The Perturbative Gross Neveu Model Coupled to a Chern-Simons

Field: A Renormalization Group Study

V. S. Alves∗ , M. Gomes, S. V. L. Pinheiro∗ and A. J. da Silva

Instituto de Fı́sica, USP

arXiv:hep-th/9810106v1 14 Oct 1998

C.P. 66318 - 05315-970, São Paulo - SP, Brazil

(1998)

Abstract

In 2+1 dimensions, for low momenta, using dimensional renormalization we

study the effect of a Chern-Simons field on the perturbative expansion of

fermions self interacting through a Gross Neveu coupling. For the case of

just one fermion field, we verify that the dimension of operators of canoni-

cal dimension lower than three decreases as a function of the Chern-Simons

coupling.

I. INTRODUCTION

Effective field theories is a subject of great interest in theoretical physics not only due to
their potential applications but also because they provide new insights into the way we look

at field theories [1]. From this perspective nonrenormalizable models have acquired a new
status as they may become physically relevants at low energies [2]. The point is that, if the
scale of energy one is interested is low enough, the ambiguities due to the virtual states of high
energy do not show up or, equivalently, are not meaningful. On the energy interval where
this happens the theory proceeds as an usual renormalizable one. Nonetheless, as observed

in [3], the use of a mass independent regularization is almost mandatory to guarantee that
high order counterterms can be effectively neglected.

1
 The ultraviolet behavior of the Green functions may be changed by a rearrangement of
the perturbative series. In fact, the incorporation of vacuum polarization effects in general

improves the convergence properties of the resumed series; this mechanism is well known
to be operative in the context of the 1/N expansion. In particular, Gross Neveu [4] or
Thirring [5] like four fermion interactions which in (2+1) dimensions are perturbatively
nonrenormalizable become renormalizable within the framework of the 1/N expansion [4].

This result has motivated a series of investigations on the properties of these theories [6].
In particular, using renormalization group (RG) methods, it has been proved that the N
component Gross Neveu model in 2+1 dimensions is infrared stable at low energies but has
also a nontrivial ultraviolet stable fixed point. These facts indicate that the theory could

be perturbatively investigated if the momentum is low enough. This actually would be the
only remaining possibility for small N.
It has recently been conjectured that in 2 +1 dimensions, besides the 1/N expansion,
there is another way to improve the ultraviolet behavior of Feynman amplitudes. By cou-
pling fermion fields to a Chern-Simons field the scale dimension of field operators could

be lowered possibly turning non renormalizable interactions into renormalizable or, better,
super-renormalizable ones. Using a sharp cutoff to regulate divergences, this idea was tested
in [7] where the effect of the CS field over massless self-coupled fermions with a quartic,
Gross–Neveu like, interaction was studied.

In this communication we will pursue this study further by considering massive fermions
and adopting dimensional renormalization [8] as a tool to render finite the Feynman am-
plitudes. In this way we evade the ambiguity problem associated with the routing of the
momentum flowing through the associated Feynman graphs [9]. Nevertheless, it should be

stressed that our calculations are valid insofar, as said above, the effect of the higher or-
der counterterms can be neglected. Otherwise, new couplings should be introduced. Our
investigations, restricted to the case of fermions of just one flavor, i. e. N = 1, show that,
differently to what happens for large N,the renormalization group beta function has only a

trivial infrared stable fixed point. Moreover, the operator dimensions of the basic field and

2
composites of canonical dimension lower than three are monotonic decreasing functions of
the CS parameter. This indicates that the Feynman amplitudes have a better ultraviolet

behavior if the underlying theory is renormalizable. However, no improvement in the ultra-

violet behavior seems to occur if the composite operators have canonical dimension bigger
than three.
Our work is organized as follows. In section II some basic properties of the model as

Feynman rules, ultraviolet behavior of Feynman diagrams and comments on the regulariza-
tion procedure are presented. The derivation and calculation of the renormalization group
parameters are indicated in section III. Section IV contains a discussion of our results as well
as our conclusions. Details of the calculations of the pole part of the relevant amplitudes

are described in the appendices A and B.

