NON PERTURBATIVE QCD

A. DI GIACOMO
Pisa University and I.N.F.N., Via Buonarroti 2, 56100 Pisa, Italy
E-mail: digiaco@[Link]

This series of lectures consists of two parts. In the first part the foundations of
arXiv:hep-lat/9912016v1 9 Dec 1999

perturbative and non perturbative formulation are discussed. The ambiguity in

the definition of vacuum condensates is then analyzed. In the second part the
symmetry patterns of the deconfining phase transition are discussed.

1 Introduction

I shall preliminary review the basic formulation of quantum field theory, to

clarify what is meant by perturbative and non perturbative. This will also
help to understand the foundations and the limitations of lattice QCD, which
is the main existing non perturbative approach.

1.1 Basic Feynman path.

A quantum system is described by a set of canonical variables q(t), p(t) and
by the Hamiltonian H. q(t), p(t) is a short symbol for q(i, t), p(i, t) the index
i running over the degrees of freedom. When the index i is not specified what
we say will apply to any system, independent of the number of degrees of
freedom. In a quantum field theory q(i, t) are the fields
q(i, t) = ϕa (~x, t) (1)
the index being i ≡ (a, ~x).
The system has a continuous infinity of degrees of freedom.
Solving a field theory means constructing a Hilbert space on which q, p act
as operators, obeying the canonical commutation relations and the equations
of motion. A ground state must exist to ensure stability.
The usual approach consists in splitting the Hamiltonian in a free Hamil-
tonian H0 and an interaction HI
H = H0 + HI (2)
H0 is quantized and solved exactly, and the resulting Hilbert space is assumed
to describe also the interacting system. HI is an operator in that space and is
treated as a perturbation. This approach is called perturbative quantization.
A more fundamental formulation can be given in terms of Feynman path
integral1 .

lisbws: submitted to World Scientific on February 1, 2008 1

 The basic quantity is the action
Z
S[q] = L dt (3)

with L the Lagrangean. S is a functional of the fields. The equations of

motion are the stationary points of the functional
δS
=0 (4)
δqi (t)
The knowledge of the partition function
 
Z Y
Z =  dq(i, t) exp[iS(q)] (5)
i,t

or better of the generating functional

 
Z Y XZ
Z[J] =  dq(i, t) exp{i[S(q) +
 dtJi (t)qi (t)]} (6)
i,t i

provides the solution of the theory.

Indeed the connected time ordered correlators of the fields in the ground
state, are expressed in terms of Z[J] as
δ n Z[J]

1 n

h0|T (q(i1 , t1 ) . . . q(in , tn )) |0i = (−i)  (7)
Z δJ(i1 , t1 ) . . . δJ(in , tn ) J=0
In terms of the correlators the Hilbert space is determined, and the solution
of the theory is known (reconstruction theorem).
Z is a functional integral, with a continuous infinity of integration vari-
ables. By definition it is defined as the limit of a sequences of ordinary inte-
grals, computed on a discretized space time, on sets of points which become
dense in space time in the limit.
The above statements are clearly illustrated by the following argument1 .
Consider the transition amplitudes
hq ′ , t′ |q, ti (8)
between eigenstates of the q’s at different times. Any physical amplitude will
be known if all the amplitudes (8) are known, the q’s being a complete set of
commuting operators.
Eq.(8) can be rewritten
′
hq ′ |e−iH(t −t) |qi (9)

lisbws: submitted to World Scientific on February 1, 2008 2

 t′ −t
Putting δ = n+1 ,
′ n+1
e−iH(t −t) = e−iHδ


as n → ∞, δ → 0.
By inserting complete sets of states, we get
X
hq ′ , t′ |q, ti = hq ′ |e−iHδ |qn ihqn |e−iHδ . . . e−iHδ |q1 ihq1 |e−iHδ |qi
qn ...q1
2
We shall consider for the sake of simplicity H’s of the form H = p2 + V (q),
but what follows can be easily adapted to momentum dependent interactions.
Then, by use of the Baker-Haussdorf formula
p2
e−iHδ ≃ e−i 2 δ e−iV (q) 1 + O(δ 2 )


(10)
h i
(qk+1 −qk )2
−i V (qk )− 2δ2
δ
−iHδ
hqk+1 |e |qk i ≃ e (11)
and
n n
Z Y " # Z Y
X
′ ′
hq t |qti = lim dqk exp i L(qk ) = lim dqk eiSn (q) (12)
n→∞ n→∞
k=1 k k=1

Eq.’s (11,12) allow to continue analitically to imaginary times (Euclidean

space time) by the Wick rotation x0 = it. The limit of the Euclidean ampli-
tude when x′0 → T , x0 → −T , and T → ∞ can be evaluated as follows
T T
X
hq ′ T |q − T iE = hq ′ |e− 2 H |En1 ihEn1 |e−[S(q)+J·q] |En2 ihEn2 |e− 2 H |qi
En1 ,En2
− T2 En1 T
X
= e ψEn1 (q ′ )e− 2 En2 ψEn2 (q)hEn1 |e−[S(q)+J·q] |En2 i (13)
En1 ,En2

The external current has been taken to be zero outside the time interval
(− T2 , T2 ), and the ψ’s are the wave functions of the states |En i which are
eigenstates of energy. As T → ∞ only the ground state survives and, apart
from irrelevant multiplicative constants which cancel in eq.(7), eq.(13) defines
the euclidean version of Z[J], and selects the true ground state of the theory.
The Lagrangean defines a field theory if the limit δ → 0 (n → ∞) exists
(ultraviolet limit), and if the limit T → ∞ exists (infrared limit).
Note that after Wick rotation Z is in fact the partition function of a
statistical system. The infrared limit exists if the theory ha a mass gap, or,
in the language of statistical mechanics if the thermodynamical limit exists.

lisbws: submitted to World Scientific on February 1, 2008 3

1.2 Lattice QCD
Lattice formulation of QCD2 is a clever construction of approximants in the
sequence of discrete integrals which define the partition function.
The theory is defined on a square lattice in space-time, and the building
block is the link
Uµ (n) = eiag0 Aµ (n) (µ = 1 . . . 4) (14)
which is the parallel transport from the site n of the lattice to the neighbor-
ing site in the direction µ. The action is expressed in terms of the parallel
transport along an elementary square, in the plane µν, the so called plaquette
Πµν as
   
