COURSE3

GAUGE THEORIES

BenjaminW. LEE
Fermi National Accelerator I,aboratory,
Batavia, III. 60510

R. Balinn and J. Zinn-Justin. edr, Les Houches. Session XXVIII, 1975 - Mkthodes en
theories des champs /Methods in fietd thmry

PUB-76/34-THY

Contents

1. Yang-Mills ficlds
1.1. Introductory remarks
1.2. Problem: Coulomb gauge
2. Perturbation expansion for quantized gauge theories
2.1. General linear gauges
2.2. Paddecv-Popov ghosts
2.3. Feynman rules
2.4. Mixed transformations
2.5. Problems
3. Survey of renormalization schemes
3.1. Necessity for a gaugcinvariant rcgularization
3.2. BPllZ renormalization
3.3. The regularization scheme of t.Hooft and Vcltman
3.4. Problem
4. The Ward-Takahashi identities
4.1. Notations
4.2. Bcccbi-Rouet-Stora transformation
4.3. The Ward-Takahashi identities for the generating functional of Green
functions
4.4. The Ward-Takahashi identities: inclusion of ghost sources
4.5. The Ward-Takahashi idenlities for the generating functional of proper
vertices
4.6. Problems
5. Renormalization of pure gauge theories
5.1. Renormalization equation
5.2. Solution to renormalization equation
6. Renormalization of theories with spontaneously broken symmetry
6.1. Inclusion of scalar ficids
. 6.2. Spontaneously broken gaugesymmetry
6.3. Gauge independence of the S-matrix
References

19
79
81
81
81
86
88
91
93
94
94
95
99
107
I07
107
109
111
112
114
117
117
117
120
124
124
127
130
135

PUB-76/34-THY
79

Gauge theories

1. Yang-Millsfields
I. 1. Introductory remarks

ProfessorFaddeevhasdiscussedthe quantizationproblem of a system

which is describedby a singularLagrangian.For the following, we shall assumethat the student is familiar with the path integralformalism, and the
quantizationof the Yang-Millstheory. The following remarksareintendedto
agreeon notations.
The Yang-MillsLagrangian,without matter fields, may be written as

For simplicity we shall assume,theunderlyinggaugesymmetry is a simple

compact Lie groupG, with structure constantsfabc.
The Lagrangian(1.1) is invariant under the gaugetransformations,

= U(e)[L,Ap)

-f

o-(e)all

U(E)] t+(E)

>

(I-2)

wherethe E arespace-timedependentparametersof the groupG, U-(E)

= M(E) and the t are the generators.
Thesegaugetransformationsform a group,i.e., if gg = g, then

(Problem: prove this statement.)

The infinitesimal versionof the gaugetransformationis
L,,SA; = +,a/

- fabcA;ebLC

Of

SA = -.-iaMea +fahcebA;.
P

(1.3)

It is preciselythis freedomof redefmingfields without altering the Lagrangian that lies in the heart of the subtlety in quantizinga gaugetheory. In the
languageof the operatorfield theory, to quantizea dynamicalsystemone has
to find a set of initial valuevariables,ps and qs, which arecomplete,in the

PUB-76134-THY

B. iv. Lee

80

sensethat their values at time zero determinethe valuesof thesedynamicai

variablesat all times. It is only in this casethat the imposition of canonical
commutation relationsat time zero will determinecommutatorsat all times
and define a quantum theory for a gaugetheory. This can neverbe done becausewe can alwaysmake a gaugetransformationwhich vanishesat time zero.
That is, it is impossibleto find a complete set of initial-valuevariablesin a
gaugetheory unlesswe removethis freedomof gaugetransformations.
To quantizea gaugetheory, it is necessaryto choose,agauge,that is, imposeconditions which eliminate the freedomof making gaugetransformatians, and seeif a completeset of initial-valuevariablesexist.
Thereis a specialgauge,called the axial gauge,in which the quantization
isparticularly simple. It is definedby the gaugecondition that
?pd(x)
P =0,

(1.4)

whereTJis an arbitrary four-vector.In this gauge,the vacuum-to-vacuumamplitude can be written as

ejW=Nf[cU~]

II
w

S@(x)-rj)

whereN is a normalizingfactor.
Thereis in principle no reasonwhy eq. (1.5) cannot be usedto generate
Greenfunctions, by the usualdeviceof addinga sourceterm in the action.
That is, WC define the generatingfunctional of the connectedGreenfunctions
WAL;l by

X exp {iJd4x{J?(x) + $(x)A pa(x)]) .

(l-6)

However,the Feynmanruleswould not be manifestly Lorentz covariantin

this gaugeand it is desirableto developquantum theory of the Yang-Mills
fields in a wider classof gauges.
As ProfessorFaddeevexplained,eq. (1.6) is an injunction that the path integralis to be performednot over all variationsof A:(x), but over distinct orbits of A:(x) under the action of the gaugegroup. To implement this idea, a

PUB-76/34-THY

Gaugetheories

81

hypersurfacewaschosenby the gaugecondition 11-A a = 0, so that the hypersurfacein the manifold of all field intersectseachorbit only once.The
problemwe poseourselvesis how to evaluateeq. ( 1.6) if we are to choosea
hypersurfaceother than the one for the axial gauge.
1.2. Problem: Coulomb gauge

The gaugedefined by ViAf = 0 is called the Coulombgauge.In this gauge

the two space-liketransversecomponentsof Ai are the qs, and the two spacelike transversecomponentsof F& arethe ps,

Expressthe Lagrangian( I. 1) in termsof the CoulombgaugevariablesA!;

0; rif; f"; F;: 81;. Referringto ProfessorFaddeevlecture,
s
comfor this gauge.

."n
2. Perturbationexpansion for quantizedgauge theories
2.1. General linear g(wgcs

The foregoingexample,the axial gaugecondition, is but one of the ways to

eliminate the possibility of gaugetransformationsduring the periodthe temporal developmentof a quantizedsystemof gauge fields is studied.Clearly
this is not the only way, and in fact, we could define a gaugeby the equation
F= [A;, cp]= 0

for all a ,

(2.1)

provided that, givenA,, and other fields which we shall collcctivcly call p
thereis one and only one gauge transformationwhich makeseq. (2.1) true.
For convenience,we shall dealonly with the casesin which Fa is linear in the
bosonfields Ai and p. In this lecture, however,we shall be concernedprimarily with the instancein which Fa dependson AZ alone.
Before proceedingfurther, let us pausehere briefly to reviewa few facts
about gloup representations.Let g,gE G. Then ggE C and

PUB-76/34-THV

The invariant Hurwitz measureover the group G is an integrationmeasureof

the groupmanifold which is invariant in the sensethat
dg= d($g) .

(2.3)

If we parametrizerl(g) in the neighborhoodof the identity as

W(g) = 1 + i@L,

+ O(f2) ,

(2.4)

g=l.

0.5)

then we may choose

dg= ? de=

Considernow the integral

where(A,h(x))g denotesthe g-transformof A:(x), as defined by eq. (1.5).

The quantity AF[Ai] is gaugeinvariant, in the sensethat

A, [(-4j)gJ =j c dg(x) n S(P[(A;(x))ggJ)

0,x
=s n d@&(x) n 6 (P [@,h(x)@])
X
ax
=$

l-J &I-$ g9 6wa [(AiybWl)

= A, [A;],

(2.7)

wherewe madeuse of eq. (2.3).

According to eq. (l.S), we can write the vacuum-tovacuumamplitude as

where5 = f d4xJ(x) is the action. Since

PUB-76134-THY

Gaugetheories

83

wemay rewrite eq. (2.8) as

eiW =N

j@4AF[Al~fl

dg(x)~ S(FIAg(dl)

X by 601 -AbO))expWWII.
,

(2.10)

In the integrandwe can make a gaugetransformationA:(x) + (A,b(x))g-.

Under the gaugetransformationof eq. (l-5), the metric [dA], the action S[A]
and AF[A] remaininvariant, so eq. (2.10) may be written as
eiV=NJ[dA]A,[A]tinx

s(Fg[A(x)])eiS[Aj
(2.1 J)

Let us assumethat

Then we have

which is a constantindependentofA. Therefore,this constantmay be ab

sorbedin N, and
,iW =N &-~]A~[A]

II 6(F[A])eiS[Al.

(2.12)

This is the vacuum-to-vacuumamplitudeevaluatedin the gaugespecifiedby

eq. (2. I>.
Let us evaluateAF [A]. Sincein eq. (2.12) this is multiplied by II6 (F, [A]),
we needonly to know AF[A] for A which satisfieseq. (2.1). Let us make a

PUB-76134-THY

gaugetransformationon A so that F= [A] = 0. For g in the neighborhoodof

the identity, then

(2.13)
whereDp is the covariantderivative
qb =6a,bap -gf,&;

(2.14)

Therefore,from eq. (2.6), we seethat

where

(2.16)
That is,
(2.17)
Mere,we can afford to be sloppy about the normalizationfactors, as long as
they do not dependon the field variablesA,,.
The factor A,[A] can be evaluatedfrom eq. (2.17) for variouschoicesof
I? The examplewe will consideris the so-calledLorentz gauge,

PUB-76134-THY

85

Gauge theories

FU = afin;

+ P(x),

whereC(X) is an arbitrary function of space-time.Under tire infinitesimal

gaugetransformation(1.3) ofA:, F, changesby
SFQ=--(6

ar
g

AC)eb
ab aP -gf ubcp

so that
bz,XlM~lb,J4

= -aiD%(x
Y

- y) -

The appearance
of the delta functionaJJJ6(Fa[A]) makeseq. (2.12) not
very amenableto practical calculations.We could havechosenas gaugecondition:
(2.18)

F[A]-c(x)=O,

with an arbitrary space-timefunction co, insteadof eq. (2.1). The determinant

AF[A] is still the sameas before, that is, is givenby eq. (2.17)and clearly the
left-handside of eq. (2.12) is independentof co. Thus, we may integratethe
right-handside of the equation
eiw =N[[dA]AF[A]

n 6(F[A]

