Tabish Qureshi

Statistical Mechanics: Lecture A1 Updated June 2, 2020

The Ising Model

One of the most interesting phenomena in the physics of the solid state is ferromagnetism. In
some metals, e.g., Fe and Ni, a finite fraction of the spins of the atoms align spontaneously in
the same direction to give a macroscopic magnetic field. This, however, happens only when
the temperature is lower than a characteristic temperature known as Curie temperature.
Above the Curie temperature the spins are oriented at random, producing no net magnetic
field. Another feature associated with the is phenomenon is that as the Curie temperature
is approached either from above, or from below, the specific heat of the metal approaches
infinity.
The Ising model is a crude attempt to simulate the physics of a ferromagnetic substance. Its
main virtue lies in the fact that the two dimensional Ising model yield to an exact treatment
in statistical mechanics. It is the only nontrivial example of a phase transition that can be
worked out with mathematical rigour.
In the Ising model the system considered is an array
of N fixed sites form a periodic lattice which could
be 1-, 2- or 3-dimensional. The geometric structure
of the lattice could be anything, cubic, hexagonal
or whatever one may want. Each lattice site has
a spin variable denoted by S i , which is a number
that is either +1 or -1. If one is more inquisitive,
this variable could represent the eigenvalue of the
z-component of a spin− 21 . If S i  +1, the spin is
said to be up, and if it is S i  −1, the spin is said to
be down. A given set of numbers {S i } specifies a
configuration of the whole system. The energy of the system in the configuration specified
by {S i } is defined to be

N
1Õ Õ
E{S i }  − Ji j S i S j − B Si (1)
2
i, j i1

where Ji j denotes the strength of interaction between the i’th and the j’th spin, B denotes an
external magnetic field, which could be present. The factor of 1/2 is introduced to account
for double-counting in unrestricted sum over i and j (i=3,j=8 and i=8,j=3 both represent
the interaction between the 3rd and 8th spin). The quantity Ji j is actually the exchange
interaction between the two magnetic atoms. The magnetic interaction between the two
magnetic atoms is too weak to give rise to ferromagnetism.
A simpler version of Ising model is generally used, where all Ji j s are assumed to be equal,
and each spin interacts only with its nearest neighbours

N
J Õ Õ
E{S i }  − Si S j − B Si (2)
2
<i j> i1

where < i j > represents a sum over only the nearest-neigbours, One can easily see that
the macroscopic magnetic moment for the whole system, for a particular spin configuration,
 Tabish Qureshi

will be given by
N
Õ
M{S i }  Si (3)
i1

Statistical Mechanics of Ising Model

Our aim is to calculate various macroscopic thermodynamic quantities using statistical
mechanics. Statistical mechanics is what connects microscopic physics to theromdynamics.
The canonical partition function can be written as
Õ
Z exp(−βE{S i }), (4)
S1 ,S2 ,...S N
Í
where the summation S1 ,S2 ,...SN denotes sum over all microstates, which happen to be all
possible values all the spins. The canonical density matrix can also be written easily:
1
ρ{S i }  exp(−βE{S i }) (5)
Z
The average value of a quantity, say, A{S i } associated with the Ising system can then be
calculated as
Õ
hAi ρ{S i }A{S i })
S1 ,S2 ,...S N
1 Õ
 A{S i } exp(−βE{S i }), (6)
Z
S1 ,S2 ,...S N

Let us use this equation to write some average quantities of interest. Average energy of
the system is given by
1 Õ
hEi  E{S i } exp(−βE{S i }), (7)
Z
S1 ,S2 ,...S N

This can be cleverly recast in the following form.

1 ∂ Õ
hEi− exp(−βE{S i })
Z ∂β
S1 ,S2 ,...S N
1 ∂
− Z
Z ∂β
∂ log(Z)
− (8)
∂β
Specific heat can then be calculated as

∂ ∂ log(Z)
C− (9)
∂T ∂β
Average magnetization of the system is given by
1 Õ
hMi  M{S i } exp(−βE{S i }), (10)
Z
S1 ,S2 ,...S N
 Tabish Qureshi

One look at equation (2) suggests that this can be recast into the form:

11 ∂ Õ
hMi exp(−βE{S i })
β Z ∂B
S1 ,S2 ,...S N
11 ∂
 Z
β Z ∂B
1 ∂ log(Z)
 (11)
β ∂B

Remembering that the Helmholtz free energy is given by F  −k B T log(Z), the above
relation can be written as
∂F
hMi  − (12)
∂B
Magnetic susceptiblity can then be calculated as

1 ∂2 log(Z)
χ (13)
β ∂B 2
One notices that the quantity of central interest is log(Z). So let us go about calculating it.

