CCNY-HEP-94/03

KUL-TF-94/12
UB-ECM-PF 94/15
UTTG-11-94
hep-th/9412228
May 1994
arXiv:hep-th/9412228v1 28 Dec 1994

Antibracket, Antifields
and Gauge-Theory Quantization
Joaquim Gomis∗1 , Jordi Parı́s♯2 and Stuart Samuel†3
∗Theory Group, Department of Physics
The University of Texas at Austin
RLM 5208, Austin, Texas
and
Departament d’Estructura i Constituents de la Matèria
Facultat de Fı́sica, Universitat de Barcelona
Diagonal 647, E-08028 Barcelona
Catalonia

♯ Instituut voor Theoretische Fysica

Katholieke Universiteit Leuven
Celestijnenlaan 200D
B-3001 Leuven, Belgium

†
Department of Physics
City College of New York
138th St and Convent Avenue
New York, New York 10031 U.S.A.

1 Permanent address: Dept. d’Estructura i Constituents de la Matèria, U. Barcelona.

E-mail: [email protected]
2 Wetenschappelijk Medewerker, I. I. K. W., Belgium.

E-mail: jordi=paris%tf%[email protected]
3 E-mail: [email protected]
 Abstract

The antibracket formalism for gauge theories, at both the classical and
quantum level, is reviewed. Gauge transformations and the associated
gauge structure are analyzed in detail. The basic concepts involved in
the antibracket formalism are elucidated. Gauge-fixing, quantum effects,
and anomalies within the field-antifield formalism are developed. The
concepts, issues and constructions are illustrated using eight gauge-theory
models.
Contents
1 Introduction 2

2 Structure of the Set of Gauge Transformations 10

2.1 Gauge Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Irreducible and Reducible Gauge Theories . . . . . . . . . . . . . . . 14
2.3 Trivial Gauge Transformations . . . . . . . . . . . . . . . . . . . . . . 17
2.4 The Gauge Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

3 Examples of Gauge Theories 24

3.1 The Spinless Relativistic Particle . . . . . . . . . . . . . . . . . . . . 24
3.2 Yang-Mills Theories . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.3 Topological Yang-Mills Theory . . . . . . . . . . . . . . . . . . . . . . 29
3.4 The Antisymmetric Tensor Field Theory . . . . . . . . . . . . . . . . 31
3.5 Abelian p-Form Theories . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.6 Open Bosonic String Field Theory . . . . . . . . . . . . . . . . . . . . 34
3.7 The Massless Relativistic Spinning Particle . . . . . . . . . . . . . . . 38
3.8 The First-Quantized Bosonic String . . . . . . . . . . . . . . . . . . . 44

4 The Field-Antifield Formalism 48

4.1 Fields and Antifields . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.2 The Antibracket . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.3 Classical Master Equation and Boundary Conditions . . . . . . . . . 51
4.4 The Proper Solution and the Gauge Algebra . . . . . . . . . . . . . . 54
4.5 Existence and Uniqueness . . . . . . . . . . . . . . . . . . . . . . . . 56
4.6 The Classical BRST Symmetry . . . . . . . . . . . . . . . . . . . . . 57

5 Examples of Proper Solutions 59

5.1 The Spinless Relativistic Particle . . . . . . . . . . . . . . . . . . . . 59
5.2 Yang-Mills Theories . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
5.3 Topological Yang-Mills Theory . . . . . . . . . . . . . . . . . . . . . . 62
5.4 The Antisymmetric Tensor Field Theory . . . . . . . . . . . . . . . . 62
5.5 Abelian p-Form Theories . . . . . . . . . . . . . . . . . . . . . . . . . 62
5.6 Open String Field Theory . . . . . . . . . . . . . . . . . . . . . . . . 63
5.7 The Massless Relativistic Spinning Particle . . . . . . . . . . . . . . . 66
5.8 The First-Quantized Bosonic String . . . . . . . . . . . . . . . . . . . 66
6 The Gauge-Fixing Fermion 68
6.1 Generalities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
6.2 Gauge-Fixing Auxiliary Fields . . . . . . . . . . . . . . . . . . . . . . 71
6.3 Delta-Function Gauge-Fixing Procedure . . . . . . . . . . . . . . . . 74
6.4 Other Gauge-Fixing Procedures . . . . . . . . . . . . . . . . . . . . . 83
6.5 Gauge-Fixed Classical BRST Symmetry . . . . . . . . . . . . . . . . 85
6.6 The Gauge-Fixed Basis . . . . . . . . . . . . . . . . . . . . . . . . . . 86

7 Gauge-Fixing Examples 89
7.1 The Spinless Relativistic Particle . . . . . . . . . . . . . . . . . . . . 89
7.2 Yang-Mills Theories . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
7.3 Topological Yang-Mills Theory . . . . . . . . . . . . . . . . . . . . . . 91
7.4 The Antisymmetric Tensor Field Theory . . . . . . . . . . . . . . . . 93
7.5 Open String Field Theory . . . . . . . . . . . . . . . . . . . . . . . . 94
7.6 The Massless Relativistic Spinning Particle . . . . . . . . . . . . . . . 96
7.7 The First-Quantized Bosonic String . . . . . . . . . . . . . . . . . . . 97

8 Quantum Effects and Anomalies 101

8.1 Quantum-BRST Transformation and Its Cohomology . . . . . . . . . 101
8.2 Satisfying the Quantum Master Equation . . . . . . . . . . . . . . . . 104
8.3 Remarks on Renormalization . . . . . . . . . . . . . . . . . . . . . . . 105
8.4 The Effective Action and the Zinn-Justin Equation . . . . . . . . . . 106
8.5 Quantum Master Equation Violations: Generalities . . . . . . . . . . 110
8.6 Canonical Transformations and the Quantum
Master Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
8.7 The Anomaly at the One-Loop Level . . . . . . . . . . . . . . . . . . 114

9 Sample Anomaly Calculations 119

9.1 Computation for the Spinless Relativistic Particle . . . . . . . . . . . 119
9.2 The Abelian Chiral Schwinger Model . . . . . . . . . . . . . . . . . . 123
9.3 Anomaly in the Open Bosonic String . . . . . . . . . . . . . . . . . . 126

10 Brief Discussion of Other Topics 134

10.1 Applications to Global Symmetries . . . . . . . . . . . . . . . . . . . 134
10.2 A Geometric Interpretation . . . . . . . . . . . . . . . . . . . . . . . 135
10.3 Locality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
10.4 Cohomological Aspects . . . . . . . . . . . . . . . . . . . . . . . . . 138
10.5 Equivalence with the Hamiltonian BFV Formalism . . . . . . . . . . 145
10.6 Unitarity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
 10.7 The Antibracket Formalism in General Coordinates . . . . . . . . . . 153
10.8 The D=26 Closed Bosonic String Field Theory . . . . . . . . . . . . . 155
10.9 Extended Antibracket Formalism for
Anomalous Gauge Theories . . . . . . . . . . . . . . . . . . . . . . . 159

A Appendix: Right and Left Derivatives 162

B Appendix: The Regularity Condition 164

C Appendix: Anomaly Trace Computations 166

References 171
This work is dedicated to Joseph and Marie,
to Pilar,
and to the memory of Pere and Francesca.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 2

1 Introduction
The known fundamental interactions of nature are all governed by gauge theo-
ries. The presence of a gauge symmetry indicates that a theory is formulated in a
redundant way, in which certain local degrees of freedom do not enter the dynamics.
Conversely, when there are degrees of freedom, which do not enter the lagrangian, a
theory possesses local invariances. Although one can in principle eliminate the gauge
degrees of freedom, there are reasons for not doing so. These reasons include manifest
covariance, locality of interactions, and calculational convenience.
The first example of a gauge theory was electrodynamics. Electric and magnetic
forces are generated via the exchange of photons. Being particles of spin 1, photons
involve a vector field, Aµ . However, not all four components of the electromagnetic
potential Aµ enter dynamically. Two degrees of freedom correspond to the two pos-
sible physical polarizations of the photon. The longitudinal degree of freedom plays
a role in interactions via virtual exchanges of photons. The remaining gauge degree
of freedom does not enter the theory. Consequently, electromagnetism is described
by a gauge theory. When it was realized that the weak interactions could be unified
with electromagnetism in an SU(2) × U(1) gauge theory [129, 266, 213] and that
this theory is renormalizable [243, 244], the importance of non-abelian gauge theo-
ries [276] grew enormously. The strong interactions are also governed by an SU(3)
non-abelian gauge theory. The fourth fundamental force is gravity. It is based on
Einstein’s general theory of relativity and uses general coordinate invariance. When
formulated in terms of a metric or any other convenient fields, gravity also possesses
gauge symmetries.
The quantization of gauge theories is not always straightforward. In the abelian
case, relevant for electromagnetism, the procedure is well understood. In contrast,
quantization of a non-abelian theory and its renormalization is more complicated.
Quantization generally involves the introduction of ghost fields. Typically, a gauge-
fixing procedure is used to render dynamical all degrees of freedom. Ghost fields
are used to compensate for the effects of the gauge degrees of freedom [101], so that
unitarity is preserved. In electrodynamics in the linear gauges, ghosts decouple and
can be ignored. In non-abelian gauge theories, convenient gauges generically involve
interacting ghosts. A major step in understanding these issues was the Faddeev-
Popov quantization procedure [98, 83], which relied heavily on the functional-integral
approach to quantization [102, 1, 165]. From this viewpoint, the presence of ghost
fields is understood as a “measure effect”. In dividing out the volume of gauge
transformations in function space, a Jacobian measure factor arises. This factor is
produced naturally by introducing quadratic terms in the lagrangian for ghosts and
then integrating them out. It was realized at a later stage that the gauge-fixed action
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 3

retains a nilpotent, odd, global symmetry involving transformations of both fields and
ghosts. This Becchi-Rouet-Stora-Tyutin (BRST) symmetry [36, 254] is what remains
of the original gauge invariance. In fact, for closed theories, the transformation law
for the original fields is like a gauge transformation with gauge parameters replaced
by ghost fields. In general, this produces nonlinear transformation laws. The relations
among correlation functions derived from BRST symmetry involve the insertions of
the BRST variation of fields. These facts require the use of composite operators and
it is convenient to introduce sources for these transformations. The Ward identities
[265] associated with the BRST invariance treated in this way are the Slavnov-Taylor
identities [233, 241]. The Slavnov-Taylor identities and BRST symmetry have played
an important role in quantization, renormalization, unitarity, and other aspects of
gauge theories.
Ghosts fields have been useful throughout the development of covariant gauge-
field-theory quantization [181, 182, 184, 205, 3]. It is desirable to have a formulation
of gauge theories that introduces them from the outset and that automatically in-
corporates BRST symmetry [32]. The field-antifield formulation has these features
[36, 277, 24, 25, 26, 27]. It relies on BRST symmetry as fundamental principle and
uses sources to deal with it [36, 254, 277]. It encompasses previous ideas and develop-
ments for quantizing gauge systems and extends them to more complicated situations
(open algebras, reducible systems, etc.) [113, 114, 172, 238, 81]. In 1975, J. Zinn-
Justin, in his study of the renormalization of Yang-Mills theories [277], introduced
the above-mentioned sources for BRST transformations and a symplectic structure
( , ) (actually denoted ∗ by him) in the space of fields and sources, He expressed the
Slavnov-Taylor identities in the compact form (Γ, Γ) = 0, where Γ, the generating
functional of the one-particle-irreducible diagrams, is known as the effective action
(see also [187]). These ideas were developed further by B. L. Voronov and I. V. Tyutin
in [263, 264] and by I. A. Batalin and G. A. Vilkovisky in refs.[24, 25, 26, 27, 28]. These
authors generalized the role of ( , ) and of the sources for BRST transformations and
called them the antibracket and antifields respectively. Due to their contributions,
this quantization procedure is often referred to as the Batalin-Vilkovisky formalism.
The antibracket formalism gained popularity among string theorists, when it was
applied to the open bosonic string field theory [56, 246]. It has also proven quite
useful for the closed string field theory and for topological field theories. Only within
the last few years has it been applied to more general aspects of quantum field theory.
In some sense, the BRST approach, which was driven, in part, by renormalization
considerations, and the field-antifield formalism, which was motivated by classical
considerations such as gauge structure, are not so different. When sources are in-
troduced for BRST transformations, the BRST approach resembles the field-antifield
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 4

one. Antifields, then, have a simple intepretation: They are the sources for BRST
transformations. In this sense, the field-antifield formalism is a general method for
dealing with gauge theories within the context of standard field theory.
The general structure of the antibracket formalism is as follows. One introduces
an antifield for each field and ghost, thereby doubling the total number of original
fields. The antibracket ( , ) is an odd non-degenerate symplectic form on the space
of fields and antifields. The original classical action S0 is extended to a new action
S, in an essentially unique way, to arrive at a theory with manifest BRST symmetry.
One equation, the master equation (S, S) = 0, reproduces in a compact way the
gauge structure of the original theory governed by S0 . Although the master equation
resembles the Zinn-Justin equation, the content of the two is different since S is a
functional of quantum fields and antifields and Γ is a functional of classical fields.
The antibracket formalism currently appears to be the most powerful method for
quantizing a gauge theory. Beyond tree level, order h̄ terms usually need to be added
to the action, thereby leading to a quantum action W . These counterterms are ex-
pected to render finite loop contributions, after a suitable regularization procedure
has been introduced. The master equation must be appropriately generalized to the
so-called quantum master equation. It involves a potentially singular operator ∆.
The regularization procedure and counterterms should also render ∆ and its action
on W well-defined. Violations of the quantum master equation are equivalent to
gauge anomalies [251]. To calculate correlation functions and scattering amplitudes
in perturbation theory, a gauge-fixing procedure is selected. This procedure elimi-
nates antifields in terms of functionals of fields. When appropriately implemented,
propagators exist, and the usual Feynman graph methods can be used. In addition,
for the study of symmetry properties, renormalization and anomalies, a modified ver-
sion of the gauge-fixing procedure is available which keeps antifields. In short, the
antibracket formalism has manifest gauge invariance or BRST symmetry, provides
the extra fields needed for covariant quantization, permits a perturbative expansion
of the quantum theory, and allows the study of quantum corrections to the symmetry
structure of the theory.
The field-antifield formalism can treat systems that cannot be handled by Faddeev-
Popov functional integration approach. This is particularly clear for theories in which
quartic ghost interactions arise [172, 81]. Faddeev-Popov quantization leads to an
action bilinear in ghost fields, and fails for the case of open algebras. An open
algebra occurs when the commutator of two gauge transformations produces a term
proportional to the equations of motion and not just another gauge transformation
[81, 27]. In other words, the gauge algebra closes only on-shell. Such algebras occur
in gravity [110] and supergravity [114, 172, 81, 258] theories. The ordinary Faddeev-
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 5

Popov procedure also does not work for reducible theories. In reducible theories, the
gauge generators are all not independent [67, 171, 6, 37, 228, 250, 260, 80, 242]. Some
modifications of the procedure have been developed by introducing ghosts for ghosts
[228, 178]. However, these modifications [228, 148, 178, 116] do not work for the
general reducible theory. Even for Yang-Mills theories, the Faddeev-Popov procedure
can fail, if one considers exotic gauge-fixing procedures for which “extraghosts” appear
[172, 199, 200]. The field-antifield formalism is sufficiently general to encompass
previously known lagrangian approaches to the quantization of gauge theories.
Perhaps the most attractive feature of the field-antifield formalism is its imita-
tion of a hamiltonian Poisson structure in a covariant way. In some instances, the
hamiltonian approach to quantization has the advantage of being manifestly unitary.
However, it is necessarily non-covariant since the time variable is treated in a manner
different from the space variables. In addition, the gauge invariances usually must be
fixed at the outset. In compensation for this, one needs to impose constraints on the
Hilbert space of states. In the field-antifield approach, the antibracket plays the role
of the Poisson bracket. As a consequence, hamiltonian concepts, such as canonical
transformations, can be formulated and used [262, 263, 264, 27, 105, 251]. At the
same time, manifest covariance and BRST invariance are maintained. Since the an-
tibracket formalism proceeds via the functional integral, the powerful techniques of
functional integration are available.
A non-trivial aspect of the field-antifield approach is the construction of the quan-
tum action W . When loop effects are ignored, W → S provides the solution to the
master equation. A straightforward but not necessarily simple procedure is available
for obtaining S given the classical action S0 and its gauge invariances for a finite-
reducible system. When quantum effects are incorporated, W must satisfy the more
singular quantum master equation. However, there is currently no known method that
guarantees the construction of W . The problem is that the field-antifield formalism
does not automatically provide the functional integration measure. These issues are
linked with those associated with unitarity, renormalization, quantum gauge invari-
ance, and anomalies. Because these aspects of gauge theories are inherently difficult,
it is not surprising that the field-antifield formalism does not provide a simple solu-
tion.
Another, less serious weakness, is that the antibracket formalism involves quite
a bit of mathematical machinery. Sometimes, a gauge theory is expressed in a form
which is more complicated than necessary. This can make computations somewhat
more difficult.
The organization of this article is as follows. Sect. 2 discusses gauge structure.
Some notation is presented during the process of introducing gauge transformations.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 6

The distinction between irreducible and reducible gauge theories is made. The latter
involve a redundant set of gauge invariances so that there are relations among the
gauge generators. As a result, there exists gauge invariances for gauge invariances,
and ghosts for ghosts. A theory is Lth-stage reducible if there are gauge invariances
for the gauge invariances for the gauge invariances, etc., L-fold times. The general
form of the gauge structure for a first-stage reducible case is determined. In Sect.
3, specific gauge theories are presented to illustrate the concepts of Sect. 2. The
spinless relativistic particle, non-abelian Yang-Mills theories, topological Yang-Mills
theory, the antisymmetric tensor field, free abelian p-form theories, open bosonic
string field theory, the massless relativistic spinning particle, and the first-quantized
bosonic string are treated. The spinless relativistic particle of Sect. 3.1 is also used to
exemplify notation. The massless relativistic spinning particle provides an example of
a simple supergravity theory, namely a theory with supersymmetric gauge invariances.
This system is used to illustrate the construction of supersymmetric and supergravity
theories. A review of the construction of general-coordinate-invariant theories is given
in the subsection on the first-quantized open bosonic string. These mini-reviews
should be useful to the reader who is new to these subjects.
The key concepts of the field-antifield formalism are elucidated in Sect. 4. An-
tifields are introduced and the antibracket is defined. The latter is used to define
canonical transformations. They can be quite helpful in simplifying computations.
Next, the classical master equation (S, S) = 0 is presented. When appropriate bound-
ary conditions are imposed, it reproduces, in a compact way, the gauge structure of
Sect. 2. A suitable action S satisfying the master equation is called a proper solution.
Given the gauge-structure tensors of a first-stage reducible theory, Sect. 4.4 presents
the generic proper solution. The last part of Sect. 4 defines and discusses the classical
BRST symmetry. Examples of proper solutions are provided in Sect. 5 for the gauge
field theories presented in Sect. 3.
Sect. 6 begins the passage from the classical to the quantum aspects of the
field-antifield formalism. The gauge-fixing procedure is discussed. The gauge-fixing
fermion Ψ is a key concept. It is used as a means of eliminating antifields in terms
of functions of fields. The result is an action that is suitable for use in the path
integral. Only in this context and in performing standard perturbative computations
are antifields eliminated. It is shown that results are independent of the choice of
Ψ, if the quantum action W satisfies the quantum master equation. To implement
gauge-fixing, more fields and their antifields must be introduced. How this works
for irreducible and first-stage reducible theories is treated first. Then, for reference
purposes, the general Lth-stage reducible case is considered. Delta-function type
gauge-fixing is treated in Sect. 6.3. Again, irreducible and first-stage reducible cases
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 7

are presented first. Again, for reference purposes, the general Lth-stage reducible
case is treated. Gauge-fixing by a gaussian averaging process is discussed in Sect.
6.4. After gauge-fixing, a classical gauge-fixed BRST symmetry can be defined. See
Sect. 6.5. The freedom to perform canonical transformations permits one to work in
any appropriate field basis. This freedom can be quite useful. Concepts tend to have
different interpretations in different bases. One basis, associated with Ψ and called
the gauge-fixed basis, is the last topic of Sect. 6. Examples of gauge-fixing procedures
are provided in Sect. 7. With the exception of the free p-form theory, the theories are
the ones considered in Sects. 3 and 5.
Quantum effects and possible gauge anomalies are analyzed in Sect. 8. The
key concepts are quantum-BRST transformations and the quantum master equa-
tion. Techniques for assisting in finding solutions to the quantum master equation
are provided in Sects. 8.2, 8.5 and 8.6. The generating functional Γ for one-particle-
irreducible diagrams is generalized to the field-antifield case in Sect. 8.4. This allows
one to treat the quantum system in a manner similar to the classical system. The
Zinn-Justin equation is shown to be equivalent to the quantum master equation.
When unavoidable violations of the latter occur, the gauge theory is anomalous. See
Sect. 8.5. Explicit formulas at the one-loop level are given. In Sect. 9, sample anomaly
calculations are presented. It is shown that the spinless relativistic particle does not
have an anomaly. In Sect. 9.2, the field-antifield treatment of the two-dimensional
chiral Schwinger model is presented. Violations of the quantum master equation are
obtained. This is expected since the theory is anomalous. A similar computation
is performed for the open bosonic string. For D 6= 26, the theory is anomalous, as
expected. Some of the details of the calculations are relegated to Appendix C.
Section 10 briefly presents several additional topics. The application of the field-
antifield formalism to global symmetries is presented. A review is given of the geo-
metric interpretation of E. Witten [273]. The next topic is the role of locality. This
somewhat technical issue is important for renormalizability and for cohomological
aspects. A summary of cohomological methods is given. Next, the relation between
the hamiltonian and antibracket approaches is discussed. The question of unitarity
is the subject of Sect. 10.6. One place where the field-antifield formalism has played
an essential role is in the D = 26 closed bosonic string field theory. This example is
rather complicated and not suitable for pedagogical purposes. Nevertheless, general
aspects of the antibracket formalism for the closed string field theory are discussed.
Finally, an overview is given of how to handle anomalous systems using an extended
set of fields and antifields.
Appendix A reviews the mathematical aspects of left and right derivatives, inte-
gration by parts, and chain rules for differentiation. Appendix B discusses in more
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 8

detail the regularity condition, which is a technical requirement of the antibracket

formalism.
At every stage of development of the formalism, there exists some type of BRST
operator. In the space of fields and antifields before quantization, a classical nilpo-
tent BRST transformation δB is defined by using the action S and the antibracket:
δB F = (F, S). From δB , a gauge-fixed version δBΨ is obtained by imposing the condi-
tions on antifields provided by the gauge-fixing fermion Ψ. At the quantum level, a
quantum version δB̂ of δB emerges. In the context of the effective action formulation,
a transformation δBcq , acting on classical fields, can be defined by using Γ in lieu of S.
Several subsections are devoted to the BRST operator, its properties and its utility.
The existence of a BRST symmetry is crucial to the development. Observables are
those functionals which are BRST invariant and cannot be expressed as the BRST
variation of something else. In other words, observables correspond to the elements
of the BRST cohomology. The nilpotency of δB and δB̂ are respectively equivalent
to the classical and quantum master equations. The traditional treatment of gauge
theories using BRST invariance is reviewed in [32]. For this reason, we do not discuss
BRST quantization in detail.
The antibracket formalism is rather versatile in that one can use any set of fields
(and antifields) related to the original fields (and antifields) by a canonical transfor-
mation. However, under such a change, the meaning of certain concepts change. For
example, the gauge structure, as determined by the master equation, has a different
interpretation in the gauge-fixed basis than in the original basis. Most of this review
uses the second viewpoint. The treatment in the gauge-fixed basis is handled in Sects.
6.6 and 8.4.
The material in each section strives to fulfill one of three purposes. A key purpose
is to present computations that lead to understanding and insight. Sections 2, 4, 6.1,
6.5, 6.6, 8 and 10 are mainly of this character. The second purpose is pedagogical.
This Introduction falls into this category in that it gives a quick overview of the
formalism and the important concepts. Sections 3, 5, 7, and 9 analyze specially chosen
gauge theories which allow the reader to understand the field-antifield formalism in
a concrete manner. Finally, some material is included for technical completeness.
Sections 6.2–6.4 present methods for gauge-fixing the generic gauge theory. Parts of
sections 2.2, 2.4 and 8.7 are also for reference purposes. Probably the reader should
not initially try to read these sections in detail. Many sections serve a dual role.
A few new results on the antibracket formalism are presented in this review.
They are included because they provide insight for the reader. We have tried to
have a minimum overlap with other reviews. In particular, cohomological aspects are
covered in [36, 89, 57, 32, 93, 8, 152, 157, 253] and more-detailed aspects of anomalies
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 9

are treated in [253]. Pedagogical treatments are given in references [157, 253]. In
certain places, material from reference [206] has been used.
This review focuses on the key points and concepts of antibracket formalism.
There is some emphasis on applications to string theory. Our format is to first
present the material abstractly and then to supply examples. The reader who is
new to this subject and mainly interested in learning may wish to reverse this order.
Exercises can be generated by verifying the abstract results in each of the sample
gauge theories of Sects. 3, 5, 7, and 9. Other systems, which have been treated
by field-antifield quantization and may be of use to the reader, are the free spin 52
field [26], the spinning string [136], the 10-dimensional Brink-Schwarz superparticle
and superstring [123, 139, 173, 190, 211, 43, 232, 44, 227]1 , chiral gravity [77], W3
gravity [161, 45, 74, 162, 72, 257], general topological field theories [66, 185, 50, 49,
127, 164, 191, 158, 79], the supersymmetric Wess-Zumino model [33, 159] and chiral
gauge theories in four-dimensions [251]. The antibracket formalism has found various
interpretations in mathematics [125, 126, 127, 209, 189, 198, 218, 219, 237]. Some
other recent relevant work can be found in [261, 21, 22, 47, 147]. The referencing in
this review is thorough but not complete. A restriction has been made to only cite
works directly relevant to the issues addressed in each section. Multiple references
are done first chronologically and then alphabetically. The titles of references are
provided to give the reader a better indication of the content of each work.
We work in Minkowski space throughout this article. Functional integrals are
defined by analytic continuation using Wick rotation. This is illustrated in the com-
putations of Appendix C. We use ηµν to denote the flat-space metric with the signa-
ture convention (−1, 1, 1, . . . , 1). Flat-space indices are raised and lowered with this
metric. The epsilon tensor εµ0 µ1 µ2 ...µd−1 is determined by the requirement that it be
antisymmetric in all indices and that ε012...d−1 = 1, where d − 1 and d are respectively
the dimension of space and space-time. We often use square brackets to indicate a
functional of fields and antifields to avoid confusion with the antibracket, i.e., S[Φ, Φ∗ ]
in lieu of S(Φ, Φ∗ ).

1
See [227] for additional references.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 10

2 Structure of the Set of Gauge Transformations

The most familiar example of a gauge structure is the one associated with a
non-abelian Yang-Mills theory [276], namely a Lie group. The commutator of two
Lie-algebra generators produces a Lie-algebra generator. When a basis is used, this
commutator algebra is determined by the structure constants of the Lie group. For
example, for the Lie algebra su(2) there are three generators and the structure con-
stant is the anti-symmetric tensor on three indices εαβγ . A commutator algebra, as
determined by a set of abstract structure constants, does not necessarily lead to a Lie
algebra. The Jacobi identity, which expresses the associativity of the algebra, must
be satisfied [258].
Sometimes, in more complicated field theories, the transformation rules involve
field-dependent structure constants. Such cases are sometimes referred to as “soft
algebras” [17, 236]. In such a situation, the determination of the gauge algebra is more
complicated than in the Yang-Mills case. The Jacobi identity must be appropriately
generalized [17, 24, 84]. Furthermore, new structure tensors beyond commutator
structure constants may appear and new identities need to be satisfied.
In other types of theories, the generators of gauge transformations are not inde-
pendent. This occurs when there is “a gauge invariance” for gauge transformations.
One says the system is reducible. A simple example is a theory constructed using
a three-form F which is expressed in terms of a two-form B by applying the exte-
rior derivative F = dB. The gauge invariances are given by the transformation rule
δB = dA for any one-form A. The theory is invariant under such transformations
because the lagrangian is a functional of F and F is invariant: δF = dδB = ddA = 0.
However, the gauge invariances are not all independent since modifying A by δA = dλ
for some zero-form λ leads to no change in the transformation for B. When A = dλ,
δB = dA = ddλ = 0. The structure of a gauge theory is more complicated than the
Yang-Mills case when there are gauge invariances for gauge transformations.
Another complication occurs when the commutator of two gauge transformations
produces a term that vanishes on-shell, i.e., when the equations of motion are used.
When equations of motion appear in the gauge algebra, how should one proceed?
In this section we discuss the above-mentioned complications for a generic gauge
theory. The questions are (i) what are the relevant gauge-structure tensors and (ii)
what equations do they need to satisfy. The answers to these questions lead us to the
gauge structure of a theory.
This section constitutes a somewhat technical but necessary prelude. A reader
might want to consult the examples in Sect. 3. The more interesting development of
the field-antifield formalism begins in Sect. 4.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 11

2.1 Gauge Transformations

This subsection introduces the notions of a gauge theory and a gauge transfor-
mation. It also defines notation. The antibracket approach employs an elaborate
mathematical formalism. Hence, one should try to become quickly familiar with
notation and conventions.
Consider a system whose dynamics is governed by a classical action S0 [φ], which
depends on n different fields φi(x), i = 1, · · · , n. The index i can label space-time
indices µ, ν of tensor fields, the spinor indices of fermion fields, and/or an index
distinguishing different types of generic fields. At the classical level, the fields are
functions of space-time. In the quantum system, they are promoted to operators. In
this section, we treat the classical case only.
Let ǫ(φi ) = ǫi denote the statistical parity of φi . Each φi is either a commuting field
(ǫi = 0) or an anticommuting field (ǫi = 1). One has φi(x)φj (y) = (−1)ǫi ǫj φj (y)φi(x).
Let us assume that the action is invariant under a set of m0 (m0 ≤ n) non-trivial
gauge transformations, which, when written in infinitesimal form, read
 
δφi (x) = Rαi (φ)εα (x) , where α = 1 or 2 . . . or m0 . (2.1)

Here, εα (x) are infinitesimal gauge parameters, that is, arbitrary functions of the
space-time variable x, and Rαi are the generators of gauge transformations. These gen-
erators are operators that act on the gauge parameters. In kernel form, (Rαi (φ)εα ) (x)
R
can be represented as dyRαi (x, y) εα (y).
It is convenient to adopt the following compact notation [82, 83]. Unless otherwise
stated, the appearance of a discrete index also indicates the presence of a space-
time variable. We then use a generalized summation convention in which a repeated
discrete index implies not only a sum over that index but also an integration over the
corresponding space-time variable. As a simple example, consider the multiplication
of two matrices g and h, written with explicit matrix indices. In compact notation,

f A B = g A C hC B (2.2)

becomes not only a matrix product in index space but also in function space. Eq.(2.2)
represents
XZ
f A B (x, y) = dz g A C (x, z) hC B (z, y) (2.3)
C

in conventional notation. In other words, the index A in Eq.(2.2) stands for A and
x in Eq.(2.3). Likewise, B and C in Eq.(2.2) represent {B, y} and {C, z}. The
generalized summation convention for C in compact notation yields a sum over the
discrete index C and an integration over z in conventional notation in Eq.(2.3). The
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 12

indices A, B and C in compact notation implicitly represent space-time variables x,

y, z, etc., and explicitly can be field indices i, j, k, etc., gauge index α, β, γ, etc., or
any other discrete index in the formalism.
With this convention, the transformation laws
XZ
i
δφ (x) = dyRαi (x, y) εα (y) (2.4)
α

can be written succinctly as

δφi = Rαi εα . (2.5)
The index α in Eq.(2.5) corresponds to the indices y and α in Eq.(2.4). The index i
in Eq.(2.5) corresponds to the indices x and i in Eq.(2.4). The compact notation is
illustrated in the example of Sect. 3.1. Although this notation might seem confusing at
first, it is used extensively in the antibracket formalism. In the next few paragraphs,
we present equations in both notations.
Each gauge parameter εα is either commuting, ǫ(εα ) ≡ ǫα = 0, or is anti-
commuting, ǫα = 1. The former case corresponds to an ordinary symmetry while
the latter is a supersymmetry. The statistical parity of Rαi , ǫ(Rαi ), is determined from
Eq.(2.1): ǫ(Rαi ) = (ǫi + ǫα ) (mod 2).
Let S0,i (φ, x) denote the variation of the action with respect to φi(x):

∂r S0 [φ]
S0,i (φ, x) ≡ , (2.6)
∂φi (x)
where the subscript r indicates that the derivative is to be taken from the right (see
Appendix A). Henceforth, when a subscript index i, j, etc., appears after a comma
it denotes the right derivative with respect to the corresponding field φi , φj , etc.. In
compact notation, we write Eq.(2.6) as S0,i = ∂∂φr S0
i where the index i here stands for

both x and i in Eq.(2.6).

The statement that the action is invariant under the gauge transformation in
Eq.(2.1) means that the Noether identities
Z n
X
dx S0,i (x) Rαi (x, y) = 0 (2.7)
i=1

hold, or equivalently, in compact notation

S0,i Rαi = 0 . (2.8)

Eq.(2.8) is derived by varying S0 with respect to right variations of the φi given

by Eq.(2.1). When using right derivatives, the variation δS0 of S0 , or of any other
object, is given by δS0 = S0,i δφi . If one were to use left derivatives, the variation
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 13

of S0 would read δS0 = δφi ∂∂φ l S0

i . Eq.(2.7) is sometimes zero because the integrand is

a total derivative. We assume that surface terms can be dropped in such integrals
– this is indeed the case when Eq.(2.7) is applied to gauge parameters that fall off
sufficiently fast at spatial and temporal infinity. The Noether identities in Eq.(2.8)
are the key equations of this subsection and can even be thought of as the definition
of when a theory is invariant under a gauge transformation of the form in Eq.(2.1).
To commence perturbation theory, one searches for solutions to the classical equa-
tions of motion, S0,i (φ, x) = 0, and then expands about these solutions. We assume
there exists at least one such stationary point φ0 = {φj0 } so that

S0,i |φ0 = 0 . (2.9)

Equation (2.9) defines a surface Σ in function space, which is infinite dimensional

when gauge symmetries are present.
As a consequence of the Noether identities, the equations of motion are not in-
dependent. Furthermore, new saddle point solutions can be obtained by performing
gauge transformations on any particular solution. These new solutions should not be
regarded as representing new physics however – fields related by local gauge trans-
formations are considered equivalent.
The Noether identities also imply that propagators do not exist. By differentiating
the identities from the left with respect φj , one obtains
!
∂l  j
 ∂l ∂r S0 ∂l Rαj
S 0,j Rα = Rαj + S0,j (−1)ǫi ǫj = 0 ,
∂φi ∂φi ∂φj ∂φ i

! 
∂l ∂r S0 
⇒ Rαj  =0 , (2.10)
∂φi ∂φj 
φ0
 
∂l ∂r S0
i.e., the hessian ∂φ i ∂φj of S0 is degenerate at any point on the stationary surface
i
Σ. The Rα are on-shell null vectors of this hessian. Since propagators involve the
inverse of this hessian, propagators do not exist for certain combinations of fields.
This means that the standard loop expansion cannot be straightforwardly applied. A
method is required to overcome this problem.
Technically speaking, to study the structure of the set of gauge transformations
it is necessary to assume certain regularity conditions on the space for which the
equations of motion S0,i = 0 hold. The interested reader can find these conditions
in Appendix B. A key consequence of the regularity conditions is that if a function
F (φ) of the fields φ vanishes on-shell, that is, when the equations of motion are
implemented, then F must be a linear combination of the equations of motion, i.e.,

F (φ)|Σ = 0 ⇒ F (φ) = S0,i λi (φ) , (2.11)

J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 14

where |Σ indicates the restriction to the surface where the equations of motion hold
[81, 28, 106, 103, 105]. Eq.(2.11) can be thought of as a completeness relation for the
equations of motion. We shall make use of Eq.(2.11) frequently.
Throughout Sect. 2, we assume that the gauge generators are fixed once and for
all. One could take linear combinations of the generators to form a new set. This
would change the gauge-structure tensors presented below. This non-uniqueness is
not essential and is discussed in Sect. 4.5.
To see explicit examples of the abstract formalism that follows, one may want to
glance from time to time at the examples of Sect. 3.

2.2 Irreducible and Reducible Gauge Theories

It is important to know any dependences among the gauge generators. Only with
this knowledge is possible to determine the independent degrees of freedom. The
purpose of this subsection is to analyze this issue in more detail for the generic case.
The simplest gauge theories, for which all gauge transformations are independent,
are called irreducible. When dependences exist, the theory is reducible. In reducible
gauge theories, there is a “kind of gauge invariance for gauge transformations” or what
one might call “level-one” gauge invariances. If the level-one gauge transformations
are independent, then the theory is called first-stage reducible. This may not happen.
Then, there are “level-two” gauge invariances, i.e., gauge invariances for the level-one
gauge invariances and so on. This leads to the concept of an L-th stage reducible
theory. In what follows we let ms denote the number of gauge generators at the s-th
stage regardless of whether they are independent.
Let us define more precisely the above concepts. Assume that all gauge invariances
of a theory are known and that the regularity condition described in Appendix B is
satisfied. Then, the most general solution to the Noether identities (2.8) is a gauge
transformation, up to terms proportional to the equations of motion:

S0,i λi = 0 ⇔ λi = R0α
i
0
λ′α0 + S0,j T ji , (2.12)

where T ij must satisfy the graded symmetry property

T ij = −(−1)ǫi ǫj T ji . (2.13)
i
The R0α 0
are the gauge generators in Eq.(2.1). For notational convenience, we have
appended a subscript 0 on the gauge generator and the gauge index α. This subscript
indicates the level of the gauge transformation. The second term S0,j T ji in Eq.(2.12)
is known as a trivial gauge transformation. Such transformations are discussed in
the next subsection. It is easily checked that the action is invariant under such
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 15

transformations due to the trivial commuting or anticommuting properties of the S0,j .

The first term R0αi
0
λ′α0 in Eq.(2.12) is similar to a non-trivial gauge transformation
of the form of Eq.(2.1) with εα0 = λ′α0 . The key assumption in Eq.(2.12) is that the
i
set of functionals R0α 0
exhausts on-shell the relations among the equations of motion,
namely the Noether identities. In other words, the gauge generators are on-shell a
complete set. This is essentially equivalent to the regularity condition.
i
If the functionals R0α 0
are independent on-shell then the theory is irreducible. In
such a case, 
i 
rank R0α 0
 = m0 , (2.14)
Σ
where m0 is the number of gauge transformations. The rank of the hessian
!
∂l ∂r S0  

rank  = n − rank Rαi  (2.15)
∂φi ∂φj  Σ
Σ

is n − m0 . Define the net number of degrees of freedom ndof to be the number of fields
that enter dynamically in S0 , regardless of whether they propagate.2 Then for an
irreducible theory ndof is n − m0 since there are m0 gauge degrees of freedom. Note
that ndof matches the rank of the hessian in Eq.(2.15).
If, however, there are dependences among the gauge  generators, and the rank
i 
of the generators is less than their number, rank R0α0  < m0 , then the theory is
Σ
reducible. If m0 − m1 of the generators are independent on-shell, then there are m1
α0
relations among them and there exist m1 functionals R1α 1
such that
i α0 ji
R0α0
R1α 1
= S0,j V1α 1
, α1 = 1, . . . , m1 ,
α0
ǫ(R1α 1
) = ǫα0 + ǫα1 (mod 2) , (2.16)
ji ij ji
for some V1α 1
, satisfying V1α 1
= −(−1)ǫi ǫj V1α 1
. Here, ǫα1 is the statistical parity
α0 i
of the level-one gauge parameter. The R1α1 are the on-shell null vectors for R0α 0
i α0  ji
since R0α 0
R 1α1 Σ  = 0. The presence of V 1α1 in Eq.(2.16) is a way of extending
this statement off-shell. Here and elsewhere, when a combination of field equations
appears on the right-hand side of an equation, it indicates the off-shell extension of an
on-shell statement; such an extension can be performed using the regularity postulate
of Appendix B. Note that, if εα = R1α α
εα for any εα1 , then δφi in Eq.(2.5) is zero on-
1 1
shell, so that no gauge transformation is produced. In Eq.(2.16) it is assumed that
i α0 α0
the reducibility of the R0α 0
is completely contained in R1α 1
, i.e., R1α 1
also constitute
a complete set
i
R0α0
λα0 = S0,j M0ji ⇒ λα0 = R1α
α0
1
λ′α1 + S0,j T0jα0 , (2.17)
2
In electromagnetism, ndof = 3, but there are only two propagating degrees of freedom corre-
sponding to the two physical polarizations.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 16

for some λ′α1 and some T0jα0 .

α0
If the functionals R1α 1
are independent on-shell

α0 
rank R1α 1
 = m1 ,
Σ

i 
then the theory is called first-stage reducible. One also has rank R0α  = m0 − m1
0 Σ
and the net number of degrees of freedom in the theory is n − m0 + m1 . Since true
and gauge degrees of freedom have been determined,

∂l ∂r S0 
rank  = n − m0 + m1 .
∂φi ∂φj Σ
α0
If the functionals R1α 1
are not all independent on-shell, relations exist among them
and the theory is second-or-higher-stage reducible. Then, the on-shell null vectors of
α0
R1α 1
and higher R-type tensors must be found.
One continues the above construction until it terminates. A theory is L-th stage
reducible [26] if there exist functionals
αs−1
Rsαs
, αs = 1, . . . , ms , s = 0, . . . , L , (2.18)
i i
such that R0α0
satisfies Eq.(2.8), i.e., S0,i R0α0
= 0, and such that, at each stage, the
αs−1
Rsαs constitute a complete set, i.e.,
αs−1 αs
Rsαs
λ = S0,j Msjαs−1 ⇒ λαs = Rs+1,α
αs
s+1
λ′αs+1 + S0,j Tsjαs ,
αs−2 αs−1 iαs−2
Rs−1,αs−1
Rsαs
= S0,i Vsα s
, s = 1, . . . , L ,
 L
X
αs−1 
rank Rsαs
 = (−1)t−s mt , s = 0, . . . , L , (2.19)
Σ
t=s
α i
where we have defined R0α−10 ≡ R0α 0
and α−1 ≡ i. The Rαs−1 arethe on-shell null
 sαs
αs−2 αs−1
vectors for Rs−1α s−1
. The statistical parity of Rsαs
is ǫαs−1 + ǫαs (mod 2), where
ǫαs is the statistical parity of the s-level gauge transformation associated with the
index αs . Finally,

∂l ∂r S0  XL
ndof = rank  = n − (−1)s ms (2.20)
∂φi ∂φj Σ s=0

is the net number of degrees of freedom.

J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 17

2.3 Trivial Gauge Transformations

As mentioned in the last subsection, trivial gauge transformations exist. Since
they are proportional to the equations of motion they do not lead to conservation
laws. This subsection discusses their role in the gauge algebra.
Given that the finite invertible gauge transformations satisfy the group axioms,
their infinitesimal counterparts necessarily form an algebra. Besides the usual gauge
transformations (2.1), there are the trivial transformations, defined as

δµ φi = S0,j µji , µji = −(−1)ǫi ǫj µij , (2.21)

where µji are arbitrary functions. It is easily demonstrated that, as a consequence

of the symmetry properties of µji, the transformations in Eq.(2.21) leave the action
invariant. In studying the structure of the gauge transformations, it is necessary to
take into consideration the presence of such transformations.
To determine their effect on the gauge algebra, consider the commutator of a
trivial transformation with any other transformation. Calling the latter δr φi = r i ,
one has
[δµ , δr ]φi = r,k
i
S0,j µjk − S0,j µji,k r k − S0,jk r k µji .
Given that δr is a symmetry transformation of S0 , it follows by differentiation by φj
that
S0,k r k = 0 ⇒ S0,jk r k + S0,k r,jk = 0 ,
so that the commutator becomes
 
j ki
[δµ , δr ]φi = S0,j r,k µ − (−1)ǫi ǫj r,k µ − µji,k r k = S0,j µ̃ji
i kj
,

from which one concludes that the commutator of a trivial transformation with any
other transformation is a trivial transformation. Hence, the trivial transformations
are a normal subgroup H of the full group of gauge transformations, Ḡ.
The trivial gauge transformations are of no physical significance: They neither
lead to conserved currents nor do they prevent the development of a perturbative
expansion about a stationary point. They are simply a consequence of having more
than one degree of freedom. On these grounds, it would seem sensible to dispense with
them and restrict oneself to the quotient G = Ḡ/H. However, this is only possible in
certain cases. In general, the commutator of two non-trivial gauge transformations
produces trivial gauge transformations. Furthermore, for reasons of convenience,
particularly when it is desirable to have manifest covariance or preserve locality, one
sometimes wants to include trivial transformations. Hence, the full group Ḡ is used
for studying the gauge structure of the theory.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 18

2.4 The Gauge Structure

In this section we restrict ourselves to the simpler cases of irreducible and first-
stage-reducible gauge theories. To avoid cumbersome notation, we use Rαi for R0α i
0
,
α α0 ji ji
Za for R1α1 , and in Eq.(2.16) we use Va for V1α1 , so that the indices α0 and α1
respectively correspond to α and a.
The general strategy in obtaining the gauge structure is as follows [81]. The first
gauge-structure tensors are the gauge generators themselves, and the first gauge-
structure equations are the Noether identities (2.8). One computes commutators,
commutators of commutators, etc., of gauge transformations. Graded symmetrization
produces identity equations for the structure tensors that must be satisfied. Generic
solutions are obtained by exploiting the consequences of the regularity conditions,
namely, completeness. In using completeness, additional gauge-structure tensors ap-
pear. They enter in higher-order symmetrized commutator identity equations. The
process is continued until it terminates.
Although this section provides some insight, it is somewhat technical so that the
reader may wish to skip it at first. If one is only interested in the irreducible case,
one should read to Eq.(2.36). For reasons of space, many details of the algebra are
omitted. As an exercise, the reader can provide the missing steps.
Consider the commutator of two gauge transformations of the type in Eq.(2.1).
On one hand, a direct computation leads to
 
[δ1 , δ2 ]φi = Rα,j
i
Rβj − (−1)ǫα ǫβ Rβ,j
i
Rαj εβ1 εα2 .

On the other hand, since this commutator is also a gauge symmetry of the action it
satisfies the Noether identity so that, factoring out the gauge parameters εβ1 and εα2 ,
one may write  
i
S0,i Rα,j Rβj − (−1)ǫα ǫβ Rβ,j
i
Rαj = 0 .
Taking into account Eq.(2.12) the above equation implies the following important
relation among the generators
i
Rα,j Rβj − (−1)ǫα ǫβ Rβ,j
i γ
Rαj = Rγi Tαβ ji
− S0,j Eαβ , (2.22)
γ ji γ ji
for some gauge-structure tensors Tαβ and Eαβ . This equation defines Tαβ and Eαβ .
β α
Restoring the dependence on the gauge parameters ε1 and ε2 , the last two equations
imply
γ β α ji β α
[δ1 , δ2 ]φi ≡ Rγi Tαβ ε1 ε2 − S0,j Eαβ ε1 ε2 , (2.23)
γ
where Tαβ are known as the “structure constants” of the gauge algebra. The words
γ
structure constants are in quotes because in general the Tαβ depend on the fields of
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 19

ji
the theory and are not “constant”. The possible presence of the Eαβ term is due to
the fact that the commutator of two gauge transformations may give rise to trivial
gauge transformations [81, 24, 27].
ij
The gauge algebra generated by the Rαi is said to be open if Eαβ 6= 0, whereas the
ij
algebra is said to be closed if Eαβ = 0. Moreover, Eq.(2.22) defines a Lie algebra if
ij γ
the algebra is closed, Eαβ = 0, and the Tαβ do not depend on the fields φi.
The gauge-structure tensors have the following symmetry properties under the
interchange of indices
ij ji ij
Eαβ = −(−1)ǫi ǫj Eαβ = −(−1)ǫα ǫβ Eβα ,
γ γ
Tαβ = −(−1)ǫα ǫβ Tβα . (2.24)
ij
In other words, Eαβ is graded-antisymmetric both in lower indices and in upper indices
γ
and Tαβ is graded-antisymmetric in lower indices. The statistical parity of structure
tensors is determined  by the sum of the parities of the tensor  indices,
 so that ǫ (Rαi ) =
γ ij
(ǫα + ǫi ) (mod 2), ǫ Tαβ = (ǫα + ǫβ + ǫγ ) (mod 2), and ǫ Eαβ = (ǫi + ǫj + ǫα + ǫβ )
(mod 2).
The next step determines the restrictions imposed by the Jacobi identity. In
general, it leads to new gauge-structure tensors and equations [172, 258, 84, 28]. The
identity X
[δ1 , [δ2 , δ3 ]] = 0 ,
cyclic over 1, 2, 3

produces the following relations among the tensors R, T and E

X  
ji
Rδi Aδαβγ − S0,j Bαβγ εγ1 εβ2 εα3 = 0 , (2.25)
cyclic over 1, 2, 3

where we have defined

 
η
3Aδαβγ ≡ Tαβ,k
δ
Rγk − Tαη
δ
Tβγ +
   
η
δ
(−1)ǫα (ǫβ +ǫγ ) Tβγ,k δ
Rαk − Tβη η
Tγα δ
+ (−1)ǫγ (ǫα +ǫβ ) Tγα,k δ
Rβk − Tγη Tαβ , (2.26)
and
 
ji ji ji δ j kj
3Bαβγ ≡ Eαβ,k Rγk − Eαδ Tβγ − (−1)ǫi ǫα Rα,k ki
Eβγ + (−1)ǫj (ǫi +ǫα ) Rα,k
i
Eβγ
 
α→β
+ (−1)ǫα (ǫβ +ǫγ ) RHS of above line with β→γ
 (2.27)
γ→α
 
α→γ
+(−1)ǫγ (ǫα +ǫβ ) RHS of first line with β→α
 .
γ→β
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 20

A useful but lengthy exercise is to derive Eq.(2.25).

If the theory is irreducible, the on-shell independence of the generators (2.14) and
their completeness (2.12) leads to the following solution of Eq.(2.25)
jδ
Aδαβγ = S0,j Dαβγ , (2.28)
jδ
where Dαβγ are new structure functions.
On the other hand, using this solution in the original equation (2.25), one obtains
jδ
the following condition on the Dαβγ
X  
ji jδ
S0,j Bαβγ − (−1)ǫj (ǫi +ǫδ ) Rδi Dαβγ ǫγ1 ǫβ2 ǫα3 = 0 . (2.29)
cyclic over ǫ1 , ǫ2 , ǫ3

Again, the completeness of the generators implies that the general solution of the
preceding equation is of the form
ji
Bαβγ + (−1)ǫi ǫδ Rδj Dαβγ
iδ jδ
− (−1)ǫj (ǫi +ǫδ ) Rδi Dαβγ kji
= −S0,k Mαβγ , (2.30)
kji
where Mαβγ is graded antisymmetric in i, j, and k. In this way, the Jacobi identity
jδ kji
leads to the existence of two new gauge-structure tensors Dαβγ and Mαβγ which, for
a generic theory, are different from zero and must satisfy Eqs.(2.28) and (2.30).
Continuing in the same way, that is to say, commuting more and more gauge
transformations, new structure tensors with increasing numbers of indices are ob-
tained. These tensors are the so-called structure functions of the gauge algebra and
they determine the nature of the set of gauge transformations of the theory. The
reader may not be aware of the higher-order tensors because in the simplest gauge
theories, such as Yang-Mills, they vanish.
ji jδ kji
The tensors Aδαβγ , Bαβγ , Dαβγ and Mαβγ are all graded-antisymmetric in α, β
ji kji
and γ. In addition Bαβγ is graded-antisymmetric in i and j while Mαβγ is graded
antisymmetric in i, j and k. Here we summarize these properties

Aδαβγ = −(−1)ǫα ǫβ Aδβαγ = −(−1)ǫβ ǫγ Aδαγβ ,

ji ji ji ij
Bαβγ = −(−1)ǫα ǫβ Bβαγ = −(−1)ǫβ ǫγ Bαγβ = −(−1)ǫi ǫj Bαβγ ,
jδ jδ jδ
Dαβγ = −(−1)ǫα ǫβ Dβαγ = −(−1)ǫβ ǫγ Dαγβ , (2.31)
kji kji kji jki kij
Mαβγ = −(−1)ǫα ǫβ Mβαγ = −(−1)ǫβ ǫγ Mαγβ = −(−1)ǫj ǫk Mαβγ = −(−1)ǫi ǫj Mαβγ .
The statistical parity of any of the above tensors is given by the sum of the statistical
parities of the indices of that tensor.
A useful device in the study of gauge-structure relations is to introduce ghost
fields C α with opposite statistics to those of the gauge parameters εα ,

ǫ(C α ) = ǫα + 1 , (2.32)
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 21

and to replace gauge parameters by ghosts, as is done in the BRST formalism [32].
The ghost fields obey the same boundary conditions as gauge parameters. The ghosts
can be used as a compact way of writing the gauge-structure equations. However, in
γ ij jδ kji
order to do this, the symmetry properties of Tαβ , Eαβ , Dαβγ , Mαβγ , etc., need to be
correctly incorporated. Note that these tensors are graded anti-symmetric in lower-
index gauge indices α, β, etc., whereas the ghosts satisfy C α C β = (−1)(ǫα +1)(ǫβ +1) C β C α .
If one is given a graded anti-symmetric tensor Tα1 α2 α3 α4 ... , then a way to make it into
a graded symmetric tensor with symmetry factors ǫα1 + 1, ǫα2 + 1, etc., associated
with indices α1 , α2 , etc., is to multiply by a factor of (−1)ǫαi for every other index αi
in Tα1 α2 α3 α4 ... . In other words, one replaces Tα1 α2 α3 α4 ... by (−1)ǫα2 +ǫα4 +...Tα1 α2 α3 α4 ... .
Using this device, one arrives at a compact way of writing the Noether identity (2.8),
the gauge commutator relation (2.22), as well as Eqs.(2.28) and (2.30) which arise
from the Jacobi identity:
S0,i Rαi C α = 0 , (2.33)
 
i
2Rα,j Rβj − Rγi Tαβ
γ ji
+ S0,j Eαβ (−1)ǫα C β C α = 0 , (2.34)
 
jδ
Aδαβγ − S0,j Dαβγ (−1)ǫβ C γ C β C α = 0 , (2.35)
 
ji
Bαβγ + (−1)ǫi ǫδ Rδj Dαβγ
iδ jδ
− (−1)ǫj (ǫi +ǫδ ) Rδi Dαβγ kji
+ S0,k Mαβγ (−1)ǫβ C γ C β C α = 0 ,
(2.36)
δ ji
where Aαβγ and Bαβγ are defined in Eqs.(2.26) and (2.27). The graded-anticommuting
nature of the ghosts automatically produces the appropriate graded-cyclic sums.
Equations (2.33) through (2.36) are key equations for an irreducible algebra.
Now let us consider a first-stage reducible gauge theory. In this case, the existence
of non-trivial relations among the generators in Eq.(2.16) leads to the appearance of
new tensor quantities.
For first-stage reducible theories there are on-shell null vectors for the generators
Rα . Let Zaα denote these null vectors. In Eq.(2.16), the Zaα are called R1a
i α
when α = α0
and a = α1 . The null vectors are independent on-shell. Their presence modifies the
solutions of the Jacobi identities in Eqs.(2.35) and (2.36) as well as higher-commutator
structure equations. In addition there are new structure equations. One of these is
Eq.(2.16) itself:
Rαi Zbα = S0,j Vbji . (2.37)
Another is derived as follows. Take relation (2.22) and multiply it by Zaβ to obtain
 
i
Rα,j Rβj − (−1)ǫα ǫβ Rβ,j
i γ
Rαj − Rγi Tαβ ji
+ S0,j Eαβ Zaβ = 0 .

