Statistical Physics.

Solutions Sheet 10.

Exercise 1.

HS 2013
Prof. Manfred Sigrist

Condensation and crystallization in the lattice gas model.

The lattice gas model is obtained by dividing the volume V into microscopic cells which are
assumed to be small such that they contain at most one gas molecule. The result is a square
lattice in two dimensions and a cubic lattice in three dimensions. We neglect the kinetic energy
of a molecule and assume that only nearest neighbors interact. The total energy is given by
X
ni nj
(1)
H =
hi,ji

where the sum runs over nearest-neighbor pairs and is the nearest-neighbor coupling. There
is at most one particle in each cell (ni = 0 or 1). This model is a simplification of hard-core
potentials, like the Lennard-Jones potential, characterized by an attractive interaction and a
very short-range repulsive interaction that prevents particles from overlapping.
In order to study the case of a repulsive interaction, < 0, we divide the lattice into two
alternating sublattices A and B. For square or cubic lattices, we find that all lattice sites A only
have points in B as their nearest neighbors.

Figure 1: Schematic view of the lattice gas model.

(a) First, show the equivalence of the grand canonical ensemble of the lattice gas model with
the canonical ensemble of an Ising model in a magnetic field.
Solution.

We consider the grand canonical Hamiltonian

X
X
H N =
ni nj
ni .

(S.1)

hi,ji

By introducing Ising spins si through the relation

ni =

1
(1 + si ) ,
2

si = 1 ,

(S.2)

we arrive at an Ising model

H N = J

si sj h

X
i

hi,ji





si h J NL = HI h J NL
2
2

(S.3)

with

, h= + .
(S.4)
4
4
2
Here, denotes the coordination number (number of nearest neighbors) and NL is the total number of
lattice sites. The grand partition function Z = Tr [exp[(H N )]] of the lattice gas is thus related to
the canonical partition function ZI = Tr [exp(HI )] of the Ising model through
J=

ZG = ZI e ( 8 + 2 )NL
with the relations (S.4) for the exchange coupling J and the magnetic field h.

(S.5)

(b) Introduce two mean-field parameters mA and mB and adapt the mean-field solution of the
Ising model discussed in Sec. 5.2 of the lecture notes for these two parameters. What are
the self-consistency conditions for mA and mB ?
Solution.

The Hamiltonian of the Ising model is

X
X
HI = J
si sj h
si .

(S.6)

hi,ji

We introduce the mean-field parameters mA and mB , which are defined as

mA = hsi iiA ,

mB = hsj ijB .

(S.7)

si = mA + i := mA + (si mA ) ,

(S.8)

Now we can write for i A

where we assume i to be small. The case j B is analog.
Now we can expand the Hamiltonian
X
X
HI = J
(mA + i )(mB + j ) h
si
i

hi,ji

= J

(mA mB + mB i + mA j + i j ) h

si

hi,ji

[mA mB + mB (si mA ) + mA (sj mB )] h

=
=

si

hi,ji

N
JmA mB J
mB si J
mA sj h
2
iA
jB
X

(S.9)

si

X
X
N
JmA mB
(JmB + h)si
(JmA + h)sj ,
2
iA
jB

where we used that nearest neighbors always belong to different sublattices and neglected the product i j .
We find that the two sublattices A and B behave as paramagnets in the effective fields
hA
eff = JmB + h ,

hB
eff = JmA + h .

(S.10)

The partition function of a paramagnet was already discussed previously, so the partition function of this
mean-field Hamiltonian is

iN/2 h

iN/2

 h
ZI = exp 12 N JmA mB 2 cosh hA
2 cosh hB
.
(S.11)
eff
eff
This immediately leads to the Helmholtz free energy

i
h
io
N
1n h
B
FI (, h, N ) =
JmA mB
log 2 cosh(hA
.
eff ) + log 2 cosh(heff )
2

(S.12)

The self-consistent solutions are given by the local minima of the free energy. The conditions are therefore
FI
=0
mA
FI
=0
mB

h
i
mB = tanh hA
eff
h
i
mA = tanh hB
eff .

(S.13a)
(S.13b)

(c) Use your results from parts (a) and (b) to calculate the grand potential for the lattice gas
and determine the self-consistency relations for the two mean-field parameters A = hni iiA
and B = hni iiB .

