IFUM 652/FT

December 1999
arXiv:hep-th/9912278v1 30 Dec 1999

SEMI-NAIVE DIMENSIONAL
RENORMALIZATION 1

M. PERNICI
INFN, Sezione di Milano, Via Celoria 16, I-20133 Milano, Italy

ABSTRACT

We propose a treatment of γ 5 in dimensional regularization

which is based on an algebraically consistent extension of the
Breitenlohner-Maison-’t Hooft-Veltman (BMHV) scheme; we de-
fine the corresponding minimal renormalization scheme and show
its equivalence with a non-minimal BMHV scheme.
The restoration of the chiral Ward identities requires the in-
troduction of considerably fewer finite counterterms than in the
BMHV scheme. This scheme is the same as the minimal naive
dimensional renormalization in the case of diagrams not involving
fermionic traces with an odd number of γ 5 , but unlike the latter
it is a consistent scheme.
As a simple example we apply our minimal subtraction scheme
to the Yukawa model at two loops in presence of external gauge
fields.

1
Work supported in part by M.U.R.S.T.
Introduction
Dimensional regularization [1, 2] is the most powerful scheme for making
perturbative computations in quantum field theory. It is particularly effective
in absence of chiral symmetries, since minimal subtraction [3] preserves the
vectorial gauge symmetries.
In [1] it has been shown that defining γ 5 as in four dimensions the ax-
ial anomaly is reproduced. In [4] it has been proven that this definition
of γ 5 leads to a consistent regularization scheme (BMHV) at all orders in
perturbation theory.
The restoration of the chiral Ward identities in the BMHV scheme has
been studied in [5].
In the BMHV scheme many of the computational advantages of the di-
mensional regularization scheme are lost, since the restoration of the chiral
Ward identities requires the introduction of non-invariant counterterms for
practically any diagram containing γ 5 .
Due to this difficulty, comparatively few quantities have been computed in
this scheme. Among them, there are some one-loop processes in the standard
model [6], two-loop anomalous dimensions of four-fermion operators in QCD
[7], the three-loop anomalous dimension of axial currents in QCD [8]. The
Adler-Bardeen non-renormalization theorem [9] has been verified at two loops
in this scheme in [10].
Systematic computations of the bare action in the BMHV scheme have
been made in non-abelian gauge theories with chiral fermions [11], where the
finite one-loop counterterms necessary to restore the Slavnov-Taylor identi-
ties have been computed; in the simplest Yukawa model at two loop in the
minimal scheme [12]; in a general class of Yukawa models at two loops, where
the finite two-loop counterterms necessary to restore the rigid Ward identi-
ties have been computed, together with some of those necessary to recover
the local Ward identities in presence of external gauge fields [13].
Most of the computations in chiral theories are made in some version of
the naive dimensional regularization scheme (NDR), first considered in [14]
and [15]; other proposals of NDR have been made in [16, 17]. The common
feature of all these versions of NDR is that in fermionic loops with an even
number of γ 5 ’s, these are considered to be anticommuting with all the gamma
matrices, thus avoiding the spurious anomalies present in the BMHV scheme.
However all these regularization schemes have some algebraic inconsistency.
To avoid these algebraic inconsistencies it has been suggested that the cyclic
property of the trace be upheld [18], but in this case bosonic symmetry is
not respected, and no consistency proof for these renormalization schemes is
available.

1
 Reviews on various proposals for dimensional regularization in chiral the-
ories can be found in [19, 20] .
In this paper we introduce a consistent regularization scheme, in which
to the usual γµ and γ 5 formal objects of the BMHV scheme two new γ 5 -like
objects η1 and η2 are added; the former is not anticommuting with all the
gamma matrices, as γ 5 , while the latter is anticommuting; they are defined
in such a way that traces with an even number of η1 or an odd number of η2
vanish.
We will call this scheme semi-naive dimensional regularization (SNDR);
due to its algebraic consistency in d = 4 − ǫ dimension, algebraic manipula-
tions on the bare quantities can be done unambiguously.
We define minimal subtraction (MS) in this scheme in such a way that,
after taking the limit ǫ → 0 on the minimally subtracted Green functions,
there exists a homomorphism from these abstract subtracted Green functions
to the usual renormalized Green functions, which preserves the algebraic
properties on convergent diagrams, like the skeleton expansion.
We prove the equivalence of the MS-SNDR scheme with a non-minimal
BMHV scheme.
The proof of the consistency of the MS-SNDR scheme is the main result of
this paper. We do not deal with the practical problem of restoring the Ward
identities, apart from an explicit example and some general considerations,
showing that in this scheme there are considerably fewer spurious anomalies
than in the MS-BMHV scheme.
As an example of the practical convenience of our scheme with respect
to the BMHV scheme, we treat in some detail the two-loop renormalization
of the Yukawa model in presence of external gauge currents, which we have
worked out in the BMHV scheme in [13]; while in the latter scheme minimal
subtraction does not preserve the (anomalous) chiral Ward identities, so that
finite counterterms must be added to restore them for every relevant term, in
the SNDR scheme, after performing minimal subtraction only two finite two-
loop counterterms are needed to preserve the anomalous Ward identities,
with the anomaly appearing in the Bardeen form [21]; the Adler-Bardeen
theorem [9] is verified.
We discuss briefly the case of chiral gauge theories and we compare SNDR
with previously proposed NDR schemes.

1 Semi-naive dimensional regularization and

minimal subtraction

2
1.1 Breitenlohner-Maison-’t Hooft-Veltman scheme
We recall how gamma matrices are treated in the ’t Hooft and Veltman
[1] dimensional regularization scheme as elaborated by Breitenlohner and
Maison (BMHV) [4] . We use Euclidean space conventions.
In the BMHV scheme one considers the Lorentz covariants I, δµν , γµ ,
pµ , etc. as formal objects, satisfying the usual tensorial rules. δµν is the
Kronecker delta in d = 4 − ǫ dimensions; a formal rule for summed indices is
given;

δµν δνρ = δµρ ; δµν pν = pµ ; δµµ = d (1)

The Lorentz indices of these formal covariants do not take a specific value.
The gamma ‘matrices’ γµ satisfy the relation

{γµ , γν } = −2δµν I (2)

where I is the identity,

I 2 = I ; Iγµ = γµ I = γµ (3)

The trace is cyclic and satisfies

tr I = 4 (4)

In the BMHV scheme additional ‘(d − 4)-dimensional’ or ‘evanescent’

tensors δ̂µν , p̂µ and γ̂µ are introduced; the Kronecker delta in the (d − 4)-
dimensional space is δ̂µν , satisfying

δ̂µν δνρ = δ̂µν δ̂νρ = δ̂µρ ; δ̂µµ = −ǫ

p̂µ ≡ δ̂µν pν ; γ̂µ ≡ δ̂µν γν (5)

The Kronecker delta in four dimensions in δ̄µν , satisfying

δ̄µν ≡ δµν − δ̂µν ; p̄µ ≡ δ̄µν pν ; γ̄µ ≡ δ̄µν γν (6)

The Levi-Civita antisymmetric tensor has no evanescent component:

δ̂µν ǫνρστ = 0 (7)

and satisfies
X
ǫµ1 µ2 µ3 µ4 ǫν1 ν2 ν3 ν4 = sign(π)Π4i=1 δ̄µi νπ(i) (8)
π∈S4

3
The ‘matrix’ γ 5 is defined by
1
γ5 = ǫµνρσ γµ γν γρ γσ (9)
4!
which implies
{γ 5 , γµ } = 2γ̂µ γ 5 (10)
γ 5 γ 5 = I ; [γ 5 , γ̂µ ] = 0 (11)
After having performed the Dirac algebra with these rules, reduced the
number of Levi-Civita tensors, eliminated traces, reduced products of gamma
matrices to the corresponding antisymmetric products and eliminated summed
indices, the BMHV elements are reduced to normal form. The Feynman
graphs are expressed in terms of meromorphic functions of ǫ, which are the
coefficients of these normal form terms.
The minimal subtraction on a l-loop 1P I Feynman graph renormalized
up to order l − 1 consists in subtracting local terms, polynomial in momenta
and masses, which are the poles of the Laurent series in ǫ of the meromor-
phic functions defining such a graph, both for hatted and non-hatted tensor
structures in normal form.
The bare action is a formal object, from which the Feynman graphs are
constructed following the usual rules. In the minimal subtraction scheme
the bare action l-loop counterterms are the poles of the l-loop 1P I Green
functions, computed using the Feynman rules derived from the (l − 1)-loop
bare action.
After making the minimal subtraction, the limit for ǫ → 0 is taken in
the entire functions which are the coefficients of the normal form terms, and
all the hatted tensors are set to zero; finally the formal sums, traces and
the normal form covariants are identified with the usual four-dimensional
operations and objects respectively.
In Euclidean space the reflection symmetry takes the place of hermiticity
[22]. Reflection symmetry is an antilinear involution, with
Θψ(x) = ψ̄(x′ )γ1 ; Θψ̄(x) = γ1 ψ(x′ ) (12)
Θγµ = −γµ ; Θγ 5 = γ 5 (13)
′ ′
where x 1 = −x1 , x µ = xµ for µ 6= 1. δµν , ǫµνρσ and ǫ are invariant under
reflection symmetry.
Considering a multiplet of fermions, the following fermionic bilinears are
reflection symmetric:
Z
ψ̄(H1 γµ + H2 γ̂µ + H3 γ 5 γ̄µ + iH4 γ 5 γ̂µ )∂µ ψ + ψ̄(iH5 A + H6 γ 5 B)ψ

+ψ̄(iH7 γµ Vµ + iH8 γ 5 γ̄µ Aµ + iH9 γ̂µ Vµ + H10 γ 5 γ̂µ Aµ )ψ (14)

4
where the scalar A, the pseudoscalar B, the vector Vµ and the pseudovector
Aµ are real fields, with

ΘA(x) = A(x′ ) ; ΘB(x) = B(x′ ) ; ΘVµ (x) = Vµ′ (x′ ) ; ΘAµ (x) = A′µ (x′ )

and Hi are hermitian matrices commuting with the gamma matrices.

A non-minimal subtraction consists in performing, after the minimal sub-
traction, the subtraction of finite local counterterms; to avoid ambiguities,
one must characterize which are the normal form terms on which the finite
subtractions are made.
In the case of a reflection symmetric theory, a convenient choice of fermionic
bilinears finite counterterms are the non-hatted terms in (14). The advan-
tage of this choice is that in the case of non-chiral theories the d-dimensional
Lorentz symmetry of the bare action is preserved.

5
1.2 Extension of the BMHV algebra
Add to the BMHV Dirac algebra the objects η and η1 satisfying the following
defining relations, assumed to be valid for arbitrary ǫ:

{η, γµ } = 2γ̂µ η1 (15)

η2 = I (16)
Iη = η (17)
[η1 , γ̂µ ] = 0 (18)
[η1 , η] = 0 (19)
tr ηγµ γν γργσ = 4 ǫµνρσ (20)
tr η12 = 0 (21)

We assume that the trace has the property (4) and that it is cyclic on this
enlarged algebra.
Under reflection symmetry, Θη = η and Θη1 = η1 .
Define

η2 ≡ η − η1 (22)

Consider first the case ǫ 6= 0.

Let us prove that

ηI = η (23)
{η1 , γµ } = 2γ̂µ η1 (24)
{η2 , γµ } = 0 (25)
η1 I = Iη1 = η1 (26)
η1 η2 = η2 η1 = 0 (27)
η12 + η22 = I ; η13 = η1 ; η23 = η2 (28)

Eq. (23) is obvious. From (15) one gets

1 1 1
ǫη1 = γ̂µ {η, γµ } = ǫη + γ̂µ ηγ̂µ (29)
2 2 2
Using (15), (18), (22) and (23) one gets the relations (24), (25) and (26).
From (19), (24) and (25) one gets easily

[η12 , γµ ] = [η22 , γµ] = 0

γµ η1 η2 γµ = (d − 8)η1 η2 (30)
[η1 , η2 ] = 0

6
From (16), (22) and (30) one has

−d = γµ η 2 γµ = γµ (η12 + η22 + {η1 , η2 })γµ =

−d(η12 + η22 ) + (d − 8){η1 , η2 }

so that

ǫ{η1 , η2 } = 0 (31)

Since ǫ 6= 0, we get {η1 , η2 } = 0, which together with (19) implies (27).

Finally (28), which follows trivially from (16) and (27), says that η12 and
η22 are orthogonal projectors commuting with all the elements of this algebra.
Let us prove the following relations for the trace:

tr η1 = tr η2 = 0 (32)
tr η12 γµ1 ...γµr = 0 (33)
tr η22 γµ1 ...γµr = tr γµ1 ...γµr (34)
tr η2 γµ1 ...γµr = 0 (35)
tr η1 γ 5 γµ1 ...γµr = tr γµ1 ...γµr (36)

for arbitrary choice of indices µ1 , .., µr , r ≥ 1. One has

−d tr η1 = tr η1 γµ γµ = tr γµ η1 γµ = (8 − d)tr η1
−d tr η2 = tr η2 γµ γµ = tr γµ η2 γµ = d tr η2 (37)

from which (32) follows. To prove the remaining relations (33-36), observe
that if for all µ

X± γµ = ±γµ X± (38)

then, for indices µ1 , .., µr all different

−d tr X± γµ1 ...γµr = tr X± γµ1 ...γµr γµ γµ =

±tr X± γµ γµ1 ...γµr γµ = ±(−1)r (2r − d)tr X± γµ1 ...γµr (39)

which for non-integer d has solution only for

tr X± γµ1 ...γµr = 0 (40)

Using this observation and the fact that η12 , η1 γ 5 and η22 commute with all the
gamma matrices, while η2 anticommutes with all of them, one gets relations
(33-36) for indices µ1 , .., µr all different. In particular one gets

tr η2 γ 5 = 0 (41)

7
 The general case for (33-36) follows from this particular case since, sim-
plifying gamma matrices with equal indices, one is reduced to the relations

tr η12 = tr η2 = 0 ; tr η22 = 4 ; tr η1 γ 5 = 4 (42)

which follow trivially from the relations (16-21) and from (32), (41).
Using (32-36) all the traces can be computed, observing that the traces
with more than two η1 or two η2 can be reduced to the cases (32-36) using
(15-28).
Observe that in d 6= 4 dimensions this algebra of covariants does not
include ǫµ1 ...µd , as might be thought due to the presence of η2 , which is a γ 5 -
like object anticommuting with all gamma matrices. In fact tr η2 γµ1 ...γµd = 0
due to (35).
In the case ǫ = 0, γ̂µ vanishes and η1 is an independent generator, since
eq.(29) becomes trivial; in that case, add the defining relations

{η1 , γ̄µ } = 0 (43)

ηη1 = η12 (44)
Iη1 = η1 I = η1 (45)

which we obtained from (15-19) in the case ǫ 6= 0.