II. A QUARTIC INTERACTION

We consider a self-interacting two–component spinor field minimally coupled to a CS

field. The Lagrangian density is given by

1 µνα 1
L= ǫ ∂µ Aν Aα + ψ̄(i 6 ∂ − m)ψ + ψ̄γ µ ψAµ − G(ψ̄ψ)(ψ̄ψ) + (∂µ Aµ )2 , (2.1)
2πα 2λ

The Dirac field ψα represents particles and anti-particles of spin up and the same mass m

(the parameter m is to be taken positive) [10]. The Gross-Neveu term in (2.1) is the most
general Lorentz covariant quartic self-interaction, for the Thirring-like vector interaction is
not independent but satisfies : (ψ̄γ µ ψ)(ψ̄γµ ψ) := −3 : (ψ̄ψ)(ψ̄ψ) :. λ is a gauge fixing
parameter but for simplicity we will always work in the Landau gauge, formally obtained by
letting λ → 0. In this gauge, the Green functions may be computed using the Feynman rules

depicted in Fig. 1. For convenience, we have introduced auxiliary dotted lines, hereafter
called auxiliary GN lines, to clarify the structure of the four fermion vertex.
Divergences show up, the degree of superficial divergence of a generic graph γ being

d(γ) = 3 − NA − NF + V, (2.2)

3
where NA and NF are the number of external lines associated with the propagators for the the
Chern–Simons and the fermion fields, respectively; V denotes the number of quartic vertices

in γ. The model is of course nonrenormalizable and the number of counterterms necessary

to render the amplitudes finite increases with the order of perturbation but, to a given
order, the number of counterterms is finite. To do calculations we will employ dimensional
renormalization starting at the space-time dimensio d. It is therefore convenient to introduce

a dimensionless coupling g and a renormalization parameter µ through G = (g/Λ)µǫ and

α → αµǫ , where ǫ = 3 − d must be set zero at the end. The massive parameter Λ must
be considered much bigger than any typical momenta and than the fermion mass m; it sets
the scale which limits the region where our results are valid. Divergences will appear as

poles in ǫ and a renormalized amplitude is given by the ǫ independent term in the Laurent
expansion of the corresponding regularized integral. However, at one loop level no infinities
will remain after the remotion of the regulator. This is so because the poles for a graph γ
may occur only at even values of the degree of superficial divergence of γ [11]. Moreover, it
is easy to check that asymptotically, i. e., for zero external momenta, one loop graphs with

even degree of divergence are odd functions of the loop momentum and therefore vanishes,
after symmetric integration.

III. RENORMALIZATION GROUP

As known, Green functions of renormalizable models, which have been made finite by
the subtraction of pole terms in the dimensionally regularized amplitudes, satisfy a ’t Hooft-
Weinberg type renormalization group equation [12]. Nonrenormalizable models require spe-
cial consideration since the form of the effective Lagrangian changes with the order of pertur-
bation. However, at sufficient small momenta, such that the effect of the new counterterms

may be neglected, the Green functions will still approximately satisfy the RG equation.
Thus, although being nonrenormalizable, for small enough momenta, the Green function of
the theory (2.1) satisfy the following renormalization group equation

4
 ∂ ∂ ∂ ∂
[Λ +µ +δm +β − Nγ]Γ(N ) (p1 , . . . pN ) ≈ 0, (3.1)
∂Λ ∂µ ∂m ∂g

where Γ(N ) (p1 , . . . pN ) denotes the vertex function of N fermion fields (since Aµ is not a
dynamical field we shall not consider vertex functions having external vector fields). The
symbol ≈ means equality in the region where all counterterms different from those terms
already present in (2.1) can be neglected. As a consequence of the Coleman–Hill theorem,

which states that all radiatives corrections to the CS term are finite, [13], the beta function
for the CS coupling α vanishes identically; that explains why the term with a derivative
with respect to α is absent from (3.1).
The coefficients δ, β and γ in equation (3.1) may be obtained by formally computing the

action of the differential operator over the two point and four point Green functions. For
the two point function, up to second order in the coupling constants, we have

g ǫ (2) (2) (2)

