You are on page 1of 161

He rea

hed a mu h higher plane of reativity when he bla ked out everything but a,
an and the. That ere ted more dynami
intralinear tensions.

, Cat h-22

J. Heller

Non-perturbative Methods
in Gauge Theory

A Set of Le tures

by

Y. M. Makeenko

Samizdat, 1995

Yuri Makeenko
 Institute of Theoreti al and Experimental Physi s, Mos ow, Russia and
The Niels Bohr Institute, Copenhagen, Denmark

Prefa e
These le ture notes are based on ourses given at:
1) Autonoma University of Madrid, Winter Semester of 1993;
2) Leipzig University, Winter Semester of 1995;
3) Mos ow Physi al and Te hni al Institute, Spring Semester of 1995.
My intention was to introdu e graduate and Ph.D. students to the methods of ontemporary quantum eld theory. The term \non-perturbative" in
the title means literally \beyond the s ope of perturbation theory". Therefore, it is assumed that the reader is familiar with quantum me hani s as well
as with the standard methods of perturbative expansion in quantum eld
theory and, in parti ular, with the theory of renormalization.
The se ond purpose was to make the ourse useful for senior people (in luding those working in ondensed-matter theory), as a survey of ideas, terminology and methods, whi h were developed in quantum eld theory in the
seventies and the beginning of the eighties. For this reason, these notes do not
go very deeply into details, so the presentation is sometimes a bit s hemati .
Correspondingly, the subje ts whi h are usually overed by modern ourses
in string theory, su h as the two-dimensional onformal eld theories, are not
tou hed. It is assumed that su h a ourse will follow this one.
The main body of the le ture notes deals with latti e gauge theories and
large-N methods. These two Chapters are pre eded by Chapter 1 whi h
is devoted to the method of path integrals. The path-integral approa h is
loosely used in quantum eld theory and statisti al me hani s. In Chapter 1,
I shall pay most attention to aspe ts of the path integrals, whi h are then
used in the next two Chapters.
In ea h Chapter, I was going to be as losed to the original papers, where
the involved methods were proposed, as possible. The list of these papers
respe tively in ludes:
i

ii
1. R.P. Feynman, An operator al ulus having appli ations in quantum
ele trodynami s, Phys. Rev. 84 (1951) 108.
2. K.G. Wilson, Con nement of quarks, Phys. Rev. D10 (1974) 2445.
3. G.'t Hooft, A planar diagram theory for strong intera tions, Nu l. Phys.
B72 (1974) 461.
The le tures were followed by seminars where some problems for deeper
studies had been solved on a bla kboard. They are inserted in the text as
the problems, whi h an be omitted at rst reading. Some more information
is also added as remarks after the main text. Both of them ontain some
relevant referen es.
The referen es, whi h are olle ted at the end of ea h Chapter, are usually given only to either a rst paper (or papers) in a series or those ontaining a pedagogi al presentation of the material. With the modern ele troni
database at qspires (SLAC), a list of subsequent papers an, in most ases,
be retrieved by downloading itations of the rst paper.
I would like to thank the students for their attention, patien e, and questions. I am indebted to Martin Gurtler for his help in preparing these le ture
notes.
1995{1996

Y. M.

Contents
1 Path Integrals

1.1 Operator al ulus . . . . . . . . . . . . . . . . .


1.1.1 Free propagator . . . . . . . . . . . . . .
1.1.2 Eu lidean formulation . . . . . . . . . . .
1.1.3 Path-ordering of operators . . . . . . . .
1.1.4 Feynman disentangling . . . . . . . . . .
1.1.5 Cal ulation of the Gaussian path integral
1.1.6 Transition amplitudes . . . . . . . . . . .
1.1.7 Propagators in external eld . . . . . . .
1.2 Se ond quantization . . . . . . . . . . . . . . . .
1.2.1 Integration over elds . . . . . . . . . . .
1.2.2 Grassmann variables . . . . . . . . . . . .
1.2.3 Perturbation theory . . . . . . . . . . . .
1.2.4 S hwinger{Dyson equations . . . . . . .
1.2.5 Commutator terms . . . . . . . . . . . .
1.2.6 S hwinger{Dyson equations ( ontinued) .
1.2.7 Regularization . . . . . . . . . . . . . . .
1.3 Quantum anomalies from path integral . . . . .
1.3.1 QED via path integral . . . . . . . . . . .
1.3.2 Chiral Ward identity . . . . . . . . . . .
1.3.3 Chiral anomaly . . . . . . . . . . . . . .
1.3.4 Chiral anomaly ( al ulation) . . . . . . .
1.3.5 S ale anomaly . . . . . . . . . . . . . . .
1.4 Instantons in quantum me hani s . . . . . . . . .
1.4.1 Double-well potential . . . . . . . . . . .
1.4.2 The instanton solution . . . . . . . . . . .
1.4.3 Instanton ontribution to path integral . .
iii

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

2
2
5
10
12
17
20
29
35
35
37
39
40
41
42
46
48
48
49
52
56
61
66
66
69
72

iv

CONTENTS

1.4.4 Symmetry restoration by instantons . . . . . . . . . .


1.4.5 Topologi al harge and -va ua . . . . . . . . . . . . .
1.5 Referen e guide . . . . . . . . . . . . . . . . . . . . . . . . . .

2 Latti e Gauge Theories

2.1 Observables in gauge theories . . . . . . . . . . . .


2.1.1 Gauge invarian e . . . . . . . . . . . . . .
2.1.2 Phase fa tors (de nition) . . . . . . . . . .
2.1.3 Phase fa tors (properties) . . . . . . . . . .
2.1.4 Aharonov{Bohm e e t . . . . . . . . . . .
2.2 Gauge elds on a latti e . . . . . . . . . . . . . . .
2.2.1 Sites, links, plaquettes and all that . . . . .
2.2.2 Latti e formulation . . . . . . . . . . . . .
2.2.3 The Haar measure . . . . . . . . . . . . . .
2.2.4 Wilson loops . . . . . . . . . . . . . . . . .
2.2.5 Strong oupling expansion . . . . . . . . .
2.2.6 Area law and on nement . . . . . . . . .
2.2.7 Asymptoti s aling . . . . . . . . . . . . .
2.3 Latti e methods . . . . . . . . . . . . . . . . . . .
2.3.1 Phase transitions . . . . . . . . . . . . . . .
2.3.2 Mean- eld method . . . . . . . . . . . . . .
2.3.3 Mean- eld method (variational) . . . . . . .
2.3.4 Latti e renormalization group . . . . . . . .
2.3.5 Monte Carlo method . . . . . . . . . . . . .
2.3.6 Some Monte Carlo results . . . . . . . . . .
2.4 Fermions on a latti e . . . . . . . . . . . . . . . . .
2.4.1 Chiral fermions . . . . . . . . . . . . . . . .
2.4.2 Fermion doubling . . . . . . . . . . . . . .
2.4.3 Kogut{Susskind fermions . . . . . . . . . .
2.4.4 Wilson fermions . . . . . . . . . . . . . . .
2.4.5 Quark ondensate . . . . . . . . . . . . . .
2.5 Finite temperatures . . . . . . . . . . . . . . . . .
2.5.1 Feynman{Ka formula . . . . . . . . . . . .
2.5.2 QCD at nite temperature . . . . . . . . .
2.5.3 Con nement riterion at nite temperature
2.5.4 De on ning transition . . . . . . . . . . . .
2.5.5 Restoration of hiral symmetry . . . . . . .
2.6 Referen e guide . . . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

76
78
82

87

88
88
91
96
99
102
102
105
111
113
117
121
123
127
127
132
135
138
141
145
149
149
151
156
158
162
165
165
169
171
174
179
182

CONTENTS

3 1=N Expansion

3.1 O(N ) ve tor models . . . . . . . . . . . . . . . . . . .


3.1.1 Four-Fermi theory . . . . . . . . . . . . . . . .
3.1.2 Bubble graphs as zeroth order in 1=N . . . . .
3.1.3 Fun tional methods for '4 theory . . . . . . . .
3.1.4 Nonlinear sigma model . . . . . . . . . . . . . .
3.1.5 Large-N fa torization in ve tor models . . . . .
3.2 Multi olor QCD . . . . . . . . . . . . . . . . . . . . .
3.2.1 Index or ribbon graphs . . . . . . . . . . . . . .
3.2.2 Planar and non-planar graphs . . . . . . . . . .
3.2.3 Planar and non-planar graphs (the boundaries)
3.2.4 Topologi al expansion and quark loops . . . . .
3.2.5 t' Hooft versus Veneziano limits . . . . . . . . .
3.2.6 Large-N fa torization . . . . . . . . . . . . . .
3.2.7 The master eld . . . . . . . . . . . . . . . . .
3.2.8 1=N as semi lassi al expansion . . . . . . . . .
3.3 QCD in loop spa e . . . . . . . . . . . . . . . . . . . .
3.3.1 Observables in terms of Wilson loops . . . . . .
3.3.2 S hwinger{Dyson equations for Wilson loop . .
3.3.3 Path and area derivatives . . . . . . . . . . . .
3.3.4 Loop equations . . . . . . . . . . . . . . . . . .
3.3.5 Relation to planar diagrams . . . . . . . . . . .
3.3.6 Loop-spa e Lapla ian and regularization . . . .
3.3.7 Survey of non-perturbative solutions . . . . . .
3.3.8 Wilson loops in QCD2 . . . . . . . . . . . . . .
3.3.9 Gross{Witten transition in latti e QCD2 . . . .
3.4 Large-N redu tion . . . . . . . . . . . . . . . . . . . .
3.4.1 Redu tion of s alar eld . . . . . . . . . . . . .
3.4.2 Redu tion of Yang{Mills eld . . . . . . . . . .
3.4.3 Rd -symmetry in perturbation theory . . . . . .
3.4.4 Twisted redu ed model . . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

187

188
188
192
202
211
213
215
215
220
227
233
237
241
248
251
254
254
260
263
267
271
274
278
280
288
295
295
300
303
304

vi

CONTENTS

List of Figures
1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
1.9
1.10
1.11
1.12

Dire tion of the Wi k rotation . . . . . . . . . . . . .


The traje tory z (t) . . . . . . . . . . . . . . . . . . .
Dis retization of the traje tory z (t) . . . . . . . . . .
Diagrammati representation of Eq. (1.2.23) . . . . .
Some of the Feynman diagrams . . . . . . . . . . . . .
Triangular diagram asso iated with hiral anomaly . .
The diagram asso iated with hiral anomaly in d = 2 .
Diagrams whi h ontribute to the s ale anomaly . . .
Double-well potential . . . . . . . . . . . . . . . . . . .
Graphi representation of the one-kink solution . . . .
The many-kink on guration . . . . . . . . . . . . . .
Graphi representation of a periodi potential . . . . .

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

8
13
14
39
40
57
59
64
67
70
77
78

2.1
2.2
2.3
2.4
2.5
2.6
2.7
2.8
2.9
2.10
2.11
2.12
2.13
2.14
2.15

Re tangular loop in the ;  -plane . . . . . . . . . . . . . .


Prin ipal s heme of the Aharonov{Bohm experiment . . . .
Two-dimensional latti e with periodi boundary onditions
A link of a latti e . . . . . . . . . . . . . . . . . . . . . . . .
A plaquette of a latti e . . . . . . . . . . . . . . . . . . . . .
Des ription of ontinuum on gurations by latti es . . . . .
Oriented boundary of a plaquette . . . . . . . . . . . . . . .
Re tangular loop of the size R  T . . . . . . . . . . . . . .
Boundaries of plaquettes with opposite orientations . . . . .
Filling of a loop with plaquettes . . . . . . . . . . . . . . . .
Lines of for e between stati quarks . . . . . . . . . . . . .
Dependen e of the string tension on 1=g2 . . . . . . . . . .
Spe i energy for a rst-order phase transition . . . . . . .
Spe i heat for a se ond-order phase transition . . . . . . .
Graphi representation of the self- onsisten y ondition . .

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

98
100
103
104
104
106
107
114
119
120
122
125
128
131
133

vii

.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.

viii

LIST OF FIGURES

2.16
2.17
2.18
2.19
2.20
2.21
2.22
2.23
2.24
2.25
2.26
2.27
2.28
2.29
2.30
2.31
2.32

Solutions of the mean- eld self- onsisten y equation .


Latti e renormalization group transformation . . . . .
Monte Carlo data [Ake93 for  . . . . . . . . . . . .
Monte Carlo data [Cre79 for the string tension . . . .
Equipotential lines at di erent values of . . . . . . .
Monte Carlo data [BS92 for the intera tion potential
Altering signs of x on a latti e . . . . . . . . . . . . .
Momentum dependen e of G 1 for latti e fermions . .
Momentum dependen e of G 1 for latti e bosons . . .
A path made out of the string bits . . . . . . . . . . .
Dependen e of -meson mass on the quark mass . . .
Monte Carlo data [HP81 for quark ondensate . . . .
Polyakov loop . . . . . . . . . . . . . . . . . . . . . . .
T dependen e of energy density of hadron matter . . .
Pressure against T 4 for hadron matter . . . . . . . . .
Breaking of the ux tube . . . . . . . . . . . . . . . .
Expe ted phase diagram of hadron matter . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

134
138
141
146
147
148
154
156
157
161
163
164
172
177
179
180
181

3.1
3.2
3.3
3.4
3.5
3.6

One-loop diagrams for the four-vertex . . . . . . . . . . . . .


One-loop diagrams for the propagator of - eld . . . . . . . .
Bubble diagram whi h survives the large-N limit . . . . . . .
Diagrams of the 1=N -expansion . . . . . . . . . . . . . . . . .
1=N - orre tions to -propagator and three-vertex . . . . . . .
Double-line representation of a one-loop diagram for gluon
propagator . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Double-line representation of a four-loop diagram . . . . . . .
Double-line representation of a non-planar diagram . . . . . .
Cutting a planar graph into two graphs . . . . . . . . . . . .
Cutting a planar graph into trees and ar hes. . . . . . . . . .
Alternative utting of the same graph as in Fig. 3.10 . . . .
A tree graph and its dual . . . . . . . . . . . . . . . . . . . .
Re urren e relation for the number of ar hes . . . . . . . . .
Re urren e relation for the number of trees . . . . . . . . . .
Generi double-line index diagram . . . . . . . . . . . . . . .
Planar and non-planar parts of three-gluon vertex . . . . . .
Conne ted and dis onne ted planar graphs . . . . . . . . . .
Graphi derivation of Eq. (3.2.52) . . . . . . . . . . . . . . .
Diagrams for gluon propagator whi h involve quark loop . . .
Generi diagram in the index spa e . . . . . . . . . . . . . . .

191
192
192
193
196

3.7
3.8
3.9
3.10
3.11
3.12
3.13
3.14
3.15
3.16
3.17
3.18
3.19
3.20

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

218
220
221
222
223
224
225
225
226
228
229
230
232
233
235

LIST OF FIGURES

3.21 Diagrams with one quark loop in the Veneziano limit . . . . .


3.22 Diagram with two quark loops in the Veneziano limit . . . . .
3.23 Demonstration of the large-N fa torization in perturbation
theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3.24 Same as in Fig. 3.23 but for quark operators . . . . . . . . . .
3.25 Demonstration of the large-N fa torization in the strong oupling expansion . . . . . . . . . . . . . . . . . . . . . . . . . .
3.26 Contours in sum over paths representing observables . . . . .
3.27 Examples of smooth ontour and ontour with usp . . . . .
3.28 Contours Cyx and Cxy entering loop equation . . . . . . . . .
3.29 Contours C  p on the RHS of Eq. (3.3.63) . . . . . . . . .
3.30 Graphi representation of Eq. (3.3.64) . . . . . . . . . . . . .
3.31 Planar diagrams for W (C ) to order 2 . . . . . . . . . . . . .
3.32 Contours Cyx rxy and Cxy ryx in the regularized loop equation
3.33 Contour integral in the axial gauge . . . . . . . . . . . . . . .
3.34 Contours with one self-interse tion . . . . . . . . . . . . . . .
3.35 Three type of ontribution in Eq. (3.3.109) . . . . . . . . . .
3.36 Simplest diagram for the quen hed redu ed model . . . . . .

ix
238
240
242
244
247
255
258
261
270
271
272
277
282
283
284
298

LIST OF FIGURES

\Yossarian? What kind of a name is Yossarian?"


He had the fa ts at his nger tips. \It's
Yossarian's name," he explained.

, Cat h-22

J. Heller

Chapter 1

Path Integrals
The path integral is a method of quantization whi h is equivalent to the
operator formalism. It re overs the operator formalism in quantum me hani s
and perturbation theory in quantum eld theory (QFT).
The approa h based on path integrals has several advantages over the
operator formalism. It provides a useful tool for non-perturbative studies
in luding:





instantons,
analogy with statisti al me hani s,
numeri al methods.

A standard way of deriving the path integral is from the operator formalism:
operator formalism

()

path integral .

We shall pro eed in the opposite dire tion, following the original paper by
Feynman [Fey51.

CHAPTER 1.

PATH INTEGRALS

1.1 Operator al ulus


The operator al ulus developed by Feynman [Fey51 makes it possible to
represent fun tions of (non- ommuting) operators as path integrals, with the
integrand being the path-ordered exponential of operators whose order is
ontrolled by a parameter whi h varies along the traje tory. This pro edure
is termed as the Feynman disentangling. It is appli able also to fun tions of
matri es (say, -matri es whi h are asso iated with a spinor parti le). When
applied to the evolution operator, this pro edure results in the standard pathintegral representation of quantum me hani s.
We rst demonstrate in this Se tion the general te hnique using the simplest example: a free propagator in Eu lidean spa e, and then onsider the
path-integral representation of quantum me hani s, as well as propagators in
an external ele tromagneti eld.

1.1.

The delta-fun tion (1) (x0 y0 ) emerges when (=t)2 is applied to the operator of the T{produ t in (1.1.1).
Derive Eq. (1.1.2) in the operator formalism.
Let us apply the operator on the left-hand side (LHS) of Eq. (1.1.2) to
the T{produ t whi h is de ned by
Problem 1.1

Solution

' (x) ' (y) = (x0 y0 )' (x) ' (y) + (y0 x0 )' (y) ' (x)

with

(x0 y0 ) =

1 for x0  y0
0 for x0 < y0 :

(1.1.5)
(1.1.6)

Eq. (1.1.3) implies a nonvanishing result to emerge only when (=x0 )2 is applied
to the operator of the T{produ t. One gets



h0jT '_ (x) ' (y) j0i
x0
y0 ) h0j [' (y ) ; '_ (x) j0i = i (d) (x y ) ;

 2 m2 h0jT ' (x) ' (y ) j0i =

1.1.1 Free propagator


Let us rst onsider the simplest propagator of a free s alar eld whi h is given
in the operator formalism by the va uum expe tation value of the T{produ t1

h0jT ' (x) ' (y) j0i

G (x y ) =

(1.1.1)

with ' being the eld-operator.


The T{produ t (1.1.1) obeys the equation

2

OPERATOR CALCULUS


m2 G ( x

y) =

i (d) (x

y) ;

(1.1.3)

and anoni al equal-time ommutators


[' (t; ~x) ; '_ (t; ~y) = i (d
[' (t; ~x) ; ' (t; ~y) = 0:

1) (~
x

~y) ;

(1.1.4)

1 The ordered produ ts of operators were introdu ed by Dyson [Dys49. This paper and
other lassi al papers on quantum ele trodynami s are olle ted in the book edited by
S hwinger [S h58.

(1.1.7)

where the anoni al ommutation relations (1.1.4) are used.

The expli it solution to Eq. (1.1.2) for the free propagator is well-known
and is most simply given by the Fourier transform:

G (x y) =

(1.1.2)

where d = 4 is the dimension of spa e-time, however the formulas are appli able at any value of d. In the operator formalism, Eq. (1.1.2) is a onsequen e
of the free equations

 2 m2 ' (x) 0 = 0 ;



0  2 m2 ' (x) = 0

= (1) (x0

dd p ip(x y)
i
:
e
d
2
p m2 + i"
(2)

(1.1.8)

An extra i" (with " ! +0) in the denominator is due to the T{produ t
in the de nition (1.1.1) and unambiguously determines the integral over p0 .
The propagator (1.1.8) is known as the Feynman propagator whi h respe ts
ausality.
Problem 1.2 Perform the Fourier transformation of the free momentum-spa e
propagator in the energy p0 :
+1
Z
p
dp0 e ip0 (t t0 )
i
G! (t t0 ) =
;
! = p~ 2 + m2 : (1.1.9)
2
2
2
p0 ! + i"
1
Solution

The poles of the momentum-spa e propagator are at

p0 =

 !  i" :

(1.1.10)

CHAPTER 1.

PATH INTEGRALS

For t > t0 (t < t0 ), the ontour of integration an be losed in the upper (lower)
half-plane whi h gives
i!(t t0 )
i!(t t0 )
G! (t t0 ) = (t t0 ) e
+ (t0 t) e
2!
2!
0
e i!jt t j
=
:
2!
The Green fun tion (1.1.11) obeys the equation


2
t2

!2



G! t t0 = i (1) t t0

(1.1.11)

(1.1.12)

In mathemati al language, the Green fun tion G (x y) is termed as the


resolvent of the operator on the LHS of Eq. (1.1.2), and is often denoted as
the matrix element of the inverse operator

G (x y) =

2



x :
2
m

(1.1.13)

The operators a t in an in nite-dimensional Hilbert spa e whose elements


in Dira 's notation [Dir58 are the bra and ket ve tors: hgj and jf i, respe tively. The oordinate representation emerges when these ve tors are hosen
to be the eigenstates of the position operator x :

x jxi = x jxi :

(1.1.14)

These basis ve tors obey the ompleteness relation


Z

dd x jxihxj

= 1;

= g(x) ;

hx j f i

(1.1.15)
= f (x) :

dd x f (x)jxi ;

hgj

dd y g(y)hyj :

(1.1.18)

Using the expansion (1.1.17), one gets


=

dd x

dd y g(y) O(y; x) f (x) :

(1.1.19)

Sin e the kernel of the unit operator is the delta-fun tion,

hy j1j xi

= hyjxi = (d) (x y) ;

(1.1.20)

the formula

hy jOj xi

O (d) (x y)

(1.1.21)

an also be written down as a dire t onsequen e of Eq. (1.1.20), where the


operator O on the right-hand side (RHS) a ts on the variable x.
Therefore, when the operator a ts on a fun tion f (x), the result is expressed via the kernel by the standard formula
(Of )(y)  hy jOj f i =

dd x O(y; x) f (x) :

(1.1.22)

Eq. (1.1.21) is obviously reprodu ed when f is substituted by a delta-fun tion,


while Eq. (1.1.19) takes the form

hg jOj f i

dd x g(x) (Of ) (x) :

(1.1.23)

O is approximated by a matrix with hy jO j xi being its elements.

1.1.2 Eu lidean formulation


(1.1.16)

These wave fun tions appear in the expansions

jf i

= O(y; x) :

If spa e-time is approximated by a dis rete set of points, then the operator

while the wave fun tions, asso iated with hgj and jf i, are given by

hg j xi

The a tion of a linear operator O on the bra and ket ve tors in the Hilbert
spa e is determined by its matrix element hy jOj xi, whi h is also known as
the kernel of the operator O and is denoted by

hg jOj f i

Remark on operator notations

OPERATOR CALCULUS

hy jOj xi

and therefore oin ides with the ausal Green fun tion for a harmoni os illator
with frequen y ! .



y

1.1.

(1.1.17)

Eq. (1.1.8) an be alternatively obtained by inverting the operator on the


LHS of Eq. (1.1.2). Before doing that, it is onvenient to make an analyti
ontinuation in the time-variable t, and to pass to the Eu lidean formulation
of QFT where one substitutes

t =

i x4 :

(1.1.24)

CHAPTER 1.

PATH INTEGRALS

The four-momentum operator in Minkowski spa e reads

pM


i M



i ; i
t ~x

Minkowski spa e ;

(1.1.25)

1.1.

We shall use the same notation v for a four-ve tor in Minkowski and
Eu lidean spa es:
 = (v ; ~v )
vM
0

vE = (~v; v4 )

while its Eu lidean ounterpart is given by

pE =

i E



; i
~x x4

Eu lidean spa e :

(1.1.26)

These two formulas together with Eq. (1.1.24) yield

 p0

i p4

(1.1.27)

for the relation between energy and the fourth omponent of the Eu lidean
four-momentum.
The passage to Eu lidean spa e results in hanging the Minkowski signature of the metri g to the Eu lidean one:2
)

(+ + + +)

Minkowski signature

Eu lidean signature

(+

(1.1.28)

As su h, one gets

p2M = p20 p~ 2

p2E = p~ 2 p24 :

(1.1.29)

The exponent in the Fourier transformation hanges analogously:

p x = Et + p~~x

! pExE = p~~x + p4 x4 :

(1.1.30)

This reprodu es the standard Fourier transformation in Eu lidean spa e

f (p) =
f (x) =

Z
Z

OPERATOR CALCULUS

dd p ipx
e f (p) :
(2)d

(1.1.31)

2 An older generation is familiar with Eu lidean notations whi h are used throughout
the book by Akhiezer and Berestetskii [AB69. On the ontrary, the two other anoni al
books on quantum eld theory by Bogoliubov and Shirkov [BS76 and by Bjorken and
Drell [BD65 use Minkowskian notations whi h are due to Feynman.

Eu lidean spa e ;

(1.1.32)

iv4 :

(1.1.33)

with

v0 =

The only di eren e resides in the metri . We do not distinguish between


upper and lower indi es in Eu lidean spa e.
Using Eqs. (1.1.24) and (1.1.26), we see that in Eu lidean spa e Eq. (1.1.2)
takes the form


 2 + m2 G (x y) = (d) (x y)

(1.1.34)

with the positive sign in front of m2 .


The passage to the Eu lidean formulation is justi ed in perturbation theory where it is asso iated with the Wi k rotation. The dire tion in whi h
the rotation is performed is unambiguously pres ribed by the +i" term in
Eq. (1.1.8), and is depi ted in Fig. 1.1. The variable t = x0 rotates through
=2 while E = p0 rotates through =2.
Figure 1.1a explains the sign in Eq. (1.1.24). Figure 1.1b and Eq. (1.1.27)
implies that the integration over p4 goes in the opposite dire tion, so that
+1
+1
Z
Z
dp4
dp0
::: = i
::: :
(1.1.35)
2
2
1
1
Thus when passing in to Eu lidean variables, Eq. (1.1.8) be omes

G (x y ) =

dd x e ipx f (x) ;

Minkowski spa e ;

dd p ip(y x) 1
e
:
p2 + m2
(2)d

(1.1.36)

Note that the RHS of Eq. (1.1.36) is nothing but the Fourier transform of the
free momentum-spa e Eu lidean propagator, and there is no need to retain an
i" in the denominator sin e the integration pres ription is now unambiguous.
It is now lear why we keep the same notation for the oordinate-spa e
Green fun tions: the Feynman propagator in Minkowski spa e and the Eu lidean propagator. They are the same analyti fun tion of the time-variable.

CHAPTER 1.

t = i

q
z

Minkowski
spa e

6E =

tm

XXnEu lidean
spa e
}|

PATH INTEGRALS

H

Eu lidean
spa e

q 

XX

ip4 Em

q 

H

a)

}|

{q

Minkowski
spa e

b)

Fig. 1.1: Dire tion of the Wi k rotation from Minkowski to Eu lidean spa e (indi ated by the arrows) for a) time and b) energy. The dots represent singularities of a free propagator in a) oordinate and b) momentum spa es.
The ontours of integration in Minkowski spa e are asso iated with ausal
Green fun tions. They an obviously be deformed in the dire tions of the
arrows.

Repeat the al ulation of Problem 1.2 in Eu lidean spa e.


A ording to Eq. (1.1.36) we need to al ulate

Problem 1.3
Solution

G! (

 0) =

+1
Z

dp4 ip4 ( 0  ) 1
e
:
2
p24 + ! 2

(1.1.37)

The integral on the RHS an be al ulated for  >  0 ( <  0 ) by losing the ontour
in the lower (upper) half-plane, and taking the residues at p4 = i! (p4 = i! ),
respe tively. This yields

G! (

!( 0  )

 0) e
2!
0
e !j  j :
2!

 0 ) = (

+ ( 0

) e

!(  0 )

2!
(1.1.38)

The Eu lidean Green fun tion (1.1.38) an obviously be obtained from the
Minkowskian one, Eq. (1.1.11), by the substitution

 = it ;

 0 = it0

(1.1.39)

1.1.

OPERATOR CALCULUS

and vise versa. G! (




 0 ) obeys the equation




 2 + ! 2 G   0  = (1)   0 
(1.1.40)
!
 2
and, therefore, is the Green fun tion for a Eu lidean harmoni os illator with frequen y ! .

As we shall see in a moment, the Eu lidean formulation makes path


integrals well-de ned, and allows non-perturbative investigations analogous
to statisti al me hani s to be arried out. There are no reasons, however,
why Minkowski and Eu lidean formulations should always be equivalent nonperturbatively.

Remark on Eu lidean -matri es


The -matri es in Minkowski spa e satisfy

f M ; M g = 2 g I :

(1.1.41)
Therefore, 0 is Hermitean while the Minkowskian spatial -matri es are antiHermitean.
Analogously, the Eu lidean -matri es satisfy
f  ;  g = 2  I ;
(1.1.42)

so that all of them are Hermitean. We ompose them from 2  2 matri es as




I 0
4 = 0 =
(1.1.43)
0 I
and


0
i~

~ =
;
(1.1.44)
i~ 0
where ~ are the usual Pauli matri es. Noti e that the Eu lidean spatial matri es di er from the Minkowskian ones by a fa tor of i.
The free Dira equation in Eu lidean spa e reads


^ + m
= 0;
^ =  
(1.1.45)
or

(ipb + m)

with p given by Eq. (1.1.26).

= 0

(1.1.46)

10

CHAPTER 1.

PATH INTEGRALS

1.1.3 Path-ordering of operators


There are no problems in de ning a fun tion of an operator A, say via the
Taylor series. For instan e,
1 1
X
eA =
An :
(1.1.47)
n
!
n=0
However, it is more ompli ated to de ne a fun tion of several non ommuting operators (or matri es), e.g. A and B having
[A; B =
6 0;

(1.1.48)

sin e the order of operators is now essential. In parti ular, one has

6 eA eB ;
e A+B =

(1.1.49)

so that the law of addition of exponents fail. Certainly, the exponential on


the LHS is a well-de ned fun tion of A + B , but sin e A and B are intermixed
in the Taylor expansion, this expansion is of little use in pra ti e. We would
like to have an expression where all B 's are written, say, to the right of all
A's.
This an be a hieved by the following formal tri k [Fey51.
Let us write


M
1
= lim
= lim 1 + (A + B )
M !1
M !1
M





1
1
1
1 + (A + B ) 1 + (A + B )    1 + (A + B ) :
M
M
M
{z
} (1.1.50)
|
M times

e A+B

The stru ture of the produ t on the RHS prompts us to introdu e an index
i running from 1 to M and repla e (A + B ) in ea h multiplier by (Ai + Bi ).
Therefore, one writes
e A+B =

lim

M !1


M 
Y
i=1

1+

1
= lim 1 + (AM + BM )
M !1
M



where the index i ontrols the order of the operators whi h are all treated
di erent. The ordering is su h that the larger i is, the later the operator
with the index i a ts. This order of operators is pres ribed by quantum me hani s, where initial and nal states are represented by ket and bra ve tors,
respe tively.
The RHS of Eq. (1.1.51) an be rewritten as
1 PM

e A+B = P lim e M i=1 (Ai +Bi )


(1.1.52)
M !1
where the symbol P stands for the ordering operation. There is no ambiguity
on the RHS of Eq. (1.1.52) on erning ordering Ai and Bi with the same
index i, sin e su h terms are O(1=M 2 ) and are negligible as M ! 1.
To des ribe the ontinuum limit as M ! 1, one introdu es the ontinuum
variable  = i=M whi h belongs to the interval [0; 1. The ontinuum limit
of Eq. (1.1.52) reads
e A+B = P e

1
1 + (A1 + B1 ) ; (1.1.51)
M

R1

d(A()+B ())

;
(1.1.53)
where A(i=M ) = Ai and B (i=M ) = Bi while the operator A() + B () a ts
at order .
Eq. (1.1.53) is in fa t obvious sin e it only involves the operator A + B
whi h ommutes with itself. For ommuting operators there is no need for
ordering so that A() + B () does not depend on  in this ase. The integral
in the exponent on the RHS of Eq. (1.1.53) an then be done, and reprodu es
the LHS.
Eq. (1.1.53) an however be manipulated as though A() and B () were
just fun tions rather than operators sin e the order would be automati ally
spe i ed by the path-ordering operation. This is analogous to the well-known
fa t that operators an be written in an arbitrary order under the T{produ t.
Therefore, we an rewrite Eq. (1.1.53) as
e A+B = P e

R1

R
d0 A(0 ) 01 dB ()

:
This is the operator analog of the law of addition of exponents.
Problem 1.4

Solution

11

OPERATOR CALCULUS

in B .

1
(A + Bi )
M i

1.1.

(1.1.54)

Cal ulate expli itly the rst term of the expansion of exp (A + B )

Expanding the RHS of Eq. (1.1.54) in B , one gets


e A+B

eA +

Z1
0

R
R1
0 0
d0 A(0 )

B ( ) e 0 d A( ) + : : :
d e

(1.1.55)

12

CHAPTER 1.

PATH INTEGRALS

There is no need for a path-ordering sign in this formula, sin e the order of the
operators A and B is written expli itly. There is also no ambiguity in de ning the
exponentials of the operator A as already explained.
Sin e the order is expli it, one drops the formal dependen e of A and B on the
ordering parameter whi h gives
e A+B = e A +

Z1

d e (1 )A B e A + : : :

1.1.

z (t)

6
z ( )

(1.1.56)

The formulas (1.1.55) and (1.1.56) are known from time-dependent perturbation
theory in quantum me hani s.

1.1.4 Feynman disentangling


The operator on the LHS of Eq. (1.1.34) an be inverted as follows:
1
(d) (x y)
G (x y ) =
2
 + m2
Z1
2
2
1
1
=
d e 2  ( m ) (d) (x y)
2
0
Z1
R
1  dt 2 (t) (d)
1 m2 
1
2
2
(x y) ; (1.1.57)
Pe 0
=
d e
2
0

where we have formally labelled the derivatives by an ordering parameter


t 2 [0;  , whi h is an analog of  from the previous subse tion. This is the
general pro edure whi h the Feynman disentangling is built on.
Sin e the operators  and  ommute in the free ase, we ould manage
without introdu ing the t-dependen e, however the operators do not ommute
in general. The simple example of the non-relativisti Hamiltonian and the
propagator in an external ele tromagneti eld are onsidered later on in this
Se tion. Other ases where the disentangling is needed are related to inverting
an operator whi h is in addition a matrix in some symmetry spa e.
Continuing with the disentangling, the RHS of Eq. (1.1.57) an be rewritten as
Z1
Z
R
2
1 
1 m2 
1
2
d e
G (x y) =
Dz (t) e 2 0 dt z_ (t)
2
0

Pe

R
0

z (0)=x
dt z_ (t) (t) (d)

(x y) ;

13

OPERATOR CALCULUS

(1.1.58)

Fig. 1.2: The traje tory z (t). The operator  (t) a ts at the order t.
where the integration runs over all traje tories z (t) whi h begin at the point
x, as depi ted in Fig. 1.2.
Sin e the operator  (t) a ts at the order t, these operators are ordered
along the traje tory z (t) with P, in Eq. (1.1.58), standing for the pathordering operator. Note, that z_ (t) and  (t) ommute sin e
d
 (t)z_ (t) =  = 0
(1.1.59)
dt
so that their order is inessential in Eq. (1.1.58). With these rules of manipulations, Eq. (1.1.58) an be proven by the \translation"

z (t)

z 0 (t) = z (t) +


Zt

dt0  (t0 )

(1.1.60)

of the integration variable z (t) in the Gaussian integral.


The integral over the fun tions z (t) in Eq. (1.1.58) is alled a path integral
or a fun tional integral.

14

CHAPTER 1.

1.1.

15

OPERATOR CALCULUS

general operator D (t) is given by

z1

PATH INTEGRALS

LL
 L


LL

L !!z3
L!z2

Pe

((( z6

 z5

R

Pe

M
Y

= lim
!0 i=1 [1 + (zi

zi 1 ) D (i) :

(1.1.63)

R
0

dt z_  (t) (t)

= exp (z  ( ) x )


;
x

(1.1.64)

whi h is nothing but the shift operator. Applying the operator on the RHS
of Eq. (1.1.64) to the delta-fun tion, one gets

-

Pe

Fig. 1.3: Dis retization of the traje tory z (t) (depi ted for M = 6).
The ontinual path integral an be approximated by a nite one. To this
aim, let us hoose M points ti = i, where  is the step of dis retization, and
M = =. We then onne t the points

z0 = x ; zi = z (i) i = 1; 2; : : : ; M

(1.1.61)

by straight lines. Su h a dis retization of the traje tory z (t) is depi ted in
Fig. 1.3. The measure in Eq. (1.1.58) an then be dis retized by

Dz (t) : : : =

dt z_  (t)D (t)

The order of multipli ation here is the same as in Eq. (1.1.51).


The expli it form of the operator  is essential when we al ulate how
it a ts on the delta-fun tion as is pres ribed by the RHS of Eq. (1.1.58). For
the free ase, when the t-dependen e of  (t) is inessential, one simply gets

z4

M Z
Y
i=1

dd zi
::: :
(2)d=2

(1.1.62)

The expli it form of the operator  in Eq. (1.1.34) was not essential
in deriving Eq. (1.1.58). If  in Eq. (1.1.34) is repla ed by an arbitrary
operator D with non- ommuting omponents, then Eq. (1.1.58) holds with
 (t) substituted by D (t). The dis retized path-ordered exponential of a

R
0

dt z_  (t) (t) (d)


(x

y) = (d) (z ( ) y) :

(1.1.65)

Therefore, z ( ) has to oin ide with y due to the delta-fun tion, whi h
disappears after the integration over z ( ) has been performed. Thus the
nal answer is
Z
Z1
R
1  dt z_ 2 (t)
1 m2
1
 : (1.1.66)
2
2
0
d e
Dz (t) e
G (x y ) =
2
0

z (0)=x
z ( )=y

This path integral goes over all traje tories z (t) whi h onne t the initial
point x and the nal point y .
Problem 1.5 Derive Eqs. (1.1.58) and (1.1.66) by introdu ing a path integral over
velo ity v (t) = z_ (t).
Solution The operator on the RHS of Eq. (1.1.57) an be disentangled by the
following Gaussian path integral

e2

R
0

dtD2 (t)

Dv (t) e

1  dtv2 (t)

2 0

R
0

dtv (t)D (t)

(1.1.67)

This formula holds for an arbitrary operator D and an be proven by al ulating


the Gaussian integral after shifting v (t).

16

CHAPTER 1.

PATH INTEGRALS

Substituting D (t) =  (t) and al ulating the a tion of the path-ordered exponential on (d) (x y ), we get
Z
Z1
R
2
1 
1 2
G(x y ) = 1 d e 2 m Dv (t) e 2 0 dtv (t)
2
0

 (d) (x +

Z 
0

dt v (t) y ):

the measure

dt v 2 (t)

!

0
Z

Dv (t) : : : =

M Z
Y
i=1

M
X
i=1

vi2 ;

dd vi : : :
(2=)d=2

(1.1.70)

(1.1.71)

obviously re overs Eq. (1.1.62) after al ulating the Ja obian from the variables vi
to the variables zi . Therefore, Eq. (1.1.68) reprodu es Eq. (1.1.66) provided

z (t) = x +

Zt

dt0 v (t0 ) :

(1.1.72)

The dis retized traje tory in Fig. 1.3 an be analyti ally written as the expansion
M
X
i=1

zi fi (t) + x (1 t=) ( t) ;

where the basis fun tions


8
< 1 + (t= i)
fi (t) = 1 (t= i)
:
0

for t 2 [(i 1); i ;


for t 2 [i; (i + 1) ;
otherwise

are nonvanishing only for the i-th and (i+1)-th intervals. The measure (1.1.62)
is de ned, therefore, via the oe ients zi as a multiple produ t of dzi 's.
While the basis fun tions fi (t) are not orthogonal:
1


Z

dt fi (t)fj (t) =

1
1
2
ij + i(j+1) + i(j
3
6
6

(1.1.73)

(1.1.74)

1) ;

(1.1.75)

the orthogonal set appears in the expansion of the velo ity

z_ (t) =
where

M
X
i=1

(zi

zi 1 )i (t) ;

(1.1.76)

1=
for t 2 [(i 1); i ; :
(1.1.77)
0
otherwise
This shows why the dis retized velo ities from Problem 1.5 are natural variables.
One an hoose, instead, another set of (orthogonal) basis fun tions and
expand

i (t) =

z  (t) =

M
X
n=1

n n (t)

(1.1.78)

with some oe ients n . Then the measure (1.1.62) takes the form

Dz  (t) : : :

Remark on de nition of the measure

z  (t) =

17

OPERATOR CALCULUS

(1.1.68)

The integration over Dv (t) in this formula has no restri tions.
To derive Eq. (1.1.66) from Eq. (1.1.68), let us note that the dis retized velo ities
read
z  zi 1
vi = i
:
(1.1.69)

Sin e
Z

1.1.

M
Y
n=1

dd n : : :

(1.1.79)

modulo a -independent Ja obian. Mathemati ally, this implies that one


approximates the fun tional spa e by M -dimensional spa es.

1.1.5 Cal ulation of the Gaussian path integral


The Gaussian path integral (1.1.66) an be easily al ulated by the following
tri k (see, e.g., the book by Feynman [Fey72, Chapter 3). Let us substitute
the variable z (t) by a new variable  (t) whi h are related by the formula

z (t) =

y x
t +  (t) + x:


(1.1.80)

18

CHAPTER 1.

PATH INTEGRALS

The boundary onditions for the variable  (t) are determined by Eq. (1.1.80)
to be

 (0) =  ( ) = 0:

1.1.

Cal ulate F ( ) from the dis retized path integral.


The dis retized version of Eq. (1.1.66) is

Problem 1.6
Solution

(1.1.81)

dt z_ 2 (t)

(y x)2
(y x)
=
+2
( ( )  (0)) +



Z

dt _2 (t) : (1.1.82)

The se ond term on the RHS vanishes due to the boundary onditions
(1.1.81) so that the propagator be omes
1
G (x y) =
2

:

Z1

1 m2
2

d e

(y x)2
2

D e

1  dt _2 (t)
2 0
:

(1.1.83)

(0)=( )=0

D e

1 
2 0

dt _2 (t)

= F ( ) :

(1.1.84)

This expression is to be ompared with the proper-time representation of


the Eu lidean free propagator whi h reads
Z

(1.1.85)

These two expressions oin ide provided that

F ( )

1
:
(2 )d=2

Z M
Y1

i=1

dd zi e
(2)d=2

1
2

PM
2
i=1 (zi zi 1 ) =

(1.1.87)

where z0 = x and zM = y . The integral an be al ulated by well-known formula


for the Gaussian integral
Z

dd z e (x2z1)2 (z2y2)2 =  1 2 d=2 e 2((x1 +y)22 ) : (1.1.88)


1 +  2
(2 )d=2
After applying this formula M 1 times, one arrives at Eq. (1.1.86). Noti e that 
an els in the nal answer.
What traje tories are essential in the path integral?
It is seen from the dis retization on the RHS of Eq. (1.1.87) that only
traje tories with
Solution

(0)=( )=0

Z1
2
 2
dd p ip(x y) 1
d e 2 (p +m )
e
G (x y) =
d
2
(2)
0
Z1
(x y)2
2
1
1
1
d e 2 m e 2
=
:
2
(2 )d=2

1
=
(2)d=2

1  dt z_ 2 (t)

2 0

Problem 1.7

The path integral over  on the RHS of Eq. (1.1.83) is a fun tion solely of
Z

Dz (t) e

z (0)=x
z ( )=y

Substituting Eq. (1.1.80) into the exponent in Eq. (1.1.66), one gets
Z

19

OPERATOR CALCULUS

(1.1.86)

jzi zi 1 j  p

(1.1.89)

are essential as  ! 0. Su h traje tories are typi al Brownian traje tories. They
are ontinuous as  ! 0 but not smooth (jzi zi 1 j   for smooth traje tories).
In mathemati al language, these fun tions are said to belong to the Lipshitz lass
1/2.

Remark on mathemati al stru ture


The measure (1.1.62) for integration over fun tions is sometimes alled the
Lebesgue measure. It was introdu ed in mathemati s by Wiener [Wie23 in
onne tion with the problem of Brownian motion. With the Gaussian fa tor
in orporated, it is also known as the Wiener measure while the proper path
integral is known as the Wiener integral3. The measure (1.1.62) is de ned
on the spa e L2 (i.e. the spa e ofR fun tions whose square is integrable, in the
sense of the Lebesgue integral, dtz 2 (t) < 1). The integration on L2 goes
over traje tories z (t), whi hR are generi ally dis ontinuous. However, the
extra weight fa tor exp f 21 0 z_ 2(t)g restri ts the traje tories in the above
path integrals to be ontinuous.
3 See,

e.g., the books [Ka 59, S h81, Wie86 for a des ription of the path-integral approa h to Brownian motion.

20

CHAPTER 1.

PATH INTEGRALS

1.1.6 Transition amplitudes


As is well-known in quantum me hani s, G(x y) is the probability for a
(s alar) parti le to propagate from x to y. A onvenient notation for a traje tory z (t) that onne ts x and y is
yx

 fz (t); 0  t  ; z (0) = x ; z ( ) = y g :

d
dt

! (t) ;

 0;

(1.1.91)

with  being a new parameter.


A onvenient parametrization is via the proper length of yx whi h is
given by

s =

ds

(1.1.92)

z_ 2()d

(1.1.93)

yx

where

ds =

and  2 [i ; f is some parametrization. For the obvious reasons the


parametrization
1
s
(1.1.94)
m
with s given by Eq. (1.1.92) is alled the proper time parametrization. Noti e
that the dimension of t is [length2 a ording to Eq. (1.1.94).
Let us denote4

t =

S [ yx

Z

2
 m2  + 12 dt z_ 2 (t) :

(1.1.95)

4 The notation S [ with square bra kets means that S is a fun tional of

with parentheses stands for fun tions.

while f (x)

21

OPERATOR CALCULUS

The sense of this notation is that the RHS oin ides with the lassi al a tion of
a relativisti free (s alar) parti le in the proper-time parametrization (1.1.94)
when
Z

(1.1.90)

Noti e that yx stands for a traje tory as a geometri obje t, while z (t)
is a fun tion whi h des ribes a given traje tory in some parametrization t.
This fun tion (but not the geometri obje t itself) depends on the hoi e of
parametrization and hanges under the reparametrization transformation

1.1.

dtz_ 2(t) = m

Z

ds = m Length[

(1.1.96)

sin e


dz (s)
ds

2

= 1

(1.1.97)

and m = Length[ by the de nition of the proper time.


Therefore, the path-integral representation (1.1.66) is nothing but the sum
over traje tories with the weight being an exponential of (minus) lassi al
a tion:

G(x y) =

X
yx

e S[

yx :

(1.1.98)

This sum is split in Eq. (1.1.66) into the traje tories along whi h the parti le
propagates during the proper time  and the integral over  .
Equation (1.1.98) implies that the transition amplitude in quantum me hani s is a sum over all paths whi h onne ts x and y. In other words, a
parti le propagates from x to y along all paths yx in luding the ones whi h
are forbidden by the free lassi al equation of motion

z (t) = 0:

(1.1.99)

Only the lassi al traje tory (1.1.99) survives the path integral in the
lassi al limit h ! 0. The reason for this is that if the dependen e on the
Plan k's onstant is restored, it appears in the exponent:

G(x y) =

X
yx

e S[

yx =h :

(1.1.100)

As h ! 0 the path integral is dominated by a saddle point whi h is given in


the free ase by the lassi al equation of motion (1.1.99).
It is worth noting that the sum-over-path representation (1.1.98) is written
entirely in terms of traje tories as geometri obje ts and does not refer to a

22

CHAPTER 1.

PATH INTEGRALS

on rete parametrization. For the free theory S [ is proportional to the


length of the traje tory :

Sfree[ = m Length[ ;

(1.1.101)

where the length is given for some parametrization  of the traje tory
Zf

Length[ =

d z_ 2 ():

by

(1.1.102)

i

The sum-over-path representation (1.1.98) with S [ given by the lassi al


a tion (Eq. (1.1.101) in the free ase) is often onsidered as a rst prin iple
of onstru ting quantum me hani s given the lassi al a tion S [ .
Problem 1.8

Represent the matrix element of the (Eu lidean) evolution operator

hy jexp f H  gj xi for the non-relativisti Hamiltonian


2

H = 2m + V (x)

H  xE

= Dv (t) e

m
2

R
0

dtv2 (t)

R
0

dtv (t) (t)

R
0

dtV (x;t) (d)


(x

y ): (1.1.104)

Here the argument t in V (x; t) is just the ordering parameter, while the same formula
holds when the potential is expli itly time dependent.
In ontrast to Eq. (1.1.67), we have put the minus sign in front of the linearin-v term in the exponent in Eq. (1.1.104), so that it agrees with Appendix B
of Feynman's paper [Fey51. In fa t it does not matter what sign is used sin e
the integral over v (t) is Gaussian so that only even powers of v survive after the
integration.
The path-ordered exponential in Eq. (1.1.104) reads expli itly
P

R
0

dt v (t) (t)

R
0

dtV (x;t)

= lim
!0

M h
Y
i=1


1 vi 
x

whi h an be rewritten as
P

R
0

dtv (t) (t)

R
0

dtV (x;t)

= lim
!0

M h
Y

1 vi

i=1

 i [1 V (x; i) ; (1.1.106)


x

if terms whi h vanish as  ! 0 are negle ted, or equivalently as


P

R
0

dtv (t) (t)

R
0

dtV (x;t)

 h
Y
t=0

1 dtv  (t)

 i [1 dt V (x; t) : (1.1.107)
x

There is no need to write down the t-dependen e of  (t) in these formulas sin e
the order of the operators is expli it.
To disentangle the operator expression (1.1.107), let us note that
[1 dtv  (t) = U 1 (t + dt)U (t)
with

U (t) = exp

Z t
0

(1.1.108)

dt0 v  (t0 )

(1.1.109)

being the shift operator. It obviously obeys the di erential equation

d U (t) = v  (t) U (t) :



dt
Now sin e

U (t) [1 dt V (x; t) U 1 (t) =

(1.1.110)

1 dt V x +

Z t
0

dt0 v  (t0 ); t



; (1.1.111)

the RHS of Eq. (1.1.107) an be written in the form


 h
Y
t=0

1 dt v  (t)

= U 1 ( )

 i [1 dt V (x; t)
x


Y
t=0

1 dt V x +

= U 1 ( ) exp 4

V (x; i) ; (1.1.105)

23

OPERATOR CALCULUS

(1.1.103)

as a path integral.
Solution
The al ulation is similar to the one whi h is already done in Subse t. 1.1.4. It is most onvenient to use a path-integral over velo ity whi h was
onsidered in Problem 1.5. The proper disentangling formula reads
D
y e
Z

1.1.

Z
0

whi h is ompletely disentangled.

Z t

dt V x +

dt0 v  (t0 ); t

Z t
0



3

dt0 v  (t0 ); t 5;

(1.1.112)

24

CHAPTER 1.

PATH INTEGRALS

The operator U 1 ( ) is now in the proper order to be applied to the variable y


in the argument of the delta-fun tion, whi h results in the shift


(d) (x y ) =) (d) x +

Z 
0

dt v  (t) y :

H  xE =

Dz (t) e

R
0

dt L(t)

(1.1.114)

z (0)=x
z ( )=y

L(t) = m2 z_2 (t) + V (z(t))

(1.1.115)

is the Lagrangian asso iated with the Hamiltonian H . The unusual plus-sign in
this formula is due to Eu lidean-spa e formalism. It is lear from the derivation
that Eq. (1.1.114) holds for time-dependent potentials as well.
Noti e that the path integral in Eq. (1.1.114) is now over traje tories whi h the
parti le propagates in the xed proper time  with no integration over  .
A spe ial omment about the operator U 1 ( ) in Eq. (1.1.112) is needed. In the
S hrodinger representation of quantum me hani s, one is interested in the matrix
elements on the evolution operator between some ve tors hg j and jf i in the Hilbert
spa e. A ording to Eq. (1.1.23) one has in the oordinate representation
D
g e

H  f E =

dd x g (x) e H  f (x) :

(1.1.116)

Integrating by parts, the operator U 1 ( ) an then be applied to g (x) whi h results


in the shift


g (x) =) U ( ) g (x) U 1 ( ) = g x +

Z 
0

dt v  (t) :

(1.1.117)

Passing to the variable (1.1.72), Eq. (1.1.116) be omes


D
g e

H  f E =

Problem 1.9

potential V (x):

Cal ulate the diagonal resolvent of the S hrodinger operator in the

R! (x; x; V ) =

Dz (t) e

R
0

dt L(t)

g (z ( ))f (z (0)) : (1.1.118)

E
1

2
2
G + ! + V x

(1.1.119)

z (0)=x
z ( )=x

As G ! 0 this path integral is dominated by the t-independent saddle-point


traje tory

z (t) = x ;

(1.1.121)

whi h is asso iated with a parti le standing at the point x. Substituting V at this
saddle-point, i.e. repla ing V (z (t)) by V (x), and al ulating the Gaussian integral
over quantum u tuations around the traje tory (1.1.121) by the use of Eqs. (1.1.84)
and (1.1.86), one gets

R! (x; x; V ) =

1
p
2 ! 2 + V (x)

(1.1.122)

in d = 1.
Equation (1.1.122) an be alternatively derived by applying the Gelfand{Di key
te hnique [GD75 whi h says that R! (x; x; V ) obeys the third order linear di erential equation
1 1 3
2 2 G

V (x) V (x) R! (x; x; V ) = ! 2 R! (x; x; V ) :

(1.1.123)

R! (x; x; V ) given by Eq. (1.1.122) obviously satis es this equation as G ! 0.


One more way to derive Eq. (1.1.122) is to perform a semi lassi al WKB expansion of R! (x; y ; V ) in the parameter G . This is explained in Chapter 7 of the
book [LL74.
Derive Eq. (1.1.123).
The resolvent

Problem 1.10
Solution

There are no restri tions on the initial and nal points of the traje tories z (t) in
this formula.

D
x

in the limit G ! 0 for d = 1.


Solution Using the formula of the type (1.1.114), we represent R! (x; x; V ) as the
path integral
Z1
Z
R
R
2
1 
1 2
1
R! (x; x; V ) =
d e 2 !
Dz (t) e 2G 0 dt z_ (t) 0 dtV (z(t)) : (1.1.120)
2
0

where

25

OPERATOR CALCULUS

(1.1.113)

This will be explained in more detail in the next paragraphs.


Passing to the variable (1.1.72), we get nally
D
y e

1.1.

R! (x; y ; V ) =

D
y

E
1

G  2 + !2 + V x

(1.1.124)

26

CHAPTER 1.

obeys the equations




2
G x
2

PATH INTEGRALS

+ ! 2 + V (x) R! (x; y ; V ) = (1) (x y ) ;




 2 + ! 2 + V (y ) R (x; y ; V ) = (1) (x y ) :
G y
!
2

(1.1.125)

It an be expressed via the two solutions f (x) of the homogeneous equation




 + ! 2 + V (x) f (x) = 0 ;
G x

2

(1.1.126)

where f+ or f are regular at +1 or 1, respe tively. Then the full solution is


f (x)f (y )(x y ) + f (x)f+ (y )(y x)
(1.1.127)
R! (x; y ; V ) = +
G W!
with
W! = f+ (x)f 0 (x) f 0 (x)f (x)
(1.1.128)

1.1.

as is pres ribed by Eqs. (1.1.101) and (1.1.102). In ontrast to (1.1.66),


this path integral is not easy to al ulate. The integration over Dz () in
Eq. (1.1.131) involves integration over the reparametrization group whi h
gives the proper group-volume fa tor sin e the exponent is parametri invariant. Eq. (1.1.66) is re overed after xing parametrization to be proper time.
How to perform this al ulation is explained in Chapter 9 of the book by
Polyakov [Pol87.
If one makes a naive dis retization of the parameter  by equidistant
intervals, the exponent in Eq. (1.1.131) is highly nonlinear in the variables zi ,
leading to ompli ated integrals. On the ontrary the dis retization (1.1.87)
of the path integral (1.1.66), where the parametri invarian e is xed, results
in a Gaussian integral whi h is easily al ulable.
Problem 1.11

sure by

R! (x; x; V ) = f+ (x)f (x)


(1.1.129)
G W!
using Eq. (1.1.126), in order to verify that it satis es the nonlinear di erential
equation



2G R! R00 + G R0 2 + 4 ! 2 + V R2 = 1 :
(1.1.130)
!

One more di erentiation of Eq. (1.1.130) with respe t to x results in Eq. (1.1.123).
It is worth noting that Eq. (1.1.130) is very onvenient for al ulating the semi lassi al expansion of R! (x; x; V ) in G . In parti ular, the leading order (1.1.122) is
obvious.

Remark on parametri invariant representation


The Green fun tion G(x y) an be alternatively al ulated from the parametri invariant representation

G(x y)

Dz () e

z (i )=x
z (f )=y

R
m0 f d z_ 2 ()
i

(1.1.131)

Cal ulate the path integral in Eq. (1.1.131), dis retizing the mea-

Dz

being the Wronskian of these solutions. Applying =x to Eq. (1.1.128), it is easy
to show that W! is an x-independent fun tion of ! .
The simplest way to prove Eq. (1.1.123) is to di erentiate

27

OPERATOR CALCULUS

1 Y
M
X

dd zi
(2)d=2
M =1 i=1

(1.1.132)

and applying the entral limit theorem as M ! 1.


Solution By making the dis retization, we represent the RHS of Eq. (1.1.131) as
the probability integral
1 Z MY1 d
X
1
d zi
G (x y ) =
(2)d=2 M =1 i=1 (2)d=2
 (x ! z1 )  (z1 ! z2 )     (zM 1 ! y) (1.1.133)
with

 (zi 1 ! zi ) = e m0 jzi zi 1 j
(1.1.134)
being an (unnormalized) probability fun tion and  is a parameter of the dimension
of [length2 . The probability interpretation of ea h term in the sum is standard for
random walk models, and means, as usual, that a parti le propagates via independent intermediate steps. The dis retization of the measure given by Eq. (1.1.133)
looks like that in Eq. (1.1.62), but the summation over M is now added.
Sin e the integral in Eq. (1.1.133) is a onvolution, the entral limit theorem
states that


M

1 X
0
G (x y ) =
(2)d=2 M (2m20 )d=2
 21 d=2 e m20 (x y)2 =(22 M )+O(M
(2 M )

2)

(1.1.135)

28

CHAPTER 1.

PATH INTEGRALS

at large M , where 0 and  2 are the zeroth and (normalized) se ond moments of :

0 =
2

dd x e jxj = 2 d=2

1
=
0

G (x y )
(1.1.136)

The sum over M in Eq. (1.1.135) is onvergent for


1=d

p 0

(1.1.137)

2

and is divergent for m0 < m . Choosing m0 > m , but


 1 in the limit
" ! 0, the sum over M will be onvergent, while dominated by terms with large
1
M  m2  :
(1.1.138)

This is easily seen by rewriting Eq. (1.1.135) as

m20

G (x y ) =

e
Ea h term with M

X

m2
2 2 M

m2

d=2

M
dM ln (mo =m ) m20 (x y)2 =(22 M )+O(M 2 ) :

 m2 ontributes O(1) to the sum, so that


G (x y )  m2 :

(1.1.139)
(1.1.140)

This justi es the using of the entral limit theorem in this ase. The typi al
distan es between the zi 's, whi h are essential in the integral on the RHS of
Eq. (1.1.133), are
jzi zi 1 j  m10  p
(1.1.141)
like in Eq. (1.1.89). The relation (1.1.138) between the essential values of M and 
is also similar to what we had in Subse t. 1.1.4.
The sum over M in Eq. (1.1.139) an be repla ed by a ontinuous integral over
the variable
2 M
 =
;
(1.1.142)
m2
whi h is O(1) for M

 m2 . Introdu ing also the variable m by


m2

d m2 m 2  > 0 ;

2 0

(1.1.143)

29

OPERATOR CALCULUS

we rewrite Eq. (1.1.139) as

(d)
;
(d=2)

dd x x2 e jxj = d(d + 1) :

m0 > m =

1.1.

m2
2

!0

Z1

d

1
e
(2 )d=2

1 2 (x y)2
2m 
2

(1.1.144)

whose RHS is proportional to that in Eq. (1.1.85) for the Eu lidean propagator.

Remark on dis retized path-ordered exponential


As is dis ussed in Subse t. 1.1.3, the order of operators Ai and Bi with
the same index i is not essential in the path-ordered exponential (1.1.52) as
M ! 1. If Eq. (1.1.52) is promoted to be valid at nite M (or at least to the
order of O(1=M )), this spe i es the ommutator of Ai and Bi . Analogously,
a dis retization of Eq. (1.1.114) spe i es in whi h order the produ t of xi pi in
the lassi al theory should be substituted by the operators xi and pi in the
operator formalism. See details in the books by Berezin [Ber86 (Chapter 1
of the Part II) and Sakita [Sak85 (Chapter 6).

1.1.7 Propagators in external eld


Let us now onsider a (quantum) parti le in a lassi al ele tromagneti eld.
The standard way of introdu ing an external ele tromagneti eld is to substitute the (operator of the) four-momentum p by

p ! p eA (x) :

(1.1.145)

Re alling the de nition (1.1.26) of the Eu lidean four-momentum, 


needs to be repla ed by the ovariant derivative

! r =  ieA (x) :
(1.1.146)
Inverting the operator r2 by the use of the disentangling pro edure, one


gets

G (x; y; A) 
Z1
1
=
d e
2
0



y
Z

1 m2
2


1

2
2
r + m x

Dz (t) e

z(0)=x
z( )=y

1 
2 0

R
dt z_ 2 (t)+ie 0 dt z_  (t)A (z(t))

: (1.1.147)

30

CHAPTER 1.

PATH INTEGRALS

Noti e that the exponent is just the lassi al (Eu lidean) a tion of a parti le
in an external ele tromagneti eld. Therefore, this expression is again of the
type in Eq. (1.1.98).
The path-integral representation (1.1.147) for the propagator of a s alar
parti le in an external ele tromagneti eld is due to Feynman [Fey50 (Appendix A).
Derive Eq. (1.1.147) using Eq. (1.1.67) with D = r .
The al ulation is analogous to that of Problem 1.8. We have

Problem 1.12
Solution

D (t)

r (t) =  (t) + ieA (x; t)

(1.1.148)

so that expli itly


P

e
=

R
0

dt v (t)r (t)

 h
Y
t=0

dtv  (t)

 h
Y
t=0

1 dt v  (t)

 + ie dt v  (t)A (x; t)i



x

 i [1 + ie dt v  (t)A (x; t) :

x

(1.1.149)

ie v  (t)A (x; t) :

(1.1.150)

Substituting this potential in Eq. (1.1.114) and remembering the additional integration over  , we obtain the path-integral representation (1.1.147).

We an alternatively rewrite Eq. (1.1.147) as

G (x; y; A) =

X0 ie
yx

yx

dz A (z)

(1.1.151)

where, pro eeding as in Subse t. 1.1.6, we have in luded the free a tion in
the de nition of the sum over traje tories:
X0
X
(1.1.152)
e Sfree [ yx ;
=
yx

yx

and rewritten the (parametri invariant) integral over dt as the ontour integral over

dz  = dt z_  (t)
along the traje tory yx.

(1.1.153)

31

OPERATOR CALCULUS

The meaning of Eq. (1.1.151) is that the transition amplitude of a quantum


parti le in a lassi al ele tromagneti eld is the sum over paths of the Abelian
phase fa tor

U [ yx = e

ie

yx

dz A (z)

(1.1.154)

Under the gauge transformation

A (z )

! A (z ) + 1e  (z ) ;

g.t.

(1.1.155)

the Abelian phase fa tor transforms as

U [ yx g.t.
! e i (y) U [ yx e i (x) :

(1.1.156)

Noting that a wave fun tion at the point x is transformed under the gauge
transformation (1.1.155) as

'(x)

This looks exa tly like the expression (1.1.107) with

V (x; t) =

1.1.

g.t.

e i (x) '(x);

(1.1.157)
we on lude that the phase fa tor is transformed as the produ t '(y)'y (x):

U [ yx g.t.
 \'(y)'y (x)" ;

(1.1.158)

where \: : :" means literally that \transforms as : : :".


As a onsequen e of Eqs. (1.1.156) and (1.1.157), a wave fun tion at the
point x transforms like one at the point y after multipli ation by the phase
fa tor:

U [ yx '(x)

g.t.

\'(y)" ;

(1.1.159)

'y (y) U [ yx

g.t.

\'y (x)" :

(1.1.160)

and analogously

Equations (1.1.159) and (1.1.160) show that the phase fa tor plays the
role of a parallel transporter in an ele tromagneti eld, and that in order to
ompare phases of a wave fun tion at points x and y, one should rst make a
parallel transport along some ontour yx . The result is, generally speaking,
dependent ex ept when A (z ) is a pure gauge. The su ient and ne essary
ondition for the phase fa tor to be independent is the vanishing of the eld

32

CHAPTER 1.

PATH INTEGRALS

strength, F (z ), whi h is a onsequen e of the Stokes theorem when applied


to the Abelian phase fa tor5.
Below we shall deal with determinants of various operators. Analogous to
Eq. (1.1.147), one gets
Z1
2
1
1 d
2
ln det r =
Sp e 2  r
2 
0
Z 1
Z
R
H

2
1 
d
1
Dz (t) e 2 0 dt z_ (t)+ie dz A (z) ; (1.1.161)
=
2 0 

1.1.

Show how the path-integral representation (1.1.147) re overs for


G (x; y ; A) the diagrammati expansion of propagator in an external eld A :

Problem 1.14

G(x; y ; A) =
Solution

ln det D = Sp ln D ;

(1.1.162)

whi h relates the determinant and the tra e of a Hermitean operator (or a
matrix) D.
Prove Eq. (1.1.162).
Let us rst redu e D to a diagonal form by a unitary transformation
and denote eigenvalues as Di . Then Eq. (1.1.162) reads
Problem 1.13

Solution

ln

Di =

ln Di

(1.1.163)

whi h is obviously true.

The phase fa tor for a losed ontour enters Eq. (1.1.161). It des ribes
a parallel transportation along a losed loop, and is gauge invariant as a
onsequen e of Eq. (1.1.156):
H



g.t.
e ie dz A (z) ! e ie dz A (z) :

(1.1.164)

This quantity, whi h plays a ru ial role in modern formulations of gauge


theories, will be dis ussed in more detail in Se t. 2.1.
5 Stri tly speaking, this statement holds for the ase when an be hosen everywhere
in spa e-time, i.e. whi h is simply onne ted. However, there exist situations when an
not penetrate into some regions of spa e like for the Aharonov{Bohm experiment whi h is
dis ussed below in Subse t. 2.1.4.

Use the formula


X0 Z

z(0)=z( )

where the path integral goes over traje tories whi h are losed due to the
boundary ondition z (0) = z ( ). To derive Eq. (1.1.161), we have used the
formula

33

OPERATOR CALCULUS

yx yx

d

(d) (

where

$
 =

z) : : : =

X0
yz

$ X0

::: ;
z
zx

!
 +  :

+ :::
y (1.1.165)

(1.1.166)

(1.1.167)

The proof is given in Appendix A of the paper [MM81.


Establish the equivalen e of the path-integral representation
(1.1.161) of ln det r2 and the sum of one-loop diagrams in an external eld A :

Problem 1.15

ln det r2 =









1
2







(1.1.168)

+ ::: ;

with the ombinatori fa tor 1=2 in the third diagram being asso iated with a
symmetry fa tor.
Solution The same as in the previous problem.

Remark on analogy with statisti al me hani s


A formula of the type (1.1.161), whi h represents the tra e of an operator via
path integral over losed traje tories, is known as the Feynman{Ka formula.
The terminology omes from statisti al me hani s where the partition fun tion (or equivalently the statisti al sum) is given by the Boltzmann formula

Z = Sp e H

(1.1.169)

(with being the inverse temperature) whose path-integral representation is


of the type in Eq. (1.1.161). The expression whi h is integrated on the RHS of
Eq. (1.1.161) over d= is asso iated, in the statisti al-me hani al language,

34

CHAPTER 1.

PATH INTEGRALS

with the partition fun tion of a losed elasti string, whose energy is proportional to its length, and intera ts with an external ele tromagneti eld. This
shows an analogy between Eu lidean quantum me hani s in d-dimensions and
statisti al me hani s in d (spatial)- and 1 (temporal)-dimensions whose timedependen e disappears, sin e nothing depends on time at equilibrium. We
shall explain this analogy in more detail in Chapter 2 (Se t. 2.5) when dis ussing quantum eld theory at nite temperature.

1.2.

35

SECOND QUANTIZATION

1.2 Se ond quantization


In the previous Se tion we have onsidered rst quantization of parti les,
where the operators of oordinate and momentum, x and p respe tively, are
represented in the oordinate spa e by

x = x ; p = i x
= rst quantization :
(1.2.1)
In the language of path integrals, rst quantization is asso iated with integrals
over traje tories in the oordinate spa e.
While propagators an be easily represented as path integrals, it is very
di ult to des ribe, in this language, a (non-geometri ) self-intera tion of
a parti le, sin e this would orrespond to extra weights for self-interse ting
paths. For the free ase (or a parti le in an external gauge eld) there are no
su h extra weights, and the transition amplitude is ompletely des ribed by
the lassi al a tion of the parti le.
In the operator formalism, self-intera tions of a parti le is des ribed by
the use of se ond quantization | this is where the transition from quantum
me hani s to quantum eld theory begins. Se ond quantization is a quantization of elds, and is asso iated with path integrals over elds whi h is the
subje t of this Se tion. We demonstrate how perturbation theory and the
S hwinger{Dyson equations an be derived by using path integrals.

