Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

Pol Chin Ski 1984

Download as pdf or txt
Download as pdf or txt
You are on page 1of 27

Nuclear Physics B231 (lq84) 269-295

' Norlh-tlolland Publishing Company

R E N O R M A L I Z A T I O N A N D EFFECTIVE L A G R A N G I A N S

Joseph POLCHINSKI*

l.vman l,ahoratorv ~!/"Phvs,w. Ilarcard Unn'er.~itv. ('amhrtdge. .'Was,~adntsett,s 0213b¢. USA

Received 27 April 1983

There is a strong intuitive understanding of renormalization, due to Wilson, in terms of the


scaling of effective lagrangians. We show that this can be made the basis for a proof of
perturbative renormalization. We first study rcnormalizabilit,.: in the language of renormalization
group flows for a toy renormalization group equation. We then derive an exact renormalization
group equation for a four-dimensional X4,4 theorv with a momentum cutoff. Wc organize the cutoff
dependence of the effective lagrangian into relevant and irrelevant parts, and derive a linear
equation for the irrelevant part. A length} but straightforward argument establishes that the piece
identified as irrelevant actually is so in perturbation theory. This implies renormalizabilitv The
method extends immediately to an}' system in which a momentum-space cutoff can bc used. but
the principle is more general and should apply for any physical cutoff. Neither Weinberg's theorem
nor arguments based on the topology of graphs are needed.

I. Introduction

The understanding of renormalization has advanced greatly in the past two


decades. Originally it was just a means of removing infinities from perturbative
calculations. The question of why nature should be described by a renormalizable
theory was not addressed. These were simply the only theories in which calculations
could be done.
A great improvement comes when one takes seriously the idea of a physical cutoff
at a very large energy scale A. The theory at energies above A could be another field
theory, a lattice, spacetime foam, or anything else. The theory just below A should
be represented by a very general lagrangian in which the various terms have
coefficients of the order of A to the appropriate power to make the dimensions
correct. Consider the physics at an energy E far below A. The non-renormalizable
terms, those with coefficients of A to negative powers, typically give contributions
that are suppressed by powers of A. This is true unless the non-renormalizable term
is embedded in a Feynman graph sufficiently divergent to make up for the small
coefficient. Power counting shows that the only n-point functions sufficiently
divergent are those which would be divergent even if they contained only renormal-

* Research was supported in part by the National Science Foundation under grant no. PHY82-15249.

269
270 J. Polchmski / Effecttt,e lagr'angians

izable interactions. We know, however, that the latter divergences can be reabsorbed
in redefinitions of the renormalizable couplings. Thus, to accuracy E / A the entire
effect of the non-renormalizable terms can be absorbed in the initial values of the
renormalizable ones, and all quantities can be calculated in the resulting effective
field theory with renormalizable interactions only.
This idea is very nice, but still perturbative and graphical in nature. Also. it does
not attempt to address the question of why arguments based on naive power
counting are correct, that is, why renormalization actually works. A further improve-
ment comes with the idea of smoothly lowering the cutoff. As this is done. the
effective lagrangian changes. The effective lagrangian at lower scales is given in
tcrms of its form at a given scale, and its change with scale is governed by a scaling
or renormalization group equation. Typically, as we scale down to smaller momcnta,
the lagrangian converges toward a finite-dimensional submanifold in the space of
possible lagrangians. That is, the scaling transformation has only a finitc n u m b e r of
non-negative eigenvalues, with deviations in the orthogonal directions d a m p e d as wc
travel to low momenta. These orthogonal directions are therefore termed "irrelevant".
In the zero-coupling limit, the negative, irrelevant, eigenvalues correspond to pre-
cisely the non-renormalizable interactions. Since there is nothing discontinuous
about the scaling transformation as the couplings are changed, those eigenvalues
which were negative at zero coupling should rcmain negative at sufficiently small
coupling. This is equivalent to renormalizability, a connection which will be devel-
oped further in sect. 2. This understanding of renormalization is due primarily to
Wilson [l ]. Ref. [2] lists an assortment of discussions from related points of view.
The classic proofs of renormalization in perturbation theory [3-5] are based on
the old idea of removing infinities. They involve detailed graphical arguments and
convergence theorems [6] that are rather far removed from the present intuitive
picture*. The understanding in terms of renormalization group flows is so com-
pelling that one must wonder whether it can be justified only by appealing to the
existing proofs. In fact, we will find that this is not so. Once we learn to discuss
renormalization in the language of renormalization group flows, a proof follows in a
straightforward way. We are concerned here only with perturbation theory, but the
proof follows the outline we would expect for a non-perturbative argument.
In sect. 2 we study a toy renormalization group equation, showing how to divide
the effective action into relevant** and irrelevant parts. We show that the toy
equation describes a renormalizable theory. In sect. 3 we make concrete the idea of a

" We should mention in particular the proof of Callan and of Blaer and Young [51. which uses the
renormalization group to simplify greatly the graphical arguments. This is still essentially a graphical
proof, as it is based on a skeleton expansion, and is global rather than local in momenta, as it uses
Weinberg's theorem [6].
*" In standard usage what we call "relevant" is usually divided further into "relevant" and "marginal".
Throughout this paper, "relevant" should be understoc,,d to mean '" relevant or marginal".
J. I'ohhmskt / l:'ffectu'e lagrangta,t.s 271

differential scaling of a cutoff, obtaining an exact rcnormalization group equation


for a ),04 theory with a momentum-space cutoff. Guided by the results of sect. 2 we
organize the action into relevant and irrelevant parts. Very simple bounds on
momentum integrals are sufficient to prove that in perturbation theory, the terms
identified as irrelevant really are. The detailed order-by-order argument is given in
sect. 4. It is then a short step to show that correlation functions depend on the cutoff
only through inverse powers.
The method extends automatically to any theory in which a momentum cutoff can
be used, and to composite opcrator renormalization. The idea is far more general,
however, and should apply for any physical cutoff. In the conclusions, we discuss the
extension to gauge theories and beyond perturbation theory.
The exact renormalization group equation for ?~04, as well as the understanding of
renormalization in terms of relevant and irrelevant operators, can all be found in the
work of Wilson [1]. Our contribution here is to note that these ideas can be used to
give a self-contained proof of renormalization in perturbation theory.
It might cause some confusion that the 2~0'~ theory which we will be studying does
not actually have a continuum limit [7], in the sense that the bare coupling ~,°(A.)
diverges when the cutoff A o is still finite. The point is that ~.°(A~) considered as a
formal power series in the renormalized coupling )~r is perfectly well defined for all
finite 3. o, as are all other quantities, and our results are concerned strictly with this
expansion. It should be emphasized that two distinct properties of the renormaliza-
tion group flow are involved here. One is purely local: is the flow converging in all
but a finite number of directions? This is the crux of our present work. The second
concerns global properties of the flow: is there a starting point, for a given A.,
which gives the desired value of )~r at some fixed lower scale? The second question is
an interesting one, but the answer is always " y e s " in perturbation theory.

