1 Complex scalar Field in d=4
Shankar starts with the partition function of the form:
Z
Z= [dd ]eS(, ) , (1)
|k|<
where:
Y dRe(~k)dIm(~k)
[dd ] = , (2)
|k|<
Z
S(, ) = (~k)J(k)(~k). (3)
|k|<
This Hamiltonian is obtained from Fourier transform of the nn interaction
Hamiltonian:
1X
S(, ) = |(~n) (~n + ~i)|2 (4)
2
n,~i
~
Vector ~n counts sites on a hypercubic lattice while ~i labels 8 nearest neighbors
(d=4). Doing Fourier transform it is easy to get that:
J(k) = 2 [(cos(k1 ) 1) + (cos(k2 ) 1) + (cos(k3 ) 1) + (cos(k4 ) 1)]. (5)
Since we are interested in low energy (small ~k) we can expand (5) and obtain:
d4 k
Z
S S0 = k 2 |(~k)|2 (6)
|k|< (2)4
This is gaussian model. In Landau-Ginzburg Hamiltonian this would corre-
spond to Fourier transform of the square of the spatial derivative of the field
(()2 ). The integration over the complex field we can write as:
Y dRe(~k)dIm(~k) Y d(~k)d (~k)
[dd ] = = (7)
2i
|k|< |k|<
Equation (7) we get by using the fact that Re() = +
2
and Im() = 2i
.
In this model the only averages that survive are the averages of the product
of even number of fileds (~k) where every field is followed by its complex
conjugate. We can show this using the properties of the Gaussian integral
1
and doing the average value of product of two complex variables (zi and zj )
with Gaussian weight:
R dxi dyi dxj dyj a z z aj zj z
zi zj e i i i e j
hzi zj i =
R dxi dyi dxj dyj a z z aj zj z
e i i ie j
R dxi dyi a z z R dxj dyj aj zj z
zi e i i i zj e j
= R dx dy dxj dyj a z z (8)
i i
e ai zi zi e j j j
R dxi dyi R dx dy
(xi + iyi )eai zi zi j j (xj iyj )eaj zj zj
= R dxi dyi dxj dyj a z z aj zj z
e i i ie j
Where we used notation Re(z) = x and Im(z) = y. Using the property that
2 2
ezz = ex ey we see that (for i 6= j)R in the numerator of equation (8) we are
2
going to have integrals of the form dxi xi eai xi which are zero. For i = j
we are gona have:
R dxi dyi 2 2 2
(xi + yi2 )eai xi eayi
hzi zi i = R dxi dyi ax2 ay2
e ie i
R 2 ai x2i ayi2
R 2 2
dx i dy i x i e e +
dxi dyi yi2 eai xi eayi (9)
=
a
a2 1
= =
a
a
i,j
We get that: hzi zj i = ai
. If we genaralize this result to a complex fields in
4d we get that:
(2)4 (~k1 ~k2 )
h (~k1 )(~k2 ) = (10)
k12
1.1 Renormalization Group approach
Renormalization group (RG) transformation is applied to find the gaussian
fixed point and then perturbations are considerd that can be relevant or
irelevant or marginal. Relevant perturbations will drive the system away
from the Gaussian fixed point upon application of RG. First we need to find
the fixed point for Hamiltonian (Action) S0 . We are gona separate fileds
into slow modes < , (k < s ) and fast modes > ( s < k < ) and integrate
over the fast modes (large ks).This is the first step in RG transformation.
2
In Gaussian Hamiltonian (S0 ) those modes are not mixed so integrating fast
modes gives us the original Hamiltonian where is replaced by < . Modes
are not mixed because of the fact that:
( (
, for k < s , for k > s
< = , > = (11)
0, otherwise 0, otherwise
R R R
Integral |k|< can be written as |k|< + <|k|< . We see that the integral
s s
of the product < > is going to be zero. Then, after the first step in RG we
have:
d4 k
Z
0
S (< ) = < (~k)k 2 < (~k) (12)
|k|< (2)4
~0
Second step of RG is to rescale ~k: ~k ks and the third step is to rescale the
~0
fields 0 (~k 0 ) = <(k ) . Doing these two steps we get that the renormalizied
Hamiltonian:
2 d4 k 0
Z
0 0
S ( ) = 6 0 (k~0 )k 02 0 (k~0 ) (13)
s |k0 |< (2)4
If = s3 then Gaussian Hamiltonian is a fixed point: S 0 (0 ) = S() = S .
Now we are gona look at perturbations from this fixed point and classify
them into relevant, irrelevant or marginal. First is quadratic perturbation:
d4 k
Z
S = < (~k)r(k)< (~k) , (14)
|k|< (2)4
where:
r(k) = r0 + r2 k 2 + .... (15)
This perturbation like the Gaussina Hamiltonian doesnt mixes up the slow
and the fast modes so we get:
k0 d4 k 0
Z
0
S = s 2
0 (k~0 )r( )0 (k~0 ) (16)
|k0 |< s (2)4
0
Shankad defines renormalized coupling as: r0 (k 0 ) = s2 r( ks ). If we compare
the Taylor expansion coefficients we see that:
r00 = s2 r0
r20 = r2 (17)
r4
r40 = 2
s
3
because:
k0 2 2 k 02 k 04
s r( ) = s (r0 + r2 2 + r4 4 + ......) (18)
s s s
We see that r0 is a relevant parameter while r2 is marginal and the rest of
the parameters are irrelevant. So the coupling function r(k) reduces to a
coupling constant r0 .