Solution. We use the mean-field approximation (S.12) derived in part (b) and the relations (S.4) in order
to write the grand potential


1

(, , NL ) = log ZG = FI (, h, NL )
+
NL

8
2

 

NL
+ +
(2A 1)(2B 1)
(S.14)
=

2
4
4




 


(A + )
+ log 2 cosh
(B + )
,
log 2 cosh

2
2
where we used the relation = 12 (1 + m). Here, the effective magnetic fields (S.10) are replaced by
hA,B
21 (B,A + ) .
eff

(S.15)

We can now reformulate the self-consistency equations (S.13) for the lattice gas by inserting the relations (S.15). Using artanh x = 21 log[(1 + x)/(1 x)] for x [1, 1], we obtain the two relations
=

1
A
1
B
log
B = log
A ,

1 A

1 B

(S.16)

which can also be written in the form

1
,
1 + e(B +)
1
.
B =
1 + e(A +)

A =

By inserting Eq. (S.17b) into Eq. (S.17a), we obtain the single condition
1




+
.
A = 1 + exp
1 + exp ((A + ))

(S.17a)
(S.17b)

(S.18)

In the following we will use the mean-field solution of the lattice gas model in order to discuss
the liquid-gas transition for an attractive interaction > 0.
(d) Argue, why in this case the mean-field results can be simplified as the two densities must be
equal, A = B = . Use your knowledge of the Ising model to define a critical temperature
Tc , below which there are multiple solutions to the self-consistency equations, and discuss
the solutions of for temperatures above or below Tc . Define also the critical chemical
potential 0 corresponding to h = 0 in the Ising model and use this for a distinction of
cases.
Solution.

The two self-consistency equations (S.17) are of the mathematical form

a = (b)

b = (a) ,

(S.19)

where the function is given by

1
.
(S.20)
1 + e(x+)
It is easy to see that for > 0 this function is monotonically increasing, while it is decreasing for < 0.
(x) =

Now if we assume b > a, this implies f (b) > f (a). This immediately leads to a contradiction, as a =
f (b) f (a) = b > a. The same contradiction follows for b < a. Therefore, for > 0 there are only
symmetric solutions A = B for the self-consistency equations and we can simplify the whole treatment
by just omitting the second mean-field parameter altogether.
From Eq. (S.4) we see that h = 0 corresponds to = /2 =: 0 . For this case we can use the knowledge
about the magnetic transition in the zero-field Ising model. In particular, there is a critical temperature
kB Tc = /4 = 0 /2 below which there exist two degenerate solutions.
3

1
liquid

0.8

> 0

l (T )

0.6
= 0

0.4
g (T )
0.2

< 0
gaseous

0
0

0.5

1
T /Tc

1.5

Figure 2: The density as a function of temperature T for different values of the chemical potential .

In the lattice gas, these solutions correspond to the liquid and to the gaseous phase and we will denote
them by l (T ) and g (T ), respectively (see Fig. 2). The third solution of Eq. (S.17) for = 0 , namely
= 1/2, is only stable above Tc .
In the general case, there is a unique solution of Eq. (S.17) for T > Tc while for T Tc there are
three solutions in the neighborhood of = 0 = 2kB Tc but only one minimizes (see Figs. 2 and 3).
The solution with d/d > 0 is stable or metastable while the solution with d/d < 0 is unstable and
corresponds to a local maximum of the grand potential . Thus, for T < Tc , the density (T, ) jumps at
0 reflecting the first-order liquid-gas transition (see Fig. 3).

(e) Find the equation of state p = p(T, ) or p = p(T, v) and discuss the liquid-gas transition
in the p v diagram. Thereby, v = 1/ is the specific volume. Compare with the van der
Waals equation of state:



a

p+ 2
v b = kB T .
v
Hint. For the lattice gas, the volume is given by the total number of lattice sites, NL .
Solution.

The pressure is given by

(, , NL )
NL


 

2
1

( ) log 2 cosh
+
,
=
2
2

2
2

p(, ) =

(S.21)

where we used Eq. (S.14). For (, ) g () and (, ) l () we can simply insert Eq. (S.16) into
the above equation and obtain

1
p(T, ) = 2 log(1 )
(S.22)
2

or in terms of the specific volume v = 1/

p(T, v) =

1
1
kB T log(1 ) .
2 v2
v

(S.23)

But for g () (, ) l () there is coexistence of the liquid and the gas. We have to set = 0 and
= g,l (T ) in Eq. (S.21) (this corresponds to the Maxwell construction) leading to a constant pressure!
This is shown in the p v diagram Fig. 4.

1
0.8

0.6

T = 2Tc
T = Tc
T = 0.5Tc

instable

0.4
0.2

metastable
stable

0
!3

!2.5

!2
[kB Tc ]

!1.5

!1

Figure 3: The density as function of the chemical potential for different temperatures. For T < Tc there is a
jump in at = 0 = 2kB Tc .

We can rewrite the van der Waals equation of state as follows:

p(T, v) =

1
+ kB T
,
v2
v b

The elementary volume of the gas (hard core volume) b equals 1 in our model. Comparing this with
Eq. (S.23), we see that the first term is identical (where a
= /2), whereas the second term diverges
either linearly (van der Walls) or logarithmically (our model) with v 1. This different behavior is
present in the limiting case of high density and can be attributed to the short-range difference of the
potential for the discrete lattice gas model and the continuous van der Waals gas.

(f) Find the phase diagram (T p diagram). Determine the phase boundary (T, pc (T )) and,
in particular, compute the critical point (Tc , pc (Tc )).
Solution.