X 1 2Nc
S= βRe T r 1 − Πµν β= 2 (15)
n,µν
Nc g

In the limit of zero lattice spacing it tends to the continuum action

1X a a
S ≃ − Gµν Gµν (16)
a→0 4
A critical point exists at β → ∞ (g 2 → 0) where the β function of the
theory tends to zero by negative values (asymptotic freedom)3 , and like in
any statistical model QCD is defined as a field theory when approaching that
point.
At sufficiently large β, by renormalization group arguments the lattice
spacing a, measured in physical units, becomes smaller and smaller
a(β) ∼ exp(−b0 β) (17)
−b0 being the first coefficient in the perturbative expansion of the β function.
At sufficiently large β the correlation length becomes large with respect
to lattice spacing, and the coarse structure of the lattice irrelevant: the limit
of eq.(12) exists.
On the other hand if the lattice is kept large compared to the correla-
tion length also the thermodynamical limit is reached, since the theory has a
fundamental scale built in.
This argument indicates that QCD exists as a field theory, even if a formal
proof of that statement does not yet exist in the sense of constructive field
theory.
At sufficiently large β, and large enough lattice a good approximant is
obtained for Z, in the sense of the limit (12), and in particular the true
ground state is selected, as shown in eq.(13).

lisbws: submitted to World Scientific on February 1, 2008 4

 The formulation is gauge invariant and needs no gauge fixing and no
Faddeev-Popov ghosts.
It would be nice if analytic solutions could be given of the lattice regu-
larized theory. Unfortunately nobody has been able up to now to produce
them. Numerical simulations are the usual technique to compute physics
from lattice. Euclidean QCD is a statistical mechanics, and like in statisti-
cal mechanics Montecarlo techniques can be used to compute the regularized
Feynman integrals and correlation functions.

1.3 Perturbative expansion.

The action S can be split in two parts: a term S0 which is bilinear in the
fields and a term SI which includes the rest
S = S0 + SI (18)
S0 can be written as
1X −1
S0 = qα Dαβ qβ (19)
2
αβ

where α = i, t spans the degrees of freedom and time. Then exp[−S] is

expanded in powers of SI as
∞
X (−SI )n
exp[−S] = exp[−S0 ] (20)
n=0
n!
and
∞
(−SI )n
Z Y Z Y X
Z= dqα exp[−S] = dqα e−S0 (21)
α α n=0
n!
SI is in general a polynomial of the fields and by inserting eq.(20)
∞
1 1 (−SI )n
Z Y X
−1
hT (qα1 . . . qαn i = dqρ exp(− qα Dαβ qβ )qα1 . . . qαn (22)
Z ρ
2 n=0
n!

All terms of the expansion are gaussian integrals of the form

1 1
Z Y
−1
dqρ exp(− qα Dαβ qβ )qσ1 . . . qσn (23)
Z ρ
2

which can be computed and give

X
Dαi1 αi2 Dαi3 αi4 Dαin−1 αin (24)
pairs

lisbws: submitted to World Scientific on February 1, 2008 5

where the sum runs over all possible pairings of indices. Eq.(24) is nothing
but Wick’s theorem with D the propagator.
In QCD such scheme of quantization produces an S matrix describing
scattering of quarks and gluons, and the ground state is Fock vacuum. The
theory can be renormalized producing for any observable quantity a power
series of the renormalized coupling constant αs , with finite coefficients
X
O= Or αrs (25)

However the series is not convergent, not even as asymptotic series. This
reflects an instability, probably due to the inadequacy of Fock vacuum as
ground state. It is phsically obvious that the theory describing scattering
of particles which are never observed cannot be a good starting point for a
perturbative expansion4,5 .
It is an empirical and not fully understood fact, however, that a few terms
of the expansion (25) give a good description of phenomena involving short
distances.

1.4 Instanton vacuum.

An attempt to go beyond perturbation theory was developed6 after the dis-
covery of classical solutions of QCD with finite action (instantons), having
non trivial topology7 .
In the computation of an ordinary integral the saddle point approximation
implies a sum over all the stationary points of the phase:
Z XZ f ′′ 2 X (∆fn )
k
dx ef (x) ≃ d(δxn )ef (xn )+ 2 δxn (26)
x
k!
n k
f ′′ 2
∆fn = f − f (xn ) − δx
2 n
Also in the computation of the Feynman integral by the saddle point approx-
imation all the stationary points of the action should be considered.
The usual perturbative expansion consists in retaining only the trivial
stationary point q = 0, and the saddle point expansion is nothing but eq.(22).
If other minima exist, with finite action, then, the gaussian approximation
reads

(∆fn )k
XZ − 21 δqαδS
S=S δqα δqβ X
0 δq 0
−Sn [q] β
Z= dδqn e e

(27)
n
k!
k

lisbws: submitted to World Scientific on February 1, 2008 6

S 0 [qn ] is the action of
 the classical configuration qn , δqn = q − qn , ∆S =
1 δ2 S 
0
S − S [qn ] − 2 δqα δqβ  δqαn δqβn .
qn
The existence of minima, i.e. of stable classical solutions, is due to topol-
ogy. Finite action requires that, in the euclidean space time7 ,
 
1
Fµν Fµν ≃ O
r→∞ r4
i.e. that the gauge field Aµ on the 3d sphere at infinity S3 is a pure gauge.
This defines a mapping of S3 into the gauge group. To each continuous con-
figuration an integer is associated which indicates how many times the group
manifold is swept Rin this mapping. If dU is the volume element of the group
normalized to 1, ( dU = 1), and dσ µ the volume element of S3
Z
dU = K µ dσ µ K µ dσ µ = n (28)

By Gauss theorem, if ∂ µ Kµ = Q(x)

Z
Q(x)d4 x = n (29)

Q(x) is known as topological charge density.

Explicit form of the instanton was produced in ref.7 .
At the end of the 70’s a big effort was done to include non perturbative
effects via eq.(27), i.e. by taking into account extra terms beyond the trivial
stationary point q = 0. The hope was that small instantons would dominate
and hence a few terms of the perturbative expansion around them would be
a good description of hadronic physics6 .
The attempt failed due to difficulties in controlling the infrared problems,
and to the difficulties in explaining confinement. It is known, however, that
instantons play an important role in understanding chiral properties of the
theory8 .

