1 The Ising model
In the Ising model, the energy of a spin configuration that accounts
for interactions between nearest neighbours and an external magnetic
field can be expressed as follows:
X X
E({si }) = −J si sj − µH si ,
<i,j> i
where < i, j > indicates a summation over nearest neighbors.
(a) What is the probability P ({si }) that a given spin configuration {si }
occurs?
A state ν is characterized by a specific spin configuration {si }. There-
fore, the partition function can be expressed as:
X X X X
Z(N, βJ, βµH) = e−βEν = exp βJ si sj + βµH si .
ν {si } ⟨i,j⟩ i
Note that the summation over {si } is given by:
X X X X
= ... .
{si } s1 =±1 s2 =±1 sN =±1
We find that the probability of a given configuration {si } is:
1 −βE({si })
P ({si }) = e .
Z
(b) Consider the case when J = 0. Show that the canonical partition
function Z(N, βJ, βµH) in this case is given by
Z(N, 0, βµH) = [2 cosh(βµH)]N .
X X
Z(N, 0, βµH) = ... exp [βµH (s1 + . . . sN )]
s1 =±1 sN =±1
X X N X
Y N
Y
βµHs1 βµHsN βµHsi
+ e−βµH
+βµH
= e ... e = e = e
s1 =±1 sN =±1 i=1 si =±1 i=1
N
= [2 cosh (βµH)] .
1
(c) We define the magnetisation as M = µ N
P
i=1 si . Compute ⟨M ⟩ for J =
0 and show that for small external magnetic fields, the system satisfies
Curie’s law, ⟨M ⟩ = CH/T , with C a constant.
We define the magnetization for specific configurations of spins as:
N
X
M =µ si =: M ({si }) .
i=1
The average magnetization, denoted as ⟨M ⟩, is calculated as follows:
N
! " N
#
X 1 X X X
⟨M ⟩ = M ({si }) P ({si }) = µ si exp βµH si
Z(N, 0, βµH) i=1 i
{si } {si }
" N
#
1 X ∂ X
= exp βµH si
Z(N, 0, βµH) ∂ (βH) i
{si }
" N
#
1 ∂ X X
= exp βµH si
Z(N, 0, βµH) ∂ (βH) i
{si }
1 ∂
= Z(N, 0, βµH)
Z(N, 0, βµH) ∂ (βH)
∂ ∂
= log (Z(N, 0, βµH)) = N log (2 cosh (βµH))
∂ (βH) ∂ (βH)
= µN tanh (βµH)
Usignt that tanh(x) ∼ x for |x| ≪ 1, we get
2 µ2 N H CH
⟨M ⟩ ∼ µ βN H = = .
kB T T
µ2 N
Within this model C = kB > 0, indeed ideal paramagnet.
(d) Consider s = N −1 N
P
i=1 si . Write si = ⟨s⟩ + δsi and expand E({si }) to
linear order in δsi . In doing so, show that
1 2
Z(N, βJ, βµH) = exp − βN Jz⟨s⟩ Z(N, 0, βµHmol ),
2
2
with z the coordination number of a single spin. Derive an expression for
the molecular field Hmol and explain why this approximation is called the
mean-field approximation.
The expression for each spin can be written as si = ⟨s⟩ + δsi . Now
let’s consider the first term in E ({si }):
N z N z
X 1X X 1X X
si sj = si sj(i) = (⟨s⟩ + δsi ) (⟨s⟩ + δsj ) ,
<i,j>
2 i=1
2 i=1
j(i)=1 j(i)=1
where z in the limits of sums denotes the number of nearest neighbours.
This can be expanded to:
N z
1X X
⟨s⟩2 + ⟨s⟩δsi + ⟨s⟩δsj + O(δs2i )
=
2 i=1
j(i)=1
Now, focusing on the first term:
N X
X z
⟨s⟩2 = N z⟨s⟩2
i=1 j(i)=1
Additionally, we have:
X z
N X N X
X z
δsi = δsj
i=1 j(i)=1 i=1 j(i)=1
Thus, it follows that:
N X
X z N X
X z N
X N
X
(δsi + δsj ) = 2 δsi = 2z δsi = 2z (si − ⟨s⟩)
i=1 j(i)=1 i=1 j(i)=1 i=1 i=1
This simplifies to:
N
X
= −2zN ⟨s⟩ + 2z si
i=1
Combining everything together, we find:
3
N N
X 1 X 1 X
si sj = N z⟨s⟩2 − N z⟨s⟩2 + z⟨s⟩ si = − N z⟨s⟩2 + z⟨s⟩ si
<i,j>
2 i=1
2 i=1
Neglecting the terms involving δsi δsj , we obtain:
N
1 2
X
E ({si }) = JN z⟨s⟩ − (Jz⟨s⟩ + µH) si
2 i=1
We define Hmol through the relation:
µHmol = Jz⟨s⟩ + µH
Thus, we can express Hmol as:
Hmol = µ−1 Jz⟨s⟩ + H
In this context, each spin is influenced by an external field, as
well as the average magnetic field generated by the surrounding spins.
This explains why this approximation is referred to as the mean-field
approximation, where we have neglected the terms involving δsi δsj .
Within this framework, we find that:
N
X " N
#
1 X
Z(N, βJ, βµH) = exp − βJN z⟨s⟩2 exp µβHmol si
2 i=1 i=1
1
= exp − βJN z⟨s⟩2 Z(N, 0, βµHmol )
2
4
P
(e) Derive from ⟨s⟩ = {si } P ({si })s that ⟨s⟩ = tanh(βµHmol ).