2. Renormalization group flows

In this section we study a toy renormalization group equation which describes one
relevant and one irrelevant coupling. This section is not needed for the proof in sect.
3, but is intended to illustrate the ideas in a simpler setting. The two couplings arc
g4, which is dimensionless, and g6, which has dimensions of (mass) 2. These have
been chosen to correspond dimensionally to a 4-point and a 6-point function in the
effective action of a 4-dimensional scalar field theory. These couplings depend on
the scale A according to

dg4
A - ~ =/34 ( ,I.~4,,"~296), ( 1a)

• dg6
A ~- = A 2136( g , , A-g 6), (1 b)
272 J. Poh'hinski / lffectwe lagrangtans

where factors of A have been inserted by dimensional analysis. Define the dimen-
sionless variables ~.4 = g4, )k6 = AZg6, so that

d)k 4
A~- = B4(~.4, ~'6), (2a)

d~. 6
A~- - 2~.~ = ,86 (X 4, ~.6). (2b)

Take a particular solution of (2), (X 4, X6), and consider small deviations e, = X i - ~,.
Then, to order e,

d E4 OqB4 O/~ 4
A dA 03~ 4 E4"{" ~ - 6 g6' (3a)

d/~ 6 a/~6 Oq/~6


A~-~- - 2 t 6 = - ~ 4 e 4 + ~ - 6 t6, (38)

where the bar means the quantity is evaluated at (~.4, X6). The term - 2 e 6 in (3b)
suggests that deviations in ~.6 are strongly d a m p e d (by O(A2/A2)) as we evolve from
a scale At) to a much lower scale A. However, before a deviation in X6 is d a m p e d
away, it will, through (3a), cause X 4 to run a little faster or slower than it otherwise
would have done. (Compare the discussion in the second paragraph of the introduc-
tion of the low-energy effects of non-renormalizable terms: ~.~, enters only through
the effective value of ~4). Fig. 1 shows the situation in the ~4 - ~,6 plane. A l and A,,
initially separated by a small amount in the ~.6 direction, have evolved to B 1 and B 2.
The point B 2 is ahead and remains so as we move along the trajectories. However,
there are other points on the B 2 trajectory, such as B;, which are much closer to B~.
To expose this we write, again to order e,

d~4 £ "d~4
(4a)

(48)

e 2 8~
_ A Az

_ AI

Fig. 1. Neighboring trajectories in the X4 -- X,~ plane.


J. Poh'hinskt / Effectme lagrangians 273

so that (~4,'~6) is the vector from B~ to the point vertically above on the B 2
trajectory. (This is a convenient definition as long as/34 ¢ 0, so that the trajectories
do not turn vertical in the )~4 - ~,~ plane.) Thus we find

d~6
A-~-2~6=
{ ~a&
6 + - -Ok4
- A In/34 ~¢6, (5)

which integrates to

A2

As long as we remain at couplings sufficiently small that the integrand in (6) remains
small and that the ratio of/~4's runs sufficiently slowly*, the behavior of ~e, at small
A is dominated by A2/A~. Thus one would say that ,~ is an irrelevant parameter.
T w o nearby traJectories a p p r o a c h each other strongly in the infrared, the separation
going as a power of A/A¢~.
Fig. 2 shows the trajectories in the ~.4 - ~6 plane. As the initial conditions vary in
a two-dimensional space, the theory in the infrared lies very near a one-dimensional
subspace: as soon as we know k4. we know ~ . to accuracy A2/A~. Actually, even
this one relevant parameter just marks where we are along a single trajectory: it is
transmuted into a scale.
N o w let us relate this to the usual language of renormalization, lmagine starting
with a bare theory at a scale Ao, with k4 set to a particular valuc and ~ , to zero; this
is the point C~ in fig. 2. At a much lower scale A R, we are at D 1. at which k,~ is
defined to have the value k~. N o w consider a larger cutoff A'¢~. We can find another

Xe

Fig. 2. ('onvergence of traJectories in tile k 4 k~ plane.

* This is juM the requirement that the anomalous dimensions do not ovcrv,helm the canonical ones.
274 J. Polchinskt / Fffectwe lagrangians

point C z, lying on a longer trajectory, which at A R arrives at D z where again ~4 =


,~r~ ,. We can proceed in this way, thus defining the bare coupling A.t] as a function of
~.~, A r, and A 0. N o w take A 0 ~ ~ holding A R and ~ fixed. We know that ~.6
depends on A 0 only as A~/A2o, so it approaches a limit. But this is just what we
mean by renormalizability: as we take the cutoff to infinity, holding the renormal-
ized couplings fixed, all other quantities in theory (in this case A.6 is all there is)
approach limits as inverse powers of the cutoff. Thus, renormalizability follows in a
very general way when dimensional analysis is applied to the renormalization group
equation for an effective lagrangian.
T w o points should be made clear. First, it is not being said that ~.6 goes to zero in
the infrared -- merely that its value is determined in terms of A.4. Second, there is
nothing special about the ~.:axis in fig. 2, it simply corresponds to the way that
calculations are usually done, with only the relevant bare couplings non-zero. We
could use another curve in place of the X : a x i s to define our bare theory, correspond-
ing to bare non-renormalizable terms with coefficients characterized by A 0 to
negative powers, and the result would be the same. The one thing we cannot do is to
take A o to infinity while keeping bare non-renormalizable terms with coefficients of
inverse powers of some smaller scale.
Let us outline, in the context of eq. (2), the strategy to be followed in sect. 3 for
scalar field theory. Starting from initial conditions ~.4 = ~.°4, ~ , = 0 at A o, eq. (2)
defines the functions ~,(A, A o, ~ ) . The vector

.,~,,~ ~,, ( A , .~,,. X°~). (7)

satisfies the linearized eq. (3), because AoO/aA o c o m m u t e s with A O/OA. Thus. as
with e, before, it should be nearly parallel in the infrared to the relevant trajectory.
We want to subtract off the relevant part. the part parallel to the trajectory, but will
do this differently from before. Rather than subtract off a multiple of the tangent
vector to the trajectory, we will subtract off an appropriate (A-dependent) a m o u n t
of

a Is)
axe,
The vector (8) also satisfies the linear eq. (3), so it too should be nearly parallel to
the trajectory in the infrared. Define, then,

a~.t (,~) a~kt (i,~' ) [ O~4(A) ,] 1 0~4(], ~ )


UI(A )=
o - 0Ao
- OXC] 1 OX~ ], A o - 0Ao
- (9)

" In fact, it is not always possible to do this, sometimes ~ ] diverges for finite Ao. However. in
p e r t u r b a t i o n theo~', our eventual interest, 3~0a always exists as a p o w e r series in ~aa (see the end of sect.
1).
J. Polch~nskt / l?ffectit:e lagrangians 275

so that v 4 ( A ) = 0. o,(A) satisfies a linear equation similar to (5). and we can again
conclude that it is driven small in the infrared• The point of the definition (9) is that

v,(A) = AOd~o)~,(a, ,4_,,, h°4(A, 2~4, A0) ) . (lo)