1.5 Non perturbative approaches.

We have no real knowledge of non perturbative methods in field theory. Only
numerical simulations on lattice allow to extract information on the mecha-
nisms of the theory, e.g. on confinement.
A truly non perturbative idea which has proved to be fruitful is the Nc →
∞ limit9 .
The idea is that the main properties of gauge theories are independent
of Nc , and the same as for Nc → ∞ (1/Nc → 0), with g 2 Nc = λ fixed.

lisbws: submitted to World Scientific on February 1, 2008 7

Corrections for finite Nc , are perturbations which do not alter the structure
of the theory.
n/2
Fermion loops are of order Nf /Nc , (Nf is the number of light quarks,
n the number of vertices). Except for n = 2, the graphs which enters in the
definition of the β function, all the rest should be a small correction. Lattice
simulation do indeed indicate that, except for a rescaling of the masses due
to the β function, numerical determination in quenched approximation differ
in general by few percent (∼ 10%) from the determinations in full QCD.
The 1/Nc limit gives an explanation of the so called U (1) problem, relating
it to a quantity, the topological susceptibility of the vacuum χ10 . Lattice
determination of χ confirm the validity of the approach11. The same idea
proves a useful guiding principle in understanding confinement (see sect.3).

2 Non perturbative effects in QCD. Vacuum condensates.

2.1 S.V.Z. Sum rules.

Consider the Wilson operator product expansion of the correlator of two cur-
rents
β(g) a a
Πµν (x) = T (j µ (x)j ν (0)) = CIµν (x) I + C4µν Gαβ Gαβ
g
+Cψµν (x)mψ̄ψ + . . . (30)

In principle this expansion is a theorem in perturbation theory12 , but is pre-

sumably valid in QCD at short distances13 . In Fourier transform and renor-
malizing
Z
Πµν (q) ≡ d4 x eiqx h0|T (jµ (x)jν (0)|0i =

= (qµ qν − q 2 gµν ) Π(q 2 ) − Π(0)

 
(31)
C4 Cψ
Π(q 2 ) − Π(0) = C1 + 2 G2 + 4 Gψ + . . . (32)
q q

G2 = h0| β(g) a a 4
g Gµν Gµν |0i has dimension [m ], is the vacuum expectation value
of the dilatation anomaly and is known as gluon condensate.
Gψ = h0|mψ̄ψ|0i also has dimension [m4 ], is related to the breaking of
the chiral symmetry and is known as the quark condensate.
C1 , C4 , Cψ are dimensionless constants, apart from log’s, and depend on
αs . G2 , Gψ are not defined in perturbation theory. A dispersion representation

lisbws: submitted to World Scientific on February 1, 2008 8

for Π̃(q 2 ) = Π(q 2 ) − Π(0) is
∞
ds R(s)
Z
2 2
Π̃(q ) = −q (33)
4m2π s s − q 2 + iε
with R(s) the well known ratio
σe+ e− →h (s)
R(s) = (34)
σe+ e− →µ+ µ− (s)
In perturbation theory, at each finite order only C1 gets contributions and is
a constant modulo logs. σ ∼ 1/s, modulo logs.
Physics, however, is not scale invariant: resonances break scale invariance
by powers, and condensates should describe that breaking. This is the basic
idea of SVZ sum rules where the equality “on average” is exploited
Z ∞
2 ds R(s) CG Cψ
−q 2
≃ C1 + 4 2 G2 + 4 Gψ (35)
4m2π s s − q + iε q q
Equality “on the average” means that the two sides are not equal at each
value of q 2 , but suitable averages are equal, which emphasize the region where
the two description are expected to be approximately valid13 . The C coef-
ficients are computed to some order in perturbation theory. A successfull
phenomenology results out of which G2 and hmψ̄ψi can be extracted.
From a recent review14 we quote
G2 = (0.24 ± 0.011) GeV4 (36)
3
hψ̄ψi1GeV2 = −0.13 GeV (37)
Two questions arise

a) Is the Wilson O.P.E. eq.(30) well defined?

b) Can the constants G2 , Cψ be determined by a non perturbative formu-

lation of the theory, like lattice?

The answer to the first question is that it is ambiguous, due to the presence of
renormalons5. The answer to the second is yes, modulo the same ambiguity.
In spite of this ambiguity, however, keeping the leading terms of the per-
turbative expansion, sum rules do work, and so does the lattice determination,
which agrees with phenomenology.
We will sketch the argument for the first question, we will show that it
applies in the same way to lattice determinations, and will present anyhow
the successful results from lattice, which agree with the values (36,37).

lisbws: submitted to World Scientific on February 1, 2008 9

 The perturbative expansion of any observable, e.g. the correlator of two
currents like Πµν , eq.(32), is finite order by order in perturbation theory, i.e.

(n)
X
C1 (αs ) = C1 αns (38)
n

(n)
where αs is the measured coupling constant, and C1 are finite numerical
coefficients. It can be shown that4

(n)
C1 ∝ n!na bn (39)

where a does not depend on the operator, b does depend on it. Of course the
dependence (39) on n makes the series non convergent. However the series
could be considered as the asymptotic expansion of a convergent series if the
Borel transform

∞ n
(n+1) b
X
C̄1 (b) = C1 (40)
n=0
n!

is convergent. In that case, since

Z
bn e−b/αs db = αn+1
s n!

the asymptotic expansion

Z ∞
C1 (αs ) − C1 (0) = dbe−b/αs C̄1 (b)db (41)
0

is just eq.(38).
Most probably this is not the case for C1 (αs ).
The resummation of a subseries of graphs shows that C1 (αs ) is ambiguous
in principle by terms which mimic the condensate15 .
Consider the QCD correction Π1 (q 2 ) to the correlator, corresponding to
the following graphs

lisbws: submitted to World Scientific on February 1, 2008 10

 +2

1 d4 k
Z
Π1 (q 2 ) = αs G(k, q 2 ) (42)
q2 (2π)4
1 b0
Replacing αs by its running value at one loop αs (k), 1/αs (k) = αs + 2π ln µk
amounts to resum a subset of graphs and gives
−1
d4 k

