Quantum Anomaly

INTERNATIONAL SERIES OF MONOGRAPHS ON PHYSICS No.
122 (ISMP)
Series Editors J. Birman City University of New York

S. F. Edwards University of Cambridge
R. Friend University of Cambridge
M. Rees University of Cambridge
D. Sherrington University of Oxford
G. Veneziano CERN, Geneva
INTERNATIONAL SERIES OF MONOGRAPHS ON PHYSICS
124. C. Kiefer: Quantum gravity

123. T. Fujimoto: Plasma, spectroscopy
122. K. Fujikawa, H. Suzuki: Path integrals and quantum anomaiies
121. T. Giamarchi: Quantum physics in one dimension
120. M. Warner, E. Tercntjev: Liquid crystal elastomers
119. L. Jacak, P. Sitko, K. Wieczorek, A. Wojs: Quantum Hall systems
118. J. Wesson: Tokamaks, Third edition
117. G. Volovik: The Universe in a helium droplet
116. L. Pitaevskii. S. Stringari: Bose-Einstein condensation
115. G. Dissertori, I.G. Knowles, M. Schmelling: Quantum chromodynamics
114. B. DeWitt; The global approach to quantum field theory
113. J. Zinn-Justin: Quantum field theory and critical phenomena, Fourth edition
112. R.M. Mazo: Brownian motion - fluctuations, dynamics, and applications
111. H. Nishimori: Statistical physics of spin glasses and information processing - an introduction
110. N.B. Kopnin: Theory of nonequilibrium superconductivity
109. A. Aharoni: Introduction to the theory of ferromagnelism, Second edition
108. H. Dobbs: Helium three
107. R. Wigmans: Calorimetry
106. J. KÜbler: Theory of itinerant electron magnetism
105. Y. Kuramolo, Y. Kitaoka: Dynamics of heavy electrons
104. D. Bardin, G. Passarino: The Standard Model in the making
103. G.C. Branco, L. Lavoura, J.P. Silva: OP Violation
102. T.C. Choy: Effective medium theory
101. H. Araki: Mathematical theory of quantum fields
100. L. M. Pismen: Vortices in nonlinear fields
99. L. Mestel: Stellar magnetism
98. K. H, Bennernann: Nonlinear optics in metals
97. D. Salzmann: Atomic physics in hot plasmas
96. M. Brambilla: Kinetic theory of plasma waves
95. M. Wakatani: Stellarat.or and heliotron devices
91. S. Chikazumi: Physics of ferramagneiisrn
91. R. A. Bertlmann: Anomalies in quantum field theory
90. P. K. Gosh: Ion traps
89. E. Simanek: Inhomoaeneous superconductors
88. S. L. Adler: Quaternionic quantum mechanics and quantum, fields
87. P. S. Joshi: Global aspects in gravitation and cosmology
86. E. R. Pike, S. Sarkar: The. quantum theory of radiation
84. V. Z. Kresin, II. Morawitz, S. A. Wolf: Mechanisms of conventional and high Tc
super-conductivity
83. P. G. de Gennes, J. Prost: The physics of liquid crystals
82. B. H. Bransden. M. R. C. McDowell: Charge exchange and the theory of ion-atom collision
81. J. Jensen, A. R. Mackintosh: Rare earth magnetism
80. R. Gastmaiis. T. T. Wu: The ubiquitous photon
79. P. Luchini, H. Motz: Undulators and free-electron lasers
78. P. Weinberger: Electron scattering theory
76. H. Aoki, H. Kamimura: The physics of interacting electrons in disordered systems
75. .J. D. Lawson: The physics of charged particle beams
73. M. Doi, S. F. Edwards: The theory of polymer dynamics
71. R. L. Wolf: Principles of electron tunneling spectroscopy
70. H. K. Henisch: Semiconductor contacts
69. S. Chandrasekhar: The mathematical theory of black holes
68. G. R. Satchler: Direct nuclear reactions
51. C. Mo11er: The theory of relativity
46. H. E. Stanley: Introduction to phase transitions and critical phenomena
32. A. Abragam: Principles of nuclear magnetism
27. P. A. M. Dirac: Principles of quantum mechanics
23. R. E. Pcierls: Quantum theory of solids
Path Integrals and
Quantum Anomalies
KAZUO FUJIKAWA
Department of Physics, University of Tokyo
and
HIROSHI SUZUKI
Department of Mathematical Sciences, Ibaraki University
CLARENDON PRESS • OXFORD

2004
OXFORD
UNIVERSITY PRESS
Great Clarendon Street, Oxford OX2 6DP
Oxford University Press is a department of the University of Oxford.
It furthers the University's objective of excellence in research, scholarship,
and education by publishing worldwide in
Oxford New York
Auckland Bangkok Buenos Aires Cape Town Chermai
Dar es Salaam Delhi Hong Kong Istanbul Karachi Kolkata
Kuala Lumpur Madrid Melbourne Mexico City Mumbai Nairobi
Sao Paulo Shanghai Taipei Tokyo Toronto
Oxford is a registered trade mark of Oxford University Press
in the UK and in certain other countries
Published in the United States
by Oxford University Press Inc., New York
© Oxford University Press, 2004
The moral rights of the author have been asserted
Database right Oxford University Press (maker)
First published 2004
KE1RO SEKIBUN TO TAISIIOSEI NO RYOSHI-TEKI YABURE
(Path Integrals and Quantum Anomalies)
by Kazuo Fujikawa
© 2001 by Kazuo Fujikawa
Originally published in Japanese in 2001
by Iwanami Shoteu, Publishers, Tokyo.
This English language edition published in 2004
by Oxford University Press, Oxford
by arrangement with the author
c/o Iwanami Shoten. Publishers, Tokyo
All rights reserved. No part of this publication may be reproduced,
stored in a retrieval system, or transmitted, in any form or by any means,
without the prior permission in writing of Oxford University Press,
or as expressly permitted by law, or under terms agreed with the appropriate
reprographics rights organization. Enquiries concerning reproduction
outside the scope of the above should be sent to the Rights Department,
Oxford University Press, at the address above
You must not circulate this book in any other binding or cover
and you must impose this same condition on any acquirer
A catalogue for this title is available from the British Library
Library of Congress Cataloging in Publication Data
(Data available)
ISBN 0 19 852913 9 (hbk)
10 9 8 7 6 5 4 3 2 1
Printed and bound
in India by
Thomson Press (India) Ltd.
PREFACE
The main purpose of this book is to provide an introduction to the path in-
tegral formulation of quantum field theory and its applications to the analyses
of symmetry breaking by the quantization procedure. This symmetry breaking
is commonly called the "quantum anomaly" or simply the "anomaly". and this
naming shows that the effect first appeared as an exceptional phenomenon in
field theory. However, as is explained in this book, this effect has turned out
to be very fundamental in modern field theory. In the path integral formula-
tion, it has been recognized that this effect arises from a non-trivial Jacobian in
the change of path integral variables. Namely, the path integral measure breaks
certain symmetries.
The study of the quantum anomaly is one of the attempts to understand the
basis of quantum theory better and, consequently, it is a basic notion which could
influence the entire quantum theory beyond field theory. The quantum anomaly
is located at the border of divergence and convergence, though the quantum
anomaly itself is perfectly finite, and thus closely related to the presence of an
infinite number of degrees of freedom. As such an example, we discuss the phase
operator of the photon which involves an infinite number of degrees of freedom
related to Bose statistics. The notion of the quantum anomaly has been mainly
developed in the applications of field theory to elementary particle physics. We
however believe that the notion of the anomaly and the machinery developed to
understand this subtle phenomenon will have important implications on other
fields such as condensed matter physics. To illustrate these points, we analyze
field theory in two-dimensional space-time in some detail. It is shown that the
phenomenon called bosonization characteristic to two-dimensions, namely, the
description of fermion theory in terms of bosons, is simply understood as a
result of the change of path integral variables. Also, the basic notions such as
the central charges of Kac-Moody and Virasoro algebras in conformal field theory
are nothing but the manifestation of the anomalies. We show explicitly how to
derive the central charges from the anomalies in the path integral method.
Another important aspect of quantum anomalies is that they are closely re-
lated to topological properties. For example, the chiral anomaly when combined
with instanton solutions in Yang-Mills theory is a quantum manifestation of
the Atiyah-Singer index theorem in mathematics. Similarly, the ghost number
anomaly which appears in the first quantization of string theory is a field the-
oretical manifestation of the Ricmann-Roch theorem in the theory of Riemann
surfaces.
As for the history of quantum anomalies, which will be explained in Chap-
ter 1, the original indication of the anomaly appeared immediately after the
modern formulation of field theory, namely, renormalization theory. It was, how-
v
vi PREFACE
ever, only in 1969 when the true significance of the quantum anomaly was clearly
recognized. On the other hand, Feynman introduced the path integrals in his for-
mulation of renormalization theory.
We now briefly describe the contents of this book. Chapter 2 covers the basics
of quantum theory. We explain the basic ideas of the Feynman path integral
and Sclrwinger's action principle which are required to understand modern field
theory and the present book, without assuming knowledge of more than standard
undergraduate quantum mechanics.
In Chapter 3 an elementary aspect of the quantum theory of the photon
initiated by Dirac is explained. The basic idea of the path integral formulation
of gauge theory is explained and the notion of the BRST symmetry is introduced.
At the same time, the basic property of the photon phase operator is explained by
using the notion of index. The notion of index reappears in the later discussions
of quantum anomalies. The essence of quantum electrodynamics is summarized
in Appendix A to further supplement the basic path integral formulation of gauge
theory. This appendix will help understand what is going on in this book better.
The phenomenon called the vacuum polarization is explained in Chapter 4.
The evaluation of this apparently simple diagram in a sense has been the most
difficult calculation in the entire renormalization program of quantum electro-
dynamics. By using the gauge invariant regularization introduced to control the
vacuum polarization diagram, we explain the simplest example of the quantum
anomaly, the chiral anomaly, by performing the calculation corresponding to
triangle diagrams but actually without relying on Feynman diagrams.
Chapter 5 is the main chapter in this book. We explain that the symmetry
breaking by quantization (the chiral anomaly) is understood as a non-trivial
Jacobian associated with a change of path integral variables. We present several
supporting arguments for the use of the gauge invariant mode cut-off to evaluate
the Jacobians throughout the present, book. The instanton solution is briefly
summarized, and the relation of the chiral anomaly with the Atiyah-Singer index
theorem is explained. In this connection, we note a problematic aspect of the
unitary transformation to the interaction picture and that the Nambu-Goldstone
theorem does not hold in general in the presence of the quantum anomaly in the
relevant symmetry.
The developments and various applications of quantum anomalies are de-
scribed in Chapter 6 and subsequent chapters. The descriptions are somewhat
condensed in these chapters. Readers may choose appropriate topical subjects in
these chapters. The description of each chapter is arranged so that readers can
read each chapter without referring to other chapters as much as possible. In
Chapter 6, the applications of quantum anomalies to modern gauge theory such
as the Standard Model of elementary particles are summarized. Almost all the
detailed calculations are explicitly performed.
The Weyl transformation is the symmetry which keeps the local angle in-
variant but the length of space-time is changed. An explanation of the Weyl
anomaly is given in Chapter 7. The trace of the energy-momentum tensor is
PREFACE vii
also involved in the analysis. The renormalization group equation is regarded as

a representation of the Weyl anomaly in terms of Green's functions. We thus
present a calculation of the Q function in QED and QCD from the viewpoint of
the Weyl anomaly. This anomaly is also related to the basis of conformal field
theory in two dimensions. To formulate the Weyl anomaly in the path integral,
some basic knowledge of field theory in curved space-time is necessary and it is
briefly summarized in Appendix B.
In Chapter 8 we explain the anomaly-related topics in two-dimensional field
theory. This chapter is aimed at possible applications to condensed matter theory
also. We thus repeat the elementary calculations of chiral anomalies and provide
several representative examples of the bosonization of two-dimensional ferrniori
theory. The conceptual basis of the bosonization. in particular, the issue related
to local counter-terms, is clarified. As a basis to understand conformal field the-
ory, we formulate the Kac-Moody and Virasoro algebras from the viewpoint of
quantum anomalies. The central extensions in these algebras are the manifesta-
tions of chiral and general coordinate anomalies, respectively. This formulation
is also compared to the operator product expansion method in conformal field
theory. The Liouville action is derived in connection with quantized bosonic
string theory and the ghost number anomaly is explained in connection with the
Riemann-Roch theorem.
Chapter 9 covers the interesting developments in the treatment of chiral sym-
metry in lattice gauge theory which took place over the past several years. The
remarkable fact that one can treat the notions of index and chiral anomaly in
lattice theory as the Jacobians just as in continuum theory is explained. We do
not discuss the practical aspects of lattice gauge theory and its numerical simu-
lations, but rather we concentrate on the conceptual aspects of chiral anomalies
and their implications in lattice gauge theory. The lattice theory which is based
on finite quantities only consolidates some aspects of the treatment of chiral
anomalies in continuum theory where the notion of index is better defined. It
is possible to formulate the contents of this chapter with certain mathematical
rigor.
We present a general calculation of chiral anomalies in arbitrary even-di-
mensional curved space-time in Chapter 10, and their relations to the Chern
character and the Dirac genus in mathematics are explained. We also explain
the possible quantum breaking of Einstein's general coordinate transformations
in the theory with chiral fermions, and some simple examples are given.
In Chapter 11 we briefly comment on a possible intuitive explanation of
quantum breaking of symmetries and also on the subjects which are not discussed
in the main chapters such as the descent formula and the global SU(2) anomaly.
As is clear from the descriptions so far we concentrate on the rather classical and
basic aspects of quantum anomalies in the present book, which can be explicitly
calculated by an elementary method in the path integral. Advanced subjects such
as the anomaly cancellation in superstring theory and supersymmetric theory in
general are not discussed, but several references to these subjects are given.
viii PREFACE
In Appendix C, some basic references directly related to the presentation of

this book are given with brief comments. The literature in this subject is vast,
and we present only a small fraction of references related to path integrals and
anomalies. We apologize to those authors whose contributions are not mentioned.
We quote some well-written reviews and textbooks which emphasize different
viewpoints on quantum anomalies to remedy the shortcomings.
One of the authors (KF) started the study of quantum anomalies in the mid-
1970s. He thanks Professor C.N. Yang and all the members of his institute at
Stony Brook, in particular. Professor P. van Nicuwcnhuizen. for encouragement
in the early developments of the path integral formulation of quantum anoma-
lies and for later hospitality at Stony Brook. He also thanks the late Professor
B. Sakita at CCNY who continuously encouraged him from the very beginning of
the investigation. He thanks Professor K. Nishijima, whose research group stim-
ulated the studies of the chiral anomaly in Japan, and Professor Y. Yamaguchi
for encouragement.
The present book is an English translation of the book originally written in
Japanese by KF, and he thanks Professor K. Kikkawa and Mr. U. Yoshida of the
Iwanami Publishing House for the encouragement to write the book.
In this translation, we added some remarks and explanations though the
basic materials are the same as in the original book. The original Chapter 10
on two-dimensional theory has been moved to Chapter 8 and the analysis of
bosonization has been expanded. One of us (HS) thanks Professor T. Fukui for
discussions on the bosonization.
Last, but not least, we thank Mr. Sonke Adlung, Senior Editor of Oxford
University Press, for his enthusiasm and support for this translation.
August 2003
Kazuo Fujikawa and Hiroshi Suzuki
CONTENTS
1 Genesis of quantum anomalies 1

1.1 Introduction 1
1.2 Is the photon massless? 2
1.3 The discovery of the quantum anomaly 3
2 The Feynman path integral and Schwinger's action
principle 7
2.1 Quantum theory of a harmonic oscillator 7
2.2 Path integral for the harmonic oscillator 8
2.3 Quantization of a sca 11
2.4 Path integral for fermions 15
2.5 Path integral for Dirac particles 20
2.6 Feynman path integral and Schwinger's action principle 23
3 Quantum theory of photons and the phase operator 31
3.1 Canonical quantization of the electromagnet 31
3.2 Path integral quantization of the electromagnetic field 35
3.3 Photon phase operator and the notion of index 40
3.4 Is there a hermitian phase operator? 42
3.5 Index theorem for a harmonic oscillator 44
4 Regularization of field theory and chiral anomalies 47
4.1 Current conservation and Ward Takahashi identities 47
4.2 Self-energy of the photon 49
4.3 Quantum breaking of chiral symmetry 57
4.4 Adler-Bardeen theorem 62
5 The Jacobian in path integrals and quantum anomalies 65
5.1 The chiral Jacobian in quantum electrodynamics 65
5.2 Ward-Takahashi identities in quantum electrodynamics 71
5.3 Chiral anomaly in QCD-type theory 74
5.4 Instantons 79
5.5 Atiyah-Singer index theorem 84
•5.6 Nambu-Goldstone theorem 86
6 Quantum breaking of gauge symmetry 92
6.1 Gauge theory with axial-vector gauge fields 92
6.2 Pauli-Villars regularization 96
6.3 Chiral gauge theory and the quantum anomaly 98
6.4 Covariant anomaly 102
6.5 Anomaly cancellation in Weinbcrg-Salam theory 109
ix
x CONTENTS
6.6 The Wess-Zumino integrability condition 112

6.7 Quantum anomalies and anomalous commutators 121
7 The Weyl anomaly and renormalization group 123
7.1 Scale transformation in field theory 123
7.2 Identities for the Weyl transformation and Weyl anomalies 124
7.3 Identities related to coordinate transformations 127
7.4 Weyl anomalies and functions in QED and QCD 132
7.5 The Weyl anomaly in curved space-time 140
7.6 The Weyl anomaly in two-dimensional space-time 144
7.7 Other applications of Weyl anomalies 147
8 Two-dimensional field theory and bosonization 149
8.1 Chiral anomalies in two-dimensional theory 149
8.2 Abelian bosonization of fermions 157
8.3 Non-Abelian bosonization of fermion theory 168
8.4 Kac Moody algebra and Virasoro algebra 174
8.5 Quantum theory of strings and Liouville action 188
8.6 Ghost number anomaly and the Riemann-Roch theorem 194
9 Index theorem on the lattice and chiral anomalies 196
9.1 Lattice gauge theory 196
9.2 Lattice Dirac fields and species doubling 198
9.3 Representation of the Ginsparg-Wilson algebra 202
9.4 Atiyah-Singer index theorem on the lattice and the chiral
anomaly 206
9.5 The operator D satisfying the Ginsparg-Wilson relation 210
9.6 Some characteristic features of lattice chiral theory 218
10 Gravitational anomalies 223
10.1 Chiral U(l) gravitational anomalies 223
10.2 Evaluation by a quantum mechanical path integral 228
10.3 Chern character and Dirac genus 230
10.4 Anomaly in general coordinate transformations 231
10.5 General properties of gravitational anomalies 234
10.6 Explicit examples of gravitational anomalies 236
11 Concluding remarks 240
A Basics of quantum electrodynamics 246
A.1 Quantum electrodynamics 246
A.2 Interaction representation and perturbation formulas 249
B Field theory in curved space-time 253
B.1 Coordinate transformation and energy-momentum tensor 253
B.2 Path integral measure in gravitational theory 258
CONTENTS xi
C References with brief comments 262

C.1 Genesis of quantum anomalies 262
C.2 The Feynman path integral and Schwinger's action
principle 263
C.3 Quantum theory of photons and the phase operator 263
C.4 Regularization of field theory and chiral anomalies 264
C.5 The Jacobian in path integrals and quantum anomalies 265
C.6 Quantum breaking of gauge symmetry 266
C.7 The Weyl anomaly and renormalization group 270
C.8 Two-dimensional field theory and bosonization 272
C.9 Index theorem on the lattice and chiral anomalies 275
C.10 Gravitational anomalies 277
C.ll Concluding remarks 279
Index 282
This page intentionally left blank
1
GENESIS OF QUANTUM ANOMALIES
1.1 Introduction
The central dogma in modern physics is the principle of quantum theory. All
dynamics, cither classical or relativistic dynamics, or all the forces such as elec-
tromagnetic and nuclear forces need to be formulated to conform to the principle
of quantum theory when applied to microscopic systems. The general theory of
relativity is not exceptional, and quantum gravity needs to be formulated in mi-
croscopic domains. Even classical dynamics is realized as a vanishing limit of the
Planck constant H in quantum theory.
The basic equation of quantum theory is given by the Schrodingcr equation
and the wave function y> has a meaning as the probability amplitude. The wave
function is an element of infinite-dimensional Hilbert space, as is expected from
the fact that the wave function of a hydrogen atom has an infinite number of
components. To emphasize this aspect, the wave function is written as the state
vector \ip). From the definition of probability, the square of the state vector needs
to be non-negative, namely, we postulate that the norm of the wave function be
positive definite
The time development of the Schrodinger equation is described by a hermitian

Hamiltonian, and thus the time development induces a unitary transformation
of the wave function
By definition, the unitary transformation does not change the magnitude (total
probability) of the wave function. What changes is, for example, that the first
component is transformed to the second component of an infinite-dimensional
state vector.
To apply quantum theory to general phenomena including photons, it is nec-
essary to quantize the field variable itself and thus formulate field theory. Intu-
itively, one distributes oscillators at each point of four-dimensional space-time in
field theory, and quantizes all these oscillators. The excitation or de-excitation
of those oscillators are interpreted as the creation or annihilation of associated
particles. There arc an infinite number of points and thus one needs to handle
an infinite number of oscillators, or if one formulates the problem suitably, one
1
2 GENESIS OF QUANTUM ANOMALIES
deals with a system where an infinite number of harmonic oscillators interact

with each other. For this reason, the field theory is called a quantum theory of
an infinite number of degrees of freedom. We have an infinite-dimensional Hilbert
space even for a single harmonic oscillator, and now we have an infinite number
of oscillators.
A representative field theory is quantum electrodynamics which deals with
the electron and the photon and their interactions. As is expected, researchers
encountered various subtle problems when they first attempted to describe the
creation and annihilation of particles. Among them, the divergence problem in
field theory is the best known. A careful examination of the divergence problem
associated with creation and annihilation led to the discovery of the phenomenon
called the quantum anomaly or simply anomaly, which is the main subject of this
book. The essential difference between the quantum anomaly and divergences is
that the quantum anomaly does not diverge, though both of them are related to
an infinite number of degrees of freedom. It is proper to understand the quantum
anomaly as a symmetry breaking by the quantization procedure. However, the
quantum anomaly usually appears in a form closely related to divergences in
interaction picture perturbation theory. For this reason, it is often said that the
quantum anomaly is associated with divergences and their regularization. This
way of characterizing the anomaly correctly reflects the historical origin of the
quantum anomaly (and its certain aspects), but one should rigorously distinguish
the quantum anomaly from divergences.
1.2 Is the photon massless?

We here briefly describe the history of the study of the quantum anomaly. Quan-
tum electrodynamics (QED) and its renormalization formulated in the late 1940s
indicated that the applicability domain of the principle of quantum theory is
quite broad and possibly covers all the energy ranges we can think of. This
quantum electrodynamics, as is well known, was formulated by S. Tomonaga,
J. Schwinger, R. Feynman, F. Dyson and others. In particular, the relativis-
tic formulation of quantum electrodynamics by Tomonaga and his associates
in isolated Japan after the Second World War was a surprise to physicists in
the United States, in addition to the meson, theory of H. Yukawa. J. Oppen-
heimer, who was director of the Institute for Advanced Study at Princeton at
that time, asked Tomonaga to send a summary of the research of his group to
him, which was later published as a Letter to the Editor in the Physical Review.
In this letter, Tomonaga explained that covariant quantum electrodynamics and
its renormalization prescription can deal successfully with all the problems asso-
ciated with the electron mass, the electric charge, and the field variables of the
electron and the photon, but he encountered a difficulty with the photon mass.
He stated, "But for this subtraction we cannot find a reasoning so natural and
plausible as that used in the case of mass-type and charge-type infinities, where
the subtraction was considered as an amalgamation. This is because it would
necessarily result in a drastic change of the Maxwell equation for the radiation".
THE DISCOVERY OF THE QUANTUM ANOMALY 3
If this difficulty should persist, it suggested that we may not be able to handle
the experiment all y established vanishing photon mass and the gauge principle,
which ensures the vanishing photon mass, in interaction picture perturbation
theory.
In retrospect, this problem of how to ensure the vanishing photon mass or
the gauge symmetry in field theory was the starting point of the study of the
quantum anomaly, the main subject of the present book. It is not obvious that
one can maintain gauge symmetry or justify the unitary transformation to the
interaction picture in field theory which deals with an infinite number of de-
grees of freedom. These are the essential aspects of symmetry breaking by the
quantization procedure we are going to discuss.
1.3 The discovery of the quantum anomaly

We briefly explain the problem associated with the photon self-energy on the
basis of relativistic quantum mechanics which can describe the creation and
annihilation of particles in perturbation theory. A more detailed analysis will be
given later in this book. We denote the interaction Hamiltonian, which describes
the interaction between the photon (usually denoted by 7) and the electron
(denoted by e) and the positron (denoted e), by HI. The photon self-energy in
second-order perturbation theory is given by
The sum over the energy of the intermediate states En = Ee + Es generally

diverges. This is because the rapid increase in the number of states allowed
for the electron-positron pair for increasing energy En and the allowed energy
is unlimited for a relativistic theory in flat, Minkowski space-time. This is a
manifestation of the infinite number of degrees of freedom in momentum space for
the creation of an electron- positron pair. The vanishing photon mass is ensured
if the energy correction A.E(k) as a function of the momentum k satisfies
In terms of the more intuitive Feynman diagram, this perturbation formula cor-
responds to a calculation of the Feynman diagram in Fig. 1.1. The photon, which
is virtually converted to a pair of an electron and a positron, recornbines after a
certain time, and the quantum correction to the photon self-energy arises in this
process. This diagram diverges (to be precise, a quadratic divergence), and the
photon mass which should be zero may become indeterminate if the renormal-
ization prescription should not be unique. As is explained later, the photon mass
is in fact kept to be zero up to any finite order in perturbation in the modern
formulation of quantum electrodynamics.
In 1949, however, this problem of the photon self-energy was the fundamen-
tal issue of renormalization theory, and the two members of Tomonaga's group,
FIG. 1.1. Photon self-energy correction
FIG. 1.2. (a) Two-photon decay of the neutral TT meson; (b) two-photon decay
of a neutral axial-vector meson
H. Fukuda and Y. Miyamoto, analyzed the next simplest Feyiiman diagrams in
Fig. 1.2. (a) and (b) for the two-photon decay of the neutral TV meson, which
is commonly denoted by TT°, namely 7r° —> 77. In terms of modern language,
they compared two processes: In one of them the neutral spinless TT meson vir-
tually splits into a quark q and anti-quark q pair in the vacuum, and those
two quarks emit two photons before pair annihilation. In the other process, the
spin 1 (axial-)vector meson Ati dissociates into a quark q and anti-quark q pair
and the pair of quarks emit two photons before pair annihilation. The predic-
tion of renormalization theory is that there exists a simple relation (a symmetry
called "Dyson's symmetry" at that time) between two processes. However, an
explicit calculation indicated that the gauge invariance in the second graph is
spoiled and that the relation between the two graphs docs not hold. The latter
graph diverges (to be precise a linear divergence) and thus a careful calculation
is required. But the linear divergence already appeared in the self-energy of the
electron in renormalization theory, and it is known that renormalization works
without any difficulty for the electron self-energy. Consequently, the calculation
of the triangle diagrams should also work.
The appearance of those discrepancies from the predictions of renormaliza-
tion theory was a very serious issue for the renormalization program itself. In
fact, Tomonaga, together with his collaborators, analyzed the triangle diagrams
by using the Pauli Villars regulator, of which a preprint was sent to Tomonaga
from Pauli. They concluded that the issue of gauge invariance can be success-
fully handled by the Pauli-Villars regulator but the relation between the two
THE DISCOVERY OF THE QUANTUM ANOMALY 5
graphs was not uniquely resolved. They stated that we have to wait for future
experimental results to resolve the issue.
A similar arid detailed analysis was performed by J. Steinberger at Princeton,
who learned of the calculation of Fukuda arid Miyamoto through Yukawa staying
at Princeton at that time.1 He also used the Pauli-Villars regulator and arrived
at a conclusion similar to that of Tomonaga.
The issues related to the gauge invariance of the photon self-energy and
the triangle diagrams were analyzed in greater detail by J. Schwinger in 1951.
He handled the photon self-energy by showing that the gauge invariance can be
consistently imposed but concluded anomalous behavior of the triangle diagrams.
The distinction between the ambiguity related to divergences and the quantum
breaking of symmetry was not clear at that time, and as a result a fundamental
understanding of this strange behavior was not achieved.
This problem of the triangle diagrams for the pion (i.e., n meson) decay came
up as a major issue again in the late 1960s. At that time, the understanding of
the pion as a Nambu- Goldstoiie particle associated with spontaneous breaking of
chiral symmetry (or PCAC) was established. If one combines this interpretation
of the pion with the (naive) calculation of the triangle diagrams, it was concluded
that the neutral pion cannot decay into two photons in the ideal limit of the
Nambu Goldstone particle. This contradicted the experimental fact that the
neutral pion predominantly decays into two photons.
This difficulty was analyzed in great detail by J. Bell and R. Jackiw at CERN.
Bell and Jackiw noticed the inevitable deviation from PCAC if one applies
the conventional Pauli-Villars regularization to the a model which incorporates
PCAC. They then showed that one can preserve both PCAC and gauge invari-
ance if one uses a modification of Gupta's implementation of the Pauli-Villars
regularization, but this spoils renormalizability. On the other hand, S. Adler at
Princeton performed a general analysis of the triangle diagrams in spinor electro-
dynamics and discussed the issue of the neutral pion decay in the appendix of his
paper. The final conclusion was that a proper understanding of the anomalous
behavior discussed by Fukuda, Miyamoto, Steinberger and Tomonaga resolves
the discrepancy between theory and experiment. At the same time, it was con-
cluded that the anomalous behavior of the triangle diagrams is unavoidable in
relativistic local field theory with gauge symmetry. Namely, it was established
that the Feynman diagrams can exhibit behavior different from a naive manip-
ulation of canonical field theory and that the anomalous behavior is consistent
with the basic postulate of local field theory and explains the experimental results
well. In effect, these analyses in 1969 marked the discovery of the breaking of cer-
tain symmetries by the quantization procedure, namely, the quantum anomalies.
1
Footnote 11 in J. Steinberger, Phys. Rev. 76 (1949) 1180, reads "Fukuda and Miyamoto.
Prog. Theor. Phys. (in press), were the first to notice that the old results were not gauge
invariant. Their work formed the starting point of this research. I wish to thank H. Yukawa for
making their results available to me before publication". Incidentally, Steinberger later turned
to experimental physics and received the Nobel Prize for the discovery of two neutrinos.
These papers influenced the entire subsequent developments of the subject. For
example, motivated by these papers T. Kimura evaluated the triangle anomaly
in the presence of the background gravitational field in the same year of 1969.
It is known that there are two main classes of symmetries which are broken
by the quantization procedure. The first is the chiral symmetry associated with
Dirac's 75 and it is related to the triangle diagrams we have discussed so far,
and it is called the chiral anomaly. The other is the Weyl transformation, which
changes the length scale of space-time, keeping the local angle invariant; this is
called the Weyl anomaly or conformed anomaly.
On the other hand, in the formulation of renormalization theory Feynman
invented the path integral methods of quantum mechanics arid quantum field
theory. The path integral and the canonical operator formulation of quantum
theory arc formally equivalent, but the path integral later found many appli-
cations in practical calculations. Combined with an intuitive understanding of
quantum processes, the path integral is becoming increasingly important in the
modern formulation of quantum theory and, in particular, quantum field theory.
It has been recognized that the quantum anomalies are understood as arising
from non-trivial Jacobians associated with the change of integration variables
in the path integral formulation. The path integral measure breaks those sym-
metries. This path integral method provides a conceptually more satisfactory
derivation of anomalous identities with the anomaly terms present from the
beginning instead of discovering the anomalies after the evaluation of current
divergences. The purpose of this book is to provide an introduction to the path
integral method of quantization and its applications to the analyses of quantum
anomalies. We show that quantum anomalies are phenomena basic to the entire
field theory, in particular, to gauge theory in general on the basis of the path
integral formulation. We thus start with an introductory account of the path
integral formulation of quantum field theory in the next chapter.
2
THE FEYNMAN PATH INTEGRAL AND SCHWINGER'S

ACTION PRINCIPLE
2.1 Quantum theory of a harmonic oscillator

In this section, we study the quantization of the most important harmonic oscil-
lator as a review of quantum mechanics.
The Hamiltonian of the harmonic oscillator is written in terms of the coor-
dinate q and momentum p as
where m is the mass arid uj is the frequency. The quantization is defined by the
Hciscnberg commutation relation
where H is the Planck constant divided by 2?r.

The time-independent Schrodiriger equation is given by
and the energy eigenvalue is given by
with a non-negative integer n. The eigenfunction un(q) for the nth eigenstate is
written in terms of a Hermite polynomial.
The description of the same problem in terms of creation and annihilation
operators is important not only in the present context but also for second quan-
tization discussed later. In this approach, we define
By using the commutation relations of p and q, we can show
7
8 PATH INTEGRAL AND THE ACTION PRINCIPLE
and the Hamiltoriian is written as
We also have
When we write the state \n) which satisfies
the absence of negative energy states requires the lowest energy state |0) to
satisfy
From the view point of quantum theory, this condition has a more fundamental
meaning as the absence of negative norm states
Starting with the ground state |0), we can generate the general state n) by
repeatedly multiplying the operator «t
The time development of the Schrodinger wave function
is written as
and it shows that the time development operator (evolution operator)

is very fundamental.
2.2 Path integral for the harmonic oscillator

The above evolution operator is written in terms of the path integral as
The right-hand side represents a sum over all the paths in phase space starting
from (ft at t = ii and ending at q/ at I — £/. In the left-hand side, the operators q
PATH INTEGRAL FOR THE HARMONIC OSCILLATOR 9
and p arc defined in the time-independent Schrodinger representation, but in the

right-hand side q(t) and p(t) stand for the time-dependent classiml coordinate
and momentum, respectively.
We now want to understand what this path integration means in more detail.
By noting the completeness of the coordinate and also momentum representa-
tions
we divide the time interval into N segments with e = (tf — tj)/7V. We then obtain
We then note for smalle
and
We thus obtain
In this expression, the path integral measure is denned by
If we consider the limit AT -*• oc by denoting the time derivative of q by q which

is defined for small e by
we obtain the final expression of the path integral

The crucial property to be noted here is that one takes a sum over connected
paths in the path integral. Even in quantum theory, the path of a particle does
not become discontinuous.
The above derivation of the path integral formula shows that the path integral
representation is valid for a more general class of Hamiltonians of the form
We can also perform the integral over the momentum variables explicitly for
general V(q).
To be more concrete, we first note (the integral here is known as the Fresnel
integral)
and then the Feynman path integral formula becomes in the limit
We here defined go = Qi, Q.N = Qf, and the path integral measure including the
normalization factor by
The last expression in equ (2.26) can be readily generalized to particle motion
in three-dimensional space whose coordinates are given by q, and it is written as
This expression shows that the time development in quantum mechanics is given
by a sum over all the paths in four-dimensional space-time starting from <fj at
t ~ ti and ending at g*/ at i = t/ with a phase factor given by an exponential
QUANTIZATION OF A SCALAR FIELD 11
of i/h times the classical action 5 (i.e.. a time integral of the Lagrangian). In
symbolic notation
This formula gives the most fundamental picture of the path integral. The clas-
sical paths appearing here do not have to satisfy the equations of motion. The
path integral of the harmonic oscillator was first evaluated by Feynman. In fact,
Feynman performed many of the calculations in quantum electrodynamics by
the path integral of the harmonic oscillator and its generalizations, and in this
way he demonstrated the usefulness of the path integral.
2.3 Quantization of a scalar field

As an application of the quantization of the harmonic oscillator we discussed
in the preceding section, we now discuss the quantization of a spinlcss particle,
which is called a scalar particle. The wave equation which describes the free
motion of a real scalar field with mass m (as a classical wave) is given in terms
of a real field 4>(t,x) by
This equation is derived by applying the variational principle to the action inte-
gral S = f d4x C, written in terms of the Lagrangian density
as
Our notational conventions arc
namely, we take a sum over indices which appear twice, and the metric of four-
dimensional space-time is given by g^ = (1, —1, —1, -1) = g^v
The Harniltonian density is thus given by
where the momentum (density) conjugate to the field variable is defined by

If one considers an orthoiiormal complete system of real plane waves
inside a box with volume V, (wg(af)} which satisfies one

can expand the field variables as
The Hamiltoriian is then written as
with ui(k) — c\lkz + (mc/H)2, namely, it, is written as a sum of an infinite number
of harmonic oscillators with frequency w(fe). When one defines annihilation and
creation operators by
the Harniltonian is written as
and the quantization condition is given by
In a specific representation of state vectors of the Hilbert space, which is

called the Fock representation, the state vector of a scalar particle is given by2
This state vector represents a state with n-\ particles with momentum tiki, n^
particles with momentum Kk-2 - n^ particles with momentum hk%, and so on.
2
We used the real basis set in eqn (2.35) to make the later transformation to the path
integral in terms of the field variable <j>(l, x) easier. To define the cigenstates of the momentum,
one needs to make a unitary transformation
QUANTIZATION OF A SCALAR FIELD 13
A crucial difference of the physical interpretation of the particle number rep-

resentation (Fock space representation) from the state vector in the first quan-
tization of the harmonic oscillator is that the state vector with n particles with
momentum hk, for example, is given by
and this state is not interpreted as the nth excitation of the harmonic oscillator
as in the quantum mechanics of the harmonic oscillator.
The path integral representation of the evolution operator is given as a gen-
eralization of a single harmonic oscillator as
One can convert this path integral into one in terms of the field variable
as follows: First one remembers
arid thus transforms the variables q%(t) into the field variable <f>(t, x) in the sense
of Dirac as
arid
One also recalls that the specification of all the variables {ql}
K
at t = ti is equiv-
alent to the specification of \(f>i(x}} at all the space points at t = ti. We thus
which preserves the form of the Hamillonian 77 = J^g ^'(^(clcg + 1/2) and the canonical
quantization condition
denote the state vector by |0j,*j}. The path integral is thus written in a rela-
tivistic invariant form
with a suitable normalization factor N, which includes the Jacobian for the above
change of variables. Namely, if one performs the path integral with the measure
with a weight factor which is given by an action integral multiplied by i/h, the
matrix element of the quantum mechanical evolution operator is given by
So far we discussed a free scalar field without any interaction but. for example,
the interaction Lagrangian density, which describes the interaction among four
scalar fields when they come together at a coincident point in four-dimensional
space-time, is given by
where the coupling constant g determines the strength of the interaction. The
interaction Hamiltonian is then given by the last term in
In this case, if one repeats a similar analysis as above the path integral is given
by a Lorentz invariant form as
This last expression is again given by a sum over all the possible field configura-
tions in four-dimensional space-time with the exponential weight factor, which
PATH INTEGRAL FOR FERMIONS 15
is given by the classical action multiplied by i/h, and the final expression is
manifestly Lorentz covariant.
As a classical field, the two scalar fields commute with each other at all the
space-time points, [4>(t, of), <p(t' ,x')] = 0. This property is converted to the Heisen-
berg commutation relation [(l/c2)d>(t, x),</>(t, f )] = (H/i)8^(x~-x') in quantum
theory, and it represents Bose particles. The path integral formula (2.53) gives
a basis of the path integral of all the Lorentz invariant Bose fields.
2.4 Path integral for fermions

To define the path integral for fermions, we first discuss a fermion with a single
degree of freedom which is described by the Hamiltonian
arid then extend the formulation to field theory later. Here Fiu stands for the
energy carried by the fcrmion and a^ and a stand for creation and annihilation
operators with anti-commutation relations
The commutation relations with the Hamiltonian are the same as in the case of
bosons
The physical states are limited to the vacuum state |0) and the one-particle state
|1), and these states are specified by using a^ arid the Hamiltonian H as
Because of a^ a t |Q) = Q, one gees that there are no states such as |2) which contain
more than one particle. Namely, the Pauli exclusion principle is satisfied.
The fact that the fermion is quantized by the anti-commutation relation
suggests that the fermionic particle at the classical level is described by the
Grassmami numbers which always anti-commute. The Grassmann numbers, for
example, £ and 77, satisfy the anti-commutation relations
by definition. Consequently, we have, for example, ry2 = 0 and the Grassmann

numbers have no notion of magnitude. This fact explains why the notion of the
fermion did not exist in classical physics, and one may say that the fermion is a
purely quantum mechanical notion.
The integral over the Grassmann numbers, namely, the linear projection from
the Grassmann numbers to complex numbers, is defined by the left derivative.
To be specific, a general function of £ and f* is written as
with complex coefficients cn ~ c3 by recalling £2 = (£*) 2 = 0. The integral is

then defined by
by noting that the complex number and the Grassmann number commute with
each other by definition. The definition of the left derivative means that we move
the relevant variable to the left-most position before integration, such as in
The integral thus defined satisfies the crucial property that the integration mea-
sure is invariant under "translation" of the integration variable. Namely, if one
defines £,'=£ + n with e another Grassmann number, one can confirm
The first equality in this relation comes from the fact that the naming of the inte-
gration variable is arbitrary (in the conventional integral. f dx f ( x ) = j dy f ( y ) ) ,
and the second equality can be confirmed by an explicit calculation. A similar
property is also satisfied by dt;*. and we have the fundamental relations
From the viewpoint of the path integral, the existence of the "translation invari-
ant measure" implies the existence of the path integral measure which ensures
the equation of motion, and thus it is the most basic property of the path integral
measure.
We here use the so-called coherent states which are convenient to avoid the
complications related to the change of orders of various fermionic operators and
Grassmann numbers. The coherent state for the fermion is defined by
and it is written as a linear combination of |0) and |1)
by expanding into powers of a and o,t by noting a|0) = 0. We used the anti-
commuting properties among the Grassmann numbers and the fcrmionic opera-
tors
By noting
one can confirm that the coherent states thus defined satisfy
One can also confirm the completeness relations for the coherent states by using
the integral for Grassmann numbers
In the path integral we deal with the matrix elements of the evolution oper-
ator
where the state na) stands for the eigenstate of the particle number. Physically,
the number representation (Fock representation) is important, but one can rep-
resent the number states \na) by using the completeness relation of coherent
states as
and thus we first consider the evolution operator between coherent states
When one divides the time interval into N small intervals

the above evolution operator is written by using the completeness of the coherent
states as
We here defined the path integral measure by
The matrix element for an infinitesimal time interval is written by using the
explicit form of the Hamiltonian H = fojjcfia and the properties of the coherent
states as
to the accuracy of linear order in e. See eqn (2.68). Namely, we neglect all the
terms of order O(e 2 ) or higher. Those terms of the order O(e 2 ) or higher are
shown to be neglected in the limit N —> oo. To the same linear accuracy in e, we
can write, by using eqn (2.68),
where we defined £j ~ (£j+i ™ £j)/ e - Irl the limit N —> oc by noting eN = tb — ta,
we obtain the path integral formula
Here we defined £0 = £a and £^ = ££.

An important feature of this path integral is that we obtain the time de-
velopment operator of quantum mechanics by exponentiating the i/h times the
classical action integral obtained from (symmetrized with respect to the time
derivative)
by the replacement a —> £, aJ —i £* and integrating over the variables £ and £*.
The action integral is defined such that the variational principle gives rise to the
equation of motion.
The partition function Z(/3) in statistical mechanics at temperature T is given

by
where /3 = l/(kT). This partition function is written by using the completeness

relation of the coherent states as
Here we used the relation (n|£&){£a n) = (£a\n)(n - £&} which is confirmed by

an explicit calculation for the state vectors {n|£t,} and (£a n}. In contrast, the
coherent states by definition contain an even number of Grassmann variables
(counting a and a^ also) and thus commute with other Grassmann numbers.
The partition function is thus obtained by the path integral with the replacement
tb — ta —> —ififi and an anti-symmetric boundary condition with respect to the
time variable. To be explicit
where we expanded e ^H in powers of J3H and we used (aô) 2 = of a valid

for the fermionic oscillator. We also used cqn (2.68). The result of the operator
formalism is thus recovered.
Incidentally, if one performs the path integral with a periodic boundary con-
dition, one obtains
where F stands for the operator counting the number of fermions in the state
vector, .F|l) = |1) and FjO) = 0. This path integral plays an important role in
the analysis of theories with supersyrnmetry.
2.5 Path integral for Dirac particles

We now apply the path integral in the previous section to field theory. The
classical action for a fermion satisfying the Dirac equation is given by
The 4 x 4 Dirac matrices 7^, fj, = 0, 1, 2, 3, are defined by
Their explicit forms arc given in terms of 2 x 2 Pauli matrices
by thefollowing4 x 4 matrices
In 7°, 1 stands for a 2 x 2 unit matrix. The hcrmitian 75 matrix which describes
the fundamental chiral property in Dirac theory is defined by
and it is given in the present explicit representation of 7 matrices as
The variables ip(t,x)a, a = I ~ 4, stand for a four-component spinor and these

components describe the physical four degrees of freedom corresponding to a
particle and an anti-particle with spin up and down, respectively. We defined
the Dirac conjugate by
When one rewrites the above Lagrangian density (2.84) as
one sees that the canonical momentum conjugate to i,h(t,x) is given by

PATH INTEGRAL FOR DIRAC PARTICLES 21
and the Hamiltonian density is given by
In the present case, we choose a complete set of four-component functions

{un(x)} defined for the hermitian operator h
arid expand the field variable as
We then obtain
We fix the canonical quantization condition so that the relation
derived from the equation motion '<<%?/;(£, x) = hw(t, x) for if) agrees with Heisen-
berg's equation of motion
Namely,
and the system becomes equivalent to an infinite set of fermionic oscillators. Here
we use the result of the spin-statistics theorem, namely, Dirac particles follow
Fermi statistics.
In the present treatment, the "energy eigenvalue" hc\n can assume both
positive and negative values. This problem of negative energy is resolved by the
ie prescription of Feynman and anti-particle interpretation (or in the canonical
quantization, one replaces an —>• b^n, aln —¥ bn for the negative energy modes). In
the path integral, one can use a naive formulation with Feynman's it prescription
m —> m — ze. 3
3
This ie prescription is regarded as imposing a positive energy condition, which is basic in
the proof of the spin-statistics theorem in the path integral framework, cf. K. Fujikawa, Int.
J. Mod. Phys. A 16 (2001) 4025.
The path integral is thus defined as a generalization of the case of a single

oscillator as
On the other hand, one may expand the classical Grassmann variable ib(t,x)
in terms of the Grassmann coefficients an (t) as
If one uses the notation un(S) = (x n), the relation
holds, and one can confirm
Also, the relation
is confirmed. In the path integral, the specification of an infinite number of

{a n (t/)} is equivalent to the specification of 'ip(if,x) for all x, and thus we write
({<xn(tf)},tf\ = ( t l > f , t f .
The path integral of a Dirac particle is thus written by using the classical
action integral
In the present derivation, we have the action which is symmetrized with re-
spect to the time derivative in the exponential. In the actual application of the
FEYNMAN PATH INTEGRAL AND SCHWINGER'S ACTION PRINCIPLE 23
path integral explained in the next section, we define general Green's functions
by choosing tf —^ oo. t, —*• —oo and we add Schwinger's source terms, which
are localized in space-time. The S-matrix element is then defined as a suitable
limit of Green's functions. In such a formulation, one can choose a symmetric
ip(tf —» oo,of) = ip(ti —$• —oo,x) boundary condition (or anti-symmetric bound-
ary condition) with respect to the time variable. We may then perform a partial
integral with respect to time, and the path integral is written as
where we used the integration variable tt>(t,x) = if>(t,x)^j° instead of y(t, x ) ^ .

