Introduction to Statistical Mechanics

and a Two-Dimensional Ising Model

J.F. Nystrom
University of Idaho
Moscow, Idaho 83843-1920
Email: [email protected]

Keywords: Statistical mechanics, Quantum statistics, Ising model,

Critical exponents, Scaling hypothesis.
PACS numbers: 01.30.Rr, 05.20.-y, 05.50.+q, 75.40.Cx

Abstract
In order to analyze macroscopic systems while taking into account
knowledge of quantum behaviors, it is required to use such techniques
as are available under the label of statistical mechanics. Herein we
present an introduction to statistical mechanics and provide a suite of
models to demonstrate the e ectiveness of the methodology.

Introduction
Quantum mechanics (QM) uses the idea of quantization to postulate the
existence of discrete states of a system. The time-dependent Schrodinger
equation
ih @@t = H

gives information about the time evolution of the system, while the timeindependent Schrodinger equation

H =E

(1)

gives the energy values of a system. Using the time-independent equation,

it is easy to show that the quantum oscillator in one{dimension has energy
levels of
En = n + 12 h!
while the oscillator in three{dimensions has energy levels of
En = n + 32 h! :
For macroscopic systems though, it is unreasonable to expect that we could
write an exact Hamiltonian to describe the systems time-evolution, or even
nd the exact energy levels available to said macroscopic system. Therefore,
quantum mechanics is only appropriate for isolated few-body systems. To
attempt to describe actual macroscopic systems, one utilizes many of the
ideas from quantum mechanics to perform calculations which fall under the
label of statistical mechanics.
The paper starts by giving an overview of the fundamentals of statistical
mechanics. This introduction, combined with the thermodynamic review
3

in the Appendix, gives a foundation for the statistical mechanical models

discussed in the remainder of the paper. In x1 the entropy representation is
utilized to show the e ectiveness of the statistical mechanics methodology.
Callen's excellent treatise 1] on thermodynamics and statistical mechanics
is utilized extensively throughout the paper.
The models in x1 are rederived in x2 using the canonical ensemble before
attention is turned to the analysis of quantum spin systems. Included in x2
is an introduction to the Ising model and a review of a full two{dimensional
renormalization group calculation. The discussion summarizes the theoretical e ectiveness of statistical mechanical techniques, including comments on
the universality hypothesis and cooperative phenomena.

The way of statistical mechanics

For macroscopic systems, with given values of extensive parameters U , V ,
and N (i.e., the energy, volume and particle number), we expect to have
many discrete states available that are consistent with the speci c values
of U , V , and N . That is, many di erent con gurations are available to
the system, such that, the systems extensive parameters are always equal to
the given values of U , V , and N . If we could write a quantum mechanical
Hamiltonian for a given macroscopic system, then using this Hamiltonian,
we theoretically could describe the time evolution of the system, and provide
probability estimates on future states. For an isolated atomic system, it is
then the act of measurement that tells us what actual state the system has
evolved into. But, no physical system can be isolated totally. Furthermore,
for transitions to occur between the quantum states of a macroscopic system,
4

it is enough that the system interacts with the so-called \vacuum" 1]. This
is a fact, because, for macroscopic systems, the energy di erence between
di ering quantum states that are consistent with the imposed constraints is
unperceptible. Thus 1]
A realistic view of a macroscopic system is one in which the system makes enormously rapid random uctuations among quantum states. A macroscopic measurement senses only an average
of the properties of myriads of quantum states.
Statistical mechanics further assumes that \a macroscopic system samples
every permissible quantum state with equal probability" 1]. This assumption
of equal probability of occupation is the fundamental postulate of statistical
mechanics.
Now, since we have made such a big deal about the number of permissible
quantum states, we naturally ask if we can somehow measure the \number"
of these states. According to Callen 1]
The number of microstates among which the system undergoes
transitions, and which thereby share uniform probability of occupation, increases to the maximum permitted by the imposed
constraints.
We now conjecture that the thermodynamic entropy should be associated
with the number of available states (consistent with the imposed macroscopic constraints). But entropy, being an extensive parameter, has an additive character when two systems are brought together, and the number of
microstates available to two systems is a product of the number of states
5

available to each, as shown in Figure 1. A unique answer to the question of

how to formulate the entropy is to identify the entropy with the logarithm of
the number of available states. Let be the number of microstates consistent
with xed values of U , V and N , then

S = kB ln

(2)

is the entropy, where kB = 1:3807 10;23J=K (Boltzman's constant) is the

prefactor chosen so that the tempature, given as
!
1 = @S
(3)
T
@U V N
agrees with the Kelvin tempature scale. This de nition nows gives a physical
foundation for the entropy associated with a macroscopic system in equilibrium.
In this representation, we are basically doing statistical mechanics in the
microcanonical formalism. That is, we set a xed energy, U , and nd the
number of available states of the system consistent with this constraint. Later
we will nd it advantageous to work in the canonical formalism, wherein
the temperature T is xed, and the fundamental relation of choice is the
Helmholtz free energy.

1 Statistical mechanical modeling

In this section we utilize the entropy representation to calculate the speci c
heat for two classical models. Both models deal with systems of identical
particles. In the rst model, the \two{state" model, the occupation number1 follows the Fermi-Dirac statistics, while the second model, the Einstein
6

model of a crystalline solid, has an occupation number that follows the BoseEinstein statistics.

1.1 Two{state model

For the two{state model, a total of N atoms comprise the system, and it is
assumed that each atom can have an energy value of 0 or . The total energy
of the system is xed at U .
Given a total energy of U , along with the fact that each atom can be in
only one of the two states, it is required that U= atoms be in the excited
state, thus (N ; U= ) are in the ground state. To calculate the entropy,
using Eqn. (2), we are required to determine the number of possible states
consistent with the current constraints. Recall from combinatorics, that the
number of ways to choose k items among n things is given by

0 1
n
B
@ CA = k! (nn;! k)!
k

i.e., \n choose k". Thus, the number of ways to distribute the U= packets
(or quanta) of energy among the N atoms is given as
=

! N!
!
U ! N; U !

