Gauge Invariant Pauli-Villars Regularization of Chiral Fermions

IU-MSTP/8; |hep-th/9603062
June 1997

Gauge Invariant Pauli-Villars Regularization of Chiral Fermions

Kiyoshi Okuyama* and Hiroshi Suzuki

Department of Physics, Ibaraki University, Mito 310, Japan

ABSTRACT

We extend the idea of the generalized Pauli-Villars regularization of Frolov and
Slavnov and analyze the general structure of the regularization scheme. The gauge
anomaly-free condition emerges in a simple way in the scheme, and, under the
standard prescription for the momentum assignment, the Pauli-Villars Lagrangian
provides a gauge invariant regularization of chiral fermions in arbitrary anomaly-
free representations. The vacuum polarization tensor is transverse, and the fermion
number and the conformal anomalies have gauge invariant forms. We also point
out that the real representation can be treated in a straightforward manner and
the covariant regularization scheme is directly implemented.

★ e-mail: okuyama@mito.ipc.ibaraki.ac.jp
f e-mail: hsuzuki@mito.ipc.ibaraki.ac.jp

1. Introduction

In this paper, we extend the idea of the generalized Pauli-Villars (PV) regular-
ization of chiral fermions proposed by Frolov and Slavnov some years ago [1]. We
analyze the structure of the regularization scheme on the basis of a regularization
of composite current operators, as has been performed [2] for the generalized PV
regularization proposed by Narayanan and Neuberger [3]. This type of analysis
provides a simple and transparent way to see the structure of the regularization of
fermion one-loop diagrams.

In the past, several studies regarding this proposal were also performed: A
characterization from the viewpoint of the analytic index [3], a verification of the
Ward-Takahashi (WT) identities and an evaluation of the fermion number anomaly
by a direct use of Feynman diagrams [4] and, more recently the generalization to
arbitrary anomaly- free complex gauge representations in curved space-time [5] . We
believe our formulation in this paper also provides a unified view concerning these
results.

A regularization based on a diagrammatical calculation, such as the PV reg-
ularization [6], in general, preserves the Bose symmetry among external gauge
vertices; thus it gives rise to the consistent [7] gauge anomaly [8]. Since the con-
sistent anomaly is not covariant nor invariant under gauge transformation on the
external gauge fields, a Bose symmetric gauge invariant regularization of chiral
fermions, if possible, exists only for anomaly-free gauge representations. How this
anomaly-free requirement emerges in the scheme is the main concern in the gauge
invariant regularization of chiral fermions. As we will see throughout this article,
the anomaly-free condition emerges in a simple way in the extension of the gener-
alized PV scheme of Ref. [1]. This is the interesting and important property of the
proposal.

The organization of this paper is as follows: In §2, we extract the essence of the
Lagrangian given in Ref. [1], which was originally constructed only for the spinor
representation of the SO(10) gauge group. We then present a general framework

2

for extension to other gauge representations. In §3, the regularized form of the
composite current operators, namely the gauge current, the vector U(l) current
and the axial U(l) current, and the trace part of the energy-momentum tensor are
summarized.

Based on the above setting, real gauge representations are studied in detail
in §4. This case allows a straightforward treatment because of the anomaly-free
nature of the representation. We also point out that the resultant regularized
operators are nothing but those in the covariant regularization scheme in Ref . [9] .

In §5, the complex gauge representation, which is important in view of appli-
cations, is studied. This part of the paper is essentially the result given in Ref. [5]
specialized to flat space-time, but we include it for the sake of completeness and
for comparison with the real representation case.

In §6, as an illustration of our formulation, the fermion contribution to the
vacuum polarization tensor [1,3,4,2] is calculated for arbitrary anomaly-free rep-
resentations and for arbitrary regulator functions. In §7, we evaluate the fermion
number anomaly [10] and the conformal anomaly [11] within our formulation. We
obtain the covariant (or gauge invariant) [12,7] anomalies.

We comment briefly on the relation of the "Weyl formulation" [1] and the
vector-like formulation [3] in §8. The final section is devoted to conclusions.

Throughout this article, we work in Euclidean spacetime, ix° = x 4 , Aq = iA^
^7° — 7 4 an d 75 = i7°7 1 7 2 7 3 = 7 4 7 1 7 2 7 3 , and in particular, 7^ = —7^, 75 = 75,
9iiv = -5 (iv and e 12U = 1.

3

2. Generalized PV Lagrangian

The PV Lagrangian due to Frolov and Slavnov [1] can be generalized as follows:

£ = ^i]/)^ - -^MC D ^ T + -ip T C\)M^ilj + 0X00 - -4>M'C D ~f + -0 r Cj ) M /t
2 2 2 2

(2.1)

In (2.1), ip and <p are the fermionic and bosonic Dirac spinors* respectively, each
of which possessing the gauge and an internal space indices. We shall call this
internal space "generation." The first generation component of ip®, is the original
massless fermion to be regularized, and other components of ip and are massive
regulator fields. The number of generations may be infinite [1]. Co is the Dirac
charge conjugation matrix/ and the covariant derivative Tj) is defined by

= >f(d ll -igAp' a P R ), PR^ 1 -^: (2-2)

where T a is in general a reducible representation of the gauge group.

The mass matrices M and M' in (2.1) possess gauge and the generation indices.
From the statistics of the fields, these matrices must satisfy

M T = M, M' T = —M'. (2.3)

The matrix X, which also has gauge and the generation indices, has been intro-
duced to avoid the appearance of tachyonic bosons. Such a tachyonic state leads
to an unwanted pole singularity in the regulator function f(t) (see (4.5)). Since
the mass squared of the bosonic fields is given by — X~ X M'X T 1 M'\ if

M' = -XM'X T , (2.4)

then the mass squared is positive definite, M'M'^. The hermiticity of the action,

* We have added free left-handed (spectator) spinors to the Lagrangian in Ref. [1]. We will

only consider regularization for a single chiral fermion.
t Cn-fC^ = -r T , C D ^C D X = 75 T , Cl = C7 D 1 and C£ = -C D .

