imposing that a
tensor transforms as a tensor product of vectors.
DiffeomrYrphi.sms. It is easy to verify that the tnmsformation law (22.4) of tensors
under reparametri~ation induces a tra.nsformatioo law for the par..Uel transportar U (C)
whlch is:
{22.18)
504 Cla ..ical and Quantum Gmvity 22.2

in which we have assumed that the curve C goes from the point 'f( o to the point .,c( 2r
This equatíon sbows that Uj(C) is a (non-local) tensor.
~ affine connection. We now explain the relation between parallel transport and
the notions of connection and covariant derivatíve. We have assumed tho,t the matrbc:
U(C) as a functional of C is differentia.ble. Due to the composítion la.w (22.14) U(C}
is then entirely determined by its value for infinitesimal curves. We thus consider an
infinitesimal differentiable curve, í.e. a straight tine C connecting two dose points <p and
.P ='f'+ 6'1' and set:
(22.19)

where r k is called tbe affine connoction on the manifold. The connection is entlrely
cbara.cterized by its action on ""ctors. We denote by r{l, the Christo!Iel symbol, the
matrbc: elements of the connection

(22.20)

Converse1y the connection field completely chara.cteri.ws parallel tra.nsport. lndeed

once a. curve C is parametrízed in terms of a para.meter t, the correspondíng parallel-
transporter U(C) a.nd I]k('P}d<p•/dt play the roles reo;pectively ofthe evolution operator
and the hamiltonia.n of quantum mechanics. Therefore, "" shown in Appendbc: A2.1,
U( C) csn he written as a path-ordered integral along the curve C

(22.21)

where P sta.nds for path-ordered.

Usíng equatíou (22.18}, it is p08síble to derive the transformation law of the affine
connection under repa:rametrlw.tion. lt is convenient to express it in terms of the 1-form
=
r r 1d..,-" asoociated with the matrix r •. One then linds
(22.22)

wbere dT is the l-form d<p"o;T. ln componeut form, in the defining representation, the
equation becomes
(22.23)
We note that the connection is not a tensor, because it.s transformation law is not lin-
ear but only affine. Let us, ho~r, decompase the te!l50t ~. in to symmetric and
antisymmetríe parta, Çj• a.nà 7;'• respectively, ín the exchange j .., k:

(22.24)

Because T} has tbe forro (22.2) the inhomogenecus term in (22.23} is symmetric in
j - k. Henoe, both quantities transform independently under (22.23). lt follaws thst
the sntisymmetric part T,·i ís a tensor. Moreover, the restrktion to connections f~ • such
=
th.U 7]. O, i.e symmetric in j ,_. k, is consistent with the transformation la.w (22.23}
Blld characteri.ws the special cla.ss of parallel transpons without !Drsion.

~l =--------------------·M~.u.,lJt·!.l•k'l
lt.•l.l:L'F\,•"-
22.2 Classical and Quantum Gmvi.ty 505

2fUU! The covariant derivative

]t isnow possible to compare the vector field at point 'f + 6cp wíth the same vector field
parallel-tTansported from the point 'f to th<· point 'f+ Ó.p. Using definition (22.19) and
equatian (22.13) we obtain:
U(C)V'('I') V' ('I'+ Ó'l')"' (a. vi ('I')+ rjâ.P) vj ('I')]""'.+ a IIE<Pll· (22.25)
The difference in the Lh.s. is a vector belonging to the tangent space at <p + 6'f. lntro-
ducing the rovariant derivative V k whose action on vectors is
v.v; ~ô.. V'+r;,v;, (22.26)
the equation can be rewritten at leading order in Óip:

U(CJV' ('I'+ 6'1')- v• ('I' +ó'l') = -6'P•v.v' ('I')+ o lfó'l'il (22.27)

The covariant derivative characterizes infinitesimal paraliel transport and allows to con·
struct new local covariant quantities, i.e. tensors, from derivatives of tensors.
Parallel transport of general tensors yields the lorm of the corresponding cova:riant
derivative
V,=ll;l+l';, (22.28)
to which ...., can associate the covariant ,form dífferentiatíon
V' d +r"" d'P' (la,+ r,). (22.29)
The explicit representation of the coYaria.nt derivatíve thus depends on the nature of
the tensor it is arling on. On scalars for example V, ô;. The forrn of the covariant =
derívative when a.ctíng oo a dual vector is
V;l'J = ô.;V;- I'~;Vk. (22.30}
From equation (22.17) one immediately t.hat 'V; obeys the usual rules of dilferential
operators. lt is linear and rnoreover, i( V a.nd W are two tensors, then for the tensor
product V ® W we find
V' (V® W) = V'(V) ® W +V® V(W).
The general form of tbe covariant derivative of a tensor with n índices follows:
• n
v-v?J•••jF_ ô·V:j, ....JF + 'rjl y.:ft ... l:r•••Jp ~ r~tyh···il" (22.31)
" '"+' ... j.. t iP+I···ift L-. t.,i ;,..,.l ... j,.. ~ L ,,, jp+l···k~d.j.. ,
t~l l-p+l

which iJ; a tensor witb n + 1 indices.

Remarks.
(i) For the class of symmetríc Chri.5toffel connectíons r}. (torsion-free transport) one
has identlty between the derívative of a form and its covariant derívatíve. ln the notation
(22.12)

as one easíly verifies. In particular the ordínary curl of a vector is a 2-forrn and thu.s tbe
oovariant curl and the ordinary cutl coincide
a,v, -8;~ v,ví -'VjV;. (22.32)
(ii) The infinitesimal form of the transformation (22.23) of the Chrístoffel oonnection
oorresponding to (22.8), which is not homogeneous, can be wrítten
ó,r;. = a;aH' + <1ô1f'h- i1!<'~k + a.lr}l + aj<1I'lk· (22.33)
Although this is not obvíous from tbís expression. li,r;. ís a tensor.

~--------------------_.,~,fA<WI'·t."l>t.'L•'•'Inll.'l,l!.,'l't,:"~
506 Classtcal and Quanlum Gmuity 22.3

22.3 The Metric Tensor

v.·, now introduce the concept of distance in !!Jl. It is characterized by the Jine element
ds, distance froü> the point <p' to the poíut <pi + d101
(ds)~ 9ij{'f')d<p'd~,

whem tbe quadratic form is non-degenerate: detg i O.

The invariani:e of distances in a repammetrization determines tbe translormation law
of the metric tensor g,j (<p):

(22.34)

It ils consisttmt with notation (22.7) to write the in....,rse (in the sense of matrices) of !
the metríc tensor as g'i ( op):
(22.35) I
The tensors 9ii and g1' can be used to lower or raíse indices: The metric tensor establi.ili.,.
nu i50morphism between the ta.ngent vector spa.ce and its dual. lt is thus a standrucd
notation to uae the same symbol for tensors wbich are deduced fiom one of them by
lowering or raising índices with the metric teD50r, for example:

(22.36)

Càtllpolibilíty of po.rollel tron:rport with !he metric. We say that parallel tra.nsport is
compatible with the metric if it leaves ínva.riant the scalar product of two voctors:

(22.37)

in which v•
mld Wi are the para.liel transporte<! o[ v• and W' from '{) to .p. Using the
definition (22.13), we can rewrite the equation:

(22.38)

where we denote by TU the transpu...:! o[ the matrix U.

The interpretation of thís equation ís simple: the metric transported from <p to. <j; is
identical to the metrie at point .p. Taking then the Jimit of an infinitesimal path we
conclude: ·
(22.39)

Detenninativn of the connectíon in torsion-free po.rollel lramport. We have shown

tha.t the consenration of the scalar product of V<!ctilrs under par aliei traru;port implies
the <JOmpatibility condltion (22.39). Using the explicit forrn (22.31) it can be written:

(22.40)

u, eliminating toroion in para!lel transport, ""' assume rh symmetric we can solve the
cornpatibility eond;tion and expr..s Ij. explicitly in tenns of the rnetric tensOr. We find:

(22,41)

Compatibility -..·lth the metric uniqu<ly detemúnes torsion-free paraliel transpurt.

~· u.•l.l:L'F\ ·--=----------------------~·L"i·•'-'l·!.hk'l'''''i'~l
22.4 Classical and Quantum G.-auity 507

r.p+if_t

F!g. 22.1 Tho loop C.

22.4 The Curvature (Riemann) Tensor

Let us consider an infiDitesimal parallelogrllln C on the manifold, joining the points:

<p, '{l + <1, <p + <1 + <2, <p + <2 , and bock to <p, and calc.ula.te the corresponding parallel
transporter (see figure 22.1 ):

u (C)= u- 1{'f!+ .,,wJ u-• ('I'+,,+ <2,., + <2) u ('P +,,+.,,'f!+ •!) u (,+ '" ""'
(22.42)
Expa.nding In U (C) up to second order in q and <2 we set:

!nU ('P+<,.P) -r,<'+ ~r,,,;.; +o(<2), (22.43)

in which r, is the ronnection.

To evaluate the r.h.s. of equation (22.42), we then repeatedly use Baker-Hausdorf
formule. {the calculatíon follows the !ines of the derivation of the commutation relations
of generators in aLie algebra., see also Section 33.2.2):
.,A eB = .,(-HB+IA,Bl/2+·.. ).

The first arder in< in the exponential of the r.h.s. vanishes as a consequence of equatíon
(22.16). At seoond order we find:

U(C) = 1 - <\~R.;i +O (i),

where the antisymmetríc tensor R.,i ís defined by

(22.44)

We have obtained the expression of the cunoature tensor R., (or Riernann tensor) in
terrns of the connection. The curvature tensor charocterizes the variatíon of tensors in a
tra.nsport along infinitesimal closed curves.
Note that we could have used the expressiou (22.21) to perform the calculation for an
a.rbitrary dosed curve. We fix one point I" on the curve and wriLe a generic point <p + •·
Expanding the path-ardered integral up to 5eeüod order we lind:

U(C) = 1- f r•('P + <)d•• +!ff dt 1 d~1 Pjr1 (tp+ <Jf:('P + <)] +0 (< 3
)

We then expa.nd r. for t small. In the 6rst integ!"al the tenn pmportional to r 1 ( 'f!)
vani:lhes because the curveis closed. In the secood integral we ca.n neg!ect the dependence
in < at this order. A short ca.iculation then yields

U(C)-1--! L R,

·~---------------------~·IA<Wf',t.'l>t.'L•'•'In ll.'l,l!.,'l't*!w=o:
508 Classical and Quantum Gravity 22A
f
where Ris the 2-form R.;;d'f' A d;p' (see equation (22.29))

a.nd D, the domain of integration, a surface which has the curve C BS bounda.ry: âD = C.
Finally the curvature tensor R.,j is characterized by it.. matrix elements, when acting
on vectors
1
R;; V• =" R~JV ,

Equatlon (22.44) in component form then becomes

A general tensor transforms like

(22.45)

C..rvature tensor and metric. When parallel transport is torsion-free and compali-
ble with t.he metric, the cui'Vl\ture tensor is determined by tbe metríc tensor. A short
calculation yields for the tensor with only lower índices, Rk!ij 9kmRf.'J, the expression

RkliJ = }(â,é),gkj- ÔtÔk9j!)- ~ (8i9mk + Ôk9mi -Ôm9ik)g"'" (ÔJ9nl + Ôl9n;- Ôn9j!)

-(i.-. j) (22.46)

A few properties. Being defined as a commutator, R~; satislies:

(22.47)

For Lhe same r<!a50D the curvature tenllOr satisfi"" the consequence of a Jacobi identity:

[V;, [V;, V•lJ + cyclic permutations (i,j,k) =O,

which in terms of R~; rea.ds:

(22.48)

and in the framework of Riemannian geometry is called the Bianchi identity.

Note that the compatibi!ity condítion (22.39) implíes

'llús equation implies that the tensor R..uj ís antisymmetríc in k ,... I,

14ttJ = -Ru:iJ- (22.49)

a property can be verified on the explícit form (22.46).

''=="_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ "'~'U.,WI·!.hk' ~'1•'1·~ J

22.4 509

v.·~ also note that R.,., is antisymmetric in (k - l) and symmetric in the exchange
(kl)-(ij):
(22.50)
lt i• ea.sy to verify a cydk identity:

ll; 3 J.z + Rkii! + R; h' = O. (22.51)

Finally let u.s define the contracted curvature tensor 11;1 :

R;; R;'jk' (22.52)

which is called the llicci teruor. lt ís symmetríc:

(22.53)

and has therefore properties similar ta tbe metric. By contracting the remaíning índices
one obtains the .scalar curvature R:

(22.54}

Remarks.
(í) An important problem isto classify local tensors which are functionals of the metric
tensor. The compatibility condition (22.39) implies that the covaríant derívative of g;;
is not a new tensor.
However we observe that when the connection ís a functíonal of the metríc tensor this
also applíes to the curvature tensor. Ali tenrors depending only on the metric can then
be obtained from the curvature tensor IUld its oovaríant derivati~.
(ii) In the ca.se of srmrnetric Christoffel connections the variation (22.33) cr.n be rewrit-
ten ín a simple way to exhibít its tensor character:

(22.55)

The sytnmetry of the r.h.s. in the exchange j - k relieo on the cyclíc identity (22.51).

21!..4.1 Holonamy variables, ho!onomy group

The parallel-transporters U(C) assocíated with closed curves C a.re often called holonomy
líai"Í<<ble.. They in particular probe some topologícal properties of the manífold. This is
speclally obvious in the case of parallel tnmsport assodated with a vanishing curvature
tenS(lr. In the latter ca.se the oontour C can be continuously defonned without changing
tbe '<orresponding holonomy varíable.
In thi:l context it can be useful to consider the group formed by holonomies U ('I'; C)
ntigiuating from a cornmon fixed poínt I{J, ca.lled the holonomy group.

Fig. 22.2

·-=----------------------"~·IA<WI'·t.'l>t.'.J•'J<'I~· JII,J'I,l!.
510 aassícal and Quantum Grovity 22.5

lndependence of the holonomy group of t.he initio.l point. Let us show tha.t the ho\onomy
groups a.ssotia.ted with dífferent inítial points a:re isornorphíc. lf <P and ij; are two different
initia.l points, we introduce a futed curve Co joining them. To e;u:h curve C passing
through cp we can now associate a curve é passing through lj; (see figure 22.2):

(22.55)

In tbis way we have constructed a one-t(}.one mapping betwren curves pa.ssing through cp
and curves passing througb ip. The corresponding relation between parallel transporters
is:
(22.57)

] t establi.shes an isornorphism between the ho!anomy groups a5SOtiated with the paints
'I' and lj;. Therefore the abstract ho!onorny group is independent of the initia.l point
and intrínsic to the set of curves ou the rnanifold equipped with the equivalence relation
{22.56).
Holonomy and metric. Let us write the conlpatibility condition (22.38) for a ciO<led
curw:
(22.58)
lf the metric is positive, equation (22.58} shows that the matrices U {cp; C) corresponding
to c!ooed curves passing thraugh a poínt <P belong to a subgroup of the O{N) orthogona.l
group (N is the dimension of tbe ma.nifold !DI). Let us now consider the curve C formed
by the infinitesimal para.llelogram of figure 22.1. If we expand equation (22.58) up to
sccond arder in<, we obtain a condition whicb, in terms of the curvature tensor, can be
written:
(22.59)
Thís result provides a geometrica.l interpretation for the antisymmetry (22,49) of the
curvature tensor: it is the antisyrnmetry of the generators of the orthogonal group.

22.5 Cawriant Volume Element

To construct for example a <:lassica.l action we need a volume e\ement invariant under
reparametrization. The euclidean measure tr8ll5forms líke:

(22.60)

Let us denote by g tbe determinant of the rnetric g;;:

(22.61)

It transforms like
2
g' detg = (det1}) g. (22.62)
The volume element dp(<p):
(22.63)

is thus invariant. Of course if we multiply dp(!p) by a sca.lar (like a function of tbe sca.lar
curvature), the volume element remains invariont.
22.6 Classical ond Quantum Gmvity 511

Remark. Let u& ca.Jculate the logarithmic deriva.tive of g:

lng: Indetg trln g. (22.64)

Dillerentiating this ldentity we find:

(22.65)

Comparíng the r.h.s. with the expression (22.41) for the connection, we can vetify the
simple relatíon:
(22.66)

This in particular implíes for the wvariant divergenre o[ a vector:

'i7, v•. = ../?;a,

l ( ,;g v•') . (22.67)

and therefore:
(22.68)

22.6 Fermions, the Vielbein

We now bríefly explain how one can construct spin 1/2 fennioDB living on Riemannían
manífolds, be<::ause tbe construction is non-trívial. !t is first necessary to introduce a
local frame in the space tangent to the manifold (this can a.lways be done locally, but
ma.y lead to topologica.l obstructions}. The set of vectors e"' (lO) which form the local
basis is called the tlielbein. We assume that the metric is positive. We can then choose
vectors ortbogonaJ wíth respect to the metric 9iJ (<p), of length 1:

(22.69)

We do not distinguish upper and Jower internai indioes a, b because as equation (22.69)
shows the internai metric is euclidean. lntroducing the vectors e';, obtained as usual from
the e"' by Jowering the index with the metric tensor, we can rewrite equation (22 .69) a.s:

(22.70)

whicb sbows tbllt the matrix e'" is the ínverse of the matríx et. Using relation (22.70),
equation (22.69) can finally be written:

(22.71)

This equation yields the expression of tbe metric tensor in t.erms of the vielbein. As
we ha.ve done above, in what fo!lows we use the first letters of the a.lphabet a, b, c, d, .. . ,
: to represent índice; corresponding to the euclidean metric and the letters i,j,k, ... to
represent tensor índices.
lt is clear that the tnetric is invariant under orthogona.l traru;formatioru; acting on the
frame:
e~ (lO)= o., (tp) (e')~ ('P), (22.72)

----------------------~·IA<WI'·t.'l>t.'.J•'J<'I~·ni,J'I,l!.
512 Glassica! and Quafltum Gmtrity

in which o.,
is aiJ orthogonal matrix. Fermions wíth spin then transform under the
spínor representation of this local O(!V) group (N is the dimension of !m). As explained
in Appendix A5.3. this ímplies that the spinors tp and ,f, transform like:

(~Jo ('f) :~~(O) 1/1~ (rp) •

(22.73)
{ (.P),. (<,c)= 1/ÍÍ;('I') A/Jo. (O),
ín which the matrix Oab is given in terms of A by:

{22.74)

With these transforma.tion laws, the quantity i/n"e ''i>

0
is a tensor undcr reparametriza-
tíon, independent of the local frame.
We recall that if 11. is written:

(22.75)

in which e., is an antisymmetríc real matríx, then:

Oab =(e").,. (22.76)

Gauge in1!llriance. Since the choice of the local Etame is arbitrary, we requíre invario.nce
of ali physical quantities under the transformations ofthe local orthogonal group (22.72).
This forces us to inttoduce a new form of parallel transportar which is a.n orthogonal
curve-dependent matrix O( C). In a change of local frame ít transforrns like:

(22.77)

in whích the curve C g<:>eS from 'P to <(;. Such tra.nsforma.tions are called !)auge traru;for-
mations (for details see Chapter 19). For infinitesimal dilferentiable curves, O(C) con
be expressed in terms of the spín conne<:tion wrb, which has the forro of .. gauge Jield.
The connection is a vector on the rnanifold, and a matrix belonging Lo the Lie a.lgebra of
O(N):.
O(C)"'l (22.78)
In the case of the gauge theories discussed in Chapter 19 this gauge field i5 a new
independent mathematical quontity. However here we notice that if we consider the
matrix (whlch is a scalar under reparnmetrization):

(22.79)

then, as a consequence of the compatibility condition (22.38) and equa.tion (22.71), it

satisfies;
(22.80)
a.nd is therefore orthogonal. Equatíon (22.72) implies that it ..Jso sa.tislies equation
(22.77). Tbe matrix u•'(C) ís therefore the parallel transportar we are looking for:
o•b(C) ;;;; U""(C). Taking in equation (22.79) the liroit of an infinitesimal curve""'
obtaio a relation between the connection wjb and the connection r;k:
(22.8!)

·-==------------------------·M~,VhLI·!.hk' ~'1•'1·~ J
22.7 Classical and Quantum Grovity 513

Equation {22.41) expresses r~k in terms of the metric tensor. lt follows that the connec-
tion t..-•fb can be expressed. in term:s of the vielbein which replaces the metrk tensor a.s
the basic geometrical quantity in theories with spin 1/2 rermions:

(22.82)

By expanding ali tensors on the hasís formed by the vielbein, we can completely réplace
the condition of covaría.nce under reparametrization by the condition of independence of
the local reference frame (gauge índependence). To the connection ,..ib is associated a
covaríant derívative '17; wbích on a vector of camponents v•:

{22.83)

acts like:
{22.84.)
Fína.Jiy we can define a general oovariant derivative whose action on euclldean índices is
given by equation (22.84) a.nd íto action on tensor índices by equation (22.31 ). Let us
then calculote for example íi';ej:

(22.85)

Wíth this definítian, equation (22.81) t~ the simple form:

V,ej =0, (22.86)

and therefore:
V,V" ejí7;V>. {22.87)
This last equatlon directly follows from the definitlon of the oovariant derivative and
equation (22. 79}.
A Iast remark: wben C is a closed curve, equation (22.79) becomeo a similarity relation
hetween matríces. In particular the curvature tensor R~t, MSOciated with the connection
w~b, is simply related to the Riemann tensor by the equation:

(22.88)

Coooriant derivative and fermums. In the case of fermíons the covariant derívatíve
17 takes a form whkh can be inferred from the considere.tions of Chapter 19, the trans-
formation law of the spínors and the expressíon of matrices of the spinor representation
(22.75):
(22.89)

22.7 Classical Gravity. Equations of Motion

!n the gravítation theory known under the name of General Relativity the rnetric tensor
f•dd Y;;('l') beeomes a dynamical variable. Using previous considerations we can con.
•truct a classkal action for a. metric tensor coupled to matter which is local ond purely
t:<"Ometrícal, i.e. independent of the parametrization of the manifoJd.

~----------------------.,,~,fA<WI'•t.'l>t.'.J•'J<'I~· JI ,J'I,l!.
514 Clas~eal and Qt<nntum Gnwity 22.7

22. 'U The elas si cal action

The pure gravíty oction S(g) must ha.ve the forro of a scalar integrated wíth an inva.riant
volume e}ement. lf we look for a. local action with only two clerivatives we see that the
scala.r curva.ture ;,. the only candidate. We then obtain Einstein'E action

(22.90)

To Eínstein's gcavity action we ca.n add a derivative-free term, called a cosmological term,

(22.91}

More generally C\7VBIÍance and locality alone allow any function of the curvatuie tensor
a.nd the scala.rs obtained by contracting ccwariant deriv&tives of the curva.ture tensor.
Finally tbe simplest covariant action for a scala.r matter field <P('P) couplecl to gravity
takes the form
(22.92)

ln the case of matW in the form of a vector field A; =ciated to some e:xternal gauge
symmetry ofthe type discussed in Chapters 18, 19, we note tha.t tbe gauge field curvature

F,s =8;Aj -éi;Ai+ [A;, A;},

is a tensor because it is a 2-form (ree (22.32)). Therefore a possible action is

(22.93)

Fermion matter couplw to grovíty. The consideratious of Sectíon 22.6, in particular

the clefmítions of the vielbein and the CQ\I!ll'Íant derivative (22.89), also allow us t.o write
an action for Dirac fermion matter coupled to gnwity:

(22.94)

22. 7.2 Classical equation ôj mation

Let us derive the equation of motion for pure gravity, obtoined from the variatíon of the
action (22.90)
Sx,ov~,, "'j d-" rp,f9R.
Note the identity (be careful about the signs)

(22.95)

Using thís identity and integrating by parts, we can also write the a.etion

-=---------------------,~.u.,wl.!.hk' ~,1•'1·~ J
22,7 Classícnl and Quantum Grovity 515

If we add a cosmo!ogjcal term (equation (22.91)) th• actíon beromes

(22.96)

Equation of motíon. We bave therefore to ca.lcu!ate tbe variation of vario115 quantities

when the metríc 9<i varies
(22.97)
Note that the varía.tíon of the cOime<:tioD is a tensor (the inhornogeneo115 part of the
transformation ca.ncels in the varíation). lndeed one finda after a ahort ca.lculation

(22.98)

In the same way one can calculate the variation of the curvature in tenm of IT. Since
6r ia a tensor one is not surprised to find

Substituting we obtain

Síoce

W<: find

Moreover
1 ..
6 ( v'9 R) =
2
..;gg'' h;; R + v'§é R
2~y'ih;iR- ,jgR'ih,i +! (g"gi'"- g'"'g>l) \1,\l,l..,.,;.

lntegt'atíng by parta in the a.ctirm and using \l i9jk =O we fina.Jly obtain

o.nd th115 the equation of motíon

(22,100)

Note that in two dimensions the curvature tensor has only one oompçnent and thus
R•i c< 9 1i. Equation (22.100) reduces to only one <'quatíon. Ta.king then the cova.ríant
trace we find (in N dimensions)
(N-2)R=O,

~---------------------·~·IA<WI'•t.'l>t.'.J•'J<'I~·ni,J'I,l!.
516 Classiroi and Qunntum Grovity 22.8

equation wbich is identica.lly satisfied in two dímensions. We recover Gauss-Bonnet's the--

orem which implies that Einstein's action in two dimensions is topologica.l (see Appendi><
A22).
Tlte cosmologica! constant. The action (22.96) with a cosmological term leads to the
equation o( motion
(22.101)
Ta.Jdng the covariant trace we find

(N -2}R+NA O,

whi<:h showo tbat the cosmological constant induceo; a non-trivia.l curvature of space even
in the aboence of matter.
Matter coupz..d to grnllity. If we couple matter to gravíty, i.e. add for example the
action tenns (22.92) and (22.94), tbe equation of motion becomes

(22.102)

wbere tbe quantity 'T';

T'' ,21is.........
6g;;
coincides in tbe limit g;.J _, li;; witb tbe energy-momentum tensor as defined in Appendb<
A13.3.

22.8 Quantiz.ation in tbe Temporal Gauge: Pure Gravity

It is easy to verify that, due to tbe covariance of the equatíons of motion under reparame--
triza.tion, it is impossible to quantize the theory in tbe standard way because the time
components of the metric tensor bave oo conjugate momenta. a.nd thus only generate
constraints. Thís is a problem we bave a.lready encountered in gauge theories and we
shail use tbe sarne strategy to solve it. We bere choose to quantize the thoory in tbe
temporal gauge o.nd we consider, for simplicity only pure gravity.
By a chenge of coordinates we ca.n reduoe the metric to the forro g00 = 1 a.nd OO< = O
for i -F O. li we tben specia.lize the action to such metrics we obtain, as equation of
motions, tbe space components of equat.ions (22.100). The remaining equations have to
impased as additional constraints. As a notation we below use the letters a, b, c, d. for
space índices.
The actíon in the tempoml gauge. Let us first calculate the components of lhe curvatura
tensor:
!
R,..ood = RoodO = ôÕYo<! - ~ Üo!lomYmnÜog,...
R,..bal Jt:bc, i
~Uo9acUo9bd + Uo9~Uo9od,
where Jt;:bc, ls the curvature tensor in the N 1 spa.ce. It follows

Roo =~g"d~9ad + iE/og"bUo9ab

R= R,..r.:dg..dl< + 2Roo
=R""·+ !âog"bâo9.b +! (râoM) + g"bô~g••.
2

, _ ,_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ ,,~•U.,WI·!.hk' ~'1•'1·~ J
22.8 Cla5sícal and Quantum Gmmty 517

We have to integra.te by pa.rts to e.limhmte the second derivatives.

2
Jgg"~BÕg.b-= -Jg [8oY<>b8o9"'+ }(g" 08og.o) ] +total derívatíve>i.

We finally obWn

St.mo ""f dN cpC

C= Jg [R"•· ~ôag.o8o9" 0 t (g"6ôoY<>b/]. (22.10~)

The corresponding oonjuga.te momenta are:

(22.104)

and conversely
(22.105)

The ha.miltonian density follows

1i = n"'Bo9ab - c
= ~ [n.oll"0 - N~2 (TI~) 2 1- ygFf'O·. (22.106)

We note that covatiance implies that the qua.dràtic farm in the momenta. is a homogeneous
function of the metrlc tensor. Therefore orderiog problems cannot he a.voided and this
rell~ts in a functional mea..sure whkh ís not llat in metric space but multiplíed by a
power of g:
Z = jfdg,igN(N-~)Ie] I1 6(90i) exp(-S{g)].
í=O~N-l

lt is then p<)&'ible to use the same funclional techniquee as in the gauge theory, and
introduce covariant ga.uges which lead to a BRS symmetric elf~íve a.ction.
Constmint... We must still study the constraints which are

Jtk = O, R - 2Jtl0 = O. (22.107)

The fim set is a simple generali~ation of Gauss's law in gauge theories

RO< =o= g""gbcR,.bOd = _ _!_v·•·n•e

.;g • ' (22.108)

where v•P· is the C<MI>iant derivative in N - 1 spa<:>l dimem;ion. Note tha< to prove this
re:sult one can use
(22.109)
These constraints imply that the wave functional is invariant in a change of space coor-
dinate.. lndeed in an infinitesimal coon!inate transformation the v&ria.tion of the metrk
tensor is ( equation (22.9))

(22.110)

-----------------------.,,~,fA<Wf',t.'l>t.'.J•'J<'I~·ni,J'I,l!.
518 Clo.ssical and Quantum Gravity 22.8

Thus the invaríllll~ of the wave functíonal is expressed by

(22.111)

which ís exactly the constraint. These constraints wmmute with the hamiltoninn since
time-independent reparametrization remains a symmetry of the thoory in the wmporal
gauge.
The last constra.int is reiated to time-independent time reparametrization and thus has
no equivalent in gauge theories:

(22.112)

A short calcnlation shows that

R-2Roo o. (22.113)

Therefore the constraint, known as Wheeler-DeWítt equatíon, ímplies that the physical
states corresponds to wave functionals whkh are eigenfunctÍOilli with eigenvalue O of the
harniltonian. In the temporal gauge there is no time evolution in the spa.ce of physical
stat.es. Dynamics ls entírely encoded in the very definition of physical states.
Remo.rks. The qua.nti:z.ation method presented her" has been criticizoo because it
relíes on tbe possibility of defining a space-like surf..,.,, a notion ill-definoo in presenc<>
of strong metric fluctuations. As for gauge fields it is indeed completely justified only
in perturbation thoory, when the background {or dMBical) metric is static and typical
devia.tions of tbe metric from the backgr<>und (or classical) metric are sma.ll.
The definition of quantum gravity, beyond the formal levei leads w a number of un-
solved problerns, rrom whlch we shalllist only a few.
Wbile in general the símplest wa.y w define a. qua.ntum lield thoory, is first to construct
the euclidean tbeory and then proceed by analytic continuation to real time, in quan-
tum gra.vity tbe euclídee.n action is not bounded from below. Moreover the connection
between euclidean and MinkoWskia.n gravity is much less obvious than in the case of
non-gravitational tbeories because a change in th" signature of the metric is involved.
Lattice regularization of the euclideo.n theory by simpliclal gravity, ca.n be easily
achieved in two dimensions (see Appendix A22) but remains a partially unsolved problem
in four dimensions.
In the real time fonnulation perturbation theory is expa.nded a.round a fixed back-
ground metric 'Ri which, in the ea.se of asymptotic flat spaee, is simply the Minkowsky
metric { +1, -1, -1,-1 ... }. How nature chooses this particular signature is a non-
trivial interesting problern. In perturbation thoory pure gravity then describes the self-
interaction of a. hypothetical spín-two massless partkle, called tbe graviwn. From the
equations of rnotions we see tbat it has ~N ( N + 1) /2 2N = ~ N (N 3) dynamical
degrees of freedom. In particular the dimension three is for gravíty the a.nalogous of
dimensioo two for gauge theories: due to reparametrization ínvarlance and constra.ints,
the metric has no dynamical degrees of freedom.
However even pure qua.ntum gravity in non-renormali:l.a.ble in four dimensíons. The
best we ca.n expect in a covariant gauge is tbat the metric ha.s canonical dimension
~(N -2}. However the action ís non-polynomial. As in ~he case of the non-linear u model

~· Jlt.J•t.l:'. '1'\r,_,_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _,,~,u.,wt·!.hk' ~'i•'t·~ J

22.8 Classical and Quantum Gmuity 51S

we thus eJ<pect that pure gravíty ís renormalizable only in two dímensions, or more exact)y
Íll 2 +< dimensian since in two dimensions the actíon is topologícal. Attempts have been
made to follow the methods successfully employed in the case of the non-linear 17 rnodel
and to look for a non-trívial UV fixed point in 2 + < dímension. However the analysis of
singularíties of perturbation theory when the dimension approaches two is complicated
because the theory does not exist in two dímensions, and therefore remains incondusive
(but the inclusion of a scalar field called dilatou seerns to ímprove the situatíon). In
fe.ct it is commonly believed that the theory, even in its supersymmetric form, remains
non-renormalízable, a property which would indicate the breakdown of local quantum
field tbeory at Planek's scale.
Note ..., additiono.l outstanding problem: Vacuum energy contributions of ali particles
autornatically generàte a cosrnolagical term in quantum gravity. However, experimentally
such a term is bound by an exceedingly sma.ll number when expressed in the natural
scale of the Planck rnass. It has been notioed that a supersyrnmetric theory h as an exact
vaníshing cosmological constant, however •ince experimentally supersymmetry, íf it ls a
symmetry of nature, must be """"rely broken, thls has not provided a solutíon to the
prohlern.
Fínally the peculiarities of two dimensõoos are discussed In the appendix.

Bibliographlca.l Notes

For more mathematical details see for example

E. Cartan, Leçons sur la géométrie des esp<>ces de Riemann (1928}, reprínted by
Gauthiers-Villarli (Paris 1950); A. Lichnerowic.z, Gt!ométrie des groupes de tronsfor-
mations {Dunod, Paris 1958); E. Nelson, Ten<!ur Analysis (Princeton Universíty Press,
Princeton 1967); S. Helga.son, Differential Geometry, Lie Groups and Symmetric Sp<>ces
(Academic Press, New York 1978); Y. Choquet-Bruhat, C. DeWitt.-Morette Md M.
Dillard-Bleiclc, Analy.ris, Manifolds and Physics (l\lorth-Holla.nd, Amsterdam 1977).
For a discussion of tw<>-dimensional models on Riemannian manifolds see
D.H. Friedan, Ann. Phys. (NY) 163 (1985) 318.
Fina.lly for an introduction to the classical thoory of grav:itation see for exa.mple
S. Weinberg, Gravitation and Cosmalogy (WIIey, New Yotk 1972).
An elementary introduction to quantum gravity is found in
M. Veltma.n, in Methods in Fie!d Thcory, Les Houches 1975, R. Balian and J. Zinn-
Just.in eds. (North-Holland, Arnsterdarn 1976).
Various íssues in quantum gravíty are discussed by B.S. DeWitt and S.W. Hawking in
theír oontributions to
Reiativity, Groups and Topology, proceedíngs of the 1983 Les Houches school, B.S.
DeWitt and R. Stora eds. (Elsevier Science, Amsterdam 1984),
and in the selected papers gathered in
Et>elídean Quantum Gnwity, G.W.Gibbons and S.W. Hawlting eds. (World Scientific,
Singapore 1993}.
For a discu55ion of matrix rnodels in the large N -límit Md its application oo two-
dímensional quantum gravity or criticai models on random surfaces see for <>xa.rnple
G. 't Hooft, NucL Phys. B72 (1974} 461; E. Brezin, C. ltzykson, G. Parisi and J.-B.
Zuber, Comm. Math. Phys. 59 (1978) 35; D, Bessis, C. Itzykson, and J.-B. Zuber, Adv.
Appl. Math. 1 (1980) 109, and refurences therein; E. Brézin and V. Kazakov, Phys. Lett.
B2l!i (1990) 144; M. Dougl,.. and S. Shenker, NucL Phys. B235 (1990) 635; D.J. Gross

_______________________ ,,~,fA<Wf',t.'l>t.'.J•'J<'I~· JII,J'I,l!. 'l"trro.-.:.-:..

520 Classical atui Quantum Grnvity 22.8

and A.A. Migdal, Phys. R.ev. Lett. 64 (llhlO) 127; M. R. Douglas, Phys. Lett. Bc38
{1990) 176; P. Ginsparg, M. Goulian, M. Fl Plesser, and J. Zinn-Justin, Nucl. Phy>.
B342 (1990) 539; E. Brézin, V. Kazakov a.nd AL A. Zamolodchikov, hi&d. Phys. B338
{1990) 573; P. Ginsparg and J. Zinn·Justín, Phys. Lett. B240 (1990) 333; D.J. Gross
and N. Miljkovic, Ph.ys. Lett. B238 {1990) 217; G. Parisi, PhyB. Lett. B238 (1990) 209.
Seo also
R andam Su'ljaCJ;s and Quantum Gravity, proceedings of the Ca.rgêse wotkshop 1990,
O. Alvarez, E. Marínari and P. Windey eds. (Pienum Press 8262, New York 1991),
and the review
P. Di Francesco, P. Ginsparg and J. Zinn-J ustin, Phys. R.ep. 254 (1995) 1.

Exercises
Exe'I'CÚ!e 22.1
To the reader who is puzzled by tbe Wheeler-DeWitt equation we propose to study the
qua.ntization oí the relativístic poínt-pnrticle of space-time coordinates x" (the metric
being MinkowskiBD, i.e. + - - · · ·) in an externai electromagnetic field A,.. We consider
the followíng lagrangían

(22.114)

where r pa.rametrizes the particle trajectory and N(-r) is a metric factor which ensures the
independence ofthe theory on the traje<:toty pa.rametrization. Vectors ha.ve components
xD)xl, ....
Thil; theory has a reparametrization !nva.riance'
M
r-'T', N'(r') = N(r) d-r'.

Of course it is possible to elíminate N from the lagrangi!UI by solving the corresponding

equa.tion of motion

One obt.ains
(22.115)
This lagrangi!UI has still a reparametrí~ation invariBilce.
Discuss quantization ftom both points of view (N(r) .., 1 defines tbe temporal gauge).

E:uTCise 1?1?.2
The exercises which fol!ow try oo motivate the description of discretized 2D quantum
gravlty in terms of triangu)ated surfaces as expla.ined in the appendix A22.
22.2.1. Random waLks. One considen; brownian ra.ndom walks on a hypercubic lattice
( o;ee tili;o A ppendix A28.1) with constant decay r ate per unit time 1 a or survival
probability a, O <a < 1. Calculate the probability U(y, :r) for a walk startíng from poínt
:r to pass through the point 11 at any later time. Study the scaling límit in which I:r - Yl
becomes large and a appraaches 1.
22.2 .2. One can write directly in the continuum a path integral which weights the paths
accordi.ng to their )ength as in the preceding discretiwd example. One can integcate with
theactiotJ
f
S(x) m dt~. (22.116)

, , __ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ !1/~,VhLI·!.hk' ~'1•'1·~ J
22.8 Cla.ssirol and Quantum GrrliJity 521

We recognize a form analogous te expression (22.115). \Vith thíE a.ctíon, however, one
again faces the problem of reparametrization invariance. Tberefore study first the con-
tinuum limit of a natural discretization of this expression, the probability distribution
U(y,x) being given by

with 7o =X, Xn+I = y! and 11- being a consta.nt.

22.2.3. Recov'er now tbe preceding result from the continuum a.ction (22.116).

·-----------------------"~·IA<WI'·t.'l>t.'.J•'J<'I~·ni,J'I,l!.
522 Glassical and Quantum Gravity A22.1

APPENDIX 22
MATRIX MODELS IN THE LARGE N LIMIT AND 2D GRAVITY

Quanturn gravity in four dimen.sions, beyond the formal levei, is !'ar from being well un·
derstood. lt is therefore quite natural to investigate the properties of quantum gravity ít1
lower dímensions. Two dimensions are of special interest because a. specula.tive extensíon
of quantum field theory, string theory, can be formulated as 2D gravity coupled to somP
ma.tter. In recent )'<'l'U'S progress has been reported in the problem Cl[ 2D gravlty coupled
to discrete matter, in the euclidea.n formulation. The solution uses original techniques:
spac&-discretized gr~<Vity ls reformula.ted in terms of matrix integrais, and the continuam
is rcrovered in a peculiar "double scaling limit" in whích the size N of the matrices b...
carnes huge. Since most af these developments are ratber technical""' here hriefly review
only the soiution of the on&-matrix problem, which is reievant for two-dirnensit>nal pure
qmwtum gravicy. The interested rea.der ls refened to the literature for rletaib. Now
that the problem of evaluating integrais over mn.tricet< of large size has also statistical
applicatíons for fluctuating surfaces, or random bamiltonians or transfer matrices.

A22.1 Quantum 2D Euclidean Gravity

A2f.1.1 Classtcal 2D gro.vity

Einstein's actíon for pure gravity in the presence of a cosmological term reads (see equa·
tions (22.00,22.91))
S(g)

in which 9ij U. the metríc tensor, 9 detg;;, R the scala.t curvature and K,A are two
coupling constants. The para.meter A, which multiplies the volume element, is called
the cosmological constant. In d 2 rlimensions classieal gravíty is trivial because the
curvature term does not contribute to the equations of motion (see Section 22. 7.2). Ao
a consequence of the Gauss-Bonnet theorem, the total curvature hiiS a topological inter·
pretatlon: the total curvn.ture of an euclídeen <:losed surface is proportional to the genue
o f the surface,
f d2 cpJ9R = 4~rx,
where x is the Euler-Poincaré character of the surface, related to the genus, i.e. the
number o! "handles" h, by h i(2- x).
2D qua.ntum gravity, ~er, is somewhat le.so trivial because large quantum lluctu,...
tions may change the g<lnus Cl[ the surfa<:e. 'Ib calcula.te tbe partition function lt U. thuo
necessary to sum over surfaces <>f ali geoera. This problem ís difficult to solve stn.rting
from t he continuum field the<lr); and surprisingly enough it can be more easlly studied
in a discretised versioo.

A22.1.2 Mo.trix representation of dís.,..tized 2D gro.uity

Following 't Hooft's analysis of the large N limit of SU(NJ gauge theories it has
recogniwd that lf une consirlers the integral

Z (a•, N) =f dM e-NtrV(Ml'
A22.2 Classical and Quantum Grovity 523

in which M is an N x N hermitian matríx and

V(~)= J.' +L a.>.k,

A:~3

then the expansion of F "" ln Z for N la.rge takes the form

(A22.1}

where F, is tbe sum of all Feynman diagrams wíth Euler-Poincaré character X = 2- 2h.
lt follows tha.t the dual of a Feynman diagra.m contributing to F, can be represented as
an orientable surface o{futed topology, the powers o{ a:3,n4, . .. ,ak countíng the number
of triangles, squares,. .,k-gons of tbe surfa.ce (taking the logarithm F <: In Z, yields the
gravity model partition function which represents the sum only over connected surfaces).
H by convention the area of ea.ch triangle, square, ., is a.ssumed to be 1, the power of
"• measures the total area of the surfa.ce. This gives a formula.tion of discrete 2D gra.vity
in terrns of a. distributíon of random ma.trícc;.
In what follows it will be convenient to change the normalitation in the integrand and
consídeJ' instead
(A22.2)

where the coupling constant g plays the role of the cosmological constant. Note that
to desenhe pure gravity only one "'• is needed, for ínstance one can use only triangles.
More general models correspond to the additíon of some new degrees of freedom on the
suriace.
T"" cantinuum limit. The continuum límit is obtained by lettíng the total area of
t.he surfaces tend towa.rds infinity. Surfa.c"" o{ large area are connect.ed with the large
order behaviour of the Taylor series expansion of Fx in powers of g a.nd tberefore to tbe
singularity of Fx(9) closest to the orígín. At g futed, the large N limit selects the surfaces
of the topology of the sphere. lt will be shown below however that ít ís possible to take
N - oo while simultaneous}y taking g to the locatíon of the leadíng síngu!arity 9c, in e
coherent way 110 that ali surfa.ce; of ínfinite area and arbitrary topology contribute.
Rema.rk. In Chapters 37-41 we shall argue that quite generally integrais of the form
( A22.2) are singular a.t g "" O. The result 9c f- O rellectll the property that the number
nf planar Feynman diagrams (selected by the large N limít) grows only geometrícally
insteed of a factorial for generic diagrlll115.

A22.2 The One-Matrix Model

In contrast with vector models where the la.rge N límit can be determined quite generally
· { - Chapter 29), the large N limít of matríx models has been obtained only in very sim-
.ple cases, although the problem is of great potentíal interest (dU.tribution of eigenVB.!ues of
raJJdom harni!tonians, QCD, critica! models on random sur[ac.,., two-dímensiona.l quao-
tum gravlty, strlngs ... ). Problems for which solutions exíst are simple integro.ls, already a
non-trivie.l problem, over one matríx or severa! matrices wíth a one-dimE!nsional nearest
11cighhour coupling and quantum mechanics of large hermitian matrices (only a partia!
·IOIUtion). M06t methods of solution are ba.sed on the possibility o{ fust diagonalizing
· matrices and being able to reduce the matrix integral ·to an Integre.! over eigenvalues.

-----------------------"'~·IA<WI'·t.'l>t.'.J•'J<'I~· JII,J'I,l!. 'l"tr~

524 Clwsical and Quantum Grnvity A22.2

In contrast with veetor problems even then the problem is not soived because a N x N
matrb< hss N eigenmues and the dependence on N is not yet explicít. At thi> point
in the case af integrais two methods oí solution have been used, stoopest descent and
orthogonal polynomi&Js. In one dimensíon (quanturn mechanics), in the sector without
uníta.ry excitations, the hamiltonia.n ca.n be transforrned into a hamiltoniaD of N inde-
pendent fermicws. ln this work we briefty review on!y the salution of the ane-matríx
prob!em, which is relevant for 2D pure quantum gravjty.

Af2.2.1 Th.• on•-matri.z model

We want to evaluate for N large the integral over a N x N hermitian matrb< M

(A22.3)

where d "'' M meo.ns integration wíth a flat measure over the N" independent real V>lri-
ables R.e M;1 , lrn Mi; and V(M) is a general po}ynomia.l potential
2
V(M) = M + z>•M".
1:~3

Since l.he integrand depends only on the N eigenvalues À; of M we factorize the measure
of integration dM into the measure of the unitary group and a measure for eigenvalues.
lntegrating over the unit.ary group one finds

Z f fi 2
dÀi 6 (À) exp [- ~(N/g)V(À;)), (A22.4)

where
6()>.) = I1 (À, -À;). (A22.5)
1Si<i.5N
6(À) can also be written as a Vandermonde deterrninant:
1
6(À) det Àt '

as one verifies ll$ing the a.ntisymmetry of the determina.nt in the interc:lumge of any two
eigenvalues (the normali~ation is determined by comparing leadíng terrns).
This lirst step is crucial for ali methods used up to now. However, in contrast w!th the
vector modelll, we still have to integrate over alarge number ofvariables. Thus additional
technical. steps are ínvolved in the evaluatíon of the Jarge N limit.
The eigenvalues measun!. Equation (A22.4) may be deriwd via the usual Faddeev-
Popov mathod: Let Uo be the unit.a.ry rna.tríx such tha.t M uJ
AUo, where A is a
diagonal matrb< wíth eigenvalues À;. The right ha.nd side of (A22.4) follows by substi-
tuting the definition 1 J dU 6(UMU 1 - A) 6 2 (Ã) (where f dU :-1). We fust perform
the integration over M, a.nd then U deoauples due to the cyclic invariance of the trace
so the integration over U is trivial, lea.ving only the integral over the eigenmues À; of A
with ftat measure. To determine 6()>), we note that only the infinltesimal neighbourhood
U = (1 + T}Uo contributes to the U íntegration, so tha.t

1 =f dUó(UMU'-A) 6 =f dT 5(lT,A])6
2
(À) 2
(À).

Now [T,A];; = T;;{À;- À;), and thus equation (A22.5) foUows (up to a sign) since we
integn>te over both real and imagína.ry parta of the oif d;agona.l T,;'s.

'-'---------------------'~•U.,WI·!.hk' ~'1•'1·~ J
A22.2 Classical and Quantum Gravity 525

A2lU!.2 The larye N limit: steepest descent

The partition functiau ( A22.4) ca.n be considered as tbe partition function o f a g85 of
ch;uges of positions >., interacting via a repulsive 2D Coulomb force generated by the
Vandermonde determina.nt (se<: Secti<m 32.3) which prevents the charges of accumulating
at the mioimum of the ooe-body potential V( A). Th.- saddie poínt equations, obtained
hy varying tbe .\, 's, yield tbe configuratíon of minimurn energy:

2_ :L-1-"" ~V'(.\;), {A22.5)

N #i Ài ->.i 9

This equation can be sol~ed by the following method: We introduce the trace of the
reoolvent w(z) of the matrix M
1 1
w(z}=-
N
tr--
M-z
=-L
1
N ,
(A22.7)

Multiplying equation (A22.6) by 1/(.\;- z) and summing over i, we find

-~"" V'(z)- V'(>.;) . (A22.8)

9 L; z -À;

This equation is analogous lú the Jliccati form of the Sclrrõdinger equation, the wave
function '1/1 being related to w by Nw(z) + NV'(z)f(2g) = '1/1'1'1/1. The eigenvalues Ài
are the zeros of the wave function. In the large N Jimit, we can neglect N- 1 w'(z) (the
WKB approximation). In this limit the distribution of eigenvalues P(À) = j;, L; 6(.l.- À;)
becomes contínuous, and
w(z) =f
p(.\)d.\ .
>.-z
(A22.9)

Note that the normalíli'Btion condition f p(>.')d>.' = ís the analogue of the Bohr-
Sommerfeld quantbation condition. Conversely the eígenvalue density p(À) is extra.cted
from w( z) via the relation

p(À) .2_ (w(z + íO)- w(z- iO)). (A22.10)

2• .-
In the continuum limit equation ( A22.8) beromes
I 1
w2(z) + -V'(z)w(z) +~
g 4 ,R(z) =O,
g
(A22.11)

where
R(z) = 4g f d.\p(>.) V'(zl- ~'(.\) (A22.12)

is a polyuomial of degree 1-2 when V is of degree I. Note that the coeffident of higbest
degree of Ris fixed by the normalization of p(>.) while the remaining coefficients depend
explicltly on the eigenvalue distríbution.
Finally the free energy In Z is abtained from tbe saddle point val~e of the integrand
(A22.4) in the continuum limit:

(A22.13)

---------------------~·IA<WI'·t.'l>t.'.J•'J<'I~·ni,J'I,l!.
526 Clo.ssíc:nJ and Quantum Grnvíty A22.3

The soluoon. The solution 1<> oquation (A22.11) is

w(z) = ~ ( -V'(z) + v'\V'(z)j2- R(z)) . (A22.14)

Generícally w(z) h a!; 2(1- 1) branch points corresponding to tbe roots of tbe polynornial
V'~ - R. Therefore the support of p(À) is fonned of !- I disconnect.ed pieces. In the
simples!. case, when the potential has only one minlmum, we expect a single connected
support and tbus only two branch points. It follows that the polynomial V' 2 -R must
have l- 2 double roots and this yields I - 2 tonditions tbat fully determine R.

A22.3 The Method of Orthogona.l Po!ynomials

The steepest descent metbod allaws a general discussion of the la.rge N limít. lt is difficult
however I<> ealculate tbe subleading orders in the 1/N expansion and therefore to discuss
perturbatíon tbeory to ali orders. We 11ow prese11t anotber method that allows us to
recaver prevjous results and to extend them to all orders in 1/N.
This alternatm method for soMng (A22.4) ma.ke6 use of a set of polynomia.ls P,.(.>.),
ortbogonal with respect to tbe measure

(A22.15)

Tbe normafu.ation of tbe ortbogonal polynomía.ls P.. is given by having tbe leading term
P,.(Ã) "' -\" + ... , hence the constMt s., on tbe r.h.s. of (A22.15). Due to tbe relation

(A22.16)

(recall that arbitrary polynomials may be built up by a<lding líueat combínations of pr...
ceding colurnns, a procedure that leaves the determinant. unchanged), the polynomiall;
P, can be employed 1<> sol"" (A22.4). We substitute the determinam det P;- 1 (>.;) =
2;(-1JxD 1 Po.-I(À•) foreach ofthe ll(Ã)'s in (A22.4) (where thesum is ~ permu-
tations ik and ( -l)x is the signature of tbe permutation). Tbe integra.ls over indMdual
À;'s factorize, &Dd due to orthogonalicy the oaly contributions are from terllll3 witb ali
P;(À;)'s p&ired. Tbere are N! such terms so (A22.4) reduees to

Z= f IIl
dÃt .-(N/•JV(~,) 2::<-W<-1)"' II Pi.-!(À>) II P;;-l (À;}
11'.~' A: J
N-1 N-1
= N! IJ Bi = N!s~ IJ f[-k, (A22.17)
t-<1 k...:.l

where we bave defined /1 =:= ••f

$t_ 1 • The solution of tbe original matrlx integral is thu&
reduced to tbe problem of determíning the normalizations ••· or equivalently the ratios
f•·
ln tbe noive large N límit, the re5cnled index kjN bocomes a rontinuous variable I
that runs from O to 1, and J.JN becomes a continuous function J(t). In thls límit, tbe
free energy (up to an irreleV!Ult addltive constant) reduces 1<> a símple one-dimensional
integral:
1
_!_lnZ=
N2
2._ L(l
N t
k/N)mft- { dt(1-t)lnf{t).
Jo0
(A22.1B)

---------------------~·V~.>Ln·!.hk' ~,1,,.~ J
A22.3 Clas$ical and Qunnt:um Crotrity 527

'lb derive the functíonal [orm for J(t), we assume for 5Ílnplícity that the potentia.l
V(.X) in (A22.15) is """"· Sine<> the P,'s from a complete set of ba.sis vectors in the
spo.ce of polynomials, it is clear that >.P,.(>.) must be expressible"" a linear combination
of lower P,'s, >.P,.(,\)"' I;:,;~ c,;P;(À) (with c,; s;-' f e-V »P., P,). Hawever, temiB
proportional tQ P, for i < n -1 vanish since J .,-v
P,. À P, = O (recai! ÀP; is a polynomial
of order at most i+ 1 ro is orthogonal tQ P,. for i+l < n). MorecM!f the term proportiona.l
to P,. also vanish<lS due tQ the ~umption that the potential is even, f.,- v À P,. P,. = O.
Thus the orthogonal po!ynomials satisfy the simple recursion relation,

(A22.19)

with r,. a scalar roefficient independent of À.

By considering the quantity P,.>.Pn-1 with ,\ paired alternately with the preceding or
succeeding polynomlal, we derive

f dÀ .,-(Nfg}V P,. .\ Pn-1 r,. s,._, = Sn.

This shows tbat the ra.tio f,. ... s.,j ,,_ 1 for this simple case is i<lentically the coefficient
defined by equation (A22.19), f., =r,.. Similady if we pair the >. in P,; À P,. before and
aherwards, an integration by parts yields

ns., f d,\ e-(Nfg)V P~ À P,. = f e-(N(g)V P,; r, P,._ 1

=r,. f dJ\ e-!N(g)V [Nfg)V' P,. P,.-1, (A22.20)

a relation that will allow us tQ determine recursively r,..

A22.3.1 The large N limit

Our intent now is to find an express.ion for f,. = T, and substitute into (A22.18) to
calculate a partition function. For definiteness, we ta.ke ss an example tbe potential

V(.\)= p 2 + p•, <=> V'(.\)=.\+ 2À3 • (A22.21)

The calculatíon ofthe r.h.s. of equation (A22.20) involves then expanding (Ã+2Ã3)P,_l
nn the basls of polynomials P,. These may be a.cromplished by using tepeatedly equation
(A22.19). For the potential (A22.21) this gives

g(n/N) =r,.+ 2T,.(rn+l +r., +r.. -tJ. (A22.22)

As explained before equation (A22.18], in the la.rge N lirnít n/N can be treated a.s a
r.ontinuous variable t n/N, and we have r, >-+ r(t) and rn±l r(t ±e), where
1 1/N. To leading order in 1/N, equation (A22.22) reduces to
gl.,.. r+ 6r2 = W(r). (A22.23)

lt can he verlfied, insertíng the oolution of equation ( A22.23) in to equation (A22.18), that
tbe result obta.lned by the method of steep€5t descent is reoovered. a.t this order, however
lhe orthogona.l polynomíal method allows to calculate much more easlly the higher arder
torrections.
We note that equation ( A22.23) he.s no solution when g is smaller that the winimum
!}c of W(r), where Yc -1/24 for t 1.

----------------------·~·IA<WI'•t.'l>t.'.J•'J<'I~·ni,J'I,l!.
528 Glassical and Quantum Gromty A22.3

.4.22.3.2 Large N and doubl• >caling limit

Let us canstruct surfa«.s in terms of squares. The corresponding potential ís then the
potential (A22.21). The la.rge N limít, i.e. the sum of ali contributians o[ surfa<oeS of genu.s
zero (planar surfaces) can be directly deduced trom the results of SuJ:.,ection A22.2.2. In
particular we have noted the exist.ence of a criticai value g, -1 /24. The limít g - g,
thus corresponds to the continuum limít. At thi$ poínt the second derivative of In Z has
a sque.re root singula.rity a.nd thus the singular part of In Z beha.ves like (g g,j51 2 .
The ali gonu.o partition functiOfl. We noW search for another solutían to equation
(A22.22) e.nd its genera.lizatians tha.t describes the contríbution of ali genus surfaces to
the partition function (AZ2.18}. We sha.li retain higher arder terms in 1/N in (A22.22)
so that e.g. equation (A22.23) iJ>BWad reads

gt = W(r) + 2r(tl{r(t +o:)+ r(t E)- 2r(t))

2
= g, + ~W"l..=r, (r(t)- r,) + 2r(t}(r(t + .:) + r(t e)- 2r(t)) + · · · (A22.24)
A6 suggested at the end of the preceding subsectian, we shall simultaneously let N ~ oo
and g- g, in a particular way. Since g g, hss dimension ~engthj 2 , it 1s convenient to
introduce a pa:rameter a with dimension length and let 9 g, = ,.-•l•a1 , with a. -. O.
Our ansatz for a coherent large N limit will be to take E 1/N =
a."f• so that the
qunntity .,-l = (g g,)•f•N rema.ins finite as g _., Yc and N- ao.
Moreov<'lr since the integral {.422.1 B) 1s domínated by t nea:r 1 in thiE 1\mit, ít i:;
conveníent ta change vatiables from t to z, defined by g, - gt = a.~ z. Our scaling ansatz
in this region ls r(!) r,+au(z). Ifwe substitute these definiti<>ns into equation (A22.23),
tbe leading terms are of arder a' and result m the relation u2 - z. To ínclude the hlgher
derivative terms, we note that

r(t +E)+ r(t- ~)- 2r(l) ~E':~= a : :2au(z) ~ a 2u",

where we have used E:(âjôt) = -ga.11'(8/âz) {which follows from the ahcm: change of
variableo; from t to z). Substituting into (A22.24), tbe vanishíng of the coefficient of a 2
implies the differentía.l equation
(A22.25)
(Bfter a suitable res<:alíng of u and z). The second derivati..,.., of tbe partition function
(the "specific heat") has a leading singular behaviour given by /(1) with t = 1, and thus
by u(z) fur z (g-g,}fa2 ,.-•!•.The solution to equation (A22.25) characterW.. the
behaviour of the partítion function of pure gravity to ali arders in tbe genus expansion.
(Notice that the leading term ís u ~ zll 2 so Bfter two integrations the leading term in
In Z is ~f'= ,.- 2 .)
Equation (AZ2.25) ís known in the mathematical \iteratute as the Painlevé I equa.-
tion, The perturbative solution in powen! of z-•1 2 = .,2 takes the form u = z 1f 2 (1
:L•=• ,.,.-••1
2
), where the u• e.re a.ll posítive. The GOlution of the Paínlevé 1 equa-
tion seems to define 2D quantum gravity beyond the perturbative topologica.l expansíon.
However, it can be verífied tbat there exists no real solution to the Painlevé equation
satisfying the proper boundary condítions, This property can be rela.ted to the non-Borel
summabílíty of the perturbative expansion. Its physical interpretation is that pure quan·
tum gra.vity il; unstable, because the number of surfa.ces increases too rapidly with the
geous.

. , . , . _ ,_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 'M~,VhLI·!.hk' ~'1•'1·~ J
23 CRITICAL PHENOMENA: GENERAl CONSIDERATIONS

Chapters 23-36 are devoted to the study of second order phase tra.nsition.s and the
determination of their universal properties througb the use of renormalization gToup
methods.
Most of the transitions we shall consider have the following character: a system of
spíns on & lattíce or a system of particles in the contínmnn interacts through short range
forces. At fixed densíty, as long as the system is containad in a finite volume, ít is
ergQdic, í.e. any finite regíon of available ph...., spo.ce has a non-vanishing probabílíty to
be explored in the course of time (in the sense of Appendix A4 the system is connected).
HOWlM!t in the ínfinite mlurne límit, depending of the value of some contrai para.meter
which usually is the temperature, the system either remains ergodíc, or experiences a
breaking of ergodicity. In tbe latter case, phase spo.ce deromposes in disjoint sets. When
the system is ínitially in one of the sets, it remalns at !ater time. For example for
lsing-like systems, the two sets correspond to the two possihle values of the spontaneous
magnetization.
In the simpl- lattioe models " tnwsfer matrlx can be delined. The thermodynruníc
limit is related to the largest eigenvalue of tbe matrlx. In the "ergodic" pbase the
corresponding eigenvector is unique while in the non_,-godic phase it is degenera.te.
Our goal is to analp.e the bebaviour of tbermodynamic quantities in tire neighbour-
hood of the phase transition, in particular their síngula.rltíes as function of the oontrol
parlllll-f-
For tbe simplest systeiDB, it is pOI!Sible to find local observables whose values depend
on the pbase in tbe ...veral phase region. We cal! one of such obse!"YRbles the order
pammeter. lt is for e><ample the spin in ferromagnetic systems.
In the one phase region, for syBtelll5 wlth short range interactions (for example decretl$-
ing exponentially or faster) the connected correlation functions decrease exponentially
with dista.n<:e when two non-empty sets of points are separated (this is directly related.to
Lhe cluster property of Section 6.L2). We cal! e<>rrelation length the inven;e of tbe small-
est decny r ate of oorrelation functions (the smallest physical mB&'l in the partícle physics
sense). For reasons which wíll beoome clearer !ater, we study only these special phase
transitions for wbich tbe correlatlon length, measured in units of the microscopic scale
(range of for<es, lattice spacing), diverges at the transitíon point. F<lr such systems it
will be sbown that, near the transition point (tbe criticai temperature for example), some
properties of tbermodyn.amic functíons are univer.al a.nd furthermore can be described
by euclídean field theories and renormalization group equBtions.
Severa! chapters will be devoted to the derivation of universal propertiet; o{ correlation
{unctions, but Iet us bem empbasize already the deep connection between non-mean field
universa!ity in phase tnullõitions with divergent correlation lengtb, and renormalization in
Field Theory. We h&ve shown that renormalnable local field tbeories are sbort distanoe
insensitive, in tbe sensethat, although they are not finite in the Jimit in whicb the UV cut-
off becomes mucb la.rger than all masses and momenta, aft.er an infinite renormaliu.tion
8 uniqne (up to finite renorrnalizations) renarmalised field is obtained. The translation
lnto the phase transition lwguage is sirnple: unlversality emerges in a regime in which
tbe correlation length and ali distances il.te much larger than the microscopic s~ale which

~=---------------------,~·IA<Wf',t.'l>t.'.J•'J<'I~·ni,J'I,l!.'1"t1f~
530 Crítica/ Phmameno.: Generol Con.riderotion.s 23.0

play» tne role of the inverse of the cut-off. However this universs.lity is non-mean field
Iike (at le..st in low dimensíons) because the microscopic scale cannot be completely
eliminated; degl'@eS of !reedom on s.ll scs.les remain coupled.
Locality of field theory ís tbe analcgue, as we shal! see Inter, of short range forces. It
remains to explron the role in phase transitions of the particular dimension in which the
fteld tbeory is just renormaliznble.
Since renonnalization group equations are a direct consequence of renormalization
theory, as we have shown in Chapter 10, we sball not be too surprised to discover that
they a.ppea.r and play a centrs.l role in the tbeory of phase transitioru;.

Tenninolo911. 'Ib describe criticai pbenomena, it has become eustomary to use the
language of magnetic syst<Jms. Although such a presentation certalnly helps our physical
lntuition, mo.ny s)'Stems to which the theory applies are non-magnetic. This language ís
therefure, in a sense, almost as abstra.ct as the langua.ge of qua.ntum field theory. Since we
shalleventually sbow that ali tbe universal problems in the theory of phase transítions can
be f<lmmlated in the lo.nguage of euclidean field theory, we could have stayed completely
within the úamework developed in the previous chapters. However, sinoe we also wish
to introduce field theory methods to readers with a background in statistkal mechanics,
we sball often use tbe statistical and magnetic langu age. In previous cha.pters we h ave
already called correlation functions wba.t often is called imágÍnary time Green's functions
or Scbwinger's functions. A short glosaary is then:

Euclidean field tbeory Classical magnetic systems

Fields Spin Vllriables
Source Magnetie fie.ld
Euc.lide.w action Con.figuration energy, barni/tonian
exp(-AtH) Dausfer matriX in the case 'of
firuu range inter=tions
F\lnctional integral Sum over spin configurations
FieJd """"umexpectation va.lue Magnetízalíon
Zero momentum two-point functioD Magnetic susceptibility
Physie&J mass ID~ correJ.ation Jengt)J
Massless tbeory Critical theory
Generatillg func:tional Z of Partit.k>n function in a space-
correl4tion fundions dependent m&gnetic iield
Generatfug functionaJ W of Fne e.nergy in a magnetic fiald
oonnect.ed correlation functlons
GeDeratiDg /imctimlal r of Thermodynamic potential, func:tion
proper vertices oi the m!lglletization

Let us now briefiy explain tbe organization o f the chapter.; devoted to criticai phenom-
ena:
In Cbit.pt.er 24 we study fet't'Omagnetic systetll!! within mean fie!d theory and sbow
how in dimension leso than or equal to four, summation of Jeadfug dhtergeDt correc-
líons to mea.n lield theory !ea.ds to the coDtillnum 1/14 field thoory. In Chapters 25-28
we then introduce renormalization group ideas. We obtain a complete da.crlption of
uniYel$ll propert.ies of second order pbase transitions in ferromagnetic systems near four
dímenaions a.nd calculate the le3ding correctlons to the universal hehaviour ín the criticai
domam.
23.1 Critica./ Phenomena: Generol ConsideratioriS &31

In the Cha.pteiS 29 and 30 we expand the N~vector model for N large a.nd at Jow
temperature and verífy RG predietions. The specia.l roles of two dimensions and the
abelian 0(2) model l!'merge. The Chapters 31 and 32 are thus devoted to a technique
spedfic to two dimensions, bosonization, which gives information about the properties
of severa.! model.s líke the sin.,.Gordon, Coulomb gas, Thi:rring or Schwinger models, a.nd
its application to the f!Uilous Kosterlitz-Thoult!ss ph!l.S€ transition.
In Cha.ptli'rs 33 and 34 we try to gain some non-perturbative insight in the properties
of gauge theories through the use of lattice regularization and RG techniques. We derive
the conditíons for large moroentum asymptotic freedom in four dimensions. Finally in
Chapter 35 we examine Critica! Dynrunics, i.e. the time evolution of criticai systems
whlle Chapter 36 deals with linite size effects.
However before we discuss phase tram;itions with more sophlsticate<l techniques, we
want to recall a few properlies, from tbe point of view of phase transitions, of simple
ferrornagnetie lattice models. ln systems with linite rBilge interactions a transfer matrix
can be define<l. We first examine its properties in a linite volume. In tbe infinite "olume
límit, low a.nd hlgh temperatun: eonsiderations provide convincing evidence of the exis-
tence of phaae transitions In Ising-like systerns, We relate the notion of order p8l'ameter
to c)uster propertíes in the low tem perature broken phase. We show in a simple example
tha.t phase transitions indeed oonespond to breaking of ergodicity. We linally extend tbe
analysis to ferromagnetic systems witb oontinuous symmetries.
The appendix contalm; a brlef discussion of quenche<l disorder.

23-1 Pha5e Tr1U15itlons and Tr-ansfer Matrix

In Cbapter 2 we bave eonstructed a patb integral representation of the partition function

in Quantum Mechanics by using the identity:

e-f'H = (e-' 11 ) 1 (23.1)

SimilaJ'Iy, In the c""" of lattice models with finite range lnteractions (for example nearest
neighbour ínteract.ions), distinguishing one "time" direction on the la.ttice, it is possible
to write the partition funetion Z with períodic boundary conditions:

Z=trT, (23.2)

in which I is a measure of tbe lattice size in the time direction. The matrix T is tbe
transfer matrix and plays tbe role of e-•H. For example in the case ol the lD Ising
model with nea.rert neigbbour (n.n.) interactions and in a magnetic field h the partition
funetion is:

Z L
{S;=:l:l}
exp [fJti=l
(JS,S,+ 1 + hS;)], (23.:>)

and the tran.sfer matrix is a 2 x 2 matrix wbich can be written:

{S'IT\S) = exp ~(JSS' + ~ (S+ S'))] (23.4)

In wbat follows we shall reBtrict oume!VBS to isotr<>pic interactions on a bypercubic la.ttiee

and therefore we ca.n a.lways choose the ttansfer matrix symmetrlc (this oorresponds to
hermitian hamiltonianB in qua.ntum mechani<>l).

-----------------------"'~·IA<WI'·t.'l>t.'.J•'J<'I~·ni,J'I,l!.
532 Critiwl Phenamena: General Consídemtions 23.1

Note the difference between the roles played hy the inverne t.emperature (:J in Quantum
and Cla.ssical Sta.tistical Mecha.nics. The parameter fJ of quantum mecha.nics is the
analogue of the si2.e of a classical system in one a.dditiona.J dimension. ln particular the
)arge /3 (í.e. zero temperature) limit of quantum mocha.nics corresponds to the large l
limit.
Since futed temperature corresponds to fixed size in one dimension, it will become ap-
parent later that from the point of view of phase transit.ions often quantum lluctuations
are irrelevant. One exception is provided by :zero temperature quarrtum statistical sys-
tems in d ~ 1 dimensions which sbare many properties with cle.ssicaJ statistical systems
in d diroensions. At the same time, as a consequence of tbe discussion of Section 2.5,
we see tbat Lhe infinite (:J quantum correlation functions are Lhe analogue; of the ínfinite
volume statistical correlation functions.
In tbe t~rroodyna.mic limit, I goes to infinity a.nd Lberefore tbe partition function is
related to the la.rgest eigenvalue of the tra.nsfer roatrix. The corresponding eigenv..ctor
plays t~ role of the ground state of a qu&ntum bamiltonian.
Simple proper1ie.s of the tmR!!/er matrix on a finite trnnsverse lattia. lf the size of
the l..ttice tra.nsvetse to the time axis is finite, using the positivity of the coefficients or
tbe transfer matrix and adaptlng tbe arguroents of Appendix A4, we can immediately
derive an important property of its spectrum. We sha.JI glve all a.rguments in the case in
whlch the spin distribution is discrete aod the transfer ma.trix is a. finite matrlx but t~
arguments generalize to continuous spin dlstríbutioll!i.
Since T is real symmetric, its eigenvalues are real. Let ). 0 be tbe eigenva.\ue with
largest modulus and lO) Lhe corresponding eigenvector:

TIO)= >..,lO). (23.5)

Let U>l cal\ Tab the matrbt elements of the tra.nsfer matr!x and v~ the components of tbe
oigenvector 10) and rewrite equation (23.li) in component form:

(23.6)

Thlcing tbe scalar product with the vector lO), we derive:

(23.7)

The deno~inator of the r.h.s. is a wm of positive tenns and the matrbt elements of T
are posi~ Taking the modulus of equation (23.7) we obtaln an inequality:

(23.8)

The r .h.s. of the inequality is the average of T taken wítb the vector \v~]. A strict
inequality would imply tbe existence of at least one positive eigenvalue huger than 141
in contra.dictíon with tbe hypothesis. Equality combined witb the property tbat T has
stril!tly poslti"" coefficients imp\les that all components of the vector 10) cau be ohosen
non-negative. Tben, according to equatíon (23.6), Ão is positi"" and all componerrts of
10) are strictly positlvé. It follows that t~ eigenvalue cannot he degenerate because two
vectors with strictly positi"" coefficients cannot he orthogonal. l.<Jt us summarize tbis
result:

, _ ,_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ ,,~,1-"I.,WI·!.hk' ~'1•'1·~ J
23.2 Cnticül Phenomena: Genernl Consideratians 533

Because the trans!er mattíx is real eymrnetrk and has strictly positive coefficíents, the
eígenva.lue wíth largest modulus is positive, the corresponding eígenvector is unique and
has strictly positive components.
ln the large I (i.e. large time) límit, the free energy W ís given by:

(23.9)

Since no crossíng of leveis can occur, ).o is a regular function of /3. For exarnple for the
Ising model one findE:

(23.10)

The connected 2-point spin correlation function behaves at large time separation as:

(23.11)

in which li) is the eigenvector correspouding to the second largest eigenvalue ia modulns
).,, which we have here assumed to be p!ISitlve, ande is given by:

(23.12)

Finally the matrix S is a matrix diE•gonal in the spin represento.tion whlch corresponds to
a multlplication hy the spin var:iflble. We have e.ssumed that its matrix element (O ISII)
does not vanish. Again in the example of the lsing model one finds:

[(sinh "'h+ e-4fJJ)''']

2
ç• = 2tanh- 1 "
coshflh
• (23.13)

We conclude from thls o.nalysis tbat in a spin model with finite rll.l1ge intera.ctions Ellld
diocrete dôstribution no phase tnmsition can occur, the free energy is a regular functíon
e
of the invetse temperature /3 and the correlation length remaim finite except for f)
infinite. These reomlts ca.n be generalized to short range ínteractions o.nd continuous spin
distributions.

23.2 Tbe lnfinite Transverse Slze Llmit: Ising-Like Systems

When the transverse size becomes infinite new phenomena may appear, whlcb we sball
examine first ou the exRlllple of tbe d dimensional lsing model with nearest neighbour
(n.n.) intero.ctíons. The partitlon funct!on can be written:

(23.14)

Using bracket notatíons to represent mo.trix elements, we find for the trlll1Sfer ma.trix:

({S~} ITI{Sp}) =exp[fjJC~-• Sps;+ ~ (SA,+S:,S~,l)]· (23.15)

p.p•EZd-t

-----------------------.,,~,fA<Wf',t.'l>t.'.J•'J<'I~·ni,J'I,l!.'l"t.ll~
534 Críti<:al Phenomena: Genem! Consídemtions 23.2

The lsing model is -ehara.cterized by a. discret€ symmetry- corresponding to spin reversal.

Let us coll P the corresponding matrix. In bracket notations:

(23.16)

We shal\ study the spectrum of the transfet matrix on a finite transverse lattice of
linear síze L, for L large. For L finíte, the general result.s on ma.trices with positive
elements apply. The matrix P commutes with the tnmsfer matrix, a.nd therefore P
and T ca.n be diagonalized simultaneously. The eigenvectors of T are eigenvectors of
P. The eigenvalues of P are ±I. The eigenvector of!argest eigenvalue of T has positive
components. Equa.tion (23. I 6) shows that P does not change the sign of the basis vectorn.
Therefore the eigenvector 10} with tbe \argest eígenvalue is an eigenvector of P with
eígenvalue +I,
PIO) =10) (23.17)
Let us now examine the infinite L limit a.t low and hlgh tempera.ture, i.e. a.t high and
low fj.
High temperature. At high temperatura (/3 - 0}, ali matri.x elements of T become
equal and T becomes a projector onto the eigenvect.or 10) which has equal components
on ali spin configurations. Ali eigen,'lllues but one wnish, thus the correla.tion length
vanishes. This property is independent ofthe volume and therefore previ.ous result.s apply
even for L infinite.
Low tempernture analysis. At low tempera.ture the domina.nt configuratíoru; correspond
to ali spins aligned. Let us call \+) aJJd 1-) the two vectors corresponding to ali spin up
and down respectively. At f3 strictly infinite both are eigenvectors o! T corre:;ponding
to the twice degenerate largest eigenvalue. The fact tha.t the eigenvalue ís degenerate
does not contra.dict the general result of Section 23.1, since at -rero temperature, if wc
normalize the largest matrix elements of T to 1, ali matrix elements but the diagonal
elements (-ITI-) and (+ ITI +) vanish, and the general arguments of Section 23.1 no
longer apply. :Note that I+} and 1-) are not eigenvectors of P since:

Pl+) H· (23.18}

At low but finite temperature the eigenvectors can no longer be exactly \+) and 1-) bu1
have alro n~ily components on conligurations in which a finite number of spins haw
been llipped, as low temperature perturba.tíon theory shows. However this still allows
two degenerate eigenstates dose to I+) a.nd 1-) and exchanged by P. What is relevant h
tbe large L beha.viour of the matrix elements of T connectlng I+) and 1-), more precíscly
the ratio 6:
i+!TI-) (23.19)
(+ITI+)"
At low temperature this ratio is rela.ted to the difference of the minimal energy of tlw
configurations corresponding to the two different bounda.ry conditions (b.c.). The cost ill
energy of imposing ''twisted" b.c., i.e. spins up on one síde and spíns down on the othe1
síde, is proportional to the area of the surface on which the spins are llipped. Therefmr
6 behaves like:
(23.20)

~· Jlt.J•t.l:'. '1'\ii'-=O_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _,,~,u.,wt.!.hk' ~'i•'t·~ J

23.2 Criti=l Phenomeno.: Generol Considemtirm• 535

The eigenvectors and eigenva!ues are qualitatively given by diagonalizing a 2 x 2 matrb<

T in the {I+), 1-)} subspace:

(23.21)

Two different situations arise depending on the dimension d of space.

{i) d"' 1
Then li is finite and the eigenvector lO) is the linear combination

lO) l+l+H,
which is also an eigenvector of P with the eigenvalue +L
(ii) d > 1
In the infmite volume limit, ó vanisbes, the highest eigenvalue ís twice degenerate. The
finite size correlation length diverges as:

(23.22)

Clearly in tbe infiníte volume limit no an~ytíc contínuation ís possible between the high
and low temperature sítuetions, Md lherefore tbermodynamic quantítíes must have at
least one singularity in {3 at some finite value {3<. We shall argue in Section 23.4 tbat at
low temperature a breaking o[ etgodidty occun; and thus f3, also corresponds to a pha.se
transition in the dynarnical sense.
Remarks
(i) Thís analysis of the infinite volume limit ís qualitatively correct in the whole low
temperature ph&se. At f3. the situation is different; an infiníte number <>f eigenvalues have
the same infinite volume lírnit à 0 • This situaiion wil! be studied in detail in Cbapter 36.
(ü) We bve bere analyzed a.lattice model, but we willshow in Section 40.! that in lhe
case of models defined in contínuum sp~, instantons lift the degeneracy oi the ground
state in the semiclassicallímit. With the conespondence:

f3 ,.__. Tí- 1 , J ,__... instanton action,

the a.nalysis is then exactly the same.

(iii) From the previous analysis we derive a. criterion of spontaneous symmelry breaking
(SSB). We consider the ratio r:

. trPT1
r= t~~ trTl '
(23.23)

in which P in general will be an operation of lhe syrnmetry group.

In the symmetric phase, the ground state of tbe transfer matrix ís invariant under a.
group operation, therefore the retio r is L
On the contrary, if the sYmmetry is spontaneously broken, P exchanges the various
ground states and thetefore r vanishes in the iufinite volume limit. In the example of the
lsing model one finds at low temperatme:

The advantage ofthe coru;ideration of the ra.tio {23.23) is that, ifwe la.ke! =L, the linear
•ize o f the lattice, we have j ust to ca(culate the ratio of two partition functions on a d

- - - - - - - - - - - - - - - - - - - - - - -. . ~·IA<WI'•I..'i>t.'.J•'J<'I ~· JII,J'I,l!. '1"t11i"'-=

536 Criticai Phenomena: General Considerations 23.3

dimensionallattice of linear size L. with different boundary condítions: the denominator

corresponds to periodic b.c. in tbe time direction, the nurnerator corresponds to twisted
b.c., i.e. the configurations on the two bounda.ries differ by a transforrnation of the
symmetry group. ln the case of the lsing model, twisted b.c. are antiperiodic b.c .. For
I L this ratio naturally incorporates the condition that the thcrmodynamic limit ís
taken by sendíng the sizes in ali dimensionB to infinit:y in the same way. ]t represent.s, as
we shal! show below, tbe probability in some dynarnics that. the s:ystem can evolve lrom
ínitial condítions in whích a!most ali spins are up to a configuration in whích most of the
spíns are reversed.

23.3 Order Parameter and Cluste.r Properties

When tbe ground state lO) (vacuum state in the sense of Partlcle Ph:ysics) is degenerate,
the determination of the infinite volume correlation functions becomes a subtle question,
which depends explicitly on the way the infinite volume limit;,; reaclled; in particular it
may depend on boundary condítions. This sensitivity to boundar:y conditíons is another
characteristic of the severa! phase regíon.
Let us examine in this case the prob\em of cluster properties of correlatíon functíons.
We oonsider the two phase region of an lsing like system and call I+), 1-) the two
ground states whích are exchanged by P {equation (23.18)) a.nd orthogonal. Any vector
la) of the form:
lo)= cosa\+)+ sina 1-), (23.24)
is also an eigenvector of the transfer matrix with the same eigenvalue. Let ·us assume that
the boundary conditions select tbe vec:tor lo). Correiation functions a.re then obtained
by calcuiating averages of matrices in tbe state ta).
lt is intuitíve tha.t the spin s.
at síte 11 ís, according to tbe definition given ín the
introduction, a.n arder pa.rameter. The correlponding matrix S, whose matrix elements
between two spin conligurations are

is odd under reflection:

(23.25)
ln the high tempeTature phase, in which the vacuum state satisfies:

I
PiO)= lO), (23.26)

tben:

(O]S]O) -{O\P- 15P\O)

In the law temperat ure phase let us define:

-(OiSIO) =0.

(+ISI+)=m,
(23.27)

(23.28)
l
f
in which m approa.ches 1 in th" lsing model in the zero temperature límit. Then; l
(23.29)
'
1

~· Jli.J•t.l:'. '1'\li"'==-O_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ ,,~,u.,WI·!.hk' ~'1•'1·~

1
]J
23.3 Criticol Phenotnena: Geneml Consideratwns 53 i

We also have:
(+ISI-l = (-IP- 1 SPI +) =- {-ISI+l. (23.30)

Since S ís a symmetric matrix:

(+ISI-) =o. (23.31)
The essential property used below ís tbAt S, restricted to the {1+),1-)} subspoce, is
diagonal in the {I+) .I-)} bosís.
lt follows from equations (23.28-23.31) in particular:

(o!Sin) = mcos2o, (23.32}

{a [SJ ,./2 +o:)= msin 2a. (23.33)

Except for a= 1f/4 (mod 71"/2), the average ofthe spin doe!õ not vanísh and thís char-
acterizes the severa! pbase region.
Let us calcu!ate wha.t one would naively expect to be tbe connected 2-point oorrela.tion
functíon of tbe spin,i.e. the 2-point fubction of 8- < S >, for nvo points separated by a
distance I in time but at tbe same posi.tion in tbe transversa direction. It is given hy:

w(2J(I)
m Co.; 2o) T (S- mcos2a) lo)
(23.34}

The transfer matrix T projects onto the ground states for large 1:

T 1 = >.~ [0+) (+l +H (-1) +o (e- 1n)].

(23.35)
=~[lo:) (oi+ 1•/2 +o} (.."/2 +a: I+ O ( .,-l/t) ].
lt then follows:
(23.36)

lt ís only for a = nw that the correlation functioru; sa.tisfy cluster properties. The
corresponding eigenve<:tors are then I+} and 1-) which are exchanged by P. According
to the definítion of Chapter 13, the reflection symmetry i.s spontaneously broken.
:Note that correlatíon functions C8lcula.ted hy summing over ali configura.tiollB (the
analogues of expressíons (2.59,2.63)) wrrespond to average over the two ground states
and do not sa1ísfy cluster properties. Tbis problem can be solved by addíng to the
ronfiguration energy a.n infinitesimal intera.ction terrn coupled to the arder para.meter
which favours one phBSe and lifts tbe degenemcy. For example in a spin system, to
generate correlatíon functions wbich satisfy cluster properties in the infinite volume lírnit,
it is possible to start wíth a finite ,·olume, to add a constant m&gnetíc field, to take lin;t
the infinite volume límit, and then tbe zero field límit. The ground state, I+} o r 1-} and
low tempera.ture, is chosen hy the sign of the m&gnetic fie!d.
In Subsection 6.1.2 we have added a small constant part to the sources whlch allowed
us to define the generating functional of connected correlation functions even in the
degenerate case.
One may wonder about the physical int,erpretation of this procedure: in next section
we s.hall argue that a system in the spontaneously broken phase is no longer ergodic.
Once prepared in one phase, it remains for ever. Therefore in the two phase region of
a spin system one should only average over configuratioru; which fluctuate a:round the
conligurations with ali spins up or aU spins down.

----------------------~·IA<WI'·t.'l>t.'.J•'J<'I ~· JII,J'I,l!. 'l"t.ll~

53S Criticai Phenomena: General Consídemtions 23.4

23.4 Stochastic Processes and Phase Transitions

To construct a stochastic process wbich converges toward the equilibrium distribution of

the Ising model we can use detailed balance (see Appendix A4) ""d set for the transition
probability p of a configuration {S;} towards a oonfiguration {S;}, r belonging now to a
la.ttice 0 of volume Ld in zd:

in which
e-~!E(s;)-E!S.)]
I otherwise,

E (S, J is tbe configuratlon energy:

for E (S,) < E(S;),
(23.37)
I
E(S) = L
n.n..
-Js,s; (23.38)
:r,r'EflCZ"

For the argument whích follows the precise description ofwhich configurations are directly
connocted by the matrix p is irrelevant provided the system is globally connected. The
relevtmt property is that the prohability to go from a configuration to a conliguration of
hígher energy is, at low temperature, of the arder of .,-~t>.E, in which l!.E is tbe energy
dilference.

+++++++++++++
+++w.++++ -+++
+++ - - - - - - + + +
++ - - - - - - - + + +
+++ - - - - - - + + +
+++ - - - - - +++
++++ + ++++
+++++++++++++
Fig. 23.1

Therefo~e, at low temperature, if we start from " configura.tion in which ali spíns are
+ l for exàmp!e, the probnhility of creating a bubble of minus spíns is proportíonal to
e~flJA, in whlch A is the area of the bubble surfa«< (figure 23.1). lf a la.tge fluctuation
creates a bubble whicb lills ba.lf of the total volume, then of com·se there ís "prohability
1/2 that aft.erwards ali spíns become equal to -1. Therefore the probabílity of reversing
most of the spins is, nt low temperature, related to the e.ma of a surface which divides
the volume 11 into two halves. Since n has linear size L, this probability is of the form
e-.aLd-t.

=
For d 1 the system is ergodic, and, as we have sbown, no phase trnnsítions .:an occur.
Fbr d > 1 the same mecha.nlsm whích allows the existence of several pbascs nt Jow
temperature is responsible for the breaking of ergodicity.
Note that the argument generalizes only to local dynamícs, i.e. dynamlcs in which the
probability of cha.nge of one spin on the lattice ís only ínftuenced by its neigbbours (for
more details see Appendix A36.1.3).

~· Jlt.J•t.l:'. 'F\r'-==----------------------llilllillft!-;,;,; ,.'1,1·,·} j

 Critical Phenomcna: Geneml Coruiderntiow 539

23.5 Continuous Symmetries

Let us now briefty discuss a model which has a continuous symmetry to here exhibit
some different-e from the c~>Se of discrete symmetríes. We again consider a dassical
spin system, but the spins Sr are N-component vectors interacting through a two-body
n.n. ferromagnetic interaction. The rnodel then has a contlnuous O(N) symmetry. The
partition function is:

(23.39)

Such a mo dei haB already been proposed as a regu!a.rization of the O( N) non-linear

o-rnodel in Chaptet 14.
At high temperature (fJ -> 0), as in the case of the Ising model, the ground state of
the transfer matrix h os uníform components on ali configuration vectors, and ís therefore
invariant under the transformation• of the O (N) symmetry group.
To undeT6tand the structure of low temperature pha.=, we now use the equivalent of
ratio (23.23). The symmetry operation is bere a rotation R(a) of angle <>. Defining:

ZL (<>)
TL = ZL(O)' (23.40)

we examine the behaviour of TL for fJ large in the large L limit. The partítion function
Z L ( <>) is the partition functlon on a d dimensionallattice, with twisted b.c. In tbe time
direction we impose:
(23.41)
For convenience we take periodic b.c. in ali other djmensíoru;.
At low temperature, the conliguratio!IS wíth mínima! energy correspond to take the
spins aligned in d-1 dimens;oru;, and rotating by an angle a/ L between two adja.cent sites
along the time axis. This has to be contrasted with the case of the discrete symmetry, in
which the transition betweo.n the two configurations imposed by the b.c. occurs between
two sites.
The cost in energy AE dueto the ratation is thus:

(23.42)

and therefore:
(23.43)
In the case of a contínuous symmetry, it is easier to pass from one minimum o[ the energy
to another. This property has the direct consequences that ít is more difficult to break
the symmetry and that Goldstone modes appear in the phase of SSB. For d 5 2, TL has a
fmite limit and the symmetry is never broken (although, as we shal! see, there is a phase
transition withont ordering for d = 2, N "'- 2). This result, for whicl). we have given a
lteuristíc argument, can be proven rigorously (Mermín-Wagner-Coleman theorem).
For d :> 2 SSB occun; at low tem per ature. There e:xist.s some finite temperature T 0 , at
which a phase tnmsítion occurs.
Note that if we oonsider a system with a longitudin,.] sire I i- L we obtaín instead:

(23.44)

-=----------------------.. ~·IA<WI'·t.'l>t.'.J•'J<'I ~· JII,J'I,l!. ' l " t r " - =

540 Crittcal Phenomena: Generu/ Considerations 23.5

Hence, us;ng Lhe results of Sections 3.3,3.4, we conclude that the correlation finite size
correlntíon Jength ~L behaves líke:

(23.45)

To relate thís result with the possibility o( a phase transitíon we remernber that íf vre
assume that the correlation length is la.rge but much smaller t ban the transwrse size of
the lattice then we have the astímate:

r.r' r,r'

and therefore:
(23.46)
Tbe ""sumption implies that (S) goes to zero in the largevolume limít. A phase trllllSition
with ordering is only possíble if:

Tbe case eL o: L, i.e. d 2, ís the m051. subtle one a.nd we shall show !ater that it
characterires the critica! temperature of a serond order phase tronsition, at whícb the
spontaneous magnetízation also vaníshes.
These high and low temperature analyses have given us some indicatíons about the
exist.ence of a phase transition, and the na.ture of the phases. However they have given
us no information about the phase transition it<;elf, and about the behaviour of thermo- ·
dynamic quantiti..s in the neighbourhood of the crítica! tempera.ture. These problems
will be exarnined in the next chapter.

Bibliographlcal Notes
The proof of sponta.neous symmetry breakíng of i;yStems with discrete symmetrles ís
giv..n in
R- Peierls, Phys. Rev. 54 (1938) 918.
The impossibility of spontMeoUS breaking of continuous symmetries in one a.nd two
dimens.ions hM been proven by
N.D. Mermin and H. Wegner, Phys. Re11. Lett. 17 (1966) 1133; S. Coleman, Cornmun.
Math. Phys. 31 {1973) 259.
Two reamt textbooks are
S.K. Ma, Stlltistical Mechanics (World Scientific, Singapore 1985);
G. Pa.risi, Statistical Field Theorg (Addison-Wes!ey, New York 1988).
The presentation of this chapter follows
J. Zinn-Justin, Baslw Polje Sumrner Sch.aol1976, Saclay preprint DPh-T /7&-99 unpub-
lished, a.nd in Recent Advances in Field Theorg and Statistical Mechania, Les Houches
1982, Appendlx C, J.-B. Zuber and R. Stora eds. (North-Holland, Amsterdam 1984).
The replica trick has heen introduced in
S.F. Edwards and P.W. Anderson, J. Phys. F5 (1975) 965.
For the spin glass phase tronsítion see for instance
M. Méza.rd, G. Pa.risi and M. Virasoro, Spin Glass and Beyond (World Sdentífic,
Singapore 1987).

'-"----------------------~.u.,WI·!.hk' ~'1•'1·~ J
A23.0 Critica! Phenomena: Genel)ll Considerutions 541

APPENDIX 23
QUENCHED AVERAGES

In this work we discuss onlv so-called a.nnealed averages, í.e. we averag-e over random
configurations of a system _;hích explores the whole available phase space in lhe wurae
of tílll€ evoiution. This leaàs to the concepl of partítion functíon. However another cla.ss
of problems is of quite different nature, disorder is reiated to frozen degrees of freedorn,
or at least degrees of freedom (for instaoce impuríties in crystals) which do not evulve
dnring the time of observat:on. Wheu a relaxation time can be defined (see Chapter 35),
the reiaxation time is large compared to the time of observation.
Then it is no longer the partition function which h"" to be averaged over disorder,
but directly the physical observables, a procedure usually called quenched aoemge. Let
us here give for completeness the simple example of the gaussian model in a quenched
random magnetic field with gaussian dístribution. For simplicity we assume continuum
space and a continuous spin distribution rr(x) corresponding to a local free ection:

S(a) = So(a)- j d~x H(x)rr(x)

Sa(u) = ~ J ddx [(Vu{x)) 2 + 1J 2 a 2 (x)].

For each magnetic field distribution the partition fuuction is gíven hy

We assume the magnetic fields a.t different points uncorrelated with a gaussian dístribu-
tion characterized by the 1- and 2-point functions:

H(x)=h,

h a.nd w being two constants.

In this model we can immediate]y calculate the statístical a'lerage of the spin variable,
at fixed field distribution, at a point x

where .ó.(x) ís the u-field propagator. lf we now av.;rage {u(x)) over the position x in a
large volume n, we find

Beca.use the propaga.tor in ab5ence of dísorder is invariant under translation we now find
in the r.h.s. the average o[ the ma.gnetic field. The rcsult is

. l--h
n-•
f d x{u(.r)) = -;;H(x)
I' 2jJ. .=

----------------------,~·IA<Wf',t.'l>t.'.J•'J<'I~·ni,J'I,l!.
542 Cn:tiool Phenomena: Generol Considerotions

We can also calcula.te the average over the sa.mple of

!r 1 f d 0 y{a(y))(a(x+y)} f
n- 1 d 0 yd"zd•tt.(y-z)t.(x+y t)H(z)H(t)

= n- f 1
d•yd•zt.(y)t.(x + y +:)f ddtH(z + t)H(t)
Again traru;lation invariance, in the absence of field, has reduced the space average to a
field avera.ge:

It follows

Note on the other hand that the a-field 2-point function is independent of the rnagnetk
field

The replica trick. In this simple example since the obrervables can be calculnte<i
explídtly, it is easy to average over the quenched random variables. This is not the
general situatíon of course, so an algebraic trick has been invented to overcome thü;
problem. lf we want to average correlation functíons we have to average the free energy
W ln Z ínstead of the po.rtitíon functíon. Then the following trick is used

Z"-
lnZ=Iim .
n-O fl

We do not know how to calcnlate Z" for general real n, but we know how to calculate it
f<>r integer n. It ís suflident to replicate n copies of the initíal varla.bles. In our exawple
we introduce a lield O'; with n comp<ments and find

The gaussia.n disúrder average is easy to perform

zn(J) = /[da,(;r;)Jexp [- ~ So(u;)

+!w 2 f d•x I: o;(x)t1J(x) +f d•x ~)h+ J(x))cr;{x)] .
.,., '
The subtle Question is then to take the n O limít. For instante to calculate the tield
expectation value we minimize the potential. We find the replica symmetrie solution:

h
(u,(x)) ,.•- nw2.
~· llt.J•t.l:'. 'F\ I/,_,_____________________,
~·V~.>Ln·!.hk'~'i',.~ J
A23.0 Criticai Phenomena: Gonerol Con$Íderntíam 543

In this simple example we set n = O and obtain the correct result

--
(cr(:r)) = 1L: o;=
lím -
on-(}n
h
1

Notíce. however, that even here the soJution does not make sense for n > !J 2 /w 2 , and
therefore a.rguments which state that under certain oonditions analytic functions known
for ali íntegers are uníquely defined, do not apply. Therefore, basicolly the domain of
valídity of the replica trick is unknown.
Similorly the replica propaga.tor can be ca.lculated

Summiog over i, j, dividíng by n and taking the n O limit yields the correct result.
Notice that for m "" O small the replica-propagator for n O is moce singular thàn the
free field propagator at low momentum. This hBS implications for critico! phenomena in
disordered systems.
For less trivial models the problem is very complicated. The famom spin-glass mode!
is difficult to solve even in the mean approximation we discuss in the next chapter, and
the mean field solution relies on a breaking of the symmetry between field replica, as
sbOWJJ by Parisi.
A final remark: to ca.lculate correlatio"" due to the disorder averaging, we have to
eva.luate instead quantities like

The replica trick obviously generallres with 2n replicas.

-----------------------.,,~,fA<WI'•t.'l>t.'.J•'J<'I~· JII,J'I,l!. '1"t1f~

4 MEAN FIELD THEORY FOR FERROMAGNETIC SYSTEMS

In Chapter 23 we have di•cussed the existence of pitas<! transitions in [erroma.gnetic

systems by comparing the phase structure at low and high temp<rra.tures. However, we
have used methods which are not valid near the transition itself and thus prOvide no
informa.tíon about the hehaviour of thermodynamícal quantities in the neighhourhood
of the criticai temperature. For this purpose we develop in thio cbapter a very simple
appro:x:imation scheme, the mean field a.pproximation. We then distinguish between fust
and second arder phase transitions. ln the latter case the correlation length diverges.
This has important physical consequences: In the mean fteld a.pproximatiou the singular
behaviour of correlation functions nea.r Te a.nd in small !ield is universal, i.e. does not
depend on the details of the interactíons.
The mean fteld approximation can be introduced by various methods, as a partia!
summation of the higb temperature expansion (see Appendix A24), as e. result of a
variational method, or as the leading order in a steepest descent calculation. We sball
adopt the latter point of vlew because it wíU allow us to cal culate systematic corrections to
the mea.n fteld approxímation and therefore to di.cus:; íts validity. The rol~ of dimension
four will tben emerge.
In the appendix we construct the mean field expansion for general la.ttice models,

24.1 Ising-like Ferromagnetic Systerns

We ftrst conslder tbe criticai behaviour of a classical spin system in a d-dimensionallattice

witb an attractive, trllllsla.tion inva.riant, short rang<>, twl}>body interaction. The spin
variable on a s.ite i is called S,. We assume the.t the system ha.s aS,_. -S symmetry. Th
discuss the effect of a magnetic field and to construct a genera.ting functional of correlation
functions it is useful to introduce in addition a lattice poinç dependent magnetic field.
We call1i the energy of a spin conftguration, f3 th<l inveree temperature and dp(S)
the one spín distribution wbich weights the spín con!igurations at ée.ch site.
The partition function can then be written:

Z(H) =!f}. dp(S,)exp(-f31i(S)], (24.1)

wítb:
- j31i {S) ""' L V;;S,Sj +L H;S, (24.2)
iJ

lt is technically conveníent to aiBO assume thll.t V;; va.nishes (see Appendbr. A24). The
intera.ctions are ferromagnetíc if V.; is non-negative. We have already met expressions
simílar to (24.1) when we constructed the Feynman-Kae representation of the quanturn
mechanical partition fuuction tr e-llH in Cba.pter 2 or íutroduced in !ield theory a lattice
regularizat.ion of the generatíng functional o[ euciidean correlation functions. Note an
ímportant difference in point of view b<ltwoon methods ínspired by hígh temperature
consídera.tions and field theoretical perturba.tíon theory: a.t high temperature the measure
dp(S) Í5 kept !ixed and the term v,,S;Sj, whlch connects d;fferent lattke sites, is the

-----------------------.,,~,u.,wl.!.hk' ~,1•'1·~ J
24.1 Mean Field Theory; 545

interaction term and ís treated as a perturba.tíon; in Fíeld Theory the term V;,S,S, ís
rather the analogue o( the ki,netic term while the measure dp (S) contains the interactions.
These remarks wíll be clarilied !ater.
H,gh temperoture sel'"ies e:zpansion. The simplest method to calculate Z (H) is to
expand expression (24.1) in powers of V (high temperature expansion) sínce ali terms
in the series CIU'l be expressed in terms of the moments of the one spin distributíon in a
magnetic field:

I
Z (H) /Zo (H) = 1 +L V;1 (Si) (S;) +-;L
lj 2.ij~
V., V~z (S,) (Sj) (Sk) (Sz) + (24.3)

with

Zo(H) =TI :(H;), (24.4)

z(h} =f dp(S)éh, {24.5)

(s;'} = z- (H;) ( ~,)" z (H,).

1
8 {24.6)

HoweV>!r thís expansion is useful only if the leading term, the inlinite temperature re-
sult, already is qualitatively correct.: the expansion gives information about the high
temperature phase and ne<:essarily diverges at the criticai temperature. It provides a
full qualítative description of the phase structure only when the critícal temperature
vanishes. Therefore we now introduce another apprrnúmation scheme, which diagram-
tnatically corresponds to perform a loopwise summation of the high temperature ser!""
expansion (see Appendix A24).
Remark. For what follows it is convenient to introduce the function A( H):

A(H} = -lnz(H). (24.7)

For the lsing model where S; = ±1 one finds for exarnple:

A( H)= -lncoshH.

Due to the syrnmetry of the dístribution, A(H) is even WJd detreasing for H > O.
Moreover A"(H) < O (Schwa.rt•'s iDequality) and thus -A(H) is ronvex. Finally we
llSSUme that the distribution decrea.ses faster than a gaussia.n for large values of the spin
S. It is th~n easy to verify that z(H) is an entire function which ímplies that A(H) is
regular On the real axis and ÍD some neighbourhood of the origin. Moreover:

lim A(H)/H' =0. (24.8)

IHI-oo

----------------------,~·IA<WI'•t.'l>t.'.J•'J<'I~·ni,J'I,l!.
545 Memt Fie!d Thoory 24.2

24.2 Mean Field Expansion

Since the partitlon function ca.n easily be cakuiated when ali points are disconnected,
a simple idea is to write the term which in the conli,guration energy connects spiiL5 on
different sites as a.n integral over disconnected terms (Hubbard's transfonnation). More
explicitly;

(24.9)

The partition function becomes:

Z(H) jiT dX,exp( -~ LX•V.j X;) fi/ dp(S;)exp [LS;(H;+X,)).

t tJ
1

' t
(24.10)
The sum over spin configurat1ons can then be performed:

z(HJ = jiT dX,exp[-~L,x.v>J'X;- LA(x, +H•J].

t ., t
(24.11)

where A(H) has heen defined by equation (24. 7).

Note that identity (24.9) is a priori valíd only for positive matrices V;;. Actually, in
the case we apply it, V \s not positive in general, a.nd equation (24.9) should only be
considered as a formal algebraic identity.
We now evaluate expression (24.11) by the steepest descent method. At lea.ding order
we replace X; by a solution of the saddle point equation:

4Lv,;'xj
;
+A'(x, +H,)= o. (24.12)

The me&n\ng of this npproximation is th"t we have replaced the interaction which con-
nects the sites by t.he bast (in a sense wbich will become clea.rer below) equh...Jent mag-
netic field. A more deta.iled analysis shows thlll: t.hís approxirolll:ion becomes exact when
the dirnension d of space becomes lo.rge, so that the action of ali sites on a given síte ca.n
índeed be rep!Med by a "mean" magnetic field (this property is related to the central
limit theorem).
The a.dva.ntage of tbis algebre.ic formulation of mean field t.heory is that it allows a
systematic discussion of the corrections to the meo.n .field approximlll:ion.
Before discussing tbe mean field approxima.tion it is useful to introduce a parameter
which cha.racterizes the expansion around mea.n field theory. Let us replace the coeffi-
cients V;j in expression ( 24.2) by V.; f I and t.he spin S; by the sum o f I independent spins
with the same dístribution dp( S):
I
s, :L:sik).
k=l

Then:
24,3 M .an Fíeld Thec;ry 547

The case I 1 corresponds to the initial dístríbution. The partition function becomes:

(24.13)

This expression dearly shows that a steepest descent calculation of the partition functíon
generates an expllllsion in powers of I /1. Mean lield theory is analogous to the tree
apprax.imation of quantum field tbeory.
Let us, for convenience, rewrite expression (2.4.13), settlng

X 1 +H1 = 'P<·

This lea.ds to:

Z(H) f n. dip; exp{ -l [1I.:., '"''v.; 'Pi l 2ip;V,j 1H; + H,V,j 1H;)+ L A(ip;)]}.
• (24.15)

24.3 Mean Field Approximation

In terms of the la.ttice variables ip;, the saddle point equation (24.12) reads:

(24.16}

The Welss free energy W (H) is then:

- L~(.,,-H,)V,j 1 (1P; H;) LA(IP;). (24.17)

ij

The magnetization M; in the mean lield appi'O>cimation (24.16) is:

aw
M; =- 8H, -A' ('P;). (24.18)

'rbe thermodynamic potential r(M), L.:geDdre transform of W(H) is then given by:

r'(M) = L M,H, W(H) = - LM,V.;M;+ LB(M;), (24.19)

i •i

In which B(M) i> tbe Legendre transforrn of -A ('t'):

B(M)- A (<p) =10M. (2.4.20)

Aga.ín in the exarnple of the lsing model we lind:

B(M) = ~(1+ M}ln(l +M}+ !(1 M)ln(l M). {24.21)

---------------------·~·IA<WI'•t.'l>t.'.J•'J<'I~·ni,J'I,l!.
548 Mean Field Theory 24.3

The relation between magnetic field and magnetization can also be written:
ar aB
H,= ôM "'-2LV.;M; + EIM. (24.22)
• j •

ln zero magnetic field, the magnetízation, i.e. tbe spin average, is given by an extremum
ofthe thermodynarnic potentia! r (M). Furthermore, since the partition function in zero
field i.s exp 1-lf(M)J, the dominant saddle points coriespond to mhúma of r(M).
In the case of translation inva.riant systems,""' expect the rnagnetization to be uníform:
M; = M. Let us cal! !1 the number of Ja.ttioe si~, then:
n-'r(M) -VM2 + B(M), (24.23)
in which V i.s a para.meter defined by:
V=LViJ>O, (24.24)

which is finite because the interaction i.s short range, and proportional to the inverse
temperature {j.
The Legendre tranoformation implies the relation:
B"(M) -1/A"('I'). (24.25)
Bocause -A(IP) is a conve1< function of <p, B(M) i.s also convex. lt i.s analytic in the
neigbbourbood of tbe orígin. Moreover the property (24.8) implies that for IMI large
enough, r(M) in equatíon (24.23) is !UI increasíng function.
Let us nO'i\' find tbe mínima o[ f(M} when the temperature, and therefore also V,
vary. For V sma!l (hígb temperature) VM 2 is negligible, and the r.h.s. of equation
(24.23} is convex. The mínimum of r(M) is M =O, the magnetizatíon vanisbes. Wben
V lncreases, in general at some value of V otber local mínima ll.ppea.r which eventually
become the absolute mínima of r(M). Wben tbis occurs, the value of the magnetization
M discontinuously jumps from zero to a fuúte value corresponding to tbis new absolute
mínima. Tbe system undergoes a firxt order phase tronsition. Figure 24.1 descríbes
the evolution. Fluctuations around tbe saddle point are governed by the value of tbe
second derivative of tbe potential at tbe mlnimum. Generica!ly in such a case tbe second
derivatlve is strictly positive and tberefore the cortelation length, wbich, as we know
from Field Theory calculatíons and sball again see explicitly !ater, is proportional to
fll/ 2 [r''(M}r 112 at the mínimum, remei"" linite.
Although first arder phose transitíons are cornmon, they are not particula.:rly interesting
for us. Wben the correl~>tlon length remains linite, no universality emerges. At tbe sa.me
time, beca.use the lluctuations are not critica!, mean field tbeory gives a satisfactory
qualitatíve descriptíon o( the phy,;ics.
However if no absolute minímum appears at a fmite distance from tbe origin, linally
at a critica! temperatura T, torrt'Sponding to the value V, of V:
2V. B"(O), (24.26)
tbe orig:in ceases to be a minimum of the potential, and below tbis ternperature two
mínima move continuously away from the origin (see figure 24.2). Sínce tbe magnetízation
remains oontinuous at V"' we bave a ~econd orcler phase lromition. Because at V, tbe
second derivative of r(M) vanisbes, the correlation length diverges.
Tbis is the situation we wil\ analyze s)'Etematically from now on, first in tbe framework
of lhe mean lield approximation and then by <'onsidering OOriections coming from higber
arder terms in tbe mean field expa.nsion.

~· llt.J•t.l:'. '1'\r.,.,._ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _,

~.u.,WI·!.hk'~'i''i'~J
 24.4 Mean Fte/d Theory; 549

Fig. 24.1 Ft"" energy: lil'St ordcr phase transition.

Fig:. 24.2 Ft<le energy: second <>rder ph.Me trii.TI.Sition.

24.4 Universality witbin lhe Mean Field Approximation

We now examine the behaviour of Vll.I"Íous thermodyna.mical quantities for a t..mperature
T dose to T,, i.e. V clo.re to V.- Since the transition is continuous, the magnetization is
•mall and we ca.n expa.nd f(M}, which as A(.p) is a. smooth even function, in a Taylor
~~Cries in M. We paramet:ri7.e f(M;) as:

f(M,) - l:ViJM,M; +L (;M; +~M: + ---).

..,
The com:sponding expa.nsion for A(.p) then is:
. (24.27)

A ('I') _2_..,2 + _b_..,~ + ... (24.2S)

2a 24a•
. The convexity of B(M) implies that a is positíve. The parameter b is a.lso generically
positive, because we have 55$umed tha.t no lir!;t order tra.nsition occurs at hígher tem-
In the Isíng model one finds:
a= 1,

IU'I.l:'. '1'11:~.·=-=---------------------""~·IA<WI'·t.'l>t.'.J•'J<'I~· JII,J'I,l!. 'l"tr."""'==

550 Mean Field Thoory 24.4

The criticai value \'•• in tbe mean field appre>o<linatíon, is tben:

(24.29)

For V > Vc. we fmd a spontaneous m&gnetlzation M:

M -[12 (V Vc)/ b] 112 for IV V. I « 1. (24.30)

Near the critical temperature r., tbe magnet!zation ha.s tbm a power !aw behaviour witb
a magnetic exponent /3: :

M ~ (r.- r)"', fi= 1/2 (24.31)

The inverse of the magnetic susceptibility X = aM / iJH is tbe second derivatíve o f r( M)

with respe<:t to M. In zer<> field we ftnd:

x+ 1 = 2 (V. - v) r:;.r,,
(24.32)
x: 1 =4(V V.) r<r,.
The magnetic susceptibility therefure diverges at r, witb susceptibility exponents ,, -y':

X+ ~c+(r-T,)-~, j=l,
(24.33)
X-~ C_ (r.- T)_,., -r= 1,
and:
(24.34)
For V close to V. and small const&.Dt applied magnetlc fieJd H, the equation of st.ate,
which is the relatíon between field, temperature and magneti%.ation, is;

H= .!!!:.._! = 2 (~ - v) M + ~Ms +O (M5 ) . (24.35)

iJM, M,=M 2 6

Fbr H, r- r,, M small, rescaling field, temperatura and magnetlzation, we can cast tbe
relation into a universal sca.ling fonn:

(24.36)

where the function f (z) is simply:

/(:r:)""' 1 +X, (24.37)

a.ud the exponent 6 b!IS the "classicaln or mBBn-fie}d value:

6=3. (24.38)

From the functional (24.27) we can alro der!"" the 2-point correlation function w,Vl:
2r-)_,
6-
w.,(2)_- ( -
8M;8MJ '
24.4 M ea.n Field Theory 551

whose FourieY transform W! 2 ) (q) is:

w< l(qJ = I:w,Vlexp(iq -r,3) =[a+ ~M2

1
2
2"ii(q)r (24.39)
J

witb the definition:

V(q)"" L v;j exp (iq · r;1). (24.40)
j

The momeutum q has, by defuútion, components satlsfying lq,.l < 1r. Tbe potentia.l v;,
is sbort ranged, posltive and translation invariant. We assume in addition that it b.as
t.he cubic symmetry of t.he lattice (this assumption is convenient but not essential since
'!e ca.n a.lways ma.ke a linear tra.II5Íormation on tbe coordina.tes). Then the expansion of
V(q) for q small ha.s the form:

(24.41)

in which ,p is the lengtb squnred of the vectór q and therefore o (d) symmetric, and r
some constant. Equa.tion (24.41) actually defines precisely what we m""" byshort tange
potential: the potential has to decrease at la:rge dístanc~)n such a way that its Fourier
tra.nsform b!l8 the expa.nsíon (24.41). Then for q small, w<2 l(q) can also be expanded:

(24.42)

This sbows that th.,t. ptopagator has a free field, or Ornstein and Zernilre form. We
bave paramettized W12 l(q) by introducing e
tbe second moment of W;j· The length (
charactetizes, up to a numerical factor, the exponential decay of the correlation functioru;,
and ca.n be taken as a measure of the correlation length. From equations (24.39,24.42)
we find:
C2 2:{1- (a+~~-2v). (24.43)

In zero magnetic field this yields:

{:t'2 = p-2 (1 V/V.) for T> T,,

(24.44}
C, 2p_, (V/Yc - 1) for T< T•.

Introducing general correlation length exponents v, v':

{+ = f+(T- T<)-P'
(24.45)
(- =/- (T.- T)-•',

(24.46)

t .
552 Mean Fíeld Th€1Jry 244

Final~v i! we assume that the 2-point fullétion beha.ves ín general at T, !ike:

wm ~ 1 /lr ld-2+~ (24.47)

Íj !r.,t-o::· 1J ,

or in Fourier transform~
(24.48)
we deriw from expression (24.39) the dassical or mean field va.lue of the exponent !J:

r J)

S[ld!ific heat. Let us calcula.te the derivative of the free energy per u.:Ut vohr with
respect to V, which is a measure of the temperature,

2. 8W(H)I ~ M'(H 0).

n av H~o

A bove T, the averagE energy V3nishes and below T, it i5 pr<>portional to tl.· :-1uare oi
tbe spontaneous magnetization. Deriving aga.in wíth respect to V we hnd specific
heat C
12
C(T-T,_)..:b. (24.50)

We find that in the mea.n fie\d a.pprmámation the &pecific heat has a. non- un; .a\jump.
Continuous •ymmetries. lf the initial spin VW'iable 81 is a. N -romp.::nen· ·tor, and
if both the interaction and the spin distribution have some continuam syr: .ry, most
of tbe previous results rema.ín cleo.rly unchanged. Tbe main difference OOl from the
appearance in zero fie!d of Go!dstrme (massless) modes below T,. Lel uo tak• example
of the O(N) symmetry. We reWrite the thermodynamic potentíal:

r(M) = _ 2:V.;M•. Mj + :z= oM~ + ~ (M:} 1

+ ···). (24.51)
., '
The re)ation between magnetizatí<>n and (constant) magnetic fie!d stiU is:

HfM =-2V +a+ tbM 2 , (24.52)

in which H, M a.re now the lengths of the vectore M, H.

From equa.tion (24.51) we ca.n compute tbe inrerse 2-point correlation function rl~)ne,

(24.53)

in which M,... (o = 1, ... , N) are the cornponents oi the constant ln8glletization ""<:tor.
Let us introduce a unit vector u along the dírection of spontanooliS [email protected]:
M=' Mu,..

Then tbe propa.gator (24.53) can 'De demmposed int.o a. transverse and longitudinal part

(24.54)

----------------------'~•1-"hWI•!.hk' ~'1' 1·~ J

24A Mean Fie!d Theory 553

The nvo components are given ín Fourier transforrn by:

fL(q) a+!bM2 2Y(q), (24.55a)

r,.(g) = -2V(q) +a+ ~bM •
2
(24.55b)

The expressíons (24.55a) and (24.39) are similar. Usmg equation (24.52) we can rewrite
rT(q):
(24.56)

Thi.s equatíon shows that, in the low t.:rnperature phase in zero !ield, the inverse trans-
verse propagator vanishes like q 2 at wro momentum for llJIY temperatme below T 0 ,
indicating the presence of (N 1) Goldstone (masslcss) modes. Let us recall brieHy in
this contm.:t the generalíty of this result. Gomg back to magnetic notations, let us assume
that the thermodynamíc pot.:ntial r(M) is a funct.íon ofthe squa.re of the magnetízation
M. Then the relatlon between magnetic field and magnetization takes the forro:

r(M) = </> (M 2 /2), .~ H, = Mo</>' (M 2 /2). (24.57)

The ínverse 2-point function at zero momentum is given by:

(24.58)

Uslng equatíon (24.57), we obtain for the transverse part of the inverse propagator at
zero momentum:
(7)]-1 (q=O)=HfM.
[W,. (24.59)

Jf M does not vanish when H goes to zero (C35e of spontaneous magnetization) one finds
that the transverse correlation lengtb diverges below T., in agreement with the general
remlts proven in Sections 13.3,13.4 (to which....., refer fM an extensive discussion of this
problem}.
La miau 's theory. Ali previous results have been ..stablished in the frrunework of a
~pecial approximation scheme, the mean field apprnxima.tion but, as shown by Landau,
they follow aJso from more general assumptíons:
(í) The thermoclynamícal pot<:ntial can be expanded ÍD powers of M around M = O.
(ii) The coefficients of the exparu;ian are regular functions of the temperature for T
tlose to T" the ternperature at whlch the coeffident of M 2 va.nishe!S.
(iií) The fundíons f'!n) whkh appear in the expansion ofr(M) in powers of M(q), the
Fourier tra.nsform of the local rnagnetiza.tion M1:

are regular for small q; and r<•J (0, O, O, O) i5 positive.

Hidden in these assumptions ís the generalldea that physics on different scaJes decouple
••nd thu• critica! phenornena CllJI be described at leeding arder in terms of a finite number
effectire mocrascopic variab!es.

----------------------,~·IA<WI'•t.'l>t.'.J•'J<'I~·ni,J'I,l!.
5M Mean Field Theory 24.5

24.5 Beyond the Mean Field Approximation

From too point of view of the representa.tion (24.15) of the partition function, mean
field tbeory it. a saddle point approximation. lt is tberefore easy to calculate systematic
corrections to tbe mean lield appr<I>".imation by expanding around the sarldle point.
Since we are only interested in the behaviour of thennodynarnical qu!Wtitíes in the
criticai region: T dose to Te, H, M and momenta lql small, we CIUI semev~hat simplify
tbe expression. The term E;; }H; V,j' H; gives ll.D arldltiw contribution to W(H) which
it. negllgible at learling order. We can r.lso take the quantíty J,:

as a measure of t.be magnetic field, since the convolutíon of H; with V,j leads just to a
s.bort range smearing.
The simplified pa.rtition function Z(J) is then givtln by:

Z(J} jfldop,exp[-t(n:'~'•V.i''~', + :L:Al<P•) J•'~'•)}· (24.60)

t 'lJ •

We recogrú:t.e a lattice regularized functional integral for a scalar boson field of the type
studied in Chapter 5. The mean lield opproxirnation ís ana.logous to the tree l.vel a.p-
proximation of Field Tbeory. Corrections to rnean field are given by perturbation theory.
Let us introduce the propagator LJ.,,:

ll.ij' ~v,j' a-1,

lts Fourier transform ll.(q) bas for q small the form:

a- 1(q) = tV- 1 (q)-a- 1 = (2V)-• [l-2V/a+p~q2+ O(q 4)] (24.61}

In tbe one-loop apprwómation (equation (6.49)) r(M) is:

f(M)
., .
~ Í::: M,V,j 1 M, + LA(M;) + ~ trln UV,j 1 +A" (M,)b,i) •
Froro thís expression ít is easy to derive the inverse 2-point function r12 l up to order I/I.
In momentum space we find:

ln particular at zero momentum we obtain:

r<•> (O) l
2V -
1~ + 2ab4l f dJq
(27r)d ll.(q) +O
(I)
fi .
Tbe criticai value Vc is now defined by:

~· Jlt.J•t.l:'. 'F\ li,______________________,

~•Vhl.'i·!.hk'~'i',.~ J
24,6 Mean Field Theory 555

Subtracting this equation fmm equation (24.64) we find:

r
(l)
(O) =
I (V, -V)
2--vv;- +
b
2la'
f 0
d q
(2")• j.ó.(q)- A,(q)] + 0 ( 121 ) · (24.66)

ln the 1/l contribution we can replace V, by its mean field value af2. Then for V- V,
and q small:
.ó.(q) = (2"~)- 1 (v,- v+ p 2 q~ +o (q•)). (24.67)
and therefore:
(V-V,)
t.(q) t.,(q)- -w;-t.(q)A.,(q). (24.68)

Discussicn. Ffom these two e:xpressions it is clea.r that we can safely take the limit V
V, inside the integrand provided tbt) integral Jd"q/(q 2 ) 2 converges at low momentum,
í.e. for d > 4. Then r<•l (O) vanishes Jínearly in (V- V,) or T- Te as in meM field thoory
for d > 4. Since r< 2 l (O) is símp!y related to the ínverse susceptibilíty in zero field, the
exponent -y rema.íns l. However, for d ::; 4 the limit V =
V, is singular. Evaluating for
V d~ to v. and 2 < d < 4 the dominant contribution to the integrol, whicb comes
entirely for the small q region, we lind: .

p2J (O)=: (V- l'c) [ 1 h r(l- df2J (v.,- v)dJ•-•] o(.!). (24.69)
2 "? + 2!a2pd (4")"/2 v, + 12
For V close to Vc the correction to r< 2 l(O) olways is ]aTget than the leading term. Tbere-
fore the predictions of mean 6e]d theory can no longet be triiBted:
(i) For dimeooon d > 4 tbe 'universal" predictions of mean field theory rema.ín un-
cha.nged.
(ii) For dimension d $ 4 "infraxed" (IR) divergences, i. e. singnla:rities due to the small
momentum behaviour of the propagator, sbaw that the predictíons of meM fie!d theory
<ll.llllot be correct in gen.,-al.
lnspection in Section 24.6 of bigher arder corrections will con6rm tbis result.

24.6 Power Counting and the Role of Dimensiun Four

In the precedíng soction we have shown that the universal predictions o f mean field theory
cannot be trusted below four dimensions. We olso note that the first dive:rgent corrections
depend on a.dditional detal1.s of the inltiol microro:Jpic model. At first sight this would
suggest that la.rge dista.nce properties a.re short distance sensitive and therefore no form
of univetsa.lity survives, in contradiction with experimental and numerícal indica.tions.
However a clooor inspection of equatk>n (24.69), for exarnple, reveols that, a.t le!!St at
leading order, only one new pa:rameter appears, re!ated to the value of the rp4 term in
lhe small cp expallsion of the interaction term. We now wa.nt to extend the analysis of
"infra.red divergences" to ali orders in the mean field expanBíon, to understand whether
the roles oi dimension four and cp' lnteraction survi"'· Since the analysis is perturbative,
we ha.ve to assume th!rt, ín some sense, deviatíons from mean field theory are not too
ltu'ge. This poínt will be further discussed In the comíng sections.
In the spirit of wbat has been done in Section 24.5, we first perfonn, order by order in
the mean field expansion, a magnetic susceptibility or mllSS (in field theoretica!language)
tenormaliza.tíon. We trode the pa.rameter V for the parameter a:
(24.70)

----------------------~·IA<WI'•I..'i>t.'.J•'J<'I~· JI ,J'I,l!. ':!:t.ll!>.=

556 Mean Fíeld Theory 24.6

by inverting order by order the relatíon betwren V and cr (see Section 10.1 ).
The cp-propugator, for wha.t concerns the most singular contributíon at low momentum,
can then oo replaced by
(24.71]

We now ooDSider the individual contributioru; coming from ali (even due to '{J - -<p
symmetry) powers of 'fJ in A ('I'), allowing even for the possibility o[ polynomial momen-
tum dependence in the corresponding vertices (thís wo.s not the case in the examples we
h ave oonsídered so far). These contributions have just the fOnn of ordinary Feynman
diagrams with the propagator (24. 71 ), integra.ted up t.o a "cut-ol!" o f order 1 since the
momenta q are limited to a Bríllouin zone;

In Chapter 8 we have associated with each dlagnun a dimension given by power counting.
If u., is the number of vertex of type o:, whlch oonneo:;t T1 0 !ines and oontain k., powers of
momenta, the dimension 6 ("f) of a diagram -, oontributing to the n- point function r<nl
is (equationll (8.14,8.22,8.26}):

(24.72)

Let us for simplicíty consider the díagrarn 1 at zero externai rnomenta and períorm the
change of variables (for ali integratíon momenta):

(:!4.73)

After tllis rescaling the contribution D(1) of the diagt-am -y takes the form:

D("l') ,ihl/' D'('Y),

by definition of 6(-y). D'(-y), whicb is calcul"ted with a propagator lj(q2 +1), is no Jonger
IR divergent when a goes to zero, but may have large momentum (UV) divergences since
the cut-off, wbich ís now of arder r:;- 112 , beoomes infinite in the snme limit.
If the condition:
(24.74)

is satisfied for ali vertíces, 6( "r) increases with the arder in perturbation theory. However,
at the same time D'{"r) he.s UV diversences proportional to o- 112 to a power given
by power counting. For tN divergent diagrams the two powers o[ a exactly cancel
Md therefore ali oontributions in perturba:tion theory have the same behaviour, i.e. the
behaviour of mean field theory. This in particular applies to the 2-point function for
which 6(1) > 2. For hígher correlation functíons a finite number of diagrams may be tN
wnvergent and therefore are more IR singular than predicted by tbe tree levei action.
Then the leadíng oontribution is given by the one--loop diagrarn witb only vertices having
the smollest poo;sible dimension.
The wndition {24.74) ís satísfied if all interactions are non renormalizable. For "
theorywith interactions even in <p this implies d > 4. The leading arder result of Section
24.5 then extends to ali orders: mean field theory is qualitatively correct order by arder
 M e.an Field Throry 557

in perturbation theory; the effect. of loop correct.ions ís to modify the non-universal

pamme'ter8 o f the mean field approximation.
If ínstead there exíst ínteractjons suoh that:

(24.75)

theo one can find diagrams in which increasing!y large negative powers of o have been
factorized. These diagrarns are superficially convergent at large moment.a. This implies
tha.t either they have no drvergent subdiagrams and their singula.rity is indeed given by
the power li(-r)/2 or they are even more singular when o goes to zero. In ali cases the
singula.rity of diagrams increases without hound wíth the order of perturbatíon theory.
Let us now try to charPCteriz.e the most dívergent diagrams order by order in pertur-
bation theory. lf we forget about the subdívergent dlagrams we have just to find the
nrlnimum of ~(-r)/2 at each arder. This mínimum is obtained for tbe least UV divergent
íntera.ctíon, i.e. for ko = (} and nu as small as possible (we a:;sume d > 2). The smallest
even value is no 4. This corresponds to the '(!4 ínteraction.
Let us then examine the ptoblem of divergent subdiagrams: remember tbat we have
performed a mass renormalize.tion. Acroroing to the analysis of Cbapters B-10, tbe .;4
diagr!L111S are thus UV finite belaw four dimensions: The form the IR singularity deduced
from the sirnple scaling (2<1.73) is the correct one.
This is no longer true for more UV divergent interactions which even below four di-
IDeDSions require edditional renormalizations. Howe1>er, it will be shown ín Chapter 27
the.t these interactions can be decomposed ínto the sum of an elfective .;4 ínteraction
!Uld a part wbich is less singular.
Therefore the most IR singular terrns, arder by arder ín perturbation theory are gen-
.erated by an effective .;4 field theory, the coefficient of the '(! 4 íntere.ction being not the
ooefficient read in the original hamiltoniiiil but a coefficient renorma.lized by the other
interactions. In Chapters 25, 26 we shall investiga.te the large distance bebaviour of
corre!ation functions ín this elfectíw rp4 field theory. In Chapter 27 we shall discuss t.he
elfect of the neg!ect.ed sub!eading divergences to show the internai oonsístency of the
method.
Note fine.lly that for d = 4, mean field behaviour is madilied by lagarithmic oorrectíons.

24. T Tt-icrltical Points

So far we h ave assumed that we could vary one control par ame ter, the temperature, !Uld
that, therefore, the coefficient b of the term M 4 in the expansion o[ r( M) was a number
of arder 1. However there are situe.tions in whicb an additional physical parameter CIIIl
be varied, IUld both the coefficients a and b of M 2 and M 4 CIIIl be made to vanish. Tb.is
occurs for instance in He3 -He4 mixtures or some metamagnetic systems. In the lsing-like
models we ba.ve oonsidered so far, this can be achieved by allowíng the spin distribution
to vary.
!f the coeffident c of M 6 is positíve:

r(M) =- LVi;M;M; +L (~Ml + ~Mt + ~M,6 +. ··). (24.76)

\J i

when b docreases, it is possible to follaw a line of critical points until b vanishes. At t.his
poínt, the tricrítical point, the M" term becomes relevant, and a new analysis has to he

----------------------~·IA<WI'·t.'l>t.'.J•'J<'I~·ni,J'I,l!.'l"t.ll"""""'=
558 Mean Fi<ld Theory 24.7

performed. The various criticai exponents bave vatues diflerent fram those found for an
ordinary criticai point. After the tricriticlll point, b become:s negative and the transition
become:s lirst order.
Corrections to tbe tricriticaJ thoory can be studied by the method of Section 24.6.
Using power counting arguments, one now finds that the most singular corrections are
reproduced by a continuum 1/!0 field theory, and tbat the upper criticai dimension is 3,
Le, above 3 dimensions mean field theory predict.o; correctly universal qua.ntities, whereas
it is definítely not valid in 3 rlimensions and below.

Bibliographical Notes
In this chapter we have closely followerl the review article of E. Brézín, J.C. Le Guillou
and J. Zinn-Justin, published in
Phase 11-ansitions o.nd Critú:al Phenomena vol. 6, C. Domb anrl M.S. Green eds. {Aca-
demic Press, London 1 976).
For earlier articles àlld reviews about meM field theory see for exa.mple
P. Weis«, J. Phys. Ro.dium, Paris6 (1907) 667; M.E. Fisher, Jl.ep. l'rvg. PhyJ. 30 (1967)
615; L.P. KedMoff, W. Gõtze, D. Ha.mblen, R. Hecht, E.A.S. Lewis, V.V. Palciauskas,
M. Rayl, J. Swift, D. Aspnes àlld J. Kane, Rev. Mod. Phys. 39 (1967) 395; H.E. Stanley,
lntm<Wction t.o Phase 11-ansition. and Criticai Phenomeno. {Oxford University Press,
New York 1971).
The scaling forro of the equation of state was proposed by
B. Widom, J. Chem. Phys. 43 (1965) 3892, 3898.
Landau '• theory is presented in
L.D. L&ndau, Phy•. Z. Sowjetunion 11 (1937) 26, reprinted in Cllllated Papers of L,D.
Londau, D. ter Haar ed. (Pergamon, New York 1965). ·
The Hubbarrl transformation appears in
G.A. Baker, Phys, Rev. 126 (1962) 2071; A.J.F. Siegert, Brnndeis Lectures vol. 3 (Ben-
jamin, New York 1963); J. Hubhard, Phys. Lett 39A (1972) 365.
See also
R. Brout, Phys. Rep. IOC (1974) I.
'The role of tbe !p4 (z) field theory was empbasized by
KG. Wílson, Phys. Rev. B4 (1971) 3174,3184.
For a discussion of Lricritical systems see
E.H. Graf, D.M. Lee and J.D. Reppy, Phys, Rev. Letl. 19 (1967) 4.17; R.J. Birge!leau,
W 13. Yelon, E. Cohen and J. Makovsky, Plry$. Rev. B5 (1972) 26(}7; M. Blume, Phys.
Rev. 141 (1966) 517; H.W. Cape!, Physiro 32 (1966) 966; E.K. Riedel and F.J. Wegner,
Phys. Rev. Lett. 29 (1972) 349.
General spin systems bave been considered by
G. Horwitz and H.B. Callen, Phys. Jl.ev. 124 (1961) 1757; F. Englert, Phys. Rev. 129
(1963) 567.
Appendix A24 follows
J. Zínn-Justin, Penpectives in Particl.es and Fidd$, Cargêse 1983, vol. B126, M. Lévy
et al eds. (Pienum Press, New York 1985 ).

'-==---------------------·~·U.,WI·!.hk' ~'1•'1·~ J
24.7 Mean Field Throry 559

Exercises
Exercise 24.1
Th.e spin 1 Jsing model. Consider the lsing model with a nea.rest neighbour (n.n.) ferro-
magnetic two-body interaction in which thespin S can take three values {0,1,-1}, the
common weight of the conligurations {I, -I} being p and tbe weight of the last conlig-
uration being therefore 1 2p. Study the model in the mean field approximation as a
function of tempera.ture and p.

Exerci.se 2./.fl
The antiferromagn<!tic lsing model. We consider ao lsing-like system on a hypercubic
lattice with antiferromagnetic interoctions:
-fi'H - - L v.1s.s; . v.1 ;:: o. v., = o.
ij

The lattice can be decomposed into two sublattices rorre;ponding to odd or even sites (as
defined by the parity of the sum of components of the position vector). We cha.racterize
tbe sublattices by a sign ~. = ( -1)' asslgned to each site i. We then as51Jllle
l'ijififj - \';.j )
a condition satisfied by tbe n.n. interaction.
Define an arder parameter for the antiferromagnetic transjtjon. Cbara.c:terize then the
effect of a smal! constant rn.agnetic field within the mean field approxima.tion.
&:errise !24.8
Mean field thenry and e!fective Landau-Ginzbmy aclion for modelo wíth ZN symmetry.
One now considers a lattice model on a cubic lattice with short range fenoma.gnetic
interaction in which tbe two- body interaction h as a Z N symmetry. For exa.mple one can
represent the degrees af freadom at each site i by a phase 6; which tai«Js N discrete
values:
8=2wn/N, p=O, ... ,N-1.
The configuration energy tben takes the fonn:
- f3'H = L V.; cos {li; - Oj).
í,j

wíth V;; -O, V., ~O. Tbe N conliguratians on a síte have the same weight.
Note that for N = 2 one simply recovers tbe Ising model. For N = 3 one finds a
model which is equivalent to the 3-state Potts model. Define a.n order parameter for tbe
Z N model, N > l. Discuss the model in tbe mean field approximatíon, in particular the
order of tbe phase transition.

E:rerci.se 24-4
The q-state Potts modeL Tbe model is defined in terms of spin Vl>ria.bles wbich can take
q values with equal weight and a configuration energy of the fonn:

-{J'H = l:Jós,s,.

For q = 2 the lsing model is recovered. Study the model for higher values of the integer
q. A- general review on the q-state Potts model and the percolation problem (q 1) can
be found in
F.Y. Wu, Reu. Mod. Phy8. 54 (1982) 235.

·----------------------~·IA<WI'·t.'l>t.'.J•'J<'I~· JI ,J'I,l!.
560 Mecm Fie!d Thenry A24.1

APPENDIX: 24
MEAN FIELD THEORY: GENERAL FORMALISM

We now explain how a mean field approximatlon can be obtained for a general spin
modei on tbe lattice. Agaln the mean field result will be identified With a saddle point
approxim..Uon, and tberefore will be the first tenn in a systematic expansicm. A brief
account of tbe metbad has already been íncluded in tbe Domb and Green series review
artic!e. We sha.ll also indicate the relation between the mean field apprcocimation and
hígh and low temperature expansions.

A24.1 Mean Field Theory

We consider a lattice model described in terms of some lattíce energy functional .4.
functlon oflatti<:e variables 5; in wbích i is the lattice site, lt ís ne<:essary for wbat follows
to coosider different powers of the same lattice variable S; as ;ndependent vari~>bl... so
tbat we writ.e .4. as .4 (S;,s;, ... ,sn.
The correspaudíng pa.rtition function L<:

Z =f ll.dp(S;Jexp [-.4.(5;, . ._,S;)], (A24.1)

in which dp {5) is the one spin distribution.

In arder to construct the mea.n field appraxilnation as the leeding order in a steepest
descent calculation, it ís oecessary to lirst transform expression (A24.1). We introduce
two sets of la.ttioe variables a) l and HF), the quantities H}ll being Lagrange multipli~,
1

to express tbe constra.ints:

I= 1, ... ,/;. (A24.2)

Then the partition function cs.n be rewritten:

Z=
f TIt
. d p(S,:). d a..!l)dH{I)
TI
t:,l
[ .4. ( (I} (2)
t expl- a-1 ,u, ! ••• ,a1(k))

+ l:: H,1'l (afl- s!) ]· (A24.3)

i.r

We can now reformulate mcne accurately what we rnean by cansidering ali powers of
the lattice vario.ble as independent: the action .4. ís a first degree polynomiBl in ali the
variables .,li)
and contains no product of the form .,;t)afml.
In expression (A24.3) tbe integra.tions over Bll variables Si are decoupled. Let us
introduce the function F0 :

(A24.4)

We here assume tha.t tbe integratiou measure dp (S) is either of compact support or
decreasíng fast enough so that the integral exists. The partition fuoction can then be

~· Jlt.J•t.l:'. 'F\ r,_,=--------------------.,,~,u.,wt·!.hk• ~'i•'t·~ J

A24.l Mean Field Theory 561

rewritten:

z= jn MÍ )dH) l exp[-.4 (a;n, ... ,,.;•>) + LH)

1 1 1laj 1l

w u
l:Fo (n; 0 , .... H)k1)]. (A24.5)
i

The mean lield appraxímation i>; the leading arder in the evaluation of this expression
by steepest descent. The saddle poínt equations are:

(A24.6a)

(A24.6b)

Tbe equation (A24.6b) shows that, at leadíng order, the original weight fl, dp(S,)e-.A
has been approrimated by a product of ~íghts for each lattice site given by the r.h.s. of
equation (A24.4), in which the va.riables H)1> have been rep)aced by their saddle poínt
values, the ~mean fields".
Remarks. lt is c)ear (l(Jm these equations that if A does not depend on a given o (I), the
correspondíng lield H! 1) vaníshes, and botb dísappea.r from the equatíons. Tberefore we
find the same result as if we had omítted them in the first pla.ce. Note also that if A is a
quadratic function of the varíables q!l), the u!1l integrations can be performed explicitly
and only integrais over the conesponding H! 1l rema.in. In Chapter 24 we have consídered
models of thís form and this explains why we díd not need the general formalísm we are
developíng here.
Discussion. Let us now explain why we have introduoed the Ir. setz of varíables u<1l
f!.nd H(l) instead of one as in the example studiad in Chapter 24. In the mean field
approxlmation the variables S; are replaoed by some average va]ue oo that the average
of a prod uct is replaced by the product of averages:

(S;,S;, ... S;.),.... (S; 1 ){S;,) ... {S;.}

lt is pl~>usible that in some limit this can be a good approximation if ali v!ll'íables in the
product are d ilferent. However thi• cannot be true if in the product appears 11 power of
Lhe sa.me variable. For exarnple in the lsing model:

A variable is always correlated with ítself, and the new La.grange pare.meters a!low to
take into a<:count these ~>utocorrelations. To illustra.te this point Iet us consider specific

Eromp!es. Let us ...Sume first that the action A is a function only of S[, then it is
obvious thot we should consider S[ as the basic dynamical variah!e in the mean field
approrima.tíon. The procedure explained above does it automatically sínce only the
JIRt!lllletets "( 2 ) and H( 2 ) will appear. This procedure also solves the following simple

-----------------------~·IA<WI'•t.'l>t.'.J•'J<'I~· JII,J'I,l!. 'l"tr"'=

562 Meo.n Field Theary A24.l

problem. Tetms in the action which rorrespond to one-body pOtentials, í.e. sum of
functions depending only on the va.riable at one site such as:

(A24.7)

could be ronsidered both ,.,; part of the rneasure p (S,) or as part of the action. Thereforc
one oould h ave the feeling tbat the results of meo.n field theory d€pend on the forrnulati011.
The introduction of additional varíables ensures tbat thio is not the case.
Let ru; for exarnple make tbe transforrnation:

.A(s,,s;, ... ,sn . . . .A(s,---,s~) +i ~s~.

Í p(S)- e•5 '1 2 p(S).

Obviously tbe lattice theory is independent of a.

(A24.8)

Consider now tbe rnodifications this transformation induces into equa.tions (A24.4-
A24.6)

exp ( -Ê'o (H!'l, ... ,Ht•l)) =f dS e•S'f2p(S)exp (- ts'H< 1

} (A24.9}

Comparing the equation witb equation (A24.4) it follows:

Fo (H(tl, ... ,H!k)) = Fo (H(ll,Hf2l - ~,HI 3 l, ... , H!'!), (A24.10)

and tbe saddle point equa.t.ions become:

(A2Ul)

We see that the equations are identical up to the change:

Ht'l
t
- ~2 - H<•J
1. •
(A2U2)

Tberefore mean field theory results are insensitíve to such changes.

Flnally we now also understand tbe teclmkal assumption made in Chapter 24:

v.. =0.
lt wa.s just to simplífy the mean fteld formula.tion a.nd avoid tbe introduct.ion of varia.b).,.
related to s'f.
lt is then possible to con·ect tbe mean fie)d apprwdmation, by systernatically expanding
the effective action in (A24.3) around the saddle points (A24.6) and integrating term l>l·
term ..s one does in usual perturbetion tbeory. We caii this expansícm the mean field
expanslon. We have ,.een in Cha.pter 24 in a particular example that a pararneter couW
be associated to tbis expansion. lt is easy to show, followlng the same idee.s, that thú;
<:ll.ll also be clone in this more general framework.

----------------------·~·U.,WI·!.hk' ~'1•'1·~ J
A24.2 Mean Field Theory 5ti3

A24.2 Mean Field Expansion

lf we set a parameter L in front of the eliective action (A24.5), then the mean field
.,xpansíon is just a formal expansion in powen; of 1/ L:

ZL j TI di7) ldHJ l {L[-A (rr)'l,aj

~
1 1
exp
2
l, ... ,")">) + L;n}'>u[1l
~

~ (Hj1>, •.• ,H/k))]}.

F0 (A24.13)

As aiready explaíned ín Chapter 24, the factor L ín front of F0 has, for integer L, the
following ínterpretation: we have reploced eacb variable S by a sum of L independent
VILtiables having the same distribution dp (S).

