INTRODUCTION TO BRS SYMMETRY

C. BECCHI
Dipartimento di Fisica, Università di Genova,
Istituto Nazionale di Fisica Nucleare, Sezione di Genova,
via Dodecaneso 33, 16146 Genova (Italy)

Abstract
This paper contains a revised version of the lecture notes of a short course
on the quantization of gauge theories. Starting from a sketchy review of
scattering theory, the paper describes the lines of the BRST-Faddeev-Popov
quantization considering the problem of a non-perturbative extension of this
method. The connection between the Slavnov-Taylor identity and the uni-
tarity of the S-matrix is also discussed.

1
Lectures given at the ETH, Zurich, May 22-24, 1996 - December 2008 revised version

1
1 Introduction

These lectures begin recalling some general results of scattering theory [1].
The reduction formulae for the S-matrix are given in terms of the Feynman
functional in the case of a massive field theory. Then the analysis comes
to gauge theories for which the concept of gauge orbit is introduced. The
Faddeev-Popov definition [2] of a finite functional measure is given. The
BRST external differential operator along the orbits is introduced together
with the full BRS operator [3]. The gauge algebra of the infinitesimal gauge
transformations is briefly discussed. Assuming the existence of a global gauge
fixing (no Gribov ambiguity [7]) the Slavnov-Taylor identity and the gauge
fixing independence of the theory is deduced from the BRS invariance of the
functional measure. The extension of the Faddeev-Popov formula to the case
of Gribov ambiguities is briefly discussed together with that of the Slavnov-
Taylor identity. The Slavnov-Taylor identity is then translated in terms of
the proper functional ( the effective action ) and the extension of the method
to the case of gauge algebras that are closed only modulo the field equations
is discussed. Limiting for simplicity the study to the case of massive fields
the Slavnov-Taylor identity is applied to the two-point functions, this leads
to the introduction of the BRS symmetry for the asymptotic fields ( in the
form of Kugo and Ojima [14] ) and to the proof the existence of a physical
Hilbert space in which the S-matrix is unitary.

2 The S-matrix

In quantum field theory [1] the scattering amplitudes are computed by means
of the reduction formula. This can be simply written using the Feynman
functional generator of the theory that is defined according:
 R 
iZc d4 xφ(x)J(x)
Z[j] ≡ e =< Ω, T ei Ω> . (1)

where φ and j in general label a set of quantized fields and corresponding

sources. Zc is the connected functional. The Feynman functional is computed
by means of the formula:
Z R
d4 xφ(x)J(x)
Z[j] = dµ ei , (2)

1
in terms of the functional measure dµ of the theory that is deduced from its
bare action by the heuristic relation; dµ = x dφ(x)eiS(φ) .
Q

Computing the two-point function according:

δ2
Zc |J=0 ≡ ∆ij (x) , (3)
δJi (x)δJj (0)
and excluding for simplicity the presence of massless fields, we can separate
from ∆ the asymptotic propagator ∆as :
XZ dp eipx
ij
∆ (x) = 4 2 2
ζλij (p) + Rij (x) ≡ ∆ij ij
as (x) + R (x) , (4)
λ (2π) mλ − p − i0+

where the Fourier transform of R has no pole in p2 . It is clear that the

asymptotic propagator is by no means unique since ζλij is defined up to a
polynomial in p2 vanishing at m2λ ; however this lack of uniqueness does not
affect the S matrix. Then one introduces the asymptotic free fields φin with
the (anti-)commutation relations:
h
i(+) j(−)
i XZ dp ipx 0
φin (x), φin (0) = 4
e θ(p )δ(p2 − m2λ )ζλij (p) , (5)
±
λ (2π)

and the asymptotic wave operator which satisfies:

Kij (∂)φiin (x) = 0 , (6)

and
˜ kj (p)|p2 =m2 = δij
Kik (−ip)∆ (7)
λ

for any λ. It should be clear that the asymptotic wave operator is by no

means unique in much the same way as the matrix ζ.
Then one computes the S matrix of the theory according:
R
d4 xφi (x)Kij (∂) δJ δ(x)
S =: e j : Z|J=0 ≡: eΣ : Z|J=0 . (8)

3 Gauge invariance and BRS symmetry

Now we come to the quantization of gauge theories. In this course we shall

disregard the crucial problem of the explicit non-perturbative construction

2
of the theory, limiting our analysis to the formal and symmetry aspects that
should simplify the construction and characterize the solution of the full
quantum theory. We shall be often concerned with the functional measure
of the theory; avoiding any consideration of its actual definition we shall
indifferently pass from the Minkowskian form

dφ(x)eiS(φ)
Y
dµ = (9)
x

to the Euclidean one

dφ(x)e−S(φ) .
Y
dµ = (10)
x

Furthermore, in order to simplify the notation we shall merge all the labels
of the fields into a single index including the space-time variable. Assuming
the usual convention of summation over repeated indices we shall also often
omit the integration symbol. However one should keep firmly in mind that
the fields are local variables and that locality is considered to play a crucial
role in field theory. For many reasons we shall also avoid discussing many
mathematical aspects that should bring our analysis too far from its purposes.
Let us call F0 the field space, that is the configuration space upon which
the gauge theory is constructed. In a gauge theory F0 is fibered by the
gauge orbits O that is the set of gauge transforms of a given configuration.
Considering the infinitesimal transformations and translating everything in
differential geometry terms ( we are freely following e.g. [4]), we are given a
system of partial differential operators {X} on F0 , that we shall label with
the index I, and that in any point of F0 define a system of tangent vectors
to the corresponding orbit. Denoting the generic field coordinate in F0 by
the φα , we can write these operators in the form

XI = PIα (φ)∂φα . (11)

In the following we shall systematically use the symbol φα , or, more ex-
plicitly φα (x), for the bosonic gauge and matter fields and we shall mention
only occasionally the spinor fields. The bosonic gauge and matter field com-
ponents are chosen real and hence they correspond to Hermitian operator-
valued distributions.
The explicit form of Eq.(11) in a pure non abelian case is:
δ K J δ
XI (x) = ∂µ − gfIJ Aµ (x) . (12)
δAIµ (x) δAK
µ (x)

3
The system {X} is often called the differential system of the orbits, whose
very existence implies:
K
[XI , XJ ] = CIJ (φ)XK , (13)

that is the complete integrability of {X}. In the standard situation the

K
algebra (13) is a Lie algebra, the structure functions CIJ are constants. As
a matter of fact, having merged the space-time variables with the discrete
indices, what we call a Lie algebra is an infinite Lie algebra. Putting the
space-time variables into evidence, that is, replacing above I, J and K with
K
the pairs (I, x), (J, y) and (K, z), CIJ decomposes into factors such as the
K
constant fIJ and the distribution δ(x − y)δ(x − z). If the discrete indices
run in the set I = 1, . . . , G we shall call G the dimension of the gauge Lie
algebra.
We shall see in the following how (13) can be weakened restricting the
integrability condition to the ”mass shell”, that is modulo the field equations.
We also assume that a X-invariant measure is uniquely defined up to an
orbit-independent normalization constant. This we shall call the vacuum
invariant measure, the measure which is associated with the vacuum state
of the theory. In general one is interested in the vacuum correlators of local
observables; these correspond to the integrals of a different class of invariant
measures that can be written as the product of the vacuum measure times
gauge invariant functionals depending on the field variables corresponding to
a suitably localized space-time domain. A generic invariant measure will be
assumed to belong to this class.
The vacuum functional measure is constant over the orbits O; in general
this makes the functional measure of F0 non integrable and the Feynman
functional ill defined. This difficulty is cured by the Faddeev-Popov trick. In
order to recall it conveniently let us assume that, even if this is in general not
the case, that the original field variables trivialize the fibration; that is let us
assume that the set of fields {φ} is decomposed according {ξ} and {η} where
{ξ} are constant along the orbits and {η} are ”vertical” coordinates along
the orbits. Then it is natural to make the measure integrable by multiplying
it by an integrable functional of η whose integral over O corresponding to the
above mentioned X-invariant measure shoould be independent of the orbit
(of ξ). One often considers the invariant Dirac δinv [η − η̄]:

δinv [η − η̄] ≡ δ [η − η̄] det|XI η J | , (14)

4
but more generally one can consider its convolution with a suitable η̄-functio-
nal.
Notice that the determinant appears in (14) since the action of a gauge
transformation does not correspond to a Euclidean transformation on the η
variables. The Faddeev-Popov measure [2] is obtained by the substitution:

dµ → dµ δinv [η − η̄] , (15)

The invariant Dirac measure can be easily written as a functional Fourier

transform. Introducing two sets of Grassmann variables {ω I } and {ω̄J } that
can be simply identified with the generators of an exterior algebra and the
corresponding derivatives that we label by {∂ωI } and by {∂ω̄J }, one introduces
the Berezin integral: Z
1
dω I ≡ √ ∂ω I , (16)
2πi
for ω and an analogous definition for ω̄. Then using the so called Nakanishi-
Lautrup multipliers {bJ } one can reproduce the right-hand side of (14) in the
form: Z Y
dω̄K ei[bI η −ω̄I ω XJ η ] .
I J I
dbI dω J
Y Y
(17)
This formula can be interpreted as an enlargement of the field space F0
with the addition of a set of ordinary fields corresponding to the Nakanishi-
Lautrup multipliers and of two sets of anticommuting fields corresponding to
the exterior algebra generators. We call FC the new space. We also introduce
the measure on FC :
Y Y Y
dµC ≡ dµ dbI dωJ dω̄K . (18)
I J K

Looking now into the details of (17), we see that the differential operator
ω I XI appearing in the exponent can be replaced by its minimal nilpotent
extension :
1
dV = ω I XI − ω I ω J CIJ
K
(φ)∂ωK , (19)
2
This operator, that is often called the BRST operator [3], is nilpotent due to
(13) and to the corresponding Jacobi identity. That is:

  
 ω I ω J XI XJ − 12 CIJ
K
(φ)XK = 0;
d2V = 0 ↔   (20)
 ω I ω J ω K CIJ
M L
(φ)CM L
K (φ) − XI CJK (φ) = 0.