4

on the other hand, requires

X+=X, [T a ,X}=0. (2.5)

Finally, the mass matrices should satisfy

T a M = ~MT aT , T a M' = -M'T aT (2.6)

to be gauge invariant. Once a certain set of matrices, M, M' and X, which satisfy
(2.3)-(2.6), and a suitable gauge generator T a , which is reduced to the original
representation T a on i/jq, are found, the gauge invariant Lagrangian (2.1) may be
constructed. Our general setting (2.1) allows various extensions of Ref. [1] which
will be discussed in the subsequent sections.

To reformulate the generalized PV regularization as a regularization of com-
posite current operators, we need the formal propagators of ip and in a fixed
background gauge field. We introduce a two component notation

In terms of these variables, the Lagrangian (2.1) is written as

c = -m T \ U + -$ T [ D U 2.8

2 \ -MC D J 2 ^ % xij) -M'Cd)
where the transpose of the covariant derivative is defined by

JD T = {-d, - igAlT aT P T R )^ T . (2.9)

T

This definition (and analogous definitions of and ffi below) is motivated by

5

the matrix notation in the functional space. Namely,

]p(x,y)=^(<%-igA«(x)T a P R )5(x-y), (2.10)

and thus

]p T (x, y) ee (flj - igA%{y)T aT Pl)'f T S{y - x)

= (~d* - igAl{x)T aT Pl)^ T 5{x - y). ^

Once writing the Lagrangian in the form (2.8), it is straightforward to find the
propagator in a background gauge field A^:

1 /At 1

-MC D ,4 ifl

and

^T$(x)$ J (y)

jjy^ jj) + M /f M' # + m'm 6(x _ v

, X - lT 0t T 1 c7 DM /t 1 I

(2.13)

(2.14)

In the above expressions, we have introduced

= ^(d II -zgA;T a P L )^]p,
where Pz, = (1 — 7s)/2, and

0t T = -^(fy + igApr aT Pl). (2.15)

It is interesting to note that the hermitian conjugate of the covariant derivative
automatically emerges in the inverse operator.

\ In deriving these formulas, it is necessary to use relations such as ffiC^M = Cj}M0 T ,
P T C^ D M^ = & D M^$, etc. These follow from the gauge invariance of the mass term, (2.6).

6

3. Composite current operators in PV regularization

The central quantity in our analysis is the gauge current, whose classical form
is defined by a functional derivative of the action (2.1) with respect to the gauge
field:

J» a (x

-T aT pT^ T '
l^T a P R

1 T

2 \ X^T a P R

T aT P^ T X T '

(3.1)

Therefore in the PV regularization by Frolov and Slavnov, the regularized gauge
current is defined by

= - lim tr

2 2/->z

l»T a P R

-T aT P^ T '

(T^(x)^ (y) /

+

T aT P^ T X T '

X^T a P R

: S(x - y)

lim tr

^T a P R %P

LT${xW (y)

t>P + MM 1

+ ^T a P R %P

0# + M'M

—A 5 ( x -y)

/ n/rn

(3.2)

where the trace is taken over the generation, gauge and Dirac indices. We note that
this definition is in accord with the standard Feynman diagrammatical calculation:
Further derivatives of (3.2) with respect to the background gauge field gives a multi-
point one loop vertex function. We will later illustrate such a calculation of the
vacuum polarization tensor. Equation (3.2) therefore summarizes the structure of
the regularization scheme in a neat way.

Strictly speaking, a PV Lagrangian such as (2.1) alone cannot definitely specify
the regularization scheme. As is well-known, one must supplement the following

7

prescriptions [6] to the Lagrangian: 1) The integrand of the momentum integra-
tion must be summed over all the generations prior to the momentum integration.
2) The momentum assignment for all the fields (the original fermion and the regu-
lators) should be taken the same. It is thus important in (3.2) that the trace over
the generation index be taken before the equal point limit y — > x, according to
prescription 1). Equation (3.2) is as it stands a formal implemention of prescrip-
tion 2): The momentum assignment is common for all the generations. However,
one should always be careful in the uniform momentum assignment in actual calcu-
lations such as (6.4). Those two underlying prescriptions in the PV regularization
are understood throughout this paper. With this caution in mind, we use the term
"Lagrangian level regularization."

Another important composite current operator in the chiral gauge theory is
the fermion number current. We define it as the Noether current associated with
a global U(l) rotation [4,2]:

ip(x) -> e ta ip(x), ip(x) -> ip(x)e ia ,
<P{x) -> e ia (f)(x), ~4>{x) -> ^(x)e~ ia ,

(3.3)

or, in terms of the two component notation,

/ e ia o \ / e ia

By localizing the infinitesimal parameter a, the Noether current is defined by

C -> C- (d^J^ix) -aB(x), (3.5)

where

J»(x) = ^ r I ) * + ^ r I ' I $, (3.6)

8

and the explicit breaking part B(x) is given by

B(x) =

T

r

-2iM'C D

(3.7)

By defining the composite current operator by the propagators, we have from (3.6),

= - lim tr

2 2/-ke

lim tr

-1

7^ T X T

+ I X7 , ) (^(*)* T (y)

+ MM 1

<J(x -y)

+ M'M ,i[

S(x - y)

(3.8)

If the composite current operator is well regularized, we may derive the correspond-
ing WT identity,

PV '

(3.9)

as a result of the naive equation of motion. Under a Lagrangian level regulariza-
tion, a possible anomaly associated with a certain global symmetry should arise as
an explicit symmetry breaking term in the Lagrangian. Thus the vacuum expec-
tation value of B(x), the right-hand side of (3.9), gives rise to the fermion number
anomaly. We will later verify that this is, in fact, the case. This situation is
analogous to an evaluation of the gauge anomaly in chiral gauge theories by the
conventional PV regularization, where the gauge symmetry is explicitly broken by
the PV mass term.