L
s,=L;s!"l, (A24.14)
a=l

(A24.15)

Lét us introduce the distributíon dpL ( S) of S:

(AZU6)

For L large, dpL (S) becomes a gaus&ian distribution of dispersion v'L. Usíng the defi-
loltion (A24.16) we can rewrite ZL:

/TI
t dpL(S;)exp [-LA (i• 1,... ,s•)]
S· S'f
1 . (A24.17)

see that each interaction term in A is multiplied by a power of L which is equal to

number of sites ronnected by the intere.ction, minus one.
The pa.ramet€t L is just a formal pa.ramet€r since really we want to calculate for
c 1, but its introduction leads to some simple consideratíons. lt shows that mea.n field
al>l>miÚn>ation is exact when we have a large number of identical independent variables
sit€ pruvided the coefficients in front of the interaction terms are scaled dawn by
"PJ>to·prí.at.!d factor. Such a situation is for instaoce approxímat€}y realized when
dimension of space, becomes large. Then in a cell of gíven linear dímension, one
an increasing numher of índependent variables. By re-expanding t€rms in the mean
expansion, one can geDerate a systematic 1/d expansion.

----------------------~·IA<WI'·t.'l>t.'.J•'J<'I~·ni,J'I,l!.
564 Mean Field Theory A24A

A24.3 High Temperature Series and Mean Field

Expression (A24.17) can also be used to show some ínteresting connectlons between mean
field expa.ruíon aud hígb temperature oi low temperature P.><pan.síon. Let US T€Íntroduce
a factor (:l in front of the action A and perlorm an expansíon in poweTS of (3, whích ís a
high temperature exparusion. At arder K in fJ we lind a sum of contribution.s of tbe form

(A24.18)

The factor LK comes from tbe L factor in front of the a.ction. The integers C1 count the
nurnber of sites linked by the corresponding interaction terrn.s. Fínally equatíon (A24.16)
shows that the average of any power of the lattice variable with the weight dpL ( S) is
proportional to L since it ís equal to:

fdpL(S)S'
(A24.!9)
f dpL (S)
Therefore in equation (A24.1B) n is the number of dif!erent lattice variables present in
tbe higb ternperature diagram. For a connected tree diagram one bas:
K
n= L:c1-(K 1). (A24.20)
l-1

Indeed eacll intera.ction term bríngs in C1 independent variables. But to ronstruct a

connected díagr= eoch interaction term must have a variable in cornmon witb nnotber
ínterection term. This suppresses exactly K - I independmt variables,
Now, each time one ndds one loop to the diagram, one suppresses one ndditíonal
índependent varíable. Callíng B the number of loops of tbe diagram one' linds:
I<
n= í:Cr-(K -1)- B, (A24.21)
1=1

The corresponding power of L for tbe diagram then ís:

J{

K LC1+n 1-B.
l-I

Mean !ield e:xpansion ís an expa.ruion in powers of L. There[ore, ata gíven order, it SUIIIS
ali blgb temperature or strong coupling díagrlliDS with the same number of !oops.

A24.4 Low Temperature and Mean Field

The mean fiel<! expansion also contains the low temperature expansion. lndeed let us
rewrite equations (A24.6) wíth a parameter f3 in front of the actíon:

{/3
~-= H~'l
ôFo
.
00(!) ••

(ll
(A24.22)
--(-1) =u; .
&H,

----------------------,~.u.,wl.!.hk' ~,1•'1·~ J
A24.4 Mean F<eld Theory

For large fJ (lav.• temperature or weak coupling), the H(l) variables become louge. To
study thís Jirnit we have thus to ew.luate Fo for H large. This will in general select for
S some classical w.lue S 0 , As a dírect oonsequenre we have:
DF (A24.23)
aH; 1)
In tlm limit the variables u and H play no longer role and one expands arouad the
oonfiguration which dominates the low temperature (la.rge fi) expansíon. Th show this
more explkitly Jet us write the measure as a Fourier transform:

dp(S) (A24.24)

To be able to perform this expansion, dp ( S~ hM to decrease fMt enough for large jS I for
p (p) to be an entíre function. Then for HI ) large, F 0 is given by the sadd)e point:
(A24.25)

which yields:
Fo(Hlll) = 'L,Hms pS+lnp(p). (A24.27)
I
The mean field sw:ldle point equations then are:

ul') = 8F;,) = S: ' (A24.28)

aH,
H!l) fJ aA {A24.29)
• au!')'
In the same nota.tion~ the relevant configura.tíon at low temperature is given by the
equations:
fJdA (A24.30)
dS,
and
d -
S,+~lnp(!') ""O. (A24.31)
p,

Now by summing equatlon (A24.29) cver I after wultiplication by I (S,)'-I, and us-
íng equatiou (A24.25), we repmduce equatíon (A:24.30), while equations (A24.26) and
(A24.31) are identir.at
The mean field expansion is also a partia! summation of the low temperature expansion.
Two final remarkB: The generaliza.tion of this fonnalism to severa! degrees of l'reedom
per site is straightforward. Mean lleld theory can also be generali2.ed to )attice models
wíth ferlllions. The equívalent oi the variables S; are tben products of the form ~.r A .p,,
in which .p, and ià are tbe ferm:ion fields and r A some element of the algebra of 1
matrices. These fields oorrespond to "c:omposite booons•. One also h"" tó use a measure
appropril>te to Grassmann varia.bles.

----------------------~·IA<WI'•t.'l>t.'.J•'J<'I~·ni,J'I,l!.
 r
!

J
ij,

~5 GENERAL RENORMAliZATION GROUP ANALYSIS.

THE CRITICAL THEORY NEAR DIMENSION FOUR

In Cha.pter 24 we have considered lsing-líke systems (a.nd more generally ferromagnetic

systems with O(N) symmetry) with short range interections and det.ermined the be-
haviour of thermodynamical functions neat a second arder phase transitíon, in the mean
lield approximation. We have shown that mean lield theory predicts a set of universal
properties, Le. properties whích are independent of the detailed structure of the ml-
croscopic hamiitonia.n. We have then systematically examíned corrections to the mean
field appraximation. We have found that above íour dimensions these corre<:tions do
not change universal qUtllltíties; on the contrary below four dimensions th" correctioru;
diverge at the eritíca.l temperature and mean field theory ca.n certainly not be correct.
Moreover we have ídentilied in an expa.nsíon around the mean field the most singular
terms at críticality and shown that tbey ca.n be formal)y summed to give a massless 4>4
field theory, a result we shall use in the seoond part of this chapter.
Such an analysis, which rellects the property th ..t different scales of physícs cannot
be deconpled, seems to indicate that below [our dimensione long dístance propertíes are
sensitiva to the detailed mícroscopic structures. Surprisingly enough, however, some
nnivernality survives, although this universalíty is less general than in mean field thoory
in the sense that long dístance propertíes dt'pend on the space dimension, on symmetries
of the system fllld some other qualitativa properties. Th understand this universality
beyond the mean field approximatíon it is necessary to call upon a new idea. The
renorrnalizatíon group will províde us witb the essential tool we need,
The RG theory, as applied to Crítical Phenomena, has been developed by Kadanoff,
Wilson, Wegner and many others. We shall first describe the basíc renormali2.ation group
ideas in a somewhat abstrect framework. The formulation wíll be rather inaccurate and
the arguments largely heuristic. The role of fixed points in hamiltonian space wiU be
stressed. A simple fu<ed point, tl"•e gaussian fixed point, whose predictions coincide with
MFT, will be described.
The abstract RG theory, altbough its formulation is vague, is extremely suggestive,
and indeed it has been implemented in many appraxlmate forros and induced a wealth of
practical calculations. lt is not our purpore to review them here. However, in tbe more
specific context of Quantum Field Theory, BSsumptions at the basis of the renormallzation
group have been clarífied and verified in many eases of physical interest, canfirmi:ng in a
very direct manner the deep connection, first recognized by Wilson, between Quanturn
Fie)d Theory and the theory of Criticai Phenomena. The methods of perturbative QFJ'
then have allowed to efficiently calculate universal quantítiC' for a large class of statístical
models.
Therefore in Section 25.3 we sha\l use what we have learned both from the analysis of
corrections to mea.n lield theory in Chapter 24 and of the relevant eigenoperators at the
gaussían fixed point near dímension (our and construct an approximate renormalization
group for an effective 4>4 field theory. We sball discorer that the RG equations which
appear as a consequenre of the necessity of reuormali?.ation of local lield theories are
directly conneeted with the ahstract RG equations we have introduoed in Section 25. L
We shall undet"Sta.nd that universaiity in the theory of criticai phenomena is directly
related to the property that local field theorico are insensltive to the short distanee
25.1 Generol RenarmaliUltion Group Analysis 567

structure, and physícs can therefore be described by renormalized oorrelatíon functíons,

Conversely, in the statistícal sense, quantum field thoories are always c!ore to crítícality
and their existence, beyond perturbatlon theory, relies, from the abstract RG point of
view, on the presence of IR fixed pQints in hamiltonian space.
Finally we shall be able to solve RG equations in d = 4 E dimem;ions and, foliQwing
Wilson and Fisher, clllculate universal quantities in an .c-expansion

25.1 The Abstract Renormali"ation Group: General Formulation

Let us consíder a general hamiltonian H (4>), function of a field .P(x) (x E JRd). We assume
that '1-1 is expandable in powers of </>:

'1-l(r/>) (25.1)

11nd tbat the Fourier transforms of the functions '1-1,. are regular at low momenta ( &S-
sumption of short range forces or locality). The hamiltonian will in general depend on
an infinite number of parrunetern or coupljng constants.
To the hamiltonian (really the configuration energy) '1-1(4>) corresponds a set of con-
nected correlation [unctions W(n) (:rt,, .. ,>:n):

w<n) (x,, "2•'., ,xn) = [f ]d<t>] <t> (xl)''. 4>{x .. ) e-1m{~)]

connect.,
(25.2)

We want to study the long dístance behaviour of critical correlation functions, i.e. the
behaviour oí w<nJ {À;:,'< .. ' Àxn} when the dilatation parameter À becomes !arge.
The renonnalization group idea. The RG idea is to construct a scale-dependent hamil-
tonian '1-1:. (4>} v.•hich has correlation functíons wt> (xi) satisfying:

(25.3)

wh.,re the functions R(n) decrease faster than any power of À for À - 00. The mapping
'H (oi>),_. '1-IJ. (4>) is called a RG transformation. We define the transformation such that
'H:.~) (<I>);;;; '1-1 (</>).
In the case of models invaríant under space translations, equation (25,3) a!ter a Fourier
transformation reads:

(25.4)

V,.rious RG transformations differ by the form oí R(") and the function Z(Ã). In explicit
ronstructions the R(n) are generated by tbe integration over the la.rge momentum modes
<>Í oi>(:.:). When R(n) =O and both spru:e and field are oontinuous variables, the RG
lrnnsformation can be implemented by a rescaling of spoce and field. However, tbis
tronsformatíon has a fixed point only in exceptional cases (gaussian models) and thus
lllOr<' general transforrnations bave to be considered (see Appendix AlO. I). Below we
olu<!l omit the terms R(n) and thus equalities between conelo.tion functions wíll mean up
lu terms decreasing [21Ster than any power.

The fixed point hamiltonian. The coupling constants appearíng in '1-1, are now ali
r>f>lidt functions of À. Let us &Ssume that we have found a RG transformation such

-----------------------"'~·IA<WI'·t.'l>t.'.J•'J<'I~·ni,J'I,l!.'l"tr_""""""'"
568 Genernl Re11ormalizatíon Group AnalyJis 25.1

that, when), becomes large. the ha.miltonian 'H., (,P) has a limit 'H."(,P), the lixed poíno
hamiltonían. If such a fixed point exísts in hamiltonian space, then the correlatim'
functions wt> have corresponding limíts w}"l and equation (25.3) becomes:

(25.é·)

lf we introduce a second scale parameter jJ and calcuJate W(n) (.\p,) from equatí<>H
(25.5) in two diffarent ways, we obtain a relation involving on1y w!"l:

(25.U)

whh:

Equation (25.6) being valid for arbitr..ry iJ immediately implies that z. forros a repre-
sentation o f the dilatation semigraup:

z. (J•,) z. (.\,) = z. (>.,>.,). (25.8)

Thus under reasonable assumptions:

(25.9)

Th2 lixed point correlation functions have a power law behavíour characterized by •
poshive number ~ which is called the dimension of the order parameter 4>(x).
Returning now to equation (25.7) we conclude that Z(>.J also has asymptotically "
power law behaviour. Equation (25.5) then •hows that the correlatiou functions w<nl
have a scaling hebavíour at l..rge distances:

(25.10)

with a power d.p which is a property of the fixed point. The r .b.s. of the equation, whic],
determines the criticai behavíour of correlation functions, therefore depends only on th"
lixed poínt hamiltonian. In other words the correlation fuuctíons corresponding to all
hamiltonia.ns wbich ftow after RG transformations into the sarne fixed point, have the
same critica! behavíour. Thi!l property is called universality. The spa.ce of hamíltonian•
is thus divided into universality classes. Unlversality, beyond mean lield theory, re!ie;
upon the exlst.ence of IR lixed points in the space o f hamiltonians.
Applied to the tW&point function this result shows in particular that lf 2d~ < d th~
correlation length € diverges and therefore that the corresponding namíltonians are nec.
essarily critica!. Criticai hamiltonians define in hamiltonian spaoe the critical surfo.cr
whích is invariant under a RG liow. In the generk case where { is finíte the correlatiou
length f./),. rorresponding to 'H>. goes to zero. The Fourier camponents of correlation func-
tiollll become momentum independent and thus carrelati<>n functions become 6-functioru;
in space. This trivial fixed point corresponds to 2d.; = d.
Renormalization group equatíons. Let us now try to write the RG ~ransformadon,
conseque~ of equation (25.5), for the harniltonian itself. For this purpose we assum<·
that the dilatation parameter can be varied continuously (on the la.ttíce this can only

-----------------------M~,u.,WI•!.hk'l'''''i·~
25.1 Geneml Renormnluation Group Analysis 569

be implemented approximately) and perfonn an additional small dílàtation which leads

from the scale À to the sca.le >. (1 + d>.j).).
We look for RG transformatlons whích <~•ill bring the traru;fonned hamiltonian doser
to a fixed point. This condition involves only 'HJ. ànd not the previous history which ha.'
led !rom 'H. to 'H.,, Therefore we ca.n write the RG transformation in differentia.l form
in terms of a mapping T of the space of hamiltonians into itsel! and a rea.l function T)
uefined on the space of hamiltonians as:

(25.11)

(25.12)

As a function of the "time" In>. these equations define a Markov's process. Note that we
have written the r.h.s. of equation (25.12) in an unnatura.l way for !ater convenience.
A fi.'<ed point hamiltonian 'H" is necessarily a solution of the fixed point "'luation:

T!'H"l 0. (25.13)

The dimension d<l> of the field rf> is then:

(25.14)

To understand, at least locally, which hamiltonians flow into the fixed point it is nocessary
tn study íts stabílity. We therefore linearize the RG equations near the fixed point.
Linearized RG equations. The RG transfonnation being determine<!, let us apply it to
n hamiltonía.n 1i dose to the ftxed point 'H.*. Setting 'H A = 'H* + l!.1i 1 we linearize the
RG equations:
(25.15)

~·hereL* is a linear operator independent of )1, acting on the hamiltonian space. Let us
IL"ume that L• hos a discrete Sét of eigenva.lues I, corresponding to a set of eigenoperators
0;. Then l::.'HA can be expanded on the O;'s:

(25.16)

;md the transformation (25.15) becomes:

(25.17)

The íntegration then yields:

(25.18)

Clasrifimtion oi eigenvectors or scaling fields. The eigenvectors O, can be classified

Imo four families depending on the va.lues of the eigenvalues !,:
(i) Eigenvalues whlch have a pooitive real part. The corresponding eigE'Iloperators
'"" ca.lled relevant. ]f 'H.~ has a component on one of these operators, this component
will grow witb >., and 'H.} will move away from the neighbourbood of 11'. Operators
"'"'ciated with a deviatíon from critica.lity are dearly relevant since adilatation decreases

-----------------------.,,~,fA<Wf',t.'l>t.'L•'•'I~· 1.'1,'11. 'l"tJ:i~

570 Generol Renormalization Group Analysís 25.2

the elfectíve correlation length. In Section 25.5 we calcnlate the corresponding eigenvalu<·
for the q,• field theory.
(ií} Eigenvalues for which R.e(l,) "' O. Then two situations ca.n nríse: either lm(l,)
does not va.nish, a.nd the corresponding component would have a periodic behaviour
(no eJUlmple will he met in this work) or exceptionally I; "' O. The eigenoperator.
corresponding t<> the eigenvalue zero are called margínal. In Section 25.5 we show tbat
in the <J>4 (x) field throry the operat<>r </>4 {z) is marginal in four dimensions. To solw
the RG equations (25.17) and determine the behaviour of the corresponding effectiV<"
components it is ne<::essary to expand heyond the linear approximation . Generically OIH'
obta.ins:
(21>.19)

which mea.D$:
,., ().) ~ 1 fln À. (25.20)

The eristence of a marginal operatúr usually leads t9 a logarithmic approach to the lixed
point.
An exceptional exnrnple is provided by the XY model ín two dimensiOill'l (0(2) sym-
metric non-linear a-model) which inst.ead of an isolated fixed point, has a line of fixed
points. The operator which corresponds to a motion along the line is obviously marginal
(see Chapter 32).
(ili) Eigenvalueswbich have negative real parts. Tbe corresponding operat<>rs are called
irreie,ant. The effective componetrt.s on these operators go to zero for large dilatations.
All these eigenvalues, which are cha.-acteristic of the fixed point, may appear in Lhe
&eymptotic expaD$íOn at large distances of the correlatíon functioWI correspondíng to
some criticai ar near-critic:a.J ha.mlltonían.
(iv) There nre also operators which do not a.Hect the ph~ics. An exaiDple is províded
by Lhe operator reallzing a canstant multiplícative renormalization of the dynarnicol
variables ,P(z). The correspouding operators are called redtindant. In quantum field
theory quantum equatíon of motl011S correspond to redundant opera.t<>rs with vanisbing
eigenvalue.
Ciassíjialtion of fi3;ed points. Fixed points can be classified according tú their local
stability properties, i.e. to the number of releva.nt operators. This number is also the
number of pa.rarneters it is necessary to fix to impose to a general hamiltonían to be on
the surface which Hows into the fixed point. For a non-triviaJ fixed palnt corresponding
tú criticaJ hamiltoníans it is the codimellSion of tbe critic:a.J surfa.ce.

25.2 The Gaussian Fixed Point

In th'Js g<!neral frarnework, we have shown tha.t universallty emerges as a consequence oi

tbe e:dstence of fixed points in hamiltonían space. To be able t<> quantitatively describc
the critícaJ behaviour, we therefore have to construct RG Hows a.nd to find their fix>:d
points. A global analyo;is ha.s never been performed. In general one is able only to exhíblt
a few futed points and study thei< local stabüity.
However a subspace of barniltonian space can be cornpletely explored and tbe RG
transformations expllcitly constructed: the subspace of quadratic hamlltoniàllS (free field
theories in the field theory languoge) which correspond to gaussian Boltzmann weight...
In tbis case fixed polnts ca.n be identified, criticai ha.miltonlallS c<>nverge towru-d Lhe

----------------------'~C."j:,),!.i•k'l
25.2 General Renormalization Grottp A nalysi.• 571

sn-called gaussían fixed point which, as we shaU see below, corresponds to mean field
theory.
We thus consider a general quadratic hamiltonian invariant under space translstions.
In tsrms of the Fourier components J,(q) of rj.>(x) it reads:

(25.21)

In this case We can take R(~) 5 O in RG tra.nsformations.

For short range intsra.ctions, the dass o[ interactíons considere<! in Chapter 24, the
Fourier transform u( 2 l(q) has a regular exparu;ion for smsll momentum:

(25.22)

For a ,_Otbe hamiltonian is non-criticsl. lf we take for renormali.zation factor Z(A) =

>.-d we obta.in the trivial lixed point h.(q) constant, with a va.uishing oorrelation
length. As functions of space variables correla.tion functions have a 6-function limit
(d~=d/2).
For a= O (criticai hamiltonian)

(25.23)

H we tske:
Z(A)=>.-(d-•l, =* d;.=!(d-2), (25.24)
we find instead the gaussian !ixed point:

(25.25)

Eigenoperotors. Let us now perform the eigenoperators analysis for the gaussia.n fixed
point (25.25) wlthin the space of even hamiltonians, which spa.n the subspace of ha.mil-
tonians having an lsing-like symmetry. We write !:,. 'H. as:

{25.26)

lt is easy to veri.lY that equation (25.3) leads to the relation:

u~") (q;) = Àd zn/2 (À) u(") (qi/.l.) .xd- ..(d-2)12,..,.. ) (q,j).}. (25.27)

TI"' equivalent o[ equation (25.17) for the fuced poínt (25.24,25.25) is then:

(2r..28}

Equations (25.27,25.28) show that the eígenoperators are homogeneous functions of the
dynamical variable q, and of the momenta q; \'·•· of the number o[ derivatives in spe.ce}.
we have assumed tbat the function'l u n) (q;) are regular, the eigenfunctions are
bomogeneous polynomials o[ degree 2k in q, a.nd the corresponding eigerivalues l,,k ate:

l,,k ""d- 2n (d- 2) - 2k. (25.29)

----------------------·~·IA<WI'•t.'l>t.'L•'•'I~· l.'l,l1.'1'tr_~
 General Renornwiizatwn Group A nalysis 25.3

(i) The operator n "' 2, k = O is relevant and corresponds t.o a deviation from the
crítica.\ tempera~ure.
(ii} The opera.tor n = 2, k = 1 is redundam, it corresponds to a simple renormalization
of tbe dvna.mícal variable.
(iii) Ábove dímensíon iour. a.ll other operators are irre!evant: on the criticai surface the
gaussil'lll fu:ed point is stab\e. At dimensíon [our, J \!14 (:r)ddx ís marginal and logarithmic
corrections are expected.
Below dimensíon four J \!14 (x)ddx is relevant 311d wben the dímension decrelL""" addí-
tional operators become relevant too. The gaussían fixed point ís IR uostable.
Jt is clear that this anal_vsís is quite similar to the analysis of conections to mean
field approxíma.tion. lt ís indeed straightforward to verify tha.t the gaussian fixed point.
reproouces ali remlts of mean fielrl theory.
lt is al60 strictly equivalent to pmver counting in qua.ntum field theory as dis<Ous.'ed in
Chapter 8. Tbe numoer of rel€Vant aperators is just the number of parameters an whlch
depends the renormalized lield tbeocy.

25.3 The Critkru Theory near Dimension Four: The Effective Field Theory

The ma.in difficulty with the previous appraach is tba.t lt requires an e>:plícit construction
o[ renormalization group transformations for hl!llliltonians, which has a chance to lead to
a fixed point. Although the general idea is to integrate over large momentum modes of
the dynamical variables, its practícal implementation is fa.r from being stralghtforward.
We shall therefore write RG equations directly for correlation functio:ns. The limítatíon
of the method we use is tha.t it is a.pplícable only when there exists a fixed point whjch,
in a sense which will l>ecome slowlr clearer, is close to the gaussían fixed point.
Jn Sectíon 24.6 we have shown that, for an lsing-like system wíth short range fer-
rom&gnetic interactions, in the criticai domain and for d ::;; 4, the sum of the mos1
dívergent contríbutions arder by order in o. me311 field expansion ca.n te reproduced by
an effective local field theory whose action is just given by tbe first relewmt terms of
Landau-Ginzburg- Wilson hami!tonia.nr

1í(tl>) e {3H f d"x [~c(V,P) 2 + ~aq\ 2 +b~4> 4 ], (25.30)

with a, b and c being regular functions o[ the temperature for T dose to T<.
Consistently the analysis of the stabílity of the gaussian fu:ed point has shown that in
four dimensions the 4>4 interaction becomes marginal while ali <>ther interactions remain
irreleva.nt. 1f tbe dimensions of operators ace oontinuous functions of the space dimension.
hamiltonian (25.30}should contain ali rel""=t operators at least in some neighbourhood
o[ dírnensíon four.
The hamiltonian (25.30) generates a perturbative expansion of field theory type whicb
can be described ín terms of Feynman diagraros. These have to be calculated with a
cut-<>ff of arder 1, rellection of the inltial microscopíc structure. We shall eventual))·
show that the precise cut-olf procedure is irrelevant except that it should satisfy somo
general conditions. For example the propagator can be modified (as in Pauli-Villars's
regula.rization) but the inverse propagator in momentum space must temain a regular
function of momentum (the forces are short range).
In this chapter we restrict ourselves to Ising-like systems, tbe field q\ has only one cOm·
ponent. Generalizations to the N-vector tnodel wíth O(N) symmetry, which is stralght-
foTWB.rd, and to several-component models with more tban one 4>4 coupling constant will
be brtelly analyzed in Section 26.6.

-----------------------.,,~,u.,wl.!.hk'l •'•'1·~
25.4 Generul Renormalüation Gmup Analysis 573

\Ve have a.lread~- seen in Chapter 24 that a convenient way to study the problem of
infrared singulariLíes isto rescale ali space or momentum variables, and mCMure distances
in units of the correlation iength, or, at the criticai temperature, in some arbítrary unit
much larger than the lattíce spacíng and correspondíng to the typical distances at whích
correlations are measured. After such a rescaüng the momentum cut-olf becomes a large
momentum A analogous to the cut-off used to regularize quantum field theory.
Let us perform such a resca.ling here, and resca.le also the field ,P(x) in such a way that
2
the coefficient of l'i7f/>(x)J becomes tbe standard 1/2:

x-Ax 1 (25.31)
<P(x) ~ (f/>(x). (25.32}

After this resca.ling ali quantíties have a dirnension in units of A. Our choice o[ normal.
ízatíon for the gradient term implies:

(25.33)

which shows that 1/> now has in terms of A its canonical dimension d/2 - L The actíon
'H (I/>) then becomes:

'H (I/>)= j d"x { ~ l'i7,P(x)J 2

+ ~r,P2 (x) + ~gA•-•q,•(x)}, (25.34)

with:
r=nA 2 /c, (25.35)
Let us ca.ll r 0 the parameter v.•hich corres1>0nds, at g fixed, to the critica.! temperature T,
at wbich the correlation l<lngth { diVerges. In terms of the scale A the crítica.! domain is
then àefined by:

physica.l mass C 1« A ~ Ir -r<I « A2

distances » 1/A or momenta « A, (25.36}
magnetizatíon M {.P(x)) «C'~ A(df 2 )-l.

Note that these conditions are met if A is identilied with the cut-off af a usual field theory.
However aiJ ínspection ofthe action (25.34) a.lso shows that, in contrast with conventiona.l
'IU"ntum field theory, the </1 4 coupling constant has a dependence ín A gíven a priarí. For
d < 4 the 1/>4 coupling i.<; very hu-ge in terms of the scale relevant for the critica.! domain.
ln the usual formulation of quantum field theory instead the bare coupling constant is
~t!so an adjustable parameter. ln the latter case the resulting quantum field theory is less
~eneric.

25.4 Renormalization Group Equations for the Critica) Theory:

The E-Expam;ion

The barniltonían (25.34) can now be studied by field theoretíca.l methods. Using a power
count.ing argument, we have shown in Chapter 24 tbat the critical theory does not exíst
in perturbat.ion theory for any dimension smaller than 4. lf we define, by dimensional
o:ontínuation, a critica.! theory in d 4 -e: dimensions, even for arbitrarily sma.ll e there
nlways exists an order in perturbation ( ~ 2/~J at which IR (infrared) divergences appear,

-----------------------"'~·IA<wn.:l>t.'L•'•'I~· 1.'1:1 .
574 Generol Renormalization Gmup Analysis 25.4

Therefore the idea, orígina.ll~· due to Wilson and Físher, ís to perform a double seríes
expansion jn powers oi tbe ooupling constant g and €. Order by arder in this expansion,
tbe critica! behaviour differs from the mean fteld beha.viour only by powers of logaríthm,
and we can construct a perturbatíve critica! theory by a.dj usting r to its critical value
rc(T Te).
To st.udy the large cut-off limit we then use method• developed for the construction
of the renormalized ma.ssless t/14 field theory. We introduce rescaled correlatíon func.
tions, defined by renormalization condítíons at a new scale IL « A, and functions of a
renormalized ooupling constant g,:

(25.37)

in whlch Ih is a numerical vector. These correlation functíons are related to the original
ones by the equatíons:

r~") (p;; g, ~-'• A) zn/ 2 (g, A fi-') r<"l (p;; g, A). (25.38)

Renorrnalization theory (more precisely a slightly extended version a.dapted to the r-

expansion) t.ells us that the functions r!"l (p;; g., ~. /1) of equation (25.38) ha.v.! ai P;,
g, and ~ lixed, a largo cut..off limít which a.re the renormalísed correla.tion functloDS
r!•l (p,;g,.p). A det8ilad a.na.lysís actua.liy shows:
(25.39)

in which the power of In h increases with the ordet in g 81l.d fi (5'*' equatí<>n (25.45)). Fur-
therrnore the renorrnalized functioiiB ri•l do not depend on the speciftc cut-olf procedure
and, given the normallza.tion conditíons {25.37), are therefore universal. Since the renor-
malized functions r!"l a.nd the initial ones f!•l are aBymptotically proportional, both
functions have the same srnall rnornentum or large distance bebaviour. The renorrnalized
functions thus conta.in the whole information about the universal crítical behaviour. lt
follows that .we could restrict ourselves frorn now on to the study of the renormalized
field theory. And indeed it is the only one to be really useful for explicit calcula.tjons o[
universal quantítieo (see Chapter 28). However, we sball eventua.lly be inter~ not only
by the asymptotic criticai behaviour, but also by the corrections to the asymptotic tbe-
ory. Furthermore, renormalized quantitíea are not directly obtained in non-perturbative
calcula.tíons. For these reasons it is quite useful to express the irnplicatíons of equatkm
(25.38) directly on the lnitial theory.
Bare RG eqtWtioru. Differentiating equatíon (25.38) with respect to A at g, 8lld fJ
lixed, and taking into account (25.39), we can derive a new identity:

(25.40)

In the first parto! our study we neglect cor..ections subleading by powers of A. We shall
retum to thls polnt in Chapter 27.

----------------------'~•1-"I.,WI·!.hk'l'''''i'~
25.4 Geneml Renarmalízation Çmup Analysi!; 575

Then, us;ng chain rule, we can rewríte equatíon (25.40) as:

(25.41)

The functions {3 and "/, which are dimensionless 8.1ld may thus depend only on the di-
mensionless quantities g and Ajp, are defineà by:

{J(g,Afp) ~ A:AI..... g, (25.42a)

rJ(g,AJp)= -A:AI 9HJ.l

lnZ(u,A/J>l. (25.42b)

However, the íunctions {3 and "'can a.lso be directly ca.lcula.ted from equation (25.41) in
terms of functions fln) wbich do not depend on p. Therefore tbe functioru; {3 and I) cannot
depend on tbe ratio Ajp.. Of course, if we examine tbe deftnitions (25.42) we see tbat
A
the r .h.s. has a weak dependence in Jp, 'hut consist.ency requires tbat this dependence,
whicb goes to ~ro líke some power of p/ A, should be negiecteà, as in equation (25.40).
Then equation (25.41) can be rewritten:

~'l(!J)] f(n) (p;; g, A)=' O. (25.43)

Equatlon (25.43) ís aD equation satlsfied when the cut-olf ls large by tbe pbysica.l corre-
lation functions of statlstíca.l mecbanics which are a.lso the bare correlation functions of
quantum field tbeory. It U;, as ""' have sbown, a direct comequence of the existence of
a renorma.lizeà theory. lt will be implicit in the solution of equation (25.43) that it a.lso
implíes the existence o[ a renorma.lizeà theory. Both statements are thus equivalent.
Beyond leading aroer. Let us fina.lly <:har3Cterize more precísely the approxima.tion
involved in writing equatiou (25.43). In a series expansion in powers of g and c:

... {25.44)

the coefficients r~~) have an asymptotic expansion for IPil/ A small of the form:

(25.45}

wjthL(n,r,s) J+r+s nj2forn 4WidL(n,r,s)=r+s-n/2forno/:4.

Tbe RG equations (25.43) are exiiCt for the sum of the terms which do not vanish for
A large, as can be verified by expanding equations (25.43) in powers of g and F:.

~---------------------·~·IA<Wf',t.'l>t.'L•'•'I~· l.'l,l1.'1'tr.~
576 Generol RenormaJüation Group Analyois 25.5

25.5 Solution of the RG Equations

Equa.tion (25.43) ca.n be solved by the method of charaderistjcs, One introduces a

dilatation parameter À and looks ror functíons g (>.) and Z (>.) such that:

(25.45)

DiJierentía.ting explidtly with respect to À, we find that equa.tion (25.46) io coDEistent

with equation (25.43) provided tha.t:

d
ÀdÀg(>.) tJ(g(!l)), g(1) = g; (25.47a)
d
>.d>. lnZ(>.} ='l(g(>.)). Z(1) = 1. (25.47b)

The function g (),) is the effective coupling a.t the scale À, a.nd is governed by the llow
equation (25.47a). Equatiow; (25.46) and (25.47) imply:

lt is actually convenient to "'""ale A by a factor 1(>. and rewrite the equa.tion:

r!nJ (p;;g,Aj>.) = z-n/ 2 (!l)r!"J (p.;g(>.),A). (25.48)

Equatioru; (25.47 ,25.48} implement approximately (becauee terms sublea.ding by powers

of A ha.ve been neglected) the RG ideas a.s presented in Section 25.1: sincé the coupling
consta.nt g (J,) characteri.zes the ha.miltonian 1-i.>., equation (25.47a) is the equivalent of
equa.tion (25.11) (up to the change), -1/-"); equa.tions (25.12) and {25.47b) differ only
by the definition of Z(À).
Let us now solve equa.tions (25.47):

l
g(!o.)~
/3(g') In>., (25.49a)

f'
9

d<T
-11(9 (c))= lnZ ('>.). (25.49b)
1 "

The equation (25.43) is the renormalization group equatlon in dilferential form. Equa-
tioru; (25.48) a.nd (25.49) are the int.egrated RG equations. In what fullows we explicitly
assume that the RG functions tJ(g) a.nd q(g) are regular functions of g for g ?. O. In
equation (25.48) woo see tha.t it is equivalent to increase A or to decrease À. To investí-
gate the large A limit we therefore study the behaviour of the effect:i"" coupling constant
g(À) when >. goes to zero. Equa.tion (25.49a) shows that g(À) increases if the function
fJ ís negative, or deereases in the opposite case. Fixed points correspond to zeros of the
p.function wbich therefore play an essential role in the a.nalysis of the critica! hehaviour.
Thoee wbere the fj-function ha.s a negatiw slope are IR repulsive: tbe effectíve coupling
moves away frow such zeros, except if the inítial couplíng has exactly a. fixed paint value.
Converscly those where the slope is positlw are IR attractive.

---------------------,~.u.,WI·!.hk•t·,·•'l·~
25.5 Geneml Renonnalization Group Analyms 577

The RG functions f3 and fi can be calculated in perturbation theory. From the relation
be~n bsre and renonnalized coupling constant:

p.'g, =A' (g- 1 6~2

3
(In;+ const.)] +O (g
3
,g 2
c), (25.50)

(e 4 d) and the delinition (25.42a} it follrm'll that:

2
3g ( 3 2 )
/3 {g,<) = -tg + 16 ..2 +o g ,g t: • (25.51)

We see from the explicit EX.pre;sion (25.51) of the {3-fnnction that above 4 dimensions,
í.e. e < O, if initia.lly g is runall, g(.\) decrea.ses approe.chíng the origin g "'O. We recover
tha.t the ga.ussian fixed point is IR stable.
Below 4 dimensions, if initially g is very small, expression (25.51) shoW13 tba.t g{.\) first
increases, behaviour reftecting the insta.bility of the ga.ussi= fixed point. There are then
three possibilities:
{i) Tbe ,9-function remains o.lways negative, g (À) increases and tends towards oo witb
À.
(ii) The 13-function remaill5 negatíve for ali g but the integra.! in equation {25.49a)
converges at infinity. Then there exists a maximal value of À for which tbe equation has
a solution. The RG method then fail.s, a.t least in this simple formulation.
(iü) The ,9-function has a zero g",

/3(y") o. (25.52)

Then equation (25.49a) shows that g p) has g• as an asympt.atic limit. Let us cal! w the
slope at su ch a zero:
w=[J'(g")>O, (25.53)
ond let us linearize equation (25.49a) a.round the fixed point:

1
9(~) dg'
-In.\. (25.54}
g w(g' g• )

lntegrating, """' get:

lo(>.)-g"l-
~-o
.\'". (25.55)

The explicit expression (25.51) shows tha.t, in the sense of the c-éXpansion, fj(g) has
htdeed snch a zero:

161f2
g" = -3-e+O(c2), (25.56)
w t+0(E 2). (25.57)

. belaw dimension 4, at least for 1: infinitesimal, this non-ga.ussian fixed point is IR stable.
dirnension 4 thls fixed point merges with the gaussian fixed point a.nd. the eigenvalne
vanumes indicating the appea.rance of a. marginal operator a.lready identified in the
o{ the gaussian fixed polnt in Section 25.2.

----------------------,~·IA<Wf',t.'l>t.'L•'•'I~· l.'l,l1.'1't11i"""""'"'
578 Geneml Renonn«lízation Gn:mp Ana.lysú 25.5

Let us now assume that r<n) (g') and Ti {y') are finite, conditions which are sa.tisfied
within the fnunework of the E-expansion. From eqnation (25.491>) we derive the behavioUT
of Z().) fm ). small. The integral in the I. h.•. is dominatoo by small values of o. lt follows:

(25.58)

·~there l'!'e he.ve set~

'l"" T/(9').
Equation (25.48) then determines the behaviour of r<nJ (p;; g, A) for A large:

r<nJ (p;;g,Aj>.) ~ >.-n•f2r<n) (p,;g',A). (25.59)

On the other hand from símple dimensional considerations we know that:

(25.60)

Combiníng this equation with equatjon {25.59) we obtain:

r!n) (.l.pi;g,ll.) >.:o >,d-(n/2}1d-Z+"lr!n) (p;; g', A). (25.61)

This equation shows that the critica! correlation functions have a power law behaviour
for small momenta, índependent of the initial value of the 41 4 coupling constB.Ilt g.
Setting n = 2 in equation (25.61) we obtain in particular the small momentum b€-
haviour of the inverse two-poínt function. For the correlation function w!2l (p) tbis
yields:
(25.62)

We therefore verify that the delinition of equation (25.58) coincides with the usual def-
inition of the critica! exponent Ti· The spectral representation of the two-po!nt function
(Section 7.4) implies fJ > O. Sínce in perturbation tbeory the lirst contribution to tbe
field renormalization Z (g, A/1-'l arises at order g\ '1(9) is of order g2 , and 'l of order ;: 2 •
A short calculation yields:
(25.63)

A semi-quantitative predíction of the theory is that 'I ;,; numerically smllll in three dí-
mensionB.
Finally let us note that equation (25.61} can De interpreted by saying that the field
ll(x), which had at the gaussian fixed point a. canonical dimension (d 2)/2, ba.s now
acquired an anomalous dimension d 0 (equatíon (Z5.14)):

Ali these results call for a few comments. Wíthin the framework of the ;:-expansion,
we have shuwn that ali correlation functions have, k>r d < 4, a long dlsta.nce l>ehaviour
di!Jerent from the one predicted by mea.n field theory. ln addition the critica! l>eba.viour
does not depend on the ínitíal value of tbe .p• coupling constant g. At least for E small
we ca.n bope that the B.llalysis of leading IR singularities of Cbapter 24 rema.ins valid and
thus it does not depend on a.ny other coupl!ng eitber (this point will be furtber discussed

'"==---------------------,~.u.,wl.!.hk't·,··'l·~
25.6 Genernl Renormalúation Group A nalysis 579

in Se<:tion Z7.4). Therefore the crítica.! beha.viour is uniuer•al, although leso univernal
than in MFT, ín the sense that it depends only on some small number of qualitatíve
features of the system under considera.tion.
Moreover the correlation functions obtalned by neglecting, in perturbatíon theory and
withín the <-expa.nsion, power la.w corrections when the cut-olf is large, and which sat-
isfy exactly RG equations (25.43), define implicitly a <me pa.rameter fa.mily of crítica.l
hamiltonia.ns which correspond to a RG trajectory which goes from tbe neigbbourhood
ofthe gaussian fixed point g O which is IR unstable below 4 dimensions to a non-trivia.l
stable fixad point g'.
Finally the conslstency of thls analysis bared on the E-<'!Xpansion relles on the followíng
observatlon: the IR divergences found in the fixed dímensíon perturbation theory are
geverated by an expansion around the wrong, since IR repulsive, fixed point. The E·
expansion allows us to interchange limits nnd to follow perturbatively the attractive IR
fixed point.

25.6 Criticai CoJTelation Functions witb ~2 (:z:} lnsertioos

In the next chapter we sha.ll •tudy the whole critica.! domain (25.36). In particular tbe
following method, already explained in Section 10.10, wi!l be used: we shall expand the
correlation functíons for T -1- T< in terms of criticai correlation functions witb J dd.:r ,P2 (>:)
insertions. Tbe operator ~ 2 (:t) has a direct phyoiCI>i interpreta.tion. It ís the most singular
part (i..e. the most relew.nt) of the energy density (25.30).
Therefore we now discuss the long dlstance properties of the criticai mixad correio,.
tion functions of the order parameter ~ e.nd the energy denslty ~~2 • We denote by
rfl,n) (q1 , ... ,ql;Pi>•" ,p,.;g,A) the IPI functions of n~(>=) fi.elds and l ~~ (>=) opera-
2

tere, (witb (i +n) ~ 2). Renormalization theory teUs us that we can define renorma.li:.ed
correla.tion function5 r!l,n) (q;;p;;g,#) which in addition to condítions (25.37) satiBfy:

rl•.2l (q-" p,· g

r ' h ' r,
P)i Ff=pJ=-112. -1
PJ":Pl'"""-j,u2 _...... '
(25.64)
rP.o) (q -q· gr, ~'>i 42-jiJ.::il =o ,
,. ' J

and are related to tbe original ones (see equation {10.62)) by:

l~"' znl• ( ~ )' [r(l,n) (q;;p3; g,A) 6,.o6,,A-< A] r!1·"l (q;;Pi;Y,,!i). (25.65)

~(g,A/Ji) and A(g,A/J.l) are two new renormalizatiov constnnts.

Differentiating with respect to A at g, aDd J.l fixed, as we have dane in Seetíon 25.4,
R!ld using chain role we obtain a set of RG equa.tions:

[A!+ f3(g) :u- ~fl(g) -lrn (g)] r!l.n) = ~.. o61,A-' B (g). (25.66)

In addition to {3 and '1 (equatíons (25.42)) two new RG functions, '12(9) and B{g), appear:

rn(g) =-A :A\ ln[Z.(g,A/t<)/Z(g,Af,u)), (25.67)

9n.J.'

B (9) = [A :AL..P- 2'12(Y} -c] A (g,A(t<). (25.68)

---------------------_.,~,fA<WI'·t.'l>t.'L•'•'I~· 1.'1,'11. 'l'tr.!>.=

580 General Renarmalizahon Group A nalysis 25.6

Again the RG functions T/2 (g) and B(g), because they are cnlculable in terms of correla-
tion functÍOilS wbich do not depend on p,, cannot depend on the r a tio A/ ll· Note that for
n = o, 1 2, the RG equation (25.66) is not hornogeneous. This is a consequence of th~
non-multiplicatí..., character of the renormall;,atíon in this case. The multiplc insertions
of operators of higher dimension líke (i}(x), lead to even more complicate RG equatíons.
We have presente<! a detailed discussion of this question in Cbapter 12 and we want only
to here wa.rn the reader against a too naíve applicatíon of RG ideas.
ln the homogeneous case equatíon (25.66) can be solved exactly in the same way
as equatíon (25.4.1), by tbe metbod of characteristícs explained in Section 25.5. lt í:s
necessary to introduce a new se ale dependent function (,(.>.) associated with tbe RG
function 112 (g):

wbich therefore satisfies:

d
>. d.\ In (2(>.} = 112{g(>.)].
The arguments given in Section 25.5 then show that the crítical behav:iaur of r<l,n) ís
govemed by the IR fixed point g•.
'Ib relate tbe RG function 112 (g) to standard exponents we also introdm:e tbe function
v(g):
(25.69)
For .), small (o behaves like:
ln(2 ~ (1/v -2) In.>.,
where we have defined a neW criticai exponent v by:

Combíning again the solutíon of equo.tion (25.66) with simple dimensional analysis W\'
finally obtain the behav:iour o[ rU,nl:

(25.70)

Using equâtions (25.67,25.69) it is easy to cruculo.te v(g) at one-loop order:

2v(g) = 1 + - g ~ + O (g') . (25.71)

32"
At the fixed point g = g• (equatíon (25.56)) we obtain the exponent v:

2" =l + •6 +o (•') .
The < 1>'1P2 > co=latíon functíon, Tbe 4>2 (energy density) two-poínt funetion rl 2 •0l
satisfies an inhomogeneous RG equation. To solve it one first looks for a particul!ll'
oolution, which ca.n be cbosen oftbe form A-'C2 (g):

{) (g) Ci (g} -I<+ 2'72 (g)l C2 (g) = B (g). (25.72)

----------------------'~•U.,WI·!.hk'l'•'•'l·~
25.6 General Renonno.lization Gro"P Analysi.s 5Bl

The solution is uniquely determine<! by imposing its regularity at g g•.

Tbe general solution of equation {25.66) is then the sum of this particular solution
BJid of the geueral solutíon of the homogeneous equation whích has a behaviour given by
equntíon (25.70):
r<>,•> (-\q;g, A) (25.73)

The dimen.sion of the q, 2 operntor. From tbe scaling behaviour {25. 70) we can derive
the dirnension of the operat.or r1>2 at tbe IR ftxed point. Let us more genera.lly consider
the n-point correlation function with insertions of ! scaling fields O, of dimensions d,.
The connected n-p!lint function in real space then behaves like:

After Fourier transformation, fo.ctorization of tbe 6-function of momentum coll5ervation

and Legendre transformatiau with respect to the field ljJ we find (neglecting possible
regular terms):
....
r~I ,o,(.l.q;,.\p;) tx >.-o·,
with D' = D (n + l- l)d -n('.M~ d) d- nd~ +I:, d;- d. lt follows in particular
that d~· = d- 1/v.
It remains to study the behaviour of rorrelation functions in the critica.! doma.in (25.36)
a.way from T, and this will be the subject o{ Chapter 26.
Remarlc. The physics we intend to descrlbe corresponds to integer valuES of e:, E 1, 2.
Althougb we can only prove tbe w.lidity of all RG results within tbe &amework of the E·
expansion, mo shall eventually assume that their va.lidity extends beyond an in&nitesima.l
neighbourhood oi dimeosion 4. The cnmparison of the results obtalned from the summed
expansioll with expeYimenta.l or numerical data thus provides a. crucial test of the theory
(see Cha.pter 28).
The ped point hamiltonian. Let us note that the fixed point hamiltonian 'H' has never
been explicitly constructed for a.ny non·gaussiM tbeory. lndica.tions from .field theory are
that such a construction is not easy. Since ftxed point correlation functions have been
ealculated, they define implidtly a cl11S5 of fixed point hamíltonians. It bD.S also been
propooed to define 'H' as a limit of a ""'luence of hamiltonians Ht whose correlation
functions W~"l converge towa.rds fixed point correlation functions wi"l in tbe following
way:

In Sectíons 2!>.3-25.5 we have constructed 'H 1•

Quantum field theory in partide physics. Let us now comment about the signilica.nce of
these remlts for field throry in the pa.rticle physic:s context. From the preceding a.nalysis
w• unden;tend that the ex.lstenoe oi a renormalízable .field theory, beyond perturbation
Lhoory, relie:s an two praperties:
(i) Mass renormalization wust have a sense beyom! perturbatian thoory (the exi8tenoe
of a critica.! temperature).
(ii) An IR ftxed point in the RG sell!le must exist, I.D eilSllre that the long dista.nce
is short-distance insensitive and can be described by a renormalizable quantum
theory.

-----------------------"'~·IA<WI'·t.'l>t.'L•'•'I~· l.'l,l1.'1"tli~
582 General Renunnalizatian Grnup Analyru 25.6

An inspection of the action (25.30) however shows that, in contrast with conventional
QIT, the q,< coupling constant in the critica! phenomena applications has a dependence
in h given a pnori. For d < 4 the 1/! 4 coupling is very large in terms of the scale relevant
for the crítica! domain. In the usual formulation of QFT inslead the !>are coupling
constant ís coruádered as an adjustable parameter: A renormàlizable QFT corresponds
to a hamiltonían maintained dose to the gaussí"" lixed poínt by adjusting the coeffidents
of ali relevant and marginal operators. The resulting theory is thus less generic.
These cocllicients then introduce a new scale in the theory (called p.- 1 in the text),
much larg...- than tbe mícroscopic scale, in such a way that lt is p06Sible to define, in
addition to tbe universal long distance phy,;ics, a. short dístance or hu·ge momentum
pbysics with p. « IP;I « h.
N ot.e however that íf we demand to particle phy,;ics tbeories to be naturol, í.e. that
conpling constants at the cut-off scale (which can be identified witb the scale of some
new pbyslcs) are numben of order 1, we Me back into the statistícal frameWDTk.
Not.e finally that the RG analysiE indícates that in four dimenJlions, in the domain of
attraction of the g O IR fur.ed point, the renorma.lized coupling constant of the rp4 field
theory goes w zero logarithmically when the cnt-off becomes infinite. Sínce no other
tixed point 5001Il8 to exist, this leads to the so-ca.lled triviality problem.

Bibliograpbical Notes

The modero formulation of the RG ideas is due to

K.G. Wilson, Phys. Rev. B4 (1971) 3174, 3184;
and presente<! in an expanded forro in
K.G. Wilson and J. Kogut, Phys. Rep. 12C (1974) 75,
The idea of RG transformations was earlier proposed in a simplilied forw in
L.P. Kadanolf, Physics 2 (1966) 263.
The systematic classilicatíon of operators can be found in
F.J. Wegner, Phys, Rev. Bfi (1972) 4529, BG (1972) 1891, and in Wegner's rontribution
to PhMe Tfunsitions o.nd Critical Pherwmena, vol. 6, C. Domb and M.S. Green eds.
( Acaderníc Press, London 1976).
The idea of tbe E-expansion ís due to
K.G. WW.on and M.E. Fisher, Phys. Rev. Lett. 28 (1972) 240,
After Wi!Bon's original a.rtides, seVeral authoJS have realized tho.t RG methods, as de-
veloped in ·Quantum Fíeld Tbeory, could be applied to Critica.l Phenomena
C. Di Castro, Le«. Nuooo Cimento. 5 {1972) 69; G. Mack, Kaisers!cmtem J97i., Lecture
Notes in Pbysics vol. 17, W. Ruhl and A Vancura eds. (Sprínger Verlag, Berlin 1972);
E. Brézin, J.C. Le Guillou 81ld J. Zinn-Justin, Phys. Rev. 08 (1973) 434, 2418; P.K.
Mitter, Phys. Rev. D7 (1973) 2927; G. Pari.si, Ca.f'f}~Se Lectun!s 1979, published in J.
Stat. Phys. 23 (1980) 49; B. Sch~oer, Ph118- Rev. B8 (1973) 4200; C • .Dl Castro, G.
Jona-Lasilúo and L. Peliti, Ann. Ph.ys. (NY) 87 (1974) 327; F. Jegerlehner and B.
Schroer, Acta Phys. Austr. Suppl. XI (1973) 389 (Springe!'-Ver)ag, Berlin).
The RG equa.tions for the bare thoory were first derived in
J. Zinn-JilStin, Caf'f}ese ~u1Es 1913, unpublished, incorporated in the contribution
of E. Brézin, J.C. Le Gulllou and J. Zinn-Justin to PhMe Tfunsitíon.~ and Criticai Ph<-
nomena vo). 6, quoted in Chapter 24.

•"""'_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _,~,u.,wl·!.hk't·,···l·~
25.6 Generol Renonnolization G>;<Jup Analysis 583

Exercises
Exen;ise 25.1
Study with the forma.lism of Chapter 25 the tricritica.l </J~ field theory near three dímen-
sions. We recai! tha.t the tricritical poínt is defined by:

Calculo.te in p!l.rt.icular the RG functions at lea.ding order. We wríte the interaction

9fl. 6 - 2 d4J6 /6! a.nd set
E 6- 2d. It is a.dvised to use dimensional regularization.
Extend this ana.lysis to the u4>2N /(2N)! ficld theory near the dimension in which it is
renortno.lrzable (the upper-crltical dimension). lt is suggested to setE= 2N -d(N -1).

Exerci se 25.2
Long mnge forces. From a mean field analysis one can show thnl. O(N) sytnmetric models
f
with two- body long range forces yield, near a continuous pbase tmnsition, a (,P 2 field
theory with however an unusua.l, non-loca.l, propagator. For a potential V decayillg a.t
large distance r as V(r) o: ,.-d-• the effectíve hamlltonian reads

with " < 2. Thls means that in momentum space the propa.gatOr is

25.2.1. Find the upper-critical dimension, the equivalent of dimension four for the short
range case.
25.2.2. What can be said about the field strength renonna.lization and the exponent f/,
and the limit " - 2?
25.2.3. Calculate the /3--function at one-loop order and the exponent v at lea.ding order
ín the ~-expB.DSion.

-----------------------"'~·IA<WI'•I..'i>t.'L•'•'I~· l.'l,l1.'1'tJ:i~
26 SCALING BEHAVIOUR IN THE CRITICAL OOMAIN

In Chapter 25 we have estab\illhed the sca.Ling behaviour of correlation functíons at crit-

icality, T = Te- We now study the criticai domain which is defined by tbe property that
the correlation lengtb ís large with respect to the microscopic scale, but finite.
Using results proven in Chapter 10, we first demoru;trate strong sca.ling above Te: in
the criticai domain above T., ali correlation functions, after resca.ling, can be expressed
in tenns of universal oorrelation functions, in wlllch tbe sca.le of distance is provided by
tbe correlatlon length.
However, because the correlation length is singular at T" thls formalism doeo not
allow to cross tbe critica! temperature and thus to describe the whole critícal domain.
Tberefore w.< theD expand correlation fuoctions in formal power series of the deviation
t "' (T- T.) (Te from the criticai temperature. Tbe coefficients are criticai correlation
functions involving 4>2 (x) insertions at zero momentum (,; one can infer from the analysis
of Chapter 10), whose sca.ling behaviour ba.s been derived ln Section 25.6. Summing the
expansion, we obtaio RG equations va.lid for T i< T•.
'Ib cross the critica! temperatura, avoiding the criticai siogularities, it is in addition
necessary te explidtly break the synunetry of the bamiltonian. We thus add a small
m&gnetic field to the spin interoctioru;. We then derive RG equations in a field, or at
fixed magnetization. In this way we are able to connect continuou•ly correlation functioru;
a.bove and below Te, and esta.blish sca.ling !aws in the whole criticai domain.
In the first. part of the chapter we restrict oursel""" to l.sing-like syStems, we then
generalize the results to N-component order panuneters in Section 26.6. In Sectioo 26.7
we show how to expa.nd the universal 2-point function when T approaches Te, usiog
the short distance expanslon. Flna.lly Úl the appendix we discuss the energy correlation
function when the sp..,ific heot exponeut a vanishes.
Remork. The temperature is coupled to the total hamíltonian or eonfiguration ehergy.
Therefore a variation of the temperature generates a variation of ali tenns contributing
to the effective hamiltonian. H~ver, as we have sbown in 25.2, the most relevant
contribution (the most IR singular) corresponds to tbe ,P 2 (x) operator. 'l're can therefore
ta.ke the difference t T - Te hetwl!en the coefficient of f/> 2 in (25.34) and its critica! value
as a. linear measure of the deviation from the critical temperature and pararoetrize the
effective hruniltonian a.s:

(26.1)

Dimensional analysis then yields the relation

(26.2)

With this parametrization the criticai domain corresponds to

-----------------------!1/~,VhLI·!.hk'l'''''i·~
26.1 Scaling Behallivur in t.he Crjtical Domain 585

26.1 Strong Scaling above T 0 : The Renonnalized Theory

We discuss in tlill- sectian '\'aJ'ious propettíes of the critical behaviour above Te· For
T > T0 , because the theory is ma.ssive, ít is possible to introduce a special nmormalized
theory with renormolization canditions imposed al zero rnomentum. Generolizing to
any dímension d ~ 4 the formalism of Chapter 1O, we define renormalizad correlation
fu nctions by:

rFl {p;m.,g,) m~ + l'2 +o (p•), (26.3)

r~4 ) (pt = O;mr,!Jr) = m:gT. (26.4)

This renormalhed theory is obtained from the original "bare" theory by taldng the large
cut-aff limit at g, and m, fixed:

(26.5}

The oondition (26.4) is written in such a way thal g, is a dimensionless parameter. A

simple dimensional analysis tben yields for tbe renormalized oorrnlation functions the
relatian
(26.6)
Therefore, the correlation length {, which characterizes the decay of the connected two-
point function, is proportíonal tom;-': it fixes the scale of the zero momentum renor-
m alized theorv.
Note thD.t ~uation {26.5) holds for any dimension d $ 4 and therefore some conse-
quences of this ,;quation are valid beyond the E-expansion. Thls is one ree.son why the
consideration of this particular renormalized theory is so useful.
As we have sh<.JWn in Chapter 10, the renormalized correlation functions satisfy CS
equations;

m~[2 'l(Ur}]rf'·"l (O;p;;m.,g,),

(26.7)
in which the renormalized ,P2 insertions are specified by the condition:

r~ 1 • 2 l (O; O, O; m., g,) = 1 . (26.8)

As usual we have gíven to different RG functions the sa.me name {p, '7} because they
play the •ame role in the equations. We examine some consequenoes of equa.tion (26. 7)
in Subsection 26.1.3.

26.1.1 Bare and renorma!ízed coupling ronstants

Let us first discuss the problem of tbe ba.re and renormalized coupling constants. Severa!
remarks are here in order:
(i) Since in dimension d < 4 the <1>4 field theory is super-renormalizable, aJI correlation
functions h a"" a large cut-olf límit after a símple ml\5S renormalli.a.tíon, Le. after 011€ ha.s
, ta.kEn the deviation from the critica! temperature as a pararoeter. The coupling constant
• ILild tbe field amplitude ren<:>rmolizations are finite and therefore are not required in
\. ~eneral. lf we only renormali.ze the m,.,.;, we obtain a finite tbrory which is a function of

-----------------------"'~·IA<WI'·t.'l>t.'L•'•'I~· l.'l,l1.'1'tr_~
5B6 Sroling Behaviour in the Criticai Domaín 26.1

the ba.re coupling constant u, the coefficient oi q,4 (x) ÍJl the hamiltonía.n. The coupling
consta.nt u, however, has a ms.ss dimension e. Therefore perturbation theory is really
an expa.nsion in powers of u/m~. It only make!l sense if this ra.tio ís kept fixed, which
irnplies lhat the coupllng constaot u (whích characteri""'" the deviation from the gaussian
fixed point) goes to zero with the inven;e correlation length as m~. This is indeed what is
implídtly s.ssumed í11 the conventíonallield theory framework. In the critica.! phenomena
situation instead, the coupling constant, which is rela.ted to microscopic pararneters of the
theory, is fixed. This rneans, as w.:> already stressed in Chapt.er 25, tha.t after introduction
by rescaling of the cUt-off A, g = ujJo.< remains fmite (see actíon (25.34)) when !o.__, oo.
Therefore, in the critica.) domain, i.e. in the la.rge cut-off limít, u becomes large for d < 4
and the IDil5S renormalized perturba.tion tbeory becomes useless. This is tbe reason
why it is necessary to introduce a. field amplitude renormalízation and 11. new expansion
param eter g, w hich, as we verlfy below, remains finíte in thís limit.
(ü) We note that by thís method (introduction of the renonnollzed theory) we ha.ve
taken the large cut-off liwit in the following way: m « u 11< « A instead of m « A ""
ui/•. We have to a.ssume that the result is the s11.01e at )eading arder.
The renonnalized caupling carn;U.nt. The function f3 (g,) in (26.7) is given by equation
(10.26):
fJ(g,) =m, _dd I
fflr A.Q
g,(g,m,fA). (26.9)

Let w; introdnce tbe varíable). = m.fA. Consídering g, as a. functlon of .\ at g jixed, we

can rewrite equatíon (26.9}:
d
À dÀ g,(À) =fJ (g,(.\)).
Thís equation, which is silnilar to equation (25.47a), is a. flow equa.tion for g,(.\). When
tbe correla.tion length increeses, the ratio m./
A decrea.ses, a.nd thw À goes to zero. The
renormalized coupllng constant wiU be drlven towards an IR stable zero of"/3 (g,) if such
" zero exists. Assuming the existem:« of such an IR fixed point:

witb w =(f (g;) >O,

m can integrate equation (26.10) in the neighbourbaod of g;, and estimate g,- g;:
lg, - 9;1- (m./A)"' ·
Tberefore, as a consequence of the spedal cooditiollS imposed by the atat.istical origin or
the problem, in the criticai domain the renormalized ooupling consto.nt is dose to the IR
!ixed point value.

:?.6.1.2 Strong scaling

At leading order we can then replace g,
by g;. Let us evaluate the behaviour of th<l
renol1lla.li2.a.tion consto.nt Z (equation (10.27)):

m.-f-1
m,. J/1
lnZ(g.,m,./11.) = IJ(g,).

Using the para.met.er À m,.f!o. a.nd tbe function g,(.l.) ofequation (26.10), wecan
equation (26.13) ""'
d
Àd); lnZ(À} = 17(g,(Ã)).

----------------------'~•1-"I.,WI·!.hk'l •'•'I·~
26.1 Scalíng BehatJiour in lhe Critic:al Domaín 587

In the case or an IR fixed point we find for À = m, I/!. •mall:

Z ~ (m,/1!.)", (26.15)

where we have ca.lled '7 the value (assumed finite) of the function IJ(.!I) at the fixed point:

(26.16}

Let us also assume that the renormalized correlation functions r}"l (p;;m,g;) are finite.
Combining then equations (26.5,26.6,26.15) we find in the critica! dome.in lhe scaling
relatíon:

r!nl (p;,t,g,A (26.17)

TJu, parameter h ha.s only been introduced to provide a scale for a.l1 qua.ntitics. Once
the form of tbe critica! behaviour has been obtained, we can rescale a.ll dimensional
parameters (here mornent.a, deviation from T<> and correlation length) to elirninate h, as
we bave done above (see equation (26.2)) ..
Equation (26.17) provides a proof of strong •c:aling a.nd universalíty in the whole
crltical dome.in above T,. The original correlation functions depend on tbe deta.iled form
of the interaction. The ~· theory bare correlation functions depend only on the way
tbe cut-off is introduced and exp!icitly on p;, !J, a.nd t. The r.h.s. of equation (26.17)
involves only tbe renormalized correlation functions at .!Ir "" g;
i.e. functions of ratios
p;jm,. Furthermore tbe result (26.17) holds for any fuced dímension d ~ 4. Resort tC>
tbe ê-expansion bas been avoided. lt has C>nly been necessary 10 BS5Ume the existence of
an IR fixed point. Of courre witbin tbe 1:-expansion, using tbe resolts of Cbapter 11, we
can ímmediately C>btain f3 (g,) at leading arder and calculate 11; and w:

9; = 16~2~ +o (~:2)' w ="'+o (€'). (26.18)

Hov.>e'lef, as first GtlggéSted by Pa.risi, it is also pos!!ible to ana.lyre numerica.lly tbe pertur·
bative expa.rudon in pawers of g, a.t fixed dimension 3 or 2. Such an a.nalysis convincíngly
demonstrates the existence of an IR fuced point and allows a precise deterrninatíC>n of
critica! expommts as will be •hown in Chapter 28 when we dísclll!S nurnerical results.
Note finally tbat tbe equations (10.2~10.31) ean be uzed to cbaracterire the divergence
of m;- 1 , and thus tbe correlation lengtb, as a function of tbe temperature or ba.re m,....
This beha.viour will be deríved in Sectkm 26.3 by a.notber, simpler, method. It confirms
that the relation between m, and t i!! singular. Therefore the methC>d of this sectíon,
bssed on the introduction of a zero-momentum renormali:.ed tbeory, does not a.llow to
cross the critical tempera.tnre. Indeed ali correlation functíons are parBllletri:.ed in terms
of the correlation length whicb is singular at T,. To a.void tbi. problem we íntroduce
in next ...::tíon a different forrna.lism whicb is a natural extension of tbe forma.lism of
Cba.pter 2~.
Let us conclude this discussion with a. few remarl<B.
(i) We ha.ve shown tbat the renorma.llzed couplíng coosta.nt g, has a finíte limit g, = g;
=
a.lthough the ba.re coupling u gh• becomes infinite. 1n addition precisely at 9; tbe fiald
IUIIpÜtude renormaliza.tion diverges as equation (26.15) shows. The conclusion is that
the IR fixed point lield theory beba""" even below four dimensiC>ns as a renormalimble
lield tbeory, a.nd a complete renormalizatíon is indead requíred.

---------------------_.,~,fA<WI'·t.'l>t.'L•'•'I~· 1.'1,'11. 'l'tr""""""""

588 Scaling Behaviour in the Criticai Damain 25.1

(ii) A second remark connected to the preYíous one, is tnat when u varies from >.era to
ínfinity, g, varies from zero to g;. This property seenlS to indicate that it is impossible
to oonstruct a theory with g, > g;. Since g; O (t) this argument suggestS that it is
somewha.t unlikely that a non-trivial rj! 4 field theory eXists in four dimensions. On the
other hand to constrm:t the renormalized tbeory we have taken the large cut-off limit at
u fu<ed. This procedure ís only legitímate íf the b~Ue renormalization group has only one
IR. fixed point. Otherwise other non-trivinl theories míght be obtained by sendíng the
cut.o!J and u to inlinity at g lixed. This poínt will be further discussed in Subsection
34.1.1.

!l6.1.S Large momentum be.luwwur

Let us retum to equation {26. 7) to investigate the behaviour of cor!'<'lation functíons at
large momenta, in the criticai dome.in m,. ~ !P.I ~ A. At leading order we Cflll replQCe
in equation (26.7) g, by íts IR. fixed point value 9;:

(26.19)

It follows from scnling relatíon (26.6) or (26.17) that the large momentum behaviour is
dírectly related to the a.pproacb to the crit\cal tbeory m, = O. lt ís thus not too surpri:sing
tho.t in order to extract some inforllUition from equation (26.19) it is necessary to retum
to the fr!IIIli<WOrk of the <-expa.nsion. A simple extensíon of Weinberg's theorem shOWI!
that, arder by arder ín g, and €, the r.h.s. of equation (26.19) is negligíble at large (non-
exceptional) momenta, or small masses, a.; one might naively guess from the f&ctor 'lll;
which has been factori?.ed. By contrast below 4 dímensions, at fixed dimenslon, tbere
always exlsts !UI arder at whích the r. h.s. ceases to be negligible. Our interpretation is that
this is a consequence of expanding perturbation theory around the wrong fixed poínt, the
IR. unstable gaussian lixed point. The true asymptotic behaviour of correlation functions
is different because it is governed by the non-trivi~I point g, g;.
Asymptotically the
rf"l therefO!'<' satisly:
(26.20)

Thgether '1\'ith relation (25.6) thls lea& to:

r!~) (Ãp. m) ~ ),d-(.,/2)(d-2Hlr(n) (p·· m,.) (26.21)

r s, r IPi.l>Tn.r r h •

Using then tbe relatíon (26.5) between bate a.nd renormnllred correlation functions, we
recover the scaiing behavjour of equation (25.61) derived in Sectíon 25.5.
lt is also possíble in this o.pproaclt to analyze the large momentum behavlour of the
r.b.s. ofequation (26.19), ~md therefore to calculate corrections to the leading behaviour,
using the short distance expansion introduced in Chapter l2 (see Section 26- 7).
Remar/;;. Note that if g, is fixed witb g, < 9;
then the large momentum behaviour of
eorrelation functions is given by the UV fixed point g, O, i.e. by perturbatíon theory.
Hawever, 8!6 we have already discussed, this oorresponds to a. situation wbere tbe bare
coupling coru;tant u goes to 'l.erO as m:,
which ís rather unnatural.
26.3 Scalíng B ehawour in the Critiml Dmnain 589

26.2 Expansion around lhe Criticai Theory

To be able to describe tbe wbole critica! domain we now use a different strategy. First
we ronsider the functional integral rcpresentation of n point correlation functíons corre-
sponding to the hRlniltonian (26.1 ). If we formally expand it in powers of t, we obtain
!
ao expansion in terms of criticai correlation functions with f ddx Çl 2 (x) (the most IR
singular part of the energy operator) insertions (see a.lso Section ]0.10). Consequently,
to be a.ble to define these correlation functíons by theír perturbative expansion, we ha.ve
to now return to the framework developed ín Sections 25.4, 25.6 and to the (-eJ<pansion.
However, even so, because the insertion of f d"x4>2 (:z:) is the insertion of the Fourier
transform of 4> 2 (:z:) at rero momentum, the corresponding correlation functions are still
IR divergent. We thus fin;t replace in the hamiltonian (26.1) the ronstant I by a field t(x).
We can then write the 1Pl rorrelatlon functions r<n), as functions of spa.ce variables, a.s;

r<nl (x;;t,g,A) (26.22)

As already noted in Section 10.10, by acting with tbe functional differential operator
f dyt(y)tl/ét(y) on eque.tion (26.22) we generote in the r.b.s. a factor l in front of r(l.nl.
lt ís then easy to verify that equatíon (25.66) implies (see equation (10.72)):

o. (26.23)

To calculate r(nl, it ís possible to perform a partia! summation of perturbation theory

in arder to introduce the mass renormalired free propagator (P' + t) -', for example by
going over to a loopwise expansion. One verilies that in this new perturbatlon theory the
oonstant t limit leads w a massíve theory and no IR divergences are generated anymore.
Equation (26.23) tben, in momentum representation, becomes:

(26.24)

Equation (26.24) is the formal analogue of equation (10. 74) and differs from equation
(26.7), whích also applies to the non-critica! (Le. massive) tbeory, by the property the.t
tt is homogeneous.

26.3 Scaling Laws above T 0

Let us first discuss aga.in the critica! bebaviour above T., which we ha"" a!ready exa.míned
In Section 26.1, with this new formalism. A1; we have done for prevíous RG equatioDS,
we integrate equation (26.24) using the method of characteristícs. ln additíon to the
Junctions g(>.) and Z(>.) af equatiom (25.49}, we now need a function t(J.) wbich we
determine by imposing the co.W.tency between equation (26.24) and:

{26.25)

----------------------~·IA<WI'•I..'i>t.'.J•'J<'I~·ni,J'I,l!.
590 Seáling Behaviour in lhe Gntieál Domain 26.3

Consist.ency implies tbe set of equations:

d
>.d>.g(À) {3(g(>.)), g(l) :g, (26.26)
d
Àd>.lnt(.>.) -'12(g(.>.)), t(l) = t, (26.27)
d
>. d). In Z().) = JJ(g(>.l), Z(l) 1. (26.28)

Dimensional 8Ilalysi5 yields:

r!n) (p,; t (>.) ,g(.>.), .\A) = (>.A)d-,(d-2)/2 r(n) ( :~; ~~i.' g(>. ), 1) . (26.29)

The criticai regíon is define<! in particular by Iti « 112 , and this is the source of the
IR singular behaviou:r which appears in perturbation theory. Let us assume that the
equation for>.:
(26.30)
hM a solution. Tben the theory at scale >. will no longer be critical. Comblning equatíons
(26.25-26.30) we find:

(26.31)

where we h ave introd uced the notation:

m"" >.A. (26.32)

Tbe solution of equation (26.27) C8ll be WTitten:

(26.33)

Substituting this relation into equation {26.30) we then obtain:

>. du 1
ln(t/A') = 1 I -;;-ll(g(u))" (26.34)

We look for a solutíon ,\ in the limlt t /A~ < 1. SinCll 11(g J is a positive function, at leest
for g small enough as can be v.!rl!ied in the e>:plicit expression (25.71), equatiou (26.34)
implles that tbe value of the pBl'a.meter >. is small, ond tbus that g(A) is dose to the lR
fixed point g'. In this limit 11 (g (u)) can be replaced at leading arder by tbe exponent 11
Md equation (26.34) can be rewrttten:

ln(t/11 2 ) ~ 1
In>.. (26.35)
v
Equation (26.28) then yields:
26.4 Scaling Behilmour in the Critie<U Domain &91

Takíng the large A, or the smo.ll À limit, and using equations (26.35,26.36) in equation
(26.31) we ftnally obta.in:

(26.37)

witb:
m(A l) C' ~t•. (26.38)
EquatioDB (26.37) omd (26.17) express the seune scaling property. Howf!Ver one new result
has been obta.ined. From equation (26.37) we infer tba.t the qua.ntity m is proportional
to the ph)"lical ma.ss or invel'l!e correlation length. Equation (26.38) then shows that the
divergence of the rotrelation length { = m- 1 at Te is characterized by the exponent v (a
result which ~ could aJso ba.ve derived with the fonnalism of Section 26.1).
For t i O, the oorrelation functions are finite a.t zero momentum a.nd behaw $S:

r(nl (O; t, g. A) "'tv(d-n(d-2+~)/2). {26.39)

In particular for n ""' 2 we obta.in the inv'erse magnetíc susceptibility:

(26.40)

Tbe exponent whlch characterizes tbe di.,....gence of X is usually called '1· Tbe equation
(26.39) establlshes the relation between exponents:

(26.41)

Finally let us note tlurl;, since we have assumed that the criticai theory exists when t goes
to zero, different powers of t ha"" to cancel in equation (26.37). From thls observatíon
we immediately """"""' equation (25.61) in the form:

(26.42)

26.4 Correlation Functlons with tf> 2 Insertions

A differentiation of the functional integral with respect to t(11:) I times before taking
the unifonn t Hmit, genera.tes cotrelation functions with I [t<!>~(x)) insertions (in the
statistical formulation insertions of the hamiltonian ot configuration energy density).
Dy differentiating equation (26.23) with respect to t (Yt) ... t (!li) before taklng the same
limit, we derive RG equations for the corresponding 1PI correlation fuoctions. It is easy
= =
to verify tha.t tbe resulting equation, except for I 2, n O, is:

a +f3(g)-
a
[A-
âA ag
(26.43)

"This equation can be solved e><actly in the same wny as equation (26.24) and leads, for
t « 1, jq;l «: 1, IP;I <t: 1, to the critica! behaviour:

rtl.n) (q;;pJ;t,g,A = 1) - mld-1/>-..{d-'+•l/>lpf·"l {q;/m;p1 fm), (26.44)

---------------------_.,~,fA<WI'·t.'l>t.'.J•'J<'I~·ni,J'I,l!.'l"t.ll""""""=
592 Srolíng Beha..Wur in the Critirol Domain 26.4

with:
(26.45)

The discussion then exact!y follows the !ines of Section 26.:>, a.nd we do not repeat it
=
here. Rather we concéntrate on the case n "" O, 1 2, which corresponds to the energy
densíty two-púint rorrelation function. Starting from equation (25.66), and using the
method explained above, we obtain:

(26.46)

The fuoction A-·c~ (g) of equation (25.72) is still a solution of the inhomogeneous equa-
tion. The equatlon (26.44) then is replaced hy:

r<2.o) (q;t,g,A = 1) ~~ m<ó-2/vlFl'·"l (q /m) + C2 (g). (26.47)

t<
1•1<1

At zero momentum r!•.o) (q = O} is u.lso the second derivative wíth respect to t of the free
energy (the connected \'acuum amplitude}, i.e. the speciftc heat. lts criticai behaviour is
thus:
(26.48)

The specífic heat exponent iB cailed a. We have therefore derived the scaling IBw:

(26.49)

1ntegrating r! 2 •0 l twice with respect to t, we obtB.in n-•r(t,g,A), the thermodynamic

potentia.l per unit volume (ca.lling n
the volume factor) for a vanishing magnetization
(and thus a.lso the free energy in zero roagnet.ic field):

(26.50)

for 11\ 4:: !1.2 .

The three first t.erms correspond to the beginning of the sma.ll t expansion of the regular
part of the free energy 1111d depend explicitly on g through three functíons Cu, C1 and
C2 while the fourtb terrn characterizes the leading bebaviour of the singulBr part of the
free energy and is uníven.al (it still depends on the normalization af the temperature but
this norma.lization can be ce.ncelled in appropriate ratios ).
Since we bave oompletely described the critica.! behaviour abow T,, we examine in tbe
next sectian the critica.l behBviaur in the ordered phase ( M i- O).

~· Jlt.J•t.l:'. '1'\r'"'=="_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _...~.u.,wt·!.hk' ~'i•'t·~ J

26.5 Scaling Behaviour in the Critico./ Domllin 593

26.5 Scaling Laws in a Magnetic Field and Below Te

In arder to pass contimwusly from the dioordered (T > Te) to the ordered phase (T < T,),
it is necessary to add to the hamiltonian an interaction whlch nplicitly breaks its sym-
metry. We thu5 consider correlation functions in the presence of an externai magnetic
field. Actually it is more con,..,nient to consider correlation functions at fixed magneti-
zetion. Therefore we fírst discuss the relatíon between field and magnetization, Le. the
equation of state.
As we have already shown in the mee.n field analysis, in th<J ordered phase some
qualita.th"' differences appear between systems which have a dlscrete and a continuous
symmetry. We shall illustrate these differenres ...•ith the example of the N -vector model
with O(N) symmetry and ( 4>~) interoction in Section 26.6.
2

26.5.1 The equ<>tion of state

Let us call M the avere.ge value of lfl(x) in a constant field H. The thermodynamíc
potential per unit volume, as a function of M, is by d<>finition:

(26.51)

The tnagnetic field H is given by:

(26.52)

Notíng that n M (â/fJM), we immediately derive from the RG equation (26.24}, the
RG equation satislied by H (M,t,g,A):

{26.53)

Th int<Jgrate equation (26.53) by the method of charocteristiCil we have to introduce, in

nddition to the functions g(À), t(>.) and Z(À} of equatio!IS (26.21>-26.28}, a new function
M(>.) which satislies:

M(l),M. (26.54)

Comparing equatio!IS (26.54) e.nd (26.28) we o;ee that M(À) is givom by:

M(>.) Mz-'0(>..).

The solution of equation (26.53) can then be written:

H (M,t,g,A): z- 112 (>.)H!M(>.),t(>.),g(>.) ,MJ. (26.55)

Dimensional analysis shows that:

H (M, t,g,A) = A"-•1• H (A,':!_1,, ;, ,g,l). (26.56)

----------------------,~·IA<WI'•t.'l>t.'.J•'J<'I~· JII,J'I,l!. 'l"tr!>.=

5!>4 Sc<iling Behauiour in the Criticai Domatn 26.5

Applying this relation to the r.h.s. of equat.ion (26.55) we obtain:

H (M,t,g,i\) (26.57)

We again use the a.rbitrariness of ). to m""" outside the criticai domain, in order to
remeM! the criticai singularities in the r.h.s. oi equation (26.57). Here a natural choice
is;
M(Ã} = (>.A) 1-'/ 2 , (26.58)
whkh implies, using the solution of equation (26.54), Í.hat:

M 1 f'da
In Al-•1~ = 2 Jt -;;
[d- 2 + 'l (g (cr)}). (26.59)

Remember tha.t the criticai domain is defined in particular by:

For g sma.ll, the expression d 2 + '1{9} is posítive because d :2;: 2 e.nd '1(9) > O, which
again implies that À should be close to. zero e.nd thus g(À) close to g". Jn this limit
equa.tion (26.59) leads to:
(26.60)
Now equation (26.27) implies;
(26.61)
and we have nlready seen that:
(26.62)
Repla.cing t(Ã} and Z(Ã) by tbeit a.ymptotic forrns (26.61,26.62) and using equa.tion
(26.60) to elimina.te >., we finnlly obtain:

(26.6g)

with:

(i 2" (d 2 + 'l), {26.64)

d+2-'l
(26.65)
d-2+')"

Equa.tion (26.63) establishes the scaling properties of the equation of state. Moreover
equa.tions (26.64,26.65) relate the traditional critical exponents which characteríze the
vanishing of the spont;an.,Qus ma.gnetization and the oingular relation between ma.gnetic
field a.nd rnagnetiza.tion at Te respectively to the exponents tJ and v introduced previously.
Valid for d < 4, these latter two relations seem to be inconslstent with the values of tbe
mean field exponent!l for d > 4. 'lb understa.nd thís poínt, it is necessary to remember
that for d > 4, g* vanish...,, and nll tenns in H, except the term linear in M, come from
corrections to expr.,.ion (26.63) (see Section 27.1).

~· Jlt.J•t.l:'. '1'\r"-'=-------------------....,,~,u.,wt.!.hk' ~'i•'t·~ J

26.5 Srolíng Behaviour in th.e Critirol Domain 595

Prupertie.s of the universal function f(x).

{í) Griffith's analyticity: equation (26.52) shows that H has a regular exparu;ion in
<>dd pnwters of M for t > O. This irnplies that when the varía.ble :r: becomes la.rge and
positive, J(x) hM the expansion:

""
J(x) = L"P"-r-"'»~. (26.66)
,.,.o
(ii) When M is different from zero, the theory remains ma.ssive when t vanishes. 1n
1
the loop expansion the corresponding ma.ss.ive propagator hM the form (~ + aM2) -
in momentum space. lt followo that we can expand [(M,t) and therefore H (M,t) in
power. of t wíthout rneeting IR divergences. Consequently f(x) is inlinitely differentiable
at x "'O.
(ili) Tbe appearance of a spontaneous magnetization, below Te, implíes that the func-
tion /(x) has a negative zero x0 :
/(:to)"'O, zo<O. (26.67)
Then equatíon (26.63) leads to the relatíqn:
M=\xorP(-t) 11 for H""O, t<O. (26.68)
Equation (26.68) gives the bebaviour of tbe spontaneous magnetízation when the tem-
perature approoches the critica! temperature from below.

26. 5.2 Com:lo.tion functiona m a field

We now examine the behaviour of eorrela.tion funotions in n 6eld. AU expressions will
8g!Úll be wriU.en for lsing-like systems. The generalization to tire N -vecu:>r model witb
O(N) symmetry is slmple and will be briefty di.scusse<l in Section 26.6.
The corre!ation functíons at fixed magnetization M ""' obtained by expandíng the
generating functional r (M (:r:)} of lPl correla.tion functions, around M (x) = M:

rt">(:z: 1, .•. ,:~:,.;t, 9 ,AJ=liM (:r:1 )ó"liM( {(M(:r:),t,g,A)I (26.69)

··· Xn M(~}-M
The expansion in powers of M of tbe r.h.s. of the equatíon then yields:
""M•
r!n) (p,,. ·- ,p,;t,M,9,A) =L -.rrn+•) (p,, ... ,p,.,O, ... ,O;t,O,g,A).
.1:1!"0 8.
(26.70)

From the RG equations satisfied by tbe correla.tíon functíons in zero magnetization (equa-
tions (26.24)) it is then eaey to derive:

[A!.+ {J (g) ~- ~'1(9) ( n + M 8~)- '12 (g) t~] rfn) (p;;t, M,g,A) =O. (26.71)

< Thls equatíon can be solved by exactly the same metbod a.s equation (26.53):
r!nl (p;;t,M,g,A 1) ~ mld-(d-2+")/>Jp(nJ (p;Jm,tm- 1 1~), (26.72)

for \p;l « 1, ltl «: 1, M «: 1 and with the definition:

m=Mvfil. (26.73)
r.h.s. of equation (26.72) now dependo on two dil!erent mass seales: m = Muffl and

---------------------~·IA<WI'•I..'i>t.'.J•'J<'I ~· JII,J'I,l!. 'l"tr'"""='=

596 Scaling Behaviaur in the Criticai Domain 26.6

2.6.5.9 CorrelatWn junctionJ below Te

We have argued above that correlation functions are regular functions of t for small t
provided M do"" not vanish. It is therefore possible to cr""' the criticai point and t~
then take the zero externai magnetic field limit. In the límit M becom...,; the spontaneous
mngnetization whicb is given, as a function of t, by equation (26.68). After eliminatíon
of M in [avour of t in equation (26.72), one finds the criticai behaviour below Te:

(26.74)

wjth:
t <o. (26.75)

We conclude that the correlation functíons have exactly the sarne scaling behaviour abow
and below T,. ln particular since traditionally one defines below Te:

(26.76)

we have thus established v''"' v and-( -y,

The universal functioru; F~nJ and F~n), instead, are of oourse different.
The extensian of these considerations to the functions ...~th ,p2 in.....,rtions, r< 1-"l is
strllightforward. In particular the soune metbad yields the beha.viaur of the specific heat
belowT.:

which similarly proves that the exponent above and below T. are tbe same, o' "" a.
Note that the oonsta.nt term A-•cz (g) whkh depends explicitly on g is the same abow
and below T0 , in oontrast with the ro<:fficlent of the singulm par!.
The deriVBtion of the equality of exponents above and below T., relies on the exlstençe
of a path which avoids the criticai point, along whicb the correlation functions are regular,
and the RG equations everywhere satisfied.
Remar/c. The universal functions characterizing tbe behavíour of correlation functione
in tbe critica.! doma.ln still depend on the nonnallza.tion of the physlcal. parameters t, H,
M, distancel or momentB- Physica.l quo.ntíties which are independent of tbese norma.li:.._
tions are universal pure numbers. Examples are provided by the ratíos of the ani•plitu<lee.
of the singularitíes above and below T. like A+ JA- for the specilic hea.t.

26.6 Tbe N-Vector.Mode!

We now gener~ previous results to models ín which tbe order parameter is a

and which have symmetries sucb tbat the La.ndau-Ginzburg-Wilson hamiltonia.n
the form of a ~' lield theory. We first oonsider a simple but importa.nt exàlDple: a
in which the harniltonian ha.s a.n O(N) symmetry, and then discuss the general case.

~· Jll.l't.l:'. ' 1 ' \ R ' -_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ '~•U.,WI·!.hk' ~'i•'t·~ J

 Scaling Behavíour in the Criticol Domain 597
26.6

!)6.6.1 The O(N) symmetric N-l!fCtur model

The O(N) eymmetric effecti"" hawiltonían has the form:

1-1. (<f>) j {! [â~<f>(x)J2 +~(r,+ t) .p2(,.-) +i'; oA' (<t>2(:r)) 2} dd:r, (26.78)

in which 1> is a N-vector.

The RG equations have exactly the same form as the lsing-like case N = L The
eoupling constant RG function bas the expansion:

N+82 (32) (26.79)

i'J{g) -Eg+ 4S1r2 9 +0 9 ,g e.

At leading order in e, {3 (g) has a zero g• which is an IR fuced poínt:

(26.80)

(2'6.81)

and therefore ali lhe scaling laws derived above and in Chapter 25 can also be proven
for tbe N-vecti:Jr model witb O(N} symmetry. We give the expr<l$Sions of the other RG
!unctions at leading order in Subsection 26.6.2 by specializing the expressions obtained
for the general N-vector model.
COfTl!lation functions in a field ar óelow Te. Tbe addition of a magnetic field term
in a O(N) eymmetric hamiltonio.n h"" various effects. First, the mtlglletization and
the magnetic lield are now vectors. Tbe scaling forms derived previously apply 1I:J the
moduliiS of tbese vectorll. Second, the continuous O(N) symmetry of the bamilwnian
ls broken linee.rly in the dynamical variables by the addition of a magnetic lield (see
Cbapter t:l}.
Since the lield a.nd the magnetization distinguish a directíon in vector space, there now
exist 2" n-point funct.ions, each spin being ei~her along the magnetization or transverse
to ít. As we have sbown in Chapter 13, these dilterent correlation functions are related
by a set of ídentities, called WT identities, whicb we have discussed there in a. general
fra.mework. Let us here recall in the C&!iC of the O(N) symmetry the simplest one,
involving the 2-point function at zero momentum, also directly establiBhed in Section
24.4 (equations (24.57,24.59)}. 1n teriiiS of rT
the inverse 2-point function transversa to
M at zero momentum or the transverse susceptibility, it reads:

(26.82)

v.é recogni.2e, in dilferent notetions, equation (13.44). As we have already discussed in

Section 13.4, it follows from this equation that if H goes 1I:J zero below Te, H f M and
· therefore r T at zero momentum vanish. Thls last result implies the existence of N - 1
Goldstone roodes eorresponding to the spontaneous breaking of the O(N) symmetry into
-1).
finally that the inverse longitudinal 2-point function rL(p) is singular at zero
lnllm•entum in zero field beJow T, as one can infer from its Feynman graph expansion
). Thls IR singularity ís not generated by crítica!. f!uctuations but ~y tbe
[G<>lde<toile modes. lt ís characteristic of cootinuous symmetries. lt implies that the

---------------------~·IA<WI'·t.'l>t.'.J•'J<'I~· JII,J'I,l!. 'l"tli~

598 Scaling Behat!Wur in the Criticai Domain 26.6

Fig. 211.1
'--0---' T
Ont>-loop Coldstone mode contribution to r L·

longitudinal 2-point correlation function does not decrease exponentially in spoce at

large distanre. We shall di.scllSS thi.s problem in detall in Chapter 30.
Remarlt. Previouo results alro apply to a model in which the hamiltonian has a sym-
metry smaller than the group O(N) but is still sucb that the elfective hamiltonia.n has
tbe forro (26.78) because the qua.dratic and quartic group ínvarlants ál'e uníquely deter-
mined. Tben tbe O(N} symmetry i.s dynamica!ly generated in the criticai domain. Only
a dose examination of the leading correctiçms to the criticai behaviour reveals tbe di!Jer-
ence. We have aJready encountered a. similar phenomenon: the hypercubic symmetry of
tbe lattice has led to a O( d) continuum space symmetry in the criticai domain.

26. 6.1! Til{; genernl N -vediJr mede/

There exm int.etesting physical syotems in which tbe hamiltonian is not O(N) invariant.
A fust category cansists in systems in which there are ..::veral correlatíon lengths. ln such
situations generical!y, when the temperature varieo;, only one wrrelatian length becomes
inlinite a.t a time. Then the companents of the dynamíc varia.bles which are non-crltical
do not contribute to the IR singularities. They can be iJJtegrated out in much tbe some
way as tbe auxiliary lields in Pauli-ViUa.rs's n:gularization scheme of Section 9.1. The
effect is to renorrnali2e the efJective local harniltonian for the critica.! componen~. This
remark is related to the decoupling theorem of Particle Physies. We can therefore restrict
ourselves to theories with only one correlation length. This second category consists in
systems ín whích tbe hamiltonian is invBria.nt under a symmetry group, subgroup of
O(N}, whicb admits e unique qua.dratic invariant 'í..<i>;t/>; but severa.! quartic invaria.nts.
The general hamiltcmian tben h!IS the forro:

(26.83)

Dueto the symmetry, the tensor !,lijkl has special properties wbich imply that the 2-point
correlation function r~;) is always proportional to the unit metrix:

(26.84)

This equation implies successive conditions on the tensor 9;jkl in perturbation theory.
ln Section 11.7 we ha"e discussed the reuorroalization and RG equations for a general
~4 field theory and we can apply the formalism here.
Rerwnnalization group equatioru. Let us first sketch tbe derivation of the RG equations
for a multi-oomponent criticai theory. Sinee the field amplitude renonnalization constant
is independent of the components, the relation between bare a.nd renorroali?.ed con:elation
functlons now takes the forro:
(2&.85)
26.6 Scolíng Behaviottr in lhe Criticai Domaín 599

in which g stands for 9ijkl and g, for g,,i,k!·

I! we differentiate the equation with respect to A at 9ro m, and I" fixed, we ohtain the
RG equation:

(26.86)

with the definitíon:

(26.87)

These equations Cl!ll be integrated by the same method as before. We have to introduoe
a scale dependent coupling constant !l<jk!(.:X) obeying the fiow equation:

(26.88)

The long distfUlce properties of such theories are then governed by fixed points oolution
of the equation: ·
A;•!(g") = o. (26.89)
The local stability properties of fixed points are governed by the eigenval~ ofthe matrix

(26.90)

If the real pnrts of ali eigenvalues are positive the lixed point is locally stable. Tbe global
properties depend on the complete solutions of eque.tíon (26.88) whích determine tbe
ba.sin of attraction in coupling space of each lR stable lixed poínt. We sha.JI not díscuss
further this problem here IUid refer to tbe litere.ture for detail>;.
Tluõ tb;(x),Pj(x) inserlions. lt is a.lso useful to consider correlatien functions with
Í</>;(x)tbi(x) insertions. Tbeir renormalization involves a multiplicatíon of each insertion
by tbe matrix çJI.~ 1 • This leads to tbe RG equation:

n
2" (g
) (
')2 !J
)t!!..)r(l.,)
éJt
< <

Jtkt•u]tttol><«i•
I
- 2: J);~.m.b..C.. r~·;;<«b-""' ... j,., .•,..... o. (26.91)
-ml

with the definítion:

(26.92)

Since the insertions of Etb1</J;, which are generated by a differenti•tion witb respect to t,
o.re multiplicative!y renormalízed, tbe matrix 'IÍ:J,
(g) bas lik1 ns eigenvector and:

(26.93)

----------------------~·IA<WI'·t.'l>t.'.J•'J<'I ~· JII,J'I,l!. 'l"t.ll"""""=

600 Sroling Behaviour in lhe Critirol Domaín

RG functiorw. At one-loop arder the RG functions are given h)~

/Jijkl = -EYijkl + 1: ,2 (g,,~n9mnk! + 2 terms) +O (g 3) , (26.94)

1 I ( ')
'l (g) Óij = 24 (S1r')' 9iklm9jk!m +O g , (26.95)

'1;;;., - 1!. 2 Yijkl + 0 (g 2 ) · {26.96)

O:mdition (26.93) then irnplies:

(26.97)

Stability of tM O (N) symmetric f""ed PQint. Among the p08Sible fi:xed points, one
a.Iways finds, in additi<m to the trívia.l gs.ussi811 lixed point, the O ( N) symmetric lixed
point. We can here study íts loca.l stabílity a.t leading order in E. l..€t us first specia.lize
expressions (26.94-26.96) to the case of the {.P2 )z field theory witb O (N) symmetry. We
then ha.ve to substituta:

(26.98)

A!ter a short ca.Iculation the expressíon (26. 79) of the iJ-function ís recovered and in
addition:
2 2 O(-')
( -)- (N + 1-9+
119-72 g, {26.99)

where the nota.tion [11.84) has been used:

As noted in Section 11.7, introdueíng the identity ma.trix I 81ld the projector P:

'l(>l - (NP + 21) ~ +O (ij') . (26.102)

The trace of the ma.trix 'l(>) yields '12 (g). The second eigenYBlue of the mstrix '72 (9),
given by its traceless part, rorresponds to a syrnmetry bre&king mass term, a.nd as we have
discussed at the beginning of this section, deseribes th~ cr056Q\'er from a situatio11 with
one oorrelatíon length to a sltuation in which some componenta of the order parameter
decouple. It ís traditional to introduce a new function 'I' (g) a.nd to para.metri.ze it as:

I ~(g)
T/2 (!I)"" v(g) -2. (26.103)

The fi:xed point YB.lue 'I' 'I' (9") ís ca!led the eros:sover exponent. Fina.Ily the stability
ronditions are given by the eigenva.lu"" of the matrix M. Setting:
(26.104)

~· Jlt.J•t.l:'. '1'\li•"="""'_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _,~,u.,wt·!.hk' ~'i•'t·~ J

26.6 Scaling Behaviour in the Çritical Domaín 601

we find at leading order:

(26.105)

Taking •;;ti proportíonal to Bi;kl we recover the exponent w. More genera.lly the eigen-
vectors can be classified a.ccording to their trace properties. L..t us v.•rít€ so;•t'

(26.106)

in whkh the tensors v;; a.nd UJ;;kl are tracei..,:

tvt;u:=O. (26.107)

The three eigenvalues correspondiug to u, w, u are respectívely:

w (
<+0<
2) •"'•nu.=€N+B:o
4- N ( 2)
é, w
I (26.108)

Note that the perturbation proportionel to tlij does not satisfythe trace condition (26.97).
lt therefore lifts the deg<:Deracy between the wrrelatioo lengtlu; of tbe different wm-
ponents of the order para.meter. It iuduces a crOS5QVet to a situatioo in which some
components decouple. H"""""r one easily verifies that the correspondiog eigenvalue w'
leads to l'lfect.s subleadi.og for < small wíth respect to the eigenvalue IJ~ (g•). Withln the
class of interactions satisfyiug equstioo (26.97), the releva.nt eigenvalue is "'"""'' We lind
the very interesting result that the O(N) symmetric fixed point is stable oga.inst a.ny
pertmbation for N sma.ller tha.n some value Nc- This is ogain a.n example of dynamically
índuced symmetry: Cotrela.tion functions in the criticai doma.in have a larger symmetry
tha.n microsoopic carrelatiou fuuctions. The calculation of "'•nio. st order E yields:

(26.10S)

Remarks. Jt can be seen in expression (26.94) ths.t the 13-function defines a gradient
fiow. Actua.lly ít ha.s been checked up to throo-loop arder that thll; property remains true
in the following sense: Calling g,. the set of ali coupli.og constants, one finds:

in which T.,ll is a symmetríc e.nd, for 9a =all, pooitive matrix. This is the best one can
expect, consideríng the trensfonnatiOJJ properties of the fj-functíon uudu reparametriza-
tion of the coupli.og consta.nt space. lt follows tha.t tbe fixed poínts are extrema of the
function B and that the mattix of derivatives of the fj-function at a fixed point, being
equivalent to a symmetrie matrb., hes real e\genvalues.
An empirícal observs.tion, in many examples, is tha.t when several fixed points are
found, then the stablest one corresponds, for g., small, to the largest value of the expohent
of 'l·

----------------------~·IA<WI'·t.'l>t.'.J•'J<'I ~· JII,J'I,l!. 'l"tli~

602 Sroling Behaviour in the Criticai Domain 26.7

26.7 ABymptotic Expansion of the 2-Point Function in the Criticai Domain

In the critic<LI domain, when points a.re separated by dístances much sm<Liler than the
correia.tion length ~. the correlation functions tend towards tiie correlation functiom of
the critic<LI theory (T =Te}
(26.110)

The r.h.s. is actunlly the ftrst term of M asymptotic expansion in the Vlll"iable p{ for
p{ large. Tbe leading term has OOell obtained by using tbe property that at large non-
exceptional momenta the derivative ar!n) (pt, •.. , Pn) / &t is asymptotically negligible
with respect to r<nl (p1, ..• ,p.,). However since:

the derivative ar!nl f&t eannot be evalua.ted with the sarne method because Lhe momenta
""' exception<LI. As we ha~ explained in Che.pter 12, it is necesse.ry to use tbe short
dístance exp8DBiOn (SDE) of operator produc:ts. Let us now concentrate on the 2-point
function. We ba'le to evaluate r!l, 2) (O;p, -p). However we ClmDot apply dírec:tly the
SDE to r!t,2} ~ause thís would involve rl•,o) whlcb needs additional renorma.lio:atíons.
We therefore difFerentia.te once more witb respect to !:

..i:....r<2l (p} =ri'·'' (o o·p -p). (26.111)

(&t)' • ' '

Th r''·'' we can DOW a.pply the SDE:

r!•.•l (O,O;p, -p) ~ B(p)r!S,O) (0,0,0) for C 1 «: p « 11. (26.112)

As shown in Section 12.5, B(p) satisfies a RG equation whích ca.n be obta.ined by a.pplying
the dUferential operator D:

(26.113)

em both sides of equation (26.112) nnd using the RG equations (26.43).

One finds:
[D +v- 1 (g) -2-'1](9)] B(p) ~o. (26.114)
Thkíng the t = O limit nnd setting g = g•, the equa.tion becomes:

[j 1
+v (26.115)
( h8A

Equa.tion (26.112) shows also that B (p) bBS caoonic<LI dimension ~. It follows:

B(p) ~ A• (p/A) 2 -~-(l-<>)/v. (26.116)

DifFerentiating equatíon (25.48} with reapect to t, we find:

(26.117)

---------------------'~·L"i·•'-'l·!.hk' ~,1•'1'~ j
26.7 Scaling Beha'!Jiour in Ih~. Criticai Domain 603

Integrating equation (26.111) twice with respect to t and llsing the set of equations
(26.112,26.116,26.117} we obtain finally:

r<2) (p) "" r?-" (a+ btp-1/v + d'-op-(1-a)/v + .. ·) . (26.118)

(-'<p<A

A systematic use of the SDE beyond leading arder allows us to calculate systematic
corrections to expression (26.118).
Note that the effect of the dífferentíation with respect to t has been símply to generate
the regular terms in the temperature which are of order comparable to the singular term;s
for e small.
Nezt to lmding tenn.s ín "field or below T~. lt is also possible to obtain expressíons in
a field o r b..low T. by expa.ndíng com!latlon functions in pow>~rs of the magnetíza.tion a.nd
applying the SDE expe.nsion to each term. The results now a.re dilferent for Ising-like
systetnS and tbe N-vector model. For Ising-like systems one finds:

r
(2)
(p, t, M) =V_ " [ a+ bpl/t .. + F,. ( Mil~
t ) (
M~l~
p )-(1-<>)Jv ]
+ . .. , (26.119)

in whlch the function F 1 can be ralated to the free energy and tbus the equation of state
by:
F1 (x) = j"" dss
1
1-l/l! [1' (O)-/'(-=--)]+
/!I/~
.,-----7:/'-+{0='-)----:c
(1//3) + 1'
Ó
(26.120)

.
In the O(N) symmetric case tbe SDE involves a second operator of dimeru;ion 2:

~·;
0;1 [<i>(z)] = ,P;(x)<Mx) - j, </> (z). (26.121)

Tbis operator is multiplicatively renormalízable. Correlation functions with the insertion

of 0,1 satisfy the RG equations:

~lj(g) ( <P(g)
v(g)
2)] rg:~ (p,;g, AJ =o, (26.122)

in which the RG function 10(9) has been defined in equation (26.103). Consequently a
new term is presen~ at the sa.me order in the asymptotic expansion of the 2-point function
whích becomes:

r<2l {!:l,t,M) p {[a+ bp:l~ +F, (M;I~) (J:18 t•-~ll~] 6;J

2-•

+
F (-t ) (-P )-d+op/• (6;; _M;Mi)} + ... .
2 Mlf/3 M•lfJ N M2 (26.123)

in whích <P is the crossover exponent and F2 a new universal function wlúch may be
calculated in an ~-expansion. At order e it is given by:

F1(x) = 1 + (Ne+ S) [(x + 3) ln(:r + 3)- (r+ 1) In( :r+ l)l +O (e2 ). (26.124)
2

---------------------·~·IA<WI'•t.'l>t.'.J•'J<'I~·ni,J'I,l!.
604 Scaling Beh.aviour '" the Criticai Domain 26.7

BibliogTaphical Notes
Most references glven at t.he end of Chapter 25 ate also relevant here and will not be
repea.ted. The cbapter is inspired by Sections VI, VII, XI a.nd XII of the review article
E. Brézin, J.C. Le Guíllou a.nd J. Zinn-Justin, in Phase Jronritions and Gritiml Phe-
i
nomena, vol. 6, already quotcd in Chapter 24,
whicb also cootaJns a more detailed discussion of the multH:omponent <1>4 theory. The
scaling fonn of the equation of state has been propooed in.
B. Widom, J. Chem. Phys. 43 (1965) 3892, 3898; C. Domb and D.L. Hunter, Pro<.
Phys. Soe. {London)86 (1965) 1147,
and vedficd in R.G calculations by
E. Brézin, D.J. Wallace and K.G. Wilson, Phys. Reu. Lett. 29 (1972) 591, Phys. Rev.
B7 (1973) 232; G.M. Avdeeva and A.A. Migdal, JETP Lett. 16 (1972) 178; see also
the contribution of D.J. Wa.llace in Phase Jhmsition.! and Critica.l Phenomena vol. 6
already quotcd.
A!so relevant is
R.B. Griftiths, Phys. Rev. 158 (1967) 176.
Tbe N-vector model with cubic anisotropy has been studied in
D.J. Wallace, J. Phys.C: Solid State Ph.ys. 6 (1973)1390; A. Aharony, Phys. Rev. BS
(1973) 334.2, 3349, 3358, 3363, 4270, P~. Re!!. Letl. 31 (1973) 1494.
See also
I.J. Ketley and D.J. Wa.llace, J. Phys. A: Math. NucL Gen. 6 (1973) 1667,
and the contribution of A. Aharony to Phase 1hlnsilioru and Critíca.l Phenomena vai. 6
already quoted.
The general N -vector model is discussed in
E. Brézín, J.C. Le Guíllou and J. Zinn-Justin, Phys. Rev. BlO {1974) 892.
The decoupling of non-critical or massive modes is discussed in
K. Symwuik, Acta Phys. Austr. Suppl. XI (1973) 199; T. Applequist and J. Carll.>lzone,
Phys. Rev. Dll (1975) 2856.
For the applicat.ion of the SDE to critical pbenoroena see
E. Bréún, D. Amit and J. Zinn-Justin, Phys. Reli. Lett. 32 (1974) 151; E. Brézin, J.C.
Le GuiUou and J. Zinn-Justin, Phys. Rctl. Lett. 32 {1974) 473.
Some resulto had been anticipated on pbysical gt<:Junds by
M.E. Fisher and J.S. Langer, Phys. Reli. Lett. 20 (1968) 665; M.E. Fishei and A.
Abarony, Phys. RetJ. BlO (1974) 2818.
The cro~r exponent bas been diseussed in
E. Riedel and F. Wegner, Z. Phys. 225 (1969) 195; M.E. Fisher and P. Pfeuty, Phyo.
Rev. B6 (1972) 1889; KG. Wilson, Phy._ Reli. Lett. 28 (1972) 548; Y. Yamazaki and
M. Suzuki, Prog. Theor. Phys. 50 (1973) 1097; Y. Yamazaki, Phys. Lett. 49A (1974)
215.
The effect of long range !Orces have been considered in
M.E. Físher, S.K. Ma and B.G. Nickel, Phys. Rev. Lett. 2!1 {1972) 917.

&ercises
E:z:ercise 26.1
Discuss • general situation in whicb the RG ,.funct.ion has the form of a gradient flow
ôV
ôg;.

'=---------------------~·U.,WI·!.hk' ~'1•'1·~ J
26.7 Scaling Behavíour in Lhe Critiooi Domain 505

Exercise 26.2
Renormaluation Group for models with cubíc •ymmet'1f. Consider a rnodel in which the
order parameter is a N-component field tf,(x) whlch bas a cubic symmetry:

q,, .... q,,, t~>i--t/1 •.

The Landau-Gin•burg action S( tf,) expanded up to fourth arder in </> then takes tbe
form:

The term wbkh breaks the O{N) symmetry of the actjon rnay be generated in a ferro-
magnetic rnodel by the interaction between tbe spin degTees o( freedom with the lattice
wbich has only cubic spatial symmetry.
26.2.1. Write the conditions relating g and). whlch guaro.nty the stability of the classical
potential.
26.2.2. Cal cu late tbe RG functions at leading order.
26.2.3. Fínd the VBrious IR fixed points and discuss their stability.
26.2.4. Relate the eigenvalues of the stability matríx at the lsing model fixed point to
Ising model exponents,

~----------------------.,,~,fA<WI'•t.'l>t.'.J•'J<'I~· JI ,J'I.l!.
606 Scaling Behaviour in the Criticai Doma.in A26.0

APPENDIX 2ti
THE SCALING BEHAVIOUR OF THE SPECIFIC HEAT FOR a= O

In the case of the 2D lsíng model the ""ponent <> vanishes. In such a situation it is

l
impossíble to fmd a function Co (g)A-• solution of equatíon (25.72) regular at g = g".
Indeed the equation for C, (g) ;,; then:

/1 (g) C2 (g) + (d- 2/v (g)) C2 (g} = B (g), (A26.I)

with:
a =0 "" v(g*) = 2/d. (A26.2)
Tbus at g = f( the Lh.s. of equation ( A26.1) vanishes if C, (g) ís regular, and the r .h.s.
does not. The general solution of equation (A26.1) has instead the form:

C2(g) = D(g)ln lg- g'l + E(g), (A26.3)

in whích D (g) a.nd E (g) are regular at g• Md

D (g") = B (g') /w. (A26.4)

As a COD5e<jUence, the functíon (q;t,g,J\)- A-•c, (g}, whích sa.tisfies a homoge-

fÍ 2 ,0)
neous RG equation, is singular at g•. When g().) approaches g•, the theory at sca!e À is
dominated hy this singularity. Sínce:

(A26.5)

equations (26.47,26.48) are replaced hy:

r( 2·0 l(q;t,g,A)-A-'C2 (g) ~A-• [F(~) -A2(g)ln ~2 ], {A26.6}

r<•.oJ (O;t,g,A)- A-'Co(g) ~ A-• [F{O} A2(g)ln ~2 ], (A26.7).

with:

A lternotit:e derivation. Alternatívely we can 50!"" directly RG equation (26.46) by the

method of chara.cteristics. We set:

(A26.8}

Applying the chain rule to evaluate explicítly the derivative ond expressing the com-
patibHity with equation (26.46), we find in a.ddítion to equations (26.21-26.28), two
equations:

(A26.9)

(A26.10)

---------------------·~·L"i·•'-'l·!.hk' ~,1•'1'~ j
A26.0 ScaliTI!I Behaviour in the .Criticai Domain 607

The solution of the first equation can be written:

(A26.ll)

lt follows ín partkular:
t(J..) t/((À). (A26.12)
Tbe solutíon of equation (A26.10) ca.n then be written:

C(>.)
1~
1
do B (g(rr))
a
(~)' (((>.)
a {(a)
)z (A26.13)

Using dimensional analysis and imposing the condition (26,30), t(>.) >.2 A 2 , we obtain
first:

r<>.oJ (q;t,g,A) = ç'(>.)(>.Ar' [r(z,a) (q/m; l,y{>.), 1)- C(>.J] , (A26.14)

with (equatíon (26.32)):

m ÀA.
Equation (26.3!'>) gives the asymptotic relatíon between >. and t, For >. small ((À) has
a power behaviour which in tenns of the exponent 11, related to 112 by equation (25.69),
has the form:
(A26.15)
In the absence of the inhomogeneous term we would recover the scaling form (26.44)
applied to i 2, n O, Let us then examine the behaviour of the ínhomogeneous term
for À small:

{A26.16}

For a > O the integral in the r.h.s, hM a límit when >. goes to zero a.nd be<x>mes a
function of g in agreement with equation (26.4BJ. When c. vanlshes, ít becomes necessary
to extract one singular term from the íntegrand before taking tbe limit. One obtaíns:

C2 (>.)À~'C(À) =B(g')exp [21 1

dcr 112 (g(u;) -!12] ln>.+ regular terms
t
A2(g) In A2 + regular terrns. (A26.l7)

fu!pla.cing C(>.) by thilo.expressíon in equation (A26.!4), we recover equation (A26.6).

----------------------,~·IA<Wf',t.'l>t.'.J•'J<'I~·ni,J'I,l!.
CORRECTIONS TO SCALING BEHAVIOUR

In pre.-ious chapters, while deríviug the scaling behaviour of corre!a.tion functions, we

have always kept only the leading term ín the criticai region. For instance when we
ha.ve solved the RG equa.tions we have always replaced the effective coupling constant
g(>.) at scale À by g', neglecting the small dilference g(À) - g', which V!Uiishes only
íf g = g• Moreover, to establish RG equations, we h,.d alre,.dy neglected corrections
subleading by powers of A, a.nd effects of other couplings of hlgher ca.nonical dimensions.
Subleading terms relate<! to the motion of g{>.), wbich give the leading correctíons for t:
small, can easily be obtained from the solutions of tbe RG equations derived preViously
a.nd will be discussed first. The situations below a.nd at four dimensions ( the upper-
critical dimension) have to examined separaLely. The second type of corrections involves
additional considerations and will be studied in tbe second part of this rhapter. Finally
a physical applico.tion is províded by syst.ens with strong dipolar forces wbich have tbree
as upper-critical dimension.

27.1 Corrections to Scaling: Generic Dímell6ions

In dimensions d < 4, to characterize the corrections to scaling due to a.n initia.l value of
tbe q,• coupling g díiferent from tbe fixad point vnlue g', it iB convenient to solve tbe
RG equations in a s!ightly dífferent manner by introducing a set of coupling constant
dependent renonnallzations:

. dg'
In Z(g) = -
1 !3
~·
M(g) = Mz-t12(g),
(!!') ['I (g') - '1),

{ t(g) ""t exp [/." ~ 1

(- - .!.)) ,
(27.1)

r fJ (g') " (g') "

and a new o::ouplíng constant ii whícb charscterizes tbe deviation o( g from g':

.9"' (g -g')exp lC dg' ({J~·}- (g' ~ g·}) 1· (27.2)

We then substitute:

r<"l (p,;t, M,g,A) = z-nf2 (g)r<nl (p,;t(g), M(g),g', A) c<nJ (p;; t(g), M(g),.ii, A).
(27.3)
The functions ç(n) satisfy the foliowing bcundary conditiom;:

ct"l(p;;t,M;O,A) 1. (27.4)

The linite renormalizations (27.J) eliminate trivial deviations fro.m the fixed point thoory
wbich correspond simply to a cbange of nonnalization of tbe díiferent scaling VBriables.
27.1 Corrections to Scnling. Behaviour 609

Equation (26. 71) then ímplies:

(27.5)

Solving this equatíon by expsnding ç(n) in powers of g we obtain:

ç(n) {p·Ht) M l::t'

;q)- -
.ç._y-•t•w• n!nl
I+ L ' (p·C"
t ,
tM-l!P) l
(27.6)
......}
in which A has been set equsl to I.
The exponent w, defined as the vulue of tbe derivatíve of /3(g) at g•, whích chatacteri.zes
the approa.ch to the fixed point, characterízes siso, therefore, the leading corrections to
the crítica! b<::haviour.
For models with one coupling constant, ,B(g) has the form:

(27.7)

In such a case one a.lways finds:

(27.8)

Not.e that to render ali t.emu; in the expansion (27.6) dimensionless, in the sense of scsling
dimensions, we can assign to g the dilnension -w.

Sroling /Qr d > 4. Up to now we have considered corrections to scaling for d < 4.
In four dimensions or above, the fuced point corresponds to g• = O, i.e. to the gaussia.n
fixed point, and therefore the leading contributions to ali correlation functions, exrept
the 2·point function, come from corrections to sca.ling since these functions vanish at
the fixed point. I t is easy to verify that this specisl feature of the gaussian fixed point
exphúns tbe apparer.t contradiction berween some RG predictions like relation berween
exponents involving explicitly the dimension d (called hyperscaling rdations} and mean
field exponents. For example if we =ign to the magnetic field H snd the magnetizatíon
M tbe climensions ,Bli/11 and ,B/11 respectiv<lly, the exponents .B and 11 being given by
equations (26.64,26.65), and, acmrding the discussion above, to the coupling constant g
the dirnension -w with:
...,,.._~ ford>4, {27.9}

ali terms in the equation of state have tbe same dimension. Indred with these assign·
ments:
[H] [!) + [M] 3[M] + \g] = 2,$6 2 + 2{J = 6{3- w,
for ali dimensions d > 4. In exactly four dimensions the situatíon is more subtle since
the ,P4 (x) operator is marginal, snd the approach to the fu<ed point is only logarithmic.
Thís wíll be the subjeet of tbe next section.

----------------------~·IA<WI'·t.'l>t.'.J•'J<'I~· JII,J'I,l!. 'l"trl!-.

610 Corrections to Scalíng Behavíour 27.2

27.2 Logarithmic Corrections at the Upper-Critical Dimension

The upper-critical dimension is the dimeDSion at which deviations from mean field theory
appea.r. For our model Üüs dimension is faur. In tlús dímcnsion there generally ,,,;,t.,
a marginal operator, heJ"é J ~ 4 (:r)dx, and therefore, as we have índicated in the general
discussion of Section 26.1, logarithmic corroctíon• to mean field behavíour a.re expocted.
Although the dimension four is not of physical relevance for sta.tistical problems, its
study is of special ped~ogkal value, bocause exact predictions can be derived from RG
arguments. Moreover a 1/>4 interactjon is present in the Híggs sector of the Standard
Model of weak electromagnetic interactions. Because the fuc.ed point corresponds to
g• = O, no as.sumption about the fuc.ed point theory ís required. Finaily let us note
some physíca.l systems bave d = 3 as Uppel'-critical dimension, for example tricritical
systems or ferroelectrics with dipolar uniaxiallong range forces, The latter exa.mple will
be dis<:ussed in Sectíon 27.5.
We study bEtlow only the equation uf state and the specific heat, the generalization to
correlatioo íunctíoM being stra'íghiforwa.rd. Let us rewrite the relation (26.57) for • 0:

- -'1'('")(..\A)a H ( M(>.)
H(M,t,g,A ) -Z t(.\) ,g(À,l.
AA '.X'A ) ) (27.LO)
2

We recall the definitions of the various functions:

-19(~)~
!n>.- • t3(g')' (27.11)

(')
1
.[~)
lnZ(Ã) :o dg'!I_!_ (27.12)
9 t3(g')'
1
M(.X) {AA = [>.z'1 2 (>.r M JA, (27.13)

t(À) t J.9Pl dg'

In ).2A2 =In A2- • i3(g') v (g')" (27.14)

It is also necessary to here recai! the expansíon ofthe RG functions for g small (equa.tioll5
2
(26.79,26.99,26.102)). In the (4>2 ) field theory, </> being a N-component vector field ,
i.e. in the O(N) symmetric N-vector model:

{J(g) (N; 8) ::2 + O (g3) '

fJ(g) = (N7; 2) (s!• )'+o (l)' (27.15)

"-l(g) = 2- (N + 2) .2._ +O (g').

6 81f2

As we have done previously we choose À such that:

M(>.)=.XA, (27.16)

or, ín zero magnetízationt solution to the equation:

(27.17)

~· Jlt.J•t.l:'. 'F\r""'"_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _...,,~,u.,wt·!.hk' ~'i•'t·~ J

27.2 Corrections to Scnlíng Beluwí<ror 611

1n both cBEes the s<J!utíon >. of these equations ís smBII in the critícBI domain. The
corresponding effective couplíng g( ~) ís then also small. lt is easy to integr,.te equation
(27.11) in this regime. One finds:

..\ -s,.•~6__1_ [lnl!n>.l]

g( l~ N +SIIn>.l + 0 (lnl~ll' · (27.18)

Dueto condítions (27.16) or (27.17), the theory at scBie À is no longer critical; further-
more g(À) is smBll: thereforé Hp) in the r.h.s. of equation (27.10) can be calculated from
perturbation theory. Note hete the pow..r of the RG rnethod: we started from a theory
witb a couplll\g constant of order 1 and perturbative ooefficients increasing like powers
of ln(t/ A2 ) or equivalent. Diroct perturbation theory WM obviously meaningless. In con-
trast the rescBled theory has a small wup!ing constant g(>.] and perturbativc coefficients
of order 1 dueto conditions (27.16) or (27.17).
For example let us use equation (27.16):

and cBlculate t(>.):

(') I
_t_..._ ~K( )-t-(11 >.i}-(N+2)/(N+sl ~K( )~I -Ml-(1'1+2){(1'1+8) (27.19)
;.,>A> g >.2A' n g M' n A '

K(g) being a finite temperature renormaliza.tion. Usíng now the perturbative expansíon;

H=tM+~M 3 +-- ,
and relations (27.10) and (27.18), it is easy to derive;

2
K(g)tM 871' M3 ( 3 1nllnMI)
H(M,t,g,ll = 1) = (N+')f(.N+B) +N &li Ml +O M - - -
2 • (27.20)
llnMI + n IInMI

At Te we thus find:
H o: M 3 /llnMI. (27.21)
The spontaneous ma.gnetization is given by:

(27.22)

Since expression (27.20) is not uniform for M smBll, it is necessary to use equa.tion (27.17)
to cBleulate the susceptibility in zero fie!d. We then find:

(27.23)

FinBily the spedfic heat satisfies the RG equation:

[A.!!_+ {3 (g) !!_- !J';l(g) (2 + ~~)] r< 2 •0 l = B(g) (27.24)

ôll ôg at '

---------------------~·IA<Wf',t.'l>t.'.J•'J<'I ~· JII,J'I,l!. 'l"tr~

612 Correctíons to Scaling Behcwíour 27.2

in which B(g) has the expansion:

N
B(g) = 16..2 + O(g). (27.25)

lt is easy to verífy thal a function C,(g) so)ution of equation (27.24) and thus of

iJ(g)C~(g)- 2'12(g)C2(9) B(g),

is necessarily singular at the origin (see alro Appendix A26). Fo> example one can take
a so!utkm of the fonn:

(27.26)

For N = 4 an additionallogarithmic singularity is present. The combination rl2,0).,.. c 2 (g)

solution of the bomogeneous RG equation is !ot 9 small domin"ted by the pole of C2(9}:

(27.27)

with:
Ç(>.) = exp [1g(J.)
•
dg' 'h (g')]
/3 (g')
O< (9(..\)r(N+2)/(N+8)
.
(27.28)

Collecting an fact.on; - obtsin the behaviour of the spedfic heat:

riU) (O;t,g,A)- C2 (g) o: jlntl(hN)/lN~) (1 + 0 cnl~:tl) 1· (27.29)

Remork.. It is apparent ftom thes;, expressioos that a parametrízation in terms of

the variabl"" t or M leads to very complicate expressions. A better way of writing
ali these results is to introduce a parametric representation in terms of the effactive
coupling constant g(..\). We can calculate À in tem111 of g(À) from equation (27.11). Lá
us parametrize /3(9) as:
(27.30)
and set:
~ = g(À). (27.31)
Then >. is given by:
(27.32)
In equ,.tion (27.32) the function .À (s) is a regular function of s for s small. The renor-
mallzatíon fllCtor Z(>.) is "regalar function of s. Finally equa.tion (27.14) yields:

~ = ..!..s 2 1 2 ~,+~•/.B.I/.B. e2f/loo [i(s)]- 1 (27.33)

À2A2 AZ '
in which i (s) is a regular function of s and 11(g) has been written:

(27.34)

Then if we impose conrlition (27.17), equation (27.33) parametrizes t as a function of •·

At Je..ding arder, ali critica! hehaviours will be descríbed by a singula.r facto r of the form
occurring in equatioos (27.32) ar (27.33) multiplied by regular series in s = g(>.).

----------------------,~.u.,WI·!.hk' ~'1•'1·~ J
27.3 Cof'1'f!<:tíoru to Scnling Behaviour 613

27.3 lrrelevant Operatars and the Question of UniverSslity

We now examine, in generic dimensions, the contributions comíng from írrelevant opera·
tors. We ..gain stra;s that tbeBe operators have been found to be irrelevant at the gaussian
fixed point, nea.r four dímensions. We stlll rely on the assumption that dímensions vary
continuou.sly when the lR fuced poínt moves away &om the gaussirm fixed point. Finslly
the anslysis ís locsl, we consider only tbe neighbourhood of the fixed poínt.
Let us first recsll pawer counting arguments for 8 genersl theory with an action S( .;1!)
(for JJlQre details see Ch8pters 8, 12):

(27.35)

in which O o (ç/1) is 8locsl monomial of degree n, in 4> with k" derívatives. The dimension
[u.. j of the coupling constant Ua ís tben:

[u.,J d- ka - !n,. (d- 2). (27.36)

We treat ali interactiom in action (27.35) in perturbation tbeary. Order by arder in

the toopwise expansion _, evaluate the divergent part of the corresponding Feynman
di..gtàlllS and add caunterterms to tbe action to render tbe theary finite. Since tbe
&ction {27.35) contains ali poosible monomials in the field, any counterterm is a linear
oombin8tion of the operators O., (,P).
In order for a product of constants u~ to appear in a oountertenn proportiona.l to an
operator 0,., it is n=ary and generica.l!y sufficíent that the condition:

11 = d (Oo(- L [ua.l ~O 1 (27.37)

I

is satisfied. Then the coefficient óu, (A) of the counterterm proportionsl to Oa diverges
lilre 11. positive ~r of the cuk>ff A (or 8 power of logatithm if 11 = 0),

(27.38)

Let us now return to tbe question of irrelevant operators. We resttict ourselves tn d;.
mensions d 4- ~.c > O o.nd small, since as we have seen, these are the only dimensians
for whieh a reliable o.nslysis is possible. The first operators we have for exa.rnple in mind
are <P", tp~ (8~4>/ ... wbich are operators of dimensi.an 6 in four dimensíons.
To introduce the cut-off A in our sta.tisticsl problem we b8ve rescsled the lengths and
the fie!d tjJ (equatians (25.31-25.34)). Therefare eacb coupllng constant u .. is the product
of a pure number g" by a power o f the cut-off wbich gíves it its dimension:

(27.39)

Equation (27.36) gíves for 6.. :

Óa = -d + ka + ÍDa (d 2). (27.40)

!f 6a is positive the rorresponding operator 0,. leads to a non-renarmslizable theory and

we have already stated that it is íttelev&nt. In tbe tree approximation tbe statement

----------------------,~·IA<Wf',t.'l>t.'.J•'J<'I~·ni,J'I,l!.
614 Gorrectíans to Scalíny Bellaviour 27.3

fo!lOWH from equa.tion (27.39): the operator gives contributions vanishlng wlth fi powe:r
of the cut-off. However in high.,- orders the statement is less trivial since divergence5
at la.rge cut-olf coming from lhe momentum integratíon can potentíally compell!!ate the
powers coming from the coupling constants. 'lli díscover what ho.ppeus it is n~
to anclyze the counterterms gen..-ated by these opemtors, using equations (27 .37) and
(27.38).
The totcl power A' of the cut-of! which multiplies the op..-ator 0 .. in a counterterm
is the sum of tbe power A generntcd by the divergence of perturbation theory (equa.tion
(27.37)) and the powers alre..dy present in the coeffidente u 0 (equation {27.39)):

ll.' A-l:ó~~>=t.+Z::lu..,j. (27.41)

I I

and therefore:
A'= d- [Oa) = [ua]. (27.42)
The result is therefore very simple: due to the dlvergences of perturbation theory, inel-
evant operators give indeed non-Vllllishing contributions, but these oontributions can be
ca.ncelled by varying the amplitudes of the relevant or margincl terms in tbe hamiltonian,
hecause ll.' ?: O is equivalent to [u.,J ~ O.
E=mpl<. The leading new corrections oome from operators Ol' whlch have dinlension
6 in four dimensions. The rorresponding intersctions have the form:

In terms of the t<lDormalized operators they have the expa.nsion:

We have denote<! by OJ(:r) the operators which haw dimension 4 in four dimellBioDB:
if>4 and (él,.!Ji) 2 • In the framework of the E-expe.nsion the CO<!fficients Zi; diverge like
powers of !nA, A;j llke A2 and B Jike A\ up to powers of !nA. Taking into account the
powers of A in front of the interactíon terms we see that oaly the oontributions propor-
tioncl to operators of dimensions 4 and 2 are divergent. lf we cancel these contrlbutioru;
by subtracting to the operators of dimension 6 a ""itable combinatíon of operators of
dimensions 4 and 2 we obtain tbe true new corT<lclions which go to zero like A-2+0(•).
Tbis diseussion aJr,o cl!ll"ifies the interpretation of the oonsta.nts r and g whicll pa.rame-
trize the q;• hamiltonian. These a.re not the parameters which for ins~.a.nce appear as
coefficients of the lji2 or </>4 terms in the mea.n field a.nclysis, but instead etfcctive pa·
rameters taking into a.ccount tbe effect of neglected irrelevant operators. The anclysis
of previous chapten; is however at Jeading order not rnodified. lndeed the change in lji2
corresponds only to a modification of the criticai temperature whicll is a non-uniVél$111
quantity. Moreover below four dimensions we have shown tbat ma.ny physical qua.ntities
{universal QUIIJltitíea) do not depem:! on g either, since g can be replaced by ite ftxed point
value g•. Finally let us note that a cha.n,ge in the cut-off procedure correaponds generally
to a cha.nge in the coefficiente of the i=levant pa.rt of the propo.gar.or (,PA 2 ,P .. .). The
effcct of such a change is obtwned from the prev:íous analysis a!Bo. We can now clarify
the concept of universality: below four dinlensions rui dimenslonless quantities in whlch 9

~· Jlt.J•t.l:'. '1'\r,_,_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _...,,~,u.,wt·!.hk' ~, J

11
,, .~
27.4 Con-ections to Smling Behaoiour 615

cen be rcplared by g• the IR fixed point value, and whích do not depend on the normal-
izations of the field rp, the devíation from the criticai tempero.ture t, and of the roagnetic
field are universal. Obvious examples are rati05 of amplitude of singularities below and
a.bove T" ratios of amplitudes of lea.díng correo:tions to scaling, the rescsled equation of
state (relation between H a.nd M), the renormaliz.ed correlation fnnctions as defined in
Chapters 25,26, etc ..
Remar*. Another simple consequence of this analysis is the following: Let us add to the
hamiltonian an irrelevant operator wbich hreaks a symmetry of the harniltonian. Then
tbe eymmetry of tbe criticai theory will be broken íf a.nd only íf the irrelevant operator
can generate relevant or marginal operators breaking the synunetry. An exarnple is the
following: on the lattice operators of tbe forro I:~ f t/>{::t) (â,.) </>(::t)dx which break the
4

O (d) symmetry of the effective <J>'(z) action are present. However these operetors have
11 hypercubic symmetry and since the only relevant operatcrs they ean generate, like
f (8p,P) 2 ddx, dueto the hypercubic eymmetry, are O(d) eymmetrit, the O(d) eymmetry
of the critical theory is not broken. Conversely the additíon of a naively irrelevant term
Jlke f t,6 5 (:r)ddx te a harniltonian whích is symmetric in 1,6 ,_, -4> generates relevant terms
linear in 4> which are equivalent to the o.d~ition of a magnetic lield.

:Z7.4 Corrections Coming from lrrelevant Operaton;. lmproved Aetion

For simplicity we oonsider the effect, in the criticai theory, of an operator Oo at first
arder only in the correspooding o:oupliog consta.nt "a· The following discussion applies
in the ftarnework of the E-expa.osioo and relies on the results of Chapter 12 eonceming
the renormalization of composite operators.

1!7.4.1 Co..-ectíom to scaling

In Section 27.3 we have shown that an operator Oo gives contributions equivalent to
ali operators of lower canonlcal dimensions. For example t,68 (:r) first gener11te!l effects
2
equivalent to 4> 2 (:r), 4>4 (x) o.nd (al'.p(x)) and ali oper11tors of dimenllion 6, snd then
genuine new oorrections. To isolate these corrections it is necessary to subtract from the
operator a linear combination of ali opemtors which have a lowe.- dimensions at d = 4,
i.e. to perform an o.dditive renormalization. Note that we can omit ali operatorB which
ue total derivatives since only the space integrals appea.r in tbe action. We define a
au btr acted operlltor Õa ( 4>) by;

Ô,.(t/>) = 0,.(4;>)- L CaB(A,g)Of!(tP)· (27.43)

P s...,L thot IO•J<[O.j

For example in the case of operatcrs of dimension 6, Jike 4>6 (x), one subtracts a linear
rombinatíon of operatcrs of dimensions 2 and 4:

The coefficienb! G 1 , c, o.nd C3 can be determined by a set of renonnalization conditions

at zero momentum;

r~~ (p) 0 (p4 X powers o{ )np) for p-< 0,

(27.44)
r~4) (Pi =O)= O.

----------------------~·IA<WI'·t.'l>t.'.J•'J<'I ~· JII,J'I,l!. 'l"tr""""""""

616 Correction.< to Scnling Behaviour 27.4

The lirst conditíon implies in particular that the criticai temperature is not changed.
These conditions are not affected by IR divergences because the correlation functíons with
Rll operator insertion ha.ve positive dirnensions. After such a.n additive renormaHza.tion,
the bare operators of dimension 6 are relaterl to the oompletely renormnlized operators
o~ hy:
~-~.f ~I0'0 (X) =L Zat.dg,A{J.<)A- 6•
~
f ddxÓI'(:r), (27.45)

in which I' i6 the renormall.:ation scn.le. Additional renormalization conditions at scn.le

J.I. for the insertion of renormalized operators determine the roa.trix Zafl·
Tbe relation between correlation functions r~l with one A-!. f d•:z 6., ínsertion and
the renormrilizerl. function5 with !'-I. f ddr 0.,(~) lnsertion ís then:

I:z..~'z"nr~1 (F;; !I, A)= [rg-2 ~p,,g,j.LJJ....•. {27.46)

IJ

This Jeads to the RG equations:

(27.47)

with
(27.48)

Effects of the insertíons of the operators 6" are then gowrned by the ,.,jues of 1Ja the
eigenvalues of the matrix 'la!J (g• ). From relation (2'7.45} we see that the renormal.i'l'atíon
matrix Za~J has the form:

(27.49)

Thuefore the eigenvalues 'lo are at leading order given by:

'The effects of the inelevant operators of C8llonic::al dimension d + 6a in four dimomsiolll;,

are suppressed by powem A -•. +O(<) of the cot-oH A. For operators of diroension 6,
á = 2. In ao infinitesimal neighbourhood of dimension four, these operatom remain
irrelevant. Our general ana.lysis, which is base<! upon tbe idea that the criticnl behaviour
of fenumo.gnetic eystems can be descríbed by an effective (jl4 tield theocy, remains valid,
beyond tbe c-expom;ion, as Jong as this property remains true.
Remarl;.. Note !inally that in some cases the irrelevant effects may be specinlly impor-
tant. An CJUUDple is províded by syst.ems where the initial theory has only a discrete
symmetJy while t.he eymmetry of the criticai theocy is continuous. In the low tempera-
ture phs.se the critica! theory has Goldstone mode singularities. TheSE singularities are
suppressed by irrelevant correctioii5.

~· Jlt.J•t.l:'. 'F\ r,______________________,

~.u.,WI·!.hk'~'i''i'~J
 27.5 Corrections to Scn./ing Eeh<>11iaur 617

27.,+.2 Fixed point in hatniltoman space and impruved attions

We have seen that by adding to the !J> 4 (x) lield tbeory írrelevant interactions, we could
modify correlation functíons by terms in the e~pansion of arder 1/.112 (up to loga-
rithrns). In the r.h.s. of RG equations (25.43) w.. have just neglected terrns of the same
arder. Symanzik has shown in perturbation theory that by adding to the hamiltonian tbe
proper linear combination of irrelemnt operaton>, it is possible to cancel exactly these
corrections. The coefficients of the linear combination are functíons of tbe q,• coupling
constant g. For example the complete set of operators of dimension 6 ca.n be used to
cancel exactly tbe corredrons of order 1/.11 2 in the r.h.s. of the RG equBtions (25.4.3), the
operaton; of dimension 8 to cancel the order 1/A• ll1ld "" on. An iteration of this proce-
dure leads to a theory which depends on only one q, 4 roupling constant and whích satis!iea
the RG equations exactly. lt is actually a "renorma.!ized" theory ronstructcd wítbout
using the renormalization procedure, constructed by consíderlng an infinite sequence of
hamiltonians.
From the gllnera.! RG point of view, the hBmiltonian.s which lead to corrdation functions
satisfyíng RG eq uations exactly belong to a one parameter line in hamiltonian spa.:e
whlch goes from tbe gaussian fixed point to the non-trivia.! IR fixed point.
Conversely by constructing directly tbe r.;,ormo.lization group for a cut-off lield tbeory,
it ú; possible to prove the existenre of the renormalimd field tbeory (see Appendbt A10.1).
Finally Jet us mention that Syrrm.nzik h as advocat.ed the use of such improved actions
(adding for example a.ll terms of dimension 6) for numerícal ca.!culations on tbe lattice. lt
should mentioned, however, that tbe applications of this ingenious idea have, up to now,
been disappointing. Because the lmproved action involves more extended interactions
on the lattioe (like second nearest neighboun;) the effective size of the la.ttice is reduced,
increa.si.ng finite size effects. The rnethod ís tbus oaly applicab)e to very la.rge lattices.
But statlstics then becomes in general the domina.nt problem, and more oomplicated
interactions slow down tbe nurnerical ca.!culatiow;.

27.5 Applieation: Uníaxial Systems with Strong Dipolar Forces

In Section 27.2 we have str=ed that the renormBiization group leads to exact predictio~
f, for criticBI systems at the upper-critica.! dlmension. Unfortunate)y in tbe case of tbe N-
'< vector model the upper-crltical dimeiil5ion !s fom and therefore the predictions are not
useful for experimental physics. The mo.in application is numerlca.l physics, in the sense
that the Higgs sector (N = 4.) of tbe Standard Mode) has been investigated numerically.
Th.,.efore we here preaent another system on which accurate measurements have been
mo.de, and whlch has dlmension three as the upper-critica.l dlmension: a unirodal ferro-
magnet or ferroelectric system with strong dipola.r forces.
Dípolar /OfUJ. We consider a spin system in d dimensions in which the d-component
spins S~' interact both through short range and dípolar forces:

-~H (S) L v,.. (x- x') s~s;, + 7 (S,.. V,.) (Sx'. v,.,) I •-2. (27.50)
"·"' lx-x'!
v,.. {x) being the short range potentia.l. We 8S8ume that the long range dipolar forces are
strong enongb to play a role in tbe part of the criticai domam accessible experiment..!ly•
.' In o.ddition we assume tbat the lattice is strongly 8.Dirotropic in such ,. way tbat only one

-----------------------,;~,fA<WI'•t.'l>t.'.J•'J<'I~· JII,J'I,l!. 'l"tr"""""""

618 Co~tíans to Srolíng Behavirmr 27.5

component of the spin S is critica! and the effective harnílt.onia.n ca.n be simplified into:

-(fH. (S) =I:

x,x'
s.s., [v (x- x') +..,a, a,. \x-x1,Id-•] , (27.51)

in which S now denotes the component of tbe spin wctor S along the z = :z;d axis.
After Fourier traru;formation tbe hamUtonlan ca.n be written:

(27.52)

in which we recai! that V (q) is a regular functionofq which, dueto bypercubic symmetry,
ha!; the exp..nsion:

(27.53)

In tbe criticai domaln, in which \'li is sma!l, the two tenns comlng from the sbort range
pot.ential and the dipola.r force$ are of tbe same mder of magnitude:

(27.54)

Tlús implies that q,, the z component of the vector q, is much sma!ler than the other
oomponents q.l:
(27.55)
We can therefore simplify the intera.ction p<>t.ential. In the case of B.Il even spin distribu-
tion, we finally obtain an effective hamiltonlB.Il 1i { t/J) of the form:

71(1/J)-" ~f ddqr/>(-q)(ql +AÕ~ +re+to)r/>(q)

+~f ddql ... ddq4.5(Eq•J<I>!q,J ... .p(q.). (27.56)

The upper-critical dimension. Usual !I<J""''" counting is now modified because space i•
no longer isotropic. ln units of the tnws~~erse companents of q, tbe dimension of q, is 2
(equation '(27.55)):
[q,] ""''2 ~ [z) = -2.
The volume elernent in configuration space dx_t dz h as thus canonical dhnension -d-I.
This implles that powtt counting a.nalysis is the same as in the conventlonal rP 4 theory
in (d + l) dimensions. In particular the upper-<:ritical dimension is given by:

d+1=4 ~ d=3.

The renCilmalization group rnethod allows to make exact l>t""'ictiom; in the physical
dimension d = 3.
Renormalizatíon group "'ll'ntions. From expression (27.56) we read off the propagator
8(q):
27.5 Correctíons to Sco.líng Behamour 619

Díagrams calculated wíth this prop..gator are regu!lll' for q small, therefore

will never a.ppear as a. counterterm.

The renormalized ha.miltDnia.n has thus the form:

1-l, =i f ddq,p(-q) [zrrl +A2 ~[+6m

2
+ z,t] ,P(q)
+ *Ag Z 9 (g) f d4 q1- .. d4q• 6(I;q,),P{q.) ... <;\ (q<), (27.58)

in which p is the renortnali2ation scale and c is now defined by:

(27.59)

Let us introduce .Uso a bare dimensionless' coupling consta.nt:

Uo =A•Aogo. (27.60)

The rela.tion between bfU'e a.nd renormalized cortela.tion functlons reads:

(27.61)

In addition comparing expressions (27.56) and (27.58) we find:

A= z11 2 Ao. (27.62)

RG equa.tions foll<JW:

(
112 g,)loâtoa] r (n) .
(p;,go,Ao,to,A) o.
(27.63}
fuo-loop calculatíon of RG frmction:J. AI< was expla.ined in Section 25.4, for practical
calcu!ations we use the renormalized theory and mínima! subtraction (see Cba.pter 11).
· In wbat follows we omít, for oonvenience, the index r on renorrruilized functions. The
renormalized RG eqoations are formally ídentical to equations (27.63):

(27.64)

but tbe RG functions are different a.t two-loop order.

Tbe calcula.tions are here similar tD the dynamical calcula.tions as described in Ap-
pendix A17, beca.use if we identify tbe :r direction witb time, tbe prop..ga.tor.; ba.ve tbe
"""'e denominatDrs. Tbe combinatorial factors of Feynm&ll diagrams are tbose of the
9~ lield theory. Only tbe ex:pressions of tbe diagrams dilfer. We need only tbeir values
at 'Ylmlsh:ing external z componeuts oí momento., and tbis simpli6es tbe integratiou over

---------------------~·IA<WI'·t.'l>t.'.J•'J<'I~· JII,J'I,l!. '1"t11i~

620 Corre~;tian.'i to Scnling Behal!iour 27.5

Q (a) (b)
Fig. 27.1 Díagranu; neoded at two-loop order.

The final result is:

1 ( 2 )'r(l+~) _,.
C, (t:)"' (lfi"')d-1 J3 -e-,- (IPI + P2IJ +O (l),

~--
1
C ()
2 e
- +0(1)
3 (1611')2 E:
The renormalization coDStant.s Z9 , z and Z2 are tben determíned up ló order g 2 :
3Nd 9 3 4 2) N•g
( !2-&lnJ'-& +0 ( g3) ,
2 1
Z9 1+-,-g+ (27.65)

2 2fJ2 ( g 3) , (27.66 )
Z= 1
27 N.-~ +O
Zi 1 = 1- Ndfl. + (-2_ + 2_ + 2_ln ~) Njg' +O (if), (27.67}
f €2&2e3

,,_,,_,__ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ ~,u.,WI·!.hk' ~'1•'1·~ J

27.5 Coi"T'tttions to Scnling .Behoviour 621

with:
(27.68}
The RG functions follow:

{3(g) d
-< [dg In (gZgz-at•)
]-1 , (27.69)
d
')(g) ~(g)dg In Z(g), (27.70)

'12(9) = v~) 2 = ! In (z,z-

1
)' (27.71)

a.nd therefore:

Nd/3 (§) = -E9+ 39 2 - ( -6ln ~ + ~) g3 +o(.§')' (27.72)

'J(.§) = ~9 +o (9 2 3
)' (27.73)

'72 (g) = -.§ + (~ + ;n ~) ii 2

+ O (93 ) , (27.74)

with:
(27.75)

The scaling belull!iour of correlation frmctians below three dimmsions. Dimensional

anBiysis in the criticai theory tells us:

r<nJ (.\. t A )
PJ.,PJlz; ,9, oi-'
.>."+!,.-2)(J-d)/>p(2-n)/2r{nJ (P P • _!_ g A.E_ ~)
J., •• ).>' • )12' À •
(27.76)
In d 3 - õ dimensions the model ha.s IR fixed point g• (c). At the fixed point we find:

r<nJ (PJ. p t A= " = 1) = t'-(n-2J<I,.rtn)

' "1 ' ,.. (R!. _P_•-)
t" t~(2-Q/l) ,
t
(27.77)

with
-y = v(2- rr),
At two-Ioop order the mtponents o:re:

g•(E) = ~ + Gln~+ :4~) <2 +0 (•3 ), (27.78)

4
'J = 243"2 +o (•3). {27.79)

1F
1
=2 20) 1!:2+O (c3) ,
g1 In 34 + 243 (27.80)
3 (

w e (~Jn~+~)e
3 3 81
2 +0(<. 3 )
. (27.81)

Logarithmic corrections to m€<Jn field behaviour in three dímensions. In three dimen-

. sions the RG equations can be solved as indicated in Section 27.2.