(4)

which is exactly equal to the number of states accessible to the macroscopic

system. Using Eqn. (4) in Eqn. (2) the entropy for the two{state system is

"

!#

S = kB ln N ! ; ln U ! ; ln N ; U ! :
7

(5)

Stirling's approximation, ln N ! ' N ln N ; N , can be applied to Eqn. (5) to

obtain the fundamental relation:
S (U N ) = Nk N ln N ; U ln U
B

; N ; U ln N ; N ; U ln 1 ; NU
= ;NkB U ln U ; N 1 ; U ln 1 ; U
N
N
N
N

: (6)

Now calculate the temperature for this system:

1 = @S = kB ln N ; 1
(7)
T @U
U
and invert Eqn. (7) to obtain an equation of state for the energy:
; =kB T
(8)
U = 1 + Ne =kB T = 1N+ ee ; =kB T
where e ; =kB T is the Boltzmann factor, which is proportional to the probability that the state will occur for a given T . Let u = U=N be the energy
per atom, then the speci c heat is given as
du = 2
e =kB T :
C = dT
(9)
kB T 2 (1 + e =kB T )2
A plot of C 0 = C=kB as a function of x = kB T= is given in Figure 2.
This gure exhibits a so{called Schottky hump, indicating the presence of a
pair of low energy states lying below some considerably higher energy states.
The two{state models assumption of only two energy levels is equivalent to
assuming that the next energy state of each atom is so high that it can not be
reached. We nd that this model, just as with the examples in the Appendix,
show that simple thermal calculations can give information on the atomic or
molecular structure of material substances.
8

1.2 Einstein model for the heat capacity of crystals

The speci c heat for a crystalline solid at high-temperature can be calculated2
from the equipartition theorem to be C = 3R. But at low temperatures it
was found experimentally that the heat capacity goes to zero. This behavior
of the speci c heat as a function of tempature is shown qualitatively in Figure
3. Classically, there is no way to explain this behavior for the speci c heat.
We now model the crystal as made up of N atoms, with each atom bound
harmonically to six nearest neighbors. Each atom is free to vibrate around an
equilibrium position in any of the three Cartesian directions with a natural
frequency !0 . This model is shown in Figure 4. As the lattice is built up, we
assign three oscillators to each atom, and neglect edge e ects. Assume now
that each of the 3N oscillators can only take on discrete values of energy,
such that the energy of each oscillator can be written as

En = nh!0 n = 0 1 2 : : :

(10)

which are the well known energy levels for the one{dimensional oscillator
(neglecting the 1/2h!0 zero point energy). Therefore, each oscillator can
be occupied by an integral number of energy quanta. In the microcanonical
formalism, as usual, we x the energy at U , giving U=h!0 quanta to distribute
throughout the system. Unlike the two{state model presented earlier, in this
model we do allow the placement of all the quanta into one oscillator.
To calculate the entropy, we require the number of possible states that
are consistent with the imposed constraint (i.e., a total energy of U ). Let
represent the number of ways to distribute U=h!0 quanta among the 3N
modes. To count the number of states, consider a diagrammatic approach
9

which uses 0's to represent each of the U=h!0 quanta, and 1 to represent
the end of a mode (which means each list has U=h!0 0's and 3N 1's). If
all the quanta are in the rst oscillator, this con guration is written as
00 : : : 011 : : : 1, i.e., U=h!0 0's followed by 3N 1's. If the quanta are distributed such that each of the rst U=h!0 modes has only one quanta, the
list would look like 0101 : : : 011 : : : 1. Using this diagrammatic technique, we
basically want to know how many ways there are to place U=h!0 markers
among (U=h!0 + 3N ) locations, that is,

U + 3N !
h!
0!
!:
=
(11)
U ! 3N !
h!0
Applying Stirling's approximation, and letting u=u0 = U=3N h!0, we can
write the molar entropy, s = S=n, as
s(u v) = 3R 1 + uu ln 1 + uu ; uu ln uu
0
0
0
0
u
u
u
(12)
= 3R ln 1 + u + u ln 1 + u0 :
0

The tempature is calculated as

!
!
1 = @ s = kB ln 1 + 3N h!0
T
@u
h!0
U
which is inverted to give the energy per mode:
U = u h!
3N
u0 0
= (e h!0=kh!B T0 ; 1) :

(13)

After rewriting s in terms of T , the speci c heat for this crystalline model
is given as
!2
@
s
h!
e h!0 =kB T :
0
Cv = T @T = 3R k T
(14)
(e h!0 =kB T ; 1)2
B
10

A graph of Eqn. (14) is shown in Figure 3. Taking the limit of Eqn.

(14) as T ! 1, we nd Cv ! 3R (the expected result), and as T ! 0,
Cv ! 3R(h!0 =kB T )2e;h!0=kB T , approaching zero as T ! 0. Even though
the Einstein model is very simple, it gives a good qualitative agreement with
the speci c heat at low temperatures. While Eqn. (14) approaches zero with
an exponential decay, which is di erent from the experimental T 3 dependence
1, 5], the model deserves attention because of its elegance, and the fact that
this was the rst quantum theory of the heat capacity of solids.

1.3 Quantum statistics

Using Eqn. (13), we nd that the average number of quanta per mode,
(U=3N )=(h!0), i.e., the occupation number 2], is given by

nboson = e h!0 =k1B T ; 1

(15)

which is just the Bose-Einstein distribution. This should be expected since

we do allow more than one quanta to occupy any given mode (to wit: these
quanta act like bosons ). Compare Eqn. (15) to the occupation number,
n = U=N , that is obtained from Eqn. (8):

nfermion = e h!0=k1B T + 1

(16)

which is just the Fermi-Dirac distribution. Again appropriate in light of the

fact that at most one quanta can occupy each mode (to wit: these quanta act
like fermions ). In the limit as T ! 1, both Eqn. (15) and Eqn. (16) give a
behavior similar in functional form to the Maxwell-Boltzmann distribution,
going e;h!0=kB T .
11

2 Ising model
In x1 we utilized the microcanonical formalism to analyze two systems: one
with quanta that behave like fermions, the other with quanta that behave
like bosons. In this section we introduce the basics of the canonical formalism
and then reformulate the two{state model and Einstein's model in this formalism. Following the introduction of the canonical formalism, the classical
magnetic interaction is discussed, and the Heisenberg Hamiltonian for quantum spin systems is derived. A spin Hamiltonian is then applied to a linear
chain of spin variables, the so-called Ising model in one{dimension. This section ends with a description of a fully two-dimensional exact renormalization
group calculation involving an interesting geometrical con guration of spin
variables.