Next, we introduce a quartic perturbation:
Z
1
S = (~k4 ) (~k3 )u(~k4 , ~k3 , ~k2 , ~k1 )(~k2 )(~k1 )
2!2! |k|<
4
Y d4 ki
(2)4 4 (~k4 + ~k3 ~k2 ~k1 ) (19)
i=1
(2)4
Z
1
4 3 2 1 u(4321)
2!2!
The renormalization of this perturbation is a bit more complicated since now
we have mixing of fast and slow modes. We can write the entire Hamiltonian
as S() = S0 () + S(). Then we get that S(< , > ) = S0 (< ) + S0 (> ) +
S(< , > ). PArtition function can be writtena s:
Z Y Y
Z= d(k) d(k)eS0 (< ) eS0 (> ) e(< ,> )
0<|k|<
s
s
<|k|<
Z Z
[d> ]eS0 (< ) eS0 (> ) eS(< ,> )
[d< ]
Z Z (20)
S0 (< )
= [d< ]e [d> ]eS0 (> ) eS(< ,> )
Z
0
[d< ]eS (< )
Where S 0 (< ) is defined as:
Z
0 S0 (< )
S (< ) = e [d> ]eS0 (> ) eS(< ,> )
R S0 (> ) S(< ,> ) Z
S0 (< ) [dR> ]e e (21)
=e [d> ]eS0 (> )
[d> ]eS0 (> )
= eS0 (< ) heS(< ,> ) i0
4
Cumulant generating function is defined as K(t) = lnhetX i. And the cumu-
2 2
lant expansion is given as K(t) = hXit + hX ihXi
2
t2 + .... From this we see
hX 2 ihXi2 2
that hetX i = ehXit+ 2
t +....
Applying this to equation (20) (t = 1) we
get:
hS 2 i0 hSi2
0 +.... 0
0 S0 (< ) hSi0 +
S (< ) = e e 2 = eS0 (< )+S (22)
Where:
hS 2 i0 hSi20
S 0 = hSi0 + + .... (23)
2
The leading termin this expansion has the form:
Z
1
hSi0 = h (< + > )4 (< + > )3 (< + > )2 (< + > )1 u(4321)i0 (24)
2!2!
There is 16 possible averages in (23): 8 possibilities with odd number of fast
field which are zero, 1 term with all fast modes (gives a constant contribu-
tion), 1 term with all the slow modes (called the tree-level term) and 6 terms
with two slow and two fast modes. The tree-level mode is obtained from
the original perturbation by replacing by < and rescale the momenta and
fields:
k40 k10
Z
1
0
ST ree = ~0 ~0
(k4 ) (k3 )u( , ..., )(~k20 )(~k10 )
2!2! |k0 |< s s
4 (25)
4 4 ~0 ~ 0 ~ 0 ~ 0
Y d4 ki0
(2) (k4 + k3 k2 k1 )
i=1
(2)4
Renormalized coupling function is given by:
k40 k0
u0 (k40 , ..., k10 ) = u( , ..., 1 ) (26)
s s
Doing a Taylor expansion it is evident that u0 term is marginal while the
rest are irrelevant. Thats why in the rest of the derivation u0 is used as a
coupling constant. Out of the six terms that have two fast modes only 4 will
give the non zero contribution (the ones not containing > > or > > ):
Z
1
S20 (< ) = u0 h [> (4)< (3)+< (4)> (3)][> (2)< (1)+< (2)> (1)]i0,>
2!2! |k|<
(27)
5
1
All of the terms give the same contribution which gets rid of the 2!2! . Let
us use this term:
Z
0
S2 (< ) = u0 h > (4)< (3)> (2)< (1)i0,> (28)
|k|<
This example can be used to ilustrate the use of diagrams. On the image
above we have a diagram for the term given by the equation (27). Arrows
going to the center correspond to and the outgoing to . S correspond
to slow modes and F to fast modes. We joined two fast modes which are
averaged out. The line conecting ~k2 and ~k4 contributes, according to the
rules of Feynman diagrams:
1
(2)4 4 (~k2 ~k4 ) 2 (29)
k2
So, we finally get:
Z Z Z Z
S20 (< ) = u0 < (~k3 )< (~k1 )
k1 k2 k3 k4
(30)
1
(2) (~k4 + ~k3 ~k2 ~k1 )(2)4 4 (~k2 ~k4 ) 2
4 4
k2
6
Two delta functions are going to kill integration over k4 and k3 and we
end up with:
d4 k1 d4 k2 1
Z Z
0
S2 (< ) = u0 < (~k1 ) ~
( k1 ) (31)
< 4 2
k1 <
s
(2)4
s
<k2 < (2) k2
Integrating the integral over k2 and rescaling the fields and momenta we get:
d4 k 0 ~ 0 0 ~ 0 2 1 1 2 2
Z
0 0 ~0 2
S2 ( (k )) = u0 s 4
( k ) (k ) [1 ] (32)
k< (2) 2 s2 (2)4