·-..-------------------~·IA<WI'·t.'l>t.'.J•'J<'I~·ni,J'I,l!.
62:2 Corrections to Scaling Behaviour 27.5

For tbe effectíve ooupling constant at •cale >. one finds:

_1_ ~ .2_ In~
[ 1 _ 2 (17 + 27ln4/3} In jln>.i + 0
g(>.) ~-o 16" >. 81 ll11 >.i In>. ·
(2.)] (27.82)

A short calculation then yields, for example, the susceptibility in zero field:

x- 1
- C± llntj-
113
(1 + 2: 3 (108m~ +41) ~~~~~+O cl~tj) 1, (27.83)

or the specific hea.t C:

2 ~3 (lOS In~+ 41) ~~~~ti +O Cnlllj)],

113 (27.84)
C= A± lln jtll [1-
witb tbe universal ratio
(27.85)

The specific beat has been measured in a high preci•ion experiment on tbe dipolar Ising
f<,rrouu•gnet LiTbF 4 by Ablers e! aL Fitting the specífic heat by:

C+='!: {[1 + bln (a/jt\)]'

tbey lind:
c_ !: { [1+ bln (a. /jt\)]"'
1}

n + B,
+ B,

A+
A- = 0.244 ± 0.009'
0.336 ± 0.024 ,
z = z'
results whlch agree beautifully witb tbe theoretical predictions.

Blbliograpbical Notes
Correction terms due to irrelevant operaton; are discussed in
F. Wegner, Phys. Reu. B5 (1972) 4529; B6 (1972} 1891, and in Wegner's contribu-
tion to Phase Tiunsitíons and Critica.l Phenomena, vol. 6, quoted in Cbapter 24; F.J.
Wegner and E.K. Riedel, PhyB. Rei!. B7 (1973) 248; E. 8réz.in, J.c. Le Guillou and
J. Zinn-Justin, Phys. RetJ. DB (1973) 434, 2418 and in Phase Tiunsitíons and Grit.irol
Ph.enomena vol. 6 alreo.dy quoted, Section VIJI.
R.ecent calcula.tiow; of dimenslons of opemtors CAII he found in
S.K. Kehrein and F. Wegner, NucL Phys. B424 (1994) 521.
The logaritbmic correctioru; in four dimensiono for the q,• tbeory and tbe uniaxíal system
witb dipolar fot-ces ha'Ye been obtsined, witb the use of the parquet approximation by
A.I. L8lkin and D.E. Khmel'nitakü, Zh. Ek3p. Tear. Fiz. 56 (1969) 647 (JETP 29,
1123).
The idea o[ imptoved action has been discussed in
K. Symllll2ik, in Mathematica.l Prnblems in Theoretirol Physics (Lecture Notes in
Phyrics J53}, R. Schro.der, R. Seiler and D.A. Ublenbrock eds. (Sprlnger- Verlag, Berlín
1982); NucL Phys. 8226 (1983) 205.
The application of tbe renonnalization group to tbe unía>tial system witb dipolal- forceo
is found in Wegner and Riede! as quoted abavc and
A. Aharony, Phys. Rev. BS (1973) 3363; E. Brézin and J.. Zinn-Justin, Phys. Rei!. 813
(1976} 251.
The experimental results are quoted from
G. Ahlen;, A. Komblit and H.J. Guggenheim, Phys. Re11. Lett. 34 (1975) !227.

~· Jlt.J•t.l:'. '1'\r---------------------·~.u.,wt·!.hk' ~'i•'t·~ J

27.5 Corrections to Sco.ling Behaviour 623

Exerclses
The tricritica.l ( </>2 ) 3 field theory is aloo IR free in tbree dimension:s. Verify that tbe RG
equations for tbe n,P, I<J>2 , ktfJ 4 correlation functions can be written:

{
{j
A-+
&A
{j
,B(v)- -
&u
lJ
k'74(v}u-
&.. &t
ân}
-I (112(11)t + 112<(v)u']- --fi( v)
2
r(t.l,n) o,

except for I =
n = O, k $ 3. Cal.culate tbe RG funciions and lltudy the logo.rithmic
deviatioru; ftom mean field theory. A review of tricritica.l b«haviour cB.D be found in
D. LawTie and S. So.rbach, in Plr.ase 1mnsitions and Critico.l Phenomena vul. D, C.
Domb and J.L. Lebowitz eds. (Academlc Press, London 1984).

----------------------~·IA<WI'·t.'l>t.'.J•'J<'I ~· JII,J'I,l!. 'l"trr,;,.......,..

28 CALCULATION OF UNIVERSAL QUANTITIES

Many modela witb second order pha.se transitions have been investigated, the effective
field theories ;dentified and then vaJ"íous universal qulllltities eolcuiated by field theoret-
icol methods. We can of course here report only " smoll number of signilkant results.
Therefore in this chapter we consider the important example of the N -ved.or rnodei, i.e.
the O(N) symmetrit (4>~) field theory,
2

'H (.p) =f U[8~,P(.:z:)] 2

+~(r<+ t) <P2 (.:z:) + ~9A" (<P2 ) 2 (;r;}} dd:t, (28.1)

in which <Pisa N-vect.or.

We present results for criticai exponents, the equation of state and some IU!lplitude
ratios. We díscuss more thoroughly critico.! exponent.B because they allow the most
detailed and accurate comparison hetween Field Tbeory, lattice models IUld experiments.
Univernal quantities have been calculated in Field Theory by two methods: the <-
expansion, which we have systematically discussed in previous chapters, lllld perturbation
theory at futed dimensíon (see Section 26.1). In both esses the expansion is divergent
for ali values o! the expansion parameter. The rate of divergente can be obtained ftom
inst.ant.on calculus, as we sholl explain in Chapt.en 41 and 42. There exist methods t.o
deal wíth divergent series (see Sections 41.4-41.6), ali of which rely on some additional
kuow!edge about the anolytic propertie; of the function which h!!.S been expanded. In
the case of the t/>• field tbeory, the Borel summation h!!.S been extensively used because
tbe infonnation drawn ftom the llll"ge behaviour analysis can easily be b]corporated. In
tbe case of the series ..t fixed dimensíon Borel summablllty h!!.S actuBI!y been rigorously
proven, For the <-expansion Botei summability has been 88Sumed. Final!y to sum the
$<!rles efliclently, it h!!.S been necessary t.o moke some plausible assurnptions about the
an.aiytic properties of the Borel transfonn ít.Belf. Therefore the reliability of the results
hBB been estírnated by checking their stability with respect tó reasonable variations of
tbe surnmation method. Furthermore the comparison between the two fa.milies of results
has provided o.n internai consístency checlr. of field theory methods. The a.greement with
series estímates coming from higb temperature exp~tnsions on lattiCI'S is then a verificatíon
of the con'cept of unlversaiity in theoreticru models. Fina!Jy of course the nltimate test
comes from the confrontation with experimental results.
The results of the N-vectór model do not apply only t.o ferroma.gnetic systems. In the
appeudix Wfl ex:ploin how the N = O Jimít oorresponds to the statistical propertie$ of
polymers and how the Ising-lil<e N = 1 model descrihes the physics of the liquid~vapour
tra.nsition.

28.1 Tbe c-Ex:panslon

Ma.ny quimtities ha\11! been calculated at throo-loop arder within tbe <-expansion. More-
over a serles of very ingenious tricks and tbe UBe of symboiic manipu!ations on computers,
have all<>Wed the determination of criticai exponents in the N.vector rnodel, up t.o order
6
E • The higher order calculations have been done using dimensiono.! regularization and
the mínima! subtraction scheme (see Cbapter 11).

'-"---------------------'~•U.,WI·!.hk' ~'1•'1·~ J
 28.1 Calculatíon of Uniuen~ Quantities 625

28.1.1 Oritiml e.zponents

Altbough the RG functions of the ( <P~) theory and therefore the criticai exponents are
2

known up to five-loop order, we here g\ve the expansions only up to three loops (four
loops for 1/), refurring to the literature for more details. In tcrrns o[ the varia.ble !J:
• 2g
(28.2)
g = (4>r)d/ 2r(d/2)'

tbe RG functions 13(§) and TJ2 (§) a.t three-loop order, TJ(g) at four-loop order are:

"(g"J • (N + 8) • (3N + 14). 3

JJ -Eg+ --6-g2 - --1-2-g
[33N2
+ 922N + 2960 + 96(5N + 22)((3)]. 4 O("")
+ 123 g + g ' {28.3)
2
n(g") (N + 2) ·2 [ 1 (N + 8). 5( -N + IBN + 100) 9·•] O("")
· ---n-u - -----u--- 9 + 576 + 9 ' (28.4)

(") (N + 2) _ [l 5 . (5N + 37). 2] O ("4) (28.5)

'119 =--6-g ]29+~9 + g '
in which Ç(s) ís the Ríemann (·function:
((3) = 1.202056903!5 ....
The zero 9"(<) of tbe P..function i5 then:
2
.• & [ 3(3N + 14) (18(3N + 14)
g (õ) = (N + S) 1 + (N + 8)2 E+ (N + 8)4
2
33N + 922N + 2960 + 98(5N + 22)((3)) •] O ( 4) (28.6)
S(N + 8)3 E + E •

The Vlllues o{ the eritical exponents '), 'Y a.nd w,

'I TJ(§"), 1= 2 :~~g•)' w=f3'(9"),

follow:
2
TJ
•'(N + 2) { 1 (-N + 56N + 272) e 1 [-sN• 230N3
2(N + 8)2 + 4(N + 8)2 + 16(N + 8) 4
2
+ll24N2 + 17920N + 46144 384(5N + 22)(N + 8)((3)] E } +o (t: 5 ) '
(28.7)
_ ~( 2
(N+2) ) 2 (N+2)
I-1+2(N+8)"+ 4(N+B)" N +22N+52 E +B(N+8)5
)( [N4 + 44N 3 + 664N 2 + 2498N + 3104- 48(5N + 22)(N + 8)((3) l E
3

4
+ O (e ) , (28.8)
3(3N + 14) 2 [33N2 + 922N + 2960 + 98(5N + 22)((3)) _,
w =E (N + 8)1 ~ + 4(N + 8) 3 "
2
(3N + 14) 3 O ( •) (28.9)
~ 18 (N + S)• c + E •

IU'I.l:'. '1'1 "~.,.------------------~·IA<WI'·t.'l>t.'.J•'J<'I ~· JII,J'I,l!.

625 Calculatían oj Univer•al Quantities 28.1

AI! other exponents can be obtalned from tbe scaling relations derived in Chapter 26.
Note that the resuJtt; presented ab""" ínvnlve ((3). At lúgbet orders ((5) and ((7)
successívely appear. In table 28.1 we gíve the value.; of the critica! exponentt; 1 a.nd 'I
obtained by simply addíng thE' successive terms of the e-expansion for E = I and N = l.

TabU. 28.1
Sum of lhe •uccessive t•"""' of IM <-<:tpanrion oJ "T and 1] for E= 1 ond N = I.

k o 1 2 3 4 5

'J ).000 1.167 1.244 ).195 1.371> 0.00

'7 0.0 ... o.o ... 0.0183 0.037'2 0.0289 0.0649

We ímmediately observe a striking pbenomenon: the sums first seem to settle near some
reosonable value a.nd then begin to diwrge with íncreasing oscillatíons. We shall argue
in Cbapter 42 that tbe t:-expansion is dlvergent for ali values of t:. Divergent series ca.n
be use<! for sma!l values of the argument. However only a finite number of tenns of the
series can be te.ken into ac<:Ount. The lset terrn !ldded gives an indication of the J>íze of
the írreducible error. For the exponents '1 and I] we therefore conclude from the seri<ls:

"J=L244±o.oro, I] = 0.037 ± 0.008 '

where the erron; are only indicative of tbe uncertainty about the resnlt.
Remarlc. The definition of the ,B-function hy minimo.l subtractíon bns an intrinsic
mea.ning, unlike other definitions, since then {3(g) can be reco.lculated in tenns of w(<).
Setting:
{3(g) = g (-c+ s(g)),
then:
g(s) = asexp [f ds' c(~)-?)].
where a is an arbitrar)' normalízation of tbe coupling constant.

28.1.2 The scaling equation of •tate

The sca.ling equation <>f state provides an interesting exe.mple of a universal function. lts
t:-expansion has been obtainad up to arder €1 for arbitruy N, and order ,;3 for N L
Weset:
(28.10)

in which the norma!ízations of x and the function /(:t) are such that:

f(O) I, J(-1)=0. (28.U)

It iB also conv~nient to set:

%=:<+3, I' •J4y. (28.12)

The expansion up to arder t:2 is then:

(28.13)

-----------------------~·V~.>Ln·!.hk' ~,1,,.~ J
28.1 Colculatíon of Uníver.sq.l Quantities 627

with:
1
h(:r) Z(N + S) I(N l)ylny+ 3z hu- 9yln3 + 6:rln2] {28.14a)

h(x}
1
[z(N + s}f {IN 1 + 6ln2- 9Jn3 + (N- 1JlnvJ [3zln• + (N- 1)11 1n 11

+6:rln2-9yln3]+ ~(10 N)y(ln2 :r Jn 2 3) +36(1n 2 z yln 2 3+:r!n 2 2)

27 -4N2 + 17 N + 212
-54ln2(1nz+:rln2 yln3)+3ln4(N-l)ylny+ N+B
x [zlnz + 2:rln2- 3yln3J + (N -l)ylny!n •- ~N(N -l)yln° !I
N 1
+ N + 8 (19N + 92)y In y- 2(N - I) [(:r+ 6)J1 (:r} - 6yJ1 (O)]
-6(N -1) [J2(x) yJ2(0)] + 4(N 1) [J3(:r} -yJ3(0)J), (28.14b)

where:
(28.15)
and:

= dulnu [
I 1 (P}'"'
d
L J/2
- - (1- u/p) 8(p- u)
u(l-u)
(28.16)

lo(p) = P dp h (p), (28.17)

la(P)"" h(P) + 2/,(p). (28.18)

The expre>sions (28.14) a:re not uniform, a.nd valid only for x of order 1. For x la:rge,
i.e. for small magnetí2ation M, the megnetic field ha.s a regular expansion in odd powers
oi M, i.e. in the variable x-IJ (S<lction 2ti.5). lt is therefore convenient to lntroduce
Josephson's pa:rametrization which leads tD a repr.,;entation uniform in both limits.

1!8.1.9 Parumetric representation of IN. equation of stau

Weset:
(28.19)
where %o is an arbitrary pasitive constant. More directly we can pa:rarnetrize M and t
in t.erms of two variables R and fi, setting:

(28.20)

Then the function h( 9):

h(8) ~ f{x(9)), (28.21)
is an odd functíon of 9 regular near 9 I, whích is x small, and near 8 .= O which is x
la.rge.
For the special choice
(28.22)

~.-------------------·~·IA<WI'•t.'l>t.'.J•'J<'I~·ni,J'I,l!.
628 Ca!culation of Universal Quantities 28.1

th.: equation of state at order e takes a rather simple forro:

h(6) = 9 (3 - 211
2) [1 +;iZ +~i In (3- 20
2)] + (E2) ,
O (28,23)

For N = 1 tbe equation is specialW simple !Uld corresponds to the so-called linea.r para-
metric model in which h(B) is a cubic odd function of 8. One verilies that it is still
possible to a.djust x~ at order e 2 to preserve this fonn. However, at order e 3 , which ís
also knawn for N = 1, the ílltroduction of a term proportional to 65 becomes necessary.
One linds:
{28.24)
in which ho is tbe lield llormalization constant, and b, c are gíven by:

~:')
12' c=
~:'
-íB (
((3)
J-1)
+ -4- ' (2S.25)

with:

~,·- itJ>'(1{3) = -2.3430072386. (28.26)

The collStant =<o is given by:

(28.27)
t
Remork. ln the case N > 1 the function h(8) lltill h115 a singularity on tbe coexistence
curve, due to the presenee o( Goldstone modes in the ordered phase. The nature of this
singularity can be obta.ined from the study ofthe non-lineara-model presented in Chapter
30. We shall sho-w that the behaviour of correlatíon functions below Te in a theory with
a spontaneously broken contiuuous symmetry is governed by the zero temperature IR
fíxed point. Tberefore the coexistence curve singularities can be obtained from a low
temperature expansion (for more deta!Js see Seetions 30.3,30.4).
In ali cases, as stated abave, the essentíal feature of the parametríc representatíon is
that it automa.tically satislies the different requírements about the regularity propertíffi
of the equati())l of state and leads to unifonn approximations.
The comparison with the numetícal results for the Jsing model (N = 1) and the Heisen-
berg model N 3 in three dimensions shows tlmt tbe suceessive • and <z correcti~ns
improve the mean field tbeory prediction.
From the parametric representation of the equation of state ít is also possíble to derive
a representation for the singular part of the free energy per unít volume. Setting:
F(M,t)"" n-'r,...(M,t) "'R2 -"g(6), (28.28)
one finds for g(8) a dilferential equation:
h( O) (1- rP + 21162 ) = 2(2- o)9g(8) +(I- 82 ) g'(6). (28.29)
The integration com;tant is obtained by requiring the regularity of g(9) at e 1. Jn the
same way the inven>e magnetic susceptibility is given by:
x-' = R"g2(6), (ZB.30)
with:
92(8) (1- e'+ 2PIJ2) 2,BMh(IJ) + (1- e') h'(ll). (28.31)
Tbese expressiom< can in particular be used to ®enlate some universal ratíos of ampli-
tudes.

---------------------·!1/~,VhLI·!.hk' ~'1''1·~ J
28.1 Caléulatian of Universal Quantities 629

28.1.4 Amplitude rn!io.s

Apart from critica! exponents some otber simple universal numbers have been calculated:
the ratios of amplitudes af singula.rití"" near T,. We first consíder two examples which
can be derh'<!d directly from the equation of state.
The specific heat. lf we w-rite the singular part of the specific heat, i.e. the 4>2 (x)
2~pointc.orreiation function at zero momentum, under tbe form:

t=T-To~ ±0, (28.32)

the ratío A+ fA- ís universal. Thio ratio is directly re)ated to the function g(O) defined
by equation (28.29]:
(28.33}

At arder .o2 one finds:

1: = za-•N{ 1 +c+ [3N2 ~ 26N + 100 + (4- N)(N 1){(2)

-6(5N + 22)Ç(3)- 9(4- N)>.]

2 (N~: S) 2} +O {i'), (28.34)

with:
((2) = rr 2 /6 =1.64493406684 ... ,
while >. i> defined in terms of the integral I given in (28.26):

>.=-I 12 = !.P!t/3)- ~rr =1.17195361934 ....

2

Th.e rnaynetic susceptihiiíly. The magnetic susceptibility in zero field can also be
calculated from the function g2 (8) defined by equation (28.31). As we know, below
T. the susceptibility diverges for systems with Goldstone modes. We reetrict ourselves
tberefore to N = 1. Delining:

t---+ ±O, (28.35)

one obtains:

c+ 2 (t + cb2 ){b2 -1) 1 -~

(28.36a)
c_ 11 - ~, (1 - 213)]

= 6~~·1 [t +C":1 - ((3)) ~]+o (c•) (28.36b)

The ratio C+fC- ·can be expresse<~ at order 1!2 enti.tely in terms of criticai exponents.
ThJs forro follows naturally from the pa.rametric representation of the equation of state.
The <3 relative correction i> of tbe order of only 3%.
The cnrrelation length. Let us here define tbe oorrelation lengtb in terms of the ratio
of the tvro fimt moments of the 2-point correlation function:

(28.37)

·-----------------·lflil--'~,fA<WI'·t.'l>t.'.J•'J<'I~·ni,J'I,l!.
 r
630 Calculation of Universal Quantities 2B.l

Tbe functíon ei has the scaling forrn:

0 1
< (M t) = M->•I"J ( -- ) [28.38)
'' ' € M1f(J ·

It shares otherwise ali the properties of the equa.tíon of state. It can be writt<m in
parametríc forrn:
(28.39)

At order ~ for N = 1 for exàlllple one finds:

(28.40)

Setting in zero field:

t ..... ±0, (28.41)
qua.ntity which exists only for N = l, one car1 use the determínation of 9( to calculate
this ratio:
fi/!)= 2" [1 + ~~ + 4~2 (~: + 21) ~ 21 +O (-< 3), (28.42)

in wbich the constant I is given by equation (28.26).

An addition.al uniuersol constant. Th the relation betw<!en exponents:

is associated a universal combination which involves only amplitudes of singularities when

T0 ili approached from above:

R(= ft (aA+)'/d. (28.43)

lndeed from the definitions (28.28,28.41)

(Rn• = {1- a)(2- a)t"- 2

F(O, t)t"d((1 )d = (l <>)(2- a)F(O,t){(J)d,

and the last product is normalization independent. The E-e><p81l6íon of R( ís:

(28.44)

with:

OU.er universal rotios. ]t is of course possíble to define an infiníte number of other

universal mtíos. We shall here give a few other examples which have been oonsidered in
the literature. Let us lirst define some additionol amplitudes. On the criticai ísotherm,
the correlation length beha.ves as

(28.45)
28.1 Calculation af Universal Quantitíes 631

the magnetíc susceptibilit~· as

(28.46)
the spontaneous magnetization vanishes as

M = B (-t) 0 , (28.47)

and the spin-.;pin correlation functíon in momentum spaee at T, behaves as

(28.48)

One can then define the follow:ing universal mtio:

(28.49)

whicb corresponds to the relation between exponents

lndeed using this relation we verify that Rc is proportiollBl to F(O, t)M-'x which is
normalization independent. The e-expansíons of Rc is:

(28.50)

Following Fisher and Thrko, one can construct the three following universal combinations:

Q, = CCój (BHc+)'fl' (28.51)

= Wllt) -~ c+ /CC,
2
Q2 (28.52)
Q3 = D r
(It)'-" c+, (28.53}

which carrespond to the relations 1' /)(6 - 1), tbe explicit exprnssion of ó and "f =
11(2- TJ). Moreover Q1 a.nd Q3 ro-e normallzation independent because Hx/M and p(.
respectively are. For Q2 this property !ollows immediately from the definition. Thus ali
three quantities are universal.
The quantity Q 1 is related to R, deftned by Aharony a.nd Hohenberg:

Tberr E-expansions are:

(28.54)

(28.55}

(28.56)

---------------------~·IA<WI'·t.'l>t.'.J•'J<'I~· JII,J'I,l!. 'l"tll"""""""=

G32 Cl>iculation of Universal Qturntities 28.2

The numerical results are gíven in table 28.10 and rompared with various high tempera-
ture (HT) sedes wd experimental determinations [or all qua.ntitíes except Q 3 ror whic:h
ane finds O. 96 ,.•hüe the HT series yield 0.896 ± 0.005.
lt should be mentíoned that universal ratios of amplitudes of corroctions to the leading
critica.! behavíour have also been calculated. Let us write a physical quantity for I =
T - T, small as:
!(t) (28.57)

where the correction exponent 8 (a.loo ca.lled Ll.J) is given by:

o,wv. (28.58)

The ratio of correction amplitudes af>jaf, corre;pondíng to two dílferent quantitíes fi

and fo is also universal. A [ew such ratios have been calculated. Let us give one exarnple
correspondíog to the correla.tion length and the sUSttptibility above Te:
2

at
a+
2{1 N
€
+8 -
[ 2À
3(N + 8) -
N 15N -124} 2 }
2(N + 8)3 <
3
+ 0 (e ) · (28.59)

28.2 Tbe Perlurbative Expansion at Flxed Dimension

Criticai exponents and various universal quantíties have also been calculated within tbe
2
framework delined in Section 26.1, i. e. in tne massive (4>2 ) field theory, as a. perturbative
expansion at flxed dimension. For example the RG ,!3-function in three dimensions, for
N = 1, lms the expansion:

fi<iíl = -!i+ u•- o.422496s7o7u3 + o.3stoo959789 4 o.376526B2ilií'

+ 0.49554751g6 - 0.749689g' + 0 (g 8 ), (28.60)

witb the normalization:

!i 3g/(161r). (28.61)
To calculate exponents or other universal qua.ntíties, we have first to find the IR stable
zero g" oi the function {J(g) which ls given by a few terms of a dívergent expansion. The
obvious ptoblem i.s that we have no small parameter in whlc:h to expn.nd and g• is a
number of órder 1. Alreedy at this stage a summation method is required. A further
problero arises from the property that RG functions, unlike the univeisal quo.ntitíes in
the e-expansion, depend explicitly on the renormaliza.tion scheme. On the other hand,
because one-loop diagrams have. in 3 dímensions, a simple ana.lytic expression, it h as been
possible to extend the calculation of RG functions and crítica.! exponents in the N-vector
model up to six-loop arder. Univen>a.l ratios of amplitude have a.lso been calculated.
Numerical results wíll be presented in Section 28.7.
Note however that in thís framework it is more dillicult to calculat.e phyoical quantitíes
in the ordered phase because the theory is parametrized in terms of the dlsoi<lered phase
correlation length m~ 1 wbich is singular at Te. Also the normalization of correlation
functions is singular at Te (equation (26.17)). For exa.mple the equation ofstate has the
form (for N 1):
(28.62)

~· Jll.l'l.l:'. '1'\ii ..._ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _, ,

~.u.,wt.!.hk'~'i•'t·~J
28.2 Calculatwn of Universal Q-uantities 633

in which g has to be set to its fixed point value g• and ]'.l is relate<! to the magncüzation
M b~· lhe field renormalízation (26.15):

(28.63)

At one-loop order, for N 1, the functicm f(z,g) is gíven by:

f(z,g)
z 3
qz 1
z +- + "--~.
j ri' 0
d p
+O (2 loops)
6 2 (27r) + 1 + z 2 /2
2 1

z + -r + -.---gz
"'N• fl)d/ ~
[( 1 +- d
-~z
2 z] . (28.64)
6 4smr.d/2 2 4

If we íntroduce the deviation from the criticai temperature t ~ T T<> we can rewrite
equation (28.62):

This expression is adequate for the description of the disordered phase, but ali terms in
the loopwise expa.nsion become singular when t goes to zero.
The ordered pha.se. This does not completely forbid calculations near the coexistence
curve, i.e. for t < O. Since at the fixed point g• ali functions have símple power law
smgula.rities at T,, it is possible to prooeed by analytíc continuatíon in the cornplex
t-plane. The scaling variable z,
(28.66)
picks up a phase below Te:

for t = ltle'~, z = lzle~<wll. (28.67)

The sca!ing variable H( ~t)-fl'i is then given by:

H(-t)- 00 = e'•IJ• /(z) = 1!(•)1 (28.68)

lt is in particular possible to evaluate ratios of amplitudes of síngula.ritie5 above and below

Tr: we can calculate the oomplex zero of f(:z, g) as a power series in 9 and substitute
it in other quantities. The result is complex but its modulus taken Bt 9 = g• converges
towards the correct resu!t. Let us take the example of the magnetk susceptibility for
illustration purpose. From equatian (28.62) we derive:

C+ f C_ ,-...~ !' {zo (g"} ,g") I r (O,g') =I!' (zo (g") ,g")l' (28.69)

in whicb zo is the eomplex zero of f(z,g). We thus get a. series expansíon for C+ I C-.
At lirst order in fj (defined by (28.61)) at d 3 we obtain

Adding the t"'l) terms we get C+fC ~ 3.9, a value below the best estimat.es (see table
28.10) but of the ríght order of magnitude.

----------------------·~·IA<WI'•t.'l>t.'.J•'J<'I~·ni,J'I,l1.'1"tll~
634 Calculation of Unlverslll Quantit.ies 2B.3

The main remaining technical difficulty is t!ms to extrapolate to t smalL Following

the !ines of Subsection 28.1.3, it ís natural to introduce the parametric representation
(28.20),

and considêt the function h( 8):

2
h(I!)"" (1 6 )1'' f (z(6)).
However in an expansion at fixed dimensíon, if we just replooe all quantities by their
perturbative el<pansion, the singularity of h(9) at 8 = 1 (Le. t = O) does not cancel
anymore. Tberefore a suggestion, inspire<! by results coming from the t expansion, is
to expand h(O) alro in powers of 6, keep at order gk ali terms up to order @'+2 k and
adjust the arbitrary constant xo to minimize the last. tenn. This sttateKY applied here at
order g yields a fonn (28.24) and then using equation (28.36) the value C+fC- 4.96 in
good agreement with the estimates of table 28.10. lt would be interesting to apply this
method to longer series.

28.3 The SeriO!S Summation

The principie and the theoretical justilication of the summa.tion method based on Borel
sumrnation and conformai me.pping wiU be explaíned in Section 41.6. We here want to
give ouly some practical details a.bout the application of the method e.s it lms heen used
ín particular for the calculation of criticai aponents. We shall alro illll8trate in a few
exa.mples the convergence of tbe transforme<! series.
TM. method. The general idea of the rnethod is explaíned in Section 41.6: In proctice a
variant of the Borel-Leroy transforma.tion is use<!. L..t R(z} be the function whose series
he.s to be summed (z here represents the coupling constant ii or •} We then tra.nsform
the serics:
(28.70)

ínto:

(28.71)

with:
1
u(z) = ,;r:tiiZ (28.72}
v'l + za.+ 1
The coefficierrts Bk e.re calculated by expanding in powers of z the r.b.s. of equa.tion
(28.71} and identi[ying with expansian (28.70). The constant a has been detêtmined by
the large arder behaviour analysis,

a(d =' 3) = 0.147774232 x N

9
+8, a(d = 2, N "" 1) = 0.238659217, (28.73)

and p and u are two free parameters which are adjusted empirically to improve the
convergence of the transforme<! series: the role of p and u is to weaken the singularities
of the Botei transform near z -a and z = +oo respectively. In additíon, in some cases,
a conforma.! tre.nsformation bas been made an the initial function R(z) in order to send
28.3 CalculatWn of Universal Quantitios 635

Tabl• 28.2
Serie.s aummed by the. mdhod ba.t:ed on Bo~l trnrufonnation and mappín~ for the .:ero 9"" of the
{J(Jl) functíon and lhe o:qJOnen!..• 1' and v tn lhe <J;, field lhenry.

k 2 3 4 5 6 7
g• 1.5899 1.4005 1.4498 1.4105 1.4211 1.4245
1.4908 1.4448 1.4248 1.4183 1.4170 1.4158
(1.6169 0.6312 0.62ti8 0.6306 0.6293 0.6303
" 0.5983 0.6253 (),6280 0.6292 0.6296 0.6299
1 ).2309 1.2421 1.2389 1.24]] ].2405 1.2407
1.2361 1.2383 1.2383 !.2401 1.2405 ].2406

away its dosest singula.rities, a.nd the procedure described above applied imtead to the
function R(z):
rz)l, (28.74)
in which 7 is also left as an adjustable parameter because the (ocation of ali singu!arití.,;
of R(•) is not known. This transformation is necessary in the case of the E-expansion ;
because the cdtical exponents, as functions of E, have close síngulru:ítí.,.: E= 2 for N > 1
and 1: = 3 for N = l. Needless to say, with three parameters and short initial series it be-
comes possible to find occasionally some transformed sedes whose apparent convergence
is deceptiVely good. I t io therefore essential to look for all set o f optimal values, and to
vary the para.meters in some range around the optimal valu.,. to aamine the sensitlvity
of the results upon their variations. Finally it is useful to sum independently series for
exponents related by sca.ling relations. An underestimation of the apparent errors leads
to irn:onsistent results. It is clear from ali these remarks that the errors quoted in the
final results should be considered as purely indicatíve.
2
The (4>2 ) field t.he<>ry ot fo>;ed dímen.5ions. The series for the crítica.! exponents are
known up to six·loop arder in 3 dimensions and four·laop arder in 2 dimensions. They
ha\'e been analyzed by two methods. In the first method tbe series of the RG ,B-function
has been first summed and its rero g• calcn!ated (ií = g(N +8)/(48.-) for d = 3, ii = 3gjB1f
for d "' 2). The series of the other RG function• h ave then been summed for ii §'.
Examples of convergence for two sets of the parameters p and a are given in table 28.2.
The main drawba.ck of this procedure is tha.t the values of the criticai exponents depend
strongly on the value of g•. Therefore an exror in the estimatíon of g• bia.ses ali exponents.
A variant has thus been used as a check. A pseudo-! pararneter has been introduced by
setting:
,B(g,E) =g(l-õ) +fi(§). (28.75)
=
The two functions {3(1J,c) and f3(ií) coineide for c 1, and the :rero of f3(g, e) is calculated
as a po~r reríes in <. Criticai exponents are tben also calculated as series in c, and these
series are summed. With tbe introductíon of E we avoid the problem mentioned aboVe
since we do not calculate g•. However there are some índications that the mappíng ij ,_. E
introduces singula.rities because the apparent convergence ís poorer. Some impravement
ÍB obtained by tbe introduction of the parameter r which ís necessary in tbe true €
expansion. lt is finally gratifyíng that all variants give consistent results. A compariwn
of all the results is belpful for the determination of the apparent errors.

-----------------------.,,~,fA<Wf',t.'l>t.'.J•'J<'I ~· JII,J'I,l!. 'l"tr!O:.=

636 Co.lcu!ation of Umver..al Quantities 28.4

The r-expansion, The e-exparusion has one ímp<>rtant advantage: it allcrn•s us to CO!l·
nect the resulto in 3 llJld 2 dimensions. In particular in the cases N = l and N = O. it
ís possible to compare the q,• results with exact remlts coming from lauice mo de L' .:nd
to test both univ€rsality and tho reliability of tho summation procedure. Morem-er it is
possible w imprmre the 3 dimensional results by imposing the exact 2 dimensional values
or the beha.viour near 2 dimensions for N :> 1. However since the series in ' are shorter
than the series at fi.xed dimensíon 3, the apparent errors are la.rgeL Fínal!y, "" already
emphasized, the comparison between the different resulto is a check of t.he consistency of
Field Theory methods combined with the summation procedures.

2!!.4 Numerical Estimates of Critkal Exponents

Fixed dimension. We give in table 28.3 the results obtained from summed perturbation
theory at futed dímenoon 2 for N 1 and compare them with the exact values of the Ising
model. The appa.rent errors are large because tho series are short. The agreement wlth
the Ising model is satisfu<tory. Note that g' is known only from HT series; only analytic
corrections to scallng have been found in the Ising model whích makes the identification
of w diffieult. However an anolysis b!ISed on ronfurmal inva.ríance predicts a correetion
exponent w = 4/m for 4J2m-> field theorles and for m > 3. One may conjedure that
the amplitudes of the singularitíes ínvolving the correctíon exponent w vanlsh when m
a.pproa.ches 3 for d 2, or for m = 3 when d approaches 2.

Thble 28.3
E•tim<1te.s of critiml e;q>Onellt$ in U... </4 field theo'1/.
g• ... 'f ... ~

<14 1.85± 0.10 1.3 ± G.2 1.79±0.04 0.96±0.04 0.18 ± 0.04

lsing 1.751 ± 0.005 4/3? ).75 1. 0.25

Thble 2S.4 displays the results obtained from summed pert.urbation series a.t fixed
dimension 3. The last exponent 9 wv characterizes corrections to scaling in the tem-
per ature varíable (equation (28.58)).

Table 28.4
E•limates oj critírol <:rponent> in the O(N) •ll"'metric (oj> 2 )~ fi<ld theory.

N o 1 2 3
g• ).421 ± 0.008 1.416± 0.005 1.406 ± 0.004 1.391 ± 0.004
"I J.l615 ± 0.0020 1.24()5 ± 0.0015 ).316 ± 0.0025 ).386 ± 0,0040
0.5880± 0.0015 0.6300 ± o. 0015 0.6695 ± 0.0020 0.705± 0.(){)3[)
" 0.0'27 ± 0.004 0.032 ± 0.000 0.033 ± 0.004 0.033 ± 0.004
"
{J 0.30'2 ± 0.0015 0.325 ± 0.0015 0,341)5 ± 0.0020 0.31;45 ± 0.0025
a 0.236 ± 0.0045 0.110± 0.0045 -0.007 ± 0.006 -0.115 ± 0.000
0.80± 0.04 0.79±0.03 0.78 ±0.025 0.78± 0.02
"'e 0.470:!: 0.025 0.498 ± 0.020 0,522 ± 0.0018 0.550 ± 0.0016

-----------------------!1/~,VhLI·!.hk' ~'1•'1·~ J
28.4 Calculation af Universal Quantitíes 637

Table 28.5
Críhr.al =ponent. tn the 1>~ field tht:Dfv from the <-exponsion.

1
"
., {3
"'
N o J.39 ±0.04 G.76 ± 0.03 {).21 ± 0.05 0.065 ± 0.015 ].7±[).2
Exact 1.34375 0.75 0.2083·-- 0.07ll! ... ?
N= l· 1.73 ± I}.Ofi 0.99 ±0.04 0.'26 ±0.05 0.120 ± 0.015 1.6± 0.2
lsing 1.75 l. 0.25 0.125 1.33 ... ?

The ~-expansion. ln table 28.5 we give the results ce>ming from the summed c-expa.nsion
for c 2 and compare them with exact results.
We see in this table that the agreement for N = l between field theory and the Jsíng
model is remarkable. We feel justili«d therefore in uslng a summatíon procedure of the
<-expansíon for N "" 1 wbich automatica.l\y incorporates the lsing values for E = 2. For
N =O the agreement ís good fOr v and 71.but only marginal for 1· In both c~, the
identification of w remains a problem.
Table 28.6 then displays the results for E = l.

Thbie 28.6
Critical ""''onents in th• (<I>')J ficld thoory from the E·apa..,íon.

N o I 2 3
1 ].]57± 0.003 I .'2390 ± 0.0025 1.315 ± 0.007 1.390 :± O.QlO
0.5S80 ± 0.0015 0.6.310 ± 0.0015 0.671 ± 0.005 0.710 ± 0.007
.," 0.0320 ± 0.0025 G.0375 ± 0.0025 0.()40 ± 0.003 0.040 ± 0.003
{3 0.3035 ± 0.0020 0.3270 ± 0.0015 0.34B.ó ± 0.0035 0.368 ± 0.004
0.82± 0.04 0.81 ± 0.04 0.80± 0.04 0.79± O.ll4
"'

Discu.ssion. We can now compare the two setB of results mming from the perturbation
series at fixed dimension, and the •-expansíon. First let us emphasíw that the agreement
is quite spectacular, a.lthough the appa.rent errors of tbe ê-expa.nsion are in generallarger
because the series are shorter. The most noticeable difference between these results
ooncerns the exponent 'I which is systematically la:rger in the E-expansion so that the
errar bars barely overlap. We noti<:<l in tahles 28.3,4 that a similar rema.rk applies at
d 2: the result at fixed dimension, N = 1, for 'I is sma!Jer than the result coming from
l.he E-expansion, however the latter result is closer to the Lsing value, even when one
takes ínto account the relative errors. At the same time it would seem that the values of
1 at fi.xed dimension are slightly higher than the <: values, the converse being true for v.
However the dífferences are small ce>mpared to the apparent errors.

-----------------------.,,~,fA<WI'•t.'l>t.'.J•'J<'I~·ni,J'I,l!.
638 Co.lculation of Universo.! Quantities 28.6

28.5 Comparison with Lattice Model Estimates

The N -vector with nearest-neighbour intera.ctions has been studied on va.rious lattices.
Most of the results for criticai exponents come from the rumlysis of HT series eJ<pB.JI-
sion by different types of ratío methods, Pedé approximants or differential approximants
(see Sections 41.6 and A41.3). Some results have aloo been obtalned from low tempera-
ture expansions, compu ter calculations using stochastic methods, and in \ow dimensions 1

transfer matri:x methods. Table 28.7 tries to give an idea af the agreement between lattice
and Field Theory remlts. A hístorical remark is here in arder: the agreement between
both types o! theoretical results has improved as the HT series beca.rne longer which is of
course encoura.ging. The main reaso:>n is tha.t it has become possible to ta.ke into account
in the analysis of longer series, the inftuence of conlluent singularities dueto correctians
to the leading power law behaviout, as predicted by the renorma.lization group. The
efEect of this improvement has been specially spectacular for the exponents 'Y a.nd v of
the 3D lsing model: the longer series obtained by Nickel on tbe BCC lattice has a.lmost
completely eliminated the dísturbing small dilferences whicb had remained between HT
serles a.nd RG results.

Tabl~ 28.7
Criticai exponents in lhe N ~uec!or modt:l an lhe laUice.

N o 1 2 3
.., 1.161 ± 0.002 1.2385 ± 0.0025 1.323 ± 0.002 1.396±0.003
0.592 ±. 0.003 0.6305 ± 0.0015 0.674 ± 0.002 0.711 ± ().002
"
a 0.224 ± 0.009 0.103 ± 0.006 -0.022 ± 0.006 -0:133 ± 0.006
{J 0.308 ± 0.006 0.329 ± 0.009 0.350 ± 0.005 0.369 ±. 0.005
9 0.465 0.52±0.07 0.60±0.08 0.54 ±O.lO

We h ave not gíven ali a.vailable results but rather only typíca.l results in arder to give a
feeling of the consistency between RG and latti.ee estírnat.es (for N of 1 a, {3 are abtained
by scaling). The obvious conclusion is that one observes no systematic dífferenccs. ln
particular the agreement is extremcly good in the case o f the lsing model where the HT
series are the most accurate. Th the best of our knowledge the la.ttíce models and the
2
(4>2 ) belong to the same universality class.

28.6 Criticai Exponents from Experiments

We hBve discussed the N -vector model in the ferromagnetic language, even tbougb most
of our experimental knowledge comes fram physical systems which are non-magnetic, but
belong to the universa.lity clw:;s of the N -vector model. The case N =
O describes the
statistica.l properties of long polymers, i.e. long not intersecting cbalns or seU-avoiding
walks (see Appendix A2S.l ). Tbecase N = 1 (lsing-like systems) describes liquid-va.pour
tra.nsitions in classical Huids or critica.! binary lluids. The helium superlluid transition
conesponds to N 2. Finally only for N 3 does the experimental information come
from truly ferromagnetic systems.

_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _,,~,u.,WI•!.hk' ~'1•'1·~ J
28.6 Calculation of Universal QtLantities 639

Table 28.8
Oritirol e;q»nenl$ in. jluid..

') {J () 6
b. 1.24 ±I
"
0.626 ± 0.0!0 0.325 ± 0.005 0.110 ± 0.005 0.50 ± 0.03
c. 1.23-1.25 0.625 ± 0.006 0.3!!;--0.327 a.Im-0.116 0.5(/± 0.03

Tabie 28.9
Fe'M"'magnet&c systems.

Critica! uponents and polymer8. In the cose of polymers only the exponent v is easily
a.ccessible. The best resuli.B are:

" = 0.586 ± 0.004'

in excellent agreement with the RG result.

Criticai fiuids N = 1. Table 28.8 gives " survey of the experimenta.! situation for
critica.! binary fluids (b.) lllld liquid-vapour transítíon in classieal lluids (c.). For the
binary mixtures we quot.e a weighted world &verage. ln the case of the Uquid-vapour
transition, we quote a range of experimental results rather than statistícal errors for ali
expaneni.B but v, the reason beifl!l that the values depend much on the method of analysis
o f the experimenta.! data. The agreement witb RG results ís clearly impressive.

Helium super:fluíd transítion, N 2. The he\ium transítíon allows extremely precise

measurements very close to T 0 and this explalns the accura.cy of the determinatíon of
crítica.! exponents. The order pa.rameter, however, is not dir~ly a.ccessible and therefore
only v and a have been determíned. Most recent reported values are

v 0.6705 ± 0.0006, 11 = 0.67016 ± 0.00008 and a -0.013 ± 0.003.

The agreement with RG values is quite remarkable and the a.ccura.cy of " becomes a
cballenge to field theory.

Fe~magnetic systems, N = 3. Fínally table 28.9 presents some results concerning

magnetíc systems.

-----------------------"'~·IA<WI'·t.'l>t.'.J•'J<'I~·ni,J'I,l!.
640 Calwlatian of Unit>erslli Quant•ties

Te.ble 28.10
Amplitude rt:~.tio:J in binary crittcal fiui.~.

c-ex:pa.nsi.on Fixed dim. d ~ 3 HT seri"' Ex:J)€rlrnem

A+ /A- 0.52 ± 0.02 0.:>41 ± 0.014 0.523 ± 0.009 0.5&-063
c+;c- 4.82 ±0.05 4.77 ±0.30 4.95 ±0.015 4.3 ± 0.3
I: !fi 1.91 2.04 ± 0.04 1.96 ± 0.01 1.9± 0.2
0.266 ± 0.001
Rt 0.2!l 0.270 ± 0.001 0.25-0.32
R. 0.066 0.0594 ± O.OOll 0.0581 ± 0.0010 004-!l.09
RtK'I' 0.667 0.692 ± o. 007 0.650 0.6(1--0.80
R, 1.6±0.1 1.7 1.75 1.75 ± 0.30
Q, 1.13 1.21 ± 0.04 1.1 ± 0.3

28.7 Amplitude Ratios

Table 28.10 conta.íns 8 comparison of amplitude ratjos as obta.ined from renormalization

group, lattice ca.lculations and experíments on bínary mixtures for N = 1. RG values
are in general much less a.ccurate than in the case of exponents because the series are
shon.er.
Some results are avaílable for tbe liquid-vapour transitíon. In particular:

R• = 1.65 ± 0.2 •

An experiment gives for A+ I A_ the v alue 0.53. The set of results shaws, wíth indeed
large erro.s, a satísf&tory agrEement with RG predícti<>ns.
Fina.lly let us give a few results concerning ratios of amplitudes of corrections to th<l
leadíng scaling behaviour. If in a.ddition to the correlation length and the susceptibility
amplítudes a( and ax we ronsider W..O the specilic hE'.at amplitude ac and tbe coexistence
curve magnetization amplitude nu, we can form three independent ra.tios. The re!mlts
for N = 1 are given in table 28.11.

Thble 28.11
C01Te<;tion amplitude ro.tio• for N 1.

• expansíon Fixed dim. d "' 3 HT series Experiment

atJtÇ 0.56± 0.15 0.65± 0.05 0.70± 0.03
atfai 2.03 1.4.5±0.11
a:!: foi ].02 0.94 ± 0.10 1.96 0.87 ±0.13
at:fac 2.:>4 0.96 ±0.25 0.7-1.35
tlj_Ja; 0.3 0.315 ± 0.013
atfa.M 1.10±0.25 1.85 ± 0.10
aufat 0.90 ± 0.21 0.0!1-1.4

~· Jlt.J•t.l:'. 'F\ r _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _,,

~.u.,wt.!.hk'~'i•'t·~J
28.7 Calculation of Universal Quantitíes 641

'Iable 28.12
Amplitttd• ralios for N = 2 and N ~ 3.

N Field theory HT series ExperJment

A+ /A- 1.029 ± 0.013 1.08 { 1.067 ± 0.003

2
1.088 ± 0.007
n-c
( 2 0.36 0.36
x• /A- 3 !.52± 0.02 !.52 1.4CH.52
R-c( 3 0.42 0.42 0.45

Systerns wilh N > 1. A few amplitude mtios have been calculated and measured for
Helium (N = 2) and ferromagnets (N 3). We give in table 28.12 the exarnples of
A+ /A- and Rt-
If one takes into acoount ali data (criticai exponents, amplitude r3tlos, ... ) one is
foroed to conclude th~<t tbe RG predictions are remarkably oonsistent with the wbole
experimental inforrnation ~<vailable. Consiaering the variety of experimental sítuations,
thls is a spectacula.r confirmation of the RG ideas and the concept of universality.

Bibliographital Notes
The E-expansion calculations were inítíated by
K.G. Wilson and M.E. Fishe:r, Phys. Rev. Lett. 28 (1972) 240.
Tbe two first terrns for 1 and '1 were obta.ined in
K.G. Wilson, Phys. &v. Lett. 28 (1972) 548.
The next arder was calculated by
E. Brézin, J.C. Le Guíllou, J. Zinn-Justin and B.G. Nickel, Phys. Lett. 44A (1973) 227.
The equation of state was derived at arder E2 by
E. Brézin, D.J, Walla.ce and K.G. Wilson, Phy•. Rev. Lett. 29 (1972) 591; Phys. Rev.
B7 (1973) 232.
For a discussion of the equation of state see also the review of
D.J. Walla.ce, in Pha.e Tmn.sitions and Critical Phenomena vai. 6,
quoted in Cbapte:r 24, and
D.J. Wallace and R.P.K. Zia., J. Phys. O Solid State Phys. 7 (1974) 3480.
The para.rnetric representation has been introduced in
P. Schofield, J.D. Litster a.nd J.T. Ho, Phys. Rt':l!. LeU. 23 (1969) 1098; B.D. Josephson,
J. Phys. C: So!id State Phys. 2 (1969) 1113.
Calculations of ratios of amplitudes have been reported in
E. Brézin, J.C. Le Guillou and J. Zinn-Justin, Phys. Lett. 47A (1974) 285; H.B. Tarko
and M.E. Físher, Phys. Rev. Lett. 31 (926) 1973; Phys. Rev. Bll (1975) 1217; M.E.
Fisher a.nd H.B. Tarko, Phys. Rl!tl. Bl1 (1975) 1131; A. Aharony and P.C. Hohenberg,
Phys. Rev. B13 (1976) 3081; Physit:a. 86-88B (1977) 611; Y. Okabe and K. ldeura,
Prog. Theor. Phys. 66 (1981) 1959; J.F. Nicoll and P.C. Albright, Phys. Rev. BJ1
(1985) 4576; C. Bervíllier, Phys. Rev. B34 (1986) 8141; G. MüDSter; J. Heitger, Nud.
Phys. B424IFS] (1994) 582.
Exponents at arder ! 5 haw been obtaíned by
A.A. Vladimirov, D.l. Kazakov andO. V. Tarasov, Zh. Eksp. Thor. Fiz. 71 (1979) 1035
(Sov. Phys. JETP 50 (1979) 521); K.G. Chetyrkin, A.L. Kataev and F.V. Tkac:hov,

----------------------·~·IA<Wf',t.'l>t.'.J•'J<'I~·ni,J'I,l!.'l"tr~
642 Calcu!ation of Universal Quantities

Phys. Lett. B99 (1981) 147; B101 (1981) 457(E); K.G. Chetyrkin and F.V. Tl<achov,
28.7
r
Nucl. Phys. Bl92 (1981) 159; K.G. Chetyrkín, S.G. Goríshny, S.A. La.rín and F.V.
Tl<achov, Phys. L€tt. 132B (1983} 351; D.I. Kazakov, Phys. Lett. 133B (1983) 406; S.G.
Gorillhny, S.A. Larin and F.V. Tkachov, Phys. Lett. lOJA {1984) 120; H. Kleíaert, J.
Neu, V. Sr.hult.,._Frohlinde, K.G. Chetyrkin lllld S.A. Larin, Phya. Lett. B272 (1991)
:19, Erratum B3HI {1993) 54&.
The use of perturbation series at fixed dimension has been suggested by
G. Parisi, Caryêse Lectures 1979, published in J. Stat. Phys. 23 (1980) 49.
Tbe calculation of the serias expansion for the RG functioru; has been inítiated by Níckel
and reported in
G.A. Baker, B.G. Niclrel, M.S. Green and D.l. Meiron, Phys. Rev. Lett. 36 (1976) 1351;
G.A. Baker, H.G. Nickel, and D.l. Meiron, Phys. Rev. Bl7 (1978) 1365.
ln these artiçles estím&tes of critica.l exponents are also given. The more accurate esti-
mates based on Borel summation and large arder behaviour ana.lysi> presented in Section
28.4 are talren from
J.C. Le Guillou and J. Zinn-Ju•tin, Phys. Reli. Lett. 39 (1977) 95; Phye. Rev. B21
(1980) 3976.
Recently the series have been published for generie values of N
S.A. Antonenko and A.l. Sokolov, Phys. Re~. E51 (1995) 1894.
Ratios o! amplitude are estimated in
C. Bagnuls and C. Bervillier, Phys. Retl. B24 (1981) 1226; J. Phys. A: Math. Gen. 19
(1986) L85; C. Bagnuls, C. Bervillier, D.I. Meiron and B.G. Nickel, Phys. Rev. B35
(1987) 3585; R. Schloms and V. Dohm, Phy•. Rev. B42 (1990) 6142; H. J. Krause, R,
Schloms and V. Dohm, Z. Physik B79 {1990} 287.
Results obtaíned by summation of the ~-expa.nsion are found in
J.C. Le Guillou and J. Zínn-Justín, J. Physiqtle Lett. (Paris) 46 (1985) L137; J.
Physique (Paris) 48 (1987) 19; 50 (1989) 1365; C. Bervülier, Phys. Reu. B34 (1986)
8141.
A comparison between experíment, hlgh temperature series and RG field theory calcu-
lation using tbe results available in 1980 ean be found in the Proceedings of the Cargese
Summer In:stitute 1980
Phase 'Il-unsítians vol. B72, M. Lévy, J .C. Le GuiUou and J. Zinn-Justin eds. (Plenum,
New York 1982).
For a revi<lw on amplitude ratios see
V. Privman, P.C. Hobenberg and A. Abarony, Phose 'II-ansitio..., and Critica.! Phe-
nomenavol. 14, C. Domb and J. Lebowitz eds. (Academic Press, New York 1991).
High temperature sedes are discussed in
A.J. GuttmWID, Phase Tmnsitioru and Oríticol Phenomena vai. 13, C. Domh and J.
Lebowi~ eds. (Academic Presa, New York 1989); A.J. Liu and M.E. Fisher, Physica
Al56 (1989) 35,
and recent resultz for the lsing model publíshed in
A.J. Guttmann, J. Phys. A: Ma.th. Gen. 20 (1987) 1855; A.J. Guttmann 8Jld I.G. Ent-
!ng, J. Phys. A: Math. Gen. 26 (1993) 807, 27 (1994) 8007; G. Bhanot, M. Creutz, U.
Glii.sBner and K. Schilling, Phys. Re~. B49 (1994) 12909.
New hlgh temperature results for N ~ 2 have been reported in
P. Butera and M. Comi, NeiJI extended high tempemture seri.., for the N vector moàels
on 8D bipanít.. lattil>es, Mila.no preprint.
The detamination of the exponent v for polymers is taken from
J.P. Cotton, J. Physique Lett. (Paris) 4.1 {1980) L231.

"'==-------------------....,,~,u.,wl·!.hk' ~,1•'1·~ J
28.7 CalCTJ..Io.tion of Universal Quantities 643

Results from Monte Carlo simulatioru; for Ising model BDd of the SAW are found in
G.S. Pawley, R.H. Swendsen, D.J. Wa.llace aDd K.G. Wüson, Phys. Rev. B29 (1984)
4030; C.F. Baillie, R. Gupta, K.A. Hawick and G.S. Pawley, Phys. Rev. B45 (1992)
10438; D.C. Rapaport, J. af PhyBics A: Ma!h. Gen. 18 (1985) 113; N. Madras and
A.D. Sokal, J. Stat. Phy•. 50 (1988) 109.
Experimenta.! determínations of universal quantitíes fOr binary fluíds are reparted in
D. Beyseru;, A. Bourgou and P. Calmettes, Phys. Rcu. A26 {1982) 3589; A. Kumar,
H.R. Krisbnrunurthy and E.S.R. Gopa.l, Phys. Rep. 98 (1983) 57.
For recent experimental results on He!ium superfluid transition see
J.A. Lipa and T.C.P. Cbiu, Phys. Rcu. Lett. 51 (1983) 2291; L.S. Goldner and G.
Ablers, Phys. Ret1. B45 (1992) 13129; J. Adriaa.ns, D.R. SWBn5()n 8Ild J .A. Lipa, Physicn
B 194-196 (1994) 733.
Tbe relation between sto.tistica.l prajrerti"' of polymers BDd tbe N = O ( .p~) ~ field theory
h"" been found by
P.G. de Gennes, Phys. Lett. 3SA (1972) 339.
We have followed the derivation of
V.J. Emery, Phys. Rcu. Bll (1975) 239; B. Duplantier, C.R. Acad. Sei. Paria 290B
(1980) 199. '
The applico.tion of RG methods to polymers is presented in
J. des Cloizeo.ux, J. Physique {Pam) 36 (1975) 281; L. Schãfer and T.A. Witten, J.
Chem. Phys. 66 (1977) 2121; J. Physíque (Paria) 41 (1980) 459.
For a comprehens.ive discussion of botb experimental and theoretical aspects of renor-
malisation group and polymer physics see
J. des Cloizeaux and G. Jannink, w Polymeres en Solulion (Les Edítioll!l de Physique,
PILfÍfl 1987), English version Po!ymers in Solulion: Model!in9 and Structure, (Oxford
Uoiv. Press 1990).
The exact two-dimensional resulto have lirst been conjectured by
B. Nienhuis, Phys. &v. Lett. 49 (1982) 1062.
For a discussion of the rela.tion between cl118Sical fluida at criticality and the q,4 fie!d
theory, see for example
F.W. Wiegel, Phys. Rep. l6C (1975) 57.
The lattice--gss model w"" introduced in
C.N. Ya.ng and T.D. Lee, Phys. Rev. 87 (1952) 404; T.D. Lee and C.N. Yang, Phys.
Rev. 87 (1952) 410.
Exercises
Eurcise !!.8.1
To the readeT wbo linds tbe relatíon between the polymer mode! and the tP• field theory
!JOmewhat mysterious the followíng exercise is propceed: Ca.lculate at one or two-loop
arder z!2 l(k, t) directly by expa.oding the meá.S\lre {A28.1) ín power$ of g. Apart the
trick which bllS already been used in the gaussian case the following remark is useful:
Write a Fourier representation for the chain self-intero.ction term:
4
64 (r(u 1) r(u2)) =f (2")
d q.,exp[iq· [' d1<(6{u1 u) 9(u2 u))r(u)]
Jo
where 8{u) is the Heaviside step function. To recover the Feynmo.n diagrams of the tP•
lield theory it is suggested to integrate OVEf r(u), then over a.l! u variables o.nd fino.Jly to
Laplace transform, keeping the q-íntegrations wbich come from the Fourier represento.tíon
for the end. Some additional inspíration can be found in Section 31.1.

---------------------~·IA<WI'·t.'l>t.'.J•'J<'I~·ni,J'I,l!.
644 Calculatian of Un'ive~sal Quantitte• A28.1

APPENOIX 28 2
NON-MAGNETIC STATISTICAL MOOELS ANO THE (4>2 ) FIELO THEORY

A28.1 Statlstics of Self-Repellíng Chains, Approximations

We here show that, as anticipated in Chapter 28, the statíatice.l properties of long polymer
2
cha.ins can be derived from the criticai behaviour of the O(N) symmetric {4> 2 ) field
theory in the N : O Jimit. More precisely Wl! study the statistical prúpertíes of long
chains with repulsive contart self-interaction. A discretized form of tbe model is provided
by the self-avoiding ra.ndorn walk (SAW) on a lattice.
We cell u tbe distance along the chain, r (u) tbe position in space of the point on the
chain of para.meter u. We tben characterize the chain by a probability distribution fur a
chaín r (u) of total length s:

[dp(r(u))! = [dr (u)Jexp { _ [~f.' r 2 (u)du +~L du,du, 6d (r (u,) _r (u,))]}.

(A28.1)
The specie.l case g = Ocorresp<>nds to the brownian cho.ín ar gaussian r andam walk, as bas
been discussed in Cbapter 4. The self-avoíding random walk (SAW) model also provideo
a lattice regularízation: at a microscopic scale A-J the chain beoomes much stiffer tba.n
wbat is implied by expression (A28.1). Note iinally tha.t any short rBllge potential would
yield, at lea.ding arder fur loog chains, to the swne results as the ó-function interaction,
(as ca.n be VEriiied by repeating in the more general case the àeriva.tion of an equiva.Jent
<P' lílre iield theory, anà using the analysis of corrections to scaling w; in the case of
criticai phenomena).
A genernting junclian. VarioU' geometríc cha.mcteristic properties of the chain ca.n be
obt.&ined from the 2-point function G(1) (k, s):

(A28.2)

The brar.b!ts mean aVErage with respect to the distribution ( A28.1 ). Indeed i f we exp&nd
G12 l (k,s) in p<>wers of k a.nd use the rotational inva.ria.nce <>f distribution (A28.1) we get:

2 2 2
G(l) (k, s) 1 - -1! -d
k ( (r (s) r (0)) 2 ) (k -
+- 1 - ( (r (s) -r (0)) 4 ) + ... ,
) --
2 4! d(d+2)
(A28.3)
lt is actually convenient to ronsíder the Laplace tra.nsform of G(7 ) (k, .s) with respect to
s:
zPl (k, t) = 1""' .-•• c<•J {k, s) às. (A28.4)

lndeed if we know the 2-point correlatíon function z<•l (k,t), we obtaín c< 7l (k, s} by
invertíng the Laplace transforrna.tion:

.2..
2iwJc
1 e'' z<•J (k) t) ctt ' (A28.5)

in wbich the rontour C is a par aliei to the ímagina.ry t a.xis at the right of all singula.ritíes
of Z( 3) (k,t). llamember that, sínce we are going to expanà in powers of k (see equatian
A2B.l Calct.latíon of Universal Quantities 645

I
lmt
I
I
•
I
le I Ret
I
I
cl
I
I
I
Fig. 28.1

(A'28.3)), k has to be considered as Infinitesimal. Tbus only tbe location of singularities

In the limit k -. O ís relevant. Let us caU te the first singularíty one meets while trying to
displace the contour to the left (figure 28.1}. On C the real part Re(t) oft thus satisfies:

Re (!) > Re (te),

lt is easy to verifythat the behaviour of zPl(k, t) near thesingularity governs the lalge
posltive s behaviour of the integral ( A28.5) and thus of the moments of r (s) - r (0).

A28.1.l Special e:wmplu. Flory's appnnimation

The gaussian random walk. In expression (A'28.2) we can su bstítute

r(s) - r(O) = f r(u)dtt.

Wethen get

G('l (k,s) j [dr (u)]exp{- f du [ir-• (u)- ik ·r(u)}}.

Shiftíng r we find:
c<•l {k,s) =e-···' (A28.6)
whlc:h implies as it ís well-known for the gaussia.n raudom walk:

(A'28.7)

The Laplace trans[ormation then yields:

(A28.8)

Th.e <>ne dimensional chain. lf the c:hain self-repulsion plays a role, it must increase
the average apatia! extens.ion of tbe c:hain for s la:rge, compared to tbe gaussian ca.se.
This is obvious in the one-dimensional case. The c:hain is then completely stretched:

r(s) = ±s, {A28.9)

---------------------~·IA<WI'·t.'l>t.'.J•'J<'I ~· JII,J'I,l!. 'l"tr"""""'=

646 Clllcutation of Universlll Quantíties A28.1

and therefore:
G' 2l (k, s) cos ks
(A28.10)
{ (2) )--t-
z (k, t - t2 + k'-

The one dimeDSiona.l example shows that the self--avoiding condition r.an r.hange the sta-
tistkal propertíes of the random walk. lt also provides an upperbound on the moments.
A sensihle guess is then in general

(!r(s)-r(O})"") -•""" fors~oot

in whir.h the exponent 11 is bounded by 1/2, the gaussie.n r.ha.in, on one side, e.nd by 1,
the stretr.hed r.ha.ín on the other side.

The upper-t:ritical dimension. Note tbat the self--avoidíng condition becomes weaker
when the dimension d of embedding space inaeases. Aetu..Uy a simple argument sug-
s-s, as in tbe case of the criticai behaviour, the existenee of an "upper-critícal" di-
mension. !f we consider a very long r.hllin, we can ronsid"" tbe elfect of self-avoídíng as
the infiuenoe of one cha.ín onto anotber. A browuian r.ha.ín has Hausdorf dimension two.
Therefore ab<:Ml dimension four (two +two) two r.ha.íns do no more Bee each other, self-
avoidíng should no longer play 8 role e.nd tbe m!IÍD should have the statistical properties
of 8 brownian r.haín.
Beyond the yaus..-ian model: Flll1'1J'& approrimation. Before solvíng the prablem by
more systematic metbods let us first describe 8 simple approximate solution. The ap-
proximation is baBed on e.n energy balance a.rgument. For s la.rge, the g8ussia.n term
should scale "" tbe ínteraction:

(f r2(u)du) ~ .2•-1' (A2S.ll)

j du du,6
1
4
(r(u1 ) - r(u 2 )) -
2
8 -.W_ (A28.12)

Balancing botb terms we obtain the equation for v:

2v-l (A28.13)

wlíd for 1 :5 d :5 4. This expression yields for d = 1 the exa.ct wlue 11 1- It predicts
tbat for d ?:: 4 tbe intera.ction can never balance the ga.ussian tenn, since 11 ís bounded by
lf2. More surprisingly, for d =- 2 it yields 11 3f4 wbir.h is also known, from conformai
fie!d tbeory a.rguments, to be tbe exa.ct value.
We sh..U discuss below other dimensioll5. Flory's approximation relíes on intuitive
argument.z and it ís dífficult to see bow ít can be irnprnved. Actually the values it yields
for the expooent 11 are so good that one ma.y wonder if tbís result is not exact. We
show below that it is iodeed ooly an approximation. For this purpose it is convenient to
completely reformulate the thoory.
 A28.1 Calculation oj Universal Quantíties 647

A28.1.2 Equivalence with the q,• field theorg

We firEt transform the ínterBCtion term in equation (A28.1} by writing it as a gaussi.an
integral over a.n auxiliary field cr (r}. li r(u) is a given cha.in, then

f !da(r)]exp [~f ddrcr 2 (r)- f duo(r(u))]

= exp [-~f du1du26d (r (u 1}- r (u,))] , (A28.14)

where we have used:

f duu(r(u)) f ddrcr(r) f duó"(r(u)-r).

Integration over irnaginary o-fields ls sssumed. After this suhstitution we identify the
r(u) measure of integration with the integmnd in the path integral representation of the
evolution operator in imaginary time u of a à-dimensional QUIIJitUm mechanieal system
with potentíal o(r) (see Chapter 2). The 2,point functlon z< 2l (k, t) Clm thUB be wrítten:

z<2l (k,t) =f !d<T(r)jexp [~f ddru2 (r)] f e-•· ds

x f ddrd9e•l.(r-r')(r'le-•Hir), (A28.15)

in w hlch H is the quantwn hamiltonia.n:

(A28.16)

The Laplace transform Cll.ll be explicítly evaluated:

z<•J (k,t) =f Ido (r)Jexp [~f ddro (rl] f ddrd"r'e•l•-•'l

2

x (r'l(-~+t+u)-'lr) (A28.17)

We now apply a variant of the so-called repli<:a trick:

}}!!;0 f !d<t> (r)] 1>1 (r) (b,(r') exp [-~f ddr ( (â,.tf>) 2 + ttf>7 +"(r) 4>>)]
(r'l(-~+l+o)- 1 1r), (A28.18)

in whlch N is the numbu of romponents of the field tf> (z). lndeed the gaussian integral
over the li.eld tf> (r) yielda the <N> propagator divided by a factor !dat(-~ + t + o)JNI•
wbích goes to 1 in the "unphysícal" N = O llmit.
Substitutíng this identity into expressían ( A28.17) we ootain:

z!2l (k, t) =f [d.p (r)j IÓI (k) </>1 ( -k) f [do (r)!
x exp [! ddr (~,.>- ~oq.•-} 2
(âp.p) - ~ltf>2)] . (A28.19)

=·=-------------------,~·IA<WI'•t.'l>t.'.J•'J<'I ~·
IU'I.l:'. '1'1lli-- JII,J'I,l!. 'l"tli~
648 Calculation of Unfuersal Quantíties A28.l

We ca.n then integmte aver "'

z<•l(k,t) f [d.PJ<Pdk)4>,(-k)exv[-S\4>JJ,

s (4>) f ddT [ t l
~ (8~4>) 2 + ~t</>2 + Wl (A28.20)

Thi.s i.s a most remlll"kable result: tbe statistical pruperties of polymers are related to tbe
2
properties of the ( .p 2 ) lield theory in tbe N O limit as fust notimd by de Gennes.

A28.1.S RG Approach to SAW and Statistics of Polymers

Tbe ]arge scale statistical properties of tbe chfÚD are related to the singularities af corre-
2
lation functions of the ( .p2 ) field theory, which alise in the massless ar criticallimit. The
short dista.nce stiHness of tbe chain provides an UV cut-ofi /1. for the theory. Near te, in
which t< is the criticai temperature or critical bare mass, the 2-point function z< 2l(k, t)
has a scaling behaviour:

(A28.21)

2
in whlch v and '1 are the N = O limits of tbe crltical exponents of the (.p2) field thoory.
Using expression (A2S.21) in integral (A28.5) we tben obtain a so::aling form for the
large 8 bebaviour of am(s,k):

a<•l (s, k) ~ e••· •• -l g (lkl •"}. (A28.22)

Expa.nding in powers of k 2 , we fínd a first term:

iJIStead of 1 whlch sbows tbat tbe dístribution is not c:orrectly normalw.d. The fimt
term is actually tbe ratio of tbe number of config.no.tions af the SAW to tbe number of
config.natíoDS of the gaussían random ws.lk. Therefare the properly normalw.d averoge
(exp [ik (r (s) -r (O)}J) has a scaling beha.viom entirely cha.racterized by the exponent v.
For eJ<ample:
((r (s} r(0)) 2 ) ~ _.2v. (A28.23)

The exponent can be obta.ined from the ~ -expansion:

V= ~ (1 + kE + ~& 2 + 0 3
(E )} . (A2B.24.)

lnstead Flory's formula (A28.13) predicts at the same order;

v(Flory) =! (1+ !H0(&2)), (A28.25)

Comparing the two expressioru;, we ore immediately that Flory's formula ís not exact in
general, altbough, flB mentioned before, ít rorrectly predicts tbe upper-critlcal dimansion
d = 4, and the values for d 1 and 2.
For d = 3 RG calculatiozn; yield:

v = 0.5880 ± 0.0015'
A28.2 Calculaticm af Univeraal Quc.ntities 649

result which ca.n be compaxed with 11(Flory) 0.6. The two values are v.ory dose although
definitlvely dilferent. Actua!ly it is easy to estimate tbe dilference approxímately. !f we
simply take into account the orders ' and c2 and the property that Flory's approximation
is exact for E = O, 2, 3 we can write:

For < I we find successively 11 = 0.590 and order € and 11 = 0.592 at arder <:2 ,
results coru;istent with the RG value abtained by summing ali terrns (tables 28.4,28.6).
Experimental results, 11 0;586 ± 0.004, HT series and Monte Carla simulations, 11 =
0.0>92 ± O.OG3, favour the RG predíction compared to Flory's approximation.
Mareover the renormalization group makes many additional pred idions and al!ows the
calculation of many universal quantíties which cannat be obtained ftom Flory's argument.

A28.2 Liquid-Vapour Phase Transition and Field Theory

That the liquid-vapour phase transition is the same universality clll8S as the Isíng model
and can he described hy the r/>4 field theory is not oompletely obvious a priori.. In
particular no apparent díscrete symmetry is broken. There exist several methods to
establísb a co~mectíon hetween tbe liquid-vapour transition and the N = 1 q,4 theory:
one can either start ftom a model which iE physically not v.ory realistic, the lattice gas
model, but rigorously equlvalent to the Ising model, or from a true gas moda! in the
continuum, and by a numher of approximations for which it iB difficult to provide a
rigorous justlfica.tion, obtain directly a 4> 4 fi.Jd theory.
The lattice gas modeL One assumes that particles are living on a lattke and that
at eacb síte i the particle occupatian number no can only be O or I. The lattíce iB
a scbematic way to repre.ent a hard core type interaction. A longer range attractive
potential ÍB represented by" nea.rest neighbour interaction which ravours the accupation
of neighbour sites. One finally adds a chemical potentíal term, i.e. a source term coupled
to the number of partícles:

-{:ní(n;) Jl::n;ni-1'2:>•· (A28.26)

t'l.n.

Tbis model is trivially related to the lsíng model by th€ transformatíon:

S; 2n;- I, (A2B.27)

where S; is an Ising spin. One finds in d dimensions:

(A28.28)

We recognize the energy of the Isíng model witb n.n. interaction and magnetic field.
A gas model in lhe rontinuum. The introduction of a lattice iE somewhat artificial. A
more direct method, although ít involves a few steps tbat are bard to justify rigorously,
is the follow:ing: one considera a real classical gas in the continuum and assumes that
the potential hetween particles is the sum of a shart range ha.rd core potentíal and a

----------------------,~·IA<WI'•t.'l>t.'.J•'J<'I~·ni,J'I,l1.'1"tr."""""=""
650 Calct<lation of Universal Qttantities A28.2

longer range attràCtive two-body potentia.l V{r). The canonical panition function for n
partides of mess m in a volume n then has the form:

Z(n,{i,fl) r;)] , (A28.29)

where a is the size of the hard core. In three dimensioos a convenlent choice for the
potential ís for ex!llilple:
V(r) = -Je-P• /4'1fr. (A28.l0)
The important hypotheses are that tbe potential depends only on the dista.nce and is
sbort range in such a way that its Fourier transform ís regular at zero mamentum.
Furthermore its Fourier transform is assumed ta have no zero. We can then define the
inverse K (r- r') of V (r- r') in the sense of kernels:

f ddr"K (r- r") V (r" -r')= 6d (r- r'). (A28.31)

In the caseof the Yukawa potentíal (A28.30), K is thedilferentia.l operator J-l (&- p~).
Using the simple identity:

exp [ -!f3 f.1 V (r; -ri)] = /tdifl(r))

x exp (!11- 1
J
ddr d"r'.p(r) K {r- r') ti> (r')+ ~rl>(r•)1, (A28.32)

we rewrlte the partition function:

Z(n,(3,0) .,n.trV(0)/2 f!ci,t.(r)] exp [ ~p-1 f ddr ddr'<? (r) K (r r') q, (r')]
x Zh.(n,(3,0,tJ>), (A28.33)

where Zh.o.(n, (3, 0,1))) is the canonicol partition function of particleo intera<Otíng through
a hard core two-body potential in a ta.ndom externai potential q, (r). The fact that in
example ( A28.30) V (O) diverges ís irrelevant sínce we can modify tbe potentia.l for r < a.
We can traJJBform equation (A28.33) int.o an identity between grand partition functions:

Z(z,(3,n) = 2: ~Z(n,(3,0),
n.
(A28.34)
"
where we have called z the fugadty. Tbe partitíon function of the hard core potentíal in
presence of a given externa.! one-body poteotial cannot be calculated exactly. However a
few rema.rks ca.n be made: since we are ínterested in distances much larger than tbe olze
a of the ha.rd core we ca.n make a local expansion of lo Zb.<. (<P)· Wben q, is repulsive the
bard core ca.n for example be neg)ected hecause the hard core then preventa the collo.pse
of the ~. The partition functioo can thus be represented approximately hy the
functiona.l integral:

Z(a,/3,0)
A28.2 Calculation of Uní11ersal Quantítíes 651

where <'> is a renormalized fugacity e.nd

U(q,) =-o j ddre<~><•l +0 (a').

This expres,;ion has to be understood with an implicit short distance cut-off a. We have
therefore been able to write the partition function apprmdmately as a functional Integral
with an efiective local action S(q,). In the ci!SE of tbe potential (A28.30) for example we
lind:
(A28.36)

Let us discuss this action at tree levei following the !ines of Section 24.3 where the
mean field appruxímation was considered. First we can shíft q, ín sucb a way that tbe
minimum of the potential oc:curs at r/! O. At high temperature the action has only one
homogeneous minimum. By varying the temperature we can reacb a situation where 8
second minimum appears and eventually the two mínima bocome degenerate. Howe""r
here, since the action has no rellection symmetry, the generic sítuation corresponds to a
first arder tr&n5ltlon. It ls only when, by "'<lryillg 8 second thermodynarnlc varlable líke
the chemlcal potential, one has been able to cancel the ~3 term in the expansion of the
action for c/1 small that one obtaíns a second arder phase transitlon. But then the tenns
odd in r/! are ali at least of dimension 5 for d 4 and therefore irrelevant nt the phase
transition. This explains that the 1/>4 field theory cnn descrihe correctly the liquid-vapour
pha.se transition. The only specific fentures of this trnnsition with rmpect to magnetic
systems are:
(i) The complica.ted, although regul8l, relntions of the thermodynarnic variables to the
more natural, from the symmetry point of view, va.riables magnetic field 811d temperatura
of the ma.gnetic systems. This introduces addition.o.l par81I1eters in the description of
experimental results.
(ii) Tbe appearance of corrections to scaling due to operators of odd C8llonical dimen-
sions like rjJ 5 .

----------------------~·IA<WI'·t.'l>t.'.J•'J<'I ~· JII,J'I,l!. 'l"tr_~

~9 THE (4>2) 2 FIELO THEORY IN THE LARGE N LIMIT

In tbe preceding chapters WlO have derived the universal properties of critícal systems
witbin the Etamework of the formal c: = 4 d expMSíou. It is therdore reassuring
to vmify in a particular model that the re!iults obtained in this way rernaín valid even
whan e ;,; no longer infinitesimal. We show ín this chapter that, in tbe case of tbe
2
( <f>2 ) O( N) syrnmetrlc lield theory, the same universal propertie!i C8.ll also be derived
at lixed dimension in the large N limit, and more genero.lly order by order in the la.rge
N -expa.nsion.

29.1 Introduction

We consider the pa.rtition function in an extern.ollield or source J:

Z (J) =f {d<f>(x)}exp [-S(<f>) +f d xJ(x)- <f>(x)],4

(29.1)

wbere S(<f>) is a O(N) symmetric action:

S (<f>)= f {i {il,.<f>(x)]
2
+ ~r<f>2 (r} + ~ 2 2
[<f> (rJ] } d"x. (29.2)

A cut-off A, consistent with the •ynunetry, is implied.

The basic idea o f the IB.rge N ""pansion iB the SBJne as mean field theory: for N large,
the O(N) invarla.nt qna.ntities self-sverage and therefore hsve smo.ll Buctnations. Thus
forexample
(<t>'(:z:)<f>'(y)) N::oo
(<f>'(x)) (4>2(y)) ·

Therefore one should try to take <f> 2 (x)""

a dynamical variable. Tecbnico.lly, in the case
2
of the (<,11 2 ) tbeory, tbis can be achieved by using an ideiJtity similar to the Hubbard
transfonnation:

exp {-f 4
d x [~r<,11 2 (x) + ~ (4>2}21} o:: f [d:.jexp U 4
d :r (fu>-'- ~À- ~>.<t>') 1•
(29.3)
where the integration contour is parallel to the imaginary axis. This identity allaws ns to
rewrite the interactíon term in the functíonal integral (29.1). The new functional integral
is then gaussiaD in <f> and the integral over the field <f> can be performed. Tbe dependente
on N of the partltion function becomeo explicit. Actuo.lly it is convenient to separate the
components of <,11 into one component u, and N - 1 components ..-, and integrate oruy
over.,. (fur T < T. it may even be convenient to integrate over only N- 2 compom•nts).
For N IB.rge the di11erence is negligíble. We co.ll H(x) the source fur a(:z:), and set all
other components of J(x) to zelO:

Z(H) =f {d>.(x)J[du(x)]exp (-sdf (>.,11) +f ddx H(z)u{x)), (29.4)

---------------------,~.u.,WI·!.hk' ~'1•'1·~ J
 2
29.2 The (.p 2 ) Field Theory ín thç Larye N Limit 653

witb:

We now take the l!iJ"ge N lirnit at Nu fuced. With thiz; condition s.fl is oi order N and
the functional integral ca.n be calculated for N large by steepest descent. We expect
cr =- O(N 112 ), À~ 0(1).
Til€ À-field mrrelation functions. In this formalism it is natural to a.lso calculare À-field
corre!ation functlons. These have a sirnple interpretation in t.he initial q)-field fonnalism.
lndeed let UB add a source j~ for À in the adiou (29.5). Then reintroducing the q)-field
and integrating over À we recover ínstead of adiou (29.2),

S(</1)- [u/6].p 2 j~ + (u/6)j~- rj~. (29.6)

Thereforej;. generates the .p2 correlatíon functions, up to a multipllcative factor and a

tramlation o f the connected 2· point function.

29.2 Tbe Large N Limit

We look for a uniform saddle point {cr(z), À(.z) space-independent},

cr(x) =a, À(.z)=À.

Differentiating then a.ctlon (29.5) witb respect to cr and À -..e obtain tbe saddle point
equations for N la.r ge:

:Au O, (29.7.-.)
n 6
2
1
---(À-r)+--
j" -ddp
-=0. (29.7b)
N Nu (2,f p2 + .l.

Rem4rk. In t.be large N lÍlllit the dornino.nt contributiol\5 in perturbation theory are
given by cbaill.'l of "bubble" diagrams of the form displayed in figure 29.1. In thís limit
tbey form a geometrical sedes whicb is summed by tbe a.lgebraic tecbniques explained
above.

Fig. 21!.1 The d<Jmiruml dlagra.JnS in the large N Umít.

---------------------~·IA<WI'·t.'l>t.'.J•'J<'I ~· JII,J'I,l!. 'l"tr~

 2
654 The (.p2 ) Fiold Theory in the Larye N Limit 29.2

29.lU The crítica/ domain

The low tempernture phru;e. Below T., a. the avernge value of tbe field, does not va.nish.
Equation (29.7a) then hnplies >.""O and equation (29.7b) yields:

6
--r (29.8)
Nu
Note that this equatíon has a solution only for d > 2. Thll; ís a manifestation of the
Mermin-Waguet--Coleillilil theorem: in a system with only short range forces a continu-
ous symmet.Jy cannot be broken for d ~ 2, in tbe sense that tbe a~rage a of the order
para.meter necessarily vaníshes. This result is also consistent with the analysis of Section
23.5. PhysícaUy the would-be Goldstone modes are responsible for this property: heing
mossless, as we kDOW from general arguments andas the propa.gator in the r.b.B. of (29.8)
confirms, they induce an IR instability for d ~ 2.
Equation (29.8) has a solution u provided:

(29.9)

The criticai value r e of r is thus:

(29.10}

Setting:
(29.11)
we can rewríte equation (29.8):

(12 = -T ~ ( -T)2f3 =? {j _, .!.. (29.12)

2

For N large the exponent {3 remalns classical, i.e. mea.n-field like, in all dimensions.
The high lempemture phase. Ab<Jve Te, "varushes. lo expression (29.5) we see that
the u-propagator then becomes
1
(29.13)
A.= P'+ >._
Therefore >. 1 / 2 ís at thís order the ph}'Sical mass, i.e. the inverse correlstion length ç- 1
of the field o. We thus bom now on set

(29.14)

Using equations (29.10,29.11) in equation (29.7b) we find:

6
-;; + (2lr)d
N j" p2 (p"ddp+ m2J = m2.,. • (29.15)

(i) Fbr d > 4 the integral in (29.15) has " límit for m = O and therefore at lea.ding
arder:
m 2
= ç- 2
~ T and thus v = ~, (29.16}
 2
29.2 The (4>2 ) Fieid Theory in !h.e Larye N Limit 655

which is the meM field result.

(ii) For 2 < d < 4 inste...:!, (setting d = 4 .<) the integral behaves for m smalllike:

1 !A ddp C(d)m-•- a(d)A _,+O (m'-•A-2 ), (29.17)

(2")<1 r(p2+m2}

with
(29.18)
The constMt a(d) which chamcterizes the lea.ding correction in equation (29.17), depends
on the cut-olf procedure.
The Lh.s. of equation {29.15) is thus dominated by the most singular contribution of
the int<>grn.l and at leading arder we obtain:
m ~-1 _ _,.l/(2-<l, (29.19)

whicb shows that the exponent 11 is not dassical:

L I
JJ=--=--· (29.20)
2-e d-2
(iii) For d = 4 the l.h.s. is still dominated by tbe integral:

1
(2")4
j"' r (]ild + m2) m::::. OS:rr21n(A/m).
4
p I

The correlation Jength has no longer a powet law behaviour but inste...:! il. mea.11-field
behaviour modifie<l by a logo:rithm. This is typical of a. situation where the ga.ussian
fu:ed point is stable, in the presence of a marginal opemtor.
(iv) If we examine equation (29. 7b) for O" =O and d = 2 we find the correlation length
can hecome large only for r _, -oo. This peculiar situation wiU be examíne<l when we
discuss the non-linear cr-model.
FiM.!ly, in the criticallimit r O, >. vanishe$ and thus from the form (29.13) of the
u-propagator we find that the critica! exponent '1 remo.ins cla.ssical for ali d

JI=O. (29.21)

We verify that the exponents {3, v, '1 satisfy tbe scn.ling relatlon (25.64).
TM scaling equation oj $tate. In a magnetic field H the saddle poínt equation (29.7a)
is modifie<l:
m 2 u=H. (29.22)
Using equations (29.10,29.11) we can rewríte equation (29.7b):

6 2 N !A d 4p
-m +--m2 (29.23)
u (2rr)d r <r+ m 2 )
Let us consider only the case 2 < d < 4. From equation (29.17) we knaw the behaviour
ofthe l.h.s. for m srnall. Keeping only theleading terrn,""' can rewTite equation (29.23):

(29.24)

---------------------~·IA<WI'·t.'l>t.'.J•'J<'I ~· JII,J'I,l!. 'l"t.ll""""=

 2
656 The ( .p2 ) Field Tlteary in the Larye N Limit 29.2

Eüminatíng m b~n equations (29.22) and (29.24) we fin!!lly obtain s.n equation of
ste.te in sc!!ling fonn (a is the magneti:z:.ation):

in which h{, is a normamation constant, 6 is given by:

é=d+2 (29.26)
d-2,

in a.greement "ith the scaling relation (26.65), 8Dd the function j(;z;) by:

(29.27)

The asymptotic forro of !(:r) for o: large implies "f= 2/(d- 2) agaín in agreement with
the scalin,g relation"'' = v(2- 'I}· Tl>lting into acrount the values of the critica.l exponent.s
"'' and {J it is then easy to verify that the function f sailifie5 ali required properties like
for example Griffith's 8lla.lyticity (see Section 26.5). In particular the equation of state
can be cast into the pe.rametric fotm of Subaection 28.1.3:

a= R''2e,
T = 3R (1 92).
H = %R6/29 (3- 2112)2/(d-2) .

The specijic heat oxponent. Di!Jerentiating twice action (29.5) with respect to .l.(:z:),
replacing then the 6eld À(x) by its expectation vlliue m 2 , wo. find the .\-propagator 6.1.[p)
above T.

~A [p) = -N
2 ( 6 1
Nu+ (211/
j" (q2 + m') (lP- q)
ddq
2 + m2 ]
] -· (29.28)

The .\-propage.tor iB negative hecause the >.-field is ima.ginary. }u; note<! in 29.1, it is
equ!!l to the 4>~ 2-point function up to 8D sdditive constant. At :oero momentum we thus
obta.in the specific heat. The r.h.s. iB dominat.ed for m small by t.he integral

r ddq
(q2+m2J'l ~(
df2 l)C(d)m-•,

where C(d) is defined in (29.18). The singular pe.rt of the specific heat thus vanishes as
m•. From the RG 8lla.lysis we expect m-af•, Usíng equation (29.20) wo. verify

-afv = 4- d := d- 2Jv,
in agreement with the general relation (26.49).
In the critíc!!l theory {m = O at thís order) for 2 :Ç d :Ç 4 the denominator is ah;o
dominated at law momentum by tbe integral
 2
29.2 The (q,2 ) Fie!d Theory in lhe La1ye N Limit 657

with
r 2 (d/21)
(29.29)
d/ 2l r(d -2)
and thus:
(29.30)

We agaín verify ronsistency with sca.ling relations. In particular we note that in the large
N limit the dimension of the field >. ís P•J = t(d -e)= 2, a result important for the 1/N
perturbation theory.
Rernarks.
(i) For d = 4 the behe.viour of the propagator is still domínated by the integral which
has a logarithnúc beha.viour Ll.l ex 1/ln(A/p).
(üj Note therefore that for d ::; 4 the contributlons generated by the term proporti<>nal
to À (x) in (29.5) always are negligible in the criticai domaín.

29.1!.2 Leadf.ng corrections to scaling and RG ftmctions

Let us again consider equation (29.15), bpt now keep also the lellding correction to the
integral for m small (equation (2!U 7}). The consta.nt a(d) depends on the specific cut-<>lf
procedure, but ,fpr E sma.ll satisfies: '
1
a(d) ~ 81f2E' (29.31)

Equation (29.15) then leads to:

m 1 cx -r 1 1l•~l [r+ const.(6/Nu- a(d)A-')t•f(>-•l + · · ·]· (29.32)

According to our previous analysis of corrections to scaling we identífy:

e
wv-= 2--; or w == .€. (29.33)

Within the framework of the é-expansíon w is assocíated wíth the leading oorrection to
scaling. In the large N limít w remains smaller than 2 for e < 2, extending this property
to ali dimensions 2 $ d ::; 4. Furthermore we notíce that for t.he special va.lue u = u":

u· = ;~~) = A•4B~'e +O (E2}' (29.34)

the lellding correction to the scaling relation (29.19) cancels, together actua.lly with ali
other conectíons decaying like powers proportlona.l to e.
Remark. For simpie regularizat.íons or quite generally for d dose to four, a( d) is positíve
and thus u· ís posítive as it should. However thís is not alwa.ys the case and it then seems
that the lellding oorrectlon to scallng cannot be ca.ncelled in this way. It ís nat knawn
whether a similar possibility also exists for N finite.
The >.-propagator. In the same way if we keep the leading correctlon to the ),..
propagator in the criticai tbeory we find:

(29.35)
658 TM {.p2 / Fóeld ThwTY in the Large N Limit 29.2

where we have neglect.ed terms of arder 11- 2 and 1/N. The leo.ding correctíons to scaling
again exactly cancel for u ~ u• ""expected.
The RG functioru. St!tting:

u ~ gA', ~ g• =u·A-• = 6/(Na), and T=ut, (29.36)

we calculate tbe RG functions. The form (29.13) of the u 2-poínt function shows that
the field ~ not renormalized at leo.ding arder aJld thus

z. =1 ~ fl(g}=O, (29.37)

result consistent wíth the value (29.21) of the exponmt 71·

The criticai renormalized u 4-point functíon at renormalization scale 1.1 ~ then gíven
by
-3ll.~(m 0,1')-
Differentla.ting with respect to A ai flxed renormalized parBmeters we obtain the {:J-
functlon ( equation (25.42))
{}(g) = -õg(l- gfg"). (29.38)
It is importaJlt to note that ín (29.35) the relevant term was the first correction to scaling.
We then calculate the a 2 ou renormalized oorrelation function at renormalizatíon scale
1.1 to obtain Z. and thus '12· From the remark (29.6) we find

-Z2 ~6,~(1')
u
=I, =* Z2ri•l(p; <X p) =gA',
and therefore ( equation (25.67))

f!z(9) = A.i_\ lnZ2 = _{J(g} -~=-E.!..

dA\,.•. fixed g g•

lt follows
.,- 1 (g) 2 + 1)2(g) = 2- cgjg•, (ZI.39}
Replacing the coupling col>Stant by íts IR fixed point value g• ""' find n::sults obviously
consistent wÍth the equations (29.20,29.33,29.34).
It is then easy to verit'y that m 2 in equation (29.15) is a RG invariant, i.e. satisfies:

a a
(11 811 + /3(9) âg -
8]
'12 (g) !{i m(t,g, A) O. (29.40)

Remarks.
(i) Again an IR fixed poínt g• e:xists only if a(d) is positive. Otherwise tbe RG metbod
fails, at least in tlús simple form.
(íí) In Chapter 27 ""' have argued that a. perturbation due to irrelevant operatol'8, is
equivalent, e.t leading order in the criticai region, to a modilication of </1 4 couplíng. Thi5
can be explicitly verified here. The amplitude of the leading oorrectíon to scaling bas
beeu found to be proportional to 6/Nu-a(d)A -• where tbe value of a(d) depends on the
cut-olf procedure a.nd thus of contributions of irrelevant operators. Let us call u' the rf}
29.3 659

coupling constant in annther schewe where a ls replaced by a'. Identifying the leading
correction to sca.ling we lind the relation:

6A' 6A'
Nu a(d) = Nu' a'(d),

homographic relation whlch i5 consístent with the specia.l form (29.38) of the P..function.
The "'nortnalizeá coupling c:an.stants. Let u.s calculate the renorma.lized couplíng con-
stant.s both in t.he massive theory reoorma.lized at zero momentum and the massless
theory renormlllli:ed at a symmetric poiut delined by (Pt + P2)2 = J.<1 • Keeping the
leading correction to scaling we find respectiVely

1 N
g,(O) = l2(d + (1g _
7)C(d)
91.) (~A)'
1 N (! 1) (~-')'
g,(l•) 6b(E) + 9 - g. A .
We recover that for generic values of g the renorma.lized coupling constants are fuced
at their IR fuced poínt value, up to corrections to the leading scaling behaviour. They
significantly deviate from the fuced point value, which t.hen provides an upper-bound,
only for non generic small values of g.

29.3 The l/N Expansion

Higher order terms in the steepest descem calculfltínn of the functional integral {29.4)
generate "systematic 1/N expansiQn. Let us first slightly rewrite action (29.5). We shift
tbe field >.(r) by its expectation value m 2 (equatíon (29.14)), .!-(.:) ~ m 2 + .!-(.:):

SeH(G,À) = ~j ddz [(&,..,) 2 + m2a 2 + >.(.:)112(.:)- ~>. 2 (.:)- ~ (m2 r)>.(.:)]

+ (N l)ttln(-6.+m2 +.>.(.:)]. (29.41)

2

Puwer countíng. Neglecting all irrelevant terms, we ce.n ana.lym the action (29.41)
from the point o[ víew of power counting. The dímension of the field cr(.:) is (d 2)/2.
From t.he critica.! bchaviaur (29.30) of the À-propagator- bave deduced the engineering
dimension (>.] of tbe lield >.(.:):

2{>.]-E d i.e. [>.] ""2.

The interaction term f >.(z)11 2(x)ddx hflS thus dimension ...ro. Jt is easy to verify tbat
tbe non-local intera.ctions involving tbe >.-field conling from tbe expatll<ion of the tr In
have a.ll the canonical dimension zero:

'I'IUs power counting property bas the following implicatioo: In contta.st with usual per-
turbation tbeory, tbe 1/N expa.nsion generates only log8J"ithmlc corrections to tbe leading

----------------------·~·IA<WI'·t.'l>t.'.J•'J<'I~·ni,J'I,l!.'l"tJl"""""""'
660 29.3

Jong distance bebaviour for a.ny lixed dimension d, 2 < d :::: 4. The sítuation is thus sim-
ilar to the situation one encounters for the t-expansion and one expect<; t.o be able to
calculate universal quantities like critica! exponents for example as power serie5 in 1/N.
However, be<:au~e the interl>dioru; are non-local, the results of standard renormalization
theory do not immediately a.pply. We devote Subseetion 29.3.2 to a thoraugh discussion
of this problem.

The non-linear o -model. We have already noticed that the term proportional to
f d~:r À2 (x), which ha.s dirnension
4 d for large N in ali dimensions, is irrelevant in
the critica! domrún for d < 4 and can thus be omitted at leading order (this Biso applies
to d = 4 where it is marginal but yields only logarithmic correetions). Actually the
constant part in the inverse propagator as written in equation (29.35) plays the role of a
large momentum cut-off. Let us thus oonsider the oct.ion (29.41) without tbe .\ 2 tenn. If
we then work backwards, reintroduce the initial field q, and integrate over .\(:r) we find

{29.42)

Under thts form we recognize the partítion function of the O(N) symmetric non-linea.r
17-rnodel m an unconventional normalization. We have therefore discovered a rem&rkable
correspondence: to ali orden; in an lfN exp8.1lSion the renormalized non-lineat u-model iE
2
identíelll to the renormalized (.p2 ) field theory f1t the IR lixed point. In Section 30.7 we
shail solve the <T-model in the latge N limit to show more explicitly the rela.tion between
the different pararnetrizations.

Anályticity in the coupling con•tant and the larye momentum apa11S"àon. We have
seen that, in the fre.mework at the lfN expansion, we can calculate at fuled dimension
d < 4 in the criticai limit (T"" T.,m 2 O). This implies that the terms of the 1/N
expMsion cannot ~ expanded in a power series of the coupling ronsta.nt, at least with
integer powers. Note that since the gaussian !ixed point is an UV !ixed point, the sma!l
coupling expansion is also a large momentum expans.ion. To understand tbe phenomenoo
we consider tbe 17 2-point function at order 1/N. At this order only ane diagram hi'I.S <,
2
to be evaluated conta.ining two ).. 17 V<>Itices. In the large cut,..o:IJ limit and after mass
renormalization we find:

{2) .:..
r a.(p)- p +
1 2
N(2lr) 4
f ddq
(6/Nu} + b(t)q E
( 1
(p + q) 2 - ?"1 ) +o ( N12 ) •
(29.43)

lf we study the integrai we find tbat ít has an expansion of the form

L O"ku*p•-h + {.huiH2k)/•p-l<. (29.44)

k

The terms with integer powers of u ca.n be obtained by expanding the integral and
calcula.ting each term by dimension regula.riza.tion. "" h"" poles at € = (2! + 2)/k for
which the corresponding pawer of p 2 is -1, i.e. an integer. One verlfies that fJ1 has a pole
at the same valm• of E and that the singular contributions cancel in the sum.
 2
29.3 The ( rl>2 ) Field Th.wry ín the Larye N Limit 661

29.3.1 The exponent ') at order 1/N

For illustration purpose let us calculate tbe exponent 'f} at order 1/N. We CWI obtaín it
from the u 2-point function at lfN order, at Te and at the IR fixed point. Expression
(29.43) OO.:omes:

(29.45)

As explained above, after mass renormalization the q-integral is loga.rithmically divergent

in any dimension. We assume a cut-off A whose precise form is irrelevant a.t this arder.
The integral then behaves for large cut-off or small momentum like:

(29.46)

Comparing with tbe expa.nsion of the 2-point function at the IR fuced point:

(29.47)

we conclude:
71 = 2Aj(Nb(c)).

Expression {29.46) shows that the coefficient of lnp can be obtained by evaluating the
divergent pa.rt of the integral. The latter contribution corresponds to large values of the
integration momentum q and is thus regular in p. We can thus expand it for p small. We
find:

{29.48)

Usíng rotational invariance we can substitute:

The divergent pa.rt of the integral is then:

(29.49)

The value of the exponent 'I'} foUows. It is convenient to set:

2r(d- 2) 4sin('II'E /2)r(2- ~)

(29.50)
r(dJ2)r(2- d/2)r 2 (d/2- I) ~ 11T(l- •/2)r(2- '"/2) ·

Then:
(29.51)

---------------------~·IA<WI'·t.'l>t.'.J•'J<'I~· JII,J'I,l!. 'l"tr_~

662 Tke (4>') ~ Field Theory in the Larye N Limit 29.3

29.9.~ An altentative criticai theory

1b be able to use tbe sta.ndard results of renonnaliz.atíon theory we reformulate the
criticai theory to dea.l with the non-local ínteractions. Omltting for d < 4 tbe irrel.,vant
term proportional to f >.2 (x)ddx In the effective a.ction (29.41), a.nd reintroducíng the
field ,P(x), we can write the functional integral:

Z =f [d.\(x)] {d.p(:r)] ""P !-S (,P,>.)], (29.52)

S (.P .\)
•
jdd:r: (!ta
2 ~ 2
2
2
2 2
.p) + !m .p + !>.,p 2
- ~'U (m 2
r) _x) . (29.53)

We now replace the action by a modílied a.ction S, obtained by &dding a tenn quadre.tic
in >. wbicb at tree levEI yields a>.- propagator of the form (29.30):

where we have set c= 3(m2 - r)/u. In the limit 11 _. oo the initíal action ís recovered.
However, now standard lield thoory method• ca.n be applied. It ís conwnient to rescale
À into v-\. The renorma.lized criticai a.ction then reads:

ISw]..,. = ! f 2
d 0.r [ Z0 (8,...p } + 6m2.p2 + v,Z.>..p2 +À( -&)-</l À 2c,.À] . (29.55)

Note that in generie díroeMions the .\-field ís not renorroa.lized because counterterms are
local wbile tbe >. inveme propagator ís non-loca.l. This situation ís a.lso encountered in
systenu; wíth long range forces. It fo\lcws that the RG equatioi>S for the critica.l thoory
take tbe fonn:
(29.56)

We ce.n then ca.lculate the RG functlon5 as power series in 1 f N. It ls r.asy to verify that
v 2 bos to be taken of ordet lfN. Therefore to generate a lfN expansion one has first
to surn the multiple insertions of on&loop À 2-polnt function contributiom whicb form
a geometrical series. Frorn the calculation of the </></> and .pq,>. correlation functions at
order 1/N, we derive

(29.57)

(29.58)

with
D(v) = 2/b(r) N1i.
The solution to the RG cquations (29.56) can be written:
r!l.~l( TP, v, A) = z-n/2(T)'Td-21-n[d-7)/2r(•.nl (p, v{r). A),

with the usual definítions

dv2
"r-
d.,-
[j(v(r)), T~"''l(v(r)).
dT
 2
29.3 The ( 4>2 ) Fie/d Thenry in _the Larye N Límít 663

We are interested in the neighbourhood of the fixed point v2

oo. Then the RG fonction
ry( v) appraaches the exponent 1) obtained by direct ca.lculation. The 8ow equation ior
tbe coupiing constant bernme<:

with

Thus
v2 (T) ~ r•,
We then note that the ÀÀ 2-point fnnction is proportional to v• for v large

lt follows tbat, altbough the field À is not renormalized, still the powe.r law is modífied.
Weftnd

with
Q

11
-<
+
4(2- e)(3 - e) X
{4-t)N I+ 0
(2..)
N2
.
29. 9. 9 Other raults
The calculations beyond the arder 1/N are ratber teclmical. The rearon ís easy to
understand: Becauae the e!fective 6eld theory ís renormalizable in ali dimensions 2 $
d :S 4, the dimensional regularization wbkh was so UBeful for practical calculations no
longe. works. Therefore either one keeps a t.rue cut-olf or one introdu(:(JS new more
sophisticated regula.riza.tion scbemes. FOr detaili; tbe reader ís referred to the literature.
Generic dimensio!L!. The exponents 1 a.nd '7 are known up to arder ljN 2 a.nd l/N 3
respecti>'cly in arbitrary dimensions but the expr,.,ions a.re toa complica.ted to be repro-
duced here. We give inste...:! the expa.nsioru of ')" and v only up to arder 1/N, which can
be directly dedured frorc the results of the preceding subsectíons:

(:2'0.60)

(29.61)

The exponents w and 8 "'", governing the leading correctioiiS to sca.!ing, ba.ve also
bcen calculated:

w ~ (1- ~ia_-c~~x~) +o(~.), (29.62)

(} = wv 2~€ ( 1 2{3; c) x,) +o ul) . (29.63)

Note that tbe exponents are regular functions of e up to e = 2 and free of renormalon
singulariti.., at c= O (see Section 42.6).

----------------------~·IA<WI'·t.'l>t.'.J•'J<'I~·ni,J'I,l!.
 2
664 The (.p2 ) Field Thcory ín lhe Larye N Límít 29.4
The oquation of state and too spín-<>pin correlation function in 7.ero lield are alro
known at order 1/N, but since the expressions ure complicated we refer the reader to the
literature for details.
1'hret dimeruional remlta. Let us give the expansíoo of 'I in thnoe dímensions at the
order presently availa.ble:

I
wíth l

1/J(x) heing the logarithmic derivative of the r fnnction.

The exponent 1' ís known only up I<> arder ljN2:

Note that the lfN expansion seems to be rapidly divergent a.nd certainly a direct surn-
matíon of these terms does not provide very good estimates of critica! exponents in 3
dimensions for useful values of N.

29.4 Generallzations

We now briefly explain how the algebraic method presented in Section 29.1 can be gen-
eralize<! to actions which have a more complicated <kpendence in one or several ve<:tor
fieids. ln Chapter 24 we have used a.n ídentíty similar to the identity (29.3) .t.o oonstruct
the mean field expansion in the specia! case of two-body !nteroclions. We bave shown In
Appendix A24 how this method can be generalized. Ao analogous strategy alro works
here. In the case of"' general O(N) symmetric field thwry, we íntroduce pairs of fields
and Lagr~~.nge multiplien! for ali !ndependent O(N) invariant scala.r products constructed
from the many-component fields.
Let m first take the exarnple of one field .p and a5Sllme that tbe interaction ís an
arbitrary function of the only invariant f/>2(:r)

(29.64)

We then introduce two fields p(x) and ,\(x) and use the identity:

exp [-f dd:rV(f/>')] IX f [dp(x)d,\(x}]exp {-f ddx [~>.(,p 2 - p) + V(p))}.

(29.65)
In the special Cl\!le in which V (p) is 11. qusdratie functlon, the integral over p can be pe<·
[ormed. In ali cases, however, the identity (29.65) transforms the action into a qusdratic
form in 4> o.nd therefure the integr ation ave r 4> can be performed and the dependence in
N becomes explicit. The effective action is then:

(29.66)

~· Jlt.J•t.l:'. 'F\11•'==--------------------~.u.,wt·!.hk• ~'i•'t·~ J

 2
29.4 The ( c/>2 ) Field Theory in the Large N Limit 665

and the ssddle point equations become:

(29.67)

The solutions of these equatians can be studíed in the same spírit as equation {29. 7b).
I f the action is H.l1 O( N) invariant function of two fields 4> 1 and c/> 2 the patentíal
depends on the three scalar products q, 1 • q,,, </>~ and </>ª· Then three pain;.of fields are
requíred.

Bibliographical Notes
As shown by Stanley the large N -limit of tbe cla.ssical N -vector model coincides with the
spherlcal model solved by Berlln and Kac
T.H. Berlin and M. Kac, Phys. &v. 86 (1952) 821; H.E. Stanley, Phys. Rev. 176 (1968)
718.
Early work on calculatíng critical properties includes
R. Abe, Prog. Tbeor. Phys. 48 (1972) 1414; 49 (1973) 113, 1074, 1877; S.K. Ma, Phys.
Rev. Lett. 29 (1972) 1311; Phys. Rev. A7 (1973) 2172; M. Suzuki, Phys. Lett.42A (1972)
5; Prog. Theor. Phys. 49 (1973) 424, 1106, 1440; R.A. Ferrei and D.J. Scalapina, Pnys.
Rev. Letl. 29 (1972) 413; K.G. Wílson, Phy~. Rev. D7 (1973) 2911.
The equation of state to arder 1/N is derived in
E. Brézin and D.J. Wallace, Phys. Reu. B7 {1973) 1967.
The spiiHõpin correlation in zero lield is obtained in
M.E. Físher 8lld A. llhamny, Phys. Rev. Lett. 31 (1973) 1238; A. Ahamny, Phys. Rev.
BlO (1974) 2834; R. Abe and S. Hikami, Prog. Theor. Phy~. 51 (1974) 1041.
Tbe exponent w has been calculated to arder 1/N in
S.K. M&, Phys. Reu. AlO (1974) 1818.
See also the rontributions af S.K. Ma andE. Bré2in, J.C. Le Guillau and J. Zinn-Justin
to
Pha.e 1hmsitions and Critica! Pnenomen<~ vol. 6, C. Domb Md M.S. Green eds. (Aca-
demic Press, London 1976).
Tbe consistency of the 1/N expansion to ali orders has been proven in
I. Ya Aref'eva, E.R. Nissimov and S.J. Paclleva, Commun. Math. Phys. 71 (1980) 213;
A.N. VBSil'ev and M.Yu. Nalimov, Teor. Mat. Fiz. 55 (1983} 163.
At present the longest lf N series for eo<ponents and amplitudes are found in
I. Kondar and T. 'Thmesvari, J. Physique Lett. (Paris) 39 (1978) 199; Y. Okabe and
M. Oku, Prog. Thenr. Phys. 60 (1978) 1277, 1287; 61 (1979) 443; A.N. v...,il'ev, Yu.M.
Pís'mak and Yu.R. Honkonen, Teor. Mat. Fu. 46 (1981) 157; 50 (1982) 195.
See also
L Kondor, T. Temesva.ri a.nd L. Herenyi, Phys. Rev. B22 (1980) 1451.
Renormalizatíon of operators ís discussed in
K. La.ng and W. Rühl, Nucl. Phys. 8400 (1993) 597; Z. Phys. C6l (1994) 459.
The caoe of long range for= has been discussed in
S.K. Ma, Phys. RetJ. A 7 (1973) 2172.

IU'I.l:'. 'L'<l'i,~.---------------------~·IA<WI'·t.'l>t.'.J•'J<'I~· JII,J'I,l1. 'l"tl'il!.=

 Th• (4>2 )
2
Field Theory in lhe Large N Limit 29.4
666

Exerdses
Exercise 29.1
Discuss the logaríthmic corroctions to mean field theory at the upper-critical dimension
d "" 4 in the large N expressions of Chapter 29. More specificaJly, consider the correlation
length, the equation of ~»a.te a.nd the spedlic hea.t.

E:tercise 29.f!
2
Use the expressions ofSection 29.4 to generalize the previons study to a potential V(4> )
and look for 1> trictitical point.

'
i'
l

tt,

~· Jl .l't.l:'. '1'\11"-=----------------------
--- ~.u.,Ln·!.hk'~'i•'t·~J
30 FERROMAGNETIC ORDER AT LOW TEMPERATURE:
THE NON-LINEAR o--MODEL

In Chapters 24-26 we have studied lhe criticai properties of Ja.ttice models described
in terms of N -vector spin variables S, of unit length on each la.ttice site i, interocting
through a short range ferromagnetic O(N) symmetric two-body interaction V.J- In zero
field the partition function of such models can be written:

Z =f f} dS,6 (S~- 1) exp[-1i (S) JT], (30.1)

in which the configuration energy 1i is:

1í. (S) JT =-; Z::V.jSi · SífT. (30.2)

'l

To derive tbe critic.al properties oftbe N-vector model we bave proceeded in tbe following
way: we have first found an appraximation for tbe partition function, tbe mean field
approximation, valid in high dimeru;ions. We have characterized the criticai properties
of ali physical quantities within this approximation scheme. We h ave then shown that
the mean ficld approx.imation was the first term in a systematic expansion. Examining
the first correction to mean lield we ha"" discovered tbe role of the dimension four,
dimension at which the validity of mean field theory ceases. We ha'"' finally explained
how a summation of the le!>ding IR diwrgene"" to ali orders in the expansion leads to
an effective 1/14 field thoory whas<:: critica! properti"" can be analysed by RG methods in
4- e dímension, i.e. near the so.c.alled upper-critical dimension. The results one derives
in this way have been presented in Chapters 25, 26.
Here, inste!>d, we discuss these models from the point of view of a low temperature
expansion. The methods we employ, however, apply only to continuou$ symmetries,
here to N ~ 2. They rely on the property that mode\5 wíth oontinuous symmetries, in
contrast to models with discrete symmetríes, have a non·triviollong distance physics at
any t.emperature below T,, due to tbe massless Goldstone modes.
We first prove universal properties of the Iow temperature, ardere<!, phase at fixed
temperature. Then, in the non-abelían case, N > 2, we show tbat !>dditional information
about crítica! propertíes can be obtalned, by analyzing the ínsta.bílity o f the ordered phase
a.t low temperature and near two dimensioiJS, due to Goldstone mode interactions. The
spedol role of two dimensions, the lower-critical dimension, in models wíth continuou•
symmetríes has olre!>dy been stressed in Chapter 23.
The analysis is based on the foUowing observation: The N·vector model (30.1,30.2) can
be considered as a lattice regularization of the non·linear u-modcl introduced in Chapter
14. The low temperature expa.nsíon of the lattire model is tbe perturbative expansion
of the regularize<! field throry (Subsection 14.3.2). Tbe formal properties of the lield
theory have been díscussed in Cbapter 14. We have expla.ined that it i• renormalízable
in dimension two. RG equations, va!id in two and more genero.lly 2 +" dimension follow.
Tbeir solutions will help us to understand the Jong distance bebaviour of correlation
functions.

------------------------"'~·IA<WI'·t.'l>t.'.J•'J<'I~· JII,J'I,l!. 'l"tJlr,.,-.

658 Fermmagnetíc Order at Low Temptroture 30.1

1t js somewhat surprising that two different contínuum field tbeories, tbe {q'>2 ) 2
and the
non-linear O"-modei de.cribe the iong distance physía; of the same lattke model. Actually
we have o.lready noted in Section 29.3, that witbin the framework oftbe 1/N-expansioo
both fieid theories ......, directly "'lated. This property, totally mysterious at the classical
levei, emphasizes t.he essential nature of qU8Jltum !luctuations.
In tbe s.ppendix we briefiy discuss a model of ..,)f.inkra.cting fermions, the G ross-Neveu
moda!, wbich shares some of these properties.
Before we begin our general discussion, let us here conlirm by a different approach
that at fixed temperature T < T., the non-linear <7·model also emerges directly in the
2
analysis of the large distance behaviour in the ordered phase ofthe [(q,2 )] field tbeory.

2
30.1 Tbe (q>2 (:r)] Field Theory at Low Temperatura

At low temperature, i.e. T fixed, T < T., in a system in wbich a discrete symmetJy is
sp<>ntaoeously broken t.he connected correlation functioDS decrease exponentialiy. The
situa.tinn is quite di!lerent, as we ha.ve a.lready noted, wben t.he symmetry is continuous
2
because Goidstone modes a.re present. Let us conaider t.he (1,62 ) field theocy:

(30.3)

in the low wmperature phase {r< r.). We ebange va.ria.bles in the funct.ional int;,gral:

Z = f!dt,D]exp(-S(q>}j,

setting
I,D(x) = p(x),j.(x) with ,j.2 (x) = 1. (30.4)
Tbe functional integral becomes:

{30.5)

with:

In the ordered phase, below T., the field p(:t) has a non-zero expecta.tion value and
is massive; its dyna.mica is therefore not critíCBl. Tbe integration over the p(x} field
generates 8Jl effective local action s.w( ,f,) for the field J.:

Moreover the functional integral (30. 7) can be calculated perturbatively. Let us call M
the expectation value of p{:t) at the tree levei:

p(x) M + p'(x). (30.8)

30.2 Ferromagnetíc Order at Lçw Thmperotttr< 669

In terms of p' the action (30.6) reads:

S (l.J.) = j dd:r{ ~ (M 2
+2Mp' +p'") [a"J,(:rl]' +~ [B,.p'(x)] 2

+lr(M + p') 2 + ~u(M

4.
4
+ p') } .

Neglecting ali fiuctuations oftbe field p', we obtain B.ff a.t leading arder:

(30.9)

We recognize the cl.assico.l actiou of the non-liuear <1-model studied in Chapter 14.
Loop correctíoru; coming ftom the inte~ation over p' renormalize tbe coeflicient M 2
in (30.9). They a.lso generate additiono.l ~intera.ctions. However, 88 longas we explore
momenta much smo.ller th8JI the ~m""" or distances much larger tba.n the corresponding
correlation length, the effective action resulting ftom the integration over the ~field can
be expanded in local terms. The term dotninant at large distance iB the term witb only
two derivatives. Due to tbe O( N) symmetry it iB proportionol to the leading arder effec-
tive a.ctlon s!~ (J.)
and renormafu.es only tbe coeflicient M 2 which appeo.rs in equation
(30.9). The other intera.ctions have four deriva.tives or more Blld cormspond to irrelevant
opera.tors. Note tba.t Cor a. tempera1ure T dose to T., tbe domain in momentum space
in which th- a.rguments apply is

In such a limit the non-liuear n-model {30.9) cornpletely describes the long distllllce
2
properties of the (4>2 ) fiold theory a.t T fuced, T < T 0 • ln addition the coeflicient ín
front of the effective a.ct.iou becomes la.rge a.t low temperature like iu the lattice model.

30.2 IR Divergences, Sponte.neons Symmetry Breaking and the Role of Di-

mension Two

In Chapter 14 ...., have shown that the perturbative phase of tbe no11-línear a model iB
automatícally a phase in which the O(N) symmatry is spontaneously brokeu, and (N -1)
components of S;, colled hereafter w;, are me5sless Goldstone modes. ThiB leads to IR
divergences for d $ 2 which we reexa.mi.ne in the context of pha.se tro.nsitioDB.
(i) We bave a.rgued ín Section 23.5 tbat in tbe N-vector model, Cor d > 2, the O(N)
symmetry iB &pontaneously broken at low temperature. This argument iB consistent
with the property tbat for d > 2 perturbation theory whích a.lso predict& spontaneous
symmetry brealting (SSB), is not IR divergent. At T < Te fuced, the large distance
behaviour of tbe theory iB donúnated by the massless or spin wave excitations. On
the other hand nothing ca.n be sa.id, in perturbatíon theory, of a possible critica! region
T~T•.
(ü) For d ~ 2 we know from the Mennin-Wagoer theorem (see a.lso the a.no.lysis of
Sectiou 23.5) that SSB with ordering ((S) !-O) is irnpossible in a model with acontinuous
symmetry a.nd short range forces, nnd thiB Í8 aga.in consistent with the appea.mnce of IR
dívergenc"" in perturbation theory. For d 'f 2 the criticai temperature T. vanishes and
perturbation theory rnakes seuse only in presence of o.n IR cut-off which breaks explícitly

----------------------~·IA<WI'·t.'l>t.'.J•'J<'I ~· JII,J'I,l!. 'l"tJl,.,.,....,.,

670 Ferromagnetic Order at Low Temperoture 30.3 r
the symmetry and orders the splns. Therefore nothlng can be said about the long distanco
properties of the unbroken theory directly from perturbation thoory.
To go somewhat beyond perturbation theory we shall use field theory RG methods. lt ís
therefore necessary to first define the model in the dimension in "·hích it ís reiiOrmalizable.
Because the non-lincar a-model is renormalizable in two dimensionslR divergences have
w be dealt with. We therefore introduce an IR cut-off in the form of a magnetic field.
2
We theu procead ln formal analogy with the case of tbe ( .p2 ) field thoory, j.e. st.udy tbe
theory in 2 + • dimension as a double series expansíon in tbe temperature T a.nd "- In
this way the perturbatíve expansion ís renormo.líza.ble and RG equatíons foii{IW.

30.3 The Non-Linear a-Model: RG Equations

By three different methods we bave shown that the large distance pbysics or' the N-
vector model can be described below T, by the non-linear a-model. We now study thís
model from the point or view or renormo.lízation and renormalization group. Th genera.te
perturbatlon theory we parametrize the field tÍ> of previcms section by (see Sectiou 14.1):

4>- {a(.:r)
~ 1t"(:r)

and eliminate locally the field a(.:r) by:

a(x) = (1- ,..~(xl) 112

.
2
As we have done for tbe ( q, 2) model, we scale ali distam:es in order w meiiSUre momento.
in units of the ínverse lattíce spacing /1.. We thl!ll consider the partition function in a.n
external magnetíc field h:

(30.10)

with

S(11,h)

Tbe properties of this action have been studied, in sllghtly different notations, in Cbapter
14. We here directly borrow the results. We have sbown that it is renormo.lizable in two
dimensioll5 and chara<:terized tbe forro of the renormalized actíon. We recall that the
presence of a non-vanishing magnetic field h is requíred, for d ::; 2, botb to select the
classical minimum of the action around which to expand perturbation tbeory and to
provlde the theory with an IR cut-off. The renormo.llzed action has tbe form:

(30.12)

in whiclt 11 is the renonno.llzation sco.le and:

a,(x) = [z-l- 'lrn'''- (30.13)

30.3 Fermmagnetic Order at LoT,JJ TemP"rotur< 671

Not<e tha.t the renormalization constants can be chosen h independent. This is automat-
ically realized in the minimaJ suhtraction scheme.
The relation:
1r,{x) z·lf',..(x), (30.14)
ímplíes
•-2'2 = Ad-2 zt/2~.
11 (30 .15 )
t, t
With our conventiom; the coupling constant, which is proportionaJ to the temperature,
is d\mensionless. The relaticm between the cut-off dependent and the renorrnaJized cor-
relation functions is:

(30.16)

Differentíating with respect to A at renormoillwd parameters fixed, we ohtain the RG

equatíons:

[A :A+ fi(t)~- %wJ + :(t)h :h] r!"l (p,; t,h,h) =o. (30.17)

We have assuwed t.hat the renormalization constants, and thus the RG functions deftned
by:

A: 11 1 t = fi(t),
rem. fixed
8
A
8
AI l'en.fixed
(-lnZ) =((t), (30.18)

11:111 ren. fbled lnh p(t),

have been chosen h independent. The coeflicient of â/IJh can he derived from equation
(30.15) which lmplies (taking the logarithm of both members):

o h.- 1 Ai!_ h+ d- 2- !ç(t)- fi(t)

âA 2 t '
(30.19)

and therefore:
1 fi(t)
p(t) = 2 - d + 2((t) + -t-. (30.20)

To be able to discuss correlation functions involving the o-field, we aJso need the RG
equations satisfied by the conoected correlation functiollS wtnJ,

a fi(!)-+
a -Ç(t)
n ( 1 fi(!) ~) h~] wtn) = o
[A-+
aA ât 2
+ -((t) + -
2 t ah '
(30.21)

in which we now have set:

(30.22}
In Section 15.6 we have caJculated the fi-function at leading order for aJI symmetric
•paces arnong which the u-model is one of the simplest examples. In this model the two
RG functíons ca.n be obtained at one-loop order ftom a caJculatíoll of the 2-point function
r<2J,

r<2J (p) A"

T (ll~ + [P2 + t (N ] !A ddq
= h) + 1) h) --d -,--h + O(t). {30.23)
(21r) q +

-='0--------------------,~·IA<Wf',t.'l>t.'.J•'J<'I~· JII,J'I,l!.
672 Fermmagnetíc Onter at Low Tempernture 30.4

2
Applying the RG equation (30.17) to f(>) and identífying the coefficients of p and h, we
derive two equations which determine fJ(t) and Ç{t) a.t one-loop order

2
P(t) = ct- (N2- )t 2 +0(t 3 ,t 2 <)' (30.24a)
"'
((t) (Nz: l) t +o (t', t~:). (30.24&)

30.4 Discussion of tbe RG Equations

From the expression of {J(t) of equation (30.24a) we immediately conclude:

For d :S 2 (c: $ 0), t = O is an unstable lR fixed point, this IR inst.ability being indu<ed
by the vanishlng mBSS of the would-be Goldstone bosons. The spectrum of the theory
thus is not given by perturbation thecry and the perturbative assumption of spontlUieous
symmetry breaking at low tempereture is inconsístent. As mentioned before, this result
agrees with rígorous argumenta. Note tha1 sincE the model depends only on one coupling
c.onstant, t O is also a UV stable futed poínt ( the property of large momentum asymp-
totic freedom}. Section 30.6 contains a sbort discussion of the pbysi"" in two dimensions
for N > 2. The abelian case N = 2 is speda.! and will be examíned in Chapter 32.
For d > 2, i.e. c: > O, t = O is a stable IR fixed point, the O(N) symmetry is sponta-
neously broken at low tem per ature in zero fie)d. Tbe effective ooupling consta.nt, whícb
determines the large distance behav'iour, approaches the origin for a.ll tem per atures t <te,
t 0 being the first non-trivial zero of {J(t). Therefore lhe large distance properties of the
model C31J be obtained from t.he low temperature expansion and renorma.!ization group,
replacing the perturbative pa.rameters by dfective parameteJ'!l obta.ined by solving the
oo~~ .
The critirol tempemture. Fina.!ly we obseJ'Ve that, at least for ~ pasitive and sm a.!l,
and N > 2, the RG function {J(t) bas a non-trivia.! zero t.:

(30.25)

Since t. is an ullStable IR fixed point, it is by definition a criticai temperature. Conse-

quences of tbis property are studied below. Let us only immediately nate that te lo a.!so
e. UV fixed point, i.e. it gcwerns the la.rge momentum behaviour of the renarma.!ízed the-
ory. The la.rge momentum behaviour of correla.tion functions is not given by penurbation
theory but by the fixed point. As a consequence the perturbative result that tbe theory
can not be rendered finite for d > 2 with a finiw number of rennrmalizatian const.ant.s,
cannot be trusted.
We now discuss more precisely tbe llO)utions of the RG equll1íons.

!10.4.1 Integmtwn of lhe RG equatíom: d > 2, t < t<

We first examine lhe implications of the RG equations for the large distance bebavlour
of correlation functíons for d > 2 where t "" O is an JR llxed point. As usual we solve
equation (30.11') by introducing a se&ling pa.rarneter >. and looking for a solution of the
fonn:
(30.26)

~· Jll.l't.l:'. '1'\r"-'=---------------------'~•U.,WI·!.hk' ~'i•'t·~ J

30.4 FerrornagnetiJ: Order at Low Temperoture 673

Compatibilíty with equation (30.17) impües:

1
'(~) dt'
(30.27a)
In.\= t {J(t')'
tp) ,{(t')
lnZ(.\) =
l, dt {J(t'), (30.27b)

h(.\) .\2-• z•t•p.) t(:J h. {30.27c)

With our conventíons r<nl has the dimension d and h dimension 2. Thking into &Count
dimensíon..J analysis WE then rewrite relation (30.26):

(30.28)

For h <t: A2 perturbation theory has IR slngularities. By choosing À solution of the

equation
h( A)."" (ÃAjl, (30.29)
we ensure that the perturbation expa.nsion in the elfectíve theory at sc..Je À ís no longer
IR singular.
It is easy to verify that, at least for t sm..Jl, h <e A2 implies À __, O. Let us then
introduce three functíons of the temperature Mn(t), {(t) and K{t):

M 0 (t) = exp [ -~ 1' ~ ~::~ dt'] , (30.30)

Wl = 1
A-'t 1• exp [l ({3 tt') - ~~) dt'}· (30.31)

K{t) = M0 (t) ]A{(t)]d- /t = 1 + O(t).

2
(30.32)

Solvíng then (30.29) we find

{30.33)
Because t Ois an IR fu<ed poínt, thesc..Je-dependent t.ernperature t{.>.) -O a.nd tbus the
leading terms in the sm..Jl h and small momento. linrit can be calculated perturbatívely.
Using equations (30.27o,b) we obtain the beha.viours of t{À) and Z(À):

t(À)- Àd- 2 tK(t)Mõ 1 ~ t[K(t)]d/l Mõ 1hld-•)1 2A2-d, (30.:?.4)

Z(Ã) "'Ml(t). (30.35)

lt follows

rlnl . - -n "I' In) P< t [K( t )]d/2 ( h ld-2)/2 )

(p,t,h,A} Mn (t)[K(t)h] r ( [K(t)h]'l•' Mo(t) 112 ) ,1,1 .

(30.36)
Actu..Uy it is easy to verify directly, usíng dimension..J an..Jysis in the form

( A .Ê_
811
+ 2h.Ê_ +
{}h 'âp;
p·_!!_) rlnl ~ ttrln) '
-=--------------------~·IA<Wf',t.'l>t.'.J•'J<'I ~· JII,J'I,l!. 'l"tr~
674 Ferromagn.tic Oníer at Law Temperoture 30.4

that equa.tion (30.36) gives the general solution of equation (30.17}.

Let us apply thís result. to the determination of the singularities near the coexibtence
curve, i.e. at t fixed below the criticai temperature when the magnetic fteld h goes to
zero.
The coe.xístence curve. The magnetizatíon satisfies:

M(t,h,h):(a(z)) (30.37)

(r(") is the magneti<: field dependent free energy). At one-Joop order in a líeld ít is given
by:
M
N -1
1 - - - A -•t--
2
1
(2"')d
!A ddq
- - +O (t2)
q2 +h

Thus using the relatíon (30.37) we find;

This resnlt shows that M0 (t) is thespontaneous magneti?;ation and gives a.n interpreta.tion
to the logarithm found in equation (28.23) and to the singule.rity for :z; = -I found in
equation (29.27).

$0.4.2 Scaliny jaTm oj oorrelt'ltion .functíom, the critiool domain

It is convenient to rewrite equation (30.36) l:l/5

(30.38)

wíth
{30.39)

Equation (30.38) shows thu.t e(t) bas in ""'O fie!d the nature of a correlation JeDgtb.
For the connected correla.tion functions thc same analysis leads to:

(30.40)

Let us calcula.te the induced ma.gnetization by:

(30,41)

Inversion of this relatíon yíelds the scaling form of tbe equation of state:

{30.42)

and the 1PI correlation functions can thus be written in terms of the magnetiz&tion u.s:

rl"l (p,, t, M,A) {30.43)

~· Jlt.J•t.l:'. '1'\li,__ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _....,,~,u.,wt·!.hk• ~'i•'t·~ J

30.4 67ó

The equatiom; (30.42,30.43) are consistent with the equatians (26.63,26.72): the ap-
peara.nce of two different functions t(t) a.nd M 0 (t) corresponds ta the existence of two
índependent criticai exponents v.fJ in the (4>2 ) 2 field theory. They extend, in the large
distance límit, lhe scaling form of correlation functiol15, w.lid in the critical region, to
ali temperatures below te. There is however one impartant difference between the RG
2 2
equalions of lhe (4>2 ) theory and of the a-model: lhe (4>2 ) thoory dependa on two
coupling constants, the coefficient of 4>2 which plays the role of lhe temperatura, and
2
lhe coeflicient of ( .p 2 ) whích h os no equíva.lent here. The correlation functioru; of the
2
continuum (4> 2 ) lheory hri11e the exacl scaling form (30.43) only at the IR fixed point.
ln contrast, in the case of the cr-model, il has been possible to eliminate ali corrections
to scaling correspondíng to irrelevant operators order by order in perturbation theory.
We are therefore led to a remarkable conclusíon: the conelation functions of lhe O(N)
2
non-línea.r model are identical to the correlation functiom; of the (.p2 ) field theory at
the IR fixed point. Tbis conclusion is supported by the analysis of the scaling behaviour
performed within tbe 1fN expansion (see equa.tion (2!1.42)).
The criticai domain: critica/ exponents. Let us now study more precisely what happens
_,.,
when t approaches te (for N > 2). The function {(t) diverges !08:

(30.44)

Cornparing with lhe scalíng fonn (26.35) we condude that the eorrelation lengtb expooent
vis given by
v= (30.45)

For d dose to 2 the exponent v thus behaves like:

·'
"~ 1/E. (30.46)

Tbe function Mo(t) vanisbes at t,:

1 ((te)
In Mo(t) = -
2 p• (te) In (te- t) + com;t.. (30.47)

This yields tbe exponent ,B e.nd tbus also '1 through the scaling reLation .B = !v(d-2+tr):
(30.48)

A leading order we find:

(30.49)

We finally note that the singularity of r(n) coming from the prefo.ctor { -d Mõ" indeed
agrees near te witb the result of equation (26.37).
The nature o! the cornlation length e!t). The length scale ((t) is a cross-over scale
hetween two different behaviours of correlatíon functions. For distances large compared
to e(t), the behaviour of correlo.tion functions is governed by the Goldstone modes (spin
wave excitations) and can thus be deduced from tbe perturbatíve low temperature ex-
(liUlsion. However when l a.pproaches te. {(t) becomes large. There then exist distances

----------------------,~·IA<WI'•t.'l>t.'.J•'J<'I~·ni,J'I,l1.'1"tJl"""""""'
676 Fen<Jmagnetic Order ot Low Temperuturr 30.5 r
largewith re.spect to the mícroscopic scale but sm..U with respect to Ç(t) in which corre-
lation functions have a criticai behaviour. In this situation we can construct continuum
correlation functions o:msistent on all scales, the criticai behaviour being also the large
momentum behaviour of the renormalized field theory.

General comment. From the considera.tion of the low temperature expansion wé have
been able to describe, for thoories wit.h a continuous symmetry, not only the complete
structure of th" low temperature phase, and this was expected, but also the criticai
behaviour near two dimensions in the non-abelio.n case.
What is somew hat surprising in thi.s result is that perturbation series is only sensítlve
to the local structure of the sphere 5 2 = 1 while the restoration of symmetry involves the
sph.,re globally. Thís exple.ins the peculiarity of th<> a.belian case N = 2 because locally
a cirde ca.nnot be distinguished from a non-compa.ct straight líne. For N > 2 the sphere
has instea.d a. local cha.ra.cteristic cuTVIl.ture. Stíll dilferent regular compa.ct ma.nifolds may
have the same local metric, and therefore the same perturbation theory. They ali have
the same low temperature physics. However the previous results conceming the criticai
behaviour are physically relevant only ií they are still valid when ~ is not infinitesimal
and t approacbes t., a condition which cannot be checked directly. ln particular t.he Jow
temperMure expansion misses in general tenns decr<msing like e>:p (const.jt) which may
in some cases be essential for the physics. Finally let us note that, a.t least, we havtl found
2
a direct connectíon b<>tween the {4?) Bnd t.he cr-model through the large N expansion
(see Section 29.3). Thls gives us some conlidence t.hat the previam considerations are
valid for the N-wctor model at least for N sufficiently large. On the other hand the
physics of N 2 is not well reproduced (see Chapter 32). Cardy and Hamber have
speculated about t.he RG fiow for N close to 2 and dimension d dose to 2, incorporating
phenomenologically the Kosterlitz-Thouless in their an&lysis.

30.5 Results beyond One-Loop

For explicit calculations of the RG functi<>tll> bey<Jnd one-loop, it is convenient to use t.he
renormaliwd a.ction (30.12), dimensional regularization Md minimalEUbtra.ction. For
example the one-loop renormalized 2-point functian rea.ds:

r<•J (p) = f (z:,~ + 2

hZ 1 1 ) + [p 2 + ~ (N 1) h) - -1d
(2?r)
f ddq
"""2.:
q +"
+ O(t). (30.50)

This expression of rm yields t.he renormalization constants at one-loop arder. In t.he

mínima! subtra.ction scheme, we fmd:

Z=l+(N-l)Ndt +O(t~),
~
(30.61)
Zt = 1 + (N 2)N:t +0 (t2 ),

in wbicb Nd is the usualloop foctor:

N-
d -
2
(4...) 0f2f(d/2) z;1 +O(e}
~· Jli.J•t.l:'. 'F\r-----------------------M~,VhLI·!.hk' ~'i•'t·~ J
30.5 Ferromagne!ic Order at Low Temperat1lre 677

The RG functions in the renorm..:Jized thoory follow:

fJ(t) ~ ~~ (t+tdl~~z·) _, ~~- (N2~2)t, +O(t3),

(30.52)
((tJ = /3(t)dmd z =
t
~~
21r
+O(t').

At the one-loop arder we rec:uver, as expected, the expressíons (30.24). We now give
directly the two RG functions and the crítice.l exponents at the arder presently available,
i.e. four loo115, wíthout presenting the detaíls of the ce.lcuiation. For convenience we
rescale the coupling constant t
tN,. (30.53)
At thU; arder ~, the correspooding /1-function, reads:

(30.54)

in which the quantity bis a constant:

1 3
b= -U (N2 - 22N + 34) + 2Ç(3)(N 3). (30.55)

We rec..:Jl the.t the value of the numeric..:J constant C(3) (which should not be confu.sed
wíth the function C(t) of equation (30.24b)) is:

((3) = 1.2020569 ....

The anom..:Jous dimemians of the composite operator of spin l (see Sectíon 14.8) is given
by:
"' -{
Ç1(t 1 = at 1 + (N 2)t, [34 + (5-
- - + -1+a
3
N
2
-{(3) ) t·]} , (30.56)

w:ith:
a= -(N +I- 2)1/2. (30.57)
The case I 1 oorresponds to the field itself. The function Ç(t) given o.t one-loop arder
by equation (30.24b) is related to {1 by:

{(f)= -2(,(1).

From these expressions the v..:Jues of the critíc..:J exponenta TI and v follow. Defining:

i=l:/(N-2), (30.58)

we obtaln:

'l=i+(N l)i2 {-l+~i+ [-b+(N-2)c~N + S~N((3))]f"}+o(~),

v= {1 €+ (
4
~ N) f 2 + l [N2 -10N + 18(3- N)Ç(3)] i.3} +O (€").

~-------------------~·IA<WI'·t.'l>t.'.J•'J<'I~· JII,J'I,l!. 'l"tr"""""=

678 Ferromagnetíc OTder at Low Tempemfure 30.6

30.6 The Dime=ion Two

Dimension two i:; of special interast from the particle physics polnt of vie'W. The RG
function f:l(t) is then:

The non-línear a-model for N > 2 is the simplest exa.mple of 8 so-called a..o;ymptotically
free field theory (UV free) since tbe first coefficíent of the 13-function ís neg8tive, in
oontrast with the </14 field theory. Therefore the large momentum b<.haviour of correlation
functions is entirely co.lculahle from perturbation theory and RG arguments. There is,
however 8 counterpart, the theory is IR unstable and thus, in zero field h, the spectrum
of the theory is not perturbative. Contro.ry to perturbative indications, it coosists of N
massive degenerate states since the O(N) symmetry is Dot brokell. Asymptotlc fremam
and the non-perturbative chMscter of the spectrum ..,.. also properties of QCD in four
dimenslons, the theory o! strong interactions (see Cha.pters 33,34).
1f we now define a function ((t) by:

e{t) 1'-1 exp l/ /3~~') 1' (30.60)

we can integrate the RG equations In the same wa.y and we find that ((f) i:; tbe correlation
length in zero field. In sdditioo we can use the explicit expression of the 13-function
(equation (30.54)) to ca.lcu!Ate the correlation length or the pbysical mass fur small t:
C 1{t) = m(t) Kl'c'/(.N-l)e- 2.. /!(N- 2 )'1 (1 + O(t)). (30.61)
However the exact value of the integration constant K, whích gives the physico.l m= in
the RG scal€, Cflll only be ca.lculated by non-perturbative tecl>niques.
Fino.lly tbe sco.llng forms (30.38,30.40) imply that the perturbative expansion at fixed
magnetic field is valid, at low momeuta or large distances, fllld for h/ho(t) large.
Elitrur's Cl)tljecture. The O(N) symmetric action (30.9) (h= O) has a sphere of degen-
erate classico.l mínima. To define perturbation theory we have been forced to add a linear
symmetry breaking term to the action, which selects one particular classico.l miuimum.
W<õ have claímed, and it is easy to verify, that for d ~ 2 the correlation functions have
IR divergen= when the parameter h goes to zero, a property tbat is consist.ent with the
absence of SSB for d ~ 2. However to co.lculate in perturbation theory another option
ÍE available, whích wíll h€ used systematico.lly in tbe case of Íll5tantoll calculations ( see
Chapters 37-43): one does not introduce a symmetry breaking term but lnstesd a set of
collective coordinates which pararoetrizes the set of clussico.l mínima. One then expands
in perturbation theory around one fixed minimum but tteats perturbatively only the
mod"" of tbe field that do not correspond to a globoJ rota.tíon. One fina.lly sums over
oJI clo.ssical mínima. C!early, after this Iam summation, only O(N) invariant correla-
tion functíons survive. As o.lrea.dy mentioned in Section 14.4, it b8S been conjectured
by Elitzur and pmven by David that in two dimensions the O(N) inva.ría.nt correlation
functions obt.aíned by this procedure have a regular low temperature expansíon: thís
meaJIS that if we ealculat.e O(N) invaria.nt correlatíon functions by perturbation theory
with a. non-vanishing field and take tho lirnit h = O, this limit is IR finite. The subtlety
of thís problem when compared to the ínsta.nton case is that in the infinite wlume the
zero momentum singularity due to the choice of one classico.l miDimum is not an isolated
pole (see o.lso Section 3.4 and Chapter 37).
30.7 FerTVmagnetic Order at Low Tempemtu,.., 679

30.7 The Large N Limit

In Chapter 29 we have shown that within the framework of the large N expa.nsion the
non-linear a-model was obtained from Lhe if>4 field theory by neglecting terms inelevant
in the large distance limít. In order to more ~plícitjy show the correspondence between
the set of parameters used in the two models, let us solve the cr-model in the large N
límit. Using the linear fonno.Iism of Section 14.9, we rewrite the partition functíon:

Z =f {d,P(:~:Jd>.(x)] exp (-S(q,,>.)l, (30.62)

with:
(30.63}

lnt.egrating, as was done in Chapter 29, over N l oomponents of <P a.nd calling CT the
remaining component, we obta.i.n:

Z f (dcr(z)d~{z)] exp (-S.~t(a, .\)I, {30.64)

with:

The le.rge N linút is uow taken at tN !ixed. Tbe saddle poínt equati<mB, ano.Iogous to
equations (29.7), nre:

(30.66)

(30.67)

where we have set (J.(;o;)) = m 2 • At low temperature cr is dilferent from zero and thus
m, whicb is the mass of the "-field, vanishes. Equat.íon (30.67) gives the spontaneous
magnetizatíon:

(30.68)

(30.59)

Therefore equation (30.68) c8.ll be rewritten:

a2 ~ 1 tf!c. (30.70}

Above t,, a ínstead vanishes and m, which is now the oommon mass of the ,._ and u-lield,
is for 2 < d < 4 given by:

(30.71)

-=--------------------~·IA<Wf',t.'l>t.'.J•'J<'I~·ni,J'I,l!.
680 Fermmagnetic Order at Low Temperotu,.., 30.8

e
We recover the &:oling form of the correlation length ~ ljm. Frorn the equatÍOIUl
(30.70.30.71 ), we can !liso derive the RG functions at le!>ding arder for N large:

N
,B(t) =et ((t): 27ft. (30.72)

The cakulation of other physlca.\ quantities and the e:xpansion in 1/N follow from the
consideJ'ations pf Chapter 29.
Two dímensicms and tM. questwa of Borel summabili!y. For d = 2 the criticai temper-
ature va.nishes and the parameter m has the form:

(30.73)

in agreement wíth tbe RG ronsidera.tions. Note that tbe field 2-point function takes in
the la.rge N -limit the form:
(30.74)
The ma.ss term vanishes to a.\1 orders in the expansion in powers of the coupling consta.nt
t, preveuting a.ny perturbative colcul&tilln Gf the m""" of the lield. Tbe pert urbation seri""
is trivia.lly not Borel summable. Most likely thiB property is !liso true for the model at
finite N. On tbe other hand if we break the O(N) symmetry by a magnetic field, !>dding
a term hc to tbe action, the physicol mass bemmes ca.lculable in perturbation theory.
Correctíons to scaling and lhe dimension /D1lr. In equation (30. 71) we ha.ve neglect.ed
corrections to scaling. If we take intp a.ccount tbe le!>ding correction we get insteoo:

where K(d) is a constant whlcb e:xplicitly depends on the cttt.off pro<:edure and can thus
be varied by cbanging contríbutions of irrelevant operators. By compa.ring with tbe re-
sults of Subsection 29.2.2, we discover that, olthough the non·linear n-model superficiolly
d€pends on one pararneter less thon the rorresponding tjJ0 lield theory, actua.lly this pa-
rameter is hidden in the cut-off functíon. This remark becomes impprta.nt in the four
dimensionol limit where most le!>ding contributíons come from the le!>ding corrections
to scaling. FN example for d = 4 equation (30.71) ta.kes a dilierent form, the dominant
term in tbe r.h.s. is proportional to m 7 1n m. We recognize in the fuctor In m the eftective
q,4 coupling'e.t ma.ss scale m. Beyond the l/N expa.!Ulion, to describe with perturbation
theory and renormalization group the physics of the non-linear u model it is necessary
to introduce tbe operator f d 4 x .\2(x), which irrelevant for d < 4, becowes marginal, and
to returo to the 4>4 field theory.