2.1 The canonical formalism

The convenient fundamental relation in the canonical formalism is the the
Helmholtz free energy, F (T V N ). To obtain the free energy, we can convert
fundamental relations from the microcanonical formalism using a Legendre
transformation, or develop the fundamental relation using the partition function. If the energy levels of a system are the sum over the energies of the
individual elements of the system, then the partition sum factors, such that
the partition function is given as

Z = z1 z2

12

zn

where zi are the partition functions for the individual elements. Thus, for N
identical elements, Z = zN , where

z=

X
j

e;

with = 1=kB T and j is the energy of the j th mode of one of the elements.
That is, z is just the sum of the Boltzmann factors associated with each
energy state. The Helmholtz free energy is then given by log of the partition
function:
F = ; 1 ln Z :
(17)
For the two{state model of x1.1, there are two separate energy states,
thus the partition function for each element is

z = e0 + e;

= 1 + e; :

Accordingly, for N atoms:

F = ; 1 ln zN
= ;NkB T ln(1 + e; ) :
The speci c heat is derivable from the free energy:

(18)

!
!
2F
T
@S
T
@
@F
T
@
@
@
C = N @T = N @T ; @T = ; N @T 2 = @T ; @ ln Z
(19)
which gives the same result as shown in Eqn. (9). For the Einstein model,
each mode can have zero, one, two, or basically an in nite number of quanta.
Thus, for each mode:
1
X
(20)
z = 1 + e; h!0 + e;2 h!0 + = e;n h!0 :
n=0

13

Recognizing a binomial expansion, Eqn. (20) is rewritten as

z = (1 ; e1;

h!0 )

which gives a free energy of

F = ; 1 ln z3N = 3NkB T ln 1 ; e;

h!0

(21)

Calculating the speci c heat for the Einstein model as is done in Eqn. (19)
gives the same result as shown in Eqn. (14).
Using the canonical formalism for the two{state model and the Einstein
model eliminated the need to count states. Recall, in the microcanonical
ensemble we x the energy and then calculate the entropy based on the
number of states consistent with the set value of energy. For the canonical
ensemble, we only need know the energy levels available to each element in
the system, and with this knowledge we construct the partition function,
wherefrom the free energy is obtained.

2.2 Heisenberg Hamiltonian

In classical electromagnetics, the interaction energy for a magnetic dipole,
m, in an applied magnetic eld, B, is given as 4]

Uint = ;m B :

(22)

Using a standard expression for the eld due to a dipole gives the dipole{
dipole interaction energy:

Uint = r13 m1 m2 ; 3 (m1 ^r) (m2 ^r)]

14

(23)

where r is the relative position vector from m1 to m2. If we limit the type of
relative orientations that the dipoles m1 and m2 can have, we gain insight
into the preferred orientations. When m1km2kr, such that m1 m2 > 0, the
energy, using Eqn. (23), is given as U = ;2m1 m2 =r3, which is the minimum energy con guration, i.e., the preferred orientation. If we specify that
m1?r?m2, then we force the dipoles into the type of con gurations shown
in Figure 5. These strict relationships between dipoles not only restricts the
relative location of the dipoles, but also only allows for two projections of
each dipole, on, say, the z-axis (e.g., mz ). The energy of the orientation
in part (a) of Figure 5 is Ua = m1 m2 =r3, while for the orientation in part
(b), the energy is Ub = ;m1 m2 =r3. Evidently, the orientation in part (b) is
preferred over that in part (a).
Unfortunately, this classical analysis of the interaction between two magnetic dipole moments is insu cient to explain some types of magnetic ordering, as these energies are on the order of the thermal energy (i.e., kB T ),
when T is near 0.03 K 3]. This means that above this temperature, thermal energy (stored in the hidden atomic modes of the spin system) will
dominate the magnetic interaction in determining the thermodynamics of
the system. In order to model the spontaneous magnetism of some metals, we must therefore consider quantum mechanical e ects. To begin, we
consider two interacting electrons, and assume that the spin interactions of
this system will dominate in determining thermal properties. Here the spin
represents the intrinsic spin of the electron, h=2. Restricting the dipoles to
only two orientations as done in Figure 5 will thus represent a spin up, or
spin down electron. We will nd that the interactions between electrons that
15

give rise to ferromagnetism are based on the Pauli exclusion principle and
the Coulomb interaction, and thus are electrostatic, not magnetic in nature.
We now commence to build the Hamiltonian for a pair of interacting,
not necessarily bound, electrons. For this two electron system, there are
four possible con gurations of spins, which in the braket notation can be
written as: j"#i, j""i, j#"i, and j##i. These four possible con gurations
partition into two distinct energy states, the singlet state, given 5, 6], for
p
example, as 1= 2 (j"#i ; j#"i), and a triplet energy state. To satisfy the
time-independent Schrodinger equation, Eqn. (1), we require a Hamiltonian
such that
and

H
H

s
t

= Es
= Et

s
t

where Es and Et are the eigenvalues, i.e., the energies, of the singlet and
triplet state, and s, for example, is the wavefunction for the singlet state.
Just such a Hamiltonian is given in 5] as
(24)
H spin = 14 (Es + 3Et ) ; (Es ; Et ) S1 S2 :
For the Hamiltonian in Eqn. (24) to give the correct eigenvalues, note that
in the singlet state, S1 S2 < 0 (i.e., the spins are antiparallel), and in the
triplet state, S1 S2 > 0 (i.e., the spins are parallel).
We will show now that the energy levels of the two states are such that
Es > Et . The Pauli exclusion principle requires that two electrons with
the same spin can not occupy the same location. To ensure this, the wave
functions for a system of two electrons should be antisymmetric under interchange of coordinates. While the triplet states do have antisymmetric wave
16

functions, the singlet state does not, which requires the singlet state to have
a larger electrostatic repulsion between electrons 7] than the triplet state
(to ensure that the two electrons do not ever occupy the same location). We
now view the quantity (Es ; Et) as an exchange energy splitting between
the two states, and accordingly, we let J = Es ; Et , where J is the so-called
exchange coupling constant. Using J in Eqn. (24) and setting a new zero for
the energy, the spin Hamiltonian for this two electron system becomes:

H 0 = ;J S1 S2 :
Note that if J > 0, parallel spins are preferred, and if J < 0, antiparallel
spins give the lower energy. Over a large system of spins, we then write the
Heisenberg Hamiltonian as

H=;

Jij Si Sj :

(25)

Technically, all the electrons in the system interact with each other, so that
the coupling constant Jij will vary, say, as the separation between spins
become larger. When an external magnetic eld is present, the Hamiltonian
will also include terms which represent the interaction of each spin with the
applied magnetic eld.

2.3 Ising model in one{dimension

The one-dimensional Ising model consists of a linear sequence of spin variables, si, where each spin variable can take on only two possible values
(si = 1). Each spin variable is connected to its neighbor's with a coupling
constant k, and potentially interacts with an externally imposed magnetic
17

eld. The Hamiltonian for the one-dimensional Ising model is a simpli cation
to the Heisenberg Hamiltonian, Eqn. (25), where if we let J be the exchange
coupling and B the external magnetic eld strength, then the Hamiltonian
for a linear chain of spins is:

H (si) = ;J

X
<ij>

si sj ; B

X
i

si

(26)

where < i j > means to sum only over nearest neighbors, and we have
assumed all nearest neighbors interact with the same e ective strength (i.e.,
Jij = Jjk = ). Note that in formulating Eqn. (26) we have simpli ed Eqn.
(25) by looking at nearest neighbors only, and have also added a term to
describe the interaction of the spins with an external eld. Let
( kB1T ) ,
and the dimensionless Hamiltonian, H, is given by:

H=; H=k

X
<ij>

si sj + h

X
i

si

(27)

where

k = J=kB T = J
h = B=kB T = B :

(28)

The dimensionless couplings k and h de ned in Eqn. (28) and used in Eqn.
(27) are referred to as a nearest-neighbor spin-spin coupling and as a singlespin ( eld) coupling, respectively.
The partition function in the canonical ensemble is given by:

Z=

X
i

e;

18

Ei

(29)

Using Eqn. (27) in Eqn. (29), we can write the partition function for the
one-dimensional Ising model in an equivalent form:
P s s ) exp(h P s ) ,
Z = Tr e; H = Tr exp(k
i j
i i
fsig
fsig
< ij >
(30)
where Tr means to trace over all possible con gurations of spins, and < ij >
says to operate on nearest neighbors only. The (Helmholtz) free energy is
given by:
F = ;kB T ln Z = ;1= ln Z
(31)
and the reduced free energy per site is given by:

f = ;( =N ) F = (1=N ) ln Z :

(32)

The use of Eqn. (30) on a system of two spins, s1 and s2, with couplings
k and h as shown in Eqn. (27) provides an instructive example:
Tr e;
fs1 s2g

= Tr ek(s1s2)+h(s1+s2)
fs1 s2g
=

+1
X

+1
X

ek(s1s2)eh s1eh s2

s1=;1 (by2) s2=;1 (by2)

+1
X
=
e| ;ks1e{zhs1e;h} + |eks1e{zhs1eh}
s1=;1 (by2)
s2=;1
s2=+1
k
;
h
;
h
;
k
h
;
h
= (e e e + e e e ) + (e;k e;heh + ek eheh)

| {z }
s1=+1
k
= 2e cosh (2h) + 2e;k :

| {z }
s1=+1

(33)

If k = 0 (no exchange coupling), using Eqn. (33) in Eqn. (30):

Z = Tr e; H = cosh(2h) + 2 = 4 cosh2 h ,
fs1 s2g
19

(34)

and using Eqn. (34) in Eqn. (32) the free energy for this system of two spins
simpli es to:

f = (1=N ) ln Z = 12 (ln cosh2 h + ln 4) = ln cosh h + ln 2 :

(35)

Evidently the ln 2 term in Eqn. (35) is the entropy contribution (F =

U ; TS ), associated with the two possible states per site. To calculate the
magnetization, M , using Eqn. (35), we nd M related to the reduced free
energy via a derivative:
1 sinh h = tanh h
=
M = @f
@h cosh h

(36)

which is the required result for the paramagnetism of a free electron (or ion)
in an imposed eld.

2.4 The two{dimensional tetrahedral gasket

Here we rst introduce the geometry of the tetrahedral gasket, shown in
Figure 6, and then the spin Hamiltonian for the gasket is developed and
the real-space renormalization group (RSRG) procedure is formulated. Using the RSRG we can identify a critical point, produce a phase diagram,
and calculate critical exponents which describe how many thermodynamic
response functions vary near the critical point. The complete calculation for
this model is contained in 9].
To nd the dimension of the tetrahedral gasket, we rst consider the idea
of dimension in general. To calculate the dimension of an object, one can
consider 8] how the area or volume changes with the length of the object's
edges. Let b be the scaling factor and d the dimensionality. When we expand
20

the length of the side of an object by a factor of b, the relationship between

the new volume V 0 and the original volume V is given by

V 0 = bd V :

(37)

When the length of all sides of a regular structure is doubled (b = 2), the
relationship between the new volume and original volume is then

V 0 = 2d V :

(38)