(2.19)

- cO)eis~AI

overc,(x) with a suitableweight, specifically with

exp(zjdx

c:(x)) )

where01is a real parameter,and obtain

eiw =NJ[d4]AF[A]exp

iS[A] - $sd4x(F[A])2

(2.20)

Eq. (2.20) is the starting point of our entire discussion.We define the generatingfunctional W,[J,] of the Greenfunctions in the gaugespecifiedby F
to be

(2.21)

PUB-7F134-THY

86

B.W.Lee

Pleasenote that the aboveis a definition. We havenot answeredyet how Green

functions in different gaugesare relatedto eachother, or to the physicalS-matrix. We shall return to thesequestionsin a future lecture.
2.2. Faddeev-Popov ghosts

As we havenoted in the precedingsection, AF[A] hasthe structure of a

determinant.Such a determinantoccursfrequently in path integrals.
Considera complex scalarfield p interacting with a prescribedexternal potential V(x). The vacuum-tovacuumamplitude is written
eiw =Nj[dlp]

[dvt]exp{iJd4x$(x)[-82

-p2+ V(x)]~(x))

(2.22)

= [-a2 - j.2 t V(x)]S4(x - y) -

(2.23)

- @et M(x,y))-

where
M(x,y)

On the other hand, we can evaluateIV in perturbation theory: it is a sum of

vacuumloop diagramsshownin the following figure:
v
W=V

o+v

ov+v

v +-.

This result can be understoodin the following way. We write

(det M(x,y))- = (det MO(x,y))-
X tdet [S4(x - Y) + A,& - Y) W)J)-*

(2.24)

where

q$GY) = t-a2 - P2P4(X-y) ,

1
-a2-p2tie

Y *
I>

(2.25)

O-26)

PUB-76/34-THY

Gauge theories

87

(The ie prescriptionin eq. (2.26) follows from the Euclidicity postulateinherent in the definition of path integrals.Seerefs. [1,2],The first factor on the
right-handside of eq. (2.24) may be absorbedin the normalizingfactor N. The
secondfactor may be evaluatedwith the aid of the formula
det( 1 + t) = exp Tr In(l t L) .
Thus,
iw= -Tr In(1 +A,V),

(2.27)

which showsvery clearly ?Vas a sum of loops.

Next, what if p and (p; were anticommutingfields?Nothing much changes,
except that eachclosedloop acquiresa minus sign. Thus, if p and V? areanticommuting fields, we have
eiCV = N

l [dvl[drltlexp(iSd4xl.lt(x)[-a2
-P + W)l~(x)~

- det M(x, y) - exp {Tr In( 1 + AF V)) .

The above.isa heuristic argumentof how integralsover anticommutingc-numbersshould be defined to be usefulin the formulation of field theory. In fact,
Berezindefinesthe integralover an elementof Grassmann
.algebraci as

dcr = 0 p

dCicj = liij .

It then follows

.v

dci eciijci

(det A)j2 .

(2.28)

(Prohlern:prove this statement.)We shall not dwell upon the integrationover

Grassmannalgebraany further, but rather refer you to Berezinstreatiseto be
cited at the end of this lecture. A nice mnemonicfor the rules of integration
over anticommutingc-numbersis, as told to me by JeanZinn-Justin,that integrationis equivalentto derivation.
For our purpose,we can write

PUB-76134-THY

or symbolically

AFL41=ffj [@Ib-hlexp
IiEMFvI,

(2.30)

wherega(x), Q(X) areelementsof Grassmannalgebra.

Note that the phaseof the exponent,$.MFq is purely conventional.
The generatingfunctional WF[$] of eq. (2.21) can now be written as
exp{ilV&~]}

=$

[dAd[dq]

wherethe effective action Se, is given by

The fields t, n areusually called the Faddeev-Popov

ghost fields. They areunphysicalscalarfields which anticommuteamongthemselves.(Sometimesit is
convenientto think of t as hermitian conjugateof n, but it is not necessary.)
In the Lorentz gauge,wherewe shall write

the term in the effective action bilinearin 2:and 7 is

d4x $5
=

Dib qb(x) =$d4x ap .&(x)D;~ nb(x)

cc

d4x[V&(x)ar

r&(X) - ga.&(x)F,b&x)Vb(x)]

(2.33)

Even when we regard$ and v as a conjugatepair, the interaction of eq. (2.33)

is not hermitian. The sole raisonde^treof this term is to createthe determinantal factor, eq. (2.29).
2.3. Feynnranrules
To describethe Feynmanrules for constructingGreenfunctions in perturbation theory, it is more convenientto couple & and nQalso to their own
sources0, and pi, which are anticommutingc-numbers.We define

PUB-76/34-THY
89

Gaugetheories

,
)I
wherewe havesuppressed
gaugegroup indices.
The Feynmanrulesare obtainedfrom eq. (2.34) in the usual way. We will
reviewbriefly the derivation of the Feynmanrulesin a simplerexample,an
interactingreal scalarfield V. The action S[cp]is dividedinto two parts,

w = qJ[IPI+ s, M *

(2.35)

whereSO[p] is the part quadraticin the field (B,and hasthe form

S&l =~d44f(acd2 - b2v21.

(2.36)

The generatingfunctional i W[J] of the connectedGreenfunctions is givenby

eiwfJ] =J[dlp]exp{iS[lp) + isd4xJq}

Nr
Therefore,we must now compute

[dpjexp IiS0 [up]+ i

eiwo[J1 = NJ [dq]exp {iSo [p] + ild4xJq}

d4xJp} _ (2.37)

(2.38)

SinceSo is quadraticin p, we can perform the integration.

The functional integralin eq. (2.38) gainsa well-definedmeaningby the
Euclidicity postulate, that the Greenfunctions eq. (2.37) generatesmust be
the analytic continuations of the well-behavedEuclideanGreenfunctions.
We obtain
ei~o[JI

where

= exp{-jji

s d4xd4~JW&

-Y)JW,

(2.39)

PUB-76/34-THY

Lt.W. Lee

90

Eq. (2.37), or
exp{il.Vo[s]) ,

exp {i W[J]) = exp

(2.37)

may be transformedinto a perhapsmore tractableform by the useof the formula

F(-i$-)G(x)=C(-i$-)FCy)eX.Yly;O,

which can be provedby Fourier analysis(seeref. [2]),

J

d4x d4u *,(x

X exp OS, [p] + i~d4xJlpl~7=0

&y

- v) &q

(2.4 I)

In much the sameway, we can developthe Feynmanrules for the gauge

theory from eq. (2.34). To be concrete,let us adopt the Lorentz gauge,
F, = -ap-4:. We defineSOto be

taP~Qall~=ti:.ptpt-fl-JP.AP

(2.42)

The remainderof the action S consistsof the cubic and quartic interactions
of the gaugefields and the interaction of the gaugefield with the ghost fields.
The Feynmanpropagatorfor the gaugebosonssatisfies
$,a,

( )I
1 -k

AF(x -u) =g;a4(x -u) ,

(2.43)

and is given by
A;!(x - r) =s - d4k

(W4

Note that in this gaugethe ghost field ,$Yalwaysappearsas ap $?. The generating functional IV, [J,,fl,@] can be written as

PUB-76/34-THY

91

Gauge theories

exp{ij~LIJP,P,P
=fll
exp(is d4xd4+%w)6~~
6
+DF(x-)w@@

6
1)

&$+--r)

exp(i tS1Pp,-5rll
(2.45)

whereDF(x - u) is the Feynmanpropagatorfor a massIess

scalarfield,
&
DFx

-~=S~2

,-ik:

b-y)

k2 $ ic

That is, in this gauge,the Faddeev-Popov

ghostsaremassless.
2.4. Mixed transformations

A few remarkson the integration over elementsof the Grassmannaigcbra:

Sincewe wish to maintain the integration rule

undera changeof variables,

fi

ci=A&

i=l

ci=,&

EjdetA,

we must have
7 dci = (det A)- y dEi Further, let us considera mixed multiple integral of the form
$rhci
i

I-I de,,
tJ

whereBs are elementsof the Grassmannalgebra.We considera changeof integrationvariablesof the form

PUB-76134-THY

B. IV. Lee

and ask howthe Jacobianmust be defined.We considerfirst the change

k 0) + 01,a
x =AY

t drle

- CUB+~Y

= (A - CfB-'p)y

+ aB-'e

Note that a! and p are anticommuting.Thus,

Now we perform the transformation(y, 8) + 0, cp),

B=Dy+Btp.

Since0 and cpareanticommutingnumbers,

As a result, we have the rule

s n dxi n dUP= j n dri n dqP det(A - cuB-l@(det B)- I
The aboveresult is in accordwith the definition of a generalizeddeterminant or Arnowitt, Nath and Zumino. They define the determinantof the
matrix

by
detC=expTrlnC,
with the convention that
Tr C=Cii - CP,, .
This definition allows the following relation to be valid:

PUB-76/34-THY

Gauge theories

93

and thereforethe product property of the determinant

det(C1C2) = det C, det Cz .
seethis, we set Ci = exp Ji, SO that det Ci = exp Tr Jp N O W Cl Cl
=exp{J1 tJztJ&
whereJ12is the Baker-Hansdorffderiesof commutators.
But Tr[JI! Jz] = 0, erc., so that det(CI C,) = exp Tr(J, + J2). Now the ma-

TO

trix Ccan be decomposed

uniquely into the form ST, where
S=

T=

a =A -CUB-/~,

b=arB-I,

u=P,

z= B.

and

Thus
det C= det S det T= (det a)(det z)- .
(The last follows from the definition det C= exp[(ln C)ii - (In C)PP].)
2.5. Problems

2.5.I. One shouldrepeatthe foregoingargumentsfor quantumelcctrodynamits, to obtain the usualFeynmanrulesin the Lorentz gauge.Let us note that
for OL= 0, one getsthe photon propagatorin the Landaugauge;for OL= I, that
in the Feynmangauge.What happensto the Faddeev-Popov
ghostfields in
those cases?