Use Eq.(2.37) to express Rβj Zaβ as a term proportional to equations of motion. Also
i
do the same with Rβ,j Rαj Zaβ and make use of the Noether identity in Eq.(2.8). After
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 22

a little algebra, one finds that the previous equation can be written in the form
 
Rγi (−1)ǫa ǫβ Za,j
γ
Rβj − Tβδ
γ ji
Zaδ = S0,j Mβa ,

ji
for some quantity Mβa . Terms proportional to the equations of motion have been
ji
collected into Mβa . Using the completeness of the null vectors Zaα , the general solution
to this equation is

(−1)ǫa ǫβ Za,j
γ
Rβj − Tβδ
γ
Zaδ = −Zdγ Adaβ − S0,j Gjγ
aβ . (2.38)

Eq.(2.38) is a new gauge-structure equation for the first-stage reducible case. Two
new structure tensors Adaβ and Gjγ aβ arise.
The null vectors also lead to modifications of the solution of the Jacobi identity.
Eq.(2.25) still holds but its solution is different. Instead of Eq.(2.35), one obtains
 
Aδαβγ + Zcδ Fαβγ
c jδ
− S0,j Dαβγ (−1)ǫβ C γ C β C α = 0 , (2.39)

where we have made use of the completeness of the null vectors Zaα . In this equation
Aδαβγ stands for the combination of terms in Eq.(2.26).
Multiplying Eq.(2.39) by Rδi and using the Jacobi identity result of Eq.(2.25) lead
ij
to a modification of Eq.(2.29) involving Bαβγ . The new result reads
 
ji
S0,j Bαβγ − (−1)ǫj (ǫi +ǫδ ) Rδi Dαβγ
jδ
+ Vcji Fαβγ
c
(−1)ǫβ C γ C β C α = 0 ,

when written using ghosts. The general solution is


ji
Bαβγ + (−1)ǫi ǫδ Rδj Dαβγ
iδ jδ
− (−1)ǫj (ǫi +ǫδ ) Rδi Dαβγ +

kji
Vcji Fαβγ
c
+ S0,k Mαβγ (−1)ǫβ C γ C β C α = 0 , (2.40)
ji
where Bαβγ is given in Eq.(2.27).
By taking more and more commutators of gauge transformations, more structure
functions and equations appear, some of which involve graded symmetrizations in the
first-stage gauge indices a, b, etc.. As in the irreducible case, it is useful to introduce
ghosts η a to automatically incorporate graded symmetrization. Equations (2.37) and
(2.38) can then be written as
 
Rβi Zaβ − S0,j Vaji η a = 0 , (2.41)

and  
(−1)ǫa ǫβ Za,j
γ
Rβj − Tβδ
γ
Zaδ + Zdγ Adaβ + S0,j Gjγ a β
aβ η C = 0 . (2.42)
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 23

To summarize, key equations for first-stage reducible theories are Eqs.(2.33), (2.34)
and (2.39) – (2.42). Besides the null vectors Zaβ , the new structure tensors are Vaji ,
Adaβ , Gjγ a
aβ , Fαβγ as well as higher-level tensors.
Needless to say, for a higher-order reducible theory the number of quantities and
equations increases considerably. The complexity of the formalism makes the study
of the gauge structure at higher levels quite complicated. A more sensible approach is
to have a generating functional whose expansion in terms of auxiliary fields produces
the generic gauge-structure tensors. In addition, it is desirable to have a simple single
equation which, when expanded in terms of auxiliary fields, generates the entire set
of gauge-structure equations. The field-antifield method [24, 25, 26] provides such
a formalism. The generating functional for structure tensors is a generalized action
subject to certain boundary conditions and the classical master equation contains
all the gauge-structure equations. Before presenting the abstract machinery, it is of
pedagogical value to consider some examples of the formalism of this section.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 24

3 Examples of Gauge Theories

This section presents eight gauge theories, which will be used in Sect. 5 to illustrate
the antibracket formalism. The theories are (1) the spinless relativistic particle, (2)
Yang-Mills theories, (3) four-dimensional topological Yang-Mills theories, (4) the four-
dimensional antisymmetric tensor field, (5) abelian p-form theories, (6) the open
bosonic string field theory, (7) the massless relativistic spinning particle, and (8) the
first-quantized bosonic string. Models (1), (2), (7) and (8) are closed and irreducible.
Models (3) and (4) are first-stage reducible. Model (5) is p-stage reducible and model
(6) is an infinitely reducible open system. For each theory, the classical action S0 and
its gauge symmetries are first presented. Then, the non-zero gauge-structure tensors
are obtained. The determination of the structure tensors is the first computational
step in the antibracket formalism. The results in this section are used in Sect. 5 to
obtain proper solutions S.
In the first subsection on the spinless relativistic particle, we illustrate the compact
notation of Sect. 2.1. Models (1) and (8) are respectively one and two-dimensional
gravity theories. Model (8), the first-quantized bosonic string, is used to explain the
construction of general-coordinate-invariant theories, i.e., gravities. In the subsection
3.7 on the massless relativistic spinning particle, we provide a mini-review of super-
symmetry and supergravity. A brief introduction to string field theory is given in
Sect. 3.6.
As exercises for the reader, we suggest the following three computations. (i)
Verify that the gauge transformations leave S0 invariant. (ii) Given S0 and its gauge
symmetries, obtain the results presented for the gauge-structure tensors. (iii) Verify
Eqs.(2.33)–(2.36) for the irreducible theories, and verify Eqs.(2.33), (2.34) and (2.39)
– (2.42) for the first-stage reducible theories.

3.1 The Spinless Relativistic Particle

One of the simplest examples of a model with a gauge invariance is the free
relativistic particle. It actually corresponds to a 0+1 dimensional gravity theory with
scalar fields. The supersymmetric generalization of the spinless relativistic particle is
presented in Sect. 3.7.
Let us use this system to illustrate the formalism in Sect. 2. The degrees of
freedom are a particle coordinate xµ and an einbein e both of which are functions of
a single proper time variable τ . The action is given by
Z  
µ 1 ẋµ ẋµ
S0 [x , e] = dτ − m2 e , φi = (xµ , e) , (3.1)
2 e
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 25

where a dot over a variable indicates a derivative with respect to proper time. The
variations of the action with respect to the fields, Eq.(2.6), are
  !
d ẋµ 1 ẋ2
S0,µ =− , S0,e = − 2 − m2 , (3.2)
dτ e 2 e

where S0,e is the variation of the action with respect to field e; in other words, we also
use e as a field index for the einbein e. If the equation of motion S0,e = 0 is used to
solve for e, and this solution is substituted into the action in Eq.(3.1), one finds that
R √
the action becomes the familiar one: S0 = − dτ m −ẋ2 . Classically, this action
and the one in Eq.(3.1) are equivalent.
The infinitesimal gauge transformations for this system can be written as
ẋµ
δxµ = ε, δe = ε̇ . (3.3)
e
It is straightforward to verify that Eq.(3.3) is a symmetry of Eq.(3.1). The Noether
identity in Eq.(2.8) reads
Z (     ! )
ẋµ d ẋµ 1 ẋ2 d
dτ − − 2
+ m2 δ (τ − τ ′ ) = 0 , (3.4)
e dτ e 2 e dτ

which is verified using integration by parts.

The transformations laws in Eq.(3.3) in the form of Eq.(2.5) are
ẋµ d
Rµ ε = ε, Re ε = ε . (3.5)
e dτ
ẋµ d
Eq.(3.5) says that Rµ is the operator that is multiplication by e
and Re is dτ
. In
kernel form using compact index notation, they are
ẋµ (τ ) d
Rσµτ = δ (τ − σ) , Rσeτ = δ (τ − σ) .
e dτ
Recall that in using compact notation the index α of Rαi in Eq.(2.5) represents not
only a discrete index labelling the different gauge transformations but also a space-
time index. Since there is only one type of gauge transformation the discrete index
takes on only one value, which we drop for convenience. Hence the index α of Rαi is
replaced by the space-time variable σ. In this subsection, we use the Greek letters ρ,
σ, τ and υ to denote proper time variables. Likewise the index i on Rαi in Eq.(2.5)
represents not only a field index µ or e but also a proper time variable τ .
The algebra of the gauge transformations is simply

[δ1 , δ2 ] = [δ(ε1 ) , δ(ε2 )] = 0 ,

J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 26

where δ(ε1 ) indicates a gauge transformation with parameter ε1 . These abelian gauge
transformations (3.3) are related to the standard reparametrization transformations
d
δR xµ = ẋµ ε , δR e = (eε) (3.6)
dτ
through the following redefinition of the gauge parameter

ε −→ εe .

The algebra of reparametrization transformations reads

[δR (ε1 ), δR (ε2 )] = δR (ε12 ) , (3.7)

with the parameter ε12 given by

ε12 = ε̇1 ε2 − ε1 ε̇2 . (3.8)

Eqs.(3.7) and (3.8) correspond to the usual diffeomorphism algebra. Since the com-
mutator of two gauge transformations is a gauge transformation, the algebra is closed
ji
and Eαβ in Eq.(2.23) is zero. This example illustrates the effect of field-dependent
redefinitions of the gauge parameters or, equivalently, of the gauge generators: An
abelian algebra can be transformed into a non-abelian one. The converse of this also
holds. One can transform any given non-abelian algebra into an abelian algebra using
field-dependent redefinitions, a result known as the abelianization theorem [27]. The
fact that this process can spoil the locality of the transformations is one of the reasons
for using the non-abelian version.
It may appear unusual that a single family of gauge transformations produces
non-abelian commutation relations. This is due to the local non-commutativity of
reparametrization transformations that arises from the time derivatives in Eq.(3.6).
Indeed, when ε1 and ε2 have non-overlapping support, i.e., ε1 (τ ) = 0 where ε2 (τ ) 6= 0
and vice-versa, ε12 = 0.
γ
It is instructive to see how a non-zero structure constant Tαβ for the diffeomor-
phism algebra arises using compact notation. In what follows, gauge indices, α, β,
etc. are replaced by proper time variable ρ, σ, τ , υ. From Eq.(3.6) one sees that the
transformation operators R for reparametrizations are

Rσµτ = ẋµ (τ ) δ (τ − σ) ,
d d
Rσeτ = e (τ ) δ (τ − σ) + ė (τ ) δ (τ − σ) = [e (τ ) δ (τ − σ)] .
dτ dτ
For the xµ degrees of freedom, a straightforward computation yields
µτ d
Rσ,νυ = δνµ δ (τ − σ) δ (τ − υ) ,
dτ
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 27

XZ d
µτ
Rσ,νυ Rρνυ = dυδνµ δ (τ − σ) δ (τ − υ) ẋν (υ) δ (υ − ρ)
ν dτ
d
= ẍµ (τ ) δ (τ − σ) δ (τ − ρ) + ẋµ (τ ) δ (τ − σ) δ (τ − ρ) .
dτ
Antisymmetrizing in ρ and σ and comparing with Eq.(2.22) one finds

τ d d
Tσρ = δ (τ − σ) δ (τ − ρ) − δ (τ − ρ) δ (τ − σ) , (3.9)
dτ dτ
which is in agreement with Eqs.(3.7) and (3.8). For e, straightforward computation
produces

eτ d d d
Rσ,eυ = δ (τ − υ) δ (τ − σ) + δ (τ − σ) δ (τ − υ) = [δ (τ − υ) δ (τ − σ)] ,
dτ" #" dτ # dτ
eτ d d d
Rσ,eυ Rρeυ = e (τ ) δ (τ − σ) δ (τ − ρ) + ė (τ ) δ (τ − ρ) δ (τ − σ) +
dτ dτ dτ
2
d d
2ė (τ ) δ (τ − σ) δ (τ − ρ) + e (τ ) δ (τ − σ) 2 δ (τ − ρ) + ë (τ ) δ (τ − σ) δ (τ − ρ) .
dτ dτ
Antisymmetrizing in ρ and σ and using Eq.(2.23), one finds T is again given by
Eq.(3.9).
Although compact notation is useful to represent the formalism of gauge theories
in full generality, it is cumbersome for specific theories, especially for those in which
more natural notation has already been established. In the examples that follow,
we do not explicitly display equations in compact form but use more conventional
notation.

3.2 Yang-Mills Theories

Yang-Mills theories [276] are perhaps the most familiar gauge theories. For each
Lie algebra G there is different theory. The fundamental fields are gauge potentials Aaµ
where there is an index a for each generator Ta of G. In a matrix representation, the
generators are antihermitian matrices which are conventionally normalized so that
T r (Ta Tb ) = − 12 δab . The generators satisfy

[Ta , Tb ] = fab c Tc , (3.10)

where fab c are the structure constants of G. They are real and antisymmetric in lower
indices fab c = −fba c and they must satisfy the Jacobi identity

fab e fec d + fca e feb d + fbc e fea d = 0 . (3.11)

J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 28

The Yang-Mills action is

Z Z
1 1
S0 [Aaµ ] = − dd x Fµν
a
(x)Faµν (x) = dd x T r [Fµν (x)F µν (x)] , (3.12)
4 2
a
where d is the dimension of space-time, Fµν (x) ≡ Fµν (x)Ta , and where the field
a
strengths Fµν are
a
Fµν (x) ≡ ∂µ Aaν (x) − ∂ν Aaµ (x) − fbc a Abµ (x)Acν (x) . (3.13)

The equations of motion, gauge transformations and gauge algebra are

(D µ Fµν )a ≡ D µ a b Fbµν = 0 , (3.14)

δAaµ = (Dµ Λ)a ≡ Dµb

a
Λb , (3.15)
 
[δ(Λ1 ), δ(Λ2 )]Acµ = δ(Λ12 )Acµ = Dµd
c
fab d Λb1 Λa2 , (3.16)

so that Λc12 = fab c Λb1 Λa2 . The covariant derivatives Dµb a

and Dµa b in the adjoint
representation are
a
Dµb = δ a b ∂µ − fcb a Acµ ,
Dµa b = δa b ∂µ + fca b Acµ , (3.17)
a
where Dµb is applied to fields φb with an upper index b and Dµa b is applied to fields
φb with a lower index b. One has
Z Z
dd xφa Dµb
a b
φ =− dd x (Dµb a φa ) φb . (3.18)

The operator Rαi in Eq.(2.5) corresponds to D µa b . The covariant derivative satisfies

[Dµ , Dν ]a b = −fcb a Fµν

c
,

[Dµ , Dν ]a b = fca b Fµν

c
. (3.19)
In using compact notation, the spatial dependence as well as index dependence of
tensors needs to be specified. For local theories, the spatial dependence is proportional
to delta functions or a finite number of derivatives acting on delta functions. When
the spatial-temporal part of a tensor structure is a delta function, it is proportional
to the identity operator in x-space when regarded as an operator. In such cases, it is
convenient to drop explicitly such identity operators.
ji
Eq.(3.16) is in the form of Eq.(2.23) with Eαβ = 0 and Tabc
= fab c where two
identity operators or delta functions are implicit. One concludes that this example
constitutes a closed, irreducible gauge algebra.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 29

It is useful to verify the key equations for an irreducible closed algebra given in
Eqs.(2.33)-(2.36). The generator of gauge transformations is the covariant derivative
in Eq.(3.17). Using Eqs.(3.14) and (3.15), the Noether identity in Eq.(2.33) reads
Z
dd x (D µ Fµν )b (D ν C)b = 0 .

To verify this equation, integrate by parts, use the antisymmetry of Fµν in µ and ν,
use Eq.(3.19), and then make use of the antisymmetry of fcd a in c and d:
Z Z
d µ ν b
d x (D Fµν )b (D C) = − dd x (D ν D µ Fµν )b C b =
Z Z
1 1
− dd x ([D ν , D µ ] Fµν )b C b = − dd x fbd a Faνµ Fµν
d b
C =0 .
2 2
i
A straightforward computation of the 2Rα,j Rj C β C α term in the commutator alge-
 β
bra equation of Eq.(2.34) produces −2fba c ∂ µ C b − Adµ fde b C e C a which, after a lit-
 
tle algebra that makes use of Eq.(3.11), leads to D µc d fab d C b C a . Using this for
i
2Rα,j Rβj C β C α in Eq.(2.34), one concludes, as expected, that Tabc ji
= fab c and Eαβ = 0.
c c
When Tab = fab is used, the gauge-structure Jacobi equation in Eq.(2.35) leads to
3Adabc = −( LHS of Eq.(3.11)) so that Adabc = 0. Finally, the other consequence of
the Jacobi identity, namely Eq.(2.36), produces the tautology 0 = 0. All terms are
ji iδ kji
zero because the tensors Bαβγ , Dαβγ and Mαβγ are all zero. Higher-level equations
(which were not displayed in Sect. 2.4) are automatically satisfied because higher-level
tensors are identically zero.

3.3 Topological Yang-Mills Theory

In four-dimensions, the action for topological Yang-Mills theory [272, 35] is pro-
portional to the Pontrjagin index
Z
1 a ∗ µν
S0 = d4 x Fµν Fa , (3.20)
4
where the dual field strength ∗Fµν
a
is given by ∗Fµν
a
= 12 ǫµνρσ F aρσ , and where ǫ0123 = 1.
The interest in this system is its connection [272] to Donaldson theory [90].
This action is invariant under the gauge transformations in Eq.(3.15) because the
lagrangian is constructed as a group invariant of the field strength. In addition, since
the theory is topological, the transformation

δAaµ = εaµ , (3.21)

J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 30

leaves the action invariant, as a short calculation verifies. The two gauge transforma-
tions form a closed algebra since

[δ(Λ1 , ε1 ), δ(Λ2 , ε2)]Aµ = δ(Λ12 , ε12 )Aµ ,

 
Λc12 ≡ fab c Λb1 Λa2 , εcµ12 ≡ fab c εbµ1 Λa2 + Λb1 εaµ2 . (3.22)
However, the gauge generators are obviously off-shell linearly dependent since the
εµ transformations include ordinary gauge transformations when εµ = Dµ Λ. Since
the gauge transformations are not all independent, one has a reducible gauge theory
and the coefficients Zaα in Eq.(2.37) and Abaα introduced in Eq.(2.38) are non-zero. Of
course, the theory can be made irreducible by eliminating ordinary gauge transforma-
tions from the set of all transformations. However, for other theories it is not so easy
to reduce the full set to an irreducible subset without spoiling locality or relativistic
covariance and often it is convenient to formulate the theory as a reducible system.
Let us use topological Yang-Mills theory to illustrate the gauge-structure formal-
ism of Sect. 2.4. The field index i corresponds to both a gauge index and a vector
Lorentz index since the field is Aaµ : i ↔ aµ. There are two types of gauge transfor-
mations so that the gauge index α of Sect. 2.4 corresponds to the group index b in the
case of an ordinary gauge transformation or to the pair cµ in the case of a topological
gauge transformation: α ↔ (b, cµ). The generator of ordinary gauge transformations
is the covariant derivative in Eq.(3.17) and the generator of topological transforma-
tions is a delta function:

Rbaµ = D µa b , aµ
Rbν = δba δνµ . (3.23)

The null vectors, denoted as Zbα in Sect. 2.4, are

Zba = δba , Zbaµ = −D µa b . (3.24)

The number of null vectors is equal to the number of gauge generators. It is easily
verified that Eq.(2.41) holds since Rcbµ Zac η a +Rcν Za η = D µb c δac η a +δcb δνµ (−D νc a η a ) =
bµ cν a

0. This computation implies that Vaji = 0 in Eq.(2.41).

The non-zero structure constants are
c
Tab = fab c , cν
Taµ cν ν
b = Ta bµ = δµ fab
c
, (3.25)
c c cν cλ c
whereas Taµ b = Ta bµ = Tab = T  aµ bν
= Taµ bν = 0. As in the example of Sect.
3.2, Aδαβγ is zero since 3Adabc = − fae d fbc e + fbe d fca e + fce d fab e = 0. One also finds
Adν dν dν ν d
aµ b c = Aa bµ c = Aa b cµ = δµ Aabc = 0.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 31

jδ ji
Eq.(2.39) holds because Aδαβγ = Fαβγc
= Dαβγ = 0. Eq.(2.40) holds because Bαβγ
kji
and Mαβγ are also zero. Finally, Eq.(2.42) is valid as long as Gjγ
aβ = 0 and

Acab = −fab c , Aca bµ =0 . (3.26)

In verifying Eq.(2.42) it is useful to note that Za,bµ cν

= −δµν fab c , and Za,j
c
= 0 both
when j = b and when j = bµ. For Eq.(2.42), there are four cases to verify: (i) γ = c
and β = b, (ii) γ = c and β = bµ, (iii) γ = cν and β = b and (iv) γ = cν and β = bµ.
Case (i) gives −fbd c δad − δdc Adabµ = 0. Case (ii) is automatically zero because each
term in Eq.(2.42)
 is zero.
 Case
 (iii), after a little algebra, results
 in the expression
c ν a b c ν a b c d c d c d a b eν
fba ∂ η C + fab ∂ η C + fad fbe + fbd fea + fed fab η C A which is zero
because of the antisymmetry of fab c in ab and the Jacobi identity for the Lie algebra
structure constants in Eq.(3.11). For case (iv), one finds δµν fab c − δµν fba c = 0. In
short, the important non-zero gauge structure tensors are given in Eqs.(3.23) – (3.26).
Eqs.(2.33), (2.34) and (2.39) – (2.42) are all satisfied.
In a topological theory, the number of local degrees of freedom is zero. One finds
that ndof is 4N (for Aaµ ) minus 4N (for εaµ ) minus N (for Λa ) plus N (for the null
vectors in Eq.(3.24)). Hence, the net number of local degrees of freedom is zero.

3.4 The Antisymmetric Tensor Field Theory

Another example of first-stage reducible theory is the antisymmetric tensor gauge
a a a
theory. Consider a tensor field Bµν in four dimensions satisfying Bνµ = −Bµν whose
dynamics is described by the action [112, 242]
h i Z  
1 1 a µν
S0 Aaµ , a
Bµν = 4
d x Aaµ Aµa − Bµν Fa , (3.27)
2 2
where Aaµ is an auxiliary vector field. The field strength Fµν
a
is given in terms of
a
Aµ as in the Yang-Mills case via Eq.(3.13). The action is invariant under the gauge
transformations
a
δBκλ = ǫκλµν D µa b Λbν , δAaµ = 0 . (3.28)
The covariant derivative D µa b is given in Eq.(3.17). The equations of motion derived
from Eq.(3.27) are

∂r S0 ∂r S0
= −Faµν = 0 , = Aaµ + D νa b Bνµ
b
=0 . (3.29)
∂Bµνa ∂Aµa

In spite of the presence of Lie-algebra structure constants fab c , the model has an
γ
abelian gauge algebra, i.e., Tαβ = 0, due to the fact that the vector field, which
appears in the covariant derivative, does not transform.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 32

The gauge transformations (3.28) have an on-shell null vector. Indeed, taking

Λbν = D νb c ξ c , (3.30)

one finds using Eq.(3.19) that

1
a
δBκλ = ǫκλµν (D µ D ν )a b ξ b = − ǫκλµν fcb a F cµν ξ b ,
2
which vanishes on-shell, since F cµν = 0 when the equations of motion (3.29) are used.
Since the null vectors are independent, this theory is on-shell first-stage reducible.
It is instructive to determine the gauge-structure tensors and verify Eqs.(2.33),
(2.34) and (2.39) – (2.42). The field index i in Sect. 2 corresponds to aµν in the case
of the antisymmetric field B aµν and corresponds to aµ in the case of Aaµ . The gauge
index α of Sect. 2 corresponds to bν which are the indices of Λbν . The null index a of
Sect. 2 is a Lie algebra generator index. The generator of gauge transformations Rβi
is
aµν aµ
Rbλ = ǫµνρσ ησλ Dρb
a
, Rbλ =0 , (3.31)
where the second equation holds because Aaµ does not transform and ησλ is the flat
space-time metric. The null vectors Zaα are the covariant derivative operators

Zabλ = D λb a . (3.32)
cµν bλ
Using Eq.(2.41), one finds Rbλ Za = 21 ǫµνρσ fab c Fρσ
b
so that

Vabρσ cµν
= −ǫµνρσ δ bd fad c ,

Vaji = 0 , if j = bν or i = bν , (3.33)
that is, Vaji = 0 if i or j corresponds to the field index of Abν . The derived quantities
ij
Aδαβγ and Bαβγ are zero. Other gauge-structure tensors also vanish:
γ ji jδ c kji
Tαβ = Eαβ = Dαβγ = Fαβγ = Mαβγ = Adaβ = 0 . (3.34)

The gauge-structure equations are all satisfied. Eq.(2.33) holds because the action
is invariant under gauge transformations, as is easily checked. Eq.(2.41) is satisfied. In
fact, it was used above in Eq.(3.33) to obtain Vaji . Eqs.(2.39) and (2.40) are satisfied
because all the tensors entering these equations are zero. Each term in Eq.(2.34) is
i
zero: The first term Rα,j Rβj is zero because Rα,ji
= 0 when j = aµν, i.e., when j is
j
a field index of B , and Rβ = 0 when j = aµ, i.e., when j is a field index of Aaµ .
aµν

The other terms vanish because the structure tensors vanish. Likewise each term in
γ
Eq.(2.42) is zero: The first term Zα,j Rβj is zero because Zα,j
γ i
like Rα,j is zero when
j = aµν.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 33

Let N be the number of generators of G, i.e., the dimension of the Lie algebra
G. The number of degrees of freedom ndof is 4N for Aaµ plus 6N for Bµν a
minus
the number of gauge transformations 4N plus the number of null vectors N, so that
Eq.(2.20) reads ndof = 7N.

3.5 Abelian p-Form Theories

It is not hard to find an example of an L-th stage reducible theory. Let A be a
p-form and define F to be its field strength: F = dA where d is the exterior derivative.
For p + 1 less than the dimension d of spacetime, an action for this theory is
Z
1
S0 = − F ∧ ∗F , (3.35)
2
where ∗ is the dual star operation that takes a q-form into a d − q form and ∧ is the
wedge product. On basis q-forms, it is defined by
∗
(dxµ1 ∧ dxµ2 ∧ . . . ∧ dxµq ) =
1
εµ1 µ2 ...µd ηµq+1 νq+1 ηµq+2 νq+2 . . . ηµd νd dxνq+1 ∧ dxνq+2 ∧ . . . ∧ dxνd ,
(n − p)!
where ηµν is the flat space-time metric and εµ1 µ2 ...µd is the antisymmetric tensor
symbol. The case p = 1 corresponds to abelian Yang-Mills theory. Using dd = 0, one
sees that the action is invariant under the gauge transformation

δA = dλp−1 , (3.36)

where λp−1 is a p − 1 form. This gauge transformation has its own gauge invariance.
In fact, there is a tower of gauge invariances for gauge invariances given by

δλp−1 = dλp−2 ,
..
.
δλ1 = dλ0 ,

where λq is a q-form. Hence, the theory is p − 1 stage reducible. The number of

degrees of freedom is
! ! ! ! ! !
d d d d d d−1
ndof = − + − . . . − (−1)p + (−1)p = . (3.37)
p p−1 p−2 1 0 p
 
d
Note that q
is the dimension of the space of q-forms in d dimensional space-time.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 34

αs −1
The gauge generators at the s-th stage, Rsα s
, correspond to the exterior deriva-
(p−1−s)
tive d acting on the space of (p − 1 − s) forms, d :

R0 ↔ d(p−1) ,
(p−2)
R1 ↔ d ,
..
.
Rp−1 ↔ d(0) .
iαs−2
Because dd = 0, Eq.(2.19) holds off-shell and all Vsα s
are zero. With the excep-
tion of the gauge generators, all gauge-structure tensors are zero.
If n(p + 1) = d = dimension of space time, a topological term can be added to the
action Z
∆S0 = F | ∧F ∧ {z. . . ∧ F} . (3.38)
n terms

For n = 2 one can consider the quantization of ∆S0 alone if d = 2 (p + 1). This would
be another example of a topological theory.
Abelian p-form theories provide a good background for covariant open string field
theory, a non-abelian generalization of p-form theory which is infinite-stage reducible.

3.6 Open Bosonic String Field Theory

The covariant d = 26 open string field theory was obtained by E. Witten [270].
It resembles a Chern-Simons theory. The fundamental object is a string field A.
Although one can proceed without a detailed understanding of A, A can be expanded
as a series in first-quantized string states whose coefficients are ordinary particle fields.
Each member of this infinite tower of states corresponds to a particular vibrational
mode of the string. In this manner, string theory is able to incorporate collectively
many particles. For example, the open bosonic string possesses a tachyonic scalar, a
massless vector field, and numerous massive states of all possible spins. For reviews on
open bosonic string field theory, see refs.[215, 168, 247]. Also useful is the discussion
in Sect. 7.7, where the first quantization of the bosonic string is treated.
Covariant open string field theory can be formulated axiomatically [270, 271].
Fields are classified according to their string ghost number. If the string ghost number
of B is g(B) = p then we say that B is a string p-form in a generalized sense. The
ingredients of abstract string theory are a derivation Q, a star operation ∗ which
R
combines pairs of fields to produce a new field, and an integration operation which
R
yields a complex number B for each integration over a string field B. These objects
satisfy five axioms:
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 35

(1) The nilpotency of Q: QQ = 0.

R
(2) Absence of surface terms in integration: QA = 0. This axiom is
equivalent to an integration-by-parts rule.

(3) Graded distributive property of Q across ∗:

Q (A ∗ B) = QA ∗ B + (−1)g(A) A ∗ QB.

(4) Associativity of the star product: (A ∗ B) ∗ C = A ∗ (B ∗ C).

(5) Graded commutativity of the star product under the integral:

R R
A ∗ B = (−1)g(A)g(B) B ∗ A.

The ghost number of a star product of fields is the sum of their ghost numbers:
g(A ∗ B) = g(A) + g(B). The derivation Q increases the ghost number by 1: g(QA) =
g(A) + 1. In some circles, including refs.[270, 271], the ghost number is shifted by
−3/2 so that A has ghost number −1/2 instead of 1.
The axioms are satisfied for non-abelian Chern-Simons theory in three dimensions.
The field is a non-abelian vector potential which is converted into a Lie-algebra-
valued 1-form by multiplying by dxµ : Aa b ≡ Aaµb dxµ ≡ Asµ (Ts )a b dxµ , where the Ts
are a set of matrix generators for the Lie group. The derivation Q is the exterior
derivative d. The star product is the wedge product and a matrix multiplication:
(A ∗ B)a b = Aa c ∧ B c b . Integration is an integral over a three-dimensional manifold
R R
M without boundary and a trace over the Lie-algebra indices: A ≡ M Aa a . Axiom
(1) is satisfied because dd = 0. Axiom (2) holds because M has no boundary. Axiom
(3) is satisfied because the exterior derivative d is graded distributive across the wedge
product. Axiom (4) holds because both the wedge product and matrix multiplication
are associative. Finally, axiom (5) is satisfied because of the graded antisymmetry of
the wedge product and the cyclic property of a trace: T r(MN) = T r(NM) for any
R
two matrices M and N. In this example and in open string field theory, B = 0
unless B is a string 3-form.
The action for the string 1-form A is given by
Z Z
1 1
S0 [A] = A ∗ QA + A∗A∗A . (3.39)
2 3
Using the above five axioms it is straightforward to show that the action is invariant
under the gauge transformation

δA = QΛ0 + A ∗ Λ0 − Λ0 ∗ A , (3.40)
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 36

where Λ0 is any string 0-form. The equation of motion for A reads

F ≡ QA + A ∗ A = 0 . (3.41)

For the open bosonic string, Q is the BRST charge QBRST of the first-quantized
theory promoted to an operator by second quantization (see Eq.(7.51) with ρ =
1). The ∗ product is intuitively described as follows. Let σ be the parameter that
determines a point on the string, so that σ = 0 corresponds to one endpoint and
σ = π corresponds to the other endpoint. Divide the string in two at σ = π/2 and
call the two halves the left and right halves. Let C = A ∗ B. Then ∗ “glues” the
left half of A to the right half of B by delta functions and what remains is C so that
the left half of C is the right half of A and the right half of C is the left half of B.
See Fig. 1. The star operator can be thought of as matrix multiplication if the range
0 ≤ σ ≤ π/2 of points of the string is associated with one matrix index and the range
π/2 ≤ σ ≤ π is associated with the other matrix index. The integral operator is a
delta function equating the left and right halves. See Fig. 2. In the matrix analogy,
R
it is the trace. Precise definitions of Q, ∗ and in terms of the vibrational modes of
the string, i.e., particle excitations, can be found in refs.[68, 214, 142, 143].

σA = π σB = 0

A B
@
σA = 0 C @@
σB = π
σC = π σC = 0
Figure 1. The String Star Product

π
σ=
 2

∫ ←→

σ=0 σ=π
Figure 2. The String Integral

Unlike the three-dimensional Chern-Simons theory, the open bosonic string theory
possesses q-forms for q < 0. As a consequence there are on-shell gauge invariances
for gauge invariances at all levels:

δΛ0 = QΛ−1 + A ∗ Λ−1 + Λ−1 ∗ A ,

J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 37

δΛ−1 = QΛ−2 + A ∗ Λ−2 − Λ−2 ∗ A ,

..
.

δΛ−n = QΛ−n−1 + A ∗ Λ−n−1 − (−1)n+1 Λ−n−1 ∗ A , (3.42)

..
.
where Λq is a string q-form. To verify these gauge invariances, note that if Λ−n+1 =
QΛ−n + A ∗ Λ−n − (−1)n Λ−n ∗ A and if one changes Λ−n by δΛ−n = QΛ−n−1 + A ∗
Λ−n−1 − (−1)n+1 Λ−n−1 ∗ A then Λ−n+1 changes by δΛ−n+1 = (QA + A ∗ A) ∗ Λ−n−1 −
Λ−n−1 ∗ (QA + A ∗ A), as a little algebra reveals. As a consequence, this change
δΛ−n+1 in Λ−n+1 is zero when the equations of motion in Eq.(3.41) are invoked. One
of the confusing aspects of quantizing the open bosonic string is the following. In the
free theory, the second term in Eq.(3.39) is dropped. Then the theory is invariant
under the tower of gauge transformations δΛ−n = QΛ−n−1 for n ≥ −1 where Λ1 ≡ A.
The gauge invariances of gauge invariances hold off-shell because of the nilpotency
of Q. So the quantization of the free-theory proceeds without any non-zero on-shell
structure tensors. In contrast, for the interacting theory there exist non-zero on-shell
tensors in the gauge structure.
Indeed, let us determine the gauge structure of the theory. The algebra is closed
(1) (2)
because the commutator of two gauge transformations Λ0 and Λ0 is equivalent to a
(12) (1) (2) (2) (1) ji
single gauge transformation given by Λ0 = Λ0 ∗ Λ0 − Λ0 ∗ Λ0 . Hence, Eαβ =0
γ
in Eq.(2.23) and Tαβ is the matrix element of the ∗ operator among three string
states. This is also known as the three-point vertex function and has been computed
in refs.[68, 214, 142, 143]. The gauge generators Rs in Eq.(2.19) correspond to the
covariant derivative operator D(−s) restricted to the space of −s forms,

Rs ↔ D(−s) , (3.43)

where D(q) is defined by

D(q) Bq = QBq + A ∗ Bq − (−1)q Bq ∗ A , (3.44)

when acting on any q-form Bq and A is the string gauge field. Eq.(2.19), when applied
to a −s form ghost η (s) , reads

D(−s+1) D(−s) η (s) = F ∗ η (s) − η (s) ∗ F , (3.45)

iαs−2
where F is given in Eq.(3.41). Hence, the Vsα s
in Eq.(2.19) are not zero and are
given by the commutator of the star operator restricted to the space of −s forms:

S0,i Vsi η (s) ↔ F ∗ η (s) − η (s) ∗ F . (3.46)

J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 38

3.7 The Massless Relativistic Spinning Particle

In this subsection, we consider an example with an anti-commuting gauge param-
eter, i.e, a system with local supersymmetry. Local supersymmetric theories are also
known as supergravity theories. In such theories, the lagrangians and transformation
laws are usually complicated. Since our purpose is not to review supersymmetry,
we treat one of the simplest examples: the massless relativistic spinning particle
[10, 41, 65]. It is a (0 + 1)-dimensional supergravity theory.
We begin with the massless flat-space version of the spinless relativistic particle,
namely, the theory discussed in Sect. 3.1 and governed by the action in Eq.(3.1) with
e = 1 and m = 0. To obtain local supersymmetry, one can proceed in two ways. One
can first implement general coordinate invariance and then supersymmetry, or one
can first implement supersymmetry and then general coordinate invariance:
Flat Space Theory
General Coordinate Invariance @ Supersymmetry
@
R
@
Gravity Supersymmetric Theory
@
Supersymmetry@ General Coordinate Invariance
R
@
Supergravity

The result of such a program is a supergravity.

Consider the first step of the left path of the above diagram. To implement
general coordinate invariance, the Poincaré group must be realized locally. In 0 + 1-
dimensions, the Poincaré group consists of translations: τ → τ + ε. The local version
of this transformation is τ → τ + ε(τ ), that is, ε becomes a function of τ . The fields
for the spinless relativistic particle are the xµ (τ ) of Sect. 3.1. Under local translations
they change as xµ (τ + ε (τ )) − xµ (τ ) = ε (τ ) ẋµ (τ ) + . . .. Hence, the transformation
law for xµ (τ ) is the one given in Eq.(3.6).
As in any gauge theory, it is necessary to introduce a gauge field and promote
ordinary derivatives to covariant derivatives [256, 276, 177]: ∂τ → Dτ so that covariant
derivatives transform in the same way as fields, namely

δ (Dτ xµ (τ )) = ε (τ ) ∂τ (Dτ xµ (τ )) . (3.47)

The gauge field is contained in the einbein e of Sect. 3.1. It transforms as in Eq.(3.6).
The covariant time derivative Dτ is
1
Dτ xµ (τ ) ≡ ∂τ xµ (τ ) . (3.48)
e
Using Eq.(3.6) it is a simple exercise to verify that Dτ xµ (τ ) transforms as in Eq.(3.47).
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 39

Any function F of xµ and Dτ xµ transforms in the same way as xµ , i.e.,

δF = ε (τ ) ∂τ F , (3.49)

as is easily checked. Given any such F , the action

Z
dτ eF (xµ , Dτ xµ ) (3.50)

is invariant since
Z Z Z
dτ δ (eF ) = dτ (∂τ (εe) F + εe∂τ F ) = dτ ∂τ (εeF ) −→ 0 ,

if ε(τ ) goes to zero at τ = ±∞. The action in Eq.(3.1) corresponds to

1 
F = Dτ xµ (τ ) Dτ xµ (τ ) − m2 . (3.51)
2
One could also add to Eq.(3.51) an arbitrary potential term V (xµ (τ )) and the theory
would still be locally coordinate invariant; however, the global symmetries xµ →
xµ + aµ , where aµ are constants, would be broken. Requiring such a symmetry forces
V (xµ (τ )) to be constant.
The above describes the general idea in constructing theories with general coordi-
nate invariance. In higher-dimensions, e has more components and a spin connection
must be introduced for fields with spin. In short, the procedure labelled “general
coordinate invariance” in the above diagram corresponds to introducing gravitational
gauge fields and promoting ordinary derivatives to covariant derivatives. For more
details, see the example in Sect. 3.8.
Let us consider the process of supersymmetrizing the massless version of the spin-
less relativistic particle, the example presented in Sect. 3.1. The action is given in
Eq.(3.1) with m = 0 and e = 1. It is necessary to have m = 0 because there does
not exist a supersymmetric generalization of the m2 term. The fields of the spinless
relativistic particle are xµ which are functions of the proper time variable τ . To su-
persymmetrize the theory, one goes from ordinary space to superspace by enlarging
the coordinate system from τ to τ and θ: τ → (τ, θ), where the anticommuting coor-
dinate θ is the supersymmetric partner of τ . Fields are then promoted to superfields
by making them functions of θ:

xµ (τ ) → X µ (τ, θ) ≡ xµ (τ ) + θψ µ (τ ) , (3.52)

where the right-hand side of Eq.(3.52) is a Taylor series expansion in θ. It terminates

at θ because θθ = 0. The θ-component of the superfield X µ is ψ µ . It is the fermionic
partner of xµ .
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 40

Although our purpose is not to review supersymmetry, we present enough back-

ground to make this section self-contained. For extensive reviews on supersymmetry
see refs.[124, 267]. Superfields form an algebra. This means that they can be added
and multiplied. Given two superfields A1 = a1 + θb1 and A2 = a2 + θb2 their sum is
A1 + A2 =  (a1 + a2 ) + θ (b1 +  b2 ) and their product is A1 A2 = a1 a2 + a1 θb2 + θb1 a2 =
ǫ(a1 )
a1 a2 + θ b1 a2 + (−1) a1 b2 . Superfields can be manipulated in a manner similar
to complex numbers. One only has to be careful about minus signs arising from quan-
tities that anticommute. Supersymmetry calculus is also straightforward. In addition
to taking derivatives with respect to τ one can take derivatives with respect to θ. In
∂
a natural way, one defines ∂θ ≡ ∂θ so that ∂θ A = b when A = a + θb. Note that ∂θ is
a left derivative. Integration is defined as the operation that selects the θ component
R
of a superfield and serves the same function as ∂θ : dθA = ∂θ A = b. For more on
Grassmann integration and supermanifolds, see refs.[38, 40, 188, 85, 39].
The generator of supersymmetry transformations is denoted by Q (not to be
confused with the BRST operator of the previous subsection). It is also convenient
to use instead of ∂θ a generalized θ-derivative Dθ . More explicitly Dθ and Q are

Dθ ≡ ∂θ − iθ∂τ ,

Q ≡ ∂θ + iθ∂τ . (3.53)
The 1-dimensional super-algebra is

{Q, Q} = 2QQ = 2i∂τ ≡ 2H ,

[H, Q] = 0 . (3.54)
These equations are easily verified using Eq.(3.53). In theories for which a super-
symmetry charge Q exists, the hamiltonian necessarily has a non-negative spectrum
as a consequence of H = QQ ≥ 0. In addition, [H, Q] = 0 implies that there is
a fermionic state for every bosonic state and vice-versa, except possibly for a zero
energy state. The (anti)commutators involving Dθ are {Q, Dθ } = 0, [H, Dθ ] = 0, and
{Dθ , Dθ } = −2H.
The supersymmetry transformations are defined by δA = iξQA where ξ is an
anticommuting parameter. For X µ , a simple calculation gives δX µ = iξQX µ =
iξψ µ + θξ∂τ xµ . Given that δX µ is defined by δX µ ≡ δxµ + θδψ µ , one obtains for the
transformation rules for the components of X µ

δxµ = iξψ µ ,

δψ µ = ξ∂τ xµ . (3.55)
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 41

The reason for using Dθ in lieu of ∂θ is that Dθ A transforms as a superfield if A

transforms as such, since Dθ commutes with Q: δ (Dθ A) = Dθ (iξQA) = iξQ (Dθ A).
The same statement holds for ∂τ A since [∂τ , Q] = −i [H, Q] = 0 so that one does not
need to generalize ∂τ .
Supersymmetric-invariant
 actions
 are constructed from the θ component of a su-
perfield: Note that δ A|θ component = ξ∂τ ( A|θ=0 ) so that if one integrates this com-
ponent over τ to obtain an action S0 , this action is invariant if fields fall off sufficiently
fast at τ → ±∞. Given that ∂τ X µ , Dθ X µ and X µ are all superfields and that su-
perfields form an algebra,the θ component of any function  of these fields transforms
µ µ µ
R
as a total τ derivative: δ F (∂τ X , Dθ X , X )|θ component = ξ∂τ ( F |θ=0 ). Since dθ
picks out the θ component of a superfield, an invariant action is
Z Z
S0 = dτ dθF (∂τ X µ , Dθ X µ , X µ ) . (3.56)

The supersymmetric generalization of the action for the massless relativistic particle
consists in taking F in (3.56) as F = 2i Dθ X µ ∂τ Xµ since, when written in component
fields, it contains the correct kinetic energy term for xµ :
Z Z Z
i µ 1
S0 = dτ dθDθ X ∂τ Xµ = dτ (∂τ xµ ∂τ xµ − iψ µ ∂τ ψµ ) . (3.57)
2 2
It is easily checked that Eq.(3.57) is invariant under the transformations in Eq.(3.55).
One could add interactions by using a more general F (∂τ X µ , Dθ X µ , X µ ).
The final goal of this subsection is to implement both local translational and local
supersymmetry and construct a supergravity theory. Following the procedures above,
we promote the einbein e(τ ) to a superfield E: e (τ ) → E (τ, θ) where

E (τ, θ) ≡ e (τ ) + θχ (τ ) . (3.58)

Here, χ is the superpartner of e. The infinitesimal gauge parameters associated with

general coordinate invariance and supersymmetry are grouped into one superfield η t
via
η t (τ, θ) ≡ ε (τ ) + θξ (τ ) . (3.59)
Using η t and the θ-partner to η t , given by

η θ (τ, θ) ≡ iE∂θ η t = ieξ (τ ) + iθχξ (τ ) , (3.60)

one writes the transformation rule for a superfield such as A (τ, θ) as

δA = η t (τ, θ) ∂τ A + η θ (τ, θ) ∂θ A . (3.61)

Notice that Eq.(3.61) reduces to the correct transformation law δA = ε (τ ) ∂τ A when

ξ = 0. Furthermore, when χ → 0, ε (τ ) → 0, e → 1, and ξ is not longer a function of
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 42

τ , the right-hand side of Eq.(3.61) becomes (iξ∂θ + θξ∂τ ) A = iξQA, which is the flat
space supersymmetry transformation. Hence Eq.(3.61) reduces to expected results in
these two limits. The transformation law for E must be generalized from δe = ∂τ (εe).
As will become clear below when we construct invariant actions, one needs
   
δE = ∂τ η t E − ∂θ η θ E . (3.62)

With this transformation law for E, it is straightforward to find covariant deriva-

tive operators Dτ and Dθ that maintain the supersymmetry transformation law
in Eq.(3.61). One requires δ (Dτ A) = η t ∂τ (Dτ A) + η θ ∂θ (Dτ A) and δ (Dθ A) =
η t ∂τ (Dθ A) + η θ ∂θ (Dθ A). A solution is
1
Dτ A ≡ (∂τ A − i∂θ E∂θ A) ,
E
Dθ A ≡ ∂θ A − iθDτ A . (3.63)
When acting on the superfield X µ , one finds
 
1 θ χ
Dτ X µ = (∂τ xµ − iχψ µ ) + ∂τ ψ µ − ∂τ xµ ,
e e e
θ
Dθ X µ = ψ µ − i (∂τ xµ − iχψ µ ) . (3.64)
e
Again, superfields form an algebra so that any function F of Dτ X µ , Dθ X µ and X µ
transforms as a superfield in curved space:

δF (Dτ X µ , Dθ X µ , X µ ) = η t ∂τ F + η θ ∂θ F .

It is now straightforward to show that

Z Z
S0 = dτ dθEF , (3.65)

is an invariant action since Z Z

δ dτ dθEF =
Z Z       
dτ dθ ∂τ η t E − ∂θ η θ E F + E η t ∂τ F + η θ ∂θ F
Z Z     
= dτ dθ ∂τ η t EF − ∂θ η θ EF

is a total derivative. The transformation rule for E in Eq.(3.62), including the mi-
nus sign on the right-hand side, was deliberately chosen so that EF would lead to
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 43

an invariant super-lagrangian. Taking F = 2i Dθ X µ Dτ Xµ produces the extension of

Eq.(3.57) with general coordinate invariance:
Z Z
i
S0 = dτ dθEDθ X µ Dτ Xµ =
2
Z  
1 1 2iχ µ
dτ ∂τ xµ ∂τ xµ − iψ µ ∂τ ψµ − ψ ∂τ xµ . (3.66)
2 e e
By construction, one is guaranteed that Eq.(3.66) is invariant under the transfor-
mations in Eqs.(3.61) (when A = X µ ) and (3.62). In components, these equations
read
δxµ = ε∂τ xµ + ieξψ µ ,
δψ µ = ε∂τ ψ µ + ξ (∂τ xµ − iχψ µ ) ,
δe = ∂τ (εe) + 2ieξχ , (3.67)
δχ = ∂τ (εχ + ξe) .
A good exercise is to check explicitly that Eq.(3.66) is invariant under Eq.(3.67).
Since the transformations rules in Eqs.(3.61) and (3.62) are defined independent
of an action, the algebra should close off-shell. The only thing to check is that the
commutator of two transformations produces an effect on E similar to the one on X µ .
A little algebra reveals that

[δ1 , δ2 ] φi = δ12 φi , (3.68)

where φi = xµ , ψ µ , e or χ, and δ12 corresponds to a transformation with

ε12 = ε2 ε̇1 − ε1 ε̇2 + 2ieξ2 ξ1 ,

ξ12 = ε2 ξ˙1 − ε1 ξ˙2 + ε̇1 ξ2 − ε̇2 ξ1 + 2iχξ2 ξ1 . (3.69)

The algebra closes off-shell but has field-dependent structure constants due to the
presence of e and ξ in Eq.(3.69). In the non-supersymmetric limit for which ξ1 =
ξ2 = χ = 0, the spinless-relativistic-particle algebra in Eq.(3.7) is reproduced. In the
flat-space limit for which χ = 0, e = 1, and gauge parameters do not depend on τ ,
Eq.(3.69) reduces to the flat space supersymmetry algebra in Eq.(3.54).
In four dimensions, the construction of supergravity theories is considerably more
complicated. Several gravitational superfields enter. The interested reader can find
introductions to four-dimensional supergravity in refs.[258, 124, 267]. The formalism
presented here is the one-dimensional analog of the covariant-Θ formalism of Chapters
XX and XXI of ref.[267].
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 44

3.8 The First-Quantized Bosonic String

The open bosonic string is the two-dimensional generalization of the relativistic
particle considered in Sect. 3.1. It is a general-coordinate-invariant theory. As such,
it can be regarded as a certain type of two-dimensional gravity theory.
Let us review the construction of general-coordinate-invariant theories involving
some scalar fields Ai . In what follows, all the Ai behave similarly and we drop the
superscript index i. We use the vielbein formulation. The gauge fields which im-
plement local coordinate invariances are vielbeins ea m , where a = 0, . . . , d − 1 and
m = 0, . . . , d − 1. These indices take on d values corresponding to time and space
coordinates. The index m is the Einstein index associated with local coordinate trans-
formations. The index a is the tangent-space Lorentz index used for implementing
local Lorentz transformations. The inverse metric g mn is the inverse of the metric
gmn : g mp gpn ≡ δnm . It is related to the vielbein via g mn!≡ η ab ea m eb n , where η ab is
−1 0
the flat-space metric. For d = 2, η ab = ηab = . Lorentz indices a, b, c,
0 1
etc. are raised and lowered using η ab and ηab . Einstein indices are raised and lowered
using g mn and gmn . Hence, the inverse vielbein ea m is ea m ≡ gmn η ab eb n and satisfies
ea n ea m = δnm and ea m eb m = δba , as one can easily check. The metric can be expressed
in terms of ea m via gmn = ηab ea m eb n .
Under an infinitesimal local translation in the m-th direction by εm , a scalar field
A changes by δA = A (xm + εm ) − A (xm ) or

δA = εm ∂m A . (3.70)

One would like to find covariant derivatives Da A that transform in the same way:
δ(Da A) = εm ∂m (Da A). The vielbein ea m allows one to do this. Letting

Da A = ea m ∂m A , (3.71)

and requiring the correct transformation law for Da A, one finds that ea m must trans-
form as
δea m = εn ∂n ea m − ea n ∂n εm . (3.72)
Any function F of A and Da A will transform like A: δF (A, Da A) = εm ∂m F . Suppose
one can find a density e whose transformation law is

δe = ∂m (εm e) . (3.73)

Then eF transforms as a total derivative δeF = ∂m (εm eF ) and is a candidate for an

action density. When the action density is integrated over space-time, it is invariant
under local coordinate transformations as long as gauge parameters vanish at infinity.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 45

The density e can be constructed from the determinant of the vielbein det (e. . ).
Note that det (e. . ) transforms as
δ det (e. . ) = δea m × cofactor matrix of ea m = δea m ea m det (e. . ) .
The derivative of det (e. . ) is
∂n det (e. . ) = ∂n ea m × cofactor matrix of ea m = ∂n ea m ea m det (e. . ) .
Using these results, a short calculation reveals that
1
e≡ .
= det (e. . ) (3.74)
det (e. )
transforms as in Eq.(3.73). The relativistic particle, presented as the example in Sect.
3.1, is a (0 + 1)-dimensional theory where the field e in that section is the same as
the density e defined here: e = det (e. . ) = e0 0 = 1/e0 0 .
In contrast to the metric gmn , which has d(d+1)/2 degrees of freedom, the vielbein
ea has d2 degrees of freedom. To compensate for this difference, one requires that the
m

theory be invariant under local Lorentz transformations that act on Lorentz indices
a, b, etc.. The transformation laws are
δea m = (εM)a b eb m , δA = 0 , (3.75)
b
where (εM)a b stands for 21 (εcd Mcd )a . Here, the d(d−1)/2 parameters εcd satisfy εcd =
−εdc and characterize the size of the infinitesimal-local-Lorentz transformations. The
d(d − 1)/2 matrix generators Mab satisfy the Lorentz algebra [Mab , Mcd ] = −ηac Mbd +
ηad Mbc + ηbc Mad − ηbd Mac and are antisymmetric in the lower indices c and d: Mcd =
−Mdc . When we write (Mcd )a b , a and b are matrix indices whereas c and d label
the different Lorentz transformations: Mcd produces an infinitesimal transformation
in the c–d plane. One explicit realization of Mcd is (Mcd )a b = ηac δdb − ηad δcb . The
extra d(d − 1)/2 gauge invariances guarantees that the physical numbers of degrees
of freedom in ea m and gmn are the same.
Given a number of fields Vai , with Lorentz indices, local invariants are constructed
by using the flat space metric η ab via η ab Vai Vbj or the flat space antisymmetric εa0 a1 ...ad−1
symbol via εa0 a1 ...ad−1 Vai00 Vai11 . . . Vaid−1
d−1 . A locally general coordinate and Lorentz in-

variant action for a scalar field A is

Z
1
S0 = − dt dd−1 x eη ab Da ADb A . (3.76)
2
For the rest of this subsection, we set d = 2. When the dimension of space-
time is 1 + 1, the action Eq.(3.76) is also classically invariant under local conformal
transformations that scale the vielbein by a spacetime dependent function ε:
δea m = εea m , δA = 0 . (3.77)
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 46

We thus have three types of local transformations: translations, Lorentz rotations and
scaling. Let us respectively associate ε. , ε.. and ε with each of these. In other words,
we use the number of indices on the infinitesimal parameter to distinguish the trans-
formations. The algebra of gauge transformations is as follows. The commutator of
two local translations is a local translation so that they form a subalgebra. Likewise
the commutator of two local Lorentz transformations is a local Lorentz transforma-
tion. The commutator of a translation and a Lorentz rotation is a Lorentz rota-
tion. Local translations and Lorentz rotations also form a subalgebra. Local scaling
transformations commute. They also commute with local Lorentz transformations.
Finally, the commutator of a scaling transformation and a translation produces a
scaling transformation. These last few statements correspond to

[δ (ε.1 ) , δ (ε.2 )] = δ (ε.12 ) , where εm n m n m

12 = ε2 ∂n ε1 − ε1 ∂n ε2 ,
 
[δ (ε..1 ) , δ (ε..2 )] = δ (ε..12 ) , where εab ac bd ac bd
12 = ε1 ε2 − ε2 ε1 ηcd ,
[δ (ε..1 ) , δ (ε.2 )] = δ (ε..12 ) , where εab n ab
12 = ε2 ∂n ε1 ,
[δ (ε1 ) , δ (ε2 )] = 0 , (3.78)
[δ (ε1 ) , δ (ε..2 )] = 0 ,
[δ (ε1 ) , δ (ε.2 )] = δ (ε12 ) , where ε12 = εn2 ∂n ε1 .
Equation (3.78) is the gauge algebra for the bosonic string. It is a closed irreducible
algebra with field-independent commutator structure constants and hence a Lie alge-
bra.
The scalar fields for the D = 26 bosonic string are target space coordinate X µ
where µ ranges from 0 to 25. The action is Eq.(3.76) for each of these fields
Z Z
1 π
S0 = − dτ dσeη ab ηµν Da X µ Db X ν , (3.79)
2 0

where ηµν is the flat-space 25 + 1 metric with signature (−1, 1, 1, . . . , 1), Da is

given in Eq.(3.71), and e = 1/(eτ τ eσ σ − eτ σ eσ τ ). The world-sheet coordinates are
denoted by τ and σ instead of t and x. Hence the first-quantized bosonic string
is a scale-invariant two-dimensional gravity theory on a finite spatial region since σ
ranges from 0 to π. For the closed bosonic string, the fields X µ are periodic in σ,
whereas for the open bosonic string, one requires ∂σ X µ to vanish at σ = 0 and σ = π.
Eq.(3.79) is invariant under the general coordinate transformations in Eqs.(3.70) and
(3.72), under the Lorentz gauge transformations in Eq.(3.75), and under the local scale
transformations in Eq.(3.77) where A stands for X µ ; however, the local translation
parameters εm must obey the same boundary conditions as X µ . It is a useful exercise
to check explicitly the gauge invariances of Eq.(3.79).
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 47

There are four gauge invariances, two translations, one Lorentz boost and one
scaling transformation. It is necessary for the target space to be 26 dimensions in
order to avoid an anomaly in one of these invariances. This is discussed in Sect. 9.3.
Hence in D =!26, the components of ea m can be fixed to constant values. Often
1 0
ea m = is used. Then Eq.(3.79) becomes a free field theory. However, in
0 1
gauge fixing, one should introduce ghosts. These ghosts play a somewhat minor role
in the first-quantized theory. However, in the second-quantized formulation, namely
the string field theory of Sect. 3.6, the ghosts as well as X µ (σ) become coordinates
of the string field A in Eq.(3.39).
Eq.(3.79) is classically equivalent to the usual Nambu-Goto action [197, 138].
Using the equations of motion of ea m to eliminate the vielbein, one finds, after some
algebra, that
q
1 ab
eη ηµν Da X µ Db X ν → (∂τ X µ ∂σ Xµ )2 − (∂τ X µ ∂τ Xµ )(∂σ X ν ∂σ Xν ) ,
2
which is the familiar Nambu-Goto action density.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 48

4 The Field-Antifield Formalism

Consider the classical system defined in Sect. 2, described by the action S0 [φi ] and
having gauge invariances. The ultimate goal is to quantize this theory in a covariant
way. The field-antifield formalism was developed to achieve this aim. In this section,
we present the field-antifield formalism at the classical level. In brief, it involves five
steps:

(i) The original configuration space, consisting of the φi , is enlarged to

include additional fields such as ghost fields, ghosts for ghosts, etc.. One
also introduces the antifields of these fields.