For the tetrahedral gasket, when the octahedral segment is removed from the
tetrahedron, this can be viewed as adding three more tetrahedrons to our base
one, in a sense doubling the length of the side of the original tetrahedron.
0
Thus, with Vgasket
= 4 Vgasket and b = 2, the dimensionality of the tetrahedral
gasket is calculated as dgasket = 2. Therefore the tetrahedral gasket is a full
two-dimensional object.
To utilize the tetrahedral gasket for this statistical mechanical calculation,
we place binary spin variables at the vertices of all the tetrahedrons. Consider
the arrangement in Figure 7 , which can be viewed as the state after the
rst octahedral segment is removed from a base tetrahedron, where the four
tetrahedrons are enclosed by the following four sets of spins:

f s1 s5 s6 s7 g f s2 s6 s9 s10 g f s3 s5 s8 s9 g f s4 s7 s8 s10 g
and the octahedral void is enclosed by the six spins labeled with dots:

fs5 s6 s7 s8 s9 s10g :
The decimation procedure of the RSRG technique consists of summing over
all con gurations of the six interior spins (si = +1 ;1 8i 5 i 10), and
21

allowing the exterior spins, fs1 s2 s3 s4g, to take on di erent con gurations.
For each con guration of spins fs1 s2 s3 s4g, we equate the decimated system to the same con guration on a base tetrahedral system consisting only
of four tetrahedral oriented spins. This decimation gives a set renormalized
couplings in terms of the original (or rather the previous) couplings. The
functions that relate the renormalized couplings to the previous set are the
recurrence relations.
The details of the model are now presented by specifying the couplings
used in the Hamiltonian and by describing the development of the recurrence
relations. For the system displayed in Figure 7, we have the following type
of (even) spin interactions:

k0 : a zero-spin coupling,
k(s1s5) : a nearest-neighbor spin-spin interaction,
l(s1s8) : a second nearest-neighbor spin-spin interaction, and
m(s1s5s6s7) : a four-spin interaction for each tetrahedron.
These even spin couplings are functions of an even number of spin variables.
Later we add the odd couplings, h and p. When the interactions k, l, and m
are used, the spin Hamiltonian for the system in Figure 7 is:

; H = k (s1s5 + s1s6 + s1s7 + s5s6 + s5s7 + s6s7 + s2s6 + s2s9

+s2s10 + s6s9 + s6s10 + s9s10 + s3s5 + s3s8 + s3s9 + s5s8
+s5s9 + s8s9 + s4s7 + s4s8 + s4s10 + s7s8 + s7s10 + s8s10) +

l (s1s8 + s1s9 + s1s10 + s2s5 + s2s7 + s2s8 + s4s6

22

+s4s5 + s4s9 + s3s7 + s3s10 + s3s6) +

m (s1s5s6s7 + s2s6s9s10 + s3s5s8s9 + s4s7s8s10) :

(39)

Compare Eqn. (39) to the Hamiltonian for the base tetrahedron below (with
spins fs1 s2 s3 s4g only):

; H 0 = 2 k00 + k0 (s1s2 + s1s3 + s1s4 + s2s3 + s2s4 + s3s4)

+m0 (s1s2s3s4) :

(40)

The zero spin coupling, k00 , contributes to the reduced free energy. For the
tetrahedral gasket, the outside spins in Figure 7 are shared between two separate decimations, therefore the four exterior spins each contribute a 1=2 k00
term to the Hamiltonian for the base tetrahedron, as shown in Eqn. (40).
The partition function for the system in Figure 7 involves both the interior
set of spins, and exterior set of spins, and is given as
=
Tr
ZN
e; H .
fs1 s2 s3 s4g fs5 s6 s7 s8 s9 s10g
(41)
For the base tetrahedron, only the exterior spins are used. The partition
function for the base tetrahedron is
ZN 0 =
Tr
e; H 0 .
fs1 s2 s3 s4g
(42)
The renormalization-group calculation requires that the partition function
be preserved (ZN = ZN 0 ). Thus we equate Eqn. (41) and Eqn. (42), which
relates the primed and original couplings:
Tr
Tr
Tr
e; H =
e; H 0 .
fs1 s2 s3 s4g fs5 s6 s7 s8 s9 s10g
fs1 s2 s3 s4g
(43)
23

Now, we can calculate the recurrence relations between the interaction

couplings given in Eqn. (39) and Eqn. (40), which give functions for the
renormalized couplings in terms of the previous couplings:

k00 (k l m) k0 (k l m) and m0 (k l m) :

(44)

The details associated with obtaining the recurrence relations is contained

in 9]. With the recurrence relations, the reduced free energy per spin, Eqn.
(32), can be calculated for the tetrahedral gasket. Note that the decimation
procedure (from the spin arrangement in Figure 7 to the base tetrahedron)
reduces the number of spins by a fourth. This factor is not two- fths since
the corners of the tetrahedron are counted as one{half of a spin. Using Eqn.
(43), we expand out Eqn. (32), where N 0 is the number of spins on the decimated lattice, with each spin contributing k00 to the free energy. Note that
k00 is a prefactor in Eqn. (40), and Z 0 the partition function, Eqn. (41), with
the primed couplings used in the corresponding Hamiltonian. This last step
corresponds to setting up for the next decimation by grouping four tetrahedrons (which resulted from the previous decimation) into a con guration as
shown in Figure 7. This, in essence, is the RSRG procedure, which yields a
derivation of the free energy:

fs =
=
=
=

(1=N ) ln ZN = (1=N ) ln ZN 0 = (1=N ) ln e (N=4) k00 ] Z 0(k0 l m0)

(1=4) k00 (k l m) + (1=N )(N 0=N 0) ln Z 0 (k0 l m0)
(1=4) k00 (k l m) + (1=4) (1=N 0) ln e (N 0 =4) k000] Z 00 (k00 l m00 )
(1=4) k00 + (1=4)2 k000 + (1=4)3 k0000 (k(3;1) l m(3;1) ) +

24

which is summarized as:

fs =

1
X
i=1

(1=4)i k0i (k(i;1) l m(i;1) )