That is, it is the total derivative of ~.,(A) holding )~4(A) fixed. For A = A R it is the
derivative of the point D in fig. 2 as the bare cutoff is changed. Once we conclude
that t:,(A R) is of order A~/A2o times slowly varying functions, eq. (10) may be
integrated to conclude that ~.,(A R, "4-o,X'](AR, ~4, A0)) has a limit as the cutoff is
taken to infinity. This is the usual statement of renormalizability. The quantity to be
studied in the next two sections is the analog of v,(A ).
There are many more useful exercises with these toy equations. Eq. (1) may be
solved order by order in gO. Divergences appear, just as in field theory, which canccl
magically in renormalized quantities (those expressed in terms of g4, not g]~). Of
course, from the point of view of fig. 2, this is not mysterious at all. One may prove
this to all orders, starting from the linearized equation for t',(A), and one is led to
the same steps as will be taken in sect. 4. Finally, additional relevant and irrelevant
couplings may be added, with similar conclusions.
Let us summarize the relation between the flow of the effective lagrangian and
renormalizability. Suppose it is known that the effective lagrangian at low scales is
strongly attracted towards an n-dimensional submanifold, where n is the number of
renormalizable couplings. Consider the n + 1 parameter theory defined by the values
of the n bare couplings and the bare cutoff ."t°. Then in general each point on the
n-dimensional submanifold is the image of a one-dimensional curve in the set of bare
theories• It is then possible (modulo the global questions discussed at the end of sect.
1) to take A° to infinity along such a curve, with the physics at a given scale thus
remaining fixed to accuracy of inverse powers of A°.

3. Scalar field theo~


We will consider a scalar field theory in four euclidean dimensions, with a
momentum space cutoff. The propagator is

( p'- + ,,,: ) - ~ K ( p ~ / A L ) . (11)

Here K(p2/A2) is a general cutoff function which we will take to have the value 1
for p2 < A 2 and to vanish rapidly at infinity• We define the theory by the propagator
(11) and vertices given by the interaction lagrangian

Lm, = f d 4 x ( _ ~ ~p,,~
,,2 {x)- , o
~p,(a,,,~(.,•)) ~
-- 1 o°,~(x) ) . (12)
276 J. Polchtnski/ Effectn,elagrangians
The bare couplings, which we label p°a, are usually called
pl~ = ~m 2 " Oil_ = Z - 1 , 03
o =
Xo • (13)

The system may be represented by a functional integral

d4P [_~_,(p),(_p)(p: +m2)~ -,(p2//~],)


Z(J) = f d , t , exp s[ (2,,)4 L

+J(p)eo(-p)] + Li,,t (q~)) , (14)

where J ( p ) is the external source used to obtain the n-point functions.


W e w i s h to integrate out the h i g h - m o m e n t u m c o m p o n e n t s of 4~, so t h a t we reduce
the cutoff in (11) to a much lower scale A R. We will take

]m21 < A R , (15a)

J ( p ) = 0, forp z > A~. (15b)

That is, A R is kept above the scale at which we are probing the physics. When we
integrate modes out, new effective interactions are generated. To see this, write the
functional integral with a general interaction lagrangian L(~, A):

Z(J.L,A)=fdrpexp f ~ d4P [ - 51q ~ ( p ) q > ( - p ) ( p2 + m2)K l(p2/.]2)

+J(p)q~(-p)] + L(,,A )I
?

=-f deoexp S( , , A ) . (16)

so that

dZ ([ d4P [_ p2 o~ '(P2/A2)]OA
A-d-~=fdep .t ~ - ) - g [ ½q~(p)ep(-p)( +m2)A

+A o~L(~,A))expS(~,A). (17)
J. Polchm~.ki / Effective lagrangmns' 277

l 1

(o) (b)
Fig. 3. (a) Replacement of propagator with (Ol./O~) 2 vertex. (b) Replacement of propagator wilh
( 3'2l./a~ 2 ) vertex.

If we choose

;¢L ' fd4p(2~r)a(pe+m 2) 'A OK(P2/A2


;~ a A = - a;.~

f aL OL #2L
(18)
x,~ a4,(_p) 04,(p) + a4,(p)a4,(-p)/"
then eq. (17) becomes

dZ aK(p2/A 2) o
A =f d p'x aa f d4, O4,(p)
×{(4,(p)K, l(p2/A2)+~(2~)a(p2+m2) -I 34,(c9')-p),

× e x p S ( 4 , , A )l = 0. (19)

(Recall that, due to (15b), J(p) has no overlap with c~K/c)A.) This naive manipula-
tion is justified because there is a cutoff. We have neglected field-independent terms
which just change Z(J, L, A) by an overall factor.
Eq. (19) says that if the cutoff A is reduced and simultaneously the lagrangian
changed according to (I 8), Z ( J ) and its functional derivatives, the n-point functions,
are left unchanged. Eq. (18) has a simple graphical interpretation. As modes are
removed from the propagator, compensating terms must be added in the interaction
lagrangian. Graphs where the differentiated propagator connects two different
vertices, as in fig. 3a, produce the first term in the bracket in (18), while graphs
where both ends of the propagator connect to a single vertex, as in fig. 3b, produce
the second term.
Although the lagrangian might start with a simple form such as (12), at lower
scales it becomes quite complicated. However, at scales A far below A0, we expect
that a great simplification will occur. That is, no matter what initial lagrangian we
start with (within limits) the lagrangian will be strongly attracted toward a three-
dimensional submanifold in the infinite-dimensional space of possible lagrangians.
Three, of course, is the number of relevant or renormalizable operators, namely 4,2
4,024,, and 4,4 *. A convenient set of coordinates for the submanifold is obtained by

* For simplicity we keep the symmet W q~ ---, -4'.


278 J. Pohhm~ki / l:2[/ectiPe la~rangtans

expanding L( ~, A ):

L(q},A)= ~2 1 ]" d"p~£..d_4p~mL 2tn ( tOl P2,,, A )


, n = l (2m)! a (2~r)*m 4

As always, we ignore the m = 0 field-independent piece. Then define

o , { A ) = - t . , ( o , o. ~,~), (21a)

1 02
P2(A) 8 Op~ L2(pI'-p1"A)Ir'~°" (21b)

o3( A ) = - 1.4(o.o.o.o, A ). (21c)

Note that in the approach we are taking, massless particles are no complication: (21)
will be suitable even if m = 0. We emphasize that on the submanifold, the lagrangian
is not expected to have any simple form such as (12), but that once the three
"coordinates" (21a-21c) are given, the rest of the action is fixed.
We need to divide the flow of the lagrangian into relevant and irrelevant parts.
Consider the vector (in the space of lagrangians)

A ,, ~ J~{,, a, A,,. o" ). ( 22 )

where the dependence on the initial condition (12) is written explicitly. This vector
satisfies the lincarized equation

A O_~,~( A ( ) ~ ) if _~(pOK, 2 + m2 ) 1

02
{ o l,o
x . 2 o e o ( - p ) o , ( p} ~" {, o~,, t+
(23)

which we will abbreviate:

{24a)
J. I'olchinskt / l-ffi, ctire lagrangtan.s 279

and in particular for the parameters (21),

- o OA.

