ON THE CONSTRUCTION OF

RENORMALIZED GAUGE THEORIES

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

Abstract
The aim of these lectures is to describe a construction, as self-
contained as possible, of renormalized gauge theories. Following a
suggestion of Polchinski, we base our analysis on the Wilson renor-
malization group method.
After a discussion of the infinite cut-off limit, we study the short
distance properties of the Green functions verifying the validity of
Wilson short distance expansion. We also consider the problem of
the extension to the quantum level of the classical symmetries of the
theory. With this purpose we analyze in details the breakings induced
by the cut-off in a SU (2) gauge symmetry and we prove the possibility
of compensating these breakings by a suitable choice of non-gauge
invariant counter terms.

1
Lectures given at the Parma Theoretical Physics Seminar, September 1-12 1991 -
Revised Version

0
1. Introduction

In a period of approximately ten years, about twenty years ago, the per-
turbative construction of renormalized quantum field theory has achieved a
remarkable level of rigor and efficiency. based on many technical achieve-
ments that are widely explained in the current literature. We mention
among others the extension to all orders of perturbation theory of subtraction
schemes suitable to avoid infinities in the Feynman amplitudes. This has put
on a rigorous basis the method of counter-terms, although in the framework
of a formal perturbation theory which is not absolutely summable.
A second significant technical progress consists in the discovery of very
clever regularization schemes, in particular the dimensional one which has
made possible calculations of renormalization effects of remarkably high or-
der, as e. g. the fourth of QCD. This has greatly improved the efficiency of
renormalized quantum field theory .
The great majority of the above mentioned achievements have been based
on a deep and complicated analysis of Feynman diagrams a typical ingredient
of which is the concept of ”forest”. The difficulties with Feynman diagrams
are amplified in the gauge models of fundamental interactions in which the
number of contributions to a given amplitude increases rapidly with the per-
turbative order and hence it is often prohibitive to push the calculations
beyond one loop. In these models dimensional regularization too can be-
come a source of problems due to the difficulty of extending the concept of
chirality to complex space-time dimensions.
Waiting for new ideas and tools to compute higher orders in gauge theo-
ries, there remains, in our opinion, the need of an essential simplification of
the proofs of the relevant general properties of renormalized quantum field
theory.
Few years ago Polchinski [1] has shown how the existence of the ultra-
violet limit of a scalar theory, regularized by means of a momentum cut-
off, can be proved using Wilson renormalization group techniques [2] . The

1
method of Polchinski is remarkably simple and can be trivially extended
to spinor and vector fields; however one still needs to recover in the same
framework the whole set of general results that have made possible the above
mentioned progresses in renormalized quantum field theory.
In these lectures we present an attempt, following the ideas of Polchinski,
to give a self-contained proof of the existence of a perturbatively renormal-
ized quantum field theory and of two ”general properties” of it. First the
validity of Wilson short distance expansion [3] which is a fundamental tool
in the analysis of Green functions. Secondly we discuss the ”quantum action
principle”, from which rather general results on the renormalized structure
of theories with continuous symmetries can be obtained [4],[5] .
The lecture notes are so organized: in section 2 we discuss the construc-
tion of the Feynman functional generator. In section 3 we present the renor-
malization group method, whose perturbative solution is discussed in section
4. In section 5 we make some comments on the construction of composite
operators and we discuss Wilson short distance expansions. In section 6 we
derive the ”quantum action principle” that we apply to the study of the
SU (2) Yang-Mills model.

2
2. The Feynman formula

The existence of a theory of scattering is one of the fundamental results of

field theory. It is based on the classical axiomatic results on the asymptotic
evolution of states and on the well known Lehmann-Symanzik-Zimmermann
reduction formulae relating S matrix elements to time-ordered Green func-
tions. If φin is a set of asymptotic ingoing fields whose wave operator is W
we can write the scattering operator Sop in the Fock space of φin as:
!
Z
δ
Sop =: exp d4 xφin (x)Wx z −1 : Z[j]|j=0 , (2,1)
δj(x)
where z is the residue of the Fourier transformed two-point function on the
mass shell pole and the functional Z[j] is the generator of the time-ordered
Green functions: R
i d4 xj(x)φ(x)
Z[j] = (Ω, T e Ω) , (2,2)
and Ω is the vacuum state.
The determination of the functional generator Z is therefore the main dy-
namical problem in the construction of a field theory. The Feynman formula
is universally considered as the solution of this problem.
This formula is usually written i terms of the classical action Scl according:
Z Y R
Z[j] = N dφ(x)ei[Scl − d4 xj(x)φ(x)]
, (2,3)
x

where it is assumed the functional integral to make sense and the measure
Q
x dφ(x) to be translation invariant.
In the first part of this course we shall discuss the possibility giving a
meaning to this formula. For this we have to overcome a sequence of diffi-
culties that can be traced back to nature of the functional measure and to
that of the integrand.
The problem with the measure is technically related to the lack of lo-
cal compactness, that is, to the presence of an infinite number of variables.
Concerning the integrand, if the measure is translation invariant as assumed,
it has to introduce a uniform convergence factor for large field amplitudes.
This is not verified in the present case since the integrand has absolute value
equal to one.

3
 The standard solution to this convergence problem is based on the trans-
formation of the minkowskian theory into an euclidean one. This is achieved
”Wick rotating” the time variables from the positive real axis to the nega-
tive imaginary one. The time-ordered Green functions are then replaced by
the euclidean Schwinger functions [6] whose functional generator is defined in
complete analogy with (2,3), absorbing the imaginary unit into the euclidean
space measure. That is:
Z Y R
Z[j] = N dφ(x)e−[Se − d4 xj(x)φ(x)]
. (2,4)
x

If the euclidean classical action Se is a positive functional increasing with

the field amplitude the wanted convergence factor is guaranteed. Of course,
we have overcome the first difficulty at the price of computing something
that is different from our goal. However, Osterwalder and Schrader have
identified the conditions ensuring the possibility of recovering the wanted
physical information from an euclidean theory. [7]
Coming back to the number of integration variables, we notice that one
can distinguish two sources of this difficulty. First, euclidean invariance re-
quires the space volume to be infinite. This is the infra-red (IR) difficulty.
Giving up the euclidean invariance one could quantize the theory in a hy-
percube choosing as integration variables the Fourier amplitudes of the field.
However these are still infinite in number and locality requires the interac-
tion to involve all the field Fourier components. This is the ultra-violet (UV)
difficulty, it can also be seen from a different point of view. If the interacting
Green (Schwinger) functions are distributions, as in the free case, the same
is true for the functional derivatives of Z and hence the interaction cannot
be written as a strictly local functional of the fields.
To avoid these IR and UV difficulties, giving up for the moment local-
ity and covariance, we introduce a system of regularizations. First, as said
above, we restrict our theory into a four-dimensional hypercube Ω. Periodic
boundary conditions are chosen for the fields in order to preserve translation
invariance. As a matter of fact in this way we are considering a quantum
mechanical system in a three-dimensional cube of side L in thermal equilib-
rium at the temperature β −1 = L1 . An infinite volume and zero temperature
limit will eventually reproduce the original relativistic euclidean field theory.
The second regularization concerns directly the UV difficulties. We mod-
ify the interaction decoupling the short wave length modes which , however,

4
continue to appear into the dynamical framework as free degrees of freedom.
Notice that in standard approaches these degrees of freedom are simply not
taken into account.
We regularize the euclidean theory by replacing into the interaction the
field φ(x) with
1 X p2
KΛ0 φ(x) = 2 k( 2 )φp~ eipx (2,5)
L p~ Λ0
where
1 Z 4
φp~ = 2 d xφ(x)eipx , (2,6)
L Ω
and, of course the integral is limited within the above mentioned four-dimensional
hypercube of side L. The ultraviolet (UV) cut-off factor k is a C ∞ function
assuming the value 1 below 1 and vanishing above 2.
It should be clear that the introduction of the UV cut-off should interfere
as weakly as possible with the observables of our theory. This requires, in
particular, that the chosen value of the cut-off be much higher than the
greatest wave number appearing in the Fourier decomposition of the sources
j and of the other sources that one could introduce to define composite
operators. More precisely let ΛR be the greatest observable wave number,
we select the sources so that

KΛR j = j . (2,7)

and we ask
Λ0  ΛR . (2,8)
Eventually Λ0 will be sent to infinity.
Now we come to the choice of the action. This is done having in mind
the short-distance properties of the free theory that we want only weakly
perturbed by the interaction. With this purpose we assign to every field φ a
mass-dimension dφ and to the corresponding source j, (dj = D − dφ ), where
D is the dimension of the euclidean space (in our case 4). dφ and hence dj
are computed from the mass-dimension of the wave operator W requiring
that the free field equations be dimensionally homogeneous. Thus, in the
scalar field case, the mass-dimension of the wave operator is two, that of the
laplacian, and hence the mass-dimension of the field is computed from:

2 + dφ = dj = D − dφ . (2,9)

5
That is one in the 4-dimensional case.
The action in (2,4) is chosen as the sum of the free part
Z
(∂φ)2 + m2 φ2
d4 x , (2,10)
Ω 2
and of the interaction L0 , an integrated local polynomial in the regularized
field KΛ φ and its derivatives. The short-distance ”power counting” condition
limits to 4 the dimension of the operators appearing in L0 . Therefore in the
scalar case we have:
m20 − m2 2
" #
Z
λ0 4 z0 − 1
4 2
L0 = d x φ + (∂φ) + φ . (2,11)
Ω 4! 2 2
A particular care should be devoted to the definition of composite operators.
To introduce for example the operator φ2 we assign him the source ω, whose
dimension is the complement to 4 of the dimension of the operator, that is
2, and we add to the interaction L0 the ω-dependent terms:
Z h i
d4 x ζ0 ωφ2 + η0 ω 2 (x) + ξ0 ω(x) . (2,12)
Ω

This functional identifies the general solution of an extended power counting

condition taking into account also the dimension of the sources.
One should wonder about the meaning of the last two terms that are field
independent. It is clear from the definition of the functional generator Z that
e. g. the ω 2 term induces a contribution proportional to a Dirac δ in the
correlation function of two composite operators. Back to the minkowskian
world this corresponds to a redefinition of the time-ordering of the product
of two operators.
Now, taking into account both UV and IR regularizations, we write the
interaction of our model according:
m20 − m2
"
Z
4λ0 4 z0 − 1 2
L0 = d x (KΛ0 φ) + (∂KΛ0 φ) + (KΛ0 φ)2 +
Ω 4! 2 2
i λ0
ζ0 ω(KΛ0 φ)2 + η0 ω 2 + ξ0 ω =
X
δ~ KΛ0 φp~1 ...KΛ0 φp~4 +
4!L4 p~1 ,..~p4 0,~p1 +..+~p4
X (zo − 1)p2 + m20 − m2
KΛ0 φp~ KΛ0 φ−~p +
p
~
2

6
 ζ0 X
δ~ ωp~ KΛ φp~ KΛ φ−~p3 +
L2 p~1 ,..~p3 0,~p1 +..+~p3 1 0 2 0
ωp~ ω−~p + L2 ξ0 ω~0 .
X
η0 (2,13)
p
~

Notice that by the IR regularization we have automatically broken the eu-

clidean invariance of the theory, since a cube is not rotation invariant. There-
fore there is no reason to preserve the euclidean invariance of every single
term of the interaction. Following the fine-tuning strategy which will be dis-
cussed in section 6 it is possible to prove the compensability of the possible
breakdown of euclidean invariance induced by the IR regularization by the
introduction into the interaction of non-invariant counter-terms. This com-
pensability holds true for all continuous symmetries if the symmetry group
is semisimple. This is a typical consequence of the quantum action principle
[4],[5].
It is apparent that in our example L0 depends on the 6 parameters λ0 ,
z0 , m0 , ζ0 , η0 and ξ0 that can be extracted from the interaction by suitable
normalization operators. That is:
∂4
ρ0,1 = L4 L0 |φ=ω=0 ≡ N1 L0 = λ0 , (2,14)
∂φ~40
and, setting
∂ ∂
L0 |φ=ω=0 = Πp~ , (2,15)
∂φp~ ∂φ−~p
Πp~ − Π~0
ρ0,2 = ≡ N2 L0 = z0 − 1 , (2,16)
p2
for some suitably chosen p~ and

ρ0,3 = Π~0 ≡ N3 L0 = m20 − m2 , (2,17)

L2 ∂ 3
ρ0,4 = L0 |φ=ω=0 ≡ N4 L0 = ζ0 , (2,18)
2 ∂φ~20 ∂ω~0
1 ∂2
ρ0,5 = L0 |φ=ω=0 ≡ N5 L0 = η0 , (2,19)
2 ∂ω~02
∂
ρ0,6 = L−2 L0 |φ=ω=0 ≡ N6 L0 = ξ0 . (2,20)
∂ω~0

7
The functional generator corresponding to the interaction (2,13) is given by
hP i
Z Y C(p) P Z Y
− φ−~p φp~ +L0 − j p φp~
~ −~
Z[j, ω] = N dφp~ e p
~ 2 p
≡N dφp~ e−S ,
p
~ p
~
(2,21)
where the normalization factor N ensures the normalization condition

Z[0, 0] = 1 . (2,22)

and:
C(p) = p2 + m2 . (2,23)
Notice that, taking into account the regularizations, the integral in (2,21)
factorizes in an infinite dimensional, purely gaussian,√ contribution corre-
sponding to the Fourier components φp~ with p~ > 2Λ0 , that is reabsorbed
into the normalization factor N , and in a finite dimensional part, that is
absolutely convergent provided that λ0 is positive.
The crucial problem in quantum field theory is to prove the existence of
an IR (L → ∞) and an UV (Λ0 → ∞) limit of (2,21). This goal has been
achieved in the framework of the perturbative method for the whole family
of theories satisfying the power counting criterion.
The major aim of these lectures is to describe the main lines of this result.
For this is convenient to remember that perturbation theory is constructed
introducing into the Feynman functional integral the ordering parameter h̄
according Z Y
Zc [j,ω] −S
Z[j, ω] ≡ e h̄ =N dφp~ e h̄ , (2,24)
p
~

and developing the connected functional Zc as a formal power series in h̄.