µ I + αµǫ I2 + α2 (1 − T )µ2ǫ I3
Γ(2) (p) = i(6 p − m) +
Λ 1
gα (2) g2 (2)
+ (1 − T )µ2ǫ I4 + 2 (1 − T )µ2ǫ I5 , (3.2)
Λ Λ

where the limit ǫ → 0 must be understood. In the above expression Ii , i = 1, . . . , 5, denote

the regularized Feynman amplitudes. In particular, the graphs ascribed to I3 , I4 and I5 have
been depicted in the figures 2, 3 and 4, respectively; T is an operator to remove the pole
(2)
term in the amplitudes to which it is applied. As mentioned earlier, the amplitudes I1 and
(2)
I2 which are associated with one loop diagrams are finite. Inserting (3.2) into (3.1) allows

us to determine the coefficients δ and γ as follows. Initially notice that as Λ enters into the
perturbative expansion only in the combination g/Λ, fixing β in lowest order as being equal
to g eliminates all contributions of the term with the derivative with respect to Λ in (3.1).
After that, up to the order we will study, in (3.1) there will be no mixing of higher order
contribution to β with those to γ and δ.

Using the expansions,

δi,j g i αj ,
X
δ= (3.3)
i,j

5
 γi,j g i αj ,
X
γ= (3.4)
i,j

where the sum is restricted to i + j ≤ 2, we get

δ1,0 = δ0,1 = γ1,0 = γ0,1 = 0, (3.5)

δ0,2 = −2i(A3 + B3 ) γ0,2 = −iB3 (3.6)

2m m
δ1,1 = − (A4 + B4 ) γ1,1 = − B4 (3.7)
Λ Λ
2im2 im2
δ2,0 = (A5 + B5 ) γ2,0 = B5 (3.8)
Λ2 Λ2
(3.9)

(2)
where Ai and Bi , for i = 3, 4, 5 are defined by writing the pole term for the amplitude Ii

as

(2) 1
Pole term of I3 = (mA3 + 6 pB3 ) (3.10)
ǫ
(2) 1
Pole term of I4 = −i(m2 A4 + m 6 pB4 + O(p2 )) (3.11)
ǫ
(2) 1
Pole term of I5 = −(m3 A5 + m2 6 pB5 + O(p2 )) (3.12)
ǫ

The appendix A presents a detailed analysis of the various contributions to these param-
eters. From (A17), (A20) and (A22), the final result is

8 11 m 7 m2 2
δ = − α2 − gα + g (3.13)
3 8π Λ 12π 2 Λ2
1 1 m 5 m2 2
γ = − α2 − gα + g (3.14)
12 8π Λ 48π 2 Λ2

Letting m → 0, we note that our determination of γ agrees with [7]. In the region m ≪ Λ,
however, our result presents corrections for nonvanishing fermion mass.
To fix β we look now at the four point Green function, which up to third order is (we
omit contributions which by power counting are finite)

g gα ǫ (4) g 2 ǫ (4) g 2 (4)

Γ(4) (p1 , p2 , p3 , p4 ) = µǫ (−i+ µ I1 + 2 µ I2 + α (1 − T )µ2ǫ I3
Λ Λ Λ Λ
g 2α (4) g 3
(4)
+ 2 (1 − T )µ2ǫ I4 + 3 (1 − T )µ2ǫ I5 ). (3.15)
Λ Λ

6
 (4) (4)
Here, again, the one loop amplitudes I1 and I2 are finite because of the use of the
dimensional regularization. Inserting (3.15), the expansion β = βi,j g i αj , and δ and
P
i,j

γ given in (3.3) and (3.4) into (3.1), we obtain β1,0 = 1, β0,1 = β1,1 = β2,0 = β0,2 = 0 and

β1,2 = −4iB3 − 2C3 (3.16)

2im 4m
β2,1 = C4 − B4 (3.17)
Λ Λ
2m2 4im2
β3,0 = 2 C5 + B5 . (3.18)
Λ Λ2
(4)
In the above expressions, Ci for i = 3, 4, 5 are related to the pole part of the amplitudes Ii

through,

(4)
Pole part of I3 = −iC3 /ǫ, (3.19)
(4)
Pole part of I4 = −(mC4 + O(p))/ǫ, (3.20)
(4)
Pole part of I5 = i(m2 C5 + O(p))/ǫ. (3.21)

In the appendix B we have collected the results of the calculations of the pole part of the
relevant graphs. Using B6 we obtain, finally,

20 2 21m 2 161m2 3
β=g+ gα − g α+ g . (3.22)
3 2πΛ 12π 2 Λ2

IV. DISCUSSION AND CONCLUSIONS

An inspection of Eq. (3.22) shows that the renormalization group beta function has g = 0
as a fixed point. As, for m ≪ Λ, β ≈ g(1 + 20α2 /3) the origin is an infrared stable fixed
point. Actually, this is the only existing fixed point. Here we are in disagreement with Ref.

