Non-Perturbative Field Theory

https://doi.org/10.
1017/9781009401654 Published online by Cambridge University Press

https://doi.org/10.1017/9781009401654 Published online by Cambridge University Press
NON-PERTURBATIVE FIELD THEORY
From Two-Dimensional Conformal Field Theory
to QCD in Four Dimensions
Providing a new perspective on quantum field theory, this book gives a ped-
agogical and up-to-date exposition of non-perturbative methods in relativistic
quantum field theory and introduces the reader to modern research work in
theoretical physics.
It describes in detail non-perturbative methods in quantum field theory,
and explores two-dimensional and four-dimensional gauge dynamics using those
methods. The book concludes with a summary emphasizing the interplay
between two- and four-dimensional gauge theories.
Aimed at graduate students and researchers, this book covers topics from two-
dimensional conformal symmetry, affine Lie algebras, solitons, integrable models,
bosonization and ’t Hooft model, to four-dimensional conformal invariance, inte-
grability, large N expansion, Skyrme model, monopoles and instantons. Applica-
tions, first to simple field theories and gauge dynamics in two dimensions, and
then to gauge theories in four dimensions and quantum chromodynamics (QCD)
in particular, are thoroughly described. This title, first published in 2010 , has
been reissued as an Open Access publication on Cambridge Core.
Yi t z h a k F r i s h m a n is a Professor Emeritus at the Weizmann Institute,

Israel, where he has served as Head of the Einstein Centre for Theoretical Physics
and Head of the Department of Particle Physics.
Ja c o b S o n n e n s c h e i n is a Professor of Physics at Tel Aviv University,
Israel, where he was Head of the Particle Physics Department from 2003 to
2007.

CAMBRIDGE MONOGRAPHS ON MATHEMATICAL PHYSICS
General Editors: P. V. Landshoff, D. R. Nelson, S. Weinberg
S. J. Aarseth Gravitational N-Body Simulations: Tools and Algorithms

J. Amb jørn, B. Durhuus and T. Jonsson Quantum Geometry: A Statistical Field Theory
Approach
A. M . Anile Relativistic Fluids and Magneto-fluids: With Applications in Astrophysics and
Plasma Physics
J. A. de Azcárraga and J. M . Izquierdo Lie Groups, Lie Algebras, Cohomology and Some
Applications in Physics †
O . Bab elon, D. Bernard and M . Talon Introduction to Classical Integrable Systems †
F. Bastianelli and P. van Nieuwenhuizen Path Integrals and Anomalies in Curved Space
V. Belinski and E. Verdaguer Gravitational Solitons
J. Bernstein Kinetic Theory in the Expanding Universe
G. F. Bertsch and R. A. Broglia Oscillations in Finite Quantum Systems
N. D. Birrell and P. C. W . Davies Quantum Fields in Curved Space †
K. Bolejko, A. Krasiński, C. Hellaby and M -N. Célérier Structures in the Universe by Exact
Methods: Formation, Evolution, Interactions
D. M . Brink Semi-Classical Methods for Nucleus-Nucleus Scattering †
M . Burgess Classical Covariant Fields
E. A. Calzetta and B.-L. B. Hu Nonequilibrium Quantum Field Theory
S. Carlip Quantum Gravity in 2+1 Dimensions †
P. Cartier and C. DeW itt-M orette Functional Integration: Action and Symmetries
J. C. Collins Renormalization: An Introduction to Renormalization, the Renormalization Group
and the Operator-Product Expansion †
P. D. B. Collins An Introduction to Regge Theory and High Energy Physics †
M . Creutz Quarks, Gluons and Lattices †
P. D. D’Eath Supersymmetric Quantum Cosmology
F. de Felice and C. J. S Clarke Relativity on Curved Manifolds
B. DeW itt Supermanifolds, 2 nd edition †
P. G. O Freund Introduction to Supersymmetry †
Y. Frishm an and J. Sonnenschein Non-Perturbative Field Theory: From Two Dimensional
Conformal Field Theory to QCD in Four Dimensions
J. A. Fuchs Affine Lie Algebras and Quantum Groups: An Introduction, with Applications in
Conformal Field Theory †
J. Fuchs and C. Schweigert Symmetries, Lie Algebras and Representations: A Graduate Course
for Physicists †
Y. Fujii and K. M aeda The Scalar-Tensor Theory of Gravitation
J. A. H. Futterm an, F. A. Handler, R. A. M atzner Scattering from Black Holes †
A. S. Galp erin, E. A. Ivanov, V. I. O rievetsky and E. S. Sokatchev Harmonic Superspace
R. Gambini and J. Pullin Loops, Knots, Gauge Theories and Quantum Gravity †
T. Gannon Moonshine beyond the Monster: The Bridge Connecting Algebra, Modular Forms and
Physics
M . Göckeler and T. Sch ücker Differential Geometry, Gauge Theories and Gravity †
C. Góm ez, M . Ruiz-Altaba and G. Sierra Quantum Groups in Two-Dimensional Physics
M . B. Green, J. H. Schwarz and E. W itten Superstring Theory Volume 1: Introduction †
M . B. Green, J. H. Schwarz and E. W itten Superstring Theory Volume 2: Loop Amplitudes,
Anomalies and Phenomenology †
V. N. Grib ov The Theory of Complex Angular Momenta: Gribov Lectures on Theoretical Physics
J. B. Griffi ths and J. Po dolský Exact Space-Times in Einstein’s General Relativity
S. W . Hawking and G. F. R. Ellis The Large Scale Structure of Space-Time †
F. Iachello and A. Arim a The Interacting Boson Model
F. Iachello and P. van Isacker The Interacting Boson-Fermion Model
C. Itzykson and J. M . Drouffe Statistical Field Theory Volume 1: From Brownian Motion to
Renormalization and Lattice Gauge Theory †
C. Itzykson and J. M . Drouffe Statistical Field Theory Volume 2: Strong Coupling, Monte Carlo
Methods, Conformal Field Theory and Random Systems †
C. V. Johnson D-Branes †
P. S. Joshi Gravitational Collapse and Spacetime Singularities
J. I. Kapusta and C. Gale Finite-Temperature Field Theory: Principles and Applications, 2 nd
edition
V. E. Korepin, N. M . Bogoliub ov and A. G. Izergin Quantum Inverse Scattering Method and
Correlation Functions †
M . Le Bellac Thermal Field Theory †
Y. M akeenko Methods of Contemporary Gauge Theory
N. M anton and P. Sutcliffe Topological Solitons †
N. H. M arch Liquid Metals: Concepts and Theory

I. M ontvay and G. M ünster Quantum Fields on a Lattice †
L. O’Raifeartaigh Group Structure of Gauge Theories †
T. O rtı́n Gravity and Strings †
A. M . O zorio de Alm eida Hamiltonian Systems: Chaos and Quantization †
L. Parker and D. J. Tom s Quantum Field Theory in Curved Spacetime: Quantized Fields and
Gravity
R. Penrose and W . Rindler Spinors and Space-Time Volume 1: Two-Spinor Calculus and
Relativistic Fields †
R. Penrose and W . Rindler Spinors and Space-Time Volume 2: Spinor and Twistor Methods in
Space-Time Geometry †
S. Pokorski Gauge Field Theories, 2 nd edition †
J. Polchinski String Theory Volume 1: An Introduction to the Bosonic String
J. Polchinski String Theory Volume 2: Superstring Theory and Beyond
V. N. Pop ov Functional Integrals and Collective Excitations †
R. J. Rivers Path Integral Methods in Quantum Field Theory †
R. G. Rob erts The Structure of the Proton: Deep Inelastic Scattering †
C. Rovelli Quantum Gravity †
W . C. Saslaw Gravitational Physics of Stellar and Galactic Systems †
M . Shifm an and A. Yung Supersymmetric Solitons
H. Stephani, D. Kram er, M . M acCallum , C. Ho enselaers and E. Herlt Exact Solutions of
Einstein’s Field Equations, 2 nd edition
J. Stewart Advanced General Relativity †
T. Thiem ann Modern Canonical Quantum General Relativity
D. J. Tom s The Schwinger Action Principle and Effective Action
A. Vilenkin and E. P. S. Shellard Cosmic Strings and Other Topological Defects †
R. S. Ward and R. O . Wells, Jr Twistor Geometry and Field Theory †
J. R. W ilson and G. J. M athews Relativistic Numerical Hydrodynamics
†
Issued as a pap erback.

Non-Perturbative Field Theory
From Two-Dimensional Conformal Field Theory
to QCD in Four Dimensions
YITZHAK FRISHMAN
The Weizmann Institute of Science
JACOB SONNENSCHEIN
Tel Aviv University

Shaftesbury Road, Cambridge CB2 8EA, United Kingdom
One Liberty Plaza, 20th Floor, New York, NY 10006, USA
477 Williamstown Road, Port Melbourne, VIC 3207, Australia
314–321, 3rd Floor, Plot 3, Splendor Forum, Jasola District Centre, New Delhi – 110025, India
103 Penang Road, #05-06/07, Visioncrest Commercial, Singapore 238467
Cambridge University Press is part of Cambridge University Press & Assessment,

a department of the University of Cambridge.
We share the University’s mission to contribute to society through the pursuit of
education, learning and research at the highest international levels of excellence.
www.cambridge.org
Information on this title: www.cambridge.org/9781009401647
DOI: 10.1017/9781009401654
© Yitzhak Frishman and Jacob Sonnenschein 2010
This work is in copyright. It is subject to statutory exceptions and to the provisions
of relevant licensing agreements; with the exception of the Creative Commons version the
link for which is provided below, no reproduction of any part of this work may take
place without the written permission of Cambridge University Press.
An online version of this work is published at doi.org/10.1017/9781009401654 under a
Creative Commons Open Access license CC-BY-NC-ND 4.0 which permits re-use,
distribution and reproduction in any medium for non-commercial purposes providing
appropriate credit to the original work is given. You may not distribute derivative works
without permission. To view a copy of this license, visit
https://creativecommons.org/licenses/by-nc-nd/4.0
All versions of this work may contain content reproduced under license from third parties.
Permission to reproduce this third-party content must be obtained from these third-parties directly.
When citing this work, please include a reference to the DOI 10.1017/9781009401654
First published 2010
Reissued as OA 2023
A catalogue record for this publication is available from the British Library.
ISBN 978-1-009-40164-7 Hardback
ISBN 978-1-009-40161-6 Paperback
Cambridge University Press & Assessment has no responsibility for the persistence or accuracy of
URLs for external or third-party internet websites referred to in this publication
and does not guarantee that any content on such websites is, or will remain,
accurate or appropriate.

To my wife Yehudith,
mother Faiga
and daughter Einat
Yitzhak Frishman
To my mother Hilda,
wife Nava
and children Nir, Ori and Tal
Jacob Sonnenschein

Contents
Preface page xv
Acknowledgements xviii
PART I NON-PERTURBATIVE METHODS IN

TWO-DIMENSIONAL FIELD THEORY
1 From massless free scalar field to conformal field
theories 3
1.1 Complex geometry 3
1.2 Free massless scalar field 4
1.3 Symmetries of the classical action 5
1.4 Mode expansion 6
1.5 Noether currents and charges 7
1.6 Canonical quantization 7
1.7 Radial quantization 9
1.8 Operator product expansion 11
1.9 Path integral quantization 12
1.10 Affine current algebra 13
1.11 Virasoro algebra 14
2 Conformal field theory 17
2.1 Conformal symmetry in two dimensions 17
2.2 Primary fields 18
2.3 Conformal properties of the energy-momentum tensor 20
2.4 Virasoro algebra for CFT 21
2.5 Descendant operators 22
2.6 Hilbert space of states 23
2.7 Unitary CFT and Kac determinant 25
2.8 Characters 28
2.9 Correlators and the conformal Ward identity 29
2.10 Crossing symmetry, duality and bootstrap 31
2.11 Verlinde’s formula 33
2.12 Free Majorana fermions – an example of a CFT 34
2.13 The Ising model – the m = 3 unitary minimal model 37

x Contents
3 Theories invariant under affine current algebras 39

3.1 Simple finite-dimensional Lie algebras 39
3.3 Current OPEs and the Sugawara construction 49
3.5 ALA characters 52
3.6 Correlators, null vectors and the Knizhnik–Zamolodchikov
equation 53
3.7 Free fermion realization 55
3.8 Free Dirac fermions and the U (N ) 58
4 Wess–Zumino–Witten model and coset models 61
4.1 From free massless scalar theory to the WZW model 61
4.2 Perturbative conformal invariance 65
4.3 ALA, Sugawara construction and the Virasoro algebra 66
4.4 Correlation functions of primary fields 67
4.5 WZW models with boundaries – D branes 71
4.6 G/H coset models 73
4.7 G/G coset models 75
5 Solitons and two-dimensional integrable models 79
5.1 Introduction 79
5.2 From the theory of a massive free scalar field to
integrable models 79
5.3 Classical solitons 81
5.4 Breathers or “doublets” 86
5.5 Quantum solitons 88
5.6 Integrability and factorized S-matrix 92
5.7 Yang–Baxter equations 94
5.8 The general solution of the S-matrix 95
5.9 From conformal field theories to integrable models 99
5.10 Conserved charges and classical integrability 101
5.11 Multilocal conserved charges 104
5.12 Quantum integrable charges in the O(N ) model 107
5.13 Non-local charges and quantum groups 108
5.14 Integrable spin chain models and the algebraic
Bethe ansatz 111
5.15 The continuum thermodynamic Bethe ansatz 125
6 Bosonization 131
6.1 Abelian bosonization 132
6.2 Duality between the Thirring model and the sine-Gordon model 136
6.3 Witten’s non-abelian bosonization 139
6.4 Chiral bosons 148

Contents xi
6.5 Bosonization of systems of operators of high conformal

dimension 159
7 The large N limit of two-dimensional models 165
7.1 Introduction 165
7.2 The Gross–Neveu model 166
7.3 The CP N −1 model 171
PART II TWO-DIMENSIONAL NON-PERTURBATIVE

GAUGE DYNAMICS
8 Gauge theories in two dimensions – basics 177
8.1 Pure Maxwell theory 177
8.2 QED2 – Schwinger’s model 178
8.3 Yang–Mills theory 179
8.4 Quantum chromodynamics 180
9 Bosonized gauge theories 183
9.1 QED2 – The massive Schwinger model 183
9.2 Abelian bosonization of flavored QCD2 185
9.3 Non-abelian bosonization of QCD2 187
10 The ’t Hooft solution of 2d QCD 191
10.1 Scattering of mesons 198
10.2 Higher 1/N corrections 201
11 Mesonic spectrum from current algebra 203
11.2 Universality of conformal field theories coupled to YM 2 203
11.3 Mesonic spectra of two-current states 206
11.4 The adjoint vacuum and its one-current state 216
12 DLCQ and the spectra of QCD with fundamental and
adjoint fermions 223
12.1 Discretized light-cone quantization 223
12.2 Application of DLCQ to QCD2 with fundamental fermions 224
12.3 The spectrum of QCD2 with adjoint fermions 228
13 The baryonic spectrum of multiflavor QCD2 in the strong
coupling limit 237
13.1 The strong coupling limit 237
13.2 Classical soliton solutions 239
13.3 Semi-classical quantization and the baryons 240
13.4 The baryonic spectrum 247
13.5 Quark flavor content of the baryons 247
13.6 Multibaryons 249

xii Contents
13.7 States, wave functions and binding energies 250

13.8 Meson-baryon scattering 252
14 Confinement versus screening 265
14.1 The string tension of the massive Schwinger model 265
14.2 The Schwinger model in bosonic form 268
14.3 Beyond the small mass abelian string tension 268
14.4 Correction to the leading long distance abelian potential 269
14.5 Finite temperature 271
14.6 Two-dimensional QCD 272
14.7 Symmetric and antisymmetric representations 276
15 QCD2 , coset models and BRST quantization 279
15.2 The action 279
15.3 Two-dimensional Yang–Mills theory 282
15.4 Schwinger model revisited 283
15.5 Back to the YM theory 285
15.6 An alternative formulation 287
15.7 The resolution of the puzzle 288
15.8 On bosonized QCD2 289
15.9 Summary and discussion 290
16 Generalized Yang–Mills theory on a Riemann surface 291
16.2 The partition function of the YM 2 theory 292
16.3 The partition function of gYM 2 theories 296
16.4 Loop averages in the generalized case 297
16.5 Stringy YM 2 theory 299
16.6 Toward the stringy generalized YM 2 301
16.7 Examples 302
16.8 Summary 304
PART III FROM TWO TO FOUR DIMENSIONS

17 Conformal invariance in four-dimensional field theories
and in QCD 309
17.1 Conformal symmetry algebra in four dimensions 310
17.2 Conformal invariance of fields, Noether currents and
conservation laws 312
17.3 Collinear and transverse conformal transformations
of fields 314
17.4 Collinear primary fields and descendants 316
17.5 Conformal operator product expansion 318

Contents xiii
17.6 Conformal Ward identities 319

17.7 Conformal invariance and QCD4 322
18 Integrability in four-dimensional gauge dynamics 329
18.1 Integrability of large N four-dimensional N = 4 SYM 330
18.2 High energy scattering and integrability 333
19 Large N methods in QCD4 337
19.1 Large N QCD in four dimensions 337
19.2 Meson phenomenology 343
19.3 Baryons in the large N expansion 346
19.4 Scattering processes 352
20 From 2d bosonized baryons to 4d Skyrmions 355
20.2 The Skyrme action 355
20.3 The baryon as a Skyrmion 361
20.4 The Skyrme model for Nf = 3 367
21 From two-dimensional solitons to four-dimensional
magnetic monopoles 371
21.2 The Yang–Mills Higgs theory – basics 372
21.3 Topological solitons and magnetic monopoles 373
21.4 The ’t Hooft–Polyakov magnetic monopole solution 376
21.5 Charge quantization 377
21.6 Zero modes, time-dependent solutions and dyons 378
21.7 BPS monopoles and dyons 381
21.8 Montonen Olive duality 382
21.9 Nahm construction of multimonopole solutions 383
21.10 Moduli space of monopoles 386
22 Instantons of QCD 389
22.1 The basic properties of the instanton 389
22.2 The ADHM construction of instantons 394
22.3 On the moduli space of instantons 396
22.4 Instantons and tunneling between the vacua of the YM theory 400
22.5 Instantons, theta vacua and the UA (1) anomaly 403
23 Summary, conclusions and outlook 407
23.1 General 407
23.2 Conformal invariance 408
23.3 Integrability 410
23.4 Bosonization 411
23.5 Topological field configurations 412
23.6 Confinement versus screening 414

xiv Contents
23.7 Hadronic phenomenology of two dimensions versus four

dimensions 416
23.8 Outlook 420
References 423
Index 433

Preface
Field theory is the framework with which one describes the theory of the standard
model of elementary particles and their interactions. The electromagnetic sector
(QED) of the standard model is understood extremely well using perturbation
theory, but the color interaction (QCD) which is responsible for hadron physics
can only be accounted for perturbatively for a limited set of observational data.
Due to the fact that at long distances the color interaction is strongly coupled,
one cannot reliably apply perturbative methods to extract, for instance, the
spectrum of the hadrons. The arsenal of tools to handle strongly coupled systems
is obviously much more limited than the one used for weakly coupled ones.
Nevertheless, several methods to handle non-perturbative field theories have been
developed. The main goal of this book is to expose the reader to those techniques
and to describe their applications in two-dimensional and four-dimensional field
theories and finally in QCD in four dimensions.
The topic of non-perturbative field theory is by itself very rich and it is clear
that one cannot cover it in a non superficial manner in one book. Thus we had
to make certain decisions about the flow of the book and about the topics that
should be addressed. As for the former issue we have decided to present the
book in three parts. In the first part we describe, in detail, the most impor-
tant non-perturbative techniques of two-dimensional field theory. The reason for
this is obvious since physical systems with one space dimension and one time
dimension are the simplest and hence it is easier to grasp the non-perturbative
tools when applied to these systems. In the second part of the book we study
two-dimensional gauge theories with the emphasis on employing the techniques
developed in the first part. The third part is devoted to the non-perturbative
aspects of gauge dynamics in four dimensions. In this part we elevate the tech-
niques of the first part to four dimensions and we examine to what extent gauge
theories in four dimensions behave like their two-dimensional simplified analogs.
There are several books on the shelves discussing non-perturbative methods in
general such as [66] and [182], there are books describing one particular method,
like conformal field theory in two dimensions for instance [77], there are books
that describe two-dimensional QCD, [2] and books that study various aspects of
four-dimensional QCD, for example [151] and of course there are books on the
basics of field theory, for example [37], [130], [173] and [215]. The aim of this
book is three-fold, to review a package of non-perturbative methods, to present
a picture which is close to the state-of-the-art in the topics described and to

xvi Preface
demonstrate application of the methods in addressing several questions of gauge

dynamics.
The particular methods we explore in Part 1 of the book associate with con-
formal field theory, with affine Lie algebras, with topological properties of fields,
solitons and integrable models, with bosonization and with the large N approx-
imation.
In Part 2 we first present the basics of gauge field theories in two dimensions
and in particular the bosonized version of them, we then describe the seminal
large N solution of ’t Hooft of the mesonic spectrum of two-dimensional QCD;
we address the mesonic spectrum using current algebra methods, we describe the
discrete light-cone quantization of QCD with quarks in the fundamental repre-
sentation and also adjoint quarks, we compute the spectrum of baryons and their
properties in the strong coupling limit, we discuss the issue of confinement versus
screening behavior, we analyze QCD2 using coset model and BRST techniques,
and finally we digress and devote a chapter to generalized Yang–Mills theory on
Riemann surfaces and their stringy nature.
In Part 3 we demonstrate the applications in four-dimensional gauge dynam-
ics of conformal invariance, techniques of integrable models, of large N expan-
sion and of topology. In particular we devote chapters to Skyrmions, magnetic
monopoles and gauge theory instantons.
As we have mentioned above we had to take decisions about what topics
related to non-perturbative field theory we should not include. We decided not
to address string theories, supersymmetric field theories and the holographic
string (gravity)/gauge duality. The main reason for this decision was that to
cover each of these topics requires a book in itself, or even more than one book.
In fact certain subjects that we do cover in the book, like conformal field theory,
magnetic monopoles or instantons would require a full book to cover properly.
What we have tried to achieve is to describe the basic ideas of each topic and to
demonstrate its application. We have also not treated subjects like anomalies, lat-
tice formulations, sigma models, chiral Lagrangians and other non-perturbative
topics.
Some topics described in the book are “fully established topics”, in the sense
that presumably the most important developments in those have been already
achieved, for instance conformal field theory in two dimensions and bosonization
in two dimensions. On the other hand some other topics of the book are under
current intensive investigation and are certainly still not fully established. An
example of the latter is integrability in four-dimensional gauge dynamics. The
reason we have decided to include topics of the latter kind is that we wanted the
book to be fairly up to date and useful to researchers investigating “modern”
topics.
In the more basic issues we have made an effort to present the material in a
pedagogical manner and to be self contained. For instance our discussion started
from a free massless scalar field theory in two dimensions and gradually evolved

Preface xvii
into general conformal field theories. In dealing with more advanced topics, like
for instance instantons in four dimensions, the reader will need to consult with
specialized references to obtain a more complete and wider picture of the topic.
Some of the content of the book, mainly in Part 2, is based on the research
work of the authors, but most of the material is a review of the work of many
researchers in the field.
The book is aimed for advanced Ph.D. students, post-docs and other newcom-
ers to the arena of non-perturbative methods in field theory. The reader should
definitely be equipped with a basic knowledge of field theory, group theory and
algebra, differential equations, geometry and topology.
Throughout the book we refer to only a limited list of references. The number
of scientific contributions to the topics discussed in this book is enormous and
since we could not cover all of them we have referred to papers that initiated the
various topics, and to review papers and books where a much more exhaustive
list of references can be found.
We have made an attempt to keep the same notations throughout the book.
However in certain instances we have changed notations during the course of
the book, mainly to be in accordance with relevant literature. In these cases we
specified explicitly the change in notation made.

Acknowledgements
We would like to thank O. Aharony, A. Armoni, M. Bernstein, S. J. Brodsky,

E. Cohen, R. Dashen, G. Date, G. F. Dell-Antonio, J. R. Ellis, O. Ganor,
D. Gepner, E. Gimon, A. Hanany, G. P. Lepage, M. Karliner, L. A. Pando Zayas,
C. T. Sachrajda, N. Sochen, M. J. Strassler, S. Yankielowicz, W. J. Zakrzewski
and D. Zwanziger who collaborated with us in the research works that are cov-
ered in this book.
We would like to thank Ori Sonnenschein for drawing the figures of the book.
We would like to thank O. Aharony, M. Karliner and S. Theisen for their
remarks on the manuscript.
The work of Jacob Sonnenschein was supported in part by the Albert Einstein
Minerva Center of The Weizmann Institute of Science.

PART I
Non-perturbative methods in two-dimensional

field theory

1
From massless free scalar field to conformal
field theories
In this chapter we analyze the simplest field theory, which is the theory of a
free massless scalar field in two space-time dimensions, one space and one time.1
The rich symmetry and algebraic structure of this theory encapsulates the basic
concepts of two-dimensional conformal field theory, which will be the topic of
the next chapter.
1.1 Complex geometry

It is convenient for the discussion of two-dimensional free scalar theory and later
conformal field theories to introduce complex coordinates as follows:2
ξ = x0 + ix1 ξ¯ = x0 − ix1 . (1.1)
We now take x0 and x1 to be in Euclidean space. Correspondingly we define the
derivatives
1 1
∂ξ = (∂0 − i∂1 ) ∂ξ¯ = (∂0 + i∂1 ), (1.2)
2 2
which is a special case of the decomposition to components of vectors, namely
1
Aξ = 12 (A0 − iA1 ) Aξ¯ = (A0 + iA0 )
2
¯
Aξ = (A0 + iA1 ) Aξ = (A0 − iA1 ). (1.3)
2 2
The metric of the flat Euclidean space-time ds2 = dx0 + dx1 translates into
¯ namely
ds2 = dξdξ,
1 ¯ ¯ ¯¯
gξ ξ¯ = gξ¯ξ = , g ξ ξ = g ξ ξ = 2, gξ ξ = gξ¯ξ¯ = g ξ ξ = g ξ ξ = 0. (1.4)
2
With this metric at hand the scalar product of two vectors takes the form
¯ 1 ¯ ¯
Aμ Bμ = Aξ Bξ + Aξ Bξ¯ = (Aξ B ξ + Aξ B ξ ). (1.5)
2
Complex components of higher-order tensors relate in a similar manner to
the real components, in particular for a symmetric two-tensor (like the
1 The content of this chapter comprises the basics of massless scalar fields in two dimensions.
This is covered in many textbooks.
2 The use of complex coordinates in the context of the bosonic string theory is described by
Polyakov in [177].

4 From massless free scalar field to conformal field theories
x0
x1
Fig. 1.1. The map between ξ and z.
energy-momentum tensor),
1
T ≡ Tξ ξ = (T00 − 2iT10 − T11 )
4
1
T̄ ≡ Tξ¯ξ¯ = (T00 + 2iT10 − T11 )
4
1
Tξ¯ξ = Tξ ξ¯ = (T00 + T11 ). (1.6)
4
Often, especially in the context of string theory, the space direction is no longer
R, but rather is compactified on S 1 so that x1 ≡ x1 + 2π. For such a geometry
it is convenient to introduce the following conformal map:
0
+ix 1
ξ → z = e ξ = ex ,
which maps the cylinder to the complex plane (see Fig. 1.1).
In particular the past x0 = −∞ is mapped into the origin and the future
x = ∞ into a circle with an infinite radius. It is clear that the relations between
0
¯ and (x0 , x1 ) derived above hold also between (z, z̄) and (Real(z), Im(z)).
(ξ, ξ)
The holomorphic and anti-holomorphic derivatives with respect to z will be
denoted by ∂ ≡ ∂z and ∂¯ ≡ ∂z̄ .
1.2 Free massless scalar field

The action S of the free massless scalar field X̂(z, z̄) is

1
2
S = d xL = d2 x∂ν X̂ ∂¯ν X̂
8π
1 1
= d2 ξ∂ξ X̂∂ξ¯X̂ = d2 z∂ X̂ ∂¯X̂, (1.7)
4π 4π

1.3 Symmetries of the classical action 5
where L is the Lagrangian density. The factor 4π 1

is used to match the normal-

ization of the bosonic string theory (with α = 2). In the complex coordinate
¯ and (z, z̄) the measure of the integral is d2 ξ = (i/2)dξ ∧ dξ¯ and
notation (ξ, ξ)
d2 z = (i/2)dz ∧ dz̄, respectively. Note that L is a local expression and thus is
the same for the Euclidean plane or for any compact two-surface.
Varying the scalar field X̂(z, z̄) → X̂(z, z̄) + δ X̂(z, z̄) induces a variation in
the action of the form

1
δS = − d2 z(∂ ∂¯X̂)δ X̂. (1.8)
2π
The action is thus extremized by configurations that solve the corresponding
equation of motion
∂ ∂¯X̂ = 0. (1.9)
It is thus clear that ∂ X̂ is a holomorphic function and ∂¯X̂ is an anti-holomorphic
function, and the most general solution takes the form
X̂(z, z̄) = [X(z) + X̄(z̄)]. (1.10)
1.3 Symmetries of the classical action

By construction the action is invariant under translations and SO(2) rotations.
Translations in x0 and x1 translate in complex coordinates to
z → z + a; z̄ → z̄ + ā, (1.11)
where a is a constant complex number, and the SO(2) rotations, in infinitesimal
form, to
δz = −iz; δz̄ = iz̄, (1.12)
where is an infinitesimal real parameter.
When we go back to Minkowski space, the SO(2) rotations turn into SO(1, 1)
transformations. In addition it is easy to realize that a shift of the field by a
constant A,
X̂(z, z̄) → X̂(z, z̄) + A, (1.13)
leaves the Lagrangian invariant. It is a special feature of two dimensions that
the symmetry group of the action is in fact much richer since one can replace
the constant A with A(z) and the constant Ā with Ā(z̄), which are arbitrary
holomorphic and anti-holomorphic functions, respectively,
X̂(z, z̄) → X̂(z, z̄) + A(z); X̂(z, z̄) → X̂(z, z̄) + Ā(z̄). (1.14)
These are the affine current algebra transformations.3
3 Affine Lie algebras describing a physical system were first discussed in [27]. More references
will be given in the next two chapters.

In a similar manner the space-time translations (1.11) can also be elevated to

holomorphic and anti-holomorphic transformations,
z → f (z); z̄ → f¯(z̄), (1.15)
referred to as two-dimensional conformal transformations. Affine current alge-

bra transformations and conformal transformations will be further discussed in
Sections 1.10 and 1.11
1.4 Mode expansion

The mode expansion of the classical solution depends on the boundary condi-
tions. For the case where the underlying two-dimensional manifold is the infinite
plane, a standard Fourier transform is used:

dk 1
0 1
X̂(x , x ) = √ √ [a(k 1 )e−ik ·x + a† (k 1 )eik ·x ]. (1.16)
2π k 0
If the range of the space coordinate is bounded, one may impose two types of
boundary conditions, associated with closed and open strings. In the case of
closed strings the boundary conditions
X̂(x0 , x1 ) = X̂(x0 , x1 + 2π) (1.17)
are automatically obeyed by X̂(z, z̄). For this case the mode expansion is
expressed in terms of a Laurent series,
∞
∞

αn ᾱn
∂X = −i ∂¯X̄ = −i . (1.18)
n =−∞
zn + 1 n =−∞
z̄ n +1
Integrating this expansion we get

∞
α ᾱm −m
z −m +
m
X̂(z, z̄) = X − iP ln(z z̄) + i z̄ , (1.19)
m m
m =−∞, m = 0
with X a constant and
P = α0 = ᾱ0 . (1.20)
For open strings the boundary conditions are of Neumann type, namely
∂1 X̂(x0 , x1 = 0) = ∂1 X̂(x0 , x1 = π) = 0 =⇒ ∂ X̂(z, z̄ = z) = ∂¯X̂(z, z̄ = z).

(1.21)
The corresponding mode expansion takes the form

∞
αm −m
X̂(z, z̄) = X − iP ln(z z̄) + i z + z̄ −m . (1.22)
m
m =−∞, m = 0

1.6 Canonical quantization 7
1.5 Noether currents and charges

Associated with the symmetries (1.14) and (1.15) are conserved Noether currents
and charges. In the Noether procedure one is instructed to elevate the global
parameters of transformations into local ones and extract the associated currents
from the variation of the action, namely δS ∼ d2 xJμ ∂ μ . Let us apply this
procedure first to the affine current algebra transformations so that we vary the
action with respect to δ X̂(z, z̄) = (z, z̄) yielding

1
δS = d2 z[∂(z, z̄)∂¯X̂(z, z̄) + ∂(z,
¯ z̄)∂ X̂(z, z̄)]. (1.23)
4π
Unlike the situation in more than two dimensions, and due to the fact that the
symmetries (1.14) are in fact not only global ones but rather “half local”, the
currents
J ≡ ∂X; J¯ ≡ ∂¯X̄ (1.24)
are holomorphic and anti-holomorphic conserved,
¯ ≡ ∂∂X
∂J ¯ = 0; ∂ J¯ ≡ ∂ ∂¯X̄ = 0. (1.25)
The classical currents are determined up to an overall constant.

A similar situation occurs with respect to the conformal transformation.
Replacing in the infinitesimal version of (1.15) δz → (z, z̄) and δz̄ → ¯(z, z̄) one
finds,

1
δS = (z, z̄)∂¯X̂(z, z̄)∂¯X̂(z, z̄) + ∂(z,
d2 z[∂¯ ¯ z̄)∂ X̂(z, z̄)∂ X̂(z, z̄)]. (1.26)
2π
The associated holomorphic and anti-holomorphic conserved energy-

momentum tensor components are
1 1
T = − ∂X∂X; T̄ = − ∂¯X̄ ∂¯X̄, (1.27)
2 2
where the coefficients were chosen in a way that will turn out to be convenient
when discussing the corresponding quantum generators.
1.6 Canonical quantization

Prior to imposing the canonical quantization condition one has to identify the
time direction. There are several options. Using x0 as the time direction, the

corresponding conjugate momentum of X̂(z, z̄) is

δL 1
Π= = ∂0 X̂,
δx0 X̂ 4π
and the standard quantization conditions are
[X̂(x0 , x1 ), Π(y 0 , y 1 )]x 0 =y 0 = iδ(x1 − y 1 )
[X̂(x0 , x1 ), X̂(y 0 , y 1 )]x 0 =y 0 = 0
[Π(x0 , x1 ), Π(y 0 , y 1 )]x 0 =y 0 = 0. (1.28)
These conditions yield the standard algebra of the creation and annihilation
operators for (1.16),
[a(k 1 ), a† (p1 )] = δ(k 1 − p1 ); [a(k 1 ), a(p1 )] = [a† (k 1 ), a† (p1 )] = 0. (1.29)
Substituting the mode expansion (1.16) into the expressions of the Noether
charges associated with the symmetries of the action (1.7) one finds that
the energy-momentum operators are proportional to a† (k)a(k) + a(k)a† (k) and
hence their vacuum expectation values are proportional to δ(0) ∼ L, where L is
the size of the space direction. It is thus clear that for the infinite Euclidean plane
(or a Minkowski space-time with space R) these expectation values diverge. One
then defines the normal ordered operators:
: O : ≡ O − <0|O|0> . (1.30)
For free fields this is equivalent to ordering annihilation operators to the right
of creation operators, and sufficient to make : O : finite.
Using the algebra of the creation and annihilation operators and the normal
ordered Hamiltonian, the construction of the Fock space is standard. One defines
the vacuum state |0> such that
a(k 1 )|0> = 0. (1.31)
The states in the Fock space are
a† (ki )n i |0>, (1.32)

i
and their energies, by applying the Hamiltonian,

H| a† (ki )n i |0> = (kj0 )ni (ki ) a† (ki )n i |0> . (1.33)
i
The canonical quantization for the scalar field on a compact space direction,
with the boundary conditions of open or closed string, (1.21) and (1.17), respec-
tively, follows very similar steps. Imposing the quantization conditions (1.28)
above implies the following algebra for the αn operators of the open string and

1.7 Radial quantization 9
for the αn and ᾱn operators for the closed string:

[αm , αn ] = mδm +n
[ᾱm , ᾱn ] = mδm +n
[αm , ᾱn ] = 0. (1.34)
It is thus clear that αn operators are related to the a(k) operators as
√ √
αm = ma(m), m > 0; α−m = ma† (m), m > 0. (1.35)
1.7 Radial quantization

For the case of a cylinder-like two-dimensional manifold, namely, where the space
0 1
direction is compactified so that x1 ≡ x1 + 2π, it is natural to use the z = ex +ix
coordinates. Space translations x1 → x1 + a take the form of multiplying by
a phase factor z → eia z, and time translations x0 → x0 + a turn into dilata-
tions z → ea z. Rotations (x0 + ix1 ) → (c + is)(x0 + ix1 ), go into z → z (c+is) ,
with (c + is) = eiθ , θ the rotation angle. Correspondingly the generators of these
transformations change their geometrical operation. For instance the Hamilto-
nian obviously goes into the dilatation generator. Moreover, generators which
are Noether charges transform into contour integrals. Recall that the Noether
charge is Q = dx1 J0 (x1 ) which in the new coordinates reads Q = dθJr (θ) so
that we can write,
1 ¯
Q= [dzJ(z) + dz̄ J(z̄)], (1.36)
2πi
where the contour integral is performed at some radius and the sign convention
we adopt is that both the dz and dz̄ integral are taken to be positive for the
counter-clockwise sense.
The infinitesimal transformation of an operator generated by the Noether
charge Q is given by:
1 ¯ (z̄), O(w, w̄)].
δ,¯ O = [dzJ(z)(z), O(w, w̄)] + dz̄[J(z̄)¯ (1.37)
2πi
Define a product R of two operators A(z)B(w) as taken radially, namely4
R(A(z)B(w)) = A(z)B(w), |z| > |w|; B(w)A(z), |w| > |z|. (1.38)
In Fig. 1.2 we show the two contour integrals that lead to a contour integral
around w, the location of the operator O, so that the infinitesimal transformation
is given by,
1
δ,¯ O = [dz(z)R(J(z)O(w, w̄)) + dz̄¯(z̄)R(J(z̄)O(w, w̄))]. (1.39)
2πi
4 The notion of radial quantization was introduced in [104]. This construction was used in the
context of complex geometry in [93].

z z
w
w w
Fig. 1.2. A contour around w from the commutator.
We now apply this formulation to the symmetry generators (discussed in Sec-

tion 2.1):
(i) The infinitesimal affine current algebra transformation X̂(z, z̄) → X̂(z, z̄) −
(z) is generated by the holomorphic current J(z) = ∂X via
1
δ X̂(w, w̄) = dz(z)R(∂X(z)X̂(w, w̄))
2πi
1 −1
= dz (z) = −(w), (1.40)
2πi z−w
where we have used for the product of operators,
R(X(z)X(w)) = − log(z − w) + finite terms. (1.41)
This is an example of the concept of operator product expansion, which is

addressed in the next section.
(ii) In a similar manner we can compute the transformation of ∂X generated by
the energy momentum tensor T
1 1
δ ∂X(w) = dz(z)R − : ∂X(z)∂X(z) : ∂X(w)
2πi 2
1 1
= dz ∂X(z)(z) = ∂(w)∂X(w) + (w)∂ 2 X(w), (1.42)
2πi (z − w)2
which is indeed the infinitesimal transformation of the holomorphic current

J = ∂X(z). The generator T is normal ordered using the following expression:

1 1 1
T (w) = − : ∂X(z)∂X(w) :≡ − limz →w ∂X(z)∂X(w) + . (1.43)
2 2 (z − w)2

1.8 Operator product expansion 11
1.8 Operator product expansion

In computing the contour integrals associated with infinitesimal transformations
we have made use of the operator product expansions of pairs of operators.5 The
singularities that occur when the points are taken to approach one another are
captured in the notion of operator product expansion (OPE),

Oi (x)Oj (y) = ckij (x − y)Ok (y), (1.44)
k
where − y) are the coefficient functions which are singular in the limit
ckij (x
of x → y. Such expansions were proven to hold in renormalizable field theo-
ries. The OPEs are an essential tool in exploring quantum field theories. Recall
that all of the information on the QFT is encoded into the values of all possi-
ble correlation functions of the complete set of local operators Oi (x), namely,
< O1 (x1 )...On (xn ) >. In particular, one is interested in the behavior of these
correlation functions when two or more points approach each other, which is
encapsulated in the OPEs. For all applications discussed here the OPEs are
treated as asymptotic expansions and only their singular terms will be specified.
For the present case of two-dimensional free massless scalar field theory the OPE
converges and in fact, as will be discussed in Section 3.7.2, a similar situation
occurs in all 2d CFTs.
The OPEs of the free massless scalar can be deduced from its propagator,
which can be evaluated from the solution. It takes the form:
< X̂(z z̄)X̂(ww̄) >= −log|z − w|2 . (1.45)
In terms of the separation of the solution into holomorphic and anti-holomorphic
parts the two propagators read:
< X(z)X(w) >= − log(z − w); < X̄(z̄)X̄(w̄) >= − log(z̄ − w̄). (1.46)
By differentiating the last relation with respect to z and to w one finds the short
distance expansion of other operators like J(z), T (z) etc. In particular the OPE
of the currents is
1
J(z)J(w) = ∂X(z)∂X(w) = − + finite terms, (1.47)
(z − w)2
with a similar result for the anti-holomorphic currents.
A different, though equivalent, approach is to write the OPE as a Taylor
expansion in (z − w) and (z̄ − w̄) in the following form:
∞
1
X̂(z, z̄)X̂(ww̄) = −log|z − w|2 + [(z − w)k : (∂ k X̂(w, w̄))X̂(w, w̄) :
k!
k=1
+(z̄ − w̄)k : (∂¯k X̂(w, w̄))X̂(w, w̄) :]. (1.48)
5 Wilson introduced for the first time the concept of an operator product expansion [219]. It
was used for two-dimensional conformal field theories in [33].

This form of expansion is based on the property that the normal ordered product
of the scalar fields,
: X̂(z, z̄)X̂(ww̄) := X̂(z, z̄)X̂(w, w̄) + log|z − w|2 , (1.49)
obeys the equation of motion, namely,
∂ ∂¯ : X̂(z, z̄)X̂(w, w̄) := 0, (1.50)
and hence can be decomposed to holomorphic and anti-holomorphic functions

and thus is non singular.
In the previous subsection we used two OPEs to determine the symmetry
transformation of X and ∂X. We will work out now two additional examples
of OPEs, involving the operator which will later be found to be very useful
: eiα X (w ) :.
(i) The conformal properties of : eiα X (w ) are being determined by its OPE with
T (z) which takes the form
1
T (z) : eiα X (w ) : = − (: ∂X(z)∂X(z) :)(: eiα X (w ) :)
2 2
( α2 ) iα X (w ) 1
= e + ∂eiα X (w ) . (1.51)
(z − w) 2 (z − w)
In language that will be developed in Section 2.2 this result will mean that
2
: eiα X (w ) : has a conformal dimension of α2 .
(ii) The OPE of two operators of the form : eiα X (w ) is
: eiα X (w ) e−iβ X (w ) :
(: eiα X (z ) :)(: e−iβ X (w ) :) = . (1.52)
(z − w)α β
1.9 Path integral quantization

So far we have been using canonical quantization. Before proceeding to the gen-
eral structure of affine current algebra and Virasoro algebra we introduce the
quantization of a free massless scalar field using the Euclidean path integral
approach. As usual the functional integration DX̂(z, z̄) can be approximated
by discretizing the two-dimensional space and representing the functional inte-
gral by products of ordinary integrals. Expectation values of operators O(X)
constructed from X are given by,

¯
< O(X̂) >= DX̂(z, z̄)O(X̂)e−S = DX̂(z, z̄)O(X̂)e− 2 π d z ∂ X̂ ∂ X̂ . (1.53)
1 2
Correlation functions have to obey the equation
∂ ∂¯ < X̂(z, z̄)X̂(w, w̄) >= −2πδ 2 (z − w, z̄ − w̄), (1.54)

as can be deduced by using the fact that the path integral of a total derivative
vanishes:

δ
0 = DX̂(z, z̄) [e−S X̂(w, w̄)]
δ X̂(z,
z̄)

−S δS
= DX̂(z, z̄)e − X̂(w, w̄) + δ (z − w, z̄ − w̄)
2
δ X̂(z, z̄)
1
= < ∂ ∂¯X̂(z, z̄)X̂(w, w̄) > + < δ 2 (z − w, z̄ − w̄) > . (1.55)
2π
Alternatively one can use (1.45) and (1.46) directly. Note that in that case care
must be exercised, as naively we would get zero rather than the delta function,
since the expression is a sum of two terms, one depending on z only and the
other on z̄ only. The point is that the expressions (1.46) cannot be taken over at
the origin. A working rule is:
1 1
∂¯ =∂ = (2π)δ 2 (z, z̄)
z z̄
1 z̄
This can be derived by going over from z to z z̄ + 2 , to regulate the singularity
at the origin.
1.10 Affine current algebra

As was shown in Section 1.3 the classical action is invariant under both affine
current algebra transformations and conformal transformations. We would like
to study the algebraic structure of the generators of these symmetries. We start
with the invariance under affine current algebra transformations. Recall that the
corresponding generators are the holomorphic and anti-holomorphic currents J
¯ given in (1.24). Expanding the currents in Laurant series,
and J,
∞
∞

J= Jn z −(n + 1) ; J¯ = J¯n z̄ −(n +1) , (1.56)
n =−∞ n =−∞
it is obvious from the mode expansion (1.18) that the algebra of the currents is
related to that of the αn operators, namely
[Jm , Jn ] = −mδm +n ; [J¯m , J¯n ] = −mδm +n . (1.57)
This form of algebra will be shown in the next chapter to be associated with level
k = 1 abelian affine current algebra (or Kac–Moody algebra as it is sometimes
referred to).
This algebra translates into the following algebra of the currents:
[J(z), J(w)] = δ (z − w). (1.58)

Using the technique developed in Section 1.7 we can derive this result also from
the operator product expansion of two currents,
1
J(z)J(w) = + finite terms. (1.59)
(z − w)2
1.11 Virasoro algebra

Next we address the algebraic structure of the generators of conformal transfor-
mations (1.27). Upon inserting (1.18) into the Laurent expansion of the energy
momentum tensor,
∞
∞
T = Ln z −(n + 2) ; T̄ = L̄n z̄ −(n +2) , (1.60)
n =−∞ n =−∞
one finds that for Ln with n = 0,

∞

Ln = 1/2 : αn −m αm : . (1.61)
m =−∞
For n = 0 the operators αn −m and αm commute, and so the product equals

the normal ordered one. The situation is different for L0 . Here one encoun-
ters an infinity in the product of chiral fields, which normal ordering removes,
resulting in,
∞
L0 = 1/2P 2 + α−m αm . (1.62)
1
We shall later see that it is sometimes necessary to shift L0 by a constant. Using

the commutation relation of αn one finds the following “naive” expression for
the commutator of Ln operators:
1
[Lm , Ln ] = [αm −k αk , αn −l αl ]
4
k ,l
1 1
= kαm −k αk +n + (m − k)αm −k +n αk
2 2
k k
= (m − n)Lm +n , (1.63)
where to get to the third line we have changed a variable in the first sum from
k → k − n. This is the classical Virasoro algebra,6 and in fact in the quantum
theory it is further corrected. The correction appears only for the case m + n = 0,
so for m = −n the classical form (1.63) is exact. For generators with m + n = 0
the two sums in the second line of (1.63) have to be brought to normal order. As
re-ordering means using the commutator, one gets divergent series for which, in
the case at hand, one cannot shift the variable of summation without changing
6 The Virasoro algebra was presented in [212]. More references will be given in the next two
chapters.

1.11 Virasoro algebra 15
the result. Taking this into account, one gets a c-number shift in the commutation
rule,
[Lm , Ln ] = (m − n)Lm +n + A(m)δm +n . (1.64)
To compute the anomaly term A(m) we introduce a cutoff function fΛ (k),

which tends to 1 in the limit of infinite regulator Λ for any k, but for every finite
Λ goes to zero sufficiently rapidly at infinite k. Thus we view the operators Ln
as regularized sums,
∞

Ln = 1/2 : αn −m αm : fΛ (m), (1.65)
m =−∞
to replace (1.61). With this regularized expression, a direct computation gives

for the anomaly,
∞

A(m) = 1/4 {k(m − k)fΛ (m − k)[fΛ (k − m) + fΛ (−k)]
k=1
+ k(m + k)fΛ (−k)[fΛ (k) + fΛ (−m − k)]}. (1.66)
If we now take fΛ (k) to 1, without being careful, we get the divergent sum,
∞

A(m) → m k. (1.67)
k=1
Using ζ-function regularization, namely replacing k by k −s , we get a convergent

sum for any s > 1, and then we continue analytically to s = −1, to get −m/12
for the right-hand side of the last equation.
To compute A(m) with the regulators fΛ , we now look at,
∞
m
A(m) + = 1/4 {k(m − k)fΛ (m − k)[fΛ (k − m) + fΛ (−k)]
12
k=1
+ k(m + k)fΛ (−k)[fΛ (k) + fΛ (−m − k)] − 4mk}. (1.68)
Only large k is relevant now, as for any finite k we can take Λ to infinity first,
obtaining zero on the right-hand side. We now take,
−p
fΛ (q) ≈ |q| |q| → ∞, (1.69)
with p → 0 as Λ → ∞. Expanding in powers of mk , and recalling that ζ(s) has a

pole only at s = 1, we get by summing first and then letting p → 0, the result,
m m3
A(m) + = . (1.70)
12 12
The anomaly term A(m) can also be determined using the Jacobi identity
[Lk , [Lm , Ln ]] + [Lm , [Ln , Lk ] + [Ln , [Lk , Lm ] = 0. One finds that for k + m +
n = 0 the anomaly term obeys (m − n)A(k) + (n − k)A(m) + (k − n)A(m) = 0.
Recall also that A(0) = 0 and A(m) = −A(−m) so it is enough to determine
A(m) for positive m. The relation derived from the Jacobi identity can be used

to get a recursion relation which is determined by values of A(1) and A(2).

In fact the general solution is of the form A(n) = b3 n3 + b1 n. The coefficient
b1 is correlated with the normal ordering ambiguity constant of L0 . One can
determine the coefficients b1 and b3 by computing the vacuum expectation val-
ues of <0|[L1 , L−1 ]|0> = 0 and <0|[L2 , L−2 ]|0> = 12 , so that altogether one finds
A(n) = 12 1
n(n2 − 1) and the full Virasoro algebra associated with the massless
free scalar field is,
1
[Lm , Ln ] = (m − n)Lm +n + m(m2 − 1)δm +n . (1.71)
12
In the next chapter the Virasoro algebra will be discussed in a broader perspec-
tive. In that context it will become clear that the algebra of (1.71) associated
with the massless free scalar is characterized by a c = 1 Virasoro anomaly.

2
Conformal field theory
Conformal invariance of two-dimensional massless scalar field theory was shown

in the previous chapter to associate with the infinite algebra of conserved charges,
the Virasoro algebra. In this chapter we describe the basic building blocks of
any two-dimensional conformal field theory (CFT). The notions of primary and
descendant operators will be introduced and the structure of the Hilbert space
of states will be described. We will discuss and classify certain classes of unitary
CFTs. Crossing symmetry, duality and bootstrap equations will be defined and
applied to computing correlators of CFTs. We then discuss the Verlinde formula
which relates the fusion rules and the S transformation. We will end up with two
examples of CFTs that demonstrate all of the concepts that have been introduced
before. The first one is the theory of a Majorana fermion and the second is the
m = 3 unitary minimal model, which is shown to be the continuum limit of the
two-dimensional Ising model.
Conformal field theory in two dimensions is covered by many review articles
and books. The former include [109] which we use intensively in this chapter,
also [25], [13], [59], [233] and many others.
Among the books that discuss 2d CFT is [140] and books on string theories
[113], [154], [174], [138], [237], [142], [30].
The most complete book on the topic is [77].
The basics of conformal field theory were stated in the seminal paper by
Belavin, Polyakov and Zamolodchikov [33]. This includes the introduction of
primary fields, the behavior of the energy-momentum tensor and the central
charge. Conformal Ward identity and the use of OPEs appears in [93], [95]
and [94].
2.1 Conformal symmetry in two dimensions

The theory of the free massless scalar field in two dimensions was shown to be
invariant under the holomorphic and anti-holomorphic coordinate transforma-
tions
z → z = f (z); z̄ → z̄ = f¯(z̄). (2.1)
Under such a transformation the metric transforms as
∂z ∂ z̄
ds2 = dzdz̄ → dz dz̄ = dzdz̄. (2.2)
∂z ∂ z̄

18 Conformal field theory
At this point we can understand why we referred to these transformations as

conformal transformations. In general in d space-time dimensions the confor-
mal group is the subgroup of coordinate transformations that leaves the metric
invariant up to a scale, namely,

gμν (x) → gμν (x ) = Ω(x)gμν (x). (2.3)
It is obvious from (2.2) that the 2d conformal transformations (2.1) indeed pro-
duce such a variation of the metric. An important property of conformal trans-
formations in any dimension is that they preserve the angle √AA·2BB 2 between two
vectors A and B.
Starting from flat space, the general infinitesimal coordinate transforma-
tions xμ → xμ + μ (x) induces a change of the metric ds2 → ds2 + (∂μ ν +
∂ν μ )dxμ dxν , so that the condition for conformal transformations reads,
2
∂μ ν + ∂ν μ = (∂ · )gμν , (2.4)
d
where gμν is ημν or δμν for a Minkowskian signature, or Euclidean signature,
respectively.
It is thus obvious that for two-dimensional Euclidean space-time = (z) and
¯ = ¯(z̄) are the unique solutions of (2.4), which reduces to the Cauchy–Riemann
equation ∂1 1 = ∂2 2 and ∂1 2 = −∂2 1 .
We would like now to put aside scalar field theory and explore the general prop-
erties of conformal field theories in two dimensions. Any theory with a vanishing
trace of the energy-momentum tensor Tμμ = 0, or in complex coordinates Tz z̄ = 0,
has necessarily an independent holomorphically (and anti-holomorphically) con-
served energy-momentum tensor components, namely,
¯ ≡ ∂T
∂T ¯ zz = 0 ∂ T̄ ≡ ∂Tz̄ z̄ = 0. (2.5)
This follows trivially from the usual conservation law ∂T¯ z z + ∂Tz z̄ = 0, and its
complex conjugation. It is also clear that in fact there are infinitely many con-
served currents, since g(z)T (z) for any analytic function g(z) is also a holomor-
phically conserved current (we sometimes call any conserved tensor “current”).
We show in the following section that indeed the energy-momentum tensor
T (z) and T̄ (z̄) generate the conformal transformations given in (2.1).
2.2 Primary fields

Conformal invariance constrains the OPEs of the theory. In particular, since T
is holomorphic, the OPE of T (z) with a general operator can be expanded in
terms of a Laurent expansion in integer powers of z. The singular part of the
OPE takes the form,
∞
1
T (z)Õ(w, w̄) = Õ(n ) (w, w̄), (2.6)
n=0
(z − ω)n +1

where the sum is usually finite, and the operators Õ(n ) (w, w̄) have to be deter-
mined. Using radial quantization as in Section 1.7 and the OPE above, we get
for the transformation generated by T (z),
1
δ Õ(w, w̄) = (∂ n )Õ(n ) (w, w̄) . (2.7)
n
n!
We now consider operators that transform under conformal transformation in a

way that generalizes the transformation of the metric, (2.2),
∂z
h
∂ z̄
h̄
O(z, z̄) → O (z z̄ ) = O(z z̄ ). (2.8)
∂z ∂ z̄
An operator with such conformal transformations is a primary field or a ten-
sor operator with conformal weights (h, h̄), which are sometimes referred to as
the holomorphic and anti-holomorphic conformal dimensions.1 The sum of the
weights h + h̄ is the total dimension that determines the behavior under scal-
ing, whereas h − h̄ is the spin that controls the behavior under rotations. The
infinitesimal transformations that correspond to (2.8) are,

¯ + ¯∂)
δ,¯ O(z, z̄) = (h∂ + ∂) + (h̄∂¯ ¯ O(z z̄). (2.9)
This form of transformation implies that the singular part of the OPE of T and
O(w, w̄) reduces to,
h 1
T (z)O(w, w̄) = O(w, w̄) + ∂O(w, w̄). (2.10)
(z − ω)2 (z − ω)
Applying these notions to the free scalar field we find that ∂X(z) has (1, 0)
2 2
weights, ∂¯X̄(z̄) has (0, 1) and the weights of : eiα X (z , z̄ ) : are ( α2 , α2 ).
In Chapter 1 the notion of OPE was discussed in the context of scalar field the-
ory. The generalization to any CFT is straightforward. Normalize the operators
with fixed conformal weights as,
1 1
Oi (z, z̄)Oj (w, w̄) = δij , (2.11)
(z − w) (z̄ − w̄)2 h̄ i
2h i
then, for a complete set, the OPE of any pair of such operators is, to leading
singularity,

Oi (z, z̄)Oj (w, w̄) ∼ Cij k (z − w)h k −h i −h j (z̄ − w̄)h̄ k −h̄ i −h̄ j Ok (w, w̄), (2.12)
k
where Cij k are the product coefficients of the theory.
1 The notion of conformal primary field and its descendants was introduced in [33] and further
discussed in [236].

2.3 Conformal properties of the energy-momentum tensor

For the free massless scalar field we found that the OPE of T (z)T (w) is not
of the form shown as (2.6), due to the anomaly term as in (1.71). The form of
T (z)T (w) OPE for any CFT is rather,
c/2 2T (w) ∂w T (w)
T (z)T (w) = + + , (2.13)
(z − w)4 (z − w)2 (z − w)
where c is the central charge (or the Virasoro anomaly), a constant that charac-
terizes the theory. The second term represents the dimensions and the third the
property of translations under T . For theories with positive semi-definite Hilbert
space c ≥ 0, as follows from,
c/2
< T (z)T (w) >= .
(z − w)4
This type of OPE implies the following infinitesimal transformation of T :
c 3
δ(z ) T (z) = ∂ (z) + 2(∂(z))T (z) + (z)∂T (z). (2.14)
12
The corresponding finite transformation T (z) → T (z ) takes the form,
c
T (z ) = (∂z )2 T (z) + {z , z}, (2.15)
12
where {z , z} is the Schwarzian derivative,
2∂ 3 f ∂f − 3∂ 2 f ∂ 2 f
{f, z} = . (2.16)
2∂f ∂f
To derive (2.16), we first note that by applying a second transformation f → ω
we get,
{w, z} = (∂z f )2 {w, f } + {f, z}. (2.17)
Then, we take ω = f + δf , thus obtaining a functional equation,
δ ∂ 3 δf
δf {f, z} = (∂z f )2 3 . (2.18)
δf ∂ f
Expressing the right-hand side as derivatives with respect to z,

1 3f 3(f )2 f
(δf ) − 2 (δf ) + 3 − 2 (δf ) ,
f (f ) (f ) (f )
we can integrate the equation to get (2.16). The first term suggests integrating to

f /f , the variation of which gives 1/f (δf ) − f /(f )2 (δf ) , while the second

term suggests −3(f )2 /2(f )2 , the variation of which gives −3f /(f )2 (δf ) +

3(f )2 /(f )3 (δf ) .
For the massless scalar case T can be written as T (z) = − 12 : J(z)J(z) :, as
we saw in (1.5). In fact, as will be discussed in Chapter 3, there is a large class
of theories that share this so-called Sugawara form. For this type of theory the
proof that the finite transformation is of the form of (2.15) is as follows. Recall

2.4 Virasoro algebra for CFT 21

that as a primary field of weights (1, 0), J(z) → ∂∂zz J(z ). If we write T (z) =
− 12 limz →w (J(z)J(w) + (z −w
1
)2 ) and substitute the transformation of the currents
we end up after some lengthy but straightforward calculation with (2.15).
2.4 Virasoro algebra for CFT

Let us use the Laurent expansion of T for CFT, following (1.60),
∞ ∞
L̄n
T = Ln z −(n + 2) T̄ = −(n +2)
, (2.19)
n =−∞ n =−∞
z̄
so that,
1
Ln = dzz n + 1 T (z). (2.20)
2πi
The expansion is chosen such that Ln has scale dimension n under z → z/a,
namely, Ln → an Ln .
The Virasoro algebra2 can now be derived using the OPE of T (z)T (w) given
in (2.13),
2
1
[Ln , Lm ] = dz dw[z n + 1 wm + 1 − z m +1 wn +1 ]T (z)T (w). (2.21)
2πi
The double
integral is performed by fixing w and transforming the difference of
the two dz integrals into one integral around w,
1
2

[Ln , Lm ] = dz dw z n + 1 wm +1 − z m +1 wn +1
2πi

c/2 2T (w) ∂w T (w)
+ +
(z − w)4 (z − w)2 (z − w)
1
= dw c/12(n3 − n)wn +m −1
2πi

+ [2(n + 1) − (n + m + 2)] wn +m +1 T (w)
c 3
= (n − n)δ(n + m) + (n − m)Ln +m . (2.22)
12
Performing identical steps for L̄n we get that L̄n obeys the same infinite algebra,
with some central charge c̄, and that [Ln , L̄m ] = 0.
Any CFT is a representation of the Virasoro algebra characterized by c and c̄.
It is straightforward to identify the following properties of the algebra:
r The generators (L±1 , L0 ) span an SL(2, R) algebra,
[L+ 1 , L−1 ] = 2L0 [L0 , L±1 ] = ∓L± (2.23)
2 The first use of the Virasoro algebra was by M. Virasoro in the context of the dual resonance
model [212]. Its application to two-dimensional CFT was presented in [33].

Table 2.1. The conformal family
Level Weight Fields
0 h φ
1 h+1 L−1 φ
2 h+2 L−2 φ, L2−1 φ
3 h+3 L−3 φ, L−2 L−1 φ, L3−1 φ
.. .. ..
. . .
N h+N P(N) fields
r For n > 0, L−n is a raising operator and Ln is a lowering one, since [L0 , Ln ] =
−nLn so that if |ψ> is an eigenstate of L0 , L0 |ψ> = h|ψ> then L0 |Ln ψ> =
(h − n)|Ln ψ>.
2.5 Descendant operators

From every primary operator φ(z, z̄) one can construct an infinite tower of Vira-
soro descendant operators,
1 1
(L−n φ(w, w̄)) = dz n −1 T (z)φ(w, w̄). (2.24)
2πi z
A distinguished descendant operator is the energy momentum tensor T (z) since,
1 dz
L−2 1 = T (z)1 = T (0). (2.25)
2πi z
The set containing the primary field φ(z, z̄) and all its descendant operators
is called a conformal family and it is denoted by [φ]. A conformal family is a
tower of operators where each layer is characterized by its level as shown in
Table 2.1, where P (N ) is the number of partitions of N into positive integer
1
parts, which can be written in terms of the generating function n =1 (1−q n) =
∞ N
N =0 P (N )q .
We can now use the conformal family to rewrite the expression of the OPE
(2.12) of two primary fields,
φi (z, z̄)φj (w, w̄)
{l ¯l} ¯l n ¯
= Cij k (z − w)h k −h i −h j + n ln
(z̄ − w̄)h̄ k −h̄ i −h̄ j + n φlkl (w, w̄),
k {l ¯l}
(2.26)
¯
where we denote by φlkl (w, w̄) the descendants L−l 1 . . . L−l n L̄−¯l 1 . . . L̄−¯l n φk (w, w̄)
{l ¯l}
with the normalization given in (2.11). The product coefficients Cij k are given
in terms of those of (2.12) Cij k as,
{l ¯l} k {l} k {¯l}
Cij k = Cij k βij β̄ij , (2.27)

2.6 Hilbert space of states 23
k {l}
where βij are determined by conformal invariance and are functions of c and
k {¯l}
hi , hj , hk , and similarly for β̄ij . This follows from a detailed analysis that we
do not show here.
The OPEs of any pair of descendant fields can also be deduced from (2.12)
which implies in fact that all the information about the OPE is encoded in the
product coefficients Cij k . Moreover since the structure of (2.26) holds for all the
primaries and their descendants, one can write the so-called fusion algebra for
conformal, families, which takes the form,

[φi ][φj ] = Nijk [φk ]. (2.28)
k
2.6 Hilbert space of states

Our next task is to construct the Hilbert space of states. First we define the
ground state |0> by,
Ln |0> = 0 n ≥ 0. (2.29)
The next step in this program is to build the highest weight states (hws).
Consider the state generated from the vacuum by a primary field φ(z) of
dimension h,
|h> = φ(0)|0> . (2.30)
It is easy to check that for n > 0, [Ln , φ(0)] = 0 since,
1
[Ln , φ(w)] = dzz n + 1 T (z)φ(w) = h(n + 1)wn φ(w) + wn +1 ∂φ(w). (2.31)
2πi
Hence the highest weight state |h> obeys
L0 |h> = h|h>
Ln |h> = 0 n > 0. (2.32)

Expanding the primary field φ(z) in a Laurent series n φn z (n −h) , one can write
the highest weight state symbolically as φh |0>.
Descendant states are generated by applying the descendant operators L−n φ
on the vacuum or alternatively by applying L−n on highest weight states, namely,
L−n |h> = L−n φ(0)|0> = (L−n φ)|0> . (2.33)
It is thus clear that the highest weight states, or equivalently the primary oper-
ators, play a major role in constructing representations of the Virasoro algebra.
In fact one can show that every representation is characterized by a primary
operator. Consider an eigenstate of L0 , L0 |ψ> = hψ |ψ>. Now act on it with the
lowering operator Ln with n > 0. The L0 eigenvalue of the new state Ln |ψ> is
hψ − n. Since we require that the Hamiltonian is bounded from below, L0 has
to be also bounded. This implies that after repeating the lowering process one
finally hits a state that is annihilated by Ln for every n > 0 and hence an hws.

It is thus clear that any state in a positive Hilbert space is a linear combination
of hws, and their descendants. The representation given in Table 2.1 is referred
to as the Verma module. Denoting it by V(c, h) and its analogous representation
for the anti-holomorphic Virasoro algebra by V̄(c̄, h̄), the Hilbert space of the
theory is a direct sum of the products V(c, h) ⊗ V̄(c̄, h̄), namely,

H= V(c, h) ⊗ V̄(c̄, h̄). (2.34)
h, h̄
The Verma module may be reducible in the sense that there is a submodule that
is by itself a Verma module. Such a submodule whose states transform amongst
themselves under any conformal transformation, is built from a |hnull>. The
latter is both an hws., namely Ln |hnull> = 0 for n > 0, as well as a descendant.
Such a state is called null state or null vector, motivated by what follows. It
generates its own Verma module which is included in the parent module. It is
orthogonal to the whole Verma module as well as to itself <hnull |hnull> = 0,
since <hnull |L−k 1 . . . L−k n |h> = <h|Lk n . . . Lk 1 |hnull>∗ = 0, and in particular it
has a zero norm <hnull |hnull> = 0 and similarly also its descendants. The null
state corresponds to a null operator which is simultaneously a primary and a
secondary field.
Let us now demonstrate the construction of a null vector. Consider a general
linear combination of the states of level 2,
L−2 |h> + aL2−1 |h>, (2.35)
we would like to check whether for certain values of the mixing coefficient a, this
state is a null state. If indeed it is |null>, then so is the state [Ln |null>]. In fact
it is easy to verify that at level 2, one has to check these consistency conditions
only for L1 and L2 . Now using the Virasoro algebra we find that,

[L1 , L−2 ] |h> + a L1 , L2−1 |h> = (3 + 2a(2h + 1))L−1 |h>,
c
[L2 , L−2 ]|h> + a L2 , L2−1 |h> = 4h + + 6ah |h> . (2.36)
2
It is thus clear that for the following values of a and c,
3 2h
a=− c= (5 − 8h), (2.37)
2(2h + 1) 2h + 1
the linear combination state (2.35) is a null state. In the unitary case we have
h and c positive (see next section). Hence in this example h < 58 .
An irreducible representation of the Virasoro algebra can be constructed from
a Verma module that contains a null vector by a quotient procedure, taking
out of the Verma module the null module. In the next section we discuss this
construction.

2.7 Unitary CFT and Kac determinant

Unitarity is obviously lost if there are negative norm states in the Verma module.
Hence, our task is to derive the conditions for having a negative norm state. In
the basis of the Verma module,
L−k 1 . . . L−k i |h> ≡ |s> (1 ≤ k1 ≤ . . . ≤ ki ), (2.38)
the matrix of inner products Iss = <s|s> is block diagonal with blocks I(N ) for

states at level N ( i ki = N ). For a given Verma module the elements of I are
functions of (h, c). It is easy to realize that unitarity dictates c > 0 and h > 0.
This follows from <h|Ln L−n |h> = [2nh + 1/12cn(n2 − 1)] <h|h>, which is pos-
itive for n = 1 only if h > 0 and for large enough n only for c > 0. To determine
the full set of constraints for unitarity let us analyze further the properties of I.

A general state |ŝ> = k ck |s> has a norm <ŝ|ŝ> = ĉ† Iĉ, with ĉ the vector of
the ck . Now since I is hermitian it can be diagonalized by a unitary matrix U

so that the norm can be written as <ŝ|ŝ> = k lk |tk |2 where t = U ĉ and lk are
the eigenvalues of I, which are real. It is thus clear that there are negative norm
states if and only if I has negative eigenvalues. A vanishing eigenvalue indicates
a null vector which means a reducible Verma module.
For the low lying levels these matrices take the following form:
I(0) = 1
I(1) = 2h
4h(2h + 1) 6h
I(2) = . (2.39)
6h 4h + c/2
The derivation of the various elements is straightforwad, for instance,
(2)
I11 = <h|L1 L1 L−1 L−1 |h> = <h|L1 L−1 L1 L−1 |h> + 2 <h|L1 L0 L−1 |h>
= 4 <h|L1 L−1 L0 |h> + 2 <h|L1 L−1 |h> = 8h2 + 4h. (2.40)
The determinant of I(2) is given by
det[I(2) ] = 32(h − h1,1 )(h − h1,2 )(h − h2,1 ), (2.41)

where h1,1 = 0 and h1,2 , h2,1 are (1/16)[(5 − c) ± (1 − c)(25 − c)]. The trace
of I(2) is Tr[I(2) ] = 8h(h + 1) + c/2. Since the trace and the determinant are the
sum and product of the two eigenvalues, unitarity is lost if either the trace or
the determinant is negative.
The determinant for I(N ) at general level N , which is referred to as the Kac
determinant,3 has the form
P (N −pq )
det[I(N ) ] = αN [h − hp,q (c)] , (2.42)
pq ≤N
3 The proof of the Kac determinant is detailed in [89], [206] and [95].

where αN are constants independent of (c, h) and hp,q (c) can be expressed in
terms of m = − 21 ± 12 25−c
1−c as,
[(m + 1)p − mq]2 − 1

hp,q (c) = . (2.43)
4m(m + 1)
Note that we can choose either the plus or the minus sign in the expression for
m, as their interchange is like interchanging p with q, which does not change the
determinant. Note also that hp,q is invariant under p → m − p, q → m + 1 − q.
Let us also mention that for N = 2 the result is identical to (2.41).
In the (h, c) plane the determinant vanishes along the curves h = hp,q (c) which
are therefore called the vanishing curves. If the determinant (2.42) is negative it
means that there is an odd number of negative eigenvalues and hence the corre-
sponding Virasoro representation is not unitary. If the determinant is vanishing
or positive one needs to further analyze the determinant as follows:
r For c > 1 and h > 0 it is straightforward to show that the determinant does
not vanish.
In the domain 1 < c < 25 the value for m has an imaginary part. Thus hp,q are
complex for p = q, and as they come in complex conjugate pairs the product of
the appropriate two factors in the determinant is positive. For p = q the value
of hp,q is negative. Thus the determinant is positive in that domain.
For c > 25 the hp,q are negative.
For large h the matrix is dominated by its diagonal elements.
Since these elements are positive, the eigenvalues for large h are all pos-
itive. Now since the determinant never vanishes in the region considered
(h > 0, c > 1) all the eigenvalues have to be positive on the entire region.
Note that in I(2) the off-diagonal element is larger at large h than the 22
element, but still the determinant is dominated at large h by the diagonal
elements, and thus also the eigenvalues, as a 2 × 2 matrix.
r For c = 1 we have hp,q = (p − q)2 /4, and so the determinant is never negative.
However, it vanishes when h = n2 /4 for some integer n.
r For 0 < c < 1, h > 0 a closer look at the determinant is required. We draw
hp,q (c) in Fig. 2.1.
By expanding the curves around c = 1 one can show that any point in the
region can be connected to the right of c = 1 by crossing a single vanishing
curve. The vanishing of the determinant is due to one eigenvalue that reverses
its sign which implies that there are negative norm states at any point in the
region that are not on the vanishing curve. In fact it turns out that there are
additional negative norm states at points along the vanishing curve except at

h h4,2
h3,1
h1,3
h2,4
1
2
h3,2
h2,1
1
4
h1,2
h2,3
1
16 h3,3
h2,2
h1,1
1 7 1 c
0 1
2 10 5
Fig. 2.1. hp , q (c) as a function of c for various values of (p, q).
certain points where they

intersect. On these points the central charge c is a
solution of m = − 2 + 2 25−c
1 1
1−c for the cases of m an integer from 3 up,
6
c=1− , m = 3, 4, . . . (2.44)
m(m + 1)
For each such unitary minimal model 4 there are m(m − 1)/2 primary fields
with h given by (2.43) where p, q are integers satisfying 1 ≤ p ≤ m − 1, 1 ≤
q ≤ p. The simplest of those models is the Ising model, given in the section
m = 3, c = 1/2 with h1,1 = 0, h2,1 = 1/2, h2,2 = 1/16. It will be described in
Section 2.13.
4 The minimal models were presented in [33] and discussed in [95].

2.8 Characters
The structure of the Verma module, and in particular the degeneracy of states
at each level, is captured in the generating function χ(c,h) (τ ), the character of
the Verma module, defined by,
∞

χ(c,h) (τ ) = Tr[q L 0 − 2 4 ] = dim(h + n)q h+n − 2 4 ,
c c
(2.45)
n =0
where q ≡ e 2πiτ
, τ is a complex number, and dim(n + h) is the number of linearly
independent states of the module at level n. The latter is equal to P (n) the
partitions of n in the generic case, but may be smaller when there are null
states. For |q| < 1, namely, τ in the upper half plane, the series is uniformly
convergent, since |q| < 1 is the domain of convergence of the inverse of the Euler
function ϕ(q) defined by,
∞ ∞
1 1
= = P (n)q n . (2.46)
ϕ(q) n = 1 1 − q n n =0
In terms of this function the character of a generic Verma module is given by,
c
q h− 2 4
χ(c,h) (τ ) = . (2.47)
ϕ(q)
The character can be expressed also in terms of the Dedekind η(τ ) function,
∞
1 1
η(τ ) ≡ q 2 4 ϕ(q) = q 2 4 (1 − q n ), (2.48)
n =1
in the form
1 −c
q h+ 2 4
χ(c,h) (τ ) = . (2.49)
η(τ )
To get the character of a minimal model one has to determine the irreducible
Verma module using the quotient procedure discussed in the previous section.
We do not give the derivation here, just the final result, which is,
c
q h− 2 4 ( 2 p p n + p r −p s ) 2 ( 2 p p n + p r + p s ) 2

χ(c(p, p ), hr s (p, p )) = = q 4 p p
−q 4 p p
, (2.50)
ϕ(q)
n ∈Z
where
(p − p )2
c(p, p ) = 1 − 6 , (2.51)
pp
and
(pr − p s)2 − (p − p )2
hr s (p, p ) = . (2.52)
4pp
Note that these are the non-unitary minimal models, except for the cases p − p =
±1, which coincide with the cases of the previous section with the identification
of p = m or p = m.

2.9 Correlators and the conformal Ward identity 29
2.9 Correlators and the conformal Ward identity

Now that the Hilbert space of states has been analyzed we would like to determine
the correlation functions of all possible operators of a given CFT. Naturally,
we first investigate correlators of primary fields and then those also involving
descendents.
A very useful tool for determining correlators are the symmetries of the system.
In the present case we obviously implement conformal invariance. In particular
we first determine the consequences of the SL(2, C) Ward identities. Recall that
the vacuum is annihilated by L0,±1 and L̄0,±1 , and hence is invariant under
SL(2, C), namely, U |0> = |0> for U ∈ SL(2, C). It thus follows that,
<0|U −1 φ1 (z1 , z̄1 )U . . . U −1 φn (zn , z̄n )U |0> = <0|φ1 (z1 , z̄1 ) . . . φ − n(zn , z̄n )|0> .
(2.53)
Recall that by definition a primary field of dimension h transforms under an
SL(2, C) transformation z → f (z) = az cz +d (with ad − bc = 1), as,
+b
U −1 φ(z, z̄)U = (∂f (z))h φ(f (z), z̄). (2.54)
Let us mention that SL(2, C) invariance holds for CFT in any dimension.
The invariance of the vacuum implies, in infinitesimal form,
<0|[Lk , φ1 (z1 , z̄1 )] . . . φn (zn , z̄n )|0> + . . . <0|φ1 (z1 , z̄1 ) . . . [Lk , φn (zn , z̄n )]|0> = 0,
(2.55)
for k = 0, ±1. Using [Lk , φ(z, z̄)] = h(k + 1)z k φ(z, z̄) + z k +1 ∂φ(z, z̄) we get Ward
identities in terms of differential equations:

k = −1 : ∂i <0|φ1 (z1 , z̄1 ) . . . φn (zn , z̄n )|0> = 0
i

k= 0: (zi ∂i + hi ) <0|φ1 (z1 , z̄1 ) . . . φn (zn , z̄n )|0> = 0
i

k = +1 : (zi2 ∂i + 2zi hi ) <0|φ1 (z1 , z̄1 ) . . . φn (zn , z̄n )|0> = 0. (2.56)
i
These Ward identities are associated with the invariance under translations,
dilations and special conformal transformations. Applying these equations to the
two point function one finds that,
c2
G2 (z1 , z̄1 , z2 , z̄2 ) ≡ <0|φ1 (z1 , z̄1 )φ1 (z2 , z̄2 )|0> = ,
(z1 − z2 )2h 1 (z̄1 − z̄2 )2 h̄ 1
(2.57)
where c2 is a constant, to be put to 1 in the normalization (2.11). Note also
that when taking two different fields φ1 and φ2 , SL(2, C) implies that h1 = h2
is necessary for a non-zero two-point function.
In a similar manner the three-point function is given by,

1
G3 (zi , z̄i ) = c123 h 1 +h 2 −h 3 h 1 +h 3 −h 2 h 2 +h 3 −h 1
z → z̄, h → h̄ , (2.58)
z12 z13 z23

w1 w1
w2 w2
w4 w4
w3 w3
Fig. 2.2. Integration along C that bounds all the operators.
where zij = zi − zj and c123 is the correpsonding product coefficient defined in

(2.12). Using the SL(2C) invariance one can set the points z1 , z2 , z3 at ∞, 1, 0,
respectively so that the constant c123 is determined from the corresponding corre-
lator via limz 1 , z̄ 1 →∞ z12h z̄12 h̄ G3 = c123 . For Gn with n > 3 the global conformal
transformations do not fully determine the correlator. For instance the four-point
function G4 can be written using these transformations as,
⎡⎛ ⎞ ⎤
−(h i +h j )+h/3 ⎠
G4 (zi , z̄i ) = f (Z, Z̄) ⎣⎝ zij z → z̄, h → h̄ ⎦ , (2.59)
i< j
4
where h = i= 1 hi and the cross ratio Z is defined as Z = zz 11 23 zz 32 44 , which is an
SL(2, C) invariant.
For a general n-point function, denoting
the power of zij by −hij , we get,
2
hij = n −2 (hi + hj ) − (n −1)(n
2
−2) h for n ≥ 3.
So far we have implemented the global Ward identities. To get the local Ward
identity one performs a conformal transformation of an n-point function of pri-
mary fields Gn . This is achieved by integrating (z)T (z) along a contour C which
bounds a region that includes all the operators (see Fig. 2.2)
Now using analyticity one can deform the contour into a sum of countours
each of which encircles one operator. The result of the integral is therefore,
% &
dz
(z)T (z)φ(w1 , w̄1 ) . . . φ(wn , w̄n )
2πi
n % &
dz
= φ(w1 , w̄1 ) . . . (z)T (z)φ(wi , w̄i ) . . . φ(wn , w̄n )
i= 1
2πi

n
= φ(w1 , w̄1 ) . . . δ φ(wi , w̄i ) . . . φ(wn , w̄n ) . (2.60)
i= 1

2.10 Crossing symmetry, duality and bootstrap 31
Using (2.9) we substitute now for δ φ(wi , w̄i ) = (wi )∂ + h∂(wi )φ(wi , w̄i ).
Since this holds for arbitrary we can get a local form of the Ward identity,
n
hi 1 ∂
T (z)φ(w1 , w̄1 ) . . . φ(wn , w̄n ) = +
i= 1
(z − wi )2 (z − wi ) ∂w i
φ(w1 , w̄1 ) . . . φ(wn , w̄n ) , (2.61)
similar to the transition from (2.9) to (2.10). It is thus clear that the correlation
function above is a meromorphic function of z with singularities at the positions
of the operators.
A useful tool for computing correlators is the use of null vectors. Rather than
discussing this for a general null vector we demonstrate this procedure on a level
two null vector. Recall that in models with a primary of weight h such that c =
2h+ 1 (5 − 8h) there is a null vector at level two of the form (L−2 + aL−1 )Φ
2h 2 (h)
=0
where a = − 2(2h+1) . As L−1 φ (z) = ∂φ (z) one can trade L−2 φ (z) with
3 (h) (h) (h)
−a∂ 2 φ(h) (z). Now L−2 φ(h) (w) is given by,

hφ(h) (w) ∂w φ(h) (w)
L−2 φ(h) (w) = limz →w T (z)φ(h) (w) − − . (2.62)
(z − w)2 (z − w)
Substituting this into (2.61) one finds the following differential equation,
−a∂w2 1 φ(w1 , w̄1 ) . . . φ(wn , w̄n )
n
hi 1 ∂
= + φ(w1 , w̄1 ) . . . φ(wn , w̄n ) . (2.63)
(w1 − wi )2 (w1 − wi ) ∂w i
i= 1
This exact differential equation will enable us to compute the four-point function
for the Ising model as we discuss in Section (2.13).
Next we would like to deduce the implications of the associativity on correla-
tion functions of primaries and descendant operators.
2.10 Crossing symmetry, duality and bootstrap

The complete package of information that specifies a CFT is its Virasoro anomaly
c, the set of primary fields φi (z, z̄), with their weights (hi , h̄i ) and the operator
product coefficients Cij k . Hence, to determine all consistent CFTs one has to find
all the allowed sets of such data. The latter have to comply with the constraints
that follow from confomal symmetry as well as with the associativity of the
operator algebra. To study the implications of associativity it is useful to consider
the four-point function,
φi (w1 , w̄1 )φj (w2 , w̄2 )φk (w3 , w̄3 )φl (w4 , w̄4 ) . (2.64)
The idea is to compare the computation of this correlator using the OPE of φi
and φj and of φk and φl , with those of φi and φk and of φj and φl , namely
calculation where (z1 → z2 ), (z3 → z4 ) versus one in which (z1 → z3 ), (z2 → z4 ).

i l
i l
Σp Clmp Cijp p
= Σq Cilq Cjmq q
j m
j m
Fig. 2.3. Crossing symmetry.
0 1
i l
−
Fijlm (p|x) F ijlm (p|x−) = p
j m
x ∞
Fig. 2.4. Single channel amplitude.
The requirement that the two ways of computing coincide, referred to as crossing
symmetry, is expressed in Fig. 2.3.5
Using conformal transformations, we can relate the diagram on the left-hand
side of Fig. 2.3 to the diagram drawn in Fig. 2.4, which corresponds to the
sum of the contributions of intermediate states belonging to the conformal fam-
ily [φp ] with the four-point function of operators located at (w1 , w2 , w3 , w4 ) =
(0, z, 1, ∞). Note that in such a situation, z is actually also the cross ratio Z. We
denote this amplitude by the conformal block Fijk l (m|z)F̄ijk l (m|z̄) which depends
on the Virasoro anomaly of the theory and the dimensions of all the operators
involved. In terms of conformal blocks the crossing symmetry condition takes
the form,

Cij m Ck lm Fijk l (m|z)F̄ijk l (m|z̄)
m
jl jl
= Cij n Ck ln Fik (n|1 − z)F̄ik (n|1 − z̄). (2.65)
n
For a given set of conformal blocks (2.65) is a set of equations that determine the
Cij k and the weights. The general set of solutions of these equations is not known,
but for a particular class of theories like the minimal models these equations can
be solved.
5 Crossing symmetry, duality and bootstrap was discussed in [33].

2.11 Verlinde’s formula 33
2.11 Verlinde’s formula

The fusion rules (2.28), namely,

[φi ][φj ] = Nijk [φk ],
k
constitute a commutative associative algebra. The commutativity implies that

Nijk = Njki and the associativity means that,

Nijk Nkml = m
Nik Njkl . (2.66)
k k
Using matrix notation in which Nijk

= (Ni )kj the associativity translates into
the commutativity of the matrices, namely Ni Nl = Nl Ni . Thus, the matrices
Ni are also members of an associative commutative algebra. Hence they can
be diagonalized simultaneously to form a one-dimensional representation. This
implies that there is a common matrix S̃,

S̃jl λi δlm (S̃ −1 )km = S̃jl λi (S̃ −1 )kl ,
(l) (l)
Nijk = (2.67)
lm l
(l)
where we denote the eigenvalues of Ni by λi .
If j is the vacuum state j = 0 then
k k
Ni0 = δi , if all the representations labeled by i are irreducible. We now multiply
from the right by S̃kn to get,

S̃0l λi (S̃ −1 )kl S̃kn
k n (l)
Ni0 S̃k =
l
(l) (n )
S̃in = S̃0l λi δln = S̃0n λi , (2.68)
l
(n ) S̃ in
which means that λi = S̃ 0n
and therefore,
S̃jl S̃il (S̃ −1 )kl
Nijk = . (2.69)
l
S̃0l
Now, for the reader who knows about the τ parameter and the characters
(discussed in Section 2.8), we recall that under the S-transformation τ → − τ1
the characters of a given CFT transform as,
1
χj (− ) = Sjk χk (τ ). (2.70)
τ
k
Verlinde’s formula6 states that the matrix S̃ above is identical to the S-

transformation matrix,
S = S̃. (2.71)
This is a remarkable relation.
6 The Verlinde formula was introduced in the seminal paper [210].

2.12 Free Majorana fermions – an example of a CFT

The theory of free massless fermions in two dimensions is an example of a 2d
conformal theory of the utmost importance. In this section we describe this
theory in detail following the steps taken in the general analysis of conformal
field theories. The well-known Dirac action of a massless free fermion in two
Euclidean dimensions is,

1
S= d2 xΨ̄ ∂Ψ. (2.72)
4π
Expressing the Dirac fermion in terms of chiral (or Weyl) fermions, a left ψ and
a right ψ̃, with Ψ ≡ (ψ, ψ̃), and using the fact that in two dimensions one can
take γ 0 = σ 2 and γ 1 = σ 1 , we rewrite the action as,

1
S= d2 z(ψ † ∂ψ
¯ + ψ̃ † ∂ ψ̃). (2.73)
4π
We remind the reader that ∂ ≡ ∂z and ∂¯ ≡ ∂z̄ . The equations of motion are,
¯ = 0 ∂ ψ̃ = 0 → ψ = ψ(z)
∂ψ ψ̃ = ψ̃(z̄). (2.74)
In analogy to the symmetries of the scalar field it is straighforward to real-

ize that the action is invariant under left holomorphic chiral and right anti-
holomorphic transformations,
ψ → ψ = eiα (z ) ψ ψ̃ → ψ̃ = ei α̃ ( z̄ ) ψ̃. (2.75)
The corresponding “affine current algebra” currents, given by,
J = iψ † ψ J¯ = iψ̃ † ψ̃, (2.76)
are holomorphically and anti-holomorphically conserved.

In addition the theory is obviously invariant under conformal transformations
z → f (z), z̄ → f¯(z̄).
Dirac (or Weyl) fermions can be further decomposed into Majorana (or Weyl–
Majorana) fermions as Ψ = √12 (Ψ1 + iΨ2 ) (or ψ = √12 (ψ1 + iψ2 )). Substituting
these, the action reads,
2
1
S= d2 z ¯ i + ψ̃i ∂ ψ̃i .
ψi ∂ψ (2.77)
8π i= 1
From this point on we discuss the theory of single Majorana fermions, namely χ
and χ̃ are a left and a right Weyl–Majorana fermion with the action,

1 ¯ + χ̃∂ χ̃).
= d2 z(χ∂χ (2.78)
8π

2.12 Free Majorana fermions – an example of a CFT 35
The equations of motion are still as in (2.74) so that χ is a holomorphic

function and χ̃ is an anti-holomorphic one (their extensions, as they are real on
the real line).
Before spelling out the conformal structure of the theory we pause for a
moment with the complex coordinate formulation and discuss canonical quanti-
zation in two dimensions with a Minkowski signature. The conjugate momentum
to χ is πχ = ∂∂0Lχ = 12 χ, and as we are dealing with a real field {πχ , χ} has a factor
1
2 multiplying a delta function,which gives,
{χ(x1 , x0 ), χ(y0 , y1 )}|x 0 =y 0 = δ(x1 − y1 ). (2.79)
Combining a pair of two Majorana fermions (each consisting of two Weyl–

Majorana) into a Dirac fermion, one finds for the latter the usual anti-
commutation relations, namely,
{Ψ† (x1 , x0 ), Ψ(y1 , y0 )}|x 0 =y 0 = δ(x1 − y1 ) {Ψ(x1 , x0 ), Ψ(y1 , y0 )}|x 0 =y 0 = 0.

(2.80)
The Noether currents associated with conformal transformations, namely the
components of the energy-momentum tensor, are given by,
1 1
T (z) = − : χ∂χ : T̄ (z̄) = − : χ̃∂¯χ̃ :, (2.81)
2 2
where : χχ : the normal ordered product, stands for the product with the sub-
traction of its OPE. The latter is given by,
1 1
χ(z)χ(w) = χ̃(z)χ̃(w) = . (2.82)
z−w z̄ − w̄
Using this basic OPE in T (z)χ(w) one finds,
1 χ(w) ∂χ(w)
T (z)χ(w) = + , (2.83)
2 (z − w)2 z−w
which implies that χ is a primary field of conformal dimensions of ( 12 , 0), and sim-
ilarly χ̃ with (0, 12 ). The Virasoro anomaly, which comes as usual from T (z)T (w),
is c = 12 , and c̄ = 12 from T̄ (z)T̄ (w).
Recall that the energy-momentum tensor of the scalar field (1.27) takes the
form of a bilinear of the “current algebra” currents (1.24). We want to examine
now if such a construction can be applied also for the fermionic fields. Since for
Weyl–Majorana fermions there are no such currents it is left only to check for
the T (z) of Weyl fermions. Let us note first the OPE of the currents and the
Weyl fermions that read,
ψ(z) ψ † (z)
J(z)ψ(w) = −i J(z)ψ † (w) = i , (2.84)
(z − w) (z − w)
where J = i : ψ † ψ : with our conventions.

Using this OPE one finds,

1 1 1
T (z) = − : J(z)J(z) := − limz →w J(z)J(w) +
2 2 (z − w)2

1 † 1 1
= − limz →w lim→0 iJ(z) ψ (w − )ψ(w + ) + +
2 2 (z − w)2

1 ψ † (z)ψ(w + ) ψ † (w − )ψ(z) 1
= − limz →w lim→0 − + +
2 [z − (w − )] [z − (w + )] (z − w)2
1 †
=− ψ ∂ψ − ∂ψ † ψ . (2.85)
2
This construction of the energy-momentum tensor in terms of a normal ordered
product of two currents, which is known as the Sugawara construction, will play
a key role in the discussion in Chapter 4.
The mode expansion of the Weyl–Majorana fermion takes the form,
ψr 1
dzz r − 2 ψ(z),
1
ψ= 1 ψr = (2.86)
r
z 2+ 2πi
r ∈Z+ν
with z = e−iw , ν is related to the boundary conditions as
ψ(w + 2π) = e2πiν ψ(w), (2.87)
so that there are two types of fermions:
Ramond fermions ν = 12 ↔ periodic boundary condition

Neveu–Schwarz fermions ν = 0 ↔ anti-periodic boundary condition.
The anti-commutation relations of the fermionic modes follow straightforwardly

upon using (2.86) and the OPEs (2.82), namely,
{ψr , ψs } = δr +s . (2.88)
The form (2.82) holds for the periodic case. For the anti-periodic case there is

an extra factor of 12 ( wz + wz ), which tends to 1 as z → w.
The canonical quantization conditions in terms of real space-time coordinates
take the following form,
1
{ψ(x1 ), ψ(y 1 )}|x 0 =y 0 = δ(x1 − y 1 ), (2.89)
2
since ψ is the conjugate momentum of itself. Combining two Majorana fermions
into a Dirac one yields the following anti-commutation relations for the Dirac
fermions,
{Ψ† (x1 ), Ψ(y 1 )}|x 0 =y 0 = δ(x1 − y 1 ) {Ψ(x1 ), Ψ(y 1 )}|x 0 =y 0 = 0, (2.90)
so that now Ψ† is the conjugate momentum of Ψ.

2.13 The Ising model – the m = 3 unitary minimal model 37
2.13 The Ising model – the m = 3 unitary minimal model

The first unitary minimal model with m = 3 has c = 1/2, just like the Majorana
fermion discussed above. We now analyze the primaries of this model, their fusion
rules and their correlators. Comparing the latter with correlation functions of
the Ising model,7 we show that in fact the m = 3 unitary minimal model is the
continuum limit of the Ising model. Recall that the set of primaries of the m = 3
model are characterized by the following conformal weights:
1 1
h1,1 = 0 h2,1 = h2,2 = , (2.91)
2 16
which determine the two-point functions,
' ( 1
Φ(1/2) (z, z̄)Φ(1/2) (w, w̄) =
|z − w|2
' ( 1
Φ(1/16) (z, z̄)Φ(1/16) (w, w̄) = , (2.92)
|z − w|1/4
where Φ(h) (z, z̄) = φ(h) (z)φ̄(h) (z̄). It turns out that at the critical point of the
Ising model, the two-point function of the spin operator σ at a lattice point n and
at the origin behaves like < σn σ0 >∼ |n |11 / 4 . Thus in the continuum, it has the
same “critical exponent” as that of Φ(1/16) , and similarly the Greens function of
the energy density falls like < n 0 >∼ |n1|2 , namely like the two-point function
of Φ(1/2) .
There are additional properties of the m = 3 unitary model that can be shown
to match those of the Ising model. Here we demonstrate this with the determi-
nation of the four-point function of Φ(1/16) , namely,
' (
Φ(1/16) (z1 , z̄1 ) . . . Φ(1/16) (z4 , z̄4 ) . (2.93)
From equation (2.63) we have that

⎡ ⎤

4
⎣ 4 ∂w2 − 1/16
+
1 ∂ ⎦ φ(w1 , w̄1 ) . . . φ(w4 , w̄4 ) = 0,
3 1 (w1 − wi )2 (w1 − wi ) ∂w i
i= 1
(2.94)
(1/16)
where φ denotes Φ .
Using the global Ward identities we express G4 as in (2.59),

−1/24
G4 (z1 , z̄1 . . . z4 , z̄4 ) = f˜(Z, Z̄) (z12 z13 z14 z23 z24 z34 ) (C.C.) , (2.95)
where Z = z1 2 z3 4
z1 3 z2 4 . Using also z1 4 z2 3
z1 3 z2 4 = 1 − Z, we can rewrite as,

−1/8 −1/24
G4 (z1 , z̄1 . . . z4 , z̄4 ) = f˜(Z, Z̄) (z13 z24 ) [Z (1 − Z)] (C.C.) . (2.96)
7 The two-dimensional Ising model has a long history. It was discussed in [137]. The relation
to Majorana fermions was discussed in [187].

Anticipating the result of the Ising model, we actually write,

−1/8
G4 (z1 , z̄1 . . . z4 , z̄4 ) = f (Z, Z̄). [z13 z24 Z (1 − Z)] (C.C.) . (2.97)
If we now substitute this ansatz into (2.94) we find the following differential
equation for f ,

Z(1 − Z)∂ 2 + (1/2 − Z)∂ + 1/16 f (Z, Z̄) = 0. (2.98)
A similar equation holds for Z̄. The solutions of this differential equation are
√ 1/2
f1,2 (Z) = 1 ± 1 − Z and so finally the four-point function takes the form,
) )1/4
) z13 z24 )
G4 (z1 , z̄1 . . . z4 , z̄4 ) = )) )
) |f1 (Z)|2 + |f2 (Z)|2 , (2.99)
z12 z23 z34 z34
where the unique combination is dictated by the requirement for a single value.
This is identical to the result found in the Ising model for G4 .
Note also that f (1 − Z, 1 − Z̄) is a solution, a result of the symmetry under
the interchange of z1 with z3 .
Note also that although Φ(1/2) is a free fermion, Φ(1/16) cannot be constructed
in a local way from the fermion.

3
Theories invariant under affine current algebras
In addition to being conformal invariant, it was shown in Chapter 1 that the

theory of a free massless scalar field admits also affine algebra currents J(z)
¯ which are holomorphically and anti-holomorphically conserved, namely
and J(z̄)
¯ ¯ z̄) = 0. The existence of these currents, as was the case with the
∂J(z, z̄) = ∂ J(z,
energy-momentum tensor, implies that the theory is invariant under an infinite-
dimensional group of transformations. Inspired by this invariance of the free
scalar theory we would like to identify and investigate field theories equipped with
an affine current algebra (ACA), which is often refered to as Kac–Moody algebra
or affine Lie algebra (ALA). Conformal field theories (CFT) are characterized by
the Virasoro anomaly, the set of primary fields and the corresponding structure
constants. It will be shown in this chapter that theories with ACA admit a
similar algebraic structure and moreover that they are necessarily also CFTs.
Thus every ACA theory will be characterized by the Virasoro anomaly as well
as its ACA analog, the ACA level. Primary fields that have been defined so far
via their operator product expansion (OPE) with the T (z) and T̄ (z̄) will have
to obey certain OPE also with currents J(z) and J(z̄).¯ The zero modes of the
free scalar affine currents J0 and J¯0 (1.57) commute, namely, they generate an
abelian group. For theories with “ordinary” non-affine currents the generalization
of the abelian group to non-abelian ones led (in four dimensions) to the standard
model of the basic interactions and in fact to an enormously rich spectrum of
interesting theories. It is thus very natural to explore the generalization of the
abelian ACA to non-abelian affine current algebras. The investigation of two-
dimensional theories which are invariant under transformations generated by
such affine currents is the subject of this chapter. We start with a brief reminder
of the properties of finite dimensional Lie algebras.
The topics included in this chapter are covered in several books and review
papers. In particular we have made use of the famous review by Goddard and
Olive [111], and the book by Di Francesco, Mathieu and Senechal [77].
3.1 Simple finite-dimensional Lie algebras

Consider the Lie algebra G,
[T a , T b ] = ifcab T c , (3.1)

40 Theories invariant under affine current algebras
associated with a group G, namely, the set T a are the generators of the group
G.1 We will consider simple groups, namely those that do not contain invari-
ant subgroups. Denote the maximal set of commuting Hermitian generators by
H i , i = 1, . . . , r so that
[H i , H j ] = 0 i, j = 1, . . . , r. (3.2)
This abelian subalgebra of G is referred to as the Cartan subalgebra. It can

be shown that any two such abelian subalgebras with generators H i and H̃ i
are conjugate under the action of the group, namely, H̃ i = gH i g −1 for some
g ∈ G. The dimension of the Cartan subalgebra, which is the maximal number
of commuting generators is defined as the rank of the algebra G, rank (G) = r.
A basis of the full algebra G constitutes H i and the step operators or ladder
operators E α defined by,
[H i , E α ] = αi E α , i = 1, . . . , r. (3.3)
The r-dimensional vector α is called a root associated with the step operator E α .
The roots are real and up to multiplication with a scalar there is a single E α
associated with α via (3.3). No multiple of a given root α is a root apart from
−α which is the root paired with E −α = E α † . The number of roots is obviously
(dimG − rankG).
The rest of the algebra are the commutation relations [E α , E β ] which follow
from the Jacobi identity,
[H i , [E α , E β ]] = (αi + β i )[E α , E β ], (3.4)
so that if (αi + β i ) = 0 and is not a root, then [E α , E β ] = 0. If on the other hand

(αi + β i ) is a root then [E α , E β ] must be a multiple of E α +β . If (αi + β i ) = 0 it
follows that [E α , E −α ] ∼ α · H.
To summarize the full algebra reads,
[H i , H j ] = 0 i, j = 1, . . . , rank(G)
[H i , E α ] = αi E α α = 1, . . . , (dim G − rank(G))
α β α +β
[E , E ] = (αβ)E if α + β is a root
2α · H
= if α + β = 0
α2
= 0 otherwise. (3.5)
This basis of the algebra is a modified version of the Cartan–Weyl basis. The
constants (αβ) can be chosen to be ±1 if all the root vectors have the same
length.
1 Finite-dimensional Lie algebra is covered in many books, for instance Cahn [54].

It is straightforward to realize that the triplet of generators E α , E −α , and αα·H

2
is isomorphic to J+ , J− , J3 of an SU (2) algebra, namely,

α · H ±α α·H
2
,E = ±E ±α , [E +α , E −α ] = 2 2 . (3.6)
α α
Consequently the eigenvalues of 2 αα·H

2 , just like those of 2J3 , are integers in any
unitary representation. The eigenvalues associated with each root β is given by

2α ·β ∧
α 2 ∈ Z. It is natural to define the notion of coroot α = α 2 .
2α
3.1.1 The Weyl group

Consider a root β such that 2α · β/α2 = 0 is its eigenvalue under the operation
of 2 αα·H
2 . There must be another step operator E
β +m α
which is a member of the
SU (2) multiplet with the opposite eigenvalue, namely,
2α · β/α2 + 2m = −2α · β/α2 . (3.7)
Then m = −2α · β/α2 , and

(α · β)α
β + mα = β − 2 = β − (α∧ , β)α ≡ σα (β) (3.8)
α2
is a root for each pair of roots α and β; σα (β) is a reflection in the hyperplane
perpendicular to α. The set of these reflections that permute the roots, generate
a finite group W (G), the Weyl group of G.
3.1.2 Cartan matrix and Dynkin diagrams

It is convenient to define the notion of simple roots as follows. Select a rank(G)
dimensional basis of the roots that consists of α(i) , i = 1, . . . , rank(G) in such a
way that any root α can be written as,
G
rank
α= ni α(i) , (3.9)
i
where the ni are integers which are either all ni ≥ 0 or all ni ≤ 0. In the former
case α is positive, while in the latter it is negative. This base is called the basis
of simple roots. Associated with the simple roots one defines the simple Weyl
reflections σα ( i ) that generate the entire Weyl group.
The scalar products of simple roots define the Cartan matrix as follows:
2α(i) · α(j )
Aij = 2 . (3.10)
α(j )
The Cartan matrix is a rank(G) × rank(G) matrix with integer components and
with diagonal elements which take the value of 2. The off diagonal elements are
either negative or vanishing.

Ar su(r+1)
1 2 3 r−2 r−1 r
r−2 r−1 r
Br so(2r+1)
1 2 3 r−2 r−1 r
Cr sp(r)
1 2 3 r−2 r−1 r
r
Dr r−2 so(2r)
1 2 3 r−3 r−1
E6
E7
E8
F4
G2
Fig. 3.1. Dynkin diagrams for all the simple Lie algebras.
It can be shown that the roots of a simple Lie algebra can have at most two
different lengths, a long root and a short one. The ratio between their lengths
are either 2 or 3. When all the roots have the same length the algebra is a simply
laced algebra. From the Cartan matrix Aij one can reconstruct a basis of simple
roots up to a scale and overall orientation. In fact constructing all the roots from
the simple roots, one finds that full information on G is encoded in Aij .
The information contained in the Cartan matrix Aij may be encoded in a
planar diagram, the Dynkin diagram. The construction of such a diagram is as
follows:
r To each simple root α assign a node in the diagram.
(i)
r If a node represents a short root mark it by a black dot, and if a long one by
a white dot.
r Join the points α and α by Aij Aj i lines. For i = j, Aij Aj i can take the val-
(i) (j )
ues of 0, 1, 2, 3. In fact since Aij Aj i = 4cos2 θij , where θij is the angle between
the two roots, then orthogonal simple roots are not connected, and those with
an angle of 120, 135, 150 degrees are connected with one, two, or three lines.
r In some conventions one does not separate between black and white dots, but
2 2
rather one draws an arrow pointing from α(i) to α(j ) when α(i) > α(j ).
The Dynkin diagrams for all simple Lie algebras are given in Fig. [3.1].

3.1.3 Highest weight states

We now consider finite-dimensional representations of G other than the adjoint
representation that has been analyzed so far, the latter being the same as the
generators. We can always choose a basis {|μ } for which,
H i |μ = μi |μ . (3.11)
The rank(G) dimensional vector μ of eigenvalues of the Cartan subalgebra gener-

ators is called the weight vector. A root is a weight of the adjoint representation.
In a similar manner to their action as roots, the triplet E α , E −α and 2α · H/α2
form an SU (2) algebra and hence {|μ } must be an SU (2) multiplet, and in
particular,
2α · μ/α2 ∈ Z, (3.12)
for any root α.

This property of any given weight defines a lattice ΛW (G), the weight lattice
of the algebra G. The weights associated with a representation must be mapped
into one another under the operation of σα and in fact the whole Weyl group.
One can choose a basis for the weight lattice ΛW (G) consisting of fundamental
weights λ(i) such that,
2λ(i) · α(j ) /α(j

2
) = δij . (3.13)

Any weight λ can then be expanded as λ = ni λ(i) with integer coefficients ni . If
all ni ≥ 0, the weight λ is called a dominant weight. Every weight can be mapped
into a unique dominant weight by action of the Weyl group. The dominant weight

ρ = i λ(i) , where i = 1, . . . , rankG, is characterized by ρ · α > 0 for any positive

α and ρ · α < 0 for any negative α. In fact ρ = 1/2 α where the sum is over
all the positive roots.
For any given finite-dimensional representation of G one defines the highest
weight state |μ0 for which ρ · μ0 is the largest. Such a state is annihilated by all
the raising operators,
E α |μ0 = 0, (3.14)
for every α > 0. All the states of a given irreducible representation can be built
by acting on the highest weight state with lowering operators, namely, each state
takes the form,
E −β 1 . . . .E −β n |μ0 . (3.15)
In fact every irreducible representation has a unique highest weight state |μ0 and
the other weights μ have the property that μ0 − μ is a sum of positive roots. The
highest weight state is a dominant weight. In the opposite direction for each dom-
inant weight there is a unique irreducible representation for which it is the highest
weight state. As was mentioned for the adjoint representation the weights are

s1 1
α2 α1+α 2
ω1
− α1 ω2 α1
s1 s 2 s2
− α1−α2 −α 2
s1 s2s1 s2 s1
Fig. 3.2. Root system and Weyl chambers of SU (3).
the roots so that the corresponding highest weight state is the highest root. The
difference between the highest root and any other root is a sum of positive roots.
We end this subsection with an example. Consider the SU (3) algebra. The
Cartan matrix for this algebra takes the form,
2 −1
. (3.16)
−1 2
The simple roots are related to the fundamental weights as α(1) = 2λ(1) −
λ(2) and α(2) = −λ(1) + 2λ(2) . The scalar products between the fundamental
weights are (λ(1) , λ(1) ) = (λ(2) , λ(2) ) = 2/3 and (λ(1) , λ(2) ) = 1/3, using the stan-
dard normalization of (α(i) , α(i) ) = 2. The full Weyl group is given by W =
{1, σ1 , σ2 , σ1 σ2 , σ2 σ1 , σ1 σ2 σ1 }. The action of the different elements of the Weyl
group on the two simple roots gives all roots.
The root system and the Weyl chambers are given in Fig. 3.2. The Weyl cham-
bers are separated by the dashed lines, and are specified here by the elements of
the Weyl group, with the latter denoted in the figure by si . The Weyl chambers
are defined by,
Cω = {λ|(ωλ, αi ) ≥ 0, i = 1 . . . r}.
3.2 Affine current algebra

In the previous subsection we acquired a certain familiarity with the notions of
roots, highest weights, Cartan matrices, Dynkin diagrams etc., in the context of
a simple Lie algebra. As was explained in the introduction to this chapter, two-
dimensional CFTs are characterized by an extended algebraic structure, that of
affine Lie algebra.2 We now describe the basic properties of the affine Lie algebra
using the notions of the previous subsection.
2 Affine Lie algebras were introduced into the physics literature by Bardacki and Halpern in
[27]. Independently V. Kac [136] and R.V. Moody [164] introduced them in the mathematical
literature.