1 1 b0 k
Z
2
Π1 (q ) = 2 + ln G(k, q 2 ) (43)
q (2π)4 αs 2π µ
At small k’s, G(k, q 2 ) is finite and equal to 4Nc /q 2 and hence (43) has a
divergence at k ≪ µ. The divergent part is
 −1
Nc αs 1 b0 k
Z
Π1 (q 2 ) ≃ 2 4 k 3 dk + ln (44)
π q αs 2π µ
The prescription of the singularity is not defined. Accordingly there is an
ambiguity
Nc αs µ4 − b 4πα
Π1 (q 2 ) ≃ e 0 s (45)
π2 q4
which mimics the terms proportional to the condensate.
The argument is only an indication, since the subset of graphs considered
is not even gauge invariant. If the indication is correct it means that the Borel
transform is singular, and condensates cannot be defined
As already anticipated, in spite of that sum rules work and give a consis-
tent phenomenology.
A similar problem arises when one tries to extract G2 from lattice16,17,18 .

lisbws: submitted to World Scientific on February 1, 2008 11

 Consider the gauge invariant field strength correlator19

Dµν,ρσ = h0|T r Gµν (x)S(x, 0)Gρσ (0)S † (x, 0) |0i


(46)

where
X
Gµν = T a Gaµν
a
 Z x 
SC (x, 0) = P exp i Aµ (x)dxµ
0,C

a
T are the generators of the gauge group and SC (x, 0) is the parallel transport
from 0 to x in the fundamental representation. S depends on the path C: our
choice for C will be a straight line.
Dµν,µν can be considered as the split point regulator of G2 , à la Schwinger.
The OPE gives
CI
Dµν,µν ≃ hIi + C4 G2 + . . . (47)
x→0 x4
< G2 > is defined if the term ∝ x14 can be unambigously subtracted. Again
like for the correlators, higher order perturbative terms containing log’s can
sum up and simulate a constant term, like G2 , with an intrinsic ambiguity in
the prescription.
The most general form of the correlator compatible with Poincaré invari-
ance is expressed in terms of two independent scalars19 D(x2 ), D1 (x2 )

Dµρ,νσ = (gµν gρσ − gµσ gνρ ) D(x2 ) + D1 (x2 ) +

 
(48)
∂D1
+ (xµ xν gρσ − xµ xσ gνρ + xρ xσ gµν − xν xρ gµσ ) 2
∂x
Calling for simplicity 0 the direction of xµ two other invariants can be
defined17
3
1X ∂D1
D|| (x2 ) = D0i,0i = D + D1 + x2 2 (49)
3 i=1 ∂x
3
1X
D⊥ (x2 ) = Dij,ij = D + D1 (50)
3 i<j

D|| and D⊥ can be computed on the lattice17 .

Their behaviour is shown in fig.1 for quenched theory.

lisbws: submitted to World Scientific on February 1, 2008 12

 12
10

10
10

8
10

6
10
0.0 0.2 0.4 0.6 0.8 1.0
FM

A parametrization
|x| a − |x|′
D(x2 ) = G2 e− λ + e λ (51)
|x|4
allows to extract from the data G2 and λ.
For quenched SU (3)
G2 = (0.15 ± 0.03) GeV4 λ = (0.22 ± 0.02) fm (52)
For full QCD and 4 quarks at different values of the quark masses (in units
of inverse lattice spacing) the result is smaller than the quenched one by an
order of magnitude and the correlation length is larger
am = 0.01 G2 = (0.015 ± 0.005) GeV4 λ = (0.34 ± 0.04) fm (53)
4
am = 0.02 G2 (0.031 ± 0.005) GeV λ = (0.29 ± 0.03) fm (54)
By use of the relation20
dG2 24
=− hq̄qi (55)
dmq 11 − 23 Nf
and of the known values of hq̄qi an extrapolation can be made to the physical
quark mass, giving18
G2 = (0.022 ± 0.005) GeV4 (56)
which compares satisfactoraly with the sum rule determination, eq.(37).
Old results for SU (2) give21
G2 = (0.10 ± 0.04) GeV4 (57)

lisbws: submitted to World Scientific on February 1, 2008 13

A few words on the technique used in ref.17,18 . Correlators measured on
a statistical ensemble of configurations are very noisy due to local quantum
fluctuations of the fields. A local cooling procedure is used to reduce quantum
fluctuations22 . After a number nt of cooling steps the size λ of the fluctuations
which are cooled away follows a diffusion equation
λ2 = Dnt (58)
If the distance to explore is large enough (e.g. 3-4 lattice spacing) cooling can
eliminate local noise and reduce the fluctutions17,18 .
Field correlators as determined from lattice are useful inputs in the sto-
castic approach19 to QCD vacuum, for which I refer to the lectures of Y.
Simonov of this school.
An alternative procedure to determine G2 is to use the plaquette as a
regulator, and measure the average of the plaquette itself16 . A perturbative
contribution appears, which is presumably non Borel summable, which must
be somehow subtracted to isolate the non perturbative part. Again there is
ambiguity. However if the perturbative subtraction is made at some order,
the resulting G2 fits what is found by other methods.
By a similar procedure the quark condensate can be extracted18 from the
correlator
C1
mhψ̄(x)S(x, 0)ψ(0)i ≃ 2 + C2 hmψ̄ψi (59)
x
The result is, at β = 5.35
 
mπ
|hψ̄ψi|1/3 = (205 ± .21) MeV am = 0.02 = 0.65(3) (60)
mρ
 
mπ
(160 ± 12) MeV am = 0.01 = 0.57(2) (61)
mρ
This approach is like sum rules: it works but it is not understood why.