In the case ǫ = 0, after identifying the formal sums (e.g. pµ γµ ) as usual

Einstein convention sums, there is a homomorphism φ between the subal-
gebra generated by γµ = γ̄µ and η, with the trace tr, and the usual four-
dimensional Dirac algebra generated by γµ′ , with unit matrix I ′ and with the
usual trace tr ′ : This homomorphism is uniquely determined by mapping the
generators,
′ 1
φ(γµ ) = γµ′ ; φ(η) = γ 5 ≡ ǫµνρσ γµ′ γν′ γρ′ γσ′ (46)
4!
and by the relations defining the homomorphism

φ(I) = I ′
φ(X1 X2 ) = φ(X1 )φ(X2 ) ; φ(aX1 + bX2 ) = aφ(X1 ) + bφ(X2 )
tr X = tr ′ φ(X) (47)

for X, X1 and X2 belonging to this subalgebra; a and b are complex numbers.

pµ , δµν and ǫµνρσ are mapped into the corresponding usual tensors in four
dimensions.

8
 To verify that this is a homomorphism with the properties (47) it is suffi-
cient to verify them on the relations (4,15,16,17,20) defining the subalgebra
for ǫ = 0.
The kernel of this homomorphism is obtained projecting this subalgebra
with (I − ηγ 5 )/2, which commutes with all the elements of the subalgebra.
γµ (I + ηγ 5 )/2 generates the orthogonal subalgebra, establishing an iso-
morphism between this subalgebra and the usual Dirac algebra.
This homomorphism cannot be extended to η1 , since
′ ′
4 = tr (γ 5 − η)η1 6= tr ′ (γ 5 − γ 5 )φ(η1 ) = 0
for any value of φ(η1 ).
Furthermore it cannot be extended to the case ǫ 6= 0; for instance
′ ′
8ǫ δ̄µν = tr(γ 5 − η)γ̄µ γρ ηγργν 6= tr ′ (γ 5 − γ 5 )φ(γ̄µ γρ ηγρ γν ) = 0 (48)
for any value of φ(γµ ).

In order to check the consistency of the defining relations (15-21), let

us give an explicit representation of this extension of the Dirac algebra for
d = 2n even dimensions. Denote by γµ′ the usual 2d/2 -dimensional gamma
′
matrices; define further γ 5 ≡ γ1′ γ2′ γ3′ γ4′ and η ′ ≡ i(d−1)d/2 γ1′ ...γd′ (the γ5′ in
the right-hand-side of the latter expression is γµ for µ = 5 and should not
′
be confused with the pseudoscalar object called γ 5 defined in the former
expression).
The extended algebra elements are given by the following tensor products:
I ≡ I ′ × diag(1, 1, 1, 1)
γµ ≡ γµ′ × diag(1, 1, 1, 1)
′
γ 5 ≡ γ 5 × diag(1, 1, 1, 1) (49)
′
η1 ≡ γ 5 × diag(1, −1, 0, 0)
η2 ≡ η ′ × diag(0, 0, 1, −1)
d
tr X = 21− 2 Sp[(I ′ × diag(1, −1, 1, 1))X]
where Sp is the ordinary trace on 2d/2+2 × 2d/2+2 matrices, while tr is a not
a matrix trace, but it is a cyclic multilinear operator on the space generated
by the gamma matrices, η and η1 .
In this representation all the defining relations for the extended Dirac
algebra hold. All the properties which we derived above hold in this rep-
resentation, including (35), which however does not follow from (39) in the
case r = 2n.

9
1.3 Semi-naive dimensional regularization
In the semi-naive dimensional regularization (SNDR) the Feynman ampli-
tudes are expressed in terms of meromorphic functions of the complex pa-
rameter d = 4−ǫ; in the case of fermionic amplitudes, there is a meromorphic
function for each element of the generalized gamma matrix algebra described
above, written in normal form (NF).
A basis for the gamma matrix normal form quantities belonging to the
generalized Dirac algebra consists in

{ΓA , ηΓA , η1 ΓA , ηη1 ΓA } (50)

{ΓA } = {I, γµ , γ[µ1 γµ2 ] , .., γ[µ1 ...γµr ] , ...} (51)

where [...] indicates antisymmetrization.

It can be decomposed in the following way:

NF = NF0 + NF1 + NF ˆ (52)

NF0 = {Γ̄A , η Γ̄A } (53)
NF1 = {η1 Γ̄A , ηη1 Γ̄A } (54)
{Γ̄A } = {I, γ̄µ , γ̄[µγ̄ν] , γ 5 γ̄µ , γ 5 } (55)

NFˆ consists of the remaining NF terms, which have at least a γ̂µ .

Denote by A0 and A1 the subalgebras generated by NF0 and NF1 respec-
tively, on the complex numbers (constant in ǫ).
Denote by  the subalgebra generated by NF ˆ , multiplied by an entire
function in ǫ, and by any NF term multiplied by an entire function with a
zero in ǫ = 0. The objects in  are evanescent.
One has

A0 A0 = A0 ; A0 A1 = A1 A0 = A1 A1 = A1
 A0 = A0  =  A1 = A1  =  (56)

Denote by T0 the algebra on constants in ǫ, whose generators are the

normal form terms in NF0 , the momenta p̄µ , δµν , ǫµνρσ and the normal forms
obtained from these taking tensor products of these (e.g. δ̄µν , I, p̄µ I ×
I, δ̄µν I ×I, γ̄µ ×η are normal forms in T0 ; these tensor products are introduced
to cover the case of Feynman graphs with more than one open fermionic line).
Denote by T1 the algebra on constants in ǫ, whose generators are the
normal form terms in NF1 and the normal forms obtained taking the tensor
product of at least one NF1 term and T0 normal forms.
Finally denote by T̂ the space formed by normal forms with an hatted
quantity, on constants in ǫ, and by any normal form multiplied by an entire

10
function with a zero in ǫ = 0, and by the elements obtained making a tensor
product of at least one of these elements and those previously defined in T0
and T1 . This is the algebra of evanescent quantities. One has

T0 T0 = T0 ; T0 T1 = T1 T0 = T1 T1 = T1
T̂ T0 = T0 T̂ = T̂ T1 = T1 T̂ = T̂ (57)

The homomorphism φ defined in the previous subsection extends canonically

to T0 , but cannot be extended to T1 or to T̂ .
After the subtraction procedure, a renormalized but still regularized 1P I
Feynman graph Gǫren n , written in normal form, belongs to the abstract algebra
T0 + T1 + T̂ . The subscript n indicates the Lorentz and internal indices of
the Feynman graph.
The algebraic consistence of the extended BMHV algebra guarantees that
it is equivalent to compute Gǫren
n directly in terms of convergent integrals, or
to compute separately the divergent integrals for Gǫunren
n and its counterdia-
gram graphs, and then sum them; the latter procedure is much easier.
Gǫren
n must be evaluated as a four-dimensional quantity Grenn , expressed
in terms of the usual Dirac algebra. This is done in two steps:
i) set ǫ and the hatted objects to zero; denote by Gnǫ=0ren the resulting
quantity, belonging to T0 + T1 ;
ii) Map the abstract objects in the generalized BMHV algebra to the
usual Dirac algebra quantities:

φ(Gnǫ=0ren ) = Gren
n (58)

where Gren n is the renormalized 1P I graph after the regulator has been re-
moved. The map φ must be the trace-preserving homomorphism previously
defined.
It is necessary that this map be a trace-preserving homomorphism to pre-
serve the algebraic properties on convergent diagrams, like making a skele-
ton expansion of a convergent 1P I Green function; these expansions involve
making products of relevant 1P I vertices and dressed propagators, sums over
Lorentz indices and Dirac traces, which are preserved by a homomorphism
preserving the traces. It is clear that this homomorphism is unique. Since it
cannot be defined on T1 , it follows that a consistent renormalization scheme
in the SNDR scheme must be such that the T1 terms are absent from Gnǫ=0ren
at all orders in loops (and hence from the ǫ0 term of the Laurent series for
Gǫren
n , which therefore must belong to T0 + T̂ , while Gn
ǫ=0ren
must belong to
T0 ).
This is a non-trivial requirement, since the T1 -dependent counterterms
can only be local, while non-local terms are produced in perturbation theory;

11
for instance in a unsubtracted two-loop fermionic two-point function (without
its one-loop counterdiagrams) there are generally both non-local poles ( this
is true also in the BMHV case) and non-local finite A1 terms; the point is
that these non-local terms are canceled by the one-loop counterdiagrams. In
the next subsection we will show that this non-trivial property holds at all
orders in perturbation theory.
The second step is trivial in the BMHV scheme, since the BMHV Dirac
algebra for ǫ = 0 is isomorphic with the usual Dirac algebra in four dimen-
sions. This is not so in the SNDR Dirac algebra for ǫ = 0.
We will call a quantity, belonging to the extended Dirac algebra, to be
weakly evanescent if it belongs to the kernel of the homomorphism φ, modulo
evanescent terms. The space of weakly evanescent quantities is called W .
5 5
The kernel of φ on the domain A0 is 1−ηγ 2
A0 . The subspace 1+ηγ
2
A0 is
isomorphic to the usual Dirac algebra. Clearly

W A0 = W W = W (59)

In the general tensor product case

W T0 = W W = W (60)

Summarizing: the regularized quantities (bare action, Gǫren n ) live in a

space with four times as many Dirac algebra elements as in BMHV; Gnǫ=0ren
lives in a space with twice as many Dirac algebra elements as in BMHV;
Gǫren
n differs from Gnǫ=0ren by evanescent quantities; Gren
n is isomorphic to
ǫren
Gn modulo weakly evanescent quantities; projecting each A0 component
5
of Gnǫ=0ren to 1+ηγ
2
A0 one gets an object isomorphic to Gren
n .

1.4 Minimal subtraction in SNDR

The renormalization by minimal subtraction (MS) consists in a minimal sub-
traction procedure and in the evaluation in four dimensions of the formal
objects defined in dimensional regularization.
We define it recursively starting from l = 1 loops.
ǫ(l)
Denote by Gn (k) a l-loop 1P I Feynman graph with momenta k, min-
imally renormalized according to our MS rules up to l′ < l loops but yet
ǫ(l)
unrenormalized at l loops. Denote by Γn (k) the corresponding Green func-
ǫ(l)
tion, which is the sum of various Gn (k).
We define the minimal subtraction procedure for the SNDR in the fol-
lowing way.
ǫ(l)
The minimal subtraction on Gn (k) consists in subtracting all the local
terms, polynomial in momenta and masses, which are:

12
 i) the poles of the Laurent expansion of the meromorphic functions defin-
ing such a graph, both for hatted and non-hatted tensor structures in normal
form;
ii) the T1 terms
ǫ(l)ren
The resulting renormalized but yet regularized 1P I Feynman graph Gn
is evaluated in four dimensions following the procedure described in the pre-
vious subsection.
The l-loop bare action counterterms consist in the local terms subtracted
ǫ(l)
from the 1P I l-loop Green functions Γn (k) according to these rules.
In a renormalizable theory without insertions of composite operators of
dimension larger than five, the only T1 counterterms are A1 terms (there are
no relevant four-fermion terms).
ǫ(l)ren
As in MS-BMHV, we have to show that Gn thus defined has no non-
local poles; furthermore we have to show that it has no non-local T1 term.
′ ǫ(l) ′ ǫ(l)
Consider Gn (k) and Γn (k) to be the corresponding quantity in the
BMHV scheme.
Suppose by induction hypothesis that in the BMHV scheme finite lo-
cal counterterms can be chosen for the renormalized (but still regularized)
′ ǫren(l′ )
relevant 1P I Green functions Γn′ up to l′ = l − 1 loops such that
′
ǫren(l ) ′ ′
ǫren(l )
Γn ′ − Γn ′ is weakly evanescent, so that this difference vanishes after
evaluation in four dimensions (in general the MS-SNDR scheme is equivalent
to a non-minimal BMHV scheme).
Let us show that the same can be done at l loops.
ǫ(l)
Let dn be the superficial degree of divergence of Gn (k). For r > dn ,
ǫ(l) ′ ǫ(l)
∂kr Gn (k) and ∂kr Gn (k) admit a skeleton expansion.
∂kr denotes r derivatives with respect to the external momenta k, with
indices not saturated neither among themselves nor with the Lorentz indices
included in the subscript n.
The difference between the renormalized and still regularized Green func-
tions in the two schemes at l′ < l loops is weakly evanescent by induction
hypothesis, so that using the skeleton expansion property and (57,60) it fol-
lows that for r > dn also
′
∂kr Gnǫ(l) (k) − ∂kr Gnǫ(l) (k) ∈ W (61)
ǫ(l) ′ ǫ(l)
This equation implies that Gn (k) − Gn (k) is a polynomial of degree dn
in the momenta plus a weakly evanescent quantity. The polynomial part de-
pends on the poles in ǫ (of degree less or equal to l) and on the normal forms,
including the T1 terms. An important point in this kind of arguments [23] is
that ∂kr commutes with the operations of extracting the poles or the ǫ0 terms

13
of the Laurent series on which the meromorphic functions are expanded, and
that it commutes also with the operation of projecting the T0 elements to
the kernel of φ and with the extraction of the T1 terms.
ǫ(l)
Making the minimal subtraction on Gn (k) using our rules and making
′ ǫ(l)
an appropriate non-minimal subtraction in the BMHV sense on Gn (k), the
ǫ(l) ′ ǫ(l)
polynomial part of Gn (k) − Gn (k) can be subtracted and the difference
between the renormalized 1P I Feynman graphs, yet regularized, is weakly
evanescent.
We have therefore proven by induction that the MS-SNDR scheme is
equivalent to a (generally non-minimal) BMHV scheme at all orders in per-
turbation theory.
The proof of polynomiality in masses of the minimal subtraction countert-
erms proceeds in the same way [24] and holds also for the T1 counterterms.
Let us show from another point of view that step (ii) is necessary to define
our minimal subtraction scheme; consider a different definition of minimal
subtraction for SNDR, in which only poles are subtracted, and in which the
evaluation procedure in four dimensions includes the rule of setting η1 to zero
and identifying η with γ 5 (this map is not a trace-preserving homomorphism,
as we saw in Subsection 1.2). Let us repeat the above argument, in which
ǫ(l) ′ ǫ(l)
now Gn refers to this modified scheme, while Gn has the same interpreta-
ǫren(l′ )
tion as above. In this case the renormalization parts Γn′ of the skeleton
r ǫ(l)
expansion of ∂k Gn differ from the corresponding renormalization parts in
the BMHV scheme by T1 terms and by weakly evanescent terms. Performing
a trace involving a weakly evanescent renormalization part and a T1 contri-
bution in another renormalization part of the skeleton expansion, one gets
′ ǫ(l)
in general a finite non-local T0 contribution, which is absent in ∂kr Gn (k).
In such a case it is not possible to choose l-loop local counterterms to get
the same renormalized Green functions in the two schemes, up to weakly
evanescent terms or local T1 terms. Therefore one cannot go from the new
scheme to the BMHV scheme with a renormalization group transformation.
This is not possible, so that the proposed scheme is wrong.
The necessity of the subtraction of the T1 terms to have a consistent
renormalization scheme makes our subtraction scheme minimal; the polyno-
miality in the masses of these T1 terms supports the notion that MS-SNDR
is really an MS scheme.
A characteristics of the usual MS scheme is that to compute the l-loop
counterterms one has to compute the O(1/ǫ) part of l-loop Feynman integrals,
the O(1) part of (l − 1)-loop Feynman integrals, the O(ǫ) part of (l − 2)-loop
Feynman integrals, and so on. In MS-SNDR the same happens. In fact the
ǫ(l)
T1 l-loop terms in Γn (k) arise from a counterterm graph with a counterterm