2.5.2. Just for the sakeof exercise,quantizeelectrodynamicsin the gauge

,F= aflAP + XA:. Derivethe Feynmanrules.

2.5.3. Show thatJ u dci n dci exp {ciMV ci)- det M , wherec and care
i
anticommuting. *

PUB-76/34-THY

94

B. W.Lee

3. Surveyof renormalizationschemes
3.I. Necessity for a gauge-invariant regufarization
In this lecture, we will developtwo subjectsthat are neededto understand
later Lectures.ThesearereguIarizationand renormalizationof Greenfunctions
in quantum field theory in general,and of Greenfunctions in a gaugetheory
generatedby the expression(2.34), in particular.
The Greenfunctions generatedby eq. (2.34) are plaguedby the ultraviolet
infinities encounteredin any realistic quantum field theory. We are going to
developa method of eliminating thesedivergencesby redetinitions, or renormalizationsof basicparametersand fields in the theory, in such a way that the
gaugeinvarianceof the original Lagrangianis unaffectedin so doing.
The gaugeinvarianceof the action.hasvariousimplications on the structure
of Greenfunctions of the theory. The precisemathematicalexpressionswhich
aresatisfied by Greenfunctions due to the gaugesymmetry of the underlying
action areknown as the Ward-Takahashi
(WT) identities. What we will show is
that theseidentities remainform invariant under renormalizationwhich eliminate the divergences.This point, that renormalizationcan be carriedout in a
way that preservesthe WT identities, is of utmost importancefor the followingreasons.First, it puts such a stringent constraint on the theory and the renormalizationprocedurethat the renormalizedtheory becomesunique, once
the underlyingrenormalizabletheory is given. Second,and perhapsmore to
the point, the unitarity ,of the renormalizedS-matrix is shownby the WT identities satisfied by the renormahzedGreenfunctions. The latter point requires
clarification.
In a perturbativeapproach,non-Abeliangaugetheoriessuffer from such
severeinfrared singularitiesthat nobody has succeededin defining a sensible
S-matrix in this framework. Consensusis that a sensiblegaugetheory arises
only in a non-perturbativeapproach,whereingaugefields and other matter
fields carrying non-Abelianchargesdo not manifest themselvesas physical particles. Physically, this conjectureis at the heart of the hope that color-quark
confinementmight arisenaturally from a non-Abeliangaugetheory of strong
interactions.Thereis an exception to this, and this is the casewhen the gauge
symmetry is spontaneouslybroken. In fact, this latter possibility is directly
responsiblefor the revival of interest in non-Abeliangaugetheoriesa fe.w
yearsago,in conjunction with efforts to unify electromagneticand weak interactionsin a non-Abeliangaugetheory. In this casethereis no difficulty in
defining the physical S-matrix, and the unitarity of such a theory is assuredby
the renormalizedversionof the WT identities. Even in unbrokengaugetheory,

PUB-76134-THY

Gauge theories

95

the S-matrix can be definedup to somelower order in perturbationtheory,

and hereagainthe unitarity of the S-matrix is a consequence
of the WT iden.
ti ties.
Why is the unitarity such a big issuein gaugetheory?After all, one does
not worry that much about the unitarity, say, in a self-interactingscalarboson
theory. The reasonis that the quantizationprocedurewe adoptedmakesuse
of a non-positivedefinite Hilbert space,aswe can readily seefrom the structure of the gaugebosonpropagator,eq. (2.44). Further, the Greenfunctions,
of the theory contain singularitiesarisingfrom the Faddeev-Popov
ghostsbeing on the massshell.Thus, in order that the theory makessense,theseunphysical particles, correspondingto the ghost fields and the longitudinal
componentsof gaugefields, must decouplefrom the physicalS-matrix. The
renormalizedWT identities arenecessaryin showingthis.
The WT identities areusually derivedby a formal manipulationof eq.
(2.34). However,the Greenfunctions generatedby eq. (2.34) arenotoriously
ill-defined objectsdue to ultraviolet divergences.It is thereforenecessaryto
invent a mcansof regularizingthe Feynmanintegralswhich define them
without destroyingsymmetry propertiesof the Greenfunctions, so that as
long as we keep a regularizationparameterfinite, the integralsarewell-defined.
It is only then that we can attach concretemeaningto the WT identities. After
renormalization,the regulatormay be removed,and if the renormalization
is to be successful,the renormaiizedGreenfunctions must be finite and independent of the reguiarization parameter.
A well-known regularizationschemein quantumelectrodynamicsis the
Pauii-Viilarsscheme,in which one addsunphysicalfields of variablemassesto
the Lagrangianin a gaugeinvariant way. After gaugeinvariant renormalization
the variablemassesare let go to infinity, and renormaiizedquantitiesare shown
to be finite in this limit. In non-Abeliangaugetheory, this deviceis not avaiiable, but an dternativeprocedure,whereinthe dimensionaIityof space-timeis
continuously varied,wasinvented by the geniusof t Hooft and Vel tman.
In the next section, we will give a brief summaryof the renormalization
theory a la Bogoliubov,Parasiuk,Hcpp and Zimmermann.This will be followed by an introduction to the dimensionalreguiarizationof t Hooft and Veitman.
3.2. BPHZ renormalization

In this section we wig give a brief surveyof renormalizationtheory developed and perfectedin recentyearsby Bogoiiubov,Parasiuk,Hepp and Zimmermann(BPHZ). Nothing will be proved,but we will try to give definitions
andtheoremsin a precisemanner.

PUB-76134-THY

96

8. W. Lee

First, we will give somedefinitions. The interaction Lagrangianis a sum of

terms1r which is a product of 6, boson fields andf,: fermion fields with di derivatives.The vertex of the ith type arisingfrom JZihas the index 6, definedas
ai = b, t zfi t di - 4 = dim(Xi) - 4 .

(3.1)

Let I be a one-particleirreducible(LPI) diagram(i.e., a diagramthat cannot

be madedisconnectedby cutting only one line). Let EB andE, be the numbersof externalbosonand fermion lines, IB and IP the numbersof internal
bosonand fermion lines, ni the number of verticesof the ith type. Then
Ee+21B=&bi,

(3.2)

EF t2$ =CH(.f;:.

(3-3)

The superficialdegreeof divergenceof P is the degreeof divergenceone

would naively guessby counting the powers of momentain the numeratorand
denominatorof the Feynmanintegral. It is
D(r)=&idit21,t31,-4V+4,
i

(3.4)

.the last two terms arisingfrom the fact that at eachvertex thereis a four-dimensionaldelta function which allows one to expressone four-momentumin
terms of other momenta,except that one delta function expressesthe conservation of external momenta.Making use of eqs.(3. I), (3.2) and(3.3), we can
write eq. (3.4) as
D=&$-EB-,1EI:t4,

(3.5)

or
DtEBt;EF-4=&$..

(3.6)

The purposeof renormalizationtheory is to give a definition of the finite

part of the Feynmanintegral correspondingto T,
Fr = jl;

j-dkl

. .. dk, Ir 9

whereI, is a product of propagatorsAF and verticesP,

(3.7)

PUB-76/34-THY

97

Gauge theories

(3.8)

The finite part of F, will be denotedby J, and written

Jr = f$

Jdkl .. . dkLR, .

(3.9)

*+

We shall describeBogoiiubovsprescriptionof constructingR, from I,.

Let us first considera simplecase,in which r is primitively divergent.The
diagramIis primitively divergentif it is proper(i.e., IPI), superficially divergent(i.e., D(F) 2 0) and becomesconvergentif any line is brokenup. In this
case,we may use the originalprescriptionof Dyson. We write
Jr = jdkl

... dkL(l -t&.

i.e.,
R, =(I -t)&

The operationt r must be defined to cancelthe infinity in J,. Jr. is a function

of EF t EB - 1 = E - 1 external momentapl, ... , pEeI,

The operation(1 - tr) onfis defined by subtractingfromf the first D(T) + 1

termsin a Taylor expansionabout pi = 0,
O(P1,

.*- , P&q) =f(O, -0-> 0) + *-*

(3.10)

E-1
id ) v

where(I = D(r). The operation(I - I r, amountsto makingsubtractionsin

the integrandI,, the numberof subtractionsbeingdeterminedby the superficial degreeof divergenceof the integral.
Somemore definitions:,A renormalizationpart is a properdiagramwhich
is superficially divergent(D > 0). Two diagrams(subdiagrams)aredisjoint,
y1 n y2 # 4 if they haveno lines or verticesin common.Let (7, 9... , yC} be a
set of mutually disjoint connectedsubdiagramsof r. Then

PUB-76/34-THY

98

B. W. Lee

is defined by contractingeach7 to a point and assigningthe value 1 to the

correspondingvertex.
We are now in a position to describeBogoliubovsR operation:
(i) if F is not a renormalizationpart (i.e., D(r) Q -l)?
R,=R,,

(3.11)

(ii) if r is a renormalizationpart (D(7) > 0),

R, ~(1 -tr)&,

(3.12)

whereRF is defined as
(3.13)

and Or = -PET, Ehere the sum is over all possibledifferent setsof (7i).
This definition of R,. in termsOCR?appearsto be recursive;in perturbation
theory thereis no problem;the R, appearingin the definition of Rr. is necessarily of lower order.
It is possibleto solve eq. (3.13). We refer the interestedreaderto Zimmermannslecturesand merelypresentthe result. Again we needsomemore
definitions beforewe can do this:Two diagrams71 and 72 are said to overlap, 71 0 72, if none of the following holds:

A P-forestII is a hierarchyof subdiagramssatisfying(a)-(c) below:

(a) elementsof U are renormalizationparts;(b) any two elementsof U, 7 and
7n are non-overlapping;(c) U may be empty. A r-forest U is futl or normal respectively dependingon whether U contains r itself or not. The theoremdue
to Zimmermannis
(3.14)

whereI; extendsovcc all possible(full, normal and empty) r forests, and in

the product [I(-IX) the factors are orderedsuch that th standsto the left of
t if h > u. If h f? (I = 9, the orderis irrelevant.A simple exampleis in order.
Considerthe diagramin fig. 3.1. The forests are I$(empty); 71 (full); 72 (normal); 71,~~ (full). Eq. (3.14) can be written in this caseas

PUB-76/34-THY

99

Gauge theories

-----1
r-7
i
I -; Ql
i Y,
--A

.(

I
I
:

Fig. 3.1. Example of the BPHZ definition of subdiagrams in a particular contribution to

the four-point function in a Aa4 coupling theory.