At small A, the vector (22) is expected to be very nearly parallel to the three-dimen-
sional relevant submanifold. So are the three vectors

0
- - L ( ,~. A. A,,, p" ). (25)
ap~

which also satisfy eq. (24). Thus, if we form the linear combination of A,,aL/OAo
and OL/Op.
" " " which vanishes in the three directions (21). it should be driven very
small in the infrared:

OL(A ) OL(A ) Op'3 (A


V(A)='a° aAo ~, ,,, aph(A)AO Op' (26)
,. / Op,, OA o

Here Opl~/Op,,(A) is the matrix inverse to Oph(A, A 0, p°)/Opl~. This inverse always
exists in perturbation theory, as the zeroth order term is just 6,h" V(A ) satisfies the
linear equation

A aV(A)
aA (
M A. aL(A)
OAo )_ E M (
..h
aL(A)
a"
. P.
Oo.
a0,,(A~A° OOh(A )
aA,,

+ OL(A) OP° Mh( C3L ) OPt" Ao OIOd(A)


Z
a, h, ¢, d
a-"
Pa
ap~(a) I ap? ap,,(A) aA,,

_ y, OL(A) OO2 (, 0L(A))


°,,, ao,'~ apb(A ) g,, ,'~ o -~S-S,,

= M(V(A))- Y~ OL(A) oOp~M,,(V(A)) ' (27)


..h ap~ aph( A )

using the linearity of M. Eq. (27) is our key. It is an equation linear in the quantity
we wish to prove irrelevant, analogous to eq. (5). When (27) is written out explicitly,
it will contain complicated non-linear terms involving products of V with L. The
exact form of these terms no longer matters, however, for these terms are the ones
which vanish fastest when the coupling is made small. Thus, at sufficiently small
coupling, the infrared behaviour of (27) will be dominated by the purely linear
terms. These terms drive V to zero according to its canonical dimensions, since we
280 J. Pohhtnski / l-ffecm,elagrangians
have constructed V so that all pieces with d i m e n s i o n s o f m a s s to n o n - n e g a t i v e
p o w e r s are absent. W e will p r o v e that this is true in p e r t u r b a t i o n theory. T h e p o i n t is
to s h o w t h a t the a n o m a l o u s d i m e n s i o n s are c o u p l i n g s t i m e s p o w e r s of l o g a r i t h m s , so
the i n t e g e r c a n o n i c a l d i m e n s i o n s d o m i n a t e .
In o r d e r to p u t b o u n d s o n V, we will n e e d b o u n d s on the r u n n i n g l a g r a n g i a n (20)
a n d its first f o u r - m o m e n t u m d e r i v a t i v e s , on the q u a n t i t y

OL(A) _ ~ OL(A) Op~ (28)


O0~(a) 002 Op,,(a) "

w h i c h a p p e a r s in (27), a n d its first f o u r - m o m e n t u m d e r i v a t i v e s , a n d finally on V a n d


its first f o u r - m o m e n t u m d e r i v a t i v e s . W e will o b t a i n s o m e f u r t h e r results a b o u t these
q u a n t i t i e s , a n d t h e n p r o c e e d in sect. 4 w i t h the o r d e r - b y - o r d e r p r o o f . T h e c u t o f f
function K(pZ/AZ) will be t a k e n to h a v e the f o r m such as

1, p2 ~ ,,i2
K(pe/A2)={exp{(l - p~/A 2) 'exp((4-p2/A 2) t)]. ,42 < p ~ < 4 A 2
0, p2 > 4A 2 "

(29)

w h i c h has b e e n c h o s e n b e c a u s e it is infinitely d i f f e r e n t i a b l e , b u t its d e r i v a t i v e


v a n i s h e s e x c e p t for A 2 < p2 < 4A 2 , . D e f i n e

Q(P'A'm2) 1 A,o~K(p2/A:).
(p2+m2) (30)

O n e m a y easily verify that there are c o n s t a n t s C a n d D,, such that for Im2l ~ A R ~ A,

f d~P [Q(p,A,m2)I<CA4. (31)


(2'n')"

0" )
-O-p; Q( p" A , m 2 < D,,A " (32)

"It might cause some concern that eqs. (16), (17) and (19) refer to K i(p), yet K(p) is taken to
,.anish for p2 > 4A2. In fact the n-point functions arc still invariant under eq (1 g). as ma'. be seen by
considering K(p) as the limit of a K(p) which no~.here vanishes. Since the propagator (11) and the
effective lagrangian (18) involve only K(p) and not K 11p). tl'fi:, limit is smooth and tile invariance
of the n-point functions under eq (18) for K( p ) implie.,, the same for K( p ). h is also possible to carp,
through the whole proof with K( pJ, but then a different norm from (33L with a non-zero weight
function falling, off as p ~ at. must be used.
J. Poh'hmskt / Effective lagrangtan.~ 281

The notation [I I[ will be used often. It is defined

I I f ( p , . . . . . p.,. A ) I I = max I f ( P , . . . . . P..... 4)1. (33)


f'," < ,.1.12

for a function of one or more momenta. This is a useful definition because the values
of the effective vertices for p ) > 4A 2 are meaningless, the propagator vanishing in
this range.
Now write the renormalization group equation (18) in terms of the dimensionless
functions A 2 . , = A 2"'- 4L2m,where L2,,, are the component functions in (20):

( A ~,I + 4 - 2m)A2.,(pl ..... p,,,,.A )

Q(P.A.m2)A21(pl . . . . . P21 i.
/=1

×A_.... , ~. 21( P21 . . . . . P2,,,, - P. A ) + 2l-2ml - 1 permutatums/,

- (2rrAd4p)
aA2 .....- ( p' . . . . . P2 .... p. - p . A ) Q ( p . A , m2). (34)

y'2/-
where P - - ,=l 1P,.
_
Referring to (31) and (32) gives

(A~+4-2nl}A2,,,(p 1. . . . . P2,,,- " )

'" 1 ( 2 m ' I
~ ,._.~, { 2 \ 2 1 - 1 ) D ° I ' A z ' ( A )[I'J'A2 .... 2 2,(A)ll, + _,CIIA2 .... 2(3.)11-

35)

Eq. (35) shows the simplicity of the present approach. Because the momentum
integrals encountered are always restricted in range, they may be cstimated naively.
r e p l a c i n g / 1 2 , , , - 2 ( P , ) with its maximum valuc. Thus. the detailed momentum depen-
dence no longer enters and eq. (35) is nearly as simple as the eqt, ation.s in sect. 2.
From (35) one can see the strategy that will be followed in the inductive proof. At
any ordcr in perturbation theory, the non-linear terms on the right-hand side of (35)
282 J. Polchmski / t'ffectwe lagrangtans

will already be known. At each order it is then necessary to proceed inductively


d o w n w a r d from large m, so the linear term on the right will be controlled (at any
given order A 2., vanishes for large enough m).
N o w consider

( 0
Op2
O
Op¢ )A 2 . , ( p 1. . . . . P2m, A)=_O~,lA~.,.
'" -
(36)

Only such a difference of derivatives makes sense, as A 2,,,( Pl . . . . . P2.,. A ) is defined


only for 52, p, = 0. Acting on eq. (34) with 0 t,I
~' and proceeding as in (35), it folk)ws
that

"1 1
/=~l {2( 2~/~F/1)( DI z/~ I['A2/(A )[' " ""~2m t-2 2/(/~ )1'