The basic ALA is given by,

a
[Jm , Jnb ] = ifcab Jm
c
+n + k̂mδab δm +n ,0 , (3.17)
a
where the central element k̂ commutes with all the generators Jm , namely,
a
[Jm , k̂] = 0.
We now use a generalization of the Cartan–Weyl basis to the affine algebra as
follows:
i
[Hm , Hnj ] = k̂mδ ij δm +n ,0
i
[Hm , Enα ] = αi Em
α
+n ,
α α +β
[Em , Enβ ] = (αβ)Em +n if α + β is a root
2
= 2 (α · Hm +n + k̂mδm +n ) if α + β = 0
α
= 0 otherwise.
i
[Hm , k̂] = [Enα , k̂] = 0. (3.18)
We have used the normalization (H i , H j ) = δ ij , (E α , E β ) = α22 δα +β , where

(X, Y ) denotes the Killing form, defined as the trace of the product in the adjoint
representation, T r(adX, adY ). The hermiticity properties are,
† α† −α
i
Hm i
= H−m , Em = E−m , k̂ † = k̂ (3.19)
Unlike the simple Lie algebra, here we have an r + 1 dimensional abelian sub-
algebra consisting of [H01 , . . . , H0r , k̂]. With respect to these generators, Em
α
are
step operators,
[H0i , Em
α
] = α i Em
α
[k̂, Enα ] = 0. (3.20)
Each of the eigenvectors (α1 , . . . , αr , 0) is infinitely degenerate since it is indepen-

dent of m. Moreover this abelian subalgebra is not maximal since [H0i , Hnj ] = 0.
Thus one has to extend the algebra by adding a grading operator which can be
taken to be L0 such that,
[L0 , Jna ] = −nJna [L0 , k̂] = 0. (3.21)
Using the generators (H0i , k̂, −L0 ) as in the Cartan subalgebra we have as the
step operators Enα corresponding to a root (α, 0, n) and Hni corresponding to
(0, 0, n).
The root system of ALA is thus infinite but spans an (r + 1) dimensional space.
We can divide the roots into positive and negative according to the following rule:
(α, 0, n) > 0 if n > 0, or if n = 0 and α > 0 (3.22)
The basis of the simple roots can therefore be taken as,
α(i) = (αi , 0, 0), 1≤i≤r

α(0) = (−θ, 0, 1), (3.23)

where αi is the basis of simple roots of the Lie algebra, and θ is the highest root.
Thus an arbitrary root of the ALA can be expressed as,
r
α= ni α(i) . (3.24)
i= 0
It is positive if ni ≥ 0 and negative when ni ≤ 0, and these are the only two
posibilities.
3.2.1 Cartan matrix and Dynkin diagrams

The first step is to define the scalar product X, Y which has to be symmetric
and obey the relation,
X, [Y, Z] + Y, [X, Z] = 0
for X, Y, Z ∈ ĝ, the ALA. Upon using a convenient normalization one can bring
the basic scalar products to the following form,
Tma , Tnb = δ ab δm +n
Tma , −L0 = 0
Tma , k̂ = 1
k̂, k̂ = 1
k̂, −L0 = 1
−L0 , −L0 = 0. (3.25)
The last relation is actually a choice, following on from the invariance of the
algebra under a shift of L0 by a multiple of k̂. Here we use T instead of δ used
previously.
In the following Hermitian basis,
(Tmi + T−m
i
) (k̂ − L0 ) (k̂ + L0 )
T0i , √ , √ , √ , (3.26)
2 2 2
the scalar product is Lorentzian since the norm of all the first three basis vectors
is +1 while that of ( k̂ +L )
√ 0 is −1.
2
The Lorentzian signature holds also for the Cartan sub-algebra (CSA) genera-
tors. One can define the scalar product of two vectors of simultaneous eigenvalues
of the CSA,
mi = (μi , μik , μi−L 0 ), mj = (μj , μjk , μj−L 0 )
to be,
mi · mj = μi · μj + μik · μj−L 0 + μi−L 0 · μjk . (3.27)
In particular the scalar product of two roots,
ai = (αi , 0, ni ), aj = (αj , 0, nj )

is
ai · aj = αi · αj . (3.28)
The root that corresponds to Enα , a = (α, 0, n) has a norm a2 = α2 > 0 and hence
is referred to as a space-like root, whereas the root that is associated with Hni ,
nδ = (0, 0, n) has zero norm (light-like) and is orthogonal to all other roots. We
have used the “unit” of δ = (0, 0, 1).
The Cartan matrix of ĝ, which is an (r + 1) × (r + 1) matrix, is defined in a
similar way to one of the Lie algebra, namely,
2a(i) · a(j )
Aij = 0 ≤ i, j ≤ r. (3.29)
a2(j )
We add to the Cartan matrix of the Lie algebra, the extra row and column Ai0
and A0i which can be found using (3.10) with α0 = −θ. Now from the definition
r
of the fundamental weight it follows that θ = − i=0 A0i λi . Since θ is a long
root of g, namely, θ2 ≥ α(i)
2
one gets that,
−Ai0 = 1 if A0i = 0
−Ai0 = 0 if A0i = 0, (3.30)
provided that θ is not itself a simple root, as happens for SU (2). The Dynkin
diagram of ĝ is obtained using that of g appended with an extra point that
corresponds to α0 connected by −A0i lines to the points a(i) . If −A0i > 1 an
arrow is drawn which points toward a(0) . We demonstrate the construction in
the following example:
SUˆ (2) - There are only two simple roots a(0) = (−α, 0, 1) and a(1) = (α, 0, 0)
so that A0i = Ai0 = −2 and the Cartan matrix is
2 −2
. (3.31)
−2 2
Thus, there are two roots of equal length connected by four lines with arrows
pointing in both directions to indicate that a2(0) is equal to a2(1) .
The Dynkin diagrams of the affine simple algebra are shown in Fig. 3.3. The
point that corresponds to α(0) is marked by a zero. The black dots relate to the
notion of twisted affine Lie algebra which we do not discuss here (see for example
[111]).
3.2.2 The Weyl group

The Weyl group of ĝ is defined in a similar way to that of g, namely it is generated
by reflections in the hyperplanes normal to a,
b·a α·β
σa (b) = b − 2 a = (σα (β), 0, na − 2 nβ ), (3.32)
a2 α2

A1 E6
1 3 4 5 6
0
Ar 2
r>2
E7
1 2 r−1 r 0 1 3 4 5 6 7
2
0
Br E8
r>3 1 2 3 r−2 r−1 r 1 3 4 5 6 7 8 0
Cr F4
r>2 0 1 2 r−2 r−1 r 0 1 2 3 4
0 r
Dr G2
r>3
1 2 3 r−3 r−2 r−1 0 1 2
Fig. 3.3. Dynkin diagrams for all the affine simple Lie algebras.
for a = (α, 0, nα ) and b = (β, 0, nβ ) a space-like root. Light-like roots are invari-
ant under such reflections σa (nδ) = nδ. In fact the Weyl group of ĝ is a semi-
direct product of the Weyl group of g and the coroot lattice of g which is the
lattice generated by the coroots αν = αα2 . The coroots form the root system of
the Lie algebra dual to g obtained by interchanging the root lengths. The simply
laced algebras An , Dn , En are obviously self-dual, as are F4 and G2 , whereas
Bnν = Cn and Cnν = Bn .
3.2.3 Highest weight representations

A highest weight state |μ̂0 is a state that is annihilated by all the raising oper-
ators for positive roots, namely,
E0α |μ̂0 = En±α |μ̂0 = Hni |μ̂0 = 0, (3.33)
for n > 0, α > 0. The eigenvalue of this state is the highest weight vector μ̂0 =
(μi0 , k, h) given by,
H0i |μ̂0 = μi0 |μ̂0 , k̂|μ̂0 = k|μ̂0 , L0 |μ̂0 = h|μ̂0 . (3.34)
We can set h to zero as a matter of convention. A highest weight representation

is characterized by a unique highest weight state. To have a unitary highest
weight representation the following necessary and sufficient conditions have to
be obeyed:
2k
∈Z k ≥ θ · μ0 ≥ 0. (3.35)
θ2

3.3 Current OPEs and the Sugawara construction 49
The non-negative integer 2k θ 2 is the level of the representation. Any state in

the representation is characterized by a weight vector μ̂ = (μi , k, h) such that
μ0 − μ is a sum of positive roots. Introduce now a set of fundamental weights
l(i) for ĝ, i = 0, . . . r, such that 2l(i) · a(j ) /a2(j ) = δij . The general solution of the
condition 2a · μ̂/θ2 ∈ Z, which is equivalent to the condition Eq. (3.35) for a =
r
a(0) = (−θ, 0, 1), now takes the form μ̂0 = i= 0 ni l(i) , where ni are non-negative
integers, apart from the indeterminate component in the L0 direction. For l(i)
one finds,
1 1
l(i) = λ(i) , mi θ2 , 0 l(0) = 0, m0 θ2 , 0 , (3.36)
2 2
where m0 = 1 and where the integers mi are defined via θ/θ2 =
r 2
i= 0 m i α(i) /α(i) . The corresponding level is given by,

r
level = ni mi . (3.37)
i= 0
Level 1 representations are thus associated with highest weights l(i) with all
mi = 1. Those are indicated by open points in Fig. 3.3.
From the definition of mi it follows that,

r
Aij mj = 0. (3.38)
i= 0
Since the Cartan matrix has the basic symmetry of the extended Dynkin dia-
gram, also the positions of the open dots have to preserve this symmetry. For
the classical groups Ar , Br , Cr , Dr the values of mi for the closed dots is 2. For
the exceptional groups the vector (m0 , . . . , mr ) is as follows
Ê 6 (1, 1, 2, 2, 3, 2, 1)
7
Ê (1, 2, 2, 3, 4, 3, 2, 1)
8
Ê (1, 2, 4, 6, 5, 4, 3, 2, 1)
4
F̂ (1, 2, 3, 2, 1)
2
Ĝ (1, 2, 1) (3.39)
3.3 Current OPEs and the Sugawara construction

In Section 1.8 for the free scalar theory it was shown that the OPE of two currents
1
J(z)J(w) takes the form of J(z)J(w) = (z −w ) 2 + finite terms. This type of OPE,
which corresponds to the abelian ALA, is generalized following the discussion in
Section 3.2 to,
kδ ab f ab J c (w)
J a (z)J b (w) = +i c + finite terms. (3.40)
(z − w) 2 (z − w)

We can now use the OPE to evaluate the infinitesimal transformation of the
current under ALA transformations,
1
δ J a (w) = dzb (z)J b (z)J a (w)
2πi w

1 kδ ab fcab J c (w)
= dzb (z) + i = ifbc J − k∂a .
a b c
(3.41)
2πi w (z − w)2 (z − w)
The same structure also holds for J¯a (z̄).
The OPE form of the ALA can be transformed into a commutator form of the
algebra. We introduce a Laurent expansion of the currents,
1
J a (z) = Jna z −(n + 1) Jna = dzz n J a (z). (3.42)
n
2πi
Substituting the OPE into the expression of the commuation relation we indeed
find the ALA of (3.17), namely
a
c
+n + k̂mδab δm +n ,0 . (3.43)
In free scalar theory there are two “currents” which are holomorphically con-
served, J and T , and moreover the energy-momentum tensor is bilinear in J,
as was shown in Section 1.5. We now elevate this special case into a general
construction of T for theories which admit ALA structure. The construction is
known as the Sugawara construction. One writes T as a normal ordered product
of the currents,
1
T (z) = : J a (z)Ja (z) : (3.44)
2κ
with a coefficient κ that has to be determined quantum mechanically. In fact,
one way to determine κ is by requiring that J a is a primary field of weight 1,
namely,
Ja ∂J a (w)
T (z)J a (w) = + . (3.45)
(z − w)2 (z − w)
Using the OPE (3.40) and the relation −fcab fbd
c
= 2Cδda , where C is the dual
(k +C )
Coxeter number, we find that κ = 1 so that the form of the Sugawara con-
structed T is,
1
T (z) = : J a (z)Ja (z) : (3.46)
2(k + C)
Note that the Casimir of the adjoint is 2C. Note also that in Section 1.5, for the
free scalar case, we had a relative minus sign, due to a difference of factor i in
defining the currents there.
In the WZW models discussed in the next section, classically one has T with a
1
coefficient of 2k . It is thus clear that for those models quantum mechanically, due
to the double contraction, we get a finite renormalization of the level k → k + C.

Here we have used currents which are in an orthonormal basis. If instead

we express the current in the Cartan–Weyl basis used in the previous sections,
the form of T in terms of the Cartan sub-algebra generators H i and the step
operators E α is,

1 |α|2 α −α
T (z) = : H i (z)H i (z) : + (E E + E −α E α ) . (3.47)
2(k + C) 2
The OPEs (3.45) and (3.40) also enable us to determine c, the Virasoro
anomaly of the model, via the computation,
1
T (z)T (w) = T (z) : J a J a : (w)
2(k + C)
* a +
1 J (z)J a (w) ∂J a (z)J a (w)
= +
(k + C) (z − w)2 (z − w)
* +
1 k(dim G) : J a J a : (w) 1 ∂ : J a J a : (w)
= + +
(k + C) (z − w)4 (z − w)2 2 (z − w)
1 k(dim G) T (w) ∂T (w)
= +2 + . (3.48)
(k + C) (z − w)4 (z − w)2 (z − w)
We thus read off the Virasoro anomaly
kdim G
c= . (3.49)
k+C
The construction of T in terms of a normal ordered product of two currents
calls for combining together the ALA and the Virasoro algebra. Substituting into
(3.46) the mode expansions, of T (z) in terms of Ln and of J(z) in terms of Jn ,
one finds that,
1
Ln = : Jna−m Jm
a
: (3.50)
2(k + C) m
where here the normal ordering implies putting the currents with positive m to
the right. In fact normal ordering is required only for L0 .
Using this relation, we write down the full Virasoro algebra and ALA,
c
[Ln , Lm ] = (n − m)Ln +m + (n − 1)n(n + 1)δ(n + m)
12
a
[Ln , Jm ] = −mJm a
+n
a
c
+n + k̂mδab δm +n ,0 . (3.51)
In mathematical terminology the Virasoro algebra belongs to the enveloping

algebra of the ALA.
3.4 Primary fields

Recall that the operators of any CFT were shown to be either Virasoro primaries
or descendants. The former were defined by their OPE with T . In a similar

manner ALA primaries Φl, ¯l (z, z̄) are defined via their OPE with J,
Tla Φl, ¯l (w, w̄)
J a (z)Φl, ¯l (w, w̄) =
(z − w)
a
T¯ Φl, ¯l (w, w̄)
J¯a (z)Φl, ¯l (w, w̄) = l , (3.52)
(z̄ − w̄)
where Tla , T¯la are the matrix T a in the l, ¯l representations, for the holomorphic
and antiholomorphic sectors, respectively. From here on we discuss only holo-
morphic properties. In terms of the Laurent components Jna the condition for a
primary field reads,
Jna Φl, ¯l (0, z̄) = 0 for n > 0; J0a Φl, ¯l (z, z̄) = Tla Φl, ¯l (z, z̄). (3.53)
In theories where the energy-momentum tensor can be constructed in a Sug-

awara construction it is easy to see that the ALA primaries are also Virasoro
primaries. Indeed, using (3.50) we see that Ln for n > 0 annihilates the ALA
primary. For L0 acting on the primaries we get,
1 C2 (l)
L0 Φl = J0a J0a Φl = Φl . (3.54)
2(k + C) 2(k + C)
Thus the primaries in theories equipped with the Sugawara construction, for
instance the WZW models that will be discussed in the next section, obey (3.53)
and also,
C2 (l)
Ln Φl, ¯l (0, z̄) = 0 for n > 0; La0 Φl, ¯l (z, z̄) = Φ ¯(z, z̄). (3.55)
2(k + C) l, l
Recall that T is not a Virasoro primary but rather is a descendant of the identity
T (0) = L−2 I. The same applies to J(z). From the mode expansion (3.42) it is
clear that,
J a (0) = J−1
a
I. (3.56)
Note however that J a (z) is a Virasoro primary. Apart from the distinguished
descendant J there are descendant operators of all the primaries. In fact all the
local operators can be written as,
ā
a1
J−n 1
aN
. . . J−n N
J¯−n̄
ā 1
1
. . . J¯−n̄
N̄
N̄
Φl, ¯l (z, z̄), (3.57)
and in the case of a Sugawara construction all the operators are of the form,
ā N̄
a1
L−m 1 . . . L−m M L̄−m̄ 1 . . . L̄−m̄ M̄ J−n 1
aN
. . . J−n N
J¯−n̄
ā 1
1
. . . J¯−n̄ N̄
Φl, ¯l (z, z̄). (3.58)
3.5 ALA characters

In Section 2.8 we introduced the notion of the Virasoro character (2.45) which
characterizes the structure of the Virasoro Verma module. In a complete analogy

3.6 Correlators, null vectors and the Knizhnik–Zamolodchikov equation 53
let us now define a character of the CFT and ALA module λ̂ as follows:

χλ̂ (z j , τ ) = e−im λ̂ δ T rλ̂ [e2π iτ L 0 e−2π i j zj hj
], (3.59)
where mλ̂ , δ and hj are the generators of the Cartan subalgebra associated with
the group, and zj are complex numbers. The character can also be expressed in
terms of the generalized theta function Θλ̂ in the following form:

w ∈W (w)Θw ( λ̂+ ρ̂)
χλ̂ (z , τ ) =
j
, (3.60)
w ∈W (w)Θw ρ̂
where the sums are over the elements of the finite Weyl group, (w) = (−1)l(w )
with l(w) the length of w.
Rather than defining the generalized theta function for any ALA at any level,
ˆ (2) level k. For this case we have,
we define here only the function for SU
λ2τ
Θλ 1 (z; τ ; t) = e−2π k t e−2π i[k n z + 2 λ 1 z −k n τ−
(k ) 1 2 1 ]
4k . (3.61)
n ∈Z
(k )
with Θλ 1 (z; τ ; o) ≡ Θ(k − λ1 , λ1 ) (see [77] for details).
In terms of this function the character of SU ˆ (2)k takes the form,
(k + 2) (k + 2)
Θλ 1 + 1 − Θ−λ 1 −1
χλ̂ = (2) (2)
, (3.62)
Θ1 − Θ−1
where λ̂ = [k − λ1 , λ1 ]. For the special point (z = 0, t = 0) the character

expressed in terms of q = e2πiτ reads,

(λ 1 + 1)2 [λ1 + 1 + 2n(k + 2)]q n [λ 1 +1+(k +2)n ]
χλ̂ (q) = q 4 ( k + 2 ) − 8 n ∈Z
1
n [1+2n ]
. (3.63)
n ∈Z [1 + 4n]q
For level one and for k = λ1 = 1 we get,

5 (2 − 4q + 8q 5 − 10q 8 + . . .)
χλ̂ (q) = q 2 4
(1 − 3q + 5q 3 − 7q 6 + 9q 10 + . . .)
5
= q 2 4 (2 + 2q + 6q 2 + 8q 3 + . . .). (3.64)
The content of the four first grades of the module [k − λ1 = 0, λ1 = 1] is

(1), (1), (3) ⊕ (1), (3) ⊕ 2(1), so that the number of the states in these different
grades is indeed 2,2,6 and 8 as in the expression of the character.
3.6 Correlators, null vectors and the

Knizhnik–Zamolodchikov equation
Correlators of Virasoro primaries were subjected to local and global Ward identi-
ties, (2.61) and (2.56), respectively. We now derive their ALA duals. Performing

a group transformation of a given correlator,

% &
dz a a
(z)J (z)φl 1 (w1 , w̄1 ) . . . φl n (wn , w̄n )
2πi
n % &
dz a a
= φl 1 (w1 , w̄1 ) . . . (z)J (z)φl i (wi , w̄i ) . . . φl n (wn , w̄n )
i= 1
2πi

n
= φl 1 (w1 , w̄1 ) . . . δ φl i (wi , w̄i ) . . . φl n (wn , w̄n ) . (3.65)
i= 1
Now from the OPE (3.52) we know that δ φl i (wi , w̄i ) = a (wi )Tlai φl i (wi , w̄i ).
Since this holds for arbitrary we can get a local form of the Ward identity in
the form,
Tlai
n
J a (z)φl 1 (w1 , w̄1 ) . . . φl n (wn , w̄n ) = φl 1 (w1 , w̄1 ) . . . φl n (wn , w̄n ) .
i= 1
(z − wi )
(3.66)
As for the global Ward identity, we use the fact that the correlator has to be
invariant under global g transformations (constant a ), namely,
δa φl 1 (w1 , w̄1 ) . . . φl n (wn , w̄n ) = 0,
leading to

n
Tlai φl 1 (w1 , w̄1 ) . . . φl n (wn , w̄n ) = 0. (3.67)
i= 1
Null vectors of CFTs were found to be useful in Section 2.9, since they lead to
differential equations for certain correlators. In a similar manner one can write
down null vectors of ALA. In the context of the Sugawara construction, due
to the link between the Virasoro algebra generator T and the ALA generators
J a , there are null vectors that combine generators from both infinite algebras.
We discuss now an important example of this class that leads to the Knizhnik–
Zamolodchikov equations. Consider, at Virasoro level one, the following null
vector,
* +
1
|null> = L−1 − a
J−1 Tlai |Φl i> . (3.68)
k+C
It is easy to see that this is indeed a null state, following (3.52). If we insert
the corresponding null operator into a correlation function of primary fields, like
<Φ1 (z1 ) . . . null(zi ) . . . Φn (zn )>, the latter must vanish and hence we get,
⎧ ⎫
⎨ 1 Tia Tja ⎬
∂i − <Φ1 (z1 ) . . . Φn (zn )> = 0. (3.69)
⎩ k+C (zi − zj ) ⎭
j = i

In the derivation, we use,

a
<Φ1 (z1 ) . . . J−1 Φi (zi ) . . . Φn (zn )>
1 dz
= <Φ1 (z1 ) . . . J a (z)Φi (zi ) . . . Φn (zn )>
2πi z i z − zi
1 dz Tja
= <Φ1 (z1 ) . . . Φj (zj ) . . . Φn (zn )>
2πi z i ,j = i z − zi (z − zj )
j = i
Tja
= <Φ1 (z1 ) . . . Φn (zn )> . (3.70)
(zi − zj )
j = i
For the case of four-point functions, as the correlator depends only on the cross-
ratio coordinate Z = zz 11 23 zz 32 44 , the partial differential equations reduce to an ordi-
nary differential equation. In Section 4.4 we will demonstrate a solution of the
Knizhnik–Zamolochikov equation for a four-point function.
3.7 Free fermion realization

In the previous chapter the theories of massless free Dirac and Majorana fermions
were analyzed as examples of CFTs. In particular it was shown that the Dirac
fermion admits an abelian ALA structure. It is thus natural to expect that the
theory of N fermions should be invariant under the transformations associated
ˆ
with non-abelian ALAs.3 Indeed, it will be shown in this section that an SO(N )
ALA, and a Û (N ) are the underlying algebraic structures of N free massless
Majorana fermions and N Dirac fermions, respectively. We start with the former
case.
2
3.7.1 Free Majorana fermions and SO(N)
Consider a generalization of the action given in Section 2.11 for N Majorana
fermions,
N
1 ¯ i + ψ̃i ∂ ψ̃i },
S= d2 z {ψi ∂ψ (3.71)
8π i= 1
where ψ and ψ̃ are left and right Weyl–Majorana fermions, respectively. Note
that this is possible in 2d, and in any other dimension that is 2 modulu 8. In 4d,
for example, we do not have a Weyl–Majorana fermion, as in the case of a single
Majorana fermion, due to the equations of motion,
ψ ≡ ψi (z) ψ̃ ≡ ψ̃i (z̄). (3.72)
3 The free fermion realization of ALA was presented for the first time in [27].

However, unlike the case of a single fermion, here the action is invariant under
2 ) affine algebra generated by the following
transformations associated with SO(N
holomorphically (anti-holomorphically) conserved currents,
1 1
J a (z) = ψ i Tija ψ j J¯a (z̄) = ψ̃ i Tija ψ̃ j , (3.73)
2 2
where T a are SO(N ) matrices which can be expressed as,
(k l)
Tija ≡ tij = i(δik δjl − δjk δil ). (3.74)
The coefficients (halfs) are not determined by the Noether procedure, but are
chosen in a manner that will be explained below.
The Tija matrices obey the relations,
T r[T a T b ] = 2δ ab

Tija Tkal = −δik δj l + δil δj k
a

fabc fabd = 2(N − 2)δcd . (3.75)
ab
The anticommutation relation and the OPE generalize in an obvious way those
of the single Majorana fermion, namely,
1
{ψ i (x0 , x1 )ψ j (y0 , y1 )}|x 0 =y 0 = δ ij δ(x1 − y1 ) (3.76)
2
and
δ ij δ ij
ψ i (z)ψ j (w) = ψ̃ i (z)ψ̃ j (w) = . (3.77)
z−w z̄ − w̄
Now using this OPE one can derive the OPE of two currents and verify that
they take the form of (3.40),
1
J a (z)J b (w) = : ψ i (z)Tija ψ j (z) :: ψ k (w)Tkbl ψ l (w) :
4

1 δ ik
= Tija Tkbl − : ψ i (z)ψ k (w) : + ψ j (z)ψ l (w)
4 z−w

δ il 1 1
+ : ψ i (z)ψ l (w) : ψ j (z)ψ k (w) = Tija Tkbl
z−w 4 z−w
[ −δ ik : ψ j (z)ψ l (w) : +δ il : ψ j (z)ψ k (w) : +δ j k : ψ i (z)ψ l (w) :
1 1 ik j l
− δ j l : ψ i (z)ψ k (w) :] + Tija Tkbl −δ δ + δ il δ j k . (3.78)
4 (z − w)2
By expanding the fields that are functions of z around w and using the relations
above one finds that indeed the OPE of the two currents take the form of (3.40),
namely,
1δ ab fcab J c (w)
J a (z)J b (w) = + + finite terms (3.79)
(z − w)2 (z − w)

It is thus clear that this is a realization of an SO(N ) ALA of level k = 1.

The Noether currents associated with the conformal transformations, the

energy-momentum tensor T (T̄ ) is just the sum of T (T̄ ) associated with each
one of the N Majorana fermions, hence,
1 1
T (z) = − : ψ i ∂ψ i : T̄ (z̄) = − : ψ̃ i ∂¯ψ̃ i : . (3.80)
2 i 2 i
Since the Virasoro anomaly of a single Majorana fermion is c = 12 it is clear that

the theory of N fermions has c = N2 .
In Section 2.12 it was shown that T of a Dirac fermion could be transformed
into a Sugawara form, T (z) = − 12 : J(z)J(z) :, where J(z) was the U (1) current.
Since we will show below that the Sugawara form is the underlying structure of
the important class of WZW models, it is a natural question to ask whether also
in the present case for the N fermions T can be put into a Sugawara construction.
2 )1 ,
Now, using the expression for the Virasoro anomaly for a theory with SO(N
we find, as we saw before, that,
dimG 1
N (N − 1) N
c= = 2 = . (3.81)
k+C 1 + (N − 2) 2
3.7.2 Primary fields

Similarly to the case of a single Majorana field, the OPE of T (z)ψ i (w) is,
1 ψi ∂ψ i
T (z)ψ i (w) = + , (3.82)
2 (z − w)2 z−w
which implies that ψ i are N primary fields of conformal dimensions ( 12 , 0), and
similarly ψ̃ i has dimension (0, 12 ).
Is the primary field Φ(1/2,1/2) (z z̄) = ψ(z)ψ̃(z̄) the only primary operator (in
addition to the identity operator that corresponds to the vacuum state)? For the
2 )1 we find (see (2.13)) that there is also one primary
primaries of the ALA SO(N
N
operator with dimension 16 for odd N , and two primary operators for even N .
These additional primaries transform in the spinor representation of SO(N 2 )1 .
Can one construct these primaries in terms of the fermionic fields ψ and ψ̃ ?
The situation here is similar to the one in the Ising model. In fact, using the spin
operator σ(z, z̄) or its dual, one indeed gets from the N independent Majorana
N
fermion theories, the dimension 16 and the number of degrees of freedom 2N ,
which are identical to the dimension of the spinor representation.
So far we have shown the free fermion construction of SO(N 2 )1 , namely, of
the ALA at level 1. We would now like to investigate the possibility of having
free fermion realization also to the affine Lie algebra at higher levels. Going
through our previous derivation it is clear that the ALA structure of the OPE
of two currents (3.40), applies to fermions at any representation. For a given
representaion ρ the corresponding level k is determined from the first term on
the right-hand side, namely T r(T a T b ) = 2kδ ab . Now since for a representation ρ,

T r(Tρa Tρb ) = 2D2 (ρ)δ ab where D2 (ρ) is the Dynkin index of the representation,
2 ) at level D2 (ρ).
it is clear that free fermions constitute a realization of SO(N

3.8 Free Dirac fermions and the U(N)
Consider the theory of N Dirac fermions described by the following action,

1 ¯ i + ψ̃ i† ∂ ψ̃i }.
S= d2 z{ψ i† ∂ψ (3.83)
4π
In terms of symmetries, the difference between this theory and the one of a single
Dirac fermion, is that now there is an invariance under U (N ) left holomorphic
and right anti-holomorphic transformations, namely,
ψ → ψ = g(z)ψ ψ̃ → ψ̃ = ḡ(z̄)ψ̃, (3.84)
where g(z), ḡ(z̄) ∈ U (N ). The associated holomorphic currents are given by,
† †
J a = ψ i T a ji ψj J = ψ i ψi , (3.85)
where J is the U (1) current, J a (z) are the SU (N ) currents and T a ji are matrices
in the adjoint of SU (N ), that obey
T r[T a T b ] = δ ab
1
Tija Tkal = δil δj k − δij δk l
a
N

fabc fabd = N δcd . (3.86)
ab
Using the OPEs of the fermions, it is straightforward to realize that the cur-
rents indeed constitute the OPEs that correspond to a U (N ) of level k = 1,
1δ ab f ab J c (w)
J a (z)J b (w) = + c + finite terms
(z − w) 2 (z − w)
1
J(z)J(w) = + finite terms
(z − w)2
J a (z)J(w) = finite terms. (3.87)
Similar to the case of Majorana fermions, the Noether current T is given by,
1 † †
T (z) = T (z)U (1) + T (z)S U (N ) = − [ψ i ∂ψi − ∂ψ i ψi ], (3.88)
2
and can be reexpressed in terms of a Sugawara form,
1 i j
T (z)U (1) = : ψ † ψi ψ † ψj :
4N
1 i j
T (z)S U (N ) = : ψ † T a ψi ψ † T a ψj : . (3.89)
2(N + 1) a

(N )
3.8 Free Dirac fermions and the U 59
Since a Dirac fermion has a c = 1 Virasoro anomaly, it is clear that the theory of
N Dirac fermions has c = N . This is also the outcome of the Virasoro anomaly
associated with the Sugawara form as follows,
N2 − 1
cU (1) + cS U (N ) = 1 + = N. (3.90)
N +1

4
Wess–Zumino–Witten model and coset models
The two-dimensional Wess–Zumino–Witten (WZW) model was introduced in

the seminal paper of Witten [224]. The model makes use of the WZ term that
was introduced by Wess and Zumino in [217]. Sometimes the model is referred
to as the WZWN model, where the N stands for Novikov, who independently
invoked a similar model [170]. Here we follow only [224].
4.1 From free massless scalar theory to the WZW model

Consider the free massless scalar theory that was described in Section 1.2, but
now with X̂(z, z̄) being an angle variable defined in the interval [0, 2π]. The
action of the scalar field can now be re-written in the following form,

1
S = d2 xL = d2 z∂ν X̂∂ ν X̂
8π

1 2 i X̂ ν −i X̂ 1 ¯ −1 ,
= d z∂ν (e )∂ (e )= d2 z∂u∂u (4.1)
8π 4π
where u = ei X̂ (z , z̄ ) is an abelian group element. Recall that the theory is char-

acterized by a Virasoro algebra and an abelian ALA structure. In terms of this
variable the currents J and J¯ can be written as,
J = −iu−1 ∂u = iu∂u−1 , J¯ = −iu−1 ∂u ¯ −1 ,

¯ = iu∂u
with
¯ = ∂ J¯ = 0
∂J
and
T =: (u−1 ∂u)2 :=: (u∂u−1 )2 : .
It is now tempting to replace the abelian u with a non-abelian group element,
u ∈ G, G = SO(N ) or SU (N ), (4.2)
and consider the action,

1 ¯ −1 ],
Ssm = d2 zT r[∂u∂u (4.3)
4π

62 Wess–Zumino–Witten model and coset models
where the trace is taken in fundamental representation so that,

1 ab
T r[T a T b ] = δ .
2
The question here is whether this action admits a similar non-abelian affine Lie
algebra and Virasoro algebra. Let us analyze the equations of motion, symmetries
and the corresponding currents of this action. The variation of the action under
u → u + δu is,

1 ¯ −1 − ∂u∂(u¯ −1 δuu−1 )]
δSsm = d2 zT r[∂(δu)∂u
4π

1
= d2 zT r u−1 δu[u−1 ∂∂(u)
¯ ¯ −1 )u]
− ∂ ∂(u
4π

1
= d2 zT r u−1 δu∂ μ (u−1 ∂μ u)
4π

1
= d2 zT r δuu−1 ∂ μ (u∂μ u−1 ) , (4.4)
4π
were we use δu−1 = −u−1 δuu−1 and ∂u−1 u = −u−1 ∂u, following from
δ(u−1 u) = 0 and ∂(u−1 u) = 0. It is easy to realize that for a constant group
element g the action is invariant under,
u → gu (u−1 → u−1 g −1 ), u → uh (u−1 → h−1 u−1 ), (4.5)
and the currents corresponding to the left and right multiplications take the
form,
1 −1 μ 1
Jμ = u ∂ u J˜μ = − ∂ μ uu−1 . (4.6)
4π 4π
Both currents are conserved. Note that the conservation of one implies the con-
servation of the other. However, unlike the massless free scalar theory, now we
do not have an ALA structure associated with a separate holomorphic and anti-
holomorphic conservation. The latter would have taken the form of JL (z) corre-
sponding to left transformation of the form u → g(z)u and JR (z̄), corresponding
to right transformation of the form u → ug(z̄). In a similar manner one finds
that the energy-momentum tensor,
Tμν ∼ T r[Jμ Jν ] − 1/2 gμν T r[J α Jα ],
and there is only the overall conservation law ∂μ T μν = 0, not ∂T ¯ = ∂ T̄ = 0,

namely not an external product of two Virasoro algebras.
Can we modify the action (4.3) so that it does have the desired ALA and
Virasoro algebraic structure? For that let us consider first the variation of the
action we are looking for. If instead of (4.4) one assumes a variation of the form,

1 −1 −1 ¯
1
δS = 2
d zT r u δu∂(u ∂u) = d2 zT r δuu−1 ∂(u∂u
¯ −1
) , (4.7)
4π 4π

4.1 From free massless scalar theory to the WZW model 63
then the global transformations of (4.5) are elevated into
u → g(z)u u → uh(z̄), (4.8)
with the corresponding currents,

k k −1 ¯
JL ≡ J = ∂uu−1 JR ≡ J¯ = − u ∂u, (4.9)
4π 4π
which have the desired ALA property,
¯ = 0.
∂ J¯ = ∂J (4.10)
Moreover, it can be shown that for an action whose variation takes the form of
(4.7) the energy-momentum takes the form,
T ∼ T r[JJ], T̄ ∼ T r[J¯J],
¯ (4.11)
and hence it also has the appropriate Virasoro behavior.

The next question is obviously what action has a variation of the form (4.7),
and in particular can it be built from Ssm plus an additional term that has
1
the standard form of S̃ = 4π d2 zL. To address this question we rewrite the
variation (4.7) in the form,

1
δS = d2 zT r u−1 δu∂ μ (gμν + μν )(u−1 ∂ ν u)
4π

1
= d2 zT r δuu−1 ∂ μ (gμν − μν )(u∂ ν u−1 ) . (4.12)
4π
Clearly the term, proportional to gμν in both forms, is exactly the variation
δSsm , so that we need to find what action S̃ has a variation that takes the form
of the μν term. It may seem that the action,

1
S̃ = d2 zμν T r[∂ μ u∂ν u−1 ]
4π
does the job, but in fact it vanishes.
It was the proposal of Witten to take for S̃ the so-called WZ action, which for
the present case takes the form of a three-dimensional integral over a ball whose
boundary is an S 2 , which is the two-dimensional space-time,

1
SWZ = d3 σij k T r[(u−1 ∂i u)(u−1 ∂j u)(u−1 ∂k u)], (4.13)
12π
where σi with i = 1, 2, 3 are the coordinates of the ball. Using the fact that
d3 σij k ∂ k (. . .) = d2 σij (. . .), it is straightforward to show that indeed the
variation of (4.13) yields the extra term to change (4.4) to (4.7).
The map u, from the Euclidean space-time that we now take to be S 2 to the
group manifold (Fig. 4.1) can be extended into a map from the ball to the group
manifold. This is based on the fact that the homotopy group associated with

Fig. 4.1. The map between the space-time S 2 and the group manifold.
maps from S 2 to the group space G vanishes, namely, π2 (G) = 01 for any non-
abelian group G.
On the other hand since π3 (G) = Z, there are topologically inequivalent ways
to extend the map u to a map from the ball to the group manifold. This implies
that there is an ambiguity in SWZ and it is well defined only modulo SWZ →
SWZ + 2π. Thus the coefficient of this term must be an integer k, and to have a
variation of the form (4.12) it is clear that the sigma term has to have the same
coefficient.
Let us now summarize. The classical action of the WZW model is,

k ¯ −1 ]
SWZW = d2 zT r[∂u∂u
4π

k
+ d3 σij k T r[(u−1 ∂i u)(u−1 ∂j u)(u−1 ∂k u)]. (4.14)
12π
The variation of this action is given by,

k k
δS = d2 zT r u−1 δu∂(u−1 ∂u)
¯ = d2 zT r δuu−1 ∂(u∂u
¯ −1
) , (4.15)
4π 4π
so that the equation of motion takes the form,
∂(u−1 ∂u)
¯ = ∂(u∂u
¯ −1
) = 0. (4.16)
The solutions of these equations of motion take the form,
u(z, z̄) = u(z)ū(z̄), (4.17)
where clearly u ∈ G, ū ∈ G.
We should state, that the form (4.14), with a term extended to one dimension
higher, follows from general properties. Equations of motion that we want, in
even space-time dimensions, imply a term in the action with one dimension
higher, otherwise the action will involve singular terms, like the introduction of
Dirac strings in the case of elementary monopoles.
The symmetries of the action are the ALA transformations,
u → g(z)u u → uh(z̄), (4.18)
1 π n (G) denotes the group of homotopy classes of maps f: S n → G

4.2 Perturbative conformal invariance 65
and the conformal transformations,
z → f (z) z̄ → f¯(z̄). (4.19)
The ALA currents are,

k k −1 ¯
J =− ∂uu−1 J¯ = u ∂u, (4.20)
4π 4π
and the classical energy-momentum tensor takes the form,
1 1
T = T r[JJ] T̄ = T r[J¯J].
¯ (4.21)
k k
4.2 Perturbative conformal invariance

In the following section it will be shown in an exact way, based on algebraic
properties, that the WZW model is a CFT. Prior to that we present now a
perturbative computation, demonstrating that to a given order indeed the theory
has a vanishing β function. Here we restrict ourselves to the one loop order. Of
course this only serves as a motivation, as the CFT is at a finite coupling, and
so exact demonstration is needed.
The idea is to use the background field method, expanding u around a solution
of the equations of motion which we denote by u0 , so u = u0 eiT π . Substituting
a a
this ansatz into the action 4.14 one finds,

k ¯ −1 ] + 1 ∂μ πa ∂ μ πa
SWZW = d2 z{T r[∂u0 ∂u 0
4π 2
1 μν
+ (η − μν )T r{(u−1 0 ∂μ u0 )[T π , T ∂ν π ]} + O(π )} (4.22)
a a b b 3
2
The one loop renormalization diagram is shown in Fig. 4.2.
The non vanishing contributions are only when both vertices are proportional
to η μν or to μν . The two contributions are the same apart from a sign since
η ρμ ηρ.ν = η μν = −ρμ .ν
ρ , so that the one loop beta function vanishes. Obviously,
this result relates to the choice of the coefficient of the sigma model term versus
the WZ term, with the latter being fixed by topological arguments. The vanishing
of the β function at this stage is an indication that we have chosen the coefficients
in a way that is compatible with conformal invariance.
Fig. 4.2. Calculation of the one loop beta function.

4.3 ALA, Sugawara construction and the Virasoro algebra

An alternative approach to the quantization methods discussed in the previous
section is the ALA and CFT approach. Notice first that under infinitesimal left
transformation δu = (z)u the left current transforms as,
k k
δ J = [∂(u)u−1 − ∂uu−1 ] = (∂ + [, J]), (4.23)
4π 4π
k
which translates into δ J a = 4π [∂
a a b c
+ ifbc J ]. Since the transformation of J is
generated by,
1
δ J a (w) = (z)J(z)J a (w), (4.24)
2πi w
it is easy to realize that the OPE that is compatible with such a transformation
is,
kδ ab ifcab J c
J a (z)J b (w) = + , (4.25)
(z − w)2 (z − w)
which is the OPE associated with the ALA discussed in Section 3.2.
Next we want to determine the conformal properties of u, in particular its
confomal dimension. The classical form of the currents (4.20) is elevated to the
quantum expression via the equations,
κ∂u(z, z̄) =: J a T a u(z, z̄) :, ¯

κ∂u(z, z̄) =: J¯a u(z, z̄)T a :, (4.26)
where κ, which is a renormalized level, will be determined shortly, and the nor-
mal ordering refers as usual to subtracting the singular parts of the product.
Assuming that u is an ALA primary field, the OPE takes the form,
∞
c2
J a (z)T a u(w, w̄) = u(w, w̄) + κ∂u(w, w̄) + (z − w)n −1 T a J−n
a
u(w, w̄),
z−w n =2
(4.27)
where (4.26) was inserted as the (z − w)0 term, and c2 is the quadratic Casimir
a a
operator in the representation of u, a T T = c2 . If we assume that u is a
Virasoro primary as well, then L−1 u = ∂u, so that combined with (4.27) one can
write down the following null vector,
vnull ≡ (J−1
a
T a − κL−1 )u = 0. (4.28)
This is a special case of the degeneracy in the combined ALA–Virasoro algebra

discussed in Section 3.3.
The null vectors obey,
L0 Vnull = (h + 1)Vnull
J0a Vnull = T a Vnull
Ln Vnull = Jna Vnull = 0 for n > 0, (4.29)

where L0 u = hu. For n = 1 the conditions L1 Vnull = J1a Vnull = 0 imply that the
renormalized level and the conformal dimension of u take the form,
1 c2
κ= (C2 + k) h = , (4.30)
2 C2 + k
where C2 is the quadratic Casimir in the adjoint representation, defined as
f acd f bcd = C2 δ ab .
The use of null vectors and the differential equations that determine cor-
relators of primary fields were introduced in the landmark paper of Knizhnik
and Zamolodchikov [143]. An elaboration of the application of these equations
appears in [77]. This direction was further developed by Gepner and Witten
[108].
The WZW has an in-built Sugawara construction. In fact it is very often taken
as the prototype model for this structure. According to the discussion in Section
3.3 the quantum version of the classical energy-momentum tensor (4.21) takes
the form of (3.46),
1
T (z) = : J a (z)J a (z) :, (4.31)
2(k + C2 )
and the Virasoro anomaly of the model is,
k dim G
c= . (4.32)
k + C2
The Sugawara construction is described in [203]. This paper, however, does
not have the correct expression that includes the finite renormalization. This
was done later in the paper of Dashen and Frishman [73].
4.4 Correlation functions of primary fields

Primary fields of theories invariant under ALA were defined and discussed in
Section 3.4. The group element of the WZW theory is an example of a primary
field in the fundamental representation of G × G. Indeed the transformation
properties of u(z, z̄) imply that it has the following OPE with the currents,
ta u(w, w̄) a
u(w, w̄)t
J a (z)u(w, w̄) = − J¯a (z̄)u(w, w̄) = − . (4.33)
z−w z̄ − w̄
Next we would like to compute the n-point correlation function of the group ele-
ment primary field of the WZW model. In Section 3.6 we presented the Knizhnik–
Zamolodchikov equation which determines the correlators of theories invariant
under ALA. We now demonstrate its use in determining the four-point correla-
tion function of the primary field u(z, z̄) of SU (N ) WZW model. We denote this
correlator as,
3 4
G4 = u(z1 , z̄1 )u−1 (z2 , z̄2 )u−1 (z3 , z̄3 )u(z4 , z̄4 ) . (4.34)

Recall from (2.59) that in general due to the conformal Ward identity the four-
point function can be written as,
G4 = [z14 z23 z̄14 z̄23 ]−2h G(x, x̄), (4.35)
where
z12 z34 z̄12 z̄34
x= x̄ = , (4.36)
z14 z32 z̄14 z̄32
with zij = zi − zj and h, the dimension of u, is given by h = 2NN(N−1

2
+k ) .
Now G(x, x̄) can be decomposed into a sum of terms, each one representing a
conformal block, the latter having the form of a product of a holomorphic and
anti-holomorphic function,
G(x, x̄) = Sum of terms of the form G(x)Ḡ(x̄). (4.37)
Since u(z, z̄) is in the fundamental representation of SU (N ), the four-point func-

tion is a product of two fundamentals and two anti-fundamentals, so each term
in the last equation can be decomposed into,
G(x) = I1 G1 + I2 G2 , (4.38)
where I1 and I2 are the SU (N ) invariant factors,
I1 ≡ δm 1 ,m 2 δm 3 ,m 4 I2 ≡ δm 1 ,m 3 δm 2 ,m 4 . (4.39)
If we now substitute this decomposed form of the four-point function into the
Knizhnik–Zamolodchikov equation (3.69) we find,
⎛ ⎞
t a
⊗ t a
⎝ ∂z i + 1 i j ⎠
[z14 z23 ]−4h (I1 G1 + I2 G2 ) = 0. (4.40)
k+N zi − zj
j = i
As was discussed in Section 2.9 conformal invariance allows us to fix three out

of the four points. Using the standard convenient choice,
z1 = x, z2 = 0, z3 = 1, z4 = ∞, (4.41)
and the equation now reads,

1 ta1 ⊗ ta2 1 ta1 ⊗ ta3
∂x + + (I1 G1 + I2 G2 ) = 0, (4.42)
k+N x k+N x−1
After introducing the explicit expressions for the various group theoretical prod-
ucts tai ⊗ taj Ik and projecting to the I1 and I2 factors we get,
−1 (N 2 − 1) G1 G2 1 G1
∂x G1 = + −
k+N N x x N x−1
−1 (N 2 − 1) G2 G1 1 G2
∂x G2 = + − . (4.43)
k+N N x−1 x−1 N x

Extracting G2 from the first equation and plugging it back into the second
equation, the latter translates into a hypergeometric differential equation,
x(1 − x) 2 2 2
[N κ ∂x + A(x)∂x + B(x)]g1 (x) = 0, (4.44)
N2
where κ = k + N and with the following two possible values for A(x), B(x) and
the relation between g1 and G1 as
N (N + κ) N2 N4 − N2 + 2
A(x) = − Nκ B(x) = − ,
x 1−x x(1 − x)
1
G1 = [x(1 − x)] κ N g1+
or
−N (N − κ) N2 2(N 2 − 1)
A(x) = − Nκ B(x) = − ,
x 1−x x(1 − x)
N 2 −1
G1 = x− (1 − x) κ N g1−
1
κN (4.45)
The solutions of the differential equations are the following hypergeometric func-
tions,
1 1 N N −1 N +1 N
g1− = F ,− ,1 − ;x g1+ = F , ,1 + ;x . (4.46)
κ κ κ κ κ κ
In a similar way the solutions for G2 are found, defining an appropriate g2 .

To fully determine the correlator we still have to fix the linear combination
of the solutions. This is done using crossing symmetry, as discussed in Section
2.10.2 The latter implies that,
G(x, x̄) = G(1 − x, 1 − x̄). (4.47)
With parametrization,
(m ) (n )
G(x, x̄) = Ii I¯j Gi,j (x, x̄) Gi,j = ξm n Gi Gj . (4.48)
i,j = 1,2 n ,m =+,−
Crossing symmetry implies that,
Gi,j (x, x̄) = G3−i,3−j (1 − x, 1 − x̄), (4.49)
which follows from the fact that under crossing symmetry I1 ↔ I2 . Single valued-
ness implies that ξ+− = ξ−+ = 0. To obey the crossing symmetry requirement
2 Crossing symmetry to determine correlators of fermions in the fundamental representation

of SU (N ) was used by Dashen and Frishman in [73].

we make use of the following property of hypergeometric functions:
F (a, b, c; x) = A1 F (a, b, a + b − c; 1 − x)
+ A2 (1 − x)c−a−b F (c − a, c − b, c − a − b + 1; 1 − x), (4.50)
where
Γ(c)Γ(c − a − b) Γ(c)Γ(a + b − c)
A1 = A2 = . (4.51)
Γ(c − a)Γ(c − b) Γ(a)Γ(b)
Finally we find that
(−) (−) c2−− − 1 (+) (+)
Gij = Gi (x)Gj (x̄) + Gi (x)Gj (x̄), (4.52)
c2+−
where
Γ( Nκ )Γ(− Nκ ) [Γ( Nκ )]2
c−− = N c+− = −N . (4.53)
Γ( κ1 )Γ(− κ1 ) Γ( N κ+1 )Γ( N κ−1 )
For k = 1 we have c−− = −1, and hence1 the second term in (4.52) vanishes.
Using F N 1+ 1 , − N 1+ 1 , N 1+ 1 ; x = (1 − x) N + 1 the four-point function takes the
form,

1 1 1 ¯ 1 ¯ 1
G(x, x̄) = [xx̄(1 − x)(1 − x̄)] N I1 + I2 I 1 + I2 . (4.54)
x 1−x x 1−x
In Section 6.3 it will be shown that the WZW theory of U (N ) at level k = 1 is
a bosonized equivalent to that of N Dirac fermions controlled by the action,

¯ α + ψ̃ † ∂ ψ̃β ].
Sf = d2 z [ψα† ∂ψ (4.55)
β
In particular the fermion bilinear is equivalent to the group element as,

α
M ũβα (z, z̄) =: ψα (ψ̃ † )β : M (ũ−1 )αβ (z, z̄) =: ψ̃β ψ † :, (4.56)
where M is a mass scale and where ũ denotes the U (N ) group element. In the
theory of free Dirac fermions the four-point function of the fermion bilinears
takes the form,

1 1 1 1
G(x, x̄) = I1 + I2 I¯1 + I¯2 . (4.57)
x 1−x x 1−x
It is easy to see that by converting (4.54) to a similar correlator of U (N ) we find
exactly the same answer. This is done as follows: define the U (N ) group element
to be,
√
ũ(z, z̄) = ei 4π/N γ ϕ(z , z̄ ) u(z, z̄). (4.58)
Then the four-point function of group elements of U (N ) is,

γ2 γ2
G̃(x, x̄) = M −2 N [xx̄(1 − x)(1 − x̄)]− N G(x, x̄). (4.59)

4.5 WZW models with boundaries – D branes 71
For γ = 1 we observe that indeed the correlator is identical to that of the

fermion bilinears. For arbitrary γ the correlator corresponds to that of fermion
bilinears of the Thirring model defined by,

γ2 − 1 ¯
S = Sf + d2 zJ(z)J(z̄). (4.60)
2γ 2
This generalized bosonization will also be addressed in Section 6.2.
4.5 WZW models with boundaries – D branes

The WZW model described in Section 4.1 was shown to be based on a map
from Σ, a compact two-dimensional manifold, in paticular an S 2 , into a group
manifold G. Let us now study the case where Σ has boundaries. For concreteness
we take it to be the upper half-plane. In the bulk the theory is invariant under
the holomorphic and anti-holomorphic ALA (4.18) and there is corresponding
holomorphic and anti-holomorphic conservation of the associated currents (4.9).
On the boundary the two types of modes mix, the symmetry is reduced to,
u → g(τ )ug(τ )−1 , (4.61)
where τ denotes the coordinate on the boundary, and accordingly there is a

relation between JL and JR ,
¯
J(z) = Ωaut J(z̄) at z = z̄, (4.62)
where Ωaut is an automorphism of the ALA.

The notion of boundary conformal field theory was introduced in [58]. The
gluing conditions used for D branes in the WZW model were introduced in [135].
From the many papers that have been written on the subject we have chosen to
describe it following [11] and [85].
Let us first address the simplest case of a level k SUˆ(2) WZW, for which
Ω = −1, and then later discuss the general case. In terms of ∂t , ∂x and the
adjoint action of G on its Lie algebra,
Ad(g)u = gug −1 , (4.63)
the gluing condition reads,
(1 − Ad(u))u−1 ∂t u = (1 + Ad(u))u−1 ∂x u. (4.64)
The tangent space to the group G at the point u can be split into Tu G =
|| ||
Tu⊥ G ⊕ Tu G, where Tu G consists of vectors tangential to the orbit of Ad through
u. On Tu G, (1 − Ad(u)) = 0 and (1 + Ad(u)) = 2, so that (u−1 ∂x u)⊥ = 0 and
⊥
the corresponding D branes, namely the submanifolds where the condition (4.62)
is obeyed, coincide with the conjugacy classes. In the case that (1 − Ad(u)) is

invertible, (4.64) can be written as,

1 + Ad(u) −1
u−1 ∂t u = u ∂x u. (4.65)
1 − Ad(u)
We can now define a two form on the conjugacy class as,
k 1 + Ad(u) −1
ω= u−1 du u du . (4.66)
8π 1 − Ad(u)
Applying an exterior derivative to this form we find,
k
dω = T r(duu−1 )3 , (4.67)
12π
namely, it is not closed. The submanifold D ⊂ G on which the WZ term is exact,
(W Z) = dω defines a D brane in G. There is a further restriction which follows
from reasoning similar to that discussed in Section 4.1.
Consider the wave functional Ψ(u(x)) on the space of closed loops u(x) in
some conjugacy class C. The latter, for the group manifold SU (2), are typically
two-spheres so that C can be constructed in two different ways, and hence there
is an ambiguity in the phase of the wave functional. It can be shown that the
phase can take the values 2πj with j = 1, . . . , k − 1. Thus there are k − 1 two-
dimensional conjugacy classes or D2 branes for the k level SUˆ(2) WZW model.
In addition there are two D0 branes associated with the two points ±e, where e
is the identity on the group space.
To address the issue of the conjugacy classes in other groups it is convenient
to rewrite the two form (4.66) as,
k
ω= T r[k̃ −1 dk̃hk̃ −1 dk̃h−1 ], (4.68)
8π
for u that belongs to the conjugacy class,
ChG = {u ∈ G|u = k̃hk̃ −1 }. (4.69)
The WZW action that corresponds to the map from the two manifold with a
boundary to the group space can be constructed as follows. Instead of considering
the map from Σ that has a boundary, we take it from Σ ∪ D where D is an
auxilliary disc that closes the hole in Σ, having a common boundary with it. The
disc is mapped into the conjugacy class allowing for its boundary (4.69). The
WZW action is now written as,

S = SWZW + w, (4.70)
D
where SWZW is the ordinary WZW action with a three-dimensional WZ term

defined now on a ball whose boundary is Σ ∪ D. It can be explicitly checked
that the new WZW action is invariant under (4.18). Similar to the topologically
distinct D2 branes of the k level SUˆ(2) WZW model, there are different embed-
dings of the disc in a conjugacy class in a general group manifold G. This is
related to the second homotopy group of the conjugacy class, which in general

4.6 G/H coset models 73
is non-trivial. The group element k of (4.69) is defined modulo a right multipli-

cation with any element that commutes with h, and the group of such elements
is isomorphic to the Cartan torus TG generated by the generators in the Cartan
subalgebra. Thus the conjugacy classes can be described as TGG and the second
homotopy group reads,
Π2 (ChG ) = Π1 (TG ) (4.71)
For a rank r algebra of G, the D2 will be characterized by an r-dimensional

vector in the coroot lattice of G. Namely, if two embeddings given by khk −1 and
−1
k hk then on the boundary of the world sheet they have to be related as,
k(τ )k (τ )−1 = t(τ ), (4.72)
where t(τ ) is an element of the subgroup isomorphic to TG which commutes

with h. This relation determines a mapping from the boundary to TG . Since the
latter is Rr modulo 2π × (coroot lattice), every such mapping belongs to the
topological sector parameterized by a vector in the coroot lattice describing
the winding of the boundary circle on the torus TG . This lattice vector deter-
mines via (4.71) the element of Π2 (ChG ) corresponding to the union of the two
embeddings. For the group element h ∈ TG of the form h = eiθ ·H , where H are
the Cartan subalgebra generators, the change in the action resulting from the
topological change of the embedding is ΔS = k(θ · s), where s is a coroot lattice
vector. Consistency of the model then implies the condition,
2πZ
θ·α∈ . (4.73)
k
This generalizes the condition that led to the set of k − 1 D2 branes for level
k SUˆ(2). It implies that θ should be 2π k × (weight lattice vector). Since a point
in TG is defined modulo 2π × (coroot lattice), the allowed conjugacy classes cor-
respond to points in the weight lattice divided by k modulo the coroot lattice.
This is also the characterization of the primary fields of the corresponding WZW
model.
4.6 G/H coset models

The concept of coset models dates back to [110] or in fact even to as early as [27].
A Lagrangian formulation in terms of a gauged WZW model was introduced for
instance in [12]. Here we follow the review of [111].
The WZW models constitute a large class of conformal field theories which are
invariant under ALA. An even larger class of CFTs can be constructed by taking
the quotient of two WZW models. Consider an ALA ĝ at level k and a subalgebra
of it ĥ at level kh . We denote the currents associated with the former as J a and
with the latter Jha h where a = 1, . . . , dim G and ah = 1, . . . , dim H. The currents

associated with the subalgebra ĥ can be expressed as a linear combination of ĝ

as,

Jha h = ma h a J a . (4.74)
a
a
Using the commutator of J and the corresponding generator of the Virasoro
algebra constructed via the Sugawara construction,
[Lm , Jna ] = −mJm

a
+n , (4.75)
it follows that,
[Lm , Jha h n ] = −mJha h m +n . (4.76)
It is also obvious that a similar relation holds with Lhm which is the Virasoro
generator built by a Sugawara construction from the currents of ĥ, namely,
[Lhm , Jha h n ] = −mJha h m +n . (4.77)
Thus it follows that,
[Lm − Lhm , Jha h ] = 0. (4.78)
Since Lhm is a bilinear of Jha h it follows that,
[Lm − Lhm , Lhn ] = 0 → [Lm , Lhn ] = [Lhm , Lhn ]. (4.79)
We can now define,
L(g
m
/h)
≡ Lm − Lhm . (4.80)
The algebra of these coset generators is a Virasoro algebra, as follows from,
[L(g
m
/h)
, L(g
n
/h)
] = [Lm , Ln ] − [Lhm , Lhn ]
(g /h) (m3 − m)
= (m − n)Lm +n + [c(ĝk ) − c(ĥk h )] δm +n ,0 . (4.81)
12
(g /h)
Thus we have just found that the Virasoro generators of the coset Lm obey a
Virasoro algebra with a central charge of
kdimg kh dimh
c= − . (4.82)
k + C2 (g) kh + C2 (h)
A special class of coset models are the diagonal coset models ĝ ⊕ĝ ĝ . The gen-
erators of the coset in this case are given by Jha = J(1)a a
+ J(2) , namely the sum
of the generators of each copy. It thus follows that the level of the coset must
a a
be the sum of the two levels since [J(1) , J(2) ] = 0. The coset therefore takes the
form,
ĝk 1 ⊕ ĝk 2
,
ĝk 1 +k 2

and its corresponding central charge is given by,

k1 k2 k1 + k2
c = dim g + − . (4.83)
k1 + C2 (g) k2 + C2 (g) k1 + k2 + C2 (g)
Consider the case where g = SU (2) and the coset is,
SUˆ(2)k ⊕ SUˆ(2)1
, (4.84)
SUˆ(2)k + 1
with the central charge,
3k 3(k + 1) 6 6
c= +1− =1− =1− , (4.85)
k+2 k+3 (k + 2)(k + 3) p(p + 1)
where in the last step we introduced p = k + 2 ≥ 3. This is exactly the central
term of the unitary minimal models discussed in Section 2.7. In fact one can
show that this is indeed an equivalence in the sense that the characters of the
minimal models are the same as those of the coset model.
4.7 G/G coset models

The concept of the G/G model was introduced in [200] and [227]. Our description
of the G/G model and in particular its BRST analysis follows [9] and [8].
A special class of the G/H models is the case where H = G, namely where we
gauge the maximal anomaly-free diagonal group. Using the gauging procedure
that will be discussed in Section 9.3.1 the classical action takes the form,

k
Sk (g, Aμ ) = Sk (g) − d2 zT r(g −1 ∂g Āz̄ + g ∂g
¯ −1 Az − Āz̄ g −1 Az g + Az Āz̄ ).
2π
(4.86)
Next we introduce the following parameterization of the gauge fields, Az =
ih−1 ∂z h, Āz̄ = ih∗ ∂z h∗−1 where h(z) ∈ Gc . In Section 15 we will elaborate more
about this formulation. The action then reads,
Sk (g, A) = Sk (g) − Sk (hh∗ ). (4.87)
The Jacobian of the change of variables introduces a dimension (1, 0) system of
anticommuting ghosts χ and ρ in the adjoint representation of the group. The
quantum action thus takes the form of,

Sk (g, h, ρ, χ) = Sk (g) − Sk + 2C G (hh∗ ) − i d2 zT r[ρ∂¯χ̄ + c.c], (4.88)
where CG is the second Casimir of the adjoint representation.3 The second term
can be viewed as S−(k + 2C G ) (h). Since the Hilbert space of the model decomposes
into holomorphic and anti-holomorphic sectors we restrict our discussion only to
the former.
3 This was C 2 (g) in the previous section; notation has changed according to the literature.

There are three sets of holomorphic G transformations which leave (4.88)

invariant,
δJ g = i[(z), g] δI h = i[(z), h]
a
δJ ( g h ) χ = a b c
ifbc χ δJ ( g h ) ρa = −ifbc
a b c
ρ , (4.89)
a
with in the algebra of G. The corresponding currents J a , I a and J (g h) =
a
ifbc χb ρc satisfy the G ALA with the levels k,−(k + 2cG ) and 2cG , respectively.
a
We now define J (tot)
a a
J (tot) = J a + I a + J (g h) = J a + I a + ifbc
a
χb ρc , (4.90)
which obeys an affine Lie algebra of level,
k (tot) = k − (k + 2cG ) + 2cG = 0. (4.91)
The energy-momentum tensor T (z) is a sum of Sugawara terms of the J a and
I a currents and the usual contribution of a (1, 0) ghost system, namely,
1 1
T (z) = : JaJa : − : I a I a : +ρa ∂χa . (4.92)
k + cG k + cG
The corresponding Virasoro central charge vanishes,4
kdG (k + 2cG )dG
c(tot) = − − 2dG = 0 (4.93)
k + cG −(k + 2cG ) + cG
The symmetry structure of the model is in fact richer. It is easy to realize
that there are also two odd conserved currents, a dimension one current which is
the BRST current J (BRST) and a dimension two operator G. These holomorphic
symmetry generators are given by

(BRST) a a 1 (g h) a
J = χa J + I + J ,
2
1
G= ρ [J a − I a ]. (4.94)
k + cG a
The reason we denote the dimension one current as a BRST current is that one
can express both T (z), J (tot) a and J (BRST) itself in terms of its corresponding
charge Q = J (BRST) (z) as follows,
T (z) = {Q, G(z)},
(tot) a
J (z) = {Q, ρa },
J (BRST) = {Q, j # (z)}, (4.95)
where j # is the ghost number current.
The fact that T (z) is BRST exact namely T (z) = {Q, G(z)} and that the total
Virasoro anomaly vanishes, are indications that the GG model is a topological
quantum field theory. These theories which were found to be very useful in
dealing with various issues in physics and mathematics are beyond the scope
4 dG is what we called dim g in the previous section; changed according to the literature.

of this book. We thus do not discuss here the topological quantum field theory
aspects of the G
G models.
By construction of the BRST procedure the space of physical states of a
G model is given by the cohomology of Q. That is to say, a physical state |phys>
G
has to be closed under Q, namely Q|phys> = 0 and not exact, namely |phys> =
Q|state> where |state> is any other state. It can be shown that taking the trace
over those states one finds the torus partition function of the G G model which
is based on the decomposition into WZW characters, discussed in Section 3.5.
The torus partition function can be expressed as

∗
−r
Z G = cτ2 duZ g (τ, u)Z hh (τ, u)Z g h (τ, u), (4.96)
G
where du is the measure over the flat gauge connections on the torus and r is
the rank of G; Z g (τ, u) is the torus partition function of the Gk WZW model,
−c
Z g (τ, u) = (q q̄) 2 4 χk ,λ L (τ, u)χ̄k ,λ R (τ, u)Nλ R ,λ L , (4.97)
λ L ,λ R
where q = e2iπτ , λR , λL denote the Gk highest weights, and for the diagonal
invariant Nλ R ,λ L = δλ R ,λ L . The character can be written as,
Mk ,λ (τ, u)
χk ,λ (τ, u) = , (4.98)
M0,0 (τ, u)
∗
with Mk ,λ defined explicitly for the SU (2) case below. Z hh (τ, u) in (4.96) is
c
the contribution of h ∈ GG at level k + 2CG or equivalently h ∈ G at level −(k +
∗ c
2cG ). This takes the form Z hh (τ, u) ∼ |M0,0 (τ, u)|−2 indicating that GG contains
just one conformal block. It is straightforward to calculate Z g h (τ, u), the ghost
contribution to the partiton function Z g h (τ, u) ∼ |M0,0 (τ, u)|4 . Thus there is a
cancellation of the |M0,0 (τ, u)| factors and the resulting character is given by
the numerator of the character of the “matter” sector. In the G G model it is
Mk ,λ . This cancellation property leads to an index interpretation for Mk ,λ . For
G = SU (2) it was found that one can express,
∞
j+ 1
* +
2 2 1
Mk ,j (τ, θ) = q (k + 2)(l+ ( k + 2 ) ) sin πθ (k + 2)l + j + , (4.99)
2
l=−∞
as
2
1 ( j( k++122)) iπθ (j + 1 ) iπθ Jˆ(0tot )
Mk ,j (τ, θ) = q e 2 T r[(−)Ĝ q L̂ 0 e ], (4.100)
2i
where θ = Reu, Ĝ is the ghost number, L̂0 is the excitation level and Jˆ(tot) 0
is
0
the J(tot) eigenvalue of the excitation. Note that Mk ,j (τ, θ) is obtained from the
torus Mk ,j (τ, u) by restricting to just one angle. This amounts to considering the
propagation along a cylinder rather than around the torus. As long as we are
interested in the spectrum it is sufficient to consider Mk ,j (τ, θ). This index inter-
pretation enables us to read important information about the physical spectrum

from (4.99). For a positive integer k, 2j = 0, . . . , k. Hence there are k + 1 zero

ghost number primary states which correspond to the first term in the q expan-
sion of the different Mk ,j s, i.e. the term corresponding to l = 0 with L̂0 = j (j +1)
k +2 .
On each of these states there is a whole tower of descendant states correponding
to the higher terms in the q expansion. For further discussion of the G/G model
the reader is referred to [9].

5
Solitons and two-dimensional integrable models
5.1 Introduction
In the previous chapters we have addressed 2D field theories with no scale. As
we discussed in Chapter 2, one cannot define an S-matrix for such theories.
Generically physical systems are characterized by certain energy scales and the
notion of S-matrix plays an important role. It is thus time to move forward
and examine non-conformal field theories. Again we start our journey with the
theory of a free scalar field, but now a massive one. We then move on and discuss
interacting theories equipped with infinite numbers of conserved charges, the so-
called integrable models, that resemble the free massive theory in a way that will
be explained below.
5.2 From the theory of a massive free scalar field to

integrable models
The classical action of a massive free scalar field is obviously the action of a
massless scalar field with an additional mass term,

1 μ 1 2 2
S = d x ∂ φ∂μ φ − m φ ,
2
(5.1)
2 2
where m is the mass scale which momentarily will be shown to be the mass of
the particle associated with the field φ. Unlike the analysis of CFTs there is no
advantage here to the use of complex coordinates, so we will use real ones.
The corresponding equation of motion,
∂ μ ∂μ φ + m2 φ = 0, (5.2)
is solved for the case of uncompactified space-time by the following Fourier trans-
form,

dk 1 1 −ik ·x
0 1
φ(x , x ) = √ √ a(k )e + a† (k 1 )eik ·x , (5.3)
2π k 0
where (k 0 )2 − (k 1 )2 = m2 .
A dramatic difference between the massless field discussed in Chapter 2, and
the massive one we discuss here, shows up when analyzing the symmetries of the
system.

80 Solitons and two-dimensional integrable models
The only transformations that leave the action invariant are the ISO(1, 1)
Poincare transformations, namely, the space and time translations and a single
Lorentz transformation. These are,
x0 → x0 + a0 x1 → x1 + a1
x0 → x0 + a01 x1 x1 → x1 + a10 x0 , (5.4)
where the transformation parameters are constants and a01 = a10 . The fact that
the parameters are constants, and not holomorphic and anti-holomorphic func-
tions of the complex coordinates, has a tremendous impact, since it implies the
absence of the powerful infinite-dimensional Virasoro algebra.
The corresponding Noether currents associated with the Poincare transforma-
tions are,
Tμν = ∂μ φ∂ν φ − gμν L,
μ μ01
JLor ≡ JLor = ρν T μρ xν . (5.5)
However, since the space-time is two dimensional, there is an additional conserved
current, the so-called topological current,
μ
Jtop = μν ∂ν φ, (5.6)
μ
which is conserved regardless of the equations of motion, since obviously = ∂μ Jtop
μν ∂μ ∂ν φ = 0. In fact this current is conserved for any interacting scalar field in
2d, and as we will see later on it plays an important role in the analysis of soliton
solutions of integrable models.
The theory of a free massive scalar field, as well as other scalar theories that
will be addressed in this chapter, are obviously invariant under the discrete
symmetry of,
φ → −φ. (5.7)
The canonical quantization was described in Section 1.6. The normal ordered
Hamiltonian and momentum expressed in terms of the creation and annihilation
operators take the form,

H = dk 1 (k 1 )2 + m2 a† (k 1 )a(k 1 ) P = dk 1 k 1 a† (k 1 )a(k 1 ). (5.8)
The state a† (k 1 )|0> is characterized by,

Pa† (p1 )|0> = p1 a† (p1 )|0> Ha† (p)|0> = (p1 )2 + m2 a† (p1 )|0>, (5.9)
and hence it is interpreted as a single free massive relativistic particle.
For the case of free single particles, the momentum and Hamiltonian can be
generalized to an infinite set of conserved charges, like Qn ≡ P n as,
Qn a† (p)|0> = (p1 )n a† (p)|0>, (5.10)
and similarly for powers of p0 .

The conserved charge Qn can also be represented as an integral in space. For

odd n = 2k + 1,

Q2k + 1 = dx[φ(2k + 1) (t, x)φ̇(t, x) + hermitian conjugate], (5.11)
where φ(n ) is n derivatives of space on φ, t is x0 and x is x1 . For even n it is a

bit more complicated, but can be evaluated similarly. Note that the expression
is local.
We elevate the field theory associated with a free massive scalar particle into
a non-trivial interacting integrable model by replacing the mass term with a
potential for the scalar field. It will be shown that identifying in such interacting
field theories, an infinite set of conserved charges similar to the one of the free
theory, will be a key ingredient in constructing integrable models. This will be
A more general construction of integrable models is based on perturbing con-
formal field theories, which were discussed in Chapter 3, with relevant primary
fields, namely, those that have conformal dimension Δ + Δ̄ < 2.1 This class of
models, which will include in particular the integrable minimal models, will be
A very basic notion in scalar theories with interacting potential is the solitonic
classical configurations, which will be the topic of the next section.
5.3 Classical solitons

We now let the massive particles interact with each other. The interaction is
introduced in the form of a potential added to the Lagrangian of the free scalar
field theory. Our first task in analyzing this type of field theories is to determine
the solutions of the classical equations of motion. We start first with solitons,
which are static solutions of finite energy, and then move on to time-dependent
solutions.
The “classical” material about solitons is described in great detail in several
books, in particular in [182], [66] and [183]. For “nontopological solitons” see
[149].
Consider a two-dimensional scalar field described by the following action,

1
S = d2 x ∂μ φ∂ μ φ − V (φ)
2

1 1
= d2 x (φ̇)2 − (φ )2 − V (φ) , (5.12)
2 2
1 So far in the context of conformal field theory we have denoted the conformal dimension by
h. Here in the chapter on integrable models it will be denoted by Δ.

where ˙ and refer to time and space derivatives and V (φ) is a positive semi-
definite function of φ. The corresponding equation of motion is given by,
∂μ ∂ μ φ + ∂φ V (φ) = φ̈ − φ + ∂φ V (φ) = 0. (5.13)
The energy associated with a given configuration of φ is,

1 1
E = dx (φ̇)2 + (φ )2 + V (φ) . (5.14)
2 2
Let us assume that the potential has a set of N absolute minima at which it
vanishes, namely V (φi ) = 0 for i = 1, . . . , N . If φi are constants independent of
space-time, then the corresponding energy vanishes, and in fact E(φ) = 0 if and
only if φ(x, t) = φi .
Static solutions of the equation of motion are determined by,
φ − ∂φ V (φ) = 0. (5.15)

Solitons which have finite energy, must have φ and V (φ) vanish rapidly enough
at ±∞, and thus must approach asymptotically one of the configurations φi that
minimizes the potential, namely
limx→∞ φ(x) → φi , limx→−∞ φ(x) → φj . (5.16)
Solving (5.15) is equivalent to solving a mechanical system where x becomes

the time, φ the coordinate of a point particle of a unit mass subjected to a
potential −V (φ), and Em ech = 12 φ2 − V (φ) is the conserved energy of the system.
The boundary conditions where at x → ±∞ V (φ) → 0 and φ → 0 implies that
Em ech = 0. The energy of the field theory (5.14) translates into the action of the
mechanical system. The particle trajectory is therefore characterized by having
finite action and vanishing mechanical energy. The virial theorem for the particle
system has the form,
1 2
(φ ) = V (φ), (5.17)
2
which is also easily derivable in field theory language by multiplying (5.15) by
φ , integrating over x and using the boundary conditions.
From the mechanical analog it is clear that:
(i) there is no non-trivial solution for a potential with a single minimum.

(ii) For a potential with n minima there are 2(n − 1) solutions associated with
trajectories starting at x → −∞ at φi and ending at x → ∞ at φi+1 and
vice versa. Trajectories where instead the particle ends at φj > i+1 or back to
n
φi are impossible, since all the derivatives ddx nφ vanish at φi+1 so the particle
that gets to this point will not be able to leave it.

The equation of motion (5.15) has solutions of the form,

φ(x)
dφ̃
x − x0 = ± , (5.18)
φ(x 0 ) 2V (φ̃)
where x0 the integration constant is any arbitrary point where the field has
the value of φ(x0 ). The integral is non-singular apart from the end-points since
everywhere else V (φ) is positive.
Classical solitons of λφ4 theory

Let us now demonstrate the general features of solitons discussed in the previous
section with the prototype model of the a potential with a quartic interaction.
Consider the potential,
2
1 m2
V (φ) = λ φ2 − , (5.19)
4 λ
which has two minima at φ = ± √mλ and is obviously invariant under φ → −φ.
Substituting this potential into (5.18) and inverting it one finds when setting
φ(x0 ) = 0 the following two possible solutions,

m m
φ(x) = ± √ tanh √ (x − x0 ) , (5.20)
λ 2
which corresponds to either starting at φ = − √mλ and ending at φ = √mλ , or vice
versa. The former will be called a “kink” and the latter an “anti-kink”. The
invariance under φ → −φ and parity transformation are easily realized in the
kink anti-kink system, namely, φkink (x) = −φanti−kink (x), and for x0 = 0 also
= −φkink (−x) (otherwise reflect around x0 ).
The energy density of the kink solution is given by

1 m4 m
(x) = (φ )2 + V (φ) = sech4 √ (x − x0 ) . (5.21)
2 2λ 2
The total classical energy, which is refered to as the classical mass of the kink is
(as our soliton is like a particle at rest),
∞ √
2 2 m3
Mcl = dx(x) = . (5.22)
−∞ 3 λ
Classical solitons of sine-Gordon theory

The sine-Gordon model, which will serve as a prototype model throughout this
chapter is defined by the action given in (5.12) with the potential,
√
m4 λ
V (φ) = − cos φ −1 . (5.23)
λ m

Later in Section 5.4 we will adopt a different convention where,

√
λ m3
= β μ2 = √ . (5.24)
m λ
In terms of this parametrization the potential reads,
μ2
V (φ) = − [cos(βφ) − 1]. (5.25)
β
The potential has an infinite set of discrete vacua at φk = 2πk √mλ and again it
is invariant under φ → −φ. As for the φ4 case, here as well the integral in (5.18)
can be solved analytically. For the soliton that goes from 0 to √mλ 2π and vice
versa for the anti-soliton, and choosing φ(x0 ) = √mλ π, we get,
m
φ(x) = 4 √ tan−1 [e±m (x−x 0 ) ]. (5.26)
λ
Adding √mλ 2nπ to this, gives a soliton that goes from √mλ 2nπ to √mλ 2(n + 1)π,
and vice versa for the anti-soliton. The soliton has a topological charge associated
with (5.6) of Q = 1, the anti-soliton Q = −1.
Substituting the explicit expression of the soliton profile (5.26) into the expres-
sion for the energy one finds that the mass of the SG soliton is
8m3 8m
MSGsol = = 2. (5.27)
λ β
Classical stability of the solitons

So far we have shown that scalar field theories with degenerate vacua admit
soliton solutions. Let us now address the question of whether these solutions are
stable against small time-dependent perturbation. Consider the field configura-
tion,
φ(x, t) = φsol (x) + δφ(x, t), (5.28)
where φsol (x) is a time-independent soliton solution and δ(x, t) is a small pertur-
bation. Substituting this configuration into the equation of motion and retaining
only linear terms in the perturbation we get,
∂ μ ∂μ δφ + V (φsol )δφ = 0. (5.29)
Since the equation is invariant under time translation, we express the perturba-
tion as a superposition of normal modes in the following form,

δφ(x, t) = Re[an eiw n t δn (x)]. (5.30)
n
The normal modes obey the equation,

d2 δn
− + V (φsol )δn = wn2 δn , (5.31)
dx2

which is in fact a one-dimensional Schrodinger equation with V (φsol ) as a poten-

tial. If this equation has eigenmodes with negative eigenvalues, the soliton is
unstable.
It is easy to construct one eigenmode. Since the soliton is invariant under space
translation φsol (x) → φsol (x + a), δ0 = dφ sdx
o l (x)
is an eigenmode with a vanishing
eigenvalue. Now since the soliton is a monotonic function of x, δo does not have
nodes. A theorem about a one-dimensional Schrodinger equation tells us that
the eigenmode with no nodes has the lowest eigenvalue and hence there are no
negative modes and the soliton for any V (φ) is indeed stable.
The topological charge

Any two-dimensional scalar field theory in two dimensions admits the topological
μ
current (5.6), Jtop = eμν ∂ν φ. Thus, the following difference is a conserved charge,

Qtop = dxφ = [φ(t, +∞) − φ(t, −∞)] ≡ φ+ − φ− . (5.32)
Often one refers to φ± as the topological indices. In fact for theories with a
potential that has a discrete number (finite or infinite) of vacua, non-singular
field configurations of finite energy have both φ+ and φ− separately conserved.
This results from the following argument. Finite energy implies that both φ+
and φ− are at absolute minima of the potential. Now since the non-singular
configurations are continuous in time, and the potential has a set of discrete
(finite or infinite) vacua, φ(t, ∞) must be stationary at φ+ , or ∂0 φ(t, ±∞) =
∂0 φ± = 0, namely the indices are conserved.
In fact this conservation can be used to show the existence of non-dissipative
solutions. For instance in the φ4 theory we can show that a configuration with
φ+ = −φ− is non dissipative. By continuity in x there must be, for any t, some x
for which φ = 0. At this point T00 ≥ V (0) and since the definition of a dissipative
solution is that the limt→∞ maxx T00 = 0 it is clear that it is non-dissipative.
Similar arguments hold for other cases of solitons.
Thus, one can divide the space of finite-energy non-singular solutions into
topologically disconnected sub-spaces that are characterized by the two indices
φ± . Such a sub-space cannot be continuously deformed into another one unless
the finite energy condition is violated. For instance, in the φ4 theory, the potential
has two minima so that φ+ = mλ and φ− = − mλ . Hence, there are four subspaces
(−, +), (+, −), (−, −), (+, +) associated with the soliton, the anti-soliton and the
two trivial constant vacuum solutions. For the sine-Gordon the solitons belong
to the subspaces characterized by φ− = 2πn √mλ and φ+ = 2π(n + 1) √mλ .
Obviously, non-trivial topological charges require multiple vacua. The latter
situation occurs if and only if there is a spontaneous breaking of a symmetry.
For instance in φ4 and sine-Gordon it is the discrete φ → −φ symmetry which
is broken.

Derrick’s theorem
Consider a scalar field theory in D + 1 space-time dimensions described by the
Lagrangian density,
1
L = ∂ μ φ∂μ φ − V (φ), (5.33)
2
where the potential V (φ) is non-negative and vanishes at its minima. The the-
orem states that for D ≥ 2 the only non-singular time-independent solutions of
finite energy are the vacua.
Let us denote by φ(x) a time-independent solution of the equation of motion.
We now introduce a one-parameter family of field configurations defined as,
φ(λ, x) = φ(λx), (5.34)
where λ is a positive real number. The energy associated with the configuration
φ(λ, x) is,

1
E(λ) = λ−D dD x λ2 (∇φ) + V (φ) .
2
(5.35)
2
By Hamilton’s principle the energy as a function of λ is stationary at λ = 1 so
that,

1 2
dD x (D − 2)(∇φ) + DV (φ) = 0. (5.36)
2
For D > 2 the two terms in the integral have to vanish separately, which occurs
only for the vacua. For D = 2, the potential term has to vanish, which again
occurs only for the vacua. This proves the theorem.
The following remarks are in order:
(i) Derrick’s theorem applies only to time-independent configurations.
(ii) It applies to field theories with only scalar fields. Once one introduces
additional fields like gauge fields or fermions the theorem is not valid (see
Section 20.3).
5.4 Breathers or “doublets”

So far we have discussed only time-independent solutions of the equations of
motion. A natural question to ask is whether the equations also admit exact
time-dependent solutions. Another question, seemingly unrelated, is that of the
interactions between solitons and between solitons and anti-solitons. We will see
shortly that these two puzzles are in fact related. We now proceed to examine
these questions in the laboratory of the sine-Gordon model.
The following periodic configuration,

4 η sin(wt)
φ(x, t) = tan−1 , (5.37)
β cosh(ηwx)

5.4 Breathers or “doublets” 87
√ 2 2
(m −w )
where η = w , with w ≤ m, is a solution of the equation of motion (5.13).
We will show now that this solution is related to a bound state of a soliton and
anti-soliton.
Consider first the simple case of small w, namely w m. For positive sin(wt)
and finite x, the argument of tan−1 is very large, and thus φ ∼ 2π β . When x
approaches −∞ we can approximate φ(x, t) as,

4 2m
φ(x, t) ∼ tan−1 exp mx + ln sin(wt) , (5.38)
β w
which looks like a soliton to the left. Similarly, it looks like an anti-soliton to
the right. The soliton and anti-soliton move further apart as sin(wt) increases to
one, and then when sin(wt) decreases they approach each other. As sin(wt) → 0
the approximation that lead to (5.38) is no longer valid, in accordance with the
fact that in this region the soliton and anti-soliton are on top of each other.
A similar discussion applies also for negative sin(wt). It is thus clear that the
solution (5.37) describes an oscillation of a soliton anti-soliton pair around their
common center of mass.
Revealing a bound state solution implies that the system must be attractive
at least in a certain region of the “coupling constant”. Indeed if one uses the
coupling constant,
πβ 2
ξ= , (5.39)
8π − β 2
then
∞ > ξ > π repulsive interaction

ξ = π free particle
π > ξ > 0 attractive interaction (5.40)
As will be clarified in the next chapter, the case of ξ = π corresponds to a free

massive Dirac fermion. This will be further discussed in Section 6.1 as a bosoniza-
tion of the free massive Dirac fermion. Also, the attractive region corresponds
to positive coupling of a four-fermion interaction, namely attraction, while the
repulsive region above corresponds to a negative coupling four-fermion. The case
of negative ξ leads to no ground state.
If indeed the breather describes a bound state, it has to have a mass which
is smaller than twice the mass of the soliton. It is easy to compute the classical
energy associated with (5.37) at t = 0, since both the potential and the φ vanish.
Thus,
5 w 2
(clas) 1 2 8(ηw)2 1
Ebreather = dx (φ̇) = dx = 2M sol 1 − ,
2 β2 cosh2 (ηwx) m
(5.41)

(a) (b) (c)

Fig. 5.1. Normal-order correction of the coupling (a), a correction to m2 (b)
and the lowest-order contribution to δm2 (c).
where Msol is the mass of the soliton. This verifies the existence of a binding
energy since the mass of the breather is less than twice the mass of the soliton.
In Section 5.5.1 the quantum description of the bound states will be addressed,
and their scattering processes in Section 5.6.
5.5 Quantum solitons

The quantization of the soliton and breather was worked out in [74].
The classical mass of the soliton is M = 8m
β 2 , as in (5.27). Quantum mechani-
cally this mass is corrected due to quantum fluctuations. We parametrize these
space-time-dependent fluctuations as,

φs → φs + Re[eiw n t δn (x)]. (5.42)
n
The normal modes δn (x) obey the equation,
[−∂x2 + m2 cos(βφs )]δn (x) = wn2 δn (x). (5.43)
The leading order of the quantum mass is then given by

1
Mquantum = Ms + wn + Vct (φs ), (5.44)
2 n
where Vct (φs ) is a counter term that one has to add to the Lagrangian.
In two dimensions the only source of UV divergences in any order of pertur-
bation theory are diagrams that contain a loop consisting of a single internal
line. Stated differently, UV divergences are due to the fact that the action is not
normal ordered. The corresponding diagrams are depicted in Figure 5.1. In fact
the corrections (a) and (b) cancel and the only corrections follow from (c).
Let us first recall the normal ordering of φ2 (x), namely,

1 dk
φ2 (x) =: φ2 (x) :m + √ (5.45)
4π k 2 + m2
where : :m indicates that the normal ordering is performed for a scalar of mass
m. The last integral in obviously divergent so one introduces a cutoff Λ m

such that
Λ
1 dk 1 4Λ2 m2
√ = ln 2 + O . (5.46)
4π −Λ k 2 + m2 4π m Λ2
For a general potential V (φ), Wick’s theorem tells us that,
2 2

1
ln 4 Λ d
V (φ) =: e 8 π m 2 d φ 2 V (φ) :m . (5.47)
If one expands the exponent one finds a term with no contraction, next, with
one contraction etc. We can pass from normal ordering at mass m to normal
ordering at m̃. This transformation is independent of the cutoff, since
1 m2
: φ2 :m =: φ2 :m̃ + ln 2 , (5.48)
4π m̃
and hence

1 m 2 d2
ln
: V (φ) :m =: e 8 π m̃ 2 d φ 2 V (φ) :m̃ . (5.49)
When applied to the sine-Gordon case, the normal-ordered potential takes the
form,
m2 (m2 − δm2 )
: [cos(βφ) − 1] :m [cos(βφ) − 1] , (5.50)
β2 β2
and where to the lowest order in β,
Λ
m2 β 2 dk
δm2 = − √ . (5.51)
4π k2 + m2
Thus the counterterm potential reads,

δm2 ∞
Vcounter (φ) = − 2 dx[(1 − cos(βφs )] − Evacuum , (5.52)
β −∞
where we further subtracted the energy of the vacuum. Finally the quantum
mass takes the form

1 δm2 ∞ 1 2
Mquantum = Ms + wn − 2 dx[(1 − cos(βφs )] − kn + m2 ,
2 n β −∞ 2 n
(5.53)
where kn = L , with L the size of the quantization length, to be sent to ∞ at
2π
the end of the calculation.

When substituting the set of all the frequencies wn associated with solutions
of (5.43) one finds that the quantum mass is finite and reads,
8m m m
Mquantum = 2
− = (5.54)
β π ξ
up to corrections of order mβ 2 .

5.5.1 Quantization of the breather

Next we discuss the quantization of the classical time-dependent breather solu-
tion. First, as a warm up exercise, compute the spectrum approximately, using
the Bohr–Sommerfeld “old quantization procedure”. Adapting this recipe to field
theory states that a one parameter family of periodic fields characterized by the
period τ = 2πw has an energy eigenstate whenever
τ
dt dxπ(x, t)∂0 φ(x, t) = 2πN, (5.55)
0
where N is an integer.
Using the relation between the Hamiltonian and Lagrangian densities, H =
π∂0 φ − L, we find after integrating over one period that,
τ
Eτ = 2πN − dt dxL. (5.56)
0
By differentiating with respect to τ (with N varying as a function of it, by

analytic continuation), and using the equations of motion we find,
dN 1 1 1
= = , (5.57)
dE w m 1 − E2
4M 2
where the expression for w in terms of E and M follows from the calculation of
the energy which can be performed most conveniently at t = 0, as was done in
(5.41).
Integrating this equation and using a natural boundary condition that N = 0
for E = 0 the Bohr–Sommerfeld procedure predicts the following spectrum,
N β2 8π
MN = 2Msol sin N = 1, 2, . . . , < . (5.58)
16 β2
Next we would like to describe the quantization procedure of Dashes, Hass-
lacher and Neveu (DHN). The classical action of the breather solution per period
τ = 2π
w is determined by substituting the breather solution (5.37) into the action
and integrating,
32π w
Scl (φb ) = 2 cos−1 −η . (5.59)
β m
The stability of the breather solution is determined by the requirement that
there are no negative eigenmodes to the stability equation,
[∂ μ ∂μ + m2 cos(βφb )]δn (x, t) = 0, (5.60)
where δn (x, t), which obeys δn (x, t + τ ) = eiν n δn (x, t), is the fluctuation of the
breather solution. The set of all the solutions of this equation was written down
by DHN [74].
The corresponding spectrum of νn reads,
2π 2
ν 0 = 0 ν1 = 0 νn = m + qn2 , (5.61)
w

where qn obeys the equations

ηw
Lqn + f (qn ) = 2πn f (qn ) = 4 tan−1 , (5.62)
qn
and where L is the size of the space direction. The two vanishing frequencies are
associated with the invariance under space and time translations.
The WKB semi-classical quantization determines the energy level of the
breather solution via the conditions,
∞
w2
Ecl (φb ) + Ect (φb ) − 1/2 ∂w ν n = E
0
2π
∞
2πE
Scl (φb ) + Sct (φb ) + − 1/2 νn = 2πN, (5.63)
w(φb ) 0
where Ect and Sct are the energy and action associated with the counterterm.
In the limit of L → ∞ the sum over qn turns into an integral. The integral has
a quadratic as well as logarithmic divergences. These divergences will be can-
celled out by the contribution of the counterterm such that Scl (φb ) + Sct (φb ) −
∞
1/2 0 νn is the same as the Scl (φb ) given above with the renormalization of
πβ 2
the coupling constant β 2 → ξ = 8π−β 2 . Using this result it is easy to determine
the energy,
∞

d 2wη
E=− Scl (φb ) + Sct (φb ) − 1/2 νn (5.64)
dτ 0
ξ
Substituting this into the second equation of (5.63) one finds that the energy
levels take the form,
2m Nξ π
MN = sin N = 1, 2, . . . , < . (5.65)
ξ 2 ξ
Note that in spite of the fact that the quantization condition permits any N ,
only if it is smaller than πξ the classical breather solution exists. Thus the inter-
pretation of this result is that there is a finite number of quantum bound states
corresponding to the classical breather solution. Even though the derivation of
the mass spectrum of the bound states was based on a Wentzel, Krames and
Brillonin (WKB) approximation, the final result turns out to be exact. This
statement follows from the analysis of the physical poles of soliton anti-soliton
scattering, and already indicated in perturbation theory from two loops. The
latter, though, works for mass ratios only, in view of scale dependence for the
normal ordering of each individual mass.
The spectrum (5.65) can be re-written in terms of the mass of the quantum
soliton. Using (5.54) this takes the form,
Nξ π
MN = 2Msol sin N = 1, 2, . . . , < , (5.66)
2 ξ

which indicates that the quantum breather states are indeed bound states of a
quantum soliton anti-soliton pair.
At weak coupling N β 2 1, the mass spectrum reads,
2

1 N β2
MN = N m 1 − 2 6
+ O(N β ) . (5.67)
6 16
Thus at weak coupling the lowest bound state has a mass of,
2

1 β2
M1 = m 1 − + O(β 6 ) , (5.68)
6 16
showing that first bound state is in fact the “elementary” boson of the theory.
2
Moreover the higher bound states have a mass which is N M1 + O[(β 2 ) N (1 −
N 2 )], namely bound states of N elementary bosons. These bound states are
2
loosely bound with a binding energy of m6 ( β16 )2 N (N 2 − 1). Using perturbation
theory one can show that each of these states is stable against decay to states
with lower N . In fact the stability turns out to be an exact statement. The source
of the stability of these states is the set of infinitely conserved charges as will be
discussed in the following section.
5.6 Integrability and factorized S-matrix

One of the first papers that discusses integrability of the S-matrix is the seminal
paper [235]. We follow this paper in describing the basic notions of integrability.
The Yang–Baxter relations were derived in [230] and [29] and S-matrix results
for solitons and breathers of the sine-Gordon model are analyzed in [198].
Consider an integrable theory with ∞ of conserved charges Qn diagonalized
in the single particle base such that
Qn |p(a) >= wn(a) (p)|p(a) >, (5.69)
where p is the momentum of the particle and (a) denotes its type. For the sine-
(a)
Gordon case the eigenvalues wn (p) are given by,
(a) (a)

w2n + 1 (p) = p2n + 1 , w2n (p) = p2n p2 + m2a . (5.70)
(a)
In general one assumes that the wn (p) form a set of independent functions. A
multiple particle in or out state obeys
(a 1 ) (a k )

k
(a 1 ) (a k )
Qn |p1 . . . pk , in> = wn(a i ) (pi )|p1 . . . pk , in>, (5.71)
i= 1
and since the charges Qn are conserved one finds that,

wn(a i ) (pi ) = wn(a i ) (pi ). (5.72)
i∈in i∈out

5.6 Integrability and factorized S-matrix 93
p2
p1
p3
p4
p4
p3 p1
p2
Fig. 5.2. Space-time picture of the multi-particle factorized scattering.
From this conservation one can deduce that,

r For any given mass ma the number of initial and final particles of this mass is
the same.
r The final set of momenta is the same as the initial one.
These two rules, that should apply also to intermediate states where particles
are far enough from each other, together with the special kinematics of two
dimensions, are behind the assertion that the multi-particle S-matrix of theories
equipped with infinitely many conserved charges, can be expressed in terms of
two-particle ones.
The factorized S-matrix corresponds to the following scattering process:
r In the infinite past a set of N particles with momenta p1 > p2 > . . . > pN are
spatially arranged in the opposite order, namely, x1 < x2 < . . . < xN .
r In the interaction region the particles collide in pairs. In each collision the
momenta are conserved and in between collisions the particles move as free
particles.
r The final state of the outgoing particles, achieved after N (N −1) pair collisions,
2
is built from the N particles arranged along the x coordinate in the order of
increasing momenta.
The factorized scattering of the N particles is represented, for N = 4, by the

space-time diagram Fig. 5.2, in which time is flowing up.

p2
p2
p1
p1
p3 p3
p3 p3
p1
p1
p2
p2
a b
Fig. 5.3. Two possible ways of three-particle scattering.
r Each line corresponds to a given value of the momentum associated with the
slope of the line.
r Each vertex corresponds to a two-particle collision. The two-particle amplitude
Sij (pi , pj ) has to be attached to each vertex.
r The total S-matrix element of the process is the product of all the N (N −1)
2
two-particle amplitudes ij Sij , and then a sum over the different kinds of
particles in the internal lines.
Take for example the case of N = 3. The same scattering can be represented in
two ways, as is shown in Fig. 5.3. These two differ only by a parallel translation
of a line, and thus represent the same process.
5.7 Yang–Baxter equations

The amplitudes and phases of the two diagrams should be the same. The require-
ment that they are indeed the same imposes cubic equations on the two-particle
matrix elements which are called factorization equations or Yang–Baxter equa-
tions. We now proceed to analyze these conditions of factorization.

It is more convenient to discuss the S-matrix in terms of the rapidities of the

massive particles. The rapidity β of a particle of mass m is defined via,2
p± = m exp(±β). (5.73)
The scattering amplitude of a system of two particles S(p1 , p2 ) is a function

of the rapidity difference β = β1 − β2 as can be seen from the fact that the
s-channel invariant
s = (p1 + p2 )2 , (5.74)
is given by
s = m21 + m22 + 2m1 m2 cosh(β1 − β2 ). (5.75)
We now analytically continue s and define the amplitude S(s) as an analytic

function in the complex s-plane. This function has two cuts along the real axis for
s ≥ (ma + mb )2 and s ≤ (ma − mb )2 . The points s = (ma − mb )2 and s = (ma +
mb )2 are square root branching points of S(s). Using (5.75), S(s) is mapped to
S(β), where β = β1 − β2 . The physical sheet is mapped into the strip of 0 <
Imβ < π, the branch cuts to the lines Imβ = 0 and Imβ = π, and the crossing
transformation of s → 2m21 + 2m22 − s to β → iπ − β. If one assumes that there
are no other cuts, then S(β) is a meromorphic function in the above strip. In
the non-relativistic limit, for mi p1i , the rapidity goes into the velocity βi →
vi = mp ii .
5.8 The general solution of the S-matrix

Consider the two-particle S-matrix,
ik Sj l = <Aj (β1 )Al (β2 ), out|Ai (β1 )Ak (β2 ), in>

= δ(β1 − β1 )δ(β2 − β2 )[δik δj l S1 (s) + δij δk l S2 (s) + δil δk j S3 (s)]
± [i ↔ k, β1 ↔ β2 ], (5.76)
where i, j, k, l = 1, 2 so that the particles are in doublets of O(2), and the ± refers
to bosons (+) and fermions (−). This can be generalized to O(N ) in a straightfor-
ward way. Here we analyze only the case of the doublet. The amplitudes S2 , S3
are the transition and reflection amplitudes, respectively, and S1 corresponds
to the process Ai + Ai → Aj + Aj for (i = j). The Ai (β), non-commutative
2 Please note, that the β we had before, in the term cos(βφ) in the action, is not to be confused
with the present one, which denotes the rapidity.

variables representing the particles, obey the relation,

Ai (β1 )Aj (β2 ) = δij S1 (β) An (β2 )An (β1 ) + S2 (β)Aj (β2 )Ai (β1 )
n
+ S3 (β)Ai (β2 )Aj (β1 ). (5.77)
Incoming states are represented by products arranged by order of decreasing
rapidities, while outgoing by increasing rapidities. The crossing symmetry rela-
tions are,
S2 (β) = S2 (iπ − β) S1 (β) = S3 (iπ − β). (5.78)
The unitarity conditions for the two-particle S-matrix are,
S2 (β)S2 (−β) + S3 (β)S3 (−β) = 1
S2 (β)S3 (−β) + S2 (−β)S3 (β) = 0
2S1 (β)S1 (−β) + S1 (β)S2 (−β) + S1 (β)S3 (−β) + S2 (β)S1 (−β)
+ S3 (β)S1 (−β) = 0. (5.79)
Unitarity (5.79) and crossing symmetry (5.78) do not fix the S-matrix. The addi-
tional conditions one has to impose are those of the factorization or Yang–Baxter
equations. The latter are obtained by considering all possible three-particle in-
products Ai (β1 )Aj (β2 )Ak (β3 ), reordering them to get out-products using (5.77),
and requiring that the results be independent of the order of successions of the
two particle commutations, one then finds,
S2 S1 S3 + S2 S3 S3 + S3 S3 S2 = S3 S2 S3 + S1 S2 S3 + S1 S1 S2
S3 S1 S3 + S3 S2 S3 = S3 S3 S1 + S3 S3 S2 + S2 S3 S1
+S2 S3 S3 + 2S1 S3 S1 + S1 S3 S2 + S1 S3 S3 + S1 S2 S1 + S1 S1 S1 , (5.80)
where, for each of the terms, the arguments for the three S factors are β, β +
β , β , respectively.
The general solution of the factorization equations is expressed in terms of one
function which we take to be S2 (β). The solution reads,
4πδ 4πβ
S3 (β) = ictg cth S2 (β)
γ γ
4πδ 4π(iδ − β)
S1 (β) = ictg cth S2 (β) (5.81)
γ γ
with γ and δ real, but so far arbitrary. This solution as well as the restriction from
unitarity (5.79) is valid also for a non-relativistic system. Crossing symmetry is a
restriction that shows up only in the relativistic case. Imposing the latter (5.78)
on the general solution fixes δ = π. A “minimum” solution for S2 (β) then takes
the form,
2 4π 2 4πβ 4π(iπ − β)
S2 (β) = sin sh sin U (β), (5.82)
π γ γ γ

where,
∞
8π 8β 8π 8β Rn (β)Rn (iπ − β)
U (β) = Γ Γ 1+i Γ 1− −i ,
γ γ γ γ n =1
Rn (0)Rn (iπ)

Γ + 2n 8π
γ i 8β
Γ 1 + 2n 8π
γ +γi 8β
γ )
Rn (β) = . (5.83)
Γ (2n + 1) γ + i γ Γ 1 + (2n − 1) γ + i 8β
8π 8β 8π
γ
It is a “minimal” in the number of singularities along the imaginary β axis, and

more general solutions can be obtained from it by multiplying with a meromor-
L shβ +i sin α k
phic function of the form f (β) = k = 1 shβ −i sin α k for arbitrary real numbers
αk .
5.8.1 The S-matrix of the sine-Gordon model

The sine-Gordon model has a hidden O(2) invariance, which is simplest to see
via the soliton solutions, where the soliton and anti-soliton are incorporated in
an O(2) doublet. In terms of the A1 (β) and A2 (β) the soliton and anti-soliton
amplitudes are,
A(β) = A1 (β) + iA2 (β) Ā(β) = A1 (β) − iA2 (β). (5.84)
In terms of A and Ā, (5.77) takes the form,
A(β1 )Ā(β2 ) = ST (β)Ā(β2 )A(β1 ) + SR (β)A(β2 )Ā(β1 )

A(β1 )A(β2 ) = S(β)A(β2 )A(β1 )
Ā(β1 )Ā(β2 ) = S(β)Ā(β2 )Ā(β1 ), (5.85)
where S(β), ST (β) and SR (β) are the scattering amplitude of identical solitons,
transition and reflection amplitude for soliton anti-soliton, which are related to
S1 (β), S2 (β) and S3 (β) as,
S(β) = S3 (β) + S2 (β)

ST (β) = S1 (β) + S2 (β)
SR (β) = S1 (β) + S3 (β). (5.86)
It follows from crossing symmetry (5.78) that,
S(β) = ST (iπ − β) SR (β) = SR (iπ − β). (5.87)

X X X X X X
X X X X X X X X X X
0
X X X X X X
X X X
0
Fig. 5.4. The zeros (crosses) and poles (dots) of ST (β) (upper) and SR (β)
(lower). All the singularities have pure imaginary values even though some of
them are displaced from the imaginary axis for clarity [235].
Substituting these relations in the general solution derived in the previous

section one finds for the SG model,

sh 8πβ
γ
ST (β) = −i SR (β)
2
sin 8πγ

sh 8π(iπ−β
γ
)
S(β) = −i SR (β), (5.88)

8π2
sin γ
where,
1 8π2
SR (β) = sin U (β). (5.89)
π γ
The zeros and the poles of ST (β) and SR (β) are shown in Fig. 5.4.
The solution (5.88) is in fact the exact solution of the S-matrix of the SG
model. This assertion is supported by the following properties:
r The poles of ST (β) are located at equidistance, and their values are in accor-
dance with the semi-classical mass spectrum if one equates γ = 8ξ.
r Note that for the value of ξ = π where the coupling of the associated Thirring
model vanishes, and the SG model is a bosonized version of a free Dirac fermion,
ST (β) ≡ S(β) = 1 SR (β) = 0. (5.90)
r At ξ ≥ π all bound states, including the “elementary particle” associated with
the field of the SG model, become unstable and the spectrum includes only
soliton and anti-soliton. This situation follows from the fact that at this region

5.9 From conformal field theories to integrable models 99
the Thirring coupling is negative and there is a repulsion between the solition
and anti-soliton.
r At ξ = π the reflection amplitude vanishes identically.
n
r Expanding (5.88) in powers of [( 8π ) − 1], which means small coupling of the
γ
massive Thirring model, matches the perturbative expansion of the latter
model.
r The limit β 2 → 0 of the exact result (5.88) is equal to the semi-classical expres-
sion of the two-particle S-matrix.
The explicit expression for the two-particle S-matrix (5.88) enables one to also
write down the S-matrix for any number of solitons and anti-solitons and the
scattering of any number of bound states. This general approach to solving the
S-matrix can be applied to the various integrable models. Here we demonstrate
it on the sine-Gordon model. For soliton and anti-solitons we find the following
S-matrix elements:
1
,1 − 1 ,− 1
S 12 , 12 (β) = S− 12 ,− 12 (β) = S(β)
2 2 2 2

π
1
,− 12 − 12 , 12
sh ξβ
S 2
1
,− 12
(β) = S − 12 , 12
(β) = − S(β)
π
2
sh ξ (β − πi)

π2 i
1
,− 12 − 12 , 12
sh ξ
S 2
− 12 , 12
(β) = S 1 1 (β) = S(β)
2 ,− 2 π
sh ξ (β − πi)
,
S 11 , 22 (β) = 0 1 + 2 = 1 + 2 . (5.91)
S can also be expressed as an exponential of an integral,

⎡ ⎤
∞ sin(κβ) sh π−ξ κ
2
S(β) = − exp ⎣−i πκ ξ κ dκ⎦. (5.92)
0 κch 2 sh 2
5.9 From conformal field theories to integrable models

So far we have analyzed integrable theories based on a scalar field theory with
an integrable interacting potential. As was mentioned in Section 1 there is a
more general scheme of constructing integrable models. This scheme is based
on perturbing conformal field theories with relevant primary fields so that the
action takes the form,

S = SCFT + λi d2 zΦi (z, z̄). (5.93)
i
Note that just as for the conformal field theories we use here complex
two-dimensional coordinates. The Φi (z, z̄) are primary fields of conformal
dimension Δi + Δ̄i < 2, namely relevant operators. Since these operators are

super-renormalizable, they do not affect the short distance behavior but do affect
the structure of the IR domain. In the analogy with statistical mechanical sys-
tems, where the CFT describes the behavior of the system at its fixed point, the
perturbation with the relevant primary fields describes the scaling region around
the fixed point.
A system described by an action of the form (5.93), is integrable provided that
one can identify a set of infinitely many conserved charges just as for the sys-
tems described previously. An important class of such theories are the integrable
minimal models, for example the tricritical Ising model M4,5 perturbed by Φ 35 , 35
and the tricritical Potts model M6,7 perturbed by Φ 17 , 17 .
The renormalization group (RG) flow of the integrable systems follows a tra-
jectory that starts at a fixed point and may end on another one in the IR, or
on a point that corresponds to a massive QFT. An important property of these
flows is the c-theorem3 which states the following:
Quantum field theories which possess rotational invariance, reflection pos-
itivity, and conservation of the energy momentum tensor admit a function
c(λi ) of the coupling constants λi which is non-increasing along the RG
trajectories and is stationary only at fixed points.
The proof of the theorem is as follows. Consider the correlators of T ≡ Tz z
and Tz z̄ ,
F (z z̄) G(z z̄)
<T (z, z̄)T (0, 0)> = , <T (z, z̄)Tz z̄ (0, 0)> = ,
z4 z 3 z̄
H(z z̄)
<Tz z̄ (z, z̄)Tz z̄ (0, 0)> = 2 2 . (5.94)
z z̄
We now use the conservation law,
¯ + ∂Tz z̄ = 0,
∂T ¯ z z̄ = 0,
∂ T̄ + ∂T (5.95)
to deduce the following differential equations for F, G and H,
Ḟ + (Ġ − 3G) = 0 Ġ − G + Ḣ − 2H = 0, (5.96)
where Ȧ = ddAlog(x)x . Since the positivity condition implies that H ≥ 0, the following
c function is non-increasing,
c = 2F − 4G − 6H, ċ = −12H. (5.97)
At the fixed points Tz z̄ = 0, hence G = H = 0 and c = 2F , as indeed it should
(c )
be. Recall that the OPE in CFT is of the form T (z z̄)T (0, 0) = z24 + . . .
One can further write down an expression for the integral of ċ, namely for
the difference of c in the UV and IR regions. For the case of a perturba-
tion with a single operator Φ, the trace of the energy-momentum tensor is
3 Zamolodchikov c-theorem was derived in [232].

Tz z̄ = πλ(1 − Δ)Φ, where the total conformal dimension is 2Δ. Using (5.97) it
was shown that,

cUV − cIR = 12πλ (1 − Δ)
2 2
d2 x|x|2 <Φ(x)Φ(0)> . (5.98)
This result has been applied to integrable minimal models yielding the cor-
rect difference in the Virasoro anomalies. Another example where this relation
between the conformal data and the properties of the non-conformal theory can
be tested is the sine-Gordon model. The model can be thought of as a pertur-
bation on a free massless scalar field which has c = 1, and the massive model
has c = 0. Let us check for this case the outcome of the relation (5.97). The
perturbation now is,
m2 β2
λΦ = : (cos βφ − 1) : 2Δ = . (5.99)
β2 4π
If we expand λΦ in β, then the leading order λΦ is 12 m2 φ2 for which Δ = 0 and,

1 2 m4 2
<λΦ(x)λΦ(0)>= m <φ(x)φ(0)>2 = K (m|x|), (5.100)
4 8π2 0
where K0 is a Bessel function. Inserting this into (5.97) we get,
∞
m4
cUV − cIR = 3π 2 2 2 2
d x|x| K0 (m|x|) = 3 drr3 K02 (r) = 1, (5.101)
2π 0
which verifies our data about the Virasoro anomalies at the two ends of the
trajectory. One can further show that higher-order terms in β are in accordance
with the fact that cUV − cIR is β independent.
5.10 Conserved charges and classical integrability

We will now show that the classical sine-Gordon theory incorporates an infinite
set of conserved charges. This will imply an exact determination of any S-matrix
element of the theory. This property of having a set of an infinite number of
conserved currents and charges is referred to as classical integrability. In the
next section we will discuss the fate of the integrability in the quantum domain.
We choose here to describe the sine-Gordon model, however this structure applies
to a class of models.
There are several methods to determine these classically conserved charges.
Here we will follow two of them: The Lax pair approach and a method based on
a generating function.
The infinitely many charges of the integrable models were analyzed using var-
ious different techniques. The Lax pair method applied to the sine-Gordon is
described in [61]. The method of multilocal charges from an integral equation
was presented in [153]. The inductive method was introduced in [45].

5.10.1 The Lax pair method

In the Lax pair approach the idea is to rewrite the sine-Gordon equation of
motion in terms of a commutator relation between two operators. Let us first
rewrite the equation of motion of the sine-Gordon system in the light-cone coor-
dinates,
∂˜+ ∂˜− φ̃ = − sin(φ̃), (5.102)
where φ̃ = βφ and ∂˜± = m∂± .

Next we define the following pair of 2 × 2 matrix Lax operators,
L = 2σ3 ∂˜− + σ2 (∂˜− φ̃)

1
B = [σ3 cos(φ̃) + σ2 sin(φ̃)]L−1 . (5.103)
2
In terms of these operators the sine Gordon equation takes the form,
∂˜+ L = [L, B]. (5.104)
The reason for rewriting the equation of motion in this form is the fact that the
spectrum of L is conserved. To realize this property of the Lax pair, notice first
that the solution of (5.104) can be parameterized as,
L(x+ ) = S(x+ )L(0)S −1 (x+ ) ∂+ S = −BS. (5.105)
Now consider the eigenvalue problem,
L(x+ )v(x+ ) = λv(x+ ). (5.106)
It is easy to check that if v(0) is an eigenfunction of L(0) with eigenvalue λ, then

v(x+ ) is an eigenfunction of L(x+ ) with the same eigenvalue λ, where,
v(x+ ) = S(x+ )v(0). (5.107)
Since x+ is the light-cone time direction, this implies the conservation of the
eigenvalues. This conservation of the spectrum is the origin of the infinite set of
conserved charges.
For the case that L can be represented by a finite matrix, it is obvious from
(5.105), using the cyclicity of the trace, that Qn = T r[Ln ] are conserved charges.
In general, and in particular in our case, L does not act only in a space of
finite-dimensional matrices but also in the continuous space whose base vectors
∞
are |x−>. Thus the trace takes the form of the integral −∞ dx− . Due to the
+
unitarity of S(x ) the cyclicity property of the trace is maintained and hence
Qn = T r[Ln ] are indeed conserved charges.
It turns out that one can map the Lax pair of the sine-Gordon system to that
of the Korteweg-deVries (KdV) equation ∂0 u(x, t) = 6u∂1 u − ∂13 u. This map is
useful since it is more convenient to express the conserved charges of the sine-
Gordon system in terms of the L operator of the KdV equation. In this format

the first four charges of the set of infinite charges which are classically conserved
are,
∞
1
Q1 = − (∂− φ̃)2 dx−
4 −∞
2 ∞
1
Q2 = + [(∂− φ̃)4 − 4(∂−2
φ̃)2 ]dx−
4 −∞
3 ∞
1
Q3 = − [(∂− φ̃)6 − 20(∂− φ̃)2 (∂−2
φ̃)2 + 8(∂−
3
φ̃)2 ]dx−
4 −∞
4 ∞
1 112 2 4
Q4 = + [(∂− φ̃)8 − (∂− φ̃) − 56(∂− φ̃)4 (∂− 2
φ̃)2
4 −∞ 5
224 64 4 2
+ (∂− φ̃)2 (∂−
3
φ̃)2 − (∂− φ̃) ]dx− . (5.108)
5 5
5.10.2 The generating function method

A second method to determine the set of infinite conserved charges is based on
a generating function. Define the generating function,
1
ψ = φ + sin−1 (β∂− ψ). (5.109)
β
When φ obeys the sine-Gordon equation, ψ obeys the following equation,

1 − 1 − β 2 2 (∂− ψ)2
∂+ − m2 ∂− (cos(βψ) − 1) . (5.110)
2
Equation (5.109) determines ψ as a power series in φ and . Upon substituting

this into (5.110), we get an infinite set of conserved charges, that are the coef-
ficients of the even powers of . A dual sequence of charges can be obtained by
interchanging ∂+ and ∂− in the equations above. The set of conserved currents
is labeled as,
+ −
∂+ J2n + ∂− J2n = 0, (5.111)
where 2n relates to the power of . The lowest-order current is the energy-

momentum tensor,
1 m2
J0+ ≡ T−− = (∂− φ)2 , J0− ≡ T+− = − (cos(βφ) − 1). (5.112)
2 β2
The second-order currents are,
1 2 2 β2 m2
J2+ = (∂ φ) − (∂− φ)2 , J2− = (φ− )2 cos(βφ), (5.113)
2 − 8 2
and similarly one can write the expressions for the higher order currents J2n .
One can map the charges derived by the Lax pair procedure to those derived
by the generating function method.