We may equally well use another definition of the regularized fermion number
current. It is defined by the Noether current associated with a global axial U(l)

9

rotation [4,2]:

i/>(x) -> e iQ75 ^(x), tf(x) -> ^(x)e iQ75 ,

_ . (3.10)

By the same procedure as above, we find the associated Noether current

= lim tr

0# + MAT + MM"

(3.11)

In what follows, we find that it is always possible to choose M and M', such that
composite operators in (3.8) and (3.11) are regularized, if the gauge representation
is free of the gauge anomaly. In this case, we can see that the currents (3.8)
and (3.11) are the same object: We first note from (2.2) and (2.14) that

11 1

+ MM* ~ L plp^ + MM^ R 2 + MM^

holds, as does an analogous relation for the bosonic part. Putting these into (3.8)
and (3.11) and noting that there exists no constant vector independent of A^, we
see that only the first term of (3.12) survives; consequently (3.8) and (3.11) are the
same operator (note 75 Pr = Pr). Of course this is an expected result because only
the right-handed fields are coupled to the background gauge field. After observing
the equivalence of (3.8) and (3.11), we use (3.8) as the fermion number current in
what follows.

Another interesting operator is the trace part of the energy- momentum ten-
sor Tff(x), which is defined in the original theory by*

$ Q i$ifo -> tfoiflifo ~ a(x)TH(x), T£{x) = ^ ? V>o, (3.13)

* This definition requires some explanation: If one simply uses the standard definition of
the energy-momentum tensor of the spinor field, —3 times our result will be obtained.
Our definition, following Ref. [13], is specified by the general coordinate invariance in the
background gravitational field (see Rcfs. [13,2] for more details).

10

where the variation of the field is

^o(x) - e - Q ^/ 2 ^o(x), Mx) - ^e"^/ 2 .

(3.14)

By generalizing the rescaling of the field (3.14) to all the regulator fields, the
regularized version of the trace part of the energy-momentum tensor is defined by

(WW

1

= - hm tr

2 y^x

+

o -\JD T X

T^(x)^ T (y)

+ (-l){T*(x)* T (y))( V ~ 20y

2^ x ■

T$(x)$ r (y)

+ (T<f>(x)<f> T (y)

o -f^^ T '

(3.15)

In (3.13) and (3.15), $ = $ - 0, $ T = $ T - $ T and

= +igA a T a P R ) ) U) T = (- d, +igA a T aT Pl)^ T .

(3.16)

Noting 5(x -y)$ y = -Ip x 5{x - y) and D ] MM l = MM' D '. and thus

1 rA tA 1

,JD = JD-

typ + MM\ pip + MMV

(3.17)

we finally have

= - lim tr 00+

2 2/->ai

1

H — lim tr

2 j/-»a;

00t + MMt
1

6(x-y)

-1

#0

+ MMt

5{x-y) + ftlp

t>p + M'M
-1

Pt> + M'M

4^ -y)

(3.18)

The composite operators, the gauge current (3.2), the fermion number cur-

11

rent (3.8), and the trace part of the energy-momentum tensor (3.18) will be ana-
lyzed in detail in the following discussion.

4. Real representations

As was noted in the Introduction, the gauge invariant regularization of a chi-
ral fermion is possible only for anomaly-free gauge representations. The situation
is simple for real representations, because the anomaly-free condition is always
fulfilled by the presence of a matrix U which transforms the original represen-
tation to the adjoint representation. As we will see below, the generalized PV
Lagrangian (2.1) can utilize this fact, and this is an advantage of the present
framework.

For any real representation T a , there exists a unitary matrix U such that

_ T aT = _ T a* = UT atf m ( 41 )

For a real-positive representation, U is symmetric, and for a pseudo-real represen-
tation, U is anti-symmetric. For both cases, we can make the choice:

T a = T a <g> 1, M = tf®m, M' = U ] <g) m', X = l®x, (4.2)

where the first index acts on the gauge and the second acts on the generation index.
It turns out that the nature of m and m! are quite different depending on whether
the representation is real-positive or pseudo-real. We thus treat them separately.

When the chiral fermion belongs to a real-positive representation of the gauge
group, we can take, for example,

m= (o i) A ' "*'=(-i o) A ' *=(! -i)' (4 ' 3)

where A is the cutoff parameter. It is readily verified that these matrices satisfy
(2.3)-(2.6). Since MAft = (J °)A 2 and M 1 M' ] = (J J)A 2 , the regularized gauge

12

current (3.2) is given by

U» a (x)) PV = limtr

^T a P R ^-^f(JpJpyA 2 )6(x - y)

(4.4)

where we have defined the regulator function

f(t) =

(t+ l)(t + 2)'

(4.5)

which vanishes rapidly as t — > oo and satisfies

/(0) = 1, Jimt/'W = Em^t) = J™^/ (3) W = 0,
lim */(*) = lim t 2 /'(t) = lim t 2 f"(t) = lim t 3 / (3) (t) = 0.

t^oo t^oo t^oo t^oo

(4.6)

Due to the rapid damping property in the second line, the gauge current (4.4) is
well regularized, and the limit y — > x can safely be taken.

Let us next consider the pseudo-real case, for which the matrix U is anti-
symmetric. In (4.2) we may choose

m

\

2
-2

4
-4

/0 1
1

A, m!

3
3

A, (4.7)

and

x = diag(l, -1,1, -!,•••)•

(4.8)

These matrices again satisfy (2.3)-(2.6). Note that in this case we have introduced

13

an infinite number of regulator fields. The mass squared is given by

/o 2

MM 1 = 1 (

\

/1 J

A 2 , M'M' ] = li

\

A 2 .