(ii) On the space of fields and antifields, one defines an odd symplectic
structure ( , ) called the antibracket.

(iii) The classical action S0 is extended to include terms involving ghosts

and antifields and is denoted by S.

(iv) The classical master equation is defined to be (S, S) = 0.

(v) Finally, one finds solutions to the classical master equation subject to
certain boundary conditions.

The key result of this section is that the solution in steps (iv) and (v) leads to a
set of equations containing all relations defining the gauge algebra and its solution.
The action S is the generating functional for the structure functions. Hence, the field-
antifield formalism is a compact and efficient way of obtaining the gauge structure
derived in Sect. 2.

4.1 Fields and Antifields

Suppose a theory is irreducible with m0 gauge invariances. Then, at the quantum
level, m0 ghost fields are needed. As in Sect. 2.4, it is useful to introduce these
ghost fields at the classical level. Hence, the field set ΦA is ΦA = {φi , C0α0 } where
α0 = 1, . . . , m0 . If the theory is first-stage reducible, there are gauge invariances for
gauge invariances and hence there are ghosts for ghosts. If there are m1 first-level
gauge invariances then, to the above set, one adds the ghost-for-ghost fields C1α1 where
α1 = 1, . . . , m1 . In general for an L-th stage reducible theory, the set of fields ΦA ,
A = 1, . . . , N, is
n o
ΦA = φi , Csαs ; s = 0, . . . , L; αs = 1, . . . , ms . (4.1)
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 49

An additive conserved charge, called ghost number, is assigned to each of these fields.
The classical fields φi are assigned ghost number zero, whereas ordinary ghosts have
ghost number one. Ghosts for ghosts, i.e., level-one ghosts, have ghost number two,
etc.. Similarly, ghosts have opposite statistics of the corresponding gauge parameter,
but ghosts for ghosts have the same statistics as the gauge parameter, and so on,
with the statistics alternating for higher-level ghosts. More precisely,

gh [Csαs ] = s + 1 , ǫ(Csαs ) = ǫαs + s + 1 (mod 2) . (4.2)

Next, one introduces an antifield Φ∗A , A = 1, . . . , N, for each field ΦA . The ghost
number and statistics of Φ∗A are
h i  
gh [Φ∗A ] = −gh ΦA − 1 , ǫ (Φ∗A ) = ǫ ΦA + 1 (mod 2) , (4.3)

so that the statistics of Φ∗A is opposite to that of ΦA .

At this stage, the antifields do not have any direct physical meaning. They are
only used as a mathematical tool to develop the formalism. However, they can be in-
terpreted as source coefficients for BRST transformations. This is made clear from the
discussion of the effective action. See Sect. 8.4. For computing correlation functions,
S-matrix elements and certain quantities, antifields are eliminated by a gauge-fixing
procedure.
The set of fields and antifields just introduced is called the classical basis.3 When
gauge-fixing is considered in Sect. 6, another basis, known as the gauge-fixed basis,
can be introduced. Then, the implications of the classical master equation change.
This is discussed in Sect. 6.6.

4.2 The Antibracket

In the space of fields and antifields, the antibracket ref.[277, 24] is defined by
∂r X ∂l Y ∂r X ∂l Y
(X, Y ) ≡ − . (4.4)
∂Φ ∂ΦA ∂Φ∗A ∂ΦA
A ∗

Many properties of (X, Y ) are similar to a graded version of the Poisson bracket, with
the grading of X and Y being ǫX + 1 and ǫY + 1 instead of ǫX and ǫY . The antibracket
satisfies
(Y, X) = −(−1)(ǫX +1)(ǫY +1) (X, Y ) ,
((X, Y ) , Z) + (−1)(ǫX +1)(ǫY +ǫZ ) ((Y, Z) , X) + (−1)(ǫZ +1)(ǫX +ǫY ) ((Z, X) , Y ) = 0 ,
gh[(X, Y )] = gh[X] + gh[Y ] + 1 ,
3
This definition differs somewhat from the one of refs.[252, 257].
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 50

ǫ[(X, Y )] = ǫX + ǫY + 1 (mod 2) . (4.5)

The first equation says that ( , ) is graded antisymmetric. The second equation shows
that ( , ) satisfies a graded Jacobi identity. The antibracket “carries” ghost number
one and has odd statistics. From these properties and the definition of right and left
derivatives, one concludes that
∂r B ∂l B
(B, B) = 2 , for B bosonic ,
∂ΦA ∂Φ∗A

(F, F ) = 0 , for F fermionic ,

((X, X), X) = 0 , for any X . (4.6)
The above expression for (B, B) is derived in Appendix A. The antibracket (X, Y ) is
also a graded derivation with ordinary statistics for X and Y :

(X, Y Z) = (X, Y )Z + (−1)ǫY ǫZ (X, Z)Y ,

(XY, Z) = X(Y, Z) + (−1)ǫX ǫY Y (X, Z) . (4.7)

The antibracket defines an odd symplectic structure because it can be written as
!
∂r X ab ∂l Y ab 0 δBA
(X, Y ) = ζ , where ζ ≡ , (4.8)
∂z a ∂z b −δBA 0

when one groups the fields and antifields collectively into z a : z a = {ΦA , Φ∗A }, a =
1, . . . , 2N. The expression for the antibracket in Eq.(4.8) is sometimes useful in ab-
stract proofs. The antibracket formalism can be developed in an arbitrary coordinate
system, in which case ζ ab is replaced by an odd closed field-dependent two-form. For
more details, see Sect. 10.7. However, locally there always exists a basis for which ζ ab
is of the form of Eq.(4.8) [218].
The antibracket in the space of fields and antifields plays a role analogous to
the Poisson bracket. Whereas the Poisson bracket is used at the classical level in a
hamiltonian formulation, the antibracket is used at the classical or quantum level in a
lagrangian formalism. One can use the antibracket in a manner similar to the Poisson
bracket. The antifield Φ∗A can be thought of as the conjugate variable to ΦA since
 
ΦA , Φ∗B = δBA . (4.9)

Infinitesimal canonical transformations [24] in which ΦA → Φ̄A and Φ∗A → Φ̄∗A , where
Φ̄A and Φ̄∗A are functions of the Φ and Φ∗ , are defined by
 
Φ̄A = ΦA + ε ΦA , F + O(ε2) , Φ̄∗A = Φ∗A + ε (Φ∗A , F ) + O(ε2) , (4.10)
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 51

where F is an arbitrary function of the fields and antifields with gh[F ] = −1 and
ǫ(F ) = 1. A short calculation reveals that the canonical structure in Eq.(4.9) is
maintained:
     
Φ̄A , Φ̄∗B = δBA + O(ε2) , Φ̄A , Φ̄B = 0 + O(ε2 ) , Φ̄∗A , Φ̄∗B = 0 + O(ε2 ) .
(4.11)
Canonical transformations preserve the antibracket, i.e.,
∂r X ∂l Y ∂r X ∂l Y ∂r X ∂l Y ∂r X ∂l Y
∗
− ∗
= − + O(ε2 ) .
A
∂ Φ̄ ∂ Φ̄A ∂ Φ̄A ∂ Φ̄ A ∂Φ ∂ΦA ∂Φ∗A ∂ΦA
A ∗

Under Eq.(4.10), an arbitrary scalar function G of fields and antifields transforms as

G → G + δG where
δG = ε(G, F ) + O(ε2) . (4.12)
An alternative approach [263, 27] uses a function which is a combination of original
and transformed fields, F2 [Φ, Φ̃∗ ], to produce a canonical transformation via
h i h i
∂l F2 Φ, Φ̃∗ ∂r F2 Φ, Φ̃∗
Φ̃A = , Φ∗A = , (4.13)
∂ Φ̃∗A ∂ΦA
where ǫ (F2 ) = 1 and gh [F2 ] = −1. Of course, one must require this change of vari-
ables to be non-singular. The advantage of using F2 is that the antibracket structure
is exactly preserved
     
Φ̃A , Φ̃∗B = δBA , Φ̃A , Φ̃B = Φ̃∗A , Φ̃∗B = 0 . (4.14)

so that F2 generates a finite transformation. A proof of Eq.(4.14) is given in ref.[251].

In using Eq.(4.13), one must solve for ΦA and Φ∗A in terms of Φ̃A and Φ̃∗A , i. e.,
ΦA = Φ̄A [Φ̃, Φ̃∗ ] and Φ∗A = Φ̄∗A [Φ̃, Φ̃∗ ]. Then the canonical transformation corresponds
to the shifts ΦA → Φ̄A [Φ, Φ∗ ] and Φ∗A → Φ̄∗A [Φ, Φ∗ ]. Infinitesimal transformations are
recovered using h i
F2 = Φ̃∗A ΦA + εf2 Φ, Φ̃∗ . (4.15)
When f2 = −F , Eq.(4.10) is reproduced to order ε.
The role of the antibracket at quantum level is discussed in Sect. 8.

4.3 Classical Master Equation and Boundary Conditions

Let S[Φ, Φ∗ ] be an arbitrary functional of fields and antifields with the dimensions
of action and with ghost number zero and even statistics: ǫ(S) = 0 and gh[S] = 0.
The equation
∂r S ∂l S
(S, S) = 2 A =0 (4.16)
∂Φ ∂Φ∗A
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 52

is called the classical master equation. This simple looking equation is the main topic
of this subsection.
Not every solution to Eq.(4.16) is of interest. It is necessary to satisfy certain
boundary conditions. A relevant solution plays a double role. On one hand, a solu-
tion S is the generating functional for the structure functions of the gauge algebra. All
relations among structure functions are contained in Eq.(4.16), thereby reproducing
the equations in Sect. 2.4 and generalizing them to the generic L-th stage reducible
theory. On the other hand, S is the starting action to quantize covariantly the the-
ory. After a gauge-fixing procedure is implemented, one can commence perturbation
theory. The latter aspects are treated in Sect. 6.
Regard S as the action for fields and antifields. The variations of S with respect
Φ and Φ∗A are the equations of motion:
A

∂r S
=0 , (4.17)
∂z a
where the collective variables z a in Eq.(4.8) are used. We assume there exists a least
one stationary point for which Eq.(4.17) holds. We let Σ denote this subspace of
stationary points in the full space of fields and antifields:
( )
∂r S
Σ ←→ =0 . (4.18)
∂z a

It turns out that, given a classical solution φ0 of S0 as in Eq.(2.9), one possible

stationary point is

φj = φj0 , Csαs = 0; s = 0, . . . , L; αs = 1, . . . , ms , Φ∗A = 0 . (4.19)

An action S, satisfying the master equation (4.16), possesses its own set of gauge
invariances. Indeed, by differentiating Eq.(4.16) with respect to z b , one finds after a
little algebra that
∂r S a
R =0, a = 1, . . . , 2N , (4.20)
∂z a b
where
∂l ∂r S
Rab ≡ ζ ac c b . (4.21)
∂z ∂z
Although there appears to be 2N gauge invariances, not all of them are independent
on-shell as we now demonstrate. Differentiating Eq.(4.20) with respect to z d , mul-
tiplying by ζ cd, using the definition of Rab in Eq.(4.21) and imposing the stationary
condition in Eq.(4.17), one finds that

Rca Rab |Σ = 0 ,
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 53

where Σ denotes the space satisfying the on-shell condition in Eq.(4.18). One con-
cludes that Rab is on-shell nilpotent. It is an elementary result of matrix theory that a
nilpotent 2N × 2N matrix has rank less than or equal to N. Hence, at the stationary
point there exist at least N relations among the gauge generators Rab and therefore
the number 2N − r of independent gauge transformations on-shell is greater than or
equal to N, where r is the rank of the hessian of S at the stationary point:

∂l ∂r S 
r ≡ rank  . (4.22)
∂z a ∂z b Σ

Necessarily,
r≤N .
A solution to the master equation is called proper if

r=N , (4.23)

where r is given in Eq.(4.22). When r < N, there are solutions to ∂∂zr Sa δz a = 0 other
than the ones given by Eqs.(4.20) and (4.21).
Usually, only proper solutions are of interest. The reason is simple. The action S
contains the physical fields φi and the ghost fields necessary for quantization. How-
ever, the antifields Φ∗A , A = 1, . . . , N, are unphysical. If r = N, then the number of
independent gauge invariances of the type in Eq.(4.20) is the number of antifields. As
a consequence, one can remove the N non-physical variables Φ∗A , while maintaining
the N fields ΦA . At a later stage, the antifields can be eliminated through a gauge-
fixing procedure. Throughout the rest of this article, we restrict the discussion to the
proper case.
We have not yet specified the relation between S0 and S. To make contact with
the original theory, one requires the proper solution to contain the original action
S0 [φ]. This requirement ensures the correct classical limit. It corresponds to the
following boundary condition on S

S [Φ, Φ∗ ]|Φ∗ =0 = S0 [φ] . (4.24)

An additional boundary requirement is


∂l ∂r S 
 αs−1
∗
 = Rsα (φ) , s = 0, . . . , L , (4.25)
α 
∂Cs−1,αs−1 ∂Cs Φ∗ =0
s s

∗ α
s−1
where Cs−1,αs−1
is the antifield of Cs−1 :
∗
Cs,αs
≡ (Csαs )∗ . (4.26)
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 54

In Eq.(4.25), for notational convenience we define

α ∗
C−1−1 ≡ φi , C−1,α−1
≡ φ∗i , with α−1 = i . (4.27)

Actually, Eq.(4.25) does not need to be imposed as a separate boundary condition.

Although it is not obvious, the requirement of being proper (4.23) and the classical
boundary condition in Eq.(4.24) necessarily imply that a solution S must satisfy
Eq.(4.25) [28].
The proper solution S can be expanded in a power series in antifields. Given the
ghost number restriction on S and the boundary conditions in Eqs.(4.24) and (4.25),
the expansion necessarily begins as
L
X
S [Φ, Φ∗ ] = S0 [φ] + ∗
Cs−1,αs−1
Rsαs
Cs + O(C ∗2 ) .
αs−1 αs
(4.28)
s=0

In the next subsection we obtain the leading terms in the solution for a first-stage
reducible theory. We also demonstrate how the gauge algebra is encoded in the
classical master equation, when the classical basis is used.

4.4 The Proper Solution and the Gauge Algebra

This subsection establishes the connection between a proper solution to the clas-
sical master equation and the equations that the gauge-structure tensors must satisfy.
For a first-stage reducible theory the minimum set of fields is φi , C0α0 (which we call
C α ) and C1α1 (which we call η a ). One expands the action S[Φ, Φ∗ ] as a Taylor series in
ghosts and antifields, while requiring the total ghost number to be zero. The result is
 
1 α
∗
S [Φ, Φ ] = S0 [φ] + φ∗i Rαi C α + Cα∗ Zaα η a + Tβγ (−1)ǫβ C γ C β
2
 
1 a
+ηa∗ Aabα η b C α
+ Fαβγ (−1)ǫβ C γ C β C α
2
 
∗ ∗ ǫi 1 ji a 1 ji ǫα β α
+φi φj (−1) V η + Eαβ (−1) C C
2 a 4
 
∗ ∗ ǫδ iδ a β 1 iδ ǫβ γ β α
+ Cδ φi (−1) Gaβ η C − Dαβγ (−1) C C C (4.29)
2
1
+ φ∗i φ∗j φ∗k (−1)ǫj Mαβγ
kji
(−1)ǫβ C γ C β C α + . . . ,
12
where, with the exception of Rαi and Zaα , which are fixed by the boundary condition
γ ji
in Eq.(4.28), one should, at this stage, think of the tensors, Tαβ , Eαβ , etc. in Eq.(4.29)
as having no relation to the tensors found in the gauge algebra structure of Sect. 2.4.
The minus signs and fractions, 21 , 14 , and 12 1
, in Eq.(4.29) are put in for convenience.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 55

The goal is to find a solution to Eq.(4.16). It is necessary to compute derivatives

of S with respect to fields and antifields. One finds
   
∂r S 1 α
i
= S 0,i + φ ∗
j Rα
j α
C + Cα
∗
Z α a
a η + Tβγ (−1)ǫβ C γ C β +
∂φ ,i 2 ,i

1 ∗ ∗  
φj φk (−1)ǫj Eαβ kj
(−1)ǫα C β C α + . . . ,
4 ,i
 
∂l S i α ∗ ǫi ji a 1 ji ǫα β α
= R α C + φ j (−1) V a η + E (−1) C C
∂φ∗i 2 αβ
 
ǫδ ǫi ∗ iδ a β 1 iδ ǫβ γ β α
+ (−1) Cδ Gaβ η C − Dαβγ (−1) C C C +
2
1
+ (−1)ǫj φ∗j φ∗k Mαβγ kji
(−1)ǫβ C γ C β C α + . . . ,
4
∂r S 1
α
= φ∗i Rαi + Cγ∗ Tαβγ
(−1)ǫα C β + (−1)ǫi φ∗i φ∗j Eαβ ji
(−1)ǫα C β + . . . ,
∂C 2
∂l S 1 α 1
∗
= Zaα η a + Tβγ (−1)ǫβ C γ C β − (−1)ǫα φ∗i Dβγδ iα
(−1)ǫγ C δ C γ C β + . . . ,
∂Cα 2 2
∂r S 1
a
= Cα∗ Zaα + φ∗i φ∗j (−1)ǫi Vaji + . . . ,
∂η 2
∂l S 1 a
∗
= Aabα η b C α + Fαβγ (−1)ǫβ C γ C β C α + . . . .
∂ηa 2
Although ghosts do not depend on φi , we write ( ),i in the first equation to avoid
some minus signs.
Given the above equations, it is straightforward to form the sum over products of
derivatives in Eq.(4.16). We leave this step to the reader and simple quote the result
γ ji
[24, 26]: The classical master equation (4.16) is satisfied if the tensors, Tαβ , Eαβ , etc.
in Eq.(4.29) are the ones in Sect. 2.4. In other words, Eq.(4.29) with the tensors
identified as the ones in Sect. 2.4 is a proper solution to the master equation. One
finds that (S, S) = 0 implies the gauge structure in equations Eqs.(2.33), (2.34) and
(2.39) – (2.42). The reason why one equation (S, S) = 0 is able to reproduce many
equations is that the coefficients of each ghost and antifield term must separately be
zero. Up to overall factors, Eq.(2.33) is the term independent of antifields, Eq.(2.34)
is the coefficient of φ∗i which is bilinear in the C α , Eq.(2.39) is the coefficient linear in
Cδ∗ and trilinear in the C α , Eq.(2.40) is the coefficient of φ∗i φ∗j which is trilinear in the
C α , Eq.(2.41) is the coefficient linear in φ∗i and in η a , and Eq.(2.42) is the coefficient of
Cγ∗ which is linear in both η a and C β . The coefficients of higher-order terms produce
the higher-order gauge structure equations. For an irreducible system, the proper
solution is the above with η a and ηa∗ set to zero.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 56

Summarizing, the antibracket formalism using fields and antifields allows a simple
determination of the relevant gauge structure tensors. The proper solution to the
classical master equation is a compact way of expressing the relations among the
structure tensors.

4.5 Existence and Uniqueness

At this point, there are two obvious and interesting questions. Does there al-
ways exist a proper solution to the classical master equation and is the proper so-
lution unique? Given reasonable conditions, which include the regularity postulates
of Appendix B, there always exists a proper solution. This result was obtained in
refs.[263, 28] for the case of an irreducible theory. For a general L-th stage reducible
theory, the theorem was proven in ref.[105].
To gain insight into the question of uniqueness, assume that a proper solution S
has been found so that (S, S) = 0. Suppose that one performs a canonical trans-
formation via Eq.(4.13). Then the transformed S also obeys the classical master
equation because such canonical transformations preserve the antibracket. For an in-
finitesimal canonical transformation, the transformed proper solution S ′ is, according
to Eq.(4.12), given by S ′ = S + ε(S, F ) + O(ε2). This is the only ambiguity in the
proper solution. Given the minimal set of fields in Eq.(4.1), the proper solution to
the classical master equation is unique up to canonical transformations. This result
was obtained in refs.[263, 28, 105]. Since our aim is pedagogical, we do not present
the proof here. The interested reader can consult the above references. Below we
illustrate the non-uniqueness question using the example of the spinless relativistic
particle. Canonical transformations which lead to field redefinitions of ghosts corre-
spond to the freedom of redefining the gauge generators. This leads to changes in the
structure tensors of Sect. 2 and corresponds to the non-uniqueness mentioned at the
end of Sect. 2.1.
When gauge-fixing and path integral quantization is considered, it is necessary
to enlarge the minimal set of fields in Eq.(4.1) to a non-minimal set. It turns out
that trivial variable pairs can be added to the theory while maintaining the classical
master equation and its properness. Let Λ and Π be two new fields and let Λ∗ and
Π∗ be the corresponding antifields. Choose the ghost numbers and statistics so that
gh [Π] = gh [Λ] + 1 ,
ǫ (Π) = ǫ (Λ) + 1 (mod 2) . (4.30)
Then gh [Λ∗ ] = −gh [Λ] − 1 = −gh [Π] and ǫ (Λ∗ ) = ǫ (Λ) + 1 (mod 2) = ǫ (Π), so that
Z
Strivial = dxΛ∗ Π (4.31)
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 57

can be added to S. Adding such a bilinear to S does not ruin the classical master
equation (S, S) = 0. When a non-minimal set of fields is employed, the proper
solution to the master equation is unique up to canonical transformations and the
addition of trivial pairs.

4.6 The Classical BRST Symmetry

An important concept in gauge theories is BRST symmetry. The BRST symme-
try is what remains of gauge invariance after gauge-fixing has been implemented. In
this sense, it can be regarded as a substitute for gauge invariance. There are three im-
portant features governing the BRST transformation: nilpotency, graded derivation,
and invariance of the action S.
Even before gauge-fixing, the field-antifield formalism has BRST symmetry. Via
the antibracket, the generator δB of this symmetry is the proper solution S itself.
Define the classical BRST transformation of a functional X of fields and antifields by

δB X ≡ (X, S) . (4.32)

The transformation rule for fields and antifields is therefore

∂l S ∂l S ǫA +1 ∂r S
δB ΦA = , δB Φ∗A = − = (−1) . (4.33)
∂Φ∗A ∂ΦA ∂ΦA
Note that the field-antifield action S is classically BRST symmetric

δB S = 0 , (4.34)

as a consequence of (S, S) = 0.
The BRST operator δB is a nilpotent graded derivation4 : Given two functionals
X and Y ,
δB (XY ) = XδB Y + (−1)ǫY (δB X) Y , (4.35)
and
δB2 X = 0 . (4.36)
4
An alternative definition for δB given by

δB X ≡ (−1)ǫX (X, S) ,

is used by some authors. In this case, the δB continues to be a nilpotent graded derivation, but acts
from the left to right, i.e., (4.35) is replaced by

δB (XY ) = (δB X)Y + (−1)ǫX XδB Y .

J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 58

The nilpotency follows from two properties of the antibracket: the graded Jacobi iden-
tity and graded antisymmetry (see Eq.(4.5)). These properties imply ((X, S) , S) =
− ((S, S) , X) + (−1)ǫX +1 ((S, X) , S) = − ((S, S) , X) − ((X, S) , S) which leads to
((X, S) , S) = − 12 ((S, S) , X) = 0. Therefore, δB2 X = ((X, S) , S) = 0.
A functional O is a classical observable if δB O = 0 and O 6= δB Y for some Y . Two
observables are considered equivalent if they differ by a BRST variation [103, 105]. A
linear combination of observables is an observable. Because of the graded derivation
property of δB in Eq.(4.35), the product of two classical observables is BRST invariant.
Thus, observables form an algebra.
For a closed irreducible theory, the BRST transformation rules for ΦA and Φ∗A ,
which depend on the original action S0 and the structure tensors R and T , can be
obtained using Eqs.(4.29) and (4.33):

δB φi = Rαi C α + . . . ,
1 α
δB C α = Tβγ (−1)ǫβ C γ C β + . . . ,
2
1
δB φ∗i = − (−1)ǫi S0,i − (−1)ǫαǫi φ∗j Rα,i
j
C α − (−1)(ǫα +ǫβ +1)ǫi +ǫα Cγ∗ Tαβ,i
γ
Cβ Cα + . . . ,
2
δB Cα∗ = (−1)ǫα φ∗i Rαi + Cγ∗ Tαβ
γ
Cβ + . . . , (4.37)
where we have displayed the terms involving S0 , R and T . When the gauge-structure
iδ kji
equations hold off-shell at all levels, then the tensors Dαβγ , Mαβγ , etc. in Eq.(4.29) are
zero, and the proper action is linear in antifields. In this case, there are no additional
terms in Eq.(4.37) for irreducible systems. We illustrate BRST symmetry in a few of
the examples in the next section.
Because Sect. 4 has introduced many ideas, it is worth enumerating the most
important ones. After reading this section, one should know the following tools and
concepts: antifields and the antibracket, canonical transformations, the classical mas-
ter equation, properness and proper solution, classical BRST symmetry, and classical
observables.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 59

5 Examples of Proper Solutions

In this section, we present proper solutions to the master equation for the exam-
ples considered in Sect. 3. Given the gauge-structure tensors, Eq.(4.29) immediately
provides the proper solution S for an irreducible or first-stage-reducible theory. Since
all but two of the example field theories fall into one of these two cases, the construc-
tion of S is straightforward. For the abelian p-form theory, Eq.(4.28) is needed. The
infinitely reducible open bosonic string requires some guesswork to obtain S. Given
S, the BRST transformations δB can be determined from Eq.(4.32). For reasons
of space, we display these transformations only for the spinless relativistic particle,
Yang-Mills theory and the open bosonic string field theory. In Sect. 5.1, it is shown
how two proper solutions to the spinless relativistic particle are related by a canonical
transformation.
For exercises, we suggest the following (i) verify (S, S) = 0, (ii) determine the
action of δB on the fields and antifields in each example, (iii) use the results of (ii) to
verify that S is invariant under BRST transformations. Since (iii) and (i) are related,
the calculations are similar and the reader may want to do only one or the other.

5.1 The Spinless Relativistic Particle

The spinless relativistic particle, considered in Sect. 3.1, is an example of a closed
irreducible theory. The minimal set of fields and antifields are
n o
ΦA = {xµ , e, C} , Φ∗A = x∗µ , e∗ , C ∗ , (5.1)

with statistical parities

ǫ(xµ ) = ǫ(e) = ǫ(C ∗ ) = 0 , ǫ(x∗µ ) = ǫ(e∗ ) = ǫ(C) = 1 , (5.2)

and ghost numbers

gh[xµ ] = gh[e] = 0 , gh[C] = 1 , gh[x∗µ ] = gh[e∗ ] = −1 , gh[C ∗ ] = −2 . (5.3)

Two versions of the gauge transformations were presented. In the first, given in
Eq.(3.3), the only non-zero structure tensors are the gauge generators Rαi given in
Eq.(3.5). Using this and the general solution in Eq.(4.29), one finds
Z  
C
∗ µ
S [Φ, Φ ] = S0 [x , e] + dτ x∗µ ẋµ + e C˙
∗
, (5.4)
e
where S0 [xµ , e] is given in Eq.(3.1). In the second version, given in Eq.(3.6), the
gauge generators Rαi and the commutator structure constant Tβγ α
are non-zero. Using
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 60

Eq.(4.29), one obtains

Z  
S [Φ, Φ∗ ] = S0 [xµ , e] + dτ x∗µ ẋµ C + e∗ eC˙ + e∗ ėC + C ∗ CC
˙ . (5.5)

By the uniqueness theorem of the previous section, the solutions in Eqs.(5.4) and
(5.5) must be related by a canonical transformation. In fact, Eq.(5.4) is mapped to
Eq.(5.5) if xµ , x∗µ and e are unchanged but

1 1
C → eC , C∗ → C∗ , e∗ → e∗ − C ∗ C .
e e
One can check that this is a canonical transformation by verifying that the antibracket
structure in Eq.(4.11) is preserved. The infinitesimal version of the transformation is
generated by the fermion
F = ln (e) C ∗ C ,
when F is used in Eq.(4.10). Infinitesimally,
1
δC = ε ln (e) C , δC ∗ = −ε ln (e) C ∗ , δe∗ = −ε C ∗ C .
e
The finite transformation is obtained by iterating the infinitesimal transformation N
times, requiring Nε = 1 and then letting N → ∞. Alternatively, the full transforma-
tion can be generated using
Z !
C˜∗ C
F2 = dτ x̃∗µ xµ ∗
+ ẽ e + (5.6)
e

and Eq.(4.13).
It is straightforward to obtain the BRST transformation rules using Eq.(4.33).
From Eq.(5.4), one finds
ẋµ C
δB e = C˙ , δB xµ = , δB C = 0 ,
e
!
∗ 1 ẋ2 2 x∗µ ẋµ C
δB e = + m + ,
2 e2 e2
!
d ẋµ + x∗µ C
δB x∗µ = , (5.7)
dτ e
x∗µ ẋµ
∗
δB C = − ė∗ .
e
It is a useful exercise to verify the nilpotency of δB when acting on any field or
antifield. One must be careful of signs. In this regard Eq.(4.35) is useful.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 61

5.2 Yang-Mills Theories

Yang-Mills theories have a closed irreducible algebra. The number of gauge
parameters is the rank of the group. There is a ghost field C a for each gauge parameter
Λa in Eq.(3.15). Hence the fields and antifields are
n o
ΦA = {Aaµ , C a } , Φ∗A = A∗aµ , Ca∗ . (5.8)

From Eq.(4.29), the proper solution is

Z  
1 a µν 1
S= dd x − Fµν Fa + A∗aµ D µa b C b + Cc∗ fab c C b C a . (5.9)
4 2
The first term is the classical action (3.12) and the second and third terms correspond
γ
respectively to the Rαi and Tαβ terms in Eq.(4.29).
As a check, let us verify the classical master equation (S, S) = 0. Using Eq.(4.16),
one finds Z
(S, S) = dd x (D ν Fνµ )a D µa b C b
Z
1 Z d ∗ µd
− dd
xA∗dµ fac d C c D µa b C b + d xAdµ D c fab c C b C a
2
Z
1
+ dd x Ce∗ fdc e C c fab d C b C a .
4
The first term vanishes when one integrates by parts, uses the antisymmetry of Fµν
in µ and ν, employs Eq.(3.19) and makes use of the antisymmetry of fab c in a and b.
The second and third terms cancel when one makes use of the anticommuting nature
of C a and C b , of the antisymmetry of fab c in a and b, and of the Jacobi identity for
the Lie group structure constants in Eq.(3.11). The last term vanishes for similar
reasons. Hence, (S, S) = 0.
The BRST transformation rules are obtained from Eqs.(4.33) and (5.9):

δB Aaµ = D µa b C b ,
1
δB C a = fbc a C c C b ,
2
δB A∗aµ = − (D ν Fνµ )a + fab c A∗cµ C b , (5.10)
 
δB Ca∗ = − D µ A∗µ + Cc∗ fab c C b .
a
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 62

5.3 Topological Yang-Mills Theory

This theory is first-stage reducible. The action given in Eq.(3.20) is invariant
under two types of gauge transformations. The gauge parameters associated with
these transformations are Λa of Eq.(3.15) and εaµ of Eq.(3.21). Correspondingly, one
needs to introduce ghosts C a ↔ Λa and C aµ ↔ εaµ . For the gauge invariances of the
gauge invariances, the ghost-for-ghost field η a must be introduced. Hence the fields
and antifield are
n o
ΦA = {Aaµ , C a , C aµ , η a } , Φ∗A = A∗aµ , Ca∗ , Caµ
∗
, ηa∗ . (5.11)

The non-zero gauge tensors are the gauge generators Rαi given in Eq.(3.23), the com-
γ
mutator structure constants Tαβ given in Eq.(3.25), the null vectors Zaα given in
Eq.(3.24), and Aabα given in Eq.(3.26). Other tensors are zero. Inserting the non-zero
tensors in Eq.(4.29), one finds [137]
Z   
1 a ∗ µν 1
S= 4
dx Fµν Fa + A∗aµ D µa b C b + C aµ + Cc∗ fab c C b C a
4 2
o
∗
+Ccµ fab c C bµ C a + Cb∗ η b − Caµ
∗
D µa b η b + ηc∗ fba c η a C b . (5.12)

5.4 The Antisymmetric Tensor Field Theory

This system has been treated by the antifield formalism in refs.[73, 29, 34, 30,
86, 103, 12]. The theory is on-shell first-stage reducible. The fields and antifields are
n o
ΦA = {Aaµ , B aµν , C aµ , η a } , Φ∗A = A∗aµ , Baµν
∗ ∗
, Caµ , ηa∗ . (5.13)

The non-zero structure tensors Rαi , Zaα and Vaij are given in Eqs.(3.31) – (3.33). The
proper solution is
Z  
1 1 ∗
S = S0 + d x Ba∗κλ εκλµν D µa b C bν + Cbµ
4 ∗
D µb a η a − Bcρσ ∗b µνρσ
Bµν ε fab c η a , (5.14)
2 8
where S0 is given in Eq.(3.27). An effect of the on-shell reducibility is the appearance
of terms quadratic in the antifields.

5.5 Abelian p-Form Theories

Recall that these are examples of p − 1 stage off-shell reducible theories. Conse-
quently there are p different types of ghosts, C0 , C1 , . . . , Cp−1 , where Cs is a p − 1 − s
form:
n o
ΦA = {A, C0 , C1 , . . . , Cp−1 } , Φ∗A = A∗ , C0∗ , C1∗ , . . . , Cp−1
∗
. (5.15)
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 63

The proper solution is

 
Z  p−1
X   
1
S= − F ∧ ∗ F + ∗ (A∗ ) ∧ dC0 + ∗ ∗
Ci−1 ∧ dCi , (5.16)
 2 
i=1

where ∧ is the wedge product and a “ ∗ ” in front of a field or antifield indicates the
dual star operation. In Eq.(5.16), some antifields have been redefined by a minus
sign factor compared to the definitions in Sect. 2. As a consequence of integration
by parts and dd = 0, as well as the definitions of the Hodge star operation and the
wedge product, one can check that (S, S) = 0. Indeed,
Z Z
∂r S ∂l S
∝ (∗ d∗ F ) ∧ ∗ dC0 ∝ d (∗ F ∧ dC0 ) = 0 ,
∂Aµ1 µ2 ...µp ∂A∗µ1 µ2 ...µp

∂r S ∂l S
∗µ1 µ2 ...µp−i−1 ∝
∂ (Ci )µ1 µ2 ...µp−i−1 ∂ (Ci )
Z Z
∗ ∗
( d (Ci∗ )) ∗
∧ dCi+1 ∝ d (∗ (Ci∗ ) ∧ dCi+1 ) = 0 , i = 0, 1, . . . , p − 1 ,
∗
where C−1 ≡ A∗ .

5.6 Open String Field Theory

This is an infinite-stage reducible theory. Hence, there are ghosts, ghosts for
ghosts, ghosts for ghosts for ghosts, etc.. The fields are the string field A and the
infinite tower Cs , s = 0, . . . , ∞:

ΦA = {A, C0 , C1 , C2 , . . .} , Φ∗A = {A∗ , C0∗ , C1∗ , C2∗ , . . .} , (5.17)

where A is a string 1-form, C0 is a string 0-form, C1 is a string −1-form, etc.. Likewise

A∗ is a string 1-form, C0∗ is a string 0-form, C1∗ is a string −1-form, etc.. In Chern-
Simons string field theory, odd forms have odd grading and even forms have even
grading. In addition, according to Eq.(4.2) of the field-antifield formalism, ε (Cs ) is
zero for s odd and one for s even. With respect to the calculus of string differential
forms, the total statistics is important. Consequently, for any fields and/or antifields
ϕi and ϕj , axioms (3) and (5) of Sect. 3.6, namely the graded distributive property
of Q across the star product and the graded commutativity of the star product under
the integral, become
 
Q ϕi ∗ ϕj = Qϕi ∗ ϕj + (−1)si ϕi ∗ Qϕj ,
Z Z
i j si sj
ϕ ∗ ϕ = (−1) ϕj ∗ ϕi , (5.18)
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 64

where si is the total statistics of ϕi :

     
si ≡ s ϕ i = g ϕ i + ǫ ϕ i (mod 2) . (5.19)

Here, g (ϕi ) is the ghost number of ϕi which is the same as the order of the string
form, and ǫ (ϕi ) is given by Eq.(4.2).
The quantization of the bosonic string was carried out by C. Thorn [246] and by
M. Bochicchio [56]. Since this subject has been reviewed in ref.[247], we keep the
discussion brief. To gain some insight in finding the proper solution, let us compute
the non-zero terms in Eq.(4.29). The classical action S0 is given in Eq.(3.39). The
γ
Rαi , Zaα , Vaij and Tαβ terms in Eq.(4.29) are respectively
Z
∗
A∗ ∗ (QC0 + A ∗ C0 + C0 ∗ A) ,
Z
∗ ∗
C0 ∗ (QC1 + A ∗ C1 + C1 ∗ A) ,
Z
∗
A∗ ∗ ∗ A∗ ∗ C1 , (5.20)
Z
∗ ∗
C0 ∗ C0 ∗ C0 ,
where ∗ before a field is the string analog of the dual star operation determined by
R
the bilinear form A ∗ B. It takes p-forms into 3 − p forms. Note that the structure
of the above terms is similar to the classical action S0 evaluated using various fields
and antifields. The sign of the term C0 ∗ A in the first equation is opposite to that of
Eq.(3.42) because C0 has odd statistics. The field-antifield statistics, ghost number
and total statistics for the fields are

ǫ (Ci ) = i + 1 (mod 2) , g (Ci ) = −i , gh [Ci ] = i + 1 , s (Ci ) = 1 ,

ǫ (∗ Ci∗ ) = i (mod 2) , g (∗ Ci∗ ) = i + 3 , gh [∗ Ci∗ ] = −i − 2 , s (∗ Ci∗ ) = 1 , (5.21)

for i ≥ −1, where we have defined
∗
C−1 ≡ A , C−1 ≡ A∗ . (5.22)

Note that the total statistics of Ci and the ∗ Ci∗ is odd, so that it makes sense to define
a field which is the formal sum
s∗+∗3 3 2 1 0 −s
Ψ ≡ ...+ Cs + . . . + ∗ C0∗ + ∗ A∗ + A + C0 + . . . + Cs + . . . ,
∞
X
Ψ= ψp , where ψ−p ≡ Cp , for p ≥ −1 , ψp ≡ ∗ Cp−3
∗
, for p ≥ 2 . (5.23)
p=−∞
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 65

In the first equation, the order of the string form is denoted above the field. The
total statistics of the field-antifield Ψ is odd: s (Ψ) = 1. The terms in Eq.(5.20) and
the classical action are both contained in the following ansatz for the proper solution
Z Z
1 1
S= Ψ ∗ QΨ + Ψ∗Ψ∗Ψ=
2 3
∞ Z ∞ Z
1 X 1 X
ψ2−p ∗ Qψp + ψp ∗ ψq ∗ ψ3−p−q . (5.24)
2 p=−∞ 3 p=−∞
q=−∞

By obtaining the L-th stage reducible proper solution, computing the non-zero
structure tensors, one could derive Eq.(5.24). Instead let us check the ansatz. One
simply needs to verify the classical master equation
Z
?
0 = (S, S) ∝ (QΨ + Ψ ∗ Ψ) ∗ (QΨ + Ψ ∗ Ψ) . (5.25)

Several terms are produced. The first is

Z Z
QΨ ∗ QΨ = Q (Ψ ∗ QΨ) = 0 ,

where axioms (1), (2) and (3) of Sect. 3.6 have been used. Note that
Z
0= Q (Ψ ∗ Ψ ∗ Ψ) =
Z Z Z
(QΨ ∗ Ψ ∗ Ψ) − (Ψ ∗ QΨ ∗ Ψ) + (Ψ ∗ Ψ ∗ QΨ) =
Z Z
3 (QΨ ∗ Ψ ∗ Ψ) = 3 (Ψ ∗ Ψ ∗ QΨ) ,
R R
where axioms (2)–(5) are used. This implies that (QΨ ∗ Ψ ∗ Ψ) = (Ψ ∗ Ψ ∗ QΨ) =
0, leading to the vanishing of two of the terms in Eq.(5.25). The last term in Eq.(5.25)
is zero when axioms (4) and (5) are combined to give
Z Z Z
(Ψ ∗ Ψ) ∗ (Ψ ∗ Ψ) = (Ψ ∗ Ψ ∗ Ψ ∗ Ψ) = − (Ψ ∗ Ψ ∗ Ψ ∗ Ψ) ,

where Eq.(5.18) is used in the last step. Since the master equation is satisfied and
the suitable boundary conditions are correctly implemented, Eq.(5.24) is a proper
solution. The classical action in Eq.(3.39) and the proper solution in Eq.(5.24),
although structurally identical, differ in that the field entering the action is different.
In S0 it is the string one-form A, while in S it is the tower Ψ, given in Eq.(5.23),
which includes ghosts and antighosts as well as A.
The BRST transformation rules for the fields and antifields are

δB Ψ = QΨ + Ψ ∗ Ψ . (5.26)
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 66

The transformation rule for a particular component ψp of Ψ in Eq.(5.23) is obtained

by selecting the p-form term of Eq.(5.26).
Let us verify the nilpotency of δB :

δB2 Ψ = QδB Ψ + Ψ ∗ δB Ψ − δB Ψ ∗ Ψ =

Q2 Ψ + Q (Ψ ∗ Ψ) + Ψ ∗ QΨ + Ψ ∗ (Ψ ∗ Ψ) − QΨ ∗ Ψ − (Ψ ∗ Ψ) ∗ Ψ = 0 .
The last equality holds due to the axioms of open string field theory: The first term
is zero because of axiom (1), the nilpotency of Q; the second, third and fifth terms
cancel because of axiom (3), the graded distributive property of Q; and the fourth
and sixth terms cancel by axiom (4), the associativity of the star product.

5.7 The Massless Relativistic Spinning Particle

This system has a closed irreducible algebra but possesses local supersymmetry. It
has been treated by the antifield formalism in ref.[135]. To the original fields, xµ , ψ µ , e,
and χ, one adds the ghosts C and Γ respectively for general coordinate transformations
and for local supersymmetry transformations. The fields and antifields are
n o
ΦA = {xµ , ψ µ , e, χ, C, Γ} , Φ∗A = x∗µ , ψµ∗ , e∗ , χ∗ , C ∗ , Γ∗ , (5.27)

where xµ , e, Γ, ψµ∗ , χ∗ and C ∗ are commuting while ψ µ , χ, C, Γ∗ , x∗µ and e∗ are

anticommuting.
γ
The only non-zero terms in Eq.(4.29) are S0 , φ∗i Rαi C α and 12 Cγ∗ Tαβ (−1)ǫα C β C α .
These quantities are determined in Eqs.(3.66)–(3.69). The result is
Z n  
S = S0 + dτ x∗µ (−ieψ µ Γ + ẋµ C) + ψµ∗ (ẋµ − iχψ µ ) Γ + ψ̇ µ C +
   
e∗ −2ieχΓ + ėC + eC˙ + χ∗ eΓ̇ + ėΓ + χC˙ + χ̇C +
   o
C ∗ CC ˙ + iχΓΓ
˙ + ieΓΓ + Γ∗ C Γ̇ − CΓ . (5.28)
where S0 is given in Eq.(3.66)

5.8 The First-Quantized Bosonic String

This system has a closed irreducible algebra. There are three types of gauge
invariances: local translations, local Lorentz boosts and scaling transformations to
which we respectively assign the ghosts C m , C ab and C where a, b and m take on the
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 67

values σ and τ and C ba = −C ab . Combining these ghosts with the original fields X µ
and ea m , one arrives at the following set of fields and antifields:
n o n o
ΦA = X µ , ea m , C m , C ab , C , Φ∗A = Xµ∗ , e∗a m , Cm
∗ ∗
, Cab , C∗ , (5.29)

where µ ranges from 0 to D − 1 = 25. The only non-zero terms in Eq.(4.29) are S0 ,
γ
φ∗i Rαi C α and 12 Cγ∗ Tαβ (−1)ǫα C β C α . These tensors are computed in Sect. 3.8. One thus
obtains Z Z π n
S = S0 + dτ dσ Xµ∗ C m ∂m X µ +
0
 
∗a
e m C ∂n ea − ea ∂n C m + ea m C + eb m ηac C cb
n m n

  o
∗ n ∗
−Cm C ∂n C m + Cab C ac C bd ηcd − C n ∂n C ab − C ∗ C n ∂n C , (5.30)
where S0 is given in Eq.(3.79).
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 68

6 The Gauge-Fixing Fermion

This section discusses the process of gauge-fixing in the field-antifield formal-
ism [24, 26], thereby paving the way to the quantization of gauge theories via the
path integral approach. The important concept is a gauge-fixing fermion Ψ. It is
a Grassmann-odd functional with ghost number −1. In Sect. 6.1, it is shown how
antifields can be eliminated to obtain an action suitable for the computation of cor-
relation functions in standard perturbation theory. Results are independent of Ψ if
the action satisfies the quantum master equation in Eq.(6.12) [23, 24]. To construct
an appropriate Ψ, additional fields and their antifields are needed. See Sect. 6.2.
Details of the gauge-fixing procedure for a general theory are presented in Sects. 6.3
and 6.4. Since Sects. 6.2–6.4 are somewhat technical, the reader may wish to read
only the irreducible case, which is discussed at the beginning of each subsection. In
particular instances, gauge-fixing can be done without Ψ by performing a cleverly
chosen canonical transformation [104, 230, 259, 44, 252, 257].
After gauge-fixing is completed, the theory is still invariant under gauge-fixed
BRST transformations δBΨ [24]. The nilpotency of δBΨ is not guaranteed off-shell
[255, 29, 30, 251]. However, when the equations of motion for the gauge-fixed action
are used, δB2 Ψ = 0. An attractive way of viewing the effects of gauge-fixing associated
with Ψ and the content of δBΨ , is to perform a canonical transformation to the so-
called gauge-fixed basis. In this basis, one retains antifields as sources of gauge-fixed
BRST transformations, and the classical master equation reproduces the algebraic
structure associated with δBΨ .

6.1 Generalities
Although ghost fields have been incorporated into the theory, the field-antifield
action S still possesses gauge invariances (See Eq.(4.20)). Hence it is not yet suitable
for quantization via the path integral approach. A gauge-fixing procedure is needed.
The theory also contains many antifields that usually one wants to eliminate before
computing amplitudes and S-matrix elements. One cannot simply set the antifields to
zero because the action would reduce to the original classical action S0 , which is not
appropriate for commencing perturbation theory due to gauge invariances. Following
ref.[24], antifields are eliminated by using a gauge-fixing fermion Ψ (Φ) via

∂Ψ
Φ∗A = . (6.1)
∂ΦA
Note that Ψ is a functional of fields only. It does not matter whether right or left
∂l Ψ ∂r Ψ
derivatives are used in Eq.(6.1) because ∂Φ A = ∂ΦA since ǫ (Ψ) = 1 (see Eq.(A.4) of
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 69

Appendix A). We denote the “surface” in functional space determined by Eq.(6.1) by

ΣΨ ,5 ( )
∗ ∂r Ψ
ΣΨ ←→ ΦA = , (6.2)
∂ΦA
so that " #
∂Ψ
X|ΣΨ ≡ X Φ, . (6.3)
∂Φ
The corresponding gauge-fixed action is S|ΣΨ ≡ SΨ . Matching the statistics and
∂Ψ
ghost numbers of Φ∗A and ∂Φ A leads to

ǫ (Ψ) = 1 , gh [Ψ] = −1 . (6.4)

To quantize the theory, let us use the path integral approach with the constraint
in Eq.(6.1) implemented by a delta function:
Z !  
∗ ∂Ψ i
IΨ (X) = [dΦ] [dΦ ] δ Φ∗A − A
exp W [Φ, Φ∗ ] X [Φ, Φ∗ ] , (6.5)
∂Φ h̄

where X is a correlation function of interest. Here W is the quantum generalization of

the field-antifield action S. By the correspondence principle, the two should be equal
in the h̄ → 0 limit. One must choose Ψ so that the theory is non-degenerate, that is,
when the action is expanded about a solution of the equations of motion, propagators
exist. Such a Ψ is called admissible. Conditions on Ψ assuring admissibility are given
in Sect. 6.3 for certain gauge-fixing schemes. When S0 is local, it is desirable that Ψ be
a local functional of the fields so as to preserve the locality of the gauge-fixed action.
Selecting a “good” Ψ is a matter of skill. A judicious choice can greatly simplify a
particular computation, and the choice can depend on the type of computation, e.g.
a correlation function, a proof of unitarity, a proof of renormalizability, etc..
The freedom in choosing Ψ corresponds to the choice of the gauge-fixing procedure.
One would like results to be independent of gauge fixing. Let us determine when this
is the case. Define the integrand in Eq.(6.5) to be I, i.e.,
 
∗ i
I [Φ, Φ ] ≡ exp W [Φ, Φ∗ ] X [Φ, Φ∗ ] . (6.6)
h̄
Under an infinitesimal change δΨ of Ψ, IΨ (X) changes by
Z " # " #!
∂Ψ ∂δΨ ∂Ψ
IΨ+δΨ (X) − IΨ (X) = [dΦ] I Φ, + − I Φ,
∂Φ ∂Φ ∂Φ
5
Do not confuse ΣΨ with Σ of Eq.(4.18). The latter (Σ without a subscript Ψ) corresponds to
the on-shell condition.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 70

Z  
∂r I ∂l δΨ 2
= [dΦ] + O (δΨ)
∂Φ∗A ∂ΦA
Z  
= [dΦ]∆IδΨ + O (δΨ)2 ,

where integration by parts (see Eq.(A.5) in Appendix A) has been used, and where
the operator ∆ is defined to be [24, 26]
∂r ∂r
∆ ≡ (−1)ǫA+1 . (6.7)
∂ΦA ∂Φ∗A
It is a kind of “nilpotent divergence operator” in the space of fields and antifields. It
formally satisfies
∆∆ = 0 ,
∆ (XY ) = X∆Y + (−1)ǫY (∆X) Y + (−1)ǫY (X, Y ) ,
∆ ( (X, Y ) ) = (X, ∆Y ) − (−1)ǫY (∆X, Y ) . (6.8)
Note that
gh [∆] = ǫ (∆) = 1 . (6.9)
According to the above calculation, the integral IΨ (X) is infinitesimally independent
of Ψ if
∆I = 0 . (6.10)
Eq.(6.10) is a requirement of a “good” integrand.
The path integral itself should be gauge independent. This corresponds to setting
X = 1 in IΨ (X), from which one finds
    
i i i 1
∆ exp W = exp W ∆W − 2 (W, W ) = 0 , (6.11)
h̄ h̄ h̄ 2h̄
so that one needs
1
(W, W ) = ih̄∆W . (6.12)
2
Eq.(6.12) is known as the quantum master equation [24, 26]. When W satisfies
Eq.(6.12), X in Eq.(6.6) must satisfy

(X, W ) = ih̄∆X (6.13)

to produce gauge-invariant correlation functions. Eqs.(6.12) and (6.13) summarize

the conditions so that a computation does not depend on the choice of Ψ [103, 152].
If W is expanded in powers of h̄ (the loop expansion) via
∞
X
W =S+ h̄p Mp , (6.14)
p=1
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 71

the quantum master equation (Eq.(6.12)) implies that the first term S and quantum
correction terms Mp must satisfy

(S, S) = 0 , (M1 , S) = i∆S ,

1 n−1
X
(Mn , S) = i∆Mn−1 − (Mp , Mn−p ) , for n ≥ 2 . (6.15)
2 p=1
By the correspondence principle, S in Eq.(6.14) is identified as the classical field-
antifield action and is consistent with Eqs.(6.12) and (6.15) because it satisfies the
classical master equation (S, S) = 0.
Likewise, when X in Eq.(6.13) is expanded in an h̄ series
∞
X
X= h̄p Xp , (6.16)
p=0

one finds that Eq.(6.13) implies

(X0 , S) = 0 ,
n
X
(Xn , S) = i∆Xn−1 − (Xn−p , Mp ) , for n ≥ 1 . (6.17)
p=1

The first equation says that the “classical” part of the quantum operator X is clas-
sically BRST invariant.