3 Confinement of colour.

3.1 Confinement.
The particles corresponding to the fundamental fields of the theory, quarks
and gluons, have never been observed in nature (confinement of colour).
In nature the relative abundance of quarks to nucleons has an upper
bound
nq
≤ 10−27 (62)
np

lisbws: submitted to World Scientific on February 1, 2008 14

coming from Millikan like analysis of ∼ 1 gr of matter.
The expectation in the absence of confinement is23
nq
≥ 10−12 (63)
np
a factor 1015 larger.
A factor 10−15 cannot be explained by fine tuning of a small parameter:
the only natural explanation is in terms of symmetry. We shall adopt this
prejudice as a starting point.
Lattice provides evidence that confinement is present in QCD. The well
known area law of Wilson loops24 , W (R, T ) ≃ exp(−RT σ) defines the static
potential between q q̄ as
V (R) = Rσ (64)
σ is known as string tension: σ 6= 0 indicates confinement, meaning that an
infinite amount of energy is necessary to pull apart q and q̄.
Finite temperature QCD can be obtained from the Euclidean partition
function, with a time extension 1/T , and periodic boundary conditions for
bosons, antiperiodic for fermions.
On the lattice this is done by using a lattice NS3 NT of extension NS ≫ NT
in each space direction and
1
NT a = (65)
T
with a the lattice spacing.
By renormalization group arguments at large enough β, a depends on β
as
1
a(β) ≃ exp(−b0 β) (66)
ΛL
−b0 is the first coefficient in the expansion of the β function, 1/ΛL the physical
length scale of the theory.
Because of asymptotic freedom a decreases exponentially to 0 as β → ∞.
The temperature is given by eq.(65,66) as
ΛL
T = exp(b0 β) (67)
NT
and is an increasing function of β. At low temperature σ 6= 0. By increasing
T (or β) σ decreases and at some temperature Tc it goes to zero. Tc is called
the deconfining temperature25,26 .

lisbws: submitted to World Scientific on February 1, 2008 15

 An alternative signal for the transition is given by the Polyakov line
I
P = hT exp(iA0 (~n, x0 )dx0 )i (68)

which is the parallel transport along the time axis from n0 to infinity and
back via periodic boundary conditions.
It can be shown that P is related to the chemical potential of an isolated
quark, µq as
µq
P = exp(− ) (69)
T
In the confined phase µq diverges and P = 0, while P = 6 0 in the deconfined
phase. Such behaviour is indeed observed on the lattice, again providing
evidence that there is a deconfining transition25 .
P is called an order parameter, being 6= 0 in the weak coupling regime.
σ is called a disorder parameter, being 6= 0 in the strong coupling regime.
A drawback of σ and P as order parameters is that, in the presence
of quarks they lose their meaning. Indeed the string should break at some
distance creating q q̄ pairs, and σ is not defined. As for P, it is an element
of the centre of the group Z3 and signals the breaking of Z3 symmetry in
the confined phase. In the presence of quarks Z3 is not a symmetry. The
order parameter usually assumed in full QCD is the chiral symmetry breaking
order parameter hψ̄ψi. A priori, however, chiral symmetry is not related to
confinement, although arguments show that it is if the philosophy of Nc → ∞
is correct. In the same philosophy, however one should be able to define
an order (or disorder) parameter which is independent of the presence of
quarks, at least for not a too large number of light quarks, which would spoil
asymptotic freedom.
Because of asymptotic freedom the confined phase corresponds to disor-
der, the deconfined phase to order. In the language of statistical mechanics
one should then look for the symmetry of the system dual to QCD27 .

3.2 Order disorder duality.

Order disorder duality is a property of d + 1 dimensional systems, with d-
dimensional extended configurations µ having non trivial topology, and car-
rying a conserved topological charge.
In the weak coupling regime the system is usually described in terms of
local fields Φ, and the symmetry of the ground state by order parameters hΦi,
which are vacuum expectation values of some field Φ. A phase transition can
occur at higher couplings, where hΦi → 0 and the system becomes disordered.

lisbws: submitted to World Scientific on February 1, 2008 16

 However a dual description can be given of the system, in which the
configurations µ become local fields, the original Φ’s become extended objects,
the coupling g → 1/g, and hence the disordered phase looks ordered, with
order parameter hµi. hµi, the order parameter of the dual phase is called a
disorder parameter (which means order parameter of the dual).
The prototype example of duality is the 1 + 1 dimensional Ising model28 .
The model is described by a field σ which takes values ±1 on the sites of a
2d square lattice, and is exactly solvable in the thermodynamical limit. The
interaction is
2 X
( 2 )
X XX
− σ(i)σ(i + k̂) Z = exp β σ(i)σ(i + k̂) (70)
k=1 i k=1 i

The model has 1 dimensional configurations with nontrivial topology, the

kinks, the topology being identified by the values of σ at x = ±∞ at fixed t.
A kink is a highly non local configuration with
σ = −1 x ≤ x0 σ = +1 x > x0 (71)
A dual variable σ ∗ can be defined which has also values ±1 on the sites of the
dual lattice, whose sites corresponds to the links of the original one.
At fixed t σ ∗ is −1 if the link joins neighbourghs sites on which σ has the
same sign, +1 otherwise.
In terms of σ ∗ a link looks as σ ∗ = +1 at x0 , σ ∗ = −1 everywhere else.
The kink becomes a local excitation in the language of σ ∗ .
In ref.28 it was shown that the partition function in terms of σ ∗ is again
an Ising model with the correspondence
Z[σ, β] = Z[σ ∗ , β ∗ ] (72)
and
1
sinh2β ∗ = (73)
sinh2β
The mapping to the dual sends β → 1/β i.e. the weak coupling region into
strong coupling region and viceversa. The high temperature phase which looks
disordered, is ordered and hσ ∗ i is its order parameter. The order parameter
of the dual phase is called a disorder parameter.
Other known systems which admit a dual description and a disorder pa-
rameter are

a) The (2 + 1)d XY model, (liquid He4 ). Here the field is θ(i) an angle, or
better Aµ = ∂µ θ(i). The topological excitations are 2d vortices. Topology

lisbws: submitted to World Scientific on February 1, 2008 17

 is given by their winding number and
I
~ · d~x
A

which is an integer if the field θ is single valued. Vortices condense in the

disordered phase29 .

b) The (2 + 1)d Heisenberg model30 . Here the topological excitations which

condense in the vacuum above the Curie temperature, when the mag-
netization goes to zero, are non abelian vortices, the instantons of the
2 dimensional O(3) σ model. Topology is the winding number of the
mapping of the 2 dimensional sphere S2 on SO(3).

c) In (3 + 1) dimensional gauge theories the topological excitations are

monopoles27 , whose topology comes from the mapping S2 on SU (2) or
S1 on U (1). Examples are the (3 + 1) dimensional U (1) gauge theory31 ,
the (3 + 1) dimensional non abelian gauge theory, the (3 + 1) dimensional
supersymmetric QCD32 .