(4.9)

With the above choice of mass matrices, the gauge current operator (3.2) becomes

(J^ a {x)) PV = limtr

y^x

lim tr

y^x

oo

-l) n $ft

$ft 0ft + n 2 A 2

5{x - y)

l»T a P R ft^f{lpft/h 2 )5{x - y)

(4.10)

where the regulator function f(t) is defined by [1]

f(t)= v tUl ^

— 2.^1 t + n 2 sinh(7r ^)'

n=— oo v v y

(4.11)

which again has the desired properties (4.6).

In (4.4) and (4.10), we see that all the divergences are made finite gauge invari-
antly. In fact these expressions are nothing but those of the covariant regularization
scheme [9], which is formulated as a gauge invariant damping factor f(IJ)ft/K 2 )

* Although it is not necessarily required, if one prefers the regulator function (4.11) in the
real-positive case,

m =

\

V

/ 1

-1 o

A, to' =

3

-3

A.

■■■/

and x in (4.8) may be chosen.

14

insertion in the original fermion propagator. The covariant regularization is known
to give the covariant anomaly [12,7]:

D, (J^(x)) PV A =™^^ Pa HT a F» v F pcT ), (4.12)

where the field strength is defined by F^ = (d^A® - d v A^ + gf abc A b ^Al) T a and
the right-hand side vanishes due to the anomaly-free condition tr(T a {T b , T c }) = 0?
Starting from (4.4) or (4.10), we can directly evaluate the gauge anomaly (4.12)
and the calculation is almost identical to the passage from (7.1) to (7.6) [15]. The
regularization scheme due to Frolov and Slavnov therefore gives a Lagrangian level
implementation of the covariant regularization scheme in Ref. [9]. This aspect of
the generalized PV regularization is studied in detail in Ref. [15].

Going back to consideration of the pseudo-real case, we have chosen the mass
matrices (4.7) simply because the explicit summation over the generation index
can be performed as (4.11). We may make other choice of m and m! in (4.2), and
it may be thought that a more clever choice could reduce the number of regulator
fields to finite value. We show below, however, that an infinite number of regulator
fields are always needed, at least when relying on the construction (4.2).^

Let us first assume the number of the generation is finite, and m, m! and x are
finite dimensional matrices. From (2.3), m T = —m (note U is anti-symmetric).
Therefore if m is an even dimensional matrix, it may be put into a block diagonal

f The vacuum overlap approach [14] in the lattice chiral gauge theory, which is closely related

to the generalized P V regularization [3] , is known to give the consistent anomaly
\ Hence it trivially satisfies the Wess-Zumino consistency condition. This is consistent with

the fact that we are treating a Lagrangian level regularization which respects the Bosc

symmetry among gauge vertices.
§ If all the fields belong to the same irreducible representation, (2.5) and Schur's lemma imply

the structure X = 1 ® x.

15

form by an orthogonal transformation of i/j:

/mi <S> s

m

, e =

(4.13)

and

/ ' raf <g> I

MM' = 1

1712 I

\

m 2 k ® I J

I =

(4.14)

The number of massless fermion fields, if such fields exist, is always even. Since
we are constructing a regularization for a single massless fermion, the matrix m
should be odd dimensional. On the other hand, from (2.5), x is hermitian, and by
a unitary transformation and a rescaling of 0, it may be put into the form

x = diag(l, 1,

l

1,-1,-1,

(4.15)

Then (2.4)implies raj ■ = —XiXjin'^, and m' has the structure

3""ir

m

Y
Y T

A,

(4.16)

where Y is a kxl matrix. If k < I, dimker Y > l—k, and if k > /, dimkery T > k—l.
For both of these cases, there exists at least one linear combination of fa which
remains massless. This is an unwanted massless bosonic field, and we should
take k — l. Therefore x and m' must be even dimensional.

16

We now have an odd number of fermions and an even number of bosons in
the same gauge representation. However, the PV condition for the ultraviolet
divergence reduction [6] always requires the same numbers of fermionic and bosonic
degrees of freedom. From the above argument, this is impossible when the number
of the generation is finite. This shows that, at least within the construction (4.2),
an infinite number of regulator fields are always needed. As we have observed, they
in fact regularize the theory.

By the same procedure as for the gauge current, the regularized U(l) global
current (3.8) becomes

{J^{x)) PV = lim tr

y^x

7 M

^-^f(]p]pVA 2 )6(x-y)

(4.17)

with an appropriate regulator function f(t). Similarly, the trace part of the energy-
momentum tensor (3.18) is given by

(T>l(x)) pv = l - lim tr \f{]J)ft/K 2 )5{x - y) + f(^0/A 2 )5(x - y)} . (4.18)

y^x

Therefore, for real representations, the generalized PV Lagrangian provides a
complete gauge invariant regularization of the gauge current (4.4) or (4.10), as well
as the fermion number current (4.17) and the trace part of the energy-momentum
tensor (4.18).

17

5. Complex representations

The generalized PV regularization [1] was originally formulated for the irre-
ducible spinor representation of SO(10), i.e., an anomaly-free complex represen-
tation, which is important for an application to the standard model. Quite re-
cently [5] the construction has been successfully generalized for arbitrary anomaly-
free complex representations. We include these results in this section (specializing
them to flat space-time) and compare the situation with that of the real represen-
tation in the previous section.

For a generalization of the PV Lagrangian to arbitrary complex representations,
it is crucial to introduce a doubled representation [5]:

rj-a _

(5.1)

With this doubling of the gauge representation, the following choice of matrices
in (2.1) satisfies (2.3)-(2.6):

M = a 1 ®m, M' = ia <g> m', X = a A <g> 1, (5.2)
where o % is the Pauli matrix. The relation to the original SO (10) model [1] is*

F n C, ia 2 -> c,

The matrices m and m! are chosen as

2

m

\

G'

/I

11-

(5.3)

A, m'

J

\

A.