6.2 Gauge-Fixing Auxiliary Fields

To eliminate antighost fields by Eq.(6.1), Ψ must be a functional of fields and have
ghost number −1. However, the fields of the minimal sector introduced in Sect. 4 all
have non-negative ghost numbers. Hence, it is impossible to construct an acceptable
Ψ unless one introduces additional fields. At this point, one takes advantage of adding
trivial pairs (c.f. Sect. 4.5) to the theory. This subsection enumerates the auxiliary
trivial pairs needed to obtain an admissible Ψ [26, 119].
For an irreducible theory, one introduces one trivial pair

C¯0α0 , π̄0α0 , (6.18)

with statistics and ghost numbers equal to

  h i
ǫ C¯0α0 = ǫα0 + 1 , gh C¯0α0 = −1 ,

ǫ (π̄0α0 ) = ǫα0 , gh [π̄0α0 ] = 0 . (6.19)

J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 72

The fields C¯0α0 are the Faddeev-Popov antighosts of C0α0 .

For a first-stage reducible theory, one needs, in addition to the above pair, two
more trivial pairs
C¯1α1 , π̄1α1 ,
C11α1 , π11α1 . (6.20)
Their statistics and ghost numbers are
   
ǫ C¯1α1 = ǫ C11α1 = ǫα1 ,
h i h i
gh C¯1α1 = −2 , gh C11α1 = 0 ,
 
ǫ (π̄1α1 ) = ǫ π11α1 = ǫα1 + 1 , (6.21)
h i
gh [π̄1α1 ] = −1 , gh π11α1 = 1 .
In general, for an Lth-stage reducible theory, for each integer s ranging from 0 to
L one introduces s + 1 trivial pairs [26]:
 
s
C¯sα
k k
, π̄sα , k = 0, 2, 4, . . . , 2 ,
s s
2
 
s−1
Cskαs , πskαs , k = 1, 3, 5, . . . , 2 +1 , (6.22)
2
L
P
where αs = 1, . . . , ms . The total number of auxiliary trivial pairs is (s + 1) =
s=0
(L + 1) (L + 2) /2. The ghosts C¯sα k
s
, Cskαs , in Eq.(6.22) for k ≥ 1 are known as ex-
k
traghosts, while the fields π̄sαs
, πskαs , are usually called Lagrange multipliers for reasons
which become clear below. The original set of fields {φi , Csαs } of Sect. 4 is called the
minimal set. The minimal sector fields together with the auxiliary fields in Eq.(6.22)
are called the non-minimal set. In Eq.(6.22), [ ] stands for the greatest integer so
that   (m
m 2
, if m is even ,
= m−1
2 2
, if m is odd .
In general, the subscript s on Cskαs and C¯sα
k
s
indicate the level of the field, whereas
the superscript k distinguishes different fields at the same level. It is convenient to
associate α−1 with the index i on φi and define
−1α−1
Cs−1αs ≡ Csαs , C−1 ≡ φi , α−1 ≡ i ,

C¯sα
0
s
≡ C¯sαs , 0
π̄sαs
≡ π̄sαs . (6.23)
The following equations also apply to the minimal sector fields when k = −1 and 0.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 73

The statistics of C¯sα

k
s
and Cskαs are the same as those of Csαs , namely,
   
ǫ C¯sα
k
s
= ǫ Cskαs = ǫαs + s + 1 (mod 2) , (6.24)

and their ghost numbers are

h i
gh C¯sα
k
s
=k−s−1 , L≥s≥0, s≥k≥0, for k even ,
h i
gh Cskαs = s − k , L ≥ s ≥ −1 , s ≥ k ≥ −1 , for k odd . (6.25)
k
The statistics and ghost numbers of π̄sα s
and πskαs are determined by the requirement
that they form a trivial pair respectively with C¯sα
k
s
and Cskαs . Using Eq.(4.30),
   
k
ǫ π̄sαs
= ǫ πskαs = ǫαs + s (mod 2) ,
h i
k
gh π̄sαs
=k−s , L≥s≥0, s≥k≥0, for k even ,
h i
gh πskαs = s − k + 1 , L≥s≥1, s≥k≥1, for k odd . (6.26)
The statistics and ghost numbers of the antifields are determined from Eq.(4.3).
The minimal sector fields along with the auxiliary C¯sα
k
s
and Cskαs can be arranged
into the following useful triangular tableau [26]:
φi
 S
7

 S
 S
C¯0α0 C0α0
7
  SS
7

/  S
C11α1 C¯1α1 C1α1
7
 7
  SS
7

/ 
/  S
C¯2α
2
2
C21α2 C¯2α2 C2α2
7
 7
 7
 7 SS


/ 
/ 
/  S
C33α3 C¯3α
2
3
C31α3 C¯3α3 C3α3

..
.
Diagram 2. The Triangular Field Tableau
The index s of Cskαs or C¯sα
k
s
labels different horizontal rows, whereas the index k
labels different rows slanting to the right and downward.
The new proper solution Snm of the classical master equation, involving the non-
minimal set of fields, is given by

Snm = S + Saux , (6.27)

J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 74

where S is the proper solution for the minimal set of fields presented in Sect. 4 and
Saux is the trivial-pair proper solution for the auxiliary fields given by

Saux = π̄0α0 C¯0∗α0 +

1∗
C1α π 1α1 + π̄1α1 C¯1∗α1 +
1 1

2 ¯2∗α2
π̄2α C
2 2
1∗
+ C2α π 1α2 + π̄2α2 C¯2∗α2 +
2 2

2 ¯2∗α3
3∗
C3α π 3α3 + π̄3α
3 3
C
3 3
1∗
+ C3α π 1α3 + π̄3α3 C¯3∗α3 +
3 3

...
or
L
X L
X L
X L
X
k ¯k∗αs k∗ kαs
Saux = π̄sα C
s s
+ Csα π
s s
. (6.28)
k=0 s=k k =1 s=k
k even k odd

The auxiliary antighost fields at level s are eliminated from the theory using
Eq.(6.1) which leads to
 
∂Ψ s
C¯sk∗αs = ¯k , k = 0, 2, 4, . . . , 2 ,
∂ Csαs 2
 
k∗ ∂Ψ s−1
Csα = , k = −1, 1, 3, 5, . . . , 2 +1 , (6.29)
s
∂Cskαs 2
at each level s.

6.3 Delta-Function Gauge-Fixing Procedure

k
Let us consider a gauge-fixing scheme for which Ψ is not a function of π̄sαs
or
kαs
πs (for k ≥ 0). This is called a δ-function gauge-fixing procedure because when
Eq.(6.29) is substituted in Saux in Eq.(6.28), the conditions

∂Ψ ∂Ψ
C¯sk∗αs = ¯k = 0 , k∗
Csα = =0 , (6.30)
∂ Csαs s
∂Cskαs

are implemented because the πskαs and π̄sα

k
s
in Eq.(6.28) act as Lagrange multipliers
k
for these equations. Indeed, in a functional
 
integral approach, integration over π̄sαs
 
∂Ψ
and πskαs leads to the insertion of δ k
∂ C̄sα
and δ ∂C∂Ψ
kαs in the integrand. Since Ψ
s s
k
does not depend on π̄sα s
or πskαs , π̄sk∗αs = ∂ π̄∂Ψ
k
k∗
= 0 and πsα s
= ∂π∂Ψ
kαs = 0.
sαs s
Not every Ψ is acceptable. For example, if Ψ is identically zero, all antifields are
set to zero and the gauge-fixed proper solution reduces to the original action S0 . This
S0 could not be used as the starting point for perturbation theory because propagators
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 75

would not exist. In the rest of this section, we determine the conditions on Ψ so that
an admissible gauge-fixing procedure arises. The discussion is somewhat technical so
that a reader may wish to skip to Sect. 6.5. For the irreducible case, one should read
to Eq.(6.40).
It is useful to define ns as the net number of non-gauge degrees of freedom for the
ghost Csαs at level s:
L
X
ns ≡ (−1)t−s mt . (6.31)
t=s

The number of gauge degrees of freedom of Csαs is ms − ns = ns+1 . If we define

m−1 = n and use the notation in Eq.(6.23), these statements also apply to φi when
s = −1: n−1 = m−1 − n0 = n − n0 = ndof is the number of non-gauge degrees of
freedom in φi .
In this paragraph, the general strategy is discussed. The basic idea is that the
conditions in Eq.(6.30) fix the gauge degrees of freedom of level s − 1 fields and non-
propagating independent degrees of freedom of level s + 1 fields. In the Triangular
Field Tableau of Diagram 2, arrows indicate how setting the antifield of one field to
zero via Eq.(6.30) eliminates degrees of freedom in another field. An upward-slanting
arrow going from ΦB to ΦA such as
ΦA
7


ΦB
Diagram 3. Elimination of Gauge Degrees of Freedom

indicates that the gauge degrees of freedom in ΦA are eliminated by the equation
∂Ψ
∂Φ∗B
= 0. To ensure their elimination, two points on the gauge slice determined
by Ψ should not be related by an infinitesimal gauge transformation. This leads to
conditions on Ψ that are presented below. A downward-slanting arrow going from
ΦA to ΦB such as
ΦA


/

ΦB
Diagram 4. Elimination of Non-Gauge Degrees of Freedom

indicates that non-propagating independent degrees of freedom in ΦB are eliminated

∂Ψ
by the equation ∂Φ ∗ = 0. When two arrows appear on a line, there is the fixing of
A
both independent and gauge degrees of freedom. In this delta-function gauge-fixing
procedure, only φi , Csαs and C¯sαs contain propagating fields. In the Triangular Field
Tableau of Diagram 2, there are no downward-sloping lines terminating on these fields
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 76

to indicate that non-gauge degrees of freedom are eliminated. In contrast, there are
upward-sloping lines meaning that the gauge degrees of freedom in φi , Csαs and C¯sαs
are fixed. The fields Cskαs , for k ≥ 1 and odd k, and C¯sαk
s
, for k ≥ 2 and even k, are
non-propagating, i.e., there is no quadratic form in the action for these fields. When
gauge-fixing is fully implemented, they are completely fixed. Like Csαs and C¯sαs , it will
be useful to think of Cskαs and C¯sαk
s
as having ns independent (or non-gauge) degrees
of freedom and ns+1 gauge degrees of freedom. Since there are both upward-sloping
and downward-sloping lines terminating on these fields, both gauge and independent
degrees of freedom are fixed. Hence, all ms = ns+1 + ns fields are removed from the
theory. The difference between Csαs and C¯sαs and the other ghosts and antighosts is
that the independent degrees of freedom in these fields are propagating. The fields
αs−1
C¯sαs play a dual role: They fix the gauge degrees of freedom in Cs−1 and serve as
αs
the antighosts of Cs , meaning that they enter quadratically in the action with these
fields.
Let Φ0 be a solution to the equations of motion for the action SΨ fixed by the
gauge conditions in Eq.(6.1). Since one is interested in performing a perturbative
expansion about this solution, write

ΦA = ΦA
0 + δΦ
A
, (6.32)

where δΦA is the quantum fluctuation of ΦA . According to the Triangular Field

Tableau of Diagram 2, the gauge degrees of freedom of φi should be fixed by the
equation 0 = C¯0∗α0 . Expanding about the perturbative solution, one obtains
! !
∂Ψ ∂Ψ ∂l ∂r Ψ
0 = C¯0∗α0 = ¯ = + δφi + . . . , (6.33)
∂ C0α 0 ∂ C¯0α0 0
∂ C¯0α ∂φi 0 0

where we use the abbreviation

()0 ≡ ()|{ΦA =ΦA } . (6.34)

0

 
∂l ∂r Ψ
The matrix ∂ C̄0α0 ∂φi
that multiplies δφi must eliminate n0 gauge degrees of free-
0
dom. This requires that Ψ be such that
!
∂l ∂r Ψ
rank = n0 . (6.35)
∂ C¯0α0 ∂φi 0

The gauge modes of φi are proportional to Rαi 0 . Hence one also needs
!
∂l ∂r Ψ i
rank R = n0 , (6.36)
∂ C¯0α0 ∂φi β0 0
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 77

to ensure that the correct degrees of freedom in φi are fixed. Using Eq.(4.28) and
∂Ψ
φ∗i = ∂φ ¯
i , one sees that the operator in Eq.(6.36) is the quadratic form for C0α0 and

C0β0 in the action SΨ :

!
∂l ∂r Ψ i
SΨ = . . . + C¯0α0 R C0β0 + . . . . (6.37)
∂ C¯0α0 ∂φi β0 0

Hence Eq.(6.36) is consistent with the fact that C¯0α0 and C0β0 should have n0 propa-
gating degrees of freedom.
If the system is irreducible, n0 = m0 and no further constraints on the gauge-
fixing fermion Ψ are necessary. Eqs.(6.35) and (6.36) are the only requirements on
Ψ. There are n − m0 propagating modes for φi and m0 propagating modes for C¯0α0
and C0α0 . Of the original n degrees of freedom of φi , m0 have been gauge fixed. The
simplest gauge-fixing fermion Ψδ is [26]
Ψδ = C¯0α0 χα0 (φ) , (6.38)
where χα0 is an arbitrary functional of φ. The subscript δ on Ψ indicates that this is
a delta-function gauge-fixing scheme. The condition 0 = C¯0∗α0 in Eq.(6.33) leads to
χα0 (φ) = 0 , (6.39)
so that χα0 is the gauge-fixing condition on φ. Eqs.(6.35) and (6.36) respectively
become  
rank χα,i0 (φ0 ) = n0 ,
 
rank χα,i0 Rβi 0 (φ0 ) = n0 , (6.40)
where n0 = m0 for the irreducible case.
Suppose the theory is at least first-stage reducible. Then C¯0α0 and C0α0 have m0
degrees of freedom but n1 of these are gauge modes and thus m0 − n1 = n0 are
independent. According to the Triangular Field Tableau of Diagram 2, the equation
C¯1∗α1 = 0 is used to fix the n1 gauge degrees of freedom in C0α0 , and C1α
1∗
1
= 0 is used to
¯
fix the n1 gauge degrees of freedom in C0α0 . Expanding the former condition about
the perturbative solution, one obtains
! !
∂Ψ ∂Ψ ∂l ∂r Ψ
0 = C¯1∗α1 = ¯ = + δC0α0 + . . . . (6.41)
∂ C1α1 ∂ C¯1α1 0
¯
∂ C1α1 ∂C0α0 0

Actually, the first term on the right-hand side of Eq.(6.41) vanishes, but we display it
to emphasize the idea that we are expanding about the perturbative solution. Since
one wants to fix n1 gauge modes of C0α0 , a necessary condition on Ψ is that
!
∂l ∂r Ψ
rank ¯ = n1 . (6.42)
∂ C1α1 ∂C0α0 0
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 78

The modes to gauge-fix are those that are non-propagating in Eq.(6.37). These modes
are proportional to 
−1α0 α0 
Z1β 1
≡ R 1β 1
 , (6.43)
0
of Eq.(2.16). Since there are n1 such modes,
 
−1α0
rank Z1β 1
= n1 . (6.44)

Because these modes need to be eliminated by the 0 = C¯1∗α1 condition

 in Eq.(6.41),
∂l ∂r Ψ
they should not be annihilated by the quadratic form ∂ C¯1α ∂C 0
α in Eq.(6.42). This
1 0 0
leads to the condition on Ψ that
!
∂l ∂r Ψ −1α0
rank ¯ α0 Z1β1 = n1 . (6.45)
∂ C1α1 ∂C0 0
∗ ∂r Ψ
Using C0α0
= in Eq.(4.28) and Eq.(6.43), one notices that the operator in
α
∂C0 0
Eq.(6.45) is the quadratic form for C¯1α1 and C1β1 . Since these fields have n1 prop-
agating degrees of freedom, the rank of the quadratic form should be n1 , which is
consistent with Eq.(6.45).
To fix the n1 gauge modes of C¯0α0 , use
! !
∂Ψ ∂Ψ ∂l ∂r Ψ
0= 1∗
C1α = = + δ C¯0α0 + ... . (6.46)
1
∂C11α1 ∂C11α1 0 ∂ C¯0α0 ∂C11α1 0

A necessary condition that these gauge degrees of freedom can be fixed is

!
∂l ∂r Ψ
rank = n1 . (6.47)
∂ C¯0α0 ∂C11α1 0
 
∂l ∂r Ψ
Next, note that, since rank ∂ C¯0α0 ∂φi
= n0 (see Eq.(6.35)) and the index α0 takes
0
0α1
on

m0 values,

there must be m0 − n0 = n1 left zero modes Z̄1α0
for the operator
∂l ∂r Ψ
∂ C¯0α0 ∂φi
:
0 !
0α1 ∂l ∂r Ψ
Z̄1α =0 , (6.48)
0
∂ C¯0α0 ∂φi 0
0α1
where we use a redundant superscript α1 on Z̄1α0 but compensate for this by requiring
 
0α1
rank Z̄1α0
= n1 . (6.49)
0α1
The Z̄1α0
are the non-propagating modes of C¯0α0 in the quadratic form in Eq.(6.37).
1∗
They must be fixed from the 0 = C1α 1
condition in Eq.(6.46), so that
!
0α1 ∂l ∂r Ψ
rank Z̄1β ¯ = n1 . (6.50)
0
∂ C0α0 ∂C11α1 0
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 79

Finally one needs to fix the n1 independent or “non-gauge” modes of C11α1 . From
Diagram 2, one sees that this is done using
! !
∂Ψ ∂Ψ ∂l ∂r Ψ
0= C¯0∗α0 = ¯ = + δC11α1 + . . . , (6.51)
∂ C0α0 ∂ C¯0α0 0 ∂ C¯0α0 ∂C11α1 0

for which one needs !

∂l ∂r Ψ
rank ¯ = n1 ,
∂ C0α0 ∂C11α1 0
which is the same condition as Eq.(6.47).
If the system is first-stage reducible then n1 = m1 and there are no gauge degrees
of freedom in C11α1 so that the above condition eliminates all the modes of C11α1 . For
φi , one has n − m0 + m1 propagating modes and m0 − m1 gauge modes fixed by gauge
conditions. For C¯0α0 and C0α0 , there are m0 −m1 propagating modes and the m1 gauge
degrees of freedom are fixed. Lastly, all m1 degrees of freedom of C¯1α1 and of C1α1 are
propagating. Summarizing, for a first-stage reducible theory the conditions on Ψ are
given in Eqs.(6.35), (6.36), (6.42), (6.45), (6.47), and (6.50).
An example of an acceptable Ψδ is [26]

Ψδ = C¯0α0 χα0 (φ) + C¯1α1 ω1α

0α1
0
(φ) C0α0 + C¯0α0 ω̄0α
1α0
1
(φ) C11α1 , (6.52)
0α1 1α0
where χα0 , ω1α0
and ω̄0α1
are arbitrary functionals of φ subject to Eq.(6.40) and
 
0α1
rank ω1α0
(φ0 ) = n1 ,
 
0α1 −1α0
rank ω1α0
(φ0 ) Z1β 1
= n1 ,
 
1α0
rank ω̄0α1
(φ0 ) = n1 ,
 
0α1 1α0
rank Z̄1α0
ω̄0α1 (φ0 ) = n1 , (6.53)
−1α0
where n1 = m1 for first-stage reducible theories. The Z1β 1
are given in Eq.(6.43)
0α1 0α1 α0
and Z̄1α0 is determined by Z̄1α0 χ,i (φ0 ) = 0.
For completeness, we present the conditions on Ψ for the general Lth-stage re-
ducible theory. There are fields from levels s = −1 to s = L. See Triangular Field
Tableau of Diagram 2.
For odd k and −1 ≤ k ≤ s, there are conditions fixing the ns+1 gauge degrees of
kαs
Cs coming from the diagram
Cskαs
7


C¯s+1α
k+1
s+1
Diagram 5. Elimination of Gauge Degrees of Freedom in Cskαs
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 80

and arising from

!
k+1∗α ∂Ψ ∂l ∂r Ψ
0 = C¯s+1 s+1 = = ...+ δCskαs + . . . . (6.54)
∂ C¯s+1α
k+1
s+1
∂ C¯s+1αs+1 ∂Cskαs
k+1
0

A necessary condition that the ns+1 gauge modes of Cskαs can be fixed is
!
∂l ∂r Ψ
rank = ns+1 , (6.55)
∂ C¯s+1αs+1 ∂Cskαs
k+1
0

where k is odd and ranges between s ≥ k ≥ −1, and s is restricted to L−1 ≥ s ≥ −1.
When there is a diagram such as
C¯s−1α
k−1
s−1



/
Cskαs
Diagram 6. Elimination of Independent Degrees of Freedom in Cskαs

the remaining ns independent degrees of freedom in Cskαs are fixed using

!
k−1∗α ∂Ψ ∂l ∂r Ψ
0= C¯s−1 s−1 = = ...+ δCskαs + . . . . (6.56)
∂ C¯s−1α
k−1
s−1
∂ C¯s−1αs−1 ∂Cskαs
k−1
0

A necessary condition that these ns non-gauge modes can be fixed is

!
∂l ∂r Ψ
rank ¯k−1 = ns , (6.57)
∂ Cs−1αs−1 ∂Cskαs 0

where k is odd, s ≥ k ≥ 1, and L ≥ s ≥ 1.

The gauge modes of Cskαs are those not fixed by the ! condition in Eq.(6.56). They
are annihilated by the quadratic form ∂ C¯k−1∂l ∂r Ψ∂C kαs in Eq.(6.56). Since this opera-
s−1αs−1 s
0
tor acts on ms -component vectors and has rank ns , it must have exactly ms −ns = ns+1
kαs
right null vectors. We denote these by Zs+1β s+1
using the redundant subscript index
βs+1 but requiring  
kαs
rank Zs+1β s+1
= ns+1 . (6.58)
More precisely, !
∂l ∂r Ψ kαs
¯k−1 Zs+1β =0 . (6.59)
∂ Cs−1αs−1 ∂Cskαs 0
s+1

kαs
A sufficient condition that these gauge modes Zs+1β s+1
can be removed from Cskαs via
Eq.(6.54) is that
!
∂l ∂r Ψ
rank Z kαs = ns+1 . (6.60)
∂ C¯s+1αs+1 ∂Cskαs s+1βs+1
k+1
0
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 81

The conditions on Ψ in Eq.(6.60) hold for odd k, s ≥ k ≥ −1 and L − 1 ≥ s ≥ −1 if

one defines 
−1αs αs 
Zs+1β s+1
≡ Rs+1β s+1
 . (6.61)
0

When k = −1 the operator in Eq.(6.60) is the quadratic form of SΨ for C¯s+1αs+1 and
βs+1
Cs+1 , as can be seen from Eqs.(4.28) and (6.29). The condition that these fields have
ns+1 propagating modes is consistent with Eq.(6.60) for the case k = −1.
The discussion for C¯sα k
s
where k is even works similarly. The ns independent
degrees of freedom of C¯sαs , via the diagram
k−1αs−1
Cs−1

/


C¯sα
k
s
Diagram 7. Elimination of Independent Degrees of Freedom in C¯sα
k
s

are fixed by
 
∂Ψ ∂l ∂r Ψ
k−1∗
0 = Cs−1α = k−1αs−1 = . . . + δ C¯sα
k 
k−1αs−1
 + ... , (6.62)
∂ C¯sα
s−1 s k ∂C
∂Cs−1 s s−1 0

which requires  
∂l ∂r Ψ
rank   = ns .
k ∂C k−1αs−1
∂ C¯sαs s−1 0

This is the same condition as in Eq.(6.55) when s → s + 1 and k → k + 1. !

The gauge modes of C¯sα
k
s
are the left zero modes of the operator k ∂C
∂l ∂r Ψ
k−1αs−1
∂ C̄sαs s−1 0
in Eq.(6.62). Since this operator acts to the left on ms -component vectors and has
rank ns , there are exactly ms − ns = ns+1 such zero modes. We denote these by
kαs+1
Z̄s+1αs
, using the redundant superscript index αs+1 but requiring
 
kα
s+1
rank Z̄s+1αs
= ns+1 . (6.63)

kαs+1
More precisely, the Z̄s+1αs
are defined by
 
kα
s+1 
∂l ∂r Ψ 
Z̄s+1α =0 , (6.64)
s
k ∂C k−1αs−1
∂ C¯sαs s−1 0

where k is even, s − 1 ≥ k ≥ 0, and L − 1 ≥ s ≥ 0. These ns+1 gauge degrees of

freedom of C¯sα
k
s
, via the diagram
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 82

C¯sα
k
s

7



k+1α
Cs+1 s+1
Diagram 8. Elimination of Gauge Degrees of Freedom in C¯sα
k
s

are fixed by
 
∂Ψ ∂l ∂r Ψ
k+1∗
0 = Cs+1α = k+1α = . . . + δ C¯sα
k 
k+1α
 + ... , (6.65)
s+1
∂Cs+1 s+1
s
¯k
∂ Csαs ∂Cs+1 s+1 0

which requires  
∂l ∂r Ψ
rank  k+1αs+1
 = ns+1 .
∂ C¯sα
k ∂C
s s+1 0

This is the same condition as in Eq.(6.57) when s → s − 1 and k → k − 1. Since these

kβs+1
gauge modes are proportional to Z̄s+1α s
, a sufficient condition for the elimination of
these modes is  
kβs+1 ∂ ∂
l r Ψ
rank Z̄s+1α k+1α
 = ns+1 , (6.66)
s
¯k
∂ Csα ∂Cs+1 s+1
s 0
where k is even, s − 1 ≥ k ≥ 0, and L − 1 ≥ s ≥ 0.
As a final remark, note that all the gauge-fixing conditions have been used: The
equations 0 = Csαk∗
s
(and 0 = C¯sk∗αs ) impose ns constraints on level s−1 gauge fields and
ns+1 constraints on level s + 1 independent fields. The total number of constraints
imposed is the number of values of the index αs . This number is ms and equals
ns + ns+1 .
For an Lth-stage reducible theory, it is necessary for Ψ to satisfy Eqs.(6.55), (6.57),
(6.60) and (6.66). All degrees of freedom in Cskαs for k odd and k ≥ 1 and in C¯sα k
s
for k
i
even and k ≥ 2 are fixed. Of the original φ , n0 of the degrees of freedom are fixed and
the remaining n − n0 modes are propagating. For C¯sαs and Csαs with 0 ≤ s ≤ L − 1, ns
fields are propagating and ns+1 fields have been gauge-fixed. For s = L, all nL = mL
modes of C¯LαL and CLαL are propagating.
One useful example of a gauge-fixing fermion is [26]

Ψδ = C¯0α0 χα0 (φ) +

L
X s
X L
X s
X
k−1α
C¯sα
k
ω kαs (φ) Cs−1 s−1 +
s sαs−1
C¯s−1α
k−1
ω̄ kαs−1 (φ) Cskαs
s−1 sαs
, (6.67)
s=1 k=0 s=1 k=1
k even k odd

where, for each upward-sloping line (like in Diagram 5 with k → k − 1 and s → s − 1),
kαs
one associates the matrix ωsα s−1
and where, for each downward-sloping line (like in
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 83

kαs−1
Diagram 6), one associates the matrix ω̄sα s
. In addition to Eq.(6.40), one requires
Eqs.(6.55), (6.60), Eq.(6.57), and (6.66), that is,
 
kαs
rank ωsαs−1
(φ0 ) = ns ,
 
kαs k−1αs−1
rank ωsαs−1
(φ0 ) Zsβs = ns , (6.68)
for k even, s ≥ k ≥ 0 and L ≥ s ≥ 1, as well as
 
kαs−1
rank ω̄sαs
(φ0 ) = ns ,
 
k−1βs kαs−1
rank Z̄sαs−1
ω̄sαs (φ0 ) = ns , (6.69)
for k odd, s ≥ k ≥ 1 and L ≥ s ≥ 1.

6.4 Other Gauge-Fixing Procedures

The previous subsection considered the most general gauge-fixing fermion Ψ which
k
is independent of π̄sαs
and πskαs . By allowing Ψ to be linear in π̄sα
k
s
and πskαs , one is
able to have gaussian averaging over a gauge condition [26]. Such possibilities arise
when
Ψ = Ψδ + Ψπ , (6.70)
k
where Ψδ is a δ-function-type gauge-fixing fermion which is independent of π̄sαs
and
kαs k kαs
πs , and where Ψπ is linear in π̄sαs and πs .
For example, in an irreducible theory, one may take
1
Ψπ = C¯0α0 σ00α0 β0 (φ) π̄0β0 , (6.71)
2
where σ00α0 β0 is an arbitrary matrix. The total gauge-fixing fermion Ψ is
1
Ψ = C¯0α0 χα0 (φ) + C¯0α0 σ00α0 β0 (φ) π̄0β0 . (6.72)
2
The antifields C¯0∗α0 are eliminated via Eq.(6.1) to give
1
C¯0∗α0 = χα0 + σ00α0 β0 π̄0β0 , (6.73)
2
which, when substituted into Eq.(6.28), results in
1
Saux |ΣΨ = π̄0α0 χα0 (φ) + π̄0α0 σ00α0 β0 π̄0β0 . (6.74)
2
Without the quadratic functional of π̄0α0 , Eq.(6.74) would give the gauge-fixing condi-
tions χα0 (φ) = 0 on φ. With the quadratic functional of π̄0α0 , one is able to perform
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 84

a gaussian averaging over these gauging-fixing conditions. If the matrix σ00α0 β0 is

completely invertible then the averaging is done over all gauge invariances. By choos-
ing σ00α0 β0 not to have an inverse, one may obtain delta-function conditions for some
gauge invariances and a gaussian-averaging of others. In some quantization schemes,
π̄0α0 may become propagating, in which case π̄0α0 is called a Nielsen-Kallosh ghost
[172, 199, 200].
For a first-stage reducible theory, one can add
1 1 1
Ψπ = C¯0α0 σ00α0 β0 π̄0β0 + C¯1α1 ρ0α1 1β1
1β1 π1 + π̄1α1 ρ0α1 1β1
1β1 C1 (6.75)
2 2 2

to the δ-function gauge-fixing fermion Ψδ in Eq.(6.52), where σ00α0 β0 and ρ0α

1β1 are
1

arbitrary matrix functionals of φ. Eliminating antifields by Eq.(6.1) and inserting

into Eq.(6.28) leads to
1
Saux |ΣΨ = . . . + π̄0α0 σ00α0 β0 π̄0β0 + π̄1α0 ρ0α1 1β1
1β1 π1 + ... . (6.76)
2

For appropriate σ00α0 β0 and ρ0α 1β1

1β1 , integration over π̄0α0 , π1
1
and π̄1α0 leads to a gaus-
sian averaging gauge-fixing procedure.
For an Lth-stage system, the most general Ψπ , satisfying Eq.(6.4) and being linear
in πskαs and π̄sα
k
s
and linear in the auxiliary ghosts and antighosts, is

1 L−1
X L
X  
Ψπ = C¯sα
k
ρkαs (φ) πsk+1βs + π̄sα
s sβs
k
ρkαs (φ) Csk+1βs
s sβs
2 k=0 s=k+1
k even

L
1 X
+ C¯kα
k
σ kαk βk (φ) π̄kβ
k k
k
, (6.77)
2 k=0 k

k even
kαk βk
where ρkα
sβs and σk
s
are arbitrary matrices. If one eliminates antifields by Eq.(6.1)
and inserts the result into Eq.(6.28), one obtains the following quadratic terms for π
fields
L−1
X L
X
Saux |ΣΨ = . . . + k
π̄sα ρkαs πsk+1βs
s sβs
k=0 s=k+1
k even
L
1 X
+ k
π̄kα σ kαk βk π̄kβ
k k
k
+ ... , (6.78)
2 k=0 k

k even

thereby allowing a gaussian average procedure for gauge invariances at all levels.
In some instances, in lieu of Eq.(6.1), one might want to eliminate certain fields
rather than antifields. The simplest way to accomplish this is to first perform a
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 85

canonical transformation that interchanges some fields for antifields [104, 230, 259,
44, 252, 257]. Then, when Eq.(6.1) is used, elimination of certain antifields will
correspond to the elimination of some original fields. Given the freedom to first
perform canonical transformations, Eq.(6.1) is quite general.

6.5 Gauge-Fixed Classical BRST Symmetry

The gauge-fixed theory inherits a remnant of the original BRST symmetry δB in
Eq.(4.32). Given Ψ, one defines the gauge-fixed BRST operator δBΨ by [24]

δBΨ X ≡ (X, S)|ΣΨ , (6.79)

where X is any functional of the ΦA . For X = ΦA , one obtains


∂l S 
δBΨ ΦA =  . (6.80)
∂Φ∗A ΣΨ

This is the same result as first performing the non-gauge-fixed BRST transformation
in Eq.(4.32) and then imposing the gauge-fixing fermion condition in Eq.(6.1). The
BRST transformation is a global symmetry of the classical gauge-fixed action since

∂r SΨ ∂l S 
δBΨ SΨ = (SΨ , S)|ΣΨ =  =
∂ΦA ∂Φ∗A ΣΨ
! 
∂r S ∂l S  ∂r S ∂r ∂r Ψ ∂l S 

 +  =0 , (6.81)
∂ΦA ∂Φ∗A 
ΣΨ
∂Φ∗B ∂ΦA ∂ΦB ∂Φ∗A ΣΨ
where the first term vanishes as a consequence of the classical master equation and
the second term vanishes due to statistical symmetry properties of the factors.
Due to the elimination of antifields, δBΨ is not necessarily off-shell nilpotent. For
example, consider
 
∂r ∂l S ∂l S  ∂r ∂l S ∂r ∂l Ψ ∂l S 
δB2 Ψ ΦA =  +  ,
∂ΦC ∂Φ∗A ∂Φ∗C ΣΨ ∂Φ∗C ∂Φ∗A ∂ΦD ∂ΦC ∂Φ∗D ΣΨ

where the second term arises from the chain rule (see Eq.(A.7) of Appendix A) via
the implicit dependence on ΦD through Φ∗C , which is now a functional of the Φ via
Eq.(6.1). By using the identity 0 = ∂l∂Φ
(S,S)
∗ , the first term can be rearranged and then
A
the two terms combine if one notes that
 
∂r (SΨ ) ∂r S  ∂r S ∂r ∂l Ψ 
=  +  . (6.82)
∂ΦC ∂ΦC ΣΨ ∂Φ∗D ∂ΦC ∂ΦD ΣΨ
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 86

The final result is [29, 30, 251]


ǫA ∂r (SΨ ) ∂l ∂l S 
δB2 Ψ ΦA = (−1)  . (6.83)
∂ΦC ∂Φ∗C ∂Φ∗A ΣΨ

The equations of motion for the gauge-fixed action are ∂r∂Φ (SΨ )
C = 0. Hence, the gauge-
fixed BRST symmetry is on-shell nilpotent. From Eq.(4.29), one sees that a necessary
ji
condition for off-shell nilpotency is that the algebra is closed, Eαβ = 0, and that there
ji
is off-shell reducibility, Va = 0.
Another way of obtaining the on-shell nilpotency is to note that the gauge-fixed
BRST transformation agrees with the ordinary BRST transformation evaluated at
ΣΨ , up to equations of motion. For functionals which do not depend on antifields
this follows from Eq.(6.79). For functionals, which depended on antifields before
gauge-fixing but are now evaluated at ΣΨ , the result follows because
∂l (SΨ )
δBΨ Φ∗A (Φ) = (δB Φ∗A )|ΣΨ + . (6.84)
∂ΦA
The last term is a gauge-fixed equation of motion. In other words, the processes
of performing the off-shell BRST transformation in Eq.(4.32) and gauge-fixing via
Eq.(6.1) commute up to equations of motion.

6.6 The Gauge-Fixed Basis

Another way to gauge-fix the theory is to first perform a canonical transformation
to the “gauge-fixed” basis and then set antifields to zero [230, 231, 259, 44, 252]. The
canonical transformation in Eq.(4.13) is done with F2 = ΦA Φ̃∗A + Ψ (Φ) to give
∂Ψ
ΦA = Φ̃A , Φ∗A = Φ̃∗A + , (6.85)
∂ΦA
∂Ψ
so that ΦA → ΦA and Φ∗A → Φ∗A + ∂Φ A . The gauge-fixed basis consists of the new

tilde fields and antifields.

In this new basis, the original gauge invariances are replaced by the classical
gauge-fixed BRST symmetry δBΨ . It emerges as the symmetry of the gauge-fixed
action SΨ . To understand how this comes about, expand the original proper solution
S in a power series in antifields via
h i 1
S [Φ, Φ∗ ] = S0 φi + Φ∗A RA + Φ∗A Φ∗B RBA + . . . , (6.86)
2
and perform the change of variables in Eq.(6.85) to obtain
h i ∂Ψ A 1 ∂Ψ ∂Ψ BA
S [Φ, Φ∗ ] = S0 φi + R + R + ...
∂ΦA 2 ∂ΦA ∂ΦB
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 87

!
∂Ψ BA
+Φ̃∗A A
R + R + ...
∂ΦB
1  
+ Φ̃∗A Φ̃∗B RBA + . . . + . . .
2
1 h i
≡ S̃0 [Φ] + Φ̃∗A R̃A + Φ̃∗A Φ̃∗B R̃BA + . . . ≡ S̃ Φ, Φ̃∗ . (6.87)
2
The terms in the Φ̃∗ expansion have an interpretation. The Φ̃∗ -independent term

S̃0 [Φ] = S|ΣΨ ≡ SΨ (6.88)

is the gauge-fixed action. The term linear in Φ̃∗

 
A ∂l S̃  ∂l S 
R̃ = ∗ 
 = ∗ 
 ≡ δBΨ ΦA
∂ Φ̃A {Φ̃∗ =0} ∂ΦA ΣΨ
B

∂r X A
is the generator of gauge-fixed BRST transformations since δBΨ X = ∂Φ A R̃ . Hence,

S̃ is a theory in which gauge transformations have been replaced by the gauge-fixed

BRST transformation δBΨ . As in Sect. 2, one can define tensors associated with the
gauge structure. It is necessary to absorb ghost fields C into their definition, i.e.,
A ˜B BA ˜C ˜D
R̃A ≡ R̃B C , Ẽ BA ≡ 12 ẼDC C C , etc.. Since δB2 Ψ is zero only on-shell, the algebra is
open and there exists a non-zero tensor field Ẽ BA . Comparing Eqs.(2.23) and (6.83),
one concludes 
BA ǫA ∂l ∂l S 
Ẽ = (−1)  . (6.89)
∂Φ∗B ∂Φ∗A ΣΨ
For an open algebra, according to Eq.(4.29), Ẽ BA (−1)ǫA is the  coefficient of the 
BA ∂l ∂l S̃  ∂l ∂l S 
proper solution quadratic in antifields. Hence, R̃ ≡ ∂ Φ̃∗ ∂ Φ̃∗  = ∂Φ∗ ∂Φ∗ 
B A {Φ̃∗B =0} B A Σ
Ψ
BA ǫA
should be equal to Ẽ (−1) . This is indeed true since it is equivalent to Eq.(6.89).
Hence, the coefficient of S̃ which is quadratic in tilde antifields is the tensor Ẽ BA
associated with the fact that the gauge-fixed BRST transformations form an open
algebra.
The classical master equation for S̃ encodes the algebraic structure of the gauge-
fixed BRST transformation [133]. A straightforward computation gives
!
1  ∂r SΨ A ∂r R̃B A ∂r SΨ
0= S̃, S̃ = A
R̃ + Φ̃∗B A
R̃ + R̃BA + ... . (6.90)
2 ∂Φ ∂Φ ∂ΦA

Since each coefficient of the antifield expansion must be zero,

∂r SΨ A
0= R̃ = δBΨ SΨ , (6.91)
∂ΦA
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 88

which is the invariance of the gauge-fixed action under δBΨ . The vanishing of the
second coefficient in Eq.(6.90) leads to the equation

∂r R̃B A BA ∂r SΨ
R̃ = −R̃ . (6.92)
∂ΦA ∂ΦA
Eq.(6.92) is related to the on-shell nilpotency of δBΨ :
!
∂r ∂r SΨ A
0 = δBΨ (δBΨ SΨ ) = R̃ R̃B =
∂ΦB ∂ΦA

∂r SΨ ∂r R̃A B (ǫA +1)ǫB ∂r ∂r SΨ

R̃ + (−1) R̃A R̃B .
∂ΦA ∂ΦB ∂ΦB ∂ΦA
The second term vanishes by the statistical properties of the factors. The first term
vanishes when Eq.(6.92) is used  inconjunction with statistical symmetry arguments.
Higher order terms in 0 = S̃, S̃ represent other consistency requirements of the
gauge-fixed classical BRST transformation.
In addition to the gauge structure, other concepts generalize to the gauge-fixed
basis. In the effective action or any other approach for which antifields remain present,
antifields are interpreted as the sources for gauge-fixed BRST transformations. See
Sect. 8.4 for more discussion. In the gauge-fixed basis, the classical limit in Eq.(4.24)
is replaced by the gauge-fixed condition
h i
S̃ Φ, Φ̃∗  = SΨ [Φ] (6.93)
Φ̃∗ =0

of Eq.(6.88). The proper condition in Eqs.(4.22) and (4.23) becomes [259, 44, 252, 133]

∂l ∂r SΨ [Φ] 
rank  =N . (6.94)
∂ΦA ∂ΦB { ∂r SΨ =0}
∂ΦD

Eq.(6.94) ensures that propagators for the ΦA are defined so that the usual pertur-
bation theory can be developed.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 89

7 Gauge-Fixing Examples
In this subsection, we present some gauge-fixing procedures for the examples
considered in Sects. 3 and 5. The first step is to introduce the auxiliary fields and
antifields necessary for gauge-fixing, as specified in Sect. 6.2. The second step is to
choose an appropriate Ψ. Here, there is quite a bit of freedom and we make specific
choices. The third step is to eliminate antifields using Ψ and substitute the results
into S to obtain the gauge-fixed action SΨ .
The reader interested in doing exercises can try the following. (i) Derive the
equations of motion of the gauge-fixed action. (ii) Determine the effect of the gauge-
fixed BRST transformation δBΨ on the fields. (iii) Check the gauge-fixed BRST
invariance of the gauge-fixed action. (iv) Compute δB2 Ψ and verify that the non-
zero terms, when present, vanish if the equations of motion are used. (v) Perform a
gauge-fixing procedure with a different gauge-fixing fermion Ψ.

7.1 The Spinless Relativistic Particle

Since this system is irreducible, one only needs to append one non-trivial pair
¯ π̄}. The non-minimal proper solution is
{C,
Z ( ! )
1 ẋ2 C
S= dτ − em2 + x∗µ ẋµ + e∗ C˙ + π̄ C¯∗ . (7.1)
2 e e

An example of a gauge-fixing fermion is

Z
Ψ= dτ C¯ (e − 1) , (7.2)

which, via Eq.(6.1), leads to

C¯∗ = e − 1 , e∗ = C¯ , x∗µ = 0 , π̄ ∗ = 0 . (7.3)

The gauge-fixed action is

Z ( ! )
1 ẋ2
SΨ = dτ − em2 + C¯C˙ + π̄ (e − 1) . (7.4)
2 e

Because the original algebra is closed, the gauge-fixed BRST transformation δBΨ ,
given by
ẋµ C
δBΨ e = C˙ , δBΨ xµ = , δBΨ C = 0 ,
e
δBΨ π̄ = 0 , δBΨ C¯ = π̄ , (7.5)
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 90

satisfies δBΨ δBΨ = 0 off-shell. After integrating over π̄, the gauge e = 1 is implemented
and Eq.(7.4) becomes
Z  
1 2 
SΨ → dτ ẋ − m2 + C¯C˙ . (7.6)
2
Since the equations of motion for π̄ and e have been used, Eq.(7.6) is no longer
invariant under Eq.(7.5). To derive the modified BRST transformations, start with
Eq.(7.1), and perform the shift, C¯∗ → C¯∗ + e − 1, e∗ → e∗ + C, ¯ to the gauge-fixed
∂r SΨ
basis. Then, determine the equation of motion for π̄ using ∂e = 0 and perform
another canonical transformation to shift π̄ by this solution: π̄ → π̄ + 12 (ẋ2 /e2 + m2 ).
The gauge-fixed BRST transformations of the transformed action at the saddle point,
e = 1, π̄ = 0, given by
1 2 
δB̃Ψ xµ = ẋµ C , δB̃Ψ C = 0 , δB̃Ψ C¯ = ẋ + m2 , (7.7)
2
constitute a symmetry of the action in Eq.(7.6), as can easily be checked. Because
equations of motion have been used, δB̃Ψ is not longer nilpotent off-shell. Indeed, a
computation of δB̃2 Ψ reveals that it is, in general, nonzero; it is zero if the equations
of motion, ẍµ = 0 and C˙ = 0 are used.

7.2 Yang-Mills Theories

Since this is an irreducible system, it is necessary to introduce one trivial pair
¯
{Ca , π̄a } for each generator index a. Frequently, π̄a is denoted by Ba and we adopt
this notation. The non-minimal proper solution is
Z  
1 a µν 1
S= d
d x − Fµν Fa + A∗aµ D µa b C b + Cc∗ fab c C b C a + Ba C¯∗a . (7.8)
4 2
To illustrate a gaussian-averaging gauge-fixing procedure, we choose
Z !
a
B
Ψ= d xC¯a −
d
+ ∂ µ Aaµ , (7.9)
2ξ
where ξ is a parameter. Elimination of antifields via Eq.(6.1) gives
B a
C¯a
C¯∗a = − + ∂ µ Aaµ , Ba∗ = − ,
2ξ 2ξ
A∗aµ = −∂µ C¯a , Ca∗ = 0 . (7.10)
The gauge-fixed action SΨ is
Z ( ! )
1 a µν Ba
SΨ = d
d x − Fµν Fa − ∂µ C¯a D µa b C b − − ∂ µ Aaµ B a . (7.11)
4 2ξ
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 91

The gauge-fixed BRST transformations are

δBΨ Aaµ = D µa b C b ,
1
δBΨ C a = fbc a C c C b ,
2
δBΨ C¯a = Ba , (7.12)
δBΨ Ba = 0 .
The nilpotency of δBΨ holds off-shell because the original gauge algebra is closed. The
gaussian integration over Ba can be performed for Eq.(7.11) to give
Z ( )
1 a µν ξ
SΨ → d
d x − Fµν Fa − ∂µ C¯a D µa b C b + ∂ µ Aaµ ∂ ν Aaν , (7.13)
4 2

which is known as the Yang-Mills action fixed in the Rξ gauge [1]. The case ξ = 1
is the Feynman gauge. When ξ → ∞, the π̄a = Ba dependence in Ψ of Eq.(7.9)
disappears and the Landau gauge ∂ µ Aaµ = 0 is imposed as a delta-function condition.
Because the quadratic forms in Eq.(7.13) are non-degenerate, Ψ is an admissible
gauge-fixing fermion. Propagators exist and Eq.(7.13) can be used as an action for
the Yang-Mills perturbation series [1, 165].
The BRST symmetry of Eq.(7.13) is determined by the procedure described at
the end of Sect. 7.1. One performs canonical transformations to the gauge-fixed basis
and then to the classical solution for B a . The latter is determined by varying the
action with respect to B a itself, and leads to the shift B a → B a + ξ∂ µ Aaµ . One finds
that the BRST symmetry of Eq.(7.13) is given by
1
δB̃Ψ Aaµ = D µa b C b , δB̃Ψ C a = fbc a C c C b , δB̃Ψ C¯a = ξ∂ µ Aaµ . (7.14)
2
The gauge-fixed BRST generator δB̃Ψ is only nilpotent on-shell.

7.3 Topological Yang-Mills Theory

By using a redundant set of gauge transformations, we have rendered this theory
first-stage reducible. The minimal-field proper solution is given in Eq.(5.12). The
gauge parameters are Λa and εaµ . The number of level 1 gauge invariances is the
rank of the group. Correspondingly, we need to introduce the trivial pairs {C¯a , π̄a }
and {C¯aµ , π̄aµ } at level 0, and {C11a , π11a } and {C¯1a , π̄1a } at level 1. In keeping with the
notation of Sects. 3.3 and 5.3, we use η instead of C for level 1 ghosts: C¯1a → η̄a and
C11a → η 1a . We also use B instead of π̄ at level 0: π̄a → Ba and π̄aµ → Baµ , and Λ
instead of π at level 1: π̄1a → Λ̄a and π11a → Λ1a .
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 92

To the minimal proper solution in Eq.(5.12), we add

Z  
Saux = d4 x Ba C¯∗a + Baµ C¯∗aµ + ηa1∗ Λ1a + Λ̄a η̄ ∗a . (7.15)

For the gauge-fixing fermion, we select

Z  
Ψ= d4 x C¯a ∂ µ Aaµ + C¯aµ Aaµ + η̄a C a + η 1a C¯a . (7.16)

The elimination of antifields using Ψ gives

C¯∗a = ∂ µ Aaµ + η 1a , C¯∗aµ = Aaµ ,

ηa∗ = 0 , ηa1∗ = C¯a , η̄ ∗a = C a ,

A∗aµ = −∂µ C¯a + C¯aµ , Ca∗ = η̄a , C¯a∗ = η 1a . (7.17)
The gauge-fixed action becomes
Z 
1 a ∗ aµν   
SΨ = 4
dx Fµν F + −∂µ C¯a + C¯aµ D µa b C b + C aµ
4
1
+ η̄c fab c C b C a + Acµ fab c C bµ C a + (η̄b − Aaµ D µa b ) η b +
2
  o
Ba ∂ µ Aa + η 1a + Baµ Aaµ + C¯a Λ1a + Λ̄a C a
µ . (7.18)
The integration over Baµ produces δ (Aaµ ) in the integrand of the functional integral.
aµ
This delta function can be used to perform  the A integral. Then the integration
over Ba , Λ1a and Λ̄a produces δ (η 1a ) δ C¯a δ (C a ) which can be used to do the η 1a , C¯a
and C a integrations. The gauge-fixed action is reduced to
Z  
SΨ → d4 x C¯aµ C aµ + η̄a η a . (7.19)

The integration over C¯aµ and η̄a leads to δ (C aµ ) δ (η a ) which can be used to do the C aµ
and η a integrals. All the functional integrals over fields have been performed, leaving
no local degrees of freedom in the action. Not too surprisingly, the integration over
the topological action leads to a trivial lagrangian. The functional integral produces
a finite number. It is important that all the terms in Ψ in Eq.(7.16) be present. If the
last term η 1a C¯a were dropped from Ψ, then the integrations over Λ1a and C¯a would
produce singular contributions.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 93

7.4 The Antisymmetric Tensor Field Theory

Like the previous example, this is a first-stage reducible theory, for which one
needs to add one trivial pair {C¯aµ , π̄aµ } at level 0 and two trivial pairs {C11a , π11a } and
{C¯1a , π̄1a } at level 1. As in the previous subsections, we use B for π̄ at level 0, and use
η and Λ respectively for C1 and π1 at level 1. To the minimal-field proper solution,
one adds the auxiliary action
Z  
Saux = d4 x Baµ C¯∗aµ + Λ̄a η̄ ∗a + ηa1∗ Λ1a . (7.20)

An appropriate gauge-fixing fermion is

Z  
1
Ψ= d x C¯aµ εµνρσ ∂ν Bρσ
4 a
+ η̄a ∂ µ Cµa + ∂ µ C¯aµ η 1a . (7.21)
2
The elimination of antifields using Ψ gives
1
C¯∗aµ = εµνρσ ∂ν Bρσ
a
− ∂ µ η 1a , ηa∗ = 0 , η̄ ∗a = ∂ µ Cµa , ηa1∗ = ∂ µ C¯aµ ,
2
Ba∗ρσ = ερσµν ∂µ C¯aν , Ca∗µ = −∂ µ η̄a , A∗aµ = 0 . (7.22)
The gauge-fixed action is
Z   
1
SΨ = S0 + d4 x − ∂ν C¯aµ D νa b C µb − D µa b C νb
2
1
−∂µ η̄a D µa b η b + ∂µ C¯cν ∂ρ C¯σb εµνρσ fab c η a +
4
  
1 µνρσ
Baµ ε a
∂ν Bρσ − ∂ µ η 1a + Λ̄a ∂ µ Cµa + ∂ µ C¯aµ Λ1a , (7.23)
2
where S0 is given in Eq.(3.27). One interesting feature of this example is the appear-
ance of a trilinear ghost term. It originates from the gauge-fixing of the bilinear term
in antifields in Eq.(5.14). The fields Baµ , Λ̄a and Λ1a are Lagrange multipliers for the
gauge conditions 21 εµνρσ ∂ν Bρσ
a
− ∂ µ η 1a = 0, ∂ µ Cµa = 0 and ∂ µ C¯aµ = 0. If one wants to
perform a gaussian averaging over these conditions, one adds
Z !
1 4 C¯aµ B aµ η̄a Λ̄a Λ1a η 1a
Ψπ = − dx + ¯ +
2 ξ ξ ξ1

to Ψ in Eq.(7.21). Here, ξ, ξ¯ and ξ 1 are constants. One could also replace ordinary
derivatives ∂µ by covariant ones Dµ in Eq.(7.21). Then Dµ replaces ∂µ in Eqs.(7.22)
and (7.23). In this case, A∗aµ 6= 0, but since A∗aµ does not enter the proper solution in
Eq.(5.14), the gauge-fixed action is as in Eq.(7.23) with ∂µ → Dµ .
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 94

7.5 Open String Field Theory

Since the open bosonic string is an infinite-reducible theory, one introduces the
extraghosts Csk for k = 1, 3, 5, . . . and s ≥ k, and C¯sk for k = 0, 2, 4, . . . and s ≥ k in
the Triangular Field Tableau of Diagram 2 for all levels s:
A
 S
7

 S
 S
C¯0 C0
7
  SS
7
/

 
 S
C11 C¯1 C1
7
 
7 7 SS

/

 /

 
 S
C¯22 C21 C¯2 C2
7
 7
 7
 7 SS

/

 /

 /

 
 S
C33 C¯32 C31 C¯3 C3

..
.
Diagram 9. The Triangular String Field Tableau
In addition, there are the trivial-pair partners πsk for k = 1, 3, 5, . . . and s ≥ k, and
π̄sk for k = 0, 2, 4, . . . and s ≥ k. Although there are no αi type indices on Csk , C¯sk ,
πsk and π̄sk , they are string fields and represent an infinite tower of ordinary particle
fields. The ghost numbers of the fields are
h i h i    
gh Csk = gh C¯sk∗ = s − k , g Csk = g C¯sk∗ = 1 + k − s ,
h i h i    
gh C¯sk = gh Csk∗ = k − s − 1 , g C¯sk = g Csk∗ = 1 + k − s ,
h i  
gh π̄sk = k − s , g π̄sk = 1 + k − s ,
h i  
gh πsk = s − k + 1 , g πsk = 1 + k − s , (7.24)
where gh[ ] is the field-antifield ghost number and g( ) is the string ghost number.
In Eq.(7.24), we use the abbreviations in Eq.(6.23) for the cases when k = −1 and
k = 0. All Csk , C¯sk∗ and π̄sk have odd total statistics, while C¯sk , Csk∗ and πsk have even
total statistics.
The auxiliary action Saux in Eq.(6.28) is
∞
X ∞ Z
X ∞
X ∞ Z
X
Saux = ∗ k
π̄s ∗ C¯sk∗ + ∗ k∗
Cs ∗ πsk , (7.25)
k=0 s=k k=1 s=k
k even k odd
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 95

where ∗ before a field is the string Hodge star operation. For an arbitrary field or
antifield ϕ, gh [ ∗ ϕ] = gh [ϕ] but g ( ∗ ϕ) = 3 − g (ϕ). The action in Eq.(7.25) is to be
added to the minimal-field proper solution S in Eq.(5.24).
The most convenient gauge for the open bosonic string field theory is the Siegel-
Feynman gauge [229] which imposes the condition c̄0 A = 0 on the string field A. When
R
implemented, the quadratic action of the massless vector Aµ becomes 12 d26 x Aµ ∂ν ∂ ν Aµ .
Above, c̄0 is the zero mode of the antighost of first-quantized open string theory in
Sect. 3.8 and Sect. 5.8. It is the conjugate momentum of the zero mode c0 of the
first-quantized ghost. These modes satisfy

{c0 , c̄0 } = c0 c̄0 + c̄0 c0 = 1 , c0 c0 = 0 , c̄0 c̄0 = 0 , (7.26)

and have string ghost numbers of g (c0 ) = 1 and g (c̄0 ) = −1. See Sect. 7.7 for more
discussion. The gauge fermion Ψsf , which implements the Siegel-Feynman gauge, has
an expansion that, for the first few levels, reads
Z
Ψsf = ∗
C¯0 ∗ c0 c̄0 A
Z Z
∗
C¯1 ∗ c0 c̄0 C0 + ∗ 1
C1 ∗ c0 c̄0 C¯0 +
Z Z Z
∗
C¯2 ∗ c0 c̄0 C1 + ∗ 1
C2 ∗ c0 c̄0 C¯1 + ∗ ¯2
C2 ∗ c0 c̄0 C11 + . . . .
For all levels,
∞
X ∞ Z
X ∞ Z
∞ X
X
∗ ¯k
Ψsf = Cs ∗ k−1
c0 c̄0 Cs−1 + ∗ k
Cs ∗ c0 c̄0 C¯s−1
k−1
. (7.27)
k=0 s=k k=1 s=k
k even k odd

Since Ψsf is not a functional of πsk or π̄sk , a delta-function type gauge-fixing pro-
cedure is implemented. Elimination of the antifields via Eq.(6.1) leads to

C¯sk∗ = c0 c̄0 Cs−1

k−1 k+1
+ c̄0 c0 Cs+1 , k = 0, 2, 4, . . . ,

Csk∗ = c̄0 c0 C¯s−1

k−1
+ c0 c̄0 C¯s+1
k+1
, k = 1, 3, 5, . . . . (7.28)
When the antifields in Eq.(7.28) are substituted into Saux of Eq.(7.25) and the inte-
gration over πsk or π̄sk is performed, the following conditions are implemented

C¯sk∗ = 0 ⇒ c̄0 Cs−1

k−1 k+1
= 0 and c0 Cs+1 =0, k = 0, 2, 4, . . . ,

Csk∗ = 0 ⇒ c0 C¯s−1
k−1
= 0 and c̄0 C¯s+1
k+1
=0, k = 1, 3, 5, . . . , (7.29)
which follow from Eqs.(7.26) and (7.28). For k ≥ 1, Csk and C¯sk are set equal to zero
because any field Φ, for which c̄0 Φ = 0 and c0 Φ = 0, must be identically zero, as
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 96

Eq.(7.26) implies. On the other hand, for Cs−1 and C¯s , only half these conditions are
imposed:
c̄0 Cs−1 = 0 , s ≥ 0 ,
c0 C¯s = 0 , s≥0 . (7.30)
When these results are used in Cs∗ ≡ Cs−1∗ = c0 c̄0 C¯s+1
0
≡ c0 c̄0 C¯s+1 , for s ≥ −1, one
finds Cs−1∗ = C¯s+1 , so that the gauge-fixed string field ΨΣsf in Eq.(5.23) is

ΨΣsf = . . . + ∗ C¯s+1 + . . . + ∗ C¯1 + ∗ C¯0 + A + C0 + . . . + Cs + . . . . (7.31)

The condition c0 C¯s = 0 implies c̄0 ∗ C¯s = 0. Hence, the constraints in Eq.(7.30) can be
written succinctly as
c̄0 ΨΣsf = 0 . (7.32)
Eq.(7.32) implies that ΨΣsf has no term proportional to c0 . Since the string integral
is zero unless there is a c0 factor present, Q must supply it. Since Q = c0 L0 + . . .,

Q → c0 L0 , (7.33)

where L0 is the string “Klein-Gordon” operator

1 
L0 = ∂µ ∂ µ − M2 , (7.34)
2
and M2 is the mass-squared operator. The gauge-fixed action SΨsf is
Z Z
1 1
S= ΨΣsf ∗ c0 L0 ΨΣsf + ΨΣsf ∗ ΨΣsf ∗ ΨΣsf , (7.35)
2 3
where ΨΣsf in Eq.(7.31) satisfies equation (7.32). This is the result obtained in refs.[56,
246]. Off-shell scattering amplitudes in the Siegel-Feynman gauge have been obtained
in refs.[128, 111, 216, 54, 55, 217].