As a more fundamental formulation, one may consider a Euclidean theory
with Euclidean time r defined by t —l —IT. One may then impose symmetric
boundary conditions both on bosons and fermions in the path integral. One
thus defines the path integral as a generalization of the partition function in
statistical mechanics and considers the limit Tf — Ti —>• oo. In this case, only the
lowest energy state survives
in the above limit. In this way, one can define a general path integral which
starts with a ground state in the infinite past and ends with a ground state
in the infinite future. Here, F stands for the operator counting the number of
fermions in the state \n). See eqn (2.83).
In field theory, the ground state is a result of very complicated interactions,
and thus this definition of path integral simultaneously defines the physical
ground state. In this formulation, one recovers the path integral with the or-
dinary action integral with Feynman's ie prescription if one applies the inverse
Wick rotation T —>• — it to the Minkowski theory after performing the path inte-
gral.
2.6 Feynman path integral and Schwinger's action principle

In this section we explain Schwinger's action principle, which provides a conve-
nient means to relate the path integral to the operator formalism of field theory.
Historically, motivated by a paper by Dirac written in 1933,4 Feynman formu-
lated the path integral arid Schwinger formulated the action principle as a basis
of quantum theory.
Încidentally, this paper also commented on a priori probability, which strongly influenced
Yukawa and through him indirectly TOmonaga when he formulated the Lorentz invariant
Schrodinger equation denned on a space-like surface.
In the following discussion we use the Lagrangian (2.51) for the scalar field
A(x) for definitcness. By noting the relation
derived from the Heisenberg equation in the operator formalism, one can derive
Since |0) is a generalization of the coordinate representation \q) in quantum

mechanics, we have a relation such as
Consequently, one can write the path integral representation
by remembering the results in Section 2.3. When one divides the time interval
into infinitesimal time intervals, the time t is assigned to the field <pi(x) and
thus (j>i(x) —^ 0(*,zO in the integrand. This relation means that a time evolution
specified by the path integral takes place from the initial state at tj to a state at
t, where the path integral includes the factor 4>(t,x), and then the time evolution
continues to the final state at i f . From this relation, one can establish
where the time ordering operation of bosonic fields is defined by using a step
function as5
5
To be precise, the T* product in the path integral avoids the coincident point t\ = i g j
and one needs to consider the ordinary T product in the canonical formalism to treat the case
ti = fe- The T product is obtained from the T* product by the Bjorkeii-Johnson-Low (BJL)
prescription, which is discussed later when we need the T product.
The step function Q(ti —12) is defined to be 1 for ti > t-2 and 0 for *2 > *i- This
time ordering is generalized for a product of n fields, and the fields are arranged
in ascending order of time starting from the right-most field. The time ordering
specified by T* in the operator formalism is realized in the path integral if one
performs the path integral with c-number fields arranged in an arbitrary order.
This is because the path integral is performed starting with a past state to a
future state following the time evolution, and the time ordering is automatically
incorporated.
In the case of fcrmionic fields, the time ordering is defined by
by taking into account the anti-commuting property of field variables. In the

path integral, this ordering with a signature factor is automatically realized if
one uses Grassrnann variables.
We next introduce a c-number source field J(x) and define
The path integral with this source field is then defined by
where the index J on the left-hand side indicates that the time evolution is
described by the Lagrangian C j .
Schwiriger's action principle states that the change of the transition amplitude
is given by
when one changes the Lagrangian slightly, which describes the time development,
by keeping the initial and final states fixed. It is possible to show that this action
principle is equivalent to the Schrodinger equation, though we do not discuss the
equivalence here.
If one applies this action principle to the case where the source term J(x) in
the Lagrangian C, is slightly changed at the space-time point x, one obtains
It is important to recognize that one has a c-numbcr operation in the left-hand

side of this relation, while one deals with an operator in the right-hand side. This
property is common to the path integral, and the precise relation between the
two is clarified later.
The field equation in the operator formalism is obtained by a variation of the
action with respect to (f> as
The matrix element of this operator equation should vanish, arid when combined
with the action principle described above gives rise to
If one can solve the functional differential equation for {(/>/,t/|</>j,t;)j in

eqn (2.120), one can evaluate all the transition amplitudes in quantum theory. In
this respect, the path integral corresponds to solving the functional differential
equation by performing a Fourier transformation in a functional space from J ( x )
to 6(x). To be explicit, one may set
and insert this expression to the above equation (2.120). One then obtains
where Sj = / d4x jCj. If one sets F[<p} = cxp[(i/7i) J d^xjC] in this equation, one
obtains
If the path integral measure is "translation" invariant in functional space,

namely, if the relation
is satisfied for an arbitrary small function e(x), one has
The first equality in this relation is a generalization of the fact that the defi-
nite integral is independent of the naming of integration variables, f dx f ( x ) =
f dy f ( y ) , and the second equality is a result of the translation invariance of the

path integral measure. If one expands this relation into powers of the infinitesimal
function e(ar), one obtains
If one chooses the arbitrary function e(x) to have ^-functional peaks at the point
x, one finally recovers the relation (2.123) required by the action principle.
Prom the above consideration, one understands that the translation invari-
ance of the path integral measure in functional space is the most fundamental
property of the path integral. In the case of bosonic fields, the path integral
is a definite integral over the conventional numbers, and the basic translation
invariance, namely, a generalization of f d(x + t) f(x + e) = f dx f(x + e) or
j dx (d/dx)f(x) = 0 holds. In the case of a fermionic path integral which is
based on Grassmann numbers, the translation invariance of the path integral
measure is shown as we discussed in Section 2.4. The path integral is thus de-
fined just as for the case of bosonic variables. (Incidentally, if one can define a
translation invariant measure for a number system other than the conventional
complex numbers or Grassmann numbers, one can define a quantum mechanical
path integral for particles satisfying the new statistics.)
If one performs a Wick rotation t —» —IT to Euclidean theory with an imag-
inary time in the path integral formula, one obtains
where the Euclidean Lagrangian in the case of eqn (2.51) is defined by
If one imposes a periodic boundary condition <£>/ = fa and performs a path

integral with respect to the variable fa also, one obtains
If one takes the limit r/ — T; —>• oo in this formula, only the lowe>st energy
state, namely, the ground state contributes and the vacuum to vacuum transition
amplitude
is obtained. Even in Minkowski theory with a real time, the same relation is
expected to hold since the amplitude rapidly oscillates for the states other than
the ground state. We thus have
The Feynman ie prescription in the propagator (see Appendix A for explicit

examples) is closely related to the absence of negative energy physical states and
also to the stability of the time evolution operator ((f>f\e~^H^f~''^\d)i) for tf —
ti —>• oo. We assume that H > 0, namely, all the eigenvalues of H arc non-
negative. In this case, if one replaces H —> H—ie = H(i—ie) with an infinitesimal
e > 0 one has
I
and the stability at t/ — <; —> oo is ensured. The divergent probability amplitude
for an unphysical process, where the vacuum decays during time evolution into
negative energy states indefinitely, does not appear. One can also increase e up to
c = 7T/2 and Wick rotate to Euclidean theory without spoiling the stability if one
notes (l—ie)(tf — ti) = e~1"e(tf — ti). In terms of the language of momentum space,
one can rotate the time component as po —>• ipp;, or if one starts with Euclidean
theory and rotates back to Mirikowski theory by PE —* —ipo, the Feynman
it prescription is automatically incorporated. Incidentally, the convention of the
Wick rotation is that one rotates the time components of both the coordinate and
momentum into pure imaginary numbers in such a way that the inner product
p^x11 remains invariant
The convention is x° —> —«x' 4 and po —>• ip4 and the metric g/^, = (I, — 1, —1, —1)
is transformed to g^ = (—1, —1, — 1, —1). In Euclidean theory, the integral be-
comes a Gaussian integral instead of a Fresncl integral and the path integral of
the time evolution operator is defined in a more reliable way. If one associates the
operator formalism with the path integral by means of Schwinger's action prin-
ciple, the it prescription is automatically incorporated as a result of the operator
formalism. However, it is important to understand that the if. prescription arises
from the physical postulate that the negative energy states do not propagate in
the forward time direction.
In the applications of field theory, the vacuum to vacuum transition amplitude
with Schwinger's source function J(x), which has a value only in the localized
region of space-time, is fundamental. Since the source function has a value only
in the localized region of space-time, the asymptotic vacua in eqn (2.134) co-
incide with the vacua in eqn (2.131). As a physical picture, one deals with the
probability amplitude (or general Green's functions) for the process where the
vacuum state at t = — oo evolves into states with many particles generated by
the source function J(x) in intermediate time and then those particles are re-
absorbed into the source function J ( x ) and finally ending at the vacuum state
at t = oc. The Green's function is denned by
The transition amplitude from a general state (different from the vacuum) to
another general state is constructed from the Green's function by the so-called
LSZ prescription. In this prescription, one sets the time coordinates of some of
the field variables appearing in the Green's function either at — oo or oo and
those field variables are associated with either initial or final states, respectively.
For the details of this prescription, readers are referred to standard textbooks
listed at the end of this book. We however note that the LSZ construction of
the physical scattering amplitude starting with Green's functions, which incor-
porate all the effects of interactions, gives rise to a more natural picture of the
physical process compared to the specification of asymptotic states by imposing
a priori boundary conditions on field variables. From a technical view point, the
rcnormalization prescription, for example, is more naturally formulated in terms
of Green's functions.
We have not specified a precise normalization condition of the path integral in
field theory. Here we would like to give a normalization which is useful in practical
applications. Without a precise specification of the path integral measure, the
left- and right-hand sides of eqn (2.131) approach for t; — ti —> oo
with a constant factor N. A simple way to take care of this normalization factor
N is to define
where Z is given by
This prescription corresponds to the normalization of the exact ground state

eigenvalue of H at EQ = 0. In fact we have
This normalization is also consistent with the normalization of the Green's func-
tion in the operator formalism. The Green's function in the Heisenberg picture is
given by (Q\T*$(xi)4>(xi) • • • 6(xn)\0}. By noting the relation H\0} = E0\0) = 0,
one can write
Our normalization of the path integral (2.137) and the definition of Green's
functions in eqn (2.135) are consistent with this operator expression in the limit
I/ —> oo, ti —> —oo.
In practical applications of the path integral, it is convenient to tentatively
include the factor Z into the path integral measure in the process of calculations
and fix the normalization factor as above only when it is required.
3
QUANTUM THEORY OF PHOTONS AND THE PHASE
OPERATOR
In this section, we first present the essence of the quantization of the electro-
magnetic field and its path integral representation. The electromagnetic field is
the simplest example of a gauge field and its quantization nicely illustrates the
technical problems associated with the quantization of gauge fields in general.
We next discuss the problem associated with the phase operator of the photon,
which appears as a result of quantizing the electromagnetic field. This problem
is analyzed on the basis of the notion of index and the postulate of positive defi-
nite Hilbert space. We explain that this problem of the phase operator is closely
related, in technical terms, to the chiral anomaly to be discussed later.
3.1 Canonical quantization of the electromagnetic field

We describe the essence of the canonical quantization of the electromagnetic
field. This quantization is basically given as a generalization of the quantization
of the harmonic oscillator, but the analysis is slightly more involved due to the
appearance of the notion of gauge invariance.
The Lagrangian density which describes Maxwell's electromagnetic field in
the vacuum is written in terms of the four electromagnetic potentials A^(x) =
Ap(t,x), ,u = 0, 1, 2. 3, as
We adopt in this book the convention that we take a sum over indices which
appear twice in the same equation. Our metric convention of space-time is g^v =
(1,-1.-1, —1). The electromagnetic tensor F!JjV = dÂv - dvA^. — —Fvti is
related to the electric field E and magnetic field B in the following way
The electromagnetic tensor F^v does not change under the gauge transformation
(i.e., change of variables)
namely. F'^v = F^v: the tensor is invariant under the gauge transformation. Here
<jj(x) = u(t,x) is an arbitrary function of space-time, and the freedom related
31
32 QUANTUM THEORY OF PHOTONS AND THE PHASE OPERATOR
to 'jj(x) is called the gauge freedom. This gauge iiivariance shows that one of
Afl(x) can be chosen to be an arbitrary function by suitably choosing w(x). This
reduction of freedom by restricting A^ (x) is called gauge fixing. This restriction
on Ay (x) itself is called the gauge condition, and the choice of the gauge condition
is rather arbitrary. The gauge condition on the spatial components of A^x).
which is called the Coulomb gauge,
is fundamental. The gauge condition called the Landau (or Lorentz) gauge, which
preserves Lorentz invariance,
is also commonly used. The gauge condition which is sometimes called the Weyl
gauge
is also convenient when one analyzes the theoretical aspects of quantization.

We here tentatively adopt the Coulomb gauge (3.4). In this gauge condition,
the action for Maxwell theory, which is defined by the space-time integral of the
Lagrangian density, is given after partial integration by
The equation of motion in the vacuum for the time component AQ becomes
where A = Y?k-i (^k)'2 is called the Laplacian. From this equation, we conclude
AQ(X) = 0 by imposing the boundary condition AQ(X) = 0 at spatial infinity.
Consequently, by remembering the space-time metric convention g^ =
(1, — 1, — 1, — 1), the Lagrangian density is reduced to
and the equation of motion is given by

CANONICAL QUANTIZATION OF THE ELECTROMAGNETIC FIELD 33
The momentum variable canonically conjugate to A& is given by
and the Hamiltoriian is given by
Here we used the notation Ak = (d/dt)Ak-

We expand the field variables in terms of a complete set of plane waves as
by keeping the Coulomb gauge condition dkAk = dkHk = 0 in mind. Here we

have k0 = \k\ as a result of the equation of motion (3.10), and as the polarization
vector eÂ' (k) which satisfies kfP*> (k) — 0 we choose
when the momentum is in the positive direction of the z axis. The polarization
for a general momentum k is defined as a suitable rotation of this expression.
The canonically conjugate momentum is then given by
and the Hamiltonian (3.12) is written as
with the notation ui(k) = c\k\ and tentatively assuming that the momentum is
discrete, as in a box normalization.
This Hamiltonian shows that the wave motion of light is equivalent to an
assembly of an infinite number of harmonic oscillators. Because of the condition
fceÂ) (fc) = 0 the light is a transverse wave polarized in the two perpendicular
directions A = 1, 2 with respect to the motion. The quantization condition is
thus given by
and
The energy carried by each plane wave is given by
by neglecting the zero point energy. The Heisenberg equation of motion gives
and the quantized operator at a general time is given by
to be consistent with the expression A^(x} in eqri (3.13).

The state vector for the photon in the Fock space representation is given by
and it represents the state which contains n\ photons with momentum arid po-
larization (Ari.Ai), 112 photons with momentum and polarization (fe,^), and
so on. This means that the photons associated with the electromagnetic wave
are Bose particles. The probability amplitude, which corresponds to the familiar
Schrodiriger amplitude in quantum mechanics, for a single free photon is given
by
If one formally introduces the coordinate and momentum variables starting

with the annihilation and creation operators by
PATH INTEGRAL QUANTIZATION OF THE ELECTROMAGNETIC FIELD 35
the Hamiltonian for the electromagnetic field is written as
3.2 Path integral quantization of the electromagnetic field

We have shown that the free electromagnetic field is equivalent to an assembly of
an infinite number of harmonic oscillators. The electromagnetic field contains two
independent transverse components, and thus the path integral representation is
equivalent to the case with two free scalar fields in eqn (2.48). We thus tentatively
write
where the field variables are actually constrained by the Coulomb gauge condition
dkAk(x) = 0, and thus we have only two integral variables.
The Coulomb gauge condition is generally understood as a decomposition of
the three spatial components of the potential Ak(x) into two transverse compo-
nents dkA^'(x) —O.i — 1, 2, and the gauge freedom dkw(x) as
In this notation, the "length in functional space" when one varies A/t(x) infinites-
imallv is written as
If one recalls that the invariant volume element in general relativity is given
by dV ~ i/dett^,, d4x when the length is given by d 2 s = <?„„ dx^dx", one can
generalize this procedure to the present functional space and the volume element
is given by6
Also in the above decomposition (3.27), the relation 6(dlAi) = 5(dldiuj) =

5(ijj)/ det[—Ac>'( 4 ) (x — y)} holds where 6 here stands for the ^-function. This rela-
tion is a functional spa,ce generalisation of the familiar relation for an ordinary
function S(f) - 6(x - x 0 )/|/'(xo)| where /(x 0 ) = 0.
By using these relations one can establish
The right-hand side of eqn (3.30) defines the correct path integral measure in
terms of the physical transverse components, and thus the path integral (3.26)
is written precisely as
We now rewrite this path integral in a Lorentz invariant manner. For a general
(hermitian) differential operator O ( x , y ) we can establish the relation
by using a bosonic auxiliary variable B(x), where N is a suitable normalization

factor. This relation (3.32) is understood as follows: We first choose the complete
set (pn(x) by
and expand
If one denotes the Jacobian for the change of variables from B(x) to {&„} by ,1,
we have 'DB = dct((x n}) [X,, dbn = JJln dbn- ê tnus nave
B
The present way of deriving the path integral measure is useful in the analysis of the path
integral measure in the presence of the iuslanton solution in non-Abeliari gauge theory, for
example. The final result of the path integral measure in eqn (3.31) is also derived starting
with the phase space path integral by using the Hamiltonian denned by the Coulomb gauge.
where N is a normalization factor. This integral is a generalization of the Fresnel

integral. On the other hand, the determinant is defined by det 0 = Y[n ^"' anc^
thus the above relation (3.32) holds.
If one chooses O(x,y) = ~A6^(x - y ) , B(x) — A0(x) in the above rela-
tion (3.32), and if one applies it to the path integral (3.31), the path integral
representation for a free electromagnetic field is finally given by
by following the transformation of the Lagrangian density for the Coulomb gauge
in cqns (3.7)-(3.9) in a reversed manner. This is the basic formula of the path
integral.
We now re-examine the path integral formula (3.36) from a more general
view point. We first recall the fact that the general gauge field variable Atl can
be written as a result of a gauge transformation parametrized by the gauge
parameter w(x) starting with the field A/j, satisfying the condition dkAk = 0 as
and that
The path integral measure in eqn (3.36) is then written as7
The determinant appearing here is regarded as a Jacobian for the transformation

from d'Af to ui, and thus we have
T
Kor an unconstrained general gauge field, the path integral measure is gauge invariant
T>A^ = TiAfj. as the gauge transformation corresponds to a translation and rotation in func-
tional space. However, the variable A^ is constrained by dkAk — 0 and thus the relation
T>A<£ - "DA,,, does not hold.
This equality in eqn (3.40) means that those three expressions of the path integral
measure give the same result when one integrates over any functional of A£ with
respect to these measures. Namely, the integration with the measure d/j.'Dw is
equivalent to the integration with the measure without any constraint f| 'DA^.
The general path integral measure is thus symbolically written as
and the measure d/i is defined by the unconstrained measure for the entire func-
tional space Ylfj, "DA p divided by the measure Vui which represents the gauge
volume.
This formal expression (3.41) shows that the path integral measure for gauge
theory is generally defined without referring to any specific gauge condition.
Consequently, the path integral measure for the Landau gauge, for example, is
defined by
The path integral measure for gauge theory (3.39) or (3.42) is written in a
more manageable form by using two real fields, which are called the Faddeev-
Popov ghost c,(x), c(x)t = c(x). and the anti-ghost c(x), c(x}^ = c(x). The
variables c(x) and c(x) stand for fermionic particles without spin and thus do riot
satisfy the spin-statistics relation. Those particles are unphysical particles with
negative or indefinite metric for the inner product in Fock space. In Chapter 2, we
explained that fermionic particles are described by ariti-commuting Grassmann
numbers in the path integral formulation. For a general hermitian differential
operator O(x. y), we have
where N stands for a suitable normalization factor. This relation is proved by fol-
lowing the same procedure as for the bosonic variable we have already explained
in eqn (3.32). We expand the variables by using a complete set

as
and note the relation
We then obtain
with a suitable normalization factor TV. Here we used the fact that the inte-
gral with respect to the Grassmann numbers {c7l,cn} is defined by the (left)
derivative.
When one applies the above formula (3.43) to the operator O(x.y) = —A x
<j(4' (x — y ) , the path integral for the electromagnetic field with the Coulomb
gauge is given by
and the path integral with the Landau gauge is given by
where 'DA^ stands for the measure for the entire functional space of the gauge
potentials AQ ~ A^,. We also used the relation
which is a generalization of
The above formula (3.47) or (3.48) for the path integral measure is called
the Faddeev-Popov formula, and it is valid not only for general non-Abelian
gauge fields (Yang-Mills fields) but also for the quantization of general relativity
and string theories. As an important property of this formulation, the effective
Lagrangian density with the Landau gauge in eqn (3.48). for example.
is invariant under the BRST (Becchi-Rouet Stora-Tyutin) transformation
where A is a real constant Grassmann number. The BRST transformation is

regarded as a gauge transformation with a gauge parameter i\c(x). Note that
[«Ac(a;)]t = i\c(x). This BRST symmetry (and its generalization) plays a crucial
role in the analysis of renormalization and uriitarity (i.e.. a positive metric for
the inner product in Fock space and thus the conservation of probability) of the
quantized gauge fields associated with non-Abelian gauge groups such as Yang
Mills and gravitational fields. (In this respect, the electromagnetic field is the
gauge field associated with the-Abelian U(l) group.) This BRST symmetry also
holds for a more general gauge fixing called the £-gauge with a non-negative real
constant
It is shown that this £-gauge leads to the same physical results as other gauge
fixing schemes, though we do not give a proof here.
3.3 Photon phase operator and the notion of index

So far we have reviewed the basics of the quantum theory of the electromag-
netic field. In this section, we discuss the issues related to the phase operator
of the photon. This problem originates from the first paper by Dirac in 1927
on the quantization of the electromagnetic field, arid the problem is also use-
ful to understand the important consequences of the positive definite norm in
quantum mechanical Hilbert space. This problem is also related to the fact that
the uncertainty relation between particle number and phase, namely, a definite
phase generally leads to ill-defined particle number, docs not always hold in the
conventional form. In this analysis, the notion of index plays a central role and
thus this problem is closely related, in mathematical terms, to the analysis of the
chiral anomaly to be discussed in later chapters. In this analysis of the photon
phase operator, the notions such as symmetry and Nother current which play a
PHOTON PHASE OPERATOR AND THE NOTION OF INDEX 41
central role in the analysis of quantum anomalies in field theory do not appear.
But this problem is regarded as a quantum anomaly in a broader sense due to
the fact that the postulate of the absence of negative norm in quantum theory
and the notion of index lead to the results, which contradict the expectations on
the basis of naive classical-quantum correspondence. In any case, this problem
is useful to learn the important notion of index by an analysis of the quantum
theory of harmonic oscillators.
We have shown in Section 3.1 that the quantization of the electromagnetic
field in the vacuum is reduced to a study of an infinite number of harmonic
oscillators. Consequently, we concentrate on a single harmonic oscillator with
frequency u; = 1 (we also set h — 1 for simplicity)
The normalized vacuum state (0|0) = 1 is specified by a|0) = 0. Under this

condition, the negative norm (or negative probability) does not appear. This is
understood for the single photon state [1} = a ^|0), for example, by
The general eigenstates of the photon number operator N = aâ are given by
in the Fock state representation, and we have
In this representation, the annihilation operator a is represented in terms of bra

(k\ and ket jfc) states as an infinite-dimensional matrix
which projects one state to another in Hilbcrt space. The creation operator of
is then given b}--
The spatial components A(t. x) of the electromagnetic field are real quantities
satisfying A^ = A and thus .the notion of phase is not defined for A itself.
Consequently, the notion of phase of the light (or photon) is defined in connection
with the complex state vector in Hilbert space. Historically, Dirac in his first
paper in 1927 on the quantum theory of light, introduced the phase (f>, which
becomes the phase operator after quantization, by
If one transforms from (a, at) to the variables (p, q) of a harmonic oscillator, one
obtains
and the invariance of the Liouville measure in classical analytical mechanics

dpdq = dNdcp suggests that the pair (N, </>) is regarded as a canonical conjugate
set of variables. Consequently, the Poisson bracket {q,p}pB = 1 implies
and thus the quantum mechanical relation
is expected.
As a physical picture for the notion of the photon, namely, the quantization
of light, Dirac explained the quantization of the conjugate variable A* to be an
integer as a consequence of the 2?r periodicity of the angular variable (p. Also, the
above commutation relation between N and cp leads to the famous uncertainty
relation
Namely, a definite photon number N leads to an undetermined phase <f>, and a

definite phase leads to an undetermined photon number.
As is clear from the above discussion, the notion of phase of the photon is
defined either by operating with e'-<* on a state vector or taking the average value
of e'"^ by state vectors. The state vector in quantum mechanics is always complex
for a real electromagnetic field as well as for general complex fields, and thus the
notion of the phase is defined for all fields. Naively, the phase operator is defined
as a variable conjugate to the particle number.
3.4 Is there a hermitian phase operator?

It is known that there is no hermitian phase operator of the photon in a strict
sense. If one defines
following Dirac (here we distinguish the general phase operator tp from the vari-
able 6, which was supposed to be hermitian), we have
This second expression (N + 1) 1//2 a shows that, as long as we stay in the re-
stricted space where the eigenvalue of N is constrained to be non-negative, the
IS THERE A HERMITIAN PHASE OPERATOR? 43
following problematic aspects of the phase operator (p have nothing to do with

the singularity at the origin associated with the polar decomposition (which is a
generalization of writing a complex number as z = re"9). One can confirm
for the above phase operator etlf (3.65). Namely, the variable (p (p is not defined
as a hermitian operator. If the operator tp<p should be hermitian, the relation
e z ^(e* v )t = (e**')te1^ = 1 should hold.
Notwithstanding the above analysis, many people looked for a "more nat-
hermitian phase operator (j>.
ural" licrmitian <$>. Among these attempts, the operator due to
D.T. Pegg and S.M. Barnett is best known. The basic idea of their approach is
to consider a subspace of Fock space where the photon number is truncated at ,s.
Namely, one considers the (s + l)-dimensional space of photon number
and later one lets s become arbitrarily large. In this case one may define the
phase operator <p
with <PO an arbitrary constant, and one can confirm
which shows that e^ indeed defines a unitary operator. In this construction, the
annihilation operator in the (s + l)-dimensional Fock space is defined by
and if one defines a\ = (a s ) t , a slightly deformed commutation relation is ob-

tained
From these relations, one can confirm
If one can show that the effects of the state |.s) disappear in the limit of
sufficiently large s, one can successfully define a herinitian phase operator for
the photon.
3.5 Index theorem for a harmonic oscillator

We analyze the above problem of the phase operator by using the notion of index,
and show that the anomalous behavior of the phase operator is closely related, in
mathematical terms, to the quantum anomaly to be analyzed in later chapters.
We recall that the ordinary representation of a harmonic oscillator satisfies the
index relation
where dim ker stands for the dimension of the kernel of the operator a. Namely,
dim ker a stands for the number of normalizable states un which satisfy
In the above example
and thus the dimensions of the kernels become 1 and 0, respectively, and the
above index relation is satisfied. The index relation is also written as
or by using the trace as
The equivalence of cqn (3.76), which uses aâ, to eqn (3.73) is concluded by
noting that au = 0 implies a'au = 0 and conversely, alau = 0 implies (aâu, u) =
(au,au) = 0 by considering the inner product, and the positive definite inner-
product implies au = 0. The equivalence to the representation which uses the
trace is shown by noting that a'a and aa1 include precisely the same number of
non-vanishing eigenvalues. Namely, if An ^ 0 in
the state defined by satisfies
and thus un and vn satisfy a 1 : 1 correspondence. Consequently, the difference

of the number of states with 0 eigenvalues appears on the right-hand side of
eqn (3.77). To be explicit, one can confirm
The notion of index has an important property that the index being an integer
does not jump to another integer under a superposition of smooth infinitesimal
INDEX THEOREM FOR A HARMONIC OSCILLATOR 45
transformations. Consequently, the index is preserved under the well-known op-

eration of squeezing of the photon in quantum optics (i.e., the compression in a
certain direction of the phase space for the photon) .
If one assumes, as Dirac suggested,
with a unitary t/(</>), we have
which shows the unitary equivalence of of a and aa*. We thus have
which contradicts the above index relation. We thus conclude that the unitary
U((j)) (or cquivalently, hermitian (!)} docs not exist.
On the other hand, if one assumes that as is an arbitrary (s + l)-dimensional
square matrix, the relation
is shown. The trace Trî) here is defined in (s + l)-dimensional space, and

this relation is concluded by rioting that a\as and asa\ share all the non-zero
eigenvalues and that, both the finite-dimensional aja s and asa| contain the same
number of zero eigenvalues, and thus all the eigenvalues coincide. This shows
that in a truncated theory to arbitrary finite-dimensional square matrices, the
notion of index is consistent with the presence of a herinitian phase operator <p.
In fact the phase operator (3.68) of Pegg and Barnett is an example of this
construction. (In this construction, however, the singularity associated with a
polar representation could appear. This comment also applies to the fermionic
operator a, which is represented by a 2 x 2 matrix.)
The next important issue is whether the operator in eqn (3.68) approaches the
conventional operator by preserving the index relation for a sufficiently large s.
In general one expects that this is impossible since the integers 0 and 1 cannot
be equivalent. Mathematically, one may examine
which is obtained by first moving the second term in eqn (3.84) to the right-hand
side and then subtracting the same quantity Tr(,,)(e~ asa -<) from both sides. Here
Ti(a) stands for the trace over the first s-dimensional subspace in the (s + 1)-
dimensional space. The right-hand side of this relation is thus the contribution
of the state s), and the index 1 appears by noting a s aj|s) = 0, which follows
from the definition of as. It is confirmed that the limit of large s of this relation
is smoothly defined, and in the limit s —> oo the index relation
is derived. The state s) does not play a special role on the left-hand side in
this limit. For example, one may start with Tr( s / +1 -|(e~ 0 '» a *) — Tr(y)(e^ aaa *) for
any s' < s, and the limit s' —> oo gives the same result as the left-hand side of
eqn (3.86). The existence assumption of a smooth limit of the hermitian phase
operator (3.68) for s —>• oe thus contradicts the notion of index. The unitary
phase operator is excluded in the limit s —>• oc.
The vanishing index in a truncated space, as in the above example, and the
recovery of non-zero index in the infinite limit of the truncation parameter (for
example, the inverse lattice spacing I/a in lattice gauge theory) is one of the
characteristic properties of the quantum anomaly in field theory to be described
in this book. In fact, the evaluation of the index in terms of the trace (3.80),
if the trace is understood in a general context, is closely related to the evalua-
tion of the Atiyah-Singer index to be explained later, which is a mathematical
representation of the chiral anomaly.
Prom these considerations, one may expect that the anomalous behavior of
the phase operator is regarded as an example of quantum anomalies. From this
viewpoint, the anomalous behavior of the phase operator is an unavoidable phe-
nomenon in quantum theory. As physical evidence of the unavoidable presence of
the anomalous behavior, one may examine the issue if the minimum uncertainty
relation is consistently described by a modified phase operator. In fact, one can
show that the hermitian operator <f> in eqn (3.68), which appears to avoid the
anomalous behavior, cannot determine if the measurement saturates the quan-
tum limit, namely, if the measurement saturates the uncertainty relation or not
in the cases where small photon numbers are involved. (The basic reason for
this phenomenon is that the state \s) appears in the intermediate states of the
product of two operators and its effect does not vanish for whatever large but
finite s.) In the cases with small photon numbers, the artificial deviation from
the minimum, uncertain relation becomes significant if one uses the hermitian
operator 0 to analyze the experimental data, even if the actual experiment sat-
urates the quantum limit. The correct quantum limit is described by using the
number operator N and the hermitian operator C(<p) = (l/2)[e* v + (e^)^], for
example.
4
REGULARIZATION OF FIELD THEORY AND CHIRAL

ANOMALIES
We illustrate the regularizatiori of field theory by taking quantum electrodynam-

ics (QED) as an example, and show that the mass of the photon is kept to be
0 even after higher-order quantum corrections. We also illustrate the simplest
example of the quantum breaking of chiral symmetry. The one-loop fermionic
diagrams become fundamental in these considerations, and the gauge covariant
regularization of one-loop Fcynman diagrams for arbitrary theory is explained. It
is shown that the chiral anomaly is defined independently of Fcynman diagrams
in perturbation theory. The basic idea of the Adler-Bardeen theorem, which as-
serts that the identity with the chiral anomaly does not receive any higher-order
corrections, is briefly explained.
4.1 Current conservation and Ward—Takahashi identities

We start with the analysis of the one-loop photon self-energy, which was men-
tioned in Chapter 1, and show that there exists a calenlational scheme which
preserves gauge irivariance and thus the vanishing photon mass. The basic La-
grangian density of QED is given by (see Appendix A)
where t\lv = dfj,Av—dvA^. The terms with a derivative of F^v in this Lagrangian
density are introduced to implement a rcgularization, which is called higher-
derivative regularization, namely a method to reduce the degree of divergence in
the theory. The parameter M with the dimension of mass provides a truncation of
the momentum integral. If one lets M —> oo before any calculation, one recovers
the unrcgularized Lagrangian density of QED. The last two terms in £eff stand
for the gauge fixing term and the Faddeev Popov ghost term.
To simplify the notation, we use the natural units
In these units, the dimensions of all the quantities are represented by the mass
dimension [M]
47
48 REGULARIZATION OF FIELD THEORY AND CHIRAL ANOMALIES
For example, length has the same dimensionality as the Compton wavelength
h/(mc) — 1/m, and the charge e2/(hc) ~ 1/137 becomes dimensionless. The
requirement that the factor \' d4x C,ef[, appearing in the exponential in the path
integral formula, should be dimensionless determines the mass dimensionality of
various field variables as given above.
The mass for the photon field
does not appear in the Lagrangian density (4.1) of QED. This is because such a
mass term changes its form (i.e., is not invariant) under the gauge transformation
of the electromagnetic field
Experimentally, it is known that the mass of the photon is 0, and thus such a mass
term not only does not appear in the starting Lagrangiari but also should not be
induced by higher-order corrections. In gauge theories such as quantum electro-
dynamics, it is of fundamental importance to perform calculations by preserving
gauge invariancc in any finite order of perturbation theory. The important rela-
tions which express the gauge invariance of the starting Lagrangian in terms of
Green's functions are called Ward-Takahashi (WT) identities.
We write the path integral with source terms added as
where the source terms are defined by
The WT identity resulting from the gauge invariance of the starting action
is derived by considering the following change of integration variables (gauge
transformation)
and the following identity in the path integral formulation holds

SELF-ENERGY OF THE PHOTON 49
This derivation is based on the fact that the value of a definite integral does not
depend on the naming of integration variables. This relation when combined with
the invariance of the path integral measure under the above change of variables
(which is established in Chapter 5)
and the change of the action in the order linear in a(x)
finally gives the identity
by keeping the terms linear in a(x). We also performed a partial integration by

using the fact that a(x) is local, namely, it vanishes at space-time infinity. We
also defined
for a general operator O(x). This notation is often used in this book.
If one chooses a(x) in the identity (4.12) to be a function with a d-functional
peak at x, one obtains
which is the basic relation known as a WT identity.

If one functionally differentiates this identity with respect to J" (y) once and
then sets all the sources to be 0. one obtains
The Fourier transformation of this last relation means that the probability am-
plitude for electron-positron pair creation from the current j^ = w-j^tl; and
then pair annihilation into a photon vanishes when multiplied by the momen-
tum. This process in the lowest order in perturbation is represented by the dia-
gram in Fig. 4.1. In common language, this corresponds to current conservation
4.2 Self-energy of the photon

The above WT identity gives a constraint on the photon self-energy if one recalls
that the current is the source for the photon. We are going to explain this fact
below.
FIG. 4.1. Feynmaii diagram corresponding to current conservation (4.15)
The actual calculation is performed in Euclidean theory, so we start with

the definition of Euclidean theory. The basic definition of Euclidean theory is to
write the time component of coordinates as
and to regard x4 as a real number. At the,

the same time, the 7 matrix of Dirac is
defined as
The hermitian 7° is thus converted to an anti-hermitian 74, arid in our convention

all the 7 matrices become anti-hcrmitian (7 M ) T = —7**. The general contravariant
vector V*M is also replaced by V° —> — iF4 in Euclidean theory. The covariant
vector is then replaced by AQ —>• iAj with a real quantity A±. Consequently, the
inner product of two vectors VÂ^, remains unchanged in Euclidean theory, but
the metric of space-time is replaced by g^ = ( — ! , — ! , — 1 , — 1 ) . To make the
path integral convergent in the Euclidean metric, one may replace B —> iB and
(: —>• ic in eqn (4.1).
The Dirac operator in Euclidean theory is given by
and its form does not change from the Minkowski one. The conjugate ib of the
Dirae spinor w is treated as an independent variable in the path integral, and
in Euclidean theory it is understood as transforming like i^ under a Lorentz
transformation (i.e.. SO(4) transformation) of Euclidean theory. Consequently,
the operator ]/) becomes a hcrmitian operator defined by the inner product
in Euclidean theory.
The path integral in Euclidean theory is given by
and the propagator of the free photon is derived by considering all the terms
quadratic in _4M in eqn (4.1). By recalling eqii (A. 20) in Appendix A. with £ = 0,
the momentum representation of the propagator when converted to the present
Euclidean metric is given by
The second propagator stands for the propagator for the auxiliary field, and it is
not used in the present discussion. The counting of the degree of divergence on the
basis of the propagators given here and the ordinary fcrmionic propagator (A. 20)
in Appendix A shows that all the diagrams except for the fermionic one-loop
diagram are either convergent or logarithmically divergent, which are made finite
when combined with gauge in variance. In fact, it is known that all the Fcynman
diagrams except for one-loop diagrams arc made finite if one uses the higher-
derivative regularization in gauge theory. Consequently, one can evaluate all the
quantum corrections without encountering divergences if one can make the one-
loop diagrams finite.
In the actual applications of gauge theory, the dimensional regularization. in
which one avoids divergences by performing calculations in space-time dimen-
sions less than four, is commonly used. However, this regularization spoils the
algebraic consistency of the chiral symmetry which is defined in terms of Dirac's
7,5 ; no consistent definition of 75 in dimensions slightly away from four is known.
In the analysis of quantum anomalies, this property of the dimensional regu-
larization is inconvenient since one cannot decide if the anomalous behavior is
coining from this insufficient definition of 7-5 or from a more fundamental origin.
In the analysis of quantum anomalies, the gauge invariance (or covariancc) of
current operators is essential. It is thus important to find a regularization. which
docs not destroy the gauge invariance of currents when one incorporates the
quantum effects corresponding to one-loop Feynman diagrams. The covariant
regularization to be discussed below gives a regularization of general one-loop
Feynman diagrams associated with current operators without spoiling gauge in-
variance. By this regularization, one can readily judge if a relevant symmetry
other than the gauge symmetry is broken by the quantum effects or not. One
might then wonder what happens if the quantum anomaly appears in gauge
symmetry itself; even in such a case, as is explained later, one can judge if the
quantum anomaly exists in gauge symmetry or not by combining a consideration
of Bose symmetry.
We first note that the electric current (or in general simply called the current)
to which the photon Afl couples in the Lagrangian density is given by
This current is rewritten as
where 1J) = 7 M (9 M — ieA^) and the trace stands for the sum over Dirac indices.
In writing this relation, we used the anti-commuting property of ijj(x) and th
propagator for the fcrmion field inside the background c-number electromagnetic
field Atl.(x) in Euclidean theory, which is given by
The derivation of this propagator proceeds in the path integral with a proper
normalization factor (2.137) as
where we performed partial integration with respect to ift in the last line. We
then obtain the desired relation (4.24) by multiplying by (i$> - m)^ 1 on both
sides.
We next expand the denominator in eqri (4.23) in powers of the electromag-
netic field eAp as
If one uses this expansion in the formula for the current (4.23) and retains only
the terms linear in Av(x) (the zeroth order term in Av vanishes in eqii (4.23)),
one obtains
FIG. 4.2. Photon self-energy correction
In this expression, the derivative 9M acts on all the x-variables appearing to the
right of it. If one functionally differentiates the last expression in eqn (4.27) with
respect to eAv (z) . one finally obtains
We here used the following representation of the (5-function
and the notation $ = j^k^. The quantity Il^^g) (4.28) obtained here contains
two fermion propagators in eqn (A.20) and it is shown that it coincides with
the one-loop photon self-energy given by Fig. 4.2 on the basis of the conven-
tional Feynman rules. Uliv(q) gives a basic quantity which is called the vacuum
polarization tensor.
The correlation functions of more than one-current tjîp operators, which
are usually calculated by expanding in powers of eA^ in perturbation theory,
are obtained by differentiating eqn (4.23) with respect to eAp suitably many
times. Those correlation functions correspond to Feynman amplitudes with one
fermion loop. The important point to be noted here is that we take the limit
y = x before we perform the momentum integration in eqn (4.28). Consequently,
the present calculational method is completely different from the point-splitting

method, which regularizes the current correlation functions by considering the
current operator ip(y}î/)(x) which is defined for y ^ x.
The vacuum polarization tensor Ii^v(q) (4.28) is quadratically divergent as is
confirmed by counting the powers of momentum variables. The regularization of
the gauge field propagator in eqn (4.21) does not improve the present divergence
since the gauge field does not appear inside the present loop diagrams. We want
to regularize this divergence. Our basic strategy is to use an arbitrary smooth
function f ( x ) , which vanishes rapidly at large values of x,
and regularize the current (4.23)_as
If one takes the limit A —>• oo before performing any calculation, one recovers
the original expression (4.23) by noting /(O) = 1. The parameter A provides the
momentum space cut-off, as will be seen later. We also used the operator
which is hermitian in Euclidean theory. This operator ]/)

$> is gauge invariant (or to
be precise, gauge covariant), and the gauge covariance is automatically ensured
if one considers this combination of the differential operator 9M and the gauge
field eAf,.
In practical calculations
for example, is convenient and the regularized current is given by
Just as in the case of unregularized current, one may expand eqn (4.34) in
powers of eA^ and retain the terms linear in eA^*
8
Those who are not familiar with the calculations of Feynman amplitudes may tentatively
proceed directly to eqn (4.39) and may examine the calculations later.
(4.35)
In this calculation, we used the relation
which is derived by using the properties of 7'' matrices {7^,7"} = 2<7A*!/ and
the definition of the electromagnetic field [D M ,jD y ] = —ieF^ in terms of the
covariaut derivative Dfl = dIJt — ieA^. Here F^ = dÂv — dvA^. In the above
expression, the differential operator <9M acts only on the field immediately after it
in (dÂ11) and Fltv. but otherwise <9M acts on all the x variables appearing on the
right-hand side of it. By considering the functional derivative of this expression
with respect to eAv(z), one obtains the regularized vacuum polarization tensor
If one uses the Fourier representation of the ^-function in eqn (4.29). the final
momentum representation of the vacuum polarization tensor is obtained:
In this final formula (4.38), the first term stands for the regularization in
terms of a naive form factor [A 2 /(—/c 2 + A 2 )] 2 , which by itself breaks gauge invari-
ancc as is known from the early days of renormalization theory. The remaining
two terms in the present regularization recover this lost gauge irivariance.
After performing the trace over the Dirac matrices and the momentum in-
tegral over kIJ' by using Feynman's parametric representation,9 the result of the
lowest-order perturbation theory is obtained in the following gauge invariant
form
in the limit of large A. This final result is proportional to q^q" — g^q'2 and
satisfies the gauge irivariance condition q^q^q" — g^q'2) = 0 in Fig. 4.1 and
equ (4.15). The mass of the photon thus remains 0. If IP"(g) contains an extra
term proportional to #'''", the condition (4.15) is not satisfied. In such a case, the
Fourier transformation of II'"'(g) back to the coordinate representation would
contain a term corresponding to the photon mass (4.4) as a result of quantum
corrections.
The present covariant regularization gives rise to a gauge invariant result
without introducing any divergence in the intermediate stages of calculation.
This clearly shows that the gauge invariance in quantum electrodynamics is not
spoiled by quantum effects. It is important to notice that we worked in strictly
d = 4-dimensional space-time at all stages of the calculation, and this property
becomes crucial in the later discussions of quantum anomalies.
It is shown that the coefficient of the term containing In A 2 in eqn (4.39)
is related to the /3-function in the renormalization group, which is explained in
Chapter 7, and it is independent of the choice of the regulator /(&). For example,
one can confirm that the choice
9For example,
QUANTUM BREAKING OF CHIRAL SYMMETRY 57
gives the identical coefficient of In A 2 . This result is also understood from the
fact that the coefficient of the term In A2 is related to the quantum anomaly
(Wcyl anomaly) to be discussed in Chapter 7 and that the Weyl anomaly is
independent of the choice of the regulator f(x). In contrast, the constant ter
— 1/3 depends on the choice of the regulating function. However, this does not
lead to any difficulty, since this constant term together with In A2 is absorbed
by the wave function renormalization factor and the final value is uniquely fixed
by the renormalization convention.
The present covariant regularization is easily extended to the one-loop calcu-
lations of multi-point functions, and it is also applied to calculations in the chiral
theory such as the Weinberg-Salam theory, which breaks parity. The covariant
regularization is easily extended to the cases with Higgs couplings, which mix
left and right components of fermions.
4.3 Quantum breaking of chiral symmetry

On the basis of the simplest gauge theory, quantum electrodynamics, we perform
a calculation of the quantum breaking of chiral symmetry (or chiral anomaly)
associated with an axial- vector current, which was first encountered by Fukuda,
Miyamoto, Tomonaga, and Steinberger.
The Euclidean path integral (4.20) for quantum electrodynamics is given by
where we set the source term at £./ = 0. We also defined
in the path integral measure. Namely, the measure [PA^] includes all the terms
related to gauge fixing. We consider the following change of path integral vari-
ables
which is called the chirai transformation. Here,

because of the anti-commuting property of 75 and 7°, we have the expression
as above. We defined the chirai matrix 75 which is hermitian
by
We have an identity under this change of variables

namely, we have the statement that the definite integral does not depend on
the naming of path integral variables, which is a generalization of the ordinary
integral j ' d x f ( x ) = f dy f(y). We look at parts which depend on fermionic
variables in this identity. Firstly, the variation of the action in the exponential
factor is given for an infinitesimal
if one notes 75 7M + 7^75 = 0. If one further assumes that the path integral
measure does not change under the above change of variables, one has
The above identity is written after those calculations as
when one retains only the terms linear in a(x) by expanding the action in the ex-
ponential factor in powers of a(x). If one considers a(x) which has a (^-functional
peak in the neighborhood of x, one obtains after partial integration the "naive"
chiral identity
The current tp(x)ÎJ-î!}(x) which contains 75 is called the axial- vector current.
In the case of the vanishing fermion mass m = 0, the axial-vector current is
conserved and such a theory is called chiral invariant.
The puzzling property, which Fukuda and Miyamoto encountered in 1949 and
which was analyzed in greater detail by Bell and Jackiw and by Adler in 1969, was
the fact that the naive chiral identity (4.49) does not hold in Lorcntz invariant
perturbation theory. The actual perturbative calculation is very complicated
and one has to deal with subtle momentum integrals which include a linear
FIG. 4.3. Feynman diagrams which give rise to the triangle anomaly
divergence. However, if one uses the covariant regularization we introduced, the

calculation becomes very simple, and one can show the deviation from the naive
chiral identity without relying on perturbative calculations.
A perturbative check of the chiral identity proceeds, just like the calculation
of the vacuum polarization tensor in eqn (4.23), bv expanding the denominator
factor ]/) = in
in powers of eA^. In the present case, the term linear in eA^ vanishes due to
charge conjugation properties and the quadratic term in eA^ gives the lowest-
order term. In terms of the language of Feynman diagrams, one evaluates the
triangular diagrams in Fig. 4.3 where the current appearing on the left-hand
side of eqn (4.49) is denoted by j£. For this reason, the chiral anomaly is also
called the triangle anomaly. The degree of divergence d of this triangle diagram
is d = 1 and the diagram diverges linearly.
If one applies the covariant regularization to this calculation, one deals with
with a regulator function f(x) and the trace is taken over the freedom of Dira
matrices. A direct evaluation of this quantity is possible, but what we want to
know is the deviation from eqn (4.49) and thus it is sufficient to evaluate the
derivative of this regularized current
This calculation is simply performed in the following manner. We first define

a complete set of eigerifurictioris of the hermitian operator ]/)
and rewrite eqn (4.51) as
The summation here converges if the function /(A^/A2) rapidly apj

for large A?,/A 2 . We now take the derivative of this expression
where we used the fact that 7M is anti-hermitian ( 7 M )^ = — 7^ and

0 and that ^<^n = A n yj re . This formula is further rewritten as
This final expression shows that the gauge covariant regularization gave rise
to an extra last term which can break chiral symmetry, in contrast to the naive
chiral identity (4.49). The chiral identity is modified from a naive form by quan-
tum effects if the above extra term does not vanish.
This last term is evaluated as follows: We first assume that the operator
/(^* 2 /A 2 ) is sufficiently convergent, and rewrite the sum over the four-component
functions ipn(x) to an integral over a complete set of plane waves
where the trace is over the Dirac indices which take four values. This calculation
is performed as
where we used the following relation
by noting {7^,7"} = 2<? A " / . We then moved the factor etkx through the operator
/(.£9 2 /A 2 ) and performed a scaling of the variable k^ -> Afc M . We next expand
the contents of the function f ( x ) around the value x = —k^k^ = \k2\ (note that
x > 0 in our Euclidean metric convention)
and use the fact that only the terms of order I/A 4 or larger survive in the
limit A —l oo in cqn (4.58). We also note that the trace with the factor 75 is
non-vanishing only when the trace contains four or more 7-matrices, The terms
satisfying these non-vanishing conditions are the third term in eqri (4.60) which
contains
In this wav we have
62 REGULARIZATION OF FIELD THEORY AND GHIRAL ANOMALIES
after calculating the 7-matrix trace
where the completely anti-symmetric symbol e^"0"3 is normalized by the conven-

tion e1234 = 1 in the Euclidean metric. We also performed the integral
by noting d4k = Tr'2d\k'2\ \k'2\ = -K^dxx. The basic property of the regulator
function (4.30) including lim^ô x f ' ( x ) = 0 and lim^-^co xf (x) = 0 was used.
We shall come back to this calculation when we analyze the Jacobian in the next
chapter.
We thus derived the chiral identity with the anomalous term (the last term)
without relying on perturbation theory. The last extra term does not rely on the
details of the regulator f ( x ) . We recognize that the anomalous term, which breaks
the naive WT identity (4.49) associated with chiral 75 symmetry, is finite, and
thus the divergence and the quantum breaking of symmetry (quantum anomaly)
are basically independent notions.
A more detailed expression of the identity which includes the effects of the
source terms for il>(x), i/)(x) is given by
by taking into account the variation of the source terms under the chiral change
of fermionic variables in eqn (4.20). The identity for general Green's functions10
is obtained by functionally differentiating this form of identity with respect to
source functions suitably many times and then setting all the source terms to
beO.
4.4 Adler— Bardeen theorem

From our derivation of the identity (4.64) which does not depend on pertur-
bation theory, it is regarded as an exact bare identity valid for the regularized
theory (4.1). It is in fact known that the identity (4.64) is exact and not mod-
ified by higher-order quantum effects even in the large regulator mass M limit
10
The result for the Minkowski metric is given by choosing the convention f1230 = 1 and
removing the imaginary factor i in front of the anomaly term.
ADLER-BARDEEN THEOREM 63
in cqn (4.1) after the ordinary renorrnalization operation. This fact is known as
the Adler-Bardccn theorem.
The Adler-Bardeeri theorem is fundamental, but for technical reasons we do
riot discuss the details of the renormalization procedure in this book. We thus
explain the essence of this theorem only briefly.11 We first notice that the opera-
tor 2im'tjj(x)-j5ib(x) is the only gauge invariant pseudo-scalar operator with mass
dimension 3 in the present theory. The operator when inserted into a Green's
function thus induces only divergences proportional to itself but it becomes fi-
nite after wave-function and mass renorrnalization in the present renormalizable
theory
where N[^>(x)"/^(x)] is called the normal product and it gives rise to a finite
result when inserted into any Green's function. Similarly, i/j(x}^'y5i/}(x') is th
only gauge invariant (to be precise, BRST invariant) axial-vector quantity with
mass dimension 3 in the present theory, and thus it induces only divergences
proportional to itself: To be precise, the divergences arising from the diagrams
which do not include the triangle diagrams in Fig. 4.3 are absorbed into the
wave-function renormalization factor of \l>. We thus write
with a divergent constant Z and a finite operator N[^}(x)^^c,'ip(x)}. Finally,

the anomalous term '2ie2/(32ir'2)fV</a0FllllFa,3(x) is a gauge invariant pseudo-
scalar quantity with mass dimension 4, but it is written as a total divergence
f^'^F^F^x]IJ\ftivaf>lAvFa&(x)\.
= 2d For this reason, it induces divergences
proportional to itself and also divergences proportional to the divergence of the
axial-vector current <9MJV[?/>(x)7A*75'!/'(a;)], but it does not induce divergences pro-
portional to the pseudo-scalar quantity N(v>(xY^(x}}. Actually, a careful anal-
ysis shows that it does not induce divergences proportional to itself, and if one
uses the conventional renormalization eA^ = erArtl in quantum electrodynamics,
one can write
with a divergent constant Z' and a finite operator

By using these relations in eqn (4.64). one obtains
11
Those who are not familiar with the details of renormalization theory may skip this section.
The main purpose of this book is to formulate the quantum anomaly in a non-perturbative
manner as much as possible.
which implies the equality between the divergent coefficients Z = Z' and those
divergent terms cancel on both sides of the identity. We thus obtain the exact
operator identity for renormalized operators
and the anomaly term does not receive higher-order corrections except for the
replacement
• 5
THE JACOBIAN IN PATH INTEGRALS AND QUANTUM

ANOMALIES
In this chapter, a general path integral formulation of the quantum anomaly

for chiral symmetry is given. The quantum anomaly is identified as the Jaco-
bian arising from the symmetry transformation of path integral variables. The
correct WT identities with an anomalous term arc thus defined from the be-
ginning. To realize this idea, it is essential to have a suitable regulator which
controls divergences by preserving as much symmetry as possible. In the path
integral formulation, the gauge invariant mode cut-off is shown to be useful. As
topics related to quantum anomalies, a brief account of iristanton solutions and
the Atiyah Singer index theorem is given. As applications of these notions, we
explain that the (naive) unitary transformation to the interaction picture does
not exist. We also discuss the general absence of Nambu-Goldstone particles
associated with spontaneous symmetry breaking when the relevant symmetry is
spoiled by the quantum anomaly and instariton effects.
5.1 The chiral Jacobian in quantum electrodynamics

We start with the Euclidean path integral of quantum electrodynamics
where the 7^ matrices are anti-hermitian
and
The chiral matrix is hermitian
and it satisfies (75 )2 = 1 and
The covariant derivative is defined by
by using the electromagnetic potential A^, and the action Suax stands for the
2
Maxwell action SMSX = — (1/4) vA^) . /d4z
We include
(0^.4^
a suitable
- d
gauge fixing term into the path integral measure of gauge fields [DA^].
65
66 THE JACOBIAN IN PATH INTEGRALS AND QUANTUM ANOMALIES
To analyze the Jacobian for the chiral transformations
in a reliable way, we expand the Dirac fields
in terms of the eigerif unctions of the hermitian
We used the notation (x\n) = fn(x) following Dirac's transformation theory.