(5.4)

/

Although the regulator fields must belong to the doubled representation (5.1) to

* The gamma matrix for the spinor representation satisfies the Clifford algebra Tj} = 25ij
and is hermitian. The gauge generator is defined by T a = i[Ti,Tj}/2 and the irreducible
representation is projected by (1 + Tn)/2. The "chiral" matrix Tu is defined by Tu =
— iTir 2 ■ ■ ■ Tio and satisfies {rn, Tj} = 0, r| x = Tn and = 1. The "charge conjugation"
matrix C has the properties CT.C- 1 = -Tf, CT^C- 1 = -T^, & = C' 1 and C T = -C.

18

have a non-vanishing mass, the original massless fermion ipo must be projected by
(l + o" 3 )/2 to have the original complex representation T a , rather than the doubled
representation T a .

Let us now consider the regularized composite operators. For example, the
regularized gauge current (3.2) is given by

{J^))pv

- lim tr

2 y^x

1 oo

:-i) n ^#

# M f + n 2 A 2

5(x - y)

H — lim tr

2

-v)

(5.5)

- lim tr

2 J/^a;

^T a P R ^^f(]p]pyA 2 )5(x - y)

H — lim tr

2 y->x

l^ 3 T a P R ±-5(x - y)
iip

where the regulator function is given by (4.11) and we have used the fact that the
first generation fermion is projected by (1 + cr 3 )/2.

Similarly, the fermion number current (3.8) and the trace part of the energy-
momentum tensor (3.18) become

(J"(x))

PV

= - lim tr

2 y->x

H — lim tr

2 y^x

o 1

-fvjpSix-y)

(5.6)

and

'f($ft/A*)S(x -y) + f(0t0/A 2 )5(x - y)] . (57)

In deriving (5.5)-(5.7), we have taken the trace over the generation index. In
the final line of (5.5) and of (5.6), the first term is completely regularized. Since the
regulator function f(t) damps rapidly enough as t — > oo, the limit y — > x can safely

19

be taken to give the finite result. It should also be noted that an infinite number of
regulator fields are always needed to balance the fermionic and the bosonic degrees
of freedom [1]. On the other hand, the last terms of (5.5) and (5.6) do not yet have
a regulator function, and the expression is ill-defined in general. Compare these
with (4.10) and (4.17) for the real representations. The origin of the complication
in the complex representation is that it is in general not anomaly-free, and it
is impossible to distinguish the anomaly-free representations from the anomalous
ones at the level of Lagrangian construction. Therefore we must supplement the
regularization scheme with the anomaly-free condition.

In fact, as first pointed out in Ref. [1] (see also Refs. [3,4]) and generalized
in Ref. [5], these unregularized terms become finite if and only if trT a = and
tr T a {T b ,T c } = 0, i.e., if free of the gauge (and Lorentz) anomaly. To see this, we
note the perturbative expansion

h = h + h [ - 3 - m h + r$ ( - 9iPR) h ( - 3iPB) h + " - (5 - 8)

Using 5(x—y) = J d A k e~ tk ^ x ^ / (27r) 4 and the momentum representation of A®(x),
we see that the last terms in (5.5) and (5.6) generate Feynman integrals with an
insertion of o" 3 . The momentum integration of the one, two, three and four-point
functions is power-counting divergent, and higher point functions are convergent.
Therefore if

tr(cr 3 r ai ...T a ") =tr(T ai ..-T a ") + (-l)^ 1 tr(T a " • • • T ai ) = 0, for n < 4,

(5.9)

the integrand of the power-counting divergent expression vanishes. As is easily
verified [4,5], this condition is equivalent to trT a = and tr(T a {T b , T c }) = 0. In
the framework of Ref. [1] , the anomaly-free complex representation is distinguished
from anomalous ones in this way, and the generalized PV regularization works only
for the anomaly-free should be the case.

* When the construction of this section is applied to real representations [5], Eq. (5.9) holds
for all n due to the presence of the matrix U.

20

6. Vacuum polarization tensor

As an illustration, we compute the fermion contribution to the vacuum po-
larization tensor [1,4,2] in our formulation, (4.10) or (5.5). For simplicity, we
first consider the case of the real representation (4.10) and comment later on the
anomaly-free complex representation (5.5).

The vacuum polarization tensor is defined by the first functional derivative of
the gauge current (4.10) with respect to the background gauge field:

5{J^ a {x))

PV

WW)

(6.1)

We assume the following form of the regulator function:

Cn.t

t + ml/A 2 '

(6.2)

The above examples, fit) in (4.5), fit) in (4.11), and the generalized PV in general,
are certainly contained in this class of functions. We also assume that c n and m n
are chosen so as to satisfy (4.6). From the definition of the covariant derivative,
(2.2) and (2.14), we have

5(J^ a (x))

PV

A=0

S(gAUz))

-- — tr(T a T b ) hmtr( P L *f j 5(x - 2)7" X]

+ ° n o \i0S(x - zW + Six - zWi

S(x-y) ).

(6.3)

All the derivatives in (6.3) act on everything to their right. We then use the mo-
mentum representation of the delta functions, 8(x — z) = f d 4 q e~ iq ( x ~ z } j '(27r) 4 etc.