7.6 The Massless Relativistic Spinning Particle

This irreducible system has super-reparametrization invariance. It illustrates the
gauge-fixing procedure for a simple supergravity theory. The system possesses one
ordinary gauge symmetry and one gauge supersymmetry, for which auxiliary trivial
pairs {C,¯ π̄} and {Γ̄, Λ̄} are needed. The fields π̄, Γ̄, C¯∗ and Λ̄∗ are commuting, while
¯ Λ̄, π̄ ∗ and Γ̄∗ are anticommuting. The auxiliary action for the pairs is
C,
Z  
Saux = dτ π̄ C¯∗ + Λ̄Γ̄∗ . (7.36)
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 97

The fermion Z  
Ψ= dτ C¯ (e − 1) + Γ̄χ (7.37)
imposes the delta-function conditions

e=1, χ=0 . (7.38)

A straightforward calculation shows that the gauge-fixed action becomes

Z Z  
1 ¯ τ C + Γ̄∂τ Γ
SΨ → dτ (∂τ xµ ∂τ xµ − iψ µ ∂τ ψµ ) + dτ C∂ . (7.39)
2
For xµ and ψ µ , this is the flat-space action in Eq.(3.57). Using the superfield formu-
lation in Eq.(3.52) and defining the ghost superfields

G ≡ C + θΓ , Ḡ = Γ̄ + θC¯ ,

the action in Eq.(7.39) can be written in supersymmetric form as

Z Z  
1
i dτ dθ Dθ X µ ∂τ Xµ − Dθ ḠDθ G , (7.40)
2
where Dθ is given in Eq.(3.53).

7.7 The First-Quantized Bosonic String

The BRST quantization of the first-quantized bosonic string was carried out in
refs.[121, 163, 174]. The field-antifield treatment is similar. In the formulation of
Sect. 5.8, there are three types of gauge transformations for this irreducible system.
Correspondingly, one needs to introduce three level 0 auxiliary trivial pairs {C, ¯ π̄},
{C¯n , π̄n }, and {C¯ab , π̄ab }, where a, b and n take on the values σ and τ , and C¯ab and
π̄ab are antisymmetric in a and b. The auxiliary action Saux is
Z Zπ  
1
Saux = dτ dσ π̄ C¯∗ + π̄n C¯∗n + π̄ab C¯∗ab . (7.41)
2
0

This action is to be added to S in Eq.(5.30) to give a total action Stotal = S + Saux .

For the gauge-fixing fermion, we use
Z Z π n
Ψ= dτ dσ C¯τ (eτ σ − eσ τ ) + C¯σ (eτ τ − eσ σ ) +
0
 o
C¯τ σ (eτ σ + eσ τ ) + C¯ eτ τ + eσ σ − 2ρ−1/2 . (7.42)
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 98

Such a Ψ leads to the delta-function-type gauge conditions

eτ τ = eσ σ = ρ−1/2 , eτ σ = eσ τ = 0 . (7.43)

This is known as the conformal gauge. The antifields are

e∗τ τ = C¯ + C¯σ , e∗σ σ = C¯ − C¯σ ,

e∗τ σ = C¯τ σ + C¯τ , e∗σ τ = C¯τ σ − C¯τ ,

C¯∗τ = eτ σ − eσ τ , C¯∗σ = eτ τ − eσ σ ,
C¯∗τ σ = eτ σ + eσ τ , C¯∗ = eτ τ + eσ σ − 2ρ−1/2 ,
Xµ∗ = 0 , C ∗ = Cn∗ = Cab
∗
=0 . (7.44)
The gauge-fixed action becomes
Z Z 
π 1
SΨ → dτ dσ ηµν (∂τ X µ ∂τ X ν − ∂σ X µ ∂σ X ν )
0 2

+ C¯τ ρ−1/2 (∂σ C τ − ∂τ C σ ) + C¯σ ρ−1/2 (∂σ C σ − ∂τ C τ )

 
+ 2C¯ C n ∂n ρ−1/2 − C¯τ σ ρ−1/2 (∂τ C σ + ∂σ C τ + 2C τ σ )
o
¯ −1/2 (2C − ∂τ C τ − ∂σ C σ )
+ Cρ , (7.45)
where the sum over n is over τ and σ: C n ∂n ρ−1/2 = C τ ∂τ ρ−1/2 +C σ ∂σ ρ−1/2 . Integrating
over C and C τ σ leads to delta functions for C¯ and C¯τ σ , so that the gauge-fixed action
becomes the first two lines in Eq.(7.45):
Z Z 
π 1
SΨ → dτ dσ ηµν (∂τ X µ ∂τ X ν − ∂σ X µ ∂σ X ν )
0 2
o
+ C¯τ ρ−1/2 (∂σ C τ − ∂τ C σ ) + C¯σ ρ−1/2 (∂σ C σ − ∂τ C τ ) . (7.46)
At this stage, one can either define a new C¯n field which absorbs the factor of ρ−1/2
or one can set ρ (τ, σ) = 1, as long as D = 26.6 In either case, since the resulting
action is a free theory, it is straightforward to quantize the system. In what follows,
we consider the open string case. The equations of motion are

∂τ ∂τ X µ − ∂σ ∂σ X µ = 0 ,

∂τ C τ − ∂σ C σ = 0 , ∂τ C σ − ∂σ C τ = 0 ,
∂τ C¯τ − ∂σ C¯σ = 0 , ∂τ C¯σ − ∂σ C¯τ = 0 , (7.47)
6
When D 6= 26, one cannot do this due to Weyl anomaly. See Sect. 9.3.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 99

and
∂σ X µ = C σ = C¯σ = 0 , at σ = 0, π , (7.48)
which follow from a careful treatment of boundary conditions. The solutions are
∞
X αnµ −inτ
X µ (τ, σ) = xµ + pµ τ + i e cos (nσ) ,
n=−∞ n
n6=0

∞
X
C τ (τ, σ) = cn e−inτ cos (nσ) ,
n=−∞
∞
X
C σ (τ, σ) = −i cn e−inτ sin (nσ) ,
n=−∞
∞
X
C¯σ (τ, σ) = −i c̄n e−inτ cos (nσ) ,
n=−∞
∞
X
C¯τ (τ, σ) = c̄n e−inτ sin (nσ) , (7.49)
n=−∞

where the modes αnµ , cn and c̄n (n = −∞ to + ∞, n 6= 0) are harmonic-oscillator-like

raising and lowering operators satisfying
h i
[pµ , xν ] = −iη µν , αnµ , α−m
ν
= nη µν δnm ,

{cn , c̄−m } = δnm , {cn , cm } = 0 , {c̄n , c̄m } = 0 , (7.50)

where η µν is the D = 25 + 1 flat space metric with signature (−, +, +, . . .). The
operators xµ and pµ correspond to the center-of-mass position and momentum of the
string. The zero-mode c0 and c̄0 play an important role in the Siegel-Feynman gauge
of the open bosonic string as discussed in Sect. 7.5. States in the theory are obtained
by applying creation operators, corresponding to modes n with n < 0, to the vacuum
states, which are eigenfunctions of pµ and the zero-mode ghost system.
In second quantization, fields become functionals of the first-quantized variables
X (0, σ), C τ (0, σ) and C¯σ (0, σ). The string fields for the open bosonic string field
µ

theory of Sects. 3.6, 5.6 and 7.5 are such functionals. Alternatively, one may expand
the string fields as a linear combination of first-quantized states whose coefficients are
ordinary particle fields. For more details, see the reviews in refs.[215, 247].
The gauge-fixed BRST symmetry of the action in Eq.(7.46) is

δBΨ X µ = C n ∂n X µ ,

δBΨ C m = −C n ∂n C m ,
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 100

 
δBΨ C¯τ = −ρ1/2 ∂τ Xµ ∂σ X µ + ρ1/2 ∂n ρ−1/2 C¯τ C n + C¯σ (∂σ C τ + ∂τ C σ ) , (7.51)
1  
δBΨ C¯σ = −ρ1/2 (∂τ Xµ ∂τ X µ + ∂σ X µ ∂σ Xµ ) + ρ1/2 ∂n ρ−1/2 C¯σ C n + C¯τ (∂σ C τ + ∂τ C σ ) ,
2
where m takes on the values τ and σ and the sum over n is over τ and σ. A
straightforward computation confirms that
∂r SΨ
δBΨ SΨ ≡ δB ΦA = 0 , (7.52)
∂ΦA Ψ
where SΨ is given in Eq.(7.46). A useful exercise is to perform this computation.
Because some fields have been eliminated using equations of motion, δBΨ δBΨ = 0 only
on-shell.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 101

8 Quantum Effects and Anomalies

In this section we discuss some of the quantum aspects of the field-antifield for-
malism. Basically, the classical quantities and concepts have quantum counterparts:
The proper solution S is replaced by a quantum action W . The quantum master
equation is used in lieu of the classical master equation. The BRST transformation
δB is generalized to a quantum version δB̂ . It is nilpotent only if the quantum master
equation is satisfied. Quantum observables are elements of the cohomology of δB̂ . A
violation of the quantum master equation corresponds to a gauge anomaly [251] and
is the subject of Sects. 8.5–8.7. In Sects. 8.1–8.4, however, we assume that W satisfies
the quantum master equation. Many equations in those subsections are of a formal
nature due to the singular operator ∆ in the context of local theories. To obtain a
well-defined action of ∆, a regularization scheme must be used. See Sect. 8.7. In
Sect. 8.4, the effective action Γ is discussed. Antifields become classical and acquire
a conceptual interpretation: They are sources for BRST transformations generated
by Γ.

8.1 Quantum-BRST Transformation and Its Cohomology

Recall that the condition in Eq.(6.13) that a functional X produces a gauge-
invariant correlation is
(X, W ) = ih̄∆X .
Introducing the operator δB̂ [186],

δB̂ X ≡ (X, W ) − ih̄∆X , (8.1)

this can be written as δB̂ X = 0. The operator δB̂ is the quantum BRST transforma-
tion. It is the quantum generalization of the classical BRST transformation δB : As h̄
goes to zero, δB̂ becomes δB in Eq.(4.32). A functional is said to be quantum-BRST
invariant if
δB̂ X = 0 . (8.2)
Although δB̂ is nilpotent as a consequence of the quantum master equation

1
δB̂ δB̂ X = −( (W, W ) − ih̄∆W, X) = 0 , (8.3)
2
it no longer is a graded derivation since

δB̂ (XY ) = X (δB̂ Y ) + (−1)ǫY (δB̂ X) Y − ih̄ (−1)ǫY (X, Y ) . (8.4)

J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 102

Suppose X is the BRST transform of another functional Y , i.e.,

X = δB̂ Y = (Y, W ) − ih̄∆Y .

Then, one refers to X as quantum-BRST trivial. By the nilpotency of δB̂ , X is

quantum-BRST invariant, but not in an interesting way. In fact, a quantum-BRST
trivial operator X produces a zero correlation function, as can be seen as follows.
The integrand I in Eq.(6.6) becomes
     
i i
exp W X = −ih̄∆ exp W Y .
h̄ h̄
 
i
Let Ĩ = −ih̄ exp h̄
W Y . Then
Z Z !
 ∂r ∂r 
[dΦ] ∆Ĩ  = [dΦ] (−1) ǫA +1
Ĩ 
 =
ΣΨ ∂ΦA ∂Φ∗A 
ΣΨ
 ! 
Z 
∂r  ∂r 
[dΦ] (−1)ǫA +1 Ĩ  +
∂ΦA ∂Φ∗A 
ΣΨ
Z !
∂r ∂r Ĩ ∂r ∂r Ψ 
[dΦ] (−1) ǫA  .
∂Φ∗B ∂Φ∗A ∂ΦA ∂ΦB  ΣΨ

The first term produces zero since it is a surface term and the second term is zero
for symmetry reasons. Hence, an interesting functional O is one that is non-trivial in
the quantum-BRST cohomology, i.e., O is quantum-BRST invariant but cannot be
written as the quantum-BRST transform of some functional:

δB̂ O = 0 but O 6= δB̂ Y , (8.5)

for any Y . An O satisfying Eq.(8.5) is called a quantum observable. Observables are

considered equivalent if they differ by a quantum-BRST trivial functional.
Unlike the situation at the classical level, quantum observables do not form an
algebra. This is due to the fact that δB̂ is no longer a graded derivation, c.f. Eq.(8.4).
This is not surprising since, in the quantum theory, the product of two observables
is singular and must be regularized. The regularization process may ruin quantum-
BRST invariance. However, if the space-time supports of O1 and O2 do not intersect,
then (O1 , O2 ) = 0. In such a case, O1 O2 is quantum-BRST invariant, if O1 and O2
are.
The result obtained above, namely,
Z  
i
[dΦ] exp W δB̂ Y = 0 , (8.6)
h̄
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 103

for various functionals Y produces identities among correlation functions. They

embody the Ward identities associated with the BRST invariance of the theory
[265, 239, 233, 241].
An example of a correlation function that produces a zero expectation value is
   
Z   ∂ X|
i l ΣΨ i ∂l (WΨ )
[dΦ] exp WΨ  A
+ X|ΣΨ  = 0 , (8.7)
h̄ ∂Φ h̄ ∂ΦA

obtained by integration by parts using Eq.(A.5). Up to a normalization factor,

Eq.(8.7) is the Schwinger-Dyson equation [94, 221]
*   +
∂l X|ΣΨ i ∂l (WΨ )
+ X|ΣΨ =0 . (8.8)
∂ΦA h̄ ∂ΦA

In the above, WΨ denotes the gauge-fixed quantum action

WΨ ≡ W |ΣΨ , (8.9)

and X is any quantum-BRST invariant functional, δB̂ X = 0. The derivatives in

Eqs.(8.7) and (8.8) involve implicit differentiation of the gauge-fixing fermion, e.g.,
   
∂l X|ΣΨ ∂l X  ∂l ∂Ψ ∂l X 
=  +  ,
∂ΦA ∂ΦA ΣΨ ∂ΦA ∂ΦB ∂Φ∗B ΣΨ

and a similar identity holds for X → W . The fact that the functional in Eq.(8.7)
produces a zero expectation suggests that it is quantum-BRST trivial. Indeed, this
is the case since
    ! 

∂l X|ΣΨ i ∂l (WΨ ) (−1)ǫ(X) ∂Ψ 
+ X|ΣΨ = δ 
B̂ Φ∗A − X  , (8.10)
∂ΦA h̄ ∂Φ A ih̄ ∂ΦA

ΣΨ

as a short calculation reveals. One obtains the interesting result that the Schwinger-
Dyson equations are a consequence of the quantum-BRST symmetry of the field-
antifield formalism [152]. Reference [2] argued that antifields originate as the antighosts
of collective fields that ensure the Schwinger-Dyson equations. This differs from one
viewpoint that antifields are the sources of BRST transformations. Equations (8.6)
and (8.10) show that the Schwinger-Dyson equations are certain quantum-BRST
Ward identities.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 104

8.2 Satisfying the Quantum Master Equation

The quantum master equation
1
(W, W ) − ih̄∆W = 0 (8.11)
2
is the most important requirement of the field-antifield formalism for several reasons:
As demonstrated in Sect. 6.1, it guarantees that computations are independent of the
gauge-fixing procedure. When
 the quantum action W is expanded in powers of h̄ via
0
W = S + O (h̄), the O h̄ term in Eq.(8.11) is the classical master equation

1
(S, S) = 0 , (8.12)
2
which yields the structure equations. The nilpotency of the quantum BRST operator
depends on Eq.(8.11) being satisfied, as can be seen from Eq.(8.3). The quantum
cohomology requires the existence of a nilpotent quantum BRST operator. It is used
to define the quantum observables of a gauge theory and to determine when two
functionals are considered equivalent. In Sect. 8.5, it is argued that the existence of
gauge anomalies is related to the violation of the quantum master equation. In short,
the field-antifield formalism at the quantum level depends crucially on Eq.(8.11) being
satisfied.
The potential difficulty is due to the operator ∆ in Eq.(6.7) which is singular
when acting on S or W because usually they are local functionals. Often terms
proportional to delta functions and derivatives of delta functions are produced. One
therefore needs to regularize ∆. If one can find a regularization such that (∆S)reg = 0,
while maintaining the classical master equation, then one can simply let W = S.7
Then the quantum master equation is satisfied. An example of this situation is a
theory without gauge invariances: Antifields are absent in S since the proper solution
is given by S = S0 . Consequently, ∆S = 0.
The quantum master equation has a symmetry given by quantum BRST trans-
formations, in which W → W − εδB̂ F , where ǫ(F ) = 1 and gh[F ] = −1 [27, 103, 152,
75, 253]. If Eq.(8.11) is true, and
′
 ′

W = W + ε W , F + ih̄ε∆F , (8.13)
′
then W also satisfies
 Eq.(8.11) to order ε2 . Eq.(8.13) has a simple interpretation. The
′
term ε W , F is the change in W due to a canonical transformation (see Eq.(4.12)).
The extra term ih̄ε∆F is due to the non-invariance of the functional integral measure
7
Strictly speaking, this statement holds for a finite or a regularized theory. Renormalization
may require one to add additional terms to S.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 105

under a canonical transformation. The measure effect can be exponentiated and leads
to the extra term.
A solution to the quantum master equation ensures that the gauge symmetries
survive the process of quantization. Although finding a solution is an important
objective, a solution does not necessarily guarantee good behavior of the theory. The
theory may still have difficulties in regard to other issues such as locality, unitarity and
the presence of Feynman-diagram infinities. These difficulties may be insurmountable
or it may be necessary to add new counterterms to achieve the desired properties.
There are systems for which the quantum master equation is satisfied but the theory
is non-renormalizable (see the next subsection). In short, the quantum theory is not
determined solely by the quantum master equation.

8.3 Remarks on Renormalization

At the one-loop level and beyond, perturbation theory produces infinities if
Feynman-diagram contributions are not regularized. If the regularization involves
a cutoff, then standard renormalizability of a field theory means that the cutoff de-
pendence can be absorbed into the coefficients of the terms in the original action
[165, 278].8 In this manner, infinities are eliminated. If a regularization procedure
can be found that respects the structure of the field-antifield formalism then renor-
malization is expected to proceed as in the usual field theory case. Infinities should
be “absorbable” into the coefficients of the terms in S. Regularization methods that
seem most convenient for gauge theories, such as the Pauli-Villars scheme, are the
best candidates for respecting the field-antifield formalism.
To be concrete, suppose a gauge theory is renormalizable, and suppose the regu-
larization of ∆ happens to give (∆S)reg = 0.9 Then the Mn in Eq.(6.14) can be chosen
to be the counterterms removing the cutoff dependence of the theory. The Mn involve
terms similar to the original action S. The quantum master equation in Eq.(6.15)
is then satisfied because (∆Mn )reg = 0 and because (Mn , S) = (Mn , Mm ) = 0.
These equations hold because of the classical master equation (S, S) = 0, because
(∆S)reg = 0, because the Mn resemble terms in S, and because of analyticity of S
in coupling constants. If (∆S)reg 6= 0 and an infinite number of local counterterms
have to be added to the action to satisfy the quantum master equation, then the issue
of renormalizability is unclear [4]. This situation arises for the closed bosonic string
field theory (see Sect. 10.8), as currently formulated.
8
In dealing with renormalizable theories, we also assume that a reasonable gauge-fixing procedure
is used, i.e., we exclude procedures leading to interactions that are non-renormalizable by power
counting.
9
A renormalizable theory without gauge invariances is a useful case to keep in mind.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 106

The question of whether the quantum master equation is satisfied and the ques-
tion of renormalizability are separate issues (although somewhat related). As argued
above, a renormalizable theory without gauge invariances is guaranteed to satisfy the
quantum master equation. Regularization and renormalization must be performed,
even if the quantum master equation is satisfied at all stages of the renormalization
process. When gauge invariances are present, the interesting issue is whether the
quantum master equation can be maintained after renormalization. A theory can be
non-renormalizable but satisfy the quantum master equation: A non-renormalizable
theory without gauge invariances is an example. An infinite number of countert-
erms must be added to the action, but the quantum master equation is satisfied
trivially because no antifields enter. In the other extreme, the two-dimensional chiral
Schwinger model is anomalous, yet the theory still makes sense and is renormaliz-
able. It is an example of a renormalizable system that does not satisfy the quantum
master equation. For more discussion on regularization and renormalization in the
field-antifield formalism, which are the natural generalizations of ideas contained in
earlier approaches [277, 187, 167], see refs.[262, 264, 249, 4, 75, 156, 191].
The Zinn-Justin equation [277] has played an important role in analyzing the
renormalizability of Yang-Mills theories. The generalization of this equation within
the antibracket formalism is the subject of the next section.

8.4 The Effective Action and the Zinn-Justin Equation

A useful concept in functional approaches to field theories [1, 165, 278] is the effec-
tive action Γ. When used in the classical approximation, it reproduces computations
at the quantum level. This is accomplished by incorporating all loop contributions
into effective interactions. These interactions sum the one-particle-irreducible dia-
grams. Hence, to compute a correlation function in perturbation theory with Γ, one
uses tree diagrams only.
The functional Γ is obtained by a Legendre transformation. One performs the
functional integral over the quantum fields in the presence of sources. Classical fields
are obtained by evaluating the expectation values of the quantum fields in the presence
of these source terms. The effective action Γ is a functional of the classical fields.
The concept of an effective action also exists in the antibracket formalism [24].
We also denote it by Γ. It makes the quantum system resemble a classical system
by summing loop effects. Since one wants the analog of the “effective antibracket
formalism”, it is necessary to have classical antifields as well as classical fields. After
this is accomplished, one can define an antibracket ( , )c in the space of classical fields
and antifields. Since computations are performed as though the system is classical,
the classical master equation for Γ incorporates the quantum master equation of the
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 107

original system.
Consider the functional integral in the presence of sources JA , i.e.,
Z  
i 
Z [J, Φ∗c ] ≡ [dΦ] exp W [Φ, Φ∗c ] + JA ΦA , (8.14)
h̄
∗
where JA are independent
  variables that do not depend on Φ and Φc , and have
statistics ǫ (JA ) = ǫ ΦA = ǫA . For reasons that become clear below, we have
replaced the Φ∗ by classical antifields Φ∗c . Actually, to compute perturbatively, one
∂Ψ
needs to shift the antifields Φ∗cA by ∂Φ A , where Ψ is an appropriate gauge-fixing

fermion. More precisely, one performs a canonical transformationhto the gauge-fixed

i
basis before setting Φ∗A equal to Φ∗cA . The net result is that Z J, Φ∗c + ∂Ψ∂Φ
is the
functional integral of interest. Although a gauge-fixing fermion is present, we omit
the dependence on Ψ for notational convenience, in what follows. Since antifields
appear in S in the form S = SΨ + Φ∗A δBΨ ΦA + . . ., they act as sources for the gauge-
fixed BRST transformations.
Define a classical field ΦA
c by

∗ h̄ ∂l ln Z [J, Φ∗c ] D A EJ
ΦA
c [J, Φc ] ≡ = Φ , (8.15)
i ∂JA

where h iJ denotes an expectation value in the presence of sources:

Z  
1 i 
hXiJ ≡ [dΦ] X [Φ, Φ∗c ] exp W + JA ΦA . (8.16)
Z [J, Φ∗c ] h̄
∗
Note that ΦA c is a functional of JB and ΦcB . In principle, it is possible to invert this
∗
relation to determine JA in terms of ΦB c and ΦcB . We indicate the solution to this
inversion by JcA : JcA = JA [Φc , Φ∗c ]. The effective action is obtained by a Legendre
transformation [165]:

Γ [Φc , Φ∗c ] ≡ −ih̄ ln Zc − JcA ΦA

c , (8.17)

where Zc is Z [J, Φ∗c ] evaluated at JA = JcA :

Zc ≡ Z [Jc , Φ∗c ] . (8.18)

As a result, Zc is a functional of classical fields and antifields. A straightforward

∂r Γ
calculation of ∂Φ A gives
c


∂r Γ ǫA ǫB ∂r JcB B ∂r ln Z [J, Φ∗c ] ∂r JcB 
= −J cA − (−1) Φc − ih̄  .
∂ΦA ∂ΦA ∂JB ∂ΦA 
c c c J=Jc
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 108

Using Eq.(8.15), one finds that the last two terms cancel so that

∂r Γ
= −JcA . (8.19)
∂ΦAc

At this stage, we have made a transition from fields to classical fields. Essentially
∗
all quantities are now functionals of ΦAc and ΦcA . Given two functionals X and Y of
∗
ΦAc and ΦcA , define the classical antibracket ( , )c by

∂r X ∂l Y ∂r X ∂l Y
(X, Y )c ≡ − . (8.20)
∂Φc ∂ΦcA ∂Φ∗cA ∂ΦA
A ∗
c

Since the classical antibracket is defined in the same manner as the antibracket, it
satisfies the same identities (4.5) – (4.7). Using Eq.(8.19), one obtains
 
1 ∂l Γ 1
(Γ, Γ)c = −JcA ∗ = −ih̄∆W + (W, W ) , (8.21)
2 ∂ΦcA 2 c

where the final equality is obtained after some algebra, which makes use of integration
by parts. In Eq.(8.21), h ic denotes the expectation value in the presence of J but
expressed in terms of Φc and Φ∗c . More precisely, if X is a functional of Φ and Φ∗
then 

hXic ≡ hXiJ  , (8.22)
J=Jc
J
where h i is defined in Eq.(8.16). Because of the quantum master equation (8.11),
Eq.(8.21) becomes
(Γ, Γ)c = 0 , (8.23)
a result known as the Zinn-Justin equation [277].
Equation (8.22) allows one to pass from a functional X of the original fields Φ and
Φ to a classical functional hXic of classical fields Φc and Φ∗c by taking the “classical
∗

expectation” of X: X [Φ, Φ∗ ] → hXic . We refer to hXic as the classical version of the

quantum functional X. The definition is consistent with the notation for ΦA c since
D E
ΦA = ΦA
c . (8.24)
c

The process X → hXic conforms to the idea that a classical variable is the expectation
value of the corresponding quantum functional.
Since Γ satisfies the Zinn-Justin equation and plays the role of S in the classical
antibracket formalism, one can construct by analogy a “classical-quantum” BRST
transformation δBcq having the same properties as δB . Define

δBcq X ≡ (X, Γ)c , (8.25)

J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 109

where X is any functional of Φc and Φ∗c . The nilpotency δB2 cq = 0 of δBcq follows from
the Zinn-Justin equation. Define X to be cq-BRST invariant if δBcq X = 0. According
to Eq.(8.23), the effective action Γ is cq-BRST invariant since

δBcq Γ = (Γ, Γ)c = 0 . (8.26)

Several of the above-mentioned classical functionals are also cq-BRST invariant. A

analysis of δBcq JcA reveals that

1 ∂r (Γ, Γ)c
δBcq JcA ≡ (JcA , Γ)c = (−1)ǫA +1 =0 . (8.27)
2 ∂ΦAc

Consider
    ∂l Γ
−ih̄δBcq ln (Zc ) = Γ + JcA ΦA
c ,Γ = JcA ΦA
c ,Γ = JcA =0 ,
c c ∂Φ∗cA

where the first equality holds because of Eqs.(8.26) and (8.27), and the last equality
follows from Eqs.(8.11) and (8.21). Hence,

δBcq Zc = (Zc , Γ)c = 0 . (8.28)

The cq-BRST operator δBcq is the effective classical version of the quantum-BRST
operator δB̂ X. To understand this statement, consider

∂r (Zc hXic ) ∂l Γ ∂r (Zc hXic ) ∂l Γ

δBcq (Zc hXic ) = (Zc hXic , Γ)c = ∗
−
∂Φc A ∂ΦcA ∂Φ∗cA ∂ΦA c
  * ! +
∂r Z hXiJ ǫA ∂r X i ∂r W
= (JcB , Γ)c + (−1) Zc ∗
+ X ∗ JA .
∂JcB ∂ΦA h̄ ∂ΦA c

The last step follows because h ic has dependence on Φc and Φ∗c through Jc . After
some algebra which makes use of integration by parts, one finds that
  
i
δBcq (Zc hXic ) = Zc δB̂ X + X ∆W + (W, W ) + (−1)ǫB Zc hXΦB
c i (JcB , Γ)c .
2h̄ c
(8.29)
If δBcq (hXic ) is computed instead, one obtains the connected part of Eq.(8.29) without
a Zc factor. Since Eq.(8.27) and the quantum master equation (8.11) are satisfied,

δBcq (hXic ) = hδB̂ Xic . (8.30)

In other words, the cq-BRST variation of the classical version of X is the classical
version of the quantum-BRST variation of X [4].
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 110

The effective action Γ in the classical antibracket formalism plays a role analogous
to the proper solution S in the ordinary antibracket formalism. Properties obeyed by
Γ are the same as those obeyed by S. Therefore, one can define a BRST structure
associated with Γ [133]. The BRST structure tensors are encoded in Γ and the
relations among them are given by (Γ, Γ) = 0. Expanding in a Taylor series in Φ∗c ,
one has
1
Γ [Φc , Φ∗c ] = Γ0 (Φc ) + Φ∗cA ΓA (Φc ) + Φ∗cA Φ∗cB ΓBA (Φc ) + . . . . (8.31)
2
Recalling that there is an undisplayed dependence on the gauge-fixing fermion Ψ, the
above terms have the following interpretation: Γ0 (Φc ) is the one-particle-irreducible
generating functional for the basic fields including all loop corrections for the action
gauge-fixed using Ψ, i.e., SΨ in Eq. (6.88), and ΓA (Φc ) is the generator of gauge-fixed
cq-BRST transformations. The gauge-fixed cq-BRST operator δBcqΨ , the analogy of
δBΨ , is defined by
δBcqΨ X ≡ (X, Γ)|ΣΨ , (8.32)
where X is a functional of the ΦA c only. In the gauge-fixed basis, ΣΨ in Eq.(8.32)
means that classical antifields are set to zero. Thus, one has δBcqΨ ΦA A
c = Γ (Φc ).
The tensor ΓBA (Φc ) is related to the on-shell nilpotency of δBcqΨ . In summary, the
quantum aspects of the classical theory described by S are reproduced by an effective
classical theory governed by Γ.

8.5 Quantum Master Equation Violations: Generalities

Suppose Eq.(8.11) is not zero. Let
i
A ≡ ∆W + (W, W ) (8.33)
2h̄
be the violation of the quantum master equation. A straightforward computation
using Eqs.(4.5), (6.8) and (8.1), but not assuming the validity of the quantum master
equation, reveals that
δB̂ A = 0 . (8.34)
This equation is consistent with Eqs.(8.21), (8.22) and (8.30) and the Jacobi iden-
tity for the antibracket: One has 21 (Γ, Γ)c = −ih̄hAic so that 0 = 2h̄i ((Γ, Γ)c , Γ)c
= (hAic, Γ)c = δBcq hAic = hδB̂ Aic .
Suppose that
h̄ h̄
iA = δB̂ Ω + (Ω, Ω) = (Ω, W ) − ih̄∆Ω + (Ω, Ω) , (8.35)
2 2
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 111

where Ω is a local functional of the fields and antifields. The last term may seem
surprising but it is necessary if A is to satisfy Eq.(8.34): When the quantum master
equation is violated as in Eq.(8.33), the nilpotency of the quantum BRST operator
no longer holds, as can be seen from Eq.(8.3). The last term in Eq.(8.35) is required
to compensate for this effect and ensures Eq.(8.34). Let
′
W = W + h̄Ω . (8.36)
′
Then, using Eqs.(4.5) and (6.8), one finds that W satisfies the quantum master
equation:
1 ′ ′
 ′
W , W − ih̄∆W = 0 .
2
Since W has the same classical limit as W , namely, its order h̄0 term is S, one can use
′

′
W in lieu of W for the quantum action. Then, since the quantum master equation
is satisfied, a quantum gauge theory can be defined.
When A cannot be expressed as in Eq.(8.35) for a local functional Ω, there is an
anomaly in the quantum master equation and an obstruction to maintaining gauge
symmetries at the quantum level. Since anomalies involve subtleties and singular
expressions, they are usually not too easy to compute. The usual approach to this
subject uses a loop expansion:
∞
X
A= Al h̄l−1 = A1 + h̄A2 + . . . . (8.37)
l=1

Using Eqs.(6.14) and (8.37), one finds the following expression for the anomaly at the
one-loop level
A1 ≡ ∆S + i (M1 , S) . (8.38)
To this order, the condition for the absence of an anomaly in Eq.(8.35) is that A1 is
a classical BRST variation, i. e.,

A1 = −i (Ω1 , S) = −iδB Ω1 . (8.39)

If A1 can be expressed as in Eq.(8.39) for some local functional Ω1 , then, by setting

′
W = S + h̄Ω1 , (8.40)

the quantum master equation in Eq.(8.11) is satisfied to order h̄. In words, Eq.(8.39)
says that if A1 is expressible as a local BRST variation then effectively there is
no one-loop anomaly. Since the second term in Eq.(8.38) is already of this form,
the requirement becomes that ∆S should be a classical-BRST variation of a local
functional.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 112

To order h̄, the equation δB̂ A = 0 in Eq.(8.34) is

δB A1 = (A1, S) = 0 . (8.41)

In view of Eqs.(8.39) and (8.41), the investigation of anomalies is related to the

local BRST cohomology at ghost number one [36, 32, 159, 251, 248]. Although not
obvious, it turns out that Eq.(8.41) embodies the Wess-Zumino anomaly consistency
equations [268]. In field-antifield formalism, the one-loop master equation anomaly
must be classically BRST invariant. If anomalies arise beyond the one-loop level,
Eq.(8.34) provides the full quantum consistency conditions: The master equation
anomaly must be quantum-BRST invariant.
Note that gh [A] = 1. Of the fields in Eq.(4.1), only C0α0 ≡ C α has ghost number
one. Hence, one may write

iA = aα (φ) C α + . . . , (8.42)

where the omitted terms involve antifields. Sometimes these terms are absent so that
Eq.(8.42) gives the structure of the quantum-master-equation anomaly. Although not
obvious, the coefficients aα (φ) are the usual gauge anomalies [251]. In other words,
a quantum-master-equation anomaly and a gauge anomaly are equivalent.

8.6 Canonical Transformations and the Quantum

Master Equation
Canonical transformations preserve the quantum master equation as long as W is
appropriately transformed [157]. Consider an infinitesimal canonical transformation
as in Eq.(4.10) governed by F . Normally a functional G transforms as G → G +
ε(G, F ). For W , however, one must add an extra term ih̄ε∆F to compensate for
“measure effects”. Hence, the transformation rule for W is taken to be
   
W → W + ε(W, F ) + ih̄ε∆F + O ε2 = W − εδB̂ F + O ε2 . (8.43)

According to Eq.(8.13), Eq.(8.43) is a symmetry of the quantum master equation

[27, 152, 103, 75, 253]. If W is a solution to the quantum master equation, changing
W as in Eq.(8.43) will not upset the solution.
The same conclusion holds for finite transformations governed by F2 in Eq.(4.13).
In this case, one transforms from {Φ, Φ∗ } variables to {Φ̃, Φ̃∗ } variables via Eq.(4.13).
The transformation rule for W is [27]

f [Φ̃, Φ̃∗ ] = W [Φ, Φ∗ ] − ih̄

W ln J , (8.44)
2
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 113

where the jacobian factor J is the berezinian governing the change from {Φ, Φ∗ }
variables to tilde variables. Then, Wf satisfies the quantum master equation exactly
in the tilde variables if W satisfies it in the {Φ, Φ∗ } variables. A detailed proof of
this result and a formula for J can be found in ref.[251]. Here, we provide a few key
steps. Define ∆e as the analog of ∆ in the transformed variables, i.e.,

e ≡ (−1)ǫA +1 ∂r ∂r
∆ .
∂ Φ̃A ∂ Φ̃∗A
If G is an arbitrary functions of Φ̃ and Φ̃∗ , then
h i h 1 i
e
∆G Φ̃, Φ̃∗ = ∆G Φ̃ [Φ, Φ∗ ] , Φ̃∗ [Φ, Φ∗ ] −
(G, ln J) . (8.45)
2
When ∆ acts on G on the right-hand side of Eq.(8.45), the tilde fields should be
regarded as functionals of Φ and Φ∗ , as indicated. The chain rule for derivatives is
then used. This produces ∆Ge plus an extra term, which is equal to 21 (G, ln J) and
needs to be subtracted to obtain the identity in Eq.(8.45). Using Eqs.(8.44) and
(8.45), one finds,
 
eWf− 1 f f 
f −1 W f , ln J
 1 f f
ih̄∆ W , W = ih̄ ∆W − W, W
2 2 2
( !) !
ih̄ 1 ih̄ 1 ih̄ ih̄
= ih̄ ∆W − ∆ ln J − W − ln J, ln J − W − ln J, W − ln J
2 2 2 2 2 2
 
1 h̄2 1
= ih̄∆W − (W, W ) + ∆ ln J − (ln J, ln J) .
2 2 4
It can be shown [27, 251] that ∆ ln J = 14 (ln J, ln J), so that the last term is zero.
Hence, if W satisfies the quantum master equation, then W f satisfies the tilde version
of the quantum master equation. For infinitesimal transformations,
   
Φ̃A = ΦA − ε ΦA , F + O ε2 ,
 
Φ̃∗A = Φ∗A − ε (Φ∗A , F ) + O ε2 ,
 
J = 1 − 2ε∆F + O ε2 . (8.46)
Then,  
f [Φ, Φ∗ ] = W [Φ, Φ∗ ] + ε (W, F ) + ih̄∆F + O ε2
W ,
so that  
f [Φ, Φ∗ ] = W − εδ F + O ε2
W [Φ, Φ∗ ] → W ,
B̂

and one recovers Eq.(8.43). The identity ∆ ln J − 14 (ln J, ln J) ∼ O (ε2 ) follows from
Eq.(8.46) and ∆2 = 0.
One can take advantage of canonical transformations in analyzing potential anoma-
lies by going to a basis for which the computation is simpler.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 114

8.7 The Anomaly at the One-Loop Level

The quantities that appear in the violation of the quantum master equation in
Sect. 8.2 involve both fields and antifields. As a consequence, one must use the action
before any elimination of antifields. On the other hand, since propagators are needed
to perform perturbative computations, a gauge-fixing procedure is required. Both
these requirements can be satisfied by working in the gauge-fixed basis described in
Sect. 6.6. It is achieved by performing a canonical transformation with the gauge-
∂Ψ
fixing fermion Ψ so that Φ∗A → Φ∗A + ∂Φ A . Throughout the rest of this section, we

assume that an admissible Ψ has been selected and that the shift to the gauge-fixed
basis has been performed. According to the result in Sect. 8.6, if the quantum master
equation is satisfied and a canonical transformation is performed to a new basis then,
by appropriately adjusting the action, the quantum master equation is satisfied in
the new basis. Hence, the existence or non-existence of an anomaly is independent of
the choice of basis, although the form of the anomaly may depend on this choice.
There are different ways of obtaining the anomaly. We mostly follow the approach
of ref.[251] and briefly mention other methods at the end of this subsection. Reference
[251] obtained general formulas for the antifield-independent part of the one-loop
anomaly using a Pauli-Villars regularization scheme. Since the derivation is somewhat
technical, we present only the final results. For more details, see refs.[87, 251, 259,
78, 75, 252, 257, 253]. In particular, refs.[75, 253] have an extensive discussion of
Pauli-Villars regularization in the antibracket formalism to which we refer the reader.
The goal of the next few paragraphs is to obtain a regularized expression for ∆S,
denoted by (∆S)reg . The anomaly A1 is essentially (∆S)reg since the M1 term in
Eq.(8.38) is eliminated as a possible violation of the quantum master equation via
Eq.(8.40) with Ω1 = −M1 , i.e., the counterterm −h̄M1 is added to the action. Define
∂l ∂r
K AB ≡ ∗ B
S [Φ, Φ∗ ] , (8.47)
∂ΦA ∂Φ
∂l ∂r
QAB ≡ S [Φ, Φ∗ ] . (8.48)
∂ΦA ∂ΦB
Note that Q involves derivatives with respect to fields and not antifields. If expanded
about a stationary point, Q becomes the quadratic form for the fields. In such an
expansion, the inverse of Q is the propagator. The properness condition in the gauge-
fixed basis guarantees that propagators exist. An operator O used to regulate ∆S is
related to Q by
 AC
OA B ≡ T −1 QCB , (8.49)
where TAC is an arbitrary invertible matrix satisfying TBA = (−1)ǫA +ǫB +ǫA ǫB TAB .
The inverse of T obeys (T −1)BA = (−1)ǫA ǫB (T −1 )AB . The Grassmann statistics of
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 115

TAB and (T −1 )AB are ǫA + ǫB (mod 2). Eq.(8.49) implies QAB = TAC OC B . In the
regularization scheme of ref.[251], the matrix T appears in the mass term for the
regulating Pauli-Villars fields. In that approach, the violation of the quantum master
equation is shifted from the ∆ term to the (S, S) term.
The regulated expression for (∆S)reg is [251]
 !B 
1
(∆S)reg = F A B O
 , (8.50)
1− M A 0

where
1
F A C ≡ K A C + (T −1 )AD (δB T )DC (−1)ǫC , (8.51)
2
and where δB T denotes the classical BRST transform of T : δB T = (T, S). In
Eq.(8.50), the sum over A and B, leading to the trace, involves the quadratic form
of fields only and not that of antifields. The subscript 0 on the square brackets in
Eq.(8.50) indicates that the term independent of M is to be extracted. Here, M de-
notes a regulator mass. The one-loop nature of Eq.(8.50) is evident by the presence
of the propagator factor −i/ (M − O) → (iM)−1 / (1 − O/M ) and the sum over the
index A indicating a trace.
When O is quadratic in space-time derivatives, one lets

R=O , M2 = M , (8.52)

where R denotes the quadratic regulator operator and M denotes the regulating
mass or cutoff. When O is linear in space-time derivatives, it is convenient to mul-
tiply on the right in the trace in Eq.(8.50) by 1/ (1 + O/M) (1 + O/M) and carry
out the multiplication of 1/ (1 − O/M) with 1/ (1 + O/M). Eq.(8.50) can then be
manipulated into the form [251]
 !B 
A 1
(∆S)reg = F ′ B R
 , (8.53)
1− M 2 A 0

where
R = O2 , M=M , (8.54)
and
1
F ′A C ≡ F A C − (δB O)A C (−1)ǫC (8.55)
2M
with δB O = (O, S).
Summarizing, in the quadratic momenta case, (∆S)reg is given by Eqs.(8.50) and
(8.51). This is the same as using Eq.(8.53) with R and M given in Eq.(8.52) and
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 116

with F in Eq.(8.51) replacing F ′ . In the linear momenta case, Eqs.(8.53)–(8.55) are

used.
For the situation in which O is quadratic in momenta or in the case where δB O
does contribute in Eq.(8.53), one can replace 1/ (1 − R/M2 ) by exp (R/M2 ). This
R
follows by writing 1/ (1 − R/M2 ) = 0∞ exp [−λ (1 − R/M2 )] dλ, and inserting in
Eq.(8.50) or Eq.(8.53):
 !!B 
Z ∞ λR
dλ exp (−λ) F A B exp  =
0 M2 A 0

Z ∞
" !n #
∞ X λ
dλ exp (−λ) fn = f0 =
0 n=−p M2 0
∞  n  "   B #
X 1 R
A
fn = F B exp ,
n=−p M2 0 M2 A 0

1
where a Laurent expansion in M2
has been performed. The resulting expression,
"   B #
A R
(∆S)reg = F B exp , (8.56)
M2 A 0

corresponds to the Fujikawa form [120] of the regularization [251] . In the limit
M → ∞, terms of order 1/Mn for n > 0 vanish, whereas terms with n < 0 blow up.
The regularization scheme consists of dropping the terms that blow up. In the Pauli-
Villars regularization, this is achieved by adding fields with appropriate statistics,
couplings and masses to cancel all n < 0 terms. As a consequence, only the n = 0
term remains.
In the quadratic momentum case, Eq.(8.50) can be manipulated into the following
supertrace form [133]
 !B 
1 1
(∆S)reg = − (R−1 δR)A B R (−1)ǫA  , (8.57)
2 1− M 2 A 0

which only involves the quadratic regulator R. Eq.(8.57) shows that if δR = [R, G]
for some G then the anomaly vanishes, as a consequence of the cyclicity of the trace
[133]. Formally, Eq.(8.57) is
"   A #
1 −R ǫA
(∆S)reg = δB − ln (−1) , (8.58)
2 M2 − R A 0

which is tantamount to demonstrating that (∆S)reg satisfies the one-loop anomaly

consistency condition δB (∆S)reg = 0 in Eq.(8.41) since (∆S)reg is a classical BRST
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 117

variation and δB is nilpotent. If the non-BRST-invariant part of the quantity corre-

sponding to [. . .]0 in Eq.(8.58) is local10 , then there is no one-loop anomaly according
to Eqs.(8.39) and (8.40). The expression in square brackets in Eq.(8.58) turns out to
be the one-loop contribution to the effective action. Eq.(8.58) says that the one-loop
anomaly is the BRST variation of this one-loop contribution.
In the approach of ref.[251], Eq.(8.50), (8.53), or (8.56) is evaluated using standard
perturbation theory about a stationary point. Antifields are finally set to zero and R
is evaluated at Σ, i.e, on-shell.
Another approach to anomalies, which retains antifields, is developed in refs.[259,
133, 252, 257]. At one loop, results agree with the above. It has the advantage of
making it easier to compute antifield-dependent terms in the anomaly, if present.
Such terms might arise if there is an anomalous non-closure of BRST transformations
or some other difficulty with a BRST-structure equation. Antifields may be retained
or eliminated at any stage of a computation. A third approach is to use the effective
action Γ in Eq.(8.21) [32, 159, 248]. In this method, antifields must be retained.
Information about anomalies can be obtained using cohomolgical arguments based
on the Wess-Zumino consistency conditions [268]. One must compute the coefficients
of candidate terms using perturbative methods [89, 57, 93, 8].
Eq.(8.57) demonstrates that fields for which δR is zero, do not contribute to
the anomaly. Non-propagating degrees of freedom, such as gauge-fixed fields and
delta-function-generating Lagrange multipliers π, are not expected to contribute [253]
because anomalies arise from loop effects. In practice, the evaluation of anomalies is
performed perturbatively. Consequently, one expands around a stationary point. In
the gauge-fixed basis, this involves expanding about Σ, that is, around Φ∗A = 0, if
the method retains antifields,11 and about ΦA = ΦA A
0 , where the Φ0 satisfy the gauge-
fixed equations of motion. One must be careful to compute F ′A C in Eq.(8.55) before
expanding about the perturbative saddle point. In a delta-function implementation
of gauge-fixing, it is advantageous, at the beginning of a computation, to perform
a canonical transformation that shifts the Lagrange multipliers π by solutions to
equations of motion. These equations are generated by the fields which are being
gauge-fixed. Such a canonical transformation ensures that gauge-fixed fields and the
π do not mix on-shell with the other fields of the system. This is illustrated in the
first and third sample computations of Sect. 9.
The choice of T in Eq.(8.49), which determines the regulator O, is at one’s dis-
posal. The requirements on T are that it be invertible and that it lead to a quadratic
10
The BRST-invariant part, which may be non-local, gives zero contribution to the anomaly since
δB is applied to it.
11
In methods for which antifields are eliminated, one does not expand about Φ∗A = 0 but simply
sets Φ∗A = 0.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 118

regulator R that is negative definite after a Wick rotation to Euclidean space. Mod-
ifying T changes the form of the anomaly. In particular, when more than one gauge
symmetry is present, varying T changes the coefficients aα in Eq.(8.42). If some non-
zero aα are made zero and vice-versa, the anomaly is shifted from being associated
with one type of gauge symmetry to another. This is analogous to the well known
situation for anomalous chiral gauge theories in four dimensions: The anomaly can be
moved from the axial vector sector to the vector sector, if so desired. See, for example,
Sect. 4.1 of ref.[251]. Although we do not present any examples of this effect, it is
well illustrated in ref.[251]. In performing anomaly calculations, it is useful to choose
T to render a computation as simple as possible. For a similar reason, it is also useful
to perform certain canonical transformations before commencing a calculation.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 119

9 Sample Anomaly Calculations

In this section, we present computations of (∆S)reg to see whether the quantum
master equation is violated at the one-loop level. In general, the analysis is compli-
cated and lengthy. For this reason, we treat only three cases: the spinless relativistic
particle, the chiral Schwinger model, and the first-quantized bosonic string. We use
the method of ref.[251], which we have outlined in Sect. 8.7. The first step in the pro-
cedure is to transform to the gauge-fixed basis. One then has the option of performing
additional canonical transformations. They can be used to partially diagonalize the
system, so that potential contributions to the anomaly can be calculated separately
from various sectors. The second step is to compute the matrices K A B and QAB in
Eqs.(8.47) and (8.48). The third step is to select a TAB matrix so that the operator
OA B in Eq.(8.49) can be obtained. A judicious choice of TAB can simplify a computa-
tion. One then obtains F A B from Eq.(8.51) and R from Eq.(8.52) or Eq.(8.54). The
final step is to use the anomaly formula in Eq.(8.56). Standard perturbation theory
is performed, in which one expands about a stationary point and sets antifields to
zero. For sample computations using a method that retain antifields throughout the
computation, see ref.[133, 257]. Other useful results for anomaly calculations can be
found in [251, 78, 75, 133, 162, 193, 14, 64, 195, 257] and references therein.