R, = (1 - t: - tfz t PI tYZ)lr = (1 - tYl)(l - tr2)Zr .

Note that in the BPH program,the R-operationis performedwith respect

to subdiagrams
which consistsof vertices,andall propagatorsin I?which connect thesevertices.I3y the BPH defmition, the subdiagramy2 abovedoesnot
contain renormalizationparts other than itself and in this sensethe present
treatment differs from Salamsdiscussion.
In formulating the BPH theoremit is necessaryfirst to reguIarizethe propagatorsin cq. (3.9) hy somedevicesuchas
A&J)

re; A&J;~,

4 = -i l
I

da exp[ia(p*

- m2 tie)]

and defineI,$-, E) as in eq. (3.9) in termsof A&, E), and then construct
R,(r, E) by the R-operation.The BPH theoremstatesthat R, existsasY-+ 0

and E -+ O+,as a boundaryvalue of an analytic function in the externalmo

ments.Another theorem,the proof of which can be found in the book by
Bogoliubovand Shirkov, sect. 26, and which is combinatoricin nature,states
that the subtractionsimplied by the (1 - tr) prescriptionin the R-operation
can be formally implementedby addingcountertermsin the Lagrangian.
A theory which has a finite numberof renormalizationpartsis calledrenormalizable.A theory in which all 6i arelessthan, or equal to zero is renormalizable.In this casethe index of a subtractionterm in the R-operationis bounded
byD+En +&!+- 4 which is at most equalto zero by eq. (3.5). In sucha theory, only a finite numberof renormalizationcountertermsto the Lagrangian
suffice to implementthe R-operation.
3.3. The regularization scheme of t Hooft and Veltman

Kecently, t Hooft and Veltman proposeda schemefor reguIarizingFeyn-

PUB-76/34-THY

100

B. IV.Lee

man integralswhich preservesvarioussymmetriesof the underlying Lagrangian.This method is applicableto electrodynamics,and non-Abeliangauge
theories,and dependson the idea of analytic continuation of Feynmanintegralsin the numberof space-timedimensions.The critical observationshere
are that the global or local symmetriesof thesetheoriesareindependentof
space-timedimensions,and that Feynmanintegralsare convergentfor suff-ciently small, or complex IV, whereN is the complex dimensionof spacetime.
Let us first review the nature of ultraviolet divergenceof a Feynmandiagram.For this purpose,it is convenientto parametrizethe propagatorsas
(3.15).

A,(p2) = $ i datexp[iol(p2 - nz2+ ie)] .

0

Making use of this representation,we can write a typical Feynmanintegral as

(3.16)

X exp {i C ai((7i2- .rf + ie)} ,

i

whereI is the numberof in ternal propagatorsin r, L the numberof loops, and

1,) ... , 1, may take any valuesfrom I to L. The momentumqi carriedby the
jth propagatoris a linear function of loop momentaki and external momenta
p,nBThe exponenton the right-handside of eq. (3.16) can thereforebe written as
I
C a&qfi=l

mf+ ie) = I CkiAij(a)kj

i,i

t ic m ki Bi,(a)p,
,

+kT+ktk-B*p-

I
- 7 i(mf - ie)

C c+(mf - ie) ,

wherek is a column matrix with entrieswhich are four-vectors.The matrices

A and B are homogeneousfunctions of fi?st degreein CYS,
and A is symmetric.
Ubon translatingthe integration variables
k+k=k

tA+Bp

PUB-76/34-THY

Gauge theories

101

and diagonalizingthe matrix A by an orthonormaltransformationon k, we

can perform the loop integrationsoverki in eq. (3.16). The result is a sum
over terms eachof which has the form

X %+-~[~p~C(or)~p

t C

c+(rnf - if)]},

(3.17)

Ai
is the ith eigenwhere TX/I . . . v is a tensor,typically a product of gP09s,
valueof the matrix A, and Si is a positive numberwhich is determinedby the
tensorialstructure of F,. Note that Ai
is homogeneous
of first degreein
ois. The matrix C is
C= BTA-B

and is also a homogeneous

function of first degreein [YSIn this parametrization, the ultraviolet divergencesof the integralappearas the singularitiesof
the integrandon the right-handside of eq. (3.17) arisingfrom the vanishingof
somefactors IIi[Ai(~)JSi as someor all (YSapproachto zero in certain orders,
for example,
ar, car2 <...<iY

where(rl, r2 , ... , rJ) is a permutation of (1, 2, *.. , I). See,for instance,a more
detailedand careful discussionof Hepp.
The t Hooft-Veltman regularizationconsistsin defining the integralF, in
n dimensions,n > 4 (one-timeand (II - I)-spacedimensions)while keeping
external momentaand polarization vectors in the first four dimensions(i.e.,
in the physical space),performing the ?I - 4 dimensionalintegralsin the space
orthogonalto the physical space,and then continuing the result in ?z.(For single-loopgraphsone may perform all n integrationstogether.)For sufficiently
small n, or complex n, the subsequentfour-dimensionalintegrationsare convergent.
To seehow it works, considerthe integral

PUB-76/34-THY

102

B. W.Lee

X n (kd * el)exp[i cc-+($ - rnf + if)],

(3.18)

where,now, the ki aren-dimensionalvectors.As beforewe can expressthe qi

as linear functions of the kj and the external momentapi, wherethe pi have
only first four-componentnon-vanishing.From now, we shall denotean n-dimensionalvector by (k,K)
A wherek^is the projection of k onto the physical
space-timeand K = k - k. Thus, p = ($, 0). Eq. (3.18) may be written as a sum
of terms of the form

(3.19)

The integralsoverkj can tie performedimmediately, usingthe formulas

d-4KK

h .. . K 42r exp(-jA K2)

Q2

)( (jA)-n/2+2-r

wherethe summationis over the elementsu of the symmetricgroup on 2~objects (aI, ~2, . .. , a>), and
%%

=n-4.

ThusF, of eq. (3.19) will have the form

PUB-76/34-THY

Gaugetheories

103

wheref(n) is a polynomial in n andri is a non-negativeintegerdependingon

the structure of II K, - K, in eq. (3.19). For sufficiently small n < 4, the singularitiesof the integrandas someor all 0;sgo to zero disappear.
The reasonsthis regularizationpreservesthe Ward-Takahashi
identities of
the kind which will be discussedare, firstly, that the vector manipulations
suchas
k'"(2p +q

= [(p +k)S-

m2] -(p2-

m2),

or partial fractioning of a product of two propagators,which arenecessaryto

verify theseidentities by hand, arevalid in any dimensions,and, secondly,
that the shifts of integrationvariables,dangerouswhenintegralsaredivergent,
arejustified for small enough,or complex n, sincethe integralin questionis
convergent.
The divergencein the original integralis manifestedin the polesof Fr(fz)
at n = 4. Thesepolesareremovedby the R-operation,so that Jr(n) as defined
by the R-operationis finite and well-definedas IZ+ 4. Actually, to our knowledgethe proof of this hasnot appearedin the literature,except for the original discussionof t Hooft and Veltman. Heppsproof, for example,doesnot
really apply here,since the analytical discussionof Heppis not tailored for
this kind of regularization.However,the argumentof t Hooft and Veltman is
sufficiently convincingand we haveno reasonto believewhy a suitablemodiftcation of Heppsproof, for example,of the BPHZ theoremshouldnot go
through with the dimensionalregularization.
The abovediscussionis fine for theorieswith bosonsonly. Whenthere are
fermionsin the theory, a complication may arise.This has to do with the occurrenceof the so-calledAdler-Bell-Jackiwanomalies.The subjectof anomalies in Ward-Takahashi
identities has beendiscussedthoroughlyin two excellent lecturesby Adler, and by Jackiw, and we shall not go into any further details here.In short, the Adler-Bell-Jackiwanomaliesmay occur when the verification of certain Ward-Takahashi
identities dependson the algebraof Dirac
gammamatriceswith y5, such as -y,,y5 + y5 7, =.O.Typically, this happens
when a propervertex involving an odd numberof axial vector currentscannot

PUB-76134-THY

104

B.W. Lee

be regularizedin a way that preservesall the Ward-Takahashi

identities on such
a vertex, and as a consequencesomeof the Ward-Takahashi
identities have to
be broken. The occurrenceof theseanomaliesis not a metter of not being
cleverenoughto ilevisea properregularizationscheme:for certain models
sucha schemeis impossibleto devise.The dimensionalregularizationdoesnot
help in such a case,due to the fact that 75 and the completely antisymmetric
tensor density ehPypareunique to four dimensionsand do not allow a logically consistentgeneralizationto H dimensions.Whenthere are anomaliesin a
spontaneouslybroken gaugetheory, the unitarity of the S-matrix is in jeopardy
since,as we shall see,the unitarity of the S-matrix, i.e., cancellationof spurious singularitiesintroduced by a particular choice of gaugeis inferredfrom
the Ward-Takahashi
identities. Grossand Jackiw haveshown that, in an Abelian gaugetheory, thc.occurrenceof anomaliesruns afoul of the dual requirementsofunitarity and renormalizabilityof the theory.
Thus, a satisfactory theory should be free of anomalies.Fortunately, it is
possibleto construct modelswhich are anomaly-free,by a judicious choice of
fermion fields to be included in the model. Thereare two lemmaswhich
make the aboveassertionpossible.One is that the anomaliesarenot renormafized, which in particular meansthat the absenceof anomaliesin lowest
orderinsurestheir absenceto all orders.This was shown by Adler and Bardeen
in the context of an SU(3) versionof the u-model,and by Bardeenin a more
generalcontext which encompasses
non-Abeliangaugetheories.The secondis
the observationthat all anomaliesare related;in particular,if the simplest
anomalyinvolving the vertex of three currentsis absentin a model, so are all
other anomalies.This can be inferred from an explicit construction of all anomaliesby Bardeen,or from a more generaland elegantargumentof Wessand
Zumino.
Let us concludewith a simpleexampleof dimensionalregularization:the
vacuumpdarization in scalarelectrodynamics.The Lagrangianis

and the relevantverticesare shown in fig. 3.2. Thereare two diagramswhich

contribute to the vacuum polarization,shown in fig. 3.3. The sum of these
contributions is
Z=e2j-

dk

1t2k+?4,(2k

+d, -

2((k

+d2 -p2kw,]

pv

We use the exponentialparametrizationof the propagatorsto obtain

(3.20)

PUB-76/34-THY

Gaugetheories

105

Fig. 3.2. Photon-scalar meson vertices in charged scalar electrodynamics.

k
Fig. 3.3. Second order vacuum polarization diagrams in charged scalar electrodynamics.