21

The vacuum polarization tensor (6.1) is then given by

U^ ab (q)

to:

2 + TO 2 —k 2 + TO 2

+ tr[P L7 ^ + ^) 7 ^]^-

-(A; + q) 2 + to 2 — k 2 + to 2 J

(6.4)

The subsequent steps are standard: We introduce the Feynman parameter to com-
bine the denominators, shift the integration momentum* and take the trace of the
gamma matrices. Noting the definition of f(t) (6.2), we have

U^ ab (q) = -ti(T a T b ) J dx J

j£k_

x <g

jJLV

2 £

Cfik^

-k 2 — q 2 x(l — x) + to 2 ^ [~k 2 — q 2 x{l — x) + to 2 ] 2

- mi - *)w - ri) E ht , _ x) +<? ]

1 oo

= --±,tr(T«T b )JdxJdt

X

(6.5)

where t = -k 2 /A 2 and s = -g 2 x(l - x)/A 2 .

Now if lim^ 00 i/(i) = as in (4.6), the quadratically divergent gauge non-
invariant term disappears after a partial integration of the first term. Therefore

* It is important to shift all the integration momenta in the same way. Otherwise, one obtains
a gauge non-invariant (but finite) piece which is proportional to g^ v .

22

we obtain the final result
W vah {q)

1 o

= ^L t r(T a T b )(gV-^V) jdxx{l-x) j dt ^ - j

fit) f dt f(t)-f(t)

s

00

24tt 2

tr(T a T 6 )(gV-^V)

-g 2 '3 "°2 ' J ' t

(6.6)

where f(t) is an arbitrary function satisfying (4.6) in the first line, and we have
used (4.11) in the last line. Since f(t) — f(t) = 0{t), we have set the limit of the
last integral to zero for A — > oo. This is a general formula for arbitrary f(t) in (6.2)
satisfiying (4.6) and reproduces the results in Refs. [1,4]. From the last expression,
it is obvious that the coefficient of log A 2 , which gives the fermion contribution to
the one loop beta function, is independent of the regulator function (6.2) [4,2]. The
last constant, on the other hand, depends on the specific choice of the function (for
example, for f(t) = f(t) in (4.5), the last integral in (6.6) is 21og(7r/2) — log 2).

Let us next consider the case of anomaly-free complex representations. In the
regularized gauge current (5.5), the second term does not contribute to the vacuum
polarization tensor due to (5.9) (n = 2). The same calculation as above therefore
gives (6.6) with the coefficient tr(T a T 6 )/(487r 2 ) = tr(T a T & )/(247r 2 ), the correct
result for a chiral fermion in a complex representation T a .

As expected, the formulas (4.10) and (5.5) give a transverse form without any
gauge non-invariant counter terms. Similar calculations of the fermion one loop

f If instead, c„ = (-1)™ and m n = y/\n\ A are used in (6.2), f(t) = t[*((i+l)/2)-*(t/2)]-l,
where ^(z) is the digamma function, and lim^oo tf(t) = 1/2^0. The partial integration
in (6.5) then has an additional surface term, — ti(T a T b )A 2 g flv /(32tt 2 ) (the last integral
in (6.6) is log(7r/2) [4]). The appearance of the gauge non- invariant piece is somewhat
puzzling, because we started with a manifestly gauge invariant expression (4.10). The
resolution in the present formulation seems to be the following: When we compute the
divergence of the gauge current (4.12), we encounter an integral J °° dttf(t) (see (7.5))
which is ill-defined for this regulator function. Therefore the WT identity (4.12), and as a
result, the transverse condition of the vacuum polarization tensor, cannot be derived with
this choice.

23

vertex functions and a verification of the WT identities [4] may be pursued. Since
the regularized form (4.10) and (5.5) for anomaly-free representations are finite and
manifestly gauge invariant, the requirements implicit in gauge invariance should
automatically be fulfilled.

7. Covariant anomalies

The generalized PV regularization also provides a reliable way to evaluate
non-gauge anomalies in the anomaly-free chiral gauge theories. Let us start with
the fermion number anomaly [10]: The Majorana-type PV mass term in the for-
mulation naturally provides the source of the fermion number anomaly. We first
take directly the divergence of the regularized U(l) current for the real represen-
tation (4.17): *

d»{J li (x)) PV = dplim tr

y >x

(7.1)

To evaluate this, we introduce the normalized eigenfunctions of the hermitian op-
erators, Pt> and $p [12],

^0<p n (x) = )&<pn(x), 0]pUn(x) = A^n(z), (7.2)

where X n is real and positive. From this definition (by appropriately choosing the
phase) we see

Jp^n = A n 0n, W<t>n = A n </?n- (7.3)

We first use the completeness relation of <p n (x), J2 n 4> n (x)4>n(y) = 8(x — y), in (7.1).

| If we had /(_0 2 /A 2 ) instead in (4.17) (this regulator is known [16] to give the consistent
anomaly), we would obtain no fermion number anomaly, (J^(x)) = 0.

24

The calculation then proceeds as follows:

(J»(x)) PV = d,

= -E>«/ A2 )

A 2 '

0n # ¥n + 4> ] n$Vn

(7.4)

n

= -i lim tr f/(^^t/A 2 )(J(x - y) - f(ft]/)/A 2 )6{x - y)

y^x L

where we have used (7.3) in several steps. The subsequent calculation is identical
to the anomaly evaluation in the path integral framework [12]: i) 8{x — y) —
j d4jfe e <fc(a!-»)/(27r) 4 , ii) shift e ifca; to the left, iii) scale -> A/c^ and expand
f{]p]pyA 2 ) by 1/A. Finally by using (4.6), we have

tr

7 (2vr)4 e \/(W/A2)

oo

4^ / A#)dimrA 4 T i^to(^)+itr(V'«') )

A^oo 1

(7.5)

and consequently,

9 M (J M (a;)>p y = -itr

(2^

327T 2 ll \ r tw r pa)-

7.6)

This reproduces the well-known gauge invariant form of the fermion number anomaly [10].

§ Since the fermion number U(l) rotation and the non-Abelian gauge transformation com-
mute, the Wess-Zumino consistency condition [7] implies a gauge invariance of the fermion
number anomaly.