[7] where a line of fixed points was found. The diverse conclusions are perhaps due to the use
of different regularizations but a more direct comparison of the methods seems unfeasible
as the calculations in [7] were not spelled out.
We will examine now the dimensions of some operators. As seen before the basic field ψ

has operator dimension dψ = 1 − α2 /12, and so at g = 0 the Green functions of the fermion

7
field have an improved ultraviolet behavior as α increases. Similar results are obtained if
one considers composite operators of canonical dimension less than three. The simplest of

them, the mass operator ψ̄ψ ha an anomalous dimension given by

γψ̄ψ = 2γ − 2Res, (4.1)

where Res is the residue coming from graphs contributing to the vertex function with the

insertion of the mass operator, ψψ and having two external fermionic lines. For practical
purpose this residue may be computed by taking the mass derivative of the contributions
calculated in the item 1 of appendix A. The result is iA3 = 5/4. Thus the dimension of ψ̄ψ
turns out to be equal to

8
dψ̄ψ = 2 − α2 . (4.2)
3

From the computation of the anomalous dimension of the operator ψ̄ψ one could easily
obtain the dimension of ψ̄ 6 ∂ψ. Indeed, the dimension of ψ̄ 6 ∂ψ is given by (4.1) but with
the replacement of Res by 3iA3 = 15/4. Thus

dψ̄6∂ ψ = 3 − 23/3α2 (4.3)

The determination of the anomalous dimension of the operator (ψ̄ψ)2 at g = 0 is more

complicated due to the fact that renormalization in general produces a mixing with other

operators of dimension lower or equal to four. However if we restrict the calculation to the
m = 0 case, as we will do, only operators of dimension four need to be considered. A further
simplification is obtained by considering only (formally) integrated operators. We have,
Z Z Z
3 2 3 2
d xN[(ψ̄ψ) ] = a1 d x(ψ̄ψ) + a2 d3 xψ̄∂ 2 ψ (4.4)
Z Z Z
d3 xN[ψ̄∂ 2 ψ] = b1 d3 x(ψ̄ψ)2 + b2 d3 xψ̄∂ 2 ψ (4.5)

where the symbol N indicates a normal product prescription corresponding to the subtrac-
C3 α2
tion of the pole terms. A direct calculation gives that a2 = b1 = 0, a1 = 1 + ǫ
and
α2
b2 = 1 − . A straightforward analysis shows now that the dimensions of d3 xN[(ψ̄ψ)2 ]
R
3ǫ

d3 xN[ψ̄∂ 2 ψ] are given by

R
and

8
 20 2
4+4γ02 α2 − 2C3 α2 = 4 + α , (4.6)
3
2 α2
4+2γ02 α2 + α2 = 4 + , (4.7)
3 2

respectively. One sees that, at least for the operators that we explicitly considered, the
operator dimension decreases with α accordingly the canonical dimension is lower or equal
to three, in accord with [7]. However, if the canonical dimension is bigger than three,
the operator dimension increases with α so that no improvement for nonrenormalizable
interactions results.

Our results are valid if the basic fermion field is flavorless. The N-flavor case is presently
under investigation.

APPENDIX A:

In this appendix we shall present a detailed analysis of the contributions to the pole part
of the two point vertex function. Due to the use of the dimensional regularization the poles
only appear at the two loop level, beginning at second order in the coupling constants. We

will examine separately each order of perturbation, i. e., the orders α2 , gα and g 2 , We have,
1. Order α2 . In this order there are three diagrams which are shown in Fig. 2. These
diagrams give the contributions,
Figure 2(a) :

d d k1 d d k2 λ α
Z
I3 (a) = 4π 2 i ǫρµλ ǫνθα T k k Tr [γ µ (6 k2 + m)γ ν (6 k2 − 6 k1 + m)]
(2π)d (2π)d 1 1
h i
γ ρ (6 p− 6 k1 + m)γ θ
×
(k22 − m2 ) [(k2 − k1 )2 − m2 ] [(p − k1 )2 − m2 ] (k12 )2
1
= (m A3 (a)+ 6 p B3 (a)) , (A1)
ǫ

where, on the right hand side of the first equality, we have introduced the operator T to

extract the pole part of the expression to which it is applied. From (A1) we obtain