14
containing a NF1 factor (T1 terms are absent at tree level), from a graph with
vertices containing only NF0 factors or from a graph containing at least a
vertex with a NF ˆ factor.
In the first case the Feynman integrals are clearly of (l − 1)-loop order;
in the second case a NF1 term is produced, in the reduction of the graph to
normal form, by anticommuting η with γµ , generating a γ̂µ η1 factor; to get a
T1 term this γ̂µ must be contracted with another γ̂µ , producing a ǫ, so that
only the pole part of the l-loop Feynman integral is involved in producing
T1 ; the same is true in the third case.
ǫ(l)
Actually the subtraction of the l-loop T1 terms in Γn (k) can be done al-
gebraically, without doing any analytic computation; in fact one can subtract
ǫ(l)
algebraically all the NF1 terms (both poles and finite parts) from Γn (k),
subtracting subsequently the remaining poles. (In practice this algebraic
subtraction can be done simply by treating η in the open fermionic lines as
anticommuting with all the gamma matrices and by setting η1 to zero on the
same lines). The analytic computation of the T1 l-loop counterterms matters
ǫ(l+1)ren
only for computing Γn (k).
ǫ(l)
On the contrary, the subtraction of the l-loop poles from Γn (k) requires
some analytic computation, so that from the analytic point of view the T1
counterterms require little effort with respect to the pole counterterms.
Let us discuss briefly finite renormalization, which are needed to recover
the chiral symmetry or more in general to renormalize the Green functions
in a non-minimal way, e.g. at a momentum subtraction point.
They belong to T0 , with the possible addition of T̂ terms to have coun-
terterms respecting d-dimensional Lorentz symmetry in the non-chiral cases;
the discussion is analogous to the one made in the first subsection for the
non-minimal BMHV scheme.
If the tree-level action is reflection symmetric, the bare action in MS-
SNDR is also reflection symmetric.
In a non-minimal SNDR scheme, reflection symmetry is preserved by
choosing reflection symmetric T0 finite renormalization terms.
Obviously non-minimal T1 counterterms are forbidden.

15
2 Applications
After having shown that the MS-SNDR is a consistent renormalization scheme,
let us show that it is convenient from a computational point of view.
In this scheme the renormalized Green functions are the same as in MS-
NDR as long as fermionic loops with an odd number of γ 5 does not occur.
In these cases the computations are exactly the same in the two schemes and
the chiral Ward identities are satisfied.
In the remaining cases, in MS-SNDR the traces with odd number of
η or η1 enter in the game ; in these cases one must check explicitly the
chiral Ward identities; if they are broken, they can be recovered as usual by
adding local finite counterterms to the MS counterterms, order by order in
perturbation theory. These are the cases in which NDR becomes ambiguous
or inconsistent.
In the case of external vector gauge fields, the vectorial Ward identities are
preserved in the MS-SNDR as well as in the MS-BMHV scheme; in fact, since
the vectorial gauge transformations do not involve neither the dimension, nor
η or η1 , the operations of extracting the poles and the T1 terms commute with
gauge transformations, so that the bare action is vector gauge invariant. The
extension of this argument to gauge theories will be discussed later.

2.1 Yukawa models

In the Yukawa models it is possible to use NDR with anticommuting γ 5 with-
out finding inconsistencies; MS-NDR preserves the rigid chiral symmetries.
However coupling these models to external gauge fields and considering the
corresponding Ward identities, NDR is not anymore a consistent regulariza-
tion scheme, due to the presence of the chiral anomalies.
In the MS-BMHV scheme the chiral Ward identities are broken by a large
number of spurious terms, which must be subtracted by introducing non-
invariant local finite counterterms in the bare action. This has been done at
one-loop and partly at two loops in [13], choosing renormalization conditions
which, in the cases in which fermionic loops with an odd number of γ 5 do
not occur, give the same renormalized Green functions as in MS-NDR.
In SNDR the tree-level bare action is
Z
(0) 1 1 1
S = (Dµ φi )2 + ψ̄γµ Dµ ψ − (Dµ ψ̄)γµ ψ + (62)
2 2 2
1
iψ̄yi ψφi + hijkl φi φj φk φl (63)
4!
We have set the dimensional regularization scale µ to one.

16
 We use the same notation as in [13], with the difference that in the tree-
level action and in the gauge transformations on the fermions we replace γ 5
with η; the difference between the two cases being weakly evanescent, this is
an equivalent starting point for making perturbation theory. We restrict our
attention to the massless Yukawa theories. Furthermore we do not require
that Âaµ = 0, which was chosen in [13] to reduce the number of counterterms
in the BMHV bare action.
The fermions have internal indices which are not indicated (ψI ). The
matrices Si , Pi are hermitian.
We will consider a group G which is not necessarily semi-simple.
The fermions belong to a (reducible) representation ta ; under an infinites-
imal transformation

δψ = iǫa ta ψ (64)

We will consider chiral representations

ta = taR PR + taL PL = tas I + tap η ; t̄a = taL PR + taR PL = tas I − tap η

taR = tas + tap ; taL = tas − tap
I +η I −η
PR = ; PL = (65)
2 2
where taR and taL belong in general to different representations.
The Dirac conjugate fermion transforms as

δ ψ̄ = −iǫa ψ̄ t̄a (66)

The scalars belong to a real (reducible) representation (θa )ij ; under an

infinitesimal transformation

δφi = iǫa θija φj (67)

The covariant derivatives are

Dµ ψ = (I∂µ + iAaµ ta )ψ
Dµ ψ̄ = ∂µ ψ̄ − iAaµ ψ̄ t̄a
Dµ φi = (∂µ δij + iAaµ θija )φj (68)

The Yukawa coupling ψ̄y i ψφi , with

yi = Si I + iPi η = Y i PR + Y i† PL
Yi = Si + iPi (69)

17
is invariant under these transformations provided

y j θji
a
+ y i ta − t̄a y i = 0 (70)

or equivalently

Y j θji
a
+ Y i taR − taL Y i = 0 ; Y j† θji
a
+ Y i† taL − taR Y i† = 0

The tree-level action is reflection symmetric.