25

We may also compute the fermion number anomaly using the WT identity (3.9).
It can be verified that direct evaluation of the right-hand side of the equation again
gives the correct anomaly. Namely,

It is obvious that this leads to the last line of (7.4), and thus (7.6).

For the anomaly-free complex representation, the regularized U(l) current is
given by (5.6). However, by repeating the above manipulations, it is easy to see
that the last, seemingly unregularized term does not contribute to the anomaly.
Namely, the divergence of the last term identically vanishes. The divergence of
the first term of (5.6), on the other hand, gives one half of (7.6), but with the
doubled gauge generator (5.1) (note the structure constant is common for T a and
—T a *). Thus the fermion number anomaly in the complex representation also
results in (7.6), when rewritten in terms of the original gauge generator T a .

Let us next consider the conformal anomaly represented by (4.18) and (5.7).
From (7.5), for both cases we have

(7.7)

PV

1

d 4 k

e- ikx [f(]p]pyA 2 ) + f(fl]p/A 2 )}e

ikx

oc

,2

(7.8)

o

which reproduces the correct gauge invariant result [11,13].

2G

8. Relation to the vector-like formalism

Finally we briefly comment on the relation of the present Lagrangian (2.1),
which may be called the Weyl formulation, to the generalized PV regularization
proposed by Narayanan and Neuberger [3], the so-called vector- like formulation.

Let us begin with the complex representations. We introduce new variables [4]

XR =

l + <r

Pr

and

¥R

1 + cr

Pr

V : /

03

V : /

XL = io — - — PlCd

a

<PL = ia

P L C D

V • J

—T

^3

V ! J

(8.2)

where we have assigned even generation indices for the fermions, and odd number
indices for bosons. Then the PV Lagrangian (2.1) with (5.2) and (5.4), discarding
the left handed spectator fields, is rewritten in a vector-like form of Ref. [3]

C = xPx - X(NP R + N^P L ) X + <plp<p - <pN'<P,

(8.3)

where the covariant derivative is vector-like:

Jp^^-igA-T-).

(8.4)

Reflecting the fact that the original theory is chiral, the mass matrix N has a
non-trivial analytic index [3], dimkeriVtiV — dimkeriViV^ = 1:

/0 2 \ /l \

N =

4
6

V

A, N' =

J

\

A.

(8.5)

Therefore, we see that the generalization [5] of the Weyl formulation to arbitrary
complex representations is basically equivalent to the vector-like formulation.

27

How does the anomaly-free requirement emerge in the vector-like formulation?
According to Fujikawa [2], the regularized gauge current for (8.3) is expressed as

(J» a (x)) = - lim tr
2 y~-*x

^T a —f(Jp 2 /A 2 )6(x - y)
tip

1 ,.

+ - lim tr

2 y->x

^ l5 T a ±-5(x-y)
tip

5.6)

Similarly, the axial U(l) current associated with x( x ) ~^ e JQ75 x(^) and (p(x)
e ta ^ip(x) is

(4( X )) = 2y^ X tT

^l 5 -^f(Jp 2 /A 2 )S(x-y)

H — lim tr

2 y->x

(8.7)

and the vector U(l) current associated with x( x ) e ia x( x ) an d <f( x ) ~^ e ia (p(x)
is,

J^(x) ) = - lim tr
2 y *x

^-0f(0 2 /A 2 )5(x-y)

H — lim tr

2 y->x

^75 0S(x - y)

For non-anomalous gauge representations, it can be argued that the first two
currents (8.6) and (8.7) are in fact regularized [3,4,2]. Demonstration of this point
requires a somewhat detailed form of the Feynman integral and goes as follows:
The last terms in (8.6)-(8.8) are evaluated as

T a 7 5

1

n=0

1 i^-d

75

\

x tr

;i

> r £ ai

III

/

^ . . . ryVn

\
/

(8.9)

28

We then change the integration variable k^ to

k,

kfi k\^ ' ' ' k

n/j,

(8.10)

and insert C^Cd = 1 in the trace. Shifting Co to the right-hand side trans-
poses the gamma matrices, and it can be expressed as a transpose of the prod-
uct of gamma matrices. Finally, by renaming all the subscripts (1,2, •■■,n) — >
(n, • • • , 2, 1), we see the integrand in (8.9) is proportional to

tr

/ ( T<1
1

1

\

+

X

r d 4 k
J W?

tr

r

\

J

75

1

75

\

+ h + --- +

"7

in.

7

M2

'7

;t n }_

\

.11

For n < 3 the momentum integration is power counting divergent, but in the first
two lines, the coefficient for n < 3 vanishes provided that (5.9), which is equivalent
to trT a = tr(T a {T b , T c }) = 0, holds. Therefore for anomaly-free representations,
it can be argued that the last terms in (8.6) and (8.7) are finite. On the other hand,
the coefficient of the last line of (8.11) is proportional to ti{T a T h ) ^ for n = 2
and n = 3, and the above argument cannot be applied (J^ and are different
objects).

What corresponds to the unregularized U(l) current J fJ/ (x) in the Weyl formu-
lation (2.1)? It is the Noether current associated with a U(l) rotation, t/j(x) — >

3 • 3

e taa t/j(x) and <j)(x) — ► e taa <p(x). Since the total Lagrangian is invariant under
this rotation, the current is conserved (d fJj J^(x) = 0) and thus cannot be used as
the fermion number current (which should be anomalous). In fact in the Weyl
formulation,

JfI{x) )p V = <^ V ^) + 07^ 3 0(*)>
= \ lim tr [ 7 V-U W7A 2 )5(:r - y)

- lim tr

2

(8.12)

29

and the last term is not regularized (note Jj) in this expression contains Pr =
(l + 75)/2).