9.1 Computation for the Spinless Relativistic Particle

In this subsection, we show that the spinless relativistic particle of Sect. 7.1
possesses no anomaly. This example is useful for illustrating the formalism of Sect.
8.7 because the computation is relatively simple.
We use Z
Ψ = dτ C¯ (e − ρ) (9.1)
for the gauge-fixing fermion, where ρ is an arbitrary function of τ . This allows us
to judge potential dependence on the gauge-fixing procedure by varying ρ. Next, a
canonical transformation is performed to the gauge-fixed basis of Sect. 6.6 using Ψ in
Eq.(9.1). One obtains
Z ( ! )
1 ẋ2 C    
S→ dτ − m2 e + x∗µ ẋµ + e∗ + C¯ C˙ + π̄ C¯∗ + e − ρ . (9.2)
2 e e

It is advantageous to perform a canonical transformation that shifts π̄ by the

solution of the equation of motion generated by e. According to the result in Sect.
8.6, canonical transformations do not affect the existence or non-existence of violations
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 120

of the quantum master equation. The variation of SΨ with respect to e yields

1 ẋ2 1 2
π̄ − − m =0 . (9.3)
2 e2 2
The relevant canonical transformation is
1 ẋ2 1 2
π̄ → π̄ + + m ,
2 e2 2
 
∗ ∗ ∗ ẋ
2
d π̄ ∗ ẋµ
e →e + ,π̄ 3 → + x∗µ x∗µ
, (9.4)
e dτ e2
with the other fields and antifields left unchanged. The action becomes
Z  
1 ρ 1
S→ dτ − 2 ẋ2 − ρm2 + C¯C˙ + π̄ (e − ρ) +
e 2e 2
!  )
C ¯∗ 1 ẋ2 π̄ ∗ C ẋµ d ẋµ
x∗µ ẋµ + e C˙ + C
∗
+ m2
− . (9.5)
e 2 e2 e2 dτ e
Let us first determine the overall structure of the computation. From Eq.(9.5),
one finds that the non-zero entries of the matrix K A B are

xµ e π̄ C¯ C
 
x∗µ ∗ ∗ 0 0 ∗
 
e∗  0 0 0 0 ∗
∗ 
K = π̄ 
∗ ∗ 0 0 ∗
, (9.6)
C¯  ∗
∗  ∗ 0 0 0

C∗ 0 0 0 0 0

where the columns and rows are labelled by the corresponding fields and antifields.
We select TAB to be proportional to the identity matrix, except in the ghost sector
for which
C¯ C
!
C¯ 0 −1
¯ =
TCC , (9.7)
C 1 0
and in the x sector for which
(Tx )µν = ηµν . (9.8)
In perturbation theory, the regulator R is evaluated at the stationary point of the
gauge-fixed action. For e, this corresponds to

e|Σ = ρ , (9.9)
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 121

where Σ indicates the stationary-point surface in field space. Using Eq.(9.5), a

straightforward calculation reveals that the non-zero entries of R|Σ are

xµ e π̄ C¯ C
 
xµ ∗ 0 0 0 0
 
e  0 ∗ ∗ 0 0
 
R|Σ = π̄  0 ∗ 0 0 0 . (9.10)
 
C¯ 
 0 0 0 ∗ 0

C 0 0 0 0 ∗

Eq.(9.10) shows that propagation is diagonal within three sectors: the xµ sector, the
e-π̄ sector and the ghost sector. As expected, the canonical transformation in Eq.(9.4)
decouples e from xµ : For the shifted action in Eq.(9.5), one has
 !
∂l ∂r S  d 1 µ d ρ µ 

 = −2 ẋ + 2 ẋ  =0 . (9.11)
∂e∂xµ Σ dτ e2 dτ e3 
Σ

Because a constant TAB matrix has been selected, the non-zero entries of F A B in
Eq.(8.51) are the same as in Eq.(9.6). From the structure of Eqs.(9.6) and (9.10), one
sees that the anomaly computation separates into contributions from the xµ sector,
the e-π̄ sector and the ghost sector. The propagating fields are xµ , C¯ and C. The
field π̄ serves as a Lagrange multiplier for setting e equal to ρ. Hence, e and π̄ are
non-propagating and should not contribute to the anomaly according to the analysis
in Sect. 8.7 [253]. For this particular system, the contribution is zero because F A B
in the e-π̄ sector is off-diagonal. It is also clear that C and C¯ do not contribute to the
anomaly since F A B is zero for all ghost entries. One only needs to consider the xµ
sector.
In what follows we use a subscript x for quantities associated with xµ . Applying
Eqs.(8.47) and (8.48) to Eq.(9.5), one arrives at
d
(Kx )µ ν = δ µ ν e−1 C , (9.12)
dτ
and
(Qx )µν = ηµν Qx , (9.13)
where Qx without µν subscripts is defined by
d −1 d d d
Qx ≡ −2 e + ρe−2 . (9.14)
dτ dτ dτ dτ
Here and below, the derivative dτd acts on everything to the right including e, ρ and
the function to which Qx is applied. Some contributions to Eq.(9.14) come from the
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 122

shifts in Eq.(9.4). We have also dropped terms proportional to π̄ ∗ and C¯∗ because
they will not contribute, when expanding about the stationary point.
Since the Tx matrix is (Tx )µν = ηµν , the regulator matrix in Eq.(8.52) is

(Rx )µ ν = δ µ ν Qx . (9.15)

Because δB Tx = 0,
(Fx )µ ν = (Kx )µ ν , (9.16)
where (Kx )µ ν is given in Eq.(9.12).
All the relevant matrices of Sect. 8.7 for the computation of the anomaly have
been obtained. At this stage, one expands about the stationary point of the gauge-
fixed action. The field e is set equal to the function ρ according to Eq.(9.9). The
regulator matrix becomes
d −1 d
(Rx )µ ν |Σ = −δ µ ν ρ , (9.17)
dτ dτ
so that on-shell
(∆S)reg =
 ! 
Z Z∞
dk d − dτd ρ−1 dτd
D dτ  exp (−ikτ ) ρ−1 C exp exp (ikτ ) , (9.18)
2π dτ M2
−∞ 0

where we have used the form of (∆S)reg in Eq.(8.56). The trace over field indices
R
A produces a factor of dτ and a factor of D (because the number of xµ fields is
D and each contributes equally). The operator trace is evaluated using a complete
set of momentum-space functions, thereby generating the factors exp (±ikτ ). The
calculation in Eq.(9.18) is performed in Appendix C, where it is shown that the
integrand is an odd function of k. Consequently,

(∆S)reg = 0 . (9.19)

The calculation of the xµ contribution is even simpler using Eq.(8.57). To compute

δB R, note that
δB e = (e, S) = C˙ . (9.20)
Using this equation, Eq.(9.14) and Eq.(9.15), one finds
!
d d d d
(δB Rx ) µ
ν =δ µ
ν 2 e−2 C˙ − 2 ρe−3 C˙ , (9.21)
dτ dτ dτ dτ

so that
(δB Rx )µ ν |Σ = 0 . (9.22)
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 123

When δB Rx is substituted into Eq.(8.57), one gets

(∆S)reg = 0 ,

in agreement with Eq.(9.19). The absence of a violation of the quantum master

equation means that the spinless relativistic particle theory is gauge-invariant even
at the quantum level.

9.2 The Abelian Chiral Schwinger Model

In this subsection, we analyze the abelian chiral Schwinger model in two-dimensions.
It is an anomalous gauge theory and a particularly simple example that illustrates the
formalism of Sect. 8.7. The model contains an abelian gauge field, i. e., “a photon”
Aµ , and a charged left-handed fermion. It is governed by the following classical action
h i Z  
1
S0 Aµ , ψ, ψ̄ = 2
d x − 2 F µν Fµν + ψ̄i/
Dψ , (9.23)
4e
where e is the electromagnetic coupling constant. Although we take ψ to be a Dirac
fermion, we use a covariant derivative that couples the photon only to the the right-
moving component:
i/
D = i/∂ + AP
/ − . (9.24)
In other words, P− ψ is charged but P+ ψ is neutral. Here P± are the chiral projectors
1 1
P− = (1 − γ5 ) , P+ = (1 + γ5 ) . (9.25)
2 2
In two-dimensions, they project onto right- and left-moving states. Hence, the left-
moving fermion P+ ψ is a free particle and decouples.
A slash through a vector Vµ represents γ µ Vµ : V / = γ µ Vµ . In two-dimensions the
γ µ are 2 × 2 matrices satisfying γ µ γ ν + γ ν γ µ = −2η µν , and γ5 is defined as γ5 ≡ γ 0 γ 1 .
The action in Eq.(9.23) is invariant under the finite gauge transformations

A′µ = Aµ + ∂µ ε , ψ ′ = exp{iP− ε}ψ , ψ̄ ′ = ψ̄ exp{−iP+ ε} . (9.26)

Although this model has not been discussed in previous sections, it is straightfor-
ward to apply the field-antifield formalism [61, 63]. The proper solution for the gauge
sector corresponds to the solution for the Yang-Mills example given in Sect. 5.2 for
d = 2 and for a U(1) group. In addition to the antifield A∗µ of the photon, one has
commuting antifields ψ ∗ and ψ̄ ∗ for the fermions. The proper solution of the master
equation is
Z h i
S = S0 + d2 x A∗µ ∂ µ C + i(ψ ∗ )t P− ψ C − iψ̄ ∗ P+t ψ̄ t C , (9.27)
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 124

where C is the ghost field associated! with the gauge parameter ε. The superscript !
ψ1   ψ̄ 1
1 2 t
t stands for transpose: ψ = and ψ̄ = ψ̄ , ψ̄ , so that ψ̄ = and
ψ2 ψ̄ 2
(ψ ∗ )t = ((ψ 1 )∗ , (ψ 2 )∗ ).
A formal computation using the expression for ∆ in Eq.(6.7) reveals that only the
fermion sector contributes to ∆S. A more detailed analysis using a regularization
procedure confirms this.12 Therefore, we focus on the contribution to ∆S from ψ and
ψ̄. Gauge-fixing is not needed because propagators for the fermions already exist.
For these reasons, it is not necessary to consider gauge-fixing auxiliary fields, nor a
non-minimal proper solution.
Using Eqs.(8.47) and (9.27), one finds that the K matrix is given by
!
P− 02
K = −iC , (9.28)
02 −P+t
!
0 0
where all entries are 2×2 matrices, e. g., 02 = , so that K is a 4×4 matrix. In
0 0
Eq.(9.28) and throughout this subsection, we label the rows and columns of matrices
in the order ψ 1 , ψ 2 , ψ̄ 1 , ψ̄ 2 . From Eqs.(8.48) and (9.23), the matrix Q for the fermion
sector is !
02 −iD /̃
Q= , (9.29)
i/
D 02
R R
where D
/̃ is defined by d2 xψ̄i/
Dψ = − d2 xψ t iD
/̃ ψ̄ t . More precisely,

iD ∂t −A
/̃ = i/ / t P+t , (9.30)

where, here, the superscript t indicates the transpose of a matrix in Dirac-index space.
For the matrix T , we choose
!
02 12
T = , (9.31)
−12 02
!
1 0
where 12 = . Because δB T = 0, the matrix F in Eq.(8.51) is equal to the
0 1
matrix K in Eq.(9.28). Using Eq.(8.49), one finds
!
i/
D 02
O= . (9.32)
02 iD
/̃
12
The computation made in ref.[78] for a pure Yang-Mills theories in four dimensions supports
the idea that gauge fields and ghosts produce a BRST trivial contribution to (∆S)reg .
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 125

Then from Eq.(8.54), one obtains

!
−/
D/D 02
R= , (9.33)
02 −D
/̃D
/̃
for the regulator matrix.
Let us use the following representation of the gamma matrices:
! !
0 1 0 1 1 2 0 1
γ =σ = , γ = iσ = , (9.34)
1 0 −1 0
so that !
5 0 1 3 −1 0
γ = γ γ = −σ = , (9.35)
0 1
and ! !
1 0 0 0
P− = , P+ = . (9.36)
0 0 0 1
With this representation,
! !
0 i∂+ 0 i∂− − A−
i/
D= , iD
/̃ = , (9.37)
i∂− + A− 0 i∂+ 0
where
∂± = ∂0 ± ∂1 , A− = A0 − A1 . (9.38)
The entries in R of Eq.(9.33) are
!
i∂+ (i∂− + A− ) 0
−/
D/D= ,
0 (i∂− + A− ) i∂+
!
(i∂− − A− ) i∂+ 0
−D
/̃D
/̃ = . (9.39)
0 i∂+ (i∂− − A− )
The derivative ∂+ acts to the right, so that it differentiates A− as well as any function
to which R is applied. Note that R is diagonal.
In general, the matrix F ′ in Eq.(8.55) has two contributions. However, the term
proportial to δB O does not contribute, upon taking the trace, because O is off-
diagonal and R is diagonal when the above gamma matrices are used. Hence, one
may take
F′ = F = K , (9.40)
where K is given in Eq.(9.28). Summarizing, for the computation of ∆S in Eq.(8.56),
one uses F = K in Eq.(9.28) and R given by Eqs.(9.33) and (9.39). Incorporating
the projectors P± in K, one obtains
     
R+ R−
∆S = −i T r C exp − exp , (9.41)
M2 M2 0
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 126

where there is no trace in Dirac-index space, only in function space. In Eq.(9.41),

R± = ∂ µ ∂µ ± i(∂+ A− ) ± iA− ∂+ . (9.42)

where we use a parenthesis around (∂+ A− ) to indicate that ∂+ acts only on A− .

In Appendix C, the computation of Eq.(9.42) is performed. One finds
Z
i
A1 = ∆S = d2 x C (ǫµν ∂µ Aν − ∂µ Aµ ) , (9.43)
4π
where ǫ10 = 1 = −ǫ01 . Note that the anomaly is consistent since
i Z 2
δB A1 = d x C (ǫµν ∂µ ∂ν C − ∂µ ∂ µ C) = 0 . (9.44)
4π
The first term vanishes by the antisymmetry property of ǫµν , while the second term
vanishes by integration by parts and the anticommuting nature of C. A local coun-
terterm Ω1 cannot be added to the action to eliminate A1 via Eq.(8.40). If one takes
1 R 2
Ω1 = − 8π d xAµ Aµ , then the second term in Eq.(9.43) is eliminated, but the term
i R
8π
d2 x Cǫµν Fµν in A1 remains.

9.3 Anomaly in the Open Bosonic String

In this subsection, we investigate the violation of the quantum master equation
for the first-quantized open bosonic string when the dimension of space-time is not
26. The bosonic contribution was explicitly computed in refs.[251, 133]. We gauge-fix
the action using the fermion Ψ in Eq.(7.42). It depends on the conformal factor ρ.
It turns out that when D 6= 26, there is an anomaly. As a consequence, the theory
depends on ρ. Although the theory is classically gauge-invariant, one of the four
gauge symmetries is violated by quantum effects. This anomalous gauge symmetry
cannot be fixed for D 6= 26.
Let us apply the one-loop anomaly analysis given in Sect. 8.7 to the bosonic string.
∂Ψ
First, we perform a canonical transformation with Ψ so that Φ∗A → Φ∗A + ∂Φ A . This

leads to the following shifts in antifield fields

e∗τ τ → e∗τ τ + C¯ + C¯σ , e∗σ σ → e∗σ σ + C¯ − C¯σ ,

e∗τ σ → e∗τ σ + C¯τ σ + C¯τ , e∗σ τ → e∗σ τ + C¯τ σ − C¯τ ,

C¯∗τ → C¯∗τ + eτ σ − eσ τ , C¯∗σ → C¯∗σ + eτ τ − eσ σ ,
C¯∗ → C¯∗ + eτ τ + eσ σ − 2ρ−1/2 , C¯∗τ σ → C¯∗τ σ + eτ σ + eσ τ . (9.45)
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 127

Additional terms are produced in the total action Stotal = S + Saux of Eqs.(5.30) and
(7.41) given by
Z Zπ n   
S→S+ dτ dσ C¯ + C¯σ (C n ∂n eτ τ − eτ n ∂n C τ ) + C¯ − C¯σ (C n ∂n eσ σ − eσ n ∂n C σ )
0
   
+ C¯τ σ + C¯τ (C n ∂n eτ σ − eτ n ∂n C σ ) + C¯τ σ − C¯τ (C n ∂n eσ τ − eσ n ∂n C τ )
        
+ C¯ + C¯σ eτ τ + C¯ − C¯σ eσ σ + C¯τ σ + C¯τ eτ σ + C¯τ σ − C¯τ eσ τ C
         o
− C¯ + C¯σ eσ τ + C¯ − C¯σ eτ σ + C¯τ σ + C¯τ eσ σ + C¯τ σ − C¯τ eτ τ C τ σ , (9.46)
and
Z Zπ n  
Saux → dτ dσ π̄ C¯∗ + eτ τ + eσ σ − 2ρ−1/2 +
0
     o
π̄τ C¯∗τ + eτ σ − eσ τ + π̄σ C¯∗σ + eτ τ − eσ σ + π̄τ σ C¯∗τ σ + eτ σ + eσ τ . (9.47)
At this stage, it is desirable to shift fields by the solutions to the equations of
motion of the ea m , by using a canonical transformation. Such a shift guarantees that
the quadratic form QAB is on-shell diagonal in the ea m sector. This avoids mixing of
the X µ and ghost sectors with the π̄-ea m sector. Variations of Stotal in the gauge-fixed
basis with respect to the ea m produce the equations for the four π̄. Using a subscript
0 to denote the solutions, and ignoring terms proportional to antifields, one finds that
!
∂S ∂S
2 (π̄)0 ≡ − τ
+ − 2π̄ =
∂eτ ∂eσ σ

(eτ τ + eσ σ )eLX − e (∂τ X µ Dτ Xµ − ∂σ X µ Dσ Xµ ) +

 
¯ n + C¯ (∂n C n − 2C) + C¯τ σ (∂τ C σ + ∂σ C τ + 2C τ σ ) +
2∂n CC
C¯τ (∂τ C σ − ∂σ C τ ) + C¯σ (∂τ C τ − ∂σ C σ ) ,
!
∂S ∂S
2 (π̄τ σ )0 ≡ − σ
+ − 2π̄τ σ =
∂eτ ∂eσ τ
−(eτ σ + eσ τ )eLX − e (∂σ X µ Dτ Xµ − ∂τ X µ Dσ Xµ ) +
 
2∂n C¯τ σ C n + C¯ (∂τ C σ + ∂σ C τ + 2C τ σ ) + C¯τ σ (∂n C n − 2C)
−C¯τ (∂τ C τ − ∂σ C σ ) − C¯σ (∂τ C σ − ∂σ C τ ) ,
!
∂S ∂S
2 (π̄τ )0 ≡ − − − 2π̄τ =
∂eτ σ ∂eσ τ
(eτ σ − eσ τ )eLX − e (∂σ X µ Dτ Xµ + ∂τ X µ Dσ Xµ ) +
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 128

 
2∂n C¯τ C n − C¯ (∂τ C σ − ∂σ C τ ) − C¯τ σ (∂τ C τ − ∂σ C σ )
+C¯τ (∂n C n − 2C) + C¯σ (∂τ C σ + ∂σ C τ − 2C τ σ ) ,
!
∂S ∂S
2 (π̄σ )0 ≡ − τ
− − 2π̄σ =
∂eτ ∂eσ σ
(eσ σ − eτ τ )eLX − e (∂τ X µ Dτ Xµ + ∂σ X µ Dσ Xµ ) +
 
2∂n C¯σ C n + C¯ (∂τ C τ − ∂σ C σ ) + C¯τ σ (∂τ C σ − ∂σ C τ )
+ C¯τ (∂τ C σ + ∂σ C τ − 2C τ σ ) + C¯σ (∂n C n − 2C) , (9.48)
where the X µ -lagrangian density LX is defined to be
e
LX ≡ (Dτ X µ Dτ Xµ − Dσ X µ Dσ Xµ ) . (9.49)
2
The canonical transformation of interest is given by

π̄ → π̄ + (π̄)0 , π̄n → π̄n + (π̄n )0 , π̄τ σ → π̄τ σ + (π̄τ σ )0 ,

∂r (π̄)0 ∂r (π̄τ )0 ∂r (π̄σ )0 ∂r (π̄τ σ )0
Φ∗A → Φ∗A − π̄ ∗ A
− π̄ ∗τ A
− π̄ ∗σ A
− π̄ ∗τ σ , (9.50)
∂Φ ∂Φ ∂Φ ∂ΦA
where in the first equation, n stands for τ or σ.
The goal of the next few paragraphs is to obtain F A B and RA B so that the
anomaly in Eq.(8.53) can be computed. One must first calculate K A B and QAB and
specify TAB . Simplifications occur for the following reasons. In the final gauge-fixed
form of the action in Eq.(7.45), the propagating fields are X µ , C¯n and C n where n
represents τ and σ.13 By the argument in Sect. 8.7, only these fields contribute; the
F A B and RA B associated with non-propagating fields do not enter the calculation
[253]. Furthermore, by a judicial choice of TAB , RA B can be made diagonal. Hence,
only the diagonal components of K A B , δB OA B and F A B for X µ , C¯n and C n need to
be computed. In what follows, we use a subscript X, C¯ and C to denote respectively
a matrix restricted to the subspace corresponding to X µ , C¯n and C n . The subscript
“ghost” is used to denote the combined C¯n and C n subspace.
After the above two canonical transformations have been performed, the compu-
tation is straightforward. From Eq.(8.47), one finds that the K A B in the three sectors
are
(KX )µ ν = δ µ ν C n ∂n ,
13
This is made clear by setting C = Cˆ + 12 (∂τ C τ + ∂σ C σ ) − 2ρ1/2 C n ∂n ρ−1/2 and C τ σ = Cˆτ σ −
1 σ τ τσ
2 (∂τ C + ∂σ C ). Throughout this subsection, one should think of C and C as standing for these
ˆ ˆτσ
combinations of fields. In terms of C and C , the gauge-fixed action in Eq.(7.45) is block diagonal
and C,¯ C¯τ σ , Cˆ and Cˆτ σ are non-propagating fields. The calculations in this subsection should be
performed in terms of Cˆ and Cˆτ σ . At the end of the computation, one sets Cˆ = C − 21 (∂τ C σ − ∂σ C τ )+
2ρ1/2 C n ∂n ρ−1/2 and Cˆτ σ = C τ σ + 12 (∂τ C σ − ∂σ C τ ), thereby returning to the original fields.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 129

3
(KC¯)τ τ = (KC¯)σ σ = −C n ∂n − (∂n C n ) + C ,
2
1 1
(KC )τ τ = −C n ∂n + (∂n C n ) + (∂τ C τ − ∂σ C σ ) ,
2 2
1 1
(KC )σ σ = −C n ∂n + (∂n C n ) − (∂τ C τ − ∂σ C σ ) , (9.51)
2 2
where the presence of a parenthesis around a derivative indicates that it acts only on
fields within the parenthesis. Throughout this subsection, the absence of a parenthesis
means that the derivative acts on everything to the right. The terms for KX and KC
arise from differentiating Eq.(5.30), whereas those for KC¯ come from Saux in Eq.(9.47)
after the π̄n shifts in Eq.(9.50) have been performed.
The quadratic forms QAB are
(QX )µν = ηµν QX , (9.52)
where
   
QX = Dτ† e 1 − eρ−1/2 (eτ τ + eσ σ ) Dτ − Dσ† e 1 − eρ−1/2 (eτ τ + eσ σ ) Dσ

+ Dτ† eρ−1/2 ∂τ − ∂τ eρ−1/2 Dτ − Dσ† eρ−1/2 e∂σ + ∂σ eρ−1/2 Dσ (9.53)

for the X µ sector, where the covariant derivatives Dτ and Dσ are given in Eq.(3.71),
and where Dτ† = −∂τ eτ τ −∂σ eτ σ and Dσ† = −∂τ eσ τ −∂σ eσ σ . Since
 derivatives act to the

right, ∂τ and ∂σ in Dτ in first the term in Eq.(9.53) act on e 1 − eρ−1/2 (eτ τ + eσ σ ) ,
†

on the vielbeins in Dτ† and Dτ , and on any function to which the operator QX is
applied. Likewise, for the other derivatives. The terms in Eq.(9.53) come from the
original action S0 in Eq.(3.79), as well as from the π̄ shifts of Eq.(9.50) in Saux of
Eq.(9.47). For the ghost sector, let
 
C¯τ
 ¯ 
 Cσ 
V = 
 Cτ  . (9.54)
 
Cσ
Then the ghost part of the action is
Z Zπ
1
Sghost = dτ dσ V t Qghost V , (9.55)
2
0

where the superscript t on V t stands for transpose. The ghost quadratic form is
 
0 0 ρ−1/2 ∂σ −ρ−1/2 ∂τ
 
0 0 −ρ−1/2 ∂τ ρ−1/2 ∂σ 
Qghost =
 ∂ ρ−1/2

 . (9.56)
 σ −∂τ ρ−1/2 0 0 
−∂τ ρ−1/2 ∂σ ρ−1/2 0 0
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 130

The terms in Qghost originate from Eq.(9.46) and from the π̄ shifts of Eq.(9.50) in Saux
of Eq.(9.47). The dependence on ea m cancels between the two contributions leaving
only a dependence on ρ.
For the matrix TAB of Sect. 8.7, we choose

(TX )µν = eηµν , (9.57)

and  
0 e−2 0 0
 
 −e−2 0 0 0 
Tghost = −i 
0
 , (9.58)
 0 0 e2 

0 0 −e2 0
where Tghost acts in the space of antighosts and ghosts given in Eq.(9.54). In the
X µ sector the operator O, defined in Eq.(8.49), is e−1 δ µ ν QX and is equal to (RX )µ ν ,
where QX is given in Eq.(9.53). In the ghost sector, O is
!
02 e2 ρ−1/2∂/
Oghost = i , (9.59)
e−2∂/ρ−1/2 02

where each entry is a two by two matrix. It is somewhat accidental that the Dirac
operator in two-dimensional Minkowski space

∂/ ≡ γ n ∂n = γ τ ∂τ + γ σ ∂σ , (9.60)

enters in Eq.(9.59), if the gamma matrices

! !
τ 3 1 0 σ 2 0 −1
γ =σ = , γ = −iσ = (9.61)
0 −1 1 0

are used. They satisfy

γ n γ m + γ m γ n = −2η nm . (9.62)
The ghost-sector regulator matrix Rghost is
!
2 RC¯ 02
Rghost = O = , (9.63)
02 RC

where
RC¯ = −e2 ρ−1/2∂/e−2∂/ρ−1/2 , RC = −e−2∂/e2 ρ−1∂/ . (9.64)
To compute F A C in Eq.(8.51), one needs the BRST variation of TAB . Using

δB e = C n ∂n e + e∂n C n − 2eC (9.65)

J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 131

AD
and Eqs.(9.57) and (9.58), it is straightforward to calculate (T −1 ) (δB T )DC . The
result is combined with K A C in Eq.(9.51) to arrive at

(FX )µ ν = δ µ ν FX ,

where
1 1
FX = C n ∂n +(∂n C n ) + e−1 C n (∂n e) − C , (9.66)
2 2
µ
for the X sector. For the ghost sector
1
(FC¯)τ τ = (FC¯)σ σ = −C n ∂n − (∂n C n ) + e−1 C n (∂n e) − C ,
2
1 1
(FC )τ τ = −C n ∂n − (∂n C n ) − e−1 C n (∂n e) + 2C + (∂τ C τ − ∂σ C σ ) ,
2 2
1 1
(FC )σ σ = −C n ∂n − (∂n C n ) − e−1 C n (∂n e) + 2C − (∂τ C τ − ∂σ C σ ) . (9.67)
2 2
Although δB Oghost is non-zero, it does not contribute because the regulator matrix
R is block diagonal in the C¯n and C n sectors. It turns out that, even in each sector,
only diagonal terms contribute because of the nature of R. Hence, in Eq.(9.67) we
display only the diagonal part of F A C . Finally, since the contribution from Cτ turns
out to be equal to the contribution from Cσ , the two ∂τ C τ − ∂σ C σ terms in Eq.(9.67)
for FC cancel. For the rest of this section, we drop these terms.
The final step is the computation of (∆S)reg . Since δB O does not contribute, we
may use Eq.(8.56). At this stage, we expand about the classical saddle point, denoted
by Σ, corresponding to the solution to the equations of motion. We set the π̄ equal
to zero and evaluate the ea m as in Eq.(7.43).
To determine the regulators and F A B at Σ, note that

e|Σ = ρ . (9.68)

It is useful to express the regulators in terms of symmetric operators R̃. For X µ ,

(RX )|Σ = ρ−1 (−∂τ ∂τ + ∂σ ∂σ ) = ρ−1/2 R̃X ρ1/2 , (9.69)

where
R̃X = ρ−1/2 ∂ n ∂n ρ−1/2 . (9.70)
For ghosts,
RC¯|Σ = ρR̃C¯ρ−1 , RC |Σ = ρ−1 R̃C ρ , (9.71)
where
R̃C¯ = −ρ1/2∂/ρ−2∂/ρ1/2 , R̃C = −ρ−1∂/ρ/
∂ ρ−1 . (9.72)
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 132

The conjugating factors in Eqs.(9.69) and (9.71) can be commuted past the oper-
ators F A B to arrive at the following equivalent form for (∆S)reg
 !!ν
 µ R̃X
(∆S)reg =  F̃X exp +
ν M2 µ
!!n !!n #
 n R̃C¯  n R̃C
F̃C¯ exp + F̃C exp , (9.73)
n M2 n
n M2 n 0
where  µ  
1
F̃X µ
= δ ν C ∂n + (∂n C n ) − C n
,
ν 2
 τ  σ 1
F̃C¯ = F̃C¯ = −C n ∂n − (∂n C n ) − C ,
τ σ 2
 τ  σ 1
F̃C = F̃C = −C n ∂n − (∂n C n ) + 2C . (9.74)
τ σ 2
The TAB matrix has been judiciously chosen so that the violation in the quantum
master equation is proportional to C. The coefficient of C n in Eq.(9.73) vanishes.
To see this, let ψr and −Er2 be the eigenfunctions and eigenvalues of any of the R̃
operators:
R̃ψr = −Er2 ψr . (9.75)
The ψr can be chosen to be real. The terms in (9.74) involving C n enter in the
combination C n ∂n + 12 (∂n C n ). Such a combination gives a zero contribution to (∆S)reg
since   !!
n 1 n R̃
C ∂n + (∂n C ) exp =
2 M2
Z Zπ   !
X 1 E2
dτ dσ ψr (τ, σ) C ∂n + (∂n C n ) ψr (τ, σ) exp − r2
n
=
r 2 M
0
Z Zπ !!
1 X E2
dτ dσ ∂n ψr2 (τ, σ) C exp − r2
n
→0 ,
2 r M
0
where, in the last step, we assume that quantities fall off sufficiently fast at large τ
and obey appropriate boundary conditions at σ = 0 and σ = π. The reader can also
verify the absence of a C n anomaly directly by using the methods in Appendix C.
To evaluate the terms in (∆S)reg proportional to C, let

Hr ≡ −ρ−(r+1)/2∂/ρr∂/ρ−(r+1)/2 . (9.76)

The tilde regulators in Eqs.(9.70) and (9.72) are

R̃X = H0 , R̃C¯ = H−2 , R̃C = H1 , (9.77)

J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 133

except that, for X µ , R̃X acts in a D-dimensional space, whereas H0 acts in a two-
dimensional space, since −/ ∂∂/ = I2 ∂ n ∂n = I2 (−∂τ ∂τ + ∂σ ∂σ ), where I2 denotes the
two-dimensional unit matrix. The non-zero terms in (∆S)reg in Eq.(9.73) all involve

   Z Zπ
Tr Hr
exp ≡ dτ dσ κr . (9.78)
2 M2 0
0

The trace T r is over both 2 by 2 gamma space and function space. The coefficients
κr , which are computed in Appendix C, are
1
κr = (3r + 1) ∂ n ∂n ln (ρ) . (9.79)
24πi
The contributions to the violation of the quantum master equation from the X µ ,
C¯n , and Cn sectors are respectively (−1)Dκ0 , (−1)2κ−2 , and (+2)2κ1 , where the
factors in parentheses are the coefficients of C in the F̃ in Eq.(9.74). The total
anomaly A on-shell is [210, 121]
π
(D − 26) Z Z
(∆S)reg =i dτ dσ C∂ n ∂n ln (ρ)
24π
0

Z Zπ
(D − 26)
=i dτ dσ C ∂ n ∂n ln (e)|Σ . (9.80)
24π
0

It is absent when D = 26. For D 6= 26 no local counter term Ω1 can be added to

cancel the violation of the quantum master equation via Eq.(8.39) and the theory is
anomalous.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 134

10 Brief Discussion of Other Topics

The following are discussed in this section: applications to global symmetries,
a geometric interpretation of the field-antifield formalism, locality, cohomology, the
equivalence of lagrangian and hamiltonian approaches, unitarity, the antibracket for-
malism in a general coordinate system, the D = 26 closed bosonic string field theory,
and the extended formalism for anomalous gauge theories. One topic not addressed
is the anti-BRST symmetry [69, 202, 3, 32, 20, 135, 137, 160, 12, 155, 164].

10.1 Applications to Global Symmetries

Certain models with continuous rigid symmetries share some of the characteris-
tics of gauge theories, such as the closure only on-shell of the commutator algebra and
the presence of field-dependent structure constants. Global supersymmetric theories
without auxiliary fields and models employing non-linear realizations of rigid sym-
metries are often examples of algebras that do not close off-shell. The antibracket
formalism can be used to assist in the analysis of such theories [33, 159]. Even though,
in the rigid-symmetry case, the parameters εα in the transformation law in Eq.(2.1)
are not functions of the space-time variable x, there is still the notion of a symmetry
structure. In other words, the analogs of the structure equations in Sect. 2, such as
the Noether identity, the Jacobi identity, etc., still exist. There are two differences for
the globally symmetric case: (a) everywhere εα appears, it is a constant and (b) the
compact notation for Greek indices, associated with gauge transformations, involves
a discrete sum but not an integral over space-time. For Latin indices, associated with
the φi , repeated indices still indicate a space-time integral. Taking into account the
above two differences, the equations in Sect. 2 hold for the globally symmetric case.
The development of an antibracket-like formalism proceeds as in Sect. 4. Since the
gauge parameters εα are not functions of the space-time variable x, one introduces
constant ghosts C α . The antifields φ∗i for the original fields φi are space-time functions,
but the antifields for ghosts are constants. Grassmann statistics and ghost numbers
are assigned as in the gauge-theory case. In the antibracket and elsewhere, functional
derivatives with respect to ghosts and antifields of ghosts are replaced by ordinary
partial derivatives. The proper solution S in Eq.(4.29) is a generating functional for
the structure tensors. The structure equations are encoded in the classical master
equation (S, S) = 0. Of course, since global symmetries do not affect the rank of the
hessian of S at a stationary point, the concept of properness has little meaning: If
one wants to treat all global symmetries via an antibracket-like formalism, one should
proceed by mimicking the gauge case.
Since global symmetries do not upset the development of perturbation theory, no
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 135

gauge-fixing procedure via a fermion Ψ is implemented. In the quantum theory, the

ghosts are only technical tools. They should not be considered as quantum fields.
Antifields are still interpreted as sources for rigid symmetries. To perform standard
perturbation theory, antifields can be set to zero. Alternatively, one can differentiate
with respect to antifields before setting them to zero to obtain global Ward identities.
A third approach is to introduce sources J for fields via Eq.(8.14), retain antifields,
and construct the effective action Γ as described in Sect. 8.4. Anomalous violations
of global symmetries can be analyzed by searching for violations of the Zinn-Justin
equation (Γ, Γ)c = 0 (see Eq.(8.21)). Examples of anomalous global symmetries are
the axial vector currents of massless four-dimensional QCD. An application to the
D = 4 supersymmetric Wess-Zumino model is given in ref.[159].

10.2 A Geometric Interpretation

This subsection discusses a geometric interpretation of the field-antifield formal-
ism, as presented by E. Witten [273]. See also refs.[175, 155, 21, 218, 176].
The geometric intepretation is made clearer if we first assume that no fermionic
fields are present. Let M denote the manifold of infinite-dimensional function space.
The classical fields ΦA of the theory are local coordinates for M. Then, ∂Φ∂ A is
a local basis for the tangent space T M of vector fields. Likewise, dΦA is a basis
for the cotangent space T ∗ M consisting of differential forms. There exists a nat-
ural quadratic form on T M ⊕ T ∗ M given by hdΦA , ∂Φ∂B i = δBA , hdΦA , dΦB i = 0,
h ∂Φ∂ A , ∂Φ∂ B i = 0. Introduce, in an ad hoc manner, two quantities z A and wA and
associate z A with dΦA and wA with ∂Φ∂ A , i.e.,

∂
z A ↔ dΦA , wA ↔ . (10.1)
∂ΦA
Consider the Clifford algebra for z A and wA determined by the quadratic form h , i,
namely
{z A , wB } = δBA , {z A , z B } = 0 , {wA , wB } = 0 , (10.2)
where { , } denotes the anticommutator: {x, y} = xy + yx. A possible representation
of the Clifford algebra regards the z A as creation operators and the wA as destruction
operators. Then, the most general state at a point Φ on the manifold is created by
Ω (Φ, z) = Ω0 (Φ) + ΩA (Φ) z A + 21 ΩAB (Φ) z B z A + . . . acting on a Fock-space vacuum
|0i, defined by wA |0i = 0 for all A, i. e., it is annihilated by all the wA . Representing
the wA as ∂z∂A , |0i can be taken to be 1 when considered as a function of the z A .
With the association z A ↔ dΦA , one sees that Ω is equivalent to an element of the
exterior algebra of differential forms on M, in which differential forms are multiplied
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 136

by using the wedge product ∧. When supplemented with the exterior derivative d,
this structure becomes the de Rham complex [107, 96]. In summary, one has an
irreducible representation of the Clifford algebra in Eq.(10.2) at each point of the
manifold.
The Clifford algebra in (10.2) is symmetrical in its treatment of the elements z
and w. Hence, one can reverse the above viewpoint and regard the wA as creation
operators and the z A as annihilation operators. In this picture, let us identify the
antifields Φ∗A of the antibracket formalism with the vector-field-like objects wA . The
most general state at a point Φ is created by
1
F [Φ, Φ∗ ] = F0 (Φ) + F A (Φ)Φ∗A + F AB (Φ)Φ∗B Φ∗A + . . . (10.3)
2
acting on a state |0i′ that is annihilated by all z A . In this picture, denoted by R′ by
∂r ∗ ′
E. Witten, z A = ∂Φ ∗ , wA = ΦA , and |0i can be taken to be 1 when regarded as a
A
function of the Φ∗A . Two elements F and G in the form of Eq.(10.3) are multiplied
∂r
using {Φ∗A , Φ∗B } = 0. Exploiting the association dΦA → z A = ∂Φ ′
∗ in the R picture,
A
∂r A ∂r ∂r
the exterior derivative d ≡ ∂Φ A dΦ becomes ∂Φ A ∂Φ∗ F = −∆F when acting on a
A
general functional F of the type in Eq.(10.3). Here, we have used the definition of ∆
given in Eq.(6.7). In computing ∆F , one treats ΦA and Φ∗B as independent variables.
Because the exterior derivative is nilpotent, ∆ satisfies ∆2 = 0. In short, one arrives
at a dual picture of the de Rham complex. It is isomorphic to the standard de Rham
complex but not in a natural way because there is no preferred manner of associating
the above two Fock-space vacuums |0i and |0i′. The state |0i′ of the R′ picture is
Q
represented as (f12... ) A z A in the first picture, where f12... is arbitrary. A natural
choice does exist if M is endowed with a measure dµ = (µ12... ) dΦ1 ∧ dΦ2 ∧ . . .. Then,
one can take f = µ.
If fermionic fields are present, M is a supermanifold. Then,   { , } appears as a
ǫx ǫy
graded commutator: {x, y} = xy − (−1) yx. Note that ǫ z A = ǫA + 1 = ǫ (wA ),
so that z A and wA have the opposite statistics of ΦA . For the bosonic case, ǫA = 0
for all A, and { , } becomes the usual anticommutator.
In the R′ picture, the Clifford algebra in Eq.(10.2) is satisfied when
∂r
wA = Φ∗A , z A = (−1)ǫA . (10.4)
∂Φ∗A
Then, −d becomes, when acting on F of Eq.(10.3),
∂r ∂r
(−1)ǫA +1 F = ∆F . (10.5)
∂ΦA ∂Φ∗A
The nilpotent operator ∆ of the field-antifield formalism is identified with minus the
exterior derivative.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 137

For elements F and G in Eq.(10.3), the antibracket ( , ) is defined by (F [Φ, Φ∗ ] , G [Φ, Φ∗ ])

∂r F ∂l G∂r F ∂l G
≡ − ∂Φ
∂ΦA ∂Φ∗A ∗ ∂ΦA . Using ∆, the antibracket can be expressed as
A

∆(F G) − F ∆(G) − (−1)ǫG ∆(F )G = (−1)ǫG (F [Φ, Φ∗ ], G[Φ, Φ∗ ]) . (10.6)

Eq.(10.6) shows that the antibracket is the obstruction of ∆ to be a derivation from

the right. Once ∆ has been defined, one can take the left-hand side of Eq.(10.6) to
be the definition of the antibracket after multiplying by (−1)ǫG .
Summarizing, one has the following analogy. If one thinks of function space as a
supermanifold, then antifields are the basis vectors in the R′ picture of the de Rham
complex. The operator ∆ is the analog of the exterior derivative. The antibracket
is the obstruction to ∆ for it to be a derivation. Interestingly, the quantum master
equation
1
(W, W ) − ih̄∆W = 0 (10.7)
2
has the same form as the equation of motion for a Chern-Simons theory [273]. The
analog of a gauge transformation in Chern-Simons theory is a quantum BRST trans-
formation [186] in the antibracket formalism. It is not always too easy to obtain
solutions W to the quantum master equation. In some sense, finding an appropriate
W is equivalent to obtaining the correct measure, i.e., specifying µ.

10.3 Locality
An important but technical aspect of quantum field theories is locality. Here,
we study this issue in the antibracket formalism [154, 157, 206, 132, 11, 257]. In
going from the classical action S0 to the proper solution S and to the quantum action
W , lagrangian terms are added. In a theory defined by a local classical action, the
question is whether these terms are also local. Local interactions involve fields and
derivatives, up to a finite order, of fields multiplied at the same space-time point.
Nonlocal terms are likely to lead to difficulties such as non-renormalizability, non-
unitarity or violations of causality.
The discussion of the gauge structure algebra in Sect. 2 used extensively the
consequences of the regularity condition in Eq.(2.11). An examination of the proof of
Eq.(2.12) reveals that certain operators need to be inverted so that nonlocal effects are
possible. Indeed, it is easy to find a λi so that the solution to S0,i λi = 0 in Eq.(2.12)
involves a nonlocal operator T ji or a function λ′α0 that does not fall off fast at large
space-time distances. Nonlocality often occurs when the quantity of interest vanishes
because it is an integral of a total derivative. As an example, consider n free quantum
R
mechanical particles governed by the action S0 = 12 dt q̇ i q̇i . Note that S0,i = −q̈i .
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 138

R
Let λi = q̇ i . Then, S0,i λi = − 12 dt dtd (q̇ i q̇i ) → 0. The solution in Eq.(2.12) is λ′α0 = 0
and ( δji
ji ′ 2
, for t > t′ ,
T (t , t) = ji
− δ2 , for t < t′ ,
R
since λi (t) = dt′ S0,j (t′ ) T ji (t′ , t). The trivial gauge transformation governed by
T ji (t′ , t) obeys the correct symmetry property T ij (t, t′ ) = −T ji (t′ , t).
Hence, an important concept is local completeness [154]. Local completeness
holds, when solutions to equations, such as in Eqs.(2.12), (2.17) and (2.19), can
be satisfied for local functionals or more precisely, in non-integrated versions. The
difficulty is that sometimes these equations are valid due to total derivatives.
In principle, it is possible that the gauge structure tensors involve nonlocal opera-
tors. This issue has been analyzed in refs.[92, 154, 132]. Given the locality of S0 and
that the gauge generators Rαi are local operators, then the proper solution S of the
classical master equation is local. Reference [154] used cohomological arguments to
obtain this result. The gauge-fixed classical action SΨ is then guaranteed to be local
if the gauge-fixing fermion Ψ is. Under these conditions, the classical BRST opera-
tors δB and δBΨ produce local variations. The question of quantum locality is more
involved. Since this must be analyzed on a case by case basis, no general statements
about the locality of W can be made.

10.4 Cohomological Aspects

In this section, we introduce the concept of cohomology. Cohomological methods
have been used to obtain certain general results [151, 103, 105, 159, 13, 11, 257, 253],
for the field-antifield formalism. For example, the existence proof of the proper
solution is based on these methods [105]. This section is intended to assist the
reader in understanding such research. Because these methods have been reviewed in
refs.[152, 157, 253], our discussion is brief.
Consider a series of spaces Fk . The integer k labels different grading levels. If
αk ∈ Fk then αk is said to have grading k. Let δ be a nilpotent map from one space
to a successive space
δ δ
. . . −→ Fk−1 −→ Fk −→ Fk+1 −→ . . . . (10.8)

Nilpotency means that

δ 2 = δδ = 0 . (10.9)
One can think of δ as carrying a grading of 1. An element αk in Fk is said to be
closed if δαk = 0. The kernel of δ for the kth space, Kerk δ, consists of the set of
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 139

closed elements of Fk , i.e.,

Kerk δ = {αk | αk ∈ Fk , δαk = 0} . (10.10)

An element α is said to be exact if it is expressible as α = δβ. The image of δ in the

kth space, Imk δ, are the exact elements of Fk , i.e.,

Imk δ = {αk | αk = δβk−1 , for some βk−1 ∈ Fk−1 } . (10.11)

Due to the nilpotency of δ, an exact element is automatically closed, so that Imk δ ⊂

Kerk δ. Consider the equivalence relation ∼k that identifies two elements of Fk if their
difference is exact:
∼ ′ ′
αk k αk if there is a βk−1 ∈ Fk−1 such that αk − αk = δβk−1 . (10.12)
∼
Define Hk (δ) to be the set of elements of Kerk δ modulo the equivalence relation k
:

Hk (δ) = Kerk δ/Imk δ . (10.13)

The cohomology Hk (δ) is equivalent to the elements in Fk that are closed but not
exact.
A standard example is the de Rham cohomology on an n-dimensional manifold M.
The spaces Fk consist of the differential forms of order k on M and δ is the exterior
derivative d. The dimension of Hk (d) is the kth Betti number for M. In this example,
more structure can be defined. Differential forms can be multiplied using the wedge
product ∧. One can add differential forms so that the formal sum of the Fk spaces
constitutes an algebra. The exterior derivative respects addition: d(α + β) = dα + dβ,
and it is a graded derivation from the left of the wedge product: d (αj ∧ βk ) = dαj ∧
βk + (−1)j αj ∧ dβk . In applications within the antibracket formalism δ has these
properties, except, with our conventions, δ is a graded derivation from the right:
δ (αβ) = αδβ + (−1)ǫβ (δα) β, where multiplication is denoted by juxtaposition of
elements.
Cohomological methods can be powerful. However, they often involve subtle is-
sues so that one must proceed with strict rigor. The question of whether a closed
element is expressible as δ( of something ) often involves global issues; usually, it can
be done “locally”. Hence, if one is not careful, one can miscalculate the cohomology.
In regard to the antibracket formalism, the pitfalls are more severe: The spaces Fk are
almost always infinite dimensional, and, in quantizing the system, the multiplication
operation becomes singular. Furthermore, an ambiguity concerning the issue of lo-
cality in Sect. 10.3 enters: one needs to decide whether local or non-local functionals
are permitted.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 140

One cohomology in the antibracket formalism uses the classical BRST operator δB
for δ. The spaces Fk consist of smooth functionals of fields with ghost number k. Note
that k ranges over all integers, both positive, negative and zero. Functionals form an
algebra since they can be added and multiplied. Furthermore, δB satisfies the correct
properties: it is nilpotent and is a graded derivation from the right. The cohomology
Hk (δB ) is the classical space of observables in the sector with ghost number k.
For proving certain results, two other cohomologies are useful. The first uses the
Koszul-Tate differential δkt [179, 58, 240]. Let G+ be the condition of setting all ghost
fields to zero: n h i o
G+ ≡ ΦA = 0 | gh ΦA ≥ 1 . (10.14)
Then, the Koszul-Tate differential is defined by

δkt X ≡ δB X|G+ , (10.15)

where δB X = (X, S). When acting on fields (and ghosts), δkt produces zero

A ∂l S 
δkt Φ =  =0 , (10.16)
∂Φ∗A G+

since δkt ΦA , having ghost number greater than one, must be proportional to ghost
fields. Hence, the interesting action is on antifields:

ǫA +1 ∂r S 
δkt Φ∗A = (−1)  . (10.17)
∂ΦA G+
2 ∗
To check the nilpotency of δkt , it suffices to compute δkt ΦA . One finds
! !!
∂r ∂r S ∂r S 
ǫA +ǫB 
2
δkt Φ∗A = (−1)  ,
∂Φ∗B ∂ΦA ∂ΦB 
G+

which can be manipulated to give

! !
∂r ∂r S ∂r S ∂r S ∂r ∂r S 
2 ∗ ǫB 
δkt ΦA = (−1) −  ,
∂ΦA ∂Φ∗B ∂ΦB ∂Φ∗B ∂ΦA ∂ΦB 
G+

which, using Eqs.(4.16), (A.4) and (10.16), produces

!
1 ∂r (S, S) 

=−  =0 .
2 ∂ΦA 
G+

The Koszul-Tate differential is zero because S satisfies the classical master equation.
Because δkt is constructed using the BRST operator it is a graded derivation, i.e.,

δkt (XY ) = Xδkt (Y ) + (−1)ǫY δkt (X) Y . (10.18)

J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 141

The action of δkt for the antifields with ghosts numbers −1, −2 and −3 is respec-
tively
δkt φ∗i = (−1)ǫi +1 S0,i ,
δkt Cα∗ = (−1)ǫα φ∗i Rαi ,
 
∗ ǫα1 +1 ∗ α 1 ǫi ∗ ∗ ji
δkt C1α1 = (−1) Cα R1α1 + (−1) φi φj V1α1 , (10.19)
2
where the tensors R and V are given in Sect. 2.
An important result is that Hk (δkt ) = ∅ for k ≤ −1, where ∅ denotes the empty
set [105, 152]. Let us formally verify this for k = −1 and k = −2. One can set terms
involving ghosts to zero because they either transform to zero or are set to zero. The
most general element α−1 , with ghost number −1 and constructed from antifields, is

α−1 = φ∗i λi ,

where λi are functionals of the φ, with ǫ (λi ) = ǫi so that ǫ (α−1 ) = 1. Since δkt α−1 =
−S0,i λi , α−1 is closed if S0,i λi = 0. Accordingly, λi must be expressible as in Eq.(2.12),
so that
′
α−1 = φ∗i Rαi λ α + φ∗i S0,j T ji .
This is the most general form for a closed element with ghost number −1. The key
?
question is whether α−1 is exact, i.e., α−1 = δkt β−2 . Let
1
β−2 = Cα∗ λ α + (−1)ǫi +1 φ∗i φ∗j T ji .
′

2
Then, a short computation reveals that δkt β−2 = α−1 . In the k = −1 sector, there are
no closed elements that are not exact, so that the cohomology is trivial H−1 (δkt ) = ∅.
In the k = −2 sector, the most general element is
1
α−2 = Cα∗ λα + (−1)ǫi φ∗i φ∗j M0ji ,
2
 
where ǫ (λα ) = ǫα and ǫ M0ji = ǫi + ǫj (mode 2), so that ǫ (α−2 ) = 0. The functional
M0ji obeys M0ji = (−1)ǫi ǫj +1 M0ij . A computation reveals that
 
δkt α−2 = φ∗i Rαi λα − S0,j M0ji ,

so that
Rαi λα − S0,j M0ji = 0 , (10.20)
?
if α−2 is to be closed. Is α−2 = δkt β−3 . It takes a little work to show that α−2 is exact.
α
For α−2 to be closed, λ must obey Eq.(2.12). Substituting the solution of Eq.(2.12)
for λα into Eq.(10.20) gives
 
λ α1 + S0,j T0jα − S0,j M0ji = 0 ,
′
Rαi R1α
α
1
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 142

or, using Eq.(2.19),

 
ji
λ α1 + (−1)ǫj (ǫi +ǫα) Rαi T0jα − M0ji = 0 .
′
S0,j V1α 1

The general solution of this equation is

ji
λ α1 + (−1)ǫj (ǫi+ǫα ) Rαi T0jα − (−1)ǫi ǫα Rαj T0iα − M0ji = S0,k N kji
′
V1α 1
,
 
where N kji must be graded antisymmetric in all indices and ǫ N kji = ǫi + ǫj + ǫk
(mod 2). In terms of the above functionals and tensors, let
1
∗
λ α1 − (−1)ǫα Cα∗ φ∗j T0jα + (−1)ǫj φ∗i φ∗j φ∗k N kji
′
β−3 = −C1α .
1
6
Then, a short computation shows that α−2 = δkt β−3 . Since any closed k = −2 element
is expressible as an exact form, H−2 (δkt ) = ∅.
Notice that the triviality of the Koszul-Tate cohomology H−k (δkt ) = ∅ for k > 0
reflects the consequences of the regularity condition used in Sect. 2.14 Although
the above discussion has been formal, a more rigorous analysis can be given. See
refs.[105, 152].
Insight into the physical significance of the Koszul-Tate cohomology is gained by
computing H0 (δkt ). A general closed element of the zero ghost-number sector is a
functional α0 of the fields φ since δkt α0 (φ) is automatically zero due to δkt φi = 0. Let
β−1 be a general element of the −1 ghost sector, i.e, β−1 = −φ∗i λi , where we include
a minus sign for convenience, and where the Grassmann nature of λi is ǫ (λi ) = ǫi .
Apply the Koszul-Tate differential to β−1 to obtain δkt β−1 = S0,i λi . One concludes
that if α0 = S0,i λi then α0 is exact. Therefore α0 and α0 ′ are not related under
the equivalence relation ∼0 if α0 and α0 ′ differ on the stationary surface Σ where the
equations of motion S0,i = 0 hold. Consequently, H0 (δkt ) corresponds to the set of
distinct functions on Σ. More precisely,
n ′ ′
o
H0 (δkt ) = α0 (φ) | α0 ∼ α0 , if α0 − α0 = S0,i λi for some λi . (10.21)

Suppose that a theory has no gauge invariances. Then the classical observables cor-
respond to functionals taking on distinct values on Σ. This space is H0 (δkt ).
If a theory has gauge invariances, then the observables should be the gauge-
invariant elements of H0 (δkt ). To facilitate the issue of gauge invariance, one in-
troduces the vertical differential δg [106, 152]. An alternative name for δg is the
“exterior derivative along the gauge orbit”. It is defined as

δg X = (δB X)|G− = (X, S)|G− , (10.22)

14
If one does not use a proper solution for S, H−k (δkt ) can be non-empty for k > 0.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 143

where G− corresponds to the condition of setting antifields to zero and going on-shell
with respect to the original fields, i.e.,
( )
∂r S
G− = Φ∗A = 0, i = 0 . (10.23)
∂φ

Because δg is defined in terms of δB , it is a derivation from the right:

δg (XY ) = Xδg Y + (−1)ǫy (δg X) Y . (10.24)

Antifields can be ignored in evaluating the vertical differential because they are either
set to zero or transformed to zero since
!
∂l S 

δg Φ∗A =−  =0 , (10.25)
∂ΦA 
G−

which follows from ghost number considerations or Eq.(10.23). On fields, one has
!
∂l S 
A 
δg Φ =  . (10.26)
∂Φ∗A 
G−

To check the nilpotency of δg , one only needs to check that δg δg ΦA = 0. A straight-

forward calculation gives
!
∂r ∂l S ∂l S 

δg2 ΦA = 
∂ΦB ∂Φ∗A ∂Φ∗B 
G−
! !
∂l ∂r S ∂l S ∂r S ∂l ∂l S 
(ǫA +1)ǫB 
= − (−1) 
∂Φ∗A ∂ΦB ∂Φ∗B ∂ΦB ∂Φ∗A ∂Φ∗B 
G−
!
1 ∂l (S, S) 
=  =0 ,
2 ∂Φ∗ A

G−

where Eqs.(4.16) and (10.25) have been used. As a consequence of nilpotency, a

cohomology with respect to δg can be defined.
The physical relevance of δg can be seen by computing H0 (δg ). The action of δg
on the original fields φi is
δg φi = Rαi C α . (10.27)
Without loss of generality, a functional α0 with ghost number 0 can be taken to be a
functional of the φi only. Such a functional is closed if δg α0 (φ) = 0. Using Eq.(10.27),
one finds
δg α0 = α0,i Rαi C α = 0 ⇒ α0 is gauge invariant .
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 144

Since any functional β−1 with ghost number −1 is annihilated by δg , a closed α0

cannot be exact: α0 6= δg β−1 . The conclusion is that

H0 (δg ) = the set of gauge–invariant functionals . (10.28)

With the above insights, one realizes that observables should roughly correspond
to H0 (δkt ) ∩ H0 (δg ). However, a difficulty arises. The intersection H0 (δkt ) ∩ H0 (δg )
does not make sense unless δkt δg + δg δkt = 0. The situation is analogous to the one
in quantum mechanics where one seeks a state that is simultaneously the eigenvector
of two different operators. Such a state is possible if the two operators commute.
Because of the graded nature of δkt and δg , the analog condition is that δkt and δg
anticommute. The difficulty can be posed as a question: Should one take the gauge-
invariant elements of H0 (δkt ) or should one take the elements of H0 (δg ) modulo
the equivalence relation of Eq.(10.21)? If δkt δg + δg δkt = 0, then the above two
procedures yield the same result. In such a case, one can define a nilpotent BRST
operator δB ≡ δkt + δg , and the observables correspond to the BRST cohomology.
Unfortunately, δkt δg +δg δkt 6= 0 in general. An inspection of δB , δkt and δg reveals that
δB = δkt + δg + extra terms. The extra terms render δB nilpotent, by compensating
for the failure of the anticommutivity of δkt and δg . The BRST operator is the
natural extension of δkt + δg . The elements of the cohomology of δB are the classical
observables [105, 152]. They are the definition of what one means by the “gauge-
invariant functionals on Σ”.
When quantum effects are incorporated, the quantum BRST transformation δB̂
is relevant. As discussed in Sect. 8.1, the quantum observables correspond to the
elements of the cohomology of δB̂ .
Because canonical transformations preserve the antibracket, the cohomology of
δB is independent of the basis, as can be seen as follows. Given a proper solution
S[Φ, Φ∗ ] in one (untilde) basis, a solution S̃[Φ̃, Φ̃∗ ] in another (tilde) basis is given
by S[e Φ̃, Φ̃∗ ] ≡ S[Φ, Φ∗ ]. Likewise, given any functional X[Φ, Φ∗ ], one can define a
functional X f of tilde fields using X[
f Φ̃, Φ̃∗ ] ≡ X[Φ, Φ∗ ]. The tilde antibracket of Xf
and S, e as a function of tilde fields and antifields, equals (X, S) as a function of
untilde fields. Hence, X f is closed if and only if X is, and Xf is exact if and only if X
is. Consequently, there is an exact isomorphism of the cohomologies.
Since the gauge-fixed BRST transformation δBΨ is not nilpotent, one cannot di-
rectly define a cohomology associated with δBΨ . However, according to Eq.(6.83),
δB2 Ψ is proportional to the equations of motion of the gauge-fixed action SΨ . Define
an equivalence relation, denoted by ≈, that equates two quantities if they differ by
terms proportional to the equations of motion for SΨ . Then, one has δB2 Ψ ≈ 0 and a
gauge-fixed BRST cohomology can be defined. What is the relation between Hn (δBΨ )
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 145

and Hn (δB )? The connection is best seen by going to the gauge-fixed basis for δB .
Let
Y (Φ̃, Φ̃∗ ) = y(Φ̃) + y A (Φ̃)Φ̃∗A + . . . (10.29)
be the antifield expansion of a functional Y in the gauge-fixed basis Φ̃A and Φ̃∗A of
Sect. 6.6. Eqs.(6.79) and (6.84) imply

∂l SΨ
δB Y = δBΨ y − y A + O(Φ̃∗ ) , (10.30)
∂ Φ̃A

so that
(δB Y )|{Φ̃∗ =0} ≈ δBΨ y , (10.31)
since the last term in Eq.(10.30) is proportional to gauge-fixed equations of motion.
Eq.(10.31) implies that if Y is δB -closed then y is δBΨ -closed, and that if Y is δB -exact
then y is δBΨ -exact. This is not enough to establish any relation between Hn (δBΨ ) and
Hn (δB ). Given a element of y of Hn (δBΨ ), one must uniquely construct an element
Y of Hn (δBΨ ), under the condition that Y |{Φ̃∗ =0} = y. In other words, one must find
the higher-order terms in Eq.(10.29). References [151, 105, 103] succeeded in doing
this. For additional discussion, see refs.[157, 253]. The cohomologies governed by δBΨ
and δB are equivalent.