This power series is obtained applying to (2,24) the method of the steepest
descent. Analyzing the terms of the series as sums of Feynman diagrams one
sees that Zc receives contributions only from the connected diagrams and
that h̄ can be interpreted as a loop counting parameter.

8
3. The Wilson renormalization group method.

A fundamental tool in the analysis of the UV limit is the Wilson renor-

malization group method. [2] This method, in its original version, is based on
a sharp UV cut-off (Λ0 ), or equivalently on a lattice regularization, limiting
the number of degrees of freedom, i. e. that of the integration variables in
the Feynman formula. It consists in the iterative reduction of the number of
degrees of freedom through the integration over the field Fourier components
φp~ with Λ0 ≥ p ≥ Λ20 . The resulting, partially integrated, Feynman formula
is brought back to the original form substituting the interaction lagrangian
with an effective one which now depends on the fields cut-off at Λ20 . The exis-
tence of the wanted UV limit is related to that of a ”fixed point” in the space
of effective lagrangians which is approached afteranalyzing a great number
of iterations of the partial integration procedure.
Following Polchinski, [1] we shall use a modified version of this method
based on a continuous lowering of the cut-off, which, in our case, corresponds
to a continuous switching-off of the interaction of the field components with
higher wave number.
This is accompanied by a continuous evolution of the effective lagrangian
Lef f replacing the bare interaction L0 in the Feynman formula as it is exhib-
ited in :
hP i
Z Y C(p) P
− φ−~p 2
φp~ +Lef f (KΛ φ,ω,Λ,Λ0 ,ρ0 )− j p φp~
~ −~
Z[j, ω] = N dφp~ e p
~ p
≡
p
~
Z Y
N dφp~ e−S , (3,1)
p
~

which holds true for ΛR < Λ < Λ0 . The identity of (3,1) and (2,21) is
guaranteed if Lef f satisfies the following evolution equation:

1X p
 
Λ∂Λ Lef f = Λ∂Λ k 2 C −1 (p) [∂−~p Lef f ∂p~ Lef f − ∂p~ ∂−~p Lef f ] , (3,2)
2 p~ Λ

with the initial condition

Lef f (KΛ0 φ, ω, Λ0 , Λ0 , ρ0 ) = L0 . (3,3)

9
In (3,2) we have introduced the simplified notation:
∂Lef f
∂p~ Lef f = . (3,4)
∂KΛ φp~

Using (3,2) it is easy to verify (3,1). Indeed

 
Z Y
dφp~  Λ∂Λ KΛ φp~ ∂p~ Lef f + Λ∂Λ Lef f  e−S ,
X
Λ∂Λ Z[j, ω] = − (3,5)
p
~ p
~

and
p
   
S −1 −S −1
e C (p)∂φ−~p e = −φp~ − C (p) k ∂−~p Lef f − jp~ . (3,6)
Λ
Thus we have:
Z Y
Λ∂Λ Z[j, ω] = − dφp~ [Λ∂Λ Lef f −
p
~

p p
     
∂p~ Lef f C −1 (p) ∂φ−~p + k ∂−~p Lef f  e−S , (3,7)
X
Λ∂Λ k
p
~
Λ Λ

where we have taken into account that from (2,7) and (2,8) one has
p
 
Λ∂Λ k jp~ = 0 . (3,8)
Λ
Now, integrating by parts the second term in the right-hand side of (3,7) we
get:
Z Y
Λ∂Λ Z[j, ω] = − dφp~ [Λ∂Λ Lef f −
p
~

p p
   
C −1 (p)k [∂−~p Lef f ∂p~ Lef f − ∂p~ ∂−~p Lef f ] e−S .(3,9)
X
Λ∂Λ k
p
~
Λ Λ

This vanishes owing to (3,2).

The evolution equation (3,2) defines a family of lines, identified by the
parameters ρ0,a of the bare lagrangian for a certain choice of the cut-off Λ0 , in
the space of the field functionals Lef f . The running parameter along the lines

10
is the cut-off Λ. However these lines do not describe the evolution toward
the limit we are interested in. In order to prove the existence of an UV
limit of our Green functional Z, we should rather study the evolution of our
theory when Λ0 increases and some ”low energy” properties of the effective
interaction are kept fixed. We can define these low energy properties by
means of the six numbers:

ρa (Λ, Λ0 , ρ0 ) = Na Lef f , (3,10)

where Na (a = 1, .., 6) are the normalization operators defined in (2,14)-

(2,20). These effective parameters are in one-to-one correspondence with the
parameters of the bare interaction and we assume that the functional relation
between ρ0,a and ρa be invertible. This will certainly be true at least for a
limited range of Λ. We denote by

ρ0,a (Λ, Λ0 , ρ) (3,11)

the inverse function of (3,10). Let us now consider the new field functional:

V (Λ) ≡ Λ0 ∂Λ0 Lef f (KΛ φ, ω, Λ, Λ0 , ρ0 (Λ, Λ0 , ρ)) =

Λ0 ∂Λ0 Lef f − Λ0 ∂Λ0 ρa ∂ρa Lef f =
Λ0 ∂Λ0 Lef f (KΛ φ, ω, Λ, Λ0 , ρ0 ) −
 !−1 a
∂ρ
Λ0 ∂Λ0 ρa   ∂ρ
0,b
Lef f (KΛ φ, ω, Λ, Λ0 , ρ0 ) . (3,12)
∂ρ0
b

This functional gives an indication of the dependence of Lef f on the UV

cut-off when the parameters ρ are kept fixed. We would like to show that V
vanishes stronger than a positive power of ΛΛ0 . This would clearly indicate
that, once the low energy parameters are fixed, the whole theory tends to-
ward a fixed point depending only on ρ. This is what is usually meant by
renormalizability of the theory.
Let us then study the evolution equation of V . Given any field functional
X, we define:
p p
   
C −1 (p)k
X
M [X] ≡ Λ∂Λ k [2∂−~p X∂p~ Lef f − ∂p~ ∂−~p X] , (3,13)
p
~
Λ Λ

where the action of ∂p~ on X is defined in the same way as on Lef f .

11
 From (3,2) we have:

Λ∂Λ Λ0 ∂Λ0 Lef f = M [Λ0 ∂Λ0 Lef f ] . (3,14)

This gives the evolution equation of the first term of V in (3,12). Considering
the second term, on gets three contributions which can be computed taking
into account that, by definition:

Λ∂Λ ρa = Na [Λ∂Λ Lef f ] . (3,15)

The first contribution is

 !−1 a
∂ρ
Λ∂Λ Λ0 ∂Λ0 ρa   ∂ρ
0,b
Lef f =
∂ρ0
b
 !−1 a
∂ρ
Na [M [Λ0 ∂Λ0 Lef f ]]   ∂ρ
0,b
Lef f =
∂ρ0
b
Na [M [Λ0 ∂Λ0 Lef f ]] ∂ρa Lef f . (3,16)

The second is:

 !−1 a
∂ρ
Λ0 ∂Λ0 ρa Λ∂Λ   ∂ρ
0,b
Lef f =
∂ρ0
b
 !−1 a  !−1 d
∂ρ ∂ρ d ∂ρ
−Λ0 ∂Λ0 ρa   Λ∂Λ   ∂ρ
0,b
Lef f =
∂ρ0 ∂ρ0,c ∂ρ0
c b
 !−1 a " " ##  !−1 d
∂ρ ∂L ef f ∂ρ
−Λ0 ∂Λ0 ρa   Nd M   ∂ρ
0,b
Lef f (3,17)
.
∂ρ0 ∂ρ0,c ∂ρ0
c b

The third is:

 !−1 a
∂ρ
Λ0 ∂Λ0 ρa   Λ∂Λ ∂ρ
0,b
Lef f =
∂ρ0
b
 !−1 a " #
∂ρ ∂L ef f
Λ0 ∂Λ0 ρa   M . (3,18)
∂ρ0 ∂ρ0,b
b

12
Combining (3,14) and (3,18) and recalling that M is a linear operator we
get M [V ]. In much the same way (3,14) and (3,15) give Na [M [V ]] ∂ρa Lef f .
Altogether we get:

Λ∂Λ V = M [V ] − Na [M [V ]] ∂ρa Lef f . (3,19)

In order to compute the initial condition to this evolution equation we notice

that, when Λ tends to Λ0 , Lef f tends to L0 and that

ρ(Λ0 , Λ0 , ρ0 ) = ρ0 . (3,20)

Thus:

Lef f (KΛ0 φ, ω, Λ, Λ0 , ρ0 ) = L0 (KΛ0 φ, ω, ρ0 ) +

 
(Λ − Λ0 )∂Λ Lef f (KΛ0 φ, ω, Λ, Λ0 , ρ0 ) |Λ=Λ0 + O (Λ − Λ0 )2 . (3,21)

Then, taking the derivative of both members of (3,21) with respect to Λ0

and setting Λ = Λ0 we have:

V [Λ0 ] = Λ0 ∂Λ0 Lef f (KΛ φ, ω, Λ, Λ0 , ρ0 ) |Λ=Λ0 −

Λ0 ∂Λ0 ρa ∂ρa Lef f (KΛ φ, ω, Λ, Λ0 , ρ0 (Λ, Λ0 , ρ) |Λ=Λ0 =
−Λ∂Λ Lef f (KΛ0 φ, ω, Λ, Λ0 , ρ0 ) |Λ=Λ0 +
Λ∂Λ Na [Lef f (KΛ0 φ, ω, Λ, Λ0 , ρ0 )] |Λ=Λ0 ∂ρa L0 =
 !
1X p 2

− Λ0 ∂Λ0 k 2
C −1 (p)∂−~p L0 ∂p~ L0 +
2 p~ Λ0
 
2 !
1 X p

Na Λ0 ∂Λ0 k 2 C −1 (p)∂−~p L0 ∂p~ L0  ∂ρ0,a L0 + c , (3,22)
2 p
~
Λ0

where c is field independent.

In deducing (3,22) we have taken into account that, for any integrated
local functional X of dimension up to 4, one has:

X = Na [X] ∂ρ0,a L0 + c , (3,23)

where c is again field independent. We still need the evolution equation of

 !−1 a
∂ρ  ∂ρ Lef f ≡ ∂ρ Lef f .

0,b a (3,24)
∂ρ0
b

13
We have
 !−1 a
∂ρ
Λ∂Λ ∂ρa Lef f =  Λ∂Λ ∂ρ
0,b
Lef f −
∂ρ0
b
 !−1 a  !−1 d
∂ρ  Λ∂Λ
∂ρ d ∂ρ  ∂ρ
 
0,b
Lef f =
∂ρ0 ∂ρ0,c ∂ρ0
c b
M [∂ρa Lef f ] − Nb [M [∂ρa Lef f ]] ∂ρb Lef f , (3,25)

and the initial condition is:

∂ρa Lef f |Λ=Λ0 = ∂ρ0,a L0 . (3,26)

It remains to discuss the solutions of the system of differential equations

(3,2), (3,25) and (3,19) whose corresponding initial conditions are L0 , (3,26)
and (3,22).
A first step in this discussion consists in the infinite volume limit. This
is harmless at the level of the effective lagrangian since the presence of a
derivative of the cut-off function limits the range of momenta in (3,2) and
(3,13) between Λ and 2Λ. This and the euclidean character of our theory
exclude any IR singularity.
We define the reduced functional L̂ef f according:
!
4 KΛ φ ω
Lef f (KΛ φ, ω, Λ, Λ0 , ρ0 ) = (LΛ) L̂ef f , , Λ, Λ0 , ρ0 . (3,27)
ΛL2 Λ2 L2

Writing Lef f and L̂ef f as (formal) power series in KΛ φ and ω whose coeffi-
cients, the effective vertices, are Γ(n,m) and Γ̂(n,m) . That is, for Lef f :
∞
X 1 X
Lef f = δ~0,P p~i +P q~j
n,m=0 n!m! p
~1 ,..,~
pn ,~
q1 ,..,~
qm

Γ(n,m) (~p1 , .., p~n , ~q1 , .., ~qm )KΛ φp~1 ..KΛ φp~n ωq~1 ..ωq~m , (3,28)

and a completely analogous equation for L̂ef f . Translating the evolution

equation of Lef f , (3,2), in terms of the effective vertices Γ̂ of L̂ef f , we get:

(Λ∂Λ + 4 − n − 2m) Γ̂(n,m) (~p1 , .., p~n , ~q1 , .., ~qm ) =

14
 X1 2 n X
m
p

2 2
)C −1 (p)
X X X
Λ Λ∂Λ k ( δp~,P p~i +P q~j
p
~
2 Λ k=1 l=1 [i1 ,..,ik ][ik+1 ,..,in ] [j1 ,..,jl ][jl+1 ,..,jm ]
 
(k+1,l)
Γ̂ (~p, p~i1 , .., p~ik , ~qj1 , .., ~qjl ) Γ̂(n−k+1,m−l) −~p, p~il+1 , .., p~in , ~qjl+1 , .., ~qjm −
X1 2
p 1