1.2.1 Integration over elds

Let us de ne the following (Eu lidean) partition fun tion 6

Z =

D' (x) e S

where the a tion in the exponent is given for the free ase by
Z


1
Sfree[' =
dd x ( ')2 + m2 '2 :
2
The measure D'(x) is de ned analogously to Eq. (1.1.62):
+1
Z
Y Z
D' (x) : : : =
d' (x) : : :
x 1

(1.2.2)

(1.2.3)

(1.2.4)

6 It is also alled the statisti al sum in an analogy with statisti al me hani s (see Sub-

se t. 1.1.7).

36

CHAPTER 1.

PATH INTEGRALS

where the produ t runs over all spa e points x and the integral over d' is the
Lebesgue one.
The propagator is given by the average

h ' (x) ' (y) i

G (x; y) =

h F [' i

= Z

D' (x) e S['F [' :

h1i = 1:

In Minkowski spa e, perturbation theory is well-de ned sin e the Gaussian


path integral, whi h determines the propagator
Z

(1.2.6)

(1.2.7)



1
x ;
2
2
 +m

h 'x 'y i

Z
Solution

(1.2.8)

d'x exp

1
2

x;y

'x Dxy 'y 'x 'y = Dxy1 :

(1.2.9)

'x

1=2 

xy 'y :


1
y
i



x :

(1.2.12)

f y ; x g

= 0;
y ; x = 0 ;

y ; x = 0 :


(1.2.13)

As a onsequen e, the square of a Grassmann variable vanishes


2
x

= x2 = 0 :

(1.2.14)

The path integral over the Fermi elds equals

Cal ulate the Gaussian integral by the hange of variable

'0x =

(1.2.11)

Path integrals over anti ommuting Grassmann variables are used to des ribe
fermioni systems.
The Grassmann variables x and y obey the anti ommutation relations

By dis retizing the (Eu lidean) spa e, derive


Z Y

d
D' e i d x 'D' ;

Here D is a proper operator in Minkowski spa e.


It an not be said a priory whether a non-perturbative formulation of
a given (intera ting) theory via the path integral in Minkowski spa e exists
sin e the weight fa tor is omplex and the integral may be divergent.

whi h oin ides with (1.1.57). Therefore, we have obtained the same propagator (1.1.85) as in the previous Se tion.
Problem 1.16

D' e iS =

1.2.2 Grassmann variables

Sin e the free a tion (1.2.3) is Gaussian, the average (1.2.5) equals


y

Remark on Minkowski-spa e formulation

equals

The notation is obvious sin e we average on the RHS of Eq. (1.2.6) over all
eld on gurations with the same weight as in the partition fun tion (1.2.2).
The normalization fa tor Z 1 provides the ne essary property of an average

G (x y) =

37

SECOND QUANTIZATION

(1.2.5)

where a generi average is de ned by the formula


Z

1.2.

(1.2.10)

Note that the integrals over '(x) are onvergent in Eu lidean spa e. If
a dis retization of spa e is introdu ed, the path integrals in Eqs. (1.2.2) or
(1.2.6) are rigorously de ned.

D D e

R d
d x

= det D

(1.2.15)

while an analogous integral over the Bose elds is


Z

D'y D' e

R d y
d x'

D' = (det D) 1 :

(1.2.16)

38

CHAPTER 1.

PATH INTEGRALS

1.2.

De ne integrals over Grassmann variables.


Assuming that and  belong to the same Grassmann algebra, the
Berezin integrals are de ned by
Z
Z

d x=

d x y=
d x y =

Z
Z

d x y = xy ;
d x y = 0 :

(1.2.17)

1.2.3 Perturbation theory


The ubi self-intera tion of the s alar eld is des ribed by the a tion

d x d x e

x x

= 1:

(1.2.18)

The average over the Fermi elds, de ned with the same weight as in
Eq. (1.2.15), equals

(x)  (y) =

1 x

(1.2.19)

whi h is the same as for bosons.


Noti e that the fermioni partition fun tion (1.2.15) an be rewritten a ording to Eq. (1.1.162) as
det D = e Sp ln D :

(1.2.20)

= e

Sp ln D :

S [' =

dd x

1
1

( ')2 + m2 '2 + '3 ;
2
2
3!

(1.2.21)

The relative minus sign in the exponents on the RHS's of Eqs. (1.2.20)
and (1.2.21) is a famous minus sign whi h emerges for losed fermioni loops
whi h ontribute to the (logarithm of the) partition fun tion.

(1.2.22)

where  is the oupling onstant.


To onstru t perturbation theory, we expand the exponential in  and
al ulate the Gaussian averages with the free a tion (1.2.3).
To order 2 , the expansion is

h ' (x)' (y) i =Z h ' (x) ' (y) ifreeZ





d
3
d
3
 1 3! d x1 ' (x1 ) 3! d x2 ' (x2 )
+


' (x)
3!


dd x1 '3 (x1 )
3!

free 

dd x2 '3 (x2 ) ' (y)

free

+ : : : (1.2.23)

The term whi h is linear in  (as well as the Gaussian average of any odd
power of ') vanishes due to the re e tion symmetry ' ! ' of the Gaussian
a tion. The last term on the RHS is depi ted graphi ally in Fig. 1.4.
Further al ulation of the RHS of Eq. (1.2.23) is based on the free average

h ' (xi ) ' (xj ) ifree

The analogous formula for bosons (1.2.16) is rewritten as


(det D)

Fig. 1.4: Diagrammati representation of the last term on the RHS of Eq. (1.2.23).

The formula (1.2.15) an now easily be derived: representing hy jD j xi in the diagonal form, expanding the exponential up to a term whi h is linear in all the Grassmann variables and al ulating the integrals of this term a ording to Eq. (1.2.17).
See more details in the book by Berezin [Ber86 (x3 of Part I).

The simplest interesting integral is


Z

d x = 0 ;

SS
S
 x2






S
x1 S
S

Problem 1.17
Solution

39

SECOND QUANTIZATION

= G (xi

xj )

(1.2.24)

and rules of the Wi k pairing of Gaussian averages, whi h allow us to represent


the average of a produ t as the sum of all possible produ ts of pair averages.
Some of the diagrams whi h emerge after the Wi k ontra tion are depi ted
in Fig. 1.5. These diagrams are alled Feynman diagrams.
The diagram of Fig. 1.5b is dis onne ted. Its dis onne ted part with
the two loops an els with the same ontribution from Z 1 (whi h yields the

40

CHAPTER 1.

PATH INTEGRALS

x1

x2

Sin e '(x) is arbitrary, Eq. (1.2.26) results in the following quantum


equation of motion

x2


41

SECOND QUANTIZATION

S [' w.s.

= h
;
' (x)
' (x)

x1

a)

1.2.

b)

Fig. 1.5: Some of the Feynman diagrams whi h appear from (1.2.23) after the Wi k
pairing.

(1.2.27)

where we have expli itly written the dependen e on the Plan k's onstant h.
It appears this way sin e the a tion S is divided by h in Eq. (1.2.2) when h
is restored.
Equation (1.2.27) is to be understood in the weak sense, i.e. it is valid
under the sign of averaging when applied to a fun tional F ['. In other
words the variation of the a tion on the LHS of Eq. (1.2.27) an always be
substituted by the variational derivative on the RHS when integrated over
elds with the same weight as in Eq. (1.2.6). Therefore, one arrives at the
following fun tional equation

fa tor in the rst term on the RHS of Eq. (1.2.23)). This is a general property
that only onne ted diagrams are left in h ' (xi ) ' (xj ) i.
Let us note nally that the ombinatori fa tor 1=2 is orre tly reprodu ed
for the diagram of Fig. 1.5a. For an arbitrary diagram, this pro edure of
pairing reprodu es the usual ombinatori fa tor whi h is equal to the number
of automorphisms of the diagram (i.e. the symmetries of a given graph).

This equation is quite similar to that whi h S hwinger onsidered in the


framework of his variational te hnique.

1.2.4 S hwinger{Dyson equations

1.2.5 Commutator terms

Feynman diagrams an be alternatively derived by iterations in the oupling


onstant of the set of the S hwinger{Dyson equations whi h is a quantum
analogue of the lassi al equation of motion.
To derive the S hwinger{Dyson equations, let us utilize the fa t that the
measure (1.2.4) is invariant under an arbitrary shift of the eld

In order to show how Eq. (1.2.28) reprodu es Eq. (1.1.34) for the free propagator, let us hoose

' (x)

! ' (x) + ' (x) :

(1.2.25)

This invarian e is obvious sin e the fun tional integration goes over all the
elds while the shift (1.2.25) is just a transformation from one eld on guration to another.
Sin e the measure is invariant, the path integral in the average (1.2.6)
does not hange under the shift (1.2.25):
Z

dd x ' (x)

D' e S['

F ['
S ['
F [' +
'(x)
'(x)

= 0:

(1.2.26)

S ['
F ['
' (x)

F ['
:
= h
' (x)

F [' = ' (y) :

(1.2.28)

(1.2.29)

Substituting into Eq. (1.2.28) and al ulating the variational derivative, one
gets


 2 + m2 h' (x) ' (y)i = h

' (y)
' (x)

= h (d) (x y)

(1.2.30)

whi h oin ides with Eq. (1.1.34).


The LHS of Eq. (1.2.30) emerges from the variation of the free lassi al
a tion (1.2.3)

Sfree
=
' (x)

 2 + m2 ' (x)

(1.2.31)

42

CHAPTER 1.

PATH INTEGRALS

1.2.

while the RHS, whi h results from the variational derivative, emerges in the
operator formalism from the anoni al ommutation relations
(x0 y0 ) [' (x0 ; ~x) ; '_ (y0 ; ~y) = i (d) (x y)
(1.2.32)
as is explained in Subse t. 1.1.1.
For this reason, the RHS of Eq. (1.2.30) and, more generally, the RHS of
Eq. (1.2.28) is alled the ommutator term. The variational derivative on the
RHS of Eq. (1.2.27) plays the role of the onjugate momentum in the operator
formalism. The al ulation of this variational derivative in Eu lidean spa e is
equivalent to di erentiating the T{produ t and using anoni al ommutation
relations in Minkowski spa e.
When the Plan k's onstant vanishes, h ! 0, the RHS of Eq. (1.2.27)
(or Eq. (1.2.28)) vanishes. Therefore it redu es to the lassi al equation of
motion for the eld ':
S ['
= 0:
(1.2.33)
' (x)
This implies that the path integral over elds has a saddle point as h ! 0
whi h is given by Eq. (1.2.33).
Another lesson we have learned is that the average (1.2.5), whi h is de ned via the Eu lidean path integral, is asso iated with the Wi k rotated
T{produ t. We have already seen this property in the previous Se tion in the
language of rst quantization. More generally, the Eu lidean average (1.2.6) is
asso iated with the va uum expe tation value of h0 j TF [' j 0i in Minkowski
spa e.

and

1.2.6 S hwinger{Dyson equations ( ontinued)


The set of the S hwinger{Dyson equations for an intera ting theory an be
derived analogously to the free ase.
Let us onsider the ubi intera tion whi h is des ribed by the a tion
(1.2.22). Choosing again F [' to be given by Eq. (1.2.29) and al ulating the
variation of the a tion (1.2.22), one gets




 2 + m2 h ' (x) ' (y) i + '2 (x) ' (y) = (d) (x y) : (1.2.34)
2
Rederive Eq. (1.2.34) by analyzing Feynman diagrams.
Let us introdu e the Fourier-transformed 2- and 3-point Green fun tions

G3 (p; q; p q ) =

G(p) =

dd x e ipx h '(x)'(0) i ;

(1.2.35)

dd xdd y e ipx iqy h '(x)'(y )'(0) i :

(1.2.36)

Let us also denote the free momentum-spa e propagator as


G0 (p) = 2 1 2 :
p +m
The perturbative expansion of G3 starts from

G3 (p; q; p q ) =

(1.2.37)

G0 (p)G0 (q )G0 (p + q ) + : : : :

(1.2.38)

It is standard to trun ate three external legs, introdu ing the vertex fun tion
(p; q; p q ) = G3 (p; q; p q )G 1 (p)G 1 (q )G 1 (p + q )

(1.2.39)

whose perturbative expansion starts from :

 3

(p; q; p q ) =

dd k G (k p)G (k)G (k + q ) + : : : : (1.2.40)


0
0
(2 )d 0

This expansion an be represented diagrammati ally as

T
""bb = ""bb + "" Tbb +
s

:::

(1.2.41)

where the lled ir le on the LHS represents the exa t vertex and the thin lines are
asso iated with the bare propagator (1.2.37).
An analogous expansion of the propagator is


+


Problem 1.18
Solution

43

SECOND QUANTIZATION















+


+ ::: ;



(1.2.42)

44

CHAPTER 1.

PATH INTEGRALS

where the bold line represents the exa t propagator. It is ommonly rewritten as
an equation for the self-energy G0 1 (p) G 1 (p) whi h involves only the 1-parti le
irredu ible (1PI) diagrams. The last diagram on the RHS of Eq. (1.2.42) is not 1PI.
Resumming the diagrams a ording to the de nition (1.2.41) of the exa t vertex and the exa t propagator G, the propagator equation an be represented
graphi ally as

G0 (p) G0 (p)G

1 (p)G




0 (p)

(1.2.43)

s



where the bold lines represent the exa t propagator G while the external (thin) ones
are asso iated with the bare propagator G0 . One vertex on the RHS of Eq. (1.2.43)
is exa t and the other is bare. It does not matter whi h one is exa t and whi h one
is bare sin e we an olle t the diagrams of Eq. (1.2.42) into the exa t vertex either
on the left or on the right. Eq. (1.2.43) an be written analyti ally as
Z
dd q

G(q ) ( q; p; q p)G(p q ) : (1.2.44)
G0 1 (p) G 1 (p) =
2 (2 )d
Multiplying Eq. (1.2.44) by G(p) and using the de nition (1.2.36), we obtain
exa tly the Fourier transform of Eq. (1.2.34).

Noti e that Eq. (1.2.34) is not losed. It relates the 2-point average (propagator) to the 3-point average (whi h is asso iated with a vertex). The losed
set of the S hwinger{Dyson equations an be obtained for the n-point averages
Gn (x1 ; x2 ; : : : ; xn ) = h ' (x1 ) ' (x2 ) : : : ' (xn ) i :
(1.2.45)
They are also alled the orrelators, in analogy with statisti al me hani s, or the n-point Green fun tions, in analogy with the Green fun tions in
Minkowski spa e.
Choosing
F [' = ' (x2 ) : : : ' (xn )
(1.2.46)
and al ulating the variational derivative, one gets from Eq. (1.2.28) the following hain of equations

 2 + m2 h ' (x) ' (x2 ) : : : ' (xn ) i



+ '2 (x) ' (x2 ) : : : ' (xn )


2
=

n
X
j =2

(d) (x xj ) ' (x2 ) : : : ' (xj ) : : : ' (xn )

(1.2.47)

1.2.

45

SECOND QUANTIZATION

where ' (xj ) means that the orresponding term ' (xj ) is missing in the
produ t. Using the notation (1.2.45), Eq. (1.2.47) an be rewritten as


 2 + m2 Gn (x; x2 ; : : : ; xn ) + Gn+1 (x; x; x2 ; : : : ; xn )
2

n
X
j =2

(d) (x xj ) Gn 2 (x2 ; : : : ; xj ; : : : ; xn )

(1.2.48)

with xj again standing for the missing argument.

Remark on onne ted orrelators


The n-point orrelators (1.2.45) in lude both onne ted and dis onne ted
parts. The presen e of dis onne ted parts is most easy to see in the free ase
when all onne ted parts disappear while Gn for an even n is given by the
Wi k pairing as is dis ussed in Subse t. 1.2.3. .
The orrelators an be alternatively de ned by introdu ing the partition
fun tion whi h is a fun tional of an external sour e J (x):

Z [J =

d
D' (x) e S+ d xJ (x)'(x);

(1.2.49)

and varying with respe t to the sour e. The n-point fun tions (1.2.45) are
then given by

Gn (x1 ; : : : ; xn ) =

   Z [J
Z [J J (x1 ) J (xn )

(1.2.50)

while the onne ted parts are given by

h ' (x1 ) : : : ' (xn ) i onn

   ln Z [J :
J (x1 ) J (xn )

(1.2.51)

The point is that

W [J = ln Z [J

(1.2.52)

involves only a set of onne ted diagrams while dis onne ted ones emerge
in Z [J after the exponentiation. We already tou hed this property in Subse t. 1.2.3 to order 2 . The fun tional W [J is alled, for this reason, the
generating fun tional for onne ted diagrams.

46

CHAPTER 1.

PATH INTEGRALS

Remark on the LSZ redu tion formula


The (analyti ally ontinued to Minkowski spa e) orrelators Gn (x1 ; : : : ; xn )
determine an amplitude of the pro ess when n, generally speaking, virtual
parti les produ es k on-mass-shell parti les. Let us denote this amplitude
as An!k (q1 ; : : : ; qn ; p1 ; : : : ; pk ) where q1 ; : : : ; qn and p1 ; : : : ; pk are the fourmomenta of the in oming and outgoing parti les, respe tively. The fourmomentum onservation says

q1 + : : : + q n = p 1 + : : : + p k :

(1.2.53)

The Lehman{Symanzik{Zimmerman (LSZ) redu tion formula reads

An!k (q1 ; : : : ; qn ; p1 ; : : : ; pk )
=

n
Y

k
Y

Z
dd pi nY1 dd qj
d
d
j =1 (2 )
i=1 (2 )

(1.2.54)

The unusual sign of the square of the parti le mass m is due to the Eu lidean
metri .
Equation (1.2.54) makes sense for timelike pi , when p2i < 0, while qj2 is
arbitrary. The amplitude for the ase of on-mass-shell in oming parti les is
given by Eq. (1.2.54) with qj2 ! m2 .

1.2.7 Regularization
The ultraviolet divergen ies (i.e. those at small distan es or large momenta)
are an intrinsi property of QFT whi h makes it di erent from the quantum
me hani s of a nite number of degrees of freedom. The divergen ies emerge,
roughly speaking, be ause of the delta-fun tion in the anoni al ommutation
relations.
The idea of regularization is to somehow smooth the e e t of the deltafun tion. The usual pro edures of regularization are to

1) smear the delta-fun tion by


{ point splitting,
{ S hwinger proper-time regularization,
{ latti izing;

47

SECOND QUANTIZATION

2) add negative-norm regulating term to the a tion


{ Pauli{Villars regularization;
3) introdu e higher derivatives in the kineti term7 ;
4) hange the dimension 4 ! 4 ;
5) regularize the measure in the path integral.
As an example of point splitting, let us onsider the regularization when
the delta-fun tion in the ommutator term is repla ed by

(d)(x y) Reg.
=) R (d)(x y) = R(x; y) :

(1.2.55)

R = e a2 ( 2

(1.2.56)

The regularizing operator R is for instan e

Z k
 Y

lim p2 + m2
2 ! m2 i
p
i
i=1
j =1
P
P
i ik=1 pi xi +i nj=11 qj xk+j
e
Gn (x1 ; : : : ; xn+k 1 ; 0)

qj2 + m2

1.2.

m2 ) ;

where the parameter a with the dimension of length plays the role of an
ultraviolet uto . The uto disappears as a ! 0 when

R ! 1;
R (x; y) ! (d) (x y) :

(1.2.57)

It is easy to al ulate how the regularization (1.2.55) modi es the propagator. The result reads
1
R (d) (x y)
GR (x y) =
 2 + m2
Z1
(x y)2
2
1
1
1
:
(1.2.58)
=
d e 2 m e 2
2
(2 )d=2
2
a

The lower limit in the integral over the proper time  is now a2 rather than
0 as in the non-regularized expression (1.1.85). This parti ular way of point
splitting oin ides with the S hwinger proper-time regularization.
A regularization via point splitting an be done non-perturbatively while
the dimensional regularization, whi h is listed in item 4, is de ned only in the
framework of perturbation theory. The regularization of the measure, whi h
is listed in item 5, will be onsidered in the next Se tion.
When a regularization is introdu ed, some of the rst prin iples ( alled
the axioms), on whi h quantum eld theory is built on, are violated. For
instan e, the regularization via the point splitting (1.2.55) violates lo ality.
7 That is in the quadrati -in- elds part of the a tion.

48

CHAPTER 1.

PATH INTEGRALS

1.3 Quantum anomalies from path integral


As is well-known, the Lagrangian approa h in lassi al eld theory is very
useful for onstru ting onserved urrents asso iated with symmetries of the
Lagrangian. Noether's theorems8 say how to onstru t orresponding urrents and when they are onserved.
An analogous approa h in quantum eld theory is based on path integrals
over elds. It naturally in orporates the lassi al results sin e the weight in
the path integral is given by the lassi al a tion.
However, anomalous (i.e. additional to lassi al) terms in the divergen ies
of urrents an appear in the quantum ase due to a ontribution from regulators whi h make the theory nite in the ultraviolet limit. They are alled
quantum anomalies.
In this Se tion we rst onsider the hiral anomaly, i.e. the quantum
anomaly in the divergen e of the axial urrent, whi h appears in the pathintegral approa h as a result of the non-invarian e of the regularized measure.
Then we brie y repeat the analysis for the s ale anomaly, i.e. the quantum
anomaly in the divergen e of the dilatation urrent.

1.3.1 QED via path integral


Let us restri t ourselves to the ase of quantum ele trodynami s (QED),
though most of the formulas will be valid for a non-Abelian Yang{Mills theory
as well.
QED is des ribed by the following partition fun tion

Z =

DA

D D e S[A; ; 

(1.3.1)

where A is the ve tor-potential of the ele tromagneti eld, i and i are


the Grassmann variables whi h des ribe the ele tron-positron eld with i
being the spinor index. They are independent but are inter hangeable under
involution
inv. 
! ;
(1.3.2)
whi h is de ned su h that9
inv.
(1.3.3)
1 2 ! 2 1 :
8 See, e.g., x2 of the book [BS76.
9 See the book by Berezin [Ber86 (x3.5 of Part I). Sometimes involution is de ned with an

opposite sign (i.e.  is substituted by i ) whi h results in a multipli ation of the fermioni
part of the a tion (1.3.4) by an extra fa tor of i.

1.3.

49

QUANTUM ANOMALIES FROM PATH INTEGRAL

In parti ular,  is invariant under involution. Therefore,  is an analogue


of i  = i y 0 in the operator formalism while involution is analogous to
Hermitean onjugation.
The Eu lidean QED a tion in Eq. (1.3.1) reads

S [A; ;  =

1 2
dd x   r + m  + F
;
4

(1.3.4)

where

r

= 

ieA (x)

(1.3.5)

is the ovariant derivative as before,

F =  A

 A

(1.3.6)

is the eld strength, and  are the Eu lidean -matri es whi h are dis ussed
in Subse t. 1.1.2.

1.3.2 Chiral Ward identity


Let us perform the lo al hiral transformation
(x) .t.
! 0 (x) =
 (x) .t.
! 0 (x) =

e i (x) 5 (x) ;
 (x) e i (x) 5 :

(1.3.7)

Here the parameter of the transformation (x) is a fun tion of x and 5 is


the Hermitean matrix
in d = 4. Note that both
Minkowski spa e
 = y 0 ;

5 = 1 2 3 4
(1.3.8)
and  have the same transformation law sin e in
5y = 5 ; 0y = 0

(1.3.9)

while 5 and 0 anti ommute.


The variation of the lassi al a tion (1.3.4) under the hiral transformation (1.3.7) reads

S =

dd x  (x) J5 (x) + 2im (x) (x) 5 (x) ;

(1.3.10)

50

CHAPTER 1.

PATH INTEGRALS

where the axial urrent

J5 = i   5
(1.3.11)
is the Noether urrent asso iated with the hiral transformation.
It follows from Eq. (1.3.10) that the divergen e of the axial urrent (1.3.11)
reads
 J5 = 2im  5 ;
(1.3.12)
so that it is onserved in the massless ase (m = 0) at the lassi al level:

 J5 m==0
Problem 1.19

0:

(1.3.13)

Verify Eq. (1.3.12) using the lassi al Dira equation




rb + m (x) = 0 ;

rb =  r :

(1.3.14)

1.3.

where sp is the tra e only over the -matrix indi es i and j . The RHS vanishes
naively sin e the tra e vanishes. A subtlety with the appearan e of the in nite
fa tor (d) (0) will be dis ussed in the next Subse tion.
Note that the in nitesimal version the transformation (1.3.7) is a parti ular ase of the more general one
(x) ! 0 (x) = (x) + (x) ;
 (x) ! 0 (x) =  (x) +  (x) ;
(1.3.19)
whi h is an analog of the transformation (1.2.25) and leaves the measure
invariant. The al ulation given in Eq. (1.3.18) is an expli it illustration of
this fa t.
The general transformation (1.3.19) leads, when applied to the path integral in Eq. (1.3.1), to the S hwinger{Dyson equations


rb + m

Cal ulate straightforwardly the divergen e of the axial urrent (1.3.11)


using Eq. (1.3.14) and the onjugate one

Solution

(x)

rb m

= 0

(1.3.15)

r =  +ieA(x) :

(1.3.16)

with

Let us now dis uss how the measure in the path integral hanges under
the hiral transformation (1.3.7). The old and new measures are related by
h
i
D D = D 0 D 0 det e 2i (x) 5 (d) (x y) ;
(1.3.17)
where the determinant is over the spa e indi es x and y as well as over the matrix indi es i and j . Noti e that the determinant, whi h is nothing but the
Ja obian of the transformation (1.3.7), emerges for the Grassmann variables
to the positive rather than negative power as for ommuting variables. This is
a known property of the integrals (1.2.17) over Grassmann variables [Ber86
whi h look more like derivatives.
The logarithm of the Ja obian in Eq. (1.3.17) an be al ulated as
h

ln det e 2i (x) 5 (d) (x y)


= Sp [2i 5 = 2i

= Sp ln e 2i 5


ddx (x) (d) (0) sp 5 = 0 ; (1.3.18)

51

QUANTUM ANOMALIES FROM PATH INTEGRAL

(x)

;
(x)

w.s.
;
=
(x)

(x) w.s.
=

r m
b

(1.3.20)

whi h hold in the weak sense, i.e. under the sign of averaging over  and .
More restri tive transformations of the type of (1.3.7), whi h are asso iated with symmetries of the lassi al a tion and result in onserved urrents,
lead to some (less restri tive) relations between orrelators whi h are alled
Ward identities. This terminology goes ba k to the fties when a proper relation between the 2- and 3-point Green fun tions was rst derived for the
gauge symmetry in QED.
The simplest Ward identity, whi h is asso iated with the hiral transformation (1.3.7), reads

 J5 (0) i (x) j (y)


m=0

= i (d)(x) ( 5 )i (0) j (y)

i (d) (y)

i (x)

 5  (0) : (1.3.21)
j

It is lear from the way Eq. (1.3.21) was derived, that it is always satis ed as
a onsequen e of the quantum equations of motion (1.3.20).
Derive Eq. (1.3.21) in the operator formalism when the averages
are substituted by the va uum expe tation values of the T{produ ts.
Solution Use Eq. (1.3.14) (whi h a quires an extra
i in Minkowski spa e, where
the spatial -matri es are anti-Hermitean rather than Hermitean as in Eu lidean

Problem 1.20

52

CHAPTER 1.

PATH INTEGRALS

1.3.

QUANTUM ANOMALIES FROM PATH INTEGRAL

53

spa e, and holds in the quantum ase in the weak sense, i.e. when applied to a
state) and anoni al equal-time anti ommutation relations for and  with the
only non-vanishing anti ommutator being


(x0 y0 )  i (y ); j (x) = ij (d) (x y ) :
(1.3.22)

reads expli itly

Remark on 5 in d dimensions

The idea of regularizing the measure is to restri t ourselves to a large but


nite number of the basis fun tions. This is analogous to the dis retization
of Subse t. 1.1.4. We therefore de ne the regularized measure as

Let us re all that

5 = 1 2    d

(1.3.23)
2d=2  2d=2.

exists only for an even d when the size of the -matri es is


For
this reason the dimensional regularization is not appli able in al ulations of
the hiral anomaly.

Remark on gauge- xing


Note that we did not add a gauge- xing term to the a tion (1.3.4). It is
harmless to do that sin e the gauge- xing term does not ontribute to the
variation of the a tion under the hiral transformation. Moreover, all gauge
invariant quantities do not depend on the gauge xing. How to quantize a
gauge theory without adding a gauge- xing term will be explained in the next
Chapter.

D D

(d)(0) 

sp 5 =

1 0:

(1.3.24)

To regularize (d)(0), one needs [Ver78, Fuj79 to regularize the measure in


the path integral over and , sin e this term omes from the hange of the
measure under the hiral transformation.
Let us expand the elds and  over some set of the orthogonal basis
fun tions, similarly to how it is done in Eq. (1.1.78):
i (x)

in in (x) ;

i (x) =

in iny (x) ;

(1.3.25)

where there is no summation over the spinor index i. Here in and in are
Grassmann variables. The measure is then similar to that of Eq. (1.1.79) and

1 Y
Y
n=1 i

(D )R (D )R =

d in

M Y
Y
n=1 i

1 Y
Y
m=1 j

d in

d jm :

M Y
Y
m=1 j

d jm :

(1.3.26)

(1.3.27)

The hange of the measure under the hiral transformation reads


(D )R (D )R
Z

kj j
0
0
d
k
y
2
i
(
x
)


5
= (D )R (D )R det d x n (x) e
m (x) ; (1.3.28)
where the determinant is both over the n and m indi es and over the spinor
indi es k and j . This is the regularized analog of the nonregularized expression (1.3.17).
Using the orthogonality of the basis fun tions:
Z

1.3.3 Chiral anomaly


As is already mentioned, Eq. (1.3.18) involves the un ertainty

ddx jny (x)im (x) = nm ij ;

(1.3.29)

and Eq. (1.2.20), we rewrite the determinant on the RHS of Eq. (1.3.28) for
an in nitesimal parameter as
det

Z

kj
dd x kn y (x) e 2i (x) 5 jm (x)

= 1 + 2i

M Z
X
n=1

dd x yn (x) (x) 5 n (x) ;

where the spinor indi es are ontra ted in the usual way.
It is easy to see how this formula re overs Eq. (1.3.18) sin e
1
X
in (x)jny (y) = (d) (x y) ij
n=1

(1.3.30)

(1.3.31)

54

CHAPTER 1.

PATH INTEGRALS

in the nonregularized ase due to the ompleteness of the basis fun tions.
In the regularized ase, the sum over n on the LHS of Eq. (1.3.31) is
restri ted by M from above so that the RHS is no longer equal to the deltafun tion. We substitute
M
X
n=1

in (x)jny (y) = Rij (x; y) ;

(1.3.32)

with the RHS being the matrix element of some regularizing operator R.
It an be hosen in many ways. We shall work with several forms:

R = e a2 rb 2 ;
1
;
R =
b2
1 a2 r
1
R =
;
b
1 + ar

or

(1.3.33)

or

(1.3.34)

::: ;

(1.3.35)

b =  r . The parameter a is the ultraviolet uto . The uto


where again r
disappears as a ! 0 when Eq. (1.2.57) holds.
These regularizations (1.3.33) { (1.3.35) are non-perturbative, and preserve gauge invarian e sin e they are onstru ted from the ovariant derivab , whi h
tive r . A onsistent regularization o urs when R ommutes with r
is obviously true for the regularizations (1.3.33) { (1.3.35)10.
Therefore, we get
Z

dd x (x)  J5 = 2i Sp 5 R
= 2i

dd x (x) sp 5 R (x; x) :

(1.3.36)

It is worth noting that an extra R in Eq. (1.3.36) is a onsequen e of a


more general formula
Sp O

Sp OR ;

(1.3.37)

whi h says how to regularize tra es of operators.


10 This an be shown by hoosing the basis fun tions to be eigenfun tions of the Hermitean
b n = En n ) and applying r
b (ir
b 1 (y))jl to Eq. (1.3.32). The LHS
b ki (x)(r
operator ir
1
b r
b 1 jyi appears on the RHS. It
then does not hange (be ause En En = 1) while hxjr
b
oin ide with the RHS of Eq. (1.3.32) when r and ommute.

1.3.

QUANTUM ANOMALIES FROM PATH INTEGRAL

55

Remark on regularization of the measure


The regularization of the measure in the path integral by Eq. (1.3.27) is
equivalent to the point-splitting pro edure where the delta-fun tion in the
ommutator term is smeared a ording to Eq. (1.2.55).
To show this, let us note that the variational derivative an be approximated for the nite number of basis fun tions by

M
X


(1.3.38)
k:

n=1
k  n
This de nition extends the standard mathemati al one11 to the ase of spinor
indi es. The sum over k is in luded in order for the regularized variational
derivative to feel variations of all the spinor omponents of n when the variation is not diagonal in the spinor indi es.
When applied to
j
R (y )

i
R (x)

it yields

or, equivalently,

i
R (x)
j
R (y )

M
X
n=1

jny (y)

M
X

in in (x) ;

(1.3.39)

in (x)jny (y) = Rij (x; y) ;

(1.3.40)

n=1

ij (d) (x y) Reg.
=) Rij (x; y) :
(1.3.41)
whi h is a fermioni analog of Eq. (1.2.55).
Thus, we on lude that the regularization of the measure in the path
integral is equivalent to smearing the delta-fun tion in ommutator terms.

Remark on regularized S hwinger{Dyson equations


A ording to Subse t. 1.2.5, the pro edure from the previous paragraph results in the following regularized S hwinger{Dyson equations
Z



rb + m (x) w.s.
=
dd yR(x; y)  ;
(y)


Z

w.s.
(x) r
b
m
=
dd yR(x; y)
:
(1.3.42)
(y)
11 See,

, the book by Levy [Lev51.

e.g

56

CHAPTER 1.

PATH INTEGRALS

1.3.

QUANTUM ANOMALIES FROM PATH INTEGRAL

These equations are understood again in the weak sense, i.e. under the sign
of averaging over  and and obviously reprodu e Eq. (1.3.20) as a ! 0.

q1

Derive Eq. (1.3.36) using the regularized S hwinger{Dyson equation (1.3.42).


Solution The al ulation is similar to that of Problem 1.19 ex ept for the additional terms whi h are due to the RHS of Eq. (1.3.42). One gets for m = 0

57

k q1

Problem 1.21

R(x; y ) 5 (x) i (x) 5


(y )
2i sp 5 R (x; x) ;

 J5 w.s.
= i

dd y


A

Aq A
AK  q + k
AA

dd yR(x; y ) 
(y )
(1.3.43)

whi h is equivalent to Eq. (1.3.36) sin e the (x) there is an arbitrary fun tion.

Fig. 1.6: Triangular diagram asso iated with hiral anomaly in d = 4. The solid
lines orrespond to R0 given by Eq. (1.3.51). The wavy lines orrespond
to the eld strength.

1.3.4 Chiral anomaly ( al ulation)


In order to derive an expli it expression for the hiral anomaly, we should al ulate the RHS of Eq. (1.3.36) for some hoi e of the regularizing operator R.
b 2 in the denominator
Let us hoose R given by Eq. (1.3.34). The operator r
an be transformed as
rb 2 = r2 + 21 [  ;  r r
ie
[ ; F
= r2
4   
e
= r2 +  F ;
(1.3.44)
2
where the tra e of the spin matri es
1
(1.3.45)
 = [  ; 
2i
is given by
sp   5 =

4 :

(1.3.46)

Expanding in e:

R = R0 + R0 (: : :) R0 + : : :

(1.3.47)

with

R0 = 1 1a2  2 ;

(1.3.48)

we get s hemati ally

eF
R0 e2F R0
2
e2
d4 x (x)
F F~ ;
162  

Sp 5 R = a4 Sp 5 R0

(1.3.49)

where

F~ = 21  F

(1.3.50)

is the dual eld strength.


The al ulation des ribed in Eq. (1.3.49) is most easy to do in momentum
spa e where it is asso iated with one-loop diagrams. The analyti expression
to be al ulated an be represented in d = 4 graphi ally by the triangular diagram in Fig. 1.6. The solid lines are asso iated with R0 given by Eq. (1.3.48),
whi h reads in momentum spa e as

R0 (p) =

1
:
1 + a2 p2

(1.3.51)

The wavy lines orrespond to the eld strength. The lower vertex is asso iated
with 5 .
The integral over the four-momentum q, whi h ir ulates along the triangular loop, an be easily al ulated by introdu ing ! = aq and transforming

58

CHAPTER 1.

the integral as

d4 q f (q)

PATH INTEGRALS

Z
 
! a14 d4 ! f !a :

1.3.

QUANTUM ANOMALIES FROM PATH INTEGRAL

(1.3.52)

Noti e that the integral involves a 4 whi h an els a4 oming from the expansion in e whose proper term is given by the intermediate expression in
Eq. (1.3.49). Therefore, the result is non-vanishing and a independent as
a ! 0. Higher terms of the expansion in e are proportional to higher powers
in a and vanish as a ! 0.
From Eqs. (1.3.36) and (1.3.49), we get nally

ie2
(1.3.53)
F F~ :
82  
The anomaly on the RHS is known as the Adler{Bell{Ja kiw anomaly. Its
appearan e is usually related to the fa t that any regularization an not be
simultaneously gauge and hiral invariant.
 J5 =

Problem 1.22 Cal ulate the oe ient in Eq. (1.3.49) and show that it is regulator
independent.
Solution The ontribution of the triangular diagram of Fig. 1.6, whi h represents
the intermediate expression in Eq. (1.3.49), reads expli itly

4e2 a4

2 Sp 5 R =

d4 x

d4 y

2e2
16 2

(1.3.54)

d4 q1
F (q )F~ (k q1 )
(2 )4  1 
Z
4
 (2dq)4 (1 + a2 q2 ) (1 + a2 (q +1q1 )2 ) (1 + a2 (q + k)2 ) (1.3.52)
=
Z1 2 2
Z
Z
d4 k ikx d4 q1
! d!
4
d x (x)
e
F (q )F~ (k q1 )
(1.3.55)
(2 )4
(2 )4  1 
(1 + ! 2 )3

2e2 a4

d4 x (x)

d4 k e ikx
(2 )4

whi h re overs the RHS of Eq. (1.3.49).


An analogous al ulation an be repeated for other regulators (1.3.33) and
(1.3.35). Let us denote

r(a2 p2 )

 R0 (p) :

q+k
r

5
Fig. 1.7: The diagram asso iated with hiral anomaly in d = 2. The solid lines
orrespond to R0 given by Eq. (1.3.51). The wavy line orresponds to the
eld strength.

Then the only di eren e with Eq. (1.3.55) is that the last integral over ! 2 is repla ed
by
Z1
! 2 d! 2 r00 (! 2 ) = r(0) = 1
(1.3.57)
0

for reasonable fun tions r whi h look like those given by Eqs. (1.3.33) { (1.3.35).

d4 z

 (x)R0 (x; y)F (y)R0 (y; z)F~ (z)R0 (z; x) :


In momentum spa e, it be omes

59

(1.3.56)

An anomaly whi h is analogous to the Adler{Bell{Ja kiw anomaly (1.3.53)


exists in d = 2 where
e
 J5 =
 F :
(1.3.58)
2  
This anomaly is given by the diagram depi ted in Fig. 1.7. It involves only
two lines with the regulator R0 (p) sin e in d = 2
Z
Z
! 
1
(1.3.59)
d2 q f (q) ! 2 d2 ! f
a
a
so that all terms with more lines vanish as a ! 0.
Cal ulate 2 Sp 5 R in d = 2.
Pro eeding as before, we see that only the diagram of Fig. 1.7 is essential
in d = 2 whi h yields
Problem 1.23

Solution

2iea2

d2 x

d2 y (x)R0 (x; y )F (y ) R0 (y; x) =

60
Z
2
2iea

CHAPTER 1.

PATH INTEGRALS

d2 k e ikx F (k)
d2 q
1


2
(2 )
(2 )2 (1 + a2 q 2 ) (1 + a2 (q + k)2 )
Z1
Z
Z
(1.3.59) ie
d! 2
d2 k e ikx F (k)
= 2
d2 x (x)


2
4
(2 )
(1 + ! 2 )2

d2 x (x)

d2 x (x)

ieF (x)
:
2

(1.3.60)

The linear-in-F term is nonvanishing sin e


sp  5 = 2i

(1.3.61)

in d = 2.
The result is again regulator independent sin e the integral over ! is repla ed
for an arbitrary R0 (p) by
Z1
d! 2 r0 (! 2 ) = r(0) = 1
(1.3.62)
0

1.3.

61

1.3.5 S ale anomaly


The s ale transformation is de ned by
x ! x0 =  x ;
' (x) ! '0 (x0 ) = l' ' (x0 ) :

(1.3.65)
(1.3.66)

The index l' is alled the s ale dimension of the eld '. The value of l' in
a free theory is alled the anoni al dimension, whi h equals (d 2)=2 for
bosons (s alar or ve tor elds) and (d 1)=2 for the spinor Dira eld, i.e. 1
and 3=2 in d = 4, respe tively. Sometimes l' is alled, for histori al reasons,
the anomalous dimension. More often the term \anomalous dimension" is
used for the di eren e between l' and the anoni al value.
The proper Noether urrent, whi h is alled the dilatation urrent, is
expressed via the energy-momentum tensor  as

D = x 

(1.3.67)

so that its divergen e equals the tra e of the energy-momentum tensor over
the spatial indi es:

where Eq. (1.3.56) has been used.

 D =  ;

Remark on non-Abelian hiral anomaly


Equation (1.3.53) holds also in the ase of a non-Abelian gauge group with
a being the non-Abelian eld strength
F
a (x) =  Aa (x)  Aa (x) + gf ab Ab (x)A (x) :
F
 
 



ig2 X a ~ a
F F :
162 a  

12 See Ref. [Mor86 for a review of anomalies in gauge theories.

 =

m

(1.3.69)

at the lassi al level.


The above formulas an be obtained from the Noether theorems whi h
state

S =

(1.3.64)

The oe ient in this formula di ers by 1=2 from that in Eq. (1.3.53) whi h
is due to di erent normalizations in the Abelian and non-Abelian ases. This
will be explained in more detail in Subse t. 2.1.1 below.
The d = 2 anomaly (1.3.58) exists for the singlet axial urrent only in the
Abelian ase.

(1.3.68)

sin e the energy-momentum tensor is onserved. For the a tion (1.3.4) one
gets

(1.3.63)

Here f ab are the stru ture onstants of the gauge group and g is the oupling
onstant. The non-Abelian analog of Eq. (1.3.53) for the axial urrent, whi h
is a singlet w.r.t. the gauge group, reads12

 J5 =

QUANTUM ANOMALIES FROM PATH INTEGRAL

dd x  (x)  D (x)

(1.3.70)

or

 D (x) =

S
:
 (x)

(1.3.71)

In the massless ase, m = 0, the RHS of Eq. (1.3.69) vanishes and the
dilatation urrent is onserved. This is a well-known property of ele trodynami s with massless ele tron whi h is s ale invariant at the lassi al level. A

62

CHAPTER 1.

PATH INTEGRALS

generi s ale-invariant theory does not depend on parameters of the dimension of mass or length. This usual dimension is to be distinguished from the
s ale dimension whi h is de ned by Eq. (1.3.66). The dimensional parameters
do not hange under the s ale transformation (1.3.65).
In the quantum ase, the s ale invarian e is broken by the (dimensional)
uto a. The energy-momentum tensor is no longer tra eless due to loop
e e ts. The relation (1.3.68) holds in the quantum ase in the weak sense,
i.e. for the averages


 D F [A; ;  =  F [A; ;  ;
(1.3.72)
where F [A; ;  is a gauge-invariant fun tional of A, and .
For a renormalizable theory like QED, the RHS of Eq. (1.3.72) is proportional to the Gell-Mann|Low fun tion B(e2) whi h is de ned by

a de2
(1.3.73)
= B e2 :
da
A non-trivial property of a renormalizable theory is that the RHS in this
formula is a fun tion solely of e2 | the bare harge.
The meaning of the renormalizability is very simple: physi al quantities do
not depend on the uto a, provided the bare harge e is hosen to be uto
dependent a ording to Eq. (1.3.73). This dependen e of e on a e e tively
a ounts for the distan es smaller than a, whi h are ex luded from the theory.
The pre ise relation between the tra e of the energy-momentum tensor
and the Gell-Mann{Low fun tion reads
2
w.s. B e
F2 ;
(1.3.74)
 =
4e2 
where the equality is understood again in the weak sense. This formula was
rst obtained in Refs. [Cre72, CE72 to leading order in e2 and proven in
Ref. [ACD77 to all orders in e2 .
Note that this formula holds in the operator formalism only when applied
to a gauge invariant state. The reason is that otherwise a ontribution from
a gauge- xing term in the a tion would be essential. It does not ontribute,
however, to gauge-invariant averages whi h an be formally proven using the
gauge Ward identity.
Prove the relation (1.3.74).
Let us absorb the oupling e into A introdu ing
A = ieA ;
F = ieF :

Problem 1.24
Solution

1.3.

63

QUANTUM ANOMALIES FROM PATH INTEGRAL

The Lagrangian density of massless QED then reads




1 2
F :
4e2

L =  b Ab

(1.3.76)

To prove Eq. (1.3.74), let us use the hain of Eqs. (1.3.68) and (1.3.71). It is
ru ial that in the absen e of other dimensional parameters the derivative = an
be repla ed by =a sin e all dimensionless quantities depend in a theory with a
uto only on the ratios of the type x=a.13 Sin e the dependen e on the uto a
2 , Eq. (1.3.74) an be
enters in Eq. (1.3.76) formally only via e 2 in front of F
heuristi ally proven by rst di erentiating with respe t to a and then expressing
the result via F again. We have used herein the fa t that F is invariant under
the renormalization-group transformation and, therefore, does not depend on a.

In the path-integral approa h, a ontribution to the s ale anomaly omes


from the regularized quantum measure. Pro eeding as in Subse t. 1.3.3, we
get

 D (x) =

sp R(x; x) :

(1.3.77)

Derive Eq. (1.3.77) using the regularized S hwinger{Dyson equations (1.3.42).


Solution The energy-momentum tensor of QED reads

Problem 1.25

 = F F


1
2 + 1  $
 F
 r
4
4

$
+   r

(1.3.78)

Taking the tra e, one gets


$
 = 1   r :
2

(1.3.79)

Using Eq. (1.3.42), it an be transformed as

 =
=

m +
m

Z

dd yR(x; y )
(x)
2
(y )
sp R (x; x) ;

(x)

dd yR(x; y ) 
(y )
(1.3.80)

whi h reprodu es Eq. (1.3.77) as m ! 0.


(1.3.75)

13 This is the reason why the Callan{Symanzik equations, whi h are nothing but the

dilatation Ward identities, oin ide with the renormalization-group equations.

64

CHAPTER 1.

q1

q1

k q1

PATH INTEGRALS

k q1

Equation (1.3.74) holds in the non-Abelian Yang{Mills theory as well if F


a given by Eq. (1.3.63).
is substituted by the non-Abelian eld strength F
The orresponding formula, reads

q+k
r

a)



b)

Fig. 1.8: The diagrams whi h ontribute to the s ale anomaly in d = 4. The wavy
line orresponds to the eld strength.

To al ulate the s ale anomaly we should therefore perform a one-loop


al ulation of

+
*



1

 x

sp x
1 + a2 i b 2
*

1
sp x
1 + a2 (i + eA )2

+


1 a2 e F x
 

(1.3.81)

whi h is again most onvenient to do in momentum spa e. The propagator is


given by Eq. (1.3.48) while the verti es, whi h emerges in the orresponding
Feynman rules for the expansion in e, ome from the operators
2iea2A  ;

e2 a2 A2 ;

1 2
2 ea  F :

The only diagrams whi h survive as a ! 0 in d = 4 are depi ted in


Fig. 1.8. The al ulation of the diagram of Fig. 1.8a is the same as in Subse t. 1.3.3 while the diagram of Fig. 1.8b gives a total derivative whi h does
not ontribute to the s ale anomaly.
The al ulation of the diagram of Fig. 1.8a yields
sp R(x; x) =

65

Remark on non-Abelian s ale anomaly

sp R(x; x) =

QUANTUM ANOMALIES FROM PATH INTEGRAL

The one-loop Gell-Mann{Low fun tion an now be al ulated using


Eqs. (1.3.77) and (1.3.74), whi h reprodu es the known result for QED. The
higher-order orre tions in e do not vanish for the s ale anomaly.

q + q1


A

Aq A
AK  q + k
A 
A

1.3.

2 (x)
e2 F
:
482

(1.3.82)

w.s.

B g2
4g2

a F~ a :
F


(1.3.83)

A heuristi proof, presented in Problem 1.24 for the Abelian ase, an be


repeated. The equality is again understood in the weak sense when averaged
between gauge-invariant states. The ontribution of gauge- xing and ghost
terms are then an elled due to the gauge Ward identity whi h is alled in
this ase the Slavnov{Taylor identity. The proof of Eq. (1.3.83) was given in
Refs. [CDJ77, Nie77.

66

CHAPTER 1.

PATH INTEGRALS

1.4.

1.4 Instantons in quantum me hani s


Instantons are solutions of the lassi al equations of motion with a nite
Eu lidean a tion. Su h eld on gurations are not taken into a ount in perturbation theory. Instantons are hara terized by a topologi al harge whi h
may result in a onserved quantum number and never shows up in perturbation theory. In Minkowski spa e, instantons are asso iated with tunneling
pro esses between va ua labelled by a distin t topologi al harge.
Instantons rst appear in Yang{Mills theory [BPST75 although this kind
of lassi al solution was used long before in statisti al physi s [Lan67. In
this Se tion we onsider instantons in quantum me hani s as an illustration
of path-integral al ulations. The onsideration follows the original paper by
Polyakov [Pol77 ex ept for te hni al details.

1.4.1 Double-well potential


Let us onsider a one-dimensional quantum-me hani al system with the
double-well potential


2 2

 2 
x
4

1 2 2 1 4 4
=
 x + x + :
(1.4.1)
2
4
4
This is nothing but an anharmoni os illator with a wrong sign for the oe ient of the quadrati term,14 whi h usually appears with a positive oe ient
!2 =2. We have introdu ed
V (x) =

2 =

!2

(1.4.2)

in order to work with real numbered values. The onstant term is added for
later onvenien e. The potential (1.4.1) as a fun tion x is depi ted in Fig. 1.9.
The (Eu lidean) a tion is de ned by


Z
1 2
S [x =
d x_ ( ) + V (x( ))
2

(1.4.3)

14 It is often alled the mass term. This terminology omes from quantum eld theory,

where the potential (1.4.1) is onsidered in the ontext of a spontaneous breaking of the
re e tion symmetry x ! x. In our quantum-me hani al problem, de ned by the Eu lidean a tion (1.4.3), the mass of the non-relativisti parti le is absorbed in  whi h has,
therefore, the dimension of [length2 . This is explained already in Subse t. 1.1.6.

67

INSTANTONS IN QUANTUM MECHANICS

6V (x)

+ p

p

-x

Fig. 1.9: The double-well potential (1.4.1). The short verti al lines represent the

position of the minima (1.4.4). The dashed lines orrespond to the energy
E0 of the lowest state in a single well, i.e. to that in the limit  ! 0.

with V (x) given by Eq. (1.4.1). The plus-sign between the kineti and potential energies is be ause we are in Eu lidean spa e.
It follows from Eqs. (1.4.1) and (1.4.3) that the parameter  has the
dimension of [length 2 or, in other words, the dimensions of x and  are
[ 1=2 and [ 1 , respe tively. Analogously, the dimension of the onstant 
is [3 .
For   3 , the potential (1.4.1) has super ially two degenerate va ua

(1.4.4)
x0 =  p ;

whose positions oin ide with the minima of the potential in Fig. 1.9.
The degenera y between the two minima is preserved at all orders of perturbation theory, where an expansion near one of the minima of the potential
(either the left or right one) is arried out:

(1.4.5)
x ( ) =  p +  ( )

p
with  ( )  = . The orrelator at asymptoti ally large  is
2

hx (0) x ( )i ! 

+::: :

(1.4.6)

Its nonvanishing value means that a parti le is lo alized at one of the two
va ua.

68

CHAPTER 1.

PATH INTEGRALS

1.4.

The next terms of the perturbative expansion in  do not spoil this result
sin e the potential (1.4.1) be omes

and


V = 2 2  3 + 4
(1.4.7)
4
after the shift (1.4.5), and has the positive sign of the quadrati term. Therefore, perturbation theory around the va uum x
0 is a usual one, and the
parti le lives perturbatively in one of the two va ua.
However, we know from quantum me hani s that (non-perturbatively)

is the lowest energy for the os illator potential (1.4.7) as  ! 0. Denoting

hx (0) x ( )i

jxn0 j2 e

(En E0 )

(1.4.8)

at imaginary time  = it with En being the energy of the n-th eigenstate of


the Hamiltonian, and xn0 being the proper matrix element. Therefore,

hx (0) x ( )i 
for large  , where
r

48 2 2 3
E = 
exp

3

E 

2 2 3
3

(1.4.9)
!

(1.4.10)

Derive Eq. (1.4.10) modulo a onstant fa tor within standard quantum me hani s.
Solution Let us use the semi lassi al formula [LL74 (Problem 3 in x50)

Problem 1.26

+a
2
a dx 2(V (x) E0 ) ;
e

where a are the lassi al turning points, whi h are determined by

V (a) = E0 ;

E0 =

h =

p 3;
2

2

z =

69

(1.4.13)

 x;


(1.4.14)

the integral in the exponent on the RHS of Eq. (1.4.11) an be al ulated by an


expansion in h whi h gives
1
2h2

1 h
Z
1+h

p
dz (1 z 2 )2 4h2 = 2 2 + ln h + O(1) :
3h

(1.4.15)

Substituting into Eq. (1.4.11), one re overs Eq. (1.4.10) modulo a onstant fa tor.

1.4.2 The instanton solution


In the path-integral approa h, the orrelator (1.4.8) is given by

is the energy splitting between the two lowest states (symmetri and antisymmetri ) for   3 whi h exponentially vanishes as  ! 0.
The appearan e of imaginary time in Eq. (1.4.8) is be ause under a barrier
parti les live in imaginary time. We may say that imaginary time is an
adequate language to des ribe a tunneling through a barrier.
Sin e the RHS of Eq. (1.4.9) vanishes as  ! 1, the re e tion symmetry x ! x, whi h is broken in perturbation theory, is restored nonperturbatively as  ! 1.

E =

INSTANTONS IN QUANTUM MECHANICS

(1.4.11)
(1.4.12)

h x (0) x ( ) i

DxRe S[x x (0) x ( )


Dx e S[x

(1.4.16)

with no restri tions on the integration over x. This is a quantum me hani al


analogue of the path integrals de ned in Subse t. 1.2.1.
At small , the path integral (1.4.16) an be evaluated by the saddlepoint method. The reason is that for x given by Eq. (1.4.4), whi h are the
minima of the a tion (1.4.3), the Gaussian u tuations around (1.4.4) are not
essential as  ! 0. This is most easily seen by making the shift (1.4.5) and
noting that ( ) is O(1) at the saddle points a ording to Eq. (1.4.7) whose
RHS is quadrati in ( ) as  ! 0.
Performing the saddle-point evaluation of the path integral (1.4.16), one
gets

2
+::: :
(1.4.17)

Noti e that x(0) and x( ) in the integrand an be substituted by the saddlepoint values after whi h the integral over Gaussian u tuations an els by
the same one in the denominator. In other words we have reprodu ed that
ea h of the trivial minima (1.4.4) results in Eq. (1.4.6).

hx (0) x ( )i

70

CHAPTER 1.

PATH INTEGRALS

1.4.

71

INSTANTONS IN QUANTUM MECHANICS

An
p analogous solution whi h interpolates between =  at  = 1 and
=  at  = +1 is alled an anti-instanton. It di ers from Eq. (1.4.19)
by an overall minus sign:

6

xainst (
0

p

0 ) =

tanh

 (

p 0 ) ;
2

(1.4.20)

and is obviously also a solution of the lassi al equation (1.4.18).

p Find all solutions pof Eq. (1.4.18) with the boundary onditions
1) = =  and x(+1) = = .

Problem 1.27

x(

Equation (1.4.18) looks like the Newton equation for a lassi al parti le,
whose mass is equal to unity, in the upside-down potential V (x) (its shape an
be obtained from the one depi ted in Fig. 1.9 by re e ting with respe t to the
horizontal axis V = 0). The rst integral of motion is the energy

Solution

p

+ p

-x

Fig. 1.10: Graphi representation of the one-kink solution (1.4.19) as a fun tion of
.

Minima of the a tion (1.4.3) an be alternatively obtained from the lassi al equation of motion

x 2 x + x3 = 0 :

(1.4.18)

The trivial minima (1.4.4) obviously satisfy this equation.


However, another solution of the lassi al equation of motion (1.4.18)
exists:
 (  )

(1.4.19)
xinst ( 0 ) = p tanh p 0 ;
2

whi h is asso iated with another (lo al) minimum of the lassi al a tion. This
solution is alled an instanton or a pseudoparti le. The arbitrary onstant 0
in Eq. (1.4.19) is the position of the enter of instanton.
The solution (1.4.19) is also known as a kink in this quantum me hani al
problem. It interpolates between the two minima (1.4.4) when  hanges
from 1 to +1 as is depi ted in Fig. 1.10. On the same gure we kept the
shape of V (x) from Fig. 1.9.

 x2
4

1
= x_ 2
2

2


2

(1.4.21)

whi h is obviously onserved due to Eq. (1.4.18).


Equation (1.4.21) an easily be solved for the velo ity

x_ =

2(E + V (x));

(1.4.22)

where we have hosen the positive sign a ording to the boundary


p ondition. It
says also that E = 0 in order for the parti le to stay at x = =  for  ! 1 sin e
this point is asso iated with the maximum of V (x). Therefore, we get
r

x_ =

 2
2 

x2

(1.4.23)

whi h results after integration in Eq. (1.4.19) with 0 being the integration onstant.
It is evident that the solution is unique.

For the instanton (or anti-instanton) minimum, one gets, substituting in


Eq. (1.4.3),

2 2 3
;
S [xinst =
3

(1.4.24)

whi h oin ides with the (minus) exponent in Eq. (1.4.10) for the energy
splitting E .

72

CHAPTER 1.

PATH INTEGRALS

1.4.3 Instanton ontribution to path integral


The ontribution of the instanton on guration looks as if it is suppressed in
the path integral by a fa tor of exp ( S [xinst ), but in fa t this exponential is
multiplied by  sin e the instanton has a zero mode. This fa tor of  appears
after an integration over the olle tive oordinate 0 | the instanton enter.
The expli it result for the one-kink ontribution to the orrelator (1.4.16)
reads [Pol77
2

h x (0) x ( ) i

2 4
2 2 3
=
1 C
exp

3

2 2 3
3

1.4.

INSTANTONS IN QUANTUM MECHANICS

The u tuations around the instanton are Gaussian ex ept for one mode, whi h
is asso iated with a translation of the instanton enter, 0 . This zero mode is given
by

0 ( )


d2
d 2

2 + 3x2

h x (0) x ( ) i

+ e S [xinst

D( ) e

 e (S [x S [xinst)

d ( 12 _ 2 +2 2 )

(1.4.26)

where the path integral in the numerator is over u tuations around the instanton
solution (1.4.19). The normalizing fa tor in the denominator is asso iated with
averaging over the Gaussian u tuations around the trivial minima (1.4.4) whose
potential energy is des ribed by the quadrati term in Eq. (1.4.7). There are two
su h trivial minima (x+ and x ) and two one-kink minima (instanton and antiinstanton) so these fa tors of 2 an el.
Keeping the quadrati term in the expansion around the instanton:

x( ) = xinst (
one gets
1
S [x S [xinst =
2

d

0 ) + (
_ 2

0 ) ;

(1.4.27)

2 2 + 3x2

inst


2

x_ inst = 0

+1
Z

1 =

(1.4.30)

d (u[x  )

(1.4.31)

in the path integral in the numerator on the RHS of Eq. (1.4.26). Here u[x is
determined by the equation
+1
Z

1

2

(1.4.29)

as a result of di erentiating Eq. (1.4.18) w.r.t. 0 .


To deal with the zero mode, let us insert the unity

Problem 1.28

2
= 


R
Dx( ) x(0)x( )

inst

where C is a (dimensional) onstant.


Derive Eq. (1.4.25) using the Faddeev{Popov tri k to deal with the
olle tive oordinate 0 .
Solution Let us approximate the path integrals in the numerator and denominator of Eq. (1.4.16) for small  by the sum of the ontributions from the trivial
minima (1.4.4) and the one-kink minima (1.4.19) and (1.4.20). Sin e the one-kink
ontribution is suppressed by exp ( S [xinst ), we an expand the denominator to
give

/ x_ inst ( ) :

This is obvious be ause

!3
5 ; (1.4.25)

73

d y (

u[x) x( ) = 0

(1.4.32)

with

y ( ) =

hR

+1

x_ ( )

2
1 dt x_ (t)

i1=2

(1.4.33)

being the normalized derivative of x( ).


Under the translation,

!  0 =  0 ;

(1.4.34)

one gets

x( )

! x( 0) = x( 0 ) :

(1.4.35)

This leaves the measure and the a tion in the path integral (1.4.26) invariant, while

(1.4.28)

u[x

! u[x + 0 :

(1.4.36)

74

CHAPTER 1.

PATH INTEGRALS

Therefore, the integration over the instanton enter, 0 , in the numerator of


Eq. (1.4.26) fa torizes and we get


Z
2
Dx( ) x(0)x( )
e (S [x S [xinst) =


Z +1
2
d0 xinst ( 0 )xinst ( 0 ) 

1
Z
R
1 d (_ 2 2 2 +3x2 2 )
inst
:
(1.4.37)
 D( ) (u[xinst ( ) + ( )) e 2
We have substituted the integration over the zero mode 0 by the integration over
the olle tive oordinate 0 . The remaining path integral is nite sin e the integration runs over the dire tions whi h are orthogonal to the zero mode.
The integral over 0 is equal to

Z +1
2
22
d0 xinst ( 0 )xinst ( 0 ) 

(1.4.38)
=


1
as  ! 0. This is be ause
(1.4.39)
xinst ( 0 ) = p sign ( 0 )

as  ! 0.
Expanding the delta-fun tion in :
Z +1
Z +1




(u[x) =
d y_ inst ( )xinst ( )
d yinst ( )( ) ; (1.4.40)
1
1
and noting that
Z +1
p
2 2 3
2
d x_ inst ( ) =
;
(1.4.41)
3
1
we get
Z

D( ) (u[xinst ( ) + ( )) e

2 2 3 D( )
3

Z +1

1
2

d yinst ( )( ) e

d (_ 2 2 2 +3x2inst 2 )
1
2

d (_ 2 2 2 +3x2inst 2 )

:(1.4.42)
1
p
Noti e the appearan e of the fa tor S [xinst .
We have thus obtained Eq. (1.4.25) with
R
R

R
2 2 2
2
2
1
D( ) +11 d yinst ( )( ) e 2 d (_   +3xinst  )
R
: (1.4.43)
C = 2
R
1 2 2 2
D( ) e d ( 2 _ +  )

1.4.

INSTANTONS IN QUANTUM MECHANICS

Cal ulate the ratio of determinants in Eq. (1.4.43).


Let us introdu e the notation
 ;
z = p
D = d :
dz
2

75

Problem 1.29
Solution

Noting that

x2inst ( ) = 2 1

1 ;
osh2 z

(1.4.44)

(1.4.45)

we an rewrite the ratio of determinants as




det0 D2 + 4 6= osh2 z
4

2
:
(1.4.46)
B = 2

det [ D2 + 4
The notation det0 means that the zero eigenvalue is ex luded. An extra fa tor of 2
omes from the normalization of the Gaussian integral in the denominator whi h
involves one integral more.
The RHS of Eq. (1.4.46) an be al ulated by the limiting pro edure




det0 D2 + 4 6= osh2 z
det D2 + ! 2 6= osh2 z
= lim
: (1.4.47)
!!2 (! 2 4) det [ D2 + ! 2
det [ D2 + 4
To ompute the ratio of the Fredholm determinants

R! [v 

det

D2 + ! 2 + v (z )
det [ D2 + ! 2

(1.4.48)

for the potential

v (z ) =
let us note that

6
;
osh2 z

(1.4.49)


i
h
1
1
 ln R [v = Sp
Sp
!
2
2
2
2
2
!
D + ! + v (z )
D +!
+
1
Z


dz R! (z; z ; v ) 1 ;
=
(1.4.50)
2!
1

where the diagonal resolvent R! (z; z ; v ) is de ned by Eq. (1.1.119) with G = 1 and
V  v . The term 1=2! on the RHS, whi h equals the diagonal resolvent in the
free ase when v = 0 (see Eq. (1.1.38)), omes from the free determinant in the
denominator on the RHS of Eq. (1.4.48).

76

CHAPTER 1.

PATH INTEGRALS

A ru ial observation is that the diagonal resolvent for the potential (1.4.49) is
given by the simple formula

R! (z; z ; v ) =

1
2!

v (z )
v 2 (z )
+
;
2
2
4! (! 1) 8! (! 1)(! 2 4)

(1.4.51)

whi h an easily be veri ed by substituting into the Gelfand{Di key equation (1.1.123) with G = 1. The reason is that the potential (1.4.49) is integrable
and possesses two bound states (see, e.g., x23 of Ref. [LL74).
Cal ulating the integral over z on the RHS of Eq. (1.4.50), by the formulas
+1
+1
Z
Z
dz = 4 ;
dz = 2 ;
(1.4.52)
2
3
osh z
osh4 z
1
1
we obtain

 ln R [v = 1  1 + 2  ;
!
! 2
! !2 1 !2 4
whi h is easily integrated over ! to give
det

(1.4.53)

D2 + ! 2 6= osh2 z
(! 2)(! 1)
=
:
det [ D2 + ! 2
(! + 2)(! + 1)

(1.4.54)

The integration onstant has been determined requiring that


lim

!!1

R! [v = 1 :

(1.4.55)

Substituting into Eq. (1.4.47), we get


r

C = 2B =

48



(1.4.56)

whi h oin ides with the onstant in Eq. (1.4.10).


For other methods of al ulating the ratio of determinants in the one-instanton
ontribution, see the original papers [Lan67, Pol77, the reviews [Col77, VZNS82
or Chapter 4 of the book by Polyakov [Pol87.

1.4.4 Symmetry restoration by instantons


At   1=E , many kinks be ome essential. A many-kink \solution" an

be approximately onstru ted from several single kinks and anti-kinks, whi h
are separated along the  -axis by the some distan e R  1=, sin e the
intera tion between kinks would be  exp f Rg. Su h a on guration is

1.4.

INSTANTONS IN QUANTUM MECHANICS

p
+= 
p

77

= 

Fig. 1.11: The many-kink on guration xM-kink ( ) whi h is ombined from the solution (1.4.19).

depi ted in Fig. 1.11 for the ase when the number of kinks is equal to the
number of anti-kinks. An analogous on guration with the number
p of kinks
p
one more than the number of anti-kinks onne ts the =  and = 
va ua.
It is not an exa t solution of Eq. (1.4.18) sin e the kink and the anti-kink
attra t and have a tenden y to annihilate. However, it is an approximate
solution as  ! 0.
Analyti ally, the M -kink on guration an be represented as

xM-kink ( ) =

p

M
Y

 i=1

sign(

i ) ;

(1.4.57)

where i are the enters of the instantons (or anti-instantons), of whi h the
M -kink on guration is built out, and

1  2  : : :  M :

(1.4.58)

Equation (1.4.57) assumes that the kinks do not intera t and are in nitely
thin as  ! 0. The a tion of the on guration (1.4.57) is therefore given by

2 23
M;
3
i.e. it equals M times the a tion for the one kink.
Summing over many-kink on gurations, one gets [Pol77

S [xM-kink =

(1.4.59)

2  E
e
;
(1.4.60)

where E is given by Eq. (1.4.10). The x ! x symmetry is now restored
as  ! 1. This restoration is produ ed by instantons = lassi al traje tories
with a nite (Eu lidean) a tion.

h x (0) x ( ) i

78

CHAPTER 1.

PATH INTEGRALS


Fig. 1.12: Graphi representation of a periodi potential.
Obtain the exponentiation of the one-kink ontribution (1.4.25)
after summing over the M -kink on gurations (1.4.57) in the dilute gas approximation when the intera tion between kinks is disregarded.
Solution The al ulation of the ontribution of the M -kink on guration (1.4.57)
to the path integral is quite analogous to that for the one kink whi h is des ribed
in Problem 1.28. One gets
Z 
Z 1
Z M 1
1
2 X
h x(0)x( ) i =  ( E )M d1 d2 : : :
dM ; (1.4.61)
0
0
0
M =0
Problem 1.30

whi h reprodu es Eq. (1.4.60) by noting that the ordered integral is equal to
Z 
0

d1

Z 1
0

d2 : : :

Z M 1
0

dM =

M :
M!

(1.4.62)

This al ulation is very similar to the one in statisti al me hani s for the exponentiation of a single-parti le ontribution to the partition fun tion in the ase of
an ideal gas.