5.11 Multilocal conserved charges

In the previous section we analyzed the set of infinitely many conserved charges
associated with local currents. It will be shown later that these conservation
laws are responsible for the fact that the system is integrable, namely there is no
particle production and the S-matrix is factorizable. We would like to show now
that this type of structure may also follow from conservation laws associated with
multilocal currents. The construction of the multi-local currents will be presented
in two ways: (i) via an integral equation, (ii) by an inductive procedure.
We then show that the two types of charges are in fact equivalent.
5.11.1 Multilocal charges from integral equation

Consider first the O(N ) non-linear sigma model defined by the Lagrangian den-
sity,

1
SO (N ) = 2 d2 x (∂+ n) · (∂− n) − u(n2 − 1) , (5.114)
2g0
where n is an N -dimensional real vector and g0 is the coupling constant. The

fact that the field n is constrained n2 = ni ni = 1 i = 1, . . . , N is incorporated via
the Lagrange multiplier u. The equations of motion that follow from this action
are,
∂− ∂+ n + un = 0, n2 = 1. (5.115)
So u is actually the action density u = (∂+ n) · (∂− n). Instead we can write the
equation of motion as,
∂μ ∂ μ n + n(∂μ n · ∂ μ n) = 0. (5.116)
The action is classically invariant under O(N ) global symmetry, by construc-

tion. For the special case of n = 3, namely O(3), the current takes the form,
jμk = ij k (ni ∂μ nj − ∂μ ni nj ). (5.117)
It is easy to check, using the equations of motion, that this current is indeed
conserved. In addition the energy-momentum is conserved,

1 1
∂+ T−− = ∂+ (∂− n)2 = ∂− n∂− ∂+ n = −u∂− (n)2 = 0. (5.118)
2 2
Thus classically the trace of the energy-momentum tensor vanishes T+− = 0.
In addition there is an infinite set of currents, the simplest of which takes the
form,
2
1 ∂− n u
J− = ∂− , J+ = − . (5.119)
2|∂− n| |∂− n| |∂− n|

5.11 Multilocal conserved charges 105
An alternative way to write down a set of non-local classically conserved

charges is the following. We define at any given t the (2 × 2) U (t, x) operator via
the equation,
w i
∂x U (t, x) = i j − wj0i σ i U (t, x), (5.120)
1 − w2 1
and the boundary condition U (t, −∞) = 1, for any Cauchy data
ni (t, x), ∂t ni (t, x), where w is the “spectral parameter” which is a com-
plex parameter with w = ±1. Following this definition and using the equations
of motion one can show that,
d d
Q(w) ≡ U (t, ∞) = 0. (5.121)
dt dt
If one expands Q in terms of the spectral parameter one finds a set of infinitely
many conserved charges,
∞
d
Q(w) = Qn wn Qn = 0. (5.122)
n=0
dt
We can now rewrite the differential equation (5.120) in terms of the integral
equation,
x
w
U (t, x) = 1 + dy j0i − wj1i (t, y)σ i U (t, y). (5.123)
1 − w −∞
2
Inserting the expansion,

∞

U (t, x) = Un wn , (5.124)
n=0
into the integral equation for U (t, x), we find the following recurrence relation,
x
Un (t, x) = i dy j0i (t, y)σ i Un −2k −1 (t, y)
−∞ ( n −1 )
0≤k ≤

2

− j1i (t, y)σ i Un −2l (t, y) , (5.125)
(n )
1≤l≤ 2
with U0 (t, x) = 1. Thus we can calculate Qn recursively deriving explicit non-

local expressions for the set of infinitely many conserved charges. The “lowest”
charges are given by,

Q01 = 0 Qi1 = dyj0i (t, y)
1
Q02 = − Qi1 Qi1
2
j
Q2 = dx dydy ij k j0 (t, y)j0 (t, y )θ(y − y ) − dyj1i (t, y). (5.126)
i k
The algebraic structure associated with these charges is the Yangian symmetry,
the description of which is beyond the scope of this book. In the reference list we

mention several that deal with these algebras. In Section 5.12 it will be further
shown that both the charges of the form (5.119) as well as (5.126) are quantum
mechanically conserved.
5.11.2 Charges by inductive procedure

The second method of obtaining non-local currents is as follows. Assume that
the system admits a (non-abelian) conserved current which is also a pure gauge,
namely,
Jμ = g −1 ∂μ g, ∂ μ Jμ = 0, (5.127)
where g is a non-singular matrix (for instance U (N ) or O(N ) matrix). It follows

that the current J is a flat gauge connection, since for Dμ = ∂μ + Jμ we find
that [Dμ , Dν ] = 0 and also ∂μ Dμ = Dμ ∂μ . Using the terminology of differential
forms these properties can be rewritten as,
DJ ≡ dJ + J ∧ J = 0 d ∗ J = 0, (5.128)
together with [D, D] = 0 and D ∗ d = d ∗ D. Now let us assume that there is an

(n )
n-th conserved current Jμ , then there is a function χ(n ) such that,
Jμ(n ) = μν ∂ ν χ(n ) . (5.129)
Then there is an (n+1)-th conserved current,
Jμ(n + 1) = Dμ χ(n ) , ∂ μ Jμ(n +1) = 0. (5.130)
The conservation follows easily,
d ∗ J (n + 1) = d ∗ Dχ(n ) = D ∗ dχ(n ) = DJ (n ) = DDχ(n −1) = 0, (5.131)
or directly,
∂ μ Jμ(n + 1) = Dμ ∂μ χ(n ) = −μν Dμ Jν(n ) = −μν Dμ Dν χ(n −1) = 0. (5.132)

(1)
The sequence of χ(n ) starts with χ(0) = 1 and Jμ = Jμ .
Associated with the set of infinitely many conserved currents, there is obviously
also a set of infinite number of conserved charges,

(n ) (n )
Q = dxJ0 (t, x). (5.133)
Consider the special case n = 2,

(2) (1)
Q (t) = dxJ0 = dx(∂0 + j0 )χ(1)
(2)

(1) (1)
= − dxj1 (t, x) + dxj0 (t, x)χ(1) (t, x). (5.134)

5.12 Quantum integrable charges in the O(N ) model 107
We can now re-express χ(1) as χ(1) (t, x) = dx j0 (t, x ). When substituting
(1)
this to the previous equation we discover the structure of multi-local charges of

(5.126).
5.12 Quantum integrable charges in the O(N) model

In Section 5.11 it was demonstrated that certain two-dimensional interacting
models have an infinite set of conserved classical charges. This property consti-
tutes the classical integrability of a given system. A natural question to ask is
whether this integrability persists also in the quantum regime. We will analyze
this question in the context of two models, the O(N ) sigma model in this sec-
tion and the sine-Gordon model in the following one. In fact we have already
seen in Section 5.8 how the quantum integrability of the sine-Gordon model fully
determines the S-matrix of the theory.4
In Section 5.11.1 it was shown that classically the O(N ) model is scale invari-
ant as the trace of the energy-momentum vanishes. Following our discussion of
the Virasoro anomaly in Chapter 2, it is clear that quantum mechanically the
classical conformal invariance of the O(N ) model is broken by an anomaly. The
right-hand side of (5.118) should not vanish any more but rather be equal to some
local terms. These terms can be determined by Lorentz x+ → ax+ , x− → a−1 x−
and scale invariance x+ → ax+ , x− → ax− . It turns out that up to a constant
the quantum relation is,

1
∂+ (∂− n)2 = β̂∂− u = β̂∂− (∂− n∂+ n). (5.135)
2
In fact one can show that the constant β̂ is the one-loop beta function.
Consider now the next conservation law. Classically it reads,

1 2 1 2 2 3
∂− u(∂− n) + ∂+ (∂− n) = (∂− n)2 (∂− u). (5.136)
2 2 2
One can show that this classical conservation is directly related to the conserva-
tion of the current in (5.119). For this make use of the classical scale invariance
x− → f (x− ) and the classical conservation ∂+ (∂− n)2 = 0, to choose a gauge
where (∂− n)2 = 1. Now the classical conservation law takes the form,

1 2 2
∂+ (∂− n) = (∂− u). (5.137)
2
This is exactly the conservation of the current (5.119), when inserting the gauge
(∂− n)2 = 1. To get to the form in general coordinates, namely the form as in
4 Quantum integrability was discussed in [176].

(5.119), make the substitution,

1
∂− → ∂− , (5.138)
|∂− n|
noting that it also implies u → |∂ −u n | .
Quantum mechanically again the right-hand side of equation (5.136) can be
corrected by several terms. To eliminate possible terms that are total derivatives
it is convenient to analyze the integrated form of this conservation law,

1 2 2 1
∂+ dx− (∂− n) = (3 + γ) dx− (∂− n)2 (∂− u), (5.139)
2 2
and finally using (5.135) to eliminate ∂− u we get the quantum conservation law,

1 2 2 (3 + γ)
I = dx− (∂− n) − (∂− n)4 , ∂+ I = 0. (5.140)
2 4β̂
One can now show that this conservation law implies the conservation of
3
P− i . Since I commutes with the S matrix [I, S] = 0 one has,
i
 = <a in|I|a in> . (5.141)
For asymptotic states with N particles and momenta P1 , . . . , PN we have

<N |I|N> = Constant P− 3i . (5.142)
i
The reason that the conserved charge on an asymptotic state has to be pro-

portional to i P− 3i is that its tensorial structure is of the form −−− , the only
conserved quantity with − Lorentz index is P− and there are no higher tensorial
charges that are not products of P− .
In a similar manner one has a similar conservation law for P+ so that,

P−3 = P−3 P+3 = P+3 . (5.143)
in out in out
If we now write these conservation laws for a 2 → N process combined with

the ordinary conservation of momenta we get four equations for two quantities
which, combined with the analyticity of the S-matrix, implies that there cannot
be any multiple production, and the only allowed process is 2 → 2.
It can also be shown that the classically conserved charges constructed by an
integral equation (5.126) are also quantum mechanically conserved.
5.13 Non-local charges and quantum groups

The discussion of quantum groups and non-local charges follows the paper [39].
For a review see, for instance, [114].
In the previous sections we have derived in various forms sets of infinitely
many conserved charges and argued that they constitute the integrability of
the corresponding models. In particular we described in Section 5.11 non-local

5.13 Non-local charges and quantum groups 109
charges. Here we will establish the algebraic structure of these charges. It will
be shown that they involve non-trivial braiding and that they obey the algebra
associated with “quantum groups”.
Rather than discussing the generalities of this algebraic structure we analyze it
in the context of our laboratory model, the sine-Gordon model. We now rewrite
the Lagrangian density of the model as,

1 2 ¯ λ̂
S= d z∂φ∂φ + d2 z : cos(β̂φ) : . (5.144)
4π π
It is straightforward to relate λ̂ of this formulation to m, β of (5.23), and β̂ here
equals β there.
Recall (Section 5.9) that this action can be considered as a conformal field
theory plus a relevant perturbation of the form (5.93). In such a case one can
identify a conserved current that obeys the relation,
¯
∂J(z, ¯ z̄) = ∂¯H̄(z, z̄),
z̄) = ∂H(z, z̄) ∂ J(z, (5.145)
¯ = ∂ J¯ = 0, and H, H̄ are defined via,
where for the conformal limit one has ∂J
Resz =w (φp ert (w)J a (z)) = ∂ha (z) H a (z, z̄) = 2λ̂ha (z)φ̄p ert (z̄), (5.146)
where the perturbation term is written, in the conformal limit, as
φp ert (z)φ̄p ert (z̄), and all this under the condition that the Res above are indeed
total derivatives. A similar construction applies also for the anti-holomorphic
current.
We can now identify a pair of non-local fields
x
1
φ̃± (t, x) = φ(t, x) ± dy∂0 φ(t, y) , (5.147)
2 −∞
with which we can write a pair of conserved currents J± of the form (5.145) as,
± 2 i φ̃ + (t,x)
J± = e β̂

β̂ 2 ±i 2
−β̂ φ̃ + (t,x)∓ i φ̃ − (t,x)
H± = λ e β̂ β̂ . (5.148)
β̂ 2 − 2
The conserved charges associated with the pair of currents are

1 1
Q± = dzJ± + dz̄H± Q̄± = dz̄ J¯± + dz H̄± . (5.149)
2πi 2πi
The charges Q± are non-local, as a consequence of being built from the non-local
field φ̃.
Using the basic canonical commutation relation [φ(t, x), ∂0 φ(t, y)] = 4πiδ(x −
y) and eA eB = e[A ,B ] eB eA (when [A,B] commutes with A and B) one finds the
following braiding relations,
1
J± (t, x)J¯∓ (t, y) = 2 J¯∓ (t, y)J± (t, x)
q
J± (t, x)J± (t, y) = q 2 J¯± (t, y)J± (t, x),
¯ (5.150)

where,
− 2 πi
q=e 2
β̂ . (5.151)
These non-trivial braiding relations of the currents imply similar relations for
the conserved charges,
Q+ Q̄+ − q 2 Q̄+ Q+ = 0
Q− Q̄− − q 2 Q̄− Q− =0
1
Q+ Q̄− − 2 Q̄− Q+ = a(1 − q 2Q t o p )
q
1
Q− Q̄+ − 2 Q̄+ Q− = a(1 − q −2Q t o p )
q
[Qtop , Q± ] = ±2Q± [Qtop , Q̄± ] = ±2Q̄± , (5.152)
where a = λ 2
2πi γ , γ −1 = Δ = −Δ̄(Q± ) and the topological charge Qtop =
β̂
= ∞) − φ(x = −∞)) (compare with (5.6)).
2π (φ(x
This algebra of the charges is referred to as “q-deformation” ŜLq (2) of the
SL(2) affine Lie algebra with zero center. Recall the basic SL(2) algebra in the
Chevaley basis (3.5),
[H, E± ] = ±2E± [E+ , E− ] = H. (5.153)
Introducing the spectral parameter w the infinitely many generators of the SL(2)

affine Lie algebra are defined via J a = n Jna wn with J a = H, E± . We then
ˆ as,
define the Chevaley basis of the affine algebra SL(2)
E+ 1 = wE+ E−1 = w−1 E−
E+ 0 = wE− E−0 = w−1 E+
H1 = H H0 = −H. (5.154)
In terms of these generators the ŜLq (2) algebra reads,
[Hi , E+j ] = aij E+j
[Hi , E−j ] = −aij E−j
q H i − q −H i
[E+i , E−j ] = δij E−j , (5.155)
q − q −1
and aij is the Cartan matrix of SL(2).
The relations between the non-local charges Q± and Q̄± and the generators
of the ŜLq (2) algebra are,
H1 H0
Q+ = cE+ 1 q 2 Q− = cE+0 q 2
H1 H0
Q̄− = cE−1 q 2 Q̄+ = cE−0 q 2
Qtop = H1 = −H0 (5.156)

2 −2
where c2 = λ
2πi γ (q − 1).

5.14 Integrable spin chain models and the algebraic Bethe ansatz 111
5.14 Integrable spin chain models and the algebraic Bethe ansatz
The discussion of the algebraic Bethe ansatz follows closely the pedagogical paper
of Faddeev [88] and also Beisert [31]. The use of the ansatz in a continuous system
that we present follows that of Zamolodchikov [234].
A very useful class of two-dimensional integrable models are the spin chain
models. In these models the space is divided into a discrete number of sites
where spin variables are placed. So far, and in fact also in the rest of this book,
we do not discuss discretized field theories. In this chapter we do since the spin
chain models will be shown, in Section 18, to be intimately related to integrable
sectors of gauge theories in four dimensions. We will demonstrate the techniques
used to solve the spin chain models by applying them to a prototype model,
the XXX1/2 model. We will describe the model, write down the Bethe ansatz
equations associated with it, solve them and extract the spectrum of the model.
We then apply the technique to the discretized sine-Gordon model.
A discrete circle with N ordered points is taken to be the space direction.
The “space” is periodic so that each site is identified with i ≡ i + N . The formal
continuum limit can be taken by introducing a lattice spacing Δ such that Δ →
0, N → ∞ while x = N Δ is kept finite.
At each site there is a dynamical variable Xiα where i denotes the site and α
is a set of finite number of values. One defines a quantum algebra of observables
A by fixing a set of commutation relations between the Xiα . When [Xiα , Xjβ ] = 0
for any i = j the algebra is called ultra local. Examples are canonical variables,
and spin variables which will be used in the XXX1/2 model we are about to
describe.
The Hilbert space of the representations of the ultra local algebra has a natural
tensor product,
N 6 6 6 6
H= hi = h1 h2 . . . hi . . . hN , (5.157)
i= 1
and the variables Xiα act nontrivially only on hi .
5.14.1 The XXX1/2 model

The XXX1/2 describes a spin chain model with N sites. At each site there is
a spin variable Siα = 2 σ α where σ α are the Pauli matrices. The Hilbert space
at each site is C 2 , the two-dimensional complex numbers. The Hamiltonian that
defines the model is based on a nearest neighbor interaction of the form,
1
Siα Si+ 1 −
α
H= . (5.158)
i,α
4
The spin of the system which is given by,

Sα = Siα , (5.159)
i

and is conserved,
[H, S α ] = 0. (5.160)
The notion XXX associates with the fact that the coefficient of (Siα Si+1 α
− 14 )
is a constant independent on i and α. In the case where the coefficient is α
1 − 4 ), the model is referred as the XYZ model.
1
dependent, namely J α (Siα Si+α
To extract the spectrum of the model we make use of the Lax operator defined
by,
6 6
Lk ,a (λ) = λIk Ia + i Skα σα , (5.161)
α
where λ is a complex parameter referred to as the spectral parameter, Ik , Skα are

the identity and spin operators acting on hk , and Ia , σ α act on an auxiliary space
V which is also L2 .
The Lax operator can also be written in the form,
i
Lk ,a (λ) = λ− Ik ,a + iPk ,a , (5.162)
2
7 2
where Pi,a is the permutation operator in L2 L , namely,
6 6
Pa b=b a, (5.163)
and is given by,

1 6 6
P = (I I+ σα σ α ). (5.164)
2 α
Equation (5.163) implies,
Pa 1 ,a 2 = Pa 2 ,a 1
Pn ,a 1 Pn ,a 2 = Pa 1 ,a 2 Pn ,a 1 = Pn ,a 2 Pa 2 ,a 1 . (5.165)
The Hamiltonian can also be expressed in terms of the permutation operator,

1 N
H= Pi,i+1 − . (5.166)
2 i 2
The idea now is to relate the Hamiltonian and other conserved charges to the
monodromy of a string of Lax operators along the whole chain. For that we
have to analyze the commuting structure of the Lax operators. This structure is
controlled by the fundamental commutation relations (FCR) which will be shown
later to be part of the Yang–Baxter relations and read,
Ra 1 ,a 2 (λ − μ)Li,a 1 (λ)Li,a 2 (μ) = Li,a 2 (μ)Li,a 1 (λ)Ra 1 ,a 2 (λ − μ), (5.167)
where,
Ra 1 ,a 2 (λ) = λIa 1 ,a 2 + iPa 1 ,a 2 . (5.168)

To prove this relation one makes use of the relations of the permutation operator
(5.165). The Lax operator can be interpreted as a connection along the chain in
the sense,
ψi+ 1 = Li ψi . (5.169)
This can be generalized to an ordered product which transports from i1 to i2 ,
Tii12,a (λ) = Li 2 −1,a (λ) . . . Li 1 ,a (λ), (5.170)
and to the full monodromy along the spin chain,
TN ,a (λ) = LN ,a (λ) . . . L1,a (λ). (5.171)
We parameterize this monodromy operator in terms of a 2 × 2 matrix in the

auxiliary space as,
AN (λ) a†N (λ)
TN ,a (λ) = , (5.172)
ãN (λ) DN (λ)
with entries in the full Hilbert space H. In analogy to the FCR of the basic Lax
operator it is straightforward to realize that there is a similar relation for the
monodromy operator,
Ra 1 ,a 2 (λ − μ)Ta 1 (λ))Ta 2 (μ) = Ta 2 (μ)Ta 1 (λ)Ra 1 ,a 2 (λ − μ). (5.173)
From this relation it follows that the trace of the monodromy operator,
F (λ) ≡ T r[T (λ)] = A(λ) + D(λ), (5.174)
is commuting, namely [F (λ), F (μ)] = 0. We can now expand both TN and F (λ)
as a polynomial of order N in λ as,
6
Ta,N (λ) = λN + iλN −1 Sα σα + . . .
α

N −2
F (λ) = 2λN + Ql λl . (5.175)
l= 0
We will see shortly that the set of N − 1 operators Ql are commuting and con-
stitute the set of conserved charges, including the Hamiltonian. Next we expand
the monodromy at λ = 2i ,
i
Ta,N = iN PN ,a PN −1,a . . . P1,a = iN P1,2 P2,3 . . . PN −1,N PN ,a . (5.176)
2
This follows from Li,a ( 2i ) = iPi,a and dλ

d
Li,a = Ii,a and then taking the permu-
tations one after the other of the term in the middle. The trace over the auxiliary
space is T r[PN ,a ] = IN so that we can now define a shift operator in H,

i
U = eiP ≡ i−N T ra TN = P1,2 P2,3 . . . PN −1,N . (5.177)
2

Using the properties of the permutation operator that P ∗ = P and P 2 = 1, it

follows that U is a unitary operator. Moreover, one can show that indeed it is a
shift operator, namely,
U −1 Xi U = Xi−1 . (5.178)
To expand F (λ) in the vicinity of λ = 2i we first observe that,
d
Ta (λ)|λ= 1/2 = iN −1 PN ,a . . . P̂i,a . . . P1,a , (5.179)
dλ i
whereˆmeans that the corresponding factor is absent. Using the same procedure
as above we find that,
d
Fa (λ)|λ=i/2 = iN −1 P1,2 P2,3 . . . PN −1,N . (5.180)
dλ i
Most of the permutations can be cancelled by multiplying with U −1 so that

d d 1
Fa (λ) Fa (λ)−1 |λ=i/2 = ln(Fa (λ))|λ=i/2 = Pi,i+1 . (5.181)
dλ dλ i i
Recalling the expression we found earlier for the Hamiltonian (5.166) we can now
see that,
i d N
H= ln(Fa (λ))|λ=i/2 − . (5.182)
2 dλ 2
We have just shown that the Hamiltonian is part of a set of N − 1 commuting
operators generated by F (λ), the trace of the monodromy. In fact there are N
such conserved charges if we add also one component, say S 3 , of the spin. The
model is characterized by its N degrees of freedom and is equipped with N
conserved charges and hence is (at least classically) integrable.
5.14.2 Bethe ansatz equations

To diagonalize the family of operators F (λ) one can generalize the procedure
used in the quantum harmonic oscillator. In that case we have a non-trivial
commutation relation [a, a† ] = 1, a Hamiltonian which is H = a† a + 1 and a
ground state which is annihilated by a, namely a|0> = 0. Let us start first with
the commutation relations. These are determined from the FCR as,
[ã(λ), ã(μ)] = 0
λ−μ−i i
A(λ)ã(μ) = ã(μ)A(λ) + ã(λ)A(μ)
λ−μ λ−μ
λ−μ+i i
D(λ)ã(μ) = ã(μ)D(λ) − ã(λ)D(μ). (5.183)
λ−μ λ−μ
The last two relations generalize the relations,
aH = (H + 1)a a† H = (H − 1)a† , (5.184)

of the harmonic oscillator. To derive the above one uses an explicit 4 × 4 matrix
7
formulation for the operators in V V . A natural basis for these matrices is
6 6 6 6
e1 = e+ e+ , e2 = e+ e− , e3 = e− e+ , e4 = e− e− , (5.185)
where,
1 0
e+ = , e− = . (5.186)
0 1
In this basis the permutation operator and the R matrix read,
⎛ ⎞
1 0 0 0
⎜0 1 0 0⎟
P =⎜ ⎝0
⎟ (5.187)
0 1 0⎠
0 0 0 1
⎛ ⎞
λ+i 0 0 0
⎜ 0 λ i 0 ⎟
R(λ) = ⎜⎝ 0
⎟. (5.188)
i λ 0 ⎠
0 0 0 λ+i
The matrices Ta 1 (λ) and Ta 2 (μ) read,
⎛ ⎞
A(λ) a(λ)
⎜ A(λ) a(λ) ⎟
Ta 1 (λ) = ⎜
⎝ a† (λ)
⎟
⎠ (5.189)
D(λ)
a† (λ) D(λ)
⎛ ⎞
A(μ) a(μ)
⎜ a† (μ) D(μ) ⎟
Ta 2 (λ) = ⎜
⎝
⎟. (5.190)
A(μ) a(μ) ⎠
a† (μ) D(μ)
Explicit multiplication of these matrices yields (5.183).
Similarly to the case of the harmonic oscillator we now define the ground state,
6
a(λ)|0> = a(λ) |0>i = 0. (5.191)
i
We choose |0>i = e+ so that,

N
S 3 |0> = |0> S + |0> = 0. (5.192)
2
Thus this state is the “highest weight state”. It also follows that (the operator
* is left unspecified),
λ+ i
∗
Ln (λ)|0>i = 2 |0>i , (5.193)
0 λ− i
2
and hence,

i N
λ+ ∗
T (λ)|0> = 2
N |0>, (5.194)
0 λ − 2i

which means that |0> is an eigenstate of both A(λ) and D(λ) and thus also of
F (λ). Higher excited states are created from the ground state |0> by a successive
action with creation operators,
Φ({λ}) = a† (λ1 ) . . . a† (λl )|0> . (5.195)
Requiring that the state Φ({λ}) is an eigenstate of F (λ) imposes a set of relations
on the λ1 , . . . , λl . In particular using the FCR relation (5.183) we find,
A(λ)a† (λ1 ) . . . a† (λl )|0>

l N
λ − λk − i i
= λ+ (λ)a† (λ1 ) . . . a† (λl )|0>
λ − λk 2
k=1

l
+ Mk (λ, {λ})a† (λ1 ) . . . â† (λk ) . . . a† (λl )|0> . (5.196)
k=1
The first term on the right-hand side of the equation has the form of an eigenstate
equations but the rest of the terms do not. The idea is to choose the set {λ} such
that these terms will cancel out against similar terms in D(λ)a† (λ1 ) . . . a† (λl )|0>.
To get the value of the coefficient M1 (λ, {λ}) we use the second term on the right-
hand side of the second equation in (5.183) when interchanging A(λ) and a† (λ1 )
and in all other exchanges we use the first term. In this way we find that,
l N
i λ1 − λk − i i
M1 (λ, {λ}) = λ1 + . (5.197)
λ − λ1 λ1 − λk 2
k=2
Interchanging now λ1 → λj we get similarly the expressions for all Mj (λ, {λ}).
The same type of manipulations yield,
D(λ)a† (λ1 ) . . . a† (λl )|0>

l N
λ − λk − i i
= λ− (λ)a† (λ1 ) . . . a† (λl )|0>
λ − λk 2
k=1

l
+ Nk (λ, {λ})a† (λ1 ) . . . â† (λk ) . . . a† (λl )|0>, (5.198)
k=1
with,
l N
i λ1 − λk − i i
Nj (λ, {λ}) = λ1 − . (5.199)
λ − λ1 λ1 − λk 2
k =2
We now observe that if the set of {λ} obey the condition,

l N l N
λ1 − λk − i i λ1 − λk − i i
λ1 + = λ1 − , (5.200)
λ1 − λk 2 λ1 − λk 2
k = j k = j

then the undesirable terms in (5.196) cancel out and we end with an eigenstate
of F (λ) and hence of the Hamiltonian. For the former it takes the form,
N l N l
i λ − λj − i i λ − λj − i
F (λ)Φ({λ}) = λ+ + λ− Φ({λ}).
2 λ − λj 2 λ − λj
j=1 k = j
(5.201)
These conditions can be rewritten as,
N l
λj + i/2 λj − λk + i
= . (5.202)
λj − i/2 λj − λk − i
k = j
These conditions on the eigenvectors were derived originally by Bethe, though

in a completely different way, and hence the names, the “Bethe ansatz equation”
(BAE) for (5.202) and the “Bethe vector” for Φ({λ}). It is straightforward to
observe that the poles in the eigenvalue of F (λ) cancel out so that it is a poly-
nomial in λ of degree N . One can further show that the full spectrum can be
recast just with nonequal λj .
We now want to determine the eigenvalues of spin, momentum and Hamilto-
nian operators of the eigenstates just found.
Using the FCR relation (5.173) in the limit of μ → ∞ the SL(2) invariance of
7
the monodromy in H V is determined via,

1 α α
Ta (λ), σ + S = 0, (5.203)
2
which means in particular that,
[S 3 , a† ] = −a† [S + , a† ] = A − D. (5.204)
The spin of the state is therefore given by,

N
S 3 Φ({λ}) = − l Φ({λ}). (5.205)
2
Furthermore it can be shown that the states Φ({λ}) are all highest weight states
provided that the BAE is obeyed, namely,
S + Φ({λ}) = 0. (5.206)
Since the S 3 eigenvalue of the highest weight states is non-negative it is obvious

that l ≤ N2 . When N is odd the spin of the state is half-integer, whereas when it
is even the spin is even and in particular for l = N2 there is an SL(2) invariant
state with vanishing spin.
Let us determine now the eigenvalue of the momentum operator. From the
definition of the shift operator (5.177) it follows that,
i λj + 2i
U Φ({λ}) = iN F Φ({λ}) = Φ({λ}). (5.207)
2 λj − 2i

The eigenvalue of the momentum operator is therefore given by,

1 λ + 2i
P Φ({λ}) = p(λj )Φ({λ}) p(λ) = log . (5.208)
j
i λ − 2i
The energy eigenvalue is determined from (5.182),

1 1
HΦ({λ}) = e(λj )Φ({λ}) e(λ) = − 2 1 . (5.209)
j
2λ + 4
The last expressions calls for a “quasi particle” interpretation. The operator
a† (λ) creates a quasi particle which reduces the spin S 3 by one unit and adds to
the momentum and energy p(λ) and e(λ), respectively. We further observe that,
1 d
e(λ) = p(λ). (5.210)
2 dλ
It is also evident that we can eliminate the dependence on the rapidity of the
energy and momentum and read directly the dispersion relation,
e(p) = cos(p) − 1. (5.211)
Since this energy is always non-positive, the highest weight state |0> can be
considered a ground state only if we take −H as the Hamiltonian rather than
H. In fact it will be shown shortly that the latter corresponds to a system of an
antiferromagnet whereas the former corresponds to that of a ferromagnet.
5.14.3 The thermodynamic Bethe ansatz

The thermodynamic limit of the spin chain models is the limit of N → ∞. Recall
that the continuum limit of the model when N → ∞ and the spacing Δ → 0. In
the BAE N appears only in the left-hand side. If we take the log of the ansatz
we find for real {λ},

l
N p(λj ) = 2πQj + ϕ(λj − lk ), (5.212)
k =1
where Qi are integers 0 ≤ Qj ≤ N − 1 that define the branch of the log and ϕ(λ)
λ+i
is a fixed branch of log( λ−i ). For large N and Qj and fixed l one finds the usual
expression for the momentum of a free particle on the chain,
Qj
pj = 2π , (5.213)
N
since the ϕ(λ) is negligible.
The second term in (5.212) associates with the scattering of these particles.
In fact by comparison with the quantum mechanics of a particle in a box we
see that ϕ(λi − λj ) stands for the corresponding phase shift of the particles with
rapidity λi and λj . Using this analogy we can now identify the S-matrix element

with,
λ−μ+i
S(λ − μ) = . (5.214)
λ−μ−i
The S-matrix also enters the large N commutation relations of the normalized
creation operators ã† (λ) = a† (λ)A−1 (λ),
ã† (λ)ã† (μ) = S(λ − μ)ã† (μ)ã† (λ). (5.215)
In addition to the quasi-particle states in the Hilbert space, there are also
bound states of the quasi-particles. These states correspond to complex solutions
of the BAE. The simplest case is with two quasi-particles l = 2. In this case the
two BAE read,
N N
λ1 + i/2 λ1 − λ2 + i λ2 + i/2 λ2 − λ1 + i
= . (5.216)
λ1 − i/2 λ1 − λ2 − i λ2 − i/2 λ2 − λ1 − i
Using (5.208) it follows that p(λ1 ) + p(λ2 ) is real. Furthermore, for N → ∞ to

compensate the exponential increase (decrease) of the left-hand side of the last
equations, it is clear that the right-hand side must have Im(λ1 − λ2 ) = i (or − i)
and thus the final form of λ1 and λ2 are,
i i
λ1 = λ1/2 + λ2 = λ1/2 − , (5.217)
2 2
where λ1/2 is real. The momentum and energy eigenvalues of the corresponding
Bethe vector are,
1 λ+i 1 d 1
p1/2 = ln e1/2 = [p1/2 (λ)] = 2 . (5.218)
i λ−i 2 dλ λ +1
The state is considered as a bound state since its energy is less than the sum of
the energies of the two constituents,
e1/2 < [e0 (p − p1 ) + e0 (p1 )], (5.219)
for any 0 ≤ p, p1 ≤ 2π.

The bound state of two quasi-particles l = 2 can be generalized for l > 2. The
roots λl are combined into complexes M , where M takes half integer values M =

0, 1/2, 1, . . . with l = M νM (2M + 1) where νM gives the number of complexes
of type M . Each complex has roots of the type,
λM ,m = λM + im − M ≤ m ≤ M, (5.220)
where λM is real and m are integers and half integers. The corresponding momen-
tum and energy are given by,
1 λ + i(M + 1/2) 1 2M + 1
pM (λ) = ln eM (λ) = . (5.221)
i λ − i(M + 1/2) 2 λ2 + (M + 1/2)2

The S-matrix for scattering of complexes M and N are

M +N
SM ,N = S0,L (λ), (5.222)
L =|M −N |
where,
λ + iM λ + i(M + 1)
S0,M (λ) = . (5.223)
λ − iM λ − i(M + 1)
To summarize, the ferromagnetic system with Hamiltonian −H in the ther-

modynamics limit has a Hilbert space HF with a ground state |0> = (e+ )i .
The excitations are quasi-particles characterized by M, M = 0, 1/2, . . . and the
rapidity λ. The dispersion relation is given by eM = 2M1+1 (1 − cos(pM )) and the
S-matrix is (5.222). Out of S α , only S 3 makes sense as an operator on HF . The
operators S ± change the ground state at each site. This may be viewed as a
symmetry-breaking phenomenon.
So far we have described the basic notions of the physics of the spin chain using
the example of the XXX1/2 . One can further generalize these considerations in
many directions such as the anti-ferromagnetic system, general spin states in the
XXX model, namely, the XXXs/2 model, the XXZ and many others.
The thermodynamic limit for the XXXs/2 model

So far we have described the basic notions of the physics of the spin chain using
the example of the XXX1/2 . One can further generalize these considerations
also to the XXXs/2 model. We state here the results without derivation. An
eigenstate characterized by the set λ1 , . . . , λl which are determined by a straight-
forward generalization of the BAE (5.202),
N l
λj + (i/2)s λj − λk + i
= . (5.224)
λj − (i/2)s λj − λk − i
k = j
The eigenvalues of the Hamiltonian and momentum are given by,

l
s 1
l
λk + 2i s
E= P = ln , (5.225)
λ2 + 14 s2
k=1 k
i
k =1
λk − 2i s
and the higher conserved charges that render the model integrable are,

i
l
1 1
Qr = − , (5.226)
r−1 (λk + 2i s)r −1 (λk − 2i s)r −1
k=1
where r ≤ N and in particular r = 2 is the Hamiltonian.

In the thermodynamic limit, N → ∞ states with low energy and zero momen-
tum can be dealt with by introducing the scaling,
1 s 1 s
l l l
2
s 1
Ẽ = N E = 2πn = 2πnk − = ,
N λ̃2
k =1 k
N λ̃
k=1 k j =1,j = k
λ̃ −
λ̃kλ̃kN
(5.227)
where λk = N λ̃k and where the second and third expressions were derived by
taking the log of the zero momentum condition U = 1 and of the BAE, respec-
tively. Using the same scaling one finds for the higher charges and the transfer
matrix the results,
1 s
l l
Qr s
1
Q̃r = = − i log T̃ (λ̃) = −i log T (N λ̃) = .
N r −1 N λ̃r λ̃ − λ̃k
N
k=1 k j =1,j = k j
(5.228)
For N → ∞ it is plausible to assume that the Bethe roots accumulate on
smooth contours (C1 , . . . CA ) ≡ C which are referred to as “Bethe strings”. Thus
we replace the discrete λ̃k locations of the roots by a continuum variable λ̃
described by a density ρ(λ̃) so that the sum of the root translates into the
integral,

1
l
→ dλ̃ρ(λ̃), (5.229)
N C k=1
with the normalization that C dλ̃ρ(λ̃) = N1 . Using the continuum formulation

we can now rewrite the expressions for the Bethe ansatz, the energy and the
higher charges as,

dλ̃ρ(λ̃) s dλ̃ ρ(λ̃ )
2πn = s 2πnλ̃ − = 2 ,

C λ̃ ũ C λ̃ − λ

dλ̃ρ(λ̃) dλ̃ρ(λ̃)
Ẽ = s Q̃r = s , (5.230)
λ̃ 2 λ̃r
C C
where nλ̃ is the mode number nk at the point λ̃ = λ̃k . It is expected to be a

constant along each contour Ca . In the second integral of the first line, a principal
part prescription is implemented. An important result is that one can determine
the set of conserved charges Q̃r using a resolvent, as,
∞
dλ̃ ρ(λ̃ )
G(λ̃) = |s| G(λ̃) = λ̃r −1 Q̃r . (5.231)
λ̃ −λ
C r =1
The resolvent is related to the transfer matrix,

1
T̃ (λ) = eiG ( λ̃) + e−iG ( λ̃− λ̃ ) = e− 2 ũ 2 cos G(λ̃) +
i i
. (5.232)
2λ̃
We wish to conclude this chapter with a model that will enable us to connect
the discussion of the spin chain models to continuum integrable models.

Fig. 5.5. Saw path on the two dimensional lattice.
5.14.4 Spin chain model in discrete time

So far we have discussed the two-dimensional integrable model with a discretized
space dimension and a continuous time dimension. It turns out that to connect
the spin chain models to continuum two-dimensional field theory it is useful to
further also discretize the time direction. The shift operator (13.77) discussed
above was determined from a trace of the monodromy at a particular value of the
spectral parameter. To describe the system with both space and time discretized,
one needs to distinguish values of the spectral parameter λ ± w for some fixed
w. We define now an inhomogeneous monodromy built from the Lax operator
Li,f acting on the quantum Hilbert space hi and the auxiliary space Vf ,
Tf (λ, w) = L2N ,f (λ + w)L2N −1,f (λ − w) . . . L2,f (λ + w)L1,f (λ − w). (5.233)
The light-like shift operators U+ and U− are given by,
U+ = trf [Tf (w, w)] U− = trf [Tf (−w, w)]. (5.234)
The monodromy is along a saw path on the two-dimensional lattice as can be

seen in Fig. 5.5.
L2n ,f (λ + w) is a transport along the NW direction and L2N −1,f (λ − w) along
the SW direction.
In analogy to the definition of the monodromy as a trace of T (λ) (13.77), we
now define the monodromy as,
F (λ, w|a, i, μ) = trf [Tf (λ, w|a, i, μ)], (5.235)
where
Tf (λ, w|a, i, μ) = L2N ,f (λ + w)L2N −1,f (λ − w) . . .

. . . L2i,f (λ + w)L−1
f ,a (μ − λ)L2i−1,f (λ − w) . . . L2,f (λ + w)L1,f (λ − w).
(5.236)
It can be shown that Tf (λ, w|a, i, μ) is subjected to the FCR relations in a similar
manner to T (λ) (13.77). Due to the commutativity of F we get a zero curvature
condition on the transport around an elementary plaquette of the space-time

lattice,
−1 −1
L2i,a (λ + w)L2i−1,a (λ − w) = U− L2i−1,a (λ − w)U− U+ L2i,a (λ + w)U+ .
(5.237)
The light-like shifts U± are related to the shift in space and time in the usual
form, namely,
U+ = e−i(H −P )/2 U− = e−i(H +P )/2 . (5.238)
As for the case of only space dimension discretized, here too one finds that the
condition of having an eigenvector of the energy and momentum is the BAE
which takes the form,
N
α(λj + w)α(λj − w)
= S(λj − λk ), (5.239)
δ(λj + w)δ(λj − w)
j = k
where α(λ), δ(λ) are local eigenvalues and S(λ) is the quasi particle phase factor.
5.14.5 The discretized version of the sine-Gordon model

As a particular example of an integrable lattice model, we consider a model with
Weyl variables rather than spin variables on the lattice sites. These variables
obey at each site the relations,
ui vi = qvi ui , q = eiγ . (5.240)
The corresponding Lax operator is

ui xvi
Li,a (x) = . (5.241)
−xvi−1 u−1
i
To reduce the number of degrees of freedom per site from two to one we impose
the constraint,
−1
u2i u2i−1 v2i v2i−1 = 1. (5.242)
The BAE for this case take the form,

N /2 l
sinh(λj + w + iγ/2)sinh(λj − w + iγ/2) sinh(λj − λk + iγ)
= ,
sinh(λj + w − iγ/2)sinh(λj − w − iγ/2) sinh(λj − λk − iγ)
k = j
(5.243)
λ w
where we have substituted x = e and κ = e .
The shift operator in the time direction related to the Hamiltonian can be
determined in the following manner. Using the explicit form of the Lax operator
we have,
1 1 x x 1 1
Li 2 ,a Li 1 ,a r =r Li 1 ,a Li 2 ,a . (5.244)
x y y y y x

zN
zW zE
zS
Fig. 5.6. Elementary plaquette.
It can be shown that this condition reduces to the functional equation,

r(x, qw) xw + 1
−1
= . (5.245)
r(x, q w) x+w
Denoting
−1 −1
w2i = u2i u2i−1 v2i v2i−1 w2i+1 = u2i+1 u2i v2i+1 v2i , (5.246)
we find that the time shift operator is given by,
e−iH = r(κ2 , w2i ) r(κ2 w2i+1 ). (5.247)
With this operator at hand we can now determine the equation of motion of the
model and show that it corresponds in the continuuum limit to the sine-Gordon
equation. To accomplish this we define now zi such that,
zi+1
wi = . (5.248)
zi−1
It is easy to see that zi does not commute only with one w, namely wi for which
we have,
zi wi = q 2 wi zi . (5.249)
Now we apply the time evolution operator on zi . It reads,
ẑi = eiH zi e−iH . (5.250)
When we substitute (5.247) we get the equation of motion,

κ2 q −1 z2i+2 + z2i
ẑ2i+ 1 = z2i+ 1 , (5.251)
q −1 z2i+2 + κ2 z2i
and similarly for z2i . This equation connects z along an elementary plaquette
(see Fig. 5.6). The equation can also be rewritten as,
(q −1 zN zW − zS zE ) = κ2 (q −1 zS zW − zN zE ). (5.252)

If we now define the variable,

z
χ = eiϕ = , (5.253)
z −1
alternatively on each second SE characteristic line on our lattice, then the equa-
tions of motion take the form,
κ 2 χW χE + 1
χN = χ − S −1 −1
, (5.254)
q χW χE + κ 2
so that for large κ2 and classical limit q = 1 we get,
χ N χS 1 1
=1+ 2 − χ E χW + ..., (5.255)
χE χW κ χE χW
and in terms of ϕ,
χ N χS Δ2 2 2
= ei 2 (∂ t ϕ−∂ x ϕ)+... , (5.256)
χE χW
1
and with the scaling κ2 = m2 Δ2 we finally discover the sine-Gordon equation,
(∂t2 ϕ − ∂x2 ϕ) + 2m2 sin(2ϕ) = 0. (5.257)
In the quantum version one modifies the scaling to take into account the mass
renormalization.
5.15 The continuum thermodynamic Bethe ansatz

Here we describe again the thermodynamic Bethe ansatz but now in the con-
text of continuous models. This will not be a straightforward transition from a
discretized to a continuous model via a certain limit, but rather a completely
different derivation. Obviously, here as well the Bethe wave-function, the ther-
modynamic limit and the interplay between the spectrum and S-matrix elements
will enter as essential ingredients.
Consider an integrable Euclidean field theory defined on a two-dimensional
torus. We denote the two cycles of the torus as cycle a and cycle b, with corre-
sponding circumferences of R and L and coordinates x and y, as shown in Fig.
5.7. Obviously, one can define in a twofold manner the states of the system and
its Hamiltonian. We can consider the space of states on a denoted by A with the
time direction along y, with the Hamiltonian,

1
Ha = Ty y dx, (5.258)
2π a
and with the momentum,

1
Pa = Txy dx, (5.259)
2π a
2πn
which has quantized eigenvalues R .

C
L
R B
y
Fig. 5.7. Flat torus generated by two orthogonal geodesics C and B of

circumference R and L, respectively.
Alternatively, one can consider the space of states B along the contour b with
the time direction along −x, with the Hamiltonian,

1
Hb = Txx dy, (5.260)
2π b
and with the momentum,

1
Pb = − Ty x dy, (5.261)
2π b
where now the quantization condition is that the eigenvalues of Pb are quantized
in units of 2πn
L .
Let us consider the cylindrical geometry via the limit L → ∞ (L R). For
this case the partition function Z(L, R) is dominated by the ground state of Ha
with the ground state energy E0 (R),
Z(R, L) ∼ e−L E 0 (R ) . (5.262)
On the other hand,
Z(R, L) = T rB [e−R H b ]. (5.263)
In B, the thermodynamic limit, namely infinite space L → ∞ which is the
analog of the large N limit in Section 5.14.3, gives the free energy f (R) at
temperature 1/R, via log Z(R, L) ∼ −Lβf (β), where β = R is the inverse of the
temperature. Hence,
E0 (R) = Rf (R). (5.264)
The ground state energy E0 (R), which can be referred to also as the Casimir
energy, can be related in the limit of conformal field theory, to the Virasoro
anomaly. Define the scaling factor c̃(r) via,
πc̃(r)
E0 (R) = −
, (5.265)
6R
where the dimensionless quantity r = m1 R, with m1 the lowest mass in the
theory. The scaling factor will be determined by the TBA. On the other hand,

in the case of conformal invariance, when R → 0, using the relation between the
Hamiltonian and L0 and L̄0 we have,
2π 1
E0 (R) = Δm in + Δ̄m in − c , (5.266)
R 12
where Δm in denotes the conformal dimension of the lowest state. This means
that the scaling factor should reduce to the effective Virasoro anomaly,
lim c̃(r) = c − 24Δm in . (5.267)

r →0
Here we took the cases of Δm in = Δ̄m in .

The TBA method enables one to compute the spectrum of energies and
momenta by combining the thermodynamic limit with the factorizable scatter-
ing amplitudes. We consider first for simplicity the case with only one type of
particle of mass m with a pair scattering amplitude S(β1 − β2 ) (see Section 5.7).
Recall (5.73) that the energy and momentum are related to the rapidities βi as
ei (β) = m coshβi pi (β) = m sinhβi , (5.268)
and that the amplitudes obey the unitarity and crossing symmetry (5.78),
S(β)S(−β) = 1 S(β) = S(iπ − β). (5.269)
For regions of configuration space where the particles are highly separated,
namely where |xi − xj | Rc , i, j = 1, . . . , N with Rc denoting the scale of the
interaction, the particles can be treated as approximately free. In these regions
it makes sense to introduce the wavefunction of the system Φ(x1 , . . . , xN ) (in
regions where the particles are not well separated, particle creation and annihi-
lation prevent the use of a single wavefunction).
For integrable systems at any free region the number of particles will be the
same, namely N , as well as the set of momenta pi . The set of particles in a
free region will be denoted by (i1 , i2 , . . . , iN ) where xi 1 < xi 2 < . . . < xi N . A
scattering process that yields a transition between (i1 , i2 , . . . , ik , ik +1 , . . . , iN )
and (i1 , i2 , . . . , ik +1 , ik , . . . , iN ) affects the wavefunctions, such that the latter
wavefunction is given by the former multiplied by the scattering amplitude
S(βk − βk + 1 ). These matching conditions on the wavefunctions combined with
the quantization of the momenta due to the fact that the “space” coordinate is
compact, lead to the relation,

eip i L S(βi − βj ) = 1, Or mL sinhβi + δ(βi − βj ) = 2πni , (5.270)
j = i j = i
where for real β we rewrote the scattering amplitude in terms of a unimodular

function S(β)eiδ (β ) , with δ(β) a real phase.
This set of transcendental equations selects the admissible sets of rapidi-
ties and hence energies and momenta. The energy and momentum of the state

(β1 , . . . , βN ) are given by,

N
N
Hb = m coshβi , Pb = m sinhβi . (5.271)
i= 1 i=1
In the thermodynamic limit of L → ∞ the set of equations (5.270) can be

simplified. In this limit the number of particles becomes large since it grows ∼ L
and the distance between adjacent levels behaves as (βi − βi+1 ) ∼ m1L . In that
case one naturally defines the notion of a continuous rapidity density of particles
ρ1 (β). If there are n particles in a Δβ we take ρ1 (β) = Δnβ . We can now exchange
the sum over βi with an integral over β of the particle density, provided that the
latter is independent on Δβ for m1L Δβ 1, so that (5.270) takes the form,

mL sinhβi + δ(βi − β )ρ1 (β )dβ = 2πni . (5.272)
If one further defines the level density ρ(β) such that n = ρ(β)dβ we get,

mL coshβ + ϕ(β − β )ρ1 (β )dβ = 2πρ(β), (5.273)
where ϕ(β) = ∂ δ∂(β

β .
)
In terms of the particle density the energy of the system is,

Hb = m coshβρ1 (β)dβ. (5.274)
At this point one has to distinguish between bosons and fermions. From the
unitarity condition one can have either S(0) = 1 or S(0) = −1. In the former case
bosons occupy each rapidity value in any number whereas for fermions the occu-
pation number is at most one. In the latter case the situation is in some sense
the opposite. S(0) = −1 implies that for two particles with the same rapidity
the wavefunction is antisymmetric in their coordinates, and since this is incom-
patible with bose statistics, this state should be excluded. Hence the bosonic
particles for S(0) = −1 behave like fermions, and indeed we will refer to this
case as “fermionic”. On the other hand identical particles that are fermion states
of identical rapidity are allowed, and will be referred to as part of a “bosonic”
system.
Now we would like to address the issue of the entropy for both the bosonic
and fermionic cases. For small intervals of the rapidity Δβα 1 but such that
m L Δβα there is a large number of levels Nα ∼ ρ(βα )Δβα and large number of
1
particles nα ∼ ρ1 (βα )Δβα that are distributed among these levels. The number
of different distributions for the two cases are,
(Nα )! (Nα + nα − 1)!
“fermionic” = , “bosonic” = . (5.275)
(nα )!(Nα − nα )! (nα )!(Nα − 1)!

The entropy S(ρ, ρ1 ) = log N (ρ, ρ1 ), where N (ρ, ρ1 ) is the total number of states,
is given by,

SFerm i = dβ[ρ log ρ − ρ1 log ρ1 − (ρ − ρ1 ) log(ρ − ρ1 )],

SBose = dβ[−ρ log ρ − ρ1 log ρ1 + (ρ + ρ1 ) log(ρ + ρ1 )]. (5.276)
Computing the partition function by performing the trace over all the states
of the system translates to the minimization of the free energy,
−RLf (ρ, ρ1 ) = −RHb (ρ1 ) + S(ρ, ρ1 ), (5.277)
with respect to the densities ρ and ρ1 , subjected to the constraints (5.273). Using
(5.273) the extremum equations take the form,

dβ
−Rm coshβ + (β) ± ϕ(β − β ) log(1 ± e−(β ) ) = 0, (5.278)
2π
where the upper sign is for the fermionic case, the lower for the bosonic, and the
“pseudoenergies” (β) are defined via,
ρ1 e−(β )
= . (5.279)
ρ 1 ± e−(β )
The extremal free energy is,

dβ
Rf (R) = ∓m (coshβ) log(1 ± e−(β ) ) . (5.280)
2π
Comparing (5.273) with (5.280) determines a useful relation,
L ∂(β)
ρ(β) = . (5.281)
2π ∂R
Finally we can also write down the TBA expressions for the expectation values
of Tμ,ν . Using the last relation and (5.279) we get,

dE(R) 2π
< Txx > = 2π = m ρ1 (β) coshβdβ,
dR L
2π d[RE(R)]
< Tμμ > =
R dR
2π 1 ρ1 ∂(β)
=m [ρ1 (β) coshβ − sinhβ]dβ. (5.282)
L R ρ ∂β
These relations generalize in a straightforward manner to the more general case
of N types of particles with masses ma , a = 1, . . . , N̂ and scattering amplitudes
Sab which are now N̂ × N̂ matrices.
For integrable models where one can take the limit of rRm1 → 0, with m1
being the lowest mass of the system, one can determine the scaling function c̃(r).
In this limit (β) become constant in the regime where − ln(2/r) β ln(2/r).
Their constant value is determined by the limit of equation (5.273) which now

for the case of N particles reads,

N̂
1
N̂
a = − dβφab (β) ln[1 + e− b ] = − [δab (+∞) − δab (−∞)] ln[1 + e− b ],
2π
b=1 b=1
(5.283)
recall that φab (β) = ∂β δab (β).
Interpolating between these values of (β) for β ln(2/r) and the values for
β → ∞ where (β) go exponentially to infinity, and taking finally the r → 0
limit one finds the effective Virasoro anomaly using the relation between the free
energy and the scaling factor (5.265),

N
6
N
1
c̃(0) = c̃a (a ) = L , (5.284)
a= 1
π2 a=1 1 + e a
where c̃a is the scaling factor associated with the particle of type a and L(x) is
the dilogarithm function,

1 x ln t ln(1 − t)
L(x) = − + dt. (5.285)
2 0 1−t t
Thus the determination of the Virasoro anomaly of the underlying CFT of our
massive integrable theory with a purely elastic S-matrix follows from the solution
for the pseudo-energies (5.283) and plugging it into the expression of the scaling
factor. One can proceed with the continuous TBA and determine not only the
ground state energy but also the full spectrum of energies as well as the spectrum
of the eigenvalues of all the conserved charges, as was described for the discretized
case discussed in Section 5.14. This is beyond the scope of this book. We refer
the reader to the relevant papers in the reference list.

6
Bosonization
In one space dimension there are obviously no rotations and hence no angular
momentum. This raises the possibility of equivalence relations between scalar
fields and fields of higher tensorial structure like spinors, vectors etc. However,
spinors and scalars seem to be distinct even in two dimensions due to their
different statistics. An equivalence between these two types of fields should
therefore incorporate the identification of operators, made out of scalars, that
are anti-commuting and vice versa. It is well known that a bilinear of fermi fields
is a commuting field, but it is less obvious how to construct a field that obeys
the Fermi–Dirac statistics from scalars. This is precisely what the bosonization
procedure does.
Coleman [63] and Mandelstam [159] introduced the concept of bosonization.
Their construction is now referred to as the “abelian bosonization”. An anti-
commuting Fermi field, constructed from the exponential of a boson, was given
explicitly by Mandelstam [159].
The fact that the theories of a free massless scalar and a free Dirac fermion are
equivalent can be proven by showing that they fall into the same representation
of the affine current algebra and the Virasoto algebra. The bosonic–fermionic
duality can also be further elevated to the free massive theories and also to
interacting ones.
It turns out that the original abelian bosonization is not convenient to accom-
modate color (or flavor) degrees of freedom and hence is inconvenient to address
systems like QCD2 . A breakthrough in that direction was achieved by Witten,
in his non-abelian bosonization [224].1
The equivalence enables one to use, as convenient, either the fermionization
of scalar fields or the bosonization of fermions. The latter is useful in several
cases. For instance in the case of duality between the Thirring model [205] and
the sine-Gordon model,2 which will be discussed in Section 6.2, the bosonization
takes the form of a strong–weak duality. For strong fermionic interactions one
finds a weak bosonic coupling. In applications to gauge theories (Section 9) it will
be shown that the one loop anomaly behavior is encoded in classical bosonized
theory. In QCD2 , as will be discussed in Section 9.3.2, the bosonic version of
1 This paper discusses Majorana fermions. The construction for Dirac fermions was done in [7].
2 This was proved by Coleman [63].

132 Bosonization
the theory admits a separation between the color and flavor degrees of freedom,
which is very useful in describing the low energy color singlet states.
We start this chapter by introducing the set of rules that span abelian
bosonization, including the rules for mass terms and the equivalence of the inter-
acting Thirring model and the sine-Gordon model. We then describe Witten’s
non-abelian bosonization of Majorana fermions. This is further generalized to the
case of massless Dirac fermions. We discuss the subtleties of the massive case and
present two methods of handling the non-abelian bosonization of massive Dirac
fields.3 We then discuss in detail the action formulation of the chiral bosonization
both abelian and non-abelian. We then depart from the applications that will be
found to be relevant to QCD2 and present topics in bosonization which are more
relevant to conformal field and string theories like the bosonization of ghost fields
and the Wakimoto bosonization [213]. We do not discuss bosonization on higher
Riemann surfaces. The interested reader can consult for instance [211] and [84].
The topic of bosonization in two-dimensional field theories has been reviewed
in several papers and books, like that of Stone [202]. Here we mainly follow the
review of Frishman and Sonnenschein [101] for the basic ingredients, and update
it to include more recent topics.
6.1 Abelian bosonization

6.1.1 Bosonization of a free massless Dirac fermion
Both the theory of a free massless real scalar field and the theory of a free massless
Dirac field are conformal field theories invariant under affine Lie algebra. Recall
from Chapter 2 that the former theory is defined by the action,

1
S = d2 xL = d2 x∂ν φ̂∂¯ν φ̂
8π

1 1
= 2
d ξ∂ξ φ̂∂ξ¯φ̂ = d2 z∂ φ̂∂¯φ̂. (6.1)
4π 4π
The solution of the equation of motion takes the form,
φ̂(z, z̄) = φ(z) + φ̄(z̄). (6.2)
The theory has holomorphically (and anti-holomorphically) conserved currents,

¯ = −i∂¯φ̄(z̄),
J(z) = i∂φ(z) J(z) (6.3)
and similarly holomorphic (and anti-holomorphic) energy-momentum tensors,

1 1
T (z) = − : ∂φ∂φ := − : J(z)J(z) :
2 2
1 ¯ ¯ 1 ¯ ¯
T̄ (z̄) = − : ∂ φ̄∂ φ̄ := − : J(z̄)J(z̄) :, (6.4)
2 2
3 A bosonization prescription for the mass term in the flavored case was suggested in [75] and
[99].

which admit a Virasoro algebra with c = 1 and affine Lie algebra with level
k = 1.
Recall also that the theory of a free massless Dirac field with the action,

1
S= d2 z(ψ † ∂ψ
¯ + ψ̃ † ∂ ψ̃) (6.5)
4π
admits conserved currents,
J(z) = ψ † ψ ¯ = ψ̃ † ψ̃,
J(z̄) (6.6)
and its energy-momentum tensor can be expressed as a bilinear of the currents
using the Sugawara construction,
1 1 1
T (z) = − [ψ † ∂ψ − ∂ψ † ψ] = − : ψ † ψψ † ψ := − : J(z)J(z) : . (6.7)
2 2 2
The correponding level of the affine algebra and of the Virasoro anomaly are
again k = 1, and c = 1, respectively.
Due to the uniqueness of the irreducible unitary k = 1 representation of the
affine Lie algebras, and the fact that the infinite-dimensional algebraic structure
fully determines the theories, we conclude that in two space-time dimensions the
theories of massless free scalar field and Dirac field are equivalent.
The equivalence implies that every operator of one theory should have a part-
ner in the other theory, in such a way that the OPEs of these dual operators
should be identical. We have just realized such correspondence for the currents
and energy-momentum tensor, namely,
Jb (z) = ∂φ(z) ↔ Jf (z) =: ψ † ψ(z) : ,
1 1
Tb (z) = − : ∂φ∂φ : ↔ Tf (z) = − [ψ † ∂ψ − ∂ψ † ψ], (6.8)
2 2
and similarly for the anti-holomorphic counterparts.
For completeness we now redescribe the currents using the “old” terminology
of vector and axial currents. The vector current reads,
1
JVμ =: ψ̄γ μ ψ := − √ μν ∂ν φ. (6.9)
π
This identification of J μ leads automatically to a conserved current,
∂μ JVμ = 0, (6.10)
independent of the equations for φ. This is a “topological” conservation, con-
nected with choosing the “vector conservation” scheme. In the applications to
follow, we will demand more freedom in the scheme choice of interacting theories,
in particular the possibility to have a vector current anomaly. The bosonization
procedure will therefore be somewhat modified. The modification will correspond
to a change of regularization scheme.
The overall coefficient of the current is such that the fermion number charge,
∞
Q= dxj0 (x) = 1, (6.11)
−∞

134 Bosonization
for the ψ-field. In addition to the “topologically” conserved vector current, the
bosonic theory has an axial current, which is equivalent to the fermionic axial
current,
1
JAμ =: ψ̄γ μ γ 5 ψ := √ ∂ μ φ. (6.12)
π
The bosonic current is the Neother current associated with the invariance of
the bosonic action under the global shift δφ = . The holomorphic and anti-
holomorphic conserved currents discussed above are (in real coordinates) nothing
μ
but the left and right chiral currents J± = Jvμ ± JAμ , which correspond to shifts
with (x+ ) and (x− ). Using the commutation relation (8.4) the ALA reads,
2i
J± (x± ), J± (x± ) = δ (x± − x± ). (6.13)
π
This is the same algebra as that of the fermionic chiral currents. The Sugawara
construction in this terminology reads,
T±± = π : J± J± : . (6.14)
These obey the Virasoro algebra,

i
T± (x± ), T± (x± ) = 2i T± (x± ) + T± (x± ) δ (x± − x± ) − δ (x± − x± ),
6π
(6.15)
which is identical to that of the fermionic energy-momentum tensor.
The equivalence of the bosons and the fermion bilinears is not only mathe-
matical. The fermion Fock-space contains those bosons as physical states. The
reason for this is that in one space dimension a massless field can move either to
the left or to the right. A Dirac fermion and its anti-particle having together zero
fermionic charge and moving in the same direction will never separate. They are
therefore indistinguishable from a free massless boson. This picture changes when
masses are introduced, and the above relations will be approached at momenta
high compared to the mass scale (including high off-mass shell).
A natural question to ask is which operator of the bosonic picture corre-
sponds to the basic Dirac ferion? Since the latter is in fact a combination of
a left chiral spinor and a right one, we would like to determine the “bosonized”
Weyl fermion ψ(z). It is a holomorphic function of conformal dimension 1/2,
that transforms under the affine Lie transformation with a unit charge, namely,
ψ(z) → ei(z ) ψ(z). Due to the fact that under the same transformation the scalar
field transforms as φ(z) → φ(z) + e(z) we are led to look for a candidate which
is an exponential in the scalar field eiα φ(z ) . We now use T (z) given in (6.4) to
compute the confomal dimension of : eiα φ(z ) : as follows,
α2
: eiα φ(w ) :
T (z) : eiα φ(w ) := 2
+ ..., (6.16)
(z − w)2

α2
where . . . stands for non-singular terms, Hence the conformal dimension is 2 .
Thus we conclude that the following equivalence should hold,
ψ(z) ↔ eiφ(z ) , ψ † (z) ↔ e−iφ(z ) . (6.17)
To confirm this bosonization rule we compute the OPEs in both descriptions and
verify that they are indeed identical,
1
ψ(z)ψ † (−z) = + :ψ(0)ψ † (0): +2z : [ψ(0)∂ψ † (0): −∂ψ(0)ψ † (0))]+O(z 2 )
2z
1
ψ(z)ψ † (−z) = + J(0) + 2zT (0) + O(z 2 )
2z
1
: eiφ(z ) :: e−iφ(−z ) : = + J(0) + 2zT (0) + O(z 2 )
2z
1
: eiφ(z ) :: e−iφ(−z ) : = + i∂φ(0) + 2z(∂φ∂φ(0)) + O(z 2 ). (6.18)
2z
The bosonic version of the fermion ψ was originally proposed by Mandelstam.
His formulation was done in terms of real coordinates and cannonical quanti-
zation. For completeness we now also present the “old” construction and proof
of equivalence. The bosonized chiral fermion in the latter formulation takes the
form,4
5 x
cμ √
ψL = : exp − i π dξ[π(ξ) + φ(x)] :
2π −∞
5 x
cμ √
ψR = : exp − i π dξ[π(ξ) − φ(x)] :, (6.19)
2π −∞
where π(x) = φ̇(x) is the conjugate momentum of φ(x), c is a constant. A compu-

tation yields c = 12 eγ ∼ 0.891, where γ is the Euler constant. The normal ordering
denoted by : : is performed with respect to the scale μ.
The equal time commutation relations of the φ-field,
[φ(x, t), π(y, t)] = iδ(x − y),
imply, upon using the formula eA eB = e[A ,B ] eB eA (for [A, B] a c-number) the
canonical anti-commutation relations for the ψ field,
: † ;
ψL,R (x, t), ψL,R (y, t) = δ(x − y). (6.20)
The fermion field ψ is therefore, an inherently non-local functional of the scalar
field. However fermion bilinears, such as the currents discussed above or the mass
terms that will be described in the next section, are local functions.
So far we have addressed the map for massless theories. Let us now discuss the
bosonization of a fermion mass bilinear operator. The mass term which mixes
4 A generalization of this bosonization to a set of N fermions was done by Halpern [121].

136 Bosonization
the left and right chiral componenets of the Dirac fermion takes the following
well-knows form,
mf [ψ̃ † (z̄)ψ(z) + ψ † (z)ψ̃(z̄)] = mf (ψL† ψR + ψR† ψL ). (6.21)
Again for completeness we write down the expression both in the complex coor-
dinates as well as in real coordinates.
Using the bosonization rules for chiral fermions (6.17) (to be justified below),
we deduce the map of the fermion bilinear to the equivalent bosonic operator,
mf [: ei φ̄( z̄ ) :: eiφ(z ) : + : e−iφ(z ) :: e−i φ̄( z̄ ) :] = mf μ : cos(φ̂(z, z̄) :, (6.22)
where we have made use of φ̂(z, z̄) = φ(z) + φ̄(z̄) and of the fact that there is no
non-trivial OPE between φ(z) and φ̄(z̄). Note that we write down the bosonic
equivalent of the mass term operator in the context of the massless theory and
hence the factorization to holomorphic and anti-holomorphic parts of the scalar
field holds. Once we identify this operation relation we will then use it to add a
fermion mass term to the bosonized action. The additional parameter which has
a dimension of mass μ is the normal ordering scale.
The derivation of the bosonized mass term in the “old language” is somewhat
more involved. We will return to this after we address the bosonization duality
between the fermionic Thirring model and the bosonic sine-Gordon model.
We now summarize the equivalence relations between the bosonic and
fermionic operators of the free theories, in both the “modern” complex coor-
dinate formulation, as well as the “old” formulation in terms of real coordinates:
Operator Fermionic Bosonic
J (z) : ψ † ψ(z) : i∂φ(z)

† ¯
J¯(z̄) : ψ̃ ψ̃(z̄) : −i∂φ(z̄)
T (z) − 12 : [ψ † ∂ψ − ∂ψ † ψ] : − 12 : ∂φ∂φ(z) :
T̄ (z̄) − 12 : [ψ̃ † ∂ ψ̃ − ∂ ψ̃ † ψ̃]: ¯ ∂φ(z̄)
− 12 : ∂φ ¯ :
fermionL ψ(z) : ei φ (z ) :
fermionR ψ̃(z̄) : ei φ ( z̄ ) :
mass term ψ̃ † (z̄)ψ(z) + ψ † (z)ψ̃(z̄) μ : cos φ̂(z, z̄) :
Bosonization in “modern” complex formulation.
6.2 Duality between the Thirring model and the

sine-Gordon model
The Thirring model is a fermionic theory with a current–current interaction,
given by the Lagrangian density,
1
L = iψ̄ ∂ψ − gJ μ Jμ ,
2

6.2 Duality between the Thirring model and the sine-Gordon model 137
J+ (x+ ) : ψL† ψL : ∂+ φ
−
J− (x ) : ψR† ψR : ∂− φ
T+ + (x+ ) − 12 : [ψL† ∂ψL − ∂ψL† ψL ] : − 12 : ∂+ φ∂+ φ(x+ ) :
T−− (x− ) − 12 : [ψR† ∂ψR − ∂ψR† ψR ] : − 12 : ∂− φ∂− φ(x+ ) :

cμ √ x
fermionL ψL (x )+
2π
: exp −i π dξπ(ξ) + φ(x) :
−∞

−
cμ √ x
fermionR ψR (x ) 2π
: exp −i π dξπ(ξ) − φ(x) :
−∞
mass term ψL† (x+ )ψR (x− ) μ : cos φ̂(x+ , x− ) :

+ ψR† (x+ )ψL (x− )
Bosonization in “old” formulation.
where Jμ = : ψ̄γμ ψ : . The model is exactly solvable and meaningful for g > −π.
The corresponding equation of motion reads,
i ∂ψ(x) = gγμ J μ (x)ψ(x). (6.23)
The theory is invariant under vector and axial U (1) global transformations.
The corresponding conserved currents are,
ν
JμV = J μ = : ψ̄γμ ψ : JμA = μν J V . (6.24)
The model can be studied by means of the operator product expansion on the
light-cone [76]. The fermionic bilinears of the model are expressed as a function
of the current, and the expressions obtained turn out to be very natural in the
light of the bosonization procedure, which we now describe.
We start with the following generalization of the bosonization formula (6.19):
⎛ ⎛ ⎞⎞
5 √ x
cμ √ 2 π β
ψL = : exp ⎝−i π ⎝ dξπ(ξ) + √ φ(x)⎠⎠ :
2π β 2 π
−∞
⎛ ⎛ ⎞⎞
5 √ x
cμ √ 2 π β
ψR = : exp ⎝−i π ⎝ dξπ(ξ) − √ φ(x)⎠⎠ :. (6.25)
2π β 2 π
−∞
The meaning of the new parameter β will be clarified shortly. In a similar man-
ner to the derivation of (6.20) we can verify that the equal-time anti-commutation
relations are still obeyed. Furthermore one can show that the Dirac operator
built from (6.25) obeys the equation of motion (6.23) provided that the bosonic
field φ(x) obeys the equation of motion of the sine-Gordon model discussed in
Section 5.3, namely,
μ2
∂μ ∂ μ φ(x) + : sin(βφ(x)) := 0, (6.26)
β

138 Bosonization
and that the parameter β is related to the coupling constant g in the following
way,
β2 1
= g . (6.27)
4π 1+ π
From this last relation it follows that the special value β 2 = 4π corresponds to
g = 0 and hence a free Dirac fermion. Indeed as we shall see below for that value
of β the sine-Gordon potential translates into the bosonized mass term. It is
interesting to note a remarkable property of (6.27) which relates the coupling
constants of the Thirring model and its bosonic equivalent, the sine-Gordon
model. The weak coupling of one theory is the strong coupling of the other.
This property often occurs in bosonized theories and hints at the usefulness
of the method in dealing with theories for strong coupling, where perturbative
methods fail.
The bosonization dictionary of the vector fermion number current is the fol-
lowing,
β μν
JμV = : ψ̄γμ ψ : ↔ J μ = − ∂ν φ. (6.28)
2π
This expression differs from the bosonized current of the free Dirac theory (6.9)
in its normalization factor. We immediately realize that for the special value
β 2 = 4π we precisely reproduce (6.9). The normalization factor can be deter-
mined from the assignment of the fermion number charge of a soliton that should
be equal to the charge of the field ψ. Recall from Section 5.3 that the classical
sine-Gordon model admits a finite energy soliton solution. It is time-independent
and interpolates between adjacent wells of the scalar potential. In quantum the-
ory this classical solution becomes a particle. The static soliton solution is given
by (5.26),
4
φ= tan−1 [expμ(x − x0 )],
β
where x0 is the “center” of the soliton. Substituting this into the integral of the
current we find the fermion number of this solution to be,
β
Q= [φ(∞) − φ(−∞)] = 1. (6.29)
2π
Thus we see that indeed the normalization factor in the vector current (6.28) is
the right one.
β2
The relation (6.28) implies that the level in the affine Lie algebra will be 4π ,
as compared to 1 for the free case.
Let us address again the issue of the bosonization of the fermion mass bilinear
(6.22). The definition of the mass term in the “old formulation”, as that of the
current, requires some care due to the appearance of the products of operators

at the same point. In fact when x approaches y one gets the following OPEs,
cμ
ψR† (x)ψL (y) = |cμ(x − y)|δ : e−iβ φ :
2π
cμ
ψL† (x)ψR (y) = |cμ(x − y)|δ : eiβ φ : (6.30)
2π
β2
with δ = − 2π
g
(1 + 4π ). The proper fermion mass term will therefore be defined
by,
∞ ∞
−δ cμ
limy →x dx |cμ(x − y)| mψ̄(x)ψ(y) = m dx : cosβφ(x) : .
π
−∞ −∞
μπ
With μ chosen such that m = cβ 2 , the mass term transforms in the bosonic
language to,
μ2
ΔL = : cosβφ : .
β2
The normal ordering is with respect to μ.5
6.3 Witten’s non-abelian bosonization

The non-abelian bosonization introduced by Witten is a set of rules assigning
bosonic operators to fermionic ones, in a theory of free fermions invariant under a
global non-abelian symmetry.6 Originally the fermions considered were Majorana
fermions and the corresponding global symmetry was O(N ). The bosonic opera-
tors are not expressed in terms of free bosonic fields as in abelian bosonization,
but rather in terms of interacting group elements. In particular, bosonic expres-
sions can be written for the energy-momentum tensor, various chiral currents,
the mass term and the complete action.
The generalization to the case of Nf Dirac fermions was introduced in [112]
and [7].
6.3.1 Bosonization of Majorana fermions

Let us start with N free Majorana fermions governed by the action,

N
i
Sψ = d2 x (ψLk ∂+ ψLk + ψRk ∂− ψRk )
2
k=1
where ψL , ψR are left and right Weyl–Majorana spinor fields, ∂± = √12 (∂0 ± ∂1 )
and k = 1, . . . , N . The corresponding bosonic action is the Wess–Zumino–Witten
5 The flavored Thirring model was studied in [73].

6 A related approach is presented in [179].

140 Bosonization
(WZW) action discussed in Chapter 4:

1
Sb [u] = d2 xT r(∂μ u∂ μ u−1 )
16π

1
+ d3 yεij k T r(u−1 ∂i u)(u−1 ∂j u)(u−1 ∂k u), (6.31)
24π B
where u is a matrix in O(N ) whose elements are bosonic fields. The second
term, the Wess–Zumino (WZ) term, is defined on the ball B whose boundary Σ is
taken to be the Euclidean two-dimensional space-time. Now, since π2 [O(N )] = 0,
a mapping u from a two-dimensional sphere S into the O(N ) manifold can be
extended to a mapping of the solid ball B into O(N ). The WZ term however is
well defined only modulo a constant. It was normalized so that if u is a matrix
in the vector representation of O(N ) the WZW term is well defined modulo
W Z → W Z + 2π. The source of the ambiguity is that π3 [O(N )] Z, namely
there are topologically inequivalent ways to extend u into a mapping from B
into O(N ).
Note that O(2) is an exception, as π3 [O(2)] = 0.
Note that the equivalence is between a fermionic theory expressed in terms
of an N -dimensional fermion in the vector representation of O(N ) and a group
element which is an N × N matrix. Nevertheless, as will be shown below the two
theories are fully equivalent.
Both the theory of N free Majorana fermions and the WZW model of (15.2)
are invariant under ALA transformations of OL (N ) × OR (N ). The latter take
the following forms for the bosonic and fermionic theories:
u → g(z)u ψi → [g(z)]ji ψ(z)j

u → uh(z̄) ψ̃i → [h(z̄)]ji ψ̃(z̄)j , (6.32)
where g(z) ∈ OL (N ) and h(z̄) ∈ OR (N ). The corresponding currents in both

pictures satisfy the ALA at level k = 1.
The two theories are also invariant under the conformal transformations,
z → f (z) z̄ → f¯(z̄). (6.33)
The associated Virasoro central charges of the two descriptions are identical,
as follows,
1 k[dim O(N )] 1/2N (N − 1) N
cf = N × cb = = = . (6.34)
2 k+N −2 1+N −2 2
For the fermions it is just N times the central charge of a single Majorana
fermion, whereas for the bosonized version we make use of the fact that the dual
Coexter number of O(N ) is N − 2. The conformal invariance of the action (15.2)
can be also shown by realizing that the corresponding β function vanishes. If one
generalizes (15.2) by taking a coupling 4λ1 2 as a coefficient of the first term and
k
24π of the WZ term (k integer), the β function associated with λ is given at the

one loop level (in the sense of expanding around u = 1), by,
2

dλ2 (N − 2)λ2 λ2 k
β≡ =− 1− ,
dlnΛ 4π 4π
namely (15.2) is at a fixed point for λ2 = 4π

k and hence exhibits conformal invari-
ance there. By showing that the energy-momentum tensor obeys the Virasoro
algebra, one can show that this property is in fact exact.
To summarize, the dictionary that translates the ALA currents and the energy
momentum tensor of the fermionic theory, into the bosonic one and vice versa,
is given by,
Ji j (z) : ψi ψj (z) : iN
4π
[u−1 ∂u]i j (z)
J¯i j (z̄) : ψ̃i ψ̃j (z̄) : iN ¯ −1 ]i j (z̄)
[u∂u
4π

T (z) − 12 N i = 1 [ψi ∂ψi − ∂ψi ψi ]: − 2 (N1−1 ) : J a J a (z) :
N
T̄ (z̄) − 2 i = 1 : [ψ̃∂ ψ̃i − ∂ ψ̃i ψ̃i ]:
1
− 2 (N1−1 ) : J¯a J¯a : (z̄)
6.3.2 Bosonization of Dirac fermions

The bosonic picture for the theory of N free massless Dirac fermions is built
from a boson matrix g ∈ SU (N ) and a real boson φ. The bosonized action now
has the form,

1
S[g, φ] = d2 xT r(∂μ g∂ μ g −1 )
8π

1
+ d3 yεij k T r(g −1 ∂i g)(g −1 ∂j g)(g −1 ∂k g)
12π B

1
+ d2 x∂μ φ∂ μ φ. (6.35)
2
Note the difference of factor two between the WZW action associated with the
SU (N ) and the O(N ).
Here again both theories are conformal invariant with an identical Virasoro
central charge,
k dim SU (N ) N2 − 1
cf = N × 1 cb = +1= + 1 = N. (6.36)
k+N 1+N
ALA transformation with respect to global SUL (N ) × SUR (N ) × U (1),
g → h(z)u, φ → φ + a(z); ψi → [g(z)]ji ψ(z)j

g → uh̄(z̄), φ → φ + ā(z̄); ψ̃i → [h(z̄)]ji ψ(z̄)j , (6.37)
leave the actions of both pictures invariant.

142 Bosonization
One way to prove the equivalence of the fermionic and bosonic theories now,
for N free massless Dirac fermions and the k = 1 WZW theory on U (N ) group
manifold, is by showing that the generating functionals of the current Green
functions of the two theories are the same. For the fermions we have,

−iW ψ (A μ ) 2
ψ
e = (dψ+ dψ− dψ̄+ dψ̄− )ei d x ψ̄ i D , (6.38)
(1)
μ ( 2 T ) + Aμ × 1 and ( 2 T ) generators of
1 A 1 A
where Dμ = ∂μ + iAμ , Aμ = AA
SU (N ). The term Wψ (Aμ ) was calculated by Polyakov and Wiegmann in a
regularization scheme which preserves the global chiral SU (NL ) × SU (NR ) sym-
metry and the local U (1) diagonal symmetry, leading to,

1
Wψ (Aμ ) = S[Ã] + S[B̃] + d2 xA(1)
μ A
μ(1)
, (6.39)
4πN
where Ã, B̃ ⊂ SU (N ) are related to the gauge fields AA
μ by
−1 −1
iAA
+ = (Ã ∂+ Ã)A , iAA − = (B̃ ∂− B̃)A .
In the bosonic theory one calculates,

e−iW B (A μ ) = [du]eiS [u ]+i d x(J − A + +J + A − )
A 2 B B B B

2 2 (1) (1)
e−iW B (A μ ) = [dφ]e 2 d x[(∂ φ) +(J − A + +J + A − )]
(1) i
(6.40)
(1) i −1
where J+B AB− and J+ A− are the appropriate parts of 4π T r[(g ∂+ g)A− ], and
(1)
similarly for the (− +) case and with A± = T r(A± ). These functional integrals
can be performed exactly, leading to,

1
WB (AA μ ) = S[Ã] + S[B̃] W B (A (1)
) = d2 xA(1)
μ A
μ(1)
.
4πN
Thus the bosonic current Green functions are identical to those of the fermionic
theory, the latter regulated in the way mentioned above.
6.3.3 The bosonization of a mass bilinear of Dirac fermions

A further bosonization rule has to be invoked for the mass bilinear. For a theory
with a U (N ) symmetry group the rule is,
√ 4π
†l
ψ+ ψ−j = c̃μNμ gjl e−i N φ , (6.41)
where Nμ denotes normal ordering at mass scale μ. The fermion mass term
mq ψ̄ i ψi is therefore,

m Nμ d2 xT r(g + g † ),
2

where m 2 = mq c̃μ, mq is the quark mass, and c is the same constant as in
(6.19). It is straightforward to show that the above bosonic operator transforms

correctly under the U (N )L × U (N )R chiral transformations. On top of that it

has the correct total dimension,
N −1 1
Δ = Δg + Δφ = + = 1, (6.42)
N N
where Δg = NN−1 and Δφ = N1 are the dimensions associated with the SU (N )

and U (1) group factors, respectively. Moreover in Section 4.4 it was explicitly
shown that the four-point function,
G(zi , z̄i ) = <g(z1 , z̄1 )g −1 (z2 , z̄2 )g −1 (z3 , z̄3 )g(z4 , z̄4 )>, (6.43)
is given by,
G(zi , z̄i ) = [(z1 − z4 )(z2 − z3 )(z̄1 − z̄4 )(z̄2 − z̄3 )]−Δ g G(x, x̄), (6.44)
where G(x, x̄) is a function of the harmonic quotients,

(z 1 −z 2 )(z 3 −z 4 ) ( z̄ 1 −z̄ 2 )( z̄ 3 −z̄ 4 )
x= (z 1 −z 4 )(z 3 −z 2 ) and x̄ = only, and in the free case is,
( z̄ 1 −z̄ 4 )( z̄ 3 −z̄ 2 )

1 1 1 1 1
G(x, x̄) = [xx̄(1 − x)(1 − x̄)] N × I1 + I2 I¯1 + I¯2 , (6.45)
x 1−x x̄ 1 − x̄
where I1 , I2 , I¯1 , I¯2 are group invariant factors. This result for the correlation
function, combined with the U (1) part gives an expression identical to that for
the fermionic bilinears. Moreover the result can be generalized to an n-point
function.
6.3.4 Bosonization of Dirac fermions with color and flavor

In his pioneering work on non-abelian bosonization Witten also proposed a pre-
scription for bosonizing Majorana fermions which carry both NF “flavors” as
well as NC “colors”, namely transform under the group [O(NF ) × O(NC )]L ×
[O(NF ) × O(NC )]R . The action for free fermions is

i
Sψ = d2 x(ψ−ai ∂+ ψ−ai + ψ+ai ∂− ψ+ai ),
2
where now a = 1, . . . , NC and i = 1, . . . , NF are the color and flavor indices,

respectively. The equivalent bosonic action is,
1 1
S̃[g, h] = NC S[g] + NF S[h]. (6.46)
2 2
The bosonic fields g and h take their values in O(NF ) and O(NC ), respectively
and S[u] is the WZW action given in (15.2).

144 Bosonization
The bosonization dictionary for the currents was shown to be,

iNC −1 iNC
J+ij =: ψ+ai ψ+aj := (g ∂+ g)ij J−ij =: ψ−ai ψ−aj := (g∂− g −1 )ij
2π 2π
(6.47)
iNF −1 iNF
J+ab =: ψ+ai ψ+bi := (h ∂+ h)ab J−ab =: ψ−ai ψ−bi := (h∂− h−1 )ab ,
2π 2π
(6.48)
where : : stands for normal ordering with respect to fermion creation and anni-
hilation operators. As for the bosonic expressions for the currents, regularization
is obtained by subtracting the appropriate singular parts.
In terms of the complex coordinates z = ξ1 + iξ2 , z̄ = ξ1 − iξ2 (where ξ1
and ξ2 are complex coordinates spanning C 2 , and the Euclidian plane (ξ1 → x,
ξ2 → −t) and Minkowski space-time (ξ1 → x, ξ2 → −it) can be obtained as
appropriate real sections), one can express the currents as
iNC iNC −1
J(z)ij ≡ πJ−ij = (g∂z g −1 )ij J(z̄)ij
¯ ≡ πJ+ij = (g ∂z̄ g)ij ,
2 2
and similarly for the flavored currents.
In a complete analogy the theory of NF × NC Dirac fermions can be expressed
4π
NFNC −i φ
in terms of the bosonic fields g, h, e now in SU (NF ), SU (NC ) and
U (1) group manifolds respectively. The corresponding action is now,

1
S[g, h, φ] = NC S[g] + NF S[h] + d2 x∂μ φ∂ μ φ. (6.49)
2

4π
NCNF −i φ
This action is derived simply by substituting ghe instead of u in (6.31).
As for the equivalence between the bosonic and fermionic theories, we note
that in both theories the commutators of the various currents have the same
current algebra, and the energy-momentum tensor is the same when expressed in
terms of the currents. But the situation changes when mass terms are introduced
(see next section). The bosonization rules for the color and flavor currents are
obtained from (6.47) and (6.48) by replacing the Weyl–Majorana spinors with
Weyl ones, and in addition we have the U (1) current,
5
√ (1) † NF NC
J (z) ≡ πJ− = : ψ−ai ψ−ai :=
(1)
∂−φ
π
5
√ (1) † NF NC
¯
J (z̄) ≡ πJ+ = : ψ+ai ψ+ai :=
(1)
∂+φ . (6.50)
π
The affine Lie algebras are given by,
A B i
Jn , Jm = if A B C JnC+m + knδ A B δn +m ,0 ,
2
where J A = T r(T A J), T A the matrices of SU (NC ), k = NF for the colored cur-

rents and J(z) is expanded in a Laurent series as J(z) = z −n −1 Jn . A similar

expression will apply for the flavor currents with T I the matrices of SU (NF ),
¯
and the central charge k = NC instead of NF . The commutation relation for J(z̄)
will have the same form.
Generalizing the case of SU (N ) × U (1) to our case, the Sugawara form for the
energy-momentum tensor of the WZW action is given by,
1 A 1 I
T (z) = : J (z)J A (z) : + : J (z)J I (z) :
2κC 2κF
A I
1
+ : J (1) (z)J (1) (z) :, (6.51)
2κ
where the dots denote normal ordering with respect to n (n > 0 meaning anni-
hilation). The κs are constants yet to be determined. In terms of the affine Lie
generators this can be written as,
∞
∞

1 A A 1 I I
Ln = : Jm Jn −m : + : Jm Jn −m :
2κC m =−∞
2κF m =−∞
∞
1 (1)
+ : J (1) J : (6.52)
2κ m =−∞ m n −m
Now, by applying the last expression on any primary field φl we can get a set of
infinitely many “null vectors” of the form,

1
0
χnl = Ln − : JA JA :
2κC m =n m n −m

1 1
0 0
(1) (1)
− :J JI I
:− :J J : φl ,
2κF m =n m n −m 2κ m =n m n −m
for any n ≤ 0 (for n > 0 holds immediately). Since each of these vectors must
certainly be a primary field, Lm χn = Jm A n
χ = Jm I n
χ = Jm χn = O, which holds
trivially for m > 0. When checking for m ≤ 0, it leads to expressions for the vari-
ous κ, for the central charge c of the Virasoro Algebra, and for the dimensions of
the primary fields Δl = Δl+ + Δl− , in terms of NC , NF and the group properties
of the primary fields,
1
κC = κF = (NC + NF ), κ = NF N C
2
NC (NF2 − 1) NF (NC2 − 1)
c= + + 1 = NF N C
(NC + NF ) (NC + NF )
(c2l± )F (c2l± )C (c2 )(1)
Δl± = + + l± (6.53)
(NF + NC ) (NF + NC ) NC NF
where (c2l± )C is the eigenvalue of the SUR,L (NC ) second Casimir operator in
the representation of the primary field φl , namely ( 12 T A )( 12 T A ) = (c2l )C I, and
similarly for the flavor group. In the cases of SU (NC ) and SU (NF ) the discussion

146 Bosonization
2 2
applies to Δl+ or Δl− separately, with Cl+ and Cl− , respectively. Note that the
expressions for κF and κC of equation (6.53) are an immediate generalization
of the case of the group SU (N ) with the central term equal to one. There the
factor was N + 1, the N being the second Casimir of the adjoint representation,
and the 1 being the central term.
The equivalence of the bosonic and fermionic Hilbert spaces was demonstrated
by showing that the two theories have the same current algebra (affine Lie alge-
bra), and that the energy-momentum tensor can be constructed from the cur-
rents in a Sugawara form. Goddard et al. [110] showed that a necessary and
sufficient condition for such a construction of the fermionic T (z), in a theory
with a symmetry group G, is the existence of a larger group G ⊂ G such that
G /G is a symmetric space with the fermions transforming under G just as the
tangent space to G /G does. Based on this theorem they found all the fermionic
theories for which an equivalent WZW bosonic action can be constructed. The
cases stated above fit in this category. Note in passing that this does not hold for
cases where the symmetry group includes more non-abelian group factors, like
for example SU (NA ) × SU (NF ) × SU (NC ) × U (1).
The prescription equation (6.49) described above , for the bosonic action that is
equivalent to that of colored and flavored Dirac fermions, is by no means unique.
In fact it will be shown that this prescription will turn out to be inconvenient
once mass terms are introduced. Another scheme, based on the WZW theory of
U (NF NC ) will be recommended.
6.3.5 Bosonization of mass bilinears in the product scheme

A natural question here is how to generalize the rule (6.41) to Majorana fermions
with action (6.46), and its analog for the case of SU (NF ) × SU (NC ) × U (1) given
in (6.49). We call the latter the product scheme. The bosonization rule for the
latter case is,

4π
†ai −i φ
ψ+ ψ−bj = c̃μNμ gji hab e NFNC
. (6.54)
Consequently, the bosonic form of the fermion mass term mq ψ̄ ia ψia is,

4π

2 2 † † −i N F N C φ
m Nμ d x(T rgT rh + T rh T rg )e , (6.55)

with m 2 = mq c̃μ. Once again the bosonic operator (6.54) has the correct chiral
transformations and the proper dimension,
NF2 − 1 NC2 − 1 1
Δ = Δg + Δh + Δφ = + + = 1.
NF (NF + NC ) NC (NC + NF ) NC NF
Unfortunately, the explicit calculation of the four-point function reveals a dis-
crepancy between the fermionic and bosonic terms in (6.54). This can actually
be understood directly. Since g and h are fields defined on entirely independent

group manifolds, then (ignoring for a moment the U (1) factor) the four-point
function of the mass term can be written as,
<g(z1 , z̄1 )g −1 (z2 , z̄2 )g −1 (z3 , z̄3 )g(z4 , z̄4 )>

<h(z1 , z̄1 )h−1 (z2 , z̄2 )h−1 (z3 , z̄3 )h(z4 , z̄4 )> .
This expression differs from the corresponding fermionic Green’s function, as

it includes independent “contractions” for the g and h factors, whereas in
the fermionic correlation function the flavor and color contractions are corre-
lated. Moreover, the expressions for the bosonic Green’s function involve hyper-
geometric functions, and do not resemble the case of free fermions, which is a
product of poles.
6.3.6 Bosonization of the U(NF × NC ) WZW action

It is clear from the previous discussion that the bosonization prescription for
our case needs alteration. A priori there can be two ways out, either modifying
the rule for the bosonization of the mass bilinear or using a different bosonic
scheme altogether. As for the first approach, (6.54) preserves the proper chiral
transformation laws under the product group SU (NF ) × SU (NC ) × U (1) as well
as the correct dimension, and therefore the number of possible modifications
is very limited. For example, one might think of multiplying the expression in
(6.54) by an operator which is a chiral singlet under the above group, with zero
dimension. We do not know of such a modification. Therefore we are going to try
a different bosonic theory than (6.49). The symmetry of the free fermionic theory
can actually be taken as UL (NF × NC ) × UR (NF × NC ) rather than [SU (NF ) ×
SU (NC ) × U (1)]L × [SU (NF ) × SU (NC ) × U (1)]R . The natural bosonic action
is hence a WZW theory of u ⊂ U (NF NC ) and with k = 1. The currents are now,
i ¯ α β = i (u−1 ∂z̄ u)α β ,
J(z)α β = (u∂z u−1 )α β J(z̄)
2 2
with α, β running from 1 to NF × NC . The expressions for the flavor and color
currents can be obtained by appropriate traces, over color for the flavor currents
and over flavor for the color currents.
The mass bilinear is now,
†
ψ+α ψ−β = c̃μNμ uα β ,
where now the U (1) term is absorbed into u.

Clearly the requirement for Sugawara construction of T, for proper chiral
transformations of all the operators and for a correct dimension for the mass
bilinear are fulfilled. Since now the flavor and color degrees of freedom are
attached to the same bosonic field, the previous “contraction problem” in the n
point functions is automatically resolved. Moreover as stated above the four-point

148 Bosonization
function and in fact any Green’s function will now reproduce the results of the
fermionic calculation.
The currents constructed from u obey the Affine Lie algebra with k = 1.
The color currents, for instance, are J A = T r(T A J), where T A are expressed as
(NC NF ) × (NC NF ) matrices defined by λA ⊗ 1, with λA the Gell–Mann matri-
ces in color space and 1 stands for a unit NF × NF matrix. The central charge
is k = NF . The same arguments will apply for the flavor currents, now with
k = NC . The central charge for the U (1) current is NC NF .
To see the difference between the present theory and the previous one let us
express u in terms of (NF NC ) × (NF NC ) matrices g̃, h̃ and ˜l in SU (NF ), SU (NC )
and the coset-space,
SU (NF × NC )/{SU (NF ) × SU (NC ) × U (1)}
respectively, through,

4π
−i φ
u = g̃ h̃˜le NC NF
.
Using the formula for expressing an action of the form S[AgB −1 ] we get

˜ 1
S[u] = S[g̃ h̃l] + d2 x∂μ φ∂ μ φ
2
1
S[g̃ h̃˜l] = S[g̃] + S[˜l] + S[h̃] + d2 xT r(g̃ † ∂+ g̃˜l∂− ˜l† + h̃† ∂+ h̃˜l∂− ˜l† ).
2π
We can now choose ˜l = l so that l∂− l† will be spanned by the generators that
are only in SU (NF × NC )/{SU (NF ) × SU (NC ) × U (1)}. This can be achieved
by taking ũ = g̃ h̃˜l, which is u but without the U (1) part, and then taking for
h̃ = h ⊗ 1 a solution of the equation ∂− hh† = N1F T rF [(∂− ũ)ũ† ], and similarly for
g with N1C T rC . These are also the conditions that the flavor currents should be
expressed in terms of g̃ and the color currents in terms of h̃. For this choice, the
mixed term in the above action, the term involving products of ˜ls with g̃s or h̃s,
is zero, and so the new action is,

1
S[u] = NC S[g] + NF S[h] + d2 x(∂μ φ∂ μ φ + S[l]).
2
Note that l is still an SU (NC NF ) matrix, while g and h are expressed as SU (NF )
and SU (NC ) matrices respectively, but the matrix l involves only products of
color and flavor matrices (not any of them separately).
6.4 Chiral bosons

So far we have discussed the bosonization of a Dirac fermion via abelian bosoniza-
tion, and N Dirac fermions using the U (N ) WZW model, or N Majorana
fermions with an SO(N ) WZW model. What about the bosonization of chi-
ral left or right Weyl fermions? In the fermionic language it is trivial to write an

action of a fermion with one given chirality. It is also easy to factorize a scalar
field into its left φL (x− )(φ(z)) and right moving φR (x+ )(φ̄(z̄)) parts since the
solution of equation of motion of a scalar field in real and complex coordinates
takes the form
φ(x+ , x− ) = φL (x− ) + φR (x+ ) or φ(z, z̄) = φ(z) + φ̄(z̄). (6.56)
However, as will be discussed shortly it turns out that it is quite subtle to write
down an action of a chiral boson which is equivalent to that of a left or a right
chiral fermion. Once we establish an action for a chiral boson, the question is
how can one couple it to abelian and non-abelian gauge fields?
In this section we will construct two seemingly independent constructions of
the action of a chiral boson. In fact, it will turn out that one formulation is a
special case of the other. We start with Siegel’s formulation which is based on
the coupling of a scalar field to fictitious gravity in a light-cone gauge [194] and
then we describe a manifestly non-Lorentz invariant action [92] which is a special
case of the former.7 In [36], [199], and in [100] the two formulations were related
and further generalizations were discussed. We follow the latter paper here.
Chiral bosons play an important role in string theories and in particular chiral
bosons on Riemann surfaces of any genus. Here we will not enter into discussions
on these constructions and describe chiral bosons only on a two-dimensional flat
Minkowsky space-time.
6.4.1 Chiral boson via coupling to fictitious “light-cone gravity”

A scalar field in two space-time dimensions couples to two-dimensional gravity
via the following well-known action,
1√
L= ggα β ∂ α φ∂ β φ, (6.57)
2
where gα β is the two-dimensional metric. In the “ light-cone gravitational gauge”
the metric has the form,
g + + = 0 g +− = 1 g −− = 2λ. (6.58)
In this gauge the action (6.57) reads,
L = ∂+ φ∂− φ + λ(∂− φ)2 . (6.59)
Since we have fixed only part of the local symmetries of (6.57) it is straightfor-
ward to realize that the last action is still invariant under the following local
transformation
δφ = − ∂− φ δλ = −∂+ − + − ∂− λ − ∂− (− λ) (6.60)
7 The anomalies of the system were analyzed in [129].

150 Bosonization
which is the transformation of the scalar field under a combination of x− coor-

dinate transformation and a Weyl rescaling,
δx− = − (x) δgα β = −gα β ∂− − . (6.61)
It is important to emphasize that we are in fact considering a flat two-dimensional

Minkowski space-time and 2λ is the −− component of a fictitious metric. The
action (6.59) is further invariant under the global shift symmetry φ → φ + a. We
denote the corresponding current as the axial current defined in (6.12) which
takes the following form,
+ −
J(ax) = ∂− φ J(ax) = ∂+ φ + 2λ∂− φ. (6.62)
−
+
One can also define J− (ax) = J(ax) and J+ (ax) = J(ax) − 2λJ− (ax) , namely using
the fictitious metric to raise and lower indices. In addition we have the topological
vector conserved current J(v ) μ = μν ∂ ν φ. The vector and axial currents defined
√
here are those defined in (6.9), times a factor of − π. Following from these
two currents we can obviously also define the left and right conserved currents
Jl/r = 12 (J(v ) ± J(ax) ).
The equations of motion derived by variations with respect to λ and to φ are,
δλ : (∂− φ)2 = 0 δφ : ∂+ ∂− φ + ∂− (λ∂− φ) = 0. (6.63)
These equations imply classically the chiral nature of the boson, namely a left
moving boson φ(x+ ) and the conservation of the axial current. Note that unlike
an ordinary scalar field the chiral scalar action (6.59), admits only the “holomor-
phic” conservation of the left current but not the “anti-holomorphic” conserva-
tion of the right one namely,
∂− J(l) − = 0 ∂+ J(r ) − = −∂− J(r ) − = 0, (6.64)
which implies that there is only left affine symmetry but not a right one.
So far we have discussed the classical system. Quantum mechanically it turns
out that the symmetry (6.60) is anomalous. There are several ways to verify this
anomaly. Probably the easiest way is to realize the resemblance of the action
(6.57) to that of the bosonic string. It is well known that the latter is consistent
only in 26 dimensions and not, as our case seems to be, in one dimension. Tech-
nically this follows from the fact that the ghost system associated with the fixing
of the fictitious diffeomorphism and Weyl invariance have a Virasoro anomaly
(see Section 6.5.1) equal to −26. In fact following the discussion in Section 6.5.1
we know that there is another way to cancel this anomaly and that is to add to
the action a background charge term of the following form,
L = Lcl + qR(2) φ = Lcl + q∂−

2
λφ = Lcl − qλ∂−
2
φ, (6.65)
where q is the background charge and R(2) is the fictitious scalar curvature.
This modification of the action yields a modified energy-momentum tensor as

will be discussed in Section 6.5.1, of the form,

φ 1
T−− = − (∂− φ)2 − q∂−
2
φ. (6.66)
2
To fix the gauge associated with the symmetry (6.60) one introduces a (b, c) ghost
system (see Section 6.5) that contributes c = −26 to the Virasoro anomaly. The
2
modified energy-momentum tensor (6.66) contributes c = 1 + 6πq so that to
25
cancel the anomaly one has to take the background charge q = 6π and hence
the quantum mechanical action of the chiral boson is (6.65) with the value of q
just quoted.
There are several possible methods to quantize this action. One approach is to
follow the BRST quantization of the bosonic string. In this procedure one uses
the Nielpotent operator associated with the Noether charge that corresponds to
the BRST symmetry

φ 1 ghost
QBRST = dxc T−− + T−− , (6.67)
2
to construct the space of physical states. The latter furnish the cohomology of
QBRST , namely,
QBRST |phys> = 0 |phys> = QBRST |state> . (6.68)
For the vanishing ghost sector this implies that,
φ (+ )
T−− |phys> = 0 → a(k)|phys> = 0 f or k > 0, (6.69)
φ (+ ) φ
where T−− is the positive frequency part of T−− and a(k) is the annihilation
operator of momentum k. Thus the space of physical states is made of only
left-moving (k > 0) states.
Once the local symmetry (6.60) has been made non-anomalous, one can safely
choose any gauge fixing. In particular we can take λ = 0 while keeping the con-
struction of the physical states discussed above, or fixing λ = −1. As will be seen
in the following section this gauge will turn out to be convenient and rather than
addressing the issues of coupling to abelian and non-abelian gauge fields of the
action (6.59) we will do it instead with the gauge fixed action.
Another approach of quantizing the system is based on the implementation of
Dirac brackets. Starting from (6.59) and its corresponding Hamiltonian density,

1 (π + λφ )2
H= + (1 − λ)(φ )2 , (6.70)
2 1+λ
we realize that the conjugate momentum of λ, πλ vanishes. This is a primary con-
straint χ1 = πλ = 0. Requiring that this constraint is preserved in time, namely,
2
χ̇1 = {πλ (x), H} = 0 we find a secondary constraint χ̃2 = (π−φ1+λ
)
= 0. The Pois-
son bracket of χ1 and χ̃2 vanish and hence they are a first-class constraint. How-
ever, if we replace χ̃2 with its classical equivalent constraint χ2 = π − φ then
the latter is a second-class constraint. If we add the additional constraint in the

152 Bosonization
form of gauge fixing χ3 = λ(x) − λ0 (x) with λ0 (x) a given function, than all the
constraints are second order with the constraint algebra,
cij (x, y) = {χi (x), χj (y)} c22 = −2δ (x − y) c13 = −δ(x − y). (6.71)
The next step is to define the Dirac bracket,

{F (x), G(y)}D = {F (x), G(y)} − dzdw{F (x), χi (z)}c−1
ij (z, w){χj (w), G(y)},
(6.72)
where cij (x, z)c−1
jk (z, y) = δ ik (x − y). The Dirac bracket rather than the Poisson
bracket is then elevated to the commutator in the quantum theory [ ] = i{ }.
Using this prescription one finds the desired result,
1
[φ(x), φ(y)] = (x − y). (6.73)
4i
Implementing the constraint quantization in the path integral formulation, one
has,

Z(J) = [dφ][dπ][dλ][dπλ ]δ(χ1 )δ(χ2 )δ(χ3 )
2
× Det[Cij (x, y)]ei d x(πφ̇+πλ λ̇−H−J φ) . (6.74)
Using the δ functions and since Det[C] is field independent we find,

Z[J] = [dφ]ei d x L̃−J φ L̃ = φ̇φ − (φ )2 .
2
(6.75)
Thus we see that this procedure yields the action (6.59) in the gauge λ = −1
which will be the topic of the following section.
6.4.2 Non manifestly Lorentz invariant classical action

Let us start with the following non-local action,

L = dzdyρ(z)(z − y)ρ̇(y) − dxρ2 (x), (6.76)
where ρ(x) is a local bosonic field. The system can also be described in terms of
the non-local bosonic field,

1
φ(x) = dy(x − y)ρ(y), φ (x) = ρ(x), (6.77)
2
with a local Lagrangian density,
L = φ φ̇ − φ = ∂− φ(∂+ φ − ∂− φ).
2
(6.78)
As we said the Lagrangian density is in fact (6.59) in the gauge 2λ = −1. The
classical equation of motion which corresponds to (6.76) is,
∂− φ = 0. (6.79)

This equation has a general solution ∂− φ = g(t), however by requiring ∂− φ = 0

at the spatial boundaries, we set g(t) = 0 and get the chiral solution ∂− φ = 0.
Even though the action (6.76) is not manifestly Lorentz invariant, it is easy to
see that it is invariant under time translation δφ = φ̇, space translation δφ = φ ,
and the unconventional Lorentz transformation δφ = (t + x)φ . For the above
invariance transformations to exist we assume vanishing surface terms. The asso-
ciated Noether charges are H = φ2 , P = −H and M = dx[(x + t)(φ )2 ]. The
system is, in fact, invariant under yet another unusual Lorentz transformation,
δ(∂− φ) = x+ ∂+ (∂− φ) − x− ∂− (∂− φ) − ∂φ .
In addition, equation (6.76) is invariant under the global axial transformation
μ
φ → φ + α. The associated current j(ax) and the vector current which as was
discussed in Section 6.1 is “topologically conserved” are given by
j(ax) + = ∂+ φ − 2∂− φ j(ax) − = ∂− φ

j(v ) + = ∂+ φ j(v ) − = −∂− φ. (6.80)
As usual from the vector and the axial currents we can write down the left
and right currents J(lr) = 12 [j(v ) ± j(ax) ], respectively. They have the following
expressions,
J(l) − = 0, J(l) + = φ; J(r ) − = −∂− φ, J(r ) + = ∂− φ. (6.81)
Note however that only the left current is holomorphically conserved namely
∂− J(l) + = 0, while the right current is not antiholomorphically conserved. This
property is related to the invariance of the Lagrangian under δφ = α(x+ ) and
not under δφ = α(x− ). As will be explained below only the left U (1) affine Lie
algebra current exists in the quantum theory. Similarly, the Lagrangian (6.76) is
invariant under the conformal transformations δφ = (x+ )φ and δφ = (t)∂− φ.
The associated Sugawara type Noether currents are,
T(l) + + = (φ )2 = J(l) 2+ ∂− T(l) ++ = 0
T(r ) 11 = (∂− φ)2 = J(r ) 21 ∂1 T(r ) ++ = 0. (6.82)
Quantization of the model

By treating the system as a constrained system, we now invoke its canonical
and path-integral quantization. We repeat here the Dirac bracket procedure dis-
cussed above. The constraint χ = π − φ = 0, is a second-class constraint since
{χ(x), χ(y)} = −2δ(x − y). The Hamiltonian density of the system takes the fol-
lowing form H = Hc + v(φ, π)χ , where Hc = φ2 = π2 = πφ . Using this form for
H it is easy to verify that the condition χ̇ = {χ, H} = 0, does not lead to further
constraints but fixes v(φ, π). The passage to the quantum theory is performed
by passing from the Dirac brackets rather than the usual Poisson bracket to

154 Bosonization
the commutator via,

−i[F (x), G(y)] = {F (x), G(y)} − dξ1 dξ2 {F (x), G(ξ1 )}

1
× − (ξ1 − ξ2 ){F (ξ2 ), G(y)}. (6.83)
4
Following this definition, the operator algebra for π and φ takes the form,
1
[φ(x), φ(y)] = (x − y),
4i
1
[π(x), φ(y)] = δ(x − y),
2i
1
[π(x), π(y)] = δ(x − y). (6.84)
2i
One also finds that, for an arbitrary operator F (φ, π), [χ(x), F (y)] = 0 so that
the constraint π(x) − φ (x) = 0, is now realized at the operator level. For example
the Hamiltonian density can now be expressed in the three forms of Hc mentioned
above. The system is solved by Fourier transforming,
0
dk
φ(x) = √ [ak e−ik x + a†k eik x ]
−∞ 2 πk
π(x) = φ (x), (6.85)
with the usual algebra,
[a(k), a† (k )] = δ(k − k ). (6.86)
Note that only k ≤ 0 appears in the decomposition of φ(x), which expresses the
chiral nature of the field. The single-particle Hilbert space is then a continuum
of states with energy E = |k|, k ≤ 0. Hence the Hamiltonian formalism has cor-
rectly implemented the chirality constraint ∂− φ = 0. Furthermore, this property
can also be deduced from the Hamiltonian equation of motion φ̇ = i[H, φ] = φ .
Note that to get the chiral solution ∂− φ = 0 as a solution of the equation of
the motion, we had to assume chiral boundary conditions. Here it looks at first
that the chirality property was derived with no assumptions, but in passing
from (6.83) to (6.84 we assumed that (π − φ )(x = ∞) = −(π − φ )(x = −∞),
so together with choosing zero surface terms we in fact assumed chiral boundary
conditions.
For the path integral quantization of the system we use the method developed
for Hamiltonian systems with constraints. The generating functional is given by,

Z[J] = dφdπδ(χ)e d x(πφ̇−Hc −J φ)
2

d 2 x(L−J φ)
= dφe , (6.87)

where a normalization by Z[O]−1 is implied. The Lagrangian density that

emerges in (6.87) is clearly in the original form of equation (6.78). The func-
tional integral (6.87) is not specified completely until we include ∂− φ = 0 on the
boundary; thus we are in the same situation as in the canonical quantization.
Using the commutation relations (6.84), it is straightforward to verify that the
Noether charges H, P, M , respectively generate the transformations δT φ, δS φ
and δM φ given above. It can easily be shown that they satisfy the Poincare
algebra,
[H, P ] = 0 [M, H] = iP [M, P ] = iH. (6.88)
Abelian bosonization of a chiral fermion

Recall (Section 3.8) that the action of one left-handed complex chiral fermion,

Sf + = d2 xψ † ∂− ψ, (6.89)
is classically invariant under both an affine chiral transformation δψ = i(x+ )ψ

and the conformal transformation δψ = i+ (x+ )∂+ ψ. The associated Noether
currents were shown to have the following quantum form,
J+ =: ψ † ψ T+ + = i : ψ † ∂+ ψ := π : J+ J+ :, (6.90)
obey the left U (1) affine Lie and Virasoro algebras, respectively with the well-
known central charges k = 1 and c = 1.
It is now straightforward to realize that the J(l) = φ and T(l) = (φ )2 have the
same k = 1 and c = 1 central charges. Using the operator algebra we can now
evaluate the commutators of the chiral current and of the energy-momentum
tensor. The results are as follows
i
[J(l) (x), J(l) (y)] = [φ (x), φ (y)] = δ (x − y),
2
[T(l) (x), T(l) (y)] = [: (φ (x))2 , (φ (y))2 ] = i(T(l) (x) + T(l) (y))δ (x − y)
i
− δ (x − y). (6.91)
24π
From the general discussion in Sections 2.4 and 3.3 it follows now that these
commutation relations correspond to central charges of c = k = 1. Hence our
bosonic theory furnishes the same irreducible representation of the affine Lie
and Virasoro algebras as one free left-handed chiral fermion. Since for k = 1
the affine Lie algebra has a unique irreducible unitary representation the two
theories on flat two-dimensional space-time are therefore equivalent. Below it
will be shown that the anomalies of the bosonized theory in coupling to gauge
and gravitational background are the same as for the fermionic theory.