10.5 Equivalence with the Hamiltonian BFV Formalism

Gauge theories can also be analyzed using a hamiltonian formalism. For the
generic theory, the Batalin-Fradkin-Vilkovisky (BFV) approach [109, 23, 108, 18, 19]
is quite useful. For the simplest theories, such as particle models, Yang-Mills theory,
and gravitation, it is not difficult to show that it yields results equivalent to the field-
antifield formulation. Demonstrating the equivalence in general, at the classical level
or formally at the quantum level without regularization, has been the subject of the
work in refs.[31, 104, 230, 231, 91, 140, 141, 201, 206, 75, 76].
A review of the BFV hamiltonian formalism is given in ref.[150]. Here, we present
only the key ideas. Let S0 be the classical action as determined from a lagrangian
R
L by S0 = dt L. The hamiltonian HS0 associated with S0 is constructed in the
standard manner using15
HS0 [φ, π] ≡ φ̇iπi − L , (10.32)
15
In this subsection, we use the convention that a field index also represents a spatial position.
An index appearing twice represents a sum not only that index but also an integration over space.
This is the hamiltonian analog of the compact notation described in Sect. 2.1. The difference, here,
is that time is not included as part of the integration.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 146

where a dot over a field indicates a time derivative, where the conjugate momentum
of φi is πi ≡ ∂∂lφ̇Si0 , i = 1, . . . , n, and where HS0 is obtained as a function of the φ and π
by solving for φ̇ in terms of the π and φ. For some systems, this velocity-momentum
inversion process is not possible due to the presence of primary constraints. Even
in this case, a hamiltonian HS0 can be uniquely constructed on the surface of these
primary constraints. We symbolically represent the procedure of obtaining a hamil-
tonian from an action diagrammatically as
S0

y
H S0
For a wide class of gauge theories, HS0 is of the form

HS0 = H0 [ϕ, π] + λα Tα [ϕ, π] , (10.33)

where the original n variables φi are split into dynamical degrees of freedom ϕa , a =
1, . . . , m ≤ n and Lagrange multipliers λα for the (secondary) constraints Tα . In
Eq.(10.33), H0 and Tα are functions of the ϕ and their momenta only. The velocities
λ̇α are usually assumed not to appear in S0 . This means that the momenta of the
λα , namely πα , are primary constraints and do not enter in HS0 . For example, in
a Yang-Mills theory, the hamiltonian density H0 is H0 = 21 Eai Eia + 41 Fija Faij , where
Eai = −Fa0i = Fa0i are the canonical momenta for the potentials Aai , the constraints
Tα correspond to Gauss’s law: Ta = −Dia b Ebi , and the Lagrange multipliers λα are
Aa0 .
For simplicity, assume that the constraints Tα and the hamiltonian H0 are first
class, i.e,
γ
{Tα , Tβ }P B = Tαβ Tγ , {H0 , Tα }P B = Vαβ Tβ , (10.34)
where { , }P B denotes the graded Poisson bracket defined by
∂r F ∂l G ∂r F ∂l G
{F, G}P B = (−1)ǫi − . (10.35)
∂φi ∂πi ∂πi ∂φi
Here, the sum over i is such that all fields and momenta are included. If the constraints
are second class, Dirac brackets [88] must be used. Note that {πi , φj }P B = −δij .
The BFV program is based on BRST invariance. One introduces ghosts and their
conjugate momenta. The ghosts needed correspond to the minimal set, introduced
in Eq.(4.1). For the irreducible case, they are the C α . We use P̄a to denote the
momentum associated with a ghost C a , where a is a label that enumerate all ghosts.
The Poisson bracket in Eq.(10.35) is then extended to include a sum over ghosts.
With these conventions, {P̄b , C a }P B = −δba . The ghost numbers and statistics of the
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 147

h i
BFV ghosts are the same as in Sect. 4.1. For momenta, gh P̄a = −gh [C a ] and
 
ǫ P̄a = ǫ (C a ). A canonical generator of the BRST transformations QB and an
extended hamiltonian H are constructed, using the requirement that QB be nilpotent
and that H be BRST invariant:

{QB , QB }P B = 0 , {H, QB }P B = 0 . (10.36)

They can be expanded as a power series in ghost fields, for which the first few terms
are
1
H = H0 + C α Vαβ P̄β + . . . , QB = C α Tα − (−1)ǫβ C β C γ Tγβ
α
P̄α + . . . . (10.37)
2
It turns out that the requirement of {QB , QB }P B = 0 reproduces the relations defining
the structure of the gauge algebra at hamiltonian level. In other words, QB plays a
role analogous to the proper solution S of the antibracket formalism.
In this approach, O is an observable if it is BRST-invariant, i.e., {O, QB }P B = 0.
Thus, the hamiltonian is an observable. Two observables O1 and O2 are considered
′ ′
equivalent if O2 = O1 + {O , QB }P B , for some O . A state | ψ i is called physical if
QB | ψ i = 0. Two states | ψ1 i and | ψ2 i are considered equivalent if | ψ2 i = | ψ1 i +
QB | ψ ′ i, for some | ψ ′ i.
Given a suitable hamiltonian H, a lagrangian can be constructed via
  Z  Z 
i [dΠ] i i
exp SH [Φ, Φ̇] = exp dt (Φ̇ Πi − H[Φ, Π]) , (10.38)
h̄ 2πih̄ h̄
where Φ denotes all degrees of freedom and Π denotes the corresponding momenta.
We indicate the process of constructing an action SH from a hamiltonian H by the
following diagram
SxH


H
In the BFV formalism, to obtain a hamiltonian HΨ , which is appropriate for
insertion in the functional integral, a fermion Ψ with ghost number minus one is
used. As in the antibracket formalism, BRST trivial pairs exist. Given two fields Λ
R
and Σ, and their conjugate momenta, P̄Λ and P̄Σ , a term dd−1 x Σ P̄Λ can be added
to QB without ruining nilpotency. The next step in the BFV program is to introduce
additional fields and their momenta and add them as trivial pairs to QB . These
fields are the analogs of the auxiliary gauge-fixing fields of Sect. 6.2. They include
antighosts, extraghosts, and the Lagrange-multiplier fields of Eq.(6.22). The fermion
Ψ in the hamiltonian formulation must satisfy conditions similar to those in Sect. 6.3
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 148

for the Ψ in the antibracket formalism. We denote the BRST charge extended by the
inclusion of the additional trivial terms by Qnm
B . The hamiltonian HΨ is given by

HΨ = H − {Ψ, Qnm
B }P B . (10.39)
R  
Let ZΨ = [dΦ] exp h̄i SHΨ , where SHΨ is constructed from HΨ via Eq.(10.38). The
Fradkin-Vilkovisky theorem [109] states that ZΨ is independent of Ψ.
The equivalence of the BFV hamiltonian and antibracket methods is established
if the remaining leg of the following diagram
Sx0 −→ SΨ

y
H S0 −→ HΨ
can be completed. In other words, is SHΨ , as constructed from HΨ via Eq.(10.38),
equivalent to SΨ as obtained from the antibracket formalism? Likewise, is HSΨ , as
constructed from SΨ via Eq.(10.32), equivalent to the BFV hamiltonian HΨ ? Another
question is whether the gauge-fixed BRST charge QNoether , as constructed from SΨ
using Noether’s theorem, coincides with the BRST charge QnmB for the BFV formalism.
The affirmative answer to the above questions, obtained in refs.[104, 91], implies that
construction processes in
Sx0 −→ SxΨ
 
y y
HS0 −→ HΨ
commute to give equivalent results.
One can also ask whether an equivalence occurs before the introduction of gauge-
fixing and Ψ:
S


y?
H
Clearly, a straightforward correspondence cannot exist because S contains antifields.
However, at least for closed irreducible theories, if certain antifields are set to zero and
others are identified with ghost momenta, then an equivalence of HS , as constructed
from S via Eq.(10.32), and the BFV H is achieved [31]. Similar results have been
obtained in refs.[230, 91].
If sources for the BRST transformations are included at the hamiltonian level, the
above correspondence can be made clearer. Then, the sources in the hamiltonian for-
mulation can be identified with antifields in the antibracket formalism. This method
was used in refs.[140, 75, 76] to establish the equivalence in the gauge-fixed basis.
An open problem is to extend all of the above analysis to the quantum case in a
rigorous manner. That situation is more difficult due to operator ordering problems
and the singular character of field theories.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 149

10.6 Unitarity
The difficulty in proving unitarity in covariant approaches to quantizing gauge
theories is due to the presence of ghosts and of unphysical degrees of freedom with
negative norms. One often deals with indefinite-metric Hilbert spaces. Unitarity
can be spoiled in theories with kinetic energy terms of the wrong sign and/or non-
hermitian interaction terms. Wrong sign kinetic energy terms almost always arise in
gauge theories with particles of spin one or higher. Due to the sign of the metric
component η00 , there are potential difficulties with the temporal components, such as
A0 in electromagnetism, Aa0 in Yang-Mills theories, and g0i in gravity. Faddeev-Popov
and other gauge-fixing ghosts enter in loops with the wrong sign, and would lead to
a violation of unitarity, if their contributions were considered in isolation.
Let us summarize how unitarity is established in certain covariant quantization
procedures. First of all, one needs to assume that there are not any non-hermitian
interactions in the original theory and that the spatial components of tensors have the
correct sign in kinetic energy terms. In other words, the theory should be “naively”
unitary.
The first approach is as follows. In some theories, there exists a unitary gauge,
in which it is evident that the unphysical excitations are not present. If one can
establish the gauge invariance of the S-matrix, then unitarity can be proven by going
from a covariant gauge to an unitary one [100]. Unfortunately, this method is only
well developed for irreducible theories with closed algebras. For reducible systems,
this approach often encounter difficulties, although for some specific examples it has
been successfully implemented [116].
Another method for checking unitarity is in perturbation theory via Feynman
diagrams [243, 245]. Using the Ward-Takahashi [265, 239] or Slavnov-Taylor iden-
tities [241, 233], as well as the Landau-Cutkosky rules [95], one tries to show that
contributions from the unphysical polarizations of the classical fields are cancelled by
contributions from ghost fields or from other sources.
A third approach proceeds via canonical quantization. The “physical sector”
is selected out by imposing some subsidiary conditions that remove negative norm
states. The physical sector should be stable under time evolution and should involve
a non-negative metric. A well-known example of this approach is the Gupta-Bleuler
procedure [53, 144] for quantizing QED. All components of the electromagnetic field
Aµ are used; however only states | ϕ iphys satisfying

(∂µ Aµ )+ | ϕ iphys = 0 (10.40)

are considered, where (∂µ Aµ )+ denotes the positive frequency components of ∂µ Aµ .

This condition determines the physical sector Hphys in the Gupta-Bleuler procedure.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 150

Unfortunately, when applied to non-abelian Yang-Mills theories, this method fails

to preserve Hphys under time evolution. To quantize covariantly non-abelian gauge
theories, refs.[70, 181, 182] proposed

QB | ϕ iphys = 0 , (10.41)

where QB is the hermitian nilpotent BRST operator. Eq.(10.41) is the basis for BRST
quantization. We use Vphys to denote the space of states annihilated by QB . In the
BRST approach, the hamiltonian is automatically hermitian so that the S-matrix is a
unitarity operator in Vphys . However, there is a possible difficulty with Vphys . Despite
the fact that QB commutes with the hamiltonian, the positive semidefiniteness of
the norm of Vphys is not ensured. The question of unitarity in BRST quantization
becomes that of proving the positive semidefiniteness of Vphys , and must be analyzed
model by model.
However, T. Kugo and I. Ojima [181, 182] (see also [196]), obtained criteria under
which unitarity does hold. They established a connection with the metric structure
of Vphys and the multiplets of the algebra generated by the conserved BRST charge
QB and the conserved ghost number charge QC .16 With our conventions, QC is
antihermitian: Q†C = −QC . These generators satisfy
1
[QC , QB ] = QB , [QC , QC ] = 0 , {QB , QB } = Q2B = 0 . (10.42)
2
Three types of multiplets are possible:
(a) “True physical states”: BRST singlets with zero ghost number.
(b) Doublets: pairs of BRST singlets related by ghost conjugation.
(c) Quartets: pairs of BRST doublets related by ghost conjugation.
Roughly speaking, ghost conjugation is the operation that interchanges ghosts and
antighosts. Under this operation, the sign of the ghost number of a state is flipped.
In the next three paragraphs, we explain the classification of the multiplets.
One can choose states to be eigenfunctions of Qc . Let | g i be a state with a non-
zero ghost number g. Then | g i has zero norm since h g |Qc | g i = gh g | g i = −gh g | g i,
the first equality arising when Qc acts to the right, and the second equality arising
when Qc acts to the left. Non-zero matrix elements occur only when bra and ket
states have opposite ghost numbers. Under application of QB , the ghost number of
16
We assume there are no anomalies associated with QB and QC . For the first-quantized string,
this is actually not the case for QC , but BRST quantization is still possible [163, 174, 115]. For
similar analyses in other models see refs.[59, 42, 48, 60].
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 151

a state is increased by one. Such states | s i = QB | s′ i also have null norms since
h s | s i = h s′ |QB QB | s′ i = 0.
Due to the nilpotency of QB , the representations are either BRST singlets or
BRST doublets. A BRST singlet | s i satisfies QB | s i = 0 and | s i = 6 QB | s′ i for any
′ ′ ′
| s i. If QB | s i = | s i = 6 0, then | s i is a member of the BRST doublet consisting
′
of | s i and | s i. The upper member of a doublet | s i is annihilated by QB , since
QB | s i = QB QB | s′ i = 0, and it carries one unit of ghost number more than | s′ i:
Qc | s i = Qc | s′ i + 1.
If | s i is a BRST singlet and carries ghost number zero, then it is of type (a). If
| s i is a BRST singlet and carries non-zero ghost number g, then it is of type (b).
Under ghost conjugation, another BRST single with ghost number −g is created, thus
forming the pair. If | s i and | s′ i constitute a BRST doublet, then ghost conjugation
produces another BRST doublet and a type (c) multiplet is obtained.
For an irreducible gauge theory, T. Kugo and I. Ojima in [181, 182] proved that (i)
if type (a) states have positive definite norm and (ii) if type (b) states are absent, then
quartets only appear in Vphys through zero norm combinations. Consequently, when
(i) and (ii) are satisfied, Vphys has a positive semidefinite norm. To obtain a unitary
theory, one mods out the null-norm states: Two states are identified if they differ by
a null-norm vector. Clearly, null-norm states are identified with the null state. The
modding-out procedure automatically restricts states to the zero-ghost number sector,
since states with non-zero-ghost number have zero norms. Furthermore, because
BRST-trivial states QB | s′ i are null-norm vectors, all that remains after modding out
are the non-trivial elements of the g = 0 BRST cohomology, i.e., states with ghost
number zero that are annihilated by QB and that cannot be expressed as QB | s′ i for
any state | s′ i. This sector is preserved under time evolution because QB and QC
commute with the hamiltonian.
In the g = 0 sector, it makes sense to identify null-norm states with the null
vector because they decouple from matrix elements involving observables, such as the
hamiltonian. Observables O are BRST-invariant operators: [O, QB ] = 0. If | t i is a
BRST-trivial state, so that | t i = QB | s′ i, and if | s i is any element of Vphys , so that
QB | s i = 0, then h s |O| t i = h s |OQB | s′ i = h s |QB O| s′ i = 0.
For reducible systems, ghosts for ghosts and extraghosts arise, some of which
have zero ghost number. Hence a third condition arises for reducible theories: (iii)
a state of Vphys involving ghosts in the g = 0 sector must be a member of a quartet
multiplet. This guarantees that they are null vectors and do not ruin the positive
semidefiniteness of Vphys .
The above conditions provide criteria for establishing the positivity of the norm
and hence unitarity in a covariant formulation. Reference [70, 181, 182] established
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 152

unitarity for Yang-Mills theories by proving (i) and (ii) for this case.
In perturbation theory and in a Fock space representation, A. Slavnov in [234]
used (i)–(iii) to obtain simpler criteria. The most important requirements, apart
from the positivity of the norm of type (a) states, were that QB be nilpotent and that
it have nontrivial action on all ghost fields or their conjugate momenta. Under these
conditions, Vphys has a positive semidefinite norm. Then, S. A. Frolov and A. Slavnov
[117] using the hamiltonian BFV-BRST formalism for lagrangians L of the form in
Eqs.(10.32) and (10.33), verified the above-mentioned conditions perturbatively. The
(0)
analysis was simplified because one could use the free BRST charge QB . The criteria
(0) (0) (0)
became that QB QB = 0 and that QB have non-trivial action on all ghosts. Given
the validity of perturbation theory, their result on the unitarity of a gauge theory
holds for the finite reducible case.
S. A. Frolov and A. Slavnov in ref.[118, 235], were able to translate the above
program into a lagrangian approach, by using an effective action Aef f . The action
(0)
and BRST charge were perturbatively expanded in a series: Aef f = Aef f + . . . and
(0) (0)
QB = QB + . . .. The term Aef f was the leading order part of the general gaussian
gauge-fixed action SΨ of the field-antifield formalism presented in Sect. 6.4. Requiring
nilpotency and BRST invariance of the action lead to a series of recursion relations
for the higher order terms in Aef f and QB . The action Aef f , thus obtained, is
constructed using unitarity requirements. Finally, when certain conditions on the
rank of the gauge generators are imposed, the free BRST charge is seen to act non-
trivially on ghosts fields and unphysical polarizations of the classical fields, thereby
yielding a unitary theory if the classical gauge-invariant degrees of freedom have a
positive norm.
The problem of unitarity in the field-antifield formalism was addressed in [130,
206, 203, 204]. A perturbative solution of the proper solution S was obtained in
[130, 206, 132] (see also ref.[13]). Then, a general gaussian gauge-fixing procedure
was performed, using a fermion Ψ of the type given in Sects. 6.3 and 6.4. It was
shown that BRST invariance of the gauge-fixed action and nilpotency of the gauge-
fixed BRST transformation lead to the same recursion relations obtained in [118, 235],
and that the leading two terms of SΨ agree with Aef f . The conclusion is that the field-
antifield formalism produces an action SΨ that coincides with Aef f of ref.[118, 235]
obtained by unitarity considerations.
The above approaches to unitarity are formal in that the difficulties with field-
theoretic infinities are not addressed. The renormalizability or non-renormalizability
is not used. To proceed rigorously, one needs to regulate the theory with a cutoff,
verify unitarity, and then make sure that unitarity remains as the cutoff is removed.
The issue of locality also enters here. For example, it may happen that Aef f or S
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 153

contains non-local terms. This does not necessarily ruin unitarity, but might signal
that the theory is non-renormalizable or ill-defined. Studies of unitarity without
using perturbation theory for general systems with finite degrees of freedom, such as
in quantum mechanics, have been carried out in ref.[194].

10.7 The Antibracket Formalism in General Coordinates

The antibracket formalism in a general coordinate system has been developed
in refs.[273, 175, 21, 176, 218, 219, 149]. A brief overview is given in [157]. This
approach sometimes goes by the name covariant formulation of the field-antifield
formalism. Possible applications are in mathematics [209, 5, 189, 198, 158] and in
string field theory [209, 274, 189, 149, 158, 225, 226].
Consider a supermanifold M of type (N, N), meaning that there are N bosonic
and N fermionic coordinates. Collectively denote these as z a , a = 1, . . . , 2N. In this
coordinate system, a local basis for the cotangent space T ∗ M consists of the 1-forms
dz a , a = 1, . . . , 2N. The Grassmann parity of a differential is the same as that of
the corresponding coordinate: ǫ(dz a ) = ǫ(z a ) = ǫa . Introduce an odd two-form ζ,
ǫ(ζ) = 1, which is non-degenerate and closed, i.e., dζ = 0. In the local basis, ζ is
expressed as
1 1
ζ = − ζab (z)dz b ∧ dz a = dz b ∧ ζba (z)dz a , ζab = (−1)ǫa ǫb +1 ζba , (10.43)
2 2
where ǫ(ζab ) = ǫa + ǫb + 1 (mod 2). Let ζ ab be the inverse of the matrix ζab . It obeys
ζ ab = (−1)ǫa +ǫb +ǫa ǫb ζ ba . One then defines the antibracket via Eq.(4.8) but using
ζ ab (z). Alternatively, let X be a function on M. Then one can define a vector field
← ← ←
∂X
V X = ∂z∂a ζ ab ∂zl b that acts from right to left on functions Y of M: [Y ]V X = ∂r Y
∂z a
ζ ab ∂∂zl Xb .
Then the antibracket can be written as
← ← ←
(X, Y ) = ζ[V X , V Y ] = [X]V Y , (10.44)

where the first equality follows from

∂r ∂r
dz b ∧ dz a [ , ] = δca δbd − (−1)ǫa ǫb δcb δad .
∂z c ∂z d
In this way, {M, ζ} becomes an odd symplectic structure. The antibracket, defined
as above, obeys the properties in Eqs.(4.5) and (4.7). It turns out that dζ = 0 is
necessary for the Jacobi identity in Eq.(4.5).
For ordinary symplectic manifolds, there exists a natural volume element dµ ob-
tained by wedging ζ with itself N times. Unfortunately, for an old sympletic manifold,
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 154

ζ ∧ ζ = 0. Hence, a measure must be introduced by hand:

2N
Y
dµ(z) = ρ(z) dz a , (10.45)
a=1
← ←
where ρ(z) is a density. The divergence of a generic vector field V = ∂z∂a V a is defined
in the usual way by
← 1 ∂l (ρV a )
divρ V ≡ (−1)ǫa . (10.46)
ρ ∂z a
Then, the laplacian ∆ρ acting on a function X is defined by taking the divergence of
the corresponding vector field via
!
1 ← 1 ∂l ∂l X
∆ρ X ≡ (−1)ǫX divρ V X = (−1)ǫX +ǫa a
ρζ ab b . (10.47)
2 2ρ ∂z ∂z
Since one would like to use ∆ρ as the general coordinate version of ∆ of Sect. 6.1,
one wants it to be nilpotent. However, this is not necessarily the case since
" !#
1 1 ∂l   ∂l
∆ρ ∆ρ = ∆ρ (−1)ǫa +ǫb a ρζ ab .
2 ρ ∂z ∂z b
Hence, one requires ρ to satisfy
!
1 ∂l  
∆ρ (−1)ǫa +ǫb a ρζ ab = 0 . (10.48)
ρ ∂z
When Eq.(10.48) holds, ∆ρ is formally nilpotent and a graded derivation of both
functional multiplication and the antibracket, i.e., it satisfies Eq.(6.8) with ∆ → ∆ρ .
It turns out that Eq.(10.48) is the necessary and sufficient condition for the exis-
tence of Darboux coordinates locally. For such coordinates, ρ = 1 and ζ ab takes the
form in Eq.(4.8). Then z a for a = 1, . . . , N can be identified with fields and z a for
a = N + 1, . . . , 2N can be identified with antifields. Hence, we employed the Darboux
coordinate system for the antibracket formalism in Sects. 4 – 8. Darboux coordinates
suffice as long as global issues are not important.
Quantization in a general coordinate system proceeds as in the Darboux case.
Everywhere ∆ appears in Sects. 6 – 8, one replaces it by ∆ρ . The functional-integral
measure also must be modified. Integration is restricted to an N-dimensional subman-
ifold N . Since little distinction is made between fermionic and bosonic coordinates
in the covariant formulation, N can be an arbitrary (k, N − k) submanifold as long
as ζ[V, V ′ ] = 0 on N for any two tangent vectors V, V ′ ∈ T N . One considers a
basis {e1 , . . . , eN ; h1 , . . . , hN } for T M, such that {e1 , . . . , eN } is a basis for T N and
ζ[ei , hj ] = δij . Then the volume element on N is
h i1/2
dµN (e1 , . . . , eN ) = dµ(e1 , . . . , eN ; h1 , . . . , hN ) . (10.49)
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 155

Since integration in function space is restricted to N , the above procedure corresponds

to a gauge-fixing procedure. The submanifold N can be defined using N linearly
independent constraints ΨA (z) = 0 satisfying
C
(ΨA , ΨB ) = TAB (z)ΨC . (10.50)
←
The vectors V ΨA are a basis for the tangent space T N of N . Furthermore, as a
← ←
consequence of Eq.(10.50), ζ[V ΨA , V ΨB ] = 0 on N , which is a consistency check. To
make contact with the gauge-fixing procedure of Sect. 6, one goes to the Darboux
∂Ψ
coordinate system and chooses ΨA = Φ∗A − ∂Φ A . One disadvantage of the general

coordinate approach, is that the concept of ghost number becomes obscure.

10.8 The D=26 Closed Bosonic String Field Theory

A review of the current formulation of the closed bosonic string has been given
in ref.[281]. Here, we present some of the salient points.
At the first-quantized level, closed strings possess holomorphic factorization. This
means that, with the possible exception of zero modes, the integrands of closed-string
amplitudes factorize into two open-string-like integrands, one for left-moving degrees
of freedom and one for right-moving degrees of freedom. At the second-quantized
level, there is a similar splitting. Hence, a closed string field A is a tensor product of
a left string field AL with a right string field AR , so that one can write A as
A = (AL ; AR ) , (10.51)
or as a sum of terms of the form in Eq.(10.51). The field AL (respectively, AR ) is
precisely of the form of the open string case, except a subscript L (respectively, R)
is appended to all quantities. One exception is the zero modes of X µ (σ), namely the
position and momentum operators. They are the same for both left and right sectors,
so that xµL = xµR ≡ xµ and pµL = pµR ≡ pµ . The total string ghost number is the sum
of the left and right string ghost numbers, i.e., g (A) = gL (AL ) + gR (AR ).
One can attempt to construct closed-string field theory along the lines of the open
string case described in Sect. 3.6. It is easy to see that not all the open string axioms
can be extended. When the axioms hold, Paton-Chan factors [208] can be appended
to the string field leading to a non-abelian Yang-Mills gauge group. However, closed
string theories cannot possess such a non-abelian gauge structure [220].
R
Define closed-string integration as the product of left and right open string
R R R closed
integrals via ≡ L R , i.e., for fields in the form in Eq.(10.51), one has
closed
Z Z Z
A= AL AR . (10.52)
L R
closed
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 156

To generate a non-zero integral in Eq.(10.52), A must have a left-ghost number of 3

and a right-ghost number of 3 and consequently a total ghost number of 6:
Z
A=0, if g (A) 6= 6 .
closed

Let σ be the first-quantized variable parametrizing the string. It varies between 0

and 2π and is periodic. To define the closed-string star operation ◦, pick two antipodal
points, e.g. σ = 0 and σ = π. This divides a string into two halves. Then, ◦ is defined
in analogy to the open string case. One half of one string overlaps with one half of
the other string and what remains is the product string. The first-quantized BRST
charge Q is the sum of the right and left parts:

Q = QL + QR , (10.53)

and it carries ghost number one: g (Q) = 1. Eq.(10.53) is equivalent to

QA = (QL AL ; AR ) + (AL ; QR AR ) . (10.54)

R
Even though , ◦ and Q have been defined, there is a difficulty in obtaining a
R
free action. Let C denote the closed-string field. The naive term C ◦ QC vanishes
because of ghost number considerations. The total ghost number of the integrand,
which is 2g (C) + 1, must be equal to 6. This constraint cannot be satisfied because
g (C) is an integer.
To correct the problem, various schemes can be used [9, 192, 168]. Let
1 L 
c̄− L R
0 ≡ c̄0 − c̄0 , c−
0 ≡ c0 − cR
0 . (10.55)
2
Impose the following two constraints on C
 
c̄−
0C = 0 , LL0 − LR
0 C = 0 . (10.56)

Anticipating that then free-theory

o equationhis QC = 0, one
i sees that
h these constraints
i
are consistent since Q, c̄0 = L0 − L0 , Q, L0 − L0 = 0 and c̄−
− L R L R
0 , LL
0 − LR
0 = 0.
 
The condition LL0 − LR L
0 C = 0 is quite natural. The operator L0 − L0 is the
R

generator for rigid rotation of the first-quantized parameter σ, which labels the points
along the string. Since, in the closed-string
 case,
 σ is periodic, there is no preferred
L R
choice of an origin. The condition L0 − L0 C = 0 does not lead to an equation of
motion for C since the 21 ∂µ ∂ µ terms in LL0 and LR 0 cancel. From Eq.(7.34), one sees
that it implies that the mass ML of the left-sector must equal the mass MR of the
right-sector, a well-known constraint of first-quantized closed-string states. However,
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 157

the operation ◦ does not preserve the constraint. One can modify ◦ to ˆ◦ by averaging
over a rigid rotation that rotates the product string over angles ranging from 0 to 2π.
The new ◦ˆ no longer is associative, as can be checked by drawing some pictures.
The ghost number problem can be fixed by inserting a factor of c− 0 . Define the
quadratic form Z
hA, Bi = Aˆ◦ c−
0B . (10.57)
closed

Let the free action be

1
(2)
S0 = hC, QCi . (10.58)
2
It is invariant under the gauge transformations δC = QΛ, where g (Λ) = 1 since
g (C) = 2. There are many ways of resolving the difficulty with ghost number of the
free action of closed-string field theory but they are equivalent to the above. The
tree-level three-point interaction is
Z
(3) 2g
S0 = ◦Cˆ
Cˆ ◦C . (10.59)
3
closed

For on-shell external states, this interaction correctly produces three-point interac-
tions.
(2) (3)
Tree-level gauge invariance is violated for the theory described by S0 + S0
due to the violation of the associativity axiom. However, by adding higher-order
terms gauge invariance can be restored [168, 169, 170, 180, 212, 183, 280]. The new
interactions can be defined by relatively simple geometrical constraints [280, 281].
This leads to a tree-level non-polynomial closed-string field theory. Unfortunately,
the classical theory needs further modification at the quantum level. One-loop and
higher-loop amplitudes are not produced by using only tree-level vertices. It is at
this stage where the antibracket formalism has been of great utility. Interaction
terms proportional to powers of h̄ need to be added in a manner similar to Eq.(6.14).
To ensure that the theory is quantum-mechanically gauge invariant, the work in
refs.[280, 224, 281, 282, 149, 225] has relied on the antibracket formalism. The guiding
principle is that the quantum closed-string field theory must satisfy the quantum
master equation.
The antibracket is defined using the quadratic form in Eq.(10.57) [224, 281, 282].
As in the open string field theory, the system is infinitely reducible so that there are
ghosts for ghosts ad infinitum. The fields can be collected into one object Ψc in a
manner similar to the open string case in Eq.(5.23)

s∗+∗4 4 3 2 1 −s + 1
Ψc ≡ . . . + Cs + . . . + ∗ C0∗ + ∗ C ∗ + C + C0 + . . . + Cs + . . . , (10.60)
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 158

where the ghost number is indicated above the field. The close-string Hodge opera-
tion, denoted by a superscript ∗ in front of a field, is defined using the quadratic form
in Eq.(10.57). It takes a p-form into a 5 − p, where the order of string form is the
same as the ghost number.
Let ϕs be a complete set of normalized first-quantized states for g (ϕs ) ≤ 2. Let
∗ s
ϕ denote the corresponding state transformed by the Hodge star operation. The
∗ s
ϕ are a normalized complete set of states for ghost numbers greater than 2. With
these definitions, Ψc in Eq.(10.60) can be written in terms of ordinary particle fields
P
ψ s via Ψc = s (ϕs ψ s + ∗ ϕs ψs∗ ) where ψs∗ are the antifields of ψ s . The quantum
master equation is then the same as the particle case, namely 12 (W (Ψc ), W (Ψc )) =
r X ∂l Y ∂r X ∂l Y
ih̄∆W (Ψc ), where the antibracket is (X (Ψc ) , Y (Ψc )) = ∂∂ψ s ∂ψ ∗ − ∂ψ ∗ ∂ψ s and ∆ =
s s
(−1)ǫs +1 ∂ψ
∂r ∂r
s ∂ψ ∗ .
s
The solution of the quantum master equation for the closed-string field theory
is presented in refs.[281, 282]. This tour-de-force work goes beyond the goals of our
review. The reader interested in this topic can consult the above references for more
discussion.
The current formulation of string fields theories is developed around a particular
space-time background. Any background is permitted, as long as it leads to a nilpo-
tent first-quantized BRST charge. Such BRST charges correspond to two-dimensional
conformal field theories with the total central charge of the Virasoro algebra equal
to zero. Usually, the flat space-time background in 26 dimensions is used. Since
string theories contain gravity, it should be possible to pass from one background to
another. It is an interesting question of whether there is background independence
of string field theory [223, 222, 274, 224, 275, 225]. A proof for bosonic string field
theories has been obtained in refs.[224, 225], for backgrounds infinitesimally close.
The antibracket formalism has played an important role in the analysis. The basic
idea is that string field theory, formulated about a particular background B, corre-
sponds to a particular solution SB of the classical master equation. Reference [224]
demonstrated that, for two nearby backgrounds B1 and B2 , SB1 and SB2 are related
by a canonical transformation of the antibracket. The conclusion is that string field
theory is background independent, although not manifestly. Barring difficulties with
singular expressions, ref.[225] has extended the result to the quantum case. For the
quantum system, a particular background B corresponds to a solution WB of the
quantum master equation.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 159

10.9 Extended Antibracket Formalism for

Anomalous Gauge Theories
In certain cases, it is possible to quantize an anomalous gauge theory. An example
is the first-quantized bosonic string for D 6= 26 discussed in Sect. 9.3. Polyakov
[210] quantized this system in the presence of a conformal anomaly. A new degree of
freedom, the Liouville mode, emerged. Another example is the chiral Schwinger model
in two dimensions. Despite its anomalous behavior, it is consistent and unitary [166].
In refs.[97, 99], additional degrees of freedom were introduced into four-dimensional
anomalous Yang-Mills theories to cancel the anomalies in the path integral. This
cancellation can be obtained by adding a Wess-Zumino term for the extra degrees
of freedom. A careful treatment of the integration measure in Faddeev-Popov path-
integral quantization shows how such a Wess-Zumino term can arise naturally [207].
For earlier approaches to this subject in the case of the Schwinger model, see [7, 145].
A treatment of anomalous chiral QCD in two dimensions within the field-antifield
formalism was obtained in refs.[61, 63]. Methods of quantizing anomalous gauge
theories using the antibracket formalism were developed in [61, 62, 131, 133, 134, 77,
63].
Let us describe in general terms the extended antibracket method of refs.[131,
133, 134]. For simplicity we consider the closed irreducible case. At the classical
level, the number of dynamical local degrees of freedom ndof is the total number of
fields n minus the number of gauge invariances m0 , i.e., ndof = n − m0 . Suppose
there are r anomalous gauge invariances. Then, due to quantum effects, r of the
m0 gauge degrees of freedom enter the theory dynamically. Hence, the true net
number of degrees of freedom at quantum level is n − m0 + r. Following the ideas of
refs.[268, 210, 279, 97, 99] for treating anomalous gauge theories, one wants to have
r extra degrees of freedom. The proposal is to augment the original set of fields, φi ,
i = 1, 2, . . . , n with r new fields φbî , î = 1, 2, . . . , r, in such a way that the original
gauge structure continues to be maintained at classical level. Roughly speaking, the
φb are fields parametrizing the anomalous part of the gauge group. In what follows, a
“hat” on a quantity indicates that the quantity is associated with the extra degrees
of freedom or that the quantity has been generalized to the extended system.
The key step is to extend the antibracket formalism to include φb variables. To
implement this idea, one demands the extra fields to transform under the action of
the gauge group: h i
δ φbî = R b εα .
b î φ, φ
α (10.61)

Let φb∗î be the antifield of φbî . Given Eq.(4.29), the classical gauge structure of the
extended theory should be governed by an action Se = S + φb∗î R b î C α . Since field indices
α
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 160

range over a = 1, 2, . . . , n, n + 1, . . . , n + r, the compact notation φa ≡ [φi , φbî ] and

b î ] is useful. To maintain the gauge structure, the extended generators
Rαa ≡ [Rαi , R α
a
Rα must satisfy the original gauge algebra. In other words, when the field content is
extended, Eq.(2.22) must still hold, where the tensors T and E are functions of the φi
only and have the same values as in the unextended theory. This requirement leads to
a set of equations for the generators R b î in Eq.(10.61). Formulas for R
b î in the closed
α α
case for which E = 0 are given in refs.[131, 133]. In the closed case, it is possible
to solve the antifield independent part of the original quantum master equation at
one-loop in a quasilocal way in the extended theory. In particular, the Wess-Zumino
term M1 [φ, φ]b can be written as

h i Z 1  h i
M1 φ, φb = −i b
dsAî F φ, φs φbî , (10.62)
0

where the Aî are the antifield-independent part of the anomalies and F i is the finite
gauge transformation of the classical fields φi under the anomalous part of the group.
The BRST variation of M1 gives the original anomaly: (M1 , S) e = iA C α .
α
e
However, in the extended antibracket formalism the action S is not proper [77].
To have a well defined perturbation expansion, it is necessary to modify Se to a new
extended action Sb that satisfies the classical master equation
 
b S
S, b =0 , (10.63)

and is proper, i.e., 

∂l ∂r Sb 
rank  = n + m0 + r , (10.64)
∂z a ∂z b on−shell
where the z a include n fields, r extended fields, m0 ghosts, and their antifields. One
proposal for Sb would be Se + h̄M1 . A difficulty is the presence of an order h̄ term.
1
However, in certain cases a canonical transformation which scales φbî by 1/h̄ 2 [77] can
be performed to overcome the problem. The general structure for Sb for anomalous
gauge theories with an anomalous abelian subgroup is
i c bĵ b∗ b î bĵ β
Sb = S − φbî D îĵ φ + φî Tβ ĵ φ C , (10.65)
2
where Dc and Tb are tensors which can be found in ref.[133]. In particular, D
c is related
îĵ
to the BRST variation of the anomalies Aî . Since Sb satisfies
 the master equation, a
classical BRST symmetry can be defined using δB X = X, S . b

The final stage is to find a solution W c to the quantum master equation in the ex-
tended space. Because of the above-mentioned scaling of φ, c is a series in h̄ 21 [77, 257]
b W
 
c =S
rather than h̄: W b + h̄ 21 M
c1 + . . .. After regulating the extended theory, ∆S b
2 reg
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 161

is computed. By appropriately adjusting Sb and M

c1 , the antifield-independent part of
2

the complete one-loop obstruction Ab1 to the quantum master equation in the extended
space can be cancelled. In terms of the unscaled extra variables, this adjustment cor-
responds to a finite renormalization of the original expression of the Wess-Zumino
term in Eq.(10.62), M1 → M c . Note that the locality of M c1 cannot be guaranteed.
1
2
Likewise, locality of the renormalized Wess-Zumino term is not assured due to the
integral over the variable s in expressions like (10.62). In some cases, such as the
first-quantized bosonic string [131, 133] or the abelian Schwinger model [207], only
local action terms are generated. Then, the anomalous theory makes sense at the
quantum level. However, for chiral QCD in two dimensions [131], the integral over s
remains. Even in these cases, the violation of locality is in some sense not severe: The
equations of motion are local, a situation referred to as quasilocal. When the quan-
tum extended theory is well-defined, the final stage, namely gauge-fixing, proceeds in
a manner similar to the non-anomalous case [131].

Acknowledgements
We thank G. Barnich, C. Batlle, F. De Jonghe, M. Henneaux, A. K. Kost-
elecký, J. M. Pons, J. Roca, R. Siebelink, A. Slavnov, P. Townsend, W. Troost, S.
Vandoren, A. Van Proeyen and F. Zamora for discussions. This work is supported
in part by the Comisión Interministerial para la Ciencia y la Technología (project
number AEN-0695), by a Human Capital and Mobility Grant (ERB4050PL930544),
by the National Science Foundation (grant number PHY-9009850), by a NATO Col-
laborative Research Grant (0763/87), by the Robert A. Welch Foundation, and by
the United States Department of Energy (grant number DE-FG02-92ER40698).
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 162

A Appendix: Right and Left Derivatives

In this appendix, we provide more details about left and right derivatives [38, 40,
188, 85, 39]. For any function or functional X of φ, they are defined as
→ ←
∂l X ∂ ∂r X ∂
≡ X , ≡X . (A.1)
∂φ ∂φ ∂φ ∂φ
Left derivatives are the ones usually used. Right derivatives act from right to left.
The differential dX(φ) of X is

∂l X ∂r X
dX(φ) = dφ = dφ . (A.2)
∂φ ∂φ
The formula for the variation δX of X with respect to φ resembles Eq.(A.2):

∂l X ∂r X
δX(φ) = δφ = δφ . (A.3)
∂φ ∂φ

What is the relation between left and right derivatives? If φ is commuting (ǫ (φ) =
0), then ∂∂φ
lX
= ∂∂φ
rX
, so that one only needs to be careful when φ is anticommuting
(ǫ (φ) = 1). Assume φ is anticommuting. Then φφ = 0. Without loss of generality
we may assume that X = φY + Z where Y and Z have no φ dependence. The left
and right derivatives of X are then ∂∂φlX
= Y and ∂∂φ rX
= (−1)ǫY Y = (−1)ǫ(φ)(ǫX +1) Y .
For all cases,
∂l X ∂r X
= (−1)ǫ(φ)(ǫX +1) . (A.4)
∂φ ∂φ
As a pedagogical exercise, let us derive the graded antisymmetry property of the
bracket in Eq.(4.5). Start with the definition of (Y, X) in Eq.(4.4) and interchange
the order of derivatives to obtain
∂r Y ∂l X ∂r Y ∂l X
(Y, X) = − =
∂ΦA ∂ΦA ∂Φ∗A ∂ΦA
∗

∂l X ∂r Y (ǫX +ǫA )(ǫY +ǫA +1) ∂l X ∂r Y

(−1)(ǫY +ǫA )(ǫX +ǫA +1) ∗
− (−1) .
∂ΦA ∂ΦA ∂ΦA ∂Φ∗A
Then using Eq.(A.4), one finds that the above is equal to
∂r X ∂l Y ∂r X ∂l Y
(−1)ǫX ǫY +ǫX +ǫY +1 ∗
− (−1)ǫX ǫY +ǫX +ǫY +1
∂ΦA ∂Φ A ∂ΦA ∂Φ∗A

= − (−1)(ǫX +1)(ǫY +1) (X, Y ) .

This is the desired result.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 163

As another exercise, let us verify the formulas for (F, F ) and (B, B) in Eq.(4.6)
When Y = X, the second term in Eq.(4.4) can be written as

∂r X ∂l X ∂r X ∂l X
∗
= (−1)(ǫX +1)(ǫX +1) ,
∂ΦA ∂Φ A ∂ΦA ∂Φ∗A

using the same manipulations as in the previous paragraph. When X = F is anti-

commuting, the sign factor is plus and the second term in Eq.(4.4) cancels the first.
When X = B is commuting, the sign factor is minus and the two add.
Another useful result concerns integration by parts. When φ is commuting, one
R R
has the standard formula dφ ∂∂φ rX
Y = − dφX ∂∂φ lY ∂r
. In such a case, ∂φ ∂l
= ∂φ , so that
the left and right subscripts on ∂ are inconsequential. Suppose ǫ (φ) = 1. Then,
R R
dφ ∂∂φ
rX
Y and dφX ∂∂φ lY
are both zero unless both X and Y are linear in φ. Without
loss of generality, we may assume that X = xφ and Y = φy where x and y are
R R R
independent of φ. Then, dφ ∂∂φ rX
Y = dφx (φy) = (−1)ǫ(x) xy and dφX ∂∂φ lY
=
R
dφ (xφ) y = (−1)ǫ(x) xy lead to the same result. Summarizing, all cases are contained
in the formula Z Z
∂r X ǫ(φ)+1 ∂l Y
dφ Y = (−1) dφX . (A.5)
∂φ ∂φ
For second derivatives of X, one has
∂r ∂l X ∂l ∂r X
′
= ,
∂φ∂φ ∂φ′ ∂φ
∂l ∂l X ′ ∂l ∂l X

′
= (−1)ǫ(φ)ǫ(φ ) ′ ,
∂φ∂φ ∂φ ∂φ
∂r ∂r X ǫ(φ)ǫ(φ′ ) ∂r ∂r X
= (−1) . (A.6)
∂φ∂φ′ ∂φ′ ∂φ
In the first equation, derivatives act from different directions and hence commute. In
the second and third equations, one must be careful of the order.
If X is a functional of Y which is a function of φ, one has the chain rules

∂r X (Y (φ)) ∂r X ∂r Y
= ,
∂φ ∂Y ∂φ

∂l X (Y (φ)) ∂l Y ∂l X
= . (A.7)
∂φ ∂φ ∂Y
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 164

B Appendix: The Regularity Condition

If a theory is invariant under the gauge transformations in Eq.(2.1) then the
lagrangian does not depend on all degrees of freedom. In other words, S0 [φ′ ] = S0 [φ]
where φ′ stands for finitely transformed fields produced by any of the infinitesimal
variations in Eq.(2.1). When this relation is expanded to first order in the gauge
parameters εα , the Noether relations in Eq.(2.8) are obtained.
Let φ0 be the stationary point about which one would like to perform the pertur-
bative expansion. Then, in a neighborhood of φ0 there are other stationary points
given by performing finite gauge transformations on φ0 . Let Σ locally be the surface
around φ0 in φ configuration space where the equations of motion vanish. The regu-
larity condition assumes that the dimension of Σ is maximal and that the quadratic
form generated by expanding the lagrangian to second order in fields has a rank ndof
on this surface [28]. Hence, the number of fields that enter dynamically in S0 is ndof .
The regularity assumption is important for implementing perturbation theory since
the propagator – which is the inverse of this quadratic form – then exists.
Summarizing, the regularity condition is
 
∂l S0,i  ∂l ∂r S0 
rank  =  = ndof , (B.1)
∂φj Σ ∂φi ∂φj Σ
where Σ is the stationary surface defined implicitly by

S0,i |Σ = 0 . (B.2)

In other words, the on-shell degeneracy of the hessian in Eq.(B.1) is completely due
to the n − ndof independent null vectors Rαi associated with gauge transformations
and does not come from some other source [27, 28]:

∂l ∂r S0 i 
R  =0 . (B.3)
∂φi ∂φj α Σ

An example of a lagrangian that does not satisfy the regularity condition is L = φ4

with no kinetic energy term for φ. The stationary point φ0 = 0 has a vanishing
quadratic form even though there is no gauge invariance. In such a case one can
proceed by arbitrarily adding and subtracting some kinetic energy term and treating
φ4 minus this kinetic energy term as a perturbation. However, throughout this review
we assume that such singular cases do not arise.
In principle, one can separate the degrees of freedom into propagating degrees of
freedom ϕs , s = 1, 2, . . . , ndof and gauge degrees of freedom χa , a = 1, 2, . . . , n −
ndof . An efficient separation is often difficult and the ϕs and χa are usually quite
complicated and nonlocal functionals of the φi .
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 165

The regularity conditions are then given by

S0,a = 0 , identically ,

∂l S0,s 
rank  = ndof . (B.4)
∂φi φ=φ0
The regularity condition assumes that the equations of motion S0,i constitute a
regular representation of the stationary surface Σ. This means that the functions S0,i
can be locally split into independent, Gs , and dependent ones, Ga , in such a way that

1. Ga = 0 are a direct consequence of Gs = 0, and

2. The rank of the matrix of the gradients dGs is maximal on Σ.

In other words, the regularity condition ensures that locally the changes of variables
φi → [ϕs , χa ] or φi → [Gs , Ga ] makes sense [27, 28, 105].
When the regularity condition is fulfilled, it can be shown that any smooth func-
tion that vanishes on the stationary surface Σ can be written as a combination of the
equations of motion [28, 103, 105, 106], i.e.,

F (φ)|Σ = 0 ⇒ F (φ) = λj S0,j , (B.5)

where the λj may be functions of the φi . No restrictions are made on the λj (φ).
Putting restrictions can lead to violations of (B.5). An example is presented in [257].
By considering only local functionals, ref.[257] found cases for which Eq.(B.5) could
not be satisfied as a local combination of the equations of motion.
For more details on regularity conditions as well as derivations of the above results
consult references [81, 27, 28, 106, 103, 105, 152].
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 166

C Appendix: Anomaly Trace Computations

In this appendix, we perform the functional trace calculations of Sect. 9. One
key idea is to use the Dyson-like expansion [102, 165]
Z1
exp [R0 + V ] = exp [R0 ] + ds exp [(1 − s) R0 ] V exp [sR0 ] +
0

Z1 Zu
du ds exp [(1 − u) R0 ] V exp [(u − s) R0 ] V exp [sR0 ] + . . . . (C.1)
0 0
Typically, R0 is independent of the cutoff M, and V goes like inverse powers of M
so that only a few terms in Eq.(C.1) need to be kept.
The anomaly equation Eq.(8.53) involves a functional trace [120]. If one uses
momentum-space eigenfunctions to saturate the sum, then expressions such as

exp (−ik · x) (O (∂µ )) exp (ik · x)

arise where O (∂µ ) is an arbitrary operator, or a product of operators, involving the

derivative ∂µ . By commuting exp (ik · x) through the expression, one arrives at

exp (−ik · x) (O (∂µ )) exp (ik · x) = (O (∂µ + ikµ )) 1 . (C.2)

When derivatives in O (∂µ + ikµ ) act on the function 1, they produce zero.
For the spinless relativistic particle system, we begin by taking Eq.(9.18) and
commuting exp (ikτ ) through the expression, using Eq.(C.2), to obtain

(∆S)reg =
 !      
Z Z∞ − d
+ ik ρ−1 d
+ ik
dk −1 d dτ dτ
D dτ  ρ C + ik exp   1 .
2π dτ M2
−∞ 0
Rescaling k by M produces
Z Z∞ " !
dk d
(∆S)reg = D dτ Mρ−1 C + iMk ×
2π dτ
−∞
! ! #
k2 k d d 1 d −1 d
exp −i ρ−1 + ρ−1 − ρ 1 . (C.3)
ρ M dτ dτ M2 dτ dτ 0
Eq.(C.3) is in the form of Eq.(C.1) where R0 = k 2 /ρ. We use the Dyson-like expansion
in Eq.(C.1) and pick out the M-independent term to arrive at

(∆S)reg =
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 167

 ! ! !
Z Z∞  d Z1
dk −1 k2 −1 d d −1 k2
−iD dτ kρ C  ds exp (1 − s) ρ + ρ exp s
2π dτ ρ dτ dτ ρ
−∞ 0

Z1 ! ! !
k2 d −1 d k2
+ ds exp (1 − s) ρ exp s
ρ dτ dτ ρ
0
Z1 Zu ! !
k2 −1 d d
+k 2 du ds exp (1 − u) ρ + ρ−1 ×
ρ dτ dτ
0 0
! ! !)
k2 d d k2
exp (u − s) ρ−1 + ρ−1 exp s . (C.4)
ρ dτ dτ ρ
The first term in Eq.(C.4) is zero because it is a total derivative. To calculate the
other two terms rotate to Euclidean space using k → −ikE . Then the expression
k 2 /ρ → −kE2 /ρ in the exponents yields gaussian damping factors, so that the integrals
are convergent. Even before evaluating the derivatives dτd , it is clear that the integrand
is an odd function of k and hence produces a zero integral.
For the chiral Schwinger model, one starts with Eq.(9.41). Using momentum-space
wave functions, the functional trace is
Z
∆S = −i d2 x C×
"Z      #
d2 k R+ R−
2 exp (−ik · x) exp 2
− exp exp (ik · x) , (C.5)
(2π) M M2 0
Equation (C.2) is used to eliminate the exp (±ik · x) factors. Then, one scales the
momentum k by M. The expression for ∆S becomes
Z Z
2 d2 k h 2      i
∆S = −i d xC M exp R̃+ − exp R̃− 1 , (C.6)
(2π)2 0

where
−2ik µ ∂µ ± A− k+ ∂ µ ∂µ ± i (∂+ A− ) ± iA− ∂+
R̃± = −k µ kµ − + . (C.7)
M M2
The parenthesis around (∂+ A− ) indicates that ∂+ acts only on A− .
The notation []0 selects the term in Eq.(C.6) independent of M. Hence, one
expands in the factors in the exponentials proportional to M−1 and M−2 . This
results in two terms, ∆S1 from the leading Taylor series term in M−2 , and ∆S2 from
the second order term in M−1 : ∆S = ∆S1 + ∆S2 , where
Z Z
2 d2 k µ
∆S1 = 2 d x C (∂+ A− ) 2 exp (−k kµ ) ,
(2π)
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 168

Z Z
d2 k
∆S2 = i d2 x C exp (−k µ kµ ) 2ik µ (∂µ A− ) k+ . (C.8)
(2π)2
Integrals are defined by analytic continuation using Wick rotation, that is, the k0
integral is performed by going to Euclidean space. The line integral contained in
R∞
dk0 can be rotated counterclockwise by 90o . Then, one sets k0 = −ik0E :
−∞

Z∞ Zi∞ −∞
Z Z∞
dk0 = dk0 = −i dk0E =i dk0E . (C.9)
−∞ −i∞ ∞ −∞

Integrals are then convergent since

h i h i h i
exp −k 2 = exp +k02 − k12 → exp −k0E2 − k12 , (C.10)

provides a gaussian damping factor. The following integration table is obtained

   



1 





2 


Z 
 k2 
  −1 
 
d2 k µ 0 i
2 exp (−k kµ )  2 
= . (C.11)
(2π) 
 k1   8π 

 1 



 
 
 
k0 k1 0 

Hence,
2i Z 2
d x C (∂+ A− ) ,
∆S1 =
4π
Z
i
∆S2 = − d2 x C (∂+ A− ) . (C.12)
4π
Adding the two contributions,
i Z 2
∆S = d x C (∂+ A− ) . (C.13)
4π
The factor ∂+ A− can be written in Lorentz covariant form using

∂+ A− = ∂0 A0 − ∂1 A1 − ∂0 A1 + ∂1 A0 = −∂µ Aµ + εµν ∂µ Aν . (C.14)

Substituting Eq.(C.14) into Eq.(C.13) produces the result in Eq.(9.43).