N
Õ ©βJ Õ Õ
Z exp ­ S i S j + βB Si ® (14)
ª
2
S1 ,S2 ,...S N « <i j> i1 ¬
Evaluating Z is not easy, because the S i S j term makes sure that the sums over different
S i and S j cannot be carried out independently.

A paradox: breakdown of ergodicity

The way magnetization in a configuration is defined by (3), if one flips every spin, it is easy
to see that the magnetization will change sign, but its magnitude will not change. So for
every value of M , there is another configuration which has magnetization −M . Suppose
there is no external field (B  0). In that case the expression for energy E , given by (2),
tells us that the energy of the system does not change if all the spins are flipped. Of course,
if you leave out flipping even one spin, the energy changes. So, for every configuration with
a magnetization M and energy E , there is another configuration which has magnetization
−M but the same energy E. Since the energy of the two configurations is the same, the
probability of the two configurations, e −βE /Z , is also the same. So when one calculates the
average magnetization using (10), the term for every M is exactly canceled by another term
with magnetization −M . The net result is that average magnetization is zero. We know
that at higher temperatures magnetization for all magnetic materials is zero. However, the
above argument is independent of temperature, and implies that average magnetization
will be zero at all temperatures! However, we do know that ferromagnetic materials show
spontaneous magnetization at low enough temperatures. So where is the catch?
What actually happens is that at low enough temperatures, when all spins are (say)
up, the system cannot wander off to the configuration where all spins are down, just by
random flipping of a few spins. The system is trapped in the mode where the spins are
predominently up. So, there are configurations which the system never attains, or in other
words, there are parts of the phase space which the system never visits. The system is no
longer ergodic! So the time-average of any quantity may not be the same as ensemble
 Tabish Qureshi

average. The ensemble average says that the average magnetization is zero, whereas the
time-average of magnetization is not zero. This indicates a breakdown of ergodicity. But
ergodic hypothesis is a central pillar of statistical mechanics. How does one reconcile the
breakdown of ergodicity with using statistical mechanics for studying phase transitions?
The answer is that one should tread carefully here, keeping in mind the possible breakdown
of ergodicity. For example, if we have a nonzero B , we will not run into the problem of
getting zero magnetization at all temperatures. This paradox teaches us about breakdown
of ergodicity which is always associated with a phase transtion. A phase transition may not
always lead to a breakdown of symmetry, but it will still show breakdown of egodicity.

Mean Field Theory

In the following we will carry out an approximate treatment of the Ising model. Let us rewrite
the energy for the Ising model in a suggestive form:

N
JÕ Õ Õ
E{S i }  − Si Sj − B Si (15)
2
i < j> i i1

where we have split the sum over pairs into a sum over all the i sites and the nearest
neigbours of i , < j > i . The term < j> i S j can be thought to be a local magnetic field,
Í
because of the neighbouring spins, acting on the spin S i . Needless to say that this local
field varies from site to site, because spin states vary from site to site. It depends on the
configuration of nearest neigbour spins of that particular site.
Now we make an approximation that the local field acting on all the sites is the same.
Mathematically this can be written as
Õ
S j  γm, (16)
< j> i

where γ is the number of nearest neigbours of spin S i and m is the average magnetization
per spin of the system. It should be emphasized that the quantity m is yet to be calculated
from the relation m  hMi/N . Using this approximation, the energy of the Ising model
now assumes the following form.