A3 (a) 1 2iπ 2 d d k1 d d k2
Z
= Tr I3 (a)|p=0 = T
ǫ 2m m (2π)d (2π)d

9
 −8m3 k12 − 8m(k12 )2 − 8mk12 (k1 .k2 ) + 8m(k1 .k2 )2
× . (A2)
(k22 − m2 ) [(k2 − k1 )2 − m2 ] [k12 − m2 ] (k12 )2

For simplicity, the trace was taken in the integrand of (A1) and calculated directly at d = 3.
This, of course, does not affect the result for the pole part of the integrals. Analogously,

B3 (a) 1 2iπ 2 d d k1 d d k2
Z
= 2 Tr(I3 (a) 6 p) = 2 T
ǫ 2p p (2π)d (2π)d
Numerator
× 2 , (A3)
(k2 − m2 ) [(k2 − k1 )2 − m2 ] [(p − k12 )2 − m2 ] (k12 )2

where

Numerator = −16ǫρµλ ǫνθα k1ρ k2µ pλ k1ν k2θ pα − 16m2 k12 (k1 · p) − 8k12 (k1 · k2 )(k1 · p)

+ 8m2 (k1 · p)2 + 8(k1 · k2 )(k1 · p)2 − 8(k1 · p)2 k22 − 8(k1 · k2 )2 p2 + 8k12 k22 p2 . (A4)

Here and in what follows we shall adopt the following procedure for performing the integrals.

We first consider the k2 integral and use Feynman’s trick,

1 Γ[α1 + α2 ] 1 xα1 −1 (1 − x)α2 −1

Z
α1 α2 = dx (A5)
a1 a2 Γ[α1 ]Γ[α2 ] 0 [xa1 + (1 − x)a2 ]α1 +α2

to reduce the denominators containing the variable of integration to only one denominator.

After integrating in k2 we use again Feynman’s formula (A5) to combine the denomina-
tors that depend on k1 . We then integrate over k1 and, finally, perform the parametric
integrations. In the present case, after integrating on k2 , we get
d
A3 (a) π (2− 2 ) 1 d d k1 1
Z Z
= T d−3 dx
ǫ 2 0 (2π) (k1 − m2 ) k12
d 2
" #
d d
−1 d 2 d −2 2
× Γ(1 − ) ∆ 2 + Γ(2 − ) k1 ∆ 2 (1 + x − x ) (A6)
2 2

and
d
B3 (a) 1 π (2− 2 ) 1 d d k1 k1 · p
Z Z
= −T 2 d−1 dx
ǫ p 2 0 (2π) [(p − k1 )2 − m2 ] (k12 )2
d
"
d   d
× 4Γ(1 − ) (d − 2)(k1 · p) − k12 ∆ 2 −1 + 8x(1 − x)
2
#
d  2 2 2
 d
−2
× Γ(2 − ) (k1 · p)k1 − (k1 ) ∆ 2 , (A7)
2

10
where ∆ = m2 − x(1 − x)k12 . Continuing our calculation, we would introduce two new
parametric integrations as there are now three different denominators (we take 1/∆ as a new

denominator) depending on k1 in each of the terms of the above expressions. However, as

the result does not depend on m and we are looking only for the pole part of the amplitudes,
we can speed up the calculation by modifying the dependence on m of some denominators.
For example, without changing the final result, we can replace the first term on the right

hand side of (A6) by

d d
π (2− 2 ) 1 dd k1 Γ(1 − d2 ) ∆ 2 −1
Z Z
T d−3 dx (A8)
2 0 (2π)d (k12 − m2 )2

Similarly, in the computation of B3 (a) one can set m = 0 in the expression for ∆ so that
one has to use only one parametric integral. Following this recipe, after integrating in k1 we
obtain (a = 1 − y − yx(1 − x)),