5
 Identifying the system {ω} with that of the X-left-invariant forms we
can interpret the differential operator dV as the vertical exterior differential
operator on F0 , that is, with the operator on F0 that in any point is identified
with the exterior differential operator on the orbit (see Appendix A).
Let us now come back to the trivializing coordinates, it is clear that these
exist globally only in very special cases, in particular when the corresponding
fibration is trivial. However in order to define a finite measure through
Eq.(15) it is sufficient to identify a global section of F0 , if it exists, intersecting
every orbit in a single point. This condition is equivalent to the existence
of a system of local functionals {Ψ(φ)} that we shall continue to label with
the index I, for which the Jacobian det|XI ΨJ | does not vanish in the points
where Ψ = Ψ̄ for some Ψ̄.
Assuming this condition, we shall replace in the exponent in (17) the
coordinate η I by a generic functional ΨI (φ) writing the exponent as:
h   i
iSGF = i bI ΨI − Ψ̄I − ω̄I dV ΨI , (21)
The above formula gives the definition of the gauge fixing action SGF .
Eq.(21) can be translated into a simpler form introducing a new exterior
derivative s acting on the algebra generated by ω and ω̄ whose action on φ
and ω coincides with that of dV and:
sω̄ = b , sb = 0 , (22)
and hence
s = dV + bI ∂ω̄I . (23)
It it clear that s is nilpotent and that (21) is written:
h  i
SGF = s ω̄I ΨI − Ψ̄I , (24)
it is also obvious that s commutes with the physical functional measure dµ.
A further generalization of the measure, that includes also the convolutions
of (24) with generic functionals of Ψ̄, is obtained extending the choice of Ψ
to b, ω and ω̄-dependent
 local functionals. In the following we shall replace
I I
ω̄I Ψ − Ψ̄ with a generic functional Θ carrying the same quantum num-
bers. In the standard situation Θ is a strictly local quadratic functional of b,
that is, the space-time integral of a second order polynomial in b, independent
of its derivatives; therefore b is an auxiliary field. However there are mod-
els, in particular in supergravity, in which [5] b corresponds to propagating
degrees of freedom that play the role of extra ghosts.

6
 The most frequently met gauge choice in the case of renormalizable the-
ories is the linear choice in which the gauge fixing function in Eq. (21)
is a linear function. Showing explicitly the dependence on the space-time
variables one sets:
" #
Z
ξ
ΨI = dx VαI (∂)φα + δ IJ bJ , (25)
2
where δ is the Kronecker symbol and the matrix V is a real linear function
of the space-time derivatives and has maximal rank, that is, rank equal to
the number of independent components of the b fields. Furthermore Ψ̄ = 0.
It follows that:
" #
Z
I α ξ IJ I α J
SGF = dx bI (Vα (∂)φ + δ bJ ) − ω̄I Vα PJ (φ)ω . (26)
2
Even within the linear gauge there are many possibilities among which one
chooses depending on the calculation purposes. The most frequent choices
are the covariant ones, typically Lorentz’s gauge in which VαI (∂)φα = kJI ∂ µ AJµ
and ’t Hooft gauges, VαI (∂)φα = kJI ∂ µ AJµ + ρIi Φi , where AIµ is a gauge vector
field and Φi is a scalar field. Among the non-covariant choice one often
meets the ligh-like axial gauge in which VαI φα = kJI nµ AJµ where n is a light-
like vector. However the covariant choices are the most convenient ones for
a general discussion.
The above mentiones condition that det |XI ΨJ | does not vanish, that
is, the condition for the gauge degrees of freedom to be completely fixed,
I
implies that det |C0,J |, the determinant of the Fourier transform of the field-
independent part of XJ VαI φα , does not vanish for a generic choice of p. In the
I
Lorentz gauge C0,J (p) = −kJI p2 and in ’t Hooft’s one C0,J
I
(p) = −kJI p2 +ρIi tij
J vj
ij
where tJ vj is the in-homogeneous part of the scalar field gauge transforma-
tion. Discussing the S-matrix unitarity we shall assume a covariant choice
I
with the further condition that C0,J (p) is real and the ghost part of the
Lagrangian is formally Hermitian together with the ghost field ω, while the
I
anti-ghost is anti-Hermitian. Since the equation det |C0,J | = 0 is an algebraic
2
equation of degree G in p whose solutions in the semi-classical approxima-
tion are the ghost masses, we assume that all these solutions are real and
positive and that the ghost particles are kinematically stable.
Thus under the standard assumption of a closed gauge algebra (13) S has
the following structure:
1
Z  
I α I J K
S = Sinv (φ) + SGF + dx γα ω PI (φ) + ζI CJK ω ω . (27)
2

7
In the case of renormalizable theories this structure is obliged by the condi-
tion that the dimension of the action should be limited by that of space-time.
A further comment on the auxiliary role of b and hence of ω̄ is here
necessary. In this study we are tacitly considering the field space FC finite
dimensional, strictly speaking this is, of course, not true since every field
corresponds to an infinite number of variables; however one assumes that
some mechanism, e.g. some kind of regularization, renders finite the number
of effective degrees of freedom. With this proviso let us consider:
Z Y Z Y P
b2
dω̄J eisΘ = lim→0 dω̄J eisΘ−
Y Y
dbI dbI K k

I J I J
Z Y Z 1 P
b2
dt s (Θ) eisΘt−
Y
= i lim→0 dbI dω̄J K k

I J 0
" Z Y Z 1 #
P
isΘt− b2
Y
= dV i lim→0 dbI dω̄J dt Θe K k , (28)
I J 0

where we have used the fact that the Berezin integral of a constant gives zero.
Eq.(28) shows that the Faddev-Popov measure corresponds to the insertion
of a dV -exact factor into the functional measure and the fields b and ω̄ are
auxiliary in the sense that they allow the explicit construction of this factor
in local terms. Furthermore we see from (28) that the resulting measure on
F0 that is: Z Y
dω K dbI dω̄J eisΘ
Y Y
dµ (29)
K I J

is an exact top form. It follows that its integral over a compact cycle, such as
a gauge group orbit of a lattice gauge theory, vanishes [6]2 . This is due to the
fact that on a cycle the gauge fixing equation Ψ = Ψ̄ has an even number of
solutions whose contributions to the above measure cancel pairwise. However
it should be clearly kept in mind that according to the Faddeev-Popov pre-
scription the functional integral should not cover the whole orbits but only a
compact subset of every orbit containing a single solution of the gauge fixing
equation. To be explicit let us consider the extreme example in which F0
reduces to a circle, a single U (1) gauge orbit. Choosing Θ = ω̄ sin ϕ, setting
s = iω∂ϕ + b∂ω̄ and integrating over the whole space, one gets
I Z I
ibsin ϕ+ω̄ωcos ϕ
dϕ dbdωdω̄e = −i dϕδ (sin ϕ) cos ϕ = 0 , (30)
2
I thank M.Testa for calling my attention to this reference

8
while with the actual prescription, one has:
Z  Z
i dϕ dbdωdω̄eibsin ϕ+ω̄ωcos ϕ
=1. (31)
−

To conclude this section let us remember [7] that in the case of covariant
and local gauge choices the condition that Ψ = Ψ̄ defines a global section of
the orbit space does not hold true, the situation is less clear for the so called
axial gauges, that however suffer even worst diseases [10]. We shall see how
this difficulty can be overcome in the situation in which F0 can be divided
into a system of cells Ua in which one can find for every cell a Ψa defining a
section in the cell.