156 Bosonization
Combining left and right chiral bosons

The canonical quantization procedure of the last section can be repeated for
a right chiral boson. Let us rename the operators for the ∂− φ = 0 case as
φL and πL and call the corresponding operators for the ∂+ φ = 0 case as ψR
and πR . The Lagrangian density for the right moving field has the form LR =
∂+ φ(∂+ φ − ∂− φ) Then the combined Lagrangian,
L = LL + LR (6.92)
and the corresponding Hamiltonian density,
H = (πL )2 + (πR )2 (6.93)
describe a single free massless scalar defined by,
φ = φL + φ R π = π L + πR (6.94)
with a Hamiltonian density,

1 2 1 2
H= π + φ . (6.95)
2 2
Using the commutation relations of the chiral bosons (6.84) we find as expected
that,
[φ(x), φ(y)] = [π(x), π(y)] = 0 [φ(x), π(y)] = iδ(x − y). (6.96)
Unsurprisingly the Noether charges associated with the space-time translation

and Lorentz transformation, H = HR + HL , P = PR + PL and M = ML + MR
obey the usual Poincare algebra. The (lr) U (1) affine Lie algebra currents of
the combined system are given by the left current of the left system and the
right current of the right system, respectively. The central charges are k = 1 for
the algebras in both sectors. Similarly, because of the Sugawara construction,
T+ + = T++ L and T−− = T−− R with c = 1 for the left and the right Virasoro
algebras.
Partition function
We would now like to compare the one loop partition function of a chiral fermion
and that of our chiral boson. We therefore pass to a two-dimensional space-time
domain with 1 ≥ x ≥ 0. The mode expansion for φ previously given by equation
(6.85) now takes the form,
1 +
a†n e2πin x + an e−2πin x .
+
φ(x, t) = φ0 + p(x + t) + √ (6.97)
n>0
2n
The one loop partition function which corresponds to this mode expansion is
Z(τ ) = Tr[e−τ 2 H +iτ 1 P ] = Tr[q π ]

H
(6.98)

where q = eiτ π and we use the fact that H = −P . Let us first calculate the
contribution to the trace of the oscillation modes,

n a †n a n − 112 )
q − 1 2 [1 − q 2n ]−1 = η −1 (τ ),
H 1
Tr[q π ] = Tr[q ( n ]= (6.99)
n=1
where the factor − 12

1
is the output of the normal ordering and η is the Dedekin
1 2
η function (see eqn. (2.48)). The Hamiltonian of the zero modes is H = 4π p .
If we now take the bosonic momenta to lie on a shifted lattice such that the
eigenvalues of p are 2π(m + α) and introduce the twist operator g = eiβ p , we
get the zero mode partition function to be equal to the Riemann theta function
Θ[(αβ)](τ |0) and so that altogether the full partition function is given by,
Θ[(αβ)](τ |0)
Z(τ ) = . (6.100)
η(τ )
This corresponds to the partition function of a Weyl fermion with the boundary
conditions ψ(x + 1, t) = −e2πiα ψ(x, t); ψ(x + Reτ, t + Imτ ) = −e−2πiβ ψ(x, t).
6.4.3 Coupling to abelian gauge fields

There are several ways to couple an abelian gauge field to a chiral boson corre-
sponding to the various regularization schemes in the fermionic theory. We start
by analyzing the bosonization in the vector conserving regularization scheme.
The vector current is coupled to an abelian gauge field via,
L(V ) = L0 + (J(V ) − A+ + J(V ) + A− ), (6.101)
where L0 is the uncoupled Lagrangian density and V stands for the vector con-
serving scheme. The vector current is still obviously conserved, but the divergence
of the axial current,
∂− J(ax) + + ∂+ J(ax) − = ∂− A+ − ∂+ A− = μν Fμν , (6.102)
is now equal to the anomaly deduced in the fermionic theory from the one loop
diagram.
Next we discuss the bosonization in the left-right scheme. For that purpose a
term bilinear in the gauge fields has to be added to the J(l) + A(l) − term. The
Lagrangian then takes the form:
√
2
L(LR) = L0 + J(l) + A(l) − + A(l) − A(l) 1 , (6.103)
4
where (LR) indicates the left-right scheme. The divergence of the left current
∂ L( L , R )
which is derived from ∂ A ±
now has the form,
1 1
∂− (J(l) − )(L,R) + ∂+ (J(l) + )(L,R) = (∂− A+ − ∂+ A− ) = μν Fμν . (6.104)
4 4

158 Bosonization
The Lagrangian (6.103) is therefore really the bosonized fermionic action regu-
larized in the left-right scheme. Obviously, a similar prescription for the (V) and
(LR) schemes can be applied to the right chiral boson. It is straightforward to
show that the vector as well as the left-right actions are invariant under “curved
space-time” Lorentz transformations as discussed above.
6.4.4 Chiral WZW and coupling to non-abelian gauge fields

The non-abelian bosonization of N Majorana or N Dirac fermions using WZW
theories of SO(N ) and U (N ), respectively was described in Section 6.3. For the
non-abelian bosonization of left chiral fermions we propose to generalize
the action (6.78) into an action that describes a map to the group manifold
of the form,
√
2
S+ [u] = d2 xT r(∂− u∂1 u−1 ) + SW Z . (6.105)
4π
In a similar manner to the abelian case this is a WZW action coupled to fictitious
chiral gravity in the gauge h+ + = −1. In fact we can consider a generalization
of this action to the so-called k-level chiral WZW namely S+k [U ] = kS+ [u]. The
equation of motion that follows from the variation of (6.105) with respect to the
variation of u can be expressed as,
∂− (u−1 ∂1 u) = 0 or ∂1 (u∂− u−1 ) = 0, (6.106)
where each form can be obtained from the other.

The global, chiral transformations u → Au and u → uB −1 where A, B ∈ G
leave the action (6.105) invariant. However, out of the invariance under the two
affine Lie algebra transformation of the original WZW action u → uB −1 (x+ )
and u → A(x− )u, only the first survives. As for the abelian case the invariance
under the second transformation is lost. The Noether currents associated with
the left-right transformations of (6.105) are,
√
i 2k −1
J(l) − = 0 J() + = u ∂1 u
2π
ik ik
J(r ) − = u∂− u−1 J(r ) + = − u∂− u−1 . (6.107)
2π 2π
The conservation of the left and right currents follow here simply from the equa-
tions of motion. However, unlike the ordinary WZW action only the left current
is holomorphically conserved a ∂− J(l) = 0; whereas ∂+ J(r ) = 0. Obviously this
is a manifestation of the invariance of the action only under the left affine Lie
algebra transformation δu = −iu(x+ ), √discussed above. The left current trans-
2k
forms as follows δJ(l) = [i(x+ ), J(l) ] + 2π , leading to the O(N )(U (N )) affine
Lie algebra with central charge equal to k.

6.5 Bosonization of systems of operators of high conformal dimension 159
The action S+k [u] is invariant under the left affine transformation δu = (x+ )u.
The corresponding energy momentum tensor has again a Sugawara form and
Virasoro central charge which are,
2π a a kdimG
T(l) = : J(l) J(l) : c = , (6.108)
(c2 + k) a c2 + k
a
where as usual J(l) = J(l) T a and T a are hermitian matrices representing the alge-
bra of the group, C2 is the second Casimir operator in the adjoint representation
and dimG, is the dimension of the group.
The coupling of non-abelian gauge fields to a chiral WZW action, is a straight-
forward generalization of the coupling of the abelian gauge fields. Again there are
several ways to couple gauge fields corresponding to the various regularization
schemes in the fermionic theory. The bosonized action (for k = 1) related to the
vector conserving regularization scheme is given by,

− +
SV [u, A− , A+ ] = S+ [u] + d2 xTr[J(v ) A− + J(v ) A+ ]
√
2
− d2 xTr[u−1 A1 uA− − A− A1 ], (6.109)
2π
where J(v ) and J(ax) are constructed from the left and right currents of (6.107)
in the usual way. Using the equation of motion one finds the conservation of the
vector current and the anomalous divergence of the axial current,
μ μ 1 μν
Dμ (J(v ) )V = 0 Dμ (J(ax) )V = π Fμν (6.110)
The coupling of a left non-abelian gauge field that corresponds to the fermionic
description in the left-right regularization scheme is given by,
√
2 1
S(LR) [u, A− , A1 ] = S+ [u] + d2 xTr[(u−1 ∂1 u + A1 )A− ]. (6.111)
2π 2π
The associated current divergence is,
1 μν
μ
Dμ (J(l) )LR = D− Jl− (LR) + D+ J−
+
l(LR) = Fμν . (6.112)
π
This expression for the anomalous divergence of the left current is identical
to the result of the loop calculation in the fermionic version regularized in the
left-right scheme.
6.5 Bosonization of systems of operators of

high conformal dimension
Can bosonization, the equivalence map between systems of dimension half
fermions and dimension zero bosons, be extended also to systems made out
of higher-dimensional operators? In particular can one map theories of odd and
even fields of higher-dimensional operators to theories built from dimension zero

160 Bosonization
scalar fields. In this section it will be shown that indeed such a map exists for
two families of theories, one with anti-commuting fields and the other with com-
muting ones with arbitrary integer and half-integer dimensions.
The bosonization of the b, c and β, γ systems was introduced in [94]
6.5.1 The bosonization of the “b,c” free CFT

We first briefly describe the system. Consider a system built from a pair of
anti-commuting fields b and c which is described by the action,

1 ¯
Sb,c = d2 zb∂c. (6.113)
2π
It is easy to see that this action is invariant under conformal transformation
z → a−1 z b → aλ b and c → a1−λ c, namely the b and c fields have conformal
dimensions,8
hb = λ hc = (1 − λ). (6.114)
The classical equations of motion,

¯
∂b(z, ¯
z̄) = 0 ∂c(z, z̄) = 0, (6.115)
imply that both fields are holomorphic, namely b(z), c(z). Similar to the deriva-
tion of the operator equations of motion using the path integral (1.55) for the
scalar field we find for the (b, c) system that,
¯
∂b(z)c(0) = 2πδ 2 (z, z̄). (6.116)
This is compatible with the OPE,

1 1
b(z)c(w) =: b(z)c(w) : + c(z)b(w) =: c(z)b(w) : + . (6.117)
z−w z−w
The OPEs b(z)b(w) and c(z)c(w) do not have singular parts when w it brought
to z. To compute the energy-momentum tensor we use the Noether procedure.
We vary the b and c fields as follows,
δb = ¯∂b + λ(∂¯
)b,
δc = ¯∂c + (1 − λ)(∂¯
)b. (6.118)
For holomorphic ¯(z) this is a symmetry transformation. Taking now ¯(z, z̄) we
read the energy-momentum tensor from the variation of the action as follows,

1 ¯T = 1 ¯[(∂b)c − λ∂(bc)].
δSb,c = − d2 z ∂¯ d2 ∂¯ (6.119)
2π 2π
8 We denote by λ the dimension of b as is common in the literature. Obviously it has nothing

to do with λ used in the previous section on chiral bosons.

Thus the energy-momentum tensor is,

T = (∂b)c − λ∂(bc). (6.120)
It is straightforward to verify that the OPE of T with b and with c indeed yields
the variation (6.118). From the OPE T (z)T (w) we read the Virasoro anomaly
associated with the (b, c) system,
c = −3(2λ − 1)2 + 1. (6.121)
The action (6.113) is also invariant under the fermion number transformation,
b → eiα (z ) b c → e−iα (z ) c. (6.122)
Using again the Noether procedure we find that the corresponding conserved
fermion number current is,
j =: bc : ¯ = 0.
∂j (6.123)
From the basic OPE of b(z) and c(w) one finds that the OPE of the energy-
momentum tensor and the fermion number current reads,
1 − 2λ j(w) ∂j(w)
T (z)j(w) = + + . (6.124)
(z − w)3 (z − w)2 (z − w)
The (b(z), c(z)) conformal field theory is fully holomorphic. Needless to say
that one can similarly write down an anti-holomorphic system b̄(z̄), c̄(z̄). In fact
in Section 2.12 we have already discussed a special case of the b, c family. For
λ = 12 we get the Weyl spinor ψ of dimension 12 such that the Virasoro anomaly of
the system is c = 1. Another “famous” case is that of the b and c ghosts associated
with the covariant fixing of two-dimensional diffeomorphism. In this case λ = 2
the dimensions of b and c are 2 and −1, respectively and the corresponding
Virasoro anomaly is −26.
Now we raise again the question, can one describe the b, c system in terms of
a scalar CFT? Since the ψ, ψ † is a special case of the b, c system and since we
have already developed the bosonization rules for Dirac and Weyl spinors (see
Section 6.1.1) we start with a similar ansatz for the bosonic version of b and c,
namely,
b(z) ↔: eiφ(z ) : c(z) ↔: e−iφ(z ) :, (6.125)
Comparing (6.18) and (6.117) it is evident that indeed this map reproduces the
algebra of the (b, c) system. It is also easy to realize that the fermion number
current has the following bosonic equivalent,
: b(z)c(z) :↔ i∂φ(z). (6.126)
What is left to determine is whether the energy-momentum tensors of the two
theories and correspondingly the dimensions of the fields match. Obviously the
free scalar action (6.1) which is the bosonic dual of the action of the Dirac

162 Bosonization
operator cannot describe the (b, c) systems which are a family of CFTs. We
also need to identify a family characterized by the parameter λ of scalar field
theories. A simple way to achieve this is to realize that the energy-momentum
of the general (b, c) system can be written in terms of the spin 1/2 fermion as
follows,
1
T b,c = T ψ − λ − ∂(bc). (6.127)
2
Thus following (6.126) the scalar energy momentum has to have the form,
1
Tλφ = T φ − λ − ∂ 2 φ. (6.128)
2
This is the energy-momentum of a scalar field with a background charge, or the
linear dilaton theory with q = −i(λ − 12 ). The central charge of these theories
was shown to be,
cφ = 1 + 12q 2 = 1 − 3(2λ − 1)2 = cb,c . (6.129)
Moreover using the fact that the dimension of an operator : eik φ(z ) : was shown
2
to be k2 + ikq it is easy to check that,
he i φ = λ = hb he −i φ = 1 − λ = hc , (6.130)
which verifies that the operators mapped by the bosonization indeed have the
same conformal dimensions.
The anti-commutative nature of the b, c system is obeyed also by their bosonic
duals, just as for the case of spin 1/2 fields, namely,
: eiφ(z ) :: eiφ(w ) := e−[φ(z ),φ(w )] : eiφ(w ) :: eiφ(z ) := − : eiφ(w ) :: eiφ(z ) :, (6.131)
at equal times, namely |z| = |w| since for that case [φ(z), φ(w)] = ±iπ.
6.5.2 The bosonization of the β, γ system

The b, c system is built from anti-commuting fields which as we have just seen are
describable in terms of a scalar field. In a complete analogy one would suspect
that a similar bosonization also holds for a system built from commuting fields
with the same structure of action. As we shall see shortly this is indeed the case.
The so-called β, γ system defined by the action (6.132),

1 ¯
Sβ ,γ = d2 zβ ∂γ. (6.132)
2π
The fact that now the building blocks are commuting introduces of course a sign
change with respect to the b, c system when one interchanges the fields, namely,
1 1
β(z)γ(w) = − γ(z)β(w) = . (6.133)
z−w z−w

The energy-momentum tensor is the same as (6.120) when replacing b and c

with β and γ, respectively. For the Virasoro anomaly one has just to reverse
the sign of that of the b, c system. A distinguished member of this family of
conformal field theories is the case of λ = 3/2 which describes the ghost system
associated with the gauge fixing of the superdiffeomorphism. In this case the
c = 11. As is well known this combined with c = −26 requires a contribution to
the Virasoro anomaly of 15 of the non-ghost fields which requires ten dimensions
for superstring theories.
The bosonization of this commuting system is slightly more involved than that
of the b, c system. It turns out that now one has to invoke two scalar fields φ and
χ with the OPEs,
φ(z)φ(w) = −ln(z − w) + . . . χ(z)χ(w) = ln(z − w) + . . . , (6.134)
where . . . stands for non-singular terms. The corresponding scalar theories have a
background charge of (λ − 1/2) for φ and i/2 for χ so that the energy-momentum
tensor of the full bosonic reads,
1 1 1 1
Tφ,χ = − ∂φ∂φ + ∂χ∂χ + (1 − 2λ )∂ 2 φ + ∂ 2 χ, (6.135)
2 2 2 2
which yields the desired Virasoro anomaly c = 1 + 3(2λ − 1)2 = −cb,c . The
bosonic operators that correspond to β and γ are,
β(z) ↔ e−φ+χ ∂χ γ ↔ eφ−χ , (6.136)
which have the conformal dimensions hβ = λ and hγ = 1 − λ .
6.5.3 The Wakimoto bosonization

One application of the β, γ system enables us to transform the WZW model
into a theory expressed in terms of free fields. Consider a combined system that
includes a (λ = 1, 0) β, γ system and an additional scalar field with a background
charge − √ i . This system is described by the following Lagrangian,
2(k + 2)
¯ + β̄∂γ̄ + ∂ϕ∂ϕ.
LWak = β ∂γ ¯ (6.137)
Alternatively, as was discussed in Section 6.4 one can add to this Lagrangian
density also a term of the form − √ i R2 φ. The corresponding energy-
2(k + 2)
momentum tensor T (z) is given by,
1 i
T (z) = −β∂γ − ∂ϕ∂ϕ − ∂ 2 ϕ, (6.138)
2 2(k + 2)
and the non-trivial OPEs of these field are given by,
1
γ(z)β(w) = + O(z − w) ϕ(z)ϕ(w) = −ln(z − w) + O(z − w). (6.139)
z−w

164 Bosonization
We define now the following holomorphic currents in terms of the β, γ and ϕ

fields:
J + = β(z),

J 0 = i 2(k + 2)∂ϕ(z) + 2 : γβ(z) :,

J − = −i 2(k + 2) : ∂ϕγ : (z) − k∂ϕ(z)− : βγγ : (z). (6.140)
Using the OPEs it is straightforward to determine the OPEs of the currents,
k J 0 (z)
J + (z)J − (w) = + ,
(z − w)2 (z − w)
2J +
J 0 (z)J + (w) = ,
(z − w)
2k
J 0 (z)J 0 (w) = ,
(z − w)2
−2J −
J 0 (z)J + (w) = , (6.141)
(z − w)
which are the OPEs of the SU (2) affine current algebra, with level k. Furthermore
the Sugawara energy-momentum tensor,

1 1 0 0
T (z) = J J + J + J − + J −J + , (6.142)
2(k + 2) 2
is identical to the energy-momentum tensor given in (6.138), and the associated
Virasoro anomaly is,
1 3k
c = 2 + 1 − 24 = . (6.143)
4(k + 2) k+2
Since β has dimension one and γ dimension zero, their mode expansion takes
the form,

β= βn z −n −1 , γ = γn z −n . (6.144)
n n
Substituting this into the expressions of the currents one finds,

Jn+ = βn ,

Jn0 = i 2(k + 2)nϕn + 2 : βm γm −n :,
m

Jn− = −i 2(k + 2) : nϕm γn −m : −knϕn − : βl γm γn −m −l : .
m lm
(6.145)

7
The large N limit of two-dimensional models
7.1 Introduction
The number of approximation techniques in quantum field theory is very limited.
1
Perturbation expansion in small interaction coupling, like αem = 137 of QED, is
obviously the most important one. Other methods include dimensional expan-
sion, high temperature expansion and large radius expansion. Quite surprisingly,
one of the most useful approximation techniques is expansion in the number of
degrees of freedom. A priori one would tend to think that the larger the number
of degrees of freedom, the more complex the system. However, it turns out that
theories with infinitely many degrees of freedom are much easier to solve than
those with a finite number of degrees of freedom. Once the system with N → ∞
is known, a systematic expansion in N1 provides an approximation procedure for
computing quantities that describe systems of finite N .
Large N methods have been applied in a very wide range of physical sys-
tems. Starting from non-critical phenomena in spin systems like the Heisenberg
ferromagnet (discussed in Section 5.14), then SU (N ) QCD theories in various
dimensions, and later matrix models associated with either string models or
two-dimensional models.
The large N approximation in field theory or correspondingly the planar
expansion of Feynman diagrams was introduced by ’t Hooft in his seminal paper
[122].
In this book we will focus on four arenas where large N approximations are
being used:
(i) Two-dimensional quantum field theory models, which include the Gross–
Neveu model and the CP N models, will be addressed in this chapter.
(ii) Quantum chromodynamics with large N SU (N ) gauge theory in two dimen-
sions. In Chapter 10 of Part 2 the solution of two-dimensional QCD, follow-
ing ’t Hooft [124], will be described,1 together with a certain generalization
of it.
(iii) The approach to four-dimensional QCD based on the N1 expansion.
(iv) Baryons in large N QCD.
1 Known as the ’t Hooft model.

166 The large N limit of two-dimensional models
The last two topics will be described in the third part of the book in Chapters
19 and 20.
In Nature the number of colors is three, and thus one may wonder whether
it makes sense to in expand a not-so-small parameter 1/3. Even though there
is no general proof that this expansion is indeed reliable, a vast literature on
the subject brings out a large amount of evidence that indeed this is the case.
It turns out that for certain quantities the N1 term vanishes and the correction
starts as N12 , and hence puts the approximation on a more solid base.
As an example of the accuracy of√the large N limit consider the Stirling formula
for N , where the leading term is 2πN (N/e)N for large N. But the correction
1
is actually 12N as compared to 1, making it only an 8 percent correction even
for N = 1.
In the next sections we describe the O(N ) model, the Gross–Neveu model and
the CP N models.
There are several review articles on large N expansions, for instance [46], [160],
[165]. In this chapter we make use of [160].
7.2 The Gross–Neveu model

The Gross–Neveu (GN) model, proposed in [117], describes a set of N Dirac
fermions interacting via a four-fermi interaction. It turns out that one can solve
the model, and prove that it is asymptotically free and admits a dynamical
symmetry breaking, using 1/N expansion. The Lagrangian of the system can be
written in the form,
λ20 a a 2
LGN = iψ̄ a ∂ψ a + (ψ̄ ψ ) , (7.1)
2
where a = 1...N and λ0 is a bare coupling of dimension zero. The correspond-
ing action is invariant under a continuous SU (N ) global transformation and a
discrete chiral transformation,
ψ a → gba ψ b g∈SU (N ), ψ a → γ5 ψ a ψ̄ a → γ5 ψ̄ a . (7.2)
The discrete chiral symmetry forbids a mass term. In fact this is the most general
action invariant under these symmetry transformations, with terms of dimension
two or less, and hence it is a renormalizable action.
Let us now check whether in this formulation of the Lagrangian, where λ0 is
fixed, one can make sense of a large N limit. Consider the scattering process of
two fermions with flavor index a that turn into a pair of fermions with a different
flavor index b. The leading Feynman diagrams that contribute to this process
are given in Fig. 7.1. The first diagram which is the basic interaction vertex is
of order λ0 , the second one is of order λ20 , the third is of order λ20 N due to the
N different flavors of the fermions that can run in the loop, and the last two
diagrams are of order λ30 N 2 . It is thus clear that the perturbation expansion

a b a b
a b
a b a b
(a) (b)
a b a b a d b
c
c d c
c d c
c
a b a b a b
(c) (d) (e)
Fig. 7.1. Leading order diagrams for the a + ā → b + b̄ scattering.
expressed in terms of the fixed coupling λ0 does not have a sensible large N
expansion. However, one can easily cure this problem by defining the coupling
λ ≡ λ0 N , which is taken to be fixed when N → ∞. The Lagrangian now reads,
λ
LG N = iψ̄ a ∂ψ a + (ψ̄ a ψ a )2 . (7.3)
2N
It is now straightforward to see that the N1 in front of the interaction term

enables a well-defined large N limit. Consider again the diagrams in Fig. 7.1.
The leading contribution is of order Nλ and the loop corrections are now of order
λ2 λ2 λ3
N 2 , N and N , respectively. Hence it is obvious that the form of any scattering
amplitude in perturbation theory is N1 A(λ, 1/N ), which becomes N1 A(λ, 0) at
the large N limit.
To further analyze the system it is convenient to introduce the auxiliary field
σ, in terms of which the Lagrangian (7.3) can be written as,
N 2
LGN = iψ̄ a ∂ψ a + σ ψ̄ a ψ a − σ . (7.4)
2λ
Integrating over σ or alternatively solving the classical equation of motion for σ

and substituting it into the action yields the Lagrangian (7.1). The introduction
of the auxiliary field, which will soon acquire a physical interpretation, is a
standard step in the large N procedure which enables a simplified counting of
powers of N1 .
The Feynman diagrams of the theory for this alternative formulation are then
given in Fig. 7.2, with the full line representing the fermion propagator, and the
dotted line that of σ.
Note that the only non-trivial interaction is the σ ψ̄ a ψ a term and that each σ
propagator contributes iλN . The diagrams Fig. 7.1 that contribute to the scatter-
ing process are converted to those in Fig. 7.3.

i
p/ 1
N
Fig. 7.2. Feynman rules of the Gross–Neveu model.
Fig. 7.3. Leading diagrams that contribute to the two-to-two scattering.
We would now like to integrate over the fermions and derive an effective
Lagrangian, Leffective (σ). Since the Lagrangian is quadratic in the fermion fields,
Leffective (σ) is given by the sum of terms (Fig. 7.4).
Note that all diagrams are with an even number of σ only, since σ is odd under
the transformation (7.2). The first term is the tree level contribution − 2λN 2
σ and
the rest are the one loop contributions. Both are of order N , the latter due to the
N fermions that can run in the loop. The N dependence of Leffective therefore
has the form,
Leffective (σ, λ, N ) = N L̂effective (σ, λ). (7.5)
This makes the counting of powers of N1 very easy. Consider a graph with E
external σ lines, I internal σ lines, V vertices and L independent loops. The
parameters (E, I, V, L) are not independent. For each internal line there is a
momentum integration and hence a loop. However each vertex introduces a delta
function in momenta that cancels one momentum apart from an overall delta
function associated with the momentum conservation.
Thus one has,
L = I − V + 1. (7.6)
1
Recall that each σ external or internal line carries a N factor, while each vertex
contributes a factor of N .
Thus the net power N of each graph is,
N −I +V −E = N −E −L +1 . (7.7)

Fig. 7.4. Diagrams of Le ff e c tive (σ).
It is obvious from this expression that adding loops and external σ lines sup-
presses the corresponding contribution due to additional powers of N1 . Since the
minimal number of σ external lines is two the leading behavior is of order N1 .
For the purpose of investigating the possibility of spontaneous breaking of
the discrete symmetry of (7.2), it is enough to compute the effective potential
V (σ) rather than the effective action, namely, the limit where all the external
lines carry zero momentum. The effective potential is given by the sum of the
diagrams in Fig. 7.4:
∞ 2n
N σ2 1 d2 p − pσ
−iV = −i −N Tr , (7.8)
2λ n=1
2n (2π)2 p2 + i
1
where 2n is the symmetry factor of the graph, N comes from summing over all
possible flavors, –1 from the fermion loop and the expression in the bracket is the
p/
product of the propagator and (i p 2 +i ) and the vertex (iσ). Using the identity,
∞
x2n 1
= − log(1 − x2 ), (7.9)
n=1
2n 2
and analytically continuing to the Euclidean space gives,

2 2
σ d pE σ2
V =N − log 1 + . (7.10)
2λ (2π)2 p2E
The momentum integral is logarithmically divergent so by introducing a cutoff
on the Euclidean momentum p2E ≤ Λ2 we find,
2
σ 1 2 σ2
V =N − σ [log + 1 . (7.11)
2λ 4π Λ2
The effective potential can be rewritten in terms of the coupling λr , renormalized
at a scale μ, defined as,
1 1 d2 V 1 1 μ2 1
≡ |σ =μ = + log + , (7.12)
λr N dσ 2 λ 2π Λ2 π

and substitute it into the V (σ) to find,

2
σ 1 2 σ2
V =N + σ log −3 (7.13)
2λr 4π μ2
The fact that the cutoff disappears obviously implies that the theory is indeed
renormalizable. The β function and the anomalous dimension can be determined
by substituting the effective potential into the renormalization group equation
(see Section 17.6),
[μ∂μ + β(λr )∂λ r − γσ (λr )∂σ ] V (σ) = 0. (7.14)
We thus find the exact (to all orders of λr ) expression of β(λr ) and γσ (λr ), which
take the form,
λr 3
β(λr ) = − γσ (λr ) = 0. (7.15)
2π
Thus we have deduced that the Gross–Neveu model is asymptotically free,

namely that the effective coupling goes to zero at high momenta. It turns out
that the minus sign ensures this.
Let us now examine whether the chiral symmetry of this theory is sponta-
neously broken. For that we determine the extremum points of the potential.
The vanishing points of the derivative,

dV σ σ σ2
=N + log 2 − 2 , (7.16)
dσ λr 2π μ
are at
π
σ=0 and σ = ±σ0 = ±μe1− λ r , (7.17)
where
σ02
V (0) = 0 and V (σ0 ) = −N < 0. (7.18)
4π
Now since the potential vanishes at σ = 0 and it is negative at σ = ±σ0 , its

global minima are at ±σ0 . Therefore the discrete chiral symmetry is broken and
the massless fermions acquire mass which to the leading order is σ0 .
A further interesting property of the model is the dimensional transmuta-
tion. The bare theory depends on one continuous dimensionless parameter and
the effective theory depends on one continuous parameter with dimensions, σ0 .
Whereas one may anticipate that observables will depend on the dimensionless
parameter in a complicated way, one finds a simple dependence on the parameter
with dimensions which follows a dimensional analysis.

7.3 The CPN – 1 model

Another model that can be solved using the large N expansion is the CP N −1
model. The model is an example of a non-linear sigma model where the fields
live in a complex projective N − 1 space,
SU (N )
CP N −1 = . (7.19)
SU (N − 1) × U (1)
The Lagrangian of the model can be written as,
λ
LC P N = ∂μ Z † ∂ μ Z − Jμ J μ , (7.20)
N
where Z † ≡ (z1 , . . . , zN ), namely, an N -dimensional “unit” vector of complex
fields that obey the constraint,
N
Z †Z = , (7.21)
λ
and Jμ is given by,
i
Jμ = − Z † ∂μ Z − (∂μ Z † )Z . (7.22)
2
The Lagrangian describes a theory of massless particles with short range inter-
action which originates from both the explicit JJ interaction as well as from the
constraint. The number of degrees of freedom of the CP N −1 model is 2N − 2.
This is the dimension of the CP N −1 coset space,

SU (N )
dim CP N −1 = dim
SU (N − 1) × U (1)
= (N 2 − 1) − ((N − 1)2 − 1 + 1) = 2N − 2. (7.23)

Differently, we count N complex numbers Zi , namely 2N real degrees of freedom,
minus one degree of freedom due to the constraint (7.21), minus one degree of
freedom due to the U (1) local symmetry,
Z → eiα Z. (7.24)
It is easy to verify that (7.20) is indeed invariant under this transformation upon
the use of the constraint. In fact one can also write the Lagrangian in the form,
LC P N = ∂μ Φ† ∂ μ Φ, (7.25)
where Φ is a traceless hermitian matrix built from Z and Z † according to,
5
λ † 1
Φ= ZZ − , (7.26)
N λ
It is clear that the local transformation of the above does not change Φ and
hence the Lagrangian is invariant under this transformation.
The first step in the large N program is to eliminate the quartic interaction
term, by introducing an auxiliary field in a similar manner to what was done

in the Gross–Neveu model. However, since the interaction has the form of a
vector times a vector, the auxiliary field should also be a vector. This shifts the
Lagrangian according to,
2
λ N
LC P N → LC P N + Jμ + Aμ
N λ
N
= ∂μ Z † ∂ μ Z + 2Jμ Aμ + Aμ Aμ
λ

= (∂μ − iAμ )Z † [(∂ μ + iAμ )Z] , (7.27)
where in the third line we have used the constraint. It is clear from its last form
that the Lagrangian is invariant under the U (1) local transformation,
Z → eiα Z Aμ → Aμ − ∂μ α. (7.28)
We now incorporate the fact that the Z are constraint variables by introducing
another Lagrange multiplier into the Lagrangian,

N
LC P N = (∂μ − iAμ )Z † [(∂ μ + iAμ )Z] + σ Z † Z − . (7.29)
λ
Obviously the path integral over σ, or equivalently using its equation of motion,
implies that Z † Z = Nλ .
The action is now quadratic in Z, so we integrate out the Z fields similarly to
what was done in the Gross–Neveu model. The Feynman diagrams that consti-
tute the leading contributions to the effective action, which is now a functional
of σ and Aμ , are drawn in Figure 7.5.
These diagrams, which include pure σ, pure Aμ and mixed diagrams, are all
proportional to N . The computation of V (σ) is similar to that in the GN model,
leading to,
σ σ σ
V (σ) = −N + log 2 − 1 , (7.30)
λ 4π Λ
where Λ is the cutoff. Again similar to the GN model the cutoff can be eliminated
by performing a renormalization at a scale μ,
1 1 dV 1 1 μ2
=− |μ 2 + log , (7.31)
λr N dσ λ 4π Λ2
so that the potential takes the form,

σ σ σ
V (σ) = −N + log −1 . (7.32)
λr 4π μ2
It is also evident that the model has a negative β function, or differently stated,
for fixed λr and μ, when Λ → ∞, λ vanishes, namely, the model is asymptotically
free.

Fig. 7.5. Leading order contributions to the effective action.
Again the model admits a dimensional transmutation. The original dimen-

sionless coupling is traded with a parameter σ0 with dimensions, at which the
potential has a minimum,

dV 1 1 σ 4π
= −N + log 2 = 0, σ0 = μ2 e− λ r . (7.33)
dσ λr 4π μ
The following remarks about the model are relevant:
(i) The model admits a dynamical generation of abelian gauge fields. From the
diagrams on the third line of Fig. 7.5 we see that the contribution to Seffective
quadratic in Aμ is,

iN 1 (1 − 2x)2
− gμν p − pμ pν
2
dx 2 .
4π 0 σ0 − p2 x(1 − x) − i
Now for long range interaction, namely, for small momenta we ignore the p2
term to obtain,
iN
− [gμν p2 − pμ pν ],
12πσ0
which corresponds to the following term in the effective action,

N
Seffective = − d2 xFμν F μν , (7.34)
48πσ0
namely an action of an abelian gauge field.

(ii) The Z fields can be interpreted as bosonic “quarks” in the fundamental rep-
resentation of the group, though transforming in a non-linear way. These Z
quarks are confined due to the dynamically generated abelian gauge inter-
action. As will be shown in Chapter 8, in two dimensions the abelian force
between a quark anti-quark pair is linear in separation distance.

PART II
Two-dimensional non-perturbative
gauge dynamics
In the first part of the book we have developed several non-perturbative tools
for analyzing two-dimensional field theories. These include methods associated
with conformal invariance, with affine Lie algebras and in particular the WZW
model, techniques of integrable massive theories including solitons, S-matrix and
the Yang–Baxter equation, sets of infinitely many conserved charges, the thermal
Bethe ansatz, methods of bosonization and the large N limit approximation.
In this second part of the book the main idea is to implement those meth-
ods in the context of two-dimensional gauge dynamics aiming at extraction of
the mesonic and baryonic spectra of two-dimensional QCD, decoding the con-
fining behavior, versus a screening one, and analyzing models with other quark
representations and models of generalized QCD.
The most important tool that will be used for this purpose will be bosoniza-
tion. It will enable us to easily solve the Schwinger model, derive the baryonic
spectrum of QCD2 using a strong coupling limit, determine the screening nature
of massless QCD models and compute the string tension for the massive ones.
Using the large N approximation we will extract the mesonic spectrum of QCD.
In two respects this part is an intermediate stage on the way to four-
dimensional gauge dynamics. Firstly, as was just mentioned, we will gain expe-
rience from applying non-perturbative methods on the simpler two-dimensional
models, which will serve us when using them in the context of “real” physi-
cal systems. Secondly, two-dimensional gauge systems will serve as a toy model
laboratory of four-dimensional ones. As will be shown in the third part of the
book, certain aspects of two-dimensional physics will survive the transition to
four dimensions. Obviously the challenge will be to identify those phenomena
and devise some additional tools to handle the other cases.

8
Gauge theories in two dimensions – basics
As an introduction to the cast of characters of two-dimensional gauge theories,

we briefly summarize here the basics of pure Maxwell theory, QED, pure YM
theory and QCD. This includes the corresponding actions, symmetries, equations
of motion and their solutions.
The basics of gauge theories in two dimensions is “standard material” which
appears in many books and review articles, for instance [66], [178], [1] and [2].
For treatment in non-covariant gauges see [28].
8.1 Pure Maxwell theory

The simplest theory of gauge fields in two dimensions is obviously the abelian
Maxwell theory defined by the classical action,

1
S = d x − Fμν F
2 μν
, (8.1)
4
where the field strength
Fμν = ∂μ Aν − ∂ν Aμ (8.2)
has, in two dimensions, only one non-trivial component E1 ≡ F10 = −F01 =

∂1 A0 − ∂0 A1 . The action is invariant under the full global two-dimensional con-
formal symmetry SO(2, 2), discussed in Section 2.1, which includes in particular
the ISO(1, 1), where I stands for inhomogeneous, namely adding the momenta,
thus going over to the Poincare group from the Lorentz group. The action is by
construction also invariant under the gauge transformation,
Aμ (x, t) → Aμ (x, t) + ∂μ Λ(x, t). (8.3)
The canonical dimension of Aμ is clearly zero. The corresponding equation of

motion reads,
∂ μ Fμν = 0 ∂ 0 E 1 = ∂1 E 1 = 0 → E1 = constant. (8.4)
Thus we conclude that the two-dimensional Maxwell theory is an empty theory

on an R1,1 manifold. On such a space-time requiring finite energy implies that
E1 = 0. This is of course not surprising. In d-dimensional space-time the number
of degrees of freedom of an abelian gauge field is d − 2 and hence there are no
degrees of freedom in two dimensions.

178 Gauge theories in two dimensions – basics
8.2 QED2 – Schwinger’s model

Next we couple the two-dimensional abelian gauge fields to a Dirac fermion. The
Lagrangian density of this model is given by,1
1
L = − Fμν F μν + Ψ̄(i∂ − e A − m)Ψ
4
1 ¯
= (∂A − ∂ Ā)2 + ψ † ∂ψ
¯ + ψ̃ † ∂ ψ̃ + eψ † ψ Ā + eψ̃ † ψ̃A − m(ψ † ψ̃ + ψ̃ † ψ),
2
(8.5)
where in the second line the action is expressed in terms of the light-cone deriva-
tives and components of the gauge fields, and the Dirac fermion is decomposed
into its left and right chiral fermions, as discussed in Section 3.8. It is evident
that with the gauge field having a vanishing dimension, the gauge coupling e has
a dimension of mass. Thus the action is not invariant any more under the two-
dimensional global conformal symmetry, but rather only under the ISO(1, 1)
Poincare group. For the massless case, the action is classically invariant under
the global transformations,
Ψ → eiα Ψ Ψ → ei α̃ γ 5 Ψ
ψ → ei(α + α̃ ) ψ ψ̃ → ei(α −α̃ ) ψ̃. (8.6)
In fact the left and right chiral transformations, for the massless case, can be
lifted also into holomorphic and anti-holomorphic transformations, as was dis-
cussed in Section 3.7.1. The corresponding vector and axial currents
J μ = Ψ̄γ μ Ψ J5μ = Ψ̄γ μ γ5 Ψ

J = ψ† ψ J¯ = ψ̃ † ψ̃. (8.7)
Again by construction the action is also invariant under the gauge transfor-
mation,
1
Ψ → e−iΛ(x,t) Ψ Aμ (x, t) → Aμ (x, t) + ∂μ Λ(x, t). (8.8)
e
Quantum mechanically the axial current is not conserved even for the massless
case due to an anomaly,
e
∂μ J5μ = μν F μν . (8.9)
2π
We will derive this result using the bosonized version, see Section 9.1. Unlike
Maxwell’s theory, this theory has non-trivial degrees of freedom. However, once
again the gauge field is not dynamical. This phenomenon can be easily demon-
strated in the axial gauge A1 = 0, where the other component A0 can be solved
1 The Schwinger model was introduced in [190] and further analyzed in [68] and [64].

8.3 Yang–Mills theory 179
as a function of the electric current. The resulting electric field is,

eθ
E1 = −F01 = −e∂1−1 J0 − , J0 =: Ψ† Ψ :, (8.10)
2π
θ is a new parameter in the theory, the vacuum angle.2 In Part 3 of the book
we will describe its four-dimensional analog which is the vacuum angle due to
QCD4 instanton tunneling.
The massless Schwinger model can easily be solved using the anomaly equa-
tion combined with the equation of motion of the system. This will be done in
Section 9.1 using the bosonized version where we also address the massive case.
In Chapter 15 we determine the spectrum of the massless case using a BRST
quantization approach. In Chapter 14 we analyze the nature of the system and
determine when it confines and when it admits a screening behavior.
8.3 Yang–Mills theory

It is straightforward to generalize the action of the Maxwell theory (8.1) to the
non-abelian case.3 The gauge fields are now in the adjoint representation of a
non-abelian gauge group G. We will mainly be interested in the groups SO(Nc ),
U (Nc ) and SU (Nc ). Thus Aμ is an Nc × Nt either orthogonal, or hermitian or
traceless hermitian matrix of the form Aμ = tB AB B
μ where t are the generators
of the group, B = 1, ..., dimG and dimG is the dimension of the corresponding
algebra [ 12 Nc (Nc − 1), Nc2 and Nc2 − 1, respectively]. The field strength is now,
Fμν = ∂μ Aν − ∂ν Aμ + i[Aμ , Aν ]
¯ − ∂ Ā + i[A, Ā],
Fz̄ z = ∂A (8.11)
where again we write it out in light-cone coordinates. The action of two-
dimensional Yang–Mills theory reads,

1 1
SY M 2 = d2 x − 2 Tr(Fμν F μν ) = d2 x − 2 F a μν F a μν , (8.12)
2ec 4ec
and the corresponding equations of motions are,
Dμ F μν = ∂μ F μν + i[Aμ , F μν ] = 0. (8.13)
Note that we have rescaled the fields A by a factor of the gauge coupling, as
compared with the abelian case. Note, however, that this does not affect the
dynamical dimensions, namely the space-time behavior of Green’s functions.
In this formulation Aμ has dimension one and so is the dimension of the color
gauge coupling ec . Again since the coupling constant has a dimension of mass
the classical theory is not invariant under the full global conformal symmetry,
2 The θ angle was introduced by Lowenstein and Swieca [152] and also by Coleman [64].
3 The Yang–Mills non-abelian gauge theory was introduced in the seminal paper [229].

180 Gauge theories in two dimensions – basics
but only with respect to the ISO(1, 1) Poincare transformations. The action is
invariant under a non-abelian gauge transformation, which in infinitesimal form
is,
Aμ → Aμ + Dμ Λ = Aμ + ∂μ Λ + i[Aμ , Λ], (8.14)
where Λ = tA ΛA . A priori this is not a free theory, but rather an interacting

one. However, in a similar manner to Maxwell’s theory, this model too on an
R1,1 manifold has no dynamical degrees of freedom just as the abelian model.
This can easily be seen by fixing a gauge, for instance A0 = 0. In this gauge the
equations of motion read,
∂0 F 01 = 0 ∂1 F 10 + i[A1 , F 01 ] = 0. (8.15)
From the first we get that ∂02 A1 = 0, and thus A1 = f1 (x1 ) + x0 f2 (x1 ). Using
the residual gauge invariance, of gauge transformations that depend only on x1 ,
we can go to f1 = 0, and then the second equation implies that f2 is a constant
C, which yields F01 = C, and then again the requirement of finite energy results
in C = 0. This will also be shown in a complicated way using a BRST approach
in Chapter 15.
When the underlying manifold has a non-trivial topology like that of a torus
then the theory is not totaly empty but instead has topological degrees of free-
dom. This will be described in Section 16.
Finally, the non-abelian case is different from the abelian in higher dimensions,
as the former is not free there. While the abelian case represents free photons,
the non-abelian case represents interacting gluons, which turn to interacting glue
balls in the physical space.
8.4 Quantum chromodynamics

The theory of non-abelian gauge fields coupled to Dirac quarks in the fundamen-
tal representation of the gauge group, QCD2 , is described by the action,
< 1 =
SQ C D 2 = d2 x − 2 Tr(Fμν F μν ) − Ψ̄ai [(i∂− A + m)Ψi ]a . (8.16)
2ec
The action is invariant under two-dimensional Poincare transformation and
the non-abelian generalization of the gauge transformation of (8.8), which in
infinitesimal form is,
δΨa = −i[Λ(x, t)]ba Ψb δAμ (x, t) = ∂μ Λ(x, t) + i[Aμ Λ], (8.17)
with the non-abelian Λ = ΛA TA . Ψ is in the fundamental representation of

the gauge group which we take to be SU (Nc ) where a = 1, . . . , Nc denote
the color indices. As was discussed in Section 6.3.4 flavor degrees of freedom
have been included by assigning a flavor index to the Dirac fermion Ψi , i =
1, . . . , Nf . For this case the theory is obviously invariant classically under a global

8.4 Quantum chromodynamics 181
UL (Nf ) × UR (Nf ) symmetry. Here there is no anomaly, as in 2d the anomaly

occurs via the abelian gauge field only.
In a similar manner to the transition from the empty Maxwell theory to
the dynamically viable Schwinger model, so is the transition from the two-
dimensional pure Yang–Mills theory to QCD2 . The difference, however, is that
in the non-abelian case, even for the massless case there is no simple way to
solve the theory. Instead we will need to implement various different techniques
developed in the first part of the book. In the next section we will describe both
QED and QCD in two dimensions using the bosonization language. This will
enable us to solve for the baryonic spectrum in the strong coupling limit. In
Chapter 14 the string tension of several two-dimensional dynamical systems will
be computed. An analysis of the spectrum of these theories will be derived using
the BRST quantization approach in Chapter 15. In Chapter 10 we present the
seminal ’t Hooft solution of two-dimensional QCD in the large N limit. A current
algebra generalization of the latter approach will enable us to solve the mesonic
spectra of certain models. Finally in Chapter 12 we will implement a discrete
light-cone quantization approach to solve QCD in two dimensions with quarks
in the fundamental as well as the adjoint representation.

9
Bosonized gauge theories
Bosonization, the equivalence map between two-dimensional fermionic and

bosonic operators, was developed in Chapter 6. In fact several such maps have
been described. The simplest one has been the abelian bosonization that maps
the free theory of a Dirac fermion into that of a single real scalar field. The map
includes in particular an explicit bosonic expression for the left and right chiral
fermions (6.19), the vector and axial abelian currents (6.3) and for a mass term
(6.22). Using these transformations it is straightforward to write the bosonized
Lagrangian or Hamiltonian that corresponds to two-dimensional QED and QCD.
By its nature the abelian bosonization is more adequate to the abelian theory
of QED. The bosonized version of QED will be discussed in the next section.
We then apply this bosonization to QCD2 . Though it is possible to write QCD2
in an abelian bosonization formulation, it will turn out not to be very useful.
Instead, we will use the non-abelian bosonization discussed in Section 6.3. For
that purpose we will need to gauge the WZW action. Once this is done the
bosonized version of massless flavored QCD2 follows easily. The massive case
requires more care, as was explained in Section 6.3.3. Using the results of that
section the full bosonized theory that corresponds to massive flavored QCD2 will
be written down.
References on bosonization were given in Chapter 6.
9.1 QED2 – The massive Schwinger model

Recall that the fermionic Lagrangian of this model is given by,
1
L = − Fμν F μν + Ψ̄(i∂ − e A − m)Ψ. (9.1)
4
The Hamiltonian density of the system in the A1 = 0 gauge takes the form,
1
H = Ψ̄(iγ1 ∂1 + m)Ψ + (F01 )2 .
2
In bosonic variables, using (8, 10), the Hamiltonian becomes1

1 2 1 cm2 √ e2 1 θ
2
H =: π + (∂1 φ)2 − cos(2 πφ) + √ −φ :m ,
2 2 π 2π 2 π
1 The treatment of the bosonized Schwinger model was done in [68] and [64].

184 Bosonized gauge theories
where c is the constant of bosonization and the normal ordering is with respect
to the mass m, as was explained in Section 6.1.1.
After a shift in the definition of φ,
1 θ
√ , φ→φ+
2 π
√
and normal ordering with respect to μ = e/ π, one finds,
1 2 1 1 cmμ √
H =: π + (∂1 φ)2 + μ2 φ2 − cos(θ + 2 πφ) :. (9.2)
2 2 2 π
From this expression the periodicity in θ is manifest. The angle θ is the conjugate
to the winding number, appearing in two dimensions for the abelian case, since
Π1 [U (1)] = Z (looking at a circle of large radius in the two-dimensional plane).
Physics is invariant under θ → θ + 2π. From (8.10) it is clear that 2π
eθ
corresponds
to a background electric field. The periodicity is due to the ability to produce
electron-positron pairs in the vacuum when | 2π eθ
| > 12 e, and these pairs create
their own electric field which reduces the original one.
When we set m = 0 we discover that the massless Schwinger model is in fact
a theory of one free bosonic field with a mass equal to μ.
In the strong coupling limit, the bosonized form of the Hamiltonian is very
useful. The theory contains a meson of a mass that is approximately μ, and the
number of bound states depends on the value of θ. It can be shown that there
are no bound states for |θ| > π/2. For 0 < |θ| ≤ π/2 there is a stable two-body
bound state, while for θ = 0 there is also a three-body bound state.
Note that even though the Hamiltonian density (9.2) resembles that of a sine-
Gordon model, it does not admit soliton solutions due to the mass term 12 μ2 φ2 .
We will come back later to analyze this bosonized Hamiltonian, in the con-
text of the question whether the system admits screening or confinement in
Chapter 14.
Finally, let us show how the anomaly arises in the bosonized version. The
equation of motion for the electromagnetic field is,
∂ μ Fμν = eJν . (9.3)
The vector fermion current, in bosonic version, is (6.9),

1
Jν = √ ν α ∂ α φ. (9.4)
π
Taking the time component, we get,
e
∂1 F10 − √ φ = 0. (9.5)
π
From here, with the vanishing conditions at space infinity,
e
F10 = √ φ. (9.6)
π

9.2 Abelian bosonization of flavored QCD2 185
Now, the axial current, in bosonic version, is (6.12),

1
J5μ = √ ∂ μ φ. (9.7)
π
For the case of a massless fermion, the scalar field is a free field of mass √e , and
π
so,
1 e2 e
∂μ J5μ = − √ φ = F01 , (9.8)
ππ π
which is the anomaly equation.
Note that in the bosonic version, the anomaly is a result of the equations of
motion, while in the fermionic one it is the result of one loop.
9.2 Abelian bosonization of flavored QCD2

Let us now apply the prescription of flavored Dirac fermions for the analysis
of QCD2 . It is convenient to start with the Hamiltonian of the theory in its
fermionic formulation which we derive from (8.16),

NC
2
NC
NF

NC
NF
H = (ec )2 Eba + Ψ̄ai γ1 iδab ∂1 − Aba Ψbi + m Ψ̄ai Ψai ,
a,b= 1 a,b= 1 i= 1 a=1 i=1
(9.9)
in the gauge,
A0 = 0; Aab = 0 f or a = b; Eba = 0 f or a = b. (9.10)
The Gauss law of the system is given by,
1 δa
NF NF NC
∂1 Eba = i[A, E]ab + Ψ†ai Ψbi − b Ψ†di Ψdi (9.11)
2 i= 1 2NC i=1
d=1
Bosonizing now the various parts of the Hamiltonian one then gets,2
H = HΨ0 + HE − H I

1 2 cmμ √
0
HΨ = Σai [π + (∂1 φai ) ] +2
: (1 − cos(2 πφai )) :
2 ai π
2
e2c
HE = (φai − φbi )
8πNc i
ab
x
2c2 μ2 √
I
H = 3 Σa= b Σij Kij,ab Nμ cos π (πai − πaj + πbj − πbi )(ξ)dξ
π2 −∞
−1
√
sin( π(φai + φaj − φbj − φbi )(ξ)) (φak − φbk ) , (9.12)
ab
2 Abelian bosonization of two-dimensional QCD was discussed in [24] and [201] and was further
elaborated in [62].

HΨ0 is the free “fermionic” part, after bosonization, thus in terms of bosonic
variables; HE is the first term of the Hamiltonian (9.9) rewritten in terms of the
boson variables corresponding to the fermions, by eliminating the electric fields
through the Gauss law. Thus although originally coming from the kinetic part of
the gauge potentials, it actually involves the interactions. This is a result of the
ab
fact that there are no transverse vectors in 1 + 1 dimensions. Kij is a properly
3
generalized ordering operator.
In the case of one flavor, i = j = 1, H I does not involve the π variables.
The interaction involves non-local terms which relate to color non-singlets.
For static and ec → ∞ approximations one finds that for NF = 2 the interaction
is field independent. For NF ≥ 3, on the other hand, the limit is singular. This
singularity should not be there in the predictions of physical quantities, but it
renders any further treatment very complicated.
It is thus clear that a different method of bosonization is required for the
treatment of flavored QCD2 . In the following it will be shown that the “non-
abelian bosonization”, based on the WZW model discussed in Section 6.3, is an
adequate tool for this purpose.
Before proceeding to non-abelian bosonization and in particular to gauge the
color symmetry group of the colored-flavored WZW model, we describe briefly
another approach, in which the flavor sector appears in the form of a WZW
model, but for the color degrees of freedom the gauged abelian bosonization is
invoked. As we have seen above one can use the Gauss law to express the gauge
fields in terms of the appropriate fermionic bilinear, which translate into bosonic
group elements as,
i
2∂1 ea = Ψ†ia Ψia = ∂1 TrF (log ga ), (9.13)
i
π
√
where ga ∈ U (NF ) is one out of NC such matrices, and ea = 2 πEaa , the diagonal
element. One can also express Aba for a = b in terms of fermion densities. Inserting
these into the QCD2 Hamiltonian one gets,
H = H0 + HI
(ec )2 −1 2 2 Tr(ga gb−1 )
HI = − Tr log ga gb − πμ
a,b
32π2 NC
a,b
Tr log(ga gb−1 )

+ mcμ NF Tr(ga ), (9.14)
a
H0 includes the fermion kinetic term. For NF = 2 the potential is free from
singularities, for NF ≥ 3 it is not. In the case of NF = 2 the low lying baryonic
spectrum can be extracted. Here we will not follow this approach further and
instead will move on to the fully non-abelian bosonization.
3 See Cohen et al. [62].

9.3 Non-abelian bosonization of QCD2

Whereas abelian bosonization has been very useful to address various abelian
systems, we have seen in the last section that the implementation of this approach
to QCD2 is quite limited. Instead the natural approach is to make use non-
abelian bosonization, namely, the WZW action.4 Recall from Section 6.3 that
the bosonized action of massless free colored flavored fermions can be expressed
either using an SU (NC ) × SU (NF ) × U (1) scheme where it reads,

1
S = NC S[g] + NF S[h] + d2 x∂μ φ∂ μ φ, (9.15)
2
or a U (NF × NC ) where the action takes the form,

1
S[u] = NC S[g] + NF S[h] + d2 x(∂μ φ∂ μ φ + S[l]). (9.16)
2
Note that l is still an SU (NC NF ) matrix while g and h are expressed now as
SU (NF ) and SU (NC ) matrices, respectively, but the matrix l involves only prod-
ucts of color and flavor matrices (not any of them separately). For massive Dirac
fermions we can use only the latter frame in which the mass term action reads,

Sm [u] = m Nm̃ d2 xT r(u + u† ).
2
(9.17)
To determine the bosonized action of two-dimensional QCD2 one needs to

couple the colored degrees of freedom to the gauge fields. Thus, we first have to
gauge the WZW model.
9.3.1 Gauging the WZW action

Since there are two possible bosonization schemes (for the massless case) we
need to invoke a gauging procedure for both of them. We start first by gauging
an SU (NC ) WZW model which is what is needed in the product scheme, we
later adopt it also to the U (NF × NC ). Gauging the colored WZW is achieved
by gauging the vector subgroup SUV (NC ) of SUL (NC ) × SUR (NC ). There are
various methods to gauge the model. Here we present two of them. One is a trial
and error method, and the other is by gauging via covariantizing the current.
Those methods are applicable also in the U (NF NC ) bosonization scheme.
The gauging of the WZW model and the full non-abelian bosonization of QCD
in two dimensions was analyzed in [75] and [99]. Bosonization of QCD in two
dimensions was reviewed in [101].
Trial and error Noether method

The WZW action on the SU (NC ) group manifold is, as stated above, invariant
under the global vector transformation h → U hU −1 , where U ⊂ SU (NC ). Now
4 The hybrid of abelian and non-abelian bosonizations was implemented in [107].

we want to vary h with respect to the associated local infinitesimal transforma-

tion U = 1 + i(x) = 1 + iT A A (x),
δ h = i[, h], δ h−1 = i[, h−1 ]. (9.18)
The variation of the action S (0) [h] ≡ S[h] under such a transformation is,

δ S (0) [h] = − d2 xTr(∂μ J μ ), (9.19)
where the Noether vector current is given by,

i
Jμ = {[h† ∂μ h + h∂μ h† ] − εμν [h† ∂ ν h − h∂ ν h† ]}. (9.20)
4π
We introduce now the first correction term S (1) given by,

S (1) = d2 xTr(Aμ J μ ) δ S (1) [h] = − d2 xTr[∂μ (J μ + J μ )]. (9.21)
The variation of S (1) is derived using the infinitesimal variation of the gauge field

δAμ = −Dμ = −(∂μ + i[Aμ , ]). J μ is found to be,
−1 †
Jμ = {[h Aμ h + hAμ h† − 2Aμ ] − εμν [h† Aν h − hAν h† ]}. (9.22)
4π

The second iteration will be given by adding S (2) , where now J μ is replacing
Jμ,

S = d xTr(Aμ J ), δ S [h] = −2 d2 xTr(∂μ J μ ).
(2) 2 μ (2)
(9.23)
It is therefore obvious that,

1 (2)
δ S + S − S
(0) (1)
= 0. (9.24)
2
Hence the action we are looking for is S[h, Aμ ] ≡ [S (0) + S (1) − 12 S (2) ], given by,

1
S[h, Aμ ] = d2 xTr(Dμ hDμ h† )
8π

1
+ d3 yεij k Tr(h† ∂i h)(h† ∂j h)(h† ∂k h)
12π B

1
− d2 xεμν Tr[iAμ (h† ∂ ν h − h∂ ν h† + ih† Aν h)], (9.25)
4π
which can also be written in light-cone coordinates,

i
S[h, A+ , A− ] = S[h] + d2 xTr(A+ h∂− h† + A− h† ∂+ h)
2π
1
− d2 xTr(A+ hA− h† − A− A+ ). (9.26)
2π
Gauging via covariantization of the Noether current

In four space-time dimensions the current, in terms of bosonic matrices, involves
up to third power gauge potentials. In D space-time dimensions the bare current

will contain (D-1) derivatives, and is gauged by replacing the ordinary derivatives
with covariant derivatives and by adding terms which contain products of Fμν
with powers of h and h† and also covariant derivatives Dμ h and Dμ h† . In two
dimensions, however, there is no room for such terms in the gauge covariant
current, as these involve μ 1 ...μ D in D dimensions, with one free index and the
others contracted with Fμν s and Dμ s, and in two dimensions they cannot be
constructed. Therefore the covariantized current is given by,
i
Jμ (h, Aμ ) = {[h† Dμ h + hDμ h† ] − εμν [h† Dν h − hDν h† ]}. (9.27)
4π
Knowing the current we deduce the action via, Jμ = δδASμ , getting (9.26) directly.
Finally, we combine the gauged WZW action of the color group manifold, the
WZW of the flavor group manifold and the action term for the gauge fields, to
get the bosonic form of the action of massless QCD2 . The well known fermionic
form of the action is (a mass term will be added later),
< 1 =
SF [Ψ, Aμ ] = d2 x − 2 Tr(Fμν F μν ) − Ψ̄ai [(i∂+ A )Ψi ]a , (9.28)
2ec
where ec is the coupling constant to the color potentials (note it has mass dimen-
sions in 1+1 space-time), and,
Fμν = ∂μ Aν − ∂ν Aμ + i[Aμ , Aν ]. (9.29)
The bosonized action is,

1
S[g, h, A+ , A− ] = NC S[g] + NF S[h] + d2 × ∂μ φ∂ μ φ
2

NF
+ d2 xTr[i(A+ h∂− h† + A− h† ∂+ h)
2π
− (A+ hA− h† − A− A+ )]

1
− 2 d2 xTrFμν F μν . (9.30)
2ec
9.3.2 Multiflavor QCD2 using the U(NF × NC ) scheme

Let us now repeat the gauging of the SUV (NC ) subgroup in the framework of
the U (NF × NC ) bosonization procedure.
Using the gauging prescription discussed in Section 9.3.1 we first get the action
in which the whole SU (NC NF ) is gauged, namely,

i
S[u, A+ , A− ] = S[u] + d2 xTr(A+ u∂− u† + A− u† ∂+ u)
2π

1
− d2 xTr(A+ uA− u† − A− A+ )
2π

+ m 2 Nm̃ d2 xTr(u + u† ), (9.31)


where we have also added a mass term with m 2 = mq cm̃. Now since we are
interested in gauging only the SU (NC ) subgroup of U (NF NC ), we take Aμ to
be spanned by the generator T D ⊂ SU (NC ) via Aμ = ec AD D
μ T . We then add
to this action the kinetic term for the gauge fields − 2e 2 d xTr(Fμν F μν ). The
1 2

c
√
coupling ec is related to the color gauge coupling ec by ec = N F ec , so that
after taking the trace over flavor we get the expected kinetic term with coupling
ec . The resulting action is invariant under local color and global flavor,
u → V (x)uV −1 (x), Aμ → V (x)(Aμ − i∂μ )V −1 (x); V (x) ⊂ SUV (NC )
u → W uW −1 ; W ⊂ U (NF ).
The symmetry group is now SUV (NC ) × U (NF ), just as for the gauged fermionic
theory. We choose the gauge A− = 0, so now the action takes the form,
S[u, A+ ] = S[u] + 1
e c 2
d2 xTr(∂− A+ )2 + i
2π d2 xTr(A+ u∂− u† )

+ m 2 Nm̃ d2 xTr(u + u† ). (9.32)

4π
NCNF −i φ
Upon the decomposition u = g̃ h̃le , we see that the current that cou-
ples to A+ is h̃∂− h̃ . In terms of u it is the color projection (u∂− u† )C =
†
† †
N F TrF [u∂− u − N C TrC u∂− u ]. Thus the coupling of the current to the gauge
1 1
i 2 †
field 2π d xTr(A+ h̃∂− h̃ ).
We can further manipulate the action to a form which will be convenient
for taking the strong coupling limit (see Chapter 13). We define H̃(x) by
∂− H̃ = ih̃∂− h̃† . We take the boundary conditions to be H̃(−∞, x− ) = 0 and
then integrate out A+ obtaining,
S̃[u] = S[u] − ( 4π
ec 2
) NF d2 xTr(H̃ 2 )

+ m 2 Nm̃ d2 xTr(u + u† ). (9.33)
In Chapter 13 this form of action will constitute the starting point of determining
the baryonic spectrum of QCD2 in the strong coupling limit. In Chapter (14)
we will use this action to analyze the string tension and the confining behavior
of massive QCD2 .

10
The ’t Hooft solution of 2d QCD
Two-dimensional quantum chromodynamics involves an SU (N ) symmetry

group. We saw that models get simplified in the large N limit, and so we would
like to examine the question of whether the large N limit QCD in two dimensions
can be solved. This question was addressed by ’t Hooft who showed that indeed
QCD2 in this limit is almost exactly soluble. The simplest Green’s functions
can be solved in closed forms and the meson spectrum can be extracted by a
non-elaborate numerical computation.
This was derived by ’t Hooft in his seminal paper [124], and it had many
follow ups. In this chapter we consider only [56], which discusses the scattering
properties of QCD in the large N model.
Recall the action of a two-dimensional QCD,
1
SQ C D = − Tr[Fμν F μν ] + Ψ̄i (i D − mi )Ψi , (10.1)
2
where the gauge fields are spanned by N × N Hermitian matrices T A such that
μ T , Fμν = ∂μ Aν − ∂ν Aμ + i N [Aμ , Aν ], the covariant derivative Dμ =
Aμ = AA A √g
g
∂μ + i √N Aμ , the fermions Ψ are in the fundamental representation of the color
group and i = 1, . . . , Nf indicates the flavor degrees of freedom. There is a sum
over the flavor indices. Note that the gauge coupling was chosen to be √gN ,
obviously to accommodate a large N approximation with g fixed.
It is convenient to impose the algebraic light-cone gauge. This gauge is advan-
tageous at least for the following two reasons:
(i) The field strength F+− becomes linear in the gauge potential,
A+ = A− = 0 ⇒ F+− = −∂− A+ . (10.2)
(ii) The theory after gauge fixing is still Lorentz invariant. This is obviously a
property of two dimensions only.
In this gauge the Lagrangian of the system becomes,
1 g
L = − Tr[(∂− A+ )2 ] + Ψ̄k i∂ − mk − √ γ− A+ Ψk . (10.3)
2 N
Recall that in the light-cone gauge there are no ghost fields.
The Feynman rules associated with this action in the so-called double line
notation follow from Fig. 10.1, as explained below.
In the following we shall be taking one flavor, for simplicity.

192 The ’t Hooft solution of 2d QCD
1 1
p2 p2
i mi k+ +k ik
m2i +2 k+ k i m2i +2 k+ k
j
2g
i
Fig. 10.1. The Feynman rules of QCD2 in the light cone.
Fig. 10.2. The quark self-energy.
The light-cone gamma matrices obey the relations,

2
γ− = γ+2 = 0 {γ + , γ − } = 2. (10.4)
Since the vertex is proportional to γ− , only that part of the propagator that is
proportional to γ+ can contribute. As a consequence we can eliminate all the γ
dependence from the Feynman diagrams. Thus the double line, representing the
−ik −
gluon propagator, is p12 , the fermion line is m 2 +2k + k − −i
, and the coupling is 2g.
−
Note that for the gauge field propagator one makes use of the principal value
such that,

1 1 1 1
D+ + (p) = P ≡ + . (10.5)
p2− 2 (p− + i)2 (p− − i)2
The dressed quark propagator and the quark self-energy, given in terms of the
diagrams in Fig. 10.2, obey the coupled equations,
ip−
S(p) =
2p+ p− − m2 − p− Σ(p) + i

dk+ dk− 1
Σ(p) = 4g 2 2
S(p − k)P , (10.6)
(2π) (k− )2

The ’t Hooft solution of 2d QCD 193
where Σ is the γ+ part, the only part that appears in the self-energy in our gauge.
If we shift the integration variables p+ − k+ → −k+ we eliminate the dependence
on p+ . Hence Σ is only a function of p− . Due to its Lorentz structure it implies
that Σ must be a constant times p1− , namely m2 + p− Σ ≡ M 2 . Thus in the
leading large N the sole effect of the interaction, for the propagator, is to replace
the quark mass m by a renormalized quark mass M .
Integrating over k+ we get,

g2 1 g2
Σ= dk− sgn(p− − k− )P = − , (10.7)
2π (k− )2 πp−
and hence,
g2
M 2 = m2 −
. (10.8)
π
In the original treatment of ’t Hooft, the regularization employed was not of
principal value, but rather of a sharp cutoff, namely integrating over |p− | > λ.
This avoids the infrared divergence as well, but introduces a new scale, which is
not gauge invariant. Obviously, one has to check that Green’s functions of gauge
invariant operators are independent of λ when λ → 0. Thus we find that,
g2 sgn(p) 1
Σ(p) = Σ(p− ) = − − , (10.9)
π λ p−
and correspondingly the dressed quark propagator is,
ip−
S(p) = 2 2 |p | . (10.10)
2p+ p− − m + gπ − g πλ
2 −
+ i
Now the pole of the quark propagator is shifted towards k+ → ∞ and hence
there is no physical single quark state.
Let us consider now the spectrum of the mesonic bound states. The propagator
of the meson is given by the sum of diagrams as is shown in Fig. 10.3.
This ladder sum is exact in the planar limit that follows from the large N
approximation. If the propagator has a meson pole, then the ladder diagrams
have to obey the Bethe–Salpeter equation as in Fig. 10.4.
The “blob” is the Fourier transform of the matrix element,
φ̃(p, q) = F.t. <meson|T ψ̄(x)ψ(0)|0>,
with external legs of a quark of mass m, momentum p, and an anti-quark of mass
m and momentum p − q (for simplicity, we take one flavor, and so the same mass
for the quark and anti-quark). The Bethe–Salpeter equation reads,

d2 k 1
φ̃(p, q) = −4ig S(p − q)S(p)
2
P φ̃(k− , q). (10.11)
(2π)2 (k− − p− )2
Defining

φ(p− , q) = dp+ φ̃(p, q),

Fig. 10.3. The Green’s function of the quark bilinear.
p-q k-q p-q
q = q
p k q
Fig. 10.4. The Bethe–Salpeter equation.
we get,

g2 1
φ(p− , q) = −i dp+ S(p − q)S(p) dk− P φ(k− , q). (10.12)
π2 (k− − p− )2
The integral over p+ can be done explicitly

I(p− , q) ≡ dp+ S(p − q)S(p)

1 1
=− dp+ 2 −i
. (10.13)
M 2 −i
2(p+ − q+ ) − M
p − −q − 2p + − p−
If p− is outside the interval [0, q− ] then the two poles are on same side of the
real axis, and the integral vanishes. When p− is inside the interval, the integral
is (taking q− > 0),

M2 M2
−iπ 2q+ − − ,
p− (q− − p− )
so that,

M2 M2 g2 q − 1
2q+ − − φ(p− , q) = − dk− P φ(k− , q).
p− q− − p− ) π 0 (k− − p− )2
(10.14)

Defining x and y by,
p− = xq− , k− = yq− ,
and,
2q+ q− = μ2 ,
one finally gets ’t Hooft’s equation,

2
M M2 g2 1 1
2
μ φ(x) = + φ(x) − dy )φ(y), (10.15)
x 1−x π 0 (x − y)2
with φ(x) defined on the interval [0, 1].

The equation cannot be solved analytically, but one can compute the wave-
functions that correspond to the various states numerically.
Before describing these solutions let us further discuss the equation. In fact
one can derive the equation using a light-cone Schrödinger equation. In the light-
cone coordinates a system is specified at x+ , and its dynamics is generated by
P+ , the generator of translations of x+ . Since the latter commutes with P− ,
the generator of translations of x− , it is useful to use the eigenspace of P− .
2
For example, for a free single particle of mass M, 2P+ = M P − . Note however
that unlike the ordinary Schrödinger formulation which is expressed in terms
of a real line, the spectrum of P1 , in the light-cone case the spectrum of P− is
the positive half-line. For a system of two particles one can always choose to
normalize the eigenvalue of P− to be one, so that the eigenvalue of the oper-
ator on one of the two particles is x and the on the other it is 1 − x, such
2
M2
that for two non-interacting particles 2P+ = Mx + 1−x . This yields the first two
terms in (10.15). The other term, the integral, is just a linear potential term.
If we interpret temporarily x as a position operator, then the operator form of
(10.15) is,
M2 M2
2P+ = + + g 2 |p|. (10.16)
x 1−x
This is the Hamiltonian of a massless particle moving in a potential and restricted
to a box [0, 1]. This guarantees that the spectrum is discrete and there is no con-
tinuum of two free particles. Moreover we can go further with this interpretation
and argue that at least for high-level states the eigenstates are like those of a
free particle in a box namely,
φn ≈ sin(πnx), μ2n ≈ g 2 πn, (10.17)
for n = 1, 2, . . . These states furnish a linear “Regge trajectory” with no con-

tinuum. We will verify shortly that for large n this is indeed the structure of
the eigenstates and eigenvalues. Since the renormalized quark mass becomes
tachyonic for large coupling constant g (eqn. 10.8) one may wonder whether the
mesonic bound states can also be tachyonic. It turns out that this cannot occur.

From (10.15) it follows that,

1 1
1 1
μ 2
|φ(x)| dx = m
2 2
|φ(x)| 2
+ dx
0 0 x 1−x

g2 1 1
|φ(x)||φ(y)|
+ dx dy . (10.18)
2π 0 0 (x − y)2
To solve the ’t Hooft equation (10.15) we need to specify the boundary condi-
tions.
At x = 0 (x = 1) the solution may behave like x±β ((1 − x)±β ) with,
πM 2
πβ cot g(πβ) + = 0. (10.19)
g2
Let us define the “Hamiltonian” of the system as the right-hand side of equation
(10.15), namely,
2
M M2 g2 1 1
Hφ(x) ≡ + φ(x) − dyP φ(y). (10.20)
x 1−x π 0 (x − y)2
This Hamiltonian is Hermitian only when acting on the space of functions that
vanish on the boundary, as can be seen from (10.18). Using the latter one can
show that φn with φn (0) = φn (1) = 0 constitute a complete orthonormal set,

φn (x)φn (x ) = δ(x − x )
n
1
φ∗n (x)φm (x)dx = δn m . (10.21)
0
Since the integral in (10.15) gets its main contribution from y close to x and
since for a periodic function we have,
1 ∞
eiw y eiw y
P dy P dy = −π|w|eiw x , (10.22)
0 (x − y)2
−∞ (x − y)2
then the configurations given in (10.17) are a good approximation of the eigen-
states of the system. The numerical solutions of eqn. (10.15) are drawn in
Fig. 10.5.
In this figure the mass spectrum of mesons is shown for various values of quark
mass. In cases when the mass of the quark and anti-quark are not equal, the term
2
M2 M2 M 22
[ Mx + 1−x ] in (10.15) is replaced by [ x1 + 1−x ].
The masses and wavefunctions cannot be determined in general in an analytic
form. However in certain limits one can write down approximate expressions. In
[52] it was shown that the highly excited states n 1, where n is the excitation
number have masses given by,
3 1
(Mm es )2n ∼ πg 2 N n+ + m2q 1 + m2q 2 ln(n) + C m2q 1 + C m2q 1 + O ,
4 n
(10.23)

μ2
61
11
1.
2.
=
=
m
m
4π2
27
1.
=
m
0
1.
=
m
3π2
56
0.
=
m
18
0.
=
m
2π2
0
=
m
π2
0
0 1 2 n 3
g2
Fig. 10.5. The spectrum of mesons. The squared masses are in units of π
[124].
where mq i are the masses of the quark and anti-quark and where the functions
C(m2q ) are given in [52].
The opposite limit of low-lying states and in particular the ground state can
be deduced in the limit of large quark masses, namely mq g and small quark
masses g mq . For the ground state in the former limit one finds,
Mm0 es ∼
= mq 1 + mq 2 . (10.24)
In the opposite limit of mq g,

5
∼ π g 2 Nc
(Mm0 es )2 = (m1 + m2 ). (10.25)
3 π
For the special case of massless quarks we find a massless meson.
In Fig. 10.6 the spectrum of meson nonets built from two triplets of flavor
with masses
(a) m1 = 0 m2 = 0.2 m3 = 0.4

(b) m1 = 0.8 m2 = 1.0 m3 = 1.2 (10.26)

μ2−nπ2
a
33
π2
23 32
22 31
13
21
12
11
n=0 n=1 n=2 n=3 n=4
0
b
Fig. 10.6. Meson nonets for Nc = 3. In case (a) the masses of the triplet
are m1 = 0.00, m2 = 0.20, m3 = 0.4 and in (b) m1 = 0.80, m2 = 1.00, m3 = 1.2
[124].
is shown, in units of √gπ . Then the ground state is at 2.7, the first excited state
at 4.16, and level n = 10 is at 20.55. It is obvious from these cases that for larger
n, the wavefunction gets more and more sharply picked around x = 0.5. For the
case of unequal masses, the wavefunction ceases to be symmetric, as can be seen
from Fig. 10.7 for m1 = 1, m2 = 5.
10.1 Scattering of mesons

In the previous section we have described the equation that governs the formation
of mesonic bound states, and the corresponding meson spectrum follows from a
homogeneous Bethe–Salpeter equation. This can be generalized to the equation
for full quark anti-quark scattering amplitude, which takes the form of the non-
homogeneous equation of Fig. 10.8.
The scattering amplitude has the following structure,
Tα β ,γ δ = (γ− )α γ (γ− )β δ T (q, q , p). (10.27)
The undressed amplitude T (q, q , p) takes the form,

ig 2 ig 2 N dk− φ(k− , q , p)
T (q, q , p) = )2 + , (10.28)
(q− − q− π2 (k− − q− )2

10.1 Scattering of mesons 199
1.5
φ(X)
0.5
0
0 0.2 0.4 0.6 0.8 1
X
Fig. 10.7. Wavefunction for m1 = 1, m2 = 5.
q q’ q q’
i i i i
T
p-q p-q’ p-q p-q’

j j j j
k
q S q’
i i j
T
p-k
p-q S p-q’
j j j
Fig. 10.8. The Bethe–Salpeter equation for quark anti-quark scattering.
where,

φ(q− , q− , p) = dq+ SE (q)SE (q − p)T (q, q , p). (10.29)
Similar to the equation for the “wave function” φ(x) we now get the generaliza-
tion to φ(x, x , p) which reads,
2
M M2
μ2 φ(x, x , p) = + φ(x, x , p)
x 1−x
1
π2 [φ(x, x , p) − φ(y, x , p)]
+
+ dy . (10.30)
N p− (x − x )2
0 (x − y)2

It is now straightforward to express φ(x, x , p) in terms of φ(x) as,

πg 2 1
1 φn (x)φ ∗n (y)
φ(x, x , p) = − 2 − p2 p
dy , (10.31)
n
p n − 0 (x − y)2
and substituting this into (10.28) we find the scattering amplitude,

ig 2 ig 2 (g 2 N ) 1
T (x, x , p) = −
p2− (x− x)2 πp2− n
p 2 − p2
n
1 1
φn (y)φ ∗n (y ) ig 2 1
× dy dy 2 (x − y )2
= − x)2
−
0 0 (x − y) p 2
− (x n
p 2 − p2
n
5
∗ 2g g2 N λ γ1 − 1 γ2 − 1
φn (x ) θ(x (1 − x ))+ + −μk2
λ π 2|p− | x 1 − x
× [(x ↔ x)] , (10.32)

Mi
where γi for i = 1, 2 are ( √gπ ) .
This clarifies the dynamics of the confinement. The infinite self-mass quark is
cancelled by the quark anti-quark interaction producing finite mass color singlet
bound states, whose mass squared, as we have seen above, increases linearly for
high excited states. The infrared behavior is determined by the dependence on
λ as in (10.9). The bound state wave function is of order λ1 as λ → 0. The fact
that the amplitude for a bound state to decay into quarks is infinite as λ → 0
compensates for the vanishing quark propagator in this limit to produce finite
bound state amplitudes, which contain no multiquark discontinuities.
To test the consistency of the model one has to examine also the hadronic
scattering processes. One has to check that these are finite in the limit of λ → 0,
unitary and Lorentz invariant. A consequence of the unitarity is the absence of
long range forces among the color singlets.
In Fig. 10.9 the three-particle vertex function and the two-particle scattering
are drawn. The three-particle vertex function, Fig. 10.9(a), is of order g ∼ √1N .
Each quark propagator is of order λ. The k+ loop momentum is of order λ1 since
it is dominated by the pole at λ1 . From the three bound state wave functions
we get a factor of ( λ1 )2 since at least one wave function must be of order unity
to conserve momentum. So altogether the factors of λ cancel out and we get a
finite result in the limit of λ → 0.
The two-particle scattering is described in Fig. 10.9(b) and 10.9(c). The for-
mer describes a hadronic exchange and the latter a quark exchange. The quark
exchange may seem to be infinite in the limit λ → 0 since now the quark and
anti-quark can move in the same direction with an amplitude that behaves like λ1 .
The total dependence on λ is as follows: λ4 from quark propagators, λ14 from the
wave functions and λ1 from the loop momentum integration. However it can be
shown that when one adds all diagrams that contribute to N1 order, the terms of

10.2 Higher 1/N corrections 201
(a)
(b)
(c)
Fig. 10.9. (a) Three-particle vertex function. (b) Hadronic exchange contribu-
tion to two-particle scattering amplitude. (c) Quark exchange contribution to
two-particle scattering amplitude.
order λ1 cancel, leaving a finite remainder. In this way we have verified unitarity
of the model to the first non-trivial order.
10.2 Higher 1/N corrections

At N = ∞ the mesons are stable since their decay rate, as will be shown shortly,
is proportional to N1 . Going to the N1 corrections, a meson has the following
amplitude to decay into two mesons1
√ * w x
4g 2 N 1 x−w
A(i, f1 , f2 ; w) = √ dxφi (x)φf 1 Φf 2
π 1−w 0 w 1−w
1 x +
1 x−w
− dxφi (x)Φf 1 φf 2 , (10.33)
w w w 1−w
1 The 1/N corrections were evaluated in [144].

where φi , φf 1 , φf 2 are the wave functions of the initial meson and first and second
final mesons, respectively. The quark ends up being in the second final meson and
the anti-quark in the first final meson. The vertex function Φ(x), with x not ∈
[0, 1], is related to the wave function as,
1
1
Φ(x) = dy φ(y). (10.34)
0 (x − y)2
The kinematic parameter w takes the values,

μ2i + μ2f 1 − μ2f 2 ∓ (μ2i + μ2f 1 − μ2f 2 )2 − 4μ2i μ2f 2
w± = , (10.35)
2μ2i
where w+ and w− correspond to the right and left moving final state f1 . The
decay can take place only provided μi ≥ μf 1 + μf 2 . It is clear that for fixed g 2 N
the amplitude is of order A ∼ O( √1N ). The amplitude (10.33) is for a partial
decay and for full-on shell amplitude one has to add the partial decays
A = (1 − (−1)σ i +σ f 1 +σ f 2 )(A(i, f1 , f2 ; w+ ) + A(i, f1 , f2 ; w− )), (10.36)
with σ+ for even parity state and σ− for odd parity state. It was found that
numerically these amplitudes for various excited states do not vanish. This also
shows that the model is not integrable.
It was further found that the amplitudes for mesons made out of massless quark
anti-quark pairs differ significantly from those of mesons made out of massive
ones. An interesting result that follows from the computations of these ampli-
tudes is that the amplitude for decay of an exited meson into a pion and another
meson vanishes, in the case of massless quarks. This is actually to be expected,
as for massless quarks the two-dimensional pion is massless and decoupled, since
there is no chiral symmetry breaking in two dimensions.

11
Mesonic spectrum from current algebra
11.1 Introduction
In this chapter we study the mesonic spectrum of various QCD2 theories. The
main idea is to use the current algebra of the underlying ungauged theories. In
addition we combine the bosonization techniques developed in Chapter 6 with
that of a large N expansion of Chapter 7 and a light-front quantization as in
Chapter 10. We will focus our attention on the massive mesonic spectrum of con-
formal field theories coupled to non-abelian gauge fields. In particular massless
multi-flavor fundamental quarks and adjoint quarks that will be shown to
correspond to the particular case of Nf = Nc .
First a universality theorem, that states that the massive mesonic spectrum
does not depend on the representation of the matter field but rather only on its
ALA level, will be derived, following Kutasov and Schwimmer [148].
We then present a detailed determination of the massive mesonic spectrum
using a ’t Hooft-like equation for the wave functions of “currentballs” states. We
will discuss in particular the special cases of Nf = 1, Nf = Nc and Nf Nc . The
last section is devoted to the spectrum of states built by the action of a single
current creation operator on the adjoint vacuum. In both cases it will be shown
that the bosonization approach leads to the introduction of current quanta as
the basic degrees of freedom. Once the mass operator P + P − = M 2 is expressed
in terms of the current quanta, the bosonization has already left the scene.
The main content of this chapter, the mesonic spectrum from current algebra,
is based on [17].1 The spectrum based on the adjoint vacuum was introduced
in [3].
11.2 Universality of conformal field theories coupled to YM2

So far we have mainly discussed the coupling of matter in the fundamental
representation to the two-dimensional Y M fields. Obviously one can also couple
other matter fields to these non-abelian gauge fields. A natural class of matter
theories that one would like to gauge are the conformal field theories which admit
on top of the Virasoro algebra also an affine Lie algebra structure. These theories
which are characterized by the corresponding Lie algebra G and the level k of the
1 This was previously also discussed in [18].

204 Mesonic spectrum from current algebra
affine Lie algebra, are candidates for coupling to non-abelian gauge fields of the
group G. A particular family of such theories are the WZW models, invariant
under G × G of level k. We have discussed in Chapter 6 the gauging of such
models. In this chapter we would like to address the issue of the spectrum of
such gauged conformal field theories, and in particular the massive sector of the
spectrum. In general the Lagrangian density of such a theory reads,
1 2
L = LC F T − 2 Tr Fμν + LI
2e
1
= LC F T − Tr (∂− A+ )2 + Tr [A+ J− ]
2e2

e2 1 e2 1
= LC F T − Tr J + 2 J + = LC F T − Tr J 2 J , (11.1)
2 ∂− 2 ∂
where we have used the light-cone gauge A− = 0. We will be using the notation
of J for J− and J¯ for J+ , and similar for other holomorphic and anti-holomorphic
quantities.
A conformal field theory invariant under the symmetry generated by a G
ALA has holomorphic currents J a in the adjoint representation of G, as well
as anti-holomorphic currents also in the adjoint representation of G. In general
the holomorphic currents obey an ALA with level k and the anti-holomorphic
currents an ALA of level k̄. However, gauging the conformal theory requires
vanishing of the chiral anomaly, namely it requires that,
k = k̄. (11.2)
Next we quantize the system on the light-front. This framework is very conve-
nient since both momenta P − and P + , or equivalently P and P̄ , can be expressed
in terms of J only (with no reference to J).¯ This decoupling of one sector (the
anti-holomorphic one) can be attributed to the fact that in a frame moving to the
right with the speed of light there is no way to interact with massless left-moving
particles. The light-cone Hamiltonian is given by,

1
+
P = dx− : J a (x− )J a (x− )
[C(G) + k]
∞
1 a a
= J J , (11.3)
n=1
n2 −n n
where in the last line we have assumed that the light-cone space direction x− = z
has been put on a circle. Thus the Hamiltonian acts inside current blocks, and
the problem of finding the massive spectrum splits into diagonalizing the decou-
pled blocks of P + on global G singlets. We want to emphasize again that the
light-front dynamics is fully independent of the anti-holomorphic sector, apart
from the constraint that k = k̄. This clearly means that we can replace the
anti-holomorphic sector with another anti-holomorphic sector, provided that

11.2 Universality of conformal field theories coupled to Y M2 205
the latter has a level that equals k. Obviously we could have fixed the oppo-
site gauge A+ = 0, leaving only the anti-holomorphic sector with currents J. ¯ In
that gauge we could have replaced the holomorphic sector with another one,
again provided that it has level k. Thus we conclude that the massive spectrum
does not depend on the representations r and r̄, but only on the gauge group G
and the level k.
We would like to demonstrate this universality in the context of a gen-
eralization of Schwinger’s model, which contains nR right moving fermions
ψiR i = 1 . . . nR and nL left-moving fermions ψiL i = 1 . . . nL [120]. Both the right-
and left-moving fermions are charged with respect to an abelian U (1) gauge
symmetry with charges qiR and qiL respectively. The system is described by the
Lagrangian density,
†
¯ R + ψ L † ∂ψ L + ĀJ − 1 (∂ Ā)2 ,
L = ψiR ∂ψ (11.4)
i i i
4e2
n R †
where J = i qiR ψiR ψiR and we are using the gauge A = 0. Upon integrating
Ā we get,
†
¯ R + ψ L † ∂ψ L − e2 J 1 J.
L = ψiR ∂ψ (11.5)
i i i
∂2
We can now bosonize the system. Note that the fermions at hand are not Dirac
fermions but rather nR right and nL left chiral fermions. The system is consistent
in the sense that there is no chiral anomaly when,
R L

N
N
k ≡
R
qiR k ≡
L
qiL k R = k L = k. (11.6)
i= 1 i= 1
One can use the prescription for chiral bosonization described in Section 6.4.
In fact it is enough to note that the interaction term takes the form,
1
Lint = −e2 J J = e2 (φ)2 , (11.7)
∂2
N R N L L L
where φ = i qiR φR i = i qi φi and φL and φL are the right and left chiral
bosons that corresponds to the right and left chiral fermions. Thus we conclude
that the spectrum includes one massive mode corresponding to φ plus nR − 1 and
nL − 1 massless right- and left-moving particles, respectively. It is now evident
that indeed in accordance with the universality theorem, the massive sector
does not depend on the explicit sequence of charges qiR and qiL but only on the
combination expressed in φ.
Another example of the universality theorem is the case of adjoint fermions.
The ALA associated with the currents built from the adjoint fermions J ab =
ψ ac ψ cb is of level Nc . The CFT based on a WZW model of SU (Nc ) of level
k = Nc is another theory with the same ALA, and hence the massive sector of
the spectrum of these theories should, according to the theorem, be the same.

In the next section we describe the massive spectrum of such models based on a
’t Hooft-like equation for the currents.
11.3 Mesonic spectra of two-current states

In this section we derive the massive meson spectrum built from two current
creation operators acting on the vacuum. In the next section we will discuss
states constructed from a single current acting on the adjoint vacuum.
The first step in the determination of the spectrum is the derivation of a ’t
Hooft-like equation for the wave functions of the “currentball” states, at arbitrary
level Nf . This equation should interpolate between the description of a single
flavor (’t Hooft model), the model Nf = Nc equivalent to adjoint fermions and
the large Nf limit. We will argue that the equation obtained suggests that the
underlying degrees of freedom in the problem are interacting “gluons” with mass
e2 N f
π . Actually, these are related to the color currents, but are color singlets.
Then we will solve the equation for the lowest massive state. Whereas the ’t
Hooft model Nf = 1 is exactly solvable, the multi-flavor case with Nf > 1 is not
solvable even in the Veneziano limit when both Nc and Nf are taken to infinity
(with a fixed ratio), since pair creation and annihilation are not suppressed.
For the case of the adjoint quarks, the results derived using the current quanta
will be shown to be compatible with those computed with fermions as the basic
degrees of freedom discussed in Chapter 12. For large Nf it will be shown that the
e2 N
exact massive spectrum is a single particle with M 2 = π f . This phenomenon
is explained by the fact that this limit can be viewed as an “abelianization” of
the model.
11.3.1 The basic setup

We now establish the basic setup. We start with the fermionic formulation of the
various theories, impose the light-cone gauge, introduce the bosonized version
and finally write down the mass operator.
In Section 8.4 the classical theory of QCD2 with Dirac fermions in the fun-
damental representation was described. Here we will address this case as well
as massless Majorana fermions in the adjoint representation. Recall that these
theories are described by the following classical Lagrangian:

1 2
L = −Tr F + iΨ̄ DΨ , (11.8)
2e2 μν
where Fμν = ∂μ Aν − ∂ν Aμ + i[Aμ , Aν ] and the trace is over the color and fla-
vor indices. For case (i) Ψ has the group structure Ψia where i = 1,. . . ,Nc
and a = 1,. . . ,Nf with Dμ = ∂μ − iAμ , whereas for case (ii) Ψ ≡ Ψij and Dμ =
∂μ − i[Aμ , ]. In both cases Ψ is two-spinor parametrized as Ψ = ( ψ̄ψ ).

As we have seen in Chapter 10 it is useful to handle these models in the frame-

work of light-front quantization, namely, to use light-cone space-time coordinates
and to choose the chiral gauge A− = 0. In this scheme the Lagrangian takes the
form,
1
(∂− A+ ) + iψ † ∂+ ψ + iψ̄ † ∂− ψ̄ + A+ J + ,
2
L=− (11.9)
2e2
where color and flavor indices were omitted and J + denotes the + component of
the color current J + ≡ ψ † ψ. This Lagrangian density is identical to (11.4) when
one replaces the complex coordinates with light-cone ones.
By choosing x+ to be the ‘time’ coordinate it is clear that A+ and ψ̄ are non-
dynamical degrees of freedom. In fact, ψ̄ are decoupled from the other fields,
so in order to extract the physics of the dynamical degrees of freedom, one has
to functionally integrate over A+ . The result of this integration is the following
simplified Lagrangian,
e2 + 1 +
L = L0 + LI = iψ † ∂+ ψ + iψ̄ † ∂− ψ̄ − J 2J . (11.10)
2 ∂−
Since our basic idea is to solve the system in terms of the “quanta” of the col-
ored currents, it is natural to introduce bosonization descriptions of the various
fields.
(i) As was discussed in Section (9.3.2), the bosonized action of colored-flavored
Dirac fermions in the fundamental representation is expressed in terms of a WZW
action of a group element u ∈ U (Nc × Nf ), with an additional mass term that
couples the color, flavor and baryon number sectors. In the massless case when
the latter term is missing, the action takes the form,

1
S0fund = S(N
W ZW
f )
(g) + S W ZW
(N c ) (h) + d2 x∂μ φ∂ μ φ, (11.11)
2

4π
i φ
where g ∈ SU (Nc ), h ∈ SU (Nf ) and e N c N f ∈ UB (1), with UB (1) denoting
the baryon number symmetry, and the WZW action was given in Section 4.1.
(ii) The current structure of free Majorana fermions in the adjoint repre-
sentation can be recast in terms of a WZW action of level k = Nc , namely
S0adj = S(N
W ZW
c)
(g), where now g is in the adjoint representation of SU (Nc ), so
that it carries a conformal dimension of 12 . Multi-flavor adjoint fermions can be
described as SN W ZW W ZW
(g) + SN (h) where g ∈ SO(Nc2 − 1) and h ∈ SO(Nf ). In
c −1
2
f
the present work we discuss only gauging of SU (Nc ) WZW so the latter model
would not be considered.
Substituting now S0fund or S0adj for S0 the action that corresponds to 11.10
becomes,

e2 1
S = S0 − d2 xJ + 2 J + , (11.12)
2 ∂−

k
where the current J + now reads J + = i 2π g∂− g † , and the level k = Nf and k =
Nc for the multi-flavor fundamental and adjoint cases, respectively.
The light-front quantization scheme is very convenient because the correspond-
ing momenta generators P + and P − can be expressed only in terms of J + . We
would like to emphasize that this holds only for the massless case.
Using the Sugawara construction, the contribution of the colored currents to
the momentum operator P + takes the simple form,

1
+
P = dx− : Jji (x− )Jij (x− ) :, (11.13)
Nc + k
√
where J ≡ πJ + , Nc in the denominator is the second Casimir operator of
the adjoint representation and the level k takes the values mentioned above.
Note that for future purposes we have added the color indices i, j = 1 . . . Nc to
the currents. In the absence of the interaction with the gauge fields the second
momentum operator P − vanishes. For the various QCD2 models it is given by,

e2 1
P− = − dx− : Jji (x− ) 2 Jij (x− ) : . (11.14)
2π ∂−
In order to find the massive spectrum of the model we should diagonalize
the mass operator M 2 = 2P + P − . Our task is therefore to solve the eigenvalue
equation,
2P + P − |ψ = M 2 |ψ . (11.15)
We write P + and P − in term of the Fourier transform of J(x− ), defined by,

dx− + −
+
J(p ) = √ e−ip x J(x− ).
2π
Normal ordering in the expressions of P + and P − are naturally with respect
to p, where p < 0 denotes a creation operator, and to simplify the notation we
will write from here on p instead of p+ . In terms of these variables the momenta
generators are,
∞
P+ = 2
N +k 0
dpJji (−p)Jij (p)
e2 ∞
P− = π 0
dp p12 Jji (−p)Jij (p). (11.16)
Recall that the light-cone currents Jji (p) obey a level k, SU (Nc ) affine Lie algebra,
1 1 1 n
Jik (p), Jln (p ) = kp δin δlk − δik δln δ(p+p )+
Ji (p+p )δlk −Jlk (p+p )δin .
2 N 2
(11.17)
We can now construct the Hilbert space. The vacuum |0, R is defined by the
annihilation property,
∀p > 0, J(p)|0, R = 0, (11.18)

where R is an “allowed” representation depending on the level. Thus, a physical

state in Hilbert space is,
Tr J(−p1 ) . . . J(−pn )|0, R .
Note that this basis is not orthogonal.
11.3.2 ’t Hooft-like equation for the two-current wave function

We restrict ourselves to the simplest case of the two-current sector of the Hilbert
space, (in Section 11.4 we will also mention the special case of one current on an
adjoint vacuum),
1
1
|Φ = dk Φ(k)J a (−k)J a (k − 1)|0 , (11.19)
Nc Nf 0
namely to states which are color singlets of two currents with total P + = 1
momentum and a distribution of P − momentum Φ(k). Note that Φ is a sym-
metric function,
Φ(k) = Φ(1 − k). (11.20)
Our task now is to find the eigenvalue (Schrödinger) equation for the wave
function Φ(k). Let us start by the action of the “Hamiltonian” P − on the state
|Φ .
The commutator of P − with a current J b (−k) yields the result,
∞
dp a 1 1 1
2
J (−p)J a
(p), J b
(−k) Nf − Nc + Nc J b (−k)
0 p 2 k
∞
1 1
+ dp − if abc J a (−p)J c (p − k)
k p 2 (p − k)2
k
dp abc c
+ 2
if J (p − k)J a (−p). (11.21)
0 p
We introduced as an IR cutoff, namely, the lower limit of integration. This is

the analog of λ in the derivation of the ’t Hooft equation of Chapter 10. We take
to go to zero at the end of the calculation.
The above expression (11.21) contains three terms on the right-hand side The
first term contains a single creation operator. The second term contains an anni-
hilation current and therefore should again be commuted with J b (k − 1). The
third term contains two creation currents and it would lead to a three-current
state. This is a manifestation of the fact that pair creation is, generically, not
suppressed in multi-flavor QCD2 .
Note that while deriving eqn (11.21) we get an “infinite” contribution
Nc 1 J b (−k). This contribution will be cancelled by a counter contribution which
comes from the regime p ∼ k in the first integral on the right-hand side of (11.21),
as below.

The commutator of the second term on the right-hand side of (11.21) with
J b (k − 1) yields,
∞
1 1
dp − if J (−p)J (p − k), J (k − 1)
abc a c b
(11.22)
k p2 (p − k)2
∞
1 1
= Nc dp − (J a (−p)J a (p − 1) − J a (p − k)J a (k − p − 1)).
k p 2 (p − k)2
Our results can be summarized by the following set of equations,

1
1 1
M |Φ =
2
dk Φ̃(k)J a (−k)J a (k − 1)|0 + 3 (11.23)
Nc Nf 0 (Nc Nf ) 2
1
× dk dp dl δ(k + p + l − 1)Ψ(k, p, l)if abc J a (−k)J b (−p)J c (−l)|0 ,
0
with,
1
2e2 (Nc Nf ) 2 Φ(l) − Φ(k)
Ψ(k, p, l) = , (11.24)
π p2
and,
e2 1 1 2Nc
Φ̃(k) = (Nf − Nc ) + Φ(k) + Φ(k) (11.25)
π k 1−k
k − 1 k
Φ(p) Φ(p) 1 1
−Nc dp −Nc dp +Nc 2 − dpΦ(p)
0 (p−k)2 k + (p−k)2 k (1−k)2 0
Ignoring the three-current term (see below), we get that Φ(k) obeys the eigen-
value equation,
M2 1 1
Φ(k) = (Nf − Nc ) + Φ(k) (11.26)
2
e /π k 1−k
1 k
Φ(p) 1 1
−Nc P dp + Nc − dp Φ(p).
0 (p − k)2 k 2 (1 − k)2 0
1
We assumed that 0 dp Φ(p) = 0, which we will justify shortly.
For general Nc and Nf , discarding the three-current term is unjustified. How-
1
ever, since the length of Ψ is |Ψ(k, p, l)|∼ e2 (Nc Nf ) 2 , in the limit of large Nc
with fixed e Nc and fixed Nf , or large Nf with fixed e2 Nf and fixed Nc , the
2
three-current contribution is indeed negligible, as compared with the two-current

term, the latter being of order 1.
The first integral in eqn. (11.26) should be calculated as a principal value
integral (denoted by P). The divergent part of this integral (arising from the
regime p ∼ k) cancels the previously mentioned infinity. In order to make contact

with the ordinary ’t Hooft equation, it is useful to integrate eqn. (11.26) with
k
respect to k and rewrite it in terms of ϕ(k) ≡ 0 dp Φ(p), to get,
1
M2 1 1 ϕ(p)
ϕ(k) = (Nf − N c ) + ϕ(k) − N c P dp
e2 /π k 1−k (p − k)2
k 1 0
ϕ(p) ϕ(p)
+ Nf dp 2 + Nf dp . (11.27)
0 p k (1 − p)2
The derivation goes as follows. First, integrating eqn. (11.20) we get ϕ(k) =
−ϕ(1 − k) + const. Then taking ϕ(1) = 0 we get,
ϕ(k) = −ϕ(1 − k). (11.28)
1
Now ϕ(1) = 0 implies 0 dkΦ(k) = 0, which was our assumption above. Then,
differentiating (11.27) we do get (11.26), and by the last equation we also get
that there is no extra integration constant.
We would like to comment on the issue of the Hermiticity of the “Hamiltonian”
M 2 . Naively, it seems that M 2 is not Hermitian with respect to the scalar product
1
<ψ|ϕ> = 0 dkψ (k)ϕ(k), since the Hermitian conjugate of (11.27) is,
†
M2 1 1
ϕ(k) = (Nf − Nc )+ ϕ(k)
e2 /π k 1−k
1 k 1
ϕ(p) 1 1
−Nc P dp − Nf 2 dpϕ(p) − Nf dpϕ(p).
0 (p − k)2 k 0 (1 − k)2 k
(11.29)
However, as we shall see in the next subsection, the numerical solution yields
real eigenvalues and eigenfunctions. Therefore, at least on the subspace which is
spanned by the eigenfunctions, namely real functions that are zero at k = 0, 1 and
anti-symmetric with respect to k = 12 , the operator M 2 is Hermitian. Note that
(11.29) is “more regular” than (11.27), as in (11.27) it is ϕ(p)/p2 that appears
in the integration from zero.
Equation (11.27) is similar to the ’t Hooft equation for a massive single flavor
e2 N
large Nc QCD2 , with m2 = π f . It differs from ’t Hooft’s equation by having two
additional terms (the two last terms in (11.27)). It suggests that the dynamics
that governs the lowest state of the multi-flavor model is given, approximately, by
a model of a massive “glueball” with an SU (Nc ) gauge interaction and additional
terms which are proportional to Nf .
Before we present our solution of (11.27) it is important to note that it is only
an approximate solution. We neglected the three-current state with, a priori, no
justification. We shall see, however, that the restriction to the truncated two-
current sector is an excellent approximation for the lowest massive meson.

11.3.3 The two-current mesonic spectrum

The most convenient way to solve (11.27) is to expand ϕ(k) in the basis,
∞
1 β +i
ϕ(k) = Ai k − [k(1 − k)] . (11.30)
i= 0
2
The value of β is chosen so that the Hamiltonian will not be singular near k → 0
or k → 1. This consideration leads to the equation,
Nf Nf /Nc
−1 − + βπ cot βπ = 0, (11.31)
Nc β+1
as derived from eqn. (11.29). Had we started with (11.27), it would have been
−β replacing β in (11.31), and constrained to β larger than 1.
Upon truncating the infinite sum in (11.30) to a finite sum, the eigenvalue
problem reduces to a diagonalization of a matrix. So, the problem can be
reformulated as,
λNij Aj = Hij Aj , (11.32)
with,
1 2
1 2β +i+j
Nij = dk k − (k(1 − k)) , (11.33)
0 2
and,
1 2
Nf 1 2β +i+j −1
Hij = −1 dk k − (k(1 − k))
Nc 0 2
1 k
Nf 1 β +i 1 1 β +j
− dk k − (k(1 − k)) p− (p(1 − p))
Nc 0 2 k2 0 2
1
Nf 1 1 β +i 1 1 β +j
− dk k − (k(1 − k)) p− (p(1 − p))
Nc 0 2 (1 − k) k
2 2
1 β +i β +j
k − 2 (k(1 − k))
1
p − 2 (p(1 − p))
1
− dkdp (11.34)
0 (k − p) 2
Hence,
B(2β + i + j + 2, 2β + i + j + 2)
Nij = , (11.35)
2(2β + i + j + 1)
and,
Nf B(2β + i + j + 1, 2β + i + j + 1)
Hij = −1
Nc 2(2β + i + j)
Nf B(2β + i + j + 1, 2β + i + j + 1)
−
Nc 2(2β + i + j)(β + j + 1)
(β + i)(β + j)B(β + i, β + i)B(β + j, β + j)

+ , (11.36)
8(2β + i + j)(2β + i + j + 1)

Table 11.1. The mass of the lowest massive meson,

2
in units of e πN c , as a function of Nf /Nc and β.
β Nf /Nc M2
0.0000 0 5.88
0.0573 0.2 6.91
0.1088 0.4 7.91
0.1552 0.6 8.91
0.1978 0.8 9.89
0.2366 1.0 10.86
0.2725 1.2 11.83
0.3050 1.4 12.77
0.3360 1.6 13.73
0.3645 1.8 14.67
where B(x, y) is the beta function,
Γ(x)Γ(y)
B(x, y) = . (11.37)
Γ(x + y)
In practice, the process converges rapidly and a 5 × 5 matrix yields the ‘contin-
uum’ results.
N
The lowest eigenvalues of (11.27) as a function of the ratio N fc are listed in
Table 11.1 (see also Fig. 11.1). Note that by β = 0, Nf /Nc = 0 we mean the limit
β → 0, Nf /Nc → 0.
These values are in excellent agreement with recent DLCQ calculations, as
will be given in the next chapter.
The typical error is less than 0.1 %.
An interesting observation is that the eigenvalues depend linearly on Nf ,
Fig. 11.1. The dependence is,
e2 Nc Nf
M2 = 5.88 + 5 . (11.38)
π Nc
We do not have a good understanding of this observation. It is not clear why the
lowest eigenvalue sits on a straight line.
In the following sections we will consider some special cases.
11.3.4 Special cases: Nf = 1, Nf = Nc and Nf Nc

We now discuss three special cases, the massless ’t Hooft model where the
fermions are in the fundamental representation with Nf = 1, the case of adjoint
fermions namely Nf = Nc and the “abelianized” model of large Nf Nc .

Mass Eigenvalues
16
14
12
Mass
Square
10
6
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Nf /Nc
Fig. 11.1. The Green’s function of the quark bilinear.
Nf = 1, currentized massless ’t Hooft model

The limit Nc → ∞ with e2 Nc fixed and Nf Nc corresponds to the well-known
’t Hooft model. In this limit QCD2 was solved exactly by ’t Hooft [124] (see
Chapter 10), using the fermionic basis. Let us see how our approach looks in the
fermionic basis in this case. In the limit Nf Nc we can neglect terms which
are proportional to Nf . Equation (11.27) takes the form,
1
M2 1 1 ϕ(p)
ϕ(k) − Nc + ϕ(k) − Nc P dp , (11.39)
e2 /π k 1−k 0 (p − k)2
which is just the ’t Hooft equation for the massless case. Note that (11.39) is
−1
exact, since in the small Nf limit the three-current state is suppressed by Nc 2
with respect to the two-current state and therefore we can neglect it. Note also
that in this eqn. (11.29) looks the same too.
Since the wave function ϕ(k) is anti-symmetric, we will recover only the odd
states in the spectrum of QCD2 (the even states can be recovered by considering
other sectors of the Hilbert space which decouple from the two-current state).
Though eqn. (11.39) is formally the same as the ’t Hooft equation, the interpre-
tation of ϕ(k) should be different. It is the integral of the function Φ(k) which
corresponds to the two-current state, namely to a mixture of 4-fermions and

2-fermions. What is the relation between the states that we find here and the
mesons in ’t Hooft’s model?
In order to answer this question let us expand the currents in terms of fermions.
It is useful to denote the current in double index notation
∞
1 i k
J (k) → Jj (k) =
a i
dq Ψ̄i (q)Ψj (k − q) − δj Ψ̄ (q)Ψk (k − q) . (11.40)
−∞ N c
We do not bother about normal ordering, as no problem for k non zero, and we
have to treat the k = 0 part in a limiting way. The state |Φ can be written as,
1
1
|Φ = dk Φ(k)Jji (−k)Jij (k − 1)|0 (11.41)
2Nc 0
1 ∞
1 1
= dk Φ(k) dq dp Ψ̄i (−q)Ψj (−k + q)− δji Ψ̄k (−q)Ψk (−k + q)
2Nc 0 −∞ N c
1 j
× Ψ̄j (−p)Ψi (k + p − 1) − δ Ψ̄k (−p)Ψk (k + p − 1) |0 .
Nc i
Note that the above expression (11.41) contains creation and annihilation
fermionic operators. Written in terms of creation operators only; (11.41) reads,
1 k 1−k
1
|Φ = dk dq dp Φ(k)Ψ̄i (−q)Ψj (−k + q)Ψ̄j (−p)Ψi (k+p−1)|0
2Nc 0 0 0
1 k 1−k
1
− dk dq dp Φ(k)Ψ̄i (−q)Ψi (−k + q)Ψ̄j (−p)Ψj (k + p − 1)|0
2Nc2 0 0 0
1 k
1
− 1− 2 dk dq Φ(k)Ψ̄i (−q)Ψi (q − 1)|0 . (11.42)
Nc 0 0
The last term in (11.42) corresponds to a meson. It can be written also as,
1 1 1
dq dk Φ(k)Ψ̄ (−q)Ψi (q − 1)|0 = −
i
dq ϕ(q)Ψ̄i (−q)Ψi (q − 1)|0 ,
0 q 0
(11.43)
which is exactly the ’t Hooft meson. We conclude that the two-current state has
an overlap with the ’t Hooft meson and this is why (11.27) reproduces exactly
the (odd part of the) spectrum of the ’t Hooft model.
Large Nf Nc limit
In the limit Nf Nc , with e2 Nf fixed, the truncation to two-current state
should again predict exact results. The reason is that the three-current state
−1
is suppressed by Nf 2 with respect to the two-current state.
In this limit eqn. (11.26) takes the form,
e2 Nf 1 1
M2 = + . (11.44)
π k 1−k

e2 N
It describes a continuum of states with masses above 2m, where m2 = π f .
The interpretation is clear: in this limit the spectrum of the theory reduces to a
single non-interacting meson (or “currentball”) with mass m.
Nf = Nc , The Adjoint Fermions Model

The case Nf = Nc is the most interesting one. It was shown that the massive
spectrum of this model is equivalent to the massive spectrum of a model with a
single adjoint fermion, due to ‘universality’ [148]. Since this model is not exactly
solvable, it is interesting to see how our approach reproduces, almost accurately,
previous numerical results.
The mass of the lowest massive meson, predicted by (11.27), is M 2 = 10.86 ×
e2 N c 2 2
π . The values reported from DLCQ calculations are M = 10.8 and M =
e2 N c
10.84, in units of π , as will be detailed in the next chapter.
This agreement is very surprising. In the regime Nf ∼ Nc , the three-current
state is not suppressed by factors of color or flavor with respect to the two-
current state. Why, therefore, is our approach so successful? The reason seems
to be that as in the fermionic basis [38], the lowest massive state is an almost
pure two-current state. However, the present approach is much more successful
than the fermionic basis, where the prediction for the mass of the lowest massive
boson of the adjoint model is twice as much as the lowest massive boson of the
’t Hooft model. It seems that the “correct” underlying degrees of freedom are
currents and not fermions, as predicted by the authors of [148].
To summarize, we have used a description of massless QCD2 in terms of cur-
rents. With this basis we wrote down a ’t Hooft-like equation (11.27) for the
wave function of the two-current states.
The equation interpolates smoothly between the description of a single flavor
model with large Nc (’t Hooft model), the adjoint fermions model Nf = Nc and
the large Nf model. The equation is derived by using an a-priori unjustified
suppression of the three-current coupling. Nevertheless, we observe an excellent
agreement with the DLCQ results for the first excited state. For higher excited
states the agreement deteriorates and it is of the order of 20%.
The accuracy of the results for the first excited state, which implies that for
this state the truncation of the “pair creation terms” is harmless, deserves further
investigation.
11.4 The adjoint vacuum and its one-current state

Next we construct the spectrum of states, which is obtained by the action of a
current on the “adjoint vacuum”, in the color singlet combination. This way we
get physical states, which are in a sense “one-current” states.

The “adjoint vacuum” is created from the singlet vacuum by applying the
adjoint zero mode, which is taken as the limit → 0 of the product of quark and
anti-quark creation operators, each one at momentum . Hence in our case,
†
|0, R = lim ψ−1
i
()ψ−1,j () |0 , (11.45)
→0
i †
where ψ−1 and ψ−1,j are the creation operators of a quark and anti-quark respec-
tively. We can represent the action of the above adjoint zero mode on the vacuum
by the derivative of a creation current taken at zero momentum. Differentiating
the current with respect to k, and acting on the vacuum we get,
5 ∞ ∞
π d †
Jj i (k) |0 k = 0 − = dp i
dqδ(k + p + q)ψ−1 (p)ψ−1,j (q) |0 k =0 −
2 dk 0 0
5
π i
=− ψ ()ψj† () |0 →0 . (11.46)
2 −1
As the currents are traceless, we have to subtract the trace part for i = j. The
latter can be neglected for large Nc . For any given Nc , results that follow are
also the same after the trace is subtracted.
The adjoint vacuum we have is a bosonic one, constructed from fermion-
antifermion zero modes, and as we show it can be written as the derivative
of the current acting on the singlet vacuum. In the case of adjoint fermions there
is another adjoint vacuum, a fermionic one, obtained by applying the adjoint
fermion zero mode on the singlet vacuum.