A3 (a) π 2−d 1 1 Z Z
= T 2d−3 Γ[3 − d] dx dy (1 − y)
ǫ 2 "0 0
(1 − 2y)m2 d−3 y −d/2 1−d/2
#
d 2 y
×[ ] + (1 + x − x ) 4−d/2
a a3−d/2 2 a
5i
=− (A9)
8ǫ

and

B3 (a) π 2−d 1 1 (1 − y)2−d/2 (ym2 − y(1 − y)p2)d−3

Z Z
= −T 2d−3 Γ[3 − d] dx dy
ǫ 2 0 0 [x(x − 1)]1−d/2
" #
(1 − 5y)Γ[1 − d/2] Γ[2 − d/2]Γ[5 − d/2]
× −2 (1 − y)(1 − 7y)
Γ[3 − d/2] Γ[4 − d/2]2
i
=− . (A10)
24ǫ

Now, for the remaining graphs of order α2 , we have,

Figure 2(b) :
Z
d d k1 d d k2 λ ρ
I3 (b) = −4iπ 2 ǫµνλ ǫαβρ T k k
(2π)d (2π)d 1 2
h i
γ µ (6 p− 6 k1 + m)γ α (6 p− 6 k1 − 6 k2 + m)γ β (6 p− 6 k1 + m)γ ν
×
[(p − k1 )2 − m2 ]2 [(p − k1 − k2 )2 − m2 ] k12 k22
1
= (m A3 (b)+ 6 p B3 (b)) (A11)
ǫ

11
Following the same steps as in previous case, we obtain

A3 (b) 1 i
= Tr I3 (b)|p=0 = − (A12)
ǫ 2m 4ǫ
B3 (b) 1 i
= 2 T r(I3 (b) 6 p) = − (A13)
ǫ 2p 12ǫ

Figure 2(c) :

d d k1 d d k2 λ ρ
Z
I3 (c) = −4iπ 2 ǫµβλ ǫανρ T k k
(2π)d (2π)d 1 2
h i
γ µ (6 p− 6 k1 + m)γ α (6 p− 6 k1 − 6 k2 + m)γ β (6 p− 6 k2 + m)γ ν
×
[(p − k1 )2 − m2 ] [(p − k1 − k2 )2 − m2 ] [(p − k2 )2 − m2 ] k12 k22
1
= (m A3 (c)+ 6 p B3 (c)) . (A14)
ǫ

The computation of this expression is a bit more complicated because one has to introduce

three Feynman parametric integrals. The final result is, nevertheless, simple,

3i
A3 (c) = − (A15)
8
i
B3 (c) = (A16)
24

Collecting these results, we obtain

5i
A3 = A3 (a) + A3 (b) + A3 (c) = −
4
i
B3 = B3 (a) + B3 (b) + B3 (c) = − (A17)
12

2. Order gα graphs. There are six diagrams which have been drawn in Fig. 3. They give
Figures 3(a) and 3(b). Both diagrams have the structure of a product of two one loop

graphs. The corresponding dimensionally regularized amplitudes does not have a pole at
d = 3.
Figure 3(c). Actually, this diagram does not contribute because the corresponding ana-
lytic expression is finite. Indeed, we have

∂ dd k1 dd k2 Tr[γ µ (6 k2 + m)γ ν (6 k2 + 6 k1 + m)]ǫµνρ k1ρ

Z
I4 (c) = T
∂m(2π)d (2π)d (k22 − m2 )((k1 + k2 )2 − m2 ) k12
∂ d d k1 d d k2 m
Z
= −4iT 2
=0 (A18)
∂m (2π) (2π) (k2 − m )((k1 + k2 )2 − m2 )
d d 2

12
since the integral in the second equality has the structure of a product of two one loop
integrals.