Λ∂Λ k 2 ( )C −1 (p) 2 4
Γ̂(n+2,m) (~p, −~p, p~1 , .., p~n , ~q1 , .., ~qm ) . (3,29)
p
~
2 Λ ΛL

1
For Λ >> L
the sums in the right-hand side of (3,29) can be safely
 4
L
replaced with integrals whose measure is: d4 p 2π
. In the same limit the
 4
2π
Kronecker δp~,P p~i +P q~j has to be replaced by the Dirac measure: L
δ(~p −
p~i − ~qj ). This yields:
P P

(Λ∂Λ + 4 − n − 2m) Γ̂(n,m) (~p1 , .., p~n , ~q1 , .., ~qm ) =

1 Z d4 p Λ2
 2 !
~ 2 p
p
~ · ∂p k Γ̂(n+2,m) (~p, −~p, p~1 , .., p~n , ~q1 , .., ~qm ) −
2 (2πΛ)4 Λ p2 + m2
n X m
Λ2 X
 2 !
1Z 4 p
d p p~ · ∂~p k 2
X
2 Λ p2 + m2 k=1 l=1 [i1 ,..,ik ][ik+1 ,..,in ]
~qj )Γ̂(k+1,l) (~p, p~i1 , .., p~ik , ~qj1 , .., ~qjl )
X X X
δ(~p − p~i −
[j1 ,..,jl ][jl+1 ,..,jm ]
 
Γ̂(n−k+1,m−l) −~p, p~il+1 , .., p~in , ~qjl+1 , .., ~qjm . (3,30)

Now, rescaling the integration variable, by setting:

!
(n,m) (n,m) p~1 p~n ~q1 ~qm
Γ̂ (~p1 , .., p~n , ~q1 , .., ~qm ) = γ , .., , , .., , (3,31)
Λ Λ Λ Λ

we can make the right-hand side of (3,30) Λ-independent at high momenta

(p2 >> m2 ).
With the purpose of analysing perturbatively our approach, we reintro-
duce into the evolution equation for γ (n,m) the loop counting parameter h̄
appearing in (2,24). This is done multiplying by a factor h̄1 both the covari-
ance C and the interaction Lef f : we obtain:

1Z 4 1
(Λ∂Λ + 4 − n − 2m − D) γ (n,m)
=− dp 2 m2
p~ · ∂~p k 2 (p2 )
2 p + Λ2

15
 n X
X m X X X X
δ(~p − p~i − ~qj )
k=1 l=1 [i1 ,..,ik ][ik+1 ,..,in ] [j1 ,..,jl ][jl+1 ,..,jm ]
 
γ (k+1,l) (~p, p~i1 , .., p~ik , ~qj1 , .., ~qjl ) γ (n−k+1,m−l) −~p, p~il+1 , .., p~in , ~qjl+1 , .., ~qjm +
h̄ Z d4 p 1
4
p~ · ∂~p k 2 (p2 ) 2 m2
γ (n+2,m) (~p, −~p, p~1 , .., p~n , ~q1 , .., ~qm ) , (3,32)
2 (2π) p + Λ2

where D is the dilation operator:

p~i · ∂~pi + ~qj · ∂~qj .

X X
D= (3,33)
i j

To proceed further, it is convenient to introduce the functional generator

lef f of the vertices γ (n,m) , that is:
∞n 
1 Z Y 
d4 pi φ(~pi )
X
lef f [φ, ω] =
n=0,m=0 n! i=1
m  
d4 qj ω(~qj ) (2π)4 δ(
Y X X
p~i + ~qj )
j=1

γ (n,m) (~pi1 , .., p~in , ~qj1 , .., ~qjm ) . (3,34)

Notice that the whole set of transformations which has led us from Lef f to
lef f consists in the introduction of dimensionless variables combined with the
infinite volume limit. Indeed setting:

φ(x) ≡ Λφ̂(Λx) , ω(x) ≡ Λ2 ω̂(Λx) , (3,35)

we have set:
 
Lef f (KΛ φ, ω, Λ, Λ0 , ρ0 ) = lef f KΛ φ̂, ω̂, Λ, Λ0 , ρ0 . (3,36)

Introducing;
1
∆0 (p) = p~ · ∂~p k 2 (p2 ) m2
, (3,37)
p2 + Λ2
and the functional differential operators:
Z
δ
N = d4 p φ(~p) , (3,38)
δφ(~p)

16
 Z
δ
M= d4 p ω(~p) , (3,39)
δω(~p)
" #
Z
δ δ
D= d p φ(~p)~p · ∂~p
4
+ ω(~p)~p · ∂~p , (3,40)
δφ(~p) δω(~p)
we can write (3,32) in the form:
h̄ Z 4 ∆0 (p) δ
" #
∂ −lef f δ −lef f
Λ − N − 2M − D e h̄ = dp 4
e h̄ . (3,41)
∂Λ 2 (2π) δφ(~p) δφ(−~p)
Representing graphically by blobs (vertices) with n legs the n-th functional
derivative of lef f and with a solid line ∆0 (p), we see that the first term in
the right-hand side of (3,32) can be represented by the fusion of two vertices
into a single one joining two lines, while, in the second term, two lines of the
same vertex joining together form a loop.

{ 
{ = −1 d4 p {]
h i
∂ R
Λ ∂Λ − N − 2M − D 2 (2π)4
[ −h̄
{ 
.
(3,42)
The infinite volume limit and momentum scale transformations performed
above can be also applied to ∂ρa Lef f and to V (Λ) giving ∂ρa lef f and:
v(Λ) = Λ0 ∂Λ0 lef f |ρ, Λ . (3,43)
The evolution equations of these functionals are easily deduced from (3,19)
∂
and (3,25), by replacing Λ∂Λ with Λ ∂Λ − N − 2M − D and the operator M
with m defined by:
∆0 (p)
" #
Z
4 δ δ h̄ δ δ
m[X] ≡ dp 4
X lef f − X . (3,44)
(2π) δφ(~p) δφ(−~p) 2 δφ(~p) δφ(−~p)
The normalization functionals Na are also changed, owing to the fact that,
after the infinite volume limit, the momentum p~ is no longer an index but a
continuous variable. One has also to take into account that both the fields
and the effective lagrangian have been rescaled. Therefore, after this limit,
one has in particular:
8
Z
1 δ4
N1 = 2 d4 p , (3,45)
(2π)4 δφ4 (~p)

17
 Z
1 δ δ
N2 = 2∂~q · ∂~q d4 p 4
|q~=0 , (3,46)
(2π) δφ(~p + ~q) δφ(~p − ~q)
while one has:
4 2
Z
1 δ2
N3 = 2 Λ d4 p ≡ Λ2 n3 , (3,47)
(2π)4 δφ2 (~p)
and in general a factor Λda appears in the normalization operator of the
parameter ρa with mass dimension da .
Therefore the normalization conditions become
ρ1
γ (4,0) (0) = ρ1 , ∂p2 γ (2,0) (0) = ρ2 , γ (2,0) (0) = , (3,48)
Λ2
and
ρ6
γ (2,1) (0) = 2ρ4 , γ (0,2) (0) = 2ρ5 , γ (0,1) (0) = . (3,49)
Λ2
Unluckily, until now, nobody has been able to solve our equations in any in-
teresting case in 4 dimensions. The analytical Wilson renormalization group
method has been successfully applied only ”near” two dimensions [8] . What
we are going to do in the following is to apply to our equations a purely
dimensional analysis and, on this basis, to discuss their pertubative solution.

18
4. Analysis of the perturbative solution.

We shall now briefly discuss the perturbative solution to (3,41).

In the perturbative framework the effective lagrangian and its initial value
L0 are considered as a formal power series in h̄. Therefore in particular the
bare constants ρ0 and the vertices Γ, Γ̂ and γ are formal power series. In this
framework the solution to (3,41) can be directly constructed as follows.
We notice, first of all, that the initial value l0 (Λ0 ) of lef f is directly ob-
tainable by applying to (2,13) the transformations (3,27) and (3,31) which
yield:
m2 − m2
" #
Z
λ0 4 z0 − 1 ξ0
l0 (Λ) = d4 x φ̂ + (∂ φ̂)2 + 0 2 φ̂2 + ζ0 ω̂ φ̂2 + η0 ω̂ 2 + 2 ω̂ ,
4! 2 2Λ Λ
(4,1)
where we have used the Fourier transformed variables:
Z
φ̂(~x) = d4 pe−i(~p·~x) φ(~p) . (4,2)

We also introduce the cut-off propagator:

k 2 ( pΛ
Λ0
) − k 2 (p)
∆(p) = m2
. (4,3)
p2 + Λ2

Now it is sufficient to notice that:

" #
∂
Λ − N − 2M − D l0 (Λ) = 0 , (4,4)
∂Λ
and the commutation relation:
"" # #
∂ Z
∆(p) δ δ
Λ − N − 2M − D , d4 p 4
=
∂Λ (2π) δφ(~p) δφ(−~p)
Z
∆0 (p) δ δ
d4 p 4
, (4,5)
(2π) δφ(~p) δφ(−~p)
to prove that;
∆(p)
 h̄
R δ δ

d4 p l0 (Λ)
lef f = −h̄ ln e 2 (2π)4 δφ(~p) δφ(−~p)
e− h̄ , (4,6)

19
is the solution of (3,41) with initial value l0 (Λ0 ). It is also easy to verify
that lef f is the functional generator of the connected and amputated Feyn-
man amplitudes corresponding to the propagator ∆(p) and to the vertices
described by l0 .
Using this result one could try to study the existence of an UV limit
of the perturbative theory. However we shall see in the following that, in
order to have a regular UV limit, we have to consider as fixed parameters
the renormalized ρ’s, given in (3,10) computed at Λ = ΛR .
From (4,6) we find in particular:
ρ0,3
(2,0) 2 ρ0,2 p2 + Λ2
γ (p ) = 
ρ0,3
+ O(h̄) ,
1 + ρ0,2 p2 + Λ2
∆(p2 )
γ (4,0) = ρ0,1 + O(h̄) , (4,7)

and
6  
γ (6,0) (~p1 , .., p~6 ) ρ20,1 ∆ (~pi1 + p~i2 + p~i3 )2
X
. (4,8)
i1 ≥i2 ≥i3 =1

The analysis of the UV limit will develop along the following lines. We shall
introduce a suitable class of norms for lef f and its Λ0 -derivative v(Λ) in the
new parametrization. We then show that, for Λ0 large with respect to ΛR
and Λ and at any finite perturbative order, this functional vanishes in the
 2−
Λ
UV limit faster than Λ0 times lef f for any positive . This will im-
mediately imply the UV convergence of the theory. Now (4,6) is explicitly
parametrized in terms of the bare constants which, written as functions of ρR ,
coincide with the counter-terms associated with the corresponding vertices.
These are h̄-ordered series. Thus the reparametrization of lef f in terms of the
renormalized constants is equivalent to the introduction of a subtraction pro-
cedure. This is however extremely cumbersome. In particular checking the
mechanism of subtraction of the UV divergent contributions of subdiagrams
with the corresponding counter-terms, requires a detailed analysis of the so-
called overlapping divergences. The advantage of the method presented here,
which is inspired by the work of Polchinski, is that, using explicitly the evo-
lution equations (3,2) and (3,19), one can reach the same results without any
diagrammatic analysis.
Concerning the just mentioned change of parametrization, let us notice
that, as shown above in the tree approximation, the two-field vertex γ (2,0)

20
involves tree parameters, m, ρ2 and ρ3 . However the quantity that is physi-
cally relevant is the correlation length that in our approach turns out to be
a function of the above mentioned parameters. This seems to leave a wide
freedom in particular in the choice of ρ2 and ρ3 that can be compensated by
that of the covariance m2 + p2 , leaving the correlation length unchanged.
However we shall see in a moment that in a loop ordered perturbation
theory, to simplify the recursive solution of the evolution equations, it is
highly convenient to choose the two-field vertex Γ(2,0) vanishing at the tree
level. This means that ρ2 and ρ3 are not ”really” free parameters. They must
be of order h̄. This assumption is perfectly compatible with our evolution
equations from which we can compute, in the case of a sharp cut-off
h̄
ρ2 = ρ0,2 − ρ0,1 , (4,9)
2(4π)2

ρ3 = ρ0,3 + O(h̄2 ) , (4,10)

ρ1 = ρ0,1 + O(h̄)ρ20,1 . (4,11)
 
∂ρ
It is also clear from (4,7) and this result that the jacobian matrix ∂ρ 0
and
the functional ∂ρ lef f remain perfectly regular after this choice.
Owing to the nature of the normalization conditions, we have to study
the Λ dependence of the vertices generated by lef f , ∂ρa lef f and v together
with their momentum derivatives.
The evolution equations for the momentum derivatives of the vertices
are immediately deducible from (3,32). One can verify in particular from
this equation how the derivatives distribute in its the right-hand side. For
purely formal reasons we briefly discuss here how the insertion of momentum
derivatives can be performed at the functional level. This point, that is not
crucial to the study of the perturbative UV limit, is motivated by the formal
need to translate the whole normalization group action into a finite number
of differential equations for the functional generators.
To define these momentum derivatives within the functional framework
we introduce the family of differential operators:
n n+m
 Y δ δ
Dν,n,m (~p1 , ..., p~n+m ) = Pν ∂~pk − ∂~pl
Y
, (4,12)
i=1 δφ(~
pi ) j=n+1 δω(~pj )

21
where Pν is a polynomial of degree d(ν). Considering in particular the evo-
lution equations for Dν lef f one finds that, due to the non-linearity of (3,32),
these equations contain a great number of terms. Let us consider for example:

~ 1,2,0 (~p1 , p~2 ) = ∂~p1 − ∂~p2
 δ δ
D . (4,13)
δφ(~p1 ) δφ(~p2 )
and notice that for any pair of functionals A and B one has:

~ 1,2,0 (~p1 , p~2 )

Z
∆0 (p) δ δ
D d4 p 4
A B=
(2π) δφ(~p) δφ(−~p)
Z
∆0 (p) δ ~ δ
d4 p 4
D1,2,0 (~p1 , p~2 ) A B+
(2π) δφ(~p) δφ(−~p)
Z
∆0 (p) ~ δ δ
d4 p 4
D1,2,0 (~p1 , p~) A B+
(2π) δφ(~p2 ) δφ(−~p)
Z
∆0 (p) δ δ ~ 1,2,0 (−~p, p~2 ) B −
d4 p 4
AD
(2π) δφ(~p1 ) δφ(~p)
Z
∆0 (p) δ δ δ δ
d4 p∂~p A B + (B ↔ A) . (4,14)
(2π)4 δφ(~p1 ) δφ(~p) δφ(~p2 ) δφ(−~p)
Iterating the same equation, we see that for a generic operator Dν :
Z
∆0 (p) δ δ
Dν,n,m d4 p 4
A B=
(2π) δφ(~p) δφ(−~p)
X Z ∆0 (p)
d4 pPρ (∂~p ) Dσ,k+1,l ADτ,n+1−k,m−l B , (4,15)
ρ,σ,τ,k,l (2π)4

where the sum is restricted to the differential operators and polynomials for
which:
d(ρ) + d(σ) + d(τ ) = d(ν) . (4,16)
We have thus the evolution equation:
" #
∂
Λ − N − 2M − D − n − 2m − d(ν) Dν,n,m lef f =
∂Λ
h̄ Z 4 ∆0 (p) δ δ
dp 4
Dν,n,m lef f −
2 (2π) δφ(~p) δφ(−~p)
1 X Z 4 ∆0 (p)
d pPρ (∂~p ) Dσ,k+1,l lef f Dτ,n+1−k,m−l lef f , (4,17)
2 ρ,σ,τ,k,l (2π)4

22
where again the sum in the right-hand side is restricted by (4,16). Analogous
equations hold true for ∂ρa lef f and v.
This completes the construction of the evolution equations. To study their
solutions we have to translate these evolution equations for the generating
functional into those for the corresponding N field and M operator vertices.
For this we have to clarify a technical point. Since the functional generators
are translation invariant, the corresponding vertices are proportional to a
Dirac delta function in the sum of the external leg momenta. To get rid of
this delta function we notice that, given any translation invariant functional
A, the (ν, n, m) momentum derivative of the corresponding (N, M ) vertex is
given by
n+m+N
Z
d4 k δ
Dν,n,m A(N,M ) (~p1 , ..., p~n+m ) = (N + M )4
Y
(2π)4 i=n+m δφ(~pi + ~k)
n+m+N
Y +M δ  
Dν,n,m p~1 + ~k, ..., p~n+m + ~k A|φ=ω=0 . (4,18)
j=n+m+N +1 δω(~pj + ~k)

For these vertices we define the norms:

kDν,n,m Ak(N,M ) ≡ supp2i <2,P p~i =0 |Dν,n,m A(N,M ) | . (4,19)

The value of these norms in the tree approximation can be easily computed
from (4,6), (4,7) and (4,8) from which it turns out that the norms of lef f
can be bounded above and below by constants for a generic choice of the
parameters ρ and for:
ΛR ≤ Λ << Λ0 . (4,20)
Furthermore from (4,6) and (3,42) one easily finds that:
2
Λ

(N,M )
kv(Λ)k < CN,M klef f k(N,M ) . (4,21)
Λ0
Notice that the condition Λ0 >> Λ is needed to guarantee the assumed lower
bound on the vertices; indeed when Λ tends to Λ0 almost all of them vanish
together with the propagator (4,3). In the following we shall disregard (4,20)
whenever we discuss upper bounds.
To push our analysis to all orders, the first step is to evaluate the norms
of lef f and ∂ρ lef f . We shall then discuss those of v.

23
 We study the evolution equations recursively for increasing order (k) in
h̄ and degrees (N and M) in the fields.
From now on we shall label by the index k the k th order term of a vertex.
For lef f we assume the induction hypothesis that:

Λ
  
kDν γk k(N,M ) < Pk ln , (4,22)
ΛR
where Pk is a polynomial of degree k with positive cofficients. (4,22) is, of
course, verified for k = 0.
Let us notice now that the essential simplification introduced by pertur-
bation theory consists in the possibility of performing a recursive analysis of
(3,32) for increasing loop order and for any given order in h̄ increasing N
and M starting from (2, 0). This is possible since the right-hand side of the
evolution equation for the (N, M ) vertex at the perturbative order k depends
on the (N − 2, M ) and (N, M − 1) vertices at the same order, while the other
vertices appearing in it have lower orders. Indeed, as already noticed, the
first term in the right-hand side of (3,32) corresponds to the product of two
vertices joined by one line, this has a tree graph structure, hence the sum
of the orders of the two vertices is k, the total number of φ-legs is N + 2,
that of ω-legs is M . Owing to the particular choice of the parametrization
discussed above, which implies the vanishing of the (2, 0) vertex in the tree
approximation, this first term does not contain vertices of order k with more
than N − 1 φ-legs and M ω-legs. In the second term in the right-hand side
of (3,32) there appears only the (N + 2, M ) vertex at the order k − 1.
A further remark which is essential to obtain an upper bound for the
absolute value of the right-hand side of (3,32) at each recursive step, is that
∆0 (p) is absolutely bounded together with all its derivatives and that it van-
ishes identically outside the hypersphere of radius 2 and inside that of radius
1.
Taking into account these remarks and (4,22) we conclude that at the
order k the absolute value of the right-hand side of (3,32) is bounded by a
polynomial of degree k in ln ΛΛR .
Coming to the study of the single vertices, let us consider, first of all:
(2,0) (2,0)
D4µνρσ γk+1 ≡ ∂pµ ∂pν ∂pρ ∂pσ γk+1 (~p) =
(2,0)
4 (δ µν δ ρσ + δ µρ δ σν + δ µσ δ ρν ) γk+1 (p2 )(ii) +

24
 8 (δ µν pρ pσ + δ µρ pσ pν + δ µσ pρ pν + δ ρσ pµ pν + δ σν pµ pρ +
(2,0) (2,0)
δ ρν pµ pσ ) γk+1 (p2 )(iii) + 16pµ pν pρ pσ γk+1 (p2 )(iv) , (4,23)
(2,0) (2,0)
where γk+1 (p2 )(x) stays for the xth derivative of γk+1 (p2 ) with respect to p2 .
We have:
!
∂ Λ
  
k Λ − D − 2 D4 γk+1 k(2,0) < Pk+1 ln , (4,24)
∂Λ ΛR

which has to be integrated with the initial condition:

(2,0)
D4 γk+1 |Λ=Λ0 = 0 . (4,25)

Now we can integrate (4,25) as follows. Setting:

2
Λ

(2,0)
D4 γk+1 (~p) = F (Λ~p, Λ) , (4,26)
Λ0
one has: 2
∂ Λ0 Λ
   
|Λ
F| < Pk+1 ln , (4,27)
∂Λ Λ ΛR
then, integrating, one gets:
2
Λ0 Λ
   
0
|F | < Pk+1 ln , (4,28)
Λ ΛR
(2,0)
and hence one recovers (4,22) for D4 γk+1 .
It is important to notice, at this point, that the initial condition (4,25) has
not been relevant for the final result (4,22). This is due to the fact that the
initial condition contributes to the solution of (4,17) proportionally to that
of the associated homogeneous equation (the one with vanishing right-hand
side). Considering a momentum derivative of order d of the vertex (N, M )
 N +2M +d−4
this homogeneous solution is proportional to ΛΛ0 : it is therefore
irrelevant if the exponent is positive. Now we compute:
1 Z (2,0)
dt(1 − t)3 pµ pν pρ pσ D4µνρσ γk+1 (t~p) =
3!
(2,0) (2,0) (2,0)(i)
γk+1 (p2 ) − γk+1 (0) − p2 γk+1 (0) . (4,29)

25
Notice that, independently of the presence of a mass, our vertices are not
affected with infra-red singularities since the propagator (4,3) is cut-off both
at high and low momenta.
Then we get:
Λ0
  
(2,0) (2,0)(i)
kγk+1 k(2,0) < Pk0 ln + |γk+1 (0)| + 4|γk+1 (0)| . (4,30)
Λ
The second and third term in the right-hand side can be computed using
again the evolution equations. However now these equations have to be
integrated between ΛR and Λ since the value of both terms is fixed by the
renormalization conditions (3,10) at ΛR .
In much the same way as above we consider:
(2,0) (2,0) (2,0)(i)
D2µν γk+1 (0) ≡ ∂pµ ∂pν γk+1 (0) = 2δ µν γk+1 (0) , (4,31)

for which the evolution equation gives:

∂ Λ
  
(2,0)
|Λ D2 γk+1 (0)| < Pk ln , (4,32)
∂Λ ΛR
T he right-hand side of this equation receives contributions only from the
term proportional to h̄ in (3,32) since ∆0 (p) vanishes identically for p < 1.
Integrating (4,32) between ΛR and Λ and taking into account (4,31) yields:

Λ
  
(2,0)(i)
|γk+1 (0)| < Pk+1 ln . (4,33)
ΛR
Notice that the degree of the polynomial P has increased by one. This will
happen every time we shall integrate a vertex evolution equation whose left-
hand side is of the form of that in (4,32).
In an analogous way we have:
!
∂ Λ
  
(2,0)
| Λ + 2 γk+1 (0)| < Pk ln , (4,34)
∂Λ ΛR

leading, after integration, to:

Λ
  
(2,0)
|γk+1 (0)| < P ”k ln . (4,35)
ΛR

26
Notice that here the normalization at ΛR has an irrelevant influence on the
value of the right-hand side.
Now, combining (4,30), (4,33) and (4,35), we get back the induction hy-
pothesis at order k + 1 for klef f k(2,0) .
We have discussed in some detail this step of the recursive proof to give an
example of the general strategy, which is essentially the following. For every
(N,M) we consider a momentum-derivative of the vertex of high enough order
to make it ”convergent” (N+2M+the order of derivation larger than 4) and
we compute this momentum derivative integrating the evolution equation
between Λ and Λ0 where it has to vanish.
The original vertex, or its derivatives of low order, are then computed
from the derivative of higher order integrating it with respect to a momen-
tum scale variable, as in (4,29). The resulting formula contains integration
constants, corresponding to the values at zero momentum of the vertex and
its derivatives of low order (its ”divergent part”). These are computed inte-
grating their evolution equations between Λ and ΛR where they are assigned
by the normalization conditions.
The procedure turns out to be much simpler in the case of ”convergent”
vertices (N + 2M > 4) which are determined directly by integrating the
evolution equation between Λ and Λ0 and in the case (N = 0 , M = 1)
where it is sufficient to integrate between Λ and ΛR .
The results that we have so far reached concerning lef f could easily be
resumed formulating a theorem. We avoid this assuming that the crucial
steps of our study have been sufficiently clarified.
The above method can be employed to evaluate the norms of ∂ρ lef f for
which we assume the induction hypothesis that:
Λ
  
(N,M ) −da
k∂ρa Dν γk k <Λ Pk ln , (4,36)
ΛR
where, as mentioned above, da is the mass dimension of the parameter ρa .
The validity of (4,36) in the tree approximation can be read directly from
(4,1). To extend (4,36) to all orders we repeat the analysis described above
for lef f noticing that the initial conditions at ΛR and Λ0 for ∂ρa lef f can be
easily deduced from those for lef f .
After infinite volume limit and momentum rescaling the evolution equa-

27
tion for ∂ρ lef f becomes:
" #
∂
Λ − N − 2M − D ∂ρa lef f =
∂Λ
m[∂ρa lef f ] − Λ−db nb [m[∂ρa lef f ]] ∂ρb lef f . (4,37)

Let us remark that in the second term of the right-hand side of this equation
the factor Λ−db coming from the normalization operator compensates Λdb
appearing in (4,36). Therefore the whole right-hand side turns out to be
proportional to Λ−da .
As above we start from the study of ∂ρa D4 lef f for which we have:
!
∂ Λ
  
k Λ − D − 2 ∂ρa D4 lef f k(2,0) < Λ−da Pk+1 ln . (4,38)
∂Λ ΛR

The integration between Λ and Λ0 (4,38) leads back to (4,36) for ∂ρa D4 lef f .
It is now evident that if ρa is a dimensionless parameter (da = 0) the recursive
proof of (4,36) is identical to that of (4,22). However, if da > 0 , we have to
change the strategy of our proof. Indeed, for example, the evolution equation
(2,0)
of ∂ρa D2 γk+1 (0) yields:
∂ Λ
  
(2,0)
|Λ ∂ρa D2 γk+1 (0)| < Λ−da Pk+1 ln . (4,39)
∂Λ ΛR
Comparing this with (4,26) and (4,32) we see that integrating down from
Λ0 we can profit of the irrelevancy of the high energy value in the present
situation. Thus we get:
Λ
  
(2,0)
|∂ρa D2 γk+1 (0)| < Λ−da Pk+1
0
ln . (4,40)
ΛR
In much the same way we can show that:
Λ
  
(2,0)
|∂ρa γk+1 (0)| < Λ−da Pk+1 ” ln . (4,41)
ΛR
(2,0)
Finally, using all these results to compute ∂ρa γk+1 (p2 ), by means of an in-
tegration procedure completely analogous to (4,25), we prove the validity of
the induction hypothesis for this derivative. Following the same strategy, one
(4,0)
can study ∂ρa γk+1 starting from a suitable second derivative.