This basis set formally diagonalizes the Dirac action
To derive an exact relation such as a WT identity, the basis set which diagonalizes
the action is essential. The path integral measure for fermions is then written as
by noting that ip, $, an and an are all Grassmann numbers which aiiti-commute
with each other and that the integral over Grassmann numbers is defined by the
(left) derivative. This gives a manageable definition of the path integral measure
and the path integral (5.1) itself in. our applications.
THE CHIRAL JACOBIAN IN QUANTUM ELECTRODYNAMICS 67
We now consider the evaluation of the Jacobian for coordinate-dependent

infinitesimal chiral transformations
To work in the above definition of the path integral measure, we expand the field
variables into the complete set
When one multiplies this expression by yj, (x) on the left and integrates over the
space-time coordinates, one obtains
Similarly, one may multiply by tpn(x) on the right to give
and integrate over coordinates to obtain
From these expansions, the transformation rules of the path integral measure arc
given by
where TV is taken to be infinity later. Here we used the fact that the integration
variables an and &„ are both Grassmann numbers and the integral is defined
a
by the (left) derivative. Consequently, the Jacobian appears as an inverse of the

case for conventional numbers. We next use the relation
valid for an infinitesimal a(x), which is derived by using the relation det M =
exp tr In M valid for a general matrix M and the relation
valid for an infinitesimal a(x).

If one writes the measure in terms of the variables ifj and if>, the Jacobian for
infinitesimal chiral transformations is given by
To evaluate the Jacobian explicitly, we consider
where f ( x ) is an arbitrary smooth function which rapidly approaches 0 at x = oo

/v
with normalization /(O) = 1. We thus replace the mode cut-off limjv_j.00 z^n=i
by a cut-off in terms of eigenvalues limjM_>oo /((An)2/M2). If one chooses f(x)
suitably, it is expected that one can approximate the mode cut-off arbitrarily
closely by this procedure. See Fig. 5.1. This cut-off in terms of the eigenvalues
preserves gauge invariance and thus defines a desirable regulator. In some basic
applications we explicitly show that the final results arc independent of the choice
of the regulator function f(x) if it satisfies a minimum requirement. In most
applications the simple choice f ( x ) = erx is convenient, and it is related to the
THE CHIRAL JACOBIAN IN QUANTUM ELECTRODYNAMICS 69
FlG. 5.1. A smooth regulator function f(x) which approximates the mode
cut-off depicted by the broken line
heat kernel or ("-function regularization. The present regularization is sometimes
called the gauge invariant mode cut-off regularization.
Since the operator j$ f (ft'2 / M2) is well regularized, we perform a unitary
transformation of basis vectors from {tpn(x}} to plane waves {elkx}, and by
doing so we can extract the gauge field dependence of the Jacobian explicitly.
This transformation to plane waves is related to the unitary transformation from
the Hcisenberg picture to the interaction picture in the operator formalism. The
Jacobian is thus given by
where tr on the left-hand side stands for the trace except for the space-time
integral which is included in IV. The remaining trace in the last expression
stands for the trace over Dirac indices. We used the relation
k/j, after moving etkx through

hghjgjhghgjhgjhgjhgjhgjhgjhghghhghgg
We here recall the trace properties of Dirac matrices
with e1234 = 1. For example, tr7 5 = tr 757^7" = 0 is shown by noting the

relations such as
We used the fact, that different 7M matrices anti-commute with each other to-
gether with (T^1)2 = -1 and the cyclic property of the trace tr 7 1 7 2 = tr-y 2 ^/ 1 .
By using these properties and noting that the terms with powers 1/M5 or
higher vanish in the expansion in powers of 1/M in eqn (5.22), we obtain
by noting e1234 = 1. We also used
where x = —k^k^ > 0 in our metric convention. The final result of the Jacobian
docs not depend on the regulating function f ( z ) , which was introduced to make
the intermediate stages of the calculation more reliable, as long as it satisfies
The regularization by f(x) preserves gauge in variance at all the intermediate

stages of the calculation. In the above integral, one recognizes that the contri-
butions from any finite domain —L < fcM < L vanish in terms of the original
momentum variables before one applied the scale transformation fc,t —» Mfe^.
In this sense, large momentum regions or short distances in space-time deter-
mine the value of the Jacobian. Precisely for this reason, the Jacobian (quantum
anomaly) is closely related to divergences in the interaction picture perturbation
theory, though the Jacobian itself is finite.
WARD-TAKAHASHI IDENTITIES IN QUANTUM ELECTRODYNAMICS 71
We summarize our result so far as
The Jacobian thus evaluated agrees with the chiral anomaly discussed in Sec-
tion 3 of Chapter 4, as will be shown in the next section. The fact that the
Jacobian does not depend on the detailed properties of the regulator f ( x ) is
consistent with the perturbative analysis of Adler, who showed that the chiral
anomaly for the triangle diagrams is uniquely determined independently of the
divergences if one imposes gauge in variance. The regulator function f ( x ) we used
here is the same as the function used to regularize the Nother current (and the
fermion propagator) in Chapter 4.
One may pause here and ask why the quantum anomaly can be evaluated
as the Jacobian? Some reasons one can think of at this stage are: The quantum
anomaly is finite and independent of regularization as long as gauge irivariance
is preserved, and in this sense it is universal. Note that the use of the basis set
{<£„}. which diagonalizes the action, and our replacement of the mode cut-off
by the eigenvalue cut-off in eqn (5.21) uniquely fixes the regulator f($>2/M'2),
which is the essence of the gauge invariant mode cut-off regulari/ation. Also,
the quantum anomaly does not depend on the detailed properties of Feynmaii
diagrams in perturbation theory, namely, the Feynman diagrams contain more
information than the quantum anomaly itself. We can thus evaluate the anomaly
without knowing Fcynman diagrams. A further supporting argument for the
present prescription to evaluate the Jacobian will be given later by using the
notion of index.
5.2 Ward— Takahashi identities in quantum electrodynamics

We have explained how to evaluate the Jacobian associated with a change of
path integral variables. We here review the formulation of identities in quantum
electrodynamics by using the result. If one considers the infinitesimal localized
chiral transformation of path integral variables
in the path integral
where 5jyax stands for the Maxwell action and the source terms are given by
we have the identity
This identity is a statement that the definite integral does not depend on the
naming of integration variables, namely, it is a generalization of f d x f ( x ) =
j dy /(?/). In this identity, the Jacobian is given by, as was shown in the previou
section
and the change of the action is given by
The change of the source terms is given by
If one combines these relations in the above identity (5.33) and keeps terms linear
in a(x), one obtains the WT identity
If one chooses the parameter a (a;) such that it has a ^-functional peak around
the space-time point x, one obtains the more familiar form of identity
which agrees with eqn (4.65) in the previous chapter. We have thus shown that
the Jacobian associated with the change of variables represents the quantum
WARD TAKAHASHI IDENTITIES IN QUANTUM ELECTRODYNAMICS 73
breaking of chiral symmetry (namely, the chiral anomaly). This derivation of the
chiral identity does not depend on perturbation theory, arid in this sense it gives
a non-perturbative definition of the identity.
A major difference between the present derivation of eqn (5.37) and the
derivation of the identity (4.65) in Chapter 4 is a conceptual one: In the derivation
of eqn (5.37), we have the anomaly coming from the Jacobian from the begin-
ning and thus "anticipate" the appearance of the anomaly when one performs
a careful evaluation of the divergence of the axial- vector current. In contrast, in
the derivation of eqn (4.65) one "discovers" the anomaly only when one performs
a careful evaluation of the divergence of the axial- vector current.
We have identified the origin of the quantum anomaly and its reliable eval-
uation, and now we apply the technique to analyze the issues related to gauge
invariance in quantum electrodynamics, which were once studied by Tomonaga
and Schwingcr. For this analysis we consider the following change of path integral
variables
The Jacobian is then evaluated in the same way as for chiral transformations as
which gives a trivial Jacobian independently of the choice of basis vectors ipn (x).
On the other hand, the change of the action is given by
We thus obtain the WT identity for gauge transformations following the same
procedure as for the chiral identity as (by suppressing source terms)
which leads to the identity in operator language
This identity shows that the gauge symmetry (current conservation) is not bro-
ken by the quantum anomaly, and thus it suggests that there exists a calcula-
tional scheme which ensures the vanishing photon mass even to non-pcrturbative
accuracy. This conclusion is consistent with the analysis in Chapter 4.
5.3 Chiral anomaly in QCD-type theory

We start with an explanation of a straightforward generalization of quantum
electrodynamics, which is called Abelian gauge theory, to a nori-Abelian gauge
theory which has the same structure as QCD (quantum chromodynamics). The
Euclidean path integral for this theory is given by
Dirac's 7-matrix convention is the same as for QED and 7'' is anti-hermitian
and the chiral matrix 75 defined by
is hermitian. The covariant derivative Dtl is defined by
by using the generator Ta of a rion- Abelian gauge group. The non- Abelian gauge
field carries the same number of components as the generators of the group.
Hereafter, we use the convention that we take a summation over the indices
which appear twice in the same expression, except when stated otherwise. We
often use the notation AjJL = A^Ta as in the last expression above. The field
strength tensor F£v of the gauge field, which is a generalization of the electric
and magnetic fields E and £>, is defined by
is an action for the non-Abelian gauge field, which is called the Yang-Mills
field,
and it gives a generalisation of Maxwell's action of the electromagnetic field. The

path integral measure for the gauge field [£>AM] in eqn (5.43) contains a suitable
gauge fixing term also, and the details of the gauge fixing are not important for
the analysis in this chapter.
We give a brief explanation of non-Abelian groups. A general compact group
is written as a direct product of simple (non- Abelian) groups and Abelian groups.
CHIRAL ANOMALY IN QCD+-TYPE THEORY 75
The generators of a simple non-Abelian group are given by hcrmitian matrices

and satisfy the commutation relation
The constant fabc which is completely anti-symmetric with respect to indices a,

b, c is called a structure constant. We normalize the generators by
As an explicit example, the generators of the group SU(2), which is familiar in

the analysis of angular momentum, are written in terms of three Pauli matrices
as
and they satisfy the relation
where tabc stands for a completely anti-symmetric symbol with respect to the
three indices with e12r! = f. The generators of the group SU(3) are written in
terms of eight 3 x 3 matrices {A0}, a = 1 ~ 8, which arc called Gell-Mann
matrices, as
In this book, we discuss only the group SU(n). which is generated by n2 — 1

generators.
Coming back to the above path integral, the Dirac field appearing there is
written for the group SU(2) in a precise notation as
and it contains two ordinary four-component Dirac fields ib\ (x) and ip% (x). Con-
sequently. Dirac matrices should also be written as 8 x 8 matrices by arranging
two conventional 4 x 4 7 matrices as diagonal components. It is however com
mon to use the simplified notation as above in eqn (5.43), since this does not
induce any confusion. The actual QCD, which describes the strong interaction,
is based on the gauge group SU(3), and thus we deal with a three-component
field consisting of three conventional Dirac fields.
In the case of the gauge group SU(2), for example, an arbitrary function
which belongs to SU(2) is written as
by using three real functions ua(x) and the generators Ta of SU(2), which are
given in terms of the Pauli matrices. The non-Abelian (local) gauge transforma-
tion is then defined by the replacement of variables
The gauge transformation for matter fields thus corresponds to a rotation in

internal space specified by a local g(x) at each space-time point; for the gauge
field Ay^ the transformation is given by a combination of rotation (the first
term) and translation (the second term). The covariant derivative D^ is the
transformed under the gauge transformation as
and consequently, the combination
is transformed in the same manner as the field variable \i>(x] itself. For this
reason, D^ is called the covariant derivative. The field strength tensor of the
gauge field, which is expressed in terms of the covariant derivative, is transformed
as
If one recalls g(x)g^(x) — 1, the action appearing in the above path integral
formula (5.43) remains invariant
under the gauge transformation, namely, the action is gauge invariant. In the
above expression, we used the convention irTaTb = (l/2)Sab. The function g(x)
itself is also often called a gauge transformation.
We examine the localized (space-time dependent) infinitesimal chiral trans-
formations
CHIRAL ANOMALY IN QCD-TYPE THEORY 77
for the non-Abelian gauge theory. To analyze the Jacobian for the chiral trans-
formations, we expand the fermioiiic variables
as in the case of QED in terms of the eigeiifunctions of the hermitian operator
The action for the fermioii is formally diagonalized by this expansion
The path integral measure for this case is written as
and it leads to the following Jacobian for an infinitesimal chiral transformation
as in the case of QED in Section 5.1.

The actual evaluation of this Jacobian proceeds by replacing the mode cut-off
by the eigenvalue cut-off as
where f ( x ) is an arbitrary function of x which approaches 0 rapidly at x = oo

with a normalization /(O) = 1. See Fig. 5.1. The condition /(O) = 1 means that
one recovers the original expression for M —>• oo before one performs any manip-
ulation, and any regulator function should satisfy this condition. The operator
is a well-regularized operator and one may use a complete set of

plane waves to extract the gauge field dependence as
where the parameter a(x) and the associated integral over the space-time coordi-
nates arc omitted. The remaining trace here stands for the trace over the indices
of Dirac's 7 matrices and the indices of the matrices of gauge group generators.
We also used
and performed the scale transformation of integration variables kp —> Mk^ after
moving eikx through /(^> 2 /M 2 ).
If one expands eqn (5.68) in powers of 1/M. one obtains
by noting tr 7 5 = tr 75 [7^,7"] = 0. We used the normalization of the anti-

symmetric symbol e1234 = I and
In the present metric convention, a; = —k^k1' > 0. In this momentum integral,

an arbitrary finite domain — L < k^ < L for the momentum variables before one
performed, the scale transformation k^ —?> Mk^ gives a vanishing contribution. In
this sense, the large momenta or short distances determine the quantum anomaly.
INSTANTONS 79
This integral is also independent of the function /(:/;), which was introduced to
avoid ambiguity in the intermediate stages of the calculation, as long as it satisfies
as in the case of quantum electrodynamics.

The Jacobian for the chiral transformation is thus evaluated as
where the remaining trace is over the indices of matrix generators of the Yang
Mills field. Consequently, the chiral identity for the general case with source
functions rj(x), fj(x) discussed in Section 5.2 is generalized to
following the same procedure as for quantum electrodynamics.

The calculation (5.70) can be readily generalized to any d = In even-dimen-
sional space-time arid the result is written as13
5.4 Instantons
The Euclidean non-Abelian gauge theory, unlike Abelian theory, accommodates
a classical solution which is called an instanton. This instanton is fundamental
since it describes tunneling from one vacuum in field theory to another. This
classical solution when combined with the chiral anomaly leads to an index
theorem, which relates the number of zero modes of the Euclidean Dirac operator
with the instanton number (or Pontryagin number).
We first briefly describe the instanton solution. In Euclidean theory, not only
the coordinates but also the field variables are replaced by
and the resulting A4 and x4 are regarded as real quantities. Consequently, the
metric of Euclidean space-time becomes g^v = —5^v and, for example,
12
To obtain the result in the Minkowski metric, one may remove the imaginary factor i in
front of the anomaly factor with the convention
13
Tho convention in d = 'In dimensions is
The gauge action then satisfies
where we included the coupling constant g into the field variables as
We use an imaginary time to describe tunneling in field theory just as in quantum

mechanics. In particular, the classical solution of Euclidean theory gives the
stationary point of the action and thus describes the path where the tunneling
probability becomes a maximum. The instanton solution precisely gives such a
stationary point of the action.
5.4.1 The instanton solution

We choose the gauge group SU(2) to explain this solution.14 An arbitrary element
of the gauge transformation which has a value in SU (2) is written as
by using four real functions (a(x), b(x)) and three Pauli matrices f. If we require
that the Euclidean action SE for the gauge field is finite, the field strength
tensor is required to satisfy F^v(x) —>• 0 at space-time infinity \x —>• oo. If
one remembers that the action SE is invariant under the gauge transformation,
the gauge field A^x) itself should approach the configuration which is gauge
equivalent to the vacuum
The instanton solution which actually satisfies this condition is explicitly given
by
The parameter p is a real constant and we defined
14
In this book we actually use only the fact that there exists a solution which satisfies
cqn (5.90).
INSTANTONS 81
where x expresses the spatial coordinates of space-time. The instanton solution

Ay for r —> co shows the behavior
and the leading first term has a magnitude of the order 1/r. namely, in the
present notation O(l/r).
Because of the condition a(x)'2 + [b(x)]2 = 1, one can regard the element g(x)
of the gauge group SU(2) as describing a unit hypersurface (which is written
as S?>) in a four-dimensional space whose coordinates are given by (a, b). On the
other hand, the function g(x) = x4- + ixf which is used to construct the instan-
ton solution shows that a point on the unit hypersurface in four-dimensional
Euclidean space-time described by (x4,x), which is also an 53, and a point on
the hypersurface described by g(x) of the gauge group SU(2), are in 1 : 1 cor
respondence. Namely, when the coordinates (x4,x) of space-time cover S3 once,
the element g(x) covers the hypersurface S3 in the gauge space once. The quan-
tity which describes this topological property is called the winding number. The
winding number v for the specific g(x) above is given by
by using the totally anti-symmetric symbol f^"aP normalized by g1234 = 1. The

integral here stands for a surface integral over the hypersurface S3 located at
infinity of four-dimensional space-time, and dS^ stands for the surface element
which is orthogonal to the /u-axis.
In fact, near the point x4 = oo, namely, near the point
3
x = 0, for example, the integrand of the integral to define v is written as
which shows that the integrand at any point of the hypersurface is given by
the surface clement for a unit hypersurface divided by 2?r 2 . If one recalls that
the surface area of the unit hypersurface (which is a volume in the conventional
sense) is given by 2?r2, the above integral over the entire hypersurface in fact
gives the winding number v = 1.
If one uses
instead of g ( x ) in the formula defining the winding number //, it is shown that
the winding number // = n is obtained. Also, the anti-instantoii with v = — 1 is
obtained by setting n = —1 in this formula, or equivalently by
The winding number v is written by using the instantori solution as
The integrand here is written as a total divergence, and the integral is written
as a surface integral at space-time infinity by using the Gauss law10
where we used the behavior of the instanton solution at infinity in the last
expression. Namely, this expression agrees with the definition of the winding-
number (5.85).
Finally, we would like to show that the instanton discussed so far in fact
satisfies the Yang-Mills equation. We define FIJtv = (l/2)e /i)ya ^F a 0 and use the
following Schwarz inequality
If one uses the formula for the winding number on the right-hand side of this
relation, we have
and the equality holds only for the case
where ± corresponds to the signature of the winding number v. If one uses the
+ signature in this relation (5.93) and uses we have
which is in fact confirmed to be satisfied by /(r 2 ) = r 2 /(r 2 +p 2 ) used to construct

the instanton in eqn (5.82).
This shows that the instanton gives the winding number v = 1 and at the
same time it gives the minimum of the action for v = 1. The field configuration
16
This property is shown by using the Jacobi identity
The quantity appearing in eqn (5.90). K^ = l/(87r 2 ) is called

the Chcrn-Simons form a,nd is very fundamental.
INSTANTONS 83
which gives the stationary point of the action is a solution of the field equation
derived from the action, and thus the instanton is in fact the solution of the
Euclidean Yang-Mills field equation.
The instanton solution
approaches the configuration ig(x)d fj/<j^(x) at infinity, which is gauge equivalen

to the vacuum. The gauge function g ( x ) , however, cannot be expressed as a
superposition of conventional "small" gauge transformations which satisfy the
boundary condition 5(00) = 1. Physically, this gauge function g(x) is interpreted
as describing tunneling starting from one vacuum at x4 = — oo to another vacuum
at x4 = oo following the Euclidean imaginary time. (For this reason, this solution
is called an "instant-on" , indicating that it appears and then disappears instantly
in time unlike the ordinary soliton.) The value of the action for this solution
SE = —8n2/g2 describes the height of the tunneling barrier, and in the path
integral the factor with this action in the exponential
is understood as giving the leading term of the tunneling probability.

The vacuum in non-Abelian gauge theory thus has a structure similar to the
motion of the electron moving in a periodic (sharp) potential, and it is described
by a wave function similar to the Bloch wave. The path integral of Euclidean
gauge theory is shown to be written as
though we here forgo the details of this derivation. In this path integral, the
sum over v runs over all the integers, and the term v — 1 corresponds to the
contribution from the instanton solution we have discussed so far. The general
v corresponds to multi-instanton solutions. The real parameter 0 in this formula
is an arbitrary constant, arid the vacuum state \9, ±oc) is called the 9 vacuum.
The gauge field configuration in the path integral measure
v is described df
by
Afj,(x) = A, L(x}(v) + a^(x) by using the instanton solution A^(x)^ with instan-
ton number (i.e., the winding number) v and the fluctuation a M (x) around the
solution.16 In analogy with the Bloch wave, the exponential factor exp(i0f) has
the following meaning: v corresponds to the (difference) of the positions of the
electron and the real parameter B corresponds to the Bloch momentum.
16
The instanton solution with v = 1 contains eight deformation parameters corresponding to
four parameters describing the position of the center and one parameter p describing the size of
the instanton and three other gauge parameters. In the path integral, one needs to perform the
integral over these parameters also. For general simple groups, it is known that the instanton
solution is constructed for each SU(2) sub-group, and no other solutions.
When one performs a change of the integration variables
with an infinitesimal constant parameter a, the calculation of the Jacobian (5.73)

shows that the path integral measure is transformed as
and thus the parameter 0 is changed as
At the same time, the Euclidean action changes as
These properties show that the 0 parameter can be freely changed in the theory
with a massless fermion, and the parameter 0 has no physical meaning in such a
theory. (In the presence of the source terms / dx (fjiji + iprj), one needs to perform
the simultaneous re-definition of sources t] —>• /7exp(—id/2) and 77 —S> cxp(—i6/2)r
to completely eliminate the 9 dependence from the generating functional.) The
9 term is written as
and ^""PFpvFap is a generalization of EB in QED and thus breaks CP sym-

metry. This breaking of CP symmetry for 0 ^ 0 in QCD is known as the strong
CP problem.
5.5 Atiyah Singer index theorem

Coming back to the analysis of the quantum anomaly, we re-examine the calcu-
lation of the Jacobian factor for the chiral transformation
If one considers a global limit of this relation by choosing a as a constant, both

sides of this relation are written as
ATIYAH-SINGER INDEX THEOREM 85
If one combines this relation with the relations valid for the eigenfunctions
with
which is derived from ^75 + J5$> = 0, one obtains
by rioting /(O) = 1. In this relation, n± respectively represent the number of

eigenstates with vanishing eigenvalue for the Dirac operator T/) with positive or
negative chirality eigenvalues
and the right-hand side of eqri (5.106) shows the instanton number (or Pon-
tryagin number). For $><f>n(x) = 0, one can show Jf)^(fn(x) = 0 and thus
Jf>\(\ ± 7o)/2](^n(^) = 0. One can thus choose the eigenfunctions of y><pn(x) = 0
to be the simultaneous eigenstates of 75.
The above relation n + — n _ = v. which holds among the number of eigenstates
of the Dirac operator with the vanishing eigenvalue and the topological property
of the non-Abelian gauge field, is known as the Atiyah-Singer index theorem in
mathematics.17 In mathematics; however, this theorem is proved for the Dirac
operator defined on a compact four-dimensional hypersurface S"1 where all the
eigenvalues are discrete. Consequently, it is not obvious that this theorem holds in
our analysis of Euclidean theory defined on R4, but a detailed explicit analysis of
the eigenfunctions of the Dirac operator in the presence of the instanton solution
shows that the relation implied by the theorem holds in Euclidean field theory.
In contrast, the local anomaly relation before integrating over four-dimensional
space-time holds in a more general situation, for example, in Minkowski theory
also.
This index theorem is also rewritten by defining
17
This index theorem was formulated in the late 1960s, about the same time as the modern
recognition of the quantum anomaly in physics.
and noting
which arises from the relation 75^+^75 = 0. We further note that (1 ± 7s)/2 is
the projection operator satisfying [(1 ± 7s)/2] 2 = (1 ± 7s)/2. The index theorem
is then written as
for M H> large. The Atiyah-Singer index theorem written in this form has the
same form as the index theorem (3.80) we discussed in Chapter 3 in connection
with the photon phase operator.
This index theorem provides another motivation and justification for identi-
fying the Jacobian with the chiral anomaly. The Jacobian for the global chiral
transformation with a constant a is given by
independently of the large N behavior, and the expression for the Pontryagin
index v = n+ — n_ in eqn (5.88) shows that the Jacobian in fact carries the
proper information about the chiral anomaly.
We have thus provided three major supporting arguments for the identifica-
tion of the Jacobian with the quantum anomalies and for the gauge invariant
mode cut-off regularization of the Jacobian: The first is the gauge invariant regu-
larization of the current operators in Section 4.3: the second is the diagonalizatior
of the basic action in the path integral; and the third is the Atiyah-Singer index
theorem. In subsequent chapters, we shall show that the gauge invariant mode
cut-off of the Jacobian gives rise to all the known local anomalies.
5.6 Nambu—Goldstone theorem

We discuss two implications of the instanton solutions and the index theorem.
The first is the general aspect of field theory, and the other is concerned with
the Nambu-Goldstorie theorem.
5.6.1 Unitary transformation to the interaction picture

We first discuss the implication on the so-called interaction picture in field theory.
In our formulation of quantum anomalies, we expand the fermionic variables as
NAMBU GOLDSTONE THEOREM 87
in terms of the eigenfunctions of the hermitian operator
This expansion formally diagonalizes the fermionic action
and in this sense the above expansion may be regarded as corresponding to

the Heisenberg representation in the operator formalism. We also obtained the
Jacobiari for an infinitesimal chiral transformation
which carries the information about the gauge field for a = constant in eqn
(5.111).
On the other hand, one could in principle expand the fermionic variables as
in terms of a complete set of eigenfunctions of the hermitian operator
This expansion diagonalizes the free Dirac operator
and in this sense this expansion is regarded as corresponding to the interaction

representation in the operator formalism. In this expansion, the Jacobian for the
chiral transformation is given by
which carries no information about the gauge field for a = constant unlike
eqn (5.115), and thus we cannot assign a physical significance to this expression.
If one uses the index theorem
but
one learns that a smooth unitary transformation from non-zero integer to 0 does
not exist.18 The naive unitary transformation from the exact basis set {<pn(x)}
to the basis set {£«(#)} for the interaction picture docs not hold in general.
This analysis shows that one needs to use a complete set of basis vectors
which diagonalizK the Dime action precisely to define exact identities. In our
formulation of the quantum anomaly, we used the unitary transformation to the
free field basis only for the well-regularized operator (5.68)
which is also the basic procedure in the proof of the index theorem in math-
ematics by using the heat kernel. In the expression with the free field basis
in eqn (5.122), the effect of the index is recovered from high-frequency sectors,
which explains why the quantum anomaly was closely related to the treatment of
ultraviolet divergences in perturbation theory. WTe emphasize that the quantum
breaking of s}'mmetry (namely, the quantum anomaly) itself is perfectly finite
and independent of divergences.
5.6.2 Nambu-Goldstone theorem

We next discuss an issue related to the quantum anomaly and the Nambu-
Goldstone theorem. The main purpose of this analysis is to show that the
Nambu-Goldstone bosons do not generally appear for a spontaneously broken
symmetry if the relevant global symmetry is broken by the effects of the anomaly
18
To define a unitary transformation, one needs to truncate the ultraviolet components of
the action in a gauge invariant way.
NAMBU-GOLDSTONE THEOREM 89
and iiistantons. We discuss this problem by using the Lagrangian of QCD. The
path integral of QCD is given by
and the gauge group involved is SU(3). We use the 9 vacuum \0) which incorpo-
rates the instanton effects. In the actual QCD we need to think of at least three
quarks u, d, s, but the essence of the U(l) problem (or the rf problem) related to
the competition between the index and the spontaneous symmetry breakdown
is analyzed by a theory with a single flavor of quark, and thus we discuss this
problem on the basis of the simplified model.
If one considers the special case with the vanishing quark mass m = 0,
the action is invariant under the global chiral transformation with a constant
parameter a
To derive a chiral identity based on this change of variables in a reliable way, we

generalize the parameter to be space-time dependent a(x) and use the following
identity (namely, the integral itself is independent of the naming of path integral
variables)
We wrote the path integral for the operator insertion

for the convenience of the following discussion. If we treat a(x) as an infinitesimal
parameter in this identity
we then have the changes of the action and the path integral measure
If one collects the terms linear in a(x) and functionally differentiates the resulting
expression with respect to a(x), one obtains the identity (we here simplify the
notation \d,-oo) and (#,+oo by \9) and (9\, respectively)
We also used
in terms of the Chern-Simons form K^.

We next analyze the spontaneous symmetry breakdown of chiral symmetry.
We start with a global chiral transformation by setting a(x) to be a constant in
the transformation law oitp(yYfflti}(y) in eqn (5.126) by assuming for the moment
that, there is no anomaly. The invariance of the action and the measure then leads
to the relation
The chiral invariance of the vacuum implies that the left-hand side and the first
term on the right-hand side are equal, as is confirmed by introducing a formal
unitary transformation, ijj' = U(a)-ij}U*< (a), ib' = V"(a}^!}^(a), and U^(a)\6) —
\0). Consequently, the second term on the right-hand side vanishes. On the other
hand, the spontaneous breakdown of chiral symmetry implies the rion-invariance
of the vacuum and thus
where we used the translational invaria,nce of the vacuum. The spontaneous sym-
metry breakdown corresponds to a second-order phase transition in condensed
matter theory, and it is generally characterized by an order parameter. The order
parameter of the spontaneous symmetry breakdown of chiral symmetry is thus
given by (<?|'0(0)</HO)|#}.
Without the presence of the quantum anomaly, we have a naive local identity
and the Fourier transform of this expression
should hold even for p^ —i 0. We thus conclude that the amplitude on the left-
hand side of eqn (5.133) contains a pole 1/p2, which fact is expressed in symbolic
notation by
with a constant fv. The axial-vector current •07A'75^) thus contains a massless
(pseudo-)scalar particle represented by the pole 1/p2- This particle is called a
NAMBU-GOLDSTONE THEOREM 91
Nambu Goldstone particle (the Nambu Goldstone theorem). The Nambu Goldstone
particle arises from the mismatch between the global symmetry and the global
limit of the localized WT identity. The localized identity holds regardless of the
presence or absence of the spontaneous symmetry breakdown.
On the other hand, if one integrates both sides of the actual chiral identity
with the chiral anomaly (5.128) by multiplying by clp^""y^ and takes the limit
pu —> 0, one obtains
by using the translational invariance. We assume that the non-vanishing order

parameter persists with or without the anomaly. For the non-vanishing instanton
number, the second term on the left-hand side does not vanish and thus the
presence of a rnassless particle in the operator (V'T^Ts^/OCp) 'IS n°t generally
concluded. If one defines a modified axial-vector current by
which satisfi.es in the operator notation
it appears that the Nambu-Goldstone theorem for the current JA' formally holds.
However this current J^ is not gauge invariant since the Chem-Sirrioiis form K1''
is not invariant under the SU(3) gauge transformation of QCD and thus a gauge
invariant physical particle for JM is not defined.
One thus concludes that the Nambu—Goldstone theorem does not hold in
general when the global chiral symmetry is broken by the effects of the chiral
anomaly and instantons. This is regarded as a reason why we do not observe a
light pseudo-scalar meson rf in actual QCD.
6
QUANTUM BREAKING OF GAUGE SYMMETRY
In this chapter we discuss the quantum anomalies which appear in gauge symme-
try itself. This class of anomalies generally appear in gauge theory where both of
the vector and axial-vector couplings of gauge fields co-exist. As a special case,
the chiral gauge theory where parity is broken is important in physical applica-
tions. The quantum breaking of gauge symmetry implies that the conservation of
probability (unitarity) fails and one cannot calculate higher-order quantum cor-
rections in a consistent manner. The absence of the quantum anomaly in gauge
symmetry is the basic requirement of consistent gauge theory, and the require-
ment imposes a stringent condition on the allowed lepton and quark multiplets
in the Weiiiberg-Salarri theory. This class of anomalies also has important appli-
cations to the study of. for example, non-linear a models. In this connection, an
interesting constraint called the integrability (or the Wess--Zumi.no consistency)
condition appears. The actual evaluation of the covariant form of anomalies,
which do not satisfy the integrability condition but have a gauge covariant form,
is much easier. Consequently, a conversion prescription of the covariant form of
anomalies to the integrable form of anomalies becomes important. In the context
of the current algebra, it is explained that the quantum anomalies generally lead
to anomalous commutation relations.
6.1 Gauge theory with axial-vector gauge fields

We analyze the Euclidean path integral
where the Dirac matrix convention is the same as in QED and QCD, and 7^ is
chosen to be anti-hermitian
The 75 matrix which satisfies
is herrnitiaii. The covariant derivative D^ is defined by
92
GAUGE THEORY WITH AXIAL-VECTOR GAUGE FIELDS 93
by using the generators Ta of a non-Abelian gauge group
The non-Abelian gauge field has the same number of components as the gen-
erators of the gauge group. To simplify the notation, we include the coupling
constant into the gauge fields, for example, gV£ —>• V®. We adopt the convention
to sum over the indices which appear twice in the same expression. We also often
write Vf, = V£Ta and .4M = A°Ta as in eqn (6.4) above.
In this chapter, the vector-like gauge field, which has been discussed so far
in this book, is denoted by V£ and the axial-vector gauge field, which appears
together with Dirac's 75 matrix, is denoted by A"^. The field strength tensors
V£v and A.^v, which are generalizations of the electric and magnetic fields E and
B, are defined by
The fcrmion action is invariant under the following gauge transformations
The last equation is also written as
where we defined
We have the gauge parameter with the 75 matrix in addition to the conventional
gauge transformation in these equations, arid thus a generalization of gauge
transformations.
When one considers the Dirac operator in Euclidean theory, the operator
with both of the vector and axial-vector fields V'M and A^
does not define a hermitian operator with respect to the following inner product
94 QUANTUM BREAKING OF GAUGE SYMMETRY
since the signature of the axial-vector field AIL changes. If one rotates the axial-
vector field AH to an imaginary field as
the operator
becomes hermitian
However, as a price for making the axial-vector field A^ pure imaginary, the axial
part of the gauge transformation is spoiled. We define a complete set of basis
functions
by using the hermitian operator ]j) (6.13), and expand the fermionic fields as
The path integral is then written in the form
which can be exactly (though formally) integrated. The Jacobiaii for the gauge
transformations (6.7)
is then given by
GAUGE THEORY WITH AXIAL- VECTOR GAUGE FIELDS 95
and the regularized Jacobian (namely, the quantum anomaly) is given by the
master formula
This formula of the Jacobian shows that the quantum anomaly associated
with the conventional gauge transformations parametrized by a(x) vanishes in
Che present method of calculation, and the gauge symmetry for the axial-vector
gauge field parametrized by /3(x)"/s is spoiled by the quantum anomaly. The
explicit evaluation of the Jacobian (6.20) proceeds by using the complete set
of plane waves just as in the case of the previous chapter. The result after a
somewhat lengthy calculation is given by (with the convention e 1234 = I)19
by rotating back to the original field A^ —» —iA^ after performing the calcula-
tion.
In the actual calculation of the Jacobian, one encounters the term
multiplied by the cut-off parameter Af 2 , but this term is cancelled by a gauge

transformation of a local counter-term M2AaÂ^ with a suitable coefficient Af 2 .
If one adds this counter-term to the original Lagrangian in eqn (6.1)
the variation of the local counter-term under eqn (6.7) cancels the part of the
Jacobian (6.22). In the analysis of the quantum anomaly, those terms in the Ja-
cobiaii which, can be cancelled by suitable local counter-terms in the Lagrangian
are generally regarded not to be a genuine anomaly. This definition of the quan-
tum anomaly is important in the analysis of quantum anomalies associated with
gauge symmetry.20
19
One may use a simple regulator function f(x) — e x. Unfortunately, this calculation is not
systematic and is very complicated compared to the calculations in Chapter 5. We shall later
present a simpler calculational method which gives rise to the covariant form of anomalies.
We shall then give a method to convert the covariant form of anomalies into the anomalies
equivalent to eqn (6.21).
20
To be precise, one says that the theory is anomaly-free if the main part of the anomaly
factor in cqn (6.21) is absent. In the presence of the main part of the anomaly, it is more general
to retain the extra term (6.22) in the Jacobian (6.21) with an arbitrary finite coefficient as
part of the anomaly by adjusting M2 suitably.
6.2 Pauli—Villars regularization

The calculatiorial scheme we described above corresponds to perturbative calcu-
lations by using the Pauli-Villars regularization. When the quantum anomaly
appears in the gauge symmetry itself, the gauge theory becomes inconsistent and
thus no non-perturbativc calculation exists. The Pauli Villars regularization is
defined in the path integral by
where the mass M of the regulator field </>(x), which was introduced to make
the calculation well defined, is set to infinity after the calculation. The field <fi(x)
is a Bose-like (namely, defined by an ordinary complex number instead of the
Grassrriaiiu number and thus unphysical) spinor field. Consequently, under the
gauge transformations
the same Jacobian factors appear in the denominator and numerator correspond-
ing, respectively, to the field variables ib(x) and (f>(x), and thus the Jacobians
completely cancel each other. We here recall that the Jacobian appears in the
denominator for the Grassmann number for which the integral is defined by the
(left) derivative. The regularized path integral thus does not produce the quan-
tum anomaly manifested as a Jacobian. However, the action for the Pauli-Villars
regulator field
contains a mass term M, and the large limit of this mass term defines a phys-
ical theory. In the present gauge theory, the mass term of the regulator field
M<j>(x)(p(x) is not invariant under the axial gauge transformation cxp[i/?a (x)Ta^\
and thus the gauge symmetry is explicitly broken. This gauge symmetry break-
ing in the limit M —?> oc induces a strong breaking of gauge symmetry, which is
a definition of the quantum anomaly in the context of the Pauli-Villars regular-
ization.
From the viewpoint of Ward-Takahashi identities, the change of variables in
the path integral with a = 0 in the gauge transformations (6.25) gives rise to
the identity (which is a generalization of the fact that the definite integral does
not depend on the naming of integration variables f dx f(x) = f dy f(y))
PAULI-VILLARS REGULARIZATION 97
In the last line we used the fact that no Jacobian appears in the path integral
measure. The action in the last expression is explicitly written for an infinitesimal
as
When one uses this expression in the above identity (6.27) arid collects only the
terms linear in j3(x), the identity is written as
The covariant derivative here is defined by
and also we define
for the general composite operator O(x). The left-hand side of the identity (6.29)
defines a regularized axial-vector current, and it is confirmed that the mass term
on the right-hand side gives the quantum anomaly in the limit M —> oo (with
the convention e1234 = I) 21
which reproduces the anomalous identity derived in the previous section.

Since the regularization is introduced simply to avoid ambiguities in the ac-
tual perturbativc calculations, it is natural that the final result agrees with the
21
One calculates all the one-loop diagrams which contain the operator 2iMip(x)^sTa<i>(x') at
one of the vertices and the gauge fields at the remaining vertices. The diagrams which do not
vanish in the limit M —> oo give the result in cqn (6.32). This calculation was first performed
bv W. Bardeen.
98 QUANTUM BREAKING OP GAUGE SYMMETRY
one obtained by a different method. To see what is going on in the Pauli-Villars

regularization in a more intuitive way, we write an identity which is based on
the transformation of only the path integral variables ip(x) and ijj(x)
and an identity based on the transformation of only the path integral variables
and
When one adds these two identities, one obtains the result of the Pauli-Villars
regularization (6.29), which avoids ambiguities in the intermediate stages of per-
turbative calculations associated with the axial-vector current. Here we wrote
the anomaly which arises from the Jacobian in a symbolic notation In J. The
change of the signature of In J between the two equations above is a result of
the fact that the Jacobian appears in the denominator and the numerator for
the Grassmann number and the ordinary number, respectively. Note that both
sides of eqn (6.34) approach zero in the limit M —>• oo.
The path integral method discussed in Section 6.1 used a regularization which
cuts off the eigenvalues of the operator Tjb (6.13), and as we explained there this
regularization generally spoils axial gauge symmetry though it preserves vector
gauge symmetry. This property is shared with the Pauli-Villars regularization.
6.3 Chiral gauge theory and the quantum anomaly

The quantum anomaly discussed so far may be called V-A-type, but from the
viewpoint of gauge theory it is natural to separate the left component ipL (x) =
[(I - ~f5)/2]il;(x) arid the right component WR(X) = [(I + 7 5 )/2]ii(x) and rewrite
the covariant derivative as
and the action as
Since we have
and thus (1 ± 7s)/2 are projection operators, one can treat the left and right
components of the gauge field as independent components.
CHIRAL GAUGE THEORY AND THE QUANTUM ANOMALY 99
In this separation, the theory with only the left components, for example,
is called a chiral gauge theory, and the action is invariant under the gauge trans-
formations
The path integral measure is not invariant under these transformations. A way
to understand this non-invariance is to introduce a dummy fermionic component
ij)H into the path integral (6.38), which does not interact with the gauge field,
and one recognizes that this theory is derived from the V-A-type theory we
discussed already by setting V^ = —A^ — L^/2 in eqn (6.4). Consequently, the
quantum anomalies are also derived by the same replacement in eqn (6.21). If
one considers the gauge transformation of the fermionic variables with gauge
field kept fixed, one obtains the anomalous identity
where we defined
The theory where the gauge field couples only to the right-handed fermion
tpn is treated in a similar way by setting Vlt = A^ — R^/2 in the V-A-typc
theory in eqn (6.21), and one obtains
6.3.1 Quantum theory of chiral gauge theory

We study the quantum theory of chiral gauge theory defined by
to understand the physical implications of quantum anomalies obtained above

for the chiral theory. As a gauge condition we set the time components of all the
gauge fields to be 0 (which is also called the Weyl gauge condition)
From the equation of motion for L^(x) we obtain the constraint
related to the Gauss law
where a non-Abeliaii generalization of the electric field in Maxwell theory is

denned by
The indices k which run over 1 ~ 3 stand for the spatial components.
The action with the condition Lg (x) = 0 still contains a gauge freedom which
does not depend on time
and the Gauss law operator is a generator (Nothcr current) of such a gauge trans-
formation. This fact is understood by considering a time-dependent parameter
u } a ( t , x ) instead of the time-independent gauge parameter uja in eqn (6.49) in the
action
which shows that Ga(x) is the Neither current for the time-independent gauge
transformation.
When one considers the above time-dependent gauge transformation uja(t, x)
in the path integral (by using the fact that the path integral measure for the
gauge field is gauge invariant)
CHIRAL GAUGE THEORY AND THE QUANTUM ANOMALY 101
one obtains an identity by including the quantum anomaly (6.41) arising from
the Jacobian
where we used e1230 = 1 in the Minkowski metric. This equation is written in

an operator notation by using the Hamiltoriiari H derived from the Lagrangian
density C in eqri (6.44) as
for the theory with the gauge anomaly. Namely, the Gauss operator Ga(x) no
longer commutes with the Hamiltonian H due to the effects of the quantum
anomaly, and the Hamiltonian H is no longer gauge invariant.
This fact shows that the basic two equations in the Schrodinger functional
representation of quantum field theory
are no longer compatible with each other since [H,Ga(x]] ^ 0. Even if one
starts with a physical state which satisfies G°* = 0 at t — 0, the physical
condition Ga^> = 0 does not hold after a time development to t / 0. The Gauss
law operator Ga, which generates the time-independent gauge transformation,
eliminates the unphysical longitudinal component of the gauge field Lak(x) when
applied as a constraint on the wave functional G°*(i;L^(x),î(x)) = 0. The
failure of the Gauss constraint by the quantum anomaly means that a consistent
unitary theory with only the transverse physical components for the quantized
gauge field is not constructed.
When one adopts a general Lorentz covariaut gauge condition, the constraint
on the state vector (in the Fock representation)
becomes fundamental. For example, a general covariant gauge fixing of eqri (6.44)
is invariant under the BRST symmetry (sec cqn (3.51))
Here ca(x) and ca(x) stand for the non-Abelian generalization of the Faddeev-
Popov ghost fields, which were briefly explained in Section 3.2, and A is a con-
stant Grassmann number; i\ca(x) replaces the non-Abelian gauge parameter.
The BRST current is derived by a localized transformation with \(x) and the
variation of the action 5 J d^x £eff = / d^x 9/i A(a;) jgRST (x). The BRST charge is
then delined by QBRST = / d3x JBRST^^)- ^n a theory with the quantum anomaly
in gauge symmetry, it is shown by considering a localized BR.ST transformation
that we have
and thus the BRST charge QBRST ceases to be a conserved quantity, and the
constraint on the state vector QBR.ST|X!'} = 0 becomes incompatible with the
time development.
We thus conclude that the quantum anomalies in gauge symmetry need to be
cancelled among fermionic variables appearing in the theory to define a consistent
theory of quantized chiral gauge theory. This gives a stringent constraint on the
allowed fermionic contents, such as the number of fields or the representations of
the gauge group, in the Weinberg-Salam theory which breaks parity symmetry.
The cancellation of the quantum anomaly in the actual Standard Model, which
incorporates the Weinberg Salam theory, is explained later.
6.4 Covariant anomaly

Among the methods to evaluate quantum anomalies in gauge symmetry explic-
itly, in addition to the method already explained which is related to the Pauli-
Villars regularization, a method which gives a gauge covariant form of anomalies
is known and the anomalies thus obtained are known as "covariant anomalies."
We explain the method for chiral gauge theory
We first note that the Dirac operator Jfii = ^(d,,, - «£ M )(1 - 7o)/2 appearing
in the Euclidean path integral is not hermitian
namely, the left-handed system characterized by (1 - 75)72 changes to a right-

handed system. In this case, the operator
COVARJANT ANOMALY 103
is confirmed to be hermitian and positive semi-definite. Consequently, one can

define a complete orthonormal system
where A n is a real non-negative number. The 1 : 1 correspondence of vanishing

eigenvalues of $>L and JPL^>L is confirmed as follows
Similarly, one can define a complete system by using the hermitian ]/)L^L
and confirm the 1 : 1 correspondence of vanishing eigenvalues of If)L and ^*L^| •

From the relation
one can define a correctly normalized eigenfuncliori ipn by
by choosing Xn > 0 for \n / 0. This shows that the two operators $>L$>L and
J/>r,J/>L share exactly the same number of non-vanishing eigenvalues.
By using the basis sets {ipn(x)} and {ipn(x)} thus defined, one may expand
the fermionic variables as22
and the fcrmion action is formally diagonalized as

22
In the framework of the covariant anomaly, one can show that the anomalous Jacobian
is independent of the choice of path integral variables, namely, either T/J£ and V'r. or y;L and
1(1 *L. In Lhc latter choice, i/ij, is defined by the operator ("tof)L)^"fof>L = W^L and 1/4 is
denned by the operator 70-Pr,(70^1,)^ = 7o^L-PJ,7o- One can then confirm that the identical
(regularized) Jacobian is obtained by either choice of path integral variables.
6.4.1 Fermion number anomaly

As an application of the present definition of the path integral, we first consider
the phase transformation related to the fermion number by using a local param-
eter a(x) (we thus discuss the non-gauge freedom for the moment) in eqn (6.59)
and then the path integral measure is transformed as
A more detailed account of the appearance of this Jacobian is given soon. This
Jacobian is evaluated by replacing the mode cut-off by the cut-off in terms of
the eigenvalues as
where we used the property of the projection operators (I ± 7s)/2
The explicit evaluation of the Jacobian (6.71) is the same as the calculations
in Chapter 5, and one finally obtains
COVARIANT ANOMALY 105
This result when combined with the Nother current for eqn (6.69) gives
which shows that the fermion number operator contains a quantum anomaly.
This was first shown by G. 't Hooft.
To understand the meaning of this anomaly, we write the path integral mea-
sure more explicitly as2s
Here we consider the general case where $>L^L and $\$L, respectively, have n0
and mo vanishing eigenvalues. This property is generally expressed as
where dimkerî, for example, expresses the number (dimension) of the normal-
izable 0 eigcnmodes (kernel) of the operator J/>i. and the above difference of zero
modes defines an important notion of index.
The index relation here has the same form as the index relation we discussed
in connection with the photon phase operator in Chapter 3, if one identifies the
creation operator of the photon with a^ = Jfii.
We discuss the meaning of the index by noting 75$) + ^75 = 0. In general
(we analyze the eigenvalues of J/) = 7M[c^ — igL^(x}])
and thus the 0 eigenmodes of ]}) can be chosen as the simultaneous eigenstates
of 75, [(l±7 B )/2]^ n (.T). Also
and the index (6.76) is expressed by using the basis set $>'il>n(x) = Xn^->n(x) as
23
If one denotes 4>n(x) — {x[<t>n), V^C1) = (<Pn\x}; îe inverse of the phase factor
det(x\<f>n)det(ipn\x) = J({4>n}, {f-n})^1 appears as the Jacobian. This phase factor is crucial
to understand the quantum anomaly in gauge symmetry from the viewpoint of the variation of
the path integral under a gauge transformation of the ga.uge field, since \n is gauge invariant.
This phase factor can be ignored when one expands ijj and •f in terms of the same basis set as
in the case of vector-like theory such as QKD or QCD, or V-A-type theory with hermitian f>.
where we defined HO = n+, m® = H_. As was explained in Section 5.5, this

index agrees with the Pontryagin index carried by the gauge field (by using the
notation including the coupling constant into the field variables)
which is known as the Atiyah-Singer index theorem.