30.8 Generali.zations

We have shown in Che.pter 15 that from the point of vi..w of the renormolizatilln group,
tbe properties of the non-linee.r 11-model genera.\ize to a.\l models defined on symmetric
spa.ces. They are a.\l UV free in two dimensions (see Section 15.6), and therefore haw
a phase transitiou at a. criticai tempera.ture of arder E in 2 + E dimension. However the
identilication af the correlatlon functions of these modela with the correla.tion functions of
a 4> 4 type tbeory at an IR fixed point is in general not easy. In particular the connectlon
thiXllljlh a la.rge N expa.nsion does not e:xist in generol. lt is likely that fur some of tbese
 30.8 Ferromagnotic Order at Low Temperotun: 681

rnodel• the transition found frorn the 2 + E expansion is actua!ly a first order transition
for .: non-infinítesírnal. Let us rnent;on that the ~-functíon h..s been calculated up to
four loops for a large d..ss of symmetric spaces.
Symmetric spaces cnrrespanding to nDn-compact groups. ·Previous considerations can
be formally extended to symmetric spa.ces G j H ill which the group G is non-compa.ct,
hut the •ubgroup H remains compa.ct. Symmetry groups obtained by comple>cification
o( oompact group generate such spaces. A simple example is O(M, N)/O(M)fO{N), in
which O(M, N) is a pseudo-orthogonal group. From the point of view o( perturbation
theory tbe only difference is that tbe sign of tbe coupling has changed. However this
means that these models are no longer UV free but IR free in two dimensions. They
have a non-trivial IR fixed point in d "' 2 - I! dimension. The physics is thus completely
different: the situation bears some Rllalogies with the behaviour of the .p4 theory in four
dbnensions. The existence of massless modes below 2 dimensions is not in contradiction
with rigorous theorems because the symmetry group is not <:ompact. Such mod"ls play
a role in the theory of localization.

Bibliographical Notes

The renormalization group properties of the non-linear u-mod<ol have been discussed in
A.M. Polyakov, Phys. Lett. 59B (1975) 79; E. Bré:!in and J. Zinn-Justin, Phys. Re~.
Lett. 36 (1976) 691; Phys. &1!. B14 (1976) 3110; W.A. Bardeen, B.W. Lee llild R.E.
Shrock, Phys. Re~. Dl4 (lll76) 985.
Higher arder calculatíons of criticai exponents a.no due to
S. Hikarni, Nucl. Phys. B215[FS7] (1983) 555; W. Bernreuther and F.J. Wegner, Ph<;•.
!c &11. Lett. 57 (1986) 1383; F. Wegner, Nucl. Phys. B316 (1989) 663.
f Speculations about a RG 8ow near dimension 2 consistent with tbe N = 2 Kosterlitz-
~ Thouless transition and tbe non-Jinear u-model results for N > 2 can be found in
J.L. Ca.rdy and H.W. Hamber, Phys. Rev. Lett. 45 (1980) 499.
Long range forces have been discussed ill
E. Brézin, J.C. Le Guillou and J. Zinn-Justin, J. Phys. A: Math. Gen. 9 {1ll76) Lll9.
For an introduction to the problem of Jocalization in electron eystems see for exarnple
E. Fradkln, G. Pw:isi in ReCEnt Advances in Field Theo'l/ and Stntislical Mechanics,
I.... Houches 1982, R. Stora and J.-B. Zuber eds. (North Holland, Amsterdam 1984);
F. Wegner, Lecture N"tes in Physics 216 (1985) 141.
Renormalization of composite opera.tore is discussed in
F.J. Wegner, Z. Phys. B78 (1990) 33; G.E. Castilla and S. ChaktRVatty, Phys. Rev.
Lett. 71 (1993) 384.
For the Gross-Neveu model discussed in the appendix see
KG. Wilson, Phys. Rev. D7 {1973) 2911; DJ. Gross andA. Neveu, Phys. Re!!. DlO
(1974) 3235; A.B. Zamolodchikov and AJ.B. Zamolodchikov, Phys. Lett. 12B (1978)
481; W. Wentzel, Phys. Left 153B (1985) 297; J.A. Gra.cey, Nucl. Phys. B367 (1991)
657.
Tbe relation with its linearized extensíon is discussed in
A. Hasenfratz, P. Hasenfratz, K Jansen, J. Kuti and Y. Shen, Nucloor Physics B365
(1991) 79; J. Zinn-Justin, Nuckar Phyfliu B367 (1991) 105.
Finally ali these techniques have a1so been applied to the supersymmetri~ extension of
tbe non-linem cr-modeJ. A recent reference is
J.A. Gracey, Nud. Phys. B352 (1991) 183.

-=---------------------·~·IA<Wf',t.'l>t.'.J•'J<'I~·ni,J'I,l!.'l"tJl~
682 Ferromagnetic Order at Low Tempemture A30.0

APPENDIX 30
THE LINEARIZED GROSS-NEVEU MODEL

To illustr8te the techniques dl"'Xlloped in Chapters 29,30, WE díscuss in the appendix the
Gross-Neveu (GN} model wlúch is describOO in terms of a U(N) symmetrit action for a
set of N mBSSiess Dira.c ferJIÚons {.P', .j;'}:

S(ôp,.p) -f ddx [;p·,.P+ !G(;í>·.P)l

The GN model has in l"'Xln dimensions a distrete chita! symmetry:
1P,..... 1s.P, ;p ,__.. -1P1s, (A30.1)
which prevents the e.ddition of a. fermion mass term while in odd dimensions a mass term
breaks spa.ce pa.ríty (see Appendíx A5). Actually the two symmetry opera.tions can oo
wrítten in a fonn

valid in ali dimensions.

Thís model illustrates the physícs of spontaneous ferrnion mass generstion and, in
even dimensíons, chi:ral symmetry breaking. However, as in tbe c~ of tbe non-linear u
model, the perturbative GN model desc:ríbes only one phase a.nd the symmetry brea.king
mechanism is non-perturbative. lt is therefore useful to first consider a more general
model, the Hnearlzed GN model, whlch has the sa.me chita! Md U(N) symmetríes, and
w hose action is:

S (1p,.p,u) = j ddx [~ij. · (~ + gu).p +! (8"u) 2

+ !m2 u 2 + ~u•]. (A30.2)
where u is an e.dditional scalar field. The action still hos a reflection syrnmetry, CJ
transforming ínto -u when too furmions transfonn by (AJO.l). We r<lCOglliuo the model
that we have introduc:ed in Sectíon 11.8 to illustr8te the calculation of RG functíons at
one-loop order in 8 mode! with booons and fermions. ln contrast wíth the GN model,
the syrnmetry breaking can then oo discussed by perturbative methods. An analogous
sitnation hll$ already been encountered wben corop8ring the (</12 ) 2 field theory With·the
non-línear u model. Even more, tbe GN model is renormaliza.ble in two dímensions a.nd
the linearized model in dimension four. The maln difference is tbat the GN perturbative
model deseribes the symmetrlo phose while the perturbstive oon-llnear u model descríbes
the broken pha.se (but in both coses the phs.!;es are mass!ess ). We now show that the
physícs of the GN model can indeed be disctlssed by the same methods as ferromagnetic
systems, that. is RG equations near two and foor dimenskms, a.nd large N expansion.
The phase trnnsition. Examining the a.ction (A30.2) we see th8t in the tree approxima-
tion wheu m 2 is nagative the chita! symmetry is spont.aneously broken. The u expectatíon
value gives 8 wass to the fermíons, a mechanism rerniniscent O( the standard model of
weak-electroma.gnetic intera.ctíons:
~=g(u), (A30.3)
while tbe u ma.ss then ís:
2 ), ~
m. = m,.. (A30.4)
392
To study the model beyond the tree approx:ímation ~ now discuss RG equations near
four d ímensions.

'-=---------------------,~.u.,WI•!.hk' ~'1•'1·~ J
A30 ..J Ferromagnetic Order at Low Temperature 683

A30.1 RG Equations near Four Dimensions

The model (A30.2) is trivial above four dimensions, renormollzable in four dimensions and
can thus bc studied near dimension 4 by RG techniques. Ce.lling 1J. the renorme.lization
scale, setting d 4 - E, we can write the ra:torme.lized action:

S,(i/J,.p,u) = j dd.x[-z.,.j,-tlii.P-IJ.'I'gZuuTfo·\b
+ ~z. (a,.u) 2 + ~(m; + Zmt)a 2 + IJ.' Z;.~u•],
4.

where m~ is the criticai bare mass sqliAl'<ld, the critica.l temperature in statistical IM-
guage, =d t characterízes the deviation from the critical temperature. The RG equations
for the renormalized functions are:

a fJ a
[IJ.-+
ÔIJ. g ôg2
ô 1
,-+fJ~--·.ll)oj>-
â). •
i D1)a-'lmf-a] r'{i n) o.
&t
(A30.6)

Remart. Note that in this section for convenience we set for the "'/ matrices tr 1 4
as in four dimensioll5. Th extrapolate the results to other dimensions one has to replace
everywhere N by N' = tr lN/4.
The RG functiuns. The RG functiollll h ave been ce.lculated at one-loop order in Seetion
11.8:

(A30.7)

(A30.8)

In four dimensíon;; the or igín ). g2 = O i.s IR stable. For d = 4- E one finds a non-trivie.l
fu:ed point:
À, (A30.9)

with
À,
3 ( 2~ + 3) [-(2N- 3) + ../4N 2
+ l32N + 9]. (A30.10)

The matrix of derivatives of the {J-functions has two eigenvahres w,w',

w <, w' = <../4N 2 + 132N + 9f(2N + 3), (A3ú.ll)

8lld thus the fixed point is IR stable. The first eigenvalue i.s e.lways the smollast.
The field renormali2.ation RG functions are at the sarne order:
2N 2
(A3ú.l2)

At the fu:ed point one finds

"" 87r2g,

2N~ t
11" = 2N + 3' 'I~ = 2(2N + 3). (A30.13)

----------------------,~·IA<Wf',t.'l>t.'.J•'J<'I~·ni,J'I,l!.
684 Ferromagnetíc Order ai Low Temperoture A3Q.2

The RG function T/m (ca.lled 'I> lu Section 11.8) corresponding to the mass operator;, at
one-loop order:

and thus tbe exponent v:

é-
(
2- 2À,

Fin...Uy we can emluate tbe ratio of masses (A30.4) at the fixed point:

m; À, -(2N- 3} + ./4N'l + l32N + 9

m~ = 3g~ = 9

The existence of an IR fixed point has a remarkable consequence: lf""' assume, as we do

in the theory of pbase transitions, tha.t the coupling cousto.nts are generic a.t the cut-off
sca.le, tben lhe rotw of fermíon o.nd scalar ma.s•es is fi:<ed, prov:ided tbe a expect~ion
Vll.!.ue is much smaUer than the cut-off. Note, however, that in four dímensions the
coupling dependent corrections vanish only logaritlnnica.lly witb the cut-olf.

A30.2 Large N Lim.Jt

We can examine the large N behaviour of previous expressions. For example ""' find
À, ~ 481r 2 f N and d- 2 +'I" 2. This reminds w; the (4>2 }, field theory and suggests
a study of the large N limit. For this purp<JOe we integra.te over N 1 fermlon fields,
resca.le for convenience ga into a, and get the effective a.ction:

(A30.14)

'Ib take the large N limit we assume a linite e.nd g 2 , À = 0(1fN).

The el<pectatíon va.lue of a ís now given by a gap equation:

(A30.15)

where N' is the number of fermions, í.e. N' = N K 4 with K 4 the t.ace of the identity
matríx (see Appeudix A5). The critica.! temperature or hare mass ís thus given by:

m<
2
g2 - -N'--f ]
(2>Jt)4
A dd
_.!
q2 '
(A30.16}

wbich shows tlw.t the fermions favour the cbira.l t.ansition. In particular when d ap-
prooches 2 we observe that m~ ..-. +oo wlúch implies that the chiral symmetry is a.lways
broken in 2 dimensiom;. Using equation (A30.I6) inside (A30.15) we obt.o.ín a.n equation
for the expectation Vll.!.ue of a, in the low teruperature ph,.,, as a function of temperature.
Setting t (m2 m~)/fl we find:

À a 2 /A d"q (A30.17)
t+ 6g•a• +N' (2,.)• q'(q•+c:r2) -o.

--------------------OO<'M~,VhLI·!.hk' ~'1•'1·~ J
 A30.2 Ferroma.gnetie Order at Lpw Tempemture 685

For d :s 4 tbe last terrn is more singular tban tbe second one for a small. In particular
for d < 4:
(A30.18)
as in the non-linear a-model.
Sillce, at leading order, the ferntion mass m,. = a, it ilnllJ.edíately follows th~<t the
exponent v is also given by:

V ~ fJ- 1/(d- 2) ~ '1c =E· (A30.Hl)

We rea.d from the effectíve action tha.t at this order '1~ = O.

Finally from the efl'ectíve action we can calculate the u-propagator at lea.ding order.
We observe that lhe term coming from the furntion determill811t is aga.in domillant at
small momentum near tbe criticai ternperature. Thus at T<
N'b(õ)
(A30.20)
(aa) - 2P"-' ,
witb
b(<) _l_r(z· d/2) r• (df2- 1). (A30.21)
(41r)d/2 r(d- 2}
Tbe result is consistent with the value of '1~ found above.
More generally, using the saddle poínt equation, one !inds for the ínver.se a-propagatm
in t.he messlve phase:
2 4
t.,-1 P ), 2 N' ( 2 4<r') !h d q (A30.22)
o lPl=92+39 4" +2(2")" P + (q'+u')l(p+q)'+a•r
In the scaling limit the propagator w.nishes for i' = -4u 2 wlrlcb is just the fi,tp threshold.
Thus, in this limit, m 0 2m,.. 'Th.king into account correctíons to scaling, tbe a mass
lies above or below t.he two furmiou thresbold depending on the value of ),.
Let "" finally note tho.t from the bebaviour of the propagator at the criticai point,
.6.0 (p) <X p 2 -d, we derive tbe ca.nonical dimension lei of the lield a in tbe lar ge N
expansion: I<T] = 1.
Correctioru tu sooling and the IR fized point. Tbe IR lixed point is determined by
demandíng the cancellation of the leading corroctions to scaling. The integral in (A30.17)
is given for large values of tbe cut-off A by equation (29.17):

C(d)"-• - a(d)A -• +O (aA2-•) 2

, (A30.23)

tE" (a(ê)- 1/S~c). Demanding the cancellation ofthe coefficíent of a 2 , we obt8.Íil a relatíon
between À 8lld g:
• ),. A• ),.~,• ( ')
g. "" 6N• a(d) "" 3N + 0 c '
a result consístent with the results of the E-e:xpansion.
In the sarne way ít is possible to calculate tbe lsading correction to t.he a-propago.tor
(A30.22). Demandíng the can.,.,Uation of tbe lea.ding correction W<t obtBin

~
g~
+~a
3g!
2 - !N'(p2 +4a2 )a(d)A-• :O.
2

---------------------,~·IA<WI'•t.'l>t.'.J•'J<'I~·ni,J'I,l!.
686 Fem>magnetic Order at Low TemP"rnture A30.2

The coellicient of cr 2 cancels ó:om the previous relation and the cancellation the coefficient
ofr? yields

in agreement with tbe t:-expansíon for N large.

The jermion a.noma.lous dimension at arder lfN. Let us now calculate the dimension
% at fim non-trivíal arder. The fermíon 2-point function at the crítica! poínt is theo:

Tbe explicit forro of the cut-off A li; irrelevant at this arder, and we just h ave to calculate
the coefficient of the term pIn A. This can be adU""ed by expanding the integrand for q
1...-ge:

Then using

1 !" qd~'(p+q)2
w<lÍ!Dd

(2..-)a
ddq (Jf +() 2 (d- 2)
-r(d/2)(4:rr)d/> -d- ,llnA.
The exponent 1)~ at arder 1/N follows

(d -2) 2 f(d -1)

% = N•d rs(df2}r(2 d/2).

For d = 4 t: we lind % ~ t:f4N, result consístent with (A30.13) for N la.ge. For
d = 2 +~ instead ooe linds%- c2 j4N.
The relatíon to lhe GN mvdel for dim.en.rions 2 '.S d '.S 4. We have seen that the tenns
(B~o-) 2 and u 4 of the effective action which have a canooical dimension 4, are irrelevam
in the IR. critica! region for d::; 4. We ~gnize a situatlon a.lready encountered in the
(f/>2 ) 2 field theory in the 1...-ge N liiJÚt. In the scaling regíon ít ls p05Bible to omit them
a.nd one tben linds the action:

(A30.24)

The integral over the 11 field can explicitly be performed and yields the action of the GN
model:
S (ij,, ,P) =-f ddo: [ij, ;i;jH ~2 (ij, · .P) 2] ·
The GN model and the linearized modelare thus equivalent for the large distance physícs.
In the GN model, in the large N límít, the u particle appears as a # boundstate at
thresbold.
Conwrsely, it would seem thnt the GN model depends on a smaller number of pa.ratn-
eters tlum ib! renormalizable extensioo. Again thís problem ls only interesting in four
A30.3 .ff!rromagnetít Order al Lqw Tempernrun 687

dimen5ions where corrections to scaling, i.e. to hee field throry, are important. Hawever,
if we examine tbe divergences of the terrn tr In (jil + u) in the effective action {A30.14)
relevant for lhe large N limit, we finà a. loco.! polynomio.l Íll (]" of tbe form:

Therefore the va.lue of the determina.nt can be modilied by a loco.! polynomi..J of tbis fonn
by cbanging tbe wa.y the cut...off ís implemented: a.ddition..J parameters, as Íll tbe case of
the non-lioear a-model, are hldden in the cut-off procedure. Near two dimensíons these
operators can be identified witb (,P,p) 2 ,[â,.(#))2 ,(.ji.p)4 . It is clear that by cbangíng
lhe cut-off procedure we cbange tbe amplitude of high& dimension opera.tors. These
bare operators in the IR límit have a component on alllower dimensionol renormalíred
operators.
Note finally tbat we could bave added to the linear model a.n explicit breakíng term
linear in the a tield, wbicb becames a fermion ffi!IBS term in the GN model, and whicb
would h ave played the role of the magnetic field of the ferromagnets.

AJO.J The Gross-Neveu Model oear Two Dimensioos

The Gross-Neveu model plays with respect to model (A30.2) tbe role tbe non-Jínear u-
model plays with respeet to the 4>4 field tbeory. ln perturbation theory it ís renorm..Jh.able
in two dimensiaiiB, and de.cribes only one phase. The main difference is tbat the role
of tbe sponta.neously broken and the explicitly symmetric pha.se are intercbanged. This
ís due to the property tbat it ís olwa.ys the massless ph...., which ís unstable in low
dimensions.
In perturhation theory the GN model is renormalizable in two dimemrlons. In tbis
dirnension the symmelry graup is really 0(2N), as one wrifies after some relabelling of
lhe fields. Therefúre tbe (,P,P) 2 interaction is multiplica.tively renormalized. A:; a function
of tbe cut-off A the hare conelution functions tbus satisfy the RG equation.s:

o. (A30.25)

A direct c..Jculation of the /3-function in d = 2 + t: dimension yields:

G2 G3
fJ(G) =eG- (N'- 2) 2 + (N'
1T
4
2}·i11·2 +0(0 ), (A30.26)

where in generic dimensions N' == NK,; (Kd is here the idenlity in the space of "f
matri~): N' counts tbe total number of fenniowc w.riables. For d 2 N' = 2N.
Note that N 1 2 corresponds, as we sb..Jl see in Chapter 32, to the sp!'Cial c""" of
the Thirring model (because for N' = 2 (.)'ryp,p) 2 = -2(.ji,p) 2 ) ror whicb the fl-function
vanísbes identica.Jly in two dimensions. The latter model ís the equiva.lent of the 0(2)
tr-model to whicb it is actu..Jiy related {see Chapter 32).
Finally the field RG fuoctioo 1J.i>(G) ís

(N'- l)(N'- 2) 03 O (G')

32lf3 + (A30.27)

----------------------~·IA<WI'•I..'i>t.'.J•'J<'I ~· JII,J'I,l!. ':!:tll"""'=o:

688 Ferromagnetic Order at Low Tr:mpernture A30.3

A30.S.1 Two Dimensíons

For d = 2 the model is osymptotically fnle. Tbe spectrum, however, is non-perturbatiw,
and many arguments lead to the conclusion that tbe clúrol symmetry is always broken
and a fennion mass generated. S-matrix oonsiderations have then led to the conjecture
tbat, for N fillit.e, tbe spectrum is:

2(N 1) . ( mr )
m..=M--,-sm 2(N-l) , n=l,2 ... <N, N>2,

wbere the parameter M is a mass-scale wbose dependence on the coupling constant

can be derived from RG arguments. This rerult Íll consistent with the spectrum for N
la.rge eva.luat.ed by semiclasaical metbods. ln particular the ratio of the masses of the
fundamentol ferroion and tbe lowest lying h<=n is:

(A30.28)

Note the.t tbe two ftrst val...:s of N are special, tbe model N = 2 is conjectured to be
equivalent to two decoupled sine-Gordon models.

A30.9.2 Dimension d = 2 +r;

As in the case of the cr-model, e.symptotíc freedom implies the existence of a non·triviàl
OV fixe<! point G., in 2 +e dimeDSion

G. is a.lso tbe critica.! coupling conste.nt for tbe transition between a pha.se in wbieh the
chlral symmetry is sponto.neously broken and a mBSS!ess sma.ll G pha.se.
At the fixed point one finds tbe conelation iengtb exponent v:

(A30.29)

o.nd the fermion field dimension 11/1]:

(A30.30)

As for the 11-model the existence of a non-trlviol UV fixed point shOW'B that the large
momentum behaviour is not giV>lD by perturbation theory &b<:MO two dimensions. How-
ever, to study tbis ptoblem in hlgber dimeosioDS a.nother method is required, like tbe
1/N exparulon wbieh we have consi<lered in the prevíous sectíon.

~· Jll.l't.l:'. ' 1 ' \ l i ' -_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _,~,u.,WI•!.hk' ~'i•'t·~ J

31 TWO-OIMENSIONAL MODELS ANO BOSONIZATION METHOD

In Chapter 30 we have díscussed the generlc O(N} non-ünear <t-rnodel. We haw noticed
that ín two dimensíons the case N "' 2 ls special because the RG 11-function vanishes.
This particular case ls specially importa.nt aod will be ex8HlÍned in Chapter 32: Tbe
0(2) invariant ferrorna.gnet in two dírnensíons provides 811 exarnple of the relebrated
Kosterlitz-Thouless phase tra.nsítion. However, a consistent and simple discll.!lSion of
this model requíres 8 new technique: We have to establish rela.tions, special to two
dirnensions, between fennion and boson rnodels, 8 method called bosoníza.tion. The
derivation involves severa.! 5teps whích we íllustrate in this cha.pter with the help of
various two-dirnensional models whích are phy>;ically interesting in their awn right.
We first study the free massl- boson and !errnion fielW.. We evalua.te the deterrninant
of the covariant fermion deriwtiV1l in presence of 811 externai ga.uge field. This will
allow to prow the basie rela.tions of the bosonization technique. We tben discuss the
sine-Gordon (SG) model. We solve the Schwinger moda!, QED with maasless fermions.
Finlllly we demonstrate the equivalence between the SG model and two fermion models
with current-current interaction: the Thirring model 811d another model with two species
of fermions.
Mass.less (unite.ry) fteld theories are conformai inve.riant. In 2 dímensions this ls a
very strong praperty because the oonformal group ls ínftnlte dimensiooal. A powerful
technology has bee.n developed to study them. A general dlscussion of conformai field
theory ( CIT) goes rnuch beyond the scope of this work 811d the interested re..der i.s
referred to the abunda.nt llterature. Here we limít ourselves to •impler considera.tions.
The functiona.l integral technique will be used everywbere although in some cases the
opera.tor formulat.ion ma.y lead to more elegant derivatíons.
The appendíx cont.n.ins some additional remarks conceming the SU(N) Thírring model
and solitons in the SG modeL
Finally note that unlike in most other cha.pters, we here work with oorrela.tion functions
expr=ed in terrns of space vari!lbles rather their Fourier transform. Renormalization
tben ta.kes the form of defining singular distributions (in the mathema.tical sense).

31.1 The Free Massless Scalar Field

Before discussing varlous models it ls necessary to examine the peculiar properties of

massless free fields in two dimensions. Let us consider the a.ction S for a free ma.ssless
se alar lield <p( x ):
S(cp) = ~ j 2
d x !ô~<p(:r)] 2 . (31.1)

The action is invaria.nt under constant trans!ations of the lield

<p(x} = <p1(x) + 9. (3L2)

To tbis syrnmetry corresponds a. conserved current J~

J~ (x) = a,cp(x). (31.3)

----------------------~·IA<WI'•t.'l>t.'.J•'J<'I ~· JII,J'I,l!. 'l"tli~

690 Two-Dimen.<ional ModelJ and Boooniza.tíon Method 31.1

Note a peculia:rity of dimeru;ion two: there exists another trivially ronserved current

(31.4)

where e~. is the a.ntisymmo:tric tensor witb <12 1.

The propaya.I.M. The field propagator 6( :z:) is:

6(:t"} = (2.-)2
1 /dfi" 2
p
e'"". (31.5)

We see that in dimension two the propagator of the massless hoson fiald is IR divergent.
The l!uctuations which tra.nslate the field <p(:t") by an almost consta.nt function ore not
damped by tbe action {31.1} a.nd I;U"e responsible for this divergence.
'Ib discuss the nature of the IR divergence we giw a small m""" m to the 6eld 'I'· For
m sma.ll tbe massive propagator behaves like:

'( )- _l_j-=!L
r+m•"'""
u:r:,m -(2~r)2
1
= - !T (lu(m2 z 2 /4) + 27) + O(m), (31.6)
4

wbere 1 "'" -1/>(1) is Euler's eonsta.nt. At z O tbe m.assive propagator is UV divergent.

Introducing a UV cut-off A we find:

1
6(0,m) A_:oo
2
1T ln(A/m) + K +o (A- 1),
where K is a constant llthich depends on tbe cut-olf procedure. For convenience we
absorh it hereafter in the delinltion of A, sett.ing:

2~ [lu(2A/m) --r]+ o (A-

1
.C.(O, m) = ). (31.7)

IR fin.ite <:t1m'lo.tion function8. Let us now shaw that, although 'I' it.self has IR divergent
oorrelation functions, some local functious of '{I bave IR finite rorrelation functions. For
instance the correlation function:s of exponentials of the field '() are given by

where tbe l<i 's I;U"e arbitra.ry ooefficients. We set:

J(x) = iL~<;6 (:r:- :t";). (31.9)

The integTal in equatíon (31.8) ta.kes the (orm:

·-==--------------------~.u.,WI·!.hk' ~'1•'1·~ J
31.1 Two-Dímensional Models and Bo3onizatian Method 691

We recognize the first functi<mal integral we have calculated in field theory:

(fi e'"'"'(••l) = exp [~ j d2 :r d~y J(:r)ó(:r y, m)J(y)] ,

= exp[-! Í::"""'ó(xi -x,,m)]. (31.10)

t,J

Replacing the propagat.ar by its small m exponslon (equation (31.6)) and collecting ali
IR divergent tenns, we find:

Í::: IC.;I<jÓ(x, -Xj,m) = - 2~ [ (2:1<.;)' (lnm -ln2 +I')- Í:: ~<;In A

t,,J t ..

+ ~l<;l<;lul:r;-:r,l] +O(m). (31.11)

'"'
Tberefore when m goes to zero only the producte 5Uch tbat L; l<i = O survive. Thi.s result
i.s related to the simple property that in iwo dimensíons the Couiomb potential created
by point-líke charges docreases at large di.sta.nce only iÍ the total system is neutra!.
Finally in equation (31.11) we see that the UV díwrgenceo; ca.n he removed by a
multiplícative renormalization (; of the composite fiel& e•••
li'(~),

e'"'"'(~)= ç, [e'""'(~)J,.•. ' (; = (A/p.)-•lt<•wl, (31.12)

where p. is a renormalization scale. Thus finally:

( rr_" e'""·<·•l) I1ren. i<j

í-1
rr <,. 1,, _ :r;ll"'"J'2• . {31.13)

Nate tbat the correlation functions of exponentials of the field decay algebralcally at
largc distance.
Remarl:. In a tra.nslation (31.2} of the ~p-field the field e''""' become;

AU non-va.nisbing correlation functíons are invaria.nt under tbis transformation, a result

in agreement with the ~>nalysis in previous chapt.ers of spontMeous breaking of continuous
S)'Iniiletrlm in rwo dímensions.
The two rJ.arye example. A particul81' """" of equation (31.13) iJ; of specio.l intere;t,
when the ~t;'s ta.ke only two values ±~<. Tben only pmducts witb an equal number of ±
signs do not vanish

n,<j (P.I"'• -x;IJ"'1.. (p \v; -yjlt'1'" (31.14)

n•.; (ptx, -y;l)"'/h
The 2-point function is givcn by:

(3U~J

----------------------~·IA<WI'·t.'l>t.'.J•'J<'I ~· JII,J'I,l!. 'l"tli~

 692 Twa-Dimenaional Madels and Basonüation Metlwd 31.1

The dimension of the operator •'""(• l follows

(31.16)

Note that for K' (2tr 2: 2 the 2-point function í.s singular at short dista.nce in the sense of
dí.stributions. It has to be renormalized as the UV divergences of its Fourier transform
show
2x--- -,-
e''"' ,.2
r(l- /•hr) (Pz)
- •' /4~-l
f d
x••f•• - r(~< 2 /41r) 4
This expression is va!id wr r< 2 < 47r. At r< 2 = 471" s divergent consta.nt has to be sub-
tracted. For 47r :S " 2 < 81r it is thus defined only up an arhitrary additive renorme.liza-
tion. For 1t2 "' 81r • second additive renormalil;ation proportional to p2 is required. We
shall agaio rneet these two s],l<)Cial values when we discuss the SG modeL
The cum!nts. The currents (31.3,31.4) provide another example of fields with IR finite
mrrelation functions. For irustance in Fourier space the 2-point functions a.re:

(3Ll7a)

(3l.l7b)

After Fourier transformation the 2-point functions beeome

(J:(z)J~(O)) (31.18<1)

(J:(:c)J~(O)) (3l.l8b)

We find proportional results (as impoaed by current conservation), although in Fourier

space the functíons are different: The r.h.s. involvea a singular diatribution 8lllbiguous
at x"' O, defined up te the addition of a 6( 2 1(:~:) function.
It is also essy to calculate the effact of the insertion of the current in a correlation
function of exponentials, since this involves calculating a on~>-point functian in an extemal
source. We find

Camplu coonlinaies. It is a peculiarity of the dimem;ion two that in real time ma.ssless
fields can be decampo.sed into left and right moving oomponents. In euchdean space this
corresponds te a. description of the z = { €1, {2} plane in terms of complex ooordínates

z= {t- iÇo, {3Ll9)

which leads to the action

(31.20)

'-=="_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ ,,~,u.,WI·!.hk' ~'1•'1·~ J

IA.,C!úCI.hibllit.JLH
31.2 Two-Dimensional Models and Bosonizalion Method 6!13

and to a decompoútíon of the field into analytic and anti-a.nalytic components

with the propagators

ó(Ç!,{l) = ('P+(z)'P+(O)) + ('1'-(i')'P-(0)},

(l"+(z)II'+(O)) _2._lnz + const., ('P-(-')'P-(0)) = _2._lnf + conat..
41r 411"

These complex variabh,. are particnlarly well suít..d for exploring the collllequences of con-
f9rrnal symmetry: the action (31.20) is obviously invaríant in the cbange z f(z'),l =
f (z'). They wUI appear naturally in the discussion of massless fermiow;.
31.2 The Free MaBSiess Dirac Fermion

Let us consider the m8S5less Dirac fennion action:

(31.21}

The rnassless dassical action has two U(l) symmetries cor,..,.ponding to phase and chiral
phase transformatiow;
~x) e•<•n•+'•l '\b'(x),
{
,p(x} ;j/ (X) ei(osOs-iO) (31.22)

A peculiarity of dlmension two is that the corresponding vector and axial currents

(31.23)

are related since equation {A5.2) implies:

i'}s")'J'o = -f:;.u/'Y..., , (31.24)

==> J; = ~f.IJ.IAJV • (31.25)

The free massless fermion propagator is (equation (5.58)):

(31.26)

Let us now decompooe the massless fermion into left and right moving components
~(1 ± 7s)'\b, !.};[1 'f l'S] (see Appendix A5.4). We denote by 1/J~, ~z the correspond-
ing components: In the represeotation in wbicb 'ls is diagonal we write

!/!= (t) . .];= (~:) (31.27)

With this parametrization the vector current becomes

(31.28)

---------------------~·IA<WI'·t.'l>t.'.J•'J<'I~· JII,J'I,l!. 'l"tli"""=

694 Two-Dtmensional Models and Bosonization Method 31.2 f
lo the complexc notation (31 .19) the a.ctíon (31.21) becomes

(31.29)

In the euclidean representation the left and right moven; bewme analytíc and anti-
analytic fields. The corresponding propagators and t.;
then reB<l: t.;,
bt = (,P+(i,z:)l/1+(0)}- _..2...,
211"2
(31.30)

b~ (i/1- (i, z) 1/J- (O)) (31.31)

The .P:t correlation funclions. Let us calculate the 2n-point correlation functioiU! o{
1/1:~., i/1±. From Wick's theorem for fermions (equation (1.96)) we obtain

( fi if,+(:r;)l/!+(xl}) =detó;(z,
•~l
zj) = (-1)"
-
2'ft'
det ~-1 1
Lii- Z;
, (31.32a)

( fi ~-(:r;)l/!-(x:J) dett.;(z, zj),.. ( 1)"

;11" det
1zj ,
X._ (31.32b)
·-I
where here z; stands for the pllir {z,, f;}.
The r.h.s. &e Cauchy determinant.s whlch satísfy the simple identíty:

(31.33)

and complex conjugated. This ís the basic relatíon with allows to relate boson lilld
fermion theories in two dimensions.

91.!!.1 Bilinear operotor correlations and boton-fermirm oorrespondence

Let us now consider the two composit.e operaton;

(31.34)

linear combinatiollB of the scalar and pseudoscalar operators:

(31.35)

From the form of the a.ction it is clea.r that only correlation functions of a product of om
equal number of "+ and "- fields do not vmúsh (conservation o( chil"ality). FUrth~nnore
they (a.ctorize into the (orm:

(-1)" /fi~+ (z;)l/!+ (x,)) /fi óf;_ (z;)iJ>- (zi)) (31.36)

\t=l \~l
where aga.in :r; st.o.nds for the pair { zt, !;}.

~· Jll.l't.l:'. '1'\R"-=--------------------·~·U.,WI•!.hk' ~'i•'t·~ J

31.2 Two-Dimen.rionnl ModeZ.. and posonization Method 695

We can now use the identities (31.32,31.33) and lind

2"0.9 1x,-x,l'lx:-xW
(II•=l"
0+ (x,)o_ (x,)
·) =
(l)
2" n.. lxi xjl'
i
(31.37)

The r.h.s. has shott distance diwrgences assoclated with the required additive renormal-
ization of the ""+" _
2-poínt function.
Comparing with equat.ion (31.14) we díscover an identity between awrages witb the
free massless ferrnion Md ·boson actions:

for ,.• = 411". This equation can be translated into a correspondence betwet.n operatcr.;

27rcr,;,(x) ,_. J,l •..

[-"''~,.(~>J., = Ae,;,;.,lb.,(z), (31.38)

with the defi.niticm (31.7). This remarkable relation between local theories of bosons
ll.lld fermions is speeiol te two dime:nsions: in higher dimensions spin degrees of freedom
distinguish between bosons and fermions.

91.!!.2 The massive jree fermion and the sine-Gordtm model

Let us wnsider the po.rtition function Z.;(M+,M-)

(31.39)

with the act.ion

(31.40)

lfW~:exp!llld Z.p in pOWI!O's ofthe sources M±, the term of degree n in M± in the expansion
is an e.verage with the action (31.21) of the form

(31.41)

Evaluatíng the averages we obtain e.n expansion of the po.rtition function Zv(M) of the
form

(31.42)

We ca.n also direetly use the correspondence {31.38). The term (31.4.1) is replaced by

(31.43)

---------------------~·IA<WI'·t.'l>t.'.J•'J<'I~· JII,J'I,l!. 'l"tli"""""""

696 Two-Dimensional Model.< and Bosonization Method 31.2
2
where ( is the renormaliza.tion constant (31.12) for K = 4r.:

(=IL/A~!L/( A, (3!.44)

wíth the definition (31.7).

We ca.n now surn this expansion in the boson tbeory. We find that the partition
function Z-.,(M) is identical to tbe partition function of a boson lield 'I' wlth the action

(31.45)

lf we tBke the constant M.,. = =

M _ M limit we find Mother remarkable correspondence:
the :free massive fermion theory is equivalent to a particular case of the sine-Gordon
model. Tbis relation will be discussed in Sectíon 31.3 and generali2ed in Section 31.5.

Sl.P..S The gauge ín110riant f•rmion dderminont

Let us now consider the !ermion action in a.n externai gauge IX:Id B":

(31.46)

Because the vector a.nd a.xíal curreot are related, the ci8S5ical action (31.46) is invariant
not only under phase gauge transformations but also under chiral gauge transformations:

(31.47)

lndeed this transformation generates the tenn:

(31.48)

which is cancelled by the tra.nsformation:

(3.1.49)

The lield B" is a. gauge field for two sets a f ga.uge transformations. Since it has only two
componeots it can be completely eliminated by ga.uge transformations from the dassical
action, whlch is tbus equíva.Jent to a free lield action. lodeed let us parametrize B., as

B,.(x) (8,.x(x) + i<,..B.<p(:r)], (31.50)

which ímplies

(31.51)

andset
1/l(x) e'l>:l•l+n,(•)J 1;'/(x),
(31.52)
{ ,P(x) = t,il(x) e•!-~!•)+·wp(•ll.

The action (31.46) then becomes

(31.53)

~· Jlt.J•t.l:'. '1'\li•"=_,_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _,~,u.,wt·!.hk' ~'i•'t·~ J

31.2 Two-Dtmensional Models and Bosonízatíon Method 697

Fig. 31.1
--·-0-..'-
The one-loop connibutlon
Q
to the curren1. 2-point function.

Quantum theory: lhe anomaly. As we have more generally discussed in Sectíon 20.3
the chiral ga.uge symmetry ís broken at the quantum leve! by the chiral anomaly. Here
we will recover the anomaly by a direct calculstion. In Appendix A31.1.1 we show the
consistency of the reo;ult obtained below witb the general iorm derived in Section 20.3
(equation (20.106)).
The integral over the fermions lieids generates a determinam which has an expansion
in terms of one-loop díagrams. In two dimensions only the term qu...:lratic in B~ shown
in figure 31.1 is divergent.
To regularize the determin!lllt we use Pauli-Vill1U'5'S type regularization, introducíng
a massive spinorial boson field 4J:

With this ...:ldition the theory ís 6nite (and tbus no longer anomalous) but the q, mass
breaks chita! symmetry explicitly. l.et us parametríze the 2--romponent gauge lield in
terms of 2 scalar fields as in equation (31.50) and perform the corresponding gauge
transformations {31.52) both on if,, f/; a.nd if,, q,. The action beoomes

S ( if,', !/!', B'",4>') =-f d :r [,j;'(:r)íilf/;'(:r)- ~'(:r) (i!+ M e"~•·*>) ,P'(:rJ].

2 (31.55)

We now integra.te over the fie!ds ~', </!'. The result i!; a. functional :D(IP),

(31.58)

which has to be evalua.ted for M la.rge- Tbe bosonization technique is useful in this
respect. Indeed :v- 1 ('1') can be writr.en as a rermion integral

In Subsection 31.2.2 we have derived a fermion-b060n equivalence. We can thus repla.ce

the fermion average with the action (31.40) by a boson a.verage with the action (31.45)
R.lld M± = e± 20'1'. Calling r1(x) the SG lield we finda modilied SG model (31.45) with r1
replaced by 't'J +V>/ ,fii in the interaction:

S(.?, IP) =f 2
d :r { ~ [a,..?(x)] 2 - ~A cos ( ,/41;~ + 2'1')}. (31.58)

We now translate (J setting .9 +'P/ ,fii ~ 11. The action becomes

(31.59)

----------------------~·IA<WI'·t.'l>t.'.J•'J<'I~·ni,J'I,l!.
696 Two-DimensioMI Model$ and Bosonization Metlwd 31.2

In the targe M limit the ll-field becomes very mfl&'iive and h"" tbus vanishing fluctuations
around d(x) O. At leadíng order we c.an set. {i= O and thuslind the finíte resu!L

(31.60)

Since V( 1P) does not go to a constant for M large, the chiral ga.uge symmetry is broken
by qua.ntum fluctuatíons: this is the simplest exarnple of a cbiral anornaly.
The adiou (31.46) is thus equivalent to

S(f/>',.,P,B,.) -f d~x [.j;'(x)IJ.P'(.r) + 2~ f d x(ô,.,.,(.r))

2 2
} (31.61)

91.~-4 Current wrrelation ft<nctilms

Note 6r.Jt that tbe expressíons (31.32) a.re singular when x, _, x:, therefore they cannot
be used to define the cunent without a short distance regularization. We can avoid thís
problem beca.use we note that in the action (31.45) the gauge field acts as a source for
tbe current: Differentiating with respect to B~ yields iJ~.
Current 9-point ft<nction. From the result (31.60) expanded at second order in tp and
the relation (3l.Sl) we obtain tbe current 2-point function

(í~(k)Ív(-k)) ; c~ . "-
k~.~·)'
(J.(x)J~(O)) =- 2,.!"'2 (6.. z"':~·),
expressions proportiona.l to the 2·point functions (31.17b,31.18b) of tbe bosonic current.
Note that tbe second expreosion is a distribution singular at :c: O where it is ambiguous.
Using then eqm•tion (31.25) we obtain the axial current 2-point function

(í;(k)í;(-k)) = ~~'
a result identieal to (31.17a). In the spa.:e represents.tion both cur-rents are proportional
but their Fourier representations ru-e dlfferent: we see that only tbe vector current is
conserve<!.
Current ínoertion. Let us now calculate u% conelations with one current J" (x) in-
sertion. In the transformation (31.52) the operators u,.(~) become D'±(.r)é 2•'<'(%). To
obtain the current illllertion we expand at 1írst order in tp. T!Us yields a fs.ttor

Differentiating with respect to B,.(:c) we finally obtain

(31.62)

~· Jlt.J•t.l:'. '1'\li•'-=--------------------_..,~,u.,wt·!.hk' ~'i•'t·~ J

31.3 Two-Dimemional Model.s and !Josonization Method 699

This expression is proportiona.l to the boson current insertíon J;(:x) for "• = ±;14;. ln
the same way the axial current Jt
is proportiona.l to J{!:

(3L63}

The second result i» not surprising: a translation of op '{) ~ '{> + (J multiplíes e±;,..;;., hy
e±',t<;;9 . To Jff. is associa.ted the transformation <7±- "± e± 2i 9 •

31.3 The Sioe-Gordon Model

The síne-Gordon (SG} model is defined by the action S(tJ)

(31.64)

where t7[:x) is a sca.lar boson field. Note one can choose o 0 > O without loos of generality.
Depending on the question ''"' want to ir;~vestigate we will sometimes normalize tbe SG
field dilferently, setting 8 1;.19 and thus

witht=n2 • (31.65)

Tbe model has been extens!vely studied. The classica.l equations of motion are completely
integrahle. Tms allows to obtaio fin.ite energy solutions of the real-time equation of
motions, solíton.s, and to lnfer the ""miclassi<:a.l spectrum. This int.egrability sutv:ives
quantization and thus, for example, the spectrum and the S· matrix can be exactly
obtained, confinning th« remid..,.,ica.l analysú;. The relevant techniques, however, are
outside the scope of this work, and wiU not be presented bere. We merely want to study
the model for some of its a.lgebraic and RG properties.

91.3.1 P•rturoativ<J ezpanrion

For" small tbe field 9 fluctuates around one of the IIUnima 8 2n,- of- cos8. We can
c:hoose one of them, 8 "" O, to eJ<pand perturbation theory because tbey are ali equivalent
and cannot be connect.ed (SCê Section 40.3). Note, however, that the degeneracy of the
classical IIUnimum is responsihle for the existence of solitODB. Expanding cos (J in powers
of O we SCê that the 8-field is massiva with a mass al/2 + O(K~).
Tbe model is super-renormalizable. From the discussion of Appendix Al 0.2 we know
that divergences, in the perturbative ex!JW>llion in powers of "• Brim only from tbe gelf.
contractions of the interaction term ros 9. Tbese divergen<:es have actuolly been ca.lcu-
lated in the Saction 31.1. Using equatlan (31.12), we can set:

e:i:íO(r-) e%1~t,.(Z') = z~l'l [e*i6(=-:)] nm. ' (31.66a)

ao= z; 112
a, (31.65b)
with z6 (A/~-<l-•'m~J, (31.66c)

in whic:h i' is the renorma.lization scale.

-=---------------------,~·IA<WI'•t.'l>t.'.J•'J<'I~·ni,J'I,l!.'l"tli""'=
700 Two-Dimensiono.l Modei. and Bosonization Method 31.3
f
91.3.2 RG equ.ations
The theory is Sllper-renormalizable ano
thus the }3-funttíon van.ishes. The field
z•.
e±iB(z) l
and the coup!ing COD5tant 0: RG functions are given in terms ar Wc find
l
17(t = r<2 ) = J.iôâJ.i ln Ze(t, A[J.i) = ..!... (31.67)
2>r

The renormalízed n-point correlation function W!n) of e±••(•l thus satisfieS the RG equa-
tion:
[ e n (t l +21J(!)o
J.iÔJ.i +2'1 1
80
J
ô W (n) (p.,t,a) =D. (31.68)

Using the dimensional relation

a a a
"ai' +2a 00 +Piap, =2(1-n),

we obtain the sca!ing relation:

W(n) (p;) = 0 tJ(B#-•lal-., Fín) (p;o-b/(8•-t), t) . (31.69)

In particular, for the 1-polnt function and the ma.ss scale or fl-ma.ss we find:

(31.70)

In these relations we soe that two values of t ,.2 play a special role:
(i) For 1<2 4w the propagator (31.15) beoomes singulll.f at short distance, in the sense
of distributions (or UV divergent in momentum spoce). Therefore the term of order a 2
in the expansian of the SG partition function requires a new additive renormalization.
This yjelds ao additive renormalization of the &ee energy proportional to cr2 •
Also the mass scale m9 and (e'") become linear in o-, results which have slmple in-
terpretations in terms of the correspandence (31.45) witb a free massive fermion model.
lndeed m 0 "' m,p = o and

The UV div.rgence is the sllJile that we bave discusoed abow since by differentiating with
respect to o we obtain the 2-point function at ,z.erc.-momentum. lt leads to a logll.fithmic
correction to the linear behaviour. We sball further discuss this relation in Section 31.5.
For rc.2 > 4,- generically the leadlng contribution to W!l) comes fram short dista.nce
effects and remàlns linea.r in a.
(ii) For 1< 2 = 811 the quantities (e'9 ) and me in equations (31.70) vanish identically for
a small. Let us find an interpretation of this result by calculating the dimension of the
aperator cos9(:t) and more generally of c~l9(:r):

(31.71)

Therefore these operators, which are relevant fot r< small a.nd give a mass to the 6-lield,
becorne irrelevant beyond a finíte value af 1<:

(31.72)
31.4 Two-Dimen8ional Models and Bo•onizotwn Method 701

For > 8>r the la.st operator cos UI also becomes irrelevant, which e:xplains why no ma.Ss
1< 2
is genera.ted. At 1<2 = 81rtbe interaction f d 2 :rcosll(:r:) is masgin!ll. Tra.nslated into field
theory language this means that the theory is just renormalizable; for r< 2 > 8,- tbe tbeory
is no longer renormalizab!e. We sball dil;cuss in Chapter 32 in more detail this peculiar
transition. Note, however, that this analysis is only valid perturbatively in o since at
2
rt = 8,- new divergentes are genemted at each arder in o and thus the dimension of
operators ca.n be modil\ed.

31.4 The Schwinger Model

We now consider two-dlmensional QED, with one Dirac fermion coupled to an abelian
gauge fie!d. The bosoni.,.tion rechnique will allow to solve the massless model exactly
and will provide us with some interesting inforrno,tion about the massive model.

S1.4.1 The mas•leso mode!

The corresponding action re..ds

(31.73)

This model, first discussed by Schwinger, exhibits the simplest example of a chirol
anoma!y, illustrates both confinernent (se<! Chepter 33) o.nd spontaneaw chimlsymmetry
bl'"Eaking in 2 dímensiollli.
The field theory is super-renonnalizable by power counting in a cavariant gauge. The
only divergent diagra.rn corre;ponds to the one-loop contribution to the gauge field 2-point
function (figure 31.1) a.nd comes from the e:xpanoion of the fernúon daterrninant which
W1l have alre..dy discussed in Subeection 31.2.3. In Subsection 31.2.3 we have shown that
the fermion pa.rt of the action with B~ = eA,., a.fter the ga.uge transformations (31.52)
and with the para.metritation (31.50), is equivalent to the free field action (31.61). We
have here simply to ..dd the contribution coming from F;v
l 2 1 ""
4F,.."' 2.?1J,.<P<TÔ,.'{).

In terms of the lields i{l, o/l, rp the Schwinger model action tben takes the free fi<:ld forrn
(in the gauge 8,.A~ = 8 2 xfe O)

(31.74)

Let us calcula.te the '1'-lield propagator in momentum space

ll..,(p) = ~ (,.+1e2/11"- ~)' (31.15)

This expression shows that the scwar rp-field propagates two modes corresponding to a
positive metric neutra! massive field with mass

m=ef..Ji, (31.76)
702 Two-Dimensionai Model.s and Bosonization Method 31.4
r
and a ma.ssless mode with negative metric. The appearance of a non-vanishing mass is a
direct conseq uence o f the chiral anomaly.
Final!y, using the representatian (31.50) ""' can calculate the ge.uge field transverse
2-paint functian. We find

~!~)IM)
c•"' (~ "" -p,.p. I p ') r +m''
1 (31.77)

a result we verify in Appendix A3Ll.2 by a direct one-loop cakulation.

'Ne can now also bosonize tbe free fermion 1{1, t/J', introducing a free massless boson {).
The action becomes

(31.78)

We then trans!ate ij, {J + 'f' /..fo - {}. The action becomes

We tben integrate over 'i' and fmally obtain

(31.79)

It is remarkable that the fennion cantribution just cancels the massless boson 6eld,
leaving only too ms.ssíve boson. Starting from the representation (31.52-) it is easy to
verífy that, in these transformations the chiral components o* of tbe neutra! composite
field ó/>'1/t are mapped to

a±(X) ÓÍ>:f(:r)1/J±(:r) ,_. ~e±'v'ii"l•l. (31.80)

Remark. &fore integration over '{! the action is fonno.lly invnriant when we translate {)
by a const.ant. Therefore tbe form (31. 79) corresponds to a choice of boundary conditions
on the invel'se oftbe Lapla.c<? operator. A different choice leads to the replacement {)(:r).....,
.?(:r) - {J=· From the correspondence (31.&1) we """ that in general space reflection
symmetry is then broken, except for l'J.., = O (mod ~ /2). Finally tbis modification
a
fonnally corresponds to adding a topalogical term proportional to 2'{! o: ,~.,F'"' to tbe
action density, i.e. a constant electric field. For a diBCussion of Lhe physical effects of such
a modüication - refer to the literature.

91.4.2 Neutml co=lation functúms: CJJnfinernent and chíml3ymmetrll /m;aking

We note that tbe gauge field (31. 77) 2-point functjon has no cut corresponding to fermion
intermediate states. Thi• i< a sign of conjinement because it mel'llls that no cha.rged
particles are emitted by the neutra! particle: The electromagnetic forces between partides
of opposite charge are strong enough to prevent the separation of charged particles: these
cannot be abserved as free particles (for details see Chapter 33).
This observation is conlirmed by the form of a±(:r) oorrelation functions. lndeed tbey
can he calculated with the massive free action (31.79) where no massless field appears.
31.4 703

Let us now calculare the EXpectatíon values of <'±· From tbe correspondence (31.80)'
and tbe equatíon (31.10) we obtain

Thkíng into account the definitíon (31.7) we ftnd a finite result

e-r e-r
,.m "'> (,b(x)w(x)) -m. (31.81}
4 271"
Sínce ~.p is a composite field which ís not chiral invariant, the non-vanishing result shows
that global chiral symmetry is spontanrous(y broken. Note that spontaneou• symmetry
hreaking wíth ordering is here pos:sible because the electromagnetic interactíon generates
long range forces.
Let us uow calculate from equation (31.10) the 2-poínt functíons of a±

(31.82)

lt follows

(.,l;(x)w(z)~(O)I/1(0))
2
= (ti:.P) cosh(4lrt.(o:, ml), (31.83a)
(~(xhs.P(x)~{O)wi>(O))
2
= {'Íit/1} sinh(4>ró.(x, ml). (3L83b)

These eJ<presslons have o;ev,ral rema.rkable properties: The 2-poínt functions have only
singularitíes assodated with the massive field. lf we expand the exponentials in pawers
of the propagator and Fourier transform we find that the expression (31.83b) h as a pole
at k 2 """ -m 2 and cuts at k 2 = -(2n 1) 2m 2 , n > 1, in momentum space, while
the expression (31.83a} has cuts at 1<2 = -(2n) 2 m 2 • Only the masoive neutral boson
appears in the intermediate states but no charged fermíons (the confinement property),
and moreover the boson ís a pseudoscalar sínce ít appears as a símple pole only in the
frts.P 2-point function.
Finally the propagator has a short distance behaviour given by (31.6), sin<:e L!.(m, x) ís
a function only of mz. Therefore the 2-point functions (cr+(x)a+(O)) and (<1-(x)cr-(0))
go to zero as x~ while
1
(a+(x)a-(0)) ~ 4112:::2 •
i. e. Hke in a free massless fermion theory. This property rellects the asymptolic fror:dam
(at Iarge momentum) of super-renormalizable theories.
These are ali properties _, aJ.o expect in the true physical world with quarks a.nd
gluons.

:il ..i.S The mt18sive Schwinger model

Let us now briefly discuss the elfect of the arldition of a fermion mass tenn. Of course
chiral symmetry ís no longer an issue since the mass term explicitly breaks chiral sym-
metry. However there ·;, still lhe problem of confinement: Will charged particles appear
in the spectrum?
We thus oonsider tbe "ction:

S (~, .p, A,.) =f d2

:t [~FJ.(x)- ~(x) (~ + ie4 + M) ,P(x)) (31.84)
7()4 Tum-Dímensional Modelo and Bosoni.zation Metlwd 31.5

We now perforoi the transformntions (31.47) and {31.50) (B~ e.A~) and find:

(31.85)

The model is no Jonger &re and cannot be solved exactly. However the action cnn be
further transformed, as showu in Subse<:tion 31.2.2. We ca.U {J the boson field assocíated
by bosoni2.a.tion with ;j/,1/J'. We then use the equívalence betw.,.,n the actions (31.40)
and (31.45). The fermion a.:tion is replaced by tbe artion (31.58)

We now translate f! setting fi+<p/ J1i ~ iJ. After tran.slation the a.ction beromes quadratíc
in 'I'· We can thus integrate over 'P and find a "massive" SG action for f!:

(31.86)

with the correspondenoe (31.80) 1>11d

no= MA/1t·

Pkysical conse.quences. We see in ~>etion (31.86) that at least for M small the result
obtained for M = O survives: The spectrum consists in a massive neutra! boson of mBSS
squared m 2 + O(mM). No cha!ged pa.rticles app€8.! in the spectrum. Note on the other
hand that for M large we expect "' non-re!a.tivistic BnAlysis tll ha valid. Tben we have a
set of fermions inter~>eting with a one-dimeru;jonal Coul01nb potential proportional to r
if ris the dista.nc:e between charged pBrticla!. This potential grows at large dista.nce Md
thus charg<ld particles can never be separated.

31.5 Tbe Mas5ive Thirring Model

In Section 31.2, romparing equations (31.13,31.37), we have discovered identities bet.w.:en

&ee ma.ssless fermion averages of po....,rs of ij,p and averages of exponentials of free
m8S5l= ~n fields. This has led to tbe correspondence derived in Subse<:tion 31.2.2
betw.:en the &ee ma55ive fermion and tire SG model for r;.2 = 411'. The Sa!De identities
wiD allow us to establish a relation between the general quantum SG and tbe mossive
Thirring model. The Thirring model is described in terms of Dira.c fermionB with an
a.ctlon:
(31.87)

where
(31.88)
For mo O the Thirríng model retains the U (1) symmetries (31.22) of the &ee massless
a.ction. Note that the interaction is tire only local interaction posslble since we have only
4 fermion variables. Fina.lly power counting shows tbat the model is renormali7.able in
two dimensíons and tbe coupling ronstant g thus dimensionless.
31.5 Two-Dimef!SÚ>JUJ/ Models and Bosonization Method 705

The ma.ssive Thirring model can be mapped onto the general SG m.odel. The corrt>-
spondence bet.ween the two modoilii cs.n be sumrnarized by the r<>l ..tions:

1+g/1f=41fjK0 , (31.89a)
1
J,. ,__. z;;:c,..â.6 (31.89b)

ófnjn---> ~ cos fi .
1(
(3L89c)

The chiral invariant model, ma = O, is mapped onto a free boson theory, as we h"""
alreBdy discussed in Section 31.3, and thus exa.ctly soluble.
We derive below thi5 correspondence, usillg tbe bO&Jnization identitié< established in
previous sections.

91.5.1 Bosonizatian: the oine-Gon:lon modcl

The fum step of the bosonization is to introduce a vectcr field A, II.Ild rewrite the
interaction term "" a gaussian integral <Mlr A,.:

(31.90)

The fermion a.ction becomes a quadratic act,ion of charged fennions coupled to an abelian
gauge lield:
(31.91)

Basonization. The a.ction then dilfers from the massive Schwinger model action (31.84)
only by the kinetic gauge field term. As we discussed in Subsection 31.2.2, the action
CIUI thus be replaced by a purely boeonie a.ction. Parametrizing tbe 2-comj)onent field
A, in terrns of 2 scalar fields x and 'I' as in (31.00) we find

f d2:rA!(:r) =f d 2:r [(â,.;t(zl) 2 - 2

(ôl''l'(x)) ]

AftE:r the gauge transformatioru; (31.52) the act,ion (31.91} then becomes:

S(x,'f'.ííf,I/J'l =-f d2 :r!jí'(6'+moe~;.,• .,)I/J'

+~f d':r U {â,.;t)2 G+;) (8,.'1')·]. (31.92)

wbere we have taken into account the anomaly.

The field x decouples from neutra! correlatíon functions. The kinetic term of the field
<p has tbe wrong sign. Averages of products of e±;., when tbe numbers of + and
signs are different are thus IR divergent. Fbrtunately, as we have indicated in Subooction
31.4.1, they always appeax multiplied by fcrwion averages whlch, dueto chiral symmetry,
'Vmlish.
Tbe correspondence between the massi'VIl Thirring and SG models now follows directly
frnm the analysis of Subsect.ion 31.2.2. We have shown tbat the fermion action can be
replaced by a SG a.ction for ,;;2 = t = 41f of the form (31.58)

f d xííf (~ +mue ''""')o,t.' ,_...f d x [Ha,..?)'- m~A cos(~ +2op)].

2 2 2
(3].93)
706 Two-Dimensíonal Models and Bosonization Methad 31.5

In the special case of the !11à55less Thirring I!lDdel mo = O we obtain an equivalent free
b<lSOD theory

(31.94)

The massless Thirring model can thus he so!ved exactly and we calculate a !ew correlation
functions in the next suhsection.
For Tll{J '#O, we change varíables 'I',_.. 9, setting

The integral aver {} is then gaussian,

and ca.n he explicitly performed. We linally obtain

(31.95)

whích est.ablishes the correspondence between the massive Thirríng model and the sin...
Gordon modcl. Comparing with the action in the normalizatíon (31.65) we see that the
relation between parameters is (t ,. 2 )

proving relations (31.89a, c). Sínce the coupling constant t is not renormalízed thls
correspondence establishes tha.t tbe coupling constant 9 is not renormalised eítber and
thus the RG tJ(g)-function vanishes.
The ficld e±>;,p ínstead has to he renormalized, the fermíon mass thus needs a renor-
malization. Witb the regula.rization and renormalization &t mass scnle ll of Section 31.1,
we find:
mo= m(ll/A)•/ID+•l.
The relation hetween renormalized parameters is then:

Let us now brielly discuss the special massless case.

91.5.2 The ma..sl••• Thirrirtg model

The msssless Thirríng model has a U(l) chirl>l symmetry. Addíng a source terrn B,. for
the current J,. we lind tbe effective action S (A.,) for the remaining A~ field:

(31.96)
31.5 Two-Dimonsícnal Modela and Bpson:i.w.ti= Method 707

We can then integrate over the A,. field to obtain the current 2-point correlatian functiori.
In moment.um space

(31.97)

Ali otbet connected correlation functions vanish. If we inste..d calculate the 2-point
cotrelatian function af •,..a.e
with the free action (31.95) for mo= O we find:

(E,.,Ôp/J(k ) <.oôv/J(-k)) '- ,.4,..

+ ( é,..- ~)
k2 , (31.9~)
9
result consistent witb the two relations (31.89a,b).
All..p-field cotrelation functions can be calculated explicitly. In particular the bosonized
form of the ~(:r)"'(:t) oorrelation functions depends only on B(x) and thus ca.n be cal-
cula.ted with the action {31.95) for mo O. The one-point (unction iP(x)IIJ(x) vanishes.
The ~"' 2-point function is given by
{~(x)IIJ(x),J;(O)"'(O)) o; {cos 6(x) cos 6(0)) o:: ,-l,
wJth: ). = t/211" 2 ... f(g + 7r).
31.5.9 Díscwsion
R G pruperties. The ma.ssless Thirring model exhibits a scaling behaviour for ali values of
the coupling constam g. This is a consequence ofthe v&llislúng of the tl(g)-function: This
model provides llll example of a line of IR fixed points. Since the ~IÚ 2-point function
decrea.ses at la.rge dist!Uice the chiral symmetry is not spcmtaneously broken, although
the correlation functions decrease only algebraicelly.
As we have discus.ed in the case of the SG model, at the w!ue 1<2 8..-, i.e. 9 - w /2,
something new happens, the mWlll term becomcs irrelevant. This suggests the possibility
o( a phase tra.nsítion, a qucstion we examine in the next chapter. The tr&llSition point,
however, is neither in the perturbative domain of the Thirring model nor of the SG
model. The properties of the bosonization metbod suggest that it should be possi ble to
map the SG model with ti' = s,. on a rree model by doubling the number of ferrnioru;.
We indeed exhibit in next section a two fermion model whicb is lllso equivalent to the
SGmodel.
Mass SJ>I'Cirum. The corrcspond<:nce between the Tbirring end SG model in particu!Br
allows te easlly calculate physical quantities in tbe SG model for g small, i.e. for t close
to 4,., or in the Thirring model for g large. Morwver we see tha.t for g > O, i. e. t < 471,
where tbe potential between fermions is attrective, tbe spectrum of the theory consists
at least in one Dirac fernúon and a boson bound state, ccrresponding to tbe fie!d of the
SG model (the fermlon appears semiclassícally in the SG model as a oolitcn). Actually
tlle exact boundstate mass spectrum is given by:
rn,. 2m(a,t)sin(n"1/16), n"' 1,2 ... <87Th,
wlth 1
7 = - -- - "' __ s_.-_
1- tf81T 1 +2g/1r'
where m(a, t) is the fermion mass:

m(a, t) = /(t)!a•/ 2 +~/ 16•, /(0) = 1. (31.99)

'Y
The Jowest mass n = 1 corresponds to the SG field; the remaining boundstates can also
be considered as collective excitations of tbe SG pa.rticle.
70S Two-Dimensianal Model.. and Bosoní.zation Method 31.6

31.6 A Two-Fermion Model

The deriwtion of tbe corresponden<:f: between the massíw Thirring and the Se models
indica.tes that it i; possible to construct a. funnion model equivalent to the se model,
such tnat tt2 g,. cor:responds to a !ree field theory, by introd ucing two Diroc fermions
and by choosing the 4-point intere.ction appropriately. We now describe such a model.

91..6.1 The model

We consider a generalized mass!eso; Thírring model wlth two fermions coupled through a
current-cutrent inter&tion. The action is:

(31.100)

where the matrlx 9a! has the form

-~(g'+g
g-2 g'-g g'+g
g'-g) '
(31.101)

and the current j: is

(31.102)
This action is the most general renormaliza.ble oction "''ith th., following symrnetties:
(i) A U(l) chiTa! symmetry which prewnts a fermion mass t..rm:

li>"(:r}>-> .,;;,e "'n(:r),

(31.103)
{ oP"(x)- ,pn •''"9.

(ii) Separa!€ fermion number conservation for eacb type of fermions and finally symmetry
between fermions 1 and 2.
Note tbat the model has an additional SU(2) invariance when g f since the int..r-
a.ction can also be written

[g' 1212 + gJ!J! +f (J!J! + J;J~)]'

where ~ and J~ are the U(l) and the SU(2) curre11ts respectively

We have denotad by t7 the three Pauli matrlces acting on the SU(2) indices.
Due t.othe U(l) chiral ínvariance ofa<:tion {31.100}, the coupling g' is not renormalized,
as will be derivai in next subse<:tion. The RG P,functions at one-loop order are (see
Appendix A31.2.2):
13. ""' -2!~ /1<' (31.104)
{ f3t""- -2fgf7r.
The equívalence whh the se model is derive<! below. The correspondence is summBrized
by the relations:

(31.105)

(31.106)
31.& Twa-Dimen~ional Modcls and Sasonimtion Melhod 709

Frotn these expressions we see tha.t for t dose to 8,- and o smo.ll, f and !l are both
sma!L The study of the phase transition is reduced t.o sta.ndard perturbation theory
with fermion 4-point renormalizable interactions.
Since the transítion taking place neo.r g = f O will be discussed extensively in
next chapter, let us here note only that in the fermion !ang1.1age the transítion occurs
between two pbases, a massless chira.l invaríant pha.se as seen in perturbation theory, and
a massive pbase in which the U(l} chira.l invariance ís spontiUleously broken.

91.5.2 Derivotton
The general idea is to separate the a.ction (31.100),

f 2
d x (i/Nl/1'- !!lobi~i~ - f~ 1 -y,...;,2;f?'Y,..P 1 ),

into the sum of two t.erms S = S0 + S1:

So (i/?,1/1") = f x(i/?,,P"- Üabj;j!)'

d
2

S1 (~".1/r") f f d'x;P'Y,.viõj?.,,,p 1
•

The lirst term S0 contains ali int.eractíon tertn5 which have a separate chiral invariance
for oocll spedes of fermions (ínteraction of Thlrring model type) and is treated in much
the 58Jlle way as the massless Thirring action: lt can be transfonned into a sum of free
fennion terms.
The remainder S1 is expanded in perturba.tion theory Jike the mass t.erm in the Thirring
model, and yields the interaction.
We thus write the interl>Ction terms of S0 "" genera.ted by gaussian integra.ls over two
A!.
vector field s The 8Ction then becomes:

where [us]a6 ís a Pauli matrix and a<:ts on the vector .pb. Parametri2.ing the vect.or fields
a.s in equation (31.50):
A!= -epx± -it,.~a.,.,'". (31.107)
and perfonning the correspondíng gatJge tra.nsformatíons (3Ui2) on both fermion fields
we obtain (see equ,.tion (31.92)):

so =-f d'"' [õín~,p• + "+

"9
11 (8 v>-) 2 +,.. + g'
.. ft!J'
(8 v>+)'-! (a -) 2
" g ,.x
_.!.g' (8 ... x+>'] ' (31.108)

(we have omitted the primes on the fermions) which shows thet g' indeed i8 not renor-
malized.
For S 1 , to be able to use directly the identitíes derived for the Thirríng rnodel, it ís
convenient to rewrite the f-term usillg a Fierz tra.nsfarmation (..,.. Appendix A5.9):

{31.109)
710 Two-Dimeruional Models and Bosonizalion Metlwd 31.6

lntroducíng the chiral components of the muss operators

we can rewrite the r.h.s.:

(31.110)

Therefore, in a perturbatíve expansion in powers of f, the integrais over the two fermions
fact.orize and each term ís just the square of the corresponding term in the expansion in
powers of m in the Thírring model.
In the transformatioru; (31."2) the chiral components .;± become

Uslng then identity (31.110) we am rewríte the f term

(31.111)

We now 1!850Cia.te two bosons ti" wlth the fermions ,P, 1/f'. In the expansíon in powers
of f m> see that we can símply replace the quantitíes <1 by their bosonic counterparts.
We then obtain a boson act.ion

Changing varia.bles, far ex!Uilple '/>- ._.. 0:

and integrating over the fields t'J' (and ,.,+, X±l we finda simple equivalem SG model

(31.112)

Comparing with the form (31.65), we see that the SG model has e parameteT 1<

(31.113)

Therefore the value 1<2 = s,.. canesponds to a fr<:e massl.ess fermion theory.
31.6 Two-Dimensional Modds and Bo.snnization Method 711

Bibliographi<:al Notes

Some references on the sine-Gordon model and its exact integrabilíty are
L.D. Faddeev and L.A. Takhtaja.n, Tear. Mat. Fiz. 21 (1974) 160; R.F. Dashen, B.
Hasslacher andA. Neveu, Phys. Rev. Dll (1975) 3424; L.D. Faddeev and V. Korepin,
Phys. Rep. 42C (1978) 1; A.B. Zamolodchikov and AJ. B. Zamolodchikov, Ann. Phys.
120 (1979) 253.
Some references on the Schwinger model and its relation to confinement are
J. Schwinger, Phys. Rev. 128 (1962) 2425; J.H. Lowenstein and J.A. Swieca., Ann. Phys.
(NY) 68 (1971) 172; A. Casher, J. Kogut and L. Susskind, Phys. Rev. DlO (1974) 732;
S. Coleman, R. Jackiw and L. Susskind, Ann. Phys. (NY) 93 (1975) 267; S. Coleman,
Ann. Phys. (NY) 101 (1976) 239.
The correlation functions of the massless Thirring model are given in
R. Kla.iber, Ledores in Theoretical Physics lOA (1967) 141, A. Ba.rut and W. Brittin
eds. (Gordon and Breach, London 1968).
The correspondence between the massive Thirring and the sine-Gordon models has been
established in
S. Coleman, Phys. Rev. Dll (1975) 208 8.
For the bosonization of tbe SU(N) Thirring model see
T. Banks, D. Hom and H. Neuberger, Nucl. Phys. mos (1976} 119.
For furtber references on bosonization techníques see
E. Witten, Commun. Math. Phys. 92 (1984) 455; I. Affieck, String, Fields and Criticai
Phenomena., Les Houches 1988, E. Brézín and J. Zinn-Justin eds. (Eisevier, Amsterdam
1989).
For a discussion of the so!itons in the sine-Gordon model see for exarnple
R. Rajaraman, Svliwns and Irutantons (North-Holland, Amsterdam 1982),
a.nd references therein. The spectrurn and the S-matrix of the SU(N) cbiral Gross-Neveu
model are discussed in
J .H. Lowenstein, Reoent AdoonC~<s in Fie/4 Theory and S!atistical Mechanics, Les
Houches 1982, J.-B. Zuber and R. Stora eds. (Eloevier 1984).

Exerclses
Ezen:ise 31.1
Study in the 2D free massless scalar field theory the short distance expansíon of the
product of operators tlt:•l'f'(z:t) e•~:~\à{x:~).

Eurd.se 31.1!
Consider 2D QED in the case of N identical charged massl""" Dirac fermions. Denotíng
the rermion fields by ..p ,ib we write the action

{31.114)

Bosonize tbe model and identify the symmetry breaking operator with non-zero expec-
ta.tion va.l ue.
712: Twa-Dimensional Model5 and Bo•~nuation M.thad 31.6

Exercise 91.!1
'Ib expllcitly verify tbe çonfarmal ínvaria.nce af the free massleso bosan 2D theory calcu-
late the two-pdnt functioo ( e"•"(•,v) e*•' ~(o,o)) in the cylindrical geometry, i.e. with a
finite si:r.e L and periodic boundary conditions in one direction (see also Section 36.~ for
the pro blem af the zero mode),

Exercise 91.4
Find the most general analytíc function f(z) which satísfies an identity of the form
(31.33)
(31.115)

One assumes f'(O) 'f O.

A31.1 Twa-Dlmmsional Models and B,o•onizatúm Metlwd 713

APPENDIX 31
A FEW ADDITIONAL RESULTS

A31.1 The Schwinger Model

In tbis section we examloe the role of the anomaly in the Schwinger model from two
other point of views.

AS1.1.1 The generol anomaly

The anomaly of the axial currcnt J% = i;j.'Ts"'~.P is (equation (20.105)):

On tbe other band the vector current J• = fyi'.p is exactly conserved:

We bave observed that in dimension two the two currelb are related (equation (31.25)),

and thiiS
ie
Õ1J.,- ÕvJ• = -;F•v.

The gauge field equation of motion yields:

ieJ,. + õ.,F., = O.

Combiniog these equatioos we get:

This equation shows that the current J• and thus the curvature F,..
are free lields of
mass m = e/.Jii, in agreement with the result derived in Subrection 31.4.1.

A31. 1.2 Th.e Schlllin9er model: one-loop rolculation

We know from the general analysis that we can regularize the fermion deterrrúnant while
preservíng the QED gauge invariance. However then we ca.nnot preserve in general the
chiral gauge ínva.ríance. For example in dimensional regularizatíon the anticommutation
properti""' of "YS together with the relation (31.24) cannot be ma.iotained (see Subrection
9.3.3). Let us here explicitly work out the one-loop contribution to the A. field 2-poiot
function {figure 31.1) in the Feynrnao gauge. Using dimensional regularization whlch
preserves non-chiral gauge invariante and giving temporarily a nu•ss M to the fermion
field to avoid IR ptoblems, we find:

(A31.1)
714 Two-Dimensional Model:; and Bosonizatíon Methad A3L2
It is """Y to verify tbat tbe dimensiono.lly regularired one-loop oontribution is transverse
as expected. Setting:

(A31.2)

and taking tbe trace or botb rnembers we abtain:

(A31.3)

The r.h.s. conto.ins a term proportiono.l to M 2 wbiob is UV oonvergent a.nd whíob goes
to zero wíth M for k i' O. The second term is proportional to d 2. In the d 2 limit
only the divergent pa.rt af the integral survives:

(A31.4)

The gauge field ínverse 2-point function at one-loop arder follows:

(A31.5)

a result consistent wítb tbe exact expressíon (31. 77).

A31.2 The SU(N) Thirring Model

ln Soction 31.6 we bave shown tba.t tbe SU(2) ínvaria.nt Tbirring model is equivalent ta
a booonic model, the SG modeL It is clea.r from the derívation oi this property that more
general models ca.n be bosonized. Let us here consíder in parUcula.r tbe SU (N) Thírrlng
model, a model des<::ribed in terms of N Dirac fermíons, with the action:

S (.,,t/.1) =~f d'x I~ ·@t/J- !J (i/> ·-r~t/.1)2 !9\ih~.;!~'l'~.P'] (A31.6)

=-f 2~((g+Nf)J~~+gl;J;)],
2 (tJ>·ilt/.1-
d o: (A31.7)

where J2 and 1; are the U(l) a.nd SU(N) currents:

~ i/>. 1'p.tP' J'; = ~·-r"ti;.pl
Thls model bBS a. SU{N) x U(l) vector together with a cblro.l U(l) syrnmetry. After a
Fierz transforma.tion the second interaction term becomw:

We then recogníze tbe actlon of clúro.l ínvariant Gross-Neveu model. Strictly speakíng
the Gl'OS&-Neveu model rorresponds to the special case f = O but suob a model is
not multipli<:atively renormalizable. Another cunent interactíon has ta be arlded for
reoonnalizatioD purpose.. Dueto the global chiro.l inva.riance of tbe model the coupling
A3L2 Two-DimensioJWI Models ond Jlosonizatíon Method 715

constant 9 + N f assoclated with the current f' is not renormalized. The remaining RC
/l9 -Function in d 2 + < at one-loop arder is then:
N •
!3. ""<g- -;9 ' (A3L8)

showing that the model is asymptotically free and has a non-perturbative spectrum for 9
positive and small. For 9 < O instearl the model is IR free: the fermions remain rn&SSless.
The model Cllll be bosonized like the Thirring model to yield a generalized SG model
of tbe form:

with
(A3L10)

The physics of the SU(N) model ís again the physies of the chiral phase tra.nsition. The
fermions are rna.s5less in perturbation tbeory but become massive for g > O (where the
for~ hetween fermions are attractive) witb (tiJ · .p) = O. Conllict with the Mermin-
Wagner theorem is avoided bocause the would-he massless Goldstone boson which is
8.SSOciated the U(l) cbiral transformation, and thus with the translation ll' ..... li +e
decouples from the other fields. This ís another exrunple of the Kostetlitz-Tbouless
mecha.nism (see Cbapter 32).
Note that the spectrum bas heen eJtaCtly obtained:

m, 2N . ("")
tx -s,n - n= 1,2 ... < N,
" 2N '
n odd corresponding to fermions and n even to borons. Thís result oonfirms that fermions
become massive.

A91.JU Derivation
The method is a simple extension of the method explained in Section 31.6 in the SU(2)
case. Intmducing the currents i!:

i! =i/Ji'"'II"V>', (A31.11)

(no sum over i) we can rnwrite oction (A31.6) "'"

~9 L: ;p•..,,.,p,"f)..,,.w•] .
r.7-j
(A31.12)

with:
g;j "" g6;j + f. (A31.13)
We then introduce vector fields A~ to rewrite tbe action

s (v;,.p,A~>) ~f d x [L i[J' (6' + i_.t') ,p' ~ !9;,;A~A~- }9L ij,i..,,..p'~-,,.1/J'],

2

t •fl1
(A3l.l4)
716 TtJJO-Dimensional Models and BosonW.tion Method A31.2

with
(A31.15)

Combining the UEua.l U(l) gauge transfonnations on tbe ferm\ons with a parametrization
of the vector lield.:

.p'(x) ,__. e<x'!•)+i"Ts..,'(") .p'(x),

,_l;•(x) ..... ~· e-••'<•l+''""'' (•!,
A~"" -a,.x'- ;,...,a""''.
wefind:

s (t/J,.p,..,,x.) j d'x[-•P.íi.P' + ~u.A.-x.'a,.x.i ! (u•i + 6: ) a,.'P'a,.l{!

- ~9 L -J,i .,<-rs(.,'-.,'l -r.;/);iJ .,-hs!<P'-.,>) 1',..P']· (A31.16)
i.,tj

lntroducing the chira.l oomponent.s of tbe mass operator ~'.P'

çl ~l.Pi.
and using tbe identity (31.111) "" can write

= -2 (e2i(.,'-.,>J u'+ ,r~+ e"(""-"''l ~+ cr~)

~'e'"!"''-.,;) 7p ;J?;j) e-i-rs(.,'-op') 1',. .p' w •
We then expand the four-fermíon term ín perturb3tion tbeory. As in tbe SU('.l) model we
see tbat the average over tbe fermioDS ~;, .P' can be replaced by Bn average aver b050US
{J'. After su1Jlll1ation tbls leads to a SG interaction of the form

-g !; L i.'#-j
cos [ J4i (ri-~}+ 2 (<P' -IP")].

Changing tben variables, tp 0 r-o tJ', setting

2op' + J:ilr17' =9',
and integmting on ali fields but Lhe fl"s"" find tbe e.ction (A31.9).

A$1.!.2 Tlw! RG /3-runction with o. 4-Fermion Cummt Intemction

We lea.ve os an exercise to calculate the RG {3-function in the case of a 4-fermion currentr-
cunent interactíon:

(A31.17)

where g.r.c.~ is chosen to be symmetric in the e:xchange (ab) <-< (cd). Tbe result is:

1t/3u!d = 9oiej giõjd- 9uijd 9ib<j +O (l} • (A31.18)

A3L3 Two-Dimensiorwl Mmkls and Bm•onization Melhod 717

A31.3 Solltons in the Sine-Gordon model

Soiitons rnrrespond to finíte energy solution6 of the real-time classícal equations of m<>-
tion. Snliton calculus should be thought as t.he field tbeory general.ization of t.he WKB
method. Snlitons ha.ve a particle interpretation, the energy of the ooliton being its mass
in the sem.icl&SSícal limit. The sine-Gordon model being ci&SSically integrable, tbe wbole
soliton spectrum can be derived sin<:AO also time-dependent solutions cau be obte.ined.
Tbe semicl&SSicBI results have been confirtned by the exact quautum a.nalysis. Let us as
a simple ex.ample calculate the mru;s a.ssociated with the stati<: solution. The equation of
motion a860ciated with oction (31.64) is a differential equation in the spoce coordinate
:r:
6"=osin9.
Finite energy solutions necessarily COWlect miníma of tbe pOtentÍI,ll. Thus

li= 4 arctan ev'ãs.

The correspondíng energy is then the spa.ce integral of the lagrangian densíty, (after
•ubtraction of the vacuum energy);

which coincides wit.h t.he Thírring fermion mass (31.64) for I small.
32 THE 0(2) 20 CLASSICAL SPIN MODEL

Havíng established In Chapter 31 a few properties of two-dimensional models we now

have the necessary technical background to discuss tbe abelian 0(2) spin model near
and in two dirnensions. As we have seen in Cbaptel' 30, at low temperature its long
di5tance properties can be described In terms of tbe 0(2} non·line,..- a-model. We reca.ll
that the 0(2) case is special because the RG tJ-fun<:tion reduces in the low temperature
expansíon to tbe dlmeJlllional term (d- 2)t and therefore the properties, from tbe RG
point of view, are quite dífFeretrt. In particular in 2 dímensions tbe 0(2) model is not
asymptotically free. The origln of tbi:; difference can be found in the local structure of the
manifold: For N"" 2 the O( N) spbere reduces to a circle which is locally a llat manifold,
i. e. which cannot be distinguished from a straight line. Therefore if we parametrize tbe
spín S(.:t) as:
(32.1)

the euclidean action (30.11) in zero field becomes:

S(&)
Ad-2
2t f dd:<[éJ,.9(.:t)J 2 , (32.2)

i.e. a free massless field action. This expla.ins the fonn of the RG tJ-function. Although
the theory is free, we are not intel'ested in the carrelation functiom of the field I) but
instead af sin 8 ar cosO ar equivalently e-±•6 • Even in a free field theory these fields have
to be renormalized and thus have a non-trivial anomalous dimension for ali tetnperatures
(equ&tiom (30.24b,31.12)}.
The simplicity of action (32.2) however lmds to a mystery: since the field Ois mMsless
the correla.tion length rema.ins infinite for ali t. By contrast a simple hígh temperature
anal:r-;is of the corresponding spin model 011 the lattice SbOW$ that the correlation lengtb
is finite for t large enough. A phase tnmsition at finite tempcrat1Jl10 is necessary to
explain this phenomenoo. Tbe action (32.2) therefore C8llllot represent the long distantes
properties af tbe lattice model for Bll temperatures.
We sha.ll find a clue in the property tbat fJ is a cyclic variable, mnditioo whích is not
incorpor&ted in actíon (32.2) aod is iodeed irrelevaot at low tempero.ture. At higber tem-
pemtures, however, cla>;sica.l configur&tioos with singularities ai isolated poitrts a.round
which 9 wties by a multiple of 21r become important. Tbe action of these configurations
(vortices) can be identified with the energy of a Coulomb g&S. Tbe neutra! Coulomb gas
exhibits a tra.nsition between a law temperatura of bouod neutra! molecuies and a high
temperature phase of a plMma af free charges.
Remarkably enough the Coulomb gBS can be mapped onto the sine-Gordon (SG) model,
mappiog in which the low and high temperature regions of tbe two models are e:xchanged.
Tbis will help us to understand some properties of tbe transitkm, tbe famous Kosterlit ..-
Thouless phase transltion (KT), wbicb separates an infinite correlation leogth phase
wilhout crdenng (tbe low temperature pbase of the 0(2) model) from a fioite correlaiion
length phase. Additional information will be obtained from tbe equlva.lence between the
SG model and several fermioo rnodels, established in Sectious 31.5, 31.6.
32.1 Tke 0{2) JID 0/ruuicrd $P,:n Model 719

In the appendb< M> recall the relntion betwt!en the stntistical medlanics of 1D non• '
relatívistic fermíoru; and the quantum lia.ld theory of ma.ssless relativistic Dir!ll: furnúoru;
in 1+1 dimellSions.

32.1 The Spin Correlation Functions

Let us first calculate the spin correlation functions in d dimensions with the free action
(32.2):

(g e"•'f••l) =f jdO]exp [- A; f d":r(tl,.0) + i~<;B(:r;)],

2
2
(32.3)

with <; = ±1. The method has been explained in Section 31.1. The result is:

(TI é•••<•.))
t"""'l
= exp [-~tA•-• L:;•••i.O.(x;
• .,
:ri)]. (32.4)

in whích Ll.{x) ís the ma.ssless propagator:

Ll.(:r} _I_
(2..-)d
fr d"p
e
ip• for d > 2. (32.5)

.0.(0) diverges with the cut-off A:

The limit d = 2 is singular, as we have discussed in Section 31.1, because the field ís
mnssless (IR divergence).
Remarl.:. To define renormalized correlation functions, we have to introduce a. renor-
malization sca.le p a.nd the oorrespanding coup\ing constnnt t,:

tA •-• t,iJ•-•,
and cancel the divt:rgent terms in the exponential in the r.h.s. of eQU&tion (32.4):

L'••,L'>(x; --x;) ~ n.O.(O) + 2L <;<;Ll.(xi -.:1). (32.6)

t,j i<.j

AE noted in Section 31.1 the fields é' 6<•1, whích are composite fields in terms e>f 11(x),
require a field renormalization Z:

Z = e-; ... ,...'2-"".o.{o). (32.7)

Dimension 2. Let us now examine more dosely what happens when the dimension
approacbes 2 (at fuced cut·off). The propa.gstor .0. has an m divergence. Setting d = 2+c
wefind:

D.(:r) 2--..!..
2>r< 4..
(lnx2 +ln>r+r) +0(.:), (32.8)

D.(O) __!_ + ..!_ In A + terms finite when < - O or A - oo (32.9)

2..-c 211
720 The 0(2) 2D Clrusica! Spín Model 32.1

where "I = -,P( 1) is Euler's constant. lt follows that the sum of the divergent contríbu-
tions in the exponential of equation (32.4) takes the form:

(32.10)

The.efo..,, only the correlatlon functions suc:h the sum :Z::: <; vanishes ha"e a finite non-
vanishing limit when d goes to 2. The other correlation functions vanish at d = 2. Thís
result has a simple interpretation in the 0(2) non-linear a-model: ali non 0(2) ínvariant
correlation functions vunish. In particular

(S(:r)} =O.

This result is in agreement with the ahsence of spontaneous symmetry brealáng wlth
ordering in 2 dimensíons. Note bowever that tbis property does not depend on tbe eom-
pact nature of the 0(2) gToup. Ali correlatian functions invariant under the translation
9(x) by a coostant bave a finite limit. ln particular, as shown in Sectiou 31.1, we can
replace the signs !; by !lD)' set af numbers witb a vaJÜBbing sum.
Let us write more explicitly the limít in the nau-trivial c....e:

(32.11)

This resnlt is rather surprising: although the 0(2) symmetry i.s not broken the correlatlon
functions decay algebrakally at large dmtam:e and the oorrelation length 1s thus inlinite.
In addition the power behaviour depends continuously on the temperature: in tbe RG
sense we have a line of fixed points and thís ís consístent with Lhe property that the RG
tl-function 'I'Mi$hes identically. Applying equation (32.11) to the 2-point function

(32.12)

we obtain the value of the exponent '1'

'1 = !/2ff. (32.13)

It remeins however to understand the absence of phase transition in Lhis model. ln

particular no invarlant relevant operator can be coostructed whic:h would modífy action
(32.2). Since we do not have y>;t ali the necessary ingTedients to examine this question
we postpone it until Section 32.4. Let us finally oote that t > 4'11' implies 11 > 2. This ;,;
an unphysícal result (tom the point af view of the O (2) spin model since it implies that,
in momentum spaee, the 2-point correlation function vanishes at low momentum. In
the corresponding lattice model the correla.tion function is tben doiiÚnated by a regui&r
oonstant (non-critic.U) contribution. t 4ff is thus the a prúJri largest possible value of
t for which the action (32.2) can represent the 0(2) model below Te•
32.2 Th.e 0(2) l/D Classimi.Spin Model 721

32.2 Correlntion Func:tions in a Field

Let us now edd a magnetic field terw in actian (32.2):

A"~>
S(O) = ~t- f {ddx ~ [ô~6(x)J
2
hcos9(x)}. (32.14)

We recognize the SG action studied in Section 31.3. Since ""' know the 2 RG functions:

p(t} = (d- 2)t' <{t) t/27r. (32.I!i.)

we can use the exp,.,íons of Chapter 30 !<> find t.he scaling foT1Jl of correlation functions.
For d > 2 the two functions Mo(t) and {(t) defined by equations (30.30,30.31) are:

M 0 (t) exp [- ~ (' ( (t') dt'] exp (- ~) (32.16)

2 lo fJ (t'J 4w.: '
1 1
1
A- t 1• exp [ (' ( -- - 2.) 1 1
lo fJ (t') ct' dt'] = A- t 1< ' {32.17)

aod therefore (equation (30.40)):

w<n) (p.,t,h) t!n-l)d/(d-2)e-n•f•u p(n) (p,t'l",he-•fb< c21•). (32.18)

Dimensúln 2. In dimension 2 the situation again is difl'erent since the p·function

vaiÜShes. In perturbation theory the action (32.14) ís super-renorme.lizable, which cor·
responds to the relevance of the magnetic field for long dilrtance propertles. Tbe RG
equations have been written in Chapter 30 Btld Section 31.3. For any temperature t the
correlation functions have a scaling behaviour. Let ru; rewrite RG equation (30.21) in
thís partic:ular ciiSe:

[ A.E_
lJA
+ ~((t)
2
+ !((t)h~]
2 âh
w<nl "'O
•
{32.19)

U&íng the dimensional relatioD:

lJ lJ lJ
A tiA + 2h lJh + Pi lJp, = 2(1 - n), (32.20)

we obtain the smling form of oorrelation functions:

wln) {p,)"' ht/(S.-o)hl-n p<n) (,.h-4w/(8•-•l). (32.21)

Thls sca.ling form is consíst.ent wítb the general scal.ing forro o( the q,4 field theory at Te
in a magnetic field and the value of the exponent t) (equation (32.13)).
In particular, calling M the ma.gnetization, we have:

M = wl 1l ~ h'/(&.-•l. (32.22)

We have a.lready notíced in Section 3UI the speci.al values t 411' a.nd t = g,.. At t 4,.
the relation between M aod h ís linear up to a logarithtnic correct.ion, M oc -h In h. Tbis
relation has been explained by tbe equi"lllomce w:ith the free ma.ssive Tbirrlng model. A8
a.rgued abow when e:mmining the behaviour of the spin 2-poi.nt funetion, for t > h tbe
model no longer represent:s the long distance physic:s of the lattice 0(2) ferromagnet.
722 The 0(2) 2D Classiaú Spin Model 32.3

32.3 Tbe Coulomb Gas in 2 Dimensions

The partítion function Z(T, z) of a globally neutra! gas of partícles of charges ±1 inter-
acting via. a Coulomb potential V (:r) in d. dímensions can be written:

Z(T, z) =f: z~"2 f (n

n~o (n.) ;
dd:z:; ddy,) exp 1-E.,(:t;, y,)}Tj,

E .. (x;,y;) L V(ix; x;I)+V{iy;-y;l)- l:VOx•-Y;I),

where T is the teroper..ture a.nd z the gas fugoclty.

We note th.at the Coulomb potential V(x) in d. dimensíons is also tbe propagaior (3:1..5).
Therefore the r.h.s. of equation (32.4) is, up to a multiplicative factor, the Boltzmann
weight for this gBB. Comparing then with expressíoo (31.42), we oee tha.t we ca.n mpr....,nt
the partition functíon ""a functíonal integral over a field ,P(:t) wlth the actíon of the SG
model:

(32.23)

We olso bere recognize the transformation (A28.32) applíed to the Coulomb potent\al.
The ldentity is troe in any dimension, bowevei we will in this chapter díscuss only two
dimensiotUl. The propertles of the SG model will then allow us to understand the physics
of the Cou!omb gas. Tbe torrespondence witb the notatlon of Section 31.3 is:

T=l/t, z=a/2t.

=
R.emark. We bave seen that for t 411 the free energy becomes íniinite in the absence
of a short d.íBta.nce cut-off. This implíes that for t <! 4.. i.e. T ~ 1/4.. the Coulomb gas
is only stable if the charged pa.rticles ha"'l a hard core.
Th• phas• transüion. We have shown in Section 31.3 that the quanturn SG model
must undergo a pbase transition at t = s,. (for a: small) between a pha.se with a finite
correlation length at low t, i.e. at hígh temperature T in the Coulomb g.,, and a pbase
with infinlte correlation length ai high t, i.e. at law T in the Coulomh gas. In the
Coulomb gBB language the nature of these pha.ses is obvíous (but not the existence of a
tra.nsition at finlte strictly positive temperature). At high T the gas is composed. of free
charges. At law T the system appr06Ches the clBSSical ground state: paire of positive and
negative charges are tightly bound {in pairs with !iníte polarízability). In the Coulomb gas
language the sio q, 2-point correlation function is tbe charge density correlation function.
Tberefore tbe correlation length characterizes the decay of the correlatlon between the
charges. lt cbaracterizes olso the decsy of tbe effective potential between 2 infinitesimal
externai charges. In tbe free charge pbase the correlation lengtb is finite whlch means
that the electrootatic potential is screened, the correlation length being the screenlng
length. In the pbase of molecular bound states instead no screening occurs, the effective
potential is proportion al to the ba.re potential, a.nd the correlation functíon hss a power
law decay.
We oba!! now try to study in more detail this phase trBilsitíon, whi<:h seems at first
síght to be un.related to the physics of the O(Z) model a.nd we sball come back to the
32.3 The 0(2) 2D Clasrical Spin Model 723

0(2) model afterwards. Since the transition point corresponds to a finite value of t11e
coupling constant T~ 1 tbe reuormalízation and RG properties of tbe model do not follow
from simple perturbatíve considerations. Tbe derivatiou of the RG equations within the
SG framework tbus involves a serles of intuitlve atguments wlúch are not easily made
rigorous. Fonunately we have derived in Cbapter 31 " remarkable relatíon between tbe
SG model BDd a two fermion rnodel whicb is a free field tbeory jnst at tbe transitíon
point. This allows to use tbe fermion model perturbatíon theory to gain some further
insight into the problem and in particular to justify the RG equatíons obtained more
díre<:tly.

92.9.1 Renonnalization and renonnalization yroup

At T = 1/871' (and z-+ O} the SG model i>; just renormalízahle. To study its renormal-
izatíon propertíes for T dose to 1/B1f we note that the deviation r- I 81f plays a role
analogous to the diHerence á- 4 in th~< rp 4 field theory. Therefore we cau try to calculate
RG quantities in a double z and 871' - T- expa.nsion. lt is convenient to introduce the
1

two dirnew;ionless coupling constants:

u ~ 1 l/81fT,
(32.24)
{ v=2z/TA 2 •

We assume thot the set {u, 11} is multiplicatively renormalízable. To tbese coupling
constants correspond two RG ti- functions. Tbe property thot tbe free ave r age o f tbe
product oi a.n odd number oi operators e*i4> vanishes ( chiral eymmetry in lhe equívalent
fermion formulation) implies a parity syrrunetry in "· Tbe RG functíon fJ, iH gíven e.t
!eading arder in u in terms of lhe function Ç(t) of equotions (32,15). Thus:

(32.25)
We know tbet the funct.ion tiu vanishes at u O. Therefore ti. sta<ts at arder v2 • Tbe
sign of lhe coefficient of u2 ís fu<ed by tbe requirement that u :::> O and u < O cannot be
connected, for v infinitesimal, by RG transfonnations. The exact value is norrnalízation
dependent. We choose:
(32.26)
tiv = -2uv.

Fíg. 32.1 Tbe RG llow of the 2D Coulomb gas.

724 Th.e 0(2) 2D Cla.ssíool Spin Modd 32.3

TM RG flow. At this order it is easy to lind curves whicb are RG invariant:

d
.>. d>. (u2 (l.J- v2 (>.l) = 2u{j.,- 2vp, O. (32.27)

For !ui > lvl we pa.ra.metrize these hyperbolas by:

1 + s2
tl =a 1 - s2 I• I < 1' (32.28)
2s
u al _....s2. (32.29)

The RG equation for the parameter s is then:

Tbe solution is:

(32.30)

In the long di6tance limit >. goes to zero. Therefore if a is positive, í.e. íf initially u > O or
T > 1/8,., then s(J.} increases as well as u(.\) and v(.\) until one leaves tbe perturbative
regime. Thls is consistent wlth the releva.nce of v for la.rge T. lf ínstead we start with
T < 1 j81r, which impUes u < O a.nd a> O, then s(À) goes to zero for small À. v(À) then
goes to zero and u(.\) to a finite limit. Thls is consisteot with the irrelevance of v for
T < 1j8lr. Thus the half-line T < 1/8,-, z ""O, is a. half-Une of IR fuced point.s.
For \ui < !vi we para.metrize the hyperboles instead hy:

28
tl""a~, \s\<1, (32.31)

1 + 82
v al-s2· (32.32)

We then find:

The solutioo Cllll be written:

a.rcta.n s(.\) = -u In À+ arcta.n •o. (32.33)

Tberefore, irrespectíve of tbe sign of a, i.e. tbe inítia.l sign of u, u(À) a.nd 11(À} always
increase nntU one lea.ves the perturbative regime. Thls a.gQ.in is a. regio11 in wbich 11 is
relevant. The conclusions are that tbe ha!f-plane 11 > O is divíded in tbree ~ons {see
figure 32.1) separated by the lines u ±u. Regíon I is an infinite correlatlon length
phase, the law temperature phase of the Coulomb problem. Regiolll! li a.nd Ill are both
finite correla.tion phases with free charges. The Une u + , = O is thus the line of phase
transition. Regi.ons n and 111 differ by tbe property that in regio11 III the fiéld theory is
asymptotlcally free a.t short distante whíle in region l1 tbe field theory is non-trivíal at
both long and short dist.ances.
32.3 Th• 0(2) !!D Clas..-icn( Spin Modd 725

9!!..!:1.!! The ccrrelation length near tJw; phase tron.sition

Let us now characterize the behaviour of the correlation length in tbe high tempera.ture
phase when one approaches the phase transition. We set:

(32.34)
(32.35)

With this p8J"a.rnetrization -r pla.ys the role of the àevíation T - Te frOin the criticai
tempera.tu:re. •
The con-elation length is a RG invariant of mass dimension -1. It therefore satisfies:

(32.36)

Th lind the relation between r and { we look for a value of À such tha.t {(>.) is of arder
1 when {(1) = { ís large:
>.~ 1/{.
We ha.ve shown that the quantity rw is ·,. RG invariant:

r(À)w(>.) = rw

The RG equation for r(.\) theo reade:

(32.37)

A sbort calculation gives the functioo T(>•)

(32.38)

We evaluate the l.h.s. when Tis small and r( À) is of arder I. The equation then becomes:

We thus obtain:

~).
e"'""P( 2v2wr (32.39)

a. result tbat ís characterístic of the 2D Coulomh ga.s pha.se transi.tion.

Remark. The RG jJ-functio!lS have a.lso been calcuh•ted to third order. The result is:

{3., = -2v2 - 4v 2 u, (32.40)

f3v ~ -2uv + 5v 3 • {32.41)

Nane of the conclusíons drawn from the analysis at leaàing order ís qualita.tively a.ffected.
726 The 0(2) 2D Classica! Spín Model 32.4

32.4 0(2) Non-Linear a-Model and Coulomb Gas

We have seen that the &tion (32.2) cannol represent the 0(2) spin model for ali tem-
peratures and we feel that somehow we have to introduce the condítion that the ti( :r) is
a cyclic variable. We also know from the analysis of Section 30.1 that at higher temper-
ature tbe non-linear 11-model appraximation fails hecause there may be poínts wbere the
2
field <J>(x) in the (.p2) Jield theory vanishes. Tbe cost in energy is minimized when these
points are isolated because thls corresponds to point def.,.,ts rather than line defects.
Then in a turn around these points the direction of the field changes hy an angle multi-
pie o[ 2n. Let us tberefore consider a configuration of the field 11( z) which l!: the sum of
a smooth background 111 (:.-) anda conftguration 8v (:z:, z;) solution of the classical field
equation and regular everywhere except at isolated points x, where v (:z:, x;) changes by e
a multiple o( 2'11":
fl(x) = 81 (:z:) + llv (x,x,), (32.42)
with:
"" (x -:::;) 2
Ov(:r,x;)=~n,arctan- _ ) ,
1
n;EZ. (32.43)
i. :t :x:-. 1
I
The tenninology is that llv (:z:, x;) is a sum of vortex excitatioru; located at points x 1
and of vorticity n;. Vortices are topolog;cal excit.ations in the sense that they cannot be
removed by a COIJtinuous deforma.tíon of tbe lield ll{:r). The energy of the oonftguration
is (setting A = 1):
S(ll) =~f d 2x[1Vt(:r) + a,.Ov(:z:)] 2 • (32.44)

We now use the identities:

(32.45)

(32.46)

We observe from expressíon (32.44) that the energy can be finite only if:

An integration by parts shows that the cross term in the r .h.s. of equation (32.44) vanishes
and yields:

f d"x IB.. 2
6'v(:r)J = f d2x L n;n;8,. In \2: - X;l a,. In lo:
•J
:z:; l

= -211 L:n;n;ln j:t:,- :r; I· (32.47)

•J

We recognize the energy of a neutra! Coulomh gas of charges n; and temperature T

t/411 2 • We know from the analysis of Subsection 31.3.2 (equation (31.71)) that the most
releva.nt tenns oorrespond to n; = ±1 wbich we assume &om now on. Of course the
fugadty A of the equivalent Coulomb gas bas to be calcul..ted from a mícroscopic model.
32.5 The 0(2) 2D Classi~ Spin Modol 727

Only results which are independent of it.s explicit value can be obtained by tbis methbd.
The relation between the Coulomb gas temperature t/41f2 a.nd the temperatura t' of the
equivalent SG mode! is:
(32.48)
In particular by iotroducing an auxiliary field IJo(x) ~md using identity (32.4) ""' can
write an effective action:

S(ll,ll,) =f d 2 ~ (~ (8.81) + 8! (8,.8

2
2 2)
2
-
4~2 00662]. (32.49)

The analy.;is of the Coulomh transit.ion t.hen shoWB that t = 1r/2 is the transition tem-
perature. For t > Tr/2, the cosO interaction is rele\'l!Jit and the correlation length finite.
For t < w[2, cos fJ is irrelevant a.nd no mass is generated.

32.5 Tbe Criticai 2-point Function in the 0(2) Model

We now want to compute the 2-point correla.tion function of tbe 0(2) model near t wJ2.
We therefore have to ca.!culate the a~rage:

<é(Bo(li)+Bv(ll)) e-i(6t(•)+Bv(•))) =l:t _ Yl-l/2% ( é(9v(ll)-9v(<))) CO • {32,5{))

To find an interpretation to tbis avera.ge in terms of a modification of tbe energy of the

Couiomb gas we make a few transformations:

(32.51)

Th perform the Coulomb gas average in the r .h.s. of equation (32.50) we want to write
the sum of tbe term {32.51) and the Coulomb gas energy (32.5Q) as ...-;sing from tbe free
field average in a source. The source J 2 (s) for the free field 82(•):

J 2 (s) = J(s) +K(s), (32.52)

generate. tbe two terms we need: AfteJ" integration over &, the J J term gives the Coulomb
gM energy, the JK cross term gives the term (32.51), but in addition a KK contríbution
D(x, y} arises of the form:

D(:r, y) =t f2,-2 dsds' K(s)K(s')D.(s- s

1
)

= -..!..
b}.
r• ds,.ds'~•...• •• a;a;' In \• - •'I· (32.53)

We can thus write the Coulomb average (32.50) as a. free fiald average in the source
(32.52) provided we multiply the expressíon by e-D, To evaluate D(:r, y) we use the
identity:
728 The 0(2) 2D Clo.s;iool Spin Model 32.5

(32.54)

The integra.tion aver s aod s' can be performed. Only one term coming from the seoond
integral, gives a contribution which depends on x - y, the othen; are cut-off dependént.
The final result is:
D(x,y) _ _!_In 1:.: y[ + const .. (32.55)
211
If we replace the Coulomb gos aversge in equation (32.50) by the free tield average with
the oource (32.52), multiplied by e-D, somethíng remarkable happens: The last fa.ctor
just canccls the factor corning from the integration over 81. Therefore the spin correlation
functions cao be entirely calculated from the SG model, the spin lield being represented
by the non-local field appearing in expression (32.52) (see also equation (31.89b)):

exp [ill{x)j .....- exp [-

2~ f" ds 14 <,.uiV~(s)] . (32.56)

The crítica! 2-pcint functíon. To calculate the 2-point function at the critica! temper-
ature we can uow use tbe RG consideration• of Section 32.3. Critico.líty COtre5ponds to
the specialllne u = -v in the notation of Section 32.3. Note that tbe rela.tíon between t
a.nd u on the criticalline is:
7r/2(1- u), (32.57)
The origin u = v = O is a.n m fixed point. The behaviour of tbe correlation function
can thm be obta.ined from perturbation theory. At leading order in the parruneter A of
equa.tion (32.49) the 2-poiot function is still given by equation (32..12).
RG equations. The criticai spin 2-point correlation function W!2 l satisfies the RG
equation:
[A!_ +/3(t)~ +<(!)] w<•>(x,t,A) =o. (32.S8)

The functíoru; ((t) and .6(t) are (equa.tíons (32.15,32.26,32.57)):

2
((t) :"' t/21f +o (t- i) '
(32.S9)
4( "')'
IJ{t) = - ; t 2 +o (t- ")3
2 .
For I < "/2 tbe theory is IR free. The effective coupllng constant at scale À beha.ves like:

t(Ã)=%(t+ 2 ~.l.)+oC:À) forÀ-+0, {32.60)

w<2l (x,t) being dimensionless, the solutíon of the FlG equa.tion can be written:
w<•l (z/>.,t) = Z2(>.)W! 2l (x,t(>.)), (32.61)
wíth:
(32.62)

Therefore, at large dúrt;ance, the critica.! 2-point oorrela.tion function w!2l bebaves like:
w<•l (z,t.) ~ x- 114 (1nx) 118 • (32.63)
•-oo
This is the celebrated KT result for the phase tra.nsition of the classical 2D XY model.
32.6 The 0(2) 2.D Cla;,sical Spin Model 729

32.6 The Generalize<! Thirring Model

Let us now use the results of Section 31.6 to justify the RG functions (32.26) and thus
the RG fiow near the phase transition. The conespondence wíth the SG model, in the
notation of the Coulomb gas (32.23) is:
z -f, 8"T = 1 + 9/". (32.64)
Ftom these expressions we see that for T close to 1/B" and z sma.ll, f and 9 are both
sma.ll. The study of the phase ttansition is reduoed to standard perturbation theory
with fermion 4-point renorma.lizable intetactions. The {j-functiom at one-Ioop arder are
{equation (A31.18)):
{jg -2/2/11. {32.65)
{ 11t -2/g/11.
We recognize the RG functions (32.26) in a differeut pa.rametrization.
In the fermíon Ianguage, tbe KT phase transition conespood• to the spontanoou.
breaking of the abelian U (1) chlral symmetry. The ph.,;e diagYam is simplest in the case of
the SU(2) synu:uetric model. Tben for g pooitive the force betw-n fennioDB ili attractive.
The model is asymptotica.lly free and the spectrum, whích is non-perturbative, conta.ins
massive partides, a massive fermion and a boson bound state: The chiral symrnetry is
spontaneously broken. However, the average of the arder pa.rameter ('i>.P) O, vu.nishes
in agteement with the Mennin-Wagner theorem. For 9 <O instead, the force between
fermion is repulsive and the chiral symmetry is unbroken.
The same picture is valid for the SU(N) Thirring model (""" Appendix A31.2).
For the two parameter model we have the three regioDB of figure 32.1, a region I in
which fields are lll8.Silless and the chira.l is unbroken, and regíoDB 11 tu~d III where the
chiral symmetry is spontaneously broken.
Note finally tha.t in the language of the simple massive Tbirríng model of Section 31.5
the situation is reversed. We have a transition at 9 = -1f/2 (for m emall) betwren a
perturbative rnaB$.iV~: phase and a ma.ssless phase where the mass operator 1j,.p becomes
irrelevant.

Bibliogra.pbital Notes
The low temperatura phase of the model was considered by
V.L. Bereúmkll, Zh. Eksp. Tenr. Fiz. 59 (1970) 907 (Sov. Phys. JETP 32 (1970) 493).
R.elevant ili aiso lhe article
J. Villain, J. Phys. C: Solid State Phya. 6 {1973) 581.
The properties of the Kosterlitz-Thoul= transition have been studied in
J.M. Kosterlitz a.nd D.J. Thou!ess, J. Phys. C: Saiid State Phys. 6 (1973) L97; J.M.
Koster\itz, J. Phys. C: Solid State Phys. 7 (1974) 1046; J.V. José, L.P. Kadanoff, S.
Kirkpatrick and D.R. Nelson, Phys. Rev. B16 (1977) 1217.
For the relation between the Coulomb gas and the 0{2) XY model see
B. Nienhuis in Ph.ase 1hlnsítions and Oritíc.al Phenamena '1'01. 11, C. Domb and J .L.
Lebowitz eds. (Academíc Press, New York 1!187).
Our discw;sion of the Kosterlitz-Thouleso; transítion follows the Une of
P.B. Wiegmann, J. Phys. C: Solid State Phys. 11 (1978) 1583; D.J. Amit, Y.Y. Gold-
sclunidt and G. Grinstein, J. Phys. A: Math. Gen. 13 (1980) 585.
For tbe applícation of these ideas to one-dimensional conductcrs {Appendix A32) see the
review article
J. Solyom, Adva.nces in Physia 28 (1979) 201.
730 The 0(2) 2D Classic:al Spin Model A32.0

APPENDIX. 32
NON-RELATIVISTIC FERMIONS IN ONE DIMENSION

A non-relativístic propagator in Fourier space for a particle of momentum k, energy E

and mass m reads
1
A(E,k) =E- k2 /(2m)

In many physical pro<esses only those non-relativlstíc fermions with energles remaining
dose to the Ferrni energy E r pla.y a role. In momentum space the Fermi sea ís in general
"'sphere:

ar some defcrmation of a sphere dependíng on the space structure (lattice... ). Hawever

in one space dimension the Fermi momenta take only two values ±.,/2mEp. Near the
FenDi surface we ca.n expand energies and momenta: l
;:.

and apprOltimate locally the dispersion curve by straight !ines. The propagator <an be
simpllfied
1
A,(w, q) = íw- ~(h/2Epjm' e:=±l.
We recognize the propagator of a massless relativistic fermion, the twu values E = ±1
corresponding to rigbt and left movers. The quantíty .j2Er/m playsthE role ofthe speed
of líght. Ir vre combine the two components E = ± 1 we obtain a rnassless Dirac fermion.
Finally the non-relativistíc spin appears as an externai quantum nurnber. Therefore a
spin 1/2 fennion yields a doublet of massless Dirac fennions. The two ferrnion model
considered in Sections 31.6, 32.6 thus describes self-lnteracting on<>-dimensional non-
relativistic fermions. Ali results derived from the point of view of the KT phase transition
have an interpretation in systems líke the Luttínger liquid ( the equivslent of the Fermi
liquid of higher dimensi<>ns), or one-dimensional conducton;. For a systematic discussion
of such rnodels the reeder ia referred to the literature.
33 CRITICAL PROPERTIES OF GAUGE THEORIES

ln this chapter ""' r:onsider lattice models with gauge symmetry and study their properties
from tbe point of view of phase transitions and spectrum structure. We ooncentrate
on pure latlice gauge theories (without fermions) a.nd study them ma.inly with la.ttice
methods. Physically this means that we cannot investigate ma.ny properties of a reo.listic
theory like QCD where fermions are coupled througb a gauged SU(3) colour group, but
we can still try to answer one important question:
Does tbe theory generate conlinement, i.e. a force betweeo ch&rged particles increasing
at large distances, so that heavy quarks in tbe fundamental representatjon cannot be
separated?
More generally can one lind charged (from the gauge group polnt of view) llSymptotic
states like massless ~or particles in the theory?
Other problellll!l which we do not consider bere, can also be discussed in this frame-
work: for example tbe a.ppearance of mBSSive group singlet bound states in tbe spectrum
(gluonium), tbe qu..,tion of a deconlinement transition at linite pbysical temperature in
QCD.
We first construct la.ttice mndels with gauge symmetry. We sbow tbat, as antícipated,
they provide a lattice regularizatíon of the cnntinuum gauge theoríes studiad in Cbapt.m;
18-21: the low temperature or small coupling e:xpansion of tbe lattice tnodel ís a reg-
ularize<! continuum perturbation theory. We then di&:UBS pure gauge theories (without
matter fields) on tbe lattíce. We discover that gauge theori"" lum< properties quite differ-
e.nt from the ferromBgJietic systei'Illi we have studíed ao far. In particular the ab..ence of
a lncal order parllllleter will force us to examine tbe behavinur of a non-local quantity, a
functional of loops called bereafter Wílson's loop to distiuguisb between the confined and
de<:onlined ph8.5eS. Results will be obtained in the higb temperature or strong coupling
limit and in tbe mean field a.pprmúmation.

33.1 Gauge lnvariance on the Lattice

Let us follow tbe arguments developed in Cbapter 19 to constmct a lattice gauge theory.
We assume that we start lrom a model possessing a global (rigid) symmetry group G,
and - wa.nt to make it gi'Wge invaríant.
Let t;'t be the dynamical variables on the site i on which acts a.n orthogonal represen-
tatíon V{G) of tbe group G:

'~'• = grp. gEV(G). (33.1)

Tbe model is ga.uge invariant if it is pnssible t<> malte an independent group transforma-
tion ou each la.ttíce site i witbout cluwging tbe partition functíon. For tbe measure of
lntegration nn rp; as well as for ali the terms in tbe action which depend only on one site,
gk>bal invariance implies lncal invaríance as in the contínuum. Problerus arise only whe.n
terms connect different sites.
Let us imagine fnr the sake of argument that the action has only nearest neighbour
interactíon terms of the form rp; 'Pí, í and j being nea.rest neighbour sites on tbe
732 Criticai Properties of Gauge Theories 33.1

lattice. Such terms are obviously not invariant since the transformations of '<'• and 'Pi
are different:
'P<'Pj ,__. ';'; Tgi gj'Pj, (33.2)
(where Tg means g tr<UlSposed). To ma.ke it locally invariant it is necessary to introduce a
parnllel transport<lr depending on both sites i, j, U;i which belongs to the representation
D ( G) and transforms like:
(33.3)
Tben the quantity
(33.4)
is gauge invaríant. There is an important differen"" betwcen the la.ttice model and the
continuum theory: in the continuum the dynamical variabies are the gauge fields which
are associated witb the Líe nlgebra, while on tbe Jattice they are the pa.rallel transporters
a.ssociated wíth the Lie group.
Th< af>e!ian ca.e. ln continuum field theory, in the abelian U(l) case, we bave already
exp!icitly constructed the pa.rallel transporter (equa.tion (18.30)). Let us consider the
product '{J' (x) 'P (y) of two complex fields at different points. We cal) A~ the gauge lield.
Then tbe foUowing expression ls gauge invariant:

tp' (x) <p (y) exp [•• [A,. (•) d.s,.], (33.5)

in which e is the gauge cCJupllng constant, and the gauge field is integrated over some
differentiable curve going from :x to y. Indeed in a gauge tr&nSformation:
'P (:r)- e'"l•l 'P (x)'
(33.6)
{ A~ (x)- ~(:r)- ~6~A (:x),
the variatíon of tbe pbase factor compensates the variation of cp' (x) <p (y) 5illce

e /.v AI' (s) ds,. ,_,e 1" A,. (s}ds• [A(y)- A(:r)]. (33.7)

The gauge field dependent phase factor which we have introduced is of cou""' an element
ofthe repr....ntation ofthe U(l) group acting on 'f!( X), which depends on tbe two points
x and y.
The non-abelian case. ln the non-abelian case the explicit relation il; more complicated
because the gauge field A: (x) !" is an element of the Lie algebra of G and the matríces
repre5enting the tield at different points do not commute. It can be formally wrítten as:

U (x,y) = P { exp [[A~ (s)t"ds")}, (33.8)

in wbich tbe symbol P means path ordered integral.

Let us now retum to our lattice theory. lt is clear that if we !lS!OCÍate to each link
f(= ij) of the lattice a gauge element Ut with the convention

u.,. = (u) 1, = (u-')w (33.9)

consistent with tbe transformation law (33.3), then ""' can alway>; render ali products
gauge invnrinnt by introducing products on curves of these group elemento.
33.2 Criticai Properties o[ Ga!'9e The.oria 733

33.2 The Pure Gauge Theory

We now discuss the pure gauge theory o.nd its formal continuum llmit as obtained ftom
a law temperature, strong coupling expansion.

99. f.J Actíon and partiiion function

We now have to construct a gauge invariant intera.ction for the gauge elements U. lt
foll<JW5 from the transformation law (33.3) that only the traces of the products of U's on
closed loops are gauge ínvariant. On a hypercubíc lattice the shortest loop is a squa.n~,
ca.lled hereafter a plequette. In what follows we consíder "pure gauge action of the form:

S(U) = L
pLaquettrs
trU;jU.;>UktUt;. (33.10)

in which {Jp is the plaquette coupliog. The appea.rance of products of para.llel transporters
a.long closed loops ís not surprising since we know quite generally that the curvature
tensor F,.. which appears in the pure gauge action of the continuum theory is associated
with infinitesimal transport along a closed loop. Note tha.t each plaquette appears with
both orientations in such a way that the sum is real when the group is unitary.
The partition junction. We can then write a partition function correspondíng to the
action (33.10):
z= f TI dU,,
l;nko{i;}
.-ll,.S(U) ' (33.11)

We integrate over Uii with the group inva.riant (de Haa.r) measure 8.SSOciated with the
group G. In contras!. to continuum gauge theoties, expression (33.11) is well-defined on
the lattice (at le!lSt as long as tbe volume iz; finite) hecause, since the group is compact,
the volume of tbe group is finite. Therefore it is not necesoary to fix the gsuge and a
oompletely gauge intJariant fonnulafúm or the theory is p05Sible.

99.f!.f! Luw lempat~ture analysjs

We wa.nt to undemtand the precise oonnection between the lattice tboory (33.11} and
the rontinuum field theory. For this purpose let us inv<lstigate the lo.ttice theory a.t low
temperatura, i.e. o.t large positi\'<l {Jp. In this lirnit the partition function is dominated
by mínima! energy configura.t.iom;.
Let us show that the tninimum of the energy corresponds to matrices U gauge tra.ns-
form of the identity. We start from a filllt plaquette 1234. Without loss of generalíty we
CB.Il set:
1
u 12 g! g2 g,, g2 e v (G). (33.12)
The matrix g, is arbitrary a.nd g 2 is calculated from U 12 and g 1 • Then we can a1so oet

(33.13)

These relations define fust ga, then g.. The tninímum of the action iB obtaíned when
the real part of ali traces is maximum, i.e. when the products of the group elements
on a pla.quette are 1. (The trace of an orthogonal matril< U is me.ximum wh"n ali its
eigenvalues are 1}.
ln particular:
(33.14)
734 Critica! Properties of Gauge Theories 33.2

;orhich yields:
(33.15)
[f we nmv ta.ke an adjacent plaquette the argument ca.n be repea.ted for ali links but one
whích h& already been fixed. In this way we can show tha.t the minímum of the action
is a. pure gauge. Thus when the coupling constant (:Jp becomes very large, ali group
elements are constraíned to stay, up to a ga.uge transforma.tion, dose to the identity.
From this a.nalysis we learn thBt the minimum of the potential is higbly degenerate a.t
low temperature, since it is para.metrized by a gauge tra.nsformation, which corresponds
to a filrite number of degree!l of freedom per sjte. This unusual property oi lattice
ga.uge theories corresponds to the property that the gauge action in classical mechanics
determines the motion only up to a gauge translúrrnation. To perform a low temperature
expansíon it il; necessary to "fix" the ga.uge in order to sum over ali mlnima..
Low temperoture e2p<lnliÍOn. Let us choose a gauge such that the nlinimum of the
energy corresponds to ali matrices U "' 1. At low temperature the matrices U are then
close to the identity:

U(x,x+ an.. ) = 1-aAp (z) +0 (a'), (33.16)

in which a is the lattice spacing, x the polnt on the lattice, lllld n,. the unit vector in the
directiOil IJ.. We know from the discussion of Section 19.1 that tbe matrix A,.(:r) is the
connection or gauge field. We have already shown that the transformation law (33.3) of
the parallel transporter implies for A,. (:z:) at leading order ín the lattice spacing:

which i• the usual gauge transformation.

Let us now expand the lattice action for small fields. To simplify the calculation we
para.metrize the otthogonal matríx U associated with the link (z, :r+ an,.) in terms of
the antísymmetric matrix A .. (•) as:

(33.17)

We shall see in what follows that we nood U up to order "~. With the pararnetrization
given o.bove equation (33.9} ímpli"" thBt the term of arder a2 vanishes. Let us define the
antisyrnmetric matríx F"" (x) by:

e-•'F ••{•) = U (x,z+ an.)U (x + an.,x + a(n,. + n.))

x U(z + a(n,.+ n.) ,x + anp) U(z + an,.,x). (33.18)

,.v
To calculate F (x) we introduce the definition (33.17) and use repeo.tedly tbe Baker-
Hau!Klorf formula:
1
ln(eAeB) = A+B+ 2 [A,B]+ ... (33.19)

Applied to the product of severa! factors it ta.kes the forro:

(33.20)
33.3 Crítica! Properties of Gauye Theories 735

and therefore:

a2 F ,.. (x) ~a (A., (x + ~an,.} +A. (x + an• + ~an~)- A,. (x+ an. + !an,.)
-A. (x + !an.)] + a2 [A.,(x),A. (x)] +O (a3 ). (33.21)
At leading arder we reccwet the curvature tensor:
(33.22)

We abta.ín one term iD the plaquette a.ction by ta.king the trace of expre5Sion (33.1~.
Since F,.. is an antisymmetric matrix tr F,..
ve.nishes. Thus

(33.23)

This result shows that the leading term of the smBII field expansion of the plaquette
a.ction (33.10) is the standard ga.uge action studied in Cbapter 19. The relation between
(J~ and the ba.re coupling constant e 0 oí contiDuum gange theories is thus:

(33.24)

As anticipated in Chaptexs 18,19, we eonctude that the law temperature expansion,

in a lixed gauge, af lattice gauge theories provides indeed a lattice regularization of
continuum gauge thoories. We have here discii8Sed ooly the pure gauge a.ction, but
the result generalizes ilnmediately to matter fietds. Higher arder tenns in the small
fieid expamlon yield additional interru::tions needed to ma.irrtain gauge invaria.nce on the
lattice. Thl.s is not surprlslng: we ha:ve already shown that the ga.uge invariant extension
of Paulí-Villa.rs's regula.rization aJso introduces additíona.l intera.ctions.

33.3 Wil&>n 's Loop and Confinement

In Subsection 20.2.2 we have cnkulated the RG (3-func:tions for non-abeli..n ga.uge theories
and shown that pure gauge throries o:re asymptotica.liy free in 4 dimenaions, which mea.ns
that the origln in the coupling constant spa.ce is llll UV lixed point llJld aJso implies that
the effe.:tive intera.ctlon increases at lo:rge distance. Therefore, as in the case of the 2D
non-línear <r-model, the spectrum of a non-a.belían gauge theory caJJnot be determined
from perturbation theory. 'lb explain the non-observation of free quarks, it has been
conjectured that the spectrum of the symrnetric phase consists only in neutra.l states, i.e.
states which are singlets for the group transforrnations.
Clearly it would be convenlent to identify alaca.l arder parameter, í.e. a local observable
whose average would distinguish between the QED pha.se of abelia.n gauge theories, in
which charge states can be produced, from the so-ca.lled confined phase. However, in
g:a.uge theories such a local order panuneter does not exist (see Elitzur's theorern). This
property follows from the sirnple rernark that physical observabies correspond to gauge
invaria.nt operators which are neutra.l by construction. F\nthermore we have seen in the
study o! oontinuum gauge theories (Chaptexs 18,19) that the only gauge independent
quantities oorresponding to non-gauge invariant operators are the S-matrix elements.
Since it is very difficult to determine S-matrix elements beyond perturbation theory, ít
has been suggested by Wilson to study, in pure gauge tboories, a gauge invaria.nt non-
tocal quantity, the energy of the vacuum in presence of largely separated atatic charges.
Let us thus first study this qu..ntity in pure abelian gauge theories, in which, in the
continuurn, ai! calculations CllJl be done explicitly.
736 Critirol Propertíes of Gcmge Theories 33.3

!J3.S. 1 Wílson's laap in continuum abelian gauge theories

In continuum field thoory, ío order to calculate the avemge energy, ít iB ne<:essary to
íntroduce the gauge hamiltonian, and tberefore conveníent to work in the temporal gauge,
We have constructed a wave functíon for two static point-!íke charg.,., in the temporal
gauge, in Section 18.4 (equation (18.4B)):

1/J (A)= exp [ -ie i, A; (•) ds;}, (33.25)

in which the charges are located at botb ends of the curve Co.
By evaluating the behaviour for large time T of the average:

W(Co) ={.p ie-HTI 1/J)' (33.26)

in whicb H ÍB the gauge hamiltonian in the temporal gauge, we obtain the energy E (Co)
of the vacuum in presence of static charges:

W(Co) ~ .-TE(Co). (33.27)

T-~

If the charges are separated by a distance R, we expect E to depend only on R and not
on Co.
The loop functionru W (Co) has a representation in terms of a functional integrru:

W (Co)= ( exp [ fcó A• (s) ds,.]),

-ie

C&, whicb is now deiioed in space and time, is the union of two curves, wbich coincide
with Co at time O, BJJd with -Co at time T respectively. The average here mearu; average
over gauge field conligarations.
Since in the temporru gauge the time component of A,. Ya~Jishes, we can add to C&
two straight !ines in the time dire<:tion whicb joín the ends of tbe curve; C 0 (t O) and
C 0 (l = T). W(Co) then be<:omes a functional of a closed loop C {see figure 3U ):

W(Co) '= W(C) = (exp l-ie fc A,.(s)ds,.]). (33.28)

The advantage of tbe representation (33.28) is that it is explicitly gauge invaríant since ít
represents the average of the p&l'allel transporter along a closed loop in space and time.

Co -Co T

Fig. 33.1 The loop C.

33.3 737
Gritical Properties of Gavge 11owrie.s

The question of confinement is related to the behaviom of the energy E when tbe
separation R between charges becomes !o.rge. In a pure abelian gauge theory i~ .tbe
continuum, which is a free field theory, tbe expr<'>'Sion (33.28) can beevaluated explicttly.
'Ib simpUfy the calculation we take for Co also a straight Iine and use the Feynman gauge.
The quantity W (C) is then given by:

(33.29)

with:
(33.30)

and:
J,. (x) -ie fc fJ (x ~ s) ds,.. (33.31)

The result is:

8~1 ,r (~ - 1) e fc.c dst · ds

2 2
In W (C)=- 2 ls 1 -s2l -d. (33.32}

The integral in the r .b.s. exhibits a short dist&nce singulnrity, and a short distance cut-off
has to be provided. F\irthennore to normalize the r.b.s. of equation (33.32), we divide
it by W (C) ta.ken for R= a, a being a fured distance. Let us now write more explidtly
the integrais:

{T
lo lu- tl 2 -d du dt +h R
lx

The first term in the r .h.s. is ca.ncelled by the normalizatíon. The second term is in-
dependent of T and therefore negligible for large T. It is a.ctually related to the scalar
product of the wa~ function .p (A) and tbe ground state eigenfunction. The third term
decrea.ses with T for d > 2 wbich we now assume. Only the last term increases with T:

2 2] l-d/2
u) +R

(33.34)

Therefore the vacuum energy E (R) in presence of the static charges bas the form:

E (R)- E (a)
4,.<~l}/ 2 r((d- 3) /2) ("3-d ~-d). (33.35)

We reoognize tbe Coulomb potential between two charges.

For d :S 3 the energy of the vacuum increa&:S without bound when the charges a.re
separated, and free charges ca.nnot exist.
738 Crítica! P.--opeTties of Gauge Thoories 33.3

For d = 3 the potential increases loga.rithmically.

For d = 2 the Coulomb potential increa.ses linearly with distance.
In more general situations the method that we have used ahove to determine tbe energy
is complicated beca. use we have to take tbe large T límit first and then evaluate the large
R behaviour. lt is more convenient to l.aJce a squaxe loop, T R, and evaluate the la.rge
R behaviour of W (C). We here obtain:

For d > 3, dimensíons in which the Coulomb potential decreases, the r.b.s. is domillated
by terms whlcll correspond to the reglon )• 1 - s2i «R in equatíon (33.32):

lnW[C(R)J lnW(C(a)J "-'const. x R. (33.37)

This is called the perimeter law since lu W (Cj Íll proportional to tbe perimeter of C.
lnstead for d 5. 3, In W (C) ÍDcreases as .R'- . The reason is tbat two cllaxges separated
on C by a distl!llce of orde:r R, feel a potential of orderw- 3•
In particular for d = 2, In W (C) increases llke R 2 , i.e. like the area of tbe surface
enclosed by C: this is the a.rea law.

99.3.! Non-abelian gouge theories

It foUows from the discussion of Section 1!1.3 that il1 the temporal gauge the wave fuuc-
tion corresponding t.o two opposite poillt--like static charges is also related to a parallel
transportei along a curve joiniug the charges. The 511Ille arguments "" in the ahelian
case, show that tbe average of the opera.tor .,-TH in th" ccrr.,pondillg state is given by
tbe average, il1 the sen.se of the functional integral, of the parallel transporter along a
closed loop:
(33.38)

in which werecall that the symbol P me8JJ5 path ordering since the matrices A,. (s} a.t
dilferent points do not cornmute.
If we calculate W(C} in perturbation tbeory we lind of coune Bt leading order the
•ame results as in the a.belian case. HDW<!VI!r we know from renormalization group, tbat
we cannot trust perturbation theory a.t large distances. Therefore to get a qualitative
idea about the pha.se structure we fu-st UBe the latt.ice mGdel to calcula.te W( C) in tbe
large coupling or bigh temperature limit fJ, -+ O.
Strong C01<pling eo:pansi<>n for Wilson'• /oop. We bem assume that the group we con-
sider bas a non-trivial centre. We sbaU take the explicit exwnple of gauge elements on
the lattíce belonging to the fundamental representation of SU(N) (wbose centre;, ZN,
witb elements the identíty multiplied by roots z of uníty, zN = 1}-
We calculate W(C) by expà.llding the integrand il1 expression (33.U) in powers of {3p-
We clwose for simplicity for the Ioop C a. rect.angle altbough tbe generalintion to other
rontours ;, easy.
33.3 Criticai Properties of Ga'!Jge Throríes 739

Any noü-vanishing contribution must be invarlant by tbe change of variables Ut - zt u;

wbere Zt belongs to tbe centre. Let us ron.sider one link belonging to tbe loop a.nd
multiply the conesponding link variable U{x, x + an~) by z.o. We now consider the set
of ali link variables U(x + y,x + 11 + an,.) wh.icb are obtained by a translation y in the
hyperplane perpendicular to n~ and also multiply them by z,. Another link belonging
to the loop belongs to the set but witb opposite oríentation. Plaquettes involving such
variables involve them in pa.irs. For a result to be ínvariant a.nd thus non-vanishing
we need thàt the number of times eàcb link va.riable appears in the dírection n,. mínus
tbe uumber of times it appears in tbe direction -n,. vaiÜEhes (mod N). Thus we st....t
adding plaquettes to satisfy tbis condition a.t point x. However tbe addition of one
plaquette does not chang<! the total difference between the nurnbers of linke in the +n,.
and -n,.directJ.:>ns. Tberefwe always Bt least one condition rema.ins urn;atisfied until the
plaquettes reach the otber link of the loop. We can then repeat tbe a.rguments for the
remalning links of the Joop and the new non integrated remaini.ng links of the plaquettes.
The number of required plaquettevariables to get a non-vanishing result, is at least equal
to the area of the rectangle, the minimal area surface having the loop 8$ boundary. We
ca.n then perform the int.egratians which are just facterized group integcations. In this
way we get a contribution to W(C) proportional to (P,)A, in whlch Ais the number
of plaquettes. The largest contribution corresponds te plaquettes covering the mimmal
area surfaces bounded by the loop. It is indeed obtained by covering tbe rectangle with
plaquettes in such a way that each llnk va.riabk: appear.; only twice in either orientation.
For a recta.ngula.r loop R x T we just get

W(C) ,._,eRTlniS•. (33.39)

Thís results indicat"' that the potentiBl between the static charges is lineru-ly rising at
large distance. St.Atic charges creating the loc.p cannot simply be &reened by the gauge
field, in which case we would aga.In get a perímeter law.

Remarlts.
(i) lf tbe centre is trivial, it i.s possible te form a tube o.long the loop and this implies
a perimeter la.w. ]f for example the group is 80(3), in the decomposition of a product of
two spin 1 represento.tions, we again find a spin l which can be coupled to a third spin 1
to form o. scalar. Thus two plaquettes can be glued te the same Jink of the loop without
constra.int on tbe orientatlon of the plaquctte.
(ü) The asymptotic form (33.39) i.s also wlid for tbe abelhm U(I) IBttíce gauge the-
ory. Therefore, in four dimensions, Wilson 's Joop bll5 " perimeter la.w at IJ.Ily order in the
weak coupling expa.nsion &nd an area law at le.rge coup(ing. We expect a pbase ttallllition
betwl!en a low coupling Coulomb pb85e, described by o. free field themy, and a strong
coupling confined pbase. Thís phase transitlon ha.s been observed in numeríco.l simula-
tions. It seems to be lirst order, but this question has not been definitively settled. The
existence of the transitioo is related to the compact nature of tbe U{l) group whicb is
only rele\'o.nt on the Iattioe {lattice QED based on gcoup elements is also called compact
QED}. Defects in which the gcoup element varie5 by a multiple of 211" around a pla.quette
gO'It:Tn the dynamics of tbe tram;itíon. They correspond in tbe continuum te magnetlc
monopoles. In four dlmensions monopole loops yie)d, for dimellllional reasoDB, logarith-
mic contributions to tbe actioo, a situat.ion reminiscent of the t;wo.dimensional Coulomb
gas discussed in Chapter 32. The separation of \'Ottíces in the Kosterlitz-Thoul""' (KT)
pha.e tram;ítion i.s here replaced by the separation of magnetic monopoles.
740 Gntiml Propertíes of Gauge Theori<!ll 33.3

The •Lring tension. Tbe coefficieut in front of tbe a.rea is called the string tension rr,

(33.40)

If no phase transition occun; when {Jp varies from zero to in.fi.nity, the gauge thoory leads
to confinement. In tbis case tbe behaviour of til<! string tension for {Jp small is predict.ed
by the renormnlization group. Since o has the dimensjon of a mass squared one ftnds:

(33.41)

in which efi is related to {Jp by the equation (33.24) and fJ.., {3, are two first coefficients of
the RG {J-function which are given in equation (34.ól). A ph)'5ical quantity relevant to
the continuum limit cal! then be obtained by dNiding ..jã by its asyrnptotie behavíour.
Let us define AL ss:

(33.42)

then AL /..fií has a continuum limit. Whcn one calculates r:r by non-penurba.tive lattice
rnethods, the verification of the scallng beha.viour (33.41} indicates that tbe result is
relevant to the continuurn field tbeory and not only a la.ttice artlfa.ct.
lt ú; possible to systematically e><pa.nd rr in powers of (J,. The possibillty of verify-
ing that oon.fi.nement is rea.li.OO in the continuum ljmit, oopends on the posslbility of
analytically eontinuing tbe strong coupling expansion up to tbe origin. Unfortunately,
theoretical arguments lead to believe that, independently of the group, the string tension
is a.ffected by n singularity 8B&lcinted with tbe roughening tra.nsition, trnnsition which
h~ is not related to bulk pmperties. At strong coupling the contributions to the
string tension come from smootb surfa.ces. When eõ decreases (/Jp increases), one passes
tbroogh a crítica! point ~R• Mter wbich the relevant 811TfBCes become rougb. At tbe
singular coupling e~R• tbe striug tension does not vanish but has a wea.k síngularity. Stlll
at tbis point the strong coup!JDg expa.nsion diverges. Therefore it is impossible to ex-
tmpolate to ubitrarlly sm..U coupl.ing. The usefulooso; of tbe strong coupling expa.nsion
then áepends on the position of the roughening tra.nsition with respect of the onse:t of
weak coupl.ing behaviour. Notice that numerica.lly in tbe neighbourhood of the roughen-
ing transition, rotational synunetry ú; approxímately restored (at least a.t la.rge enough
dístance). ·
One can a.lso calculate other qulllltities which are IISSOCiated to bulk properties, a.nd are
therefore not a.ffected by rougbening singularities, such as the fn!e energy (the connected
va.cuum amplitude) or the plaquette-plaquette conela.tion function. Howerer even for
these quantities tbe enrapolation is not easy becaUEe the tra.nsltion between strong a.nd
weak coupllng behaviours is in general very sharp. This is confirmed by results coming
from Moute C arlo simulations a.nd is interprete<! as iudicating the presence of singularities
in the complex {jp plane close to tbe real axls. F\:om the numerkal point of view it seems
that the plaquette-plaquette correlatlon functíon is the most prornising case for strong
coupling exprulSion.
Remark. We note that tbe potential between static charges in the confined pbase il;
llnearly incressing in the 88llle WRY as the Coulmnb potentia.l in one spa.ce dimension.
Thls leads to tbe following physical pieture: m QED the gauge field responsible of the
potential has no charge and propa.ga.tes essentia.Jly like a free field isotropically in ali
33.4 Crití=l Propertíes of G<Wge Theones 741

spa.ce directions. Consenation of Hux on a spbére then yields the R 2 -d force between
the charges. However in the non-abelian """" the attractive fon:e between the gauge
partides generates instead a Hux tube between static charges in such a way that the
force remll.Íns the same as in one space dímensíon.
Gauge svmmetry breaking: E!itzur's theorem. Let us add a simple c<>ttllllent about the
absence of " local e>rder pHta.meter in gauge theoríes. We have seen that in the témporal
gnuge the ground state is invariant under space dependent gauge trl!.llllfonnations. This
property is incompatible with the existénce of a local order pararneter which is nece!'Sarily
non-gauge invariant. Therefore the question is: can a phase transition on the lattice leail
to a spontaneous brenking of gauge invariance? The answer to this question can obto.ined
by generaii'l.ing the arguments devcloped for orrlinary symmetries in Sections 23.2 and
23.5. We consider the transition pr<>bability at low temperature between two states,
ooncentrated one Htound the minimal energy configuration A,. =
O and the other one
around a pure non-trivial gauge. H the gauge function is different from zero only in a
fhüte spaoe volume, the cost in energy is the same "" in a one dimensional system and
therefore the transition probability always remains finits independently of the number
of space dimenaions. Therefore the qunntum mechanical ground staté is ga.uge lnvariant.
N e>te that this argument does not apply to ga.uge transformations which do not w.nish
at lmge distances. Therefore it does not forbid a spoDtaneous breaking of the global
symmetry associated witb tbe gauge group.