28
 Then, increasing step by step N and M , one can prove (4,36) for all
the coefficients of ∂ρa lef f at the order k + 1 completing the proof of this
hypothesis.
We now come to the study of v defined in (3,43) and notice, first of all,
that in its evolution equation every factor Λda coming from a normalization
operator compensates the one appearing in the norms of ∂ρa lef f .
In much the same way as in the already discussed cases, the analysis be-
gins in the tree approximation where v can be computed directly from (3,43)
and (4,6)-(4,11). Indeed, in this approximation, lef f depends on Λ0 only
through the propagator (4,3) since the bare and renormalized parameters
coincide. Now we have:
2 ! 2
∂ 1 pΛ Λ pΛ
   
Λ0 ∆(p2 ) = − 2 m2
p~ · ∂~p k 2
=− ∆0 , (4,42)
∂Λ0 p + Λ2
Λ0 Λ0 Λ0

thus: 2
∂ Λ

sup|Λ0 ∆| = sup|∆0 | . (4,43)
∂Λ0 Λ0
One has also to notice that a cut at low momenta appears in Λ0 ∂Λ∂ 0 ∆ at
p = ΛΛ0 (while in ∆ it appears at p = 1), therefore the Λ0 -derivative of a
vertex vanishes unless the momenta of a great number of external momenta
add together to overcome this cut. Thus this derivative does not vanish only
for vertices whose external legs increase proportionally to ΛΛ0 .
Anyway, recalling the structure of the tree diagrams and (4,38), we can
conclude that:
Λ 2
 
(N,M )
kDν v0 k < CN,M,ν . (4,44)
Λ0
In order to study the higher perturbative orders using, as above, the evolution
equation, we have to discuss, first of all, the initial condition at Λ = Λ0 .
Before the infinite volume limit this is given in (3,22). After this limit and
the rescaling of momenta shown in (3,31) we have:

1 Z 4 ∆0 (p)
"
δ δ
v(Λ0 ) = dp l0 (ρ 0 ) l0 (ρ0 )−
2 (2π)4 δφ(~p) δφ(−~p)
" # #
δ δ
Na l0 (ρ0 ) l0 (ρ0 ) ∂ρa l0 (ρ0 ) + h̄C , (4,45)
δφ(~p) δφ(−~p)

29
where C is field independent. Recalling (4,22) we find for this initial condition
the upper bound:
Λ0
  
(N,M )
kDν vk (Λ0 )k < Pk ln . (4,46)
ΛR
It is now possible to apply the evolution equations for v repeating the analysis
already developed for lef f and ∂ρa lef f . In the present case the recursive
procedure will be based on the induction hypothesis:
2
Λ Λ0
   
kDν vk k(N,M ) < Pk ln , (4,47)
Λ0 ΛR
which is suggested by (4,44) and by the fact that the initial condition at
Λ gives irrelevant contributions to v (in this case irrelevant means of order
 0 2
Λ
Λ0
).
Considering the right-hand side of the evolution equation of v, that has
essentially the same structure of that of lef f but is linear in v, we find that:
" #
∂
k Λ − N − 2M − D − n − 2m − d(ν) Dν,n,m vk+1 k(N,M ) <
∂Λ
Λ 2 Λ0
    
Pk+1 ln . (4,48)
Λ0 ΛR
We then begin, order by order and in complete analogy with lef f , from D4 vk ,
with D4 defined in (4,23), and we find:
2
Λ Λ0
   
(N,M ) 0
kD4 vk+1 k < Pk+1 ln . (4,49)
Λ0 ΛR
Notice that here the initial value at Λ0 gives a relevant contribution to (4,49).
(2,0)
Applying on D4 vk+1 the integration procedure exhibited in (4,29) we find
(2,0) (2,0)(i)
directly kvk+1 k(2,0) since vk+1 (0) and vk+1 (0) vanish. Indeed from (3,12)
one finds
Na V (Λ) = 0 . (4,50)
Then, step by step, we consider the coefficients of v with increasing N
and M completing the proof of (4,47).

30
 From this inequality we can prove the existence of an UV limit for the
theory. Indeed from the definition of v we have:

lef f (ΛR , Λ0 , ρ0 (ΛR , Λ0 , ρR )) − lef f (ΛR , Λ0 0, ρ0 (ΛR , Λ0 0, ρR )) =

Z Λ0 0
dx
v (ΛR , x, ρ0 (ΛR , x, ρR )) . (4,51)
Λ0 x
This equation can be translated in terms of the coefficients and, using (4,43),
it leads to:

klef f [ΛR , Λ0 , ρ0 (ΛR , Λ0 , ρR )] − lef f [ΛR , Λ0 0, ρ0 (ΛR , Λ0 0, ρR )] k(N,M ) <

2
2 ΛR Λ0
  
2
|Λ0 − Λ0 0 | 4 Pr ln , (4,52)
Λ0 ΛR
which proves, to any perturbative order r, the wanted convergence property.
Indeed (4,52) shows that, if the effective parameters ρR,a are kept fixed, the
vertices of the effective theory satisfy the Cauchy convergence criterion for
Λ0 → ∞.
A further comment about the nature of the UV limit is here possible if
we translate (4,22) in terms of the effective vertices Γ in (3,28). Indeed we
have that, for p < Λ and q < Λ the coefficients in the expansion of Lef f in
series of KΛ φ and ω satisfy, at the order r:
Λ
  
|Γ(n,m)
r | ≤ Pr log (Λ)4−n−2m . (4,53)
ΛR
For the momentum derivatives of order ν of these vertices the power behavior
in the right-hand side of (4,53) is increased by ν.
From this inequality we can see in particular the effect of the introduction
into L0 of terms with dimension d higher that 4 (e. g. 6). Indeed these
terms can be introduced first as composite operators together with their
sources of negative dimension (e. g. -2) and then they can be inserted
into the bare interaction by replacing their source with a coupling coefficient
proportional to Λ4−d0 . Now (4,53) shows that at the first order in the coupling
(n,m)
the effect on Γ of the insertion into the bare lagrangian of an operator of
dimension d > 4 is proportional to Λd−n−2m , thus, the ratio of this versus the
 d−4
unperturbed effective coefficient vanishes at low energy as ΛΛ0 . Therefore
the effect of a ”non-renormalizable” term in L0 is ”irrelevant” at low energies.

31
 This result agrees with the previous remark that in the evolution of ”fi-
nite” vertices or derivatives of vertices a change in the initial conditions at
Λ0 gives irrelevant contributions to lef f for Λ << Λ0 .
As a further example of irrelevant operators that will be important in the
study of quantum breaking of symmetries. Let us consider a bare operator
that in the limit Λ0 → ∞ is local and has finite dimension d, while for finite
Λ0 it is non-local and contains contributions of any dimension. Suppose also
that the operator satisfies vanishing normalization conditions at ΛR at all
perturbative orders. It is possible to show that the operator vanishes in the
UV limit.
Indeed let us consider the evolution equation of the corresponding effec-
tive operator, the derivative of the effective lagrangian with respect to the
corresponding source in the origin. The evolution equation is directly ob-
tained from (3,41) and it is strictly analogous to it. However it is linear
in the effective operator. This equation can thus be recursively solved in-
creasing the dimension and the perturbative order of the vertices. In the
first step, increasing the dimension of the vertices and keeping their order
fixed, one begins integrating the evolution equation from ΛR . Due to the
vanishing normalization conditions and to the linearity of the equation one
finds vanishing vertices. However, if the dimension of the vertex overcomes
that of the operator, one has to integrate down from Λ0 and hence one finds
non-vanishing results due to the non-locality of the of the bare operator.
The vertices of high dimension are however proportional to a positive integer
power (n) of ΛΛ0 as are the irrelevant contributions to the solution of (3,41).
 n
Thus one finds a global upper bound proportional to ΛΛ0 for the vertices
of the effective operator. It is clear that this upper bound holds true to all
orders. This proves our claim.

32
5. Composite operators and Wilson expansion.

Having so concluded the general study of the UV limit, we have now to

deepen the analysis of composite operators. Until now the general method
to introduce these operators in the framework of the functional technique
has been illustrated referring to the operator φ2 . The method is based on
the introduction of a source (ω) of given dimension (2). We have also seen
that, in order to identify completely an operator one has to assign a suitable
set of normalization conditions. Among these conditions those correspond-
ing to lagrangian terms non-linear in the source concern possible multi-local
contributions to the Green functions of many composite operators.
We shall now make some comments on this procedure, limiting however
our attention to the Green functions of a single composite operator and an
arbitrary number of fields.
The first point to be considered is the choice of the dimension of the
source. In our example we have requested the bare lagrangian to have di-
mension 4 and therefore we have assigned dimension 2 to the source of an
operator of dimension 2. Following our study of the effective lagrangian and
of the corresponding evolution equation, it is possible to have a better un-
derstanding of this choice. Indeed the dimension of the source has influenced
its rescaling in the infinite momentum limit (3,31) and hence the evolution
equation of lef f . On the other hand, in the tree approximation, the bare
structure of the composite operator determines directly the scaling behavior
of the vertices containing its source; this behavior has to match with the
evolution equation in order to be reproduced to all orders of perturbation
theory. In this way we have an upper bound for the source dimension.
As we have seen in the analysis of the evolution equation of lef f , once
the dimension δ of the source has been fixed, in order to define the operator
completely, we have to assign a normalization condition for every ”divergent”
vertex involving the source, that is, for every monomial in the quantized field
and its derivatives with dimension up to 4 − δ, that of the operator we are
willing to define.
Concerning the bare structure of an operator, that fixes the initial condi-
tions at Λ0 , it has been repeatedly remarked that, after infinite volume limit
and momentum rescaling, the addition of terms of higher dimension with
Λ0 -dependent coefficients gives irrelevant contributions to the operator.

33
 Thus, in conclusion, in the UV limit a composite operator is specified by
assigning its dimension 4 − δ, or, equivalently, the dimension of its source δ,
and a polynomial operator P with dimension up to 4 − δ which defines its
normalization conditions at ΛR . An analogous characterization of composite
operators has been introduced by Zimmermann [3] who has represented them
with the symbol N4−δ [P ]. We can use here the same symbol provided we take
into account the normalization point ΛR which in Zimmermann’s definition
is equal to zero.
It is interesting to notice here that, given an operator of dimension 4−δ, if
one considers  only
η the vertices linear in tis source and if multiplies all these
Λ
vertices by ΛR for η > 0, one gets the vertices of a new operator with
dimension 4−δ +η. Indeed the new vertices satisfy the evolution equations of
the vertices of an operator of this dimension since these evolution equations
are linear in the operator vertices. The normalization conditions for the
vertices of this new operator can be directly computed from the vertices of
the original one at ΛR . The number of the conditions for the new operator is
larger than that of the original one since the dimension is higher. Comparing
the vertices of the two operators one gets a linear relation analogous to that
connecting in Zimmermann’s scheme operators with different dimension.
A further important remark concerning our approach to composite op-
erators is that nowhere in our analysis the locality of the bare operator has
played an essential role. Indeed, to identify the relevant normalization pa-
rameters of a certain operator, we have used the evolution equations of its
vertices. These equations are obtained, independently of the locality of the
operator, by taking the source derivative of both members of (3,2) and hence
they have the same structure of (3,14). Once the evolution equations have
been determined one has to identify a recursive hypothesis for the upper
bounds of the vertices. As we have said above these bounds are set together
with the dimension of the source in the tree approximation. Hence, in par-
ticular, if a bare operator is non-local involving fields in split points at a
distance of order Λ10 , in our analysis it appears as a strictly local operator.
Then the analysis follows strictly that of the previous section identifying
the relevant normalization parameters with the zero momentum value of the
vertices whose dimension does not exceed 4 taking into account also the
dimension of the source.
We can therefore apply our results to multilocal products of operators

34
to study their Wilson short-distance expansion. We can therefore apply our
results to To put into evidence the origin and the nature of this expansion we
simplify as much as possible our discussion referring to the product of two
field operators, that is: to the operator whose bare structure is the product of
two bare fields at different points (φ(x)φ(0)). The crucial new aspect of our
analysis consists in considering this product, for what concerns the effective
lagrangian, as a single composite operator. In particular we associate a source
ωX to it and we study the dependence of lef f on this source.
Notice that considering the product of two fields in different points as a
single composite operator implies a modification of the concept of connect-
edness of a diagram.
As discussed above, the first step in the analysis of our composite op-
erator consists in assigning a dimension to ωX . Owing to the fact that the
contributions of Feynman diagrams to the tree vertices involving our opera-
tor depend on x through a phase factor eipx , where p is a partial sum of the
external momenta, we have that the largest possible value of the dimension
of ωX is 2, the same the local operator φ2 .
Having specified the dimension of the source of our bilocal operator, we
consider now its normalization conditions. According to the above discussion
in order to define (φ(x)φ(0)) as a single composite operator, we have to assign
its vacuum vertex, that is: the coefficient of lef f linear in the source ωX
and of degree zero in the field, and the vertex with two legs with vanishing
momentum. These vertices can be easily deduced from the vertices of lef f
involving only φ-legs. It is sufficient to consider respectively the vertices
γ (2,0) (~p) and γ (4,0) (~p, −~p, 0, 0), to multiply then by ei~p·~x ∆(p)2 and to integrate
with respect to p~.
It is therefore possible, in our theory, to define for every x a renormalized
operator of dimension 2 corresponding to the product of two bare fields at dif-
ferent points. This is identified with an operator of dimension two satisfying
the following normalization conditions:
i) For the vacuum vertex, :
Z
d4 k δ Z
d4 p ipx 2 h
(2,0) 2
i
l |
ef f φ=ωX =0 = e ΛR ∆(p) 1 + γ (p )∆(p) .
(2π)4 δωX (~k) (2π)4
(5,1)