Coming back to the path integral of chiral gauge theory, the path integral
measure is transformed for a global constant fermion number a in eqn (6.70) as
where we have a difference of zero modes, since {<fn} and {</>«} appear in 1 : 1
correspondence for the non-vanishing eigenvalues. In chiral gauge theory the
fermion number is not conserved due to the quantum anomaly in the presence
of the non-vanishing Pontryagiii index of the gauge field (i.e., instantoiis). In a
theory such as the Weinberg Salam theory, where the quarks and leptons couple
to the same Yang-Mills field (which is called the TV-boson), it is shown that
the conservation of (mark number B/3 and lepton number L is simultaneously
broken by the effects of the quantum anomaly, and only the difference B — L is
preserved.
In terms of Green's functions for the theory (6.59), we have
which shows that np number of tyi and rno number of 'tpi are absorbed into the
vacuum in the presence of the general iustanton solution. Since the integral over
the Grassmanri numbers is defined by the (left) derivative and the Grassmann
coefficients of the expression in eqn (6.67) corresponding to the zero modes do
not appear in the action of the path integral, we need to insert at least the
above number of ipL and 'I/JL into the integrand to obtain the non-vanishing
path integral. If one inserts more I/JL and -I/JL, one obtains a generalization of
conventional Green's functions.
COVARIANT ANOMALY 107
6.4.2 Covariant anomaly in gauge transformations

We next examine the gauge transformation of fenriioii variables defined by
The path integral measure is transformed by using eqn (6.67) as
with a(x) = aa(x)Ta, and the Jacobian is evaluated just as in eqn (6.71)
This quantum anomaly for the gauge transformation is called the covariant
anomaly because of its gauge covariant appearance.
We show that this covariant anomaly is associated with a specific regulariza-
tion of the source current of the gauge field. A regularized current which gives
rise to this covariant anomaly is defined by using the complete set
and the construction of covariantly regularized current discussed in Chapter 4

as
When we have \n = 0 we need to control the infrared divergence by using,

for example, the Higgs mechanism to be discussed in the next section without
changing the quantum anomaly. The quantum anomaly itself is free of infrared
divergences as is shown shortly. The gauge transformation is written as
with oj(x.) = (jja(x)Ta. By noting the gauge invariance of the eigenvalues in

eqn (6.86), we have a covariant form of transformation law of the current
for an infinitesimal w. The quantum anomaly is given by using e1234 = 1
which agrees with eqn (6.85). In fact this identity is derived by considering the
transformations (6.83) in the path integral of the chiral gauge theory (6.59) and
by using the Jacobian (6.85). This shows that the calculational method of the
Jacobian for the covariant anomaly is understood as a covariant regularization
of the current.
This result (6.90) shows that gauge symmetry is broken by the quantum
effects. Conversely, if all the quantum anomalies (6.85) thus evaluated are can-
celled, the chiral gauge theory is consistent even in the quantum level. If one
recalls that the quantum anomaly (6.85) is symmetric under the replacement of
F^vTb and F^T0, a formulation of quantized chiral theory is consistent if the
following relation holds24
It is known that the orthogonal gauge group SO(ri) with n > 2 and n ^ 6,
for example, leads to anomaly-free gauge theory. As an explicit example, if one
chooses the group SO(3) or locally equivalent SU(2), it is confirmed that all
24
From the viewpoint of Feynman diagrams, Ihc covariant anomaly corresponds to a calcu-
lational scheme where one collects all the effects of the anomaly to a specific current vertex
while imposing the ga,uge invariance on all the remaining vertices coupled to the gauge field. I
is thus clear thai chiral gauge theory is consistent if the covariant form of anomalies all cancel.
In the case of the gauge group SU(2), one needa to include the analysis of the globa.1 anomaly,
which is briefly mentioned in Chapter 11.
ANOMALY CANCELLATION IN WEINBERG-SALAM THEORY 109
the anomalies cancel by using the properties of the Pauli matrices which satisfy
eqn (6.91). A simple way to understand the absence of anomalies in the realistic
Weinberg Salain theory may be to use the fact that all the quarks and leptons
are precisely accommodated in a representation of the anomaly-free gauge group
SO (10) in a grand unified model.
In the general case where the gauge field couples to the left-handed fermion
field with a representation T£ and the right-handed fermion field with a repre-
sentation Tft, the above anomaly cancellation condition is written as
For example, the vector-like theory, where the gauge field couples to the left-
handed and right-handed fermion fields with an identical representation, defines
a consistent theory since the quantum anomaly is automatically cancelled. The
quantum electrodynamics and QCD belong to vector-like theory.
6.5 Anomaly cancellation in Weinberg—Salain theory

We examine the anomaly cancellation in the Weinberg Salam theory which is
based on the gauge group SU(2)^ x C7(l)y. When we denote the generators of
the gauge group SU(2)i x U(l)y by (I\. T2, T3) and Y, respectively, the electric
charge of fermions satisfies the Gell-Mann-Nishijima-type relation
When we write a doublet of quarks as
the Lagrangian for the quarks is given by
If one recalls that Ta = (l/2)r a is written in terms of Pauli matrices, the gener-
ator of U(l) is written as
by remembering that the electric charges of quarks (u,d) are (2/3, —1/3).
Consequently, the formula (6.90) for the quantum anomalies gives the anom-
aly for SU(2) gauge transformations (by using the notation W^ = W^Ta)
arid for U(l) gauge transformations
We multiplied all the expressions by the factor 3 since three quarks appear for
each case due to the color degrees of freedom.
Similarly, for the doublet of leptons
(by writing the right-handed component of the neutrino explicitly, though it does
not couple to gauge fields), the U(l) charge is given by
by recalling the electric charge (0. —1) of leptons (v,e). The Lagrangian for the
leptons is written as
Consequently, the leptons induce the quantum anomalies for the SU(2) gauge
transformations
and for the U(l) gauge transformations
The gauge field belonging to SU(2) docs not contribute to the anomaly for
the SU(2) gauge transformations, since SU(2) is an anomaly-free group, arid the
properties trT" — 0 and 3F/ + Y'L — 0. which are confirmed by using the explicit
expressions in eqn (6.96) and eqn (6.100), ensure the absence of the anomaly for
all the SU(2) gauge transformations. The remaining anomaly for the U(l) gauge
transformations cancels if one uses
The Weinberg-Salam theory thus defines a consistent quantum theory if one

considers quarks and leptons together.
ANOMALY CANCELLATION IN WEINBERG SALAM THEORY 111
6.5.1 Effects of Higgs particles

As an application of the covariant anomalies, we show that the Higgs (or scalar)
particles do not contribute to the quantum anomalies. We analyze this issue by
using the following Lagrangian
where 4>(x] stands for the Higgs field (or in general, a scalar field). Note that
the operator 0r, = 7''(<9;j ~ i/î), for example, docs not contain the projection
operator (1 — 7s)/2. By noting that the left- and right-handed fermions are
coupled to each other mediated by the Higgs field, we use the following general
operator
In Euclidean theory we expand the field ?/; in terms of the complete set of eigen-
functions of 'D~'D and the field '</.> in terms of the complete set of 'D'D~ in the
calculational scheme of covariant anomalies.
The quantum anomaly for the gauge transformations of left-handed compo-
nents, for example.
is given by the master formula
We now use the following relation
The terms linear in the Higgs field, for example.

show that the differential operator acts on the Higgs field only and does not act
on fields coming to the right of it. As a result, in the expression with the plane
wave basis of the master formula
the terms with the Higgs field is always expanded in powers of 1/M2, when
one moves c.'lkx through the differential operators in, for example, TjD\ and then
rescales as k^ —> Mk^ and expands all the terms in the exponential in powers of
1/M.
Also, the terms linear in the Higgs field convert the right-handed components
to left-handed ones and vice versa, and thus only the even powers in these terms
contribute to the master formula.2''1 For this reason, the identical form of terms
appear in both of the anomaly factors coming from ip and ip in eqn (6.108), and
consequently we take the trace over these terms together with 75. For example,
in d = 4 space-time, we have
which vanishes since it contains only two 7^ matrices.

The master formula eqn (6.108), after eliminating the terms odd in powers
of the Higgs field in eqn (6.109), thus gives
One can confirm that those terms with the Higgs field in eqn (6.113), which do
not contain Dime's 7 matrices, do not contribute to the anomaly if one performs
the same calculation as in the previous section.
6.6 The Wess Zuinino integrability condition

We examine the following path integral by setting A/t(x) = A®(x)Ta
The covariant current defined in eqn (6.87) with f) = 7^(9^ — iA^)
as was already explained, contains the covariant form of anomaly

25
This fact is seen by writing the trace over the 7^ matrices in the first term in eqn (6J 11),
for example, as tr[(l - 7o)/2] cxp(-£>t?3/Af 2 )[(l - 75)72].
WESS-ZUMINO INTEGRABILITY CONDITION 113
In the applications of quantum anomalies, the anomalies for a finite transfor-

mation often become important. We thus have to superpose the Jacobian for an
infinitesimal transformation to realize a finite transformation. In this analysis,
we make the following correspondence between the Jacobian and the transfor-
mation of the fermionic variable for an infinitesimal parameter a(x) — aa(x)Ta
as
The composition law eta^x'e^x'e ia(x>e IP(X> = exp(—[a,/?]) of group theory

then imposes the following integrability condition on the Jacobian
The first term on the left-hand side is given by the Jacobian evaluated for the pa-
rameter /? by using the gauge field, which is a result of the gauge transformation
parametrized by —a, and the second term is given by the Jacobian obtained by a
reversed order of calculation. For the above covariant anomaly (6.116), however,
because of the gauge covariance of F^ we obtain twice
on the left-hand side, namely, the integrability condition is spoiled by a factor

of 2.
On the other hand, it is shown that the anomaly related to the Pauli Villars
regularization discussed in Sections 2 and 3 of this chapter satisfies the integra-
bility condition (the Wess-Zumino condition), after a somewhat lengthy calcu-
lation. The Wess-Zumino condition becomes important when the gauge theory
is not consistent due to the presence of quantum anomalies in gauge symmetry,
and thus only the perturbative calculation with one-loop fermion diagrams be-
comes relevant. The Wess-Zumino condition is then shown to correspond to the
Bose symmetry with respect to the vertices of the one-loop Feynman diagrams.
The covariant anomaly, which collects the effects of the anomaly to only one
of the vertices, does not satisfy the Bose symmetry and thus the integrability
condition. In an actual explicit calculation of anomalies in arbitrary space-time
dimensions d (which in general includes the gravitational field) . one can calculate
only the covariant form of anomalies. It is therefore important to give a prescrip-
tion 'which converts the covariant anomalies to intcgrable anomalies which satisfy
the Wess -Zumino condition.26 There are several ways to do this conversion, and
we give here a method which uses a current to define a result of the path integral
(or the partition function) in the presence of the background gauge field.
26
This anomaly is also called a consistent anomaly, but it is clear that the covariant anomaly
is not "inconsistent."
If one replaces the gauge field as A^ —> sA^ by introducing the parameter s
in cqn (6.114), the quantity defined by W(sAp,) = lnZ(sA/t) satisfies
We thus define the regularized path integral by means of the covariantly regu-
larized current J°''(x-) (6.115) by27
The current Jap'(x} is a finite quantity and thus W(Ali} is also finite.
If one regards the finite W(A^) as a functional of Ap and expands it in powers
of AH, the expansion naturally satisfies the Bose symmetry. Consequently, the
current jav(x] defined by
satisfies the Wess Zurnino condition. From its definition, we have
The comparison to eqn (6.121) shows that W thus defined docs not, depend on
the class of currents we use to define it.
To be more explicit, the current ja"(x) is given by adopting the notation
Ja»(x.s) = Ja>l(x,sA
which combined with the fact that the current Jal*(x,s) depends on s only
through the combination sA
finally gives
The last term in this expression is a manifestation of the fact that the current
J afi (x) does not satisfy the integrability condition (if it is written in a form
27l
['he imaginary part, of eqn (6.121) contains interesting information about the quantity
called the '/-/-invariant.
Ja^(x) = 5W/6A^(x), the last term vanishes) arid at the same time it directly
depends on the quantum anomaly, since in the absence of the anomaly, these
two currents should coincide.28
In the following, we would like to explain the evaluation of the last term in
eqn (6.126) above,29 by using the regulator function f ( x ) = e~x in eqn (6.115)
as an example. We first consider the case s = 1. If one defines
the "curvature" 5Jb"(y]/5A^1(x} — (a -B- b, p.• O v, x •<-?• y) is obtained from the

combination
by identifying the subscripts as 1 —> (a,fj,,x) and 2 —> (b,v,y). In deriving the
above expression, we used the identity
valid for a general operator X. By noting the relation
and the cyclic property of the trace Tr, it can be confirmed that all the non-local
terms in eqn (6.128) which contain 1/0 cancel each other after integration by-
parts with respect to a. Thus we have
28
In the absence of gauge anomalies, one can impose the gauge invariance on all the vertices
in the one-loop Feynman diagram without spoiling the Rose symmetry. In fact, this is the only
sensible calculation and thus these two currents coincide.
29
Rcadcrs who are interested in the final formula may proceed directly to eqn (6.135).
where we have noted that the terms which do not contain 75 vanish under the
anti-symmetrization of 1 4-» 2. From this expression, we have (by recalling that
In writing the last expression, we discarded the commutator {j",e ^ a^ /M"]

and the anti-commutator {7^,7"} = 2<?''" because these terms give rise to van-
ishing contributions in the limit of large M.30 Then by recalling the basic prop-
erty (5.69) and comparing the above expression with the covariant anomaly in
eqn (6.90) and eqn (6.116) with f ( x ) = e x, we arrive at the expression
by noting eqn (6.129). It can be shown that this relation holds for a general
regulator function /(x). 31
If one uses SJbv(y)/5A°(x) - 5Ja»(x)l5Abv(y) thus obtained in eqn (6.126)
(the index s indicating that sA™ is used) one obtains
where Str indicates that we take a trace of the expression which is arranged to
be symmetric with respect to the generators of the gauge group. The quantum
anomaly is then given by the covariant divergence of this expression'32
30
Powers in 1/M do not change but the number of 7 matrices is reduced by two in these
operations.
31
To show this, we introduce the Laplace transformation
and the result for a general regulator function f(x) is obtained by multiplying
liniM->oo f£° dpg(p) by the above evaluation in eqn (6.132) with the scaled regulator
mass M2 —5- M 2 /p. The terms which survive in the large M limit are independent of M and p,
and we obtain the result identical to the case f(x) = e x, because f£° dp g(p) = /(O) = 1'.
32
Quantities inside brackets { and } are treated as a block, namely, they are not subject to
svmnietrization under Str.
which gives a formula to derive the integrable anomaly (6.41) starting with
the covariant anomaly (6.116). In this derivation we used the Jacob! identity
^vafi\^^Âa,Ao^ = 0 and it is important to recognize that the symmetrized
trace Str is used.
As a special example, we obtain
for Abclian gauge theory, and the factor 1/3 is the Bose symmetrization factor
of the triangle Feynman diagram.
It is clear that these formulas are generalized to arbitrary even d = 2n di-
mensional space-time, and by using the covariant, anomaly, which is derived in a
manner identical to the U(l)-type anomaly (5.75) in Chapter 5,
one can write the current
and the anomaly factor
As is clear from the above derivation, -4"ons = 0 is automatically satisfied if

-4£ov = 0.
6.6.1 The Wess-Zumino term
We define the Wess-Zumino term I\vz by33
33
Under an infinitesimal gauge transformation parametrized by a, SL^ = l)â, SR^ — 0,
5U = iaU, the left-hand side in eqn (6.141) is invariant and the second factor on the right-
by considering the background fields L^(x)T" = L^ and R^(x)Ta = R^ belong-

ing to the group, for example, SU(n). The factor U(x) is an element of the group
SU(n), and it is given explicitly by
We define the path integral measure in eqri (6.141) by the V-A prescription in
Section 6.1. In the defining path integral of I\yz, one may perform a change of
variable
and then the U(x) dependence is completely removed if the Jacobian is ignored.
Consequently, I\vz contains only the effects of the anomaly. The path integral
can also be written as
if one performs the vector-like transformation ib —>• Uîp. which does not contain
the anomaly as was shown in Section 6.1.
To evaluate FWZ explicitly, we introduce a real parameter s and consider the
gauge field which, after a certain amount of gauge transformation, becomes
Starting with this configuration, one performs a further infinitesimal chiral trans-
formation
and an anomaly-free vector-like transformation
alternatingly. The result of this operation is expressed as an integral over the

variable s of the anomaly appearing for the chiral transformation in eqn (6.146)
as
hand side produces the gauge anomaly by considering combined chiral and vector-like gauge
transformations of the fermionic variables. In this way, the Wess Zumino term which turns
out to be a local functional reproduces the anomaly under the same gauge transformation,
where A(L/Jl(x),Rlj(x)) stands for the integrable form of anomaly34 discussed in

Section 6.1
with the replacements
When one adds this FWZ (6.148) evaluated for the group SU(3) to the non-
linear <T model, which describes the dynamics of Nambu-Goldstone bosons -Ka(x)
as a low-energy effective theory of QCD with three flavors of quarks ( u , d , s ) ,
one can incorporate all the effects of the quantum anomalies arising from the
interactions with external fields. To take into account the color degrees of freedom
in QCD correctly, F'wz needs to be multiplied by the color factor Nc = 3.
We now discuss two applications of FWZ • Firstly, the two-photon decay 7r° ->•
77 of the neutral TT meson is described by choosing the group SU(2) and setting
An = 0. When one retains only the terms linear in TT(X) and sets
one obtains the effective interaction among the neutral pion and electromagnetic
fields after multiplying by the color factor Nc = 3 as
Here v^ stands for the electromagnetic field and FIW = 9Miv - dvv^.
To obtain the same result from a more fundamental consideration, one may
consider two flavors of quarks (u,d) with the electric charges (2e/3, —e/3). re-
spectively, and the Lagrangian
The axial-vector current corresponding to the third component of isospin sym-

metry
34
Equation (6.148) is written in the Minkowski metric with e123n = 1. The imaginary factors
i arising from the Minkowski convention of e 12 ^ 0 — l and the volume element d^x compensate,
and the expression (6.149) still gives the correct result.
is shown to contain the following chiral anomaly (by including the color degrees
of freedom Nc = 3 which is not written explicitly in the above Lagrangian)
The Nambu-Goldstone theorem in the presence of the spontaneous breakdown

of chiral symmetry states
When one combines the anomaly relation eqn (6.155) with this theorem (at
p2 ~ 0), one has
which agrees with the effective Lagrangian given by the Wess-Zumino term
(6.152). This analysis of eqii (6.157) was performed by Bell and Jackiw, and
Adler in 1969, and it is known that the result describes well the experimental
result.
As the second application, we consider the case where all the external gauge
fields vanish (L M (x) = R/j.(x) = 0) and the case of group SU(3)
where we introduced the new notation
with U(x,s) = exp{2i[7r(x)// ff ](l - s)}. When one uses <9 M t/t = -U^(dÛ)U^
which arises from U(x, s)U^(x, s) = 1 and defines a five-dimensional completely
anti-symmetric tensor by e12305 = 1 by regarding the variable s formally as the
fifth coordinate of space-time, one can write FWZ(^) in a symmetric way as
The integration domain D (which means a disc) shows 0 < s < 1. When one
expands this Tw7,(U) in powers of TT(.T), one can describe the new interactions
among the eight Nambu-Goldstone bosons induced by the quantum anomaly.
QUANTUM ANOMALIES AND ANOMALOUS COMMUTATORS 121
6.7 Quantum anomalies and anomalous commutators

Anomalous commutation relations generally appear in the theory which contains
the quantum anomalies. This is based on the fact that a quantum field theory
is defined by the commutation relations of various operators and their represen-
tations in the operator formalism, and thus if all the commutation relations are
normal we have no space to accommodate the quantum anomalies. An example
of the anomalous commutator already appeared in cqn (6.53) in Section 6.3, and
further examples arc discussed in detail in Chapter 10 later in connection with
the Kac-Moody and Virasoro algebras. In this section, we would like to briefly
explain the difference between the Goto-Imamura— Schwinger term, which ap-
pears in the absence of the quantum anomalies, and the anomalous commutators
associated with the quantum anomalies.
In quantum electrodynamics, the correlation function of the current j ^ ( x ) =
ip(x)îj}(x) is given by the calculation of the vacuum polarization tensor as (in
Minkowski metric)
which was explained in eqn (4.39) of Chapter 4.35 In the analysis of canonical
commutation relations, one implicitly assumes a theory which is defined by a
cut-off in momentum space (instead of renormalized quantities) and thus we
write the bare correlation function with an explicit cut-off parameter A.
We now apply the BJL (Bjorken-Johnson-Low) prescription which states the
rule: If the right-hand side of the above Fourier transformation approaches 0 for
go —>• oo, one can replace the Lorentz covariant T* product by the canonical
T product. If the Fourier transformation does not approach 0, one defines the
T product after subtracting the non-vanishing quantity, which is expressed by
a polynomial of q0, from both sides. A physical reason for this prescription is
that in the canonical formulation which is based on the equal-time commutation
relations the correlation functions of general currents should be well defined and
smooth in x^ — j/M around x° = y°- The limit </o —^ oo of the Fourier transfor-
mation should thus vanish (one may recall the Riemann—Lebesgue lemma).
In the present example both of (T*j°(x)j°(y)) and the correlation function
with the spatial components of the current (T*j°(x)jk(y)) in eqn (6.161) satisfy
the above condition, and thus one can replace the T* product by the T product.
For the T product one generally has
35
To be precise, the result in eqn (4.39) is written for the case A 2 ;§> |<?2 ; while the present
formula is valid for the case A 2 -C \q2\ also. A direct way to derive eqn (6.161) is to use
the Pauli—Villars regulator in Section 6.2 and set the regulator mass at A. Explicitly, one can
obtain the result (6.161) as n""(g,m 2 . A 2 ) - W(q, M2, A 2 ), by using W'1'(q,m2^\'2) given
in eqn (4.39) and then by redefining M —} A.
since one obtains the S function when the derivative acts on the T product and
the term with derivative operation inside the T product vanishes in the limit
(?o —> oo by the definition of the T product.36 One then concludes by using
eqn (6.162) for the correlation functions in eqii (6.161)
where C — i/(27T 2 )(A 2 — m 2 ) is a divergent c-number in the present example.

The term which contains C is called the Goto-Imamura-Schwinger term.
On the other hand, if one applies the same analysis to the correlation func-
tions containing the Gauss law operator for the gauge condition AQ = 0 (a
non-Abelian version of the Gauss operator is given in eqn (6.47), while we use
here an Abelian version)
where Ak (x) is the time derivative of Ak (x), one obtains
after applying the BJL prescription. The basic difference between [ G ( x ) , j k ( y ) ] x

6(x° - y°) and [j°(x), j k ( y ) ] S ( x ° - y°) is that the contribution from the term
where the gauge field contained in G(x) couples to the current jk(y) via the
photon propagator cancels the C term in eqn (6.163). This shows that the Gauss
law operator defines a proper generator of the gauge transformation in anomaly-
free theory and the current j ^ ( y ) is a gauge invariant operator.
On the other hand, in a, theory with the anomalies in gauge symmetry, the
commutators containing the Gauss law operators themselves have anomalous
terms. See. for example, eqn (6.53) in Section 6.3. A detailed analysis of the
commutation relations among the Gauss law operators in non-Abelian gauge
theory is rather technical, and we forgo the discussion here.
36
This effectively gives a prescription to separate the equal-time commutator from the cor-
relation function.
7
THE WEYL ANOMALY AND RENORMALIZATION GROUP
In this chapter we discuss the quantum anomaly associated with the scale trans-
formation of space-time coordinates or the transformation generally called the
Weyl transformation. In flat space-time, this anomaly is related to the renormal-
ization group and the calculation of the ft function in the renormalization group
equation is related to the calculation of the Weyl anomaly. In other words, the
renormalization group equation is regarded as an expression of the Weyl anomaly
in terms of Green's functions. We illustrate the calculation of the one-loop ft func-
tions in QED and QCD by means of the Jacobians for the Weyl symmetry. The
Weyl anomalies in curved space-time arc briefly explained. We also mention an
improved finite energy-momentum tensor in renormalizable theory on the basis
of an analysis of the Weyl anomaly.
7.1 Scale transformation in field theory

We explain the scale transformation in field theory by taking QED as an example.
The starting action is given by
We consider the following transformation of coordinates and field variables pa-

rametrized by a constant a, which is called the scale transformation,
If one writes the action in terms of the variable x', one has
If one uses the above scale transformation laws and the relations such as & =
exp(a)9jj. d4x' = d 4 xexp(—4a), one obtains
123
124 THE WEYL ANOMALY AND RENORMALIZATION GROUP
which shows that the action is invariant except for the mass term, which spoils
the scale transformation. In this sense, the field theory for massless particles
is generally scale invariant. The characteristic constants appearing in the scale
transformation such as 3/2 for i/j and 1 for AM are called the canonical mass
dimensions. As is well known in renormalization theory, the above naive scale
symmetry is broken since a new mass parameter is introduced in the renorrnal-
ization procedure even into a massless theory.
The purpose of the present chapter is to study these properties of field the-
ory from the viewpoint of quantum anomalies. It is known that QED is free
of anomalies associated with general coordinate transformations. Consequently,
it would be strange if one has an anomaly for the coordinate transformation
x1-1 —> exp(—a)x^. It is also shown that the Jacobian for the scale transforma-
tion with naive canonical dimensions does not give rise to the correct anomalies.
Those apparently paradoxical properties are all resolved if one analyzes the above
scale transformation as a combination of coordinate and Weyl transformations
in a slightly curved space-time. In fact all the anomalies of the scale transforma-
tions arise from the Weyl transformations when one defines a general coordinate
invariant measure, and the coordinate transformation docs not give rise to the
anomalies in QED in agreement with the general expectation.
7.2 Identities for the Weyl transformation arid Weyl anomalies

We formulate the Ward-Takahashi identity associated with the Weyl transfor-
mation in the flat limit of curved space-time. The Wick rotation to the Euclidean
metric in curved space-time is defined by e°(x) -> —ie^(x), and consequently we
have \f^g = dete* —»• — zdete* = —i^/g. To derive an identity associated with
the general coordinate transformation in Euclidean theory, we first recall that
the action for QED in curved space-time is written as (see Appendix B) 37
. 37In Euclidean theory, one needs to rotate the auxiliary field B(x) —> iB(x) (and thus c >. ic)
when performing the path integral,
THE WEYL TRANSFORMATION AND WEYL ANOMALIES 125
where F^v = dÂ^—dÂ^ and we used the Landau gauge explained in Chapter 3.
As is explained in Appendix B. this theory is invariant under the local Wcyl
transformation
except for the mass term. It is important to notice that ordinary gauge fields
(including Yang-Mills fields) are not transformed under the Weyl transformation,
which arc sometimes called the Weyl scalars, and thus the ghost field c(x), which
appears in gauge fixing and is associated with the gauge field, is not transformed
under the Wcyl transformation.
We now consider the path integral
where the path integral measure is given by the general coordinate invariant
measure in eqn (B.4f)
The Weyl transformation property of each variable in eqn (7.8) is given by
It should be noted that the naive Weyl transformation laws in eqn (7.6) and
the Weyl transformation laws of those variables in eqn (7.9) which define the
coordinate invariant measure arc quite different. In particular, the anti-ghost and
the auxiliary field B are no longer transformed under the Weyl transformation.
Incidentally, we have the relation 'DBtt = VBVc.
To write the identity related to the Weyl transformation in the present Eu-
clidean path integral in a compact manner, we define a generic variable for QED
by
The general coordinate invariant measure is then written as
and the general Green's function is defined by
The path integral (7.7) and the Weyl transformation laws (7.9) with an in-
finitesimal a(x) give the following identity
where the variables with primes represent the variables after the Weyl transfor-
mation, in particular, e'^'(x) = exp[a(z)]e^(x). The first equality in this relation
shows the fact that the path integral itself is independent of the naming of path
integral variables. The transition from the first expression to the last expression
is based on the fact that the action is invariant under the Weyl transformation
except for the mass term together with the evaluation of the Jacobian J(a) for
the transformation of variables.
The last two expressions in eqri (7.13) show that the change of the action
for an infinitesimal Weyl transformation of the vierbeins gives the same effects
as the Jacobian and the explicit Weyl symmetry breaking by the mass term
together with the transformation of variables in the integrand. The variation of
the action with respect to the vierbeins defines the energy-momentum tensor,
and in particular, the variation of the vierbein by the Weyl transformation defines
the trace of the energy-momentum tensor. We use the definition
and T%(z) = fkk(x) = (e%(x)/^/g)5S/6e%(x) with the variables $, c, c and B

kept fixed.
IDENTITIES RELATED TO COORDINATE TRANSFORMATIONS 127
We thus obtain the following identity (in the flat space-time limit)
where wn stands for the weight factors in eqn (7.9) for the Weyl transformation,
and to be explicit, wn = -1/2 for the fermion and wn = -1 for the gauge field. If
one integrates this identity over the coordinates x (or equivalently setting a = 1
in eqn (7.13)). one obtains the identity for the global Weyl transformation
In this equation Nf stands for the number of fermion fields appearing in the
Green's function and NA stands for the number of gauge fields. If one can eval-
uate the Jacobian exactly in this expression (7.16), one obtains the exact Weyl
identity. In practice, no exact non-perturbative evaluation of the Jacobian for
the Weyl transformation is known, and in this sense this identity is a formal one.
However, we can still extract some interesting physical results from eqns (7.15)
and (7.16), as is illustrated in the following sections.
7.3 Identities related to coordinate transformations

To derive an identity related to the general coordinate transformation, we use
the following variables of QED-type theory defined in eqn (7.10)
The general coordinate invariant measure is then given by cqn (7.11), and a
general Green's function is defined by eqri (7.12). The general coordinate trans-
formation law for the vierbeins is given by
and all the path integration variables transform as
This transformation law of $(x) is determined by the weight factors in eqn (7.9).
Note that we do not integrate over the vierbeins e^k (x) .
One can understand that the measure (7.11) is free of anomalies under the
general coordinate transformation (7.19) by expanding the variable Aa(x) as
and then the Jacobian vanishes
irrespective of the choice of the basis sets {4>an(x}}- For the complex fermioiis ij)
and -il>, one expands
and the Jacobians for these two variables added together vanish
irrespective of the choice of the basis sets {ipn(x)}. (In a chiral theory which
could contain a genuine gravitational anomaly, this simple argument fails since
one generally expands ip and ?/) in terms of different complete sets in such a
theory if one follows the calculational scheme of the covariant anomalies.)
The identity for the above general coordinate transformation (7.19) is given
by
The first equality in this expression shows that the value of the integral is in-
dependent of the naming of integration variables, and the transition from the
first line to the last line expresses the fact that the path integral measure is
anomaly-free and the action is invariant under the coordinate transformation.
We first note the relation
by using the metric condition in eqn (B.I7) and by noting that the tensor T^v
is not symmetric in general. When one combines this relation with $'(:/;) =
$(:/;) + £"c^$(a;) + (l/2)(d^)$(a;) in the last two lines of the identity (7.24),
one obtains the identity
We here collected only the terms linear in the infinitesimal parameter £"(x)
and then differentiated those terms with respect to £"(x). If one takes the
flat space-time limit in the identity (7.26) and uses the renormalized fields
$(#) —>• \/~Z$ r(x) with the wave function renormalization factor Z in the case
of renormalizablc theory, one obtains an identity for the renormali/ed Green's
functions
where we used the fact that the spin connection A%® (x) vanishes in the flat space-
time limit. Since the right-hand side of this relation is finite in renormalizable
theory, we learn that the energy-momentum tensor we defined gives a finite
operator. A concrete expression of this finite T^ will be given later in cqn (7.35).
When one multiplies by xv on both sides of the above renormalized iden-
tity (7.27) and integrates over x. one obtains
which shows that the trace of the energy-momentum tensor generates a scale
transformation of space-time coordinates. The renormalized Green's function
satisfies the following relation, which is based on a dimensional analysis in renor-

malizable theory.
where n stands for a new mass parameter specifying the renormalization point.
m stands for the renorrrialized fermion mass, and N/ and NA, respectively, show
the numbers of fermion fields and gauge fields appearing in the Green's function
with N = Nf + NA- One thus concludes
The identity (7.28) for the renormalized Green's function is then written as
The renormalization group equation for the renormalized Green's function

is derived from the fact that the bare Green's function does not depend on the
renormalization point [i
and it is written by using the renormalized coupling constant g as
By using the relations (7.33), the identity (7.31) is also written in the form
The function I3(g) which appears inside DN in eqn (7.33) is a fundamental quan-
tity in field theory and it is called the /? function. We emphasize that eqri (7.34)
is an exact relation valid up to any finite order in perturbation theory.
7.3.1 Improved energy-momentum tensor

We here give an explicit form of the improved energy-momentum tensor for a
QED-type theory in eqn (7.5), which is finite in reiiormalizable theory as we
explained in eqn (7.27), in the flat space-time limit
where !),„ is the conventional naive energy-momentum tensor. The difference

between 2)tJ, and T^v arises from the fact that we defined the tensor Tltv(x)
by differentiating the action S with respect to the vierbeins by keeping the
variables $(z) with the weight, factors fixed. One recognizes that the difference
is proportional to the equations of motion. When one writes the identity for the
change of the variable <&'(x) — $(x) + a(x)$(x), one obtains
where J(a) is the Jacobian for the change of the variable, and the difference of
two tensors is thus basically the difference of (Weyl) anomalies. A characteristic
feature of this improved tensor in eqn (7.35) is that, the mass term does not
appear explicitly.
If one recalls the local Lorentz transformation of the vierbeins parametrized
by an infinitesimal anti-symmetric parameter ujmn(x)
and uses the definition of the energy-momentum tensor in eqn (7.14) (see also
Appendix B), one can see that the anti-symmetric part of the energy-momentum
tensor generates the Lorentz rotation.38 Consequently, if one separates the sym-
metric and anti-symmetric components as
the symmetric part TS^V satisfies the identity
which explicitly separates the Lorentz rotation of $(xn) generated by S1'1 „(«,)
from the identity (7.27).
7.4 Weyl anomalies and j3 functions in QED and QCD

In Section 7.2 the identity containing / d ^ x T I J - ^ ( x ) was derived from the iden-
tity for the Weyl symmetry which contains the Weyl anomaly, and the identity
which contains the same object / d 4 x T ^ ^ ( x ) was also derived from the identity
for general coordinate transformations in Section 7.3 which is anomaly-free. By
combining these two identities (7.16) and (7.34) we obtain
where the symbol N[ip^](x) defines a quantity called the normal product, which
gives a finite result when inserted in any Green's function without inducing any
divergence. The relation moil}ip(x) = mN[ipii>](x) is known to be valid in QED
and QCD-type theories. The index c in [J(l) + J d'ixmoipip(x)]c indicates a
connected component, namely, one retains only those components connected by
the propagator to other variables when inserted in any Green's function.
To express the right-hand side of the equation (7.41) in terms of operators,
we write the action in terms of renormalized variables (by including the coupling
38
If one transforms the variables <S>(x) as *'(xj) = e.xp[^Smn(l)ujmn(x}l^(xi) simultaneously
as eqn (7.37), one obtains the identity
where TA^^X) stands for the anti-symmetric part of the energy-momentum tensor. We used
the fact that the localized Lorentz transformation is anomaly-free in the present model; the
Lorentz anomaly for a chiral theory will be explained in a later chapter. This Lorentz identity
combined with eqn (7.27) leads to eqn (7.40).
WEYL ANOMALIES AND /3 FUNCTIONS IN QED AND QCD 133
constant into field variables as g0A^(x) -*• A/_l(x) and rewriting the gauge fixing
term anew in terms of A(x)
where we omitted the subscript "r" to field variables. By using Schwinger's action
principle (in a renormalized form) we can operate the derivatives with respect
to g and m in the identity (7.41), which was derived by using T^, directly on
the action in the path integral arid we obtain
This relation, which was derived here in a heuristic manner, is known to be jus-
tified up to any finite order in perturbation theory by a more detailed analysis in
the framework known as the normal product algorithm.39 Accepting the result of
this analysis, we can identify the Weyl anomaly in eqn (7.41) with the operators
containing field variables as
We derived this relation for the QED action in eqn (7.42), but the same analysis
is known to be valid for QCD-type theories also. The relation (7.44) shows that
the part of the Wcyl Jacobian J(l) a that depends on the gauge field gives rise to a
term containing the j3 function, and the part of the Jacobian that depends on the
ferinioiiic field gives the term with the mass renormalization factor S(g), although
we cannot evaluate Weyl anomalies exactly. In the following we illustrate the one-
loop level evaluation of the 0 function by calculating the Wreyl anomaly in QED
and QCD-type theories.
39
See S.L. Adlcr, J.C. Collins and A. Duncan, Phys. Rev. D 15 (1977) 1712; N.K. Nielsen,
Nucl. Phys. B 120 (1977) 212.
7.4.1 Lowest-order 13 function in QED

The Weyl anomaly to one-loop accuracy is evaluated by the background field
method by separating the field variables into the background c-number parts
and the quantum fluctuations around the c-number parts. In the following we
evaluate only those terms which depend on the background gauge field and thus
we obtain the one-loop j3 function.'10
We start with the analysis of QED. The term which contains the field vari-
ables higher than quadratic order appears only in the fermionic action in QED,
and thus the lowest-order 8 function is given by the evaluation of the Weyl
anomaly generated by fermions in the presence of the background gauge field.
This calculation of the Weyl anomaly proceeds just like the calculation of the
chiral anomaly in Section 5. We first expand the path integral variables in terms
of Grassmann numbers as (in the Euclidean metric)
with the notation goA^(x) —>• Atl(x), and we define the path integral measure by
The Jacobian for the Weyl transformation in the flat space-time limit 41
for an infinitesimal a(x) is given by the gauge invariant mode cut-off regular-
ization (the variables tb and tb give the same Jacobian, and an extra minus sign
arises from the fact that these variables are Grassmann numbers)
40
The evaluation of the Jacobian with the c-number fermionic parts requires an evaluation of
the "supertrace." Such a calculation can be performed systematically in the superfield technique
in the theory with supersymmetry. We are here content with the calculation only with the gauge
field background, which is sufficient to give rise to the one-loop /? function in QED and QCD.
11
The identity associated with the Weyl transformation of the fermionic variables can be writ-
ten in a form similar to the chiral identity as | (1/117'*D^ 1/1 — (Df,.ip)i"tîp) — (l/247T2)-FA,l,
where Ihe right-hand side stands for the Weyl anomaly to be evaluated below.
WEYL ANOMALIES AND fl FUNCTIONS IN QED AND QCD 135
A more detailed account of this calculation is given later in cqn (7.54) again.
Consequently, the connected components which contain the field variables are
given by (using the result of eqn (7.56) below)
and the lowest-order 0 function in QED is obtained by using the relation

as
which reproduces the well-known result.
7.4.2 Lowest-order 3 function in QCD

We next analyze QCD-type theories with the notation gÂ^x) —?• A^(x)
where we denoted the bare coupling constant go explicitly by a suffix 0. The gauge
fixing procedure is important in the present application and it is explained later.
The calculation of fermion contributions to the lowest-order 13 function in QCD
type theory is evaluated in the presence of the background gauge field, and the
actual calculation is almost identical to the calculation in QED except for the
appearance of the trace over the color freedom in the Jacobian.
We here examine the calculation of the Jacobian for the Weyl transforma-
tion of fermionic variables (7.48) in more detail in the actual case of QCD-type
theories, so that we can apply the calculational technique to the calculation of
gauge field contributions later. We start with the formula
where we rcscalcd /% ->• Mk^ in the last expression after passing the factor
e%kx through the regulator arid tr here stands for the trace over the Dirac 7^
matrices and color indices. Only the terms of order 1/M4 or larger survive in
the expansion in terms of 1/M. If one defines
with the generators Ta of the gauge group SU(n) normalized by tiTaTb — \8ab
(in actual QCD we have SU(3)), one obtains
where we used the fact that terms odd in fcM vanish and that only F contains the
Dirac matrices and tr F = 0. The notation f^(x) stands for the n-th derivative
of/(x).
We next note that we can replace
when one performs a symmetric integration for the terms with k^. Also, by noting
k2 < 0 in the present metric convention we have
WEYL ANOMALIES AND 13 FUNCTIONS IN QED AND QCD 137
We thus have
and the connected component, which is finite, is finally given by using [D/^, Da] =
-iFlllr as
The important property in this calculation is that the finite connected component
does not depend on the details of the regulator function f ( x ) except for the fact
that f ( x ) approaches 0 rapidly for large x and /(O) = 1 together with (see
Fig. 5.1)
In QCD-type theories the gauge field also contributes to the lowest-order /?

function. The Yang-Mills Lagrangian is given by
and we split the field variables as

to apply the background field method. Here jE?° stands for the background field,
and a™ for the quantum fluctuation around it. The field strength tensor is then
given by
where the covariant derivative is defined by
The gauge transformation of the gauge field parametrized by uia(x) is defined by
by recalling that a^ is the dynamical variable.

If one uses the above field strength tensor in the action and assumes that the
variable B1^ formally satisfies the field equation (this is the basic ingredient of
the background field method), the action is written as
since the terms linear in a^ disappear. To simplify the notation and also the
evaluation of the Jacobian later, we write the covariant derivative as
and regard the gauge field a M (x) as a vector with n 2 — 1 components for the group
SU(n). The operation of the covariant derivative in eqn (7.63) is then written in
the simple form
As for the gauge fixing term and the compensating Faddeev-Popov term, we
employ the background Feynman gauge with £ = 1 (the BRST transformation
of the gauge field atl(x) is determined by eqn (7.64) by the replacement <jja(x) ->•
i\ca(x))
where we used vector notation for the ghost variables c and c just as for a M , and
eliminated the auxiliary field B by integrating over it in the path integral.
WEYL ANOMALIES AND /3 FUNCTIONS IN QED AND QCD 139
The Lagrangian for the quadratic parts in the dynamical variables related to
the Yang-Mills field and ghost variables is obtained from cqns (7.65) and (7.68)
as
where we used the notation
by using eqn (7.66). We now evaluate the Weyl anomaly by using this quadratic
Lagraiigian. The Weyl transformation laws of integration variables are fixed as
was explained in eqn (7.9)
in the flat space-time limit.

Consequently, the connected components of the Jacobian for the Weyl trans-
formation are given by the general formula (7.58), which was explained for
fermion variables, by suitably adjusting the transformation coefficient and the
operator appearing in the regulator. For the gauge field we have
where we used the notation (F)^ — —1iF^v(B) and replaced $92 —>• -DT>;, + F
inside the regulator, which formally has the same form as in eqn (7.52). In the
third line in this calculation we evaluated the trace with respect to the Minkowski
indices of a^ and then evaluated the trace over the adjoint representation of the
gauge group by using
For the group SU(n) we have C2(G) = n.

140 THE WEYL ANOMALY AND RENORMALIZATTON GROUP
Similarly, the Weyl anomaly for the ghost variables is evaluated as (by re-
placing f)2 -t D^DH inside the regulator in eqn (7.58))
where we used the fact that the ghosts are described by Grassmann variables
and thus there is an extra minus sign in the Jacobian.
One may replace the background field B^ (x) by the full variable A/( (x) in the
Jacobians evaluated so far to the accuracy of the present one-loop approximation.
If one adds those results of the Wcyl anomalies in eqns (7.58). (7.72) and (7.74),
the /3 function for a QCD-type theory with / fermions is given by using eqn (7.44)
as
namely, one recovers the well-known result of the lowest-order 3 function
where 62 (G) = n for the gauge group SU(n). Note that J3(g) < 0 and thus the
theory is asymptotically free for / < 33/2 for SU(3).
7.5 The Weyl anomaly in curved space-time

We illustrate a typical example of the Weyl anomaly, which appears in the back-
ground gravitational field, by taking QED as an example. Actual calculations of
the Wcyl anomaly rely on the technically involved regularization schemes such
as the heat kernel method and the C function regularization. We here simply
state the final results. We employ the action in the Euclidean metric defined by
the Feynman gauge
We first analyze the Weyl anomaly generated by quantized fermion fields whose
Weyl transformation rules are given by
THE WEYL ANOMALY IN CURVED SPACE-TIME 141
We expand the fermionic variables as
and the Jacobian for the Weyl transformation is given by (by writing a sum of
'(/.' and -0 contributions, which are Grassmann numbers)
where we used the gauge invariant mode cut-off regularization, and the final
expression uses the common notation in the heat kernel regularization. This
calculation is performed by using a formula which appears in the intermediate
stage of the C function regularization. and the Weyl anomaly is obtained as42
In this expression we write only the gravitational contributions. The fermions

couple to the gauge field and thus the Weyl anomaly actually contains the con-
tributions from the gauge field which were discussed in the previous section.
The quartic divergent term ^/gM4 . which corresponds to the cosmological con-
stant, could be subtracted by a suitable local counter-term. It is known that
42
The calculational scheme by converting the complete set to plane waves is not convenient
here since we have to take into account the effects of the curved space-time. The heat kernel
or £ function regularization provides a more systematic method to evaluate eqn (7.80).
the quadratically divergent term ^/gRM2 disappears in the final result of the C
function regularization.
We next note that the action for the gauge field and Faddeev-Popov ghosts
in eqn (7.77) is written as
where Aa(x) — e^(x)AIJi(x) and the covariant derivative acting on the gauge field
is defined by
as explained in Appendix B.
The Weyl transformation rules of the variables which define the general co-
ordinate invariant measure are given by (see eqn (7.9))
and we expand the variables as
arid
The Jacobiaii for the Weyl transformation of the gauge field is

THE WEYL ANOMALY IN CURVED SPACE-TIME 143
\uJA(a)
where tr stands for the trace over the vector indices of the gauge field, and
the Jacobian for the ghosts is (by recalling the fact that the ghost fields are
Grassmann numbers)
These results are based on the formulas which appear in the intermediate stage
of the £ function regularization. We omitted the cosmological term ^/TjM4. which
could be subtracted by a suitable counter-term, and the term ^/gRM2 which is
eliminated in the final result of the £ function regularization.
The final contributions of the gauge field and the ghost fields to the Jacobian
are thus given by
It is known that, the Wcyl anomalies (7.81) and (7.89) agree with the results of
other calculational methods.
7.6 The Weyl anomaly in two-dimensional space-time

The Weyl anomaly in Euclidean two-dimensional curved space-time is important
in connection with conformal field theory. In this section we discuss the Weyl
anomaly generated by massless scalar and fermion fields. We employ the confor-
mal gauge for the metric variables, which is convenient for practical calculations.
This gauge condition is specific to two-dimensional space-time and is defined by
We define the Euclidean metric by a Wick rotation from the Lorentz metric as
Gmn = (-!,-!) = ??M,,.
In this gauge the action for the scalar field X(x) is written as
and the action for the Dirac field is written as (by using the definition of the
spin connection A^/, in Appendix B)
These expressions of the action show that the scalar theory is invariant under
the Weyl transformation
THE WEYL ANOMALY IN TWO-DIMENSIONAL SPACE-TIME 145
and the Dirac theory is invariant under the Weyl transformation
These classical symmetries are broken by the quantum anomaly if one uses
the path integral measure which preserves general coordinate invariance. If one
defines the path integral measure, which is invariant under the general coordinate
transformation, by using the weight 1/2 variables X(x) = (g)l^X(x) = ^fpX(x)
and il! (x) = (g)1^4"4>(x) = \fp'4!(x] the path integrals are respectively written as
The complete orthonormal sets which define the path integral measure in the
gauge invariant mode cut-off are then respectively given by
The Jacobians (quantum anomaly) for the Weyl transformations
arc, respectively, given by
and
where the symbol tr stands for the trace over the (two-component) spinor free-
dom.
Since one can write #0/2) = (If p * - / * ) f i ( l / p l / z ) $ ( l / p 1 / * ) in eqn (7.96), the
Jacobians for the scalar and Dirac fields are evaluated bv
if one sets n = 0 or n = -1/2, respectively, after the calculation and adjusting

the factor 2 arising from the trace over 7'* matrices.
The evaluations of Weyl anomalies are performed by
where the overall factor 2 arises from the trace over the spinor freedom which
should be omitted for the scalar field.
The actual calculation of the Jacobian proceeds as follows: We define
and we expand
OTHER APPLICATIONS OF WEYL ANOMALIES 147
in powers of 1/M to the second order by imitating the calculation in interaction

picture perturbation theory. In this calculation
is obtained from Hj in cqn (7.102) by the replacement
When one uses the above expansion in
and uses the relations (by noting k2 < 0 in our metric convention)
together with the fact that terms odd in k^ vanish, one obtains the final re-
sult (7.101).
We thus obtain the Weyl anomalies for massless fields in two-dimensional
space-time from eqns (7.98). (7.99) and (7.101) as
where we used ^fg = p and ^/gR = —d^d^lnp with R standing for the scalar
curvature. The term ^/gM'2 could be subtracted (or made finite) if one adds a
suitable cosmological constant to the starting action as a counter-term.
7.7 Other applications of Weyl anomalies

Other important applications of the calculational method of Weyl anomalies ap-
pear in theories with supersymmetry (boson fermion symmetry), though we do
not discuss supersymmetry in this book. The first example is called the Kon-
ishi anomaly, which is regarded as a supersymmetric generalization of the Weyl
anomaly. (The Konishi anomaly is also regarded as a supersymmetric general-
ization of the chiral anomaly.) The other example is a generalization of the cal-
culation of the ,3 function discussed in this book to supersymmetric theories. A
characteristic feature of supersymmetric theories is that renormalization proper-
ties are greatly improved compared to ordinary theory (the non-renormalization
theorem and its generalization). As a result one can evaluate quantum correc-
tions up to all orders in perturbation theory in some special cases. It is known
that the present calculational method is generalized to supcrsymmetric theories
defined in terms of superfields. which ensures manifest supersyrnmetry, and in-
teresting results have been obtained. For details of these calculations, readers
are referred to the references at the end of the present book.
8
TWO-DIMENSIONAL FIELD THEORY AND BOSONIZATION
In this chapter we discuss the quantum anomalies in two-dimensional field theory.