33.4 Mea.n Field Theory

We have shown that the pure gauge lattice model yields a.t low temperature or coupling
the continuum gauge theory. The continuum model aliOWB, in perturba.tion theory, the
separation of charges at large distances. On the contrary, at high temperature, charges
are confined in the lattice model.
lt is therefore nec""""-'l".Y to investigaté the possibility of phase tra.nsitions in latl.ice
gauge thoories. In the case of spin model5, mean field theory hss given us a semi-
quantitative understo.nding of the phase structure at least for d > 2. It is therefote
natural to study gauge theories in the mean field approxímation.
The general mean field formalism bas been described ln Appendix A24. We introduce
two sets of real matrices lf>t and Ht, in which the index t stends for link. We then rewrité
the partitlon function Z:

Z fn t
dU, exp ha.s (U)], (33.43)

in which S (U) is the lattice action (33.10), as:

Z =f fi
UnU l
dlf>,dH,dU,exp[-{3.,8(4>)+ í:trH1 (rh -U1)].
llnb
(33.44)

The introduction of the variables 1/J allows te> expreos the a.ction in terms of an average
link variable. Since tbe average of an orthogonal (unitary) matrix is not orthogonal
(resp. not unitary), we bave defined <b as an arbitrary real (resp. complex) matrix. The
variables H 1 represent directly at leading order the mean lield wbich approximates the
effect of tbe plaquetté intéra.ction.
742 Criticai Propertíes of Gauge Theones 33.4

The integral over the matrices V now factorizes ínto a product of integrais over each
!ink variab!e:
f
dV e-"HU =e-p(H), (33.45)

in which p{ H) ís thus a G x G group invariant functíon of H (H tre.nsforming under

independent right and !eft multiplication).
The partitíon function becomes:

(33.46)

We tben look for saddle points in tbe variables H and q,. Since H aJJd tf> are g~>neral real
or complex ma.trices, we expect. to find many saddle poínts. However both for simplicity
and :symmetry reasons we look for solutions in which Ht and tf>t are constant on the
lattice and multiple of the identity (up to a gauge transformation ):

tf>t ipl' Ht=hl, (33.47)

in which I is the identity matrix. Calling S ('P,h) the !attice action per Jink we then find:

(33.48)

in which we ba.ve defined V(h) by:

v (h)= p(hl) (33.49)

tri '
with V(h) = -~h 2 +O (h 4 ) for SU(2). The sadd!e point equations are:

'I' V'(h), (33.50)

We ca.n elimínate 'P and obtain:

(33.51)

For h small; V'(h) is a.t least linear in h (as in SU(2)). We realize immediately the
essential dillerence with the spin models we had coll.6Ídered so flll'. The r.h.s. of equation
(33.51) is at least cubic in h instead ofbeing linear. Thus the equation has newr a non·
trivíalsolution a!'bitrarily close to ~ero. For 13. small there exists only tbe trivial solution
h O, which, according to the strong coupling or bígh temperature analysls, com:sponds
to the confined pbase ín which Wilson's loop follCIII's an a.rea law. For a criticai value (3 0 , h
jumps froro wro to a 6nite value, indicl>ting a first orrkr phase transitíon. We recall that
at a first ordet transition the correlation length, at least above the transition, remains
finite. Therefore tbe neighbourhood of the transltíon temperature does oot define a
continuum fiald theory, in contrast with the non.linear u-model. Above (3. the average
of Wilson's loop is given by:

W(C), (u TI q,,) ~ q,P<cl,

allliob lEC
(33.52)
33.4 Criticai Properties of Gq."9e Theorie. 743

in which P (C) is the perimeter of the loop. Therefore Wilson'S loop follows a perimete~
la.w and thE phase ís decnnfined. Above /3, we are in the low temperature phase which
can be described by a oontinuum fie!d theory and perturbation theory.
Discussion. As we h1n<e shown in Appendíx A24, mean fi;,Jd theory ís valid in high
dimensions. Omtinuum field theory tclls us that the zero tem perature (/3, oo) is an
IR stable fixed point for d > 4. Thus the mean field result can only apply for d > 4.
HoweV<lr we would na.ively expect a second arder phase transitkm in 4 + ~ dirnensíons
with a Critica! temperature of arder "• or /Jp ~ 1 f E:, in analogy witb the non-lineW" <f•
model. The open questíon is whether in MY integer dimension d > 4 the transition is
really second order.
For d :5 4 the zero temperature is a UV fixed point. The simplest consistent scheme
is one in which the critica! temperature vanishes and the model alway>~ rernoJ.ns in the
confined high temperature phase. The dimension d "" 4 for gauge theories plays the role
of the dimension d = 2 for the non-linear a-model. The l.arge momentum beha.viour
of correlation functions can be deterrnlned from perturbation tbeory, but no analytical
method yields directly their low momentum behaviour and therefore for example the
spectrum of the theory. The only other. analytical piece of informa.tion available is the
small coupling expansion in a finite volume of tbe eigenstates ofthe qua.ntum hamiltonian,
which one ca.n try to extrapolate by numerical methods towards the lnfinite volume limlt
(see Chapter 36). However, agaín there i.s nuwerica.l evidence of a sharp tra.nsition
between the finite volume aod infinite volume rcsults, makíng the extrapolation dillicult.
The most promising quantíties seem to be rati06 of m......,... This lock of m!iable analytica.l
methods expla.í.ns the popularity of nwnerical simulations based on stochastic methods
of Monte Ca.rlo type in this problem.
Monte Carlo methOiis. We shall not describe in any deta.il the numerica.l methods
which have been used in !att.ice gauge theoríes. They o.re genera.lizations of the methods
whlch we have briefly described in the case of simpler systerns in Appendix A4.4. In pure
gauge tbeories the existeMe of phase transltions has been inY1lstigated for maoy lattice
r.ctions. For tbe gauge group SU(3), relevant to the physícs of Strong Interactions, the
string tensíon hiiS been carefully mea.sured, the plaquette-plaquette oorrelation function
has been studiad to determine the mass of low Jying gluonium states. Finally ca.lculations
have been performed a.t finite phy>~ica.l temperature, i.e. on a 3+1 dimension.e.l lattice in
the limit in whlch the size of the lattice remalns linite in one dimension, this BiUJ being
relate<! to the temperature. In this way the temperature of a deconlinement transition
has been determíned.
Fennioll$ in numerical simvlatíÍms. One importa.nt qualitative feature of Strong Inter-
action physiai ís the approximate spontMeous breaking of chira.l symmetry (see Section
13.6). However we have already shown in Section 9.2 that non-trivia.l prob)enu; arise
wben one tries to construct a chira.l inva:ria.nt lattice r.ction. One has the choice only
between writing ao r.ction which is not explicitly chita! symmetric and in which one tries
to restare clúral symmetry by adjusting the ferrnion mass term (Vo'il110n's fernúons), or
writing a chiral symmetric action with too many fermiom; (staggered or Kogut-Susskind
fermions). Th.i$ is a theoretical problem. An importa.nt practical diffi.culty aJso arises
with fermiolll3: because it is impossible to slmulate oumerically fermions, it ís nece&-
sary to integrate over fermions explicitly. Th.i$ geoerates an effective gauge field action
which oontainB a contribution proportional to the ferrnim1 determinant and is therefore
no longer local. The speed of numerical metbods crudally depends 011 the k>cality of
the action. Thís expla.ins that most numerical símulations witb ferrnions ha~ been up
744 Critical Properties of Ga.uge Theoríes 33.4

to now perlonned in the so-called quenched approximation in which the determina.nt is

neglected. This approxímation corresponds to the neglect of ai! fermion l<>ops and bean
s<>me similarity wJth thE> eikonal approximation. In this approximation the a.ppraximat.e
spontaneous breaklng of chiral symmetry has been ~rified by measuring the decrea.se of
the pion mass for decreasing quark muses. Owing to the difliculty of the problem the
numerical study of dynamícal femúons is still in a preliroinary stage.

Bibliograpbical NotEs
Lattice ga.uge theories bnve been introd uced in
K.C. Wilson, Phys. R.ev. DlO (1974) 2445; F.J. Wegnet, J. Ma.th. Phys. 12 (1971) 2259.
Compnct QED is studled in
A.M. Polyakov, Phys. Lett. 59B (1975) 82.
High temperature expo.nslon is discussed in
R. Balian, J.M. Drouffe and C. Itzykson, Phys. Rev. Dll (1975) 2098; J.M. Droulfe
and J.-B. Zuber, Phys. Rep. 102 (1983) 1.
Elitzur'• theorem is derived in
S. Elitzur, Phys. R.e!J. 012 (1975) 3978.
For a review of the ea:rly development of lattice gauge theories see
J.B. Kogut in Recent Aduances in Fiel.d Theo'11 and Statú;tirol Meclw.ni.cs, Les Houches
1982, J.-B. Zuber and R. Stora eds. (North-Holland, Amsteroa.m 1984); R.ev. Mod.
Phys. 55 (1983) 775.
Calculations of the spectrum in non-a.beli.a.D gauge theori,; can be found in
M. Liischer, Nucl. Phys. B219 (1983) 233; M. Lüscber and C. Münster, Nucl. Phfl".
B232 (1984) 445; J. Koller and P. van Baal, Nucl. Phys. 302 (1988) 1; P. van Baal,
Nucl. Phys. B307 (1988) 274.
Monte Carlo simulatíons were initiated by
M. Creuts, Phys. Reu. Lett. 45 (1980) 313.
Ear!y articles on lattice gauge theories and Monte Carla simulations are reprinted in
Lattice Gauge Theories and Monte Corlo Simulations, C. Rebbi ed. (World Scientific,
Singapore 1983).
See also
M. Creutz, L. Jacobs a.nd C. Rebbi, Phys. R.ep. 95 (1983) 201; 1. Montvay, Rev. Mod.
Phys. 59 (1987) 263.
For Monte. Carla methods in statistical physics see for example
Monte Carlo Methods in Srotistical Physics, K. Bind& ed. (Springet-Verlag, New York
1979).
Relevant !ll'e also the lectures
M. Lüscher in Critica! Phenonuna, Random Sys!ems, Gauge Thoorie.s, Les Houches
1984, K. Osterwa.lder and R. Stora eds. (Elsevíer, Amsterdam 1986) o.nd Fields, Strings
anti Critical Ph.enomena, Les Houches 1988, E. Brézin and J. Zinn-Justin eds. (Elsevier,
Amsterdam 1989).
Finite ternperature transitions B.Ie discus.sed in
B. Svetitsky, Ph.ys. Rep. 132 (1986) 1; E. Marinari, Phys. Rep. 184 (1989) 131.
For pure gauge tbeoríes in two dimensions see for eJ<lUilple
D.J. Cross o.nd E. Witt.en, Phys. Rev. D21 (1980) 446.
33.4 Orític.al Propertí"" of Ga119e Throi"Íe.!J 745

Exercises
Exercise !!3.1
The Landau-Ginzbury model of supereonductivit!l. The Landau-Ginzburg model of a
superconductur is, in field theory language, the abelia.n Higgs model, i.e. a fie!d theory
for a charged scalar fie!d coupled to ll.D electromagnetic gauge lield (for more details see
the classical hookE on superconductivity). We propose to study by RG rnethads the
nature of the phase trllllSition:

with:
D,. =a,. + ieA,.. F". =a,. A.. -a. A,..
Calculate the RG tHunctions at one-loop arder, solve the f!aw equations for the coupling
constants in 4 and in 4 - < dimensions, Discu~ the RG f!aw, the loca.tions a.od stability
of the fixed paints.
746 Critica! Prnperties of Gauge Theories A33.0

APPENDIX 33
GAUGE THEORY ANO CONFINEMENT IN TWO DIMENSIONS

Two dímensions is from the point of view of gauge theoríes peculiar in the seru;e tha.t
a gauge field bas no real dynamicnl degrees of freedom. Still the gauge field generates
a force between particles which, as we have seen in Section 33.3, leads to conftnement,
even in tbe abelian case. An example of sucb a situation has been encountered in tbe
Scbwinger model in Sei:tion 31.4. Tbis ís a questíon tbat we here examine in more detail
on tbe lattire for pure gauge tbeories.
Lattice ga..ge theoríes in two dimensions. Let us consider a pure gauge lattice actíon
(33.10) in two dimensioll5. The partition fuoction is given by:

Z=! n Iinb{ij}
dU;;exp(/3 L
plaqUI!!ttri&
trU;juj.ukluli)· (A33.1)

We assume free boundary conditioJIS in the time direction (to amid closed loop variables
surviving dueto the topologic porperties of the two dimensional spa.ce manifold). We can
then use the gauge invariance to set to 1 ali link varia.bies in the time direction. Tbis is
the equivalent of tbe temporal gauge of the continuum theory. Ho""""" in twa dimensians
this has the remarkabie e!fecr af decoupling ali links in the orthogonal directíon. The
partition function factarizes. lf we cnll L tbe size in this direcrion, we find:

z zt, (A33.2)

wíth:

Z, =f l) dU, exp ~ ~ tr U,+l Tu} (A33.3)

in which i is now tbe coordinate in the lime direction. We recognize the partition function
of a simple one dimeru;ionnl model with twe>-body nearest neigbbaur interactions. The
partítion function can be co.lculated hy the transfer matrix method. ln the case of free
space baundary conditions we o1so can ret:

(A33.4)

The integrais over the matrice5 V. factaríze and we find:

(A33.5)

in w hicb n is the area of the lattice, and z(/3) is given by:

•(.B} f dU exp (/3 tr U). (A33.6)

We hereafter assume that the volume of the group has been normalized to l. The
function z(p) is, as expected, a regular positive function: no ph""" transition occurs in
two dimensions. lt is o1so ea.sy to cnlculate the avera.ge of Wilson 's loop. Calling R and
T the sizes of the loop in space and time, one finds:
W(C) = .-RTo(~), (A33.7)
A33.0 Critica! Praperlies of Gp.v.ge Theories 747

with:
z'(i1}) (A3l.8)
a(/3) -ln ( Nz({3) •
We have again assumed that tbe centre o[ the group is non-trivial a.nd called N the trace
of the unit matrix. As e:xpected, Wílson 's loop has an area law ror ali groups and ali
couplings in two dimensi.ons.
For {J small, z(fl) has the expansion:

z(/3) = 1 + ,,{3
2
+O (tr) . (A33.9)

This yields for tbe string tension:

cr(fl)- -ln/3, (A33.l0)

in agreement with expressíon (33.39). For fllarge, z{fl) can be calculated by steepest
descent which is also perturbation theory, and one linds:

lnz(fl} =N (fi~ Klnn) +0 (Ir'), (A3Ul)

in which K is a eonstant. Therefore:

(A33.I2)

in agreement with perturbation theory in the continuum. Finally let ns note that since in
two dimensions the gauge field h"" no pbysical degrees of freedom, no particle propagate!l,
and no gluonium state ean be found.
14 LARGE MOMENTUM BEHAVIOUR IN FIELD THEORY

Up to now we have majnly dis<:ussed the IR behaviour of field theories. In this chap-
ter we use the RG equations ro charocterize ill6tead the large momentum behaviour of
ranormalized field theories.
Tt hru; been observed experimentally thet the quarks, fundamental particles of the the-
ory of Strong InteractíOns, bebave líke free particles B1 tbe sbortest dibtances presently
acc<=ible. Tberefore the discussion of tbe large momentum behaviour and tbe identifica-
tíou of field theories whích behave approximately as free field theories at short distance
is dírectly relevant for Particle Physics. Examining the large momentum behaviour of
all field theoríes renonnalizable in four dímensiollS, we show that only tbrories having a
non-abelian gauge symmetry can be asymptoticully free, i.e. bebave as a froo field tbeory
a.t large euclidea.n momenta (Colema.n-Gross throrem). We begin our investigation with
scalar q,• -like lield throries and then add fermíons a.nd gauge fields. As an applicatíon we
calculate the total cross ll<:Ct.ion of electron-positron annihilstion into hadrons at large
momentum.
Another more theoretical rea.son for discussing the Isrge momentum behavi<lur ú; the
a.pparent connection between the existenre of non-trivial renormalized quantum field
theories a.nd the presence of IN fixad points. The absenre of identified IN tixed points
in theories like the 4>4 field theory or QED leads to the so-called triviality problem whích
we examine in Subsection 34.1.1. lf we then consider the tj>4 field theory as the mo.in
responsible of the Higgs mass, ""' obtain a ~;emi-quantita.tíve hound on the Higgs mass
{see Subsection 34.1.3).

34.1 The (.P2) 2 Fíeld Theory: Large Momentum Behaviour and 'fiiviality

Before discussíng the problem of la.rge momentum beh~>viour of correla.tion functions in

renormalízed field tbeories, let us fust empbasize a difference between the <1>4 field theory
as it a.rises in Statistícal Mechanics, a.nd was discussed in Chapters 25 and 26, and tbe 4>4
theory as consídered from the traditional point of view o! Particle Physics: in the case
of phase transitiona, the "bare" coupling constant 90 is fixed a.nd related to the initial
microscopíc" phys.ics, and the renormafu.ed ooupling ronsta.nt 9 is a function of 90, the
mass m and the cut-<Jff A. Tnstead in quantum field theory one traditíonally assumes that
the renormalized coupling consto.nt g is fixed, a.nd 90 varies when the cut-off A increasl.'$.
Below fonr dimensions the q,4 field theory i5 super-renormalizable and, st g fixed, IJD
has a finite limit when A becomes inlinite:

90 = m'f(g). (34.1)

The coupling consta.nt g = O ú; a IN stable fixed point and the large momentum be-
ha.viour of correlation functions is simply given by the leading order of perturbation
theory. 'l'here is only one special point, the IR fixed point g g", where a non-trivial
scalíng behsviour is obtajned.
In four dimenslons ill6tead the •ituation is di!Jerent, the ~· field theory is renormal-
izable and the determination of the la.rge momentum behaviour of correlation functlons
34.1 Lo:rye Momentum Behamour_ín Fíeld Thoory 749

relies on the study of RG equatiom;, líke for exa.tnple the CS equations. At large momen-·
tum the r.h.s. of the CS equatíons can be neglect.ed, reducing them to the homogeneous
equàtíons (10.50) for the asymptotic maasless theory (see Section 10.9):

I 8 8 N
méml+f3(g)f)g-2t](g) ] ri;:_l(p;;m,g)=Ü. (34.2)

Equatian (34.2) ca.n be solve<! by the metbod of cha.rocteristícs which we have íntmduced
in Section 25.5. Calling >. the scaling facto r, we look for a solutian o[ equation (34.2) of
theform:
(34.3)
with:
g(l) = g; Z(l) = l. (34.4)

Differentiatíng the r.h.s. with respect to A Bnd using equation (34..2), we lind:

d
"'.uoP·l iJ(g(A)J, (34.5a.)
d
Ad>. lnZ (.l.) = -'1 (g (.l.)). (34.56)

The solutians o[ equatiotJS (34.5) are:

(34..6)

{34.7)

Resca.ling in equation (34.3) >.m into m, and using the dimensional equation {26.6), we
obtain for d 4:

(34.8)

Thís equatioo shows that g (>.) Lo the effective coupliug 81. scale À. U we ~tert with a
coupling constant g small enough such that perturbation theory ;. applicable, then {J(g)
ls positive and wbeu >. increases, g(>.) increases. Jf p(g) has an UV zem, i.e. a zero g' at
whlch the derivatiw p'(g') is negative, g(>.) wíll haw g' ""limit. li p(g) has no zero,
the hehaviour of g (>.) depends on the behaviour of p (g) for g large. lf the integral u(g):

+<o dg
u(g) = ! /3(9)' (34.9)

diverges, g (.l.) tends toward• +oo. H instead thís integral conwrges, the equatíon (34.6}
bBS no solutíon for >. la:rge.
In tbe case of generic pha.se transitions, the special form o[ the bare coupling constant,
as s function of the cut-off, h~ ímposed to g to sit on an IR fixed point.
750 Large Momenlum Behaviour in Field Theory 34.1

3,j.1.1 Existence of the renormalized ~1 Jieid theory

lu; we shall show, tbe existence of a non-trivial renonnalized ~· field tbeory in four di-
mensíons relies on tbe same properties. Tbe problem isto be able to find a bare coupl\ng
corutant g0 which yields a given renormalized coupling corutant g for arbitrarily large
values of the cut-off 11. This problem has always of course a formal perturbative solution
but we want to discuss thís question beyond perturbatioo theory. Using the RG equa-
tions, we can discnss tbe behaviour of renormalization constants at fixed renormolized
po.rameters wben the cutroff varies.
Setting;
m
9o = 9o(s,g), s =h' (34.10)

and using tbe definition of the tl-function (for example equation (10.26)), we find:

O= (a :s + {j'(g) ~) 9o (s,g). {34.11}

The solutíon of the equa.tion can be exprt'Sied in tenns of the efiective coupling constant
g(>.), define<! by equation (34.5a), as:
go(s/>.,g) =go(s,g(>..)), (3U2)

When 11 increases, the parameter .>. increases. For 9 small, {3 (g) is positíve and tbus
g (.>.) increases uatil perturbation theory is no longer v:alid. The lacge cut-off behaviour
of renormallzation constants is related to the IN bebaviour of the reooriii1llli.ed tbeory
( oot surprisingly) and the possible existence oi IN fured points. If a IN lixed point e:xlsts,
g(.\) ftows ínto tbe fixed point. In tbe ~ 4 6eld thoory, there are numerical indications
tbat no IN fixed point e:xlsts. In the absence of a. :lixed point we bave to again examine
the convergence of integral {34.9). H the integral diverges, we can say that tbere is a UV
fixed point at infinity. lf tbe integral convergES, tbe equation (34.12) has no oolution for
large enough values of tbe cut-off and the renonnalized theory does not exist.
The problem can also be fonnulated from the point of víew of tbe bate theory. We
ha.ve sbown tbat for tbe values of the bare coupling constont go which ere in the domain
of application of perturbation theory, the corresponding renorma.lized coupling consta.nt
vaoishes at large culrolf. Again the existence of a oon·trívial renormalized theory relies
oo the existence o[ a non-trivial IR unstable fixed poínt. As we discussed In Chapter
30 in the ·case of the non-linea.r 11-model, when a tboory iH IR free, a non-trlvial the-
ory at smaller distance implies the ex:istence oi 811 intermediate scale la.rge compa.red
with tbe mlcroscopíc scale but small compareci with the correlatíon leugtb, at wbich a
crDB50Vt!l between a trivial m bebaviour ond a non-trivial IN behavlour te.ires place.
Renormalizatíon grouptells us that this scale mnst ba.ve tbe form (see equation {30.31}):

e(go) ~~-l exp [! !1<> dg' ]

~ (g') ' (34.13)

in which ~ (go) iH now tbe /3-function of tbe bare theory. The lengtb scale ( diwrges only
a.t a UV lixed point.
An equiva]ent condition ia obtained by calculating the effective coupling constant at
scale p. <1:: 11 (wbich is one delinition of tbe renormaiízed coupling consta.nt):
{"" dg'
ln(A/!L) = J. fJ(g'). (34.14)
34.1 Lo;f!le Momentum Beha11Ít1Ur.in Fidd Theory 751

H we increase A at g fixed, w.o find &. solution 90 only if the then: existo; a value at·
whicb the integral diverges, i.e. an UV fixed point. !f the integral converge:s for 9o la.tge,
A/1" - oo implies that g goes to the IR fixed point g = O.
In the case of &. one component ,;ystern (lsing-like) we have some definlte informatíon
about this question. In the limito[ inlinite ba.re ooupling constant, the lattice regularized
</!4 theory becomes the lsing model. We have presented conclusive numerical evidence
in Chapter 28 that in twn and tlrree dimensiom the lsing model and the q,• belong w
the same universality class. This means tha.t. no IR unstable fixed point exi8tS in these
dimensions. !n fÔur dimensions the evidence is sornewhat wea.ker. This ls expectet;j
within the RG framework sinee the 11pproach to scaling is only logarithmic. In bigher
dimensions !lgain the lsing model and the q,• fe.ll in the sa.me universallty class. It would
thus be hlghly •urpriHing if only in lour dimensions these models would behave differently,
with morrowr rather dose exponents. Finally Monte Carla numerical simulatíons are
completely consistent with triviality. Some of these remarks can Biso be extended w the
2
(rp2 } thoory for a number of components N :<;; 4. Furthermore we have calculated the
fj-function in the la.rge N llmit in Chapter 29 a.nd found tha.t. it is proportional to ~.
Therefore no fixed point exlsts in the domain of the 1/N expa.n5ion, i.e. for 9o of arder
2
1/N. On the other ho.nd for la.tge 90 the'(,p2) thoory becomcs the non-linear a-model
which we have exam.ined in the la.tge N limit in Section 30.7. Agaln we have found that it
becomes a free field thoory in four dimensions. Therefore we have strong evidena! that no
2
continuum renormalized (<J>2} theory exists in fom dimensions. Rigorons rosults would
Biso imply triviallty if it wen: possible w prove the divergence of the field renonnalization.
One may then wonder e.bout the meaning of the correlntion functions !IS defined by
renonnalized pmturbation theory. We sball e.rgue in Chapter 42 tbat there ore intrinllic
diflkulties in the reoonstruction of corn:la.t.ion functions from the knowledge of their
perturbative expa.tJBion. However, irrespective of thi!i problem, it foUaws from the RG
e.rguments that to a finite n:normalized coupling consta.nt can only correspond a complex
ba.re coupling consta.nt, the imagina.ry part vanishing at ali ordem in perturbation theory.
As a consequence, the rorrele.tion functions, depending on the summation procedure, will
either be complex, ar wiU not SBtisfy the field equations beyond perturbation theory.

$4.1.2 The q,• field theory far neyatiw coupling

In the prevíous analysis we have assumed that the renormalized coupling consta.nt is
alwa.ys positlve. Jt has been argued, althc:>ugh not very convincingly, that the renormal-
ized couplmg constant could Biso be negative. If we start from g < O and small, the
RG equations still teU us that g increases, but it now converges towu.rds 9 O wbich
is tbus a UV lixed point. Tbis procedure then defines a noo-trivíal theory. However a
question arlses immediately: doeo this theory corre:spond to a hamiltonlan bounded from
below? We have shown in Sectíon 6.6 tha.t. the IPI functional iH related w the a~age
of the ha.rníltonian in states with fuced field average. we bave """" in Section 26.5 that
RG equations relate the thermodynamical potential for small magnetization to IR fixed
points. Not surprisingly the la.tge consta.nt field behavioUJ' of the JPI functional is gcrv-
erned by UV fixed points. Following the a.rguments of Section 26.5, ""' co.n derive the
RG equation satislied by r(rp), the tbermodynamical potential a.t. consta.nt field o.nd per
unit volume. We eonsider for simplicity a 1IU18SI""" theory r<>normalízed at sca.le ll, and
subttect r(IP) at rp =O. We obtain the equation:

(34.15)
752 Lo.rge Mamentum Behamour in Fie!d Thea"!! 34.1

Solving the equation by the method of cha.racteristics and usíng dimensional analysis we
find:
r (-X'P, 9 , ~<) = ;.•r ( z'''(>.)<f,g(>.),IJ), (34.16)

in which g(>.) and Z().) are defined by equations (34.4,34.5) in terniS of the corresponding
RG functions. We ..., that to study the Iarge lield behavíour of r( 'P) we ha."" to increase
>. a.nd therefore study the UV limit of the tbeory. lf we start froru g < Oa.nd smali, g(>.)
a.pproaches the origin for>. la.rge. Tbe field renormalization Z(Ã) then tends towards 1,
and r!'Pl ca.n be taken from perturbation theory. Tberefore, in the case of tbe O(N)
2
ínV!l.CÍant (o;IJ2 ) field throry, 'P being then the length of tbe vector '{), we obtain:

(34.17)

with:
g(Ã) ~ (34.18)
(N +8)ln.>.'
The coru;equenc"" of this result are obv:ious: by increasi.ng >., r can be made a.rbitrarlly
1...-ge and nega.t.ive. Jt foDows that tbe correspondlng hanliltonian is not bounded from
belaw.

s,p.9 Non-l1S1/ffiptotic frwiom. A bound on the Higgs masJ

2
A first mnclusion emerges, importllllt for Particle Physics: tbe ( .p2 ) field theory cannot
be asymptotically free at la.rge momenta and is tberefore, alone, not a good candidate to
represent tbe physics of Strong lnteractions at experimentally accesoible short dista.nres.
A seoond conclusion seems unavoidable, considerlng the argumenta we have giV<>n a.bow
a.nd the numerical ~vidence a.vaila.ble: the (4>2 ) field theory is ll free fieM theory in tbe
2

infinite cuk>lf limit. This a problem for tbe Higgs rector of the Standard Model of weak
2
ele:ctroma.gnetic interactions which contains a ( .p2 ) intera.ction. Of course then other
coupling constants contribute to the t/>4 coupl.ing ~functíon, a.nd the RG 8ow may be
dífferent, as we di..:uss ín the coming sectlons. However the most líkely conclusíon is that
in t.he Sta.ndard Model this problem cannot be avoided. Considering the phenomenologi-
cal success of the model this conclusion is romewhat. surprising. Howe\'ef we reafu.e that
the model ha.s OD!y been tested in some limitOO range of energies. Therefore, demo.nding
that the field theory should be cousistent on ali scales he.s no real pbysical ju.tífication.
lf we keep the cuk>ff l&rge but lixed, at tbe cut-olf scale th.e theory will break down
but at lower scllles it will give reasonable BIIBWetS. Correspondingly the renormalized
coupling will be allowed to va.ry in s limite<! range, which goos to zero (logarithmlcaliy)
for large cut-off. ln this seme the Sta.ndard Model is llll effective low energy field theory,
in which the cut-off, rcllection of a !ruge mass scale of new physia;, cannot be completely
eliminated, even thougb the tbeory is perturbst.ively renormaliza.b!e: Tbe theory is not
consiEtent on ali scales.
This condusion is specia.lly una.voida.ble if tbe .p• Higgs coupling is not very small.
lndeed sinoe the gauge and YukaW!J. couplings are quite small (even in tbe CBSB of the top
quark), the Higgs coupling then probably dominates the Híggs sector and a. purely ,P4
analysis is relevant.
An upper bound on the Higgs mass. The Higgs fleld through its various couplings gives
massas to ali otber fields. The observe<! masses determine the correspondlng coup)lngs.
34.1 Lo.rye M omentum BehatJiot!r in Field Theory 753

Only the Higgs mass and thus the Higgs seif-coupling are unknown parameters. Note,
ho"""'er, that it is likely that the renormalized q,• roupling g is such that pertnrbation
theory remains at least semi-quantitatively applicable. Otherwise the succe>SeS of the
Standard Model are difficnlt to understand. In the perturbative regime the Higgs ma.ss
incre""es with g. 'Ib obta.in an upper-bound on the Higgs mllSS one has to examine wbat
happens when g iDcreases. As we ha"" argued above for g large enough probably tbe
Higgs mass is mostiy determined by the Higgo self-coupling. We tberefore exa.mine below
the pure 1/>4 field theory. Since we Nma.in in the perturbatíve regime RG argumO>nts a.re
still applicable. A bound on mll, the Higgs mas., c.w then be derive<!.
We solve equation (34.14), using the f!l(pa.nsion of the ,6-function,

for g sma.ll. We lind:

ln(A/Il) = ~g + ~ lng + K(go) + O(g), (34.19)

where the function K(go), aceording to tbe prev:ious discussion, is bounded but can
only be determined by non-perturbati-re methods. For g sma.ll we can use perturbation
theory to relate tbe 4> (=Híggs) expecte.tion wlue, which is known from the Z rnB.SS
{(4>)- 250GeV), and the Higgs mass (""'1 Section 27.2). At leading arder we find:

m1 = !u {4>)• +o (g~) . (34.20)

To minimize higher order conections we ch()QS() for g the renormalized coupling constant
at scale (<[J). We can then elimine.te g between equations (34.20) and (34.19), and lind

H we assume that we CBn neglect in tbe r.h.s. a.ll terms but the two ftrst ones, we obtain
a relation between the two ratios A/ {4>) and mH I (lf>}. Moreover if the Higgs is really
associated to a physical partide it.s me.ss must be smaller than the cut-olf (whlch at tbis
point only repres€nts the onset of new physícs beyond tbe Standa.rd Model). Taking
for the two coefficient.s of tbe 1)-funetion the wlues from eque.tíon(Z8.3) for N = 4,
81< 2 {3, ~ 2, (J3 /(3~ ~ -13/24 we obtain an upper bound for me

mH < 2.6 {4>} =1> mH < 640 GeV,

wlue whicb can be compared w!th romputer runulation ve.lues wbich vary in the range
67(}-700GeV. Note that the corresp<Jnding vn.lue o( g is such tbat perturbation theory at
leading arder should still be remi-quantitatively corNCt.
Conversely if the know the pbysical coupling constant at scale J.l. we can infer from
equation (34.19} an upper bound on the cut-off ar scale of new physics. Of course thú;
bound is very sensitive to sma.ll corrections since the equatíon determines ln(A/Jl).
754 Large Momentum Belulvíour in Fielà Theory 34.2

34.2 The General r/14 Field Theory

We now study the large momentum behaviour in a general 4>4 -lilre field theory in four
dimensions, We consider an a.ction for a msssless field </Ji(x) whích a.t the tree levei has
the forrn:

S(tjl) =f [~ Í : (ô,_rp;) 2+ ~~:;d Í:Y•;•t</J;<I>;<Po<I>J

ddx
1 i;~l J
. (34.21)

The corresponding RG equatíons have been established in Section 11.7. The n-poínt
correlation functions ...tísfy eq\Ultíon (11.98):

(34.22)

We also need the expllcit leadíng order of IJ;;•1 (equation (11.103}) in 4 dimensioos:

(34.23)

Furthermore we lwow that fl<i ís of order g'. FoUow:ing the arguments given in Chapter
26, it l.s easy to verífy that the thermodynamical potential per unit volume in a ronstant
field <p, and subtracted at <p O, sl>tisfies:

(34.24)

We aga.ín solve this equa.tion by the method of chara.cteristics (see Chapters 25 and 26).
Introducing n scale par61Deter À, scale dependent coupling constant.s 1/ijk!(À) and field
renorrnaliza.tion ma.trix Z,;(À) defined by:

d
ÀdÀYi;A-1 (À)= l1;;kl(g(À)), g;;J.t(l) = 9<ikl, (34.25)

1
ÀdZII• z-•1•) -- Tf<i (g(À)), Z;;(l) = 6;;, (34.26)
( dÀ ij 2

we obtain:
(34.27)

We know from the ana.lysis of the previoUll sectlon tha.t the large field behaviour of
r( <p) is governed by the UV fixed points of the theory. About the zerOB of the RG (j-
function, little is known in the general case. However there is a problero of int.erert. fur
Particle Physics we can investigate: the existence of asymptotically free field theories. If
for some initial Vl>lue 9iJkl "f the renormalized ooupling constant the efEective coupling
constMt 9ijkt(À) fioWS into the origin for large scale, then r(À<p) ca.n be ca.lculated froro
perturbation theory for À large:

(34.28)
34.3 Large Momentum E<hat!iour. ín Field Tkeory 755

We have immediately taken into account that for 9<;kl smali tbe renormalization matrix
Z,; goes to L The boundness of the hamiltonian implies tberefore thlli 9i;kz(>.)'f';'P;'Pk'PI
must be e. non-negative quartic form. Let us examine the motian of the quantitY v(.\),

(34.29)

using the explicit form o f the {3 function given by equation ( 34.23 ):

(34.30)

Tbe r.b.s. of the equation is a sum of squares. Therefore v(>.) is a positive increasing
functíon. This is clearly incompatible witb the 118Sumed property that, at least for >.
large enough, ali functiODS g1;tl(>.) go to zero. The only other possibility is that o.ll tenns
in the r .h.s. vanish:
Vm~n. (34.31)
However v(À) then vanishes ídentically. The argument is valid for ali vectors V'· This
implies that we sta.rted from a free field theory.
We can condude therefore that no stable (/>4 -like theory can be asympt.otícally free.

34.3 Theorles with Scalar Bosons and Fermions

Let us now consider theori..s renormalizable in foux dimenskms and involvlng only scalar
bosons and fermions. We fim discuss a theory with one scalar and one Diroc fermion
field for which we have already calculated the RG (3-functions in Section 11.8.
A nmple example. In the C6Se of one scalar boson field coupled to one spin 1/2 Ditac
field through an interoction term of the &mn:

(34.32)

the RG {3-functions a.t one-loop order in four dimensions are given by equations (11.139):

(34.33}

We see that the fermions also generate negative contributions to the 1/14 coupling RG
functíon, which ís no longer obviously positive. lnstead (3,.• now is strictly positi....,.
Therefore for u 2 smo.ll tbe running roupling constant u 2 (>.) increases for large >.. Since
u 2 is posítive, it grows in a.bsolute value a.nd the tbeory cannot be asymptotico.lly free.
There is however one esse which must be exa.míned separately: if u is of order 9 then the
tw.rloop contribution oi order u 2g2 is compa.rable to the one-loop term. Thís two-ioop
term which comes entírely from Z;,, the 4>-field renormalization in the purely q,• theory,
is given by equation (11.78) a.nd has to he added in equation (11.133). The function A.•
then becomes:
(3•• (34.34)
756 ÚJTyE Momentum Behaviour in Field Theo"!f 34.3

Therefore this edditional term is a.lso positíve BDd the condusion is the same.
The general case. The most general interactíon, renormaliza.ble in four dimensions,
has the form:

(34.35)

in wbich {u').b BDd (v').b IUehermítian matriees (see Appeodix A5). Sincethediagramf;
contributing to the RG functions ha.ve been ca!culated in the one component case in
Section 11.8, we just have to take into accounL the additional geometrical factors. It is
conveníent to set:
(34.36)
A mma.rk simplilies the calculation. 1f one caleulates the Feynman dia.grams in the
ma.ssless thoory, then each time a iir>J;q, vertex commutes with a. fermion propagator, the
matrix w• is changed into its hermitian oonjuga.te w'l. The renormalization constants
are tben:
1 . .
Z,1 = [- } 11'2E w'lw• + 0(2 Joops), {34.37)
6
(Z~) 11. = fi,i - ...2__
2
tr {w'lw1 + wilw') + O(Z loops), (34.38)
80' E

(I is the identity matrix) wbile the divergent pa.rt of the #<P 3-point function iB pro-
portional to wiwilwi. A sbort ca!cuiation tbeu leeds to the expression of the RG
)3.,-function:

l&.r 2 ,8'.., = ~ (wiwilw' + w'wilwi) + tr(w''w'')wi + tr (w'wil) wi + 2wiw 11wi.

(34.39)
Let UB wríte tbe How equation for tbe qua.ntity trw''w':

811' 2 À dd). trw'lw 1 = j (trwiwHw'w" + trwilwjw<lwi) +trw•twi trwii..,J

+ trw'wi'ltrw''wi + 2trw'w'lw3w'l. (34.40)

The ma.trices w1 w''a.nd witwi are positíve tberefore the two first terinB ín tbe r.h.s. oí
eque.tion (34.40), heing of the form of the tra.ce of the sque.re of a posítíve me.trix, are
positíve. Tbe fuurth term is la.rger than tbe third one. lndeed we ha.ve:

trw'lwitrw'I..,J Re (trw'twi trw''wi)

:$ (trw 1twi)" (trw'lwJ) = trw'wi' trw'lwi. (34.41)

lt follows that:

a,.2 .). ddÀ trw''w' ~ 2 [(trw'fwi) (trw''wi) +trwiw'1wiw''l (34.42)

Let. us introduce the 4-point wrtex Mobal:

{34.43)
34.4 Large Mumentum Behatliour in Field Theory 757

lt is easy to verify that equation (34.42) can then be rewritten:

871" .\ d~ tr
2
w'1w' ?: (M...... +Modo.:) (M;bcd + M;db<), (34.44)

whích proV<'S that the r.h.s. ls positíve. We conclude that trwótw'(.\) is a positíve
increasíng function for À large, except as before if the two-loop contribution corning from
the lj\-fteld renormalization can ca.Dcel the one-locp term - have just considered. This
two-loop term is proportional to g ...1,..gki"'Jw!b• with a posítiw coef!icient as we have seen
before. lt contribut.e< in equatian (34.40) by a term proportional to Diklm901mjw!bw~
which is pooitive:
(34.45)

Therefore the inequalíty remalns valid and the conclusion is unchanged: a field theory
contaíning only scalar bosons and fermions cannot be asymptotkally free in four dimen-
síons. lf such a theory ha.s no other IN fixed poínt, it is probably "triviBl" in the sense
of the trivialíty of the q, 4 field theory.

34.4 Gauge Theorles

We have calcula.ted in Section 18.9 the RG ()-function for QED. In four dírnensíons, in a
theory witb np fermions and n 8 bosons of charge e the (3 function rea.ds:
l e•
(3 (e2) (4np + ns)3 B1r' +O (e6 ) , (34.45)
4
therefore QED is IR free in four dimeiiSÍons and like the q, theory it is doubtful that
it existo as a theory consisteot at all scales. Of course, since the physical coupling
constant is very small, the physical predictiom of QED are not affected by thil; possible
incoll.'listency whose effects are much too small. In other words the size of the cut-off
required to define QED can be enormous (the Planck ma.ss?).
In Suhsection 20.2.2 we have calculated the ()-function for purely non-abeli8Jl gauge
theories and found:
• ll
2
f3(g ) "" - :,.2 6C(G) +O (o") , (34.47)

in which C(G) is the Casímir of the group G. As we haw alrea.dy emphru;ized, noo·
abelian gauge theories correspondíng to semi-sirnple groups are !l.S)'tDptotically free in
four dímensions. What we h!lve lea.med in a.ddition in this chapter is quite rernarkable:
only theories possesslng a non-abelio.n g!luge symmetry may share this property. In the
language of criticai phenomen!l these theorles are the only ones for which dimension four
is the lower criticai dimension, in the same sense as dimension two is the lower criticai
dimension for thoories w hich have a global continuous symmetry.
Gauge theories and fermíofiJJ. In Subsection 20.2.2 ""' heve also calculated the contri-
butíon of fermions to the (3-function. lf the t'ermioiiS belong to the representation R and
T{R) is the trace of the square of the generators of the Lie algebra in tbe representation:
t.rt•t• = -ó.bT(R), (34.48)
the (3-function rea.ds (equation (20.73)):
4
11
/3 (g2 ) " " - [ aC(G)- 4 T(R)] g,., +O (96 ). (34.49)
3 8
758 La.ye Momenw.m Behaviour in Fiold Theory 34.4

Tbis result is sometimes expressed in terms af the equivalent of the fine 5tructure constant
a. g2 /4.,,
Before any calculation we knew that the contribution of ferrnions would be positive
as in the abelian case. Therefote a gauge theory with enough fermions is no longer
asymptotically ftee, Actually with the conventional normalization used for this problem:
G(G) = N fur SU(N), T(R) = }NF,
in which N F is the number of ftavours, i.e. the number of fermion multiplets belonging
to the fundamental representation of SU(N). For the physícal colour group SU(3),
a.symptotic treedom imposes:
33
Nr<2. (34.50)
For completenesB let us bere also give the (j-funct.ion at two-loop order:
(3 (g') = i>Jg. + i>Jg6 + o (g8) ' (34.:H)
with:

132= [ .!.!.C(G)
3
~T(R}]
3
_.!_
g,.2'
(34.52)

- [ ~C2 (G)- C(R)T(R) (34.53)

We recall that, as shown In Section 10.11, these two lirst coeffidents are independent of
the renorma!i2ation scherne. For the SU(N) group, with Nr Havours in the fundamental
representation one linds:
l32"" _ (17N6 2
_ ~NNF
6
(34.54)

Gauge fields and scalar bosons. The sit1llltion ís more oomplicated in the cose of a
theory also containiug scalar bosons, like in tbe Higgs model. In general. the scalm fields
have a. tendency to destroy asyroptotic freedom. Fim they yield a positiw contn'bution
to the ga.u.ge coupling (3-function which, as can be seen in equation (18.128), is 1/4
of the fermion contribution corresponding to t.be same l'i'presentation. However, more
important, they introduce a f/! 4 coupling whlch, as we know from the ana.l.ysis of Section
34.2, by ítself does not allow asymptotic freedom. Let us assume for simplicity thll1 we
need only one <P" coupling constant u. The corresp<lnding (3-function has at one-loop
arder the form:
(3., =ou'+ 2buy' + cg4 • (34.&>)
We know that a is positiw. lt is eaey to wrify that the contrihutions to c of the two
diagrams (a) of figure 34.1 are positive. li the gauge group is SU(N) and the sealar field
belongs to the fundamental representation, a short calculation shoWB that bis nege.tiw
(diagrnms (b) of figure 34.1).

I
I \
I l
\ I
I
(a) (b)
Fig. 34.1 Doshed lineo corre!;p<lnd to gauge fields.
34.4 Large Mornentum Behavi<>'!T in Fíeld Theory 759

From the form (~4.54) of the 13-function we.,.,., tha.t wheo g and u are small, if g 2 • is
much stnaller tba.n u <>r the converse, then u increBSeS. Therefore tlle only possibility
for a.symptotic freedom is tllat u and g' remain of tbe same order. It is tben natura.! to
introduce the ratio:
v= ujg2 (34.56)
Setti.ug:
13.• dg 4 + ... ' (34.57)
and using equation (10.80) which relates the (3-functiollE in different parametrizatiom;:

ag -
/3i (g) -8 ' - .6d§), (34.58)
u;
we can ca.lculate f3v:

(34.59)

Weobtain:
(34.60)
The thoory can only be asymptotically free if the second degree polynomia.l in v has
posltive zeros. Thís requires:

2
(b-d/2) >ac, d 2b >o. (34.61)

The two zeros v1 and v, a.re then posíti'l'tl. Assuming:

we verífy thot v, is UV stable whlle v, is IR stable. Therefore, depending on the starting

poínt in the (9, u)-plane, the effect.ive coupling constants may be driven towards .the
origin. The conditíollE (34.61) are rather stri.ugent. In pa.rtlcular one can veriiy by
explicit ca.lculation of the c:oefficients (a, b, c, d) that it is ímpossible to add enough acalars
belouging to the fundamenta.! representetion of SU ( N) to give masses to ali vector booom;.
A genera.l analysis of a system involving gauge fields, fermionB and !!Calar bosom; is
ratller complex .. However a few results have been ohte.ined,
(i) It is necessary to render the oodlicient d of equation (34.57) smali by adding enough
fermions.
(íi} Generica.lly ít is impossible to give a mass to e.ll gauge fielda through the Higgs
mechanism without losing asymptotlc fteedom. There exist however theories in which
one can finda manifold of measure ZEro in the space of coupling coll5tants which leads to
asymptotic freedom. Tbis situation NQUiies a fine tuning of the Yukawa-type interact.ioJIB
between scalars arul fermions. Only in some supersymmetric theories is this fine tuning
automatically realized and therefore natural.
The conclusion is that most probably tbe weak electromagnetic sector in Particle
Physics is not asymptotically free. This is completely consistent with tbe observatlon
that the gauge couplings are smali, as one wonld expect in an IR free theory.
760 Lo:rge Momentum Behaviour in Field Theory 34.5

34.5 Applications: The Theory of Strong Interactíons

At the shortest dist.Mces preoently experimentally accessible, Strong lnteractions are well
described by a set o( fermions. quarks, transfornúng under the fundamental representa·
tion of the group SU(3), and intera.cting via SU(3) gauge fields. Presently six quarks
have been observed indirectly, corresponding to six .fi«TJ<YUrs. The sixth quark (the top)
has bren confirmed only recently at Fermilab, but it.s exist.ence had been generally pos-
tulated because llavours are paíred in successive generations. LEP and Fermlla.b recent
experiments yield a top quark with a mass ín the 150-200 GeV range, and exdude an
additional generation of quarks associa.ted with ligbt leptons (neutrínos of mass below 45
GeV). With this six quarks the theory is stUI asymptotica!ly free a:nd &symptotic frredom
alone would leave much roam for additlona! quarks.
Azymptotic freedom has first emerged to provide an explana.tion for tbe experimental
observations of point-like structure in deep inelastic scatteríng. One measures the inclu-
sive croas section for the sca.ttering of leptons (electrons, muons or neutrinos) off nucleons
to give Jeptons plus any number of unobserved hedrons at large momentum tra.nsfer. In
this way ooe probes the matrix elements between nucleon sta.tes of the product of two
electromognetic or weak currents nea.r the ligbt rone. We have shown in Chapter 12 that
information about the behavíour neor the light cone can be obtoined from RG w:guments.
It ;,. clea.r from the discussion of Section 27 .2, which can he ímmediately transposed to
IN stable fixed points, the.t this behaviour is characterized by Iogaríthmic deviatíons
frorn a ftee field behaviour. Asymptotic freedom thus provides a simple and elegant ex-
planation to tbe results obtained in deep inelastic scattering experimenta. We here do
not give a detaíled discussion of tbe theoretical predietions and refer the ínterested read er
to the abundant literature. We ratber examine .., an illustratíon a somewhat simpler
ex:arnple: electron-positron annihilation.
E!ectron-positron annihiiation. The total cross section of annihílatíon' of electron-
positron palll! into hedrons is related, at Jeading arder in tbe electromagnetic charge, to
the expectation V>Jlue o f the product of two hedroDic electromagnetic currents J ,.. Sínce
the electromagnetic current is conserved, we can wrlte in momentum space:

(34.62)

The cross section is proportional to 1m F ( q2 ) for if < O, At la.rge values o( q,, in an

asymptotically free theory, the bebaviour of F (q2) can be estima.ted from renormali2ation
group and perturba.tion theory. Strictly spea.king we ean only obt.Un tbe bebnviow of
F (q2 ) at sbort distance, i.e. in tbe eudidean region if large and positive. This behaviow
is valid ln any direction in the complex .f-plane at tbe possible exception of the pbysical
region q2 < O because F (q2 ) b011 a cut on the negative real axis corresponding to the
intermedi&te hedron sta.tes we are looking foc. lt is easy to construct analytic functions
whlch dea:ease faster than any powe:r in ali diz,ctions of the complex plane and oscillate
on the eut; an exa.mple is provided by the funct.ion exp( -v'?). We hereafter ignore this
dil!iculty but remember tbat the bebaviour we obt.Un might only a.pply to some local
average of F (if) which smoothes such 06cill.Biions.
The electroma.gnetk current J,.(q) requireo no renormalization because it is exactly
conserved, emd hes thus no anomalous dimens!on (see Appendix Al3.2). However the
product of two currents whlch have dimension 3 has dimell!líon 2. Since, due to curr<Jnt
conservation, we·have bee.n able to extract a factor of dimension 2, F (q2 ) is logB.rítbml-
caliy dívergent. This situation is simila.r to tbe tf>2 2-poínt function in the rfl 4 field theory.
34.5 Lo.rye Momentum Behavíour in Fíeid Th.t!Dry 761

The 2-point function ne<e<is an additive renormalizatíon which induces an inhamogeneous

term in the RG equations (see Section 10.7). CaDing 1-' the reoormalization scale, and
assocíating a pal"ameter m to the masses of the qu;u-ks, we ca.n write the RG equation:

(34.63)

The solution of a similar equation has alrea.dy been díscUBSed in Section 27.2 (see equa-
tions (27.24-A26.10)). Let us caD C(g) a particular solutíon of the inhomogeneous equa-
tion: ...
a
f3(g) ag C(g) =- B(g). (34.64)

Theo F (q2 ) -C satisfies a homogeneous equation which cWJ be solved in the usual way:

{34.65)

Since the thoory ís asymptotically free, the effectiv-e coupling constant g(.\) goes to rero
for large .\. The l.h.s. term B(g) does nót vanish for g O (even in a free theory F (<f!)
ís divergent),
2
B(g) "'Bo + B10 +O (g'), (34.66)
and thus C (D) behaveo for g smaD líke:

C(g) Bo- + (B1

= -- Bofjo;,) lng + 0(1).
- --;:;;_- (34.67)
2
2fjo;,g fjo;, >'2

Therefore at large scale ,\ the r.h.s. ar equation (34.65) ís domin<>ted by the singuliU'
terms of C (g(.\)):
F(>.V,o,m,p) "'-C(g(>.)) + 0(1). (34.68)
From the definitian of g(>.J and equation (34.64) it follows that:

d
>. d>. C(g(>.J) = B(g(.\)), (34.69)

a.nd therefore, using the expansíon of o(>.) for >. large:

B
C(g(>.J) = B 0 1n.\- ~ lnln.\ + 0(1}.
2 (34.70)

The final result ís:

F(ql) 8
~-In
0 (q- 2) + -loln
B (q2)
- 1
+ 0(1) ' (34.71)
2 p2 2/h. JJ2

It follows that:

(34.72)

One usua!lyexpresses thís result in terms ofthe ratío R (of) ofthe cruas ;...,uoo for"+ e_
into hadrons to e+ e_ into I'+J.i-· The latter cross section is given by the imagioary piU't of
762 Large Momentum Behavicmr ~" Field Thenry 34.5

tbe one-loop correction to the photon inven;e propagator due to muons. The expression
of the corresponding diagram has been given in Section 18.9 (equation (18.109)). It
behaves for q2 la.rge as In q 2 , its imagina.ry partis just 11 cansta.rtt. In a free qua.rk theory,
tbe ha.dronic cross section for Q2 is given in terrns o! the sarne diagram, the only difference
being the ooeffident whích involves the charges Q, o! the quarl<s. Therefore for q'J large
and negative, this ratio iB just the sum o! the squa.res of the quark charge<, the charge o f
the electron being talren ss unit. In an asymptot.ica.lly frtl<l tbeory the result. is the ~e
at leading arder. The gauge interaction bet:ween quarks lea.ds to logArithmic correct.ions
to the leading tAlrm:

R (q') = ~ Q; [ B~~ In (q!/p

1
2) +o Cn(q!/JJ 2))] ·
(34.73)

A twn.loop caicula.tion yields the coefficient B 1 • The final result is usua.lly e:xpre<Sed in
tetms of the effective coupling ronstant g(qfji) a.t scale qj ~''

R(q2) I:;Q~ [1+ 4!2g2(q/p)+O(g'(q/s.<))]. (34.74)

'
For the SU(3) mlour group and for the six flavoure already observed the coefficient o[
tbe leading tenn is:

(34.75}

Tbis result is valld when q is large compareci to ali quack masoos. In fa.r:t in ,experilnents
one measores R for momenta wge compareci to some quark massas and comparable or
smaller thaa others. If the miiSS<lS are well separated one expect.s, aad indeecl observes,
R to be slowly 'I'RI'J'ing in interrnediate regions and dose to the value obtalned by taking
into acmunt only the qua.rks of sma.ller masses. Fbr instance a.ll measurements have been
made below the top threshold and therefore the largest relevant value is 11(3.

Bibliogt'aphical Notes
For a discussion ofthe applicat.ions o! renormali:r.at.ion group to Strong Interactiou Pbyllics
"""D.J. Gross in Methods in Field Theory, Les Houches 1975, already qunted in Chapter
11.
Tbe RG P,.fundioD for ga.uge theories b.as been calculated at OD!>'loop order by
H.D. Politzer, Phys. Rev. Lett. 30 (1973) 1346; D.J. Gross and F. Wilczek, Phyo. R.ev.
Lett. 30 (1973) 1343;
The general a.nalysis o! RG properties of fou:r dimensional field tbeories is due to
S. Coleman a.nd D.J. Gross, Phys. R.ev. Lett. 31 (1973) 851.
See also
S. Colemaa in Properties oj the rvndamentGllntemctions, Erice 1973, A. Zichlchi ed.,
reprinted in Aspocts of Symmetrie3 ( Cambridge Univmsity Press, Cwnbridge 1!l85).
Applications of asymptotica.lly free theories to the moment.B o! the deep ineiBSt.ic structu:re
functlons are ronsidered by
H. Georgi a.nd H.D. Polítzex, Phys. Rev. D9 (1974) 416; D.J. Gross and F. WUcrek,
Phys. &v. 09 (1974) 920.
34..5 La.ye M cnnentum Be!Ulviou~ in Field Theory

An elementary textbook is
M. Le Bellac, Des Phénomenes Critiques aux Champs de }auge (Editions du CNRB,
1988), english versíon: Quantum and Statistíml Fietd Thenry (Oxford Unív. Pr.ss,
Oxford 1992).
Calculation of the RG /3-function to b.igher ordem is due to
W.E. Ca.swell, Phys. Rev. Lett. 33 (1974) 244; D.R.T. Jones, Nucl. Phys. B75 (1974)
531; A.A. Belavin and A.A. Migdal, JETP LeU. 19 (1974) 181; E.Sh. Egorya.n and
O.V. Tarasov, Tror. Math. Phys. 41 (1979) 863; O. V. Tarasov, A.A. Vladimirov and
A.Y. Zharkov, Phys. Lett. 93B (1980) 429.
For a discussion of the trivíality of q,4 see also
D.J.E. Caliaway, Phys. Rep. 167 {1988) 24.1.
The upper bound on the Higgu m...., is discussed foc exarnple in
M. Lüscher a.nd P. Weisz, NucL Phys. B318 (1989) 705.
CRITICAL DYNAMICS

U p to now we have been concerned ouly with statístical properties of criticai systems ai
equilíbríum. We shall now study their time-dependem properties (note the.t from the
point of view of Particle Physics thís time is completdy unphyskal and can be thought
of as, for example, the computer time in Monte Carla simu)ations).
Typical quantities of ínterest are relaxation rates towards equillbrium, time-dependent
correlation functions and transport coefficients.
The main motivation for such a st udy is that, ín systems in which the dynarnics is local
(on short time &:ales a modífication of a dynamic variable h85 an influence only locally in !
spa.ce), when the correlation length of a system becomeslarge, a large time scale emerges !
which charB.Cterires the rate of time evolutlon. This phenomenon called critirol slawín.g \.
doum lei>dE to universal behaviour and scalíng laWE for time-dependent quantities. l
Note however that many teclmiques we have developed in Cbapter 17 aJoo apply to •.
general local dynamical equatíons without refereoce to a possíhle equilihrium state.
·In contrast to the sltuation in static criticai phenomena, there is however no dean
and systematic derivation of the dynamícal equationE gaverning tbe time evolution ín
the crítica! domaín. One reason ís tbat ít ís often neeessary to dynamica!ly couple the
arder para.meter to other variables which have automa.tícally a slow dyna.mics and whlch
corre.;pond to coru;erved densitiCE, i.e. to densities whooe integral over whole spaoe is a.
coDBtant of motion: typical sucb quantíties are energy, :momentum, angular momentum ....
Examples of such couplíngs will be given !ater.
One can argue however that the dynamics of ali these Quantities can be descríbed by
coupled Langevín equations of the type considered ín Chapters 4 and 17. We bave alre..cty
shown in Sectíon 4.3 that tbe equilibrium distríbution does not determine the drivíng
force in the Langevin equa.tion. Only the dissipative couplíngs whicb are generated by
tbe derivative of tbe effective hamiltonian are relat.ed to the static properties. Indeed the
La.ngevín equation;

(35.1)

in whith v;(t) is the usual gaussian nolse:

(v,(!))=. O, (v;(t)v;(t')) "' n,;.S (t- t')' (35.2)

lei>ds to the equilibrium dístrihution e-li1'< íf the "streaming" term F; ('P} satisfies the
conservatíon equatíon:
O, \ (35.3)

as can be verílied ímmedíately on the corresponding Fokker-Planck equation (see for

exa.mple equatíons (4.27)).
This equation does not determine the streamíng term. A!l a direct consequence, with
eaclJ static univerEality dass are associa.ted "" ínfinite numher of dyna.mical universality
ciMSeS and Critica! Dyna.mícs has no longer the bea.utíful simplicity of the statics.
35.1 765

CoTTeiation and Tesponse ftmctions. A!; mentil:lned above we can be illterested in tbe
c:itical behaviour of rela:xation t<JWards equilib~ium, time dependent correlation func-
uon:s:
{35.4}
and also response functíons whicb cbarad.erize th., rnsponse of the system to inJilútesimal
time-dependent perturbations. They ca.n be genorated by adding to the hamiltonían 'H
(<p(t)) à souri:e:
1-i (<p (t)) ._.'H (~p(t)) f dt h(t)O ('P (t)) (35.5)•

The variation R(n) of the correlation function W(nl under an infinitesimal perturbation
proportíonal to the function O ('P (t)) is then given by:

(35.6)

The causality of the Langevin equation implies:

R(n) =O for t. < 11) '11 Si S n.

lf we take h constant, then the modifice.tion (35.5) is just a modification of the hamilto-
nian allowing at equilibrium to generate O (<p) eQrrelation functions. Therefore:

(35.8)

irl whicb the r.b.s. is now ""average taken with the equllibrium dietribution e-!!7-I('Pl.
We can of course ealculate bigber arder derivatives which respect to h (t) and generalize
propertíes (35.7,35.8).

35,1 RG Equations near 4 Dimensions: The PUl'ely Dissípative Case

We first discuss a model wíth a purely dissiplltive dynamics and witbout coll8el"Vntion
law5. In the classification of the revíew article of Halpexin a.Dd Hohenberg, whlch we
follow in this cbaptcr, we conside~ modelA. We haw developed in Cbapter 17 moot of
the technology we need to derive RC equat.ions for tbe dynamics. Moreover we have
already calculated in Appendb< Al7 the dynMlic RG functions botb for the ~· field
tbeory and the non-linear c>-model.
The N -vector madel near 4 dímensians. We first consíder 11 dissipative dynMlics for
tbe N-vecwr model:

(35.9)

Tbis model ha.s been briefiy dlscussed in Appendb< A17.1. lt leads to a supersymmetric
etfective a.ction S (til} whlch in terms of the superlield

(35.10)
766 Crit<cal Dynamics 35.1

can be written:

(35.11)

in which 'H chara.ct.erizes the equilibrium distríbution:

'H (q>) =f ddx [ ~ (Ôp<p)2 + ~To<p2 + 9o ~; (cp2)2]. (35.12)

Sv.persummetry and the fiv.ctuation-dissípatíon theorem. Note tha.t the addition to the
static action o( a perturbation proportional to the order parameter itself:

'H (<,o (x,t)) ,_.'H (<,o (x, t)) j dtddx b (x, t) · <p (x, t), (35.13)

is equiva.lent to the additkm of a term to the noôse in equation (35.9):

1-'(:t,t) ~->V(:c,t) + 10h(x,t). (35.14)

Therefore differentiations with respect to h(:c, t) generate >.(x, !) field correlation func-
tions.
In Subsection 17.5.2 we have derived the supersymmetric WT identities for the con-
nected time-<lependent rorrelation functions (equatiou (17.47)):

t (a:
J;:::;;l J
+ &1 ;
]
) w<nl (~<;, t,, ii;, 6',) = o. (35.15)

We have tben obtained tbe general form of a 2-point function consistent with supersym-
metry and causa.líty (equation (17.52)):

In our example A(t) is in addition an even real function of t. In terms of tbe Fourier
transform over time:
B(lt,w) ""L+oo e"-' A(k,t)dt,
1
(35.16)

tbe Fourier tra.nsform of Wl'l tBkes the form:

Wl 2 l (k,w,9,9') = B(k,w) + B'(k,w} -iw (8 8') !B'B(k,w) + /JB'(k,..,•)]

+ A(k,0)6 2 {6- e'), (35.17)
in which we denote by B' the complex conjuga.te of B and by fJ the set {9,0}. /
Tbe sum B + B' wbicb is the {<,o<p) connected 2-point functíon, is a real functioo. Tbe
response functioo {.X<p) is the coefficient of fl'6'. Taking íts ímaglnary part we find the
relB1íon:

lm [ 6h (!,w) {<,o ( -k, -w))] L:o := (>. (k,w) <p(-k, -w})

'2"' (<p (k, w) cp ( -k, -w)), (35.18)
35.1 Criticai Dpnami,; 767

known under the narne of fluctuation-dissipation theorem.

RG equatiorlS at Te. We hnve shOWll in Subsection 17.5.3 that the static and super~
symmetric dynamic theories bavé the sa.me upper criticai dimensíon. Fluctuations are
therefore only relevant for dímensions d $ 4. The reno~malized actíon S, ( .P) then has
the form:

(35.19)

in wbich tf> is now the renonnalized field and 'Ji, (,P) is the static renormalized hamil-
tonian. It il; characterized by the appearance of a new independent renormalí>.ation
comtant Z~ e.nd therefore a new RG function Tlw (g) which in the minimal subtraction
scheme and in the renormali>ed theory is given by:

d
'l.. (!I)"" {3(g) dyln (Z.,/Z). (35.20}

The RG equations for tbe criticai theory t.hen read:

éJ a a
[ll iJp. + .8 (g) Ôg + 'lw {g)!! 00 (35.21)

in which p. is the renormalizatíon scale.

At the IR fixed point g•, they reduce to:

(35.22)

witb:
'l.. 'l.. (g").
Dimensional analysü; yields the relation:

r(n} (Àp·
H
w· 6· " f!) = Àd-n(d-2)/Zpl-nr(n}
h h.r-)
(p· ~p' IJ·tVP').'
f.;
;;;p /:!. ill
p
') • (35.23)

Combining botb equatíons, setting:

(35.24)

and choosing:
P = n>.•p-"-', (35.25)
we find tbe dynw:nic scn.liug relations:

n)
r (n) ('r.p;,w;, (J ,,p."" 1 ,.. (~ ~ ,_, (J.nl{h•!>)
",d-n(d-2+•)/2-r(n-l)nl-nF(n}
.. ,..., n" , •" "
(35.26)
A few algebra.ic manipulations yield the equlvalent relation for connected correlation
functiOIIS:

Wl"l (Ãp;,w;, 9,,1'"" l,n =1) = >.í4+•Hl-n)+n(<i-2+q)J>clnl (p;,w,>.-z, 9,À•1•).

(35.27)
768 Critica.! Dynamics 35.1

For the ;p-field 2-point correlatíon function, l'l"hich corresponds to fJ O, we therefore

get:
w< 2l (v ' w' e= o)~ -p2-o+'
1
-G( 2l (::::.) (35.28)
p• .
The equal-time cotrelation function is obta.íned by integrating over "-"'. We thus verify
the corn;ístency witb the static scaling.
The clynamic critical2-point function thus depends on a. frequency scale which vanishes
at small momeutum like p• or a time scale which diverges like v-•. The RG function 'J..,
for this model has been calculat.ed in Section Ali.2 (see equation (Al7.21)):

'1.-(ii) (N ;
7
2
) (6ln~ 1)§2 +0(§ 3), (35.29)

witb as usual:
2
g "' -----::7-;Ú (35.30)
f(d/2)
The dynamic criticai exponent z follows:

z (35.31)

At this order it CE>D also be wrítten:

(35.32)
+
with:
c.: 6ln(4/3) -1. (35.33) I
The scalíng behBviour of the respome function can be obtained by consklering for example
the coefficient of 88 in W!'l.
I

I
Correlation junction$ above T, in lhe criticnl domain. Let us just wríte the RG equa-
tions at tbe lR fixed point:

(1':1-' +'lwfl~ -l)2a! - ~'l) rl"l (p;,w,,B,,a,l-',rt) =O, (35.34)

in which o is a measure of the deviation from the criticai t.emperature:

a ~T-T,. (35.35)
The dimensional rela.tion (35.23) becomes:

" !l)"" ;..d-n(d-2)/2pt-nr(")

r (n) (ph wiJ B·,, <1 ,.,...-") (!'i ~ 9 ínp !!_ !!_ fi,\')
À 1 p} typ, À21 À' {J (35.36)

Combining this equation with the RG equation (35.34) and choosing:

(35.37)
in which { is the correlation length, we finally obtain:

rlnl (p;,w;,fJ., a,l', fi}~ {-d+n(d->+•ll•+•ln~tJF(n) (P,e, w;Ç', ll,ç-•12). (35.38)

We now see that ali times are measured in terms of a correlation time .,. which diverges
at the criticai t.emperat ure as t,;•:
(35.39)
35.2 Critica! Dynamic• 769

35.2 RG Equ:•tions near 2 Dimensions: The Purely Dissipatlve Case

l"ear two dimensions the N -vector model is descríbed by a Langevin equation of the form
(see Section 17.7. Appendix A17.3):

(35.40)

v(x, f.) heing the noíse of Se<:tíon 35.1 and 'P satislyíng:

(35.41)

Settíng:
{J = go/A•-•, (35.42)
we can derive from equation (35.40) an effective ..,tion S((j>):

(35.43)

wíth:
"H (c/>)"" ! j d•x (o"q, (x, t)] 2 . (35.44)

The superlield q, satislies the constraint:

q,• = ]. (35.45)

In the superlield form (35.43) we see that the dynamical actions for (q, field thOOry 2) 2

and the non-linear o-model are related in the same way as the correspondíng static
hamiltonians.
The renormali2ed action S, (q,) then reads:

{35.46)

in which p. is tbe renormalization scale and d = 2 + ç.

ln the minimal subtraction scheme, the new RG function ""' (g) is given by:

(35.47)

The RG equations in zero magnetic field then read (see equa.tion (30.17)):

a a a
[J1. Ôjl + il (g) &g +I)"' (9) (I â!l (35.48)

Dimensional analyois yields {equatíon (:>5.23)):

r {n) (p,,w;,B;,g,p.,fl)-
. -
.\ d Pt-nr{n) (E! "": 8· r,;.
.\, p, •vP, m• ~'-) •
P ): (35.49)
770 Oritical Dynarmcs 35.3.

We introduce in addition to the functions Mo (g) and Ç (g):

(35.BO)

(35.51)

(see equations (30.30.30.31)) the new RG functíon:

r(g) n-l e (g) exp [ [o I)~{j(g')

(g') dg')
.
(35.52)

Combiníng equatíops (35.48-35.52) and choosing:

>. C'(g), (35.53)

we obtain:

(35.54)

Nee.r the criticai t.emperature g•, this expression agroes wlth the scaling form (35 .38).
The exponent .t is then given by:

z = 2 + 'lw (g .) . (SB.55)

Equatíon (A17.41) givt"S l)w (g) at two.loop arder. The exponent z is then:

(35.56)

or again in terms of 1):

(35.57)

35.3 Conset:Ved Or-der Parameter (Model B)

A •imple modífication of equation (35.9} ensures that the order para.meter is conserve<!,
i. e.:
ft f d
4
.:"' (:t, t) = o. (35.58)

We write the equation:

(35.59)

For the Langevw equation to genera.te the same equílibrium distTibution, the noise 2-
polnt function now has to be modilied occordingly:

(35.50)
35.4 Critiw/ lJynamics 711

The effective act'Jn S (t/>) is still supersymmetric:

S(t/>) = j d6dÕdt [-f ddx~0 7:; -ll-l (~:-li~) +1í(q,)]. (35.61)

The appearance of 11 non-local term in the action reminds us of the elfective field theory
for uniaxi11l systerns with dipolar forces.
Power counting is now different. Since the propagator reads:

flk 2 [1- !iw (i/- B') (6 + 8') + fí!P (k' +r) b2 (8- 8')]
~(k,w,0,8'J (35.62)
w 2 + ifl2 (k'):í (k 2 +r/'
w h"-'l the dimension of k 4 • Above 4 dimensions, the charaeteristic frequency diverge;
like k 4 wbich mea.ns that the dynarnical expcnent z is 4 instead of 2 as in model A.
Above two dimensions Feynman diagtams calculatad with the propagator (35.62) are
not singular at zero momentum for w I O. Therefore no counterterm singular in k CI!Jl be
generated and thus fl 0 in the effectíve action remajns unrenormalized. The renormalized
action S,{q,) is:

s,(q,) f dedãdt [-f ddx~: -~-~(~~-e~) +1t,(t/>)]. {35.63)

Tbe field amplitude renormalization implies the relation:

no~ zn, (35.64)

and therefore tbe RG function rtw {g) is:

d
'~<.; (g) -iJ(D) dg lnZ = -11 (g). (35.65)

The analysis then becomes quite similar to tbe ptevíous case. However relation (35.65)
implíes that tbe exponent :t: is no longet 8Jl independent exponent but given instead by:

z 4 -t). (35-fi6)

35.4 Relaxational Model with Energy Conservation (Model C)

Still in the &amework of the N-vector model we now assume that the total energy is
conserved. We know that in the criticai doma.in the most singular part of tbe energy is
tp 2 • We shall therefore coupie a field e (x) to '{> 2 {x) and write an equi)ibrium hamiltonian
1i (<,O(:~:), e (:r)):

1i (tp,e) f dd:r [~ {8,.<,0)2 + ~rtp2 + ~ ('{> 2 ) 2 + ~A 312ve(x)tp2 (x) + ~e2 (x)]

(35.67)
The static properties are not affected by this modiikation. Let w; call K (x) the source
for the correlation functions of e(;:;). I f we integrate o"" r e (:r) we obtaln a new effective
hamlltonian 1i ( tp, K):

1t(<,O,K) =f ddx [~ (â,.tp) 2

+ ~rtp2 + ~ (u 3A'tP) ('P2 ) + A'I2~Ktp 2
2
- ~'].
(35.68)
772 Critiwl Dynam,cs

The e-field correlations are given ín terms of the usua1 static r.p';. r::or:relation functions
aDd are thus the standard energy correla.tion lunctions. The e-field amplitude and tbe .,,
coupling c<mstant renormalizatioo:; can be e>..1Jress..:l in term:; of ~ insertion and {tp2 <p2 )
additíve renonnalízatíon <:onst-ants. More predsely we hnve the relation:

(35.69)

Ali other conelation functions of the field e (:r) are identical to the torrelation functions
of Nl'vtp 0 (:r) /2. Ftom the RG equations satisfied by the cnrrelation functions, we <:an
derive the RG equations for e (x) insertions. At Te they take the form:

a a a I n
2T}.(g,v)-2'1(g) ) r,(!,n) (q,p,g,v,/\)=0,
.
[I\ a11. + /3 (g) ag +Pu (g, v) 8,; (35.70)

in which r~Lnl is the correlation function with i e( :r) insertions.

Applying this equation W the case I+ n > 2 and using equation (25.66) we obtain the
relation:
!'/• (g, v)=<+ 2~
v
+ 2'12 (g). (35.71)

Then applying equation (35. 70) to relation (35.69) and again using equation (25.66) we
lind:
'V•
B (g), (35.72)

and therefore:

'l• = v 2 B(g}, (35. 73a)

11. =-~[e +2'12- v1 B (g)]. (35.7~b)

At the IR fixed point g• tbe function {J. beeomes:

O~= -i[; 2
v B(g')}. (35.74)

At leading arder in the e-expansion B (9') is positive:

B(g')

Therefore two cases have to be envisaged:

{i) a< 0:
The origin v = O is tbe unique IR fu<ed point and the roupling w e (x) 1s irrelevant,
we are back W the model A of Section 35.1.
(ii) a>O:
The IR fixed point is nan-trivial, it corresponds to:

v=v·=± [vB~g·Jf'2 (35.75)

At leading arder a vanlshes for N = 4:

4-N •) (35.76)
a= 2 (N+B)e+O(e.
35.4 Critica! Dynamics 773

In three dimensíons nurnerical calcuiations (see Chapter 28) shmv that o is alread)-
slightly negative for N 2. For o > O the dvmunícs differs from the simple d;vnamics of
Section 35.1 as we shall see below. •
Of course in both case. the values of '~• lead to a behaviour consistent with prevíous
resu!ts conceming the ( ipz ip 2 ) correlation funct íons.
The Langevín equation. Let us therefore examine the dynamícs of the model in the
case <> > O. The Langevin equation has the fonn:

íl 611
r{;= +v(x,t), (35.77)

ê= +v1 (x,t). (35.78)

The new gaussian noise v (x,t) is defmed by:

(v}= o, (v'(x,t)v1 (x',t')) -í1'6 (I- t 1 ) ;;f. é (x- x'). (35.79)

The elfective action now written in terllll! of 2 superftelds ,P(x,t) and E(x,t) is still
supernymmetríc and reads:

S(q, E) =Jded8dt{Jddx
'
[-.! tJE ~- 1 (tJEalJ -e 6&tE) +!.n aq,.
íl' 89 ao
(!!2.- 98atrl>)]
89

+'H (q,, E)}· (35.80)

For the reasons we have already given, the {EE) correlation function is not singular at
zero momentum, the pararneter !l' is unrenormalized. However the renormaüza.tion of íl
is now modified by the presence of loops with E fields, Settixlg agaín:

fl=íl,fZw, (35.81)

at on,..Joop order we go>t;

z.. (35.82)

while for íl' we haw:

rr = z,n:, (35.83)
in which Z, is the e-lield renormalization. The corresponding RG function ~. is glven by
equatíoo (35,73a).
The RG equations tbeo read:

(35.84)

Equation (35.83) leads to the identlty:

(35.85)
774 35.5

while a.t leading order:

(35.86)

To separate variables we set:

s= n';n. (35.87)

and take for example s and !!' as índependent variables. We then bave:

(35.88)

wíth:
fJ, =1)• 1/w. {35,89)

At the IR fixed in the (g, v) plane, the functíon (J. takes the form:

1
.!_ - -] s +O (e2 ). (35.90)
Nl+s

The fixed points in the variables s are:

s' =2/N -1, s = oo. (35.91)

Slnce o is p<>sití...,, t.he fixad point • = O ís stable for N > 2 + O (E). However the
s = O limit is 8- peculiar limit and it is not clear whether thls result ís consistent with
the ê-<'Xpansion.
The fixed poínt s = ao is never stable for a > O. lt also rorresponds to decoupling of
the E sector.
Finally for N < 2, s• = 2/N -1 ís the ste.ble fixed poínt. The ratío fi' /fi is finite and
therefore the dynamics of E and t/> are coupled. The function 'I• corresponds to 'flw in
mooel A and therefore, since TI< e.t the fixed point hOl! t.he vuJue o/11, the exponent z is:

2 a
s' N- I +o (é)~ : = 2 +v. (35.92)

35.5 A Non-Rela:xnlional Model (Model E)

Geneml l"<marks. Let w; consider the Langevin equation (35.1) (we have set f3 = 1):

,P(t) = -!íl.;1 6 !;t) + F,(t) + "<(t), (35.93)

with a noise (35.2):

(v;(!)vj (t')) = n,j b (!- t')'
1f the function F; ss.tísfies equation (35.3):
35.5 Critica[ Dynamja 775

the equilibrium distribution U; e-><. A particular solution of this equation is províded.

by:
(35.94)

~here R,.J is an anti.s~rmmetric matrix:

R;, =-R;,

In particular the mode-roupling of Kawasaki and Ka.danoff...Swift has thiJ; form. In con"..
crete exa.mples the rnatrix F/.;j is linear in the field 1p, and associated with transformatioru;
corr€sponding to symmetries of the harnilt.onian.
&ample. We now give an e;xample of a model with non-<lissipative couplings. Since
the effective action in such cases is no longer supersymmetric, we expect the number
of independent renormalizatio!lE to increase su bstantially and therefore the analysis to
become more complex. Moreover, by Josing the supersymmetry we lose a !JOW'lrful and
elegant technique to solve the renormalízation problem.
We ccnsider the so-callad model E: the ordet parameter is a complex field <p (x), nnd
as in prev:ious section there is a consened density e (x). The Langevin equation however
rea.ds:

(35.95)

(35.96)

The hamiltonian is:

(35.97)

Tbe noise 2-point functions are the same as in Section 35.4 (for the N = 2 case):

(ll{x,t)ll" (x',t')} ~ n6(x-x')6(t -t'),

(w) = (11'11') =O, (35.98)
v (x',t')} -11'fl,6 (:r- x')ó (t- t').
(v 1 (x, t)

lt ís easy w verify that this model provides one example or Langevin equation ( 35.93) with
a strearning term of the form (35.94) which ensures that e-><(") remains the equilibrium
diJ;tribution.
The model has a U(l) symmetry correspondi~ to the multiplication of cp by a phase.
From the invariance of the hamiltonian under an infinitesimal U(l) tra.ru;formatíon foi·
lows:
O,

and therefore e(x, t) i.s a conserved quantity.

Th simplify the discllSSíon, we have omitted the generically expected e j1p2 j coupling
and this generate; an a.dditional reflection symmetry e (x) '-' -e (:r). Ac<:Ordíng to the
analysis of Section 35.4, this simplilication is really justilied only if the exponent or the
776 35.5

specific heat o ís negatiw. The analysis can be general.iud to the CMe where the e I"'I I
coupling is induded. The interestecl reader is referred t.o the líterature.
Rmannalizatíon. Power countíng teUs us tha~ the theory ís renormalízable in four
dimensions and that the canonical dimensiono of the fields are:

I'PJ 1, !e] = 2

Ftom the general results of Section 17 .4, bawl on the BRS symmetry of the elfe<:tive
a.c;tion, we know that the Langevin equatlons renormaliu as predicted by power counting.
The form of the renormalized equations is further restricted by the U(J) symmetry,
the parity symmetry and the conservation of e(:t, t). Huwever these conditions are not
sufficiently restrictive. They do not forbid a tenn proportional to ifl (<p"<p) in equation
(35.96} and do not imply the equality of the coupling constants s in equations (35 .95) and
(35.96). We here need a rather indirect argumeut: sínce the regularize<! dynamic theory
has the regularired 5tatíc thoory as equilibrium distríbution the same must be true for
the renormalired thoories. We then ha.ve three reoormalization constants given by the
z.
statlcs, Z"', = 1 a.n.d Zu. In addition we have to renormalize fi, íl' and •· However a
WT ídentity follows from the remark that e(:t,t) is coupled to phase transformations on
the field <p. lndeed if we perform the transformation:

<p(x,t) e'D(t)'l''(x,t), e(x,t) =e'(x,t)-ó(t)f•,

equation (35.95) is unchanged while a.n additional field independent term is added to the
r.h.s. of equation (35.96): iifs. Therefore t.he renormalixatíon of s is connected to the
renormalization of the Lagrange multiplier .\0 MSOciated with e in the effective action.
lt is then eMy l<l verify that s is not renorm ali.OO.
The RG 11-furn:tírms. The modej depends on three independent dimensionless roupling
constante which we eM choose t<l be u and:

v s~/flfl', w=fl/íl'. (35.99)

The function f3v. is glven by the statics and determines the IR fu<ed polnt value u• "'
3'/40n- 2 +O (€2 ). From tbe prevl.ous di5cussion, and dimensional considerations, it
follows that the two other tl-functions can be written in 4 - ' dimensions:

{1."' -v(c+TJ., +'lw<), (35.100)

IJ,. = w (IJ..,- 'lw'). (35.101)

The coupling constant v has one obvious fixed point value v =O which decouples IP(x, t}
and •(x, t). Then 11~· ""'O and fi.., assumes the val.ue of modelA (equation (35.29)) and
ls thus positive. The stability matrix constructed wíth the derivati"'lS of the tl-functions
has an eigenvalue w given by:

w = (a{3.jlJv) - (" + '1w) <O,

showing that model A is unstab\e with respect to the introduction of the coupling ~.
lf t> does not vanísh, equa.tion (35.100) implies:

(35.102)
 35.5 Critica! Dynamics 777

Equation (35.101) has three type of solutions, w = 00 , w =O ar 11., = TJ.,•· To find the
stable fixed points we then need the RC functions at leading arder:
v (35.103)
'7·· fi.,• = -l6,r2.

It follows:
(i) The fu.ed point w oo is unstable because the stabilit;v matrix has one nega.tive
eigenv&lue -<.
(li) The fu.ed point w O, at leading arder, also appears to be unstable beeause one
eigenvalue is negative w -c'f'J. However the next term in the €-expansion has been •
c&lculated aud is positive. Therefore one cannot excl1>de that this fixed point beoomes
stable for < = 1. Thls fixed point exhibits an interesting violation of dynrunic scaling
since the renorm&lized ratio of the time scales n- 1 and n•-l vanishes.
(iii) The last fu.ad point w = t:/2 + O (<2 ) rorresponds to normal dyna.mic scalíng
since the two time scales are relate in a finite way. Ali cigenvalues of the sta.bility matrix
are posítive at leading arder in <. In thís case TJ., = 'f/w' and equation (35.102) holds.
Therefore, to a.ll orders:
'f/w = TJ.,•, = -</2. (35.104)
The exponent z is then ""actly calculable:
2: d/2. (35.105)
For a more detaílad discussion ...., refer the interested rea.der to the literature.

Bibliographical Notes
The classílication of the various dynamíc models i,; taken from the review article
P.C. Hohenberg and B.I. Halperin, Rev. Mod. Phvs. 49 (1977) 435.
The effective action associated with the Langevin equation has been introduced in
P.C. Mtmin, E.D. Síggia o.nd H.A. Rose, Phys. Rev. AS (1978) 423.
For an early discussion of the renorma.lization of dynamlc theories in a field theory
language see for example
C. De Dominicis and L. Peliti, Phys. Rev. B18 (1978) 353,
and references thereín.
The non~linear a~model has been considered ln
R.. Bausch, H.K. Ja.nssen and Y. Yama.:akí, Z. Phyi!. B37 (1980) 163.
For a revíew on early work on the criticai dynami.:.< of He see
P.C. Hohenberg, Physie<J 109 & llOB (1982) 1436.
A recent review with ma.ny relerences can be found in
V, Dohm, J. Low Temp. Phy8. 69 (1987) 51.
The exponent z for model A is given a.t arder e 3 in
N.V. Antonov o.nd A.N. Vasil'ev, Theor. Math. Phys. 60 (1984) 671.
The dynamics of various other models has also been discussed, like the interface model
of Wallace and Zia.:
f R. Bausch, V. Dohm, H.K. Janso;en and R.K.P. Zia, Phys. Rev. Lett. 47 (1981) 1837,

l·,··· and theNiel,

J.C. Yang-Lee
Phys. edge
IlerJ. singu)arity
B37 {1988) 2156,

l
N. Breuer and H.K. Janssen, Z. Phy8. B41 (1981) 55.
Finally the criticai behaviour in driven diffusive systerns has been studied in
H.K. Janssen and B. Schnúttmann, Z. Phys. B64 (1986) 503, K. Leung and J.L. Cardy,
J. Stat. Phys. 44 (1986) 567.
.fi FIELD THEORY IN A FINITE GEOMETRY: FINITE SIZE
SCALING

Many numerical calculatjons, like Monte Carlo ar tronsfer ma.trix calcula.tions, are per-
formed with systems in which the size in severa! or ali dim.,nsions is finit;e. Th extrapolate
the results to the infinite system, it ís tbus neoess"'Y to have some ldea about how the
inlinite si7.e limít ís reached. In particular in a system in which the forces are short
raJige no phase transition can occur in a finite volume, or in a geometry in which tbe
size is in!inite only in one dimension. This indicates tbat the infiníte sim extrapolation ís
wmewbat non-trivial. We present in this chapter an analysís of tbe problem in the case
of second arder pbase tronsitions. We fil"5t díscuss it from tbe RG point af vieW (in lhe
case there[Qre of second order pbase transitions) and derive the eximence of a finite size
scaling. We then distinguish between the finite volume geometry (in ex.plicit calculations
w.. ta.ke the example of the hypercube) and the cylindrical geometry in wbich the stl.e
is finite in all dimensions except one. We explain how to modify the methods uzed in
the case of infinite systems to calculate tbe new universal qua.ntities appearing in finíte
síze eJfects, for emmple in d = 4 - t: or d = 2 + & dirnen.síons. Special propertíes of the
commonly used periodic bound>UY condítiorul are emphasi.zed. Finally botb static lWd
dyna.mical finite sire eff..ct.s we discussed.
Note finally that two-dímensional rnodels with one finíte size IUid periodic boundaxy
conditions also describe finite temperature 2D qua.ntum field theory.
The appendix contains a few remarkl! about finite sh.e eff..ct.s in the several phase
regiou when til<! rorrelation lengtb is finite, and about tbe calculation of one-loop finite
size Feynmo.n diagnun.s.

36.1 Renormalization Group in Flnit.e Geometrles

We alwnys áSSume ín this chapter that the si.., o( our system is characterized by one lengt.h
L which is large in the microscopíc scale, for example much larger than the lattice spacing
in la.ttice models. Wben tbe correlation length is also large, the urúversal properties of
the system "can be described by a continuum field theory. V."e use boundaxy conditions
whicb do not brea.k translation symmetry to avoid surf~ effects which are o( a diJferent
nature. Periodíc boundary conditions oort,a;nly satisfy such a criterlon. Depending of
the specilic symmetries of a model, other bound>UY conditíons are also available (líke
a.nti-periodic bound"'Y conditíons for lsing-like systems).
The crucial ohservation wlúch explains fmite size scalíng is tha.t the renormalizatipn
tbeory whlch leads to RG equations is rompletely ínsen.oítiue ta finíte súe effects si!1ce
renormalizations are entirely due to shon diztance síngularíties. As a consequen(:<l RG
eqnations are not modilied. However tbeir solutíon is diHerent because correlation func-
tions now depend on one additional dimensional pararoeter L. We discuss below lhe
solution of RG equatíons both in the examples of the q,• field theory and of the non-
furear u-model.
36.1 Fíeld Theon; ín a Fínik GeomeiTlf: Finíte Size ScoJ.ing 779

86.1.1 The (q, 2 ) 2 field thoory for d < 4

L<!t us first con.sider the exa.mple of the field theory action

(36.1)

where t characterizes tbe deviation fmm the criticai te.mpera.ture. In terms of tbe dimen-
sionless coupllng con:;t8llt g uA-• the correspanding correlation functions satisfy for
d < 4 the RG equa.tions (Section 26.5):

a
[AfJJ\. + iJ(g] fJg ~'I (9) ( n+ M â~)- 112 (g) t~] rinl (p;;t,M,g,L,A) o.
EJ

(36.2)
They can be solved in the usual WRY by settíng:

r!~ i (p;; t, M,g,L,A) = z-n/ 2 (À) r<nl (p,; t (>.), M (>.), g (>.),L, >.A), (36.3)

the various functions of >. being defined bY {equations (26.26-26.28,26Ji4))

d
>. d>.g(>.) = p(g(>.l), g(l) = 9'

>.dd>.t(>.) =2 -11-l(g(>.)), t(l) t'

d
>.d.\ z (>.) = '1(9(>.)), Z(l) = 1,
M(>.) = z-l/2 (.\)M

However the presence of tbe new length scale L now modifies the dimensional relations:

(36.4)

We use this relation in the r.h.s. of equation (36.3) 8lld then choose >. such that:

>.LA=l or >.l/LA. (36.5)

When tbe product AL becomes la.rge, >. goes to zero and therefore g (>.) approa.ches the
IR fixed point g•. This implies for the various functions of >. tbe behaviour:

(36.6)

We have therefore derived tbe s<.aling of finite size correlation functioru;:

r(n) ln,·t
~)),)
M g L t 1)- L-4+n(d-H~l/ 2 r<nl (Ln··tL
:Yiy
1f" MLM• g• 1
' ' ' '
1) • (36.7)
780 Field Thl1l>TY in a Finite Geometry: Finite Size Scaling 36.]

It is chara.cterized by the appeara.nce of a n""· scaling variable tL 11" whlch ís a pcrwer of

E./ L.
From equation (36.7) the usual inlinite size scaling form is recovered by expressíng
that rínl (L) has a limit for L :» { ~ t-•. On the other hand, in the lirnít ~ :» L,
the correlation functíons have a regular expansion in t even in zero magnetizntion in the
finite volume or cylindrical geometry, since no phase transitions ca.n occur in both cases
(for short range interactions).
Note that. all combinations which are independent of tbe normalization of the field q,,
of the temperature t, a.nd of the m~etiç lield are universal for the reasons explained in
the infinite volume case, once the geometry and boundary conditíoru; are fu<ed.
Adapting the usual a.nalysis of correo:tíoll5 to scaling to thls new situation, one imme-
diately linds that the leading corrections to the scaling fonn (35.7) have near d 4 the
form of a scalíng function multiplied by a fa.ctor L-~.
Example•. Let us now give two exarnples of qua.ntities which have been considered in
practical calculations. ln a. finite volume V the ratio 7<.o:

(36.8)

has been used to determine the criticai temperature. It has in zero m~tic field the
scaling form:
(36.9)

By calcula.ting "R. (t, L) for ditferent values o f L and looking for a tem per aturo at which
or
it is independent L, one finds the critical temperature t =o, provided tbe correctiorn>--
in L -w are negligible. The quantity f (O) is a universal number which in principie can be
calculated from the continuum field theory. We examine this problem !ater.
In the cylindrical geometry, the correlation length h in the infinite direction is anotber
quantity of interest. FTom equation (36. 7) one concludes:

(36.10)

e
In particular at t O, f. L grows like L and the ratio L/ L is universal. Since f.L is related
to the ratio of the two largest eigenvalues of the transfer matrix (equation (23.12)) -'o
and Àt we learn a.lso:
>-o/Ã1 = 1 + Ljg(D).
With this knowledge, it is inte~ing to return to the ilnalysís of the existence of phase
transitions in Chapter 23.
Since for t > O, {L goes to a constant fOT large L, and since for t < O it grows faster
than L, as ca.n be easily verified, the ra.tio eL/ L can be used to determine the criticai
temperat ure in transfer matrix calculations.

36.1. 2 Low temperotu~ expansion and finite size effects

We have shown in Chapter 30 tha.t in models with continuous symmetríes, at low tem-
perature the long distance behaviour is described by the elfectiVE interactions betwren
Go}dstone modes. In the c""" of the N-vector model the universal phymcal observ-
ables ca.n be derived from the low temperature or low coupling expan.sion of the O(N)
36.1 Fie/d Theory in a Finite Gwmetry: Finite Size Sculing 781

non-linear a-model. Let us thus shortly examine the prohlem of finite size effects in
thís framework. Previous consíderations concerning RG equations also e.pply to the RG
equatlons derived for the o-model: the equatíons are not modilied, only the solutíon is
changed by the finite size: The general solution (30.38,30.40) of the RG equations (30.21)
now depends on a.n additional scaling varie.ble L/((t) where for d ;> 2, t < t. the lengt.h
{(t) is defined by eQuation (30.31):

1 ) d t '] ' (36.11)

(d-2)t'

and for d 2 by

Ç(t) ocA - l exp [J t dt']

{:) (t') . (35.12)

Another convenient way to pe.rametrize the solution is to intrcduce a size..dependent

temperature tL, obt.ained by solving the equation:

t(1) = (' {36.13)

at scale À = 1/ AL: )
tL :st(l/AL). (36.14)

1'
Then
dt'
ln(AL) = '• {J(t') , (36.15)

which shows in pa.rticula:r that tL is a functíon of t and L only through the expectad
combination L/{(t).
When AL increases, Á l/LA goes to zero. For d > 2 and t < t 0 fu<ed, tL approa.ches
the IR fu<ed point t 0:
(36.16)

Therefore finite size effects can be calculated from the low temperature expansion and
renormalization group.
At te, 8Jid more generally in the critica! domrun, physical quantities can be calculated
in an ~ = d- 2 expansion, as has been shown in Chapter 30. Since te is a RG fu<ed point
k(tc) =f •.
Finally calculations can be perlormed in two dimenrions even in zero magnetic lield
h, L providíng an IR cut-off. However, because t = O is then a UV fu<ed point, !L goes
to zero for L/{(t) smal~
(36.17)
LL..., (N 2)ln(f,;(t)/L)'
and tbis is the limit in wlúcb physical QUantities can be calculated.
Notice here that a field theory in two dimensions with a finite size in one of the di-
mensiollB, and p<rriodic boundary coDditions .is also a finite temperature field theory.
Therefore the considerations of the Subsoction 56.~.2 are slso relevant fo~ finite temper-
ature.
i82 Field Theory in a Finite Geome!Ty: Fínite Size Scaling 36.2

36.2 Perturbatíon Theory in Finite Geometries

The scaling properties (36.i) are valid for ali boundary conditioru, but the explicit uni-
versal finite size expressíons change. Even the te<:hnical details of the calculation when
the temperature approaches Te vary. In this chapter we rnainly consider periodic bound-
ary conditíons, but we briefiy indicate in the next suboectíon how the method can be
adapred to other examples. In Section 3.4 we have octually gíven one example of twisted
bounda.ry conditions.
The cha.racteríztic feature of finíte goometries is that, in Fourier space, tbe momenta
which corrcspond to directions in which the size of the system is finite are quentized.
As mentioned above, to avoid extra complications due to surla.re elfects, we here ccm-
sider only boundary conditions which do not break translation symmetry. For generic
systems only periodic boundary conditions sa.tisfy this condition. However for systems
wíth symmetries other boundary conditions are pO&'lible.

!16. 2.1 Periodic boundary c.onditioru and the zero mode

In the case of a d dimensional hypercube of linear size L with periodic boundary condi-
tions, the quantized momenta p,. have the forro:

(36.18)

When the products tL 11" or LJ€ are positíve and not small, the usual methods of caJ..
culation of the inlinite volume are applicahle and finite size effects due to momentum
qua.ntizatíon are only quantitative, decreasíng like exp!-oonst. LJ€]. Wben t.he prOdnct
tUI• is nega.tive a.nd not small (the ordered phase) the physics of the infinite and finite
systems are very different, and tbis problem will be examine<! \ater. At T< in a finite
volume t.he prapagator h8.6 an isolared pole at p,. = O and therefore IR divergences will
appear in perturbation theory for € » L.
In the cylindrical geometry one component of the momentum w varies continuously but
still when aD other components vanish, rnassless Feynman diagrams receive a divergent
contribution ofthe fonn f dwfw 2 • Fínally a geometry in which tbe sizes in twod~ions
or three dímensiollS (this ia the case of finlte tempernture quantum field theory) among
d are ínlinite stillleads to IR divergences. We shaD not discuss these latter casE\'; further
in thls chapter because t.he IR problem cannot be solved exactly.
As a consequence even in hlgh dimensions, for which in the infinite geometry mea.n
field tbeory is exact, IR divergences are generated. Th overcome this dífficulty ít is nec-
essary to separate the zero momenturn Fourier components of the field. The components
k,. # O can be treated by the methods developed in the ínfinite geometry (perturbation
theory and RG), the oomponent k,. = O whose Huctuations are damped at T< only by
interaction terms, has to be treated exactly. In the case of the finite volume we thexefore
construct an effective integral over the component ~ (p.,. = O) by integiating out all other
components. In the cylindrical geometry we construct an effective quantum mechanícal
problem corresponding to the component ~ (PT =O, -r), denoting by r tbe coordínate in
the infmite directíon. Finally the geometry with two or three inlinlte dimensions leads
to effective 2D or 3D theories. Note that simlla.r eonsiderations will apply to the zero
modes of instanton calcula.tíons (see Chapters 37-43).
We examine in t.he next section the two fust geometlies sepa.rately, beginning with the
simpl...rt. case of tbe periodic hypercube.
36.3 Field Theo.-y tn a Finit~ Gwmet.-y: Fínite Size Scalíng 783

96.2.2 Twisted boundal"!f conditions

For systems with symmetríes additional boundary conditions do not brea.k translation
invoo-iance: conditions such that the values of the arder parameter at both boundaries
(for each direction in space) differ by a constant group transfonnation (often called
twisted boundary conditiom;). For illStance for lsing-like systems one can use antiperiodíc
boundary conditlons, for the N-vector mede! with O(N) symmetry one can impose a
rotation of a given angle around some a.xís.
In such cases the quantized momenta p~ are shifted by oome additional oollStants.
Let us verify this explicitly for a lield theory invariant under the transformations of lt
Lie group G.
We impooe to the field tJ> the boundary conditions

where the A,.'s are constant (i.e. space independent) commuting matrices, [A,., A.] O,
elements of the Lie algebra of G (A,. is a curvature-free gauge field),
To return to the s!tuat!on of periodic boundary conditions, we perform a gauge trans-
formation on the field t/1. We set ·

The new field c/>' then sa.tisfies periodic boundary oonditions. However the derivativas
have now been replaced by covariant derivatives:

8,. ._. 810 + A,.jL.

Aftet Fourier transfonnation tbe derivative; yield quantized momenta of the form {},_ ,_.
(2i1tk,. +A,.)/ L. For a unitary or ortbogonal group tbe matrices A~ are antihermitian and
thus have imagínsry eigenwlues iU,:. Writing the field rfl in a base where ai! A,. (which
commute) are diagonal, we finally see that the effect of twi5ted boundary conditlons has
been to generate a set of quantized momenta af tbe form

for each a component of the field. We can clearly choose ali angles O:: to belong to the
íntenral [-1r, .-].

36.3 The ,P4 Field Theory: The Periodic Hypercube

We first stud~ the cases of dimensions d > 4 and d = 4 - < in the framework of the
effective {4>2 ) field theory. As explaíned above, we expand ,P(x) in Fourier components,
separating the zero momentum oomponent:

(36.19)

The symbol I:' me""!' that the ''!ilue k = O has been omitted in the sutwnation. The
integration ovet the <J>t,., k # O, modes is performed as in tbe infinite goometry limit:
this generates a perturbative expansion which has RG properties. An integral remains
7B4 Pield Therrry in o Finíte Geome!ry: Finíte Si:ze Scalíng 36.3

on the last k O modes whlch must be calculated exactly. Note that the first pa.rt of the
procedure ll; formally equivalent to the shift of the expectation value of the fieid .p(x) in
the iolinite geometry. The main difference, apa.rt fr<>m tbe replacement af integrais by
díscrete sums in Feynman diagrams, ís that the avera.ge L -d J .p(:z;)dd:z: here rema.ins a
dynamical varia.ble (sce also the discussion of Section 6.5).
As an illustra.tion let us calculate avere.ges of the fonn (36.6), moments of the average
spin per unit volume distribution, in a spin syotem. We set

exp[-E(<p)] = N- 1 j Il dcP (P = 271"k/L) exp[-S(.p)J,

ko'O

where S(.p) is the a.ction (36.1) and the normalization N is chosen such that E{O) =O
for t "' O. The momento are then glven by

(36.20)

where Z is tM pa.rtition function

Z= J d<pexp[-E(<p)].

For q = 2p they are related to zero momentum correie.tion functions:

(35.21)

lt follows from the discussion of Section 6.6 thn.t in the infinite volume limit E(<P) =
I'('f') - r(O) where I'(<p) is the thermodyna.mic potencial as obtllinOO in perturbn.tion
theory. lt se.tisfies the RG equa.tíon

1
2'1(Y)'Pa.p ô)
éJ ~fll(g)ttjt E(t,<p,g,L,A) =R(t,g,A), (36.22)

where we re<:all the.t R is a. sewnd degree polynomial in t. Tbough for q ;:: O generic,
mn is not reln.ted to correie.tion functions, it se.ti.slies a. usual RG equa.tion which can be
easily derived from the equn.tion (36.22):

(36.23)

Another quantity of interest is tbe specillc heat C( L, t)

(36.24)
36.3 Fíeld Thenry in a Finite Geom€try: Fínite Size Sco!ing 785

96.9.1 Dimensíon.o; d > 4

We first discuss the case d > 4, which in the infinite geometry is ,..,ry
simple. To
obtain the zero mode integral at leading order, we simply neglect the <Í>k, k 'I O, modes
altogether. The action (36.1) then red uces to

(36.25)

Aftet the change of variables

'P ...... (uLdrlt• 'P, (36.26)
we find that the moment.s m, can be written:

(36.27)

in which f.,(z) is given by

/.,(<) (36.28)

with
g.. (z) = fo"" dip 10"- 1
exp [- Gz'P' + ~\O•)] . (36.29)
_;'
Equaüon {36.27) shows that above four dimensions the finite size scaling relations, proven
for a non-trivíal fixad point, and which ptedict instead for the moment ·m., a behaviour

do not hold. ln particular, instead of the axgument tLl oc (L/{) 2 one finds tL' L(d-i)/2. '·'
The extra factor L(d-•)12 ru-ises because the leading arder result depends explidtly on
u, which bas a dimension 4- d, and characterizes the vio!Ation of the naive sca.ling (see
the problem ofhypersca.ling in Section 27.1).
The result (36.27) leads to the prediction of some universal quantities. For instance
let us calculate the moments a.t T = Te (t = 0). Evalnating integral (36.29) explicitly,
we find:
g,(o) H24)"1 4 r(a/4).
Dimen.sionless ratjos are universal, for example

(36.30)

In particular for lsing-like systems the quantity 1?.4 is:

(36.31)

Anotber qua.ntity of interest is m2/mr:

n->_m2 v'2.
1 - m? (36.32)
786 Field Theory in a Finito Geometry: Fimtc Size Scaling 36.3

Note finally that at leadíug order the specilic heat C (equation (35.24)) is símply given
by
C(L,t) 1 d (m -m2) _!_ {f. - fi) + const. ,
=
4L 4 2
4U
where the constant comes from the regul.u part of the free energy.
Remark. The expreosion (36.27) is a priori only valld for tLdl 1 finite. .It is however
ea.sy to verífy that it h"" both for I < () and t > O futed the correct L - oo beha.•iour.
lndeed for t < O one tinds

m.(L,t)- (-6tfu)" 12 '""']Mo(t)]",

where Mo(t) is the infinite size spontaneous magne~ation at this order. For t > O one
obtains instea.d
L t) ~ f\(c + N)/ 2] (2 (t)L-d)"12
mu ( , r(N/ 2) X •

where ;,:(t) = 1/t is the inlinite si:z.e magnetíc suseeptibility.

The correction terms to this Jeading beha.viour bowever are inconect. To calcu)a.te
them it is necesslY)' w ta.ke ínto accouut higher order contributions.
Higher order <OntlCtiuns. lt now remains to shcw that higher otder contributians do
not invalidate the leading arder results. The contribution to :E( <p) comiog fram loop
diagrams contaíns a part divergent (UV} for la.rge cut-olf, 6:Ediv('P, t), which is the same
as in the infinite volume limit and which, according to the a.nalysis of Chapter 27, is
regular in .p 2 and t. lt is actually a polynomial at a.ny finite arder in perturbation
throry:
6:Edi• = L• LE,,..t"'<p2•.
k,m

After the change of variables (36.26), and taking int.o account that t is af order L -d/2
we find that a term proportíonal to <p 2*t"' gives a contribution of arder Ldi 2-"'-k)/ 2.
Tberefore only the terms m + k ~ 2 have to be discussed. The terms k :> O simply
renarmalize r c, t and u. The two remaínlng terms are proportianal w t a.nd t2. They
cancel in conelation functions. The latter yie)ds a consta.nt non-universal contribution
to the specilic heat.
Let us now examine the UV finite contributians, generated by tbe low momentum
region. For insta.nce the on~loop correctians are:

l:u~p (.p,L, t, u} = ! k"O -- -2 ((t + '!!...p2 ) 6;; + ~3 'I'<'I'J )]

1
L trln (6;; + -(2,.k/ L) 6

= L (1 + +!u<P:) +
4""o [In
1
(N -l)ln (1 + +lu<)].
t (36.33)
(21rk/L) (2,.k/L)

Arter subtract.ian of the UV divergent terrns, :Elloop ca.n be expandad in powers of <p and
t. Mo""""'r u.L2.p2 , !lfter the change af variables (36.26), is of order u 112 L( 4 -dl/2 . lo tbe
same way equation (36.27) shows that the cornbination tL 2 is also of arder u 112 L(<-d)J•
in the scallng region. lt followo; that a term proportion..l w .p2 ktm is of arder (uL•- 4 ) 5
w:ith o<= ~(k + m) and thus goes to zero for L large.
36.3 Field Theory in o Finí!E Geometry: Finite Sue Scolóng 787

The argument ca.n ea..ily be generalized to hígher orders in the Joop expansion because
at L fixeó perturbation theory is IR convergent and ali terms are regular functíons for
<P a.nd t small. Moreover we note that e&.b loop ín the perturbative expansion yields a
factor u which is the loop expa.ru;ion parameter, and thus a factor uL•-d for dimensional
reasons. It follows that a term af 1-loop arder proportianal to cp2•t- is of arder (uL•-•t
with 1< 1-1 + !(k+m).
The conclusian is that for d > 4 the elfective action (36.25) ca.n be simply derived fram
mea.n fidd theory, as in the infinite volume limit, the only modification coming from
the last íntegration over the avera.ge field (36.20). In particular the expresoion (36.30).,.
ind eed universal.
Finally the leading term at two-loop arder has 1< = 3(2 and thus the only signifi-
cant rorrections at one-loop arder correspond to m + k 5 2. A short calculation then
shows that for the dimensionless ratios considered above the leading corte<:tians <= h€
reproduced by repln.cing t by a quantity IL which has the form

A 1 , A 2 being twa constants. The specific heat receives an additional contribution af order
L4-à_

96.3.2 Dimensiot!Jl d = 4- •
We now use the RG arguments presenred in Sectian 36.1: instead of calculating physical
quantities as functian of {t, M, u, L}. "'e co.n, in the criticai domaín, set L = 1, u ~ u'
tbe IR foc.ad point value, then replace t by tL 11•, M (if""' íntrodure a magnetk field)
by M JPI• and tbus 'P by <pL~I· in E(<p).
At lead ing arder the fum:tion E( <p) i.s

E(I{J,L 1, t, u')

At the same arder u• can be replaced by its value a.t order e:

(36.34)

Tben replacíng t by tLll• and cp by cpLIJI•, and integratíng OYer 'P we find the moments
"'• a.t leading arder

(36.35)

wbere f, has been defined in (36.28). The equation shows that tbe •-expansion is not
uniform. Tbe method used here, in which tbe zero mode is treated separately, gives the
correct leading order only if tis assumeó to be af arder •''2 (this candítian is realízed in
particular at t e<: T- Te= 0).
Note tbe appearance ofpowers of (u') 1/ 2 wbicb, for c small, is equivalent to !:lf2 • This
ouggests that physical quantities will have an expansion in powers of E 112 tatber than c.
Tbe analysis of hígber order corrections confirms tbi.s obGervation. Let us exhibit this
phenomenon in the ane-loop approximatlon.
788 Fie1d Theory in a Finite Geometry: Finite Size Smliny 36.3

One-loop calct.Zation. The integration over the k f. O components of the lield in the
gaussian approxlrnation y:ields the one-loop contribution Í:ll=p to !; gíven in equaUon
(36.33). The divergent part of the one-loop term hM first lo be subtracted. We then set
L c: 1, u = u'. A• WE have already discussed at the end of the preceding subsection, at L
fixed ali teTffiS in perturbatíon throry can then be expande<! in powers of 'F 2 and t. After
the cltange of variables {36.26} '{' 2 bM a coefficíent proportíonal to u••l2 ~ ,•12. In the
same way, tis of arder''''· A term contributing to the 1-loop arder and proportional
to 4'2kt- is of arder ,l-l+(i<+ml/ 2 . The leading two-loop correction will come from the
terrn proportional to .,l and thus be of arder e'I'. Therefore at on,..loop arder on!y the
terms proportional to .p2 , .p2 t, cp4 t and t 2 have to be considered. The form of E(IP) a.t
one-loop order wíll thus he

E {cp,L l, t, u")- E(O, l, t, u") = ~ (t(l + n1 u•)

•
+ n2u").p2 + u"(l + aau") ~!.
Note that the correction to the coefficient of <p 4 can be eliminated by a rescaling of 'f'.
The only relevant elfect is to change a1 into a. 1 - aJ/2. But the complete coeffident
of t can be absorhed into a finite change of norrnalizatíon of t. The conclusion mthe
only correction relevant at one-loop arder is related to the coefficient a 2 whiclt we now
calculate.
The 'f'2 contribution. The coefficient ã 2 of u"<P 2 /2 in the expansion of expression
(36.33) is:
N+2t 1
- 6 - ·~ (21rk)0 •
(36.36)

The coefficíent a 2 is obtained from ã; by subtracting lhe infiníte size contribution, whiclt
is a criticai temperature shift.
Let us introduce a useful function (related to elliptic functions, see Appendix A36.2):

(36.37)

As shown in Appendix A36.2, for s ..... O

A(s)- (n/s) 112 ~ 2(1r/s) 11Ze-w't•, (36.38)

where (1r/s) 112 i• the infinite size limit.

1n terms of this function we can iormally write:

(36.39)

The integral in the r.h.s. converges exponentíally for s large, but the behaviour (36.3S)
shows that it diverges for s srna.ll. However if we subtra.et the infinite síze limit ""' obtain
a fmite result:

(36.40)
36.4 Fíeúl Theory in <> Finite Geomefrll: Finite Size Scnling 789

The coefficient a2 follows:

(,.-js)"1']. (36.41)

lntroducing rescll.led temperature t' C< t and lield rf> we can rewrite the moments at
one-loop order:
m. (L, t') L -o!a-2+,)/> J. L' I• + b) , (t' (36.42)

where the constant b is given by (36.34,36.41 ),

(36.43)

The rotio 1i4 (T0 ). From expressian (36,42), setting t = O we immediately derive the
universal dimensianless ratío 'R. 4 (T =Te) at order c11•:

g<+N(b)gt<(b)
2 (36.44}
IBHN(b)] '

g. being delined by equation {36.29). Using the value of the integTI•.I:

1
+~

ds [A4 (•)-1-7! 2 js2 ] = -l.765084B012 ... ,

0

we obtain in three dimensions for N = 1:

(36.45)

This ,.,;ult should be rompared to the mean lield value 2.188 anda Monte Carla numerical
estimate 1.6. The agreement is comparable to other results at arder E.
The apecific hetrt. The t~ tertns, both dlvergent and convergent in I: have to be taken
into account. A short c!llculation yields

C(L,t) "'L"lv [t (Mt'L 1

'" + b) J](t'L 1 1• + b)) + 4~N] + const.,

where o.gain the constant oomes fram the non-universal regular contribution.

36.4 The <1>4 Field Theory: The Cyllndrical Geometry

In what follows we consider a system infinite in one dimension hereafter called time a.nd
of linite size L witb periodic bounda:ry conditions in the rema.ining d-1 space dimensions.
Th isolate tbe zero modes W<O now expand the fields in Fourier components in the d 1
spaoe dimensions:

!f>(:t,T) = Í:: (2rr/L)d-! .,i'h<k·z/L !f>•( r}. (36.46}

I:.,.E:Z.od-1
790 Field Th.eary m a Finite G eometn;: Fínite Size Scaling 36.4

Let us again consider only the simpie exampie of correlation functions of space integrais

These CQll be calculated using only the effective actjon :E( 1p) obtained by integrating over
all componen.s .Pc{-r), k I 0:

exp(-:E!~p(r)] = N- 1 j TI [d.p(21rk/L)exp(-S(.p)],
k ..n

where the normalization N is now chosen in such a way that :E{O) = O. The effective
action I: has, at leàding orcler, the form of a quantum mecha.nical a.ction, function of the
time-.dependent patb ~p('r).
We illustrate the method witb the calculation of tbe finite size correlation length €I.·
We first examine the case d > 4 whicb is described by meau field tbeory in tbe infinite
volume limit.

36.4.1 Dimeruiom d >4

The leading order approxiroation is obta.ined by neglecting all corrections due the inte-
gration over the k 1 O components of the field:
t
l

!
(36.47)
~-
To this action is associated tbe quantum mechanical hamiltonian H (see Chapter 2):

H (36.48)

The 1\níte size correla.tlon lengtb {L is related to the two lowest eigenvalues Eo and E1
of H by:
(36.49)
After the cananícal transformation:

(36.50)

t~e eigenvalues E, can be expressed in terms oi tbe eígenvalues Ê;(z) of the hamiltonian
H:
(36.51)

by:
E;(t,u,L) = ulfJL-(4-lJ/•Ê, (tL'(d-1)/3u-2f3). (36.52)

The finite sire correlation length ~L h as thus the ft>rm:

(36.53)
36.4 Field Theory in a Finite Geometf'y: Finite Size Scaltng 791

where the function X {z) is de!ined by:

(36.54)

In ~erms of the infinite size correlatioo lengt.h ç..,, this expressíon can be rewritten:

Y(z) vzX(z). {36.55)

This expressíon exhibíts a violation of the naive ext<lnsion of the scaling from (36.10)
proven for d < 4. Tbe reason is bere, "" before, that in case af the gaussian fuced point,
the amplitude of expected leading terms vanishes and the lirst non-trivial contributions
correspond to what are ford < 4 only corrections toscaling (see Section 27.1).
Note that the canonical transformation (36.50) is equivalent in the Jangur.ge of the
effactive &etion (36.47) to:

(36.56)

"" can be verilied directly. The second t ransformation can ai50 be written in terms of
the Fourier variable w asaociated 'With r:

(36.57)

Fínally we observe that, in the limit of interest, t has to be takeo of order u 2i 3 L2!l~d)/ 3 .
Loop con-ectivns. Let us now show that loop corrections due to the integmtion over
the non-zero modes do not modify tbe scaling form (36.55) at Jeading order. These
corrections are of two types:
(i) Corrections corresponding due to UV divergent terms already present in the ;nfi..
nit<l volume limít. At any finite order divergences generate a polynomi.J in rp and its
derivatiws. After the change~; of variables (36.56) a monomíal with 2/r powers of rp and
2n derivatives is of order L~ with I( = (d- 1)(4 2/r 2n)/3. Therefore terms wbich
are allsent from the leading arder action have vanishing roeffidents. The simplest exam-
ple is 4>6 whích índuces a correction _proportíonal to Ld- 1 q6 in the hamiltonian H and
thus proportional to L -•ld-lJ/ 3 in H. The remaining terms 'With 2k + n :$ 4 simply
renormalize tbe roefficients of the leading order action.
(ii) Corrections which are finite in the large cut-olf limit. The one-loop correction
I:uoop('P) tothe elfective action, analogous toexpression (36.33), is given by:

(36.58)

The action now becomes a non·local íunctional of the lield rp(.,.) and it would seem at
first sight that we can no longer use s hamiltonian formalism to evaluate corrections to
the mean field approximatíon. ·
However we first notice that rp o.ppears, after the ch8Jige (36.56), only in the combina-
tion uL 2 rp 2 which after the chan!'." (36.56) is of arder u 213 L 2í 4 -dJ(', whlch g<>es to zero
792 Field Theory in a Fin;te Geome!ry: Finite Size Scaling 36.4

for L large. Therefore because L provides an IR cutroff we can expand in powen; of 'I'·
]n the sarne way l'lippears only ín the combination tL' which, according to the lorm
(36.53), ís of arder u 2f3Ll(<-d)f• and thus a.lso goes Lo lero for L large. For the same
reason we can expand in powers Of t. Finally we lace the problem of t.he non-local oper-
&; 2
ator !-L 2 + (2;,k) }- 1 • After the change of variables (36.56), the differential operator
L 7 8~ becomes also o( or<ler u 213 L 2 c•-dl/ 3 and thus is small compared to k 2 • 'We then
know from general arguments tbat the action (36.58) has a local expansion. Actually tbis
problem is related to the dassical problem of expanding the resolvent o! the Schrôdínger
operator at high energy:

tr [-d! + U(q)- Er' =L E- 1 2

/ -nnn{U),
n""'O

where the lln(U) are local polynomíals in the potential U.

We have thus shown that for L large the effective oction I: has, "t one-Ioop order, a
local exvansion. Similar argum ..nts apply at bígher orders in the loop expansíon, and
therefore :E( <p) h as a local expansíon to any finite order. Mo reaver we have a kind of
power counting property

(36.59)

ln additíon each new loop hrings a foctor uL<-d. Therefore at loop-order I a tertJl
with 2k powers of <p, m poWers of t and 2n derivatives is of arder (uL 4 -•)~ with " "'
l + 2(k + m + n 2)/3. For d > 4. all loop corrections have a negatiw power of L. Tbe
J
domínant correction for large L comes from the term proportional to dT 'F2 (T) which
vaníshes like L- 2 (d-<)/3.
R•marl;. Here again, although the scaling relation (36.53) has been proven only for
!L2 (d-l)/ 3 finíte, ít has the correct la.rge argurnent behaviour. For t >O

For t < O the behavíour changes drastica.lly depending whether the symmetry is contin-
uous or discrete as we discuss in detail in Subsections 36.5.2 and A36.1.2. For N > 2
we observe tl)at the lowest eigenvalues of the hamiltonirm (36.51} can, for z - -oo, he
obtained by approximatíng H hy the angular moment part, fudng the radial coordiD>Lte
at Jql = J=Gz. Tbe corresponding eigenvalues are then I!( f+ N- 2)/{ -12z). lt folloWS

12z (L 12 t d 2 Ld- 2
X(z) ,_':::""- N- 1 ;; L- --;;NJ.L - <X (~(t)'

a result which can be compared with (36.90).

For N = 1 instead, in•tantons are responsible for the splitting between the two lowest-
lying states (Chapter 40). A WKB analysís yields

(36.60)

a result whieh can be compared with (A36.18).

36.4 Fieiá Theo'), in a Finik Gwmet111: Finíte Size Scal.íng 793

36.4.2 Dimensions d
For d 4- <. a.t leading ordet, we can replace u by its IR fixed point '/alue u• which is of
arder '. We then """ the pre~eding consideratíons to analyze the leading corrections for
'small. '"" ha...., seen that the expansion parameter is actually u.L•-d, which for d > 4
is small beca use L is large, "'hile here it is small because u is of order e. Therefore the
power counting argument given above transforrns into an argument about the powers -of
e. An interaction term generated at Joop order l with m powers of L, 2k fullds tP and 2n
time derivativos is multiplied by .:•:

"= l + 2(k + m + n- 2}/3,.

The leadíng olll71oop term is thus proportional to I dqo 2(T) and of order "'/ 3. Note
that the leading tw.,..Joop correction, which we oeglect in this calculation, is o( order
e:•ll. Since the one-loop contribution to the tvro-point function i$ a constant, no term
proportional to (p2 is generated. The one-loop terrns proportional to dT (<,e2 (T}) and f
2

tI dT <,e (1') are o( order <. The f dr (<,e2 (r)) term and the term with two derivatives
2 3

coming from the 4-point funttion nre of order .!'1 3 and can be neglected.
Ftom now on the discussion closely follows the lines of Section 36.3. We set L = 1 and
u = u•. Calling a 1 u• the coefficíents oftfd7 <,e2 (T), a 2 u• of fdr\1' 2 (1') aJJd a3u• 2 o(
f dr (<,e2 (r}) 2 , we obtain a one-loop corre<:ted hamiltonian (36.48}:
2
p
H= Ld-t 1
+L d-1 [ :;(t(l 1 (u• +nau •2 ) (q2 ) ']
+ a,u•) + a 2 u")q 2 + 4i •
2

Using equation (36.53) and the RG result (36.1 O), we find the linite 5ize conelation length
{L

(L(l') "'L (u• + n3u•l) -t/l X (t' L 1" + b) ,

1 (36.61)
113
b= a2u"
2
(u" +a 3u" )- '
23
02 (~:~) +0 (<4/3),

whete t'
t' u·- 213 [1 + t(a 1 jna)] ,
is a renorrnalized temperature.
As in the esse of the hypercubic geometry we observe that the contribution coming
from n3 , which is of the same order as the two-loop contribution (<413 ), is negligible at
this order. We thus only need the coefficient i,iz of u• J dT 11' 2 /2.
The weffir:ient of 11'2 • The coefficient õ.2 of u• f dr<,e' J2 in the expansion o{ the ex-
prcssion (36 .:>8) is:
2/dw '
2'11' w2 t 1
+ (2?rk) 2 '
(36.62)

We can integrate aver w. We then use the formal identity:

(36.63)
794 Field Theory in a Finite GMmetry: Pinote Size Scaling 36.5

in which the functíoo A(s) hru; been defined hy equation (36.37). The integral has a
divergente for • small wlúch, as we have seen in Subsection 36.3.2, is cancelled by the
criticai temperature shift. We then find:

(36.64)

The finíte ~ize correiatian length at T<. Substítuting the value of a, in (36.61) we
obtain in particular tbe finite size canelation lengtb at T. at on.,.loop order:

g_ = (48,..2e) -l/3 X [K.--~1" N+ 2 (~) 1/>) (1 O(~)) (36.65)

L N +8 12 N +8 + '
with:
(36.66)

36.5 Finite Size Effects in tbe Non-Line!lr u-Model

In models with continuous symmetries, below T<, the propagator corresponding to the
Goldstone rnodes bas a pole a.t zero momentum . In a finite volume or in a cyündrical
geornetcy wítb p€riodic boundary conditions, tbe zero mode leads to IR divergences for
ali temperatura below T.. We thus have to separate the zero mode and treat it non- í
perturbatively. Let us oonsider the action of the non-linear u-model (see Chapter 30):
!
'
(36.67) ~

l
with:
q:.'(:r) = l. (36.68)
The non-linelll' u· model can be used to descríbe the IR beho.viour at fixed tempera.ture
below Te aod fixed dimension 2 < d < 4 in wlúch case a cut-olf A is required, or the wbole
pbysical range up to Te hut only within the framework of the E = d- 2 expansion. ln the 1.
la.tter case we shall tra.de tbe cut-olf se ale for a renormalized scale p. and use dímemional
regularization.
We shall ex:amine separately the two cases ofthe hypercnbíc and cylindrlcal geometries.

36.[>.1 The h117""""Ubic geometry

The zero mode can here be associated with the set of collective coordinates whích
parametrize the degenerate cla,.;ícal minima crf the potentlal. The integration over the
zero mode is then equivalent to 8Jl awra.ge over ali dired.ioll5, whicb restares the O ( N)
symmetl)' brolmn by the choice o{ a classical minímum of the a.ction around whích p€r-
turbation theory is expanded. In the case of O(N) invariant observables, the awraging is
trivial o.nd the result is simply that the zero mode has to be omitted in the perturbative
expansion. We recalJ that for d ? 2 these observables then ha.ve a finite Jímit when the
síze L becomes iDfinite. Let us, for illustration purpose, first oonsider the second moment
m, of the spin distribution (equation (36.20)):

m2 "" L -d (j ddx ofl(x) • .P(O)). (36.69)

36.5 Field Theory in a Fini!e Geometry: FiniU Size Scalíng 795

In terms of the fune\ions (30.30,30.31) m 2 has tbe s~ailing form:

(36.70)

For d > 2 mo is finíte at the crítica! tempemture te and thus

To calculate m, we pa:rametrize the field o;t>(x) in terms of 1r(x):

o;t>(x) { -vh- 1r 2{x) , 1T(.:z:)} • (36.71)

lt is convenient to set

20. (35.72)

At leàdíng order

(36.73)

{7r;(x)7r,;(O)} Õ;jll.(x), (36.74)

We thetefote lind at one-loop order

r (ei:>xk·x/L 1)
m2 = 1- 2{0) = 1 + t(N -l)(AL) 2-d f ddx ~ ( 1rk)'-
2
+O (t 2 ) (36.75)

After integration over x the BXprBSSion becomes:

m, = 1 -t{N -1}
-- 2 -(AL)
>-d
L .ç... V1 +0 (t ') , (36.76)
4
" k

a. regula.riza.tion being implied. At this order for d > 2 the dívergern:e is removed by
dividing m2 by tbe squa:re of the spontaneous magnetization M0

tA >-d ["" " •1•

1- 2(27r)d ds (;) lo . (36.77)

As usuíll we transform the sum over k ínto an integríll invohing the function (36.37).
Taking into account equa.tion (36. 71) we find for the sca.ling function (36. 70}:

~'• 1- t{~...~ 1) (AL)2 -d J.'"' ds ( Ad(s) -1- (7r/s)df>) +O (t 2). (36.78)

RG a.rguments tell us tha.t we can set L = 1 8lld replace t by tbe size.dependent coupling
tL (equation (36.14)), whicb at \eMíng arder a.t d > 2 is given by

h = t{AL) 2 -~ +O (t2 ),
796 FieLd Thenry In a Finíte Geometry: Finite Size Scalíng 36.5

We then obwn

tL(N
JiZ = 1 - - --1)
411" 2
-
1"" ( •
o
d• A (s}

expressíon valíd for 2 < d < 4 and t < t< fixed. Since t.hen t,. goes to zero for L large,
thís expression showo how Ji• approaches 1. U~ing the estimate (36.16) we find

(36.79)

'Th generate a universal amplitude we again consider the ra.tjo (36.30) n. ,... m 0 {(m2 )•12 .
Witb the definition (36.72)

Therefore
1 ~ !a(a- 2) [{0 2 ) - ((0))
2
'R,. } •

Using equations (36.73,36.74) we obtain

W" then use RG argument.s, set L= 1 and repla.ce t by t1..:

(35.80)

For d > 2 &nd t < te fixed, tL goes to zero for L la.rge, Using the estimate (35.16) we
thus find how 'R,. goes to 1:

(36.81)

The neighbourhood of the criticai temperature can be studied only within the < = d- 2
expa.nsion. Setting tL te we obtain the universal ratio 'R.. at arder i':

n. "'1 + <• a(ct N -1 1 r"" dss rA •(s) 1J.

2} 4 (N _ 2 ) 2 (211")' Jo (36.82)

lndu.;ed magnetizatíon in a small field. ln a srnall magnet.ic field h, i.e. such that hLd
is small, we can easily calculate the m<>gnetization M at Jeading arder. Since (q,(:z:)) is
not an O(N) invatiam observable, tbe average over the zero mode is not trivial. RG
equations predict:
M(h,t,L) (36.83)
36.5 Field Theo'll in a Finite Gromef'll: Finite Size Sroling 797

At leading arder, the partitlon functian in a lield ís given by:

Z(h,t,L)= f d"'<f>oõ(if>~-l)exp(L"b·</>1)/t)
1
= [d0sin"'- 2 f!exp(Ldhcos0ft). (36.84)
1T o

The integral ís a rnodified Bessel function. The magnetization is the logarithn:úc deriva-
tive of Z. At this arder M depends only on the scaling variable '"

I! (36.85)

and ís given by:

(36.86)

36.5.2 'I1u cylindriCI1! geometry

Let us now consider, as an illustration, the ca)culation of the ftnite size correlation length
in a cylindrícal geometry. At leading order the action of the zero mode is the action of
the O(N) rigid rotator which we have already discussed in Section 3.4:

A"-2Ld-l
S[<p) = __2_t_
f dT .,0 2(T). (36.87)

From equation (3.61) we infer the correlation length at low temperature, at leading order:

2 A4-2Ld-t
N-1--t-- (36.88)

C<lmbining this result witb the RG arguments of Subsection 36.1.2, we can rewrite thls
equation
{L(t} 2
(36.89)
~~ (N-l)tL.
For d > 2 in the ordered phase tL goes to zero and we find the scaling {orm:

(36.90)

We immediately learn tbat in a system in which a oontinuous symmetry is broken, at any

temperature below t,,the finite size correlation length grOWBlike Ld-l. This behaviaur
has to be contrasted with the behaviour of the correlatlon length in the case of discrete
symmetry (ooe Subsectíon A36.1.2}.

I
The criticai temperature is a RG fixed poínt and thus IL = t,. We then obtr.in tbe
universal rrotio {L/ L at leading order in t: = d 2
798 Field Theory in a Finite Geomet'll: Finite Sizr Scalíng 36.5

Finally in two dimensions for L/{00 (!) small we finrl

~ ~ ~ ln(Ç.,.,(t)/L).
L (N -1)"

One·!oop correction.s. We now calculate Lhe one-loop correctioiiS to this result for
d > 2 below te, at te in 8ll c--expa~~Sion or for d = 2 in a low Wmperature expansion.
Since we need tbe size-dependent temperature tL definerl by (36.14) at one-loop order.
let us relate it dírectly t.ó the divergence of one-loop diagra.ms. In what follows it will be
convenient to define the renormalizerl coupling constant in terms of the coefficient of e~
in the angle-dependent free energ:y. At one-loop arder the relation between bare t and
renormalized coupling tr a.t 5eale p, takes naturally tbe form

pd-2 = Ad-2
t,
(!.-
t
/32(d)
d-2
+ O(t)) • i32(d) N2: 2
"
+ O(d 2).

Differentíating with respect to A, at renormalizerl pa.rameters lixed, we obtain an equa.tion

for the RG ,!J(t) function (Section 30.3) which yie!ds

,!J(t) (c!- 2)t- /32(d)t~ +o (t3 ).

Solving then equation (36.14) perturbatively we find

2:_ = (AL)d-2 - 132(d) [(AL)d-2- I) + O(t). (36.91)

tL t d -'};

We note aJso that for d- 2 small t, ~ (d- 2)/i32(d).

1b calculate the on..-loop corrections to (36.89) we extend to higher dimensioiiS the
metbod explained in Section 3.4. lnstea.d of explicitly oonstructing the effective hamil-
tonía.n H, we evaluate tbe eigenvalues of tbe hamiltonian dírectly from the functional
integral. We con.sider a system hypercubic in d - l dimensions of linear size L, a.nd of
size tJ in the "time" direction, with tJ » L. In the time direction we ítnpose fixerl twisted
boundary conditions (whi<:h of course break translation inva.ria.nce):

· f/>(1'= O,x) r/> 1, rp(T={3,x)=!f>2, (36.92)

in whích r/>1 a.fld <1>2 are two coll5tant vectors such the.t:

(36.93)

The cornlSponding pa.rtition function ís given by:

{36.94)

By projecting Z(L,,!J,8) over the hyperspherku.l polynomiala P/"(cos6) (see eque.tion

(3.58)), we obtain tbe eigenvalues of H. We choose a fr=e su<:h that:

rl>l [1,0;0],
(36.95)
4>2 [cas8,sin8;0].
36.5 Field Theory in a Finíte Geametry: Finite Size Swling 799

"We then parametrize the lield 4>(r.x) by:

cos (9r/IJ) u 1 (r,x)- sín (9r//J)a2{T, x),

4>(r, r) sin (OT /iJ) o 1(r,x) + c:os (6T/i3) o,(;, r), (36.96)
{
'II"(T,x),

in which the field 1r(r,x) bere has only N 2 components. Wíth this parametrizatíon
the boundary conditíons (36.92) take the simple fonn:

a 1 = 1, a2 O, ,.. =O for r O and r I}, {36.97)

and the solution to the cl!13sicallield equation satisfying (36.97) is then cr, = 1, "> O,
,.. = O. Finally the transformation (36.96) is a rotatíon. Therefore the three fields
a,,a,,'IT still satisfy the constraint:

(36.98)

and the integration measure ln the functional integral is left inV&riant.

The action 5(4>) in the new fields reads:

S(a,, "'• w) ~;f drdd-lx[~ (a?+ ui)+ Út 2

+ dl + 1f 2

+ z%(&2" 1 - ú 1u2) + (ã,o- 1)2 + {a,a,)2 + (8,.,.) 2]. (36.99)

Th calculate the eiTects of small tluctuations around the classical minimum a, 1, u 2 =

O,,.= o, we climinate the field cr 1 using the constraint (36.98):

(36.100)

and expand the action in powi!IE of cr 2 and ,... The QUadratic part ofthe act\on 5 2 (., 2 , 11')
needed for the one-loop calculation is then:

S,(u2, :>r) IPA/3< Ld-t +h' jdrdd- 1 x [-

2t 2t
e:,.
I}
2 +d2 2 + w2 + (o;cr2) 2 + (8;.,.f].
(36.101)
At this order the íntegrationover o-2 gives afactor independent of6 which can be absorbed
ínto the norma!i2ation of the functíonal integral The integral over 11' yiclds a determinant
to the power (N ~ 2)/2. Hence the result a.t ont>-loop order is:

In Z(L,/},8)
Z(L,/9,0)
.!!!._A'Ld-l
211}
+ l(N-
2
2)trln [(-~- 2
13' a - T
iff) (-lf! an-'].
< T

(36.102)
The leading term of the fre.; energy shows tha.t tbe relevant values of in the integral
2
e e
1 2
are of arder t 1 • At on.,.Joop order we thus need lnZ only at arder 9 • We set

-In Z(L,f3,1J) 62 F 2 (t,L,/}}+0(0 4 ) . (36.103)

Z(L,I},O)
800 Field Th""'1! ín o Finite Geometry: Finite Size Scaling 36.5

The eígenmodes of the operator -a; - a?, given the boundary conditious ( 36. 97) in the
time direction, and períodic houndary conditions in the ·remainíng d- 1 dimensians, are

m'?r2 + (~)2
{32 L '
k
E
zd-1
'

and m is a posítive integer. lt follaws:

(36.104)

The sums nave to be understood with a large momentum cut-off. \'le naw ~uate
expression (36.104) for {Jlarge. lt is canvenient ta treat the k =O mode ooparate\y. We
ewntually find:

(36.10:'>)

lt follc-.s
F2{t,L,f1)='-Nl-22+ L ((AL)M
t
_!:!...::3.t2.).
41t !kl
k

Compadng thit expression with expressian (3.~7), expa.nded in powers of 9, we find that
the only effect of the one-loop correct.ions is to resca.le the panuneter {1. Using equatious
(3.60,3.61) we ronclude:

(L(t) = _2_ ((AL)d-

L N -1 t
2
N-2 '
-r,;:- ~ lkl .
1)
The cut.-off dependence ca.n be is ellminated by subtracting the infinite size lirnit. We
use the identity:

(36.106)

Therefore:

~ = _2_ [(LA)d-2 (~
L N-1 t
+O(t).

We fina.lly rewríte this expression in terms of the sioze.dependent roupling constant tL

given by (36.91)

= --- -
fl.fd\
d- 2
N- 2 J.~ ds (
4lf)f2 • .,fi
d 1
A- (s)-1- -
(11')\•-I)/2)] +O(tL).
8
(36.107)
36.6 Field Theory in 11 Finit< Gernnetry; Finite Size Smliqg 801

We now spedalize thís expression to various situations. For d > 2 8lld t < t, fixed ot
d 2, t - O, we obtain the correction to the leading behaviour when tL goes to zero. To
obtain the remit at t, we have to expand ín e = d- 2. The subtracted integral diverges
for s íarge for d ~ 2

lor~ fi
ds ( CJ:)
A(s}-1-0(11'-•)v,./s - - , - .
2ft

Note that the expression in the r.h.a. of (36.107) has thus, as expected, a limit for d-> 2.
The remaining integral can be explicitly calculated

l [~ ds ( . r.;T.\
ft}o fi A(s}-l-6'(1T-S)v1t/sJ"''Y-in41r,

where 1 ís Euler's constant. At the fixed point !, we have t1. = t,. We need te at arder
<2 , and it ís easy to verify that the combination l/te - fh/c has a limit independent
of the regularization scheme. Using the expression (30.54), coming from the minimal
subtraction scheme, we then find the universal ratio:

(L{t,}
-
L
-=- N -2 - e - --('Y-In4>r}+0
1 [1 + -
N -11rt:
€
N -2
(
E
2
l)] • (36.108)

36.6 Finite Size Effeds and Dynami<:6

w., have díscus.OO dynamics in Cbapters 17 and 35 from the renormalízation group poínt
of view. Previous considerations immediately apply to the dynamícs in a. finite volume.
Like in the case of sta.ti<:: properties, finíte size efrects are characterized by the dependence
in the scalíng vaTiable L/{(T). For example if the IR fixed point ís not gaussian, in the
criticai domain, tbe correlation time T(T, L) in a finite volume of linear Sí»e L has a.
scaling form which generafu.es equation (35.39):

(36.109)

Physical quantitíes, for example in a periodic hypercube, can be calculated by metbods

analogous to those u.OO in tbe case of the cylindrical geometry in pre-'Íous oections, the
time being now lhe physical time ínstead of one of the spatial directions. We bere give a
few examples of calculations of correlation times in the simple case of a purely dissipative
model without conservation laws, based on the static N -vector model.
Notalion. Because in thís oection t denotes time, we denote by t!.T oc T - Te the
deviation from tbe criticai temperature.
2
96.6.1 The ('f 2 ) theory
We consider the Langevin equation (35.9) with gaussian white noise v:

(36.110)

and:
(v;(t)) =0, (v; (t) Vj (t')) !15,j 6 {t- t'}. (36.111)
802 Field Theo'1i in a Fmite Geometry: Fínite Size Scol;ng 36.6

Vie recall tbatto the Langevin equation (36.110) we can associate a supers,vmmetrk
dynamic action S (<f>}:

(36.112)

in whích 1{ (<P) is the static eudidean 6<'tion:

(36.113)

and <f> tbe superfield:

(36.114)
In a periodic hypercube, at the criticai temperature we have to separate the 2ero mode.
At leading order tbe e!fective action for the zero made is just obtalned calculating the
6<'tíon (36.112) for a. space-independent field. We tben recognize the a.ction assodated
with a stochastic dilferential equation of a type studied in Chapter 4. The corresponding
Fokker-PJsnck hamiltonian Hyp, in hermitian form, is (equation ( 4.40)):

n [ 2 + i('VE(q)
HFP = 2Ld p
1 ) , (36.115)

with, at le.Ming order:

(36.116)

Rescaling q:
(36.117)
we find that the eigenvalues E, of HFP, as functions of L and the devlation of the criticai
temperature Cl.T, take tbe forro:

E, rzulf2L-di2e, [Ldl2.c:.Tu-ll•]. (36.118)

The scaling for d > 4 of the relaxation time 'f, whkh ís the inverse of the difference aí
the two first eígenvalues, follows:

(36.119)

As for the finite síze correlatjon length (see equatíon (36.53)}, the naive extrapolatíon of
the scaling forro valid for d < 4 is incorrect. Ao aoalysis similar to tbe ooo perforroed in
the static case shows tha.t loop corrections do not modify the scaling form for d > 4.
As for the correlation length, let us examine the behaviours of T when ~TLd/l ís large.
For Cl.T >o we find t.he expected limit T - l - n.c:.T. For ~T <o and N > 1 we obta.in

Note, howeW!r, that for N ""1 the behaviour found in tbis way is incorrect, as we discllilS
in Subsection A36.1.3.
36.6 Field Theory in a Finite Geom•try: Finde Sue Scalíng 803

The "'laxation time in d = 4 E dimensions. Let us now calcu)ata the relaxation

time in 4 - e dirnensians, at the IR fu<ed puint. The calcu!a.tions are performed in the
mínima! subtraction scheme, usíng the supersyrnmetric formulatíon of Appendíx A17.l.
Followíng the !ines of Section 36.4, we perform a rcscaling in the actíon, equivalent to
(36.117):

(36.120)

end oount the powers of e. We agaín verify that only the coellicients oi rp', t:.Trp
2
aml
(rp2 ) 2 are relevant at one--loop arder. In ..Odition the scaling of the time t and Grassm."nn
variables IJ shows that we need only the contributions proportíonal to (1, .ó.T) f dtd6d94>'
2
and f dt d9 dê (rp 2 ) • lt is easy to verify that the calculation then becomes identical to
the statíc calculation in a finíte volume. The main result is given by equation (36.4.3).
The relaxation time, at tbe IR fu<ed poínt, in the one--loop appmrimation, is then:

(36.121)

in which f(z) is the function implicitly d~fined by equatian (36.1 19), and íl'- and .ó.T'
1

are a renormalized time scale and a renormalized deviation from the criticai temperature
respeetívely.

86. 6.2 The non-linear a -model

Witb the non-linea.r a-model, we can calcu!ate for example the relaxation time at fixed
temperature below T., dose to T. In a d- 2 expansion or in two dimensions. The forro
of the dynarnic action has been given in Appendíx Al7.3:

S(rp) = A• JdGdÕdtfd"x
g
[3Jirp.
íl 88
( 8ôfJtf> 84>) + 21 (8,.rp) 2] •
8 at (36.122)

with:
(36.123)
'Th avoid oonfusions we here denote the coupling constaut or temperature by 9·
The RG equatíons of tbis model have been discussed in Section 35.2. The relaxation
time, defined in the inliníte volume by equatiotJ (35.52), satisfies the RG equa.tion:

o. (36.124)

The linite size relaxation time -r satisfies the dimensional relation:

r(A,L,g,fl) n- 1 A-2,.-(1,AL,g,l). (36.125)

The RG equation (36.124) can thus be rewritten:

(L :L+ !l(g) :g- 2 -11.., (g)) r(A,L,g,íl) =O. (36.126)

We have now setA = 1 and símplify the notatíon T(A, L,g,íl),.... r(L,g) becnuse the
dependence in n is trivial. Solving thís equation by the method of characteristics, we
find:
(36.127)
804 Field Thev'1} in a Finite Gcome!l'1}: Finite Size Sroling 36.6

with the notation YL =g(l/L) (see Subsectíon :'16.1.2) and

. dg'
ln(L)
1 -
"" {:l(g')'
(36.128)

Ç(L,g) = exp [1: ';):;;) dg'}· (36.129)

The equation (36.127) can also be rewritten:

r( I., g) = L 2 Ç(L,g)'I (LJ~(g)), (36.130)

in which {(g) ís the infinite volume correlation length. A!; we ha.ve alrea.dy sh<'lWn in
Section 36.1, at fixed coupling g < g., the elfective tempera.ture YL goes to zero. Therefore
-r r.a.n be derive<:! from perturbation theory.
Note finally that the function 7/w(Y) begins at order i' (see Section Al7 .3). Since we
calculate only at one--loop order, the functiou {(L, g) can be replaced in what follows by
a. constant renormalization factor Ç(g ):

Ç(g) = exp [f ';J(~~dg'). (36.]31)

'Ib calculate .,. we can use the method explo.lned in SubSI!Ction 36.5.2. The only differ-
ence r.ppea.rs in the form of the solution of the field equation which satisfies the boundary
conditions. lf we impose:

r,f>(-r = 0,8 = O,x) = \l,O;OJ,

(36.132)
tf>(-r = {3,8 = O,x) = !cosa,sína:; 0),

the rotation which tra.nsfarms the so!utian of the lield equation into a oonsta.nt is:

cos (~ (1 + y)]a,(t,9,~) sín [~ (t + y)] rr2 (t,8,x),

r,f>(I,B,x) =
{
sin [~ (t + ~)] "l(t,9,x) + cos (~ (t + ~)] a 2 (t,8,x),
6 133
(S . )

· ,.-(t,B,x)

We theu note that, a.s a consequence of equation (35.123), the terms of the action which
explicitly depend an a are

o? ( d
2{J'!lg L /3
We immediat.ely obto.ln the form of the rela.xatíon time, at Jea.díng arder for large L, from
the dassical cont.ributíon and renorrnalization group:

L' _ N -1
r(L,g)- fl'- 1 (g) , with fl'(g) !l( 1
(g)-2-, (36.134)
9L

where <(g) is defined by equation (36.131).