X [Ox: +dp

+P) - 2((k +pj2 -p2jgpv].

(3.21)

The exponentis proportional to

(CY+ P)k2 + 2k . pa + ap2 - (a + /.I)(/.? - ie)

so we may write

=-e2bgp,
- p2g,,)
0r da0i $3(s)2j Enn
Xexpi

(a+/3)k2+ SP2

-(cY+p)(p2-ie)

PUB-76/34-THY

(3.22)
The first term is explicitly gaugeinvariant and only logarithmicaffydivergent,
so that a subtraction will make it convergent.it is the secondterm that requiresa careful handling.We needthe formulas
-

dk

exp(iAk2) =

Gw
k2 exp(iXk2)

(2&Q

= $ (-in)

7
i expb 4),

(2&X)

s
(3.23)
so that tfie secondterm, 12, is

X exp i -$p2~-(atp)(&ie)
I

A.11 (a+P)

x (i(l -in)-[-.&+(otp)p~)

= -2ie 2g

e iml4
-----ldadfl6(1

-u-~jj~ei~lap-P2fl

pv (2&r)
x [il-l(1

- $2) + i(orpp2 -/.l)]

0 if
.

(3.24)

PUB-76/34-THY

Gauge theories

107

For sufficiently small n, n < 2, the A-integration is convergent, and

J-

dh

,ik(A+iel(I

-2

+i~)

hrtl2-I

p dh -& {h-/2 exp[iX(A t ic)]) = 0.

(3.25)

So the dimensionalregularizationgivesthe gaugeinvariantresult,

12=0.

3.4. Problem
Repeatthe vacuumpolarizationcalculationin spinorelectrodynamicsusing
the dimensionalregularization.

4. The Ward-Takahashi identities

4.1. Notations

One of the problemsin discussinggaugetheoriesis that notationswill get

cumbersomeif we are to put explicitly space-timevariables,Lorenb indices
and grouprepresentationindices.We will thereforeusea highly compactnotation. For simplicity in notation, we will assumethat the gaugegroup in
questionis a simpleLie group. Extensionto a product of simpleLie groups,
suchasSU(2) X U(I), is not too difficult.
We will agreeto denoteall fields by r&. Again for simplicity in notation we
will assume#j to be bosons.Inclusion of fermionsdoesnot presentany diffculty in our discussions,but we will haveto be m indful of their anticommuting nature.Thus, for the gaugefield A;(x) i standsfor the groupindex a, the
Lorentz index 1-1,
and the space-timevariablex. Summationand integration
over repeatedindiceswill be understood.Thus
(piz=sd4x q A;(x)A~~(x) + 0..,

(4-I)

wherethe dotted portion includescontribution from other speciesof fields.

PUB-76/34-THY

108

B. W. Let

The infinitesimal local gaugetransformationmay be written as

l#Jp$;=$it(A;

tt~~j)ea,

(4.2)

whereO. = 8,(x,) is the space-timedependentparameterof the groupG. We

choose@ito be real, SO that
t;

= ~(xa - xp4(xn - Xi)

(4.3)

is real antisymmetric, where7; is the representationof the generatorLa of G

in the basis&. Theinhomogeneous
term q is non-vanishingonly for the
gaugefields
hi =$ars4(xi - xa)ija, b
=0

for & = A:@$

otherwise.

(4.4)

We shall also define

Notice that
(4.5)

where
f le =pw(xo
$g

- <gk

- Xb)64(Xa - xc) ,
=ft;

(4.6)

pk being the structure constant of G. The proof of eq. (4.5) is simple: we

will just show

SinceI$!is non-vanishingonly whenj refers to a gaugefield, let us write

i = (CJA, Xi), i = (d, V,xi>. Then

t; = gpy.pd64(Xi- xa)S4(xa- Xi) ,

PUB-76/34-THY

109

Gauge theories

g(t; ij! - t; A.) =fabcj-d4x E(xa - x)S4(x - xi) -& s4(x - xb)

t S4(Xb- x)S4(x - Xi)-&

S4(x - xa)

=pbcld4x

[(x~ - x)S4(x - xb)]

S4(x - xi) a
i3Xfl

=pbbe&sd4x

S4(x - xj)b4(x -mx,)S4(x - xb) ,

which is equaltog times the right-handside of eq. (4.5).

The gaugeinvarianceof the action can be statedin a compactform,
(4.7)
The linear gaugecondition we discussedin lecture 2 may be written as

Fa[@I= fai@3i

(4-g)

where

fai = a;

for Qi = A$xi) ,

(4.9)
(4.10)

a; = tPbap64(xi - xa) .
Enthis notation, the effective action is written as
f&-jth f>d = s[@l -&F;

[@I+ $Mab[@hb 3

(43)

with

4.2. Becchi-Rouet-Stora transformation

The WT identities for gaugetheorieshavebeenderivedin a numberof different ways. The most convenientway that I know of is to considerthe re-

PUB-76/34-THY

110

B. IV. Lee

sponseof the effective action, eq. (4.1 l), to the so-calledBecchi,Rouet,

Stora(BRS) transformation.It is a global transformationof anticommuting
type which leavesthe effective action invariant. Here 1 shall follow a very elegant discussionof Zinn-Justin givenat the Bonn SummerInstitute in 1974.
The BRS transformationfor non-Abeliangaugetheory is definedas
$=DiQX,

(4.13a)

when?6 X is an anticommutingconstant.Note that if we identify O. = T)~Sh,

we seeimmediately that the action S[$] is invariantunder(4.13a).Thereare
two important propertiesof the BRS transformationwhich we shall describe
in turn.
(i) The transfonnatiorrs on 9t and pa are nilpotent, i.e.,

s2ei=0,

(4.14)

s2qa= 0.

(4.15)

Proof: Eq. (4.14) follows from

(D;Q=O.

(4.16)

Indeed

Since?jOand vb anticommute, the coefficient of vaqn in the first term on the

right-handside may be antisymmetrizedwith respectto a and b. It vanishesas
a consequenceof eq. (4.5).
To show eq. (4.15), we note that
(4. I 7)
by the Jacobi identity. QED.

PUB-76/34-THY

Ill

Gauge theories

(iij The BRS transformations leave the effective action SEn of eq. 14.II)
invariant

Proof: As noted aboveS[$] is invariant under eq. (4.13a).We further note

that
(4.18)

s<Mab[@h&lb)
= 0.
by eq. (4.16) and the definition of Mab, eq. (4.12). Thus

=~FaMabqbsA-~Fa~D;qb6hi0
P

(4.19)

by the definition of Mab, eq. (4.12).

For,later use,we remarkfinally that the metric [d&dta dn,] is invariant
under the BRS transformationof eqs.(4.13). I want you to verify it.
4.3. The Ward-Takahashi identities for the generating functio?zal of Green
functions

We will first derivethe Ward-Takahashi

identity satisfiedby lV,[JI of
eq. (2.3 l),
Z,[~~eiWr;Il=NS[d~d5:dq]exp{id,ff[~,~,q]+iS-~i).

(4.20)

We first note that, accordingto the rule of integration over anticommuting

numbers
(4.21)
we have
(4.22)

s [WdE dq] 4,expCiSeffMl + iJi 3> = 0 ,

becauseSe-rcontainst and q only bilinearly. In eq. (4.22) we makea change

of variablesaccordingto the BRS transformations(4.13). Since a changeof
integrationvariablesdoesnot changethe value of an integral,we have
O=j[d@d[dq]

-~a[$j+iJiE,D,![@]qb

exp(iS,ff+iJi$i)
I

I (4.23)

PUB-76/34-THY

112

B. W. Lee

Eq. (4.23) is the WT identity as first derivedby Slavnovand Taylor. We can

rewrite it in a differential form involvingZF. We define
(ZFlba z Ni s Id&Wtl

Carib exp We, + iJ&$ e

(4.24)

It is the ghost propagatorin the presenceof externalsourcesJi- It satisfies

M

=tinei+
.
[Ls1(zj&

(4.25)

ab iSJ

I wIl1leavethe derivation of eq. (4.25) as arrexercise.Eq. (4.23) can be written as

aFa

c1
+&

=o*
[ 1tzj7)ba

ZF[J] - J,D,! +&

(4.26)