1.4.5 Topologi al harge and -va ua


Let us onsider a periodi potential whose period equals 1, whi h is depi ted
in Fig. 1.12. It an be viewed as being de ned on a ir le S 1 of a unit length.
The boundary onditions are

x(1) = x(0)

in perturbation theory ;

x(1) = x(0) + n

for n-instanton solution :

INSTANTONS IN QUANTUM MECHANICS

(1.4.63)

(1.4.64)

79

where k (M ) is the k-th homotopy group whose elements are lasses of ontinuous maps of the k-sphere S k onto M . Eq. (1.4.64) says that an (integer)
winding number n 2 Z is asso iated with the mapping S 1 ! S 1 , whi h
ounts how many times the target is overed.
We see the di eren e between the M -kink on guration for the doublewell potential and the multi-instanton solution for the periodi potential.
The former was not an exa t solution of the lassi al eld equation (1.4.18).
Only a single instanton or anti-instanton was a solution that onne ts the
two va ua. This is why we need a periodi potential for the multi-instanton
solution to exist due to the topologi al argument.
The value of n in the boundary ondition (1.4.63) is alled the topologi al
harge of the instantons, while n < 0 is asso iated with anti-instantons. The
va uum states are labelled by n: jni. The n-instanton on guration onne ts
the jmi and hm + nj states. Therefore, instantons are asso iated in Minkowski
spa e with the pro ess of tunneling between topologi ally distin t va ua16
rather than with real parti les. For this reason, they are sometimes alled
pseudoparti les in Eu lidean spa e.
It is onvenient to onsider another representation of va uum states
1
X
(1.4.65)
ji =
e in j ni ;
n= 1
whi h are alled the -va ua. The -va ua are orthogonal
1 X
1
X
0
h j  0 i =
e i(n  m) hm j ni = 2 ( 0 ) ; (1.4.66)
m= 1 n= 1
where 2 is a periodi delta-fun tion with period 2. Here we used the
orthogonality of the n-states:

hm j ni

The multi-instanton solution always exists be ause of the topologi al


formula15

1 (S 1 ) = Z ;
15 See, e.g., the book [DNF86 (x17.5 of Part II).

1.4.

= mn :

(1.4.67)

16 The Minkowski-spa e interpretation of instantons is attributed to V.N. Gribov (unpublished). It is based on the fa t that when the parti le is lo alized in one of the two wells
its momentum is inde nite and an sometimes be very large so that the proper energy is
above the barrier between the two wells. Su h a parti le jumps from the given well to the
other one. The hara teristi time of this pro ess is small in the typi al units given by
. In other words this pro ess is instantaneous that explains the term \instanton" whi h
was introdu ed by 't Hooft. The exponential suppression with  of the one-instanton ontribution (1.4.25) represents quantitatively the fa t that the probability of having large
momentum is small.

80

CHAPTER 1.

The -va uum partition fun tion reads

Z =

PATH INTEGRALS

Dx e S[x+i

d x_ ( )

d x_ ( ) = x (1) x (0)

(1.4.68)

(1.4.69)

whi h never shows up in perturbation theory. Therefore, the partition fun tion (1.4.68) an be alternatively represented as

Z =

n x(1)=x(0)+n

INSTANTONS IN QUANTUM MECHANICS

81

Correspondingly, the topologi al harge is given by


R1

Here  is multiplied in the exponent by the topologi al harge


Z1

1.4.

Dx e S[x+in :

(1.4.70)

The se ond term in the exponent in Eq. (1.4.68) is known as the -term.
The parameter  plays a role of a new fundamental onstant whi h does not
show up in perturbation theory. Amplitudes of physi al pro esses generated
by instantons may depend on the .

Remark on des ription of instantons


A des ription of instantons in the rst-quantized language an be done only
in quantum me hani s (where the rst and se ond quantizations do not di er
essentially). The path-integral representation (1.4.16) is more in the spirit of
se ond quantization, whi h is dis ussed in Se t. 1.2, where x( ) plays the role
of a eld whi h depends on the one-dimensional oordinate  .

Remark on instantons in Yang{Mills theory


In the Yang{Mills theory, instantons are onveniently des ribed by a (Eu lidean) path integral over elds. The saddle-point equation, whi h des ribes
instantons in the SU(2) Yang{Mills theory, reads [BPST75
a (x) = F~ a (x);
F
(1.4.71)

whose non-trivial solutions exist due to the fa t that the mapping of the
asymptoti s S 3 of four-dimensional Eu lidean spa e onto SU(2) is nontrivial:

3 (SU(2)) = Z :

(1.4.72)

n =

g2
322

d4 x

3
X

a=1

a (x)F~ a (x) ;
F


(1.4.73)

whi h equals one half of the non onservation of the axial harge given by the
Minkowski-spa e integral of the hiral anomaly (1.3.64). This expression is
also known in topology as the Pontryagin index or the se ond Chern lass.
See, e.g., Refs. [Col77, VZNS82 for an introdu tion to instantons in the
Yang{Mills theory.

82

CHAPTER 1.

PATH INTEGRALS

1.5 Referen e guide


The operator formalism in quantum eld theory is des ribed in the anoni al
books [AB69,BS76,BD65 whi h were written in the fties or the beginning
of the sixties.
The Feynman disentangling is ontained in the original paper [Fey51
whose appendi es are espe ially relevant. A lassi al book on path integrals in quantum me hani s is the one by Feynman and Hibbs [FH65. The
path-integral approa h to a very losely related problem of Brownian motion
is dis ussed in the books [Ka 59, S h81, Wie86. Many useful information on
path integrals an be found in the book by Kleinert [Kle95.
An introdu tion to path integrals in quantum me hani s and quantum
eld theory an be found in many books. I shall list some whi h I have on
my bookshelf: [Ber86, Pop76, SF88, IZ80, Ram81, Sak85. The ordering is
a ording to the appearan e of the rst edition. The book by Berezin [Ber86,
whi h pretends to be mathemati ally more rigorous, ontains an ex ellent
des ription of operations with Grassmann variables.
An introdu tion to path integrals in statisti al me hani s an be found in
Refs. [Ka 59, Fey72, Pop76, Wie86. The well-written book by Parisi [Par88
des ribes a modern view on the relation between statisti al me hani s and
quantum eld theory. A very good, while a bit more advan ed, book where
ontemporary problems of quantum eld theory and statisti al me hani s are
dis ussed in an uni ed language of Eu lidean path integrals is the one by
Polyakov [Pol87.
The derivation of quantum anomalies from a non-invarian e of the measure
in path integral is ontained in the original papers [Ver78, Fuj79 (see also
the review [Mor86). It an also be found in Se t. 8.9 of the Se ond edition
of the book by Ramond [Ram81.
Instantons in the Yang{Mills theory were dis overed by Belavin, Polyakov,
S hwartz and Tyupkin [BPST75. The role of instantons in quantum me hani s is lari ed in the original paper by Polyakov [Pol77. Their des ription is
given in the books [Sak85, Pol87. The review arti les [Col77, VZNS82 are
also useful for an introdu tion to the subje t.

Referen es to Chapter 1
[AB69

A.I. Akhiezer and V.B. Berestetskii, Quantum ele trodynami s,


Nauka, M., 1969 (English transl.: Oldbourne Press, L., 1962).
[ACD77 S.L. Adler, J.C. Collins and A. Dun an, Energy-momentum-tensor
tra e anomaly in spin-1/2 quantum ele trodynami s, Phys. Rev.
D15 (1977) 1712.
[BD65 J.D. Bjorken and S.D. Drell, Relativisti quantum elds, M GrawHill, N.Y., 1965 (Russian transl.: Nauka, M., 1978).
[Ber86 F.A. Berezin, The method of se ond quantization, Nauka, M., 1986
(English transl.: A ad. Press, N.Y., 1966).
[BPST75 A.A. Belavin, A.M. Polyakov, A.S. S hwartz and Yu.S. Tyupkin,
Pseudoparti le solutions of the Yang{Mills equations, Phys. Lett.
B59 (1975) 85.
[BS76
N.N. Bogoliubov and D.V. Shirkov, Introdu tion to the theory of
quantized elds, Nauka, M., 1976 (English transl.: Inters ien e,
N.Y., 1959).
[CDJ77 J.C. Collins, A. Dun an and S.D. Joglekar, Tra e and dilatation
anomalies in gauge theories, Phys. Rev. D16 (1977) 438.
[CE72 M.S. Chanowitz and J. Ellis, Canoni al anomalies and broken s ale
invarian e, Phys. Lett. 40B (1972) 397.
[Col77 S. Coleman, The uses of instantons, in Pro . of Eri e Int. S hool of
Subnu lear Physi s 1977, Plenum, N.Y., 1979, p. 805 (Reprinted
in S. Coleman, Aspe ts of symmetry, Cambridge Univ. Press, 1985,
pp.265{350).
[Cre72 R. Crewther, Nonperturbative evaluation of the anomalies in lowenergy theorems, Phys. Rev. Lett. 28 (1972) 1421.
[Dir58 P.A.M. Dira , The prin iples of quantum me hani s, Oxford Univ.
Press, 1958 (Russian transl.: Nauka, M., 1979).
83

84

REFERENCES TO CHAPTER 1

[DNF86 B.A. Dubrovin, C.P. Novikov and A.T. Fomenko, Modern geometry. Methods and appli ations, Nauka, M., 1986 (English transl.:
Springer, N.Y., Part I 1984, Part II 1985).
[Dys49 F.J. Dyson, The radiation theories of Tomonaga, S hwinger, and
Feynman, Phys. Rev. 75 (1949) 486.
[Fey50 R.P. Feynman, Mathemati al formulation of the quantum theory
of ele tromagneti intera tion, Phys. Rev. 80 (1950) 440.
[Fey51 R.P. Feynman, An operator al ulus having appli ations in quantum ele trodynami s, Phys. Rev. 84 (1951) 108.
[Fey72 R.P. Feynman, Statisti al me hani s, Benjamin, Mass., 1972 (Russian transl.: Mir, M., 1975).
[FH65 R.P. Feynman and A.R. Hibbs, Quantum me hani s and path integrals, M Graw-Hill, N.Y., 1965 (Russian transl.: Mir, M, 1968).
[Fuj79 K. Fujikawa, Path integral measure for gauge invariant fermion
theories, Phys. Rev. Lett. 42 (1979) 1195.
[GD75 I.M. Gelfand and L.A. Di key, Asymptoti s of development of the
Sturm{Liouville equations and the Korteveg{de Vries algebra, Usp.
Mat. Nauk 30 (1975) v.5, pp.67-100 (English transl.: Russ. Math.
Surv. 30 (1975) v.5, 77).
[IZ80
C. Itzykson and J.-B. Zuber, Quantum eld theory, M Graw-Hill,
N.Y., 1980 (Russian transl.: Mir, M., 1984).
[Ka 59 M. Ka , Probability and related topi s in the physi al s ien e, Inters ien e, N.Y., 1959 (Russian transl.: Mir, M., 1972).
[Kle95 H. Kleinert, Path integrals in quantum me hani s, statisti s, and
polymer physi s, 2nd Ed., World S i., 1995.
[Lan67 J.S. Langer, Theory of the ondensation point, Ann. Phys. 41
(1967) 108.
[Lev51 P. Levy, Probleems on rets d'analyse fon tionnelle, Paris, 1951
(Russian transl.: Nauka, M., 1967).
[LL74
L.D. Landau and E.M. Lifshits, Quantum me hani s. Nonrelativisti theory, Nauka, M., 1974 (English transl.: Pergamon, N.Y.,
1977).
[MM81 Yu.M. Makeenko and A.A. Migdal, Quantum hromodynami s as
dynami s of loops, Nu l. Phys. B188 (1981) 269 (Russian version:
Yad. Fiz. 32 (1980) 838).
[Mor86 A.Yu. Morozov, Anomalies in gauge theories, Usp. Fiz. Nauk 150
(1986) 337 (English transl.: Sov. Phys. Usp. 29 (1986) 993).

REFERENCES TO CHAPTER 1

[Nie77

85

N.K. Nielsen, Energy-momentum tensor in a non-Abelian quark


gluon theory, Nu l. Phys. B120 (1977) 212.
[Par88 G. Parisi, Statisti al eld theory, Addison-Wesley, 1988.
[Pol77 A.M. Polyakov, Quark on nement and topology of gauge groups,
Nu l. Phys. B120 (1977) 429.
[Pol87 A.M. Polyakov, Gauge elds and strings, Harwood A ad. Pub.,
1987.
[Pop76 V.N. Popov, Fun tional integrals in quantum eld theory and statisti al physi s, Atomizdat, M., 1976 (English transl.: Reidel, Dordre ht, 1983).
[Ram81 P. Ramond, Field theory. A modern primer, Benjamin/Cummings,
1981 (Russian transl.: Mir, M., 1984); 2nd ed. Addison-Wesley,
1989.
[Sak85 B. Sakita, Quantum theory of many-variable systems and elds,
World S i., 1985.
[S h58 J. S hwinger, Sele ted papers on quantum ele trodynami s, Dover,
N.Y., 1958.
[S h81 L.S. S hulman, Te hniques and appli ations of path integration,
Wiley, N.Y., 1981.
[SF88
A.A. Slavnov, L.D. Faddeev, Introdu tion to quantum theory of
gauge elds, Nauka, M., 1988 (English transl.: Benjamin, Mass.,
1980).
[Ver78 S.N. Vergeles, unpublished, as quoted in A.A. Migdal, E e tive
low-energy Lagrangian for QCD, Phys. Lett. 81B (1979) 37.
[VZNS82 A.I. Vainshtein, V.I. Zakharov, V.A. Novikov and M.A. Shifman, ABC of instantons, Usp. Fiz. Nauk 136 (1982) 553 (English
transl.: Sov. Phys. Usp. 25 (1982) 195).
[Wie23 N. Wiener, Di erential spa e, J. Math. Phys. 2 (1923) v.3, pp.131{
174.
[Wie86 F.W. Wiegel, Introdu tion to path-integral methods in physi s and
polymer s ien e, World S i., 1986.

86

REFERENCES TO CHAPTER 1

\I never said it."


\Now you are telling us when you did say
it. I'm asking you to tell us when you
didn't say it."

, Cat h-22

J. Heller

Chapter 2

Latti e Gauge Theories


Latti e gauge theories in their modern form were proposed in 1974 by Wilson [Wil74 in onne tion with the problem of quark on nement in quantum
hromodynami s (QCD).
Latti e gauge theories are a non-perturbative regularization of a gauge
theory. The latti e formulation is a non-trivial de nition of a gauge theory
beyond perturbation theory. The problem of non-perturbative quantization
of gauge elds is solved in a simple and elegant way on a latti e.
The use of the latti e formulation lari es an analogy between quantum
eld theory and statisti al me hani s. It o ers a possibility to apply the nonperturbative methods, su h as the strong oupling expansion or the numeri al
Monte Carlo method, to quantum hromodynami s and to other gauge theories, whi h provide an eviden e for quark on nement.
However, the latti e in QCD is no more than an auxiliary tool to obtain
results for the ontinuum limit. In order to pass to the ontinuum, the latti e
spa ing should be many times smaller than a hara teristi s ale of the strong
intera tion.
We shall start this Chapter by a des ription of the ontinuum formulation
of non-Abelian gauge theories, and will return to it from time to time when
dis ussing the latti e approa h. The point is that some ideas, e.g. about a
possibility of reformulating gauge theories in terms of gauge invariant variables, whi h were originally introdu ed by Wilson [Wil74 on a latti e, are
appli able for the ontinuum theory as well.

87

88

CHAPTER 2.

LATTICE GAUGE THEORIES

2.1 Observables in gauge theories


Modern theories of fundamental intera tions are gauge theories. The prin iple of lo al gauge invarian e was introdu ed by H. Weyl for ele tromagneti
intera tion in analogy with general ovarian e in Einstein's theory of gravitation. An extension to non-Abelian gauge groups was given by Yang and
Mills [YM54.
A ru ial role in gauge theories is played by the phase fa tor whi h is
asso iated with a parallel transport in an external gauge eld. The phase
fa tors are observable in quantum theory, in ontrast to the lassi al theory.
For the ele tromagneti eld, this is known as the Aharonov{Bohm e e t.
We rst onsider in this Se tion the matrix notation for the non-Abelian
gauge elds and introdu e proper non-Abelian phase fa tors. Then we dis uss
the relation between observables in lassi al and quantum theories.

2.1.1 Gauge invarian e


The prin iple of lo al gauge invarian e deals with the gauge transformation
of a matter eld , whi h reads
(x)

g.t.

0 (x) =
(x) (x) :

while an extension to other Lie groups is obvious. Then we have


0 y (x) = y (x)
y (x) :

(2.1.3)

In analogy with QCD, the gauge group G = SU(N ) is usually asso iated
with olor while the proper index of is alled the olor index.
The gauge transformation (2.1.1) of the matter eld an be ompensated
by a transformation of the non-Abelian gauge eld A whi h belongs to the
adjoint representation of G:

A (x)

g.t.

! A0 (x) =
(x) A (x)
y (x)

OBSERVABLES IN GAUGE THEORIES

(x) 
y (x) : (2.1.4)

89

We have introdu ed in Eq. (2.1.4) the anti-Hermitean matrix A (x) with


the elements
[A (x)ij = ig

Aa (x) [ta ij :

(2.1.5)

Here [ta ij are the generators of G (a = 1; : : : ; N 2 1 for SU(N )) whi h are


normalized so that
1
tr ta tb = ab ;
(2.1.6)
2
where tr is the tra e over the matrix indi es i and j , while g is the gauge
oupling onstant. This normalization is due to histori al reasons, in parti ular
a
(2.1.7)
ta =
2
for the SU(2) group, with a being the Pauli matri es.
Equation (2.1.5) an be inverted to give

Aa (x) =

(2.1.1)

Here
(x) 2 G with G being a semisimple Lie group whi h is alled the gauge
group (G = SU(3) for QCD). Equation (2.1.1) means that belongs to the
fundamental representation of G.
We restri t ourselves to a unitary gauge group when

1 (x) =
y (x) ;
(2.1.2)
y (x) g.t.
!

2.1.

2
tr A (x) ta :
ig

(2.1.8)

Substituting

(x) = e i (x) ;

(2.1.9)

we get for an in nitesimal :

A (x) g.t.
= i radj
 (x):

(2.1.10)

radj
   [A ;

(2.1.11)

Here
is the ovariant derivative in the adjoint representation of G while

 
rfun


A

(2.1.12)

is the one in the fundamental representation. It is evident that

radj
 B (x)

= [rfun
 ; B (x)

(2.1.13)

90

CHAPTER 2.

LATTICE GAUGE THEORIES

where B (x) is a matrix-valued fun tion of x.


The QCD a tion reads in the matrix notation as

A; ; 

where

d4 x

  (

A )

+m

1
2 ; (2.1.14)
tr F
2g2

F

or

F (x) g.t.


=
i [F (x); (x)
(2.1.17)
for the in nitesimal gauge transformation.
For the Abelian group G = U(1), the above formulas re over those of
the previous Chapter for QED where we have already used the alligraphi
notations in Problem 1.24.
Rewrite lassi al equations of motion in the matrix notations.
The non-Abelian Maxwell equation and the Bian hi identity read, respe tively, as
(2.1.18)
radj
 F = 0
and
radj
(2.1.19)
 F~ = 0 ;
where the dual tensor is de ned by Eq. (1.3.50). Rewriting Eq. (2.1.15) as
(2.1.20)
F = [rfun ; rfun

and using Eq. (2.1.13), we represent the Bian hi identity as
fun
fun
 [rfun
(2.1.21)
 ; [r ; r = 0
whi h is obviously satis ed due to the Ja obi identity.
We have thus proven the well-known fa t that the Bian hi identity is expli itly
satis ed in the se ond-order formalism, where F is expressed via A by virtue of
Eq. (2.1.15). On the ontrary, A and F are onsidered as independent variables
in the rst-order formalism, where both equations (2.1.18) and (2.1.19) are essential.
The on ept of the rst- and se ond-order formalisms omes from the theory of
gravity.
Problem 2.1

91

OBSERVABLES IN GAUGE THEORIES

2.1.2 Phase fa tors (de nition)

=  A  A [A ; A
(2.1.15)
is the (anti-Hermitean) matrix of the non-Abelian eld strength.
The a tion (2.1.14) is manifestly invariant under the lo al gauge transformation (2.1.1), (2.1.4) sin e
F (x) g.t.
!
(x) F (x)
y (x)
(2.1.16)

Solution

2.1.

In order to ompare the phases of wave fun tions at distin t points, one needs
a non-Abelian extension of the parallel transporter whi h was onsidered in
Subse t. 1.1.7. The proper extension of the Abelian formula (1.1.154) is
written as
R
dz A (z)
U [ yx = P e yx
:
(2.1.22)
While the matri es A (z ) do not ommute, the path-ordered exponential
on the RHS of Eq. (2.1.22) is unambiguously de ned by the general method of
Subse t. 1.1.3. This is obvious after rewriting the phase fa tor in an equivalent
form
R
R1

dz A (z)
P e yx
= P e 0 d z_ ()A (z()) :
(2.1.23)
as1

Therefore, the path-ordered exponential in Eq. (2.1.22) is to be understood

U [ yx =


Y
t=0

[1 + dt z_  (t)A (z (t)) :

(2.1.24)

We already used this notation for the produ t on the RHS in Problem 1.8.
Using Eq. (1.1.153), Eq. (2.1.24) an also be written as

U [ yx =

y
Y
x

[1 + dz  A (z ) :

(2.1.25)

If the ontour yx is dis retized as is shown in Fig. 1.3, then the nonAbelian phase fa tor is approximated by
M
Y

1 + (zi zi 1 ) A
U [ yx = Mlim
!1 i=1

zi + z i
2



; (2.1.26)

whi h obviously reprodu es (2.1.25) in the limit  ! 0.


Noti e that the non-Abelian phase fa tor (2.1.22) is, by onstru tion, an
element of the gauge group G itself, while A belongs to the Lie algebra of
G.
1 Sometimes the phase fa tor is de ned by a similar formula but with the inverse order
of multipliers. Our de nition by Eq. (2.1.24) is exa tly equivalent to the Dyson's de nition
of the P{produ t (see the footnote on p. 2) whi h an be seen by hoosing the ontour yx
to oin ide with the temporal axis.

92

CHAPTER 2.

LATTICE GAUGE THEORIES

Problem 2.2 Write down an expli it expansion of the non-Abelian phase fa tor
(2.1.22) in A .
Solution Let us use the notation

Zy

dz  : : :

dz  : : :

Ry 
x dz A (z)
1 Zy
X

as

k=0 x

dz 1

Zy

Zy

For the Abelian ase, when A(zi ) ommute, Eq. (2.1.32) results in
Zy

zk 1

z1

dt1

dt1

dt2 : : :

t1

0
Z

Z

Z
0

dt2 : : :

Z

dzkk Ak (zk )    A2 (z2 )A1 (z1 ) :

(2.1.28)

tk 1

dtk (tk ; tk 1 ; : : : ; t2 ; t1 ) z_ 1 (t1 )z_ 2 (t2 )    z_ k (tk )

Ak (z(tk ))    A2 (z(t2 ))A1 (z(t1 )) ;

(2.1.29)

where
1

tk 2 )    (t2 t1 ) (2.1.30)

orders the points along the ontour. We shall also denote this theta in a parametrization independent form as

(k; k 1; k 2; : : : ; 2; 1)

Zy

1
k!

dzkk Ak (zk )    A2 (z2 )A1 (z1 )

zk 1
1k
0 y
Z
 dz  A (z )A

(2.1.33)

Problem 2.3 Disentangle the non-Abelian phase fa tor using a path integral over
Grassmann variables living in a ontour.
Solution Let us de ne the average

 (tk ; tk 1 ; tk 2 ; : : : ; t2 ; t1 ) :

(2.1.31)

It satis es the obvious identity

(k; k 1; k 2; : : : ; 2; 1) + (k 1; k; k 2; : : : ; 2; 1)
+ (other permutations of k; : : : ; 1) = 1:

F [ ; 

  _

D (t)D (t) e 0 dt (t) (t) (0) (0) F [ ; 
R
: (2.1.34)
R
 _

D (t)D (t) e 0 dt (t) (t) (0) (0)

The path integral in this formula looks like those of Se t. 1.2 with i (t) and j (t)
being Grassmann variables whi h depend on the one-dimensional variable t 2 [0; 
that parametrizes a ontour, and i and j are the olor indi es.
The simplest average, whi h des ribes a propagation of the olor indi es along
the ontour, is


(2.1.35)
= ij (t2 t1 ) ;
0  t 1 ; t2   :
i (t2 ) j (t1 )

dtk z_ 1 (t1 )z_ 2 (t2 )    z_ k (tk )

(tk ; tk 1 ; tk 2 ; : : : ; t2 ; t1 ) = (tk tk 1 ) (tk

dz 2 : : :

so that the Abelian exponential of the ontour integral is reprodu ed.

Ak (z(tk ))    A2 (z(t2 ))A1 (z(t1 ))

Z

Zy

z1

The ordered integral in this formula an be rewritten in a more symmetri form


Z

dz 1
1

yx

dz 2 : : :

93

OBSERVABLES IN GAUGE THEORIES

(2.1.27)

for the integral along the ontour yx . Then we get


P

2.1.

This an be easily he ked, say, by deriving the S hwinger{Dyson equation



(t )  (t ) = (1) (t t ) ;
0 < t 1 ; t2 <  :
(2.1.36)
ij
2
1
t2 i 2 j 1
as was done in Se t. 1.3. We see now that we need the Grassmann variables be ause
the operator in the a tion in Eq. (2.1.34) is =t.
A spe ial omment is needed on erning the term (0) (0) in the exponent
whose appearan e in the disentangling pro edure is lari ed in Ref. [HJS77. The
need for this term an be seen from the dis retized version of the exponent:
Z 
0

dt (t) _ (t) + (0) (0)

M
X
n=1

(n) [ (n)

(n ) + (0) (0):(2.1.37)

For this dis retization we immediately get


(2.1.32)

i (n) j (m)

ij
0

for n  m ;
for n < m :

(2.1.38)

94

CHAPTER 2.

LATTICE GAUGE THEORIES

The term (0) (0) is needed to provide nonvanishing integrals over (0) and (0).
It is also seen from the dis retized version that the path integral in the denominator
on the RHS of Eq. (2.1.34) is equal to unity.
The fermioni path-integral representation for the non-Abelian phase fa tor
reads (see, e.g., Ref. [GN80) as


R


e 0 dt z_ (t)A (z(t))

ij

 R

 
e 0 dt z_ (t) (t)A (z(t)) (t) i ( ) j (0)

:(2.1.39)

There is no path-ordering sign on the RHS sin e the matrix indi es of A are
onta ted by and .
In order to prove Eq. (2.1.39), one expands the exponential in A and al ulates
the average using Eq. (2.1.35) and the rules of Wi k's pairing whi h yields
1
k!
=

Z

0 
Z
i ( ) 

dt1

Z
0

1k

dt z_  (t) (t)A (z (t)) (t)A j (0)

dt2 : : :

Z

 [Ak (z(tk ))    A2 (z(t2 ))A1 (z(t1 ))ij ;

(2.1.40)

Invert ( r2 + m2 ) when r is in the fundamental representation.


Solution The al ulation is quite analogous to that of the Problem 1.12. We rst
use the path-integral representation of the inverse operator:
Problem 2.4




fun + m2 x
rfun
r
 

1 m2
2

Dz (t) e

2
1 
2 0 dt z_ (t)

z (0)=x



R z( )

 rfun
dz

x : (2.1.41)

y P e x

The integral over z ( ) | the nal point of the traje tory | of the matrix element
on the RHS equals
Z

dd z ( )

G(x; y ; A) =
where

P0


R z( )  fun 

y P e x dz r x

Ry 
dz A (z)
:
Pe x

X0

R
P

yx

yx

dz A (z)

(2.1.43)

is de ned by Eq. (1.1.152).


Invert ( r2 + m2 ) when r is in the adjoint representation.
Let us introdu e

Problem 2.5
Solution

ab gf ab A
rab
 = 

ab
and the Green fun tion G (x; y ; A) whi h obeys
b
2 ab  bd
ad (d)
ra
 r + m G (x; y ; A) = (x y ) :
X0
yx

2 tr tb U [ yx ta U y [ yx ;

(2.1.44)
(2.1.45)
(2.1.46)

where U [ yx is given by Eq. (2.1.22).

where (; tk ; : : : ; t2 ; t1 ; 0) is given by Eq. (2.1.30). It is ru ial in the derivation of


this formula that only onne ted terms ontribute to the average (2.1.34). Equation (2.1.40) reprodu es Eq. (2.1.28) from the previous Problem, whi h ompletes
the proof of Eq. (2.1.39). Moreover, we an say that the path integral (2.1.34) is
nothing but a ni e representation of the thetas (2.1.30).

G(x; y ; A) 
Z1
1
d e
=
2

Therefore, the result reads

Then we get

dtk (; tk ; : : : ; t2 ; t1 ; 0) z_ 1 (t1 )z_ 2 (t2 )    z_ k (tk )

95

OBSERVABLES IN GAUGE THEORIES

Gab (x; y ; A) =



y

2.1.

(2.1.42)

Sin e matri es are rearranged in an inverse order under Hermitean onjugation, one has2
U y [ yx = U [ xy :
(2.1.47)
In parti ular, the phase fa tors obey the ba ktra king ondition

U [ yx U [ xy = 1 :
(2.1.48)
We have hosen A in the dis retized phase fa tor (2.1.26) at the enter
on the i-th interval in order to satisfy Eq. (2.1.48) at nite .
Problem 2.6 Establish the relation between non-Abelian phase fa tors and the
group of paths.
Solution The group of paths (or loops) is de ned as follows. The elements of the
group are the paths yx . The produ t of two elements zx and yz is the path yx
whi h is a omposition of zx and yz . In other words one passes rst the path zx
and then the path yz . The produ t is denoted as

yz zx =

yx :

(2.1.49)

2 The notation yx means that the ontour is oriented from x to y while xy stands
for the opposite orientation from y to x. In the path-ordered produ t (2.1.25), these two
ontours result in the opposite orders of multiplying the matri es.

96

CHAPTER 2.

LATTICE GAUGE THEORIES

The multipli ation of paths is obviously asso iative but non ommutative. The inverse element is de ned as
1

yx = xy ;
i.e.

(2.1.52)

2.1.3 Phase fa tors (properties)


Under the gauge transformation (2.1.4) the non-Abelian phase fa tor (2.1.22)
transforms as
U [ yx g.t.
!
(y) U [ yx
y (x) :
(2.1.53)
This formula stems from the fa t that
g.t.


1 + dz  A0 (z )

(z + dz ) [1 + dz  A (z )
y (z )

(2.1.54)

under the gauge transformation, whi h an be proven by substituting


Eq. (2.1.4), so that
y (z ) and
(z ) an el in the de nition (2.1.25) at the
intermediate point z .
One of the onsequen es of Eq. (2.1.53) is that (x), transported by the
matrix U [ yx to the point y, transforms under the gauge transformation as
(y):

U [ yx (x) g.t.
 \ (y)" ;

(2.1.55)

and, analogously,
(y) U [ yx g.t.
 \ (x)" :
Therefore, U [ yx is, indeed, a parallel transporter.

(y) U [ yx (x) g.t.


! (y) U [ yx (x) :

(2.1.57)

(2.1.58)

These properties of the non-Abelian phase fa tor are quite similar to those
of the Abelian one whi h was onsidered in Subse t. 1.1.7.
Cal ulate U [ yx =x and U [ yx =y .
It is onvenient to start from Eq. (2.1.26). Then only (z1 x) in the
last element of the produ t should be di erentiated with respe t to x or (y zM 1 )
in the rst element of the produ t should be di erentiated with respe t to y . As
 ! 0, we get
Problem 2.7

Solution

 P e xy dz A (z) = P e xy dz A (z) A (x) ;



x
Ry 
Ry 
 P e x dz A (z) = A (y ) P e x dz A (z) :
(2.1.59)

y
These formulas are exa tly the same as if one were to just di erentiate the lower and
upper limit in the path-ordered integral keeping in mind the ordering of matri es.
One an rewrite Eq. (2.1.59) via the ovariant derivatives as

rfun (y) U [ yx = 0 ;
U [ yx r fun
 (x) = 0 :

(2.1.60)

It is the property of the parallel transporter whi h is annihilated by the ovariant


derivative.
Prove that the su ient and ne essary ondition for the phase fa tor
to be independent on a lo al variation of the path is the vanishing of F .
Solution Let us add to yx at the point z 2 yx an in nitesimal loop Czz whi h
lies in the ;  -plane and en loses the area  (z ). Then the variation of the phase
fa tor is
Problem 2.8

U [ yx

(2.1.56)



g.t.
tr P e dz A (z) ! tr P e dz A (z) :

(2.1.51)

In other words the paths of opposite orientations an el ea h other in the phase


fa tors.

[1 + dz  A (z )

It follows from these formulas that (y) U [ yx (x) is gauge invariant:


Another onsequen e of Eq. (2.1.53) is that the tra e of the phase fa tor
for a losed ontour is gauge invariant:

The ba ktra king relation (2.1.48) then reads

U [ yx xy = 1 :

97

OBSERVABLES IN GAUGE THEORIES

(2.1.50)

the path with opposite orientation.


It follows from de nition (2.1.25) that

U [ yz U [ zx = U [ yz zx :

2.1.

 U[

yz Czz zx

U [ yx = U [ yz F (z ) U [ zx  (z ) : (2.1.61)

We an rewrite Eq. (2.1.61) as

U [ yx =

U [ yx F (z )  (z )

(2.1.62)

98

CHAPTER 2.

2.1.

OBSERVABLES IN GAUGE THEORIES

99

Remark on analogy with di erential geometry

LATTICE GAUGE THEORIES

The formulas of the type of Eq. (2.1.61) are well-known in di erential geometry where the parallel transport around a small losed ontour determines
the urvature. Therefore, F in Yang{Mills theory is the proper urvature
in an internal olor spa e while A is the onne tion.

6dz

dz 

A histori al remark
x

Fig. 2.1: The re tangular loop Czz , whi h is added to the ontour
intermediate point z in the ;  -plane.

yx at the

sin e the P{produ t will automati ally put F (z ) at the point z of the ontour
yx .
A onvenient way to prove Eq. (2.1.61) is to hoose Czz to be a re tangle
whi h is built on the ve tors dz  and dz  , as is depi ted in Fig. 2.1. Using the
representation (2.1.42), we see that the phase fa tor a quires the extra fa tor
[1 + dz  r [1 + dz  r [1 dz  r [1 dz  r = 1 dz  dz  [r ; r (2.1.63)
at the proper order in the path-ordered produ t. Then Eq. (2.1.20) results in
Eq. (2.1.62). Alternatively, one an prove Eq. (2.1.62) using the dis retized formula (2.1.26).
Derive a non-Abelian version of the Stokes theorem.
The ordered ontour integral an be represented as the double-ordered
surfa e integral [Are80, Bra80

Problem 2.9
Solution

exp

Cxx

dz  A (z ) =

  exp

d  \F (x)"

(2.1.64)

where  and  parametrize the surfa e S (spanned by C but arbitrary otherwise)


whose element is given by


d  = dd z z z z :
(2.1.65)
   
\F (x)" in Eq. (2.1.64) means that F (z (;  )) is parallel-transported, say, to the
initial point x.

An analog of the phase fa tors was rst introdu ed by Weyl [Wey19 in his
attempt to des ribe gravitational and ele tromagneti intera tion of an ele tron on equal footing. What he did is asso iated in the modern language
with the s ale rather than gauge transformation, i.e. the ve tor-potential was
not multiplied by i as in Eq. (1.1.154). This explains the term \gauge invarian e" | gauging literally means xing a s ale. The fa tor of i was inserted
by London [Lon27 after the reation of quantum me hani s and re ognition
of the fa t that ele tromagneti intera tion orresponds to the freedom of a
hoi e of the phase of a wave fun tion and not with a s ale transformation.
However, the terminology was left.

2.1.4 Aharonov{Bohm e e t
The simplest example of a gauge eld is the ele tromagneti eld whose
transversal omponents des ribe photons. Otherwise, the longitudinal omponents of the ve tor-potential, whi h are hangeable under the gauge transformation, are related to gauging the phase of a wave fun tion, i.e. permit to
ompare its values at di erent spa e-time points when an ele tron is pla ed
in an external ele tromagneti eld.
As is well-known in quantum me hani s, the wave-fun tion phase itself
is unobservable. Only the phase di eren es are observable, e.g. via interferen e phenomena. For the ele tron in an ele tromagneti eld, the urrent
(gauged) value of the phase of the wave fun tion at the point y is related,
as is dis ussed in Subse t. 1.1.7, to its value at some referen e point x by
the parallel transport whi h is given by Eq. (1.1.159). Therefore, the phase
di eren e depends on the value of the phase fa tor for a given path yx along
whi h the parallel transport is performed.
It is essential that the phase fa tors are observable in quantum theory,
ontrary to lassi al theory. This is seen in the Aharonov{Bohm e e t. The
orresponding experiment is depi ted s hemati ally in Fig. 2.2.

100

CHAPTER 2.

LATTICE GAUGE THEORIES

2.1.

101

OBSERVABLES IN GAUGE THEORIES

passes through the lower one.


The intensity of the interferen e pattern is given by j (x; y)j2 whi h ontains, in parti ular, the term whi h is proportional to
ele tron  XXbeam

XXXXXz

XXXXX


*

XXX interferen e
solenoid


 plane

H
H 6= 0
sour e

HH

HHH
*
j HH


HH 
ele tron H  beam






s reen
Bohm e e t. Ele trons do not pass inside the solenoid where the magneti
eld is on entrated. Nevertheless, a phase di eren e arises between the
ele tron beams passing through the two slits. The interferen e pi ture
hanges with the value of ele tri urrent.

It allows one to measure the phase di eren e between ele trons passing through the two slits and, therefore, going a ross opposite sides of the
solenoid. The ne point is that magneti eld is nonvanishing only inside the
solenoid where ele trons do not penetrate. Hen e the ele trons pass throughout the region of spa e where the magneti eld strength vanishes! Nevertheless, the ve tor potential A itself does not vanish whi h results in observable
onsequen es.
The probability amplitude for an ele tron to propagate from the sour e at
the point x to the point y at the interferen e plane is given by the Minkowskispa e analog of Eq. (1.1.151):
(x; y) =

yx

where the ontour

+
yx

yx

dz A (z)

yx

dz A (z)

ie

yx

dz A (z)


= e ie dz A (z) ;

(2.1.67)

where the losed ontour is omposed from +yx and xy . This is nothing
but the phase fa tor asso iated with a parallel transport along the losed
ontour .
For the given pro ess this phase fa tor does not depend on the shape of
+ and
yx
yx . Applying the Stokes theorem, one gets
H



e ie d A = eie d F = eieHS ;

(2.1.68)

where HS is the magneti ow through the solenoid. Therefore, the interferen e pi ture hanges when H hanges3.

Remark on quantum vs. lassi al observables

Fig. 2.2: Prin ipal s heme of the experiment whi h demonstrates the Aharonov{

X0 ie

ie

X0 ie

yx

dz A (z)

A moral from the Aharonov{Bohm experiment is that the phase fa tors are
observable in quantum theory while in lassi al theory only the ele tri and
magneti eld strengths are observable. The ve tor potential plays, in lassi al theory, only an auxiliary role to determine the eld strength.
For the non-Abelian gauge group G = SU(N ), a quark an alter its olor
under the parallel transport so the non-Abelian phase fa tor (2.1.22) is a
unitary N  N matrix. A non-Abelian analogue of the quantity, whi h is
measurable in the Aharonov{Bohm experiment, is the tra e of the matrix of
the parallel transport along a losed path. It is gauge invariant a ording to
Eq. (2.1.58).
It looks suggestive to reformulate gauge theories entirely in terms of these
observable quantities. How to do that will be explained in Chapter 3.

(2.1.66)

yx

passes through the upper slit while the ontour yx

3 A detailed omputation of the interferen e pi ture for the Aharonov{Bohm experiment

is ontained, e.g., in the review by Kobe [Kob79.

102

CHAPTER 2.

LATTICE GAUGE THEORIES

2.2 Gauge elds on a latti e


The modern formulation of non-Abelian latti e gauge theories is due to Wilson [Wil74. Independently, gauge theories were dis ussed on a latti e by
Wegner [Weg71 as a gauge invariant extension of the Ising model and in an
unpublished work by A. Polyakov in 1974 whi h deals mostly with Abelian
theories.
Pla ing gauge elds on a latti e provides, rstly, a non-perturbative regularization of ultraviolet divergen ies. Se ondly, the latti e formulation of
QCD possesses some non-perturbative terms in addition to perturbation theory. A result of this is that one has a nontrivial de nition of QCD beyond
perturbation theory whi h guarantees on nement of quarks.
The latti e formulation of gauge theories deals with phase-fa tor like quantities, whi h are elements of the gauge group, and are natural variables for
quantum gauge theories.
The gauge group on the latti e is ompa t therefore o ering the possibility
of non-perturbative quantization of gauge theories without xing the gauge.
The latti e quantization of gauge theories is performed in a way to preserve
the ompa tness of the gauge group.
The ontinuum limit of latti e gauge theories is reprodu ed when the
latti e spa ing is many times smaller than a hara teristi s ale. This is
a hieved when the non-Abelian oupling onstant tends to zero as it follows
from the renormalization-group equation.
We onsider in this Se tion the Eu lidean formulation of latti e gauge
theories. We rst introdu e the latti e terminology and dis uss the a tion
of latti e gauge theory at the lassi al level. Then, we quantize gauge elds
on the latti e by the path-integral method, where the integration is over
the invariant group measure. We explain Wilson's riterion of on nement
and demonstrate it by al ulations in the strong- oupling limit. Finally, we
dis uss how to pass to the ontinuum limit of latti e gauge theories.

2.2.1 Sites, links, plaquettes and all that


The rst step in onstru ting a latti e gauge theory is to approximate the
ontinuous spa e by a dis rete set of points | a latti e. In the Eu lidean
formulation, the latti e is introdu ed along all four oordinates, while the
time is left ontinuous in the Hamiltonian approa h4. We shall dis uss only
4 A Hamiltonian formulation of latti e gauge theories was developed by Kogut and

Susskind [KS75.

2.2.

103

GAUGE FIELDS ON A LATTICE

2q

3q

4q

5q

6q

a6
? a

Fig. 2.3: Two-dimensional latti e with periodi boundary onditions. The sites labelled by the same numbers are identi ed. The latti e spa ing equals a
while the spatial size of the latti e orresponds to L1 = 6 and L2 = 4.

the Eu lidean formulation of latti e gauge theories.


The latti e is de ned as a set of points of the d-dimensional Eu lidean
spa e with the oordinates

x = n  a ;

(2.2.1)

where the omponents of the ve tor

n = (n1 ; n2 ; : : : ; nd )

(2.2.2)

are integer numbers. The points (2.2.1) are alled the latti e sites.
The dimensional onstant a, whi h is equal to the distan e between the
neighboring sites, is alled the latti e spa ing. Dimensional quantities are
usually measured in the units of a, putting therefore a = 1.
A two-dimensional latti e is depi ted in Fig. 2.3. A four-dimensional latti e for whi h the distan es between sites are the same in all dire tions (as
for the latti e in Fig. 2.3) is alled the hyper ubi latti e.
The next on ept is the link of a latti e. A link is a line whi h onne ts two
neighboring sites. A link is usually denoted by the letter l and is hara terized
by the oordinate x of its starting point and its dire tion  = 1; : : : ; d:

l =

fx; g :

(2.2.3)

104

CHAPTER 2.

LATTICE GAUGE THEORIES

l =  + n1 d + n2 dL + : : : + nd dLd 1 ;

x + a^ + a^
e

(2.2.6)

where n = x =a and  is the dire tion of the link fx; g.

p
e

105

Cal ulate the numbers of sites, links and plaquettes for a symmetri
hyper ubi latti e with periodi boundary onditions.
Solution Let us denote L1 = L2 = : : : = Ld = L. Then
d(d 1) d
Nl = dLd ;
Np =
L :
(2.2.5)
Ns = Ld ;
2
Problem 2.11 Label the latti e links by a natural number l 2 [1; Nl .
Solution One of the hoi es is as follows

Fig. 2.4: A link of a latti e. The link onne ts the sites x and x + a^.
e

GAUGE FIELDS ON A LATTICE

Problem 2.10

x + a^

x + a^

2.2.

2.2.2 Latti e formulation


e

x + a^

Fig. 2.5: A plaquette of a latti e. The plaquette boundary is made of four links.
The link l onne ts sites with the oordinates x and x + a^, where ^ is a unit
ve tor along the -dire tion, as shown in Fig. 2.4. The lengths of all links are
equal to a for a hyper ubi latti e.
The elementary square en losed by four links is alled the plaquette. A
plaquette p is spe i ed by the oordinate x of a site and by the two dire tions
 and  along whi h it is built on:

p = fx; ;  g :

(2.2.4)

A plaquette is depi ted in Fig. 2.5. The set of four links whi h bound the
plaquette p is denoted as p.
If the spatial size of the latti e is in nite, then the number of dynami al
degrees of freedom is also in nite (but enumerable). In order to limit the
number of degrees of freedom, one deals with a latti e whi h has nite size
L1  L2     Ld in all dire tions (see Fig. 2.3).
Usually, one imposes periodi boundary onditions to redu e nite-size
e e ts whi h are due to the nite extent of the latti e. In other words, one
identi es pairs of sites whi h lie on parallel bounding hyperplanes. Usually
the sites with the oordinates (0; n2; : : : ; nd ) and (L1 ; n2 ; : : : ; nd) are identi ed
and similarly along other axes.

The next step is to des ribe how matter elds and gauge elds are de ned on
a latti e.
A matter eld, say a quark eld, is attributed to the latti e sites. One
an just think that a ontinuous eld ' (x) is approximated by its values at
the latti e sites5 :

' (x) =) 'x :


(2.2.7)
It is lear that, in order for the latti e eld 'x to be a good approximation
of a ontinuous eld on guration ' (x), the latti e spa ing should be mu h
smaller than a hara teristi size of a given on guration. This is explained
in Fig. 2.6.
The gauge eld is attributed to the links of the latti e:

A (x)

=) Ux; :
(2.2.8)
It looks natural sin e a link is hara terized by a oordinate and a dire tion
(see Eq. (2.2.3)) | the same as A (x). Sometimes the notation U (x) is
alternatively used for Ux; .
The link variable Ux; an be viewed as

Ux; =

R x+a^ 
P e x dz A (z) ;

where the integral is along the link fx; g. As a ! 0, this yields
Ux; ! e aA (x)

(2.2.9)
(2.2.10)

5 We write arguments of fun tions whi h are de ned on a latti e as subs ripts keeping

the standard notation for their ontinuum ounterparts.

106
q

CHAPTER 2.

LATTICE GAUGE THEORIES

b)

a)

Fig. 2.6: Des ription of ontinuum eld on gurations by latti es a) \ oarse" and b)
\ ne". The latti e a) an represent the given ontinuum eld on guration
very roughly, while the latti e b) has the spa ing whi h is small enough.

so that Ux; is expressed via the exponential of the -th omponent of the
ve tor potential, say, at the enter of the link to agree with Eq. (2.1.26).
Sin e the path-ordered integral in Eq. (2.2.9) depends on the orientation,
the on ept of the orientation of a given link arises. The same link, whi h
onne ts the points x and x + a^, an be written either as fx; g or as
fx + a^; g. The orientation is positive for  > 0 in the former ase (i.e.
the same as the dire tion of the oordinate axis) and is negative in the latter
ase.
We have assigned the link variable Ux; to links with positive orientations.
The U -matri es whi h are assigned to links with negative orientations are
given by
(2.2.11)
Ux+a^;  = U y :
x;

This is a one-link analog of Eq. (2.1.47).


It is lear from the relation (2.2.9) between the latti e and ontinuum
gauge variables how to onstru t latti e analogues of the ontinuum phase
fa tors | one should onstru t the ontours from the links of the latti e.
An important role in the latti e formulation is played by the phase fa tor
for the simplest losed ontour on the latti e: the (oriented) boundary of a
plaquette, as is shown in Fig. 2.7. The plaquette variable is omposed from
the link variables (2.2.9) as
Up = U y U y
Ux+a^; Ux; :
(2.2.12)
x; x+a^;

2.2.

107

GAUGE FIELDS ON A LATTICE

x + a^

?
x

x + a^ + a^

x + a^

Fig. 2.7: A ontour in the form of an oriented boundary of a plaquette.


The link variable transforms under the gauge transformation, a ording
to Eq. (2.1.53), as

Ux; g.t.
!
x+a^ Ux;
yx ;
(2.2.13)
where the matrix
x is equal to the value of
(x) at the latti e site x. This
de nes the latti e gauge transformation.
The plaquette variable transforms under the latti e gauge transformation
as
g.t.
Up !
x Up
y :
(2.2.14)
x

Therefore, its tra e over the olor indi es is gauge invariant:


tr Up

g.t.

tr Up :

(2.2.15)

The invarian e of the tra e under the latti e gauge transformation is used
in onstru ting an a tion of a latti e gauge theory. The simplest (Wilson)
a tion is


X
1
(2.2.16)
Re tr Up :
SLat [U =
1
N
p
The summation is over all the elementary plaquettes of the latti e (i.e. over
all x, , and  ), regardless of their orientations.
Sin e a reversal of the orientation of the plaquette boundary results, a ording to Eq. (2.1.47), in omplex onjugation:
reorient
! tr U y = (tr U ) ;
tr U
(2.2.17)
p

p

p

108

CHAPTER 2.

LATTICE GAUGE THEORIES

one an rewrite the a tion (2.2.16) in the equivalent form

SLat [U =

1 X
1
2 oriented p

1
;
tr U
N p

Expanding the exponential on the RHS of Eq. (2.2.19) in a, we get


Z
X
1
2 (x) ;
d4 x
tr F
(2.2.21)
SLat a!!0
4N
;
whi h oin ides modulo a fa tor with the a tion of the ontinuum gauge
theory.
Problem 2.12

(2.1.18).

Solution

Derive the latti e version of the non-Abelian Maxwell equation

: Let us perform the variation of the link variable


y ! (1 + ix; ) U y ;
Ux; ! Ux; (1 ix; ) ;
Ux;
x;

X 

(2.2.18)

exp a2 F (x) + O(a3 ) ;


(2.2.19)
where F (x) is de ned by Eq. (2.1.15).
In the Abelian theory, the expansion (2.2.19) is easily found from the
Stokes theorem. The ommutator of A (x) and A (x), whi h arises in the
non-Abelian ase, omplements the eld strength to the non-Abelian one, as
is ensured by the gauge invarian e. Equation (2.2.19) was in fa t already
derived in Problem 2.8.
The transition to the ontinuum limit is performed by virtue of
X a!0 1 Z
X
! 2 d4 x :
a4
(2.2.20)
p
;

(2.2.22)

where x; is an in nitesimal tra eless Hermitean matrix.


A given link fx; g enters 4(d 1) plaquettes p = fx; ;  g in the a tion (2.2.18).
One half of them has boundary with positive orientation and the other half with
negative one. The variation of the a tion (2.2.18) under the shift (2.2.22) is
i X
y :
tr Up x; tr x; Up
S [U =
(2.2.23)
2N  =6 

109

GAUGE FIELDS ON A LATTICE

Sin e x; is arbitrary, we get

where the sum is also over the two possible orientations of the boundary of a
given plaquette.
In the limit a ! 0, the latti e a tion (2.2.16) be omes in d = 4 the a tion
of a ontinuum gauge theory. In order to show this, let us rst note that

Up

2.2.

 6=

or, graphi ally,


0
B
B
X B
B
B
B
 6= B


U fx;; g Uyfx;; g


? 6
x 

= 0;

(2.2.24)

1
C

 CCCC
? 6 CC
- A

= 0:

(2.2.25)

In the last equation we have depi ted only plaquettes with the positive orientation,
while those with the negative one are re overed by the sum over  for  < 0.
Equation (2.2.24) (or (2.2.25)) is the latti e analog of the non-Abelian Maxwell
equation.
In order to show how this equation reprodu es the ontinuum one (2.1.18) as
a ! 0, let us rewrite the se ond term on the LHS of Eq. (2.2.25) using (2.2.11):
0
B
B
X B
B
B
B
 6= B



? 6
x 


?6? 6
-



x a^

C
C
C
C
C
C
C
A

= 0

(2.2.26)

or, analyti ally,

 6=

U fx;; g


Ux a^; U fx a^;; g Uxy a^; = 0 :

(2.2.27)

It is now lear that the plaquette boundary in the se ond term on the LHS, whi h is
the same as the rst one but transported by one latti e spa ing in the  -dire tion,
is asso iated with F (x a^). Using Eqs. (2.2.10) and (2.2.19), we re over the
ontinuum Maxwell equation (2.1.18).

Remark on the naive ontinuum limit

The limit a ! 0, when Eqs. (2.2.10) and (2.2.19) hold reprodu ing the ontinuum a tion (2.2.21), is alled the naive ontinuum limit. It is assumed in the

110

CHAPTER 2.

LATTICE GAUGE THEORIES

naive ontinuum limit that A (x) is weakly u tuating at neighboring latti e links. Flu tuations of the order of 1=a are not taken into a ount, sin e
dis ontinuities of the ve tor potential in the ontinuum theory are usually
asso iated with an in nite a tion.
Another subtlety with the naive ontinuum limit is that next order in a
terms of the expansion of the latti e a tion (2.2.16), say the term / a2 tr F 3 ,
are asso iated with non-renormalizable intera tions and the smallness of a2
an be ompensated, in prin iple, by quadrati divergen ies.
The a tual ontinuum limit of latti e gauge theories is in fa t very similar
to the naive one modulo some nite renormalizations of the gauge oupling
onstant. The large u tuations of A (x) of the order of 1=a be ome frozen
when passing to the ontinuum limit. How to pass to the ontinuum limit of
latti e gauge theories is explained in Subse t. 2.2.7 below.

Remark on ambiguities of the latti e a tion


The Wilson a tion (2.2.16) is the simplest one whi h reprodu es the ontinuum a tion in the naive ontinuum limit. One an alternatively use hara ters
of Up in other representations of SU(N ), e.g. in the adjoint representation

adj (U ) = jtr U j

1;

(2.2.28)

to onstru t the latti e a tion.


The adjoint-representation latti e a tion reads

Sadj [U =

X

X

1
jtr Up j2 :
N 2

X
1
1
Re tr Up + A
N
2 p

GAUGE FIELDS ON A LATTICE

(2.2.29)

1
jtr Up j2 : (2.2.30)
N 2

The ratio A = is a onstant  1 whi h does not a e t the ontinuum limit.


This a tion is alled the mixed a tion.
The latti e a tion (2.2.29) for N = 2 is asso iated with the a tion of the
SO(3) latti e gauge theory. Sin e algebras of the SU(2) and SO(3) groups

111

oin ide, these two gauge theories oin ide in the ontinuum while they di er
on the latti e.
One more possibility is to use the phase fa tor asso iated, say, with the
boundary of two plaquettes having a ommon link, or the phase fa tors for
more ompli ated losed ontours of nite size on the latti e to onstru t the
a tion. These a tions will also reprodu e, in the naive ontinuum limit, the
a tion of the ontinuum gauge theory.
The independen e of the ontinuum limit of latti e gauge theories on the
hoi e of latti e a tions in alled the universality. We shall say more about
this in the next Se tion when dis ussing renormalization group on the latti e.

2.2.3 The Haar measure


The partition fun tion of a pure6 latti e gauge theory is de ned by

Z ( ) =

Z Y

x;

dUx; e S[U ;

(2.2.31)

where the a tion is given by Eq. (2.2.16).


This is an analog of a partition fun tion in statisti al me hani s at an
inverse temperature given by7

The naive ontinuum limit will be the same as for the Wilson a tion (2.2.16).
Moreover, one an de ne the latti e a tion as a mixture of the fundamental
and adjoint representations [BC81, KM81:

Smixed [U

2.2.

2N
:
g2

(2.2.32)

This formula results from Eq. (2.2.21).


A subtle question is what is the measure dUx; in Eq. (2.2.31). To preserve
the gauge invarian e at nite latti e spa ing, the integration is over the Haar
measure whi h is an invariant group measure. Invarian e of the Haar measure
under multipli ation by an arbitrary group element from the left or from the
right:
dU = d(
U ) = d(U
0 ) ;
(2.2.33)
guarantees the gauge invarian e of the partition fun tion (2.2.31).
This invarian e of the Haar measure is ru ial for the Wilson formulation
of latti e gauge theories.
6 Here \pure" means without matter elds.
7 One has instead = 2N =g2 a4 d on a d-dimensional latti e sin e the Yang{Mills

oupling g is dimensionful for d 6= 4.

112

CHAPTER 2.

LATTICE GAUGE THEORIES

It is instru tive to present an expli it expression for the Haar measure in


the ase of the SU(2) gauge group. An element of SU(2) an be parametrized
by the unit four-ve tor a (a2 = 1) as

U = a4 I + i~a~

(2.2.34)

with ~ being the Pauli matri es. The Haar measure for SU(2) then reads
4
1 Y
dU = 2
da (1) a2
 =1 

1 ;

(2.2.35)

sin e det U = a2 .


Rewrite the Haar measure on SU(2) via a unit three-ve tor ~n ( ~n2 =
1) and an angle ' (' 2 [0; 2 ).
Solution An element of SU(2) reads in this parametrization as
U = e i'~n~=2 = os ' + i~n~ sin ' :
(2.2.36)
2
2
The geometri meaning of this parametrization is simple: the element (2.2.36) is
asso iated with a rotation through the angle ' around the ~n-axis. The Haar measure
for the SU(2) group then is
d2 ~n d' 2 '
dU =
sin :
(2.2.37)
4 
2
This formula an be obtained from Eq. (2.2.35) by integrating over j~aj.
Problem 2.13

For the U(N ) group represent the Haar measure as a multiple


integral over the matrix elements of U .

Solution Elements of a unitary matrix U are omplex numbers Uij = Uji . The
Haar measure an be represented as
Z
Z1 Y

2
dU : : : =
d Re Uij d Im Uij (N ) UU y 1 : : : (2.2.38)
1 i;j
The integral in this formula goes over unrestri ted Uij as if U were a general omplex
matrix while the delta-fun tion restri ts U to be unitary.

Problem 2.14

The partition fun tion (2.2.31) hara terizes va uum e e ts in the quantum theory. Physi al quantities are given by the averages of the type of
Eq. (1.2.6):

h F [U i

= Z

1 ( )

Z Y

x;

dUx; e S[U F [U ;

(2.2.39)

2.2.

GAUGE FIELDS ON A LATTICE

113

where F [U is a gauge invariant fun tional of the link variable Ux; . The averages (2.2.39) be ome the orresponding expe tation values in the ontinuum
theory as a ! 0 and is related to g2 by Eq. (2.2.32).

Remark on the latti e quantization


On a latti e of nite size, the integral over the gauge group in Eq. (2.2.39) is
nite sin e the integration is over a ompa t group manifold, in ontrast to the
ontinuum ase, where the volume of the gauge group is in nite. Therefore,
the expression (2.2.39) is a onstru tive method for al ulating averages of
gauge invariant quantities, though the gauge is not xed.
The gauge an be xed on the latti e in the standard way by the Faddeev{
Popov method. This pro edure involves extra ting a ( nite) ommon fa tor,
whi h equals to the volume of the gauge group, from the numerator and
denominator on the RHS of Eq. (2.2.39). Therefore, the averages of gauge
invariant quantities oin ide for a xed and un xed gauge, while the average
of a fun tional whi h is not gauge invariant vanishes when the gauge is not
xed.
The xing of gauge is onvenient (though not ne essary) for al ulations
in a latti e perturbation theory. A Lorentz gauge an not be xed, however,
outside perturbation theory be ause of Gribov opies [Gri78. In ontrast,
the latti e path integral (2.2.39) with an un xed gauge is a method of nonperturbative quantization.
A pri e for the ompa tness of the group manifold on the latti e is the
presen e of u tuations A (x)  1=a whi h do not o ur in the ontinuum
(say, the values of the ve tor potential A and A + 2=ae are identi ed for
the Abelian U(1) group). However, these u tuations be ome unimportant
when passing to the ontinuum limit.

2.2.4 Wilson loops


As is already mentioned in Subse t. 2.2.2, latti e phase fa tors are asso iated
with ontours whi h are drawn on the latti e.
In order to write down an expli it representation of the phase fa tor on
the latti e via the link variables, let us spe ify the (latti e) ontour C by its
initial point x and by the dire tions (some of whi h may be negative) of the
links whi h the ontour is built of:

C = fx; 1 ; : : : ; n g :

(2.2.40)

114

CHAPTER 2.

(R; 0)

?
-

(0; 0)

LATTICE GAUGE THEORIES

2.2.

segments of the re tangle in Fig. 2.8 ontribute to URT . Denoting

(R; T )

we then have

(0; T )

we get

= Z

1 ( )

Z Y

1
dUx; e S[U tr UC :
N

x;

(2.2.42)

This average is often alled the Wilson loop average.


A very important role in latti e gauge theories is played by the averages of
the Wilson loops asso iated with re tangular ontours. Su h a ontour lying
in the x; t plane is depi ted in Fig. 2.8.
This Wilson loop average is related for T  R to the energy of the
intera tion of the stati (i.e. in nitely heavy) quarks whi h are separated by
the distan e R by the formula
=R e E0 (R)T :
WRT T 
(2.2.43)
Derive Eq. (2.2.43) by xing the gauge A4 = 0.
In the axial gauge A4 = 0, we have Ux;4 = 1 so that only verti al

Problem 2.15
Solution

WRT

(2.2.41)

For the links with a negative dire tion it is again onvenient to use
Eq. (2.2.11).
A losed ontour has ^1 + : : : + ^n = 0. The tra e of the phase fa tor for
a losed ontour, whi h is gauge invariant, is alled the Wilson loop.
The average of the Wilson loop is determined by the general formula (2.2.39) to be

WC

dz1 A1 (z1 ;:::;t)


Pe 0

ij

(2.2.44)

1 tr (0) y (T E
(2.2.45)
ij
ij ) :
N
Inserting in the last equation a sum over a omplete set of intermediate states

Then the latti e phase fa tor UC reads as

1
tr U
N C

RR

WRT =

Fig. 2.8: Re tangular loop of the size R  T .

ij (t) 

UC = Ux+a^1 +:::+a^n 1 ;n    Ux+a^1 ;2 Ux;1 :

115

GAUGE FIELDS ON A LATTICE

jn ih nj = 1 ;

1 tr h (0)j ni
n y (T )
ij
ji
N
n
X 1
=
jh ij (0)j nij2 e En T
N

n

(2.2.46)

(2.2.47)

where En is the energy of the state jni. As T ! 1, only the ground state with the
lowest energy survives in the sum over states and we nally nd
large T
WRT ! e E0 T ;
(2.2.48)
whi h results in Eq. (2.2.43).
Noti e that nothing relies in this derivation on the latti e. Therefore,
Eq. (2.2.43) holds for a re tangular loop in a ontinuum theory as well.

Equation (2.2.43) an also be understood as follows. Let us onsider the


Abelian ase when the intera tion is des ribed by the Coulomb law. The
ontour integral an then be rewritten as the integral over the whole spa e

dz  A (z ) =

dd xJ  (x)A (x) ;

(2.2.49)

where

J  (x) = e

dz  (d) (x z )

(2.2.50)

is a four-ve tor urrent of a lassi al parti le moving along the traje tory C
whi h is des ribed by the fun tion z (t).

116

CHAPTER 2.

LATTICE GAUGE THEORIES

ln W (C ) =

ln e

R
E
i d4 xJ  (x)A (x)

(2.2.51)

determines a hange of the a tion of the lassi al parti le due to ele tromagneti intera tion in a ordan e with Eq. (2.2.43).
A similar interpretation of Eq. (2.2.43) in the non-Abelian ase is somewhat more ompli ated. For a heavy parti le moving along some traje tory
in spa e-time, olor degrees of freedom are quantum and easily respond to
hanges of the gauge eld A (x), whi h intera ts with them. Let us suppose
that a quark and an antiquark are reated at the same spa e-time point in
some olor state. Then this state must be singlet with respe t to olor (or olorless) sin e the average over the gauge eld would vanish otherwise. When
quarks are going away, their olor hanges from one point to another simultaneously with hanging the olor of the gauge eld, in order for the system of
the quarks plus the gauge eld to remain olorless. Therefore, the averaging
over the gauge eld leads to an averaging over u tuations of quark olor
degrees of freedom. E0 (R) in Eq. (2.2.43) is asso iated with the intera tion
energy averaged over olor in this way.
Derive a non-Abelian analog of Eq. (2.2.50).
The proper non-Abelian extension of Eq. (2.2.50) is [Won70

Problem 2.16
Solution

Ja (x) =

Z

dt z_ (t) (d) (x z (t)) I a (t)

(2.2.52)

where I a (t), whi h des ribes the olor state of a lassi al parti le moving along the
traje tory z  (t) in an external Yang{Mills eld A (z ), is a solution of the equation

I_a (t) + gf ab z_  (t) Ab (z (t)) I (t) = 0 :

(2.2.53)

It is onvenient to use again Grassmann variables to des ribe olor degrees of


freedom as in Problem 2.3. Then [BCL77, BSSW77

Ja(x) = (t)ta (t)

(2.2.54)

and (t) is a solution of


_ (t) z  (t) A (z (t)) (t) = 0:

GAUGE FIELDS ON A LATTICE

117

Remark on mass renormalization

It is lear that
D

2.2.

(2.2.55)

By de nition, E0 (R) in Eq. (2.2.43) in ludes a renormalization of the mass of


a heavy quark due to the intera tion with the gauge eld and whi h is thus
independent of R. To the rst order in g2 , it is the same as in QED and reads
Emass =

g2 N 2 1
4a 2N

(2.2.56)

as a ! 0.8
The potential energy of the intera tion between the stati quarks is therefore de ned as the di eren e

E (R) = E0 (R) Emass :

(2.2.57)

If g2 =4a in Emass did not be ome in nite as a ! 0, the term resulting


from the mass renormalization would not have to be subtra ted, sin e it
simply hanges the referen e level for the potential energy.

2.2.5 Strong oupling expansion


We already mentioned in Subse t. 2.2.3 that the path integral (2.2.39) an
be al ulated by the latti e perturbation theory in g2. As was pointed out
by Wilson [Wil74, there exists an alternative way of evaluating the same
quantity on a latti e by an expansion in 1=g2 or in sin e they are related
by Eq. (2.2.32). This expansion is alled the strong oupling expansion. It is
an analog of the high temperature expansion in statisti al me hani s sin e
is an analog of an inverse temperature.
In order to perform the strong oupling expansion, we expand the exponential of the latti e a tion, say in Eq. (2.2.42), in . Then the problem is to
al ulate the integrals over the unitary group of the form
im ;k1 kn
I ij11
jm ;l1 ln

dU Uji11 : : : Ujimm U y kl11 : : : U y klnn ;

(2.2.58)

where the Haar measure (given for SU(2) by Eq. (2.2.35)) is normalized by
Z

dU = 1 :

8 The al ulation is presented below in Problem 3.20.

(2.2.59)

118

CHAPTER 2.

LATTICE GAUGE THEORIES

It is lear from general arguments that the integral (2.2.58) is nonvanishing


only if n = m (mod N ), i.e. only if n = m + kN where k is integer.
For the simplest ase m = n = 1, the answer an easily be found by using
the unitarity of U 's and the orthogonality relation:
Z
1 i k
k
dU Uji U y l =
:
(2.2.60)
N l j
Prove Eq. (2.2.60) for the U(N ) group.
From the general arguments we get
Z

dU Uji U y l

= A li jk + B ji lk :

(2.2.61)

Contra ting by il , using the unitarity of U , and Eq. (2.2.59), we have

AN + B = 1 :
(2.2.62)
One more relation between A and B omes from the fa t that the hara ter in
the adjoint representation is given by Eq. (2.2.28). Contra ting Eq. (2.2.61) by ij
and kl , and using the orthogonality of hara ters whi h says
Z

dU jtr U j2 1

= 1;

(2.2.63)

we get

AN + BN 2 = 1 :
Therefore, A = 1=N and B = 0 whi h proves Eq. (2.2.60).

(2.2.64)

The simplest Wilson loop average, whi h is nonvanishing in the strong


oupling expansion, is the one for the loop whi h oin ides with the boundary
of a plaquette (see Fig. 2.7). It is alled the plaquette average and is denoted
by


1
Wp =
tr U
:
(2.2.65)
N p
In order to al ulate the plaquette average to order , it is su ient to retain only the terms O( ) in the expansion of the exponentials in Eq. (2.2.42):
R Q

Wp =

x;

dUx;

R Q

x;

P
1 + N1 Re tr Up0
p0

dUx; 1 +

1
N tr Up

P 1
N Re tr Up0
p0

GAUGE FIELDS ON A LATTICE

?6

-0
p

-

119

?6p

Fig. 2.9: Boundaries of the plaquettes p and p0 with the opposite orientations: p
and p0 , respe tively.

Problem 2.17
Solution

2.2.

The group integration an then be performed remembering that


Z
 y k
1
i k
dUx; [Ux; ij Uy;
(2.2.67)
l = N xy  l j
at di erent links.
Using this property of the group integral in Eq. (2.2.66), we immediately
see that the denominator is equal to one (ea h link is en ountered no more
than on e), while the only nonvanishing ontribution in the numerator is from
the plaquette p0 , whi h oin ides with p but has the opposite orientation as
is depi ted in Fig. 2.9.
It is onvenient to use the graphi notations9 for Eq. (2.2.60) at ea h link
of p:

i - j = 1   i   j ;
N
l
k
l  k
where the semi ir les are asso iated with the Krone ker deltas:

i  = i :
(2.2.69)
l  l
This notation is onvenient sin e the lines whi h denote the deltas in the last
equation an be asso iated with propagation of the olor indi es. Analogously
a losed line represents the ontra ted delta, whi h is summed up over the
olor indi es,
j

2

+ O : (2.2.66)

(2.2.68)

= ii = N :

(2.2.70)

9 A al ulation of more ompli ated group integrals (2.2.58) in the graphi notations
is dis ussed in the le tures by Wilson [Wil75 and in the Chapter 8 of the book by
Creutz [Cre83. An alternative way of al ulating the group integrals by the hara ter
expansion is des ribed in the review by Drou e and Zuber [DZ83.

120

CHAPTER 2.

LATTICE GAUGE THEORIES

- - 6- ?6- ?6- ?6- - ?6 ?6 ?6 ?6 ?6 ?6 ?
- - - - - - 6
6 ?6 ?6-?6 ?6 ?6 ?

WC = (Wp )Amin(C ) ;


2N 2

Wp =
4

= 1;
(2.2.71)

for SU(N ) with N  3 ;


for SU(2) :

(2.2.73)

to the leading order in .


More ompli ated surfa es, whi h do not lie in the plane of the re tangle,
will give the ontribution to WC of the order of Area . They are suppressed
at small sin e their areas are larger than Amin.