For the first-quantized bosonic string theory, we need to compute κr of Eq.(9.78).
Using momentum-space eigenfunctions, it can be expressed as [121, 122]
"Z   #
tr d2 k Hr
κr = 2 exp (−ik · x) exp exp (ik · x) , (C.15)
2 (2π) M2 0
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 169

where tr is a trace over two-by-two gamma-matrix space. In Eq.(C.15), k · x =

−kτ τ + kσ σ. Using the definition of Hr in Eq.(9.76) and commuting exp (ik · x)
through the expression via Eq.(C.2), Eq.(C.15) becomes
"Z ! #
tr d2 k ρ−(r+1)/2 (/ k ) ρr (/
∂ + i/ k ) ρ−(r+1)/2
∂ + i/
κr = exp − 1
2 (2π)2 M2 0
" Z ! #
tr d2 k k2 A B
= M2 2 exp − −i + 1 ,
2 (2π) ρ M M2 0
where we have performed the rescaling k → Mk, and where

A ≡ ρ−(r+1)/2 (/
∂ ρrk/ + k/ρr∂/) ρ−(r+1)/2 , (C.16)

B ≡ −ρ−(r+1)/2∂/ρr∂/ρ−(r+1)/2 . (C.17)
For the next step, we use the Dyson-like expansion in Eq.(C.1). Recalling that
[ ]0 indicates that the M-independent term is to be selected, κr becomes a sum of
two terms
κr = B-term + AA-term ,
where
Z Z1 Zu
tr d2 k
AA-term = − du ds ×
2 (2π)2 0 0
" # " # " #
k2 k2 k2
exp − (1 − u) A exp − (u − s) A exp −s 1 ,
ρ ρ ρ
and " # " #
Z Z1
tr d2 k k2 k2
B-term = ds exp − (1 − s) B exp −s 1 .
2 (2π)2 0
ρ ρ
In AA-term, carry out the differentiations ∂/ in both A operators using the defini-
tion of A in Eq.(C.16) to obtain
Z Z1 Zu (
d2 k h i ∂ n ∂n ρ  2 
AA-term = − du ds k 2 exp −k 2 2sk − 1 +
(2π)2 0 0
ρ
 )
∂ n ρ∂n ρ 1 2 
− r − 5 − 7sk 2 − uk 2 + 2suk 4 ,
ρ2 2
where another rescaling k → kρ1/2 has been performed. Next, rotate the k integration
∞
R
from Minkowski space to Euclidean space. The line integral contained in dkτ can
−∞
be rotated counterclockwise by 90o . One sets kτ = −ikτE and uses Eq.(C.9) for
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 170

k0 =h kτ . The iexponential factor exp [−k 2 ] = exp [+kτ2 − kσ2 ] in AA-term becomes
exp −kτE2 − kσ2 . One can then do the s, u and k integrations since the latter are
now convergent. The result is
!!
1 ∂ n ∂n ρ 1 ∂ n ρ∂n ρ r2 + 1
AA-term = − . (C.18)
4πi ρ 6 ρ2 4

The B-term is treated similarly. One carries out the differentiations ∂/ in the B
operator and rescales k by ρ1/2 to arrive at
1 (
tr Z d2 k Z h i ∂n∂ ρ  r+1

2 n 2
B-term = ds exp −k sk − +
2 (2π)2 0 ρ 2
!)
∂ n ρ∂n ρ (3 − r) (r + 1)
− 3sk 2 + s2 k 4 .
ρ2 4
After rotating to Euclidean space, all integrations can be formed. The B-term is
!!
1 ∂ n ∂n ρ r ∂ n ρ∂n ρ r2 r 1
B-term = + − + . (C.19)
4πi ρ 2 ρ2 4 2 12

The sum of the AA- and B-terms in Eqs.(C.18) and (C.19) is the quoted result for
κr in Eq.(9.79).
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 171

References
[1] E. S. Abers and B. W. Lee, Gauge Theories, Phys. Rep. C9 (1973) 1.

[2] J. Alfaro and P. H. Damgaard, Origin of Antifields in the Batalin-Vilkovisky La-

grangian Formalism, Nucl. Phys. B404 (1993) 751;
Generalized Lagrangian Master Equations, Phys. Lett. B334 (1994) 369.

[3] L. Alvarez-Gaumé and L. Baulieu, The Two Quantum Symmetries Associated

with a Classical Symmetry, Nucl. Phys. B212 (1983) 255.

[4] D. Anselmi, Predictivity and Nonrenormalizability, SISSA preprint SISSA-147-

93-EP, (Sept., 1993), hep-th/9309085.

[5] S. Aoyama and S. Vandoren, The Batalin-Vilkovisky Formalism on Fermionic

Kähler Manifolds, Mod. Phys. Lett. A8 (1993) 3773.

[6] C. Aragone and S. Deser, Consistency Problems in Hypergravity, Phys. Lett.

86B (1979) 161.

[7] O. Babelon, F. A. Schaposnik and C. M. Viallet, Quantization of Gauge Theories

with Weyl Fermions, Phys. Lett. 177B (1986) 385.

[8] G. Bandelloni, Yang-Mills Cohomology in Four Dimensions, J. Math. Phys. 27

(1986) 2551; Nonpolynomial Yang-Mills Cohomology, J. Math. Phys. 28 (1987)
2775.

[9] T. Banks, D. Friedan, E. Martinec, M. E. Peskin and C. R. Preitschopf, All Free

String Theories Are Theories of Forms, Nucl. Phys. B274 (1986) 71.

[10] A. Barducci, R. Casalbouni and L. Lusanna, Supersymmetries and the Pseudo-

classical Relativistic Electron, Nuovo Cim. 35A (1976) 377.

[11] G. Barnich, F. Brandt and M. Henneaux, Local BRST Cohomology in the An-
tifield Formalism: I. General Theorems, preprint ULB-th-94/06, NIKHEF-H
94-13, hep-th/9405109.

[12] G. Barnich, R. Constantinescu and P. Grégoire, BRST–Anti-BRST Antifield

Formalism. The Example of the Freedman-Townsend Model, Phys. Lett. B293
(1992) 353.

[13] G. Barnich and M. Henneaux, Consistent Couplings Between Fields with a

Gauge Freedom and Deformations of the Master Equation, Phys. Lett. B311
(1993) 123.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 172

[14] G. Barnich and M. Henneaux, Renormalization of Gauge Invariant Operators

and Anomalies in Yang-Mills Theory, Phys. Rev. Lett. 72 (1994) 1588.

[15] F. Bastianelli, Ward Identities for Quantum Field Theories with External Fields,
Phys. Lett. B263 (1991) 411.

[16] F. Bastianelli, A Locally Supersymmetric Action for the Bosonic String,

Phys. Lett. B322 (1994) 340.

[17] I. A. Batalin, Quasigroup Construction and First Class Constraints,

J. Math. Phys. 22 (1981) 1837.

[18] I. A. Batalin and E. S. Fradkin, A Generalized Canonical Formalism and Quan-

tization of Reducible Gauge Theories, Phys. Lett. 122B (1983) 157.

[19] I. A. Batalin and E. S. Fradkin, Operator Quantization and Abelization of Dy-

namical Systems Subject to First-Class Constraints, Riv. Nuov. Cim. 9 (1986)
1.

[20] I. A. Batalin, P. M. Lavrov and I. V. Tyutin, Covariant Quantization of Gauge

Symmetries in the Framework of Extended BRST Symmetry, J. Math. Phys.
31 (1990) 1487;
An Sp(2)-Covariant Quantization of Gauge Theories with Linearly Dependent
Generators, J. Math. Phys. 32 (1991) 532.

[21] I. A. Batalin and I. V. Tyutin, On the Multilevel Generalization of the Field-

Antifield Formalism, Mod. Phys. Lett. A8 (1993) 3673.

[22] I. A. Batalin and I. V. Tyutin, On Possible Generalizations of Field-Antifield

Formalism, Int. J. Mod. Phys. A8 (1993) 2333.

[23] I. A. Batalin and G. A. Vilkovisky, Relativistic S-Matrix of Dynamical Systems

with Boson and Fermion Constraints, Phys. Lett. 69B (1977) 309.

[24] I. A. Batalin and G. A. Vilkovisky, Gauge Algebra and Quantization, Phys. Lett.
102B (1981) 27.

[25] I. A. Batalin and G. A. Vilkovisky, Feynman Rules for Reducible Gauge Theories,
Phys. Lett. 120B (1983) 166.

[26] I. A. Batalin and G. A. Vilkovisky, Quantization of Gauge Theories with Linearly

Dependent Generators, Phys. Rev. D28 (1983) 2567; Errata: D30 (1984) 508.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 173

[27] I. A. Batalin and G. A. Vilkovisky, Closure of the Gauge Algebra, Generalized

Lie Algebra Equations and Feynman Rules, Nucl. Phys. B234 (1984) 106.

[28] I. A. Batalin and G. A. Vilkovisky, Existence Theorem for Gauge Algebra,

J. Math. Phys. 26 (1985) 172.

[29] C. Batlle, Teories de Camps de Cordes, Thesis, Universitat de Barcelona (1988).

[30] C. Batlle and J. Gomis, Lagrangian and Hamiltonian BRST Structures of the
Antisymmetric Tensor Gauge Theory, Phys. Rev. D38 (1988) 1169.

[31] C. Batlle, J. Gomis, J. Parı́s and J. Roca, Lagrangian and Hamiltonian BRST
Formalisms, Phys. Lett. B224 (1989) 288;
Field-Antifield Formalism and Hamiltonian BRST Approach, Nucl. Phys.
B329 (1990) 139.

[32] L. Baulieu, Perturbative Gauge Theories, Phys. Rep. 129 (1985) 1.

[33] L. Baulieu, M. Bellon, S. Ouvry and J.-C. Wallet, Batalin-Vilkovisky Analysis of

Supersymmetric Systems, Phys. Lett. B252 (1990) 387.

[34] L. Baulieu, E. Bergshoeff and E. Sezgin, Open BRST Algebras, Ghost Unification
and String Field Theory, Nucl. Phys. B307 (1988) 348.

[35] L. Baulieu and I.M. Singer, Topological Yang-Mills Symmetry, Nucl.Phys. B

(Proc. Suppl) 5B (1988) 12.

[36] C. Becchi, A. Rouet and R. Stora, The Abelian Higgs Kibble Model, Unitarity of
the S-Operator, Phys. Lett. 52B (1974) 344;
Renormalization of the Abelian Higgs-Kibble Model, Comm. Math. Phys. 42
(1975) 127;
Renormalization of Gauge Theories, Ann. Phys. 98 (1976) 287.

[37] F. A. Berends, J. W. van Holten, P. van Nieuwenhuizen and B. de Wit, On Spin

5/2 Gauge Fields, Phys. Lett. 83B (1979) 188.

[38] F. A. Berezin, The Method of Second Quantization, (Academic Press, New York,
1966).

[39] F. A. Berezin, Introduction to Superanalysis, Mathematical Physics and Applied

Mathematics, no 9, (Reidel, Dordrecht, 1987).

[40] F. A. Berezin and D. A. Leites, Supermanifolds, Sov. Math. Dokl. 16 (1975) 1218.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 174

[41] F. A. Berezin and M. S. Marinov, Particle Spin Dynamics as the Grassmann

Variant of Classical Mechanics, Ann. Phys. 104 (1977) 336.

[42] E. Bergshoeff, J. de Boer, M. de Roo and T. Tjin, On the Cohomology of the

Non-Critical W String, Nucl. Phys. B420 (1994) 379.

[43] E. Bergshoeff and R. Kallosh, Unconstrained BRST for Superparticles,

Phys. Lett. B240 (1990) 105.

[44] E. Bergshoeff, R. Kallosh and A. Van Proeyen, Superparticle Actions and Gauge
Fixings, Clas. Quantum Grav. 9 (1992) 321.

[45] E. Bergshoeff, A. Sevrin and X. Shen, A Derivation of the BRST Operator for
Non-Critical W -Strings, Phys. Lett. B296 (1992) 95.

[46] N. Berkovits, M. T. Hatsuda and W. Siegel, The Big Picture, Nucl. Phys. B371
(1992) 434.

[47] M. Bershadsky, S. Cecotti, H. Ooguri and C. Vafa, Kodaira-Spencer Theory of

Gravity and Exact Results for Quantum String Amplitudes, Harvard Univ.
preprint HUTP-93-A025, (Sept, 1993), hep-th/9309140.

[48] M. Bershadsky, W. Lerche, D. Nemeschansky and N. P. Warner, A BRST Oper-

ator for Non-Critical Strings, Phys. Lett. B292 (1993) 35;
Extended N = 2 Superconformal Structure of Gravity and W -Gravity Coupled
to Matter, Nucl. Phys. B401 (1993) 304.

[49] D. Birmingham, M. Blau, M. Rakowski and G. Thompson, Topological Field

Theory, Phys. Rep. 209 (1991) 129.

[50] D. Birmingham, M. Rakowski and G. Thompson, BRST Quantization of Topo-

logical Field Theories, Nucl. Phys. B315 (1989) 577.

[51] M. Blagojević and B. Sazdović, Off-Shell BRST Quantization of Reducible

Gauge Theories, Phys. Lett. B223 (1989) 325.

[52] A. Blasi and G. Bandelloni, Antisymmetric Tensor Gauge Theory,

Clas. Quantum Grav. 10, (1993) 1249.

[53] K. Bleuler, A New Method for the Treatment of Longitudinal and Scalar Pho-
tons, Helv. Phys. Acta 23 (1950) 567.

[54] R. Bluhm and S. Samuel, The Off-Shell Koba-Nielsen Formula, Nucl. Phys.
B323 (1989) 337.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 175

[55] R. Bluhm and S. Samuel, Off-Shell Conformal Field Theory at the One-Loop
Level, Nucl. Phys. B325 (1989) 275.

[56] M. Bochicchio, String Field Theory in the Siegel Gauge, Phys. Lett. 188B
(1987) 330;
Gauge Fixing for the Field Theory of the Bosonic String, Phys. Lett. 193B
(1987) 31.

[57] L. Bonora and P. Cotta-Ramusino, Some Remarks on BRST Transformations,

Anomalies and the Cohomology of the Lie Algebra of the Group of Gauge Trans-
formations, Comm. Math. Phys. 87 (1983) 589.

[58] A. Borel, Sur la Cohomologie des Espaces Fibres Principaux et Espaces Homo-
genes de Groupes de Lie Compacts, Ann. Math. 57 (1953) 115.

[59] P. Bouwknegt, J. McCarthy and K. Pilch, BRST Analysis of Physical States for
2-d Gravity Coupled to c ≤ 1 Matter, Comm. Math. Phys. 145 (1992) 541.

[60] P. Bouwknegt, J. McCarthy and K. Pilch, Semi-infinite Cohomology of W Alge-

bras, Lett. Math. Phys. 29 (1993) 91.

[61] N. R. F. Braga and H. Montani, Batalin-Vilkovisky Lagrangian Quantization of

the Chiral Schwinger Model, Phys. Lett. B264 (1991) 125.

[62] N. R. F. Braga and H. Montani, The Wess-Zumino Term in the Field-Antifield

Formalism, Int. J. Mod. Phys. A8 (1993) 2569.

[63] N. R. F. Braga and H. Montani, BRST Quantization of the Chiral Schwinger

Model in the Extended Field-Antifield Space, Phys. Rev. D49 (1994) 1077.

[64] F. Brandt, Antifield Dependence of Anomalies, Phys. Lett. B320 (1994) 57.

[65] L. Brink, S. Deser, B. Zumino, P. di Vecchia and P. Howe, Local Supersymmetry

for Spinning Particles, Phys. Lett. 64B (1977) 435.

[66] R. Brooks, D. Montano and J. Sonnenschein, Gauge Fixing and Renormalization

in Topological Quantum Field Theory, Phys. Lett. B214 (1988) 91.

[67] E. Cremmer and J. Scherk, Spontaneous Dynamical Breaking of Gauge Symme-

try in Dual Models, Nucl. Phys. B72 (1974) 117.

[68] E. Cremmer, A. Schwimmer and C. Thorn, The Vertex Function in Witten’s

Formulation of String Field Theory, Phys. Lett. B179 (1986) 57.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 176

[69] G. Curci and R. Ferrari, Slavnov Transformations and Supersymmetry,

Phys. Lett. 63B (1976) 91.

[70] G. Curci and R. Ferrari, An Alternative Approach to the Proof of Unitarity for
Gauge Theories, Nuovo Cim. 35A(1976) 273.

[71] O. F. Dayi, Odd Time Formulation of the Batalin-Vilkovisky Method of Quan-

tization, Mod. Phys. Lett. A4 (1989) 361.

[72] O. F. Dayi, BV and BFV Formulation of a Gauge Theory of Quadratic Lie Al-
gebras in 2-D and a Construction of W3 Topological Gravity, Tubitak Research
Institute preprint, (Jan., 1994), hep-th/9401148.

[73] S. P. de Alwis, M. T. Grisaru and L. Mezincescu, Quantization and Unitarity in

Antisymmetric Tensor Gauge Theories, Nucl. Phys. B303 (1988) 57.

[74] J. de Boer and J. Goeree, The Effective Action of W3 to All Orders, Nucl. Phys.
B401 (1993) 348.

[75] F. De Jonghe, The Batalin-Vilkovisky Lagrangian Quantization Scheme: With

Applications to the Study of Anomalies in Gauge Theories, Leuven Univ.
preprint, (Dec., 1993), hep-th/9403143.

[76] F. De Jonghe, Schwinger-Dyson BRST Symmetry and the Equivalence of Hamil-

tonian and Lagrangian Quantisation, Phys. Lett. B316 (1993) 503.

[77] F. De Jonghe, R. Siebelink and W. Troost, Hiding Anomalies, Phys. Lett. B306
(1993) 295.

[78] F. De Jonghe, R. Siebelink, W. Troost, S. Vandoren, P. van Nieuwenhuizen and

A. Van Proeyen, The Regularized BRST Jacobian of Pure Yang-Mills Theory,
Phys. Lett. B289 (1992) 354.

[79] F. De Jonghe and S. Vandoren, Construction of Topological Field Theories Using

the Batalin-Vilkovisky Quantization Scheme, Phys. Lett. B324 (1994) 328.

[80] S. Deser, P. K. Townsend and W. Siegel, Higher Rank Representations of Lower

Spin, Nucl. Phys. B184 (1981) 333.

[81] B. de Wit and J. W. van Holten, Covariant Quantization of Gauge Theories with
Open Algebra, Phys. Lett. B79 (1979) 389.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 177

[82] B. DeWitt, Dynamical Theory of Groups and Fields, in Relativity, Groups and
Topology, Les Houches Summer School, Session XIII, edited by C. DeWitt and
B. DeWitt, (Gordon Breach, 1964).

[83] B. DeWitt, Quantum Theory of Gravity. I. The Canonical Theory, Phys. Rev.
160 (1967) 1113;
Quantum Theory of Gravity. II. The Manifestly Covariant Theory, Phys. Rev.
162 (1967) 1195;
Quantum Theory of Gravity. III. Applications of the Covariant Theory,
Phys. Rev. 162 (1967) 1239.

[84] B. S. DeWitt, The Spacetime Approach to Quantum Field Theory, in Relativity,

Groups and Topology, II, Les Houches Summer School, session XL, edited by
B. S. DeWitt and R. Stora, (North-Holland, Amsterdam, 1984).

[85] B. S. DeWitt, Supermanifolds, (Cambridge Univ. Press, Cambridge, 1984).

[86] A. H. Diaz, The Nonabelian Antisymmetric Tensor Field Revisited, Phys. Lett.
B203 (1988) 408.

[87] A. Diaz, W. Troost, P. van Nieuwenhuizen and A. Van Proeyen, Understand-

ing Fujikawa Regulators from Pauli-Villars Regularization of Ghost Loops,
Int. J. Mod. Phys. A4 (1989) 3959.

[88] P. A. M. Dirac, Lectures on Quantum Mechanics, (Yeshiva University, New York,

1964).

[89] J. A. Dixon, Cohomology and Renormalization of Gauge Theories I,II,III,IV,

unpublished (1976-1979); Comm. Math. Phys. 139 (1991) 495.

[90] S. Donaldson, An Application of Gauge Theory to the Topology of Four Mani-

folds, J. Diff. Geom. 18 (1983) 279;
Polynomial Invariants of Smooth Four-Manifolds, Topology 29 (1990) 257.

[91] A. Dresse, J. M. L. Fisch, P. Grégoire and M. Henneaux, Equivalence of the

Hamiltonian and Lagrangian Path Integrals for Gauge Theories, Nucl. Phys.
B354 (1991) 191.

[92] M. Dubois-Violette, M. Henneaux, M. Talon and C. M. Viallet, Some Results on

Local Cohomologies in Field Theory, Phys. Lett. B267 (1991) 81.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 178

[93] M. Dubois-Violette, M. Talon and C. M. Viallet, B. R. S. Algebras. Analysis of

the Consistency Equations in Gauge Theory, Comm. Math. Phys. 102 (1985)
105.

[94] F. J. Dyson, The S-Matrix in Quantum Electrodynamics, Phys. Rev. 75 (1949)

1736.

[95] R. J. Eden, P. V. Landshoff, D. I. Olive and J. C. Polkinghorne, The Analytic S-

Matrix, (Cambridge University Press, Cambridge, UK, 1966).

[96] T. Eguchi, P. B. Gilkey and A. J. Hanson, Gravitation, Gauge Theories and Dif-
ferential Geometry, Phys. Rep. 66 (1980) 213.

[97] L. D. Faddeev, Operator Anomaly for the Gauss Law, Phys. Lett. 145B (1984)
81.

[98] L. D. Faddeev and V. N. Popov, Feynman Diagrams for the Yang-Mills Field,
Phys. Lett. 25B (1967) 29.

[99] L. D. Faddeev and S. L. Shatashvili, Algebraic and Hamiltonian Methods in the

Theory of Non-Abelian Anomalies, Theor. Math. Phys. 60 (1984) 770;
Realization of the Schwinger Term in the Gauss Law and the Possibility of
Correct Quantization of a Theory with Anomalies, Phys. Lett. 167B (1986)
225.

[100] L. D. Faddeev and A. A. Slavnov, Gauge Fields: Introduction to Quantum The-

ory, (Benjamin-Cummings, Reading, MA, 1980).

[101] R. P. Feynman, Quantum Theory of Gravitation, Acta Phys. Pol. 24 (1963) 697.

[102] R. P. Feynman and A. R. Hibbs, Quantum Mechanics and Path Integrals,

(McGraw-Hill, New York, 1965).

[103] J. M. L. Fisch, On the Batalin-Vilkovisky Antibracket-Antifield BRST Formal-

ism and Its Applications, Univ. Libre de Bruxelles preprint ULB-TH2-90-01,
(Jan., 1990).

[104] J. M. L. Fisch and M. Henneaux, The Antifield-Antibracket Formalism for Con-

strained Hamiltonian Systems, Phys. Lett. B226 (1989) 80.

[105] J. M. L. Fisch and M. Henneaux, Homological Perturbation Theory and the Al-
gebraic Structure of the Antifield-Antibracket Formalism for Gauge Theories,
Comm. Math. Phys. 128 (1990) 627.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 179

[106] J. M. L. Fisch, M. Henneaux, J. Stasheff and C. Teitelboim, Existence, Unique-

ness and Cohomology of the Classical BRST Charge with Ghosts for Ghosts,
Comm. Math. Phys. 120 (1989) 379.

[107] H. Flanders, Differential Forms, (Academic Press, 1963).

[108] E. S. Fradkin and T. E. Fradkina, Quantization of Relativistic Systems with Bo-

son and Fermion First- and Second-Class Constraints, Phys. Lett. 72B (1978)
343.

[109] E. S. Fradkin and G. A. Vilkovisky, Quantization of Relativistic Systems with

Constraints, Phys. Lett. 55B (1975) 224.

[110] E. S. Fradkin and G. A. Vilkovisky, Unitarity of S-Matrix in Gravidynamics and

General Covariance in Quantum Domain, Nuovo Cim. Lett. 13 (1975) 187.

[111] D. Z. Freedman, S. Giddings, J. Shapiro and C. B. Thorn, The Nonplanar One-

Loop Amplitude in Witten’s String Field Theory, Nucl. Phys. B298 (1988)
253.

[112] D. Z. Freedman and P. K. Townsend, Antisymmetric Tensor Gauge Theories and

Non-Linear σ-Models, Nucl. Phys. B177 (1981) 282.

[113] D. Z. Freedman and P. van Nieuwenhuizen, Properties of Supergravity Theory,

Phys. Rev. D14 (1976) 912.

[114] D. Z. Freedman, P. van Nieuwenhuizen and S. Ferrara, Progress Toward a Theory

of Supergravity, Phys. Rev. D13 (1976) 3214.

[115] I. B. Frenkel, H. Garland and G. J. Zuckerman, Semi-Infinite Cohomology and

String Theory, Proc. Natl. Acad. Sci. 83 (1986) 8442.

[116] S. A. Frolov and A. A. Slavnov, Quantization of Non-Abelian Antisymmetric

Tensor Field, Theor. Math. Phys. 75 (1988) 470.

[117] S. A. Frolov and A. A. Slavnov, Physical Subspace Norm in Hamiltonian BRST-

Quantization, Phys. Lett. B218 (1989) 461.

[118] S. A. Frolov and A. A. Slavnov, Construction of the Effective Action for General
Gauge Theories Via Unitarity, Nucl. Phys. B347 (1990) 333.

[119] S. Frolov and A. Slavnov, Lagrangian BRST Quantization and Unitarity,

Theor. Math. Phys. 85 (1990) 1237.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 180

[120] K. Fujikawa, Path-Integral Measure for Gauge-Invariant Fermion Theories,

Phys. Rev. Lett. 42 (1979) 1195;
Path Integral for Gauge Theories with Fermions, Phys. Rev. D21 (1980) 2848.

[121] K. Fujikawa, Path Integral of Relativistic Strings, Phys. Rev. D25 (1982) 2584.

[122] K. Fujikawa, T. Inagaki and H. Suzuki, BRS Current and Related Anomalies in
Two Dimensional Gravity and String Theories, Nucl. Phys. B332 (1990) 499.

[123] S. J. Gates Jr., M. T. Grisaru, U. Lindström, M. Roček, W. Siegel, P. van

Nieuwenhuizen and A. E. van de Ven, Lorentz-Covariant Quantization of the
Heterotic Superstring, Phys. Lett. B225 (1989) 44.

[124] S. J. Gates Jr., M. T. Grisaru, M. Roček and W. Siegel, Superspace or One Thou-
sand and One Lessons in Supersymmetry, (Benjamin/Cummings, Reading, MA,
1983).

[125] M. Gerstenhaber, The Cohomology Structure of an Associative Ring,

Ann. Math. 78 (1962) 267.

[126] M. Gerstenhaber, On the Deformation of Rings and Algebras, Ann. Math. 79

(1964) 59.

[127] E. Getzler, Batalin-Vilkovisky Algebras and Two-Dimensional Topological Field

Theories, Comm. Math. Phys. 159 (1994) 265.

[128] S. Giddings, The Veneziano Amplitude from Interacting String Field Theory,
Nucl. Phys. B278 (1986) 242

[129] S. L. Glashow, Partial Symmetries of Weak Interactions, Nucl. Phys. 22 (1961)

579.

[130] J. Gomis and J. Parı́s, Unitarity and the Field-Antifield Formalism, Nucl. Phys.
B368 (1992) 311.

[131] J. Gomis and J. Parı́s, Field-Antifield Formalism for Anomalous Gauge Theo-
ries, Nucl. Phys. B395 (1993) 288.

[132] J. Gomis and J. Parı́s, Perturbation Theory and Locality in the Field-Antifield
Formalism, J. Math. Phys. 34 (1993) 2132.

[133] J. Gomis and J. Parı́s, Anomalies and Wess-Zumino Terms in an Extended

Regularized Field-Antifield Formalism, Nucl. Phys. B431 (1994) 378, hep-
th/9401161.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 181

[134] J. Gomis and J. Parı́s, Anomalous Gauge Theories Within the Field-Antifield
Formalism, in Proc. of the International EPS Conference on High Energy
Physics, HEP93, Marseille, (July, 1993), p. 101-102.

[135] J. Gomis, J. Parı́s, K. Rafanelli and J. Roca, BRST and Anti-BRST Symmetries
for the Spinning Particle, Phys. Lett. B246 (1990) 435.

[136] J. Gomis, J. Parı́s and J. Roca, BRST Structures of the Spinning String,
Nucl. Phys. B339 (1990) 711.

[137] J. Gomis and J. Roca, The Anti-BRST Symmetry in the Field-Antifield Formal-
ism, Nucl. Phys. B343 (1990) 152.

[138] T. Goto, Relativistic Quantum Mechanics of One Dimensional Mechan-

ical Continuum as Subsidiary Condition of Dual Resonance Model,
Prog. Theor. Phys. 46 (1971) 1560.

[139] M. B. Green and C. M. Hull, Covariant Quantum Mechanics of the Superstring,

Phys. Lett. B225 (1989) 57.

[140] G. V. Grigorian, R. P. Grigorian and I. V. Tyutin, Equivalence of Lagrangian

and Hamiltonian BRST Quantizations: Systems with First Class Constraints,
Sov. J. Nucl. Phys. 50 (1991) 1058.

[141] G. V. Grigorian, R. P. Grigorian and I. V. Tyutin, Equivalence of Lagrangian and

Hamiltonian BRST Quantizations: the General Case, Nucl. Phys. B379 (1992)
304.

[142] D. Gross and A. Jevicki, Operator Formulation of Interacting String Field The-
ory (I), Nucl. Phys. B283 (1987) 1.

[143] D. Gross and A. Jevicki, Operator Formulation of Interacting String Field The-
ory (II), Nucl. Phys. B287 (1987) 225.

[144] S. N. Gupta, Theory of Longitudinal Photons in Quantum Electrodynamics,

Proc. Phys. Soc. A63 (1950) 681.

[145] K. Harada and I. Tsutsui, On the Path-Integral Quantization of Anomalous

Gauge Theories, Phys. Lett. B183 (1987) 311.

[146] H. Hata, Construction of the Quantum Action for Path Integral Quantization
of String Field Theory, Nucl. Phys. B339 (1990) 663.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 182

[147] H. Hata, “Theory of Theories” Approach to String Theory, Phys. Rev. D50
(1994) 4079.

[148] H. Hata, T. Kugo and N. Ohta, Skew-Symmetric Tensor Gauge Field Theory
Dynamically Realized in the QCD U(1) Channel, Nucl. Phys. B178 (1981)
527.

[149] H. Hata and B. Zwiebach, Developing the Covariant Batalin-Vilkovisky Ap-

proach to String Field Theory, Ann. Phys. 229 (1994) 177.

[150] M. Henneaux, Hamiltonian Form of the Path Integral for Theories with a Gauge
Freedom, Phys. Rep. 126 (1985) 1.

[151] M. Henneaux, On the Algebraic Structure of the BRST Symmetry, Lectures

given at Banff Summer School in Theoretical Physics on Physics, Geometry
and Topology, in Banff NATO ASI 1989, p.81-104.

[152] M. Henneaux, Lectures on the Antifield-BRST Formalism for Gauge Theories,

Nucl. Phys. B (Proc. Suppl.) 18A (1990) 47.

[153] M. Henneaux, Elimination of the Auxiliary Fields in the Antifield Formalism,

Phys. Lett. B238 (1990) 299.

[154] M. Henneaux, Spacetime Locality of the BRST Formalism, Comm. Math. Phys.
140 (1991) 1.

[155] M. Henneaux, Geometric Interpretation of the Quantum Master Equation in the

BRST Anti-BRST Formalism, Phys. Lett. B282 (1992) 372.

[156] M. Henneaux, Remarks on the Renormalization of Gauge Invariant Operators in

Yang-Mills Theory, Phys. Lett. B313 (1993) 35, (Erratum Phys. Lett. B316
(1993) 633).

[157] M. Henneaux and C. Teitelboim, Quantization of Gauge Systems, (Princeton

University Press, Princeton, 1992).

[158] P. Horava, Space Time Diffeomorphism and Topological W∞ in Two Dimen-

sional Topological String Theory, Nucl. Phys. B414 (1994) 485.

[159] P. S. Howe, U. Lindström and P. White, Anomalies and Renormalization in the

BRST-BV Framework, Phys. Lett. B246 (1990) 430.

[160] C. M. Hull, The BRST and Anti-BRST Invariant Quantization of General

Gauge Theories, Mod. Phys. Lett. A5 (1990) 1871.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 183

[161] C. M. Hull, Matter-Dependent W -Gravity Anomalies of Non-Linearly Realised

Symmetries, Phys. Lett. B265 (1991) 347.

[162] C. M. Hull, W Gravity Anomalies with Ghost Loops and Background Charges,
Int. J. Mod. Phys. A8 (1993) 2419.

[163] S. Hwang, Covariant Quantization of the String in Dimensions D ≤ 26 Using

a Becchi-Rouet-Stora Formulation, Phys. Rev. D28 (1983) 2614.

[164] H. Ikemori, Extended Form Method of Antifield BRST Formalism for Topological
Quantum Field Theories, Clas. Quantum Grav. 10, (1993) 233.

[165] C. Itzykson and J. B. Zuber, Quantum Field Theory, (McGraw-Hill, New York,
1980).

[166] R. Jackiw and R. Rajaraman, Vector-Meson Mass Generation by Chiral Anoma-

lies, Phys. Rev. Lett. 54 (1985) 1219.

[167] S. D. Joglekar and B. W. Lee, Gauge Theory of Renormalization of Gauge In-

variant Operators, Ann. Phys. 97 (1976) 160.

[168] M. Kaku, Introduction to Superstrings, (Springer-Verlag, New York, 1988).

[169] M. Kaku, Geometrical Derivation of String Field Theory from First Principles:
Closed Strings and Modular Invariance, Phys. Rev. D38 (1988) 3052;
Nonpolynomial Closed-String Field Theory, Phys. Rev. D41 (1990) 3734.

[170] M. Kaku and J. Lykken, Modular Invariant Closed String Field Theory,
Phys. Rev. D38 (1988) 3067.

[171] M. Kalb and P. Ramond, Classical Direct Interstring Action, Phys. Rev. D9
(1974) 2273.

[172] R. E. Kallosh, Modified Rules in Supergravity, Nucl. Phys. B141 (1978) 141.

[173] R. Kallosh, Super-Poincaré Covariant Quantized GS Heterotic String as a Free

Conformal Theory, Phys. Lett. B224 (1989) 273.

[174] M. Kato and K. Ogawa, Covariant Quantization of String Based on BRS In-
variance, Nucl. Phys. B212 (1983) 443.

[175] O. M. Khudaverdian, Geometry of Superspace with Even and Odd Brackets,

J. Math. Phys. 32 (1991) 1934.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 184

[176] O. M. Khudaverdian and A. P. Nersesian, On the Geometry of the Batalin-

Vilkovisky Formalism, Mod. Phys. Lett. A8 (1993) 2377.

[177] T. W. Kibble, Lorentz Invariance and the Gravitational Field, J. Math. Phys. 2
(1961) 212.

[178] T. Kimura, Antisymmetric Tensor Gauge Field in General Covariant Gauges,

Prog. Theo. Phys. 64 (1980) 357;
Quantum Theory of Antisymmetric Higher Rank Tensor Gauge Field in Higher
Rank Dimensional Space-Time, Prog. Theo. Phys. 65 (1981) 338.

[179] J. L. Koszul, Sur un Type d’Algébres Differéntielles en Raport avec la Trans-

gression, Bull. Soc. Math. France 78 (1950) 5.

[180] T. Kugo, H. Kunitomo and K. Suehiro, Non-Polynomial Closed String Field

Theory, Phys. Lett. B226 (1989) 48.

[181] T. Kugo and I. Ojima, Manifestly Covariant Canonical Formalism of Yang-Mills

Theories, Physical State Subsidiary Conditions and Physical S-Matrix Unitar-
ity, Phys. Lett. B73 (1978) 459.

[182] T. Kugo and I. Ojima, Local Covariant Operator Formalism of Non-Abelian

Gauge Theories and Quark Confinement Problem, Prog. Theor. Phys. Suppl. 66
(1979) 1.

[183] T. Kugo, and K. Suehiro, Nonpolynomial Closed String Field Theory: Action
and Gauge Invariance, Nucl. Phys. B337 (1990) 343.

[184] T. Kugo and S. Uehara, General Procedure of Gauge Fixing Based on BRS In-
variance Principle, Nucl. Phys. B197 (1982) 378.

[185] J. M. F. Labastida and M. Pernici, A Gauge Invariant Action in Topological

Quantum Field Theory, Phys. Lett. B212 (1988) 56.

[186] P. M. Lavrov and I. V. Tyutin, Effective Action in Gauge Theories of a General

Form, Sov. J. Nucl. Phys. 41 (1985) 1049.

[187] B. W. Lee, Gauge Theories, in Methods in Field Theory, Les Houches Summer
School 1975, edited by R. Balian and J. Zinn-Justin (North-Holland, Amster-
dam, 1976).

[188] D. A. Leites, Introduction to the Theory of Supermanifolds, Russ. Math. Surv. 35

(1980) 1.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 185

[189] B. H. Lian and G. J. Zuckerman, New Perspectives on the BRST-Algebraic

Structure of String Theory, Comm. Math. Phys. 154 (1993) 613.

[190] U. Lindström, M. Roček, W. Siegel, P. van Nieuwenhuizen and A. E. van de Ven,

Lorentz-Covariant Quantization of the Superparticle, Phys. Lett. B224 (1989)
285;
Gauge-Fixing Redundant Symmetries in the Superparticle, Phys. Lett. B228
(1989) 53.

[191] C. Lucchesi, O. Piguet and S. P. Sorella, Renormalization and Finiteness of

Topological BF Theories, Nucl. Phys. B395 (1993) 325.

[192] J. Lykken and S. Raby, Non-Commutative Geometry and the Closed Bosonic
String, Nucl. Phys. B278 (1986) 256.

[193] N. Maggiore and S. P. Sorella, Perturbation Theory for Antisymmetric Tensor

Fields in Four-Dimensions, Int. J. Mod. Phys. A8 (1993) 929.

[194] R. Marnelius, General State Spaces for BRST Quantizations, Nucl. Phys. B391
(1993) 621.

[195] R. Mohayee, C. N. Pope, K. S. Stelle and K. W. Xu, Canonical BRST Quantiza-

tion of World Sheet Gravities, Texas A&M preprint CTP-TAMU-18-94, (April,
1994), hep-th/9404170.

[196] N. Nakanishi and I. Ojima, Covariant Operator Formalism of Gauge Theories

and Quantum Gravity, World Scientific Lecture Notes, Vol. 27 (1990).

[197] Y. Nambu, Duality and Hadrodynamics, Lectures at the Copenhagen Summer

Symposium, (1970), published in Strings, Lattice Gauge Theory, High Energy
Phenomenology, Proc. of the Winter School Panchgani, edited by V. Singh and
S. R. Wadia, (World Scientific, 1987).

[198] A. P. Nersesian, On Geometry of Supermanifolds with Odd and Even Kahlerian

Structures, Theor. Math. Phys. 96 (1993) 866.

[199] N. K. Nielsen, Ghost Counting in Supergravity, Nucl. Phys. B140 (1978) 494.

[200] N. K. Nielsen, BRS Invariance of Supergravity in a Gauge Involving an Extra

Ghost, Phys. Lett. 103B (1981) 197.

[201] Kh. S. Nirov and A. V. Razumov, Field-Antifield and BFV Formalisms for
Quadratic Systems with Open Gauge Algebras, Int. J. Mod. Phys. A7 (1992)
5719.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 186

[202] I. Ojima, Another BRS Transformation, Prog. Theo. Phys. 64 (1981) 625.

[203] C. Ordóñez, J. Parı́s, J. M. Pons and R. Toldrà, BV Analysis for Covariant and
Noncovariant Actions, Phys. Rev. D48 (1993) 3818.

[204] C. Ordóñez, J. M. Pons and R. Toldrà, Unitarity and Batalin-Vilkovisky Path

Integral Quantization for Gauge Theories, Phys. Lett. B302 (1993) 423.

[205] F. R. Ore, Jr. and P. van Nieuwenhuizen, Local BRST Symmetry and the Ge-
ometry of Gauge-Fixing, Nucl. Phys. B204 (1982) 317.

[206] J. Parı́s, Anomalies i Unitarietat en el Formalisme de Camps i Anticamps, The-

sis, Universitat de Barcelona (1992).

[207] J. Parı́s, Faddeev-Popov Method for Anomalous Quasigroups, Phys. Lett. B300
(1993) 104.

[208] J. E. Paton and H.-M. Chan, Generalized Veneziano Model with Isospin,
Nucl. Phys. B10 (1969) 516.

[209] M. Penkava and A. Schwarz, On Some Algebraic Structure Arising in String

Theory, UC Davis preprint PRINT-93-0001, (Dec., 1992), hep-th/9212072.

[210] A. M. Polyakov, Quantum Geometry of Bosonic Strings, Phys. Lett. 103B

(1981) 207.

[211] M. Roček, W. Siegel, P. van Nieuwenhuizen and A. E. van de Ven, Covariant

Superparticle Quantization in a Super Maxwell Background, Phys. Lett. B227
(1989) 87.

[212] M. Saadi and B. Zwiebach, Closed String Theory from Polyhedra,

Ann. Phys. 192 (1989) 213.

[213] A. Salam, Weak and Electromagnetic Interactions, in Elementary Particle The-

ory, edited by N. Svartholm, (Almquist and Forlag, Stockholm, 1968), p.367-
377.

[214] S. Samuel, The Physical and Ghost Vertices in Witten’s String Field Theory,
Phys. Lett. B181 (1986) 255.

[215] S. Samuel, Introduction to String Field Theory, in Strings and Superstrings,

edited by J. P. Mittelbrunn, M. Ramón-Medrano and G. S. Rodero, El Escorial,
Spain, June, 1987, (World Scientific, Singapore, 1988);
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 187

Also see Appendices A-C of Mathematical Formulation of Witten’s Superstring

Field Theory, Nucl. Phys. B296 (1988) 187.

[216] S. Samuel, Covariant Off-Shell String Amplitudes, Nucl. Phys. B308 (1988)
285;
Off-Shell String Physics from Conformal Field Theory, Nucl. Phys. B308
(1988) 317.

[217] S. Samuel, Solving the Open Bosonic String in Perturbation Theory, Nucl. Phys.
B341 (1990) 513.

[218] A. Schwarz, Geometry of Batalin-Vilkovisky Quantization, Comm. Math. Phys.

155 (1993) 249.

[219] A. Schwarz, Semiclassical Approximation in Batalin-Vilkovisky Formalism,

Comm. Math. Phys. 158 (1993) 373.

[220] J. H. Schwarz, Superstring Theory, Phys. Rep. C89 (1982) 223.

[221] J. Schwinger, On the Green’s Function of Quantized Fields. I.,

Proc. Nat. Acad. Sci. 37 (1951) 452;
On the Green’s Function of Quantized Fields. II., Proc. Nat. Acad. Sci. 37 (1951)
455.

[222] N. Seiberg and S. Shenker, A Note on Background (In-)dependence, Phys. Rev.

D45 (1992) 4581.

[223] A. Sen, On the Background Independence of String Field Theory, Nucl. Phys.
B345 (1990) 551;
On the Background Independence of String Field Theory II. Analysis of On-
shell S-Matrix Elements, Nucl. Phys. B347 (1990) 270;
On the Background Independence of String Field Theory III. Explicit Field
Redefinitions, Nucl. Phys. B391 (1993) 550.

[224] A. Sen and B. Zwiebach, A Proof of Local Background Independence of Classical

Closed String Field Theory, Nucl. Phys. B414 (1994) 649.

[225] A. Sen and B. Zwiebach, Quantum Background Independence of Closed String

Field Theory, Nucl. Phys. B414 (1994) 485.

[226] A. Sen and B. Zwiebach, A Note on Gauge Transformations in Batalin-

Vilkovisky Theory, Phys. Lett. B320 (1994) 29.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 188

[227] E. Sezgin, Aspects of Kappa Symmetry, Texas A&M preprint CTP-TAMU-28-

93, (Oct., 1993), hep-th/9310126.

[228] W. Siegel, Hidden Ghosts, Phys. Lett. 93B (1980) 170.

[229] W. Siegel, Covariantly Second-Quantized String, Phys. Lett. 142B (1984) 276;
Covariantly Second-Quantized String II, Phys. Lett. 151B (1985) 391;
Covariantly Second-Quantized String III, Phys. Lett. 151B (1985) 396.

[230] W. Siegel, Batalin-Vilkovisky from Hamiltonian BRST, Int. J. Mod. Phys. A4

(1989) 3951.

[231] W. Siegel, Relation Between Batalin-Vilkovisky and First Quantized Style

BRST, Int. J. Mod. Phys. A4 (1989) 3705.

[232] W. Siegel, Lorentz Covariant Gauges for Green-Schwarz Superstrings, in Strings

89, eds. R. Arnowitt et al. (World Scientific, Singapore, 1990).

[233] A. A. Slavnov, Ward Identities in Gauge Theories, Theor. Math. Phys. 10

(1972) 99.

[234] A. A. Slavnov, Physical Unitarity in the BRST Approach, Phys. Lett. B217
(1989) 91.

[235] A. A. Slavnov, BRST-Quantization and Unitarity, Nucl. Phys. B (Proc. Suppl.)

B15 (1990) 107.

[236] M. F. Sohnius, Soft Gauge Algebras, Z. Phys. C18 (1983) 229.

[237] J. Stasheff, Closed String Field Theory, Strong Homotopy Lie Algebras and the
Operad Actions of Moduli Space, Univ. of North Carolina preprint UNC-MATH-
93-1, (April, 1993), hep-th/9304061.

[238] G. Sterman, P. K. Townsend and P. van Nieuwenhuizen, Unitarity, Ward Iden-

tities, and New Quantization Rules of Supergravity, Phys. Rev. D17 (1978)
1501.

[239] Y. Takahashi, On the Generalized Ward Identity, Nuovo Cim. 6 (1957) 371.

[240] J. Tate, Homology of Noetherian Rings and Local Rings, Illinois J. Math. 1
(1957) 14.

[241] J. C. Taylor, Ward Identities and Charge Renormalization of the Yang-Mills

Field, Nucl. Phys. B33 (1971) 436.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 189

[242] J. Thierry-Mieg, BRS Structure of the Antisymmetric Tensor Gauge Theories,

Nucl. Phys. B335 (1990) 334.

[243] G. t’Hooft, Renormalization of Massless Yang-Mills Fields, Nucl. Phys. B33

(1971) 173.

[244] G. t’Hooft, Renormalizable Lagrangians for Massive Yang-Mills Fields,

Nucl. Phys. B35 (1971) 167.

[245] G. t’Hooft and M. Veltman, Diagrammar, in Louvain 1973, Particle Interactions

at Very High energies, Part B, (New York, 1973), p.177-322.

[246] C. B. Thorn, Perturbation Theory for Quantized String Fields, Nucl. Phys.
B287 (1987) 61.

[247] C. B. Thorn, String Field Theory, Phys. Rep. 175 (1989) 1.

[248] M. Tonin, Covariant Quantization and Anomalies of the G–S Heterotic σ–

Model, Int. J. Mod. Phys. A6 (1991) 315.

[249] M. Tonin, Dimensional Regularization and Anomalies in Chiral Gauge Theories,

talk presented at Topical Workshop on Nonperturbative Aspect of Chiral Gauge
Theories, Rome, Italy, (March, 1992), in Rome Chiral Gauge, p.137-151.

[250] P. K. Townsend, Gauge Invariance for Spin 1/2, Phys. Lett. 90B (1980) 275.

[251] W. Troost, P. van Nieuwenhuizen and A. Van Proeyen, Anomalies and the
Batalin-Vilkovisky Lagrangian Formalism, Nucl. Phys. B333 (1990) 727.

[252] W. Troost and A. Van Proeyen, Regularization and the BV Formalism, in

Proc. of Strings 93, Berkeley, (May, 1993).

[253] W. Troost and A. Van Proeyen, An Introduction to Batalin-Vilkovisky La-

grangian Quantisation, Leuven Notes in Math. Theor. Phys., in preparation.

[254] I. V. Tyutin, Gauge Invariance in Field Theory and Statistical Mechanics, Lebe-
dev preprint no 39 (1975), unpublished.

[255] I. V. Tyutin and Sh. Shvartsman, BRST Operator and Open Algebra, Phys. Lett.
B169 (1986) 225.

[256] R. Utiyama, Invariant Theoretical Interpretation of Interaction, Phys. Rev. 101

(1956) 1597.
J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 190

[257] S. Vandoren and A. Van Proeyen, Simplifications in Lagrangian BV Quantiza-

tion Exemplified by the Anomalies of Chiral W3 Gravity, Nucl. Phys. B411
(1994) 257.

[258] P. van Nieuwenhuizen, Supergravity, Phys. Rep. 68 (1981) 189.

[259] A. Van Proeyen, Batalin-Vilkovisky Lagrangian Quantization, talk presented in

Proc. of Strings and Symmetries Stony Brook, Strings: Stony Brook, 1991,
p.388-406.

[260] M. A. Vasil’ev, “Gauge” Form of Description of Massless Fields With Arbitrary

Spin, Sov. J. Nucl. Phys. 32 (1980) 439.

[261] E. Verlinde, The Master Equation of String Theory, Nucl. Phys. B381 (1992)
141.

[262] B. L. Voronov, P. M. Lavrov and I. V. Tyutin, Canonical Transformations and

Gauge Dependence in General Gauge Theories, Sov. J. Nucl. Phys. 36 (1982)
292.

[263] B. L. Voronov and I. V. Tyutin, Formulation of Gauge Theories of General

Form. I, Theor. Math. Phys. 50 (1982) 218.

[264] B. L. Voronov and I. V. Tyutin, Formulation of Gauge Theories of General

Form. II. Gauge-Invariant Renormalizability and Renormalization Structure,
Theor. Math. Phys. 50 (1982) 628.

[265] J. C. Ward, An Identity in Quantum Electrodynamics, Phys. Rev. 78 (1950) 182.

[266] S. Weinberg, A Model of Leptons, Phys. Rev. Lett. 19 (1967) 1264.

[267] J. Wess and J. Bagger, Introduction to Supersymmetry, (Princeton Univ. Press,

Princeton, NJ, 1983).

[268] J. Wess and B. Zumino, Consequences of Anomalous Ward Identities,

Phys. Lett. 37B (1971) 95.

[269] P. L. White, Anomaly Consistency Conditions to All Orders, Phys. Lett. B284
(1992) 55.

[270] E. Witten, Non-Commutative Geometry and String Field Theory, Nucl. Phys.
B268 (1986) 253.

[271] E. Witten, Open Superstrings, Nucl. Phys. B276 (1986) 291.

J. Gomis, J. Parı́s and S. Samuel — Antibracket, Antifields and . . . 191

[272] E. Witten, Topological Quantum Field Theory, Comm. Math. Phys. 117 (1988)
353.

[273] E. Witten, A Note on the Antibracket Formalism, Mod. Phys. Lett. A5 (1990)
487.

[274] E. Witten, On Background Independent Open String Field Theory, Phys. Rev.
D46 (1992) 5467.

[275] E. Witten, Some Computations in Background Independent Off-Shell String

Field Theory, Phys. Rev. D47 (1993) 5467.

[276] C. N. Yang and R. L. Mills, Considerations of Isotopic Spin and Isotopic Gauge
Invariance, Phys. Rev. 96 (1954) 191.

[277] J. Zinn-Justin, Renormalization of Gauge Theories, in Trends in Elementary

Particle Theory, edited by H. Rollnik and K. Dietz, Lecture Notes in Physics,
Vol 37, (Springer-Verlag, Berlin, 1975).

[278] J. Zinn-Justin, Quantum Field Theory and Critical Phenomena, (Oxford Uni-
versity Press, Oxford, England, 1989).

[279] B. Zumino, Chiral Anomalies and Differential Geometry, in Relativity, Groups

and Topology, edited by B. S. deWitt and R. Stora, (North-Holland, Amster-
dam, 1983), and references therein.

[280] B. Zwiebach, Quantum Closed Strings from Minimal Area, Int. J. Mod. Phys.
A5 (1990) 275;
Consistency of Closed String Polyhedra from Minimal Area, Phys. Lett. B241
(1990) 343;
How Covariant Closed String Theory Solves a Minimal Area Problem,
Comm. Math. Phys. 136 (1991) 83.

[281] B. Zwiebach, Closed String Field Theory: An Introduction, MIT preprint MIT-
CTP-2206, (May, 1993), hep-th/9305026.

[282] B. Zwiebach, Closed String Field Theory: Quantum Action and the B-V Master
Equation, Nucl. Phys. B390 (1993) 33.