Eq. (4.26) is in the form written down by &n-Justin and Lee. It is the WT
identity for the generatingfunctional of Greenfunctions, and as such it is
rather cumbersomefor the discussionof renormaiizability,since,as we have
seenin lecture 3, the renormalizationprocedureis phrasedin terms of (singleparticle irreducible) propervertices.Nevertheless,eq. (4.26) was used to deduce by hand consequences
of gaugesymmetry or renormalizationparts by
Zinn-Justin and myself. We do not haveto do this, sincewe know better now.
Eq. (4.26) will be useful in discussingthe unitarity of the S-matrix later, however.
4.4. The Wurd-Takuhushi identities: inclusion of ghost sources

To discussrenormalizationof gaugetheories,we haveto considerproper

verticessome of whoseexternallines areghosts.For this reason,the ghost
fields 5 and q should havetheir own sources.We thereforereturn to eq. (2.34),

X exp[iCQ#, t, 4 + U + @a+ J#,.i,l

(4.27)

For the ensuingdiscussion,it is more convenientto consideran object

(4.28)

and define

PUB-76134-THY

Gaugetheories

113

(4.29)
In eq.(4.28), Ki and -Laare sourcesfor the compositeoperatorsDffr$]qa and
$?f,a Q, Q, respectively.It follows from eqs.(4.16) and (4.17) that Z is invariant under the BRS transformations(4.13). Note further that Ki is of anticommuting type* and
(4.30)

(4.3 1)

The invarianceof Z is expressedas

or
(4.32)

We needone more equation,

(4.33)

g=ftig.
i

Let us examinethe consequences

of eqs.(4.32) and (4.33). We perform a
changeof variables
(4.34a)
(4.34b)

&$=-iFa&

(4.34c)

Eq. (4.32) tells us that Z is invariant undersuch a transformation,and the integrationmeasure[d&d$dp] is also, thanks to

PUB-76/34-THY

B.W. Lee

114

A?&(-)
qiKi~

(4.3%)

(4.35b)
Thus, the changeof variables(4.34) in eq. (4.29) leadsto

X exp[i{X + t - /I + /3t - 9 +Jr.$>] = 0.

(4.36)

Next, the equationof motion for n is

(4.37)
Combiningeqs.(4.33) and (4.37), we obtain
(4.38)
Eqs.(4.36) and (4.38) are the basisfor deriving the WT identity for the
generatingfunctional of propervertices.
4.5. The Ward-Takahashi identities for the generating functional of proper
vertices

The generatingfunctional of properverticesis obtainedfrom IV,

W=-itnZ,
by a Legendretransformation.We define
(4.39a)
(4.39b)

PUB-76/34-THY

115

Gauge theories

(4.39c)
wherewe haveusedthe samesymbolsfor the expectationvaluesof fields as
for the integrationvariables.The generatingfunctional for properverticesis

P-C?.

(4.40)

As usual,we havethe relationsdual to eqs.(4.39),

(4.41a)
(4.41b)
(4.41c)
It is easy to verify that if Wand F dependon parametersQ, such as K or L
in our case,which arc not involvedin the Legendretransformation,they satisf-Y
(4.42)

From eqs.(4.36) and (4.38) we can derivetwo equationssatisfiedby F,

(4.43)

(4.44)

It is important to observethe correspondence

betweeneqs.(4.32) and (4.33),
and eqs.(4.43) and (4.44).
If we now define
(4.45)

PUB-76/34-THY

B. W. Lee

116

f&a%,=%.

(4.47)
a

The functional r carriesa net ghostnumber zero, wherewe define the

ghostnumberNp as
N&II = 1 3

NJK]=-1

N,LEl = -1,

NJL]

=-2,

NJ4

=o.

Clearly I may be expandedin terms of the ghost numbercarryingfields:

Substituting the expression(4.48) in eqs.(4.46) and (4.47), differentiating

with respectto Q, and setting all ghostnumbercarryingfields equalto zero,
we obtain
(4.49)

(4=50)

Theseare the equationsfirst derivedby me from (4.26) by a complicated

functional manipulation.Theseare the fully dressedversionsof eqs.(4.7) and
(4.12),

PUB-76/34-THV

Gaugetheories

I17

4.6. Prob Iems

4.6.1. Convinceyourself that the measure[dddldq] is invariant under the

BRS transformation.
4.6.2. Show that

Proveeq. (4.25).
4.6.3. Show that
dci Af(c)=O,
ac

i not summed,

whereci is an anticommutingnumber.Proveeq. (4.37).

4.6.4. Derive the WT identity for Z[J, fl, $1,

5. Renormalization of pure gauge theories

5. I. Renormalization equation

We areready to discussrenormalizationof non-Abeliangaugetheories

basedon the WT identity for properverticesderivedin the last lecture.
Let us recall that our Feynmanintegralsare regularizeddimensionallyso
that for a suitably chosenn not equalto 4, all integralsareconvergent.Thus,
we can perform the BogotiubovR-operationafter the integralhasbeendone,
insteadof making the subtractionof eq. (3.10) in the integrandasZimmermanndictates. In fact this is the procedureusedby Bogoliybov,Parasiukand
Hepp. Further, insteadof making subtractionsat pi = 0, we will choosea
point whereal1momentaflowing into a renormalizationpart areEuclidian.

PUB-76/34-THY

118

B. W.Lee

For a vertex with II externallines, this point may be chosento be --pi= a2,
- I). Th is is to avoidinfrared divergences.
At this point the
squareof a sum of any subsetof momentais alwaysnegative,so that the amplitude is rea1and free of singularities.
For simplicity we first considera pure gaugetheory.Inclusion of matter
fields, such as scalarand spinor with renormalizableinteractionspresentno
difficulty. In particular, couplingof gaugefields with scalarmesonswill be
treatedin chapter6.
Wemay write down the propervertex as a sum of terms,eachbeinga product of a scalarfunction of externalmomentaand a tensorcovariant,which
is a poiynomial in the componentsof externalmomentacarrying available
Lore& indices.All renormalizationparts in this theory haveeitherD = 0 or 1.
The self-massof a gaugebosonis purely transverseas we shall see,so that it
alsohaseffectively D = 0. Thus, only the scalarfunctions associatedwith tensor covatiantsof lowest order aredivergentas n + 4. (Note also that vertices
involving external ghost Iineshavelower superficial degreesof divergencethan
simplepower counting indicates.This is because.C,alwaysappearsas 8 E,&,.)
The basicproposition on renormalizationof a gaugetheory is the following.
If we scalefields and the couplingconstant accordingto
pi - pi = c&z

Q1
i = z/2(e)@
i

K.I = .i?/2(~)Ki

r;a = S2(e)ga

L a = zqE)L'

a = Z(E)c? ,

g=

X(e). g
+z(e)z12(e)

(5.1)

wheree = n - 4 is theregularizationparameter,and chooseZ(e), X(E) and

Z(e) appropriately, then

is a finite (that is, as e = D - 4 4 0) functional of its arguments$, 5. $, K,

L and or. Under the renormalizationtransformationof (5.1). eqs.(4.45) and
(4.46) become

PUB-76/34-THY

119

Gauge theories

We will expandloopwise,

Wehave

Supposethat our basicpropositionis true up to the (n - 1) loop appioximation. That is, up to this order, all divergences are removed by rescahg OF

fields and parametersasin eq. (5.1). We supposethat we havedeterminedthe

renormalizationconstantsup to this order,

(x)n-,

= 1 +X(r) + .** + X(n-1) "

P-7)

We haveto show that the divergencesin the n-loop approximationare also

removedby suitably chosenz,r, 2n andx,r.
Following Zinn-Justin,we introduce the symbol

where the superscriptr denoteshere the quantities rcnormalizedup to the

(rr - 1) loop approximation.We can write eq. (5.8) as

with
(5.10)

PUB-76/34-THY

120

B. W.Lee

The right-handside of eq. (5.9) involvesonly quantitieswith lessthan II loops,

it is finite by the induction hypothesis.Further, divergencesin subdiagramsof
r(,, are removedby renormalizationsup to (n - 1) loops. Thus, the only remainingdivergences
in r(,l) are the overall ones.Let us denoteby Odiv the
divergentpart. If we adjust finite parts of I& appropriately,we have

(5.Ii)
(5.12)

5.2. Solution to renormalization equation

The divergent part off&) is a solution of the functional differential equations (5.1 I), (X1.2). We recall that

(5.13)

(5.14)
(5.15)
wherethk functional operator9 is given by
9= clot61

(5.16)

(5.17a)

(5.17b)

PUB-76/34-THY

Gaugetheories

121

From now on, for the interest of notational economy,we will drop the superscript r, until further notice.
An important aid in solvingeq. (5.14) is the observationthat
p=0.

(5.18)

We will prove this in steps.First, we verify by direct computation that

$g=o,

90

=D;qa+f
i

6.
ahcq b q c&qa

(5.19)

Eq. (5.19) is a direct consequenceof eqs.(5.16), (5.17) that the BRS transformation on 4 and qa is nilpotent. Next we note that

(5.20)

wherewe haveusedthe fact that

(5.21)

(5.22)

Direct computationsyield

SD;6r(())
63050,=3
sqo)-s*r(o)
=
&K@.
3
1

qa63

PUB-76/34-THY

122

B.IK Lee

Qo

qo)

= -

s%

s50)
(
Qf.

s2r(0)
-++----

s r(o)

sKjstl,

sV*

s 2 r(o)
6L*6Qa

Thus

which,proveseq. (5.18).
The fact that 9 is nifpotent meansthat, in general,99 for arbitrary
9 = Y(i$, & 1, K, L) is a solutibn of eq. (5.14)
8(sw).= 0 *

(5.23)

The questionis whether there areother solutionsnot of the form ~7. This
questionalso arisesin renormalizationof gaugeinvariant operators,and has
beenstudied,in particular, by Kluberg-Sternand Zuber.They alsoadvanceda
conjecture: they suggestedthat the generalsolution to eq. (5.14) is of the
form
W(,)P

=GM+99[$,t,tl,KLl,

(5.24)

whereG[$] is a gaugeinvariant functional,

Recently, Jogtekarand I were able to prove this, mostly by the effort of the
first author. The proof is tedious,and I believethat it can be improvedas to
rigor, eleganceand length. For this reason,I will not presentthe proof. It is
easyto seethat the form (5.24) satisfieseq. (5.14) and that G[$] of eq. (5.25)
is not expressibleas gS, in this caseat least. It is the completenessof eq.
(5.24) which requiresproof.
Eq. (5.15) [or (5.4)] is immediately solved.It meansthat
r&J fh Es~5K, Ll= r&l [A O,V,Kj f a;$, L] + jQ[ti* -5V, L.1,(5.26)
where Q is transverse:3%Qj = 0 -

PUB-76/34-THY

Gaugetheories

123

The quantity (l&} div is a local functional of its arguments.If we asignto

K and L the dimensionsD(K) = 2 and D(L) = 2, then Z. hasthe uniform dinmsion 0, and SOdoes Irfn)ldiv. It hasNg [rfil,] = O. SinceNg [ 91 = t.1, it
follows that iVg[ 9]= -1 in eq. (5.24). In order that the right-handside of
eq. (5.24) is local, both 9 and Qmust be separatelylocal. The most general
form of {r{,r)}div satisfying the aboverequirementsis

wherecr,p, y arein generaldivergent,i.e., e-dependent,constants.