4 The Slavnov-Taylor identity

The particular structure of the functional measure allows an immediate proof,

up to renormalization effects, of the Slavnov-Taylor (S-T) identity. That is:
for any measurable functional Ξ :
Z
dµC eiSGF s Ξ = 0 . (32)

Indeed, using the same arguments as for (28), we get:

Z Z h i
dµC eiSGF s Ξ = dµC s eiSGF Ξ
Z Z Y
dω I dV dω̄K eiSGF Ξ = 0 ,
Y Y
= dµ dbJ (33)
I J K

since the last expression apparently corresponds to an exact top form whose
support, according to the general prescription, is contained in the integration
domain, and hence it vanishes on the boundaries of this domain. Considering
the extreme example given in the last section, one has for any continuous
A(ϕ):
Z  Z
i dϕ db dωdω̄eibsin ϕ+ω̄ωcos ϕ
s [ω̄A (ϕ)]
−
Z  Z
=i dϕ db dωdω̄eibsin ϕ+ω̄ωcos ϕ
[bA (ϕ) − iω̄ωA0 (ϕ)]
−

9
 Z 
dϕ Z
= db eibsin ϕ+ω̄ωcos ϕ [bA (ϕ) cos ϕ − iA0 (ϕ)]
− 2π
Z 
dϕ d Z
= −i db eibsin ϕ A (ϕ) = 0 , (34)
− 2π dϕ

since db eibsin ϕ A (ϕ) vanishes at ϕ = ±.

R

The identity (32) can be interpreted saying that all correlators between
elements of the image of s and s-invariants ones vanish. Indeed, according to
the definition given in section 3, the s-invariant functionals can be considered
to be generic local factors in the invariant measure dµ. Considering the s
operator as the natural extension of dV , the exterior derivative corresponding
to the gauge transformations, it is natural to assume as a basic principle of
gauge theories the identification of observables with s-invariant functionals.
Due to the nilpotency of s this set contains the image of s, whose elements,
however, correspond to trivial observables according to (32). Therefore the
non-trivial observables belong to the quotient space of the kernel of s versus
its image, that its to the cohomology of s.
It remains to verify that the cohomology of s is equivalent to that of
the vertical exterior differential operator dV . Indeed consider the functional
differential operator:
∆ ≡ − (bI ∂bI + ω̄I ∂ω̄I ) . (35)
Let P be the projector on the kernel of ∆, that is on the ω̄ and b-independent
functionals. It is apparent that ∆ and hence P , commute with s. Therefore
a generic functional X which is s-invariant, that is satisfying:

(dV + bI ∂ω̄I ) X = 0 , (36)

is the sum of two terms: P X and (1 − P )X, satisfying:

dV P X = 0 , s (1 − P ) X = 0 . (37)

In much the same way, an element of the image of s: Z = sY , is decomposed

according:
P Z = dV P Y , (1 − P )Z = s(1 − P )Y . (38)
Therefore the cohomology of s is the union of that of dV in the kernel of ∆ and
that of s in the kernel of P . We want to verify that this second contribution
is trivial. Indeed, consider the differential operator: ω̄I ∂bI , satisfying:

{ω̄I ∂bI , s} = −∆ , (39)

10
for s-invariant X, this yields:
sω̄I ∂bI (1−P )X = −∆(1−P )X → (1−P )X = −s∆−1 ω̄I ∂bI (1−P )X . (40)
Indeed, on account of the definition of P , ω̄∂ω (1−P )X belongs to the domain
of ∆−1 . Thus (1 − P )X belongs to the image of s.
A second consequence of (32) is the gauge fixing independence of the
correlators of observables. Indeed let us compare the expectation values
of the same s-invariant functional Ω computed with two different measures
corresponding to choices of Θ differing by δΘ. To first order in δΘ the
difference of the expectation values is given by
Z
i dµC eiSGF s (δΘ) Ω = 0 . (41)

Of course this implies the independence of the expectation values of the choice
of Θ in a certain class of measurable functionals. Even in perturbation theory
this is not enough to prove that the expectation values in a renormalizable
gauge coincide with those in a non-renormalizable one.
We now come to the problem of extending the functional measure to the
situation in which the gauge fixing is defined only locally. In general the
orbit manifold has to be divided into cells, each corresponding to a different
choice of Θ. Every cell in the orbit space corresponds to a cell Ua in F0 .
Let χa (φ) be a suitable smooth positive function with support in Ua and
such that the set {χ} is a partition of unity on the union of the supports
of the gauge-fixed measures dµC eisΘa . This is shown in the figure where the
dotted lines corresponds to the support of the measures and the circles to
the cells. Explicitly a cell will be defined giving its center, that is a special
configuration φa (background field), and defining the characteristic functions
χa according:
θ (R2 − kφ − φa k2 ) θ2 (R2 − inf c kφ − φc k2 )
χa (φ) ≡ 2 2
. (42)
b θ (R − kφ − φb k )
P

11
 where θ is a smoothed Heavyside function and kφ − φa k is the L2 norm
of the difference φ − φa . Hints about the values of Ra can be found in [8].
The BRS invariant functional measure corresponding to this local gauge
choice is given by [9]:
 n(n−1)
∞
(−1) 2  
dµC  χa eisΘa n
X X
−i sχa1 ...sχan χan+1
a n=1 n+1 A
  i
∂Θ Θa1 ...Θan+1 eisΘa1 ,...,an+1 , (43)
where we have used the definitions:
n
(−1)l+1 Θa1 ...Θ̌al ...Θan .
X
∂Θ (Θa1 ...Θan ) ≡ (44)
l=1

and:  
n
Z ∞Y n Pn
eisΘa1 ,...,an ≡ tj − 1 eis t Θ
X
dti δ  k=1 k ak . (45)
0 i=1 j=1

In Appendix B [(137)] it is proven that under the hypothesis of a finite

multiplicity of cell intersections the measure (43) satisfies the Slavnov-Taylor
identity.
The lack of gauge invariance of the characteristic functions of the cells
induces new contributions to the measure localized on the cell (regularized)
boundaries. Of course the above measure could be ill defined if the cells
would accumulate around some singularity of F0 . This could perhaps induce
instabilities of BRS symmetry in the sense of [11].
Another version of the S-T identity concerns the Feynman functional
involving the sources jα of φα , J I of bI , σ̄I of ω I and σ I of ω̄I . The new
functional is defined according:
Z R
dx[jα φα +J I bI +σ̄I ω I +σ I ω̄I ]
Z [ j, J, σ̄, σ ] ≡ dµC eiSGF ei , (46)

where we have explicitly shown the space-time integral symbol. This more
explicit form will be essential for the discussion of unitarity.
The new form of the S-T identity is a particular version of Eq¿(32), that
is: Z R
dµC eiSGF s ei dx[jα φ +J bI +σ̄I ω +σ ω̄I ] = 0 .
α I I I
(47)
It is possible, exploiting the nilpotency of s, to translate (47) into a functional
differential equation for Z; this requires the introduction of further sources

12
for the composite operators generated by the action of s on the fields. On
these are the source has to introduce the source γα for sφα and ζI for sω I .
These sources, which are often called anti-fields, appear in a further factor
in the functional measure:
R R
dx[γα sφα +ζI sω I ] dx s[γα φα −ζI ω I ]
dµC eiSGF → dµC eiSGF ei = dµC eiSGF e−i
(48)
which remains s-invariant due since s is nilpotent. Notice that the intro-
duction of the sources for the fields and their variations has enlarged the
functional exterior algebra upon which the Feynman functional is defined.
In particular σ, σ̄ and γ are odd elements of this algebra. In the following
formulae many derivatives are in fact anticommuting derivatives, this induces
some obvious changes of sign.
Now, inserting the new measure into (47) we get:
Z R R
dx[γα sφα +ζI sω I ] dx[jα φα +J I bI +σ̄I ω I +σ I ω̄I ]
dµC eiSGF ei s ei =0, (49)

that is:
Z R Z
dx[γα sφα +ζI sω I ]
h
iSGF i
dµC e e dx jβ sφb − σ̄J sω J
R
dx[jα φα +J I bI +σ̄I ω I +σ I ω̄I ]
i
−σ K bK ei =0. (50)

This is equivalent to the first order partial differential equation for the ex-
tended Feynman functional:
" #Z
Z
δ δ K δ
R
iSGF i dx[γα sφα +ζI sω I ]
dx jβ − σ̄J −σ dµC e e
δγβ δζJ δJ K
R
ei dx[jα φ +J bI +σ̄I ω +σ ω̄I ]
α I I I

" #
Z
δ δ δ
≡ SZ = dx jβ − σ̄J − σK K Z = 0 . (51)
δγβ δζJ δJ

From now on, whenever we shall make explicit dependence on the space-time
coordinates, we shall use for the functional derivative the notation used in
Eq.(51). This equation translates the S-T identity in terms of the Green
functions. The same equation holds true for the connected functional:

Zc [ j, J, σ̄, σ , γ, ζ] ≡ −i logZ [ j, J, σ̄, σ , γ, ζ] . (52)

13
It is very useful to translate (52) into a functional differential equation for
the proper functional [1]. In perturbation theory the proper functional is the
functional generator of the 1-particle-irreducible amplitudes and is generally
defined as the Legendre transform of Zc . It is often called the effective
action, although this name is also shared by completely different objects.
Introducing the collective symbol J for the field sources (j, J, σ̄, σ ), K for
the other sources (γ, ζ) and Φ for the fields (φ, b, ω, ω̄), one defines the field
functional:
δ δ
Φ [J , K] ≡ Zc [J , K] − Zc [0, 0] , (53)
δJ δJ
then, assuming that the inverse functional J [Φ, K] is uniquely defined, one
has the proper functional:
!
Z
δ
Γ [Φ, K] ≡ Zc [J [Φ, K] , K] − dxJ [Φ, K] Φ + Zc [0, 0] . (54)
δJ

It is easy to verify that:

δ
Γ [Φ [J , K] , K] = ∓J , (55)
δΦ
δ δ
Γ [Φ, K] |Φ=Φ[J ,K] = Zc [J , K] . (56)
δK δK
Therefore:
" #−1
δ δ δ δ
0
Γ [Φ, K] |Φ=Φ[J ,K] = ∓ Zc [J , K] . (57)
δΦ(x) δΦ (y) δJ (x) δJ 0 (y)

That is: the second field-derivative of Γ gives the full wave operator. Notice
that the minus sign in (55) and (57) refers to commuting fields while in the
anti-commuting case one has the plus sign.
Using the above identities one can immediately write the S-T identity for
the proper functional:
" #
Z
δ δ δ δ δ
dx α
Γ Γ + I Γ Γ + bI Γ =0. (58)
δφ δγα δω δζI δ ω̄I

This identity is a crucial tool in many instances, we shall exploit it in the

analysis of unitarity, even more important is however its role in renormaliza-
tion.