+2Do [OQ,A2t(A)II'IIA2.,, 2- 2t(A)l[)} + ~Cll3~,A2 .... 2(-"~)11 -

(37)

Again this will involve only lower orders or larger m. Similar results hold for any
number of derivatives acting on A 2.,-
Turning now to the quantity OL(A)/Opt,(A). eq. (28). one finds, using the fact
that M is a linear operator,

4 0_0_(OL(A))=M(OL(A)) OL(A) (OL(A)) (38)


• O,,~ aO~(A), Op~(A) -,~, O0.(A)M" O0~(a) "

Define an expansion in terms of dimensionless functions Bh,2m:

OL(A)_ ~. A 4-2" 28~,


Oo6(A )
.,=1
(2rn)! f d 4 P l ' " "daP2m Bh,2m(
(2¢r) s" "'
Pl ..... P2,,,' A )

×64( ~ P i ) ¢ ( Pl)'.'q~( P2m) • (39)


J. Pohhtnshi ,/ t:ffecttve lagrangtans 283

Eq. (38) becomes

(o
A-~-+4-2m-28
)
m Bh,2m ( p l . . . . . P2 .... A)

= -- /'--~1:={ Q ( P ' A ' m 2 ) A 2 I ( P ' . . . . . P2I" 1" P ' A )

×Bin2 .... 2 _w( P21 . . . . . P2m, - P, A ) + (\ 21_12m ) - l p e r m u t a t i o n s }

_ "-i f ~ Bd4P
h.2 .... 2 ( P , . . . . . P : . , . P . - P . A ) Q ( P . A . n'2)

+ B, .... (~,, . . . . . p2,,,. :~)~ f d'~q l?',, 4( 0.0, q, --q, ~'~)e(q, A, m 2 )


'- (2~rA)4 -

/k2
+/32.2,,,( [1 ..... P2 ..... '~") 16

×f d4q 02 B, 4 ( q l . - q t . q , - q . A ) [ , , , .Q(q,A m:)


(2zrA)4 a q ( - "

+ B3.2,,,( P~ ..... P2,,,. A }

×f d4q 4 Hb 6 { 0 ' 0 " 0 ' 0 " q" -- q" A ) Q ( q. A . D'I2),


G {40)
(2~rA) "

so that

. . . . .

+ ~ CIIB,,.2 .... 2(A)II + ! C!jB,.2,,,(.~* )[I-IIB,, ~(A )!1

4- ,',. ('AX]I B2.2.,( A )]l' 113~.231'.x B,,.4 (A)l[

+ 5 CllBs.>,,( A )1{-11B,,.,,(A )1[. (41l

Similar results, parallel to (37). hold for any number of derivatives acting on B/,.:,,,.
284 J. Polchmskt / Fffi,cnce lagrangians

Finally, expand V(A) in dimensionless functions V2.,:


A4 - 2., d4Pt ... dap2,,,

,42)
Eq. (27) becomes
( a ~ 03 X + 4 - 2m) V2., (P, ..... P2., . A )

=- ~
/=1
(Q(P,A,m2)A2t(P,p 2 t - I , P , A )
.....

×V2"'z-2t(P2/ ..... Pz"'-P'A)+(21-2ml ) - l p e r m u t a t i o n s }

-~f d4----C-
p V2-,* 2( P~ . . . . . P2m' P' -- P, A ) Q ( p , A , m 2 )
(27rA)4

+B,.2.,(p~ ..... p2.,,a)~ f d4--A--q V4(0'0' q" - q" A )Q( q. A, m 2)


(27rA)4

42
+B~.~,.(p~ ..... p~.,,a)~6 f d4q
(2~ra)4
0z
× ~ V 4 ( q l , - q l , q , - q , A ) l v , = o Q ( q , A , m2)
Oq?
+ B ~ ( p , ..... p2.,.a)!f d4q
' (2~rA)4

X I:6(0,0,0,0, q, - q , A)Q(q, A, m2). (43)


From this
( A o-~ + 4 - 2 m ) V 2 . , ( p l ..... p2.,,. A )

~ {( 2m )DoIIA2/(A)I[.I[V2 m 2:( A)II}


t=l 21-1 *2-

+ ½CIIV2,,,, 2(m)ll + ~CIIB~,2.,(A)II [IV4(A )11


+ ~t~CA211Bz.2,.(A)II.IIO~.20~,2Va(A)[I
+ tflln3.2.,(A)[]. IIV6(A)II, (44)
and so on for any number of derivatives.
.I. Poh'htnskt / l?ffectit,elagrangtan.~ 285

4. Perturbative renormalization

In this section we will prove, in perturbation theory, that V(A ) is driven to zero in
the infrared. At the end of the section we give the steps needed to convert this into a
statement of the cutoff-independence of n-point functions•
Theorem• Consider the field theory defined by the propagator (I1) and vertices
(12). Define the effective lagrangian L(O, A, A o, po) and the relevant parameters
0h(A ' 3.o. po) as before. Define p~(A R. ~ a Ao ) implicitly by

o , ( A R , Ao, O") = O, (45a)

p: (Art, A o, po) = 0, (45b)

o~(A R, A,), po) = ~.R. (45c)

Then order by order in perturbation theory in /~R the limit

lim L ( ¢ , A R, ~R, Ao ) = L(~, A R,)xR, oc), (46)


• *1 II "~ 9C

exists, where L(¢, A R, )~R Ao ) _= L(q~, Art, A o, p°(A R, )xrt, Ao) ), and at rth order in
h g•

II L~',).( AR, A,,) - L~r2,(AR. ~o)11

~<A~ 2"( AR ) 2
,~o p2r . , ( I n ( A , , / A R ) ) " r+l-m>~0

=0, r+l-m<0,

(47)

where p2r ,,, is a polynomial of degree 2 r - m whose coefficients, taken to be


non-negative, are pure numbers*• The degree of the polynomial is not essential to
the proof and the reader is free to ignore the superscript on P throughout. We retain
the degree only because it is easy to do so; in fact it can be reduced to r + I - m, but
this requires distracting additional steps.
Note that eqs. (45) actually do define p0 implicitly in perturbation theory, for at
order r in AR the left-hand side of (45) is pl~(r) plus terms which are already known.
Lemma (i). At order r in ~.R,

11¢91*t'Zt• . . 0 ~ f . , / 'a2<, ,r,)P¢t ..... P2-, . 3-)II~A '- p e 2 r ,,,(InCAo/AR) ) + r+ 1 -m>~0,

=0, r+l--m<0. (48)

* P" is defined to be zero for n < 0.


286 .I. Poh'htn.~ki / t:ffectipe lagrangtans

Proofoflemma (i). N o t e that the initial v a l u e of A2,,,, at A 0, is d e f i n e d by eq.