We use a modified subtraction scheme [26] , in which the bare constants
are chosen of the form
X (l)
cA = h̄l Ndl cA (ǫ)
l≥1
ǫ (l)
X (l)
Nd = (4π)ǫ/2−2 Γ(1 + ) ; cA (ǫ) = cA,−r ǫ−r (71)
2 r≥0

(l)
For all counterterms which do not involve NF1 , in minimal subtraction cA (ǫ)
has no ǫ0 term.
The considerations made in the first section are easily generalized to the
modified subtraction schemes.
Let us discuss the MS-SNDR renormalization at one and two loops, com-
paring it with the MS-NDR computations in the MS-NDR scheme [25] and
in the non-minimal BMHV scheme [13].
Consider first the scalar Green functions; since it is not possible to pro-
duce renormalization parts involving ǫµνρσ , these Green functions have only
fermionic loops with an even number of η’s, which behave in these traces as
the anticommuting objects η2 . Therefore in this case our MS scheme is the
same as the MS-NDR scheme; only poles are subtracted, respecting the chiral
Ward identities. All the spurious terms present in BMHV [13] are absent.
The same is true for the ǫµνρσ -independent parts of the Green functions
involving external gauge fields with or without external scalar fields. There
is none of the spurious terms present in BMHV, with or without Âaµ (the
former can be avoided in the case of external gauge fields setting Âaµ = 0,
as in [13], but they must be included when the gauge fields are promoted
to quantum fields, leading to many other spurious terms in BMHV, besides
those computed in [13]). Let us now consider the relevant fermionic Green
functions. At one loop, multiplying them by η22 , they become identical to
those in NDR (after the replacement γ 5 → η2 ), and are minimal subtracted
as in that case; multiplying them by η12 , they become identical to those in
BMHV (after the replacement γ 5 → η1 ).

18
 Expressing the sum of the two sectors in the base of NF terms involving
η and η1 , MS consists in subtracting, besides the poles, the finite NF1 terms,
that is the A1 terms. Projecting the MS subtracted graphs with η22 , one
has MS-NDR; projecting the MS subtracted graphs with η12 , one has BMHV
graphs renormalized in a non-minimal way (in the BMHV sense), with a
finite subtraction such that in four dimensions the renormalized graphs are
the same as in MS-NDR. This is precisely the renormalization choice made
in [13].
To renormalize the fermionic Green functions at two loops in our scheme
it is sufficient to multiply them by η22 and perform MS-NDR. The terms
(2)
SN F1 are useful only to compute renormalized Green functions at three or
more loops, so we will not compute them here; let us only remark that all
(2)
contributions to SN F1 apart from those with a fermionic loop subdiagram can
be read out from the corresponding spurious terms in [13], replacing γ 5 with
η1 and projecting them with η12 . The difference in the remaining diagrams is
that in [13] the fermionic loop contains γ 5 , not η2 , leading to some difference
in the coefficients of those counterterms, but to the same renormalized Green
functions.
The last renormalization parts to be considered are those involving ǫµνρσ ,
which are those with fermionic loops involving an odd number of η’s, which
behave as the non-naive γ 5 -objects η1 . At one loop the relevant graphs are
the ǫµνρσ -dependent parts of the three-vector and four-vector graphs, which
are the same as in BMHV, since η1 acts as γ 5 in these cases. In our MS
scheme, there are no spurious terms in the bare action, corresponding to the
fact that the anomaly appears in the Bardeen form [21], preserving vectorial
gauge invariance (for a discussion, see [13]).
At two loops, the ǫµνρσ -dependent part of the three-vector graphs involves
again η1 ; the fermionic counterterms are therefore those obtained projecting
the one-loop fermionic counterterms with η12 , that is are the same as in BMHV
in [13]. Therefore the two-loop anomaly terms cancel as in [13], in agreement
with the Adler-Bardeen theorem [9].
The complete bare action in MS-SNDR scheme is
Z
1 1
S= cij Dµ φi Dµ φj + ψ̄γµ (PR cψ + PL c̄ψ )Dµ ψ −
2 2
1
(Dµ ψ̄)(PL cψ + PR c̄ψ )γµ ψ + iψ̄ci ψφi +
2
1 1 a b
cijkl φi φj φk φl + cab Fµν Fµν + SN F1 + ∆S[Aaµ ] (72)
4! 4

19
where

c̄ψ ≡ c̄ψ (Yi , Yi† ) ≡ cψ (Yi† , Yi )

cψ ≡ cψ (Yi , Yi† ) (73)

are hermitian due to reflection symmetry and ∆S[Aaµ ] is the non-naive part
of the pure gauge part of the bare action. It is due to Feynman graphs
containing one or more fermionic traces with odd number of η or η1 , giving
each a Levi-Civita tensor.
At one loop one has

(1) 4
cij = − Yij
ǫ
(1) 1 †
cψ = − Y i Y i
ǫ
(1) 2 †
ci = y j y i y j
ǫ
(1) 48 3
cijkl = − Y(ijkl) + hrs rs
(ij hkl)
ǫ ǫ
(1) 8 1
cab = − S2 (F ) − S2 (S)ab
ab
(74)
3ǫ 3ǫ
and
Z
(1) i i
SN F 1 = − ψ̄η1 γ̂µ Pi yi Dµ ψ + (Dµ ψ̄)η1 yi Pi γ̂µ ψ +
ǫ ǫ
†
ψ̄η1 yj Pi yj ψφi − iψ̄η1 yi tap yi γ̄µ ψAaµ
∆S (1) [Aaµ ] = 0 (75)

where Yi1 i2 ...i2n−1 i2n , S2 (F ) and S2 (S) are defined as in [13], as well as the
group covariant K2ab in the next formula.
We checked explicitly that SN F1 is invariant under vectorial gauge trans-
formations, i.e. under transformations satisfying the constraint ǫa tap = 0 (see
(65)), as expected by the argument given at the beginning of this section.

20
 At two loops one has [25, 13]
(2) 1 8 2 4 3
cij = − hikmn hjkmn + (− 2 + )Yikjk + (− 2 + )Yijkk
12ǫ ǫ ǫ ǫ ǫ
(2) 1 1 † 2 † 2 3
cψ = (− 2 + )Yj Yi Yi Yj − 2 Yj Yi Yj Yi + (− 2 + )Yij Yj† Yi
† †
2ǫ 8ǫ ǫ ǫ 2ǫ
(2) 4 2 † 2 1 † †
ci = ( 2 − )Yjk yk yi yj + ( 2 − )yk yj yi yj yk +
ǫ ǫ ǫ ǫ
1 1 2
( 2 − )yk (yi yj yj + yj yj yi )yk + 2 yk (yi† yj yk† + yj† yk yi† )yj +
† † † †
ǫ 2ǫ ǫ
1 † † 1 †
yk yj yi yk yj − hijkl yk yj yl
ǫ ǫ
(2) 192 96 48 96 48
cijkl = (− 2 + )Yninjkl + Ynijnkl + (− 2 + )Ynijkln +
ǫ ǫ ǫ ǫ ǫ
96 48 24 12 6
hmnij [− 2 Ymkln + (− 2 + )Ymknl + ( 2 − )Ymp hpnkl +
ǫ ǫ ǫ ǫ ǫ
3 6 3
hmnpq hpqkl + ( 2 − )hnpqk hmpql ]
ǫ2 ǫ ǫ
(2) 2 ab
cab = − K2 (76)
ǫ
(2)
where in the expression for cijkl the symmetrization in the indices i, j, k, l is
understood.
We didn’t compute ∆S (2) [Aaµ ], which might be different from zero; in that
case, it must be an ǫµνρσ term, and it is the same as in BMHV scheme used
in [13]. Observe that ∆S (2) [Aaµ ] does not have to do with the anomaly, since
it does not involve the totally symmetric tensor dabc .
(2)
We didn’t compute SN F1 which, as explained in the previous section, is
not necessary to compute the two-loop renormalized Green functions.
In absence of the external gauge fields MS-SNDR preserves the rigid chiral
Ward identities at all loops. In fact it is not possible to produce the relevant
terms with ǫµνρσ , which are at the origin of the breaking of the symmetries.
In presence of the external gauge fields the chiral Ward identities might be
broken by the ǫµνρσ contributions, so that a finite counterterm ∆S (2) [Aaµ ]non min
might be necessary to recover the chiral Ward identities; at more than two
loops ∆S (l) [Aaµ ] can contain also parity even terms, coming from an even
number of Levi-Civita tensors.