35
 ii) For the coefficient with two field legs at zero momentum:
Z
d4 k δ δ2 Z
d4 p ipx 2
lef f |φ=ω =0 = e ∆ (p)γ (4,0) (~p, −~p, ~0, ~0) .
(2π)4 δωX (~k) δφ2 (~k) X
(2π)4
(5,2)
As it is evident already at the tree level, these normalization conditions
are by no means uniform in x, in particular the vacuum vertex is proportional
to x12 at short distances. This lack of uniformity in x can however be taken
into account in the following way. We consider the linear combination of
composite operators:
(ΛR ) (ΛR )
N2 [φ(x)φ(0)] ≡ φ(x)φ(0) − A(x) − B(x) N2 [φ2 (0)] . (5,3)

Here A(x) and B(x) are c-numbers chosen in such a way that the new com-
(Λ )
posite operator N2 R [φ(x)φ(0)], satisfies normalization conditions uniform
in x. More precisely:
i) Its vacuum vertex has to vanish.
ii) The two field vertex at zero momentum has to be equal to two.
Now, remembering the analysis of the evolution equations, we notice that
the new operator has vertices uniformly bounded in x owing to the unifor-
mity of the normalization conditions at ΛR . Therefore the introduction of
(Λ )
N2 R [φ(x)φ(0)] can be seen as a decomposition of a singular operator prod-
uct into the sum of a singular and a regular part:
(ΛR ) (ΛR )
φ(x)φ(0) = A(x) + B(x) N2 [φ2 (0)] + N2 [φ(x)φ(0)] . (5,4)

The first two terms in the right-hand side give the singular part. Indeed
taking into account (5,1) and (5,2) it is possible to verify that:
1
A(x) ∼ P (ln(xΛR )) , (5,5)
x2
and that:
B(x) ∼ Q (ln(xΛR )) , (5,6)
where P and Q are polynomials of increasing degree with the perturbative
order. This requires a little work on the evolution equation (4,22) that, after
UV limit, can be transformed into the Callan-Symanzik equation [7] and then
used to study the presence of log’s.

36
 The decomposition (5,4) is the essential part of Wilson operator prod-
uct expansion [3] , it can be obviously extended to multilocal products of
composite operators. Starting from this expansion it is possible to analyze
the Green functions of the theory [4] as distributions recovering the results
of Bogoliubov, Parashiuk, Hepp, Zimmermann, Epstein, Glaser and Stora
and hence proving that our theory leads to the same Green functions as the
subtraction methods mentioned in the introduction.

37
6. The quantum action principle.

Now we discuss how one can characterize the invariance of a theory under
a system of, in general non-linear, continuous field transformations.
In the Feynman functional framework, the invariance of the theory cor-
responds to that of the functional measure and its consequences are simply
exhibited by introducing the field transformations as a change of variables in
(2,21).
A simple but significant example of a model with non-linear symmetry is
provided by the two-dimensional non-linear σ-model. With a simple choice
of coordinates the classical action of this model is written:
Z    √ 2  √ 
2 2 2 2 2 2 2
S = d x z (∂~π ) + ∂ F − π +m F −π −F , (6,1)

where ~π is an isovector field with vanishing dimension. This action is in-

variant under the linear field transformations corresponding to rotations in
isotopic space. However to characterize completely (6,1) we have to exploit
the transformation property of the action under the non-linear transforma-
tion: √
~π → ~π + ~ F 2 − π 2 , (6,2)
where ~ is an infinitesimal, space independent isovector. It is easy to verify
that the corresponding variation of the classical action is:
Z
− d2 x m2 ~ · ~π . (6,3)

If one asks, together with the linear isotopic invariance, that the variation of
the action under (6,2) be (6,3), one identifies the action (6,1) up to the choice
of z. This example is presented to put into evidence the role of symmetries
in the identification of the action and hence in the construction of quantum
field theories.
In the general case we shall consider a multicomponent field φi and a set
of infinitesimal, non-linear, rigid field transformations, that we shall write,
disregarding the indices, according:
p
   
φp~ → φp~ +  T φp~ + k P0,~p (KΛ0 φ, ω, 0) , (6,4)
Λ0

38
where the matrix T is assumed to be traceless and to satisfy:
T C(p) ± C(p)T T = 0 , (6,5)
where the sign has to be chosen in agreement with the field statistics. It is
apparent that this equation is equivalent to the invariance of the free part of
the action under the linear part of the field transformation. The polynomial
P0,~p (KΛ0 φ, ω, 0) accounts for the strictly non-linear part of the transforma-
tion, let us indicate by dP the canonical dimension of the corresponding
operator.
To define the composite operator P within the functional framework, we
introduce its source γ, with dimension 4 − dP , by the substitution:
γp~ P0,−~p (KΛ0 φ, ω)+O(γ 2 ) ≡ L0 (KΛ0 φ, ω, γ) .
X
L0 (KΛ0 φ, ω) → L0 (KΛ0 φ, ω)−
p
~
(6,6)
Notice that the introduction of a composite operator of source γ is in
general accompanied by terms of degree higher than one in γ which appear
in the term O(γ 2 ) in (6,6). The presence of these terms induces a change of
the infinitesimal transformation law (6,4) that we shall write according:
p
   
φp~ → φp~ +  T φp~ − k ∂γ−~p L0 (KΛ0 φ, ω, γ) ≡
Λ0
p
   
φp~ +  T φp~ + k P0,~p (KΛ0 φ, ω, γ) . (6,7)
Λ0
In general the substitution shown in (6,6) produces symmetry breaking terms
of O(γ). We shall see however that in the case of gauge theories the nilpotency
of the transformation (6,4) allows a precise control of these O(γ) terms.
Introducing (6,7) as a change of variables into the Feynman formula (2,30)
and requiring the invariance of the functional integral, we get, to first order
in :
Z Y X   p  h  i
dφp~ k jp~ − φp~ C(p) − ∂φ−~p L0 ∂γp~ L0 + ∂φ−~p ∂γp~ L0
p
~ p
~
Λ0
i
−jp~ T φ−~p + φp~ T T ∂φp~ L0 e−S = 0 . (6,8)
Let us now introduce the composite operator:
X   p  h  i
∆≡ k φp~ C(p) + ∂φ−~p L0 ∂γp~ L0 − ∂φ−~p ∂γp~ L0
p
~
Λ0
i
−φp~ T T ∂φp~ L0 , (6,9)

39
or, using the simplified notation introduced in section (2):
X  p   
∆≡ k φp~ C(p)∂γp~ L0 − T T ∂p~ L0
p
~
Λ0
p
  
2
+k ∂γ−~p L0 ∂p~ L0 − ∂γ−~p ∂p~ L0 , (6,10)
Λ0
and define Z Y
∆·Z ≡N dφp~ ∆ e−S (6,11)
p
~

the generator of the Schwinger functions with the insertion of the operator
∆. This is computed in the functional framework adding to the bare action
∆ and taking the derivative of Z with respect to  at  = 0. Remembering
that the support of j is contained in a sphere of radius
 ΛR , which is smaller
p
than Λ0 , we can disregard the cut-off factor k Λ0 in the first term in the
left-hand side of (6,8) which can be written:
X  
jp~ ∂γp~ − T ∂jp~ Z = ∆ · Z . (6,12)
p
~

Now, if the couplings of the source γ are suitably normalized and therefore
the theory has UV limit in the presence of γ (the operator P is finite), from
(6,12) we trivially see that the operator ∆ is finite.
Equation (6,12) is a broken Ward identity, it is sometimes referred to as
”the Quantum Action Principle”. For an introduction to its consequences in
a regularization independent framework we refer to [4],[5].
Of course, the finiteness of ∆ is interesting since it ensures the meaning-
fulness of the broken Ward identity (6,12). However this is not the final goal
of our study which is devoted to the construction of quantum theories with
symmetry properties. Therefore the equation that we have to discuss and
possibly to solve is:
∆ = O(γ) . (6,13)
This is a system of equations for the bare, or better, the normalization con-
stants of the theory that should allow to identify the invariant theory. The
general idea is that the symmetry breaking effect induced by the regular-
ization of the Feynman integral should be compensated fine-tuning non-
invariant terms in the bare action. This is why we shall call (6,13) the
fine-tuning equation.

40
 Until now we have considered on equal footing the possibility of defining
the left-hand side of (6,13) before and after the UV limit provided that the
theory be regular in this limit. The first point to be clarified in our analysis
of this equation is to decide if we are going to study it before or after UV
limit. In the case in which the existence of the UV limit is guaranteed by
simple power counting arguments, as it is in the framework of perturbation
theory, it is natural to consider the situation after this limit. In this case one
has to study, instead of (6,12), the corresponding equation for the effective
action since the bare action is UV-singular.
Let us then discuss the fine-tuning equation after the UV limit. To extend
our results to the effective theory we repeat the previous analysis referring
to the effective lagrangian and we introduce the effective breaking:
X p  
∆ef f ≡ k φp~ C(p)∂γp~ Lef f − T T ∂p~ Lef f
p
~
Λ
p
  
2
+k ∂γ−~p Lef f ∂p~ Lef f − ∂γ−~p ∂p~ Lef f , (6,14)
Λ
Then (6,13) is naturally replaced with:

∆ef f = O(γ) . (6,15)

This fine-tuning condition (6,15) is equivalent to an infinite number of

equations for ∆ef f . However it is apparent from (6,9) that ∆ has limited
dimension provided the field transformations (6,7) change the field dimension
by a limited amount. More precisely, in the limit Λ0 → ∞, the bare operator
∆ becomes a local operator of dimension 3 + dP .
Therefore, according to the remark at the beginning of section 5 con-
cerning composite operators, ∆ is completely equivalent to a strictly local
operator whose dimension is equal to 4 plus the increase of field dimension
in the transformations (6,4).
Furthermore, computing explicitly the Λ-derivative of the right-hand side
of (6,14) and taking into account (3,2) it can be seen that ∆ef f satisfies, before
the UV limit, the evolution equation:

Λ∂Λ ∆ef f = M [∆ef f ] , (6,16)

if the theory is regular in the UV limit (6,16 ) is expected to hold true

also after this limit. Therefore we can conclude that ∆ef f is determined

41
by its normalization conditions and hence the vanishing condition for ∆ef f
is equivalent to a finite number of equations concerning the normalization
conditions for ∆ef f .
Let us notice that (6,16 ) can be verified directly using the evolution
equation for Lef f . Indeed one can compute the left-hand side of (6,16 )
according:
q
  
2
X
Λ∂Λ ∆ef f = Λ∂Λ k ∂γq~ Lef f ∂−~qLef f − ∂γq~ ∂−~qLef f
q~
Λ
 
2 q
p Λ∂Λ k
  h
XX
2 Λ
k ∂γp~ ∂q~Lef f ∂−~qLef f ∂−~p Lef f
p
~ q~
Λ C(q)
1 
+∂γp~ Lef f ∂−~qLef f ∂−~p ∂−~qLef f − ∂γp~ ∂q~∂−~qLef f ∂−~p Lef f + ∂q~∂−~q∂−~p Lef f ∂γp~ Lef f
2
1

−∂−~p ∂γp~ ∂q~Lef f ∂−~qLef f − ∂q~∂γp~ Lef f ∂−~p ∂−~qLef f + ∂−~p ∂γp~ ∂−~q∂q~Lef f
 
2
2 q 
p Λ∂Λ k Λ 1
XX    
k φp~ C(p) ∂γp~ ∂q~Lef f ∂−~qLef f − ∂γp~ ∂q~∂−~qLef f
p
~ q~
Λ C(q) 2
1
 
T
−φp~ T ∂p~ ∂q~Lef f ∂−~qLef f − ∂p~ ∂q~∂−~qLef f , (6,17)
2
and, taking into account the definition of M given in (3,13), the right-hand
side is computed according:
 
2 q
p Λ∂Λ k Λ h
 
k2
XX
M [∆ef f ] = ∂γp~ ∂q~Lef f ∂−~p Lef f ∂−~qLef f
p
~ q~
Λ C(q)
+∂γp~ Lef f ∂−~p ∂q~Lef f ∂−~qLef f − ∂γp~ ∂−~p ∂q~Lef f ∂−~qLef f − ∂γp~ ∂−~qLef f ∂−~p ∂q~Lef f
1 
− ∂q~∂−~q∂γp~ Lef f ∂−~p Lef f + ∂q~∂−~q∂−~p Lef f ∂γp~ Lef f − ∂γp~ ∂−~p ∂q~∂−~qLef f
2  
q
X X  p  Λ∂Λ k 2 Λ h  
+ k φp~ C(p) ∂γp~ ∂q~Lef f ∂−~qLef f − ∂γp~ ∂q~∂−~qLef f
p
~ q~
Λ C(q)
i
−φp~ T T (∂p~ ∂−~qLef f ∂q~Lef f + ∂p~ ∂q~∂−~qLef f )
 
q
X Λ∂Λ k 2 Λ
h  
C(q) ∂γq~ Lef f ∂−~qLef f − ∂γq~ ∂−~qLef f
q~
C(q)
−T (∂q~Lef f ∂−~qLef f − ∂q~∂−~qLef f )] . (6,18)