Two-dimensional field theory is important in connections with conformal field
theory and its applications to string theory and condensed matter theory. We
first summarize essentially all the steps in evaluating the chiral anomalies in
the path integral formulation so that a reader who starts with this chapter can
understand the subject without excessive reference to earlier chapters. We then
formulate the description of fermionic theory in terms of bosonic theory, namely,
the bosonization in the path integral formulation. An issue related to a local
counter-term is clarified. We next explain that the central extensions in Kac-
Moody and Virasoro algebras are the algebraic representations of chiral and
general coordinate anomalies, respectively. We also explain the connection of
the identities written in terms of the operator product expansion in conformal
field theory with the identities in conventional field theory. Finally, we discuss
the calculations! method of Wcyl anomalies in string theory and its implications.
The ghost number anomaly in the first quantization of string theory is related
to the Riemann-Roch theorem.
8.1 Chiral anomalies in two-dimensional theory

We start with the fermionic path integral of a theory which contains Abelian
U(l) vector V^(x) and axial-vector Afl(x) gauge fields
where we regard Vtl(x) and A^x) as the background c-number fields for the
moment. We adopt natural units by setting H = c — 1. Our Minkowski metric
conventions are
150 TWO-DIMENSIONAL FIELD THEORY AND BOSONIZATION
where ak stands for the Pauli matrices arid r]IJ/v = rf" — (1,— 1). We adopt
the convention of summing over the indices which appear twice in the same
expression.
The actual path integral is performed in the Euclidean metric convention
with imaginary time. The Euclidean theory is defined by the Wick rotation
and x2 is treated as a real number. The Euclidean inner product is then defined
bv
and the metric tensor is now given by 'r\^v ~ rîv = (—1, —1). The 7-matrices are
defined in Euclidean theory by the replacement 7° —>• —/'7 2 .
and they are aiiti-hermitian (7''')^ = ~7 // but 7? = 75. We also defined the Wick
rotation of the gauge fields
so that Tf) retains the same form as in the Minkowski metric.

The Dirac conjugate of il>(x) is defined by ip(x) = t/J^(x)j°, but in the path
integral these two quantities are treated as independent variables. In Euclidean
theory, ip(x) is transformed in the same manner as the spinor ['</>(z)]t under
the Euclidean rotation group SO (2). The Dirac operator in Euclidean theory
f) = 7^(3,, — iVp. — iApjs) is not hermitian in the Euclidean inner product
To define a hermitian operator, which renders the path integral manipulation

more reliable, we rotate all the components of the axial gauge field in Euclidean
theory into pure imaginary variables
and rotate them back to the original field A^ —?> —iA^ after the calculation. The
operator after the rotation of A/L
CHIRAL ANOMALIES IN TWO-DIMENSIONAL THEORY 151
becomes hermitian ffi = $ in the Euclidean sense, and one can define a complete
orthonormal set and the expansion of fermionic variables
where {an} and {&„} are the Grassmann numbers. The path integral measure is
then written as
and we define the path integral by
This path integral is performed exactly though in a formal sense. The integration
with respect to Grassmann numbers is defined by the left derivative.
We now examine the Jacobian under gauge transformations. Under an in-
finitesimal vector-like transformation
we have by remembering the expansion (8.10)
The Jacobian for the transformation of {an}, for example, is evaluated by
for an infinitesimal a(x) by remembering that the integral over the Grassmann
numbers is defined by the left derivative and thus the inverse of the ordinary
Jacobian appears. Similarly, one has
We thus define the total Jacobian by
which vanishes identically, and thus we have no anomaly for the vector-like trans-
formation in the present formulation.
On the other hand, for the infinitesimal chiral transformation43 generated by
7s,
we obtain the Jacobian following a procedure similar to the vector-like transfor-

mation as
This Jacobian is evaluated by the gauge invariant mode cut-off,44 namely, by

replacing the cut-off in N by the eigenvalue cut-off with a simple regulator / (x) =
e.~\
where in the last line we transformed {(pn(x)} to the plane wave basis for the
well-defined operator 75 exp(—$> 2 /M 2 ) to extract the gauge field dependence,
and the trace is taken over the freedom of Dirac indices.
43
In accord with eqn (8.8) one may also set a(x) —} io(x) in the chiral transformation, but
we forgo this refinement.
1/1
Actually, we preserve only the vector-like gauge symmetry for the present case.
This Jacobian is evaluated as follows: When one defines D,,, = d^ — iV^, as

a covariant derivative with only the vector-like gauge field (by noting if =
(—1, —1) in Euclidean metric) we have
where the differential operator in (dÂ^) acts only on A^ and F^ = d^v—dvV^,.

If one recalls the definitions of 75 = cr3, 71 = iu'2 and 72 = ial in terms of Pauli
matrices in Euclidean theory and the relation
one finds that the last two terms in eqn (8.21) do not contain the 7 matrices and
thus do not contribute to the Jacobian (8.20) in d = 2 dimensions. Consequently,
one can use
in the calculation of the Jacobian (8.20).

In this way one can evaluate the integrable form of anomaly
by applying the scale transformation k^ -» Mfe,( after passing e*fc:£ through the
regulator as in four-dimensional theory and then retaining the terms to order
1/M2 in the expansion of the exponential factor in powers of 1/M. We also used
by recalling that fc2 < 0 in our metric convention.

When one first goes back to the original variable A^ -» —iA^ in eqn (8.5)
in the Euclidean metric, and then if one Wick rotates back to the Minkowski
metric by noting the convention of eM" in the Minkowski metric
the Jacobian (8.24) is finally written as
for the theory (8.1) defined in the Minkowski metric. We here also used <fx —>
i d2x in rotation to the Minkowski metric.
In the case of 11011-Abelian theory defined by
with YH = V®Ta and A/( = A^Ta, the Jacobian (8.27) is generalized to
where tr stands for the trace over the generators T° of the group SU(n), for
example. The generators are defined by
We also defined
in eqn (8.29). The calculation of eqn (8.29) is essentially the same as eqn (8.27),
and the expression (8.23) in Abelian theory is replaced by
in non-Abelian theory with the definition D^ = d^ — iV*Ta. The trace in

eqn (8.20) is then taken over the freedom of the non-Abclian generators arid
Dirac matrices by using a(x) = aa(x)Ta.
8.1.1 Local counter-term in gauge theory

Prom the viewpoint of the general analysis of the anomaly in gauge theory,
the first term of the Jacobian In J(a), namely, the term (dÂ^) in Abelian the-
ory (8.27) or (DÂ^) in non-Abelian theory (8.29) could be cancelled if one adds
a suitable mass term AÂ1* for the gauge field A^ in the starting Lagrangian.
We illustrate this property for the Abcliari theory (8.1) by adding a mass term
to the Lagrangian
In fact we have
where the second equality states that the path integral is independent of the
naming of path integral variables, arid we set tj)'(x) = exp[ia(x)75]'i/>(2;) and
tp'(x) = i,b(x) exp[ia(a;)75] for an infinitesimal a(x). The Jacobian for this trans-
formation is given by our result in eqn (8.27) and thus the first term in the
Jacobian is in fact cancelled by the local counter-term.
It is customary not to assign a physical significance to these terms cancelled
by a local counter-term in the ordinary treatment of the anomaly. However, those
terms do not diverge in two-dimensional theory and they play a basic role by
giving the kinetic terms for boson fields in the path integral bosonization to be
discussed below. In fact, we shall later show that those terms are not cancelled
by a local counter-term in the context of the path integral bosonization. The
usual argument of the local counter-term thus needs to be carefully examined
depending on the physical context one is studying.
Another important issue in the path integral bosonizatiori is the symmetry
relation with the anti-symmetric symbol e1*", eM" — —(v>i and e10 = 1,
which is valid in strictly d — 2 dimensional Minkowski space-time. This relation

is used in an essential way inside the action of the path integral. One thus has
to preserve the corresponding relation in the current level
for the sake of consistency. This relation is generally satisfied if one regularizes
the currents in terms of the regularized correlation function (T*i^(x)ip(y))ieg in
d = 2 dimensions.
The representative regularization schemes of the Jacobian in the path in-
tegral, namely, the V-A-type regularization presented in Section 8.1 and the
covariant regularization discussed in Section 6.4, both satisfy the above condi-
tion (8.36). As for the covariant current associated with the covariant form of
anomalies, we already explained that it is defined in terms of the regularized
correlation functions in Section 4.3 and also in eqn (6.87).
As for the V-A-type regularization in eqns (8.17) and (8.20), it is related to
the regularized currents in the Euclidean metric defined by
and similarly
where we used the complete set { i f n } in eqn (8.10).

The divergence of the current (8.37) gives
ABELIAN BOSONIZATION OF FERMIONS 157
where we used the fact that 7^ is anti-herniitian (7^)^ = —7 M and 7^75 + 757'* =
0 and that tf)ipn = Xn<fn. Similarly, one can show
We can thus correctly generate the anomalies identical to the Jacobians from
the regularized currents. For the regularized currents defined in eqn (8.37) and
eqn (8.38), the relation (8.36) or
in the Euclidean metric is trivially satisfied.

In the path integral formulation, the V-A-type regularization satisfies the
Wess-Zumino integrability (or consistency) condition45 and also the relation
(8.36). We thus adopt the V-A scheme for the path integral bosonization. Besides,
the V-A-typc regularization preserves the fermion number symmetry as is seen
in eqn (8.17), which is essential in the bosonization since the fermion number
should be well defined by definition when one talks about the bosonization of a
fermion theory.
8.2 Abelian bosonization of fermions

We diseuss the bosonization related to Abelian gauge transformations. We can
learn the basic idea of bosonization by the analyses of these simple models. There
are various versions of bosonization, and we here discuss three representative
examples.
4
°The Wcss-Zumino condition is related to Bose symmetry in the Feynman diagram lan-
guage. The Bosc symmetry in this sense is not sufficient for the path integral bosonization. For
example, it is shown that the currents related to the consistent anomaly defined in Section 6.6
do not satisfy the relation (8.36).
8.2.1 Bosonization of free fermions

We study the path integral
where the path integral measure is specified by the regularization in Section 8.1.
We next define a generating functional of connected Green's functions W(v^) by
by regarding v^x) as a source function. In this setting any local counter-term

should be expressed as a local polynomial in v^(x).
Now we observe that the vector field v^x) in two-dimensional space-time
contains two independent components and it can be decomposed into two arbi-
trary real functions n and £ as
After this decomposition, we can write cqn (8.43) as
where we used eqn (8.35), £^7''' = 71/75, which is valid in a strictly d = 2

Miiikowski metric. This path integral has formally the same form as eqn (8.42)
if one denotes Vp, = d^r) and A/,, = <9^£.
We extract the functions r; and £ as integrated Jacobians associated with the
transformations of integration variables ip and il>. Using the results of Section 8.1
for the infinitesimal transformations
we have the Jacobians in eqn
where Vp, = d^rj and AIL = d^. Using these Jacobians, the r\ and £, dependences
in the action arc extracted as
where F(w |U ) stands for the integrated Jacobian (or anomaly), and it has the form
in the Minkowski metric
The derivation of this integrated Jacobian proceeds as follows: One first elimi-
nates the component r)(x) by a vector-like transformation, which is anomaly-free
and thus without any Jacobian. One then performs an infinitesimal transforma-
tion parametrized by dn£(x) in the intermediate result where the axial-vector
component d^(x] is partially extracted and given by AIL = (1 - s)9 /t ^(x) arid
Vp — 0 in the general formula of the Jacobian (8.47). One then obtains the above
formula r(i^) by integrating over the parameter s of infinitesimal transforma-
tions from 0 to 1.
We can also rewrite eqn (8.48), where the fermionic path integral gives a
numerical constant, as
since the absolute normalization of the path integral does not matter in the defi-
nition of W(v^. We next shift the variable <p -*• ip+£, and note the "translational
invariance" of the path integral measure 'D((p + £) = T>ip. We then have
In writing the second line, we used the relation
and in the last line we rescaled if -> ^/TT^ so that the kinetic term of <p(x) has
the canonical normalization.
The expressions (8.43) and (8.51) show that the theory of a free Dirac fermion
if) and the theory of a free real Bose field tp define the identical generating func-
tional Wivn) of Green's functions. The correspondence in the normalized nota-
tion in eqn (8.51) is
By differentiating the generating functional with respect to the source field

Vp,(x) twice and then setting v/Ji(x) — 0, we obtain
which shows the equivalence of the correlation functions described by a Dirac

fermion and a real boson in d = 2 dimensional space-time.
. The part of the Jacobian dÂ^ in eqn (8.27). which could be subtracted by a
local counter-term l/(27r)^ in the context of gauge theory in eqn (8.33), plays
a central role in the above bosonization. In fact it gives the kinetic term of the
bosonized field.
We examine this connter-term in more detail. The term i/(2ir)A^l, which is
local in terms of the axial vector field A^ appearing in eqn (8.33), is actually
not local in the context of eqn (8.43). The local counter-term in the.context of
eqn (8.43) should be expressed as a local polynomial of the source field v^(x). as
we have emphasized there. The would-be counter-term (in the context of gauge
theory) is written in terms of v^ as
which is not local. It is thus not allowed to add this term as a counter-term to the
definition of the original partition function (8.43). This term, if added, modifies
the physical contents of the original fermionic theory even in the conventional
understanding of local counter-terms.
8.2.2 The massless Thirring model

We study a theory of interacting fermions, which is called the massless Thirring
model,
where the first representation is obtained from the second one when one elimi-
nates the auxiliary field B^ by using the equation of motion. The path integral
formula is given by
where we used the decomposition generally valid in two-dimensional space-time
by using the relation (8.35). We also used
which is a generalization of the derivation of the volume element ^fgtf'x from

the formula for the length (line element) ds2 = g^lv dx^dxv, and discarded the
constant factor det(—d^d^).
We first eliminate the coupling between p and $ in eqn (8.57) by a change
of variables corresponding to a vector-like gauge transformation
which do not give a non-trivial Jacobian. We next perform the change of variables
corresponding to a chiral gauge transformation
which eliminates the coupling between the fcrmion and <j>, but we obtain a non-
trivial Jacobiaii
The evaluation of this Jacobian proceeds just as the calculation (8.49) in the
preceding subsection. In this way, the path integral is written as
which shows that the interacting theory of the i'ermion *& is rewritten as a sum
of the theories of a free Dirac fermion tjj and two real scalar particles p and d>.
But the variable rp corresponds to a theory with negative metric.
By using the above result we can, for example, write the propagator (or
correlation function) for the original variable in eqn (8.56)
as
The Green's function which satisfies
in the minkowski metric is given by
where p, is an arbitrary parameter with a unit mass dimension. This expression

when combined with
gives a correlation function of a free fermion (by using
The calculation for the field <j> by adding a source function J(x)
gives (by choosing J ( x ) with a (^-functional peak at x and y such that J dx J ( z ) x

(j)(z) = 4>(x) — o(y) and by noting that the combinations x and y appear twice)
where z is a cut-off dependent constant arising from the propagator at the co-
incident point. Similarly we obtain (by a replacement 1 + g2 fir ->• -1 in the
exponential factor for the case of 6(x) in eqn (8.71))
where z' is a cut-off dependent constant.

We can thus derive the exact correlation function for the interacting *
after absorbing the cut-off dependent factor (zz1)2 into the left-hand side. Note
that the ju-deperidence on the right-hand side of eqn (8.73) carries information
about the anomalous dimension of ^S'. More general 2n-point functions are simi-
larly evaluated exactly.
8.2.3 The massive Thirring 'model and the. sine-Gordon model

We now discuss the equivalence between the massive Thirring model and the
so-called sine-Gordon model. This analysis is a combination of the analyses in
Subsections 8.2.1 and 8.2.2. The massive Thirring model is defined by the La-
grangian
and the sine-Gordon model is defined by
where ao, /3 and 70 are real parameters.

We start with the generating functional of connected Green's functions for
the massive Thirring model
where we added the source term v/^ (x). We now use the decomposition
where we used eqn (8.35). The path integral (8.76) is then written in the form
of eqn (8.57) as
164 TWO-DIMENSIONAL FIELD THEORY AND BOSOMZATION
We first apply a vector-like transformation to the fermion variables
which eliminates the couplings of p(x) and r/(x) to the fermion without generating
a non-trivial Jacobiari. We next perform a chiral transformation
which eliminates the couplings of 4>(x) and £(x) to the fermion in the kinetic
term but generates a Jacobian
The derivation of this integrated Jacobian proceeds just as in eqri (8.49).

In this way, the generating functional of connected Green's functions is writ-
ten as
We next define
and rewrite the path integral as
by noting that T>[<f —-((I) — £)] = T>(p and the absolute normalization of the path
integral does not matter in the definition of W(v^).
This formula is further modified to
where we used / d2xd^d11'^ = - f d2xvfte^"dIJif and dropped the path integral

of p since it gives a constant. Finally, we define
and write the path integral as
which gives a basis of our analysis.

We now fix the variable (p and perform the path integral over the remaining
variables. We first observe that the mass term can be written as
with I!JR,L = [(1 ± 7s)/2]V ; > an(l expand the path integral (8.87) in powers in
the mass term.46 Since the massless fermion is chiral invariant and also the
Lagrangian for the massless 0 is invariant under the translation in <•/>, only the
following terms survive in the expansion
where the numerical coefficient l/(fc!) 2 appears from the binomial coefficient
We show below that the expression (8.89) is equal to

46
This perturbative treatment of the mass term is consistent -with our calculation of the
Jacobian (8.27), which treats the fermion action without the mass term in eqn (8.74) as the
basic action. Tf one treats the mass term non-perturbatively, one generally needs to include the
scalar and pseudo-scalar fields in the basic operator (8.20) for the present V-A-type evaluation
of the Jacobian.
where A stands for the ultraviolet cut-off to define the propagator for the scalar
field 0 at the coincident point. The generating functional is thus written by
following the reverse steps to the above and by noting that the free field part of
(f is invariant under the translation ip —> <p + constant as
where we added an arbitrary constant 70- If one defines the variable (p with the
standard normalization by
we have the generating functional expressed in terms of the sine-Gordon model
where the parameter 0 is given by
and ao/fi2 = Am 0 /7r. By comparing eqn (8.94) to eqn (8.76). we have the iden-
tification
We thus recover the well-known bosonization rules. Note that the coupling con-
stant g in some of the literature corresponds to our g2.
We now briefly explain the transformation of eqn (8.89) to eqn (8.91). The
correlation function of the scalar field <j> is given by
for generic points \Xi} and {yj}- The constant A stands for the ultraviolet cut-off
in the propagator for the coincident point. This result is obtained by a generaliza-
tion of the calculation (8.70) with suitable adjustments of numerical coefficients.
The free fernaion propagator is evaluated by the same procedure as in the path
integral in Appendix A (A.22) (by using the correct normalization Z(0, 0) = 1)
where we used the explicit form of iSp(x - y ) = (T*if)(x)ip(y)) in eqri (8.69). To

simplify the following calculation, we write the two components of the fermions
explicitly
In this notation the result of the above path integral is written as
where we used the light-cone coordinates x± = XQ ± %i to be denned in eqns

(8.106)-(8.108) below. We also used the 7-matrix convention in the Minkowski
metric in eqn .(8.2). in particular, 75 = a^. In the present notation the source
terms in eqn (8.98) are written as rjiV-'i + ^2^2 + Wi7?! + 'ta'te, an(i thus the
correlation functions of component fields arc written as
In this notation we also have
We then evaluate
where the determinant is defined for the k x k matrix specified by the indices i
and j. We next apply the identity (Cauchy's lemma)47
to the two determinant factors in eqn (8.103). We thus obtain
where we have used the relation x+X- = x2 and introduced an arbitrary mass
parameter //. By combining eqn (8.97) and cqn (8.105) in eqii (8.89), we finally
establish the desired form of cqn (8.91).
8.3 Non-Abelian bosonization of fermion theory

To analyze the non-Abelian bosonization, we first briefly summarize the notation.
We use the light-cone coordinates in Minkowski space-time defined by
47
This identity may be proved by examining the residues of poles of both sides and invoking
the Liouville theorem. See. for example, M. Stone, Bosonization. World Scientific, Singapore
(1994).
NON-ABELIAN BOSONIZATION OF FERMION THEORY 169
where our convention is e10 = 1. The following relations hold in these coordinates
and similarly
8.3.1 Non-Abelian bosonization of free fermions

We define the generating functional of Green's functions by
where the fermions belong to a representation of the group SU(n) and the source
field v_(x) is Lie-algebra valued, v~(x) = vaL(x)Ta. In two-dimensional space-
time, the two independent components of WM can be written as
where h(x) = ha(x)Ta and l(x) = la(x)Ta.

By using the above relations, the generating functional is written as
Formally, this path integral can be regarded as a non-Abelian chiral gauge theory
in which vector and axial gauge fields are given by
In eqn (8.111), we. repeatedly apply an infinitesimal transformation of inte-

gration variables,
and integrate the resulting Jacobian by using eqn (8.29) with respect to the
parameter s. In the intermediate steps, the gauge fields are given by
where
This operation then extracts the v- (x) dependence from the action of the path
integral. The result is
where
Various factors in this expression are given explicitly as
Thus we can also write
By using
the first term in eqn (8.119) is written as

where only the value at a = 0 contributes. The second term in eqn (8.119) is
written as
where we used dÛ(h,s)^ = -U(h,,rfd,iU(h,s)U(h,s)^ which follows from

dn(U*U) = 0. The last equation can be written in a symmetric way by in-
troducing a three-dimensional notation
for eqn (8.115) and by using a totally anti-symmetric symbol eM"A with the
normalization e102 = 1 as
where the integration domain D = disc indicates 0 < x2 = 1 — s < 1. We thus

finally obtain
where the Wcss-Zumino-Witteii action FWZW(^) is given by
It can be verified that Fwzw(£0 defined above satisfies the composition law
for general U = U(£). To show this relation we start with

where £/(£) = exp[-i£a (x)Ta]. We now consider the change of variables
and we obtain
Here iT = lnJ stands for an integrated Jacobian, and it is given by
The difference of zT(u_, f/(£)) from the previously evaluated iT(U(h)) (8.117) is
that besides the replacement U(h) —> U(£) we ha,vc
instead of We thus obtain from eqn

(8.119)
where the second term is independent of v- . The first term in this expression is
written as
where only the value at s = 0 contributes.

On the other hand we have
which is a result of the Wess-Zumino integrability condition. We thus have
which gives the desired result (8.127) when written in terms of FWZW-
We have shown in eqn (8.116) and eqn (8.125) that
since the absolute normalization of the path integral does not matter in the
definition of W(v-). We can rewrite this relation by multiplying by the constant
J 'DU exp[jTwzw(£0] on the right side as
We then change the integration variable as U —> U(h)U. By using the gauge
invariance of the integration measure, T>(U(h)U) = DU, and the composition
law in eqn (8.127), we recover the non-Abelian bosonization formula
Comparing eqn (8.139) to eqn (8.109) we obtain the bosonization rule of the
free Dirac fermion theory
and the free fermion theory is translated into a non-linear a model with a Wess-
Zumino term. This bosonization of a fermion theory plays a fundamental role in
conformal field theory in d = 2 dimensions.
The first term of r\yzw(^ r (7 1 )) in eqn (8.126) comes from the first term in
the anomaly in eqn (8.29). which could be subtracted by a local counter-term
in the context of gauge theory. The would-be counter-term (in the context of
gauge theory) is not local in the context of the present bosonization. In fact, by
noting the relation, which arises from the fact that U(h)^dÛ(h) is a pure gauge
and thus FIJ/V = 0,
the would-be counter-term is written in terms of V- as (by recalling A^ =

(i/2)U(h^d,U(h))
which is non-local with regard to V-(x). The kinetic term in I\vzw(k r ) cannot
be subtracted by a local counter-term.
8.4 Kac—Moody algebra and Virasoro algebra

We discuss some of the basic subjects in conformal field theory in d = 2 Euclidean
space-time. We show that the central extensions of the Kac-Moody and Virasoro
algebras are the algebraic representations of quantum anomalies. We also explain
the connection between the identity expressed in terms of the operator product
expansion, which is commonly used in conformal field theory, and the more
conventional identity in field theory.
8.4.1 Kac-Moody Algebra

The starting Lagrangian of a fermion theory in two-dimensional Euclidean space-
time is given by
in the path integral
we consider the chiral transformation for an infinitesimal aa (x)

KAC- MOODY ALGEBRA AND VIRASORO ALGEBRA 175
We then have the identity
where the first equality is the statement that the path integral itself is indepen-
dent of the naming of the path integral variables, and in the last line we evaluate
the Jacobian and the variation of the action. By noting tiTaTb = (l/2)<5a{,, we
then obtain
where we used the result of the anomaly (8.29) with AM = 0 (written in Euclidean
space-time). Similarly, the vector-like transformation
which is free of anomaly as was explained in Section 8.1, gives rise to the identity
We normalized e12 = 1 and defined the Nother currents
As a special property of two-dimensional theory, the vector and axial-vector

currents (in the Euclidean metric) are related by
in the present path integral formulation. See eqn (8.41).

When one functionally differentiates both sides of the identity (8.148), which
contains the axial-vector current, with respect to V^(y) and then sets V°' = 0,
one obtains
where jbv(y) appeared from the action in the exponential of the path integral.
Note the definition
Similarly, one obtains from the identity with vector currents (8.150)
We now recall the Bjorken-Johnson-Low (BJL) prescription. This prescrip-

tion tells us how to convert the time ordering in Lorentz covariarit calculations,
which is denoted by T*, to the conventional time ordering operation T, which is
not necessarily Lorentz covariant. To be explicit, if the relation
holds for arbitrary operators A(x) and B(y) (by taking into account the trans-
latiorial invariance), one can replace it by
Intuitively, the canonical formulation, which is based on the equal-time commu-

tator, should give rise to a smooth x2 (= ix°) dependence of the correlation
function near x'2 ~ 0, and thus its Fourier transform vanishes for large ki (recall
the Riernann-Lcbesgue lemma). If the Fourier transform does not vanish in the
large momentum limit, we define the time ordering T by subtracting it
The subtraction term is generally a polynomial in k%.

When we apply the BJL prescription to eqns (8.153) and (8.155), we obtain
The derivation of the second relation in eqn (8.159), for example, proceeds as
follows: We have a time derivative of the (J-function on the right-hand side of
but such a term does not appear in the time derivative of the left-hand side
if it is defined in terms of the T product, as can be confirmed by considering
KAOMOODY ALGEBRA AND VIRASORO ALGEBRA 177
the Fourier transform of both sides. Such a term on the right-hand side is thus
eliminated when one converts the left-hand side to the T product. In contrast,
the spatial derivative of a <5-function on the right-hand side does not contradict
the T product on the left-hand side and thus it remains on the right-hand side
of the first relation of eqn (8.159). The relation with the term dtl(Tja>1(x)jbl(y)}
is in general inflicted with the Goto-Imamura- Schwinger term we mentioned in
Chapter 6 and thus its analysis is involved, but a careful analysis shows that, it
gives a relation equivalent to the first relation in eqn (8.159). Namely, the chiral
anomaly and the Goto-Imamura-Scrrwinger term gives the same effect in the
current algebra in d = 2 dimensions.
The identities (8.159) are summarized in terms of operator language as48
where we used j b i ( y ) = ijf (y) and J 5 cl (z) = i j c 2 ( x ) , and d^ = 8^ = 0

for free fermions. We also used the fact that the time derivative acting on the T
product symbol converts it into a 5-function, for example,
by noting
When we define
eqri (8.161) gives rise to equal-time commutation relations
If one understands that these relations arc written for a theory defined on the
surface of a cylinder
and if one defines for an integer n

48
Thc relations (8.159) are valid for the path integral with source terms for fermions
added, namely, for general malrix elements. They are thus converted to the operator state-
ments (8.161).
FIG. 8.1. Time arid space coordinates in the complex plane
in the first relation of eqn (8.164). for example, one obtains
which is called the Kac Moody algebra. The last c-rmrnber term which arises
from the chiral anomaly gives a central extension of the algebra.
8.4.2 Method in conformal field theory

To compare the above result of the Kac-Moody algebra with the derivation in
conformal field theory, we define the complex coordinates
in Euclidean theory. We also understand that, the above coordinates are in fact
obtained from the coordinates (£1, C 2 ) on the cylinder, —TT < C1 < if and —oo <
C2 < oo, by the projection
on the entire two-dimensional plane. Sec Fig. 8.1. The point £2 = — oo is then
mapped to the origin and the point £2 = oo is mapped to z = oo, and the time
development is described by the clock which proceeds along the radial direction
starting from the origin (denoted by T in Fig. 8.1). The spatial freedom is then
described by a circle with center at the origin (denoted by x in Fig. 8.1). This
is called radial quantization. As a result, the field variable i/j(x), for example, is
obtained from the variable on the cylinder by a projection, and thus its definition
in general differs from the variable naively defined in two-dimensional theory.
KAC-MOODY ALGEBRA AND VIRASORO ALGEBRA 179
For general vector quantities, we define
When one remembers the definitions of Dirac matrices
the currents for fermions are written as (by noting ip = ^7°)
where i/>i, 02 are respectively the first and second components of the spinor ip(x).
See also eqn (8.163). In conformal theory ip and ^ are treated as independent
variables unlike the covariaiit formulation with the variables tp and ip so far in
this chapter. As a special property of two-dimensional theory, the axial-vector
and vector currents are directly related to each other by (see eqn (8.152))
In terms of the present notation, the identities (8.153) and (8.155) for j^
and jatt are written as (by using tzz = —*/2)
By considering the sum and difference of these "two identities we obtain
When one applies the BJL prescription to the present problem, the identi-
ties (8.175) are written as
In the first identity we regard z as the time variable, and we directly obtained
the T product since there are no terms which diverge when the momentum
conjugate to the coordinate z becomes large. In the second identity we regard z
as the time variable, and the time derivative of the 5-function on the right-hand
side is absorbed by the left-hand side when defining the T product.
To relate those identities (8.176) in conventional field theory to the operator
product expansion (OPE) in conformal field theory, we use the relation
which is explained in eqn (8.183) later and we write the first relation in eqn
(8.176) as
where we used the fact that d=jcz(x) = dzjcz(x) — 0 holds for free fermions, as
is seen in eqn (8.181) below. In general the relation dsF(z,z) = 0 implies that
F(z.z) does not depend on z and F(z,z) has no poles in z. Consequently, we
have the relation from eqn (8.178) which shows that both sides have the same
pole structure
and similarly
In conformal field theory we consider the Euclidean Lagrangian

The equations of motion then imply dzipi = 0 and Qî = 0. If one uses the
complex coordinates z = (xl + ix2)/2, w — (yl + % 2 )/2 in the definition of the
Green's function in Euclidean theory
one obtains the Green's function in two-dimensional theory without boundaries
When one combines eqns (8.181) and (8.183), one obtains
which is consistent with the Minkowski relation

y)+] in eqn (8.100) if one remembers t/»i = (V-^7°)i = V4- If orie considers
a contour integral with respect to z along an infinitesimal circle around, w in
eqn (8.184), one obtains (since the T* product in eqn (8.184) is replaced by the
T product by the BJL prescription)
One can convert this integral to a difference of two contour integrals along the
circles \z = w ± f. around the origin z = 0 if one recalls that the correlation
function has a singularity only at z = w. Since the Grassmann variables anti-
commute, one thus obtains in the limit c —¥ 0
which gives (a part of) the canonical quantization condition of the fermion.
This relation (8.186) is generalized for two arbitrary operators A(x) and B(y)
as
where ± corresponds to Fermi or Bose statistics. The function f ( z ) is a test

function with possible singularities only at z = 0 and z = oo. One can thus
recover all the information about equal-time commutation relations once one
knows the pole structure of the operator product expansion of two operators.
We now come back to the current algebra for the current
By using (T^2(z)ipl(w)} = i/[4ir(z — w)] and counting all the possible combina-
tions (or contractions) of tp and ip^, one obtains the operator product expansion
which agrees with the identity (8.179) in conventional field theory including the
c-rmmber term. The symbol ~ implies that the pole structures of both sides are
equal. A contour integral after multiplying the test function f ( z ) gives
if one defines
by choosing f ( z ) = zn, one recovers the Kac-Moody algebra
after multiplying eqri (8.190) by wm and contour integrating around the origin.
8.4.3 Virasoro algebra
We show that the central extension of the Virasoro algebra is a manifestation
of the quantum anomaly. The analysis is somewhat lengthy and involved, but
our main purpose is to show that the central extension of the Virasoro alge-
bra arises from the general coordinate anomaly when one defines a traceless
energy-momentum tensor. We consider an action for a scalar field X(x) in two-
dimensional Euclidean curved space-time
which is invariant under the Weyl transformation

Since X(x) is invariant under the Weyl transformation, the Weyl invariant path
integral measure is denned in terms of X (x) and the path integral is given by
To specify the measure more definitely we define
and expand
The measure is then given by
When one defines (pn(x) = (g)l^tpn(x) one has
The identity for general coordinate transformations
is given just as for other identities we discussed so far by
The first equality shows the fact that the path integral is independent of the
naming of integration variables, and the last expression is obtained from the first
expression by noting the invailance of the action and evaluating the Jacobian.
The identity is thus obtained by comparing the last two expressions
where the energy-momentum tensor T^v is defined by keeping the integration

variable X(x) fixed as
The tensor T^ thus defined satisfies the relation
identically in d = 2. The Jacobian is evaluated as
In the conformal gauge which is specified by g^ = p(x)r]IJ/v, we have
and the actual calculation of the Jacobian becomes identical to that of the Weyl
anomaly for the scalar field in Section 7.6. For the general case with d scalar
fields Xa(x), the anomaly is multiplied by d.
In this way the identity is written as
where the covariant derivative is given by using the affine connection F^ given
in Appendix B
We next derive the correlation functions of !),„. For this purpose we define
an infinitesimal e^"(x) by
where ^"(x) is assumed to be symmetric and satisfies r/ MJ /e MZ/ (a;) = 0. (In the
present discussion of the Virasoro algebra, we use e^v(x) to denote the infinites-
imal deviation of the metric ffv from the flat metric. Elsewhere in this book e'i!/
KAC MOODY ALGEBRA AND VIRASORO ALGEBRA 185
is used for an anti-symmetric symbol.) In this case we have det g = 1 to accuracy

linear in t^ (x]. To the same accuracy linear in ^ (x) we have
and we can write the energy-momentum tensor (8.203) as
where Tji,, is zeroth order in e MI/ (x). We can then write
From this equation we obtain (by writing T^J on the right-hand side simply by
Tpv for notational simplicity after setting eî,(x) = 0)
In the following equations, all TMl, stand for quantities zcroth order in efH/(.7;).
By using the definitions of the Ricmanri-Christoffcl curvature tensor and the
quantities Rpv = Rapal, and R = g^R^v in terms of the affine connection F$
(see Appendix B), we have
We next functionally differentiate both sides of the identity (8.207) with re-
spect to eai3(y} and use the fact that another Tctfj(y) appears from the functional
differentiation of the action on the exponent in the path integral. By considering
the flat space-time limit e^v (x) = 0 after the functional differentiation we have
We here recall that the second-rank tensors are generally written in complex
coordinates z = (xl + ix2')/2, z = (xl - ix2)/2 as
In complex coordinates, the tracelcss property of T^ is expressed by
and the non-vanishing components are given by Tzz = 8ZXOZX and Tzz —
dzXdsX, which satisfy in flat space-time
The above identity (8.215) which contains the correlation functions of Tp,v is
written in complex coordinates by setting v = z,a = j3 = was
In this expression, if one notes dzTzz = ~(1/2)8ZTZZ and interprets z as time and
z as space coordinates, the time derivative does not appear on the right-hand
side and the correct T product is defined. If one uses
in the identity (8.219) and writes the expression after one removes the operator
d, from both sides, one obtains
which expresses the fact that the pole structure of both sides is identical. In this
last expression we give a result valid for a theory with d scalar fields Xa(x).
This relation has been derived from the identity with the anomaly, and it gives
a relation corresponding to the operator product expansion in conformal field
theory. It is important to recognize that the last c-number term arises from the
anomaly associated with general coordinate transformations when one imposes
the Wcyl invariance.
On the other hand, the path integral in conformal field theory is defined by
and one obtains the correlation function by taking cqn (8.183) into account
and thus
which is also valid for a T product. The traceless energy-momentum tensor TzZ =
0 is given by (see eqn (8.217))
or TZz = d'gXdzX, and the operator product expansion of the energy-momentum

tensor is calculated as
This relation including the last c-numbcr term agrees with the identity (8.221)
with d = I derived in conventional field theory.
By multiplying the test function £,z(x) by the identity (8.221) or (8.226)

derived by the operator product expansion and contour integrating with respect
to the variable z around an infinitesimal circle around w, one obtains
If one defines
by choosing £ 2 (x) = zn+1, and if one contour integrates (8.227) by multiplying

wm+\ one ob|;ains the Virasoro alebra
where the last term stands for the central extension. Note that the central ex-
tension in eqn (8.221) arises from the general coordinate anomaly.
The essential point in the present consideration is that the central exten-
sions of the Kac-Moody and Virasoro algebras give the algebraic manifestation
of chiral and general coordinate anomalies. In this connection, the conserved
energy-momentum tensor T^. which is defined by a variational principle with
the variable X = (g^^X kept fixed (see Appendix B), is related to the traceless
tensor TIW by
and the conservation condition D^f^ = 0 gives £"%„ = -(l/48n)dvR (8.207).

On the other hand the operator 7),,, which is conserved contains the trace
anomaly T^ = J?/(24?r).
8.5 Quantum theory of strings and Liouville action

In this section we discuss the basic aspects of the first quantization of a bosonic
string. The Polyakov-type Lagrangiau of a string is given by
where Xa(x'), a — 1 ~ d, stand for the coordinates of a bosonic string in

d-dimcnsional Euclidean space-time. The two-dimensional internal coordinates
(x 1 ,^ 2 ) = (cr, r) stand for the parameter cr, which describes the spatial exten-
sion of the string, and the parameter r, which describes the extension in the time
direction (or time development). See Fig. 8.2.
QUANTUM THEORY OF STRINGS AND LIOUVILLE ACTION 189
FIG. 8.2. Free motion of a closed string

A naive path integral which defines the first quantization of a string is given
by
We present a precise definition of this path integral below. Formally, this path
integral is equivalent to a theory of d scalar fields Xa(x) defined in a curved
two-dimensional space. A characteristic property of string theory is that the
two-dimensional space takes various topological shapes. A two-dimensional space
with a cylindrical shape describes the free motion of a closed string, and a two-
dimensional space with a doughnut in the middle describes the motion of a string
with higher-order (one-loop) quantum corrections. To perform the path integral
on the space with holes (or handles) is an important but in practice very difficult
problem in string theory.
Here we discuss the quantum effects (namely, quantum anomalies) instead of
those topological issues. The BRST invariant path integral measure with respect
to general coordinate transformations in curved two-dimensional space is given
by
Sec cqn (B.41) in Appendix B. The actual calculation is performed in the con-
formal coordinate defined by
where 77^ is the flat metric. In this coordinate the variables with weight factors
become
The conformal coordinate conditions are written in terms of metric variables as

and thus the gauge conditions are fixed by the Lagrangian
When one uses the equations of motion for BI and B%. one obtains the gauge
conditions (8.236) written in terms of variables with weight factors in eqn (8.235).
When one defines the new anti-ghost variables
and similarly iBzz = BI +iB^ and —iBzS = BI —iB^, the BRST invariant gauge
fixing Lagrangian which contains the Faddeev-Popov ghost fields is given by
In deriving £g, we used the supcrfield notation derived from eqn (B.34) in Ap-
pendix B
If one sets /j• — v = z in this expression, for example, the first term gives gzz
in the gauge fixing term, and the term proportional to 0 gives the last term in
eqn (8.239) proportional to bzz. After the integration over the auxiliary fields Bzz
and Bzz in Cg, we have gzz = g^z = 0 and gzz — -2^/p. If one uses these relations
in the Faddeev-Popov ghost parts, the gauge fixing Lagrangian is simplified as
although the BRST symmetry is no longer manifest.

In this way the path integral for a bosonic string is given after the integrations
over Bzz, Bsz and then over gzz, g^ by
An important property of this path integral formula is that the Weyl freedom
p(x) explicitly appears in the action as a result of the choice of the BRST in-
variant path integral measure (8.233) with respect to general coordinate trans-
formations.
When one makes the change of variables (the Weyl transformation)
where
the Weyl freedom p(x) is extracted from other variables in the action in eqn
(8.242). To evaluate the Jacobian for this transformation, we fix the hermitiaii
operators, which define the basis vectors to expand the integration variables, by
(using the calculational method of covariant anomalies)
The operators for the variables bzz(x) and cz(x) are given by replacing z by z in
HI, arid HC, respectively, and it is shown that the Weyl anomaly does not change
under this replacement.
All these operators in eqn (8.245) are obtained from the general operator
by choosing the parameter n at n = 0, -2, 1, respectively. Consequently, the

general calculation of the Weyl anomaly for an infinitesimal parameter a(x) is
given by
where we defined
and used a generalization or the gauge invariant mode cut-off. This calculation
agrees with the one already performed in eqn (7.101), and its result is given by
where R is the scalar curvature and we used ^/g = p and ^/gR = —d^d^ \np =
dzdzlnp. From the viewpoint of the Weyl anomaly, the term ^/gM2 could be
removed by adding a suitable local-counter term (the cosrnological term) to the
starting Lagrangian.
The integrated Jacobian by extracting the freedom p from the action is eval-
uated as follows: We start with the intermediate stage after some Weyl trans-
formation where p = cxp[(l — s)a], and we apply a further infinitesimal Weyl
transformation ds a and integrate the resulting anomaly factor (by taking into
account the weight factors in cqn (8.243) and also the fact that b and c arc
Grassmann numbers) as
The path integral of the bosonic string (8.242) is then given by

where the Liouvillc action SL is defined by
In the last term of SL, which corresponds to the cosmological term, we give
the result after adjusting the coefficient to be a finite constant \i? by adding a
suitable counter-term to the original string Lagrangian. From eqn (8.250) one
might think that the main part of the Liouville action is subtracted by a local
counter-term. If one uses the gauge condition other than the conformal gauge,
the main part of the Liouville action is written as
which is not local, and thus it is not subtracted away by a local counter-term.
A specific quantum bosonic string at d = 26, which is defined without con-
sidering the Liouville freedom p ( x ) , is called a critical string and d = 26 is
called the critical dimension. (When one considers the superstring, which con-
tains the same number of bosonic and fermionic freedoms, the critical dimension
is given by d = 10.) On the other hand, the string theory for d < 26 is called
a non-critical string and a BRST invariant quantum string theory is defined
only when one includes the Liouville freedom p ( x ) . In the quantum theory of
strings, it is convenient to consider the motion of string coordinates Xa(x) on
a two-dimensional parameter space, which is called the world sheet, and this
parameter space is treated as a Riemaim surface. One can naturally treat the
topological notion (genus) if one introduces a background c-numbcr metric for
this Ricmann surface. In the calculations so far we used the specific metric spec-
ified by the conformal gauge gîx) = p(x)r]ll,v^ but it is more natural to treat a
more general situation
where p'(x) carries the dynamical freedom and the background gfu, (x) carries the
topological information. The variables Xa(x), bzz(x), cz(x) are thus influenced
by the presence of the background g^' (x). Even for the critical string for d = 26
where the dynamical freedom p'(x) is discarded, the topological effects of the
background metric remain. For the non-critical string for d < 26. the freedom
p' (x) behaves like an extra scalar freedom in the path integral. It is thus expected
that one unit of Liouville action is induced by p ' ( x ) , and as a result, the coefficient
in front of the Liouville action (8.252) is modified to (25 - d)/(487r). It is in fact
known that this modified coefficient leads to a more natural quantum theory of
Liouville theory and thus to a quantum theory of non-critical strings.
8.6 Ghost number anomaly and the Riemann—Roch theorem

We now come back to the analysis of the quantum anomaly. When one considers
an infinitesimal ghost number transformation in eqri (8.242)
The associated Jacobian, which is obtained from the above calculation of the
Weyl anomaly, leads to an identity (by recalling that bzz and cz are Grassmann
numbers)
and similarly
and the sum of the right-hand sides of these two relations (3/^Tr)^/gR gives the
ghost number anomaly.
When one integrates the sum of these two identities over a closed two-
dimensional surface and uses the Gauss-Bonnet theorem, one obtains
where x = 2(1 — g) is the Euler number with g standing for the genus; g = 0 for
a sphere and g = I for a torus, for example. We thus obtain a two-dimensional
ghost number identity whose integral has a form analogous to the chiral anomaly
in the presence of the iristanton in four-dimensional non-Abelian gauge theory.
The index theorem related to the ghost number anomaly is written by denning
HcLpn = \^<pn, Hi,(j)n = Xîpn f°r the operators in eqn (8.245) and also for those
operators with the replacement z «->• z as
where we used the relation in eqri (8.247) with a = constant and the 1 : 1
correspondence of (pn and </>* for non-vanishing An for the operators Hc and H£
defined in eqri (8.245), and similarly for H* and HI,. If one defines the operators
GHOST NUMBER ANOMALY AND THE RIEMANN-ROCH THEOREM 195
which appear in eqn (8.242) the above index relation is written as
which has the same form as in eqn (3.77) if one introduces the two sets of
operators.
This index relation is known to correspond to the Riemann-Roch theorem in
the theory of Riemann surfaces. The Riemann-Roch theorem is thus understood
as the ghost number anomaly in string theory. The contents of this index relation
and the Riemann-Roch theorem are understood from.eqn (8.241) as follows. If
one uses the zero modes for the ghosts (f or cz in the BRST symmetry, the
conformal gauge conditions
are not modified. Thus the ghost zero modes correspond to the conformal Killing
vectors. On the other hand, the zero modes of the anti-ghosts bzz or bzz are or-
thogonal to the variations of gzz or gzz induced by coordinate transformations.
The zero modes of the anti-ghosts bzz or bzz thus generate the variations of
the metric which are not generated by coordinate transformations. This freedom
of the metric variations other than the coordinate transformations is called the
modulus. The number of zero modes of bzz and bzz thus corresponds to the num-
ber of the moduli parameters of the Riemann surface. The index relation (8.259)
then expresses the Riemann-Roch theorem.
9
INDEX THEOREM ON THE LATTICE AND CHIRAL
ANOMALIES
Recently there has been important progress in the treatment of Dirac fields in
lattice gauge theory and we can now discuss the chiral anomaly on the lattice in
a manner similar to that in continuum theory. In particular, we can discuss the
index theorem on the lattice. The analysis of the index theorem on the discrete
lattice itself has certain subtle aspects, but lattice theory deals with completely
regularized quantities and thus some of the subtle aspects in continuum theory
are now given a more rigorous basis. We explain that all the results of chiral
anomalies in continuum theory are reproduced in a suitable continuum limit of
lattice gauge theory, and thus we can give a uniform and consistent treatment
of both continuum and lattice theories.
9.1 Lattice gauge theory

We first briefly summarize the basic aspects of lattice gauge theory. We con-
sider a hypercubic lattice with a lattice spacing a in four-dimensional Euclidean
space-time. The gauge field (connection) is defined on the link in the /i direction
connecting the points h and n + ap, as in Fig. 9.1
and it takes a value in a gauge group, for example, SU(n) in our treatment. Here
fl stands for a unit vector in the fj, direction, jj, = 1 ~ 4. and Tb stands for a
generator of SU(n)
FIG. 9.1. Smallest plaquette specified by the fiv plane
196
LATTICE GAUGE THEORY 197
with the normalization tiTaTb = (l/2)<5 at , and g is the gauge coupling constant.
We also define
and the gauge transformation rule is given by
(9.4)
where U(u}(n)) takes a value in the group SU(n). In Fig. 9.1 we show the smallest
plaquette P^ lying on the plane specified by // and v. The Wilson action of the
lattice gauge theory is defined by
where tr stands for the trace with respect to the gauge freedom. We consider
a product of gauge fields defined on each link of a plaquette P^v in Fig. 9.1
by taking into account the direction of the link, and we define the action by
summing over all the possible P/j,v. It is confirmed that the action thus defined
is invariant under the gauge transformation (9.4) on the lattice.
In the limit of small lattice spacing a, we have
and thus the trace in the action gives
Consequently, the above lattice action in the naive, continuum limit a —>• 0 be-
comes
and the Yang-Mills action in the continuum is realized. Namely, the lattice gauge
theory defines a regularization by approximating the action of continuum Yang-
Mills theory, which has an infinite number of degrees of freedom, by a theory with
a finite number of degrees of freedom on the link. This regularization clearly pre-
serves the gauge symmetry, but the rotational symmetry of the four-dimensional
Euclidean space-time is spoiled for a finite a. The hypercubic symmetry defined
by a 90-degree rotation is preserved, and it is believed that the hypercubic sym-
metry is sufficient to recover the Lorentz symmetry in the continuum limit.
198 INDEX THEOREM ON THE LATTICE AND CHIRAL ANOMALIES
The path integral quantization of lattice gauge theory is defined by
where the integration is performed for all the link variables U/j,, and the path
integral measure is denned by a Haar invariant measure specified by the followin
rules
The function f(Ufl) is an arbitrary function of U^ and UQ £ SU(n) is an arbitrary

group element. Since Ut,U^ = I, the volume of the entire gauge field can be
normalized to be unity by j T>U = 1. Namely, the present lattice gauge theory
defines a compact gauge theory. For this reason, the path integral in lattice gauge
theory is finite even without gauge fixing and the quantization can be defined
without gauge fixing. This integration rule gives for SU(3), for example.
The vanishing of the first integral is understood by noting that the value of the
integral does not change when one applies independent gauge transformations,
which belong to the representations 3 and 3* of SU(3), to the indices i, j of
(Up)ij, respectively. One can understand the second integral intuitively by setting
j = k and summing over
9.2 Lattice Dirac fields and species doubling
The Dirac field is defined to be located on each lattice point and the action is
defined by
where the sum over x runs over all the lattice points. Also a1' = aft and the sum
is taken from fj, = 1 to /j, = 4. The last term in the action with the coefficient r
LATTICE DIRAC FIELDS AND SPECIES DOUBLING 199
is called the Wilson term, and its meaning will be explained later. The metric
convention is chosen as
This action is gauge invariant if one performs the transformation
simultaneously with the gauge transformation of the gauge field (9.4).