(45)

where k0i and the couplings ki and mi are calculated using the recurrence
relations. Note that, k1 in Eqn. (45) is k0(k l m), the rst renormalized
coupling k2 k00 (k0 l m0), is the second renormalized coupling, and k(0) is
just k. The reduced free energy per spin, Eqn. (45), can be used to calculate
the free energy surface for a segment of the km{plane around the origin. A
plot of this surface, using l = 0:1, is shown in Figure 8 . The free energy
surface e ectively shows the value of the free energy, written as fs(k l m),
at each point on the km{plane. This gure has many features in common
with the phase diagram of the model (see 9]).
Now, the Hamiltonian for the base tetrahedron, Eqn. (40), is expanded
to include the odd interaction couplings, h and p. Therefore, the interactions
for the base tetrahedron now include:

k0 : a zero-spin coupling,
k(s1s2+ s1s3+ s1s4+ s2s3+ s2s4+ s3s4) : a nearest-neighbor coupling,
m(s1s2s3s4) : a four-spin (tetrahedral) coupling,
h(s1 + s2 + s3 + s4) : a single-spin ( eld) coupling, and
p(s1s2s3 + s1s2s4 + s1s3s4 + s2s3s4) : a triangular (face) interaction.
The Hamiltonian for the system in Figure 7 is derived in the same way as
in Eqn. (39), and is shown in 9] along with the Hamiltonian for the base
25

tetrahedron. According the RSRG program, we again desire the recurrence

relations (cf. Eqn. (44)):

k00 (k m l h p) k0(k m l h p) m0 (k m l h p)
h0 (k m l h p) and p0 (k m l h p) :

(46)

The novelty associated with the tetrahedral gasket spin model is that it is
a two-dimensional model for which exact recurrence relations can be found,
even in the presence of a non-zero applied eld (i.e., h 6= 0).

2.5 Critical exponents for the tetrahedral gasket

According to the scaling hypothesis, near a critical point, some thermodynamic properties diverge according to strict power laws. For example, the
critical exponents , , , and can be de ned 1, 5] in terms of how
various thermodynamic quantities behave near a critical point. Let Tc be the
critical temperature, and de ne t such that it gives a measure of how far one
is from the critical temperature:

(Tc ; T ) :
Tc

(47)

Then, the equations,

t
t;
t;
t;
B 1=

C
=a
I
26

(48)

describe the temperature dependence near the critical point of the thermodynamic quantities: C , the speci c heat =a, the scaled correlation length M ,
the magnetization , the susceptibility and I , the magnetic moment (with B
the external magnetic eld strength, I is related to M (in Gaussian units) as
R
I = 21c MdV ). If all the exponents are positive, Eqn. (48) says, for instance,
that the speci c heat has a singularity at the critical point and diverges as
t; . Typically, the correlation length and susceptibility also have singular
behavior at the critical point. One of the great theoretical achievements of
this century, shows that, evidently, the critical exponents, de ned in Eqn.
(48), are not necessarily independent of each other. This is manifested in the
so-called scaling laws:
+2 + =2
= ( ; 1)

(49)

where Eqn. (49a) is the Essam-Fisher scaling law 10] (also referred to as
Rushbrooke's scaling law 1]) and Eqn. (49b) is Widom's scaling law. In light
of Eqn. (49), it is natural to seek exponents which are independent 11]. Using
scaling arguments, one such pair of exponents are found to be the thermal
exponent, yT , and the magnetic exponent, yH . The critical exponents are
then given as:
= 2 ; yd
T
1
=
yT
= d ;y yH
T
2
y
H ;d
=
(50)
yT
27

where d is the dimensionality of the model. Therefore, to determine the

critical behavior of a system, it su ces to nd yT and yH .
A critical point for the tetrahedral gasket in the h = p = 0 phase
space (i.e., the km-plane) is an unstable equilibrium point of the renormalization, e.g., if (k m ) is the critical point, then k0 (k m ) = k and
m0(k m ) = m . For the tetrahedral gasket the critical point, (k m ), is
found numerically to be given approximately as:

k ' 0:1525
m ' ;0:0208 :

(51)

Th critical point shown in Eqn. (51) compares favorably to those found

for other two-dimensional models (and the three{dimensional face centered
cubic), as seen by comparing Eqn. (51) with the values given in Table 1 .
Also note that it is standard practice to relate the temperature dependence
of an Ising model to the nearest neighbor coupling, k. Using Eqn. (28), we
can x a relationship between k and T by scaling the exchange coupling J
such that
T = k1 :
(52)
Therefore, in Eqn. (47), we use Tc = 1=k for the critical tempature at which
the model undergoes, in this case, a nontrivial phase transition.
To obtain values for the thermal exponent, yT , and the magnetic exponent, yH , we consider a linearization of the renormalization about the critical
point. From nonlinear system theory 12], we know we can linearize around
a critical point, and if both eigenvalues of the Jacobian are positive we have
an unstable equilibrium point. This unstable equilibrium point is in fact a
28

critical point of the RSRG calculation. For the full tetrahedral gasket model
with odd interactions, the linearization is

2
66 k0
66 m0
66
66 h0
4 0
p

3 0
77 BB @k0 =@k
77 BB @m0 =@k
77 = BB
77 BB @h0 =@k
5 @ 0
@p =@k

@k0 =@m
@m0 =@m
@h0 =@m
@p0 =@m

@k0 =@h
@m0 =@h
@h0 =@h
@p0 =@h

1
@k0 =@p C
C
@m0 =@p C
CC
C
@h0 =@p C
CA
@p0 =@p k

m h p

2 3
66 k 77
66 m 77
66 77 (53)
66 h 77
4 5
p :

To calculate the magnetic exponent, consider the eigenvectors of the Jacobian

in Eqn. (53), evaluated at the critical point (k m h = 0 p = 0). If one
of these eigenvectors is associated with the direction of the renormalization
ows away from the critical point towards larger h, then we use the associated
eigenvalue to nd the magnetic exponent. This same technique is used when
only even spin couplings are present to calculate the thermal exponent, yH .
The use of yT and yH in Eqn. (50) then shows how the thermodynamics of
the system behave near the critical point.