N N
Jmγ Õ Õ
E{S i }− Si − B Si
2
i1 i1
N
Jmγ Õ
−( + B) Si (17)
2
i1

Let us calculate the partition function using this simpler form of the energy. Z now assumes
 Tabish Qureshi

the form
N
!
Õ Jmγ Õ
Z exp β( + B) Si
2
S1 ,S2 ,...S N i1
N  
Õ Ö Jmγ
 exp β( + B)S i
2
S1 ,S2 ,...S N i1
+1
N Õ
Ö Jmγ
 e β( 2 +B)S i

i1 S i −1
N  
Ö Jmγ
 2 cosh β( + B)
2
i1
   N
Jmγ
 2 cosh β( + B) (18)
2

Therefore, log(Z) is given by

  
Jmγ
log(Z)  N log(2) + N log cosh β( + B) . (19)
2

Now we are all set to calculate any quantity. Let us start by evaluating the average
magnetization of the system

1 ∂ log(Z)
hMi−
β ∂B
 
Jmγ
N tanh β( + B) . (20)
2

But hMi/N  m , which leads to

 
Jmγ
m  tanh β( + B) . (21)
2

This is a trancendental equation, which is

not easy to solve. We are looking for spon-
taneous magnetization, i.e., magnetization 1

without an external magnetic field. For B  0 0.8

the equation reduced to

0.6

m  tanh(β Jmγ/2). (22)

0.4

One can look for a solution by plotting the 0.2

L.H.S. and R.H.S. of this equation on the

0
same graph. The points where the two 0 0.2 0.4 0.6 0.8 1
m
curves intersect, will be the solutions of the
above equation. One such plot is shown in
the figure. It is clear that for some values of J, γ, β , a nonzero value of m is a solution.
This shows that Ising model yields spontaneous magnetization even in the mean-field
approximation.
 Tabish Qureshi

Let us try to find out an analytical expression for m in some approximation. can have small
non-zero value. Expanding tanh in a series for small argument and for B  0, we obtain,

m ≈ β Jmγ/2 − (β Jmγ/2)3 /3 (23)

Denoting Tc  Jγ/2k , we can rewrite the above equation as

3
m3
  
Tc Tc
m≈m − (24)
T T 3

One solution of this equation is m  0, and the other one is

 1/2
√ T
 
T
m± 3 1− (25)
Tc Tc

where Tc  γ J/2k can now be identified as a critical temperature. At temperatures close

to Tc , we can approximate the above relation as

 1/2
√

T
m ≈± 3 1− (26)
Tc

This relation implies that at temperature goes below Tc , the magnetization starts from
zero, and grows as (1 − T/Tc )1/2 , even in the absence of an external field. Generally
speaking, the order parameter in a phase transition, close to the transition temperature,
goes as ψ  (1 − T/Tc )β , where β is a critical exponent. Ising model in mean field theory,
yields β  0.5. Real experiments on ferromagnetic materials show that β ≈ 0.33. So, our
simplified model gives a value of β which is not drastically different from the experimental
value. This shows that the Ising model, despite its simplicity, captures the essential physics
of phase transitions.
We will now attempt at determining the behavior of magnetization m at all temperatures
below Tc . Equation (22) can be written as m  tanh(mTc /T). We can obtain m by
numerically finding the zeros of the function tanh(mTc /T) − m , for various values of T .
This can be done through a simple computer program using bracketing and bisection, and
the result is displayed in the figure on the left below. Compare this with the experimental
data of three ferromagnets, iron, nickel and cobalt, shown in the figure on the right below.
Our mean-field curve qualitatively agrees quite well with the experimental data.
1

0.8

0.6
m(T)

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1
T/T_c

A better agreement is expected if the Ising model is solved without approximations, or with
a better approximation.
 Tabish Qureshi

Magnetic susceptibility
Let us now look at the magnetic susceptibility of the Ising model. To do that we should look
at the case B , 0. Equation (21) can be written as
 
mTc B
m  tanh + . (27)
T kT
∂m 

Magnetic susceptibility is defined as χ  ∂B B0
. We differentiate both side of the above
equation with respect to B

∂m ∂
 
mTc B
 tanh +
∂B ∂B T kT
∂m Tc
 
1 1
 + (28)
∂B T
 
kT cosh2 mTc + B
T kT

Putting B  0 on both sides, we get

 
Tc 1 1
χ χ +   (29)
T kT cosh2 mTc
T

For T > Tc , without any external field, the magnetization m is zero. The above equation
then simplifies to yield
1 1
χ · (30)
k T − Tc
This is the well-know Curie-Weiss law, which is valid for temperatures above the transition
temperature. For T < Tc , m has no closed form and hence an analytical expression for χ
cannot be obtained.