36.6 Fielrl Theory in a Finite Geometry: Finíte Size Smling 8()5

Funhermore, in the one-loop order calculation, we can omit the lield llo a.nd repla.ce
ali components of 1r(t, 8,z) other than 1r(t, O,z) by their classica.l '18lue. In terms o( this
JIISt field the action reduces to:

(36.135)

The integration over the 'IT-field yields the partition at one-loop order. Expanding for o
small we lind

F2 (L 11) ~- _1:_ In Z{L, /3,o)

' c.t2 Z(L,I1,0)

=.!!__N-2 -Ç-[m 2 2 (2'1r2!"1k2) 2]-l

w
211flg 2f32 L L
k '"""l
11' + L 2
(36.136)

From here on we follow very closely the method of Subsectíon 36.5.2. For large 11 the
r .h .s. has the expansion:

(36.137)

The relaxation time follOWli

{36.138)

The sum has to be understood with a cut-off. As we have explo.ined in Subsection 36.5.2,
lf we subtra.ct to the sum lts infinite slze limlt we obto.in o. linlte result

We then introduce the sb:e-dependent temperature YL (equation (36.91) with t ,_. g and
A 1) a.nd find

w:ith !"1'(9) defioed in (36.134). We have thus obtaioed the lirst correction to the leading
term for g < g 0 • Note ..gain that thls expressíon has a finite limit when d __, 2. lndeed

(rr /s)dl~] = -~ + 1~ ds [A 2 (s)- I- 8(,.- s)::_] {36.140)

d-2 o •

IHence we also obto.in the form of the leading corroction for d =

2 and {(t)(L lmge.
Finally we ca.n calculate the '18lue at Ih in a.n õ-expansion but the re5ult is proportional
to a time sca.le.
806 Fídd Theory in a Finite Geometry: Finite Size Scal.ing 36.6

Bibliographical Notes
Early artícles and reviews about finite size effects are
M.E. Fisher in Criticai. Phenomena, Proceedings o! the Intemattonal School of Physics
Enricn Fermi, Va.nmna 1971, M.S. Green ed, (Academic Press, New York 1972); M.E.
Fisher and M.N. Barber, Phys. Rev. Lút. 28 (1972) 1516; Y. lmry and D. Bergman,
Ph.ys. R""- A3 (1971) 1416.
See Wso
M.N. Barber in Phwe Ttnnsitions and Criticai. Phenomena vol. 8, Ç. Domb and J.
Lebowítz eds. (Academic Press, New York 1983),
and the papers oollected in
Finite Sue Sca!ing, J .L. Cardy ed. (North-Holland, Amsterdam 1988).
The field theoretical Bnalysis has been performed in
E. Bréz.in, J. PhysiqtU (Paris) 43 (1982) 15; E. Bré?.in and J. Zinn-Justín, Nud. Phy•.
B257\FS14] (1985) 867; J. Rndnick, H. Guo and D. Jasnow, J. Stat. Phys. 41 (1985)
353; M. Lüscher, Phys. utt. 118B (1982) 391; Nu& Phys. B219 (1983) 233.
The corre!ation length is calculated with twist.ed boundary conditions in
E.G. Floratos and D. Petscher, Nud. Phys. B252 (1985) 689.
A RG analysis is also given in
V. Privman and M.E. Fisher, J. Stat. Phys. 33 (1983) 385.
The almost plana.r interface model is studied in
D.J. Walla.ce and R.K.P. Zia, Phys. Rev. utt. 43 (1979} 808; D.J. Wallace in Rocent
Advanres in Field Theory and Stalistical. Mechanics, Les Houches 19S2, R. Stora Md
J.-B. Zuber eds. (North-Holland, Amsterda.m 1984).
Fínite si.u ocaling laws in Criticai Dynamics have been postulated in
M. Suzuki, PTvg. Theo~. Phys. 58 (1977) 1142.
A l!)'Stema.tic RG analy•is ís given in
J.C. Niel and J. Zinn-Justin, Nud. Phys. B280 1FS18J (1987) 335; Y.Y. Goldschmidt,
Nucl. Ph:ys. B280\FSl8j (1987) 340.
A36.1 Fidd Thwry in a Fín;te Gemn.etry: Finde Size Scaling 8()7

APPENDIX 36

A36.1 Discrete Symmetries and Finite Size Elfects

Vle have already chaxacterízed finíte size effects in the case of second order phare transi-
tions, in the criticai doma.in. We now consider, for completenes., a few examples of finite
size effects in the absence of criticai tluctuations: first order phase transitio<>S, general.
phase tra.nsitions below the criticai temperature. We recall that in the case of dlscrete
•ymmetries, the infinito sire correlatian length·ís finite in the ordered phase.

AS6.1.1 Finite volume

We here cansider first arder phase transitians in whiclt the arder parameter jumps fram
one consta.nt value to another. We restrict oursclves to the calculatíon of homogeneous
quantities like the magneti:oation. Then the basic fu nction we need is tbe free energy
-S(cp) at fixed field a.verage

(A36.l)

a quantity wh<>se physkal meaning has already been disclli!Sed in Section 6.6. In the
case of first arder trllllsitions, fluctuations are not critica.l because the correlation length
remains finito at the transition. Therefore the integration over the zerO-momentum
camponent does not solve a non-exlstent zero-made problem, but insteM the symmetry
restaration problem.
The free energy, ln presente of a constant magnetie lield h, ís given by:

(A36.2)

In a translation lnvariant finite system, 6 at. large size L behaves like:

(A36.3)

in wbich E( lp) is independent a f L.

For L laxge the integral can thus be calculated by steepest descent. In the ca.re of a
unique saddle point one finds

W(h)/Ld = -E(cp) + [Jhcp, (A36.4)

with
E'(<p) = fJh.
Note tha.t 6(cp,L) i.s such tba.t the corrections for L large to (A36.4) are exponentially
small in L beclt.use the infinite size correlation length remains finite.
Degenemte minima. 'Vhen instead severa! sBddle paints are fouud, W(h) i.s the sum
of saddle paint cantributions. As an example let us ...Sume that - are discussing an
lsíng-like system (6('1') = 6( -<,o)), in the ordered phase, in an infinitesimal mf>gnetic
806 Field Theo'1! in a Finitc Geomet.,: Finite Siu Scaling A36.l

field h. The minimum is tben almOSt degenerate. We call ±Mo the two mínima of :f:( <P)
(the generalization to any discrete set of mínima is straightforward):

(A36.5)

The free energy is now the sum of the two saddle point contríbutions:

(A36.6)

The magnetization M and the zero-field susceptibility X are then given by:
M (op) "" (t/Ld)- 1W'(h) =Mo tllllh (IJhL" Mo)+ (Jh/L-" (Mo), (A36.7)
X {3/L." (Mo), (A36.8)

where xis here defined as X=(,.?)- ((l<pl)) 2 .

In equation (A36. 7), for hLd finite, the second term is negligible, and the finite volume
magnetization takes tbe univerSal form:

M = Mo tanh (,ClhL d Mo) • (A36.9)

Let us note that in zero-field the situat\on is more subtle, as we discuss in next section.
For l'PI < Mo the dominant confígurations are no longer· almost uniform configuratkms
but clusters of one phase inslde the other. Let us call p the fraction of the total volume
occupied by the -Mo phase, O< p < 1. The cost in «Dergy ls then proportional to the
surface tension cr(T) multiplied by the minimal area which is of tbe arder Ld-lp(d-1)/d.
The average magnetization is M 0 (1 - 2p). For p small the region of traru;ition between
uniform and non-uniform wnfigurations corresponds to the balance

where the factor {(T) is implied by dimensional conslderations. When L> {(T) S(<p) is
at leading arder of the form

6(<p, L)"' L di:( Mo)+ const. a(T)L4 - 1(M Mo)(d-l)/4 .

Finally for (i la.rge enougb in a finite wlume with periodic bounda.ry conditions tbe
dominant configuration consists in two phases sepa.rated by two Hat interfaces and then
the cost in energy becomes constant and equal for an hypercube to 2a(T)L4 - 1 •

A86.J.f!. Finite size co~latian length in Jsing-like systems be!owT,

In Chapter 40 we shall relate the restoratíon of discrete symmetries in one dimension to
the exjstence of instantons. Assuming the results of Chapter 40 known, we here indicate
how tbese arguments extend to sca\ar field theories in higher dimensions in order to
demonstrate the complete parallel between the low temperature analysis of Section 23.2
for the lsing mode! and the instanton analysis. The form of the !inite size correlàtkm
length below T, will follow.
We consíder therefore the exa.mple of an effective eudidean actian in d dímensioDS of
the form:
S(rfi) (A36.10)
A36.1 Field Theory ín a Finite GeomettlF Finite Size Sroling 809

where V (<i>) is a potential which is invariant in the reftection:

(A36.11)

and has degenerat.e minima. lt could for instance have the form of the double--well
potential
(A36.12)
We want to evaluate the correlation ~L in tbe cylindrical geometry. We can deduce it
frorn tbe ratio r (see equations (23.23) and (40.6)):

(A36.13)

in whlch we recai! that P ís the operator which implements the reflection (A36.11 ). The
path integral representation of tr Pe-I!Ht. is

trPe-PHt. = j ldr!J{x,t)]exp[-S(,P)], (A36.14)

with t/J(x,-{3/2) = ~,P(x,(j/2), (A36.15)

where in Jtd we have chosen a time direction and called t the corresponding variable. With
the bounda.ry oondítions (A36.15) the leading configurations correspond to instanton
configuration in the time direction

r!J.(x, t) = fc(t), (A36.16)

in which in the large {3 limit tbe function fc(t) interpolates between the two minirna of
the potentiaJ. The corresponding ratio r then h"" the form:

(A36.17)

We haV<! identified the dassical oction S (!.) witb a surface tension which is tr aditionally
denoted hy 11(T). This msult ís anaJogous to results ohtained in Sections 23.2 for lsíng-
like S)"'tems. lnstantons replace, in continuous systems, the walls of the lattice models.
From r we derive the correlation length:

ln{L(T) ~ u(T)Ld-t. (A36.18}

This beha.viour of the linite síze corralation length, chacacteristic of the brealúng of
díscrete symmetries, has to be contr...ted with the power law behaviour (36.90) found
in the case of continuous aymmetries. The equlliion (A36.18) iH valid for temperatures
T < T<, and the surface tension vanishes at Te· Tbe hehaviour near T, iH given by
equation ( 36.60} for d > 4 and by RG arguments for 2 < d < 4:

a(T) (l( (Te- T) 3 f 2 (l( [{00 (T)J 3 for d > 4,

(l( (Te- T)"(o-L) (l( [{00 (T)]d-l for d < 4.

Remar*. In the case of one dimen.sional ínstantoru;, to integrate out tlie fluctuations
around the saddle point, one h"" to introduce collective coordinates corresponding to
the position of the saddle point. ln d dimensions, the poaition of the wall is defined
810 Field Theory in a Finite Geametry: Finíte Size Scalíng A36.1

by a function ll(:z:), whicb has also to be considere<:! a.s a set of collective coordina.tes.
'1\-anslation invaría.uce implies that the a.ction can only depend on the derivatives of
9(:z:). 1t ba.s thus t.he form:

s.H{I1) - j dd-'x [<a;e{:z:})2 +o (ta,e(r)l4)) . (A36.19)

The term with only tw0 derívatives gíves the leading rontribution a.s long as ·Lhe surface
tension i>; IArge. Wben one approoches the criticel point, terms with more derivatives
become important. General eudídean invariance tells us tbat the effective action begics
with a term proportional to the area oi tbc wall and has thus tbe form:

(A36.20)

lt has been conjectured by Wallace and Zia that such thi>; model, the almost planar
interface model, could describe tbe critica! properties of the Ising model in d 1 +E
dimensiom;. lt i>; ea.sy to verify that tbe model has an UV lixed point of order E and that
the correlation exponent 11 can be calculated os a serles in !. These properties remind us
the non-linear a-model.

A36.I.!l Dynamics in lhe orrlered phase

Let us consider a purely dissipative dynamics associated with a statíc action of tbe form
(A36.l0), for example:

A(rp) = j ddx [ ~ (a.cp(x)) 2

+ 9 ('I'' M~ 2} . [A36.2l)

We assume that our system evolves inside a bypercube of linear size L with periodic
boundary oonditians.
lt is shown in Sect.ion 40.6 that tbe di.fference ~ the two first eigenvalues of the
corresponding Fokker-Planck hamíltonian is of the order of
.,-<LO\, with .ó.A (A,.,. Am;n),

in whích A.n1n is the value of A(<p) at the degenerate mínima a.nd Am.. the value at the
saddle poiut which separatas them. We have already brielly analy.ed thi>; problem in
Sec.tion 23.4: Let us start from a conóguration in which cp is closed to Mo, and create
a bubble of tbe phase <p = - M0 • Since tbe rost in energy is proportional to the area
of tbe bubble, the saddle poínt will correspond to the situation in which the hypercube
is evenly divided between the two phases. Due to the periDdic boundary oonditions,
the minimel area surface which evenly divides the hypercube consists in two parallel
sections perpendicular to the a.xes. Such a conliguration oorresponds to an instanton
o.nti-instanton pllir of the static action. Calling o the instanton actíon in one diroeMion
we thus lind that the relaxation time TJ., which is the inversa of the second eigenvalue,
behavcs líke:
(A36.22)
Comparing with equatíon (A36.18) we find a simple relatíon between the rela.xation time
in a cubic geometJy and tbe finlte size correlation length SL in a cylindrlcal geometry:
(A36.23)
A36.2 Fi.!d Th.mry in a Finite Geometry: Fínite Síze Srolin.g 811

A36.2 Perturbation Theory in a Finite Volume

Let us add a few remarlu; concerning the calculation of Feynman diagratDJS in a finite
volume. lt is convenient to introduce Schwinger's param.eters a.nd write the momentum
space propagator [1. (p),

Ll.(p) = r~ 1-'2 1"' ds .-.(p•+... l'

a method already used in the infinite volume limit. Aft.er this transforma.tion gauS5iDll
integral6 over momenta are repla.c:ed by infinite sum over integers which ca.n no longer be""
cakulated exactly. However, dimensional continuation can be defined, and tbe infinite
size limit studied. In the chapt.er we have considered only simple one-loop diagrams D.,
which cau be wrítten
D.,-,L-d L (p•+~')-u=....!:::!..._loodss~l•-1 L .,-•(,'+,.').
p=2-<k/ L r(O' / 2) O p=2wk/L

In tenns of the function A(s) defined by (36.37),

+oo
A(e)= •L (A36.24}

the sums ca.n be written

D
.,
=- -
L-d
r(a/2)
1""
o
dss"/ 2 -le-•"' Ad(4ow2 fL 2 ) .

Poí.sson's formulais useful in thls context. Let J{x) be a function which ha.s a Fourier
tnmsform

Then from
+ro +oo
L e""'b = L .l(x-1),
k=-t'CI l-":"--00
follows Poí.sson's formula
+oo +""
2: j(kJ 2: I<l). (A36.25)
k=-oo l:;:-oo

Applying this relation to the function e_,.,, one finds tbe identíty:
A(s) (Tr/s) 1' 2 A (1r2 /s). (A36.26)
This identity shows, in particular:, tha.t tbe infinite size limit is approached exponentially
wben the mE>SS 1-' is finite

A(s)-/dpe-••' ~2(1r/s) 1 1 2 e-•'l•=l> D.,.(L)-D.,(L=oo) oc (,..L)(u-d-l)f'e-"L.

L-oo
More generally tbe fuDction
A(s;u, >.) = .,-••'113 (,\ + ivsf,.,e-•) =L e-•(n+•)'+2lcn>, (A36.27)
n
where fJ3 is an elliptic J acobi's function, satisfies
A(s:u,A) = A(s; -v, -À)= (Tr/s) 112 A{1r2 js; >-.,-v).
The exponential approach to the infinite size limit in the case of twisted l>oundar:y con-
ditions follows.
INSTANTONS IN QUANTUM MECHANICS: THE ANHARMONIC
OSCILLATOR

Up to now, to expand euclídean functlonal integral!; in perturbation theory, in the absence

of sources we have always looked for a saddle point (a. solution to the classice.l equa.tíons
of motion) corresponding to a constant lield a.nd then exp=ded the adiem around such
a. rolution. However classicaJ field equations may have non.çonsta.nt oolutions. In sta.ble
tbeories the value of the classicaJ action for these solutions ís larger tha.n for the constant
solutions beca use the gradient term always gives a positive contribution. If the difference
in BCtioo is infinite the solutions gíve no contribution to the functional integral, However
we shall find example:s in which the dilierence is linite. These oolutions are caJled, in the
case of euclidean lield theories, instanton ~olutiom. We shell show that they e.re related
to the physics of barrier penetrabon. The expa.nsion around these non-trivial solutions
IBBds to a semlclassical calcule.tion of tunnelling effects, alternative to the WKB rnethod
of Quantum Mech=ics, which can be ea.sily generalízed to Field Tbeory.
Dyoon bad conjectured that the behaviour of perturbatlon theory a.t large orders in
QED was related to the vw:uum instability for negative values of e' (e being the electric
chaq;e). Considering the quantum mechanicaJ example of the quartic anhe.rmonic osc!l·
!ator, Bender and Wu showeà that the behaviour of perturbatiOn theory at large orders
could indeed be obta.ined from a semiclassícel caJculation of b&rier penetration elfects
occurring w hen tbe sign of the coup ling constant is changed. We shall sbow in Chapter
41 how these results can be recovered from instantODB, and how tbey can be generalize<!
to a le.rge class of potentials (provlded one considers also complex fmlte octieo solutions).
As lirst shown by Lipe.tov, instanton calculus allows also to extend this e.nalysis to field
theory. We sball therefore derive in Chapters 41, 42 the le.rge order behavíour of tbe
perturba~ expansion of several lield theories. We sha.ll show how this knowledge can
be used to efficiently sum perturbation tbeory. Remnrkably enougb the main applicatioo
of these summation methods is the a.ccurate celcule.tion o f criticai exponents. The results
have hi!ea given in Cbapter 28.
Finally let us mention that, although we sha.ll only deo.l witb euclidel>tl lield theory,
ma.ny liiSpeets of the techniques we shall develop apply elso to the calculatiOn of elfect.B
coming from·finíteenergy solutions ofthe real time field equatiOns, called ~liton solutions
in tbe líterature.
In this chapter we explain the role of instantons in unstable theories on tbe simple
example of the quartic anhnrmonic oscillator in quantum meehanics snd calculate tbeir
contrlbution at leading order,
The appendiX contains some additíonal remarks about collective coordina.tes.

37'.1 Tbe Anharmonic Oscillator for Negative Coupling: Prelimlnary Consid-

erations

L<rt us ronsider the hamiltooian H of tbe quartlc a.nharmonic oscillator:

(37.1)

The ground state energy Eo(!l) can be obtained from the Jarge /3 límit ofthe partltion
37.1 lnstantons in Quantu~ Mechanics 813

function tre-/Jll:
Eo(g) lim -.!.In tre-llH.
ll-+oo {J
Moreover a sym.ematic expansion of the partition functi<>n for J3 large a.lso yields til€
energies of the excited states. We can thus deri'lle the eigenva.lues of H from tbe path
integral representation of the partition functíon:

tre-1'" = f (dq(t)] exp [-S (q(t))l, (37.2)

jo(-N2l~o(/JJ2l
where S (q) is tbe euclidean IV'tion:

(37.3)

Genera.lization of arguments appllcable to finite dimensional integTa.ls indicates that

tbe patb integral (37.2) defines s function of g ana!ytic. in the hali plane Re (g) > O. In
this domain the integral is domlnated for g small by the saddle point q(t) =O. Therefore
it can be calculated by expanding tbe integrand in powers of g and integrating term by
term. This leads to the perturbative cxpansion of the partition function and thus of the
ground state Eo (g} by taking the large {j limit.
Remarks.
(i} Note tha.t we always expand in g before tuking tbe large J3 limlt. Since EN(g), the
Nth eigenva.lue of H, sa.ti!lfies:

EN (g) = N + ~ + 0 (g),
the perturbative expa.nsion can be writl.en:

tre-PH L .,-llEw(g)
N~o

=-L ..-(N+lf2)1lL ~(-}3}. (EN l N)k. (37.4)

N~O k..O k.

We observe: that EN(g} can be deduced from the coefficient of e-(N+If 2 lll, that til€
coefficient of gk is a polynomia.l of degree k in }3.
(ü) As we have a.lready mentioned when discussing the l/>4 field theory, by resca.ling
q(t):
q(t) .... q(t)g-lf2,
we factorire the whole dependence ín 9 in front of the actlon:

S(q) (37.5)

The coupling constant g plays the same formal role as 1\. in the semiclwcsl exparu;ion.
Negati!Je coupling. For g < O the ha.miltolli.an is unbounded úom below for ali va.lues
of 9· Therefore the energy leveis, considered as analytic functions of g, must have a
singularity at g = () and the perturbation series is always dívergent.
814 lnstontons in Quantum Mechanics 37.2

A wave funct.ion ,P(t), locali~d at initial time t = O (t is hete the real physi.ml time
of tbe Schrõdinger equation) in tbe well of tbe potential near q = O, decays due to
barrier penetration. To calculate the decay rate of the wave !unction 1/J(t) we can use the
íollowing method: For g positive the time-dependent solution 1/Jo(t) of the Schrõdinger
equatian asoocíated wítb the ground state energy Eo behaves like:

We then proceed by am;lytic continuatian in the complex g plane ta go to g < O. After

ana.lytic continuation Eo becomes comple:x a.nd thus 1/Jo(t) decreases e:xponentia.lly with
time at a rate
11/Ju(t)l _ e-1 bnEol• •

IIm Eol is the inverse life..time of the wave function ,P(t). Act.ually the decay of ,P(t) also
involv"" the ímagina.ry parts of lhe rontinut<tions of a.lJ excited states. However '"" expect
on intuitive grounds that, when lhe real part of the energy increases, the corresponding
Iifetime decreases (this can easily be verified by examples). Thus at large times anly
the component rom:spondíng to the ground state survives. We •ha!l therefore herea.fter
c&.lculate 1m Eo for g sma.ll and negative.
To now understand how we can define and emlnate Eo (g) for 9 negativa, we first study
this problem on the example of a simple integral: t.he ·~ro-dimensional q,4 field theo<y».

37.2 A Simpie Integral

We cansider the integral which connt. the number of Feynman diagra.ms contributing to
the partitíon function in the <P 4 field theory:

1 =_I_ f+"" e-(z'/2+gz'/4)dx. (37.5)

J2if -oo

For g positi"" Md sma.ll, the integral is dominated by the sad.dle point at lhe origin:

I= 1 +ú(g). (37.7)

The function J(g) is ana.lytic in a cut plane. Th be ahle to perform the analytic <:ontin-
uatinn of the integral to g < O, it is nec=ary to rotate tbe contou r of integration C as
one changes the argument of g:

C: Argx=-iArgg (mod1t),
4
so that Re (g:r ) remains positive. As a consequence, one abtains two dilferent e:xpressions
for I(g) depending on the direction in wbich ane has ratated in the g-plane:

for g =- lg[ + íO : I(g) 1c+

e-(•'12 +••'1•) dx
,.
withC+: Arg:z: = -4 (mod "'), (37.8)

for g= [gl íO: I (g) = 1c.

e-(•' (l+••'!•) d:z:

with c_, Arg:z: = 4" (mod ,.). (37.9}

37.2 Instanto"-' in Quantum. M edwnic.s 815

For g - 0_, the two integral;; are still dominated by the saddle point at the origin ~ince
the contribution of the other sadàle points:

x+gx 3 =0~ x 2 = -1/g, (2.7.10)

are of the order

(37.11)

Let us however calculate the discontinuity of J(g) on the cut, which is given by tbe
difference between the two integrais:

I (g + iO) - I (g- iO) (37.12)

Flg. 37.1 The conloun; of integration c+' C-, c, Md c •.

It corresponds to the contour C+ C_ which, as figure 37.1 shows, can be deformeà

into the sum of the contours C 1 and C 2 passing through the non-trivial sadàle points
S 1 and S 2 : x ±lf A- This means that the oontríbution of the saddle point aL the
orígín cancels, and that the integral is 11ow àomillateà by these saddle poj11ts:

(37.13)

Thus for g negative an.d small, Lhe real part of the integral is gíven by perturhation theory,
while tbe expoDeDtially small imaginary part is given by the contrihutio11 of non-trivia.l
sadàle points.
Let us now try to use the same strategy for the path integral (37.2).
816 Jnstantons in Quantum Mechanics 37.3

37.3 Quantum Mecha.nics: Tbe Saddle Point

lnspired by the previous exarnple we rotate the contour in the functional q(t) space, while
we change the argument g l'rom g positive to g negatíV€'

q(t) >--> q(t) .-••,

in which 11 is time independent. If we return to the delinition of the path integral as a

limit of integrais in a discretized time (see Cbapter 2), we understand tbat this is a quite
sensible prooedure.
There is hawever one difference with the case of the simple integral: we have to stay in
a domain in which Re [f(t)] :>O, since the kinetic part J q2 (t)dt, as we have discussed
in Cba.pter 42, favours the smooth paths and ensures therefore the existence of tbe
continuum linút of the discretized path integral.
For g negative we thus integra.te along the po.th:

Arg q(t) = -0, (37.14)

which satislies the two conditlons:

Re [l(t)] > o. (37.15)

For g small, the two po.th integrais corresponding to the two o.n a!ytic continuations are
here also dominatei! by the saddle point at the origin

q(t) o'
but ii we calcula te the difference between the two integrais, this coutribution canceis. We
ha.ve to look for non-trivial saddle points, which are solntions of the euclidean equation
ofmotion:

-ij(t) + g(t) + gq3 (t) O (g<O), (37.16)

q (-/3/2) q (13!2). (37.17)

Tbe ccntributjon of the constant saddle point:

l{t) -1{g' (37.18)

is of the order of and therefore negliglble in the large /3 limit. We have to look
ef'/4 •
for solutions which have a.n action which remains fialte for /3 -+ +""· These are ca.Iled
ímtantons.
Tbe solutions of equations (37.16,37.17) represent ll. periodic motion in real time íu the
potentia! -V(q)
V(q) (37.19)

lt is clear that solutions to the equa.tion of motion eXist which correspúnd to oscillations
around the mínima. of -V, q = ±F/9. Integra.ting once equatíon (37.16) WE find:
37.4 ln.5tantons in Qu.antuf?l M echanic.s 817

witb e < O. Callíng qo a.nd q1 the points wíth q > O where tbe velodty q vanishes, we
find for the period of such a solutíon:

.8 = z1"'•, Jq + !u<t +
2
dq
2<
.

.8 canbecome la.rge only if the constant < and thus Qo go to zero. The classical trajectory
then comes closer and closer to the origin. ln the infmite {3 limit the dá.SSical rolution
beoomes:
2)112 1
9e(t) ± ( --g (37.;!0)
cosh(t-to).
The corresponding d=ical action ís:

(37.21)

Since the eudidean action is invariant under time translatíons, the classica.l solution
depends on a free parameter io whlch fdr í3 finite varies between O and í3,

Therefore, in contrast to the slmple integral, ..., do not fmd two degenerate saddle points
but two one-plll"ameter familles.
Notice also that we could bave considered trajectories oscillating n times around q 2 ~
-ljg in the time ínterval íJ. lt is msy to verify that the corresponding action in the
large 13 limit becomes:
S (q.,) ,;: -n4/3g, (37.22)
=d yields therefore a cantribution proportional to e.-.41 3 •. For g small tbe patb integrol
is dominated by the tenn n = 1.

37.4 lnstanton Contribution at Leadlng OTder

The gaussian approrimation. Th evaluate the contribution, at leaàing order, of the

saddle poínts let us expand the action around a saddle poiot and perform the gaussian
integration (one-loop order). Setting:

q(t} = q.(t} + r(t) ,

W<>find:

Imtre-~H a: 7e•l 3g f ldr(t)]exp (-~f dt 1dt 2 r (ft) M (tt.t2 ) r(~)) ,

oc!i e"l"•(clet M)-112,
for g _, 0-, 13 _, oo. The operator M ís giren by:

(37.23)
BIS lnstantons in Quantum Medianics 37.4

The zero ,.,de. Let us differentiate equatíon (37.16) with respect to !:

2
- ( ~) 'Íc (!) + <ic{t) + 3gq;(t)q,(t) =O. (37.24)

Since the function r),(t) is square integrable, this equation implies that q,(t) is an eigen-
vector of M with eig•mvalue zero:
(37.25)
1 2
Hence the naive gaussian àpproximatíon yields a result, proportional to (detMj- 1 ,
whlch is in6nit<J!
We should not be too surprised by this result: as _, have noted abo•e, due to trans-
lation inva.ria.m:e in time, Vt."e have two one-parameter fa.rnilies of continnously connected
degenerate saddle points. An infinitesimal va.ristion of q(t) which rorresponds to a varj..
ation of the parameter t 0 , i.e. proportional to <Íc• leaves the action unchanged. The
problem W<! her-e face is by no means special to path int<Jgrals as the follow'ing example
shaws.

91..f1 Zero modes in stmple integroLs

Let us agaín consider a simple integral:

1= f dvx .,x'-o(x
2 2
) , (37.26)

in whicb x ill an v-comp<.ment vector, and tbe integrand i!; O( v) inve.tiant. For g smBII,
tbis integral can be calculated by steepa.st descent. Tbe saddle point is gíven by

Xc (l- 2gxD O, (37.27)

that is
(37.28)
We here alro find n v - I parameter fe.mily of degenerat<J saddle points, since only the
lengtb of tbe vector Xc is determine<! by tbe saddle point equation. If we single out on.-
saddle point a.nd evaluate its contribution in tbe gaussían approx:imation, we are led to
calculate tbe determinant of the matrix M,~:

(37.29)

which i!; the projector on x and has therefore (11- 1) zero eigenvalues,
The solution to this problem is here obvious: it is necessary to calcu!ate the integral
over angular variables exactly; only tbe integral over the radial variable can be evalue.ted
by the st....pa.st descent method.
In the case of the patb int<Jgral it i!; al.so necesse.ry to integrate =tly over the pa-
rarneters whicb de!!cribe the ...ddle points, in the example above the time translation
pa.ramet<Jr. A suitable set of integration variables has to be [ound in whlch the time
parameter explicltly appears. This i.s tbe method of so-called collective roordinates.
Remarlc. We bave alteady studied theories possessing a symmetry group in which the
2
dassical minimum is not invariant under the group, for example tbe ( .P2) field theory
in the ordered phase:

S[rJ>] =~f ddr [~ (â,.rj>(r)) 2 + ~.p2 (r)+ 1g(rJ>2 (x)) 2] . (37.30)

37.4 Imtantotu; ín Quanl""! M•chanics 819

In thls case we h ave general ly chooen one da.ssical minlmum and made a syst.ematic
expansion around it. However tlús is a sensible prooedure only when tbe symmetry is
spontaneously bmken. We bave a.ctually seen in Chapter 30 tbat the absence ofsymmetry
breaking ID&llifests itself in perturbation theory by the appearance of IR singula.rities. ln
tbe case of instllllton solution.s we bave shown above that the propagator in an instanton
6eld has an isolated pale at the origin. TherefCJre ali terms in perturbation theory are
in6nit.e. We conclude that the time translation symmetry is not spontaneously broken
and Lhat it is necessary to sum over ali degenerat.e sadd!e points (see also Chapt.er 23).

37.J,.l? Callective coordinate8 and gaussian integrotivn

In order to take a va.riable a.ssociated with time translation as an integration variable,
we set:
q(t) q:O(t- to)+ r (t to),
(37.31)
{ r(t) = Lq.,f.,(t),
,_L
in which to is no longer a parameter, but !Qrms, together with tbe set {q,} the new set of
integratíon va.riables. We impose to the functions {!.. } to forro an orthonormal basis up
to a polnt. An infinitesimal variation of to aclds to q(t) a term proportional to q,. lf we
want our new set {to, {qn)] to consist only of independent variables, we have to impose
a oonstta.int. We choa;e:

f q, (t -to} fn (t -to) dt O Vn. (37.32)

We see tba.t the normalised eigenvectors of the hetmlti= operator M (!t, !2) correspond-
ing to the non-vanishing eigenvalues satlsfy ali the condítions imposed to the functions
fn(t) slnce they are orthogonal to the last eigen.....ctor 'ic and form witb q, a complete
basis. With this choire the argument of the exponential in the r.b.s. takEO> a very símple
furm:
(37.33)

in which {m,.} forms the :set o f a.!] non-vanishing eigenvalueo; of M.

lt is useful at this point to remember th.at the functional measure (dq(t)] can be defined
in tbe continuum as the flat int.egratiou measure on the coefficients c,. of the expansion
of q{t) on an orthonormal basis of square integrable functions (see equation (2.41)):

00

q(t) L C,.9m(t),
m-0

(dq(t)) = N n
00

m..O
de,..

The jacobian of the transformation which goes from the set {c,} to the set {to, {q.,}} ís
given by (see Appendix A37):

J = li !o(!- to) d~o [q, (t to)+ f: f,.

n=l
(t- lo) qn] dtl. (37.34)
B20 }nstantons in Quantum Mech.anic.s 37.4

At leading arder in g we can neglect the dependence in q.:

J -ll<icll [! q~(t}dt] 112

(37.35)

Since tbe integrand doe>; not depend on Lo, the integration over the collective coordinate
2
to simply yíelds a fact.or {3. The in!A3gratíon over the varíables qn ylelds ( det' M) -l/ , in
whlch det' M is the product of ali non-vanisbing eígen,•alues of M, i.e. the determlnant
of M in the subspace orthogonal to q,.
The nonnalization. To normalize tbe path integral, we compare it to its va.IU€ at g O
(the harmonic oscilla.tor), which, in the large fJ limit, reduces to e·N 2 . At 9 = O the
operator M becomes the operator M 0:

(37.36)

Wben comparing the instaoton contribution and the reference path integral correspond-
ing to the harmonic osdllator, we h ave to remember that we have excluded one mede
from the ga.ussian integration in the case of the insta.nton. We have therefore a difference
of one gaussian integration between the two pa.th integrais. We have to divide by a. fa.ctor:

the ínsta.nton contribution.

Taking into acc.ount the two fa.milies of sa.ddle points and collecting all factors we
obt.aln:

(37.37)

S1..f. 9 The re.rult at hading arder

In Appendix A3S .2, we sha!l compare this calculation with the corresponding WKB cal-
culation. We shall indirectly show tha.t for all systems for which we can solve explicítly
the cla.ssical equations of motíon with arbitrary boundary conditions, we can also expüc-
itly cal cu late the determina.nt or the opera.tor gavernlng the small fluctúations a.round the
cla.ssical trajectory, In the special case considere<! here, M ha.ppens to bethe harnUtonlan
correspondíng to an eJ<BCtly soluble Bargrnann potential, and one finds:

-1 ~ -1 v"f"+"ê 2
det (M +c) (Mo+ õ) = ,; ,...,..-:; (37.38)
l+t+lvl+!+ 2

Simple arguments teU us that the ground state wa"" function has no node, the wave
fullctíon of the first excited state one nade, etc. The wave function ric(t) vanishes once
at the turning point, therefore it does not correspond to the ground state. There eocists
one other state whlch is the ground state, ..nd M has therefore one negative eigenvalue,
as is nppa.rent on expression (37.38).
37.4 Instanton• ín Quantum .Meclumíco 821

lt can be verified, by expressing the deteruúnant o f M as a product of eigenvalues that:

det' M (detMo)- 1 = lim! det (M +e) (Mo +e)- 1 • (37.39)

E-DE

Then from (37.38) we conclude

det' M (detM0 )- 1 (37.40)

We still have an a.mbiguity of sign sjnce we have to take the square rool of expression
(37.38). This amblguity can only be rerolved by followíng the nnalytic contlnuation from
g positive to g uegative, However in both cases the square root of tbe determinant is
imaginary, so tbat the final resu)t is real, as expected.
Tbe jarobian J is easily calculable, but tbe important point is tbat P is proportional,
as is the classical solution itself, to lfg. As we shall see !ater, tlús is tbe first e:xample of
a general situe.tion: ee.ch time the instanton solution breaks some continuous symmetry
of tbe classica.l e.ction, tbe oolution depends on pararnet.ers generated by tbe action of the
symmetry group on the solution. Ee.ch paro.metei has to be taken as an integration vari-
able and the corresponding je.cohian gtmerates as a factor the )oop expansion parameter
to tbe power -l/2. Here we find:

J (37.41)

The expression then becomes:

-pe-IJ/ 2 ~--e413s [1 +o (g,e-11)]'

1
lmtre-IIH
.j2; ...;=g
for 9-+ o_' f3-+ oo.
(37.42)
For {3 large, the l.h.s. has tbe form:

We have learned that fur g small tbe imaginary part of Eo is exponentia!ly small. Since
the small g limít has alway,; to be taken befure tbe large f3 limit, we can write:

Imtre-11H ~ -{3lm(Eo(g))e-i'ReEo(oJ ~ -f3e-ai'ImEo. (37.44)

Equation (37.42) then leads to:

lmEo(g) (37.45)

A systematic expansion around the saddle point then generates a.n expansion in powers
of g.
Remark. We have obtained the behaviour, for 9 sme.ll e.nd negative, of tbe imaginary
part of the ground state energy and therefore the decay rate of a state lo<&ll:red in tbe
unbounded potentio.l corresponding to the anharmonic oscillator with negátive <:OUpling.
1n Chapter 41 we she.ll show that this re5Ult also leads to a.n evaluation of tbe large arder
behaviour of perturbation seri"" lbr the Bllhannonic oscillator.
822 Jrutanton.s in Quanlum Mechani""' 37.4

Bib1iographical Notes
Common references to Chapters 37-43 are
J. ZiDD-Justin, Recent Advances in Field Tluwry and Statisiirol MechaniC&, Les Hauches
1982, R. Stora and J.-R Zuber eds. (Eisevier, Amsterda.m 1984),
whicb we follow cloo;e]y n.nd w hich also contains additional references and
S. Coleman, The Whys of Subnudror Physics, Erit:e 1977, A. Zichícbi ed. (Pienum,
New York 1979).
For an early investiga.tion of the analytic properties of the anharmonic osci!lator see
C.M. Bender a.nd T.T. Wu, Phys. Rev. 184 (1969) 1231.
The content of Cbapter 37 is ba.sed on
E. Brézin, J.C. Le Guillou and J. Zinn-Justin, Phys. Retl. 015 (1977) 1544.
Tbe method of c:olleetive coordinates wa.s first described in thiil context by
J. Zít~ and J.S. La.nger, Phys. RetJ, 14S (1966) 741; J.S. Langer, Ann. Phya. (NY)
41 (1967) 108.
The analytic contlnu,.tion in the g-p!ane is di.scussed in
G. Paris\, Phys. Lett. 66B {1977) 167; E.B. Bogolmony, Phys. Lett. 67B (1977) 193.
For the calculation of tbe determina.nt a:risíng in the gaussio.n integr<ltion around the
instanton see for example
R.G. Newton, Scattering Th<ory of Waves and Pamd~ {Me Graw-Híll, New York
1965, 2nd edn. Springer· Verlag, New York 1982) p. 433.
Rigorous results conceming the spectrum of the quartíc anharmonic oscillator can be
found in
J.J. Loe!Iel, A. Martin, B. Símon and A.S. Wíghtman, Phy1. Lett. 30B (1968) 1656;
S. Gro.ffi, V. Grecchi and B. Simon, Phys. Lett. 32B (1970) 631; B. Simon, F\mctionol
Jntegrntion and Qullnturn Physia (Academic Pre!S, New York 1979).

Exerci....s
&erdse 37.1
Calculate tbe imaginary part of the ground state energy for tbe hamiltooian

H -~ (d/dd drf + jgq~N,

for g negative.
A result usef~l for ali exercises: if M (tJ, t~) is of the form:

M(t1,t2) [-(d~J+l >.~2 ~~l5(tl-t2),

in the uotation of equatian (37 .38) we have:

Exercise 37.1!
Calculate the imaginary part of th" ground state energy for tbe hamiltonian witb O( v)
intemalsymmetry
H=
A37.1 lnstantcns in Quantum Mechanics 823

APPENOIX 37
THE JACDBIAN, CONSTRAINTS

1n the appendix we first calculate the jacobian of the transformation from q(t) to the set
to, qn !llld then indicate bow collecti\lll coordinates can also be introduced by a conmaint
equation.

A37.1 The Jacobian

Let u• expand q(t) on a complete set of real orthonormal functions (in the 1:.2 sense)
satísfyíng periodic boundary conditions on [0, P],
00

q(t) = L Cm9m(t), {A37.1)

6nm = l o,
dtg,.(t)gn(t). (A37.2)

In the explictt eva.luation of Section 2.3 (see equation (2.41)), we have defined the func-
tional measure as
ldq(t >I ""' N
"" de.. •
TI (A37.3)
"m.=O

in which N ís the usual constant normSJ.lliation factor. We now change variab!es

q(t) q, (t -to) +L f,. (t to) q.. , (A37.4)

n=l

in such a way that the set {9e (t ~to} ,f,. (t- to)} forms an ortbogonal basil;, and the
in'• are normed. Tbe new variables are t 0 and the set {q.. ] .
Let us calculate the varia.bles em in terms of the new variables:

c,= f dtg..,(t)q, (t- to)+ f::q ..

nw.l
f dtOm(t)fn (t to). (A37.5)

The jacoblan of the transformation is the detsrmina.nt of the matrix

élc,
[ Bto' õq,.
&c,] ' (A37.6}

with:

to),

(A37.7)

Because we calcula.te at leading order only we can neglect tbe dependence of tbe jacobian
on the {q.} . Since the set ·

{
qt(t -to)
l<lcl
f}
' n '
824 Imtanton.s in Quantum Mechanics A37.2

forrns a.n orthonormal basis, the determina.nt of the matrix

(A37.8)

is the determína.nt of an orlhogonal matrix whkh is one. Therefore the iacobian of our
transforrnation is just ll4c (t !o)ll: •

(A37.!1)

A37 .2 Constraint Equatlons

Another method ca.n be used to introduce collective coordinates, which yields identical
results. Inspired by the Faddeev-Popav quanti'Oati<m method o f gauge theories, we start
from the identity:

(A37.10)

We insert (A37.10) into the p&th integral and replaces the 6-function by its Fourier
representatíon. Calling >. tbe varíable which ímposes (A37.10), we can write the total
action S{q(t), >.]:

S[g(t),).] = S[q(t}] +i>. f dtq.(t)(g(t +to)- q,(t)). , (A37.11)

The variatíon of the total action S[q(t), >.] then vanishes for q(t) = qc(t- t 0 ), >. O. The
detenninant generated by the gaussian integratioo around the saddle point is now a 2 x 2
matrix with non-diagonal elements

hq~~a>.. S[q,>..] = iq.(t to),

This modifies tbe determínant

The modífied operator has ali the same eigenvalues as the initial one but one: The
eigenvalue which corresponds to the cigenvectcr rj, is now lltiell 2 instead of O. Ta.king into
account tbis modificat;on of tbe determínant and the first factor in (A37.10) at leading
order one exactly recovers the j!LCObia.n of previo11S section. Note finally that the factor
{:l1r)- 1/ 2 in (37.37} a.rises as the ratío of a factor (27r) 112 generated by the additional
gaussia.n integcaticm dueto>. a.nd the factor l/(27r) of the Fourier representation of the
6-function.
38 INSTANTONS IN QUANTUM MECHANICS: GENERALJZATION

We now want to calculate for a general class of potentials, still in Quam um Mecllanics, the
decay rate of a wave packet located at initial time at a relative mininmm of a potential,
generaliziug the method of Chapter 37. To guide our intuition, we a.gain imagine that wé
start from a situation in which a gíven minimum of a potential is an absolute minímum,
and after some analytic continuation becomes a relatíve minimum oi the potential. As
we h ave discussed in the previous chapter, the correspondÍllg ground state energy then
becomes complex in the aoalytic contlnuation, and its imaginary part yíelds the inverse
lifutime of a wave packet concentrated around the minimum ofthe potential and decaying
through barrier penetration. In tbe semidassical !;mit the decay rate is again re]ated
to finite action, í.e. instanton solutions of the euclidean classical equatíons. We shall
calculllte, at leading arder, the contribut~n of instantons to the imaginary part of the
energy.
The appendix contains, to compare, analogous calculations using WKB methods, and
some remark about the average oction in path ÍllWgtals.

38.1 The Instantan Contribution

The ínstanton salution. We consider h•miltonians oi tbe form:

(38.1)

The fuuction V(q) is an ano.lytic function oi q which, for q small, behaves like:

V[q) !t/ +0 (q3 ). [38.2)

Again in the harniltonian (38.1) the pot.cntial has been pararnetrized in such a way that
g plays the formal role of li.
The path integral representation of the partition function tre-tJH is:

(38.3)

(38.4)

In the situation that we are considering we know that instanton solutions exist: because
q = O is only a relatlve minimum of the potential, the function V[q) which we haVe
assumed regular and thus continuous, has at ieo.st another zero. For {3 inlinite, one
lnstanton solution q,(t) starts from the origín at time -oo, is reflected on the zero of the
potential, and comes back to the origin at time +oo.
Let us write the eudidean equation of mation:

1 av
Qc(l) .;{&J (9c ./9) • (38.5)
826 Jnstantons in Quantum Mechanics: Genemlízation 38.1

Integrating once, we obtain for a finite action solution (for {3 = co):

(38.6)

Calling :to the relevant zero of V (x), we can write the corresponding action:

+"" A
s {g.)
1-coo
ett)dt =-
c g'
(38.7)

A 2 [ " J2V(x)dx. (38.8)

We notice iliat tbe dassical action is positive and proportlonal to g-1.

The gaUJsian integrotian. To calcula te the instanton contribution at leading arder, we
have to integrate around tht> saddle point q.(t) in the gaussíWJ approximation. However,
as we have shown in previous section, we must before separate a collective coordinate
corresponding to time transla.tion. This yields a factor f3 and a jacobia.n J:

J [ L., q~(t)dt]1/2
+oo
(38.9)

Since the functíon q,(t), which is an eigenfunction of 62 SJ (6q.) 2 witb eigenvalue zero,
has a nade at tbe turning point :ro j.,;g, iliere exists an eigenfunction associated with
a negatjve eigenvalue. Consequently, the determinant which appears in the gaussian
integration a.round tbe saddle point is negative.
Collecting ell factors we obtain:

lmtre-PH ~ ~_p_ e-f1!• ~ [det Mo (-det MJ- 1] 112 e-AI• 1

2y'21T V9
for g __, O , f3 -HJO • (38.10)

This yiekls for the imegina.ry part of the "ground state" energy E 0 :

ImEo=-I~[
- 1
detMo(-det'M) )1/2 e-AJg. _ (38.11)
2 21fg

The definitions of Mo, M, det' are tbe same as in the preceding chapter:

2
Mo(!" t,) = [- (d fdtt) + 1]6 (t1 t2),
(38.12)
and det' means determina.nt in the subspa.ce orthogonal to q0 •
To complete the calculation we sball explicitly evaluate dee M in qua.ntum mecha.nics.
In Appendix A38.2, to compare, we sball recall how the calculatíon is dane with the
WKB method.
38.2 Instantons in Quanturn Mechanics: Genernlizatúm 827

38.2 Calculation of the Determinant: The Shifting Method

For reasons which will become apparent !ater, we first calculate the general matrix ele-
ment:

J.q{-fi/Z)~•·
q(fJ/2)~%

(x' ie-ilHI x) [dq(t}] exp [-S(q{t))]. (38.13)

Let us call S. (x', x; t3) the dMSical action oorresponding to the classical solution w:ith
the prescribed hounda.ry conditions:

{
S, (x',x;P) =~:: [~g~(t) + g
1
V(q,(t)yg)] dt, (38.14)
Qc ( -{J/2)- X , Qc {/3/2) =X,

Setting:
q(t) = q.(t) + r(tl, (38.15)
we obta.in, at Jeading arder, the path intel?'"al:

L
(fjf~)~
{:r' ie-BH j x) .,-s. (dr(t)] exp [-E(r(t))],
r(-8/2)-0
(38.16)
f!/2 1
E(r(t)) =
J-f!/2
2 [r 2 (t) +v" (q.-/9) r 2 (t)j dt.

We now ca.lculnte the gaussian íntegra.! over r(t) using the so-called shifting melhod. The
main dr:awback of this method is tbat it ínvolves a dangerous cha.ng<! of variables and
the final result is, as we sball see, slightly íll-defined. On the otber hand it allows a
ra.ther strnightforward evalua.tion of the determina.nt. The idea behind the calculotion is
that lf we know tbe solutions of the classica.l equation of motion for a.rbitrary boundary
oonditions, we can construct a canonical tra.nsforma.tlon which mops any hamiltonian
system onto a stnndard one (here we chOO<!e a fr.., hamiltonian). For det.ails see Appendix
A38.L

38.fU The shí/ting metlwd

We first set:
V" (ygq.(t)) = i<(t) /tt(t) • (38.17)
We know at Jeast one possible choice for tt{t). [f we differentiate the equation of motion
(38.5), we find:

(ft)' 1/c{t) V'(qcJ9)qc(t). (38.18)

]f q.(t) does not vanísh on tbe dassica.l tra.jectory, we choose

><(t} = Qe(t) • (38.19)

Otherwlse we look for o linear combination of tbe two independent solutions of equotíon
(38.17), q.(t) and
828 lnstonton.s ín Quantum Mechani.c.s: G enerolization 38.2

which does not vanish on the dassical trajectory.

The actíon in e:xpression (38.16) then becomes:

(38.20)

In the integration by parts, the integrated terms vanisb dueto the boundary conditions.
The evaluation of the r .h.s. of equation (38.20) leads to the rema:kable identity:

(J/2 1 [ ' ]2 dt
l - r(t)- ~r (t)
-IJ/2 2 "
(38.21)

This suggests an obvious linear change of variable, r(t)- a(t):

r(t) - :i!; r(t) = â(t)' u( -{1/2) = O, (38.22)

whlcb tra.nsforms the herniltonian o f the time-dependent harmonic o.sdllator, into the
free hamiltonian:

1 -~/2
2
~1 -l [r 2 (t) + =(t)r2(t)
2
i<(t1
"
]
dt =
t/ 2

-(J/2 2
1
- [& 2 (1)] dt. (38.23)

The change of ''anables (38.22) reminds us of the Langev;n e<juation we discussed in

Chapter 4, r(t) playíng the role of the dynernical variahle a.nd iT the role of tbe ooíse.
We are therefore not too sutprised to encounter the same difliculty as in the derivation
of the Fokker-Planck equatíon. Iotegratlng e<jllation (38.22), we obtain

r(t) = t<.(t)j' dr~((r) ~u(t)+J<(t)j' dn:r(r) ~(r)). (38.'24)

-0/2 "'r -N2 K (r
The jacobian J of tbis transformation is formally the determinant of the kernel (see
Section 4.5)

(l8.25)

Us.ing:
lndet(l+M) trln(l+M)=trM ~ttM'+ ... (38.26)
we get:
2
lo J = 6(0) ff;/ dt i<((tt)) (38.27}
}_fJ/2 "
As we have shown in Section 4.2, bnd a!ready used in Section 4.6, symmetry coRBidera-
tions lead to the prescription:
6'(0) = ~. (38.28)
Integratiog over t we then obtbin the jacobian J:

J (38.29)
38.2 Instanwns í'1 Qua'1tum Meclumics; Genernlizatian S29

In the path integral we stlll have to impose the boundary condition:

fi/ 2 ú(t)
o T (f'/2) = tt(f'/2)
l -IJ/2
df -() ·
X. t
(38.30)

Tlm can be achieved by introducíng a 8-function fur which, as usual, we use a Fourier
repres('ntation

6Hfl/2)) = ~t(iJ/2)
1 f 2;-dÀ
exp
( {"
12
i~ J-fi/2 dt ~<(t)
&(t))
.

The complete expression tben reads.:

(38.31)

with
(~.:r•(t) -i~ ú(t)) .
S(a,À)
l fJ/2
-/3/2
d!
J<(t)
(38.32)

Th elirninate the term linear in ér(t) in equation (38.32), we shíft é(t),

À
ú(t) =i l<(t) + ú'(t). (38.33)

After this shift the path integral becomes

{x'le _JlHI x) ~ e-s. 1u(-8/2)-D

[du(t)] dJ.
2..
I
v'x(/3 /2)1<( -1)/2)
x exp 1 2
--À
lfJ/2
--
dt 1
--
jfi/ 2
b 2 (t)dt] . (38.34)
[ 2 -IJ/2 1<2 (t) 2 -fi{Z

We integrate over ~:

(38.35)

The constant N (I>) does not depend on :>: and z', a.nd ís proportional to the matrix
element (O!e-PHolo) in wbich H0 ls the free ha.miltonia.n:

(38.36)

1b determine N(i>) We set H= H 0 in equation (38.35) and note that in this case tt(t) ís
a. constaut.
Tbe final r""nlt i.:

(38.37)
S30 lnstantons i1l Quanturn Mechanics: Genernlization 38.2

We leave as an exercíse to show that the .result (38.37) is fonna.lly independent of the
partícula.r línea.r combination of the two solutíons of equatíon (38.17) one has chasen.
To obtai.n a more explicit expression we then substítute for example ~<(t) = q,(t). We
'·ctegrate the equation of motion (38.5), taking into account the baundary conditions:
~oi~ "'9-l 1V (q,(t)Jg) + EJ' {38.38)

1'""
and therefare
13 dq (38.39)
'.fi [2 (E+ V(q)WI 2 •
Differentiating equation (38.39) witb respect to f3 we obtai.n:
.fi âE
(38.40)
1 =- [ .fi 12 (E+
whicb can be rewritten:
aE (38.41)
â{J =
The result ( 38.37) can then also be written

(x' !e-8H lx) ~ e-S.(•, ,z:8) 1

y'2lfli<{/3/2)<ic(-f3/2)
(-aE) 11
'--
8{3
{38.42)

We lea.ve as an exercise to verify that one also has tbe identity

)-I
1
~'' dt ( a•s
"(/3/'2) "(-/3/2) -P/ 2 _.2(t} = - âxâ~ (38.43)

Substituting equation (38.43) into equation (38.37) one tben obtains Va.n Vleck'' formula:
112
1 IPS )
(x'je-llHix)~ ( -2,.axa~ exp!-S,(x',:r;/3)]. (38.44}

Seveml degrees of frer.dom. Tbe calculation of tbe instanton contribution by the shifting
method can be generali:red to d > 1 degrees of frea.dom provided one can find a oon-
singular matrix K solution of the equation:
k- = av (q,(t)) K (38.45)
•J aq,aqk *1 ·
The cbange of variables (38.22) then takes the forro:
(38.46)
The matri:x K can be chosen in such a way that k x-• is cymmetric. 1t is then easy to
verify that ali arguments can be repeated and one finally obtains an expression simila.r
to (38.42):

(x' 1·-~Hix)
r
~ {(21t)ddet lK (~) K (-o L~/2 dt {TK)- 1 x - l
M2 ] }-l/2

(38.47)
This expression is again equivnlent to Van Vleek's formula (see Appendix A38.2) and can
be derived in the same conditions, i.e. if the classícal equations of motion can be solved
for a.rbitrary írútial and final conditions. Fot more tban one degree of freedom this is
no longer tbe generic situation and corresponds instead t.o the special dass of integrable
hamiltonians. A simple example is provided by O( d) invariant •ystems.
38.2 lnstantons in Quantum Mechanícs: Generolízation 831

98.2.2 The partition function

In order to calculate tre-!lH we can now irnpose periodic bounda.ry conditions. Then

[q,(íJ/2)g.(-tl/2)r 112 {g2 [V (:rJg)+ E] }-1/2 · (38.48)

lntegrating over x we obta.iu the trace. Using equatíon (38.39) we find

]-l/2
f [g
2
d:r (V (zy'g) +E) (38.49)

We now collect ali factors and obtain a more e...plidt expressíon:

12
13
lmtre-IIH ~- ( -&E
--1 )' e-A(fi)/9 (38.50)
2i &p 2lfg '

where E(P) u.nd A(p) a.:re defined by:

P 21'+ [2 '
r_
dz
(E (PJ +li (x})] 1/ 2 '
(38.51)

A(p)=21~" d:r[2(E(p)+V(z))J 112 fjE(/3). (38.52)

The quantities "'+ u.nd :r_ a.:re the zeros of [E+ V (:r)). Notice the useful relatíon

ãAffJP ""-E(/3) (38.53)

lt is clea.:r, a.t least for p large enough, tha.t E (P') is a negative jncrea.síng function of {3.
Therefore -8E /8{5 is negativa and the result is real as expected:

fJ (8E
Imtre-/iH- - - -- 1 ) ''' e-AWl!• for g ...... O. (38.54)
2 &{3 21rg

This completes the calculation at finite (3.

Romari:. At f5 finite the calculation is valid only above some critica! value {50 • Indeed
when f3 decreases, "'+ and z_ approach a common value x 0 whicb corresponds to a
maxlmum aí V(:r):

V (x) ~ Vo- !w2 (x- xo)2 +O [<x- xo) 3] , (38.55)

{ V0 >O.
Let us parantetrize E, z+ and x_,

E"' -Vo + !w2 <2 , :I:±= .xo±e, (38.56)

tben
f5 (38.57)

For f5 $ {3., no instanton solution can be [ound and it is on the cantrary the perturbative
expansion around the classical extremum z -= z 0 of the potential whicb is rel€'VB.Ilt.
832 ]rutantons ín Quantum Mf!t:.ha.nie.s: Ge.nern.lizatio-n 38.3

38.3 The Large (3 Lhnit

Writing equation (38.51) as

•+
{3 = 2
1
_ { [2 (V (x) + E)]- 112 (38.58)

we can explicitly evaluate the lr>St tenn and neglect E in the dilference between the first
two terms. Thi.s lelllis to

(38.59)

wh<!re x.,. i.s now tbe zero of tbe potential. In tbe same way equation (38.52) becames:

•+
a= 2
1
0
v'2V(x). (38.60)

Substituting int.o equation (38.54) we obtAin at le!iding order

(3.B.61)

and thus
lmE,:,(g) -
o-o
--1 -
2
(c)'' e-•19.
1r9
2
(38.62)

We bave here calculated only the imaginary part of the would- be ground state euergy.
'Ib obtain the imaginlll)' part of the exdted leveis we have to keep the correction of
arder e-IJ in A{,B) for {3 la.rge. We then expand e:xpl-g- 1 A(/>ll in powers of e-IJ. The
coefficient of e-Nfl in the expansion yields the imagínary part of the Ntb levei at leading
arder.
Two Remo.rks.
(i) We haw assumed tbat. we have only one inst.anton solution carresponding to a given
zero of tbe potential. U we find several instanton solutions corresponding to different
zeroes of tbe potential, we have to look for tbe solution of mínima! action, which gives
the la.rgest contribution in the small ooupling limit.
(ii) At the end of Chapter 37 we have a.rgued that the imaginary part of the energy
leveis which we evaluate, is tbe inven;e lifetime of a state whooe wave function is orig-
inally concentrated near the bottom of tbe unstable minimum of the potential. Thi.s
interpretation is not problematíc for potentials whlch are either unbounded or have a
continuous speetmm in which case the complex energy leve! corresponds to a resonaoce
in the potential. For potentials "'hich have apure discrete spectrum (a.nd ali eigenvalllel
are real} the situation is Jll()re puzz.ling. The intuitive solution to this puzzle is that ;n
the sernidassicallimit, the Jifetime of the metastable state is very long. For times whlch
a.re not too long, the decay process is exponential and ignores effects coming from the
shape of the potential outside of the barrier. Eventually lnverse tunnelling will occur and
the decay law will he modified.
38.3 lnS'tantons in Quantum Mechan_ics: Generoliz:atí01'1 833

Bibliogrnphical Notes
The ca.lcuia.tion of the instanton contributíon for an arbitrary potentia.J can be found in
E. Brézin, J.C. Le Guillou anel J. Zinn-Justin, Phys. lle!J. D15 (1977) l55S.
The shifting method is exposed in
I.M. Gel'fand and A.M. Yaglom, J. Math. Phys. 1 (1960) 48; R.F. Dashen, B. Hass-
lru:her andA. Neveu, Phys. Rev. DlO (1974} 4114.
Tb.. appendix ís inspired by the classlcal paper
J.H. Van Vleck, Proc. No.tl. Amd. Sei. USA 14 (1928) 178.

Exercises
Exerc-ise 38.1
Use the shifting method, ratber than the method of Sectíon 2.3, to calcul&te the path
integral giving the matrix elements of the euciidean evolution operat.or U(x', x; {3) for the
simple harmonic oocillator

lP)=•'
U(x',x;{3) =-
L . (dr(t)jexp [-S(r(t})],
r{G):o.:z

S(r) ! J." dt (r2 {t) + w2r 2 (tl).

Exereise SB. 2
38.2.1. Ca.Jculate the imsginary part ofthe Nth levei above the ground state at leading
arder.
38.2.2. Calculate the ima,ginary part of the ground state energy for a systetn with O(v)
internai symmetry, with the actíon

S(q} =f dt [!4 2
+V (gq~ /g].

E:rerc-ise 38. S
Establísh, using the results o! Appendíx A38.2, the semiclassical expression or the scat-
tering amplitude for an O( á} invnri&.nt hnmilt.oni&.n with a radial potentia.l V(lxl):

H b 2 + V(rj, r lxl.
834 Jnstantons in Quantum Mechanics: Genemlization AJB.l

APPENDIX 38
THE WKB METHOO: MISCELLANEOUS REMARKS

A38.1 The Classical Equatiom of Motion

We consider general hamíltonian systems for whích the classical equations of motion
can be so\ved with arbitrary boundary conditions. Tbis means in particular that we
can exp\iclt\y calcu\ate the classícal a.ction on the trajectory "" o. function of initial and
final positions x' and x, and times T" and T. The classical action A corresponding to a
hami\tonían H(p,q,t) is:

A(p,q) f dt !p(t) · q(t)- H(p(t) ,q{t);t)j, (A38.!)

lr·
wíth q; (T') =X: and q;{T) = :>:;.
The action A, correspondíng to a trajectory {Pe(tJ,q.(t)} in phase space solution of
the dassical of motions is a function of the initial and final positions, and time:

Ae(x',x;T) =A(p.,q.) (A38.2)

ln tbis appendix, in otdet to obtain the usual expressions of claosical mechanics, V>'e
shall work in real time. The analytic continuation of ali expressions to imaginary tnne is
straigbtforward.
As is well known and will he verified below, the c\assical action A. satisfi.,. the
Hamilton-Ja.cobi equations.
Pnliminary Remarks. Let u.s first recall a rew ciaBSicai resuits. Let us call S (Q, q; t)
a function satisfyíng the Hamilton-J acohi equations:

as -H (::.q;t), (A38.3)
i:Jt
witb S (Q, Q; T") O and with additional implicít boundary conditions at t T' whích
will be explained below. We then use S to ~enera.te a time--dependent canonical tram;.
formation in phase spa.ce, traiJSforming the oet (p,q) into (P,Q):

P; = -éJS /ôQ; . (A38.4)

Tbe implidt boundary conditions come from the cooditions:

t = T' - P; = p; and Q; = q; . (A38.5)

We make the transformation (A38.4) in the action (A38.1)

A (p, q) =
((as.
Il" as)
8q, q; + i:Jt dt. (A38.6)

The quantities q and Q are now considered as time--dependent. Expression (A3B.6) can
be rewritten as

A(p,q)
l r -s
T• dt
d
(q(tJ,Q(t);t}dt-
hr as.
"'Q· Q,dt.
T• v ,
(A38.7)
A38.1 Jn.stantons in Qt«~ntum Mochanícs: Genémhzation 835

Using again rela.tíons (A38.4), 'W<' fmally obtain

A(p,q) S(x,Q(T);T)+ {T P;Q;dt. (A38.8)

lr·
The equations of motion are now trivial,

Q; =O= Q;(t) Q; (T'). (A38.9) •

The conditíons ( A38.5) determine the so!ution

(A38.10)

We have therefore sbown that

A.{x',x;T);;; S(x,x';T), (A38.1I)

and found a canonical trans!ormatíon whích maps the inítial bamiltonian syst.em onto a
trivial one with a vanil<bing hlllDiltonian. By performing an additional inven;e ttansfor-
mation based on a standa.rd hnmiltorúan like 8 free ha.miltonian of the form

(A38.12)

or 8 harmonic osdllator
(A:?-8.13)

we can map the original hamiltorúan syst.em onto any convenieot hamiltonian.
Let us also verify that trllllsformatíon (A38.4) leaves the measure in pbase space in·
variant. We perform the transformatíon in two steps. Fínlt we go from p; to Q;:

I1 dp,dq; !f a>s
dq;dQ, det lJq;l!Q;' (A38.14)

We now eliminate q; in favour of P,:

l!q, = [ap]<-•) = (A38.15)

l:JP; 8q i;

Therefore the second jacobian cancels the first one (note that a similar argument di-
rectiy shows the ínVffi'iance already proven above of the symplectíc form dp; fldq;). This
analysis suggests that we can perform two traruformations (A38.4} on t_he path integral
representation of the evolution operator, to reduce it to a path integral conespanding to
a standard system for whích the evolution operator il< exactly known:

(x' IU(T',T)jx) j llfdq;(t)dp

. 0(t)]exp{iA(p, q)J, (A38.16)

witb q, (T'} = :rl, g; (T) "i·

836 ln•tantons in Quantum M echanics: Genemluat:ion A38.2

In this way, ít would seem that we a.re able to colcula.te exa<tly the evolution operator
of a.ny system for w hic:h we know bow to solve the dassica.l equations of motion with
arbítrary bound3.1)' condítions. This would in particular apply to systems with one
degree of freedom, with a hàmlltonian H of the furm

(A38.17)

Unfortunately it is easy to verify that the result is wrong at least in QuantumMechanics.

Actually the whole procedure is somewhat ill-deJined. Thi5 comes from the íact that
changes <>f variables on the path integrol on phase space are even more IJJJJbiguous thao
transformations on ordinary path íntegrals in configuration space. We have given some
indicatíons about thi5 problem in Section 3.L Let us just mention that if we dilscretize
time, we díscover that the transformation ís not reolly canonical because to a variable
q (t 1 ) corresponds 8 momentum variable p (tk + ll.t) of a slightly displaced time. Thi5
effect, invisible in the naive continuum limit, completely changes the result. It is thus
nece;sary to work on the discretized form of the phase space path integrol. It should be
possíble in thi5 way to derive the SEmicla..sícol result which we shali now establish, but, to
our knowiedge, no one bas yet dane so. On the other hand, ao often, the situatlon seems
to be more favourable in field theory. The prob\em we h ave described above comes mainly
from commutation of quantum operators. We have emphasized in Chapter 5 that the
comruutators are inliníte in lield theory and disappear in the renormolízation. Therefore
one expects, as this has been verilied on examples, that the semíclassical approximations
of classically int.egrab\e fu.ld theorie!5 reproduce features of the exact solution.

A38.2 The WKB Method

We now explícitly write the Schrõdinger equation for the evolution operator

H " a ) U(x,x;t)"''"fii"•";t,
( i&'";t , ·tau < , l (A38.l8)

with the deJinitions

U(x',x;t) (x'IU(t)lx), (A38.19)
and the boundary condition
U(t T') 1. (A:\8.20)
In order to write equation (A38.l8), it has b""" necessary to associate wíth the classical
function H(p,q,;t) 8 qua:ntum operator.
We sha.ll see below how much the semiclassical result depends upon the quantization.
We have reintroduced tbe qua.ntity li which we usually set equal to one, to make the
expansion pararneter explicit. The on.satz "1ll be to set

U (x',x;t) G(x',x;t) e'A(x',,.;l)/~{1 +O (li)] (A38.21)

Introducing the ansat• into equation (A3B.18) and keeping the two first ternu; in li we
obtain two equations. The first equ ation is

H ( 8A
i:Jx' '
x:t) = _ élA.
8t
(A38.22)
A38.2 /nstanttmS" in Quantum M e.cha.niCB: Genemlization 837

Tbis equa.tion ínvulves only the classical bamíltonian a.nd is the Hamilton-Jocobi equation
for the classical a.ction on the cla.ssical trajectory. Togethet with the boundary conditions
implied by the condition (A38.20), it determines A completely. The decivation of the
seçond equation involves some more work~ Let us Jirst note that

li 8 ) . BH âG 2 (A3B.23)
H ( iox'x;t G=GH-tn&p;Ô:L;+O(Ii).

Here agrun only the da.ssical hamiltonian is needed. The term contaíníng ôH /BP; is.
already multiplied by a facto r li, thus we can replace the operator p; by BA /&x;, For
the first term we now use the identity:

H (âA
ôx. '
x·t) _~ &pj8qj
2
éPH (ôA x·t)
ôx' '
ili éPH (&A ) oA
2
2
(A38.24)
-2 &pjâpk &x ,x;t ô:r.iB"'• +O (A).
The socond term in the r.h.s. comes from commuting ali the ®rivatives completely on
the ríght. lt relies on the assumption that the ciMSical hamiltonian H is real and can
be quantiwd to generate a hermítian operator. lndeed let us first Bssume that we have
symmetrized ali monomials:

(A38.25)

Then a contribution to this term arlses ea<:h time an operator p of pnq"' acts on q"' and
!
the factor comes from the ,;ymmetrizatíon. lfwe choose another hermítían quantization
procedure, we can start comruuting ali opere.tors p and q untíl the harniltonian is again
a sum af terms (A38.25). Eacl> comruutation introduces a factor ilí. Sínce the dífference
between the two expressions is hermithm, it can only invo\ve (i li) •, whicb can be neglected
at this arder.
The thírd term in expre5Sion (A38.24) nrises from two derívatives acting on the action.
The factor 1f2 is a counting fact<:>r. We then obtain an equation for G:

(A38.26)

Let us now differentiate equation (A38.22) wíth respect to xj and :z:; successively:

éi2 A âH
(A38.27)
ax;axk &p. - &t8x~·
éPA i:JH lPA 8 2 A éPH ô 2 A éPH {i3A
- - - - (A38.2B)
a:z:;Eix;â:r.t âpk + a:r:ax. axja"'l âpk&pl + ôxiâxt âpt&q; &tâxjâx/
L<lt us intmduce a matrix notation
H. _ éPH - éPH
., - &p,&p;' H;; = &p;aqj • (A38.29]

The equation (A38.28) <:an then be rewritten:

~ôH +MHM+MH=-aM. (A38.30)

âxkâp. &t
838 Jnstantan.<l in Quantum Mechanics: Generolization A38.3

Ali multiplications are meant in a matrix sense. We now multíply equation (A38.30) by
M -I on tbe left o.nà ta.lre tbe trace:
tJH _ 8M ( -) _ 1 aM
-trM 1 - + t r HM+H =~trM -. (A38.31)
apk axk at
Let us note that:
a a
az lndetM(z) âz trlnM(z) (A38.32)

The equation (A38.31) can then be rewritten as

BH 8 â2 H ffiH â2 A 8
- - JndetM+ ,_ +~a - " =~-;;:IndetM, (A38.33)
0 p; 8.:Z:; vyj 8 qi vyj P> 8.:z:;v.:Z:k u<

while equation (A38.26) can be written as

8H lJ l &H I &H &3A 8

--lnG+---+---~~--=~-InG. (A38.34)
{jpi 8x.; 2 {jpj8'Ji 2 Õpi8P1 8.:z:jÜX.k lJt

A comparison between these two equations sbows that a solutíon to equation (A38.34)
is
lnG ilndetM+ const .. (A38.35)
'Thking into a<:count the boundary conditions we futally obtain Van Vlecl:'s formula:

~- -- - (~dat &3A )'/2 eí<fh}A.

1
(x'jU(T' T)lx} (A38.36)
' (2 .. i/i)n/2 fJx{Jx!

Jt is straightforward to derive from this equation the corre<lponding expression for imag-
inazy time.

ASB.S The A verage Action in Path Integrals

Since we have emphasized finite action oolutions, one could have the impression that the
re:mlt of tbe path integral is domínated by finite act.ion paths. As the two examples below
show, tbis is n~er the case. lnstead the pa.tb integral is dominated by paths wbicb are
in the neighbourhood of finite action solutions.
(i) Tb a functionol integml Wfltribute only infinite action conjigurutions. Let us give a
very simple exo.mple to illustrate this point:

] . ., J~o/N• IN = (2·nrNt• f d.:z: 1·.. dx.N exp (-~ t x;)

t~l
(A38.37)

Obviously I = l. Setting

(A38.38)

we obtain
I 11 = (21r)-N/2 _2.._N/2_ !."" R"-'dR e-R,f2. (A38.39)
r(N/2) 0
A38.3 ]n!;tantons in Quantum Mechanics: Genernlization 839

Finíte action means that R is bounded:

(A38.40)

Therefore:
(A38Al)

More precisely the integral (A38.39) c110 be calculated by steepest descent. The saddl~
poínt R, is
R. Nl/2, (A38.42)
and the classical octions A which are releva.nt to the path integral satisfy:

(A38.43)

(ii) Ca101J.latú>n of the averoge ac!ion in Quantum Mechnnics.

tre-fiH = 1 q(-/J/2}~q(fJ/2)
[dq(t)]exp[-A(q(t))J, (A38.44)

.A (q(tJl = jtJt• (!li• (t) + 9 - v

-ll/2
dt
2
1
(q(t)v'Yl). (A38.45)

To calculate the average action we use the expresslon

(.A)= -:>.1 >=1 tnj[dq(t)]exp[-U(q(t))]. (A38.46)

We now rescale q(t) in trus expression

>.lf2q(t) ..... q(t).

We get a factor coming from the jacobian. II we as:;uroe thst we have discretized the
time in the path iutegral with a spac~ tifi we find

{.A)= -!_\â>. À=l

In { À-j/lfM trexp (-{JH (gj>.)l}. (A38.47]

In the la.rge fi línüt this yields

(.A} (A38.48)

The first term !filtifi 1s just the term N/2 we fouod in the previous example. The second
term iB still pwportíonal to the "volume" fi but no longer depends on the díreretization.
J UNSTABLE VACUA IN FIELD THEORY

We begill with this cbapter a semicli!Ssical study of barrier penetration effects in quantum
field theory. We first discu55 a !!Calar field theory with a ,P4 interaction, generalization of
tbe quartic a.nharmonic osctllator considere<! in Cbapter 37, in the dimensions in whicb
it is s\Iper-remmnalizable, i.e. two and three dimensjons. New problems a.rise in four
dimensions, the dimension in whlcb the theory is only renormalizable, which will be
""amined !ater.
We have shown that in qua.ntum mecha.nics barríer penetration is associated with
classical motion in imaginary time; we here consider therefore qua.ntum field theory in
its euelidean fonnulation.
We have ~>rgued that the lifetime of metastable statas is relate<! to the ima.ginary
part of the "ground state" energy. Hawerer, for !ater purpose, it is useful to caleulate
the imaginary part not only of the vacuum IUDplitude but also of cortelatíon functions.
We therefore start from the functional integral representation of euclldean conelatian
functions. As in quantum mecluu:úcs, barrier penatration in the semiclaosieal límit is
related the ~nce of instanton solutioDS wbicb we try to characterize. We then e><pla.in
how to evaluate the inst.anton contributions at leading arder, the maio new problem
aríslng from UV divergences.
In Section 39.2 we wnsider more general scalar field theories, of a fonn a.nslogous to
the qua.ntum mecho.nical models discussed in Chapter 38.
In tbe case of the vacuum BIDplltude we find that the iDBtanton wntribution iB pro-
portional to the sp6Ce-tíme volume. Díviding by the volume we therefore abtain the
probability per unit time and unit volume of the meti!Stable pseudo-vacuum to decay.
The lset section is devoted to a brief discUS5ion of a speculati~ cosmalogical application
af this result.

39.1 The q,4 Field Theory for Negative Coupling

We consider tbe d-dimensional field theory for a scalar field <J>, oorresponding to the action
S(l/>)
(39.1)

m bei.ng the mass and 9 the dímensíonless coupllng const.ant (the power of m whicb
appears in front of tbe interaction term 4>4 t.a.kes care of the dimeusion ).
The complete n-point correlation function has the functionsl representation:

{39.2)

We normalize the functionsl integral wlth respect to the vacuum amplitude (partítion
function) at g = O, to avoid introducing a non-trivial g dependence through the normsl-
ization. Following the method used in Chapter 37, we &SSume that we start from positive
values of g and proceed by analytic continuation to define tbe functional integral for g
negative. The imaginary part of correlation functions iB givm~ by the difference between
39.1 Unstable Vacua in Field Theory 841

the continuations above and below the negativo: g-axis. For g small, only non-trivial
saddle points oontribute to thE imagínary prut. Therefore we Jook for non-trivial finite
action solutions of the eudidean field equatioDS, Le. instanton configurations, a.rJd then
calculate the co~onding coutxibutions.

89.1.1 ln.otanto113: C/assícal. solutwru Dnd dassical action

The instanton solutions. The field equa.tion cotresponding to the action (39.1} is'

(39.3j

Let us set (g is negativ.. ):

(39.4)

In terms of f the classi<:a.l a.ction (39.1) reads

(39.5)

The function f (x) then satisfies a parameter- &ee equation,

(-Ll. + l)f(x)- P(z) =O. (39.6)

Jt. ca.n be shown (for deta.ils see Appendix A39) tba.t the solution with the smallest e.ction
is spberi<:a.lly syrnmetríc. We therefore choose an arbitrary origin z 0 and set:

r= iz- xol- (39.7)

A function f(x) wlüch depends only on the radial varía.ble r satisfies the differential
equation
2
d) d--1 -
d
+ 1]
( - ( -dr - - f(r) (39.8)
r dr

Interpretlng r ss a time, we note thnt this equation descríbes the motion of a particle in
a potential -V(!)
(39.9)
submitted in a.ddition to a vlscous damping force.
Since we look far finite action solutions we have to impose the boundary condition

/(r}-+0 for r-+00. (39.10)

Equation (39.8) shows that if !{r) goes to zero at lnlinity it goes exvonentially. The
equation h"" wlutíons even In r, whicb "'" thus detennlned by the value of f at the
origin. For a generic value of f(O), the corresponding solution tend. at infinity towards
a minímum o[ the patential f ±1. The condition (39.10) is satlsli.ed only for a discrete
set of inltial values of /(0). It can moreover be shown that the mínima! a.ction solution
corresponds to the function for which lf(O)I is minimal in tbe set, and wlnch vanishes
only a.t infiníty.
 Unstable Vacua in Field Theory 39.1

The classical aclion. Let u> denote by fc a solution. Since g is dimensionless the
oorresponding das;ical acti<>n ha.s the form

S{4>c) s S(/c) = -Ajg, {39.11)

in which A i5 apure number, whose si,gn can be determined by the following a:rgument:
If Slfc(x)] is finite sois S[fc(Àx)J. After the change of variables À:r- X, Sífc(.\:r)}
t.akes the forro

This expressioo is statíonary at À = 1 because fc(:t:) is a solution of the lield equation.

This yields the identity:

(39.12)

We can use thís relation to eliminate the potential term. We then get an expression for
A
(39.13)

which 5hows that A is always positive.

Remark. lt is dear that the relation (39.12) has nn analogue for arbitrary pQtentials. In
the case of interoction terms which are proportional to a simple power af </>, an additianal
relatian can be derived.
lf S (10 ) ís linite, sois S (J•fc)· Moreover S (>-.fc) is stationary at ~ = 1 béCause fc is
n so)ution of the lield equation. Let us write the oorresponding relation,

{39.14)

We then lind ·the relations

We immediately note that these relations can anly be true for d < 4 nnd thus the
dimension four is singular (see Appendix).
Let us here give the numerkal results for d = 2 and d 3:

d= 2: fc(O) 2.20620086465 A ~ 35.102689573 (39.15)

d=3: !c( O) 4.33738767997 A 113.383507815 (39.16)
39.1 Unstable Vacua in Field Thenry 843

39. L2 The gaussian in!egrution

We want to perform the gaussían íntegration in the neighbourhood of the saddle poirrt.
This inmlves studying the eigenvalues of the operator M:

é'S (:>.9.17)
(r!Mix') M (x,x')
61J>.(x)6,P. (x')'
[(-A,+ mz) + ~gm4-dq,; {:z:)) ó (:z:- .:r')' (39.18)

or, introd\Jcing the solution f(.:r),

(39.19)

Differentia.ting the equation of motion (39.3) with respect to :tp, we lea.m that the func·
tions li,.q, • are d eigenvecton; of M with vanlshing eigenvalue:

(39.20)

This property had to be expected. Due ~ translation symmetry we have found a fa.mily
of degenerat<> saddle point.s ~.(x) depending on d parameten; .:ro., (equation (39.7)). A.;
in quantum mechllnia;, we have to sum over ali •addle points. We thus cb ange variables,
taking the collective ooordinates xo,. as d of our new variables. For instance""' can set:

ti>(x) = 1/>c (x- xo) +L 'Pnfn (x- xo), (39.21)

n~O

wíth the constraints:

(39.22)

in whicb {xo,., '(),.}are the newvariahles. We then integrate over x 0 ,. exa.ctly rmd evaluate
the integral over the other variabl"" '{),. by tbe steepest descent method.
The gaussian integration involves the determinant of M in the aubspace orthogonal
to the zero eigenvalue sector. Tbe change of Vlll'iables generates a jacobian J, whicb, at
leading order, is:
d
J = li 11&,.4>. (x)JI.
,_, (39.23)

Since the solution is spherically symmetric, we can rewrite J as

J (39.24)

Using the relatíon (39.13) we can express the ja<:obian in terrru; of the cla.ssical a.ction

(39.25)

Note one important feature of this expression: each component of zo has generated a
fllctor ( -g) -•;•. Finally in expression (39.2) we can replace at leading order the lield
.P(x) by 1>o(x) in the product IJ:=,
oi> (x;). Since we normalize the functional integral
844 Unstab!e Vacua in Fidd Thoory 39.}

(39.2) by comperíng it wltb tbe functional integral of tbe free theory, vre still haw to
divid<e by ll factor .,;2; for eacb missing gaussian int.egration. Collecting ai! factono we
find

(39.26)

with
Fn (:rh ... ,J:n) = m•+n(d->)/
2
6"/2 f ddxo TI f (m (:r; ~xo)),
i:e:.l
(39.27)

a.nd
(:riMo! :r')= (~A. +m2 ) 6(:r ~x 1 ), (39.28a)
112
n (dat'MM0 1 ) - 1 = lim•-à det((M+•)Mõ 1]1 _ . (39.281>)
m=l •-O m-1

While for t.he vocuum amplitude the integration owr :rc generates o. fa.ct.or proportional
to the volume, for non-trivial oorrelation functions the integration restares translation
invarlance.
Wave function arguments of the kind used for the Schrõdillger equation sbow tbat 8pr/Jc
is not t.he ground state of M. There is one state with o. negative eigeuvalue so t.hat the
final result is real r..s expected. In Appendix A39 we give a proof of this property usillg
Sobolev inequalities.
Discusuíon. A few comments concerníng exprcssion (39.26) are here in order. We
have obta.ined a result for tbe complete correlation functious, bnproperly normalized, for
convenience, with respect to tbe free lield tbeory. We notice, b~, that, because
cf;,(:r) is proportional to 1/A, the ímagioary pa.rt o{ the n·point function increases
wlth n for g small. This shows that at leading order the correlatíon functions nonnalbed
wlth respect to the pa.rtitiou fuuction correopondillg to the complete a.ction (39.1) ba.ve
the same beha.viour "" thase renormaliz.ed wít.h respect to the free field theory.
Moreover, for the same reB.SOn, when we rousider a complete n-point function, the
imaginary pa.rt coming Erom disconnected parts is subleading by at least a pOWilr of g. If
we call w<nl (:r 1, ••• ,:r.,) t.he connected n·point function, we thus find at leading order:

lmW<"l ~ImZ<"l,

a result that is consistent with the observation that the explicit expression (39.26) ia
indeed connected. Tb pass from connected correlatioo functioos to IPI functions, we have
first to subtra.ct the reducible contributious which involve fuoctions with a sma.ller number
of arguments and which are therefore negligible at Jeading arder, and then to amputate
the rema.ining pa.rt. Again for the •ame reason only the real pa.rt of the propagator
matters; to amputal.e expression (39.26) we therefure have simply to multiply it by the
product of the inverse free propagators cottespondillg to each externa! line. Introducing
j (p Jm ), t.he Fourier transform of f (rrcr:), and writillg the n-point 1Pl function f'(n) in
momentum space representatioo we obta.in:

lmr
-(n)
(ph•••,Pn)--2i
1(A211" )di• '\-g)(d+n)/2m
eA/ 9 d-n(d/2+1) rr"
j;Jv6f
r-- (Pi) • •
m (p,+m)
(39.29}
39_1 Unst<tble Vacua in Fiçld Tlu;orv S45

The structure, at \eacling order, of the imaginary part of the n-point functioo is par-·
ticular\y simp\e in momeotum sp~, in particular it depends only on the square of the
momenta a.nd not of their scalar products_
Up to this point the discussions of the f/>4 fie\d theory a.nd of the a.nharrnooic oscilla.tor
have been rema.rkably similar_ Now comes one important dilference: as we shall sce the
determinant of the operator M is actual\y UV divergent and we shall ba.ve to deal with
this new problem_

39.1.3 Renonnalitation
To define properly tbe r/! 4 in 2 and 3 dimensions, we have first to introduce a UV cut-off
a.nd then to a.dd te the classical action a m&o coUul.ertettu befote t..tdhg tbe iofltiite
cut-off \irnit: the cut-<>ff dependent action SA b85 the form:

SA (t/>) J dd:t Htf>(x) (-a+ a 2 /A2 + m2) ,P(:t) + f,g,P 4(:t) +!6m2 (A) tf>•{:r)j
(39.30)
Let us examine, at cut-off A Jarge but fixed, the effect of tbese modlficatlons. The
a.dditional term

modifies the equa.tion of motion but wben A bi!COmes large this modification becomes
smalllike 1/A 2 •
Tbe countenerm on the other band increases wlth the cut-olf but ís proportlonal te
at least one power of g. Hence, because we take the small g limit before taking tbe large
cut-ofr limit, tbe oounterterm does not oontribute to tbe classieal equatíon of motion.
On tbe otber hand, if we calcu late tbe cootribution of the counterterm te the classícal
action we find that the one-loop counterterm, wbicb is proportional to g, gives a contri-
bution of ordet 1 in 9 because 1/lc(:~:) is proportional to 1/A. lt therefore groerates
ll.tl a.ddltional multiplica.tiw factor. Let us now consider the determina.nt of M in the
regula.rized theory:

(39.31)

We ean expll.tld it in powers of q,;(:~:) using the identity tr In= In det:

JndetM ""Jndet (-a+ a 2/A 2 + m2 )

f:~ tr [!gm•-d.p~(:t) (a- 62/A2 m2)-lr. (39.32)
,_,
The first term is cancelled by the free determina.nt det Mn. Ali t.erms for I; ~ 2, are UV
finlte in 2 and 3 dítnensiow;. Let us rewrlte more explicitly the k = 1 term:

(39.33)

in whlcb D is the free regularize<! propagator at coinciding arguments,

D (39.34)
846 Un.stable Vacua in Field Theorv 39.2

The determínant of the operator M therefore contains a lactor whích div<'rges at large
cut-aft:
(39.35)

Thio fa.ctor exa.ctly cancels the infinite factor coming from the counterterm, so that the
final expression for the imagínary part il; finite. The fa.ct that realty we have to calcu\a.te
det' M does not chllllJl:e thil; a.rgument, because UV divergences are insensitive to the
omíssion of a finite number of cigenvalues of M, as the seoond form (39.28b) explicitly
shows,
The decay af the false vaettum. The special case n = O oorresponds to the imaginwy
part of the vacuum amplitude. We bavé expa.nded perturhation theory around the min-
ímum rf> O of the potential. The perturbative gmund state corresponds to a wave
function concentrated around small fields. However, because we have expanded a.round
a relative mínimum of tbe pot.ential, this state a.ctually is unatab\e. We have calculated
its decay rate dueto barríer penetration. VW. note that the integral over :~: 0 ~ in equation
(39.27) yields a space-time volume fuctor. Th obtain a finíte decay amplitude we have
to divide the result by thil; volume factor. We thiiE obtain the probability per unit time
and unit volume for the would-be ground stat.e ("false vaeuum") of the theory to decay.
Some implications of such a result will be discussed in a slightly more general context in
Section 39.:>.

39.2 General Potentials: lmtanton Contributions

We now extend the analysis o{ Chapter 38 to more general scalar field theories, IIEÍng the
techniques developed in Section 39.}. We con.sider an euclidean oction of the form:

(39.36)

in which the potential V(,P) has a stable and an unstable minlmum, a.nd il; of the type
discussed in Cbaprer 38. Assuming that at some initial time the quantum mechanical
state corresponds to a field concentra.ted a.round the un.stab\e minimum of the potenti81,
the "false" vacuum, we wllllt to evaluate semiclassícally the probability for the false
vacuum to decay into the true vacuum of the theory. The calculation, at leadlng order,
agaln involve!i the deterrnination of an instanton solution and a gaussian integration
around the instanton.

$9.!J.l Calculatíon of the instanton contribution

VW. define tbe lield In such s way that tbe unstable minimum corresponds to rf> O.
The discussion of the existence of an instanton solution is similar to the one gíven in
Section 39.1. A theorem establishes, under mlld assumptions, tba.t spherícally symmetric
solutions glve the mínima! actíon. We therefore look for such a sotution and set:

r 1:~:- x 0 1, /(r)= vg,P,(z).

The cla.<sical equation of motion reduces to:

2
d f + d- 1 df , V' (f). (39.37)
dr2 r dr
39.2 Unstaille Vacua in Field Theo111 847

This again i• the equatie>n govetning the motíon of a particle in a potential V (f) and
submitted to a viscous damping force. The solution depends on its value Bt the origin
/(0).
Let us coll f+ the abso!ute mirúmum of the potentlol. If we cboose !(O) too dose to
f+, /(-r)- j{O} will remain very smoll until r bocome>; vezy large. When r is Jarge, the
dampiug force is smoll so tbat energy js almost conserved and the partiele will overshoot.
If /{0) is toa eloS<" to zero, the partide willlooe too much energy aud therefore unde.rsbot,
the asymptotlc value /{r) then cnrresponding to the mll.lCÍmum of V (f). Thus somewhere
in between we expect to find a value /{0), which corresponds to a solution which goes to
zero at infinity and tberefore has a finlte a.ction. The viria! theorem, as used in Section
39,1, implíes that the corrcsponding actlon ís posítivc. Let 4->c(:r) be a solution. We
calculate S i!jlc (>.:r)],

S [1'>.{.\:r)] )..2-d f~ (&p!jlc)2 ddz +)..-d~ f V (l/>c.J9)ddx. (39.38)

Expressing that the action is statlonary for ).. = 1 y:ields the relation

(39.39)

Therefore
S[~j>c(:r}]= A' (39.40)
9
witb again:
A=~ f ({)~j)2do:r >O. (39.41)

The second deriwtive of S I.Pc (Àz)! is

--
d2
2 S[.P.()..x)Jl~ 1 =(2-d)
{d)..) -
f W~<<J>.(z)]
.d
d:r. (39.42)

This equation y:ields another proof at least for d ;::> 3 that the second deriva.tive of the
action at tbe sa.ddle point bas at least one negative eigenvalue. To get a more precise
result, one has til make use of a generol theorem established by Coleman e! al that
states that 1/>c(:r) is an absolute minimum of S(Q!) at fixed integral of the potential
J d~x V ( .;g.p,(x)). lndeed if
IJ,P.(z)ó.p. (:r')
ha.d two negative eigenvalues, it would he possíble to a.dd to .P.(:r) a linear comhination
of the correspOndíng two eigenvectors leavíng a.t first otder the integral of the potentiol
unchanged, a.od decreasíng S(,P), contradictíng tbe theorem. Thus ti 2 Sj6,P.(z) 61/>.(:r')
ha.s at most one negative e~nvalue.
Since equation (39.42) or more generolly the argument about the elgenvalue, shows
that it has at least one, l.2 S/ (1!<1>.) 2 has one o.nd only one nagative eigenva!ue.
We now ho.ve to take :ro as an íntegration variable. This leads to a jacobían J whicb,
as we have sea> in Section 39.1, can be wrítten as

{39.43)
S48 Uoutable Vacua in Field TI..Qry

where equations (39.40,39.41) ha"" been used.

All other details ofthe calculation can be borro""'d from the $ 4 example, a.nd we finall,y
obtaín a.n explidt expression for tbe imnginary part o f the n-polnt correlation function,
Let us just make a few remarks concerning the renormalization. {If we suppose tbe
theory super-renormali~ahle or renormallzahle).
Renarmalization. To construct tbe renormatized tbeory we can proceed in iUl inductive
way: we regularize the theory, we tben add the counterterrns wbich make the theory
finit.e order by order in a loopwise expansion, i.e. bere an expansion in powers of g. The
renormalli.ed oction S, (<i>) bas tbe form:

At leading order only the one-loop counterterrns mntt.er in tbe instanton calculatíon. To
evaluate them we expand the efiective potential f(.P) (the generating functional of IPI
correlation functions) at one-loop order using the regularized action, and calculate its
divergent part. In Chapter 5 we have deríved:

(39.44)

To render correlation functions fioite we have to subtra.ct to the regularized oction th.,
divergent part af the one loop tenn oH' (<i>}:

1 (
- 2 IPS) _
trlnp:<
'I"''ÍJ dn.
.

When evaluated for </> = if!e, this contribution exactly cancels the divergence in the de-
t.ermin8llt coming from the gn.ussian integration around the saddle point. This argument
ca.n be generallzed to arbitre.ry orders.

39.3 The Decay of the Fa.lse Vacuum: CosmologicallnterpJ"<>tation

In preceding """tions we have in particular deterrnined the probability for a "false vac-
uum• of a Field Theory to decay tbrough ba.níer penetration effects. It has been spec-
ulated that such " pbenomenon could be linked to the dynarnics of tbe early universe.
When the universe started to coai down, some symm..tries started to be spontaneously
broken. Some regíon might bave been trapped in the wrong phase. The false vscuum
must eventually decay in the t.rue vacuum, but íf tbe process is s1ow enough, it míght
have occurred at a much !ater time when the universe v.'BS already cool. This ís tbe kind
of physical &pplicn.tion that we here ha.ve in mind.
According to the pre\•íous discussion, if the uníverse ís in the wrong vacuum, there is
some proba.bility at eacb point in space for some bubble of true va.cuum to be created,
and if the bubble is large enough, it becomes favourable for it to expand, absorbing
eventually the wbole spsce. Th diseuss what happens once a bubble bas been created, ít
is useful to oonsider fust the n.nalogous problem in ordíne.ry quantum mecllanics.
Quantum Mechan.ics. In the ln.nguage of pa.rticle physics, a semiclassical descriptiou of
the decay process would be the fullowíng: a particle is sítting in tbe weU of the poteatial
corresponding to the unstable minimum. Ata given time, it makes a quantumjump and
39.3 U...table Vacu<> in Freld Theory 849

reappears outside of the barrier, at tbe point wbere the potential bas the same value as
in the bottom of the well, with :~.ero V<>locity (by energy oonservatioo). Then its !urther
trajeetory can be entirely d.,cribed by Classical Mechanics.
Field Thoory. Let us apply the same ideas to the fielà theoreticaJ model we discuss
here. At time zero tbe system makeo a quantum jump. Accoràing to the previous
discussion, tbe value of the field at time zero is then (with the cholce "'Op O)

r/J(t O,x) '"'r/Jc (:td '"'O,x), (39.45)

and its time derivative van~es.

814>(t O,x) O. (39.46)

At a la.ter titnEo <fJ (t, x) then obeys the minkowskio.n equation of motion

[V~- a:) .Z,(t,x) =~V' ( yg~} (39.47)

The fust equation (39.45) tells us that the same function describ"" the form of the in-
stanton in euclidean space, and its shape in ordinary (d - 1) space when it materializes.
Let us now consider the continuation in real time of the solution of the euclidean equa-
1
tion of motion </Jcl(x2 - t 2 ) f2) (since ,Po(r) is an even function, the sign in front of the
square root is irrelevant). It satisfi"" the condltions (39.45,39.46} and obviously obeys
the equation of motion (39.47). It is therefore the solution of our problem for positive
tiro....
Since the si..e of the bubble is gíven by micropbysics, the interior of the bubble corre-
sponds to smell valu"' of r on a macroscopic scale,

O :<:: x 2 t
2
=r2 « 1•

Therefore after a short time the bubble starts expanding at almost the speed of light.

Bibllographleal Notes
The first calculations of instanton contributions in Field Tbeory are due to
L.N. Lipatov, JETP Lett. 24 (1976) 157, Sou. Phys. JETP 44 (1976) 1055, JETP Lett.
25 (1977) 104, Sou. Phya. JETP 45 {1977) 216.
We here have followed tbe method presented in
E. Brézin, J.C. Le Guillou and J. Zinn-Justin, Phys. Rev. D15 (1977) 1558; J. Zinn-
Justin, Phys. Rep. 70 (1981) 109.
To simplify expressions we have used scaling argumenta
G.H. Denick, J. Malh. Phys. 5 (1964) 1252.
The field equations for the .p• field theory have been 50lved numerically by Zinn-Justin
unpublished, and the determinant calculated in
E. Brézin and G. Parisi, J. Sta!. Phys. 19 (1978) 269.
The relativistic generalization of the metbod of collective coordinal.el is described in
J-.L. Gerva.ís and B. Sakita, Phys. Reu. Dll {1975) 2943; V.E. Korepin, P.P. Kulish
arad L.D. Faddeev, JETP Lett. 21 (1975) 138; N.H. Christ and T.D. Lee, Phys. Rev.
Dl2 (1975) 1606.
The standard early reference on wcuum deca.y is
 Unstable l''acua in Fíeld Theory 39.3

M.B. Voloshín, 1. Yu. Kobzarev and L.B. Okun, Yad. Fiz. 20 (1974) 1229 (Sov. J.
Nuel. Phys. 20 (1975) 644).
The instanton approacll to this problem has been developed in
S. Coleman, Phys. Rev. D15 ( 1977) 2929; C.G. Callan and S. Colernan, Phys. Rev. D16
(1977) 1762; M. Stone, Phys. Rét!. Dl4 (1976) 3568, Phys. Lett. 67B (1977) 186; S.
CoJeman a.11d F. De Luccia, Phys. Rev. D21 ( 1980) 3305; P.J. Steínhardt, Phys. Rev.
D24 (1981) 842.
Propertíes of the mínima! action are proven in
S. Coleman, '\{. Glaser andA. Martin, Cnmmun.. Math. Phys. 58 (1978) 211.
For a discussíon of Sobolev inequalitíes see for example
E.M. Stein, Singular lnt~ anrl Dijjerentiability Propertíes nf Punctiom (Prínceton
Urtiversity Press, Princeton 1970); J.R. Klauder, Acta Phys. Austr. Suppl. XI (1973)
341.
The relevance of Sobolev inequalití"" for the instanton problem in the <P' fietd theory has
been emphasized by
G. Parisi, Phys. Lett. 66B (1977) 167; 68B (1977} 361.

Exercises
E:un:íse 99.1
Fínd an estimate for the instanton action in the.p4 field theory in d = 2, 3,4 dimell5ions by
using a first ga.ussian trial límction, then a function proportional to the one-dimensional
sotution, [(r) o; 1/cosh(r/ro), and compare wíth the exact results given in Section 28.3,
for d 2, 3 and in Section 42.1 for d = 4.

E:terrise 99. :l
Consider the unphysioal field theory in d dimensions, whose actiou is

in whi<:h .ó. is just the Lapl&<e OJJerator in d dimenoions. In this theory the free prop.
a.gator is in momeutum space e-?, thus the theory is UV finite in any dimensíon, and
the Feynroan diograms can be ealculated algebraically. lt is also possible to calculate
analytically the contríbution of the presumabiY,. domínant ínstanton solutíon.
Ansatz: Look for a solution of tbe forme-o% . Soiution ín Bervillier et aL, Phys. Rev.
017 (1978] 2144.
A39.0 Umtable Vacua in Field Thenry 851

APPENDIX 39
SOBOLEV INEQUALITIES

l.Et us consider the following functional .R{<p),

{f d"x [<&,.<p(x))2 + 'l'•(:rJ]

f tp4 (x)d":r
r (A39 .•1)

lf the dirnension d is not larger than four:

d :54. (A39.2)

Sobolev inequalities tell us that

R("':)<':R>O. (A39.3)

ln addition, for d < 4, there exists a spherical!y symmetric, zero-free function rp.(z)
whicb sntura.tes the bound
R(rpc)=R, (A39.4)

and is a solution of the variational equatlon

6Rf6'f'c(I) ""O. (A39.5)

Dimen.sicmal -~m>aller lhan four. LEt us write thís last equation explidtly,

{A39.6)

in which we he.ve defined

(A39.7)

This equation is, up to a rescaling of 'Pc(x), the equatíon of motion (39.6). Both equatioru;
become identical if we cboose the scale of 'Pc(x) which is otherwise arbítrary such that

K 1. (A39.8)

For eech ln.stanton solution we have derived the identity (39.14), whlch for f(x) can be
written as
f 2
ddx [ca~t(x)) + J (xJ]
2
=f
ddx f 4(x). (A39.9}

Therefore we can a.lso write the classical action as

A (A39.10)
852 Unstable Vacua in Fieid Theory A39.0

The smal!est actJon solution thus corresponds to the rninímurn of R ('I'):

A= 3R/2, (A39.11)

and the solutíon /(:t) we are looking for js given by

f{x) 'P,(x) for K 1. (A39.12)

The introduction ofthe functional R( 'I') h"" tht' following advantage: the funttional S (rp)
is obviously not baunded from beiow. But ifwe restrict ourse!ves to fields .P(x) solutions
of the equation of motim> with finite action, then S (.P) can be relared to the functional
R(.p) which is bounded from below for all fields.
Let us then derive the properly that the operator M has one and only one negative
eigenvalue from the form of R( 'I') and the assumption that 'Pc corresponds to an absolute
minimum of R. ]n the operatoc 5ense:

flR >O. (A39.13)

6'1'c{z)6r,o, (z') -

Calculating the second derivative of R explicitly, we ohtain:

(A39.14)

We have again 5et K = 1. Let us write below M (x, x') in terms of f(x) or 'Pc(x) for
m=l,
M(x, :~:') = [-LJ.+ 1- 3r,o~(x)] ó (x- x'). (A39.lf>)
We therefote deri"'l the relation:

(A39.16)

(i) Since R(op) is inva.riant in the change rp,(x) in À'l'c(x), 'l'c is a.n eigenvector of
2
62 R j(5'1'c) with eigenvalue zero, thus

(A39.17)

Tbe operator M has at least one negative eigenvalue.

(ii} Since M is the sum of a positive operator and a projector of rank 1, it can have
at most one negative eigenvalue. lndeed if it had two negative eigenvalues, we could find
a linear combination of the conesponding two eigenvect.ors whicb would decrease M at
2
an avera.ge of the projector fixed. This would imply that é1 R/ (li<Pc) is not a posítive
opetat.or.
We oonclude that M has rme and only one negative eigenvalue.
Dimeruion four. Let us calculate R !op(>.x)] for d :<:: 4. Then, cbanging >.:r in x' in the
va.rious integrais, we obta.i.n:

(A39.18)
A39.0 Unstab!e Vacua in Field ~1"11 853

V./e can now writ.e

R= lllin nún R [<P(À:r:)]. (A39.19)
{ IP(:r)} >.
The núnimum in>. af expressíon (A39.18) iB obtained for

À (A39.20)

alld equati.on (A39.19) becoines

R (A39.21)

FbT d = 4, we see that t~ solution is >. oo alld expressian (A39.21) i.sjust tbe equivalem
of expression (A39.1) for the massless q,• lield tbeory. Since fot- d 4 tbe massless 4>4
is scale invariant tbe contribntion of the mess term can be arbitrarily decreased by 11
rescaling o[ the variable "'·
We can draw two lnteresting conclusions from thls analysis: tbe wiaimal value of R(<P)
is tbe same in four dirnensions for tbe massive and the mBSS!ess theory. Tbe sa.me will
apply toS (1/J).
The minimum of R(<P) will be obtained frODl a solution of the massless lield equa.tion.
The me.ss.ive field equation has no solution.
Tbese temarks will become useful when we di.scuss tbe <1>4 lield thoory in Chapter 42.
DEGENERATE CLASSICAL MINIMA ANO INSTANTDNS

We con.sider in this chapt.er a somewhat dífferent situatíon in which instantonspl&y atole:

dassical actíons with degener&te ísolated mínima. The simplest examples are provided
by quantum mechanics ín the form of hamiltonian systems correspondíng to potentials
with degenerate mínima. Classical.ly the st&te of minimum energy corre.sponds to a
partide sitting at any of the mínima of the potential. The position of the parti ele breaks
(spontaneously) the syrnmetry of the system. In quantum mechanics on the contrary, we
expect the modulus o[ the ground state wave functíon to be large near ali mínima of tbe
potentíal, a.s a consequeoce of b&rr:íer penetration effects. We discuss in thiB chapter this
phenomeoon on two t;ypical examples: the dou ble-well potential and the periodíc cosine
potential.
In fielrl tbeory the problem is more subtle a.s the study of pha.se transitions has shown.
However the presente of ínstantons agaln índícates that the classical mínima are con-
nected and that the symmetry between tbem ís not spontaneously broken. Exa.mples of
ínstantons of this type are provided in two dímensíons by the C P(N- 1) modela and in
four dímensíons by SU(2) gauge theories.
Fínally in Section 40.6 we díscuss the role of ínstantons in the context of st.ochastic
dynamícs.

40.1 The Double-Well Potentia.l

We first discuss the simple example of the double-well potentíal:

H= -4 {d /d:r )2 +!:r' (1 - x.f§)2 • (40.1)
The potential has two degenerate mínima located at the origin and at 1/.f§, In addition
the hamiltonían commutes with the refiectíon operator P which exchanges :r and g- 1 /~­
:r. This last property ís not essential for the existence of ínstanton solutions. It is jll5t a
simplifying feature, which moreover is present in most examples of physical interest.
The strocture of fhe ground state. Due to lhe symmetry of the potential, we can
expand around .each of the mínima of the potential, and we find the same perturbative
expanBion to ali orders. lt would therefore seem that the quantum hamiltonian also
has a douhly degenerat.e ground st.ete, corresponding to two eigenfunctions respectively
concentrated around each of the dassical miDima of the potential. Actually we know
that, due to barrier penetration effects, this ;, not tbe case and the true elgenstlltes nre
also eigenstates of the refiection operator P, the ground state being ao even state. The
refiection symmetry is not spantaneously broken in quantum mechanics in the case of
regular potentials: as we ha'"" extensivcly discussed, correla.tion functioru constmcted
with a hamilton!an of thís type have, from the point of view of pha.çe tranllitions, thc
properties of correlation functions of the JD lsing model (see Section 23.1 ). Neverthel=

a use the quantity tr e-flH to calculate the ground state energy, since we expand in
ali first, we find ít difficult to separate the two first eigenstates in the large p limít:
tre-llH ~ e-fJR.,
for!/ -+O,
+ e-llR-
{J-+ oc·.
~ 2e-~(li++.L)I 2 coshjj(E+ -E-) /2
(40.2)
40.1 Degenerate Giassical Minim~ and Instantons 855

The partítion function is dominated by the perturbative expa.nsion of the half sum
~(E++ E-), Md is only sensitiw to the non-perturbative difference betvroen the eigen-
values E+ a.nd E_ at orde:r (E+- E_)2 :

1
lntre-eH ~(E++ E_)-~ ln2+0 [e- 6 ,,8(E+- E_)']
forg-->0, /J-+oo. (40.3)

To calculate the difference (E+ E_), it is therefore muro easier to evaluate the quantity
tr p.,-~H (see Se<:tion 23.2): In the same limits we now find:

tr Pe-fJH ~.,-~E, -e-~E- ~ -2sinb !IJ (E+- E_) .,-S(E,.+E-)1 2 (40.4)

g-> CJ, IJ-+ 00.

Since E.,. -E_ \'allishes in perturbation theory, the r.h.s. is dominated by

(40.5)

It is actually more convenient to oonsider the ratlo between tlte quantitíes (40.2} and
(40.4):

(40.6)

By evaluatíng this ratio, we can distinguish between a situatlon in whlcll the ground
state is degenerate and t.he synunetry spontaneously broken, and a situation in whicll
quantum fiuctuations restore the symmetry a.nd líft the degeneracy between the two
lowest lying states. Since t.he ratio vanish"" in perturbation theory, we have to look for
non-pe:rturbative effects: they are here dueto instantons.
lnslanton contrib-utions. The pa.th integral representatíon of tr Pe-.BH differs from the
representation of the partition functíon only by boundary conditíons:

trPe-ilH = 1 o! -~/2)+q(/J/2)~9 -•r•

ldq(t)]exvi-S(q)], (40.7)

wíth
.B/2
S(q)
l -fl/2
[ H2 (t) + fq 2 (t)(1- y'yq(t)) 2 ] dt. (40.8)

The path integ-ral represente.tion of ue-~H is domínated by the trivial saddle points:

q(t) =O or q{t) g- 112 ,

and this lea.ds to the usual perturbative expansion. However these paths do not contribute
to the path integral (40.7) because they do not sa.tisfy the boundary condítions. Thls
is not t"oo surprising since we have ahea.dy seen that t.he differenoe E+ - E_ vaníshes
to all orders in an expansion in powers of g. We therefore have to look for non-triYínl
solutíons of the equa.tion o f motion whicll he.ve a finite action in the inlinite {3 limit. The
boundary conditioru; then impose

q('f-00) =o and q(±oo) = g-112.

856 Degcne=u Qassical Minima cmd lnstantons 40.2

The non-degeneracy of the gTOUnd sta.te thus depends on the mústence of on instanton
solution connecting the two mínima of the potentíal.
The euclidean equation of motion is

-q(t) + q(tl(l- q(t)Jg) (1- 2q(t)Jg) =o. (40.9)

In the infinite fJ limit, t.bis equo.tion indeed has two solutions witb fioite de.ssical action
whlcb we cal! generica!ly ÍT15tantons or iTI5tanton and anti-iru;tanton when it is necessary
to distinguish betw.!en tbem:

±(t)- .2_ 1 (40.10)

q. - ..ft {I+ e=H•-•ol)"

Let us calculate the corresponding classical &etíon. lt ís easy to adapt the method of
Section 38.2 to the present problem. From equations analogous to (38.51,38.54), we
obtain the expansions of the classical energy arul ectíon for {j large:

E (,8) = -2e-ll +0 (e- 2 ~'), (40.11)

S(q.) =; [~- 2e-ll +0 (e- 2 ~)]. (40.12}

The determinnnt resulting from the integTation around the saddle point eM also be
evaluated by the method explained in Cho.pter 38. The only notíceable modification
comes from the fact tbat lj.(t) has no zero. lt is thus corresponds to the ground state of
2
tbe harniltonian éPSj (llqc) , whicb is thernfore a positi"" operator. Tbe final result is
real, as expected. We can use eXpresslon (38.50) to obtain it, exrept that no 1/2i fact.or
is needed bere, but instead one has to multiply by a factor 2 since the two solutions qi
a.nd q; gi"" identical contributions:
2
trPe-~H ~ ,...,.f3e-ll1 2 e- 1/ 6 §(1+0(g}), g.-0, {3-oo. (40.13)
v1f9

Using then equation (40.6) we find tbe asymptotíc behaviour of E+- E- for g small,
2
E+-E- = - --e- 116 g(l+O(g}). (40.14)
§-0 ..fi'9
The dífference·is exponentially small in 1/g and this iB ccnsistent with tbe property that
it ~to a!l orders in g.

40.2 'f!;jPeriodic Cosine Potential

We now consider a slightly more complicated problem:

(40.15)

Tbe potentia.l is períodic ond has therefore an irúinite number of degenerate minima. We
can eXpand perturbation series around each of these mini ma !llld therefore, to ali orden;
in perturbatíon thoory, the quantum macbanical hamiltoni8JI also has 8JI infinite number
of degenerate ground states. Actua!ly we know that the spectrum of the ha.míltonian H
is continuou.s a.nd has, at le&St for g =all enough, a ba.nd structure: this property, for g
small, is a.gain due to barrier penetration effects.
40.2 Degenerote Cl4Ssical Minim_e> and Instantons 857

40.2.1 The •truclun of the ground staíe

To malte our rema.rks more precise let uo introduce the translatíon operator T, corre-
sponding to an elementary translation of one period 2" /.,jfi
IT,H]=O. (40.16)
Tbe operators T and H oan be diagonalized simultaneously. Siuce we want the eigen-
functio= of H to be bonnded at infinity, the eigenvalm"' of T must be pure phases. Each
eígenfunction of H is tbus ohftl'aoterized by an angle 'f' (pseudo-momentum) eigenvalue
ofT:
{40.17)
ln the limit g - O, H hns the spectrum o! the harmonic oecillator, each state being
inlinitely degenerate. In tbe basis in whíoh Tis diagonal, a!l tbeeígenstates corresponding
to the same eigenvalue are indexed by the angle '(1. Wben g ,P O, each point of tbe
spectrum becomes a band. ln a band tbe eoergy elgenvalue is a peTíodic funct.ion o! '(I
whioh can be e:xpanded in a Fourio•r serles:
-!-CO

E('f'l=Z::::,E!e01"', E1 E-1· (40.18)

_.,.,
Ali coeffidents E 1 except Eo vanisb to ali orders in a perturbative expansion in g.
Let us now collBider the partition functioo whioh is here tr' T e-I!H. The notatíon tr'
has tbe fo!low:ing memúng: since the diagonal matrix elements of e- /iH in conl\guration
space are pe:riodíc functions, we integrate only over one perlod.
The large f3 l!mít selects the lowest band and we obtain (see Appendix A40.1 ):