When a is small we have
and the above Dirac action in the naive continuum limit a —> 0 becomes
where the covariant derivative is defined by
In the naive continuum limit a —S> 0, the Wilson term with the coefficient r
vanishes. To understand the meaning of the Wilson term, we go back to the
action (9.12) with a finite a and consider the free Dirac field obtained by setting
r = 0 and II^ = 1. By performing the Fourier transformation
we rewrite the free fcrmioii part of the action (9.12) as
The propagator for the Dirac particle is given by the inverse of the factor between
•0(fe) and V-'(fc) in the momentum representation of the Lagrangian
As is well known in condensed matter theory which deals with the matters of
lattice structure, the momentum in lattice theory is defined in the fundamental
Brillouin zone
for each component specified by /K with yti = 1 ~ 4. When one considers the limit
a —> 0 in the above propagator (9.20) of the Dirac particle with all the momen-
tum components k/j, kept fixed within —n/(2a) < kfl < 7r/(2a). one obtains the
ordinary propagator in continuum theory with a mass mo
In addition to the conventional propagator, however, if one considers the limit

a. —T> 0 by choosing the first component k\ of the momentum at fci = TV /a + k[
with fixed k\ and the remaining components of kIL being kept fixed, one obtains
Since the components of the momentum run between — oo and oo in the con-
tinuum limit, this second propagator (9.23) also describes the propagation of a
conventional Dirac particle with massTOO. Namely, the lattice Dirac action which
was introduced to describe one Dirac particle in fact describes two physical Dirac
particles in each momentum direction when rotated back to the Minkowski met-
ric. In four-dimensional space-time there appear 24 = 16 Dirac particles alto-
gether. This phenomenon is called species doubling.
On the other hand if one keeps the Wilson term with r ^ 0 the Dirac prop-
agator in lattice theory is given by
and this propagator describes the ordinary Dirac propagator with mass m0 only
when one takes the limit a —>• 0 with all the momentum components fc,, kept
fixed within — ?r/(2o) < kp, < 7r/(2a). The propagators for other momentum
domains such as cqn (9.23) describe particles with mass of the order m<) + r/a
and thus with an infinite mass in the continuum limit. The extra particles are
thus eliminated from the physical Hilbert space in the continuum limit and the
problem associated with the species doubling is resolved.
LATTICE DIRAC FIELDS AND SPECIES DOUBLING 201
The species doubling is understood from the viewpoint of symmetry as fol-

lows: The lattice Dirac action without the Wilson term r = 0
is invariant under the global chiral transformation with a constant a
if one sets m,Q = 0. The above action (9.25) is also confirmed to be invariant
under the transformations of the Dirac field (even with mo 7^ 0)
Here Tn represents one of the following 16 operators
and
We define T^ for by
which satisfies the following relation
This algebraic property is the same as that of the four ordinary Dirac matrices
17^, and thus we can construct the above 16 independent operators from TM. For
the anti-hermitian 7^ in the present Euclidean theory, we have
We denote those 16 operators by Tn, n = 0 ~ 15, and define in particular

TO = 1- If one notices that the operator TM adds momentum -K/O, to ?/>(fc M ) in
eqn (9.18), one can cover the entire fundamental Brillouin zone by starting with
the momentum domain
which is called the physical domain, if one applies the operation (9.27) by using
Tn.
The important property of Tn is that eight operators in cqn (9.28) commute

with 75 and eight operators in eqn (9.29) anti-comrnute with 75 and thus change
the signature of the chiral 75 charge. Because of this property, eight species
douhlers give the chiral anomaly with a positive signature and the remaining
eight species doublers give the chiral anomaly with a negative signature when
evaluated in the theory with r = 0. In this way the chiral anomaly vanishes in
the naive latticized Dirac theory.
In other words, the mechanism which produces the chiral anomaly disappears
in a regularized theory with only finite degrees of freedom such as the lattice
theory. To be consistent with the fact that the chiral anomaly is inevitable,
the theory adjusts by itself such that it is anomaly-free. Namely, the 15 species
doublers are generated so that the sum of all the chiral charges vanish
This suggests that it is generally difficult to regularize a continuum theory and

preserve exact chiral symmetry by simply replacing the differential operator by
the difference operator in lattice theory. Namely one cannot avoid species dou-
bling by a naive discretization. This fact is known as the Nielsen-Ninomiya
theorem.
From this viewpoint, the ordinary chiral symmetry is broken by the Wilson
term in the lattice Dirac action with r ^ 0. The Wilson term is understood as a
generalized (infinite) mass term which breaks chiral symmetry, analogous to the
mass term in the Pauli-Villars regularization. It is known that the chiral anomaly
corresponding to one Dirac fermion is correctly reproduced in the continuum
limit by the lattice Dirac action with the Wilson term.
9.3 Representation of the Ginsparg—Wilson algebra

We write the action for a Dirac particle on the lattice in the general form
where D(x,y) generally contains the gauge field and the sum runs over all the
lattice points x. The Dirac operator D(x,y) is not ultra-local in general, namely,
the operator has a non-vanishing value even at points far apart on the lattice but
we assume that the operator D(x, y) decreases exponential!}' for a large value of
where ga with a constant Q is the localization range of the Dirac operator. For a
hennltiaii 75, we assume the following hermiticity property
when we regard D(x,y) as a matrix specified by the raw indices x arid column
indices y. In lattice theory the operator D contains both kinetic and (generalized)
REPRESENTATION OF THE GINSPARG-WILSON ALGEBRA 203
mass terms, and the operator can be hermitian only after being multiplied by
75; this property is shared with the continuum Euclidean Dirac theory with a
mass term.
The recent progress in the treatment of the lattice Dirac operator is based
on the operator which satisfies the algebraic relation
which is called the Ginsparg-Wilson relation19 where k stands for a non-negative

integer. This relation shows that the continuum chiral symmetry is modified by
the term on the right-hand side which depends on the lattice spacing a. In the
naive continuum limit a — ¥ 0. this relation is reduced to the ordinary continuum
relation
for a massless Dirac field. An explicit construction of the lattice Dirac operator
which satisfies this algebraic relation will be discussed later.
Here we discuss the general characteristic properties of the operator D, which
are implied by the algebraic relation (9.38). We analyze all the possible finite-
dimensional representations of the algebraic relation (9.38) by considering a her-
mitian operator
defined on a finite lattice. As the boundary condition we employ the periodic

boundary condition.
We first define the operator
which is hermitian by its construction, and the Ginsparg-Wilson relation (9.38)

is written as
This relation implies if
then
where the positive definite inner product of eigenfunctions is defined by
49
The relation corresponding to k — 0 is commonly called the Ginsparg-Wilson relation. We
can discuss the algebraic properties and chiral anomaly for all these operators in a uniform
manner.
with the summation running over all the lattice points. If Xn = 0. these relations
imply that H(j)n = 0 and thus H(j56n) = 0. namely, one has
We can thus choose the eigenstates with An = 0 as the simultaneous eigenstates

of
We next consider the case \n ^ 0. If \n ^ 0 and (F5<;6n, F B </i n ) ^ 0, the

eigenstatc <f>n with \n and the state Yc,<f>n with — \n always appear pairwise, as
is shown in cqns (9.43) and (9.44). We thus examine the norm of F5
Since the eigenvalue An of the hermitian operator H is real, the state <j>n satis-
fies the "highest state" condition familiar in the representation of the angular
momentum
only if 1 — (a\n)2 = 0. Here we used the fact that the inner product in Euclidean
theory is positive definite. We thus understand that the states tf>n with \n = ±l/o
do not give rise to paired states by the operation T^(j>n in general. We also see
from eqn (9.49) that these states are the simultaneous eigenstates
of 7,5. By using the relation valid for the states with
which is derived by sandwiching the Ginsparg-Wilson relation (9.38) by <£>,], and

4>n, one can show that these eigenvalues Xn = ±l/o are the largest or the smallest
eigenvalues of j&D. Namely.
Finally, the states with 0 < |An < I/a always appear pairwise as
and satisfy the relation
where <£>_-„ stands for the state with eigenvalue —Xn, and we note the relation
FsCFsaC) = [1 - (aXn)2(2k+ind).T, oc <£„. Since 0 < |aAJ < 1, these states
cannot be the eigenstates of
REPRESENTATION OF THE GINSPARG WILSON ALGEBRA 205
On the other hand, the relation tr~/5 = 0 generally holds on a finite lattice,
and it implies
where in the last line we used the fact that all the states with Xn ^ 0 except for
those with \n = ±l/a appear pairwise with positive and negative eigenvalues
±|A n | and thus cancel in the summation. Here N± respectively stands for the
number of states §n with eigenvalues \n = ±l/a and 75<pn = ±<pn- in this way
we conclude the chirality sum rule
This relation shows that the chiral asymmetry n+ — n_ in the eigcnstates with
zero eigenvalue is always balanced by the chiral asymmetry N+ — N- in the
states with the largest eigenvalues |A n | = I/a for the operator satisfying the
Ginsparg-Wilson relation.
The analysis in this section is valid both for Abelian as well as for non-Abeliau
theories. As for the boundary condition, we assumed the periodic condition for
simplicity, but the analysis is valid for more general boundary conditions when-
ever there appear non-trivial zero eigenstates. The analysis of the index (i.e.,
the difference n+ — ri_ of eigenstates with vanishing eigenvalue) in this section
is formal, since it is shown that the index or the chiral anomaly is not uniquely
specified by the Ginsparg-Wilsoii relation alone. We shall show this property
later in the present chapter.
To summarize the analysis of this section, the normalized eigcnstates of
7oD = H/a on a finite lattice arc classified into the following three categories:
(i) ri± states;
(ii) N-J- states (highest states,
respectively.
(iii) The remaining states 0 < |A n | < I/a (paired states):
and the chirality sum rule n+ + N+ = n_ + N- holds.

All the states n± and N± are also the eigenstates of the operator D. and
they satisfy D(f>n = 0 and D<pn = (l/a)<pn, respectively. If one denotes the total
number of states in (iii) above by 2A7o, the total number of fermionic states on
the lattice is given by N = 2(n + + N+ + NO), and the number N is independent
of the background gauge field configurations.
9.4 Atiyah Singer index theorem on the lattice and the chiral
anomaly
The starting path integral for the Dirac field in lattice theory is given by
where the operator D satisfies the Ginsparg Wilson relation (9.38) and the gauge
field is treated as the c-number background field. The action in the exponent
is invariant under the generalized chiral transformation with an infinitesimal
constant parameter
if one writes the Ginsparg-Wilson relation (9.38) in the form
with
which is shown to satisfy 75 = 1 by noting ^H2 - H2^5 = 0. Under the transfor-

mation (9.59) one obtains the following Jacobian from the path integral measure
(by remembering that the fermionic variables are described by Grassmarm num-
bers)
where FS is defined in eqn (9.41).

We have the following relation for the trace appearing in the Jacobian
by using the representation of the Ginsparg-W7ilson algebra in the preceding

section. Here n± stands for the number of zero eigenvalue solutions
ATIYAH-SINGER INDEX THEOREM AND ANOMALY 207
of the hcrmitian operator 75!? with 75^»n = ±(f>n, respectively, and n+ - n_ is

called the index of the operator 75!). We also used eqn (9.52). In lattice theory
the inner product is denned by a sum over all the lattice points (®n,&n} =
Y^xa4'<t>n(x)^(!)n(x)- To define the zero cigenmodes clearly, one may consider a
finite lattice L = Na = finite with N standing for the total number of lattice
points in the x-dircction, for example, of the hypcrcubic lattice.
The analysis of the index of the Dirac operator implies transferring the in-
formation about topological properties of the gauge field to those of the Dirac
field, namely, the number of zero modes of the Dirac field. Consequently, the
path integral measure of Dirac fields becomes non-trivial even in lattice theory
when one considers the index.
To show that the index (9.63) associated with the Jacobian agrees with the
Pontryagin index, we start with the relation
where f ( x ) is an arbitrary smooth function with /(O) = 1 and rapidly approach-

ing 0 at infinity.50 By using the result of the. representation of the Ginsparg-
Wilson algebra in the preceding section, one can confirm that the index n+ — n_
(9.63) is not modified for any smooth function f ( x ) with /(O) = 1. In our analysis
of the index, the hcrmitian operator 7,5!? plays a special role.
We now analyze a local version of the index
by omitting the sum (or integral) over the coordinates in the formula for the
index. The trace tr here stands for the sum over the Dirac and Yang-Mills
indices. Since the local version of the index is insensitive to the precise boundary
condition, we consider the infinite lattice limit L — Na —> oo in the following
analysis of the index. (The limit N —> oo is in any case required in the continuum
limit.)
We now consider the continuum limit a —> 0 in eqn (9.65), which corresponds
to a "naive" continuum limit in lattice gauge theory. We first note that
approaches 0. Because of the presence of the regulator function /(x), which

rapidly approaches 0 for large x (see Fig. 5.1), the eigenvalues of 75!) are cut-off
s
°Wc here present this calculational scheme to show that the lattice calculation in the contin-
uum limit is reduced to the continuum calculational schemes discussed in preceding chapters.
A more direct evaluation of the Jacobian (9.62) by the lattice formulation will be given later.
at ~ M. Consequently, the above expression (9.66) approaches 0 in the limit

a —> 0, or in terms of the integrated index it vanishes as
We thus analyze the small a limit of
where the operator is well-regularized by f ( x ) , and thus one can use the plane
waves for the Dirac fields to evaluate this trace and in this way one can extract
the gauge field dependence of a local version of the index.
The basic momentum domain for the hypercubic lattice is defined by the
Brillouin zone (9.21)
We also assume that the operator D blows up as ~ I/a for small a in the
momentum domain of species doublets, namely, the operator D is free of species
doublcrs. See eqn (9.20) for species doublets. (An explicit construction of D of
this property is given later.) Consequently, in the limit a —5- 0 with, for example,
in the momentum domain of the would-be species doublers which corresponds to

outside — ?r/(2o) < k^ < 7r/(2a). and it does not contribute to the trace because
of the regulator f ( x ) . We are thus allowed to consider only the momentum
domain for the physical species
in the evaluation of the above trace.

In the limit a —>• 0 we obtain
where we first considered the limit a —>• 0 with the momentum variables k^ fixed
in — L < ku < L, and then took the limit L —t oo. This procedure is justified
ATIYAH-SINGER INDEX THEOREM AND ANOMALY 209
if the integral is well convergent.51 In this calculation we also assumed that the
(difference) operator D in the limit a -> 0 satisfies
for an arbitrary fixed momentum fc;,,, —vr/(2a) < ktl < ?r/(2a), and for an arbi-
trary smooth function h(x). The function h(x) in the present calculation corre-
sponds to the gauge potential. The gauge potential is thus assumed to be smooth
and vary very little over the distances of the basic lattice spacing a. We shall
show that the explicit example of D given by Neuberger (and its generalizations)
in fact satisfies all these conditions in eqri (9.114) later.
In this way we arrive at the same formula as in Chapter 5. The formula (9.64)
combined with eqn (9.72) and eqn (5.104) gives
in the limit M —> oo. The index theorem in the presence of topologically non-
trivial gauge fields and the chiral anomaly in the continuum are thus recovered. It
is important to recognize that we used only the general properties of the operator
D, namely, the existence of a hcrmitiaii 75!) which satisfies the Ginsparg-Wilson
relation with a smooth continuum limit a —>• 0 without producing the species
doublers. This analysis shows that the chiral Jacobian (9.62) in lattice gauge
theory contains the correct chiral anomaly. (We however implicitly assumed that
the index defined in lattice theory does not change in the process of taking the
continuum limit.) A direct evaluation of the Jacobian for the Neuberger operator
will be given later.
We would like to add a comment on the present analysis of the index. The
relation Tr7s = 0 contains important information about the Hilbert space in
lattice gauge theory. In the perturbative calculation of the chiral anomaly in
B1
To be precise, we are dealing wi
f~^ /(2 dx fa(x) where fa(x) depends on the parameter a. (The generalization to four-
dimensional integrals is straightforward.) Consequently, we need to show that both
lim a _Kj/£/ a dx fa(x) and lim a _j,o J_^/(2 a ) ^x f&(x) cari ê made arbitrarily small for
large L. The integral we encounter in lattice gauge theory generally has a .structure
\ii^ôf"ff$a-)dxe~3in2aX/(a2M''i) = ^mrôo/^ê-^^ 2 and the convergence con-
dition is satisfied for the choice f(x) = e~x. Note that J^/(2a) dxtTsin2 ax/(a2M2) <•
J^ a dxe~c x IM for a sufficiently small constant c, and the right-hand side goes to
zero for a ->• 0 and then L -> oo. Incidentally, we have Iim0.^0 f^/ffi dx e"1/1"2^1'2'1 =
7r 0; e 1 l <1 M
a_>o( / ) ^ '' - •* = 0 for the momentum domain of the would-be species doublers be-
cause of our assumption about the property of the operator D, which was used in eqn (9.70).
lattice gauge theory one usually assumes the relation Tr7g = 0. In this case, the
relation
holds by remembering F5 = 7.5 - (aj5D)2k+L. Namely, only the states N± in

eqn (9.56) contribute to the Jacobian (9.62) if one uses the results in Section 9.3,
since other states give either vanishing contributions or cancelling contributions
of positive and negative values. When one combines the above expression (9.75)
with the chirality sum rule Tr7g = n+ — n_ + Ar+ — N- = 0 in eqn (9.53), the
correct index theorem Tr Fg = n+ — n~ is derived. This chirality sum rule also
shows that the relation Tr 7.5 = 0 has no well-defined meaning in the continuu
limit a —> 0, since one cannot count the number of states N±, which have in-
finitely large eigenvalues in the continuum limit, in a reliable way. It is in fact
shown that the states N± correspond to the heavy topological states related to
the would-be species doublers. in contrast to the physical topological states n±
which represent the states with the vanishing eigenvalue. Consequently, one has
the relation
in the physical Hilbert space with the species doublers excluded, which agrees
with the continuum analysis.
9.5 The operator D satisfying the Ginsparg—Wilson relation
An explicit form of the operator D, which satisfies the Ginsparg-Wilson rela-
tion (9.38) with k = 0, has been given by Neuberger and it has the form
It is confirmed that this operator D satisfies the relation

by a direct insertion of D, independently of the detailed form of the hermitian
operator HW- In the explicit construction of D, one uses DW = 7&Hw which
was discussed in Section 9.1 as the Wilson operator (9.12) for the Dirac field on
the lattice
but the mass term is chosen as mo —* rn^/a and the signature of the Wilson
term is reversed. Namely,
OPERATOR D SATISFYING THE GINSPARG WILSON RELATION 211
It has been proven that the operator D thus constructed satisfies the locality
condition in the sense of eqn (9.36) for sufficiently smooth gauge field configura-
tions, though we do not give the proof here. One way to specify the smoothness
of gauge fields is to impose the so-called admissibility condition
on lattice gauge fields, where \\O\\ is the operator norm and e is a certain small
fixed number. This condition divides the space of lattice gauge fields into topo-
logical sectors; the index (9.63) takes a well-defined value for each topological
sector. The necessity of such a restriction on gauge field configurations is thus
suggested by the consideration of the index theorem (9.63) also.
The chiral Jacobian (9.62) in the preceding section is given for this operator
by
The conventional calculation of this Jacobian and the chiral anomaly in lattice
gauge theory proceeds as
where we used the relation Tr75 = 0 valid on a finite lattice. The local version
of the index (i.e., the chiral anomaly) is given by
where the trace runs over the Dirac and Yang-Mills indices. The factor D\y x
rD I D — is a phase factor, and its behavior in the continuum limit is somewhat
V w W
subtle. But this calculation illustrates an interesting aspect of the lattice calcu-
lation, and we present its details below.
9.5.1 Explicit lattice evaluation of the chiral Jacobian
We first note
where B is the Brillouin zone (scaled by a)
We then move the plane wave Klkx/a through the operator to the left-hand side
of the operator. For the Wilson operator, we have the identity
where f(x) is a smooth function, and the operators QM and R are defined by
and
Thus, by using we have
where we have used our Euclidean metric convention

With this form of the ohiral Jacobian, it is straightforward to find the naive
continuum limit by noting the following facts. Firstly, the operators Q^ and R
are of O(a°) in the naive continuum limit (9.15),
and thus
OPERATOR D SATISFYING THE GINSPARG- WILSON RELATION 213
Secondly, the trace over Dirac indices requires at least four gamma matrices; see
eqn (5.24). Also it is sufficient to retain terms to the magnitude O(ai) in the
integrand.
After some rearrangements, we have
where the equality holds in the limit a —i 0. The coefficient J(mo,r) is given by
the integral
By comparison with our previous result (9.74), one expects that I(mo,r) = 1 if
the Dirac operator D is free from species doubling, namely, for 0 < mo/r < 1
as will be explained in the next subsection. We now confirm this expectation by
evaluating /(mo,r) explicitly.
We first change the integration variables from fcM to
and split the Brillouin zone (9.85) into -vr/2 < k,L < K/2 and vr/2 <kll< 3?r/2
in each direction. We thus split B into 24 = 16 domains. Then we have
where
The four quantities e^ which assume the values cfl = ±1 specify the 16 domains
in the Brillouin zone and we sum the contributions from those 16 domains.
We can first confirm that I (mo, r) is invariant under an infinitesimal change

of parameters mo and r. This follows from the identities
and
By integration by parts, we then have (d/dmQ)I(mo,r) = (d/dr)I(m,0,r) = 0.

The surface terms do not contribute because they are independent of e^ and thus
vanish due to J^f =±i(n,u £ M) = 0- Once the invariance of I(mo,r) has been
established, we may evaluate /(TOO , r) by considering some convenient limiting
cases of the parameters.
We set a. = mo/r and regard /(TOO,'') as a function of a arid r. After the
rescaling s^ —>• rs M , we have
One can see that the integrand becomes singular near s2p ~ 0 only when a = 0,
2. 4, 6 and 8. Therefore, by fixing a avoiding these values, we may consider the
r —> 0 limit as a convenient limit of parameters. It is shown that the r —> 0 limit
of eqn (9.99) is given by first taking the r —> 0 limit in the integrand and then
taking the r —>• 0 limit of the integration domain.52 Therefore, we have
52
To show this, we first divide the integration domain [— 1/r, 1/r] into a four-dimensional
cylinder of size L with L < 1/r, C(L) = Ss x [-L, L], and the remaining part R(L).
The radius of S'3 is L and the direction of the cylinder is taken along the /^-direction
of the numerator of eqn (9.99). If the r —f 0 limit of the integral over R(L) vanishes
for L —j- oo, lim^-joo lim,—>o I(cvr,r)\R(L<f — 0, our assertion that lim,,_s.o I(ar, r) =
limi.-j.oo linv—x) '(o"v)lc(Z,) holds. Now, one can see tha,t \l(oir,r)\R^L) is bounded by a
linear combination of
OPERATOR D SATISFYING THE GINSPARG-WILSON RELATION 215
By examining the last expression, we finally obtain
where 6(x) is the step function. As we expected, when

and the present direct calculation of the chiral Jacobian (9.92) coincides with
our previous analysis (9.74).
9.5.2 Physical meaning of the operator D

The definition of the Ncuberger operator D (9.77) is somewhat abstract, and
to clarify its physical meaning we consider its continuum limit by the following-
two steps. This analysis also supplements the explicit calculation of the Jacobian
presented above. We first take the limit a —5- 0 with
kept fixed, and then consider the continuum limit by letting 2r/a, m 0 /a -> oo.
This two-step procedure is known to simplify perturbative calculations in the
lattice theory defined by the Wilson operator.
and fL ' /T dp f_L dz with the same integrand. After performing the integration over p, these
integrals are bounded by
Similarly, the integral over fL dpf^Ldz is bounded by 47T/L. In this way,

lim£,_j.i:x; limr_,,o I(ar, r)1 R(!,) = 0 is established.
We represent the fundamental Brillouiri zone in momentum space by starting

with the basic domain
and applying the operators Tn, n = 0 ~ 15, defined in eqns (9.28) and (9.29).
We first observe that the relation for Dw in cqn (9.79)
holds in the continuum limit a —> 0 for any fixed k/j in the domain
kp. < 7r/(2a) and fixed MO = —mo/a and for a sufficiently smooth h(x). We next
define
where we used the fact that @ = 7^CM in eqn (9.79) and Tn commute. The
second term on the right-hand side of (Dw)n gives the mass term generated by
the Wilson term for a -> 0 with &M kept fixed in — 7r/(2a) < &M < 7r/(2a). If one
recalls the momentum representation of DW with the vanishing gauge field
the explicit expression of the 16 mass terms Mn is given by MO = —rrio/a and

one of the following 15 expressions
for n = 1 ~ 15. We indicate inside the brackets of eqri (9.109) the (multiplicity,
chiral charge) of the would-be species doublers for each mass parameter. For
example, Mn = 2r/a — mo/a appears four times with the chiral charge —1.
These multiplicities and chiral charges appear as the coefficients in cqn (9.103).
Consequently, the value of (Dw)n (9.107) in the continuum limit a -» 0 is
given by
If one recalls the definition

OPERATOR D SATISFYING THE GINSPARG-WILSON RELATION 217
one obtains in the same limit as eqn (9.110)
When one chooses the mass parameter such that
namely 0 < mo < 2r, one can show that eqn (9.112) gives rise to
for any k^ fixed in —7r/(2a) < /% < 7r/(2a) in the limit a—>• 0 now with IT and
mo kept fixed. In the last expression we chose the parameters as 0 < 2mo =
1 < 4r. The property (9.114) provides sufficient information to prove the index
theorem (9.74) in the preceding section.
These considerations, namely, the explicit evaluation of the Jacobian and the
present analysis of the continuum limit show that the Ginsparg-Wilson relation
by itself does not fix the coefficient of the chiral anomaly uniquely, since the
Ginsparg- Wilson relation (9.38) with k = 0 is satisfied independently of the
details of the hermitian operator H\y in eqn (9.77), whereas the chiral anomaly
depends on the details of HW- For example, if one chooses mo < 0 then MO > 0
and thus 70~1D'7o ~ I/a, and the chiral anomaly vanishes. See eqn (9.103).
When one imposes the condition 0 < mo < 2r which ensures only the physical
species in D and the absence of species doublers. the chiral anomaly is uniquely
fixed.
In passing, we would like to briefly explain the construction of the solution
for eqn (9.38) with general k. We first note that the condition (9.38) is equivalent
to the two relations
When one defines H^k+i) = H2k+1, one has
which has the same form as eqn (9.38) with the simplest choice k = 0. One can
thus construct the solution following eqn (9.77) as
where we denned
by using eqri (9.79). One may thus define the Dirac operator H by
in the representation where H(2k+i) is diagonal. If one chooses 0 < m 0 < 2r and
one obtains a lattice Dirac operator which is free of species doublers and gives
the correct Dirac operator iSfi in the (naive) continuum limit. The general Dirac
operator thus constructed gives the correct index theorem and chiral anomaly
as was shown in Section 9.4. But the locality condition (9.36) for k / 0 has not
been proved though we expect the locality of D for a sufficiently smooth weak
background gauge field, since the locality for the vanishing gauge field is proved
by an explicit analysis of D in the momentum representation.
9.6 Some characteristic features of lattice chiral theory

The Dirac operator on the basis of the Ginsparg-Wilson relation thus resolved
many of the basic issues associated with fcrmions on the lattice. We would like
to comment on several important properties of this formulation.
First of all, we have noted the existence of the chirality sum rule
where AT± in eqn (9.56) stands for the number of states with eigenvalues
with 756n = ±<pn- Also we explained the index relation
for the trace Tr constrained to the physical Hilbert space, and that this index
relation has a smooth continuum limit a. —>• 0 as was discussed in eqn (9.72).
This phenomenon is the same as the behavior of the index discussed for the
phase operator of the photon in Section 3.5. Namely, the index vanishes Tr 75 =
in the truncated Hilbert space, but a non-trivial index generally appears after
the elimination of the cut-off.
The Dirac field exists only on the lattice point in a discretized theory such as
the lattice theory, and thus it is not obvious if the notion of index in the Atiyah-
Singer index theorem is rigorously defined. The Atiyah-Singer index theorem is
valid for a continuum Dirac equation with non-Abelian gauge fields defined on,
for example, the smooth compact four-dimensional space such as the 4-sphere S4.
The analysis of the index in this chapter shows that one can formulate the index
theorem in the continuum limit of the lattice Dirac operator for sufficiently
SOME CHARACTERISTIC FEATURES OF LATTICE CHIRAL THEORY 219
smooth gauge fields, and that the result is confirmed by detailed perturbative
and non-perturbativc calculations. In this sense, the notion of the index theorem
is valid on the lattice also.
The notion of the index is somewhat subtle on the lattice, but the treatment of
the Jacobian in the lattice path integral is based on well-defined finite quantities
only. One can thus analyze the chiral Jacobian and the anomaly on the basis of
mathematically well-defined quantities.
9.6.1 Chiral gauge theory and fermion number anomaly

When one defines
by using D which satisfies the Ginsparg-Wilson relation (9.38), one can confirm
•yj? = 1 by using eqn (9.115). One can thus define the projection operators
with
One can also confirm that the operator satisfying the Ginsparg-Wilson rela-
tion (9.38) is decomposed into left- and right-handed components as
As a result one can define the right-handed or left-handed Dirac field by
One of the important implications of this decomposition into chiral compo-

nents is that the pseudo-scalar operators Pa(x) representing the pseudo-scalar
particles such as TT, r\ and t]' mesons in QCD are given by
where Xa, a = 1 ~ 9, standing for the Gell-Mann matrices of flavor SU(3) and
the unit matrix, which express the flavor freedom of three light quarks. In this
construction, the states N± in eqn (9.56) are automatically eliminated by the
presence of the operator r 5 .
The chiral theory which breaks parity such as the Wcinberg-Salam theory is
defined by the Lagrangian03
and the path integral on a finite lattice is given by
A lattice non-perturbativc formulation of chiral gauge theory has many potential

applications, but its final formulation for non-Abeliari gauge theory has not been
given yet. We here briefly sketch the basic idea of how to define the path integral
measure in the present scheme.
The functional space in our problem is naturally spanned by the eigenfunc-
tions of the basic hermitian operator
However this eigenvalue equation is gauge covariant as are all the quantities
in the gauge invariant lattice regularization. To accommodate the gauge non-
covariant quantities such as a consistent form of anomaly, one defines the path
integral in a specific topological sector specified by M by
where we expanded fermionic variables as
The basis vectors {wn} and {wn}, which satisfy
are suitable linear combinations of {(/)„,} and {4>\,}-, respectively; wn oc P-6n and
wn K-<j>llP+. See also eqn (9.52). The measure factor 7?^ stands for the Jacobian
for the transformation from the ideal bases {vn} and {vn}, which automatically
ensure the Wess- Zumino intcgrability condition, to the bases specified by // and
thus <&M crucially depends on the ideal bases.54 In practice, the ideal bases are
53
Note that the basic lattice operator D ( x , y ) is always vector-like and we do not define the
generators T° (1 — 75)/2 for SU(2)]J, for example. We define the chiral theory by the projection
operators P_ and P+.
°4The measure factor is thus chosen to be a constant for the ideal bases.
SOME CHARACTERISTIC FEATURES OF LATTICE CHIRAL THEORY 221
not given explicitly and one determines the measure factor by imposing several
physical conditions.
The lattice path integral (9.131) corresponds to the formulation of the covari-
ant anomaly in continuum theory in eqn (6.75). When one considers the change
of fermionic variables which corresponds to the gaug_e transformations (including
the U(l) phase transformation), -0 -» 1/1' and t/5 ->• '<//, the expansion coefficients
with the fixed basis vectors are transformed as {an} —>• {a'n} and {an} —» {a'n}-
Since the naming of path integral variables does riot matter, one obtains the
identity
In this form of identity, the Jacobian of the path integral measure gives a lat-
tice version of the covariant anomaly and the variation of the action gives the
divergence of the covariant current.
As an important application of this derivation of the anomaly, one can calcu-
late the gauge invariant ferrnion number anomaly in chiral gauge theory. When
one considers the fermion number transformation with a constant a
the Jacobian is given by (by noting that both of (1 — 7s)/2 and (1 + 7s)/2 are
projection operators)
which shows that there appears an asymmetry between the number of freedoms
Tr(l — 7s)/2 describing the particle ibi, and the number of freedoms Tr(l + 7s)/2
describing the anti-particle 4'L- We also used the index in eqn (9.63). There
thus appears an asymmetry between fermions and anti-fermions proportional
to the index when one defines chiral fermions on the lattice. As was discussed
in Chapter 6, the fermion number is not conserved in chiral gauge theory in
the presence of the instantori background. The proton can thus decay in the
Weinberg Salam theory. The fermion number anomaly is naturally built-in in
the present lattice formulation of chiral gauge theory.
If one performs the simultaneous gauge transformation of the link variables U
in the above path integral ZM(U) (9.134), the action becomes invariant but one
needs to take into account the variation of the measure factor MM induced by the
gauge transformation of U. This variation S'&M converts the covariant anomaly
given by the Jacobian to a lattice form of consistent anomaly, which is one of the
requirements on the measure factor. In the theory with an anomaly-free gauge
group, the measure factor should completely cancel the non-vanishing Jacobian
arising from the lattice artifacts. Unlike the continuum theory, the Jacobian in
lattice theory does not vanish completely even for an anomaly-free gauge group
for a finite lattice spacing a. The current associated to ST^M should be local
and satisfy several other requirements: The existence proof of such a measure
factor $M amounts to a definition of lattice chiral gauge theory.
The construction of the path integral measure with the desired measure factor
for chiral non-Abelian gauge theory has not been given yet, and interested readers
are referred to the references at the end of this book.
10
GRAVITATIONAL ANOMALIES
In this chapter we discuss chiral anomalies associated with the gravitational

field. We first discuss a gravitational generalization of chiral U(l) anomalies
for ordinary gauge fields. We thus give a method to evaluate the Dirac genus
and Chern characters in general 2n-dimensional Euclidean space-time. We next
explain that there generally appear quantum anomalies in general coordinate
and local Lorentz transformations for chiral Dirac fields in (4n + 2)-dimensional
curved space-time, and thus the basic principles of general relativity and quan-
tum theory become incompatible with each other if these anomalies arc not
cancelled.
10.1 Chiral U(l) gravitational anomalies

In this section we study the Dirac field in the presence of the background grav-
itational field and QCD-type SU(JV) gauge fields. We consider an arbitrary 2n-
dimensional (Euclidean) space-time. The action for the Dirac field is given by
(see Appendix B)
This action except for the last mass term is invariant under a global chiral
transformation
with a constant parameter a. Here 72™+1 is a generalization of Dirac's 75 in

four-dimensional space-time to 2n-dimensional space-time, and it is defined as a
product of all 2n Dirac 7 matrices
All the In 7-matriccs are anti-hermitian iri our convention
223
224 GRAVITATIONAL ANOMALIES
and we define 7^ = e%ik, and Gab = (—1, —1, • • • , -1). In this case, the most
general Dirac operator (see eqn (B.3))
becomes a hermitiari operator for the Euclidean inner product
(10.6)
When this operator acts on the Dirac field itself which carries no Minkowski
indices, the generator of general coordinate transformation U® is set to 0.
We define a complete orthoiiormal set
to define the path integral measure, and expand the field variables as
If one defines the path integral measure by using weight 1/2 variables such as
•tp — (gY^tp-, we have
and the path integral is given by
If one considers a localized chiral transformation in this path integral
one obtains the following identity
The first equality in this identity expresses the fact that the path integral itself is
independent of the naming of integration variables, and the transition from the
CHIRAL U(l) GRAVITATIONAL ANOMALIES 225
first line to the last line expresses the result of expanding the action in powers of
the infinitesimal parameter a(x) combined with the evaluation of the Jacobian.
We also defined
When one uses the notation J(a) = cxp[-2i J dZnx v/ga(x)^45(x)], the chiral
identity is written as
We thus need to evaluate A&(x). We obtain the master formula for the quan-
tum anomaly by using the gauge invariant mode cut-off with the simple regulator
function
just as the case in flat space-time. The trace here runs over the spinor and
Yang -Mills indices. We next notice the following formula55
For an arbitrary spinor ipn (x), we have
Here -ft""lMJ, stands for the Riemann-Christoffel curvature tensor. By using the
symmetry properties of the curvature tensor, one can further rewrite
BB
7" commutes with Dp. if one uses the metric condition (B.17).
The master formula (10.15) is thus written as
where d(x,x') is a generalization of the 6 function to a curved space-time and it

is given by the representation
The quantity a(x, x') is called a geodesic biscalar, and it is a generalization of

(1/2)(x — x')2 to a curved space-time.
We perform the calculation of the Jacobian around a point x$ where the
space-time becomes flat. The coordinate conditions are specified by (see Ap-
pendix B)
We then set
and treat y]Jl as coordinate variables. The actual calculation is performed for an
infinitesimal r/ M and y^ is set y^ —>• 0 after the calculation. If one defines
the master formula becomes
a,nd the desired result is obtained by setting j/M to 0 after the calculation. We here
replaced 5(x, x') in the neighborhood of XQ by the ^-function in a flat space-time,
which is justified in the present calculation of the chiral anomaly. It is possible
CHIRAL U(l) GRAVITATIONAL ANOMALIES 227
to perform the calculation (10.24) directly, but to derive the result for arbitrary
space-time dimensions in a simple way we set
and take the large M limit simultaneously as setting y11 to 0. When one expands
various variables in powers in y^/M one obtains
where to simplify the calculations we used the fact that the relations such as
9a9ii.v(%o) = 0 and A™n(x0) = 0 hold in the present choice of coordinates and
that the final result depends on quantities anti-symmetric with respect to the
indices fj, and v (the symmetric components are eliminated by using the gauge
freedom). We thus have
If one rescales fc^ —> Mk^ in the master formula (10.24), the chiral anomaly
is written as
where we used the fact that only the terms with n-th powers in aab /M2 survive at
the end since only those terms which are larger than or equal to (l/M) 2n with at
least In 7-matrices survive at the end. In the derivation of this formula, we elim-
inated various terms at intermediate stages, for example, the terms (1/4M2)J?
and (IjM)A^(x^ + y»/M')Ta in eqn (10.27) were omitted in eqn (10.28). These
terms do not carry a sufficient number of 7-matrices relative to the powers in
1/M. It is confirmed that these eliminated terms do not modify the final result
of the chiral anomaly to be given later.
When one defines
in the above final formula for the anomaly (10.28), RVfl/M'2 can be treated
as quantities commuting with each other; the commutator of two Rvff/M'2's
increases the powers in 1/M but it does riot increase the number of 7 matrices.
Similarly, RV^/M2 and F/M2 are treated as quantities that commute with each
other. Consequently, the master formula is also written as
We perform an explicit evaluation of this formula by using a pa,th integral tech-

nique of first quantization in the next subsection.
10.2 Evaluation by a quantum mechanical path integral

We formally define a Hairiiltonian
with y^ as a variable conjugate to p^, and we define the time evolution operator
as in Chapter 2. After a Legendre transformation, we define a path integral of
the first quantization for a unit time interval
and the normalization constant Ar is an undetermined constant at this moment

and it is fixed at the end of the calculation. By recalling that we can treat
Rnti(xo)/M2 as a commuting constant we perform the path integral over the
momentum variables
We thus confirm that only the anti-symmetric components of Rv(1 contribute.