14
 It is apparent from Eq.(26) that the Lagrangian density depends on the
field b only through linear and bilinear terms. Thus this field behaves like a
free field and satisfies a linear equation of motion which is not affected by
the perturbative corrections and in the functional language turns out to be:
" #
I δ δ
J + VαI (∂) + ξδ IJ J Zc = 0 , (59)
δjα δJ

and hence:
δ
Γ = VαI (∂)φα + ξδ IJ bJ . (60)
δbI
Equation (59) combined with (51), which holds true also for Zc , gives:
δ
σ I + VαI (∂) Zc = 0 , (61)
δγα
that is:
δ δ
Γ + VαI (∂) Γ=0. (62)
δ ω̄I δγα
We shall use these equations in order to simplify the analysis of the S-matrix
unitarity.
A third interesting application of Eq.(58) is the search for generalizations
of the geometrical setting of gauge theories. This is based on the fact that
(58) is verified by the classical action S of a gauge theory. Indeed the classical
action is the first term in the loop-ordered perturbative expansion of Γ.
The search for generalizations is justified by the fact that the low energy
effective actions of more general theories, such as e.g. supergravity, are free
from dimensional constraints; this allows the introduction of terms of higher
degree in the sources γ and ζ. Disregarding the gauge fixing, setting ω̄ =
b = 0, let us consider for example:
1 αβ 1
S = Sinv + γα ω I PIα (φ) + ω I ω J γα γβ RIJ I
+ ζI CJK ωJ ωK , (63)
2 2
that inserted into (58) gives:

ω I PIα ∂α Sinv = 0 ,
1
 
γβ ω I ω J PJα ∂α PJβ − PKβ CIJ
K αβ
+ RIJ ∂α Sinv = 0 ,
h
2 i
ζK ω I ω J ω L PIα ∂α CJL
K K
+ CM M
I CJL = 0 ,

15
 1 γ
  
I
γα γβ ω ω ω J L
∂γPIβ RJL
γα
+ αβ
PI ∂γ RJL αβ
+ RM C M
I JL =0,
h
2 i
βγ
γα γβ γγ ω I ω J ω K ω L RIJ
δα
∂δ RKL =0
h i
γα ζK ω I ω J ω K ω L RIJ
δα M
∂δ CKL =0. (64)

In order to simplify the notation we have written ∂α instead of ∂φα . The first
line above implies the invariance of Sinv under the action of the differential
system X given in (11), the second one takes place of the first equation in
(20). It is indeed clear that the first term in this line corresponds to the
commutator of two X’s, the second one prescribes the structure functions
of the algebra while the last one is new; it defines the deviation from a
closed algebra that, being proportional to the field derivative of the physical
action, vanishes on the mass shell. The third equation in (64) prescribes
the deformed structure of the Jacobi identity, that is, of the the second
αβ
line of Eq.(20), while the remaining lines give constraints for RIJ . These
constraints depend on the particular choice of Eq.(27) which excludes terms
of second order in ζ. Eq.(27) shows the simplest example of the extensions
of our method to open algebras that have been introduced by Batalin and
Vilkovisky [12].
A very simple example of a mass-shell closed gauge algebra can be found
if one tries to use the BRS algorithm to compute a n-dimensional Gaussian
integral in polar coordinates3 . Let ~x with components xi , (i = 1, .., n) be
2
the variable and Sinv = − x2 define the invariant measure under the action
of the gauge group O(n) corresponding to the BRS transformations:

sxi = ωij xj (65)

where ωij is the O(n) ghost antisymmetric in its indices and, as usual, the
sum over repeated indices is understood. The polar coordinate gauge choice
corresponds to the vanishing of n − 1 components of ~x. This configuration
has a residual O(n − 1) invariance and therefore the Jacobian matrix in (14)
is highly degenerate. To overcome this difficulty one has to enlarge the BRS
structure adding ghosts for ghosts (γij ); and hence introducing the ghost
transformations:

sωij = γij − ωik ωjk

sγij = γik ωjk − ωik γjk . (66)
3
This exercise has been suggested by J.Fröhlich

16
With this choice s is not nilpotent; indeed s2 xi = γij xj . It is mass-shell
nilpotent since the ”field equations” are: ∂xi Sinv = xi = 0. Disregarding
the structure of the gauge fixing and introducing a suitable set of sources,
we identify the form of the action (27) which is adapted to the present case
according:
1
S = Sinv + µi sxi + ζij sωij + ηij sγij + µi µj γij . (67)
2
Eq.(67) satisfies the Slavnov-Taylor identity (58) ( for b = 0 ) The last term
in Eq.(67) corresponds to the second order term in γ in Eq.(27).

5 Unitarity

The first step in the analysis of S unitarity is the study of the asymptotic
propagators and wave operators of a gauge theory [13].
For simplicity we shall limit our study to the situations in which the whole
gauge symmetry is spontaneously broken and hence no vector particle is left
massless. We shall also assume that, contrary to the Electro-Weak model,
all physical and unphysical particles are stable, and we choose a gauge fixing
prescription of the ’t Hooft kind in which no unphysical particle, in particular
Faddeev-Popov ghost, is mass-less.
First of all let us now consider how one can extract information about
asymptotic particle states from the second derivatives of the proper func-
tional. We label by Φi the fields appearing in our theory, that is the variables
upon which Γ depends. In our case the index i = 1, · · · , N distinguishes vec-
tor field components from ghost and matter field ones. We choose the fields
Φ Hermitian, with the exception of the anti-ghost which are chosen anti-
Hermitian. Defining the Fourier transformed field:
Z
d4 x −ip·x
Φ̃i (p) ≡ e Φi (x) , (68)
(2π)4

and setting:
Z
δ δ
∂Φ̃i (p) ≡ d4 x eip·x , ∂Φi ≡ , (69)
δΦi (x) δΦi (0)

17
we get the Fourier transformed wave matrix of the theory whose elements
are the second derivatives:
∂Φ̃i (p) ∂Φj Γ|Φ=0 ≡ Γi,j (p) . (70)
With the exception of the ghost field components this is in general an Her-
mitian matrix. Due to translation invariance one has Γi,j (p) = ±Γj,i (−p)
where the upper and lower signs refer to commuting and anti-commuting
fields respectively.
For simplicity we limit our discussion to the cases in which the determi-
nant ∆(p2 ) ≡ det |Γ(p)| is an analytic function of p2 for p2 < M 2 and its
zeros in the analyticity domain lie on the positive real axis.
Let (p̄λ )2 = m2λ for λ = 1, · · · , a be solutions of order kλ < N of the
equation ∆(p2 ) = 0. We assume that the matrix equation Γ(−p̄λ )i,j vj (p̄λ ) = 0
has kλ independent, non-trivial solutions. That is, the matrix Γ(−p̄λ ) has kλ
independent null eigenvectors.
This is a simplifying hypothesis which excludes the presence of dipole, or
even worst, singularities in the unphysical propagators. These singularities
are often met in gauge theories with particular gauge choices, e.g. QED in
the Landau gauge and in the example we shall present in the following. As a
matter of fact it is shown in Appendix C that dipole singularities correspond
to mass-degenerate asymptotic states with opposite norm, that is, they cor-
respond to a pair of simple poles with opposite residue and degenerate in
mass. Thus our analysis is easily extended to the dipole case, however here
we prefer to begin considering the simplest case of simple poles.
Under the above assumptions near a zero p̄λ of the determinant one has:
Γ(p) = (p2 − m2λ )ζ̃λ (p) + Rλ (p) , (71)
where the matrices ζ̃λ and Rλ are Hermitian and by no means unique. As
already noticed in section 2 this lack of uniqueness does not affect the scat-
tering theory. Still, in a suitable neighborhood of p̄λ , we make the choice
ζ̃λ Rλ = Rλ ζ̃λ = 0.
With this choice ζ̃λ (p) has rank kλ and, if Pλ (p̄l ) is the projector upon
the space spanned by the null eigenvectors of Γ(−p̄λ ), one has:
Pλ (p)ζ̃λ (p) = ζ̃λ (p)Pλ (p) = ζ̃λ (p) (72)
Furthermore another matrix ζλ (p) with rank kλ exists such that ζ̃λ (p)ζλ (p) =
ζλ (p)ζ̃λ (p) = Pλ (p) . Therefore one has:
Γ(p)Pλ (p) = (p2 − m2λ )ζ̃λ (p) + O((p2 − m2λ )2 ) , (73)