Using the
explicit form of 9, eqs.(5.17) we can write

(5.28)

In eq. (5.27) G[$] is equalto S[I$). This is so becausethe action is the only
local functional of dimensionfour which satisfieseq. (5.25).
Because

(5.29)

Thus, combiningeqs.(5.28), (5.29), we obtain

(5.30)

Recall that in cq. (5.30) d, f, 77,K, L andg are renormalizedquantitiesup to

the (n - 1) loop approximagon.We shall denotethem by (#),-t, etc.
If we now define (Z),, (Z),, (X), by

PUB-76/34-THY

124

93.IV. Lee

<a,=(+Q-l+qn)5

etc. ,

(5.3 I)

and renormalizethe fields and coupling constantaccordingto

etc. ,

(5.32)

(5.33)

and choose+), :(,+ xtn) to be

%) = -4 a * lw)
?(a = 37 + IN4
%)

,
9

- q,z) - :2(n) = 2+j,

(5.34)

then {~fnj)drv is eliminated: I&, is a finite functional in terms of (@j,,, . .. I

and ($),,. Furthermore,since(@), = (Zji2$, .. . , eqs.(5.3), (5.4) are ako
true for the newly renormnlizedquantities. This completesthe induction.
Note further that
(5.35)

is finite. The renormalizedfield $* transformsunder the gaugetransformation

as

6. Renormalizationof theories with spontaneouslybroken symmetry

6.1. hclusion of scalar fields

In chapter5, we detailedthe renorrhalizationof pure gaugetheories.Let us

considernow a theory of gaugebosonsand scalarmesons.Let
!4j = (.qxj,

s&)) 9

PUB-76/34-THY

Gauge theories

wheres, arescalarmesons,and let

be the vector which defies the gauge.The action for the scalarfields is of the
form

where(Ds), is the covariantderivativeacting on S, V(s) is a G-invariantquartic polynomial in s and of dimensionat most zero.
We shall write
(6.2)

[W, ta] = [6M,t] = 0,

(6.3)

whereM& is the renormalize$massmatrix for the scalarmesons.We shahas.sumefor the moment that Mr is a positive semi-definitematrix.
Let us discussrenormalization.Almost everythingwe discussedin the last
section holds true. In particular we have

WCn,ldiv = G [$I + 9 W> E,rl, K t3 ,

wherewe have written @i= {A,, sa) and Ki = {K,, &},A,

fields, A, = A;(x), t = (a, p, x). Now we have

beingthe gauge

(6.4)

where Yis a G-invariantquartic polynomial in s. This term is eliminated by

renormalizationsof coupling constantsappearingin V(s). 9 takes the form

PUB-76/34-THY

B. IV. Lee

126

whered$ is a G-covariantcoefficient. (It could be i$, or somethingelse,such

as the d-type couplingin SU(3), for example.)This gives

{r&p=1& +pl)
[A& +ia$]-a a
t

2 Gag

where

Thesedivergencesareeliminatedif we renormalizeA, &q, K,, La andg as before and

%!

= 3/z 9
s a

- &. l12p
a=
Q- ( z s >

which leaves

invariant, and shift the si fields by

s; = Sk - u,w P
and choose

p = -$- 12(c3r ,
u ( 1
s

PUB-76/34-THY

Gaugetheories

An important lesson to be learned here is that in a generallineargauge,scalar

fields can developgauge-dependent
vacuumexpectationvalues,which areinnocuousfrom the renormalizationpoint of view.
6.2. Spontaneously broken gauge symmetry

Let us considerthe casewhereM is not positive semi-definite.It is by now

well known that under such circumstancesspontaneousbreakdownof the
gaugesymmetry takesplace, and someof the scaIarfields and someof the
(transverse).gauge
bosonscombine to form massivevector bosons.We will give
herea very brief discussionof the Higgsphenomenon.
We define V0 by

If M is not positive semi-definite,So= 0 is no longera minimum of the potential V,. Let sa = 11,be the absoluteminimum of Vo,
(6.7)

2Vo
q$sp

s=u

= Nl$ ,

CR,, positive definite.

G-invarianceof the potential Vo is expressedas

6 Vo
s ta -=o.
Q d 6sP
Differentiating this with respectto s7 and setting s = U, and making use of
eqs.(6.7), (63, we obtain
(6.9)
Therefore,thereare as many eigenvectorscorrespondingto the eigenvalue
zero as there arelinearly independentvectorsof the form C&U,. If the dimension of G is N and the little groupg which leavesu invariant hasdimensionm,
thereareiV - ~11
eigenvaluesof 7K2 which vanish.
For fltture use,it is useful to define a vector 1: by

PUB-76/34-THY

128

B. W. Lee

where ~4;is to be defined. Since all representations,except the identity representation, of a Lie group are faithful, there areN - m independentvectors of
this form. Now we define
(6.10)
This matrix is of a bloc; diagonalform; moreoverif a or b refer to a generator
of the little group g, (+)ab vanishes.Now form
(6.11)
This is a projection operator,P2 = P, onto the vector spacespannedby vectors of the form t~pt$. This spaceis N - m dimensional,
trP=N-

tn.

Eq. (6.9) may be written as

We renormalizc the gaugefields, ghost fields and gaugecoupling constant

as before, and renormslize s, accordingto
sa =Z/J(s
a (Y+ua +su a) 5
and determine6 U, = (6 t.ta)l + (6 I& t .._by the condition that the divergencesof the form -(6 S[$]/S s,)A(e)u, in the n-loop approxkation be cancelled by the displacementof the renormalizedfields s:, (6 u&. (See the discussionin subsect.6.1.) The renormalizedvacuum expectation value z& is to
be determined by the condition

s2rr
I

6 f#J;s4; #=u

positive semi-definite.

PUB-76/34-THY

128

Gaugetheories

We equatethe gaugefixing term C,!Jto (numerically)

Then the termsin the action quadraticin renormalizedfields and coupling

constants(excludingrenormaiizationcounter terms) are

wherewe havesuppressed
the superscriptr altogether.The propagatorsfor
the gaugebosom, scalarsand ghostfields are, respectively,
[Aj$(k,c~)]~~=~

k2-p2+ie

,
cb

which can be written, in a representationin which p2 is block diagonal,as

01

[&12a)
the latter holding for a, b beingone of the m indicescorrespondingto generators of the little groupg;

[AF(k2s
a)], =(1-Jlq ( k2- A2

+ ,),

tz~(~)ab(kl_:2+iQ)hlpe9

(6.12b)

PUB-76/34-THY

130

B. W. Lee

tAdk2p

a&

= ( k2 _ b,z + , ).,

(6.12~)

If the theory is to be sensible,and gaugeinvariant, then the poleswhose

locations dependon the gaugeparameterLYcannot be physical, and the particles correspondingto such polesmust decouplefrom the Smatrix. If this is
the case,as we shall show, then there areN - m massivevector bosons,m
masslessgaugebosons,and (N - m) lessscalarbosonsthan we started out
with. This is the Higgsphenomenon.
is a theory of this kind renormalizable?The answeris yes, becausethe
Feynmanrules of the theory, including the propagatorsaboveare those of a
renormalizabletheory, and the WT identity, eq. (6.3), and the ensuingdiscussion in chapter 5 and subsect.6.1 hold true whetheror not#is positive definite. That is, by the methodsdiscussed,we can construct a finite I in this
casealso. The expansioncoefficients of T about r$:= uf, where

62r
646 (b;IQ=uZ

positive definite,

then are the reducibleverticesof the renormalizedtheory. I shall not describe

the details of the renormalizationprogramsince they havebeendescribedin
many papers,most recently in my paper(ref. [SO]), but the principle involved
should be clear.
But an additional remarkis in order: the divergentparts of variouswavefunction and coupling-constantrenormalizationconstantsare independentof
MF. This has to do with the fact that theseconstantsareat most logarithmically divergent,and insertion of the scalarmassoperators(whosedimensionis
2) rendersthem finite. For detailedarguments,seeref. [50].
6.3. Gauge indepeizdemzeof the S-matrix

What remainsto be done is to demonstratethat the unphysicalpolesin the

propagatorsineq. (6.12). which dependon the parameterL\!and some of which
correspondto negativemetric particles,do not causeunwantedsingularitiesin
the renormalizedSmatrix. We shall do this by proving that the renormalized
S-matrix is independentof the gaugefixing parameterCX.To ensurethat the S-

PM-76/34-THY

131

Gauge theories

matrix is well-defined,I shall assumethat after the spontaneousbreakdown

of gaugesymmetry there is at most one masslessgaugebosonin the theory.
Beforeproceedingto the proof, the following ihustration is useful. For
simplicity let us considera X@4theory. The generatingfunctional of Green
functions is

where

What happensif we insteadcouplethe external sourceto $J+ @? We can write

the generatingfunctional as

Z[iJ =~~[WJexpW[~l+iit@+ d~~)l-

(6.15)

We can expressZ in termsof Z,,

(6.16)
where
F(4) = c#J
-)-(b3.
Let us considera four-point function generatedby Z[i],
GJ1,2,3,4] =(-Q4

S4-mJ
Sit lPSWA3)6jW

What eq. (6.16) tells us may be pictured as foliows:

wherewe haveshown but a classof diagramsthat emergein the expansionof

the right-handside of eq. (6.16). The part of the diagramenclosedby a dotted

PUB-76134-THr

132

B. W. Lee

squareis a Greenfunction generatedby ZO[q. Let us now considerthe two

point functions Aj and AJ generatedby Z[j] andZu [J],

. ..a

So, if we examinethe propagatorsnearp2.= pf, we find

lim Ai = - zi

ZJ

lim AJ=-------

P-P,2

P2+

P2-$

(6.17)

P2 -$

wherethe ratio
u = (zj/zJ)12

(6.18)

is given diagrammatically by
0=1*-i-

e+...