To understand disorder one can adopt two strategies.

1) Perform the transformation to dual. This has been done in ref.28 for the
Ising model, in ref.33 for the U (1) compact Villain action, in ref.32 for
the SUSY QCD.

2) Guess the symmetry of the dual phase and write the disorder parameter
hµi in terms of the original local fields Φ. This line has been pursued in
ref.34 on the continuum and by the Pisa group on the lattice27,30,31,35,36 .

The advantage of strategy n.2, which we shall adopt, is that the symmetry
can be explored without going through the hardly manageable procedure of
constructing the dual.
The general principle is to create by an operator µ, constructed in terms
of the local fields Φ, a dual excitation carrying non trivial topological charge,
and investigate dual order by directly computing hµi, the disorder parameter.
This is done by the field theoretical analog of translation operator,
eipa |xi = |x + ai (74)
The operator37
" #
XZ
µ(~y , t) = exp i d3 ~xΠk (~x, t) Φ̄k (~x, ~y ) (75)
k

lisbws: submitted to World Scientific on February 1, 2008 18

with Πi (~x, t) the conjugate momentum to ϕi (~x, t) and Φ̄i (~x, ~y ) the field config-
uration produced by a topological excitation located at ~y . In the Schrödinger
representation, where the fields ϕi (~x, t) are diagonal,
µ(~y , t)|ϕi (~x, t)i = |ϕi (~x, t) + Φ̄i (~x, ~y)i (76)
A topological configuration Φ̄ has been added to the original one.
Care is needed to adapt eq.’s(75,76) to compact theories, where the fields
cannot be arbitrarily shifted. Another important point is that hµi is only
defined in the thermodynamical limit. Indeed by eq.(75), if the number of
degrees of freedom is finite hµi is an analytic function of β, and cannot vanish
identically in the ordered phase without vanishing identically. This is only
possible when V → ∞ and Lee-Yang singularities develop in the partition
function. All that can be mastered. We have computed hµi for a number of
models (in the 3d XY model, the Heisenberg magnet, the (3 + 1)d compact
U (1)) where the symmetry of the dual is known. We have thus a tool to
explore the deconfining transition in QCD.
As we shall discuss in QCD the precise symmetry of the dual system is
not known. Configurations candidate to condense in the confined phase are
monopoles. It is indeed a long standing idea38,39 , that confinement could be
produced by dual Meissner effect. Dual here is intended in the sense electric-
magnetic, i.e. that the roles of electric and magnetic fields are interchanged
with respect to an ordinary superconductor. The chromoelectric field acting
between a q̄q pair is expected, by this mechanism, to be confined by dual
Meissner effect into Abrikosov flux tubes, with energy proportional to the
length.
The string tension σ is then the energy of the tube per unit length.
The disorder parameter would then be the v.e.v. hµi of any operator which
creates a monopole. hµi = 6 0 signals that monopoles condense in the vacuum,
or, under very general assumptions, that the system is a dual superconductor.
The preliminary question is then to identify the monopoles.

3.3 Monopoles in non abelian gauge theories.

~
Let us first consider SU (2) gauge group, for the sake of simplicity. Let Φ(x)~
σ
be any local operator in the adjoint representation, in the usual notations.
We can define the unit vector parallel to Φ ~ as Φ̂ = Φ/|
~ Φ|,
~ everywhere in
~
a configuration, except at the zeros of Φ(x).
Let us define40 the field operator Fµν as

~ µν − 1 (Dµ Φ̂ ∧ Dν Φ̂) · Φ̂
Fµν = Φ̂G (77)
g

lisbws: submitted to World Scientific on February 1, 2008 19

 ~
Fµν is a colour singlet, defined everywhere except at zeros of Φ(x), and is
invariant under non singular gauge transformations.
Here
~ µν = ∂µ A
G ~ µ + gA
~ ν − ∂ν A ~ µ ∧ Aν (78)
is the gauge field strength and Dµ = ∂µ − ig A ~ µ ∧ is the covariant derivative.
The coefficient in front of the second term is chosen in such a way that
the bilinear terms in A~ µ cancel between the two terms. Indeed one has

Fµν = Φ̂ · (∂µ A ~ µ ) − 1 (∂µ Φ̂ ∧ ∂ν Φ̂) · Φ̂

~ ν − ∂ν A (79)
g
Define
∗ 1
Fµν = εµνρσ F ρσ (80)
2
and
jνM = ∂µ Fµν
∗
(81)
Then jνM is zero wherever Φ̂ is defined, i.e. everywhere except at the zeros of
~ and is in any case conserved since Fµν
Φ, ∗
is antisymmetric.
∂µ jµM = 0 (82)
Eq.(82) identifies a conservation law, a magnetic U (1) symmetry.
One can operate a gauge transformation bringing Φ̂ along a fixed axis.
By eq.(79) then, in all nonsingular points
Fµν = ∂µ A3ν − ∂µ A3ν (83)
is an abelian field. That is why such transformation is called an abelian
projection.
It can be shown that at the singularities of Φ̂ a non zero magnetic cur-
rent is present. The regular field is a magnetic field of a monopole sitting
at the singularity. The singularity of the gauge transformation which per-
forms the abelian projection generates the Dirac string which ensures flux
conservation41 .
The abelian projection of the t’Hooft Polyakov classical monopole
configuration40 does indeed demonstrate that Fµν is the field of a static Dirac
monopole.
~ a conserved magnetic charge can
In general, for an arbitrary choice of Φ,
be defined. So a functional infinity of conserved monopole charges exists.
Which of them condense in the vacuum to produce dual superconductiv-
ity, if any?
There are in literature two different attitudes on the problem.

lisbws: submitted to World Scientific on February 1, 2008 20

 ~ are physically equiv-
a) (t’Hooft)27 All monopoles, whatever the choice of Φ,
alent. All of them should condense in the confined phase, to produce dual
superconductivity of the vacuum, and go to normal state above Tc .
b) Some monopole are more equal than others. The maximal abelian pro-
jection identifies the relevant monopoles42 .
We have checked the two alternatives on the lattice as explained in the fol-
lowing sections36 .
A similar construction works for SU (3), where two different monopoles
a
species can be defined for each operator Φ = λ2 Φa in the adjoint representa-
tion, carrying two different magnetic charges36.