2.2.6 Area law and on nement

where the ontra ted Krone ker deltas are asso iated with the four sites of
the plaquette.
The nal answer for the plaquette average is

Wp =

121

where Wp is given by Eq. (2.2.72) and Amin(C ) is the area (in units of a2 )
of the minimal surfa e.
For the re tangular loop, whi h is depi ted in Fig. 2.8, the minimal surfa e
is just a pie e of the plane bounded by the re tangle. Therefore, we get
WRT = (Wp )R T
(2.2.74)

Using the graphi representation (2.2.68) for ea h of the four links depi ted
in Fig. 2.9, we get
y 0 = 14 
dUx; tr Up tr Up
N
x;

GAUGE FIELDS ON A LATTICE

takes on the smallest possible value. This yields

Fig. 2.10: Filling of a loop with elementary plaquettes.

R Q

2.2.

(2.2.72)

The result for SU(2) di ers by a fa tor of 1=2 be ause tr Up is real for SU(2)
so that the orientation of the plaquettes an be ignored.
The graphi representation (2.2.68) is useful for evaluating the leading
order of the strong oupling expansion for more ompli ated loops. A ording
to Eq. (2.2.67), a nonvanishing result emerges only when plaquettes, oming
from the expansion of the exponentials of Eq. (2.2.42) in , ompletely over
a surfa e en losed by the given loop C as is depi ted in Fig. 2.10. In this ase
ea h link is en ountered twi e (or never), on e in the positive dire tion and
on e in the negative dire tion, so that all the group integrals are nonvanishing.
The leading order in orresponds to lling a minimal surfa e, whose area

The exponential dependen e of the Wilson loop average on the area of the
minimal surfa e (as in Eq. (2.2.73)) is alled the area law. It is ustomarily
assumed that if an area law holds for loops of large area in pure gluodynami s (i.e. in the pure SU(3) gauge theory) then quarks are on ned. In other
words, there are no physi al jini or houtj quark states. This is the essen e of
Wilson's on nement riterion. The argument is that physi al amplitudes,
e.g. the polarization operator, do not have quark singularities when the Wilson riterion is satis ed. I refer the reader to the well-written original paper
by Wilson [Wil74, where this point is lari ed.
Another, somewhat oversimpli ed, justi ation for the Wilson riterion is
based on the relationship (2.2.43) between the Wilson loop average and the
potential energy of intera tion between stati quarks. When the area law

W (C ) large!C e KAmin(C )

(2.2.75)

holds for large loops, the potential energy is a linear fun tion of the distan e
between the quarks:

E (R) = KR :

(2.2.76)

The oe ient K in these formulas is alled the string tension be ause


the gluon eld between quarks ontra ts to a tube or string, whose energy
is proportional to its length, as is depi ted in Fig. 2.11a. The value of K

122

CHAPTER 2.

LATTICE GAUGE THEORIES

2.2.

123

GAUGE FIELDS ON A LATTICE

the gauge eld between quarks would be distributed over the whole spa e as
is depi ted in Fig. 2.11b. The Wilson loop average would have the behavior

W (C )
r

a)

large C

e onstL(C ) ;

(2.2.79)

where L(C ) stands for the length (or perimeter) of the losed ontour C .
This behavior of the Wilson loops is alled the perimeter law. To ea h
order of perturbation theory, it is the perimeter law (2.2.79), rather than the
area law (2.2.75), that holds for the Wilson loop averages. A perimeter law
orresponds to a potential whi h an not on ne quarks.

Remark on the Creutz ratio

b)

Fig. 2.11: Lines of for e between stati quarks for a) linear and b) Coulomb intera tion potentials. For the linear potential the lines of for e are ontra ted into a tube, while they are distributed over the whole spa e for the
Coulomb one.

To distinguish between the area and perimeter law behavior of the Wilson
loop averages, Creutz [Cre80 proposed to onsider the ratio

(I; J ) =

W W
ln I J (I 1)(J
W(I 1)J WI (J

1)
1)

(2.2.80)

is the energy of the string per unit length. This string is stret hed with the
distan e between quarks and prevents them from moving apart to ma ros opi
distan es.
Equation (2.2.74) gives

where WI J is as before the average of a re tangular Wilson loop of the size


I  J . The exponentials of the perimeter, whi h is equal to

1 2N 2
ln
a2


1
= 2 ln N g2
(2.2.77)
a
for the string tension to the leading order of the strong oupling expansion.
Next orders of the strong oupling expansion result in orre tions in to this
formula.
Therefore, on nement holds in the latti e gauge theory to any order of
the strong oupling expansion.

an el out in the ratio (2.2.80). In parti ular, the mass renormalization


(2.2.56) an els out, whi h is essential for the ontinuum limit.
The Creutz ratio (2.2.80) has a meaning of a for e of intera tion between
quarks, whi h an be seen by stret hing the re tangle along the \temporal"
axis (as is illustrated by Fig. 2.8). If the area law (2.2.75) holds for asymptoti ally large I and J , then

K =

Remark on the perimeter law

(I; J )

large I; J

a2 K ;

(2.2.81)

(2.2.82)

i.e. does not depend on I or J and oin ides with the string tension. This
property of the Creutz ratio was used for numeri al al ulations of the string
tension.

2.2.7 Asymptoti s aling

For the Coulomb potential

E (R) =

L(I  J ) = 2I + 2J ;

g2 N 2 1
;
4R 2N

(2.2.78)

Equation (2.2.77) establishes the relationship between values of the latti e


spa ing a and the oupling g2 as follows. Let us set K to be equal to its

124

CHAPTER 2.

LATTICE GAUGE THEORIES

2.2.

experimental value10

K = (400 MeV)2

1 Gev/fm :

(2.2.83)

Then the renormalizability pres ribes that variations of a, whi h plays the role
of a latti e uto , and of the bare harge g2 should be done simultaneously
in order that K does not hange.
Given Eq. (2.2.77), this pro edure alls for a ! 1 as g2 ! 1. In other
words, the latti e spa ing is large in the strong oupling limit, ompared to
1 fm | the typi al s ale of the strong intera tion. This is a situation of the
type shown in Fig. 2.6a. Su h a oarse latti e an not des ribe orre tly the
ontinuum limit and, in parti ular, the rotational symmetry.
In order to pass to the ontinuum, the latti e spa ing a should be de reased
to have a pi ture like that in Fig. 2.6b. Equation (2.2.77) shows that a
de reases with de reasing g2 . However, this formula eases to be appli able
in the intermediate region of g2  1 and, therefore, a  1 fm.
The re ipe for further de reasing of a is the same as in the strong oupling
region: further de reasing of g2 . While no analyti formulas are available at
intermediate values of g2, the expe ted relation between a and g2 for small
g2 is predi ted by the known two-loop Gell-Mann{Low fun tion of QCD.
For pure SU(3) gluodynami s, Eq. (2.2.77) is repla ed at small g2 by


1 162
K = onst  2
a 11g2

102
 121

162
11g2 :

(2.2.84)

The exponential dependen e of K on 1=g2 is alled asymptoti s aling. An


asymptoti s aling sets in for some value of 1=g2 as is depi ted in Fig. 2.12.
For su h values of g2 , where an asymptoti s aling holds, the latti e gauge
theory enjoys a ontinuum limit.
The knowledge of the two asymptoti s says nothing about the behavior
of a2 K in the intermediate region of g2  1. There an be either a smooth
transition between these two regimes or a phase transition. Numeri al methods were introdu ed to study this problem. Some of them are des ribed in
the next Se tion.
10 This value results from the string model of hadrons where the slope of the Regge
traje tory 0 and the string tension K are related by K = 1=2 0 . This formula holds
even for a lassi al string. The slope 0 = 1 Gev 2 say from the  { A2 { g traje tory. A
similar value of K is found from the des ription of mesons made out of heavy quarks by a
nonrelativisti potential model.

125

GAUGE FIELDS ON A LATTICE

strong oupling
ln a2 K

JJ
J asymptoti s aling
JJ
J
1=g2
JJ
JJ

Fig. 2.12: Dependen e of the string tension on 1=g2 . The strong oupling formula
(2.2.77) holds for small 1=g2 . The asymptoti s aling formula (2.2.84) sets
in for large 1=g2 . Both formulas are not appli able in the intermediate
region of 1=g2  1 whi h is depi ted by the dashed line.

Remark on dimensional transmutation


The QCD a tion (2.1.14) does not ontain a dimensional parameter of the
order of hundreds MeV. The masses of the light quarks are of the order of
a few MeV and an be disregarded. The only parameter of the a tion is the
dimensionless bare oupling onstant g2 . At the lassi al level, there is no
way to get the dimensional parameter of the order of hundreds MeV.
In quantum theory, these is always a dimensional uto (like a for the
latti e regularization). The renormalizability says that a and g2 are not
independent but are related by the Gell-Mann{Low equation (1.3.73). It an
be integrated to give the integration onstant
QCD =

1
exp
a

dg2
B(g2) :

(2.2.85)

Up to now there was no di eren e between QCD and QED. The di eren e
stems from the fa t that the Gell-Mann{Low fun tion B(g2 ) is positive for
QED and negative for QCD. In QED e2(a) in reases with de reasing a while in
QCD g2 (a) de reases with de reasing a. The latter behavior of the oupling

126

CHAPTER 2.

LATTICE GAUGE THEORIES

onstant is alled asymptoti freedom. In both ases the Gell-Mann{Low


fun tion vanishes when the oupling onstant tends to zero. Su h values of
oupling onstants where the Gell-Mann{Low fun tion vanishes are alled the
xed points. Sin e the infrared behavior of e2 in QED is inter hangeable with
the ultraviolet behavior of g2 in QCD, the origin is an infrared-stable xed
point in QED and an ultraviolet-stable xed point in QCD. In QED the nestru ture onstant (= 1=137) is measurable in experiments while in QCD the
onstant QCD is measurable.
This phenomenon of an appearan e of a dimensional parameter in QCD,
whi h remains nite in the limit of vanishing uto , is alled dimensional
transmutation. All observable dimensional quantities, say the string tension
or hadron masses, are proportional to the orresponding
p powers of QCD .
Therefore, their dimensionless ratios, say the ratio of K to hadron masses,
are universal numbers whi h do not depend on g2 . The goal of a nonperturbative approa h in QCD is to al ulate these numbers but not the
overall dimensional parameter.

Remark on se ond-order phase transition


It is usually said in statisti al physi s that ontinuum limits of a latti e system
are rea hed at the points of se ond-order phase transitions when the orrelation length be omes in nite in latti e units. This statement is in perfe t
agreement with what is said above about the ontinuum limit of latti e gauge
theories.
A orrelation length is given by a formula similar to Eq. (2.2.85):

1
QCD
= a exp

Z

dg2
B(g2 ) :

(2.2.86)

The only han e for the RHS to diverge is to have a zero of the Gell-Mann{
Low fun tion B(g2) at some xed point g2 = g2 . Therefore, the bare oupling
should approa h the xed-point value g2 to des ribe the ontinuum.
As we have dis ussed, B(0) = 0 for a non-Abelian gauge theory so that
g2 = 0 is a xed-point value of the oupling onstant. Therefore, the ontinuum limit is asso iated with g2 ! 0 as is said above.

2.3.

LATTICE METHODS

127

2.3 Latti e methods


Analyti al ulations of observables in the non-Abelian latti e gauge theories
are available only in the strong oupling regime g2 ! 1, while one needs
g2 ! 0 for the ontinuum limit. When g2 is de reased, the latti e systems
an undergo phase transitions as it often happens in statisti al me hani s.
To look for phase transitions, the mean- eld method was rst applied to
latti e gauge theories [Wil74, BDI74. It turned out to be useful for studying
the rst-order phase transitions whi h very often happen in latti e gauge
systems but do not a e t the ontinuum limit.
The se ond-order phase transitions are better des ribed by the latti e
renormalization group method. The approximate Migdal{Kadano re ursion
relations [Mig75, Kad76 were the rst implementation of the renormalization
group transformation on a latti e, whi h indi ated the absen e of a se ondorder phase transition in the non-Abelian latti e gauge theories and, therefore,
quark on nement.
A most powerful method for pra ti al non-perturbative al ulations of
observables in latti e gauge theories is the numeri al Monte Carlo method.
This method simulates statisti al pro esses in a latti e gauge system and is
often alled for this reason the numeri al simulation. The idea of applying it
to latti e gauge theories is due to Wilson [Wil77 while the pra ti al implementation was done by Creutz, Ja obs and Rebbi [CJR79 for Abelian gauge
groups and by Creutz [Cre79, Cre80 for the SU(2) and SU(3) groups.
We brie y des ribe in this Se tion the mean- eld method, the latti e
renormalization group method and the Monte Carlo method. Few results
of the Monte Carlo simulations will be dis ussed.

2.3.1 Phase transitions


As is pointed out in the previous Se tion, analyti al ulations of the string
tension are available only in the strong oupling regime g2 ! 1, while one
needs g2 ! 0 for the ontinuum limit. A question arises what happens with
latti e systems when g2 is de reased. In parti ular, does an a tual pi ture of
the dependen e of the string tension on g2 look like that in Fig. 2.12?
We know from statisti al me hani s that latti e systems an undergo
phase transitions with hanging the parameters, say the temperature, whi h
ompletely alter ma ros opi properties. A simplest example is that of a
rst-order phase transition whi h o urs in a teapot.
First-order phase transitions very often happen in the latti e gauge theories. They are usually seen as a dis ontinuity in the (or 1=g2) -dependen e of

128

CHAPTER 2.

LATTICE GAUGE THEORIES

2.3.

129

LATTICE METHODS

equation (2.2.24). The proper solution reads

6 Wp

SP = Z ;
Ux;


where Z is an element of the Z(N ) group, the enter of the SU(N ),

Z = I  e 2in =N ;

(2.3.3)

/ 1=g2

Fig. 2.13: Typi al -dependen e of the plaquette average for a rst-order phase transition whi h o urs at =  .

the plaquette average (2.2.65) as is depi ted in Fig. 2.13. The form of Wp ( )
at small is given to the leading order of the strong oupling expansion by
Eq. (2.2.72), while that at large is pres ribed by the latti e perturbation
theory11 to be
dG
Wp ( ) = 1
+ O( 2 ) ;
(2.3.1)
d
where dG is the dimensionality of the gauge group (dG = N 2 1 for SU(N ),
dG = N 2 for U(N )) and d is the dimensionality of the latti e as before.
This behavior of the plaquette average is quite analogous to a dependen e
of the internal energy per unit volume ( alled the spe i energy) in statisti al
systems. In order to see an analogy between the spe i energy and (1 Wp ),
let us remember that is an analog of the inverse temperature and rewrite
Eq. (2.2.65) as
1 
ln Z ( ) ;
(2.3.2)
Wp ( ) = 1 +
Np 
where the partition fun tion is given by Eq. (2.2.31) and the number of plaquettes Np is an analog of the volume of a statisti al system.
Derive Eq. (2.3.1) for the SU(N ) gauge group.
The partition fun tion (2.2.31) an be al ulated at large by the saddlepoint method. The saddle-point on gurations are given by solutions of the lassi al
Problem 2.18

Solution

11 It is often alled, for obvious reasons, the weak oupling expansion.

n = 1; : : : ; N :

(2.3.4)

It is evident that this is a solution be ause elements of the enter ommute so that
Z and Z  an el ea h other in Ux;; .
In order to take into a ount u tuations around the saddle-point solution (2.3.3), let us expand
SP e i ta ax; ;
Ux; = Ux;

(2.3.5)

where the order of multipli ation is inessential sin e Z ommute with the generators
ta . The expansion of tr Up to the quadrati order in a reads
1
1 2
tr U
= 1
E ;
(2.3.6)
N x;;
4N x;;
where

Exa;; = ax; + ax+a^; ax+a^; ax; :

(2.3.7)

Due to the lo al gauge invarian e, we an always hoose, say, x;d = 0 so that there
are only Nl Ns independent 's.
Substituting into Eq. (2.2.31) and expanding the Haar measure, we get
+1
N
d X
Y
Y Z
2
(2.3.8)
Z ( ) /
dax; e Ex;; =4N :
 =1 n =1 x;<d;a 1
The sum over n , whi h is due to the degenerate saddle points, is just an irrelevant
onstant.
We see from Eq. (2.3.8) that only
1
ax;  p
(2.3.9)

are essential whi h justi es the expansion in . Res aling the integration variables
in Eq. (2.3.8), we get, therefore,

Z ( )

(Nl Ns ) dG =2 :

Substituting into Eq. (2.3.2) and remembering that (Nl


Eq. (2.2.5)), we obtain Eq. (2.3.1).

(2.3.10)

Ns )=Np = 2=d (see

130

CHAPTER 2.

LATTICE GAUGE THEORIES

2.3.

Problem 2.19 Repeat the derivation of the previous Problem for the adjoint a tion (2.2.29).
Solution The only di eren e from the Wilson a tion (2.2.16) is that the saddlepoint solution (2.3.3) is now modi ed as

SP = Z ;
Ux;
x;

 n =1

=)

N
d X
Y
x; nx; =1

(2.3.12)

whi h only hanges an irrelevant overall onstant. Therefore, Eq. (2.3.1) remains
un hanged providing the plaquette average is also taken in the adjoint representation. This supports the expe tation that the ontinuum limits for the both a tions
oin ide.

The rst-order phase transitions of the type in Fig. 2.13 are usually harmless and are not asso iated with de on nement. They are related with dynami s of some latti e degrees of freedom (say, with large u tuations of the
link variable Ux; whi h o ur independently at adja ent links) whi h do not
a e t the ontinuum limit and are alled latti e artifa ts. Moreover, these
latti e degrees of freedom be ome frozen for >  whi h is ne essary for
the ontinuum limit to exist. A brief des ription of some latti e artifa ts and
of the related phase transitions an be found in the Se tion 3 of the review
arti le [Mak84.
Another possibility for a latti e system is to undergo a se ond-order phase
transition in analogy with spin systems. In this ase Wp is ontinuous but
the derivative Wp = be omes in nite at the riti al point =  as is
depi ted in Fig. 2.14. Given Eq. (2.3.2), this derivative is to be onsidered
as an analog of the spe i heat of statisti al systems. Its behavior at small
and large is governed by Eqs. (2.2.72) and (2.3.1), respe tively.
Di erentiating Eq. (2.2.65) with respe t to , the derivative Wp = an
be expressed via the sum of the onne ted orrelators:


6 W p

(2.3.11)

i.e. may take on di erent values at di erent links. It is evident that this is a
minimum of the a tion (2.2.29).
The only modi ation of Eq. (2.3.8) is

N
d X
Y

131

LATTICE METHODS

1
1 X
Wp
1
:
tr Up tr Up0
=

2 oriented p0 N
N
onn

(2.3.13)

/ 1=g2

Fig. 2.14: Typi al -dependen e of Wp = for a se ond-order phase transition


whi h o urs at =  .

This formula shows also that Wp = is positive de nite, sin e the RHS an
be rewritten using translational invarian e as


1 X
1
1
tr U
tr U 0
2 oriented p0 N p N p onn
1
=
4Np
 0;

1
tr Up
N
oriented p
X

!2 +

1
4Np

1
tr Up
N
oriented p
X

!+2

(2.3.14)

where the equality is possible only for a Gaussian averaging, i.e. for a free
theory. This repeats the standard proof of the positivity of spe i heat in
statisti al me hani s.
Sin e ea h term of the sum in Eq. (2.3.13) is nite (remember that a tra e
of a unitary matrix takes on values between N and N ), the only possibility
for the RHS to diverge is for the sum over plaquettes p0 to diverge. This is
possible only when long-range (in the units of the latti e spa ing) orrelations
are essential or, in other words, the orrelation length is in nite. Thus, we
have on e again reprodu ed the argument that the ontinuum limit of latti e
theories is rea hed at the points of se ond-order phase transitions.
Su h a se ond-order phase transition seems to o ur in ompa t QED (i.e.
the U(1) latti e gauge theory with fermions) at e2  1. It is asso iated there
with de on nement of ele trons. Ele trons are on ned for e2 > e2 , similar
to quarks in the latti e QCD, and are liberated for e2 < e2. The intera tion

132

CHAPTER 2.

LATTICE GAUGE THEORIES

potential looks like that of Fig. 2.11b for e2 < e2 and like that of Fig. 2.11a in
the on nement region e2 > e2 . 12 In order to rea h the ontinuum limit with
de on ned ele trons, the bare harge e2 should be hosen slightly below the
riti al value. Then the renormalized physi al harge an be made as small as
the experimental value ( = 1=137) a ording to the renormalization group
arguments whi h are presented in the Remarks to Subse t. 2.2.7.
The nature of the phase transition in a four-dimensional ompa t U(1)
latti e gauge theory without fermions was investigated by numeri al methods.
While the very rst paper [LN80 indi ated that the phase transition is of
se ond order, some more advan ed re ent investigations say that it may be
weakly rst order (see, e.g., the talks at the Latti e Conferen e [Lat94, Part
2.7). Anyway, we need fermions whi h usually weaken a phase transition that
happens in a pure latti e gauge theory.
There are no indi ations that a se ond-order phase transition o urs in
non-Abelian pure latti e gauge theories at intermediate values of . This
very strongly supports a behavior of the string tension of the type depi ted
in Fig. 2.12. The se ond-order phase transition o urs in four dimensions at
= 1 (or g2 = 0) a ording to the general arguments of Subse t. 2.2.7,
whi h is ne essary for the ontinuum limit to exist.

2.3.

2.3.2 Mean- eld method


The idea to apply the mean- eld method, whi h is widely used in statisti al systems, to study phase transitions in the latti e gauge theories was
proposed by Wilson [Wil74 and rst implemented for Abelian theories by
Balian, Drou e and Itzykson [BDI74. A mean eld usually works well when
there are many neighbor degrees of freedom, intera ting with a given one.
12 The latter statement is not quite orre t for the reasons whi h are dis ussed in Sub-

se t. 2.5.4.

m
m

Fig. 2.15: Graphi representation of the self- onsisten y ondition (2.3.17). The link
variables are repla ed by m  I at all links ex ept for a given one denoted
by the bold line.

In the simplest version of the mean- eld method, the link variable Ux;
is repla ed by the mean- eld value m  I everywhere but at a given link (see
Fig. 2.15) at whi h the self- onsisten y ondition
D

[Ux;ij

Remark on on nement in 4 +  dimensions


In 4 +  dimensions ( > 0), a se ond-order de on ning phase transition
always o urs in non-Abelian pure latti e gauge theories at some nite value
of < 1 (or g2 > 0). The ase of   1 an be onsidered analogous to
the -expansion in statisti al me hani s [WK74. An ultraviolet-stable xed
point now exists at g2   sin e the theory is asymptoti ally free in d = 4.
This phase transition is asso iated with de on nement quite analogously to
ompa t QED in d = 4. The de on ning phase is asso iated with g < g
while the on ning phase is asso iated with g > g .

133

LATTICE METHODS

= m ij

(2.3.15)

is imposed.
The average on the LHS of Eq. (2.3.15) is al ulated with the a tion whi h
is obtained from (2.2.16) by the substitution of m  I for all the link variables
(or their Hermitean onjugate) ex ept at the given link. Sin e the given link
enters 2(d 1) plaquettes, the average on the LHS of Eq. (2.3.15) is to be
al ulated with the a tion

S0 [U = 2 (d 1) m3 Re tr Ux; + onst :
(2.3.16)
Therefore, the self- onsisten y ondition (2.3.15) an be written by
the substitution of the mean- eld ansatz into the latti e partition fun tion (2.2.31) as
R

with

 Re tr U 1
dU e N
N tr U = m
R

dU e N Re tr U

(2.3.17)


 = 2 (d 1) m3 2 :
N

(2.3.18)

134

CHAPTER 2.

LATTICE GAUGE THEORIES

2.3.

135

LATTICE METHODS

or for the SO(3) group), agreement with numeri ally al ulated positions of
the phase transitions is remarkable.

6m

Cal ulate  for the SU(1) latti e gauge theory, when the group
 2 for   1 (a strong oupling phase)
integral on the LHS of Eq. (2.3.17) equals: =
and 1 1=2  for   1 (a weak oupling phase).
Solution For the strong oupling phase, the self- onsisten y equation

(d 1) m3 2 = m
(2.3.20)
N
has the only solution m = 0. The other solutions are una eptable due to stability
onsideration.
The nontrivial solutions of the self- onsisten y equation appear in the weak
oupling phase when dm=d = 1 or d =dm = 0. Di erentiating, we get in d = 4
Problem 2.20

/ 1=g2

Fig. 2.16: Typi al behavior of the solutions of the self- onsisten y equation (2.3.17).

The only solution with m = 0 exists for <  . Two more solutions
appear for >  . The one depi ted by the dashed line is unstable. The
a tual value of m versus is depi ted by the bold lines. A rst-order phase
transition is asso iated with =  .

The meaning of Eq. (2.3.17) is very simple: the average of the tra e of the
link variable at the given link should oin ide with m whi h is substituted for
all other links of the latti e.
In order to verify whether the self- onsisten y ondition (2.3.17) admits
nontrivial solutions, one should rst al ulate the group integral on the LHS
and then solve the self- onsisten y equation for m versus . A typi al behavior of the solution is depi ted in Fig. 2.16. For all values of , there exists a
trivial solution m = 0 whi h is asso iated with no mean eld. At some value
 , two more solutions of the self- onsisten y equation appear. The upper of
them is asso iated with positive spe i heat while the lower one orresponds
to negative one. This an be seen by noting that

Wp = m4

(2.3.19)

in the mean- eld approximation whi h follows from the substitution of the
link-variables in the de nition (2.2.65) by the mean- eld values. This nontrivial solution is preferred for >  sin e the partition fun tion for it is
larger (or the free energy is smaller) than for the m = 0 solution. The value
of  is often asso iated with the point of a rst-order phase transition.
The mean- eld method in su h a simple form was rst applied to nonAbelian latti e gauge theories in Ref. [GL81, CGL81. For the ases when a
rst-order phase transition o urs (say, for the SU(N ) groups with N > 3

 1 N 2 = 12 3m2
m

4m3 ;

(2.3.21)

whi h yields

 = 43 = 0; 79 :
m = 3 ;
(2.3.22)
4
N 2
34
It is still left to verify that the proper  is indeed asso iated with the weak oupling
phase. From Eq. (2.3.18), we get  = 2 and this is the ase.

2.3.3 Mean- eld method (variational)


There are some puzzles with the simplest mean- eld ansatz des ribed above.
First of all, the average value of the link variable Ux; in a latti e gauge theory must vanish due to the gauge invarian e (remember that Ux; hanges
under the gauge transformation a ording to Eq. (2.2.13) while the a tion
and the measure are gauge invariant). This is in a ordan e with Elitzur's
theorem [Eli75 whi h says that a lo al gauge symmetry an not be spontaneously broken, so that any order parameter for phase transitions in latti e
gauge theories must be gauge invariant.
A way out is to reformulate the mean- eld method in latti e gauge theories
as a variational method [BDI74 whi h is similar to that proposed by R. Peierls
in thirties. It is based on Jensen's inequality13

F
 e hF i0
(2.3.23)
e
0

13 More detail an be found,

to Chapter 1.

, in the books [Fey72, Sak85 ited in the referen e list

e.g.

136

CHAPTER 2.

LATTICE GAUGE THEORIES

whi h is due to the onvexity of the exponential fun tion, where h: : :i0 means
averaging with respe t to a trial a tion.
Let us hoose the trial partition fun tion

Z0 =

Z Y

x;

dUx; e


N

x; Re tr Ux;

(2.3.24)

as a produ t of one-link integrals. Adding and subtra ting the trial a tion,
we write down the following bound on the partition fun tion (2.2.31):

P

 P Re tr Ux;
Re tr Up N
N

x;
p
0
Z0 e

X
X

Re tr Up N
Re tr Ux;
N p
x;
 2 Nl m ;
= Np m4 N

where Eq. (2.3.19) has been used.


The idea of the variational mean- eld method is to x  from the ondition
for the trial ansatz (2.3.24) to give the best approximation to Z in the given
lass. Cal ulating the derivative of the RHS of Eq. (2.3.25) with respe t to 
and taking into a ount that m depends on  a ording to Eq. (2.3.17), we
nd the maximum at  given by Eq. (2.3.18), whi h reprodu es the simplest
version of the mean- eld method des ribed above.
To restore Elitzur's theorem, a more sophisti ated trial ansatz [Dro81 an
be onsidered:

Z0 =

Z Y

x;

dUx; e N

P
y
x; Re tr Bx; Ux;

Problem 2.21

(2.3.27).

Solution

(2.3.27)

where we hoose Bx; to be an arbitrary omplex N  N matrix. Now the


best approximation is rea hed for
Bx; = 
x
yx+a^ ;
(2.3.28)
where  is given by exa tly the same
as before, while
x 2 SU(N )

ij equation
but is arbitrary otherwise. Now Ux;
vanishes
after summing over equiv0
alent maxima whi h results in integrations over d
x .

Perform the variational mean- eld al ulation with the ansatz

Let us denote
ij =
Mx;

RQ

P
y
ij
x; Re tr Bx; Ux; Ux;
P
y Ux;
Re tr Bx;
N
dUx; e x;

x; dUx; e

RQ

x;

Then the analog of Eq. (2.3.26) reads


*

X Re tr U
p
N p

X Re tr M
=
p
N p

x;

y Ux;
Re tr Bx;

x;

(2.3.29)

+
0

y Mx;
Re tr Bx;

(2.3.30)

so that the inequality (2.3.25) takes on the form

(2.3.26)

137

LATTICE METHODS

(2.3.25)

Here h: : :i0 means averaging with respe t to the same a tion as in Eq. (2.3.24).
Sin e the expression whi h is averaged in the exponent in Eq. (2.3.25) is
linear in ea h of the link variables, it an be al ulated via the one-matrix
integral given by the LHS of Eq. (2.3.17). Therefore, we get
*

2.3.

P
y
P Re tr M
p N x; Re tr Bx; Mx; :
p

 Z0 e N

(2.3.31)

Bx; an now be determined by maximizing with respe t to Bx; and taking into
a ount Eq. (2.3.29).
It is easy to see that if Bx; =   I is a solution as before, then (2.3.28) is also
a solution. Therefore, we get

ij
U

x; 0

= m

d
x+a^ d
x
x+a^
yx = 0 ;

(2.3.32)

where the integration over


takes into a ount di erent equivalent maxima. Thus,
all gauge invariant quantities for the ansatz (2.3.27) are the same as for the
ansatz (2.3.24) while gauge noninvariant quantities now vanish in agreement with
Elitzur's theorem.

Remark on the riterion for phase transition


Another puzzle with the simplest mean- eld method is why the point of the
rst-order phase transition is hoosen as is explained in Fig. 2.16 but not
when the free energy of both phases oin ide (the standard Maxwell rule in
statisti al physi s). Perhaps, the riterion of Fig. 2.16 should be hoosen if
a barrier between two phase is impenetrable whi h happens at large N or
if quantum u tuations are not taken into a ount as for the simplest mean
eld. The mean- eld al ulations of Ref. [FLZ82, whi h take into a ount
u tuations around the mean- eld solution (2.3.28), agree for the Maxwellrule riterion with numeri al data. These results are reviewed in Ref. [DZ83.

138

CHAPTER 2.

LATTICE GAUGE THEORIES

r.g.

b)

Fig. 2.17: Latti e renormalization group transformation (2.3.33). The thin lines of
the old latti e a) represent links on whi h integration is performed. The
new latti e b) has the latti e spa ing 2a but the same spatial extent La.

2.3.4 Latti e renormalization group


While the mean- eld method is useful for studying the rst-order phase transitions, the se ond-order phase transitions in latti e statisti al systems are
better des ribed by the renormalization group method (see, e.g., the review
by Wilson and Kogut [WK74). The idea to apply a similar method to latti e
gauge theories is due to Migdal [Mig75.
A simple renormalization group transformation in latti e gauge theories is
asso iated with doubling of the latti e spa ing a. One has originally a latti e
depi ted in Fig. 2.17a. The latti e renormalization group transformation
onsists in integrating over the link variables Ux; on the links shown by the
thin lines whi h results in a latti e with spa ing 2a,

a =r.g.
) 2a ;

(2.3.33)

whi h is depi ted in Fig. 2.17b. The spa e size of the latti e is L before the
transformation and be omes L=2 after the transformation,

L
;
(2.3.34)
2
so that the latti e extent is L  a in both ases whi h is expe ted to redu e an
in uen e of nite size e e ts on the transformation.
The Wilson a tion on the latti e of Fig. 2.17a be omes a more general one
r.g.

L =)

139

LATTICE METHODS

under the renormalization group transformation:


X 1
S [U = tr Up
p N
X
1
r.g.
=) S 0 [U =
10 tr Up
N

p
X
X
1
1
+ 20 tr Up2 + 30 tr Up3 + : : : :
N
N


p2
p3

=)

a)

2.3.

(2.3.35)

The new a tion S 0 [U is not ne essarily a single-plaquette a tion and an


involve tra es of the Wilson loops for boundaries of double plaquettes, triple
plaquettes and so on.
The new a tion would be the same as the old one only at a xed point.
This usually happens after the renormalization group transformation is applied several times when the latti e theory does have a xed point. The
resulting a tion is then asso iated with an a tion of a ontinuum theory.
The great su ess of non-Abelian latti e gauge theories with the Wilson
a tion in des ribing the ontinuum limit even at a relatively small spatial
extent or, whi h is the same, at relatively large g2 and a, is be ause it is
not far away from the xed-point a tion of the renormalization group. The
proper numeri al results are presented below in this Subse tion.
If both a tions S [U and S 0 [U are the single-plaquette Wilson a tions,
then
r.g.
=) 0 = 
(2.3.36)
under the renormalization group transformation on the latti e.
Sin e the Gell-Mann{Low fun tion B(g2) in the ontinuum is known, 
versus is determined by the equation
Z

dx
B
x3=2

 

6
x

2 ln 2
p :
6

(2.3.37)

Here the ln 2 on the RHS is due to Eq. (2.3.33) and the relation (2.2.32)
between and g2 is used with N = 3.
For the pure SU(3) gauge theory, we get from Eq. (2.3.37)
 = 0; 579 +

1
0; 204
+O 2

(2.3.38)

140

CHAPTER 2.

LATTICE GAUGE THEORIES

2.3.

141

LATTICE METHODS

at asymptoti ally large .


One an integrate over the thin links in Fig. 2.17a either approximately
or numeri ally. The following pro edure for an approximate integration is
known as Migdal{Kadano re ursion relations.
Let us expand the exponential of the old a tion in hara ters

e S[U =

fr dr r (U )

(2.3.39)

where

dr = r (I )

(2.3.40)

is the dimension of a given representation r and fr are the oe ients whi h


depend on the form of S [U .
Migdal [Mig75 proposed to approximate the new a tion, whi h appear
after

a =)  a ;

(2.3.41)

by the formula
0 0
e S [U =

"

fr  dr r (U 0 )
2

# d 2

(2.3.42)

whi h is exa t in d = 2 dimensions. Kadano [Kad76 slightly modi ed the


re ursion relation (2.3.42).
The study of the Migdal{Kadano re ursion relations was histori ally the
rst argument that se ond-order phase transitions do not o ur in the nonAbelian latti e gauge theory when g2 is de reased. Moreover, these relations
in d = 4 are the same as for spin systems (with the same symmetry group)
in d = 2 where this phenomenon is known. A disadvantage of the method is
that it is di ult to estimate its a ura y.
A nal answer to the question of whether or not a se ond-order phase
transition o urs in the non-Abelian latti e gauge theory was done by the
numeri al integration. This is known as the Wilson Monte Carlo renormalization group. Some typi al results [Ake93 for the  , whi h is de ned by
Eq. (2.3.36), versus are depi ted in Fig. 2.18. The solid line represents
the asymptoti s (2.3.38). The agreement on rms that the ontinuum limit
is rea hed already at these values of , while the deviation of the Monte
Carlo data from the asymptoti s for smaller values of is due to latti e
non-perturbative e e ts.

Fig. 2.18: Monte Carlo data by Akemi et al. [Ake93 for  . The errorbars represent
statisti al errors. The solid line represents the asymptoti s (2.3.38).

2.3.5 Monte Carlo method


The idea of the Monte Carlo method is to al ulate the partition fun tion (2.2.31) and the averages (2.2.39) for arbitrary values of numeri ally,
using the fa t that the multipli ity of the integral is large. For a L  L  L  L
latti e in four dimensions, a typi al multipli ity of the integral is as large as
4  (N 2 1)  L4 ( 107 for L = 24). It is hopeless to al ulate su h an integral
exa tly. On the ontrary, the larger the multipli ity the better the Monte
Carlo method works.
As usual, the Monte Carlo method is applied not to sequential integrals
over Ux; at ea h link but rather to the multiple integral as a whole, whi h
an be viewed as a sum over states of a statisti al system.
A state is identi ed with a gauge eld on guration whi h is des ribed by
the values of the link variables at all the links of the latti e:

C =

 ij
U

x; ; : : : ; : : : ; : : : ; : : : ; : : : ; : : : ; : : : ; : : : ; : : : ; : : : ; : : :

: (2.3.43)

There are as many positions in this row as the multipli ity of the integral.
Then the sequential integral an be represented as
Z Y

x;

dUx; : : : =

::: :

(2.3.44)

142

CHAPTER 2.

LATTICE GAUGE THEORIES

The averages (2.2.39) an be rewritten as


P

h F (C ) i

e S(C )F (C )

CP

e S(C )

(2.3.45)

where S (C ) and F (C ) are the values of F and S for the given on guration
C.
The task of Monte Carlo al ulations is not to sum over all possible on gurations whose number is in nite but rather to onstru t an ensemble, say,
of N on gurations

E =

fC1 ; : : : ; CN g

(2.3.46)

su h that for a given on guration Cn to en ounter with the Boltzmann


probability

P Bol (Cn ) = Z 1 ( ) e S(Cn ) :

(2.3.47)

Su h a sample of on gurations is alled the equilibrium ensemble.


Given an equilibrium ensemble, the averages (2.3.45) take the form of the
arithmeti mean

h F [U i

N
1 X
F (Cn )
N n=1

(2.3.48)

be ause ea h on guration \weighs" already as mu h as is required. In parti ular, the Wilson loop average for a re tangular ontour is given by

WRT =

N
1
1 X
tr URT (Cn ) :
N n=1 N

(2.3.49)

If all on gurations in the equilibrium ensemble are independent, then the


RHS of Eq.
p (2.3.49) will approximate the exa t value of WRT with an a ura y  N .
An analogy of this method of al ulating averages with statisti al physi s
is obvious. The equilibrium ensemble simulates a tual states of a statisti al
system while the index n des ribes a time evolution.
A ru ial point in the Monte Carlo method is to onstru t the equilibrium
ensemble. It is not simple to do that be ause the Boltzmann probability is
not known at the outset. A way out is to establish a random pro ess for

2.3.

143

LATTICE METHODS

whi h ea h new on guration in the sequen e (2.3.46) is obtained from the


previous one by a de nite algorithm but sto hasti ally. In other words, the
random pro ess is ompletely determined by the probability P (Cn 1 ! Cn )
for a transition from a state Cn 1 to a state Cn and does not depend on the
history of the system, i.e.

P (Cn

! Cn )

= P (Cn 1 ; Cn ) :

(2.3.50)

Su h a random pro ess is known as the Markov pro ess.


The transition probability P (C; C 0 ) should be hosen in su h a way to provide the Boltzmann distribution (2.3.47). This is ensured if P (C; C 0 ) satis es
the detailed balan e ondition
0
e S(C )P (C; C 0 ) = e S(C ) P (C 0 ; C ) :
(2.3.51)
Then
1) an equilibrium sequen e of states will transform into another equilibrium sequen e,
2) a nonequilibrium sequen e will approa h an equilibrium one when moving through the Markov hain.
Problem 2.22 Prove the statements 1) and 2) listed in the previous paragraph
using the detailed balan e ondition (2.3.51).
0
Solution Let a state C is en ountered in ensembles E and E with the probability
0
densities P (C ) and P (C ), respe tively. Then a distan e between the two ensembles
an be de ned as
X
jP (C ) P 0 (C )j :
(2.3.52)
kE E 0 k =

For a Markov pro ess when Eq. (2.3.50) holds, we have

P 0 (C ) =

C0

P (C; C 0 ) P (C 0 )

(2.3.53)

if E 0 is obtained from E by a Monte Carlo algorithm. The transition probability


P (C; C 0 ) is nonnegative and obeys
X

P (C; C 0 ) =

C0

P (C; C 0 ) = 1

sin e ea h new state is obtained from an old one and vi e versa.

(2.3.54)

144

CHAPTER 2.

LATTICE GAUGE THEORIES

It is now easy to prove the statement 1). Summing the detailed balan e ondition (2.3.51) over C 0 , we get
X
P (C; C 0 )P (C 0 ) ;
(2.3.55)
P (C ) =
Bol

Bol

C0

0
i.e. the Boltzmann distribution is an eigenve tor of P (C; C ). Comparing with
Eq. (2.3.53), we see that the new distribution is again the Boltzmann one, whi h
proves 1).
To prove the statement 2), let us ompare the distan es from E and E 0 to some
equilibrium ensemble E Bol asso iated with the Boltzmann distribution (2.3.47).
We have the inequality
X

P 0 (C ) P
kE 0 E k =
(C )
Bol

Bol


X X

P (C; C 0 ) P (C 0 )


C C0
X

P (C; C 0 ) P (C 0 ) P

X
P (C 0 )

kE E Bolk ;

Bol

CC 0
C0


P Bol (C 0 )


(C 0 )


P Bol (C 0 )

(2.3.56)

where Eqs. (2.3.53), (2.3.55) and (2.3.54) are used. Thus, 2) is proven.

Spe i Monte Carlo algorithms di er in the hoi e of the transition probability P (C; C 0 ) while the detailed balan e ondition (2.3.51) is always satis ed. The two most popular algorithms, whi h a t at one link, are
0 is sele ted randomly from
Heat bath algorithm: A new link variable Ux;
the group manifold with a probability given by the Boltzmann fa tor

P U0
/ e S(C 0 ) :
(2.3.57)
x;

Then this pro edure is repeated for the next link and so on until the
whole latti e is passed. This an be imagined as if a reservoir with a
temperature 1= tou hes ea h link of the latti e in su ession. It is
lear from physi al intuition that the system will be brought to the
thermodynami equilibrium sooner or later.

Metropolis algorithm: This algorithm is used in statisti al physi s sin e


the fties and onsists of several steps.

2.3.

LATTICE METHODS

145

0 is sele ted (suppose randomly on the


a) A trial new link variable Ux;
group manifold).
b) The di eren e between the a tion for this trial on guration and
that for the old one is al ulated:
S = S (C 0 ) S (C ) :
(2.3.58)
) A random number r 2 [0; 1 is generated.
d) If
e S > r ;
(2.3.59)
0 is a epted. Otherwise, Ux;
0 is reje ted and the old value
then Ux;
Ux; is kept.
e) All this is repeated for the next links.
A new on guration C 0 , whi h is obtained by applying either Monte Carlo
algorithm to ea h links of the latti e on e (this pro edure is often alled the
Monte Carlo sweep), will be strongly orrelated with the old one C . This
is be ause the latti e a tion depends not only on the variable at the given
link but also on those at the neighboring link whi h form plaquettes with the
given one. In order for C 0 to be ome independent on C , this pro edure should
be repeated many times or spe ial tri ks to redu e the orrelations should be
implemented. Then this new on guration an be added to the equilibrium
ensemble (2.3.46) as Cn .
More detail about the Monte Carlo algorithms as well as their pra ti al
implementation in latti e gauge theories an be found in the review [CJR83
and the book by Creutz [Cre83.

2.3.6 Some Monte Carlo results


The rst Monte Carlo al ulation in non-Abelian latti e gauge theories, whi h
is relevant for the ontinuum limit, was that by Creutz [Cre79 who evaluated the string tension for the SU(2) gauge group. His result is reprodu ed
in Fig. 2.19 and looks very mu h like what is expe ted in Fig. 2.12. This
al ulation was the rst demonstration that the ontinuum limit sets in for
relatively large g2  1:9 (  2:2) and that results for the ontinuum an
therefore be extra ted from relatively small latti es.
The restoration of rotational symmetry for these values of g2 was expli itly demonstrated by Land and Rebbi [LR82. They al ulated equipotential

146

CHAPTER 2.

LATTICE GAUGE THEORIES

2.3.

LATTICE METHODS

147

Fig. 2.19: Monte Carlo data by Creutz [Cre79 for the string tension in the SU(2)
pure latti e gauge theory.

surfa es for the intera tion between stati quarks. In the strong oupling
region g2 ! 1, they appear as in Fig. 2.20a sin e the intera tion potential
is given by
E (x; y; z ) = K ( jxj + jyj + jz j )
(2.3.60)
be ause the distan e between the quarks is measured along the latti e. This
is asso iated with a ubi symmetry on the latti e (i.e. rotations through
an angle whi h is a multiple of =2 around ea h axis and translations by a
multiple of the latti e spa ing along ea h axis) rather than with the Poin are
group. The rotational symmetry must be restored in the ontinuum limit.
The Monte Carlo data by Land and Rebbi [LR82 are shown in
Figs. 2.20b, . They demonstrate the restoration of rotational symmetry
when passing from = 2 (Fig. 2.20b) to = 2:25 (Fig. 2.20 ).
The old Monte Carlo al ulations played an important role for developing
the method. A dramati improvement of the Monte Carlo te hnology in
latti e gauge theories happened during the last ten years.
As an example of modern Monte Carlo al ulations, the Monte Carlo
data by Bali and S hilling [BS92 for the intera tion potential in the SU(3)

Fig. 2.20: Behavior of equipotential lines at di erent values of : a) the strong ou-

pling limit = 0, b) = 2, ) = 2:25. Figs. b) and ), taken from the


paper by Lang and Rebbi [LR82, show how the rotational symmetry is
restored as is in reased.

148

CHAPTER 2.

LATTICE GAUGE THEORIES

2.4.

149

FERMIONS ON A LATTICE

2.4 Fermions on a latti e

Fig. 2.21: Monte Carlo data [BS92 for the ontinuum intera tion potential E (R).
The solid urve is a sum of linear and Coulomb potentials.

pure latti e gauge theory are shown in Fig. 2.21. The solid urve represents
how the data are des ribed by a sum of linear and Coulomb potentials. The
maximal latti e size was 324 and the spatial extent of the latti e was up to
3.3 fm.
The Pro eeding of the Latti e Conferen e [Lat94 are very useful, as is
already mentioned, for a today's look at the subje t.

It turned out to be most di ult in the latti e approa h to QCD to deal with
fermions. Putting fermions on a latti e is an ambiguous pro edure sin e the
ubi symmetry of a latti e is less restri tive than the ontinuous Lorentz
group.
The simplest hiral-invariant formulations of latti e fermions lead to a
doubling of fermioni degrees of freedom, as was rst noted by Wilson [Wil75,
and des ribe from 16 to 4 relativisti ontinuum fermions, depending on the
formulation. One half of them has a positive axial harge and the other
half has a negative one, so that the hiral anomaly an els. There is a nogo theorem whi h says that the fermioni doubling is always present under
natural assumptions about a latti e gauge theory.
A pra ti al way out is to hoose the fermioni latti e a tion to be expli itly
noninvariant under the hiral transformation and to have, by tuning a mass
of the latti e fermion, 1 relativisti fermion in the ontinuum and the masses
of the doublers to be of the order of the inverse latti e spa ing. The hiral
anomaly is re overed in this way.
We onsider in this Se tion various formulations of latti e fermions and
the doubling problem. We dis uss brie y the results on spontaneous breaking
of hiral symmetry in QCD.

2.4.1 Chiral fermions


The quark elds are generi ally matter elds, whose gauge transformation in
the ontinuum is given by Eqs. (2.1.1) and (2.1.3), and an be put on a latti e
a ording to Eq. (2.2.7). Then the latti e gauge transformation is
g.t.
! x0 =
x x ;
x
x g.t.
! 0 = x
y :
(2.4.1)
x

The latti e analog of the QCD a tion (2.1.14) reads14


X


x x
S U; ;
= SLat [U + M
1 X 
Uy
+
2 x;>0 x  x; x+a^

x+a^  Ux; x  :

(2.4.2)

14 A standard formula di ers from this one by an inter hange of U and U y due to the
inverse ordering of matri es in the phase fa tors (see the footnote on p. 91). It does not
matter how to de ne Ux; sin e the Haar measure is invariant under Hermitean onjugation.

150

CHAPTER 2.

LATTICE GAUGE THEORIES

The rst term on the RHS is the pure gauge latti e a tion (2.2.16). The
se ond term is a quark mass term on a latti e. The sum in the third term is
over all latti e links (i.e. over all sites x and positive dire tions ). This a tion
is Hermitean and invariant under the latti e gauge transformation (2.2.13),
(2.4.1) at a nite latti e spa ing.
The partition fun tion of latti e QCD with fermions is de ned by

Z ( ; M ) =

Z Y

x;

dUx;

d x d x e S[U; ; ;

(2.4.3)

F U; ;
Z

1 ( ; M )

Z Y

x;

dUx;

d x d x e

S [U; ;

F U; ;

; (2.4.4)

whi h extends Eq. (2.2.39) to the ase of fermions. Sin e both the a tion
and the measure in Eq. (2.4.4) are gauge invariant at nite
 latti e spa ing, a
nonvanishing result is only when the integrand, F U; ; , is gauge invariant
as well.
In order to show how the latti e a tion (2.4.2) reprodu es (2.1.14) in the
naive ontinuum limit a ! 0, let us assume that the latti e quark eld x
slowly varies from site to site and substitute
x
x+a^

! a3=2 (x) ;
! a3=2 ( (x) + a (x))

(2.4.5)

in d = 4. Here (x) is a ontinuum quark eld and the power of a is due to


dimensional onsideration (remember that x is dimensionless).
Equation (2.4.5) together with Eq. (2.2.10) yields
y x+a^
x  Ux;

! a3  

5
+ a4 rfun
  + O a ;

(2.4.6)

where there is no summation over  in the se ond term as before in this


Chapter. The rst term an els when substituted into Eq. (2.4.2) while the
se ond one reprodu es the fermioni part of the ontinuum a tion. The mass
term is also reprodu ed if M = am.

151

FERMIONS ON A LATTICE

The fermioni latti e a tion (2.4.2) was proposed in Ref. [Wil74. For
M = 0 it is invariant under the global hiral transformation
x

:t:

e i 5 x ;

t:  i 5
x :!
:
xe

(2.4.7)

For this reason, these latti e fermions are alled hiral fermions. Sin e the
latti e a tion is both gauge and hiral invariant, there is no Adler{Bell{Ja kiw
anomaly a ording to the general arguments of Se t. 1.3.
Problem 2.23

where the a tion is given by Eq. (2.4.2). The integration over Ux; is as in
Eq. (2.2.31), and the integral over the quark eld is the Grassmann one. The
averages are de ned by

2.4.

Show that the latti e a tion (2.4.2) is invariant under


x

! i 4 5 ( 1)t=a x :

Find 15 more similar transformations.


Solution Let us de ned the generators T by
x ! x T y :
x ! T x ;

(2.4.8)

(2.4.9)

The transformation (2.4.8) an be performed for ea h of the d = 4 axes whi h gives

T = i  5 ( 1)x =a :

(2.4.10)

The other generators are given by produ ts of (2.4.10). Their expli it form
is [KS81a

T = I ; i  5 ( 1)x =a ; i   ( 1)(x +x )=a ( >  ) ;


4 ( 1)(x+y+z)=a ; : : : ; 1 ( 1)(y+z+t)=a ; 5 ( 1)(x+y+z+t)=a :(2.4.11)
There are all together 1 + 4 + 6 + 4 + 1 = 16 independent transformations whi h
form a dis rete subgroup of the U(4) group.

2.4.2 Fermion doubling


As is pointed out at the end of the previous Subse tion, the latti e fermioni
a tion (2.4.2) is both gauge and hiral invariant (for M = 0) so that there
is no axial anomaly in the ontinuum. Sin e the anomaly is present for one
ontinuum fermion, this suggests that the a tion (2.4.2) is asso iated with
more than one spe ies of ontinuum fermions.
In order to verify this expli itly, let us al ulate the poles of the latti e
fermioni propagator.
As usual, it is easier to work with the Fourier image of x :
k

= a5=2

xe

ikx :

(2.4.12)

152

CHAPTER 2.

LATTICE GAUGE THEORIES

The free fermioni a tion then reads




S0 ;

=a
Z

=a

d4 k  1
k G (k ) k
(2)4

(2.4.13)

with

G 1 (k) =

4
1X
i sin k a
a =1 

(2.4.14)

for M = 0.
In the naive ontinuum limit, the sin in Eq. (2.4.14) an be expanded in
the power series in a, whi h results in the free (inverse) ontinuum propagator

G 1 (k)

! i

4
X

=1

 k = i k^ :

(2.4.15)

The Lorentz invarian e has been restored after summing over .


When passing from the latti e expression (2.4.14) to the ontinuum
one (2.4.15), it was saliently assumed that the momentum k is not of order
of 1=a be ause otherwise the sin an not be expanded in a. The doubling of
relativisti ontinuum fermioni states o urs exa tly for this reason.
To nd the poles of the propagator, let us return to Minkowski spa e
substituting k4 = iE with E being the energy. The poles are then determined
by the dispersion law
sinh2 Ea =

3
X

=1

sin2 p a :

(2.4.16)

Let us look for solutions of Eq. (2.4.16) with positive energy E > 0
(solutions with negative energy are asso iated as usual with antiparti les).
Suppose that a parti le moves along the z -axis so that omponents of the
four-momentum

p(1) = (E; 0; 0; pz )

(2.4.17)

sinh Ea = sin pz a ;

(2.4.18)

are related by

2.4.

153

FERMIONS ON A LATTICE

whi h follows from the substitution of (2.4.17) into the dispersion law.
Sin e the sin is a periodi fun tion, the four-ve tor



p(2) = E; 0; 0;
(2.4.19)
pz
a
is also a solution of Eq. (2.4.18) if (2.4.17) is. Quite analogously, the fourve tors
 

p(3) = E; ; 0; pz ;
a
: : : ;

  
p(8) = E; ; ;
pz ;
(2.4.20)
a a a
whi h are obtained from p(1) and p(2) by hanging zeros for =a, also satisfy Eq. (2.4.18). Therefore, a quark state with the energy E is eightfold
degenerate.
The quark states with the four-momenta p(1); : : : ; p(8) are di erent states.
Their wave fun tions equal
(j) (t; x; y; z )

exp iEt ip(xj) x ip(yj) y ip(zj) z :

(2.4.21)

The wave fun tion in the state with the momentum p(3) di ers, say, from the
wave fun tion in the state p(1) by an extra fa tor ( 1)x=a . In other words, it
strongly hanges as a ! 0 with one step along the latti e in the x-dire tion.
One more step returns the initial value.
For su h fun tions, the naive ontinuum limit of the latti e a tion (2.4.2)
is as good as for the slowly varying fun tions when Eq. (2.4.5) holds. In order
to see that, let us rewrite the a tion (2.4.2) as
X


x x
S U; ;
= SLat [U + M
x

1 X h  y
+
U
2 x;>0 x  x; x+a^

Ux;y  x a^

i

(2.4.22)

Even if x has opposite signs at neighboring latti e sites along the -axis, as
is illustrated by Fig. 2.22a, i.e.
(2.4.23)
x+a^ !
x;
then the di eren e x+a^
x a^ on the RHS of Eq. (2.4.22) is still of the
orre t order in a:
3=2 (x) ;
2a5=2  (x) ; (2.4.24)
x ! a
x+a^
x a^ !

154

CHAPTER 2.

LATTICE GAUGE THEORIES

+r

+r

+r

+r

+r

+r

+r

+r

+r

+r

+r

+r

+r

+r

+r

+r

+r

+r

+r

+r

+r

+r

+r

+r

+r

+r

+r

+r

+r

+r

+r

+r

+r

+r

+r

+r

+r

+r

a)

x on a latti e along a) one axis and b) two axes.

so that the ontinuum fermioni a tion is reprodu ed ex ept for the sign of
the  -matrix whi h is opposite to that in Eq. (2.1.14).
This extra minus sign an be absorbed in the rede nition of the ontinuum fermioni eld (x) ! i  5 (x), whi h hanges its hirality. Therefore,
the axial harge of the doublers is opposite. Analogously, four of the eight
doublers have positive axial harge and four others have negative one dependently on whether the sign of x alters at neighboring sites along even or odd
number of axes (see Fig. 2.22). In Eu lidean spa e the doubling o urs also
along the temporal axis, so the number of doublers is equal to 2d = 16: 8 of
them with positive and 8 with negative axial harge. This explains why the
hiral anomaly an els.
Problem 2.24 Cal ulate the ve tor and axial harges of the doublers deriving the
ve tor and axial urrents on a latti e.
Solution The ve tor and axial urrents on a latti e an be derived using a latti e
analog of the Noether theorem. The invarian e of the latti e fermioni a tion under
i
x ! x e i x
(2.4.25)
x ! e x x;

results in the latti e ve tor urrent



y x+a^ + x+a^  Ux; x  ;
Vx; = 1 x  Ux;
2
whi h is onserved in the sense that

(2.4.26)

(Vx;

(2.4.27)

>0

Vx a^; ) = 0 :

155

FERMIONS ON A LATTICE

This an be proven using the latti e (quantum) Dira equation


1 X Uy
2 >0  x; x+a^


Ux;y  x a^ + M x w.s.
=  ;
x

(2.4.28)

whi h is the latti e analog of Eq. (1.3.20).


Analogously, the latti e hiral transformation
x

! e i x 5

x;

x ! x e i x 5

(2.4.29)

results in the latti e axial urrent

b)

Fig. 2.22: Altering signs of

2.4.


y x+a^ + x+a^  5 Ux; x  ;
Ax; = i x  5 Ux;
2

(2.4.30)

whi h reprodu es (1.3.11) as a ! 0. The urrent (2.4.30) is onserved for M = 0.


It is now easy to verify that 16 generators (2.4.11) ommute with the latti e U(1)
transformation (2.4.25) so that the latti e ve tor urrent (2.4.26) is left invariant.
Analogously, the latti e axial U(1) transformation (2.4.29) ommutes only with
1 + 6 + 1 = 8 of 16 generators (2.4.11) whi h are built out of the produ ts of an
even number of the generators (2.4.10) and does not ommute with the 4 + 4 =
8 other generators. Therefore, the axial urrent (2.4.30) is invariant under the
1 + 6 + 1 = 8 transformations, whi h are the produ ts of an even number of the
generators (2.4.10), and alters its sign under the other 4 + 4 = 8 transformations,
whi h are the produ ts of an odd one. Thus, the ve tor harge of all the doublers is
the same while the axial harge is positive for 8 and negative for 8 other doublers.

It is worth noting that the mass term in Eq. (2.4.2) is not 5 invariant
but does not remove the fermioni doubling.
One might think of removing the doubling problem by modifying the
expression for the inverse latti e propagator G 1 (k) in the free fermioni
latti e a tion (2.4.13), for instan e, by adding next-to-neighbor terms. It is
easy to see that it does not help if the fun tion G 1 (k) is periodi as it should
be on a latti e. A typi al form of G 1 (k) as a fun tion of, say, k4 is depi ted
in Fig. 2.23. The behavior around k4 = 0 is pres ribed by Eq. (2.4.15) and is
just a straight line with a positive slop. Therefore, G 1 (k) will have inevitably
another zero at k4 = =a due to periodi ity.
This is the di eren e between the fermioni and bosoni ases. For bosons
G 1 (k) is quadrati in k4 near k4 = 0 rather than linear as for fermions. A
typi al behavior of G 1 (k) for bosons is shown in Fig. 2.24. There is no
doubling of states in the bosoni ase.

156

CHAPTER 2.

6G

LATTICE GAUGE THEORIES

2.4.

6G


a

157

FERMIONS ON A LATTICE

+ a

k4

Fig. 2.23: Momentum dependen e of G

1 for the hiral latti e fermions. The periodi ity leads to an extra zero at k4 = =a.

Remark on the Nielsen{Ninomiya theorem


A general proof of the theorem whi h says that there is no way to avoid the
fermion doubling under natural assumptions about the stru ture of a latti e
gauge theory was given by Nielsen and Ninomiya [NN81. It is formulated
sometimes as an absen e of neutrinos on a latti e. In other words, this is a nogo theorem for putting theories with unequal number of left- and right-handed
massless Weyl parti les on a latti e, su h as in the standard ele troweak
theory.
A way to bypass the Nielsen{Ninomiya theorem is, say, to hoose a
fermioni latti e a tion whi h is highly nonlo al. Then it is possible to repla e
sin k a in Eq. (2.4.14) by k a itself to get an expression whi h is similar to
the ontinuum propagator (2.4.15). However, su h a nonlo al modi ation is
useless for pra ti al al ulations. Some other modern approa hes to resolving
the fermion doubling problem an be found in the Pro eedings of the Latti e
Conferen e [Lat94.

2.4.3 Kogut{Susskind fermions


The number of ontinuum fermion spe ies is not ne essarily equal to 16. It
an be redu ed down to 4 by the tri k whi h was proposed for the Hamiltonian formulation in Refs. [KS75, Sus77 and elaborated for the Eu lidean
formulation in Refs. [STW81, KS81b.

+ a


a

k4

Fig. 2.24: Momentum dependen e of G

1 for the latti e bosons. No doubling of


states is asso iated with this behavior.

Let us substitute

= 1x=a 2y=a 3z=a 4t=a x


(2.4.31)
into the free fermioni a tion (2.4.13). Then it takes the form


h
ii
 y i


1 XX X
i
i
y

x+a^ [x ; (2.4.32)
x [x+a^
=
S0 ;

2 x i >0 x;
x

whi h is diagonal with respe t to the spinor indi es sin e

x; = ( 1)(x1 +:::+x 1 )=a ;

(2.4.33)

or expli itly

x; = 1
for  = 1 ;
x
x; = ( 1) 1 =a
for  = 2 ;
::: ;
x; = ( 1)(x1 +:::+xd 1 )=a
for  = d ;
(2.4.34)
does not depend on spinor indi es.
The idea is to leave only one omponent of ix in order to redu e the
degenera y:
1
0
1
0
x
x
B 0 C
B 0 C
C
C
ix = B
ix = B
(2.4.35)
 0 A;
 0 A:
0
0

158

CHAPTER 2.

LATTICE GAUGE THEORIES

These latti e fermions are known as the staggered fermions, sin e x; is
staggering from one latti e site to another. They are more often alled the
Kogut{Susskind fermions due to their relation to those of Refs. [KS75, Sus77.
The a tion of the Kogut{Susskind fermions is


S U; ;
+

= SLat [U + M

1 X
  U y 
2 x;>0 x; x x; x+a^

2d=2

x x


x+a^ Ux; x :

(2.4.36)

2d=2 1

2.4.

159

FERMIONS ON A LATTICE

Substituting the expansion (2.4.5) in the a tion (2.4.37), we get, in the


naive ontinuum limit, the ontinuum fermioni a tion (2.1.14) with the mass
being

4
:
(2.4.38)
a
Therefore, the Wilson latti e fermions des ribe a relativisti fermion of the
mass m in the ontinuum when

m =

M ! 4 + ma :

(2.4.39)

It des ribes
= 4 spe ies for omplex x or
= 2 spe ies for Majorana x. Components of a ontinuum bispinor are distributed in this approa h
over four latti e sites.
There is no axial anomaly for the Kogut{Susskind fermions as with the
hiral fermions.

In order to see that there are no other relativisti fermion states in the
limit (2.4.39), let us onsider the fermioni propagator whi h is given by

Remark on four generations

Introdu ing the Minkowski-spa e energy E =


dispersion law

It is suggestive to identify four spe ies of Kogut{Susskind fermions with four


generations of quarks and leptons (see, e.g., Ref. [KMNP83). Remember that
one of the motivations to add the fourth generation to the standard model is
to an el the anomaly. However, there are problems with this idea on erning
the splitting of fermion masses for the four generations.

2.4.4 Wilson fermions


The hiral latti e fermions were proposed by Wilson [Wil74. Soon after
that he re ognized [Wil75 the problem of fermion doubling and proposed a
latti e fermioni a tion whi h des ribes only one relativisti fermion in the
ontinuum. The latter fermions are alled Wilson fermions.
The latti e a tion for the Wilson fermions reads


S U; ;

= SLat [U + M

x x

1
y x+a^ + x+a^ (1 +  ) Ux; x : (2.4.37)
 (1  ) Ux;
2 x;>0 x
X 

G 1 (k ) = M

osh Ea =

3
P

1+ M

=1

2 M

os p a
3
P

=1

!2

ik4, we get the following


3
P

=1

sin2 p a

os p a

(2.4.41)

Let a parti le be at rest, i.e. p1 = p2 = p3 = 0 and E = m > 0. Then


Eq. (2.4.41) redu es for ma  1 to the relation (2.4.39). It is easy to show
that a parti le at rest is the only solution to Eq. (2.4.41) whose energy is
nite as a ! 0.
The di eren e between the dispersion laws for the hiral and Wilson
fermions is be ause the fun tion on the RHS of Eq. (2.4.41) is no longer
periodi . It redu es for a ! 0 and M ! 4 to a usual relation

E =

The di eren e between this a tion and the a tion (2.4.2) for hiral fermions
is due to the proje tor operators (1   ) whi h pi k only one fermioni state.

4 

1X
(1  ) e ik a + (1 +  ) e ik a : (2.4.40)
2 =1

~p2 + m2

(2.4.42)

between the energy and momentum of a relativisti parti le.


Problem 2.25

M  4.

Show that a solution to the dispersion law (2.4.41) is unique for

160

CHAPTER 2.

LATTICE GAUGE THEORIES

2.4.

161

FERMIONS ON A LATTICE

For M  4, we an repla e the LHS of Eq. (2.4.41) by 1 and substitute M = 4 on the RHS. Then Eq. (2.4.41) redu es to the equation for spatial
omponents of the four-momentum:

Solution

3
X

=1

!2

os p a

+ 3

3
X

=1

os2 p a

= 0;

(2.4.43)

It is instru tive to dis uss what happens with the fermion doublers under
the hange of   by (1   ) in the latti e fermioni a tion. Let us onsider
one of su h states, say, that with p1 = =a, p2 = p3 = 0. Its energy is
determined by Eq. (2.4.41) to be  1=a so that this state is inessential as
a ! 0.
The hiral anomaly is orre tly re overed by the Wilson fermions. The 15
states of the mass  1=a play thereat a role of regulators whi h result in the
anomaly as a ! 0.
Cal ulate the masses of all 16 fermioni states.
Substituting Eqs. (2.4.23), (2.4.24) and so on into the a tion (2.4.37),

Problem 2.26

we get

m =
where

s = e ip a

M


Therefore, 1 state is relativisti as M

P4

=1 s

+1 p = 0
1 p = =a :

(2.4.44)

(2.4.45)

! 4 while 15 others have the masses  1=a.

Remark on ba ktra kings for Wilson fermions


Another way to understand why the doubling problem is removed for the
Wilson fermions is to onsider how they propagate on a latti e. The proje tors
1  
Wilson fermions
(2.4.46)
2
restri t the propagation of the Wilson fermions. One half of the states propagates only in the positive dire tions and the other half propagates only in

P =

whose only solution is p1 = p2 = p3 = 0 sin e both terms on the LHS are nonnegative.

Solution

Fig. 2.25: A path

yx made out of the string bits, whi h leads to a nonvanishing term


of the hopping parameter expansion for the quark propagator (2.4.49) on
a latti e. Ea h site involves at least two quark elds (depi ted by the
ir les). Otherwise the Grassmann integral at a given site vanishes.

the negative ones. In parti ular, there are no ba ktra kings in the (latti e)
sum over paths sin e
P+ P = 0 :
(2.4.47)
This removes the doubling.
Problem 2.27 Represent the fermion propagator in an external Yang{Mills eld
as a sum over paths on a latti e, performing an expansion in 1=M .
Solution Let us res ale the fermion eld, absorbing the parameter M in front of
the mass term. The fermioni part of the new a tion reads
X
X 
y x+a^ + x+a^ P + Ux; x  ; (2.4.48)
x x 
x P Ux;
S =


x;>0

where  = 1=M is usually alled the hopping parameter. The large mass expansion
in 1=M is now represented as the hopping parameter expansion in .
It is onvenient to depi t ea h of the two terms in the square bra kets in
Eq. (2.4.48) by a string bit as in Fig. 2.4 with the quark elds at the ends and
the gauge variable at the link. The rst term orresponds to the negative dire tion
of the link, and the se ond term orresponds to the positive dire tion. Substituting Eq. (2.4.48) into the de nition (2.4.4) and expanding the exponential in ,
we get a ombination of terms onstru ted from the string bits. A nonvanishing
ontribution to the quark propagator

i
j
Gijmn (x; y ; U ) =
(2.4.49)
m (x) n (y ) ;
where i; j and m; n represent, respe tively, olor and spinor indi es, emerges when
the links form a path yx that onne ts x and y on the latti e as is depi ted
in Fig. 2.25. Otherwise, the average over and  vanishes due to the rules of
integration over Grassmann variables des ribed in Problem (1.17).
Therefore, we get

Gijmn (x; y ; U ) =

X
yx

1
U ij [ yx 4
M L( )+1

Y
yx

P 5

;
mn

(2.4.50)

162

CHAPTER 2.

LATTICE GAUGE THEORIES

where P+ or P are asso iated with the positive or negative dire tion of a given
link 2 yx . For the Wilson fermions, they are given by Eq. (2.4.46), while

P =  
hiral fermions
(2.4.51)
2
for hiral fermions. The sum in Eq. (2.4.50) runs over all the paths between x and y
on the latti e, while L( ) stands for the length of the path yx in the latti e units.
Eq. (2.4.50) is a latti e analog of the ontinuum formula (2.1.43) for the ase of
fermions.
Represent the integral over fermions in Eq. (2.4.3) as a sum over
losed paths on a latti e, performing an expansion in 1=M .
Solution The al ulation is analogous to that of the previous problem. The result
reads

Problem 2.28

Z Y

d x d x e S = e Sind [U

(2.4.52)

2.4.

am

!!!
!
!
!!
-

M

Sind [U =

tr U [
sp
L( ) M L( )

P ;

(2.4.53)

where the ombinatori fa tor 1=L( ) is due to identity of L links forming the losed
ontour , and the minus sign is be ause of fermions.
Equation (2.4.53) de nes an e e tive a tion of a pure latti e gauge theory,
whi h is nonlo al sin e involves arbitrary large loops. However, it an be made
the single-plaquette latti e a tion (2.2.16) by introdu ing many avors of latti e
fermions [Ban83, Ham83.