By taking into account the boundary condition for
we expand £ M (T) by a complete orthonormal set
EVALUATION BY A QUANTUM MECHANICAL PATH INTEGRAL 229
The path integral measure is then expressed in terms of the expansion coef-
ficients
where we defined
If one performs the path integral over 6^ and then a%, one obtains
where we used the matrix notation
and the determinant in eqn (10.37) is defined with respect to these indices. We
also used the orthogonal property56
and an infinite product representation of sinhx. 57 The normalization constant

N is fixed by considering the special case J2,,,, = 0
and we obtain the final result
56
Em=i A(n,m)A(l,m) = 4 £)^_ 0 / 0 drcfrr' sin 2mrr cos m?rT sin 2!7rr'cos mirr' = Sn.i
where we extend the sum over odd integers m to the sum over all the integers m by using
JQ dr sin 2ri7TT cos 2m7rr = 0.
B7
siiih x =
10.3 Cherri character and Dirac genus

By using the result (10.41) of the previous section, the chiral U(l) anomaly is
written in a general coordinate system as
although we performed the actual calculation in a specific choice of the coor-

dinates (10.21). The trace in front of the entire expression stands for the trace
over the Dirac matrices and Yang-Mills indices, and the trace in the exponent
is taken over the Minkowski indices of R,j,a.
In this formula, the term with exp(F/M2) is called the Chern character and
the determinant with R is called the Dirac genus. The expression (10.42) when
combined with the classical solutions of the Dirac equation gives rise to
where n±, respectively, stand for the number of normalizable solutions with
the vanishing eigenvalue flipfl(x) — 0 and jzn+i'-Pn = ±¥>n- We here used
f d2nx ^fg(p\(x)^n+\(pi(x) = 0 for A; / 0. This relation (10.43) gives a gen-
eralization of the Atiyah Singer index theorem.
If one recalls the following definitions in the formula (10.42)
and if one expands the formula (10.42) in powers of aab/M2 by using
the term with the rt-th power in aab /M2 combined with tr 72,^1 gives the final
explicit result of the chiral anomaly in 2n-dimcnsional space-time.
From the expansion of sinhx/x one recogni/es that the chiral U(l) anomaly
with the gravitational field appears only in d = 4fc (k = 1, 2, • • • ) dimensional
space-time. In contrast the chiral anomaly with the gauge field F appears in
arbitrary even-dimensional space-time. For example, in four-dimensional space-
time, the contribution of the gravitational field to the identity (10.14) in QED-
type theory is written as
In a general higher even-dimensional space-time the cross-terms of F and R.

which are called the mixed anomaly, appear.
ANOMALY IN GENERAL COORDINATE TRANSFORMATIONS 231
In mathematical literature the notation of differential forms is often used. To

convert our formula to the notation of differential forms, one first replaces
in the formula (10.42). and then the chiral anomaly is written as
where the remaining trace is over the Yang-Mills indices. When one extracts
2n-forms from this formula by noting
the result expressed in terms of differential forms is obtained in arbitrary 2n-

dimcnsioiial space-time. The extraction of the 2ri-forms replaces the trace with
10.4 Anomaly in general coordinate transformations

We have explained in Chapter 6 that the gauge symmetry itself can contain
an anomaly in chiral gauge theory which breaks parity. It was pointed out by
Alvarez-Gaume arid Witten that Einstein's general coordinate transformation
itself can contain an anomaly in chiral theory, which plays a basic role in the
quantum theory of superstring theory. In general, the self-dual tensor fields, for
example, can contain gravitational anomalies besides the chiral fermion, but we
discuss only the simplest chiral fermion in this book.
We consider a general 2n-dimensional Euclidean space-time and quantize
only the Dirac field by treating the gravitational and gauge fields as classical
background fields. The action for chiral theory is given by
The Einstein equation for the quantized Dirac field is written as (see Ap-
pendix B)
where (T^v(x)} stands for the energy-momentum tensor generated by the quan-
tized Dirac field. This (Tllv(x)} with a correct normalization is given by
We now analyze the symmetry properties of the quantized (TnV(x)}. Since the
action S = f d4x ^/g.C is invariant under the local Lorentz and general coordinate
transformations (see Appendix B), we have
and thus we obtain the identity
where the first equality expresses the fact that the path integral itself is inde-
pendent of the naming of integration variables, and the transition from the first
expression to the last one depends on the evaluation of the Jacobian and the
invariance of the action. When one writes an infinitesimal transformation of the
vicrbein as eJA = eA + JeA and retains the lowest-order term in <5e A . one obtains
the identitv
It is important to recognize that the variables such as ip instead of ip are kept

fixed when one takes the functional derivative with respect to the vierbein.
If one uses the variation corresponding to the local Lorentz transformation
in the above identity, we obtain
By noting the anti-symmetry ujml = -wj m of the local Lorentz transformation

arid the general structure .of the Jacobian
ANOMALY IN GENERAL COORDINATE TRANSFORMATIONS 233
the identity is rewritten as
If the Jacobian (i.e., the quantum anomaly) for the local Lorentz transformation
does not vanish, the symmetric property of the left-hand side of the Einstein
equation (10.51) and the anti-symmetric components of the energy-momentum
tensor generated by the chiral Dirac field on the right-hand side are not compat-
ible with each other.
An infinitesimal form of general coordinate transformations is given by
but to deal with the energy-momentum tensor, which is not symmetric in general,
in a natural manner we consider a more general coordinate transformation
where Dv is the covariant derivative and thus satisfies the metric condition
D,,e% = 0. See Appendix B. Namely, we consider a combination of the general
coordinate transformation and the local Loreritz transformation which contains
the spin connection as a parameter
For this variation the above identity (10.55) is written as
If one assumes a general form of the Jacobian
one concludes
since ^(x) is arbitrary. If one combines this identity with the fact that the
anti-symmetric parts of Tvp,(x) satisfy the identity for the Lorentz transforma-
tion (10.58), the symmetric components T,swi(x) satisfy
which is called the WT identity for general coordinate transformations. If the

anomaly on the right-hand side of eqn (10.65) is not 0. the Einstein equation
becomes inconsistent for the quantized chiral Dirac field since the left-hand side
of the Einstein equation (10.51) satisfies the (Bianchi) identity
10.5 General properties of gravitational anomalies

In the calculational scheme of a covariant form of anomaly in Chapter 6, we used
the complete orthonormal system defined by f)\f>L and ^>i^L to expand the
path integral variables ip arid ip, where
An equivalent but more direct way of expanding the variables is to start with
and to note the properties
We then expand
In this case, the action in the path integral is formally diagonalizcd
where we ignored the phase factor which appears as the Jacobian for the change
of variables from ip and il> to their expansion coefficients. This phase factor,
which does not appear in vector-like theory, is important in the present chiral
theory when one analyzes the consistent form of anomalies. Since we evaluate
the covariant form of anomalies, we can ignore this phase factor in the present
analysis.
The Jacobian for the local Lorentz transformation
is given by
GENERAL PROPERTIES OF GRAVITATIONAL ANOMALIES 235
where we sum over all the eigenvalues (negative as well as positive and 0 eigen-
values) in the last expression of cqn (10.72).
Similarly, the Jacobian for the general coordinate transformation (by denning
is given by
where we sum over all the eigenvalues in the last expression. An important prop-
erty of this last expression is that a spurious gravitational anomaly which does
not contain 72n+i is automatically eliminated as a result of the use of integration
variables •</> = (g)l/4ip with the weight factor (fl) 1//4 . (See Appendix B.)
In this way, the quantum anomalies for the local Lorentz and general coor-
dinate transformations are respectively given by
where we choose the simple regulator function e x. These two anomalies are
related to each other in the following manner (by omitting to write the explicit
regulator e~x^/M and by noting
This relation shows that the anomaly for general coordinate transformations for
the Dirac field is automatically obtained once one can evaluate the local Lorentz
anomaly.
It is known that the anomaly for local Lorentz transformations exists only
in d = (4tk + 2)-dimcnsional space-time. An intuitive way to understand this
fact is to decompose the d = (4k + 2)-dimensional space-time as a product of
d = 2-dimensional and d = 4/fc-dimensional space-times
In this case, if one considers Ai2 in eqn (10.75) and notices
the local Lorentz anomaly A12 is decomposed into a product of the two-dimen-
sional Weyl anomaly (generated by 1) and the 2(n — l)-dimcnsional gravitational
chiral U(l) anomaly. As was explained in Section 7.6, the Weyl anomaly appears
in two-dimensional space-time and the gravitational U(l) anomaly exists only in
d = 4&-dimensional space-time as was explained in Section 10.3. Consequently,
one can at least intuitively understand that the local Lorentz anomaly exists
only in d = (4k + 2)-dimensional space-time.
10,6 Explicit examples of gravitational anomalies

We first consider a chiral U(l) gauge theory in two-dimensional Euclidean curved
space-time, though the gravity in this space-time is trivial. The local Lorentz
anomaly is given by
EXPLICIT EXAMPLES OF GRAVITATIONAL ANOMALIES 237
and this expression agrees with (—z/4)e m ™ times the Weyl anomaly in two-
dimensional space-time discussed in Chapter 7. By using the result of the Weyl
anomaly in Section 7.6 (the gauge field does not contribute to the Weyl anomaly
in two dimensions) and the relation between the Lorentz anomaly and general
coordinate anomalies (10.76), we obtain
where we used
In the space-time of general dimensions, one can use for example the heat
kernel method to obtain the explicit results of gravitational anomalies by a gen-
eralization of eqn (10.19) but the actual calculations are very involved. As for the
anomaly of general coordinate transformations, a calculational scheme for general
space-time dimensions is known. This method is based on the formula (10.74) in
the preceding section which is written as
and the following rules based on this expression: The Dirac matrix 72^+1 m this
formula is denned in d = 2n = (4.k + 2)-dimensional space-time, but one may
formally regard it as 72n+3 multiplied by <r"m
in d = 2n + 2 = (4fc + 4)-dimensional space-time which contains two more

dimensions. One may then regard ^ as the spin connection associated with
these extra two dimensions by formally assigning the two indices £^n. We then
use the gravitational chiral U(l) anomaly (10.48) written in differential forms in
(4A; + 4)-dimensional space-time, which was derived in Section 3 of this chapter,
In this formula we make the following replacement of

or. in terms of the curvature 2-form
and we retain only the terms linear in £^ and multiply the. result by 2-iri. In
this way one finally obtains the gravitational anomaly Â^x) in cqn (10.81).
From this rule, one can understand that the anomaly for the general coordinate
transformations appear only in d = (4k + 2)-dimensional space-time. (This rule
also shows that the mixed anomaly of gravitational and gauge fields appears in
more general 2n-dimeiisional space-time.)
An intuitive understanding of the rule (10.84) is obtained if one recalls the
evaluation of the gravitational chiral U(l) anomaly in eqns (10.19) and (10.30).
If one indicates the two extra dimensions by a tilde, one can write the covariant
derivative as
where we assumed that there is no spin connection with mixed indices such as
Apmn- We also assume that the spinor field and the spin connection depend
only on the first 2n — (4fc + 2)-dimensional coordinates. If one considers that
Aiann(x) = £/j,mn(%) as an infinitesimal quantity, and expands the formula (10.19)
in powers of Al2.7jlfl(x) and retains the term linear in AllT-nn(x), one obtains the
formula (10.81) if one takes into account the numerical constant factor properly.
This gives an intuitive understanding of the above rule (10.84).
The rule (10.84) in fact reproduces the gravitational anomaly in d=4k+2=
2 dimensions, which was obtained in eqn (10.80) by using a different method.
For this case, one starts with the gravitational chiral U(l) anomaly in d = 4,
and according to eqns (10.83) and (10.45). when
Under the replacement (10.85), the term linear in £ is given by
In two dimensions, the Rierriann-Christoffcl curvature has only one independent

component which is given by the scalar curvature
{Jua-gvp)- By using this fact, we have
and consequently the replacement rule gives

EXPLICIT EXAMPLES OF GRAVITATIONAL ANOMALIES 239
This is identical to f <Px ^fq^'A^x) where the gravitational anomaly A,L(x) is

given by eqn (10.80).
We need to cancel gravitational anomalies arising from various fields in a
given theory since the basic principles of .quantum theory and general relativ-
ity become incompatible with each other if the gravitational anomalies appear.
In this analysis of anomaly cancellation, one generally deals with the cancella-
tion of general coordinate anomalies, and if the general coordinate anomalies
arc cancelled the residual local Lorentz anomalies are superficial and they can
be eliminated by a suitable counter-term. One way to understand this fact may
be to formulate the gravitational theory for the Dirac fcrmion by eliminating
the local Lorentz freedom by gauge fixing and realize the general coordinate
transformations in a non-linear manner, which is known to be possible in pertur-
bation theory. In this formulation, one deals with only general coordinate trans-
formations and thus the above rule is obtained. (The general relation (10.76)
shows that a reverse reasoning is trivially valid, namely; the cancellation of the
Lorentz anomalies automatically ensures the cancellation of the general coordi-
nate anomalies.)
As for practical implications of gravitational anomalies, in addition to the
fact that even the Einstein general coordinate transformations could be broken
by quantum effects, one obtains an important consistency condition of quan-
tized superstring theory defined in d = 10 (= 4 x 2 + 2) dimensions. In fact,
the gravitational anomalies played an important role in showing the existence of
anomaly-free (consequently quantum mechanically consistent) superstring the-
ory and also in counting all the possible consistent snperstring theories.
11
CONCLUDING REMARKS
In this book we have discussed the basic ideas of quantum anomalies and their
applications in the path integral formulation of quantum field theory. The first
quantum anomaly was discovered in the detailed analysis of triangle Feynman
diagrams. In this sense, the triangle diagram has a special meaning in four-
dimensional space-time (in d = 2n dimensional space-time, the (n + l)-gon di-
agrams have a special meaning). We have explained that all the known (local)
quantum anomalies are formulated as the non-trivial Jacobians associated with
the symmetry transformations of integration variables in the path integral for-
mulation of field theory. Namely, the path integral measure (or the quantization
procedure itself) breaks some classical symmetries. In this sense we have given
an answer to the question of what is the quantum anomaly. In particular, the re-
cent demonstration that the chiral anomaly in lattice gauge theory is formulated
as the Jacobian for chiral transformations has given additional support for this
view of quantum anomalies. At the same time we have also explained that the
quantum anomalies are characterized by anomalous commutation relations (or
central extensions in commutation relations) in the operator formalism. In either
case,. one has to go beyond the naive canonical formulation (or its straightfor-
ward path integral transcription) to accommodate quantum anomalies in local
field theory.
We have thus learned that various characterizations of quantum anomalies arc
possible. One might still ask if a more intuitive and elementary characterization
of quantum anomalies is possible. In the following, we briefly describe such an
elementary characterization.
We first recall the fact that quantum anomalies are closely related to diver-
gences but a careful analysis shows that they arc perfectly finite and independent
of divergences. In this sense the presence of an infinite number of degrees of free-
dom is essential for the anomaly.
For example, in the representation which diagonalizes 75 we have
and the chiral anomaly in a naive interaction picture corresponds to the evalua-
tion of oscillating infinite series
240
CONCLUDING REMARKS 241
The value of this infinite series generally depends on how to sum the series. In
the evaluation of chiral anomaly in the present book we evaluated this series as
by using the complete basis set {ipn(x)} for a Dirac operator lf>. The basis set
plays a fundamental role to give a definite value to the oscillating scries. We thus
understand that the quantum anomaly is located in the border of divergence
and convergence. In the case of a finite or absolutely convergent series the sum
is uniquely given, and thus there is no freedom to control the sum of the series
by a choice of the basis set. This shows that the quantum anomaly is closely
related to the presence of infinite degrees of freedom.58 In connection with the
phase operator of the photon discussed in Chapter 3, we also explained that the
infinite-dimensional space spanned by the creation operator of a bosonic particle
gives rise to the notion of index and provides important information.
As another characteristic feature of quantum anomalies, we observe the com-
petition of gauge symmetry and chiral symmetry in the chiral anomaly, for ex-
ample, and we observe the competition of gauge symmetry (including general
coordinate transformations) and Weyl symmetry in the Weyl anomaly. This is
reminiscent of a certain kind of "uncertainty relation." For example, the fact
that we cannot diagonalize 75 and Tf> simultaneously is closely related to the ap-
pearance of the chiral anomaly. To diagonalize lf> implies that we impose gauge
invariance, and in this case the relation
implies that 75 cannot be simultaneously diagonalizcd. In the sector of the 0

eigenvalue of $>, 75 is simultaneously diagonalized and the notion of the index
arises from this sector. This property at first sight might imply that the ap-
pearance of the index is independent of the property that 75 and T/) are not
diagonalized simultaneously. But as we explained above the infinite degrees of
freedom is fundamental to render the notion of index to a square matrix, and
the non-commuting property of two operators plays a basic role in the analysis
of quantum anomalies.
In fact, when one evaluates the average of the right-hand side of the above
commutation relation (11.4) by using the single-particle states in Hilbert space
(see eqn (4.54)), one obtains
°8The presence of tke-notion of index on-the finite lattice does not contradict this view. We
have Tr75 = N+ + n+ — N— — n_ = 0 on the finite lattice. We find a non-vanishing index
TrFo = n+ — n _ only when we use a modified operator FS which projects out the freedom N±.
See Chapter 9.
242 CONCLUDING REMARKS
which gives both of the explicit breaking of chiral symmetry by the mass term and
the quantum breaking of chiral symmetry by the anomaly. In the last step of the
calculation, however, we need to specify how to sum the series in increasing order
of the magnitudes of eigenvalues Xn (as a rather weak form of regularization).
Similarly, when one defines the generator of the Weyl transformation formally
by w, one has
and thus for the Dirac operator $> = 7ae^(a;)[3^ - (i / 2) Af aat> - igA^} defined
in a slightly curved space-time, one has
by noting [w,A^] = 0. namely, the gauge fields are Weyl scalars. The average of
the right-hand side of this commutation relation in the flat space-time limit as
in eqii (11-5) gives
where the last result is obtained as in eqn (7.48) by applying a suitable cut-off
in terms of the eigenvalues of ]/) and retaining only the connected components as
the operator. We thus correctly evaluate both of the explicit breaking of Weyl
symmetry by the mass term arid the quantum breaking of Weyl symmetry by
the anomaly.
These examples illustrate that we can understand (certain aspects of) the
quantum anomalies as a kind of uncertainty relation originating from the fact
that two basic operators are not simultaneously diagonalized.
We now briefly comment on several other topics related to quantum anoma-
lies which we did not discuss in the body of the book. We have so far concentrated
on the explicit evaluation of Jacobians in the path integral. There are elegant
though somewhat abstract constructions of the anomalies related to gauge trans-

formations, which are sometimes called the descent formula. In this construction
of the integrable form of gauge anomalies in d = In dimensions, one starts with
the chiral U(l)-type anomalies in d = In + 2 dimensions. The differential form
is convenient for this construction and we define
We start with the, relation (see also the Chern character in Chapter 10)
where W2n+i (A, F] is a In + 1 form. This relation is derived by a generalization

of the derivation of the Chern-Simoiis form in Chapter 5. We next define
where Au stands for the gauge transformation of A
We note that
by recalling the definition of ^/(U, A, F) and the gauge invariance of tr Fn+l. For
an infinitesimal a (a;), we have
where Aa(A. F) gives the gauge anomaly in eqn (6.140) (or eqn (6.136) in the
case d = 4) written in form notation. The Wess-Zumino term is then written as
where the integration domain D (= disc) is a generalization of the domain in four-

dimensioiial theory in cqn (6.160) (or eqn (6.148) with R/j, = 0) for the Euclidean
metric. As for a justification of these constructions, readers are referred to the
references at the end of this book.
We next comment on the non-perturbative (global) SU(2) anomaly which
appears in the Euclidean path integral
where L^ = La^(x)Ta stands for the gauge field belonging to SU(2). We impose
the boundary condition on the allowed set of gauge transformations U(\x =
244 CONCLUDING REMARKS
oo) = 1. The Euclidean space-time is then regarded as the compact S4. The
chiral SU(2) gauge theory is anomaly-free for a gauge transformation which
is smoothly connected to the identity, as we already explained. However, it is
known that the homotopy from the space-time S4 to the group manifold of SU(2)
is given by
and thus there are two kinds of "large" gauge transformations: one is smoothly
connected to the identity and the other is a non-trivial one which is not smoothly
connected to the identity. The non-trivial one applied twice is reduced to the
trivial one which is smoothly connected to the identity. Because of 7r/i(SU(3)) = 0,
the non-trivial gauge transformation in SU(2) when embedded in SU(3) is now
smoothty connected to the identity.
We thus formally extend the SU(2) gauge theory to an SU(3) theory by
enlarging the SU(2) doublet ib to the SU(3) triplet ip, but the gauge field L0^ still
couples to only the generators of SU(2). The non-trivial gauge transformation
in SU(2), which we denote by g, is embedded in SU(3) in the form
where g is a 2 x 2 matrix. When one varies the gauge transformation from the
identity to g, the gauge transformation assumes values in SU(3) in the intermedi-
ate stages. In the definition of the Wess Zumino term, the gauge transformation
assumes the specific value g at the boundary of the disc D. We then consider
the gauge transformation
and obtain the identity
where we extracted the Wcss-Zumino term as an integrated Jacobian for the

gauge transformation inside SU(3), where one can smoothly connect the identity
to g. See eqn (6.141). A detailed analysis shows that,
The path integral over all the possible gauge field configurations thus gives
rise to the result
where we adopt the path integral prescription to take a sum over the (large)
gauge freedom associated with Zi. Note that the path integral is invariant un-
der an ordinary SU(2) gauge transformation which is smoothly connected to the
identity, and thus the ordinary gauge transformation is gauge fixed in a conven-
tional manner when one performs the path integral over the gauge field with
the Yang-Mills action added. The path integral of a single chiral SU(2) dou-
blet thus has no physical meaning, which was first pointed out by Witten. In the
Wcinberg-Salam theory we have an even number (= 4) of SU(2) doublets in each
generation by counting lepton and quark multiplets, and thus the above phase
factor (11.21) becomes unity and the difficulty associated with the global SU(2)
anomaly does not appear. A detailed analysis of the global SU(2) anomaly re-
quires some mathematical background, and we give references and a monograph
on this subject at the end of the present book.
The anomaly cancellation in superstriiig theory is another important subject
which we did not discuss in this book. This issue was analyzed by Green and
Schwarz to show the existence of quantum mechanically consistent superstriiig
theory. The basic idea and the machinery to analyze this problem are explained
in the present book, but a concrete analysis of this problem requires further
detailed knowledge of gravitational anomalies. It is also interesting to evaluate
the quantum anomalies directly in superstring theory, and interested readers are
referred to references given at the end of the present book.
We did not discuss supersymmetry and associated quantum anomalies, which
are subjects of great interest among particle physicists recently. When one con-
siders supersymmetric theory in terms of its component fields, all the evaluations
of quantum anomalies are performed by the methods given in the present book.
However, some of quantum anomalies which appear to be independent of each
other are often related to each other in the framework of supersymmetric the-
ory. To understand such a relation it is necessary to generalize the calculational
schemes in this book to evaluate the quantum anomalies in the superfield for-
mulation. It is in fact known that the anomalies can be evaluated as non-trivial
Jacobians in the path integral formulation in superspacc. Interested readers are
referred to the references at the end of the present book.
APPENDIX A
BASICS OF QUANTUM ELECTRODYNAMICS
In this appendix we summarize the basic aspects of quantum electrodynamics.

Though the knowledge summarized here is not essential to understand the main
part of the present book, it helps to understand some of the analyses better.
A.I Quantum electrodynamics

The basic action of quantum electrodynamics is defined as a combination of the
Maxwell action and the Dirac action
An important characteristic of quantum electrodynamics is that the electromag-

netic field An. couples to the Dirac field -0 only through the covariant derivative
As a result, the action is invariant under the gauge transformation, which is a

generalization of the gauge transformation in Maxwell theory,
where u>(x) is an arbitrary function. The co-variant derivative is transformed

under this gauge transformation as
The basic principle of gauge theory is formulated, following the reversed

way of reasoning, as a requirement that the action is invariant under the phase
transformation i}>'(x) — U(x)ip(x) of the matter field tp. Both the gauge field
and covariant derivative are then introduced, and the gauge transformation rule
of the gauge field (the photon in the present case) is determined. Consequently,
the action of the gauge field itself is fixed so that it is invariant under this gauge
transformation. Moreover, it is known that the action is uniquely determined if
one requires the renormalizability (i.e., the consistent treatment of higher-order
246
QUANTUM ELECTRODYNAMICS 247
quantum effects) of the theory thus constructed. This is the reason why the gauge
principle (and gauge theory) is so powerful and useful.
In the present example the gauge transformation is composed as
namely, following the rules of addition. The gauge group thus forms an Abelian
U(l) group, and for this reason the electromagnetic field A^ is called an Abelian
gauge field.
In general, one can introduce the gauge field consistently for an arbitrary
compact group and the gauge field associated with a non-Abelian group is called
the Yang-Mills field.
The name of the covariant derivative comes from the fact that D^(x) is
transformed
in the same manner as the original field ijj(x) under the gauge transformation.
The electromagnetic field strength tensor is generally defined as the "curvature"
by using the covariant derivative. From this definition one can understand that
the electromagnetic field strength is gauge invariant
From the Jacob! identity
one can derive two of the four Maxwell's equations
Another characteristic feature of the action of quantum electrodynamics is

that the interaction term which is gauge invariant by itself (the Pauli term)
gFnVtf}[jlt,jl']^j is not included. This is because the action with the Pauli term
spoils the renormalizability of the theory. The action such as that of quan-
tum electrodynamics where the matter fields interact with the gauge field only
through the covariant derivative is called a theory with minimal coupling.
The path integral quantization of quantum electrodynamics is given by the
analyses in Chapters 2 and 3 by
248 BASICS OF QUANTUM ELECTRODYNAMICS
We employ the £ gauge condition (3.52) and also write the Faddeev Popov fields
explicitly, though they are not essential in the present Abelian theory. The inte-
gral over the Dirac field is defined by the left derivative in terms of Grassmaim
numbers. Following Schwinger we introduce the ordinary c-number source Jli(x)
for the gauge field A^ and the Grassmann number sources fj(x) and 77(0;) for
Dirac fields. When one writes the action appearing in the exponential of the
path integral as Sj, the equations of motion in quantized theory arc given by
considering the variational derivative of 5j with respect to, for example, A,L and
ip as
The matrix elements of those quantum operator equations should vanish.

To be precise, if one replaces the operator field variables in the matrix element
by derivatives with respect to source functions by using the action principle of
Schwinger and if one uses the above path integral representation, the equations
of motion written in terms of c-number fields appear in the integrand of the path
integral as
and
where we used the "translational invariarice" in functional space of the path

integral measure
both for bosons and fermions, namely, the fact that the definite integrals of
derivatives vanish. The quantum operator equations are thus ensured in the path
integral. The path integral formula thus defines the correct quantum theory for
interacting gauge and Dirac fields.
INTERACTION REPRESENTATION AND PERTURBATION FORMULAS 249
A.2 Interaction representation and perturbation formulas

We now separate the (effective) Lagrangian density in the above path integral
formula (A. 11) as
where £j stands for the free part consisting of only the terms quadratic in field
variables and
stands for the interaction part. The perturbative calculations are then reduced
to the evaluation of
In the last two expressions the interaction is specified by the derivatives with
respect to source terms, but when one brings the exponential terms with deriva-
tives inside the path integral the functional derivatives act on the source func-
tions contained in Cj' and produces the corresponding field variables. We thus
reproduce the starting formula. The important property of the last formula in
eqn (A.18) is that the path integral is performed only for the free fields. This
corresponds to the interaction representation in the operator formalism. In the
interaction representation all the field variables satisfy the equations of motion
of free fields. In fact, if one applies the Schwinger action principle to the path
integral denned in terms of Cj the Dirac field ^>, for example, satisfies the free
field equation
The above perturbation formula (A.18) thus corresponds to the Dyson formula
for the S-matrix. It is interesting that the operation related to the Wick theorem
in operator formalism is realized by the functional derivative with respect to

source functions.
We can perform the path integral explicitly for free fields. To show this we
define the following Feynman propagators
where the derivative <9A acts on the coordinate x. By using those functions we
make the following change of variables in the path integral
If one uses the fact that the path integral measure is invariant under the above
change of variables, for example, T>ip = 'Dt/j', the path integral for free fields is
given by
When one uses the normalization of the path integral in the interaction picture
(0. +oo|0, — oo}j_i0 = 1, the path integral for the free fields is given by
When one applies the Schwinger action principle to the above path integral
of free fields (A.23) one obtains
INTERACTION REPRESENTATION AND PERTURBATION FORMULAS 251
where we used the representation 6^\x —2'y)?4=exp[—

/ , ip(x — ?y)] of the
delta function. The last formula is obtained by performing the integral over po by
deforming the contour to the direction where the exponential function decreases
and the result is known to agree with the result in the operator formalism. The
Feynman ie prescription dictates that the negative-energy solutions of the Dirac
equation propagate only in the negative time direction and, as a result, they
represent the anti-particle (the positron in the present context) with positive-
energy propagating in the positive time direction. (The Feynman ic prescription
is also understood as imposing the positive energy condition in the path integral
formulation, as was mentioned in Chapter 2 in the present book.) Similarly, we
obtain
where we give the result of the simplest Feynman gauge with £ = 1 in the last
expression. This result is also known to agree with the operator formulation
for the Feynman gauge £ = 1, though we do not demonstrate it here. The ie
prescription specifies that the negative-energy solutions of the Maxwell equation
propagate only in the negative time direction and, as a result, they describe
the anti-particle (in the present case of a real field A^, the anti-particle and
particle are not distinguished) with positive-energy propagating in the positive
time direction.
The h'nal formula for perturbation theory in quantum electrodynamics is thus

given by
If one expands this formula in powers of the coupling constant e up to a suit-

able order and then performs the functional derivative of (0, +oo|0, -oo)j with
respect to source functions by suitable times (and setting the remaining source
functions to zero), the formula for the Green's function which is expressed as
the vertex functions expressed by 67^ connected by the Feynman propagators
Dp and SF is obtained. The Feyrirnan diagrams give an intuitive picture of
this representation of the Green's function.
APPENDIX B
FIELD THEORY IN CURVED SPACE-TIME
In this appendix we briefly summarize the basic properties of field theory in

curved space-time.
B.I Coordinate transformation and energy-momentum tensor

The Lagrangian of Einstein's general theory of relativity in the presence of QCD-
type gauge theory is defined by
where K = 8?rG with the Newton constant G. The Greek indices which describe
the Minkowski coordinates of the curved space and the Roman indices which
describe the Lorentz coordinates of the flat space attached to each point of the
curved space are related to each other by the vierbein e^(x). The metric of the
flat Lorentz frame is denoted as Gmn. The vierbein is connected to the metric
by
The raising and lowering of the Minkowski indices p is realized by the metric
9tj,v(x) and the raising and lowering of the flat Lorentz indices m is realized
by Gmn. The basic symmetry of Einstein's theory, namely, the general coordi-
nate transformation, does not admit a double-valued representation such as the
rotation group for spin 1/2 in flat space, and thus we represent the Dirac field
as a double-valued representation of the Lorentz group of the flat space attached
to each point of the curved space-time.
The covariant derivative in the curved space is generally given by
and contains the generators Smn of the Lorentz group and the generators U& of
the general coordinate transformation GL(4, R) in addition to the generators Ta
of the ordinary gauge group. These generators are defined by
253
254 FIELD THEORY IN CURVED SPACE-TIME
To be more explicit, the generators for spin 1/2 and 1 are, respectively, given by
On the other hand, the generators U® of the general coordinate transformation

are given for the covariant vector Av and the contravariant vector A" respectively
by
Consequently, the covariant derivatives are defined by
The gauge field A™n appearing in the covariarit derivative is called the spin
connection and the gauge field Fjg is called the affine connection, respectively,
and these are expressed in terms of e^ and g^ as
The fields A™n and Pg in the present formulation, in which the torsion free-
dom is ignored, arc also called respectively the Ricci rotation coefficient and the
Christoffel symbol.
By recalling the basic relation of Riemann which states that the geometrical
length is independent of the choice of coordinate systems
the general coordinate transformation laws of the quantities with Minkowski

indices are given by
For an infinitesimal transformation
the general coordinate transformation is expressed in terms of the generators as

COORDINATE TRANSFORMATION 255
The transformation law of the metric tensor is derived from the transformation
law of the vierbein as
and the transformation law of the affirie connection is derived from its definition
as
The infinitesimal local Lorente transformation is given in terms of the pa-

rameter ojmn(x) as
by using the generators of the Lorentz group, just as the ordinary gauge trans-
formations. The transformation law of the spin connection in the last expression
is obtained from the transformation of the covariant derivative
or from the definition of A™n in terms of e% and the transformation law of e% .

Einstein's theory contains two gauge fields F^ and A™n, but both of these
fields are expressed in terms of the vierbein e'k. As the gauge freedom, the vierbein
contains the localized Poincare transformations with 10 parameters, namely, four
localized translations £ M (x) and six localized Lorentz transformations umn(x).
In the present formulation, the condition called the metric condition
is automatically satisfied. This metric condition implies that the inner product
of two vectors parallel transported from a point x to the point x + dx is defined
by the metric g^, (x + dx) .
As an application of this metric condition, we can show that the vierbein e%
at an arbitrary point, for example, at the origin, is chosen to be
This choice of the vierbein is used in the body of the present book to evaluate
quantum anomalies. The proof of this choice of the vierbein proceeds as follows:
We first apply a coordinate transformation x'v- — a^x" to the vicrbein e'Jk by

choosing a suitable constant a^ near the origin such that
We next apply a specific infinitesimal coordinate transformation near the origin
such that F^(0) = 0 without changing e'£(Q)- Finally, we apply a local Lorentz
transformation parametrized by 'jjmn(x) - -A"m(Q)x^' so that A'™n(Q) = 0
without changing e^*(0). When one uses the two relations 1^(0) = 0 and
A'™n(Q) = 0 in the metric condition (B.I7), we obtain the desired result.
The metric condition shows that we can freely commute the covariant deriva-
tive with e% and g^v. For example, when one applies the commutator of covariarit
derivatives to an arbitrary covariant vector Ap = e™An, one obtains
This relation shows that the Riernann-Christoffel curvature tensor evaluated in

terms of the affinc connection Pg^ .
agrees with the curvature tensor Rmn^ evaluated in terms of the spin connec-
tion A™n.
As an important local symmetry of the matter and gauge fields in gauge
theory, we have the Weyl symmetry defined by
where the transformation law of the spin connection A™n is derived from the
transformation law of the vierbein e£, since A™n is expressed in terms of e£. The
Weyl transformation changes the length as
but the local angle is preserved, and for this reason it is also called a conformal
transformation. When one defines g = det.g/u,, it is confirmed that the matter
COORDINATE TRANSFORMATION 257
part of the action / d4x ^f^gC- in terms of the Lagrangian (B.I) is invariant un-
der the Weyl transformation if one sets the fermion mass m = 0. The action for
a massless scalar theory is also rendered invariant under the Weyl transforma-
tion if one chooses £ = ^g^d/^dup + \R<t>1. However, the action of Einstein's
gravitational theory itself is not Weyl invariant.
The scalar curvature R in the Einstein action, which is also called the Einstein-
Hilbert action, and also the Ricci tensor R^v are defined in terms of the curvature
tensor by
When one considers the variation with respect to the vierbein e%(x) in the
Einstein-Hilbert action (B.I), the Einstein equation
is obtained. The (naive) energy-momentum tensor T^v(x) generated by the mat-

ter fields in this equation is defined by°9
wtu,re we symmetrized the expression with respect to the replacement of ?/> and ijj.
We also normalized the completely anti-symmetric symbol e1230 = 1 and used
the property of the Dirac matrix {^k,Smn} = ekmnljij5.
To study the symmetry requirement on the energy-momentum tensor, we re-
call the basic property of the Einstein Hilbert action. We first note the following
relation
valid for an arbitrary variation 5gp'v(x) of the metric tensor. As a specific vari-
ation of the metric, we consider the variation induced by a coordinate transfor-
mation
S9
ln deriving this expression, it is convenient to note the relation SA™n —
(l/2)cmXen>>(8CXpl, - <5CpV - <5C, iAp ), where 5CXpl, -= e*(DpSek^ - Dp,5ekp). This relation
follows from the fact that the difference of two spin connections behaves as a tensor.
When one recalls that the Einstcin-Hilbert action is invariant under the coordi-
nate transformation and that the covariant derivative commutes with the metric,
the above relation (B.29) valid for an arbitrary variation implies
Since £"(#) is arbitrary we conclude
When one combines this relation with the Einstein equation (B.27). one con-
cludes the requirements on the classical energy-momentum tensor
B.2 Path integral measure in gravitational theory

It is necessary to define the path integral measure carefully when one analyzes
the quantum theory of gravity and the quantization of fields in curved space-
time in the path integral formulation. In particular, the treatment of general
coordinate transformations is subtle and a naive treatment could induce spurious
anomalies. Also the Weyl anomaly is uniquely specified only when one uses the
general coordinate invariant measure. In the following, we discuss the definition
of the path integral measure which does not induce spurious breaking of general
coordinate transformations.
It is convenient to use the BRST symmetry in this analysis, though the
BRST symmetry itself is somewhat technical. The basic idea of the BRST trans-
formation is based on writing the parameter f M (a;) of general coordinate trans-
formations in the form £^(x) = i\dl(x) where A is a constant real Grassmann
parameter and c>L (x) is the Grassmann variable called the Faddeev-Popov ghost
field. When one uses the BRST supcrfield notation, one writes the original field
variable together with its variation under the BRST transformation by using a
constant Grassmann parameter 0 as
whore we used the transformation law of tf>, 5tp(x) = ip'(x) — ib(x) = £.pdp'ip(x),
and wrote the anti-ghost field c-n(x) together with its variation, the auxiliary
field B^ (x).
PATH INTEGRAL MEASURE IN GRAVITATIONAL THEORY 259
The BRST transformations are denned as a translation in the Grassmann

parameter 9 ->• 9 + X. Consequently, the first component of the superfield, which
does not contain the parameter 8, varies by an amount proportional to the pa-
rameter A and the second component, which contains the parameter 0, remains
invariant. To be explicit, the transformation law 6\ of the ghost field cll(x) is
given by (noting that both CM(O;) and A are Grassmann numbers)
Namely, the, first component receives a transformation proportional to the second

component, and if one uses its result the second component becomes invariant.
The BRST transformation is thus consistently denned. The treatment of the anti-
ghost is exceptional, and its variation is defined by fi\c^(x) = \B(x). 5\B(x) = 0
and. as a result the measure X>cMX>£>M is BRST invariant. When one uses the
notation of the BRST superficld, the gauge condition d^g^ = 0. for example, is
defined by collecting the terms linear in 9 in the expression
which defines both the gauge fixing and compensating terms simultaneously.
In passing we note that the transformation law of the ghost (B.34) does not
appear as a transformation law of a contravariant vector quantity, but if one
considers the differential of the superfield itself one obtains
which corresponds to the replacement. £p (a;) —> i9cp (x) in the general coordi-
nate transformation of the contravariant vector dcIJ-(x). This property plays a
fundamental role when we discuss the BEST invariant path integral measure.
We now discuss the path integral measure which is invariant under the BEST
symmetry associated with general coordinate transformations. We start with the
simplest fields which do not carry the Minkowski indices such as the fermions
?/>(x) and $(x). We define the weight 1/2 fields60
and examine the path integral of the mass term of the fermion
In this relation, the action on the left-hand side is invariant under the general
coordinate (or associated BRST) transformation and the right-hand side is a
60
The weight 1/2 fields are defined as the field variables (without the Minkowski indices)
multiplied by the weight factor (—g) 1 / 4 . and these variables transform under general coordinate
transformations as in eqn (7.19).
constant independent of the metric variable. We thus conclude that the path
integral measure on the left-hand side is invariant under the general coordinate
(or BRST) transformation. One can in fact confirm that the Jacobian for the
BRST transformation becomes trivial, i.e., 1, independently of the choice of basis
sets, and thus free of general coordinate anomalies. See eqn (7.21).
We now generalize this analysis to the cases such as the vector field Aa(x)
and the second-rank tensor field Aap(x). We first convert these fields into fields
without the Minkowski indices Aa(x) = eÂa(x) and Aab(x) = e%e%Aa/3(x) by-
using the vierbein. We then consider the weight 1/2 field prescription as above
which define the general coordinate (or associated BRST) invariant path integral
measure. When we give two definitions of the measure in eqn (B.40), the latter
definition is obtained from the first one by first extracting the vierbein as the
Jacobian and then distributing the Jacobian to all the degrees of freedom equally.
This second definition may not be said to be precise, but we cannot convert
the metric g^v into a quantity without the Minkowski indices by multiplying
the vierbein. The second definition of the measure works for the case of the
metric tensor also. We also give the definitions for the vector and tensor fields
in arbitrary d = n dimensional space-time in eqn (B.40) for the convenience of
applications in the present book.
In this way the path integral measure invariant under the BRST symmetry
associated with the general coordinate transformation is given by
The part containing the metric is defined by d/j,(gal3) = T)[(-g}kgai3] as we

explained above. The measure for the ghost variable is determined by c^ =
(-gr) 1 / 4 e^c M by using the fact that the differential of the ghost variable is trans-
formed as a contravariant vector under the BRST transformation. In this path
integral measure the BRST invariancc of the part dfj^(ga^)Vc>J is not manifest,
but its invariance is confirmed by showing that the result after BRST transfor-
mation agrees with the result before the BRST transformation as
PATH INTEGRAL MEASURE IN GRAVITATIONAL THEORY 261
The above proof proceeds as follows: The measure for the metric is invariant
for a fixed <?'' arid thus one can set 9 = 0 in ga'B(x,0) in the third line. We
next fix the metric, and then the Jacobian, det{[—g(x,0)]'Lêl^(x,d)}~1, is the
one which renders a general (fcrmionic) corltravariant vector invariant and thus
when combined with T>c^(x, 0) we can set 0 = 0. It is important to keep in mind
that one needs to consider the metric and the ghost variables always together in
the analysis of the gravitational path integral measure.
APPENDIX C
REFERENCES WITH BRIEF COMMENTS
We first give some references which are directly related to the descriptions of the
present book with brief comments. We then present some references to further
advanced research in path integrals and quantum anomalies. We also briefly
comment on the subjects or applications of quantum anomalies which were not
covered in the present book.
C.I Genesis of quantum anomalies

The basic references to modern field theory (renormalization theory) are found
in:
1. J. Schwinger, Quantum, electrodynamics. Dover (1958).
The letter of Tomonaga to Oppenheimer is printed in:
2. S. Tornonaga, Phys. Rev. 74 (1948) 224.
The evaluation of the two-photon decay of the neutral TT meson which eventually
led to the discovery of quantum anomalies was first performed by:
3. H. Fukuda and Y. Miyamoto, Prog. Theor. Phys. 4 (1949) 49.
This calculation was refined by:
4. H. Fukuda, Y. Miyamoto, T. Miyajima, S. Tomonaga, Prog. Theor. Phys.
4 (1949) 385; H. Fukuda. Y. Miyamoto, T. Miyajima, S. Tomonaga, S.
Oncda and S. Sasaki, Prog. Theor. Phys. 4 (1949) 477.
5. J. Steinberger, Phys. Rev. 76 (1949) 1180.
6. J. Schwinger. Phys. Rev. 82 (1951) 664.
The Pauli-Villars rcgularization was introduced in:
7. W. Pauli and F. Villars. Rev. Mod. Phys. 21 (1949) 434.
8. S.N. Gupta, Proc. Phys. Soc. A 66 (1953) 129.
The detailed analyses of the decay of the soft TT meson and the triangle diagrams
in spinor electrodynamics led to the modern formulation of quantum anomalies:
9. J.S. Bell and R. Jackiw, Nuovo Cim. A 60 (1969) 47.
10. S.L. Adler, Phys. Rev. 177 (1969) 2426.
We here give some well-written readable reviews of quantum anomalies:
11. S.L. Adler, in: Lectures on elementary particles and quantum field theory.
S. Deser et al. (eds.), MIT Press (1970).
12. S.B. Treirnan, R. Jackiw, B. Zumino and E. Witten, Current algebra and
anomalies. World Scientific (1985).
262
PATH INTEGRAL AND ACTION PRINCIPLE 263
13. P. van Nieuwenhuizen, Anomalies in quantum field theory: cancellation of

anomalies in D = 10 super-gravity. Leuven University Press (1988).
14. R. Bertlemann, Anomalies in quantum field theory. Oxford University Press
(1996).
In these books and reviews together with a more recent lecture note by J. Zinn-
Justiii, Chiral anomalies and topology, hep-th/0201220, readers can learn various
viewpoints on quantum anomalies complementary to those in the present book.
C.2 The Feynman path integral and Schwinger's action principle

The classic paper of Dirac is found in:
15. P.A.M. Dirac, Phys. Zeit. Sow. 3 (1933) 64.
The Feynman path integral is given in, for example:
16. R.P. Feynman, Rev. Mod. Phys. 20 (1948) 367; Phys. Rev. 80 (1950) 440.
Other references are found in [1]. For Schwinger's action principle, see, for ex-
ample:
17. J. Schwingcr, Phys. Rev. 91 (1953) 713.
Further references are found in [1]. A readable account of the action principle is
given in:
18. C.S. Lam. Nuovo Cim. 38 (1965) 1755.
The fermioriic path integral is discussed in, for example:
19. F.A. Berezin, The method of second quantization. Academic Press (1966).
20. T. Kashiwa, Y. Ohnuki and M. Suzuki, Path integral methods. Clarendon
Press (1997).
21. M.S. Swarison, Path integrals and quantum processes. Academic Press (1997).
22. M. Chaichian and A. Demichcv, Path integrals in physics. Inst. of Phys.
Pub. (2001).
C.S Quantum theory of photons and the phase operator

The quantization of the electromagnetic field was clearly formulated in:
23. P.A.M. Dirac, Proc. Roy. Soc. of London A 114 (1927) 243
where the notion of the photon phase operator was introduced. The general
treatment of constrained systems is found in:
24. P.A.M. Dirac, Lectures on quantum field theory. Yeshiva University, New
York (1966).
The path integral formulation of gauge theory was given in:
25. L.D. Faddeev and V.N. Popov, Phys. Lett. B 25 (1967) 29.
The BRST symmetry was introduced and its applications were discussed in:
26. C. Becchi, A. Rouet and R. Stora, Commun. Math. Phys. 42 (1975) 127.
27. J. Zinn-Justin, Lecture notes in physics 37 Springer-Verlag (1975).
264 REFERENCES WITH BRIEF COMMENTS
For reviews of the BRST symmetry sec, for example:

28. N. Nakanishi and I. Ojima, Covariant operator formalism, of gauge theories
and quantum gravity. World Scientific (1990).
29. J. Gomis. J. Paris and S. Samuel, Phys. Rep. 259 (1995) 1.
For an attempt to construct the photon phase operator (including the past ref-
erences) see:
30. D.T. Pcgg and S.M. Barnett. Phys. Rev., A 39 (1989) 1665.
The connection between the phase operator of the photon (and harmonic oscil-
lator) and the index theorem was discussed in:
31. K. Fujikawa, Phys. Rev. A 52 (1995) 3299.
C.4 Regularization of field theory and chiral anomalies

A brief summary of U(l) gauge theory (quantum electrodynamics) is given in
Appendix A, and we here give several standard textbooks:
32. J.D. Bjorkeii and S.D. Drell, Relativistic quantum fields. McGraw-Hill (1965).
33. K. Nishijima, Fields and particle,?. Benjamin (1969).
34. N.N. Bogoliubov arid D.V. Shirkov. Introduction to the theory of quantized
fields (3rd edn). John Wiley (1980).
35. ,T.C. Taylor, Gauge theories of weak interactions. Cambridge University
Press (1976).
36. L.D. Faddeev and A.A. Slavnov. Gauge fields. Benjamin/Cummings (1980).
37. P. Ramond, Field theory—A modern primer. Addison-Wesley (1989).
38. M. Kaku, Quantum field theory: a modern introduction. Oxford University
Press (1993).
39. M.E. Peskin and D.V. Schroeder, Quantum, field theory. Addison-Wesley
(1995).
40. S. Weinberg, The quantum theory of fields, I and II. Cambridge University
Press (1995).
41. J. Zinn-Justin, Quantum field theory and critical phenomena (4th edn).
Oxford University Press (2002).
The notation of the present book follows that of the textbook [32].
As for the contents of Chapter 4, the higher-derivative regularization is ex-
plained in detail in [36]. A readable account of the Wick rotation is found in:
42. P. van Nieuwenhuizcn and A. Waldron, Phys. Lett. B 389 (1996) 29.
The covariant regularization of the currents discussed in Chapter 4 follows the
references:
43. K. Fujikawa, Phys. Rev. D 29 (1984) 285.
44. K. Fujikawa. Nucl. Phys. B 428 (1994) 169.
The covariant regularization is closely related to the generalized Pauli-Villars
regularization:
THE JACOBIAN IN PATH INTEGRALS AND QUANTUM ANOMALIES 265
45. S.A. Frolov and A.A. Slavnov, Phys. Lett. B 309 (1993) 344.
46. R. Narayanan and H. Neuberger, Phys. Lett. B 302 (1993) 62.
47. S. Aoki and Y. Kikukawa, Mod. Phys. Lett. A 8 (1993) 3517.
In ordinary perturbative calculations, the dimensional regularization in:
48. G. 't Hooft and M. Veltman, Nud. Phys. B 44 (1972) 189.
49. C.G. Bollini and J.J. Giambiagi, Nuovo Cim. B 12 (1972) 20
is often used, but this regularization is not convenient for the analyses of quantum
anomalies.
The analyses of triangle Fcynman diagrams in Chapter 1 are explained in
detail in [9] [10]. The Adler-Bardeen theorem is given in:
50. S. Adler and W.A. Bardccn, Phys. Rev. 182 (1969) 1517.
51. A. Zee, Phys. Rev. Lett. 29 (1972) 1198.
52. K. Higashijima, K. Nishijima and M. Okawa, Prog. Theor. Phys. 67 (1982)
668.
C.5 The Jacobian in path integrals and quantum anomalies

Chapter 5 gives a path integral reformulation of classic Feynman diagrammatic
analyses, which directly leads to the chiral identities with anomaly terms present.
This path integral method was given in:
53. K. Fujikawa, Phys. Rev. Lett. 42 (1979) 1195.
54. K. Fujikawa, Phys. Rev. D 21 (1980) 2848; 22 (1980) 1499(E).
The proof of the Atiyah-Singer index theorem, which is based on a technique
closely related to the method in physics, is found in:
55. M. Atiyah, R. Bott, and V.K. PatodL Invent. Math. 19 (1973) 279.
An explicit analysis of this theorem in the context of flat four-dimensional Eu-
clidean space is given in:
56. R. Jackiw and C. Rebbi, Phys. Rev. D 16 (1977) 1052.
Our analysis of the index theorem closely follows these two references.
The instanton solution is found in:
57. A.A. Belaviri, A.M. Polyakov, A.S. Schwartz and Yu.S. Tyupkin, Phys.
Lett. B 59 (1975) 85.
The physical implications of this solution have been clarified in:
58. G. 't Hooft, Phys. Rev. Lett. 37 (1976) 8.
59. G. 't Hooft, Phys. Rev. D 14 (1976) 3432.
60. R. Jackiw and C. Rebbi, Phys. Rev. Lett. 37 (1976) 172.
61. C.G. Callan, R.F. Dashen arid D.J. Gross, Phys. Lett. B 63 (1976) 334.
Reference [59] analyzes the so-called U(l) problem related to the 77' meson from
the viewpoint of the competition of the Nambu-Goldstone theorem and the
quantum anomaly, and references [60] [61] introduced the notion of the 0 vacuum.
The U(l) problem, strong CP a,nd axions are reviewed in:
62. S. Coleman, The uses of instantons. in: Aspects of symmetry. Cambridge

University Press (1985).
63. G. 't Hooft, Phys. Re.pt. 142 (1986) 357.
64. J.E. Kirn, Phys. Rc.pt. 150 (1987) 1.
The analyses of the Jacobian in a general context have been given in:
65. A. Diaz, W. Troost, P. van Nieuwerihuizen and A. Van Proeyen, Int. J.
Mod. Phys. A 4 (1989) 3959.
66. W. Troost. P. van Nieuwerihuizen and A. Van Proeyen, Nud. Phys. B 333
(1990) 727.
The regularization of the Jacobian has been discussed in:
67. A.A. Andrianov, L. Bonoraand R. Gamboa-Saravi. Phys. Rev. D 26 (1982)
2821.
68. D.W. McKay and B.-L. Young, Phys. Rev. D 28 (1983) 1039.
69. G.A. Christos, Z. Phys. C 18 (1983) 155.
70. R.E. Gamboa-Saravi, M.A. Muschietti, F.A. Schaposnik and J.E. Solomin,
Annals Phys. 157 (1984) 360.
71. M. Renter, Phys. Rev. D 31 (1985) 1374.
72. R.I. Nepomechic, Phys. Rev. D 31 (1985) 3291.
73. S.D. Joglekar and A. Misra, Phys. Rev. D 36 (1987) 642.
74. D.V. Vassilevich, Heat kernel expansion: User's manual, hep-th/0306138.
The index and related matters in path integrals have been discussed in:
75. R. Musto, L. O'Raifeartaigh and A. Wipf, Phys. Lett. B 175 (1986) 433;
Nud. Phys. D 293 (1987) 559.
76. A.J. Niemi and G.W. Semenoff, Nud. Phys. B 269 (1986) 131.
77. P.O. Mazur and V.P. Nair, Nud. Phys. B 284 (1987) 146.
78. T.P. Branson. P.B. Gilkey, D.V. Vassilevich. J. Math. Phys. 39 (1998)
1040; 41 (2000) 3301(E).
C.6 Quantum breaking of gauge symmetry