We have observed that the Weyl formulation in §5 (complex representation)
is basically equivalent to the vector- like formulation in Ref. [3]. An advantage
of the Weyl formulation is, however, that the requirement of the anomaly-free
nature emerges in a rather simple way; it requires only a power-counting, and the
above argument based on the change of the momentum integration variable and
the charge conjugation invariance is effectively shortcut by the introduction of T a ,
the doubled representation, and the matrix cr 3 .

Let us now turn to real representations. It is obvious that for real-positive
representations, the Weyl formulation can be non-equivalent to the vector-like for-
mulation because it only requires a finite number of regulator fields ((8.3) on the
other hand always requires an infinite number of such fields to have a non-trivial
analytic index). Also, for the real representations we can construct the Lagrangian
such that the seemingly unregularized terms do not appear from the beginning
((4.10) and (4.17)). In the Weyl formulation, the fact that all the real representa-
tions have no gauge anomaly can be incorporated into the Lagrangian construction.
This is another advantage of the Weyl formulation.

9. Conclusion

In this paper, we have studied the general structure of the generalized PV
regularization proposed by Frolov and Slavnov on the basis of a regularization of
composite operators. We have observed that the PV Lagrangian provides a gauge-
invariant regularization of the chiral fermion in arbitrary anomaly-free gauge rep-
resentations. The generalization [5] to the arbitrary complex representation is ba-
sically equivalent to the vector-like formulation in Ref. [3] , and real representations
can be treated in a straightforward manner. As the gauge current is regularized
in a gauge invariant way, the vacuum polarization tensor, for example, is found
to be transverse. We have also computed the fermion number anomaly and the

30

conformal anomaly within our formulation and the gauge invariant form of the
anomalies were reproduced.

A practical calculation of multi-point vertex functions is simpler if one starts
directly from the covariant regularization [9], because one can then choose a con-
venient form of the regulator function f(t). Nevertheless, the very existence of a
Lagrangian level gauge invariant regularization makes the renormalizability and
the unitarity proofs of the anomaly-free chiral gauge theories (at least conceptu-
ally) simpler: No gauge non-invariant counter term is needed to compensate for
the breaking of gauge symmetry by the regularization. In particular, the standard
model may be directly treated in the scheme.

It seems to us, however, that the real importance of a possibility to construct
such a Lagrangian level gauge invariant regularization lies in a possible implication
on the lattice chiral gauge theory, in which a consistent treatment of chiral fermions
has been a long standing problem. In fact, several proposal have been made on the
basis of the generalized PV Lagrangian [17]. A remark on this problem is found in
Ref. [18].

It would be interesting to consider a supersymmetric extension of the general-
ized PV regularization, as a supersymmetric, gauge invariant one-loop regulariza-
tion.

Acknowledgements: We thank Prof. T. Fujiwara and H. Igarashi for enlightening
and helpful discussions. We are grateful to Prof. K. Fujikawa for his series of
lectures, 'Application of path integral in quantum field theory," given at Ibaraki
University, which motivated the present work. We are also grateful to Prof. A. A.
Slavnov for clarifying some crucial points we had misunderstood in the previous
version. The work of H. S. was supported in part by the Ministry of Education
Grant-in-Aid Scientific Research Nos. 07740199 and 07304029.

31

REFERENCES

1. S. A. Frolov and A. A. Slavnov, Phys. Lett. B309 (1993), 344.

2. K. Fujikawa, Nucl. Phys. B428 (1994), 169; Indian J. Phys. 70A (1996), 275.

3. R. Narayanan and H. Neuberger, Phys. Lett. B302 (1993), 62.

4. S. Aoki and Y. Kikukawa, Mod. Phys. Lett. A8 (1993), 3517.

5. L. N. Chang and C. Soo, Phys. Rev. D55 (1997), 2410.

6. W. Pauli and F. Villars, Rev. Mod. Phys. 21 (1949), 434.
S. N. Gupta, Proc. Phys. Soc. A66 (1953), 129.

7. J. Wess and B. Zumino, Phys. Lett. 37B (1971), 95.

W. A. Bardeen and B. Zumino, Nucl. Phys. B244 (1984), 421.

8. S. L. Adler, Phys. Rev. 177 (1969), 2426.

J. S. Bell and R. Jackiw, Nuovo Cim. 60A (1969), 47.
W. A. Bardeen, Phys. Rev. 184 (1969), 1848.

9. K. Fujikawa, Phys. Rev. D29 (1984), 285.

10. G. 't Hooft, Phys. Rev. Lett. 37 (1976), 8.

11. R. Crewther, Phys. Rev. Lett. 28 (1972), 1421.

M. Chanowitz and J. Ellis, Phys. Lett. 40B (1972), 397; Phys. Rev. D7
(1973), 2490.

12. K. Fujikawa, Phys. Rev. D25 (1982), 2584; Phys. Rev. D21 (1980), 2848;
D22 (1980), 1499(E); Phys. Rev. Lett. 42 (1979), 1195.

13. K. Fujikawa, Phys. Rev. Lett. 44 (1980), 1733; Phys. Rev. D23 (1981), 2262.

14. R. Narayanan and H. Neuberger, Phys. Rev. Lett. 71 (1993), 3251; Nucl.
Phys. B412 (1994), 574; Nucl. Phys. B443 (1995), 305.

15. K. Okuyama and H. Suzuki, Phys. Lett. B382 (1996), 117.

16. L. Alvarez- Gaume and P. Ginsparg, Nucl. Phys. B243 (1984), 449, and
references therein.

32

17. S. A. Frolov and A. A. Slavnov, Nucl. Phys. B411 (1994), 647.

A. A. Slavnov, Phys. Lett. B319 (1993), 231; Phys. Lett. B348 (1995), 553.

18. K. Haga, H. Igarashi, K. Okuyama and H. Suzuki, Phys. Rev. D55 (1997),
5325.

33