Figure 3(d). The same reasoning can be applied to this situation since no external
momentum flows through the diagram. We conclude that there is not a pole term.
Figure 3(e). By Furry’s theorem this diagram cancels with its charge conjugated partner.
Figure 3(f ). We have the following contribution
Z
d d k1 d d k2 µ
I4 (f ) = 4πǫµνλ T d d
γ (6 p− 6 k1 + m)k1λ
(2π) (2π)
(6 k2 − 6 k1 + m)γ ν (6 k2 + m)
×
[(p − k1 )2 − m2 ] (k22 − m2 ) ((k2 − k1 )2 − m2 ) k12
(m2 A4 (f ) + m 6 pB4 (f ))
= −i (A19)
ǫ
from which one obtains
9
A4 = A4 (f ) =
16π
1
B4 = B4 (f ) = (A20)
8π
3. Order g 2 graphs. There are only the two diagrams shown in Fig f5 . We get,
d d k2 d d k1 6 p+ 6 k1 + m
Z
I5 = −4iT
(2π)d (2π)d [(p + k1 )2 − m2 ](k22 − m2 )[(k2 − k1 )2 − m2 ]
× {(6 k2 + m)(6 k2 − 6 k1 + m) − Tr[(6 k2 + m)(6 k2 − 6 k1 + m)]}
(m3 A5 + m2 6 pB5 )
=− , (A21)
ǫ
where the two terms on the second equality refers to the graphs 4(a) and 4(b), respectively.
After a lengthy calculation one determines,
3i
A5 = −
16π 2
5i
B5 = − (A22)
48π 2

APPENDIX B:

In this appendix we will discuss the calculation of the pole part of the four point vertex

function which is needed for fixing the renormalization group beta function. Actually, since

13
all we need is the constant part of the residue, the calculation of the relevant graphs will be
done at zero external momenta.

The first observation is that, up to the order we are interested, i. e., third order, there
are too many graphs. To be systematic, we will separate them accordingly they have or have
not closed fermionic loops. Moreover, if they do not posses fermionic loops we group them
accordingly the number of CS or auxiliary GN lines linking the two fermion lines crossing

the diagram. Many diagrams cancel because of Furry’s theorem; this is the case if there
is a fermionic loop with an odd number of attached CS lines. Other diagrams have the
structure of a product of two one loop graphs and therefore are finite. We shall not consider
these two types of graphs any longer. The anti-symmetrized amplitude for a graph γ has

the generic structure of a product, (A ⊗ B) C, where A and B refers to the propagators and
vertices associated to the two fermion lines and C to the others factors (⊗ indicates the
anti-symmetrized direct product). Using this notation, one can verify that

T
Z Z
d d
Pole Part of d k1 d k2 (A ⊗ B)C = dd k1 dd k2 (Tr[A] Tr[B] − Tr[AB])C, (B1)
2

For example, from the analytic expression for the graph shown in Fig. 5,
Z
k2λ
λ dd k1 dd k2 [γ µ SF (k2 )] ⊗ [γ ν SF (−k2 )]Tr[SF (k1 )SF (k1 − k2 )]ǫµνλ , (B2)
k22

where λ is a combinatorial factor, we determine A, B and C as

A = [γ µ SF (k2 )] (B3)

B = [γ ν SF (−k2 )] (B4)
k2λ
C = λǫµνλ (B5)
k22

The results for the pole parts are summarized in the Tables A and B, which corresponds
to the two cases mentioned above. In the table A are listed the results from graphs with one
closed fermionic loop; these have been arranged accordingly the number of CS vertices in
the loop. Table B exhibits the pole part of graphs without fermionic loop. They have been

collected into types (i,j), where i and j are the number auxiliary GN and CS lines,respectively,

14
linking the two fermion lines crossing the diagram. Notice that there are not contribution
from graphs of type (0,1) since they are not proper. Figure 6 furnishes examples of each

one of these sets of diagrams. The final result for each order is obtained by summing the
corresponding entries in each table. Thus we have,

7 11i 13
C3 = − C4 = C5 = (B6)
2 2π 2π 2

ACKNOWLEDGMENTS

This work was supported in part by Conselho Nacional de Desenvolvimento Cientı́fico e

Tecnológico (CNPq) e Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP).

15
 REFERENCES

∗
On leave of absence from Universidade Federal do Pará.

[1] S. Weinberg, “The Quantum Theory of Fields”, Cambridge University Press, 1995; G.
P. Lepage proceedings of TASI-89, 1989; G. P. Lepage et al., Phys. Rev. D 46, 4052

(1992); M. Luke, A. V. Manohar, Phys. Rev. D55, 4129 (1997).

[2] W. A. Bardeen, C. T. Hill and M. Lindner, Phys. Rev. D41, (1990) 2197; M. Atance
and L. Cortés, Phys. Rev. D 54, 4973 (1996).