3 Discussion
The statistical mechanical methodology has been employed to study simple models that have energy quanta which behave according to statistics
which are associated with fermions and bosons. The methodology was also
employed to study the phase transition associated with the spontaneous
magnetization of a two-dimensional fractal lattice. In all the models presented, basic quantum mechanical ideas are extended from the microscopic
realm for use in a macroscopic calculation. It should not be surprising that
once we leave the microscopic realm of analysis, we no longer speak of the
29

time-development of a system, but rather talk in terms of thermodynamic

measurements, thus the statistical mechanical methods discussed herein are
appropriate to equilibrium situations only.
The Ising model is concerned with the dynamics of spin-variable systems,
and as such lends itself directly to the modeling of ferromagnetic and antiferromagnetic systems. Evidently the ferromagnetic phase transition of a
magnetic system has much in common with the more familiar liquid-gas phase
transition. Experiment and theory have shown that the relationship between
the ferromagnetic phase transition and the liquid-gas phase transition is such
that the two systems in fact share common values of their critical exponents
1, 13]. These critical exponents describe how the systems order parameter
varies near the critical temperature. An order parameter is a variable that
is non-zero only in the ordered phase of the system. While one systems order parameter might be di erent from another systems order parameter (for
instance, let l and g represent the liquid and gas densities, then ( l ; g )
is the order parameter for the liquid-gas phase transition, while the magnetization, M , is the order parameter for ferromagnetic systems), the apparent
similarities in behavior near the critical point of two such disparate systems
eventually has led to the so-called universality hypothesis. The universality
hypothesis states that if two systems have the same (spatial) dimensionality
and have order parameters with the same dimensionality, then these two systems will have similar values for their critical exponents 1, 13]. This quite
remarkable hypothesis has been applied across a wide range of physical systems, including binary alloys, binary mixtures, super uids, absorbed lms
and of course uids and ferromagnets 13]. The study of phase transitions
30

in general is now commonly grouped into a subject refereed to as critical

phenomena, and the statistical mechanical methodology is currently the best
theoretical tool available to study such critical phenomena.

Appendix
Some basic de nitions and techniques from thermodynamics are reviewed.
Speci c details are included on how thermal measurements can be used to
uncover the molecular topology of simple gases.
An extensive parameter is a parameter whose value in a composite system
is equal to the sum of the values over all subsystems. Examples of extensive
parameters include S , U , V and N (i.e., the entropy, energy, volume, and
particle number). Thus, it is easily seen that S ( U V N ) = S (U V N ).
An intensive parameter is a value that is constant throughout a macroscopic
system, e.g., the temperature T , the pressure P and the chemical potential .
Fundamental relations are functions that contain all possible thermodynamic
information about a system. In the microcanonical formalism, S (U V N ) is
the fundamental relation in the entropy representation, while U (S V N ) is
the fundamental relation in the energy representation. Taking a full di erential in the energy representation we obtain:
!
!
!
@U
@U
@U
dU = @S
dS + @V
dV + @N
dN
(54)
VN
SN
VS
= T dS ; P dV + dN :
From Eqn. (54), we can identify, e.g., that
!
@U
T
@S V N = T (S V N ) :
31

Such a function for the intensive parameter T is referred to as an equation of

state. If all the equations of state are known for a system, the fundamental
relation can be constructed.
In the microcanonical formalism, we can use the entropy (or energy) representation, which assumes that the energy (or entropy) of the system is
xed. There are two principles in the microcanonical formalism that guide
the system to various equilibrium conditions. The entropy maximum principle states that the equilibrium value of any unconstrained internal parameter
is such as to maximize the entropy for a given value of the total internal energy. The energy minimum principle states that the equilibrium value of
any unconstrained internal parameter is such as to minimize the energy for
a given value of the total entropy.
In the canonical formalism we assume the tempature is xed and the
fundamental relation is the Helmholtz Free Energy, F = F (T V N ). To
move between di erent formalisms, the Legendre transformation can be employed. Recall that in classical mechanics, the Lagrangian is a function of
coordinates and velocities: L = L(xi vi). To obtain the Hamiltonian from
the Lagrangian, we employ a Legendre transformation:
N @L
X
;H = L ; @v vk = ;H (xi Pi)
k=1 k

where Pk = @L=@vk are the generalized momenta. To move from the microcanonical to the canonical formalism, note that F = U ; TS , where
T = (@U=@S )V N , therefore the transformation

!
@U
F = U ; @S S
32

eliminates the variable S to give F = F (T V N ). Now consider the full

di erential of F :

!
!
!
@F
@F
@F
dF = @V dV + @T dT + @N dN
= ;P dV ; S dT + dN :

We can readily identify, for instance, that S = ;(@F=@T ), which gives a

Legendre transformation from the canonical to microcanonical formalism:

!
@F
U = F + TS = F ; @T T :
The molar heat capacity, at constant volume, Cv , in the microcanonical
formalism is given as
!
!
T
@s
@S
(55)
Cv = T @T = N @T :
V
V
For a xed particle number, dU = TdS ; PdV , such that at constant volume,
from Eqn. (55),
!
1
@U
Cv = N @T :
(56)
V

The heat capacity is the quasi-static heat ux per mole required to produce
unit increase in the tempature of a system maintained at a constant volume. Heat is evidently a form of energy transfer. Thermodynamics describes
the macroscopic properties apparently a result of interactions involving the
atomic coordinates of a system. The \hidden" atomic modes of a system
thus act as repositories for energy. An energy transfer via the hidden atomic
modes is called heat 1].
Thermodynamics shows that if we can measure how a system stores heat,
we can obtain information about the structure of these hidden modes. An
33

equation of state for the energy of an ideal gas is given as

U = c nRT

(57)

which describes an ideal gas fairly well at high tempature. In Eqn. (57),
R = NA kB = 8:3144 J=mole K is the universal gas constant, n is the
number of moles, and c = 3=2 for monatomic ideal gases. Using Eqn. (57)
in Eqn. (56), we nd that Cv = 3R=2 for a monatomic gas. This value of
Cv is due to a R=2 contribution from each quadratic term in the classical
energy. For a monatomic gas, the energy is due to the kinetic energy of the
molecules, thus the Hamiltonian is
p2 + p2 + p2
U x 2my z :
Classically, for conservative systems, we can write the energy (or Hamiltonian) as a sum of quadratic terms (of the momenta and coordinates):