As we showed already, (J a ) (0) |0 represents the adjoint zero mode
b† (0)d† (0) |0 (indices suppressed), for any Nf and Nc , so in particular also
for Nf = Nc . But in the latter case the theory is equivalent to that of adjoint
fermions, as follows from the equivalence theorem discussed in Section 11.2. As
also stated there, states built on the adjoint vacuum above, cannot be distin-
guished from those built on the fermionic adjoint vacuum, the latter obtained
by applying the adjoint fermions on the singlet vacuum.
The adjoint bosonic vacuum can also have flavor quantum numbers, when the
fermion has flavor. This does not change our results about the mass of the new
state we have. Our “currentball” will have flavor too in such a case. In our scheme
of bosonization, which is the “product scheme”, especially convenient when the
quarks are massless, the flavor sector is decoupled, and so the flavor multiplets
are given by the action of flavor zero modes, not changing the mass values.
Let us introduce the notation,
5
2 a
Z ≡−
a
(J ) (0).
π
The state we have in mind is,
|k = J b (−k)Z b |0 .

This state is obviously a global color singlet, but in our light-cone gauge A− = 0
it is also a local color singlet, as the appropriate line integral vanishes.
Now,
5
π a 1
J (p), Z b = Nf δ ab δ(p) − if abc (J c ) (p), (11.47)
2 2
and thus, for p > 0,
5
2 abc c
J (p)Z |0 = Z J (p)|0 − i
a b b a
f (J ) (p)|0 = 0.
π
Hence the state Z b |0 is annihilated by all the annihilation currents, and so it is
indeed a colored vacuum.
Using,
+ b
P , J (−k) = kJ b (−k), (11.48)
we get that our state |k is indeed of momentum k.

Note that when quantizing on a circle of radius R, the adjoint vacuum would
be an eigenstate of P + with eigenvalue Nc /R. As we work in the continuum
limit, we get zero.
11.4.1 The action of M2 on the one-current states

First, we evaluate the commutator of P − with a creation current,
∞
dpφ(p)J a (−p)J a (p), J b (−k)
0
k
1 1 b
= Nf J (−k) + if abc
dpφ(p)J a (−p)J c (p − k)
2 k 0
∞
+ if abc
dp (φ(p) − φ(p − k)) J a (−p)J c (p − k),
k
−
note that in P (and in P + ) we ignore contributions from zero-mode states, that
is, we cut the integrals at , and then take the limit.
As P + and P − act on a singlet state, and as J a (0), being the color charge,
annihilates this state, the contribution from the zero modes in both P + and P −
is zero. Therefore it is legitimate to cut the integration limit above the zero mode
and then take the cutoff to zero, as we have done. Note also that the integral
of φ(p) around p = 0 is finite, and in fact zero when integrating over the whole
line, therefore there are no divergences when we take the limit.
It is important, however, to remember that the zero mode does contribute
when we act upon non singlet states, like the adjoint vacuum Z b |0> itself.
When quantizing on a circle of radius R one gets that P + is of order 1/R.
And then, with P − of order e2 R, M 2 is R independent, and so remains finite
in the continuum limit. However, this is subtle, as P − becomes IR divergent in

the continuum and needs to be regularized. This subtlety does not affect our
calculation as we work in the singlet sector only.
Actually, the argument connected with P − acting on singlets should be some-
what sharpened. Let us put the lower limit at , and let it go to zero at the end.
This IR cutoff is similar to the one introduced in the derivation of ’t Hooft’s
model discussed in Section 10. Then J(), when acting on a singlet, would go
like . We have two currents in the integral, so we get 2 . But then we have 1/2
from the denominator, so a finite integrant. But the region for integration is of
order , so indeed the total contribution goes to zero.
Now apply P − on our state,

P − J b (−k)Z b |0 = P − , J b (−k) Z b |0 , (11.49)
as the Hamiltonian annihilates the color vacuum as well.

Using the commutator of the Hamiltonian with a current, we get,
π − b 1 1
P J (−k)Z b |0 = Nf J b (−k)Z b |0
e2 2 k
k
+ if abc dpφ(p)J a (−p)J c (p − k)Z b |0 .
0
Note that we use the fact that annihilation currents do annihilate the colored
vacuum also.
Let us apply the operator M 2 to our one-current state,
M 2 J b (−k)Z b |0 = 2P − P + J b (−k)Z b |0 2kP − J b (−k)Z b |0

k
e2 Nf 2e2
= J (−k)Z |0 +
b b
k if abc dpφ(p)J a (−p)J c (p − k)Z b |0 .
π π 0
(11.50)
So it seems that, in the large Nf limit, the state J b (−k)Z b |0 is an (approximate)
e2 N
eigenstate, with eigenvalue π f .
To see the exact dependence of the two terms in the equation above (the
one- and two-current states) on Nf and Nc , we should normalize them. The
normalization of J b (−k)Z b |0 is,

0| Z a J a (k)J b (−k)Z b |0 = 0| Z a J a (k), J b (−k) Z b |0
1
= Nf kδ(0) 0| Z b Z b |0 + if abc 0| Z a J c (0)Z b |0 (11.51)
2
1
= Nf kδ(0) 0| Z b Z b |0 + Nc 0| Z b Z b |0 .
2
The second term in the last line can be neglected compared with the first, as it
is a constant to be compared with δ(0) [the space volume divided by 2π].
Now,
0|Z b Z b |0 = (Nc2 − 1) 0|Z 1 Z 1 |0 ,

and the factor kδ(0) is the normalization of a plane wave of momentum k. So

the normalized state is, for Nc 1,
1
J b (−k)Z b |0 , (11.52)
1
Nc 2 Nf
relative to 0|Z 1 Z 1 |0 .
The normalization of the second term is more complicated. A lengthy but
straightforward calculation gives,
> k >2
> >
> >
> if kdef
dqΦ(q)J (−q)J (q − k) Z e |0
d f
>
> 0 >
1
2
3 ) ) 4
= Nc Nc2 − 1 Nf kδ(0) 0 )Z 1 Z 1 ) 0 (11.53)
2

k k k −p
Nc
×k dpp(k − p)Φ(p) (Φ(p) − Φ(k − p)) − dpΦ(p) dqqΦ(q) .
0 Nf 0 0
Using the following relations,
f abc f abd = Tr(T c T d ) = N δ cd

abc a bc aa d cc d
f f f f = Tr(T b T d T b T d )
1 2 2
= if bde Tr(T e T b T d ) + Tr(T b T b T d T d ) = N (N − 1),
2
we have evaluated only the terms proportional to δ(0) as they are the dominant
ones.
The various momentum integrals (including the ones for the non dominant
terms) are divergent for → 0, thus they should be regulated. We leave this
problem for now, and assume henceforth that they are regulated and finite. For
simplicity the integrals (including the factor k) appearing in the two dominant
terms will be denoted R1 and −R2 in the following expressions. Note that we
2
have 12 and 1 divergences and also ln( k 2 ). It seems that these are cancelled in
R2 .
Define now the normalized states,

|S1 = C1 J b (−k)Z b |0 (11.54)
k
|S2 = iC2 kf abc
dpΦ(p)J a (−p)J c (p − k)Z b |0 , (11.55)
0
where,
√
2
1 Nf N c3
C1 , C2 = . (11.56)
1 Nc
Nc 2 Nf
R1 + R2 N f

The mass eigenvalue equation of the normalized states is,

5
e2 e2 Nc √ Nf
M 2 |S1 = Nf |S1 + 2 R1 + R2 |S2 , (11.57)
π π Nc
thus, we see that in the large flavor limit, our state |S1 is an eigenstate with
mass
5
e2 Nf
M= . (11.58)
2π
In the large color limit, however, we actually get that the second term dominates
by a factor of Nc . Moreover, while the first term goes to zero in the large Nc
limit, due to the factor of e2 , the second term survives in that limit.

12
DLCQ and the spectra of QCD with fundamental
and adjoint fermions
12.1 Discretized light-cone quantization

So far we have analyzed the mesonic spectra of QCD in low dimensions using the
methods of the large Nc limit (’t Hooft model), and bosonization or currentiza-
tion. In both these methods we have chosen a light-cone gauge and implemented
a correlated light-front quantization. To get a broader perspective of the spec-
tra in this framework, and in particular to extract the mesonic spectra using
fermionic degrees of freedom with finite Nc , we now invoke another tool, the
discrete light-cone quantization. We first describe the method and then apply it
to both QCD with fundamental quarks and with adjoint quarks.
The discretized light-cone quantization (DLCQ) is a method devised to com-
pute spectra and wave functions of physical states of quantum field theories.1 It
is based on the following ingredients:
r A Hamiltonian formulation of the theory.
r Calculations in momentum representation.
r Periodic boundary conditions and hence discretized momenta.
r Light-front quantization.
The Hamiltonian approach is used since it is more convenient for analyzing

the structure of bound states. The periodic boundary conditions assure that
charges associated with symmetries are strictly conserved. For a conserved cur-
rent ∂ + J+ + ∂ − J− = 0 the light-front charge is conserved,
L
dQ(x+ )
Q(x+ ) ≡ dx− J + (x− , x+ ) = 0, (12.1)
−L dx+
provided that,
J + (x+ , +L) − J + (x+ , −L) = 0, (12.2)
which is guaranteed by the periodic boundary conditions.

Note that, the light-front plane of constant x+ , serving as “time”, is gener-
ally called “light-cone quantization”, although the plane is only tangential to
the light cone. In general d-dimensional space-time one may view the DLCQ
1 For reviews see [48] and [47].

224 DLCQ and the spectra of QCD with fundamental and adjoint fermions
approach as a projection into non-relativistic dynamics since for a fixed light-cone

momentum P + the Hamiltonian H = P − is quadratic in the transverse momenta
P2 M2 i
H = 2Pi + + 2P + where P are the transverse momenta.
In spite of a lot of progress in handling problems faced by the DLCQ there
are still several unresolved issues such as:
r There is no proof that the light-front dynamics is fully equivalent to that of
ordinary time evolution, in particular for massless chiral fermions.
r There are subtleties with the renormalization of a Hamiltonian matrix with a
cutoff such that all physical results are independent of the cutoff.
r The implementation of a proper quantization for the system, with constraints
which emerge from gauge fixing.
12.2 Application of DLCQ to QCD2 with fundamental fermions

Instead of describing the method in general, we demonstrate the application
of the DLCQ method to the case of QCD2 with fundamental fermions. The
light-front action of two-dimensional SU (N ) YM gauge fields coupled to Dirac
fermions in the fundamental representation of the gauge group in the light-cone
gauge A+ = 0 reads,2

−1 g
SQ C D 2 = dx dx Tr (∂− A+ ) + Ψ̄ i ∂ − m − √ γ− A+ Ψ , (12.3)
+ 2
2 N
ψL
where Ψ = , with ψL and ψ̄R are Weyl fermions, the trace is over the
ψ̄R
color indices which are not written explicitly. To simplify the analysis we restrict
ourselves to the case of a single flavor.
The corresponding equations of motion take the form,
i∂− ψL = mψR , (i∂+ + gA+ )ψR = mψL , ∂− A+ = gψR† T a ψR .

2 a
(12.4)
One can then express ψL and A+ in terms of ψR only,

−im +L − −
ψL (x− , x+ ) = dy (x − y − )ψR (x+ , y − )
2 −L
+L
g
Aa+ (x− , x+ ) = dy − |x− − y − |ψR† T a ψR (x+ , y − ), (12.5)
2 −L
where (x− − y − ) is +1 for positive argument and −1 for negative.
2 The application of the discrete light-front quantization to two-dimensional QCD was done
in [128] and in [127].

The light-cone momentum and energy are given by,

+L
+
P = dx− ψR† i∂− ψR (x+ , x− )
−L

−im2 +L − + L − † −
P− = dx dy ψR (x )(x− − y − )ψR (y − )
2 −L −L

g2 + L − + L − † − a
− dx dy ψR (x )T ψR (x− )|x− − y − |ψR† (y − )T a ψR (y − ).
2 −L −L
(12.6)
When one substitutes into these expressions the expansion of the fields in anti-
commuting modes, subjected to anti-periodic boundary conditions, one gets,
2π
P+ = n(b†n bn + d†n dn ), (12.7)
L 1 3
n = 2 , 2 ,...
where,
1 n πx − n πx −
ψR (x− ) = √ bn e−i L + d†n ei L . (12.8)
2L n = 1 , 3 ,...
2 2
The creation and annihilation operators b†n , d†n , bn , dn are all taken to be in the
fundamental representation of SU (N ) and obey the usual algebra,
{b†n , bm } = δn m {d†n , dm } = δn m . (12.9)

2π
Since the eigenvalues of the momentum are proportional to L it is natural to
define a dimensionless momentum,
L +
K= P . (12.10)
2π
Similarly one defines a dimensionless Hamiltonian,
?
2π 1 − λ̂2 − 1
H≡ P λ̂ ≡ 2 , (12.11)
L m2 1 + πm
g2
where we introduce the dimensionless coupling λ̂. The rationale behind this
parameterization is that the spectrum and wave function depend, apart from
an overall mass scale, only on the ratio of mg . The Hamiltonian H is decomposed
into a free kinetic term H0 and the potential V ,
H = (1 − λ̂2 )H0 + λ̂2 V, (12.12)
where,
1 †
H0 = (b bn + d†n dn ), (12.13)
n n
n = 12 , 32 ,...

and,
∞
1 a 1
V = J (k) 2 J a (−k), k=0 (12.14)
π k
k =−∞
with the currents given by,

∞

J a (k) = [θ(n)b†n + θ(−n)dn ]T a [θ(n − k)b†n −k + θ(k − n)dk −n ]. (12.15)
k =−∞
We already have the expression for the potential (11.16).

Note that there is no contribution from k = 0, since we apply the Hamiltonian
P − only on singlet states, and j a (0) on these vanishes. Normal ordering of the
potential gives,
λ̂2 CF In †
V =: V : + (b bn + d†n dn ), (12.16)
π n n
n = 12 , 32 ,...
where CF = N2N−1 is the second Casimir operator in the fundamental representa-

2
n +1/2
tion, and In is the self-induced inertia In = − 2n
1
+ m =1 m12 . The self-induced
inertia terms cancel the infrared singularity in the interaction term in the con-
tinuum limit. : V : involves a sum of eight quartic terms in the fermionic creation
and annihilation operators. For instance one such term is,
1 c1 c3 1 c4 c3
− N δc 2 δc 4 − δcc41 δcc23 δn −n +n −n ,0 b† n 4 b† n 3 bn 2 ,c 2 bn 1 ,c 1 , (12.17)
4N (n4 − n2 )2 4 2 3 1
where ci are the color indices that have been suppressed before and there is an
implicit summation over the half integers ni such that the momentum is con-
served. Now since P − and P + (or H and K) commute they can be diagonalized
simultaneously. One fixes the value of K = 1, 2, 3 . . . and the corresponding Fock
space is finite dimensional. One then diagonalizes H in the restricted subspace
of gauge singlets such that the masses are given by,
2m2
M 2 = 2P + P − = KH(K). (12.18)
1 − λ̂2
Notice that the dependence of the invariant masses on L the size of the space
drops out.
In Fig. 12.1 the DLCQ spectrum of low-lying mesons is drawn as a function
of m/g for N = 2,3,4 and compared with the t’ Hooft large N calculation. A
comparison with lattice calculation is presented in Fig. 12.2.
In performing these calculations it was found that, except for very small quark
masses, there is a quick convergence of the numerics. This is a manifestation of
the fact that the lowest Fock states dominate the hadronic state. It was found
out that typically the momentum carried by sea quarks is less than one percent.

SU(2)
SU(3)
SU(4)
Hamer: SU (2) Lattice
M/g
m/g
Fig. 12.1. Comparison of the DLCQ meson spectra for N = 2,3,4 and the
spectrum derived from lattice calculations [127].
LARGE N
SU(4)
SU(3)
SU(2)
(2π/N)1/2 M/g
(2π /N )1/2 m/g
Fig. 12.2. Comparison of the DLCQ meson spectra for N = 2,3,4 and the
’t Hooft large N spectrum [127].

Several further properties were extracted from the DLCQ spectrum and wave
functions:
r The scaling of the lightest mesonic and baryonic masses with N . It was found
that there is fair agreement with the result deduced from bosonization for small
m
g , namely, that

Mm eson π
= 2 sin . (12.19)
Mbaryon 2(2N − 1)
The results “measured” were found to be 1, .62(5), .46(4) for N = 2, 3, 4 com-
paring with bosonization result 1, .618, .445. In the large N limit this result
implies that the baryon mass is proportional to N times the mass of the meson.
r The mesonic form factors were shown to be in accordance with analytical work.
r The “deuteron”, a loosely bound state of two nucleons, was shown to be stable,
in QCD2 with two colors and two flavors.
r The “anti Pauli-blocking” effect, for which the sea quarks with the same flavor
as that of the majority of the valence ones, are not suppressed in spite of their
fermionic nature.
12.3 The spectrum of QCD2 with adjoint fermions

Our starting point is the action of two-dimensional SU (N ) YM theory coupled
to Majorana fermions in the adjoint representation.3 The latter is expressed in
terms of a traceless Hermitian matrix ψij . The action reads,

1
Sadj = d xTr iψ γ γ Dμ ψ − mψ γ ψ − 2 Fμν F
2 T 0 μ T 0 μν
4g

−
√ 1
= dx dx Tr i(ψ∂+ ψ + ψ̄∂− ψ̄) − i 2mψ̄ψ + 2 (∂− A+ ) + A+ J ,
+ 2 +
2g
(12.20)
where we have parameterized the Majorana fermions as follows,
1 ψij
ψij = √ Jij+ = 2ψik ψk j , (12.21)
2 ψ̄ij
where ψ and ψ̄ are Weyl Majorana spinors written as N × N traceless Hermitian

matrices. In the second line we have imposed the light-cone gauge A− = A+ = 0
and used γ 0 = σ2 and γ 1 = iσ1 . Note that the action does not include time
(x+ ) derivatives of A+ and of ψ̄ and hence both of them are non-dynamical. The
equal time (x+ ) anti-commutation relation for the dynamical Majorana fermions
is given by,
1 1
{ψij (x− ), ψk l (y − )} = δ(x− − y − ) δil δj k − δij δk l . (12.22)
2 N
3 Two-dimensional QCD with adjoint fermions was analyzed in several papers. Here we follow
[38], [72], [147].

In analogy to the expression of P + and P − given for the fundamental fermions

in Section 12.2 and for the bosonized case in Section 11.3, we have now,

P + = dx− Tr[iψ∂− ψ]

− − −im2 1 1 2 + 1 +
P = dx Tr ψ ψ− g J 2 J . (12.23)
2 ∂− 2 ∂−
Denoting by Φ the physical states of the system, that obey the zero charge
condition,

dx− J + |Φ> = 0, (12.24)
which is simultaneously an eigenstate of both P + and P − since [P + , P − ] = 0,

the spectrum is then determined as usual in the light-cone quantization via,
2P + P − |Φ> = M 2 |Φ> . (12.25)
Next we introduce the mode expansion and transform the expressions from
the configuration space to the space of momenta. In Section 12.2 we have done
this directly in the discretized formalism. Here for completeness we first consider
a continuous momentum and then perform the discretization.
The mode expansion reads,
∞
1 + − + −
−
ψij (x ) = √ dk + [bij (k + )e−ik x + b†ij (k + )eik x ], (12.26)
2 π 0
and the non-trivial part of the algebra of the creation and annihilation operators
is given by,
1 1
{bij (k + ), bk l (q + )} = δ(k + − q + ) δil δj k − δij δk l . (12.27)
2 N
From here on we will omit the + of k + and denote it as k. Plugging the mode
expansion into (12.23) we get,
∞
P+ = dk kb†ij (k)bij (k), (12.28)
0
and

m2 ∞ dk † g2 N ∞
P =−
b (k)bij (k) + dkC(k)b†ij (k)bij (k)
2 0 k ij π 0

g2 N ∞
+ dk1 dk2 dk3 dk4 [A(ki )δ(k1 + k2 − k3 − k4 )b†k j (k3 )b†j i (k4 )bk l (k1 )bli (k2 )
2π 0
+ B(ki )δ(k1 + k2 + k3 − k4 )(b†k j (k4 )bk l (k1 )bli (k2 )bij (k3 )
− b†k j (k1 )b†j l (k2 )b†li (k3 )bk i (k4 ))], (12.29)

where
1 1
A(ki ) = − ,
(k4 − k2 )2 (k1 + k2 )2
1 1
B(ki ) = − ,
(k2 − k3 )2 (k1 + k2 )2
k
k
C(k) = dp . (12.30)
0 (p − k)2
From these expressions of P + and P − it is obvious that the vacuum is annihilated

by both P + and P − ,
P + |0> = 0 P − |0> = 0. (12.31)
The bosonic and fermionic states of the system take the following form,
∞ P+
2j

|Φb (p )> =
+
dk1 . . . dk2j δ ki − P +
j=1 o i=1
f2j (k1 , k2 , . . . k2j )N Tr[b† (k1 ) . . . b† (k2j )]|0>,

−j
∞ P+
2j

|Φf (p )> =
+
dk1 . . . dk2j +1 δ ki − P +
j=1 o i=1
−j
f2j (k1 , k2 , . . . k2j )N Tr[b (k1 ) . . . b† (k2j +1 )]|0>, (12.32)
†
where the wave functions obey the cyclicity relation due to the fermionic nature
of the creation and annihilation operators,
fi (k2 , k3 , . . . , ki , k1 ) = (−1)i−1 fi (k1 , k2 , . . . ki ). (12.33)
Unlike the case of fundamental fermions, pairs of adjoint fermions are not sup-
pressed by additional factor of N1 and hence the eigenstates are generated by
applying operators on the vacuum with a mixture of different numbers of cre-
ation operators. This renders the extraction of the spectrum for adjoint fermions
much harder to determine than that of the fundamental ones. These states are
obviously eigenstates of P + . We will have to ensure that they are also eigenstates
of P − . Following the same procedure as for ’t Hooft’s model of Chapter 10 and
of Chapter 11 one derives a set of equations for the wavefunctions fi by applying
(12.25) on the bosonic and fermionic eigenstates which take the form,
m2
M 2 fi (x1 , x2 , . . . xi ) = fi (x1 , x2 , . . . xi )
x1
x 1 +x 2
g2 N
+ dyfi (y, x1 + x2 − y, x3 , . . . xi )
π(x1 + x2 )2 0

g 2 N x 1 +x 2 dy
[fi (x1 , x2 , . . . xi ) − fi (y, x1 + x2 − y, x3 , . . . xi )]
π 0 (xi − y)2

x1 x 1 −y
g2 N 1 1
dy dzfi+ 2 (y, z, x1 − y − z, x2 , . . . xi ) −
π 0 0 (y + z)2 (x1 − y)2

1 1
− ± cyclic, (12.34)
(x1 + x2 )2 (x2 + x3 )2
k+
where xi = P i+ and the last term of the equation stands for cyclic permutations
of (x1 , x2 , . . . , xi ) which for odd i comes with a + sign and for even i with
alternating signs. Similar to what happens in the ’t Hooft model, the equation
does not have an ambiguity once we incorporate a principal value prescription
to the Coulomb double pole since at x1 = y the numerator also vanishes.
At this point we implement the idea of discretizing the light-cone momenta in
the following way,
1
2
K
n
x→ dx → , (12.35)
K 0 K
o dd n > 0
where n is an odd positive integer and K → ∞ is the continuum limit. The

i
constraint j = 1 xj = 1 eliminates all states with over K partons, where a par-
ton is a state created from the vacuum by a single creation operator. In this
way the discretized eigenvalue problem becomes finite dimensional. With this
discretization the Fourier transform (12.26) translates into a sum,
1 + − + −
ψij (x− ) = √ [bij (n)e−ik x + b†ij (n)eik x ]. (12.36)
2 π o dd n > 0
Similar to (12.8), the creation and annihilation operators of (12.27) also take
discretized values, and obviously the Dirac delta function in (12.27) is replaced
by a Kronecker delta function. The eigenvalue problem now reads,
2
g N
2P + P − = K T + m2 V , (12.37)
π
where the mass term is given by,
1 †
V = b (n)bij (n), (12.38)
n
n ij
and,
n −2
T =4 n b†ij (n)bij (n) m
1
(n −m ) 2 +

2
N m {δn 1 +n 2 ,n 3 +n 4
1
(n 4 −n 2 ) 2 − 1
(n 1 +n 2 ) 2 b†k j (n)b†j i (n)bk l (n)bli (n)

+ δn 1 +n 2 +n 3 ,n 4 1
(n 3 +n 2 ) 2 − 1
(n 1 +n 2 ) 2
b†k j (n4 )bk l (n1 )bli (n2 )bij (n3 ) − b†k j (n1 )b†j l (n2 )b†li (n3 )bk i (n4 )}, (12.39)
where all the summations are over positive odd integers.

20
15
M 10
5 10
<N>
Fig. 12.3. The spectrum of fermionic states for K = 25, m = 0 [38].
One chooses a basis of states normalized to 1 in the large N limit,
1
i
√ Tr[b† (n1 ) . . . b† (ni )]|0> nj = K. (12.40)
N i/2 s j =1
The states are defined by ordered partitions of K into i positive odd integers,
modulo cyclic permutations. If (n1 , n2 , . . . , ni ) is taken into itself by s out of i
possible cyclic permutations, then the corresponding state receives a normaliza-
tion factor √1s . Otherwise s = 1. For even i, however, all partitions of K where
i/s is odd do not give rise to states.
Using the discretized Hamiltonian and the basis of states (12.40) one can
diagonalize the Hamiltonian and compute the spectrum for a range of values
of K and then extrapolate the results to infinite K, the continuum limit. One
can extract certain properties of the spectrum also from the results at a fixed
large K. In particular the dependence of the spectrum on the mass of the adjoint
quark m is also of interest and the special cases of m = 0 and m2 = g 2 N/π where
the model is supersymmetric.
The fermionic spectrum found by diagonalizing the system with K = 25 for
2
the massless case and for m2 = g πN is described in Figs. 12.3 and 12.4 in the
form of the mass of the bound state as a function of the expectation value of
the parton number. The bosonic spectrum using K = 24 for the two masses is
drawn in Figs. 12.5 and 12.6.

15
10
M
5 10
<N>
Fig. 12.4. The spectrum of fermionic states for K = 25, m2 = g 2 /π [38].
20
15
M
10
5 10
<N>
Fig. 12.5. The spectrum of bosonic states for K = 25, m = 0 [38].
The characteristic features of the spectra are the following:

r The density of states increases rapidly with the mass, and almost all the states
lie within a band bounded by two < N >∼ M lines. The system admits a
Hagedorn behavior,
ρ(m) ∼ mα eβ m , (12.41)

where ρ(m) is the density and from the data it follows that β ∼ 0.7 g 2πN .

15
10
M
5
5 10 15
<N>
Fig. 12.6. The spectrum of bosonic states for K = 25, m2 = g 2 /π [38].
r The mass increases roughly linearly with the average number of partons. Such a
behavior characterizes a system of large N non-relativistic particles connected
into a closed string by harmonic springs.
r For the low-lying states the wave function strongly peaks on states with a
definite number of partons. For instance, for K = 25 the ground state has a
probability of 0.9993 of consisting of 3 partons, and the first excited state has
a probability of 0.99443 of consisting of 5 partons.
r Thus the low-lying states can be well approximated by truncating the diago-
nalization to a single parton number sector. For instance the bosonic ground
state can be derived from a truncation of (12.34) to a two-parton sector which
yields the following equation,

2 2 1 1 2g 2 N 1
φ(x) − φ(y)
M φ(x) = m φ(x) + + dy , (12.42)
x 1−x π 0 (y − x)2
with φ(x) = f2 (x, 1 − x). Note that this equation is the ’t Hooft equation dis-
cussed in Chapter 10 with the replacement of g 2 → 2g 2 . This difference stems
from the fact that unlike for mesons built from fundamental quarks, here there
are two color flux tubes connecting two partons.
r Due to the fermionic statistics φ(x) = −φ(1 − x) half of the states of the ’t
Hooft model including the ground state are now excluded. In particular for
m = 0 the state φ(x) = 1 which associates with a massless bound state is
missing. The absence of a massless ground state even in the limit of m → 0
can be explained huristically as follows. For m = 0 the mass of the states is
measured in units of the coupling constant g and hence the massless limit can

be achieved in the strong coupling limit g → ∞ for which the action takes the
form,

S = d2 xTr[iΨT γ 0 γ μ ∂μ Ψ + Aμ J μ ]. (12.43)
Now the left and right currents Jij± constitute two independent level N affine
Lie algebras for which we have seen in Chapter 3 the corresponding Virasoro
anomaly is,
k N2 − 1 k
c = c0 − (N 2 − 1) = − (N 2 − 1) , (12.44)
k+N 2 k+N
where c0 is the central charge before gauging and k is the ALA level. Since
k = N it is obvious that c = 0 and hence there is no massless bound state. For
fundamental quarks in the same limit we get, by taking k = 1 and c0 = N ,
that c = 1, which means that for this case there is a massless bound state.

13
The baryonic spectrum of multiflavor QCD2
in the strong coupling limit
We are now going to compute the baryonic spectrum of QCD2 , for which as
it turns out the bosonic formulation is very convenient. The mesonic spectrum
was found earlier, using large N with quark fields as variables in Chapter 10,
as well as using currents as building blocks in Section 11.3. For the baryon
spectrum, however, the large N limit, in terms of fermionic fields, is not the
natural framework to use since in such a picture the baryon is a bound state of
a large number N of constituents. Instead, it will be shown in this chapter that
the bosonized version of QCD2 in the strong coupling limit provides an effective
description of the baryons.1 We will start by deriving the effective action at
the strong coupling limit. It will be argued that for the purpose of extracting
the low-lying baryons, one can in fact use the product scheme instead of the
U (Nc × Nc ) scheme, with the former being more suitable for our purposes. Once
the effective action is written down we will search for soliton solutions that carry
a baryon number. It will be shown that for a static configuration the effective
action reduces to a sum of sine-Gordon actions. Using the knowledge acquired
on solitons, in Chapter 5, it will be easy to write down the classical baryonic
configuration. We will then semi-classically quantize these solitons. This problem
will be mapped into a quantum mechanical model on a CP (N f −1) manifold.
The energy and charges of the quantized soliton can be derived and thus the
spectrum of the baryons is determined. We then analyze the quark flavor content
of the baryons and discuss multi-baryon states. Finally, we include meson-baryon
scattering, this time also for the case of any coupling.
13.1 The strong coupling limit

It turns out that the mass term plays an essential role in the determination of
classical soliton solutions in 1+1 space-time dimensions. It is therefore required
to switch on this term before deducing the low energy effective action. As was
explained in Chapter 6, we know how to do this rigorously only in the scheme
of U (NF NC ). It will turn out, however, that the product scheme can be used for
the low mass states in the strong coupling limit.
1 The spectrum of baryons of two-dimensional QCD extracted in the strong coupling limit was
derived in [75].

238 The baryonic spectrum of multiflavor QCD2 in the strong coupling limit
Our starting point is the last equation of Chapter 9. In the strong coupling limit
ec
mq → ∞, the fields in h̃ which contribute to H̃ will become infinitely heavy. The
sector g̃l ⊂ S US U(N(NF NC )C ) , however, will not acquire mass from the gauge interaction
term. Since we are interested only in the light particles we can, in the strong
coupling limit, ignore the √ heavy fields, if we first normal order the heavy fields
e c√ N F
at the mass scale μ̃ = 2π
. Using the relation, for a given operator O,
Δ
μ̃
Nμ̃ O = Nm̃ O, (13.1)
m̃
to perform the change in the scale of normal ordering, and then substituting
hab = δba , we get for the low energy effective action,

1
Seff [u] = S[g̃] + S[l] + d2 x∂μ φ∂ μ φ
2

−i N 4 π φ +i N 4 π φ † †
+ cmq μ̃Nμ̃ d2 xTr(e CNF g̃l + e CNF l g̃ ). (13.2)
We can now replace the two mass scales mq and μ̃ by a single scale, by normal
ordering at a certain m so the final form of the effective action becomes,

1
Seff [u] = S[g̃] + S[l] + d2 x∂μ φ∂ μ φ
2
m2 2 −i N 4 π
NF φ +i N 4 π
CNF
φ ˜† †
+ Nm d xTr(e C g̃l + e l g ), (13.3)
NC
with m given by,
√ ΔC
1 + 1Δ
C
ec NF
m = NC cmq √ , (13.4)
2π
N C2 −1
here ΔC , the dimension of h̃, is N C (N C +N F ) . For the l = 1 sector, defining g =

4π
−i φ
g̃e NCNF
⊂ U (NF ) one gets the effective action,

Seff [g ] = NC S[g ] + m2 Nm d2 xTrF (g + g † ). (13.5)
Thus, the low energy effective action in the l = 1 sector coincides with the result
of the “naive” approach of the product scheme.
In the strong coupling limit ec /mq → ∞ the low energy effective action reads,2

S[g] = NC S[g] + m2 Nm d2 x(Trg + Trg † ), (13.6)
with g in U (NF ). Note that the analog of our strong coupling to the case of 3+1
space-time, would be that of light current quarks compared to the QCD scale
ΛQ C D .
2 From here on we omit the prime from g so we denote g ∈ U (N F ).

13.2 Classical soliton solutions 239
13.2 Classical soliton solutions

We now look for static solutions of the classical action. For a static field config-
uration, the WZ term does not contribute. One way to see this is by noting that
the variation of the WZ term can be written as,

δW Z ∝ d2 xεij Tr(δg)g † (∂i g)(∂j g † ), (13.7)
and for g that has only spatial dependence δW Z = 0. Without loss of generality
we may take, for the lowest energy, a diagonal g(x),

4π 4π
−i NC ϕ1 −i NC ϕN F
g(x) = e ,...,e . (13.8)
For this ansatz and with a redefinition of the constant term, the action density
reduces to,
5

NF
1 dϕi
2
4π
S̃d [g] = − dx − 2m cos
2
ϕi − 1 . (13.9)
i= 1
2 dx NC
This is a sum of decoupled standard sine-Gordon actions for each ϕi . The

well-known solutions of the associated equations of motion are,
5
4NC ( N8 π m x)
ϕi (x) = arctg e C , (13.10)
π
with the corresponding classical energy,
5
2NC
Ei = 4m , i = 1, . . . , NF . (13.11)
π
Clearly the minimum energy configuration for this class is when only one of the
ϕi is nonzero, for example,

4π
−i NC ϕ(x)
g◦ (x) = Diag 1, 1, . . . , e (13.12)
Conserved charges, corresponding to the vector current, can be computed using

the definition,

A 1
Q [g(x)] = dxTr(J0 T A ), (13.13)
2
where 12 T A are the SU (NF ) generators and the U (1) baryon number is generated
by the unit matrix. This follows from Jμ = JμA T A , and in the fermionic basis
JμA = ψ̄γμ 12 T A ψ.
In particular, for eqn. (13.10), we get charges different from zero only for QB
and QY corresponding to baryon number and “hypercharge”, respectively,
?
◦ ◦ 1 2(NF − 1)
QB = NC , QY = − NC , (13.14)
2 NF

these charges are determined solely by the boundary values of ϕ(x), which are,
5 5
4π 4π
ϕ(∞) = 2π, ϕ(−∞) = 0. (13.15)
NC NC
Under a general UV (NF ) global transformation g◦ (x) → g̃◦ (x) = Ag◦ (x)A−1
the energy of the soliton is obviously unchanged, but charges other than QB and
QY will be turned on. Let us introduce a parametrization of A that will be useful
later,
⎛ ⎞
z1
⎜ .. ⎟
⎜ Aij . ⎟
⎜ ⎟
A=⎜ ⎜ . ⎟. (13.16)
.. ⎟
⎜ ⎟
⎝ z(N F −1) ⎠
Y1 . . . . . . Y(N F −1) zN F
Now,

4π
−i ϕ
g̃◦ = 1 + (e NC
− 1)z, (13.17)
N F
where (z)α β = zα zβ∗ , and from unitarity α =i zα zα∗ = 1. The charges with g̃◦ (x)
are,
1
(Q̃◦ )A = NC Tr(T A z). (13.18)
2
Only the baryon number is unchanged. The discussion of the possible U (NF )
representations cannot be done yet, since we are dealing so far with a classical
system. We will return to the question of possible representations after quantizing
the system.
13.3 Semi-classical quantization and the baryons

The next step in the semi-classical analysis is to consider configurations of the
form,
g(x, t) = A(t)g◦ (x)A−1 (t), A(t) ∈ U (NF ), (13.19)
and to derive the effective action for A(t).3 Quantization of this action cor-
responds to doing the functional integral over g(x, t) of the above form. The
effective action for A(t) is derived by substituting g(x, t) = A(t)g◦ (x)A−1 (t) in
the original action. Here we use the following property of the WZ action,

S AgB −1 = S AB −1 + S g, Ãμ , (13.20)
3 The semi-classical quantization makes use of the Polyakov–Wiegmann formula [179].

where S[g] is the WZW action and S[g,Ã] is given by (9.25), respectively, with
the gauge field Ãμ given as,
iÃ+ = A−1 ∂+ A, iÃ− = B −1 ∂− B; A, B ∈ U (NF ). (13.21)
Using the above formula for A = B, noting that S(1) = 0, and taking A = A(t),
Ȧ
∂+ A = ∂ − A = √ , (13.22)
2
we get,
< =
NC
S̃ A(t)g◦ (x)A−1 (t) − S̃[g◦ ] = d2 xTr [A−1 Ȧ, g◦ ][A−1 Ȧ, g◦† ]
8π
NC < =
2 −1 †
+ d xTr (A Ȧ)(g◦ ∂1 g◦ ) . (13.23)
2π
This action is invariant under global U (NF ) transformations A → U A, where
U ∈ G = U (NF ). This corresponds to the invariance of the original action under
g → U gU −1 . On top of this it is also invariant under the local changes A(t) →
A(t)V (t), where V (t) ∈ H = SU (NF − 1) × UB (1) × UY (1), with the last two
U (1) factors corresponding to baryon number and hypercharge, respectively. This
subgroup H of G is nothing but the invariance group of g◦ (x). In terms of g◦ (x)
and A(t) the charges associated with the global U (NF ) symmetry, eqn. (13.13),
have the form,
<
NC =
B
Q =i dxTr T B A g◦† ∂1 g◦ − g◦ ∂1 g◦† + g◦ , A−1 Ȧ, g◦† A−1 .
8π
(13.24)
The effective action, eqn. (13.23), is an action for the coordinates describing the
coset space,
G/H = SU (NF ) × UB (1)/SU (NF − 1) × UY (1) × UB (1)
= SU (NF )/SU (NF − 1) × UY (1) = CP N · (13.25)
To see this explicitly we define the Lie algebra valued variables q A through

A−1 Ȧ = i T A q̇ A . In terms of these variables (13.23) takes the form (the part
that depends on q A ),
⎡ ? ⎤
2(N F −1)

1 2(N − 1)
Sq = dt ⎣ q̇ Y ⎦
F
(q̇ A )2 − NC
2M NF
A=1
∞
5 √
1 NC 4π 2 NC 3/2
= (1 − cos ϕ)dx = ( ) . (13.26)
2M 2π −∞ NC m π
The sum is over those q A which correspond to the G/H generators and q Y
is associated with the hypercharge generator. Although the q A seem to be a
“natural” choice of variables for the action (13.23), which depends only on the
combination A−1 Ȧ, they are not a convenient choice of variables. The reason for
that is the explicit dependence of the charges (13.24) on A−1 (t) and A(t) as well
as on A−1 Ȧ(t).

Instead we found that a convenient parametrization is that of (13.16). One

can rewrite the action (13.23), as well as the charges (13.24), in terms of the
N F
z1 ,. . . ,zN F variables, which however are subject to the constraint α =1 zα zα∗ = 1.
Thus,

S̃ A(t)g◦ A−1 (t) − S̃[g◦ ] = S[zα (t), ϕ(x)], (13.27)
where,

S [zα (t), ϕ(x)] = NC
2π d2 x{(1 − cos 4π
ϕ)[żα∗ żα
NC
−(zγ∗ żγ )(żβ∗ zβ )] − i N4πC ϕ zα∗ żα }. (13.28)
We can do the integral over x and rewrite (13.28) as,

1 NC
S[zα (t)] = dt[żα∗ żα − (zγ∗ żγ )(żβ∗ zβ )] − i dt(zα∗ żα − żα∗ zα ), (13.29)
2M 2
where 1/M is defined in eqn. (13.26). The first term in (13.29) is the usual
CP(N F −1) quantum mechanical action, while the second term is a modification
due to the WZ term.
Similarly we express the U (NF ) charges in terms of the z variables, using
eqn. (13.24),
1
QC = TβCα Qα β
2
i
Qα β = NC zα zβ∗ + [zα zβ∗ (zγ∗ żγ − żγ∗ zγ ) + zα żβ∗ − zβ∗ żα ]. (13.30)
2M
Of course the symmetries of S[z] are the global U (NF ) group under which,
zα → zα = Uα β zβ , U ∈ U (NF ), (13.31)
and a local U (1) subgroup of H under which,
zα → zα = eiδ (t) zα . (13.32)
As a consequence of the gauge invariance one can rewrite the action in a covariant
form,

1 †
S[zα ] = dtTr(Dz) Dz + iNC dtTrż † z, (13.33)
2M
where,
(Dz)α = żα + zα (żβ zβ ). (13.34)
Constructing Noether charges of the U (NF ) global invariance of (13.31) out of
the action (13.33 leads to expressions identical with (13.30)). Note that in eqn.
(13.34) we can view żβ zβ = ia(t) as a composite U (1) gauge potential.
Now let us count the degrees of freedom. The local U (1) symmetry allows

us to take one of the zs to be real, and the constraint α zα zα∗ = 1 removes
one more degree of freedom, so altogether we are left with 2NF − 2 = 2(NF − 1)

physical degrees of freedom. This is exactly the dimension of the coset space
S U (N F )
S U (N F −1)×U (1) . The corresponding phase space should have a real dimension
of 4(NF − 1). Naively, however, we have a phase space of 4NF dimensions and,
therefore, we expect four constraints.
There are several methods of quantizing systems with constraints. Here we
choose to eliminate the redundancy in the z variables and then invoke the canon-
ical quantization procedure.4
But before following these lines let us briefly describe another method, through
the use of Dirac’s brackets. We outline the classical case. The quantum case is
obtained by replacing { , } with i[ , ].
The first step in this prescription is to add to the Lagrangian a term of

the form λ( α zα zα∗ − 1), in which case the conjugate momentum πλ of the
Lagrange multiplier vanishes. By requiring that this condition be preserved in

time one gets the secondary constraint Φ1 = ( α zα zα∗ − 1) = 0. Further impos-
ing Φ̇1 = {Φ1 , H}P = 0, where { }P denotes a Poisson bracket, one finds another
second-class constraint Φ2 = Π · z + z † · Π† . In addition there is a first-class con-
straint Φ3 = Π · z − z † · Π† , which corresponds to the local U (1) invariance of the
model. Fixing this symmetry one gets an additional constraint Φ4 . For instance
∗
one can choose the unitary gauge Φ4 = zN F − zN F
. The next step is to compute
the constraint matrix {Φi , Φj }P = cij . In the constrained theory, the brack-
ets between F and G are replaced by the Dirac brackets of those operators,
given by
{F, G}D = {F, G}P − {F, Φi }P (c−1
ij ){Φj , G}P , (13.35)
where c−1 ij is the inverse of the constraint matrix. Imposing the constraints as
operator relations it is easy to see that zN F , ΠN F and their complex conjugates
can be eliminated. The brackets for the rest of the fields coincide with the results
we derive below, when eliminating the constraints explicitly.
We now describe in some detail the quantization of the system using uncon-
strained variables. We want to choose a set of new variables so that the constraint
N F ∗
α = 1 zα zα = 1 is automatically fulfilled. There is a standard choice of such vari-
ables, namely (f or i = 1, . . . , NF − 1),
ki ki∗ eiχ
zi = √ , zi∗ = √ , zN F = √
1+X 1+X 1+X
F −1
N
X= ki∗ ki . (13.36)
i= 1
The ki , ki∗ and χ are 2NF − 1 real variables with no constraints on them. The
phase space will now have dimension 2(2NF − 1) and we still have two extra
4 The quantization of the system including its constraint was done in [75]. For an alternative
procedure of quantization in the presence of constraints see [181].

constraints. After some straightforward algebra we can write,

S[k, k , χ] = dtL(k, k∗ , χ)
∗
1 ∗ NC ki∗ k̇i − k̇i∗ ki

L(k, k∗ , χ) = k̇i hij k̇j − i
2M 2 1+X
@ A
1 X 2 i ki∗ k̇i − k̇i∗ ki NC
+ χ̇ + χ̇ + , (13.37)
2M (1 + X)2 2M (1 + X)2 1+X
where,
δij ki kj∗
hij = − . (13.38)
1+X (1 + X)2
The local U (1) transformations of the z variables transcribe into the transfor-
mations,
δχ = (t), δki = i(t)ki , δki∗ = −i(t)ki∗ , (13.39)
and δL = −NC ˙ just as in terms of the z variables. This local U (1) symmetry
can be made manifest by defining the covariant derivatives,
Dki = k̇i − iχ̇ki Dki∗ = k̇i∗ + iχ̇ki∗ . (13.40)
The Lagrangian can then be recast in a manifestly gauge-invariant form,
1 NC ki∗ Dki − (Dki∗ )ki
L(k, k∗ , x) = Dki∗ hij Dkj − i + NC χ̇. (13.41)
2M 2 1+X
Although one can now fix the gauge, for instance χ̇ = 0, we will continue to work
with (13.41). The conjugate momenta are given by,
∂L 1 NC ki∗
πi = = Dkj∗ hj i − i
∂ k̇i 2M 2 1+X
∗ ∂L 1 NC ki
πi = = hij Dkj + i
∂ k̇i∗ 2M 2 1+X
∂L i 1
πχ = = (k ∗ hij Dkj − Dki∗ hij kj ) + NC . (13.42)
∂ χ̇ 2M i 1+X
Since hij is invertible we can solve for Dki∗ , Dki in term of the phase space
variables,
NC kj∗
∗
Dki = 2M πj + i h−1
2 1 + X ji

NC kj
Dki = 2M h−1
ij π ∗
j − i , (13.43)
2 1+X
where,
h−1 ∗
ij = (1 + X)(δij + ki kj ). (13.44)
Also,
πχ = i(ki∗ πi∗ − πi ki ) + NC , (13.45)

giving the constraint equation,

ψ = πχ − i(ki∗ πi∗ − πi ki ) − NC = 0. (13.46)
The canonical Hamiltonian is given by,
Hc = πi k̇i + πi∗ k̇i∗ + πχ χ̇ − L

NC ki∗ −1 ∗ NC kj
= 2M πi + i h πj − i
2(1 + X) ij 2(1 + X)
+ χ̇[πχ − i(πi∗ ki∗ − πi ki ) − NC ], (13.47)
and this can be further simplified to,

Hc = 2M (1 + X) πi πi∗ + (πi ki )(πi∗ ki∗ )

NC 1 NC2 X
−i (πi ki − π∗i ki∗ ) + + χ̇ψ. (13.48)
2 4 (1 + X)
Here Hc is obtained explicitly in terms of the canonical variables ki , ki∗ , πi , πi∗ .
The χ̇ψ term indicates that χ̇ also behaves as a Lagrange multiplier since, fol-
lowing the Dirac procedure, we should define,
HT = Hc + λ(t)ψ, (13.49)
where λ is a priori an arbitrary function of t. We could absorb the χ̇ in λ.
Quantization of this Hamiltonian is now essentially straightforward. Let us
first consider the symmetry generators Qα β , which in terms of the new canonical
variables take the form,
Qij = i(ki πj − πi∗ kj∗ )

−iχ NC ki ∗
Qi,N F = e − i(πi + ki πj kj )
2

NC ki∗
QN F ,i = eiχ + i(πi + kj∗ πj∗ ki∗ ) = Q∗i,N F
2
QN F ,N F = NC − i(πi ki − πi∗ ki∗ ). (13.50)
We will now show that the HT can be expressed in terms of the second Casimir
operator of the SU (NF ) group.
The second U (NF ) Casimir operator is related to charge matrix elements Qα β
as,
1
QA QA = Qα β Qβ α . (13.51)
2
A straightforward substitution gives,
1
Qα β Qβ α = (1 + X)[πi∗ πi + πi ki πj∗ kj∗
2
NC 1 X
−i (πi ki − πi∗ ki∗ )] + NC2 1 + . (13.52)
2 2 2

Therefore, the Hamiltonian is,

NC2
HT = 2M Q Q −
A A
+ λ(t)ψ. (13.53)
2
Denoting the SU (NF ) second Casimir operator by C2 , and using QA QA = C2 +
1 2
2N F (QB ) we get (also applying the constraint ψ = 0),

2 (NF − 1)
HT = 2M C2 − NC . (13.54)
2NF
The fact that HT is, up to a constant, the second Casimir operator, is another
way to show that the charges Qα β are conserved. These conserved charges will
generate symmetry transformations via,
δki = i[Tr(Q), ki ], δki∗ = i[T r(Q), ki∗ )]
δχ = i[Tr(Q), χ], (13.55)
and similar equations for the momenta πi , πi∗ , πχ . Here ij = 12 A TijA is the matrix
of parameters. The transformation laws are derived using the constraint equation
ψ = 0 after performing the commutator calculations. Notice that Qij and QN F ,N F
are linear in coordinates and momenta and therefore the SU (NF − 1) × UY (1)
transformations they generate are linear. The QN F ,i and Qi,N F charges, on the
other hand, have cubic terms as well (quadratic in coordinates), so that the coset-
space transformations of S U (NSFU−1)×U
(N F )
(1) are non-linear. This is a well-known
property of CPn models. Substitution of Qα β in eqn (13.55) gives,
δkl = i[j i ki δj l + eiχ iN F δil − e−iχ N F i ki kl − N F N F kl ], (13.56)
where we use [k, π] = i.
Inversely, starting with these transformation laws it is easy to verify the invari-
ance of the action. The standard Noether procedure then gives the charges Qα β
in terms of the coordinates and velocities, which (not suprisingly) coincide with
those given in eqn. (13.50). One could also deduce these transformation laws by
making the change of variables zα , zα∗ → ki , ki∗ , χ in (13.30) directly.
One can verify that,
[QA , QB ] = if A B C QC , (13.57)
where f A B C are the structure constants of the U (NF ) group.
Do we have further restrictions on the physical states? We shall see now that
in fact we do have. Remember that our Lagrangian (13.41) includes an auxiliary
gauge field A◦ ≡ χ̇ and thus has to obey the associated Gauss law,
∂L ∂L
= = πχ = NC − i(πi ki − πi∗ ki∗ ) = 0. (13.58)
∂A◦ ∂ χ̇
Since πχ is a linear combination of QB and QY , and the first is constrained to
be QB = NC , the QY is restricted as well. More specifically, QY = Q̄Y , with,
?
1 2
Q̄Y = NC . (13.59)
2 (NF − 1)NF

13.4 The baryonic spectrum 247
13.4 The baryonic spectrum

The masses of the baryons (13.11) and (13.54), and the two constraints on the
multiplets of the physical
states, namely QB = NC and that the multiplets con-
tain QY = Q̄Y = 12 (N F −1)N
2
F
NC , are the main results of the last section. All
states of the multiplet with QY = Q̄Y will be generated from the state QY = Q̄Y
by SU (NF ) transformations as in (13.19). Using the above constraints we can
investigate now what possible representations will appear in the low energy
baryon sector. Considering states with quarks only (no anti-quarks), the require-
ment of QB = NC implies that only representations described by Young tableaux
with NC boxes appear. The extra constraint QY = Q̄Y implies that all NC quarks
are from SU (NF − 1), not involving the NF th. These are automatically obeyed
in the totally symmetric representation of NC boxes. In fact, this is the only
representation possible for flavor space, since the states have to be constructed
N F n i
out of the components of one complex vector z as i=1 zi with i ni = NC .
See also more detailed discussion in the next section. For another way of deriving
this result see Section 13.7.
Thus for NC = 3, NF = 3 we get only 10 of SU (3). This is understandable,
since there is no physical spin in two dimensions.
What about the masses of the baryons? The total mass of a baryons is given
by the sum of (13.11) and (13.54), namely,
5 ?
√ 3
2NC π 2 (NF − 1)
E = 4m +m 2 C2 − NC . (13.60)
π NC 2NF
For large NC , the classical term behaves like NC , while the quantum correction
like 1. This will be worked out in Section 13.7.
That the total mass goes like NC for large NC , and that the quantum fluctu-
ations are N1C of the classical result, is in accord with general considerations.
13.5 Quark flavor content of the baryons

A measure of the quark content of a given flavor qi in a baryon state |B is given
by5

q̄i qi B = dx gii B − dx gii 0 (13.61)
% &
−i N4 π φ c
= dxzi∗ zi e C −1 (13.62)
B
= const. zi∗ zi B. (13.63)
In order to make contact with the real world, we take here NC = 3 and NF = 3,
getting the baryons in the 10 representation of flavor. Similarly, for SUF (2) there
5 Quark solitons as constituents of hadrons were discussed in [86]. Following that, the flavor
content of the baryons was discussed in [97].

is only the isospin 32 representation. This is what we would expect from naı̈ve
quark model considerations. The total wave function must be antisymmetric.
Baryon is a color singlet, so the wave function is antisymmetric in color and
it must be symmetric in all other degrees of freedom. There is no spin, so the
baryon must be in a totally symmetric representation of the flavor group, a 10
for three flavors. Therefore, strictly speaking there is no state analogous to the
proton. On the other hand, there is a state which is the analog of the Δ+ , namely
the charge 1 state in the 10 representation, z12 z2 .
The 10 is the lowest baryon multiplet in QCD2 . In the following we shall be
dealing with the relative weight of a given flavor in some baryon state. Thus,
q̄q B will henceforth stand for the ratio,
q̄q B
3 4
¯ + s̄s .
ūu + dd
(13.64)
B
For Δ+ ∼ z12 z2 we obtain

(d2 z1 )(d2 z2 )|z3 |2 (z12 z2 )(z12 z2 )∗
1
s̄s Δ+ = = , (13.65)
2 2 2 2 ∗ 6
(d z1 )(d z2 )(z1 z2 )(z1 z2 )
as well as,
1 3 4 1
ūu = , ¯
dd = . (13.66)
Δ+ Δ+
2 3
In evaluating the integral in the numerator in eqn. (13.65) we have used |z3 |2 =
1 − |z1 |2 − |z2 |2 , which follows from the unitarity of the matrix A in (13.19).
Similarly, for Δ+ + ∼ z13 we have,
2 3 4
ūu Δ + + = , ¯ + + = 1 , s̄s + + = 1 .
dd (13.67)
Δ Δ
3 6 6
In the constituent quark picture Δ+ + contains just three u quarks. Both the
d-quark and the s-quark content of the Δ++ come only 3 4from virtual quark
¯ + + , and s̄s + =
pairs. Therefore in the SU (3)-symmetric case s̄s Δ + + = dd Δ Δ
s̄s Δ + + , as expected.
From eqn. (13.67) one can also read the results for Ω− ∼ z33 , by replacing
u ↔ s. In the general case of NF flavors and NC colors, one obtains,
1
(q̄q)sea B = , (13.68)
NC + N F
where (q̄q)sea refers to the non-valence quarks in the baryon B. Moreover, one can
also compute flavor content of valence quarks. Consider a baryon B containing
k quarks of flavor v. The v-flavor content of such a baryon is,
k+1
v̄v B = . (13.69)
NC + N F

13.6 Multibaryons 249
This implies an “equipartition” for valence and sea, each with a content of
1/(NC + NF ). It also follows that the total sea content of NF flavors is,

NF
NF
(q̄q)sea B = , (13.70)
q=1
NC + N F
which goes to zero for fixed NF and NC → ∞, as expected.

It is interesting to compare these results with the Skyrme model in 3+1 dimen-
sions. For the proton,
3+ 1 2 3 43+ 1 11 3+1 7
ūu p = , ¯
dd = , s̄s p = , (13.71)
5 p 30 30
and for the Δ,
3+ 1 7 3+ 1 5
s̄s Δ = , s̄s Ω − = . (13.72)
24 12
The qualitative picture is similar, although the s̄s content in the non-strange
baryons is lower in 1 + 1 dimensions. One may speculate that in 1 + 1 dimensions
the effects of loops are smaller than in 3 + 1 dimensions, since the theory is super-
renormalizable and there are only longitudinal gluons. In the SUF (3)-symmetric
limit the strange quark content of baryons with zero net strangeness is significant,
albeit smaller than that of either of the other two flavors. The situation obviously
is reversed for Ω− .
In the real world the current mass of the strange quark is much larger than
the current masses of u and d quarks. It is natural to expect that this will
have the effect of decreasing the strange quark content from its value in the
SUF (3) symmetry limit. We do not know the exact extent of this effect, but
it is likely that the strange content decreases by a factor which is less than
two. This estimate is based on both explicit model calculations and what we
know from PCAC, namely that the analogous quark bilinear expectation val-
ues in the vacuum are not dramatically different from their SU (3) symmetric
values,
s̄s 0
0.5 ≤ ≤ 1. (13.73)
ūu 0
13.6 Multibaryons
Let us now explore the possibility of having multi-baryons states.6 The pro-
cedure follows similar lines to that of the baryonic spectrum, namely, we look
for classical solution of the equation of motions with baryon number kNC , and
then we semiclassically quantize this. The ansatz for the classical solution of the
6 Multibaryonic states were studied in [102] and [103].

low-lying k-baryon state is taken now to be,

⎛ (N F −k ) ⎞
B CD E
⎜1 ⎟
⎜ ⎟
⎜ .. ⎟
⎜ . ⎟
g0 (k) = ⎜
⎜ k
⎟.
⎟ (13.74)
⎜ B CD E ⎟
⎜ 4π 12
exp[−i( N C ) ϕc ] ⎟
⎝ ⎠
..
.
For the semi-classical quantization we generalize the parametrization given in
(13.16) to,
⎛ ⎞
A = ⎝ Aij ziα ⎠ , (13.75)
where i represents the rows (1, . . . , NF ) and α the columns (NF − k + 1, . . . , NF ).

The effective action in its covariant form (13.33) becomes,

1
S[zα ] = dtTr(Dz)† Dz + iNC dtTrż † z, (13.76)
2M
where now, instead of (13.34),
(Dz)iα = żiα + ziβ (żjβ zj α )eDz. (13.77)
Using the same steps as those which led to (13.54) one finds now the
Hamiltonian,
NC2
H = 2M C2 (NF ) − k(NF − k) + kEc , (13.78)
2NF
with Ec the classical contribution for one baryon, the first term in (13.60).
13.7 States, wave functions and binding energies

It was shown in [102] that the allowed k-baryon states contain (kNC ) boxes in
the Young tableaux representation of the flavour group SU (NF ). Let us recall
that this result followed from the constraint implied by the local invariance,
ziα → eiδ (t) ziα . (13.79)
Performing a variation corresponding to this invariance we find that the action
S changes by

ΔS = (kNC ) δ̇ dt. (13.80)
This means that the Nz number is equal to (kNC ). Thus for any wave function,
written as a polynomial in z and z , the number of zs minus the number of z s
must equal (kNC ). Note that for k = 1 the transformation (13.79) represents also
the NFth flavor number. Thus (13.80) entails that the representation contains a

13.7 States, wave functions and binding energies 251
state with NC boxes of the NF flavor, and therefore must be the totally symmetric
representation.
Now, the effective action (13.76) is invariant under a larger group of local
transformations. In fact, we have extra (k 2 − 1) generators, which correspond to
SU (k) under which (13.76) is locally invariant. This can be exhibited by defining
“local gauge potentials”,
Ãβ α (t) = −(z † ż)β α , (13.81)
so that,
Dz = ż + z Ã. (13.82)
Under the local gauge transformation corresponding to Λ(t), Ã transforms as,
Ã(t) → eiΛ Ãe−iΛ + (∂t eiΛ ) e−iΛ , (13.83)
which implies,
(Dz)iα → (Dz)iβ (e−iΛ )β α , (13.84)
and so ΔS = 0. If we perform the U (1) transformation (13.79) we obtain a con-

tribution (13.80) from the Wess–Zumino term, which implies Nz = (kNC ). But
due to the larger local symmetry we have more restrictions; they imply that the
allowed states have to be singlets under the above mentioned SU (k) symmetry.
This is analogous to the confinement property of QCD, which tells that, due to
the non-abelian gauge invariance, the physical states have to be color singlets.
Here we have an analogous singlet structure of the SU (k) in the flavor space.
Taking a wave function that has zs only (analogous to quarks only for QCD), it
must be of the form,
NC
ψk (z) = α 1 ...α k zi 1 α 1 ...zi k α k , (13.85)
i= 1
for a given set of 1 ≤ i1 , ..ik ≤ NF .

The most general state will then be of the form,
ψ̃(z, z ) = ψk (z)[
(ziα zj α )n i j ], (13.86)
{i,j }
and the products are over given sets of indices.

Using the explicit formula from [103], we obtain the mass of the state repre-
sented by (13.85),
E[ψk ] = M k(NF − k)NC + kEc . (13.87)

To obtain binding energies, consider our k-baryon as built from constituents

kr , such that k = r kr . Then,

B[k|kr ] = −(M NC ) k 2 − i ki2

= −(2M NC ) r > s kr ks . (13.88)
When all kr = 1, the sum gives us 12 k(k − 1), i.e. the number of one-baryon pairs
in the k-baryon state. Note that the binding energy is always negative, thus the
k-baryon is stable. The maximal binding corresponds to the case when all kr = 1.
Note also that in the NC → ∞ limit, the binding tends to a finite value, since
then,
1
1 2NF 4
3
lim (2M NC ) = (Cmec ) 2 π2 . (13.89)
N C →∞ π
Let us take as an example an analog of a deutron, namely a di-baryon k = 2.
Then for NC = 3, NF = 2 we find that its representation is a flavor singlet (this
is the limiting case of k = NF ). The ratio of the binding to twice the baryon
mass is given by,
1
2 = = 0.29. (13.90)
1 + π242
For k = 2, NC = 3 and NF = 3 we find that the di-baryon is represented by, 10
and the ratio is given by,
1
3 = = 0.23. (13.91)
2 + π242
For general NF we obtain,
1 1
F = = . (13.92)
(NF − 1) + 24
π2
NF + 1.43
Finally, let us make the following comment. The ratio of the quantum fluctu-
ations term to the classical term, in the expression for the mass, eqn. (13.87), is
given by,
(Quantum corrections) π2 NF − k
= . (13.93)
(Classical term) 8 NC
Thus, we do not expect our approximations to hold in the region NF ≥ (NC + 1).
We expect it to start for NC ≥ NF , and to be good in the region NC NF .
13.8 Meson-baryon scattering

So far we have analyzed, using semiclassical quantization of the bosonized theory
in the strong coupling limit, the spectrum of the baryons and their flavor content.
Applying the same technique one can also study the scattering processes of
mesons from baryons. The idea is to introduce perturbations around the classical
soliton solutions and to compute the forward phase shifts. We start with the

computation in the strong coupling limit [98] and then we discuss the general
case of any coupling [87].
Our starting point is the soliton solution that describes the static classical
baryon gc (x) = exp[−iΦc (x)] where,
⎛ ⎞
φc (x)
⎜ 0 ⎟
⎜ ⎟
⎜ . ⎟
⎜
Φc (x) = ⎜ ⎟, (13.94)
. ⎟
⎜ ⎟
⎝ . ⎠
0
and,
5
μx 8π
φc (x) = 4arctg (e ) , μ=m . (13.95)
NC
Note that we have shifted the non-trivial phase factor to the upper left-hand
corner, whereas in (13.77) it was put in the lower right-hand one.
We introduce a fluctuation around it of the form,
g = exp {−i [Φc (x) + δφ(x, t)]} (13.96)
1
−iΦ c (x)
g≈e −i dτ e−iτ Φ c (x) δφ(x, t)e−i(1−τ )Φ c (x) . (13.97)
0
Actually, to avoid integrals as in eqn. (13.97), which yield rather complicated

expressions for fluctuations, we will adopt a different expansion, namely,
g = e−iΦ c (x) e−i δ̃ φ(x,t) ,

(13.98)
≈ e−iΦ c (x) − ie−iΦ c (x) δ̃φ(x, t),
where we have denoted by δ̃φ the new variation, different from the δφ of eqn.
(13.97), but still a fluctuation about the classical solution. Now,
Nc
∂+ e−iΦ c (x) ∂− δ̃φ(x, t) eiΦ c (x)
4π
+ m2 e−iΦ c (x) δ̃φ(x, t) + δ̃φ(x, t)eiΦ c (x) = 0. (13.99)
Obviously the two expressions coincide in the abelian case. In fact, the relation
between δφ and δ̃φ is
1
δ̃φ(x, t) = dτ eiτ Φ c (x) δφ(x, t) e−iτ Φ c (x) . (13.100)
0
Physical quantities should obviously come out to be the same for both types of
fluctuation.

13.8.1 Abelian case

We start with the abelian fluctuation δφ that commutes with Φc . Denote this
case by δφab , where the subscript “ab” stands for “abelian.”
Then the fluctuation reads
δg = −iδφab (x)e−iφ c (x) , (13.101)
where,
δφab + μ2 (cos φc )δφab = 0, (13.102)
and,

2
cos φc = 1 − . (13.103)
cosh2 μx
This equation of motion can be derived from the following effective action,
1 2 1 2
Leff = (∂μ δφab ) − V (x) (δφab ) (13.104)
2 2

2
V (x) = μ2 cos φc (x) = μ2 1 − . (13.105)
cosh2 μx
For a solution with an harmonic time dependence of the form,
δφab (x, t) = e−iω t χab (x), (13.106)
the spatial part has to solve,
−ω 2 χab − χab + V (x)χab = 0. (13.107)
Note that asymptotically the potential approaches x → ±∞, the potential
→ μ2 , and so asymptotically,
χab (±∞) + ω 2 χab (±∞) = μ2 χab (±∞). (13.108)
For asymptotic behavior of the form,
χab (x) −→ eik x , (13.109)
|x|→∞
with,
ω 2 = k 2 + μ2 , (13.110)
the two asymptotic solutions are,
χab (x) ∼ A(ω) sin kx + B(ω) cos kx, (13.111)
and the S-matrix is,
1
Sforward = (B − iA) (13.112)
2
1
Sbackward = (B + iA), (13.113)
2
for an incoming wave eik x from x = −∞.

Fig. 13.1. The phase shift δ = 2ctg−1 (k/μ), eqn. (13.117), as a function of the
normalized momentum k/μ, for the potential (13.105), governing the small
fluctuations around the soliton in the abelian case. The phase shift is smooth
and monotonically decreasing with momentum, indicating that no resonance
is present. Note logarithmic momentum scale.
We can now proceed to derive the scattering matrix, using the standard
procedure. The solution for x → ∞ contains only the transmitted wave,
ψ(x → ∞) ∼ eik x .7
It turns out that for the particular potential (13.105) there is no reflection at
all, i.e. the wave function for x → −∞ contains only the incoming wave,
1 + ik/μ
ψ(x → −∞) ∼ eik x−δ = − eik x . (13.114)
1 − ik/μ
Thus,
1 1 + ik/μ
=− (13.115)
T 1 − ik/μ
1 − ik/μ
T =− = eiδ (13.116)
1 + ik/μ
1 k
ctg δ = . (13.117)
2 μ
As shown in Fig. 13.1, δ varies smoothly and decreases monotonically from δ = π
at k = 0 to δ = 0 at k = ∞, indicating that there is no resonance.
The no-reflection potential we found is a special case of a well-known class of
reflectionless in quantum mechanics.
7 We take the convention where the scattering phase is taken to be zero at x → ∞ and is
therefore extracted from the wave function at x → −∞.

13.8.2 The non-abelian case

We got a no-reflection potential in the previous section, in the case of one flavor.
We want to examine now the non-abelian case.
Following eqn. (13.99), we get,
1
δ̃φ − i (∂+ Φc ) ∂− δ̃φ + i ∂− δ̃φ (∂+ Φc ) + μ2 δ̃φe−iΦ c (x) + eiΦ c (x) δ̃φ = 0.
2
(13.118)
The equation for δ̃φij with i, j = 1 is as for the free case,
δ̃φij + μ2 δ̃φij = 0 , i and j = 1, (13.119)
whereas the i = 1, j = 1 matrix element is as in the abelian case,
δ̃φ11 + μ2 (cos φc (x)) δ̃φ11 = 0. (13.120)
with no reflection and no resonance.

So in order to proceed beyond these results, we need to consider δ̃φ1j , j = 1,
or δ̃φi1 , i = 1. As δ̃φ is Hermitian, it is sufficient to discuss one of the above.
Thus we take,
δ̃φ1j = e−iω t uj (x) j = 1, (13.121)
resulting in,

uj (x) − iφc (x)uj (x) + ω 2 + ωφc (x) − 12 μ2 1 + eiφ c (x) uj (x) = 0. (13.122)
Defining,
i
u j ≡ e 2 φ c vj , (13.123)
we find,

vj + ω 2 + ωφc − 12 μ2 (1 + cos φc ) + (φc )
1 2
4 vj = 0. (13.124)
Using,
(φc ) = μ2 (1 − cos φc ) ,
1 2
2 (13.125)
we get,

vj + ω 2 + ωφc − μ2 cos φc vj = 0. (13.126)
This can be rewritten as,
−vj − ω 2 vj + V (x)vj = 0, (13.127)
where,
V (x) = −ωφc + μ2 cos φc =

(ω/μ) 1
= μ2 − 2μ2 + , (13.128)
cosh μx cosh2 μx

Fig. 13.2. The normalized potential V (ω/μ; x)/μ2 of eqn. (13.128), for ω/μ =
1.01 (upper), 2 (middle) and 3 (lower).

with ω = k 2 + μ2 , as before. Note that the potential depends on the momen-
tum of the incoming particle, as shown in Fig. 13.2.
Next we proceed to solve numerically for the reflection and transmission coef-
ficient. It turns out that for numerical solution of the scattering problem it is
more convenient to take the coefficient of the outgoing wave at x ∼ +∞ to be
1, instead of the T prefactor, and integrate eqn. (13.127) backward, reading off
the T and R amplitudes from the solution at x ∼ −∞.
We thus use,
vj (x) = eik x , x → +∞
−ik x
(13.129)
vj (x) = 1
T e ik x
+ R
T e , x → −∞.
Since the potential is symmetric, the symmetric and anti-symmetric scattering

amplitudes don’t mix, yielding two independent phase shifts δS and δA , respec-
tively. This leads to,

T = 12 eiδ S + eiδ A
(13.130)
R = 12 eiδ S − eiδ A .
Defining,
1
δ± = (δS ±δA ) , (13.131)
2
we find that,
T = eiδ + cos δ−
(13.132)
R = ieiδ + sin δ− .

Fig. 13.3. Scattering by the potential eqn. (13.128) as a function of the nor-
malized energy ω/μ. Upper plot: transmission probability |T |2 ; lower plot:
phase of T , δ+ (continuous line). Also shown is the approximate result for δ+
from WKB (dot-dashed line).
Note that R/T is purely imaginary. The transmission and reflections probabilities
are,
|T |2 = cos2 δ−
(13.133)
|R|2 = sin2 δ− .
The numerical results for the transmission probability |T |2 and for the phase of
T , δ+ are presented in Fig. 13.3. For comparison and as an extra check we also
plot the WKB result for δ+ . Note that no resonance appears.
Note that the asymptotic value of the phase shift is π. This can also be obtained
from a WKB calculation, which becomes exact at infinite energies.

13.8.3 Extension to arbitrary coupling

To analyze the system at any gauge coupling we go back to bosonized action
prior to the implementation of the strong coupling limit, which we now rewrite
in the form,

e2c Nf −1 2
Seff [u] = S0 [u] + d x Tr ∂− u ∂− u c + m Nm̃ d 2 x Tr u + u† .
2 † 2
8π2
(13.134)
The strong coupling limit eliminates the second term of (13.134), for arbitrary
coupling,

e2c Nf −1 2
2
d 2 xTr ∂− u ∂ − u† c , (13.135)
8π

where u ∂− u† c is the color part of M ≡ u∂− u† , to be computed as,
Mc = Trf M − 1/Nc Trf &c M. (13.136)
As already mentioned, this term represents the interactions, as it arises from

integrating out the gauge potentials. However, we will see that, for the physical
situation we discuss, this term does not contribute to meson-baryon scattering
for any coupling. As a result, the latter is described by the effective action S̃eff [u],
whereas in the strong coupling limit it is described by S̃eff [g].
In a similar manner to (13.95) we take u to be of the form,
u = exp(−iΦc ) exp(−iδΦ), (13.137)
corresponding to a classical soliton Φc , and a small fluctuation δΦ around it,

representing the meson. The resulting action is then expanded to second order
in δΦ, yielding a linear equation of motion for δΦ in the soliton background. The
latter serves as an external potential in which the meson is propagating.
We start by evaluating,
M ≡ u∂− u† =
= exp(−iΦc ) ∂− (exp iΦc ) + exp(−iΦc ) exp(−iδΦ) [∂− exp(iδΦ)] exp(iΦc ),
(13.138)
and obtain the equations of motion for the meson field by varying with respect
to δΦ. The variation of (13.135) with respect to δΦ is proportional to,
δMc −2
∂ Mc . (13.139)
δ(δΦ)
To compute its variation with respect to δΦ, we need only the second term M2
of M , as the first term M1 is independent of δΦ.

We take for the soliton a diagonal ansatz (13.94) now in the form of a u matrix
rather then a g one,
√
[exp(−iΦc )]aa j j = δaa δj j exp (−i 4πχ α j ) :
a = 1, . . . , Nc , (13.140)
j = 1, . . . , Nf ,
so that,
{exp(−iΦc )[exp(−iδΦ) ∂− exp(iδΦ)] exp(iΦc )}aj,aj

√ √
= exp(−i 4πχaj ) [exp(−iδΦ) ∂− exp(iδΦ)]aj,a j exp(i 4πχa j ). (13.141)
The part of M that contributes to the effective action is its color projection
(13.136). We note that Trf &c M2 = 0, and thus,
√ √
[(M2 )c ]a,a = exp(−i 4πχaj )[exp(−iδΦ) ∂− exp(iδΦ)]aj,a j exp(i 4πχa j ).
j
(13.142)
The mesons δΦ have to be diagonal in color, so,

[(M2 )c ]a,a = [ exp(−iδΦ) ∂− exp(iδΦ) ]aj,aj δa,a . (13.143)
j
We recall that the flavor structure of the mesons is independent of their color
indices, and restrict our attention to mesons that have no U (1) flavor part. In
this way, we may be sure that classical solutions lead to stable particles, since
their non-vanishing flavor quantum numbers put them in a different sector from
the vacuum. We then have,

[ exp(−iδΦ) ∂− exp(iδΦ) ]α j,α j = 0, (13.144)
j
as shown earlier, and the effective meson-baryon action is,

S̃m-b [δΦ] = S0 [u] + m Nm d2 x Tr u + Tr u† ,
2
(13.145)
with u depending on δΦ for fixed Φc as in (13.137).

Next we would like to evaluate the potential. The equation of motion for δΦ
is obtained from (13.145), by first varying with respect to u and then varying u
with respect to δΦ. To first order in δΦ, we find,
δu = −i[exp(−iΦc )]δΦ. (13.146)
The resulting equation of motion is then,

1
∂+ (∂− u) u† + um2 − m2 u† = 0, (13.147)
4π
where m is the diagonal mass matrix: m = δij mj with (possibly different) entries
mj corresponding to flavors j. We note that there is the possibility of an overall

scale ambiguity in m, since, when the masses are different, there is a question of
which normal-ordering scale to use. The resulting equation of motion for δΦ is,
δΦ − i (∂+ Φc ) (∂− δΦ) + i (∂− δΦ) (∂+ Φc )

1
+ δΦμ2 exp(−iΦc ) + exp(iΦc )μ2 δΦ = 0, (13.148)
2
√
where μ ≡ m 8π.
As discussed before, both Φc and δΦ are diagonal in color. Moreover, Φc is
diagonal in flavor too. So, taking the aajj matrix element of the equation of
motion (13.148), we find,
δΦaj j − i(∂+ Φc )aj (∂− δΦ)aj j +i (∂− δΦ)aj j (∂+ Φc )aj

1
+ {δΦaj j μ2 j [exp(−iΦc )]aj + [exp(iΦc )]aj μ2 j δΦaj j } = 0. (13.149)
2
Examining the classical solutions for the quark solitons inside the baryons, we see
that, for a given color index a, there is only one flavor for which Φc is non-zero.
We can now distinguish three cases:
r The first is when an index a and indices j and j are chosen in such a way that
both (Φc )aj and (Φc )aj are zero. In such a case,
1
δΦaj j + [μ2 j + μ2 j ]δΦaj j = 0, (13.150)
2
where (Φc )aj = 0 and (Φc )α j = 0.
Thus δΦaj j is a free field with squared mass given by the average of m2j and
m2j in this case, which we do not discuss further.
r The second case is that of j = j , with a such that (Φc ) is a quark soliton
aj
inside the baryon. In this case,
δΦaj j + μ2 j cos[(Φc )aj ]δΦaj j = 0. (13.151)
r The third case is when j is different from j , now with one of the Φc being a
soliton and the other vanishing. Taking (Φc )aj to be the soliton, we obtain,
1
δΦaj j − i(∂+ Φc )aj (∂− δΦ)aj j + {μ2 j + μ2 j [exp(iΦc )]aj }δΦaj j = 0,
2
(13.152)
where j = j and (Φc )α j = 0.
Next we want to proceed and evaluate the meson-baryon scattering. For that
purpose we need to analyze the equations that determine the static solution
(Φc )aj . First one defines,
√
(Φc )aj = 4π(χc )aj , (13.153)

where the (χc )aj are canonical fields, whose equations of motion are,
⎛ ⎞
1 √ √
χα j − 4αc ⎝ χα − χ β l ⎠ − 2 4πm2j sin 4πχ α j = 0.
Nc
l βl
Note the extra factor 2 in front of the mass term, as compared with eqn. (22)
of [86], due to an error in this reference.
Choosing the boundary conditions χaj (−∞) = 0, we get as constraints for
χaj (+∞), denoted hereafter simply by χaj ,
1
√ χ α j = nα j integers, (13.154)
π
and

nα = n independent of a. (13.155)
l
The baryon number8 associated with any given flavor l is given by,

Bl = nα .
a
Combining the last two equations, we find,

B= Bl = nNc ,
l
for the total baryon number.

We now continue in a similar manner to the discussion in the strong coupling
limit, starting with the first non-trivial case (13.151) identified above. As the
soliton solutions are such that there is a unique correspondence between the
color index a and the flavour index j, we suppress a in what follows. Putting,
δΦj j = e−iω j t uj (x), (13.156)
with,
uj (x) −→ eik x , (13.157)
x→∞
we find,
ωj2 = k 2 + μ2j , (13.158)
and the equation for uj (x) is,
uj (x) + ωj 2 uj − μ2 j [cos (Φc )j ]uj = 0. (13.159)
We define the potential Vj for this scattering process via,
uj (x) + ωj 2 uj − Vj uj = 0, (13.160)
8 In our normalization, a single quark carries one unit of baryon number.

and find,
Vj = μ2 j [cos (Φc )j ]. (13.161)
In our normalization the outgoing wave has coefficient 1, which is more conve-
nient for numerical calculations, and the wave for x → −∞ is now,
1 ik x Rj −ik x
uj (x) = e + e , x → −∞, (13.162)
Tj Tj
in this case.
In the second non-trivial case (13.152), we put,
δΦj j = e−iω j j t uj j (x), (13.163)
so that,
uj j (x) − i(Φc )j (x)uj j (x) + {ωj2j + ωj j (Φc )j (x)
1
− {μ2 j + μ2 j [exp(iΦc )]j }}uj j = 0. (13.164)
2
To eliminate the first derivative term in u, we substitute,

i
uj j = exp Φc vj j . (13.165)
2 j
This results in,

1
vjj (x) + {ωj2j + ωj j (Φc )j (x) − μ2 j [cos(Φc )]j }vj j + (μ2 j − μ2 j )vj j
2
* +
1 1
+ [(Φc )j (x)]2 − μ2 j (1 − [cos(Φc )]j ) vj j
4 2
i: ;
+ (Φc )j (x) − μ2 j [sin(Φc )j ] vj j = 0. (13.166)
2
We note that the last three lines vanish when all the quark masses are equal, as
then the soliton is a sine-Gordon one. Thus, the scattering would then only be
elastic.
The potential of the scattering is defined here via,
vjj (x) + ωj2j vj j − Vj j vj j = 0, (13.167)
so that,
Vj j = −ωj j (Φc )j (x) + μ2 j [cos(Φc )]j
1
− (μ2 j − μ2 j )
2
* +
1 1
− [(Φc )j (x)]2 − μ2 j (1 − [cos(Φc )]j )
4 2
i
− {(Φc )j (x) − μ2 j [sin(Φc )j ]}. (13.168)
2

Taking again,
vj j (x) −→ eik x , (13.169)
x→∞
we get,
1 2
ωj j = (μ j + μ2 j ), (13.170)
2
and the wave for x → −∞ is,
1 ik x Rj j −ik x
vj j (x) = e + e , x → −∞, (13.171)
Tj j Tj j
in this case.
To summarize we have shown that meson-baryon scattering in QCD2 in the
large-Nc limit is non-trivial for non-zero quark masses, and is described by two
distinct effective potentials when the quark masses are unequal. These effective
potentials are not of the sine-Gordon type found in previous cases, and we expect
the scattering amplitudes also to be non-trivial. Their calculation will require
numerical analysis.

14
Confinement versus screening
One of the most challenging problems of gauge dynamics in four dimensions is

how to show that QCD admits confinement. One of the measures of confinement
is the fact that the potential between an external quark and an external anti-
quark placed at a separation distance L, as in Fig. 14.1, is dominated by a linear
dependence, namely,
V = σL. (14.1)
The coefficient in this linear dependence is the string tension. Thus, a non-
confining behavior, which will be referred to as a screening behavior, implies
a vanishing string tension. Whereas in four dimensions the computation of the
string tension is a formidable task, in two dimensions, as will be shown in this
chapter, it is a fairly easy one. In this chapter we describe the extraction of the
string tension in various two-dimensional gauge systems.
We start by calculating the string tension for the massive Schwinger model in
both the fermionic and the bosonic languages. This is done in the small mass
limit and then we discuss the corrections due to going beyond this limit. We then
discuss the short range corrections to the confining potential. We focus on the
abelian case, believing that the non-abelian case is very similar. Next we com-
ment on the behavior of the string tension when finite temperature is introduced.
Then we move to non-abelian generalization. We compute the string tension for
the cases of matter in the fundamental and adjoint representations, followed by
the symmetric and anti-symmetric representations.
Much of this chapter is based on [15] and [16].
The string tension of the massive Schwinger model was calculated using
bosonizaton in [68]. The massless cases in gauge theories were analyzed in [116].
The next-to-leading order in small mass was computed by [4].
14.1 The string tension of the massive Schwinger model

We start with the derivation of the string tension in the massive Schwinger model,
in the fermionic language. Consider the partition function of two dimensional
massive QED2 ,
Z= (14.2)

1 2
DAμ DΨ̄DΨ exp i d2 x − 2 Fμν + Ψ̄i∂Ψ − mΨ̄Ψ − qdyn Aμ Ψ̄γ μ Ψ ,
4e

266 Confinement versus screening
q L q
Fig. 14.1. Quark anti-quark separated at a distance L.
where qdyn is the charge of the dynamical fermions. Gauge fixing terms were not
written explicitly. Let us add an external pair with charges ±qext at ±L, namely,
j0ext = qext (δ(x + L) − δ(x − L)), (14.3)
so that the change of L is −jμext Aμ (x). Note that by choosing jμext which is
conserved, ∂ μ jμext = 0, the action including the coupling to the external current
is also gauge invariant.
Now, one can eliminate this charge by performing a local, space-dependent
left-handed rotation,
1
Ψ → eiα (x) 2 (1−γ 5 ) Ψ (14.4)
−iα (x) 12 (1+γ 5 )
Ψ̄ → Ψ̄e , (14.5)
where γ 5 = γ 0 γ 1 . We choose a left-handed rotation (or equally well a right-

handed one) rather than an axial one, since in the non-abelian case the former
will be easier to implement.
The new action is,

1 2 1
S = d x − 2 Fμν
2
+ Ψ̄i∂Ψ − Ψ̄∂μ α(x)γ μ (1 − γ5 )Ψ − mΨ̄e−iα (x)γ 5 Ψ
4e 2

α(x)qdyn
− qdyn Aμ Ψ̄γ Ψ − qext (δ(x + L) − δ(x − L))A0 +
μ
F , (14.6)
2π
where the last term is induced by the chiral anomaly,

α(x)qdyn
δS = d2 x F, (14.7)
2π
with F the dual of the electric field, F = 12 μν Fμν .

The external source and the anomaly term are similar, both being linear in
the gauge potential. This is the reason that the θ-vacuum, to be discussed in
Chapter 22, and electron-positron pair at the boundaries are the same in two
dimensions.
In the following we assume θ = 0, as otherwise we absorb it in to α. Choos-
ing the A1 = 0 gauge and integrating by parts, the anomaly term looks like an
external source,
qdyn
A0 ∂1 α(x). (14.8)
2π

14.1 The string tension of the massive Schwinger model 267
This term can cancel the external source by the choice,

qext
α(x) = 2π (θ(x + L) − θ(x − L)). (14.9)
qdyn
Let us take the limit L → ∞. The form of the action, in the region B of −L <
x < L is,

1 2 q
−i2π q e x t γ 5
SB = d2 x − 2 Fμν + Ψ̄i∂Ψ − mΨ̄e dyn Ψ − qdyn Aμ Ψ̄γ μ Ψ .
B 4e
(14.10)
Thus the total impact of the external electron-positron pair is a chiral rotation
of the mass term. This term can be written as,
q
−i2π q e x t γ 5 qext qext
Ψ̄e dyn Ψ = cos 2π Ψ̄Ψ − i sin 2π Ψ̄γ5 Ψ. (14.11)
qdyn qdyn
The string tension is the vacuum expectation value (v.e.v.) of the Hamiltonian
density in the presence of the external source relative to the v.e.v. of the Hamil-
tonian density without the external source, in the L → ∞ limit,
σ = <H> − <H0 >0 , (14.12)
where |0>0 is the vacuum state with no external sources. The change in the
vacuum energy is due to the mass term. The change in the kinetic term which
appears in (14.6) does not contribute to the vacuum energy.
Thus,
qext qext
σ = m cos 2π <Ψ̄Ψ> −m sin 2π <Ψ̄iγ5 Ψ> −m <Ψ̄Ψ>0 .
qdyn qdyn
(14.13)
The values of the condensates <Ψ̄Ψ> and <Ψ̄γ5 Ψ> are needed. The easiest
way to compute these condensates is bosonization, but it can also be computed
directly in the fermionic language. We state here the final result for the m = 0
case (the derivation can be found in the references of this chapter),
exp(γ)
<Ψ̄Ψ>m = 0 = −e (14.14)
2π3/2
<Ψ̄γ5 Ψ>m = 0 = 0. (14.15)
Equation (14.15) is due to parity invariance (with our choice θ = 0). The
resulting string tension, to first order in m,
exp(γ) qext
σ = me 1 − cos 2π . (14.16)
2π3/2 qdyn
Though this expression is only the leading term in a m/e expansion and might
be corrected, when qext is an integer multiple of qdyn the string tension is exactly
zero, since in this case the rotated action (14.10) is not changed from the original
one (14.2).

14.2 The Schwinger model in bosonic form

Next we derive the same result in the bosonized formulation.
The bosonized Lagrangian, in the gauge A1 = 0, is given by,
1 1 √ qdyn
L = 2 (∂1 A0 )2 + (∂μ φ)2 + M 2 cos(2 πφ) + √ A0 ∂1 φ − A0 jext , (14.17)
2e 2 π
where M 2 = mμ, μ = exp(γ )

2π μ(φ) with μ(φ) =
√e qdyn the mass of the photon, for
π
e m.
Chiral rotation corresponds to a shift in the field φ. Upon the transformation,
√ qext
φ = φ̃ + π (θ(x + L) − θ(x − L)) . (14.18)
qdyn
The Lagrangian (14.17) takes, in the region B, the form,
1 1 √ qext qdyn
LB = 2 (∂1 A0 )2 + (∂μ φ̃)2 + M 2 cos 2 πφ̃ + 2π + √ A0 ∂1 φ̃.
2e 2 qdyn π
(14.19)
Hence, similarly to the previous derivation, a local chiral rotation was used to
eliminate the external source. The calculation of the string tension is exactly the
same as in the previous section.
The relevant part of the Hamiltonian density is,
√ qext
H = −M 2 cos 2 πφ̃ + 2π . (14.20)
qdyn
2
To zeroth order in ( Me ) , the vacuum is φ̃ = 0. Setting this choice in (14.20) and
subtracting the v.e.v. of the free Hamiltonian, we arrive at,
qext
σQ E D = mμ 1 − cos 2π , (14.21)
qdyn
where m is the electron mass, μ = e exp(γ

2π3 / 2
)
, e the gauge coupling, γ the Euler
number and qext , qdyn are the external and dynamical charges, respectively (we
measure charge in units of e, thus qext and qdyn are dimensionless).
14.3 Beyond the small mass abelian string tension

The expression (14.21) contains only the leading me contribution to the abelian
string tension. This expression was computed in the previous section, using a
classical average. However, as we used the normal ordering scale μφ which is the
photon mass for e m, taking φ̃ = 0 actually gives the full quantum answer,
as is evident by comparing with the fermionic calculation in the section before
that.
The full perturbative (in m) string tension can be written as,
∞
l−1
m qext
σQ E D = mμ Cl 1 − cos 2πl . (14.22)
eqdyn qdyn
l= 1

14.4 Correction to the leading long distance abelian potential 269
The value of the first coefficient is C1 = 1 and the next was found to be
C2 = −8.9 exp(γ
8π1 / 2
)
. Higher coefficients are not calculated yet.
Note that for finite me we have to minimize the potential,
√ qext 1
V = M2 1 − cos 2 πφ + 2π + μ2φ φ2 . (14.23)
qdyn 2
The minimum φ = φm obeys,

√ √ qext
2 πM 2 sin 2 πφm + 2π + μ2φ φm = 0. (14.24)
qdyn
√
Thus, for the first-order ( eqmd y n ) correction, we get a C2 which is −( 12 ) π(exp γ).
This has the same sign, but a factor 1.41 larger, than the instanton contribution.
Note that all above results for the string tension are symmetric under change
of sign of the external charge, as expected on general grounds. However, when
a θF term is introduced, we get odd terms as well, like sin(lθ) sin(2πl qqde yx nt ). The
even terms are multiplied by cos(lθ).
Finally, let us remark that for very large me , the abelian case has a string
tension which is 12 e2 qext
2
.
14.4 Correction to the leading long distance abelian potential

The potential (14.1) is the dominant long-range term. However, there are, of
course, corrections. In this section we present these corrections.
The equations of motions which follow from the bosonized Lagrangian (14.17)
are, in the static case,
1 2 qdyn
− ∂ A0 + √ ∂1 φ − jext = 0 (14.25)
e2 1 π
√ √ qdyn
−∂1 φ + 2 πM 2 sin 2 πφ + √ ∂1 A0 = 0.
2
(14.26)
π
In order to solve these equation, it is useful to eliminate the bosonized matter

√ √
field φ. Using the approximation sin 2 πφ ∼ 2 πφ, we arrive at (in momentum
space),
e2 (k 2 + 4πM 2 )
A0 (k) = jext (k), (14.27)
2 2
k2 k 2 + 4πM 2 + eπ qdyn
where k is the Fourier transform of the space coordinate. We will discuss the
validity of our approximation for φ later in this section. The last equation can
be rewritten as,
m21 1 m21 1
A0 (k) = + 1 − e2 jext (k), (14.28)
m22 k 2 m22 k2 + m22

where,
m21 = 4πM 2 , (14.29)

2
e 2
m22 = 4πM 2 + q . (14.30)
π dyn
Note that the photon propagator has two poles, a massless pole that repro-
duces the string tension and a massive pole which adds a screening term to
the potential. Note that there is no const.
L correction, which appears in higher
dimensions, since in the present case the string cannot fluctuate in transverse
directions.
Note also that in the massless case, when M 2 = 0, only the second term sur-
2
vives and the photon has only one pole with mass square eπ qdyn
2
. This result is
of course exact, independent of our approximation.
The resulting gauge field is,
2π2 M 2 qext
A0 (x) = 2 (| x + L | − | x − L |)
qdyn
√
e π qext − √eπ q d y n |x+L | − √e q |x−L |

− e − e π dyn , (14.31)
2 qdyn
where we took M 2 e2 for simplicity.

In order to calculate the potential we will use,

1
V = A0 (x)jext (x)dx. (14.32)
2
Hence the potential is,
√ 2
2
qext e π qext − √e q 2L
V = 2π M 2 × 2L +
2 2
1 − e π dyn . (14.33)
qdyn 2 qdyn
The first term is the confining potential which exists whenever the quark mass
is non-zero. On top of this, there is always a screening potential.
The string tension which results from the above potential is,
2
qext
σ = mμ × 2π2 2 , (14.34)
qdyn
which is exactly (14.21) in the approximation 2π qqde yx nt 1. This turns out to be

√ √
also the condition for sin 2 πφ ∼ 2 πφ that we assumed at the start of this
section. To see that, we solve for φ from eqn. (14.25) as,
qdyn e2 1 1
φ(k) = −ik √ − 2 jext (k). (14.35)
π m22 k2 k + m22
Define φ = φ1 + φ2 , where φ1 is the part with k12 , and φ2 with k 2 +m

1
2 . The φ2
2
part goes to zero at long distances, i.e. k → 0. As for the φ1 part, its x-space

14.5 Finite temperature 271
form is,
e2
φ1 (x) = √ qdyn qext (θ(x + L) − θ(x − L)), (14.36)
πm22
m
which for small e reduces to,
√ qext
φ1 (x) ∼ π (θ(x + L) − θ(x − L)). (14.37)
qdyn
√
Thus 2 πφ small means,
qext
(2π) 1, (14.38)
qdyn
the condition mentioned before.

Note that we could generalize the argument to values of 2π qqde yx nt that are close
to 2πn, with integer n.
14.5 Finite temperature

In this section we would like to comment on the behavior of the string tension in
the presence of finite temperature. It is interesting to check whether the string
is torn due to high temperature and whether the system undergoes a phase
transition from confinement to deconfinement.
The prescription for calculating quantities at finite temperature T is to for-
mulate the theory on a circle in Euclidean time with circumference β = T −1 .
For the purpose of calculating the string tension, we can follow the same steps
which we employed previously, leading to a modification of eqn. (14.16) as,
qext
σ = −m <Ψ̄Ψ>T 1 − cos 2π . (14.39)
qdyn
It is enough to calculate <Ψ̄Ψ>T , the condensate at finite temperature, in the

massless Schwinger model.
The chiral condensate behaves as,
e
<Ψ̄Ψ>(T →0) → − eγ , (14.40)
2π3/2
and
π3 / 2 T
<Ψ̄Ψ>(T →∞) → −2T e− e . (14.41)
This result indicates that the string is not torn even at very high temperatures.
The explicit expression shows that <Ψ̄Ψ>T is non-zero for all T . Thus, the system
does not undergo a phase transition. It is just energetically favorable to have the
electron-positron pair confined.

14.6 Two-dimensional QCD

The action of bosonized QCD2 with massive quarks in the fundamental repre-
sentation of SU (N ) (see Chapter 8) is

1
Sfundam ental = d2 x tr ∂μ g∂ μ g † (14.42)
8π Σ

1
+ d3 yij k tr(g † ∂i g)(g † ∂j g)(g † ∂k g)
12π B

1 1 a aμν
+ mμf u n d d2 x tr(g + g † ) − d2 x 2 Fμν F
2 4e

1
− d2 x tr(ig † ∂+ gA− + ig∂− g † A+ + A+ gA− g † − A+ A− ),
2π
where e is the gauge coupling, m is the quark mass, μ = e exp(γ3) , g is an N × N

(2π) 2
unitary matrix, Aμ is the gauge field and the trace is over U (N ) indices. Note,
however, that only the SU (N ) part of the matter field g is gauged.
When the quarks transform in the adjoint representation, the expression for
the action is,

1
Sadjoint = d2 x tr ∂μ g∂ μ g † (14.43)
16π Σ

1
+ d3 yij k tr g † ∂i g g † ∂j g g † ∂k g
24π B

1 †
1 a aμν
+ mμadj d x tr g + g − d2 x 2 Fμν
2
F
2 4e

1
− d2 x tr ig † ∂+ gA− + ig∂− g † A+ + A+ gA− g † − A+ A− .
4π
The action (14.43) differs from (14.42) by a factor of one half in front of the
WZW and interaction terms, because g is real and represents Majorana fermions.
Another difference is that g now is an (N 2 − 1) × (N 2 − 1) orthogonal matrix.
The two actions (14.42) and (14.43) can be schematically represented by one
action,

1
S = S0 + mμR d2 x tr g + g † (14.44)
2

ikdyn a
− d2 x g∂− g † Aa+ ,
4π
where A− = 0 gauge was used, S0 stands for the WZW action and the kinetic
action of the gauge field, kdyn is the level (the chiral anomaly) of the dynamical
charges (k = 1 for the fundamental representation of SU (N ) and k = N for the
adjoint representation).
Let us add an external charge to the action. We choose a static charge (with
respect to the light-cone coordinate x+ ) and therefore we can omit its kinetic

term from the action. Thus an external charge coupled to the gauge field would
be represented by,

ikext a
− d2 x u∂− u† Aa+ .
4π
Suppose that we want to put a quark and an anti-quark at a very large separation.
A convenient choice of the charges would be a direction in the algebra in which
the generator has a diagonal form. The simplest choice is a generator of an
SU (2) subalgebra. Since a rotation in the algebra is always possible, the results
are insensitive to this specific choice. As an example we write down the generator
in the case of fundamental and adjoint representations,
⎛ ⎞
1 1
3
Tfund = diag ⎝ , − , 0, 0, . . . , 0⎠
2 2 D EB C
N −2
⎛ ⎞
⎜ 1 1 1 1 1 1 ⎟
3
Tadj = diag ⎜
⎝1, 0, −1, 2 , − 2 , 2 , − 2 , . . . , 2 , − 2 , 0, 0, . . . , 0⎟⎠.
D EB C D EB C
2 (N −2)
2(N −2) dou blets
Generally T 3 can be written as
T 3 = diag(λ1 , λ2 , . . . , λi , . . . , 0, 0, . . .),
where {λi } are the ‘isospin’ components of the representation under the SU (2)
subgroup.
We take the SU (N ) part of u as,
3
u = exp −i4π θ(x− + L) − θ x− − L Text , (14.45)
3
for N > 2, and a similar expression but with a 2π factor for N = 2. Text represents
the ‘3’ generator of the external charge and u is static with respect to the light-
cone time coordinate x+ . The theta function is used as a limit of a smooth
function which interpolates between 0 and 1 over a very short distance. In that
limit u = 1 everywhere except at isolated points, where it is not well defined.
The form of the action (14.44) in the presence of the external source is,
*
1
S = S0 + mμR d2 x tr g + g †
2
+
ikdyn a a
+ − g∂− g † + kext δ a3 δ x− + L − δ x− − L A+ .
4π
The external charge can be eliminated from the action by a transformation of
the matter field. A new field g̃ can be defined as follows,
ikdyn a ikdyn a
− g̃∂− g̃ † = − g∂− g † + kext δ a3 δ(x− + L) − δ(x− − L) .
4π 4π

This definition leads to the following equation for g̃ † ,

kext
∂− g̃ † = g̃ † g∂− g † + i4π (δ(x− + L) − δ(x− − L))Tdyn
3
. (14.46)
kdyn
The solution of 14.46 is,
* +
kext
g̃ † = P exp dx− g∂− g † + i4π (δ(x− + L) − δ(x− − L))Tdyn
3
kdyn
i4π k e x t θ (x − +L )T d3y n −i4π k e x t θ (x − −L )T d3y n
k k
=e dyn g† e dyn , (14.47)
3
where P denotes path ordering and we assume that Tdyn commutes with g∂− g †
for x− ≥ L and with g † for x− = −L (as we shall see, this assumption is self
consistent with the vacuum configuration).
Let us take the limit L → ∞. For −L < x− < L, the above relation simply
means that,
k
i4π k e x t T d3y n
g = g̃e dyn .
Since the Haar measure is invariant (and finite, unlike the fermionic case) with
respect to unitary transformations, the form of the action in terms of the new
variable g̃ reads,

ikdyn a
S = SWZW (g̃) + Skinetic (Aμ ) − d2 x g̃∂− g̃ † Aa+
4π

1 i4π
k ext
T3
k
−i4π k e x t T d3y n †
+ mμR d2 x tr g̃e k d y n d y n + e dyn g̃ , (14.48)
2
which is QCD2 with a chiraly rotated mass term.
The string tension can be calculated easily from (14.48). It is simply the vac-
uum expectation value (v.e.v.) of the Hamiltonian density, relative to the v.e.v.
of the Hamiltonian density of the theory without an external source,
σ =<H> − <H0 > .
The vacuum of the theory is given by g̃ = 1. In terms of the variable g, this

configuration points in the ‘3’ direction and hence satisfies our assumptions while
solving eqn. (14.46). The v.e.v. is,
1 k
i4π e x t T 3
k
−i4π k e x t T d3y n
<H> = − mμR tr e k d y n d y n + e dyn
2
kext
= −mμR cos 4πλi .
i
kdyn
Therefore the string tension is,

kext
σ = mμR 1 − cos 4πλi , (14.49)
i
kdyn
which is the desired result.

We expect that similar corrections as those in eqn. (14.22) will occur also in
non-abelian systems. For the fundamental/adjoint case, the following expression
may correct the leading term,
∞
l−1
m kext
σQ C D = mμR C̃l 1 − cos 4πλj l . (14.50)
ekdyn j
kdyn
l= 1
A few remarks should be made:
(i) The string tension (14.49) reduces to the abelian string tension (14.21) when
abelian charges are considered. It follows that the non-abelian generalization
is realized by replacing the charge q with the level k.
(ii) The string tension was calculated in the tree level of the bosonized
action. Perturbation theory (with m as the coupling) may cause changes,
eqn. (14.49), since the loop effects may add O(m2 ) contributions. However,
we believe that it would not change its general character. In fact, one fea-
ture is that the string tension vanishes for any m when kkde yx nt is an integer,
as follows from eqn. (14.48), since the action does not depend then on kext
at all.
(iii) When no dynamical mass is present, the theory exhibits screening. This is
simply because non-abelian charges at the end of the world interval can be
eliminated from the action by a chiral transformation of the matter field.
(iv) When the test charges are in the adjoint representation kext = N , eqn.
(14.49) predicts screening by the fundamental charges (with kdyn = 1).
(v) String tension appears when the test charges are in the fundamental rep-
resentation and the dynamical charges are in the adjoint. The value of the
string tension is
4π 2π
σ = mμadj 2 1 − cos + 4 (N − 2) 1 − cos (14.51)
N N
as follows from eqn. (14.49) for this case.
The case of SU (2) is special. The 4π which appears in eqn. (14.49) is replaced
by 2π, since the bosonized form of the external SU (2) fundamental matter differs
by a factor of a half with respect to the other SU (N ) cases. Hence, the string
tension in this case is 4mμadj .
(vi) We would like to add, that when computing the string tension in the pure
YM case with external sources in representation R, the Wilson loop gives
1 2 1 2 2
2 e C2 (R), while our way of defining external source gives 2 e kext . Thus we
C 2 (R )
need a factor k 2 to bring our result to the Wilson loop case. Analogous
ext
factors should be computed for the other cases, when dynamical matter is
also present.

14.7 Symmetric and antisymmetric representations

The generalization of (14.49) to arbitrary representations is not straightforward.
However, we can comment about its nature (without rigorous proof).
Let us focus on the interesting case of the antisymmetric representation. One
can show that the WZW action with g taken to be 12 N (N − 1) × 12 N (N − 1)
unitary matrices, is a bosonized version of QCD2 with fermions in the antisym-
metric representation.
The antisymmetric representation is described in the Young-tableaux nota-
tion by two vertical boxes. Its dimension is 12 N (N − 1) and its diagonal SU (2)
generator is,
1 1 1 1 1 1
3
Tas = diag , − , , − , ..., , − , 0, 0, ..., 0 , (14.52)
D2 2 2 EB2 2 2C
(N −2) doublets
and consequently k = N − 2. When the dynamical charges are in the funda-

mental and the external in the antisymmetric the string tension should vanish
because the tensor product of two fundamentals include the antisymmetric rep-
resentation. Indeed, (14.49) predicts this result.
The more interesting case is when the dynamical charges are antisymmetric
and the external are fundamentals. In this case the value of the string tension
depends on whether N is odd or even.
When N is odd the string tension should vanish because the anti-fundamental
representation can be built by tensoring the antisymmetric representation with
itself 12 (N − 1) times. When N is even string tension must exist.
Note that (14.49) predicts,
2π
σ = 2mμas (N − 2) 1 − cos , (14.53)
N −2
which is not zero when N is odd, contrary to expectation.
The resolution of the puzzle seems to be the following. Non-abelian charge
can be static with respect to its spatial location. However, its representation
may change in time due to emission or absorption of soft gluons (without cost
of energy). Our semi-classical description of the external charge as a c-number
is insensitive to this scenario. We need an extension of (14.45) which takes into
account the possibilities of all various representations. One possible extension is,

a
jext = δ a3 kext (1 + lN ) δ x− + L − δ x− − L , (14.54)
where l is an arbitrary positive integer. This extension takes into account the
cases which correspond to 1 + lN charges multiplied in a symmetric way. The
resulting string tension is,
kext
σ = mμR 1 − cos 4πλi (1 + lN ) , (14.55)
i
kdyn
which includes the arbitrary integer l. What is the value of l that we should pick?

14.7 Symmetric and antisymmetric representations 277
The dynamical charges are attracted to the external charges in such a way
that the total energy of the configuration is minimal. Therefore the value of l
which is needed, is the one that guarantees minimal string tension.
Thus the extended expression for string tension is the following,
@ A
kext
σ = min mμR 1 − cos 4πλi (1 + lN ) . (14.56)
l
i
kdyn
In the case of dynamical antisymmetric charges and external fundamentals

and odd N , l = 12 (N − 3) gives zero string tension. When N is even the string
tension is given by (14.53).
The expression (14.56) yields the right answer in some other cases also, like
the case of dynamical charges in the symmetric representation. The bosoniza-
tion for this case can be derived in a similar way to that of the antisymmetric
representation, and T 3 is given by
1 1 1 1 1 1
m = diag 1, 0, −1, , − , , − , . . . , , − , 0, 0, . . . , 0 .
3
Tsym (14.57)
D2 2 2 EB 2 2 2C
(N −2) doublets
Hence k = N + 2. When the external charges transform in the fundamental rep-

resentation and N is odd, eqn (14.56) predicts zero string tension (as it should).
When N is even the string tension is given by
4π 2π
σ = 2mμsym m 1 − cos + (N − 2) 1 − cos . (14.58)
N +2 N +2
We have discussed only the cases of the fundamental, adjoint, anti-symmetric
and symmetric representations, since we used bosonization techniques which are
applicable to a limited class of representations.

15
QCD2, coset models and BRST quantization
15.1 Introduction
In Chapter 9 we realized that the structure of the bosonized non-abelian massless
QCD2 is that of a gauged WZW model with an additional F 2 term of the gauge
fields. Apart from the pure gauge term, this is therefore a special form of a
two-dimensional coset model, discussed in Section 4.6. This naturally calls for
a treatment of the system similar to that for a coset model. Using the form of
the gauge fields in terms of scalars f and f¯ as A = if −1 ∂f , Ā = if¯∂¯f¯−1 with
f (z, z̄), f¯(z, z̄) ∈ H c , the complexification of H ≡ SU (NC ), leads to a convenient
formulation of the model.1 The main advantage of this approach is that one can
then easily decouple the “matter” and the gauge degrees of freedom.
In this chapter we point out that the F 2 term requires a special treatment.
The formulation of pure YM theory in terms of the f variables seems naively to
contain unexpected “physical” massive color singlet states. This result is obvi-
ously neither in accordance with our ideas of the degrees of freedom of the model
nor with the lattice and continuum solution of the theory. We show that similar
“naive” manipulations in the case of QED2 do reproduce the Schwinger model
results. Using a coupling constant renormalization we show that in the limit of no
matter degrees of freedom the coupling constant is renormalized to zero. In this
case the unexpected states turn into unphysical massless “BRST” exact states.
In the flavored QCD2 case a similar analysis shows the existence of physical
flavorless states of mass m2 = N2πF e2c .
This chapter is based on [96].
15.2 The action

The bosonized version of QCD2 was shown in Chapter 9 to be described by the
action,

1
SQ C D 2 = S1 (u) − d2 zTr(iu−1 ∂uĀ + iu∂u
¯ −1 A + Āu−1 Au − AĀ)
2π

m2 2 −1 1
+ d z : TrG [u + u ] : + 2 d2 zTrH F 2 , (15.1)
2π ec
1 This parameterization of the gauge field was previously introduced in [9].

280 QCD2 , coset models and BRST quantization
where u ∈ U (NF × NC ), Sk (u) is a level k WZW model,

k k
Sk (u) = d2 xTr(∂μ u∂ μ u−1 )+ d3 yεij k Tr(u−1 ∂i u)(u−1 ∂j u)(u−1 ∂k u),
8π 12π B
(15.2)
A and Ā take their values in the algebra of H ≡ SU (NC ), F = ∂A ¯ − ∂ Ā +
i[A, Ā], m2 equals mq μC, where μ is the normal ordering mass and C = 12 eγ
with γ, Euler’s constant. Apart from the last two terms which correspond to the
G
quark mass term and the YM term, the rest of the action is a level-one H coset
model with G = U (NF × NC ).
We now introduce the following parameterization for the gauge fields
A = if −1 ∂f, Ā = if¯∂¯f¯−1 with f (z, z̄), f¯(z, z̄) ∈ SU (NC )c . These type of vari-
ables were used frequently in dealing with gauged WZW actions, for instance
in computing the effective action of QCD2 and in the G G models discussed in
Section 4.7. They may be interpreted as Wilson lines along the z and z̄ direc-
tions. The gauged WZW part of the action, first line of (15.1), takes the form,
S1 (u, A) = S1 (f uf¯) − S1 (f f¯).
The Jacobian of the change of variables from A to f introduces a dimension

(1, 0) system of anticommuting ghosts (ρ, χ) in the adjoint representation of H.2
The WZW part of the action thus becomes

¯ ¯ i
S1 (u, A) = S1 (f uf ) − S1 (f f ) + d2 zTrH [ρD̄χ + ρ̄Dχ̄], (15.3)
2π
where Dχ = ∂χ − i[A, χ]. Our integration variables in the functional integral
are if −1 df and if¯df¯−1 . This action involves an interaction term of the form
TrH (ρ̄[f −1 ∂f, χ̄]) and a similar term for ρ, χ. By performing a chiral rotation,
like those of Chapter 14, ρ̄ → f −1 ρ̄f and χ̄ → f −1 χ̄f with ρ → f¯ρf¯−1 and χ →
f¯χf¯−1 , one achieves a decoupling of the whole ghost system. The price of this
(H )
is an additional S−2N C (f f¯) term in the action (here trace over H only) resulting
from the corresponding anomaly. This result can be derived by using a non-
abelian bosonization of the ghost system. A different bosonization of a (1,0)
ghost system was described in Section 6.5.
In this language the ghost action takes the form,
Sg h = SN c (l1 , A, Ā) + SN c (l2 , A, Ā) + Stwist (l1 ) + Stwist (l2 ),
where l1 and l2 are in the adjoint representation and Stwist is a twist term given,

Nc ¯ −1 f −1 ∂f ].
Stwist = − d2 zTr[l∂l (15.4)
2π
2 The ghost action was introduced in [8].