4 Some details on the disorder parameter hµi.

The recipe of eq.(75) amounts to say that

Z[S + ∆S]
hµi = (84)
Z[S]
Indeed
1
Z Y
hµi = [ dϕ] exp[−βS − β∆S] (85)
Z
Let us consider U (1) for simplicity
1
exp[−β∆S] = exp {−β [(Π0i (θ0i + bi ) − Π0i (θ0i ]} β= (86)
e2
Here 1/ebi (~x − ~
y ) is the vector potential describing the field of a monopole.
Π0i = [cos(θ0i ) − 1] is the plaquette 0, i.
In the weak coupling θ0i = Ei , the electric field and
1
∆S ≃ − [eEi ]bi
e2
gives exactly the definition (75) for µ.
A similar definition can be given for QCD36 .
Instead of computing hµi, which is affected by wild fluctuations, it proves
convenient to compute
d
ρ= lnhµi (87)
dβ
From eq.(84)
ρ = hSiS − hS + ∆SiS+∆S (88)

lisbws: submitted to World Scientific on February 1, 2008 21

In terms of ρ

"Z #
β
′ ′
hµi = exp ρ(β )dβ (89)
0

since by definition hµiβ=0 = 1.

In the thermodynamical limit one expects in finite temperature QCD that,
if monopole condensation is related to confinement,

lim hµi = 0 β > βc (90)

V →∞
lim hµi =
6 0 β < βc (91)
V →∞
hµi ≃ (βc − β)δ β ∼ βc (92)

At finite volume the qualitative behaviour of hµi is shown in fig.2, the quali-
tative behaviour of ρ(β) in fig.3.

<µ>
βC

β
Fig.2

lisbws: submitted to World Scientific on February 1, 2008 22

 ρ β

βc

Fig.3
For β > βc (weak coupling) ρ can be computed in perturbation getting,
in terms of the volume

hµi = exp(−c1 V + c2 ) c1 > 0 (93)

hµi tends to zero as V → ∞.

For β < βc ρ stays finite at V → ∞, or hµi is 6= 0.
Around βc the phase transition takes place, which is known to be second
order for SU (2) gauge group, weak first order for SU (3). This means that
the correlation length goes large for both, since both of them look like second
order. On dimensional grounds
ξ a
µ ≃ µ( , ) ξ ≃ (βc − β)−ν (94)
L ξ
ξ being the correlation length and a the lattice spacing.
Near the critical point a/ξ ∼ 0 and
ξ
µ = µ( , 0) = f (L1/ν (βc − β)) (95)
L
Eq.(94) has been used.

lisbws: submitted to World Scientific on February 1, 2008 23

 d
This gives the scaling law for ρ = dρ ln β

ρ
1/ν
= F ((L1/ν (βc − β)) (96)
L

From that scaling ν and βc can be determined, together with the critical index
δ.
If condensation is related to confinement the determination should be
consistent with other methods.
The behaviour of ρ for monopoles defined by different choices of Φ is shown
in fig.4: it looks the same for all of them, in agreement with the statement of
t’Hooft that all monopoles are physically equivalent.
0.0

−100.0

−200.0
ρ

−300.0

−400.0 Polyakov gauge

butterfly gauge
plaquette gauge

−500.0
2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0
β

Fig.4
Also the two different monopoles corresponding to the same abelian pro-
jection look identical.

lisbws: submitted to World Scientific on February 1, 2008 24

 0.0

-100.0

-200.0
ρ

-300.0

-400.0 T3
T8
V3-U3

-500.0
0.0 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0
β

Fig.5

−1.5

−2.0
1/ν

−2.5
ρ/NS

NS = 12
NS = 16
−3.0 NS = 20
NS = 24
fit

−3.5
−5.0 0.0 5.0 10.0 15.0
NS (β−βc)
1/ν

Fig.6
The quality of scaling law for SU (2) is shown in fig.6, for the optimal

lisbws: submitted to World Scientific on February 1, 2008 25

determination of βc and ν.
The result for ν is

ν = 0.62 ± 0.05 (97)

and for δ

δ = 0.18 ± 0.06 (98)

For SU (3) the transition fig.7

−0.020

−0.040
1/ν

−0.060
ρ/NS

NS = 16
NS = 20
−0.080 NS = 24
NS = 32
fit

−0.100
−100.0 0.0 100.0 200.0 300.0 400.0 500.0
NS (βC − β)
1/ν

Fig.7
is first order with

ν = 0.33 ± 0.07 δ = 0.54 ± 0.04 (99)

Monopoles do condense in the vacuum below Tc , supporting thus the mech-

anism of confinement by dual superconductivity. Monopole identified by dif-
ferent abelian projections behave in the same way. As a check the finite size
scaling analysis of hµi provides determination of βc and of the critical indices
consistent with other methods.
The disorder parameter hµi can be defined and determined also in full
QCD, contrary to σ and P . Preliminary results give evidence for monopole
condensation also there.

lisbws: submitted to World Scientific on February 1, 2008 26

 The method has been checked on 2d Ising model, on 3d XY model, on
Heisenberg magnet, on U (1) with similar results, consistent with completely
different determinations.

5 The symmetry of the confined phase in QCD.

What we have learned from lattice is that QCD vacuum is a dual supercon-
ductor in the confined phase, and goes normal at the deconfining transition.
And this independently of the choice of the magnetic charge.
This solves the problem that in a given abelian projection some colored
states could be neutral with respect to the abelian electric charge, and hence
not confined.
However the situation is not clear. We do not understand yet the symme-
try pattern of the ground state, except for the fact that any magnetic charge
condensate is a good disorder parameter. There must exist for sure a more
synthetic way of describing this situation.
An attractive feature of our order parameter is that it is independent
of the presence of quarks, and this fits the idea that the Nc → ∞ theory
is a good basic description of physics, O(1/Nc ) corrections and higher being
perturbations which do not alter its basic features.