tr' e-/iH ~ _1 12.- d<,<> e-llE('P). (40.19}

IJ~oc21< o
Like in the case of the double-well potential, we observe that it ís difficult to determine
the dependente on 'f' of the ene.-gy l!M!ls from the partition function. Let us instead
consider:
tr 1 Te-il 8 ~ -
1 12.- d<,<> . e'"'e-~E(\Pl. (40.20)
~~~~2lr o
For g small E(g) Eo(g) vanísbes faste.- than any power of g. We can therefore expand
eque.tion (40.20):

tr' T e-llH ~ e-PI>o J d'(l e'"' [1- (J (E

21<
Eo) + forg-0, /3-oo. (40.21)

The integratíon over 'P seleds E 1 :

tr'Te-IJH ~-(Je-P&.El{g), g--+0, p-oo. {40.22)
This equation can be more ronveniently rewritten:
tr'Te-PH /tr'e-I'H ~-fiE, (g). (40.23)
As indica.ted a.bove, if E, does not \'anisb this ímplíea that the traru;lation symmetry ís
not spontaneously broken.
Remark. To e\'aluate tbe other Fourier seTies coefficients E,, E., ., for g sm.all, the
most convenient method i& to consider tr' T" .,-IJH for n = 2, 3, .... This evaluation leads
to a.dditiona.l problems wbioh w.. sha!l solve in Chapter 43.
858 Deyenemte Classieal Minima and lnstantons 40.2

40.2.2 The instanton contrí~utians

The path integral representatíons of the partítion function tr' e-/JH and of tr' Te-tm
again differ only by the boundary conditions. The operator T has the effect o[ translating
the argument q in the matrix element < q'[ tr' .-~H [g > before taking the trace:

tr'Te-~H =1 q(~f2)=q( -IJ/Z)·U,/.;i

{dq(l)]exp[-S(q)], (40.24)

13/Z
S (q) =
!-/3/2
[!4 2 (t) + g- 1 (1- cosJ9"q(t))) dt. (40.25)

We recall that q ( -/3 j2) varies <M!r only one period of the potential. For {3 large and 9
small, dueto the boundary conditions, the path integral is dominated by instanton con-
figuratíons which connect two co!l5ecutive mínima of the potential. Solving the equation
of motion explicítly we find:

(40.26)

and th" corresponding clas&cal action, in the infinite /3limit, is:

(40.27)

For ali potentials for which the mínima can be exchanged by a refiection, the a.nalogue
of expressíon (38.59) is

E(/3)~-e-ll...!!.exp
:r2
2
[2 1"•1' (
o
~
1
y2V(:r)
(40.28)

in which :ro is the lo;;;;qm

of the other minímum. Applying equation (40.28) to the
analogue of equation~O) we obtain:

(40.29)

or:
~ - --e-
4 81"
E1(9) (40.30)
•-o ..[ff9
\
Although we shall not evaluate En for n ;>:: 2 here, it is easy to see that the corresponding
boundary conditions for trT"e-13H which are:

211"
q ((3 f2) q(-(3J2)+nJ9,

select an instanton solution which for (3 large has an action 8n f g. Therefore E 1 gives
the dominant non-perturhative contríbution for !I small, and

E., (g} = Eo (g) _8_ .-6/R [1 +0 (g)J CO.S\0 + Q (·-lfi[g) • (40.31)

.ftY
40.3 Deyenerate Cla.osical Mínima and lnstantons 859

Canclusions. We have shown on two examples that, a.s antidpa.ted, in a theory in

which, at the cla.ssical levei, a discret~ symmetry ís spontaneously bYoken because the
dassical potentia.l ha.s degenera.te minima, the existence of insta.nton solutions ímplies
that quantum Huctuatloru; restore the svmmetrv. An a.nalogous a.nalysis has been made
on the lattice in Sections 23.1,2l.2 wh.,;.e we h~ve also showu that, in contrast, sponta·
neous symmetry brealúng of discrete symmetties is possible ín h\gher dimensions.
Note that, in contrast to discrete svmmetries, where quantum Huctuations lead to
exponentially small elfects in 1/h or the equivalent coupling consta.nt, in the case of
continuous symmetries the elfects of quantum lluctuations show up already at first ordé.r
in perturbation theory as a conoequence of the Goldstone phenomenon (see Chapter 30).
Finally, as we btieHy discuss below, while in theories in which the dynamical variables
live in ftat eudídean space, instwrtoru; are always aswdated with a degenera.cy of the
cla.ssical minimum of the potential, thís is no longer necessa.rily the case when the space
has curvature or ia topologically non- trivial.

40.1!..3 In.stantons in stable theorns wi!h unique c/nssicaiiJàcuum

Let us agaln consider tbe example of thé cosine potential but v.ith a dílferent interpre-
tation. We now assume tha.t the toordinate x represents a point on a. circle of radius
'21r /..,lg. The hamiltonian then represents a 0(2) rota.tor in a potentíal (Section 3.3) or
a one-dimensional classical spin chain in a magnetic field. The classical minimum is no
longer degenerate heca.use ali minima are identified to one point on the círde. The quan-
tum ground state ia equally unique since the Hilbert "!'""" consi.sts in strictly periodic
eigenfunctions (I" 0). Still instanton oolutions exíst but they start from and retum
to the sa.me classical nünimum, winding a.round the cirde. They are st&.ble because the
circle ia topologically non-trivial. Tbey generate the same exponentially small cortections
to the perturbatlve expansíon that we have described before.
Finally to gain some further insight ín the problem we can generalize the problem to
the O(N) rotator of Section 3.4. The classical action becomes

S(x) ; f dt [!:;.2(t) + 1- ZJ (t)]. (40.32)

The classical so!utions a.re the same, however the degeneracy and tbe stability properties
a.re different. For N > 2 the so!utions whlch wind around the sphere have N -2 dírectíons
of instability. Theír contributions have to díscussed in the framework of tbe large order
beha.viour of perturbation theory (see Chapter 41).

40.3 Instantons in Stable Boson Field Theories; General Remarks

We now briefly discuss instantons in stable t.heoriw, connecting for exarnple degener-
ate clBBSica! minima. The most interesting examp!es correspond unfortunately to scale
inva.riant classical theories a.nd the evaluation of the instanton contributions at leading
arder, whlch formally follows the Unes presented in Chapte: 39, leads to difficu!ties due
both to UV and IR divergeoces. Some of them will be exarnined in Chapter 42. Sínce
for the two examples we consider in Section< 40.4,40.5, they have not been satisfactorily
solved yet, we here restrict ourselves to classical considerations.
We first ma.ke a fuw elementary and general remarks about the possible existente of
instantoM.
860 Degenerote Cla.ssical Mimma and lnsiantan.< 40.3

ScaJ.ar field thearies. Let us assume that the action h&s the form

(40.33)

in which 1/;' is a multicomponent scalar boson field, 9;;(4Í) a positive ma.tiÍ:'( (positive
delinite almost everywhere) and

minV(<f>) =0. (40.34)

{<lo}

Let us Write the a.nalogue of equn.tion (39.39):

(2 d)
f 2i (9i;(4i)ô,..fiA,<P.
• ') d d :z: = d f V (tf>e) d d :Z:.

We see that this equation ha:s no solution for d > 2. For d = 2, ít bos only solutions for
wbicb
v (tf>,(:r)) =o. (40.35)
The condition (40.34) then implies that t/>e(:r) is for ali :r a mínimum of the potential:

fi}•) =O,

and therefore tf>e(:r) is a solutlon of the field equations:

These two equations a:re in general incompatible, except if V(tf>) vmrishes identice.lly. In
the latter case action (40.33) corresponds to a two.-dimensional model on a Riemanninn
manifold. A particula:r class of such models, mo deis based on homogeneous spaces, has
been díscussed in Chapters 14, 15. Among them, the CP(N- 1) models are known to
admit instanton solutious a.nd we 5he.ll describe them in Section 40.4.
Gauge !Mories. lf, in additlon to scalar fields, the theory contains gauge fields A~, the
gauge invariant action has the furm

. S(tf>,Apl =f ddz [i F,;.~.+! (D.</>,) 2

+ V(if>)]. (40.36)

Let us assume that there exists a linite action solution {4><, A~} (in wbich A~ iB not a
pure gauge), and let us calculate the action for >.A~ (Ãz) a.nd tf> 0 (>.:r). We obtain:

S[t/>'(>.:r),.>.A~(>.:r)) =À4-d f t(F,...)2ddz+>.2-d~ f (Dptf>}'ddz

+>.-d f V(t/>)ddz_ (40.37)

Stationa.rity at >. = 1 itnplies:

(4 d)t. f d 4.,(F.,.)1 + (2- d) ~f (Dptf>) 2 dd:r- d f V(tf>)d•x =O. (40.38)

40.4 Deg•nemu Classioal Mínima and l....U.ntoru 861

We ree that no solutlon can exist for d > 4, .since a sum of negative termS cannot va.nish ..
For d 4 we get two conditions:

(40.39a)
(40.39/J)

Wríting the fteld equations, we conclude that A~ is the solution of the pure gauge field
equatiom. AI; we show in Section 40.5, instantons can indeed be found in pwe gauge the-
aries. Equatian (40.390), which is now a.n e<,juation for ,p•, then leads to the integrability
conditions: •
(40.40)
in whlch the matrices t• are the generators of the Lie algebra. Tbe conditiom (40.40}
together with the equatian {40.3911) show that in general the S}'Stem bes only the trh•ial
solution tf>• = O.

40.4 Instantons in CP(N- 1) Models

To illustrate prevlous considerations let us 6rst give one exarnple of instantons in a two.
dimensional model, tbe CP(N -1} model which is baBed on one ofthe symmetric spo.ces
presented in Appendbt A15.4.3. We just dt9:ribe the nature of the solutions and refer
the reader to the literature for a deeper and more detalled a.nalyois.
We consider a set af N wmplex fields '{>0 , subject to the condition

(40.41)

In addition two vectors <Po and 'P'~ are equivalent if

'P~(:t) = e>/1(•) 'Po(:t}. (40.42}

These conditions charncterize the manifolds CP(N- I) which are isomorphic to the
symmetric spscesU(N)/U(I)/U(N-1) and thus a special caseofcomplex Grassmannian
m!!.IÚfo!ds. One form of the unique classical action is:

(40.43}

in which the oovaria.nt derivative DP is:

D~ =a~ - .p. a,."'. (40.44)

The field .p · a,.'P is a composite gauge lield for the U(1) gauge inva!:i.a.nce defined by the
transformation (40.42). Note tha.t the CP(1) model i.s locally isomorphlc to the 0(3)
non-lineax <1-model, witb tbe identffication

(40.45)

The existence of locally stable non-trivíal minima of the action, ín two dímensions,
follows from símple considerations. The folJowing inequality ho!ds

jd 2
x ID,."' 'f i<~.D·"'I';::: o, (40.46}
862 Deyenemte Classical Mínima and Jnstantons 40A

(s,.. being the antisymmetric tensor). Expanding this expression we obtaín

S('P) 2 D~ 1 Q(rp), (40.47}

with
Q(tp) = lt<,.. f d,x D,.rp. D.rpl (40.48)

The integrand in the r .h.s. is a pure dívergence. Indsed using equation (40.4i), we ca.n
traruform equation (40.48) into

Q('P) = lic,... j d x8,. [r,il..a.r)l

2
(40.49)

The quantity Q(l") is a topological charge: ít d~ only on the behaviour of the

classícal rolution for lxlla.rge. Finlteness of the actíon implíes that at large distances tp,
bewmes a gauge transform of a constant vacuum solutíon,

(40.50)

The topological charge measures the variatíon of the angle A(x) on a large círcle, whlch
is a multiple of 2,.,

Q('P) IJ d:r,.a,.A(:r)l = 2wn, n 2 o, (40.1H)

~ S(tp) 22rrnjg. (40.52)

We have been thus led to the cousideration of the homotnpy classe6 of mappings from
U(l), i.e. 5 1 tn S 1 • •
The local minímum corresponds to the equallty

(40.53)

whlch implies that the classical solutions must satísfy first arder partia! differential equa-
tions (self-duality equatíons):
D,.'l' = ±iE,..D.<p. (40.54)
It is easy tn show that the vectors soluti<>n of the equations (40.54) are proportional to
holomorphic or· anti-bolomorphic (depending on the sign) vectors in the variable z =
:r, + i:r,. Using theD equation (40.41) and the eqnlvalence (40.42) the holomorphic
solution can be C8St into the form (up to a gauge transformation)

'Pn P.,(z) j .JP:P , (40.55)

where the property (40.50) implies that the quantities P.,(z) are polynomials in z without
common roots. This form is consisteDt with the conformai invariance of tbe classical field
theory. The anti-holomorphic solution corresponds to intetchange <p 8lld fÍl· Translating
the CP(l) minimal solutíon (polynomíals of degree I} in the 0(3) u modella.nguage ooe
fiods
z +z .z - z 1 - z•
<Pt = 1 + zz ' <h 1 1 + iz ' q,3 1 + iz ·
We recognize the stereographlc mapping af the sphere S2 onto tbe piWJe.
40.5 Degenemte Clo.ssical Minima and Jnstantons 863

40.5 lnstantons in tbe SU(2) Gauge Theory

Let us now give aJl eJ<ample of instaJJtons in four dimensions. According to the a.nalysis
of Secti<:m 40.3, we can restrict ourselves to gauge theories. Actually it i.s sufficlent to
consider the gauge group SU (2) sim" a general theorem states that for a Lie group
containing SU(2) as a subgroup the inst.antons are those of the SU(2) subgroup.
ln 0(3) notation the gauge action is:

S (A,.) = -}g j 2
!F,..(x)l d4 x, (40.56)

with
(40.57)
The existenre and some properties af instantons in this theory follow frnm simple con·
siderations very similar to those presented for the CP ( N 1) model.
Let us define the dual of the tensor F,.. by

(40.5S)

Then the inequallty

(40.59)

yields
j d x(F,..) <:: I! d'xF,.,. ·F!'VI·
4 2
(40.60)

The quantity F,.. ·F,.. again iB a pure divergence:

F,.•. F,.."" 8,.V,.. (40.61)

wíth
(40.62)
As in the case of the CP(N- 1) model, the r.h.s. of equation (40.60) is a topological
quantíty, which depends only on the behaviour of the gauge l:ields at large dista.nces.
We have met it alrea.dy in Subsectiou 20.3.3 (equation (20.110)) in the calculatiou of
the axial anomaly. We have in particular shown that its values are qmmtized ( equation
(20.115)).
The finit.eness of the action, implies tha.t the classical solution should a.symptotically
become a pure ga.uge, i.e. wíth our conv.entians)

(40.63)

iu whkh cr are Pauli rna.trices a.nd g (x} i• an element of SU(2). Since SU(2) is topolog-
ícally equivalent toSa, we are now led to the study of the homot<>py classes of mappings
from S3 tos., which are also classi6ed by an integer called the wmding number.
The simplest one to one mapping corresponds to an element g(:r.) of the form

r {40.64)
854 Deyenerote Classíca/ Mínima and Instantons 40.5

and thus
(40.65)

f . - f n"-
lt follows that:
d x F~· F~· = dQ n~ · V,. = 32" 2 , {40.66)

in which dfl is the measure on the sphere and the unit vector normal to the sphere.
lf we rompare this result witb equation (20.115) we see that we bave indeed .found tbe
minimal action solution. In general we then expect

(40.67)

and therefore
S [A,.l ~ Sw:2 nfg. (40.58)
The equality, whlch conesponds to a local minimum of the action, ís ohtained for fi.,}ds
satisfyíng the ..,lf-<luality equations

FIJ."' =±F~~ov, (40.69)

which are first order partia.! differentia.l equatíons. The one-insta.nton solutlon, wlúch
depends on a scale parameter À, ís

(40.70)

The .tructure of the semiclassú:al uacuum. In contras!. to our ana.lysís of tbe quantum
mecha.nica.l problems, we ha.ve here diBcussed the existence uf instantons without reference
to the structure of the cla.ssica.l vacuum. Tb find an interpretation of the il18tantowl in
gauge theories, it ls com.,ni<lnt to express the results in the temporal gauge ~ = O. In
this gauge the cla.ssica.l mínima of the potential correspond to space components ga.uge
fields which are pure gauge functions of the three space variables:

{40.71)

However we knów that the wa.ve functions have to satisfy GaUBii's law. This bnplies that in
the absence o! externai static charges the wave function is invariant under infinitesimaJ
gauge Uansformatíons (Section 19.3). Therefore two gauge functions g(x;) which are
ínflnitesima.lly dose do not correspcnd to different \18CU"· Tbi$ lmplies th"t the structure
of the dassica.l vaeuum iB conneeted to the homotopy classes C>f mappíngs of the gtoup
elements g ínto oompactífied R 3 , i.e. again of S3 into Sa and that the cliiSSica.l potential
hBB a periodic structure reminiscent of the periodic cosine potentlaL One can verify
that the ínstanton solutlon (40.70), trllilSported into the temporal gauge by a ga.uge
transformation, connect8 vacua with different winding numbers. Tberefore, as in the
"a.se of the periodíc potentíal, íf W\> want to project onto a proper Quantum mechanical
eigel18tate, the "6-vacuum•, we have to add, in analogy witb exprestrlons (43.65, 43.56),
a term to tbe classical actíon of gauge theories:

(40.72)
40.6 De11enernte Classical Minima anti ]nstnntons 865

and we can then integrate over ali fields A~ without restriction. A similar propen:y holds ·
for the CP(N 1) models. At leBBt at the semiclassícallevel, the field theory depends
on an a.dditional pa.rameter, the angle G. For non-vanishing values of (J the a.dditional
term violates CP COD8ervation, and is at the origin of tbe strong CP violation problem.
lntrodudian of fenníons. If we now consider the case oí QCD, gauge fields coupled to
fermioru; Q, Q through an action

severa! comments can be ma.de. First., if one fermion field i5 massless, then according to
the analysi.s of Subsection 20.3.4, the determinant resulting from the fermion integratioJJ
va.nishes because the Dirac operator Ju.s a vanishing eigeJJvalue. Then the instzmtol15 do
not contribute to the functional integral and the strong CP violation prohlem is solved.
However it aeems that it is inconsistent with experimental data to assume thllt one quark
is massli'SS.
Second, in the limit in which the quark,s are massli'SS, the B)'Jilmetry of the action is
the chiral U (NJ) x U (Nt) group which has a.n axial U(l) subgroup. As we a!rea.dy
discu&;ed in Section 20.5, if the instantons contrjbute, they solve the U (1) problem, i.e.
the absence of a Goldstone boson associated with the a!most spontBJJoous brea.king of
the axial U(l) current.
The quantum mechanical vacuum. In both examples, CP(N l) models, and noJJ-
abelia.n gauge theories, the dassical theory is seale invariant. Therefore the solutions
depend on a scale parameter >. which has to be taken as a collective coordinate in the
quantízation. This lea.ds to ""ry serious problem.s 8S the analysis of the massless <fl~ field
theory will reveal (see Chapter 42). In an aeymptotically free theory, lilre the two theories
considere<! ahove, the difficulties come Eram the infrated region, í.e. from ínstantons of
la.rge size for whicll tbe semiclassical approximation is no longer legitimate because the
interaction increases at large distance (see Chapters 33,34). Only with an IR cut-off is
the ca.lculation possible. A solution ís to coruJider a system in a finite volume.
The role of instanton.s in the infinite volume is thw; not clearly understood. The only
piece of information presently available concems the 0(3) non-linear a-model, wbose
instantoru; are derived from tbose of tlu:i CP(l)-model. It ha.s been recently ra.ther in-
directly argUed, by mapping the c-model auto a one dimensional quantum spin cha.in,
that instantons ll.l'e only relevant for e " but then lea.d to " quite difJerent physks.

40,6 lnsta.ntow and Stochastic Differential Equatíow

In this oection we discuss the role of instanton.s in the contat of stocllastic dynamics.
We consider tbe problem of evaluating the decay prohability of a metastahle state by
thermal fluctua.tions. At first sight one could imagine that thi.s topic shou!d bave con-
sidered a!rea.dy in Chapter 39, simultaneou.sly with the problem of decay by quantum
fluctuations. We show here that technica!ly the pmblem has n more dírect relation with
quantum ground state degeneracy. At tbe end of the section we also briefly examine the
role or instantons when the equilibrium distribution has degenerate mínima.
We recall (aee Chapter 4) that with the Langevin equation (4.37)

q; -!ã;E(q) + "•(t), (40.73)

866 Degenero-te ClaSsit:al Mínima and Jnstantons ~.6
with gaussia.n noise (4.2):

(dp(v)j = !dvlexp [-~11- 1 j vf(t)dt],

is associated the lwrmíiían bamíltonian H (4.40):

H ~ [-nll. + {11- 1
{'li'E (q))
2
~M(êl)], (40.74)

and therefore the effect.ive action:

(40.75)

The classic.al limit here corresponds to the small 11 ( also low temperature) linrlt. At
leading order in a semiclassical analysís the term ll.E can thw; be omitted. It follows
that to each extremum of the function E(q) corresponds a minimum of the classical
actíon ( 40. 75). In particular, íf E(q) has more than one extremum, t.he cl8B5ical actlon
has lnstanton solutions. Let us dis<:uss using a simple example tbe role of instanton
solutíon.s in this context.
An example. Let us consider tbe exa.tnple of a function E(q) which is such that the
corresponding distribution e-E(•l/0 is not nomuiliz.able:

(40.76)

Physically we then know that if we put a pa.rt.icle at time O a.t the relatiw minlmum
q O of the distributíon, tbe:n after some time the pa.rt.ide will escape from the well, as
a resuit of the therma.l fluctuations described by the Langevin equation. Our problem is
to evaluate in the smrill 11 limit the escape probability per unit time. We know that it
ís given by tbe inven;e oi the smrillest eigenvalue of tbe hamiltonia.n (40. 74)(see Section
4.3). When the distribution is not norma.lizable this eigenvalue is strictly posit.ive.
However to ali orders ín a parturbative expansion in powers o{ 11, e-Ef(2!l) iB the
ground sta.te eigenvector a.ssociated with the eige:nvalue O. lt follows that the eigenvnlue
must be related to non-perturbative eHect. We indeed note that the function E(q) has
two extremalocated at q: O and q = 1. Hence E'l(q) has two mínima. We can su.spect
thAt the instantons connecting these minima provide a oolution to our problem.
The classical equation of motion. Since we only look for solutions wbich connect
mínima of the potentia.l in ínfinite time, we CllD immediately integrate a ftrst time the
equation of motion !Ulrl find:
q = ±E'(q)/2. (40.77}
Therefore the ínsta.nton action is:

.A(q) n-t f dtq2 ~n- 1 f dtqE'(q) ~n-l (E(l)- E(O)) (40.78)

However this is not tbe end of the story. Indeed, ín Section 4.S we have shown that the
degeneracy between tbe mínima and maxima of the function E(q) iB lifted by the first
qnantum correction. Therefore the two mínima of the actlon a:re not rerilly degeuerate
and no ínstanton can oonnect them. What ,..,a)ly happens ís that liA! have to cons:íder
40.6 D"9enerote Classical MiriirtUl and I~ 867

only closed trajectories passíng tbrough tbe origin. If we consider a finite time interval
/3 we can find sucb trajectories. In the ínfinite p limít they decomp06e into a succession
of instantoiLS and anti-instantons. The limit of the classical BCtion is a.n even multiple of
the insta.nton action. Tbe leading contribution is thus:

A(q) = (E(l) E(O)) jfl. (40.79)

These arguments obviously are independent of the precise farm oi the function E(q).
Therefore we conclude quite generally that if the function E(q) b.e.s a relative mínirnUJII
wbere E E.,;0 separated from a lower minimum (p05Sibly E- -oo) by a local mB.Jt-
imum E = Em,.., theo the eigenvalue corresponding to a eigenfunctíon concentrated
around the first minimum is of the order e-D.E[fl in whicb llE is the variation of the
function E:
llE Emax- Emin ·
The time ..,D.E/fl has the following interpretation: it cbaracterize:s tbe exponential decay
of the probability of finding q(t) near the orlgin when the initial conditions at t = O are
q(t "'O)"" O.
Note finally that in order to complete 'the calculation of the eigenvalue it is necessary
to use multi-instanton techniques which will be developed in Chapter 43.
Generulúation. Prev:ious analysis can be generali..ed to severa! degrees of fmedom.
The classical e<,~uation of motíon tben takes the form:

(40.80)

If we con.sider only finite BCtion solutions and infiníte time, the e<,~uation can be integra.ted:

(40.81)

The classical a.ctioo ia then:

.A(q) = n-• f dt<l =~fl- 1 f dtq ·\!E ~fl-1 (Emax- Emm), (40.82)

in which Emtn and Em., a.re respectively the valucs of the functíon E (q) at the relat:ive
minimum and tbe saddle point one has to pass to be able to go lower.
Quantum Field Theory. We now consíder a dynamics governed by a purely dissipative
Langevín equation, whicb formally con....,rges towards an equilibrium distrihution cone-
sponding to the path or the fnnctional integral of a d dimensiono.! euclidean quantum
field theory (see also Sobsection A36.1.3). The effective action theo is d + 1 dimensional.
Let us again examine the problem of an unstable field theory wjth euclidean action

(40.83)

We know that a quantum state concentrated around the minimum ,P(:.:) = O will decay
due to qua.ntum lluctuatlc>ns, and we have calculated the rate by inst!l.llton methods.
We now wa.nt to evaluate the decay ptobn.bility due to thermal ftuctualions. The
relevant effective a.ction at lea.ding arder reads:

(40.84)
868 Deyenero.te Classiool Minima and Jnstantons 40.6

Forma.lly the discussion follows the same líne as in the case of a finite number of degrees
of freedom. The ptoblem is to identify the minimum aod the maximum of the a.ctíon.
The minimum is easy to find: 4> ;;;; O. The maxjmum requires some more thought. It
does not correspond to a ronstant field configuration: 4>(:z:) = -2m2 jg. lndeed it is
sufficient that some part of the lield starts passing the barrier. This mea.ns that the
relevant maximum of the actíon instead oorresponds to a static instanton configuration.
Using the a:rguments of the preceding subsection, the result follows:

r~ exp[S; ..,.;n].

Degenernte minimum. Anotber problem arises when the function E(q) hM a degenerate
minimum. Let us assume that the corresponding distribution Íll normalizable. Then
the ground state eigenvalue vanishes. The interesting question is how to calculat.e the
düference betw....n the two lirst eigenva.Jues, ditference whlch vanisbes to ali orders in
perturbation theory. This is the problem we have solved in Section 40.1 for the double-
wdl pot.entia.l. However here the set-up is slightly ditferent because, if E(q) is regula.-, as
we always assume, tbe two mínima are neres•arily separat.ed by a maximum and tberefúre
E 12 (q) are thr"" minimo.. An ex~>mple is:

(40.85)

and therefore:
(40.S6)
This time we look fúr instaoton solutions wbicb connect q = O to q = 1. However
in the ínftnite time limit only instantons which go from O to 1/2 or 1/2 to 1 surv:ive.
From the ana.lysis af tbe previous problem we guess that the relevant conftgurations will
conespond to glue togetber two ínstantons. Therefore the dilference betwoên the two
leading eigenva.lues, whicb is a.lso the 5eeond elgenva.lue <1 , is again of the form:

(40.87)

in which Em~. and Ernu are respectively the values of the funcbon E{q) at the degenerat.e
mínima and at the maximum whicb connects them.

Bibliographical Notes
Polyakcw has einpbasized the doubJ.e.well as the cypical instanton problem in
A.M. Polyakov, Nucl. Phy•. B120 (1977) 429.
The role of instantons in the question of spontaneous syrnmetry breaking of discret.e
symmetries has b""n díscussed in
J. Zinn-Justin, Ba.sko Polje Summer Schoal 1976, Saclay preprint DPh-T/76{99, un-
publisbed.
We ha"" used S<'Aiing arguments to discuss the existence of ihstantons
G.H. Derrick, J. Math. Phys. 5 (1 964) 1252.
References ahout instantons in the CP( N - 1) model include
A. Jevickí Nucl. Phys. Bl27 (1977) 125; D. Fõmer, Nucl. Phys. Bl30 (1977) 38; M.
Liischer, Phys. LetJ.. 78B (1978) 465; A. D'Adda., P. Di Vecchia and M. Lüscber, N~<ei.
Phys. Bl46 (1978) 63; B152 (1979) 125; H. Eichenherr, Nud. Phys. Bl46 (1978) 215;
V.L. Golo andA. Perelemov, Phys. Lett. 79B (1978) 112; A.M. Perelemov, Phys. Rep.
146 (1987) 135.
40.6 Degemrou Cla.osícal Mínima and Instantons 869

For in.sta.ntons in gauge theories see

A.A. Belavín, A.M. Polyalrov, A.S. Schwaru: and Yu S. Tyuplcin, Phys. Lett. 59B (1975)
85; G. 't Hooft, PhyB. Rev. Lett. 37 (1976) S; Phys. Rei!. 014 (1976) 3432 (Erratum
Phys. Rev. Dl8 (1978) 2199); R. Jackiw a.nd C. Rebbi, Phys. Rev. Lett. 37 (1976) 172;
C.G. Callan, R.F. Da.shen and D.J. Gross, Phys. Lett. 63B (1976) 334; A.A. Belavin
nnd A.M. Polyakov, Nucl. Phys. B123 (1977) 429; F.R. Ore, Phys. Rev. Dl6 (1977)
2577; S. Chadha, P. Di Vocchia, A. D'Adda and F. Nicodemi, Phys. Lett. 72B (1977)
103; T. Yoneya, Phys. Lett. 71B (1977) 407; LV. FroiOY and A.S. Schwarz, Phys. Lett.
SOB (1979) 406; E. Corrígan, P. Goddard and S. Templeton, Nucl. Phys. B151 (1979)
93.
For a recent dísCU55ion of the U (1) pr<>l>lem """
G. 't Hooft, Phys. Rep. 142 (1986) 357.
A pos.<;il>le solution to the strong CP violation prol>lem, whích h"" been shown since then
to lead to difficulties wíth experiment, was proposed in
R.D. Pecc::ei. and H.R. Quinn, Phys. Rev. Dl6 (1977) 1791,
and dlscussed in
S. Weinberg, Phys. Rev. Lett. 40 (1978) 223; F. Wilczek, Phys. Rev. Lett. 40 (1978)
279.
Another lield theory wíth interesting pmperties ís the two-dimensional abelia.n Higgs
model
C.G. Ca!lan, R.F. Dashen and D.J. Gross, Phv$. Lett. 66B (1977) 375.
o Degenerate Cla.ssical Mintma and lnstantons A40.l

PPENDIX 40 4

40.1 Trace Formula for Periodic Potentials

'e consíder a hamiltonian H correspondlng to a real períodic potential V (x) with period

V (x +r) = V (x) (A40.1)

igenfunctions !/;~ (x) are then also eigenfunctions of the translation operator T:
(A40.2J
et us put the system in a bol< of size N T with periodk houndary conditions. This
nplies a quanti<ation of rp

(A40.3)

:alling 1/;p,n the eigenfunction of H corresponding to the band n and the p5eudo-
lomentum 'f'p> we can write the matrix elements of Te~BH:

(x' iTe-PH I:r) L .P;,.. (x')e-PE.(.,_) 1/;p,n (:t +r)' (A40.4)

p,n
nd therefore:
(x' ITe-IIHI:t) L .P;,,. (x') .-llE.(IP,)+i"'• >/-·•·" (:t). (A40.5)
p,n

rhis implies for the diagonal elements:

(x ITe-PH[x) L l!/;p,n (:z:)lz.,-llE.(,.,)+•"• (A40.6)
P•"
3oca.use in a transla.tion of a period (equation {A40.2)), the eigenfunctions are multi·
llied by a phase, equation (A40.6) shows that the J.h.s. is a periodic function of :>: with
Jeriodicity T. Tberefore:

(A40.7)

rhe functians 1/1.... (:r) are orthonormal over NT:

(A40.8)

We now tske the large N limít and obt.ain expression (40.51):

(A40,9)

since reality implics:

(A40.10)
1 PERTURBATION THEORY AT LARGE ORDERS ANO
INSTANTONS. THE SUMMATION PROBLEM

In Chapter 37 we have discussed the analytic struct ure of the gr<>und state energy E(g)
of the anharmoníc 05Cillator. We have a.rgued that E(g) is analytic in a cut-plane, and •
calculated hy instanton meth<>ds its imaginary part on the cut for g small and negative.
On the other hand, perturbation theory yields E(g) for g smaD as a power sedes in g:
00

E(g) = L Ekgk. (41.1)

k=O

We explain in this chapter h<>w the small g behaviour of Im E(g) is related to the h(;.
haviour of the coefficients Ek when the arder k becomes large. We then generalize the
method to the class o[ potentials for which· we have calculated instanton contributions.
The sarne method can be readíly applíed to hooon field theoríes, while tbe extension to
field theories involving femúons requires, as we shaw, some additional considerations.
We already know that tbe expansion (41.1) is divergent for ali valu"" of g. This impli""
that, even for 9 small, the series d<>es not determine the functíon E(g) uníquely. We thus
examine the ímplicatíons of the large order bebaviour for the problem of the summation of
the series. Finally we descrihe a few practical methods rommonly used to sum divergent
series of the type met in quantum mechanics and quantum field tbeory. Some of these
2
methods have been successfully applied to the ( ~ 2 ) field theory in 2 a.nd 3 dimensions
and have led to the accurate predictions of criticai exponents display.<i in Chapter 28.

41.1 Quantum Mechanics

41.1.1 Real instanton solutions
The anhannonic oscillator. We first consider the ground state energy o[ the quartic
anharmonic oscillator. Since E(g) is analytic in the cut.plane 8.1Jd behaves llke 9 1/ 3 for
g large, it has a Cauchy representatjon for E(g) of the form

E(g) ! +~
2 1T
1"
-oo
ImE(g') dg'.
9' (g' - 9)
(41.2)

This equation leads to an integral representation for the coefficíents E• ohtained by

expanding the integrand in powers of g:

fork>O. (41.3)

Let us now examine what happens when k, the arder in the expansion, becomes !ruge.
Dueto the factor g-*. the dispersion integral (41 .3) is dominated by thé small negative
9 values. In Chapter 37 we have calculated lm E(g} for g small and negative. We can
here use thís result to estimate the large k behaviour of Ek:

~ -
1j"- (8) -1- - - d
E• k-oo11' -71'
112 4
e /lo
.;=ggk+l g, (41.4)
 f2 Peri'urbation Thevry nt Larye Orders nnd Irurtm>.tons 41.1

he íntegration can be done explicitly and we find:

(41.5)

uccessíve correctiono to the semiclassical result yield a series in powers of g which,

>tegrated, genetates a systematic expansion in powers of 1 1k •
Generol potenti<Jls. The same argument is applicable w the situation described in
:hapter 38. We can calculate the energy of the unsta.b)e state in power series of the
oupling constrmt g by maldng a systematic expansion around the relotive miuimum of
be potential. On the other band we can, as above, derive ftom the knowledge of the
maginHry part of the energy levei !ar small coupling, B!'l esthnate of the l:>ehaviour of the
>arturbative coefficient.s at large order. Let us oonsider the action;

S(q) =f dt IW<t) + g-ly (q..,íg)J. (41.6)

rhe analogue of the dispersion relation (41.3) is:

The bebaviour of lmE(g) for 9 smell il! gM>n by expression (38.62). lntegrating near
g = O we obtain

whe1ee Ais tbe clBSSical aciiOD

(41.7)

(41.8)
I
f
We now see generic features emerge: at large ordels, the perturbative coefficient.s Ek
behave like
Ee ~ Ckb- 1k!A-k. (41.9)
·-""
Tbe fact<>r kl is universal and characteristíc of the semiclassical or loop e><p8.ll8ion. lt
shows that the perturbation seties is a divergent eeries. Tbe fact<>r A-& depends only on
the action, siDre it is the action of the classical solution; in particular it also characterízes
the behaviour at large ordert of the excited energy l"""ls or of eorrelation functions. The
power /c 0 comes from the power o f g ín front of the resnlt. It in particular depends
on tbe number of continuous symmetrill5 braken by the classical solution, but it would
also cbange íf we considere<! an excited stote rather than the ground state. Thíll can be
verilied by explicitly calcnla.ting the imaginary parts of the energy of the excíted leveis
as explained in Chapter 37, &nd using equatjon (41.3). The parBilleter bis in general a
balf ínteger. Finally there is a const.ant multiplicatiw factar c wbich depends in a more
complicated way ou ali tbe specific featum~ of the expanded quantity.
DisctiSsion.. In both examples we baw been able to cal enlate the large order behaviour
of perturbo.tion series from the decay rate due to barrier penetration of an unstable
4l.l 873

minímum oi lhe potential. For the potentials considered in Chapter 38 lhe action A is
positive and therefore a.ll terms in the perturbative expansion bave the sa.me s\gn.. The
same property bolds for the a.nharmonic oodllator in the uostable case, i.e. when g is
oegative. However for g positive, in whích ca.se perturbation series has been expa.nded
around tbe stable mínimum of the potential, Wl' observe lhBt the perturbative eoefficients
oscillate in sign. Also we note tbat for g > O the insta.nton solution becomes purely
ima.gino.ry. Thls wlll belp to understand how to obtain tbe lBTg<> behaviour in the generic
stable case.

41.1.2 Complex instantt>M

We have up to now cbarBCteri•.ed the la.rge order bebaviour of perturhatíon theory in two
cases, in the generic case in whlcb we expand around a relative minimum ofthe potentlal,
a.nd in one special case in whlcb we were expa.ndiog BJou.nd an ahsolute minimum of
the potential, hut whlcb by aoalytic continuation in the coupling coosta.nt could he
transforme.:! ioto lhe ull5table one. We now consider oct.ionsoftbe form (41.6), in which
the potential V(q) is an entire function of q and satisfies tbe ooodition:
V(q) '- ~l +O (,f'),
=
and 85!;ume that perturbation theory is ex~anded a.round q O, the absolute minimum
of the potential. Then cleo.rly no real instantoo solutions can be [ound. Following tbe
example of the anharmonie oscillator, we are going to 888Wlle tbat we can introduce )
parameters in tbe potential whlcb a.llow us to go to an unstBble si.tua.tion by analytic
continuation in these parameters. We ca.n then obtai.D the large arder beha.viou:r from
e:xpression (41. 7). We then use the inverse analytic continuation and return tothe original
situation. If notbíng droonatic ha.pperu;, tbe large behavioru of tbe original expansion will
be given by the analytic continnation of expressíon (41.7). )
We can now fonnula.te the principies of the la.rge order behavioru calcnlation In lhe
original tbeory: to the oomplex zeros (at finite or infinite disl.a.nce) of the potential V(q)
are aasociated complex instanton solutions, with, in genero.l, complex (or exreptiona.lly
negative) oct.iou. Tbese imtanton solntions are cBDdidates to give contributions to tbe
'
)
)
large order behaviour. From e:xpressíon (41. 7) we see tbe.t tbe action(s) with tbe smallest
modulus (wben the action iB oomplex, there will be at least two complex conjugate
actions) giVES the dominant contribution to tbe la.rge arder behaviour. Note that tbe
difference we have found between the anharmonic oscillator and tbe unrt.able case Ls
generic. In the stable case, the classical action is non-rea.l posíti~, a.nd the phase of the
perturbative coeffi.cients at lBTg<> order varies with the arder.
When we diseusB the question of summability of divergent series we shall see that thls
dilference in the behaviour o[ the series plays a.n esse~~tial role.
A special ca.se. Tbe previous di&:ussion does D<>t immediately apply to the case of
potentials with degenere.te mínima.. Let us indeed coru;ider •ncb a potential as the limit
of a potential whlcb has two mínima at wbicb the va.lues oi the potential a.re very close.
From the explicit form of the action, we see that the cl.assic.aJ BCtimi has a limit whicb is
twice tbe a.ctioo of the instaotoo whicb connects the two mínima of the potential

A = 2 [ • J2V (x)d:t.

However the amplitude in front ofthe expression (41.7} diVI!Tg<:S wben "'+
iB a.n e:xtremum
of the potential. Thls result can be e88ily undeiStood. When tbe values at the two mín-
ima approach each other, tbe time spent close to tbe second minimum of tbe potential
874 Perturbatwn Theory at LaT!Je Orders and lnstantons 41.2

by the classical trajectocy correspondíng to tbe ín.stant.<m solution diverges. Therefore

fluctuation.s whicb tend to change this time leaw the actíon almost stationacy. Corre-
spondingly one eigenvalue of the operator second deríwtíve oí the action at the saddle
point goes to zero, and this explain.s the divergence of expression (41.7) in this case.
Only by atlow!ng tbis time to fluctuate, and introducing e.n additional time collectíve
coordinate can we obtaln the correct answer. Let us al,;o note that here, as in the case
of unstable mínima, the cla.ssical action Íll pooitive. This ls the snurce of oorious difficul-
ties when one tries to sum tbe perturbation series. We sball dlscuss tbese problems in
Chapter 4.3.

41.2 Scalar Field Theory

In Chapter 3g we have evaluated the contributions of instaotons to the decay rate of

metastable states. We can here apply thrne results to large order behaviour estimates.
In the case of the r/1 4 lield theocy the discontinuity across the cut o[ the n-point function
reads (equatíon (39.26)):

.
dtscZ
lnl
(%~o····:r,.)~
(A2.. )"I" g(à+n)/>(detMM
e-AJo ' - t -In
0 ),.. F.(:rl,•··•x,.), {41.10)

with
F,. (:q, ... ,x,.) = md+n(d-2)/26n/2 f IT f
ddxo
f-=1
(m (:ri :to)). (41.11)

Using previnus arguments, we CM immedíately trMslate tbis result into a large order
behaviour estimate for correlation functions

{z(n) (:rl,···,Z'nl:-
l} - .2_
2
i7r f ~d· glr:+llSC zln) ( Xf,- .. ,%n~ l
and therefore
1
{ zfnl{x 1 , ••• ,:r,.)} - - ~12 F,.(x 1 , ... ,x,.)A-"12- 11T(k+d/'2+n/2).
u-oo 2i'lf(2")
(41.12)
In a general scalar boson fleld theory, íf we = find instanton solutions, the same atgu•
ments wíll lelíd to:

Í: c. {xJ, ... , x,.) k&-l A-"k!, {41.13)

dominant
&Midle poínt6

in which:
(i) A is the instwlt.<Jn action, whích ís in general complex;
!
(ii) b = (n + 5) and 6 is the number of symmetries broken by the dassical solution;
(iíi) C,. (:r" ... , :r,.), whicb does not depend on k, contains the whole dependenre in
the externai arguments.
2
E:rample: thf RG 13-functíon in the ( .p2 ) ín 3 dimensions. The large order behaviour
ha5 been determinE<! by solving nwnerically tbe fi.eld equatíons to lind the dassical ac-
tion A ond calculating numerically the determína.nt. Tbe predictions of the aeymptotic
41.3 Perturba.tion Theary at Larye Qrners and ln.!!tantons 875

Table 41.1
The en<fficients 13. of th.e wuplí"9 comtant nmonnali<al•on vroup fun<tion f'J(g) dimdcd by
the lorge Onli!T .,.toma•• for th.e o (N) svmm.etric.p; field th.ea"!!-

k 4 6 7
N;Q 3.53 1.55 1.11!15 0.967 0.951
N=J 3.98 1.75 1.32 1.050 1.023
N~2 4.82 2.09 1.53 ].29 1.20 L!5
N=3 6.14 2.58 1.1!6 1.55 1.41 1.35

formulae have been compared witb the terms of the reries whicb h ave been ca.lculated
(see Section 28.2). The agreement is quite rea.sonable and gives us some confidence that
the large order behaviour estima.tes are indeed correet (see table 41.1).

41.3 Field Theoríes witb Fermions

In the case of boson field theories we have related the large order behaviour of pertur-
bation theory to the decay of the false vacuum for, in general, unphy>;ícal values of the
coupling collfitAnt. We e.xpect therefot<! oome modifications if we oonsider a system of
self-interacting fermioru;, or of fermlon.s intera.cting with bCISOns whicb themselves have
no self.interBCtion. {Actually the first case ean be reduced to the second one by introduc-
ing an auxiliary booon lield.) Indeed the Pauli principie wiU make the decay of the false
vacuum more difficult because it is not pOS5ible to h ave rnlllly iermions in the same state
to produce a classícallield, and this effect should be especially strong in low dimensions.
Of cuuree if the bosons have a self-intera.ction, this interaction will drive the decay of the
vacuum, and the fennions will no Jonger play any role.
Soon from the point of view of lntegrals, the difierence between fermions IUld bosons is
alsn immediately apparent. We h ave shown that the simple integra.! countíng the number
o f Feyrunan diagra.ms, whicb is also the ~· fteld theory in d =O dimensions, a.li"eady h as
the characteristic k! behaviour at large orders. Let us instead consider a zero-dimensíuna.l
fennion theory, i.e. an íntegra.! over a finite nurnber of fermion degrees of freedom:

I(.\)= f fi d~.~;exp
•=>
!(,D;;{; +ÃC;;~ç~{,(;,;,.e1 ). (41.14)

The quantities (i and {; are anticommuting variables and D,1 and C;;•t are a set of
numbers. Because we assume that we have a finite nurnber of anticommuting variables,
if we expand tbe exponential, we get only a linite number of non-vanishing teniJB in the
expansion. Therefore J(À) is a polynomia.l in À.

41.9.1 &:ampk of a Yukawa-!ike fie/4 theary

We now consider the VBCUum ampUtude of tbe following theory:

(41.15)
876 Perturbation. Theory at Larg• Orders and lnstantOilS 41.3

in whic:h the action is:

(41.16)

In the functíonal integral we haw d€noted by 4>(x) the scalar boron field, by ,J;(x) and
,.:•(x) the fennion fields. The pa.rameter g is a loopwíse €Xpansíon pa:rameter. Since there
is "" classicallimit for a feTmion field, the expression (41.15) is not very well rulapted
to the study of the vacuum decay. In fact we expect the fennioti fields to g€Derate an
elfect.ive interoction for the boson !ield 4J(:r), aud thís elfective ínteraction w:ill lerul to
the decay of tbe vacuum. This suggests tbat we should integrate over the ..;: and ;j,
variables, and study the instantons of the effective thoory for <i>(x). In ruldition the zero-
dimensional example has shown that tbe lermion integration gives some biDts about the
analytic structure of the thoory. After the integration over tf; and ~ we obtain:

{41.17)

w., are faced with a new difficulty arising from tbe integra.tion, the effective action ís now
non-local in 4J(x), a.nd leads to non-local field equatioos. However, because we are con-
cerned only with the detennínation of the la.rge behaviour, we ca.n simplify the effectíve
action. The determinam generated by the fermion integration ís, at least for the class of
relevaut 4>(x) lields, a.n entire function of the coupllng comtant .;g. As a consequence,
essemial singulo.ritíes can only be genero.ted by the range of tbe integratlon. lt is suf-
ficient to calculate the contribution to the functional integral of large fields oii(x). Tbis
situation ha.s to be contrasted with what would have happened íf .,P(x) and ,j,(:r) would
have been commuting variables. The íntegration then would have generated· the inverse
of the determioa.nt function wbicb boa singula.rities for ali zeros in g of the determinant.
These singula.rities would have yielded essential singularlties in the coupling constant
after integration. Finally let us note tbat thís difference, determinant versus inveTSe de-
terminant, ís respoosíble for the minus sigo for eacb fermion loop in perturbation theory,
which allow• for ea.ncellations.

41.9.!! Evalual:ion of lhe fermion detenninant for larye field$

As an eJ<ercise we first solve a similar prohlem in whicb, however, some sligbt additi(mal
complications due to the spin structure are absent.
The F'redhnlm determinant of a Schrõdinger apernl<rr for iarge potentiols. Let us eval-
uate, in the límit of la.rge (smooth) potentials V(x), the Ftedholm determinant of the
Scbrõdinger operator:

(41.18)

On intuitive grounds we expect the determinant to oooverge toWards a local functional.

Tbe límit can be obtained for instance by tbe following method: We rewríte the deter-
minant using the identity

(41.19)
41.3 PeTturoativn Theory at Large OrrieTS and Jnstantons 877

applied tú the case:

H= -1:;. + 1-1 2 + V(:z). (41.20)

When jV(:z}l becomes large, the integral over tis dominated by the small t region. The
evaluation for t small of the evolution operator e-<A, corresponding tú the Schrõdinger
operator A, is a problem we have faced in Section 2.2, in the construction of the path
integral representatíon. We have seen that for t I>IlloillWé can ro>place in V(z) the operatúr
X by a ronstant. Using directly equatíon (2.14} we obta.i.n

(4J.:!J)

As we have indicated there, this evaluation is valid only if V(x) i• •.t. leaat continuous.
It then follows

trluHH- 1 ~ - - -1- {"" ~f.tx [e-tiV(,.)+,.'l-e-"'']. (41.22)

o (41<)4/> ) 0 tl+d/2

For d < 2 we can ai&J integrate over t and finoilly obta.i.n:

(41.23)

For d ~ 2 we know tbat the quantity (41.18), whicb has the form of" one-loop diagram
in a scalar field theory, has tú be renormalired. For d = 2 we hflve tú add " mass
counterterm. Since V(:z) is large we can neglect ~· We then obtain the evaluation

lnD(V)- ~ {- (4, 1)df 2 f dd:z [rt-d/2) V 412 (x) + r(l- d/2) V(x)l},
~ _....!:_f d2 x V(x) In V(x). (41.24)
4..
In tbe same limit we obtain for d = 3

(41.25)

For d = 4 a. counterterm quadra.tic in V(:z) is required. We then find:

(41.26)

The fermion de!erminanL The same method can be used tú evaluate the fennion
determinant D(9):

D(g) det(~+ M +yg.P(:r))(!Í'+M)- 1 . (41.21)

Using reftection symmetry we first rewrite D{g):

lnD(g) = 1trln[;! + M + ygol>{o:)]l-sl +M + yg.p(x)](-a' +M 2 )- 1 ,

878 Pe..turbatian Theory at Large Orders and ln•tantoru 41.3

(the arder of factoro does not matter in a determinant). Evaiuating the product we find:

We then again use identity (41.19) with now:

H -éP + (M + ../9<1>) 2 + .Jg~</>, (41.28)

Ho -fP +M 2 . (41.29)

For g large and .p(x) smooth enough, we can neglect theterm ..;g~.p in (41.28). Then the
trace over 'r matrices yields a I..ctor N tr 1 and the remaining part of the ca.lculation
ís identica.l to the case of the Schrõdinger equation "~th the potentíal:
2
V(x) = 91Jl (o:) + 2../9M<t>(:r). (41.30)

Substitutíng thís expression into equa.tión (41.23) m obtain the large field behaviour:

(41.31)

41. 9.9 The lm:qe ordeT behaviour

We can now study the essentia.l singula.rity of tbe theory at g O sma.ll from the properties
of the effective loca.! action:

(41.32)

It should be understood that for d even, tbe ne<:essary counterterms are provided to
make the actíon finite. We bave to look for ínstanton so)utíons of the rorresponding field
equatíons. Since tbe particular model ís not interesting in itself, we shall not do ít here.
Let us just assume that we have found a solution and determine the dependence in tbe
eoupliog constant g. We resca.le the lield <I> to fuctoriM the dependence on g in Iront of
the classica.l action:
(41.33)

The classica.l a.ction ca.lcu!ated for a solution taktl!l thus the forro:

(41.34)

lntroducing this IOrm ínto the Cauchy formula we find:

(41.3.5)

Tbe integration y'ields tbe )lltge arder estimate:

(41.36)
41.3 Perturbation Thenry at Larye Orders and Instantons 879

We observe that, as expected, this theory i>; less divergent than a purely booonic fie)d
theory. However for d large the Pauli principie becomes decreasingly elle<;tive Md we
recover the boo;on result (in a cut-offfield thoory}. For d = 2 expression (41.36) becomes:

(41.37)

in agreement with rigorous bounds which imply:

(41.38)

Remarlc. To compare the contribution of bos.m interactjoru; and fermion interactions,

we ha.ve implicitly 8S$umed that. we were dealing only with loopwise expa.nsions. In
such circumstances, the boson contrlbutions a.lways dominate the large arder behavíour.
Il we group the diagrruns in anotber way, the answer can be different. Let us again
consider the theory defined by the action (41.16) in four dimensio!IB. In four dimensíom,
thís theory cannot be renorma.lized · without the a.ddition of a >.~• counterterm. Thus
renorma.lization forces us to introduce a OOSQn self-interaction. But it is co!IB\stent with
renorma.lization to consider >. as being nf arder g 2 • Then ooth interacti<>n terms i[ni?.P
and </1~ give contributio!IB of the same order to the large arder behaviour .

./I.S•.j The case of QED

Of course the potentia.lly most interesting applica.tion of the prevíous ana.IY»is i!; QED,
which satlsfies a.ll previous requirements. One a.dditiona.l complication tben arises. The
fermion integration lea.ds to the determina.nl

(41.39)

To estimate D(e) for large charge e we can use the equa.tions (18.66,18.67) (we U6ume
d even):
(41.40)
In the large e lirnit the last terrn, whicb is of order e, is negl.igi.b)e with respect toD!
whi.cb is of order e2:
lnD(e) ~ !Ndtrln{m2 - D~)- (41.41)
We then use the representation (41.19) to evalu ate the determinant ofthe electrom agPetic
Schrõdinger operator. The pa.th integral representation of e-tA has been given in Section
3.2. The determinati<>n of the large roupling oonstant behaviour is however more subtle
than hefore. The electromagnetic term in tbe path integral depends only on the geometry
oi the loop one integmtes a.long and not <JD the time spent on the loop trajectory (see
equation (3.19)). Therefore the large coupling constant Iimit does not select the short
time contributimJS in the representation (41.19). As a consequence the determinant no
longer generates a local term. A direct calculation of the determinant has not been ·
perf<>rmed. The difficulty seems to be rela.ted to the fact that, due to gauge invariance,
the ga.uge degree <>f freedom of the gauge field cannot be considered as slowly varying.
Jt has theref<>re been cpnjectured, on the basis of studying the determinant for specia.l
gauge fields, that the determin&nt is equiva.lent for large e to
880 PerturbatWn Theory a! Lary• Orders and Instantoru 41.4

Thls result is gauge invariant, as it should, but non-local except in the gauge a"A" = o.
lt agrres for d 2 witb tbe exa.ct result (31.60) obtained from the abelian w>omaly
(C(2) = 1/2..-). For d = 4, tbe esse of physical mterest, C(4) 1/12..-2 • The effective
clas:sica.l field theory tben is scale inw.riant. Arguments related to conformai invatiance
can be used to construd some ansstz for the instsnton solutioru;. Two kind of solutions
have been explored by Ba.!ian et a!., or Bogolmony !Uld Fateev. Thking tbe nrinímal
a.ction so!ntion one obtains an eva.!u11tion of the form:

(41.42)

the expa.nsion pa.rameter being: a = e2 f4..-. lt ls wortbwhile rnentioning that this eval-
ue.tion is probe.bly not ""ry useful as a pra.ctica.! mee.n to predict new orders in QED
for ..,.,.,,a)
reasans. First the theory is not asymptotically free e.nd tbus has a potential
renormalon problem, whicb c.on be underst.ood by inserting Íll 11 Feynman diagrarn the
one-loop oorrected pboton propagu.tor. Second the canrellation coming from the sigo of
fermion loops does not seem to be very effective 11t low ordem. Tberefore a.n alternativa
calcuiBtion, whlcb leails to 11 la.rge order behaviour at a fixed number of fermion loops,
seems to be more useful. PredictiollS of this kind mlllle for diagrams with one fermion
loop, seem to agree well with numerical estimat.es.

41.4 Diwrgent Series, Borel Summabiljty

Aaymptotic serões. Let us oonsider a function /(z), ana.lytic in a sector S:

!Argzj Sa/2, (41.43)

We assume that f{z) has in S the followíng asymptotic expe.nsion:

(41.44)

This meanB tha.t the series (4.1.44) diverges for ali 110n-trivial values of z and that in S
the following bound is satisfied

l
f(z)- t /Fkl
k-0
:S CN+liz!N+t for aJI N, (41.45)

in which:

Though the series (41.44) diverges, it is possible to use it to estima.te the Fnnction f(:z)
for l:zl srna.!J. At lzi fu<ed we can look for a minimum in the bound (41.45) when N
varies. 1f lzl is small enougb, the bound first decreases with N and then, sínce the series
is divergent, fina.lly increases. !f we trunca.te the series a.t the rninimurn, we get tbe best
41.4 Perturbation Theury at Larye Onlers and lnstantons 881

passible estimate of /(z), with a finite error E(z). Let us assume for definiteness that the.
coellicients C N h ave the form:
Cn MA-N(N!) 11 . (41.46)

We can then estimate c(z) explícltly and find:

.:(z) minCnlziN ~ exp [-.B(A/Jzl)' 1"']. (41.47)

{N}

We see that an asymptotic series does not in general define a unique function. lndeed if
we have found one function, we CBII add to it any (unction analytíc in the oector ( 41.43)
and smaller tha.n e (z) in the whole sector. The new function still satiBfies the rnndition
(41.45). However tbere is one sit.uation in which the asymptotic series defines a unique
function. lf the angle o satisfies: " ? ",B, then " cl..ssical theorem about analytíc
functions teUs us that a function analytic in the sector ~U~d bounded by e(z) in the whole
sector vanishes identically. In particular in the special case .B = I, w hieh ís typical for
perturbative expansions, we find
(41.48)
In the marginal case in which tbe series i& asymptotic only in the open interval I Arg zj E
( -7r,Bf2, tr/3/2), additional conditions have to be impozed to prove uniqueneos.
Borel tmnsfonnatíon. We specialize from now on to /3 = 1, since it is tbe most useful
cx.ample, but the generalization to arbítrary {J is stra.ightforwa.rd. Under the condition
(41.48), the function f(~<) is uniquely defined by the series. In addition there then exist
metbods to ":sum" tbe series, which means that one can xeconstruct tbe function from
tbe knowledge of tbe terms of the series. One set of methods is bMed upon the Borel
transformation.
Let us define the Borel transform B,(z) of f(z) by

B1(z) =
"'
Í:B•zk: L"" ff,zk. (41.49)
o (f •

The bound (41.45) and the BStimate (41.46) give usa hound on the COEfficients fk:
IA fklj < M A-•. (41.50)
Thus Bt(z) is analytic at least in a circle of radius A and uniquely defined by the serie3.
Furthermore, in the sense of power series:

(41.51)

As a consequenre of the inequallty (41.48), it can be shown tba.t B 1 (z) is alw analytic
in a oector
(41.52)
and does not increase faster than an exponential in the sector, so that integral (41.51)
converges for ]zl small enough and inside the sector
1Argzj < o/2.
In addition it ca.n be shown that the r.h.s. of equation (41.51) satisfies a bound of type
(41.45 ). Hence this integral representation yíelds tbe unique function whicb has tbe
asymptotic expansion (41.44) in the domam (41.43).
882 Perturbation Theory at Large Order• and Instantoru 41.5

41.5 Largé Order Behaviour and Borel Summability

We have learned that, for a large cla.ss of potentíals in quantum mechanics and for a
number of field theories, the perturbative coefficients !k in the loopwíse expansion of any
physical quantity f(g),
(41.53)

behave at large orders as

(41.54)
Therefore the coeffidents B• of the Borel transform B1 (equation (41.49}) of f(g), behave

(41.55)

This asymptotic estimate tells us that the síngularity of B J(9) closest ro the origin ls
located a.t the point g = Ifa. Consequently, the Borel triiJlBformation does not exist if
the classical &ct.ion A = 1/a is positive. The perturbation series in such theories is not
Borel surrunable. Let us, in the Iíght of this result, discuss the various situa.tions we have
enccuntered:
(i) The field equations have no real instanron solutions. This is iD particular the case
if we have e:xpanded around the uni que absolute minímum of the potential. U romplex
instanron solutions exlst, the oorresponding classical action is non-positive, and the per-
turbative expansion is pre;umably Borel summable. It is only a presumption because
varíous features o! the perturbative e><pansion, invisible a.t large orders, could prevent
Bo~l summa.bility. The perturbatm expa.nsion could contain for instance rontributions
ali of the same sígn, growing fBBter than any exponential of the arder k, but much smaller
than k! (for example v'ki). Then BJ(.z) would grow too rapidly for large argument z
(lnBt(z)- z2 in the example) and the Borel integral would not converge at infinity.
(ii) We have found real instanrons in the theory because we expanded around a rela.tiw
minimum of the potential: the perturbatiw expansion is not Borel summable.
Hawever ín this case we can provide one additional piecll of information useful for
determlning tbe solution: the unstable situation can be considere<! as coming from a
stable situation by analytic continuation. Therefore a pOOlible solution could be to
integrate in tbe Borel transform just above the cut which is on the real positive axis.
As a consequénce, from a real perturbatíve expansion we would obtain a complex result,
but this is ex&ct.!y what we expect. It ís e.asy to wrífy tbat the im&ginary part is for 9
small ex&ct.ly what we haw calculated directly. Actually this is only the solution of the
problem ín the sUilplest case, when no other instaurou singularities cross the contour of
integation in the analytic continuation.
(iü) There are real instanrons connecting degenerate clo.ssical mínima.
The theory is not Borel sumrnab!e. lntegration above ar below the axis yields a complex
mrult for a real qua.ntity. Thís cannot be the correct prescription. The half sum of the
integral above and below is real, hut even ill the simple example of the double well-
potential, one can ,.,.ify numerically, and argue analytically, that it ís not the correct
solution. We shall sbow, in quantum rnechanics, that the additional information needed
tD determine the sum af the perturbative expansion ís provided by the consideration of
many instanton eontdbutions (see Chapter 43). The corresponding problem has not been
solved in field theory examples yet.
41.6 Pertu.,.bation Themy at Larye Orders and Instantons 883

Remarlr:s. We have given 6eld theory exampies of soch a situation in Sections 40.4,
40.5: the tw<>-dimensional CP(N -1) models and four-dimensional SU(2) gauge theory.
In these models real instantons conne<::t degenerate minima of the d8S<ice.l action and
the corresponding classlcal action is P05ltlve. Therefore the coefficient.s of the perturba-
tive expansion rontain a non-Borel summable contribution. This contríbution does not
necessa:rily dominate the large arder behaviour, because, as the exrunple of of the </>~
massless field theory in Cbapter 42 wíll illustrate, when a field theory is classicaJJy scale
inva:riant, th€ perturbative expansion might be dominated by contributions unobto.inable
by semicla.ssícal methods, and related to the UV or IR singu)arities.

41.6 Practical Summatlon Methods

Va:rious pra.ctical summation methods rely upon a Borel transformation.

The Borel transformation reduces the problem of determining the function to tbe
analytic continuation of the Borel transform. The Bor!!l tr!liiBform is gi'•eo by a Taylor
series in a clrcle and an Malytic continuation of the series in a neighbourhood of the real
pOOlitive !iXIs is required. This a.nalytic continuatioo ean be performed by roany methods
!ll'ld the optiroal choice depends somewhat on tbe additional information one po5Sesses
about the fuoction, Let us gíve two examples:
Padó approximants. In the absence of accurate knowledge of the positions of ali singu-
larities of the -Borel tmnsform in the complex plane, one can U!le the Padé approximation,
Froro the series one derives Pad é approximants which are ratlonal functioRB PM / Q"' •at-.
isfying:
Bt(z) = Pu(z) +O (zN+M+I), (41.56}
QN(Z)
wbere PM and QN are polynornials of degrees M and N respectlvely. Ir one knows K
terms of the series, one can construct ali IM, N] Padé approximants with N + M ic K.
Thls roethod is well adapted to meromorphic functions. The main disadvantage of the
method is that for a rather gmeral class of functions Padé apprmdmants are known to
converge only in measure and thus spurious poles may occasionally appear cJoo;e to or on
the real positive &.:>cis.
Even if the Padé approximants converge, this property may lead to instabillties in the
results.
Crmfonnal mapping. li we know the domaio of analyticity of the Borel transform, we
can 6nd a mapping whicb preserves the origin, and roapo; the domain of analyticity onto
a círcle. In tbe trllllSfonned variable the new series converges in the whole domain of
aoalyticity. Let us explain the method on an example.
lfthe Borel transform is analytic in a cut-.plMe, the cut ruoning e.long the real negative
&.:>cis from -oo to -1/a, we cao map the cut-plane onto a circle of radius 1:
v'l + az- I
z ,_. u, u{z) (41.57)
v'l+az+l ·
From the original series for the Borel transform we derive a series in powers of the new
variahle u:
""'fk k
BJ(Z) L; kf% •
00

Bt [z(u)] = LBkuk. (41.58)

o
884 Perturbatíon Theory at Larye Orders and ln.otr<ntcms 41.6

lntroducing this expansion in the Borel transformatíon we obtain " new elCpansion for
f(z),
""
f(z) = LBtlk(z), l41.59)

in which the functions J0 (z) have the integral representation:

(41.60)

lt is possible to stucly the natural domBin of convergence of this new expansion. Using
for u(z) the explicit expression (41.57), '"" can eva.luate Io(z} for k large by steepest.
The ..,ddle point equation is:

(41.61)

which for k large yields:

t ~ k2f3 /(az)lf3. (41.62)
lt follows that I•(z) beh.aves for k Iarge as:

h(z) ~ exp [ -3k213 /(ru:J 1' 3]. (41.63)

Three situat.ions can now arise:

(i} The cOéfficients B• either clecreose or at least do not gow too rapidly,

Then tbe expansion (41.59) converges at least in the region

Rez- 113 >O ..,. IArg( < 3>r /2. (41.64)

This in particular implies that the function f(z} must be analytic in the oorresponding
region w!Jicb contruns a part of the serood sheet.
{ii) The éoofficients behave like

Bk ~ exp ( ck 213 ) for k large. (4.1.65)

The domBin of canvergence is then:

(41.66)

Thís condition implies analyticity in a finíte domaln containing a part o f the second sheet
since for \zl small the r.b.s. is negligible.
(iü) Tbe coefficients Bk gow faster than exp(ck2 13 ). Tlris is quite possible since the
only constraint on the coefficients Bt ls that the series ( 41.58) has a todius of convergence
l. Ft>r instance the coefficíents B• could grow Jike exp( cJ<</ 5 ). In such a situa.tion the new
series is also divergent. Such a sitm•tion arises when the singularities on the boundary of
41.6 Perturlmli<>n Theury C1( Large Ordeno and Jn.stantons 885

the dornain of ll.llalyticity are too strong. One should map a smaller part of the domain
of analyticity onto a cirde.
Application to the rolculation of criticai eo;pon<nts. In the summation method ba.ed
on Borel transformation and mapping it is easy to incorporate the ínformation coming
from the large arder behaviour analysis. This is one reason why jt ha.s been used quite
systematkally in the fra.mework of the <1;4 field theoty to calculate critical exponents
and other universal quantities. Jn Chapter 28 we have given values for vsrious criticai
exponents, obtained by app)yíng variants of the Borel summation method to the knawn
terms of the perturbatíve expansíon, i.e. síx successive terms in lixed dimension 3 ansl
up to arder .:5 for the o:-expansion.
Let us now summarize the irúormation available for in tbe </>4 field theoty that justifies
tbe use of this summation method.
(i) The Borel sumrnability of perturbatíon theory in .P~ and <,b~ has been rigorously
establisbed.
(ü) Tbe large arder behavíour has been detennined in ali cases and compares favourably
with the first terms of the series available (see Section 41.2).
(lii) SinCE ali known instrurton oolutions in tbe <,b4 tbeory give negative octions, it is
plausíble that the Borel transform is a.Ju!.lytic in a cut plane, the location and nature of
the singulo.ríty dosest to the orígin being given by the large order estimates.
Consequentiy the methods based upon a Borel transformation and a conformai map-
ping of the cut-plane onto a circle, have appea.ood as ideal candidates to sum tbe pertur-
bation reries and the E-expansion.
For completeness let us finally give at least one exarnple of a summation method not
ha.sed on a Borel tran.sforma.tion.
Onter dependent mappings (ODM). The ODM metbod requires, to be applícable, some
kllowledge of tbe analyticity properties of the function itself. As we have dis<;ussed, the
series diverge beca.use the function bal> singularities accumula.ting at the orlgin. However
the strengths of the síngulo.rities have to decrease fast enougb for tbe function to have a
series expanslon. In the examples we have met, tbe discontinuity ofthe function decreases
exponentially near tbe origin. The idea is then to pretend thBt tbe function is a.nalytic,
in addition to its true domain of analytícity, in a small círcle centred at the origin of
adjusta.ble radius p and to map this extended domain onto a drcle centred at tbe origin,
keeping the origín lixed. lf the function would really be analytic in such a doroain, the
expansíon in the transformed variable would converge in the whole domain of analyticity
and our problem would be oolved. Since tbe original seríes is in fact only asyrnptotic,
the series in the transformed vo.riable is also asymptotic. However, a.s a result of thís
transformatíoo, tbe coefficíents of tbe new series now depend on an adjustable pa.rameter
p.
Let us assume for instance that f(z) ís analytic in a cut plane. We then use the
mapPing
2
z 4puj (1 - u) • (41.67)
The transformed seriéS has the forro

f(: (u))
"" (p) u\
L}\ (41.68)
o
in which the coeffidents }\ (p) llie polynomials of degree k in the parameter p. Io more
general sit uatioru;, one can ofteD use a mapping of the form
ph(u), h(u} = O(u). (41.69)
886 Perruroatíon Theory at Laf)Je Orders and Jnstantans 41.6

The ktb arder approximation is obtained by truncating the series at order k, õnd choosing
p as one of the zems o( the polynomia.l Pk (p). The zerO ca.nnot a.ctually be chooen
arbitrarily, but roughly speaking must be the zero of largest modulus f~r which the
derivatíve P~ (p) ÍB small. The ideJJ behind the method is the following: wit~ the original
series the best approximation ís obta.ined by tnmcating the series a.t z fixe<\ at a.n arder
dependent on z sucb the modulus of the last term ta.ken into account is ~inimal. By
introducing BD additiona.l pe.rameter we haw modlfied the situation. W.-, fix fust the
order of truncation $.11d then try to a.djust the parameter p in sucb a W!'ly that, at z aga.in
fixed, the last tenn taken intD account is minima.l.
The lrth arder apprmcimant has the fonn
k
{/(z)h L F\ (Pk) [u(z)t, 1\(Pk) =O. (41.70)
1~o

lt ciUl be show:n under certa.in conditions thst if the terms of the original "'!rles grow Iike

(41.71)

then the sequence Pk decreases like

(41.72)
Sucb a method has been successfully applied to test probleJilS like the Qnart;c anharrnonic
asdllator with a me.pping:
z = puj(l
and to ane physical example, the bydragen atom in a strong magnetic field·.

Bibliographical Notes
The cannectian between barrier penetratian eilects in tbe semiclrussical limit and Jarge
bebllVíour in perturbation theory was first rea.lized by
F.J. ~n. Phys. Rev. 85 (1952) 631; C.M. Bender and T.T. Wu, .f>hys. &v. 184
(1969) 1231; D7 (1973) 1620; Phys. Rev. Lett. 37 (1976) 117; T. Ban!ag, C.M. Bender
iUld T.T. Wu, Phys. Rev. 08 (1973) 3346; T. Banks and C.M. Bender, Phys. Rev. DS
(1973) 3300;
These authars ealcule.te the barrier penetratian coefficientS in Quantum t..techsnics using
standa.rd WKB metbods. Tbe idea of using instead patlt integra.ls can be fonnd in 8D
early paper
C.S. Lam, NUOIJO Cimento, 55A (1968) 258;
however the fieJd theory calculatian (the ,p~ theory) is dueto Lipatov ~ady quot.ed
L.N. Lipatov, JETP Lett. 24 (1976) 157, Sov. Phys. JETP 44 (1976) ICJs5, JETP Lett.
25 (1977) 104, Sov. Phys. JETP 45 (1977) 216.
The metbod has been generalized in
E. Brézin, J.C. Le Guillau and J. Zinn-Justin, Phys. ]lev. D15 (1977) 1558,
and tben applíed to many theories like the a.belian gauge field coupled t, scala.rs
C. ltzykson, G. Paris! and J .-B. Zuber, Phys. Reo. Lett. 38 (1977) 306; 1\..P. Bukbvnstov
and L.N. Lipatov, Zh. Ek3p. Teor. Fú. 73 (1977} 1658;
to tbe ,P3 lield tbeary witb Potts symmetry
A. Houghton, J.S. Reeve and D.J. We.llace, Phys. Reli· Bl7 (1978) 29~6;
41.6 Perturbation Theory at La.rge Orders ornllnstantons 887

theories wíth degenerate minilllà (..,., Chapter 43)

E. Brézin, G. Parisí and J. Zinn-Justin, Phys. Rev. D15 (1977) 408;
non-abelie.n gauge theories
E.B. Bogomolny and V.A. Fateyev, Phys. Lett. 71B (1977) 93; L.N. Lipatov, A.P.
Bukhvostov andE.!. Malkov, Phys. Rev. DI9 (1979) 2974.
The idea of evaluatíng tbe large order bebaviour by lirst calculating tbe discontinuity of
functions ís due to
G. Parisi, Phys. Lett. 66B {1977) 167; E.B. Bogomolny, Phys. Lett. 67B (197'r) 193.
Let us aiso q uote the review artide
E.B. Bogomolny, V.A. Fateyev and L.N. Lipatov, Sov. Sei. Rev., I.M. Kha.la.tnikov ed.,
section A: Pbys. Rev. 2 (1980) 247,
and the reprint volume
lArge Onler Behaviour oj Pert<.<rbation TheC>ry, CuM"ent Phvsics vol. 7, J .C. Le Guillou
and J. Zinn-Justin eds., (North-Ho!land, Amsterdam 1990).
Large order in field tbeories with fermiom; has first studied by
G. Parisí, Phys. Lett. 66B (1977) 382.
QED has been considered in
C. ltzykson, G. Parisí a.nd J.-B. Zuber, Phys. Rev. D16 (1977) 996; R. Balian, C.
ltzykson, J.-B. Zuber and G. Parú;i, Phys. Rev. DI7 (1978) 1041; E.B. Bogomoluy and
V.A. Fateyev, Phys. Lett. 76B (1978) 210; E.B. Bogoroolny and Yu A. Kubyshin, Yad.
Fiz. 34 (1981) 1535; 35 (1982) 202 (Sov. J. Nucl. Phys. 34 (1981) 853; 35 (1982) 114).
The Yulmwa intera.ction in two dimensiom; has been discussed in
M.P. Fry, Phys. Lett. SOB (1978) 65.
The mat.hematical discussion of divergent series c!lll he found in
E. Borel, Leçon sur les Séries Divergentes, 2nd edn (Gauthier-Viliars, Paris 1928); G.
Hardy, Di11ergent Series {Oxford University Press, Oxford 1949).
For a textbook concerning Padé ~praxima.nts see
G.A. Baker Md P. Gravw-Morris, Encyclapedia oj Ma.thematics and its Application.!
vols. l3 and 14, Gian-Carlo Rota ed. (Addioon-Wesley, New York 1981).
The method of summation of series by Borel transformatíon IUid mapping was proposed
by
J.J. Loeffel, Saclay Report, DPh·T/76/20 unpublished.
The summation method based on order-dependent mappíngs has been studied in
R. Seznec Md J. Zinn.Justin, J. Math. Phys. 20 (1979) 1398; J.C. Le Guil!ou a.nd J.
Zinn-JIIStin, Ann. Phys. {NY} 141 (1983) 57,
and proofs of convergence in special cases given in
R. Cuida, K. Konisi.Ji a.nd H. Suzuki, Ann. Phys. {NY} 241 {1995) 152.
Tbe Borel summa.bility of the ~ 4 ín two and three dimensions has been estab!ished in
J.P. Eckmann, J. Magnen and R. Sénéor, Commun. Mnth. Phys. 39 (1975) 251; J.S.
Feldman and K. Osterwa.lder, Ann. Phys. (NY) 97 (1976) 80; J. Magnen and R. Sénéor,
Commun. Ma.th. Phys. 56 (1977) 237; J .•P. Eckmann a.nd H. Epstein, Commun. Math.
Phys. 68 (1979) 245.
For difi'erential approximants see
A.J. Guttmann and G.S. Joyce, J. Phys. A: Math. Gen. 5 (1972) 181; M.E. Fisher
and J.H. Cben, Phll.se 'lm118itions, Cargêse 1980, vol. B72, M. Lévy, J.C. Le Guillou
and J. Zinn-Justin eds. (P!enum, New York 1982),
and reference5 tberein.
888 Perturbation Theory oi Large Order$ and J~tanto'l.S A41.2

APPENDIX 41

A41.1 Large Order Behaviour for Simple lntegrals

Consider tbe integral of Cbapter 37:

1
l(g)=-
ffi
1+"' [ (:r2- +gz4)]
-oo
exp- -
'l 4
dz. (A41.1)

We ca.n expand I (9) in power series,

(A41.2)

The coeflicient.s lk count tbe number of vacuum Feynman diagrams with tbe proper
""'ights in a q,• field theory. We have argued in Cha.pter 37 tha.t the imaginary part of
I(g) for g nega.tive !Lild small WBS dominàted by the non-trivia.l saddle points

x2 =-l/g. (A4l.3)

Tbe contribution o( the saddle points yields

Im/(g) ~ 2-1/2 el/4o. (A41.4)

g-0-

Tberefore 1• behaveo for k la.rge a.s:

(A41.5)

Thís result leads to tbe following interpretation of tbe large arder behaviour formulue
obtained in tbe Cbapter 41: in the case of the anb!lllllonic osci.llator o.nd the .p4 field
theory, tbe number of Feynmàn diagrams .ís oi the arder of 4~k! for k la.rge and a. typical
diagram beha.ves at la.rge ordel'li as (4A)-k where k is tbe arder but alro, up to an additive
constant, the number of loops and A the cla.ssien.l action.

A41.2 Nou.Loopwise ExpBD.SiOI15

Althongh we have discussed la.rge order bebaviour o:stimatcs only for Joopwise expansions,
it ís stra.igbtfo,...ard to generalize tbe analysis for perturbation expa.nsions in àifferent
pa.rometers. For example Jet us consider the acti.on S (q},

(A41.6)

in whicb V(q) is a polynomial interaction. If we study the large arder behaviour of tbe
exp!I.IISion in powers of .l., it is ea.sy to verify tha.t """ need only to consider tbe insta.nton
solutions of o.n action in whicb V(q) ha.s been repla<:ed by its tenn of higho:st degree. Let
us assume that
(A41.7)
A41.3 Per!U'rbation Thet>ry at Larg• Orders and Jnstant"ns 889

Then •imply by rescaling, we see that the dassícal solution q<(t) ha.s the form
g<(t) =). -1/(21"-2) f(t)' (A4I.8)
and the terrn of degree n in V (q) will give a contríbution to the classical action propor-
tional to ).l-n/(>N -2). We verífy that tbe term of highest degree gíves indeed tbe Ja.rgest
oontribution to the action for À smal!. The saddle point in >. in the dispersion rela.tíon
for la.rge order k is of the order
.\-k-(N-1).

Thus the term of degree n in the potmtial generates a. fa.::tor of the form
exp [enk'/(,.+2-2N)] ,

which is relevant, at leading order, only for n <:: 2N- 2.

A4L3 Linear Differential Approximants

Padé approximants provide the simplest example of a general class of approxímlll1ts,

which are obtained as solutions to equations (algebralc or di1ferential) with polynomial
coefficients. These polynomials are cbósl::n to be tbe polynomials of lowest degree far
which the solution of the equa.tion has the """"' power series expansion up to a given
order as the function one wants to a.pproximate. To be more ooncrete let "" gíw the
exa.mple of the linear dífferential approximsnts.
Let /(z) be a function for whlch we know a power serie> expansion. We can construct
a.pproximants [{z) to this function by looking for oolutions of the differential equation:

L P,.(<) (a)"
N
a.. f(.t) = R(z), (A4L9)
n-o
in which the polynomials Pn(z) a.nd R(z) form a set of polynomials of lowest poB5ible
total degree chosen such that
(A41.JD)
In tbe generic situation the degrees \P.. ) and (R) of the polynomials P,. and R satisfy:
N
L[P,.]+[R] =k. (A41.11)

The advantage of these kinds o[ a.pproximants is that they are extremely llexible. lt
;.. possible to """ a lot of additional information one possesses a.bout the function by
imposing additional constralnts on the polynomials P,. and R.
Furthermore whíle Padé approximants generate only a.pproxim!lDts with poles, the
more general approxímants can ha.ve a la.rge class of new singularities. Tbere is of oou""'
a price to pa.y: this approximatíon I!; much more unstAble. lt is necess;,ry to select
among the lar!l" number of approximants one can oonstruct, those for which one has
some reasons to beliew that they are espedally well adapted to tbe original function one
wants to approximate.
Due to the generality of the problem, a systematic study of this cl8ss of approximants
ís la.cking. Notice that the met.hod can be generalized to power oories In more than one
variable. One tben writes partia! di!Ierential equations with polynomíal coefliclents in
sewral varia.bles.
INSTANTONS: THE (1 4 FIELD THEORY IN DIMENSION FOUR

We have shown in Appendix A39, using Sobolev lnequalities, that in dimension four the
massive lield equation has no instanton solution and the relevant insta.nton is a solution
of the massless field equation instead. We therefore lirst study the massless <J; 4 thoory a.nd
cornment at the end ahout the massive theory. The pril:<'! to pa.y for snéh a símptilkation
is the appearance of some suhtle infrared (IR) problems.
Tbe t/>4 field tbeory in dimension four is just renormali2able. We ha~ thus to deal
at Jea.ding arder with the ooe-loop coupling consta.nt renormalization in addition to the
mass renormalízation. Tiús fact, togetber With the scale invaria.nce of the classical theory,
leads to the àppeara.nce of a.n effective coupling consta.nt at the BCale of the ínstaoton,
and therefore the calcula.tion of the contribution of t.he instanton depends on global
raoormalizatJon group properties of the thooty. Finally ""' sha.ll discaver that, as a
consequence of tbeir la.rge moment.a properties, individual diagrama at arder k grow
themselves lilre kl, introducing some new complica6ons in the la.rge arder behavíour
analysis. Similac!y IR síngularities in t.he massless theory also yield contributions oí
order kl, hut witb a different sign.
Note that we bere explain the cal~ulation only for one-component <J;' field theory, but
the extension to the O(N) symmetríc model is simple.

42.1 The Euclidean Equation of Motion. The lnstanton Actlon

The euclidean action of the massless theory <J;4 theory <:ao be written:

(42.1)

and the corresponding-equation of motion reads:

(42.2)

Note tbe unconventional normalization of the coupling consto.nt. To return to the usual
convention one há!! to set g ..... gf6.
We know that the sol ution of mínima! action is spberically syrnmetric, thus we set:

(42.3)

with
r= lz-;>:ol· (42.4)

We then obtain a dilferentlal equation

(42.5)
42.2 lnstantoru: The f/14 Foeid Th~ in Dimen.ri<m F<YitT 891

We nuw use the scale invariance of tbe dassical tbeory (the theory is actually confonnal
invariant. see Appendi"< A42.2). !f <I>( :r:) is the solution to the equation, then '\li( X) is also
a solution with
q'J(:r:) = Àt/J(Àx). (42.6)
This suggests the following change:

f(r) = .,-• h(t), (42.7)

which transfcrms equation (42.5) into:

(42.8)

We recognize tbe equation of motion of the anhermonic osdllator that we have solved in
Cbapter 37:

h..(t} = ± cosh (t to)· (42.9)

Tbe solution 4>c(x) of equatíon (42.2) th~ iE

(42.10a)

(42.10b)

where we have defined À e-to. Tbe rorresponding classical actian S{if>c) is

S(</>c) = -A/9, (42.11)

Wilh the standard normalization of 9 one finds A =

l61r2 •
Because the d.o.ssical tbeory ís scale lnvariMl, tbe instanton solution now depends on a
scale perameter À, in addition to the faur translatíon pBrameters Xo~· We have therefore
t<> introduce five collective coordinates to calculate the instMton contributíon.

42.2 The Instanton Contributioo st Leading Order

The geneml strotegy. The second derívative M(x, x') of the action at the saddle point is

tPS [ (42.12)
M(:r:,:r'} = Ó</Jc(x)liof>c (:r') = -t>

To find tbe eigenvalues tbe operator of M, one has to solve a 4-dimensional Schrodínger
equation with a spherically symtnetric potential. We immediately note a.t this stage
two serious problems. Til€ operator M ha:s of oourse five eigenve<::tors, ô~,P.(z) and
(d jd>.) .P.(x), with eigeuvalue zero. The la.st of these eigenvectoTS ís not norma.lízable
with the natural measure of this problem,

(42.13)
892 Jnstanloru: The <{! 4 Fíeld Throry in Dimension Four 42.2

This ís an IR problem which arises because the theory is massless.

Another difficulty comes from the mass countenerm which has to be added to the
a.ction. lt has the form:
(42.14)

The integral of ql~("') is also IR divergent, and this IR dívergence is expected to cancel
with an IR divergence of detM. Thus we need in general some kind of IR regu!arization.
In the particular case of the dimeJL'Jional regulari2ation, this problem is postponed to
two-loop order.
These problems will be solved in severa! steps. First we realiz.e that we do not need
the eigenva.lues of M but only the determínant det1 MM.J 1 (equations (39.28)). We
can rnultiply M a.nd Mo by the sarne operator. A speciftc choice which makcs full
use of the scale invariance of the cl.,..ical theory, then transforms M into an operator
whose eígenvalucs can be calculated analytícally. Bec9.use the calculations are somewhat
tedious, we here indicate only the various stepo;, without gíving all details.
The tmnsfonnatíon. We extend the transforma.tioo (42.7) to arbitrary fields, ..,ttiog:

q'J(:r) = e-• h(t, n), with t = In 1,.,1, (42.15)

The classical &et.ion then berome;

(42.16)

Tha symbol Jdn rneans lntegration over the angular variables n,

and 1 2 is the square
of the angula.r mcmentum operator with eigenvalues l(i + 2) and degeneracy (i+ 1)2 .
Expression (42.16) can be rewrltten

(42.17)

Tbe integral of hh w.nlshes dueto boundary conditíons.

With the pa.r..metrization

the classical solution (42.10~) transforms into hc(t):

h - ±2J2 (42.18)
h c{t) - e!t-to) -2v. n + e-«-'<l(v 2 + 1) ·
We oote that in these new va.riables translaticms take a complicated form, unlike dila.ta-
tion which simply corre;ponds to a translation o[ the variable t.
The second derivative of the classical action at the saddle point now takes the forro
(for t 0 = Xop = O)

M = 6hc6h,
~2s
=-
(d)
dt
2
6
+L'+ 1 - cosh2 t' (42.19)
4.2.2 Jrutantons: The ,p• Field Tlu!trry in Dimenaion Four 893

The naJ;ural measure associated to this hamiltonian problem is

f dtdfl'

whicb in the original language means

d4 x
f X
,.
This measure is not translation invariant, and thus the jacobian result\ng from the in-
troduction of collective coordinates, and the determinant depend individually on :z: 0 ,..
However the product of the corresponding oontributio"" to the linal r.,.ult should not,
thus we perform the calcuta.tion for :z:o,. = O.

42.2.1 The jacobian

With the new mea.sure di/J 0 jd>. is normaliza.ble:

(4.2.20)

(42.21)

This lea.ds to alirnt factor:

J, (42.22}

The second ja.cobian J~ comes from the collective ooordintltes :z:o,.:

(42.23)

>.4 16 4
!f)(97f. (42.24)

The complete ja.cobian J is thus

(4.2.25)

,/2.2.fl The detem.inant

Using the reoult of equation (37.38) it is possible to calculate the determinant of M for
eacb value i of the angular momentum:

(-d)'
dt
+(1+1)~---. 6
oosb 2 t
(42.25)
894 Instantons: The (/! 4 Field Theary in Dimension Faur 42.2

The determinant is thus:

(42.27)

in whích Mm ís the operator of the corresponding free theory. As we know, this deter-
minant ís UV dh..,rgent and we ha"" to renormaliz,e it. Hovrever let us first calculate
formally the unren~rmalized determinant:
-1 1(1 1)
i;:: 2: det MtM01 = (I+ 2) (i + 3), (42.28)
1
1: lím! det(M, +c) (Mot +•f' = - , (42.29)
1-0E 48
1
~-
12 .
1
Jim det (Mbo + e:)(Mor=o + •)- 1
(42.30)
,-o c
As .,xpected tbe determinant ís negative and we obtain the formal expression

det'MM0 = -
1 l (1)
12 4sx
4

x !!
""[ 1(1~1}
(I+ 2) (I+ 3)
]{!+!)' (42.31)

Renornllllization. In these varíables, the UV divergences appear as divergences of the

ínfinite product on 1. Let U5 thus use in an intermediBte step a ma.ximum value L of
l as a cut-olf. From the general analysis m know the UV divergent part of In det M
is completely contained in the two li.rst terms of the expansion in powers o( We ,g
therefore proceed in the following wa.y: the determinant of the operator M(s),

M(s) = (42.32)

is exactly knawn:

-1 r(l+Ji)r(Ji)
<letiM(s) + zJ!Mo + >:] = r(l + s+ Ji) r{ ,fi_ s) (42.33)

Setting:
s(s+ 1) =&r, (42.34)
ít ís easy to expand In det M( s) in powers of -y. We deduce immediately from this expan-
sion, the expansion up to second arder of In detM in powers of the potential -6/ oosh 2 t
in the repre«mtation (42.27). We then subtroct these two terms ftom In detM as ob-
tained from the representation {42.31). lt is easy to verify that indeed the large L limít
of the subtrected quantity:

lim i2,f3x(48)2n
L [(I )(
+2 1+3)
](1+1}'/2 L
ne-3U+l)
L-+cc !=l (!- 1) !~O

"n ,.-t•{l+l)'lf
1=0 sl+l/;2
2_- _1_
1+1
1
2(1+1)2'
) {42.35)
42.3 lnstantoru: The <{; 4 Fieúl Theory; in Dimensíon Four 895

is finíte. We set:
{42.36)

Takíng ínt<J account the jacobians, the factor {2n }-!f> for eacll collective mode, the facto r
(2i)- 1 anda fru:tar two for the two saddle points, we get a lirst foctor c2 of the form

(42.37)

which w.. write as

A"
-----r:72c" . (42.38)
(-g)
We then have to add to the clasoical action the two terms we have subtracted above to
In det M. However w.. can now write them in the normal space representation, regularízed
as we have regularíred the perturba.tive rorrela.tion functions, and take ínto account
the one-loop counterterms. The first ter,m in the expansion in powers of t/J~ is exactly
cancelled by the mMs rounterterm, as we have alre&Jy discussed. The seoond term in the
expansion, which is the on<>-loop contribution to the four point function, islogarithmlcally
divergent. In next wction we calculate explicitly the finite difference between this term
aod the coupling constant counterterm which cancels the divergence.

42,3 The Coupling Constant Renormalization

The terlll5 we want te calculate involve the renormalized 4-point function. We have to
cboose n renormalizatíon scbeme: we assume therefore that we ha.ve renormalized the
field theory by mínima! subtraction after dimensional regularization. The rt'normali•a-
tíon constants have been calcula.ted in Section 11.6, Notice the dílferent normalizatíon of
the coupling con.stant. Let us wríte the rontribution 682 which we have to add te the ac·
tion, coming from the subtroction oflndetM and the on.,..loop coupling renormaliza.tion
constant:

(42.39)

ín whicb Nd is the usual geometrical foctor:

(42.40)

and d = 4 E- The e:xpression ca.n be rewritten:

68, = -~g 2j d":rd 4 y d•p4 e•Pl>-~) <f;2{x),p2 (y) lim

( / ddq -~11._•_ _ N~)
4 (211") < < (2.-) 0 q2(p- q)>
d-4 E '
{42.41)
in which 11. is the renormalizatíon scale. The integral over q has been performed in Section
9.3 {see equation (9.29)):

(42.42)
896 Instanlons; The <!>4 Fi•ld Theo'11 ín Dimensíon Four 42.3

Let us also jntroduce the Fouriert.rans[orm ofthe function /'(r) (f( r) is given by equation
(42.10a)):

"(p) _l_jd·:x~ {42.43)

(2r.) 4 (l+:rz)''
The solutíon .p,(z) depends on the scale ,\. Rescaling the variabies x, y, &nd p, we can
then write the total expression more explicitly:

6S0 ""'-
9>r
2
2
f. d pv 2 (p) [12 (42.44)

From the definition of v(p) we deduce after a short calculation:

(42.45)

(42.46)

in which 'Y is Euler's constaot: "f = -1/J(l) 0.577215 .... We then obtain:

(42.47)

with:
In c. = 1 - 3ln 2- 3-y. (42.48)

We note that the r.h.s. of equaúon (42.47) now depends on the scale para.meter A. The
interpretation of this resu)t. is the following: the coupling constant renormalization brea.ks
the scale invaria.nce of the classical tbeory, and therefore tbe scale parameter ), remains
in the expressíon. Moreover the term proportional to In À together witb the contribution
from the classícal a.ctíon CM be rewritten as:

(42.49)

in which g (?.) is tbe effective coupling at the scale À, solution of the renormali:<ation
group equatíon,
dg (>.) i3 [g (.1.)], (4UO)
dlnÀ
with
{3(g) = g,9 2 g 2 + O (g') . (42.51)

Thls property is expected. The renormalization of the perturbative expamion renderS

tbe instanton contributíon, before integration over dilatation, finite. Consequently tlús
contrlbution sbou)d satisfy a renormalization group equatíon, and the coupling consta.nt
g cao be present only in the combinatlon g(À), since À fixes tbe scale in the calculatlon
(for more datails- Appendix A42.1).
42.4 Jnstanwns: Tlu. ,p• Field Th""'11 in Dimensíon Four 897

42.4 The lmaginary Part of the n-Point Function

We can now writ.e the complete contributíon to the imaginary part of the n-point function,

where we have set:

It is convenient to Fourier transform expression (42.52), introducíog

u{p) (42.53)

Then, after f~orizing the é-functíon of lllómentum conservation,

Imztn) (p
!, ... ,
p,) ~ Cs
(-g)(n+5)f2
!.""
O
d.\.>."-3" n" u (!:!.) é~';3o!Al
i:J ). .
(42.54)

We CBil express this result on lPI correlation functíons f'Cnl (p1 , ••• , p,):

(42.55)

lt is stra.ightrorward to verify that. ru

(p) goes to a constant ror IPI 5tna.Jl.
I u oontrMt to the super-renormalizable case, becau"" the theory is only renormo.Iiz..
able the final result is not totally explicít, but invo!ves instesd a fino.I int.egration over
dilatations whose convergence is not obvious. Let us now di5Cuse this point.
The small instnntan contríbution. Small ínstantons correspond to À large. For .>. large,
the integral behaves líke:
(42.56)

and therefore we h ave to examine the behaviour of g {À) for À large. From eq uation
{42.51) we see that the theory is asymptotica.Jly because for g negative, i.e. g (..\) goes to
zero for .\ large. Thus perturbation theory is app!icable and we can use the n.pprrndmation
(42.49). The argument rem&iru true even i f we take g slightly complex. Thus the integral
has the form
(42.57)

We oee that the power bebaviour in ). depends explicitly on the coefficient "f the g" term
of the /1(g)-function. Without tbe rontribution coming from g(;\), the integro.I (42.57)
would have a UV divergence similar to the one found in the rorresponding perturba-
tive expansion. Dueto the sdditional power ar). coming from g (.\), only the va.cuum
amplitude Í6 divergent.
The convergence of the dilatation integral is thus better than expected; indeed the
renormafu.ation consta.nts are now thelll9elves given by divergent wries and are com-
plex for g negative. Their imaginary part contribut.es directly to the imaginary part of
898 Jnstantons: The <t>4 Field Theory in Dimensivn Four 42.5

f{n) (p~, ... , Pnl for n ~ 4. In the q,6 field theory in climension 3 for exa.mple, tbese
contribntions cancel the dlvergcnces comíng from the integral over .>.. Here instead the
integrals over À are finite at this arder. This implies in particular that in the mínimal
subtraction scheme tbe imaginary parts of the renormalization conStants vanish at leacl-
ing arder. In another renormalization scheme (fixed momentum subtractíon for example)
these imaginary parts are finíte at lea.ding order.
The large instantan contribution. Let us now examine the oonvergence of t.be .\ integral
for .\ sm.aU. Tbe behaviour of g (.\) is totally unknown. On the other hand, it i• """'Y to
verify tbat the f&Ctors u (Pi f>.) decrease exponentially for .>. small. Thus, if the behaviour
of g (>.) is not roo dramatic, the íntegrals will converge and it will be justifted to replace
g (.>.) by the expansion {42.49). For the vacuum amplitude, this argument does not apply,
and so the result is unknown.
Tbis analysís shows that, although this calculation seems to be a sirnple fonnal exten-
sion of the calculation for lower tlimensions, couplingconstant renormalization introduces
a set of new problems which a.re not ali completely under control. The fact tha.t the the-
ory is massless only makes matters worse. ConBideration of the ma.ssive theory improves
the situatíon in this respect, but the instanton calculation becomes more complicated.

42.1> lnstanton Contribution to Large Order Behavíour

As a hyproduct of the calculation, we get the instanton oontribution to the large arder
behaviour;
(42.58)

Thls yields a result of the form

(42.59)

Aft.er integration we obta.in

(42.60)

Ftom this expre5Sion, it is stra.íghtforwa.rd to derive the l&ge order behaviour of various
renormali•ation group functions in for example the fuced momentum subtraction scheme.
A comparison between ia.rge order behaviour anà .xplícit calculations can be fo<.md in
table 42.1, in tbe cose of the RG 1:)-function.

Thble 42.1
The we,fficient.s 13> of lhe RG 13-funclíon dillided by tM wymptolic estímole, in the oa.oe af
the O{N} symmelric <1>: Jield theory.

k 2 3 4 5
N=l 0.10 0.66 1.08 1.57
N 2 0.06 0.49 0.1!7 1.32
N:3 0.04 0.33 0.66 l.!J9
42.6 Instantons: The <1> 4 Fielcl ThetJTY. in Dimension Four 899

The large order behavíour of Wilson-Fisher ~-expansion, which is important for thê
theory of Criticai Phenornena, ca.n instearl only be gu.,...,d at becauae, a.s discussed
above, the RG functions in the ntinimal 6Uhtraction scheme vanish Bt leading order. A
cakulation of the next arder would be necesso.rv and tbío ha.s vet not been dane. Since
at leadíng order the fixed poínt consta.nt g• (e} i'.: .
u•(t:) ~ (s-rr2 ) J(N + s),
except if for some unknown reason the accident of leading arder persists, the <-expansíon
is likely to involve a fadar (-3/(N + B))kk! multiplíed by ll.Il unknown power of k. ·

42.6 UV and IR Contributlons to Large Order Behaviour

lmplicit in the large arder behaviour calculation is the assumption that the •ingularities
of correlation functiom come entirely, in the neighbourhood of the origin, from barrier
penetratíon effects. lf thi>; assumption il; certainly correct in quantum mechanics, íf there
are very strong indications that it is valid for super-renormali:.able theoriffi, it is much
more questionahle for renormalizable theories, not to mention massless renorma.lizable
theories. Let us lirst di""uss tbe large m~mentum problem and then the IR problem of
massless theories.
UV singularities: renonnalons. If the argum~nts go througb, without modifications,
for the regularired field theory, they become extremely formal for the renoTinAli:.ed theory
in tbe ínfurlte cut-off limit. We ha""' already seen that even in the naíve calculation, non-
trMal questions arise about the global renormalization group properties of the theory.
Direct investigation of the perturbative expa.nsion raises new questions and suggests that
UV singularitieo yield additional contributions to the large order behaviour.
2
Let us consider tbe ( •f>1) field theory in dímerurion four, in which .P ís an N component
vector, and tbe model has an O(N) symmetry,

8 i.PJ"' Jd~x u (â,..p)2 + ~2 .p2 + ~ (.p2)'). (42.61)

A S)'lltematic 1/N expa.nsion allows us to rearrange the perturba tive expansíon by per-
forming partia! summations. At leading arder, the 4-point function is repla.::ed by the
sum of the bubble diagrams,
2g
(42.62)
l+gNl(p)'
where tbe bubble diagra.m is:

l(p) --4
l
(2")
f [ (p
d•q
+q)• +m•] {q2 + m2]
- subtraction. (42.63)

The important point ís that for large momenta I(p) behav... like
1 m
l(p)- ,.2 In
8
iPi, IPI - oo. (42.64)

Therefore the sum of the bubble diagr81llll hBB a síngularity for g small and positive at
momentum
(42.65)
 4
900 Instnntons: The </> Field Theory in Dimension Four 42.6

This is one exn.mple of the fa.mous Land1m gbost. Since the theory ís IR frre, and not
asymptotically free, this singularity occurs for pooitíve values of the coupling consta.nt.
Once tlris surn of bubbles is inserted at next order, for instance in the 2-point functíon,
it produces a. cut for g srnall and positive,

F,(p) =f d•q su btra.ctíons . (42.66)

[(v+ q) 2 + m 2J [1 + Ngl {q)]
After subtr&Ction, and for \q\ la.rge, the íntegrand behaves lilre

F, (p) jql>l ~ 1 + ~ ln ~ + ... (42.67)

8.-2 q

Setting
t = ln(q/m), (42.68)
we tranoforrn equation (42.67) ínto

f oo d
te
-21 1
1 - N gt/(8Tr 2 )'
(42.69)

This ylelds e.n írnaginary contribution to the rorrelatíon functíons for g smo.ll and pos-
itíve of the form e:xp(-161< 2 /Ng). Altemati....,ly, we oould have consideted individual
dí..grams contaíning bubble ínsertions. By eJt]>&nding for instance equation ( 42.66) in
powen; of g, we would have concluded that these diagrarns behave like (N /161f 2 )• k! at
large order k. Tberefore, in rontra..«t to •uper-renormalizahle theori"" in which an indi-
vidual di..gram bebaves like a power in k and the kl CQDleS from the number .of diagra.ms,
here individual diagrarn~; have a kl b<;haviour, witbout the sign oscillations cha.racteristic
of tbe semidassical result.
Further ~gations show that if a non-perturbative contribution exists, it sbould
satisfy the bomogeneous RG equations. Let us for simplicity consider t.he case of a
dimensíonless ratio of correlation functions R (p /m, g) witbout anomalous dimen&ionB,

(42.70)

Tbe RG equ.ation tells us that the function R(p/m,g} is a.ctually a function of only one
variable s(g)p/m, in which s(g) then satislies:

)3 (O) •' (g)"' • (g), (42.71)

whicb afte:r integration yie]ds:

s (g) - exp [f ~ )3dg' ]

(g') . (42. 72)

For g sma.ll, • (g) behaves like:

. N+8
)3 (g) = (3,g2 +Q (gl) ' Wltb fh. «= 8'lf2 ' (42.73)

s (g) "'exp [- ~g) (42.74.)

42.6 In.stantono: Tho .p• Field Theory tn Dimension Four 901

Thís ls exactly the singularity that we obtained from the set of bubbles in the large N
limit.
Sin<:e the correlation function depends only on the mass squared, only s2 (g) can enter
the calculation, and the contribution to the large arder behaviour is of the form

(!32)•
f.
.-2/fJ.o
---dg~ - k!. (42.75)
gl•+• 2

Thls potential contribution has to he compared with the semiclassical result (42.60).
Ali these problems are of course related to the question of the existence of the renor-
malizcd ,P 4 field theory in four dímensíans. lf the theory does not exist, then prohably
the sum af perturbatlon theory is complex for 9 pasitlve, and these singular terrns, SOille-
times called renormalon elfects, are the small coupllng evidence of this situation. More
generrilly, the e:xistence of renormalons sbows that the perturbatlon series is not Borel
summahle and does not define unlque correlation functions.
Finally let us note that, at leading order in the 1/N expansion, for tbe Wilson-
Fisher E-expamrion, and thus also for suitahly defined renormalization group functions,
tbe renorrnalon singularlties cancel. We conjecture on thls basis and on the basis of the
numerica.l evidence present.ed in Chapter 28 that the €"""pansíon is free of renormalon
singula.ritles.

Mas•le&s renormalizable theories, Let w; again illustrate tbe problem with the (,P2 ) 2
field theory in the large N limit. We now work in a massless theory witb fixed cut-olf A.
We evaluate the contribution of tbe small momentum region to the mass renormalizatíon
constant. The bubble diagram ( 42.63) hehaves Uke

1
I{p) ~ a,..21n(A/pJ.

Tbe sum of hubhles yields a oontrihutioo to the mass renormalization proportional to

Expanded in powers of g thls yieids a contribution of arder (-l)k(N/16or 2 )"kt for large
order k. This contribution has the sign osdllatíons of tbe semiclassica.l term. More
gene:rally for ftnite N one finds (-tl./Z)kkl. IR singularities yiald an additional Borel
summahle contrihution to the large arder behaviour.
For massless, but asymptotically "rree tbeories tbe role of the IR and UV regions are
intetcllanged. UV renormalons are expacted yieldin,g addition.a! singularlties to the Borel
transform on tbe real negatlve axis, while IR contrihutions destroy Borel summahility.
When these theories have real instantons like QCD or tbe CP(N -1) model.s, the Borel
transform has also semiclassical singularities on the real positive a>;is.
902 Instantons: The </! 4 HE!d Theory in Dimen.non Four 42.7

42.7 The Massive Theory

We have shown in Appendix A39 that the massive field equations can have no so!ution,
and that tbe minimum of the sction can be obtained from tbe massle>s theory. To study
the massive theory, we thus start from tbe instanton solution of the massless theory,
with its sca1e parameter À. However we notice a dil!iculty: as expla.ined in Section 42.2
the integral of 4>~ is IR divergent. We have thus to modify the field configuration at
\arge distiUlces, by connecting it smoothly to the so!ution of the massive free equa.tion
Witb mass m. Qualitatively speaking we oonsider a oonfiguratíon which up to a distance
R, >.R ::2> 1, mR « 1, is ÀrPo(Àx) and for lx! :> R, is proportional to the free massive
solution. An analogous problem will be discussed in Cbãpter 43 in the case of multi-
instanton configuration. Altbough the theory is no longer sca.le invaxiant, À bas to be
kept as a collectíve coordinate. The mass term then acts as an IR cuwff, and restrkt
the domain of integration in À to va.lu"" large with respect to m. Tbe classical action
has the fonn

-91(8""' À)
2
Sm(4>.) "" 3 + 81r2 );2
m In;;; for À::2> m, (42.76}

where the In m term is directly related to the initial IR divergence of the cp 2 integral.
Tbe remaining part of the calcul!ltion closely follcrws the calculatiOil for the mru;sless
case and the reader is referred to tbe literature for deto.ib;.
In tbe massless theory tbe inst&nton contrlbution to the vacuum energy couid not
be eval.uated without some knowledge of the non-perturbative IR beh!lviour of tbe {3-
function. In the masslw theory the problem is solved because the À integral is cut at a
scale m/ ..;::::õ. For correlation functions the integral will be cut by the largest between
momenta and m/ R. Tbis implies that the limits m ..... O and g _,O do not commute.
In the large order estimate the À integral will be cut at a scale of arder m..;k.

Bibliographica1 Notes
References to the work of Lipatov and its generalization by Brézin, Le Guillou and Zinn-
Justin ha.s already been given in previous chapters. See also
A.J. McKane and D.J. Wal!ace, J. Phys. A: Math. Gen. 11 (1978) 2285,
and for a discussion of the massive theory
Y. flishmán and S. Yankielowicz, Phys. Re~. D19 (1979) 540; l. Affieek, Nucl. Phys.
BHil (1981) 429.
Renormalons have been discussed in
G. Parisi in Cargue Lecturee 1977, vol. B39 (Plenum, New York 1979) anel Phys. Lett.
76B (1978} 65; B. Lautrup, Phys. Lett. 69B (1977) 109;
G. 't Hooft, Erice Lectures 1g77, A. Zichichi ed. (Plenum, New York 1979); P. Olesen,
Ph:gs. LeU. 73B (1978) 327.
Renormalons have IR a.nalogues in massless field theories
G. Parisi, Nucl. Phys. B150 (1979) 163; F. David, NucL Phy5. B209 (1982) 433; B234
(1984) 237; B263 (1986) 637.
 903
42.7 lnstantoTIB: The tf>' Fíeld 7'h.e.orri m Dimensíon Four
Exercises
Generalize the calculations of Cha.pter 42 to the mamless f/?" field theory in the dimension
in which ít i.s renormalizable:

S(cf>) = fddx [~ (& rJ;)2 + __

2
"
9__ q,>n] .
2(n- 1} 2

The eigenvalues of L 2 in d dimensíons are 1(1 + d- 2) with a degeneracy 61(d):

We also recai! that if M (t 1,t 2 ) ís ofthe form:

in the nota.tíon of equation (37.38) w.: have:

(42.77)

with
" v'f+i:Jw. (42.78)
 Jnstantons: Tlle <P" Fidd Theory in Dimension Four A42.1
904

APPENDIX 42

A4'2.1 lnstantons and RG Equations

The instanton contribution to the lPI n-poínt functíon ca.n be written as:

( ) {"" d),
Jmr n (p,;Jl,g) fo >:F(n) (p;;!',g,>.), (A42.1)

in whicb Jl represents the subtraction •cale, and ;. the dila.tatio!l parameter. The ooun-
terterms wlúch renorma.!ize the perturbative expan.síon, also render F(n) finite for reasous
we have already exple.íned. Therefore p(n) satísfies a RG equation:

a a O, (A42.2)
[1'-+Jl(g)-
fJ!' 8g

wbich, integra.ted by the metbod of cbara.deristíos, yields:

(A42.3)

with the definítiom:

r(•} dg'
lur= J. P(g')' (A42.4)
, _ r<•) "'g')
nZ- }, Jl(g')'
The coupling constant g (r) is the effoctive coupling constant at the scale ..-.
Ordina.ry dimensional coru;ideratlons now tell \lS that:

p(n) (p,;Jl;g;À) (A42.5)

Applled to the r.h.s. of equatíon (A42.3), thi• identíty yields:

(A42.6)

The cboíce:

Iinally lead5 to the relation:

(A42.7)
A42.2 lnJtanions: The q,< Field Theo"1J in Oimension Four 905

A42 .2 Conformai ln'YBriance

The scale invariance of the classical ~i lield theory b85 allowed us to obtain ao aoalytic
instanton solution. Moreover by introducing the special coordinates {t, n~) we ha"" been
able to use tbe results obtaíned for the anhannonic oscillator Íll Cbapt€r 37, a.nd calcu-
late explicltly the instanton contribution at ]eading arder. Actua.lly the scale invariant
classica.l ~~ throry is also conformai ínvaríant (see Appendix AI3.5). Thís property,
whicb also holds for other scale invariant field tbeories like gauge theories, can be used
more directly to calculate the instanton contríbution. Tbe conformai group is i.somorphíc •
to S0(5,l). It ls expected that the mirúmal acti<>n solution wi!l be invariant under a.
maximal compe.ct subgroup af SO(fl, 1), i.e. 80(5). lt is then convenient to perfonn a
stereagraphíc mapping af R4 <>ata tbe spbere S4 to simp]ify the SO(S) transformatiaDS.
One sets:
l - x2
"" 2x" (' = ')':i:X2 ' (A42.S)
' = ')':i:X2 '
so that
(A42.9)

lt is alsc useful to introduce a lield which has simple tra.ru;formation properties under
80(5). In the <P: theory, the conformai transfarma.tian prapertles af tbe Q>-lield lead to
set:
(A42.10)

Let us rewrite l.he clBSSical ~>etíon (42.1) in terms of tbese new vatiables. We perform
the transfarmations in two steps: finrt we keep l.he variables x", but naw considered as
coordina""' on S4 , and perform only the substitution (A42.JO). The metric g,.., an 8 4 in
the coordínates x" \s:
6,.,
11••""4---,. (A42.11)
(1 + x2 )
The invariant m<'asure on the sphere irrvolveo the square raot of the determinant of the
metr:ic g (see Section 22.5):
16
y'detg = - - 2-•. (A42.12)
(1 +x )
Finally, after an integration by parts the kinetic term ca.n be rewtitten:

(A42.13)

The classícal e.ctíon then reads:

(A42.14)

In thís rovariant form l.he change of caordinates (A42.8) is straightfonvard and ha.rdly
necessa.ry. One solution of minimal a.ction is a canstant:

w• = -1/Bu. (A42.15)
906 ]nstantoru: The di• Field Theory in Dimension Four A42.2

The da.ssical actlon is proportíonal to the surfa.ce of s.

which ls 81r2 f3. The opetator M 4
second derivative of the actíon is given by (see Sectíon 3.4):

(A42.16)

in which L ;,; the angular momentum in 5 space dimensions. The eígenvu.lues of L 2 are
!(! + 3) wíth à degeneracy li1:

6 =
1
! (21 + 3)r(i + 3) (A42.17)
6 r(!+ 1) ·

The form of M shows thàt it has O a.s eígenvll.l.ue, corresponding to l 1, with degeneracy
5, in agreement with the eonsiderations of Chapter 42. We '""""' up to the reader, as an
exercise, to verify the other resulto of Chapter 42.
3 MUlTI-INSTANTONS IN QUANTUM MECHANICS

A linear combination of instanton solutions is not a solution of the equatiollll of motion

smce these eqt.ations are non-lioear. ~r, in the limit in which ali instMtoos are
largely separated, such o. configuration rendera the actioo almost stationary because each-
inatanton solution differn from a constant solution only by exponentially small correct;ons
at !ruge distances (in a field theory this is only true ii the theory is massive). We here
examine in the context of Quantum Mechanícs the signilicance of such multi-ínstanton
quasi-soluiions. The generalization to Fffild Theory, howev<!T, is non-trivial a.nd has sti!l
to be worked out.
We have found in preceding chaptero severa! situations in which we can expect multi-
instantons to play a role. In the case in which instantons are found, the ground state
energy Eo tal<$ the form:
+
E, = ~O) ~l) + ... '
in which EÓ ) and E~ 1 ) are the perturhative and one-instanton contributions respectively,
0

a.nd the dots represent possible multi-instanton contributioos. The ground state energy
has been derived from a semidassical calculatíon at {3 large of the partition function
wlúch then has the form

tre-tlH ~ .,-tJEo ~ e-tJE~o} f (-~)" (EÕIJ) n • (43.1)

n=O
Thu5 tbe existence of IUl on<>-lnsta.nton contribution to the energy implies the existence
of n-instanton contributions to the partition function proportional to /3".
When we have evaluated tr e-tlH at finite fJ we alw&ys have kept only the solution
which descrihes the classlcal trajectory once. We have argued that the other solutions,
in which the tra.jectory is described n-times, have in the large fJ límit an actíon n-times
larger, and therefore give subleading contributions to the path integral. In the infiníte
{3 limit these configura.tions have the properties we expect from o. n-instanton. However,
there is a subtlety: mtively we expect these configurations to give a contribution of order
(3 because a classícal solution depends only on one time parameter, in contrast with the
n 1h t.erm in (43.1) wlúch is of order fi".
Another example is provided by the estimation of the large order behaviour of pertur-
bation theory for potentials with degenerat.e mínima. We have soon that wben we start
from a situation in which the mmima are alrnost degenerate, Wl! ohtaln in the degenerate
limit a typical two-instanton contribution but with an infinite multiplicative coefilcjent.
Tlús divergence has the following interpretatíon: In the degenerate llmit the cla.ssical so-
lution decoro poses into the superpooition of an instiUlton and an a.nti-ill5tarrton inli.nltely
separated and thus fluct uatioru, which tend to change the distance be-tween the insta.nton
and the anti-instanton induce a vanishingly smaU vruiation of the action. Jt follows tha.t,
to properly study the limit, one has to introduce a secolld collective coordinat.e which
describes these fluctuations, although there is no corresponding symmetry of the action.
lt can then also be unden;tood where, in the lirst example,. the factor fi" comes from. Al-
though a given classical trajectory can only generate a factor fJ, tbese new conligurations
depend on n independent rnllecti\>e coordinates cwer wbich oDe has to integrnte.
90S Multt-lnstantoll$ ín Quantum M~hanícs 43.1

'Jb summarire: we know tbat n-instanton contributions do exist. However these con-
tributions do not in general correspond to solutions of the cJassícal equa.tion of motion.
They correspond t" conliguratíons of Jargely separated instantons connected in a way
which we shall discu.s, which become solutions of the equatioo of motion only asymptot-
ically, in the limit of infinite 5epatation. Tbeae configurations depend on n times more
collectiV<l coordinat.es than the one-instanton configuratíon.
1n Sections 43.1,43.2 we first study eXplicitly two examples wbich we have already
considered in Cbapter 40: the double-weil potential and tbe periodic c.osine potentíal.
We then discuss general potentia.ls with degenerate minima. We a.lso calculate the large
order beba.viour in the case of the O(v) symmetric anba.rmonic oscillator. Finally, from
the results obtalned for the many instanton contribulions, we !ire !ed to conjecture the
exa.ct form ar the semic!BS'!.ical expa.nsion for potentia.ls witb degenerate mínima.
Tbe a.ppendix conta.ins in particull't some remarks about constra.ined ínsta.nton solu-
tions B.lld a simple example of a non-Borel summable expansion.

43.1 The Double- Well Potentlal

We fust consider the hami!tonlB.Il of the double-well potential, already di.scussed in Sec-
tion 40.I:
2
H=--I ( -d ) 1
+-V(q.Jg) (43.2)
2 dq g •
We have 8een that, in the inlinite {J limít, the instanton solutions are (equat!on (40. 10)):

1
'l±(t) ,;gf(=F(t- to)), (43.3)

/(t) = 1/ (1 +e') = I- f(-t), (43.4)

where the constant ll) cbaracterizes the instanton position .

.j3.J.l The two-in.stanton configurotion

We firnt COIIBUuct the tw~instanton configutflt.ion (really an insta.nton-anti-instanton).
We look for a configuration depending on 8Jl additíonal time parameter, the separation
between instantoliB, which in the limlt of ínlinite sepaution decomposes into two insta.n-
tons, and whlch for large separations minimizes the variatíon of the e.ction (figure 43.1).
For this purpose we could introduce a constra.int in the path integral lixing the separa-
tion between instantons, and solve the equation o f mot.ioll with a Lagrange multíplier for
the constrwnt (for details see Appendices A37.2 and A43.3). We use instead a method
whích, at Jeast at leading arder, is simpler and shows that the result ís universal.

Fig. 43.1 The two-instanton configuration.

43.1 Multi-Instantaru in Quantum Mechanics

We coruider 1< configuratíon qc(t) which;, the sum of instantons separated by a dístance
IJ, up to an addltíve constant adjusted in such a way as to sa.tisfy the boundary condítíoll!i.
It is converllent to introduce some notation:

u(t) = /(! fi/2), (43.5)

_y(t) !(-t- 6/2)' (43.6)
v(t) = 1 - _y(t) = u(t + 8). (43.7)

where /(t) ís the functíon (43.4). We thEn take

1 I
q.(t) = yg (u + y- 1) = yg (u- v). (43.8)

(A,gein time translation by a constant t 0 generates e. aet of degenerate configuratiom;).

This path has the following properties: It is contínuous e.nd differentiable 8Jid wben 8
is large it díffern, near eacll instanton, from the instanton solution only by e>q>onen-
tially small tetlnll of order e- 8 • Although the calculation of the corrnsponding oction ís
straíghtforward, let us perform it step wise to show tbe generality o( tbe previous ansatz.
Tbe action of the path ( 43.8) is •

S(q.) f dt [!<I!+ ~V(q.v'Y)]

= 2>< _!_ +!jdt[i!ú+V!1!+u -1)- V(!!.) -V(u)]. (43.9)
6g g
The parity of Qc allows to restrict. the í11tegration to the region t > O, where v is 5IIlall.
After an integre.t.ion by parts of the term úi! -ilú we find

1 2 { v(O)ú(O)+
S(q.)=lg+g l+oo dt[!!íi+V(u-v)-V(u)-V(v)).
} (43.10)
0

We now expand in powers of v, 8Dd take into ru;count the equation of motion for u. The
leading terms will be of order e- 8 and we thus stop at order v 2 • We obta.in:

S(qc) ~ + ~v(O)ú.(O)+ ~{f"" dt[~v2 V"(u)-V(v)] }· (43.11)

The function t1 decreases exponentially away from the orlgin so the main contributions
to the integral come from tbe neighbourhood of t = O, where V"(u) ~ 1. V(u) ca.n be
replru:ed at leading arder by !v2 • Therefore the int<pal. canoel and we are left with the
integrated terrn:
v(O)u(O)- -e-•,
and thus
{43.12)

lt will become clearer }ater wby we need the classical action only up to ord<>r .-e. It is
convenient in the calcu!!ltion to keep {3 large but finite. Symmetry between e and {3- 8
implíes
(43.13)
910 Mu!tí~Instantons in Quantum Meehanics 43.1

As a check, we can co.lcu!ate the extremuJn of S (q) at {3 fixed a.nd we obto.ln:

e.= {3/2, ,., s (q.) = g- 1 [~ -4 .-~12 +O(e-11)] _

In Chapter 40, for the same hamíltonian, we haw found for f3 large {equation (40.12}):
S(q<) g- 1 [~- z.,-~ +O(e-2P)]. {43.14)

Both results are consistent. Indeed to compare them we have to replace {3 by {3/2 in
equation (43.14) and mnltíply the action by a factor 2, since the a.ction corresponds to a
trajectory described twíce in the total time {3.
The wriation of lhe action. We now shtJW tha.t ifwe infinlteoimally (for (J large) modify
the ronfiguration to further decrease the variation of the action, the change r(t) of the
path will be of arder .,-o
a.nd the variation of tbe a.ction of arder e-•• at least. Setting
q{t) = q.(t) + r(t), (43.15)
we find, expanding the action up to second arder in r(t),

S (qc +r)= S(q.) + j [<;i.(t)r(t) +~V' {q.(t)Jg)r(t)] dt

+~ j dt [f 2
(t) +V" (q,Vu)r2 (t)] +o (!r(t)J
3
). (43.16)

In tbe term linear in r(t), we integrate by parts f(t), in arder to use the pl'Operty tbat
q,(t) approximately satisfies the equation of mot.ion. In the term quadratic in r(t}, we
rep)ace V" by one, since we expect r(t) to be large only far from the instantom. We
then verify that the term linea:r in T ~ of arder e- 8 while the quadratic term is of order
L Shiftlng r to e)iminate the linear term would then give a contribution' of order e- 29
and thus negligible at the arder we consider.

4!1.1.2 1"he
n~iMtantan c.onfiguration

We now consíder a succession of n instantons, separated by times 8,,

r::.~ o, f3. (43.17)
At leading order we need consider "interactions" only between nearest neighboun;. Thls
is an esoentio.l simplify:ing feature of quantum mecha.nics. The dassícal a.ction s. (li;) can
then be directly infetred !Tom expression (43.13}:

s. {8 1) = ~ [~ -
g
2 t
·~]
e-•· +0 ( e-<6•+ 5•) )] (43.18)

Otber interactions are negligible because tbey are of higher order in e- 6 •

Note tbat for n even tbe n~ínstanton configuratíon contributes to tre-~H, whUe for n
odd it oontributes to tr (Pe-PH) (P ís the pa.rity operator). Tben calculatíng
Z, = ! tr [(1 + EP)e-llH], (43.19)
we obtain for < = +1 and E = -1 rontrihutions to the even and odd eigenstate energies
respectívely.
Remark. Since we keep in the ru:tion a.ll terrns of order .,-~ we expect to find the
contributions not only to the two lowest energies but B.loo to ali energies which remaln
finite when g g<'e$ to zero (see the remark after equation (38.62) in Section 38.3).
43.1 Multi-ln&tantons in Quantum Mechanits 911

1,9.1.3 The n-in.otanton contribtdíon

We have cakulated the n-instanton actioo. Let us oow evaluate, at learling arder, the
contribution to the path integral af the neighhourhnad af the dasskal path. Although
the configuration is not a soluüon of the equation of motion, we have constructed it in
such a way that we can neglect the linear ternlS in the gaussian íntegration. The second
derivative of the action a.t the classkal path M (t', t),

ft)
2
M (t', t) [- ( + V"(ygq.(t))] 6(t t'), (43.20)

is a bamíltonian with a potential wbich oonsists of n wells almost identical to the ""'ll
arising in the one-instanton prohlem, and which are largely .eparated. At learling ar-
der tbe conespondmg spectrum is therefore the spect.rum arising in the one-instantan
proh!em n-tímes degenetate. Corrections are exponentially small in the repatation (for
detalls see Appendix A43.1). At tbe same time, by introducing n collective time vari-
ables, we have suppressed n times the zero eigenvalue, and genersted the jooobian of the
one-instMton c""" to the power n. We can therefore write the n-ínstMton oontribution
to Z, (equation (43.19)):

(43.21)

Ali the factors have already been explalned except the factor {3 which comes from the
lntegration over a global time translation, a.nd the factor 1/n whlch arises because the
conJiguration is invariarrt under a cydic permutation of the 8;. The factor e-t1J 2 is the
usual normali.:.ation factor.
Let us define the quMtity À, the "fuga.city" of the instanton gas:

(43.22)

whlch ís half of the one-inst.anton oontribution e.t learling order. lt is convenient to

lntroduce the sum E({3, g) of the lee.ding arder n-instanton oontrihutioUB:
co
E ({3,g) =.-ti/>+ L z!n)({3,g). (43.23)

If we neglect the instMton intemction (the dilute gas spprmcima.tion), we can integrste
aver the &;'s and calculate the sum

(43.24)

We recognize the perturhative a.nd ane--inste.nton contribution, at learling arder, to E,(g),

the ground stste e.nd the first excited state energíes:

E,(g) ~.,-tl"•(l +O(g)).

=! + O(g)- v'lf9 (43.25)
912 Mul!i-Instantons in Quan!um Malwnics 43.1

DiscU,Ssion. Th go beyom:l the one-instant.on appro>úmation we have to take into at-

count the ínteraction between instant.ons. We then face a problem: examining expression
( 43.21 ), we di.scover that the inter&etion be~;Ween instantons is attrnctive. For g small
the domínant contributions to the integral therefore come from configmatíons in which
the ínstantons e.re ciose. For such configuratíons the concept of instanton is no Jonger
vruíd, since such configurations cannot be distinguished from the lluctuations around the
constant or the one-instanton snlution.
We should not be too surprised by this phenomenon. lndeed lhe large arder behaviour
analysis has indicated that the perturbatlve expansion in the case of pot.entials wíth de-
generate núníma is not Borel summable. An ambiguity is expect.ed at the two-instanton
arder. But ií the perturbatlve expansion i.s ambiguous 111 the two-instanton arder, we
should not expect to be able to calculat.e a contribution af tbe same arder or smaller. To
proceed any further we fust. have to make a choice about how we define the sum of per-
lnrbation theory, ll is possible to sbow that tbe perturbat.íon tieries is Borel summable for
g negative by te!atiog it to the perturbat.íve exprmsion ofthe 0(2) anha.rmonic oscillator.
We therefore define lhe sum of the perturbation r;erÍe!i as the analytic ccntinuation of this
Borel sum from g negatíve to g = IYI ± iO. This corresponds in the Borel traosformatíon
to int.egrate above or below the real positíw axis. We theo not.e that for g negative the
interoctíon between instantons is n!ptdsive, and ex:pressioo (43.21) beromes meaningfu!.
We therefore calculate, for g srnall and negative, both the rom of the perturbation se-
ries and the instanton contributions, and perform &D analytic oontinuation to g positive
of ali quantities consist.ently. In the same way the perturbative expá.nsion around each
multi-instanton configuration is aJro non-Borel oummable and hM to be summed by the
same procedure.

,JS. Lj 1'/u! calculation

Let us agoin writ.e the n-instanton contribution:

(43.26)

To factorize the integral over the variables 9;, we replace lhe 6-function by an integral
representation:

6 (to,
,_, fJ) f-ur j'""-" ds exp (-• ({J- t 9•)] , 'l > o
-tco-I'J i=l
(43.27)

In t.errns of the functioo

J(s) ""'l+oo eB-1'•-• d6, (43.28)

integral (43.26) can be rewrítt.en:

Zlnl ~
[Je-~/2 ).n
--.-. -n
j""'-• ds e-ll•!I(s)t, with IA= -2/g. (43.29)
Zt1r _,00 _ 71

By giving to s a small negative real pe.rt, we have ensured the convergence of integral
(43.28). To evaluat.e the integral (43.28) we set:

{43.30)
 Multi-Irutantan.o in Qua.ntum Mechanics 913

rmd the integral becomes:

1 (s) = [ ~ (!'j)' e-• = [,., dtt (~)' .,-• +0 (e-• /!L). (43.31)

for IL positive a.nd !BJ"ge, i. e g ~ O_. U p to an exponentially small correction we thus

obtaln:
J(s) -~L'r( -s) (43,32)
The generating function L:(/l, g) of the leading order multi-instanton contríbutions (43.211)
then becomes:

E(fi,g)

The sum can easi!y be performed:

(43.34)

Weset:
(43.35)
We then integmte fJ .,- B• by parts a.nd obtaín

I:(fJ,g) = -~ 1+'~ dEe-PE rtf (E). (43.36)

2tll" -ioo </J (E)

The asymptotic behav:iour of the r-functíon (given by the Stírling formula) ensures the
convergence of the integral, and moreover the oontour can be deformed to enclose the
poles of the integrand in the half-p!ane R.e(E) > O. Integmtlng we obtain a sutn of
residucs:
E (fJ,g) =L
e-Pliiu 0 (43.37)
N~D

where the energies EN are rolutions of the equation:

r/>(E) = 1- ÀI'E-If>r(l/2- E)= 0. (43.38)

Since >. is small, a zero E of this equation is close to a pole of r(l/2- E):

(43.39)

We can then expand the solutions of equation (43.38) in a syst.ematic power seríes in >.:

(43.40)

We obtam at once the IDilllY instanton contributions to ali energy levcl!, EN(g) of the
double-well potentia.l at )eading order. lt is oonvenient to rewrite equatiou (43.38) 35

e-l/6g (
- - --
2)E
rt!-E) -d .,. C051fE = EÍ~ (-:_)E __ 1_ _
(43-41)
.j21i !J ,. v'2i u r(~+E)'
914 Multí-lnstantaru in Quantum M echanics 43.1

For example the one--instanton contribuUon ís

(1)
EN (g) = - -
E (2)N+l/2
-
e-1f6g
~(1 + O(gJ). (43.42)
N! g v'21T

The two-ínstanton contribution then ís:

Ej;l(g},. -~2
1 (2)'N+l
-
e-!f3o [in(-2/g)- t/J(N + 1) + O(glng)l,
- - (43.43}
(N!) g 2,.

where .p is the logarithmic derivative of the f -function. Tbe appearance of a factor

In {I can be &imply understood by noting that tbe ínteractíon terms are only relevant for
g-l e-• of arder 1, i.e. 6 of arder -In g.
More generally it C<ill be wrilied that the n-iru;tanton rontribution ha& at leadíng arder
the form:

(43.44)

in which P: (a) is a polynomial of degree n- L For example for N =O- find:

3 2 ,.2
l'l(o) = 2 (a +-r) + ]2' (43.45)

in which "'l' is Euler'• constant 1 = -t/1(1)-- 0.577215 ....

Discu.s.rion. When we naw perform mil' analytic continuation from g negative to g
positive, two things happen: the Borel sums become complex a.nd get ll.ll Ímagin81}' part
exponentially smaller by about a foctor e- 1/"" than the real part. Símultaneom;ly the
functionln(-2/g) also becomes complex and gets a.n imagínary part ±i11. Since the sum
of ali the contríbutions is real, the imagínary parts should ca.ncel. This argument leads
to an evaluation of the imaginary part of tbe Borel swn of the perturbation series, or of
the expansion around one instanton for example.
From the imagin"'J' part of P;, we derive

(43.46)

ano therefore:
(43.47)

Expanding naw Et0 ) (g) in perturbation series

E(O) (g) 'LE~0ll, (43.48)

k

and again using a dispersion relation to calculate the coefficients E~01 , we obta.in the
large order behaviour of the perturbe.tive expa.nsion:

(43.49)
 Multi-Instantons in Quantum Mechanics 915

From the irnaginary part of Ps we can derive the large order behaviour of the expansion
of E(ll (g):

(43.5-0)

We ex:press that the imaginary part of Etll (g) aiJd E\3 ) (g) have to cancel at leading
arder:
e-I/6o )3
lmEOl (g) ~ ~ ( - - lmjP3 (ln (-g/2))). (43.5!)
..pg
For the toel!icients Ek'l we write a díspersion integral:

(43.52)

Using then equations (43.45) and (43.51) we find

(43.53)

At leading arder for k large, we can teplace g by its saddle point value 1 /3k ín In g IUld
finally obta.in:
3k+2 [ ln6k+1+0 (lnk)]
--,.-/;:! T . (43.54)

Both results (43.49) and (43.54) have been checked against tbe numerical behaviour of
the ccrresponding series for whlch about 100 term• are known.
To check the real part of P:z, the foUowing quantity has been evaluated numerically:

4
{! (E++ E_)- Re [Borel sum E! 0 l (g)J}
ó(g) (43.55)
(E+- E_J 2 (ln2g- 1 +1') .

In this expression E+ a.nd E_ are respectively the ground state and the first excited
state energy. In the sum (E+ +E-) the ccntributions corresponding to an odd number
nf instantons canceJ, Therefore the nurnerator ís dominated for g srnall by the real part
of the two ínstanton contribution. Tbe difference (E+ -E~), as we know, is dominated
by the one-instanton oontribution. Using the voríous e:x.pressions given above, it is easy
to verify the.t Ó (g) should go to 1 when g goes to zero. Thble 43.1 sbows whe.t happens
in the range of values of g for which the nuroerical calculation was reasonably accurate.

Table 43.1
The rotic l!i(g) tu a function of g: For 9 small il •hould btlha~• like I+ 3g. (Th• V<llue for
9 = 0.05 .. 0.938807.)
916 MulttCJnstan.!ons ín Quantum Mechanics 43.2

43.2 The Periodíc Cosine Potential

To avoid the pro\iferation of big integer factors ít i.s oonvenient to use a romewhat non-
standtrrd normalizatíon of the coupling constant . We write the hamiltonian
2
H""--1 ( -d ) 1
+-(1 (43.56)
2 dq 16g

A;, we have already d~uSI;ed in Cbapter 40, to eacb state of the harmonic oscilla.tor is
associated a band for g small. A state in the band il; cbaracterized by an angle <P,

(43.57)

The operator T is the operator which translatefl by one period of the potential.
H we expand in a Fourier seriffl the corresponding energy E: [g, <p)

+oo
E:(g,op)"" L f!{g)eól", (43.58)
1=-oo

then, as we have seen, for g sma.ll t:1 (g) is domlnated by i-instanton configurations. ln
particular for the ground state energy in the <p sector, E 0 (g, 1p), the I =. 1 term beha.ves
lilre
Eu l-I (g) - _1_ .,-t/29. (43.59)
. ..fiõ

The pnrliti<m funclion ín the '{> sector. Let us recall a useful formula. We first define

Z[{3,g,op) ""I:e-~E"'( 9 •"1, (43.60)

N

the partition function ir. the sectOJ" corresponding to angle <p. In particular for {3 Jarge

(43.61)

Tben in tbe notaLion of Chapt.er 40 we have

(43.62)

This is the expression we sha.ll use to calculate Z({J,g,<P)· Tbe patb integral representa-
tion of tr' (T1e-I'H) is

tr' (T e-PH) - f.
q(P/>)=o( -6/2)+!2~/ Vfi
[dq(t)] exp [-S(q)], (43.63)

with
[1 1 ]
S (q) =
!
-M>
6/2
dt -q(t)2
2
+-
16g
(1 - cos4qyg) . (43.64)
43.2 Multi-lnstanto'IIJI in Qt<antum Mechanics 917

Note that for fJ large the sum (43.62} can also be written:

e-~Eo(g,.,l f!~=
~ J!dq(t)]exp{-s(q)+i,;g'f'[q(f3/2)
211"
-q(-(3/2)]}. (43.65)

lndeed the configuratians for whícb q(f3/2)- q ( -(3 /2) ís nat a multiple af the period
are suppressed in the large (3 limít, sin<:e tbeir classical actíon ís necessarily infinite. The
additional term can also be trivially rewrltten as the integral of a. local density:
+fN2
q (+!3/2)- q(-fJ/2)
1-tm dtrj(!). (4.3.6&)

These expressions ha.ve a. slraíghlforward generalizati<>n in lhe ca.se of lhe B-YB.Cuum af

lhe CP(N- 1) madel and gauge thearies (see &cti<>ns 40.4,40.5).
Th.e n-írutanton configurations. There is one important difference between this ex-
a.mple, and the dauhle wel! potent.ial. In tbe previous case ea.cb oonfiguration was a
succession of instantons and anti-instantons. Here in contrast we canal each step decide
ta go to the next minimum of the polential or to go <>ne step hac:k. We assign therefore
a sign é= +1 to an ínstaJJton aod a sign' é= -1 to an anti-instanton. A otraightforward
calculation, similar to the calculation prerented above (for det!Lils see Appendix A43.2)
ylelds the following interaction term hetw...n two consecutive inslantons of types €J aod
E2 separated by a dlstance 612,

(43.67)

The force between instantons of the same kind is repult;ive, wbiie it 1s atlractive for
differenl kinds. Let us ag!Lin cal! À the one-iDStanton contribution at leading arder,

}. = _1_ e-•/29. (43.68)

_fig
With tbís notation the n-ínstanton contribution reads

pe-fJ/l ~f.
n t,~-o
li (ts-t
6;- {3) ln (8;), {43.69)

with
J,. (6;) = L
~.:=±1
exp (t -=~•"•+' -í<;'f') .
i=l g
e-O; (43.70)

..The
,.
addílional term -if:;<p comes from lhe formula. (43.62). We have identified En+l and

In eontrast with the case <>f the double-w..ll potential, the interaction between in-
staJJtons cont!Lins bath altractlve and repult;ive tenns. Thus we ha.ve to ~gin with !I
complex to perform the analytic oontinualion o! bolh the Borel sums lUld the instaoton
contríbutions.
Following the same 5teps IIS in the case of the double-well potent.ial we lhen obtaln:

z<nl(p,g,<p) = :11.-13/2 ~f ds e-~S•rn (-•)

x L
{E~~::t1}
exp [t
'"""J
-if:;<p- sln (~"'''+' )] . (43.11)
9!8 Multi-Jnstantorn; in Quanturn Mechanics 43.2

Let us introduce the notation:

"=
ln(g/2). (43.72)
and dPcide to make the analytic continuation from above so that;

In (~ec;<,+I) a - ~i" (1 - E;ei+,). (43.73)

Expression (43.71) can then be rewritten:

zlnl((3,g,'fi) ~: ...-~1' ~"f ds e-~· [r(-s) .,~.-•r

" ~
{<;~±!)
exp t
l.=l -iE;<P + ~· (1
r €;ei+ll1· (43.74)

The summation over the set {<i} corresponds to t.he calculation of the pwtition function
of a one-dimensional Ising model whose transfer matrix M is:

M (43.75)

Tbe sum then is just tr M". Til€ eigenwlues m± are:

m: cos-cp ± (e 21.wo11_ sin 2 cp) l/7. (43.76)

(43.77)

The sum E (/3, g) o( allleading order multi-ínstantoll contributíons can now be cakulated:

E(tJ,g)=e-ll/2_~
/3e-fJ/2 f dse-li•Jn{[I

x [I ->.r( -s) e-"' m_ (s))} (43.78)

The argument of the logaritlun can also be wrítten as:
[1- >.r( -s)e_ .., m+ (s)] {1- Àf{-s)e-'"m- {•)]
2i" e<•~-..-)•]
=1->.r(-s)e-~•
[2c:os<P+J. r(!+ 8 ) (43.79)

An integration by pa.rts of (Je-1'1• Íil integral (4.3.78) yields our final result:
00

E ({J,g) ~ ~ .-ilEN(9). (43.80)

11-11
in which E11(9) ~ + sN(>.,o) is a solutíon expa.ndable in powers of À of the equatioll:

(-~)E e-l/2g
( ~)-E~
g r(! E)+ g rt! +E)
(43.81)

Note the symmetry in the change g, E- -g, -E- This symmetry is, however, sligbtly
fictítíous because the equation is actually quadra.tíc in r( i-
E) and only one root,
corresponding to m+, is rel...,ant for g > O.
43.3 Multi-lnstantOTJ8 in Quantum M ochanícs 919

43.3 General Potent\als with Degenerate Mínima

We now consider a general analytic potential having two degenerate mínima located at
the origin and another point x 0 :
V(x) =- ~x 7 +O (:r 3 ) ,
{ V(x) = ~"" 2 (x- x 0 ) 2 +O (x- xo) ( l) (43.82)

For definiteness we assume"" :> 1.

ln such a situation the cl&Ssical equat.ion of motion have instanton solutions connectillg
the two mínima of the potential. However there is no gnound state degeneracy beyond
the classicallevel. Therefore the one-ínSI.anton solution does not contribute anymore to
the path integTal. Only periodic classical path.s are relevant: the leeding contribution
now comes from the two-imtanton configumtion.
To calculate the potential between instantollS and the norm.elization of the patb inte-
gral it i.6 conV€nient to first calculate the contrlbution at (3 finite of a trajectory described
TI times and take the Jorge (3 limit of this expres.ion. Using the expressions derive<! in
Chapter 38 we iind:

(43.83)

with the definítions:

x~w 21('+"') exp { ~ ~)]}.

1
C [ f"" dx ( - -- (43.84}
1 +"-' lo vf2V(x) X

a.nd
A ((3) = 2 f'' J2V (:r)dx- 2C(l + "-') e-WnJw/(l+"'l + ... (43.85)
lo ""
Note that n has not the saDie meaning here"" in Section 43.1. Since"" is different f:rom 1,
the one-instanton configuration does not contribute and TI inBtead counts the number of
irwt.anton WJti.instanton psirs in tbe language of Section 43.1. Therefore n here IICtually
corresponds to 2n in the w 1 limit.

49.3.1 The n-instanton a,clion

We now ca!l B; the successive amounts of time the dassical tmjeetory spends near x 0 ,
""d "'' near the orígín. The n-ínstanton action take5 the form

A(íl;,'!'i) =na " 1 e-w•. +C2 e-"''),

í:)C (43.86)
i~l

with r;;:_, (6; +\Oi) = (3 a.nd

a= 21•o J2V (:r)d:r. (43..87)

By comparing the value of the ection at the saddle point

6; __
(3_ - _;:jJ_ (43.88)
n(l+w)' <Pt-n(l+w)'
with the expression ( 43.85 ), we see that we can cboose
C,= 2Cjw, C2 =2C, (43.89)
by ndjusting the definitions of 8 and I"·
920 Multi-lnstantono in Quantum Medumics 43.3

43.S.f! Tlle n-instan!on rontributíon

The n-inste.nton oontribution then has the form:

(43.90)

The ad<litional term Z:, ~ {1 -w) o, in the integtand comes from the determinant gen-
erated by the gaussian integration around the classical pe.th. The normaliza.tion cap be
obte.ined by performing a 5teepest desctmt integre.tion over the variables e, and 'l'i and
compare the result with expression (43.83). The result is:

(43.91)

The factor 1 /n comes from tbe symmetry of the actioo under cyclic permute.tions of the
8; e.od 'f'i·
We now set:
2G
!l=--. (43.92)
g
As ín Section 43.1, we introduce e.n integral representation for the 6-function:

õ (~(Ih+ 'P;) - /3) (43.93)

witb n >O.
Expression (43.90) ca.n be rewritten as:

(43.94)

w here we ha,"" now defined

I (•) = l+oo e••-,..-• dll, (43.95]

J(•) {= exp{[~(l-w)+s]e-~e-"'6 }do. (43.96)

In the small g límit these integrais can be eva.luated:

/(s) = 11'r(-•), (43.97)

J (s) ~_I_ (!!.)(o+l/2)/w r (1• (•+2l)f w.
) (43.98)
.jiíW w

We cal} as before E (/3, g) tbe generatíng functionru of the many-instanton contrlbutions.

43.3 Multi-ln•tantons in Quanium Meclw.nics 921

Summing over n a.nd íntegrating by parts, w.> obtain E (/3,g) as a sum of residues:

(43.99)

in whicl> tbe values E. ~ ! + '" are the solutions expandable for g small of the equatíon:
JJ,)Ejw e-•Jg
( ;;_; r(i-E/w)JlEr(!-E)~ -I. (43.100)

We now notire that we find two series of energy leveis corresponding to the poles o! the
two r -functions:

E,."" N +! +0(>.), (43.101)

EN = (N +~)w+O(.>.). (43.102)

The same expression contains the instanton contributions to the two dilferent oots of
leveis.
One can verify that the many-inst.anton contributions are singular for w ~ 1. But if
one directly set.B w = 1 in equation (43.100) one obtains:

-1,

equation whlch ca.n be rewritten as

(43.103)

This is exactly tbe set of two equations obtained in Section 43.1.

49.9.9 Large arder estimates of perturixltion themy

Expr~ion (43.100) can be used to determine the large order behaviour of perturbatíon
theory by calculating the ímaginary pnrt of the leading instanton contribution a.Dd writing
a dispersíon integral as we have done in Section 41.1.1. For the energy EN(Y) = N + + l
O(g) one finds
(43.104}
wíth

KN = ( ~~;;' w-CN+lf>l/"'(2C)(N+l/2)(1+1/"') sin [" {N + !) (1 + lfw)]

x r[!- (N +i) fw]. (43.l05)

From the imagi03TJ' part of EN(U) one deduces at large order k:

E ~K r (/c+ (N + 1/2)(1 + 1/w)) ( O (k-l))

Nk N ak+(N+l/2)(1+!/..,} 1+ . (43.106)

Note tbat this expression, in contrast with the instanton contribution to tbe re&l part, is
uniform in the limit w I in whlc:h the result ( 43.49) is recovered.
 Multí-Imtantons in Quanlum Mechanics 43.4

43.4 The O{v) Symmetric Anharmonic Ü5cillator-

lt is interesting to ronsider a last exrunple, the analytíc continua.tion of the energy leveis
of the O( v) symmetríc anharnmnic oscillator for g < 0:

(43.107)

We first discuss the v = 2 example, and the genetalízation will be straightforwa.rd.

The 0{2) anharmonic oscillator. The instanton rotution is of the form

q(t) = uf(t), (43.108)

in which u is a fixed unit vector. The on~>-instanton contribution to the ground state
energy is:
41
_E(Il(g) = e1 f3•(1 +O(g)), forg __. 0_ (43.)09)
g
It js easy to calculate the instanton interaction. The n-instanton action then is:

A(6';) -~n-4Le- 9'co•<P•· (43.110)

•
in which IJ, is the disto.nce between two successive instantons and ~· the angle between
them:
(43.111)
lt is convenient to consider the quantlty (Sections 3.3,23.5):

tr[R(cr)e-W]= jldq(t)]exp(-S(q)),withq(-!l>)·ii(!,a) coso. (43.112)

The matrix R(o) is a rotatjon matrbc which rotates vectors by a.n angle a. lt leads to
tbe boundary conditíon that q(t) at initial and final times should di!fer by ao angle o:.