15.2 The action 281
Now using the Polyakov–Wiegmann formula we get,
Sgb h = SN c (f l1 f¯) − SN c (f f¯) + Stwist (l1 ) + SN c (f l2 f¯) − SN c (f f¯) + Stwist (l2 )

i
= −S2N c (f f¯) + d2 zTrH [ρ̄ ∂ χ̄ + ρ ∂χ
¯ ], (15.5)
2π
where the last line has been transferred back to the ghost language. Notice that
unlike the ghost fields in (15.3) the new ghost fields ρ and χ are gauge invariant.
It is interesting to note that the action given in (15.5) is non-local in terms of
the local degrees of freedom A and Ā. Note that had we done the right and left
rotations separately, we would have got S−2N c (f ) + S−2N c (f¯), which however is
not vector gauge invariant, but rather a left–right symmetric scheme.
The full gauge invariant action including the anomaly contribution of the anti-
commuting part now reads,

1 2 m2
SQ C D 2 = S1 (u) + 2 2
d zTrH F + d2 z : TrG f −1 uf¯−1 + f¯u−1 f :
ec 2π

(H ) i ¯

¯
+ S−(N F +2N C ) (f f ) + 2
d zTrH ρ̄ ∂ χ̄ + ρ ∂χ . (15.6)
2π
In deriving eqn. (15.6) we used a redefinition f uf¯ → u. This does not require an
extra determinant factor. Also, as S (H ) (f f¯) involves TrH rather than the TrG
in S1 (f f¯) of eqn. (2.4), a factor of Nf appears. Note that had we introduced
the special parameterization of the gauge fields in the fermionic formulation of
QCD2 , we would have arrived at the same action after decoupling the fermionic
currents from the gauge fields, by performing chiral rotation and then bosoniz-
ing the free fermions. Equation (15.6) was derived without paying attention to
possible renormalizations. The latter will be treated in Section 15.7.
At this point one may choose a gauge. A convenient gauge choice is
Ā = if¯∂¯f¯−1 = 0. Notice that since the underlying space-time is a plane this
is a legitimate gauge. The gauge fixed action can be written down using the
BRST procedure, namely,
SG F = SQ C D 2 + S (g f ) + S (g h) = SQ C D 2 + δB R S T (bĀ) =
= SQ C D 2 + TrH [B Ā] + TrH [bD̄c], (15.7)
where SG F , S (g f ) and S (g h) are, respectively, the gauge fixed action, the gauge
fixing term and the ghost action. The (b, c) fields are yet another (1, 0) ghost
system and B is a dimension-one auxiliary field, all in the adjoint representation
of SU (NC ). The integration over B introduces a delta function of the gauge
choice to the measure of the functional integral. In addition we integrate over
the ghosts b and c.
It is interesting to note that the QCD2 action can be related to a “perturbed”
topological HH coset model. To realize this face of QCD2 we parameterize u as

4π
i NCNF φ
ghle and rewrite (15.7) accordingly. The Polyakov–Wiegmann relation

implies,

1
S[u] = S[ghl] + d2 x∂μ φ∂ μ φ,
2

1
S[ghl] = S[g] + S[l] + S[h] + d2 xTr(g † ∂+ gl∂− l† + h† ∂+ hl∂− l† ). (15.8)
2π
Since l is a dimension-zero field with an associated zero central charge we have
S[l] = 0 and thus,

(H ) i ¯
SG F = SN F (h) + S−(N F + 2N C ) (f ) + d2 zTrH [ρ̄∂ χ̄ + ρ∂χ]
2π

1 ¯
+ SN C (g) + d2 z[∂φ∂φ]
2π

m2 2 −1 i N 4πNF φ −i N 4 π N F φ −1 −1 −1
+ d zTrG : f ghle C +e C l h g f :
2π

1 ¯ −1 ∂f ))2 ].
+ 2 d2 zTrH [(∂(f (15.9)
ec
It is now easy to recognize the first line in the action as the action of SS UU (N C)
(N C )
topological theory.
It is interesting to note that a WZW term S−2N C (f ) appears in the action
even without the introduction of quarks. We therefore digress to an analysis of
the pure YM theory in the formulation introduced above.
15.3 Two-dimensional Yang–Mills theory

Pure Yang–Mills theory has attracted much attention recently along the lines of
an underlying string theory. Here we restrict our discussion to the 2D Minkowski
or Euclidean space-time, where the rich structure of the model on a compact
Riemann surface does not show up. In terms of the parameterization introduced
in eqn. (15.6) the gauge invariant action of the pure YM theory is,
SY M 2 = S−(2N C ) (f f¯) + 2π
i ¯
d2 zTrH [ρ̄∂ χ̄ + ρ∂χ]
+ e12 d2 zTrH [F 2 ]. (15.10)
c
Here again we remind the reader that the coupling constant undergoes a mul-
tiplicative renormalization. This will be discussed in Section 15.7. Let us first
discuss the corresponding equations of motion for f and f¯,
δf : ∂A¯ − DĀ + 2 DD̄F = 1 + 2 DD̄ F = 0,

m2A m2A
2 2
δ f¯ : ∂ Ā − D̄A − 2 D̄DF = − 1 + 2 D̄D F = 0, (15.11)
mA mA

where D = ∂ − i[A, ·], mA = ec NπC and ¯ In fact these two equations
= 2∂ ∂.
are identical, as [D, D̄]F = 0. The equation is that of a massive gauge field with
self interaction. Note that in this approach, unlike the equations that follow from

15.4 Schwinger model revisited 283
varying the action with respect to the gauge fields, one gets two derivatives of
F. In deriving the above, it is convenient to remember that,
1 : ;
δSW Z W (f ) = Tr (f −1 δf )∂(f
¯ −1 ∂f ) .
2π
The YM action equation (15.10) is obviously invariant under the original gauge
transformations,
f → f v(z, z̄) f¯ → v −1 (z, z̄)f¯,
with v ∈ SU (NC ). In addition the action is invariant separately under the holo-
morphic and anti-holomorphic “color” transformations,
f → u(z̄)f f¯ → f¯w(z),
where u, w ∈ SU (NC ). These are “spurious” transformations since they leave
A and Ā invariant. The corresponding holomorphic and anti-holomorphic color
currents are,
¯ f¯)−1 −
J¯s = − Nπc [i(f f¯)∂(f 2
f D̄F f −1 ],
m 2A
J s = − Nπc [i(f f¯)−1 ∂(f f¯) + 2

m 2A
f¯−1 DF f¯]. (15.12)
The gauge fixed (f¯ = 1) action takes the form,

1 ¯ −1 ∂f ))2 ]
SY M 2 = S−(2N C ) (f ) + 2 d2 zTrH [(∂(f
ec

i ¯
+ d2 zTrH [ρ̄∂ χ̄ + ρ∂χ]. (15.13)
2π
As is expected the equation of motion at present is just that of eqn. (15.11)
after setting Ā = 0. Naturally, the action now lacks gauge invariance, neverthe-
less, it is invariant under the following residual holomorphic transformations,
f → u(z̄)f f → f w(z),
with the corresponding holomorphic and anti-holomorphic currents,

NC 2 ¯ ¯ NC 2 ¯
JG = − A + 2 D(∂A) JG = − Ã + 2 D̃(∂ Ã) ,
π mA π mA
¯ −1 . Notice that in spite of the similar structure, Ã is not related
and Ã = if ∂f
to Ā which was set to zero. To better understand the physical picture behind
these currents we defer temporarily to the abelian case.
15.4 Schwinger model revisited

Since in the pure Maxwell theory there is no analog to the (−2NC ) level WZW
term of eqn. (15.10), we study instead the Schwinger model in its bosonized form,
√ √
1 ¯ − 2∂X Ā + 2∂XA
¯ π ¯ 2 .
S(Sch) = d2 z ∂X ∂X + 2 (∂ Ā − ∂A)
2π e

In analogy to the change of variables in the non-abelian case, we now introduce

the following parameterization of the gauge fields A = ∂ϕ, Ā = ∂¯ϕ̄. In terms of
these fields the action takes the form,
2 < √ =
S(Sch) =
d z ¯ − 2X∂ ∂(ϕ
[∂X ∂X ¯ − ϕ̄) + π [∂ ∂(ϕ
¯ − ϕ̄)]2 + i[ρ̄∂ χ̄ + ρ∂χ]
¯ .
2π e2
In the gauge Ā = 0 and after the field redefinition X̃ = X + √1 ϕ, the action is
2
decomposed into decoupled sectors,
S(Sch) = S(X̃) + S(ϕ) + S(ghost) ,
* +
1 ¯ 1 2 ¯ 2 ¯
S(X̃) = 2
d z[∂ X̃ ∂ X̃] S(ϕ) = 2
d z [∂ ∂ϕ] − ∂ϕ∂ϕ , (15.14)
2π 4π μ2
e2
where μ2 = π . The corresponding equations of motion are,

2 ¯
∂ ∂ 1 + 2 ∂ ∂ ϕ = 0, ∂ ∂¯X̃ = 0.
¯
μ
The invariance under the chiral shifts δϕ = (z̄) and δϕ = (z) are generated by
the holomorphically conserved currents,
2 ¯ ¯ + 2 ∂ ∂¯∂ϕ.
¯
JG = ∂ϕ + ∂ ∂∂ϕ, J¯G = ∂ϕ
μ2 μ2
To handle this type of “hybrid” current we suggest the following decomposition
of the massless and massive modes ϕ = ϕ1 + ϕ2 with,
¯ 1 = 0 [2∂ ∂¯ + μ2 ]ϕ2 = 0.
∂ ∂ϕ
In the holomorphic quantization,
δL 1 μ2 1 ¯
Π= = − 2 ∂¯ ∂ ∂¯ + ϕ= ∂(ϕ2 − ϕ1 ). (15.15)
δ(∂ϕ) πμ 4 4π
A unique solution to the commutation relations [ϕ(z, z̄), Π(w, w̄)]z =w =
iδ(z̄ − w̄), [ϕ, ϕ] = 0 and [Π, Π] = 0 is,
[ϕ1 (z, z̄), ϕ1 (w, w̄)]z =w = πi(z̄ − w̄)
[ϕ2 (z, z̄), ϕ2 (w, w̄)]z =w = −πi(z̄ − w̄)
[ϕ1 (z, z̄), ϕ2 (w, w̄)]z =w = 0

X̃(z, z̄), X̃(w, w̄) = −πi(z̄ − w̄), (15.16)
z =w
where is the standard antisymmetric step function. Notice that the massless
degree of freedom has commutation relations which correspond to a negative
metric on the phase space. These relations can also be translated to the following
OPEs (choosing the part ϕ1 (z) of ϕ1 ),
ϕ1 (z)ϕ1 (w) = log(z − w),
ϕ2 (z, z̄)ϕ2 (w, w̄) = − log |(z − w)|2 + O(μ2 |z − w|2 ),
ϕ1 (z)ϕ2 (w, w̄) = O(z − w). (15.17)

15.5 Back to the YM theory 285
It is thus clear that the model is invariant under a U (1) affine Lie algebra of level
k = −1 since JG (z)JG (w) = (z −w 1
) 2 , as JG = ∂ϕ1 with no contribution from ϕ2 .
The physical states of the model have to be in the cohomology of the BRST
charge. Due to the fact that the current is holomorphically (and the other anti-
holomorphically) conserved, it follows that the same property holds for the BRST
charge, and thus the space of physical states is an outer product of the cohomol-
ogy of Q and Q̄. The latter are given by,
Q = χJ = χ(i∂ X̃ + ∂ϕ1 ),
Q̄ = χ̄J¯ = χ̄(−i∂¯X̃ + ∂ϕ
¯ 1 ). (15.18)
Expanding the fields i∂ X̃ and ∂ϕ1 in terms of the Laurent modes X̃n and
(ϕ̃1 )n with [Xn , Xm ] = nδn +m and [(ϕ1 )n , (ϕ1 )m ] = nδn +m we have

Q= χ̃n X̃−n − i(ϕ1 )−n .
n
Since J0 = {Q, ρ0 }, physical states have to have a zero eigenvalue of J0 . The

general structure of the states in the ϕ1 , X̃, ρ, χ Fock space is,
(X̃n )n X (ϕ1 m )n f (χk )n χ (ρl )n ρ |vac ,
where obviously nχ and nρ are either 0 or 1. It is straightforward to realize that

only the vacuum state and states of the form (X̃0 )n X (ϕ1 0 )n f are in the BRST
cohomology. Recall that being on the plane we exclude zero modes and thus only
the vacuum state remains. Since there is no constraint on the modes of ϕ2 , the
physical states are built solely of ϕ2 which are massive modes. This result is
identical to the well-known solution of the Schwinger model.
15.5 Back to the YM theory

Equipped with the lesson from the Schwinger model we return now to the YM
case and introduce a decomposition of the group element f so that again the
gauge currents obey an affine Lie algebra. Let us write f = f2 f1 which implies
that,
A = if −1 ∂f = if1−1 ∂f1 + if1−1 (f2−1 ∂f2 )f1 ≡ J1 + J2

¯ 1 = 0 implying also ∂J
With no loss of generality we take ∂f ¯ 1 = 0. Inserting these
expressions into JG of eqn (15.3) one finds,

NC 2 ¯ 2 + i[∂J
¯ 2 , J1 + J2 ]) .
JG = − J1 + J2 + 2 (∂ ∂J
π mA
If one can consistently require that,
2 ¯ 2 + i[∂J
¯ 2 , J1 + J2 ]) = 0,
J2 + (∂ ∂J
m2A

then, in a complete analogy with the abelian case, JG = − Nπc J1 . The lat-
ter is an affine current of level k = −2NC . One can in fact show that
(15.5) can be assumed without a loss of generality. ∂J ¯ G = 0 implies that
2 ¯ ¯
J2 + m 2 (∂ ∂J2 + i[∂J2 , J1 + J2 ]) = u(z), where u(z) is some holomorphic func-
A
tion. We then introduce the shifted currents J˜2 = J2 − u(z), J˜1 = J1 + u(z). Now
∂¯J˜1 = 0 as does J1 , and J˜2 obeys eqn (15.5) with J˜1 replacing J1 . It is easy to
check that the shifts in the currents correspond to f1 → v(z)f1 , f2 → f1 v(z)−1
with u(z) = if1−1 (v(z)−1 ∂v(z))f1 .
Note that the equation for J2 involves a coupling to J1 . This is related to
the fact that, unlike the abelian case, one cannot write the action as a sum of
decoupled terms which are functions of J1 and J2 separately.
Once the color current JG is expressed in terms of the holomorphic current
J1 , the analysis of the space of physical states is directly related to that of the
topological GG model at k = 0. The physical states have to be in the cohomologyof
the BRST charge, which corresponds to the following holomorphically conserved
BRST current,
a
1 NC a 2 ¯ i a b c
Q(z) = χa JGa + Jgah = − χ A + 2 D(∂A) + fbc ρ χ .
2 π mA 2
An anti-holomorphic BRST current Q̄(z) determines the condition for physical
states in the analogous manner to Q. From here on we restrict our description
to the latter. We define now the zero level affine Lie algebra current,
a
J(tot) = JGa + J(g
a a a
h) = JG + i, fbc χb ρc ,
and the c = 0 Virasoro generator T ,

1
T (z) = − : JGa JGa : +ρa ∂χa ,
NC
as well as dimension (2,0) fermionic current,
1
G=− ρa JGa ,
2NC
and realize the existence of the “topological coset algebra”,
: ;
T (z) = {Q, G(z)} , Q(z) = Q, j # (z) , J(tot)a
= {Q, ρa (z)} ,
{Q, Q(z)} = 0, {G, G(z)} ≡ W (z),
W (z) = {Q, U (z)} , [W, W (z)] = 0, (15.19)
where J # = χa ρa is “ghost number current”,
1
W (z) = fabc JGa ρb ρc + ∂ρa ρa ,
4Nc
and,
1
U= fabc JGa ρb ρc .
12Nc

15.6 An alternative formulation 287
A direct consequence is that any physical state has to obey,
J(tot) 00 |phys> = 0, L0 |phys> = 0, W0 |phys> = 0, (15.20)
where J(tot) in , L̃n and Wn are the Laurent modes of J(tot) i the Cartan sub-algebra
currents, T and W , respectively. In fact the BRST cohomology of the present
model is a special case of the set of G/G models.
We therefore refer the reader to those works [9], [200], [229] and present here
only the result. On the plane where no ghost zero modes are allowed, the only
state in the cohomology is the zero ghost number vacuum state of J1 .
This state can be a tensor-product with oscillators of the massive modes of J2 .
Unlike the abelian case, JG does not commute with J2 so that in general the J2
modes are not obviously in the BRST cohomology. However, there is no reason
to believe that all the J2 modes will be excluded by the BRST condition. Those
J2 modes that remain are by definition color singlets.
This result contradicts previous results on Y M2 . Usually one believes that
pure gluodynamics on the plane is an empty theory since all local degrees of
freedom can be gauged away.
15.6 An alternative formulation

To get a better understanding of the subtleties of the Yang–Mills theory when
expressed in terms of A = if −1 ∂f , Ā = if¯∂¯f¯−1 , and for future application, we
compare now with another formulation of the theory. A similar approach will be
used in the discussion of generalized YM theories in Chapter 16. Consider the
following functional integral,
Z= DADĀDBeiS (A , Ā ,B ) ,
S = − d2 zT rH [ e1c F B + 14 B 2 ], (15.21)
where B is a pseudoscalar field in the adjoint representation. Obviously the

integration over B produces the usual Tr[F 2 ] action. It is also easy to realize
that the action is invariant under the ordinary gauge symmetry provided that
δB = i[, B]. In terms of the f variables after imposing the gauge f¯ = 1 one finds,

i −1 ¯ 1
SY M 2 = S−(2N C ) (f ) + d2 zTrH (f ∂f )∂B − B 2 + S (g h) ,
ec 4
i
where S (g h) = 2π ¯
d2 zTrH [ρ̄∂ χ̄ + ρ∂χ]. One should again bear in mind that
the coupling constant undergoes a multiplicative renormalization. This will be
discussed in the next section. Using Polyakov–Wiegmann we get,
SY M 2 (B) = Γ2N C (B) − 14 TrH [B 2 ]

−S(2N C ) (vf ) + S (g h) , (15.22)

¯ −1 ). The second line in (15.22) is

¯ = 2N C (iv ∂v
where Γk (B) = Sk (v) with e1c ∂B 2π
a c = 0 “topological system”. Since the underlying Minkowski space-time does
not admit zero modes we can safely integrate over the corresponding fields. We
can further pass from functional integrating over B to iv ∂v ¯ −1 . This involves
D̄
the insertion of det ∂¯ which will introduce a Γ−2N C (B) term with no additional
ghost terms. The functional integral (16.1) thus takes the final form,

Z = D[v]e−i ( 4 d z Tr H [B ] ) .
1 2 2
It is thus clear that in the present formulation there is no trace of the massive
“physical modes” discussed in the previous section.
15.7 The resolution of the puzzle

Encouraged by the result of the last section, we proceed now to reexamine the
steps that led to the unexpected massive modes in the pure Y M2 theory. In
particular, we would like to check whether in addition to the implementation of
proper determinants there is no coupling constant renormalization that has to
be invoked when passing to the quantum theory expressed in the f variables.
For this purpose we turn on again the matter degrees of freedom. We introduce
NF quarks in the fundamental color representation and explore the behavior of
the system in the limit NF → 0. Recall that the action of this model is given in
eqn. (15.9). Starting actually from eqn. (15.1), taking the massless limit, writing
A in terms of f in the action but still with A as an integration variable, and
using the formulation presented in the previous section, the path integral of the
colored degrees of freedom now reads

(co l)
Z (col) = [DA][DB][Dh]eiS

(col) NF ¯ −1 f −1 ∂f ]
S = SN F (h) + d2 zTrH [h∂h
2π

i −1 ¯ 1 2
2
+ d zTrH (f ∂f )∂B − B , (15.23)
ec 4
where we have also gone from u to h as in Section 15.2.
It was found out that quantum consistency imposes finite renormalization on
the coupling constant of the current-gauge field interaction.3 This renormaliza-
tion is expressed in the following equality,

¯ ≡ DAei [S k (f )+ 21π d 2 z Tr H [i(f −1 ∂ f ) J¯]
Z(J)

e ( −k ) 2 −1 ¯ (g h )
= Df ei[S k −2 N C (f )+ 2 π d z Tr H [(f ∂ f ) J ] D(gh)eiS

e ( −k )
iΓ J¯ (g h )
= e −k + 2 N C −k + 2 N C D(gh)eiS , (15.24)
3 The finite renormalization of the coupling was introduced by D. Kutasov in [146].

15.8 On bosonized QCD2 289
where k is an arbitrary level and Γk (L) = Sk (w) for L = iw∂w ¯ −1 . The renor-
e(−k −2N C ) k
malization factor e(k) has to satisfy e(k ) = k +2N C . In addition it is clear
from eqn. (15.24) that it has
to be singular at the origin. It can be shown that
e(k) takes the form e(k) = k + 2Nk
C
. Implementing this renormalization in our
case, eqn. (15.23) takes the form,
S (col) = SN F (h̃) + S−(N F + 2N C ) (f )

5
NF + 2NC 1 −1 ¯ 1 2
2
+ d zTrH i (f ∂f )∂B − B
NF ec 4
+ S (g h) , (15.25)
where h̃ = f h. After integrating the auxiliary field B the action becomes,
S (col) = SN F (h̃) + S−(N F + 2N C ) (f )

NF + 2NC ¯ −1 ∂f )]2 ] + S (g h) .
+ d2 zTrH [∂(f (15.26)
e2c NF
It is now straightforward to realize that the equation of motion which follows
the variation with respect to f is that of eqn. (15.11) where now mA =
from
ec N2πF . Thus, the coupling constant renormalization turns the massive modes
into massless ones in the case of pure YM theory (NF = 0). Notice that to reach
this conclusion it is enough to use the fact that e(k) has to be singular at k = 0
and the explicit expression of e(k) is really not needed. Following the arguments
presented in Chapter 5, it is clear that these states that became massless are not
in the BRST cohomology and thus not in the physical spectrum.
A somewhat similar derivation of the triviality of the model in the NF = 0
limit is the following. We integrate in eqn. (15.25) over the ghost fields and over
f , using again the coupling constant renormalization, and find,
* +
e2 N 2
− iS N F (v )+ c 4 F d 2 z Tr H [B 2 (v )]
Z = D[v]e .
It is now clear that the action vanishes at NF = 0 and hence again, on triv-
ial topology, the theory is empty. Notice, however, that the implementation of
renormalization modifies also the result of the previous section.
The final conclusion is that in both methods one finds that indeed the pure
YM theory has an empty space of physical states as of course is implied by the
original formulation in terms of A. We have demonstrated that in this formulation
it follows only after taking subtleties of renormalization into account.
15.8 On bosonized QCD 2

To resolve the puzzle of the YM theory we were led to analyze the color and
flavor sectors of QCD2 . The full bosonized QCD2 includes in addition the baryon
number degrees of freedom. The corresponding action is given by eqns. (15.6),

(15.7) or by eqn. (15.9). In the past the low-lying baryonic spectrum in the strong
m
coupling limit e cq → 0 was extracted using a semi-classical quantization. In this
chapter our analysis was based on switching off the mass term, mq = 0. This
limit cannot be treated by the semi-classical approach, as the soliton solution
is not there for mq = 0. In our case here one finds a decoupled W ZW action
for the flavor degrees of freedom SN C (g) and a decoupled free field action for
the baryon degree of freedom, in addition to the action of the colored degrees of
freedom which is given in eqn. (15.26) or eqn. (15.7). The general structure of a
physical state in this case is that of a tensor product of g and φ with the colored
degrees of freedom f , h and the ghosts. The structure of QCD2 which emerged
from the semi-classical quantization for mq = 0 involves g and φ only. In our case
here the f colored degrees of freedom acquire mass mA = ec N2πF while the h
degrees of freedom remain massless. In the limit ec → ∞ the f degrees of freedom
decouple. It is thus clear that one has to introduce the mass term which couples
the three sectors. The massless limit of QCD2 can then be derived by taking the
limit mq → 0 after solving for the physical states. Indeed, it was shown in the
limit of ec → ∞ that turning on mq = 0 results in a hadronic spectrum where
the flavor representation and the baryon number were correlated. The analysis
of the spectrum of the massive multi-flavor QCD2 in the approach of this work
remains to be worked out.
15.9 Summary and discussion

In this chapter we have analyzed 2D YM and QCD theories using a special
parametrization of the gauge fields in terms of group elements. In the mq = 0
case it enabled us to decouple the matter and gauge degrees of freedom. How-
ever, this formulation led, in a naive treatment, to unexpected massive modes.
Even though we did not present a full solution of the theory we had reason-
able arguments to believe that the BRST projection would not exclude these
modes. The fact that a similar approach to QED2 reproduced the known results
of the Schwinger model, enhanced the puzzling phenomenon. Eventually, we
showed that a coupling constant renormalization, renders the unexpected mas-
sive modes into massless un-physical states. The benefit of this detective work is
the appearance of “physical” massive states in massless QCD2 .

16
Generalized Yang–Mills theory on
a Riemann surface
16.1 Introduction
Pure gauge theory in two dimensions is locally trivial and has no propagating
degrees of freedom. This was discussed in the first chapter of this part of the
book and now in the last chapter we will describe the global properties of gauge
theories in two dimensions. For the latter to be non-trivial we will either take
the underlying manifold to be a compact Riemann surface or introduce Wilson
loops external sources. We will show that in those cases the gauge theory has a
rich structure and in fact is almost a topological field theory, which is a theory
with no propagating degrees of freedom (see also Section 4.7). Moreover, it will
be shown that the theory has an interpretation in terms of a string theory.
It is easy to realize that in two dimensions the pure YM theory is in fact the
simplest member of a wide class of renormalizable theories that incorporate only
gauge fields. These will be referred to as the generalized gauge theories gYM.
In Chapter 15 we introduced an alternative formulation of the YM theory using
the action1

S = − d2 zTr[iFB + g 2 B 2 ]. (16.1)
Now it is easy to realize that the B 2 term can be generalized to an arbitrary

invariant function Φ(B). This will constitute the family of gYM.2 The partition
function of the generalized theories on Riemann surfaces and the computation
of Wilson loops for these theories will also be described in this chapter.
Pure YM2 theory defined on an arbitrary Riemann surface is known to be
exactly solvable. In one approach the theory was regularized on the lattice and,
using a heat kernel action, explicit expressions for the partition function and
loop averages were derived.3 Identical results were derived also in a continuum
path-integral approach.
In the following sections we briefly review the former derivation of the partition
function and determine in a similar way the results for the partition function and
1 In fact in (16.1) we have used a slightly different formulation which is however equivalent to
the one used in Chapter 15.
2 The notion of the generalized QCD theory in two dimensions was introduced and analyzed
in [82].
3 A lattice version of two-dimensional Yang–Mills theory was shown to be exactly solvable by
Migdal in [161]. Correlators of Wilson lines on this formulation were computed in [139].

292 Generalized Yang–Mills theory on a Riemann surface
Wilson loops in the gYM 2 case. Since this chapter is based on non-trivial two-
dimensional topology which has not been dealt with intensively in this book,
to fully understand its content the reader will need to consult the references to
this chapter. The reader who is not interested in the topological aspects of two-
dimensional gauge dynamics may skip this chapter and proceed directly to the
third part of the book.
This chapter is based mainly on [115], [118], [119] and [105].
16.2 The partition function of the YM2 theory

The partition function for the ordinary Y M2 theory defined on a compact Rie-
mann surface M of genus H and area A is,

1
Z(N, H, λA) = [DAμ ]exp − 2 d2 x det Gμν trFμν F μν , (16.2)
4g Σ
where the gauge group G is taken to be either SU (N ) or U (N ), g is the gauge
coupling constant, λ = g 2 N , Gμν is the metric on M, and tr stands for the trace
in the fundamental representation.4
As was emphasized in Chapter 8 the pure YM theory defined on a flat
Minkowski space-time with trivial topology is empty since one can gauge
away the gauge fields. However this does not hold if the underlying manifold
M is topologically non-trivial. If M contains a non-trivial cycle γ such that
μ
tr[P e γ A μ dx ] = 1, where P stands for path-ordering, one cannot gauge Aμ away
along γ. Thus, the partition function depends on the topology of M and in fact
only on the topology and its area. This follows from the fact that the theory is
invariant under area preserving diffeomorphism. The field strength can be written
in the form Fμν = μν F and hence the action takes the form,

1
S=− 2 d2 x det Gμν F 2 . (16.3)
4g Σ

Apart from the area form d2 x det Gμν this action is independent of the metric
and is therefore invariant under area preserving diffeomorphism.
The lattice partition function defined on an arbitrary triangulation of the
surface, as described in Fig. 16.1, is given by,

ZM = dUl Z [U ], (16.4)
l

where l denotes a product over all links, U is the holonomy around a plaquette

U = l∈ Ul , and Z is a plaquette action. For the latter one uses a heat kernel
4 The partition function on any Riemann surface of the discretized theory was written down
in [184]. An identical result was found also in the continuum formulation [228].

U2
U3
U1
Fig. 16.1. A triangulation of the Riemann surface M. A group matrix is placed

on each link.
− g12 tr(U +U † )
action rather than the Wilson action (which is Z [U ] = e ), i.e.

Z [U ] = dR χR (U )e−tc 2 (R ) , (16.5)
R
where the summation is over the irreducible representations R of the group;

dR , χR (U ) and c2 (R) denote the dimension, character of U and the second
Casimir operator of R, respectively, and t = g 2 a2 with a2 being the plaquette
area. The character, which was also discussed in Section 3.5 in relation to the
ALA algebras, is defined here as χR (U ) ≡ trR [U ]. The holonomy U (its subscript
is omitted from here on) behaves as U ≈ 1 − iaF when a is small. Note that
the region of validity of (16.5) is not only a → 0 with F fixed, but actually also
a → 0 with F going to infinity as a−1/2 because this is the region for which the
exponential − 4t1 a2 Tr F 2 is of order unity.
We will briefly review the derivation that singles out (16.5) as a conve-
nient choice among the different lattice theories which belong to the same
universality class. Let us look for a function Ψ(U, t) that will replace the contin-
uum e− 4 t a Tr F .
1 2 2

The requirements which we impose on Ψ are:
1. As t goes to zero (and, therefore, for finite g also a goes to zero) we want the
holonomy to be close to 1,
Ψ(U, 0) = δ(U − 1). (16.6)
2. For any V ∈ G we have,
Ψ(V −1 U V, t) = Ψ(U, t). (16.7)
In other words Ψ is a class function.

3. Ψ satisfies the heat kernel equation,
⎧ ⎫
⎨∂ ⎬
− gab ∂ a ∂ b Ψ(U, t) = 0, (16.8)
⎩ ∂t ⎭
a,b
where gab is the inverse of the Cartan metric,
g ab = tr(ta tb ), (16.9)
which was defined and discussed in Section 3.2.1.
To see that (16.5) is an approximate solution to the heat-kernel equation we

note that any class function is a linear combination of characters. The differ-
entiation of a character in the direction of a Lie algebra element ta is given
by,
ik
∂ a 1 ∂ a 2 · · · ∂ a k χR (U ) = χR (U t(a 1 ta 2 · · · ta k ) ) + O(U − 1). (16.10)
k!
The notation χR (U ta 1 ta 2 · · · ta k ) stands for the trace of the multiplication of the
matrices which represent U and ta 1 , . . . , ta k in the representation R. The brackets
(· · ·) imply symmetrization with respect to the indices. The term O(U − 1) means
that the corrections are of the order of U − 1 ∼ aF ∼ t1/2 . Since,
1
gab ∂ a ∂ b χR (U ) ≈ − χR (U gab t(a tb) ) = −c2 (R)χR (U ), (16.11)
2
a,b a,b
we see that (16.5) is the correct answer up to terms of the order of O(t3/2 ) which
drop in the continuum limit. Using (16.5) as the starting point, we finally find
the following form of the partition function,
λ A c 2 (R )
2−2H − 2 N
Z(N, H, λA) = dR e . (16.12)
R
To get from (16.5) to (16.12) we take the following steps. First we make use of
the additivity property of the heat-kernel action. Consider two triangles glued
along U1 as depicted in Fig. 16.2.

U3 U4
U1
U2 U5
Fig. 16.2. Integrating over U1 on a link which is common to the two triangles.
U1
V 2+ V1
U 2+ U +1
V2 V 1+
U2
Fig. 16.3. Opening up of a genus-two surface.
Using the orthogonality of characters,

χR 1 (V W )
dU χR 1 (V U )χR 2 (U † W ) = δR 1 ,R 2 , (16.13)
dim R1
we find,

dU1 Z1 (U2 U3 U1 )Z2 (U1† U4 U5 ) = Z1 +2 (U2 U3 U4 U5 ). (16.14)
This relation can be used to argue that the lattice representation is exact and
independent of the triangulation since using this we can add as many trian-
gles as desired, thus reaching the continuum limit. We can also use this rela-
tion to reduce the number of triangles to the minimum needed to capture
the topology of M. Describing a genus H manifold in term of a 4H-polygon
with identified sides as described in Fig. 16.2 for a genus-two Reimann surface
a1 b1 a−1 −1 −1 −1
1 b1 . . . aH bH aH bH . The partition function on such a manifold can be
written as,

λ A c 2 (R )
ZM = dR e − 2N
DUl DVl χR [U1 V1 U1† V1† . . . UH VH UH† V1† ]. (16.15)
R

We can simplify this expression using again the orthogonality of the characters
and the relation,

1
DU χR [AU BU † ] = χR [A]χR [B], (16.16)
dimR
to arrive at (16.12).
16.3 The partition function of gYM2 theories

Pure YM 2 theory is in fact a special representative of a wide class of 2D gauge
theories which are invariant under area preserving diffeomorphisms. These gener-
alized YM 2 theories are described by the following generalized partition function,

Z(G, H, A, Φ) = [DAμ ][DB] exp[ d2 x det Gμν tr(iBF − Φ(B))], (16.17)
Σ
where F = F μν with ij being the antisymmetric tensor 12 = −21 = 1. B is
μν
an auxiliary Lie-algebra-valued pseudo-scalar field.5

We wish to generalize the substitution (16.5) for the plaquette action (16.17),

Z [U ] = DBetr{iaB F −tΦ(B )} → Ψ(U, t).
?
(16.18)
Here B is a Hermitian matrix and Φ is an invariant function (invariant under

B → U −1 BU for U ∈ G). The quadratic case Φ(X) = g 2 tr(X 2 ) obviously corre-
sponds to the YM 2 theory. We will take Φ to be of the form,

Φ(X) = a{k i } tr(X i )k i , (16.19)
{k i } i
(e.g. tr(X 3 )2 + tr(X 6 )) For SU (N ) (U (N )), tr(X i ) can be expressed for i ≥ N

(i > N ) in terms of tr(X i ) for smaller i. Thus the summands in (16.19) are
not independent. This does not affect the following discussion. Moreover, in the
large N limit that we will discuss in the following section, the terms do become
independent.
We define the general structure constants dabc...k to be,
def
dabc...k = gaa gbb gcc · · · gk k tr(ta tb tc · · · tk ). (16.20)
For every partition r1 + r2 + · · · + rj , we define the Casimir,
def
C{r 1 +r 2 +···+r j } =
1 (1) (1) (2) a(j ) )
da ( 1 ) ...a ( 1 ) da ( 2 ) ...a ( 2 ) · · · da ( j ) ...a ( j ) t(a 1 · · · ta r 1 ta 1 · · · t r j
(r1 + r2 + · · · + rj )! 1 r1 1 r2 1 rj
(16.21)
5 1
In principle, we could perturb the ordinary Y M 2 with operators of the form g 2 k −2
tr(F k ),
without the need for an auxiliary field.

16.4 Loop averages in the generalized case 297
Note that the index of C{·} will always pertain to a partition. Thus C{p} =
C{r 1 +r 2 +···+r j } even if p = r1 + r2 + · · · + rj . The brackets in the t-s mean a
total symmetrization ( (r1 + r2 + · · · + rj )! terms).
Cρ can easily be seen to commute with all the group elements and so, by
Schur’s lemma, is a constant matrix in every irreducible representation.
We claim that the correct lattice generalization of (16.5) is,

dR χR (U )e−tΛ(R ) , (16.22)
R
where,

Λ(R) = a{k i } C{k 1 ·1+k 2 ·2+k 3 ·3+···} (R). (16.23)
{k i }
This results from the requirements that Ψ(U, t) must satisfy:

1. Ψ(U, 0) = δ(U − 1).
2. Ψ is a class-function.
3. Ψ satisfies the equation,
⎧ ⎫
⎨∂ ∂l
kl ⎬
− a{k j } (ia)−l da 1 a 2 ...a l Ψ(U, t) + O(U − 1) = 0
⎩ ∂t ∂Fa 1 · · · ∂Fa l ⎭
{k j } l
(16.24)
For the U s that are important in the weight for a single plaquette, U − 1 is
of the order of magnitude of aF which, in turn, is of the order of magnitude
of O(t1/ν ) where ν is the maximal degree of Φ. Thus, the corrections to Ψ are
O(a−(1+ 1/ν ) ) and drop out in the continuum limit.
The partition function for the generalized Y M2 theory is therefore,

(dim R)2−2H e− 2 N Λ(R ) ,
λA
Z(G, ΣH , Φ) = (16.25)
R
where A is the area of the surface and Λ(R) is defined in (16.23).
16.4 Loop averages in the generalized case

The full solution of the YM 2 theory includes, in addition to the partition func-
tion, closed expressions for the expectation values of products of any arbitrary
number of Wilson loops,
n
i A dx
W (R1 , γ1 , . . . Rn γn ) = < TrR i Pe γi
>, (16.26)
i= 1
where the path-ordered product around the closed curve γi is taken in the rep-
resentation Ri . Using loop equations, one can derive an algorithm to compute
Wilson loops on the plane. This can be further generalized into a prescription for

c1 = 6 c2 = 6 c3 = 4 c4 = 2 c5 = 2 c6 = 1 c7 = 1 c8 = 1
n1 = 8
n2 = 8
n3 = 3
n4 = 3
n5 = 2
n6 = 2
Fig. 16.4. Wilson loops on a torus.
computing those averages for non-intersecting loops on an arbitrary two man-

ifold. Let us briefly summarize the latter. One cuts the 2D surface along the
Wilson loop contours, see Fig. 16.4, forming several connected “windows”. Each
window contributes a sum over all irreducible representations of the form of
(16.12). In addition, for each pair of neighbouring windows, a Wigner coefficient,

DR 1 R 2 f = dU χR 1 (U )χR 2 (U † )χf (U ), (16.27)
is attached. Altogether, one finds,
1 1 Nw
−λ A i C 2 ( R i )
W (R1 , γ1 , . . . Rn γn ) = .... DR 1 ...R n d2−2G i
e 2N ,
ZN n
i=1
Ri
R1 Rn
(16.28)
where Nw is the number of windows, 2 − 2Gi is the Euler number associated with
the window i and DR 1 ...R n is the product of the Wigner coefficients for neigh-
boring windows. For the case of intersecting loops a set of differential equations
provides a recursion relation by relating the average of a loop with n intersections
to those of loops with m < n intersections.
Generalizing these results to the gYM 2 is straightforward. The only alteration
−λ A C 2 ( R )
that has to be invoked is to replace the e 2 N factors that show up in those
algorithms with similar factors where the second Casimir operator is replaced
by the generalized Casimir operator (16.23). For instance the expectation value
of a simple Wilson loop on the plane is given by,
−λ A γ Λ ( R )
W (R, γ) = e 2N , (16.29)
where Aγ is the area enclosed by γ.

16.5 Stringy YM 2 theory 299
Fig. 16.5. Wilson loops on a torus.
It is interesting to note that for odd Casimir operators the expectation values
of real representations (like the adjoint representation) equal unity due to the
fact that the corresponding Casimirs vanish.
16.5 Stringy YM2 theory

Before dwelling into the stringy description of the generalized YM 2 thoeries,
we first review the proof of the stringy nature of pure YM 2 theory.6 We then
generalize this construction to the generalized YM 2 and present several examples
to demonstrate the nature of the maps that contribute to the gYM 2 and their
weights.
The partition function expressed as a sum over irreducible SU (N ) (U (N ))
representations (16.12) can be expanded in terms of powers of N1 . This involves
expanding the dimension and the second Casimir operator of the various rep-
resentations. The representation of U (N ) or SU (N ) are described by Young
tableau Y (R), see for instance Fig. 16.5, composed of r ≤ N horizontal lines
each with ni boxes so that n1 ≥ n2 ≥ . . . ≥ nr . The U (N ) and SU (N ) second
Casimir operators of a representation R and its dimension are given by,
U (N )
r
r
C2 (R) = N ni + ni (ni + 1 − 2i) = N n + 2P̂{2} (R)U (N ) (R),
i=1 i= 1

r
r r
S U (N ) ( i=1 ni )2
C2 (R) =N ni + ni (ni + 1 − 2i) − = N n + 2P̂{2} (R)S U (N ) ,
i=1 i= 1
N

i≤j ≤N (ni − i − nj + j)
dR = , (16.30)
i≤j ≤N (i − j)
6 The stringy description of Yang–Mills theory in two-dimensional Riemann surfaces was intro-
duced by D. Gross and W. Taylor in [115], [118] and [119]. The formulation of the two-
dimensional Yang–Mills theory in terms of topological string theories was done in [126] and
[69].

MH
Mh
Fig. 16.6. A map from Mh to MH .
r
with n = i= 1 ni . For every representation R there is a conjugate representa-
tion R̄ whose Young tableau Y (R̄) has its rows and columns interchanged. To
determine the Casimir operator of the conjugate representation we use (16.30)
with 2P̂{2} (R) = −2P̂{2} (R)U (N ) (R).
Using the Frobenius relations between representations of the symmetric group
Sn and representations of SU (N ) (U (N )), the coefficients of this asymptotic
expansion were written in terms of characters of Sn . The latter can be shown
to correspond to permutations of the sheets covering the target space. The final
result takes the form of,

∞ (n + + n − )(2 H −2 )+ (i + + i − )
1
Z(A, H, N ) ∼
N
n ± , i ± = 0 p ± , . . . , p ± ∈T 2 ⊂S ± s ± , t ± , . . . , s ± , t ± ∈S ±
1 i± n 1 1 H H n
+ −
(−1)(i +i ) − − − 2 + −
(λA)(i +i ) e− 2 (n +n )λA e 2 ((n ) +(n ) −2n n )λA /N
+ 1 + 1 + 2 2
+ − + −
i !i !n !n !
H H
+ − − 2−2H
1 · · · pi + p1 · · · pi − Ωn + ,n −
δS n + ×S n − p+ [s+ +
j , tj ] [s− −
k , tk ] , (16.31)
j =1 k =1
where [s, t] = sts−1 t−1 . Here δ is the delta function on the group algebra of
the product of symmetric groups Sn + × Sn − , T2 is the class of elements of Sn ±
consisting of transpositions, and Ω−1 n + ,n − are certain elements of the group algebra
of the symmetric group Sn + × Sn − .
The formula (16.31) nearly factorizes, splitting into a sum over n+ , i+ , · · · and
n− , i− , · · ·. The contributions of the (+) and (−) sums were interpreted as arising
from two “sectors” of a hypothetical worldsheet theory. These sectors correspond
to orientation reversing and preserving maps, respectively. One views the n+ = 0
and n− = 0 terms as leading order terms in a 1/N expansion. At higher orders
the two sectors are coupled via the n+ n− term in the exponential and via terms
in Ωn + n − .
Thus, the conventional Y M2 theory has an interpretation in terms of sums of
covering maps of the target space, see Fig. 16.5. Those maps are weighted by the
factor of N 2−2h e− 2 n λA where h is the genus of the world-sheet and A is the area
1
of the target space. The power of N 2−2h is obtained from the Riemann–Hurwitz
formula,
2h − 2 = (n+ + n− )(2H − 2) + (i+ + i− ), (16.32)

16.6 Toward the stringy generalized YM 2 301
where B = i+ + i− is the total branching number. The number of sheets above

each point in target space (the degree of the map) is n and λ = g 2 N is the
string tension. Maps that have branch points are weighted by a factor of λA.
The dependence on the area A results from the fact that the branch point can
be at any point in the target space.
16.6 Toward the stringy generalized YM2

Note that the stringy description of the YM 2 theory does not attribute any
special weight for maps that have branch points of a degree higher than one, nor
is there a special weight for two (or more) branch points that are at the same
point in the target space. The latter maps are counted with weight zero, at least
for a toroidal target space, since they constitute the boundary of map space.
The main idea behind a stringy behavior of the gYM 2 is to associate nonzero
weights to those boundary maps, once one considers the general Φ(B) case rather
than the B 2 theory. In other words, we anticipate that we will have to add for
the tr(B 3 ) theory, for example, maps that have a branch point of degree 2 and
count them, as well, with a weight proportional to A. From the technical point
of view the emergence of the YM 2 description in terms of maps followed from
a large N expansion of the dimensions and the second Casimir operators of
(16.12). Obviously a similar expansion of the former applies also for the gener-
alized models and therefore what remains to be done is to properly treat the
Casimirs appearing in the exponents of (16.25).
In [119], the expansion of the second Casimir operator C2 (R) of a represen-
tation R introduced the branch points and the string tension contributions to
the partition function. The C2 (R) was expressed in terms of the eigenvalue of
the sum of all the n (n2−1) transpositions of n elements (permutations containing
a single cycle of length 2), where n is the number of boxes in R. This is the
outcome of the following formula,
C2 (R) = nN + 2P̂{2} (R), (16.33)
where P̂2 (R) is the value of the scalar matrix representing the sum of transposi-

tions i< j ≤n (ij) in the representation R of Sn (the matrix commutes with all
permutations and thus is scalar). In the partition function, C2 (R) was multiplied
λA
by 2N . The resulting term 12 nλA arises from the action and is proportional to
the string tension. The term λAP̂{2} arises from the measure and is interpreted
as the contribution of branch points to the weight of a map.
Our task is, therefore, to express the generalized Casimirs Cρ of (16.21) in
terms of P̂ρ , the generalizations of P̂{2} (R). This is expressed as,
ρ
Cρ (R) = αρρ N h ρ P̂ρ (R), (16.34)
ρ


where αρρ are coefficients that are independent of R and the power factors hρρ ,
are adjusted so that a string picture is achieved.
The P̂{ρ } factors are associated with ρ which is an arbitrary partition of
certain numbers, namely,
k1 k2
B CD E B CD E

ρ : ki · i = 1 + 1 + · · · + 1 + 2 + 2 + · · · + 2 + · · · , (16.35)
i
P̂{ρ } (R) is the product of two factors. The first is the sum of all the permutations
in Sn (n is the number of boxes of R) which are in the equivalence class that
is characterized by having ki cycles of length i for i ≥ 2. Just like the case of
P̂{2} (R), the matrix P̂{ρ } (R) commutes with all permutations and thus is a
scalar. The sum is taken in the representation R of Sn . The second factor is,

n − i=2 iki
, (16.36)
k1
which can be interpreted later as the number of ways to put k1 marked points
on the remaining sheets that do not participate in the branch points.
16.7 Examples
A complete diagrammatic expansion of the operators was determined in [105].
Using this expansion one can write down the stringy description of the partition
function for any generalized YM theory. We end this chapter with a few examples
of the Casimir factors for various choices of Φ(B) for both U (N ) and SU (N )
groups.
λ
1. For N tr(B 2 ) which is the conventional Y M2 theory we get (16.33),
2λ
P̂{2} + λP̂{1} . (16.37)
N
The first term means that we give a factor of 2λA
N for each branch point, and
the second term means that we have a factor of λ for each marked point (i.e.
this is the string tension).
2. For αN −2 tr(B 3 ) in U (N ) we get,
1 1
3αN −2 P̂{3} + 3αN −1 P̂{2} + 3αN −2 P̂{1+1} + αP̂{1} + αN −2 P̂{1} . (16.38)
2 2
The first term is the contribution from branch points of degree 2 (the simple
branch points are of degree 1). The next term is a modification to the weight
of the usual branch points. The third is the weight of two marked points at
the same point (but different sheets), which will translate into n+ (n+ − 1)
+ n− (n− − 1) in the weight of a map for which (n+ , n− ) are the numbers
of sheets of each orientability. The last two terms are modifications to the

16.7 Examples 303
cosmological constant (or, in our terminology, to the weight of the single

marked point).
Note that, because of the original power of N −2 we do not get the usual
2−2h
N stringy behaviour in the partition function. We can overcome this
problem by interpreting the last term not as the usual marked point, but as
a microscopic handle that is attached to the point. By interpreting certain
marked points as actually being microscopic handles (or higher Riemann sur-
faces) we can always adjust the power of N to be N 2−2h . Similarly, we should
interpret the term 3αN −2 P̂{1+ 1} as a connecting tube. We will come back to
this interpretation towards the end of this section and investigate it further
in the next subsection.
3. For N −2 tr(B 3 ) in SU (N ) we obtain the following corrections to (16.38),
6 12 12
− P̂{2+ 1} − 3 P̂{2} + 4 P̂{1+1+1}
N3 N N
6 12 3 2
− 2
− 4 P̂{1+ 1} − 2
− 4 P̂{1} . (16.39)
N N N N
These terms and the terms in the previous example (16.38) do not mix chi-
ralities (i.e. sheets of opposite orientations). In the full theory (chiral and
anti-chiral sectors) there is the corresponding anti-chiral term:
6 12 12
+ 3
P̂{2̄+ 1̄} + 3 P̂{2̄} − 4 P̂{1̄+ 1̄+ 1̄}
N N N
6 12 3 2
+ − 4 P̂{1̄+ 1̄} + − 4 P̂{1̄} . (16.40)
N2 N N2 N
For SU (N ) there are additional terms that do mix chiralities. They are,
6 12
− 3
(P̂{2̄+ 1} − P̂{2+ 1̄} ) + 4 (P̂{1̄+ 1̄+1} − P̂{1̄+ 1̄+1} ). (16.41)
N N
The first term is the contribution of maps that have a branch point in one
orientability and a marked point in the other (at the same target space point).
The second term is the contribution of maps with three marked points – two
for one orientability and one for the other.
To illustrate the content of these formulae in terms of representations, we
will calculate the value of the third Casimir 16 dabc t(a tb tc) for a totally anti-
symmetric representation of SU (N ) with k boxes. The term P̂{3} is translated
into the sum of all the permutations of the k indices of a totally antisymmetric
tensor that are 3-cycles, this gives 13 k(k − 1)(k − 2). The term P̂{2} gives the
sum of all the permutations that are 2-cycles, that is − 21 k(k − 1) (a minus

sign comes from antisymmetry). All in all we get,

1
C{3} (k) ≡ dabc t(a tb tc)
6
3 3
= k(k − 1)(k − 2) − k(k − 1) + N k(k − 1)
2 2
1 1 2 3
+ k + N k + k(k − 1)(k − 2)
2 2 N
6 2 6 2k
− 3k(k − 1) + k(k − 1) − 3k + 2 k(k − 1)(k − 2) + 2 k(k − 1) + 2
N N N N
k
= (N + 1)(N + 2)(N − k)(N − 2k). (16.42)
2N 2
4. For N −3 tr(B 4 ) in U (N ) we get,
8
4N −3 P̂{4} + 6N −2 P̂{3} + 6N −3 P̂{2+1} + N −2 P̂{1+1}
3
4 −1 −3 1 5 −2
+ N + 6N P̂{2} + + N P̂{1} .
3 6 6
The terms that have an extra microscopic handle are,
8 5
6N −3 P̂{2+ 1} + N −2 P̂{1+ 1} + 6N −3 P̂{2} + N −2 P̂{1} .
3 6
5. For N −4 (tr(B 2 ))2 in U (N ) we get,
24N −4 P̂{3} + 8N −4 P̂{2+ 2} + 4N −3 P̂{2+1}

16 −3 8 2 −2 1 −4
+ N P̂{2} + 2 + N −4 P̂{1+1} + N + N P̂{1} . (16.43)
3 3 3 3
and additional terms,
8N −4 P̂{2+ 2̄} + 4N −3 (P̂{2̄+1} + P̂{2+ 1̄} ) + 2N −2 P̂{1+ 1̄} ,
that mix the two chiral sectors.

+ −
The meaning of the term 2N −2 P̂{1+ 1̄} is a factor of N −2 e−2α n n A , where
α is the coefficient of the N −4 (tr(B 2 ))2 term in the action. n+ is the number
of sheets of positive orientability and n− is the number of sheets of negative
orientability for a given map.
16.8 Summary
In this chapter we studied the generalized two-dimensional Yang–Mills theory
on Riemann surfaces. We reviewed the exact formulae for the partition function
and Wilson loop averages of the conventional YM theory. We then presented
the generalization of these results in the context of the generalized YM theories.
These expressions are based on a replacement of the second Casimir operator
with more general Casimir operators depending on the particular model. There

16.8 Summary 305
is another method to obtain these results [228], i.e. by regarding the general
Yang–Mills actions as perturbations of the topological theory at zero area.
Using the relations between SU (N ) representations and representations of the
symmetric groups Sn , we wrote down the generalizations that have to be made
in the Gross–Taylor string rules for 2D Yang–Mills theory, so as to make the
generalized Yang–Mills theory for SU (N ) or U (N ) a local string theory as well.
The extra terms are special weights for certain maps with branch points of a
degree higher than one.
An obvious extension of the results presented in this chapter is to consider
other gauge groups. The conventional YM 2 theory with gauge groups O(N ) or
Sp(N ) which were shown to be related to maps from non-orientable world-sheets.
One can further couple the gYM 2 theories to fermionic matter in analogy to
’t Hooft’s analysis presented in Chapter 10. This domain of research is far from
being fully explored. A particularly interesting question is to find out certain
Φ(B)s that lead to a special behaviour of the coupled system. For example,
in the U (1) case, the representations R are labeled by an integer n and for
Φ(B) = −α log(1 + λB 2 ) we get,

Z(U (1), A) = (1 + λn2 )−α A
n
which has a singularity for 2αA = 1.

PART III
From two to four dimensions

17
Conformal invariance in four-dimensional field
theories and in QCD
Conformal symmetry was shown to be extremely powerful in two-dimensional

field theories and obviously also in string theories. This is due to the fact that
the conformal algebra is infinite dimensional and hence supplies a set of infinitely
many conserved charges. In four dimensions the conformal algebra is finite and
therefore less powerful.1 The purpose of this chapter is to explore the use in
4d conformal field theories of notions and tools that we encountered in two-
dimensional CFT, such as primary fields, conformal operator expansion, confor-
mal anomalies and Ward identities.
Free massless theories are obviously scale and conformal invariant. However,
field theories that describe the elementary particles of nature and their interac-
tions, are interacting field theories. The question is thus, to what extent can one
apply the techniques of CFT to those theories and in particular QCD in four
dimensions? QCD with massless quarks is a prototype model of theories which
are classically conformal invariant. In fact even for theories with masses and
other dimension-full parameters, in certain cases these can be neglected in the
high energy and high momentum transfer regime of the theory. However, even in
the massless case and with only dimensionless couplings, it is easy to realize that
the quantum picture lacks conformal invariance. This follows from the fact that
one has to introduce dimension-full parameters such as UV cutoff, which turns
into a scale where the coupling is defined, after renormalization. Thus there is
an anomaly in the conformal symmetry in the sense that it is a symmetry of the
classical system but not of the quantum one.
We will investigate in this chapter conformal symmetry in four-dimensional
field theories and in particular its applications in the context of QCD in four
dimensions. We start with the description of conformal transformations, their
corresponding generators and the SO(2, 4) conformal algebra. We then analyze
the Noether currents that follow from the conformal transformation and their
conservation laws. Next we present the SL(2, R) collinear subgroup associated
with light-cone conformal transformations. In a similar manner to the treatment
of two-dimensional conformal symmetry, we define primary and descendant oper-
ators of the collinear group. We then define and study the conformal operator
product expansion (COPE). We proceed and describe the conformal Ward iden-
tity, the Callan–Symanzik equation. We then make use of the conformal toolbox
1 The full conformal algebra in four dimensions was introduced in [156]

310 Conformal invariance in four-dimensional field theories and in QCD
in the study of four dimensional QCD which is conformal only at the classical
level. We analyze the non-local operators built from a quark and an anti-quark
and expand them in terms of Gegenbauer polynomials. We use the COPE to write
down the operator product of two electromagnetic currents. Finally we deter-
mine, in the limit of large momentum exchange, the pion distribution amplitude.
Conformal invariance in four dimensions was described in several review papers
and books. The original studies on conformal symmetry are summarized in [207]
and [66]. A modern review that we follow in this chapter is [43].
17.1 Conformal symmetry algebra in four dimensions

In general in d space-time dimensions the conformal group is the subgroup of
coordinate transformations that leaves the metric invariant up to a scale, namely,

gμν (x) → gμν (x ) = Ω(x)gμν (x). (17.1)
It is obvious from (2.2) that the 2d conformal transformations (2.1) indeed pro-
duce such a variation of the metric. An important property of conformal trans-
formations in any dimension is that they preserve the angle √AA·2BB 2 between two
vectors A and B.
Starting from flat space-time, the general infinitesimal coordinate transforma-
tion xμ → xμ + μ (x) induces a change of the metric,
ds2 → ds2 + (∂μ ν + ∂ν μ )dxμ dxν , (17.2)
so that the condition for conformal transformations reads,
2
∂μ ν + ∂ν μ = (∂ · )gμν , (17.3)
d
where gμν is ημν or δμν for a Minkowskian signature or Euclidean signature,
respectively, and the factor of d2 is fixed by tracing both sides of the equation
with g μν . To check what are all the possible solutions for μ we differentiate (2.4)
twice to find that,
[(d − 2)∂μ ∂ν + gμν ∂α ∂ α ]∂β β = 0. (17.4)
This equation together with (2.4) implies that the third derivatives of μ vanish,
which means that μ can be of order 0, 1, 2 in xν . Obviously the parameters
associated with the Poincare group, since they are an isometry and hence do
not change the metric, obey (17.1). These transformation parameters together
with additional infinitesimal parameters which are linear and quadratic in xμ
are summarized as follows:
μ = μ0 space-time translations
μ
= μν0 xν Lorentz transformations
μ
= 0 xμ scale transformations
μ
= ˜μ0 x2 − 2xμ ˜ν0 xν special conformal transformations (17.5)

17.1 Conformal symmetry algebra in four dimensions 311
where μ0 , μν ˜μ0 are vector, antisymmetric tensor, scalar and vector infinites-
0 , 0 ,
imal constants, respectively. The corresponding finite transformations are,
xμ → xμ = xμ + aμ
xμ → xμ = aμν xν
xμ → xμ = axμ
xμ + ãμ x2
xμ → xμ = , (17.6)
1 + 2ãμ xμ + ã2 x2
where the various forms of a are the finite parameters of transformation that
correspond to the infinitesimal ones above. The last transformation is referred
to as the special conformal transformation. It can in fact be decomposed into
μ
an inversion transformation xμ → − xx 2 , a space-time shift transformation and
another inversion. The sum of all these transformations is d + d(d−1)
2 +1+d=
(d+ 1)(d+ 2)
2 , which is the dimensions of SO(2, d), the algebra of the conformal
group in Minkowski space-time.
Let us analyze now the generators of the conformal transformations and show
that indeed they obey the SO(2, d) algebra. The generators are,
Pμ = −i∂μ space-time translations

Lμν = i(xμ ∂ν − xν ∂μ ) Lorentz transf ormations
D = −ix ∂μ μ
scale transf ormation

Kμ = −i x2 ∂μ − 2xμ xν ∂ν special conf ormal transf ormations (17.7)
Using these expressions for the generators it is a straightforward exercise to

realize that they obey the following algebra,
[Pμ , Pν ] = 0
[Pμ , Lν ρ ] = i(ημν Pρ − ημρ Pν )
[Lμν , Lρσ ] = −i(ημρ Lν σ − ημσ Lν ρ + ην σ Lμρ − ην ρ Lμσ )
[D, Pμ ] = −iPμ [D, Kμ ] = iKμ
[Pμ , Kν ] = 2i(Lμν − ημν D)
[D, Lμν ] = 0
[Kμ , Kν ] = 0, (17.8)
which is indeed the SO(2, 4) algebra. The first three lines constitute the Poincare
algebra in four dimensions. It is well known that (17.8) is not the most general
form of the SO(2, 4) algebra. One can further generalize the construction of the
generators by modifying Lμν in the following way,
Mμν = Lμν + Σμν , (17.9)
where Σμν does not act on the space-time points and obeys,
[Σμν , Σρσ ] = −i(ημρ Σν σ − ημσ Σν ρ + ην σ Σμρ − ην ρ Σμσ ). (17.10)

Shortly the role of these generators in the conformal transformation of fields will
be discussed.
17.2 Conformal invariance of fields, Noether currents and

conservation laws
So far we have discussed the conformal transformations as they act on the points
in space-time. Now we would like to consider field theories in four dimensions
that are classically invariant under the group of conformal transformations. The
fields associated with conformal invariant theories may be scalar fields, spinors,
vectors or tensors. We denote such a field by Φ(x).
The transformation of the field under the conformal transformations,
δΦ(x) = δxμ ∂μ Φ(x) + δI Φ(x), (17.11)
is composed of two parts, the one due to that of the space-time point with
δxμ given in (17.5) and an “internal transformation” δI Φ(x), which vanishes
for space-time translations, while for Lorentz transformations, dilatations and
special conformal transformations, takes the form,
δI Φ(x) = −eμν Σμν Φ Lorentz transformations

δI Φ(x) = e lΦ scale translations
δI Φ(x) = 2ẽμ (lx − xν Σμν )Φ
μ
special conformal transformations
(17.12)
where l is the conformal dimension of the field and the internal Lorentz generators
Σμν are given by,
i μ ν
Σμν = [γ , γ ] − Dirac spinors [Σμν ]βα = η μβ δαν − η ν β δαμ − gauge f ields
4
(17.13)
Recall that all the parameters of transformations are global, namely space-time
independent. To determine the Noether currents associated with the various
symmetry transformations, one elevates the transformations into local ones and
reads the currents from the variation of the action,

δS = d4 xJμa ∂ μ ea (17.14)
where ea is any of the parameters of transformations given in (17.5). The outcome

of the Noether procedure are the following conserved currents,
μ
J (P ) ν ≡ Tνμ = Πμ ∂ν Φ − δνμ L
μ
J (M ) ν ρ = xν Tρμ − xρ Tνμ − Πμ Σν ρ Φ
μ
J (D ) ≡ Dμ = xν T μν + lΠμ Φ
μ
J (K ) ν = (2xρ xν − ην ρ x2 )T μρ + 2xρ Πμ (lδνρ − Σρν )Φ, (17.15)

17.2 Conformal invariance of fields 313
∂L
where Πμ = ∂ (∂ μ Φ) . In fact the variation of the action with respect to dilatations
μ
may lead in addition to the divergence of J (D ) , to another total derivative
term ΔD . However for Lagrangians that are polynomials in the fields and their
derivatives, this term vanishes. For the special conformal transformations the
additional term is defined by

δK S = d4 xeμ (x) [−∂ν K μν + 2xμ ΔD + Δμk ] , (17.16)
with Δμk = 2Πν Φ(lg μν + Σμν )Φ. For invariance we need ΔD = 0 and Δμk =
μ
2∂ν σ μν . For l = 1 and l = 3/2, σμν vanishes, so in these cases J (K ) ν are really
the generators of conformal transformations.
An interesting observation is that all the Noether currents associated with the
full conformal group can be expressed in terms of a modified energy-momentum
tensor. First note that the energy-momentum tensor defined above in not
μ
necessarily symmetric. In fact from the conservation of J (M ) ν ρ the antisymmetric
part of Tμν can be determined since,
μ
∂μ J (M ) ν ρ = Tρν − Tν ρ − ∂μ (Πμ Σμν Φ) = 0. (17.17)
Using this result it is now easy to define a modified conserved symmetric energy-
momentum tensor,
1
(S )
Tμν = Tμν + ∂ ρ (Πρ Σμν Φ − Πμ Σρν Φ − Πν Σρμ Φ). (17.18)
2
The current associated with the Lorentz transformations can be expressed in
(S )
terms of Tμν as,
μ μ μ
J (M ) ν ρ = xν T (S ) ρ − xρ T (S ) ν . (17.19)
One can further modify the symmetric energy-momentum tensor to render it

also traceless,
(T L ) (S ) 1
Tμν = Tμν + ∂ ρ ∂ σ Xρσ μν , (17.20)
2
where Xρσ μν is defined such that the energy-momentum tensor is traceless and
(S )
conserved and η μν ∂ ρ ∂ σ Xρσ μν = 2∂ ρ ∂ ν σρν = −2η μν Tμν . In terms of this trace-
less energy-momentum tensor the dilatation current and the current associated
with the special conformal transformation are given by,
μ μν μ μρ
J (D ) = xν T (T L ) J (K ) ν = (2xν xρ − x2 ηρν )T (T L ) . (17.21)
μ μ μν
It is thus clear that J (D ) and J (K ) ν are conserved only provided that T (T L )
is conserved and traceless.
Note that in the latter form, scale invariance, namely a traceless energy-
momentum tensor, implies also conformal invariance.

17.3 Collinear and transverse conformal transformations of fields

Recall that in 2d conformal field theories it is very useful to employ light-cone
coordinates (in Minkowski space-time) or holomorphic and anti-holomorphic
coordinates (in complex Euclidean space-time). We want to argue now that it is
also quite useful to use light-cone coordinates, when analyzing four-dimensional
conformal field theories.
Consider the two light-like vectors nμ+ and nμ− ,
nμ+ n+ μ = nμ− n− μ = 0 nμ+ n− μ = 1. (17.22)
We can then decompose any Lorentz four-vector Aμ as follows,
Aμ = A− nμ+ + A+ nμ− + AμT , (17.23)
where,
A+ ≡ Aμ nμ+ A− ≡ Aμ nμ− , (17.24)
and the transverse part of the four-vector AμT is defined using the transverse part
of the metric defined as,
T
ημν = ημν − nμ− nν+ − nμ+ nν− AμT ≡ ημν
T
Aν . (17.25)
Using this decomposition we find that Aμ Aμ = 2A+ A− − A2T .
We can now also consider transformations associated with a subgroup of the
full conformal group. In particular consider the special transformations associ-
ated with a light-like parameter ãμ = ãnμ− . The transformation of x− takes the
form
x−
x− → x− = . (17.26)
1 + 2ãx−
Combining this transformation with the translation along the x− direction x− →
x− + a− and scaling x− → ax− these transformations constitute a subgroup of
the full conformal group, the collinear subgroup which is an SL(2, R).2 To verify
this group structure we define the following generators,
i
L+ = −iP+ L− = K−
2
i i
L0 = (D + M−+ ) E = (D − M−+ ). (17.27)
2 2
The generators L± and L0 obey the algebra,
[L0 , L± ] = ±L± [L− , L+ ] = −2L0 , (17.28)
which is indeed the SL(2, R) ∼ SO(2, 1) algebra; E commutes with them. It is,
actually, the L0 of the other SL(2, R), the one in the x+ direction.
2 The use of the SL(2, R) group in applications of conformal symmetry to QCD was introduced
in [150] and [83]

17.3 Collinear and transverse conformal transformations of fields 315
The SL(2, R) collinear subgroup is particularly useful for collinear fields, where
for instance Φ(x) takes the form Φ(α) ≡ Φ(αnμ+ ), with α a real number and nμ+
the light-cone direction defined above. In particular, as will be shown below,
this will apply to parton description of quarks. The field Φ(α) is taken to be an
eigenstate of the spin operator Σ+− ,
Σ+− Φ(α) = sΦ(α), (17.29)
so that s is the spin projection to the n+ direction. The collinear subgroup of the
conformal group now acts on the coordinate α as an SL(2, R) transformation,
aα + b
α → α = , (17.30)
cα + d
where a, b, c, d are real numbers with ad − bc = 1, and correspondingly the field
Φ(α) transforms as,
aα + b
Φ(α) → (cα + d)−2j Φ , (17.31)
cα + d
with,
1
j= (l + s). (17.32)
2
Thus Φ(α) is a representation of SL(2, R) or an SL(2, R) form of degree j.
The generators of the SL(2, R) group and E act on the collinear field as,
[L+ , Φ(α)] = −∂α Φ(α)

[L0 , Φ(α)] = (α∂α + j)Φ(α)
[L− , Φ(α)] = (α2 ∂α + 2jα)Φ(α)
1
[E, Φ(α)] = (l − s)Φ(α) (17.33)
2
where t = (l − s) is referred to as the collinear twist. In addition Φ(α) is an
eigenstate of the Casimir operator with,

[Li , [Li , Φ(α)]] = j(j − 1)Φ(α). (17.34)
i= 0,1,2
Another subgroup of the conformal group is the transverse subgroup SL(2, C)

acting on the transverse coordinates xμT = (0, x1 , x2 , 0) or in complex coordi-
nates z = x1 + ix2 and z̄ = x1 − ix2 , with fields Φ(z, z̄). This is in fact identi-
cal to the SL(2, C) discussed in Chapter 3 where the conformal symmetry of
two-dimensional field theories is discussed. In terms of the conformal generators
(17.7) the generators of the SL(2, C) are the PTμ , MTμν , D, KTμ . The coordinate z
transforms under the SL(2, C) transformation,
az + b
z → z = , (17.35)
cz + d

which implies the following transformation of Φ(z, z̄),

az + b
Φ(z, z̄) → (cz + d)−2h Φ , z̄ , (17.36)
cz + d
where h = 12 (l + λ) with λ is the helicity defined by Σz z̄ Φ = λΦ. Similarly for

the transformation of z̄ and the corresponding transformation of Φ(z, z̄), with
h̄ = 12 (l − λ).
17.4 Collinear primary fields and descendants

In two-dimensional conformal field theories fields were classified into primary
fields and descendant ones. The classification was based on their conformal trans-
formations. Correspondingly the states were put into Verma modules each con-
taining a highest weight state and its descendants. Recall the definition of the
former,
L0 [φ(0)|0>] = h [φ(0)|0>] Ln [φ(0)|0>] = 0, n > 0. (17.37)
In a similar manner the primary operator and the highest weight state of the
four-dimensional collinear group are defined [171], [42] as,
[L0 , Φ(0)] = jΦ(0) [L− φ(0)] = 0 ⇒

L0 [Φ(0)|0>] = j[Φ(0)|0>] L− [Φ(0)|0>] = 0, (17.38)
Φ(0) is by definition collinear since it is defined at the origin of the light-cone

direction. The fact that in the 2d case the conformal algebra is infinite while in
4d it is finite is manifested by the fact that in the former case there is an infinite
set of annihilation operators Ln , n > 0 that annihilate the highest weight state,
whereas in the latter case it is the single operator L− .
The descendant fields and correspondingly the descendant states are obtained
by repeatedly applying the creation operators, which are L−n in 2d while in 4d
it is the operator L+ . So in 4d,
On = [L+ , . . . [L+ , [L+ , Φ(0)]]] = (−∂+ )n Φ(α)|α =0 . (17.39)
Note the difference in notation, as it is L− in 2d while it is L+ in 4d, both raising.

The descendant operators obey the following commutation relations,
[L0 , On ] = (j + n)On [L+ , On ] = On +1 [L− , On ] = −n(n+2j − 1)On −1 .

(17.40)
In two-dimensions we discussed the Verma module that includes a highest weight
state and all its descendants, and similarly, in four dimensions we consider the
so-called conformal tower which also includes the highest weight state and all its
descendants. Recall however that there is an essential difference between the two
cases due to the fact that in the 2d the algebra is infinite dimensional whereas

17.4 Collinear primary fields and descendants 317
in 4d it is finite dimensional. In particular the notion of null vectors that played

an important role in the 2d case does not exist in four-dimensional CFTs.
We can associate the complete sets of Φ(α) and On by the following Taylor
expansion,
∞
(−α)n
Φ(α) = On . (17.41)
n=0
n!
An interesting and useful map relates the descendant operators and polyno-
mials. Consider for instance the descendent operator defined in (17.39), which
can be re-expressed as,
On = Pn (∂α )Φ(α)|α = 0 Pn (u) = (−u)n . (17.42)
It is straightforward to realize that in terms of these polynomials the operation

of L0 , L± takes the form,
L+ → L̃− = −u
L− → L̃+ = (u∂u2 + 2j∂u )
L0 → L̃0 = (u∂u + j). (17.43)
This correspondence can be viewed as first mapping,
∂α → u α → ∂u , (17.44)
then interchanging the + and − components, and finally some “normal ordering”
of taking the derivatives with respect to u to the right of the factors of u.
The representation in terms of polynomials is referred to as the ‘adjoint repre-
sentation’. Note that since in the original algebra L− includes a term proportional
to α2 , the new algebra includes a second-derivative term ∂u2 in L̃+ . This can be
avoided by introducing a different argument of the polynomials defined as,
un
→ κn , (17.45)
Γ(n + 2j)
so that the action L0 , L± on P̃(κ) is the same as (17.33) with α → κ and the
interchange of L− and L+ .
We now discuss composite operators built from two “elementary” operators of
the form,
O(α1 , α2 ) = Φj 1 (α1 )Φj 2 (α2 ), (17.46)
with α1 = α2 . The operator product expansion with |α1 − α2 | → 0 is expressed

in terms of the composite operators,
On (0) = Pn (∂1 , ∂2 )Φj 1 (α1 )Φj 2 (α2 )|α 1 =α 2 =0 , (17.47)
where Pn (∂1 , ∂2 ) is a homogeneous polynomial of degree n. It can be shown that

the complete set of local operators with which one can perform the conformal

operator expansion (COPE) takes the form,

→
− −
←
(2j 1 −1,2j 2 −1) ∂+ − ∂+
On (x) = ∂+ Φj 1 (x)Pn
j 1 ,j 2 n
→
− ←
− Φj 2 (x) , (17.48)
∂+ + ∂+
(a,b)
Pn (x) are the Jacobi polynomials, and we are back to x space here.3 One can
further generalize this construction to a product of three or more operators.
17.5 Conformal operator product expansion

The conformal operator expansion in two dimensions, discussed in Section 3.7.2,
was shown to be a very powerful tool in determining correlation functions. Obvi-
ously, we anticipate that in four dimensions the COPE will be less powerful.
We discuss now the general structure of the COPE.4 Consider the OPE of two
local conformal operators A(x)B(0) with twists and spin projection along the +
direction (tA , sA ), (tB , sB ), respectively. We perform an expansion for fixed x−
and x+ , xT → 0, namely x2 → 0. We want to expand the product in terms of a
complete set Onj 1,n,j+k
2
and to leading order in the twist. Such an expansion takes
the form,
∞
∞ 1/2(t A +t B −t n )
1
A(x)B(0) = Cn ,k xn−+k +Δ Onj 1,n,j+k
2
(0) + . . . , (17.49)
n=0 k=0
x2
where Δ = s1 + s2 − sA − sB , On ,n +k = (−∂+ )k On , s1 and s2 are the spin pro-

jections of the constituent fields in the local operators Onj 1,n,j+k
2
, t n = ln − n −
s1 − s2 = l1 + l2 − s1 − s2 the twist of the operator On , actually independent of
n, and the dots refer to higher twist contributions. We want to check to what
extent conformal invariance enables us to determine the coefficients Cn ,k . For
this purpose we act on the OPE with L− as,
1
[L− , A(x)B(0)] = x− (2jA + x · ∂x )A(x) − x2 n̄ · ∂x A(x) B(0) + . . . (17.50)
2
Inserting (17.49) and taking into account that,
[L− , Onj 1,n,j+k

2
(0)] = −k(k + 2jn − 1)Onj 1,n,j+k
2
−1 , (17.51)
3 The Jacobi polynomial is given by,

n
(a , b ) Γ(a + n + 1) n Γ(a + b + m + n + 1) z − 1 m
Pn (z) = .
n!Γ(a + b + n + 1) m Γ(a + m + 1) 2
m =0
4 COPE in four dimensions was introduced in [90] and used in QCD in [49], [50], [51].

with jn = j1 + j2 + n, we find the following recursion relation for the coefficients

Cn ,k ,
jA − jB + jn + k
Cn ,k + 1 = − Cn ,k , (17.52)
(k + 1)(k + 2jn )
which is solved by,
1 Γ(jA − jB + jn + k) Γ(2jn )
Cn ,k = (−1)k Cn , (17.53)
k! Γ(jA − jB + jn ) Γ(2jn + k)
where Cn = Cn ,0 . Plugging this into (17.49) we get the following form for the
OPE,
∞
1/2(t A +t B −t n )
1 xn−+s 1 +s 2 −s A −s B
A(x)B(0) = Cn
n=0
x2 B(jA − jB + jn , jB − jA + jn )
1
× duu(j A −j B +j n −1) (1 − u)(j B −j A +j n −1) Onj 1 ,j 2 (ux− ), (17.54)
0
Γ(a)Γ(b)
where B(a, b) = Γ(a+b) is the beta function.
17.6 Conformal Ward identities

The application of conformal invariance in determining correlation functions was
discussed in Section 2.8 for the case of two-dimensional conformal field theo-
ries. It includes both the use of the Ward identities associated with the global
SL(2, C) transformation5 and the full holomorphic conformal transformation.
Here in discussing four-dimensional field theories we will encounter two major
differences:
(i) Due to the fact that the conformal symmetry group is finite dimensional,
there are Ward identities only associated with global transformations.
(ii) When discussing theories with conformal anomalies, there will be modifica-
tions of the Ward identities.
Let us start by reminding ourselves of the concept of Ward identities and in
particular the conformal ones. Associated with any infinitesimal transformation
of a given field φ(x) → φ(x) + δφ(x), the action that describes the system is
transformed into,

S → S + δS = S + d4 xΔ(φ, ∂μ φ). (17.55)
Associated with this transformation there is a current Jμ such that ∂ μ Jμ = Δ.

Obviously where Δ = 0 (or a total derivative) the transformation is a symmetry
and the corresponding Noether current is conserved. Associated with such a
5 Conformal Ward identities which were studied in [168] are identical to the Callan–Symanzik
equation [55] and [204].

transformation of the field there is a constraint on correlation functions of this

field. This constraint which is referred to as a Ward identity takes the form,
∂y μ <T Jμ (y)φ(x1 )...φ(xN )> = <T Δ(y)φ(x1 )...φ(xN )>

−iδ 4 (x1 − y) <T δφ(x1 )...φ(xN )> ...
−iδ 4 (xi − y) <T φ(x1 )...δφ(xi )...φ(xN )> ...
−iδ 4 (xN − y) <T φ(x1 )...δφ(xN )> . (17.56)
This relation can be derived straightforwardly using the path integral formulation
of correlation functions. The Ward identity takes a simpler form when integrating
over y μ ,
<T δφ(x1 )...φ(xN )> +...+ <T φ(x1 )...δφ(xi )...φ(xN )> +...
<T φ(x1 )...δφ(xN )> + <T iδSφ(x1 )...φ(xN )> = 0. (17.57)
In particular in analogy with (2.56) the Ward identities associated with dila-
tion and special conformal transformation take the form,
N
(lφ + xi ∂i ) <T φ(x1 )...φ(xN )> = −i d4 x <T ΔD (x)φ(x1 )...φ(xN )>
i

N
(2xμi (lφ + xi ∂i ) − 2Σμν xνi − x2i ∂iμ ) <T φ(x1 )...φ(xN )>
i

= −i d4 x2xμ <T ΔD (x)φ(x1 )...φ(xN )>, (17.58)
where lφ is the canonical dimension, namely that of the free field. Similarly
to the way we extracted information about the structure of correlators in 2d
CFT in Section 2.9, we can now constrain the form of correlators in 4d. The
Ward identities associated with the Poincare transformations imply that any
correlation function is in fact not a general function of the N coordinates xμi ,
but only of the invariants x2ij ≡ (xi − xj )2 .
To understand the implication of the dilatation transformation on the correla-
tion function let us first study the theory at its fixed point, namely at a coupling
g ∗ = g(μ∗ ) such that β(g ∗ ) = 0. Recall that the β function is defined as,
∂
β(g(μ)) = μ g(μ), (17.59)
∂μ
and hence the vanishing β function implies a fixed point of the coupling constant
g. This will be further discussed below for the case of 4d QCD. In this case the
dilatation Ward identity takes the form of that of a free theory, like the one given
in (17.58), apart from the change of scaling dimension,

N
(lΦ + γ(g ∗ ) + xi ∂i ) <T φ(x1 )...φ(xN )> = 0, (17.60)
i

where γ(g ∗ ) is the anomalous dimension of the filed Φ. As a consequence of this

form of the Ward identity, the two-point function of two scalar fields at the fixed
point has the form,
l φ +γ (g ∗ )
∗ ∗ −2γ (g ∗ ) 1
<φ(x1 )φ(x2 )> = N2 (g )(μ ) . (17.61)
(x1 − x2 )2
For particles with spin s and the same projection on the light-cone,
l φ +γ (g ∗ ) s
∗ ∗ −2γ (g ∗ ) 1 (x1 − x2 )+
<φ(x1 )φ(x2 )> = N2 (g )(μ ) ,
(x1 − x2 )2 (x1 − x2 )−
(17.62)
where it is assumed that (x1 − x2 )T = 0. At the fixed point, namely β(g ∗ ) = 0,
the Ward identity associated with the special conformal transformation takes the
form of that of a free theory with lφ again shifted by the anomalous dimension
lφ → lφ + γ(g ∗ ).
In two dimensions (see Section 2.8) it was found that the three-point function
of primary fields is fully determined by the SL(2, C) symmetry and any four-
point function of primary fields is determined up to a function of the cross ratio
(or anharmonic ratio) zz 11 23 zz 32 44 and its complex conjugate coordinate. Based on
the Poincare, dilation and special conformal transformation in four dimensions,
the three-point function is determined here too. For instance, the three-point
function of a scalar field is,
[l φ +γ (g ∗ )]/2
∗ 1
<φ(x1 )φ(x2 )φ(x3 )>=N3 (g ∗ )(μ∗ )−3γ (g ) ,
(x1 −x2 )2 (x1 −x3 )2 (x2 −x3 )2
(17.63)
x x
and any correlator of n > 3 operators depends only on the ratios xiijl x jk ll .
It is worth noting that the Ward identity associated with the dilation (the
first equation of (17.58)) is in fact the same as the Callan–Symanzik renor-
malization group equation. First note that based on dimensional counting and
Lorentz invariance the dependence of the N point function on the scale μ takes
the form,
<T Φ(x1 )...Φ(xN )> = μN l Φ G(x2ik μ2 ; g(μ)), (17.64)
which means that the following relation holds,

N
∂
(lΦ + xi ∂i ) <T Φ(x1 )...Φ(xN )> = μ <T Φ(x1 )...Φ(xN )> . (17.65)
i= 1
∂μ
It is easy to realize that the right-hand side of the conformal Ward identity can
be rewritten in the form,

∂
i d4 x <T ΔD (x)Φ(x1 )...Φ(xN )> = −M <T Φ(x1 )...Φ(xN )>, (17.66)
∂M

as follows from,
∂
ΔD (x) = −M Leff , (17.67)
∂M
for the cases with no explicit dimension-full parameters. We will show this explic-
itly for the effective theory of 4d QCD below. On the other hand the dependence
of the correlator on M follows from the dependence of the field renormalization
factor and the dependence of the coupling constant so that,

∂ ∂ N
−M <T Φ(x1 )...Φ(xN )> = β(g) + γΦ <T Φ(x1 )...Φ(xN )> .
∂M ∂g i=1 i
(17.68)
Combining this together with (17.58) and (17.60) we get the Callan–Symanzik
equation,

∂ ∂ N
μ + β(g) + γΦ <T Φ(x1 )...Φ(xN )> = 0. (17.69)
∂μ ∂g i= 1 i
17.7 Conformal invariance and QCD4

So far we have discussed the implications of conformal invariance in general, and
in particular the invariance properties under the SL(2, R) collinear group and
conformal Ward identities. We are now in the position to examine the application
of conformal symmetry to four-dimensional QCD. Recall that the action of four-
dimensional SU (N ) gauge theory with massless quarks takes the form,
1 a μν a
LQ C D 4 = − Fμν F + iψ̄D ψ, (17.70)
4
where Dμ = ∂μ − igta Aaμ is the covariant derivative, and ta as usual are the
N × N matrices in the fundamental representation of the SU (N ) algebra. It is
straightforward to check that the corresponding classical action is invariant under
the full set of fifteen transformations associated with the SO(2, 4) symmetry
group. In particular it is invariant under the scale transformation given by,
xμ → λxμ Aμ (x) → λAμ (λx) ψ(x) → λ3/2 ψ(λx). (17.71)
The invariance under these transformations manifests itself in the form of con-
servation of the corresponding Noether current,

μν i ↔
Dμ = xν T (T L ) = xν F μρ a Fρν a + ψ̄(D)(μ γ ν ) ψ , ∂ μ Dμ = 0, (17.72)
2
↔ → ←
− −
where (D) ≡ D − D. The classical invariance is not maintained quantum
mechanically. This situation of having classical conformal symmetry but not
a corresponding quantum mechanical one is referred to as the conformal

anomaly.6 In string theory the two-dimensional conformal symmetry is local.

Having an anomaly in a local symmetry renders the theory into an inconsistent
one. This implies that (at least in flat space-time) the theory will be defined in a
critical dimension where the conformal anomaly vanishes. In the four-dimensional
field theories discussed here, like QCD4 , conformal invariance is a global sym-
metry and the theory is consistent even when having an anomaly. There are
several ways to show that the quantum theory is not scale and hence also not
conformal invariant. One may say that the anomaly follows from the fact that
the theory has infinities that are cured just by the introduction of a renormal-
ization procedure. The latter involves the introduction of a cutoff scale. Once a
scale is introduced the theory is not any more scale invariant.
To see it more explicitly let us consider the low energy effective action of mass-
less QCD4 . We expand the gluons and quarks in terms of modes and distinguish
the low energy (momentum) modes and the high energy modes. Next we inte-
grate the high energy modes to derive the one loop low energy effective action.
It takes the form,7

1 1 β0 M2 μν a
SLE = − d4 x − ln F a
μν F + ... , (17.73)
4 g02 16π2 μ2 low
where M is the UV cutoff, and β0 = 11 3 Nc − 3 Nf is the coefficient of the one

2
loop beta function. It is easy to check that this one loop renormalized action is
not invariant under the scale transformations of (17.71). The variation of the
action under those transformation reads,

1 1 a μν a
δS = − β0 lnλ d4
x F F + ... . (17.74)
32π2 g02 μν low
Thus the quantum mechanically (unlike the classical case) dilatation Noether
current is not conserved,
1
∂μ Dμ ≡ ΔD = − 2
a
β(g)Fμν F μν a low , (17.75)
32π
and in deriving the right-hand side of the equation we have used the equations
of motion.
The effective action admits also an anomaly with respect to the special con-
formal transformations.
In (17.46) we discussed the general structure of non-local operators of four-
dimensional conformal field theory. In QCD in many cases we encounter a non-
local operator built from a quark and an anti-quark at light-like separation, with
6 The conformal anomaly was introduced in [169] and [6].

7 The explicit calculation is a one loop perturbative calculation. Since we do not deal with
perturbative methods in this book, we do not present here the derivation and refer the
reader to references that deal with perturbation theory in QCD 4 .

a line integral connecting them,

α1
ig dtA + (t)
Qμ (α1 , α2 ) = ψ̄(α1 )γμ P e α2
ψ(α2 ) (17.76)
where P stands for path ordering. The path integral factor will be denoted
[α1 , α2 ]. In performing the short distance expansion we need now to identify
the corresponding conformal operators. To relate the operator ψ to a primary
operator we first have to make a spin projection in the following way,
ψ+ = Γ+ ψ ψ− = Γ− ψ ψ = ψ+ + ψ − , (17.77)
where,
1 1
γ− γ+ ,
Γ+ = Γ− = γ+ γ− , Γ− + Γ+ = 1. (17.78)
2 2
The spin projected parts are,
ψ+ (s = +1/2, j = 1, t = 1) ψ− (s = −1/2, j = 1/2, t = 2). (17.79)
With this identification we define the quark anti-quark operators:
twist − 2 : Q+ = ψ̄+ γ+ ψ+ ≡ Q(1,1)
twist − 3 : QT = ψ̄+ γT ψ− + ψ̄− γT ψ+ ≡ Q(1,1/2) + Q(1/2,1)
twist − 4 : Q− = ψ̄− γ− ψ− ≡ Q(1/2,1/2) . (17.80)
The corresponding local conformal operators are,
↔
Qn1,1 (α) = (i∂+ )n ψ̄(α)γ+ Cn3/2 D+ /d+ ψ(α) ,
↔
Qn1,1/2 (α) = (i∂+ )n ψ̄(α)γ+ γT γ− Pn1,0 D+ /d+ ψ(α) ,
↔
Qn1/2,1/2 (α) = (i∂+ )n ψ̄(α)γ− Cn1/2 D+ /d+ ψ(α) , (17.81)
where
↔ →
− ←
− →
− ←
−
D+ = D + − D + d+ = D + + D + , (17.82)
and where the Jacobi polynomials with two identical indices were replaced by
3/2 1/2
the Gegenbauer polynomials P (1,1) ∼ Cn and P (0,0) ∼ Cn .
A similar analysis can be carried out for the gluons. The various components
of the gluon field have the following properties,
F+T (s = +1, j = 3/2, t = 1) FT T , F+− (s = 0, j = 1, t = 2)
F−T (s = −1, j = 1/2, t = 3). (17.83)
Local operators built from two-gluon fields with leading twist are,
↔
Gn3/2,3/2 (α) = (i∂+ )n F+T (α)Cn5/2 D+ /d+ F+T (α) . (17.84)
Another application of conformal invariance to QCD is the determination of

the OPE of two electromagnetic currents jμEM = i ei ψ̄i γμ ψi where the ei are the

charges of the u, d and s quarks. At the tree level only the transverse components
are of interest. The latter have spin sj = 0 and twist tj = 3. The quark operators
Qn1,1 are the relevant basis for the expansion, with conformal spin jn = (ln + 1 +
n)/2 = n + 2 and tn = (ln − 1 − n)/2 = 2. As Δ = 1 we find
J T (x)J T (0) ∼
∞ (6−t n )/2 1
n=0 Cn 1
x2 (−ix− )n + 1 Γ(jΓ(2j n)
n )Γ(j n ) 0
du[u(1 − u)]j n −1 Qn1,1 (ux− ).
(17.85)
The coefficients Cn can be extracted from deep inelastic scattering via the fol-
lowing matrix element of forward scattering,
∞
(6−t n )/2
1
 ∼ Cn (−ix− )n +1 .
n=0
x2
(17.86)
Another application of the COPE is the determination of the short-distance
expansion of the operator Q+ ( 17.76). This case is characterized by sA = sB =
s1 = s2 = 12 so that Δ = 0 and lA = lB = l1 = l2 = 32 and we find,
∞
1
n
Q+ (α1 , α2 ) ∼ C̃n (−i) (α1 −α2 )n duun + 1 (1−u)n +1 Qn1,1 (uα1 +(1−u)α2 ),
n=0 0
(17.87)
C n Γ(n +2) 2
where C̃n = Γ(2n +4) which can be determined again from forward matrix ele-
ments and are found to be C̃n = 2(2n + 3)
(n + 1)! .
Conformal invariance can be used at short distances to give predictions for the
quark distribution amplitudes for flavor non-singlet mesons, namely the wave
functions which control the behavior of the exclusive mesons processes at large
momentum transfer. Here we discuss as an example the pion distribution ampli-
tude in the leading twist order.
The basic ingredient in computing exclusive reactions including a large
momentum transfer to a pion is the matrix element of a quark anti-quark between
the vacuum and a one pion state. By using the light-cone gauge A+ = 0 the
Wilson line (17.76) is set to unity. We choose a frame where pμ = p+ n−μ and
xμ = x− nμ+ + xμT , x+ = 0 so that x2 = x2T . The matrix element can then be
written as,
¯
<0|d(0)[0, ∞n]γ+ γ5 [∞n + x, x]u(x)|π+ (p)> =
1
ifπ p+ 0
dye−iy (p·x) f (y, lnx2 ) + O(x2 ). (17.88)
This matrix element is the probability amplitude to find the pion in the valence
state consisting of a u-quark carrying a momentum y and an anti-d quark of
momentum ȳ = 1 − y and have a transverse separation xT . This amplitude is
intimately related to the pion electromagnetic form factor for large momentum
transfer Q2 and small separation distance of the order xT ∼ 1/Q2 . To approach

this limit, one defines the pion distribution amplitude taken at exactly light-like
separation where xT = 0. This amplitude reads,
1
¯
<0|d(0)[0, α]γ+ γ5 u(α)|π+ (p)> = ifπ p+ dye−iy (α p +) φπ (y, μ). (17.89)
0
The distribution amplitude φπ (y, μ) is scale and scheme dependent. In fact the
small transverse distance behavior of the valence component of the pion wave
function is traded for the scale dependence of the distribution amplitude.
It can be shown that the evolution equation of φπ (y, μ) is given by,
1
d
μ2 2 φπ (y, μ) = dỹV (y, ỹ, αs (μ)φπ (y, μ), (17.90)
dμ 0
where to leading order in αs the integral kernel is given by,

1−y 1 ỹ 1
V0 (y, ỹ) = CF 1+ θ(y − ỹ) + 1+ θ(y − ỹ) ,
1 − ỹ y − ỹ y y − ỹ +
(17.91)
where ]+ stands for,

1
[V (ỹ, y)]+ = V (ỹ, y) − δ(y − ỹ) dtV (t, ỹ). (17.92)
0
Instead of solving this evolution equation one can alternatively proceed by

expanding both sides of (17.90) in powers of α. In this way moments of the distri-
bution amplitude are related to matrix elements of renormalized local operators
in the following form,
1
↔
¯
<0|d(0)γ n +
+ γ5 (iD + ) u(0)|π (p)> = ifπ (p+ )
n +1
dy(2y − 1)n φπ (y, μ).
0
(17.93)
This is similar to the leading twist operators that enter the OPE for the unpo-
larized deep inelastic scattering apart from the flavor, the additional γ5 factor
and the fact that now one has to take into account mixing with operators that
contain total derivatives of the form,
↔
¯ n −k
On −k ,k = (i∂+ )k d(0)γ+ γ5 (iD + ) u(0). (17.94)
The mixing matrix is in fact triangular since operators with fewer total deriva-
tives can only mix with operators with more total derivatives but not the other
way around. The components of the matrix on the diagonal are true anomalous
dimensions, which are identical to those of inelastic scattering,

2
n+1
1
γn = CF 1 −
(0)
+4 , (17.95)
(n + 1)(n + 2) m =2
m

where,
(0)
γn
αs (μ) 11 2
β0
 = Nc − Nf . β0 =
αs (μ0 )3 3
(17.96)
Conformal invariance is useful in finding the eigenvectors of the mixing matrix
since conformal operators with different conformal spins cannot mix under renor-
malization to leading order. This happens since to leading order the renormaliza-
tion is determined by counter terms of the tree level which is conformal invariant.
Thus the mixing eigenvector operators are Q1,1 (x) defined in (17.80) with the
right flavor and γ matrices structure,
↔
¯
Qn1,1 (x) = (i∂+ )n d(x)γ 3/2
+ γ5 Cn D+ /d+ u(x) . (17.97)
Note that because of their flavor content these operators cannot mix with oper-
ators made out of gluons and they also cannot mix with operators with more
fields since they have higher twist. Thus the operators (17.97) are the only rel-
evant ones and they must be multiplicatively renormalized. Comparing (17.93)
with (17.97) one concludes that the Gegenbauer moments of the pion distri-
bution amplitudes are given in terms of reduced matrix elements of conformal
operators,
1
ifπ pn++ 1 dyCn3/2 (2y − 1)φπ (y, μ) = <0|Qn1,1 (0)|π+ (p)> . (17.98)
0
As was mentioned above these operators are renormalized by a multiplication

and the corresponding anomalous dimension is given by (17.95). Thus the final
picture is that the distribution amplitude φπ (u, μ) can be expanded in a series
of Gegenbauer polynomials,
∞

φπ (u, μ) = 6u(1 − u) φn (μ)Cn3/2 (2u − 1)
n=0
2(2n + 3)
φn (μ) = (ifπ pn++ 1 )−1 <0|Qn1,1 (0)|π+ (p)>
3(n + 1)(n + 2)
(0)
γn
αs (μ) β0
φn (μ) = φn (μ0 ) . (17.99)
αs (μ0 )
This example demonstrates the application of conformal invariance to solve the
problem of operator mixing. There are other applications of conformal symmetry
to four-dimensional QCD. We refer the interested reader to [43]. The predictions
based on conformal symmetry beyond one loop, for the sector with β = 0 [51],
turned out to be in contradiction with explicit calculations [79]. This paradox
was resolved in [166].

18
Integrability in four-dimensional gauge dynamics
Integrability was discussed in Chapter 5 in the context of two-dimensional mod-

els. In particular solutions of spin chain models based on the Bethe ansatz
approach were described in some detail. Integrable models are characterized by
having the same number of conserved charges as the number of physical degrees
of freedom. Furthermore, the scattering processes of those models always involve
conservation of the number of particles.
A natural question at this point is whether integrability is a property of only
two-dimensional models or whether one can also identify systems in four dimen-
sions that admit integrability. Four-dimensional gauge theories have generically
infinite numbers of degrees of freedom and their interactions do not conserve the
number of particles. Thus four-dimensional gauge theories like the YM theory
are not integrable theories. However, it turns out, as will be shown in this chap-
ter, that various sectors of certain four-dimensional gauge theories, which are
derived upon imposing certain limits, do admit integrability.
The two-dimensional integrable models discussed in Chapter 5 were non-
conformal ones and were characterized by a scale and hence also with particles
and an S-matrix. On the other hand the integrable sectors of four-dimensional
gauge theories that we are about to describe are conformal invariant. The main
idea is that these special conformal invariant sectors can be mapped into two-
dimensional spin chains that were described in Section 5.14.
The investigation of this issue is far from complete. Nevertheless, a large body
of knowledge has already been accumulated. In recent years this has followed
the lines of the AdS/CFT duality [158] which is not covered in this book.1 The
purpose of this section is just to demonstrate the idea of the map between gauge
theories and in particular QCD and integrable spin chain models. This will be
done by describing the following cases:
(i) N = 4 super YM theory in four dimensions.

(ii) Scale dependence of composite operators in QCD.
N = 4 super YM theory is known to be the maximal global supersymmetric

theory in four dimensions. Since supersymmetry is beyond the scope of this book
we will not discuss it in the context of the N = 4 SYM. Thus the description
1 For a review of the AdS/CFT the reader can refer to [10].

330 Integrability in four-dimensional gauge dynamics
of the integrable aspects of the theory will be incomplete and will be missing
certain essential parts. However, since N = 4 SYM is the simplest interacting
four-dimensional non-abelian gauge theory we start with this and then proceed
to a certain limit in non-supersymmetric QCD.
There are several review papers on integrability in four-dimensional gauge
dynamics. In this chapter we follow [31] about the integrability of N = 4 SYM
theory and [34] for the scale dependence of composite operators of QCD.
18.1 Integrability of large N four-dimensional N = 4 SYM

The Lagrangian of N = 4 SYM is given by,
1 1 1
2
LN = 4 = − Tr [Fμν F μν ] + Tr [Dμ Φn Dμ Φn ] − g 2 Tr [Φm , Φn ]
4 2 4
i i
+Tr ψ̇α̇a σμα̇ β Dμ ψβ a − gTr ψα a σm [Φm , ψβ b ] − gTr ψ̇α̇a σab
ab α β m α̇ β̇
Φm , ψ̇β̇b ,
2 2
(18.1)
where Fμν is the field strength associated with an SU (N ) gauge group, Φm is a
set of six m = 1, . . . , 6 scalar fields, and ψ and ψ̇ are doublets of SU (2) × SU (2).
Both the scalars and the spinors are in the adjoint representation of SU (N ). The
matrices σ μ and σ m are the chiral projections of the gamma matrices in four and
six dimensions, respectively and is the totally antisymmetric tensor of SU (2).
It is convenient to write the corresponding action as,
4
d x
S=N LN =4 , (18.2)
4π 2
g2 N
where the coupling constant is taken to be g 2 ≡ Y8π M
2 .
It is well known that the theory, on top of being invariant under SU (N )
gauge symmetry and SO(6) global symmetry, is also conformal invariant and in
fact superconformal invariant. The β function of the theory which vanishes to
all orders in perturbation theory is believed to vanish also non-perturbatively
and hence the theory is assumed to be conformal also in the quantum level. In
Section 17.1 we have described the conformal symmetry algebra in four dimen-
sions. Recall the SO(2, 4)2 conformal transformations (see 17.7) which are being
generated by P μ , S μν , D, K μ , the generators of space-time translations, Lorentz
transformations, dilation and special conformal transformation, respectively.
A major player in the structure of the N = 4 is the dilatation operator D.
Whereas the generators of the Poincare group do not get quantum corrections,
the dilatation operator does so that in fact,
D = D0 + δD(g), (18.3)
2 In fact the N = 4 SYM admits a superconformal algebra of psu(2, 2|4) which we do not
discuss here.

18.1 Integrability of large N four-dimensional N = 4 SYM 331
Fig. 18.1. A single trace operator as a spin chain.
where D0 is the classical operator and δD is the anomalous dilation operator

which obviously depends on the gauge coupling g.
The “states” of the theory take the form of multi-trace gauge invariant oper-
ators,
Tr [W . . . .W] . . . Tr [W . . . .W] W ∈ {Dk Φ, Dk Ψ, Dk Ψ̇, Dk F }, (18.4)
where D stands for the covariant derivative and F ≡ Fμν is the field strength.
The Hilbert space of states is built, as for any conformal field theory, from
Verma modules each characterized by a highest weight state or a primary state,
which were defined in (2.8). An example of a highest weight state is |K> =
η m n Tr[Φm Φn ]. The rest of the Verma module includes the descendant states
which are derived by acting with lowering operators on primary states. Needless
to say the general structure of correlation functions of four-dimensional conformal
field theories discussed in Section 17.6 applies also for the case of the N = 4 SYM.
In particular recall (2.8) that the two-point function of two operators is given by,
M (g)
<O(x1 )O(x2 )> = . (18.5)
|x1 − x2 |2D(g )
The anomalous dimension can be computed perturbatively as a power series
in g. As was discussed in Chapter 7 the perturbation expansion becomes much
more tractable in the large N limit, namely, in the planar limit. In this limit the
dominant diagram has a vanishing Euler number χ = 2C − 2G − T = 0 where
C, G, T stand for the number of components, genus, namely the number of
handles, and the number of traces, respectively. Since each component requires
two traces, one incoming and one outgoing, it implies that the planar limit
projects onto diagrams with G = 0 and T = 2C. This means that only single
trace operators are relevant.
We have seen above that in the planar limit we deal with single trace operators.
Pictorially, (see Fig. 18.1) a single trace operator looks like a cyclic spin chain.
This map can be made precise. Spin chain as integrable models were discussed
in Section 5.14. Recall that a spin chain includes a set of L spins with cyclic
adjacency property.3 The spin at each site is a module of the symmetry algebra
of the system. The Hilbert space of the whole system is the tensor product
of L modules. In Section 5.14 we discussed only chains with a fixed number
3 In Section 5.14 we denoted the number of spins by N . Here to avoid confusion with the rank
of the gauge group we will refer to the number of spins as L.

Table 18.1. N = 4 SYM theory to spin chain dictionary
planar N = 4 SYM spin chain
Single trace operator Cyclic spin chain

Field operator Spin at a site
Anomalous dilatation operator g −2 δD Hamiltonian
Anomalous dimension Energy eigenvalue
Cyclicity constraint Zero momentum condition U = 1
of spins. One can generalize this situation to incorporate also a dynamic spin
chain with an unfixed number of spins. In this case the Hilbert space is a tensor
product of all Hilbert spaces of a fixed length. In the Heisenberg model each
(L )
spin has two possible states and the Hilbert space is therefore C (2 ) . In general
the spin in the chain can point in more than two directions and in particular
also in infinitely many directions, as is the case for the spin chain of the N = 4
SYM theory. In the latter case the spin is mapped into a field operator and the
possible spin states to the components of the gauge symmetry multiplet. The
cyclicity of the single trace operators maps into a constraint on the spin chain
so that states that differ by a trivial shift are identified and hence states with
non-trivial momentum are unphysical. In the language of Section 5.14 we have to
impose U = 1 as a constraint. In the Heisenberg model this renders the Hilbert
(2(L ) )
space into C Z L . The Hamiltonian of the spin chain model translates into the
dilatation operator and the energy eigenvalues to the anomalous dimensions. The
full correspondence between the spin chain and the planar limit of the N = 4
SYM theory is summarized in Table 18.1.
Once the correspondence with a spin chain model has been established, one
can proceed in a similar way as for the Heisenberg spin chain model. The next
step is to write down the algebraic Bethe ansatz which now corresponds to an
SO(6) symmetry if one considers operators constructed only from the fields Φm
or in general the psu(2, 2|4) for the full N = 4 SYM theory. The algebraic Bethe
ansatz, the analog of (18.6) now reads as follows,
L K
λk + i/2Vj k λk − λl + iMj k ,j l
= , (18.6)
λk − i/2Vj k λk − λl − iMj k ,j l
l= k
where L is the size of the chain (N in (5.224)), the total number of excitation
is K (l in (5.224)) and where for each of the corresponding Bethe roots λk one
specifies which of the simple roots is excited by jk which takes the values of
1, . . . ., #sr with #sr being the number of simple roots which for the SO(6) case
is three and for the psu(2, 2|4) is seven. M is the Cartan matrix of the algebra
(1 in (5.224)) and V are the Dynkin labels of the representation (s in (5.224)).

The condition of zero momentum now reads,

K
λk + 2i Vj k
1=U = . (18.7)
k=1
λk − 2i Vj k
The energy of a configuration of roots that satisfies the Bethe equation is,

K
Vj k
E= . (18.8)
k=1
λ2k + 14 Vj2k
This is of course the analog of (5.225) and the higher conserved charges are,

i
K
1 1
Qr = 2 r −1 − 2 r −1 . (18.9)
r−1 λk + 2i Vj k λk − 2i Vj k
k=1
The leading order part of the transfer matrix reads,

K
λ − λk + 2i Vj k
T (λ) = + . . . ., (18.10)
k=1
λ − λk − 2i Vj k
It was shown that these generalized Bethe equations provide a solution to the
planar anomalous dimensions of the N = 4 SYM theory [162]. This is just the tip
of the iceberg. The integrable structure of the planar N = 4 has been investigated
very thoroughly in recent years. For an early review on the topic the reader can
consult [31]. As an epilog let us mention that, as was shown in [106], one can
identify in a similar manner with what was done in N = 4 SYM [35], a spin
chain structure in gauge theories which are confining and with less or even no
supersymmetries. In that case the spin chain Hamiltonian would not correspond
to the dilatation operator but was rather associated with the excitation energies
of hadrons.
18.2 High energy scattering and integrability

High energy scattering is characterized by the fact that the Mandelstam parame-
ter s = (pA + pB )2 is the largest scale of the system, and in the limit of s → ∞ the
energy dependence corresponds to a renormalization group flow of the dynamical
system that “resides” on the two dimensions transverse to the scattering plane.
It is convenient to study the properties of the high energy scattering amplitude
A(s, t) using the Mellin transform,
δ +i∞
dw w
A(s, t) = is s Ã(w, t), (18.11)
δ −i∞ 2πi
where the integration contour goes to the right of the poles of A(w, t) in the
w complex plane. The high energy asymptotic of A(s, t) is determined by
the poles of the partial wave amplitudes, namely if Ã(w, t) ∼ (w −w1 0 (t)) , then
A(s, t) ∼ is1+w 0 (t) . Poles in the w plane are referred to as reggeons and the posi-
tion of the pole is called the reggeon trajectory.

The partial wave amplitude Ã(w, t) can be written using the impact parameter
representation as follows,

Ã(w, t) = d b0 e
2 i(q b 0 )
d2 bA d2 bB ΦA (bA − b0 )Tw (bA , bB )ΦB (bB )

≡ d2 b0 ei(q b 0 ) <Φ(b0 )|Tw |Φ(0)>, (18.12)
where the impact factors ΦA (bA ) and ΦB (bB ) are the parton distributions which
are functions of the transverse coordinates bA = b1A , b2A , . . . , bnA for the A col-
liding hadron and bB = b1B , b2B , . . . , bnB for the B hadron, and Tw (bA , bB ) is the
scattering (partial wave) amplitude for a given parton configuration. The idea
behind this representation of the amplitude is that the transverse coordinates
of the partons can be considered as “frozen” during the interaction. It implies
that the structure of the poles in the w-plane does not depend on the parton
distribution in the colliding hadrons but rather on the general properties of the
gluon interaction of the t-channel. It was shown [145], that the propagators of
the t-channel gluons develop their own Regge trajectory due to interactions. A
t-channel gluon “dressed” by the virtual corrections is referred to as reggeized
gluon. The reggeized gluons are the relevant degrees of freedom of the high energy
scattering. The partial waves Tw (bA , bB ) can be classified according to the num-
ber of the reggeized gluons propagating in the t-channel. The minimal number
required to get a colorless exchange is two gluons. We will discuss here only this
case. It can be shown that the amplitude Tw (b1A , b2A b1B b2B ) satisfies the so-called
BFKL equation that reads [23], [145],
αs Nc
wTw = Tw(0) + HB F K L Tw , (18.13)
π
(0)
where Tw corresponds to the free exchange of two gluons. Formally one can
write the solution as,
−1
αs Nc
Tw = w − HB F K L Tw(0) , (18.14)
π
so that the singularities of Tw are determined by the eigenvalues of the operator,
HB F K L Ψα (b1 , b2 ) = Eα Ψα (b1 , b2 ), (18.15)
where Ψα is the eigenstate. The high energy behavior of the scattering ampli-
tude is dominated by the right-most singularity of Tw , namely on the max-
imal eigenvalue (Eα )m ax . The equation (18.15) has the interpretation of the
two-dimensional Schrödinger equation of two interacting particles. The inter-
acting particles can be identified with reggeized gluons and Ψα (b1 , b2 ) is the
wavefunction of a colorless bound state of them. Defining the holomorphic and

anti-holomorphic coordinates of the reggeized gluons as follows,

zj = xj + iyj z̄j = xi − iyi , (18.16)
where bj = (xj , yj ), we can split the BFKL Hamiltonian into a sum of two terms,
one that acts only on the holomorphic coordinates and another that acts only
on the anti-holomorphic ones, as follow HB F K L = H + H̄ where,
H = ∂z−1
1
ln(z12 )∂z 1 + ∂z−1
2
ln(z12 )∂z 2 + ln(∂z 1 ∂z 2 ) − 2ψ(1), (18.17)
where z12 = z1 − z2 , ψ(x) is the Euler digamma function defined by ψ(x) =
d ln Γ(x)
dx and in H̄ we replace all the zi s by z̄i s.
The BFKL Hamiltonian is further invariant under SL(2, C) transformations.
Denoting the SL(2, C) generators (see Section 2.9) by,
Lj − = −∂z j Lj 0 = zj ∂z j Lj + = zj2 ∂z j La = L1a + L2a , (18.18)
and similarly for the anti-holomorphic generators, the invariance takes the form,
[HB F K L , La ] = [HB F K L , L̄a ] = 0. (18.19)
This implies that HB F K L depends only on the Casimir operators of SL(2, C)
algebra of the two particles, namely, with,
L212 = −(z1 − z2 )2 ∂z 1 ∂z 2 L̄212 = −(z̄1 − z̄2 )2 ∂z̄ 1 ∂z̄ 2 , (18.20)
the Hamiltonian must take the form,
H = H(L212 ) H̄ = H̄(L̄212 ). (18.21)
It thus follows that the eigenstates of the Hamiltonian must also be eigenstates
of L212 and of L̄212 ,
L212 Ψn ,ν = h(h − 1)Ψn ,ν L̄212 Ψn ,ν = h̄(h̄ − 1)Ψn ,ν , (18.22)
where the complex dimensions h and h̄ are given by,
1+n 1−n
h= + iν h̄ = + iν. (18.23)
2 2
The non-negative integer n and the real parameter ν specify the irreducible
representation of the SL(2, C) group to which Ψn ,ν belongs. The wave functions
which are eigenstates of the Casimir operators take the form,
1+ n 1 −n
+iν +iν
z12 2
z̄12 2
Ψn ,ν (b) = , (18.24)
z10 z20 z̄10 z̄20
where zij = zi − zj and b0 = (z0 , z̄0 ) is the center of mass of the state. The con-
formal dimension of the state is h + h̄ = 1 + 2iν and the spin h − h̄ = n. Upon
substituting these eigenstates into (18.15) and using the explicit form of the
BFKL kernel we find the following eigenvalues,
1+n 1+n
En ,ν = 2ψ(1) − ψ + iν − ψ − iν . (18.25)
2 2

The maximal eigenvalue corresponds to n = ν = 0 or h = h̄ = 12 E0,0 = 4 ln 2.