18
and hence:

Pλ (p) = (p2 − m2λ )Γ−1 (p)ζ̃λ (p) + O((p2 − m2λ )) , (74)

and
ζ(p)
Γ−1 (p) = + Qλ (p) , (75)
p2 − m2λ − i0+
where Qλ is analytic in the mentioned neighborhood of p̄λ . This last equation
leads to Eq.(4) and hence allows the construction of the scattering matrix
with the asymptotic fields satisfying the (anti-)commutation relations (5).
In particular the expression for the scattering matrix, Eq.(8), can be
written in the alternative form involving the Fourier transformed asymptotic
˜
fields φ̃in and corresponding sources J:
Z
d4 p i δ
S =: exp Φ̃in (p)Γj,i (−p) : Z|j=0 , (76)
(2π) 4 ˜
δ Jj (p)

where, taking into account Eq. (5), we have replaced the asymptotic wave
operator K with the proper two point function Γ.
Now we consider how the Slavnov-Taylor identity for the proper func-
tional (58) and Eq.s (60) and Eq.s (62) constrain the wave matrix of our
gauge theory. Setting:

δ2
Γ|Φ=0 ≡ Γαβ (p) , (77)
δ φ̃α (p)δφβ (0)

and
δ2
Γ|Φ=0 ≡ ΓβI (p) , (78)
δω I (0)δγ̃β (p)
and computing in the origin of the field manifold the second functional deriva-
tive of Eq.(58) with respect to φ̃α (p) and ω I (0) we get:

Γαβ (p) ΓβI (p) = 0 . (79)

In QED this corresponds to the transversality condition for the vacuum po-
larization.
In general from Eq’s (11) and (27) we see, up to quantum corrections,
that Eq. (78) defines G vectors ΓI (p), where G is the dimension of the gauge
Lie algebra mentioned in section 3. These vectors span the tangent space to

19
the orbits in the origin. Therefore they must be independent and they must
remain such beyond the quantum corrections. Eq. (79) shows that, for a
generic p the matrix Γ has G null eigenvectors and hence its determinant is
identically zero. We shall call Γ a degenerate matrix. For a generic choice of
the momentum Γ’s rank is equal to F − G where F is the number of its rows
and columns .
From Eq.(60) one get:

δ2
Γ|Φ=0 = VαI (ip) , (80)
δ φ̃α (p)δbI (0)

thus setting, as above:

δ
Γ| ≡ CIJ (p) . (81)
˜ J (p) Φ=0
δω I (0)δ ω̄

and taking the second derivative of (58) with respect to b̃I (p) and ω J , we get

ΓαJ (p)VαI (−ip) = −CJI (p) . (82)

We have furthermore from (60):

δ2
Γ|Φ=0 ≡ ξδ IJ . (83)
δ b̃J (p)δbI (0)

The wave matrix of gauge and matter fields is then given by:
!
Γαβ VαI
(p) , (84)
VβJ∗ ξδ I,J

where, taking account that V (∂) is a real linear matrix, we have set VαI ≡
VαI (ip) and VαI∗ = VαI (−ip).
The wave operator of the Faddeev-Popov ghosts has matrix elements
J
CI (p) .
Let us now look for a null eigenvector of the above gauge and matter field
wave matrix. We have to solve the system Γi,j (−p)Φj (p) = 0, that is:

Γαβ (−p)φβ (p) + VαI (−ip)bI (p) = 0

VβI (ip)φβ (p) + ξbI (p) = 0 (85)

20
which implies:
" #
1
Γαβ (−p) − VαI (−ip)VβI (ip) φβ (p) ≡ Γ̄αβ (−p)φβ (p) = 0 . (86)
ξ
and
ξbIin (p) = −VαI (ip)φαin (p) . (87)
If the gauge fixing procedure is complete, the matrix Γ̄ must be non-degene-
rate, and the solutions to the equation det |Γ̄|(p2 ) = 0 correspond to the
masses of the bosonic gauge and matter fields. Notice that the spinor de-
grees of freedom are not considered since they only carry physical degrees of
freedom.
Now, taking into account Eq.(79), Eq.(82) and Eq.(86) we get:
VαJ (ip)CIJ (p)
Γ̄αβ (p)ΓβI (p) = . (88)
ξ
This equation relates the masses in the ghost sector to those in the gauge
matter sector. The matrix Γ̄ is Hermitian and hence it has the structure
appearing in Eq.(71).
If p̄g is a solution of the equation det |C| = 0, the linear system

CJI (−p̄g )wJ (p̄g ) = 0 (89)

has non trivial solutions. Let Mg be the number of the independent solutions
of (89) that we label by waJ (p̄g ) for a = 1, · · · , Mg . From (88) we have

Γ̄αβ (−p̄g )ΓβI (−p̄g )waI (p̄g ) ≡ Γ̄αβ (−p̄g )φβa (p̄g ) = 0 . (90)

The Mg vectors φa are linearly independent since the vectors ΓI are linearly
independent. In other words, if one had a non-trivial set of coefficients ca
such that ca φαa = ca waI (p̄g )ΓαI (−p̄g ) = 0, this would contradict, either the
hypothesis of linear independence of the wa ’s, or that of independence of the
ΓI ’s.
In the same situation also the linear system

w̄I (p̄g )CJI (p̄g ) = 0 (91)

has Mg linearly independent solutions that we label by w̄a (p̄g ). For every
solution of det |C|(p) = 0 one has the same number of solutions of Eq.(89)
and of Eq.(91).

21
 Notice that, if there are solutions to Eq.(86) with bI 6≡ 0, they correspond
to solutions of Eq.(91). Indeed under the above condition VαI (ip)φα (p) 6≡ 0,
however one has:

ΓαI (p)Γ̄αβ (−p)φβ (p) = −CIJ (p)VβJ (ip)φβ (p) = ξCIJ (p)bJ (p) = 0 . (92)

These solutions with non-vanishing b do not necessarily exist, indeed, as it

will be shown in a moment, they may be replaced with dipole singularities. In
any case the number of solutions of Eq.(86) with independent non-vanishing b
cannot exceed G. Let it be G−X. We claim that Eq.(86) has X independent
dipole solutions.
In order to prove this claim we better specify our framework. We are
considering models in which the bosonic fields are either gauge vector or
scalar fields. Among the scalar fields there are G Goldstone bosons associ-
ated with the spontaneous breakdown of gauge symmetry and required by
the Higgs mechanism, and the same number of vector fields. Indeed in this
situation and with the ’t Hooft gauge choice, one can exclude massive asymp-
totic fields. Furthermore the bosonic wave operator is the sum the space-like
vector field operator, that of the Higgs scalar fields, and the gauge fixing
wave operator for the mixed the Goldstone fields and scalar components of
the vector fields, that is ∂ µ Aµ . We concentrate on this wave operator for
the moment forgetting the rest. With our Hermitian choice of the field basis
the Fourier transform of the gauge fixing term of the wave operator defined
in Eq.(77) is a real p-dependent matrix which has the following 2 × 2 block
structure: !
Γ11 Γ12
, (93)
ΓT12 Γ22
where the suffix T means ”transposed”. The blocks are G×G matrices and in
particular Γ11 is the restriction of the wave matrix to the ∂ µ Aµ components
and Γ22 that to the Goldstone bosons. According to Golstone’s theorem
Γ22 = p2 K and, in our framework, K, which is a real symmetric matrix, is
also invertible, thus we can write:

Γ22 = OT p2 d O (94)

where d is a real, diagonal and strictly positive matrix and O an orthogonal

one.