[3] A. V. Manohar, Phys. Rev. D56 230 (1997).

[4] D. Gross in Methods in Field Theory, Les Houches 1975, R. Balian and J. Zinn-Justin
eds., North-Holland, Amsterdam, (1976); G. Parisi, Nucl. Phys. B100, 368 (1975); B.
Rosenstein, B. J. Warr and S.H. Park, Phys. Rev. Lett 62 1433, (1989).

[5] M. Gomes, R. S. Mendes, R. F. Ribeiro and A. J. da Silva, Phys. Rev. D43, (1991)
3516; ; M. Gomes, V. O. Rivelles and A. J. da Silva, Phys. Rev.D41, 1363 (1990); N.
V. Krasnikov and A. B. Kyatkin, Mod. Phys. Lett. A6, (1991) 1315.

[6] B. Rosenstein, B. J. Warr and S.H. Park, Phys. Rev. D 39 3088, 1989 L. Del Debbio,
S. J. Hands, J. C. Mehegan, Nucl. Phys. B502 (1997) 269; S. Hands, “Fixed Point
Four-Fermi Theories”, hep-lat/9706018.

[7] W. Chen and M. Li, Phys. Rev. Lett. 70, 884 (1993); W. Chen, Nucl. Phys. B435, 689
(1995).

[8] C. G. Bollini and J. J. Giambiagi, Phys. Lett. 40B, 566 (1972); G. ’t Hooft and M.
Veltman, Nucl. Phys. B44, 189 (1972); J. F. Ashmore, Lett. Nuovo Cimento 4, 289
(1972).

[9] R. S. Willey, Phys Rev. D48, 2877 (1993); T. Ghergheta, Phys. Rev. D50, 5985 (1994)

[10] We use natural units (c = h̄ = 1) and our metric is g00 = −g11 = −g22 = 1. The fully

16
 antisymmetric tensor ǫµνλ is normalized such that ǫ012 = 1 and we define ǫij ≡ ǫ0ij .
Repeated greek indices sum from 0 to 2, while repeated Latin indices from the middle

of the alphabet sum from 1 to 2. For the γ-matrices we adopt the representation γ 0 =
σ 3 , γ 1 = iσ 1 , γ 2 = iσ 2 , where σ i , i = 1, 2, 3, are the Pauli spin matrices.

[11] E. Speer,in “Renormalization Theory”, Proceedings of the International School Ettore

Majorana, Erice 1975, edited by G. Velo and A. S. Wightman, D Reidel Publishing Co,
1976.

[12] S. Weinberg, Phys. Rev. D8 3497 (1973); G. ’t Hooft, Nucl. Phys. B61 455 (1973).

[13] S. Coleman and B. Hill, Phys. Lett. 159B, 184 (1985).

17
 FIGURES
FIG. 1. Feynman rules for the interaction vertices. Continuous and wavy lines represent the

fermion and vector propagators, respectively.

FIG. 2. Order α2 graphs contributing to the two point function.

FIG. 3. Fermionic self-energy graphs of order gα.

FIG. 4. Order g2 fermionic self-energy graphs.

FIG. 5. Example of a two-loop diagram.

FIG. 6. Diagrams illustrating the various classes of graphs in the four point vertex function

18
 TABLES

TABLE I. Pole part for four legs graphs with a closed fermionic loop.

Order of perturbation Number of diagrams Pole part

gα2 12 −i/2ǫ

g2 α 7 −4i/πǫ

g3 8 4i/π 2 ǫ

TABLE II. Pole part for four legs graphs without closed fermionic loops. The first column

lists different types of diagrams (i,j), where i and j are the number of GN and CS lines joining the

two fermion lines crossing the graph; the digit in parenthesis after the pole parts is the number of

contributing graphs.

Diagram type Order gα2 Order g2 α Order g3

(0,2) i/2ǫ (8) — —

(1,0) −5i/2ǫ (12) 3i/πǫ (18) −3i/4π 2 ǫ (6)

(1,1) 8i/ǫ (24) 4i/πǫ (8) —

(1,2) −2i/ǫ (12) — —

(2,0) — −i/πǫ (14) i/π 2 ǫ (12)

(2,1) — −15i/2πǫ (9) —

(3,0) — — 5i/4π 2 ǫ (4)

19