The r.h.s. of equatíon (43.112) can be rewritten:

tr(R(o)e-i9Hj = Le-ün-PE,,N. (43.113)

l,N

ln this expression l is the angular momentnm. The honndary condition imposed on the
path integral (43.112) implies a constraint on the many-instanton configuratioo:

(43.114)

constra.int which can be implemented through the identity:

6 (t o)
~~
'P; - ~ I: [il (t ~~) iol] .
~-~
exp
~l
(43.11D)
43.4 Multi-Jnstanto"" ín Quant-um Medwn;cs 923

The 11-instanton contribution to expression (43.113) then take. the form:

(43.116)

where we h ave set:

À = 4i ell3o I' = 4 (43.ll7)

g ' 9

..!_ /. ~ d<p j"""" d6' exp (sB + it'P- I" e-e cosop).
2

h (s) = (43.118)
2:n- o -O<

We cal! E 1 ({J,g) the generating function of n-insl.ànton contributions at fu<E>d angular

momentum 1:
(43.119)

Let us evaluate 11 ( s ). The integmtion over 6' yields:

(43.120)

Performing the last integral and using various relations among r functions, we finally
obtain:
lt (s) (43.121)

lt is easy to verify that:

lt(s) Lt(s). (43.122}
From e<Juation (43.119) we now derive the result:
00

tre~PHl = L e-fJEN_~' (43.123)

N=O

with EN,I = •N,I +I the solution of the equation:

el/(39) (-~)E e'"(E+I)/2 r(! (1 + 1- E)) (43.124)

g r(!(l+l+EJ) l,

(43.125)
Note that checks about these expression.s """ provided by the surprising perturbatiV<!
rela.tion between the 0(2) anhaJ"monic oscillator wlth negative couplíng and the double-
well potential.
The O( v) symmetric hamiJtonian. One can extend this result to the general O(v) case
since, at fixed angular rnomentum I, the hamiltonian depends only on the cornbination
l +f; v. Hence, rnaking in equation (43.124) the corresponding substitution, we obtain:

;e'/(30) (-~)E é"(E+I+v/2)/2 r H

(l + "1 2-
E)) ~l (43.12õ)
g r(~(l+v/2+E)) '
924 Multí.Jns!antons in Quantum Mechanics 43.5

At leading arder in J. we recover the imaginary part of the ener~· leveis for 9 small and
nega.tive;

(43.127)

Using the Cauchy formula we can derive fram thiE exprcssian large order estimates for
perturbatian thoory. At next order in J. we obtain the two-instanton contribution which
is related by the same di.spersion rel&tíon to the large orrler behaviour of the petturbative
expansion around one instanton.

43.5 A General Conjecture

Up to now we have considered instanton contributions ooly at leading order. However

the form of the result is extremely suggestive and has Jed us to conjecture the general
form of the semidassico.l expansion for potentials with degenerate mínima.
To be spe<:ífic let us explain the conjecture for the double-weli potential, o.lthough it
can be easily generalized to the other problems discussed in this cha.pter (see Appendix
A43.5).
We introduce the functíon D(E,9)

D(E,g) ""E+ LY}Dk+l(E),

"" (43.128)
·-1
in terms of whiclJ the petturbation expansion l'<lr an energy levei E~l can be obta.ined
by invetting:
(43.]29)

Tb take ínto account instanton contríbutions we write instead another equatíon:

D(E,g)
_l_r(l D) ( -~ ) ,-A(g,EJn (43.130)
V2ir 2

with
1 «>
A(g, E) = 3 +L l At+l (E). (43.131)
9 k~l
The functioll5 A(E,g), D(E,g) can be calculated by expanding the rorresponding WKB
seriel, which is an expansion at gE fixe<!, for E small (for deta!ls see Appendix A43.5).
The eaefficients D. and Ao are polynomials af degree k í.n E of defined parity.
If we solve (43 .130) in the ane-instanton approximation and substitnte ínto equatlon
{43.128) we find

...lO)
E = .:::• (g) - E:
(2)
-
N
-
] .,-.A(g,E(Oí)/2 ÔD (
E
(0))-l
. (43.132)
9 N! ..;;rg ôE

We ...e that tbe knawledge of the two functíoos D and A is equivalent to the knowledge
Hlld the perturbative and one-instanton cxpansions for allievels and to ali orders.
4.3.5 Multi-lMtantons in Quantum Mechanics 925

1f we now systemati<:ally expand equation (43.!30) """ !ind for an energy level the
fol!owing expansion,

E(g) (43.133)

Ali the series in powers of g appearing in this expa.nsion are determined by the pertur-
bati"" expansion of A and D. This phenomenon has recently found an explanati<m in
lhe framework of lhe theory of resuf"!lent functioDS. ·
Moreover we conjecture that till these series have to be summed for g negatíve first.
and the value of eaclt instantou contribution for g p06itive is then obtained by analytic
contlnuation. Tbe property that the infmite number of perturbation series around ali
instantoDS are related may, at Jeast in quanlum mechanics, 5implify the probiem of the
summation of the many-instanton contríbutíons.

Bibliographical Notes
The two-instanton contribntion and its application to large order behavíour for potential
wítb degenerate mínima háS been discussed in
E. Brézin, G. Parisi and J. Zinn-Justin, Phys. Rev. 016 (1!177) 408; E.8. 8ogomolny,
Phys. Lett. 9!8 (1980) 431.
The multi-instanton contributions have been calculated in
J. Zínn-Justin,NucL Phys. 8192 (1981) 125; B21B (1983) 333; J. Math. Phys. 22 (1981)
511; 25 (1984) 549.
In these articles a re!ation between tbe 0(2) anhnrmonic oscülator and the double-well
potential wa.s conjectu.OO, wbich was !ater proven in
A.A. Andrianov, Ann. Phys. (NY) 140 (1982) 82.
Multi-instantoru; in the 0(3) non-linear cr-model ha.ve been co115idered in
V .A. Fateev, I.V. Frolov and A.S. Scbwarz, Nucl. Phys. 8154 {1979) l, Yad. Fiz.
30 (1979) 1!34; B. Berg and M. Lüscher, Gommu!L Math. PhyB. 69 (1979) 57; A.P.
Bukhvostov a.nd L.N. Lipatov, Nuc/. Phys. 8180[FS2] (1981) 116.
The peculiar equations (4.3.128,43.13ll) which imply relations between lhe pertnrbative
expa.nsioM around multi-instantons !lave been recently pTO\>en using tbe theory of resur-
genl functions in
F. Pbarn, &suf"!lena, Quantized Canonicai 1hlnsfo.matíon and Multi-Irutanton, Ai-
gebraic Analysis, vol. 11 (1988), C.R. Acad. Sei. Pari.s 309 (1989) 999; E. Delabaere et
H. Dillinger, thesis Universitê de Nice (Nice 1991 ).
For th• WKB rnethod "" discussed in the appendix see alro
A. Voros, Annales JHP 29 (1983) 3.

Exen:ises
&emse 1,9.1
Complete the calculatíon of Appendix 40.6 by evaluating e:xplicitly the grouud state
energy at leadillg arder for lhe hamiltonian H:

H=~ [-n(d/dq)2 + ~n- 1 (E'(q)) 2 ~E"(q)l,

iD whkh 11 is the expansion pararneter a.nd E(q) = q 2
- 2q3 /3.
926 Multi-Insto.ntons in Quantum Mechanics A43.1

APPENDIX 43

A43.1 Multi-instantons: The Determlnant

We can wríte the operator M defined by equa.tion (43.20) a.s:

M=- (
d
dt
)2 +l+I;v(t-t,),
n
(A43.1)
i=l

in which u(t} is a potentia.l Iocalized around t = O, and t, are the posítíons of the
instantons:
v{t) = O ( e-lil) , ltl __, oo. (A43.2)

We want to calculate

detM Mõ 1 = det {1 + [- (d fdt) 2

+ lr' t t;)}.
v(t- (A43.3)

Using the identity lndet

·-·
trln, we expand the r.h.s. in powen; of v(t):

(A43.4)

with the definition:

(A43.5)

lt is clear from the h<lh,.viour of v(t) and ó(t), that when the instantons are largely
separated, only the terms in which one reta.ins from each potentía.l the same instanton
contribution survivo;. Therefore

JndetMMõ' = n L -(-l)k+l
00 f .
- k - ó(u,- uo)v(u,}· «Ó (uk- u,)v(ul) rr du;,
k-1 j=l

for 14 - t1i » 1. (A43.6)

We recogni2€ n times the logarithm of the one-instanton determinant.

A43.2 Multi-Instantons in Quantum Mechantes 927

A43.2 Tbe lnstanton Interaction

We assume, os in Section 43.3, that the potential has two degenerate mínima at the
poínts x O and x = Xo with:

V (:r) = !x 2 +O (x 3 )
(A43.7)
{ V (x) = {w~ (x- x 0 ) 2 +O ( (:r- x 0 ) 3 ) .

Let us write the one-insta.nton solution q,(t) whicb goes from Dto go x 0 J..;9 as

q,(t)"' f(t)/..;9 (A43.8)

We cboose the function /(t) in sucb a way that it satisfies:

xo- f(t) ~ Fce-"'' fw for t ..... +oo

(A43.9)
{ /(t) ~./C e' for t--+ -oo.

By solving the equation of motion, it is easy to calcu]ate the constant C:

c x>w>f(l+"')
0 ex {
p l+w
~ [1'"• dx
0
(-1 __: _ 1 )]}
J2V(x} X w(xo-x) ·
(A43.10)

We recognize the consta.nt (43.84).

We now construct insta.nton-anti-instanton pair configurations q(t) whkh correspond
to tra.jectories starting from, and returning to, q = qo or q O. Since we wa.nt a.Lso to
consider the case of twO successive instantom, we assume, but only in tbis )ast CBS€, that
V(x) is an """n function a.nd has therefore a thírd minimuw at x = -x 0 •
According to tbe di.<;(:ussion of Chapter 43, we ca.n take as a two-insta.nton configura-
tion:
1
01 (f+(t] + ~f-(t)). E ±1, (A43.11)

with:
f+(t) j(t-9/2), 1-(t)=f(-t 0/2). (A43.12)
s measure of the instanton separ&tion. Th,. case E = 1 rorresponds to an imtanton-
f) i$
anti-instanton pair starting from q = qG at time - oo, approaching q "' O at intermediate
times, Blld returning to q0 • The case<= -1 corresponds to a sequence of two instantons
going from -go to qo. Fínal!y for the d=ical tmjectory whicb goes instead from the
origin to q0 and ba.ck we ca.n take:

g2{t) ~ [/(1 + 8/2) +f (fJ/2- t)- zo) /..;9. (A43.13)

Let us naw calcula.te the classical a<:tion corresponding to q 1 (t). We sepa.rate the action
into two parts, corresponding at leadíng order to the two insta.nton crmtributions:

(A43.14)
92B Multi-Instantom in Quantum Mechanícs A43.2

wíth:
s~ (qd =- [= H4~ +~v (Jgq 1(t))] dt.
(A43.15)
S_(q,) lJ~q~+~V(Jgq 1 (t)))dt.
The value t = O of the sepa.ration point is somewhat arbitra>:y and ca.n be rephu:ed by
any value whlch remains finite when 8 becomes ínfinite. We then use the properties that
for 1/}arge f+(t) is small for t < O, and f_ (t) is small [or t > O, to expa.nd both terms.
For example for S+ we find:

S+ (q1) ~ ["" dt {[!i;(t) +V (f+(t))] H [i-(t)Í+(t) +V' (f+(t)) f_(t)J

+~ [j:(t) +V" (f+(t)) /_:(t)]}. (A43.16)

Since f-(t) decreases exp<mentially, only values of t small oompared to 11 f2 contribute to
the last term of equntion (A43.l6) which is proportional to V". For such values o! t, we
h ave:
{A43.17)
For the terms linear in f-(t), we integrate by parts the kinetic term and use the equntion
of motion
/(t) = V' [f(t)]. (A43.18)
Only the integrated term survives a.nd yields:

["" dt [í-tt)Í+(t) +V'(!+(!)) /-(tJ] =-i (-8/2) f (-9(1). (A43.19)

'Ibe conttibution 8 _ oa.n be evaluated by exactly the same method. We note that the
sum of the two conttibutions reconstructs twic" the dassical action "· We tben find:
S(q 1) ~ [2a-2ef(-!1/2)Í(-8/2) +···J, (A43.20)
with
a = [ • v'2V (:r)dx (A43.21)

Replacing, for 9 la.rge, f by its osymptotic form (A43.9), we linally obta.in the dassical
action:
(A43.22)
a.nd therefore the instanton interaction.
Following the sa.me steps, we can calculate the classical action correspondlng to qz(t).
The result is:
(A43.23)

which for O la.rge is equivaleot to:

1
S(qo)= [2a 2(C/w)e-'"' 9] . (A43.24)
g
Finally, íi ....., oonsider tbe case of a finite time interval {3 with periodic boundary con-
ditions, we can combine both results to find the action o[ a periodic trajectoty passiog
close to q = o and q = qo:
S(q) = g- 1 (2a 2C (e-~+• +e-wf /w)J, (A43.25}
in agreement with equ,.tions (43.86,43.89).
A43.4 Multi-Instantons 1n q,antum Mochamcs 929

A43.3 Multi-instantons from Constraints

Althougb multHnstanton configurations do not correspond to solutiODB of the equation of

motion, it ís nevertheless possible to modify the classieal action by introducing constraints
and integrating over ali passib!e constraints, generali:~.ing the method oí Appendíx A37.2.
The main problem "•ith such a method is to find a system of conetraints which are botb
theoretícally reasonable, and convenient for pra.ctical calculations.
One can for instance fuc the positions af the instantons by introducing in the patb
integral (in tbe example of the double--well):

(A43.26)

where ti are the insta.nton positions and '' a successions of ± indicating instantoru; a.nd
antHnstantons. One tben uses a.n integral representation of tbe 6-functions, so tbat tbe
path integrlll becomes

C'~:! r
2

f lldt;d)., fJdq(t)j llexp [-S (q,Ã;)]

a=:J t=l
with
(A43.27)
S {q,A;) =S{q) +f:>., f
t=l
dt q,,(t- t;)(q(t) -tJ,,(t t;)}.

The arguments of Appendix A37.2 can then be generlllized to recover the results of
Section 43. 1.

A43.4 A Sirople Example of Non-Bom Summability

Let us try to iiJustrate the problem of non-Borel summabi!ity with the example of a
simple integral, which shares some of the features of the problem in quantum mechanics
wbiclt we have studied in O:tapter 43. We consider the function:

l(g) = ~1+""
y211"
_..,
dq exp r-~V(q,íg)],
g
{A43.28)

wbere V(:r) is an entire functíon with an ahsolute minimum at x =O, V(O) =O. For g
small 1 (g) can be cruculated by steepest descent, expanding V around q 0:

I(g) LI~yk. (A43.29)

k~O

It is easy to write a finite dimensionlll integral o( the forro (A43.28} as a generlllized

Borel ar Laplace transform,

J(gl = ~f dqdt6 IV (q,ígJ _ tJe-tJ•. (A43.30)

We integrate over q,

J( )-
g -
l
J21'9 1""
o
d
t e
-•/•~
'7 IV' (:t;(t)]l'
I (A43..3!)
930 Multi-lnstantons in Quantum Mechaníc• A43.4

ín which {:t;(t)} are the solmions of the equation

Vj:r,(t)] =t. (A43.32)

>\'hen the functiou V(:t) !s monotonous both for :r p<>Sitlve and negative, the equatíon
(A43.32) has two solutions for all values of t and the equation (A43.31) is directly the
Borel representation of the function J(g), which h"" a Borel summable power series
expansiOD.
Wenow assumeinsteadthat V(:r) ha.s asecond local minlmum which gives a negligible
contributít.o to l(g) for 9 small. A simple example is

(A43.33)

which has a minimum at x =- I. Between its two mínima the potential V(:t} ha.s a.
maxímum, located at x a, whose contribution domjnates th" large order behaviour of
the e:xpa.ll>i;ion jn powen; of 9

I* cx f(k) [V (a)J-k, V( a)> O, (A43.34)

k-oo

(in the ""6.mple (A43.33} V(a) ~ n2 (l a/2)/6) e.nd the seríes is not Borel summab)e.
The na~.,e Borel tr311Sform of 1(9) is obtained by retaining in equation (A43.31) only
the roots of equation (A43.32) which exlsts for t small. The singularities of the Borel
transforlll then correspond to the zeros of V"(:r).
For the potential (A43.33) the expression (A43.3l) has the form

I _ 1 {+""' -tJg ( 1 6(V(Il) t) + O(V(Il}- t)l!(t- V(l))

(g)- ...p:;rg lo dt" IV' (:z:, (t))l + IV' (x2(t))l IV' (:r3(tJ)I
9(t V(I))]
+ IV' (:r.(t))l • ( A43.35)

with the t:lefinitkms (see figure 43.2): :r 1 (t) <;;O::; ;r:2 (t)::; a::; x,(t)::; 1 5 :>:4(!).

Fig. 43.2 The four root.. of equation (A43.32).

The id.ea of the analytic oontinuation is to integrale ea.ch contribution up to t +oo

followínil; a contour ":hich passes below or above the cut along the positive real ax:is. This
means t!:.at we consider x 2 (t} to be solution of the equation:

V J:to(t)J = t ± ic. (A43.36)

A43.5 Mul!i-Instantons in Q!Uln!um Me~l!anic5 931

The sign ís a.rbítrary. Let us for ínstance choose lhe positive sign. We then have to
subtrect thll; addítional contribution. We proceed in the same way for :ra(t) for t > V (a).
Since :r2 (t) and x 3 (t) meet at t = V(a), the analytic continuation will correspond to take
for xa(t) tbe other solution:
V (:ra(t)l e'f ic. (A43.3i)
We therefore have to subtract from tbe total expression the contributions of two roots
of the equation. But it ís easy 1.o verífy that this is just the contribution of the saddle
point located at :r = a, which corresponds to "' maxirnurn of the potential. .
We have therefore succEeded in writing expressíon (A43.35) a.s the sum of three saddle
point contributions (see figure 43.3}. There ís some arbitra.riness in thís decomposition
wbkh here corresponds to the choice E = ±1.

Imx I I
c.,
I I
l
1 te,
' I I
----~-----1 I
c l
"
I
c.t
I
I
I
Fig. 43.3 The different oontours in the :r plane.

In tbe cornplex :r plane, we have replru:ed tbe ínitial contour C on the real posítive
axis, by a sum of three contours C1, C2 and C 3 corresponding to the three saddle points
located at O, a, 1.

A43.5 Multi-Instantons and WKB Approximation

Let us give a more general form of the conjecture presenteei in Sectíon 43.5 and indícate
how a few properties of the functions A(E,g) and D(E,g) can be obtrillted from the
WKB expansion.

A-/9.5.1 The conjecture

The àouble-well potentiol. For the double-well potential we have conjectured the equation

l 2)D(E,g)
.;21lr (! - D) ( - g e-A(g.E)/ ~ €i,
2
(A43.38)

with
00

D(E,g) =- -D{ -E, -g) E+ L gk Dt+J (E), (A43.39)

(A43.40)
932 M..Zti-Irutantons in Quantum Mechanics A43.5

The coefficients Dk(E) and A0 (E) are polynomials of E of degree 1<. This conjecture has
been geuerali~ to a potentiel with two Mymmetric wells:

l "r(~ -D.(E,g))
(- C)D 2 (E,g) ( C)D 2
(E,g)

2 9
f(!- D2(E,g)) -
9 .,-A(g,E) :c -1,
(A43.41}
where D 1 (E,g) a.nd D~(E, g) are deterrnined by the perturbative expansíons around eacli
of the two mínima of the potentíal and C is a numerical consta.nt.
Thf cosin<: potential. For the casine potential i\(1 - cos4q) the conjecture ís
(A43.42)

The O(v)-oymmetric anharmonic osciilalor. For the O(v) RDha.nnonic osdllator the
conjecture corresponds to the expa.nsion of the energy leYO!Is for g < O and, in particular,
yields the installton ccntributions to the large arder behaviour.

(A43.43)

A4.'1.5.l? The WKB apprvximation

Consider the Schr5dinger equation:

-~rt>'' + (V( ..,t9q)jg)!/J = E.p. (A43.44)

We first assume that q =Ois the absolute minimum of V{q) RDd moreover V(q) ~ !.f
for q small. Setting x = ,j§q the equation cBI! he rewritten

-g2 1/J" + 2V(:.:).,I = 2gE.p. (A43.45)

The WKB expansion is an expansíon for g - O at Eg fu<ed, in contrast with the perturba-
tive expa.nsion where E is fixed. lt can be oonstructed by introducing the corresponding
Riccatl equation, set.ting:
S[:r) -g,P'IT/J, (A43.46)
where S satisfies:

S~(x) = 2V(:r) 2gE. (A43.47)

One then expands systematically in powers of g, at Eg fixed, starting from S(x) = S 0 (x).
lt is con...,nient to decomp05e S(:r) into an odd and even part set.ting:

S(:.:,g,E) s.. [:r,g,E) +5-(x,g,E), S,i(x, -g,-E) ±S±(x,g,E). (A43.48)

It follows:

gS'__ S! - S: + S~ = O, (A43.49)
gS'+- 2S+S- =O, (A43.50)
A43.5 Multí-lrutantona in Quantum Meckanícs 933

whkh o.llows to wrile the wave function in terms of S+ only:

(A43.51)

The spectmm can then be determtned by the eondition

(A43.52)

where N is the number oí nodes of the eigenfunction a.nd C a oontour whic:h encloses
tbem. In the semiclassicallimit C encloses the cut of So(:r) whic:h joins the tw<> turning
points solutions of So(:z:) ~ 0 (Xt,X2 in figure 43.2). In terms of S+ equation {A43.52)
becomes:
__ I_ f dzS+(z) = N +i. (A43.53)
2•,.ulc
)f we reploce S+ by its WKB expans:ion and expand eaclllerm in a power series of Eg
we get tbe function D(E,g) (the pertur~a.ti"" expansion):

_....!._ J dz S+(z) = D(E,g) = -D(-E,-g}. (A43.54)

2•"!1 !c
Potentials witk degenero!< mínima. In the case of potentials with degenerate minima
tw<> flmctíons D 1 and D 2 (in the notation (A43.41)) appea.r, corresponding to the ex-
pansioru a.round eac:h minimum. An additional contour integral a.rise!oJ corresponding to
barrier penetration effects. The expansion fur Eg smo.ll of its WKB expans:ion yíelds tbe
function A(g,E):

D,(E,g)) + D;{E,g) In( -gfC), (A43.55)

where C' encloses !:r 2 ,:r2 ] in figure 43.2. In the WKB expansion the functions r'(~- D;)
appear in tbe fonn of an asymptotic expa.nsion for Di large. Stíll a calculatlon of A( E, g)
at a finite arder in g requires the WKB expansion and the asyrnptotic expansion of the
[' function only at a finite arder.
O(v) symmetri< potentíals. Tbese expressions can be generalize<! to the case of O( v)
symmetric potentials. The pert.urbative expansion can be obtained by inverting a relation
ofthe form
I'+ 2N + 1 = D(E,g,!'), 1J =I+ t'/2 1, (A43.56)
where the function D(E,g,p.) is given by a contour integral. surrounding ali tetOlS of the
wave functioo on the real axis (including the negative real a.xis) of the even part (In the
sense of equation (A43.48)) o{ -g(,P(/tJil +(v- I)/2lql). The followíng propertíes can
lhen Verified

D(E,g,p.) = -D(-E, -g,p.), D(E,g,/.L) D(E,g,-p.),

Md the coefficient of arder gk in the expansion of D is a polynomial of degree [(k + 1 )/2)

In IJ· ln the WKB e:xp8.1l5Íon the fnnctions D and A again correspond to dilJer-ent oontour
integrais around turning points.
44 EXERCISES: SOME SOLUTIONS

Chapter 1: Algebraic Preliminaries

Exen:ise 1.1
One COil.Biders the fermion integral

f (n t
dO,d9;) exp (.t
l~:]
a.;iJ;tJ.! + 2:: s.;ll,e,) .
J

lntegration directly ylelds tbe I.b.s. of equation (1.99). Expanding in perturbe.tion tbeory
tbe term proportional to the s;'s and using ferminn's Wick theorem (1.96) yields the r.h.s..

E:rercise 1.!!
First
D(ABC) =DA BC+ P(A)D(BC)
=DA BC+ P(A)DB C+ P(A)P(B)DC.
Also
D(ABC) = D(AB)C + P(AB)DC
DA BC+ P(A )DB C+ P(AB}DC.
Comparlng both ,..,.ults one findo
P(A)P(B) = P(AB).

Exen::i3e 1.9
Iti>; of rourse sufficient w chara.cterize the action of V on x•. We can set Dx = 1. We
have w use the be.sic relation
D(AB) = DAB + H(A)DB,
where H being a homomorphism of algebra, eonsistent with tbe grading is characterized
by a complex constant h
H(:r) =hx,
whíle
D(x•) ~- 1 •
lt is now ee.sy w verify tha.t the solution i>;
l h•
D(:r") = 1- h a;P-1.
Tben the consistency condition D(x") = O implies h" I with h = oi 1.
Note tha.t D" O and also in the sense of operawrs on 21
l hP+l
~xD+Dx=O.

Finally o.n linear integratíon operator on 21 satisfying

Dl ID=O,
is
I= D"-1.
44.2 E:x.err::ises: Some StJlution.s 935

Exercise 1..1
Let us express the condition for any pair or generators. lt is then sufficient to consider
the sum of any two of them llke x 1 and :.; 2 . The result relies on the proof by induction
of the identity:
p m-1 _ cP-1
1
(x 1 + x 2 )" I; x;"~-~ ld+'' TI
m.~o l=O
Specializing this result for p = n we obtaín the necessary condition. We then note that
any linear combination oi :tl and x 2 also satisfies
X3(<11X1 + aox2) c(a 1 x1 + a2:z:2):z:3,
with the condition (1.100). We can then prove tbe result for any triplet of generators.
The final result fo!lows from this •econd induction.

Chapter 2: Euclidean Path Integrais io QUAOtum Mechanics

Exercise 2.1
2.1.1. One finds:

Z = (2sinb (íl/2))-
1
(1 ~ j 3.\
1312

-P/2
dt K 2 (t, t) +o (.\2 ) ) ,

in whích:
K(t,t) = !rotanb ({J/2).
Therefore:
Z= [2sinh(fJ/2lr' [1 ~:l.fJ(cotanb2 ({J/2l) +0(.\2)].
Using two algebra.íc identities:

I; exp [- (n + ~) fJ] l/l2sinh (!3/2)],

n=O
and, dilferentiating twice with respect to {J,
~ l)' _ ! _ 2cotanh (fJ/2)- 1
2
;':n (n + 2 exp [ (n + 2) fJ] - 8sinh (fJ/2) '

one can verify that Z ís the expam;ion to first arder in.\ ofi:exp(-fJEn) with:
En = n +~+~À (n2 + n +!)+O (.\ 2 ).
2.1.2. Setting " = n + 4, tbe Te$ult can be written
E= v+~ (v2 + i).\- ív (17v2 + §j) A2 +O (:1.3 ),
Note tbat more generally for tbe potential
V(q) = jq2 + v 1 q~ + v,q 6 +O (q8 ) ,
one finds
n +! E- v.{jE2 + ~) + & (ll]v~ ~v,)+ E (~v~- ~112)
2.1.3. For N = 6 and N = 8 the results are respectiwly
En n+ ~ + ~À(nHJ(n 2 +n H)
En = n+ t +~À (n 4 +2n3 +5n 2 +4n+ D.
936 Exen:ises: Same Solution..< 44.2

E:r:ercise JU!
Let K(q,q') be the kernel

K(q,q') = exp [-ja (q 2 + q' 2 ) + bqq']. (44.1)

lt is convenient to pararnetrize the coefficients a, b as:

_ coshli b _ _ P_
0
- P sinhíl' - sinh6

One then easily verifies that KP(q,q') bas the form (44.1) with e replaced by p8 (up to
a global normalizatíon).
Startlng wltb the approximt>tíon

where). is a free parameter and c fJ/n one finds

p= 1 + W·- tlc• +o <,•)

9 E [1- ~(.\ f;)c2] + 0 {E:S).

Comparlng Kn(q, q') with tbe exact result (2.38) one sees that the torrections come from
the deviations of p frow 1 IWd niJ from /3, H.Ild are of order l/n2. No value of .\ cancels
botb rorrections. A good oompromíse is À = 1 /3 whicb minimizes both correctíom;. We
note that this value. correspond• to tbe result (2.13) obtaínad from tbe small c: expansion.

E:r:ercise 1!.9
We lirst brielly explaín how the S-matrix can he calcula.ted exactly. We consider tbe
correspnnding Scbrõdinger equation:

(44.2)

tb(x) Q = -ik, k >o.

(:! cosh :z:}" 'I''
We llnd:
-<;P + 2ot1Wbxl?' + 2.\ +o( o+ 1) 'P,.., O,
cosh 2 x
For la.ter purpose it is nseful to note

x- -oo
(e"+ e ~)& - { x- +oo

We then set t tania 1111d find

44.2 Exerr:ises: Some So./uticm.J 937

The solutions of thls equation are hypergeometric functíons, which in specia.l cases de.
genera.te int<J Jacobi's polynomíals. The sa!utíon regular at t = -1 for e.xample is

<p(t) = F (a, q(a + b+ 1); !(1 + t)),

with a, b given by:

For JU, !(l + n + b) > Re b > O, which ímplies À > O, we have Euler's formula

Wefínd

- ) = r( o+ 1- d)r(d)
( 1
'{) r(1 +O) '

u.mg a = d +o, we cen write the behaviour of <p(t) near t = 1 as

The singular contribution corresponds w the incoming wave. Note that

We conclude that the forward and backward S-matrix elements, S+ and S_ Me:

r(d +a )r(a+ 1 -á) s_ sín1rd f{d +a)f(o+ 1- d)

r(l +o)r(o) sín:r<o f(1 +a)r(a)

These exptessions, estab!Ji;hed for À > Ohave an ano.lytíc continuation for À < O and give
the complete answer.
We can check the unitaríty relation for d real and o imaginary

·2 0 (
iS 12 JS-1 2 = sm " 1
+ + sin,.(d-a)sín.,.(d+a)

For À > 1/8 d is complex but the relation .r + d = 1 Jeads to the same ideotity.
Note that the potential is reflectionle!$ for integer values of d, i.e. for À ~n(n- 1 ), =-
situation in whicb the potentia.l is attractive and the S-matrix a rationo.l fraction of
a=ik
n-1
S(n) (-lt-~n~.
~I
m-a
The S-matrbt has n- 1 poles correspondiDg to n 1 bouD<I sto.tt"S.
Note fina!ly that in the nonnalizatlon of exercise 2.1 we haw the relation

n+ 1 = E(1 + l/4)-l/l + !9 (n 2
+ n +i), with g
2
"" -1/2>.,
938 Exerr:ues: Some Solutions 44.2

for lhe potential ~ tanh 2 (q). This provicles a check of the corresponding perturbative
resulto.
SemidClSsica! limit. The semiclassical limít is obLained by Ietting lc and >. becoming
simultaneously large k ex ..,15.. Then d- -~ and the r functions are replaced by
theír Stirling's a.pproximation. Let us consider the different cases,
(i) .>.>O, f!= ffifk < 1 with the choice d. -iffi. We find:

lnS+ ~ -ik [(1 + <r)ln(l +a)+ (1 -a )In( I <r)],

lnS_: ~ -ilc{{l + t!) ln(1 +<r)+ (l-a) ln(l cr)J 1fk(1- o).

We note tha.t, as expected, the classically forbidden ba.ckwmd sca.ttering is exponentially

suppressed.
(li) .>. > O, q = ffifk > 1 with the choice d -iffi. We find:

lnS+ ~ -ik [(<r+ l)ln(o + 1)- (a -1) ln(<r 1)} 1fk(<r- 1),
lnS_- -ik [(o+ l)ln(<r + l)- {a- l)ln(<1 l)]

We now note that the claseically forbidden forwatd ~~<:attering is exponentially suppr.,...,J.
(iü) À< O, q = ~/k. We find:

lnS+--ik[(l+ia)ln{l+ía)+(l io-)ln(l ic)],

lnS_- -ilc[(l + io)ln(l +ia)+ (1- i<1)ln(l- it:r)]-11k.

The usymptotic forro of s_ ís only valid for d. non integer. We note agaln that the
classically forbidden ba.cl<wmd scattering ls exponentially suppr.,...,J. ,
Semíclassical calculation. We now perform the expliclt ca.lculation of the correspond-
ing path intt!gral for tbe evolutíon operator by steepest descent. We first assume a
generic analytic potential, decre ....ing fset enough for large a:rguments. We consider the
hamiltonia.n

Fonuard srotlering. We =ume that the energy is larger than the maximal value of
the potential so the.t class.ical forward scattering is p<=ible. We then solve the equa.tion
of motion. Integrating once we obtain:

witb the bounda.ry conditlons

x(r') "'x', x(r") x".

We set X = x" - x', T = r" -r'. We know from the general analysis of Sect!on 2.6 that
m traje<:tories which correspond to classícal scattering, witb X /T = k
we are interested
linite when T'- -oo, -r" - oo. In tbese limits the bounda.ry condition

•" dx
T=
1r• .jK.:i- 2V(:t)
•
44.2 Exerrises: Som-e Sqlutions 939

leads to
k + _!_ 1+~ d" (..jk•- k - 1) +o (r-') .
T -~ 2V(:z:)
The action of the trajectory is then

A= ~" 2 T- 2j V(x)M = !~<2 T 21'" y'r;:

r'
V(x)dx
2 - 2V(x)

= !k T
2
+L:"" dx ( -./k2- 2V(x) k) +O (r-l).
We then calculate the Fourier transform. The result only depends on z" x' and thus
a factor of momentum comervation li (k" k') follows. The remaining integral over X ís
calculated by steepffit descent. At leading order for T large we need only ronsider the
terms of order T for the saddle point. We find

X k'T ~ k'=k.

As shown in Section 2.6 the terms proportional to T cancel with the free motion factors,
and we finally Clbtain the semicla.ssical result.

In the special c""" of the hamiltonian (44.2) a short calculation yields for lnS_.(k) a
result in agreement with the expa.nsion of the exact result.
Backwarà scnttering. We now assume that the energy is lower than the rniiXirnum
value of the potential in such a way that a classi<:al refiection is allowed. lf we imagine
starting from -oo we obtain after a similar calculation the classi<:al action:

=
where xt. is the rellection point k 2 = 2V (xo), and k (2;~;0 x'- x")/T.
Since the result only depends on the combination x' + :r:" we find after Fourier tra.no-
formation as expected a factor ó(k" + k'). The rernaining integral over X x' + x",
calculatad by steepest descent, yieldl;:

2xo-X=k'T, k'=k,

and thus

The applicatlon to the hamiltonian (44.2) again shows agreement with the expansion of
the exact expression. Note finally that the results ín the forbidden region can be obtained
by a proper analytic rontinuation in the potential.
940 Exercises: Some So!utions 44.3

Chapter 3: Path Integrais in Quantum Mechanics: Genera!izatian

Exercise 3.1
For levei n and angular momentum !, setting J.i I+ vj2- 1, one finds

Exereise S.!!
We shall e:xplain the idea in the case of the lu.grangi!Ul

C(q,q) = ~9ij(q)q'<f.

The short time evolution operator can then be written

U(q", q') Af(q") exp [~9ii(q") (q"'- q'') (q"1 - q'i)} .

We then use the unitarity relation

f dq' U(g'',q')U.(q, q') = If 6(q'- q"').

We use the property that for c small q' can be rep!B.Ced by g or q" in 9ii and Af. The
remaíning integrei over g' is then gaussian. One linds

The quadratic term in q' has cancelled, and the resulting integration }~elds a 6-function:

f dq' U(q",q')U· (g,q') = Jf'(q) det-l (Y<j(qJ} TI 6(q'- q'").

i

Unitaríty thus ímplies

Jf'(q) = det(g;;(qJ},
as we have found by direct calculation.

Exercise S.S
If we start from the discreti:red free evolution operator we obtain of course the action

with a measure nk
f'(qk)dqk·
lf we change variables in the rontinuum path integral we find the actiou
44.3 Ezercisuc Same Solutiuns 941

which we could imagine discretizíng in a hermitian way as

with the same measure as previous!y.

Let us thus calculate the dilference betw..en the terms in So and S,. We find

s,- s, = 2: ;t.í9>+• q.).f'(q•)f"'(q.) .

•
=
We that since Qk+l - q• 0(~'' 2 ) this term is of order é, i.e. comparable to a potentia.l
term. Returning to the initial variables and treating this dilference at first order in
perturbation we find that the two actions dilfer by the potentia.l

f
s,- dtq~{t) ~-f dt/"'(q(t))[/'(q(t)Jt",
which exhibits the dilierence in the order of operatol10 in the two QUantization methods .

.&ercise 3.-/
3.4.1. Let us first consíder the effect of" translation on the evolution operator. In the
path intégra.l we change variables, setting

q(t) ...... q(t) +a,

where a is a const.ant vector. The tmundary conditions then become

q(-.0/2) ~ q' a, q(iJ/2) = q"- a, .o t"- t'.

The action receives an additive contribution !ieB >< a(q"- q') dueto the magnetic fie!d,
which leads to the relation

(q" IU(I3)iq') exp(!ie(B >< a)(q" -q')l (q"- ajU(il)lq'- a}.

We cnn use this relation for 11 = q' to simplify calcula.tions. The additiona.l phase then
becomes ~ieB · (q' X q"). We have to ca.lculate (q"- q' jU(.O)I O).
We thus solve the classica.I equation of motion. The motiOn along B being free, ""'
restrict ourselV€8 to the plane perpendicular to B. We find

eB
8 c~ 4tanh(eB/9/2m) ( , ')2
q -q ·

We then evaluate the determinant resu!ting from the gaussian integration. We have to
calculate the eigenvslues of the differentia.l operator
942 Exercises: Some Solutians 44.3

with the boundazy conditions that the eigenfunctions vanisb at ±/}/2. After Fourier
transformatíon thís operator ta.kes tbe form
rrn.• 2
( weB
-weB)
m.w2
'

where w í.s the conesponding frequency, The eigenvalues Án then follow:

n 2 1r2 /{P + e2 B 2 /4m2 ,
>.n =
each eigenvalue being twice degenerate, Tbe product of eigenvalues divided by tbe value
in zero field then follows from the identíty (2.35), Norma.llz.ing witb respect to tbe free
haroiltonian one finally <>btaí!ll<:

(c{' IV(/J)I q') 4.,. sinh~;Bf 2m) exp [~ieB · (q' x q")] exp!-S.].
3.4.2. The classical action theo becomes
Se = ~
2smb(Jw'
[{ q' 2 + q"2 ) cooh /}w' - 2q'q" cooh (eB/})
2m

-2isinh ( '2~) B (q' x q")],

where
B= BJB,
The gaussian íntegratí<>n then yields the factor N
eB
N = 4" sinh(/}w')
B:tercise S. 5
Let u• first calculate tbe classical acti<ln of the va.rious trajectories which satisfy the
proper boundary conditions. lt is possíble to go from q' to r/' directly or after being
rellected an arbitrary number of times by the two walls. One find two series
So (q" r/+ 2n1r) 2 /2T, Se (q'' - q' + 2n" ) 2 /TI',
After a shift of the classical trajectory the remaining gaussian integral Í5 independeo< of
the path a.nd yields a constant normalizatioo. If we would just add the two ..,ries we
would find

U{q",q';T) (21tT)- 112 (~:el(o"-q'+>n-.)'/2T +eí(q"+q'+ 2"•)'/2T).

The sum can be evaluated, as in Sectíon 3.3 using Poissou's formula nnd the remlt Í5

U(q",q';f}) =2::::-e•Tk'/2 (e•A(o"-•'1 +eík(o"+o')). (44.3)

k
This sum does not coincide wíth the exru;t ~ult wbich can be written
U(g",q';/}) I::•'n'o
(ék(o"+o') e•t<q''-<'l).
k
The eigenfunctiom ,P( q) corresponding to expression (44.3) are cos kq and tbus satí.sfy the
boundary conditions .P'Io-o .• = O while tbe conect eigenfunctiom satisfy .Plo-0.• = O.
The correct result is obtaíned by multiplying the amplitude by a sign for each reflection.
This phase factor which can only be obtained by a CMefui study of tbe limit of a smootb
potential converging towa:rds the square well using tbe eemiclassicol expressions of Section
A38,2,
 44.3 Ezercises: Some Soiution.o 943

&ercise 9.6
3.6.1. We titst write the equation of motion

i(t) + w.z(t) = j(t),

-í(t) +wz(t) = ](t).

The solutions are

.z(t) = ze-w(t-t') +i',. dr .,-w(t-r) j(r),

i(t) z"e-w(t"-t)+ 1,u dr "W(T-IJ](r).

Taking into account the equatíon of mution the classical action simplifies

Se= -z''z(! 11) - ["' ](t)z(t)dt.

],,
l We then substitute the explidt solutions. Artes shifting .t(t}, z(!) by the cl"""ical solution
we ha""' to perform a gaussian inl.egration which depeods only oo t', t'~. It )ields a
normalization which ean be directly inferred from the grolllld state energy.
3.6.2. It ill convenient to introduce the notation

a= j dt e""' j(t), ij = J dt .,-'-'~ ](t).

The unitarity relation reads

J -dz'dz'-
'. S(Z',z) _.,._.
.-S(z,z')e-u
2;.- .
We note
2 Re U.:"" dtd1' J(t)B(t

We then ha'lll! to eva.luate

dz'dZ'
K(z,z) =
J ~ exp[zz'- i(itz' +ai)- z'z' + í (üz+ ai')+ Z'z -i1aJ.

This a simple ga.uas.ian integral which leads to the result.

3.6.3. In the tirst cose the interaction term yields a contribution to the action

p. j dt (z(t) + z(tl) 4 •

At this arder only the term proportional to (iz) 2 contributes. Therefore, using Wick's
theorem, we find
944 E:r:ercise.: Some Solutions 44.4

where (:Z(t)z(t)) is given by the propagator (3.77) at time O.

(z(t)z(t));;; ti.(O) = 11 +•-D

e-ll/1(0).

With the convention 8{0) "' ! we find

which coincides witll the result o f exercise 2. L L

ln the second case we note quite generally (as a consequence of Baker-Hausdorf's
formula)

lt follows
4
(a +a') = at• +a4 + 4a!u3 + 4a' 3 a + 6a + 6at + 3 + l2n'a + 6a1 2 a 2 •

We now replace the quantum opera.tors a, at by the classical variables z, f. At this order
only the terms with ao equal number of a and ai factors coutribute. We then obtam

e-flH - 1 [1 ~.l.iCJ>. (1 + 4t:.{O) + 4l:. 2 (0l) +O(>.')].

tr - 2sinh(wiJ/2)

We thus !ind that to rocover the correct result we ha"" to assign to 0(0) the value
B{D+) =O.

Exercise 9.1
We first calculate the norm of .P( z) and find e~". We can then repi1'Sent the position
and momentum operators by

A short c:a.lcu!ation then yields

Chapter 4: Stochastic Dilfe:rential Equations

E:r:ereise J.l
To derive coupled differential equ ations for moments we differentiate the moments witb
respect to time and use the equation.

(44.4)

Setting (xk) = mk we get

mk"" -~akmk+n-l +k(xk-l{t)"(t)).

 Eurcises: Some Solutions 945

To evaluate the last tenn we use the same stro.tegy ,., in the derívation of the Fokker~
Pianck equation and find

Therefore
1Ít4 = -}akmk+n-1 + 1k(k- l)flm•-2·
ln particular if we assume the EXistente of an equilibrium we find the relations

wbich determines ali moments once ma:, k lt ···~ n are koown. 4.1.1. The Fokker-
Planck reads:
F(x,t) (44.5)

The possible equilibrium distribution is the static solution

P(x) oc exp (- Q(nn+ 1) xn+l).

For P(z) to be =rma.lizable n mu..t be odd 8.nd a positíve. Ttansforrning the FP hSIIIii-
tonian into a hermitian harniltonin.n one ~ifi~ ít hiiS a complettl discrete specuum and
the oonvergence towards equilibríum is exponentia.l.
4.1.2. What cn.n be sa.ld if the parametel a a.nd x(!) are complex and v(t) is still the
real noise (4.2). Begin with n"' 1; for n > 2 this is a. research problem.
For n 1 we can solve the Langevin equatíon

z(t) ~ x(O) .-••1' + 1' dt' v(t') e•t•' -t)/l.

It is clear the.t the condit.ion for the existence of an equilibrium distribution is Re a > O.
Then ali avera.g~ are determined by the two-point function.

Exerri.5e 4.2
The solution of the stochastic differentia.l equation can be written

wbere r 1 , r 2 are tbe solutions o( the equation

First it is c!ea.r that an equilibrium ca.u he rea.ched only if both r1, r 2 h ave a. posítive
reo.l pa.rt, which implies bja > O, cfa > O. Then o.t large times we cà.n forget about
the boWldary conditimis at t O, or equiw.lently we can integre.te the equa.tion frorn
I= -oc.
1. The two-point function. ABswning that the gaussia.u noise distributlon is normali2ed

{v(t)v(t')) = 26(1 t'),

946 Exen:ises: Some. Solutian...c; 44.5

we can calculate the two-poínt correlation function:

(x(t)x(t')) = 2 21 2 (}_ e-r,lt-t'l _ _l:_ e-•,1<-<'1). (44.6)

a (r2 - r1 ) r1 r2

Note that here, in contrast with tbe ordinary Langevin equation, also the (:i:(t)x(t'))
correlation function is continuous at t = t! and thus the typical paths are smoother.
2. The equi!ibrium distribution. The equilibrium d istribution ís of courSe gaussian ín
the two variables :r:(t) and i:(t). lt is thus characterized by

{r 2 (t)) = lf [a 2r 1n (r1 + r2)] 1/(bc),

2
{:i:2(t)) 1/ [a (r 1 + r2)] lj(ab).
(:i:(t)x(t)) =- O.
9. The autocorreiation time. We observe that tbere are two pol'<lible situations:
(i) &th r 1 a.nd r 2 a.re real and then at large times the largest value gives negligíble
contributions. The second order díHerential equation can he well appmximated by an
ordinary Langevin equation. The autocorrelatíon time.,. is given by

b+~
r=
2c
(ii) r 1 a.nd r 2 are complex conjugated. Both terms then contribute and the behaviour
ditrers from the ordinary Langevin equation. The autocorrelation time .,. is tben

r= 2afb.

(íii) The border case n r2 is spedally interesting and corresponds to criticai de.mv-
ing. At b,c fixed (and thus o.lso (x2 (t))} it leads to the fastest decay of correlation or
rela:xation towards equilibrium:

Chapter 5: Functionallntegrals in Field Theory

Exercise 5.1
5.1.1. The field equation of motion reads:

[(-ll~ + m 6 J~x) + ~ ( 6 J~xJ ') Z(J)

2
) J(x)Z(J).

5.1.2. Calling zCn) the n-point correlation function we 5UCC<!S5ively obtain

(-&, + m 2 )zlll(x) + ~ 9 z<•l(x,x} =o,

(-a; +m 2 jz< 2l(z,y) + !gZ< 3>(x,x,y) ó(x -y),
(-a;+ m 2 )z(3l(x,y1 ,y2 ) + ~gz<•l(x,x,y 1 ,y2 ) ó(x y1 )Z(Il(Y:!)
+<~(:r y,)zPl(y 1 ),
(-a;+ m 2 )z<•>(:r,y~, l/2,!/a) + !gZ15l(x,x, li~>li>,Ya) =- é(x y,)Z( 2l(!l:!, Yl)
+é(x y 2 )Z 12l(ya,!/ll + é(x -yz)Z<'l(y,,yl).
44.5 Exercise.o: Some Solutioru; 947

5.1.3. The first equation yields Z( 1 l(z)

which by tran51ation ínvarjance ís constant from
the 2-poínt function at leading arder. Denoting by ó(x- y) the prop~ator we obtaín

Fram the last equation we e>btain the 4-point function at !eadíng order, which is, a.o;
expected a disconnected sum of products of 2-poínt functiol'l5 (property of the gaussia.n
measure).
zC•l(yJ,!/2, !13, !14) ""Ll.(y1 - !,12}LI.(y,- Y•) + 2 temlS.
We introduce these expressions in the equation for z<3 l, and find

Z( 3l(llh!i2•!13) ~g f <b: ó(:t !Jt)Ó(:t -112)LI.(x !12)

- _!,ó(O) (ó(y 1 !,12) + 2 terms) .

2m•

We can now calculate the two-point funt;tion

2
Z( l(lll !lz)=ó(yl~!t.z) !g/dxó(yt-:z:)z<'l(x,:r,!/2)

= ó(yl !t.z) + !g2 f dxdz ó(!l! 2

x)ó (x- z)ó(z- !12)

+ 2g2
m2
f dx ó(!lt - x}ó(:z: !12) + 4g> 2
m 4 6 (0)

Exercise 5 .~
5.2.1. The equatíons of motion are

[9J+M +9 6J~:z:)] 6 ~x)Z(IJ,'1,J) 'l(x)Z(i!,'I,J) (44.7a)

[rH~+M)+gé~:z:J 617~:z:)Z(ij,IJ,J) =fi(x)Z(i),'l,J) (44.7b)

[<-&; + m•) óJ~x) + g 6'1~x}líl)~x)] Z(i), 'I,J)"' J(x)Z(i!,'),J), (44.7c)

where T means transposed.

5.2.2. Setting J 'I ij =O in (44.7c) we immedíately obtaín the ,P..field average:

(-<f.+ m') {~(x))- 9 (t);(:r)t/>(x)) =O,

a.nd thus, beca.use (~(x)) is x independent (translation invaria.nce)

(.P(x)) = ~tróp(O).
m
If instead we first di!Jerentíate with respect to J we find

(-~ + m 2 ) (.P(x)ql(y))- g (.P(y)t)}(:.)f/!(:r)) 6(x- y).

948 Exerdseo: Som< Solulioru 44.6

Differentiaüng (44.ía) with respect to 17(Y) y:ie!ds

We a1so need equations (44.7c) differentiated ..,,.;th respect to~ and ij, and (44.7a) with
respect to ij and twice '1

(-8~ +m 2 ) (<J>(x)tP(Yd.P(!12)} g(,j;(y,),P(!12).,I;(x),P(x)) =O,

(~. + M) (oi>(Ya),P(x).Ji(y,},P(!12)) + g (t,P(x).,i>(!12).P(x).,i>(y1 ),P(y,J)
ó("- !ill U'lml.P(!12l) ó(:r- 11d (.j;t!13l~,(y,J) .

In the last equation we set g O, replsce the fermion 2-point íunction by the propagator,
and ohtain the fennion 4-point function at leading order (in the gaussian approxímation).
From the previous equation we then find the <P# vertex

We then substitute in the equations for tbe 2-point functions. The fine.! answers are:

(t,P(:r:).P(!I}) = Ãs(x -y) + (t,6) 2 l j dz dz, Ll.e(x- z,)

1

X tr [fl.F(ZJ >2)fl.F(z1- >tl]f1B(Z2- !1),

(.,i>(y)tJI(x)) = Ll.F(x- y)(l- g (<,6)) + g2 f dz1 d%2Ap(x z,)6F(z 1 - >2)

X Áp(Z>- 1/)fl.g(Z! - Z2),

where 6e and Ll.F are the boson and fermion propa.gators.

Chapter 6: Generating Functions of Correlation Fnn<:tions

Exerr;ise lU
The field equatíon of motion tflads:

[(-a"+ m') l!J~x) + ~ (6J~x)Y] Z(J) J(x)Z(J).

For the generating functional of connected correlation functions it becomes

Fine.lly for the lPI functional we find

44.6 Exetcises: Some Svlulions 949

lf we Omit the inv..n;e two-paint function we ca.n integra.te and obtain r in the trce
a.pproxímation:
r(;o) S('f').
Then iterating the equation we obtain the derivatíve of the one.loop functiona.l:

Expandíng this oocpression in powers of 9 we find

kg~(O) !92 f dy~ (x-

2 y)l(J(y)

+ !93 f dydz Ll(z y);>(y)6(y z)>p(z)~{z- :r)+ O {94),

where we have called L> the propaga.tor. Compa.ring then with the expansion of the IPI
functional we conclude

r~'l = !gt>(O)
r(2l =~tg262(z-y)
r('l = l6(x !I)~(Y z)~(z- x).

Note fiually that in expn,..;ion (44.8) we recogníze the functional derivative of the ex-
pected tr In:

&erdse 6.2
The equation for the complete correla.tion functions has been given in Chapter 5. For
the connected functions....., obtain

[
2 2 ~
{ -fJ + m ) !JJ(:r:) +
g (
6
6
l!J(x)
)a] W(J) +
g (
6
6W )
óJ(:r.}
3
g óW
2
ó W
+ 26J(x) (6J{x}) 2 = J(x).

For the tPI functional we finally obtain

where the three-point connected function Wt'l(J) has the symbolic form
950 Exerci>es: Some Solutíom 44.9

One easily verifi"" that the last term does not contribute at one--loop arder. Then, aftet
integration 1 ane finds at tree order

r(<p) = S(<,o),
and thus at one-loop order

Jr,,.,..,0
O<p(x)

which, after integratíon, yields tbe general tr ln term

rlloop

E:urríse 6.9
The equations of motion for the partition function are given in the exercise 5.2.1. From
them one derives

ér !'l
ó.p(x) = I~+ M + gr/>(x)l ,P(x) + gWJ~ (x,x)

óJ'
ó<)>(:r)

where we ha"" defined

2
(2) li W (2) _ 62 W
wJfl(.:r,v) = 6J(:r)6fi{v)' w"" (x,y)- Ó7](.r)ófj(y)

The second functional derivatives of W form an operator and a 3 x 3 matrix in the

{ J, 71, ij} space, inveree ofthe analogous matrix for r. The explícit expr=iona in terms
of r are to o complicated to be reprod uced here.

Cbapter 9: Regularization Methods

E:z:ercíse !1.1"
The equatioo reads
2
(-ai+ m)
2 "
liJ;(x) 9 óJ 6(:r)
+ 6N ( 6Jj(x)
6 ) ]
Z(J) = J;(:r)Z(J).
[ 0

9.1.1. Using the O(N) invariance (BS5Umed not spontaneously broken) we obtain

(-&; +m 2 ) z{"ll(x y) + 6~.z~1i(y,x,x,x) = ó(x y).

9.1.2. lfthe a.ction is quaclra.tic, the four-point function roduces to the sum of products
of twp..point functions:

z;;l,{Xt,X2,X3,X4) = Ó,j6klz!')(:z;l Z2)z{>)(Xa X4) + 2 terms.

44.9 951

Introducing this relation ín the equation we find:

[<-&; + m 2) + g(~; 2
l z<2l(o)] z!'l(:r- y) = 6(x y).
Weset
11
2 m2 g(N + 2 ) z<'l(o)
+ 6N '
and thus alter Fourier transformation
-p)
z (p) p•
1
+ ll2 => ll
2
m
2 g(N +2)
+ ~ (21f)d
I f" r ddp
+ jll •

We note that this equation is meaningful only in the presence of a cut-off A for dimew;ions
d ~ 2, which we indicate syrnbolically. lt y\elds a relation betwren the "bare" ma.ss m,
the renormalized or physical ma.ss Jl aod the cut-off. Thls relation becomes exact in the
large N limit 85 the analysis of Chapter 29 shows.

Exercise 9.2
Calling '1 and ij the aources for ij; and 1)1 respectively we find:
6 6 6• 6 ]
[ j\t"6>j;(:r) + 9 ófj;(:r) ÓIJk(x) 6flk(x) Z(ij,IJ) = f/;(:r)Z(fj,IJ)·
A relation betWeen the two and four-point functions follows by differentiating with respect
to 11;(11} and then setting the sources to zero:
fJz (.p.(:r}rh(v)) + 9 {t,b;{x)ib;(y)IPt(x)ij;k(x)) = -6(x y)6;;.
Again assuming that the functions are calculated wíth a gaussian measure we can fa.;-
tori:re the four-point function
( l)i(x)ib;(Y)tPk(x).ti~(x)) "' ( l)l,(z)ib,;(Y)) ( tPk(x)ib•(x)) ( IPt(x)!b,; (y)) ( l)li(;,;),Pi< (x)) ·
The function < l)l(x)!j;(:r) > being a constant, we can solve the eqnation. After Fourier
transformatíon, and ta.king into nccount U(N) invariance, we find

(IP;(phh{ -p)) = 6;j í,l ~ .u ,

w here we have set
g(N 1) -
p= - N - - (IPk(x).P<(x)) = g(N
We have here assumad that the regularization was chiral invariant. We see that this
equati<>n, called often the gap equation, has always the solution Jl O which preserves =
the chiral symmetry. However, m d > 2 dimensioD8, when the coupling satisfies

g>g,, 9,
_
1
tr 1
(N-1)(21r)d
JA 7'
ddp

i.e. is sufficiently attra.ctive, there exists another solution wlúcb bteaks the chíral sym-
metry
(N-1)trl !A~
g (21r)d pZ + 112 '
and wbich, in addition, hss a Iower action. Within tbis approximation scheme, this
mode) thus provides an exrunple of non-parturbative spontaneous syrnmetry breaking
and fermian mass generation, the Narnbu-Jona-Lasiltio mechanisrn, dueto ferrnion pair
condensa.tion. Here again the appraxlma.tion can be justified only in the large N limit
(see Appendix A30).
952 Exercises: Some Salution.; 44.11

Chapter 10: lntroduction to Renormalization Theory

E:tercise 1U.l
One verifieo tb.at the reoult ís
r!2l , c • g,
,,... "'P+ 1Pl-l12g,!n(pfm,)

Thís eJCpression hBB a pale when the momentum reache:s the value

(for 112 > O) often called the Landau po!e. Tbis singula.rity .bows that even if g, is small,
perturhation theory will break at sulliciently hlgh momentum. To go beyond this simple
rem,.,-k, more sophisticated RG arguments are required (see Chapt.er 25). Conversely if
in some other tbe<>ry 112 < O then perturbe.tion theory breaks down at sufficiently low
momentum, indicating that the perturbative prediction of a massle:ss spectrum is not
reliable.

Chapter 11: Dimensional Regularization and Minimal Subtraction

&ercise 11.1
Let us take tbe example of the coupling comtant renormalization. Then tbe bare ~oupling
constant g is re!ated to the renormalized coupling constant by

g g, + lnAjm,G2(g,) + (lnA/m,iG,(gt) + · .. ,
where Gn(g,) is of order g~. We can then calculate the iJ-function by differentiating wíth
respect to m, at g, A fixed

Since the {3 function is finite we conclude

In analogy with dimensional regularization ali qua.ctities are determined by the coellicient
of In( Afm,).

Exerci•e 11.1?
Pseudoscalar Yukawa interoction. Consider the action

S (ri>, ,P, ;j;) j d".r{ -;j;(x) li!+ M + ig-rsrf>(x)) 1/>(.r) +! (o~oi>(.r)) 2

+ ~m2 ,P2 (x) + ~,\oj>4 (x)}. {44.9)

in which rp is a pseudoscal,.,- boson field and t/>, ;j; Dirac fermions.

44.11 Exen:i.ses: Some So!utions 953

The factor i in !Tont of "IS is impo.;ed by inva:riaru::e under reHection hermiticity, i.e.
euclidean hermitian conjugation foll~ by sp..ce reHection (see Appendíx A5.6) which
implies:
1111 (igtf>(x))* 11 = -rsig.P(x).
In addítíon the action ís reftection s~"lllffietric (see Appendix A5.5):

,P(x) .... ,,,P(z),

iJ,(x) ,_. .]l(:é)l,,
.P(:z:)- -p(:é),

with:é (-:r 1,x,, ... ,xd)-

Tbe transfarmatian of p(x) is imprn;e<l by the factor "'s
in front of the .P'b.P interaction
and justifies the denomination pseudoscalar for tbe ficld p {x).
Calculations can be clone using dimensional regu!arization and mínima! subtractian
because it is never nec"""ary ta evaluate any trace af products of "'S
by an even number
of "'~• matrice5.
11.2.1. One finds in the notation of Setti<>n 11.8,

where we have set:

(44.11)

11.2.2. The divergent part of the 4> 2-point function is:

(44.12)

11.2.3. We cal! Mo, mo, go, ~ the bare pararneters, M, m, g, À the renormalized
parameters, J.t the rmormalízation scale.
Substituting in the Feynman diagrams and identifying the dívergent parts we obtain
equations for the renormaliz.ation constants:
954 E:r.err:i.ses: Some Solutions 44.11

Jt follows that:

g' (44.13,)
Zop 1 - N•2;_ +O (2 loops),
1
2Ng (44.14)
Z~ 1- Nd-~- + 0(2 loops),

go =- m' /> g ( 1 (2N+3)g')

+ Nd--- - - ; + 0(2 loops), (44.15)
2

.\o=- m• [À+ ~d G.\ 2

+ 4N J,.g
1
-
4
24N g ) 1
+ 0(2 loop5), (44.16)

~ m
2
[1 + ~d ( ~ + 2g 2) 1 + 0(2loops). (44.17)

The {3-filTlctíon. When bare and renorm..Uzed coupling consta.nts are related by ""
eQUation ofthe form (11.115), the coupling oonstant RG functíons j'}, (.\)are given by:

(44.18)

and thus usíng eQUation {11.115):

(44.19)

A short calculation then yields:

/39 • = -~o• + Nt~5g + · · ·,

4 (44.20)

{31 -~.\+N•(~.\2 +4.\g2 -24o4 )+· (44,21)

11.2.4. For M = O let us set

Then the action becomes identical to tbe actíon of the model (11.113). Tbi.s explalru;
why tbe renormalizatíon ronsta.nts are the same.
A few addítional results are:

Mo= M (1- io2 ~4 ) + 0(2 loops), (44.22)

m5 =- m [1 + G>. + 29') ~d]- 4M 2 l~d + 0(2 io0p5).

2 (44,23)
44.11 Ezereises: Same. ~olutions 955

Exercise 11.3
The general problem is to prove that renormalízed correlation functions have a finíte
!imit when e = 4 d goes to zero.
Dimensional analvsís is modífied and the dimensions of correlation functions is now
!r1•nJ = d- n(d- 2J/2- 21.
To define a dimensionless coupling constant we have to modify the third renonnaliza-
tíon condítion ( 10.9)

Wíth this delinitíon the roefficient functions of th" CS equations are only functíons of g,
nnd E.
Equation (10.33) ís modified and becomes

Th take care of the additkmal term in the iJ-function it is useful to define

ln the prevíous equation only Í3 appean;.

Then we keep in the r.h.s. of our inductíon equation (10.35) Band add the remaining
part in the th.s.:
m _!!_
[ r-amr
- cg ..!!..]
TÔ9r
f(l,n)
7
•

Wíth this modí!icatíon many arguments remain valid. Let us solve for example the
equation
â â ] r,(<) -- • (4) .
[m, Ôfny -cg, âg, m,g,f (p,jfny,g,e),

where the Lh.s. is assumed to be /inite when E - Q and satisfies

1 â
m,
-t
~ ôm (g,m,<) =O.
g, t

The íntegration of the equation then yields

The integral converges at p "" O becauoo Jl•) is expandable in powers o{ g, and is of arder
p' for p small. Hence no singularity in E is generated.
956 Exercise<: Same Soiutions 44.12

Chapter 12: Renormalization of Composite Operators

Exercís• 12.1
12.1.1. Let us first note that, unlike in the rJ>4 theory, the action is no Jonger reflection
symmetric in rJ> and the rJ> has an non-vnnishing expectation value. lt is thus conrenient
to shift rJ> by a constant in such a way that (4>) O. The action then takes the form

In six dímensions the field rJ> ha.s dímensíon two and the lPl functions r(!,n) have dí-
mension 6 2n - 2l. Power countíng then tells us that the renorrnalized action ba.s the
form
2
S,(tl>.) =f d x [ ~Zq, (ô~Q1,)
6
+ ~(m: +óm )rJ>: + ~g,Z9 4>~
2
C,P,] .
The re!ations between bare and renormalized quMtities are then

Possible renormalization conditions are (Ql,(x)) = O aDd

f} 2 l(p =O) m~,

fJ'(2)
&r r, (P)Ip=o = 1,
f~al(o, o, o) = g,.

The other primitively divergent functions are r!l.ll, r<l,2J and r<>. 1l.
We now write the equntion of motion of the ba.re theory. We define the insertion of
the 4>2 operator as the correlation function of -!4>2 • Then we find

~ 2 6Z
( -<T + m ) 6J (x) - gZ"'<•í (J(x) + c)Z(J).

The equation for W(J) is tben

~ 2 5W
(
-<T +m } 6J(x) gW~'í•) = J(x)+c.

After Legendre transformation we obtain

<lT 2 1
ó,P(x) +c (-8 +m )Q>(x) + gr.,'(•J(rl>).

Tbe equation for the renormalized IPI function~l i5 then

44.12 Exercise.: Some SoiutionJ 957

This equation relate. the renonna.!ization oonstants due to the insertion of the ql(x)
2
operator to the renormalization constants already present in the action: .P (x) is not
really a new operator but ís related to ,P(x). Introducjng the notation (in Fourier space)

we can define renormalized correlation functions with one insertion as

which implies
r~n+ll(q,p,, ... ,pn) = g,rp,nl(q;P~>· .. ,p,).

In particular for n =2 we find a renormalization condition for the function r!'·2 l

rP·'l(o,Q,O) = 1.

For n =- 1 instead we set

which oorresponds to

and the renormalization conditions

12.1.2. The relations between nmormalized and bare functions lead to CS equatíons.
Delining tbe operator

ô a n
Dcs = m, ,._ + .Ll(o,)-;;-- 2fl(g,) -i'12(!1r),
Ulr&r UUr

we find for n > J:

valid except for n = 1,1 = 1, 2. Using renormalíza.tion conditions one imme<Hately finds
u(g,) = 2 - 'l(g,). Moreover, since we have related the renonnalization of q,2 to the
renormalization of 1/> and g, the corresponding RG function '1•(!1,) ís related to f](g,) and
.Ll(g,)
'12(!1,) !fJ(g,)- .Ll(g,)jg,.

A simple algebra. yields the CS equatians for the renuúning functions.

958 Exercises: Some Solution.s 44.12

Exeri:ise 12.2
We start from the action (11.3) in the mínima! subtractjon scheme to establísh the
relation betw<;en dimension four operators. Following the method of Sedíon 12.2 we find

-z.,r<I>(•JB'<t>(•J + Z2(g)m'r"'(•l + ~gm' z.r o•(•J = <t>(x) ~!~x). (44.24)

By definition of Zz(g) the ínsertion of Z2r/>2 , where we reoall that rp is the renormalized
field, is finite
l<t>'(x)], = Z2(g)rp 2 (:r).
Since the r.h.s. oi equa.tion (44..24) is finite we immediately conclude

Dlfferentiating with respect to g we obtain

~: = ~Z~(g)r!B.<I>l' + !m'~(g)f~, + ~m'(Z.(g) +gZ;(g))r q,•. (44.25)

Tbe J.h.s. is finíte. We conclude that the ínsertion at zero momentum of the sum of the
operators in the r .h.s. is also finíte. We can thus define

[4>4 (x)j, (Z9 (g) + gZ;(g))q14 (:~:) 12m-•Z~(g)<!i(x)a1 q.(x)

2 2
+12m - ' Z2(g)4> (x) + m-• A(g)ô2 <f}(r).
Note that we cWJ also diff.,..,ntiate wíth respect to the mi>S5:
ar
m- (44.26)
am
Equation (44.25) ca.n be rewritten
ar = !Tl(Y)Z<~or<a.<>}' + ~m (T)(g) + '12(9))~(g)r~~>' +
1
41 m'(271(9 )- E)gZ9 r ~·
2
,!l(g)
89
By integratíng equatíon ( 44.24) over :r and eliminating the dimension four operator inser-
tions between (44.24,44.26,44.25) we recover the CS eqoations for correlation functions.
Moreover' we Cllll then directly c:alculate the elements ar the 2 x 2 matrix which ap-
pea.rs in the CS equations with the insertion nf the two operators of dime1111íon four. In-
deed the renormalized (â,,P) 2 operator at xero momentum from (44.24) just corresponds
f dx ,P(x)6/6,P(x), i.e. to the multiplícation of r<nl by n, an operation whích coromutes
with the CS dífferential operator. Thus the corresponding matrix elements are zero.
For what concerns 1)14 at zero momentum, from (44.25) we see that it ís equívalent to
m -c â / ôg. The commuta.t;on of thls operator with the CS operator Dcs

yields
[Dcs,m-'8/ôgl =-c fJ'(g) + 4nr/(g).
The diagonal element is thus -E fJ'(g) o.nd the component on tbe operator .p6flup n
is ~Tl'(g). >e re
44..13 E:rercises: Some Solutions 959

Chapter 13: Symmetries and Renormalization

E:rercise 1S.l
The action cao be wrítten

S(M)= f ct•x[tr(&,.M'&.M+m2 M 1M+411(M1M)

2
) +~g(trM 1 M)
2

- u (detM' + detM)]. (44.~7)

An interaction proportional to det M is allow..d because for a 3 x 3 matrlx it is equivalent

to .a tubic interaction, owíng to the identity:

6detM --o (trM) 3 + 2trM3 - 3trMtrM2 •

Note tha.t for the last two terrns in (44..27) we have made a conventional choice of global
phase for M. When u (which we ca.n choose positive} vanishes the symmetry group
beromes ínstead U(3) x U(3) (cf. the U(l) anomaly). Finally since
2 2 2
tr (M'M) :::; (trM'M) :::; 3tr (M'M) ,

Wl! ha"" the restrictíons or g, h p<»itive or if one ís negative

h- 3lgl <:O, or g lhi<:O.

Spontaneous "1/mmetry breaking. A general complex 3 x 3 matrix M can be written

M=e16i 3 UlAU.,

where U1 aod U, are two SU(3) mntrices, andAis diagonal with positíve matrlxelements
À;. Only the detenrunant<; depend on the angle 9:

detM' +detM = 2coslldetA.

For u > O the minímum of the corresponding term ís at (! = O. Mirúrnizing the potential
with respect to À; we then get

We now assume that the parameters are such that the equation has non-trivial roots.
~ then find two cases
{i) h> 0: At detA, tr A2 fixed, the equation h"" only one positive root in À~, and sínce
À; > O all three À< are equal, >.; À, the matríx A ls proportional to the ídentity and
the residual symmetry ís SU (3):

u;À1U2'"' Àl => UI u,.

Equation (44.28) becowes:
960 Eien:ises: Some Solution5 44.13

We know 3g + h 2:: O and thus tbe equation ha.s pooitíve solutions either if m < O or if 2

u 2 4m 2 (3g +h) <!O.

(ií) h< O: Again at detA,trA2 fixed, if the equation has one positive root in >.J, it
has two, and the >., ca.u take two dífferent values. Let us cal) >. the value which appean;
twice >.' the other one. Thus a new pattern of SSB is possible where SU(3) x SU(3) is
broken down g<merically to U(2):

U\AUo ~h with A= (~o ~o )/~)

=1- U1 U 0 = U with U;;a = Ua. = O for i 1, 2.

Expressing that >.i" >.' we derive from equation {44.28) a simple relation

h>.'+ u =0. {44.29)

(üi) There i.l; an exceptional case when the double value >. vanishes which lea.ds to SU(2) x
SU{2) x U(l). Equation (44.28) tben reduees to

(g + h)>.'2 + m 2 =O.

Eiercúe 19.2
(i) SU(3) ha.s 8 generators. ln case (i) the number ofGoldstone particles is thus 8+8-8 =
8. To identify them we consider infinitesimal SU(3} transformations which do not beiong
to the SU(3) subg-roup. We cBD para.metrize them in terrns of a traceless hermitían matrix
H:
Uo l+í~H.
Acting on the vacuuw expectation value A we get i>.H, í.e. a traceless antihermitia.n ma-
trb<:. The conclusion are that the massless states correspond to an octet of pseudoscalar
particles. There are l.hree types of ma.ssive states, one se alar and one pseudoscalar singlet,
one scalar octet.
Remark. Note tbat in the special case u = O the symmetry group becomes SU (3) x
SU(3) x U{l) and the srune vacuum expectation value leeds to a nonet of Goldstone
pseudopartícles. Since such a symmetry group i.l; generated in QCD by a.ssuming the
quarks u,d·,s m!ISS!ess, thís leads to the famous U(l) problem.
X...t us now consider the m88S matrb<: corresponding to action {44.27). We set M =
>.+(E+ iii)/../2 and expand the different tenns up quadratic arder in tbe fields:

trMIM- trA~ = tr ( -/2>.E + ~]12 + tE2 )

2
tr (MIM) - tr A4 = ). 2 tr ( 2-/2),E + U2 + 3E2 ) + ...
(tr A 2 )~ =
2 2
(trMIM} - >. 2
tr (6../2>.E + 3II2 + 3E2 ) + 2>.2 (trE) + .. ·
detMI + detM- 2detA = ~), [tr (2h>.E +Il 2
- E')- (trU) 2
+ {tr 1:) 2] + ... ,
where we have used for the last equa.tion In det = tr In. By definítion of À the coe!lident of
tr E must Vl!JlÍSb. We then observe that tr II2 only appears in the combination 2;Í2>.E -t
44.13 Exen:ises: Some Solutíon> 961

Il 2 .Thus the ooellident of tr IJ2 also vanishes. The only rewaining dependence in II
ís through (tr II) 2 Thus. as anticipated, the deternúnants gtre a mass to the SU(3)
pseudoscalar sínglet. Finally we note that the scalar fields are all massiw but with a
dillerent mass for the SÍI!glet and octet. Eliminating the parameter m 2 with the he!p of
equation (44.28) we find
(6g + 2h)A 2 - uÀ, mi;, = 2h>. 2
+ 2u),
(ii) We now expect 8 + B - 4 12 mBSsless Goldstone particles. To identify them we
now set
Ut=l-i~H 1 , U,=l+i!Hz,
where H 1 • H, are traceless hermitian matrices, and impose
(Ht)ij (Hz);i for i,j E (1,2} and i= j = 3.
Acting on A we find that the Goldstone particles form a singlet, two doublets and a
triplet of SU(2) for what ooncerns the pseudoscalars, and twu doublets of scalars. Again
we verify these pmperties on Lhe rn8S8 lllAttix. We set M = ), +(E+ iii)/ v'2 and expand
the different tem"' up to quadro.tic order in the fields:

trMfM tr A2 = tr ( VZAE + ~II 2 + ~E2 )

2
tr(MfM) trA 4 tr(2v2A3 E+2A2 II 2 (All) 2 +2A2 E 2 +(AEJ 2 )+· ..
2 2 2
(trM 1 M) - (trA 2 ) =trA2 tr (2v'2AE+ll 2 +E') +2(tri\E) + ···

detMt + detM- 2detA =! detA [tr (2v'2A- 1 E + (A- 1II) 2 - {A- 1 E)

2
)

~(trA-'n) +(trA- E) ] +···.

2 1 2

Using eqs. (44.28,44.29), ooe verilies that ali tenos depending on the n field cancel except
the term proportional to u(tr A-tn) 2 whích corresponds to a SU(2) singlet. In the same
way one checks that the coefficients of the terms IE;3j2, i= 1, 2, vaoish.
(íii) The last case.,...., be tre,.ted in a rompletely similar way. Tbe 9 Goldstone bosons
are:) pseudoscale.rs, a singlet anda (!,O) Ell (0, ~) representatlon of SU(2) x SU(2) and
4 scalars in the sarne representation.

E:rercise 13. S
With the linear symmetry breaking term, the residual symmetry is the diagonal SU(2)
subgroup of case (ü) of exercise 13.2. The matrix C is a diagonal matrix similar to A.
Its addition allows to w;e .\, >.' as free paro.meters, related to C by:
C (m 2 + gtrA2 ) A+ hA3 udetAA- 1 .
One finds for the pseudoscalo.r particles the following analytic e><pressions:
m\'t, =m~ =m2 + g (2>. 2 + .l! 2 ) + h.\2 - u>.',
m~, :m';.: = m! + (u+h>.')(>.'- >.),
m~"' m! +}h (>.' 2 + ~u>.'- ~ [(h (.\ 2 -
1
>.2 ) 2
.\' ) + u>.') 2 + su2 .\2 ] n
112
m~, m; +~h (>.12 - >. 2 ) + ~u.\1 +~[(h (A2 >.'') + u.\') 2 + 8u 2 >.2 ]
962 Exerci.ses: SDme Solution.s 44.14

Again in the limit u ~ O one finds results inconsistent with experimental data since t.hen
m"l m1!" and mK ,....111n'·
For the scalars the masses are
mi, m'; + 2hÀ 2 + 2uÀ 1 ,
mt m~ = m;+ (u+ h>.')(>.+Ã'),
mt± = m; +!h (Ã 2
+ 3À'2 ) + 9 (2>. 2 + À 12 ) + ~uÀ'y ±
with we have defined
A= (3h + 4g)J. 2 uA'
B = v'2>.(2gJ.' -u)
c= (3h + 2g),\.12
Consistency with experimental masses requires
)<'>À, _.!_<~<_I_.
À' u "' À
lf we determine the parameters by imposing the physical values of the pseudoscalar
ma.sse. we find tbe solution
=
À' (À= 1.80... , u)..fm. 10.4 ... , h>.2 Jm; 2.42 .... =
The last coupling constant only appears in the scalar ma.sses and is thus poorly deter·
míned gÀ 2 ;m;- 10. Finally the matrix C has on its diagonal Àm~ ,>-m;, (À+ -'')mk
Àm~.

Chapter 14: The Non-Linear o-Model: An Example of Non-Linearly Realized

Symmetries
&errise 14.1
One first verifies that the spi!.Ce depend<lnce plays no role and can thus be omitted. Tben
IJS &s àS à2 s ] a [as à1 S as à2 s ] à
[ó;, ó;[ [ àw; aw;8H + IJH éiTr;Ülf; ôH + 811; ôHôH + aH aw.aH 811;
- (i,_. j).
Dilferentiating equation (14.35) with respect to "'i and H respectively we obtain
as &s as a•s
a"• &;aH + aH aw,a,,. +H~.;= o,
as aos as a~s
ÔTr; éJHaH + âH 8T!;âH + "'' O.
It follows
[ó;, ó,.) = -Ó;j
To calculate tbe commutatot lóii, ó•] we note that

ó;;S =O,
as =o,
ó,; âH
as a consequence of the O(N- 1) invariaJJce of the action. Also
as a as fJS
ó,; ... -ó.;S = -6<~<-a + 6;k- . 8 "'
8 • 011} 11j
Therefore
[ó;;,ók)
44.14 Exerciões: Some Solutians 963

Exercise 14.2
We consider the parLítion function of the 0(4)- SU(2) x SU{2) oon-linear a-model

(44.30)

with the action

(44.31)

It what follows it wlll be convenient to set c = IJ. 2 f.

We propose to study this model at one-loop arder, to verify in particular that it can
l>e renormalized wítb ouly two renormalizatíou consta.nts Z, Z1 as proven íu Cbapter 14.
It will be conveníent to set d 2 + ~ and to define a dimensionless couplíng consta.nt
g:
r 2 r(l-c/~)i-!'{4,.)-·l• ,_ g.
Dimensional regularization will everywhere he used. 14.2.1.

1
t.,(p) =r?+ 1-'2 '
v!•l = 8~6,,,,~•••• ( (p + P.>)' + 112),
1

V( l = ~]4 ó,,,,ó,.;,ó<.io ( {1lt + P.>) + 1-1 ).

1
6 2 2

14.2.2. The renormalized action rea.ds

(44.32}

in which cr is now giwn by:

Thus in terms of the reuormalized fields !.i 1 gets ao e>:tra factor z,z, V(c) a factor
zz
1
2 and v{•J a facwr 3z 2
1 Z . Moreover 1J. is replaced by !J.'z-
112Zj 1 .

14.2.3. The vacuum co:peetation. (a) is given by

where I ís ao integral which appears everywhere at one-loop

I - 1
-j~= •-•-
(2")" r?+ 1'2 !J.
1
-r(I-d/2)
(4.-)d/2 .

lf we then choose
964 Exerr:íse3: Some Sol.utions 44.14

The !!-poínt junction. The 2-poínt. function ;. proportiona.l to l!;1 which we omit in
what follows. Then

We verífy tho.t the Vlllue of Z obtained above renders the coefficient of 1"2 finite. From
the coeffident of p 2 we find that if we choose

the 2-point is finite and reduces to its tree approximation

14.2.4. For symmet.ry reosons the 4-poiDt function Wi!/o. has the form
(w~p.) <;kt = C(PI,P2,,.,p.)<'i,16., +c CP"""'Pl,P•l 6,k6;1 +c (p~.v•• P:J.P2)<'~ü6kj,
(44.33)
wheJ"e C (pl,P2.1'3,P4) is symmet.ric in theexchangep1 ,... P2 with Pa .... P• a.nd {P~oP2} ....
{p,,p.}. In what follows we only gi"" C.
lt is convenient to introduce the noto.tion

Pi!=P,+Pi·

At tree levei f(•J is immediately obto.ined from the 4-point vertex:

C=-p1 ( Z1 z•P2z+Z• 1 JJ. =-p1 [(1+ 24g)

1
2 •) -2
"" P12+ (
l+z9 2Yn ) iJ •] ·

At one-loop both the 4-point and the 6-point iDteractíons contribute. For whBt concerns
the 6-point vertex we ho.ve just to oontract 2 externai legs. One finds

Then we have the contributions of two 4-point vertices. The following identity ís useful

with
(d- l)S =F Hl +41'') B(p)
T = (~(d- 2)1 + ~ (p2 + 41' 2 ) B(p)j
44.14 Exercises: Some Solutions 965

where we defm!!d tbe bubble díagram B

Not.e the useful reiation

Hd 2)1 = -J,<
2
B(O).
We then find that C gets four contributíons, whir:h ín the pictore of faithful diagrams
correspond to a bubble, a triangle a.nd a square box dimct and crossed. The som of the
two lirst contributioru; is

The d irect bax yields

The terms have been written the way they appe,... in the ealeulation. They ean be slightly
rearranged

The last contribution Í>i simply 63 C(p,,P:z,p•,P:z).

After renorma.lizstion the total one-loop contributlon can be wrítten

expressíon ftnite for ali d < 4.

966 Exe.rcises: Some Solutions 44.14

14.2.5. lt is convenient to introduce some additional notation to ta.ke into account the
0(3) symrnetry. Let us set

r~~l(p) = 6,1G(p),
f!:l (p.,P:<;P>) 6;;V (pl,P2;p3),

in which our c;onventions are that índices correspond of course to "-fields, and in rnixed
,__, correlation functions the argurnent.:l of the ,.--1\elds are placed 1\rst.
In the non-línear u-model we ba'"" the relation

for which we use the parametrízation (44.33).

They are two ways to derive WT identities for correlation functions. One can either
start from the WT identities for oonnected correlation functions which are tbe same 85
for the linear u-model, or from the WT identities for the lPl functional. In tbe latter
case dífferentiatíng respectívely witb respect first to" (and setting H(:t) =c), and tben
"' and H one finds, after Fourier tra.nsformation,

G(O)(n) =c, (44.34)

fC2l(pJfl?(pJ +ó;; = (u)fiJ>to,-p;p). (44.35)

The second equatíon can be rewritten

f< 2l(p)G(p) + 1 = (u} V(O, -p;p). (44.36)

F;nally differentia.ting three times with respect to " we find a relation between 3 and
4-point rorrelation functions:

- (u) tj;~l {0,P:>,P3,~) + f;;(P:<lfL~l (P:<,P4;P:<) + 2 ternu; O, (44.37)

whlch implies
(u) C(O,P:!,!"',P•) '"'G {P:<) V (P>,p,;p,).
Settlng p~ ;,. -m; one recovers from the lwrt equation Adler's consistency conditions

Elímlnatmg V between eqs. (44.36) and (44.37) one derives Weinberg•s rela.tion

The verification af these identities on the explicit expressions ís the straightforward.

A few additional details C3ll be found in
D. Bessis and J. Zinn-Justin, Phys. Reli. D5 (1972) 1313; J. Zínn-Justin, in the proceed-
ings ofLes Houchcs summer schooii971, C. ltzykson ed., Gordon anà Breach, New-York
1973,
wbere the model is studíed as a limít of the linear o-model.
44.15 Exerr:ises: Some Snlution.s 967

Chapter 15: Models on Homogeneous Spaces in Two Dimenslons

Exercise 15.1
Let us fi:rot verify that equation (15.42) is indeed equivalent to equation (15.43):

The only surviving terms corr.,.pond to the most left derivativas acting on D and tbllS

With the explicit form (15.128) we find

Antisyrnm~trí.zing ín a, f3 we find commutators of genera.tors t 0 except for the tertm;

proportiona.l to Ô<Jk/â"•· However we then use equa.tion (15.29). Tbis reduce oga.ln the
ca.kulation to wmmutation relations and we find the anticipated result.

Exerr:i.se 15.1!
Let )(' be another set of coordinates on the manifold. Then the infinitesimal group
transformatíons ta.ke the forrn

We want thu.s to prove that the new functions Ll.~:

satísfy the commutation IE'Iations (15.43).

We now see that the first term ;. symmetric in the exchange a .... f3 and thus cancels in
the wmrnuta.tor, while the sewnd term is proportiona.l to the tenn in (15.43). Tbe result
follows. Actua.lly using the results of Cha.pter 22, one can verify tha.t, as in the case of
the curl of a vect.or, the rela.tion (15.43) can be rewritten by replacing the derivative by
a. oovaria.nt d erivative when tbe Cbrist<;>fi'el connection is Symmetric.
968 Exernses: Sanu; Salutions 44.18

Exereise llUI
Let us ""rify the identity (Al5.7) for the tensor

g'i = D~(rp)J1('1')·

In temu; of g'i it rea.ds

·k a
gil~ -
~
fl1 aD!
o D' âg'
Q ôrp• é!,P ü,P
J:1 (v~él;~- D~&,IY,) + ~ (D~a,~- D)Ji11D!).
We now use the commutation relations (15.43) '1\'e find
O= D~f<>lhD!, + ~J.,i»v;,
whích ;. satislied because the structure constants are antisymmetric in f3 - -r.
Exercise 15...1
The Lie algebra of 0(4) is the same as the Lie algebra oi 0(3) x 0(3). The infinitesimal
transformations of parameters w, w' can be written

lnri. - fiikWj'7T.t + wia, => 6tr = -.c...{"·'~

where we have chosen '!f; and a = ~ as the coordinates on the sphere. We thus
find
"rij =L (t;kl11'1íjkm'll'm + Ô;k11ÔjkU) 6<;- 7r;'lr; 1

•
or calculating the i"""rse
2
9ii "'é,,+ 11'i'tr;/(l - 11' ),
in which we recogníze the metric on S 3 •

Chapter 18: Abelian Gauge Theories

Ettrcise 18.1
Let us díl!erentiate Gauss's law
1 6
iô'M,(:r} 'lr(A) Jd(:r, t)W(A),

with respect to time, and use Schrõdínger's equation

1 ó
ia' óA;,(x) 'H'Ir(A) ii4 (:r,t)W(A) + Jd(x,t)"H'Ir(A). (44.38)

We then use the property that in the abseore of sourre tbe generator of time-independent
gauge transformations commutes with the harniltonlan 1i
[ ~a- - 1i] = ia.J..
6
i '6At{:t)
1
' '

After commutation in the l.b.s. of (44.38) and again application of Gauss's law we find:
-i8;;.'11(A) = iÍd(x,t)W(A),
whích foll"""' from current conservation.
·.·.
44.18 Exercises: Some Solutío~
969

E:xercíse 18.2
The photon propagator reads
(44.39)

The fermíon propagator reads:

{M if)
l!.F(p) = r?+ M• '

and finally there ís the "ÍJ,PA vertex v.:

(44.41)

At one-loop order, in the Feynm= gauge, and for a mass1= vector field, the fermion
2-point functíon iB then given by:

We then use the identities:

Projectíng then r( 2) onto its components

{44.42)

we can replace ~ by pk · p/'fÍ'. We obtain:

Replacing M + í"JI by Z-..(ZMM + ip) we derive the renormaliza.tion constants ín the

mínima! subtraction scheme:

In the notation of Section 18.9 it follows (a e2 /4rr}:

3 l
TJM = 2,.<>, TJ,p=-a.
21r
970 ExercL•es: Some Solution' 44.18

E:rerci5e 18.3
We ncrw consíder the QED a.ctíon (18.26). with a mass term for the photon and quantízed
in a general covarlant gaug€

S (A~, w,
o op') = f d d x ['4 F~"
2 + 21 m 2 A#+
' 2€1 (a.,A.) 2 (44.43)

18.3.L From ídentity (18.81} we derive

k.,r~J(k) = k. (k2 /E,+ m 2 ) ,

k.rL3 l (k; p; q) = erí'l (p + k; q) ~ erPl (p;q + k),
k w('l(k) = ,_k_._
• "" 'k 2 +E,m2 '
(k'!e + m 2
) k.,W!'l (k;p;q} = eWí2 l {p + k; q}- ewl') (p;q + k).
The lirst equation implies
r~~(k) = rT(k) (k 2 li,..- k,_k.) + m 2 6,.. + k.,kvf~ (44,44}

Thís identity fOf" r;;'l can made even mo~ explicit by pararnetriZíng r<'l as:
fí 2 l(p) = A (p') + iiB (p2 ) • (44.45)

where:
A=M+O(o), B=1+0(o).
18.3.Z. The ínverse fetmion 2-point function rí•l (p} at one-loop order is

r('l(p) = M + i'i +e' f ddk '• (M + ;~- ífj 1'•

(211f (k- p)2 + M 2
6,.. ({- l)k.,k. )
( k2+m2 + (k'+m 2 )(P+E,m 2 ) •
(44.46)

We set
rPl(p) M A 1 (p) + i:rJA2(p). (44.47)
We then use a few standard tricks:
"~•"~• d, "("'"(" (2-d)'i. ~fl 2pk,.-k
2
i,
f ddk~f (k2 ,p2 ,kp) f
=i ddk~ f 2
(k ,,l,kp)'

Repeated use of the trivial identity

2pk= k2 +mi- [(k-p) 2 + ~~ +p 2 +m~ m~,

44.18 Exert:ises: Some Solutíons 971

lt is easy to obtaín the coefficient of M in eq. (44.47). Setting m' mv'e one linds

A, I+ e 2 j(d -1)B(p;M,m) + ÇB(p;M,m')).

Let us then write A 2 as the sum of terms
.,.
A, = l + 21" (A2 1 + A22 ),
where A,, corresponds to { = 1, i.e. ís the result in the Feynman gauge, and A22 thê
rema.ining part. Using the previous remarl<s A 21 can easily be evaluated

A21 = (2- d) [I(M) J(m) (p2 - M 2

+ m 2 ) B(p; M, m)].
The remaining contributlon can be written

We then use the identities

Weobtain
A22 {e[l(M)- J(m') + (3~ +M 2 - m
0
) B(p;M,m')]
[I(M)- J(m) + (3p + 2
M2 - 2
m ) B(p;M,m)]} +(I- {)J(M)
+ ~2 { [- (zi' + M>- m' 2) l(m'} + (p2 + M2 m'2 )
2
B(p; M, m')]

+ [(~> + M
2 2 2 2 2
- m ) l(m) (V+ M - m2) B(p;M,ml)}.

Our final e:>q>J:e55ion for A2 ís then:

2-d
A,=1+e 2 p2 [I(M) l(m)-(rJ M 2 +m2 )B(p;M,m)]
2

2~:1" {(P + M 2 2
2 2 2
+ ) IJ(m)- J(m')] +[{r?+ M ) + (p - M2 ) m"]

><B(p;M,m')- [(zi' + M 2 ) 2 + (p2 M 2 ) m 2] B(p; M,m)}.

18.3.3. At one-loop arder the fermion mass MF is gíven by

A short calculatíon leads to

~ 1 +e 2e~~ [l{M)- I(m)l + [2 + (d 2) 2: , ] B(iM; M,m)}

972 Eun::i:;es: Some Solutiom 44.18

One verifies that the gauge dependeuce h as cancelled. lndeed the furmion mass iE a phyo-
ical observable and it has been shown that pbysical observables are gauge índependent.
18.3.4. At one-loop arder r13l reads
rj."l (q;p,;P:I) = -ie')'~

+ (-ieJ 3 j (2w) ddkdl• [M +i(~+ 1hW'-r~ [M + i(j{ i2lr 11p[.C.d •• (k).

Let us now calculate the one-loop contribution to q.rJ?> (q;p1;!'2)· We note

lt follows:

q,.r~3 l (q;p 1;P2) e[(M + i,IL)- (M- i,l,)[

+ e3 f dd\-,. {
(2w)
[M + i(JI + ,I,Jr' [M + i(j{- ,1,)]- 1 } /p[ótJ •• (k).

We recognize in the r.h.s. the dlfference between two inverse propagators.

Calculatíon of the verte:r. Let us set

H we specialíze to the Feynma.n gauge ({ 1) C,. becomes

C _f ctdk -,.(M- i(j( +i•})-r,.(M- i(j(- ,lo) )-r.

~- (2")" [Ck + p,)2 + M2] (k2 + m2) [!k !'2)2 + M2].
We then write C,. as the sum of three terlllli:

where the denominators of the integrands of C,, C2, Ca successively are

(2~d)1,.,

(2- d) [(lf + 711)1,. + -r,.(j{- Jlo)l- 21,.(~ + 7h}- 2(j{- ,lo )I,.,
- 27~(1(- Ji,)(ll + 71.) + 2(~ + ,1 1 )(~- ,1,)1,.- (2- d)(j{ + ,1!)-r,.Cll- Ji,).
44.21 Exen:ises; Some Solutions 973

Chapter 21: Renormalization of Gauge TheoriéS. General Formalism

Exereíse 21.1
21.1.1. Because lhe gauge variation o! the gauge conditíon (21.91} is field-dependent

the Faddeev-Popov determínant is no longer a ronstant a.nd it is ne<:essary to introduc.;:

ghost fields. To express the ronstraínt it is also necessary to introduce a LagriUlge
multiplier field .\(x). The qu~IDtized ection ca.n then be written:

(44.48)

Note the relation

21.1.2. One immedíately notes a drastic simplification wíth respect to the case of non-
abelian ga.uge symmetries: The BRS tra.nsfonnations are linear in the fields and therefore
no sources for composite fields are required. lf we add to the action (44.4B) the !ollawing
source terms

we find that the generatlng functional W for connected rorrelation functions satisfies

Because the BRS transformations are linear the Legendre tmnsformation is strwghtíor-
ward. The JPI functional r satisfies

- .
'Vr(A1., >.,C, C) =O WJth 'V= -jdx (8"C
d ó
M,.(x) 6) .
+ A(x) éC(x) (44.49)

21.1.3.
The additional terrns ín (44.48) generated by the quantization procedure are the BRS
varíatíon of ~:

<l>(A,.\,C,C) f dd:cê(x)G(A)

=} 'V~= j dd:r: [A (o.A. + ieA!)- é (& + 2ieA~8~) c]. (44.50)

974 E:r:ercises: Some Solutíans 44.21

Exercise fl1.f
21.2.1. The êC propagator is 1/k2 in momentum space. The fields A" and À ar<'
coupled at tbe gaussi11.11 levei and the propogators thus fonn a 2 x 2 matrix:

(44.51)

21.2.2. First let us note that the large momenturn bebaviour of the propagators show-s
that th~t quantized action (44.48) is renormall?.able by power counting in four dimensions
and that tbe canonícal dimensions of fields are

d 4 :;. !Al 1. [>.1 2. !ê] + ICJ = 2'

Let us then write the most general form o( the renormalized action ju.st taking lnto
account power countíng and ghost number conservatian. A term que.rtic in tbe gbost
fields is allowed but by power counting it is necessarily proportional to [C(:r:)C(:c)j2 and
tbus vanishes because ghosts are fermíons. TheD

(44.52)

where G, has dimeDSion two and is thus at

two and a is a constant.
m-
linear in CC, L( A) has aJso dímensioD

We have in full genenility shown that íf the lPl functional r satisfies li""ar WT
identities the reDormalized action satislies the same. Thus

We now intraduce the decomposition (44.52) and identi:fy the coefficients af powelll of >.:

""'
2 •• éG_, =0, óG,_,
"' : -;--A u~
C - L(A)C = O,
fJC u "
6L(A)
f
d -
d :r:C8,C óA,(:r:)C =O,

Ftom the filllt equation it follows that G, is only n function of A,. and from the seoond
that L( A) ís determined as a function of G,. The most general form of G, compatihl<>
with power counting is

Thus the renormalized ghost action is

44.21 E:rercises: Some Solutions 975

The third equatíon ís now automatica.lly satísfied and the Jast equation tdhl us that I;( A)
is gauge invariant. The renarrna.l.ized action thus reads (with the change in notetion
a,__. ~uZ~)

S,'"" f d 0 x [Z.<iP;.- !{oZl..\

2
+ Z>...\ (v,.A~ + ieZ,A! + Za)
-z,c (&" + 2ieZ,AA.) C). (44.53)
21.2.3. We see in action (44.53) that a term propottional to >. 2 has been generated
by renorma.lization. The condition G(.4) is thus not stable under renorma.lization but
is replaced by a gaussían distribution centred araund G(A) = O. Since the action is
quadralic in >. the gaussian integral over >. results in a.n action where >. ls replaced by
the solution of the corresponding field equation:
Z>.>. =(8•A• + ieZ,A! + Za) /&,,

21.2.4. The induced variation of ~ ís

8~ = i6e f dd:rC(x)A!(:r).

The variation Df the generating functiona.l Z( J) of gauge field corre)ation functions foi·
lows:

liZ(J) = -ilie jldAd..\dCdCJ ex.p [-s(A, >.,C, C)+ f d"y J,.Ap] V f d•z A!(x)C(:r).
We then integrate by parts to let V art on the actiou and source:

6Z(J)=i.5e fidAd..\dCdC'Jf d•xA!(:r)C(x)Vex.p[-s(A,>.,C,C)+ f ddyJ,.A,.]

= i<le f ldAd>.dédCj f dd:r A!(;,;jê(o:) f ddy&,.C(y)Jp(ll)

xexp[-s(A,>.,C,C)+ f J~A~J·
Differentiat.ing then twice wíth respect to J and setting J "' O we obtain the variation of
the two-point funetion:

61-V!~(:<, !I)"'- ilie (f ddz A~(z)C(z)A,.(:r)8~C(y)),

where the a-verage symbol mean.s functíona.l average 'With the weight e-s. Covariance and
translatiou invariance teli"" that the two-point functiDn (f ddzA!(z)ê(z)A~(:r)C(y})
must be of tbe form
(f ddzA!(z)ê(z)A,.{:z;)C(y)) = 8,./(:r -y).

It follows after FOurier transformatiou that the variation of W~~ has only a longitudinal
component and thus the transverse part is independent of e. Therefore

W,..
- (2) (k) = ( ~... k~lt·) kO
- -;;'J I + k•kuw(k,e).
In particular Z A l.
97fi Exercúles: Some Soluticms 44.22

Cbapter 22: Classical and Quantum Gravity

E:z;en:íse 22.1
By a change in the pa.rametrizo.tion we can set N to 1. lt is then possible to construct a
harniltonian
ac.
p,. = &3:• = -x,..
and thus
X(p,x) = p,.x" C=~ [m2 (p+ eA) 2 ] .

In addition we must impooe the N equation of motion as a constraint:

After quantization we thus look for waw functions .p(x) whích satis!'y

We recognize the Klein-Cordon equation whkh is, in tbls framework, the Malogue o!
the Wheeler-DeWitt equation.
In this example we recognize that T il; not really a time but simply an au:xiliary ps-
rameter, and the absence o! evolutlon in this parameter is not too surprising: the Klein-
Gordon equation i& the real dynamical equation.
lf we cons.ider instead the lagrangian C' we could have the illusian that conjugated
momenta p,. can be delined because C' depends on aJI time derivatives:

However we note that tbls equation implies

Md therefore the conjugated momenta are not independent. We thus instead define o
temporal gauge by the conditian
x 0 (r) T.
We then restrict ourselves again to space components a.nd introduce the notation p, Á
for the vectors of components P•, A,., I' :> O, and v for ±", p, > O. We !ind

mU A . ií+eÁ
p ± JJ _ f12 - e A => ±mv =-r===:===
..j
m

(fi+ eÂ)2 + mZ

The hamiltonian follows

11: ~ ±V(i>+ eÂ)' + mZ+ eA0 •

We see that if we take into account the two signs we recover the Klein-Gordon equation.
We note also th!rt only one sign corresponds to a harniltonia.n bounded from below.
44.22 Exereís""' Some Solutions 977

Exerrise 22.2
22.2.l. Random wolh. Let us call Pn(z) the probabílity to be at point z at time n. lt
satisfies a recursion relation (in d dimensions)

2d

Pn(z) =~L Pn-I (:-e,.),

"""""1
where the points z- e" are all the neighhours of z an the lattice. The Fourier transfonu
Pn(k) af Pn(z) then SB.ti.sfies

Po(z)=é(z :r).

After a shart calculatíon we find

ln the long distance scaling limít the integral ;, domínated by the small values af k. The
prabability i.s non-vanishing only if ais dose to 1. We thus seta 1- m 2 j2 a.nd expand
the cosines

It follows that U(y, x) has the form of a propagat<:>r of a scalar partícle of mass m:

U(:y,x) =
1
( 2 ~r)"m
2
f e••tw-•l
ddk 1;• + m'. (44.55)

This interpretatíon of the euclidean propagator associates random wa.]ks to Feynman

diagrams.
22.2.2. The expre5Sian can be rewritten

wíth
(:yiTI:r) = .,-,.l•-:1.
After a Fourier transformation the operator acts by multiplicatían:

with
T(p) =f d•x e'"" .,-,.1•1.
The method of calculation is quite similac to tbe preceding exrunple. The continuum
Jímit is again obtained when the parameter /J is a.djusted in snch a way that the ]owest
978 Exercises: Some Solution• 44.22

eigenvalue f(O) appr~ches 1. The long distM(e asymptotic scaling form depends onl~·
on the expansíon of T(p) for p small:

We recover the Fe:.-nma.n propagator (44.55) and note that the result is quite robust
(using the language of criticai phenomena we can call it "universal") since it does no!.
depend on the explicit form of {yiTix} exrept that it is a function of lx- !11 and decrease"
fast enough at large separa.tion.
22.2.3.% [ace the problem ofparametriza.tion and moreover at lirst sight we know of
no dírect recipe to calculate a patb integral wíth such an act.ion.
Tb solve the problem of the reparametrisation invariance we can choose the length
along the path s:
(ds) 2 =dx·dx.
The path integral then C8Jl be formally written

U(y,x) r= doe-"'' J~(O)•z

lo
("!•l-~ldx(o))ó(x(o)·x(o)-1).

As usual we introduce a Fourier representation fOI' the 6-fnnction

U(y,x) 100

ds e-"'• j[dxk))[d.\(a)J exp l-~ [do (x- :X -l)À(a)], (44.56)

where tbe integratíon over À runs par aliei to tbe ;maginary axís. We now chsnge variables
q ........ t, setting
do
dt = .\(u).
The action ín (44.56) becomes

with O t(a 0), T= t(s). Usíng then

s= [da= 1T .\(o(t))dt,

we find
U(y,x) = r= dT l:r:(o)~
lo
r~<Tl=•ldx(tl][d.\(t)].\(T)

x exp [ -1T dt (!x(t). :ic(t) ~.\2 (t) + m>.(t))].

The integral over À is gaussie.n e.nd call be performed. We finally obta.in

U(y,x) = m
1"'
(}
dT 1z(T)=t jdx(t)lexp [ -~
~(0)-:r:
1T
o
dt (x(t) ·i.{t) + m
2
)
]
•
44.24 Exerr:íses: Some $olutions 979

Under this form we recognize a natural representation of thc ptopagator (44.55) in the
continuum, when we consíder the operator -d 2 + m 2 a.s a quantum hamlltonian:

Hawever the geometrical meaning in terms of a length of a path and the invariance under
parametrization are no Jonger obvious. Returning to the exercise 22.1 we understand that
thls action results from the ínvariant action integral ofthe lagrangian (22.114) (for A O)
by ga.uge fixing. ,.,

Chapter 24: Mean Field Theory

Exercise i?4.1
We use everywhere the notation of Chapter 24. We find

wjth À ~ l/2p- 1 > o.

The magnetizatlon i.s then

sinh 'P
M -A'('P)
À+ CO<ih 'P.

Hence B{M), the Legendre transformed of -A(<p) satisfies

B'(M) = <p(M) =In (>.M +VI-M'+ >,2M2) .

1-M

Ftom this expression we derive

B"(M) I (1
1-M~ + \f'J-Mli+>.2M2 '
>. )
B(M) = !(1 + M)lnD(M) + ~(1-M)InD(-M),
with
D(M)

The thermodynarnical potential r(M) is given by

r(M) = -V M 2 + B(M).

N<:>l-e tbat the cases P = ~ (À O) and p = l (.\

= 1) are identical ín the mean field
approximation, they correspond to I 1 and I = 2 respectively in the notatíon of Section
24.2.
lt is useful to expand B(M) up to order M 6 :

B(M) = (l +À) UM 2 + ;h{l + >.)(2 >.)M 4 + ,!0 {1 + >.) 2 (s- 9.\ + 3.\ 2 ) ~]
+0(M8 ).
980 Exereíses: Some Solutions 44.24

Fínally we note
B"(M) 2 B"{O) for A:$ 2.
We ímmediately condude the,t for >. < 2 or p > I /6 r(M) remains convex until V reaches
the criticai value V,
Vc ~(I+ A) l/(4p),
which defines the criticai temperatura of • second arder ph...., transition. The line of
second order phese transition 1;erminates at the special point À = 2 which corresponds
to a different physics and is called a trkritical point. For V - V, and M small we can
expand r(M)

The tricritical point 1s characterized by new exponents llke {3 1/4, 6 = 5, new universal
ratios lilre
C+JC_ 4.
Finally the "·bole scaling region is not limited to ), "" Ác = 2 hut contains also P• -À c!
sma.ll. Aft.er rescaling of temperature, field, magnetization and the additional thermo-
dynamical variable >. - Ác the scaling equation of state can be east into the universal
form
H M 5 f(:s:,y). f(z,y)=l+z+y
z (V Vc}/M~, y = (Àc- >..)/M 2 •

Exerrise 2j.l!
The antíftl1'1TJ'IfiJJgnetic I:ring m<>de~ In the absence of a magnetic field, the antíferro-
mag~~etíc lsing model 011 a bipa.rti\e )attice (a lattice whích can be decomposed into
two sublo.ttices such that in the two-spin interadion the two spins belong to' different
sublattices) can be simply related to a ferromagnetic Ising model by setting

(44.57)

In the new variables the mteraction reads

-{31í = L VijDiUj'
ij

and is thus ferromagnetjc. The alternating spíns o; define an order pammeter for the
antiferromagnetíc system.
In mean field theory the thermodynamíc po1<entíal r( M;) reads

r(M,) = L V;;M;M; +L B(M,).

H

A ccnstant magnetíc field a.fter the transformation (44.57) becom"" an alternating field
E; H. In such a field the minimum of tbe potential corresponds to the sum a constant
and alternating magnetízation
M;=M+EiM.
The equation
ar
8M;
44.24 Ezercises: Some Soiution.s 981

for small field H reduces to two equations

H= 2VM +aM +O(M3 ,MM 2 )

0= -2VM +aM +~bM 3 + tbM 2 M +O (M5 ,MM 3 ,M~M),

We see in the first equation that M alwa)'E remains linear in H

M ~H/(2V +a).
Suhstituting this solution m the oecond equatian we observe that the only elfect, a.t
(eadíng arder, is to modífy the criticai tempera.ture

V.- !a~ ftibH 2 /a 2 .

The a.nai)'Eis confirms that sínce a consta.nt rnagnetic field is not coup]ed to the order
para.meter, it does not destroy the phase trlj.DSition, but simply modilies non-univernal
fea.tures, (ike the critica! temperature.

E:ureise 24.9
Mean jield theory and effectiue Landa.tt-Cinzburg actíon /o'r modela with ZN symmet'1/.
An order parametec for the system is theexpectation value M = (eiD). Thevariab!e M is
thus a complroc nmnber, on which the symmetry group actB by multiplication. Symmetry
argmnents, teii us that the thermodynamical potential, is a real function of tbe ínvariant
po!ynomía.ls M'M, whleh is U(l);;;; 0(2) invviant, MN o.nd M'N. If we asswne the
existence of a. second arder phase trll.llSition, we ca.n expand the poteDtlal for M smaJI.
Severa! cases then ha.ve t.o he considered:
(í) If N 2: 5 then the terms which are not functions on!y of M' M can be negleeted
a.nd the scaling theory is tbe same as for the 0(2) symmetric model (a model with a
continuous symmetry). Only correctioru; to tbe leadíng t..rms reveal the dffi'erence. Note
however that the correctíoo terms play here a more important role than usual, &ince they
suppress below Te the vrould-be Goldstone modes of the 0(2) symmetry.
(ü) U N = 4 inst..ad we find the most general model depending on two order parameters
Mt.M2 and symmetric In M1 - M 2 , which Í5 equivalent t.o the 0(2) model w\th cubic
a.nísotropy.
{ili) For N = 3 a cubic term is ali~. Then we expect generkally a first arder phase
trD.ll!!ítion.
Let us now calculate the exparu;íon of the fuoction A(<p, <p') up the fourth order:

and thus

-A(<p,<p') = <p"<p +i (.p 3 + <p' 3 ) 6Na +i:. ('P4 + <p" 4 ) 6N 4 - ~ ('P"<p) 2

After Legendre trBDSfarmation

982 Exercises: Some Solu.tions

we obtain

The thermodynamical potential at fourth order then is

f(MM') VM'M B(M,M•).

confirming the generàl an.Jysis.

Other results. The N ~ 3 madel is known to have a secand order phnse tnmsitiou
in 2 dimensions (with ~ponents derived from CFT). In 3 dimensians the nature of th<'
transitian (first or second order) has been an object of debate and of heavy numerical
investigntions in recent time. The motívation was the following: The Z N group is tlw
centre of tbe SU( N) group and is involved in the question of conlinement (Wilson's loop
arguments). The confinement phase transition at high physical temperature in lattk<'
QCD in four dimensions is thus related to tbe phase transition of the Zs model in 3
dimensions (the physical gauge group is SU(3)). Most recent data suggests tbat th•·
transition is lirst order.

Cbapter 25: General Renormalization Group Analysis

E:rercise 25.1
The weigbt factor for the 2N-point function a.t lea.ding {N - 1-loop) order is (2N -
l}!f(N!) 2 /(N 1)!. For the 2-point function one finds I/{2N - 1)!. These fa<:tofl'
correspond to tbe symmetry fa.ctors of the corresponding diagrams. To then cakulak
the dia.grams in the massless theory it is conveníent to use real spa.re variables becauS<·
the diagrarns are just powers of the propaga.tor. One starts from

1
pí r(d/2 -1)
4?rd/2
Jd~ XX 2-d e -··· .

Calling K(d) the combina.tion f'(d/2 1)"-d/2 /4 one finds for the diagra.m with m +l
propagator atta.cbed to two points in which Hows a momentum p:

D - xm+t d .,di• f'(l- m(d 2)/2) (2)- 2

-=(d-Z)
=+t(Pl- { l r((m+ l)(d- 2)/2) p

The 2N-poínt funotion corresponds to m N - 1 a.nd the two-point function to m

2N -2.
For N = 3 K(3) =
lf(4>r). The weíght ofthe oontribution to tbe 6-point function ;,
5(3 and the divergent part then 5C/3t with C= lf(lfur2 ). For the 2-point function th<·
results are respectively 1/120 and C 2 /(12t). Aft.er a short calculation one obta.ins tlw
RG functions at lea.ding order
 Exercises: Sam€ Solutio~ 983

Exercíse 25.2
25.2.1. Lang range force•. We set u = A'•-•g. The function fj(g) comes from the
divergent part of a uníque bubble díagram. One finds

At this order the calculatíon of Z 2 and thus 'I> involves the sarne diagram. One thus
Jinds
'1>(9') = -< %1i
v- 1 = u + 'I> = u -c~
cN 2
-r=vu=l+ +

Chapter 26: Scaling Laws in the Criticai Domain

Ezercise 26.1
First note that the gradient llow property ís parametrization dependent. The properties
W>! derive here will be valíd more genera.lly for ali equations which are obtained from a
gradiomt llow by reparametrízatíon.
The corresponding llow of the coupling ronstant takes the form

av (44.58)
/3;(g) = ôg; •

Let us multiply the equation by dg,fdlnÀ. Then

dg; ) 2 dV
( dln>. d!n>. >O.

We tecover tbe essentia.l property of a gradient flow: fixed poínts are mínima or mexima
of the potentia.l V; patbologies like limiting cycles ate excluded. Morem-er if we calcula.te
the matrix B of deriVI'tÍveo; Bt à fixed point we Jind

which is a symroetric matrix. Therefore a.ll its eigenvalues are rea.l.

E:rercise !!6.2
26.2.1. Renmmalí.zation group for modé!s with cubic symmetrg. The stabi!ity condi-
tions are: g +À ::>:O, Ng + >. ;:: O, the second relatíon relying on Schwartz's inequa.lity.
26.2.2. The two (3 functions are

/3(g) = -tg + 8~2 ( N; 8 92 + g.X)

8 ~ 2 (2g.X + ~À )
2
tJ(>.) -CÀ +
984 Exerci.ses: Some Solu!ions 44.28

26.2.3. One finds four lixed paints, the gaussian fixed poínt g À = O, tbe Ising fixed
poínt g ~ O,>. "' l&7r 2 /3, the isotropíc fixed point À = O, g 48E7r 2 /(N + 8) and a new
cubíc fbted point
= 16?r (N - 4)~
2
À
3N .
Note that for N ? l the cubic fixed point satisfies the stability condítion g +). ?: O. Also
ali lixed points have 9 ;:,. O. The !ines À O and g O cannot be crossed.
Tn lind the local stabílity of tbe four fixed points one has to calculate the eigenvalues
of the matrix M of derivatives of the ~functions

__ 2_ (!ipg+À )
M - c + 81r2 2.À 2g ! 3À .

The set of cigenvalues are

gaussian E ,-c:'

lsíng ~ j-t tt'

4-N
isotropic
N+S"'"'
cubíc
N 4
C, ---:ii{ E .

We recover a parti.,ular example of a more general result, the isotropic fixed paint is
stab!e for N < N, with N, : 4 +O( c). The cubic fixed point is stable only for N > N,.
At N = N, the two fixed points merge and excba.nge role. The lsing lixed pnint ho.s
alwaY>I one direction of instability. Note that for some initial values of g,), (like g < O)
the effective ooupling constants are not attracted to a fixed point but run away. It is
generally believed that this signals a first arder phase transition.
26.2.4. The perturbatinn of tb.: lsing modd by tbe isotropic O(N) symmetric interac-
tion involves tbe operators (if>2 ) 2 • At the fixed point al1 .P components are decoupled.
Therefore the (4>2)2 is equivalent to" linear oombination of .pf and 4>~4>5 wbich factorizes.
Tbe corresponding exponents Me "''oing and {a/ ..)Ioi..:·

Chapter 28: Calculation of Universal Quantities

PDiyme.-.. Let us calculate the 2-poínt function z< 2l(k,3) at on~loop arder. At leading
order - have fnund

The onE>-loop contribution to Z('l(li:, t) reads

[z<•l(li:,t)J.Loo"~ -~ j (~:~d fo""' dH-"'((k,q,s),

((k,q, s) =fo' du 1du2 (eik·[r(•)-•(O)( e•q·[r(,.,)-•("•ll)

"'21' du21u' du, .-k'(•-u,+-.J-(k+o)'(••-••).

44.29 &e-r<:i:Jes: Some Salutíons 985

We bav-e used the property tbat the average of a product of exponentiall; is given by a
simple gaussian integral, as we discuss in 31.1.
1
lt is then convenient to change variables, s, tt2, -u1 ;....., S , u', u1:
I
u.

The integrais OVér u 1, s', u' can be explicitly performed. Shifting q ín tbe remaíning
integral, q + k ,__. q, we then obtain

whicb is índeed the one-loop contribution to the 2-point fuw:tion in the (t/>0 ) 2 field tbeory
in tbe N = O limit.
2
Chapter 29: The (t/>2 ) Field Theory in the Large N Limit

Exercíse 29.1
The COI"relation Jengtb above T, is given by the limit of oquation (29.15):

N • N lnT
T- 4:or2 m•In(A/m) '* e= m-• ~ -Ifur'--;-.
One vermes that this result is in agreement with the analy.;is of Section 27.2. Similo.rly
the oquation of state (oquation (29.23)) becomes

o r:
H_ 32~ u(u2 + T)
N m(A/a) '
again In agreemeut with tbe large N limit of the general result. Fína.lly the specific hea.t
above T. is proportional to 1/lnm, i.a. at leadingorderto l/lnT as expected.
We stress again that even in four dimensioru the bare coupling coDStant has played no
role Md the same resulta would have obtained from the O(N) non-lineo.r u-model.

Exereise 29.!!
Let us explore further the sa.ddle point equations (29.67), for a general potential V

(44.59a)
(44.59b)

(44.59c)

At low temperature lT >F O and thus .>. vanish.,. Equatlon (44.59c) has a solution only
for }J > }Jc,
986 Exereises: Some Solutioru 44.31

Equation (44.59b) then yields the criticai temperatura. Setting V(JL) = W(i') + fTI', we
find
To= -2W'(I<c)•

Th find the expooent f3 we need the relation between the r and I' near the criticai poínt.
In a generíc sítuation we find
t =r -r,~ -2W"(Pc)(l'- pc) ~ fJ = 4
Thís ís the case of a slmple criticai point which we have examined in the section. A
new situation aríses íf we can adjust a parameter of the potentia.l in such a way that
W"(>tc) = O. This provides an example of a tricritícal point. It can be achieved on!y
if tbe potential V ís o.L least cubic. Higher criticai points can be obtaiued when more
derivatíves vanish. However thls ís no longer ínteresting for oontínuous oymmetrí"" sinee
mean·field behaviour ís then obtained for d;:: 3 (and symmetry breakíng ís possible only
for d > 2).
At hígh temperatura o = [) and >., which ís the o-fie!d mass squared, ís given by

2W'(JL)+r= >., P.-1'•

N
=- (2w)d>.
f ddp
Ji'(p" +>.).

For d < 4 we find p.- ,11c ~ >.(d-2 l/2 • In the tricritical case then
>. t + const. x W"'(p,).>.d·>.
For d = 3 we find a mean field behaviour without Iogruithmic oorrections. We recai! that
d = 3 ís the upper·critical dímension.
DiBcus.W... Jn the mean tield approximation the function W(p) = p 3 is not bounded
from below, h~er JJ = O is tbe miuimum because by delinit.ion i' ;:> O. Here ínstead
we are in the sitllatioo where W(JJ) ~ (p- pe)' but ~ is pooitlve. Thus this extremum
of the potential ís unstable. Tbe tricritical pt>int aad tbe (.p2 )~ cannot be studied by the
large N limit method.

Chapter 31: Two-dimensional Models and Bosonizatlon Method

Eurcúe 91.1
Distinguishing two operaton: in equation (31.13) we find

(ll e'"'"'(z,))l
&--1 reo.
(IJ lo:t- x21J"'"' 12 ~ TI (p lxt- xd)"'"''~7
1ó2

x (JJIX2- x,l)"•"'i2• TI (IJI:r;- xill"'"'/2•

'Th.lcing· the limit x 2 --> x 1 we see that the firat factor is the singular roefficient function
and thus

with
cl'fl/'r;.~(z) N%"1"'1}211'.
M oreover the behaviour of the coeflldent function C is relate<! in a way expected from tbe
discussion t>f Section 12.7 to the dimensions of the three operators (see equation {31.16))
since
44.31 &ercises: Some Solutio"" 987

Exercise 91.2
The model (31.114) possesses, in additíon to gauge ínvaríance, a. U(N) X U(N) chlra.l
symmetry because the fermions are massless. The model is a símple extensioü of the
Schwinger model and can be solved exs.rtly. &; we shall see, however, for higber values
of N its structure is somewhat more complica.ted.
The firnt. step is tbe evaluation of anomaly. This is simple because the fermion deter-
mina.nt is replaced by tbe determinant to the power N. The generalization of the free
aciion (31.61) is thus

S(ip',.p',B~) - jd 2
.r [ô}.'{x) ·i!.P'(x) +::f d 2:r(8.,'P(xl)
2
]. (44.60)

Adding the gauge field free action which is uncllanged we obtain

(44.61)

with
(44.62)
The 'I' propagator has the form:

('P(X)lf!(O)) "" - lf d p e'"-

41r2
. N11(-
2 1- - -p21) ,
p2 + m 2
(44.63)

and therefore propagates a neutra! massí"" state together with a nagative roetric massless
state, 88 in the Schwinger model. Sinoe this bosoo is &ee it ca.nnot prod uce cha.rged
fermions Md this is an indication of con.fioement.
Let us oow bosonize the fermion fields f/1, f//, introducing N scalar bosons fJ;. The
equivalent boson s.rtion is

S(fJ<,'P) =~f d2 :r [a~t91 ô,..?;- ~a.. ., (m2 - &')a• .,]. (44.64)

Then we proceéd as for the Schwinger model, transla.ting .?; + 'P{../i- {); and integrate
over !p: The action takes the form

(44.65)

Through ali th""" transformations the chiral romponents o~ of .fj,,.p, become símply

<T~(x) = .j,i (:r).p~(x) ,...._ ~ ,.±<......--ri<(•). (44.66)

r-r 2ft
The first cemark is that in contrll8t to the Schwinger modeJ, besides one massive field
Í::; !l;, we haw in addítion N - 1 massless booon fields.
The <T~ =lation functiom. Due lo the presence for N > 1 of massless boson.s
averages of the form ( cr~) ali vanish. However let us cal cu late the average of the operator
988 Exercises: Some Solutions 44.31

Let us write the operator ín the ferrniou representa.tion we see

n
JV

i=l
11~(%) = ( -l)N[N-1)/2 n n
N

t.-1
~~(%)
N

'i=l
w+(x).

Dueto the antisymmetry ofthe fermiollll, ifwe perfonn a U(N) x U(N) transformation
of matrices U+,U-, the operator transforms by multiplication by detU+fdetU_. The
operator is thus invariant under the SU(N) x SU(N) x U(l} group. lts non-vanishing
expectatio.n value breabi only the remaining U{l) cblra.l component. .
Finally the confumation of oonlinement is here a more complicat..d problem since
spe<:trum considerations are not suflident.

E:rerrise 91.9
We first expand the fíeld ,P(:z:, y) in Fourier modes:

The corresponding action takes the form

As usual in the finite size prob!em we have to ronsider separately the zero momentum
mode whicb wollld naively lead to IR divergences. The zero mode corresponds to a free
particle in quo.ntum mechanics. A simple calculation sbows that its contribution to the
tw~point funct.íon is a factor

(21r)b(k, + k,) e-!k~!•l!L.

The .5-funct.íon is expeeted from the ana.Jysis of Section 31.1. The propagator for the
other modes then take5 the form

11 ( ) 1 "" 2iwn11/L f~-. e""•

L x,y 2..-L L" ""'w2 + (21fnjL) 2 '
n .. O

4~ L~ e-2'11'nlr}JL ( e2i~nlf/L +e-:Zi•n'lfi/L)

._
n>O

=- ~ In [ ( 1 - e-•~1•1/L e•;~;.(L) ( 1 _ e-•"1-"I/L .-••••JL)].

4
The propaga.tor at coíndding points is UV divergent:

llL(O,O} = 217T ln(LA) + 0(1).

Performing the gaussi811 integration we obtaín the two-point functíon

( .,,,.,.(,.,•) ei•,4>(0,0)) oc é(k, + ko) .,-~kilzi/L .,-klb>(Lh)f2" éi.t.~(-"••J

+ k2) [A 2L 2 sinh (";') sinh (":)rk~/!

4
oc 6(kl •l,
44.31 Exernses: Some Soluti 0ns 989

where we bave set w X+ iy, and w is the complex conjugate of w.

w.. now observe that the conformai transformatíon z = _,2wwfL, maps the complex
plane onto a periodíc strip of size L. The ptopagator in the inítial r variahle is

~(z,z} = -~ lnzi.
41r

Then the propagator in the new variables between the points w = x + iy and w = O has
the form
~ - 4l"ln(l-e2n•/L) (! e2x1ll/L)'

whích coincides with the r.,;ult obtaineâ by direct ca.lculation. One can actually verify
t.hat in a conforma.lmapping z >-+ w the two-point function of eü.P(•) tra.nsforrns by the
su bstítution z .... w MJd is multiplied hy a factor coming the dimension k1/ (
811) of the
operator

Exerrise SJ.,;
Let us first determine the function f(z) by exploiting the identity for n = 2. Multiplying
by the d•mominalonl one linds

Setting z, z2 O one linds /(0) =O. Setting tben z 2 = 4 one concludes that /(z) iB
an odd fnnctíon. One linally differentiates twice with respect to •I and once witb respect
to z~. One setB z2 •1 = z; = O. A differentia.l equation for f follows

/"(z)/'(0)- /(r}f"'(O) O.

If f"' (O) vanishes we re<:mW tbe !in.,.,. fnnctíon we bave cunsidered sa far. However we
have another solution
f(z) = sính(oz).

It is then easy to verify that the functioD satísfies the identity for all n by noting

2e0 z eoz'
sính(a(z ela.z - e2<r~:' ~

We can factori2e the numerators in front of the deteTminant, and in the denominator!l
we recognize the Cauchy determinant wíth variables Z = e•a•. Tberefore we can a.lsa
hosonize a system in a strip or at linite temperature.
990 Exorcíse.: Some Solutions 44.37

Chapier 33: Criticai Propertles of Gauge Theories

The Landau-Girubu1!J model of superoonductivi.ty. !ntroducing the parameters g. ii

we obtain for the couesponding ;él-functions

{39 -Eg + ~_g' 6gô + l&l , (44.67a}

l3ii = -<&+ !&2• (44.67b}

Setting t = In .l., where >. is the scale parameter, and õ õs, we can write the fiO"'
equations at d = 4:
d& 1-2
dt 110'
ds
dt = ã(~s -
2
.!f• + 18).
The fust gives the llow of êt which goes to rero at large distance (t - -oo), ii -
-3ft. One verifies that the factor in the r.h.s. of (44.67a) mNer vanishes. Therefore s(t)
decreases at large diBtan~ and passes through the orígin for a finite t. Since stability
implies that B(t) should remain positíve, we face a case of runaway coupling constant, a
strong indication of a lirst arder phase transition.
One then verifies that for the same reason in d 4- E there is no rolution to the
IR fixed point equatlom; and again one predict. a first order phase transítion. Th!s first
order character is not ob.served experimenta!Jy, the belief being tbat the critica! domain
is too smal!.
Fina.lly by increasing enough the number N of charged scalar fields eventually one
reaches a situation where the transition becomes second order, a. result whicb can also
be cbecked in the large N limit.

Chapter 37: lnstantons in Quantum MEChanics

Exercise 37.1
A few elements of tbe ro!ution are:
(i) The classical salution

1 1
v'=Ycosh[{N -l){t -to)]'

(ii) Tbe cla.ssical action S,

S, = A(N)f(-g) 1f(l>l-t),
2
A(N) ~r(Nj(N 1)) = 41 /(I"-1Jr (N/(N 1)).
2r((3N -l)/2(N -1)) r(2N/(N -1))

(iü) The operator M, second functional of the action a.t the sáddie point, is

2 N(2N 1)
M -d, + 1 - cosh~[(N- 1)t] ·
44.37 Exe..-ci:u;s: Some So!utions 991

Then, in the notation of fl<!Uatioc (37.38), we ha're:

-1 r(l +z)r(z)
det(M+E)(Mo+E) =r(l+>.+z)r(z >.)'

wíth
>. Nj(N -1), z = ,f!~j(N- 1).
Then
det (M +E) (Mo+ E)- 1 ~ -2-(N+ 1l/(N-llA(N)E.
(iv) It again ha.s one zero eigenva!ue assodated with time trans!ation. The jacobían J
induced by tal<.ing time as a collective coordinate ís

J' =f dtq'(t) = A/(-g)l/(N-1).

Using fl<!Uation {37.37) and oollecting ali factors, we obtain the imaginary part of the
ground state energy for negatíve g:

lmE(g) = C(-y)-~exp [ -A(N)/(-g)1f(N-1)], (44.68)

with
21/(N-1)
13 lfi2(N- 1)], C=--
,fi

Exercise 37. f!
We only exp!ain the main new features of the solution, compa.red to the " = 1 case. The
classical !!O]utions are
q,(t) nq<(v l,t),
where n ís an arbitrary consta.nt unit vector. Therefore the classícal action ís the same
as in the " = l case. The maio difference comes from the degeneracy of the saddle point:
We have to include the unit vector n in our set of collectív.. coordinates. Thís leacls to a
new jacobian and to a factor whicb is the surfa.ce of the 11 dimensional sphere.
The second derivative of the action is

M;i = ( -d~ + 1) Ó;j - N(óij + 2(N- 1)n;ni) h2 [ l

cos (N 1)t]
The determinant can be expn:ssed in terms of the determinants of a tra.nsverse and a
longitudinal operator ( with respect to the vector n)

M da N('lN-l)
L - t + l- cosb2 [(N l)tj'
N
MT = -d~ +1 cosh2 [(N -l)t]"
The longitudinal determinant is the same "" before. Jts zero mode corresponds to time
translation, The transwrse determinant which appearE v - 1 tinles corresponds to >. =
1/( N - 1) and its wro mo de is related to rotations of t h e vector n:

det(MT +.:)(Mo+ c)- 1 ~ ~2-(H+l)/(N'-tlA(N)c.

N-1
992 Exrn:ises: Som• Solufíons 44.38

Note that ali cigen,..(ues of MT are positive. in such a way that the final result remains
real as expected.
The eigenvecton; corre.<;pondíng to a rotation are ali unit vectors orthogonal to n mul-
tiplied by q<{ v !, t). The corresponding jacobian is thus

!•-l)/2 (N + 1 )!•-1)/~
JT = (f dtq~(t) ) - N=}A (-g)!• l)/2(N 1)·

We have of eourse to divide by a factor v'21f for each zero-mo de. Finally the integratio o
over n yields the surface of the sphere. Thking aho ínto account the factor l/2í to obtaín
the imaglnary part W1l find a addítional factor:

Collecting alJ factorS we finally obtain, in the notation (44.6S):

{) = vf(2(N- 1)], C= _1_2vf(N-l]

r(v/2) .

Chapter 3!!: lnstantons in Quantum Mechanie&: Genera!ization

ExefT:Íiie 98.1
The action S(r} can be rewrítten

S(r) =!lor" dt(f(t) + wr(t)) 2 + !"' (x"- 7'2).

which corresponds to the boundary condítíon cr(O) = x.

The jacobian J of the transformation is (equation (38.2Q))

We bave still to impose the condition r(tl) x'. We introduce a variable ), to write the
correspondíng 6-fnnction in exponential form. We obtain the action

Shifting tben cr(t)

á(t) &'(t) +i.\e"'(t-Pl, c'(O) x,
44.38 Exercises: Some Solutions 993

we find

5(~>" • .1.)

The integral ovei <7(t) is now

The integral over ). th"" Jesds to

U(x I ,x;,6) = J
'"-..).
y21f
vwh pexp[-;wA( x,x),
sm w
1 ']

with
(r+ x' 2 ) cosh w/3- 2xx'
sinhwtl
which is the oorrect result. The factor (2")-'12 is the product of a factor l/(211") from
the representation of the 6-function by the factor (211") 112 of the gaussian integration.

Exercise 98.2
AU the relevant equatioru; can alresdy be found in Section 38.3. 1b calculate the im agi-
nary part of the Nthe levei we have to identífy the ooef!icíent of e-(N+ 1 / 2 l 8 in the la.rge tl
exparu;ion of the partition functíon. The lesdíng oontribution then comes from the term
with the la.rgest poweT of lfg. 1f we expatld E(tl) Blld A.(tl) in (38.50) we observe that
the lesding oont.-ibutíou comes from the correction of order e-P in A(f3). Ftom equation
(38.60)

we then lind

ImE,...(g) ~~~(~f lmE,(g).

We notice that, as expected, the ímagínary patt íncreases with N and thus the lifetime
decreases. Of course eventually for N large enough the expressíon is no longer be valid
and has to be replaced by the correspondíng WKB valid for large quantum numbers.
38.2.1. Barrier penetration effects are relat.ed to euc!idean trajectories with go from
the orígin to a turning point and bounctl back. lf the action is O(v) ínvariant such
trajectories wiU take the form of the 0(1) t.-ajectories multiplied by a consta.nt unit
vector n:
qc(t) = nq.(!).
Therefore the cii>Bijjcal actíon is independem of 11 a.nd the a.nalysis of the " 1 can be
repeated. The main dif!erence comes from the g~ussian integration. Note first that it
is simpler to calculate the contribution to tre-I'H than to the matrix elements, because
the dassícal solutíoru; are símpler. The classical rolution, being períodic, is determined
by one boundary rondition for example the value x = q( -/3/2). The variable x plays the
role of collective coordinates corresponding to tínte translation and direction in q space.
994 E:z:ereises: Some Solution.s 44.38

We have then to lindo non-singular matrix K satísfying equation (38.45). The equation
for K takes the form

We have to find enough lndependent solutions. Here the problem is rather sirnple. We
decompose the matrix K as

K1; =KLn;n; +(6iJ -nini)KT.

For what concerns KL one solution is qc{!} and the problem ls the same as in the v ~ 1
case. For KT we recognize the equation of lll<ltion and t.herefore a solution is qc[t) itself.
The cakulation is tben an extension of the 11 1 example and the rooult is, with the
notation (38.59):

lmE.J(g) ~ - -
f(v/2)
1 (-2C)"'
- ' 9
e•f•.

Ezercise 98.9
As we have seon in Sectioo 2.6 we have to e\oaluat.e tbe evolution operat.or witb dassical
scattering houndary conditíons. We recall the expres.iion of the evolution opera.t.or at
leading WKB order (V..n Vleck's formula for 1i = 1), as gi~reo in A38.2:

(x"]U (T",T')I x') ~ __2__12 (-det :::.., )l/2 exp [í.A(x ,x';1' ,T')],
1 1
11 11

(21rit ""' '-"'"

where .A is the classical action evaluated for the solutions of t.he equation of motion.
The da!;Sícal equation of motions, after integration, yield ·

where J<, I are two constants of the motion a.nd I} characterizes the angular positíon of
the vector x in the plane of the motion. Taking as houndary conditions x(t') = x',
x(t") = x", setting T = 111 t' and

1<(r) = 1<2[ - 2V(r)-

t]
~
1(2
,

we can write the two equatio!"l.S determining 1<, I

r dr
)L ~t(r)
I r dt
}L r 2 ~<(r)
0
T' '

where a is the angle between x' and x" and L means the domain of integration

r= 1r' +[",
JL 'to ro
~<(rol O.

Fínally the cla.ssical action .A is given by

.A"'
f dt (ix2 2 V(r)dr
V(r)j = 4T~< - 2 L~
f
44.3S E:rercises: Some Solutions 995

lt is useful to calculate the variation of the a.ctíon when the end~points va.ry

6A = lho+ t<(r")6r"- 1<(r 1)6r'.

As a consequence:

1
B.A = - --
r"sína
(cosa=f
ru
A few other formula are useful:

.
fPa
ôx~lJxtf]
1 [
= - r'r'' sino liiJ

We note that the result is a projectot orthogonal to x 1 and x 11 • Finally we can derive

det
éi A
2

ô;,;;ax'; =
1 (-aA
r1r" 75;;- r 1r" sina
1. )"-
2
(li'ôrxélr'A aaélr''
â A
2

A short calculation then yields

2
a• A ( I )"- tJ!
det iJ:I:iâx'j = - r 1r 11 sina r'r''~<(r')K(r")l él,;.'
with
],., -·~] r~
I< a,._ ljL 1<(r).
Larye time expansíon. Knowíng from the general analysis that the boundary condltions
should correspond to classicalscattering,- can expand for T la.rge (we assume that V(r)
decreases fast enough). Setting

k (r'+ r 11 - 2ro)/T, k(r) = k2 [ V(r)-

t•}
;:>
1/2
,
where ro is replaced by the limiting v&lue solution of k(r 0 ) = O we find

T,

and thus
11= k+~ L"' dr(k~r) -1) +O(T-
2
).

lt follows aftet a short calcu!ation

Q = 21 L"" dr

A= !Tk 2 +21"" dr[k(r) k)+lt:t.

ro
996 Ezert:íses: Some Soiut.ions 44.39

In this limit I is related to the impact parameter b by h 1/k. We then celculate

the Fourier transform of the evolutíon operator. At leading order for T large the action
depends only on r +r". We can tbus integrate over the angles. We find Bessel's functiollS
1

with large arguments pr1 and r/' r" whích can be rep]aced by theír asymptotic forros.
These correspond to the angle b~n x and p being O or r.. The integral over r'- r"
then ;rields the energy conservatlon delta-functionl p1 ~ prr'l. The int.egral over r 1 +r"
then implies k p'. Thus finally we can substitute

Ali these terms give only a pha.se </! obtained by adding to the result the contdbution
~p' 2 coming &om the free terms

ft>-= 2 (" dr [k(r)- k] +la- 2ikro,

Jro
while the amplitude completely given by the determinant. Thus the final result for the
crO>Wsection is proportionel to

8a(k,o) ()( ab/1-2

ao aa '
which ís the result obtained directly by a dassicel celculation.

Chapter 39: Unstable Vacua in Fiekl Theory

Ezercise 99.1
The expression (A39.21) is independent of the normelization of <p and the se ale of dís-
tances. Therefore we just substítute

op(r) = e-••t•'

and lind A(d) 247rdf 2 ( 4 - d)(d-•lf• and thus

A(2) ""37.699 ... , A(3) = 133.639 ... , A(4) 24,- 2 •

For the seco!Jd tríel function one finds instead

ln2 +~
d 2, A(2) = l h l n 2 - - - ""35.178 ...
2ln2-21

d 3, A(3) = 87r (((2) + 2)3J2 ('/2(2) = ll5 934 .

3 C(2) 1 . . .
2
A( ) ,• (9((3} + 24ln2) _ 2
d=4, 4 16 9((3)-12ln2 _,. x 18 ·840 ... '

where <{8) is Riemann's ( functíon.

This can be eompared with the exact results A(2) = 35.10268..., A(3) 113.38350...
and A(4) = 16:or2 • We see that both VMiationaJ estimate; are better for smaller dimen-
sions, but also tbat the JD triel function is much better.
44.42 Exen::i.ses:- Some. Solutions 997

Chapter 42: lnstantons: The r/! 4 Field Theory in Dimension Four

Ezerciu 4 2.1
We first note tbat the dímension ín which the theory is renormalizable, and thus the
d~ca.l theory sca.le ínvariant, is d -= 2nf{n 1). The calculations then are a simple
genera.lizatíon of the f/>4 field thoory.
The dominant insunton still corresponds to a field depending only on a radia.l coordi-
nflte r. We set
<i>(x) = ( -g)-lf(2n-2) /(r).

The classica.l actíon becomes

s"
S(f/>)-= (-g)'/ln-t)E(/)

E{/)= L"" r"- 1 2

dr ({/' (r)- ~<.,!.. 11 >f2n(r)),
with Sd = 2,.-d/2 /L(d/2). The actíon is ínvariant under the sca.le transformation f( r)~
).(d->)/ 2 f(-\r). Therefore we set r= e' and

(44.69)

The action tben becomes

We recognize a quantum mecbe.nica.l problem, which we have solvod in the exercises of

Chapter 37. The instanton solution of the equatíon of motion is

The dassica.l action 8, 1s

S, A{d)/( -g)l/(n-t)'
2
2
A(dl = zd- 3 s. b(d)(d- 2{ !d/ l
rcd)
lf now more gcnera.lly we change variables in tbe a.:tíon

x - t=ln,-, n xjr,
and a.Jlow tbe field {44.69) to depend on n we lind a scaled a.:tion

!:(h)= ~b{d) j dfldt [h {tj + hL h +

2 2
(n2l)'h 2(t) ("_'l),h2 n{tJ],

where L 2 is the angular momentum operator with eigenva.lues 1(1 +2/(n-1)). Tbe second
deriwtive M of the a.ctíon at the sa.ddle poínt at angular momentum I ls proportíona.l
to:
-d~ + (l + lf(n
998 Exercises: Some Soiutions 44.43

ln the notation (42.77,42.78)""' find

.\ = n/(n 1) [
z= (1+1/(n-l)) +< 2 ]V2 .
We observe that M hBS vanishing eígenvalues for I = O and l = 1, as in the simple rjJ 4
example, corresponding to the dílatation and translatíon modes respectively. One has
to introduce collective roordinatel>, and calculate the jacohian. One important difference
betwren tbe n = 2 and n > 2 case is tbe absence of one-loop retJormalization constant of
the coupling constant. The remaining part of tbe ca!culation can be found in
E. Brézin, J.C. Le Guil\ou and J, Zinn-lustin, Phys. Rev. D15 (1977) 1544, where
r
r~'lllt.s are more genera!ly given for the ( rp 2 O(N) symmetric theory.

Chapter 43: Multi-instantons in Quantum Mecl:umlcs

Ezen::ise 49.1
The corresponding eudidean actíon is

S(q) = 2~ f dt [oi 2 (t) + ~ (E'(q)) 2 - ~E''(q)].

At !eadíng arder for n SJnall E"(q) can be neglected in tbe classical equation of mo-
tion. However, evaluated for the classical solution, it will yield a multiplicatíve fa.ctor
in the result. The potential ~ (E'(q)) iE exactly the double-well potential. Thus the
2

classical solutíon is the same. Tbe subtle point is that, due to the E"(q) contribution
the wells are not rea!ly degenera.te and the leading terrn in the ground sta.te energy is a.
iostanton-antíinstanton pair_ A similar situation has been e:xamined in Sectíon 43.3 and
the calculation is quite Similar. Here we have w = l but we bave to add to the instanton
actíon terrru; corresponding to tbe values of E" near the well multíplíed by the time spent
ih tbe well. In the notation of Section 43_3 we find

(44_70)

The effect' of these additional terms is slmply to shift by ± ~ the a.rgument o f the function
l(s}. The spectral equation then reads

(-g2)2E f(-E)f(l
e-1/Jo
E)~=-1. (44.71)

Let us extract from th!s equation the ground state energy

.,-1/Sg
E~~(l+O(g)).

By expllllding to next arder in E we also obtain the instanton contribution to the ímag-
inary part of E a.t leading arder
e-2/39
lmE(g)~~.
44.43 Ettrr;ises: Some Solutions 999

a result consistent with the property that theleading contributíon to Eis alrea.dy of order
€-•!3•. Tlili; result ís peculiar to the gmund state. E:xcited states ha.ve a non-trivial per-
turbative expansion. Note that at this order the equa.tion see1115 to índícate that startíng
from the first excited states the energies are paírwio;e degenerate in perturbation theory
and that the ínstanton contribution corresponds to the splitting between states in the two
wells. Actually this is a feature of tbe leading arder. There is no symmetry hetween the
wel!s and the degenere.cy willlifted in perturbatinn theory. In the r-functíons E will be
replaced by two different functions. The calculatinn of the ínstanton contributíon, which
is proportinnal to .,-l/ 3• / g2 N+I, requires the knowledge of these functions at order g. ·"'
IN OEX

o.bellim anomaly, 472. Cabibbo anglo, 465.

aim<Jst planar interface model, 810. Calla.n-Symanzik (CS) equation, 204.
amplitude ratiM, 629. canonical (engineexing) dimension, 168.
amputated correlation functions, 124. ClUlonical inWli'iance, 42t 491.
<UJharmoni< oocillator, 871, 922. canonícal t.rnnsformaUon in phase spa.ce, 834.
anomalh~ ('ee a1so chiml), 459, 464. canonica.l transformations, 42.
a.noma.lous dimension, 57S. Cosimir effect, 438.
anticommutation relationa.) 56. c:ha:racteristi<:s: method of, 576.
antiperiodic bounda.ry oonditions, 59. c:ha:rge conjugatíon, ll2, 155.
area law 1 738. chiral anomaly, 187, 476, 698, 701.
asymptotic frwdom, 338, 471, 678, 703, 735, chiral components, ll L
748. chiral gauge invariance, 713.
asymptotic series, 880. chiral invaríant Gross-Neveu model, 714.
axial current; partia! conservatíon (PCAC), chiral modeffi, 333, 382.
2!lll. chiral symmetry, 114, 189, 245, 286, 682, 706,
708.
bacl<ground field, 147, !56, 227,339.
chiral symmetry: spontaneous bteakíng, 701,
band •t.ructure, 856.
703.
bare aetion, 196.
bare correlation functions, 202. Chrisloflel symb<>l, ó04.
bare parametero, 195. classical lield equation, 445.
bare RG equations, 574. Clifford algebrs, 8, 108.
Bargmann potentlal, 820. clusleT properties, 121> 536.

barrior penetration, 8!2. coherent sta.tes~ 61.

berezinian, 1 I. Colem<UJ-Gross theorem, 748.
Bianchi identity, 508. collective ooordinate, 818, 826, 843.
b;na.ry Jluids, 638. rolour quantum number, 463, 466, 731, 758.
Borel summability, 524, 680, !lll2, 912. cornmut.ation relations, 9l.
Borel :summe.tion, 634t 914~ compatibility onnditions, 271, 358.
Borel tra.nsformation, 881. composite opere:to.rsl 174.
Borel-Leroy tr8.]1lll;format.ion; 634. oonfinement, 466, 701 1 703, 731, 735.
bosoniza.Uon 1 689~ conformai invaríance, 636, 689, 905.
bounda<y conditions: twisted, 783. conformai transformationst 300~
Brillouin wne, 181. · connecteà correlation functions, 120.
brownian cba.in, 644. connected conelation functions: decay of1 161.
brownian motlon~ 20~ 67. oonnection, 442, 504.
BRS exact, 355. co:nserved r:nrrent) 406.
BRS operator, 487. constrai:nts~ integrnl represento.tion, 4.

BRS eymmetry, 330, 352, 355, 376, 448, eom'exlty, 125.

449, 486. correl11.tfon functíons, 28, 71 1 93.
 1002 Ind=

oorrelatíon length, 529, 551. dipolar fon:;:cs; slrong., 617~

.;:onelo.tion time, 7681 801. Dinu: .i-function. 4 1 13} 103.
coset space, 322. Dirac "t matrh::es, 108.
c=ine potenlía.l, 855, 916. Dirac fermion, 104, 1 08, 153.
cnsrnologica.l t.erm, 522. Dirac operator • 477 ~
Coulomb gas, 718, 722. discr-e~ symmetries, 285~
Coulomb potential, 35. dísslpative couplíngs, 764.
Coulomb's gauge, 413. díssipalive Langevin equation, 73.
rounterterms~ l9fi. di..ergence from operat.or ordering, 46.
conpling constant renorma.lization, 196. dí""rgences: UV, 164.
covarlance of RG functions, 218. doubl.,.well potentia.l, 854, 908.
covariant cnrl, 505. dressed okcleton diagram, 199.
covaFiant derivativo, 334, 409, 442, 505. drift. driving,. forcef 73~ 764.
oovariant gauge, 44fL dynam.ic action, 371.
CP(N-1) modd, 881. dynamk oorrelation functions, 37'2.
criticai dimenBion, lower-, 667, 757. dynamic crítica! exponent >:, 768.
critical dimension: upper-, 583, 610~ dynrunic ><:o.líng, 767, 777.
criticai domain, 573, 564. Dyoon-Sehwinger equation, 99.
criticai e:xponent;- see exponent, 550.
eikona.l, 34.
criticai slowíng do.,m, 764.
Einstcin's action, 514..
criticai aurrace, 568.
electromagnetic decay of the 1f'o meson, 471.
criticai t.empera.ture, 548.
electromagnetlc hamiltonianj 44.
crossíng symrnetry, l-47.
electron-positron annihllation, 760.
(:rQssover ex:ponent 1 600.
Elil2ur's conjecture, 309, 678.
current 1 295_
Elilll:ur'e theorem, 735, 741.
rurrent non~conservation, 480.
energy momentum tensor, 297~
cu...Uure tensor, 409, 443, 507, 733.
epsilon expansíon, 574, 597.
cyclic ldentity, 509.
equation of motlon, 255.
cylindrical geometry, 789.
equation of state, 550, 555, 674.
decay of Lhe fa.lse vacuum, 348. equation of state: S<:aling, 594:
decay rate, 814. equilibrlum distribution, 70.
deep ine}llStÍc scattetíngt 760. ergodicity, 529, 538.
degenerate mínima, 854, 873, 907. euclidean action, 20, 92.
degenerate minima.: general potential, 919. euclidean f<>rmillation, 17.
detailed balance, 73, 87. Euler-Poincan! character, 522.
detenninant of oper-atOts, 6, 23, 76. evolution operator, 18, 91.
dífferential formst 502. exc.lptional moment.., 213, 215, 259.
differentlal formst closed, exa.ctt 7. eJ<j)<lnent a, 592.
dilatation current1 298.. exponoent a: a =O, 606.
dilute gas appr<>xímation, 91 L exponent (3, 550, 594, 654, 67S.
dimension of composite operators, 175. exponent 6, 550, 594.
dimeru:ion of fields, 168. exponent A,, 632.
dímoension of vertices, 169. apon<:nl 1), 552, 578, 661, 677.
dimensional regularization, 183, 307. -exponent -y, 550, 591~ 663.
dim"""iooal <egularization: fermíona, 187. exponent v, 55] 1 580, 663, 6'75, 677.
 1003
Indu

expooent wt 577. 663. gap "CJuation, 951.

exponent 8~ 632 1 663. gauge coupllng to )eptom, 461.
exponents; classical values, 551. ga.uge coupling ta quarlw, 463.
gauge dependence, 423, 440.
Faddeev-Popov '11ghost::;" t 447.
gauge field, 409, 442, 512.
Foddeev-Popov qua.ntize.tion, 448.
gauge field fermion verta:, 470.
Fenní constant, 462.
gauge independence, 497.
Fermi model of low energy che.rgsd wee.k
ga.uge invarianc:e) 44, 512.
int.eractions1 462.
g.o.uge invariant ~tion, 444.
fennion loops 1 106.
gauge inve.ri8JII operators, 425, 497.
fermion mass generatíon, 682, 715.
ga.uge seclíon1 445.
fermion propagator, 105.
gauge symmeb'y, 409.
fermiom, 56t ] 04.
gauge theories, 404, 441.
fermions: Jarge arder behaviour, 875.
ga.uge transformatione, 442.
fermlons: proper ve:rticest 131~
ga.uges, 401.
Feynma.n diagrams, !:17, 122, 451.
gauges: covariant, 446~
Feymmm gauge, 407t 4.SL
gauges: general, 4!)2.
Feynman pa.rBllletm;, 235.
gauges; linear, 496.
Feynman-1<..: formula, 17.
gaussia.n fixed point, 571.
lield renormalization, 150, 151, 196.
gaussían integro.], 1, 3, 4, 12, 22, 45, 58, !M, 104.
field renormalization constant,. ] 52.
gaussia.n random wa.lk, 644.
lie)d RG function: positívlty, 241, 249.
geussiaD white noise, 67.
lieid theory: holomorphic representation, ]43.
Gau~'s law, 4141 446,
Fien transformBtion, 117.
General Relativity, 513.
finite si:Ze: hypercuhic geometry~ 783.
~enera.ting functianal, 93} IlB.
finite J5Íze scaling, 778~
ghost .equa.tion of motian1 496~
lirst order phase transition, 548.
ghoot: quartic tenns, 495.
fixed poínt ha.miltoniM, 567.
global symmetry, 271.
flavout quantum number, 466l 75S..
gluonium, 73L
Flory's ap-prmdma.tíon, 646.
gluons, 466.
lluctuation-<impation theorem 766
Goldbetger-Treiman relation, 289, 292.
fiux tube, 741. ' •
Goldstone bosons, 279.
Fock sp..::e, 142.
Fokker-Plandc "'luation, 7{1, 78, 370• Goldstone fennions, 391.
Fbkker-Planck hamíltonian, 802. Go]dstone mod..,, 3112.
Fourier transform, 126. Gr,.,.,mann algebra, 105, 108.
ftee energy, 533, 547. Grossmann a.lgebra, 6, 56.
ftee field Lheory, 94, 140. Grlll3Smann algebra: differentíalion 7
functional differentiation, 5, 26, 68. Grassmann algebra: gausslan ~a;lon, 12.
functional integml, 91. Grassmo.rm algebra: inte,gration, 9.
functional integml' perturbative definítion, 98 • Gr,...mann ooordina.te, 356.
functional meaoure, 25. gTBVJty< classical equationo, 514.
Furryts theorem, 429, gTavity~ qua.ntization, 516.
Grlbov's amblguity, 369, 449.
gammn ma.trices: traces1 115. Gros&--Neveu model, 189, 682.
garru:nas matrix, 108. c,_-Neveu model, chiral invariant, TI 4.
1004 Indu

Groos-Neveu model: línearized, 682. jocobian, 5, !0, 819, 828.

Jooephson•s parometri2ationl 627.
bamiltonisn form.W.m, 444.
Ham.ílton-Jacobi equations1 834. Kãllen-Lehmann representation, 39, 150.
harmonic osciUatO'f1 2L Kobayaohi-Maskawo. ma.trlx, 464.
Heisenberg representation~ 29. Kosterlitz-Thoul.,... phase tnwsltion, 715, 718.
heliurD Buperftuid transition, 638.
Landau-Ginzburg-Wilsoo bam.iltonian, 572.
hennitian canjugation, Jl3. La.ndau 's gauge, 407, 451.
hermitian hamíltonlan, 73. Landau's ghOilt, 900.
hierarchy problem, 389, 465. Larnlau 'a theory of phase transitions, 5&3.
Híggs maas: bound, 753.
Langevin f!<luation, 57, 370, 764, 865.
Higgs mechanism: abelian, 430.
Langevin "'luat.ion: discretized, 80.
Higgs m~anism, boson, 452, 460.
l....angevín equatJon~ dissipative> 377.
higb temperature oeries, 545. large field behaviour, 878.
hígber spíns: renonnallzability, 409.
]arge momentwn bebr:nriour ~ 748.
holomorphíc represEiltB.tíon, 51.
largo N llmit, 522, 652, 679, 684.
holomorphic representation: field theory, 140. large arder behaviour, 871,872,898,914,921.
holonomy group, vuríables, 509.
large order estímate, 878.
homogeneous CS equatJon, 213~
lattíce gas modo!, 649.
bomogeneous RG f!<lUations, 217.
lattice gauge theories, 73!.
homogeneous sp&ee:l:it 321, J4fL
lattice QED, 739.
homo~eneous spaces~ non-oomp11.Ct, 341.
la.ttice regularizatíon, 180, 307, 420.
homot.opy classes, 862.
Legendre transformation, 91, 122.
Hubbard's transfonnation, 546.
leptons, 461.
hyperscaling relatwno, 609.
Lie a.lgebras' conventions, 270.
improved action, 617. light cone expa.nsion 1 267.
infinil.e!limal chonge of variables, 101. linear differentia.l n.pproximants, 889.
infrared: se<! IR. linear u-model, 287.
insta.nt.on, 478, 825, 84!. linear sym-try bl"t!aklng, 274.
iru;t.ant.on inter&.ctjon, 927. lin~ly rea.lízed oontlnuous symmetries, 269.
instanton solutJonsJ 812. Líouvllle meaaun, 42.
insta.ntons a.nd renormalization, 845. liqwd-vapour phase t.ransitíon, 638, 649.
inslantons: CP(N-l) model, 861. little group, 3()3.
instant.ono: general scalar lield theories, 346. local group transforrnations, 441.
inslantons: SU(N) model, 663. ]ocality, 18, 90, 118.
integrahle lield theorie'" clas:sical, 836. localization, 681.
integra.ble hamlltonian systems; clossica.l, 834. loga:rithmrc corrections; 610.
fnteraction representation> 31. 38~ Joop expansion, 138.
interl8(;t! model' see ahnost planar. ]ow temperature ex:pansion, 667..
IR di'iergences, 308. LuLtinger Iiquid, 730.
IR regulator, 451. me.gnetic monopole, 66, 739.
irrelevant operator, 570, 613. mognetic suaeeptíbilíty, 550.
J8ing; model, 531. many-body quantum mechanics, 158.
lt.ô calculus, 83. marginal opeTator, 570.
JBI:Obi identlty, 494, 508. Markuv's p~, 67, 77, 85.
 1005

m""' insertíon, 213. overlapPing dlvergencee, 201.

mass operator, 125. Padé approxlnuwts., 883.
mBSS r.enonna.lizationt 196. Palnlevé I equation, 1;28.
=lesa field theoríes, 213. parallel transport, 441, 506, 732.
matrix modeiB, 523. pBrity: see rei!ECtí<>I1S.
mean lield .a.pproximation, 544. partition function, 17, 25, 27, 29.
mean field theory: gauge model, 741. path lntegrnl, 18, 21, 41, 59, 75, 80.
mean field theory: general fotrnalism, 560. Pauli matrices, 109.
Mermín-W~~gner--ColemWl lheotem, 53!l, 654. Pauli-Vi!laro's regul&íza.tíon, 177, 306, 417.
metric tensor, 79, 508~ PCAC,288.
mínima! subtniCtíonl 168, 232. percolation, 559.
minimal subtrection: modified, 236. perlmeter la.w, 738.
Monte Corlo methods, simulations, 743, 764. perlodic boundary conditions, 25, 27~ 29, 778,
multi~instanton, 907_ 782.
multiplicative renonnalization, 213. perturbation theory, 3, 14, 26, zr,38, 59, 77, !}4,
105.
N !>Dlbu--úoldstone mod€8, 280.
pfuffian, 13.
neut.Ta.l currenta, 463.
phase space path integral, 40.
nilpotency, 491.
ph""" tr~V~Bition: breaklng of ergodicity, 529,
Noether's theorem, 295.
538.
noise fie!d, 370.
plaquette, 420, 733.
non-abelia.n anomaly, 478.
Poisson brockeUI, 42.
n<>n· Botei summability, g29,
Poisson's fonnula~ 811.
non·linear u-model, 302, 668.
polymen!, 638.
non~linear a-model: finite size efl'ect.s 1 794.
polymen!: statistical properties, 644.
non-linear u-model: 0(2), 718.
positivity of the field RG function, 241.
non-locaJ coru;erved currenta~ 334.
Potts model, 559.
non· relatívistic limit, 140, 156.
power counting, 173.
non-relax.a.tiona.l modeiJ 774. power lnw bohaviour, 578.
non-renonnaliza.bJe theorjes 1 172.
prOpOJl;ator, 95.
normal-arder, 52t 54, 59~ proper tirne method, 177.
normal-ordered vertex, 224. proper time regularization. 190.
0(2) non-linear u-model, 718. proper vertice:s, 122~
one particle irreducible (IPI), 133. QED: compact, 739.
one part.icle irreducible functionaJ: phy;.ical QED: large order behaviour, 879.
int.erpret:ation, }35. QED: one-lDop P..function, 428.
opend.or inse:rtíom CS equations, 254. QED, ZA = 1/Z., 422.
opera.tor l:nSertionr re:normaliza.tion, 251, 255~ qua.<batic symmetry breaking, 280.
operator ordering: di...,..gence, 46. quantiz.a'til"'D., 78.
operotor product exp!O:I>SÍon (OPE), 267. Quo.ntum Chromodyna.mics (QCO), 466, 73L
opEJ"ators~ powu oounting, 174~ Quaotum Electrodynamics: see QED.
order dependent l)lapping, 885. quantum gravlty, 500, 51fô, 518.
order pa:rameter, 529. quo.ntum gravity: dlmeDBion 2, 522.
Ornstein end Zerníke form, 551. quantum Jump, 84.9.
orlhogonal poJynomiaJo, 526. quwttum olatistkal <>perat.or, 00.
1006 lnda

qua.rk generaíiOJlBt 464. rigid symmetry, 271.

quark m......,., 464. roughenlng tmnsit.íon, 740.
guart.ic mbarrnonic osdllator, 812.
saddle point, 3, 58.
quenched a.pproximation, 744.
scalar curvature, 346, 609.
quenched average, 541.
scale in....-acianoe, 890.
rBDdom magnetíc field, 368. scaling equation of st..te, 550, 626.
random matricee, 523. scaling laws, 589, 592, 593.
ratio of amplitudes, 551. Schwinge:r model, 701.
real time quantum mechanJcs, 30. screening lengtb, 722.
redundant opera.tor, 57t:l. second order phaae transitiont 548.
rellection hermiticlty, 114. self-e.voiding random wa.lk, 644.
reflecliono, 112. self-<luality equations, 862.
regularizaf.íon, dimensional, 183, 307. ó<lilidasoica.l expansion, 127.
regularization: lattk.e, 180, 307, 420. semíd.as:sical vacuum, 864.
,.,gulàrlzo.tiom non-abelian gauge theories, 45l.shífting method, 827.
regulari•ation' non-linear a model, 306. short distance fO!pansion (SDE), 259, 602.
regularioation: Paulí-Villan;, 177. ellort distance e.'<pMSÍou: RG equatious, 263.
regulari.W.ion: proper time, 190. sigma model, 287, 302, 667.
regulator fields, 178. sine-Gordon (SG) model, 699, 717, 722.
Telaxation rates, 764. skelet<>n di..gram, 199.
releva.nt operator, 570. Slavnov-To.ylor (ST) symmetry, 352, 374, 486.
renorma.lízahle theories, 173. S~rnat.-ix, 31, 38, 93,

renonnalízation, 168. s.matrix: fuoctiona.l integral, 140.

renonna.lize.tion conditions, 198. S-m~>trix: gauge independence, 497.
renorrua.líza.tion constant.s, 196, S-matr.bt~ perturbative expan&ion, 31.
renorma.liretion gl'OUp, ( - also RG), 566. Sobolev ínequa.lities, 851.
renormali:r:a-tion; muJtiplica.tivet 213. :ooft brea.lcing, 269.
renorma.litation of gauge theories, 483. soliton, 7!7, 812.
renorma.ljZ8j a.ction, 196. solitoo scatlering, 148.
renorma.lized wrrelation Iunctíoru;l 202. specific beBI exponent .:., 592.
renormalízed field, !96. spin oonnectlon, 512.
r€llonna.lízed fonn of the action, 452. spin group, 1!0.
reJJOTma.lízed parameters, 195. spin-statistics connection, J04l 155, ]6:2.
tenonnalon, 663~ 901. spinors, I Hl.
replica trick, 542, 647. .spontaneous magnetization, 550.
response functlons, 765. spontaneous symmetry brealúng, 278, 3()5, 854.
RG P-fun<:tion, 338, 468. stabi!ízer subgroup, 303.
RG equstlons, 466. staggered fermions, 183.
RG equations: b!ITe, !174. Standa.rd Model ofweak ood electromagnetic in-
RG functions, 24, 244. temclions, 459.
Ricci tensor, 509. eta.t.istical mechanics: quantum1 classical~ '28.
Riemann tensor, 507. steepest descent method, 3, 127.
Riemannian monifold, 45, 78, 226, 321, 341, ste~graphíc mapping, 905.
385, B60. :stochastic rlílfe:reníial eqna.tioru;~ 67.
rigld rota.tor, 47, 48, 797. stochastíc dynamk:s, 448.
 Ind"'C 1007

stochastk: field equation, 365. transport coefficients, 764.

stochastic proc:ess: phase transiUon, 538. lree }evel &ctjon, 196.
otoclJBSti<: variables, 359. tricritical behaviour, 623.
Strat.a.novich CC'.Invention, 78. tricritical point, 557.
streaming term, 764. trivlality of the 4>: field theory, 750.
stress tensor, 299. tríviality problem, 289, 532, 7 48.
string tension) 740. twlst, 267.
strong ooupling expansion, 738. twlst.ed boundary condítions, 783.
strong CP violation problem, 865. two--i:nstanton configtuation, 008.
strong scalíng, 587. two--point correla.tion functíOn 1 39, 246.
swnmat.ion method.s 1 863_
U(l) problem, 400, 855.
super-diagra.m.s 1 398.
uniaxi;al ferromagnet, 617.
superficial degree of divergente ( Feynman
unitarity, ]45, 156.
diagram), )70.
unitaríty relatíons, 425.
supomeld, 377.
unitary g~>nge, 407, 431, 454.
supomeld: SC«<tu', 391.
universal critkal behaviour, 574T
supafield: vector, 393.
unlversality, 529, 568, 587.
super-renorma.lízable field tlH<ories, 172, 223.
universality c].......,, 568.
supersymmetrie: ac:tion, 392.
unstable vacua in field theory, 340.
supersymmetric curvature tensor 1 394.
UV divergences, 164.
supersymm~&tüc f.ree gauge action, 395.
UV fixeà points, 748.
supe.rsymmetric gauge trans!onna.tiont 394.
UV singularitíes, 164.
supersymmelry, ]55, 364, 317, 766.
fõupel"5ymmet.ry in four dimensions, 389. vacuum, !)JT
surface tension, 809. vacuum energy, 437.
symmetric space, 332, 453, 680. VM VIEck'o formula, 830, 838.
~ymmetries; linear realization, 269. vec:tor Held: maBSive, 404, 454, 457.
symmetry breaking; roft, 285, 269. vielhein, 79, 511.
symmetry: 1ocal, gauge, 409. volume element~ inva.riant 1 5]0.
symmetry resto:rat.on~ 859. vortex excito.tions, 72ft

't Hooft's ga.uge, 433, 455. Ward-Takahashi:- WT identíties.

tempora.l gauge, 413, 416, 445, 736. weak angle Ow, 461.
tensotst 500. weak lnteractions; see Standard Model.
thsta-vacuum, 864, 917. Weinberg's theor<mi, 213.
Thirring model, 704. We53-Zumino cormistency conditions, 481.
Thirring model: SU (2) in..,.,.íant, 708. Wess-Zumino (WZ) gauge, 394.
Thirring model: SU(N), 114. W -g,.uge boson, 461.
thne-ordered produd, 29, 38. Wheeler-DeWitt equation, <>18.
topological oction, 62. W.ick.~s t.heoTern, 2~ 41 14, 26~ 95.
Lopologica.l charge, 862. Wigner qutmtization, 41.
topological propertíes, 478. Wil.<;on's fennions, 1B2.
lop~>logica.l quantity, 863. Wil.<;on'• loop, 731.
torsion~ 504T wínding number,. 863.
T-product: see tlme-ordenod, 38. WKB expansion, 931.
transfer ma.trixí 19, 531. WT identlli..,, 269, 272, 309, 326, 330, 37!1, 452.
XY modek classical 2D, 728. Yukawa type inte:ri>Ction, 245.

Yo.ng-Lee edge singu!wit,y, 164.

Yang-Mills lield, 442. z.-ga.uge booon, 461.
Yukawa roupllngs, 464. z.ero modes, 782.