(12), so the L H S o f (48) v a n i s h e s at A 0 for m / > 3, for m = 2, p > 1, a n d for m = 0,
p > 3. T h e l e m m a is trivially true for r = 0, as (45) i m p l i e s "2,,a(°)= 0. S u p p o s e it to be
true for r = s - 1, for s o m e s > 1. W e n o w p r o c e e d d o w n w a r d in m. T h e l e m m a is
true for m >/s + 2, that is Ac,) "2,,, t h e n vanishes, for the o p e r a t i o n s s h o w n in fig. 3 c o u l d
h a v e linked t o g e t h e r at m o s t s vertices, l e a v i n g 2s + 2 e x t e r n a l fields*. S u p p o s e it is
true for m >i n + 1. T h e r i g h t - h a n d side of (35) at o r d e r s is

,1 .s- 1

-~-I Y" ! 2 - 1~) D°llA(Sz)(A)[I" A~;'"z) 2z(A)ll + - ;l C l l A ~ ' ; ) ' 2 ( A ) I I " (49)
/= t=l

All q u a n t i t i e s in (49) are b o u n d e d by the i n d u c t i o n h y p o t h e s i s , so (49) is less t h a n or


e q u a l to

n ~ 1 1(
~p2, ,(in(Ao/XR))P2, 2. . . . : ,(In(A()/AR))+p2 ..... In(A()/AR) )
/-1 t=l

= p2 ..... I ( I n ( A ) , / A R )) "

(50)

In a s i m i l a r way, all q u a n t i t i e s o n the r i g h t - h a n d side o f (37) and the c o r r e s p o n d i n g


e q u a t i o n s for h i g h e r d e r i v a t i v e s are b o u n d e d by the i n d u c t i o n h y p o t h e s i s , so

A ~ +O4 - 2 n )0 u'
,,,/,'" .3~":A~;](pl
:, ..... p2,,,A) <~A PP2' " '(ln(.~,,/a,)).

(51)

I n t e g r a t e (51). F o r n >/3
,u-i . . :,1,%.,4(-~ ) [ n
Ila,,.,,. %,.,/.2,~,-1 ..... P2,. ,'1)1i

.. : . ' . d , " (' :~_,)'-"':'


< A ,,p2, .- ,(In(A,)/AR))jA ~7{7 ,.~.

<A pp2 . . . . ,(In(Ao/AR) )

<~A pp2 .... (in(A,)/3. R)). (52)

* It is curious that this one simple fact cannot be determined from eq. t34). To see the problem,
consider the simpler system dr:( x ) / d x =f, ~ 1( x ). i = 1..... oo,/i(0) = 0. This has many ,,,olution.',,
such asf,(x) = 0, orfl(x) = exp( l/xZ),f,(x) = 2exp( l / x 2 )/x 3. . . . . Eq. (34) becomes complete
with additional information, such as the vanishing of At,'~], for large m.
J. Pohhmski / Effecttt'e la,wan.gians 287

This establishes the induction step down to n = 3. In thc same way it carries to n = 2,
p >/ 1, but for n = 2, p = 0 there is an extra term from the initial value of ,4 4, and we
proceed differently. By (51),

A A~4"(O,O,O,O,A)[<~P2' 3(ln(A,,/,'~ r~)). (53)


I

Integrate (53) up from the initial value A ]'~= - - 8 '1 at A = A~, eq. (45c). to obtain

IA~4"(O,O,O,O,A)I <~6't + P2' Z(In(A,,/ARj)=p2, :(In(A,,/AR)) ' (54)

N o w reconstruct A] '~ via Taylor's theorem:

.I
A~a'l( PI" P2, P3" P4" ,4_ ) = A ~ " ( 0 , 0 , 0 , 0 , A )+ E
t,]=l
P,~P;'~d),(1
I - k)

×(0" a,,
~..~,.,.4..,~ P; , P z' - P ; , p4, A))l,,'-~,r, . ( 5 s )-

Both terms in (55) are b o u n d e d by p 2 , - 2(In(Ao/AR)). the first from eq. (54) and
the second from eq. (52) for n = 2, p = 2, so the induction step follows for all of A] '~.
For n = 1, p/> 3, eqs. (51) and (52) follow immediately, while from (51),

A ~ -OX + 2 )A~]~ (O,O,A ) ~ p2., -(In(Ao/AR)),


,

/ ..~ \
}~/ AOA av n , )zt . .,.y l , p ~ , A
v + 2 / a ~vl.zv,,z,,
' )e~=ps=0 <~A- Zp2, - = ( I n ( A , , / A ~ ) ) . (56)

Again integrating up from the initial conditions (45a, 45b) one obtains

IA~,,(0,0 ' ,4_)I ~< p 2 , -2(In(Ao/AR)),

10~.2 (~"
Y l , 2 "a' 2, , " ( p ; , p ~ , A)p;=p~_0l <~a 2pZ, l(ln(Ao/A R)) (57)
288 J. Polchinski / l'.ffectwe lagrangtans

N o w use Taylor's theorem to reconstruct A~2'*:

A(')t
2 ~Pl,-Pl, a)=AC')(O,O,A)+ .~p,p,(O~.2a~.zA~2O(pl,pz,
~' . . . . A))p;=v,.=o

+ ~ .1 P l P~t x Pv l Po l wf) I dX(1 - X ) 3

×(O~.20'l'2O'~.2O,'.2A(2~)(p~,p~,,A))p;__p. xp. (58)

Eq. (52) for n = 1, p = 4 and eq. (57) establish the induction step for n = 1, and the
lemma is proven.
It is somewhat contrary to the spirit of our proof that we have had to integrate up
from A R and not just down from A o, and this deserves some comment. In the case
of A4(0) and 02,42(0), this was just a convenience. The initial conditions for these at
A o involve 0 ° and 0 °, so proceeding as we have, we avoid carrying po and 0 ° around.
(This is also why In A R appears so early: it comes entirely from the dependence of p0
and p~ on ln(Ao/AR).) In the case of A2(0), however, integrating from A R was
essential. The point is that our theory is not actually on a typical trajectory. On a
typical trajectory, A 2 ( 0 ) - A{j/A 2, that is, the scalar mass is of order A o, and this is
all we would have learned integrating down from A o. By imposing (45a) we have
forced the initial values of 0° to be finely adjusted so as to produce a scalar with
m << A0. We must therefore integrate up from (45a) to produce the needed bound
on A 2. This is of course the famous naturalness problem for light scalars [8].
Lemma (ii). At order r in Xa,

- , . , , ' ' ' v , , ,..~'p


[}~', . , p U h .R{')
2,,,'Pl l ..... P2,,," A ) [[~ A I,p2," "'-l+a"'(ln(Ao/"~n)) .

r + 2 - m > 0,

= 0. r + 2 - m < 0. (59)

Proof of lemma (ii). From the definition (28), the initial condition is

P; (6o)
Bt,,em(Pl . . . . . P2,,,,A(,) = -~/,1~,,,1 - ~t,2~,,,1 Z - ~,3~m2 .

