LECTURE 5: RENORMALIZATION GROUPS (CONTINUED) David Gross

Notes by P.Etingof and D.Kazhdan In this lecture we will return to the Wilsonian point of view on the method of renormalization group, which was discussed in the rst lecture. Namely, we will discuss the exact renormalization group equation, which describes the renormalization group ow in the space of all Lagrangians. This equation together with asymptotic freedom provides a tool for constructing nonperturbative eld theories. We follow J.Polchinski's paper \Renormalization and e ective Lagrangians", Nuclear Physics B231, 1984, p. 269-295. For simplicity we will consider the theory of a scalar bosonic eld in 4 dimensions, as in the rst lecture. We consider the theory with a cuto propagator Pp(p+= ) , m where P is a smooth positive nonincreasing function, such that P (a) = 1 for a 1 and P (a) ! 0 su ciently fast as a ! 1. The free action for a massive scalar associated to this cuto can be written in momentum space as
2 2 2 2

(5.1)

1 Z ^(p) ^( p) p2 + m2 d4 p S0( ; 0) = 2 P (p2 = 2) (2 )4 0

Z 1 0 0 0 (5.2) ) = ( g1 2 + 1 g2 (r )2 + g3 4)d4 x; I 2 2 4! where gi0 are formal variables, and consider the total action S ( ; 0) = S0( ; 0) + 0 SI ( ). Consider the theory whose Green's functions are de ned by the path integral Z R (5.3) Z (J ) = D e S( ; )+ J d x:
S 0(
0 4

(hat denotes the Fourier transform). This action is very big if high Fourier modes in are present, so it \suppresses" high Fourier modes in the path integral. Also, when 0 ! 1, this action obviously tends to the standard action for a massive scalar. Introduce a perturbation of the action S0 by \interaction" terms:

Because we are considering a cuto theory, all Feynman diagrams are well de ned, so we get a perturbation series. Now consider some smaller scale R < 0. We are interested in physics at the scale R. This means, we assume that m << R and we are interested in values ^ of Z (J ) for such J that J (p) = 0 for p2 > 2 . R A S
Typeset by

-T X

The physics at the scale R will be described by an e ective Lagrangian, which is obtained from the original Lagrangian by \integrating out" degrees of freedom between R and 0. To study the e ective Legrangian, we introduce an arbitrary interaction term SI ( ; ) and consider the integral (5.4)

Z (J; S; ) = D e

Z

S0 ( ; )+SI ( ; )+R J d4 x

d Z D e S ( ; )+SI ( ; )+R J d x = 0: (5.5) d Di erentiating, we get Z 1 Z ^(p) ^( p)(p2 + m2) d P (p2= 2) d4p dSI ( ; ) D 2 d (2 )4 d R ^ ^ d p (5.6) = 0: e S ( ; ) SI ( ; )+ J (p) ( p)
0 4 0 4 (2 )4

Let us look for such SI ( ; ) that Z (J; S; ) is independent on , and write down a di erential equation for such SI ( ; ). The independence of Z of yields

>From this one can get

(5.7)

@SI = 1 Z d4p(2 )4 (p2 + m2) 1 @P (p2 = 2) @ 2 @ 2 SI SI SI + ( p) (p) ( p) (p) :
2 2

Indeed, substituting (5.7) into (5.6), we get the integral of a complete derivative, p ^ i.e. zero; here we heavily use the fact that J (p) has disjoint support with @P (@ = . And it can be shown that (5.7) follows from (5.6). Note that equation (5.7) corresponds to a linear equation for eSI : (5.8)
2 eSI @eSI = 1 Z d4p(2 )4 (p2 + m2) 1 @P (p2 = 2) @ 2 @ ( p) (p) :

Equation (5.8) can be regarded as an in nite-dimensional heat equation. It is called the exact renormalization group equation. If SI ( ; ) is the solution of this equation 0 such that SI ( ; 0) = SI ( ), then SI ( ; R ) is the e ective action at the scale R. This equation allows to prove the renormalizability of 4 theory without use of graph techniques, in particular of Weinberg's theorem (see Polchinski's paper).

2