Let us finally discuss what would have happened if, as discussed at the
end of the previous section, we had not subtracted the A1 terms.
In the computation of the anomaly, the difference from the correctly MS-
subtracted case and the one now discussed consists of the counterterm graph

21
 Figure 1:

in Figure 1, in which there is the insertion of the A1 counterterm in (75), lead-

ing to a non-local finite contribution, which conspires to the absence of two-
loop corrections to the anomaly; apart from non satisfying the Adler-Bardeen
theorem, the suggested theory would have the following inconsistency: in the
one-loop approximation it has the same renormalized Green functions as in
a consistent renormalization scheme, the non-minimal BMHV scheme used
in [13]; from general renormalization theory arguments it follows that the
renormalized Green functions in two different renormalization schemes can
be connected by a renormalization group transformation; being the one-loop
relevant terms equal, it follows that the marginal Green functions can differ
at two loops only by a local term; however we just found a case in which this
difference is non-local; this is not possible, so that the proposed scheme is
inconsistent, in agreement with the arguments in the previous subsection.

2.2 Gauge theories

In the Yukawa theories only the pure gauge Green functions do not satisfy
the chiral Ward identities in the MS-SNDR scheme; the remaining Green
functions satisfy them at all orders in perturbation theory. The vector Ward
identities are respected in this scheme.
In chiral gauge theories no Green function satisfies the chiral Ward iden-
tity in this scheme (at order high enough in loops); however SNDR breaks
the chiral symmetries in a gentler way than BMHV.
As long as graphs involving fermionic loops with an odd number of γ 5 -like
objects are absent, the chiral Ward identities will be satisfied. In general, we
can expect that the level at which MS is not sufficient in SNDR is the same
at which NDR reveals its inconsistencies. Apart from the graphs mentioned
in the previous section, it is reasonable to expect that other superficially di-
vergent graphs containing the subgraph in Figure 2(a) (giving non-vanishing
contributions to the two-loop anomaly [13]), like the one in Figure 2(b) need

22
 (a) (b) (c)

Figure 2:

to be subtracted in a non-minimal way.

In presence of composite operators the need of introducing a finite renor-
malization appears already at low loop order; for instance coupling an ex-
ternal gauge current to the theory, the minimally subtracted triangle graph
involving it and two chiral gauge fields gives the anomaly in the Bardeen form
which does not respect the chiral Ward identities so that a finite counterterm
must be added to restore them.
Apart from this issue, the MS-SNDR scheme for a chiral gauge theory
differs qualitatively from that in Yukawa theories for the appearance of γ 5
itself (not η or η1 ) in the bare action. It is only at this point that the
bare action starts living in the extended algebra four times bigger than the
BMHV algebra, as can be seen observing that, for instance, in absence of
counterterms with γ 5 in a chiral gauge theory the relevant bare vectorial
part of the vertices contains γµ and ηη1 γµ , but not ηγ 5 γµ and η1 γ 5 γµ .
For example consider the graph in Figure 2(c), contributing to the vertex
for an external gauge field (it is chosen to be external to avoid anomaly
cancellation); in MS it generates counterterms involving 3!1 ǫµνρσ γν γρ γσ /ǫ =
γ 5 γ̄µ /ǫ, while at lower loop level only vertices involving η and η1 appeared in
the bare action. Similar counterterms can be expected in the graph in Figure
2(b); this counterterm cannot be replaced by ηγ̄µ /ǫ, since the difference is
not weakly evanescent (see the example in eq. (48)).
As far as vector gauge invariance is concerned, let us argue that MS-
SNDR preserves it in the case of a non-abelian vector gauge theory with
chiral Yukawa couplings (i.e. the gauged version of the Yukawa model in the
previous subsection, with tap = 0).
The bare action is BRST-invariant at tree level. Suppose that it is BRST-
invariant at (l − 1) loops; then the l-loop functional generator of the 1P I
Green functions renormalized up to (l − 1) loops, called Γǫ(l) , satisfies the

23
Slavnov-Taylor identities, which in the Zinn-Justin form [27] read

X
l−1
(0) ǫ(l) ǫ(l) (0)
S ∗Γ +Γ ∗S =− Γǫ(m) ∗ Γǫ(l−m) (77)
m=1

′
In MS-SNDR Γǫ(l ) , for l′ < l, is finite and belongs to T0 for ǫ = 0; due to
(57), for ǫ = 0 the right-hand-side of (77) belongs also to T0 ; it follows that
the T1 local term of Γǫ(l) is BRST invariant, as well as its poles; therefore the
bare action is BRST invariant also at l loops.

2.3 Comparison with other NDR schemes

The NDR prescriptions proposed in the past have, as regularization schemes,
some inconsistency, so that they require some care in using them, usually on
a diagram-by-diagram basis, while in a true regularization scheme one should
be able to get unambiguous results by using the bare action and the Feynman
rules, whatever computational routine is followed.
Furthermore no proof has been given, in any of these schemes, that the
renormalized Green functions are equal, modulo finite local renormalization
terms , to those in some consistent scheme, e.g. the BMHV scheme, at all
orders in perturbation theory.
The issue of preserving the Ward identities comes only after settling these
points, and it is solved by adding local finite terms to the bare action. In
absence of a consistent regularization system it is not true that checking the
Ward identities on the relevant vertices one is guaranteed that the theory
is renormalized consistently. In fact consider a set of ‘renormalized’ Green
functions which differs from any consistent set of renormalized Green func-
tions by non-local quantities 2 Then check the Ward identities on the relevant
terms defined at a subtraction point. If there are no anomalies, one can add
local finite counterterms to restore the Ward identities on the relevant terms.
Clearly the new set of Green functions is still inconsistent, since non-local
l-loop terms cannot be canceled by local l-loop counterterms.
2
i.e. such that there is no renormalization group transformation transforming one set
of Green functions in the other; one can construct easily similar pathological cases by
choosing, within a consistent regularization system, two different one-loop sets of renor-
malization conditions, say A and B, and by computing a subset of the two-loop Green
functions in system A, another subset in system B; the resulting mixing of Green func-
tions differs by non-local terms from the set of Green functions computed within a single
scheme. Working without a consistent regularization system without great care it is likely
that a similarly inconsistent system is produced at some point.