X
⟨s⟩ = s({si })P ({si })
{si }
N
! " #
1 1 XX X X
= si exp βJ si sj + βHµ si
Z(N, βJ, βµH) N i=1 <i,j> i
{si }
" #
1 X 1 ∂ X X
= exp βJ si sj + βHµ si
Z(N, βJ, βµH) N ∂(βµH) <i,j> i
{si }
1 ∂
= Z(N, βJ, βµH)
Z(N, βJ, βµH)N ∂(βµH)
Now, we apply the mean-field approximation:
1 ∂
⟨s⟩ ≈ Z(N, 0, βµHmol )
Z(N, 0, βµHmol )N ∂(βµH)
1 ∂
= log (Z(N, 0, βµHmol ))
N ∂(βµH)
The partition function is given by:
Z(N, 0, βµHmol ) = [2 cosh (βµHmol )]N = [2 cosh (β (Jz⟨s⟩ + µH))]N
Finally, we arrive at the expression:
⟨s⟩ = tanh (βJz⟨s⟩ + βµH) .
(f ) Show that the condition from (e) follows also from F ′ (⟨s⟩) = 0, with
F (s) the Helmholtz free energy of the system. Here a prime denotes a
differentiation with respect to the argument.
In the mean-field approximation, the partition function can be ex-
pressed using the following formula:
1
Z(N, βJ, βµH) = exp − βJN z⟨s⟩2 Z(N, 0, βµHmol )
2
5
From this, we can derive the expression for the Helmholtz free en-
ergy:
1
βF (N, βJ, βµH) = βJN z⟨s⟩2 − N log [2 cosh (βJz⟨s⟩ + βµH)]
2
The condition for the vanishing of the derivative of F with respect to
⟨s⟩
∂F
= 0,
∂⟨s⟩
results in the following condition:
2 sinh (βJz⟨s⟩ + βµH)
βN Jz⟨s⟩ = N βJz
2 cosh (βJz⟨s⟩ + βµH)
Thus, we arrive at the equation:
⟨s⟩ = tanh (βJz⟨s⟩ + βµH) .
(g) Consider the situation when H = 0. Argue that a second order phase
transition occurs when F ′′ (⟨s⟩) = 0. Derive the critical temperature, Tc =
Jz/kB . Evaluate this expression for a cubic lattice in one, two, and three
spatial dimensions. How do your results compare to the exact results?
At high temperatures, where T > Tc , the system behaves like a para-
magnet. This means that the free energy F has a single minimum at
⟨s⟩ = 0. However, at low temperatures, where T < Tc , the system
exhibits magnetization, leading to two degenerate minima in the free
energy F . Consequently, at the critical temperature Tc , F (⟨s⟩ = 0)
transitions from being a global minimum to a local maximum, as illus-
trated in Fig.1.
6
60
40
20
T < Tc
-3 -2 -1 1 2 3
-20
-40
T = Tc
-60
T > Tc
-80
Figure 1. Illustrative plot of the free energy F .
The condition
∂ 2F
=0
∂⟨s⟩2 ⟨s⟩=0; T =Tc
implies that
∂ 1
1= tanh (βJz⟨s⟩) = 2 .
∂⟨s⟩ ⟨s⟩=0; T =Tc cosh (βJz⟨s⟩) ⟨s⟩=0; T =Tc
From this, we find that
βc Jz = 1,
leading to the critical temperature:
Jz
Tc = .
kB
7
For a cubic lattice:
2J
In 1D z = 2 ⇒ Tc = ,
kB
4J
In 2D z = 4 ⇒ Tc = ,
kB
6J
In 3D z = 6 ⇒ Tc = ,
kB
For cubic lattice, z = 2d, where d denotes spatial dimension.
The higher the dimensionality, the more stable the magnetized state
becomes. The exact results are kBJT = 0, 2.27, 4, for 1D, 2D, and 3D,
respectively. We observe that there is no phase transition in 1D.
(h) Show that close to the critical point ⟨s⟩ ∼ (Tc − T )1/2 . Note that the
mean-field critical exponent is 1/2.
In subsections e) and f), we derive the following formula:
e2βJz⟨s⟩ − 1
⟨s⟩ = tanh (βJz⟨s⟩) =
e2βJz⟨s⟩ + 1
Starting from the above equation, we can rearrange it as follows:
⟨s⟩e2βJz⟨s⟩ + ⟨s⟩ = e2βJz⟨s⟩ − 1
This simplifies to:
(1 − ⟨s⟩) e2βJz⟨s⟩ = 1 + ⟨s⟩
From here, we can express e2βJz⟨s⟩ as:
1 + ⟨s⟩
e2βJz⟨s⟩ =
1 − ⟨s⟩
Next, we derive the expression for β:
1 1 + ⟨s⟩
β= log
2Jz⟨s⟩ 1 − ⟨s⟩
8
Near the transition point, we assume ⟨s⟩ is small, allowing us to
expand log(1 ± ⟨s⟩):
⟨s⟩2 ⟨s⟩3 ⟨s⟩2 ⟨s⟩3
1 + ⟨s⟩
log ∼ ⟨s⟩ − + − . . . − −⟨s⟩ − − − ...
1 − ⟨s⟩ 2 3 2 3
This results in:
⟨s⟩3
1 + ⟨s⟩
log ∼ 2⟨s⟩ + 2 + ...
1 − ⟨s⟩ 3
Thus, we find:
⟨s⟩3
1
β= ⟨s⟩ + + ...
Jz⟨s⟩ 3
Rearranging yields:
1
βJz = 1 + ⟨s⟩2 + . . .
3
This leads to:
1 2 1
⟨s⟩ = Jz(β − βc ), where βc = .
3 Jz
Finally, we obtain the scaling relation:
⟨s⟩ ∼ (Tc − T )1/2
Here, 1/2 is identified as the critical exponent.