The renormalizedS-matrix is defined by

S(k,, . . . >=n
lim
i=l k&,2

k;-$ _

(6.19)

----gyG(kl....),

whereG is the momentum spaceGreenfunction. Let us considerthe unrenormalizedS-matrix defined from (?j,
Si(kl, ..e ) = n lim(kf - pz)Gj(kl,

... ) .

(6.20)

Clearly only thesediagramsOf ej in which there arepolesin all momentum

variablesat $ will survivethe amputation process.(In fig. 6.1, thereare poles
in k: and k: at cl,.) Thus
S;(kl,

. .. ) = crN2SJ(k, 9 a.. ) )

(6.2 1)

PUB-76/34-THY
Gaugetheories

133

whereN is the number of the externalparticle;. It follows from eqs.(6.1 Q(6.21) that
SpSfES,

(6.22)

and we reachan impor tant conclusion:if two Zs differ only in the external
sourceterm, both of them yield the samerenormalizedS-matrix.
We now come back to the original problem,and ask what happensto
ZF [.Il if an infinitesimal changeis madein F,

We are dealingwith unrenormalizedbut dimensionallyregularizedquantities

in eq. (6.23). To first orderin AF, we have

Now, making use of the WT identity (5.23),

[&%dVl IF, - iJi~~D~[~]~~)exp{i~~~ +iJ&}

we can write eq. (6.24) as

Since
:AF,

[ 1
i&

iJi=-.z,

= 0,

PUB-76134-THY

134

B.M Lee

we obtain

z F+AF

- F

=iJiN s [dtdrld@Jexp{iSeK+ iJ&.i,

X G-iAFabPJ~a~~I~J~&~.

(6.25)

But, eq. (6.25) meansthat to lowest order in AF,

F+AF

=fVj [d$dEdqJexp{is,, + iJiai} ,

(6.26)

where

Thus,an ir@ itesimal changein the gaugecondition correspondsto changing

the sourceterm by an infinitesimalamount. But we havealreadyshown that
the renormahzedS-matrix is invariant under such a change!Thus

W lF+AF= W IF :

(6.27)

A few final remarks:(i) Inthe previouslectureswhen we discussedrenormalization,we definedthe renormatizationconstantsin respectto their divergent parts. The wave-functionrenormalizationconstantsusedin this lecture
aredefined by the on-shellcondition (6.17). Thesetwo arerelatedto each
other by a finite mdtiplicative factor. To seethis, observethat we can make
the propagatorsfinite by the renormalizationcounter terms definedin the
pteviouslectures.The propagatorsso renormalizeddo not in generalsatisfy
the on-shellcondition
hm Ak(p) = L!-P-P,2

P2-$

but a finite, final renormalizationsuffices to make them do so. (ii) We can define the couplingconstantsto be the value of a relevantvertex when all physical external lines are on massshell. Then

PUB-76/34-THY

Gauge theories

13s

References
Lecture I
For the geneialdiscussionof path integral formalismappliedto gaugetheories,seeProf. Faddeevlectures,
s
and
[l J E. Abers and B-W. Lee, Phys. Reports 9C (1973) 1.
[2] S. Coleman, Secret symmetry, Lectures at 1973 Int. School of Physics &tore
Majorana, to be published.
[SJ N.P. Konopleva and V.N. Popov, Kalibrovochnye polya (Atomizdat, Moscow,
1972).

Originalliterature on the quantizationof gaugefields includes

(4 J R.P. Feynman, Acta Phys. Polon. 26 (1963) 697.
[S] B. De Witt, Phys. Rev. 162 (1967) 1195,1239.
[6J L-D. Faddeev and V.N. Popov, Phys. Letters 25B (1967) 29.
[7J V.N. Popov and L.D. Faddeev, Perturbation theory for gauge invariant fields. Kiev
ITP report, unpublished.
[S] S. Mandelstam, Phys. Rev. 175 (1968) 1580.
(91 M. Veltman, Nucl. Phys. B21 (1970) 288.
IlOJ G. t Hooft. Nucl. Phys. B33 (1971) 173.

The axial gaugewas first studiedby

111J R. Arnowitt and S.I. bkkler, Phys. Rev. 127 (1962) 1821.

In conjunction with L. Faddeevlectures

s
includedin this volume,see
[12] L.D. Faddeev, Theor, Math. Phys. 1 (1969) 3 [English trabslation I (1969) 11.

Lecture 2
Gauge theories cnn be quantized in other gauges than the ones discussed in
this chapter.In particular the following papersdiscussquantizationand/or renormalizationof gaugetheoriesin gaugesquadraticin fields:
[13] G. t Hooft and M. Veltman, Nucl. Phys. B50 (1972) 318.
[14J S. Joglekar, Phys. Rev. DlO (1974) 4095.

Differentiation and integrationwith respectto anticommutingc-numbers

arestudied and axiomatizedin
[15] F.A. Berezin, The method of second quantization (Academic Press,New York,
.1966) p. 49;
R. Amowitt, P. Nath and B. Zumino, Phys. Letters 56 (1975) 81.

Lecture 3
For renormalizationtheory see:
[ 161 F.J. Dyson, Phys. Rev. 75 (1949) 486,1736.
1171 A. Salam, Phys. Rev. 82 (1951) 217; 84 (1951) 426.
(181 S. Weinberg, Phys. Rev. 118 (1960) 838.
[ 191 N.N. Bogoliubov and D-V. Shirkov, Introduction to the theory of quantized fields
(Interscience, NY, 1959) ch. N, and references therein.
[ZOJ K. Hepp, Comm. Math. Phys. 1 (1965) 95; Th&orie de fa renormatisation (Springer,
Berlin, 1969).

PUB-76/34-THY

136

B. W. Lee

(211 W. Zimmermann, Lectures on elementary particles and quantum field theory, ed.
S. Deser, M. Grisaru and H. Pendleton (MJT Press,Cambridge, 1970) p. 395.

The dimensionalreguhuization,in the form discussedhere,is due to:

1221 G. t Hooft and M. Veltman, Nucl. Phys. E44 (1972) 189;
1231 C.G. Bollini, J.-J. Giambiagi and A. Gonzales Dominguez, Nuovo Cimento 31
(1964) 550;
[24] G. Cicuta and E. Montaldi, Nuovo Cimento Letters 4 (1972) 329.

A closely reiatedregularizationmethod (analytic regularization)is discussed

in:
[25] E.R. Speer, Generalized Feynman amplitudes (Princeton Univ. Press, Princeton,
1969).

Excellent reviewson the Adler-BelEJackiwanomaliesare:

[26] S.L. Adler, Lectures on elementary particles and quantum field theory, ed. S.
Deser, hi. Grisaru and 11.Pendleton (MIT Press,Cambridge, 1970).
[27] R. Jackiw, Lectures on current algebra and its applications (Princeton Univ. Press,
Princeton, 1970).

For a completelist of anomalyvertices,involving only currents(not pions)

see:
[28] W.A. bnrdeen, Phys. Rev. 184 (1969) 1848.
[.29] J. Wess and B. Zumino, Phys Letters 37B (1971) 95.

The following papersdiscussthe problemof anomaliesin gaugetheories:

(301 D.J. Gross and R. Jackiw, Phys. Rev. D6 (1972) 477.
[31] H. Ceorgi and S.L. Glashow,.Phys. Rev. D6 (1972) 429.
1321 C. Bouchiat, J. Iliopoulos and P. Meyer, Phys. Letters 38B (1972) 519.
(331 W.A. Bxdeen, Proc. 16th Conf. on high-energy physics, 1972, vol. 2, p. 295.

The last referencegivesa concisealgorithm for dimensionalregularizationvalid

for scalarloops. Our prescriptionagreeswith it for this case.
For an excellent discussionof dimensionalregularizationand a diagrammatic
discussionof the Ward-Takahashi
identities, see:
i34] G. t Hooft and M.J.G. Veltman, Diagrammar, CERN yellow preprint 1973.
Lecture 4

TheWard-Takahashi
identities werefirst discussedin the context of quantum electrodynamicsin
135) J.C. Ward, Phys. Rev. 78 (1950) 1824.
[36] Y. Takahashi, Nuovo Cimento 6 (1957) 370.

The WT identities for the non-Abeliangaugetheorieswere first discussedin:

(371 A.A. Slavnov, Theor. Math. Phys. 10 (1972) 152 [English translation 10 (1972) 99.
(381 J.C. Taylor. Nucl. Phys. B10 (1971) 99.

and usedextensively to study renormalizationin

[39] B.W. Lee and J. Zinn-Justin, Phys. Rev. D5 (1972) 3121,3137; D7 (1973) 1049.
[40) G. t Hooft and M. Veltman, Nuci. Phys. B50 (1972) 318.

The WT identity (5.49) (4.50) was obtainedin:

1411 B.W. Lee, Phys. Letters 46B (1974) 214; Phys. Rev. 9 (1974) 933.