22
 The matrix ΓβI (p) defined in Eq.(78) has the corresponding block struc-
ture: !
γ1
(95)
γ2
where, due to the assumption of complete symmetry breakdown, both γ1 and
γ2 must be invertible matrices.
Eq. (79) is written in terms of the above block matrices in the form:

Γ11 γ1 + Γ12 γ2 = 0 , ΓT12 γ1 + Γ22 γ2 = 0 . (96)

Changing the basis of the Goldstone bosons one can transform Γ22 into
the unity
q matrix leaving Γ11 and √
γ1 unchanged and transforming Γ12 into
Γ12 O 1/(p d) ≡ Γ12 and γ2 into p2 dOγ2 ≡ γ20 . After this transformation
T 2 0

Eq.(96) becomes:
0
Γ11 γ1 + Γ012 γ20 = 0 , Γ12T γ1 + γ20 = 0 , (97)

which is solved by:

0
Γ012 = −(γ1−1 )T γ2T ≡ R , Γ11 = RRT . (98)

The gauge fixing wave matrix Γ̄ defined in Eq.(86) can be written as the
difference Γ̄ = Γ−∆ where ∆ is deduced from Eq.(86). Here we generalize the
gauge choice replacing the matrix ξδIJ with the symmetric real and invertible
matrix ξ IJ getting the new matrix ∆ in the block form:
!
ξ −1 ξ −1 V T
. (99)
V ξ −1 V ξ −1 V T

With the new choice

q of the Goldstone field basis ∆ keeps the same form with
V replaced with 1/(p2 d) O V ≡ W .
The ghost wave matrix is computed from Eq.(82) getting:

C = −γ1T − γ2T V = −γ1T (I − RW ) . (100)

Therefore, in order to study the masses of the asymptotic fields corresponding

to the Goldstone bosons and scalar components of the gauge fields, we have
to compute the determinant of the matrix whose block structure is:
!
RRT − ξ −1 R − ξ −1 W T
Γ̄ ≡ . (101)
RT − W ξ −1 I − W ξ −1 W T

23
We shall use the following formula for a matrix in block form:
!
A B
det = det(A − BD−1 C) det D , (102)
C D

where, of course, we have assumed that det D does not vanish4 . In order to
exploit this formula let us consider the matrix:

RRT − ξ −1 − (R − ξ −1 W T )(I − W ξ −1 W T )−1 (RT − W ξ −1 )

∞
= RRT − ξ −1 − (R − ξ −1 W T ) (W ξ −1 W T )n (RT − W ξ −1 )
X

n=0
∞
= −(I − RW )ξ −1 (W T W ξ −1 )n (I − W T RT )
X

n=0
= − − (I − RW )ξ W T (W −1 )T (I − W ξ −1 W T )−1 (I − W T RT ),(103)
−1

where we have assumed the convergence of the operator power series. This
does not limit the generality of our analysis since the operator series certainly
convergences for a suitable choice of the matrix ξ and, from Eq.(102) we get:

det Γ̄ = (det ξ)−1 (det C)2 /(det γ1 )2 (104)

which is independent of det(I − W ξ −1 W T ) ≡ det D.

In order to simplify the discussion let us now assume that on a given
ghost mass-shell, say p2 = m̄20 , there is a single ghost solution w0 (Eq.(89))
and a single anti-ghost one w̄0 (Eq.(91)). This hypothesis does not limit the
generality of our discussion since it can always be met through a suitable
choice of the ghost fixing parameters. Our formula (104) shows that on the
same mass-shell det Γ̄ has a double zero Therefore, either Eq.(86) has two
solutions, that is, φ̂α = ΓαI (−p̄0 )w0I and a solution with bI = w̄0,I , or the first
solution is unique and the corresponding asymptotic field propagator has a
dipole singularity. It is shown in Appendix C that in this case the positive
frequency part of the asymptotic field creates two degenerate particle states
with opposite norm.
The analysis of S-matrix unitarity can be pusher forward in both situa-
tions, it is however simpler in the case of two solutions [3]. Therefore, in the
following, we shall consider this particular case.
4
I thank C. Imbimbo for suggesting this particular equation among many equivalent
forms

24
 Let us now compute, in the framework of our gauge theory, the operator
Σ appearing and defined in Eq.(8). Limiting the analysis to the gauge and
bosonic matter fields and to the ghost fields, that is, disregarding the spinor
fields, one has:
d4 p
"
Z
δ δ
ΣB = φαin (p)[Γβα (−p) + VαI (ip) I ]
(2π) 4
δ j̃β (p) δ J˜ (p)
δ δ δ
+bIin (p)[VβI (−ip) + ξ I ] − ωin I
(p)CIJ (−p) J
δ j̃β (p) δ J˜ (p) ˜ (p)
δ σ̄
#
δ
+ω̄in,J (p)CIJ (p) . (105)
δσ̃I (p)
Using Eq.(87), we can transform this equation into the simpler form:
d4 p
"
Z
δ δ
ΣB = 4
φαin (p)Γ̄βα (−p) I
− ωin (p)CIJ (−p) J
(2π) δ j̃β (p) δσ̃ (p)
#
δ
+ω̄in,J (p)CIJ (p) . (106)
˜ I (p)
δ σ̄
It is clear that the operator ΣB acts on the asymptotic states contain-
ing ghost and bosonic particles, since we are forgetting spinors. Among
the bosonic particles there are those corresponding to the asymptotic fields
φαa (p) = ΓαI (−p)waI (p) considered above in Eq.(90) and the asymptotic com-
ponents of b. There also are the particles corresponding to the remaining in-
dependent solutions of Eq.(86) that we call physical together with the spinor
particles.
Computing the commutator [S, ΣB ]Z where;
" #
Z
δ δ δ
S≡ dx jβ − σ̄J − σK K , (107)
δγβ δζJ δJ
is the functional differential operator appearing in Eq.(51) and Z is the Feyn-
man functional, we get:
d4 p
" #
Z
δ δ
[S, ΣB ]Z = − φαin (p)Γ̄βα (−p) I
+ ωin (p)CIJ (−p) J Z,
(2π) 4 δγ̃β (p) δ J˜ (p)
(108)
R d4 p J δ δ
where we have omitted the term (2π)4 ω̄in,J (p)CI (p) δζ̃ (p) Z since δζ̃ (p) Z is
I I
a regular function of p2 at p2 = m2g for any ghost mass mg and the momen-
tum integral in Eq.(108) is restricted to the union of ghost mass-shells where

25
 δ
a pole in δ ζ̃I (p)
Z should be needed in order to compensate the zeros of the
ghost wave operator C. For the same reason the functional derivative δγ̃βδ(p) Z
appearing in the same integral should be replaced with its mass-shell singu-
lar part, which, using Eq.(78), is easily seen to be given by ΓβI (p) δσ̄˜ Iδ(p) Z.
δ2
Indeed one finds that the two-point function Z|
δσ I (0)δγ̃α (p) c J =0
is equal to
δ2
ΓαJ (p) δσI (0)δ ˜ J (p) Zc |J =0
σ̄
where the second factor is the ghost propagator.
Furthermore, from Eq.(59) one has that δJ˜Iδ(p) Z should be replaced with
− 1ξ VIα (−ip) δj̃ αδ(p) Z. Hence, taking into account Eq.(88), Eq.(108) should be
written:
d4 p
"
Z
δ
[S, ΣB ]Z = − φαin (p)Γβα (−p)ΓβI (p)
(2π) 4 ˜ I (p)
δ σ̄
#
1 I δ
− ωin (p)CIJ (−p)VJα (−ip) α Z
ξ δ j̃ (p)
d4 p
"
Z
I δ
= 4
ωin (p)ΓαI (−p)Γ̄βα (−p)
(2π) δ j̃β (p)
#
1 δ
− φαin (p)VαJ (ip)CIJ (p) Z. (109)
ξ ˜ I (p)
δ σ̄

We have used the relation Γαβ (p) = Γβα (−p). This result is the kernel of the
first rigorous unitarity proof given in [3]. Here we shall follow the simpler
analysis given in [14].
Following Kugo-Ojima, we introduce an operator Q acting on the asymp-
totic state space and annihilating the vacuum state: Q|0i = 0. The new
operators satisfies the following relations

[Q, φαin (p)] = −i ΓαI (−p)ωin

I
(p)
1 J
{Q, ω̄in,I (p)} = i Vα (ip)φαin (p) = −ibIin (p) . (110)
ξ
We remind that with our conventions all the asymptotic fields are Hermitian
except ω̄in,I which is anti-Hermitian, and the gauge fixing action is Hermitian.
The operator Q generates a nilpotent transformation on the Fock space. Its
kernel KerQ is the subspace generated by the action of the vacuum state of
(+)
the positive frequency (creation) part of the physical fields and by bin,I and
(+)I (+) (+)I
ωin , while its image ImQ is the subspace generated by bin,I and ωin .