This maximal eigenstate defines the right-most singularity of the partial wave
amplitude. This determines the asymptotic behavior of the scattering amplitude
in the leading logarithmic approximation,
αs N
A(s, t) = is1+ π 4ln2
(18.26)
which is referred to as the BFKL Pomeron. Using the explicit form of the eigen-
value one can reconstruct the operator form of HB F K L acting on the represen-
tations of the SL(2, C) group,
1
HB F K L = [H(J12 ) + H(J¯12 ) H(j) = 2ψ(1) − ψ(j) − ψ(1 − j), (18.27)
2
where L212 = J12 (J12 − 1), and similarly for L̄212 . This is a special case of the
Heisenberg spin chain of spin s operators whose Hamiltonian takes the form,

L
Hs = H(Ji,i+ 1 ) Ji,i+ 1 (Ji,i+1 + 1) = (Si + Si+1 )2 , (18.28)
i
where Ji,i+ 1 is related to the sum of two spins of the neighboring sites, Si2 =
s(s + 1) and H(x) is the following harmonic function,

2s−1
1
H(x) = = ψ(2s + 1) − ψ(x + 1). (18.29)
1+l
l=x
To connect it to the analysis of Section 5.14 we check this for s = 1/2. For this
case Ji,i+ 1 can take one of the two values 0, 1 for which we have H(0) = 1 and
H(1) = 0, so that the Hamiltonian is a projection into Ji,i+1 = 0 subspace with
H(Ji,i+ 1 ) = 14 − Si · Si+ 1 which is identical to (5.158).
One can generalize the exchange of colorless boundstates of two reggeized
gluons to exchange of multireggeon boundstates built from Nr reggeized gluons.
This is beyond the scope of this book and can be found for instance in [34].

19
Large N methods in QCD4
We have encountered the large N expansion, or the N1 expansion, in Chapter 7 in

the context of two-dimensional field theory, and in particular its application by
’t Hooft to solve the mesonic spectrum of QCD in two dimensions [10]. A natural
question is thus what does this limit tell us about four dimensional QCD, and in
particular can one also solve the mesonic spectrum of QCD in four dimensions in
the large N approximation. These questions will be the topics of this chapter. We
start with the rules of counting powers of N in four-dimensional QCD, and the
relations between Feynman diagrams in the double line notation and Riemann
surfaces. We then briefly discuss certain applications of the expansion to the
mesonic physics and then follow Witten’s seminal analysis of baryons in the
planar approximation [222].
The large N technique was introduced by ’t Hooft in [122]. Since then there
have been many follow-up papers and there is a very rich literature on large N
approximation including review papers and books like [223], [66], [165], [46] and
[160]. In this chapter we use mainly the latter.
19.1 Large N QCD in four dimensions

Let us remind the reader the basic notations and the classical action of QCD in
four dimensions. The two-dimensional ones were presented in (8),

1
SQ C D = d x − Tr[Fμν F ] + Ψ̄i (i D − mi )Ψi ,
4 μν
(19.1)
2
where the gauge fields are spanned by N × N Hermitian matrices T A such

μ T , Fμν = ∂μ Aν − ∂ν Aμ + i N [Aμ , Aν ], the covariant derivative
that Aμ = AA A √g
g
Dμ = ∂μ + i √N Aμ , the fermions Ψ are in the fundamental representation of the
color group and i = 1, . . . , Nf indicates the flavor degrees of freedom. The gauge
coupling was chosen to be √gN , to accommodate a large N approximation with
g fixed. This can be shown for instance in applying the large N expansion to the
β f unction. The latter, when g is used in the covariant derivatives, is given by
(17.73),
3
dg 11 2 g
μ =− N − Nf + O(g 5 ), (19.2)
dμ 3 3 16π2

338 Large N methods in QCD4
(a) (b)
(c) (d) (e)
Fig. 19.1. Four-dimensional Feynman rules in the usual form and in the double
line notation.
which obviously is not suitable for a large N expansion whereas if one replaces
g → √gN the β function takes the form,
3
dg 11 2Nf g
μ =− − + O(g 5 ). (19.3)
dμ 3 3N 16π2
The rules of the Feynman diagrams in two-dimensional QCD (see Fig. 10.1)
include the fermion propagator, the gluon propagator and the quark gluon vertex,
all expressed in the double line notation. Using the light-cone in two dimensions
one eliminates the three- and four-gluon vertices. In four dimensions due to the
transverse directions the gluon vertices cannot be eliminated by choosing a gauge.
Thus all together the four-dimensional Feynman rules are expressed in Fig. 19.1.
The figures a,b,c are identical to the two-dimensional ones (Fig. 10.1) whereas
figures d and e are the three- and four-gluon vertices. The quark propagator
(19.1a) is given by,
<ψ a (x)ψ̄ b (y)> = S(x − y)δ ab . (19.4)

Fig. 19.2. Color flow in the double line notation associated with two traces.
The SU (N ) gluon propagator in the double line notation reads,

1 1 a c
<(Aμ )ab (x)(Aν )cd (y)> = Dμν (x − y) δda δbc − δ δ . (19.5)
2 N b d
For a U (N ) gauge group the propagator does not include the term which is
proportional to N1 . One can view the SU (N ) color indices as those of a U (N )
theory plus an additional “ghost” U (1) gauge field that cancels the contribution
of the U (1) gauge field in the U (N ) gauge group. For many applications the
distinction between the U (N ) and SU (N ) in the 1/N expansion is sub-leading.
The three and four gluon vertices (see Fig. 19.1 d,e) emerge obviously from the
Tr[Fμν F μν ] term in the action. Note that it is a single trace operator and hence
the four-gluon vertex is the one depicted in (19.1 e) and not in Fig. 19.2 which
corresponds to the color flow of a two-trace operator.
To compute the N dependence of Feynman diagrams it is convenient to re-scale
the gluon field and the quark field as follows,
gAμ √
√ → Âμ ψ → N ψ̂ (19.6)
N
In terms of the re-scaled field the QCD action reads,
⎡ ⎤
1
Nf
¯
L = N ⎣− 2 Tr[F̂μν F̂ μν ] + ψ̂i (i D − mi )ψ̂i ⎦ . (19.7)
2g i= 1
From this Lagrangian we can read off the powers of N and λ ≡ g 2 = N of each

part of a Feynman diagram. The vertex operator scales like N , the propagator as
1
N and every color index loop gives a factor of N . If we combine the dependence
on λ we find for the gluon, that the vertex behaves as Nλ , the propagator as Nλ
and the color loop as N , while for the fermions neither the propagator nor the
quark-gluon vertex depend on the coupling.
Thus a connected vacuum diagram with V vertices, E propagators, namely
edges and F loops, namely faces, is of order (see (7.7)),
N V −E +F λ(E ( G ) −V ( G ) ) = N χ λ(E ( G ) −V ( G ) ) , (19.8)

λ N2
λ2N 2
λ3N 2
λ2
Fig. 19.3. Examples of Feynman diagrams with only gluons.
where,
χ ≡ V − E + F = 2 − 2h − b, (19.9)
is the Euler character of the surface, h is the genus, namely, the number of
handles and b is the number of boundaries. E(G ) and V(G ) are the appropriate
qualities for gluons. For instance the sphere has χ = 2 since it has no handles and
no boundaries, the disk has χ = 1 since it has no handles and one boundary and
the torus has one handle and no boundaries and hence it has χ = 0. Thus the
Feynman diagrams look like triangulated two-dimensional surfaces. In fact all
possible gluon exchange may fill the holes of the triangulated structure forming
a smooth surface with no boundaries for gluon only diagrams, and with bound-
aries for diagrams that include quark loops. It was conjectured that the two-
dimensional surface is the world sheet of a string theory which is dual to QCD.
There has been tremendous progress in this string/gauge duality in recent years
following the seminal AdS/CFT duality of Maldacena [158]. This is beyond the
scope of this book and we refer the reader to the relevant literature, for instance
the review [10].
To further demonstrate the determination of the order of a diagram consider
first the diagrams that involve only gluons which appear in Fig. 19.3. The dia-
gram in (a) has V = 2, E = 3, F = 3 and thus it behaves as N 2 λ. Similarly in
(b) and (c) V = 4, E = 6, F = 4 and V = 5, E = 8, F = 5 so that they behave as
N 2 λ2 and N 2 λ3 , respectively. The three diagrams (a), (b) and (c) are all planar
diagrams and have a topology of a sphere. Note however that diagram (d) which
is non-planar behaves as N 0 λ2 , namely of genus one. In the large N limit this
last diagram is obviously suppressed.
So far we have only discussed diagrams with gluons. Quarks propagators are
represented (see Fig. 19.1a) by a single line. A closed quark loop is a boundary
and hence using (19.9) it contributes to the diagram a factor of N1 . Consider for
example the diagram drawn in Fig. 19.4.

Fig. 19.4. Four-dimensional Feynman rules in the usual form and in the double
line notation.
(a) (b) (c)
Fig. 19.5. Non-planar diagram with gluons and quarks.
It is a diagram of order N . This follows trivially from the fact that it has zero
genus, h = 0 and one boundary b = 1. Alternatively we have one gluon vertex,
three gluon propagators and three loops and hence N 1−3+3 = N . Obviously this
is also the result when one uses the unrescaled operators where each vertex con-
tributes √1N and each index loop N , so that we get ( √1N )4 × N 3 = N . Similar to
the non-planar gluon diagram (19.3d), Fig. 19.5 describes a non-planar diagram
that includes both gluons and quarks. This diagram scales like N1 since there is
no gluon vertex, two gluon propagators and one index loop N 0−2+1 = N1 .
As was mentioned above the difference between the SU (N ) case versus the
U (N ) can be accounted by adding a ghost U (1) gauge field whose role is to cancel
the extra U (1) part of the U (N ). The U (1) commutes with the U (N ) gauge fields
and therefore does not interact with them and hence one has to incorporate only
the coupling of the quark fields to the U (1) gauge field. When we consider a con-
nected diagram with gluons and U (1) ghost gauge fields the contribution to the

Fig. 19.6. U (1) ghost propagator connecting two otherwise disconnected diagrams.
counting of orders of N due to the gluons is not affected, whereas each U (1) ghost
field contributes a factor of N12 . The latter follows from the fact that the ghost
U (1) propagator contributes a factor of 1/N and another 1/N factor due to the
two coupling constants at the end of the propagator. This can easily be seen from
the unrescaled action (19.1). For diagrams that are connected with the ghost field
and otherwise disconnected as is depicted in Fig. 19.6 the situation is different.
For instance that diagram is of order N 0 since it has N × N × N12 . Note however
that even in this case there is a difference of order N12 between the SU (N ) and
U (N ) cases, (actually, in Fig. 19.6 the contribution of SU (N ) vanishes).
19.1.1 Counting rules for correlation functions

So far we have described the counting rules for vacuum diagrams. We now
proceed to the counting rules of correlation functions of gluons and quarks
which are vacuum expectation values of gauge invariant operators made out
of gluon and quark fields. The latter should be color singlets, not necessar-
ily local, that cannot be split into color singlet pieces. Thus operators like
y
ψ̄ψ, T r[Fμν F μν ], ψ̄(y)e−i x A μ (z )dz ψ(x) are allowed whereas (ψ̄ψ)2 is not. As
μ
usual the procedure to compute correlation functions of such operators is to add

appropriate source terms to the action and differentiate the generating function
with respect to sources that correspond to the operators. If we denote by Ôi a
gauge invariant operator made out of the rescaled fields, we shift the Lagrangian

density as follows, L0 → L0 + i N Ji Oi , where L0 is the Lagrangian density
without the sources and Ji is the source that corresponds to Oi . Thus any cor-
relator can be determined as follows,
1 ∂ 1 ∂
<Ô1 . . . .Ôn > = ... W (J). (19.10)
iN ∂J1 iN ∂Jn
In terms of N counting, for correlation functions of only gluon fields W (J) is of
order N 2 and hence the correlator is of order N 2−n . The leading order of W (J)
in the case where quarks fields are also involved is of order N which means
that the correlator is of order N 1−n . Let us denote by Ĝi and M̂i glueball and
meson gauge invariant operators, respectively built from the rescaled fields Âμ
¯
and ψ̂, ψ̂. The leading order in N of
√ the various correlators are summarized in
the following table. The operator N M̂ is the operator that creates a meson

Table 19.1. The N counting for glueball and

meson correlators
Correlator N counting
<Ĝ1 Ĝ2 >c N0

<Ĝ1 . . . Ĝn g>c N 2 −n g
√ <M̂1 √ M̂2 >c N −1
< N M̂1 N M̂2 >c N0
√ √ n
1 − 2h
< N M̂1 N M̂n h>c N
√ √ nh
<Ĝ1 . . . Ĝn g N M̂1 N M̂n h>c N 1 −n g − 2
with a unit amplitude. In particular we read from the table that the glueball
meson interaction is of order √1N .
19.2 Meson phenomenology

The picture that emerges from N counting is that of mesons and glueballs inter-
acting weakly with a coupling of √1N . At the tree level the singularities are
poles. At one loop, namely at order N1 the singularities are two particle cuts, at
two loops three-particle cuts and so on. We now describe certain phenomena of
meson physics that are accounted for by N1 arguments and quite often cannot
be explained in any other way.
r The spectrum at low energies of QCD in the large N limit include infinitely
many narrow glueball and meson resonances. The fact that the number of res-
onances is infinite follows from the need to reproduce the logarithmic running
of QCD correlation functions. A meson two-point function can be written as a
sum of resonances,
Zi
d4 xeiq x <M (x)M (0)>c = , (19.11)
q − m2i
2
since single meson exchange dominates in the large N limit. The logarithmic
dependence on q 2 of the left-hand side can be recast only provided that the
sum on the right-hand side includes infinitely many terms. The resonances are
narrow since their decay width goes to zero in the large N limit. This follows
from the fact that the phase space factor is N independent and the coupling
constant behaves like √1N .
r In Chapter 17 we encountered the pion decay constant fπ . Let us check how
it scales with N . Recall its definition <0|ψ̄γ5 T A√
ψ|π b (p)>= ifπ pμ δ ab . The cor-
responding gauge invariant correlator is <N M̂1 N M̂2 >, where√the first oper-
ator N M̂1 corresponds to the axial current and the second N M̂2 to the

q q
q q
q q
q q
Fig. 19.7. Zweig’s rule for the decay of a meson into two mesons.
pion√produced from the vacuum with a unit amplitude. This correlator scales
like N and hence,
√
fπ ∼ N . (19.12)
r The suppression of exotic states of the form q q̄q q̄ and the fact that the meson
is almost a pure q q̄ state with little impact of the q q̄ sea are straightforward
consequences of large N . Since a quark loop, as we have seen above, is sup-
pressed by a factor of N1 the q q̄ sea is irrelevant. Since in the leading order the
mesons are non interacting in large N , there is no interaction that will bind
two mesons into a q q̄q q̄ exotic state.
r Consider the two diagrams of Fig. 19.7.
Using the counting rules it is obvious that the right-hand diagram is N1 sup-
pressed in comparison to the one on the left. Correspondingly the meson will
preferably decay into two mesons of the left-hand side of the figure, what is
referred to as Zweig rule conserving decay, and not to the two mesons on the
right which is a Zweig rule suppressed decay. In this sense large N predicts the
Zweig rule. The same mechanism is in charge of the fact that there is almost
flavor singlet and octet degeneracy. In the leading order in large N the whole
nonet is degenerate since the diagrams that split singlets from octets involve
a q q̄ annihilation which is order of N1 . In the large N for instance the vector
mesons (ρ, w, φ, K ∗ ) are degenerate.
r It is known that meson decay proceeds mainly via decay into two body states
and not into states of more mesons. Large N tells us that the decay into two
mesons behaves as √1N , whereas a decay into four mesons is of order N 13 / 2 .
This can also be compared to the decay of a meson via creation of a quark
anti-quark pair in the mesonic flux tube [60].
r The N counting rules tell us also that meson scattering amplitudes are given by
an infinite sum of tree diagram of exchange of physical mesons. This fits nicely
the so-called Regge phenomenology, where strong interactions are interpreted
as an infinite sum of tree diagrams with hadron exchange.
r Another very important phenomenon is related to the axial U (1), the theta
term and the mass of the η . This will be described in detail in Section 22.5,
but here we present the picture in the large N .

19.2.1 Axial U(1) and the mass of the η

Consider the full action of four-dimensional YM theory, which includes also the
θ term,

1 θg 2
SYM = d4 x − Tr[Fμν F μν ] + Tr[F μν F̃ μν
] . (19.13)
4 16π2 N
The normalization of the θ term is such that θ is an angular variable, namely,
under the shift of θ → θ + 2π the action is shifted by 2πn where n is some integer,
so that eiS is unchanged. This result follows from the quantization of the θ term,

g2
d4 x Tr[Fμν F̃ μν ] = integer. (19.14)
16π2 N
We would like to explore the dependence on θ of the vacuum energy in the
pure YM theory and in the theory with massless quarks. In particular we would
2
like to determine ddθE2 |θ = 0 .
Using the path integral formulation we find that,
2
d2 E 1 g2
|θ = 0 = d4 x <T (Tr[Fμν F̃ μν (x)]Tr[[Fμν F̃ μν ])(0)> . (19.15)
dθ2 N 2 16π2
Let us introduce an IR cutoff and take,

2
d2 E g2
|θ = 0 = lim U (k)
dθ2 16π2 N k →0

U (k) = d4 xe(ν k x) <T (Tr[Fμν F̃ μν ](x)Tr[[Fμν F̃ μν (0)])> .
(19.16)
It is easy to check that in perturbation theory U (k) is of order N 2 due to
the contribution of the N 2 degrees of freedom of the gluons. However, perturba-
tively, limk →0 U (k) = 0 since F F̃ is a total derivative. One concludes that per-
turbatively the vacuum energy is θ independent. To better understand (19.16)
we rewrite U (k) in terms of a sum over intermediate single particle states,
N 2 (agb )2 N (am es )2
n n
U (k) = + , (19.17)
k 2 − (mgb )2n m es
k 2 − (m
m es )2n
gb
√
where gb stands for glueball and mes for meson. N agb and N am es are the
amplitudes for Tr[F F̃ ] to create a glueball and meson state, respectively. This
result again follows from the N counting rules,
√
<0|Tr[F F̃ ]|mes>∼ N <0|Tr[F F̃ ]|gb>∼ N. (19.18)
The fact that only single states and not multi-states are taken in the interme-
diate states is since the latter are suppressed in the large N . In the pure YM
2
without quarks the first term vanishes and hence U (0) ∼ N 2 and ddθE2 |θ =0 ∼ 1.
In the presence of massless quarks we know that there could not be any θ depen-
dence and thus we should be able to show that the first term is cancelled out.
However, it seems that there is no way that the second term can cancel the first.

Fig. 19.8. Perturbative correction to the free propagator due to an exchange

of one gluon (left) and two gluons (right).
In fact it is possible if there is one meson state with mass mm es ∼ √1N and if the
two terms have opposite sign. This obviously can cancel the k = 0 term in U (k)
and does not cancel for non-trivial k, but this is exactly what enters (19.16). The
opposite sign follows from the fact that an additional equal time commutator
term has to be added to (19.15) (see the appendix of [221]). Assuming such a
state with mass mm es ∼ √1N the form of U (0) is,
a2η
U (0) = N . (19.19)
Mη2
Using the axial anomaly equation which will be further discussed in (22.5),
g2
∂μ J5μ = Nf Tr[F F̃ ], (19.20)
4π2 N
we get,
g2 N N
<0|Tr[F F̃ ]|η > = <0|∂μ J5μ |η >= fη Mη2 , (19.21)
8π2 2Nf 2Nf
From this relation we get the Veneziano–Witten formula or the mass of the η ,
2
2N(f ) d2 E
Mη2 = |θ =0 . (19.22)
f(π ) dθ2
The picture that emerges from this discussion is that the η is a Goldstone boson
in the large N limit. It has a mass of the order of Mη ∼ √1N . The dependence
on η of nonzero amplitudes can be obtained from the dependence on θ in the
theory without quarks by the following replacement,
2N(f )
θ→θ+ η . (19.23)
f(η )
1
Note that f(η ) = f(π ) to leading order in N .
19.3 Baryons in the large N expansion

Whereas we have seen that the large N expansion is very useful in discussing
mesons, it may seem that it is not the case for baryons. Baryonic diagrams
depend on N both via the combinatorial factors associated with the diagrams,
as well as the fact that the diagrams themselves include N quarks.
The problem is clearly demonstrated when computing the perturbative cor-
rection to the free propagator of an N quarks state. The correction occurs due to
an exchange of a gluon between two quarks (see Fig. 19.8). The gluon exchange

diagram scales as N1 . However since there are 12 N (N − 1) possible pairs the net
effect is of order N . In a similar manner the exchange of two gluons is of order
( √1N )4 N 4 ∼ N 2 , where the first factor comes from the four vertices and the sec-
ond from the number of ways to choose the four quarks. Higher-order exchange
diagrams will have higher order divergence in large N . We will now show, that
in spite of this fatal obstacle, there is a large N approximation to the problem of
the baryons. The idea is to divide the problem into two parts, in the first one uses
diagrammatic methods to study the problem of n quark interaction, and then
the effect of these forces on an N quark state. Let us first apply this approach
for determining the dependence of the mass of the baryon on N in the quark
model. Assuming that the mass gets contributions from the quark masses, quark
kinetic energy and quark–quark potential energy, the mass of the baryon reads,

1
MB = N m q + T q + V q , (19.24)
2
where mq is the quark mass, Tq is the kinetic energy of the quark and Vq is
the quark–quark potential energy. Thus we observe that the mass of the baryon
scales as N . This result will be shown to hold even beyond the quark model.
Again we have made use of the fact that the potential energy is combined from
the N 2 combinatorial factor and the N1 factor that comes from the vertices, or
gluon propagator.
Leaving aside the quark model, we want to address first the baryonic system
made out of very heavy quarks.
19.3.1 The Hartree approximation

In the baryons, the quarks are anti-symmetric in color. Thus they are symmetric
in flavor, space and spin combined. Hence they act like bosons. A natural frame-
work to analyze such a system of bosonic charged particles that are subjected to
a central potential is the Hartree approximation, in which each particle moves
independently of the others in a potential which is determined self consistently by
the motion of all the other particles. The justification of the use of this approxi-
mation is the large N limit which renders the interactions to be weak. Therefore
neglecting the fact that the particle trajectory affects the state of all the other
particles and hence the potential that it feels, is justified. Also it is obvious that
taking the potential created by all particles and not the one created by all the
particles apart from the one we consider, is a N1 effect. Since, as mentioned above,
the particles in the non-color degrees of freedom are bosons it implies that in
the ground state of the baryon all the particles will sit in the ground state of the
Hartree potential.
Let us take the Hamiltonian of the system to be,
1 1 2 1 3
H= |pa |2 + V (ra , rb ) + V (ra , rb , rc ) + . . . ,
2m a 2N 6N 2
a= b a= b= c
(19.25)

where we have suppressed the flavor and spin degrees of freedom, V n stands
for the n body interaction and is independent of N and its strength is of order
N 1−n , as explained above. In fact since the number of clusters of n quarks is of
order N n , each term in the Hamiltonian is proportional to N .
Next one takes for the ground state wave function a product of the wave
functions of each of the particles, namely,
ψ(ra , . . . rN ) = φ(ra ) (19.26)

a
where the particle wave functions are determined by a variational method. The
expectation value of the Hamiltonian,

1 3 2 1
<ψ|H|ψ> = N d r|∇φ| + d3 r1 d3 r2 V 2 (r1 , r1 )|φ(r1 )φ(r2 )|2 +
2m 2

1 3 3 3 3 2
+ d r1 d r2 d r3 V (r1 , r2 , r3 )|φ|(r1 )φ(r2 )φ(r3 )| + . . . , (19.27)
6
has to be minimized with respect to φ(r) subjected to the constraint that,

d3 r|φ|2 = 1. (19.28)
The minimization translates to,

∇2
− + V (r) φ = φ, (19.29)
2m
where is the Lagrange multiplier associated with the constraint, and the Hartree
potential is,

1
V = d3 r1 V 2 (r, r1 )|φ(r1 )|2 + d3 r1 d3 r2 V 3 (r, r1 , r2 )|φ(r1 )φ(r2 )|2 + . . . .
2
(19.30)
We will now treat first the case of heavy quarks and subsequently will address,
in a less rigorous manner, the case of light quarks.
19.3.2 Baryons made out of heavy quarks

A non relativistic Schrodinger equation is an adequate framework to deal with
very heavy baryons. In this setup for short distances the quark–quark potential
is an attractive Coulomb potential so that the Hamiltonian takes the form,
N
∂i2 g2 1
H = N mq − − . (19.31)
i= 1
2m q N i< j
|xi − xj |
The system is effectively that of N bosons with a Coulomb interaction with a

strength of N1 .

In spite of the fact that the interaction potential behaves like N1 , we can-
not treat this term as a perturbation since each quark interacts with N other
quarks and hence the total quark–quark interaction of each quark is of order one.
This situation calls for a Hartree approximation where, as explained above, the
quark is exposed to an average effective potential. The fluctuations of the effec-
tive potential are negligible and hence we can consider a background c-number
potential.
For heavy quarks where the potential is taken to be a Coulomb potential we
find,

)| |φ(r
d3 r |φ(r |r
2
g2 )|2
<ψ|H − N |ψ>= N [M + 1
2m d3 r|∇φ|2 − 2 d3 r −r |
− d3 r|φ(r)|2 ]. (19.32)
The main point here is that each of the terms is proportional to N and hence
the result of the minimization is N independent. The variation with respect to
φ∗ results in the following Schrodinger equation,

∇2 φ |φ(r )|2
− − g φ d3 r
2
= φ. (19.33)
2m |r − r |
One can convert this integro-differential equation into a fourth-order differential
equation
1 2 ∇2 φ
− ∇ + 4πg 2 |φ|2 = 0. (19.34)
2m φ
For radial solutions, for instance, the ground state of this equation takes the
form,
2
1 d 2 1 d2 2
− + ddr + ddr φ + 4πg 2 |φ|2 = 0, (19.35)
2m dr2 r φ dr2 r
which is derived by dividing (19.33) by φ and acting with ∇2 .

Even without solving this equation, it is clear that the mass of the baryon is
linear with N and that the charge distribution of the baryon which implies in
particular its size is N independent.
19.3.3 Baryons made out of light quarks

Up to this point we have used a non relativistic Hartree approximation which is
valid only for heavy quarks. However, phenomenologically, one is more interested
in baryons made out of light quarks and in scattering processes that involve
relativistic quarks. We will show now that even for the light quark baryons, a
Hartree-like approximation, namely that each quark moves independently of the
others in a potential which is determined self consistently by the motion of all
the other particles, is still justified. Moreover, it will be argued that just as for the

q q
Fig. 19.9. Baryon stringy configuration for N = 3.
heavy quark baryons, also for the baryons made out of light quarks, the mass is
linear with N , whereas the size and the shape of the baryon are N independent.
A major difference between the case of heavy quarks versus that of light ones
is that in the latter case one has to introduce on top of the two-body interaction
also a three-body interaction and, in general, n-body interactions. In addition one
has to use a relativistic analog of the Hartree approximation. In two dimensions
one can solve the relativistic Hartree approximation. Unfortunately, the four-
dimensional analog is not known. Let us first discuss a non relativistic Hartree
approximation with n-body interactions and then argue about the relativistic
analog. The Hamiltonian for the case with any n-body interaction takes the form,
1 1 2 1 3
H= |pa |2 + V (ra , rb ) + V (ra , rb , rc ) + . . . ,
2m a 2N 6N 2
a= b a= b= c
(19.36)
n
where we have suppressed the flavor and spin degrees of freedom, V stands
for the n-body interaction and is independent of N . The strength of V n is of
order N 1−n since breaking the n quark line costs a factor of N −n and since the
baryon is in a totally antisymmetric representation, each quark line carries a
different color index.
We now substitute this Hamiltonian into <ψ|H|ψ> and use a variational
method as above. Since for each V n term there are N n ways to choose a set
of n quarks, here again the expectation value of the Hamiltonian is linear in N .
Next we have to introduce a four-dimensional relativistic Hartree approxima-
tion. In two dimensions in the large N limit the Hartree approximation is exact.
The generalization to four dimensions, however, is not known and hence one can
make only the qualitative statement that even in this case the mass is linear in
N and the size and shape are independent of N .
The Hartree approximation of light quarks moving in an effective potential
can be also related to a string model of the baryon. In this model, the N quarks
are attached to a common junction as can be seen in Fig. 19.9 for the case of
N = 3.1 In the large N approximation the junction can be regarded as a heavy
object and its motion can be ignored. The interaction of the quarks with the fixed
junction can be thought of as an interaction with an effective Hartree potential.
1 The modern picture of the latter is that of a wrapped D brane.

19.3.4 Baryonic excited states

Low-lying excitations of the baryons are described by wave functions where out
of N single particle states, a number nk N of states are placed in the kth
excited state of the Hamiltonian. The corresponding mass of the excited baryon

is M = M0 + n k nk k where k is the energy of the kth excited state.
Highly excited states have a finite fraction of single particle excited states.
Denote by p this fraction, namely there are (1 − p)N particles in the ground
state and pN ones in excited states which we take to be φ1 so that the baryon
wave function is,
pN (1−p)N
ψ(ra , . . . rN ) = (−1)P φ1 (ra ) φ0 (ra ). (19.37)
a a
Inserting this ansatz into the expectation value of the Hamiltonian one gets a set
of two coupled nonlinear equations for φ0 and φ1 . This structure can obviously
be generalized to states with higher single particle excited states.
Another approach to studying excited states is to apply a time-dependent
Hartree approximation. It is easy to check that starting with the Hartree ansatz
for the wave function but now with single particle wave functions that are
also time dependent, one finds instead of (19.33) the following time-dependent
Schrodinger equation,

∇2 φ |φ(r )|2
− − g 2 φ d3 r = i∂t φ(r, t). (19.38)
2m |r − r |
This equation is solved by φ(t, r) = e−it φ(r) where φ(r) is a solution of the time-
independent equation. By Galilean boosting along, for instance the x direction,
a static baryon solution, we find the solution,
1 2
φ(r, t) = φ(x − vt)ei(M v x−t− 2 M v t)
, (19.39)
which is a baryon travelling with a constant velocity. This is an additional solu-

tion to the time-dependent equation. In fact starting with any function φ(r, 0)
and substituting it into (19.38) a new solution will be generated. These solutions
will generically be excited states, but not in energy eigenstates, since they will
not have a harmonic form.
To generate excited baryon solutions which are in eigenstates of the energy, we
make use of the DHN procedure discussed in Section 5.5.1 in the context of two-
dimensional field theories. The idea is to look for solutions which are periodic in
time and to quantize them by requiring that the action during a period will obey,
T
dt <ψ|H − i∂t |ψ> = 2πn, (19.40)
0
where n is some integer number. Recall that this condition follows from the fact
that the solutions are invariant under time translations, so from ψ(t) we can
also generate a solution ψ(t − t0 ) for any t0 and also any linear combinations of
T
them, and in particular a harmonic varying solution 0 dt0 e−it 0 E ψ(t − t0 ).

i j i i
i j i i
Fig. 19.10. Baryon–baryon scattering. Exchange of a quark on the right, while

on the left, such an exchange plus an exchange of a gluon.
In analogy to the discussion in Section 5.5.1 here as well one can introduce a
non-abelian flavor group, namely construct baryons made out of several flavors.
In this case one introduces into the Hartree wave function a separate single
particle wave function for each flavor.
For very heavy quarks one can neglect the spin-dependent forces, and hence
anticipate that the baryons are spherically symmetric. However, for less heavy
quarks this is no longer the case. For a baryon made out of a single flavor, in the
ground state all the spins are aligned and hence the total spin is 12 N . Due to the
fact that the total spin is very large, the effect of the coupling of this large spin
to the orbit is significant and hence the ground state will no longer be spherically
symmetric. If one takes the large N analog of the baryon to be composed of N 2+1
quarks of one flavor and N 2+ 1 − 1 of the other flavor, then the net spin will be
1
2 since the spin–spin interaction will align the spin of the different flavors in
an antiparallel way. Unlike the one flavor case where the spin is N2 , the spin 12
will be too small to affect the spherically symmetric ground state via spin orbit
interaction, and hence for that case it will remain symmetric.
19.4 Scattering processes

In Section 19.2 it was shown that in the leading order of the large N there
is no meson–meson scattering. The same applies also for meson–glueball and
glueball–glueball scattering. Let us now address the question of baryon–baryon
and baryon–meson scattering.
Baryon–baryon scattering is dominated by an interchange of one quark
between two baryons. Whether the process involves only an interchange or also
in addition an exchange of a gluon, as is shown in Fig. 19.10, the amplitude is of
order N . In the case of no gluon exchange, there is a choice of the interchanging
quark in one of the baryon, which goes like N . Once a quark in one baryon
is chosen it can be interchanged only with a quark in the second baryon that
carries exactly the same color index, hence there is no additional N dependence.
Thus altogether the amplitude is order N . Note also that the diagram (19.10)
comes with a factor of (−1). The amplitude for an interchange that is accom-
panied with an exchange of a gluon is also of order N , which follows from the
fact that there is a factor of N from choosing the quark in the first baryon,

19.4 Scattering processes 353
i j
i j
Fig. 19.11. Annihilation of a quark coming from the baryon and anti-quark
coming from the anti-baryon.
another factor of N from the other baryon and a factor of N1 from the quark
gluon vertices.
This fact that the amplitude is order N is behind why there is a smooth large
N limit to the baryon–baryon scattering. Recall that the mass of the baryon and
hence also the non-relativistic kinetic energy of the baryon are linear in N . Thus
the total Hamiltonian can be written as H = N Ĥ where Ĥ is N independent.
The eigenvectors of Ĥ and hence of the scattering process are N independent.
Quantitatively one addresses the question of baryon–baryon scattering using a
non relativistic time-dependent Schrodinger equation for a system of 2N quarks.
Due to the exclusion principle the total wave function should be a product of
two orthonormal space and spin wave functions φi (x, t) where i = 1, 2 in the
following way,
N N
ψ(x1 , . . . , x2N , t) = (−1)P φ1 (xi , t) φ2 (xj , t). (19.41)
P i= 1 j =1
Using again the time-dependent variational principle, we find that in the case
where all the spins of the quarks are parallel so we can ignore them,

∇2 dyφ∗1 φ1 (y, t)
i∂t φ1 (x, t) = φ1 (x, t) − g 2 φ1 (x, t)
2M |x − y|
∗
dyφ 2 φ1 (y, t)
−g 2 φ1 (x, t) , (19.42)
|x − y|
and another equation where φ1 ↔ φ2 . Apart from the last term this equation
is identical to the one describing a single baryon (19.38), hence the last term
obviously describes the interaction between the two baryons. To describe baryon–
baryon scattering we start with inital conditions where the wave functions φ are
localized at two far away regions of space, but heading for a collision. When the
two wave functions overlap the interaction term is important and determines the
scattering via (19.42).
The baryon anti-baryon scattering is dominated by an annihilation of a quark
coming from the baryon and an anti-quark from the anti-baryon. The amplitude
of this process is of order N since choosing one quark is order N , choosing an
anti-quark is order N and the coupling is order N1 (see Fig. 19.11). Again this
is like the scaling of the kinetic term and hence there is a smooth limit. The

Fig. 19.12. Exchange of a quark and a gluon in meson–baryon scattering.
variational procedure now involves a wave function composed of N quark and N

anti-quark wave functions, namely,
ψ(ra , . . . , rN ) = φ(ra ) φ̄(r̄b ), (19.43)
a b
where φ̄(r̄b ) is the wave function of a single anti-quark. The minimization now
yields a pair of coupled equations for φ and φ̄.
The meson–baryon scattering is described in a diagram like Fig. 19.12. The
diagram is of order N 0 since there is a factor of N1 from the coupling and N
from the number of ways to choose the quark from the baryon. Recall that the
baryon kinetic energy is order N and that of the meson is order one. Hence the
interaction term is negligible from the point of view of the baryon and it does
not feel the meson but the meson motion is affected by the interaction and thus
there is a meson baryon non-trivial scattering. Denoting again the wave function
of a quark of the baryon as φ(x, t) and that of the meson as φM (xM , yM , t), the
trial many body wave function reads,
ψ(ra , . . . , rN , xM , yM , t) = φ(ra , t)φM (xM , yM , t). (19.44)
a
Again we substitute this wave function into the variational principle

dt <ψ|H − i∂t |ψ>. The solution for φ of the corresponding equations is iden-
tical to the solution of the baryon and hence indeed the baryon is not affected
by the presence of the meson. On the other hand the equation for φM is affected
by the presence of φ. The equation will be that of a free meson plus two additional
terms describing the interaction that take the form,

dzφ∗ (z, t)φM (z, yM , t) dzφ∗ (z, t)φM (z, yM , t)
Hint = −g φ(x)
2
+ .
|xM − z| |z − yM |
(19.45)
Thus the interaction term and hence the whole equation is linear in φM .

20
From 2d bosonized baryons to 4d Skyrmions
20.1 Introduction
Low energy effective actions associated with four-dimensional QCD, and in par-
ticular the Skyrme model have been very thoroughly studied with an emphasis
on both formal aspects such as anomalies as well as phenomenological ones like
the spectrum of baryons. This chapter is devoted to the Skyrme model. We first
derive the various terms of the Skyrme action. These include the sigma term,
the WZ term, the mass term and the Skyrme term. The first three terms we
have encountered already in the two-dimensional analog, the bosonized QCD
(Chapter 13) whereas the fourth one, the Skyrme term, shows up as a stabi-
lization term only in the four-dimensional case. We then construct the classical
soliton solution, the Skyrmion, of the corresponding equations of motion. Next
we determine the classical mass and radius of the baryon. In a similar manner to
the procedure taken in the two-dimensional model, we quantize the system semi-
classically. This yields mass splitting between the nucleons and the delta particles
and the axial coupling of the nucleons. Most of the discussion will be done for
SU (Nf = 2) but we will also discuss certain properties of the three-flavor case.
The Skyrme model was introduced in [195], [196], [197] and [91].
The topic of baryons as Skyrmions was discussed and reviewed in several
books [53], [157] and reviews [22], [186], [231]. In several sections of this chapter
we follow the latter review.
20.2 The Skyrme action

In two dimensions using the bosonized version of QCD, we were able to integrate
in the strong coupling limit the color degrees of freedom and derive the low energy
effective action of the flavor degrees of freedom. The latter took the form of a
WZW action for the group U (Nf ) of level Nc for massless QCD and modified
flavored WZW action with a mass term for massive QCD. The main point there
was that the action derived was exact. In four dimensions the situation is quite
different. For once we do not have a bosonized version of QCD which enables
us (at least in the massless case) to decouple the flavor and color degrees of
freedom and then integrate over the latter. However, due to confinement, the low
energy degrees of freedom of four-dimensional QCD are also only flavor degrees
of freedom and hence it is natural to use an Nf × Nf group element g(xμ )
of the flavor group U (Nf ). Since we do not have a systematic way to derive

356 From 2d bosonized baryons to 4d Skyrmions
the corresponding action, we will now consider various terms that eventually
construct the full Skyrme model.
20.2.1 The Sigma term

Recall that the group element g(x) transforms under left and right transforma-
tions as follows,
g(x) → U g(x) g(x) → g(x)U. (20.1)
As we have shown in two dimensions the sigma term is the term with the lowest
number of derivatives. In two dimensions it takes the form,

1
S2d = d2 x Tr [∂μ g∂ μ g −1 ]. (20.2)
12π
It is easy to see that the analog in four dimensions has the form,

1 2
S= fπ d4 x Tr [∂μ g∂ μ g −1 ], (20.3)
16
where fπ has dimensions of mass (it will be shown below that by comparison
to experimental data, for Nf = 3, it has to be taken to be ∼ 93M eV ). This
coefficient is needed since our group element still does not carry classically any
conformal dimension. The sigma term which is clearly the one with the lowest
number of derivatives can also be expressed as,

1 2 1 2
S= fπ d4 x Tr [Lμ Lμ ] = fπ d4 x Tr [Rμ Rμ ], (20.4)
16 16
where,
Lμ = g −1 ∂μ g Rμ = g∂μ g −1 . (20.5)
It is important to note that by construction the Lμ obey the so-called Maurer–
Cartan equation,
∂μ Lν − ∂ν Lμ + [Lμ , Lν ] = 0. (20.6)
and so does Rμ . Note that they are both like pure gauges in a non-abelian gauge
theory, and thus they have Fμν = 0.
Before proceeding to the WZ term let us check the symmetries of this action
in comparison with the known symmetries of QCD. It is easy to check that it
is invariant under global SU (Nf )L × SU (Nf )R × U (1)B transformations. It is
further invariant under the following three discrete transformations,
T ranspose : g ↔ gT x ↔ x, t ↔ t
P0 : g↔g x ↔ −x, t ↔ t (20.7)
(−1)NB
: g ↔ g −1 x ↔ x, t ↔ t.
The second transformation P0 is a parity transformation and the third is the
number of bosons modulo two. To check whether these discrete symmetries

are also shared by QCD we first expand g(x) around unity,

N 2 −1
2i a a
f
g(x) = 1 + T π (x), (20.8)

fπ 1
in terms of the Goldstone bosons π(x) and then consider Nf = 3. It turns out
that P0 and (−1)N B are not conserved separately but only the combination
P = P0 (−1)N B . This is demonstrated by the process K + K − → π + π 0 π − . Obvi-
ously the number of bosons modulo two is not conserved as well as the parity
P0 since the π i are pseudo scalars. It is thus clear that the action (20.3) cannot
describe the effective action of QCD and another term that does conserve P and
not P0 and (−1)N B separately. It is well known that the parity transformation
P0 is violated by a term which is proportional to the Levi Civita tensor which in
four dimensions reads μν ρσ . However, it is very easy to verify that the only term
proportional to μν ρσ , namely μν ρσ Tr [g −1 ∂μ ggg −1 ∂ν gg −1 ∂ρ gg −1 ∂σ g] vanishes
due to the antisymmetric nature of μν ρσ and the cyclicity of the trace.
20.2.2 The WZ term

Experienced with the two-dimensional WZ term it is clear that this situation
naturally calls for a four-dimensional WZ term: [225], [226]. Recall that the two-
dimensional WZ term was written as a three-dimensional integral over a three-
dimensional ball or a three disk whose boundary is the two-dimensional space-
time. In a similar manner we can construct a term defined on a five-dimensional
disk D5 whose boundary is the four-dimensional space-time and has the form,

λS
SWZ = −i d5 xij k lm Tr [g −1 ∂i gg −1 ∂j gg −1 ∂k gg −1 ∂l gg −1 ∂m g], (20.9)
240π 2 D 5
where now i, j, k, l, m denote coordinates on D5 and λS is a coefficient that has to
be determined. Extending the map g(xμ ) from the four-dimensional space-time
to the SU (N ) group manifold into a map from D5 to the group manifold is based
on the fact that π4 (SU (N )) = 0 and π1 (SU (N )) = 0. Now let us check whether
there are any constraints on λS . The analogous two-dimensional case tells us
that λS = Nc . We will now verify this result in two steps. First we show that
on general topological grounds it has to be an integer and then by relating the
action to QCD we show that this integer has to be equal to the number of colors.
To understand the topological nature of the WZ term it is convenient to use a
compactified Euclidean four space of a topology S 1 × S 3 where the S 1 factor
corresponds to a compactified time direction. Now the five-dimensional disk D5
can be taken now to be S 3 × D2 . However as is clear from Fig. 20.1 there are in
fact two options of choosing the disk D2 namely Dn2 and Ds2 . Requiring that the
independence of the physics on choice translates into,

n s
e iS W Z =e iS W Z → 0
w5 = w50 = 2π integer, (20.10)
(D n2 +D s2 )×S 3 S 2 ×S 3

M Q Q
Fig. 20.1. Two options of choosing the disk D 2 .
where we have used the fact that the sum of the two disks (Dn2 + Ds2 ) is topo-
logically equivalent to S 2 , that the five cycles in the group manifold SU (N ) of
the topology S 2 × S 3 can be represented by the cycles of topology S 5 and that
π5 (SU (N )) = Z and hence any S 5 in SU (N ) is topologically a multiple of the
basic S 5 on which w can be normalized such that S 5 w50 = 2π.
0
We thus conclude that the coefficient λS has to be an integer. As we mentioned
above in two dimensions we have shown that this integer has to be Nc . We will
show below when discussing the phenomenology of the Skyrme model that this
is the case also in four dimensions. Thus we will take from here on that λS = Nc .
20.2.3 The Skyrme term

Baryons were described in the context of the bosonized theory of two-dimensional
QCD in terms of soliton solutions of the WZW theory in flavor space (see Chapter
13). In a similar manner we anticipate that also in four dimensions solitons are
intimately related to baryons. However, in Section 5.3 it was shown that accord-
ing to Derrick’s theorem, there are no stable solitons in the space dimension
larger that one. To recapitulate this theorem let us analyze the scaling behavior
of the energy of a soliton solution in four dimensions. It is easy to realize that in
D space dimensions the energy that corresponds to the action (20.3) reads,

f2
E = dD x π Tr [Lμ Lμ ]. (20.11)
4
where a change in normalization was made.
Under a scaling of x → λx, g(x) → g(λx) since as mentioned above g(x) has
a zero scaling dimension, and the energy scales as,
Eλ = λ2−D E. (20.12)
Thus for D = 3 the energy of the system can shrink to zero by large scaling and
hence the solutions are not stable against scale transformations. To avoid this
problem we add a term which is quartic in Lμ so that the total Lagrangian is,
fπ2 1
L = L2 + L4 = Tr [Lμ Lμ ] + e2 T r([Lμ , Lν ]2 ). (20.13)
4 4

Under scale transformation the energy scales as,
Eλ = λ2−D E(2) + λ4−D E(4) . (20.14)
It is easy to see that for D ≥ 3 there is a minimum at λ = 1 and with

dEλ E(2) 4−D
=0→ =−
dλ E(4) 2−D
d2 Eλ
>0→ 2(D − 2)E(2) > 0. (20.15)
dλ2
In fact the energy is bounded from below by the topological charge. The energy
associated with the action (20.13) takes the following form for a static configu-
ration,
* 2 +
fπ 1 2
E = d x − Tr [Li Li ] − e Tr ([Li , Lj ] ) ,
3 2
(20.16)
4 4
which can be rewritten as,

fπ2 e2 √
E=− d x Tr Li L + 2 ( 2ij k L L ) ≥
3 i j k 2
4 fπ
) √ )
fπ2 ) ) √
3 ) 2 2e i j k )
− d x ) Tr ij k L L L ) = 12 2π2 efπ |B|. (20.17)
4 ) fπ )
This is referred to as the Bogomol’ny bound. It is interesting to note that unlike

other cases like instantons (see Section 22.1) there is no configuration that satu-
rates the
√ bound. The configuration that does saturates the bound has the form
Li = f 2e
π
eij k Lj Lk , however it is easy to see that it does not obey the Maurer–
Cartan equation (20.6).
20.2.4 A mass term

In two dimensions the mass term was shown to be a key ingredient to having
soliton solutions of strong coupling of QCD2 . In four dimensions this is not the
case. Stable soliton solutions do not require a mass term. However, to incorpo-
rate the fact that the pions are not massless, one adds a mass term to the full
Lagrangian. The mass term has the same form as that of the two-dimensional
theory, namely,

m2 f 2
Sm = π π d4 x Tr [g + g −1 − 2]. (20.18)
16
Upon substitution for g(x) in the ansatz (20.8) this term takes the form of a
mass term for the π fields,

1
Sm = d4 x m2π (π a )2 . (20.19)
2

For Nf > 2 one can use a mass term that breaks flavor symmetry by assigning
different masses to different flavors. One can also generalize the mass term by
using general functions of g which in the limit of g → 1 approach g.
20.2.5 Gauging the Skyrme action

In Section 9.3.1 we discussed the gauging of the WZW action. In that case we
were interested in gauging the diagonal SUD (Nc ) ∈ SUL (Nc ) × SUR (Nc ). We
presented there two methods for the gauging procedure: (i) Noether trial and
error method, (ii) covariantization of the associated currents. Here in the case
of the Skyrme action we would like to gauge the U (1) diagonal global abelian
symmetry that corresponds to electromagnetism, as well as the full SU (Nf ) ×
SU (Nf ) global symmetry of the Skyrme action. Let us first identify the diagonal
abelian symmetry that we want to gauge to incorporate EM gauge fields. In the
particular case of Nf = 3 the EM charge matrix of the u, d, s quarks is given by,
⎛ ⎞
2/3
Q=⎝ −1/3 ⎠. (20.20)
−1/3
Thus a local transformation that corresponds to the EM gauge transformation

is,
g(x) → U g(x)U −1 U ∼ 1 + i(x)[Q, g]. (20.21)
As usual the local transformation can be a symmetry of the action only provided
we add to the action gauge fields that transform under the EM gauge transfor-
mation as Aμ → Aμ − 1e ∂μ (x) where e is a unit EM charge. For the sigma term
and the Skyrme term it is obvious that gauge invariance is achieved by replac-
ing the ordinary derivative with covariant ones, namely ∂μ → Dμ = ∂μ + ie∂u .
The gauging of the WZ term Γ ≡ SW Z is more subtle and as was done for the
two-dimensional case; we use a trial and error method. First we compute the
variation of the term under the global U (1) symmetry. We find that,

Γ → Γ − d4 x∂μ (x)J μ
1 μν ρσ
Jμ = { Tr [−Q(Rν Rρ Rσ )]
48π2
+ Tr [Q(Lν Lρ Lσ )]} (20.22)
where Q is defined in (20.20). The next step in gauging the WZ term is to replace
the original term with,

Γ → Γ − e d4 xAμ J μ . (20.23)

It turns out that the action after this replacement is still not gauge invariant but
it is invariant with the following addition,

fπ2
S= d4 x Tr [Dμ gDμ g −1 ] + Nc Γ̃,
16

ie
Γ̃(g, Aμ ) = Γ(g) − e d4 xAμ J μ + d4 xμν ρσ ∂μ Aν Aρ
24π2
× Tr [Q2 (Lσ − Rσ ) − QgQg −1 Rσ ]. (20.24)
In a similar manner one can gauge the full global symmetry or its subgroups.
Since we will not need it in this chapter we refer the reader to references, for
instance [231].
20.3 The baryon as a Skyrmion

The two-dimensional solitonic baryon was analyzed at two levels, firstly the clas-
sical configuration and then at the semi-classical level. At both levels properties
of the baryon such as its mass and conserved charges were computed. In this
section we present the analogous calculations for the 4d Skyrmion and then we
compare the four- and two-dimensional results.1
20.3.1 The classical Skyrmion

The Skyrmion is by definition a solitonic solution of the Skyrme action.
Soliton solutions in two dimensions were discussed in general in Section 5.3 and
in particular the solitons of the low energy effective action of QCD2 in the strong
coupling limit in Chapter 13. In fact the classical solitonic baryons were solu-
tions of a sine-Gordon equation that was derived from an action that included a
sigma term and a mass term since for static configuration the WZ vanishes. The
latter property holds also in four dimensions so the relevant action now includes
the sigma term and the Skyrme term. In fact we will describe here the case of
two flavors and as mentioned above the WZ term vanishes for the SU (2) group
manifold anyhow. The equation of motion derived by computing the variation of
the action with respect to g −1 δg is,
e2 μ
∂ μ Lμ − 2 ∂ [Lν , [Lμ , Lν ]] = 0. (20.25)
fπ2
Obviously, an equivalent equation can be written by replacing Lμ → Rμ . Param-
eterizing the general static configuration as,
g(x) = ei τ ·r̂ F (r ) = cos(F (r)) + iτ · r̂ sin(F (r)). (20.26)
1 The classical properties of the SU (2) baryonic Skyrmion were analyzed in [133], [5] and
afterwards in many other papers.

For this ansatz the equation of motion reads,

2 sin(2F ) e2 sin(2F ) sin2 F F 2 sin(2F ) 2F sin2 F
F + F − + 8 − = 0.
r r2 fπ2 r4 r2 r2
(20.27)
The boundary conditions are taken to be,
F (r = 0) = π F (r → ∞) = 0. (20.28)
The mass of the classical Skyrmion is derived by substituting a solution of the
equation of motion into (20.16) getting,
∞
2 fπ2 2 2 sin2 F 2
2 sin F 2 sin2 F
Ms = 4π r dr F + + 4e 2F + . (20.29)
0 2 r2 r2 r2
Using the virial property this reduces to,

√ ∞ 2 dF
2
2 sin2 F
Ms = 4π 2e x dx + , (20.30)
0 dx r2
where x = ef√π 2 r is dimensionless. The value of the integral is ∼ 11.7. One can
either use the mass of the proton combined with the mass of the delta, to deter-
mine fπ and the coefficient of the Skyrme term, or use the experimental values
of fπ and the axial coupling to be discussed shortly.
Let us now analyze the radial profile of the soliton F (r). Asymptotically for
r → ∞ only the terms inside the square brackets can be neglected leading to
2
A2
a solution of the form F (r) → 16e f π2 r 2 , where again A can be determined by
comparing to experimental data and is found to be A ∼ 1.08. On the other limit
around the origin it is easy to see that the equation is solved by F (r) ∼ nπ − ar.
The numerical solution of F (r) that interpolates between these two boundary
conditions is drawn in Fig. 20.2.
In addition to the mass, we have also extracted in two dimensions from the
classical soliton the flavor properties and baryon number. For the Skyrmion we
should also be able to determine these properties as well as its spin. When we
insert the classical soliton solution in the baryon density we get,
i ij k 1 2 F
B0 = L i L j Lk = sin F , (20.31)
24π2 2π2 r2
so that the baryonic charge is,
∞
1
B = 4π drr2 B 0 (r) = (F (0) − F (∞)) + 12π[sin(2F (∞)) − sin(2F (0))] = 1.
0 π
(20.32)
where we have used the boundary conditions of F (r) specified above. Using the
distribution of the baryonic charge, one can define the rms radius of the baryon
as follows,
∞ 1/2
e
rrm s = − dxx2 sin2 F F , (20.33)
πfπ 0
which is of the order of 0.48 for the B = 1.

F (r)
0
0 1 2 3 4
r (fm)
Fig. 20.2. The numerical solution of F (r).
Fig. 20.3. The Skyrmion hedgehog configuration.
The hedgehog configuration, see Fig. 20.3, used as an ansatz for the Skyrmion,
is by construction invariant under the operation of,
K ≡ J + I = (L + S) + I, (20.34)
where L, S and J are the orbital angular momentum, the spin and the total
angular momentum, and I is the isospin.

It follows from,

τ
[K, g(x)] = i sin F [(r × −∇) , τ · r̂] + , τ · r̂ = 0. (20.35)
2
Hence the Skyrmion carries a charge of K = 0. It is straightforward to notice
that it is also an invariant under the parity operator defined in (20.7), so that
altogether it is K = 0+ state.
20.3.2 Semiclassical quantization of the soliton

Recall that the quantization of the collective coordinates of the two-dimensional
soliton was performed by elevating the static group element to a space-time
dependent one in the following way,
g(x) → g(x, t) = A(t)g(x)A−1 (t) A(t) ∈ U (Nf ). (20.36)
Nothing in this prescription is two dimensional and hence we now use the same
ansatz also for the four-dimensional soliton. In two dimensions we discussed the
general Nf case, here we start with the simplest case of Nf = 2 and then we
discuss the Nf = 3 and comment about the general case. As discussed above for
SU (Nf = 2) there is no WZ term, thus we have to substitute (20.36) into the
action that includes the sigma term and the Skyrme term (for simplicity we do
not add the mass term). The collective coordinates A(t) can be parameterized
in the following ways either,
A(t) = a0 (t) + ia(t) · σ (aμ )2 = 1, (20.37)
or,
i
A−1 Ȧ ≡ σ · w. (20.38)
2
It is easy to verify that in terms of A(t) the Lμ defined in (20.5),
L0 = A(t)g −1 (x)(A−1 Ȧ)(t)g(x)A−1 (t) Li = A(t)g −1 ∂i g(x)A−1 (t). (20.39)
The result of the substitution of the Lμ expressed in terms of w into the

Lagrangian is,
1
L = Lcl + α2 w2 , (20.40)
2
where the constant of proportionality α2 is computed as a spatial integral over
the chiral angle to be e53.3
3 f . In terms of spin and isospin operators the Hamiltonian
π
can be rewritten in terms of a second Casimir operator,
1 1
H = Ecl + 2 J 2 = Ecl + 2 I 2 . (20.41)
2α 2α
In terms of the aμ variables the Lagrangian density takes the form,
L = Lcl + λ(ȧμ )2 , (20.42)

where λ is the moment of inertia given by,

8π 2 ∞ 8e2 sin2 F
λ= fπ drr2 sin2 F 1 + 2 F2 + . (20.43)
3 0 fπ r2
The corresponding Hamiltonian is therefore,
1 2 1 ∂2
H = πμ ȧμ − L = Ecl + π = Ecl + − , (20.44)
8λ μ 8λ ∂a2μ
where in the last step we introduced the canonical quantization namely,
i∂
[aμ , πν ] = iδμν πμ = − , (20.45)
∂aμ
subjected to the constraint (aμ )2 = 1.
The Noether charges associated with the angular momentum and isospin
expressed in terms of aμ take the form,
i ∂ ∂ ∂
Ik = a0 − ak − k lm al
2 ∂ak ∂a0 ∂am
i ∂ ∂ ∂
Jk = ak − a0 − k lm al . (20.46)
2 ∂a0 ∂ak ∂am
Choosing a fermionic rather than bosonic wave function, namely, odd under
changing A → −A, implies that the wave function of baryons of I = J = 12 is
linear in aμ , for instance the proton wave function is |p> = π1 (a1 + ia2 ) and
√
those of I = J = 32 cubic in aμ like |Δ+ + > = π2 (a1 + ia2 )2 . It is thus obvious
that the mass difference between the Δ and the nucleon is,
3
MΔ − M N = . (20.47)
2α2
Expectation values of flavor charges can be computed in the semi-classical
approximation by expressing the Noether currents and charges in terms of the
aμ (t) and their corresponding momenta πμ = − ∂i∂a μ (20.46) and sandwiching
these operators in between quantum states, like N and Δ. For instance the
space components of the Baryonic current is given by,
eij k sin2 F
Bi = i r̂k Tr [Ȧ−1 Aτj ]. (20.48)
2π 2 r
We use this expression to compute the isoscalar magnetic moment, defined by,

1
μI = 0 = d3 rr × B, (20.49)
2
of the proton as follows,

2i
(μI = 0 )3 = − drr2 F <p, 1/2| Tr (τ3 Ȧ−1 A)|p, 1/2>
3π
i 1
<r2 >I = 0 <p, 1/2| Tr (τ3 Ȧ−1 A)|p, 1/2> = <r2 >I =0 .
3 6I
(20.50)
In a similar manner one can compute the isovector magnetic moment.

Another important property of nucleons that can be extracted from the

description of baryons as semiclassical solitons is the axial coupling gA defined
via,
a τa
<N (p2 )|JμA |N (p1 )> = <ū(p2 ) [gA (q 2 )γμ γ5 + hA (q 2 )qμ γ5 ]u(p1 )>, (20.51)
2
a
where JμA is the axial current and q = p2 − p1 . In the chiral limit hA (q 2 ) has a
pion pole whose residue is −2fπ gπ N N namely,
dA (q 2 )
hA (q 2 ) = dA (0) = −2fπ gπ N N , (20.52)
q2
where gπ N N is the pion nucleon coupling. Current conservation implies that,
2MN gA (q 2 ) + q 2 hA (q 2 ) = 0. (20.53)
In the nucleon rest frame the nonrelativistic limit q → 0, taken in a symmetric
form qi qj → 13 δij q 2 yields,
τa
lim <N (p2 )|JiA a|N (p1 )> = lim < u(p2 ) [gA (0)σi
q →0 q →0 2
+hA (q 2 )σ · q̂ q̂i ]u(p1 )> =
τa
= lim gA (0)(δij − q̂i q̂j ) <N (p2 )|σi |N (p1 )>
q →0 2
2 τa
gA (0) <N (p2 )|σi |N (p1 )>,
= (20.54)
3 2
where we have made use of the Goldberger–Triman relation,
gπ N N fπ
gA (0) = . (20.55)
MN
In the Skyrme model we can extract the axial coupling gA (0) in the following
way. First we compute the space integral over the axial current,

a 1
d3 xJ A i (x) = − d Tr [τi A−1 τ a A], (20.56)
2
where d is the space integral over a function that depends only on the classical
soliton configuration. We then sandwich this operator in between nucleon states
to find,

a 2 τa
lim d3 xei q ·x <N |J A i (x)|N> = d <N |σi |N> . (20.57)
q →0 3 2
Equating the last expression with (20.54) it was found that the Skyrme model
value of gA (0) = 0.61 whereas experientially it is equal to 1.33.
20.3.3 The Skyrme model and large Nc QCD

In Section 19.2 it was shown that the scattering amplitude or quadrilinear cou-
pling of mesons in the large Nc limit behaves like N1c . Recall that in this limit we

take Nc → ∞ while keeping λt Ho oft = g 2 Nc finite. The amplitude that is pro-

portional to g 2 behaves like N1c . By comparing this result to the quadrilinear
coupling of the Skyrme model we are able to determine the dependence of the
Skyrme coefficients on Nc . Let us start by expanding the sigma term in terms of
the pion fields,
fπ2 1 1
Lσ = Tr [∂μ g∂ μ g −1 ] ∼ ∂μ π · ∂ μ π+ 2 (π · ∂ μ π)2 − π 2 ∂μ π · ∂ μ π +O(π 6 ).
4 2 6fπ
(20.58)
The quadrilinear coupling behaves like f12 . If we expand the Skyrme term in a
π
similar manner we find that in that case the coupling behaves like e 21f 4 , and
π
hence we conclude in agreement with (19.12) that,
1
fπ ∼ Nc e∼ √ . (20.59)
Nc
This enables us to check the Nc dependence of the classical Skyrmion mass and
its semi-classical extension,
fπ 1 1
Mcl ∼ ∼ Nc Msc ∼ 2
∼ e3 fπ ∼ . (20.60)
e α Nc
Recall for comparison that the two-dimensional solitonic baryons were shown to
have classical mass which is also order Nc , but the semi-classical correction term
behaves like Nc0 and not N1c .
20.4 The Skyrme model for Nf = 3

Phenomenologically we should obviously be interested in the case of Nf = 3
rather than only two flavors. Moreover, to let the WZ term play a role we
also have to go beyond Nf = 2. So we have two reasons to discuss now the
U (Nf = 3) classical solitons and their semi-classical quantization. The action
is now the sum of a sigma term (20.3), the Skyrme term (20.13) and the WZ
term (20.9). We can further add a mass term (20.18). The latter can be used
to introduce an explicit breaking of the flavor symmetry by assigning different
masses to the different flavor degrees of freedom. In fact one can add additional
flavor symmetry breaking terms which, to simplify the treatment, we would
not do.
We first need to choose a parametrization for the U (Nf = 3) group element.
The analog of (20.8) including the U (1) factor now reads,
√ √
a=8 √
i√ 2
η 0 (x) iΦ(x) i√ 2
η0 i 2
λa φa
g(x) = e 3fπ e =e 3fπ e a=1 fπ , (20.61)

where λa are the SU (3) Gell–Mann matrices. In terms of the pions, kaons and
η we have,
⎛ √1 0 √1 8 ⎞
2
π + 6η π+ K+
Φ≡⎝ π− − √12 π 0 + √16 η 8 K0 ⎠ . (20.62)
− 0 2
√ η8
K K̄ 6
Next we have to choose an ansatz for the static classical configuration g0 (x).
Recall that in the two-dimensional case we took

an embedding of the U (1) in
−i 4π
φ(x)
U (Nf ) of the form g0 (x) = Diag(1, 1, . . . , e Nc
). In analogy in the four-
dimensional case we embed the SU (2) hedgehog configuration in the SU (3)
group element as follows,
⎛ ⎞
0
g0 (x) = ⎝ eiF (r ) τ ·r̂ 0⎠. (20.63)
0 0 1
Since the WZ term vanishes for an SU (2) group the solution for F (r) is iden-
tical to that discussed in Section 20.3 and hence the elevation to Nf = 3 shows
up basically only in the semi-classical quantization of the collective coordinates.
Recall that the latter are introduced via g0 (x) → A(t)g0 (x)A−1 (t). In two dimen-
sions we parameterized the quantum fields A(t) in terms of the Zi , i = 1, . . . , Nf
variables which was adequate for the CP N f −1 that the collective coordinates
span in that case. Clearly in the present case since the g0 (x) ∈ SU (2) a differ-
ent ansatz is required. The most straightforward one is in terms of the angular
velocities wa that generalize those of (20.38) as follows,
i a
8
A−1 (t)Ȧ(t) = λ wa . (20.64)
2 a=1
When substituting this into the Lagrangian one finds,

1 2 2 1 2 2
3 7
B
LS U (3) = Lcl + α w + β w − Nc √ w8 . (20.65)
2 a= 1 a 2 a=4 a 2 3
Note that the WZ term which is proportional to Nc and linear in the angular
velocity w8 associated with λ8 that commutes with the classical ansatz [λ8 , g0 (x)].
This is very reminiscent of the structure of the WZ term in two dimensions that
is also linear in the angular velocity associated with the hypercharge with Nc as
a coefficient.
The quantization of the system is performed as if it is an “SU (3) rigid top”
namely one defines the right generators,
∂LS U (3)
Ra = − = −α2 wa = −Ja , a = 1, 2, 3
∂wa
−β 2 wa , a = 4, .., 7 (20.66)
NC B
√ , a = 8,
2 3

and imposes the quantization condition,

[Ra , Rb ] = −ifabc Rc , (20.67)
where fabc are the structure constants of SU (3).
The Hamiltonian of the system takes the form,

1 1 1 1 3
H = Ecl + 2
− 2
J 2 + 2 C2 − 2 . (20.68)
2 α β 2β 8β
Again this form of the Hamiltonian is similar to the one we found in two dimen-
sions which is also proportional to the second Casimir operator. One can now
apply the Hamiltonian on states associated with representations of the SU (3) and
compute the corresponding masses. Using the eigenvalues of the second Casimir
operator of the representations 8, 10, 1̄0, 27 which are given by 3, 6, 6, 8,
respectively we can determine the masses of the various Skyrme hadrons. The
full analysis of the Nf = 3 baryons is beyond the scope of this book. We refer the
interested reader to literature. For a review of the topic see for instance [231].

21
From two-dimensional solitons to
four-dimensional magnetic monopoles
21.1 Introduction
Solitons play an important role in non-perturbative two-dimensional fields as
we have seen in the first part of this book. They are intimately related to non-
trivial topology, they are an essential ingredient in integrable models, and they
enable the phenomenon of fermion boson duality-bosonization. When passing
to four-dimensional field theories the topology may be even richer and thus
we would anticipate having topological solitons as static solutions also in four-
dimensional space-time. As we have seen in Section 5.3, Derrik’s theorem does
not permit the existence of solitons of scalar field theory in space dimensions
higher than one, however, they are not prohibited in theories that include higher
spin fields, in particular in theories of scalar fields coupled to non-abelian gauge
fields. Indeed as we will see in this section certain theories of this type that admit
spontaneous symmetry breaking, admit soliton solutions. These configurations
carry a conserved topological charge which guarantees their stability against
decay to the vacuum. As it will turn out this charge is in fact a magnetic charge
and hence these solitons are magnetic monopoles, or in the more general case
dyons with both magnetic and electric charge. The construction of dyons from
static solutions will be the analog process of building up breathers from two-
dimensional solitons.
In the next section we present the basics of the Yang–Mills Higgs theory. We
then show the relation between magnetic monopoles and topological solitons both
for the simplest case of SU (2) (and SO(3)) as well as for a general non-abelian
gauge group. The next topic is the seminal solution of ’t Hooft and Polyakov.
Then we discuss zero modes, time-dependent solutions and dyons. In the follow-
ing section we discuss the very important limit of BPS. We then describe the
construction of multi-monopole solutions that was proposed by Nahm. We show
its application to the construction of BPS monopoles of charge one and two. The
next topic is the moduli space of monopoles. We determine the metric on this
space for the case of widely separated monopoles.
The topic of magnetic monopoles and dyons has been covered by several review
papers, proceedings of meetings and books, for instance [21], [214], [67], [7], [182]
and [193], respectively. Here in this chapter we made use of mainly the former
two references.

372 From two-dimensional solitons to four-dimensional magnetic monopoles
Monopoles and dyons play an important role in N = 2 SYM. They admit a

mathematical non-perturbative structure that can be determined exactly. Since
we do not discuss supersymmetry in this book the monopoles and dyons of that
theory will not be addressed in this chapter. We refer the reader to [191] and
[192].
21.2 The Yang–Mills Higgs theory – basics

Consider the Yang–Mills Higgs system described by the following Lagrangian
density,
1
L = − Tr[Fμν F μν ] + Tr[Dμ ΦDμ Φ] − V (Φ), (21.1)
2
where the (non-abelian) gauge group is G, Φ is in the adjoint representation of
the group, namely,
Aμ = T a Aaμ Φ = T a Φa , (21.2)
where T a are the generators of G (see Section 3.1), the covariant derivative reads,
Dμ = ∂μ Φ + ie[Aμ , Φ], (21.3)
and V (Φ) is given by,
V (Φ) = −μ2 Tr[Φ2 ] + λ(Tr[Φ2 ])2 , (21.4)
where λ is taken to be positive so that the energy is bounded from below, and
we also take μ2 > 0. In general one can discuss a similar system where Φ is
in any representation of G but here we consider only the case of the adjoint
representation.
Let us start with the simplest case where G = SU (2) and Φ is in the triplet
(adjoint) representation. For such a case the vacuum solution can be put in the
form,
5
σ3 μ2
Φ(x) = v ≡ Φ0 v ≡
2 λ
Aμ (x) = 0. (21.5)
In this case the vacuum expectation value of the Higgs field breaks the SU (2)
symmetry down spontaneously to a U (1) symmetry along the a = 3 direction.
The physical fields will be denoted as follows,
A1μ + iA2μ
Aμ = A3μ Wμ = √ ϕ = Φ3 , (21.6)
2
which associate with the massless “photon”, pair of mesons W, W ∗ with a mass
√ of
eV and charges ±e and an electrically neutral scalar boson with mass mH = 2μ,
respectively.

For a general group G which we take to be a simple Lie group of rank r (for
the basic definitions see Section 3.1). In this case the expectation value of φ = Φ0
can be taken to lie in the Cartan subalgebra of the group G. Using the notation
H for the r-dimensional vector of the elements of the Cartan subalgebra, Φ0 is
characterized by a vector h such that,
Φ0 = h · H. (21.7)
The generators of the unbroken subgroup are those generators of G that commute
with Φ0 . These are all the generators of the Cartan subalgebra together with
ladder operators associated with roots orthogonal to h. If none of the γ are
orthogonal to h the unbroken symmetry is U (1)r , whereas if there are some

roots γ such that γ · h = 0 then the unbroken symmetry is U (1)r −r × K where
K is of rank r and it has γ as its root diagram.
21.3 Topological solitons and magnetic monopoles

In Section 5.2 when discussing two dimensional solitons, we identified a topolog-
ical conserved current and an associated topological charge. Configurations that
carry a non-trivial value with respect to this charge cannot, due to charge conser-
vation, decay to vacuum. These configurations were shown to be stable solutions
of the equations of motion and to have finite energy. Thus we anticipate that
also in four-dimensional field theories, and in particular in the Yang–Mills Higgs
theory we discuss here, there should be solutions of the equations of motion that
associate with non-trivial topological charges and one can determine their exis-
tence even without solving the equations of motion. It is easy to verify that for
G = SU (2) the following current,
1
kμ = μν ρσ abc ∂ ν Φ̂a ∂ ρ Φ̂b ∂ σ Φ̂c , (21.8)
8π
a
Φ
where Φ̂a = |Φ| , is conserved for any configuration whether it solves the equations
of motion or not. It follows trivially from the total anti-symmetry of μν ρσ that
∂ μ kμ = 0. The corresponding charge is,

1
Q = d3 xk0 = d3 xij k abc ∂ i Φ̂a ∂ j Φ̂b ∂ k Φ̂c
8π

1 3 abc i a j b k c 1
= d xij k ∂ (Φ̂ ∂ Φ̂ ∂ Φ̂ ) = d2 Si xij k abc Φ̂a ∂ j Φ̂b ∂ k Φ̂c .
8π 8π
(21.9)
This topological charge is the winding number associated with the map,
G /H
Φ0 (∞) : S2space → S2 , (21.10)
where S2space is the boundary of the space at r = ∞ and where the coset space
G/H is in our example G/H = SU (2)/U (1) = S2 . It is thus an integer charge
Q = n and as such it must be invariant under smooth deformations of the surface

of integration that do not cross any of the zeros of Φ. In fact a configuration with
Q = n must have at least |n| zeros of the Higgs field. If we distinguish between
a + zero and a – anti-zero then the net number is precisely n. Obviously if
we consider Higgs configurations of Q = n = 1 there must be one with minimum
energy. This cannot be smoothly deformed to a vacuum since the winding number
is quantized. Hence such a configuration must be a local minimum of the energy
and therefore a static classical solution.
The next question we want to address is what is the connection between these
non trivial soliton solutions and magnetic monopoles? The field strength of the
abelian gauge field Aμ defined in (21.6), Fμν = ∂μ Aν − ∂ν Aμ is the outcome of,
1
F̃μν = Φ̂a Fμν
a
− abc Φ̂a Dμ Φ̂b Dν Φ̂c , (21.11)
g
when we take Φ̂ = (0, 0, 1), namely Φ = ϕ. What distinguishes F̃μν from an ordi-
nary abelian field strength is that it does not obey the Biachi identity,
∗ 1 4π
dF̃ = eμν ρσ ∂ ν F̃ ρσ = ekμ = kμ , (21.12)
2 g
where g is the magnetic charge which will be shown to be equal to 4π/e. Defining
now the magnetic field Bi associated with F̃μν as usual as,
1
Bi ≡ ij k F̃ j k , (21.13)
2
we find that,

4π 1 4π 4π
∇·B = k0 QM = d3 xk0 = Q= n. (21.14)
g g e e
We have thus realized that the non-trivial soliton configurations carry a magnetic
charge and hence are magnetic monopoles. We can further determine the classical
mass of the monopole since the total energy of such a solution is,

E = d3 x Tr[Ei2 ] + Tr[(D0 Φ)2 ] + Tr[Bi2 ] + Tr[(Di Φ)2 ] + V (Φ) . (21.15)
For a static configuration that does not carry electric charge, the first two terms
are expected to vanish. Then one can show that the form of the mass has to be,
4πv λ
M= f , (21.16)
e e2
where f ( eλ2 ) should be of order one.
So far we have discussed the topological charges of the group G = SU (2) case.
Let us now address the general case. Instead of the map (21.10), the asymptotic
Higgs field constitutes in the general case a map,
G
Φ(∞) : ∂M → , (21.17)
H
where ∂M is the boundary of the space which for ordinary flat Minkowski space-
time is S2 and G/H is the coset of the unbroken symmetry group H and the

original group G, which in the discussion above was SU (2)/U (1) = S2 . These
maps from the boundary of space to the coset, fall into equivalent classes which
form the homotopy group π2 (G/H). For simply connected group G, namely with
π1 (G) = 0 the classification of the maps is in fact done by π1 (H) since for this
type of G,
π2 (G/H) = π1 (H). (21.18)
This follows from the following exact sequence1
. . . → π2 (G) → π2 (G/H) → π1 (H) → π1 (G) → . . . . (21.19)
The image of a given homomorphism equals the kernel of the next one in the
sequence. It is well known that for any semi-simple group G, π2 (G) = 0 and
hence,
π2 (G/H) ∼
= Ker[π1 (H) → π1 (G)]. (21.20)
Now since for a simply connected group π1 (G) = 0 we find (21.18). Let us
describe now several cases of physical interest:
r The ’t Hooft–Polyakov solution that will be discussed in the next section, is
slightly different since in that case G = SO(3) which is not simply connected
π1 (SO(3)) = Z2 and hence only the even elements of π1 (H = U (1)) are in the
kernel of the homomorphism of (21.20). This is the source of the fact that
the quantization condition is twice the one given by Dirac (see (21.27)) even
though in both cases H = U (1).
r For a simply connected G of rank r and with H which is the full Cartan sub-
algebra, namely H = U (1)r , the homotopy group that classifies the magnetic
monopoles is π1 (H = U (1)r ) = Z r .
r In the spontaneous symmetry breaking of the electro-weak theory we have
G = SU (2) × U (1) and H = U (1) such that G/H ∼ = S 3 . Since π2 (S 3 ) = 0 mag-
netic monopoles are excluded in this theory.
r On the other hand a wide class of grand unified theories do admit mag-
netic monopoles. The most prominent example is the G = SU (5) grand unified
model with H = SU (3) × SU (2) × U (1). This is an example of the case that
H = U (1) × K where K is a semi-simple and simply connected group. In this
case there is only a single component of the magnetic charge that is topologi-
cally conserved.
r Another interesting scenario is the case where the group G twice undergoes
a spontaneous symmetry breaking namely, G → H1 ⊂ G → H2 ⊂ H1 . This is
relevant to an evolution of the universe where at as early stage magnetic
1 The reader who is not familiar with the notion of an exact sequence can refer to any text
book on topology or alternatively to the book of Coleman [66] where an elegant proof of this
theorem is presented.

monopoles associated with π2 (G/H1 ) are being created and then the question
is what is their fate when the universe undergoes a second phase transition to
H2 ? This can be determined by an exact sequence similar to (21.19).
21.4 The ’t Hooft–Polyakov magnetic monopole solution

Equipped with knowledge based on the topological arguments of above, that
magnetic monopoles do exist for theories associated with non-trivial π2 (G/H) we
want to proceed to the determination of explicit configurations of the non-trivial
topological solutions. It turns out that this is a non-trivial task and only for a
limited set of cases can it be accomplished analytically. We start with the simplest
case where G = SU (2), or G = SO(3) as was done in the original solution of
’t Hooft and Polyakov [123], [175]. We will see in the next section an explicit
solution using a special limit of the theory. Without this limit one can simplify
the procedure by searching for spherically symmetric solutions. This implies that
the fields must be invariant under a combination of rotations and a compensating
gauge transformation. The latter can be space independent by using the so called
“hedgehog” gauge where rotational invariance requires that the fields be invariant
under a combined rotation and global internal SU (2) transformation. Stating
it differently, one looks for a configuration which is symmetric under a mixed
angular momentum,
J = L + I, (21.21)
where L = −ir × ∇ is the ordinary spatial part of the angular momentum and
I are the generators of the SU (2) gauge group. With this definition of J the
ansatz should obey,
[Ji , Φ] = 0 [Ji , Aj ] = iij k Ak . (21.22)
Using this gauge ’t Hooft and Polyakov suggested the following ansatz for the
fields,

1 − u(r)
Aai = iam r̂m Φa = r̂a h(r). (21.23)
er
To write down the equations that determine u(r) and h(r) one can substitute
these expressions into the Lagrangian density (21.1) and vary with respect to
u(r) and h(r). This is easier than the usual procedure of substituting (21.23)
into the equations of motion derived from (21.1). The resulting equations are,
(u2 − 1)u
u − + e2 uh2 = 0
r2
2 2u2 h
h + h − 2 + λ(v 2 − h2 )h = 0. (21.24)
r r

21.5 Charge quantization 377
Fig. 21.1. The hedgehog configuration of the Higgs field. The orientation in
isospace is aligned with the position vector.
The primes denote derivatives with respect to r. Finiteness of the energy asso-
ciated with the solution requires that,
Φ0 (∞) = 0 → u(∞) = 0
Di Φ0 (∞) = 0 → h(∞) = v. (21.25)
Similarly requiring that the solutions are non-singular at the origin implies that,
u(0) = 1 h(0) = 0. (21.26)
Qualitatively, the profile of the Higgs field is that of a hedgehog, as can be seen
in Fig. 21.1. The orientation in isospace is aligned with the position vector.
Analytic solutions of these equations will be derived in the next section using
a special (BPS) limit. In general one has to solve these equations numerically.
The physical picture that comes out from these calculations is that there is a
central core of radius Rcore ∼ ev
1
, outside of which u(r) and |h − u|(r) decrease
exponentially. The mass of the monopole takes the form M = 4πv λ
e f ( e 2 ) where
f (0) = 1 and f (∞) = 1.787.
21.5 Charge quantization

In his seminal paper on magnetic monopoles in quantum mechanics [78] Dirac
found out that the magnetic charge g and the electric charge e must be related
via the famous charge quantization condition,
eg = 2πn, (21.27)

where n is an integer. This implies that the existence of a magnetic monopole

explains the observed quantization of all electric charges. He proposed a solution
of a magnetic potential of the following form,
eg r̂ × n̂
eA = , (21.28)
4π r(1 − r̂ · n̂)
which has in addition to the singularity at the origin also a singularity extending
from the origin out along the n̂ direction. The quantization condition follows
from the requirement that physical charges should not be able to detect the
string.
Yang–Mills Higgs models with spontaneous symmetry breaking such that
π2 (G/H) = 0, as we have seen above, admit magnetic monopole solutions asso-
ciated with each element of the Cartan subalgebra of the unbroken group H. For
large distance the corresponding magnetic fields take the form,
eg 1
eAi = (1 + cosθ)∂i φ + O . (21.29)
4π r2
The generalization of the quantization argument of Dirac to the non-abelian case
is straightforward. This follows from the demand that the electrically charged
fields of the theory are single valued if acted upon by a group element eeg .
This quantization condition can be solved in terms of the simple roots γi where
i = 1, . . . , r where r is the rank of H, of the root system of H. From the simple
roots one constructs a convenient basis (Hi ) for the elements of the Cartan
subalgebra with the property that each element has half-integer eigenvalues when
acting on the basis vector of any representation. This is achieved by taking,
γi
Hi ≡ H. (21.30)
|γi |2
In this basis the solution of the quantization condition is,
eg r
= ni Hi . (21.31)
4π i=1
This solution can be represented as an r-dimensional lattice dual to the weight

lattice of the group (see Section 3.1). For the simple example of SO(3) the rank
is one and thus one gets eg = 4πn twice as the condition of Dirac due, as was
explained above, to the fact that SO(3) is not simply connected. For the group
SU (3) which is of rank r = 2 the charge lattice is drawn in Fig. 21.2. In general
it was shown that the charge lattice is the weight lattice of the dual gauge group.
21.6 Zero modes, time-dependent solutions and dyons

From the static SU (2) monopole solution discussed above we can generate obvi-
ously (infinitely) more solutions by applying gauge transformations. This of
course will be avoided by fixing a gauge. However even in that case there is

21.6 Zero modes, time-dependent solutions and dyons 379
γ2
γ1
Fig. 21.2. The charge lattice of the SU (3) monopole.
one unfixed global gauge U (1) phase. This adds up to the parameters associated
with the translation of the monopole. Infinitesimal transformations of this set of
four parameters generate field variations δAi and δΦ that preserve the equations
of motion and leave the energy unchanged. These variations in general will be
referred to as zero modes.
Time-dependent excitations of the translational zero modes can be derived by
substituting r → r − vt into the static solution. This has to be done together
with ensuring the Gauss law constraint,
Di E i = ie[Φ, D0 Φ], (21.32)
which also implies that for most choices of gauge A0 = 0. Substituting the solu-
tions of the Gauss law into the expression of the energy (21.15) one can show
that the change of energy for non-relativistic velocities is as one expects, given
by,
1
ΔE = M |v|2 . (21.33)
2
Next we want to describe the excitation of the zero mode associated with the
fourth parameter, that of the global gauge phase. The Noether charge associated
with this transformation is the electric charge that corresponds to the unbroken
U (1) gauge symmetry. In terms of the physical variables defined in (21.6) this
takes the form,

∗
QE = −ie d3 x[W j D0 (Wj − Dj W0 ) − W j D0 (Wj∗ − Dj W0∗ )], (21.34)
where the U (1) covariant derivative is defined by Dμ Wν = (∂μ − ieAμ )Wν and
where a string gauge is used where the Higgs field direction is uniform.

A dyon [134] is defined to be a configuration that carries both magnetic as well

as electric charges. To construct such a solution we start with a static solution
in the string gauge and multiply the W field by a uniformly varying U (1) phase
factor eiw t . In analogy to the magnetic charge found above, the electric charge
has the form,

QE = d2 Si F̃0i = d2 Si Φ̂i Eia . (21.35)
The time-dependent configuration can be transformed into a static solution

by a U (1) transformation of the form,
w
Wi → e−iw t Wi Ai → Ai A0 → Ā0 = A0 − . (21.36)
e
In this static form we can transform the solution into the non-singular hedgehog
gauge with Aai and Φa given by (21.23) and,
Aa0 = r̂a j(r) = r̂a Ā0 . (21.37)
For this case the static field equations, which are the analog of (21.24) become,
2 2u2 h
h + h − 2 + λ(v 2 − h2 )h = 0,
r r
(u 2
− 1)u
u − − e2 u(h2 − j 2 ) = 0,
r2
2 2u2 j
j + j − 2 = 0, (21.38)
r r
where the first equation is identical to the one in (21.24), in the second there is
a 2e2 uj 2 addition and the third equation is the Gauss law.
The dependence of QE on w can be determined by substituting the ansatz
into (21.34) and recalling that W0 = 0 we find,

8πw j(r)
QE = dru(r)2 ≡ Iw. (21.39)
e j(∞)
The integral can be estimated since u(r) falls off exponentially outside a region
of radius ∼ 1/v so that,
4π ¯
I= I, (21.40)
e2 v
where I¯ is of order unity. In analogy to (21.33) one can show that the correction
to the energy is,
Q2E e 2 Q2 M Q2E
ΔE = = E
¯ ∼ M. (21.41)
2eI 4π 2If 2Q2M

21.7 BPS monopoles and dyons 381
21.7 BPS monopoles and dyons

A special limit of the monopole configuration occurs when one takes the limit
of,
μ2
λ→0 μ2 → 0 v2 = fixed. (21.42)
λ
In this limit the energy of the system,

E = d3 xTr[E · E + B · B + (D0 Φ)2 + (DΦ)2 ]

= d3 xTr[(E ± sinαDΦ)2 + (B ± cosαDΦ)2 + (D0 Φ)2 ]

±2 d3 x[cosαTr[B · DΦ] + sinαTr[E · D]]. (21.43)
The passage from the first line to the rest is of course an identity for arbitrary
α. Next we perform an integration by parts in the last line and use the Gauss
law D · E − ie[Φ, D0 Φ] = 0 and D · B = 0 we find,

E = d3 xTr[(E ± sinαDΦ)2 + (B ± cosαDΦ)2 + (D0 Φ)2
± cos αQM ± sin αQE

E ≥ ± cos αQM ± sin αQE , (21.44)
where the magnetic and electric charges QM = vQM and QE = vQE are given
by (21.14) and (21.35)

QM = 2 d2 S · Tr[ΦB] QE = 2 d2 S · Tr[ΦE]. (21.45)
Recall that so far α is arbitrary. The most stringent bound is achieved when
QE
one takes tan α = Q M
, for which the bound reads,

E ≥ Q2M + Q2E . (21.46)
It is easy to realize that the bound is saturated if,
E = cos αDΦ B = sin αDΦ D0 Φ = 0. (21.47)
These first-order equations are referred to as the Bogomolny Prasad Sommerfeld

equations or BPS equations ([41] and [180]). The configurations that obey these
equations have the minimal value of energy,

E = Q2M + Q2E , (21.48)
for given magnetic and electric charges QM , QE respectively and hence are also
solutions of the (second-order) equations of motion of the system. In particular

the magnetic monopole which carries no electric charge and is static and hence
QE = 0 and D0 Φ = 0 (in the A0 = 0 gauge), obeys the Bogomolny equation,
B = DΦ. (21.49)
The BPS limit seems to be unnatural and artificial since we have introduced a
potential to give a non-trivial expectation value to the field Φ and then we tuned
the potential to zero keeping the expectation value. It turns out that in certain
suspersymmetric models the BPS equations follow from the requirement of the
invariance under supersymmetry. Since supersymmetry is not discussed in this
book we refer the reader to the literature, for instance [214].
If we go back to the equations of motion and substitute λ = 0 the equations
take the form,
2 u2 h
h + h − 2 2 = 0,
r r
u(u 2
− 1)
u − − e2 uh2 = 0. (21.50)
r2
The solution of this set of equations is given by,
evr 1
u(r) = h(r) = v coth(evr) − (21.51)
sinh(evr) er
The fact that Φ falls off as 1/r and not exponentially is due to the fact that in
the BPS limit it has a vanishing mass and hence associates with a long range
force.
The BPS equations can also be solved for the case of a dyon, namely a config-
uration that carries both a magnetic as well as an electric charge. In that case
the solution reads,
eṽr
u(r) =
sinh(eṽr)

Q2M + Q2E 1
h(r) = ṽ coth(eṽr) −
QM er

QM 1
j(r) = − ṽ coth(eṽr) − . (21.52)
QE er
21.8 Montonen Olive duality

In two dimensions we have seen an equivalence between a soliton configuration
and an elementary field. This was the essence of bosonization manifested for
instance in the equivalence between the sine-Gordon theory and the Thirring
model (see Section 6.2). Montonen and Olive [163] conjectured an analogous
duality between the spectrum of states created from the elementary field and
from those created by the solitons. Since the former are electrically charged
and the latter are magnetically charged, the duality is in a sense a non-abelian

Table 21.1. The spectrum of the SU (2)

YM Higgs theory in the BPS limit
Mass QE QM
photon 0 0 0
φ 0 0 0
W± ev ±e 0
4 πv
Monopole e
0 ± 4eπ
generalization of electric–magnetic duality of the Maxwell equations. To under-

stand this duality notice that under the following operation,
4π
QE ↔ QM e↔ , (21.53)
e
the entries associated with the W mesons and those of the magnetic monopoles
in Table 21.1 are interchanged.
It turns out that on top of the self-duality of the spectrum, there is a similar
duality also in the low energy scattering. It is a well-known property of BPS
states in general and in particular the magnetic monopoles that there is no net
force between them. This follows up from an exact cancellation between the
magnetic repulsion and the attraction due to an exchange of a Higgs scalar. The
zero velocity limit of the scattering amplitude of two W bosons also admits a no
force behavior. The exchange of a single photon is cancelled out by the exchange
of a Higgs boson.
In the non-supersymmetric YM Higgs theory the duality of the spectrum and
the scattering amplitudes cannot be lifted to a duality of the full theory. A simple
indication of this is the fact that the W boson carries a spin one, whereas the
quantum state built from a spherical monopole carries spin 0. In supersymmetric
analogs of the YM Higgs theory this difficulty may be overcome since both the W
bosons as well as the magnetic monopoles are members of supersymmetric mul-
tiplets that contain states with several different spins. It turns out that the YM
theory with 16 supercharges, the so-called N = 4 SYM admits a complete invari-
ance under the Olive Montonen duality. Since we do not discuss supersymmetry
in the book we refer the interested reader to the review papers mentioned above,
for instance [214].
21.9 Nahm construction of multimonopole solutions

This construction [167] maps the Bogomolny equation in three variables into a
nonlinear equation in one variable. We present the construction for the SU (2)
case. To simplify the notations we set the coupling constant e = 1. It can be
restored when needed.

The construction of an SU (2) k monopole is built from three steps:

1. We look for a quartet of Hermitian k × k matrices Tμ (s), where μ = 0, i =
1, 2, 3 which satisfy the Nahm equation,
dTi i
+ i[T0 , Ti ] + ij k [Tj , Tk ] = 0, (21.54)
ds 2
where s is an auxiliary variable that takes its value in the interval
−v/2 ≤ s ≤ v/2 and where v is the vacuum expectation value of the Higgs
field. For k = 1 since the commutators vanish we get that Ti are constants.
In fact due to the ordinary gauge invariance one can choose a gauge where
T0 = 0 and hence the equation reads,
dTi i
+ ij k [Tj , Tk ] = 0. (21.55)
ds 2
The boundary condition that one should impose for the multimonople case is
that the Ti (s) have poles at the boundaries of the form,
L±
Ti (s) = − i
+ O(1). (21.56)
s ∓ v2
The Nahm equations implies that the L±

i form a k-dimension representation
of the SU (2) algebra,
[L± ± ±
i , Lj ] = iij k Lk . (21.57)
These representations should be irreducible, namely, must be equivalent to
the (k − 1)/2 representation.
2. The next step is to solve the construction equation for the 2k component
vector w(s, r),

† d
Δ w(s, r) ≡ − − Ti ⊗ σi + ri Ik ⊗ σi w(s, r) = 0. (21.58)
ds
We denote by wa (s, r) a completely linearly independent set of normalizable
solutions that obey the orthonormality condition,
v /2
dswa† (s, r)wb (s, r) = δab . (21.59)
−v /2
3. It can be proved that for the SU (2) case there are only two normalizable
solutions wa (s, r). The space-time fields are given in terms of these as follows,
v /2
Φab (r) = dswa† (s, r)wb (s, r),
−v /2
v /2
Aab
j (r) = dswa† (s, r)∂j wb (s, r). (21.60)
−v /2
We do not bring here the the details of the proof of this construction (see for
instance [71]) we just mention that it includes the following elements: (i) Using

the expressions for Φab and Aab ab

i given above in (21.60) one can show that Bi and
ab
Di Φ are identical and hence the Bogomolny equation is obeyed. (ii) Showing
that the solutions lie in SU (2) and (iii) computing the long-range behavior and
showing that the solutions indeed have a magnetic charge equal to k. We refer
the interested reader to, for instance, [214] for the detailed proof. Instead we pro-
ceed now to a demonstration of the application of the method both for k = 1 and
k = 2. As was mentioned above for the former case the Ti are constants indepen-
dent of s. In fact these constant values of T enter the construction equation as
(r − T ) · σ namely, they are the coordinates of the center of mass of the monopole
and thus by shifting to a frame of coordinates which is centered at the monopole
center T = 0. The construction equation (21.58) takes the form,
dw
= r · σw. (21.61)
ds
The two normalizable solutions of this equation are,
wa (s, r) = N (r)es r ·σ ηa , (21.62)
where N is a normalization factor and ηa are orthonormal constant vectors. From

the orthonormality condition we find that,
5
r
N (r) = , (21.63)
sinh(vr)
where we have made use of the first of the following integrals,
v /2
sinh(vr)
e2s r ·σ ds = I2
−v /2 r
v /2
r·σ
se2s r ·σ ds = 3 [vr cosh(vr) − sinh(vr)]. (21.64)
−v /2 r
Using the integral we find that the Higgs field and the corresponding gauge field
are given by,
1 1 r·σ
Φab = v coth(vr) − ηa† ηb ,
2 r r
1 v
Aab = −iηa† ∂i ηb − iij k r̂j ηa σk ηb − . (21.65)
2r 2 sinh(vr)
Upon setting η1t = (1, 0) and η2t = (0, 1) we retrieve the hedgehog solution of
21.51).
21.9.1 SU(2) two-monopole solutions

The k = 2 solutions are characterized by a priori eight parameters out of which
one corresponds to the U (1) phase and as for the case of k = 1 does not enter the
Nahm data. Three parameters relate to the translation and three to the rotations

of the solution. Thus we end up with one non-trivial parameter. Intuitively this
parameter should relate to the separation distance between the two centers of
the solution. Let us see if the Nahm construction verifies this intuition and to
what extent one can write an explicit solution for this case. The Ti (s) which now
are not constants can be decomposed into the following form,
1 v
Ti (s) = T (s) · σ + Tis (s)I2 . (21.66)
2 i
Substituting this into the Nahm equation implies that the Tiv (s) have to obey,
dTiv (s) 1
= ij k Tjv (s) × Tkv (s). (21.67)
ds 2
It is easy to realize that there is a relation between the Tkv (s), namely the fol-
lowing matrix,
1
Tij = Tiv (s) · Tjv (s) − δij Tkv (s) · Tkv (s), (21.68)
3
is s independent. After some tedious algebra one can show that the most general
form of Tkv (s) takes the form,
1
Tiv (s) = Aij fj (s + v/2, κ, d)τj + Tis I2 , (21.69)
2 i
where Aij is an orthogonal matrix of constants that diagonalize Tij , f (s +

v/2, κ, d) are the Euler top functions2 and the parameter d can be shown for
widely separated monopoles to be the inter-monopole distance as the original
intuition taught us.
21.10 Moduli space of monopoles

As we have seen above the monopole configurations are parameterized by certain
moduli. One defines a space of these moduli, the moduli space. The properties
of this space are intimately related to the low energy behavior of monopoles
and dyons. In Section 22.3 we will describe the moduli space of YM instantons.
Denoting the collective coordinates that parameterize a monopole configuration
2 The Euler top function can be expressed in terms of the elliptic functions SN κ (x), C N κ and
DN κ (x) as follows:
C N (D x )
f1 (x, κ, D) = −D S N κ (D x )
κ
D N κ (D x )
f2 (x, κ, D) = −D S N κ (D x )
f3 (x, κ, D) = −D S N 1(D x ) (21.70)
κ

21.10 Moduli space of monopoles 387
by zr , the Lagrangian of the system is approximated by,

1
L = −(total rest mass of monopoles) + gr s (z)żr żs . (21.71)
2
The metric on the moduli space gr s (z) can be determined from the background
zero models of the gauge fields as follows,

gr s (z) = 2 d xTr[δr Ai δs Ai + δr Φδs Φ] = 2 d3 xTr[δr Aa δs Aa ],
3
(21.72)
where a takes the values a = 1, . . . , 4 with A4 = Φ, and where,

∂(Acl )a
δr Aa = − Da r , (21.73)
∂zr
and where r is defined via A0 = żr r which follows from the Gauss law,
Da F a0 = 0.
In the case of a single monopole, as we have seen above, there are four zero
modes associated with the location of the center of the monopole rcm and the
global U (1) phase so that,
I
gr s (z)żr żs = M (ṙcm )2 + α̇2 , (21.74)
e
where M is the mass of the monopole and I is defined via (21.39 ) QE = Iw.
The moduli space of BPS monopoles is hyper-Kähler. This property that
implies that there are three almost complex structures with correspondingly
three closed Kähler forms will be discussed in detail in Chapter 22, so we will
not discuss it here for the BPS monopoles.
An important part of the structure of the moduli space is encoded in its isome-
tries, namely the symmetries that preserve the form of the metric. Naturally since
the underlying space where the monopoles reside is an R3 the isometries include
three translations of the center of mass of the collective coordinates. The same
applies to the spatial rotation of the monopole. It takes a monopole solution
to another monopole solution and hence it maps one point in the moduli space
into another one. Another type of isometries is those associated with the unbro-
ken U (1) gauge groups. These isometries, unlike the rotational isometry pre-
serve the complex structure. One can choose a coordinate basis where the gauge
transformations act as translations of the angular variables ξ A , with the corre-
sponding Killing vectors being KA = ∂ ξ∂A . Denoting by y p the rest of the coor-
dinates, the Lagrangian associated with the moduli space approximation can be
written as,
1 1
L= gpq (y)ẏ p ẏ q + g̃A B (y)[ξ˙A + ẏ p wpA (y)][ξ˙B + ẏ q wqB (y)]. (21.75)
2 2
Thus the coordinates ξ A are cyclic coordinates and their conjugate momenta are
conserved. In fact the latter are the electric charges of the dyonic cores.

For monopoles that are separated by large distances the task of determining
the metric of the moduli is much easier. Consider a system of N fundamental
monopoles all well separated from each other. In such a layout only abelian inter-
actions are relevant and there is an enhanced gauge symmetry. Instead of having
a conserved electric charge for each unbroken U (1), there is an effective con-
served charge for each monopole core. The moduli space is spanned in this case
by 3N coordinates of the positions of the cores and N angles ξj , j = 1, . . . , N .
The enhance symmetry is the translation along each of the ξj . The approximated
Lagrangian then takes the form,
1 1
L = Mij (x)ẋi · ẋj + g̃ij (x)[ξ˙i + wki (x)ẋk ][ξ˙j + wli (x)ẋl ]. (21.76)
2 2
By computing the pairwise interactions between the separated dyons, one can
determine the functions Mij (x), g̃ij (x) and wji (x). We refer the reader to [214]
for the derivation and we cite here the results,
4π α ∗ ·α ∗
Mij = mi − k = 1 e 2 ri i k k , i = j (21.77)
∗
4π α ·α ∗
Mij = k = 1 e 2 ri i j j , i = j (21.78)

Wij = − k = 1 αi∗ · αk∗ wik , i = j (21.79)
Wij = αi∗ · αi∗ wij , i=j (21.80)
2
−1
and K = (4π)
e4 M . It can be further shown that the metric of the moduli space
of a two monopole BPS solution can be determined exactly and it takes the
form of a Taub–Nut metric or Atiya–Hitchin metric [19] depending whether the
monopoles are distinct or the same. This is beyond the scope of this book and
we refer the reader to for instance the review [214].

22
Instantons of QCD
In Chapter 5 we saw that solutions of the classical equations of motion, which are
characterized by a topological number, play an important role in two-dimensional
QFT. Derick’s theorem (5.36) forbids scalar field soliton solutions in higher than
two-dimensional space-time. However, for gauge fields one can bypass the theo-
rem, and indeed, as we have seen in Chapter 21, there are solitons in the form of
magnetic monopoles in four-dimensional gauge theories. The topic of this chap-
ter will be solutions of the Yang–Mills theory defined on a Euclidean space-time
which have finite action and are topological in their nature, the instantons. We
will start with a description of the basic properties of one instanton solution
including the topological charge that characterizes it. We then describe the con-
struction of multi-instanton solutions and the moduli space of instantons includ-
ing its dimension, complex nature, singularities and symmetries. When Wick
rotated to Minkowski space-time the instanton describes a tunneling process
between different vacua. We will elaborate on this phenomenon in the context of
the four-dimensional YM theory. Various properties of QCD and hadron physics
were thought to be related to instantons. In certain cases like confinement, the
relation to instantons is still a mystery. One case where the role of instantons
is clear is the U (1A ) problem. This will be described in the last section of this
chapter.
The one instanton solution was derived in [32]. The basic properties of instan-
tons were worked out by many authors including [125], [57]. The instantons of
SU (N ) gauge symmetry were derived in [218]. There are several review papers
such as [65], [208] and [189] and books [182] and [188] that describe the basic
instanton solutions.
22.1 The basic properties of the instanton

The action of the four-dimensional YM theory in Euclidean space-time can be
rewritten in the following form,

1 1
S= d4
x Tr [F μν
F μν ] = d4 xF a μν Fμν
a
2g 2 4g 2

1 1
S= 2 d4 x ± Tr [F μν ∗ Fμν ] + Tr [(Fμν ∓ ∗ Fμν )(F μν ∓∗ F μν )] , (22.1)
2g 2

390 Instantons of QCD
where as usual Fμν = Fμν a

T a 1 and ∗ Fμν = 12 μν ρσ Fρσ is the dual field strength,
and g is the YM coupling. The first term in the second line is a topological
invariant, or a topological charge2 since,

1 4 μν ∗
Q= d x Tr [F Fμν ] = d4 x∂μ Kμ = dσμ Kμ , (22.2)
16π2
where,

1 3
Kμ = μν ρσ Tr Aν ∂ρ Aσ + Aρ Aσ . (22.3)
8π2 2
In fact Q is the Pontryagin index or the winding number of maps from the
sphere at space-time infinity to the SU (2) group manifold which is also the
three sphere, namely Ss3 → Sg3 . This topological invariant is the homotopy π3 (S 3 )
which is an integer π3 (S 3 ) ∈ Z. This can be shown as follows. Since the self-dual
field is asymptotically a pure gauge, namely on σμ Aμ = U ∂μ U −1 and Fμν = 0
hence,
1
Q= dσμ μν ρσ Tr [Aν Aρ Aσ ]
24π2
1
Q= dσμ μν ρσ Tr [(∂ν U )U −1 (∂ρ U )U −1 (∂σ U )U −1 ]. (22.4)
24π2
If we take for U the following ansatz that will be shown below to correspond
to the one instanton solution,
U (x) = x̂μ σμ = x0 + ixi σi , (22.5)
we found,

1 −12xμ
Q = dσμ Kμ = − dσμ
24π2 |x4 |

1 xμ 1
= dΩxμ |x|2 4 = dΩ = 1. (22.6)
2π2 |x| 2π 2
Thus we have shown that Q is indeed the winding number that measures how
many times we wind Sg3 when we integrate over Ss3 .
Let us now return to (22.1). Since it is a sum of a topological charge and a
positive semi-definite quantity, it is clear that it is minimized when the latter
vanishes namely,
Fμν = ±∗ Fμν . (22.7)
The corresponding gauge fields Aμ (with a + sign) will be referred to as instanton

or self-dual gauge field and those with a − sign as anti instanton or anti self-dual
1 In this chapter we denote the SU (N ) adjoint indices with a = 1, . . . , N 2 − 1 whereas in

Chapter 19 we used A and not a.
2 Recall in analogy the topological charge defined in two-dimensional scalar field theories (5.3).

gauge field. It is straightforward to show that the self-duality condition implies

the equation of motion,
Dμ Fμν = 0. (22.8)
A solution of the equation of motion is not necessary self-dual but it can be shown
that the non-self-dual configurations are saddle points and not local minima of
the action.
Comparing the expression of the action (22.1) and the topological charge (22.2)
it is clear that a (anti) self-dual configuration that carries an instanton number
(Q = −k), Q = k has an action of,
8π2
S= |k|. (22.9)
g2
One can add the topological charge as an additional term to the action. To be
more precise one adds a θ term,

θ
Sθ = i d4 x Tr [F μν ∗ Fμν ] = iθk. (22.10)
16π2
Whereas the ordinary YM action is the same for the instanton and anti-instanton,
the θ term obviously distinguishes between them by assigning opposite charges
to them. We will further discuss the theta term in Section 22.5. For the self-dual
solution up to a constant the action is equal to the topological charge which by
definition does not depend on the metric. This exhibits the topological nature of
the instanton. Another indication of this nature is the fact that it has vanishing
energy-momentum tensor as follows from,

2 1
Tμν = − 2 Tr Fμρ Fν ρ − δμν Fα β F α β = 0. (22.11)
g 4
This clearly implies that instantons do not curve the space-time they
reside in.
The one instanton solution for the SU (2) can be constructed from U (x) given
in (22.5) via,
ρ2
Aaμ (x) = U −1 ∂μ U , (22.12)
(x − X)2 + ρ2
which yields the explicit form,
a
ημν (x − X)ν
Aaμ (x) = 2 , (22.13)
(x − X)2 + ρ2
a
where ημν is the ’t Hooft antisymmetric symbol defined by,
a
ημν = aμν μ, ν = 1, 2, 3 a
ημ4 = −η4μ
a
= δμa
a
η̄μν = aμν μ, ν = 1, 2, 3 a
η̄μ4 = −η̄4μ
a
= −δμa . (22.14)

a a
It is easy to check that ημν and η̄μν are self-dual and anti self-dual respectively.
The corresponding field strength takes the form,
ρ2
a
Fμν = −4ημν
a
. (22.15)
[(x − X)2 + ρ2 ]2
a a
Obviously since ημν is self-dual so is Fμν .
The one instanton solution is characterized by eight parameters, four corre-
spond to the center of the instanton Xμ , one to the size of the instanton ρ and
three to three global SU (2) gauge transformations. Recall that fixing a guage
we fix only the local gauge transformations. The space of parameters of the k
instanton solutions will be further addressed in Section 22.3 where it will be
shown that in general for SU (N ) the dimension of the moduli space is 4kN .
The instanton solution (22.13) falls off asymptotically as x1 and hence it con-
tributes a finite amount to the integral of the topological charge (winding num-
ber) which is of the form A3 x3 dΩ. The field strength falls off as 1/x4 such
that indeed the corresponding action is finite. However due to the x1 asymptotic
behavior it is difficult to form square integrable expressions that contain it. For
that purpose one can use the following singular gauge transformation,
σμ† (x − X)μ
U= , (22.16)
|x − X|
1
which renders the instanton to have a x3 fall off as can be seen from,
1 2ρ (x − X)ν η̄νa μ
2
Aaμ = . (22.17)
g (x − X)2 [(x − X)2 + ρ2 ]
The singular instanton is obviously singular at the location of the instanton
xμ = Xμ . This singularity is not physical and can be removed by a gauge trans-
formation or by puncturing the Euclidean space with the singular point being
removed.
Instantons of SU (N ) gauge theory can be constructed by embedding SU (2)
instantons in SU (N ) for instance,

S U (N ) 0 0
Aμ = S U (2) , (22.18)
0 Aμ
where the instanton is the 2 × 2 matrix on the lower right. The most general
SU (N ) one instanton configuration can be derived from (22.18) by the following
transformation,

† 0 0 SU (N )
S U (N )
Aμ =U S U (2) U U∈ . (22.19)
0 Aμ SU (N − 2) × U (1)
Operating with U ∈ SU (N − 2) × U (1) obviously leaves the basic configuration

invariant and thus only transformations with elements of the coset are relevant.
This is in accordnace with the fact that for a k instanton solution the stability
group is S(U (N − 2k) × U 1)) as will be shown in Section 22.3. The dimension

of the coset is N 2 − 1 − ((N − 2)2 = 4N − 5. Together with the 5 parameters

of the location and the size we have 4N collective coordinates. Indeed in Sec-
tion 22.3 we will see that in general the dimension of the moduli space is
4N k for instanton number equals k solution. To demonstrate this counting con-
sider the case of SU (3) for which the generators are the Gell–Mann matrices
{λa }, a = 1, . . . , 8. The first three generators λa , a = 1, 2, 3 form the SU (2)
k = 1 instanton. λ4 , . . . , λ7 form two doublets under this SU (2) so they can gen-
erate new solutions while λ8 commutes with the SU (2) and hence leaves the
basic SU (2) solution invariant.
One can express the instanton solution in terms of quaternionic notation. This
will turn out to be convenient for the ADHM construction of multi-instanton
solutions (see Section 22.2). The idea is to make use of the representations of
the covering group SU (2)L × SU (2)R of the Lorentz group of four-dimensional
Euclidean space-time rather than the SO(4) Lorentz group itself. In particular we
represent any four vector of SO(4) as a (2, 2) representation of SU (2)L × SU (2)R ,
for instance the four vector xμ is denoted by xα α̇ or xα̇ α defined as follows,
xα α̇ = xμ σαμ α̇ xα̇ α = xμ σ̄μα̇ α , (22.20)
where σαμ α̇ is a 2 × 2 matrix of (iσ, 1) and σ̄α̇μ α = (σαμ α̇ )† . In terms of the quater-
nionic notation the one instanton solution (22.13) for SU (2) gauge theory is
given by,
1 2(x − X)ν Σν μ
Aμ = , (22.21)
g (x − X)2 + ρ2
where Σμν , which were introduced in Section 17.1, are the part of the Lorentz
generators that do not act on the space-time coordinates but only on the internal
degrees of freedom. Here using the SU (2)L × SU (2)R notation we define them
as follows,
1 1
Σμν = (σμ σ̄ν − σ̄ν σμ ) Σ̄μν = (σ̄μ σν − σν σ̄μ ). (22.22)
4 4
The self-duality property of the instanton configuration follows trivially from the
fact that Σμν is self-dual, namely,
1 1
Σμν = μν λρ Σλρ Σ̄μν = − μν λρ Σ̄λρ . (22.23)
2 2
The corresponding field strength reads,
1 4ρΣμν
Fμν = . (22.24)
g ((x − X)2 + ρ2 )2
As was discussed above, the instanton solution (22.21) falls off asymptotically as
1
x and hence it is difficult to form square integrable expressions that contain it.
The solution in the singular gauge now reads,
1 2ρ2 (x − X)μ Σ̄ν μ
Aμ = . (22.25)
g ((x − X)2 (x − X)2 + ρ2 )

22.2 The ADHM construction of instantons

The vacua of the YM theory is given by pure gauge configurations (which can
be written in terms of a double index notation,
1 l
Aμij = U † i ∂ μ Ulj , (22.26)
g
where i, j, l = 1, . . . , N . It is straightforward to check that this gauge field obeys
the self-duality condition. The idea of the ADHM construction3 is to generalize
this configuration also to the k instanton case by taking now the matrices U to
be of the form UIi where I = 1, . . . , N + 2k with the orthonormality condition,
I
U † i UIj = δij . (22.27)
The U matrices are the basis vectors of a null space,
α̇ I I
Δ† I UIi = 0 = U † i ΔII α̇ , (22.28)
where I = 1, . . . , k and ΔII α̇ is a (N + 2k) × 2k complex valued matrix which is
taken to be linear in the space-time coordinate xμ , namely takes the form,
α̇ I α̇ I
ΔII α̇ = aII α̇ + bαII xα α̇ Δ† I = a† I + xα̇ α bII α , (22.29)
α̇ I
and with Δ† I
≡ (ΔII α̇ )∗ .
The ADHM k instanton solution of the form (22.26) is self-dual if one further
requires that ΔII α̇ obeys the following condition,
α̇ I
Δ† I ΔIJ β̇ = δβ̇α̇ (f −1 )I J , (22.30)
where f is an arbitrary x-dependent k × k dimensional Hermitian matrix. Since
fI J (x) is arbitrary there are three “ADHM constraints” on a, a† , b and b† ,
α̇ I 1 †
a† I aIJ β̇ = a a δβ̇α̇
2 IJ
α̇ I βI
a† I bII β̇ = b† I aα̇I J
I 1 †
b† α I bβII = a a δαβ . (22.31)
2 IJ
It is straightforward to realize that the ADHM construction is invariant under,

Δ → AΔB −1 U → AU f → Bf B † , (22.32)
where A ∈ U (N + 2k) and B ∈ GL(k, C). Thus by construction there is a redun-
dancy in a and b. One can choose a simple canonical form for b and a as follows,
0 I
bβIJ = b† β J = (0 , δβα δj i )
δαβ δI J
α̂iJ α̇ α̇ I † α̇ al
aIJ α̇ = a† J = (âα̇J i , (a )I J ). (22.33)
(aα α̇ )I J
where I = i + Iα.
3 Multi-instanton solutions were presented in [220], [132] and other papers. Our discussion of
the construction of multi-instanton is based on the paper of ADHM [20]. This approach was
further discussed in [70]. We follow the description of the construction given in [81].