References

1. R.P. Feynman, Rev. Mod. Phys.20, 267 (1987); R.P. Feynman A.R.
Hibbs: Quantum Mechanics and Path Integrals (Mc Graw Hill, NY 1965)
2. K.G. Wilson, Phis. Rev.D 10, 2445 (1974).
3. D. Gross, F. Wilczeck, Phys. Rev. Lett. 30, 3497 (1973); H.D. Politzer,
Phys. Rev. Lett. 30, 1346 (1973).
4. Large Order Behaviour of Perutrbation Theory, eds.J. Le Guillou, J. Zinn-
Justin, (North Holland Amsterdam 1990).
5. A.H. Müller, Nucl. Phys. B 250, 327 (1985).
6. C.G. Callan, R.F. Dashen, D.J. Gross, Phys. Rev. D 17, 2717 (1978),
Phys. Rev. D 18, 4684 (1978), Phys. Rev. D 19, 1826 (1979).
7. A. Belavin, A. Polyakov, A. Schwartz, Yu. Tyupkin, Phys. Lett. B 59,
85 (1975).
8. E.V. Shuryak, Nucl. Phys. B 203 93 (1982), B 214 237 (1983); D.I.
Dyakonov, V.Y. Petrov, Nucl. Phys. B 245 259 (1984).
9. G. ’t Hooft, Nucl. Phys. B 72, 461 (1974);
G. Veneziano, Nucl. Phys. B 117, 519 (1976);
E. Witten, Nucl. Phys. B 149, 285 (1979).

lisbws: submitted to World Scientific on February 1, 2008 27

10. E. Witten, Nucl. Phys. B 256 269 (1979);
G. Veneziano, Nucl. Phys. B 159 213 (1979).
11. B. Alles, M. D’Elia, A. Di Giacomo, Nucl. Phys. B 434 481 (1997)
12. W. Zimmerman, in Lectures on Elementary Particles and Quantum
Field Theory - 1970 Brandeis University Summer Institute in Theoret-
ical Physics, eds. S. Deser, H. Pendleton and M. Grisaru, (MIT Press,
Cambridge 1970)
13. M.A. Shifman, A.I. Vainshtein, V.I. Zakharov, Nucl. Phys. B 147,
385,448,519 (1979).
14. S. Narison, Phys. Lett. B 387, 162 (1996).
15. H.G. Dosh in Selected Topics in non Perturbative QCD, (S.I.F. Varenna
1995, eds. A. Di Giacomo and P. Dyakonov).
16. A. Di Giacomo, G.C. Rossi, Phys. Lett. B 100, 481 (1981);
M. Campostrini, A. Di Giacomo, Y. Gündüc, Phys. Lett. B 225, 393
(1983).
17. A. Di Giacomo, H. Panagopoulos, Phys. Lett. B 285, 133 (1992);
A. Di Giacomo, E. Meggiolaro, H. Panagopoulos, Nucl. Phys. B 483,
371 (1997).
18. M. D’Elia, A. Di Giacomo, E. Meggiolaro, Phys. Lett. B 408, 315 (1997).
19. H.G. Dosh, Phys. Lett. B 190, 177 (1987);
H.G. Dosh, Yu. A. Simonov, Phys. Lett. B 205, 339 (1988).
20. V.A. Novikov, M.A. Shifman, A.I. Vainshtein, V.I. Zakharov, Nucl. Phys.
B 191, 301 (1981).
21. M. Campostrini, A. Di Giacomo, G. Mussardo, Z. Phys. C 25, 173
(1984).
22. M. Campostrini, A. Di Giacomo, M. Maggiore, H. Panagopoulos, E. Vi-
cari, Phys. Lett. B 225, 403 (1989).
23. L.B. Okun, Leptons and quarks (North Holland, Amsterdam) 1982.
24. M. Creutz, Phys. Rev. D21 2308 (1980).
25. J. Engels, J. Jersak, K. Kanaye, E. Laerman, C.B. Lang, T. Neühaus, H.
Satz, Nucl. Phys. B 347 441 (1980).
26. A. Di Giacomo, M. Maggiore and Š. Olejnı́k, Nucl. Phys. B 347, 441
(1990).
27. G. ’t Hooft, Nucl. Phys. B 138, 1 (1978), Nucl. Phys. B 190, 455
(1981).
28. H.A. Kramers, G.H. Wannier, Phys. Rev. 60, 252 (1941); L.P. Kadanoff,
[Link], Phys. Rev. B 3 3918 (1971).
29. G. Di Cecio, A. Di Giacomo, G. Paffuti, M. Trigiante, Nucl. Phys. B
483, 739 (1997).
30. A. Di Giacomo, D. Martelli, G. Paffuti Phys. Rev. D 60, 094511 (1999).

lisbws: submitted to World Scientific on February 1, 2008 28

31. T.A. DeGrand, D. Toussaint, Phys. Rev. D 22, 2478 (1980).
32. N. Seiberg, E. Witten, Nucl. Phys. B 341, 484 (1994).
33. J. Fröhlich, P.A. Marchetti, Commun. Math. Phys.112 343 (1987).
34. E. Marino, B. Schroer, J.A. Swieca, Nucl. Phys. B 200, 473 (1982).
35. A. Di Giacomo, G. Paffuti, Phys. Rev. D 56, 6816 (1997).
36. A. Di Giacomo, B. Lucini, L. Montesi, G. Paffuti, Colour Confinement
and Dual Superconductivity of the Vacuum, I and II, to appear in Phys.
Rev. D 1 (1999).
37. L. Del Debbio, A. Di Giacomo, G. Paffuti, P. Pieri, Phys. Lett. B 355,
255 (1995).
38. G. ’t Hooft, in High Energy Physics, ed. A. Zichichi (EPS International
Conference, Palermo 1975).
39. S. Mandelstam, Phys. Rep. 23C, 245 (1976).
40. G. ’t Hooft, Nucl. Phys. B 79, 276 (1974);
A.M. Polyakov, JETP Lett. 20 894 (1974).
41. A. Di Giacomo, M. Mathur, Phys. Lett. B400, 129, (1997), Nucl. Phys.
B 531, 302 (1998).
42. T. Suzuki, Nucl. Phys. B 30 (Proc. Suppl.), 176 (1993).

lisbws: submitted to World Scientific on February 1, 2008 29