2.4.5 Quark ondensate


The latti e a tion (2.4.37) is not invariant under the hiral transformation.
Therefore, the hiral symmetry is expli itly broken for the Wilson fermions.
Nevertheless, one expe ts a restoration of hiral symmetry as a ! 0 when
the relativisti fermion is massless (say, for M = 4 in the free ase) while
heavy states with m  1=a play the role of regulators. For the intera ting
theory, this restoration happens at some value M = M whi h is no longer
equal to 4. A signal of this restoration is the vanishing of the mass of the
-meson (as is illustrated by Fig. 2.26). m = 0 is usually asso iated with
the fa t that the hiral symmetry is realized in a spontaneously broken phase
and the -meson is the orresponding Goldstone boson.

Fig. 2.26: Dependen e of the -meson mass on the latti e quark mass M . At

M = M the -meson be omes massless and the hiral symmetry is restored.

For the hiral or Kogut{Susskind fermions with M = 0, the latti e a tion is invariant under the global hiral transformation (2.4.7). The order
parameter for breaking the hiral symmetry is


with

163

FERMIONS ON A LATTICE

:t:

 e 2i 5 ;

(2.4.54)

whi h is not invariant under the hiral transformation. Therefore, the average
of  must vanish if the symmetry is not broken spontaneously15. Su h
spontaneous breaking results in



6= 0 :
(2.4.55)
This nonvanishing value of the average of x x does not depend on x due to
translational invarian e and is alled the quark ondensate.
The spontaneous breaking of the hiral symmetry in QCD was demonstrated by the Monte Carlo al ulations of the quark ondensate. This quantity has a dimension of [mass3 and should depend on g2 at small g2 as is
pres ribed by the asymptoti s aling. The Monte Carlo data for the quark
ondensate from the pioneering paper by Hamber and Parisi [HP81 are shown
in Fig. 2.27. Its agreement with asymptoti s aling demonstrates that the hiral symmetry is spontaneously broken in the ontinuum QCD.

Remark on Monte Carlo with fermions


Monte Carlo simulations with quarks are mu h more di ult than in a pure
gauge theory. Integrating over the quark elds by using Eq. (1.2.15), one is
15 Spontaneous symmetry breaking usually o urs when the va uum state is not invariant

under the symmetry transformation.

164

CHAPTER 2.

LATTICE GAUGE THEORIES

2.5.

165

FINITE TEMPERATURES

2.5 Finite temperatures

Fig. 2.27: Monte Carlo data by Hamber and Parisi [HP81 for the quark ondensate
in the quen hed approximation.

left with the determinant, say for the Kogut{Susskind fermions, of the matrix

D [U = M x;y +

1 X
 Uy
2 >0 x; x; x(x+a^)

x; Ux;x(x a^) (2.4.56)

for a given on guration of the gluon eld Ux;. This results in a pure gauge eld problem with the e e tive a tion given by
e Seff [U = det D [U e S[U :

(2.4.57)

The matrix that appears in this determinant has at least N L4  N L4 elements, and is to be al ulated at ea h Monte Carlo upgrading of Ux; .
Several methods are proposed to manage the quark determinant approximately. The simplest one is not to take it into a ount at all. This approximation is known as the quen hed approximation when only valen e quarks are
onsidered, while the e e ts of virtual quark loops are disregarded. Re ently
a progress in the full theory has been a hieved using some tri ks for evaluating
the quark determinants (see Ref. [Lat94 for a review of the subje t).

Finite-temperature quantum eld theories at thermodynami equilibrium are


naturally des ribed by Eu lidean path integrals. The time-variable in this
approa h is ompa ti ed and varies between 0 and the inverse temperature
1=T . Periodi boundary onditions are imposed on Bose elds while antiperiodi ones are imposed on Fermi elds in order to reprodu e the standard Bose
or Fermi statisti s, respe tively.
The latti e formulation of QCD at nite temperature is espe ially simple,
sin e the Eu lidean latti e has a nite extent in the temporal dire tion. The
Wilson riterion of on nement is not appli able at nite temperatures and is
repla ed by another one, whi h is based on the thermal Wilson lines passing
through the latti e in the temporal dire tion. They are losed due to the
periodi boundary ondition for the gauge eld.
When the temperature in reases, QCD undergoes [Pol78, Sus79 a de on ning phase transition whi h is asso iated with the liberation of quarks. At
low temperatures below the phase transition, thermodynami al properties of
the hadron matter are well des ribed by a gas of nonintera ting hadrons while
at high temperatures above the phase transition these are well des ribed by
an ideal gas of quarks and gluons.
The situation with the de on ning phase transition be omes less de nite
when e e ts of virtual quarks are taken into a ount. The de on ning phase
transition makes stri t sense only for large values of the quark mass. For light
quarks, a phase transition asso iated with the hiral symmetry restoration at
high temperatures o urs with in reasing the temperature. It makes stri t
sense only for massless quarks.
We rst derive, in this Se tion, a path-integral representation of nitetemperature quantum eld theories starting from the Boltzmann distribution.
Then we apply this te hnique to QCD and dis uss the on nement riterion at
nite temperatures as well as the de on ning and hiral symmetry restoration
phase transitions whi h o ur with in reasing the temperature.

2.5.1 Feynman{Ka formula


Thermodynami properties of an equilibrium system in 3 + 1 dimensions are
determined by the partition fun tion

Z (T; V ) =

e En =T

Sp e H=T

(2.5.1)

166

CHAPTER 2.

LATTICE GAUGE THEORIES

whi h is asso iated with the Boltzmann distribution at the temperature T .


Here H is a Hamiltonian of the system and Sp is al ulated over any omplete
set of states, say, over eigenstates of the Hamiltonian whose eigenvalues are
hara terized by the energy En .
For a quantum theory of a single s alar eld ' (~x; t), the (S hrodinger)
states are des ribed by the bra- and ket-ve tors hgj and jf i:

hgj ~xi

h~x jf i

= g (~x) ;

= f (~x) ;

(2.5.2)

as is explained in Subse t. 1.1.1. The matrix element of the evolution operator


exp ( H=T ) is given by the formula

hgj e

H=T

jf i

D' (~x; t) e

R 1=T

(2.5.3)

'(x~;0)=f (~x)
'(~x;1=T )=g(~x)

where L is the proper Lagrangian, say for example,

L ['

dt L['

d3~x

1

1
( ')2 + m2 '2 + '3
2
2
3!

Sp e H=T =
=

Df (~x) hf j e H=T jf i
Z

D' (~x; t) e

real time for a system at equilibrium. The variable t in Eq. (2.5.5) is just
the proper time of the disentangling pro edure. This analogy between the
partition fun tions of statisti al systems and the Eu lidean formulation of
quantum eld theory is already mentioned in the Remark to Subse t. 1.1.7.
Derive the Feynman{Ka formula for a quantum parti le with nonrelativisti Hamiltonian (1.1.103).
Solution The matrix element hx jexp ( H=T )j xi is determined by Eq. (1.1.114)
to be
Z
R 1=T

H=T
x
x e
=
(2.5.7)
Dz (t) e 0 dt L(t) ;
Problem 2.29

z (0)=x
z (1=T )=x

where the Lagrangian L(t) is given by Eq. (1.1.115). Integrating over dd x, we


get [Fey53

(2.5.4)

dt L['

(2.5.5)

'(~x;1=T )='(~x;0)

Noti e that the path integral in Eq. (2.5.5) is taken with periodi boundary
onditions for the eld ':

' (~x; 1=T ) = ' (~x; 0) :

(2.5.6)

It reprodu es as T ! 0 the standard Eu lidean formulation of quantum eld


theory whi h is dis ussed in Se t. 1.2. The point is that nothing depends on
16 Its derivation in the modern ontext of non-Abelian gauge theories, whi h extends the

Feynman derivation [Fey53 for statisti al me hani s, is due to Bernard [Ber74.


dd x x e H=T x

Dz (t) e

R 1=T
0

dt L(t)

(2.5.8)

z (0)=z (1=T )

This integral is over the traje tories with periodi boundary onditions

z (0) = z (1=T ) :

(2.5.9)

Cal ulate the partition fun tion (2.5.8) for the free ase.
The Gaussian path integral with the boundary onditions

Problem 2.30
Solution

R 1=T

167

FINITE TEMPERATURES

Sp e H=T =

for the ubi self-intera tion of '. The derivation is quite analogous to that
of Problem 1.8.
In order to al ulate the tra e over states, one should put g(~x) = f (~x) and
perform the additional integration over f (~x). This yields the Feynman{Ka
formula16
Z

2.5.

z (0) = z (1=T ) = x

(2.5.10)

is al ulated in Subse t. 1.1.5 with the result given by Eq. (1.1.86). In order to
al ulate the partition fun tion (2.5.8), we need to integrate this expression over x
whi h yields [Fey53

Z (T; V ) =

dd x F (1=mT ) = V

mT d=2 :
2

(2.5.11)

The formula (2.5.11) is to be ompared with that given by the Boltzman distribution in lassi al statisti s. Sin e the energy of a free non-relativisti parti le
is
2
E (p~) = ~p ;
(2.5.12)
2m

168

CHAPTER 2.

LATTICE GAUGE THEORIES

the Boltzmann distribution is given by the sum over positions of the parti le in a
box of volume V and the integration over its momentum p~:
Z
dd p~ e E (p~)=T = V  mT d=2 ;
Z (T; V ) = V
(2.5.13)
(2 )d
2
whi h oin ides with Eq. (2.5.11) derived from the path integral.
Cal ulate the partition fun tion (2.5.5) for the free ase.
Sin e the path integral over '(~x; t) is Gaussian, it an be represented as

1
ln det 2 + m2
ln Z (T; V ) =
2

1
=
Sp ln 2 + m2
2 Z
dd p~ Sp ln D2 + ! 2  ;
1
V
(2.5.14)
=
2
(2 )d t

Problem 2.31
Solution

where

(2.5.15)
! = ~p2 + m2 :
We have used the fa t that the ~x variable is not restri ted while the remaining
tra e of the one-dimensional operator is to be al ulated with periodi boundary
onditions.
We shall perform the al ulation by expressing the tra e via the diagonal resolvent of the same operator as it was already done in the Problem 1.29. The
Green fun tion G! (t t0 ) is no longer given by Eq. (1.1.38) be ause of the periodi
boundary onditions. We get, instead, the sum over even frequen ies:
+1
0
X
e i2nT (t t)
G! (t t0 ) = T
;
(2.5.16)
(2nT )2 + ! 2
n= 1
whi h satis es G! (1=T ) = G! (0), as it should be for periodi boundary onditions,
and reprodu es Eq. (1.1.38) as T ! 0. The diagonal resolvent is given by
+1
X
1
!
1
(2.5.17)
G! (0) = T
2 + ! 2 = 2! oth 2T :
(2
nT
)
n= 1
Therefore,

D2 + !

Spt ln
=

Z!

2

Z!

d! 2

1=T
Z
0


e !=T

(2.5.18)

169

FINITE TEMPERATURES

modulo an ! -independent onstant. Substituting into Eq. (2.5.14), we get


ln Z (T; V ) =

dd p~ h ! + ln 1
(2 )d 2T


e !=T ;

(2.5.19)

whi h is the standard result for an ideal Bose gas in quantum statisti s modulo the
rst term on the RHS asso iated with the zero-point energy of the va uum.

2.5.2 QCD at nite temperature


QCD at nite temperatures is des ribed by the partition fun tion

Z (T; V ) =

DA D  D e

R 1=T
0

dt V d3 ~x L[A ; ; 
;

(2.5.20)

whi h is the proper analog of Eq. (2.5.5). The path integral is taken with the
boundary onditions

A (~x; 1=T ) = A (~x; 0) ;


(~x; 1=T ) =
(~x; 0) ;
 (~x; 1=T ) =
 (~x; 0) ;

(2.5.21)
(2.5.22)
(2.5.23)

whi h are periodi for the gauge eld (gluon) and antiperiodi for the Fermi
elds (quarks). The antiperiodi ity of the Fermi elds is related, roughly
speaking, with the famous extra minus sign of fermioni loops in the va uum
energy.
Cal ulate the partition fun tion for free massive one-dimensional
fermions with antiperiodi boundary onditions
 (1=T ) =
 (0) :
(1=T ) =
(0) ;
(2.5.24)

Problem 2.32

Solution

The al ulation is analogous to that of Problem 2.31. We get

ln Z (T; V ) = ln det (D + m) = Sp ln (D + m) :
(2.5.25)
The fermion Green fun tion Gm (t t0 ) is given by the sum over odd frequen ies:

Gm (t t0 ) = T

dt G! (0)

!
!
1
= + 2 ln 1
d! oth
T
2T
T

2.5.

whi h satis es Gm (1=T ) =


onditions.

+1
X

e i(2n+1)T (t t)
;
i(2n + 1)T + m
n= 1

(2.5.26)

Gm (0), as it should be for antiperiodi boundary

170
As T

CHAPTER 2.

! 0, we get
Gm (t t0 ) =

+1
Z

LATTICE GAUGE THEORIES

0

d e i(t t)
=  t t0
2 i + m

(2.5.27)

sin e the ontour of integration over  an be losed for t > t0 (t < t0 ) in the lower
(upper) half-plane. We have thus reprodu ed the fermioni Green fun tion (2.1.35)
from Problem 2.3.
The diagonal resolvent is given by
+1
X
1
1
m
Gm (0) = T
= tanh ;
(2.5.28)
i
(2
n
+
1)
T
+
m
2
2
T
n= 1
whi h di ers from Eq. (2.5.17) by the hange of the oth for the tanh. Therefore,
Zm


(2.5.29)
dm 1 tanh m = m + ln 1 + e m=T
T
2T
2T
modulo an m-independent onstant. The se ond term on the RHS involves the
plus sign whi h hara terizes the Fermi statisti s (remember that ! = m if there is
no spatial dimensions). If we were hoose periodi boundary onditions instead of
antiperiodi ones, we would get the minus sign as in Eq. (2.5.19) whi h is wrong for
fermions. The rst term on the RHS is asso iated again with the zero-point energy
of the va uum.

ln Z (T; V ) =

The dis ussion of the previous Subse tion about the relation between the
nite-temperature and Eu lidean formulations explains why the latter allows
to al ulate in QCD only stati quantities, say hadron masses or intera tion
potentials, whi h do not depend on the time. It is also worth noting that
we did not add a gauge xing term in Eq. (2.5.20) having in mind a latti e
quantization as before.
The latti e formulation of nite-temperature QCD is espe ially simple.
One should take an asymmetri latti e whose size along the temporal axis is
mu h smaller than that along the spatial ones:
1
Lt =
 L:
(2.5.30)
Ta
This guarantees the system to be in a thermodynami limit. Then the temperature is given by
1
;
(2.5.31)
T =
aLt

2.5.

171

FINITE TEMPERATURES

i.e. oin ide with the inverse extent of the latti e along the temporal axis.
The periodi boundary onditions are usually imposed on the latti e by onstru tion.
Sin e the latti e spa ing a and the bare oupling onstant g2 are related
by Eq. (2.2.85), the temperature (2.5.31) an be rewritten as

T =

1

exp
Lt QCD

Z

dg2
B(g2) :

(2.5.32)

Therefore, one an hange the temperature on the latti e varying either the
size along temporal axis Lt or g2 .

2.5.3 Con nement riterion at nite temperature


The Wilson's riterion of on nement, whi h is dis ussed in Subse t. 2.2.6, is
not appli able at nite temperatures. A proper riterion for on nement at
nite temperatures was proposed by Polyakov [Pol78.
The Polyakov riterion of on nement at nite temperature uses the thermal Wilson loop whi h goes along the temporal dire tion:

L(~x) = tr P e

R 1=T
0

dt Ad (~x;t)

(2.5.33)

It is gauge invariant be ause of the periodi boundary onditions for the gauge
eld and is alled the Polyakov loop or the Wilson line. One an imagine that
the time-variable t  xd is ompa ti ed so that the Polyakov loop winds
around the torus in the temporal dire tion as is shown in Fig. 2.28.
The latti e Polyakov loop

L~x = tr

xd

Ux;d

(2.5.34)

is just the tra e of the produ t of the link variables along a line whi h goes in
the temporal dire tion through the latti e with imposed periodi boundary
onditions.
Using the latti e gauge transformation (2.2.13), almost all link variables,
asso iated with links pointing in the temporal dire tion, an be put equal 1
ex ept for one time sli e sin e the gauge transformation is periodi :

(~x; 0) =
(~x; 1=T ) :

(2.5.35)

172

CHAPTER 2.

LATTICE GAUGE THEORIES









 A  Polyakov A 


loop








? t 6*~x



2.5.

While the latti e a tion is invariant under the transformation (2.5.39), the
Polyakov loop transforms as

L~x

Z =



Fig. 2.28: Polyakov loop whi h winds around ompa ti ed temporal dire tion.
The average of the Polyakov loop is related to the free energy F0 (~x) of
a single quark (minus that of the va uum) nailed at the point ~x of a threedimensional spa e by
= e F0 =T :

(2.5.36)

If F0 is in nite, whi h is asso iated with on nement, then

hL(~x)i

= 0

on nement :

de on nement

(2.5.37)
(2.5.38)

is asso iated with de on nement. This is the Polyakov riterion of on nement at nite temperature.
This riterion establishes a onne tion on a latti e between on nement
and the Z(3) symmetry | the enter of the SU(3). The Z(3)-transformation
of the link variables

Ux;d

! Zxd Ux;d

( Zxd 2 Z(3) )

( Z 2 Z(3) ) ;

(2.5.40)

(2.5.39)

leaves the latti e a tion to be invariant. This transformation is not of the


type of the lo al gauge transformation (2.2.13) sin e only the temporal linkvariables are transformed. The parameter Zxd of the transformation (2.5.39)
depends on xd but is independent of the spatial oordinates ~x so that the
symmetry is a global one.

xd

Zxd :

(2.5.41)

Therefore, Eq. (2.5.37) holds if the symmetry is unbroken, while Eq. (2.5.38)
signalizes about spontaneous breaking of the symmetry. Thus, on nement
or de on nement are asso iated with the unbroken or broken global Z(3)
symmetry, respe tively.
On a latti e of a nite volume, the number of degrees of freedom is nite
and spontaneous breaking of the Z(3) symmetry is impossible. It is more
onvenient to use then the riterion whi h is based on the orrelator of two
Polyakov loops separated by a distan e R along a spatial dire tion. This
orrelator determines the intera tion energy E (R) between a quark and an
antiquark by the formula

On the ontrary,

hL(~x)i 6=

! Z L~x

where



hL(~x)i

173

FINITE TEMPERATURES


L(~x)Ly (~y) onn = e E(R)=T :

(2.5.42)

A nite orrelation length is now asso iated with on nement while an in nite
one orresponds to de on ned quarks.
More about the Z(3) symmetry in nite-temperature latti e gauge theories
an be found in the review by Svetitsky [Sve86.
Cal ulate the orrelator (2.5.42) to the leading order of the strong
oupling expansion.
Solution The al ulation is analogous to that of Subse t. 2.2.5. The group integral
is nonvanishing when the plaquettes ompletely ll a ylinder, spanned by two
Polyakov loop, whose area is R=T . This is an analog of the lling shown in Fig. 2.10.
Contra ting the indi es, we get

Problem 2.33


L~x Ly~y onn = (Wp )R=T ;

(2.5.43)

where Wp is given by Eq. (2.2.72). This yields the same intera tion potential E (R)
as before (see Eqs. (2.2.76) and (2.2.77)).

174

CHAPTER 2.

LATTICE GAUGE THEORIES

Remark on high temperatures

g32d = g2 T ;

(2.5.44)

whi h has a dimension of [mass in three dimensions. The three dimensional


QCD and QED always on ne. If we take a Wilson loop in the form of a
re tangle along spatial dire tions in four-dimensional QCD at high temperature, its average oin ides with that in three dimensions and obeys the area
law. This does not mean, however, that we are in a on ning phase sin e the
on nement riterion at nite temperature is di erent [Pol78.

2.5.4 De on ning transition


The e e ts of nite temperatures are negligible under normal ir umstan es
in QCD where the typi al energy s ale is of the order of hundreds MeV while
the temperature, say, of T  300 K is asso iated with the energy17 kT 
3  10 8 MeV. However, for the times  s in the very early universe, energies
of thermal u tuations were  200 MeV, i.e. of the order of the mass of the meson. Therefore, -mesons an be reated out of the va uum at those times
while their density in a unit volume is des ribed by thermodynami s of an
ideal gas. Heavier hadrons are suppressed at these energies by the Boltzmann
fa tor.
The energy density E (T ) of the hadron matter is given by the standard
thermodynami al relation


E (T ) = V1  (1=T ) ln Z (T; V ) ;
V

(2.5.45)

with Z (T; V ) given by Eq. (2.5.20).


When the density of hadrons is small, E (T ) is given by the formula
=

X


T
gi m3i K1 (mi =T ) + 3m2i K2 (mi =T ) ;
2
2 i=;;!;:::

(2.5.46)

where g = 3, g = 9, g! = 3, : : : are the statisti al weights of the , , !,


: : : mesons while K1 and K2 are the modi ed Bessel fun tions.
17 Here k = 8:6  10 11 Mev/ K is Boltzmann onstant.

FINITE TEMPERATURES

175

Derive Eq. (2.5.46) starting from the partition fun tion (2.5.19).
For a dilute gas, the logarithm in Eq. (2.5.19) an be expanded in
exp ( E=T ). Therefore, we get
Z
p2 2
onst
d3 p~
ln Z (T; V ) =
+V
e ~p +m =T
3
T
(2 )
onst V T m2
=
+
K (m=T ) :
(2.5.47)
T
2 2 2
The se ond term on the RHS des ribes the lassi al statisti s of an ideal gas of
relativisti parti les. Equation (2.5.46) an now be derived by di erentiating this
formula with respe t to 1=T a ording to Eq. (2.5.45) and taking into a ount
statisti al weights of the hadron states. The zero-point energy term gives a T independent ontribution to Eh , whi h only hanges the referen e level of energy.
Problem 2.34

At high temperatures T ! 1, the temporal dire tion shrinks and the partition fun tion (2.5.20) redu es to a three-dimensional one with the oupling
onstant

Eh (T )

2.5.

Solution

At low temperatures, the hadron matter is in the on nement phase.


However, when the temperature is in reased, a phase transition asso iated
with de on nement o urs at some temperature T = T as was rst pointed
out by Polyakov [Pol78 and Susskind [Sus79. For T < T the intera tion
potential between stati quarks is linear, as is shown in Fig. 2.11a, while for
T > T the potential is de on ning, as is shown in Fig. 2.11b. The state
of the hadron matter with de on ned quarks and gluons is often alled the
quark-gluon plasma.
There exists a very simple physi al argument why the de on ning phase
transition must o ur in QCD when the temperature is in reased. It is based
on the string pi ture of on nement whi h was onsidered in Subse t. 2.2.6.
The string is made of the gluon eld between stati quarks in the on ning
phase, whi h are asso iated with the string end points. With in reasing the
temperature, a ondensation of strings of an in nite length will inevitably
o ur due to a large entropy of su h states, whi h orresponds to a de on ning
phase transition.
Problem 2.35 Derive the temperature of a phase transition for an elasti string
analyzing the temperature dependen e of its free energy.
Solution Let us onsider the thermodynami s of an elasti string with nailed end
points. For low temperatures, thermal u tuations of the length of the string are
suppressed by the Boltzmann fa tor sin e the energy is proportional to the length.
Therefore, the string is tightened along the shortest distan e between the quarks
whi h leads to a linear potential.
When the temperature is in reased, entropy e e ts asso iated with u tuations
of a shape of the string be ome essential. An in rement of the string length l by l
looses the energy
E l = K l ;
E =
(2.5.48)
l

176

CHAPTER 2.

LATTICE GAUGE THEORIES

where K is the string tension as before, but gains the entropy


S = S l :
l
The hange of the free energy is given by
h
i
F = E T S = K T S l :
l
A phase transition o urs at the temperature

2.5.

6 E =T 4

(2.5.49)

The energy density E (T ) is des ribed by a free gas of hadrons for low
temperatures, as is already mentioned, and by a free gas of quarks and gluons
at high temperatures. The latter statement is due to asymptoti freedom,
whi h says that the e e tive oupling onstant des ribing strong intera tion
at the temperature T is given by

g2 (T )

b ln

QCD
T

(2.5.52)

with

2
(2.5.53)
11 + Nf
3
and Nf being the number of fermion spe ies (or avors) whose mass is mu h
less than T . This formula has the same stru ture as the running onstant
g2 (Q) whi h des ribes strong intera tion at the momenta Q. Sin e Q  T for
thermal u tuations, these two oupling onstants oin ide with logarithmi
a ura y.18
The energy density E (T ) of the quark-gluon plasma is given by Boltzmann
law
2
(2.5.54)
Ep (T ) = gp 30 T 4 + B

b =

18 The perturbative al ulations in QCD at nite temperature are des ribed in the book

by Kapusta [Kap89.

hadrons

gh

82

plasma

gp

(2.5.50)

 1
;
(2.5.51)
T = K S
l
when the hanges of the energy and entropy ompensate ea h other, so that the free
energy eases to depend on l. Therefore, the phase transition is asso iated with
a ondensation of arbitrary long strings.

177

FINITE TEMPERATURES

Fig. 2.29: Temperature dependen e of the energy density of the hadron matter. E (T )

for the hadron and plasma phases are given by Eqs. (2.5.46) and (2.5.54).
The di eren e Ep Eh at the temperature T of the de on nement phase
transition is equal to the latent heat E .

where

gp = 2  8 +

7
 2  2  3  Nf
8

(2.5.55)

is the statisti al weight, i.e. the number of independent internal degrees of


freedom of the parti les of the ideal gas. There are 2 spin and 8 olor states
of gluons, and 2 spin, 2 parti le-antiparti le, 3 olor and Nf avor states of
quarks (Nf = 2 for the u- and d-quarks). The fa tor 7=8 is a usual one for
fermions.
The T -independent onstant B > 0 in Eq. (2.5.54) is asso iated with
the fa t that the va uum energy in the plasma phase is higher than in the
hadron one. In other words, the energy density of the perturbative va uum is
larger by B than that of a non-perturbative one. It is be ause of this energy
di eren e that hadrons are stable at low temperatures. Su h a shift of energy
densities between perturbative and non-perturbative va ua is typi al for the
bag models of hadrons.
Numeri al Monte Carlo simulations of latti e gauge theory at nite temperature indi ate that the de on ning phase transition is of the rst order
and o urs at T  200 MeV. The a tual dependen e of the energy density on
T , al ulated by the Monte Carlo method, is well des ribed by Eq. (2.5.46) for
T < T and Eq. (2.5.54) for T > T . This behavior is illustrated by Fig. 2.29.

178

CHAPTER 2.

LATTICE GAUGE THEORIES

Problem 2.36 Cal ulate T and the latent heat E approximating Eh by an ideal
gas of massless  -mesons.
Solution It is reasonable to disregard the mass of the  -mesons for T > 200 Mev.

Then

Eh (T ) = gh 30 T 4

(2.5.57)

and
2

Pp (T ) = gp 90 T 4 B :

(2.5.58)

The positive onstant B in the energy density (2.5.54) leads to a negative pressure in the plasma state at low temperatures. Therefore, the hadron phase is preferable at low temperatures. This is in a spirit of the bag model of hadrons. At high
energies the pressure is higher for the plasma phase, sin e

gp = 37 > gh = 3 ;

(2.5.59)

so that the plasma phase is realized. The pressure against T 4 is shown in Fig. 2.30
for both phases of the hadron matter.
The de on nement phase transition o urs when the pressures in both phases
oin ide. Therefore, we get

T 4 = 2 B
90 (gp gh )

(2.5.60)

and
E  Ep (T )

Eh (T ) = 4B :

(2.5.61)

If we were put gh = 0 in Eq. (2.5.60), this would hange the value of T by few
per ents. This justi es the approximation of massless  -mesons.

179

FINITE TEMPERATURES

(2.5.56)

with gh = 3 being due to the three isotopi states ( + ,  , and  ). Eh (T ) for the
plasma state is given by Eq. (2.5.54).
The pressure for the relativisti gases with the energy densities (2.5.56) and
(2.5.54) is given, respe tively, by

Ph (T ) = gh 90 T 4

2.5.

plasma








hadrons




T 4

T4

Fig. 2.30: Pressure against T 4 for the two phases of the hadron matter. The solid

and dashed lines represent Eqs. (2.5.57) and (2.5.58), respe tively. The
hadron phase is stable for T < T , while the plasma phase is stable for
T > T .

Remark on de on ning transition in osmology


The on ning phase transition from quark-gluon plasma to hadrons happened
in the early universe when its age was  s. The equation of state of the
hadron matter is des ribed by Eqs. (2.5.54), (2.5.58) before that time and by
Eqs. (2.5.46), (2.5.57) after that time. There are presumably no osmologi al onsequen es of this phase transition, whi h survive till our time, sin e
it happened too long ago. For instan e, u tuations of the hadron matter
density whi h might o ur just after the phase transition were washed out by
the further expansion. The on ning phase transition in the early universe is
onsidered in the review [CGS86, Se tion 6.

2.5.5 Restoration of hiral symmetry


The hiral symmetry is spontaneously broken in QCD at T = 0, as is dis ussed
in Subse t. 2.4.5. With in reasing the temperature, the hiral symmetry
should restore at some temperature T h (whi h not ne essarily oin ides with
T ) sin e perturbation theory

is
appli able at high T . This restoration o urs
as a phase transition with  being the proper order parameter. Therefore,
the quark ondensate is destroyed at T = T h. Monte Carlo simulations
indi ates this phase transition to be of the rst order.
However, there is a subtlety in the above string pi ture of quark on nement when virtual quarks are taken into a ount. The e e ts of virtual

180

CHAPTER 2.

LATTICE GAUGE THEORIES

2.5.

Fig. 2.31: Breaking of the ux tube by reating a quark-antiquark pair (depi ted by
quarks are suppressed when their mass m is in nitely large and the pi ture
of on nement is the same as in pure gluodynami s: quarks are permanently
on ned by strings made of the ux tubes of the gluon eld. This is asso iated
with a linear intera tion potential.
For light virtual quarks, the ux tube an break reating a quarkantiquark pair out of the va uum as is shown in Fig. 2.31. This happens
when the energy saved in the ux tube is large enough to ompensate the
kineti energy of the produ ed parti les. Hen e, the linear growth of the
potential will stop at su h distan es.
The average of the Polyakov loop (2.5.33) is no longer riterion for quark
on nement in the presen e of virtual quarks. The test stati quark an
always be s reened by an antiquark reated out of the va uum (a quark
reated at the same time will go to in nity). Therefore, the free energy F0 in
Eq. (2.5.36) is always nite so that hL(~x)i 6= 0 in both phases.
The e e ts of virtual quarks usually weaken a phase transition in a pure
gauge theory. If the de on ning phase transition in the SU(3) pure gauge theory was of the se ond order rather than of the rst order, it would presumably
disappear for an arbitrary large but nite value of m. Su h a phenomenon
happens in the Ising model where an arbitrary small external magneti eld
(whi h is an analog of the quark mass) destroys the se ond order phase transition. A dis ontinuity of hL(~x)i at the rst order de on ning transition ontinues in the T; m-plane as is illustrated by Fig. 2.32. It seems to terminate
at some value m of the quark mass.
This situation with the order parameter for the de on nement phase

transition is somewhat similar to that for the hiral phase transition.  vanishes in the unbroken phase only for m = 0. If m 6= 0 but is small, there is a
small expli it breaking of hiral symmetry due to the quark mass. Sin e the
hiral phase transition is
of the rst order for m = 0, it is natural to expe t
that a dis ontinuity of  ontinues in the T; m-plane up to some value
m h of the quark mass.

the open ir les) out of the va uum.

181

FINITE TEMPERATURES

Hadrons

Plasma

m h

0
0

T h
T

Fig. 2.32: Expe ted phase diagram of the hadron matter in the T; m-plane. The

de on ning phase transition starts at T = T for m = 1. hL(~x)i is its


order parameter for m > m . The hiral phase transition starts at T = T h
for m = 0.  is its order parameter for m < m h .

If m h < m , the phase diagram in the T; m-plane may look like that
shown in Fig. 2.32. In

the
intermediate region m h < m < m , the behavior
of neither hL(~x)i nor  an answer the question whether a phase transition
(or two separate transitions) o urs. A proper parameter, whi h signals about
a phase transition is this region, ould be the temperature dependen e of the
energy density E (T ) that undergoes dis ontinuities at the points of rst order
phase transitions.
It is worth noting that an alternative behavior of the phase diagram in the
T; m-plane, when m h > m , is not on rmed by Monte Carlo simulations.

182

CHAPTER 2.

LATTICE GAUGE THEORIES

2.6 Referen e guide


There is a lot of very good introdu tory le tures/reviews on latti e gauge theories. For a perfe t des ription of motivations and the latti e formulation, I
would re ommend the original paper [Wil74 and le tures [Wil75 by Wilson.
Various topi s of the latti e gauge theories are overed by the well-written
book by Creutz [Cre83. The book by Seiler [Sei82 ontains some mathemati ally rigorous results. The re ently published book [MM94 ontains a
ontemporary look at the latti e theories.
I shall also list some of the reviews on latti e gauge theories whi h might
be useful for deeper studies of the subje t. The strong oupling expansion
and the mean- eld method are dis ussed in [DZ83. The Monte Carlo method
and some results of numeri al simulations are onsidered in [CJR83, Mak84.
The fermion doubling problem and the Wilson fermions are dis ussed in the
le tures [Wil75.
An introdu tion to quantum eld theory at nite temperature is given
in the book by Kapusta [Kap89, whi h ontains, in parti ular, a dis ussion
of perturbation theory in QCD at nite temperature. Latti e gauge theory
aspe ts of the de on nement phase transition at nite temperature is onsidered in the review [Sve86. A des ription of the phases of the hadron matter at
various temperatures, a omparison with results of the Monte Carlo simulations and a dis ussion of the de on ning phase transition in the early universe
is ontained in the review [CGS86.
Most of the reviews mentioned above were written in the beginning or
middle of the eighties. Sin e that time a great progress happens in the
Monte Carlo simulations of latti e gauge theories. New al ulations are
performed on larger latti es and with better statisti s. A vast amount of
papers on latti e gauge theories was published during the last ten years.
There are more than 1000 subs ribers to an ele troni preprint ar hive
hep-latftp.s ri.fsu.edu. The pro eedings of the Latti e Conferen es
([Lat94 and for the pre eeding years) provide an updated information about
this subje t.

Referen es to Chapter 2
[Ake93

K. Akemi et al., S aling study of pure gauge latti e QCD by Monte


Carlo renormalization group method, Phys. Rev. Lett. 71 (1993)
3063.
[Are80
I.Ya. Aref'eva, Non-Abelian Stokes theorem, Theor. Math. Phys.
43 (1980) 353.
[Ban83 M. Bander, Equivalen e of latti e gauge and spin theories, Phys.
Lett. 126B (1983) 463.
[BC81
G. Bhanot and M. Creutz, Variant a tions and phase stru ture
in latti e gauge theory, Phys. Rev. D24 (1981) 3212.
[BCL77 A. Bardu i, R. Casalbuoni and L. Lusanna, Classi al s alar
and spinning parti les intera ting with external Yang{Mills eld,
Nu l. Phys. B124 (1977) 93.
[BDI74 R. Balian, J.M. Drou e and C. Itzykson, Gauge elds on a latti e.
I. General outlook, Phys. Rev. D10 (1974) 3376.
[Ber74
C.W. Bernard, Feynman rules for gauge theories at nite temperature, Phys. Rev. D9 (1974) 3312.
[Bra80
N. Brali , Exa t omputation of loop averages in two-dimensional
Yang{Mills theory, Phys. Rev. D22 (1980) 3090.
[BS92
G.S. Bali and K. S hilling, The running oupling from SU(3)
gauge theory, in Latti e 92, Pro . of the Int. Symp. on latti e
eld theory, eds. J. Smit and P. Van Baal, Nu l. Phys. B (Pro .
Suppl.) 30, 1993.
[BSSW77 A.P. Bala handran, P. Salomonson, B. Skagerstam and
J. Winnberg, Classi al des ription of a parti le intera ting with
a non-Abelian gauge eld, Phys. Rev. D15 (1977) 2308.
[CGL81 P. Cvitanovi h, J. Greensite and B. Lautrup, The rossover point
in latti e gauge theories with ontinuous gauge groups, Phys. Lett.
105B (1981) 197.
183

184
[CGS86
[CJR79

[CJR83
[Cre79
[Cre80
[Cre83
[Dro81
[DZ83
[Eli75
[Fey53
[FLZ82
[GL81
[GN80
[Gri78
[Ham83
[HJS77

REFERENCES TO CHAPTER 2

J. Cleymans, R.V. Gavai and E. Suhonen, Quarks and gluons at


high temperatures and densities, Phys. Rep. 130 (1986) 217.
M. Creutz, L. Ja obs and C. Rebbi, Experiments with a gaugeinvariant Ising system, Phys. Rev. Lett. 42 (1979) 1390; Monte
Carlo study of Abelian latti e gauge theories, Phys. Rev. D20
(1979) 1915.
M. Creutz, L. Ja obs and C. Rebbi, Monte Carlo omputations
in latti e gauge theories, Phys. Rep. 95 (1983) 201.
M. Creutz, Con nement and riti al dimensionality of spa etime, Phys. Rev. Lett. 43 (1979) 553; Monte Carlo study of quantized SU(2) gauge theory, Phys. Rev. D21 (1980) 2308.
M. Creutz, Asymptoti -freedom s ales, Phys. Rev. Lett. 45 (1980)
313.
M. Creutz, Quarks gluons and latti es, Cambridge Univ. Press,
1983 (Russian transl.: Mir., M., 1987).
J.M. Drou e, The mean- eld approximation in the SU(2) pure
latti e gauge theory, Phys. Lett. 105B (1981) 46.
J.-M. Drou e and J.-B. Zuber, Strong oupling and mean eld
methods in latti e gauge theories, Phys. Rep. 102 (1983) 1.
S. Elitzur, Impossibility of spontaneously breaking lo al symmetries, Phys. Rev. D12 (1975) 3978.
R.P. Feynman, Atomi theory of the  transition in Helium, Phys.
Rev. 91 (1953) 1291.
H. Flyvbjerg, B. Lautrup and J.B. Zuber, Mean eld with orre tions in latti e gauge theory, Phys. Lett. 110B (1982) 279.
J. Greensite and B. Lautrup, Phase transition and mean- eld
methods in latti e gauge theories, Phys. Lett. 104B (1981) 41.
J.L. Gervais and A. Neveu, The slope of the leading Regge traje tory in quantum hromodynami s, Nu l. Phys. B163 (1980)
189.
V.N. Gribov, Quantization of non-Abelian gauge theories, Nu l.
Phys. B139 (1978) 1.
H.W. Hamber, Latti e gauge theories at large NF , Phys. Lett.
126B (1983) 471.
M.B. Halpern, A. Jevi ki and P. Senjanovi , Field theories in
terms of parti le-string variables: spin, internal symmetries, and
arbitrary dimension, Phys. Rev. D16 (1977) 2476.

REFERENCES TO CHAPTER 2

[HP81

185

H. Hamber and G. Parisi, Numeri al estimates of hadron masses


in a pure SU(3) gauge theory, Phys. Rev. Lett. 47 (1981) 1792.
[Kad76 L.P. Kadano , Re ursion relations in statisti al physi s and eld
theory, Ann. Phys. 100 (1976) 359.
[Kap89 J.I. Kapusta, Finite-temperature eld theory, Cambridge Univ.
Press, 1989.
[Kob79 D.H. Kobe, Aharonov{Bohm e e t revisited, Ann. Phys. 123
(1979) 381.
[KM81
S.B. Khokhla hev and Yu.M. Makeenko, Phase transition over
the gauge group enter and quark on nement in QCD, Phys.
Lett. 101B (1981) 403.
[KMNP83 H. Kluberg-Stern, A. Morel, O. Napoly and B. Petersson, Flavors
of Lagrangian Susskind fermions, Nu l. Phys. B220 (1983) 447.
[KS75
J. Kogut and L. Susskind, Hamiltonian formulation of Wilson's
latti e gauge theories, Phys. Rev. D11 (1975) 395.
[KS81a L.H. Karsten and J. Smit, Latti e fermions: spe ies doubling,
hiral invarian e and the triangle anomaly, Nu l. Phys. B183
(1981) 103.
[KS81b N. Kawamoto and J. Smit, E e tive Lagrangian and dynami al
symmetry breaking in strongly oupled latti e QCD, Nu l. Phys.
B192 (1981) 100.
[Lat94
Latti e 94, Pro . of the XIIth Int. Symp. on latti e eld theory,
ed. F. Kars h et al., Nu l. Phys. B (Pro . Suppl.) 42, 1995.
[LN80
B. Lautrup and M. Nauenberg, Phase transition in four-dimensional ompa t QED, Phys. Lett. 95B (1980) 63.
[Lon27
F. London, Quantenme hanis he Deutung der Theorie von Weyl,
Z. Phys. 42 (1927) 375.
[LR82
C.B. Lang and C. Rebbi, Potential and restoration of rotational
symmetry in SU(2) latti e gauge theory, Phys. Lett. 115B (1982)
137.
[Mak84 Yu.M. Makeenko, The Monte Carlo method in latti e gauge theories, Usp. Fiz. Nauk 143 (1984) 161 (English translation: Sov.
Phys. Usp. 27 (1984) 401).
[Mig75
A.A. Migdal, Re ursion equations in gauge theories, ZhETF 69
(1975) 810 (English transl.: Sov. Phys. JETP 42 (1975) 413).
[MM94 I. Montvay and G. Munster, Quantum elds on a latti e, Cambridge Univ. Press, 1994.

186
[NN81
[Pol78
[Sei82
[STW81
[Sus77
[Sus79
[Sve86
[Weg71
[Wey19
[Wil74
[Wil75
[Wil77
[WK74
[Won70
[YM54

REFERENCES TO CHAPTER 2

H.B. Nielsen and M. Ninomiya, Absen e of neutrinos on a latti e,


1. Proof by homotopy theory, Nu l. Phys. B185 (1981) 20; 2.
Intuitive topologi al proof, Nu l. Phys. B193 (1981) 173.
A.M. Polyakov, Thermal properties of gauge elds and quark liberation, Phys. Lett. 72B (1978) 477.
E. Seiler, Gauge theories as a problem of onstru tive quantum
eld theory and statisti al me hani s, Le t. Notes in Physi s 159,
Springer-Verlag, 1982 (Russian translation: Mir., M., 1985).
H.S. Sharat handra, H.J. Thun and P. Weisz, Susskind fermions
on a Eu lidean latti e, Nu l. Phys. B192 (1981) 205.
L. Susskind, Latti e fermions, Phys. Rev. D16 (1977) 3031.
L. Susskind, Latti e models of quark on nement at high temperature, Phys. Rev. D20 (1979) 2610.
B. Svetitsky, Symmetry aspe ts of nite-temperature on nement
transitions, Phys. Rep. 132 (1986) 1.
F.J. Wegner, Duality in generalized Ising models and phase transitions without lo al order parameter, J. Math. Phys. 12 (1971)
2259.
H. Weyl, Eine neue Erweiterung der Relativitatstheorie, Ann. der
Phys. 59 (1919) v.10, pp.101{133.
K.G. Wilson, Con nement of quarks, Phys. Rev. D10 (1974)
2445.
K.G. Wilson, Quarks and strings on a latti e, in New phenomena
in a subnu lear physi s (Eri e 1975), ed. A. Zi hi hi, Plenum,
N.Y., 1977.
K.G. Wilson, Quantum hromodynami s on a latti e, in New developments in quantum eld theory and statisti al me hani s,
eds. M. Levy and P. Mitter, Plenum, N.Y., 1977.
K.G. Wilson and J. Kogut, The renormalization group and the
-expansion, Phys. Rep. 12C (1974) 75 (Russian transl.: Mir.,
M., 1975).
S.K. Wong, Field and parti le equations for the lassi al Yang{
Mills eld and parti les with isotopi spin, Nuovo Cim. 65A
(1970) 689.
C.N. Yang and R.L. Mills, Conservation of isotopi spin and isotopi gauge invarian e, Phys. Rev. 96 (1954) 191.

\Can I s hedule parades and then all


them o ?"
\But just send out announ ements postponing the parades. Don't even bother to
s hedule them."

, Cat h-22

J. Heller

Chapter 3
1=N

Expansion

In many physi al problems, espe ially when u tuations of s ales of di erent


orders of magnitude are essential, there is no small parameter whi h ould
simplify a study. A typi al example is QCD where the e e tive oupling,
des ribing strong intera tion at a given distan e, be omes large at large distan es so that the intera tion really be omes strong.
't Hooft [Hoo74a proposed in 1974 to use the dimensionality of the gauge
group SU(N ) as su h a parameter, onsidering the number of olors, N ,
as a large number and performing an expansion in 1=N . The motivation
was an expansion in the inverse number of eld- omponents N in statisti al
me hani s where it is known as the 1=N -expansion, and is a standard method
for non-perturbative investigations.
The expansion of QCD in the inverse number of olors rearranges diagrams
of perturbation theory in a way whi h is onsistent with a string pi ture of
strong intera tion, whose phenomenologi al onsequen es agree with experiment. The a ura y of the leading-order term, whi h is often alled multi olor
QCD or large-N QCD, is expe ted to be of the order of the ratios of meson
widths to their masses, i.e. about 10{15%.
While QCD is simpli ed in the large-N limit, it is still not yet solved.
Generi ally, it is a problem of in nite matri es, rather than of in nite ve tors
as in the theory of se ond-order phase transitions in statisti al me hani s.
We shall start this Chapter by showing how the 1=N -expansion works for
the O(N )-ve tor models, and des ribing some appli ations to the four-Fermi
intera tion, the '4 theory and the nonlinear sigma model. Then we shall
on entrate on multi olor QCD.
187

188

CHAPTER 3.

1=N

EXPANSION

3.1 O(N ) ve tor models

3.1.1 Four-Fermi theory


The a tion of the O(N )-symmetri four-Fermi theory in a d-dimensional Eu lidean spa e1 is de ned by


dd x  ^ + m 

G 
2

2

(3.1.1)

1 In d = 2 this model was studied in the large-N limit in Ref. [GN74 and is often alled

the Gross{Neveu model.

(N ) VECTOR MODELS

189

Here ^ =   and

The simplest models, whi h be ome solvable in the limit of a large number of
eld omponents, deal with a eld whi h has N omponents forming an O(N )
ve tor in an internal symmetry spa e. A model of this kind was rst onsidered by Stanley [Sta68 in statisti al me hani s and is known as the spheri al
model. The extension to quantum eld theory was done by Wilson [Wil73
both for the four-Fermi and '4 theories.
In the framework of perturbation theory, the four-Fermi intera tion is
renormalizable only in d = 2 dimensions and is non-renormalizable for
d > 2. The 1=N -expansion resums perturbation-theory diagrams after whi h
the four-Fermi intera tion be omes renormalizable to ea h order in 1=N for
2  d < 4. An analogous expansion exists for the nonlinear O(N ) sigma
model. The '4 theory remains trivial in d = 4 to ea h order of the 1=N expansion while has a nontrivial infrared-stable xed point for 2 < d < 4.
The 1=N expansion of the ve tor models is asso iated with a resummation
of Feynman diagrams. A very simple lass of diagrams | the bubble graphs
| survives to the leading order in 1=N . This is why the large-N limit of the
ve tor models is solvable. Alternatively, the large-N solution is nothing but
a saddle-point solution in the path-integral approa h. The existen e of the
saddle point is due to the fa t that N is large. This is to be distinguished from
a perturbation-theory saddle point whi h is due to the fa t that the oupling
onstant is small. Taking into a ount u tuations around the saddle-point
results in the 1=N -expansion of the ve tor models.
We begin this Se tion with a des ription of the 1=N -expansion of the
N - omponent four-Fermi theory analyzing the bubble graphs. Then we introdu e fun tional methods and onstru t the 1=N -expansion of the O(N )symmetri '4 theory and nonlinear sigma model. At the end we dis uss the
fa torization in the O(N ) ve tor models at large N .


S ;

3.1.

= ( 1; : : : ; N )

(3.1.2)

is a spinor eld whi h forms an N - omponent ve tor in an internal-symmetry


spa e so that


N
X
i=1

i i :

(3.1.3)

The dimension of the four-Fermi oupling onstant G is


dim [G = m2 d :

(3.1.4)

For this reason, the perturbation theory for the four-Fermi intera tion is
renormalizable in d = 2 but is non-renormalizable for d > 2 (and, in parti ular, in d = 4). This is why the old Fermi theory of weak intera tions was
repla ed by the modern ele troweak theory, where the intera tion is mediated
by the W  and Z bosons.
The a tion (3.1.1) an be equivalently rewritten as

S ; ;  =


dd x

 ^ + m 

2
 +
;
2G

(3.1.5)

where  is an auxiliary eld. The two forms of the a tion, (3.1.1) and (3.1.5),
are equivalent due to the equation of motion whi h reads in the operator
notation as
(3.1.6)
 = G :  :;
where : : : : : stands for the normal ordering of operators. Equation (3.1.6) an
be derived by varying the a tion (3.1.5) with respe t to .
In the path-integral quantization, where the partition fun tion is de ned
by

Z =

DD D e S[ ; ;

(3.1.7)



with S ; ;  given by Eq. (3.1.5), the a tion (3.1.1) appears after performing the Gaussian integral over . Therefore, one alternatively gets

Z =

D D e S[ ;

(3.1.8)

190

CHAPTER 3.

1=N

EXPANSION



with S ; given by Eq. (3.1.1).
The perturbative expansion of the O(N )-symmetri four-Fermi theory an
be onveniently represented using the formulation (3.1.5) via the auxiliary
eld . Then the diagrams are of the type of those in Yukawa theory, and
resemble the ones for QED with  and being an analog of the ele tronpositron eld and  being an analog of the photon eld. However, the auxiliary eld (x) does not propagate, sin e it follows from the a tion (3.1.5)
that

D0 (x y)  h (x)(y) iGauss =

G (d)(x

y)

(3.1.9)

or

3.1.

(N ) VECTOR MODELS

I 

6 6
 I


I - 
I
  


I  
I
 - 


a)

b)

 

(3.1.10)

in momentum spa e.
It is onvenient to represent the four-Fermi vertex as the sum of two terms

I 

I
 


I 
I
 

I 
 I



 

- 

I
  
I

where the empty spa e inside the vertex is asso iated with the propagator (3.1.9) (or (3.1.10) in momentum spa e). The relative minus sign makes
the vertex antisymmetri in both in oming and outgoing fermions as is pres ribed by the Fermi statisti s.
The diagrams that ontribute to se ond order in G for the four-Fermi vertex are depi ted, in these notations, in Fig. 3.1. The O(N ) indi es propagate
through the solid lines so that the losed line in the diagram in Fig. 3.1b
orresponds to the sum over the O(N ) indi es whi h results in a fa tor of N .
Analogous one-loop diagrams for the propagator of the - eld are depi ted
in Fig. 3.2.
Cal ulate the one-loop Gell-Mann{Low fun tion of the four-Fermi
theory in d = 2.
Solution Evaluating the diagrams in Fig. 3.1 whi h are logarithmi ally divergent
in d = 2, and noting that the diagrams in Fig. 3.2 do not ontribute to the wavefun tion renormalization of the - eld, whi h emerges to the next order in G, one
gets

Problem 3.1

B (G) =

(N

1) G2
:
2



d)

(3.1.11)

(3.1.12)







D0 (p)  h ( p)(p) iGauss = G

191




I  
I
 - 




e)

Fig. 3.1: Diagrams of the se ond order of perturbation theory for the four-Fermi
vertex. The diagram b) involves the sum over the O(N ) indi es.

The four-Fermi theory in 2 dimensions is asymptoti ally free as was rst noted by
Anselm [Ans59 and redis overed in Ref. [GN74.
The vanishing of the one-loop Gell-Mann{Low fun tion in the Gross{Neveu
model for N = 1 is related to the same phenomenon in the Thirring
model. The
2

latter model is asso iated with the ve tor-like intera tion 
of one spe ies
of fermions with  being the -matri es in 2 dimensions. Sin e a bispinor has
in d = 2 only two omponents 1 and 2 , both the ve tor-like and the s alar-like
intera tion (3.1.1) for N = 1 redu e to 1 1 2 2 sin e the square of a Grassmann
variable vanishes. Therefore, these two models oin ide. For the Thirring model, the
vanishing of the Gell-Mann{Low fun tion for any G was shown by Johnson [Joh61
to all loops.

Remark on auxiliary elds


An introdu tion of the auxiliary eld is often alled, in English s ienti terminology, the Hubbard{Stratonovi h transformation in analogy with statisti al me hani s. In spite of the original paper by Stratonovi h [Str57 being
published in a Russian journal, the proper Russian term is just \auxiliary

192

CHAPTER 3.

1=N

EXPANSION

3.1.

(N ) VECTOR MODELS



 

?
-6 -

6 ?
- -

a)

b)

I





 

193


I
6




I


6

I

Fig. 3.2: One-loop diagrams for the propagator of the - eld. The diagram b)




involves the sum over the O(N ) indi es.

I




I




 







 



Fig. 3.3: Bubble diagram whi h survives the large-N limit of the O(N ) ve tor models.

eld".
The perturbation-theory expansion of the O(N )-symmetri four-Fermi theory
ontains, in parti ular, the diagrams of the type depi ted in Fig. 3.3 whi h are
alled bubble graphs. Sin e ea h bubble has a fa tor of N , the ontribution of
the n-bubble graph is / Gn+1 N n whi h is of the order of
Gn+1 N n  G
(3.1.13)
as N ! 1 sin e
1
:
(3.1.14)
G 
N
Therefore, all the bubble graphs are essential to the leading order in 1=N .
Let us denote
n loops


= G + : : : + G2

+ Gn+1





+:::




(3.1.15)


6
6

I





Fig. 3.4: Some diagrams of the 1=N -expansion for the O(N ) four-Fermi theory. The
wavy line represents the (in nite) sum of the bubble graphs (3.1.15).

In fa t the wavy line is nothing but the propagator D of the  eld with the
bubble orre tions in luded. The rst term G on the RHS of Eq. (3.1.15) is
nothing but the free propagator (3.1.10).
Summing the geometri series of the fermion-loop hains on the RHS of
Eq. (3.1.15), one gets analyti ally2

D 1 (p) =

3.1.2 Bubble graphs as zeroth order in 1=N


I
6
6


1
G

dd k
(2)d



sp k^ + im k^ + p^ + im
:
(k2 + m2 ) ((k + p)2 + m2 )

(3.1.16)


This determines the exa t propagator of the  eld at large N . It is O N 1


sin e the oupling G is in luded in the de nition of the propagator.
The idea is now to hange the order of summation of diagrams of perturbation theory using 1=N rather than G as the expansion parameter. Therefore,
the zeroth-order propagator of the expansion in 1=N is de ned as the sum over
the bubble graphs (3.1.15) whi h is given by Eq. (3.1.16). Some of the diagrams of the new expansion for the four-Fermi vertex are depi ted in Fig. 3.4.
The rst diagram is proportional to G while the se ond and third ones are
proportional to G2 or G3 , respe tively, and therefore are of order O(N 1 ) or
O(N 2 ) with respe t to the rst diagram. The perturbation theory is thus
rearranged as the 1=N -expansion.
The general stru ture of the 1=N -expansion is the same for all ve tor
models, say, for the N - omponent '4 theory whi h is onsidered in the next
Subse tion.

2 Re all that the free Eu lidean fermioni propagator is given by S0 (p) = (ip^ + m) 1
due to Eqs. (3.1.5), (3.1.7) and the additional minus sign is asso iated with the fermion
loop.

194

CHAPTER 3.

1=N

EXPANSION

The main advantage of the expansion in 1=N for the four-Fermi intera tion, over the perturbation theory, is that it is renormalizable in d < 4 while
the perturbation-theory expansion in G is renormalizable only in d = 2.
Moreover, the 1=N -expansion of the four-Fermi theory in 2 < d < 4 demonstrates [Wil73 an existen e of an ultraviolet-stable xed point, i.e. a nontrivial zero of the Gell-Mann{Low fun tion.
Problem 3.2 Show that the 1=N -expansion of the four-Fermi theory is renormalizable in 2  d < 4 (but not in d = 4).
Solution In order to demonstrate renormalizability, let us analyze indi es of the
diagrams of the 1=N -expansion.
First of all, we shall get rid of an ultraviolet divergen e of the integral over the
d-momentum k in Eq. (3.1.16). The divergent part of the integral is proportional
to d 2 (logarithmi ally divergent in d = 2) with  being an ultraviolet uto . It
an be an elled by hoosing

2
G = g 2 d ;
N

(3.1.17)

where g 2 is a proper dimensionless onstant whi h is not ne essarily positive sin e


the four-Fermi theory is stable with either sign of G. The power of  in Eq. (3.1.17)
is onsistent with the dimension of G. This pres ription works for 2 < d < 4
where there is only one divergent term while another divergen y / p2 ln  emerges
additionally in d = 4. This is why the onsideration is not appli able in d = 4.
The propagator D(p) is therefore nite, and behaves at large momenta jpj  m
as
(3.1.18)
D(p) / 1d 2 :
jpj
The standard power- ounting arguments then show that the only divergent diagrams appear in the propagators of the and  elds, and in the -- threevertex. These divergen ies an be removed by a renormalization of the oupling g ,
mass, and wave fun tions of and .
This ompletes a demonstration of renormalizability of the 1=N -expansion for
the four-Fermi intera tion in 2  d < 4. For more detail, see Ref. [Par75.
Cal ulate in d = 3 the value of g in Eq. (3.1.17).
To al ulate the divergent part of the integral in Eq. (3.1.16), we put
p = 0 and m = 0. Remembering that the -matri es are 2  2 matri es in d = 3,
we get

3.1.

(N ) VECTOR MODELS

Note that the integral is linearly divergent in d = 3 and  is the uto for the
integration over jkj. This divergen e an be an elled by hoosing G a ording to
Eq. (3.1.17) with g equal to

g =  :

Cal ulate in d = 3 the oe ient of proportionality in Eq. (3.1.18).


Let us hoose G =  2 =N  as is pres ribed by Eqs. (3.1.17), (3.1.20) and
put in Eq. (3.1.16) m = 0 sin e we are interested in the asymptoti s jpj  m. Then
the RHS of Eq. (3.1.16) an be rearranged as

Solution

D 1 (p) =

Z
sp k^k^
=
2
k2 k2

d3 k
(2 )3

1
1
= 2
k2


Z 

djkj =


: (3.1.19)
2

2N

= 2N

d3 k
k2 + kp
3
2
(2 ) k (k + p)2
d3 k p2 + kp :
(2 )3 k2 (k + p)2

1
k2

(3.1.21)

This integral is obviously onvergent.


To al ulate it, we apply the standard te hnique of the -parametrization, whi h
is based on the formula
Z1
2
1 =
d e k :
(3.1.22)
2
k
0

We have
Z

d3 k p2 + kp =
(2 )3 k2 (k + p)2

Z1

d 1

Z1

d 2

d3 k p2 + kp e 1 k2 2 (k+p)2 (3.1.23)


(2 )3

after whi h the Gaussian integral over d3 k an easily be performed. We get then
Z1
Z1
1 2 2
N
1
1
2
1 + 2 p :
D (p) = 3=2 p d 1 d 2
(3.1.24)
e
4
( 1 + 2 )5=2
0

The remaining integration over 1 and 2 an easily be done introdu ing the new
variables 2 [0; 1 and x 2 [0; 1 so that

1 = x ;
2 = (1 x) ;
 ( 1 ; 2 ) = :
 (x; )

Solution

d3 k
(2 )3

(3.1.20)

Problem 3.4

Problem 3.3

Z 

195

(3.1.25)

This gives nally

D(p) =

8
:
N jpj

(3.1.26)

196

CHAPTER 3.

1=N

EXPANSION

3.1.

(N ) VECTOR MODELS

197

The diagram, whi h gives a non-vanishing ontribution to the three-vertex in


order 1=N , is depi ted in Fig. 3.5b. It reads analyti ally



 R
R



a)

(p1 ; p2 ) = 1 +

Fig. 3.5: Diagrams for the 1=N - orre tion to the - eld propagator (a) and the
three-vertex (b).

Equation (3.1.26) (or (3.1.18) in d dimensions) is remarkable sin e it shows that


the s ale dimension of the eld , whi h is de ned in Subse t. 1.3.5 by Eq. (1.3.66),
hanges its value from l = d=2 in perturbation theory to l = 1 in the zeroth
order of the 1=N expansion (remember that the momentum-spa e propagator of a
eld with the s ale dimension l is proportional to jpj2l d ). This appearan e of s ale
invarian e in the 1=N -expansion of the four-Fermi theory at 2 < d < 4 was rst
pointed out by Wilson [Wil73 and implies that the Gell-Mann{Low fun tion B(g )
has a zero at g = g whi h is given in d = 3 by Eq. (3.1.20).
Find the (logarithmi ) anomalous dimensions of the elds , , and
of the -- three-vertex in d = 3 to order 1=N .
Solution
The 1=N - orre tion to the propagator of the - eld is given by the
diagram depi ted in Fig. 3.5a). Sin e we are interested in an ultraviolet behavior,
we an put again m = 0. Analyti ally, we have

Problem 3.5

8i
N

Z 

d3 k k^ + p^
:
(2 )3 jkj(k + p)2

(3.1.27)

The (logarithmi ally) divergent ontribution emerges from the domain of integration
jkj  jpj so we an expand the integrand in p. The p-independent term vanishes
after integration over the dire tions of k so that we get


Z 

d3 k 1
(2 )3 jkj3


2
2
= ip^ 1 + 2 ln 2 + nite :
3 N p

S 1 (p) = ip^ 1 +

8
N

Z 

d3 k (k^ + p^1 )(k^ + p^2 ) ;


(2 )3 jkj(k + p1 )2 (k + p2 )2

(3.1.29)

where p1 and p2 the in oming and outgoing fermion momenta, respe tively. The
logarithmi domain is jkj  jpjmax with jpjmax being the largest of jp1 j and jp2 j.
This gives

b)

S 1 (p) = ip^ +

8
N

(3.1.28)

2
2
ln
+ nite :
 2 N p2max

(p1 ; p2 ) = 1

(3.1.30)

An analogous al ulation of the 1=N orre tion for the eld  is a bit more ompli ated sin e involves three two-loop diagrams (see Ref. [CMS93). The resulting
expression for D 1 (p) reads


N D(p)

 1


= 2+
g

 jpj
1
+
+ 2 2
2 8
N

jpj

2 2
ln + nite
3 p2



:(3.1.31)

The linear divergen e is an elled to order 1=N providing g is equal to




g =  1 + 1 ;
N

(3.1.32)

whi h determines g to order 1=N . After this D 1 (p) takes the form

D 1 (p) =

N jpj 1
8

16
2
ln 2
2
3 N p

(3.1.33)

To make all three expressions (3.1.28), (3.1.30), and (3.1.33) nite, we need logarithmi renormalizations of the wave fun tions of - and - elds and of the vertex
. This an be a hieved by multiplying them by the renormalization onstants
2

Zi () = 1 i ln 2
(3.1.34)

where  stands for a referen e mass s ale and i are anomalous dimensions. The
index i stands for , , or v for the - and -propagators or the three-vertex ,
respe tively. We have, therefore, al ulated
2
=
;
3 2 N
2
v =
;
2 N
16
 =
(3.1.35)
3 2 N

198

CHAPTER 3.

1=N

EXPANSION

to order 1=N . Due to Eq. (3.1.6)  oin ides with the anomalous dimension of the
omposite elds 

 =  :

(3.1.36)

Note, that

Z 2 Zv 2 Z = 1 :
(3.1.37)
This implies that the e e tive harge is not renormalized and is given by Eq. (3.1.32).
Thus, the nontrivial zero of the Gell-Mann{Low fun tion persists to order 1=N (and,
in fa t, to all orders of the 1=N -expansion).

Remark on s ale invarian e at xed point


The renormalization group says that


dg2
 =  exp
(3.1.38)
B(g2) ;
whi h is the same as Eq. (2.2.86) sin e the orrelation length  . If B has a
nontrivial xed point g2 near whi h

B(g2)

= b g2

g2

(3.1.39)

with b < 0, then the substitution into Eq. (3.1.38) gives


 b
:
(3.1.40)
g2 = g2 +

Therefore, the approa h to the riti al point is power-like rather than logarithmi as for the ase of g2 = 0 when
B(g2) = bg4 :
(3.1.41)
The latter bahavior of B results, after the substitution into Eq. (3.1.38), in
the logarithmi dependen e
1
g2 =
(3.1.42)
b ln 

when b < 0 whi h is asso iated with asymptoti freedom.


If g is hoosen exa tly at the riti al point g , then the renormalizationgroup equations

 d ln i
(3.1.43)
= i g 2 ;
d

3.1.

(N ) VECTOR MODELS

199

where i stands generi ally either for verti es or for inverse propagators,
possess the s ale invariant solutions
g2
(3.1.44)
i /  i( ) :
This omplements the heuristi onsideration of Subse t. 1.3.5 on the relation
between s ale invarian e and the vanishing of the Gell-Mann{Low fun tion.
For the four-Fermi theory in d = 3, Eq. (3.1.44) yields


1 p2
;
(3.1.45)
S (p) =
ip^ 2
 2  
p
8
;
(3.1.46)
D(p) =
N jpj 2
 2  v  2


p pp
(p1 ; p2 ) =
f 22 ; 1 2 2 ;
(3.1.47)
2
p1
p1 p1
where f is an arbitrary fun tion of the dimensionless ratios whi h is not
determined by s ale invarian e. The indi es here obey the relation
1
(3.1.48)
v = + 
2
whi h guarantees that Eq. (3.1.37), implied by s ale invarian e, is satis ed.
The indi es i are given to order 1=N by Eqs. (3.1.35). When expanded
in 1=N , Eqs. (3.1.45) and (3.1.46) obviously reprodu es Eqs. (3.1.28) and
(3.1.33). Therefore, one gets the exponentiation of the logarithms whi h
emerge in the 1=N -expansion. The al ulation of the next terms of the 1=N expansion for the indi es i is ontained in Ref. [Gra91.

Remark on onformal invarian e at xed point


S ale invarian e implies, in a renormalizable quantum eld theory, more general onformal invarian e as is rst pointed out in Refs. [MS69, GW70. The
onformal group in a d-dimensional spa e-time has (d + 1)(d + 2)=2 parameters as is illustrated by Table 3.1. More about the onformal group an be
found in the le ture by Ja kiw [Ja 72.
A heuristi proof [MS69 of the fa t that s ale invarian e implies onformal invarian e is based on the expli it form of the onformal urrent K ,
whi h is asso iated with the spe ial onformal transformation, via the energymomentum tensor:


K = 2x x x2   :

(3.1.49)

200

CHAPTER 3.

Group

Transformations

Lorentz
Poin are

d(d 1)
2

rotations

+ d translations

1=N

EXPANSION

# Parameters

x0 =
 x

d(d 1)
2

x0 = x + a

d(d+1)
2

3.1.

(N ) VECTOR MODELS

201

Equation (3.1.51), whi h results from onformal invarian e, unambiguously xes the fun tion f in Eq. (3.1.47). In ontrast to in nite-dimensional
onformal symmetry in d = 2, the onformal group in d > 2 is less restri tive.
It xes only the tree-point vertex while, say, the four-point vertex remains an
unknown fun tion of two variables.
Problem 3.6

1=N .

Cal ulate the integral on the RHS of Eq. (3.1.51) in d = 3 to order

The integral on the RHS of Eq. (3.1.51) looks in d = 3 very mu h like that
in Eq. (3.1.29) and an easily be al ulated to the leading order in 1=N when only
the region of integration over large momenta with jkj >
 jpjmax  maxfjp1 j; jp2 jg is
essential to this a ura y.
Let us rst note that the oe ient in front of the integral is / v  1=N , so
that one has to peak up the term  1= v in the integral for the vertex to be of order
1. This term omes from the region of integration with jkj >
 jpjmax . Re alling that
jp1 p2 j < jpjmax in Eu lidean spa e, one gets
Z 3
dk
k^ + p^2
1
k^ + p^1
1+

=
2
2

2 [(k + p1 )
[(k + p2 )2 1+  =2 jkj1+2  =2
Z1
1
dk2 =
=
;
(3.1.52)
2
[k 1+ v
v (p2max ) v

Solution

Weyl

+ 1 dilatation

Conformal + d spe ial onformal

x0 =  x
x0
(x0 )2

d2 +d+2
2

= xx2 + 

(d+1)(d+2)
2

Table 3.1: Contents and the number of parameters of groups of spa e-time symmetry.
Di erentiating, we get

 K = 2x  ;

(3.1.50)

whi h is analogous to Eqs. (1.3.67) and (1.3.68) for the dilatation urrent.
Therefore, both the dilatation and onformal urrents vanish simultaneously
when  is tra eless whi h is provided, in turn, by the vanishing of the
Gell-Mann{Low fun tion.
Conformal invarian e ompletely xes three-verti es as was rst shown by
Polyakov [Pol70 for s alar theories. The proper formula for the four-Fermi
theory reads [Mig71
( d2 ) ( d2 v )
( v )
dd k
k^ + p^2
k^ + p^1
1+

=
2
d=
2
2

 [(k + p1 )
[(k + p2 )2 1+  =2 jkjd

(p1 ; p2 ) = 2 v

2+2

 =2

; (3.1.51)

where the oe ient in the form of the ratio of the -fun tions is pres ribed
by the normalization (3.1.45) and (3.1.46) and the indi es are related by
Eq. (3.1.48) but an be arbitrary otherwise3.