26
 Now it is interesting to define the cohomology of Q, that is, the quotient
space KerQ/ImQ. This is the linear space of the Q-equivalence classes of
the elements of KerQ. We consider Q-equivalent two elements of KerQ if
their difference belongs to ImQ. It is easy to verify that the cohomology of Q
coincides with Q-equivalence classes of the states of the subspace of the Fock
space generated by the action on the vacuum state of the positive frequency
part of the physical fields. Notice that the asymptotic properties of these
field components are completely determined by the invariant part of Γ and
hence it is expected that they generate a positive norm space. It is also a
direct consequence of the nilpotency of Q that the states in the image of Q
have vanishing scalar product with those in its kernel, they are in particular
zero norm states.
Using (110) one has

i [Q, ΣB ] Z = [S, ΣB ]Z . (111)

It follows that:
h i h i
[Q, S] = Q, : eΣB : Z|J =0 =: Q, eΣB : Z|J =0 = −i[S, : eΣB :]Z|J =0
= i : eΣB : SZ|J =0 = 0 . (112)

Notice that Q acts separately on the positive and the negative frequency
parts of the asymptotic fields and this justifies the second identity in Eq.(112)
which shows that the commutator [Q, S] vanishes. Under the assumption that
the measure corresponding to a Hermitian Lagrangian defines a ”pseudo”uni-
tary S-matrix in the asymptotic Fock space:

SS† = S† S = I , (113)

and that the physical space is a positive metric space - both assumptions
are true in perturbation theory - we conclude that, owing to (112) and to
the above discussed properties of Q, the S-matrix is unitary in the physical
space identified with the cohomology of Q. That is: if the initial state |ii
is a physical state and hence it belongs to the kernel of Q and has positive
norm, S|ii belongs to the kernel of Q and has the same norm as |ii, therefore
it defines a Q-equivalence class corresponding to a positive norm physical
state.
We consider for example an SU (2) Higgs model in the tree approxima-
~ µ , a triplet of
tion [3]. This model involves an iso-triplet of vector fields A

27
Goldstone particles ~π , and the Higgs field σ that appears in our calculations
only through its vacuum expectation value V . Therefore now the symbol φα
corresponds to A ~ µ and ~π . We add a further iso-triplet of Nakanishi-Lautrup
multipliers ~b. In the tree approximation the free Lagrangian density is given
by:

F~µν · F~ µν 1  ~ 2 + ~b · ∂ A
   2
~ + ρ~π + ξ b − ω̄
L=− + ∂~π − gV A ~ · (∂ 2 + gρV )~ω ,
4 2 2
(114)
From now on we shall disregard the isotopic indices since all the wave oper-
ators are diagonal in the isotopic space. The wave matrix Γ defined in (77)
is given by: !
pµ pν − g µν p2 + g 2 V 2 g µν −igV pµ
, (115)
igV pν p2
where the first row and line correspond to A and the second ones to π. The
matrix VIα defining the gauge fixing in (25) corresponds to:
!
−ipµ
, (116)
ρ

Notice that the isotopic indices are hidden. The ghost wave operator C(p)
is given by p2 − gρV .
The wave operator of the gauge and scalar fields Γ̄αβ (−p) appearing in
Eq. (86) is:
 
pµ pν (1 − 1ξ ) − g µν p2 + g 2 V 2 g µν i(gV − ρξ )pµ
 2  . (117)
−i(gV − ρξ )pν p2 − ρξ

Forgetting the isospin degeneracy its determinant is given by:

1
det |Γ̄| = − (p2 − (gV )2 )3 (p2 − gρV )2 , (118)
ξ
however Eq.(86) has only four independent solutions, three of them corre-
spond to p2 = m2p = (gV )2 , the remaining one corresponding to p2 = m2g =
gρV where the above determinant has a double zero. Looking at the solu-
tions of Eq.(86) and considering in particular the longitudinal ones in which
A(l)
µ (p) ∼ pµ , one sees that for a generic choice of ρ there is a single solution
which, in contrast with the hypothesis made at the beginning of this chapter,

28
corresponds to a dipole singularity of the gauge and scalar field propagator
which is degenerate with the simple pole of the ghost propagator. This is by
no means surprising since the asymptotic longitudinal vector field satisfies
the free field equations (p2 −ξgV )A~(l) µ (p) = ipµ (ξgV −ρ)~π which is analogous
to the field equation in Landau’s gauge QED. One can find some details on
the structure of the asymptotic state space in Appendix C and in Landau’s
gauge QED in [15]. The dipole singularity disappears and one finds a fifth
independent solution of Eq. (86) mass degenerate with the ghosts if ρ = ξgV .
This is called the special ’t Hooft choice. It is easy to verify that Eq. (86)
has solutions with b 6≡ 0 only in this special case.
The matrix ΓIα defined in (78) in the tree approximation is:
!
ipµ
. (119)
gV

It is apparent that its columns correspond to solutions of Eq. (86) on the

ghost mass-shell in agreement with Eq. (90). The Q operator is defined by
the conditions:
h i
~ µin (p) = − pµ ω
Q, A ~ in (p) , [Q, ~πin (p)] = −igV ω ~ in (p)
1
~ in (p)} = [pµ Aµin (p) + iρ~πin (p)] ≡ −ibin (p) .
{Q, ω̄ (120)
ξ
This operator is apparently nilpotent since on the ghost mass-shell
h i
~ µin (p) } = 0 .
{Q, Q, A (121)

In order to have a look at the physical content of the theory we introduce

three space-like polarization vectors µa (p) orthogonal to the momentum p
~ µin (p) =
and such that µa (p)bν = −δab . Due to Eq.(86) the asymptotic field A
P ~ µ 2 2 2
a φa,in (p)a (p) has support on the mass-shell p = g V and its positive
frequency part generates a positive norm subspace of the asymptotic space.
Furthermore from the above relation it is apparent that [Q, φa,in (p)] = 0 and
hence the Q-equivalence classes of these positive norm states identify the
cohomology of Q, that is the physical state space.
Notice that the presence of dipole singularities remarked above, which,
as shown in Appendix C, corresponds to ghost-degenerate unphysical states
with opposite norm, does not affect the conclusions of our analysis [3][15].

29
A Appendix
The differential system {X} being integrable one can choose local charts
of coordinates trivializing the fibration; let us indicate by {ξ} the gauge
invariant coordinates, that are constant along the orbits, and by {η} those
parametrizing points on the orbits; the transition functions between two
neighbouring charts are:
ξa = ξa (ξb ) , ηa = ηa (ξb , ηb ) . (122)
In a given chart the elements of the differential system are:
XI ≡ XIJ (ξ, η) ∂ηJ . (123)
On every fibre we define the adjoint system of differential 1-forms according
ω I ≡ ωJI (ξ, η) dη J , (124)
with:
XIL ωLJ = δIJ . (125)
Taking into account the commutation relation (13) one verifies directly that
these 1-forms satisfy the Maurer-Cartan equation:
1 I
dV ω I ≡ dη L ∂ηL ω I = − CJK ωJ ωK . (126)
2
The above equations define explicitly and, according to (122), globally on
very orbit, the exterior algebra involved into the definition of the Faddeev
Popov measure. However they cannot be directly translated into their field
theory equivalent due to locality. Indeed, even if the differential system {X}
is given as a set of local functional differential operators in the gauge field
variables, the trivializing coordinates are non-local with respect to the fields.
This difficulty is overcome replacing the system of generators of the ver-
tical exterior algebra {dη} with {ω} that in this way appear into the theory
as new Grassmannian local field variables. This however requires that the
vertical exterior derivarive dV be written according (19):
1 I
dV = ω I XI − CJK ω J ω K ∂ωI , (127)
2
where the first term in the right-hand side accounts for the action of dV on
functions while the second term acts on the exterior algebra generators.

30
B Appendix
In this appendix we prove (43) exploiting (32). The first step will be the
proof of two lemmas whose recursive use will lead to (43). Let us consider
the set of cells {Ua } and the corresponding partition of unity {χa } and gauge
fixing functionals {Θa } We define:
  n
(−1)k−n χak sχa1 ...sχ̌ak ...sχan ,
X
sχa1 ...sχan−1 χan ≡ (128)
A
k=1

where the check mark above χak means that the corresponding term should
be omitted. It is fairly evident that the functional (128) is antisymmetric
with respect to permutations of the indices (a1 , ..., an ) and that its support
is contained in the intersection of the corresponding cells.
It is apparent that:
 
s sχa1 ...sχan−1 χan = (−1)n+1 n sχa1 ...sχan . (129)
A

Furthermore, taking into account that {χa } is a partition of unity on the

support of the functional measure, one has:
X   X n+1
(−1)k−n−1 χak sχa1 ...sχ̌ak ...sχan+1
X
sχa1 ...sχan χan+1 =
A
an+1 an+1 k=1
= sχa1 ...sχan , (130)
indeed only the term with k = n+1 contributes to the second member giving
the right-hand side of this equation. Comparing (129) with (130), we get:
  X  
s sχa1 ...sχan−1 χan = (−1)n+1 n sχa1 ...sχan χan+1 . (131)
A A
an+1

Let now Aa1 ,...,an+1 be antisymmetric in its indices, one has:

Aa1 ,...,an+1 ∂Θ (Θa1 ...Θan ) eisΘa1 ,...,an
X

a1 ,...,an+1
(−1)n  
Aa1 ,...,an+1 s ∂Θ Θa1 ...Θan+1 eisΘa1 ,...,an+1 . (132)
X
=i
n+1 a1 ,...,an+1

Indeed, using the identity:

(−1)n n+1
(−1)k+1
X X X
Aa1 ,..,an+1 Va1 ,..,an = Aa1 ,..,an+1 Va1 ,..,ǎk ,..,an+1 ,
a1 ,..,an+1 n + 1 k=1 a1 ,..,an+1
(133)