The quantum breaking of gauge symmetry was evaluated by using the Pauli-
Villars regularization in:
79. W.A. Bardeen, Phys. Rev. 184 (1969) 1848.
Other related calculations are found in:
80. R.W. Brown, C.-C. Shih and B.-L. Young, Phys. Rev. 186 (1969) 1491.
81. A.P. Balachandran, G. Marmo, V.P. Nair and C.G. Trahern, Phys. Rev. D
25 (1982) 2713.
82. S.-K. Hu, B.-L. Young and D.W. McKay, Phys. Rev. D 30 (1984) 836.
The covariant anomaly in the path integral method was calculated in [54]. As
for the integrablc form of anomalies, see:
83. M.B. Einhorn and D.R.T. Jones. Phys. Rev. D 29 (1984) 331.
QUANTUM BREAKING OF GAUGE SYMMETRY 267
84. A.A. Andrianov and L. Bonora, Nud. Phys. B 233 (1984) 232; 233 (1984)
247.
Reference [84] discussed the rotation of the gauge field to a pure imaginary value.
The Pauli-Villars regulari/ation in the path integral was discussed in [54].
For the anomaly cancellation in gauge theory (in particular, in Weinberg-
Salam theory), see:
85. C. Bouchiat, J. Iliopoulos and P. Meyer, Phys. Lett. B 38 (1972) 519.
86. D.J. Gross and R. Jackiw, Phys. Rev. D 6 (1972) 477.
The analysis of anomalous gauge theory in the present book follows:
87. K. Fujikawa, Phys. Lett. B 171 (1986) 424.
The algebraic analyses of the Gauss operators in anomalous theory are found in,
for example:
88. L.D. Faddeev, Phys. Lett. B 145 (1984) 81.
89. S.G. Jo, Nud. Phys. B 259 (1985) 616.
90. M. Kobayashi, K. Sco and A. Sugamoto, Nud. Phys. B 273 (1986) 607.
There are attempts to give a consistent physical meaning to anomalous gauge
theories. See, for example:
91. L.D. Faddccv and S.L. Shatashvili, Phys. Lett. B 167 (1986) 225.
92. R. Jackiw and R. Rajaraman, Phys. Rev. Lett. 54 (1985) 1219; 2026(E).
93. E. D'Hoker and E. Farhi, Nud. Phys. B 248 (1984) 59.
94. K. Harada and I. Tsutsui, Phys. Lett. B 183 (1987) 311.
95. 0. Babelon, F.A. Schaposnik and C.M. Viallet, Phys. Lett. B 177 (1986)
385.
A calculational method of covariant anomalies is found in [54] arid
The fermion number non-conservation in chiral gauge theory, which is naturally
evaluated in the covariant regularization, was pointed out in:
97. G. 't Hooft, Phys. Rev. Lett. 37 (1976) 8.
The connection between the covariant and integrable anomalies is discussed
in detail in:
98. J. Wess and B. Zumino, Phys. Lett. B 37 (1971) 95.
99. W.A. Bardccn and B. Zumino, Nud. Phys. B 244 (1984) 421.
Reference [98] introduced the notion of the Wess—Zumino iiitegrability condition
and the Wess- Zumino term. The method to convert the covariant anomaly to
an integrable form of anomaly in the present book follows the analyses in:
100. H. Bancrjcc. R. Banerjee and P. Mitra, Z. Phys. C 32 (1986) 445.
101. H. Leutwyler, Phys. Lett. B 152 (1985) 78.
A more mathematical differential geometrical method, such as cohomology

and descent formulas, which were only briefly commented on in the present book,
are found in [99] and the following references
102. R. Jackiw, in: Relativity, groups and topology II. B.S. DeWitt and R. Stora
(eds.), North Holland (1984) p. 221.
103. B. Zumino, in: Relativity, groups and topology II. B.S. DeWitt and R. Stora
(eds.), North Holland (1984) p. 1291.
104. R. Stora, in: Progress in gauge field theory. H. Lehman et al. (eds.), NATO
ASI Series B, Physics, 115, Plenum (1984) p. 543.
105. B. Zumino. Y.-S. Wu and A. Zee, Nud. Phys. B 239 (1984) 477.
106. L. Alvarez-Gaume and P. Ginsparg, Nud. Phys. B 243 (1984) 449.
107. T. Sumitarii, J. Phys. A 17 (1984) L811.
For the analyses of the Wess-Zumino term in chiral effective theory, sec the
following references in addition to [98]
108. E. Witten, Nud. Phys. B 223 (1983) 422.
109. E. Witten, Nud. Phys. B 223 (1983) 433.
110. K.-C. Chou, H.-Y. Guo, K. Wu and X.-C. Song, Phys. Lett. B 134 (1984)
67.
111. H. Kawai and S.H.H. Tye, Phys. Lett. B 140 (1984) 403.
The Goto-Iinarnura-Schwinger term was introduced in:
112. T. Goto and I. Imamura. Prog. Theor. Phys. 14 (1955) 196.
113. J. Schwinger, Phys. Rev. Lett. 3 (1959) 296.
The analysis of the cancellation of the Goto-Imamura-Schwinger term in the
Gauss operator for quantum electrodynamics in four-dimensions follows:
In the two-dimensional theory such as in the analysis of the Kac-Moody algebra
in Chapter 8, the distinction between the quantum anomaly and the Goto-
Imarmira-Schwinger term becomes less clear.
Further analyses of the general aspects of path integrals and anomalies are
given in:
115. R.I. Nepomechie, Annals Phys. 158 (1984) 67.
117. H. Banerjee arid R. Banerjee, Phys. Lett. B 174 (1986) 313.
118. D. Evens, J.W. Moffat, G. Kleppe and R.P. Woodard, Phys. Rev. D 43
(1991) 499.
119. K. Okuyama and H. Suzuki, Phys. Lett. B 382 (1996) 117.
They are reviewed in:
120. R.D. Ball, Phys. Rept. 182 (1989) 1.
The detailed analyses of commutator anomalies and related issues arc found
in:
QUANTUM BREAKING OF GAUGE SYMMETRY 269

122. S. Hosono and K. Seo, Phys. Rev. D 38 (1988) 1296.
123. A.Yu. Alekseev, Ya. Madaichik, L.D. Faddeev and S.L. Shatashvili. Theor.
Math. Phys. 73 (1988) 1149.
124. T. Nishikawa and I. Tsutsui, Nucl. Phys. D 308 (1988) 544.
125. I. Tsutsui, Phys. Lett. B 229 (1989) 51.
126. J. Alfaro and P.H. Daingaard, Annals Phys. 202 (1990) 398.
127. T. Fujiwara, Y. Igarashi and J. Kubo, Nucl Phys. B 358 (1991) 195.
128. J. Gomis and J. Paris, Nucl Phys. B 395 (1993) 288.
The baryon number violation and related issues arc discussed in:
129. II. Yamagishi, Phys. Rev. D 28 (1983) 977.
130. O. Espinosa, Nucl. Phys. B 343 (1990) 310.
131. K. Funakubo, Prog. Theor. Phys. 96 (1996) 475.
The anomalies and path integrals are useful in the effective descriptions (and
rion-liriear cr-models) of weak and strong interactions.
132. A. Dhar and S.R. Wadia, Phys. Rev. Lett. 52 (1984) 959.
133. A. Manohar and G.W. Moore, Nucl. Phys. B 243 (1984) 55.
134. G.W. Moore and P. Nelson, Phys. Rev. Lett. 53 (1984) 1519.
135. J. Balog, Phys. Lett. B 149 (1984) 197.
136. G.W. Moore and P. Nelson, Commun. Math. Phys. 100 (1985) 83.
137. A. Dhar, R. Shankar and S.R. Wadia, Phys. Rev. D 31 (1985) 3256.
138. A.A. Andrianov, Phys. Lett. B 157 (1985) 425.
139. P. Simic, Phys. Rev. D 34 (1986) 1903.
140. M. Wakamatsu and W. Weise, Z. Phys. A 331 (1988) 173.
141. M. Wakamatsu, Annals Phys. 193 (1989) 287.
142. D. Espriu, E. de Rafael and J. Taron, Nucl. Phys. B 345 (1990) 22; 355
(1991) 278(E).
143. J.G. Korner and G. Thompson, Phys. Lett. B 264 (1991) 185.
144. J. Preskill. Annals Phys. 210 (1991) 323.
145. M.M. Islam, Z. Phys. C 53 (1992) 253.
146. H. Ito, W.W. Buck and F. Gross, Phys. Lett. B 287 (1992) 23; Phys. Rev.
C 45 (1992) 1918.
147. J. Wudka, Int. J. Mod. Phys. A 9 (1994) 2301.
148. R. Alkofer and C.D. Roberts, Phys. Lett. B 369 (1996) 101.
•149. P. Herrera-Siklody. J.I. Latorre, P. Pascual and J. Taron, Nucl Phys. B
497 (1997) 345.
150. F. Araki, M. Musakhanov and H. Toki, Phys. Rev. D 59 (1999) 037501.
151. T. Feldmann, Int. J. Mod. Phys. A 15 (2000) 159.
Several reviews:
152. I. Zahed and G.E. Brown, Phys. Re-pi. 142 (1986) 1.
153. M. Bando, T. Kugo and K. Yamawaki, Phys. Rept. 164 (1988) 217.
154. R. Alkofer, H. Reinhardt and H. Weigel, Phys. Rept. 265 (1996) 139.
155. J. Bijnens, Phys. Rept. 265 (1996) 369.
As for chiral perturbation theory, see:
156. J. Gasser and H. Leutwylcr, Annals Phys. 158 (1984) 142; Nud. Phys. B
250 (1985) 465.
C.7 The Weyl anomaly and renormalization group

A readable account of the scale transformation in field theory is found in:
157. S. Colcman and R. Jackiw, Annals Phys. 67 (1971) 552.
The Weyl (or conformal or trace) anomaly was discussed in:
158. R..T. Crewther, Phys. Rev. Lett. 28 (1972) 1421.
159. M. Chanowitz and J. Ellis, Phys. Lett. B 40 (1972) 397.
The path integral formulation of the Weyl anomaly is given in:
160. K. Fujikawa. Phys. Rev. Lett. 44 (1980) 1733.
The description of the present book is mainly based on [161], where the improved
energy-momentum tensor discussed in the present book is also given.
A detailed analysis of the renormalization properties of the Weyl anomaly in
gauge theory is given in:
162. S.L. Adler, J.C. Collins arid A. Duncan, Phys. Rev. D 15 (1977) 1712.
163. N.K. Nielsen, Nud. Phys. B 120 (1977) 212.
The evaluation of the 13 function in the present book is based on:
164. K. Fujikawa, Phys. Rev. D 48 (1993) 3922
which is closely related to the calculation in:
165. G. 't Hooft, Nud. Phys. B 62 (1973) 444.
As for the details of the background field method including earlier references,
see:
166. S. Ichinose and M. Omote, Nud. Phys. B 203 (1982) 221.
The (3 function in QCD was first evaluated in:
167. D. Gross arid F. Wilezek, Phys. Rev. Lett. 30 (1973) 1343.
168. II.D. Politzer, Phys. Rev. Lett. 30 (1973) 1346.
As for the Callan- Symanzik equation and the renormalization group discussed
in the present book, see:
169. C.G. Callan, Phys. Rev. D 2 (1970) 1541.
170. K. Symanzik, Commun. Math. Phys. 18 (1970) 227.
171. G. 't Hooft, Nud. Phys. B 61 (1973) 455.
172. S. Weinberg, Phys. Rev. D 8 (1973) 3497.
THE WEYL ANOMALY AND RENORMALIZATION GROUP 271
The evaluations of the Weyl anomaly in curved space-time are found in:
173. S. Dcser, M.J. Duff and C.J. fsham, Nud. Phys. B 111 (1976) 45.
174. L.S. Brown and J.P. Cassidy, Phys. Rev. D 15 (1977) 2810.
175. S.W. Hawking, Commun. Math. Fhys. 55 (1977) 133.
176. R. Endo, Prog. Theor. Phys. 71 (1984) 1366.
The last reference gives a readable account of the heat kernel method. Further
analyses of the Weyl anomaly are found in:
177. L. Bonora, P. Pasti and M. Bregola, Class. Quant. Grav. 3 (1986) 635.
178. N.K. Nielsen and P. van Nieuwenhuizcn, Phys. Rev. D 38 (1988) 31.83.
179. F. Bastianelli, Nud. Phys. B 376 (1992) 113.
180. F. Bastianelli and P. van Nieuwenhuizen, Nud. Phys. B 389 (1993) 53.
181. M.J. Duff, Class. Quant. Grav. 11 (1994) 1387.
182. S. Ichinose, Phys. Rev. D 57 (1998) 6224.
Applications of the energy-momentum tensor and scaling properties are found
in:
183. A. Cappelli and A. Coste, Nud. Phys. B 314 (1989) 707.
184. T. Jacobson, Phys. Rev. D 44 (1991) 1731; 48 (1993) 728.
185. J.F. Donoghue and H. Leutwylcr, Z. Phys. C 52 (1991) 343.
186. V. Moretti, Phys. Rev. D 56 (1997) 7797.
187. M. Yoshimura, Phys. Rev. D 30 (1984) 344.
188. S. Ichinose and S.D. Odiiitsov, Nud. Phys. B 539 (1999) 643.
189. S. Arnone, C. Fusi and K. Yoshida, JHEP 9902 (1999) 022.
Induced gravity and related topics arc discussed in:
190. S.L. Adler, Rev. Mod. Phys. 54 (1982) 729; 55 (1983) 837(E).
191. A. Zee, Annals Phys. 151 (1983) 431.
192. E.T. Tomboulis, Nud. Phys. B 329 (1990) 410.
193. K. Hamada, Prog. Theor. Phys. 105 (2001) 673.
As for the Konishi anomaly in supersymmetric gauge theory, which is related
to the Weyl (and also chiral) anomaly, see:
194. I.E. Clark, O. Piguet and K. Sibold, Nud. Phys. B 159 (1979) 1.
195. K. Konishi, Phys. Lett. B 135 (1984) 439.
196. K. Konishi and K. Shizuya, Nuovo Cim. A 90 (1985) 111.
197. K. Shizuya, Phys. Rev. D 35 (1987) 1848; 35 (1987) 2550.
The last two references give the path integral formulation in superspacc. As for
more recent analyses, see for example:
198. N. Arkani-Hamed and H. Murayama, JHEP 0006 (2000) 030.
C.8 Two-dimensional field theory and bosonization

The bosonization of two-dimensional fermion theory was formulated in:
199. S.R. Colcman, Phys. Rev. D 11 (1975) 2088.
The path integral bosonization discussed in the present book was initiated in:
200. R. Roskies and F. Schaposnik, Phys. Rev. D 23 (1981) 558.
201. R.E. Gamboa-Savari, F.A. Schaposnik and J.E. Solornin, Nucl. Phys. B
185 (1981) 239.
202. K. Furuya. R.E. Gamboa-Savari and F.A. Schaposnik, Nucl. Phys. B 208
(1982) 159.
The description in the present book is based on an unpublished note by K. Seo
and
203. C.M. Naon, Phys. Rev. D 31 (1985) 2035.
The non-Abelian bosonization was first given in:
204. E. Witteii, Commun. Math. Phys. 92 (1984) 455
on the basis of a different formulation.
There are many references to bosonization, and some of the early references
are:
205. A.M. Polyakov and P.B. Wiegmann. Phys. Lett. B 131 (1983) 121; 141
(1984) 223.
206. D. Gonzales and A.N. Redlich, Phys. Lett. B 147 (1984) 150.
207. P. Di Vecchia and P. Rossi, Phys. Lett. B 140 (1984) 344.
The analysis of the local counter-term in the context of the bosonization in
this book is based on:
208. K. Fujikawa and H. Suzuki, hep-th/0305008.
209. N. Banerjee and R. Baiierjee, Nucl. Phys. B 445 (1995) 516.
210. E.M.C. Abreu, R. Banerjee and C. Wotzasek, Nucl. Phys. B 509 (1998)
519.
Further references related to the path integral formulation are:
211. M. Reuter and W. Dittrich, Phys. Rev. D 32 (1985) 513.
212. J.N. Webb, Z. Phys. C 31 (1986) 301.
213. A.K. Das and V.S. Mathur. Phys. Rev. D 33 (1986) 489.
214. K.D. Rothe, Nucl. Phys. B 269 (1986) 269.
215. R. Banerjee, Phys. Rev. Lett. 56 (1986) 1889.
216. S. Miyake arid K. Shizuya, Phys. Rev. D 36 (1987) 3781.
217. S. Miyake and K. Shizuya, Phys. Rev. D 37 (1988) 2282.
218. M.N. Saniclcvici, G.W. Semenoff and Y.-S. Wu, Nucl. Phys. B 312 (1989)
197.
219. K. Shizuya, Phys. Lett. B 213 (1988) 298.
220. X-Q. Luo and Q-Z. Chen, J. Phys. G 16 (1990) 1181.
TWO-DIMENSIONAL FIELD THEORY AND BOSONIZATION 273
221. C.M. Naon, M.C. von Reichenbach and M.L, Trobo, Nucl. Phys. B 435
(1995) 567.
222. J.V. Steele, A. Subramanian and I. Zahed, Nucl. Phys. B 452 (1995) 545.
223. D.J. Gross, I.R. Klebanov, A.V. Matytsin and A.V. Srnilga, Nucl. Phys. B
461 (1996) 109.
224. S.D. Joglekar and G. Saini, Phys. Rev. D 43 (1991) 1355.
225. M. Faber and A.N. Ivanov, Eur. Phys. J. C 20 (2001) 723.
For reviews, sec:
226. E. Abdalla, M.C.B. Abdalla and K.D. Rothe, Non-perturbative methods in
two-dimensional quantum field theory. World Scientific (1991).
227. E. Abdalla and M.C.B. Abdalla, Phys. Kept. 265 (1996) 253.
There are a vast number of references to conforriial field theory, and see, for
example, the following review:
228. P. Ginsparg, Applied conformal field theory, in Proc. of Les Houches Sum-
mer School (1988).
The analysis of anomalies and central extensions of infinite-dimensional algebra
in this book is close to that in [114].
As for the applications of conformal theory to condensed matter theory, sec
[41] and
229. N. Kawakami and S.K. Yang, Conformal field theory and one-dimensional
quantum systems. Iwariaiiii Publishing House (1997).
The B.TL prescription has been given in:
230. J.D. Bjorken. Phys. Rev. 148 (1966) 1467.
231. K. Johnson and F.E. Low, Prog. Theor. Phys. Suppl. 37-38 (1966) 74.
Very early references to the first quantization of string theory by the path
integral method are:
232. A.M. Polyakov; Phys. Lett. B 103 (1981) 207.
234. O. Alvarez, Nucl. Phys. B 216 (1983) 125.
235. D. Friedan, E.J. Martinec and S.H. Shenker, Nucl. Phys. B 271 (1986) 93.
The treatment of the present book is based on [233], where the BRST formula-
tion was first introduced into string theory and the ghost number anomaly was
discovered. The connection of this ghost number anomaly to the Riemann-Roch
theorem was recognized in [235].
Some of the early quantum analyses of the Liouville action which suggested
the replacement d — 26 —¥ d — 25 are:
236. A.M. Polyakov, Mod. Phys. Lett. A 2 (1987) 893.
237. K. Fujikawa, T. Inagaki and H. Suzuki, Phys. Lett. B 213 (1988) 279.
238. F. David, Mod. Phys. Lett. A 3 (1988) 1651.
239. J. Distler and H. Kawai, Nucl. Phys. B 321 (1989) 509.
The gravitational path integral in [233] is based on the prescription in Ap-

pendix B. of which details are found in:
240. K. Fujikawa, Nucl. Phys. B 226 (1983) 437.
241. K. Fujikawa and 0. Yasuda, Nucl. Phys. B 245 (1984) 436.
As for supersymmetry aspects, see:
242. U. Lindstrom, N.K. Nielsen, M. Rocek and P. van Nieuwcnhuizen, Phys.
Rev. D 37 (1988) 3588.
The justification of the use of plane waves in curved-space, in particular in two
dimensions, which is essential for our discussion of the bosoriic string, is given
in:
243. A. Ceresole, P. Pizzochero, and P. van Nieuwenhuizen. Phys. Rev. D 39
(1989) 1567.
244. J. Kowalski-Glikman and P. van Nieuwenhuizen, Int. J. Mod. Phys. A 6
(1991) 4077.
The quantum anomaly and energy-momentum tensor in two-dimensional the-
ory by the path integral formulation are illustrated in:
245. K. Fujikawa, U. Lindstrom, N.K. Nielsen, M. Rocek and P. van Nieuwen-
huizen, Phys. Rev. D 37 (1988) 391.
Further careful analyses of the path integral measure for first quantized
strings are found in:
246. P. Bouwknegt and P. van Nieuwenhuizen, Class. Quant. Grav. 3 (1986)
207.
247. M. Rocek, P. van Nieuwenhuizen and S.C. Zhang, Annals Phys. 172 (1986)
348.
248. A. Eastaugh. L. Mezinccscu, E. Sezgin and P. van Nieuwenhuizen, Phys.
Rev. Lett. 57 (1986) 29.
249. F. Bastiarielli and P. van Nieuwenhuizen, Phys. Rev. D 41 (1990) 2525.
and also related analyses in:
250. N. Ohta, Phys. Rev. D 33 (1986) 1681.
251. U. Lindstrom, M. Rocek and P. van Nieuwenhuizen, Phys. Lett. B 199
(1987) 219.
252. W.I. Weisberger, Nucl. Phys. B 284 (1987) 171.
253. Y. Tanii, Nucl. Phys. B 289 (1987) 187.
254. D. Chang and A. Kumar, Phys. Rev. D 35 (1987) 1388.
255. L. Baulieu, W. Siegel and B. Zwiebach, Nud. Phys. B 287 (1987) 93.
256. S.D. Mathur and S. Mukhi, Phys. Rev. D 36 (1987) 465.
257. D.W. Dusedau, Phys. Lett. B 188 (1987) 51.
258. K. Fujikawa, Nucl Phys. B 291 (1987) 583.
259. A. Rebhan and U. Kraemmcr. Phys. Lett. B 196 (1987) 477; Nud. Phys.
B 315 (1989) 717.
INDEX THEOREM ON THE LATTICE AND CHIRAL ANOMALIES 275
260. D.N. Petcher and J.W. van Molten, Phys. Lett. B 194 (1987) 221.
261. C. Batlle. J. Gomis and J. Roca, Phys. Rev. D 40 (1989) 1950.
262. J. Govacrts, Int. J. Mod. Phys. A 4 (1989) 173.
263. U. Ellwanger and J. Fuchs, Nud. Phys. B 312 (1989) 95.
264. N. Ohta and S. Osabe, Phys. Rev. D 39 (1989) 1641.
265. T. Fujiwara, Y. Igarashi and J. Kubo. Nud. Phys. B 341 (1990) 695.
266. M. Henrieaux, Phys. Lett. B 238 (1990) 299.
267. K. Fujikawa and H. Suzuki, Nud. Phys. B 361 (1991) 539.
268. J. Gomis and H. Suzuki, Phys. Lett. B 278 (1992) 266.
269. E.S. Fradkin and V.Ya. Linetsky, Phys. Lett. B 261 (1991) 26.
270. J. Gomis and ,7. Paris, Nud. Phys. B 395 (1993) 288.
271. T. Fujiwara, Y. Igarashi, J. Kubo and K. Maeda, Nud. Phys. B 391 (1993)
211.
272. F. De Jonghe. J. Paris and W. Troost, Nud. Phys. B 476 (1996) 559.
and reviews in:
273. C.B. Thorn, Phys. Kept. 175 (1989) 1.
274. W. Siegel, Fields, hep-th/9912205.
Some aspects of the path integral measure for quantum gravity are discussed
in:
275. M. Ebner, R. Held and J. Lopes, Z. Phys. C 37 (1987) 85.
276. D. Anselmi, Phys. Rev. D 45 (1992) 4473.
277. A.O. Barvinsky. Phys. Re.pt. 230 (1993) 237.
278. K. Hamada, Prog. Theor. Phys. 103 (2000) 1237.
C.9 Index theorem on the lattice and chiral anomalies

For the earlier analyses of chiral anomalies and fermions on the lattice, see:
279. L.H. Karsten and J. Smit, Nud. Phys. B 183 (1981) 103.
280. H.B. Nielsen and M. Niiiorniya, Nud. Phys. B 185 (1981) 20; Phys. Lett.
B 105 (1981) 219.
281. N. Kawamoto and J. Smit, Nud. Phys. B 192 (1981) 100.
282. W. Kerler, Phys. Rev. D 23 (1981) 2384.
283. E. Seller and 1.0. Stamatescu, Phys. Rev. D 25 (1982) 2177.
284. S. Aoki, Phys. Rev. D 30 (1984) 2653; Phys. Rev. D 35 (1987) 1435.
285. F. Karsch, E. Seller and I.O. Stamatescu, Nud. Phys. B 271 (1986) 349.
286. P. Hernandez and R. Sundrum, Nud. Phys. B 455 (1995) 287.
The domain-wall fermions arc formulated in:
287. D.B. Kaplan, Phys. Lett. B 288 (1992) 342.
288. Y. Shamir, Nud. Phys. B 406 (1993) 90.
289. V. Furman and Y. Shamir, Nud. Phys. B 439 (1995) 54.
290. P.M. Vranas, Phys. Rev. D 57 (1998) 1415.
291. H. Neuberger, Phys. Rev. D 57 (1998) 5417.

292. Y. Kikukawa and T. Noguchi, hcp-lat/9902022.
293. M.F. Golterman, K. Jansen and D.B. Kaplan, Phys. Lett. B 301 (1993)
219.
The overlap formula for fermions was discussed in:
294. R. Narayanan and H. Neuberger, Nucl. Phys. B 412 (1994) 574; Phys.
Rev. Lett. 71 (1993) 3251; Nucl. Phys. D 443 (1995) 305.
295. S. Randjbar-Dacmi and J. Strathdee, Phys. Lett. B 348 (1995) 543; Nucl.
Phys. B 443 (1995) 386; 466 (1996) 335.
The recent developments are based on the algebraic relation noted in:
296. P.H. Ginsparg and K.G. Wilson, Phys. Rev. D 25 (1982) 2649
and the construction of an explicit solution (the overlap operator):
297. H. Neuberger, Phys. Lett. B 417 (1998) 141; 427 (1998) 353
which contains an analysis of the index. See also:
298. P. Hasenfratz, Nucl. Phys. B 525 (1998) 401.
The index theorem on the lattice was further clarified in:
299. P. Hasenfratz. V. Laliena and F. Niedermayer, Phys. Lett. B 427 (1998)
125
and the recognition of the chiral anomaly as the Jacobian even for the lattice
theory in
300. M. Liischer, Phys. Lett. B 428 (1998) 342.
was important. For reviews of these developments, see:
301. F. Niedermayer, Nucl. Phys. (Proc. Suppl.) 73 (1999) 105.
302. H. Neuberger, Ann. Rev. Nucl. Part. Sci. 51 (2001) 23.
303. M. Liischer, Lectures given at the International School of Subnuclear Physics.
Erice 2000, hep-th/0102028.
304. M. Creutz, Rev. Mod. Phys. 73 (2001) 119.
For the detailed evaluation of the chiral anomaly on the lattice, see, for ex-
ample:
305. Y. Kikukawa and A. Yarnada, Phys. Lett. B 448 (1999) 265.
306. D.H. Adams, Annals Phys. 296 (2002) 131.
307. H. Suzuki, Prog. Theor. Phys. 102 (1999) 141.
308. K. Fujikawa, Nncl. Phys. B 546 (1999) 480.
309. T. Fujiwara. K. Nagao and H. Suzuki, JHEP 0209 (2002) 025.
A generalized form of the Ginsparg-Wilson algebra and its solutions are rioted
in:
310. K. Fujikawa, Nucl. Phys. B 589 (2000) 487.
GRAVITATIONAL ANOMALIES 277
The proof of the locality of the fermion operator, when a suitable admissibility
condition is imposed, exists at this moment only for the simplest operator given
in [297]:
311. P. Hernandez, K. Jansen and M. Luscher, Nucl. Phys. B 552 (1999) 363.
312. H. Neubergcr, Phys. Rev. D 61 (2000) 085015.
The analysis of the locality for the general case is incomplete in
313. K. Fujikawa and M. Ishibashi, Nucl. Phys. B 605 (2001) 365.
314. T-W. Chiu, Nucl. Phys. (Proc. Suppl.) 94 (2001) 733
but sufficient to analyze the chiral anomaly:
315. K. Fujikawa and M. Ishibashi, Nucl. Phys. B 587 (2000) 419.
The meaning of the heavy topological excitations N± appearing in the Ginsparg-
Wilson operators was clarified in:
316. T-W. Chiu, Phys. Rev. D 58 (1998) 074511.
Abelian chiral gauge theories were formulated on the basis of the Ginsparg-
Wilson relation in:
318. M. Luscher, Nucl. Phys. B 549 (1999) 295.
For non-Abelian theories, however, the formulation is still in the process of con-
struction. See, for example:
319. H. Neuberger, Phys. Rev. D 59 (1999) 085006.
320. H. Suzuki, Prog. Theor. Phys. 101 (1999) 1147.
321. M. Luscher, Nucl. Phys. B 568 (2000) 162; JEEP 0006 (2000) 028.
322. H. Suzuki, Nucl. Phys. B 585 (2000) 471; JHEP 0010 (2000) 039.
323. Y. Kikukawa and Y. Nakayama, Nucl. Phys. B 597 (2001) 519.
324. Y. Kikukawa, Phys. Rev. D 65 (2002) 074504.
A non-commutative technique in lattice theory is noted in:
325. T. Fujiwara, H. Suzuki and K. Wu, Nucl. Phys. B 569 (2000) 643.
For some interesting applications of the Ginsparg-Wilson fermions, see:
326. H. So and N. Ukita, Phys. Lett. B 457 (1999) 314.
327. O. Bar and I. Campos, Nucl. Phys. B 581 (2000) 499.
328. W. Bietenholz and J. Nishimura, JHEP 0107 (2001) 015.
329. L. Giusti, G.C. Rossi, M. Testa and G. Veneziano, Nucl. Phys. B 628
(2002) 234.
C.10 Gravitational anomalies

The calculation of the chiral U(l) anomaly in the presence of the background
gravitational field was performed in:
330. T. Kimura, Prog. Theor. Phys. 42 (1969) 1191.
331. R. Delbourgo and A. Salam, Phys. Lett. B 40 (1972) 381.
332. T. Eguchi and P.G.O. Freund, Phys. Rev. Lett. 37 (1976) 1251.
The appendix in reference [330] gives a concise account of the basic properties
of curved space-time.
The generalization to spin 3/2 particles is given in:
333. N.K. Nielsen, M.T. Grisaru, H. Romer and P. van Nicuwenhuizen, Nud.
Phys. B 140 (1978) 477.
334. R. Endo and T. Kimura, Prog. Theor. Phys. 63 (1980) 683.
335. R. Erido and M. Takao, Phys. Lett. B 161 (1985) 155.
For the analysis of the Atiyah Singer index theorem by using supersymrnetric
quantum mechanics, see:
336. L. Alvarez-Gaume, J. Phys. A 16 (1983) 4177.
A detailed review of topological aspects of anomalies is found in:
337. T. Eguchi, P.B. Gilkey and A.J. Hanson, Phys. Rep. 66 (1980) 213.
The calculation of chiral anomalies in arbitrary dimensions in the present book
is based on:
338. K. Fujikawa, S. Ojima and S. Yajima, Phys. Rev. D 34 (1986) 3223
which is a simplified version of the calculation in [336] without using supersym-
metry.
The quantum breaking of general coordinate and local Lorentz transforma-
tions has been found and analyzed in detail in:
339. L. Alvarez-Gaume and E. Witten, Nud. Phys. B 234 (1984) 269.
As for related early analyses, see:
340. L. Alvarez-Gaume and P. Ginsparg, Ann. Phys. 161 (1985) 423.
341. L.N. Chang and N.T. Nich, Phys. Rev. Lett. 53 (1984) 21.
342. S. Yajima and T. Kimura, Phys. Lett. B 173 (1986) 154.
and references [99] [335]. The simple relation between the general coordinate and
local Lorentz anomalies was noted in:
343. K. Fujikawa, M. Tomita arid 0. Yasuda, Z. Phys. C 28 (1986) 289.
Further general aspects of gravitational path integrals are analyzed in:
344. H.T. Nieh, Phys. Rev. Lett. 53 (1984) 2219.
345. R. Erido. Prog. Theor. Phys. 80 (1988) 311.
346. M. Hatsuda, P. van Nieuwenhuizen. W. Troost and A. Van Proeyeri, Nud.
Phys. B 335 (1990) 166.
347. Z. Bern, E. Mottola, S.K. Blau, Phys. Rev. D 43 (1991) 1212.
348. 0. Chandia and J. Zarielli, Phys. Rev. D 58 (1998) 045014.
Various path integral evaluations of higher-dimensional anomalies have been
discussed in:
349. T. Matsuki, Phys. Rev. D 28 (1983) 2107.
350. R. Delbourgo and T. Matsuki, J. Math. Phys. 26 (1985) 1334.
351. R. Endo and M. Takao, Prog. Theor. Phys. 73 (1985) 803.

352. J.M. Gipson, Phys. Rev. D 33 (1986) 1061.
353. R. Endo and M. Takao, Prog. Theor. Phys. 78 (1987) 440.
354. J. de Boer. B. Peeters, K. Skenderis and P. van Nieuwenhuizen, Nucl. Phys.
B 459 (1996) 631.
355. S. Yajima, Class. Quant. Grav. 13 (1996) 2423.
The anomalies in a space with torsion have been analyzed in:
356. S. Yajima and T. Kimura, Prog. Theor. Phys. 74 (1985) 866.
357. S. Yajima, Prog. Theor. Phys. 79 (1988) 535.
358. 0. Chandia and J. Zanelli, Phys. Rev. D 55 (1997) 7580.
C.ll Concluding remarks

The argument related to "uncertainty relations" in Chapter 11 is based on
[54][160].
The details of the global SU(2) anomaly, which we just briefly mentioned,
are found in [108] [109] and
359. E. Witten, Phys. Lett. B 117 (1982) 324.
360. S. Elitzur and V.P. Nair, Nucl. Phys. B 243 (1984) 205.
361. L. Alvarez-Gaume, S. Delia Pietra, V. Delia Pietra, Phys. Lett. B 166
(1986) 177.
362. R.A. Baadhio, Quantum topology and global anomalies. World Scientific
(1996).
The Green-Schwarz mechanism of the anomaly cancellation in superstring
theory is given in:
363. M.G. Green and J.H. Schwarz, Phys. Lett. B 149 (1984) 117.
364. M.G. Green, J.H. Schwarz and P.O. West, Nucl. Phys. B 254 (1985) 327.
365. D.J. Gross, J.A. Harvey, E..J. Martinec and R. Rohm. Nucl. Phys. B 256
(1985) 253.
A direct evaluation of the anomaly in superstring theory is found in, for example:
366. H. Suzuki and A. Sugamoto, Phys. Rev. Lett. 57 (1986) 1665.
367. K. Pilch, A.N. Schellekens and N.P. Warner, Nucl. Phys. B 287 (1987)
362.
The detailed references to string theory are found in:
368. M.G. Green, J.H. Schwarz and E. Witten. Superstring theory I,II. Cam-
bridge University Press (1987).
369. K. Kikkawa, Quantum theory of strings. Asakura Publishing Co. (1991).
370. J. Polchinski, Superstring theory I.II. Cambridge University Press (1998).
There are a vast number of references to anomalies in supersymmctric theo-
ries. We just mention:
371. S. Ferrara and B. Zumiiio. Nucl. Phys. B 87 (1975) 207.
372. T.E. Clark, O. Piguct and K. Sibold, Nucl. Phys. B 143 (1978) 445.
373. N.K. Nielsen, Nucl. Phys. B 244 (1984) 499.
374. M.T. Grisaru and P.C. West, Nucl. Phys. B 254 (1985) 249.
375. E. Guadagnini, K. Konishi, M. Miiitchev, Phys. Lett. B 157 (1985) 37.
376. V.A. Novikov, M.A. Shifman, A.I. Vainshtein and V.I. Zakharov, Phys.
Lett. B 157 (1985) 169.
377. H. Itoyama, V.P. Nair and H.-C. Ren, Nucl. Phys. B 262 (1985) 317.
378. K. Harada and K. Shizuya. Phys. Lett. B 162 (1985) 322.
379. H. Suzuki, Phys. Rev. Lett. 56 (1986) 1534.
380. F. Cooper, A. Khare arid U. Sukhatme, Phys. Kept. 251 (1995) 267.
For recent analysis and related references, see for example:
381. Y. Ohshima, K. Okuyama, H. Suzuki and H. Yasuta, Phys. Lett. B 457
(1999) 291.
The standard textbooks on supcrsymmetry are:
382. J. Wess and J. Bagger, Supersymmetry and super gravity. Princeton Uni-
versity Press (1983).
383. S. Wcinberg, The quantum theory of fields, III. Cambridge University Press
(1999).
Supergravity is reviewed in:
384. P. van Nieuwenhuizen, Phys. Rept. 68 (1981) 189.
The anomalies in fuzzy physics have also been discussed in path integrals:
385. A.P. Balachandran and S. Vaidya, Int. J. Mod. Phys. A 16 (2001) 17.
386. J.M. Gracia-Boridia and C.P. Martin, Phys. Lett. B 479 (2000) 321.
387. E.F. Moreno and F.A. Sdiaposnik, JHEP 0003 (2000) 032.
388. J. Nishimura and M.A. Vazquez-Mozo, JHEP 0108 (2001) 033.
389. T. Nakajima, Phys. Rev. D 66 (2002) 085008.
390. C.P. Martin, Nucl. Phys. B 623 (2002) 150.
391. H. Aoki, S. Iso and K. Nagao, Phys. Rev. D 67 (2003) 065018.
For other applications, the anomaly matching condition by 't Hooft and its
implications on the rnassless composite particles are discussed in:
392. G. 't Hooft, in: Recent developments in gauge theories. G.'t Hooft et al.
(eds.), Plenum (1980) p. 135.
393. Y. Frishman, A. Schwimmer, T. Banks and S. Yankielowicz. Nucl. Phys.
B 177 (1981) 157.
394. S.R. Coleman and B. Grossman, Nucl. Phys. B 203 (1982) 205.
The anomaly in stochastic quantization has been studied in:
395. J.D. Breit, S. Gupta and A. Zaks, Nucl. Phys. B 233 (1984) 61.
396. M. Namiki. I. Ohba, S. Tanaka and D.M. Yanga, Phys. Lett. B 194 (1987)
530
and anomalies in topological field theory in:

397. R.K. Kaul and R. Rajaraman, Phys. Lett. B 249 (1990) 433.
398. D. Birmingham, M. Blau, M. Rakowski and G. Thompson, Phys. Rept.
209 (1991) 129.
The anomaly-related phenomena in condensed matter theory are discussed
in:
399. K. Ishikawa, Phys. Rev. Lett. 53 (1984) 1615; Phys. Rev. D 31 (1985) 1432.
400. R. Jackiw, Phys. Rev. D 29 (1984) 2375.
401. Z.-B. Sn and B. Sakita, Phys. Rev. Lett. 56 (1986) 780.
402. D. Boyanovsky, E. Dagotto arid E.H. Fradkin, Nucl. Phys. B 285 (1987)
340.
403. C. Mudry, C. Chamon and X.-G. Wen, Nucl. Phys. B 466 (1996) 383.
404. D.H. Kim and P.A. Lee, Annals Phys. 272 (1999) 130.
Anomaly (and parity) related phenomena in odd dimensions are discussed in:
405. L. Alvarez-Gaume, S. Delia Pictra and G.W. Moore, Annals Phys. 163
(1985) 288.
406. D. Boyanovsky and R. Blankenbecler, Phys. Rev. D 31 (1985) 3234.
407. S.G. Naculich, Nucl Phys. B 296 (1988) 837.
408. M. Gomes, R.S. Mendes, R.F. Ribeiro, A.J. da Silva, Phys. Rev. D 43
(1991) 3516.
409. C. Fosco, G.L. Rossini arid F.A. Schaposnik, Phys. Rev. Lett. 79 (1997)
1980; 79 (1997) 4296(E).
The quantum anomaly in molecular physics was also noted in:
410. H.E. Camblong, L.N. Epele, H. Fanchiotti and C.A. Garcia Canal. Phys.
Rev. Lett. 87 (2001) 220402.
411. R. Jackiw, in: M.A.B. Beg memorial volume. A. All and P. Hoodbhoy
(eds.), World Scientific, Singapore (1991).
412. B.R. Holstein, Am. J. Phys. 61 (1993) 142.
INDEX
Abelian bosonization of fcrmions, 157 chirality sum rule, 206
Abelian gauge field, 246 coherent state. 16
Adler. S., 5 color degrees of freedom, 110
Adler-Bardcen theorem, 62 compact gauge theory. 198
afflne connection, 254 compact group, 7-1
Alvarcz-Gaume, L., 231 completely anti-symmetric symbol, 62
anomalous commutator, 121 conformal field theory, 144, 178
a priori probability, 23 conformal gauge, 144
Atiyah Singer index theorem, 84, 106 connected component, 132
on the lattice, 206 contravariant vector, 259
axial-vector current, 58, 90 correlation function, 162
axial-vector gauge field, 93 cosmologica.1 term, 143
background field method, 138 Coulomb gauge, 32
Bardeen, W., 97 cova.ria.nt anomaly, 107, 117
bare identity, 62 covariant derivative, 74, 76
basic principle of gauge theory, 246 covariant regularizatioii, 51
basis vectors which diagonalize the Dirac critical string, 193
action, 88 current conservation, -19
Bell, J.S., 5 descent formula, 243
P function. 130 differential form, 231
for a QCD-typc theory. 140 dimensional regularization, 51
in QED, 135 Dira,c equation, 20
Bjorken-Johnson-Low (BJL) Dirac genus, 230
prescription, 121, 176, 180 Dirac matrices, 20
Bloch wave, 83 Dirac operator, 50
Bose particle. 15 Dirac's transformation theory, 66
Bose symmetrization factor, 117 doublet of leptons, 110
Bose symmetry, 113 doublet of quarks, 109
Bose-like spinor field, 96 -Dyson. F., 2
bra state, 41 Dyson's formula for the S-matrix. 2-19
Brillouin zone, 200 Einstein equation, 231, 257
BRST charge, 102 electromagnetic tensor, 31
BUST invariant path integral measure, energy-momentum tensor. 126, 183
189 naive, 257
BRST transformation, 258 trace of, 129
canonical mass dimension. 124 equal-time commutation relation, 177
canonical quantization of the Huclidean theory, 50
electromagnetic field, 31 in curved space-time, 124
canonical T product, 121 2n-dimcnsional, 223
central extension, 178, 188 Killer number, 194
Chern character, 230 evolution operator, 8, 17
Chern-Simons form, 82, 90 Faddeev Popov formula, 40
chiral gauge theory, 98 Faddeev Popov ghost, 38
chiral identity, 62, 225 fermion number, 104
chiral Jacobian, 05 asymmetry in. 221
chiral symmetry ferrnion, 15
spontaneous breakdown of, 90 Feynman gauge, 138
chiral transformation, 57 Feynman path integral, 23
identity for, 79 Feynman, R.P., 2
chiral WT identity, 72 Feynman's it prescription, 21, 251
282
INDEX 283
Fcynmaivs parametric representation, 56 interaction representation, 249

Feynman's propagator, 250 unitary transformation to, 86
field strength tensor, 74 Ja.ckiw, R., 5
finite operator, 129 Jacobi identity, 82
first quantization of a bosonic string, 188 Jacobian. 68
flavor freedom, 219 Kac-Moody algebra, 174
Fock space representation, 12 ket state, 41
four-component spinor, 20 Kimura, T., 6
Fresnel integral, 37 Konishi anomaly, 147
Fukuda, H., 4 Landau gauge, 32
fundamental Brillouin zone, 200 lattice gauge theory, 196
gauge condition, 32 left derivative, 16
gauge invariancc, 76 Legendre transformation, 228
gauge transformation. 31 length in the functional space, 35
of left-handed component, 111 light pseudo-scalar meson, 91
Gauss operator, 101 local counter-term, 95
Gauss Bonnet theorem, 194 local Lorentz transformation. 232
Gell-Mann matrices, 75 local Weyl transformation, 125
Gell-Mann-Nishijima-type relation, 109 Lorentz covariant T* product, 121
general coordinate transformations, 231 Lorentz indices, 253
anomaly associated with, 187 LSZ prescription. 29
identities related to, 127 mass term for the photon, 48
measure invariant under the BRST Maxwell's electromagnetic field, 31
symmetry associated with, 260 metric condition, 255
WT identity for, 233 metric of space-time, 50
general theory of relativity, 253 minimal coupling, 247
minimum uncertainty relation, 46
generators of a simple nou-Abclian group,
Minkowski indices, 253
75
Miyamoto, Y.. 4
genus, 193 mode cut-off, 68
geodesic biscalar, 226 naive chiral identity, 58
ghost number anomaly, 194 naive continuum limit, 197
Ginsparg Wilson relation, 203 naive energy-momentum tensor, 257
global SU(2) anomaly, 243 naive form factor, 56
Goto-Imamura-Schwinger term, 121, 122, Nambu—Goldstohe boson, .119
177 natural unit, 47
Grassmann number, 15 Xeuberger. H., 210
Green's function, 29 Nielsen—Ninomiya theorem, 202
Grecn-Schwarz mechanism, 245 non-Abelian bosonization, 168
Haar invariant measure, 198 nou-Abelian gauge transformation, 76
harmonic oscillator, 7 non-critical string, 193
heat kernel method, 237 non-linear a model, 119, 174
Ileisenberg, W., 7 non-renormalization theorem, 148
Heisenberg's equation of motion, 21 norm of the wave function, 1
hermitian phase operator, 42 normal product, 63, 132
Higgs particles, 111 normalization condition of the path
higher derivative regularization, 47 integral, 29
hypcrcubic lattice, 196 normalization of generators, 75
improved energy-momentum tensor, 131 Mother current, 100
index, 44, 105 operator product expansion, 180
infinitesimal chiral transformation, 76 Oppenheimer, J., 2
instanton, 79, 80 orthogonal group, 108
instanton number. 85 overlap Dirac operator, 210
integrability (Wess Zumiiio) condition, particle number representation, 13
112 partition function in statistical
integrable anomaly, 117, 153 mechanics, 23
284 INDEX
path integral measure, 26 Steinberger. J., 5

Paul! matrices, 20. 75 structure constant, 75
Pauli, W,, 15 superstring theory, 239
Pauli-Villars regularization, 96 supersymmetry, 147
photon phase operator. 40 supertrace, 19
physical Hilbert space, 210 symmetric component, 132
plaquette, 197 6 vacuum, 83
point-splitting method, 54 Thirring model
Poisson bracket, 42 massive, 163
pole, 180 massless, 160
Polyakov, A.M., 188 't Hooft, G., 105
Pontryagiu index, 106 time ordering, 24
Pontryagin number, 85 Tbmonaga. S., 2
positive definite inner product, 204 translation invariant measure, 26
projection operator, 98 triangle anomaly, 59
proton decay, 221 T* product, 176
pseudo-scalar operator, 219 • tunneling, 83
quantum anomaly uncertainty relation, 42
associated with general coordinate unit hypersurface, 81
transformations, 187 vacuum polarization tensor, 53
cancellation condition, 109 vector-like gauge field, 93
cancellation of, 102 vierbein, 253
identity, 97 Virasoro algebra, 182, 188
integrable, 117, 153 Ward-Takahashi (WT) identity, 47, 71
master formula for, 152, 225 for general coordinate transformations,
quantum chromodynamics (QCD), 74 233
wave function renormalization factor, 129
f) function, 140
weight 1/2 variables, 224, 259
quantum electrodynamics (QED), 246
Weinberg-Salam theory. 109
0 function, 135
Wess, J., 112
perturbation formula, 252
Wess-Zumino term, 117
quark mass, 89
Wess Zumino-Witten action, 171
radial quantization, 178
Weyl anomaly, 145
rcnormalization group equation, 130 in two-dimensional space-time, 144
renormalization point, 130 Weyl gauge, 32
Riemann surface, 193 Weyl transformation, 256
Riemann-Christoffel curvature tensor, 225 Jacobian for. 139
Riemann-Roch theorem, 194 local, 125
scalar curvature, 147 Wick rotation, 23, 28
scalar field, 11 Wick's theorem, 250
scale invariance, 124 Wilson, K., 197
scale transformation, 123 Wilson term, 199
Schrodinger equation, 1 winding number, 81
Schrodinger functional representation, 101 VVitten, E.. 231
Schwarz inequality, 82 £-gauge, 40
Schwinger, J., 2 Yang-Mills field, 74
Schwinger's action principle, 23, 250 f function regularization, 140
Schwinger's source function, 28 Zumino, B., 112
short distance, 78
sine-Gordon model, 163
small gauge transformation, 83
SO(10) grand unified model, 109
source field, 25
species doubling, 200
spin connection, 254
squeezing of the photon, 45

Quantum Anomaly

Uploaded by

Copyright:

Available Formats

Quantum Anomaly

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Quantum Anomaly

Uploaded by

Copyright:

Available Formats

What topics are covered in the document?

What topics are covered in the document?

What is the structure of the document?

What is the structure of the document?

INTERNATIONAL SERIES OF MONOGRAPHS ON PHYSICS No.

Series Editors J. Birman City University of New York

124. C. Kiefer: Quantum gravity

CLARENDON PRESS • OXFORD

also involved in the analysis. The renormalization group equation is regarded as

In Appendix C, some basic references directly related to the presentation of

1 Genesis of quantum anomalies 1

6.6 The Wess-Zumino integrability condition 112

C References with brief comments 262

The time development of the Schrodinger equation is described by a hermitian

deals with a system where an infinite number of harmonic oscillators interact

1.2 Is the photon massless?

1.3 The discovery of the quantum anomaly

The sum over the energy of the intermediate states En = Ee + Es generally

FIG. 1.1. Photon self-energy correction

THE FEYNMAN PATH INTEGRAL AND SCHWINGER'S

2.1 Quantum theory of a harmonic oscillator

where H is the Planck constant divided by 2?r.

and the energy eigenvalue is given by

By using the commutation relations of p and q, we can show

and the Hamiltoriian is written as

When we write the state \n) which satisfies

The time development of the Schrodinger wave function

and it shows that the time development operator (evolution operator)

2.2 Path integral for the harmonic oscillator

and p arc defined in the time-independent Schrodinger representation, but in the

We then note for smalle

In this expression, the path integral measure is denned by

If we consider the limit AT -*• oc by denoting the time derivative of q by q which

we obtain the final expression of the path integral

2.3 Quantization of a scalar field

Our notational conventions arc

where the momentum (density) conjugate to the field variable is defined by

If one considers an orthoiiormal complete system of real plane waves

inside a box with volume V, (wg(af)} which satisfies one

The Hamiltoriian is then written as

the Harniltonian is written as

and the quantization condition is given by

In a specific representation of state vectors of the Hilbert space, which is

A crucial difference of the physical interpretation of the particle number rep-

2.4 Path integral for fermions

by definition. Consequently, we have, for example, ry2 = 0 and the Grassmann

with complex coefficients cn ~ c3 by recalling £2 = (£*) 2 = 0. The integral is

and it is written as a linear combination of |0) and |1)

When one divides the time interval into N small intervals

We here defined the path integral measure by

Here we defined £0 = £a and £^ = ££.

The partition function Z(/3) in statistical mechanics at temperature T is given

where /3 = l/(kT). This partition function is written by using the completeness

Here we used the relation (n|£&){£a n) = (£a\n)(n - £&} which is confirmed by

where we expanded e ^H in powers of J3H and we used (a^o) 2 = of a valid

2.5 Path integral for Dirac particles

The 4 x 4 Dirac matrices 7^, fj, = 0, 1, 2, 3, are defined by

Their explicit forms arc given in terms of 2 x 2 Pauli matrices

and it is given in the present explicit representation of 7 matrices as

The variables ip(t,x)a, a = I ~ 4, stand for a four-component spinor and these