 0 is imposed by the Kallen{Lehmann representation of the


propagator while there is no su h restri tion on  sin e it is a omposite eld.
3 The only restri tion

p2max

where Eq. (3.1.48) has been used and

2 v :
(3.1.53)
(p1 ; p2 ) =
p2max
While the integral in Eq. (3.1.52) is divergent in the ultraviolet for v < 0, this
divergen e disappears after the renormalization.
Equation (3.1.30) is reprodu ed by Eq. (3.1.52) when expanding in 1=N . This
dependen e of the three-vertex solely on the largest momentum is typi al for logarithmi theories in the ultraviolet region where one an put, say, p1 = 0 without
hanging the integral with logarithmi a ura y. This is valid if the integral is fast
onvergent in infrared regions whi h is our ase.

Remark on broken s ale invarian e


S ale (and onformal) invarian e at a xed point g = g holds only for large
momenta jpj  m. For smaller values of momenta, s ale invarian e is broken
by masses. In fa t, any dimensional parameter breaks s ale invarian e. If the
bare oupling g is hosen in the vi inity of g a ording to Eq. (3.1.40), then
s ale invarian e holds even in the massless ase only for jpj   while it is
broken if jpj <
 .

202

CHAPTER 3.

1=N

EXPANSION

3.1.3 Fun tional methods for '4 theory


The large-N solution of the O(N ) ve tor models, whi h is given by the sum
of the bubble graphs, an alternatively be obtained by evaluating the path
integral at large N by the saddle-point method. We shall restri t ourselves
to the s alar O(N )-symmetri '4 theory while the analysis of the four-Fermi
theory is quite analogous.
The a tion of the O(N )-symmetri '4 theory reads

S ['a =

dd x

1
1

( 'a )2 + m2 'a 'a + ('a 'a )2
2
2
8

(3.1.54)

where


'a = '1 ; : : : ; ' N :

(3.1.55)

The oupling  in the a tion (3.1.54) must be positive for the theory to be
well de ned. The verti es of Feynman diagrams are asso iated with .
Cal ulate the one-loop Gell-Mann{Low fun tion of the O(N )symmetri '4 theory in d = 4.
Solution The orresponding diagrams are similar to those of Fig. 3.1, though now
the arrows are not essential sin e the eld is real. The diagrams are logarithmi ally
divergent in 4 dimensions. Ea h diagram ontribute with the positive sign while
the diagram of Fig. 3.1b has now an extra ombinatori fa tor of 1=2. The diagrams of Fig. 3.2 result in a mass renormalization and there is no wave-fun tion
renormalization of the '- eld in one loop so that one gets

Problem 3.7

B () =

(N + 8)2
16 2

(3.1.56)

The positive sign in this formula is the same as for QED and is asso iated with
\triviality" of the '4 theory in 4 dimensions. It is also worth noting that the
oe ient (N + 8) is large even for N = 1.

Introdu ing the auxiliary eld (x) as in Subse t. 3.1.1, the a tion (3.1.54)
an be rewritten as


Z
 a
2
1 a
2
2
a
d
:
(3.1.57)
S [' ;  =
d x '  + m +  '
2
2

(N ) VECTOR MODELS

203

In other words  is again a omposite eld.


The orrelators of ''s and 's are determined by the generating fun tional

Z [J a ; K

D (x)
"R

D'a (x)
R

S ['a ;+ dd xJ a (x)'a (x)+ dd xK (x)(x) ;

e

(3.1.59)

whi h is a fun tional of the sour es J a and K for the elds 'a and  and
extends Eq. (1.2.49).
To make the path integral over (x) in Eq. (3.1.59) onvergent, we integrate at ea h point x over a ontour whi h is parallel to imaginary axis. This
is spe i to the Eu lidean formulation. The propagator of the - eld in the
Gaussian approximation reads

D0 (p) =

h ( p)(p) iGauss

;

(3.1.60)

whi h reprodu es the four-boson vertex of perturbation theory.


Sin e the integral over 'a is Gaussian, it an be expressed via the Green
fun tion




1

x
G (x; y; ) = y 2
(3.1.61)
 + m 2 + 
as

Z [J a ; K
=

"

d 2 1
d d a
a
d
D (x) e d x 2 + 2 d xd yJ (x)G(x;y;)J (y)+ d xK (x)(x)
N Sp ln G 1 [
2
:

e

(3.1.62)

We have used here the obvious notation

G 1 [ =

2 + m2 +  :

(3.1.63)

It will also be onvenient to use the short-hand notation

g f = hgjf i

dd x f (x)g(x) :

(3.1.64)

Then, Eq. (3.1.62) an be rewritten as

The two forms are equivalent due to the equation of motion

 = 2 : 'a 'a : :

3.1.

(3.1.58)

Z [J a ; K

"

D (x) e

 + 1 J a G[J a +K  N Sp ln G 1 [
2 2
2
:

(3.1.65)

204

CHAPTER 3.

1=N

EXPANSION

The exponent in Eq. (3.1.65) is O(N ) at large N so the path integral an


be evaluated as N ! 1 by the saddle-point method. The saddle-point eld
on guration

 (x) = sp (x)

(3.1.66)

is determined (impli itly) by the saddle-point equation

N
sp (x)
G (x; x; sp )
2

+ J a G (; x; sp ) G (x; ; sp ) J a + K (x) = 0 : (3.1.67)
2
If K  1=, ea h term here is O(1) sin e
1
 
(3.1.68)
N
in analogy with Eq. (3.1.14).
When the sour es J a and K vanish so that the last two terms on the LHS
of Eq. (3.1.67) equal zero, this equation redu es to
N
G (x; x; sp ) = 0 :
(3.1.69)
2
Its solution is x independent due to translational invarian e and an be
parametrized as
sp

sp = m2R m2

(3.1.70)

where m and mR are the bare and renormalized mass, respe tively. Equation (3.1.69) then redu es to the standard formula [Wil73

m2

m2R

N
2

1
dd k
d
2
(2) (k + m2R )

(3.1.71)

for the mass renormalization at large N .


To take into a ount u tuations around the saddle point, we expand

 (x) = sp +  (x)


where

 (x)

 N

1=2 :

(3.1.72)
(3.1.73)

3.1.

(N ) VECTOR MODELS

205

The Gaussian integration over  (x) determines the pre-exponential fa tor


in (3.1.65).
To onstru t the 1=N expansion of the generating fun tional (3.1.65), it
is onvenient to use the generating fun tional for onne ted Green's fun tions, whi h was already introdu ed in Eq. (1.2.52). It is usually denoted by
W [J a ; K and is related to the partition fun tion (3.1.59) by
a
Z [J a ; K = e W [J ;K :

(3.1.74)

Then we get

W [J a ; K =

1
N
 
Sp ln G 1 [sp
2 sp sp 2
1
+ J a G [sp J a + K sp
2


1
Sp ln D 1 [sp + O N 1 ;
2

(3.1.75)

where
1 (d)
N
D 1 (x; y; ) =
(x y )
G (x; y; ) G (y; x; )

2
a
+J G (; x; ) G (x; y; ) G (y; ; ) J a :
(3.1.76)
This operator emerges when integrating over the Gaussian u tuations
around the saddle point. The orresponding (last) term on the RHS of
Eq. (3.1.75) is asso iated with the pre-exponential fa tor and, therefore, is
 1.
The next terms of the 1=N expansion an be al ulated in a systemati
way by substituting (3.1.72) in Eq. (3.1.65) and performing the perturbative
expansion in .
If the sour es J a and K vanish so that the saddle-point value sp is given
by the onstant (3.1.70), than the RHS of Eq. (3.1.76) simpli es to

N
1 (d)
(x y)
G (x; y; sp ) G (y; x; sp ) :(3.1.77)

2
Remembering the de nition (3.1.61) of G and passing to the momentum-spa e
representation, we get
D 1 (x; y; sp ) =

D 1 (p) =

1


N
2

1
dd k
: (3.1.78)
2
d
2
(2) (k + mR ) ((k + p)2 + m2R )

206

CHAPTER 3.

1=N

EXPANSION

The sign of the rst term on the RHS is onsistent with Eq. (3.1.60).
Equation (3.1.78) is an analog of Eq. (3.1.16) in the fermioni ase and an
be alternatively obtained by summing bubble graphs of the type in Fig. 3.3
for
D (p) = h ( p)(p) i :
(3.1.79)
The extra symmetry fa tor 1=2 in Eq. (3.1.78) is the usual ombinatori one
for bosons. Therefore, the large-N saddle-point al ulation of the propagator (3.1.79) results pre isely in the zeroth-order of the 1=N -expansion.
We see from Eq. (3.1.75) the di eren e between perturbation theory and
the 1=N -expansion. The perturbation theory in  an be onstru ted as an
expansion (3.1.72) around the saddle point sp given again by Eq. (3.1.67),
with the omitted se ond term on the LHS, whi h is now justi ed by the fa t
that  is small (even for N  1). The se ond term on the RHS of Eq. (3.1.75),
whi h is asso iated with a one-loop diagram, appears in perturbation theory
as a result of Gaussian u tuations around this saddle-point.

Remark on the e e tive a tion


The e e tive a tion is a fun tional of the mean values of elds
W
W
'a l (x) = a ;
 l (x) =
(3.1.80)
J (x)
K (x)
in the presen e of the external sour es. The e e tive a tion is de ned as the
Legendre transformation of W [J a ; K by
['a l ;  l  W + J a 'a l + K  l ;
(3.1.81)
a
a
where the sour es J and K , whi h are regarded as fun tionals of ' l and  l ,
are to be determined by an inversion of Eq. (3.1.80). To the leading order in
1=N we get

J a (x) = G 1 [ l 'a l (x) + O N 1 ;

(3.1.82)
 l (x) = sp (x) + O N 1 :
When Eq. (3.1.82) (with the 1=N orre tion in luded) is substituted into
Eq. (3.1.81) and a ount is taken of the 1=N terms, most of them an el and
we arrive at the relatively simple formula
N
1
  + Sp ln G 1 [ l
['a l ;  l =
2 l l 2


1
1
+ 'a l G 1 [ l 'a l + Sp ln D 1 [ l + O N 1 ; (3.1.83)
2
2

3.1.

(N ) VECTOR MODELS

207

where

D 1 (x; y;  l ) =

1 (d)
N
(x y)
G (x; y;  l ) G (y; x;  l )

2
+'a l (x)G (x; y;  l ) 'a l (y)
(3.1.84)

oin iding with (3.1.76) to the leading order in 1=N .


The se ond and fourth terms on the RHS of Eq. (3.1.83), whi h involve
Sp, are asso iated with one-loop diagrams of the elds 'a and , respe tively,
in the lassi al ba kground elds 'a l and  l . Higher orders in 1=N are given
by diagrams whi h are one-parti le irredu ible with respe t to both ' and .
It follows immediately from the de nitions (3.1.80) and (3.1.81) that

= J a (x) ;
'a l (x)

= K (x) :
 l (x)

(3.1.85)

Therefore, 'a l (x) and  l (x) are determined in the absen e of external sour es
by the equations

['a l ;  l
= 0
'b l (x)

(3.1.86)

['a l ;  l
= 0:
 l (x)

(3.1.87)

and

Substituting (3.1.83) into Eqs. (3.1.86) and (3.1.87), we get to the leading
order in 1=N , respe tively, the equations


2 + m2 +  l (x) 'a l (x) = 0

(3.1.88)

and

 a
N
' l (x)'a l (x) +
G (x; x;  l ) :
(3.1.89)
2
2
The rst equation is just a lassi al equation of motion in an external eld
 l (x) while the se ond one is just the average of the (quantum) equation (3.1.58). Equation (3.1.89) is often alled the gap equation.
A solution to Eqs. (3.1.88) and (3.1.89) depends on what initial (or boundary) onditions are imposed.
 l (x) =

208

CHAPTER 3.

1=N

EXPANSION

Find translationally invariant solutions to Eqs. (3.1.88) and (3.1.89)


and al ulate the orresponding e e tive potential.
a
Solution The e e tive potential V (' l ;  l ) is de ned via the integrand in the
a
e e tive a tion [' l ;  l for translationally invariant

Problem 3.8

'a l (x) = 'a ;

 l = 

(3.1.90)

by the formula
= Vol:  V ('a ; ) :

(3.1.91)

From Eq. (3.1.83) we get at large N

V =

1 2 N
 +
2
2

dd k ln k2 + m2 +  + 1 m2 +  '2 ;


(2 )d
2

(3.1.92)

whi h obviously re overs Eqs. (3.1.88) and (3.1.89) after varying with respe t to
onstant 'a and .
It is onvenient to perform renormalization by introdu ing, in d = 4, the renormalized oupling R given by
1
1 1
= +
R
 2

Z 

1
N
2
d4 k
1
ln 2
= +
2
4
2
2
2
(2 ) k (k + mR )  32
mR

(3.1.93)

and

R
Assuming that R  2 (also mR
the form

R 1

(3.1.94)

  as usual) and representing Eq. (3.1.71) in

1 2 m2R R
N 2
 +
+
2R R
R
64 2 R

'a = 0 ;
R = m2R for m2R > 0 ;
(3.1.99)
2
2mR
2
2
' =
(3.1.100)
;
R = 0
for mR < 0 :
R
The rst of them is asso iated with an unbroken O(N ) symmetry, while the se ond
one orresponds to a spontaneous breaking of O(N ) down to O(N 1). Both
formulas look like the proper tree-level ones while the only e e t of loop orre tions
at large N is the renormalization of the oupling onstant and mass.
A subtle point is a question of stability of these solutions. For small deviations of '2 from the mean value given by Eqs. (3.1.99) and (3.1.100), the e e tive
potential VR is a monotoni ally in reasing fun tion of '2 , as an be shown for
R N < 32 2 eliminating the auxiliary eld R from Eq. (3.1.98) by solving the gap
equation (3.1.97) iteratively in '2 , and the solutions are lo ally stable. Both solutions are, however, unstable globally with respe t to large u tuations of the elds.
This an be seen by eliminating '2 from VR by solving the gap equation (3.1.97)
for '2 whi h yields


VR = 1 2R 1
2
R

N ln R
32 2 m2R

N 2 :
128 2 R

(3.1.96)
(3.1.97)

1
1

+ ln R2 + R '2 ;
2
mR
2

whi h obviously reprodu es Eqs. (3.1.96) and (3.1.97).

(3.1.98)

(3.1.101)

This fun tion is monotoni ally de reasing for very large

R > m2R e R N ;

(3.1.102)

where the theory be omes unstable. This is related to the usual problem of \triviality" of the '4 theory whi h makes sense only for small ouplings R N as an e e tive
theory and annot be fundamental at very small distan es of the order of

r
Problem 3.9

Equation (3.1.92) results then in the renormalized e e tive potential

VR =

209

Equations (3.1.96) and (3.1.97) possess the solution

(3.1.95)

R 'a = 0 ;

R N ln R
= m2R + R '2 :
32 2 m2R
2


(N ) VECTOR MODELS

322

=  + m2 :

m2R = m2
N 2 ;
R
 32 2
we rewrite Eqs. (3.1.88) and (3.1.89) as [S h74, CJP74


3.1.

de reases as

 mR 1 e

(3.1.103)

Find a solution to Eqs. (3.1.88) and (3.1.89) whi h exponentially

'a l (x) =  a mR e mR 
where 

162
R N

for 

! 1

(3.1.104)

 x4 and a is an O(N ) ve tor.

Solution The di eren e from the previous problem is that ' l is no longer translationally invariant along the time-variable due to the initial ondition (3.1.104). Let
us denote

'a l (x)
 l (x)

 a ( ) ;
 v( ) :

(3.1.105)

210

CHAPTER 3.

1=N

EXPANSION

and

v ( ) =  a ( )a ( ) + N
2
2

= 0

(3.1.106)

d3 k G (;  ; v ) ;
(2 )3 !

D  d ;
(3.1.108)
d
p
k 2 + m2
! =
(3.1.109)
and we have introdu ed the Fourier image of the Green fun tion (3.1.61)


=

d3 ~x e i~k~x G (; ~x); (; ~0); v

D


E
1


D2 + ! 2 + v

211

The nonlinear O(N ) sigma model4 in 2 Eu lidean dimensions is de ned by


the partition fun tion

(3.1.107)

where

G! (;  ; v )

(N ) VECTOR MODELS

3.1.4 Nonlinear sigma model

The the saddle-point equations (3.1.88) and (3.1.89) an be rewritten as



D2 + m2 + v ( ) a ( )

3.1.

(3.1.110)

with respe t to the spatial oordinate.


The solution to Eqs. (3.1.106) and (3.1.107) an easily be found to be
 a mR e mR  ;
(3.1.111)
a ( ) =
1  R162 e 2mR 

(3.1.112)
v ( ) = R a ( )a ( )
2
where the renormalized oupling
R
 R =
(3.1.113)
1 + 16RN2

di ers from Eq. (3.1.93) only by an additional nal renormalization and the renormalized mass mR is de ned in Eq. (3.1.71). This solution is nontrivial for  2  N
and obviously satis es the initial ondition (3.1.104).
The solution is so simple be ause the diagonal resolvent (3.1.110) takes on the
very simple form
1
v ( )
G! (;  ; v ) =
(3.1.114)
2! 4! (! 2 m2R )
for the potential v ( ) given by Eqs. (3.1.111), (3.1.112). This an be easily veri ed
by substituting into the Gelfand{Di key equation (1.1.123) with G = 1. This is a
feature of an integrable potential whi h was already dis ussed in Problem 1.29.
The fun tion a ( ) given by Eq. (3.1.111) des ribes large-N amplitudes of multiparti le produ tion at a threshold [Mak94.

Z =

D~n

~n2

1
e
g2

1
2

R 2
d x( ~n)2

(3.1.115)

where

~n = (n1 ; : : : ; nN )

(3.1.116)

is an O(N ) ve tor. While the a tion in Eq. (3.1.115) is pure Gaussian, the
model is not free due to the onstraint
1
~n2 (x) = 2 ;
(3.1.117)
g
whi h is imposed on the ~n eld via the (fun tional) delta-fun tion.
The sigma model in d = 2 is sometimes onsidered as a toy model for
QCD sin e it possesses:
1) asymptoti freedom [Pol75;
2) instantons for N = 3 [BP75.

The a tion in Eq. (3.1.115) is  N as N ! 1 but the entropy, i.e. a


ontribution from the measure of integration, is also  N so that a straightforward saddle point is not appli able.
To over ome this di ulty, we pro eed as in the previous Subse tion,
introdu ing an auxiliary eld u(x), whi h is  1 as N ! 1, and rewrite the
partition fun tion (3.1.115) as
Z

"

Du (x)

D~n (x) e

1
2


R 2 
d x ( ~n)2 u ~n2 g12
;

(3.1.118)

where the ontour of integration over u(x) is parallel to imaginary axis.


Doing the Gaussian integration over ~n, we get

"

Du (x) e

N Sp ln(  2 +u(x))+ 1

2
2g2

R 2
d x u(x)

(3.1.119)

4 The name omes from elementary parti le physi s where a nonlinear sigma model in
4 dimensions is used as an e e tive Lagrangian for des ribing low-energy s attering of the
Goldstone -mesons.

212

CHAPTER 3.

1=N

EXPANSION

The rst term in the exponent is as before nothing but the sum of one-loop
diagrams in 2 dimensions

N
Sp ln
2

2 + u (x) =

1
n
n






 



NG (x; x; usp ) = 0

(3.1.121)

is quite analogous to Eq. (3.1.69) for the '4 theory while G is de ned by

G (x; y; u) =



y

1
x ;
2 + u

(3.1.122)

whi h is an analog of Eq. (3.1.61).


The oupling g2 in Eq. (3.1.121) is  1=N as is pres ribed by the onstraint (3.1.117) whi h involves a sum over N terms on the LHS. This guarantees that a solution to Eq. (3.1.121) exists. Next orders of the 1=N -expansion
for the 2-dimensional sigma model an be onstru ted analogously to the
previous Subse tion.
The 1=N -expansion of the 2-dimensional nonlinear sigma model has many
advantages over perturbation theory, whi h is usually onstru ted solving
expli itly the onstraint (3.1.117), say, hoosing
v
u

NX1
1u
nN = t1 g2
n2a
g
a=1

(N ) VECTOR MODELS

(3.1.123)

and expanding the square root in g2 . Only N 1 dynami al degrees of freedom


are left so that the O(N )-symmetry is broken in perturbation theory down to

213

O(N 1). The parti les in perturbation theory are massless (like Goldstone
bosons) and it su ers from infrared divergen ies.
On the ontrary, the solution to Eq. (3.1.121) has the form

usp = m2R  2 e

(3.1.120)

where the auxiliary eld u is denoted again by the wavy line. Equation (3.1.119) looks very mu h like Eq. (3.1.65) if we put there J a = K = 0.
The di eren e is that the exponent in (3.1.119) involves the term whi h is
linear in u, while the analogous term in (3.1.65) is quadrati in .
Now the path integral over u(x) in Eq. (3.1.119) is a typi al saddle-point
one: the a tion  N while the entropy  1 sin e only one integration over u
is left. The saddle-point equation for the nonlinear sigma model
1
g2

3.1.

4
Ng2

(3.1.124)

where  is an ultraviolet uto . Therefore, all N parti les a quires the same
mass mR in the 1=N -expansion so that the O(N ) symmetry is restored. This
appearan e of mass is due to dimensional transmutation whi h says in this
ase that the parameter mR rather than the renormalized oupling onstant
gR2 is observable. The emergen e of the mass ures the infrared problem.
Show that (3.1.124) is a solution to Eq. (3.1.121).
Let us look for a translationally invariant solution

Problem 3.10
Solution

usp (x)

 m2R :

(3.1.125)

Then Eq. (3.1.121) in the momentum spa e reads


1
= N
g2

Z 

1
N
d2 k
=
(2 )2 k2 + m2R
4

Z 2
0

dk2
2
k + m2R

N 2
ln
: (3.1.126)
4 m2R

The exponentiation results in Eq. (3.1.124).


Equation (3.1.126) relates the bare oupling g 2 and the uto  and allows us
to al ulate the Gell-Mann{Low fun tion yielding

Ng 4 :
(3.1.127)
2
The analogous one-loop perturbation-theory formula for any N reads [Pol75
2

=
B(g2 )  ddg


2)g 4
:
(3.1.128)
2
Thus, the sigma-model is asymptoti ally free in 2-dimensions for N > 2 whi h is
the origin of the dimensional transmutation. There is no asymptoti freedom for
N = 2 sin e O(2) is Abelian.

B(g2 ) =

(N

3.1.5 Large-N fa torization in ve tor models


The fa t that a path integral has a saddle point at large N implies a very
important feature of large-N theories | the fa torization. It is a general
property of the large-N limit and holds not only for the O(N ) ve tor models.
However, it is useful to illustrate it by these solvable examples.

214

CHAPTER 3.

1=N

EXPANSION

The fa torization at large N holds for averages of singlet operators, for


example

h u (x1 ) : : : uZ(xk ) i
 Z 1 Du e
"

N Sp ln[  2 +u+ 1

2
2g2

R 2
d xu

u (x1 )    u (xk ) (3.1.129)

in the 2-dimensional sigma model.


Sin e the path integral has a saddle point at some5

u (x) = usp (x) ;

(3.1.130)

we get to the leading order in 1=N :

h u (x1 ) : : : u (xk ) i

= usp (x1 ) : : : usp (xk ) + O

whi h an be written in the fa torized form

1
; (3.1.131)
N

h u (x1 ) i : : : h u (xk ) i + O N1 : (3.1.132)


Therefore, u be omes \ lassi al" as N ! 1 in the sense of the 1=N h u (x1 ) : : : u (xk ) i

expansion. This is an analog of the WKB-expansion in h = 1=N . \Quantum"


orre tions are suppressed as 1=N .
We shall return to dis ussing the large-N fa torization in the next Se tion
when onsidering the large-N limit of QCD.

3.2.

215

MULTICOLOR QCD

3.2 Multi olor QCD


The method of the 1=N -expansion an be applied to QCD. This was done
by 't Hooft [Hoo74a using the inverse number of olors for the gauge group
SU(N ) as an expansion parameter.
For a SU(N ) gauge theory without virtual quark loops, the expansion
goes in 1=N 2 and rearranges diagrams of perturbation theory a ording to
their topology. The leading order in 1=N 2 is given by planar diagrams, whi h
have a topology of a sphere, while the expansion in 1=N 2 plays the role of a
topologi al expansion. This reminds an expansion in the string oupling onstant in string models of the strong intera tion, whi h also has a topologi al
hara ter.
Virtual quark loops an be easily in orporated in the 1=N -expansion. One
distinguishes between the 't Hooft limit when the number of quark avors Nf
is xed as N ! 1 and the Veneziano limit [Ven76 when the ratio Nf =N
is xed as N ! 1. Virtual quark loops are suppressed in the 't Hooft limit
as 1=N and lead in the Veneziano limit to the same topologi al expansion as
dual-resonan e models of strong intera tion.
The simpli ation of QCD in the large-N limit is due to the fa t that the
number of planar graphs grows with the number of verti es only exponentially
rather than fa torially as do the total number of graphs. Correlators of gauge
invariant operators fa torize in the large-N limit whi h looks like the leadingorder term of a \semi lassi al" WKB-expansion in 1=N .
We begin this Se tion with a des ription of the double-line representation
of diagrams of QCD perturbation theory and rearrange it as the topologi al
expansion in 1=N . Then we dis uss some properties of the 1=N -expansion
for a generi matrix-valued eld.

3.2.1 Index or ribbon graphs


In order to des ribe the 1=N -expansion of QCD, whose extension to N
olors has already been onsidered in Subse t. 2.1.1, it is onvenient to use
the matrix- eld representation (2.1.5). In this Se tion we shall use a slightly
di erent de nition
X
[A (x)ij =
Aa (x) [ta ij ;
(3.2.1)
a

5 This saddle point is, in fa t, x-independent due to translational invarian e.

whi h is similar to that used by 't Hooft [Hoo74a and di ers from (2.1.5) by
a fa tor of ig:
Aij (x) = igAij (x) :
(3.2.2)

216

CHAPTER 3.

1=N

EXPANSION

The matrix (3.2.1) is Hermitean.


The propagator of the matrix eld Aij (x), in this notation, takes on the
form



ij

1 il kj 1 ij kl
kl
A (x) A (y) Gauss =

D (x y) ; (3.2.3)
2
N
where we have assumed, as usual, a gauge xing to de ne the propagator in
perturbation theory. For instan e, one has

D (x y) =

1 
42 (x y)2

(3.2.4)

in the Feynman gauge.


Equation (3.2.3) an be immediately derived from the standard formula

a
A

 (x)

Ab (y)

Gauss

= ab D (x y)

(3.2.5)

multiplying by the generators of the SU(N ) gauge group a ording to the


de nition (2.1.5) and using the ompleteness ondition
2
NX
1

a=1

(ta )ij (ta )kl =

1 il kj

2

1 ij kl

N

for SU(N ) ; (3.2.6)

where the fa tor of 1=2 is due to the normalization (2.1.6). Eq. (3.2.3) an
be derived dire tly from a path integral over matri es.
We on entrate in this Se tion only on the stru ture of diagrams in the
index spa e, i.e. the spa e of the indi es asso iated with the SU(N ) group.
We shall not onsider, in most ases, spa e-time stru tures of diagrams whi h
are pres ribed by Feynman's rules.
Omitting at large N the se ond term in parentheses on the RHS of
Eq. (3.2.3), we depi t the propagator by the double line

i - l
:
(3.2.7)
j  k
Ea h line represents the Krone ker delta-symbol and has orientation whi h
is indi ated by arrows. This notation is obviously onsistent with the spa etime stru ture of the propagator whi h des ribes a propagation from x to
y.
The arrows are due to the fa t that the matrix Aij is Hermitean and its
o -diagonal omponents are omplex onjugate. The independent elds are,

Aij (x) Akl


 (y )

Gauss

/ il kj

3.2.

217

MULTICOLOR QCD

say, the omplex elds Aij for i > j and the diagonal real elds Aii . The
arrow represents the dire tion of the propagation of the indi es of
the omplex
ij 
eld Aij for i > j while the omplex- onjugate one, Aji
 = A , propagates
in the opposite dire tion. For the real elds Aii , the arrows are not essential.
The double-line notation (3.2.7) looks similar to that of Subse t. 2.2.5.
The reason for that is deep: double lines appear generi ally in all models
des ribing matrix elds in ontrast to ve tor (in internal symmetry spa e)
elds whose propagators are depi ted by single lines as in the previous Se tion.
The three-gluon vertex, whi h is generated by the a tion (2.1.14), is depi ted in the double-line notations as
j1

i1
... ....
... ...
i2 ............................................j3
.
........
.
.
.
.
.
...
j2
i3

i1

j1
.... .....
... ...
j2 ...........................................i3
.
.
........
.
.
.
.
...
i2
j3

/ g i1 j3 i2 j1 i3 j2 i1 j2 i3 j1 i2 j3

(3.2.8)

where the subs ripts 1, 2 or 3 refer to ea h of the three gluons. The relative minus sign is due to the ommutator in the ubi in A term in the
a tion (2.1.14). The olor part of the three-vertex is antisymmetri under
inter hanging the gluons. The spa e-time stru ture, whi h reads in the momentum spa e as

1 2 3 (p1 ; p2 ; p3 )
= 1 2 (p1 p2 )3 + 2 3 (p2 p3 )1 + 1 3 (p3 p1 )2 ; (3.2.9)
is antisymmetri as well. We onsider all three gluons as in oming so that
their momenta obey p1 + p2 + p3 = 0. The full vertex is symmetri as is
pres ribed by Bose statisti s.
The olor stru ture in Eq. (3.2.8) an alternatively be obtained by multiplying the standard vertex
a1 a2 a3
1 2 3 (p1 ; p2 ; p3 )

= f a1 a2 a3 1 2 3 (p1 ; p2 ; p3 )

(3.2.10)

by (ta1 )i1 j1 (ta2 )i2 j2 (ta3 )i3 j3 , with f ab being the stru ture onstants of the
SU(N ) group, and using the formula

f a1 a2 a3 (ta1 )i1 j1 (ta2 )i2 j2 (ta3 )i3 j3


1 i1 j3 i2 j1 i3 j2 i1 j2 i3 j1 i2 j3 


;
=
2
whi h is a onsequen e of the ompleteness ondition (3.2.6).

(3.2.11)

218

CHAPTER 3.

1=N

EXPANSION

3.2.

models and is pres ribed by the asymptoti -freedom formula

.....................
....... ....................................
................
...... ....
.
.
.... ...
.......
.. ...
..... ...
.. ..
.
..
... ....
..
... ...
.... .....
.
... ....
.
.
.
. ..
.... ......
...... ............ .........................
.......... .... ...........
...........

The four-gluon vertex involves six terms | ea h of them is depi ted by a


ross | whi h di er by inter hanging of the olor indi es. We depi t the olor
stru ture of the four-gluon vertex for simpli ity in the ase when i1 = j2 = i,
i2 = j3 = j , i3 = j4 = k, i4 = j1 = l but i; j; k; l take on di erent values.
Then only the following term is left
i

l .....
..... i
....
.....
k
j
... ...
k j

/ g2
(3.2.12)

and there are no deltas on the RHS sin e the olor stru ture is xed. In other
words, we pi k up only one olor stru ture by equaling indi es pairwise.
The diagrams of perturbation theory an now be ompletely rewritten
in the double-line notation [Hoo74a. The simplest one whi h des ribes the
one-loop orre tion to the gluon propagator is depi ted in Fig. 3.6.6 This
diagram involves two three-gluon verti es and a sum over the N indi es whi h
is asso iated with the losed index line analogous to Eq. (2.2.70). Therefore,
the ontribution of this diagram is  g2 N .
In order for the large-N limit to be nontrivial, the bare oupling onstant
g2 should satisfy

g2

 N1 :

(3.2.13)

This dependen e on N is similar to Eqs. (3.1.14) and (3.1.68) for the ve tor
6 Here and in the most gures below the arrows of the index lines are omitted for

simpli ity.

(3.2.14)

of the pure SU(N ) gauge theory.


Thus, the ontribution of the diagram of Fig. 3.6 is of order

The sum over the N indi es is asso iated with the losed index line. The
ontribution of this diagram is  g2 N  1.

... ....

242
11N ln (=QCD )

g2 =

Fig. 3.6: Double-line representation of a one-loop diagram for the gluon propagator.

l.

219

MULTICOLOR QCD

Fig. 3.6

 g2 N 

(3.2.15)

in the large-N limit.


The double lines of the diagram in Fig. 3.6 an be viewed as bounding
a pie e of a plane. Therefore, these lines represent a two-dimensional obje t
rather than a one-dimensional one as the single lines do in ve tor models.
These double-line graphs are often alled in mathemati s the ribbon graphs
or fatgraphs. We shall see below their onne tion with Riemann surfa es.

Remark on the U(N ) gauge group


As is said above, the se ond term in the parentheses on the RHS of Eq. (3.2.6)
an be omitted at large N . Su h a ompleteness ondition emerges for the
U(N ) group whose generators T A (A = 1; : : : ; N 2 ) are

TA =

tr T AT B =

ta ; p2IN ;

1 AB
:
2

(3.2.16)

They obey the ompleteness ondition


2

N
X
A=1

TA

ij

TA

kl

1 il kj

2

for U(N ) :

(3.2.17)

The point is that elements of both the SU(N ) group and the U(N ) group
an be represented in the form

U = eiB ;

(3.2.18)

where B is a general Hermitean matrix for U(N ) and a tra eless Hermitean
matrix for SU(N ).
Therefore, the double-line representation of perturbation-theory diagrams
whi h is des ribed in this Se tion holds, stri tly speaking, only for the U(N )
gauge group. However, the large-N limit of both the U(N ) group and the
SU(N ) group is the same.

220

CHAPTER 3.

1=N

EXPANSION

..................
...................................................
...........
............... ...
.
.
.
.... ...
.. ...
... ...
... ...
.. ..
.... ....
.. ..
. ...
..
... ..
.... ...
.. ...
.
... ..
.
. ...
..........
............
..............
.
.
.
.
.
.
......... ...................... ..........
............................

The sum over the N indi es is asso iated with ea h of the four losed index
lines whose number is equal to the number of loops. The ontribution of
this diagram is  g8 N 4  1.

The double-line representation of perturbation theory diagrams in the index


spa e is very onvenient to estimate their orders in 1=N . Ea h three- or
four-gluon vertex ontributes a fa tor of g or g2 , respe tively. Ea h losed
index line ontributes a fa tor of N . The order of g in 1=N is given by
Eq. (3.2.13).
Let us onsider a typi al diagram for the gluon propagator depi ted in
Fig. 3.7. It has eight three-gluon verti es and four losed index lines whi h
oin ides with the number of loops. Therefore, the order of this diagram in
1=N is
4

 1:

(3.2.19)

The diagrams of the type in Fig. 3.7, whi h an be drawn on a sheet of


a paper without rossing any lines, are alled the planar diagrams. For su h
diagrams, an adding of a loop inevitably results in adding of two three-gluon
(or one four-gluon) vertex. A planar diagram with n2 loops has n2 losed
index lines. It is of order

n2 -loop planar diagram

 g2 N

n2

Fig. 3.8: Double-line representation of a three-loop non-planar diagram for the

gluon propagator. The diagram has six three-gluon verti es but only
one losed index line (while three loops!). The order of this diagram is
 g6 N  1=N 2 .

The order of this diagram in 1=N is

3.2.2 Planar and non-planar graphs

 g 2 N

221

MULTICOLOR QCD

...............
...................... .........
.................. .......................
.
.
.
.
... .............
....... ...
... ... ................... ................ ..........
.......................
... ....
.. .
...
... ....
... ......................
... ...
.
.
.
.
... ...
.
.
.
.
.
.
..
... .... .................. ...........................
.... . ....
.... ..........
.................. ..................
........... ...... ..........
...........

Fig. 3.7: Double-line representation of a four-loop diagram for the gluon propagator.

Fig. 3.7

3.2.

 1;

(3.2.20)

so that all planar diagrams survive in the large-N limit.


Let us now onsider a non-planar diagram of the type depi ted in Fig. 3.8.
This diagram is a three-loop one and has six three-gluon verti es. The rossing
of the two lines in the middle does not orrespond to a four-gluon vertex and
is merely due to the fa t that the diagram annot be drawn on a sheet of a
paper without rossing the lines. The diagram has only one losed index line.

Fig. 3.8

 g6 N  N12 :

(3.2.21)

It is therefore suppressed at large N by 1=N 2.


The non-planar diagram in Fig. 3.8 an be drawn without line- rossing on
a surfa e with one handle whi h is usually alled in mathemati s a torus or
the surfa e of genus one. A plane is then equivalent to a sphere and has genus
zero. Adding a handle to a surfa e produ es a hole a ording to mathemati al
terminology. A general Riemann surfa e with h holes has genus h.
The above evaluations of the order of the diagrams in Figs. 3.6{3.8 an
now be des ribed by the unique formula


1 genus
genus-h diagram 
:
(3.2.22)
N 2
Thus, the expansion in 1=N rearranges perturbation-theory diagrams a ording to their topology [Hoo74a. For this reason, it is referred to as the topologi al expansion or the genus expansion. The general proof of Eq. (3.2.22)
for an arbitrary diagram is given in Subse t. 3.2.4.
Only planar diagrams, whi h are asso iated with genus zero, survive in the
large-N limit. This lass of diagrams is an analog of the bubble graphs in the
ve tor models. However, the problem of summing the planar graphs is mu h
more ompli ated than that of summing the bubble graphs. Nevertheless, it
is simpler than the problem of summing all the graphs, sin e the number of
the planar graphs with n0 verti es grows at large n0 exponentially [Tut62,
KNN77
#p (n0 )  # of planar graphs

onstn0 ;

(3.2.23)

222

CHAPTER 3.

1=N

EXPANSION

..
..
..
.. ..
.. ..
.... ...
. . .
.. ...
... ... .... ............
.
..
.
.
.. ....
... ... ........ ......
... .. .... ... ....................................................................
.
............... . .
................................................ .... .......... ... .....
. .... . .
.... ........
......... .............. .... ...
................. ...
................. ......................................... ..................................
........
.
. . ...... ..
... ......... .........................
.
.......
.... ..... ............ ....................................................................... .....................
..............
.
.
.
............................................... ...................
.
.
.
.
.
.
.
.
.
...........
..... ....
.. ..
.. . ..... ..........
....
.. ..
.. ... ........
....
..... ..... ..... ... .......
....
...
.. .. ... ....
..
..
.
..

line. The numbers of verti es of ea h part and of the whole graph obey
Eq. (3.2.24).

while the number of all the graphs grows with n0 fa torially. There is no
dependen e in Eq. (3.2.23) on the number of external lines of a planar graph
whi h is assumed to be mu h less than n0 .
It is instru tive to see the di eren e between the planar diagrams and,
for instan e, the ladder diagrams whi h des ribe e+ e elasti s attering in
QED. Let the ladder has n rungs. Then there are n! ladder diagrams, but
only one of them is planar. This simple example shows why the number of
planar graphs is mu h smaller than the number of all graphs, most of whi h
are non-planar.
In the rest of these le ture notes, we shall dis uss what is known about
solving the problem of summing the planar graphs.
Problem 3.11 Show that Eq. (3.2.23) for the number of planar graphs is onsistent
with its independen e of the number of external lines.
Solution Let us split a planar graph in two parts by utting along some line as is
depi ted in Fig. 3.9. The numbers of verti es of ea h part, n00 and n000 , are obviously
related to that of the original graph, n0 , by

(3.2.24)

We assume that both n00 and n000 are large.


p
The number of ut lines is  n0 for a planar graph in ontrast to that for a

223

MULTICOLOR QCD

. . . .... . .. . . . . . .
.. .
..
... .....................
.
......... ...
.. ..........
.
.......
..
.
.
.
.
... ..
.. ....
... ....... . . . .
... ..
. . . . .. ...
.. . . ......... . . .. ..
..
.
...
....
..
...
..
....
.
..
.
.
.. . . ..... . . ..
.
.
.
.
...
.
.
.
.
. . . . ..... ..
... ...... . .... .
......
......
....
.
.
.
.
.
.
.. ......
.. .....
.... .......
.. .
.. .
.
.
.
.
...
..
..

Fig. 3.9: Cutting a planar graph into two graphs. The utting is along the dotted

n00 + n000 = n0 :

3.2.

.......
........
.........
.........
.... ...
.... ...
... ...
... ...
............................................................
.... ..... .... ..... .... ..... .... ....
.... ..... ....
..... ....
.......... ...
.... .....
.
..
.
.... ....
.... .......
.
.
.... ....
.
.
.
.
.
.
.... ....
.... ...
.... ....
.... ....
.... ....
.... .... ....... .......
.... .... .... ...
.... .... ....
....
.... ......
.... ....
....

b)

a)

Fig. 3.10: Cutting a planar graph into trees and ar hes. The line of utting is depi ted

in a) by the dotted line. The ombination of tree and ar hes in b) is


obtained from a) by a ontinuous distortion.

generi non-planar one, when it would be  n0 . Disregarding the ut lines, we get




(3.2.25)
#p (n0 ) = #p n0  #p n00 ;
0

whi h is obviously satis ed by the formula (3.2.23) a ounting for Eq. (3.2.24).
Problem 3.12

Cutting all loops of a planar graph, obtain the upper bound


#p  (1024)n2

(3.2.26)

for the number of planar graphs with n2 loops.


Solution Sin e #p does not depend on the number of external lines (see Problem 3.11), let us onsider a one parti le irredu ible planar graph with one external
line and ut all the loops as is depi ted in Fig. 3.10a. By a ontinuous distortion,
it an be depi ted like in Fig. 3.10b, where below the dotted line we have a tree
with n0 verti es and above the dotted line we have n2 ar hes. The latter number
oin ides with the number of loops of the planar graph. The number of tips of the
tree is 2n2 .
Sin e ea h planar graph an be ut is several ways, #p is bounded from above
by
#p

 #A (n2 ) #T (n0 ; 2n2 ) ;

(3.2.27)

224

CHAPTER 3.

.. .
.. .

1=N

EXPANSION

3.2.

.. .. ..
..
..
.. .. ..
.. ..
... .. ... ....... .. .. ... ... ..... .. .. ... .... ..... ... ....... ..... ... ..
.
.
.
.
.
.
.
.
.
... .... . .. ..
.
.. .. ... .... ...... ......... .. .. ... ......
.
.
.... ..
... .... ......
.. ...
... ..
.... ... ... ...
.... .......
.... ..
.. .... ...
..... ...
.
.
.
.
.
.
.
.
. ..... .... .... ....
.. . .... .... ..... ........
..
....
...
...
...
....
....
.
...
.
....
.
.
.
...
....
....
...
.... ......
....
....
.... ...
.
.... .
.
.
.
.....
..
... ...
.... ..
. .... ...
. ..... ..... .... .... .... ....

.......
.......
........... .............
........... ...........
...... .............. ..........
...... ............. .........
....
.
...
. .
............................................................
.....
. .
. ...
. ....... ......
.... ...... ...................
... ...
........
.... ...
..
.... .... .............. .... .....
....
....... .. ..
.... ...
.
.
.
..........
.... ...
.
.
.... ....
.... ....
.... ...
... .....
.... ...
..
.... ...
.....
........
....
.......
.
.
.
....... ....
..... ...
..

. . . . ......................................
.....
......
.... . . . . . . . . . . . . .......
...
.
...
.
.
.
. .... .
...
... .... . . . .. ... .. ..... ...
..
...
.
.
....
..
...
.
.
.
... ...... . . . . . . . . . . . . . ..
.
.
.
. . . . . . ....
.. ....
....
.... . . . . . . .. ......
.. .... .
...
. . . ...... ......
.. .
.... .....
.
.
....
..

225

MULTICOLOR QCD

Fig. 3.12: A tree graph (the solid lines) and its dual (the dashed ar hes).
a)

..
..
..
..
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.............
...
..
..........
.......
.
.
.
.
.
.
.
.
.
.
.
.
.
.
..... ...
. .....................
.
......
.
.
. .
.. . ..............
.
......... ..........
... ............ .............
....... ... .. ... ....... .... ........... ...........
.. ... .. ... ... ... .. ... ... ... .. ... .. .. ... .. ... ... ... .. ... ...
..
..
..
..
.

b)

Fig. 3.11: Alternative utting of the same planar graph as in Fig. 3.10 into trees and
ar hes.

where #A (n2 ) stands for the number of ar hes and #T (n0 ; 2n2 ) stands for the
number of trees with n0 verti es and 2n2 tips. An alternative way of utting the
same planar graph, whi h leads to a di erent ombination of ar hes and trees, is
depi ted in Fig. 3.11.
The number of ar hes is well-known in mathemati s and is given by the Catalan
number of order n:
2n!
n!1 n
#A (n) =
! 4 :
(3.2.28)
n!(n + 1)!
The number of trees is not independent sin e a graph, dual to a tree graph, onsists
of ar hes as is illustrated by Fig. 3.12. The number of ar hes of this dual graph
equals the sum of the number n0 of verti es and the number 2n2 of tips, i.e. equals
n0 +2n2 . Given the number n2 of loops, the number n0 of verti es is maximal when
all verti es are trivalent, so that

n0  2n2

(3.2.29)

(n0 = 2n2 for trivalent and n0 = n2 for fourvalent verti es when n2 is large).
Therefore, the number of ar hes of the dual graph is bounded by 4n2 , so that
#T (n0 ; 2n2 )

 #A (4n2 ) :

(3.2.30)

Substituting in (3.2.27), we get [KNN77 the inequality (3.2.26). Finally, the inequality (3.2.23) an be obtained by noting that n0  n2 for large n2 .

...

....... .......
... ... ... ...

Fig. 3.13: Re urren e relation for the number of ar hes. The dotted line separates

a on guration of n ar hes in two pie es: n0 to the left and n n0 to the


right.

Derive Eq. (3.2.28) for the number of ar hes.


Let us onsider a general on guration of n ar hes like is depi ted in
Fig. 3.13. Let us pi k up the leftmost ar h, splitting the on guration in two pie es:
n0 ar hes to the left and n n0 ar hes to the right of the dotted line. The number
of ar hes obviously satis es the re urren e relation

Problem 3.13
Solution

#A (n) =

n
X
n0 =1

#A n0

1 #A n n 0 ;

(3.2.31)

where the number of ar hes to the left of the dotted line is des ribed by #A (n0 1)
be ause one ar h en ir les n0 1 others. Eq. (3.2.31) expresses #A (n) re urrently
via #A (0) = 1.
Introdu ing the generating fun tion
1
X
fA (g ) =
g n+1 #A (n) ;
(3.2.32)
n=0

226

CHAPTER 3.

1=N

EXPANSION

3.2.

227

MULTICOLOR QCD

where g n 1 orresponds to n 1 verti es of ea h tree, we rewrite Eq. (3.2.36) as the


quadrati equation

. . . . . . . . . ..
. . . . . . . . . . . . . ..
...
.. . .
.. .
...
.
..
.
....
..
.. ...
..
... ...... ..... ...... ..... .. .... ......... ..... ... ..... .... ....
.
.
.
.
.
.
.
.
.
.
. .......
....... .. .... .......
..... ....
..
......
... .. .. ...
..
. ..
..
......
... ..
..
..
...
... ......
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...... ...
..
.
.
..
.
.
.
.
.
.
.
...... ... ....
... .....
..
..
...... .. .. .
.
.
.
.
..
.
......... . ..
.
..
..
....... .. ...........
..
..
..
....... ... ......
...
.
.
.
.
.
. .... ...... .
.. . .
... ..... . . .. .... . . . ..
. .. .... . . . ..
...

fT (g ) 1 = gfT2 (g ) :
Its solution

fT (g ) = 1

p1 4g X
1
(2n 2)!
=
gn 1
2g
n
!(n 1)!
n=1

(3.2.38)
(3.2.39)

gives
#T (n) =

Fig. 3.14: Re urren e relation for the number of trees. The trees inside the left and
right dotted ir les have n0 and n n0 tips, respe tively.

fA (g ) g =
Its solution

fA (g ) = 1

fA2 (g ) :

(3.2.33)

p1 4g X
1
(2n)!
=
g n+1
2
n!(n + 1)!
n=0

(3.2.34)

gives Eq. (3.2.28) for the number of ar hes.


Improve the inequality (3.2.26), al ulating the number of trivalent
tree graphs with n tips.
Solution Let us rst note that the number of verti es of a trivalent tree graph
with n tips equals n 1. Hen e, we are interested in

Problem 3.14

#T (n)

 #T (n 1; n)

(3.2.35)

in the notation of Problem 3.12. Pi king up the rst vertex in a tree as is depi ted
in Fig. 3.14, we get the following re ursion relation
#T (n) =

n 1
X
n0 =1

#T n 0 #T n n 0 ;

whi h expresses #T (n) via #T (1) = 1.


Introdu ing the generating fun tion
1
X
fT (g ) =
g n 1 #T (n) ;
n=1

(3.2.36)

(3.2.37)

(3.2.40)

Returning to Problem (3.12), it is shown that


#T (n0 ; 2n2 )

we rewrite Eq. (3.2.36) as the quadrati equation

(2n 2)!
:
n!(n 1)!

 #T (2n2 1; 2n2 ) = #A (2n2 1) :

(3.2.41)

The inequality here is due to (3.2.29) and the equality is be ause of the expli it
formulas (3.2.28) and (3.2.40). We have improved, thus, the estimate (3.2.30) having
al ulated the number of tree graphs. The inequality (3.2.26) is now improved by
#p

 (64)n2 :

(3.2.42)

The a tual number of the trivalent planar graphs was evaluated by Tutte [Tut62
and reads

n2
#p  256
(3.2.43)
27
for asymptoti ally large n2 .

3.2.3 Planar and non-planar graphs (the boundaries)


Equation (3.2.22) holds, stri tly speaking, only for the gluon propagator while
the ontribution of all planar diagrams to a onne ted n-point Green's fun tion is  gn 2 whi h is its natural order in 1=N . Say, the three-gluon Green's
fun tion is  g, the four-gluon one is  g2 and so on. In order to make ontributions of all planar diagrams to be of the same order  1 in the large-N
limit, independently of the number of external lines, it is onvenient to ontra t the Krone ker deltas asso iated with external lines.
Let us do this in a y li order as is depi ted in Fig. 3.15 for a generi
onne ted diagram with three external gluon lines. The extra deltas whi h
are added to ontra t the olor indi es are depi ted by the single lines. They
an be viewed as a boundary of the given diagram. The a tual size of the

228

CHAPTER 3.

1=N

EXPANSION

...................................
...............
........
........
.......
.
.
.
.
.
......
......
.
.....
.
.
.
.
.
....
.
.
.
.
....
.
.....
...
.
...
....
.....
...
..........................................................
.
.
.
.
...
....
.
. ......
...........
.
.
.
...
.
.
.
.
.
.......................
....
.
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
. . ............. ........... ... ...
...
......................
...
.
... ....
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.....
.
...
... ...
.
.
.
.
.
.
.
... .
..
.. ..
...
... ...
...... ..........
...
... .... ................ ...............................
.
...
.
.
.... ...........
...
..
...... ............ ..........................
..
...
............ ..... ............
...
...
... ...
.
.
....
...
....
...
.
....
.
.
.....
....
......
.....
.......
.....
.
.
.
.
.........
.
.
......
................
.... ........................

n2 = 4 losed index lines, and B = 1 boundary. The olor indi es of


the external lines are ontra ted by the Krone ker deltas (represented by
the single lines) in a y li order. The extra fa tor of 1=N is due to
the normalization (3.2.44). Its order in 1=N is  1=N 2 in a ord with
Eq. (3.2.22).

boundary is not essential | it an be shrunk to a point. Then a bounded


pie e of a plane will be topologi ally equivalent to a sphere with a pun ture.
I shall prefer to draw planar diagrams in a plane with an extended boundary
(boundaries) rather than in a sphere with a pun ture (pun tures).
It is lear from the graphi representation that the diagram in Fig. 3.15
is asso iated with the tra e over the olor indi es of the three-point Green's
fun tion
3
 (igN) h tr [A1 (x1 ) A2 (x2 ) A3 (x3 )i :

(3.2.44)

We have introdu ed here the fa tor g3 =N to make G3 of O(1) in the large-N


limit and inserted i3 for later onvenien e. Therefore, the ontribution of the
diagram in Fig. 3.15 having one boundary should be divided by N , while the
fa tor g3 is naturally asso iated with three extra verti es whi h appear after
the ontra tion of olor indi es.
The extension of Eq. (3.2.44) to multi-point Green's fun tions is obvious:

G(n1)n (x1 ; : : : ; xn )

(ig)n
h tr [A1 (x1 )    An (xn )i :
N

(3.2.45)

229

MULTICOLOR QCD

....... .................
.........
......
.
......
....
..... ......
............
.....
.
..
........
..
...
....
..
.
.
..
.
.
.
.
.
.
....... .
.
.
.
.
...
.
.
.
..
.
... ................. . .................. .................. ....
............ ........
....... ..............
........
....
....
............
.....
.......
.............. ...........................
........

Fig. 3.15: Generi index diagram with n0 = 10 verti es, n1 = 10 gluon propagators,

G(3)
1 2 3 (x1 ; x2 ; x3 )

3.2.

ppp pppp p ppp p p pppp p ppppppppp p ppppp pp pp p


pp p pp pppppppp pppp
ppppppppp
ppppppp
pp ppppppp
p
p
p
p
p
.. ............ ..
p ppppp . .... .... .... .... ... ..... .. .... .... .... .... p.p.pppppppp
p
p
p
p
......ppppp
ppp ppp ..... ...
...p.p.pp
ppp ...
...pppp
.
.
.
.p..pp.p..
.
.
..... ................
.. ppp
.
.
.
.
.
.
.
ppp ..
.
.
.
.
. p
.
.
.
.
.
.
.....
.
.
.. pppp
.
.
.
pppp ....
ppp .. ppppp ppp pppp p pppppp pp ppppp ... pp
... ppp
pppppp ..ppp
p.p.p. ppppppp
.
ppp ....
ppp pp pp.
pp ..
. pppppp p
... ppp
p
pp ...
.. ppppppppppppppp.p ppppppppp ppppp p ...
... pppp
ppp
.........
.
.
p
p
p
p
p
.
.
.
p
p
.
p
p
p
p
p
...
ppp
pp
.. .......
..
......
....
..... ......
ppp
........... .. .......... . .........
.......... pp ppp
.
.
.
.
.
.
.
.
.
.
.
.
.
.
...... ..
.
pppp
....
pp
........ ......... .......... ...........
pppp
ppp
....................................... . .............
ppppp
pppp
.... ...................................
p
pppppp
p
p
.
p
..... .
. .
ppp
ppppppp
... .
pppp pp
pppppppp .......
.
pppppppppppp..p..pp
ppppppppp
ppppppppppppp p p p p p p pppp p pppp.p..p.p p ppp pppppppppppp

a)

b)

Fig. 3.16: Planar a) and non-planar b) ontributions of the two olor stru tures in
Eq. (3.2.8) for three-gluon vertex to G(3) in the lowest order of perturbation theory.

The fa tor 1=N , whi h normalizes the tra e, provides the natural normalization

G(0) = 1

(3.2.46)

of the averages.
Though the two terms in the index-spa e representation (3.2.8) of the
three-gluon vertex look very similar, their fate in the topologi al expansion
is quite di erent. When the olor indi es are ontra ted anti- lo kwise, the
rst term leads to the planar ontributions to G(3) , the simplest of whi h is
depi ted in Fig. 3.16a. The anti- lo kwise ontra tion of the olor indi es in
the se ond term leads to a non-planar graph in Fig. 3.16b whi h an be drawn
without rossing of lines only on a torus. Therefore, the two olor stru tures
of the three-gluon vertex ontribute to di erent orders of the topologi al
expansion. The same is true for the four-gluon vertex.

Remark on oriented Riemann surfa es


Ea h line of an index graph of the type depi ted in Fig. 3.15 is oriented.
This orientation ontinues along a losed index line while the pairs of index
lines of ea h double line have opposite orientations. The overall orientation

230

CHAPTER 3.

...
............. ...................
...........
..
.
.
.. ........
.............
.............. .....
... ..................
.
.
.
.
....
.......... ........
..
................
..
....
.
.......................
.
...
.
.
.
...
......
.
.
.
.
.
.
.
.
.
.
.
...
.
.
.
......
........... ..
.
.
.
.
......... ..
.... .........
............
...........
......
..
...................................

...
....
...
.

1=N

..........................
............
......
.....
......
.
.
.
.
...
...

...
...
....
....
.......
............ ..................
..............

a)

EXPANSION

..
..
.
...
...

b)

Fig. 3.17: Example of a) onne ted and b) dis onne ted planar graphs.
of the lines is pres ribed by the orientation of the external boundary whi h
we hoose to be, say, anti- lo kwise like in Fig. 3.16a. Sin e the lines are
oriented, the fa es of the Riemann surfa e asso iated with a given graph are
oriented too | all in the same way | anti- lo kwise. Vi e versa, su h an
orientation of the Riemann surfa es unambiguously xes the orientation of
all the index lines. This is the reason why we omit the arrows asso iated with
the orientation of the index lines: their dire tions are obvious.

Remark on y li -ordered Green's fun tions


The y li -ordered Green's fun tions (3.2.45) naturally arise in the expansion
of the tra e of the non-Abelian phase fa tor for a losed ontour, whi h was
onsidered in Problem 2.2. One gets


H

1
tr P e ig dx A (x)
N
xZn 1
Zx1
1 I
X


2
1
=
dx1 dx2 : : : dxnn G(n1)n (x1 ; : : : ; xn ) :

n=0

x1

3.2.

planar while the graph in Fig. 3.17b is dis onne ted planar.
The usual relation (1.2.52) between the generating fun tionals W [J and
Z [J for onne ted graphs and all graphs, whi h is dis ussed in the Remark
on p. 45, does not hold for the planar graphs. The reason is that an exponentiation of su h a onne ted planar diagram for the y li -ordered Green's
fun tions (3.2.45) an give dis onne ted non-planar ones.
The generating fun tionals for all and onne ted planar graphs an be
onstru ted [Cvi81 by means of introdu ing non- ommutative sour es j  (x).
\Non- ommutative" means that there is no way to transform j 1 (x1 )j 2 (x2 )
into j 2 (x2 )j 1 (x1 ). This non- ommutativity of the sour es re e ts the
y li -ordered stru ture of the Green's fun tions (3.2.45) whi h possess only
y li symmetry.
Using the short-hand notations (3.1.64) where the symbol in ludes the
sum over the d-ve tor (or whatever available) indi es ex ept for the olor
ones:

j A 

(3.2.47)

The reason is be ause the ordering along a losed path implies the y li ordering in the index spa e.

Remark on generating fun tionals for planar graphs


By onne ted or dis onne ted planar graphs we mean, respe tively, the graphs
whi h were onne ted or dis onne ted before the ontra tion of the olor
indi es as is illustrated by Fig. 3.17. The graph in Fig. 3.17a is onne ted

XZ

dd x j  (x)A (x) ;

(3.2.48)

we write down the de nitions of the generating fun tionals for all planar and
onne ted planar graphs, respe tively, as

1 1
X
n
Z [j 
tr (j A)
(3.2.49)
n=0 N
and

W [j

x1

231

MULTICOLOR QCD

1
X
n=0

1
tr (j A)n
N

onn

(3.2.50)

The planar ontribution to the Green's fun tions (3.2.45) and their onne ted ounterparts an be obtained, respe tively, from the generating fun tionals Z [j and W [j applying the non- ommutative derivative whi h is
de ned by

j (y) f (j ) =  (d) (x y) f (j ) ;
j  (x) 

(3.2.51)

where f is an arbitrary fun tion of j 's. In other words the derivative pi ks


up only the leftmost variable.

232

CHAPTER 3.

1=N

EXPANSION

... ...

...

Fig. 3.18: Graphi derivation of Eq. (3.2.52): Z [j is denoted by an empty box, W [j


is denoted by a shadow box, j is denoted by a lled ir le. By pi king
a leftmost external line of a planar graph, we end up with a onne ted
planar graph, whose remaining external lines are somewhere to the right
interspersed by dis onne ted planar graphs. It is evident that j Z [j plays
the role of a new sour e for the onne ted planar graph. If we instead
pi k up the rightmost external line, we get the inverse order Z [j j , whi h
results in Eq. (3.2.53).

The relation whi h repla es Eq. (1.2.52) for planar graphs is

Z [j = W [j Z [j ;

(3.2.52)

while the y li symmetry says

W [j Z [j = W [Z [j j :

(3.2.53)

A graphi derivation of Eqs. (3.2.52) and (3.2.53) is given in Fig. 3.18. In


other words, given W [j , one should onstru t an inverse fun tion as the
solution to the equation

j  (x) = J  (x)W [j ;

(3.2.55)

More about this approa h to the generating fun tionals for planar graphs an
be found in Ref. [CLS82.
Solve iteratively Eq. (3.2.54) for the Gaussian ase.
In the Gaussian ase, only G(2) is nonvanishing whi h yields

Problem 3.15
Solution

(3.2.56)
W [j = 1
where the propagator D is given by Eq. (3.2.4). Using Eq. (3.2.52), we get expli itly
Z

dd x dd y D (x y ) j  (x) Z [j j  (y ) Z [j :

a)

b)

Fig. 3.19: Diagrams for the gluon propagator with a quark loop whi h is represented

by the single lines. The diagram a) involves one quark loop and has no
losed index lines so that its order is  g2  1=N . The diagram b) involves
three loops one of whi h is a quark loop. Its order is  g6 N 2  1=N .

While this equation for Z [j is quadrati , its solution an be written only as a


ontinued fra tion due to the non- ommutative nature of the variables. In order to
nd it, we rewrite Eq. (3.2.57) as

Z [j =

1
R
;
1 + g 2 dd x dd y D (x y ) j  (x) Z [j j  (y )

(3.2.57)

(3.2.58)

whose iterative solution reads [Cvi81

Z [j =

1 + g2 j

:
D
j
D j
1 + g2 j
D j
1 + g2 j
.

(3.2.59)

..

3.2.4 Topologi al expansion and quark loops


It is easy to in orporate quarks in the topologi al expansion. A quark eld
belongs to the fundamental representation of the gauge group SU(N ) and its
propagator is represented by a single line

g2 j D j ;

Z [j = 1 g 2

............................
....................... ...........................
.
.
.
.... ...
... ...
... ...
.. ..
... ... ........ ........
... ..
..
.. ....
..
.
.
...
... ...
.
.
.. ....
..... .....
.
... ...
.
.
.
.
.
.
.
.
.
... ....
. ..
.... ......
... ...
...... ......... .....................
............. .............
............

(3.2.54)

after whi h Eq. (3.2.52) says

Z [j = W [J :

233

MULTICOLOR QCD

.......................
......
.........
.....
....
.
.
.
...
..
...
.....
...
..
...
...
...
...
...
.
.
....
....
......
.....
..........
......................

....

3.2.

i j

/ ij

(3.2.60)

The arrow indi ates, as usual, the dire tion of propagation of a ( omplex)
eld . We shall omit these arrows for simpli ity.
The diagram for the gluon propagator whi h involves one quark loop is
depi ted in Fig. 3.19a. It has two three gluon verti es and no losed index

234

CHAPTER 3.

1=N

EXPANSION

lines so that its order in 1=N is


Fig. 3.19a

 g2  N1 :

(3.2.61)

Analogously, the order of a more ompli ated tree-loop diagram in Fig. 3.19b,
whi h involves one quark loop and two losed index lines, is
Fig. 3.19b

 g6N 2  N1 :

(3.2.62)

It is evident from this onsideration that quark loops are not a ompanied
by losed index lines. One should add a losed index line for ea h quark loop
in order for a given diagram with L quark loops to have the same doubleline representation as for pure gluon diagrams. Therefore, given Eq. (3.2.22),
diagrams with L quark loops are suppressed at large N by


1 L+2genus
L quark loops 
:
(3.2.63)
N
The single-line representation of the quark loops is similar to the one of
the external boundary in Fig. 3.15. Moreover, su h a diagram emerges when
one al ulates perturbative gluon orre tions to the va uum expe tation value
of the quark operator
1 
;
O =
N

(3.2.64)

where the fa tor of 1=N is introdu ed to make it O(1) in the large-N limit.
Therefore, the external boundary an be viewed as a single line asso iated
with valen e quarks. The di eren e between virtual quark loops and external
boundaries is that ea h of the latter omes along with the fa tor of 1=N due
to the de nitions (3.2.45) and (3.2.64).
In order to prove Eqs. (3.2.22) and its quark ounterpart (3.2.63), let
us onsider a generi diagram in the index spa e whi h has n(3)
0 three-point
(4)
verti es (either three-gluon or quark-gluon ones), n0 four-gluon verti es, n1
propagators (either gluon or quark ones), n2 losed index lines, L virtual
quark loops and B external boundaries. A typi al su h diagram is depi ted
in Fig. 3.20. Its order in 1=N is
(4)
1 n(3)
g 0 +2n0 N n2
B
N

 N n2

(4)
n(3)
0 =2 n0 B

(3.2.65)

3.2.

235

MULTICOLOR QCD

..................................
...............
.........
........
.......
.
.
.
.
.
......
......
.
.....
.
.
.
.
.
....
.
.
.
....
.
.
.....
...
.
...
....
..................
.
.
.
.
...
.
.
.
.
.
.
.
. .................................
.
.
.
...
.
.
....
...........
... .....
..
.
.
.
.
.
.........................
....
.
.
.
.
..
.
.
.
.
.
.
.
.
.
.
.
.
.
..
.
. . ............. ............ ... ...
......................
.
...
..
... ...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
............
.
...
... ...
.
.
.
.
.
.
.
.
.
.
.
..
.
.
.
.
.
.
.
.
...
... ...
.
. . .. ..
..
... .... ................. ..............................
.
...
.
.
.
.... ..........
...
..
...... ........... .........................
...
..
..............................
...
... ..
....
...
...
....
...
.... .......
.....
.
.
.
.
.
.....
.
...
...
......
.....
..... .......
.......
.....
..
.
.
.
.
........
.
.
.....
...............
...... ..........................

Fig. 3.20: Generi diagram in the index spa e whi h has L = 1 quark loop and B = 1

loop asso iated with the external boundary. Its order in 1=N is des ribed
by Eq. (3.2.71).

as is already explained. The extra fa tor of 1=N B is due to the extra normalization fa tor of 1=N in operators asso iated with external boundaries.
The number of propagators and verti es are related by
(4)
2n1 = 3n(3)
0 + 4n0 ;

(3.2.66)

sin e three- and four-point verti es emit three or four propagators, respe tively, and ea h propagator onne ts two verti es. Using the relation (3.2.66),
we rewrite the RHS of (3.2.65) as
(3)
(4)
N n2 n0 =2 n0 B = N n2 n1 +n0 B ;

(3.2.67)

where the total number of verti es


(4)
n0 = n(3)
0 + n0

(3.2.68)

is introdu ed.
The exponent on the RHS of Eq. (3.2.67) an be expressed via the Euler
hara teristi s  of a given graph of genus h. Let us rst mention that a
proper Riemann surfa e, whi h is asso iated with a given graph, is open and
has B + L boundaries (represented by single lines). This surfa e an be losed
by atta hing a ap to ea h boundary. The single lines then be ome double

236

CHAPTER 3.

1=N

EXPANSION

lines together with the lines of the boundary of ea h ap. We have already
onsidered this pro edure when dedu ing Eq. (3.2.63) from Eq. (3.2.22).
The number of fa es for a losed Riemann surfa e onstru ted in su h a
manner is n2 + L + B , while the number of edges and verti es are n1 and n0 ,
respe tively. Euler's theorem says that

2 2h = n2 + L + B

n 1 + n0 :

(3.2.69)

Therefore the RHS of Eq. (3.2.67) an be rewritten as

N n2 n1 +n0 B

N 2 2h L 2B :

(3.2.70)

We have thus proven that the order in 1=N of a generi graph does not
depend on its order in the oupling onstant and is ompletely expressed via
the genus h and the number of virtual quark loops L and external boundaries
B by
generi graph

1
N

2h+L+2(B 1)

(3.2.71)

For B = 1, we re over Eqs. (3.2.22) and (3.2.63).

Remark on the order of gauge a tion


We see from Eq. (3.2.45) that the natural variables for the large-N limit are
the alligraphi matri es A whi h in lude the extra fa tor of ig with respe t
to A (see Eq. (3.2.2)). For these matri es
1
h tr [A1 (x1 )   An (xn )i = G(n1)n (x1 ; : : : ; xn ) (3.2.72)
N
so that they are O(1) in the large-N limit sin e the tra e is O(N ).
In these variables, the gluon part of the QCD a tion (2.1.14) takes on the
simple form
Z
1
2 (x) :
S =
dd x tr F
(3.2.73)
2g2
Sin e g2 in this formula is  1=N and the tra e is  N , the a tion is O(N 2 )
at large N .
This result an be anti ipated from the free theory be ause the kineti
part of the a tion involves the sum over N 2 1 free gluons. Therefore, the
non-Abelian eld strength (1.3.63) is  1 for g2  1=N .

3.2.

237

MULTICOLOR QCD

The fa t that the a tion is O(N 2 ) in the large-N limit is a generi property
of the models des ribing matrix elds. It will be ru ial for developing saddlepoint approa hes at large N whi h are onsidered below.
Rederive the formula (3.2.71) using the alligraphi notations.
The propagator of the A- eld is  g 2 while both three- and four-gluon
verti es are now  g 2 as a onsequen e of Eq. (3.2.73). The ontribution of a
generi graph is now of the order

Problem 3.16
Solution


g 2 n1 n0 N n2 B

 N n2

n1 +n0 B

(3.2.74)

for g 2  1=N . This oin ides with the RHS of Eq. (3.2.67) whi h results in
Eq. (3.2.71).