42
In (6,18) we have left to the reader the bookkeeping of the field indices. If
this is correctly made it is apparent that the last line of (6,18) vanishes due
to (6,5), in which one has to replace C with C −1 , and that (6,17) and (6,18)
coincide, thus confirming (6,16 ).
Therefore, after the general discussion concluding section 4, we can say
that (6,15) is guaranteed in the UV limit if ∆ef f satisfies vanishing normal-
ization conditions.
At this point of the analysis it is convenient to perform the infinite volume
limit. As discussed in section 3 this limit is completely harmless, at least
in perturbation theory. Therefore, from now on, we understand that all
the sums over the momenta in the previous formulae are replaced with the
corresponding integrals.
Now we can express the vanishing of these normalization conditions in
a simple form introducing Td F : the local approximant of dimension d to
a translation invariant functional F . This approximant is defined as the
integrated local functional of dimension d of fields and sources whose vertices
coincide with those generated by F up to terms with momentum degree equal
to d minus the total dimension of the fields and sources corresponding to
the external legs of the vertex. Here, as usual, we assign the source of a
local operator the complement to four of the dimension of the operator. For
example if F depends only on φi we have:
"
Z
4 δ 1 δ δ
T3 F = F [0] + d xφi (x) F + φj (x) F
δφi (0) 2 δ φ̃j (0) δφi (0)
i δ δ
+ ∂µ φj (x)∂pµ F+
2 δ φ̃j (0) δφi (0)
δ2
#
1 δ
φj (x)φk (x) F |φ=0 , (6,19)
6 δ φ̃j (0)δ φ̃k (0) δφi (0)

where we have set Z

φ(x) = d4 pe−ipx φ̃(p) . (6,20)
Given a local operator O with source η and dimension dO , this definition
automatically induces that of the effective local approximant of dimension d
to O, according:
δLef f
Td Oef f ≡ Td . (6,21)
δη

43
In perturbation theory the effetive local approximants to a local operator
have a very special property. Let us introduce the perturbative order ωO of a
local operator, as the first order in h̄ at which the operator contributes to the
Green functional generator Z. In general this order appears as the h̄-power
of a factor in front of the bare operator.
The above mentioned special property is the following: all the effective
local approximants to O with dimension higher or equal to dO coincide up
to terms of order h̄ωO +1 That is:
 
if d ≥ dO → Td Oef f = TdO Oef f + O h̄ωO +1 . (6,22)
This property is easily understood recalling the interpretation of Lef f ,
given at the beginning of section 4, as the functional generator of the con-
nected and amputated Feynman amplitudes corresponding to the propagator:
   
p p
k2 Λ0
− k2 Λ
, (6,23)
p2 + m2
and the vertices correspnding to L0 . Then the contributions of order h̄ωO to
the vertices generated by Oef f are given by the tree approximation connected
and amputated amplitudes and the same holds true for their momentum
derivatives. Owing to the fact that the above propagator vanished with all
its momentum derivatives for vanishing momentum we see that the only
contribution of order h̄ωO to Td Oef f come from connected and amputated
tree diagrams without internal lines, that is those coinciding with a single
vertex. These vertices are determined by the bare structure of the operator
and their local approximant do not depend on d provided d ≥ dO . This
proves the mentioned property.
In the case of Td∆ef f ∆ef f and of T4 Sef f we shall, from now on, understand
the dimension of the local approximant.
Since the couplings appearing in T ∆ef f are in one-to-one correspondence
with the normalization parameters of ∆ef f it is now clear that if the theory
is finite in the UV limit, the fine tuning equation (6,15) is equivalent to
T ∆ef f = O(γ) , (6,24)
that is to:
"
Z
4 δLef f δLef f δLef f
T d p φ(p)C(p) +
δγ(p) δφ(p) δγ(−p)

44
 δ 2 Lef f
#
δLef f p
 
T
−φ(p)T − h̄k = O(γ) , (6,25)
δφ(p) Λ δφ(p)δγ(−p)

where we have taken into account that k(0) = 1 and that all the partial
derivatives of kvanish
 at the origin. We have also considered that Lef f is a
p
functional of k Λ φ(p) and γ(p). Notice that in (6,25) we have introduced
in the last term the h̄ coefficient that appears in a loop-orderd perturbation
theory in which one introduces h̄ as loop counting parameter.
For future purposes it is convenient to introduce here two further fuction-
als:
" #
Z
4 φ(−p)C(p)φ(p)
Sef f ≡ dp − γ(−p)T φ(p) + Lef f , (6,26)
2

and
p δ2 −Sef f
Sef f
Z  
2 4
Def f ≡ −h̄ e d pk
h̄ e h̄
Λ δφ(p)δγ(−p)
δ 2 Sef f
 " #
Z
4 p δSef f δSef f
d pk − h̄
Λ δφ(p) δγ(−p) δφ(p)δγ(−p)
 " ! !
Z
4 p δLef f T δLef f
= d pk − φ(p)T C(p)φ(−p) +
Λ δγ(−p) δγ(p)
2
#
δ Sef f
−h̄ + O0 (γ) , (6,27)
δφ(p)δγ(−p)

It is apparent that:
T ∆ef f = T Def f + O0 (γ) , (6,28)
and hence the fine tuning equation can be written and discussed in terms
of Def f instead of ∆ef f . We shall present the further developments of our
method in the next section applying them directly to a non-abelian gauge
theory.
Beyond the perturbative level the existence of an UV limit is far from
being guaranteed. However there is some evidence of a rather general con-
nection between the existence of this limit and asymptotic freedom [10]. In
this case the study of (6,13) or (6,15) must precede the UV limit since the
theory with broken symmetry in general is not expected to be asymptotically
free.

45
 Notice however that in the presence of an UV cut-off Λ0 the wanted
symmetry can be unreachable at the bare level. This happens typically in a
gauge theory in which (6,13) is unsolvable for finite UV cut-off.
Therefore at first sight the construction of a fully quantized gauge theory
seems impossible. However in the framework of the effective theory it might
be possible to fine-tune the low energy parameters under the condition that
∆ef f given in (6,14) satisfy normalization conditions vanishing with ΛΛ0 .
If this condition is met, considering the solution of the evolution equations
discussed in section 4, we see that the breaking is irrelevant and hence, if the
theory has a regular UV limit, ∆ vanishes in this limit at least as ΛΛ0 . To
satisfy this last condition one should ask the fine-tuned theory to be asymp-
totically free. Postponing the dream of the non-perturbative construction of
a gauge theory to better times, we shall limit our present discussion to the
perturbative situation.

7. Analysis of the SU (2) Yang-Mills model

To proceed further with the analysis of (6,15) it is convenient to refer

to the highly non-trivial example of a pure SU (2) gauge theory, whose field
content consists of the gauge field Aaµ , of the Lagrange multiplier ba and of the
anticommuting Faddeev-Popov ghosts ca and c̄a . All the fields are isotopic
vectors. In terms of the gauge field strength:

Gaµν = ∂µ Aaν − ∂ν Aaµ − gabc Abµ Acν , (7,1)

where abc is the antisymmetric Ricci tensor, and of the covariant derivative:

Dµ ca = ∂µ ca − gabc Abµ cc , (7,2)

the ”classical action” of the theory is given:

ba ba
" #
Z
4 z a a
Scl = dx Gµν Gµν + + iba ∂µ Aaµ + ∂µ c̄a Dµ ca . (7,3)
4 2

46
From this equation one can easily deduce the covariance matrix of the fields.
It is also possible to single out the symmetries characterizing the theory at
the quantum level. It turns out that, given the ba -dependent part of (7,3),
Scl is identified by the invariance condition under the non-linear, nilpotent
field transformations:
Aaµ → Aaµ + ηDµ ca , (7,4)
g
ca → ca + η abc cb cc , (7,5)
2
c̄a → c̄a + iηba , (7,6)
ba → ba . (7,7)
Notice that here η is an anticommuting constant that replaces  in the defi-
nition of ∆.
A further necessary comment concerning (7,3), is that the field ba plays
the role of an auxiliary field since the action is at most quadratic in it and
hence it does not appear into the interaction lagrangian. It follows that the
Wilson renormalization group starting from a bare action with this property
will generate a b-field independent effective lagrangian.
Owing to the anticommuting character of the transformations (7,4- 7,7),
quantizing the theory, we have to introduce an anticommuting vector-isovector
source γµa of dimension two, coupled to −gγµa abc Abµ cc and a commuting isovec-
tor ζ a of dimension two, coupled to the variation of ca . Adding to the action
these source terms together with the sources of the fields we have:

ba ba
"
Z
4 z a a
Scl = dx Gµν Gµν + α + iba ∂µ Aaµ + ∂µ c̄a Dµ ca
4 2
abc
#
a abc b c a g b c a a a A a a ¯ a a
−gγµ  Aµ c − ζ c c − jµ Aµ − J b − ξ c̄ − ξ c . (7,8)
2

Notice that ξ a and ξ¯a are anticommuting sources and that there is a conserved
ghost charge that vanishes for the vector field, is equal to +1 for the c field,
−1 for the c̄ and γ source and −2 for the ζ source.
In order to quantize this theory we introduce a cut-off in much the same
way as for the scalar field, choosing the same function k for all the field
components. In this way we maintain at the quantum level the invariance
of the theory under rigid isotopic symmetry. however this automatically

47
introduces an unavoidable breaking into the Ward identity corresponding to
(7,4-7,7). Indeed, considering the cut-off transformations:
p
   
Aaµ~p → Aaµ~p + η k ∂γµ~
a L0 + ipµ c
a
−~p , (7,9)
Λ0 p

p
 
cap~ → cap~ + ηk ∂ζp~a L0 , (7,10)
Λ0
c̄ap~ → c̄ap~ + iηbap~ , (7,11)
and following the analysis described above, we get the broken Ward identity:
X   
a
jµ~
p ∂γµ~
a + ipµ ∂ξ a
p p
~
− ξ¯p~a ∂ζp~a − iξp~a ∂Jp~a Z ≡ SZ = ∆ · Z , (7,12)
p
~

where:

∆ =
X    p  
2
(δµν p − pµ pν )Aaν~p + pµ bap~ + ∂Aaµ−~p L0 k ∂γµ~
a L0 + ipµ ca−~p −
p
~
Λ0 p

p
     
p2 c̄ap~ − ∂ca−~p L0 k ∂ζp~a L0 − i p2 cap~ + ∂c̄a−~p L0 ba−~p −
Λ0
p 

  
k ∂Aµ−~p ∂γµ~p + ∂c−~p ∂ζp~ L0 ,
a a a a (7,13)
Λ0
Substituting the classical lagrangian into this equation and disregarding the
cut-off factors, one finds that at the classical limit ∆ is equal to:

∆cl = −igabc a b a
XX X
pµ γµ~
p cq~ c−~ p = −i
q −~ pµ ∂ζp~a Scl . (7,14)
p
~ q~ p
~

For a generic choice of the bare action it is apparent that this breaking
is b-dependent. Now this is a true disaster since, even if the breaking were
compensable by a suitable modification of the action, this would introduce
b-dependent interaction terms. However this difficulty can be avoided if one
chooses the bare action so that:

−ipµ ∂γµ~
a L0 = ∂K
p
a L0 .
Λ0 c̄p
~
(7,15)

Indeed substituting (7,15) into (7,13) it is apparent that the b-dependent

terms in ∆ annul each other.

48
 Now, assuming that the theory remain finite in the UV limit, we translate
the above results in terms of the effective theory in the infinite volume limit.
We have in particular that the constraint (7,15) remains true if we replace
in them L0 with Lef f . That is

−ipµ ∂γµ~
a Lef f = ∂K c̄a Lef f .
p Λ p~
(7,16)

Indeed it is apparent that both sides of (7,16) satisfy the same linear evolution
equation. In particular for the left-hand side this equation reads:
h i
Λ∂Λ ∂γµ~
a Lef f = M ∂γ a Lef f
p µ~p
. (7,17)

Taking into account (7,14) we assume for ∆ef f the following fine tuning
condition: Z
δT Lef f
T ∆ef f = −i d4 ppµ γµa (p) a . (7,18)
δζ (p)
Now, following the line shown in the previous section we introduce the
effective action:
δµν p2 − pµ pν a
Z "
a
Sef f [φ, γ] = d p Ãaµ (−p)
4
Ãν (p) + ¯c̃ (−p)p2 c̃a (p)−
2
i
iγ̃µa (−p)pµ c̃a (p) + Lef f , (7,19)

and the modified breaking Def f defined in (6,27) that in the present case is
written: in terms of it:
 
p  δ δ δ δ −Sef f
Sef f
Z  
Def f = −h̄2 e h̄ d4 pk +  e h̄ =
Λ δ Ãaµ (p) δγ̃µa (−p) δc̃(p)a δ ζ̃(−p)a

p  δ δ δ δ
Z  
d4 pk Sef f a Sef f + a
Sef f Sef f −
Λ a
δ õ (p) δγ̃µ (−p) δc̃(p) δ ζ̃(−p)a
  
δ δ δ δ
h̄  a a
+ a
 Sef f  . (7,20)
δ õ (p) δγ̃µ (−p) δc̃(p) δ ζ̃(−p)a

Comparing the true breaking (6,14) with the modified one given above
we see that: Z
δT Lef f
T Def f = T ∆ef f + i d4 ppµ γµa (p) a , (7,21)
δζ (p)

49
and hence we are left with the homogeneous fine-tuning condition:

T Def f =
   
p  δ δ δ δ −Sef f
Sef f
Z  
−h̄2 T5 e h̄ d4 pk +  e h̄  =
Λ δ Ãaµ (p) δγ̃µa (−p) δc̃(p)a δ ζ̃(−p)a
  
Z
δ
δ δ δ
d4 p T5  a Sef f a Sef f + a
Sef f Sef f  −
δ õ (p) δγ̃µ (−p) δc̃(p) δ ζ̃(−p)a
  
p  δ δ δ δ
 
h̄T5 k a
+ a
 Sef f  = 0 . (7,22)
Λ δ õ (p) δγ̃µ (−p) δc̃(p) δ ζ̃(−p)a
a

Notice that the cut-off factor in the first two terms under integral has dis-
appeared since, by momentum conservation, the T operator restricts the p
variable corresponding to these terms to zero.
It is important to notice here that, as it will be apparent in a moment, we
can choose this homogeneous condition due to the nilpotency of the classical
transformations (7,4-7,7).
In the present case the differential operator
p δ δ
Z  
4
d pk ≡
Λ δφ(−p) δγ(p)
 
p  δ δ δ δ
Z  
4
d pk a
+ a
 , (7,23)
Λ δ õ (p) δγ̃µ (−p) δc̃(p) δ ζ̃(−p)a
a

is nilpotent owing to the anticommuting nature of the ghost field ca and of

the source γµa .
This nilpotency of the differential operator (7,23) leads to the following
equation for Def f :

p δ δ −Sef f
Z  
4
−h̄ d pk Def f e h̄ =
Λ δφ(−p) δγ(p)
 "
Z
p δ δ δ δ
d4 pk Sef f Def f + Def f Sef f −
Λ δφ(−p) δγ(p) δφ(−p) δγ(p)
#
δ δ −Sef f
h̄ Def f e h̄ = 0 , (7,24)
δφ(−p) δγ(p)

50
where we have taken into account (7,20) and the anticommuting character
of Def f .
The last identity, that in the gauge case is written:

p  δ δ δ δ
Z  
d4 pk S ef f a
D ef f + a
S ef f Def f +
Λ δ Ãaµ (p) δγ̃µ (−p) δγ̃µ (−p) δ Ãaµ (p)
δ δ δ δ
a
Sef f Def f + Sef f Def f −
δc̃(p) δ ζ̃(−p)a δ ζ̃(−p)a δc̃(p)a
  
δ δ δ δ
h̄  a +  Def f  = 0 , (7,25)
δ õ (p) δγ̃µa (−p) δc̃(p)a δ ζ̃(−p)a

plays the role of a consistency condition for the breaking analogous to the
Wess-Zumino [11] consistency condition for the chiral anomaly.
Now we show how in the perturbative case this consistency condition
turns out to be sufficient to prove the solvability of the fine-tuning equation
(6,15).
As repeated many times in the previous sections, perturbation theory is
built by a recursive procedure in h̄. Hence we have to apply this procedure to
the fine-tuning equation. Therefore we develop the effective action Sef f and
the effective breaking Def f in power series of h̄ and we label the k th -order
terms of these series by the index k.
Now, if we assume the tree approximation normalization condition
gabc b c
" #
Z
1 a a  
T Sef f 0 = d4 x Gµν Gµν + ∂µ c̄a − γµa Dµ ca − ζ a cc , (7,26)
4 2
we have

Z
δ δ
T Def f 0 = d4 p  T Sef f 0 T Sef f 0 +
δ Ãaµ (p) δγ̃µa (−p)
!
δ δ
a
T Sef f 0 T Sef f 0 = 0 . (7,27)
δc̃(p) δ ζ̃(−p)a
Indeed one can directly verify using the fact that Sef f 0 is invariant under the
transformations (7,4-7,5).
Assuming that
T Def f k = 0 for k < n , (7,28)

51
it follows that Def f k = 0 and hence the consistency condition

p  δ δ δ δ
Z  
d4 pk Sef f 0 a Def f n + a Sef f 0 a Def f n +
Λ a
δ õ (p) δγ̃µ (−p) δγ̃µ (−p) δ õ (p)
#
δ δ δ δ
a
Sef f 0 Def f n + Sef f 0 Def f n = 0 . (7,29)
δc̃(p) δ ζ̃(−p)a δ ζ̃(−p)a δc̃(p)a

Selecting the local approximant of dimension 6 to (7,29) and taking into

account (7,26) and (7,32) we get the perturbative consistency condition:

Z
δ δ δ δ
d4 p  T Sef f 0 T Def f n + T S ef f 0 T Def f n +
δ Ãaµ (p) δγ̃µa (−p) δγ̃µa (−p) δ Ãaµ (p)
#
δ δ δ δ
a
T Sef f 0 T Def f n + T Sef f 0 T Def f n ≡
δc̃(p) δ ζ̃(−p)a δ ζ̃(−p)a δc̃(p)a
DT Def f n = 0 . (7,30)

Here we have used (6,22). From (7,26) we have the explicit form of D:

gabc b c δ
"
Z
δ
D=− d x Dµ ca
4
− c c a+
δAaµ 2 δc
#
  δ δ
Dν Gaνµ + gabc γµc cc + Dµ γµa . (7,31)
δγµa δζ a

It is important to notice here that the differential operator D defined

above is nilpotent. Therefore any element of the image of D solves (7,30).
At this point of our analysis it is convenient to write explicitly the local
approximant of dimension 5 to Def f n . From now on we shall put c̄ = 0
without any loss of generality since the dependence on this field is strictly
connected to that on the source γµa . Owing to the euclidean, isotopic (SU (2))
and translation invariance and recalling that Def f has ghost charge +1, we
have:
Z h
T Def f n = d4 x an abc γµa cb ∂µ cc + bn γµa Abµ ca cb +
cn Aaµ ∂ 2 ∂µ ca + dn Aaµ ∂µ ca +
en abc Aaµ ∂ν Abµ ∂ν cc + fn abc Aaµ ∂µ Abν ∂ν cc +

52
 gn Aaµ Aaµ ∂ν Abν cb + hn Aaµ Abµ ∂ν Aaν cb +
kn Aaµ Aaν ∂µ Abν cb + jn Aaµ Abν ∂µ Aaν cb +
i
ln Aaµ Abν ∂µ Abν ca . (7,32)
Thereby we see that this functional does not contain monomials of dimension
lower than 3.
Inserting into (7,30) (7,26) and (7,32) one finds, after some algebraic
manipulations, the following constraints:
bn = −gan , fn + en + gcn = 0
hn = kn = jn , ln = 2gn , (7,33)
therefore the general solution of the perturbative consistency condition con-
tains six free parameters.
Now, given the most general local approximant to Sef f n satisfying (7,15)
is:
gabc b c
" !
Z
4 z1 a a
T Sef f n = dx Gµν Gµν − z2 γµa Dµ ca + ζ a cc
4 2
 
+z3 Dµ Aaν Gaµν + γµa ∂µ ca + x1 abc γµa Abµ cc + x2 ∂µ Aaµ ∂ν Aaν
 2 
+x3 Aaµ Aaµ + x4  abc
Aaµ Abν ∂µ Acν + x5 Aaµ Abµ Aaν Abν + x6 Aaµ Aaµ .(7,34)

The first three terms in the right-hand side are annihilated by D; this is
obviously true for the first term that depends only on the gauge field and is
gauge invariant; it is also obvious for the second term that added to the first
one gives the action (7,26). The third term is obtained from the action intro-
ducing the scaling transformation: Aµ → ZAµ , γµ → Z −1 γµ and taking the
Z-derivative for Z = 1; it is annihilated by D since the scaling transforma-
tion leaves the differential operator (7,23) invariant. It is a little lengthy but
straightforward to verify that T Def f n in (7,30) with the constraints (7,33)
satisfies:
T Def f n = DT S¯ef f n , (7,35)
if the coefficients of T S¯ef f are chosen according
n

an = −x̄1 , cn = −2x̄2
dn = 2x̄3 , en = x̄4
gn = −x̄4 − 4x̄6 , hn = x̄4 − 4x̄5 . (7,36)

53
Let us now consider the structure of T Def f at the nth order in h̄ as it can be
computed from (7,22). This is written:

T Def f n =

Z
δ δ δ δ
d4 p  T Sef f 0 T Sef f n + T Sef f 0 T Sef f n +
δ Ãaµ (p) δγ̃µa (−p) δγ̃µa (−p) δ Ãaµ (p)
δ δ δ δ
a
T Sef f 0 T Sef f n + T Sef f 0 T Sef f n +
δc̃(p) δ ζ̃(−p)a δ ζ̃(−p)a δc̃(p)a
 
n−1
X δ δ δ δ
T5  a Sef f k a Sef f n−k + S ef f k Sef f n−k  −
k=1 δ õ (p) δγ̃µ (−p) δc̃(p)a δ ζ̃(−p)a
  
p  δ δ δ δ
 
h̄T5 k +  Sef f
n−1
 . (7,37)
Λ δ Ãaµ (p) δγ̃µa (−p) δc̃(p)a δ ζ̃(−p)a

On account of (7,35), (7,37) can be written:

DT Sef f n +
 
n−1
Z
δ δ
d4 p 
X
T5  Sef f k Sef f n−k +
k=1 δ Ãaµ (p) δγ̃µa (−p)
!
δ δ
S ef f k Sef f n−k −
δc̃(p)a δ ζ̃(−p)a
  
p  δ δ δ δ
 
h̄T5 k + n−1 ≡
 Sef f 
Λ a a
δ õ (p) δγ̃µ (−p) δc̃(p) δ ζ̃(−p)a
a

DT Sef f n + Ωn = DT S¯ef f n , (7,38)

from which we see that there exists a 4-dimensional integrated local func-
tional Σn = T S¯ef f n − T Sef f n such that

Ωn = DΣn . (7,39)

Taking into account this result and (7,37) the nth order fine-tuning equation
is written:  
D T Sef f n + Σn = 0 , (7,40)
and it is solved by
T Sef f n = −Σn . (7,41)

54
This proves the iterative solvability of the fine-tuning equation.

8. Further comments.

As it is apparent from the above analysis the crucial step of this proof
has been the fact that the consistency condition can be written in the form
DT Def f n = 0 , (8,1)
and the result that the general solution of this equation is
T Def f n = DT S¯ef f n . (8,2)
In the framework of a more general gauge theory one can always reduce
the perturbative problem to the study of an equation completely analogous
to (8,1) and in particular to the comparison of the kernel of the nilpotent
operator with its image. In this way the solvability of the fine-tuning equation
is in general reduced to the triviality of a certain class of the so-called BRS
cohomology. This is exactly what we have verified in the present example.
Of course, our proof is strongly dependent on the chosen perturbative
framework. However the idea of making the low-energy vertices of the break-
ing to vanish by a fine tuning of the parameters of the theory could well work
beyond perturbation theory.
The first question to answer in this direction is how this possibility is
related to the linear consistency condition (7,24). In particular it can be
interesting to notice that, if we write (7,24) in the form:
D̂Def f = 0 , (8,3)
the differential operator D̂ is not nilpotent. Indeed in the case of anticom-
muting γ and commuting φ one has (h̄ = 1)
" #
Z
δ δ δ δ
D̂2 = d4 p Def f − Def f . (8,4)
δγ(p) δφ(−p) δφ(−p) δγ(p)
Here we have assumed φ commuting and γ anticommuting.
Thus the nilpotency of the perturbative coboundary operator D is vio-
lated beyond the pertubative level by terms of order Def f .

55
References
[1] J. Polchinski, Nucl. Phys. B231 (1984) 269.
G. Gallavotti, Rev. Mod. Phys. 57 (1985) 471.
[2] K. Wilson, Phys. Rev. B4 (1971) 3174.
K. Wilson and J. Kogut, Phys. Reports 12 (1974) 75
[3] W. Zimmermann, Local Operator Products and Renormalization, in:
Lectures on Elementary Particles and Fields. Eds. S. Deser, M. Grisaru,
H. Pendleton (MIT Press. Cambridge, Mass 1970).
[4] C. Becchi, An introduction to the renormalization of theories with con-
tinuous symmetries, to the chiral models and to their anomalies, in:
Fields and Particles, Proceedings of the XXIX Int. Universitästwochen
für Kernphysik, Schladming, Austria 1990, Eds. H. Mitter and W.
Schweiger (Springer-Verlag Berlin Heidelberg 1991).
[5] C. Becchi, Lectures on the renormalization of gauge theories, in: Rel-
ativity, Groups and Topology II (les Houches 1983) Eds. B.S. DeWitt
and R. Stora (Elsevier Science Pub.B.V. 1984).
[6] K. Symanzik, Euclidean Quantum Field Theory, in: ”Local Quantum
Theory” (Varenna 1968) Ed. R. Jost (Academic Press, New York 1969).
[7] see e. g.: J.Glimm and A.Jaffe, Quantum Physics, Springer-Verlag,
Berlin Heidelberg 1981.
[8] J. Feldman, J. Magnen, V. Rivasseau and R. Sènèor, Commun. Math.
Phys. 98 (1985) 273.
K. Gawedzki, A. Kupiainen, Commun. Math. Phys 99 (1985) 197
[9] J. Hugues and J. Liu, Nucl. Phys. B307 (1988) 183.
[10] K. Gawedzki, A. Kupiainen, Asymptotic freedom beyond perturbation
theory, in: Critical Phenomena, Random Systems, Gauge Theories (les
Houches 1984) Eds. K. Osterwalder and R. Stora (Elsevier Science
Pub.B.V. 1985).
[11] J. Wess and B. Zumino, Phys. Letters 37B (1971) 95.

56