The critical pressure is given by Eq. (S.22) for = 0 = 2kB Tc and = g,l (T )
pc (T ) = 2kB Tc 2g,l (T ) kB T log(1 g,l (T )),

(S.24)

as shown in Fig. 5. In particular, for T = Tc we have g,l (Tc ) = 1/2 and

pc (Tc ) =

kB Tc
(log 4 1) .
2

(S.25)

Instead of the liquid-gas transition, which we have observed for an attractive interaction > 0,
a crystallization transition (sublimation) can be observed for nearest-neighbor repulsion, < 0.
In this case, we will find that the two mean-field parameters are different, A 6= B , below some
critical temperature Tc .
(g) Discuss the solutions above and below the critical temperature for < 0. Plot the densities A and B , as well as the average, (A + B )/2 for both attractive and repulsive
nearest-neighbor interaction at low temperature, T < Tc . Interpret the result in terms of
compressibility.

0.4
0.35
T > Tc

p(T, v) [kB Tc ]

0.3
0.25

T = Tc
critical point

0.2

T < Tc

0.15
0.1

Coexistence

0.05
0

10

10

Figure 4: The isotherms p(T, v). The shaded region denotes the region of liquid-gas phase coexistence.

Phase diagram of the lattice gas model

pkB TC
0.20

0.15

Liquid

0.10

0.05

Gas
0.2

0.4

0.6

0.8

1.0

TTC

Figure 5: p-T phase diagram of the lattice gas model. The two phases coexist when = 0 and T < Tc
(equilibrium line). Above Tc there is only one phase (a single density for a given pressure).

Solution. Below the same critical temperature kB Tc = ||/4 as for an attractive interaction and in a
certain range [0 , 0 + ] around 0 = /2, we find three different solutions for the selfconsistency relations (S.17). There are two degenerate asymmetric solutions A 6= B , which are related
by A B , B A , and one symmetric solution A = B .
The range is defined by the condition
0 ()|()= < 1

(S.26)

where () is the function in the self-consistency equations defined in Eq. (S.20). This can be understood
by looking at the plot of (A ) and (B ) shown in Fig. 6. As () > 0, there have to be two asymmetric
solutions whenever 0 () < 1 at the symmetric solution. By inserting into Eq. (S.26) and solving for ,
one obtains
r



1
1+
4
=
+ log
,
= 1+
.
(S.27)
2

The asymmetric solutions, which are generally lower in energy, correspond to a crystal structure, where
(at T = 0) one of the sublattices is occupied while the other one is empty.
The densities for attraction and repulsion are shown in Fig. 7. While for a nearest-neighbor attraction the
densities of the sublattices are identical, there is a symmetry-broken phase for nearest-neighbor repulsion.
The compressibility can be written as
T =

1
,
2

(S.28)

1.0

0.8

0.6

0.4

0.2

0.0
0.0

0.2

0.4

0.6

0.8

1.0

Figure 6: Plot of (A ) and (B ) for [0 , 0 + ]. The two graphs cross at the symmetric solution
in a way that there must be two additional crossings. The condition is given in Eq. (S.26).
1.0
0.8
A H<0L

0.6

B H<0L

0.4

H<0L
H>0L

0.2
0.0
-3

-2

-1

Figure 7: Densities on the two sublattices for attractive ( > 0) and repulsive ( < 0) nearest-neighbor interaction
at T = 0.5 Tc . The thick lines show the average densities, the dashed and dotted lines the densities of the two
sublattices.

see Sec. 1.5.2 in the lecture notes. The crystallization transition for < 0 is of second order (except at
T = 0, where it is of first order). For the average density, which is related to the total particle number,
there exists a plateau around 0 . On this plateau the compressibility is small (T = 0 for T = 0), see
Fig. 8. This indicates that it costs a lot of energy to add additional particles, as one sublattice is almost
completely filled (so no additional particles fit in) while it is very difficult to add particles to the second
sublattice due to the repulsive interaction.
The liquid-gas transition for > 0 is of first order. Therefore, there is a jump in the density (see Fig. 7)
which is related to a diverging compressibility. The compressibility in the liquid phase is strongly reduced
compared to the gaseous phase.
For both transitions, the compressibility vanishes for large chemical potentials, where the lattice is almost
completely filled. In contrast, at low chemical potentials, the lattice is almost empty and the compressibility
is large due to the factor 2 in Eq. (S.28).

104

104

100
T

100
1

1
0.01

0.01
-1

10-4
-2.5

-2.0

-1.5

-1.0

-0.5

0.0

0.5

Figure 8: Compressibility for a nearest-neighbor repulsion < 0 at T = 0.2 Tc (left) and for a nearest-neighbor
attraction at T = 0.5 Tc (right). The second-order phase transitions for the crystallization are clearly visible
as jumps in the compressibility. Around 0 there is a range where T 0, indicating the crystallization. The
diverging compressibility at the condensation for > 0 is not shown in the plot. However, the compressibility is
decreased by a factor of 100 at the transition from gas to liquid.