3.2.5 t' Hooft versus Veneziano limits


In QCD there are several spe ies or avors of quarks (u-, d-, s- and so on).
We denote the number of avors by Nf and asso iate a Greek letter or
with a avor index of the quark eld.
The quark propagator then has the Krone ker delta with respe t to the
avor indi es in addition to Eq. (3.2.60):
D


i j

/ ij :

(3.2.75)

Their ontra tion results in


Nf
X
=1

= Nf :

(3.2.76)

Therefore, an extra fa tor of Nf orresponds to ea h losed quark loop for


the Nf avors.
The limit when Nf is xed as N ! 1, as was onsidered in the original
paper by 't Hooft [Hoo74a, is alled the 't Hooft limit. Only valen e quarks
are left in the 't Hooft limit. Hen e, it is asso iated with the quen hed
approximation whi h was dis ussed in the remark on p. 164. In order for a
meson to de ay into other mesons built out of quarks, say for a -meson to
de ay into a pair of -mesons, a quark-antiquark pair must be produ ed out
of the va uum. Consequently, the ratios of meson widths to their masses are
total

 NNf

(3.2.77)

238

CHAPTER 3.

1=N

EXPANSION

a)

Nf

b)

Fig. 3.21: Diagrams with quark loops in the Veneziano limit. The olor and avor
indi es of a quark loop whi h are represented by the solid and dashed single
lines, respe tively. The diagram a) is  g2 Nf  Nf =N . The diagram b)
is  g6 N 2 Nf  Nf =N .

in the 't Hooft limit. The ratio on the LHS of Eq. (3.2.77) is experimentally
10{15% for the -meson. A hope to solve QCD in the 't Hooft limit is a hope
to des ribe QCD with this a ura y.
An alternative large-N limit of QCD when Nf  N as N ! 1 was
proposed by Veneziano [Ven76. Some diagrams for the gluon propagator,
whi h involve one quark loop, are depi ted in Fig. 3.21. The dashed single line
represents propagation of the avor index. Ea h losed loop of the dashed line
is asso iated with the fa tor Nf a ording to Eq. (3.2.76). This is analogous
to the ve tor models whi h exa tly des ribe the O(Nf ) avor symmetry in
these notations.
The diagrams in Fig. 3.21 ontribute, respe tively,
N
(3.2.78)
Fig. 3.21a  g2 Nf  f
N
and
N
Fig. 3.21b  g6N 2 Nf  f
(3.2.79)
N
in the limit (3.2.13). Likewise, a more general diagram with L quark loops
will ontribute

L quark loops


 
Nf L 1 genus
:
N
N 2

239

MULTICOLOR QCD

We see from Eq. (3.2.80) that quark loops are not suppressed at large N
in the Veneziano limit

...................
............... ...........................
................
..... ....
.
.
..
.... ..
... ...
... ...
..... ............
.
.
.
.... ...
.
.
... ...
.
.
.
.
.
.
.
.
.
.
.
.
.
.
... ...
..
.
... ..
... ... ............. ...........
.... ....
... ....
... ..
.... .....
..... ....
...............
.............. ...........................
...............

..................
......... .... .... .................
.... ....
......... ...
........ .
... ....
.. ...
........
.. ..
..
... ..
.. ..
... ..
.. ....
... ...
.
.
. ...
.... ...
...... . .... . . .... ............
.......... .... ..............
............

3.2.

(3.2.80)

This formula obviously follows from Eq. (3.2.63) sin e ea h quark loop results
in Nf .

 N ! 1

(3.2.81)

if the diagram is planar. Furthermore, the representation of a avored quark


by one solid and one dashed line is obviously similar to the double-line representation of a gluon. All what is said above about the topologi al expansion
of pure gluodynami s holds for QCD with quarks in the Veneziano limit.
It is the Veneziano limit (3.2.81) that is related to the hadroni topologi al
expansion in the dual-resonan e models. Hadrons an have, in the Veneziano
limit, nite widths a ording to Eq. (3.2.77). I refer the reader to the original
paper by Veneziano [Ven76 for more detail.
There is an alternative way to show why virtual quarks are suppressed in
the 't Hooft limit and survive in the Veneziano limit. Let us integrate over
the quark elds whi h yields
Z

D D e

R 4

b +m 
d x r

= e Sp ln rb +m

(3.2.82)

as is dis ussed in Subse t. 1.2.2. The tra e in the exponent involves summation both over olor and avor indi es, so that
Sp ln

rb + m  N Nf :

(3.2.83)


The order in N of the pure gluon a tion is O N 2 as is dis ussed in the


Remark on p. 236. Hen e, the quark ontribution to the a tion is  Nf =N
in omparison to the gluon one. The quark determinant an be disregarded
in the 't Hooft limit, but is essential in the Veneziano limit.
The onsideration of the previous paragraph also explains why ea h quark
loop ontributes a fa tor  Nf =N . The exponent on the RHS of Eq. (3.2.82)
is asso iated with one-loop diagrams. A diagram with L quark loops orresponds to the L-th term of the expansion of the exponential. This explains
the fa tor (Nf =N )L in Eq. (3.2.80). A diagram with two quark loops, whi h
appears to the se ond order of this expansion, is depi ted in Fig. 3.22.

Remark on asymptoti freedom in the Veneziano limit


Though the number of avors be omes large in the Veneziano limit, this does
not mean that asymptoti freedom is lost. The leading-order oe ient of

240

CHAPTER 3.

...............
...... .... .....
.. ...... .... ...
..
.
.. ..
..
.. ..
..
... ..
... ..
..
... ...
.... . .. ..
...... .. .... ...
..............

1=N

EXPANSION

the B-fun tion of QCD with N olors and Nf avors reads


2
11
N + N :
3 3 f
It is still negative if Nf =N < 11=2 in the Veneziano limit.

b =

241

MULTICOLOR QCD

3) The oupling onstant of meson-meson intera tion is small at large N .


4) The widths of glueballs are  1=N 2 , i.e. they should be even narrower
than mesons built out of quarks. The glueballs do not intera t or mix
with mesons at N = 1.

.........
.. ......................
.... ....
..
..
.. ...
.. ....
..
..
.. ..
.. ....
..
.
.. .... .........
..... .... .....
...............

Fig. 3.22: Diagram with two quark


 loops in the Veneziano limit. The diagram is

 g6 N Nf2  Nf =N

3.2.

All these hadron properties (ex ept the last one) approximately agree with
experiment, and were well-known even before 1974 when multi olor QCD was
introdu ed. Glueballs are not yet dete ted experimentally (maybe be ause of
their property listed in the item 4).

3.2.6 Large-N fa torization


(3.2.84)

Remark on phenomenology of multi olor QCD


While N = 3 in the real world, there are phenomenologi al indi ations that
1=N may be onsidered as a small parameter. We have already mentioned
some of them in the text | the simplest one is that the ratio of the -meson
width to its mass, whi h is  1=N , is small. Considering 1=N as a small
parameter immediately leads to qualitative phenomenologi al onsequen es
whi h are preserved by the planar diagrams asso iated with multi olor QCD,
but are violated by the non-planar diagrams.
The most important onsequen e is the relation of the 1=N -expansion
to the topologi al expansion in the dual-resonan e model of hadrons. Vast
properties of hadrons are explained by the dual-resonan e model. A very
lear physi al pi ture behind this model is that hadrons are ex itations of a
string with quarks at the ends.
I shall brie y list some onsequen es of multi olor QCD:
1) The \naive" quark model of hadrons emerges at N = 1. Hadrons
are built out of (valen e or onstituent) quark and antiquark qq, while
exoti states like qqqq do not appear.
2) The partial width of de ay of the -meson, whi h is built out of ss (the
strange quark and antiquark), into K +K is  1=N , while that into
+  0 is  1=N 2. This explains Zweig's rule. The masses of the and !-mesons are degenerate at N = 1.

The va uum expe tation values of several olorless or white operators, whi h
are singlets with respe t to the gauge group, fa torize in the large-N limit
of QCD (or other matrix models). This property is similar to that already
dis ussed in Subse t. 3.1.5 for the ve tor models.
The simplest gauge-invariant operator in a pure SU(N ) gauge theory is
the square of the non-Abelian eld strength:

O (x) =

1
2 (x) :
tr F
N 2

(3.2.85)

The normalizing fa tor provides the natural normalization




1
tr F 2 (x)
N 2 

1 a
a (x)
F (x) F
2N 2 

 1:

(3.2.86)

In order to verify the fa torization in the large-N limit, let us onsider


the index spa e diagrams for the average of two olorless operators O (x1 )
and O (x2 ), whi h are depi ted in Fig. 3.23.
The graph in Fig. 3.23a represents the zeroth order of perturbation theory.
It involves four losed index lines (the fa tor N 4 ) and the normalization fa tor
1=N 4 a ording to the de nition (3.2.85). Its ontribution is
Fig. 3.23a

 N12 N 2  N12 N 2  1 ;

(3.2.87)

i.e. O (1) in a ord with the general estimate (3.2.86).


The graph in Fig. 3.23b involves a gluon line whi h is emitted and absorbed
by the same operator O (x1 ). It has ve losed index lines (the fa tor N 5 ),

242

CHAPTER 3.

......................
......................................
.........
... ..
.
... ..
.. .
... ...
.. ....
.
... ......
.
.
.
.
.
............................
....................

......................
......................................
.........
... ..
.
... ..
.. .
... ...
.. ....
.
... ......
.
.
.
.
.
............................
....................

......................
..... ................... .....
........................ ...........
... ... ................ .... ..
... .... . ..
... ...
.......... .....................
............................
..............

a)
........
....................... ......................
..............................................................................................
.
.
.
.
.
.
.............
.... .....
..... ....
.........
... ...
.. ..
.. ..
.... ....
.. ..
... ...
... ...
........
.
.
.........
..... ...
............
...........
................
................................ ............................................
.................... ......... ...................
.................

1=N

EXPANSION

......................
.....................................
.... .....
... ...
.
... ..
.
... .
... ..
.......
...........
.
.
.
................................
..............

b)

3.2.

Therefore, it is suppressed by 1=N 2 in the large-N limit. For this graph, the
gluon line is emitted and absorbed by di erent operators O (x1 ) and O (x2 ).
This lowest-order example illustrates the general property that only (planar) diagrams with gluon lines emitted and absorbed by the same operators
survive as N ! 1. Hen e, orrelations between the olorless operators
O (x1 ) and O (x2 ) are of order 1=N 2, so that the fa torization property holds
as N ! 1:


........................
...................................
...........
... ...
... ...
..
... ...
... ..
.......
...........
.
.
.
.
.............................
..............

........................
....................................
...........
... ...
.
... .....
... ..
.. ..
... ....
....................................
.......................

d)

Fig. 3.23: Demonstration of the large-N fa torization to the lowest orders of perturbation theory. The losed double line represents the average of the operator (3.2.85) to zeroth order in g. The diagrams a) and b), whi h are asso iated with the fa torized part of the average on the LHS of Eq. (3.2.91),
are O(1). The diagrams ) and d), whi h would violate the fa torization,
are suppressed by 1=N 2 .

the normalization fa tor 1=N 4 , and g2 due to two three-gluon verti es. Its
ontribution is
Fig. 3.23b  g2 N  1 ;
(3.2.88)
i.e. also O (1) in the limit (3.2.13).
The graph in Fig. 3.23 is of the same type as the graph in Fig. 3.23a,
but the double lines now onne t two di erent operators. It has two losed
index lines (the fa tor N 2 ) and the normalization fa tor 1=N 4 , so that its
ontribution
1
(3.2.89)
Fig. 3.23  2
N
is suppressed by 1=N 2.
The graph in Fig. 3.23d, whi h is of the same order in the oupling onstant as the graph in Fig. 3.23b, involves only three losed index lines (the
fa tor N 3 ) and is of order 1=N 2:
1
 N12 :
(3.2.90)
Fig. 3.23d  g2
N

243

MULTICOLOR QCD

1
1
tr F 2 (x1 ) 2 tr F 2 (x2 )
N 2
N





1
1
1
2
2
:
tr F (x1 )
tr F (x2 ) + O
=
N 2
N 2
N 2

(3.2.91)

For a general set of gauge-invariant operators O1 , . . . , On , the fa torization property an be represented by

h O1    O n i

h O1 i    h O n i + O

1
:
N 2

(3.2.92)

This is analogous to Eq. (3.1.132) for the ve tor models.


The fa torization in large-N QCD was rst dis overed by A.A. Migdal in
the late seventies. An important observation that the fa torization implies a
semi lassi al nature of the large-N limit of QCD was done by Witten [Wit79.
We shall dis uss this in the next two Subse tions.
The fa torization property also holds for gauge-invariant operators onstru ted from quarks like in Eq. (3.2.64). For the ase of several avors Nf ,
we normalize these quark operators by

O =
Here

1 
Nf N

(3.2.93)

stands for one of the ombination of the gamma-matri es:

1
[ ; ; : : : :
(3.2.94)
2i  
The lowest-order diagrams of perturbation theory for the average of two
quark operators (3.2.93) are depi ted in Fig. 3.24. Estimation of their order
in 1=N is analogous to that for the pure gluon graphs in Fig. 3.23.
The graph in Fig. 3.24a represents the zeroth order of perturbation theory
for the average of two quark operators. It involves two losed olor and
= I; 5 ;  ; i  5 ;  =

244

CHAPTER 3.

... .... .
... ..........................
.... ..
..........
.. .
..
.. ....
... ...
.
.
.
... ....
.
..
.
.
....................... ... .
.... .... ....

... .... .
... ..........................
.... ..
..........
.. .
..
.. ....
... ...
.
.
.
... ....
.
..
.
.
....................... ... .
.... .... ....

1=N

... .... .
... ............................
... ...
.........
.. .
..
..
... ....
.. ..
.
.
.
.
... ....
..
.
.
.
....................... ... .
.... .... ....

... .... .
... ..................... ...
.................... ...........
. . ... ...
..
... .... ................... .... ...
.
.
.
.
.
.
.
.
.
.
... ....
.. ... ..
......................... ... .
.... .... ....

a)

EXPANSION

b)

........
....................... ......................
............. .... .... .... .... .... .... .... ........................
.
.
.
.
.
.
.
.... .......
.... ..
.... ...
.... ...
... ...
.. ...
.... ..
... ...
.
... ...
.. ....
... ...
. ...
.... ....
.
.
.
..... ....
..
....... .... .
.. .. .....
............ .... .... .. .... .... .... ....................
.................... .. .... ...................
.................

........................
..... .... ..... .... ........
.. ...
....... ..
... ..
..
... .
.. ..
. .....
...... ...
.
.
........ .... ...........
..............

........................
..... .... ..... .... .........
.. ..
....... ..
. ..
... ..
.. .....
... ...
..... .... .... .... .........
.......................

d)

Fig. 3.24: Same as in Fig. 3.23 but for quark operators (3.2.93). The solid and dashed

lines are asso iated with olor and avor indi es, respe tively. The diagram
a), whi h ontributes to the fa torized part of the average on the LHS of
Eq. (3.2.99), are O(1). The diagrams b) and ), whi h would violate the
fa torization, are suppressed by 1=N 2 and 1=(Nf N ), respe tively.

Nf2 N 2 )

two losed avor index lines (the fa tor


and the normalization fa tor
1=(Nf N )2 a ording to the de nition (3.2.93). Its ontribution is
Fig. 3.24a

 N 21N 2 Nf2 N 2  1 :
f

(3.2.95)

This justi es the normalization fa tor in Eq. (3.2.93).


The graph in Fig. 3.24b involves a gluon line whi h is emitted and absorbed
by the same quark operator. It has three losed olor and two losed avor
index lines (the fa tor Nf2 N 3 ), the normalization fa tor 1=(Nf N )2 , and g2
due to two quark-gluon verti es. Its ontribution is
Fig. 3.24b

 N 21N 2 Nf2 g2 N 3 
f

(3.2.96)

in full analogy with the pure gluon diagram in Fig. 3.23b.


The graph in Fig. 3.24 is similar to the graph in Fig. 3.23 | the lines
onne t two di erent quark operators. It has one losed olor and one losed
avor index lines (the fa tor Nf N ) and the normalization fa tor 1=(Nf N )2 ,

3.2.

245

MULTICOLOR QCD

so that its ontribution


Fig. 3.24

 N 1N
f

(3.2.97)

is suppressed by 1=(Nf N ).
Finally, the graph in Fig. 3.24d involves a gluon line whi h is emitted by
one quark operator and absorbed by the other. It has one losed olor and two
losed avor index lines (fa tor Nf2 N ), the normalization fa tor 1=(Nf N )2 ,
and g2 due to two quark-gluon verti es. Its ontribution
Fig. 3.24d

 N 21N 2 Nf2 g2N  Ng  N12




f

(3.2.98)

is suppressed by 1=N 2 in the limit (3.2.13).


We see that the fa torization of the gauge-invariant quark operators holds
both in the 't Hooft and Veneziano limits:


(3.2.99)
h O 1    O n i = h O 1 i   h O n i + O N 1N :
f
The non-fa torized part, whi h is asso iated with onne ted diagrams, is
 1=N in the 't Hooft limit. This leads, in parti ular, to the oupling
onstant of meson-meson intera tion of order 1=N , larifying the property of
multi olor QCD listed in item 3 on p. 241. The Veneziano limit is analogous
to pure gluodynami s as is already mentioned.
It is worth noting that the fa torization an be alternatively seen (at all
orders of perturbation theory) from Eq. (3.2.71) for the ontribution of a
generi onne ted graph of genus h with B external boundaries whi h are
pre isely asso iated with the quark operators O , as is explained in Subse t. 3.2.4. The diagrams with gluon lines emitted and absorbed by the same
operator like in Fig. 3.24b are produ ts of diagrams having only one boundary. Hen e, their ontribution is of order 1. Otherwise, the diagrams with
gluon lines emitted and absorbed by two di erent operators like in Fig. 3.24d
have two boundaries. A ording to Eq. (3.2.71), their ontribution is suppressed by 1=N 2 . Alternatively, the diagrams like in Fig. 3.24 (in luding its
planar dressing by gluons) have one boundary.7 Their ontribution is O(1)
times 1=(Nf N ) oming from the normalization of the operator (3.2.93). This
proves the fa torization property (3.2.99) at all orders of perturbation theory.
7 In the dual-resonan e model, they are asso iated with the meson-meson intera tion

due to an ex hange of onstituent quarks.

246

CHAPTER 3.

1=N

EXPANSION

3.2.

Remark on fa torization beyond perturbation theory


The large-N fa torization an be also veri ed at all orders of the strong
oupling expansion in the SU(N ) latti e gauge theory. A non-perturbative
proof of the fa torization will be given in the next Se tion by using quantum
equations of motion (the loop equations).
Problem 3.17 Prove the fa torization of the Wilson loop operators within the
strong oupling expansion of the SU(N ) latti e gauge theory as N ! 1.
Solution Let us rst estimate the order in N of the Wilson loop average (2.2.42).
The expli it result to the leading order in is given by Eqs. (2.2.73), (2.2.72), where

 N 2

(3.2.100)

in the limit (3.2.13) as is pres ribed by Eq. (2.2.32). Therefore, WC  1 in the


large-N limit.
To be pre ise, we rst perform the strong oupling expansion in and then
set N ! 1 in ea h term of the strong oupling expansion. As we shall see in a
moment, the a tual parameter is =N 2 , so that the limits ! 0 and N ! 1 are
inter hangeable.
It is easy to estimate the order in N of any graph of the strong oupling
expansion for WC , whi h looks like that in Fig. 2.10. Let the plaquettes ll an
arbitrary surfa e en losed by the loop C , with n2 , n1 , and n0 being the number of
plaquettes, links, and sites whi h belong to the surfa e. Ea h plaquette ontributes
=N sin e it omes from the expansion of the exponential of the latti e a tion,
ea h link ontributes 1=N due to Eq. (2.2.60), and ea h site ontributes N sin e it
is asso iated with summing over the olor indi es due to Eq. (2.2.70). A ounting
for the normalization fa tor 1=N B where B = 1 is the number of boundaries, the
ontribution is of order


n2 N n1 +n0 B   n2 N n2 n1 +n0 B




N
N 2

n2

n2  2h+2(B
1
 N 2 N n2 n1 +n0 B  N 2
N

C1

247

MULTICOLOR QCD

..
.......
...........
.
.
.
.
.
......... .
........ ........
........... ..............
......... ...
........ ........
....... ...............
...
....... ........
........... .................
.....
.........
.............
.

.
......
...........
.
.
.
.
.
........
......... .........
........... ...............
...... ..
......... .........
....... ...............
. ...... ....
....... ..........
...... ................
........
..........
......

C2

C1

....... .... ...... .... .... . .... ...... .... .... . .... ......
............. .......... . ......... .. . ......... . . .......... .. . ..........
......... . .................. ................. ................ . .................. ...................
.
.
.
.
.
.
.... .. .. ...... . .. ........ .. ...... . .................
......... ............. ............... .. ........... .. ........... ... ..........
........ ...... ..... ...... .. .. ...... .. .. ...... .. ... ........
..................... ................... ......................................
.... . . . . . ..
. . . . . ..
...... .... .... .... .... .... .... ....
.
............................................................................................................................... .................
...... .... .......... .... ......... ..... .......... .... .......... .... ...........
.
.
....... .... .......... .... ......... .... ......... .... .......... ................
....... . .. ...... . . ....... . ....... . . ...... . ...........
.... .... .... ....... .... ....... .... ....... .... ........ .... .........

C2

b)

a)

Fig. 3.25: Fa torization of the Wilson loop operators in the strong oupling expansion

as N ! 1. The surfa es are onstru ted from plaquettes whi h ome


from the expansion of the exponential of the latti e a tion. Ea h link in the
surfa e is passed at least twi e: otherwise the result vanishes. a) involves
two separate surfa es en loses by the Wilson loops. It ontributes to the
fa torized part of the average on the LHS of Eq. (3.2.102). The surfa e in
b) onne ts two di erent Wilson loops and would violate fa torization. It
has two boundaries and its ontribution is suppressed by 1=N 2 a ording
to the general formula (3.2.101).

spanned by ea h individual loop. Its ontribution is O(1) as N ! 1. A nonfa torized part emerges from surfa es of the type depi ted in Fig. 3.25b, whi h onne t two di erent Wilson loops. They look like a ylinder and have two boundaries.
Their ontribution is suppressed by 1=N 2 a ording to the general formula (3.2.101).
Thus, we have proven the fa torization property
D

1 tr U 1 tr U E =
N C1 N C2

1 tr U E D 1 tr U E + O  1  (3.2.102)
N C1 N C2
N 2

at all orders of the strong oupling expansion.


1)

; (3.2.101)

where we have used Euler's theorem (3.2.69). In the limit (3.2.100), the ontribution
does not depend on the order of the strong oupling expansion and is ompletely
determined by the number B of boundaries and the genus h of the surfa e. This
is analogous to the perturbation theory. For the minimal surfa e, we reprodu e
previous results.
We are now in a position to analyze the order in N of di erent terms in the
strong oupling expansion of the average of two Wilson loop operators. The fa torized part results from the surfa es of the type depi ted in Fig. 3.25a whi h are

Find the relation between the Wilson loop averages in the fundamental and adjoint representations for a SU(N ) pure gauge theory at large N .
Solution The hara ters in the fundamental and adjoint representations are related by Eq. (2.2.28). Using the fa torization formula (3.2.102) with oin iding
ontours C1 and C2 , we get

Problem 3.18

Wadj (C ) = [Wfun (C )2 + O

1 
:
N 2

(3.2.103)

As is dis ussed in Chapter 2, the Wilson loop average in the fundamental representation obeys area law (2.2.75). The same is true at N = 1 for the Wilson loop
average in the adjoint representation due to Eq. (3.2.103). In parti ular, the string

248

CHAPTER 3.

1=N

EXPANSION

tensions in the fundamental and adjoint representations at N = 1 are related by

Kadj = 2Kfun :

(3.2.104)

On the other hand, the adjoint test quark an be s reened at nite N by a gluon
produ ed out of the va uum. This is similar to the breaking of the ux tube
in the fundamental representation by a quark-antiquark pair, whi h is dis ussed
in Subse t 2.5.5. Therefore, the perimeter law (2.2.79) must dominate for large
ontours. The point is that the perimeter law appears due to onne ted diagrams
whi h are suppressed as 1=N 2 :

Wadj (C ) ! e 2KAmin (C ) + 12 e L(C ) :


(3.2.105)
N
These properties of the adjoint representation were rst pointed out in Ref. [KM81
ited in Chapter 2.
large C

3.2.

249

MULTICOLOR QCD

The idea is to rewrite the path integral over A for the Yang{Mills theory as that over a olorless omposite eld  [A, likewise it was done in
Subse t. 3.1.4 for the sigma-model. The expe ted new path-integral representation of the partition fun tion (3.2.106) would be something like

A

  [A
Aa

DAa e

R 4 1 a a
d x 4 F F

(3.2.106)

The a tion,  N 2 , is large as N ! 1, but the \entropy" is also  N 2 due


to the N 2 1 integrations over Aa :

DAa

e N :

(3.2.107)

Consequently, the saddle-point equation of the large-N Yang{Mills theory is


not the lassi al one whi h reads8
S
= (r F )a = 0 :
(3.2.108)
Aa
8 It was already dis ussed in Problem 2.1.

2
e N J [

(3.2.110)

in Eq. (3.2.109) is related to the old entropy fa tor, so that

J [

The large-N fa torization in QCD assumes that gauge-invariant obje ts behave as -numbers, rather than as operators. Likewise the ve tor models, this
suggests that the path integral is dominated by a saddle point.
We already saw in Subse t. 3.1.5 that the fa torization in the ve tor models does not mean that the fundamental eld itself, for instan e ~n in the
sigma-model, be omes \ lassi al". It is the ase, instead, for a singlet omposite eld.
We are now going to apply a similar idea to the Yang{Mills theory whose
partition fun tion reads

(3.2.109)

The Ja obian

3.2.7 The master eld

Z =

2
1
D  [A e N S[ :
a

(3.2.111)

in the large-N limit.


The original partition fun tion (3.2.106) an be then rewritten as

2
2
D e N J [ N S[ ;

(3.2.112)

where S [ represents the Yang{Mills a tion in the new variables. The new
\entropy" fa tor D is O(1) be ause the variable  [A is a olor singlet. The
large parameter N enters Eq. (3.2.112) only in the exponent. Therefore, the
saddle-point equation an be immediately written:

S
J
=
:


Remembering that  is a fun tional of A :


 [A ;

(3.2.113)

(3.2.114)

we rewrite the saddle-point equation (3.2.113) as

S
J
= (r F )a =
:
(3.2.115)
Aa
Aa
It di ers from the lassi al Yang{Mills equation (3.2.108) by the term on the
RHS oming from the Ja obian (3.2.110).

250

CHAPTER 3.

1=N

EXPANSION

Given J [ whi h depends on the pre ise from of the variable  [A,
Eq. (3.2.115) has a solution

A (x) = A l (x) :

(3.2.116)

Let us rst assume that there exists only one solution to Eq. (3.2.115). Then
the path integral is saturated by a single on guration (3.2.116), so that the
va uum expe tation values of gauge-invariant operators are given by their
values at this on guration:

hOi

= O A l (x) :

(3.2.117)

The fa torization property (3.2.92) will obviously be satis ed.


An existen e of su h a lassi al eld on guration in multi olor QCD was
onje tured by Witten [Wit79. It was dis ussed in the le tures by Coleman [Col79 who alled it the master eld. Equation (3.2.115) whi h determines the master eld is often referred to as the master- eld equation.
A subtle point with the master eld is that a solution to Eq. (3.2.115) is
determined only up to a gauge transformation. To preserve gauge invarian e,
it is more reasonable to speak about the whole gauge orbit as a solution of
Eq. (3.2.115). However, this will not hange Eq. (3.2.117) sin e the operator
O is gauge invariant.
The onje ture about an existen e of the master eld has surprisingly ri h
onsequen es. Sin e va uum expe tation values are Poin are invariant, the
RHS of Eq. (3.2.117) does. This implies that A l (x) must itself be Poin are
invariant up to a gauge transformation: a hange of A l (x) under translations
or rotations an be ompensated by a gauge transformation. Moreover, there
must exist a gauge in whi h A l (x) is spa e-time independent:

A l (x) = A l (0) :
(3.2.118)
In this gauge, rotations must be equivalent to a global gauge transformation,
so that A l (0) transforms as a Lorentz ve tor.
In fa t, the idea about su h a master eld in multi olor QCD may be
just wrong as was pointed out by Haan [Haa81. The onje ture about an
existen e of only one solution to the master- eld equation (3.2.115) seems to
be somewhat in orre t. If several solutions exists, one needs an additional
averaging over these solutions. This is a very deli ate matter, sin e this
additional averaging must still preserve the fa torization property. One might
better think about this situation as if A l (0) would be an operator in some
Hilbert spa e rather than a -valued fun tion. This is simply be ause A l (0) is,

3.2.

251

MULTICOLOR QCD

in the matrix notation (3.2.1), a N  N matrix whi h be omes, as N ! 1,


an in nite matrix, or an operator in Hilbert spa e. Su h an operator-valued
master eld is sometimes alled the master eld in the weak sense, while
the above onje ture about a single lassi al on guration of the gauge eld,
whi h saturates the path integral, is alled the master eld in the strong sense.
The on ept of the master eld is rather vague until a pre ise form of
the omposite eld  [A, and onsequently the Ja obian  [A that enters
Eq. (3.2.115), is not de ned. However, what is important is that the master
eld (in the weak sense) is spa e-time independent. This looks like a simpli ation of the problem of solving large-N QCD. A Hilbert spa e, in whi h
the operator A l (0) a ts, should be spe i ed by  [A. We shall onsider in
the next Subse tion a realization of these ideas for the ase of  [A given by
the tra e of the non-Abelian phase fa tor for losed ontours.

Remark on non- ommutative probability theory


An adequate mathemati al language for des ribing the master eld in multi olor QCD (and, generi ally, in matrix models at large N ) was found by
I. Singer in 1994. It is based on the on ept of free random variables of non ommutative probability theory, introdu ed by Voi ules u [VDN95. How to
des ribe the master eld in this language and some other appli ations of non ommutative free random variables to the problems of planar quantum eld
theory are dis ussed in Refs. [Dou95, GG95.

3.2.8

1=N

as semi lassi al expansion

A natural andidate for the omposite operator  [A from the previous Subse tion is given by the tra e of the non-Abelian phase fa tor for losed ontours | the Wilson loop. It is labeled by the loop C in the same sense as the
eld A (x) is labeled by the point x, so we shall use the notation
 (C )

 [A =

H

1
tr P e ig C dx A (x) :
N

(3.2.119)

Nobody up to now managed to reformulate QCD at nite N in terms


of  (C ) in the language of path integral. This is due to the fa t that selfinterse ting loops are not independent (they are related by the so- alled Mandelstam relations [Man79), and the Ja obian is huge. The reformulation was
done [MM79 in the language of S hwinger{Dyson or loop equations whi h
will be des ribed in the next Se tion.

252

CHAPTER 3.

1=N

EXPANSION

S hwinger{Dyson equations are a onvenient way of performing the semi lassi al expansion, whi h is an alternative to the path integral. Let us illustrate an idea how to do this by an example of the '3 theory whose S hwinger{
Dyson equations are given by Eq. (1.2.47).
The RHS of Eq. (1.2.47) is proportional to the Plan k's onstant h as is
explained in Subse t 1.2.5. In the semi lassi al limit h ! 0, we get




12 + m2 h' (x1 ) : : : ' (xn )i + '2 (x1 ) : : : ' (xn ) = 0 ; (3.2.120)
2
whose solution is of the fa torized form

h' (x1 ) : : : ' (xn )i

provided that

= h' (x1 )i : : : h' (xn )i + O (h)

h' (x)i  ' l (x)

(3.2.121)
(3.2.122)

obeys


 2 + m2 ' l (x) + '2 l (x) = 0 :
(3.2.123)
2
Equation (3.2.123) is nothing but the lassi al equation of motion for
the '3 theory, whi h spe i es extrema of the a tion (1.2.22) entering the
path integral (1.2.2). Thus, we have reprodu ed, using the S hwinger{Dyson
equations, the well-known fa t that the path integral is dominated by a lassi al traje tory as h ! 0. It is also lear how to perform the semi lassi al
expansion in h in the language of the S hwinger{Dyson equations: one should
solve Eq. (1.2.47) by iterations.
The reformulation of multi olor QCD in terms of the loop fun tionals
 (C ) is, in a sense, a realization of the idea of the master eld in the weak
sense, when the master eld a ts as an operator in the spa e of loops. The
loop equation of the next Se tion will be a sort of the master- eld equation
in the loop spa e.

Remark on the large-N limit as statisti al averaging


There is yet another, pure statisti al, explanation why the large-N limit is a
\semi lassi al" limit for the olle tive variables  (C ). The matrix U ij [Cxx ,
that des ribes the parallel transport along a losed ontour Cxx , an be redu ed by the gauge transformation to


U [Cxx =
[Cxx diag e ig 1 (C ) ; : : : ; e ig N (C )
y [Cxx : (3.2.124)

3.2.

253

MULTICOLOR QCD

Then  (C ) reads
N
1 X
e ig j (C ) :
 (C ) =
N j=1

(3.2.125)

The phases j (C ) are gauge invariant and normalized so that j (C )  1 as


N ! 1. For simpli ity we omit below all the indi es (in luding spa e ones)
ex ept olor.
The ommutator of 's an be estimated using the representation
(3.2.125). Sin e
[ i ; j

/ ij ;

(3.2.126)

one gets
[ (C ) ;  (C 0 )

 g2 N1  N12

(3.2.127)

in the limit (3.2.13), i.e. the ommutator an be negle ted as N ! 1, and


the eld  (C ) be omes lassi al.
Note that the ommutator (3.2.127) is of order 1=N 2. One fa tor 1=N is
be ause of g in the de nition (3.2.125) of  (C ), while the other has a deep
reason. Let us image the summation over j in Eq. (3.2.125) as some statisti al
averaging. It is well-known in statisti s that su h averages weakly u tuate
as N ! 1, so that the dispersion is of order 1=N . It is the fa tor whi h
emerges in the ommutator (3.2.127).
We see that the fa torization is valid only for the gauge-invariant
quantities whi h involve the averaging over the olor indi es, like that in
Eq. (3.2.125). There is no reason to expe t fa torization for gauge invariants whi h do not involve this averaging, for instan e for the phases j (C ).
Moreover, their ommutator (3.2.126) is  1, so that j (C )'s strongly u tuate even at N = 1. An expli it example of su h strongly u tuating
gauge-invariant quantities was rst onstru ted in Ref. [Haa81.
The resume to this Remark is that the fa torization is due to the additional
statisti al averaging in the large-N limit. There is no reason to assume
an existen e of the master eld in the strong sense in order to explain the
fa torization.

254

1=N

CHAPTER 3.

EXPANSION

3.3.

3.3 QCD in loop spa e


QCD an be entirely reformulated in terms of the olorless omposite eld
 (C ) | the tra e of the Wilson loop for losed ontours. This fa t involves
two main steps:

x1

This approa h is espe ially useful in the large-N limit where everything
is expressed via the va uum expe tation value of  (C ) | the Wilson loop
average. Observables are given by summing the Wilson loop average over
paths with the same weight as in free theory. The Wilson loop average obeys
itself a lose fun tional equation | the loop equation.
We begin this Se tion with presenting the formulas whi h relate observables to the Wilson loops. Then we translate quantum equation of motion
of Yang{Mills theory into loop spa e. We derive the losed equation for the
Wilson loop average as N ! 1 and dis uss its various properties, in luding
a non-perturbative regularization. Finally, we brie y omment on what is
known about solutions of the loop equation.

3.3.1 Observables in terms of Wilson loops


All observables in QCD an be expressed via the Wilson loops  (C ) de ned
by Eq. (3.2.119). This property was rst advo ated by Wilson [Wil74 on a
latti e. Cal ulation of QCD observables an be divided in two steps:
1) Cal ulation of the Wilson loop averages for arbitrary ontours.
2) Summation of the Wilson loop averages over the ontours with some
weight depending on a given observable.
At nite N , observables are expressed via the n-loop averages

h  (C1 )     (Cn ) i ;

(3.3.1)

whi h are analogous to the n-point Green fun tions (1.2.45). The appropriate
formulas for the ontinuum theory an be found in Ref. [MM81.
Great simpli ations o ur in these formulas at N = 1, when all observables are expressed only via the one-loop average

W (C ) =

h  (C ) i  N1 tr P e ig


C dx A

...........................
.....
.....
..
...
....
...
.
.
.....
..
.
.
.
.
.
.
.
.
.
.
.
.
.......
.....
.....
...
................................

(3.3.2)

........................
.....
....
.
...
....
.
.
..
.
.
...
.....
.
.
.
.
.
.
.......
......
.......... x
...
...... .......................
3
......

x2 ............

x2
x1

i) All the observables are expressed via  (C ).


ii) Dynami s is entirely reformulated in terms of  (C ).

Wn (C1 ; : : : ; Cn ) =

255

QCD IN LOOP SPACE

b)

a)

Fig. 3.26: Contours in the sum over paths representing observables: a) in Eq. (3.3.3)

and b) in Eq. (3.3.4). The ontour a) passes two nailed points x1 and x2 .
The ontour b) passes three nailed points x1 , x2 , and x3 .

This is asso iated with the quen hed approximation dis ussed on p. 164.
For example, the average of the produ t of two olorless quark ve tor
urrents (3.2.93) is given at large N by

  (x1 )   (x2 ) =

C 3x1 ;x2

J (C ) h  (C ) i ;

(3.3.3)

where the sum runs over ontours C passing through the points x1 and x2 as is
depi ted in Fig. 3.26a. An analogous formula for the ( onne ted) orrelators
of three quark s alar urrents reads

 (x1 )  (x2 )  (x3 )


onn =

C 3x1 ;x2 ;x3

J (C ) h  (C ) i ;

(3.3.4)

where the sum runs over ontours C passing through the three points x1 , x2 ,
and x3 as is depi ted in Fig. 3.26b. A general ( onne ted) orrelator of n
quark urrents is given by a similar formula with C passing through n points
x1 , . . . , xn (some of them may oin ide).
The weights J (C ) in Eq. (3.3.3) and J (C ) in Eq. (3.3.4) are ompletely
determined by free theory. If quarks were s alars rather than spinors, then
we would get

J (C ) = e

1 2
2m 

1 
2 0

dt z_2 (t)

= e mL(C )

s alar quarks ; (3.3.5)

where L(C ) is the length of the ( losed) ontour C , as was shown in Subse t. 1.1.6. Using the notation (1.1.152), we an rewrite Eq. (3.3.4) for s alar

256

CHAPTER 3.

quarks as

y (x1 ) y (x2 ) y (x3 )

onn

X0

1=N

EXPANSION

h  (C ) i :

C 3x1 ;x2 ;x3

(3.3.6)

Therefore, we get the sum over paths of the Wilson loop, likewise in Subse t. 1.1.7 and Problem 2.5.
For spinor quarks, an additional disentangling of the gamma-matri es is
needed. This an be done in terms of a path integral over the momentum
variable, with k (t) (0  t   ) being an appropriate traje tory. The result
reads9

J (C ) =
and

J (C ) =

Dk (t) sp P e

R
0

dt[ik (t)(x_  (t)  (t))+m

R

Dk (t) sp P  (t1 )  (t2 ) e

(3.3.7)

dt[ik (t)(x_  (t)  (t))+m

;
(3.3.8)
where the values t1 and t2 of the parameter t are asso iated with the points
x1 and x2 in Eq. (3.3.3), and the symbol of P{ordering puts the matri es 
and  at a proper order.
0

Derive Eqs. (3.3.7) and (3.3.8).


Solution Sin e the spinor eld
enters the QCD a tion quadrati ally, it an be
integrated out in the orrelators (3.3.3), (3.3.4), so that they an be represented, in
the rst quantized language, via the resolvent of the Dira operator in the external
eld A , with subsequent averaging over A . Pro eeding as in Se t. 1.1, we express
the resolvent by


Z1 
 
1
 r
b +m

e

y
(3.3.9)
x
=
d
y

x
rb + m
Problem 3.19

and represent the matrix element of the exponential of the Dira operator as

 

y e  rb+m x

= e m P e

R

dt r
b (t) (d)
0
(x

y) :

(3.3.10)

In order to disentangle the RHS, we insert the unity represented by


1 =

z(0)=x
9 See,

, Ref. [BNZ79.

e.g.

Dz (t)

Dp(t) e

R
i 0 dt p (t)z_ (t)

(3.3.11)

3.3.

257

QCD IN LOOP SPACE

where the path integration over p (t) is unrestri ted,


and p ( ) are in luded. Then we get

 

y e  rb+m x

P e

= e m

Dz (t)

i.e.

the integrals over p (0)

Dp(t)

z(0)=x
dt
ip
(
t
)
z
_
(
t
)
ip
[
(
 
 (t)+A (t))  (t)+ (t)z_ (t) (d) (x
0

R

y ) ; (3.3.12)

whose equivalen e to the original expression is obvious sin e everything ommutes


under the sign of the P{ordering (so that we an substitute p (t) = i (t) in the
integrand).
By making the hange of the integration variable

p (t) = k (t) + iA (t)

(3.3.13)

and pro eeding as in Problem 1.12, we represent the RHS of Eq. (3.3.12) by

 

b +m
y e  r
x

= e m

Dz (t)

Dk(t)

z(0)=x
dt
ik
(
t
)
z
_
(
t
)
[  (   (t)) z_ (t)A (t)+ (t)z_ (t) (d) (x
0

R

P e
= e m

Dz (t)

z(0)=x
z( )=y

Pe

Ry
Dk(t) P e x dz A (z)

R
i 0 dt k (t)[z_ (t)  (t)

y)

(3.3.14)

where the rst P{exponential on the RHS depends only on olor matri es (it is
nothing but the non-Abelian phase fa tor), and the se ond one depends only on
spinor matri es. In Ref. [BNZ79, Eq. (3.3.14) is derived by dis retizing paths.
Equation (3.3.14) leads to Eqs. (3.3.7) and (3.3.8).

Remark on renormalization of Wilson loops


Perturbation theory for W (C ) an be obtained by expanding the path-ordered
exponential in the de nition (3.3.2) in g (see Eq. (3.2.47)) and averaging over
the gluon eld A . Be ause of ultraviolet divergen ies, we need a (gauge
invariant) regularization. After su h a regularization introdu ed, the Wilson
loop average for a smooth ontour C of the type in Fig. 3.27a reads
2 (N 2 1) L(C )
W (C ) = e g 8N a Wren (C ) ;

(3.3.15)

258

CHAPTER 3.

.........................
....
....
...
....
...
....
.
...
.
.. .
....
.
.
...
.
..
...
.....
...............................

1=N

EXPANSION

.........................
..
......
....
...... ... .
....
........
....
... . ..
..
... .. ....
...
.....
.
.
.
....
..
...
.....
....... ....................
.....

a)

b)

Fig. 3.27: Examples of a) smooth ontour and b) ontour with a usp. The tangent
ve tor to the ontour jumps through angle at the usp.

where a is the uto , L(C ) is the length of C , and Wren (C ) is nite when
expressed via the renormalized harge gR . The exponential fa tor is due to the
renormalization of the mass of a heavy test quark, whi h was already dis ussed
in the Remark on p. 117. This fa tor does not emerge in the dimensional
regularization where d = 4 ". The multipli ative renormalization of the
smooth Wilson loop was shown in Refs. [GN80, Pol80, DV80.
If the ontour C has a usp (or usps) but no self-interse tions as is
illustrated by Fig. 3.27b, then W (C ) is still multipli atively renormalizable [BNS81:

W (C ) = Z ( ) Wren (C ) ;

(3.3.16)

while the (divergent) fa tor Z ( ) depends on the usp angle (or angles) (or
's) and Wren (C ) is nite when expressed via the renormalized harge gR .
Cal ulate divergent parts of the Wilson loop average (3.3.2) for
ontours without self-interse tions to order g 2 . Consider the ases of a smooth
ontour C and a ontour with a usp.
2
Solution Expanding the Wilson loop average (3.3.2) in g (see Eq. (3.2.47) and
Problem 2.2), we get

Problem 3.20

W (C ) = 1 + W (2) (C ) + O(g 4 )
with

W (2) (C ) =

g 2 (N 1)
4N

dx

dy D (x y ) ;

where D (x y ) is the gluon propagator (3.2.4).

(3.3.17)
(3.3.18)

3.3.

QCD IN LOOP SPACE

259

Sin e the ontour integral in Eq. (3.3.18) diverges for x = y , we introdu e the
regularization by
Reg: 1
D (x y ) =
(3.3.19)
) 42 (x y)2 + a2
with a being the ultraviolet uto . Parametrizing the ontour C by the fun tion
z ( ), we rewrite the ontour integral in Eq. (3.3.18) as
I
I
Z
Z
z_ (s)z_  (s + t)
1
dx dy
=
ds dt
: (3.3.20)
(x y )2 + a2
[z (s + t) z (s)2 + a2
C

Choosing the proper-length parametrization (1.1.97) when

z_ (s)z (s) = 0 ;


(3.3.21)
expanding in powers of t, and assuming that the ontour C is smooth as is depi ted
in Fig. 3.27a, we get for the integral (3.3.20)
Z
Z
Z
p
ds x_ 2 (s) dt 2 12 2 =  ds x_ 2 (s) =  L(C ) :
(3.3.22)
x_ (s)t + a
a
a
Typi al values of t in the last integral are  a, whi h justi es the expansion in t:
next terms lead to a nite ontribution as a ! 0.
Thus, we get
(N 2 1) L(C )
+ nite as a ! 0
(3.3.23)
W (2) (C ) = g 2
8N
a
for a smooth ontour. This is pre isely the renormalization of the mass of a heavy
test quark due to the intera tion.
If the ontour C is not smooth and has a usp at some value s0 of the parameter, as is depi ted in Fig. 3.27b, then an extra divergent ontribution in the
integral (3.3.20) emerges when s  s0 , t  t0 . Introdu ing s = s s0 and
t = t t0 , we represent this extra divergent term by
Z
Z
1
x_  (s0 + 0)x_  (s0 0) ds dt
[x_  (s0 + 0)s x_  (s0 0)t2 + a2
L(C ) ;
(3.3.24)
= ( ot 1) ln
a
where is the angle of the usp ( os  x_  (s0 + 0)x_  (s0 0)) and the upper limit
of the integrations is hosen to be L(C ) with logarithmi a ura y. Colle ting all
together, we get nally for the divergent part of W (2) :


(N 2 1) L(C ) 1
L(C )
W (2) = g 2
+ ( ot 1) ln
8N
a

a
+ nite as a ! 0 :
(3.3.25)

260

CHAPTER 3.

1=N

EXPANSION

3.3.

261

QCD IN LOOP SPACE

The se ond term in this formula is asso iated with the bremsstrahlung radiation of
a parti le hanging its velo ity when passing the usp. The answers in the Abelian
and non-Abelian ases oin ide to this order in g 2 .

3.3.2 S hwinger{Dyson equations for Wilson loop


Dynami s of (quantum) Yang{Mills theory is des ribed by the quantum equation of motion
rab Fb (x) w=:s: h Aa (x)
(3.3.26)

whi h is analogous to Eq. (1.2.27) for the s alar eld, and is again understood
in the weak sense, i.e. for the averages
D

rab Fb (x) Q [A

= h

Aa (x)

Q [A :

(3.3.27)

The standard set of S hwinger{Dyson equations of Yang{Mills theory emerges


when the fun tional Q [A is hosen in the form of the produ t of A's as in
Eq. (3.2.45).
Stri tly speaking, the last statement is in orre t, sin e we have not added,
in Eqs. (3.3.26) and (3.3.27), ontributions oming from the variation of
gauge- xing and ghost terms in the Yang{Mills a tion. However, these two
ontributions are mutually an elled for gauge-invariant fun tionals Q[A. We
shall deal below only with su h gauge-invariant fun tionals (the Wilson loops).
This is why we have not onsidered the ontribution of the gauge- xing and
ghost terms.
It is also onvenient to use the matrix notation (2.1.5), when Eq. (3.3.26)
for the Wilson loop takes on the form


H

1
tr P r F (x) e C d A
N

 2
H


g
tr
P e C d A ;
=
2N A (x)

(3.3.28)

where we have restored the units with h = 1.


The variational derivative on the RHS an be al ulated by virtue of the
formula


Aij (y)
1 ij kl
(d) (x y ) il kj
(3.3.29)


Akl
N
 (x)

Cyx

6 Cxy

Fig. 3.28: Contours Cyx and Cxy whi h enter the RHS's of Eqs. (3.3.28) and (3.3.32).
whi h is a onsequen e of
Aa (y)
=  (d) (x y) ab :
(3.3.30)
Ab (x)
The se ond term in the parentheses in Eq. (3.3.29) | same as in Eq. (3.2.6)
| is be ause A is a matrix from the adjoint representation of SU(N ).
By using Eq. (3.3.29), we get for the variational derivative on RHS of
Eq. (3.3.28):
I
H

d A
C
tr
=
dy (d) (x y) 
Pe
A (x)
R

R

d A
1
1
1
C d A : (3.3.31)
tr P e Cyx
tr P e Cxy
tr
P
e
3
N
N
N
The ontours Cyx and Cxy , whi h are depi ted in Fig. 3.28, are the parts of
the loop C : from x to y and from y to x, respe tively. They are always losed
due to the presen e of the delta-fun tion. It implies that x and y should be
the same points of spa e but not ne essarily of the ontour (i.e. they may be
asso iated with di erent values of the parameter ).
We nally rewrite Eq. (3.3.28) as


H

1
tr P r F (x) e C d A
N


I
1
(
d
)
h  (C ) i (3.3.32)
=  dy (x y) h  (Cyx )  (Cxy ) i
N2

d A

where we have introdu ed

 =

g2 N
:
2

(3.3.33)

262

CHAPTER 3.

1=N

EXPANSION

Noti e that the RHS of Eq. (3.3.32) is ompletely represented via the ( losed)
Wilson loops.
Prove the an ellation of the ontributions of the gauge- xing and
ghost terms in the Lorentz gauge.
Solution The Yang{Mills a tion, asso iated with the Lorentz gauge, reads

3.3.

263

QCD IN LOOP SPACE

We have thus proven that the ontribution of gauge- xing and ghost terms in
Eq. (3.3.36) are mutually an elled, when applied to the Wilson loop (and in fa t
to any gauge-invariant fun tional).

Problem 3.21

Sg:f : =

1
g2

h
i
2 + 1 tr ( A )2 :
dd x 1 tr F
 
2

(3.3.34)

Sin e  (C ) is gauge invariant, the (in nite) group-volume fa tors, in the numerator
and denominator in the de nition of the average, an el when xing the gauge (see
the Remark on p. 113), and we get

W (C )

DA e S  (C )
R
DA e S
DA det  r e Sg:f :  (C )
R
;
DA det  r e Sg:f :

(3.3.35)

where det  r is asso iated with ghosts.


The S hwinger{Dyson equation for the Yang{Mills theory in the Lorentz gauge
is

w:s:
b
rab
 F (x) = h Aa (x)


+ 1   Aa +  f ab Gb x0 = x; x; A ;
(3.3.36)

x
where Gb (x0 ; x; A) is the Green fun tion of the ghost in an external eld A .
Applying this equation to the Wilson loop and using the gauge Ward identity (the
Slavnov{Taylor identity), we transform the ontribution from the se ond term on
the RHS to
E
1D1
tr   A U (Cxx )
N
g:f :
I
D
E
 
1 
2
b ( ) G a (; x; A)
d
= g
tr U (Cx ) ta U (Cx ) tb
r
N
x
g:f :
C

g2

g2

D
1 tr U (C ) ta U (C ) tb   Gba (; x; A)E
d 
x
x
 N
x
g:f :

1 tr U (C ) ta ; tb   Gba x0 = x; x; AE


xx
N
x
g:f :

(3.3.37)

whi h exa tly an els the ontribution from the ghost term in Eq. (3.3.36).

3.3.3 Path and area derivatives


As we already mentioned, the RHS of Eq. (3.3.32) is ompletely represented
via the ( losed) Wilson loops. It is ru ial for the loop-spa e formulation of
QCD that the LHS of Eq. (3.3.32) an also be represented in loop spa e as
some operator applied to the Wilson loop. To do this we need to develop a
di erential al ulus in loop spa e.
Loop spa e onsists of arbitrary ontinuous losed loops, C . They an
be des ribed in a parametri form by the fun tions x () 2 L2 ,10 where
0    f and  = 1; : : : ; d, whi h take on values in a d-dimensional Eu lidean spa e. The fun tions x () an be dis ontinuous, generally speaking,
for an arbitrary hoi e of the parameter . The ontinuity of the loop C
implies a ontinuous dependen e on parameters of the type of proper length

s() =

Z

0

d0 x_ 2 (0 )

(3.3.38)

where x_  () = dx ()=d.


The fun tions x () 2 L2 whi h are asso iated with the elements of loop
spa e obey the following restri tions:
i) The points  = 0 and  = f are identi ed: x (0 ) = x (f ) | the
loops are losed.
ii) The fun tions x () and  x () +  , with  and  independent
of , represent the same element of the loop spa e | rotational and
translational invarian e.
iii) The fun tions x () and x (0 ) with 0 = f (); f 0 ()  0 des ribe the
same loop | reparametrization invarian e.
An example of fun tionals whi h are de ned on the elements of loop spa e is
the Wilson loop average (3.3.2) or, more generally, the n-loop average (3.3.1).
The di erential al ulus in loop spa e is built out of the path and area
derivatives.

10 Let us remind that L2 stands for the Hilbert spa e of fun tions x () whose square is
R
integrable over the Lebesgue measure: f dx2 () < 1. We already mentioned this in
0
the Remark on p. 19.

264

1=N

CHAPTER 3.

EXPANSION

The area derivative of a fun tional F (C ) is de ned by the di eren e


2

F (C )
 (x)

x 
>- A

1 4 
j j F

F

13
A5

(3.3.39)
where an in nitesimal loop C (x) is atta hed to a given loop at the point x
in the  -plane and j j stands for the area en losed by the C (x). For
a re tangular loop C (x), one gets

 = dx ^ dx ;

(3.3.40)

where the symbol ^ implies antisymmetrization.


Analogously, the path derivative is de ned by
2

x F (Cxx ) 

1 4 
jx j F

- A

xq

F

13
A5

(3.3.41)
where x is an in nitesimal path along whi h the point x is shifted from the
loop and jx j stands for the length of the x .
These two di erential operations are well-de ned for so- alled fun tionals
of the Stokes type whi h satisfy the ba ktra king ondition | they do not
hange when an appendix passing ba k and forth is added to the loop at some
point x:
0

F

1
A

F

(3.3.42)

This ondition is equivalent to the Bian hi identity of Yang{Mills theory and


is obviously satis ed by the Wilson loop (3.3.2) due to the properties of the
non-Abelian phase fa tor (see Eq. (2.1.48)). Su h fun tionals are known in
mathemati s as Chen integrals.11
A simple example of the Stokes fun tional is the area of the minimal
surfa e, Amin(C ). It obviously satis es Eq. (3.3.42). Otherwise, the length
11 See,

, Ref. [Tav93 whi h ontains de nitions of path and area derivatives in this

e.g.

language.

3.3.

QCD IN LOOP SPACE

265

L(C ) of the loop C is not a Stokes fun tional, sin e the lengths of ontours
on the LHS and RHS of Eq. (3.3.42) are di erent.
For the Stokes fun tionals, the variation on the RHS of Eq. (3.3.39) is
proportional to the area en losed by the in nitesimally small loop C (x)
and does not depend on its shape. Analogously, the variation on the RHS of
Eq. (3.3.41) is proportional to the length of the in nitesimal path x and
does not depend on its shape.
If x is a regular point (like any point of the ontour for the fun tional (3.3.2)), the RHS of Eq. (3.3.41) vanishes due to the ba ktra king
ondition (3.3.42). In order for the result to be nonvanishing, the point x
should be a marked (or irregular) point. A simple example of the fun tional
with a marked point x is
R
tr  a
d A () 
a [Cxx 
(3.3.43)
t P e Cxx
N
with the SU(N ) generator ta inserted in the path-ordered produ t at the
point x.
The area derivative of the Wilson loop is given by the Mandelstam formula
H
H



1
1
tr P e C d A =
tr P F (x) e C d A : (3.3.44)
 (x) N
N
In order to prove it, it is onvenient to hoose C (x) to be a re tangle
in the  -plane, as was done in Problem 2.8, and straightforwardly use the
de nition (3.3.39). The sense of Eq. (3.3.44) is very simple: F is a urvature
asso iated with the onne tion A , as we dis ussed in the Remark on p. 99.
The fun tional on the RHS of Eq. (3.3.44) has a marked point x, and is of
the type in Eq. (3.3.43). When the path derivative a ts on su h a fun tional
a ording to the de nition (3.3.41), the result reads
H
H


1
1
tr P r B (x) e C d A ; (3.3.45)
x tr P B (x) e C d A =
N
N
where
r B =  B [A ; B
(3.3.46)
is the ovariant derivative (2.1.11) in the adjoint representation (see also
Problem 2.7).
Combining Eqs. (3.3.44) and (3.3.45), we nally represent the expression
on the LHS of Eq. (3.3.28) (or Eq. (3.3.32)) as
H
H


1
1

tr P r F (x) e C d A = x


tr P e C d A ; (3.3.47)
N
 (x) N

266

CHAPTER 3.

Ordinary spa e

EXPANSION

 (x)

ovariant derivative

The standard variational derivative, =x (), an be expressed via the path
and area derivatives by the formula

area derivative

m
X

= x_  ()
+  xi ( i ) ;
x ()
 (x()) i=1 

x path derivative

r^F =0

Bian hi identity

Stokes fun tionals

r F

S hwinger-Dyson
equations

Loop
equations

= =A

loop spa e.

i.e. via the a tion of the path and area derivatives on the Wilson loop. It is
therefore rewritten in loop spa e.
A resume of the results of this subse tion is presented in Table 3.2 as a
vo abulary for translation of Yang{Mills theory from the language of ordinary
spa e in the language of loop spa e.

Remark on Bian hi identity for Stokes fun tionals


The ba ktra king relation (3.3.42) an be equivalently represented as

 (x)

 (C ) = 0 ;

(3.3.48)

by hoosing the appendix in Eq. (3.3.42) to be an in nitesimal straight line


in the -dire tion and geometri ally applying the Stokes theorem. Using
Eqs. (3.3.44) and (3.3.45), Eq. (3.3.48) an in turn be rewritten as



H

1
tr P r F (x) e C d A = 0 :
N

(3.3.50)

where the sum on the RHS is present for the ase of a fun tional having m
marked (irregular) points xi  x(i ). A simplest example of the fun tional
with m marked points is just a fun tion of m variables x1 ; : : : ; xm .
By using Eq. (3.3.50), the path derivative an be al ulated as the limiting
pro edure
Z+0

x
(

)
d0
 =
:
(3.3.51)
x (0 )
 0

Table 3.2: Vo abulary for translation of Yang{Mills theory from ordinary spa e in

 x

267

QCD IN LOOP SPACE

Remark on relation to variational derivative

 (C ) loop fun tional

F (x) eld strength

3.3.

Therefore, Eq. (3.3.48) represents the Bian hi identity (2.1.19) in loop spa e.

Loop spa e

 [A phase fa tor

rx

1=N

(3.3.49)

The result is obviously nonvanishing only when x is applied to a fun tional
with x() being a marked point.
It is nontrivial that the area derivative an also be expressed via the
variational derivative [Pol80:

=
 (x())

Z+0

d0 (0

 0

)

:
x (0 ) x ()

(3.3.52)

The point is that the six- omponent quantity, = (x()), is expressed via
the four- omponent one, =x (), whi h is possible be ause the omponents
of = (x()) are dependent due to the loop-spa e Bian hi identity (3.3.48).

3.3.4 Loop equations


By virtue of Eq. (3.3.47), Eq. (3.3.32) an be represented ompletely in loop
spa e:
E
D
 (C )
x
 (x)


I
1
(
d
)
=  dy (x y)  (Cyx )  (Cxy )
 (C ) ; (3.3.53)
N 2
C

268

CHAPTER 3.

1=N

EXPANSION

or, using the de nitions (3.3.1) and (3.3.2) of the loop averages, as

x
W (C )
 (x)


I
1
(
d
)
W (C ) : (3.3.54)
=  dy (x y) W2 (Cyx ; Cxy )
N 2
C

This equation is not losed. Having started from W (C ), we obtain another


quantity, W2 (C1 ; C2 ), so that Eq. (3.3.54) onne ts the one-loop average with
a two-loop one. This is similar to the ase of the (quantum) '3 -theory, whose
S hwinger{Dyson equations (1.2.47) onne t the n-point Green fun tions with
di erent n. We shall derive this omplete set of equations for the n-loop
averages in this Subse tion later on.
However, the two-loop average fa torizes in the large-N limit:


1
;
(3.3.55)
W2 (C1 ; C2 ) = W (C1 ) W (C2 ) + O
N 2
as was dis ussed in Subse t. 3.2.6. Keeping the onstant  (de ned by
Eq. (3.3.33)) xed in the large-N limit as is pres ribed by Eq. (3.2.13), we
get [MM79
I

x
W (C ) =  dy (d) (x y) W (Cyx ) W (Cxy ) (3.3.56)
 (x)
C

as N ! 1.
Equation (3.3.56) is a losed equation for the Wilson loop average in the
large-N limit. It is referred to as the loop equation or the Makeenko{Migdal
equation.
To nd W (C ), Eq. (3.3.56) should be solved in the lass of Stokes fun tionals with the initial ondition

W (0) = 1

(3.3.57)

for loops whi h are shrunk to points. This is a onsequen e of the obvious
property of the Wilson loop
H 
(3.3.58)
e 0 d A = 1
and the normalization

h1i

= 1

(3.3.59)

3.3.

269

QCD IN LOOP SPACE

of the averages.
The fa torization (3.3.55) an itself be derived from the hain of loop
equations. Pro eeding as before, we get
1 x

W (C ; : : : ; Cn )
   (x) n 1
=

C1

dy (d) (x y) Wn+1 (Cxy ; Cyx ; : : : ; Cn )


I

h
1
(d) (x y ) W
dy


n
2
j 2 N Cj

1
W ( C ; : : : ; Cn )
N 2 n 1

C1 Cj ; : : : ; Cj ; : : : ; Cn

W n ( C1 ; : : : ; C n ) :

(3.3.60)

Here x belongs to C1 ; C1 Cj stands for the joining of C1 and Cj ; Cj means


that Cj is omitted.
Equation (3.3.60) looks like Eq. (1.2.47) for the '3 -theory. Moreover,
the number of olors N enters Eq. (3.3.60) simply as a s alar fa tor N 2 ,
likewise Plank's onstant h enters Eq. (1.2.47). It is the major advantage of
the use of loop spa e. What is said in Subse t. 3.2.8 about the \semi lassi al"
nature of the 1=N -expansion of QCD is expli itly realized in Eq. (3.3.60). Its
expansion in 1=N is straightforward.
At N = 1, Eq. (3.3.60) is simpli ed to

x

W (C ; : : : ; Cn )
 (x) n 1

= 

dy (d) (x y) Wn+1 (Cyx ; Cxy ; : : : ; Cn ) :

(3.3.61)

C1

This equation possesses a fa torized solution




1
Wn (C1 ; : : : ; Cn ) = h (C1 )i : : : h (Cn )i + O
N 2


1
 W (C1 )    W (Cn ) + O N 2

(3.3.62)

provided W (C ) obeys Eq. (3.3.56) whi h plays the role of a \ lassi al" equation in the large-N limit. Thus, we have given a non-perturbative proof of
the large-N fa torization of the Wilson loops.

270

1=N

CHAPTER 3.

. . .

...........

...
.....

. . .

. . .

C + p
a)

3.3.

. . .

xn.pp.p.pp pppppppp pppppppppppppppppppppppp p pp p p p p pppppppppppppppppppppppppppp

...
.....

..... .. .......
.. . .. .......
.......... ... .........
........ ....
.. ..
.....
...........

x1

p

p pp p
ppp p ..
pppp
ppppp ............
ppp
.
.
.
pppp pp
.
ppp
..
.....
.
pp
..
pp
.......
pp
.
.
pp p
pp
...... .......
pp
pp
.
.
.
.
.
......... ........
ppp
pppp
.
.
.
.
p
...
..
.
ppp
(n) .................................................ppppp
p..p................................................
.
p
ppp
...
..
.....
pp
pp
........................
ppp
pp
p
p
ppp
p
..
........
pp
ppp
..
pp
......
ppp
.. ..
ppp pp
......
pppp
p
.
p
ppppp
.
pppp
ppp pp ..........
ppppppp
p
p
pp ppppp ..
p
p
p
p
p
p
p
p
.p.pppppppp ppppppppp pppppppppppppppp pp pppppppppp
ppppppp pp

xi

x2

b)

Fig. 3.29: Contours a) C + p and b) C p on the RHS of the latti e loop equation (3.3.63).

Fig. 3.30: Graphi representation of the terms on the RHS of Eq. (3.3.64).

3.3.5 Relation to planar diagrams

Derive a latti e analog of the loop equation.


Solution The derivation is similar to that in Problem 2.12 for the lassi al ase.
We perform the shift (2.2.22) in the de nition (2.2.42) of the latti e Wilson loop
average. Likewise Eqs. (2.2.24) and (3.3.56), we get

Problem 3.22

X
X [W
 (l) WCyx WCxy :
C +p WC p =
2
2N p
l2C

271

QCD IN LOOP SPACE

x............

x ...................

....
..
...........
...
............. ...........
........ ...
.. ..
....
...........

...........

EXPANSION

(3.3.63)

Here the ontours C + p and C p are obtained from Cxx by adding the boundary
of the plaquette p ( p means that the orientation of the boundary is opposite)
and the sum over p goes over the 2(d 1) plaquettes involving the link at whi h the
shift of Ux; is performed. This ontours are depi ted in Fig. 3.29.
The sum on the RHS goes over the links belonging to the ontour C . The unit
ve tor  (l) = 0; 1 stands for the proje tion of the (oriented) link l 2 C on the axis
 ( (l) = 1, 1 or 0 when the dire tions are parallel, antiparallel or perpendi ular,
respe tively). The point y is de ned as the beginning of the link l if it has the
positive dire tion, or as the end of l if it has the negative one. Su h an asymmetry
is due to the fa t that we have performed the right shift (2.2.22) of Ux; . The
Krone ker symbol xy guarantees that Cyx and Cxy are always losed.
Equation (3.3.63) is a latti e regularization of the ontinuum loop equation
(3.3.56). The loop equation on the latti e was rst dis ussed in Refs. [Foe79, Egu79.

The perturbation-theory expansion of the Wilson loop average an be al ulated from Eq. (3.2.47) whi h we represent in the form
I
I
1 I
X


1
2
W (C ) = 1 +
dx1 dx2 : : : dxnn
n=2 C

C
C
(n)
 (1; 2; : : : ; n) G1 2 n (x1 ; x2 ; : : : ; xn ) ;

(3.3.64)

where  (1; 2; : : : ; n) orders the points x1 , : : :, xn along ontour in the y li


order and G(n1)n is given by Eq. (3.2.72). This -fun tion has the meaning
of the propagator of a test heavy parti le whi h lives in the ontour C (see
Problem 2.2).
We assume, for de nitiveness, the dimensional regularization throughout
this Subse tion to make all the integrals well-de ned.
Ea h term on the RHS of Eq. (3.3.64) an be onveniently represented
by the diagram in Fig. 3.30, where the integration over the ontour C is
asso iated with ea h point xi lying in the ontour C .
These diagrams are analogous to those dis ussed in Subse t. 3.2.3 with
one external boundary | the Wilson loop in the given ase. This was already mentioned in the Remark on p. 230. In the large-N limit, only planar
diagrams survive. Some of them, whi h are of the lowest order in , are depi ted in Fig. 3.31. The diagram in Fig. 3.31a has been already onsidered in
Problem 3.20 (see Eq. (3.3.18)).

272

CHAPTER 3.

1=N

EXPANSION

3.3.

273

QCD IN LOOP SPACE

Solve Eq. (3.3.56) by the ansatz (3.3.64) to order .


To order , we an restri t ourselves by the n = 2 term in the
ansatz (3.3.64). For the -fun tion, we have
1
 (1; 2)  [(1; 2) + (2; 1) = 1 :
(3.3.66)
2
2
The meaning of this formula is obvious: there is no y li ordering for two points.
We rewrite therefore the ansatz as
I
I

W (C ) = 1 1  dx dy D (x y ) + O 2
(3.3.67)
2

Problem 3.23
Solution

a)

b)

Fig. 3.31: Planar diagrams for W (C ): a) of order  with gluon propagator, and of
order 2 b) with two nonintera ting gluons and ) with the three-gluon
vertex. Diagrams of order 2 with one-loop insertions to gluon propagator

with some unknown fun tion D (x y ). Its tensor stru ture reads

are not drawn.

The large-N loop equation (3.3.56) des ribes the sum of the planar diagrams. Its iterative solution in  reprodu es the set of planar diagrams for
W (C ) provided the initial ondition (3.3.57) and some boundary onditions
for asymptoti ally large ontours are imposed.
Equation (3.3.64) an be viewed as an ansatz for W (C ) with some unknown fun tions G(n1)n (x1 ; : : : ; xn ) to be determined by the substitution
into the loop equation. To preserve symmetry properties of W (C ), the fun tions G(n) must be symmetri under a y li permutation of the points 1, : : :,
n and depend only on xi xj (translational invarian e). A main advantage of
this ansatz is that it automati ally orresponds to a Stokes fun tional, due to
the properties of ve tor integrals, and the initial ondition (3.3.57) is satis ed.
The a tion of the area and path derivatives on the ansatz (3.3.64) is easily
al ulable. For instan e, the area derivative reads
I
1 I
X
W (C )
=
dx1 1 : : : dxnn  (1; 2; : : : ; n)
 (z )
n=1
C
C
h
 (n+1)
z
z
     G 1 n (z; x1 ; : : : ; xn )
i
n+2)
+ (    ) G( 
1 n (z; z; x1 ; : : : ; xn ) :

D (x y ) =  D (x y ) +   f (x y ) :

(3.3.68)

The se ond (longitudinal) term in this formula does not ontribute to W (C ) sin e
the ontour integral of this term vanishes in Eq. (3.3.67). We an thus write
1

2

W (C ) = 1

dx

dy D (x y ) + O 2 :

(3.3.69)

The area derivative an be easily al ulated by using the Stokes theorem, whi h
gives

 (z )
=

dx

dy D (x y )

0
I
2

dy  D (z y )

and
sin e

z

 (z )

dx



dy  D (z y )A

(3.3.70)

dy D (x y ) = 2

(3.3.65)

The analogy with the Mandelstam formula (3.3.44) is obvious.