24
 Figure 3:

The dangers of using algebraically inconsistent regularization systems

have been reviewed in [19], where it is emphasized the fact that the renormal-
ization conditions are compatible with the Ward identities does not guarantee
the validity of the quantum action principle, in absence of a true regulariza-
tion scheme.
Let us consider briefly some of the NDR schemes discussed in the litera-
ture.
To avoid confusion with previously defined objects, call Γ5 the alge-
braically undefined γ 5 -like object; γ 5 is defined in (9).
In [15] Γ5 is chosen to be anticommuting in open fermionic lines and in
loops with an even number of Γ5 , while in the loops with an odd number of
Γ5 the trace rules are given modulo some evanescent term, to be fixed by the
Ward identities. The authors emphasize that this is not a true regularization
scheme, but that it is a convenient prescription for one-loop computations.
In [17] Γ5 is chosen to be anticommuting in open fermionic lines and in
loops with an even number of Γ5 , while in the loops with an odd number of
Γ5 one reduces the traces with an odd number of fermions, before evaluating
them, to the case of one Γ5 only, by using anticommuting Γ5 ; this rule does
not preserve the Bose symmetry.
In [18] a non-cyclic definition of the trace is used, together with a totally
anticommuting Γ5 . Bose symmetry is broken in this scheme; to avoid this
problem one can add the rule of choosing an appropriate reading point in
the graphs.
In [28] it is suggested the use of γ 5 together with a trace rule which, in
the case of even number of γ 5 , acts as if the γ 5 were naive, while in the
case of odd number of γ 5 it acts as the BMHV trace. In this case there are
ambiguities in the treatment of objects like tr({γ 5 , γµ }{γ 5 , γµ })γν γρ , which
is zero or not depending on the fact that ({γ 5 , γµ }{γ 5 , γµ }) is given or not
its value 4ǫ. For instance using this trace rule the renormalization of the
two-loop graph in Figure 3 requires some care due to this ambiguity.
Trueman [29] has used the trick of renormalizing the axial vector-fermion-
fermion vertex as the corresponding vector-fermion-fermion vertex multiplied
by γ 5 , as a way of restoring the Ward identities and for treating γ 5 as if

25
it were anticommuting, while keeping the BMHV rules. Unlike the other
prescriptions reviewed in this subsection, this one is consistent; however,
apart from the fact that this trick has been defined only for certain cases, it
has the drawback of using the BMHV scheme, with all its spurious anomalies
(i.e. finite counterterms which must be added to restore the Ward identities);
for instance in the case of the Yukawa theory in [13] a trick similar to the
one proposed by Trueman has been used, with the result that the number of
spurious anomalies is conspicuous. Trueman’s trick works well as long as one
is able to make computations without resorting to the full form of the bare
action, implicitly defined; for a three-loop application of this trick see [8].
Let us finally describe the MS-SNDR as a NDR prescription.
Γ5 is chosen to be anticommuting in open fermionic lines and in loops
with an even number of Γ5 , while in the loops with an odd number of Γ5 it
is considered to be equal to γ 5 .
Apart from subtracting the poles, the MS-SNDR requires a finite sub-
traction in the fermionic subdiagrams which occur in the fermionic traces
with an odd number of Γ5 ; this finite subtraction is chosen such that the cor-
responding fermionic renormalization part has the same value, apart from
evanescent terms, as the fermionic renormalization part occurring in an open
fermionic line, or in a closed fermionic line with an even number of Γ5 (min-
imal subtraction of the T1 terms).
In MS-SNDR the BMHV γ 5 does not appear in the tree-level action, but
can be produced in divergent graphs, by application of the Levi-Civita tensor
on the gamma matrices. The resulting counterterms must not be confused
with those with Γ5 ; in particular the counterterms (Γ5 − γ 5 )/ǫ cannot be
neglected ((η − γ 5 )/ǫ is not weakly evanescent).

I thank M. Raciti and F. Riva for help and discussions.

References
[1] G. ‘t Hooft and M. Veltman, Nucl. Phys. B44 (1972) 189.

[2] C.G. Bollini and J.J. Giambiagi, Phys. Lett. 40B (1972) 566;
J.F. Ashmore, Nuovo Cimento Lett. 4 (1972) 289;
G.M. Cicuta and E. Montaldi, Lett. Nuovo Cimento 4 (1972) 329;
D.A. Akyeampong and R. Delburgo, Nuovo Cimento 17A (1973) 578.

[3] G. ‘t Hooft, Nucl. Phys. B61 (1973) 455.

26
 [4] P. Breitenlohner and D. Maison, Comm. Math. Phys. 52 (1977) 11.

[5] G. Bonneau, Nucl. Phys. B171 (1980) 477; B177 (1981) 523

[6] J.G. Korner, N. Nasrallah and K. Schilcher, Phys. Rev. D41 (1990) 888;
R. Ferrari, A. Le Yaouanc, L. Oliver and J.C. Raynal, Phys. Rev. D52
(1995) 3036.

[7] A.J. Buras and P.H. Weisz, Nucl. Phys. B333 (1990) 66.

[8] S.A. Larin, Phys. Lett. B303 (1993) 113.

[9] S. Adler and W. Bardeen, Phys. Rev. 182 (1969) 1517.

[10] D.R.T. Jones and J.P. Leveille, Nucl. Phys. B206 (1982) 473;
M. Bos, Nucl. Phys. B404 (1993) 215.

[11] C.P. Martin and D. Sanchez-Ruiz, ‘Action principles, restoration of

BRS symmetry and the renormalization group equation for chiral non-
Abelian gauge theories in dimensional renormalization with a non-
anticommuting γ 5 ’ , preprint hep-th/9905076.

[12] C. Schubert, Nucl. Phys. B323 (1989) 478.

[13] M. Pernici, M. Raciti and F. Riva, ‘Dimensional renormalization of

Yukawa theories via Wilsonian methods’, preprint hep-th/9912248.

[14] W.A. Bardeen, R. Gastmans and B. Lautrup, Nucl. Phys. B46 (1972)
319

[15] M. Chanowitz, M. Furman and I. Hinchliffe, Nucl. Phys. B159 (1979)

225.

[16] S. Gottlieb and J.T Donohue, Phys. Rev. D20 (1979) 3378.

[17] M.I. Adelhafiz and M. Zralek, Acta Phys. Pol. B18 (1987) 21.

[18] H. Nicolai and P.K. Townsend, Phys. Lett. B93 (1980) 111 ;
D. Kreimer, Phys. Lett. B237 (1990) 59;
J.G. Korner, D. Kreimer and K. Schilcher, Z. Phys. C54 (1992) 503.

[19] G. Bonneau, Int. Journ. Mod. Phys. A5 (1990) 3831;

Phys. Lett. 96B (1980) 147.

[20] P.A. Baikov and V.A. Il’in, Theor. Math. Phys. 88 (1991) 789.

27
[21] W. A. Bardeen, Phys. Rev. 184 (1969) 1848.

[22] J. Collins, ‘Renormalization’ Cambridge Monographs on Mathematical

Physics, Cambridge University Press (1984) .

[23] W.E. Caswell and A.D. Kennedy, Phys. Rev. D25 (1982) 392.

[24] J.C. Collins, Nucl. Phys. B92 (1975) 477.

[25] M. E. Machacek and M.T. Vaughn, Nucl. Phys. B222 (1983) 83; B236
(1984) 221; B249 (1985) 70.

[26] E. Braaten and J.P. Leveille, Phys. Rev. D24 (1981) 1369.

[27] see e.g. J. Zinn-Justin, ‘Quantum Field theory and Critical Phenomena’,
Clarendon press, Oxford (1989).

[28] M. Veltman, Nucl. Phys. B319 (1989) 253.

[29] T.L. Trueman, Phys. Lett. 88B 331 ; Z. Phys. C 69 (1996) 525.

28