Note FixedPoints RG Gitterman Halpern
Note FixedPoints RG Gitterman Halpern
Note FixedPoints RG Gitterman Halpern
Chapter 5
In fact, this idea was already used in Eq. (4.15), where we compared
the free energies of site and block lattices. In this chapter, we con-
sider the connection between Hamiltonians rather than between free
energies.
The key idea in the use of the renormalization group is that the
transition from H to H can be regarded as a rule, K = f (K), for
obtaining the parameters K of the Hamiltonian H of a block lattice
from those K of the Hamiltonian H of the site lattice. This pro-
cess can be repeated, with the lattice of small blocks being treated
as a site lattice for a lattice of larger blocks, so that for the latter
K = f (K ) = f [f (K)], and so on indefinitely for larger and larger
49
June 25, 2004 14:18 WSPC/Book Trim Size for 9in x 6in chap05
50 Phase Transitions
blocks until their size reaches the correlation length. This is an exam-
ple of a map, i.e., a uniquely defined rule relating two successive
members of a sequence, xn and xn+1 by means of xn+1 = f (xn ). In
such a map, it can happen that as n → ∞ this process may stop at
some stage, when x∗ = f (x∗ ), in which case x∗ is called a fixed point
of the map. For instance, one of the simplest examples of a non-linear
map is the logistic map [19]
so that the fixed points are the roots of the equation x∗ = µx∗ (1−x∗ ).
Since this is a quadratic map, it has two fixed points, namely x∗ = 0
and x∗ = 1 − 1/µ. This situation is shown in Fig. 5.1 where the fixed
points are the intersection of the graph of the parabola, f (xn ), and of
the diagonal of the unit box. Starting from some initial point x0 we
make a vertical shift to the point x1 = f (x0 ) lying on the parabola,
which (after the horizontal shift to the diagonal) will be the starting
point x1 for the next approximation, and so on, reaching finally the
stable fixed point x∗ = 1 − 1/µ. In order to examine the stability of
the fixed points, we introduce a perturbation ξn from x∗ in the form
xn = x∗ + ξn (5.2)
Fig. 5.1 Approach to the fixed point for the logistic map.
June 25, 2004 14:18 WSPC/Book Trim Size for 9in x 6in chap05
K = f (K ) = f [f (K)] = · · · ,
52 Phase Transitions
and the same procedure can be performed for each two subsequent
terms. In this way, we exclude from the sum all the spins with even
indices. Thus, we can compare the partition functions for the site
and block lattices, and write
N
Z(K, N ) = F 2 Z(K , N/2). (5.12)
54 Phase Transitions
Let us now compare the free energy per site, f1 (K), for the site
lattice and that, f2 (K ), for the block lattice. Since
f1 (K) = (1/N ) ln[Z(K, N )] and f2 (K ) = (2/N ) ln[Z(K , N/2)],
it follows from Eq. (5.12)–(5.14) that
f (K ) = 2f (K) − ln 2 cosh(2K) . (5.15)
Moreover,
1
K = ln [{exp(2K) + exp(−2K)}/2]
2
1 1
< ln [exp(2K)/2] = K − ln 2,
2 2
i.e., K is smaller than K, and so by repeatedly doubling the size
of the blocks one can proceed to a very small value of interac-
tions between spins far away from each other. Therefore, on going in
the reverse direction from small to large interactions, one can start
with a very small value of the interaction, say K = 0.01, so that
Z(K ) ≈ 2N and f ≈ ln(2). Using the above equations, one can find
the renormalized interaction and the appropriate free energy. It turns
out [18] that it takes eight steps to proceed from K = 0.01 to the
exact result.
The one-dimensional Ising model does not show a phase transition
at a finite temperature, and our analysis was only designed to show
by means of a simple example how the RG method works. We now
turn to the two-dimensional Ising model, for which we have seen in
Chapter 2 that a phase transition does occur.
s1
s4 s0 s2
s3
Fig. 5.2 Ising model on a square lattice with alternate sites removed.
with the spins of the sites that we omitted. Since the spin S0 located
at the site (0, 0) interacts with the spins S1 , S2, S3 , S4 , located at
the sites (0, 1), (1, 0), (0, −1), (−1, 0) respectively, we must consider
the terms in the partition function arising from S0 = ±1 with all
the 16 possible values of (S1 , S2, S3 , S4 ). As we will see immediately,
these 16 configurations are divided into four different groups which
will give four different equations instead of the two equations (5.13)
for the one-dimensional lattice. In order to satisfy these equations we
are forced to introduce three different interactions in the partition
function of a block lattice, i.e., to replace the cell of the site lattice
shown in Fig. 5.2 by the following interactions in a block lattice
exp[K(S1 + S2 + S3 + S4 )] + exp[−K(S1 + S2 + S3 + S4 )]
= F {exp[(K1 /2)(S1 S2 + S2 S3 + S3 S4 + S4 S1 )
+ K2 (S1 S3 + S2 S4 ) + K3 S1 S2 S3 S4 ]}. (5.16)
56 Phase Transitions
ln 2 = ln F − 2K2 + K3 (5.19)
This set of four linear equations for the four unknowns has the unique
solution
1
K2 = K1 /2 = ln[cosh(4K)],
8
1 (5.21)
K3 = K2 − ln[cosh(2K)],
2
1 1
F = 2[cosh(2K)] 2 [cosh(4K)] 8 .
In contrast to the one-dimensional Ising model, for the two-
dimensional Ising model, as well as in some others cases, the resulting
interactions in a block lattice are more complicated than those in the
original lattice, and so they are not suitable for an exact RG analysis.
To proceed, one must use some approximation. The simplest approx-
imation is to neglect K2 and K3 , but then we come back to the equa-
tions considered in the one-dimensional case, where there is no phase
transition. To obtain a physically meaningful result, we assume that
the four-spin interaction is so weak that it can be ignored. The terms
in K1 and K2 correspond, respectively, to interactions between near-
est neighbors and next-nearest neighbors in the block lattice, and
both favor the spins being parallel, and so we write K1 + K2 = K .
Thus,
too bad for such an approximate theory, and suggests that its basic
physical ideas are correct.
5
µa = sgn Sia . (5.23)
i=1
As can be seen, the block lattice √ is also a square lattice with a dis-
tance between blocks of L = 5. For µa = 1, there are sixteen
possible values of S a , which can be grouped into the six different
configurations shown in Fig. 5.4, having characteristic energies 0,
±2K, and ±4K. For µa = −1 one obtains the same energies as for
µa = 1 with signs reversed.
In order to calculate the partition function Zµ for the Hamiltonian
Hµ of the block lattice it is convenient to write Hµ = H0 + V , where
H0 contains the interactions between the spins within a block and V
1
4 0 b
1
L 2
4 0
3
a 2
3
Fig. 5.3 Ising model on a square lattice with five sites in a block.
June 25, 2004 14:18 WSPC/Book Trim Size for 9in x 6in chap05
58 Phase Transitions
Fig. 5.4 The different spin configurations for the five sites in a block on the
square lattice, where black and white circles represent spins of opposite signs.
The number of states in each configuration and their energy are shown.
60 Phase Transitions
Hence, to first order in V , βHµ = Const + K a,b µa µb , where
2
cosh(4K) + 2 cosh(2K)
K = 3K . (5.34)
cosh(4K) + 4 cosh(2K) + 3
5.6 Conclusion