22.2 The ADHM construction of instantons 395
In this parametrization the third ADHM constraint is automatically obeyed

while the other two take the form,
β̇ β̇ β̇
σβ̇α̇ (a† aα̇ ) = σβ̇α̇ (ˆ(a† ) âα̇ + a† aα̇ ) = 0, (22.34)
where we made use of the fact that a must be Hermitian. In this canonical
parametrization the matrix f reads,
α̇
f = 2((â† ) âα̇ + (aμ + xμ 1[k ]×[k ] )2 )−1 . (22.35)
The field strength Fμν that corresponds to the ADHM k instanton configura-
tion (22.26) can be written in the following form,
1 1
Fμν = ∂μ Aν − ∂ν Aμ + g[Aμ , Aν ] = ∂[μ (U † ∂ν ] U ) + (U † ∂[μ U )(U † ∂ν ] U )
g g
1 1
= ∂[μ U † (1 − U † U )∂ν ] U ) = ∂[μ U † Δf Δ† ∂ν ] U
g g
1 † 1 1
= U ∂[μ Δf ∂ν ] Δ† U = U † bσ[μ σ̄ν ] f b+ U = 4 U † bσμν f b+ U, (22.36)
g g g
where we have made use of,
J α̇ J
PIJ ≡ UIi U † = δIJ − ΔII α̇ fI J Δ† J . (22.37)
To get an explicit expression for Aμ we make use of the decomposition

Ûij âiJ α̇
UIi = ΔIJ α̇ = . (22.38)
(Uα )I i (Δα α̇ )I J
From the completeness condition (22.37) Û can take the form,

α̇ −1
U = (i[N ]×[N ] − âα̇ f (â)† ) U = Δα̇ f (â)† (û)† .
α̇
(22.39)
We next show that for the particular case of k = 1 the ADHM solution (22.26)
is identical to (22.25). For k = 1 we have to drop the indices I, J. One can verify
that in that case the parameters aμ can be identified with the center of the
instanton Xμ . From the ADHM constraint (22.34) we get that,
β̇
(â† ) âα̇ = ρ2 δα̇β̇ , (22.40)
and,
1
f= , (22.41)
(xμ − Xμ )2 + ρ2
where ρ will naturally be the size of the instanton.
From the relation (22.39) we deduce that,
?
1 (x − X)2
Û = 1[N ]×[N ] + 2 − 1 (â† )β̇ âα̇ ,
ρ (x − X)2 + ρ2
(x − X)α α̇ (â† )α̇
U = − . (22.42)
|x − X| (x − X)2 + ρ2

Plugging these expressions into U we finally get the singular form of the one
instanton solution (22.25).
In general for k = 1 one can show that the gauge configuration given in (22.36)
indeed carries an instanton of charge k. To derive this result one makes use of
the following relation,
g 2 Tr [Fμν
2
] = ∂μ ∂ μ trk [log f ]. (22.43)
This relation can be proven by expanding the two sides of the equations using
the explicit expression for Fμν (22.36). Upon integrating (22.43) over the whole
1
Euclidean space-time divided by 16π 2 and making use of the fact that asymp-
totically f (x) → x 2 , we find that indeed it is equal to the instanton charge k.

1
22.3 On the moduli space of instantons

The moduli space of instantons4 M is the space of inequivalent self-dual Yang–
Mills configurations. The notion of moduli space of solutions is an important tool
in general and in particular for instantons. We have encountered it in the context
of magnetic monopoles in (21.10). We will elaborate in this section about the
basic properties of the moduli space of instantons such as its dimension, complex
structure, metric, symmetries, and singularities.
Consider a small fluctuation δAμ (x) around an instanton solution Aμ (x) which
is also a self-dual solution of the YM equation, namely, it obeys to linear in
δAμ (x),
Dμ δAν − Dν δAμ = μν ρσ Dρ δAσ , (22.44)
where Dμ is the covariant derivative in the instanton background. In terms of
the quateronic notation this equation reads,
β̇ α
σβα̇ D † δAα α̇ = 0, (22.45)
where D = σ μ Dμ . Next we would like to guarantee that the fluctuation is not a
local gauge transformation. This can be done by requiring that the fluctuation
be orthogonal to any gauge transformation, namely,

d3 x Tr[δAμ Dμ Λ] = 0 → Dμ δAμ = 0, (22.46)
where we have made use of an integration by parts to derive the last expression.
α̇ α
In the quaternionic notation this condition takes the form of D † δAα α̇ = 0
which combined with (22.45) is given by,
α̇ α
D† δAα β̇ = 0. (22.47)
4 The properties of the moduli space of instantons were discussed by many authors. In partic-
ular [155], [141] and [80]. The review about the moduli space that we are using is [81].

The fluctuation that obeys this equation is referred to as a zero mode since it is
a physical fluctuation that does not change the action. Note that this is exactly
the equation of motion of a Weyl spinor in the background of the instanton
Aμ (x). The zero modes defined by (22.47) are the collective coordinates of the
instantons.
Since the YM instantons are characterized by the topological charge defined
in (22.2) so is the corresponding moduli space. We thus discuss the moduli space
of instantons of charge k which we denote by Mk .
It can be proven that the moduli space of instantons is a manifold. In fact we
will see below that Mk has some conical singular points associated with zero size
instantons. The coordinates on the moduli space are the collective coordinates
that were just shown to be equivalent to the zero model (22.47). We denote by Xn
the collective coordinates where n = 1, . . . , dim Mk . A trivial set of coordinates
are the space-time coordinates of the center of the instanton Xμ accordingly the
moduli space is a product of the form,
Mk = R4 × M̂k . (22.48)
The collective coordinates Xμ follow from the fact that the instanton solu-
tion breaks the symmetry of the action under space-time translations. There are
other collective coordinates that associate with symmetries of the theory that the
instanton configuration breaks. However, not all symmetries yield non equivalent
collective coordinates and not all the coordinates associate with broken symme-
tries.
From the ADHM construction it follows that the moduli space is identified
with the variable aα̇ subject to the ADHM constraints (22.31) quotiented by the
residual U (K) symmetry transformation (22.32) with,
1[N ]×[N ] 0
A= , B=C C ∈ U (k), (22.49)
0 C1[2]×[2]
which preserve the canonical form of b (22.33) and transform a as follows,
âiI α̇ → âα̇ C aμ → C † aμ C. (22.50)
Thus the dimension of the moduli space is,
dim Mk = 4k(N + K) − 4 dim U (k) = 4k(N + k) − 4k 2 = 4N K. (22.51)
This result can be derived also by using an index theorem that counts the zero
modes at a point in the moduli space. In fact as we will see shortly the space Mk
is a hyper-Kahler quotient of the flat space R4k (k +N ) by the U (k) group. The
one instanton solution of SU (2) is indeed characterized by the four coordinates
of its center, its size and three global SU (2) gauge transformations.
The moduli space is a complex manifold. A complex manifold is an even-
dimensional manifold that admits a complex structure I a linear map of the

tangent space to itself such that I2 = 0. There are always local holomorphic
coordinates (Z i , Z̄ i ), i = 1, . . . , n (see Section 1.1) for which,
iδj̄ī 0
I= g = gi j̄ dZ i dZ̄ j̄ w = igi j̄ dZ i ∧ dZ̄ j̄ . (22.52)
0 −iδji
where g is an Hermitian metric and w is referred to as the fundamental 2 form.
In the case that the fundamental 2 form is closed namely dw = 0 it is called the
Kähler form and the associated manifold is a Kähler manifold. The latter is also
characterized by the fact that the complex structure is covariantly constant and
the Kähler metric can be derived from a Kähler potential,
∇μ I = 0 gi j̄ = ∂i ∂j̄ K. (22.53)
The moduli space of instanton is not only a Kähler manifold but in fact a hyper
Kähler manifold which means that it admits three linearly independent complex
structures, I(c) , c = 1, 2, 3 that satisfies the algebra
I(c) I(d) = −δ cd + cde I(e) . (22.54)
The four-dimensional Euclidean space R4 is hyper Kähler and the three complex
structures are,
μν = −ημν
I(c) c
(I · x)α α̇ = ixα β̇ σα̇β̇ , (22.55)
c
where ημν is the ’t Hooft η symbol defined in (22.14) and the expression in the
left-hand side is the quaterionic formulation. Now recall that by the definition
of the zero modes (22.47), if δn Aα α̇ is a zero mode so is also δn Aα α̇ Cα̇β̇ for any
constant matrix C and in particular also to σ and hence if δn Aα α̇ so is also
(I · δn A)α α̇ = iδn Aα α̇ σα̇β̇ . Since the zero modes form a complete set there must
exist Inm such that,
(I · δm A)α α̇ = δn Aα α̇ Inm , (22.56)
from which it implies that Inm satisfies the algebra (22.54).
The Kähler potential which is common to the three complex structures of the
moduli space of instantons takes the form,

g2
K=− d4 xx2 TrFμν F μν . (22.57)
4
Using the form of I(c) on R4 given in (22.52) for instance I (3) associated with
the complex coordinates on R4 ix3 + x4 and ix1 − x2 , we find that,
(I(c) · ∂Z i A)α α̇ = i∂Z i Aα α̇ , (22.58)
for instance for I(3) we get ∂Z i Aα 2 = ∂Z̄ ī Aα 1 = 0. Furthermore the derivative
with the respect to the holomorphic and anti-holomorphic coordinates of the
gauge fields obey the equations of zero modes namely,
δi Aμ ≡ ∂Z i Aμ δ̄ī Aμ ≡ ∂Z̄ ī Aμ . (22.59)

It further follows that ∂Z̄ j̄ ∂Z i Aμ = 0 and hence,

2
∂Z̄ j̄ ∂Z i Tr [Fμν ] = ∂μ ∂ μ Tr [δi Aμ δ̄j̄ Aμ − 2∂μ ∂ν ] Tr δi Aμ δ̄j̄ Aν . (22.60)
Upon integrating by parts twice we find the metric on the moduli space,

∂Z̄ j̄ ∂Z i K = −2g 2 d4 x Tr δi Aμ δ̄j̄ Aμ = gi j̄ . (22.61)
Next let us now discuss the symmetries of the moduli space, in particular the
realization of symmetries of the gauge theory which are broken by the instanton
configuration. We start with the four-dimensional conformal group (see Section
17). In the quaternionic formulation the basic variable of the ADHM construction
Δ is transformed as follows,
AB
x → x = (Ax + B)(cX + D)−1 det =1
CD
Δ(x; a, b) → Δ(x ; a, b) = Δ(x; aD + bB, aC + bA)(Cx + D)−1 (22.62)
In fact the term (Cx + D)−1 in the right-hand side of the last equation is irrele-
vant since the gauge field depends on U and U † defined in (22.26) is redundant.
We can now use transformations of (22.32) that keeps the canonical structure
of b (22.33). Upon applying this transformation a goes into,
a → A(aD + bB)B −1 . (22.63)
A particular example of transformations which belong to the conformal group
are the translations. For this case,
Δ(x; a, b) → Δ(x; a + b, b), (22.64)
from which it follows that,
aμ → aμ + μ 1[k ]×[k ] âα̇ → âα̇ . (22.65)
It is thus clear that indeed the components aμ are proportional to the coordinates
of the center of the instanton,
trk aμ = kXμ . (22.66)
Global gauge transformations act non trivially on the ADHM variables if N ≤
2k, while if N ≥ 2k there are transformations that leave the instantons fixed.
This is the stability group of the instanton. One can embed the k instanton
solution in an SU (2k) subgroup of SU (N ) and show that the stability group is
S(U (N − 2k) × U (1)).
The moduli space M̂k is in fact not a smooth manifold due to certain singu-
larities. However, these singularities do not signal any pathology of the moduli
space and integrals over the moduli space are well defined. It can be shown that
M̂k is a cone. For the moduli space of single instanton k = 1 the apex of the
cone is the point ρ = 0 where the instanton has a zero size. This structure can
be generalized also to the k = 1 instantons.

Topological characteristics of the moduli space can be described by a topologi-

cal field theory where the observables of the theory are the topological invariants.
This is beyond the scope of this book and we refer the interested reader to the
list of references for this chapter.
22.4 Instantons and tunneling between the vacua of the

YM theory
The vacua of the YM theory in Minkowsi space-time are defined to be the gauge
configurations for which the energy vanishes. Using the temporal gauge A0 = 0,
the Hamiltonian of the theory is

1
H= 2 d3 x Tr [E 2 + B 2 ]. (22.67)
2g
Thus a classical vacuum has a vanishing field strength,
Fμν = 0 → Ai (x, t) = iU (x, t)∂i U (x, t)† . (22.68)
Thus the vacuum gauge configuration is that of a pure gauge. Prior to a dis-
cussion of how to tunnel between two vacuum states, we have to classify and
enumerate the vacua namely following (22.68) the group elements U (x). This
translates to the equivalence classes of maps from S 3 to the SU (N ) group man-
ifold. This is done by the topological charge or winding number or Pontryagin
number defined in (22.2). Since this step is very essential in the discussion of the
tunneling let us clarify this point. Let us analyze the tunneling between a vacuum
state Ai (x, t1 ) at t = t1 into another vacuum state Ai (x, t2 ) at t = t2 . On top of
fixing A0 (x, t) = 0 we can use the residual gauge symmetry to set Ai (x, t1 ) = 0.
Next we consider a path in the space of gauge configurations that connects the
two vacua points and has a finite energy H (in Minkowski space-time). Finite
energy implies that for large |x| → ∞ it has to be a pure gauge,
Aμ →|x|→∞ U ∂μ U † . (22.69)
Since A0 = 0, U (x, t) = U (x) is time independent. Because U (x, t1 ) = 1 we

obtain that for all t and |x| → ∞, U = 1 and hence also Ai (x, t2 ) −→|x|→∞ 0. The
fact that asymptotically in |x| all Ai = 0 allows us to compactify the spacelike
hypersurface at fixed t into S 3 . The following Fig. 22.1 describes the situation.
On the boundary of hyper-cylinder the gauge fields Ai vanish apart from on
the hyper-disk at t = t2 where Ai is a pure gauge. Consider now the topological
charge which we have seen (22.2) is in fact a surface term. Since on the boundary
Fμν = the contribution to the surface integral takes the form,
1 μν ρσ
Q= dσμ Tr [Aν Aρ Aσ ]
24π2

1 0ij k
= d3 xr[U ∂i U † U ∂j U † U ∂k U † ], (22.70)
24π2

22.4 Instantons and tunneling between the vacua of the YM theory 401
Aj = e–α je
α
t = t2
Aj = 0
t = t1
Aj = 0
Fig. 22.1. Compactification of the space coordinates on S 3 at fixed t.
where in the last expression the integration is over the three sphere at t = t2
since at all other parts of the surface of the hyper-cylinder Ai = 0. Thus
the configurations in Minkowski space-time that connect a vacuum state at
t = t1 to one at t = t2 are classified by the winding number of the maps of
S 3 (space) → S 3 (group) just as the maps of instanton in Euclidean space-time.5
In the latter case S 3 (space) is the boundary of R4 whereas in the former it is a
compactification of R3 at t = t2 .
It is easy to realize that there is no way to interpolate a vacuum at t = t1 of
zero winding number with a one at t = t2 with non vanishing winding number
with a configuration of zero energy. The latter corresponds to a pure gauge
configuration which has Fμν = 0 everywhere and hence also vanishing topological
number. Thus the energy of the tunneling configuration as a function of time
should look as in Fig. 22.2.
To identify the configuration that has the largest tunneling rate we consider a
family of gauge configurations characterized by the collective coordinates asso-
ciated with a coordinate transformation from t to λ(t) such that,
(λ)
Ai (x, t) = Ai (x, λ(t)), (22.71)
with the requirement that λ(t1 ) = t1 and λ(t2 ) = t2 . Next we compute the elec-
tric and magnetic fields,
∂Ai
Ei = Fi0 = −∂0 A(λ) (x, t) = (x, λ(t))λ̇
∂λ
1 1
Bi = ij k Fj k = ij k (∂j Ak (x, λ(t)) + Aj (x, λ(t)(Ak (x, λ(t)) − (j ↔ k) (22.72)
2 2
5 The role of instantons in tunneling between different vacua was proposed in [131]. It was also
discussed in [26], [40] and [44]. This topic is reviewed in [185] and in [209]. We follow the
latter.

H (t)
t
t1 t2
Fig. 22.2. The energy of the tunneling configuration as a function of time.
and substitute them into the Lagrangian L = d3 xL = d3 x[− g12 Tr [E 2 − B 2 ]]

which can be written in the following form,
1
L = − m(λ)(λ̇)2 − V (λ),
2
2
2 ∂Ai
m(λ) = 2 d3 x Tr ≥ 0,
g ∂λ

1
V (λ) = − 2 d3 x Tr (B i )2 ≥ 0. (22.73)
g
The Lagrangian (22.73) is the Lagrangian of a particle that moves from one
vacuum at t = t1 where V (λ) = m(λ) = 0 to a vacuum at t = t2 , where again
V (λ) = m(λ) = 0. The quantum mechanical tunneling rate is proportional to
e−2R where R is given by,
λ2
R= 2m(λ)(V (λ) − E)
dλ
λ1
F
λ2 G G 2
1 ∂A 1
dλH
i
= d3 x Tr d3 x Tr (Bi )2
λ1 g2 ∂λ g2
t2 ?
2 1
= 2 dt d3 x Tr (Ei )2 d3 x Tr (Bi )2 . (22.74)
g t1 g2
Using the triangle inequality we can relate the tunneling rate to the winding or
instanton number as follows,
t2
2 8π2
R ≥ 2| d4 x Tr [E i B i ]| = 2 |k|, (22.75)
g t1 g

where the instanton number is Q = k. Thus we see that the tunneling rate is
bounded by,
2
− 8gπ2 |k |
e−R ≤ e , (22.76)
and the bound is saturated for instanton configurations. To be more precise

the most probable tunneling paths are given by a Minkowski gauge configura-
tion with E = ±B which when viewed as a configuration in Euclidean space
are instantons. Conversely given an instanton AE
μ (x, t) in Euclidean space one
can construct a set of paths in Minkowski space-time AM μ (x, λ(t)) such that
M ,(λ) M ,(λ)
Ai (x, t) = AE
μ (x, λ(t)) and A0 (x, t) = AE
4 (x, λ(t)).
22.5 Instantons, theta vacua and the UA (1) anomaly

It was shown in the previous section that the instantons connect different vacua.
This means that the vacuum of the YM theory cannot be described by any of
the states of zero energy and a specific topological charge, but instead has to be
a superposition of all these states, namely,

|θ> = eik θ |k> . (22.77)
k
The generator of large gauge transformation that changes the winding number by
one unit, namely, T |k> = |k + 1> has to be a symmetry generator that commutes
with the Hamiltonian so that T |vac> = eiϕ |vac> for some phase ϕ. Indeed for

the θ vacuum we get T |θ> = k eik θ |k + 1> = e−iθ |θ>.
The energy associated with the θ vacua given by,
E(θ) = −2K cos(θ)e−S . (22.78)
This follows from the following steps. Consider the amplitude to tunnel from a
vacuum |i> to a vacuum |j> is given by,
δN −N −(j −i)
<j|e−H t |i> = N± + − (Kte−S )N + +N − , (22.79)
N+ !N− !
when the instantons are sufficiently dilute and where K is the pre-exponential
factor in the tunneling amplitude, and N± are the number of instantons and anti-
instantons. We introduce the parameter θ via a representation of the Kroneker
delta function,
2π
1
δN + −N + + (i−j ) = eiθ (N + −N + +(i−j )) . (22.80)
2π 0
Upon performing the summations over N+ and N− we get,
2π
−S
<j|e−H t |i> = eiθ (i−j ) e2K t cos(θ )e , (22.81)
0

which implies that the energy of the θ vacuum is as given in (22.78). Note that
this does not imply that the YM theory has a continuous spectrum without mass
gap since the θ parameter is fixed for a given theory and it cannot be changed.
Fixing the value of θ can be achieved by adding a θ term to the action (22.10),

1 1 θ
SYM = 2 d x Tr Fμν F → 2 d x Tr Fμν F +
4 μν 4 μν
d4 x Tr [Fμν ∗ F μν ].
2g 2g 16π2
(22.82)
The additional θ term is a surface term and hence does not affect the equations
of motion, however it is not invariant under CP or T transformations.6 As will be
discussed below, with no massless quarks indeed the θ term implies a strong CP
violation. The most severe restriction on CP violation comes from the electric
dipole moment of the neutron. This sets the upper bound to theta to be,
θ < 10−9 . (22.83)
The puzzle of why θ is so tiny is referred to as the strong CP problem. One
proposal to handle this problem is the introduction of the axion, χ, a pseudo
scalar field with a coupling of the form χ Tr [Fμν ∗ F μν ] so that the effective θ is
√
the sum of < χ > and the θ term. As will be discussed below there is in fact
an even simpler mechanism to resolve the strong CP problem and that is having
a massless u quark. This brings us to the next topic which is the incorporation
of light quarks to the game.
In the presence of light quarks there is a simple physical observable
that distinguishes between the different topological vacua, the axial current.
Recall that for Nf massless quarks the theory is classically invariant under
global UL (Nf ) × UR (Nf ) ≡ SUL (Nf ) × SUR (Nf ) × UB (1) × UA (1) symmetry.
The SU (Nf ) × UV (1) symmetry group factors are realized in nature also quan-
tum mechanically. The invariance under the axial SU (Nf ) transformations is
broken spontaneously and there are Nf2 − 1 Goldstone bosons. For Nf = 2 these
are the pions. The UA (1) axial symmetry is not conserved quantum mechanically.
In analogy to the anomaly of the axial symmetry in two dimensions discussed
in Section 9.1, in four dimensions as well one can show using various different
methods that,
Nf
∂ μ Jμ5 = Tr [Fμν ∗ F μν ], (22.84)
8π2

where Jμ5 = i ψ̄i γμ γ5 ψi is the axial current and ψi i = 1, . . . , Nf are the fields of
the various flavored quarks. This resolves the so-called UA (1) puzzle, namely the
absence of the fourth Goldstone boson for Nf = 2 or the ninth one for Nf = 3.
Indeed for the former case one could associate the η pseudo-scalar meson with
the fourth Goldstone boson, however it has a mass of 478M eV whereas a current
6 A proposal for resolving the strong CP problem was proposed in [172]. The U A (1) problem
has been resolved by ’t Hooft. The mass of the η was proposed by Witten [221] and by
Veneziano [216]. Our discussion of this topic follows the review [185].

√
algebra theorem states that it has to be lighter than 3mπ . The right-hand side
of (22.84) is proportional to the divergence of the topological current (22.2) ∂μ Kμ
so one can define a modified conserved axial current j̃μ5 = jμ5 − Nf Kμ . However,
unlike the topological charge, the topological current is not gauge invariant. A
massless pole in the correlator of Kμ does not necessarily correspond to a massless
particle. One may wonder also about the fact that the right-hand side of (22.84)
is a surface term and hence cannot have a physical significance. However, as was
emphasized above due to instantons the surface term is relevant. Let us see this
explicitly. We start by computing the change in the axial charge,

ΔQ5 = Q5 (t = +∞) − Q5 (t = −∞) = d4 x∂ μ Jμ5

= Nf d4 x∂ μ Tr [S(x, x)γμ γ5 ], (22.85)
where S(x, y) is the fermion propagator S(x, y) = <x|(i D )−1 |y> that can be
determined from the eigenfunction equation i D ψλ = λψλ in the form S(x, y) =
ψ λ (x)ψ λ† (y )
λ . Substituting this expression we get,
λ

ψλ (x)ψ † (y)
ΔQ5 = Nf 4 μ
d x∂ Tr λ
2λγ5 = 2Nf (nL − nR ), (22.86)
λ
λ
where we have used the fact that ψλ and γ5 ψλ are orthogonal so only the nL (nR )
left (right) zero modes contribute.
Integrating the left-hand side of (22.84) we get the topological charge Q which
is thus related to ΔQ5 . The latter counts the number of left-handed zero modes
minus the number of right-handed zero modes. This is obviously associated with
instantons. Each instanton contributes one unit to the topological charge and
has a left-handed zero mode, whereas an anti-instanton has a right-handed zero
mode and Q = −1. This is the way the instantons contribute to the axial anomaly
and hence to the resolution of the UA (1) problem. For the case of Nf = 3 this
implies that this would be the ninth Goldstone boson, the η is massive even if
the quark masses vanish. It was shown that the mass of the η is related to the
topological susceptibility in the following form,

2Nf 2Nf
χ top = d4 x <Q(x)Q(0)>= m2η + m2η − 2m2K . (22.87)
fπ2 fπ2
The combination of masses on the right-hand side corresponds to the part of the
η mass which is not due to the strange quark mass.

23
Summary, conclusions and outlook
23.1 General
Relativistic quantum field theory has been very successful in describing strong,
electromagnetic and weak interactions, in the region of small couplings by per-
turbation theory, within the framework of the standard model.
However, the region of strong coupling, like the hadronic spectrum and various
scattering phenomena of hadrons within QCD, is still largely unsolved.
A large variety of methods have been used to address this question, includ-
ing lattice gauge simulations, light-cone quantization, low energy effective
Lagrangians like the Skyrme model and chiral Lagrangians, large N approxima-
tion, techniques of conformal invariance, the integrable model approach, super-
symmetric models, string theory approach, QCD sum rules, etc. In spite of this
major effort the gap between the phenomenology and the basic theory has only
been partially bridged, and the problem is still open.
The goals of this book are to provide a detailed description of the tool box
of non-perturbative techniques, to apply them on simplified systems, mainly of
gauge dynamics in two dimensions, and to examine the lessons one can learn
from those systems about four-dimensional QCD and hadron physics.
The study of two-dimensional problems to improve the understanding of four-
dimensional physical systems was found to be fruitful. This follows two directions,
one is the utilization of non-perturbative methods on simpler setups and the
second is extracting the physical behavior of hadrons in one space dimension.
Obviously, physics in two dimensions is simpler than that of the real world
since the underlying manifold is simpler and since the number of degrees of
freedom of each field is smaller. There are some additional simplifying features
in two-dimensional physics. In one space dimension there is no rotation symmetry
and no angular momentum. The light-cone is disconnected and is composed of
left moving and right moving branches. Therefore, massless particles are either
on one branch or the other. These two properties are the basic building blocks
of the idea of transmutation between systems of different statistics. Also, the
ultra-violet behavior is more convergent in two dimensions, making for instance
QCD2 a superconvergent theory.
In this summary chapter we go over several notions, concepts and methods
with emphasis on the comparison between the two- and four-dimensional worlds
and what one can deduce about the latter from the former. In particular we deal

408 Summary, conclusions and outlook
with conformal invariance, integrability, bosonization, solitons and topological

charges, confinement versus screening and finally the hadronic spectrum and
scattering.
23.2 Conformal invariance

From the outset there is a very dramatic difference between conformal invari-
ance in two and four dimensions. The former is characterized by an infinite-
dimensional algebra, the Virasoro algebra, whereas the latter is associated with
the finite-dimensional algebra of SO(4, 2). This basic difference stems from the
fact that whereas the conformal transformations in four dimensions are global,
in two dimensions the parameters of conformal transformations are holomorphic
functions (or anti-holomorphic), see Section 17.5 versus 2.1. Nevertheless there
are several features of conformal invariance which are common to the two cases.
We will now compare various aspects of conformal invariance in two and four
dimensions:
r The notion of a primary field and correspondingly a highest weight state is
used both in two-dimensional conformal field theories as well as for the four-
dimensional collinear algebra. It is expressed in the former as (17.38),
L0 [φ(0)|0>] = h[φ(0)|0>] Ln [φ(0)|0>] = 0, n>0 (23.1)
and for the latter,
L0 [Φ(0)|0>] = j[Φ(0)|0>] L− [Φ(0)|0>] = 0. (23.2)
The difference is of course the infinite set of annihilation operators Ln

versus the single annihilation operator L− in four dimensions.
r The COPE, the conformal operator product expansion has a compact form in
two dimensional CFT (Section 2.12)

Oi (z, z̄)Oj (w, w̄) ∼ Cij k (z − w)h k −h i −h j (z̄ − w̄)h̄ k −h̄ i −h̄ j Ok (w, w̄),
k
(23.3)
where Cij k are the product coefficients, while in four dimensions it reads,
∞
1/2(t A +t B −t n )
1 xn−+s 1 +s 2 −s A −s B
A(x)B(0) = Cn
n=0
x2 B(jA − jB + jn , jB − jA + jn )
1
× duu(j A −j B +j n −1) (1 − u)(j B −j A +j n −1) Onj 1 ,j 2 (ux− ), (23.4)
0
where the definitions of the various quantities are in Chapter 17. Again there
is a striking difference between the simple formula in two dimensions and the
complicated one in four dimensions.

23.2 Conformal invariance 409
r As an example let us compare the OPE of two currents. Recall from the dis-
cussion of Chapter 3 the expression in two dimensions reads,
kδ ab fcab J c (w)
J a (z)J b (w) = + i + finite terms, (23.5)
(z − w)2 (z − w)
for any non-abelian group, and in particular for the abelian case the second
term on the right-hand side is missing. For comparison the OPE of the trans-
verse components of the electromagnetic currents given in Chapter 17 takes
the form,
J T (x)J T (0) ∼
∞ 1 (6−t n )/2 1
n = 0 Cn x 2 (−ix− )n + 1 Γ(jΓ(2j n)
n )Γ(j n ) 0
du[u(1 − u)]j n −1 Qn1,1 (ux− ).
(23.6)
r The conformal Ward identity associated with the dilatation operator in four
dimensions (17.60),

N
(lφ + γ(g ∗ ) + xi ∂i ) <T φ(x1 )...φ(xN )>= 0, (23.7)
i
where lφ is the canonical dimension and γ(g ∗ ) is the anomalous dimension,
seems very similar to the one in two dimensions,

(zi ∂i + hi ) <0|φ1 (z1 , z̄1 )...φn (zn , z̄n )|0>= 0. (23.8)
i
In both cases one has to determine the full quantum conformal dimensions of
the various operators. However, as was shown in Section 2.7, in certain CFT
models, like the unitary minimal models, there are powerful tools based on
unitarity which enable us to determine exactly the dimensions hi of all the
primary operators and hence all the operators of the model. On the other
hand, it is a non-trivial task to determine the anomalous dimensions in other
models in two dimensions, and of course four-dimensional operators. In certain
supersymmetric theories there are operators whose dimension is protected, but
generically one has to use perturbative calculations to determine the anomalous
dimensions of gauge theories to a given order in the coupling constant.
Using the Ward identity one can extract the form of the two-point function
of operators of spin s in four dimensions. It is given by,
l φ +γ (g ∗ ) s
∗ ∗ −2γ (g ∗ ) 1 (x1 − x2 )+
<φ(x1 )φ(x2 )>= N2 (g )(μ ) .
(x1 − x2 )2 (x1 − x2 )−
(23.9)
The corresponding two-point function in two dimensions, which depends only
on the conformal dimension of the operator h, reads,
c2
G2 (z1 , z̄1 , z2 , z̄2 ) ≡<0|φ1 (z1 , z̄1 )φ1 (z2 , z̄2 )|0> = .
(z1 − z2 ) (z̄1 − z̄2 )2 h̄ 1
2h 1
(23.10)

r As for higher point functions, we have seen in Section 2.9 that one can use the
local Ward identities together with Virasoro null vectors to write down partial
differential equations. The result for a four-point function (2.63) was later used
to determine the four-point function of the Ising model (2.94). Two dimensional
conformal field theories are further invariant under affine Lie algebra transfor-
mations, and as we have shown in Section 3.6 those can be combined with
null vectors to derive the so-called Knizhnik–Zamolodchikov equations (3.69),
which were later used to solve for the four-point function of the SU (N ) WZW
model in Section 4.4. These types of differential equations that fully determine
correlation functions are obviously absent in four-dimensional interacting con-
formal field theories.
23.3 Integrability
Integrability was discussed in Chapter 5 in the context of two-dimensional mod-
els and in Chapter 18 in four-dimensional gauge theories. For systems with a
finite number of degrees of freedom, like spin chain models, there is a finite num-
ber of conserved charges, equal of course to the number of degrees of freedom.
For integrable field theories there is an infinite countable number of conserved
charges. Furthermore, the scattering processes of those models always involve a
conservation of the number of particles.
In two dimensions we have encountered continuous integrable models like the
sine-Gordon model as well as discretized ones like the XXX spin chain model.
The integrable sectors of gauge dynamical systems discussed in Chapter 18 are
based on identifying an exact map between certain properties of the systems and
a spin chain structure. In two dimensions the spin chain models follow from a
discretization of the space coordinate, by placing a spin variable on each site
that can take several values and imposing periodicity. In the four-dimensional
N = 4 super YM theory discussed in Section 18.1 the spin chain corresponds to
a trace of field operators and in the process of high-energy scattering of Section
18.2 it is a “chain” of reggeized gluons exchanged in the t-channel of a scattering
process. A summary of the comparison among the basic two-dimensional spin
chain, the “spin chains” associated with the planar N = 4 SYM, and the high-
energy scattering in QCD, is given in Table 23.1. A powerful method to solve
all these spin chain models is the use of the algebraic Bethe ansatz. This was
discussed in detail for the the XXX1/2 model in Section 5.14. The solutions of
the energy eigenvalues needed for the high-energy scattering process was based
on generalizing this method to the case of spin s Heisenberg model (see Section
18.2) and for the N = 4 to the case of an SO(6) invariance.
There is one conceptual difference between the spin chains of the two-
dimensional models and those associated with the N = 4 SYM in four dimen-
sions. In the former the models are non conformal, involving a scale, and hence

23.4 Bosonization 411
Table 23.1. Spin chain structure of the two-dimensional model and the
four-dimensional gauge systems of N = 4 SYM and of high-energy behavior
of scattering amplitudes in QCD.
Spin chain Planar High energy

N = 4 SYM scattering in QCD
Cyclic spin chain Single trace operator Reggeized guons in t-channel

Spin at a site Field operator SL(2) spin
Number of sites Number of operators Number of gluons
Hamiltonian Anomalous dilatation HBFKL
operator
1 lo g A
Energy eigenvalue Anomalous dimension ∼ λ lo g s
g −2 δD
Evolution time Global time The total rapidity log s
Zero momentum U = 1 Cyclicity constraint
also with particles and an S-matrix. The integrable sectors of four-dimensional

gauge theories, however, are conformal invariant.
The study of integrable models in two dimensions is quite mature, whereas the
application of integrability to four-dimensional systems is at an infant stage. The
concepts of multi-local charges described in Section 5.11 and of quantum groups
discussed in Section 5.13 have been applied only slightly to gauge dynamical
systems in four dimensions.
23.4 Bosonization
Bosonization is the formulation of fermionic systems in terms of bosonic variables
and fermionization is just the opposite process. The study of bosonized physical
systems offers several advantages:
(1) It is usually easier to deal with commuting fields rather than anti-commuting
ones.
(2) In certain examples, like the Thirring model, the fermionic strong coupling
regime turns into the weak coupling one in its bosonic version, the sine-
Gordon model (see Section 6.2).
(3) The non-abelian bosonization, especially in the product scheme (see Section
6.3.4), offers a separation between colored and flavored degrees of freedom,
which is very convenient for analyzing low lying spectrum.
(4) Baryons composed of NC quarks are a many-body problem in the fermion
language, while simple solitons are in the boson language.
(5) One loop fermionic computations involving the currents turn into tree level
consideration in the bosonized version. The best-known example of the latter
are the chiral (or axial) anomalies (see Section 9.1).

In four dimensions, spin is obviously non-trivial and one cannot constitute

generically a bosonization equivalence. However, in certain circumstances a four-
dimensional system can be described approximately by fields that depend only on
the time and on one space direction. In those cases one can apply the bosonization
technique. Examples of such scenarios are monopole induced proton decay, and
fractional charges induced on monopoles by light fermions. In these cases the
relevant degrees of freedom are in an s-wave and hence taken to depend only
on the time and the radial direction. This enables one to use the corresponding
bosonized field. There is a slight difference with two dimensions, as the radial
coordinate goes from zero to infinity, so “half” a line. Appropriate boundary
conditions enable us to use a reflection, so as to extend to a full line.
23.5 Topological field configurations

r The topological charges in any dimensions are conserved regardless of the equa-
tions of motion of the corresponding systems. In two dimensions it is very easy
to write down a current which is conserved without the use of the equations of
motion. This is referred to as a topological conservation. Consider a scalar field
φ or its non-abelian analog φa that transforms in the adjoint representation
of a group, then the following currents are abelian and non-abelian conserved
currents,
Jμ = μν ∂ ν φ Jμa = μν ∂ ν φa . (23.11)
Recall that for a system that admits, for instance an abelian case, also a current
Jμ = ∂μ φ that is conserved upon the use of the equations of motion, one can
then replace the two currents with left and right conserved currents J± = ∂± φ
or J = ∂φ and J¯ = ∂φ,¯ as was discussed in Chapter 1. The charge associated
with the topological conserved current is given by,

Qtop = dxφ = [φ(t, +∞) − φ(t, −∞)] ≡ φ+ − φ− , (23.12)
where the space dimension is taken to be R. For a compactified space dimen-

sion, namely an S 1 this charge vanishes, except for cases where the field is
actually an angle variable, in which case the charge is 2π. The latter appears
in the case of U (1) gauge theory in two dimensions, where there is a winding
number.
r Obviously one cannot have such topologically conserved currents and charges in
four dimensions. However, for theories that are invariant under a non-abelian
group, one can construct also in four dimensions a topological current and
charge, as for the cases of Skyrmions, magnetic monopoles and instantons. For
the Skyrmions the topological current is given by,
μ iμν ρσ
Jskyre = Tr [Lν Lρ Lσ ]. (23.13)
24π 2

23.5 Topological field configurations 413
Table 23.2. Topological classical field configurations in two and four dimensions
classical field dim. map topological current

1
soliton two μ ν ∂ν φ
baryon two S1 → S1 μ ν T r[g −1 ∂ν g]
iμ ν ρ σ
Skyrmion four S3 → S3 24π 2
T r[Lν Lρ Lσ ]
monopole four Ss2p a c e → SG2 / H 1
8π
μ ν ρ σ a b c ∂ ν Φ̂a ∂ ρ Φ̂b ∂ σ Φ̂c
iμ ν ρ σ
instanton four Ss3 → Sg3 16π 2
T r[Aν ∂ρ Aσ + 23 Aν Aρ Aσ ]
r The topological charges, for compact spaces, are the winding numbers of the
corresponding topological configurations. For a compact one space dimension,
we have the map of S 1 → S 1 related to the homotopy group π1 (S 1 ). In two
G /H
space dimensions, the windings are associated with the map S2 → S2 , as
for the magnetic monopoles. For three space dimensions, it is S → S 3 for
3
the Skyrmions at Nf = 2, and the non-abelian instantons for the gauge group
SU (2). The topological data of the various models is summarized in Table 23.2.
r According to Derrick’s theorem (see Section 5.3), for a theory of a scalar field
with an ordinary kinetic term with two derivatives, and any local potential at
D ≥ 2, the only non singular time-independent solutions of finite energy are the
vacua. However, as we have seen in Chapters 20, 21 and 22, there are solitons
in the form of Skyrmions and monopoles and instantons. Those configurations
bypass Derrick’s theorem by introducing higher derivative terms or including
non-abelian gauge fields.
r As was emphasised in Chapter 20, the extraction of the baryonic properties in
the Skyrme model is very similar to the one for the baryons in the bosonized
theory in two dimensions. Unlike the latter which is exact in the strong cou-
pling limit, one cannot derive the former starting from the underlying theory.
Another major difference between the two models is of course the existence of
angular momentum only in the four-dimensional case.
r A non-trivial task associated with topological configurations is the construc-
tion of configurations that carry multipole topological charge, for instance a
multi-baryon state both of the bosonized QCD2 as well as of the Skyrme
model, a multi-monopole solution and a multi-instanton solution. For the
two-dimensional baryons (as discussed in Section 13.6) the construction is a
straightforward generalization of the configuration of baryon number one. For
the multi-monopole solutions we presented Nahm’s construction, and for the
multi-instantons the ADHM construction. These constructions, which are in
fact related, are much more complicated than that for the two-dimensional
muti-baryons.
1 Depends on the type of the soliton. See Section 5.3.

r A very important phenomenon that occurs in both two and four dimensions
is the strong-weak duality, and the duality between a soliton and an elemen-
tary field. In two dimensions we have encountered this duality in the relation
between the Thirring model and the sine-Gordon model, where the coupling
of the latter β is related to that of the former g as (6.27),
β2 1
= g . (23.14)
4π 1+ π
This also relates the elementary fermion field of the Thirring model with the
soliton of the sine-Gordon model. In particular for g = 0 corresponding to
β 2 = 4π, the Thirring model describes a free Dirac fermion, while the soliton
of the corresponding sine-Gordon theory is the same fermion in its bosonization
disguise. An analog in four dimensions is the Olive–Montonen duality discussed
in Section 21.8, which relates the electric charge e with the magnetic one
±
eM = 4π e , where the former is carried by the elementary states W and the
latter by the magnetic monopoles.
23.6 Confinement versus screening

Naive intuition tells us that dynamical quarks in the fundamental representa-
tion can screen external sources in the fundamental representation, dynamical
adjoint quarks can screen adjoint sources, but that dynamical adjoint cannot
screen fundamentals. The picture that emerged from our two dimensional calcu-
lations (Chapter 14) showed that this was not the case. We found that massless
adjoint quarks could screen an external source in the fundamental representa-
tion. Moreover we have seen that any massless dynamical field will necessarily
be in the screening phase. The argument for that was that in all cases we have
considered we have found that the string tension is proportional to the mass of
the dynamical quarks,
σ ∼ mg, (23.15)
where m is the mass of the quark and g is the gauge coupling, and hence for the
massless case it vanishes. This was shown in Chapter 14 based on performing a
chiral rotation that enabled us to eliminate the external sources and computing
the string tension as the difference between the Hamiltonian of the system with
the external sources and the one without them namely (14.12),
σ =<H> − <H0 > . (23.16)
It seems as though the situation in two dimensions is very different from that in
four dimensions. From the onset there is a dramatic difference between two and
four dimensions relating to the concept of confining theory. In two dimensions
both the Coulomb abelian potential and the non-abelian one are linear with
the separation distance L, whereas obviously in four dimensions the Coulomb
potential between two particles behaves as 1/L. The confining potential is linear

23.6 Confinement versus screening 415
with L in both two and four dimensions. However, that does not explain the
difference between two and four dimensions, it merely means that in two dimen-
sions the coulomb and confining potentials behave in the same manner.2 The
determination of the string tension in two dimensions cannot be repeated in four
dimensions. The reason is that in the latter case the anomaly is not linear in the
gauge field and thus one cannot use the chiral rotation to eliminate the external
quark anti-quark pair. That does not imply that the situation in four dimensions
differs from the two-dimensional one, it just means that one has to use different
methods to compute the string tension in four dimensions.
What are the four-dimensional systems that might resemble the two-
dimensional case of dynamical adjoint matter and external fundamental quarks?
A system with external quarks in the fundamental representation in the context
of pure YM theory seems a possible analog since the dynamical fields, the gluons,
are in the adjoint representation, though they are vector fields and not fermions.
An alternative is the N = 1 SYM where in addition to the gluons there are also
gluinos which are Majorana fermions in the adjoint representation. Both these
cases should correspond to the massless adjoint case in two dimensions. The lat-
ter admits a screening behavior where as the four-dimensional models seem to be
in the confining phase. This statement is supported by several different types of
calculations in particular for the non supersymmetric case this behavior is found
in lattice simulations.
At this point we cannot provide a satisfactory intuitive explanation why the
behavior in two and four dimensions is so different. There is also no simple
picture of how the massless adjoint dynamical quarks in two dimensions are able
to screen external charges in the fundamental representation.
It is worth mentioning that there is ample evidence that four-dimensional
hadronic physics is well described by a string theory. This is based for instance
on realizing that mesons and baryons in nature admit Regge trajectory behavior
which is an indication of a stringy nature. Any string theory is by definition a
two-dimensional theory and hence a very basic relation between four-dimensional
hadron physics and two-dimensional physics.
In addition to the ordinary string tension which relates to the potential
between a quark and anti-quark in the fundamental representation, one defines
the k string that connects a set of k quarks with a set of k anti-quarks. This
object has been examined in four-dimensional YM as well as four-dimensional
N = 1 SYM. These two cases seem to be the analog of the two-dimensional QCD
theory with adjoint quarks and with external quarks in a representation that is
characterized by k boxes in the Young tableau description. In Chapter 14 we have
derived an expression for the string tension as a function of the representation
of the external and dynamical quarks and in particular for dynamical adjoint
2 Note that the linear potential in two dimensions is already there at lowest order, while
obviously in four dimensions it is a highly non-perturbative effect.

fermions and external quarks in the k representation. If there is any correspon-

dence between the four-dimensional adjoint matter field and the two-dimensional
adjoint quarks it must be with massive adjoint quarks since for the massless case,
as was mentioned above, the two-dimensional string tension vanishes whereas the
four-dimensional one does not. Thus one may consider a correspondence for a
softly broken N = 1 case where the gluinos are massive.
In two dimensions for the pure YM case we found that the string tension
behaves like σ ∼ g 2 kext
2
whereas a Wilson line calculation yields σ ∼ g 2 C2 (R)
where C2 (R) is the second Casimir operator in the R representation of the
external quarks. For the QCD2 case of general k external charges and adjoint
dynamical quarks, one can derive from (14.49) that,
πk
σk2d ∼ sin2 , (23.17)
Nc
whereas in four dimensions it is believed that for general k, the string tension
either follows a Casimir law or a sinusoidal rule as follows,
k(Nc − k) πk
σkcas ∼ σksin ∼ sin . (23.18)
Nc Nc
It is an open problem which of these holds.
As expected all these expressions are invariant under k → N − k which cor-
responds for antisymmetric representations to replacing a quark with an anti-
quark.
23.7 Hadronic phenomenology of two dimensions versus

four dimensions
QCD2 was addressed first in the fermionic formulation. In his seminal work ’t
Hooft deduced the mesonic spectrum in the large NC limit as is described in
Chapter 10. We further presented three additional approaches to the hadronic
spectra in two dimensions, the currentization method for massless quarks for the
entire plane of NC and Nf , the DLCQ approach to extract the mesonic spectrum
for the case of fundamental as well as adjoint quarks and finally the bosonized
formulation in the strong coupling limit to determine the baryonic spectrum. As
for the four-dimensional hadronic spectrum we described the use of the large NC
planar limit and the analysis of the baryonic world using the Skyrme model. It
is worth mentioning again that whereas in the four-dimensional case the Skyrme
approach is only an approximated model derived by an “educational guess”, in
two dimensions the action in the strong coupling regime is exact.
23.7.1 Mesons
As was just mentioned the two-dimensional mesonic spectrum was extracted
using the large NC approximation in the fermionic formulation for Nf = 1

23.7 Hadronic phenomenology 417
(’t Hooft model), the currentization for massless quarks and the DLCQ approach
for both cases of quarks in the fundamental and the adjoint representation. For
the particular region of NC >> Nf and m = 0 the fermionic large Nc and the
currentization treatments yielded identical results. In fact this result is achieved
also using the DLCQ method for adjoint fermions upon a truncation to a single
parton and replacing g 2 with 2g 2 (see (12.42)). For massive fundamental quarks
the DLCQ results match very nicely those of lattice simulations and the large
Nc calculations as can be seen from Figs (12.1) and (12.2).
In all these methods the corresponding equations do not admit exact ana-
lytic solutions for the whole range of parameters and thus one has to resort to
numerical solutions. However, in certain domains one can determine the analytic
behavior of the wavefunctions and masses.
The spectrum of mesons in two dimensions is characterized by the dependence
of the meson masses Mm es on the gauge coupling g, the number of colors Nc ,
the number of flavors Nf , the quark mass mq and the excitation number n. In
four-dimensional QCD the meson spectra depend on the same parameters apart
from the fact that ΛQ C D , the QCD scale, is replacing the two-dimensional gauge
coupling and of course some additional quantum numbers. The following lines
summarize the properties of the spectrum
r The highly excited states n 1, are characterized by,
Mm2 es ∼ πg 2 Nc n. (23.19)
This seems to fit the behaviors of mesons in nature. This behavior is referred

to as a Regge trajectory and it follows easily from a bosonic string model of
the mesons. Following this analogy, the role of the string tension in a two-
dimensional model is played by g 2 Nc . This seems to be in contradiction with
the statement that the string tension is proportional to mq g, as seen in the
discussion of screening versus confinement.
It is very difficult to derive the Regge trajectory behavior from direct cal-
culations in four-dimensional QCD.
r The opposite limit of low-lying states and in particular the ground state can
be deduced in the limit of large quark masses mq g and small quark masses
g mq . For the ground state in the former limit we find,
Mm0 es ∼
= mq 1 + mq 2 , (23.20)
where mq i are the masses of the quark and anti-quark. In the opposite limit of
mq g
5
0 2 ∼ π g 2 Nc
(Mm es ) = (m1 + m2 ). (23.21)
3 π
For the special case of massless quarks we find a massless meson. This is very
reminiscent of the four-dimensional picture for the massless pions. For small

masses this is similar to the pseudo-Goldstone boson relation where,

<ψ̄ψ>
m2π ∼ (m1 + m2 ). (23.22)
fπ2
Note that in two dimensions the massless mesons decouple.
r The ’t Hooft model cannot be used to explore the dependence on Nf the number
of flavors. This can be done from the ’t Hooft-like equations derived in Chapter
11. It was found that for the first massive state there is a linear dependence of
the meson mass squared on Nf
Mm2 es ∼ Nf . (23.23)
We are not aware of a similar behavior of the mesons in four dimensions.

r The ’t Hooft model (Chapter 10) provides the solution of the meson spectrum
in the planar limit in two dimensions. The planar, namely large Nc limit, in
four dimensions is too complicated to be similarly solved. As we have seen in
Chapter 19 one can extract the scaling in Nc dependence of certain hadronic
properties like the mass the size and scattering amplitude but the full determi-
nation of the hadronic spectrum and scattering is still an unresolved mystery.
r Tremendous progress has been made in the understanding of the supersym-
metric theory of N = 4 partly by demonstrating that certain sectors of it can
be described by integrable spin chain models (Section 18.1).
r As was demonstrated in Chapter 12 the DLCQ method has been found very
effective to address the spectrum of mesons of two-dimensional QCD. This
raises the question of whether one can use the DLCQ method to handle the
spectrum of four-dimensional QCD. This task is clearly much more difficult.
On route to the extraction of the hadronic spectrum of QCD4 an easier system
has been analyzed. It is that of the collinear QCD (see Chapter 17) where in the
Hamiltonian of the system one drops off all interaction terms that depend on
the transverse momenta. In this effective two-dimensional setup the transverse
degrees of freedom of the gluon are retained in the form of two scalar fields.
This system which was not described in the book has been solved in [14] where
a complete bound and continuum spectrum was extracted as well as the Fock
space wavefunctions.
23.7.2 Baryons
In Chapter 13 we have described the spectrum of baryons in multiflavor two-
m
dimensional QCD in the strong coupling limit e cq → 0. The four-dimensional
baryonic spectrum was discussed in the large Nc limit in Chapter 19 and using
the Skyrme model approach in Chapter 20. We would like now to compare these
spectra and to investigate the possibility of predicting four-dimensional baryonic
properties from the simpler two-dimensional model. In the former case the mass
is a function of the QCD scale ΛQ C D , the number of colors Nc and the number

23.7 Hadronic phenomenology 419
Table 23.3. Scaling of baryon masses with Nc in two and four

dimensions
two dimensions four dimensions
Classical baryon mass Nc Nc

Quantum correction Nc0 Nc−1
of flavors Nf and in the latter it is a function of ec , Nc and Nf . Thus it seems

that the dimensionful gauge coupling in two dimensions is the analog of ΛQ C D
in four dimensions.
r In two dimensions the mass of the baryon was found to be,
5 ?
√ 3
2Nc π (Nf − 1)
E = 4m +m 2 C2 − Nc2 , (23.24)
π Nc 2Nf
where the classical mass m is given by
√ ΔC
1 + 1Δ
C
ec Nf
m = Nc cmq √ , (23.25)
2π
N 2 −1
with Δc = N c (Nc c +N f ) . Due to the fact that in two dimensions there is no spin,
the structure of the spectrum with respect to the flavor group is obviously
different in two and four dimensions. For instance the lowest allowed state for
Nc = Nf = 3 is in two dimensions the totally symmetric representation 10,
whereas it is the mixed representation 8 in four dimensions.
r Let us discuss now the scaling with Nc in the large Nc limit. The classical
term behaves like Nc , while the quantum correction like 1. This classical result
is in accordance with four dimensions, derived when the large N expansion is
applied to the baryonic system (see Chapter 19), that the baryon mass is linear
in Nc and with the Skyrmion result (see Chapter 20). However, whereas in two
dimensions the quantum correction behaves like Nc0 , namely suppressed by a
factor of N1c compared to the classical term, in four dimensions it behaves like
1 1
N c namely a suppression of N c2 . This is summarized in Table 23.3.
r In terms of the dependence on the number of flavors, it is interesting to note
that both in two dimensions and in four dimensions, the contribution to the
mass due to the quantum fluctuations has a term proportional to the second
Casimir operator associated with the representation of the baryonic state under
the SU (Nf ) flavor group (compare (23.24) with (20.68)).
r Another property of the baryonic spectrum that can be compared between the
two- and four-dimensional cases is the flavor content of the various states. In
Chapter 13 we have computed the ūu, dd ¯ and s̄s content for the Δ+ and Δ++
states. Recall that in the two-dimensional model for Nc = Nf = 3 we do not
have a state in the 8 representation but only in the 10 so strictly speaking there

Table 23.4. Flavor content of two-dimensional and four-dimensional baryons
two dimensions four dimensions

state value state value
ūu Δ+ 1
p 2
3 4 2 5
¯
dd Δ + 1
p 11
3 30
s̄s Δ + 1
6
p 7
30
s̄s Δ+ + 1
6
Δ 7
24
s̄s Ω− 5
24
is no exact analog of the proton. Instead we take the charge = +1Δ+ as the
two-dimensional analog of the proton. In the Skyrme model one can compute
in a similar manner the flavor content of the four-dimensional baryons. The
two- and four-dimensional states compare as is summarized in Table 23.4.
23.8 Outlook
We can imagine future developments associated with the topics covered in the
book in three different directions: Further progress in the application of the
methods discussed in the book to unravel the mysteries of gauge dynamics in
nature; applications of the methods in other domains of physics not related to
four-dimensional gauge theories; and improving our understanding of the strong
interaction and hadron physics due to other non-perturbative techniques that
are not discussed in the book. Let us now briefly fantasize on hypothetical devel-
opments in those three avenues.
23.8.1 Further progress in the application of the methods

discussed in the book
r A lesson that follows from the book is that the exploration of physical systems
on one space dimension is both simpler to handle and sheds light on the real
world so there are plenty of other unresolved questions that could be explored
first in two dimensions. This includes exploration of the full standard model
and the physics beyond the standard model including supersymmetry and its
dynamical breaking, large extra dimensions, compositeness etc.
r There has been tremendous development in recent years in applying meth-
ods of integrable models and in particular of spin chains, like the thermal
Bethe ansatz, to N = 4 SYM theory, namely, in the context of supersymmetric
conformal gauge theory. We have no doubt that there will be further develop-
ment in computing the anomalous dimensions of gauge invariant operators and
correlators.

23.8 Outlook 421
r Moreover, one can identify in a similar manner to N = 4 SYM theory spin

chain structures in gauge theories which are confining and with less or even
no supersymmetries. In that case the spin chain Hamiltonian would not corre-
spond to the dilatation operator but rather be associated with the excitation
energies of hadrons.
r It is plausible that the full role of magnetic monopoles and of instantons has
not yet been revealed. They have already had several reincarnations and there
may be more. For instance there was recently a proposal to describe baryons
as instantons which are solitons of a five-dimensional flavor gauge theory in
curved five dimensions.
23.8.2 Applications to other domains

r A very important application of two-dimensional conformal symmetry has been
to superstring theories. A great part of the developments in superstring theories
is attributed to the infinite-dimensional conformal symmetry algebra. In fact
it went in both directions and certain progress in understanding the structure
of conformal invariance has emerged from the research of string theories. A
similar symbiotic evolution took place with regard to the affine Lie algebras.
r String theories and in particular the string theory on AdS5 × S 5 have recently
been analyzed using the tools of integrable models like mapping to spin chains,
using the Bethe ansatz equations, identifying a set of infinitely many conserved
charges and using structure of Yangian symmetry.
r Spin chain models have been suggested to describe systems of “real” spins in
condensed matter physics. As was discussed in this book the application of
the corresponding tools to field theory systems has been quite fruitful. The
opposite direction will presumably also take place and the use of properties
of integrability that were understood in field theories will shed new light on
certain condensed matter systems.
r The application of conformal invariance to condensed matter systems at crit-
icality has a long history. There has been recently an intensive effort to fur-
ther develop the understanding of systems like various superconductors, the
fractional Hall effect and other systems using modern conformal symmetry
techniques.
23.8.3 Developments in gauge dynamics due to other methods

r An extremely important framework for analyzing gauge theories has been
supersymmetry. Regardless if it is realized in nature or not, it is evident that
there are more tools to handle supersymmetric gauge theories and hence they
are much better understood than non supersymmetric ones. One can gain
novel insight about non supersymmetric theories by introducing supersymme-
try breaking terms to well understood supersymmetric models. For instance

one can start with the Seiberg Witten solution of N = 2 [192] where the struc-
ture of vacua is known and extract confinement behavior in N = 1 and non
supersymmetric theories.
r A breakthrough in the understanding of gauge theories in the strong coupling
regime took place with the discovery by Maldacena of the AdS/CFT holo-
graphic duality [158]. The strongly coupled N = 4 in the large N and large
’t Hooft parameter λ is mapped into a weakly curved supergravity background.
Thousands of research papers that followed develop this map in many different
directions and in particular also in relation to the pure YM theory and QCD
in four dimensions. There is very little doubt that further exploration of the
duality will shed new light on QCD and on hadron physics.
r String theory has been born as a possible theory of hadron physics. It then
underwent a phase transition into a candidate for the theory of quantum grav-
ity and even a unifying theory for everything. In recent years, mainly due to
the AdS/CFT duality there is a renaissance of the idea that hadrons at low
energies should be described as strings. This presumably combined with the
duality seems to be a useful tool that will improve our understanding of gauge
dynamics.
r The computations of scattering amplitudes in gauge theories has been boosted
in recent years due to various developments including the use of techniques
based on twistors, on a novel T-duality in the context of the Ads/CFT duality
and on a conjectured duality between Wilson lines and scattering amplitudes.
One does not need a wild imagination to foresee further progress in the industry
of computing scattering amplitudes.
To summarize, non-perturbative methods have always been very important tools
in exploring the physical world. We have no doubt that they will continue to be a
very essential ingredient in future developments of science in general and physics
in particular.

References
[1] E. Abdalla and M. C. B. Abdalla, “Updating QCD in two-dimensions,” Phys. Rept. 265,
253 (1996) [arXiv:hep-th/9503002].
[2] E. Abdalla, M. C. B. Abdalla and K. D. Rothe, Nonperturbative Methods in Two-
Dimensional Quantum Field Theory, Singapore: World Scientific (2001).
[3] A. Abrashkin, Y. Frishman and J. Sonnenschein, “The spectrum of states with one
current acting on the adjoint vacuum of massless QCD2,” Nucl. Phys. B 703, 320 (2004)
[arXiv:hep-th/0405165].
[4] C. Adam, “Charge screening and confinement in the massive Schwinger model,” Phys.
Lett. B 394, 161 (1997) [arXiv:hep-th/9609155].
[5] G. S. Adkins, C. R. Nappi and E. Witten, “Static properties of nucleons in the Skyrme
model,” Nucl. Phys. B 228, 552 (1983).
[6] S. L. Adler, J. C. Collins and A. Duncan, “Energy momentum tensor trace anomaly in
spin 1/2 quantum electrodynamics’” Phys. Rev. D 15, 1712 (1977).
[7] I. Affleck, “On the realization of chiral symmetry in (1+1)-dimensions,” Nucl. Phys. B
265, 448 (1986).
[8] O. Aharony, O. Ganor, J. Sonnenschein and S. Yankielowicz, “On the twisted G/H
topological models,” Nucl. Phys. B 399, 560 (1993) [arXiv:hep-th/9208040].
[9] O. Aharony, O. Ganor, J. Sonnenschein, S. Yankielowicz and N. Sochen, “Physical states
in G/G models and 2-d gravity,” Nucl. Phys. B 399, 527 (1993) [arXiv:hep-th/9204095].
[10] O. Aharony, S. S. Gubser, J. M. Maldacena, H. Ooguri and Y. Oz, “Large N field theories,
string theory and gravity,” Phys. Rept. 323, 183 (2000) [arXiv:hep-th/9905111].
[11] A. Y. Alekseev and V. Schomerus, “D-branes in the WZW model,” Phys. Rev. D 60,
061901 (1999) [arXiv:hep-th/9812193].
[12] D. Altschuler, K. Bardakci and E. Rabinovici, “A construction of the c < 1 modular
invariant partition function,” Commun. Math. Phys. 118, 241 (1988).
[13] L. Alvarez-Gaume, G. Sierra and C. Gomez, “Topics in conformal field theory,” contri-
bution to the Knizhnik Memorial Volume, L. Brink, et al., World Scientific. In Brink, L.
(ed.) et al.: Physics and Mathematics of Strings 16-184 (1989). Singapore.
[14] F. Antonuccio and S. Dalley, “Glueballs from (1+1)-dimensional gauge theories with
transverse degrees of freedom,” Nucl. Phys. B 461, 275 (1996) [arXiv:hep-ph/9506456].
[15] A. Armoni, Y. Frishman and J. Sonnenschein, “The string tension in massive QCD(2),”
Phys. Rev. Lett. 80, 430 (1998) [arXiv:hep-th/9709097].
[16] A. Armoni, Y. Frishman and J. Sonnenschein, “The string tension in two dimensional
gauge theories,” Int. J. Mod. Phys. A 14, 2475 (1999) [arXiv:hep-th/9903153].
[17] A. Armoni, Y. Frishman and J. Sonnenschein, “Massless QCD(2) from current con-
stituents,” Nucl. Phys. B 596, 459 (2001) [arXiv:hep-th/0011043].
[18] A. Armoni and J. Sonnenschein, “Mesonic spectra of bosonized QCD in two-dimensions
models,” Nucl. Phys. B 457, 81 (1995) [arXiv:hep-th/9508006].
[19] M. F. Atiyah and N. J. Hitchin, “The geometry and dynamics of magnetic monopole.
M.B. Porter lectures” Princeton, USA: Princeton University Press (1988) 133p.
[20] M. F. Atiyah, N. J. Hitchin, V. G. Drinfeld and Yu. I. Manin, “Construction of instan-
tons,” Phys. Lett. A 65, 185 (1978).
[21] F. A. Bais, “To be or not to be? Magnetic monopoles in non-abelian gauge theories,”
in ’t Hooft, Hackensack, G. ed. Fifty years of Yang–Mills Theory, New Jersey World
Scientific, C 2005.

424 References
[22] A. P. Balachandran, “Solitons in nuclear and elementary physics. Proceedings of the

Lewes workshop,” World Scientific, 1984.
[23] I. I. Balitsky and L. N. Lipatov, “The Pomeranchuk singularity in quantum chromody-
namics,” Sov. J. Nucl. Phys. 28, 822 (1978) [Yad. Fiz. 28, 1597 (1978)].
[24] V. Baluni, “The Bose form of two-dimensional quantum chromodynamics,” Phys. Lett.
B 90, 407 (1980).
[25] T. Banks, “Lectures on conformal field theory,” presented at the Theoretical Advanced
Studies Institute, St. John’s College, Santa Fe, N. Mex., Jul 5 – Aug 1, 1987. Published
in Santa Fe: TASI 87:572.
[26] T. I. Banks and C. M. Bender, “Anharmonic oscillator with polynomial self-interaction,”
J. Math. Phys. 13, 1320 (1972).
[27] K. Bardakci and M. B. Halpern, “New dual quark models,” Phys. Rev. D 3, 2493 (1971).
[28] A. Bassetto, G. Nardelli and R. Soldati, Yang-Mills Theories in Algebraic Noncovari-
ant Gauges: Canonical Quantization and Renormalization, Singapore World Scientific
(1991).
[29] R. J. Baxter, Exactly Solved Models in Statistical Mechanics, London: Academic press
(1989).
[30] K. Becker, M. Becker and J. H. Schwarz, String Theory and M-Theory: A modern intro-
duction, Cambridge, UK: Cambridge University Press (2007).
[31] N. Beisert, “The dilatation operator of N = 4 super Yang-Mills theory” Phys. Rept. 405,
1 (2005) [arXiv:hep-th/0407277].
[32] A. A. Belavin, A. M. Polyakov, A. S. Shvarts and Yu. S. Tyupkin, “Pseudoparticle
solutions of the Yang-Mills equations,” Phys. Lett. B 59, 85 (1975).
[33] A. A. Belavin, A. M. Polyakov and A. B. Zamolodchikov, “Infinite conformal symmetry
in two-dimensional quantum field theory,” Nucl. Phys. B 241, 333 (1984).
[34] A. V. Belitsky, V. M. Braun, A. S. Gorsky and G. P. Korchemsky, “Integrability in QCD
and beyond,” Int. J. Mod. Phys. A 19, 4715 (2004) [arXiv:hep-th/0407232].
[35] D. E. Berenstein, J. M. Maldacena and H. S. Nastase, “Strings in flat space and pp waves
from N = 4 super Yang Mills,” JHEP 0204, 013 (2002) [arXiv:hep-th/0202021].
[36] M. Bernstein and J. Sonnenschein, “A comment on the quantization of the chiral bosons”
Phys. Rev. Lett. 60, 1772 (1988).
[37] J. D. Bjorken and S. D. Drell, Relativistic Quantum Field Theory, New York: McGraw-
Hill (1965), ISBN 0-07-005494-0.
[38] G. Bhanot, K. Demeterfi and I. R. Klebanov, “(1+1)-dimensional large N QCD coupled
to adjoint fermions,” Phys. Rev. D 48, 4980 (1993) [arXiv:hep-th/9307111].
[39] D. Bernard and A. Leclair, “Quantum group symmetries and nonlocal currents in 2-D
QFT,” Commun. Math. Phys. 142, 99 (1991).
[40] K. M. Bitar and S. J. Chang, “Vacuum tunneling of gauge theory in minkowski space,”
Phys. Rev. D 17, 486 (1978).
[41] E. B. Bogomolny, “Stability of classical solutions,” Sov. J. Nucl. Phys. 24, 449 (1976)
[Yad. Fiz. 24, 861 (1976)].
[42] V. M. Braun, S. E. Derkachov, G. P. Korchemsky and A. N. Manashov, “Baryon distri-
bution amplitudes in QCD,” Nucl. Phys. B 553, 355 (1999) [arXiv:hep-ph/9902375].
[43] V. M. Braun, G. P. Korchemsky and D. Mueller, “The uses of conformal symmetry in
QCD,” Prog. Part. Nucl. Phys. 51, 311 (2003) [arXiv:hep-ph/0306057].
[44] E. Brezin and J. L. Gervais, “Nonperturbative aspects in quantum field theory. Proceed-
ings of Les Houches Winter Advanced Study Institute, March 1978,” Phys. Rept. 49, 91
(1979).
[45] E. Brezin, C. Itzykson, J. Zinn-Justin and J. B. Zuber, “Remarks about the existence of
nonlocal charges in two-dimensional models,” Phys. Lett. B 82, 442 (1979).
[46] E. Brezin and S. R. Wadia, “The large N expansion in quantum field theory and statistical
physics: From spin systems to two-dimensional gravity,” Singapore, Singapore: World
Scientific (1993) 1130 p
[47] S. J. Brodsky, “Gauge theories on the light-front,” Braz. J. Phys. 34, 157 (2004)

References 425
[48] S. J. Brodsky, H. C. Pauli and S. S. Pinsky, “Quantum chromodynamics and other field
theories on the light cone,” Phys. Rept. 301, 299 (1998) [arXiv:hep-ph/9705477].
[49] S. J. Brodsky, Y. Frishman, G. P. Lepage and C. T. Sachrajda, “Hadronic wave functions
at short distances and the operator product expansion,” Phys. Lett. B 91, 239 (1980).
[50] S. J. Brodsky, Y. Frishman and G. P. Lepage, “On the application of conformal symmetry
to quantum field theory,” Phys. Lett. B 167, 347 (1986).
[51] S. J. Brodsky, P. Damgaard, Y. Frishman and G. P. Lepage, “Conformal symmetry:
exclusive processes beyond leading order,” Phys. Rev. D 33, 1881 (1986).
[52] R. C. Brower, W. L. Spence and J. H. Weis, “Effects of confinement on analyticity in
two-dimensional QCD,” Phys. Rev. D 18, 499 (1978).
[53] G. E. Brown, ‘Selected papers, with commentary, of Tony Hilton Royle Skyrme,” Sin-
gapore: World Scientific (1994) 438 p. (World Scientific series in 20th century physics,
3).
[54] R. N. Cahn, Semisimple Lie Algebras and their Representations, Menlo Park, USA:
Benjamin/Cummings (1984) 158 p. (Frontiers In Physics, 59)
[55] C. G. Callan, “Broken scale invariance in scalar field theory,” Phys. Rev. D 2, 1541
(1970).
[56] C. G. Callan, N. Coote and D. J. Gross, “Two-dimensional Yang-Mills theory: a model
of quark confinement,” Phys. Rev. D 13, 1649 (1976).
[57] C. G. Callan, R. F. Dashen and D. J. Gross, “Toward a theory of the strong interactions,”
Phys. Rev. D 17, 2717 (1978).
[58] J. L. Cardy, “Boundary conditions, fusion rules and the Verlinde formula,” Nucl. Phys.
B 324, 581 (1989).
[59] J. L. Cardy, “Conformal invariance and statistical mechanics,” Les Houches Summer
School 1988:0169-246.
[60] A. Casher, H. Neuberger and S. Nussinov, “Chromoelectric flux tube model of particle
production,” Phys. Rev. D 20, 179 (1979).
[61] A. Chodos, “Simple connection between conservation laws in the Korteweg-De Vries and
sine-Gordon systems,” Phys. Rev. D 21, 2818 (1980).
[62] E. Cohen, Y. Frishman and D. Gepner, “Bosonization of two-dimensional QCD with
flavor,” Phys. Lett. B 121, 180 (1983).
[63] S. R. Coleman, “Quantum sine-Gordon equation as the massive Thirring model,” Phys.
Rev. D 11, 2088 (1975).
[64] S. R. Coleman, “More about the massive Schwinger model,” Annals Phys. 101, 239
(1976).
[65] S. R. Coleman, “The uses of instantons,” Subnucl. Ser. 15, 805 (1979).
[66] S. Coleman, Aspects of Symmetry, selected Erice lectures of Sidney Coleman, Cambridge,
UK, Cambridge Univ. Press (1985).
[67] S. R. Coleman, The magnetic monopole fifty years later, In the Unity of Fundamental
Interactions, ed. A. Zichichi: New York, Plenum (1983).
[68] S. R. Coleman, R. Jackiw and L. Susskind, “Charge shielding and quark confinement in
the massive Schwinger model,” Annals Phys. 93, 267 (1975).
[69] S. Cordes, G. W. Moore and S. Ramgoolam, “Large N 2-D Yang-Mills theory and topo-
logical string theory,” Commun. Math. Phys. 185, 543 (1997) [arXiv:hep-th/9402107].
[70] E. Corrigan, D. B. Fairlie, S. Templeton and P. Goddard, “A Green’s function for the
general selfdual gauge field,” Nucl. Phys. B 140, 31 (1978).
[71] E. Corrigan and P. Goddard, “Construction of instanton and monopole solutions and
reciprocity,” Annals Phys. 154, 253 (1984).
[72] S. Dalley and I. R. Klebanov, “String spectrum of (1+1)-dimensional large N QCD with
adjoint matter,” Phys. Rev. D 47, 2517 (1993) [arXiv:hep-th/9209049].
[73] R. F. Dashen and Y. Frishman, “Four fermion interactions and scale invariance,” Phys.
Rev. D 11, 2781 (1975).
[74] R. F. Dashen, B. Hasslacher and A. Neveu, “Nonperturbative methods and extended
hadron models in field theory. 2. Two-dimensional models and extended hadrons,” Phys.
Rev. D 10, 4130 (1974).

426 References
[75] G. Date, Y. Frishman and J. Sonnenschein, “The spectrum of multiflavor QCD in two-
dimensions’” Nucl. Phys. B 283, 365 (1987).
[76] G. F. Dell-Antonio, Y. Frishman and D. Zwanziger, “Thirring model in terms of currents:
solution and light cone expansions,” Phys. Rev. D 6, 988 (1972).
[77] P. Di Francesco, P. Mathieu, D. Senechal Conformal Field Theory, Series:Graduate Texts
in Contemporary Physics. New York: Springer-Verlag (1997).
[78] P. A. M. Dirac, “Theory Of Magnetic Monopoles,” In *Coral Gables 1976, Proceedings,
New Pathways In High-energy Physics, Vol. I*, New York: Plenum Press, 1976, 1-14.
[79] F. M. Dittes and A. V. Radyushkin, “Two loop contribution to the evolution of the pion
wave function,” Phys. Lett. B 134, 359 (1984); M. H. Sarmadi, Phys. Lett. B 143, 471
(1984); S. V. Mikhailov and A. V. Radyushkin, Nucl. Phys. B 254, 89 (1985); G. R.
Katz, Phys. Rev. D 31, 652 (1985).
[80] N. Dorey, T. J. Hollowood, V. V. Khoze, M. P. Mattis and S. Vandoren, “Multi-instanton
calculus and the AdS/CFT correspondence in N = 4 Nucl. Phys. B 552, 88 (1999)
[81] N. Dorey, T. J. Hollowood, V. V. Khoze and M. P. Mattis, “The calculus of many
instantons,” Phys. Rept. 371, 231 (2002) [arXiv:hep-th/0206063].
[82] M. R. Douglas, K. Li and M. Staudacher, “Generalized two-dimensional QCD,” Nucl.
Phys. B 420, 118 (1994) [arXiv:hep-th/9401062].
[83] A. V. Efremov and A. V. Radyushkin, “Factorization and asymptotical behavior of pion
form-factor in QCD,” Phys. Lett. B 94, 245 (1980).
[84] T. Eguchi and H. Ooguri, “Chiral bosonization on a Riemann surface,” Phys. Lett. B
187, 127 (1987).
[85] S. Elitzur and G. Sarkissian, “D-branes on a gauged WZW model,” Nucl. Phys. B 625,
166 (2002) [arXiv:hep-th/0108142].
[86] J. R. Ellis, Y. Frishman, A. Hanany and M. Karliner, “Quark solitons as constituents of
hadrons,” Nucl. Phys. B 382, 189 (1992) [arXiv:hep-ph/9204212].
[87] J. R. Ellis, Y. Frishman and M. Karliner, “Meson baryon scattering in QCD(2) for any
coupling,” Phys. Lett. B 566, (2003) 201 [arXiv:hep-ph/0305292].
[88] L. D. Faddeev, “How algebraic Bethe ansatz works for integrable model,” arXiv:hep-
th/9605187.
[89] B. L. Feigin and D. B. Fuchs “Skew symmetric differential operators on the line and
Verma modules over the Virasoro algebra” Funct. Anal. Prilozhen 16, (1982) 47.
[90] S. Ferrara, A. F. Grillo and R. Gatto, “Improved light cone expansion,” Phys. Lett. B
36, 124 (1971) [Phys. Lett. B 38, 188 (1972)].
[91] D. Finkelstein and J. Rubinstein, “Connection between spin, statistics, and kinks,” J.
Math. Phys. 9, 1762 (1968).
[92] R. Floreanini and R. Jackiw, “Selfdual fields as charge density solitons,” Phys. Rev. Lett.
59, 1873 (1987).
[93] D. Friedan, “Introduction To Polyakov’s string theory,” Published in Les Houches Sum-
mer School 1982:0839.
[94] D. Friedan, E. J. Martinec and S. H. Shenker, “Conformal invariance, supersymmetry
and string theory,” Nucl. Phys. B 271, 93 (1986).
[95] D. Friedan, Z. a. Qiu and S. H. Shenker, “Conformal invariance, unitarity and two-
dimensional critical exponents,” Phys. Rev. Lett. 52, 1575 (1984).
[96] Y. Frishman, A. Hanany and J. Sonnenschein, “Subtleties in QCD theory in two-
dimensions,” Nucl. Phys. B 429, 75 (1994) [arXiv:hep-th/9401046].
[97] Y. Frishman and M. Karliner, “Baryon wave functions and strangeness content in QCD
in two-dimensions,” Nucl. Phys. B 344, 393 (1990).
[98] Y. Frishman and M. Karliner, “Scattering and resonances in QCD(2),” Phys. Lett. B
541, 273 (2002). Erratum-ibid. B 562, 367, (2003). [arXiv:hep-ph/0206001].
[99] Y. Frishman and J. Sonnenschein, ”Bosonization of colored-flavored fermions and QCD
in two-dimenstions,” Nucl. Phys. B 294, 801 (1987).
[100] Y. Frishman and J. Sonnenschein, “Gauging of chiral bosonized actions’” Nucl. Phys. B
301, 346 (1988).

References 427
[101] Y. Frishman and J. Sonnenschein, “Bosonization and QCD in two-dimensions,” Phys.

Rept. 223, 309 (1993) [arXiv:hep-th/9207017].
[102] Y. Frishman and W. J. Zakrzewski, “Multibaryons in QCD in two-dimensions,” Nucl.
Phys. B 328, 375 (1989).
[103] Y. Frishman and W. J. Zakrzewski, “Explicit expressions for masses and bindings of
multibaryons in QCD(2),” Nucl. Phys. B 331, 781 (1990).
[104] S. Fubini, A. J. Hanson and R. Jackiw, “New approach to field theory,” Phys. Rev. D
7, 1732 (1973).
[105] O. Ganor, J. Sonnenschein and S. Yankielowicz, “The string theory approach to gener-
alized 2-d Yang-Mills theory,” Nucl. Phys. B 434, 139 (1995) [arXiv:hep-th/9407114].
[106] E. G. Gimon, L. A. Pando Zayas, J. Sonnenschein and M. J. Strassler, “A soluble string
theory of hadrons,” JHEP 0305, 039 (2003) [arXiv:hep-th/0212061].
[107] D. Gepner, “Nonabelian bosonization and multiflavor QED and QCD in two-
dimensions,” Nucl. Phys. B 252, 481 (1985).
[108] D. Gepner and E. Witten, “String theory on group manifolds,” Nucl. Phys. B 278, 493
(1986).
[109] P. H. Ginsparg, ”Applied conformal field theory,” Published in Les Houches Summer
School 1988:1-168, arXiv:hep-th/9108028.
[110] P. Goddard, A. Kent and D. I. Olive, “Virasoro algebras and coset space models,” Phys.
Lett. B 152, 88 (1985).
[111] P. Goddard and D. I. Olive, “Kac-Moody and Virasoro algebras in relation to quantum
physics,” Int. J. Mod. Phys. A 1, 303 (1986).
[112] D. Gonzales and A. N. Redlich, “The low-energy effective dynamics of two-dimensional
gauge theories,” Nucl. Phys. B 256, 621 (1985).
[113] M. B. Green, J. H. Schwarz and E. Witten, “Superstring theory. Vol. 2: Loop amplitudes,
anomalies and phenomenology,” Cambridge, UK: Univ. Pr. (1987) 596 P. (Cambridge
Monographs On Mathematical Physics)
[114] C. Gomez, G. Sierra and M. Ruiz-Altaba, “Quantum groups in two-dimensional physics,”
Cambridge, UK: Univ. Pr. (1996) 457 p
[115] D. J. Gross, “Two-dimensional QCD as a string theory,” Nucl. Phys. B 400, 161 (1993)
[116] D. J. Gross, I. R. Klebanov, A. V. Matytsin and A. V. Smilga, “Screening vs. Confinement
in 1+1 Dimensions,” Nucl. Phys. B 461, 109 (1996) [arXiv:hep-th/9511104].
[117] D. J. Gross and A. Neveu, “Dynamical symmetry breaking in asymptotically free field
theories,” Phys. Rev. D 10, 3235 (1974).
[118] D. J. Gross and W. Taylor, “Two-dimensional QCD is a string theory,” Nucl. Phys. B
400, 181 (1993) [arXiv:hep-th/9301068].
[119] D. J. Gross and W. Taylor, “Twists and Wilson loops in the string theory of two-
dimensional QCD,” Nucl. Phys. B 403, 395 (1993) [arXiv:hep-th/9303046].
[120] I. G. Halliday, E. Rabinovici, A. Schwimmer and M. S. Chanowitz, “Quantization of
anomalous two-dimensional models,” Nucl. Phys. B 268, 413 (1986).
[121] M. B. Halpern, “Quantum solitons which are SU(N) fermions,” Phys. Rev. D 12, 1684
(1975).
[122] G. ’t Hooft, “A planar diagram theory for strong interactions,” Nucl. Phys. B 72, 461
(1974).
[123] G. ’t Hooft, “Magnetic monopoles in unified gauge theories” Nucl. Phys. B 79, 276
(1974).
[124] G. ’t Hooft, “A two-dimensional model for mesons,” Nucl. Phys. B 75, 461 (1974).
[125] G. ’t Hooft, “Computation of the quantum effects due to a four-dimensional pseudopar-
ticle,” Phys. Rev. D 14, 3432 (1976) [Erratum-ibid. D 18, 2199 (1978)].
[126] P. Horava, “Topological strings and QCD in two-dimensions,” hep-th/9311156, talk given
at NATO Advanced Research Workshop on New Developments in String Theory, Con-
formal Models and Topological Field Theory, Cargese, France, 12–21 May 1993.
[127] K. Hornbostel, “The application of light cone quantization to quantum chromodynamics
in (1+1)-dimensions,” Ph.D. thesis, SLAC-R-333, Dec 1988.

428 References
[128] K. Hornbostel, S. J. Brodsky and H. C. Pauli, “Light cone quantized QCD in (1+1)-
dimensions,” Phys. Rev. D 41, 3814 (1990).
[129] C. Imbimbo and A. Schwimmer, “The Lagrangian formulation of chiral scalars,” Phys.
Lett. B 193, 455 (1987).
[130] C. Itzykson and J. B. Zuber, Quantum Field Theory, New York, USA: McGraw-Hill
(1980) 705 P.(International Series In Pure and Applied Physics)
[131] R. Jackiw and C. Rebbi, “Vacuum periodicity in a Yang-Mills quantum theory,” Phys.
Rev. Lett. 37, 172 (1976).
[132] R. Jackiw, C. Nohl and C. Rebbi, “Conformal properties of pseudoparticle configura-
tions”, Phys. Rev. D 15, 1642 (1977).
[133] A. D. Jackson and M. Rho, “Baryons as chiral solitons,” Phys. Rev. Lett. 51, 751 (1983).
[134] B. Julia and A. Zee, “Poles with both magnetic and electric charges in nonabelian gauge
theory,” Phys. Rev. D 11, 2227 (1975).
[135] N. Ishibashi, “The boundary and crosscap states in conformal field theories,” Mod. Phys.
Lett. A 4, 251 (1989).
[136] V. G. Kac, “Simple graded algebras of finite growth,” Funct. Anal. Appl. 1, 328 (1967).
[137] L. P. Kadanoff, “Correlators along the line of two dimensional Ising model,” Phys. Rev.
188, 859 (1969).
[138] M. Kaku, Strings, Conformal Fields, and M-Theory, New York, USA: Springer (2000)
531 p.
[139] V. A. Kazakov, “Wilson loop average for an arbitrary contour in two-dimensional U(N)
gauge theory,” Nucl. Phys. B 179, 283 (1981).
[140] S. V. Ketov, Conformal Field Theory, Singapore, Singapore: World Scientific (1995) 486
p.
[141] V. V. Khoze, M. P. Mattis and M. J. Slater, “The instanton hunter’s guide to supersym-
metric SU(N) gauge theory,” Nucl. Phys. B 536, 69 (1998) [arXiv:hep-th/9804009].
[142] E. Kiritsis, String Theory in a Nutshell, Princeton, USA: Univ. Pr. (2007) 588 p.
[143] V. G. Knizhnik and A. B. Zamolodchikov, “Current algebra and Wess-Zumino model in
two dimensions,” Nucl. Phys. B 247, 83 (1984).
[144] W. Krauth and M. Staudacher, “Non-integrability of two-dimensional QCD,” Phys. Lett.
B 388, 808 (1996) [arXiv:hep-th/9608122].
[145] E. A. Kuraev, L. N. Lipatov and V. S. Fadin, “The Pomeranchuk singularity in non-
abelian gauge theories,” Sov. Phys. JETP 45, 199 (1977) [Zh. Eksp. Teor. Fiz. 72, 377
(1977)]. “Multi-reggeon processes in the Yang-Mills theory,” Sov. Phys. JETP 44, 443
(1976) [Zh. Eksp. Teor. Fiz. 71, 840 (1976)].
[146] D. Kutasov, “Duality off the critical point in two-dimensional systems with nonabelian
symmetries,” Phys. Lett. B 233, 369 (1989).
[147] D. Kutasov, “Two-dimensional QCD coupled to adjoint matter and string theory,” Nucl.
[148] D. Kutasov and A. Schwimmer, “Universality in two-dimensional gauge theory,” Nucl.
[149] T. D. Lee and Y. Pang, “Nontopological solitons,” Phys. Rept. 221, 251 (1992).
[150] G. P. Lepage and S. J. Brodsky, “Exclusive processes in quantum chromodynamics:
evolution equations for hadronic wave functions and the form-factors of mesons,” Phys.
Lett. B 87, 359 (1979).
[151] L. N. Lipatov, The Creation of Quantum Chromodynamics and the Effective Energy,
Bologna, Italy: Univ. Bologna (1998) 367 p.
[152] J. H. Lowenstein and J. A. Swieca, “Quantum electrodynamics in two-dimensions,”
Annals Phys. 68, 172 (1971).
[153] M. Luscher, “Quantum nonlocal charges and absence of particle production in the two-
dimensional nonlinear Sigma model,” Nucl. Phys. B 135, 1 (1978).
[154] D. Lust and S. Theisen, “Lectures on string theory,” Lect. Notes Phys. 346, 1 (1989).
[155] A. Maciocia, “Metrics on the moduli spaces of instantons over Euclidean four space,”
Commun. Math. Phys. 135, 467 (1991).
[156] G. Mack and A. Salam, “Finite component field representations of the conformal group,”
Annals Phys. 53, 174 (1969).

References 429
[157] V. G. Makhankov, Y. P. Rybakov and V. I. Sanyuk, The Skyrme model: Fundamentals,

methods, applications, Berlin, Germany: Springer (1993) 265 p. (Springer series in nuclear
and particle physics).
[158] J. M. Maldacena, “The large N limit of superconformal field theories and supergravity,”
Adv. Theor. Math. Phys. 2, 231 (1998) [Int. J. Theor. Phys. 38, 1113 (1999)] [arXiv:hep-
th/9711200].
[159] S. Mandelstam, “Soliton operators for the quantized sine-Gordon equation,” Phys. Rev.
D 11, 3026 (1975).
[160] A. V. Manohar, “Large N QCD,” arXiv:hep-ph/9802419, Published in *Les Houches
1997, Probing the standard model of particle interactions, Pt. 2* 1091-1169.
[161] A. A. Migdal, “Recursion equations in gauge field theories,” Sov. Phys. JETP 42, 413
(1975) [Zh. Eksp. Teor. Fiz. 69, 810 (1975)].
[162] J. A. Minahan and K. Zarembo, “The Bethe-ansatz for N = 4 super Yang-Mills,” JHEP
0303, 013 (2003) [arXiv:hep-th/0212208].
[163] C. Montonen and D. I. Olive, “Magnetic monopoles as gauge particles?,” Phys. Lett. B
72, 117 (1977).
[164] R. V. Moody, “Lie algebras associated with generalized Cartan matrices,” Bull. Am.
Math. Soc. 73, 217 (1967).
[165] M. Moshe and J. Zinn-Justin, “Quantum field theory in the large N limit: A review,”
Phys. Rept. 385, 69 (2003) [arXiv:hep-th/0306133].
[166] D. Mueller, “Constraints for anomalous dimensions of local light cone operators in phi**3
in six-dimensions theory,” Z. Phys. C 49, 293 (1991); Phys. Rev. D 49, 2525 (1994).
[167] W. Nahm, “A simple formalism for the BPS monopole,” Phys. Lett. B 90, 413 (1980).
[168] N. K. Nielsen, “Gauge invariance and broken conformal symmetry,” Nucl. Phys. B 97,
527 (1975).
[169] N. K. Nielsen, “The energy momentum tensor in a nonabelian quark gluon theory,” Nucl.
Phys. B 120, 212 (1977).
[170] S. P. Novikov, “Multivalued functions and functionals, An analogue to Morse theory,”
Sov. Math. Dock. 24, 222 (1981)
[171] T. Ohrndorf, “Constraints from conformal covariance on the mixing of operators of lowest
twist,” Nucl. Phys. B 198, 26 (1982).
[172] R. D. Peccei and H. R. Quinn, “Constraints imposed by CP conservation in the presence
of instantons,” Phys. Rev. D 16, 1791 (1977).
[173] M. E. Peskin and D. V. Schroeder, An Introduction To Quantum Field Theory, Reading,
USA: Addison-Wesley (1995) 842 p.
[174] J. Polchinski, String Theory. Vol. 2: Superstring Theory and Beyond, Cambridge, UK:
Univ. Pr. (1998) 531 p.
[175] A. M. Polyakov, “Particle spectrum in quantum field theory,” JETP Lett. 20, 194 (1974)
[Pisma Zh. Eksp. Teor. Fiz. 20, 430 (1974)].
[176] A. M. Polyakov, “Hidden symmetry of the two-dimensional chiral fields,” Phys. Lett. B
72, 224 (1977).
[177] A. M. Polyakov, “Quantum geometry of bosonic strings,” Phys. Lett. B 103, 207
(1981).
[178] A. M. Polyakov, Gauge Fields and Strings, Chur, Switzerland: Harwood (1987) 301 p.
(Contemporary concepts in Physics, 3).
[179] A. M. Polyakov and P. B. Wiegmann, “Goldstone fields in two-dimensions with multi-
valued actions,” Phys. Lett. B 141, 223 (1984).
[180] M. K. Prasad and C. M. Sommerfield, “An exact classical solution for the ’t Hooft
monopole and the Julia-Zee dyon,” Phys. Rev. Lett. 35, 760 (1975).
[181] E. Rabinovici, A. Schwimmer and S. Yankielowicz, “Quantization in the presence of
Wess-Zumino terms,” Nucl. Phys. B 248, 523 (1984).
[182] R. Rajaraman, An Introduction to Solitons and Instantons in Quantum Field Theory,
Amsterdam, Netherlands: North-holland (1982) 409p.
[183] C. Rebbi and G. Soliani, Solitons and particles, World Scientific 1984.
[184] B. E. Rusakov, “Loop averages and partition functions in U(N) gauge theory on two-
dimensional manifolds,” Mod. Phys. Lett. A 5, 693 (1990).

430 References
[185] T. Schafer and E. V. Shuryak, “Instantons in QCD,” Rev. Mod. Phys. 70, 323 (1998)
[arXiv:hep-ph/9610451].
[186] J. Schechter and H. Weigel, “The Skyrme model for baryons,” arXiv:hep-ph/9907554.
[187] T. D. Schultz, D. C. Mattis and E. H. Lieb, “Two-dimensional Ising model as a soluble
problem of many fermions,” Rev. Mod. Phys. 36, 856 (1964).
[188] M. A. Shifman, Instantons in Gauge Theories, Singapore, Singapore: World Scientific
(1994) 488 p.
[189] E. V. Shuryak, “The role of instantons in quantum chromodynamics. 1. Physical vac-
uum,” Nucl. Phys. B 203, 93 (1982).
[190] J. S. Schwinger, “Gauge invariance and mass. 2,” Phys. Rev. 128, 2425 (1962).
[191] N. Seiberg and E. Witten, “Monopoles, duality and chiral symmetry breaking in N=2
supersymmetric QCD,” Nucl. Phys. B 431, 484 (1994) [arXiv:hep-th/9408099].
[192] N. Seiberg and E. Witten, “Monopole condensation, and confinement in N=2 super-
symmetric Yang-Mills theory,” Nucl. Phys. B 426, 19 (1994) [Erratum-ibid. B 430, 485
(1994)] [arXiv:hep-th/9407087].
[193] Y. M. Shnir, Magnetic Monopoles, Berlin, Germany: Springer (2005) 532 p.
[194] W. Siegel, “Manifest Lorentz invariance sometimes requires nonlinearity,” Nucl. Phys.
B 238, 307 (1984).
[195] T. H. R. Skyrme, “A Nonlinear theory of strong interactions,” Proc. Roy. Soc. Lond. A
247, 260 (1958).
[196] T. H. R. Skyrme, “Particle states of a quantized meson field,” Proc. Roy. Soc. Lond. A
262, 237 (1961).
[197] T. H. R. Skyrme, “A unified field theory of mesons and baryons,” Nucl. Phys. 31, 556
(1962).
[198] Smirnov, F. A. “Form factors in completely integrable models of quantum field theory,”
Adv. Ser. Math. Phys. 14:1–208 (1992).
[199] J. Sonnenschein, “Chiral bosons” Nucl. Phys. B 309, 752 (1988).
[200] M. Spiegelglas and S. Yankielowicz, “G/G topological field theories by cosetting G(K),”
Nucl. Phys. B 393, 301 (1993) [arXiv:hep-th/9201036].
[201] P. J. Steinhardt, “Baryons and baryonium in QCD in two-dimensions,” Nucl. Phys. B
176, 100 (1980).
[202] M. Stone, Bosonization, Singapore: World Scientific (1994) 539 p.
[203] H. Sugawara, “A Field theory of currents,” Phys. Rev. 170, 1659 (1968).
[204] K. Symanzik, “Small distance behavior in field theory and power counting,” Commun.
Math. Phys. 18, 227 (1970).
[205] W. E. Thirring, “A soluble relativistic field theory,” Annals Phys. 3, 91 (1958).
[206] C. B. Thorn, “Computing the Kac determinant using dual model techniques and more
about the no-ghost theorem,” Nucl. Phys. B 248, 551 (1984).
[207] S. B Treiman, R. Jackiw and D. Gross Current Algebra and its Applications (Princeton,
University Press, New Jersey, 1972).
[208] A. I. Vainshtein, V. I. Zakharov, V. A. Novikov and M. A. Shifman, “ABC of instantons,”
Sov. Phys. Usp. 25, 195 (1982) [Usp. Fiz. Nauk 136, 553 (1982)].
[209] S. Vandoren and P. van Nieuwenhuizen, “Lectures on instantons,” arXiv:0802.1862
[hep-th].
[210] E. P. Verlinde, “Fusion rules and modular transformations in 2d conformal field theory,”
Nucl. Phys. B 300, 360 (1988).
[211] E. P. Verlinde and H. L. Verlinde, “Chiral bosonization, determinants and the string
partition function,” Nucl. Phys. B 288, 357 (1987).
[212] M. S. Virasoro, “Subsidiary conditions and ghosts in dual resonance models,” Phys. Rev.
D 1, 2933 (1970).
[213] M. Wakimoto, “Fock representations of the affine Lie algebra A1(1),” Commun. Math.
Phys. 104, 605 (1986).
[214] E. J. Weinberg and P. Yi, “Magnetic monopole dynamics, supersymmetry, and duality,”
Phys. Rept. 438, 65 (2007) [arXiv:hep-th/0609055].

References 431
[215] S. Weinberg, The Quantum Theory of Fields. Vol. 1: Foundations, Cambridge, UK:
Univ. Pr. (1995) 609 p.
[216] G. Veneziano, “U(1) Without instantons,” Nucl. Phys. B 159, 213 (1979).
[217] J. Wess and B. Zumino, “Consequences of anomalous Ward identities,” Phys. Lett. B
37, 95 (1971).
[218] F. Wilczek, “Inequivalent embeddings of SU(2) and instanton interactions,” Phys. Lett.
B 65, 160.
[219] K. G. Wilson, “Nonlagrangian models of current algebra,” Phys. Rev. 179, 1499 (1969).
[220] E. Witten, “Some exact multipseudoparticle solutions of classical Yang-Mills theory,”
Phys. Rev. Lett. 38, 121 (1977).
[221] E. Witten, “Instantons, the Quark model, and the 1/N expansion,” Nucl. Phys. B 149,
285 (1979).
[222] E. Witten, “Baryons in the 1/N expansion,” Nucl. Phys. B 160, 57 (1979).
[223] E. Witten, “Large N chiral dynamics,” Annals Phys. 128, 363 (1980).
[224] E. Witten, “Nonabelian bosonization in two dimensions,” Commun. Math. Phys. 92,
455 (1984).
[225] E. Witten, “Global aspects of current algebra,” Nucl. Phys. B 223, 422 (1983).
[226] E. Witten, “Current algebra, baryons, and quark confinement,” Nucl. Phys. B 223, 433
(1983).
[227] E. Witten, “On holomorphic factorization of WZW and coset models,” Commun. Math.
Phys. 144, 189 (1992).
[228] E. Witten, “Two-dimensional gauge theories revisited,” J. Geom. Phys. 9, 303 (1992)
[229] C. N. Yang and R. L. Mills, “Conservation of isotopic spin and isotopic gauge invariance,”
Phys. Rev. 96, 191 (1954).
[230] C. N. Yang, “Some exact results for the many body problems in one dimension with
repulsive delta function interaction,” Phys. Rev. Lett. 19, 1312 (1967).
[231] I. Zahed and G. E. Brown, “The Skyrme model,” Phys. Rept. 142, 1 (1986).
[232] A. B. Zamolodchikov, “Renormalization group and perturbation theory near fixed points
in two-dimensional field theory,” Sov. J. Nucl. Phys. 46, 1090 (1987) [Yad. Fiz. 46, 1819
(1987)].
[233] A. B. Zamolodchikov, “Exact solutions of conformal field theory in two-dimensions and
critical phenomena,” Rev. Math. Phys. 1, 197 (1990).
[234] A. B. Zamolodchikov, “Thermodynamic Bethe anzatz in relativistic models, scaling three
state Potts and Lee-Yang models,” Nucl. Phys. B 342, 695 (1990).
[235] A. B. Zamolodchikov and A. B. Zamolodchikov, “Factorized S-matrices in two dimensions
as the exact solutions of certain relativistic quantum field models,” Annals Phys. 120,
253 (1979).
[236] A. B. Zamolodchikov, A. B. Zamolodchikov and I. M. Khalatnikov, “Physics Reviews,
Vol. 10, pt. 4: Condormal field theory and critical phenomena in two-dimensional sys-
tems,” London, UK: Harwood (1989) 269–433. (Soviet scientific reviews, Section A, 10.4)
[237] B. Zwiebach, A First Course in String Theory, Cambridge, UK: Univ. Pr. (2004) 558 p

Index
ADHM construction, 394–396 BPS

adjoint fermions magnetic monopole, 381
current states, 216 breather, 86
adjoint vacuum, 216–221 quantum, 90
affine current algebra, 5, 13, 39 BRST quantization
affine Lie algebra, see affine current 2d QC D, 281
algebra, 44
WZW model, 63 c-theorem, 100
algebra Callan–Symanzik equation, 321
simply laced, 42 Cartan matrix, 41
algebraic Bethe ansatz affine algebra, 47
N = 4 SYM, 332 Cartan sub-algebra, 40
anomalous dimension, 170, 321, 409 Cartan–Weyl basis
N = 4, 332 affine Lie algebra, 45
anomaly Lie group, 40
axial, in two dimensions, 178 Casimir operator
axial U (1), 346 YM in 2d, 296
axial, bosonized, 184 central charge, 20
character, 28
baryon spectrum YM in 2d, 296
2d, 247 ALA, 52
baryons WZW model, 77
4d large N, 346 closed string, 6
β function collective coordinates
C P N −1 model, 172 instantons, 397
Gross Neveu model, 170 collinear symmetry, 314–316
WZW model, 40, 65 Confinement versus screening
beta function, 323 2d, 265–277
large N , 337 conformal
Bethe ansatz dimension, 19
algebraic equations, 114–125 family, 22
continuum, 125 map, 4
Bethe–Salpeter equation, 194 transformation, 6
BFKL equation, 334 conformal anomaly, 309
bootstrap, 31 4d, 323
bosonization, 131–164 conformal dimension, 81
(β, γ) ghosts, 162 of A μ , 177
(b, c) ghosts, 160–162 conformal field theory 17–38
abelian, 132–139 conformal invariance
Dirac fermion, 132 4d QCD, 322–327
Mandelstam construction, 135 conformal symmetry
chiral, 148–159 in 4 dimensions, 309–327
Thirring model, sine-Gordon model, 136 in low dimensions, 17
Wakimoto, 163 conformal Ward identity
Witten’s non-abelian, 139 4d, 319–322
Majorana fermions, 139 conformal weights, 19

434 Index
conjugacy class, 72 Hartree approximation, 347

conserved charges highest weight state, 23
generating function, 103 affine algebra, 48
multilocal, 104 Lie algebra, 43
coset mode holomorphic
G/G in 2d QC D, 281 function, 5
C P N −1 homotopy group, 63
2d baryons, 242 D brane, 72
model, 171–174 hyper Kähler manifold
critical exponent, 37 monopoles, 387
crossing symmetry, 31, 96 hyper-Kahler manifold
instanton moduli space, 397
D branes, 71–73
Dashen–Frishman, 67, 69 instanton, 390
Dedekin η function, 157 4d QCD, 389–405
Derrick’s Theorem, 86 integrability, 92–99
descendant operator, 22 N = 4 4d SYM, 330–333
DHN, 90 4d, 329
dilatation operator classical, 101
N = 4 SYM, 330 quantum
dilogarithm function, 130 O(N ) model, 107
Dirac bracket, 151 integrable models, 79
2d baryons, 243 Ising model, 37
Dirac fermion, 34, 58
DLCQ, 223–235 Jacobi identity, 40
adjoint fermions, 228–235 Jacobi polynomials, 318
fundamental fermions, 224–228
duality, 31 Kac determinant, 25
Dynkin diagram KdV equation, 102
affine algebra, 47 kink, see solitons
Lie group, 42 Knizhnik–Zamolodchikov equation, 53–55,
dyons, 378 67–71
effective action large N

Gross Neveu model, 172 4d QCD, 337–343
energy momentum tensor, 4 counting rules, 342–345
eta’ mass, 344 meson phenomenology, 343
Euler character, 340 large N limit
exact sequence, 375 of QC D 2 , 191
exceptional groups, 49 two dimensional models, 165–174
Lax operator
fixed point spin chain, 112
integrable model, 100 Lax pair
flavor content sine-Gordon, 101
2d baryons, 247 light-cone
Frobenius relations, 300 discrete quantization, 223
fusion algebra, 23 light-cone gauge, 191
G/G coset models, 75 magnetic monopole

G/H coset models, 73–75 ’t Hooft Polyakov solutions, 376
gauge group Majorana fermion, 34, 55
non-abelian, 179 Mandelstam parameter, 333
Gegenbauer polynomials, 324 Maurer–Cartan equation, 356
Goldberger-Triman relation, 366 Maxwell theory
Gross-Neveu model, 166–170 in two dimensions, 177

Index 435
Mellin transform, 333 primary state

meson-baryon scattering N = 4 SYM, 331
2d, 252–264 principal value, 192
meson spectrum propagator
DLCQ in 2d, 232 gluon in 2d, 192
two current states, 212
minimal model, 27 q-deformation, 110
moduli space QCD
4d instantons, 396–400 2d, non-abelian bosonization of,
monopoles, 386 187
monodromy QC D 2
spin chain, 113 multiflavor, 189
Montonen Olive duality, 382 quantization
multibaryons BRST
2d, 249 G/G model, 76
canonical, 7
N = 4 4d SYM, 330 path integral, 12
Nahm construction, 383 radial, 9
Neveu-Schwarz fermion, 36 semi-classical of solitons, 91
Noether current, 7 quantum chromodynamics(QCD)
4d conformal symmetry, 312–313 in two dimensions, 180
normal ordering quantum groups, 108–110
2d QC D, 238 quasi particle, 118
null state, 24
null vector, 31 R matrix, 115
WZW model, 66 Ramond fermion, 36
rank, 40
OPE rapidity, 95
4d conformal symmetry, 318 Regge trajectory, 195
electromagnetic currents in 4d, reggeon, 333
324 relevant operator, 99
WZW model, 66 Riemann θ function, 157
open string, 6 Riemann surface
operator product expansion YM theory in 2d, 295
definition, 10 root, 40
product coefficients, 19 simple, 41
Weyl–Majorana, 35
S matrix
partition function factorized, 92–99
chiral boson, 156 sine-Gordon, 97
generalized 2d YM, 291 solution of, 95
parton distribution, 334 S-transformation, 33
pion distribution amplitude, 325 scalar field, 4
planar limit Schwinger Model, 178
N = 4 SYM, 331 bosonized, massive, 183
Poincare transformations, 80 BRST quantization, 283
Polyakov Wiegmann Formula, string tension, 265
142 semi-classical quantization
pomeron, 336 2d baryons, 240
Pontryagin index, 390 sine-Gordon
primary field, 19 discretized version, 123
4d collinear symmetry, 316 Skyrme model, 355–369
ALA, 51 action, 355–361
primary fields baryons, 361–369
Majoran fermion, 57 Skyrme term, 358

436 Index
soliton, 79 Virasoro anomaly, 15, 20

Skyrmion, 361 Bethe ansatz, 126
classical, 81–86 ALA, 51
sine-Gordon, 83 G/G model, 76
stability of, 84 G/H model, 74
quantum, 88–89 WZW model, 67
spectral parameter, 112
spin Ward identity
dimension, 19 SL(2, C ), 29
spin chain, 111–125 weight
N = 4 SYM, 331 lattice, 43
step operators, 40 vector, 43
string tension fundamental, 43
2d QC D, 272 highest state
Schwinger model, 265 spin chain, 115
stringy YM theory Wess Zumino (WZ) term, 63, 140
2d, 299 Wess Zumino Witten model, 61, 140
strong coupling chiral, 158
2d QC D, 237 with boundaries, 71
strong CP problem, 404 Weyl chambers, 44
Sugawara construction, 36, 50 Weyl fermion, 34
WZW model, 66 Weyl group, 41
Sugawara form, 20 affine algebra, 47
Weyl-Majorana, 34
theta function, 53 Wilson loop
theta term, 391 YM in 2d, 297
large N , 345 winding number, 184
Thirring model, 136 instanton, 400
’t Hooft equation WZ term
for 2d QCD, 195 Skyrme model, 357
’t Hoof equation WZW model see Wess Zumino Witten model
current states, 210 gauging of, 187
’t Hoof model
1/N corrections, 201 X X X 1 / 2 , 111
of 2d QCD, 191–202
scattering of mesons, 198 Yang–Mills
topological charge, 85 generalized in 2d, 291
bosonization, 133 Yang–Baxter equation, 94
magnetic monopoles, 373 spin chain, 112
topological current, 80 Yang–Mills theory
torus partition function in two dimensions, 179
G/G model, 77 Yangian symmetry, 105
tunneling
instantons, 400 zero modes
instantons, 397
Verlinde’s formula, 33 monopoles, 378
Verma module, 24 ζ-function regularization, 15
Virasoro algebra, 14 Zweig rule, 344

Non-Perturbative Field Theory

Uploaded by

Copyright:

Available Formats

Non-Perturbative Field Theory

Uploaded by

Document Information

Original Description:

Original Title

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

Non-Perturbative Field Theory

Uploaded by

Copyright:

Available Formats

https://doi.org/10.

1017/9781009401654 Published online by Cambridge University Press

Yi t z h a k F r i s h m a n is a Professor Emeritus at the Weizmann Institute,

https://doi.org/10.1017/9781009401654 Published online by Cambridge University Press

S. J. Aarseth Gravitational N-Body Simulations: Tools and Algorithms

https://doi.org/10.1017/9781009401654 Published online by Cambridge University Press

https://doi.org/10.1017/9781009401654 Published online by Cambridge University Press

https://doi.org/10.1017/9781009401654 Published online by Cambridge University Press

Cambridge University Press is part of Cambridge University Press & Assessment,

https://doi.org/10.1017/9781009401654 Published online by Cambridge University Press

https://doi.org/10.1017/9781009401654 Published online by Cambridge University Press

PART I NON-PERTURBATIVE METHODS IN

https://doi.org/10.1017/9781009401654 Published online by Cambridge University Press

3 Theories invariant under aﬃne current algebras 39

https://doi.org/10.1017/9781009401654 Published online by Cambridge University Press

6.5 Bosonization of systems of operators of high conformal

PART II TWO-DIMENSIONAL NON-PERTURBATIVE

https://doi.org/10.1017/9781009401654 Published online by Cambridge University Press

13.7 States, wave functions and binding energies 250

PART III FROM TWO TO FOUR DIMENSIONS

https://doi.org/10.1017/9781009401654 Published online by Cambridge University Press

17.6 Conformal Ward identities 319

https://doi.org/10.1017/9781009401654 Published online by Cambridge University Press

23.7 Hadronic phenomenology of two dimensions versus four

https://doi.org/10.1017/9781009401654 Published online by Cambridge University Press

https://doi.org/10.1017/9781009401654 Published online by Cambridge University Press

demonstrate application of the methods in addressing several questions of gauge

https://doi.org/10.1017/9781009401654 Published online by Cambridge University Press

https://doi.org/10.1017/9781009401654 Published online by Cambridge University Press

We would like to thank O. Aharony, A. Armoni, M. Bernstein, S. J. Brodsky,

https://doi.org/10.1017/9781009401654 Published online by Cambridge University Press

Non-perturbative methods in two-dimensional

https://doi.org/10.1017/9781009401654 Published online by Cambridge University Press

1.1 Complex geometry

https://doi.org/10.1017/9781009401654 Published online by Cambridge University Press

Fig. 1.1. The map between ξ and z.

1.2 Free massless scalar ﬁeld

https://doi.org/10.1017/9781009401654 Published online by Cambridge University Press

where L is the Lagrangian density. The factor 4π 1

1.3 Symmetries of the classical action

https://doi.org/10.1017/9781009401654 Published online by Cambridge University Press

In a similar manner the space-time translations (1.11) can also be elevated to

z → f (z); z̄ → f¯(z̄), (1.15)

referred to as two-dimensional conformal transformations. Aﬃne current alge-

1.4 Mode expansion

X̂(x0 , x1 ) = X̂(x0 , x1 + 2π) (1.17)

Integrating this expansion we get

with X a constant and

∂1 X̂(x0 , x1 = 0) = ∂1 X̂(x0 , x1 = π) = 0 =⇒ ∂ X̂(z, z̄ = z) = ∂¯X̂(z, z̄ = z).

The corresponding mode expansion takes the form

https://doi.org/10.1017/9781009401654 Published online by Cambridge University Press

1.5 Noether currents and charges

J ≡ ∂X; J¯ ≡ ∂¯X̄ (1.24)

are holomorphic and anti-holomorphic conserved,

The classical currents are determined up to an overall constant.

The associated holomorphic and anti-holomorphic conserved energy-

1.6 Canonical quantization

https://doi.org/10.1017/9781009401654 Published online by Cambridge University Press