This also gives the full zeroth-order term,

(61)

as may be verified by inserting (61) into (40). Thus the lemma is established for
J. Pok'hmski / Effectu'e lagrangians 289

r = 0. N o t e that the definition (28) gives certain parts of B h exactly:

B~ r) (0, 0, A ) = - (~ rO(~/, 1 , (62a)


Ou " R(") [
1,2 81,2 ~h,2 ', P l ' P2' A ) m - v ~ - 0 = - 28~08u"81,2 • (62b)

B~'7,(0,0,0.0, ,'1 ) = - 8~°8,,,. (62c)

S u p p o s e now that the l e m m a is true for r = s - 1, for some s >I 1. A g a i n proceed


d o w n w a r d in m. T h e vanishing of B~~~,~ for m >/s + 3 follows from the definition
(28) and the vanishing of ~(") ,.2m for m > ~ s + 2. S u p p o s e the l e m m a is true for
m = n + 1. Eq. (41) becomes

(A ~ + 4-2,1-28m)B~,">z,,(pl ..... p2,,. A )

<~k Y] (( 2 / -n 1 ) D°I]AV)(A)II'¢ [['-t,.2


o" . .". .2
= t " 1

+_, C•liB,,,,)2,,
_ , )
2(A)iI+'C -
Y'. '~~l ,'2"n (.1)ll'lJOh('4 " ( A ) [ I
t=[)

•, l

R(t} /" .,,~ ht ,u (~ - t)


+ ,',,C-12 ~ 11--2.2,,,-)11' la~ 2a~ 28,, 4 (A)II
t-O

+ , 'C~ ~ 3 , 2 n ~~A)II'I B~,'o ')(.~)11 .


II m" (63)
/=0

Because of the z e r o t h - o r d e r term (61), there are some terms on the right side of (63)
which involve B ~'~ and not just B ~ l ) and so are p o t e n t i a l l y not b o u n d e d by the
i n d u c t i o n hypothesis. These are the last three terms when t = 0 and the third to last
term when t = s. The latter p r o b l e m involves only h -- 3 and the u n k n o w n b o u n d is
on BI~),,. so there is no p r o b l e m if at cach value of s and n we b o u n d B[~,),, before
(•)
B3,,,. The former p r o b l e m is not present if n > 3, because of the form of (61), so for
n>3

, A(~-A + 4 - 2 n - 2 8 m B('I
t~. 2 ~I ( P i "" --./)2,, " -1

Z ~, {e-"' '(1n(.i,)/,1.~)) P2' 2, ,,.,-~,.,(in(Ao/.lR)) }


/.11 1

+p2, ,,-a,,,(In(A,)/AR))

+ L pZ, ,,.l(ln(Ao/.1R))pe,.'-, l.~,.,(ln(.~,)/AR))


t=O

+ like terms ~ p : , ,,. s,.,(ln( . 1 o / ,~R ) ) . (64)


290 J. Polchinski / kffectiee lagrangians

In a similar way,

(Ao~+4-2n-23h1)O~' t l , I 1" " " 0 ~i p . j p B ch .~2 n) ( P , . . . . . P2,, , A )

<~A - " e 2...... '"' ( l n ( A 0 / A R ) ) .


(65)

Integrating

I]O,~',j,... O t~'~
p , Jp B~S)2
, gl (p~ . . . . . P2n * A)II

~< •4_- pp2,- .... , +8~, (In( A o / A R )) (66)

which establishes the induction step down to n = 3. Once (66) is known for n<.~ ~ h . 6 ' the
right-hand side of (63) is bounded for n = 2 and (64) and (66) then follow for n = 2.
Then with (66) for n<s~u h , 4 , (63) is bounded for n = 1 and (64) follows; eq. (66) now
follows for n = 1, p >/2 and with (62a) the rest of R{,~ ~t,.2 can be reconstructed using
Taylor's theorem, completing the lemma.
L e m m a (iii). At order r in )~

II0~, '. ,, - • - O~,~,~v~r'(P~.


.., . . . . . P2.," A ) I I ~ A ,.T~,, e 2 " ' ( l n ( A " / A " ) ) '

r+l -m>_-O,
=0, r+l -m<0. (67)

Proof o f l e m m a (iii). T o obtain the initial condition on V, given the definition


(26), use


4- o -L-a
0A o
L(A.A o,
o°)l,,~,,+Aa-~-sC(a ,
4-0.p°)l ..'~_,.

= Ao~L(Ao, Ao, pO) = 0. (68)

so that with eq. (51) of iemma (i), the initial condition satisfies

-,,.,, . . . . , , . j ~ . 2 , . , p , ..... P 2 , . , A 0 ) l l ~ A 0 " e ~-. . . . ~(ln(Ao/aR)),


r+l -m>_-O,
=0, r+ l-m<0. (69)
J. Polchinski / l-ffective lagrangians 291

Also from the definition (26). it follows that ,t"m~2,,= 0. Suppose the lemma to be true
for r = s - 1, some s >/ 1. Again we argue inductively d o w n w a r d in m. The vanishing
of V,~;~,~ for m > s + 2 follows by the same argument as for a~.~ Suppose the lemma
_ Jl2/q I"

to be true for m >/n + 1. Then from (44):

( A ~ + . - 2.)<,:,(~, ..... ~:.,. A)

y, n DoIIA~9(A)II . "v2-n-' 2, I 2z(A)l I +~CIIV~i[~z(A)II


/=1 t=l 2 -1

D(t)
+ ~c ~ {11,,, -,,,(A)II.IIV2' "(A)li
t=0

+ ~A 2 t~2.2,,(
"~,1 A )ll" 0~ 20'~ 2V4~' "(,4,)11

+ m,,
~3,2m (A)II'IIV,,'"-"(A)II} (70)
As in lemma (ii), the induction hypothesis bounds the right-hand side of (70) if the
t = 0 terms vanish. In particular, for n >/3,

~" 8A + 4 - 2 n V~;;'( Pl . . . . . P2"" A ) ,T. ,

and with derivatives

(A ~.81 + 4 - 2 n ) 8"',,. , , . . . O*','j,


1 . , V2~;"(p
. . . . . . p2,,..~_ )

( ')2
~<A p 7~"o p2, ,, l(ln(A,,/,~ R)). (72)

Integrating (72) gives

, ?

01'.' , . " " ,,,. j,,V2,, ( Pl . . . . . P 2 . . "X )11 ~< A P ,~ p2, ,, l(]n( ,io/.~. R ))

Ja A' ~ A '

' A 2m. 1' 4


+.., ,{, ~) p:, ,, l(,n(.,,,/,,, II
292 J. Polchinskt / l:ffeutiue lay,rangians

(the last term is from the initial value (69))

A )2
<~A ~[ S~,, p2, "(ln(,~,,/,,~,~)), (73)

establishing the induction step down to n = 3. As in lemma (ii), eq. (73) for n = 3
gives us eq. (71) and (72) for n = 2 . Eq. (73) then follows for n = 2 . p > / 2 but
I/4(0, 0, 0, 0, A ) is known (it is zero by construction) and Taylor's theorem establishes
the induction step for all of n = 2. Eq. (71) and (72) now follow for n = 1, eq. (73)
follows for n = 1, p >/4. Again, n = 1, p = 0 and p = 2, vanish by construction at zero
m o m e n t u m and we use Taylor's theorem to establish (73) for all of n = 1 and thc
iemma is proven.
P r o o f o f theorem. By definition of V,