More about solving the loop equation by the ansatz (3.3.64) an be found
in Refs. [MM81, BGSN82, Mig83.

dy  2 D (z y )

(3.3.71)

dy  D (x y ) = 0 :

(3.3.72)

Substituting into the loop equation (3.3.56), we get


I

dy

 2 D (x

y) =

dy (d) (x y )

(3.3.73)

274

CHAPTER 3.

1=N

EXPANSION

whi h is equivalent to

 2 D (x

y) =

(d) (x

y)

(3.3.74)

sin e the ontour C is arbitrary. The solution to Eq. (3.3.74) is unique, provided
D (x y ) de reases for large x y , and re overs the propagator (3.2.4).

3.3.6 Loop-spa e Lapla ian and regularization


The loop equation (3.3.56) is not yet entirely formulated in loop spa e. It is
a d-ve tor equation whose both sides depend expli itly on the point x whi h
does not belong to loop spa e. The fa t that we have a d-ve tor equation for
a s alar quantity means, in parti ular, that Eq. (3.3.56) is overspe i ed.
A pra ti al di ulty in solving Eq. (3.3.56) is that the area and path
derivatives, = (x) and x, whi h enter the LHS are ompli ated, generally
speaking, non- ommutative operators. They are intimately related to the
Yang{Mills perturbation theory where they orrespond to the non-Abelian
eld strength F and the ovariant derivative r . However, it is not easy
to apply these operators to a generi fun tional W (C ) whi h is de ned on
elements of loop spa e.
A mu h more onvenient form of the loop equation an be obtained by
integrating both sides of Eq. (3.3.56) over dx along the same ontour C ,
whi h yields
I

dx x
W (C )
 (x)
C

= 

dx

dy (d) (x y) W (Cyx )W (Cxy ) :

(3.3.75)

Now both the operator on the LHS and the fun tional on the RHS are s alars
without labeled points and are well-de ned in loop spa e. The operator on
the LHS of Eq. (3.3.75) an be interpreted as an in nitesimal variation of
elements of loop spa e.
Equations (3.3.56) and (3.3.75) are ompletely equivalent. A proof of
equivalen e of s alar Eq. (3.3.75) and original d-ve tor Eq. (3.3.56) is based
on the important property of Eq. (3.3.56) whose both sides are identi ally
annihilated by the operator x. It is a onsequen e of the identity (see Subse t. 2.1.1)
(3.3.76)
r r F = 21 [F ; F = 0

3.3.

275

QCD IN LOOP SPACE

in the ordinary spa e. Due to this property, the vanishing of the ontour
integral of some ve tor is equivalent to vanishing of the ve tor itself, so that
Eq. (3.3.56) an in turn be dedu ed from Eq. (3.3.75).
Equation (3.3.75) is asso iated with the so- alled se ond-order S hwinger{
Dyson equation
Z

a (x)
dd x r F
Aa (x)
Z

w:s:
(3.3.77)
= h dd x dd y (d) (x y) a
a
A (y) A (x)
in the same sense as Eq. (3.3.56) is asso iated with Eq. (3.3.26). It is alled
\se ond order" sin e the RHS involves two variational derivatives with respe t
to A .
The operator on the LHS of Eq. (3.3.75) is a well-de ned obje t in loop
spa e. When applied to regular fun tionals whi h do not have marked points,
it an be represented, using Eqs. (3.3.51) and (3.3.52), in an equivalent form
I

 
dx x
 (x)
=

C
Zf

0

d

Z+0

 0

d0

:
0
x ( ) x ()

(3.3.78)

As was rst pointed out by Gervais and Neveu [GN79b, this operator is
nothing but a fun tional extension of the Lapla e operator, whi h is known
in mathemati s as the Levy operator.12 Equation (3.3.75) an be represented
in turn as an (inhomogeneous) fun tional Lapla e equation
W (C ) = 

dx

dy (d)(x y) W (Cyx )W (Cxy ) :

(3.3.79)

We shall refer to this equation as the loop-spa e Lapla e equation.


The form (3.3.79) of the loop equation is onvenient for a non-perturbative
ultraviolet regularization.
The idea is to start from the regularized version of Eq. (3.3.77), repla ing
the delta-fun tion on the RHS by the kernel of the regularizing operator:
D

Reg:
ab (d) (x y) =
) y Rab x = Rab (d)(x y)

(3.3.80)

12 See the book by Levy [Lev51, ited in the referen es to Chapter 1, and a re ent

review [Fel86.

276

CHAPTER 3.

with

Rab =

1=N

EXPANSION

2 2 ab
e a r =2 ;

3.3.

(3.3.81)

where r is the ovariant derivative in the adjoint representation. The regularized version of Eq. (3.3.77) is
Z

a (x)
dd x r F
w:s:

h

dd x dd y y Rab x

:
Ab (y) Aa (x)

(3.3.82)

To translate Eq. (3.3.82) in loop spa e, we use the path-integral representation (see Problem 2.5)
E
D
y ab x

Dr(t) e

R 2

1 a
2 0

dt r_ 2 (t)

2 tr

a
t U (r

b
yx )t U (rxy )

r(0)=x
r(a2 )=y

(3.3.83)

with

U (ryx ) =

(3.3.84)

Dr(t) e

2
1 a2
2 0 dt r_ (t)



2 tr ta tb

r(0)=x
r(a2 )=y
2 2
= ab e a  =2 (d) (x y) = ab

1
e
(2a2 )d=2

(x y)2
2a2

(3.3.85)

Cal ulating the variational derivatives on the RHS of Eq. (3.3.82), using
Eq. (3.3.83) and the ompleteness ondition (3.2.6), we get as N ! 1:
Z

dd x dd y

E
D
y ab x

r(0)=x
r(a2 )=y

Dr(t) e

Ab (y) Aa (x)

 (C ) = 

dx

R 2
1 a dtr_ 2 (t)
2 0
 (Cyxrxy )  (Cxy ryx )

dy

(3.3.86)

rxy

ryx

6 Cxy

Fig. 3.32: Contours Cyx rxy and Cxy ryx whi h enter the RHS's of Eqs. (3.3.86) and
(3.3.87).

where the ontours Cyxrxy and Cxy ryx are depi ted in Fig. 3.32. Averaging
over the gauge eld and using the large-N fa torization, we arrive at the
regularized loop-spa e Lapla e equation [HM89
W (C ) = 

Ry
P e x dr A (r) ;

where the integration is over regulator paths r (t) from x to y whose typi al
length is  a. The onventional measure is implied in (3.3.83) so that
Z

Cyx

Aa (x)

277

QCD IN LOOP SPACE

dx

dy

Dr(t) e

R 2
2
1 a
2 0 dt r_ (t) W (C r )W (C r )
yx xy
xy yx

r(0)=x
r(a2 )=y

(3.3.87)
whi h manifestly re overs Eq. (3.3.79) when a ! 0.
The onstru ted regularization is non-perturbative while perturbatively
reprodu es regularized Feynman diagrams. An advantage of this regularization of the loop equation is that the ontours Cyx rxy and Cxy ryx on the RHS
of Eq. (3.3.87) both are losed and do not have marked points if C does not
have. Therefore, Eq. (3.3.87) is written entirely in loop spa e.

Remark on fun tional Lapla ian


It is worth noting that the representation of the fun tional Lapla ian on the
RHS of Eq. (3.3.78) is de ned for a wider lass of fun tionals than Stokes
fun tionals. The point is that the standard de nition of the fun tional Lapla ian from the book by Levy [Lev51, ited in the referen es to Chapter 1, uses
solely the on ept of the se ond variation of a fun tional U [x, namely the
term in the se ond variation whi h is proportional to (x ())2 :

2 U [x =

1
2

Z f

0

00 [x(x ())2 + : : : :
d Uxx

(3.3.88)

278

CHAPTER 3.

1=N

EXPANSION

The fun tional Lapla ian  is then de ned by the formula


U [x =

Z f

0

00 [x :
d Uxx

(3.3.89)

Here U [x an be an arbitrary, not ne essarily parametri invariant, fun tional. To emphasize this obsta le, we use the notation U [x for generi
fun tionals whi h are de ned on L2 spa e in omparison to U (C ) for the
fun tionals whi h are de ned on elements of loop spa e. It is easier to deal
with the whole operator , rather than separately with the area and path
derivatives.
The fun tional Lapla ian is parametri invariant and possesses a number
of remarkable properties. While a nite-dimensional Lapla ian is an operator
of the se ond order, the fun tional Lapla ian is that of the rst order and
satis es the Leibnitz rule
 (UV ) =  (U ) V + U  (V ) :

(3.3.90)

The fun tional Lapla ian an be approximated [Mak88 in loop spa e by


a (se ond-order) partial di erential operator in su h a way to preserve these
properties in the ontinuum limit. This loop-spa e Lapla ian an be inverted
to determine a Green fun tion G (C; C 0 ) in the form of a sum over surfa es
SC;C 0 onne ting two loops:

G (C; C 0 ) =

SC;C 0

::: ;

(3.3.91)

whi h is analogous to the representation (1.1.98) of the Green fun tion of the
ordinary Lapla ian. The standard perturbation theory an then be re overed
by iterating Eq. (3.3.79) (or its regularized version (3.3.87)) in  with the
Green fun tion (3.3.91).

3.3.7 Survey of non-perturbative solutions


While the loop equations were proposed long ago, not mu h is known about
their non-perturbative solutions. We brie y list some of the results.
It was shown in Ref. [MM80 that area law
W (C )  h (C )i / e K Amin(C )
(3.3.92)
satis es the large-N loop equation for asymptoti ally large C . However, a
self- onsisten y equation for K , whi h should relate it to the bare harge and

3.3.

279

QCD IN LOOP SPACE

the uto , was not investigated. In order to do this, one needs more detailed
information about the behavior of W (C ) for intermediate loops.
The free bosoni Nambu{Goto string whi h is de ned as a sum over surfa es spanned by C

W (C ) =

S :S =C

e K A(S) ;

(3.3.93)

with the a tion being the area A (S ) of the surfa e S , is not a solution for intermediate loops. Consequently, QCD does not redu e to this kind of string,
as was originally expe ted in Refs. [GN79a, Nam79, Pol79. Roughly speaking, the ansatz (3.3.93) is not onsistent with the fa torized stru ture on the
RHS of Eq. (3.3.56).
Nevertheless, it was shown that if a free string satis es Eq. (3.3.56), then
the same intera ting string satis es the loop equations for nite N . Here
\free string" means, as usual in string theory, that only surfa es of genus
zero are present in the sum over surfa es, while surfa es or higher genera are
asso iatedwith a string intera tion. The oupling onstant of this intera tion
is O N 2 .
A formal solution of Eq. (3.3.56) for all loops was found by Migdal [Mig81
in the form of a fermioni string

W (C ) =

S :S =C

D e

R 2
d [ k k +  m4 pg

(3.3.94)

where the world sheet of the string is parametrized by the oordinates 1 and
2 for whi h the 2-dimensional metri is onformal, i.e. diagonal. The eld
( ) des ribes 2-dimensional elementary fermions (elves) living in the surfa e
S , and m stands for their mass. Elves were introdu ed to provide fa torization
whi h now holds due to some remarkable properties of 2-dimensional fermions.
For large loops, the internal fermioni stru ture be omes frozen, so that the
empty string behavior (3.3.92) is re overed. For small loops, the elves are
ne essary for asymptoti freedom. However, it is un lear whether or not the
string solution (3.3.94) is pra ti ally useful for a study of multi olor QCD,
sin e the methods of dealing with the string theory in four dimensions are
not yet developed.
A very interesting solution of the large-N loop equation on a latti e was
found by Egu hi and Kawai [EK82. They showed that the SU(N ) gauge
theory on an in nite latti e redu es at N = 1 to the model on a hyper ube,

280

CHAPTER 3.

1=N

EXPANSION

Z Y
d

=1

dU e

P
y y
1
> Re N tr(U U U U ) :

(3.3.95)

Here the matri es U ( = 1; : : : ; d) depend only on the dire tion , rather


than on a spa e-time point x. The equivalen e is possible only at N = 1,
when the spa e-time dependen e is absorbed by the internal symmetry group.
More about this large-N redu tion will be said in the next Se tion.

3.3.8 Wilson loops in QCD2


Two-dimensional QCD is popular sin e the paper by 't Hooft [Hoo74b as a
simpli ed model of QCD4 .
One an always hoose the axial gauge

A1 = 0 ;

(3.3.96)

so that the ommutator in the non-Abelian eld strength (2.1.15) vanishes


in two dimensions. Therefore, there is no gluon self-intera tion in this gauge
and the theory looks, at the rst glan e, like the Abelian one.
The Wilson loop average in QCD2 an be straightforwardly al ulated via
the expansion (3.3.64) where only dis onne ted (free) parts of the orrelators
G(n) for even n should be left, sin e there is no intera tion. Only the planar
stru ture of olor indi es ontributes at N = 1. Diagrammati ally, the
diagrams of the type depi ted in Fig. 3.31a and Fig. 3.31b are relevant for
ontours without self-interse tions, while that in Fig. 3.31 should be omitted
in two dimensions.
The olor stru ture of the relevant planar diagrams an be redu ed by the
virtue of the formula
X
N
(3.3.97)
(ta )ik (ta )kj = ij ;
2
a
whi h is a onsequen e of the ompleteness ondition (3.2.6) at large N . We
have
I
I
I
I
1
X
k



1
k
1
W (C ) = 1 +
( ) dx1
dx2    dx2k 1 dx2kk
k

C
 (1; 2; : : : ; 2k) D1 1 (x1

281

QCD IN LOOP SPACE

where the points x1 , : : :, x2k are still y li ordered along the ontour. Similar
to Problem 2.2, we an exponentiate the RHS of Eq. (3.3.98) to get nally

whi h is des ribed by the partition fun tion

ZEK ( ) =

3.3.

C
C
C
x2 )    Dk k (x2k 1 x2k ) ;(3.3.98)

W (C ) = e

2 C dx

 H dy D (x y)
C
:

(3.3.99)

This is the same formula as in the Abelian ase if  stands for e2 .


The propagator D (x; y) is, stri tly speaking, the one in the axial gauge
(3.3.96) whi h reads
1
jx y j (1) (x2 y2 ) :
(3.3.100)
2 2  2 1 1
However, the ontour integral on the RHS of Eq. (3.3.99) is gauge invariant,
and we an simply hoose

D (x y) =

D (x y) =  D (x y) :

(3.3.101)

In two dimensions13 we have

D (x y ) =

`2
1
ln
;
4 (x y)2

(3.3.102)

where ` is an arbitrary parameter of the dimension of length. Nothing depends


on it be ause the ontour integral of a onstant vanishes.
The ontour integral in the exponent on the RHS of Eq. (3.3.99) an be
graphi ally represented as is depi ted in Fig. 3.33, where x2 = y2 due to the
delta-fun tion in Eq. (3.3.100) and the bold line represents jx1 y1 j. This
gives
I

dx

dy D (x y) = A (C )

(3.3.103)

where A (C ) is the area en losed by the ontour C . We get nally

W (C ) = e

 A(C )
2

for the ontours without self-interse tions.


13 In d dimensions

D (x y ) =


1
d
2
4d=2

1
d=2 1 :
(x y)2

1 

(3.3.104)

282

CHAPTER 3.

x2 =

1=N

EXPANSION

.....................
........
......
.....
....
.
.
..
...
....
..
...
..
y2 ......
.....
.
....
...
......
................................

Fig. 3.33: Graphi representation of the ontour integral on the LHS of Eq. (3.3.103)
in the axial gauge. The bold line represents the gluon propagator (3.3.100)
with x2 = y2 due to the delta-fun tion.

Therefore, area law holds in two dimensions both in the non-Abelian and
Abelian ases. This is, roughly speaking, be ause of the form of the twodimensional propagator (3.3.102) whi h falls down with the distan e only
logarithmi ally in the Feynman gauge.
Prove Eq. (3.3.104) in the Feynman gauge.
To prove Eq. (3.3.103) in the Feynman gauge (3.3.101), (3.3.102), we
note that the area element in two dimensions an be represented by

Problem 3.24
Solution

 dx ^ dx = e d2 x ;

d  (x)

(3.3.105)

where e is the antisymmetri tensor e12 = e21 = 1. Therefore, the area an be
represented by the double integral

A (C ) = 1
2

d  (x)

S (C )

d  (y ) (2) (x y )

(3.3.106)

S (C )

whi h goes along the surfa e S (C ) en losed by the (non-interse ting) loop C .
Applying the Stokes' theorem, we get
I

dx

dy  D (x y ) =

d  (x) 
Z

d  (x)

S (C )

=
1
=
2

1
2

dy  D (x y )

........................
.........
.....
....
.
...
.
.
...
.....
...
.
...
A
1
...
...
.
.
..
...
.
....
.
......
.....
......... .............
.
.
.
.. . ...
..... .....
.... A2 ....
.... ....
.............

d  (x)

S (C )

S (C )

d  (y )   D (x y )
Z

d  (y )  2 D (x y )

S (C )

d  (x)

S (C )

d  (y ) (2) (x

b)

Fig. 3.34: Contours with one self-interse tion: A1 and A2 stand for the areas of
the proper windows. The total area en losed by the ontour in Fig. a) is
A1 + A2 . The areas en losed by the exterior and interior loops in Fig. b)
are A1 + A2 and A2 , respe tively, while the total area of the surfa e with
the folding is A1 + 2A2 .

Using Eq. (3.3.106) we prove Eq. (3.3.104) in the Feynman gauge.


It is worth noting that Eq. (3.3.107) is based only on the Stokes' theorem and
holds for ontours with arbitrary self-interse tions. On the ontrary, Eq. (3.3.106)
itself is valid only for non-interse ting loops.

The di eren e between the Abelian and non-Abelian ases shows up for
the ontours with self-interse tions.
We rst note that the simple formula (3.3.103) does not hold for ontours
with arbitrary self-interse tions.
The simplest ontours with one self-interse tion are depi ted in Fig. 3.34.
There is nothing spe ial about the ontour in Fig. 3.34a. Equation (3.3.103)
still holds in this ase with A (C ) being the total area A (C ) = A1 + A2 .
The Wilson loop average for the ontour in Fig. 3.34a oin ides both for
the Abelian and non-Abelian ases and equals
 (A1 +A2 )
2

(3.3.108)

This is nothing but the exponential of the total area.


For the ontour in Fig. 3.34b, we get
I

y ) :(3.3.107)

.............
........... ..................
.....
......
.
.
.
.
....
...
...
.....
...
A
1
..
....
...
...
...
...
.
...
.
.
.
.
.
.
.
.
.
.... ......
..
...
...
.
..
....
..... ..... A2 .... ........
....... .... ..... ........
...........................

a)

W (C ) = e

S (C )

283

QCD IN LOOP SPACE

S (C )

3.3.

dx

dy D (x y) = A1 + 4A2 :

(3.3.109)

This is easy to understand in the axial gauge where the ends of the propagator
line an lie both on the exterior and interior loops, or one end at the exterior

284

CHAPTER 3.

.............................
.........
.....
.....
....
.
.
.
...
...
....
...
...
...
..
.
...
...
...
.
.
.
.
.
.
.
.
.
...
... ......
.
.
.
.
.
... ...
....
..... .....
. .
....... ..... ....................
.
.............. ......

...........................
.........
.....
.....
....
.
.
.
...
...
.....
..
....
..
...
.
...
...
.
...
.
.
.
.
.
.
.
.
. .....
.
.
.
.
.
.
...
..
..
..... ....
..... .... .... ..........
.................................
...

...........................
..........
.....
.....
....
.
.
.
...
...
.....
...
....
...
...
...
...
.
...
.
.
.
.
.
.
.
..
.... .......
...
..
.... ....
.
...... ... .... .........
...................................
...

b)

a)

1=N

EXPANSION

......
.............. ...................
......
.....
.
.
.
...
..
...
...
...
....
..
...
.
...
...
...
.
.
.
.
.
.
.
.
.
.
...
.. ... ......
.
.
....
.. ....
.
. .
..... ....
....... ....... ....................
........... ......

3.3.

where P is a polynomial of the areas of individual windows and Area is the


total area of the surfa e with foldings, were rst al ulated in Ref. [KK80 by
solving the two-dimensional loop equation and in Ref. [Bra80 by applying
the non-Abelian Stokes' theorem. The latti e version is given in Ref. [KK81.

line lie both on a) exterior and b) interior loops, or ), d) one end on the
exterior loop and another end on the interior loop.

Demonstrate that Eq. (3.3.99) satis es the Abelian loop equation

Problem 3.25

x

d)

Fig. 3.35: Three type of ontribution in Eq. (3.3.109) The ends of the propagator

285

QCD IN LOOP SPACE

 (x)

W (C ) = 

dy (d) (x y ) W (C ) :

(3.3.113)

The al ulation is the same as in Problem 3.23. In d = 2 one an


alternatively use [OP81 the expression on the RHS of Eq. (3.3.107).

Solution

Obtain Eqs. (3.3.108) and (3.3.111) for the ontours with one selfinterse tion by solving the loop equation (3.3.56).

Solution Let us multiply Eq. (3.3.56) in d = 2 by e and integrate over dx along
a small (open) pie e C 0 of the ontour C in luding the point of self-interse tion. We
get

Problem 3.26

loop and the other end on the interior loop. These ases are illustrated by
Fig. 3.35. The ontributions of the diagrams in Fig. 3.35a,b, ,d are A1 + A2 ,
A2 , A2 , and A2 , respe tively. The result given by Eq. (3.3.109) is obtained
by summing over all four diagrams.
For the ontour in Fig. 3.34b, the Wilson loop average is

W (C ) = e

 (A1 +4A2 )

(3.3.110)

in the Abelian ase and

W (C ) = (1 A2 ) e

 (A1 +2A2 )
2

(3.3.111)

in the non-Abelian ase at N = 1. They oin ide only to the order  as


they should. The di eren e to the next orders is be ause only the diagrams
with one propagator line onne ting the interior and exterior loops are planar
and, therefore, ontribute in the non-Abelian ase. Otherwise, the diagram is
non-planar and vanishes as N ! 1. Noti e, that the exponential of the total
area A (C ) = A1 +2A2 of the surfa e with the folding, whi h is en losed by the
ontour C , appears in the exponent for the non-Abelian ase. The additional
pre-exponential fa tor ould be asso iated with an entropy of foldings of the
surfa e.
The Wilson loop averages (3.3.108) and (3.3.111) in QCD2 at large N as
well as the ones for ontours with arbitrary self-interse tions, whi h have a
generi form

W (C ) = P (A1 ; : : : ; An ) e

 Area
2

(3.3.112)

e

C0

dx x

 (x)

W (C ) = e

C0

dx

dy (2) (x y ) W (Cyx ) W (Cxy ) :

(3.3.114)
The RHS of Eq. (3.3.114) an be al ulated analogously to the known representation for the number of self-interse tions of a loop in two-dimensions. For the ase
of one self-interse tion, we have

e

C0

dx

dy (2) (x y ) W (Cyx ) W (Cxy ) = W (C1 ) W (C2 ) ; (3.3.115)

where C1 and C2 stands, respe tively, for the upper and lower loops in Fig. 3.34a
or the exterior and interior loops in Fig. 3.34b.
The LHS of Eq. (3.3.114) an be transformed as

e

C0

dx x

 (x)

W (C ) =

C0

dx x W (C ) ;
 (x)

(3.3.116)

where = (x) denotes the variational derivative with respe t to the \s alar" area
(3.3.117)
 (x) = 1 e   (x) :
2
The integrand on the RHS of Eq. (3.3.116) is a total derivative and the ontour
integral redu es to the di eren e of the
-variations at the end points of the ontour

286

CHAPTER 3.

1=N

EXPANSION

C 0 , whi h would vanish if no self-interse tions. The RHS of Eq. (3.3.114) also
vanishes if no self-interse tions so W (C ) is determined in this ase by Eq. (3.3.113)
rather than Eq. (3.3.114).
For the ontour in Fig. 3.34a, this gives
.........
........... ..............
...
.....
...
.....
...
...
A
1
...
...
...
..
.
.
...
.
.....
...
............. ............
.............
.
.
.
.. ..
.... A2 ....
....... ......
......

.............
......... .............
.....
...
....
.
A1 ......
.....
.
...
...
...
..
.
.....
.
.
.
........ ........
...................
..... ...
.... A2 .....
.................


  W (C ) :
(3.3.118)
A1 A2
The
-variation of the ontour on the LHS represents the variational derivative.
The minus sign in front of =A1 on the RHS is be ause adding the
-variation
in the rst term on the LHS de reases the area A1 while that in the se ond term
in reases A2 . Then Eq. (3.3.114) takes for the ontour in Fig. 3.34a the form
=


  W (C ) = W (C ) W (C ) :
1
2
A1 A2
For the ontour in Fig. 3.34b, we get quite similarly
..........................
.....
..........
.....
.....
...
......
...
A
...
1
...
..
...
...
...
..
.
...
.
........
..
...
..... ....
..... .... A2 .... ......
..... ... .. ... .......
.......... .......................
......

(3.3.119)

..........................
.....
..........
.....
.....
...
......
...
...
A1
...
..
...
...
...
..
.
...
.
........
..
...
..... ....
.... .... A2 .... ......
..... .... ... .......
...... ........................
........


A1


A1

  W (C ) =  e
A2

 (A1 +A2 )
2



 
W (C ) =  e 2 (A1 +2A2 ) ;
(3.3.123)
A1 A2
respe tively. Their solution is uniquely given by Eqs. (3.3.108) and (3.3.111).
It is worth nothing that the linear Abelian loop equation (3.3.113) an be written
for the ontours in Fig. 3.34a,b as


 
ln W (C ) =  ;
(3.3.124)
A1 A2


 ln W (C ) =  :
2 
(3.3.125)
A1 A2
The operators on the LHS's are always the same for the non-Abelian and Abelian
loop equations, whi h is a general property, but the RHS's generi ally differ: Eqs. (3.3.122) and (3.3.124) for the ontour in Fig. 3.34a oin ide while
Eqs. (3.3.123) and (3.3.125) for the ontour in Fig. 3.34b di er. The solution to
Eq. (3.3.125) is given by (3.3.110).

Prove Eq. (3.3.115) for the ontours with one self-interse tion.
Let the interse tion orresponds to the values s1 and s2 of the parameter
s, i.e. x (s1 ) = x (s2 ). Noting that only the vi inities of s1 and s2 ontribute to
the integral on the LHS of Eq. (3.3.115), we get

Problem 3.27
Solution

C0

dx

dy (2) (x y ) W (Cyx ) W (Cxy )

(3.3.121)

The RHS's of Eqs. (3.3.119) and (3.3.121) are known sin e C1 and C2 have
no self-interse tions so that Eq. (3.3.104) holds for W (C1 ) and W (C2 ). Equations (3.3.119) and (3.3.121) take nally the expli it form [KK80


and

= e x_  (s1 ) x_  (s2 )

  W (C ) = W (C ) W (C ) :
1
2
A2

(3.3.122)

287

QCD IN LOOP SPACE

e

  W (C ) :
(3.3.120)
2 
A1 A2
Now adding the
-variation in the rst term on the LHS in reases A1 and de reases
A2 while that in the se ond term de reases A1 . Equation (3.3.114) takes the form
=

3.3.

ds

dt (2) ((s s1 )x_ (s1 ) (t s2 )x_ (s2 ))

W Cx(s2 )x(s1 ) W Cx(s1 )x(s2 )



e x_  (s1 ) x_  (s2 )
W
C
W
C
x
(
s
)
x
(
s
)
x
(
s
)
x
(
s
)
2
1
1
2
x_ 2 (s1 ) x_ 2 (s2 ) (x_  (s1 ) x_  (s2 ))2


= W Cx(s2 )x(s1 ) W Cx(s1 )x(s2 )
(3.3.126)

whi h is pre isely the RHS of Eq. (3.3.115).

Remark on the string representation


A ni e property of QCD2 at large N is that the exponential of the area en losed by the ontour C emerges14 for the Wilson loop average W (C ). This
14 This is not true, as is already dis ussed, in the Abelian ase for ontours with self-

interse tions.

288

CHAPTER 3.

1=N

EXPANSION

is as it should for the Nambu{Goto string (3.3.93). However, the additional


pre-exponential fa tors (like that in Eq. (3.3.111)) are very di ult to interpret in the stringy language. They may be ome negative for large loops whi h
is impossible for a bosoni string. This expli itly demonstrates in d = 2 the
statement of the previous subse tion that the Nambu{Goto string is not a
solution of the large-N loop equation.

3.3.9 Gross{Witten transition in latti e QCD2


The latti e gauge theory on a two-dimensional latti e is de ned by the partition fun tion (2.2.31) with d = 2:

Z2d ( ) =

Z Y Y

x =1;2

dUx; e S[U ;

(3.3.127)

where the a tion is given by Eq. (2.2.16).


A spe i s of two dimensions is that the number of the latti e sites is equal
to the number of the plaquettes. For this reason, we an always perform su h
a gauge transformation that the link variables are hosen to be equal unity
along one of the axes, say

Ux;1 = 1 :

(3.3.129)

where Np stands for the number of plaquettes of the latti e and Z1p is the
one-matrix integral

Z1p ( ) =

1
dU e ( N Re tr U 1) :

(3.3.130)

In other words, the plaquette variables of the latti e gauge theory an be


treated in two dimensions as independent.
The orre t interpretation of Eq. (3.3.130) is that it is the partition fun tion of the one-plaquette model, i.e. the latti e gauge theory on a single
plaquette. This is onsistent with the gauge invarian e.
The one-matrix model (3.3.130) an be easily solved in the large-N limit
by loop equations.

289

QCD IN LOOP SPACE

Let us rst introdu e the proper `observables' for the one-matrix model:


1
n
Wn =
;
(3.3.131)
tr U
N
1p
where the average is taken with the same weight as in Eq. (3.3.130). The
interpretation of Wn 's in the language of the single-plaquette model is that
these are the Wilson loop averages for ontours whi h go along the boundary
of n sta ked plaquettes.
In order to derive the loop equation for the one-matrix model, we pro eed
quite analogous to the derivation of the loop equation in the latti e gauge
theory (given in Problem 3.22). Let us onsider the obvious identity
0 =

h tr ta U ni1p ;

(3.3.132)

and perform the (in nitesimal) hange

! U (1 ita a ) ; U y ! (1 + ita a ) U y

(3.3.133)

of the integration variable on the RHS of Eq. (3.3.132). Sin e the Haar
measure is invariant under the hange (3.3.133), we nally get


(W
2N 2 n

(3.3.128)

Hen e, the partition fun tion (3.3.127) fa torizes:

Z2d = (Z1p)Np ;

3.3.

where

Wn+1 ) =

n
X
k=1

Wk Wn k

for n  1 ;

W0 = 1 ;

(3.3.134)


1
=
N 2


(3.3.135)

 1;

(3.3.137)

and   1 as N ! 1.
Equation (3.3.134) has the following exa t solution
1
; Wn = 0 for n  2
(3.3.136)
W1 =
2
whi h reprodu es the strong oupling expansion. The leading order of the
strong oupling expansion turns out to be exa t at N = 1.
However, the solution (3.3.136) an not be the desired solution at any
values of the oupling onstant. Sin e Wk are (normalized) averages of unitary
matri es, they must obey

Wn

290

CHAPTER 3.

1=N

EXPANSION

3.3.

291

QCD IN LOOP SPACE

whi h is not the ase for W1 , given by Eq. (3.3.136), at small enough values
of .
In order to nd all solutions to Eq. (3.3.134), let us introdu e the generating fun tion
1
X
f (z ) 
Wn z n
(3.3.138)

single-plaquette model (3.3.130) in the large-N limit. In other words, the


number of degrees of freedom is now in nite due to the internal symmetry
group rather than an in nite volume.
Finally, we mention that sin e plaquette veriables are independent in latti e QCD2 , the Wilson loop average for a non-interse ting latti e ontour C
takes the form

and rewrite Eq. (3.3.134) as the quadrati equation

where A is the area (in the latti e units) en losed by the ontour C . W1 in
this formula is given by Eq. (3.3.136) in the strong oupling phase (  1)
and Eq. (3.3.141) in the weak oupling phase (  1).
The ontinuum formula (3.3.104) is re overed for small  from
Eq. (3.3.142) as follows:

WC = ( W 1 ) A

n=0

1
(f 1) + W1 = 2 f 2
z
A formal solution to Eq. (3.3.139) is

fz

f :

(3.3.139)

(1 + 2z + z 2 )2 + 4z 2 (2W1 1)
1 2z
f (z ) =
+
; (3.3.140)
4z
4z
where the positive sign of the square root is hosen to satisfy f (0) = 1.
The RHS of Eq. (3.3.140) depends on an unknown fun tion W1 () whi h
must guarantee for f (z ) to be a holomor fun tion of the omplex variable z
inside the unit ir le jz j < 1. This is a onsequen e of the inequality (3.3.137)
whi h stems from the unitarity of U 's.
There exist two solutions for whi h f (z ) is holomor inside the unit ir le:
the strong oupling solution given for   1 by Eq. (3.3.136) and the weak
oupling solution given for   1 by

z2


:
(3.3.141)
2
A omparison with Eq. (2.3.1) for d = 2 shows that the leading order of
the weak oupling expansion is now exa t. Therefore, f (z ) is given by two
di erent analiti fun tions for  > 1 and  < 1.
At the point  = 1, a phase transition o urs as was dis overed by Gross
and Witten [GW80 who rst solved latti e QCD2 in the large-N limit. This
phase transition is of the third order sin e both the rst and se ond derivatives
of the partition fun tion are ontinuous at  = 1. The dis ontinuity resides
only in the third derivative. This phase transition is pretty unusual from the
point of view of statisti al me hani s where phase transitions usually o ur
in the limit of an in nite volume (otherwise the partition fun tion is analyti
in temperature). Now the Gross{Witten phase transition o urs even for the
W1 = 1

(3.3.142)

a2 A=a a!0


(3.3.143)
! e 2 A ;
WC = 1
2
where we have restored the a-dependen e as is pres ribed by the dimensional
analysis.
The solution of N = 1 latti e QCD2 by the loop equations, whi h is
des ribed in this Subse tion, was given in Refs. [PR80, Fri81.
Cal ulate the density of eigenvalues for the matrix U in the onematrix model (3.3.130).
Solution Let us redu e U to the diagonal form

Problem 3.28

U = diag e i 1 ; : : : ; e i j ; : : : ; e i N :
(3.3.144)
The density of eigenvalues (or the spe tral density), ( ), is then de ned as a
fra tion of the eigenvalues whi h lie in the interval [ ; + d . In other words,
introdu ing the ontinuum variable x = j=N (0  x  1) in the large-N limit, we
have
 ( ) = dx  0
(3.3.145)
d
whi h obey the obvious normalization
Z 

d  ( ) =

Z 1

dx = 1 :

(3.3.146)

d  ( ) os n :

(3.3.147)

Given  ( ), we an al ulate Wn by

Wn =

Z 

292

CHAPTER 3.

1=N

EXPANSION

It is now lear from the de nition (3.3.131), (3.3.138) that


Z 

f (z ) =

d  ( )

i
2

1
:
1 z e i

(3.3.148)

Z 

r

os +  +

1 2W1



os + 

Cal ulate the density of eigenvalues for the Gaussian Hermitean


one-matrix model whi h is de ned by the partition fun tion

Z1h =

Z Y

ij

2
M
dij e N 2 tr  :

(3.3.155)

The di eren e from the previous problem is that ij is now a Hermitean
N  N matrix whose eigenvalues pi an take on values along the whole real axis.
Let us de ne
Z
D
E
1
n
Wn 
tr 
=
d  ( )  n :
(3.3.156)
N
1h
The S hwinger{Dyson equations for the Hermitean one-matrix model an be derived [Wad81 using the invarian e of the measure in (3.3.155) under the hange
ij ! ij + ij
(3.3.157)
where ij is (in nitesimal) Hermitean.
Pro eeding as before and using the large-N fa torization, we get the set of
equations
Solution

d  ( ) ot ! :
2


(3.3.149)

The dis ontinuity of this analyti fun tion at ! =  i0 then determines  ( ).


Using the expli it solution (3.3.140), we formally obtain

 ( ) = 1
2

293

QCD IN LOOP SPACE

Problem 3.29

Choosing z = exp (i! ), we rewrite Eq. (3.3.148) as



1
e i! = +

3.3.

1 2W1 : (3.3.150)

For W1 given by Eqs. (3.3.136) and (3.3.141) for the strong and weak oupling
phases, we get nally

1
1 
1 + os
for   1
(3.3.151)
 ( ) =
2

and
1
q sin2
 ( ) =
os
for   1
(3.3.152)

2
2
for the strong and weak oupling solutions, respe tively. Noti e, that (3.3.151) is
non-negative for   1 as it should due to the inequality (3.3.145). For  < 1,
the strong oupling solution (3.3.151) be omes negative somewhere in the interval
[ ;  whi h an not happen for a dynami al system. This is the reason why the
other solution (3.3.152) is realized for  < 1. It has the support on the smaller
interval [ ; where 0 < <  is determined by the equation

sin2 = 
(3.3.153)
2
whi h always has a solution for  < 1. The weak oupling spe tral density (3.3.152)
is non-negative for   p1.
For small , = 2  so that
p
 ( ) = 1 4 2 :
(3.3.154)
2
As  ! 0,  ( ) ! ( ) and U freezes, modulo a gauge transformation, near a unit
matrix. This guarantees the existen e of the ontinuum limit of QCD2 .
The spe tral densities (3.3.151) and (3.3.152) were rst al ulated in
Ref. [GW80 by a dire t solution of the saddle-point equation at large N .

M Wn+1 =

n 1
X
k=0

for n  0 ;

Wk Wn k

W0 = 1 :

(3.3.158)

Introdu ing the generating fun tion

W (p)

1
1
tr
N p 


1h

1
X

Wn ;
n+1
p
n=0

(3.3.159)

we rewrite Eq. (3.3.158) as the quadrati equation

Mp W (p) M = W 2 (p)
whose solution reads

(3.3.160)

Mp
M 2 p2 M ;
W (p) =
(3.3.161)
2
4
where we have hosen the minus sign of the root to satisfy the asymptoti s W (p) !
1=p for large p as is pres ribed by the de nition (3.3.159).
W (pp
) given by Eq. p
(3.3.161) is an analyti fun tion of p with the ut from
p = 2= M to p = +2= M along the real axis. The dis ontinuity of W (p) at the
ut determines the spe tral density:
W (  i0) = M  i ( ) :
(3.3.162)
2

294

CHAPTER 3.

We have

 ( ) = M
2

1=N

EXPANSION

3.4.

NC

LARGE-

295

REDUCTION

3.4 Large-N redu tion


4
M

2 :

(3.3.163)

Noti e
p that the
p spe tral density is non-negative and has support on a nite interval
[ 2= M; 2= M in analogy with the unitary one-matrix model it the weak oupling
regime. The spe tral density (3.3.163) was rst al ulated by Wigner [Wig51 and
is alled Wigner's semi ir le law.
We have already semi ir le law in the previous Problem for the spe tral density
of the unitary one-matrix model at small  (see Eq. (3.3.154)). This is be ause we
an always substitute U = exp (i) where U is unitary and  is Hermitean and
expand for small  in  up to the quadrati term. We then obtain the Hermitean
model (3.3.155) with M = 1= from the unitary model (3.3.130).

The large-N redu tion was rst dis overed by Egu hi and Kawai [EK82
who showed that the Wilson latti e gauge theory on a d-dimensional hyper ubi latti e is equivalent at N = 1 to the one on a hyper ube with periodi boundary onditions. This onstru tion is based on an extra (ZN )d symmetry whi h the redu ed theory possesses to ea h order of the strong
oupling expansion.
Soon after it was re ognized that a phase transition o urs in the redu ed model with de reasing the oupling onstant, so that this symmetry
is broken in the weak oupling regime. To ure the onstru tion at weak
oupling, the quen hing pres ription was proposed by Bhanot, Heller and
Neuberger [BHN82 and elaborated by many authors. An elegant alternative redu tion pro edure based on twisting pres ription was advo ated by
Gonzalez-Arroyo and Okawa [GAO83. Ea h of these pres riptions results in
the redu ed model whi h is fully equivalent to multi olor QCD, both on the
latti e and in the ontinuum.
While the redu ed models look as a great simpli ation, sin e the spa etime is redu ed to a point, they still involve an integration over d in nite
matri es whi h is in fa t a ontinual path integral. It is not lear at the
moment whether or not this is a real simpli ation of the original theory
whi h an make it solvable. Nevertheless, the redu ed models are useful and
elegant representations of the original theory at large N .
We shall start this Se tion by a simplest example of a pure matrix s alar
theory. The quen hed redu ed model for this ase was proposed by Parisi
[Par82 on the latti e end elaborated by Gross and Kitazawa [GK82 in the
ontinuum, while the twisted redu ed model was advo ated by Egu hi and
Nakayama [EN83. Then we on entrate on the Egu hi{Kawai redu tion of
multi olor QCD both on the latti e and in the ontinuum.

3.4.1 Redu tion of s alar eld


Let us begin with a simplest example of a pure matrix s alar theory on a
latti e whose partition fun tion is de ned by the path integral

Z=

Z YY

x ij

d'ijx e

P
x N tr

V ['x +


P
 'x 'x+a^

(3.4.1)

296

CHAPTER 3.

1=N

EXPANSION

Here 'x is a N  N Hermitean matrix eld and V [' is some intera tion
potential, say
M 2 3 3 4 4
V [' =
' + ' + ' :
(3.4.2)
2
3
4
The pres ription of the large-N redu tion is formulated as follows. We
substitute
'x ! Sx S y ;
(3.4.3)
x

where

[Sxkj = e ipk x kj




= diag e ip1 x ; : : : ; e ipN x

F ['x

 Z1

Z Y

d'x e

P
x N tr


P
 '(x)'(x+a^) F ['x ;

V ['x +

(3.4.5)

F ['x

Z  Y
N
a d Y
N
d

a

a =1 i=1

F ['x

2

x Sx

Redu ed

ij kl

(3.4.6)

where the average on the RHS is al ulated [Par82 for the quen hed redu ed
model whose averages are de ned by
D
E
1
F [

Redu ed ZRedu ed
Z Y
P
 
2P
dij e N tr V [+N ij jij j  os ((pi pj )a) F [ : (3.4.7)

ij
The partition fun tion of the redu ed model reads
Z Y
P
 
2P
dij e N tr V [+N ij jij j  os ((pi pj )a) (3.4.8)
ZRedu ed =
ij

! F [Sx s Sxy ;

(3.4.9)

where the momenta pi are uniformly distributed in the hyper ube. Therefore,
this saddle point on guration plays the role of a master eld in the sense of
Subse t. 3.2.7.
In order to show how Eq. (3.4.6) works, let us demonstrate how the planar diagrams of perturbation theory for the s alar matrix theory (3.4.1) are
re overed in the quen hed redu ed model.
The quen hed redu ed model (3.4.8) is of the general type dis ussed in
Se t. 3.2. The propagator is given by
D

dpi D
F [S

297

REDUCTION

whi h an be dedu ed, modulo the volume fa tor, from the partition fun tion
(3.4.1) by the substitution (3.4.3).
Noti e that the integration over the momenta pi on the RHS of Eq. (3.4.6)
is taken after the al ulation of averages in the redu ed model. Su h variables are usually alled quen hed in statisti al me hani s whi h lari es the
terminology.
Sin e N ! 1 it is not ne essary to integrate over the quen hed momenta
in Eq. (3.4.6). The integral should be re overed if pi 's would be uniformly
distributed in a d-dimensional hyper ube. Moreover, a similar property holds
for the matrix integral over  as well, whi h an be substituted by its value
at the saddle point on guration s :

an be al ulated at N = 1 by
D

NC

LARGE-

(3.4.4)

is a diagonal unitary matrix whi h eats the oordinate dependen e, so that


 does not depend on x.
The averaging of a fun tional F ['x whi h is de ned with the same weight
as in Eq. (3.4.1),
D

3.4.

Gauss

1
G (pi pj ) il kj
N

(3.4.10)

with

G (pi pj ) =

1
os
(pi


pj )a

(3.4.11)

It is onvenient to asso iate the momenta pi and pj in Eq. (3.4.11) with


ea h of the two index lines representing the propagator and arrying, respe tively, indi es i and j . Remember, that these lines are oriented for a
Hermitean matrix  and their orientation an be naturally asso iated with
the dire tion of the ow of the momentum. The total momentum arried by
the double line is pi pj .
The simplest diagram whi h represents the orre tion of the se ond order
in 3 to the propagator is depi ted Fig. 3.36. The momenta pi and pj ows
along the index lines i and j while the momentum pk ir ulates along the

298

CHAPTER 3.

1=N

EXPANSION

...........
..........

..........
.................. .............
...... ............................. ...........
.
.
.
.
.... ..
... ......
k ......... .......
... ...
.. ..
.... ....
..
.
... .....
... ...
.
.
... ...
. .
... ..
... .....
.... ....... k .......... ......
...... ............................ ......
........
...................................

...........
...........

Fig. 3.36: Simplest planar diagram of the se ond order in 3 for the propagator in

the quen hed redu ed model (3.4.8). The momentum pi ows along the
index line i. The momentum pi pj is asso iated with the double line ij .

index line k. The ontribution of the diagram in Fig. 3.36 reads

23
2X
G
(
p
p
)
G ( p i pk ) G ( p k
i
j
N 2
k

pj ) ;

(3.4.12)

where the summation over the index k is just a standard one over indi es
forming a losed loop.
In order to show that the quen hed-model result (3.4.12) reprodu es the
orre tion to the propagator in the original theory on an in nite latti e, we
pass to the variables of the total momenta owing along the double lines:

pi pj = p; pk

pj = q ; p i pk = p q ;

(3.4.13)

whi h is obviously onsistent with the momentum onservation at ea h of the


two verti es of the diagram in Fig. 3.36. Sin e pk 's are uniformly distributed
in the hyper ube, the summation over k an be substituted as N ! 1 by
the integral
1 X
f (pk ) ) ad
N k

Z 
a

a

dd q
f (q) :
(2)d

(3.4.14)

The pres ription (3.4.6) then gives the orre t expression


Z
23
2
d
a G (p)
N


a

a

dd q
G (q) G (p q)
(2)d

(3.4.15)

3.4.

NC

LARGE-

REDUCTION

299

for the se ond-order ontribution of the perturbation theory for the propagator on the latti e.
It is now lear how a generi planar diagram is re overed by the redu ed
model. We rst represent the diagram by the double lines and asso iate
the momentum pi with an index line arrying the index i. Then we write
down the expression for the diagram in the redu ed model with the propagator (3.4.11). Passing to the momenta owing along the double lines, similar
to Eq. (3.4.13), we get an expression whi h oin ides with the integrand of the
Feynman diagram for the theory on the whole latti e. It is ru ial that su h
a hange of variables an always be done for a planar diagram onsistently
with the momentum onservation at ea h vertex. The last step is that the
summation over indi es of losed index lines reprodu es the integration over
momenta asso iated with ea h of the loops a ording to Eq. (3.4.14). It is
assumed that the number of loops is mu h less than N whi h is always true
for a given diagram sin e N is in nite.
We thus have shown how planar diagrams of the latti e theory de ned by
the partition fun tion (3.4.1) are re overed by the redu ed model (3.4.8). The
latti e was needed only as a regularization to make all integrals well-de ned
and was not ru ial in the onsideration. This onstru tion an be formulated
dire tly for the ontinuum theory [GK82, DW82 where the propagator turns
into
1
G ( p i pj ) =
(3.4.16)
( pi p j ) 2 + m 2
and a Lorentz-invariant regularization an be a hieved by hoosing p2 < 2 .

Remark on the twisted redu ed model


An alternative redu tion pro edure is based on the twisting pres ription
[GAO83. We again perform the unitary transformation (3.4.3) with the
matri es Sx being expressed via a set of d (unitary) N  N matri es  by

Sx = x1 1=a x2 2 =a x3 3 =a x4 4 =a

(3.4.17)

where the oordinates of the (latti e) ve tor x are measured in the latti e
units. The matri es  are expli itly onstru ted in Ref. [GAO83 and ommute by
(3.4.18)
  = Z  
y being elements of ZN .
with Z = Z

300

CHAPTER 3.

1=N

EXPANSION

For the twisting redu tion pres ription, Eq. (3.4.6) is valid providing the
average on the RHS is al ulated for the twisted redu ed model whi h is de ned
by the partition fun tion [EN83

ZTRM =

d e

N tr V [+N

y
 tr     :

(3.4.19)

Now the perturbation theory for the unredu ed model (3.4.1) is re overed due
to the expli it form of Sx given by Eq. (3.4.17).
We an hange the order of 's in Eq. (3.4.17) de ning a more general
path-dependent fa tor
Y
Sx = P
(3.4.20)
:
l2Cx1
The path-ordered produ t in this formula runs over all links l = (z; ) forming
a path Cx1 from in nity to the point x.
Due to Eq. (3.4.18), hanging the form of the path multiplies Sx by the
Abelian fa tor
Y
Z (C ) =
Z (2)
(3.4.21)
22S :S =C
where (;  ) is the orientation of the plaquette 2. The produ t runs over any
surfa e spanned by the losed loop C whi h is obtained by passing the original
path forward and the new path ba kward. Due to the Bian hi identity
Y

22 ube

Z (2) = 1

(3.4.22)

where the produ t goes over six plaquettes forming a 3-dimensional ube on
the latti e, the produ t on the RHS of Eq. (3.4.21) does not depend on the
form of the surfa e S and is a fun tional of the loop C .
It is now easy to see that under this hange of the path we get
[Sx ij [S y kl ! jZ (C )j2 [Sx ij [S y kl
(3.4.23)
x

x
jZ (C )j2

and the path-dependen e is an eled be ause


= 1. This is a general
property whi h holds for the twisting redu tion pres ription of any even (i.e.
invariant under the enter ZN ) representation of SU (N ).

3.4.

NC

LARGE-

redu ed model whi h is nothing but its redu tion to a point. The a tion of
the redu ed model is given by
1

SEK =

2g2d

tr [A ; A 2 ;

(3.4.24)

where A are d spa e-independent matri es and  is a dimensionful parameter.


A naive statement of the Egu hi{Kawai redu tion is that the averages
oin ide in both theories, for example,


H 
1
tr P e i d A ()
N

d dim

H 
1
tr P e i d A
N

(3.4.25)

EK

where the LHS is al ulated with the a tion (2.1.14) and the RHS is al ulated
with the redu ed a tion (3.4.24). Stri tly speaking, this naive statement is
valid only in d = 2 or supersymmetri ase for the reason whi h will be
explained in a moment.
The pre ise equivalen e is valid only if the average of open Wilson loops
vanish in the redu ed model:
R


i C d A
1
yx
= 0;
(3.4.26)
tr P e
N
EK
as it does in the d-dimensional theory due to the lo al gauge invarian e under
whi h


d A ()
i
P e Cyx

ij

y (y)P e



Cyx d A ()
(x)

ij

(3.4.27)

The point is that this gauge invarian e transforms in the redu ed model into
(global) rotation of the redu ed eld by onstant matri es
:
A !
y A
:
(3.4.28)
whi h does not guarantee su h vanishing in the redu ed model.
There exists, however, a symmetry of the redu ed a tion (3.4.24) under
the shift of A by a unit matrix15:

3.4.2 Redu tion of Yang{Mills eld


The statement of the Egu hi{Kawai redu tion of the Yang{Mills eld says
that the theory on a d-dimensional spa e-time is equivalent at N = 1 to the

301

REDUCTION

Aij

! Aij + a ij ;

(3.4.29)

15 This symmetry is rigorously de ned on a latti e where it is asso iated with a dire tion-

dependent ZN transformation.

302

1=N

CHAPTER 3.

EXPANSION

whi h is often alled the Rd -symmetry. Under the transformation (3.4.29),


we get


Pe

i C d A
yx

ij

! e i(y

x )a

Pe

i C d A
yx

(3.4.30)

ij

whi h guarantees, if the symmetry is not broken, the vanishing of the open
Wilson loops

WEK (Cyx ) 

i
1
tr P e
N

R
 
Cyx d A

EK

=0

(3.4.31)

in the redu ed model.


The equivalen e of the two theories an then be shown using the loop
equation whi h reads for the redu ed model

x


1
d A
i
tr P[A ; [A ; A e Cxx
WEK (C ) =
N
 (x)


H
 i Cxx d A
1
d
tr P
e
= 
N
A
EK
= d

dy WEK (Cyx ) WEK (Cxy ) :

EK

(3.4.32)

The RHS is pretty mu h similar to the one in Eq. (3.3.56) while (d)(x y)
is missing.
This delta fun tion an be re overed if the Rd symmetry is not broken
sin e
(d)(x y)
WEK (Cyx )
(3.4.33)
WEK (Cyx )  (d)
(0)
due to Eq. (3.4.31) for the open loops.
This is not a rigorous argument sin e a regularization is needed. What
a tually happens is the following. If we smear the delta fun tion introdu ing

(d) (x) =
then

p

2

d

2 2
e x  =2 ;

2 2
1  (d) 2
 (0) / d e x 
(d)
 (0)
reprodu ing the delta fun tion.

! (d) (x) ;

LARGE-

303

REDUCTION

3.4.3 Rd -symmetry in perturbation theory

Sin e N is in nite, the Rd -symmetry an be broken spontaneously. The


point is that the large-N limit plays the role of a statisti al averaging, as
is mentioned already in Subse tion 3.2.8, and phase transitions are possible for in nite number of degrees of freedom. This phenomenon o urs in
perturbation theory of the redu ed model for d  3.
The perturbation theory an be onstru ted expanding the elds around
solutions of the lassi al equation
[A ; [A ; A = 0 :
Any diagonal matrix

(3.4.36)

(N )
A l  p = diag p(1)
 ; : : : ; p

(3.4.37)

is a solution to Eq. (3.4.36).


The perturbation theory of the redu ed model an be onstru ted expanding around the lassi al solution (3.4.37):

A = A l + gAq ;

NC

3.4.

(3.4.34)
(3.4.35)

(3.4.38)

where Aq is o -diagonal.


Substituting (3.4.38) into the a tion (3.4.24), we get


1
1
[p ; Aq 2 + higher orders :
SEK = tr [p ; Aq 2
2
2
To x the gauge symmetry (3.4.28), it is onvenient to add


1
q
2
Sg:f : = tr [p ; A + [p ; b[p ; ;
2
where b and are ghosts.
The sum of (3.4.39) and (3.4.40) gives

S2 = tr

1
[p ; Aq 2 + [p ; b[p ;
2  

(3.4.39)

(3.4.40)

(3.4.41)

up to quadrati order in Aq .


Doing the Gaussian integral over Aq , we get at the one-loop level:
Z

dp dAq e S2 : : : =

Z Y
N

k=1

dp(k)

Yh

i<j

(p(i)

p(j) )2

i1 d=2

::: ;

(3.4.42)

304

CHAPTER 3.

1=N

EXPANSION

where the integration over p a ounts for equivalent lassi al solutions.


For d = 1 the produ t on the RHS of Eq. (3.4.42) reprodu es the Vandermonde determinant. For d = 2 it vanishes and does not a e t dynami s. For
d  3 the measure is singular and the eigenvalues ollapse. This leads us to
a spontaneous breakdown of the Rd in perturbation theory.
The equivalen e between the N = 1 Yang{Mills theory on a whole spa e
and the redu ed model an be provided [BHN82 introdu ing a quen hing
pres ription similar to the one des ribed in Subse tion 3.4.1. Then no ollapse
of eigenvalues happens and d-dimensional planar graphs are reprodu ed by
the redu ed model. More about the quen hing pres ription in Yang{Mills
theory an be found in the reviews [Mig83, Das87 and ited there original
papers.

Remark on supersymmetri ase


In a supersymmetri gauge theory, there is an extra ontribution from
fermions to the exponent on the RHS of Eq. (3.4.42). Sin e the integration over fermions results in the extra fa tor [(p(i) p(j) )2 tr I=2 , this yields
nally the exponent 1 d=2 + tr I=2. It vanishes in d = 4 for either Majorana or Weyl fermions and in d = 10 for the Majorana{Weyl fermions.
Therefore, the Rd-symmetry is not broken and no quen hing is needed in the
supersymmetri ase [MK83, IKKT97.

3.4.4 Twisted redu ed model


The ontinuum version of the twisted redu ed model an be onstru ted [GAK83 by substituting A ! A  into the a tion (3.4.24),
where the matri es  obey the ommutation relation
[  ;  = B I ;
(3.4.43)
where B is an antisymmetri tensor and d is even. This is possible only for
in nite Hermitean matri es (operators). An example of su h matri es is x
and p operators in quantum me hani s. Eq. (3.4.43) is a ontinuum version
of Eq. (3.4.18).
The Wilson loop averages in the twisted redu ed model are de ned by
R
R


d  1
d A
i
i
1
WTEK (Cyx ) =
: (3.4.44)
tr P e Cyx
tr P e Cyx
N
N
TEK
They vanish for open loops whi h is provided by the vanishing of the tra e
of the path-ordered exponential of  in this de nition. For losed loops

3.4.

NC

LARGE-

REDUCTION

305

this fa tor does not vanish and is needed to provide the equivalen e with
d-dimensional Yang{Mills perturbation theory, sin e the lassi al extrema
of the twisted redu ed model are A l =  and the perturbation theory is
onstru ted expanding around this lassi al solution.
The proof of the equivalen e an be done using the loop equation quite
similarly to that of Subse t. 3.4.2 for the Egu hi{Kawai model with an unbroken Rd symmetry.

306

CHAPTER 3.

1=N

EXPANSION

Referen es to Chapter 3
[Ans59
[BGSN82
[BHN82
[BNS81
[BNZ79
[BP75
[Bra80
[BW93
[CJP74
[CMS93
[CLS82

A.A. Anselm, A model of a eld theory with nonvanishing renormalized harge, Zh. Exp. Teor. Fiz. 36 (1959) 863 (English transl.:
Sov. Phys. JETP 9 (1959) 608).
R.A. Brandt, A. Go ks h, M. Sato, and F. Neri, Loop spa e, Phys.
Rev. D26 (1982) 3611.
G. Bhanot, U. Heller and H. Neuberger, The quen hed Egu hi{
Kawai model, Phys. Lett. 113B (1982) 47.
R.A. Brandt, F. Neri and M. Sato, Renormalization of loop fun tions for all loops, Phys. Rev. D24 (1981) 879.
R.A. Brandt, F. Neri and D. Zwanziger, Lorentz invarian e from
lassi al parti le paths in quantum eld theory of ele tri and magneti harge, Phys. Rev. D19 (1979) 1153.
A.A. Belavin and A.M. Polyakov, Metastable states of twodimensional isotropi ferromagnet, Pis'ma v ZhETF 22 (1975)
503 (English transl.: JETP Lett. 22 (1975) 245).
N. Brali , Exa t omputation of loop averages in two-dimensional
Yang{Mills theory, Phys. Rev. D22 (1980) 3090.
E. Brezin and S. Wadia, The large N expansion in quantum eld
theory and statisti al physi s: from spin systems to 2-dimensional
gravity, World S i., 1993.
S. Coleman, R. Ja kiw and H.D. Politzer, Spontaneous symmetry
breaking in the O(N ) model at large N , Phys. Rev. D10 (1974)
2491.
W. Chen, Y. Makeenko, and G.W. Semeno , Four-fermion theory
and the onformal bootstrap, Ann. Phys. 228 (1993) 341.
P. Cvitanovi , P.G. Lauwers, and P.N. S harba h, The planar se tor of eld theories, Nu l. Phys. B203 (1982) 385.
307

308
[Col79

REFERENCES TO CHAPTER 3

S. Coleman, 1/N, in Pro . of Eri e Int. S hool of Subnu lear


Physi s 1979, Plenum, N.Y., 1982, p. 805 (Reprinted in S. Coleman, Aspe ts of symmetry, Cambridge Univ. Press, 1985, pp.351{
402).
[Cvi81 P. Cvitanovi , Planar perturbation expansion, Phys. Lett. 99B
(1981) 49.
[Das87 S.R. Das, Some aspe ts of large-N theories, Rev. Mod. Phys. 59
(1987) 235.
[Dou95 M.R. Douglas, Sto hasti master elds, Phys. Lett. B344 (1995)
117.
[DV80 V.S. Dotsenko and S.N. Vergeles, Renormalizability of phase fa tors in non-Abelian gauge theory, Nu l. Phys. B169 (1980) 527.
[DW82 S. Das and S. Wadia, Translation invarian e and a redu ed model
for summing planar diagrams in QCD, Phys. Lett. B117 (1982)
228.
[Egu79 T. Egu hi, Strings in U(N) latti e gauge theory, Phys. Lett. 87B
(1979) 91.
[EK82 T. Egu hi and H. Kawai, Redu tion of dynami al degrees of freedom in the large-N gauge theory, Phys. Rev. Lett. 48 (1982) 1063.
[EN83 T. Egu hi and R. Nakayama, Simpli ation of Quen hing Pro edure for Large N Spin Models, Phys. Lett. 122B (1983) 59.
[Fel86
M.N. Feller, In nite-dimensional ellipti equations and operators
of the type by P. Levy, Sov. J. Usp. Mat. Nauk. 41 (1986) 97.
[Foe79 D. Foerster, Yang{Mills theory | a string theory in disguise, Phys.
Lett. 87B (1979) 87.
[Fri81
D. Friedan, Some nonabelian toy models in the large N limit, Commun. Math. Phys. 78 (1981) 353.
[GAO83 A. Gonzalez-Arroyo and M. Okawa, The twisted Egu hi-Kawai
model: A redu ed model for large N latti e gauge theory, Phys.
Rev., D27 (1983) 2397.
[GAK83 A. Gonzalez-Arroyo and C. P. Korthals Altes, Redu ed model for
large N ontinuum eld theories, Phys. Lett. B131 (1983) 396.
[GG95 R. Gopakumar and G.J. Gross, Mastering the master eld, Nu l.
Phys. B451 (1995) 379.
[GK82 D.J. Gross and Y. Kitazawa, A quen hed momentum pres ription
for large-N theories, Nu l. Phys. B206 (1982) 440.

REFERENCES TO CHAPTER 3

[GN74
[GN79a
[GN79b
[GN80
[Gra91
[GW70
[GW80
[Ja 72

[Joh61
[Haa81
[HM89
[Hoo74a
[Hoo74b
[IKKT97
[KK80

309

D.J. Gross and A. Neveu, Dynami al symmetry breaking in asymptoti ally free eld theories, Phys. Rev. D10 (1974) 3235.
J.L. Gervais and A. Neveu, The quantum dual string wave fun tional in Yang{Mills theories, Phys. Lett. 80B (1979) 255.
J.L. Gervais and A. Neveu, Lo al harmoni ity of the Wilson loop
integral in lassi al Yang{Mills theory, Nu l. Phys. B153 (1979)
445.
J.L. Gervais and A. Neveu, The slope of the leading Regge traje tory in quantum hromodynami s, Nu l. Phys. B163 (1980) 189.
J.A. Gra ey, Cal ulation of exponent  to O(1=N 2 ) in the O(N )
Gross Neveu model, Int. J. Mod. Phys. A6 (1991) 395, 2755(E).
D. Gross and J. Wess, S ale invarian e, onformal invarian e, and
the high-energy behavior of s attering amplitudes, Phys. Rev. D2
(1970) 753.
D. Gross and E. Witten, Possible third-order phase transition in
the large-N latti e gauge theory, Phys. Rev. D21 (1980) 446.
R. Ja kiw, Field theoreti investigations in urrent algebra, in Le tures on Current Algebra and its Appli ations by S.B. Treiman,
R. Ja kiw, and D.J. Gross, Prin eton Univ. Press, 1972, p. 97; also
in Current Algebra and Anomalies by S.B. Treiman, R. Ja kiw,
B. Zumino, and E. Witten, World S i., 1985, p. 81.
K. Johnson, Solution of the equations for the Green's fun tions of
a two dimensional relativisti eld theory, Nuovo Cim. 20 (1961)
773.
O. Haan, Large N as a thermodynami limit, Phys. Lett. 106B
(1981) 207.
M.B. Halpern and Yu.M. Makeenko, Continuum-regularized loopspa e equation, Phys. Lett. 218B (1989) 230.
G. 't Hooft, A planar diagram theory for strong intera tions, Nu l.
Phys. B72 (1974) 461.
G. 't Hooft, A two-dimensional model for mesons, Nu l. Phys.
B75 (1974) 461.
N. Ishibashi, H. Kawai, Y. Kitazawa and A. Tsu hiya, A large-N
redu ed model as superstring, Nu l. Phys. B498 (1997) 467.
V.A. Kazakov and I.K. Kostov, Non-linear strings in twodimensional U(1) theory, Nu l. Phys. B176 (1980) 199.

310
[KK81

REFERENCES TO CHAPTER 3

V.A. Kazakov and I.K. Kostov, Computation of the Wilson loop


fun tional in two-dimensional U (1) latti e gauge theory, Phys.
Lett. 105B (1981) 453.
[KNN77 J. Koplik, A. Neveu and S. Nussinov, Some aspe ts of the planar
perturbation series, Nu l. Phys. B123 (1977) 109.
[Mak83 Yu.M. Makeenko, Large N , in Gauge Theories of the Eighties, eds.
R. Raitio and J. Lindfors, Le ture Notes in Physi s 181, Springer{
Verlag, 1983, p. 67.
[Mak88 Yu.M. Makeenko, Polygon dis retization of the loop-spa e equation, Phys. Lett. 212B (1988) 221.
[Mak94 Y. Makeenko, Exa t multiparti le amplitudes at threshold in largeN omponent 4 theory, Phys. Rev. 50 (1994) 4137.
[Man79 S. Mandelstam, Charge-monopole duality and the phases of nonAbelian gauge theories, Phys. Rev. D19 (1979) 2391.
[Mig71 A.A. Migdal, On hadroni intera tions at small distan es, Phys.
Lett. 37B (1971) 98.
[Mig81 A.A. Migdal, QCD = Fermi string theory, Nu l. Phys. 189 (1981)
253.
[Mig83 A.A. Migdal, Loop equations and 1/N expansion, Phys. Rep. 102
(1983) 199.
[MK83 R.L. Mkrt hyan and S.B. Khokhla hev, Redu tion of the U(1)
theory to a model of random matri es, JETP Lett. 37 (1983) 160.
[MM79 Yu.M. Makeenko and A.A. Migdal, Exa t equation for the loop
average in multi olor QCD, Phys. Lett. 88B (1979) 135.
[MM80 Yu.M. Makeenko and A.A. Migdal, Self- onsistent area law in
QCD, Phys. Lett. 97B (1980) 235.
[MM81 Yu.M. Makeenko and A.A. Migdal, Quantum hromodynami s as
dynami s of loops, Nu l. Phys. B188 (1981) 269 (Russian version:
Yad. Fiz. 32 (1980) 838).
[MS69 G. Ma k and A. Salam, Finite- omponent eld representation of
the onformal group, Ann. Phys. 53 (1969) 144.
[Nam79 Y. Nambu, QCD and the string model, Phys. Lett. 80B (1979)
372.
[OP81 P. Olesen and J.L. Petersen, The Makeenko{Migdal equation in a
domained QCD va uum, Nu l. Phys. 181 (1981) 157.
[Par75 G. Parisi, The theory of non-renormalizable intera tion, Nu l.
Phys. B100 (1975) 368.

REFERENCES TO CHAPTER 3

[Par82

311

G. Parisi, A simple expression for planar eld theories, Phys. Lett.


112B (1982) 463.
[Pol70 A.M. Polyakov, Conformal symmetry of riti al u tuations,
Pis'ma v ZhETF 12 (1970) 381) (English transl.: JETP Lett.
12 (1970) 381).
[Pol75 A.M. Polyakov, Intera tion of Goldstone parti les in two dimensions. Appli ations to ferromagnets and massive Yang{Mills elds,
Phys. Lett. 59B (1975) 79.
[Pol79 A.M. Polyakov, String representations and hidden symmetries for
gauge elds, Phys. Lett. 82B (1979) 247.
[Pol80 A.M. Polyakov, Gauge elds as rings of glue, Nu l. Phys. B164
(1980) 171.
[PR80 G. Pa uti and P. Rossi, A solution of Wilson's loop equation in
latti e QCD2 , Phys. Lett. 92B (1980) 321.
[S h74 H.J. S hnitzer, Nonperturbative e e tive potential for '4 theory
in the many-parti le limit, Phys. Rev. D10 (1974) 1800.
[Sta68 H.E. Stanley, Spheri al model as the limit of in nite spin dimensionality, Phys. Rev. 176 (1968) 718.
[Str57
R.L. Stratonovi h, DAN SSSR 115 (1957) 1097.
[Tav93 J.N. Tavares, Chen integrals, generalized loops and loop al ulus,
preprint DF/IST 5.93 (May, 1993), hep-th/9305173.
[Tut62 W.T. Tuttle, A ensus of planar triangulations, Can. J. Math. 14
(1962) 21.
[VDN95 D.V. Voi ules u, K.J. Dykema and A. Ni a, Free Random Variables, AMS, Providen e 1992.
[Ven76 G. Veneziano, Some aspe ts of a uni ed approa h to gauge, dual
and Gribov theories, Nu l. Phys. B117 (1976) 519.
[Wad81 S.R. Wadia, Dyson{S hwinger equations approa h to the largeN limit: Model systems and string representation of Yang{Mills
theory, Phys. Rev. D24 (1981) 970.
[Wig51 E.P. Wigner, Ann. Math. 53, 36 (1951); Pro . Cambridge Philos.
So . 47, 790 (1951).
[Wil73 K.G. Wilson, Quantum eld-theory models in less than 4 dimensions, Phys. Rev. D7 (1973) 2911.
[Wit79 E. Witten, The 1/N expansion in atomi and parti le physi s, in
Re ent developments in gauge theories, eds. G. 't Hooft et al.,
Plenum, 1980, p.403.