31
the left-hand side of (132) is written:
n+1
(−1)n
(−1)k+l Θa1 ..Θ̌ak ..Θ̌al ..Θan+1
X X
Aa1 ,..,an+1
n+1 a1 ,...,an+1 k<l=1
 
isΘa1 ,..,ǎk ,..,an+1 isΘa1 ,..,ǎl ,..,an+1
e −e
n+1
(−1)n
(−1)k+l Θa1 ..Θ̌ak ..Θ̌al ..Θan+1
X X
=i Aa1 ,..,an+1
n+1 a1 ,...,an+1 k<l=1
isΘa1 ,...,an+1
s (Θak − Θal ) e , (134)
from which one reaches (131).
Let us now consider the extension of (33) to the case of a cell decompo-
sition of FC , one has:
Z Z
χa eisΘa sX = − sχa1 eisΘa1 X
X X
dµC dµC
a a1
Z
(sχa1 χa2 )A eisΘa1 X
X
= dµC
a1 ,a2
1Z  
(sχa1 χa2 )A eisΘa1 − eisΘa2 X
X
= dµC
2 a1 ,a2
i Z
(sχa1 χa2 )A s∂Θ (Θa1 Θa2 ) eisΘa1 ,a2 X .
X
=− dµC (135)
2 a1 ,a2

In (135) we have used (130) and (132) with n = 1. From (135) we have:
" #
Z
isΘa iZ
(sχa1 χa2 )A ∂Θ (Θa1 Θa2 ) eisΘa1 ,a2 sX
X X
dµC χa e + dµC
a 2 a1 ,a2
iZ
s (sχa1 χa2 )A ∂Θ (Θa1 Θa2 ) eisΘa1 ,a2 .
X
=− dµC (136)
2 a1 ,a2

Using recursively the same equations we get:

 n(n−1)
m
n (−1)
2  
dµC  χa eisΘa
X X
− i sχa1 ...sχan χan+1
a n=1 n+1 A
  i
∂Θ Θa1 ...Θan+1 eisΘa1 ,...,an+1 sX
m(m−1)
(−1) 2 Z
m
 
=i dµC s sχa1 ...sχam χam+1

m+1  A
isΘa1 ,...,am+1
∂Θ Θa1 ...Θam+1 s e X . (137)

32
It is clear that, if the maximum effective number of intersecting cells is N
the right-hand side of (137) vanishes for m ≥ N .

C Appendix
The aim of this Appendix is to clarify the structure of the Fock space asso-
ciated with a generalized free Hermitian field φ whose propagator presents a
dipole singularity, that is such that:
Z
dp eipx
h0|T (φ(x)φ(0))|0i = i . (138)
(2π)4 (m2 − p2 − i0+)2
Using the Lehmann spectral representation it follows that the Wightman
function is
Z
dp ipx 0 0 2
h0|φ(x)φ(0)|0i = e θ(p )δ (p − m2 ) , (139)
(2π)4
and hence, if the field is Hermitian, the corresponding Fock space must be an
indefinite metric space. Indeed δ 0 (x) is not a positive distribution. We shall
call pseudo-Hermitian an Hermitian operator in an indefinite metric space
and we shall label the pseudo-Hermitian conjugate by a dagger. It is clear
that our field satisfies the linear equation:

(∂ 2 + m2 )2 φ(x) = 0 , (140)

whose general solution is

Z
dp ipx
φ(x) = 2
e [α(p)δ(p2 − m2 ) + β(p)δ 0 (p2 − m2 )] . (141)
(2π)
Here α(p) and β(p) have analogous p ≡ (p0 , p~) dependence, in particular,
α(p) = α+ (~p)θ(p0 ) + α− (~p)θ(−p0 ). The stability condition for the vacuum
state with our metric choice, px = p0 x0 − p~ · ~x, implies that α− (~p)|0i = 0 and
β− (~p)|0i = 0. Thus α− and β− are annihilation operators. An alternative
expression for the field is given integrating over p0 and taking into account
that, if the function f (x) continuous with its derivative has n non-degenerate
zeros xi one has:
n n
δ(x − xi ) 1 X δ 0 (x − xi )
and δ 0 (f (x)) =
X
δ(f (x)) = 0
. (142)
i=1 |f (xi )| f 0 (x) i=1 |f 0 (xi )|

33
 q
Setting E(~p) = |~p|2 + m2 one gets:
Z
dp ipx
h
φ(x) = e α+ (~p)δ(p0 − E(~p)) + α− (~p)δ(p0 + E(~p))
(2π)2 2E(~p)
#
β+ (~p) 0 0 β− (~p) 0 0
+ δ (p − E(~p)) + δ (p + E(~p))
2p0 2p0
Z
d~p β+ (~p)
= 2
eipx |p0 =E(~p) [α+ (~p) + 2
(1 − ix0 E(~p))]
(2π) 2E(~p) 2E (~p)
Z
d~p β− (~p)
+ 2
eipx |p0 =−E(~p) [α− (~p) + (1 + ix0 E(~p))] . (143)
(2π) 2E(~p) 2E 2 (~p)
Therefore we see that the field is pseudo-Hermitian if
†
α+ (~p) = α− (−~p) , β+† (~p) = β− (−~p) . (144)
Writing Eq(139) in terms of α(p) and β(p) one gets from Eq.(141)
0 0
Z
d4 p ip(x−y) δ (p − E(~p))
h0|φ(x)φ(y)|0i = 4
e 0
(145)
(2π) 2E(~p) 2p
Z
d~p
= eipx |p0 =E(~p) (1 − i(x0 − y 0 )E(~p))
(2π)4 2E 3 (~p)
Z
d~p d~q
= 4
ei(px+qy) |p0 =E(~p),q0 =−E(~q)
(2π) 4E(~p)E(~q)
! !
β+ (~p) β − (~
q )
h0| α+ (~p) + (1 − ix0 E(~p)) α− (~q) + (1 + iy 0 E(~q)) |0i ,
2E 2 (~p) 2E 2 (~q)
from which one finds the commutation conditions:
[α+ (~p), β− (~q)] = [β+ (~p), α− (~q)] = 2E(~p)δ(~p + ~q)
[α+ (~p), α− (~q)] = −E −1 (~p)δ(~p + ~q) [β+ (~p), β− (~q)] = 0 . (146)
Taking into account the pseudo-Hermiticity conditions, these commutation
relations can be diagonalized introducing the annihilation and creation oper-
ators Aσ (~p) and A†σ (~p) with σ = ± defined by:
s √
2 5 + σ1
Aσ (~p) ≡ √ (E(~p)α− (~p) + σ) β− (~p) , (147)
5 4E(~p)
whose commutation rules are:
[Aσ (~p), A†σ0 (~q)] = δσ,σ0 σ2E(~p)δ(~p − ~q) [Aσ (~p), Aσ0 (~q)] = o . (148)
It is apparent that the corresponding Fock space has indefinite metric corre-
sponding to the operator (−1) with N− = d~p/(2E(~p))A†− (~p)A− (~p) .
N−
R

34
References
[1] C.Itzykson e J.-B.Zuber, Quantum Field Theory, MacGraw-Hill, New-
York 1980.

[2] L.D.Faddeev and V.N.Popov, Phys.Letters B25 (1967) 29.

[3] C.Becchi, A.Rouet and R.Stora, Phys Letters. B32 (1974) 344.
C.Becchi, A.Rouet and R.Stora, Commun.Math. Phys. 42 (1975) 127.
L.V.Tyutin, Lebedev preprint FIAN n.39 (1975).
C.Becchi, A.Rouet and R.Stora, Ann. Phys. 98 (1976) 287

[4] S.Sternberg, Lectures on Differential Geometry, Prentice Hall Inc. 1964

[5] N.K.Nielsen, Nuclear Phys. B140 (1978) 499.

[6] H. Neuberger, Phys Letters. B183 (1987) 337.

[7] V. N. Gribov, “Instability of non-abelian gauge theories and impossibil-

ity of choice of Coulomb gauge”, SLAC Translation 176, (1977).
I.M. Singer Commun.Math.Phys. 60 (1978) 7.

[8] D.Zwanziger, Nuclear. Phys. B209 (1982) 336.

G.Dell’Antonio and D.Zwanziger, Nuclear. Phys. B326 (1989) 333.

[9] C. Becchi and C. Imbimbo, A Lagrangian formulation of Two-

Dimensional Topological Gravity and Cech-De Rham Cohomology; in:
BRS Symmetry, M.Abe, N.Nakanishi, I.Ojima ed.s, Universal Academy
Press, Tokyo, 1996.

[10] see e.g.: P.Gaigg, W.Kummer and M.Schweda (Eds.) Physical and Non-
standard Gauges, Lecture Notes in Physics, Vol.361, Springer-Verlag
Berlin Heidelberg 1990.

[11] K. Fujikawa, Nuclear. Phys. B223 (1983) 218.

[12] I.A.Batalin and G.A.Vilkovisky Nucl.Phys. B234 (1984) 106.

35
[13] see e.g.:
C.Becchi, Lectures on the renormalization of gauge theories, in: Rel-
ativity, Groups and Topology II (les Houches 1983) B.S. DeWitt and
R.Stora Eds. (Elsevier Science Pub.B.V. 1984).

[14] T.Kugo and I.Ojima, Prog. Theor. Phys. 66 (1979) 1.

[15] Carlo Maria Becchi and Camillo Imbimbo (2008), Scholarpedia,

3(10):7135
http://www.scholarpedia.org/article/Becchi-Rouet-Stora-Tyutin symmetry

36