V((#. A R , )~R, AO ) = ,1_,,d.~, 1. (¢" "4-R, Ao" P°( '1-R" )~R.,4,,)). (74)

so that integrating (74) and using lemma Off),

IIL~'2,(AR, ao) - ~2n,l{r)/, A, R ,%)11

(75)

x{ pZr mt'n,Ao,AR,'_ __
A'o,
p2,-m(ln(,Vo/A

r+l-m>~0,

r+l-m<0.
R)) },

The existence of a limit (46) with the property (47) follows from elementary
properties of limits (Cauchy's criterion), completing the theorem.
Ultimately we are interested not in the effective lagrangian but in the n-point
functions. A general n-point function G ( A o, A R, ~R) is defined initially with the
propagator (11) cut off at A 0 and the bare vertices (12), with po defined as a function
of ~R by the renormalization conditions (45). The point of integrating out modes was
that we get the identical answer if we use P ( p , A R) cut off at A R, and the vertices of
the effective lagrangian at A R. Thus at order r, G¢'~(Ao, A R, AR) is given by a finite
n u m b e r of terms of the form

f d4,pL~%~(AR,Ao)...L~%~(AR,
Ao)p(p1,AR)...p(p,,,Ak), (76)
.I. I ' o k h m s ' l ~ t / Kl][ecm'e lagrangtan.~ 293

where Y,r, = r. Thus, at order r in ~3.

IG'r'(Ao)-G'"(oc)' ~ ( f d4'pP( pL,AR)...P(p,,,AR ))

X II-2.,,
/-(rtI(AR, Ao). • . /*{, 2rt ~~ lI (t A R , . ' ~ o )

_ f~(2,-,h, t~a~ "R ' ~)...L%',,'(A


^ R" at)l] • (77)

The separate integral over the propagators has no A° dependence and is obviously
convergent for m 2 > 0, converges at rn = 0 if we avoid the IR divergent point
p~,~,,,j = 0. and converges for m 2 < 0 if we shift to the correct minimum. The shift in
the last case is no problem, as the theorem for L(~. A~. ~R. A0 ) implies the same
result for [.(~ + v. A R. ),u. Ao). The difference in lagrangians in (77) is bounded bv
the theorem, so

[ G ~ , ( A o ) _ G~r~(oc)l ~< Ao 2 × polynomial in I n ( A J A r < )

x finite, A o-independcnt quantities (78)

and so the n-point functions have limits as A o --, zc. which they approach as 1/.l{j.

5. Discussion and conclusions

The argument leading from the linear equation for V(A ), eq. (27). to renormaliza-
bility, eq. (78). is lengthy. We claim, however, that the result was a foregone
conclusion, for the reasons discussed following (27). ()nce loop integrals can be
treated naively, as in going from eq. (34) to eq. (35), the difference between the field
theory and the toy equations such as in sect. 2 all but disappears. Since the toy
equations can be understood even for small finite coupling, we always had a map
through the perturbative woods.
Two things that we have come to expect are missing from this proof. The first is
Weinberg's theorem, which justifies naive power counting for multiloop Feynman
graphs with many potential subdivergences. The second is a discussion of such ideas
from graph topology as skeleton expansions and overlapping divergences. The point
we have tried to make is that renormalizability is a general property which does not
hinge on these particular technical points. In our case, naive power counting for
integrals was justified because they were always over the limited range A e < p2 < 4.~e.
The renormalization group equation automatically disentangles the overlapping
divergences, for in building an arbitrary Feynman graph with the operations shown
in fig. 3, it always constructs the subgraphs with the largest momentum first.
294 J. Poh'hinskt / I:.ffectice lagrangians

The method here can be applied to composite operator renormalization as well, by


including ~,O, (O, = composite operator) and identifying the terms of order e in the
effective lagrangian. Operators of any dimension can be studied, since by including
in the bare L all operators (of given symmetry) up to dimension d, we may separate
from V not just the relevant parts, but the subleading parts up to A4~ ,/.
The proof here extends immediately to any system whose symmetries are pre-
served by a momentum-space cutoff. The principle is very general, however, and
should apply to any physical cutoff, such as Pauli-Villars for abelian gauge theories
and the lattice (probably with discrete rescaling) or higher covariant derivatives for
non-abelian theories. The main problem is to obtain the renormalization group
equation appropriate to the cutoff. It is unclear whether dimensional regularization
can be adapted to this purpose, as in that case it is d - 4 and not the scale p, which
plays the role of cutoff. It may actually be possible to treat gauge theories with a
momentum-space cutoff". The point is that a momentum cutoff need not change the
physics, as the missing modes can be replaced by effective terms in the action. In
terms of the new action, the gauge invariance should still be present in some
disguised form.
Our key equation (53) and its component form (60) were derived without using
perturbation theory in an essential way. It may be possible to make some progress
understanding these at finite coupling, starting perhaps with truncated versions. Of
course, for a ~.~4 theory at finite coupling the continuum limit cannot actually be
taken, as discussed at the end of sect. 1, but it may be possible to cstablish more
limited results.

I am indebted to C. Arabica, D. Friedan, and J. Preskill for inspiration. I would


like to thank S. Coleman, S. della Pietra and N. Sakai for suggestions on the
manuscript.

References
11] K.(J. Wilson. Phys. Rev. IM (1971) 3174, 3184:
KG. Wilson and J.G. Kogut, Phys. Reports 12 (1974) 75
[21 L.P. Kadanoff, Physics 2 (1966) 263:
J. Glimm and A. Jaffc, Fort. Phys. 21 11973) 327;
E. Brczin. J.C Le Guillou and J. Zinn-Justin in Phase transitions and critical phenomena, eds. C.
Domb and M.S. Green (Academic Press, London, New York) 1975:
T. Appelquist and J. Carazzone, Phys. Rev. D l l (1975) 2856:
S. Weinberg, Physica 96A (1979) 327
[3] F.J. Dyson, Phys. Rev. 75 (1949) 486, 1736;
A. Salam, Phys. Rev. 82 (1951) 217; 84 (1951) 426
[4] N.N Bogoliubov and O.S. Parasiuk, Acta Math. 97 (1957) 227;
K. ttepp, Comm. Math. Phys. 2 (1966) 301,
W. Zimmermann, Comm. Math. Phys. 15 (1969) 208

" I v,'ould like to thank D. Friedan and S. Shenker for discussions of this point.
J. P~,hhm~'kt / l:~([e¢tn'e ktgrangtan.~ 295

[5] C.(i. Callan, Jr., Phys. Rev. I)2 (1970) 1541:


A. Blaer and K. Young, Nucl. Phys. B83 (1974) 493:
C.G. ('allan, Jr., in Methods in field theot3, ed~. R. Balian and J. Zinn-Justin (North-tlolland, 1976)
I6] S. Weinberg, Phys. Rcv. 118 (1960) 838:
Y. Hahn and W. Zimmermann, Comm. Math. Phy,,,. 10 (196~) 330
[7] J. Fr6hlich, Nucl. Phys. B2OO[FS4] (1982) 281
[8] S. Weinberg, Phys. Rev. D13 (1975) 9 7 4 : D I 9 (1979) 1277:
I.. Su,,,skind. Phys Rcv. I)20 (1979) 2619:
K Wilson, Phy~,. Rev. I)3 (1971) 1818

You might also like