p2 + q2

where p is the generalized momenta and q a generalized coordinate. The

equipartition theorem of classical statistical mechanics (with N~ = n NA) then
says 1]:
At su ciently high tempature every quadratic term in the energy
~ B to the heat capacity.
contributes a term 1=2 Nk
We can now use the equipartition theorem to investigate the molecular topological structure of some simple gases. Consider the molecular shapes shown
in Figure 9. If we quantify the type of modes each of these molecules possess, as in done in Table 2, we nd that the speci c heat can tell us something
34

about the topology (or shape) of the molecule. In this way, we are basically
measuring how many ways the system can store energy. For a diatomic gas,
the molecule still has three momentum terms in the Hamiltonian for the
molecule as a whole, but for this bonded molecule, we also attribute a vibrational mode which has a potential and momentum term in the Hamiltonian.
Lastly, since we require two angles to specify the orientation, we need two
more terms in the Hamiltonian. When all the modes are totaled up, the
speci c heat for a diatomic gas is given at high tempature as Cv = 7R=2 per
mole.

35

1 That

is, the average number of quanta occupying each state.

2 Table 2 shows that this is due to the vibrational modes of 3N oscillators.

36

References
1] H.B. Callen, Thermodynamics and an Introduction to Thermostatistics
(John Wiley & Sons, 1985).
2] K. Stowe, Introduction to Statistical Mechanics and Thermodynamics
(John Wiley & Sons, 1984).
3] J.R. Hook, H.E. Hall, Solid State Physics (John Wiley & Sons, 1991).
4] J.D. Jackson, Classical Electrodynamics (John Wiley & Sons, 1975).
5] N.W. Ashcroft, N.D. Mermin, Solid State Physics (W.B. Saunders,
1976).
6] A. Goswami, Quantum Mechanics (Wm. C. Brown Publishers, 1992).
7] C. Cohen-Tannoudji, B. Diu, F. Laloe, Quantum Mechanics: Volumes I
and II (John Wiley & Sons, 1977).
8] J.T. Sandefur, \Using Self-Similarity to Find Length, Area, and Dimension," The Am. Mathematical Monthly, 101 (2), 107-120 (1996).
9] J.F. Nystrom, \An Exact Finite Field Renormalization Group Calculation on a Two-Dimensional Fractal Lattice," Int. J. Mod. Phys. C, 11
(2), (2000).
10] G.M. Bell, D.A. Davis, Statistical Mechanics of Lattice Models Volume
1: closed form and exact theories of cooperative phenomena (Ellis Horwood, 1989).
37

11] G.L. Sewell, Quantum Theory of Collective Phenomena (Oxford University Press, 1986).
12] M. Vidyasagar, Nonlinear System Analysis (Prentice Hall, 1993).
13] G. Careri, Order and Disorder in Matter (Benjamin/Cummings, 1984).

38

Figure 1: Entropy is additive.

Figure 2: Speci c heat for a two{state system.
Figure 3: Speci c heat for a crystalline solid.
Figure 4: Bound oscillator model for a crystalline solid.
Figure 5: Sample spin con gurations.
Figure 6: Three levels of the tetrahedral gasket.
Figure 7: Tetrahedral decimation.
Figure 8: Free energy surface using l = 0:1 .
Figure 9: Simple molecular gases.
39

Lattice Critical Value

khoneycomb
0:6585
ksquare
0:4407
ktriangular
0:2747
kfcc
' 0:1
Table 1: Critical value of the spin-spin coupling for common lattices.

40

Type of molecule Di erent modes

pi
Monatomic
3 translational 1=2kB
Diatomic

3 translational
1 vibrational
2 rotational

1=2kB
1=2kB
1=2kB

Triatomic

3 translational
3 vibrational
3 rotational

1=2kB
1=2kB
1=2kB

Linear Triatomic 3 translational

4 vibrational
2 rotational

1=2kB
1=2kB
1=2kB

qi (mode) Total
0 = 3=2kB
! 3=2kB
0 = 3=2kB
1=2kB = kB
0 = kB
! 7=2kB
0 = 3=2kB
1=2kB = 3kB
0 = 3=2kB
! 6kB
0 = 3=2kB
1=2kB = 4kB
0 = kB
! 13=2kB

Table 2: Hidden atomic modes of ideal gases.

Entropy

W 1 states

W 2 states

S2

W 1 W 2 states

S1 + S2

0.4

0.3

0.2

0.1

0
0

0.2

0.4

0.6

0.8

1.2

1.4

C
R
3
2.5
2
1.5
1
0.5

0.5

1.5

2.5

111
000
000
111
000
111
000
111

1
0
0
1
0
1
0
1
0
1
0
1
000
111
0
1
000
111
0
1
000
111
0
1
000
111
0
1
0
1
0
1
0
1
0
1
0
1
1
0
0
1
0
1
0
1
0
1
0
1
0
1
000
111
0
1
000
111
0
1
000
111
0
1
0
1
0
1
0
1
0
1
0
1

(a)

(b)

1
0
0
1
0
1
0
1
0
1
0
1
000
111
0
1
000
111
0
1
000
111
0
1
000
111
0
1
0
1
0
1
0
1
0
1
0
1
1
0
0
1
0
1
0
1
0
1
0
1
0
1
000
111
0
1
000
111
0
1
000
111
0
1
0
1
0
1
0
1
0
1
0
1

S1

S6
S5

S7
S2
S10

S4

S9

S8

S3

m
-2

-1

6
f(k,0.1,m)

4
2
0
-2

-1

0
k

Molecule
Diatomic
Triatomic
Linear
triatomic

Shape
00
11
11
00
00
11
00
11

000
111
111
000
000
111
000
111
00
11
00
11
000
111
0000
1111
00
11
000
111
0000
1111
00
11
000
111
0000
1111
000
111
0000
1111
000
111
0000
1111
00
11
000
111
000
111
0000
1111
00
11
000
111
00
11
000
111
11 11
00
00
000
111
00 11
11
00
000
111
00 11
11
00
000
111
00 11
11
00
000
111