Cluster Expansion (a.k.a.

High Temperature Expansion or Hopping

Expansion)
Cluster Expansion is a type of power series used to expand partition function to compute virial
coefficients. The methodology was formulated by Joseph E Mayer and Elliott Montroll in 1941.
Consider a system of interacting monoatomic gas. In the limiting case of no interactions between
particles (atoms in this case) the gas acts as an ideal gas. For a real interacting gas the equation
of state may be expressed in by what is known as a virial expansion. The gas law is thus given by
the following formula

𝑝 𝑁 𝑁2 𝑁3
= + 𝐵2 (𝑇) 2 + 𝐵3 (𝑇) 3 + ⋯ 1
𝑘𝐵 𝑇 𝑉 𝑉 𝑉

where the functions 𝐵𝑖 (𝑇) are the virial coefficients.

For the atoms in the given system of gas the underlying potential energy 𝑈(𝑟) between the atomic
dipoles of length 𝑟,may be modeled by either using the Leonard-Jones potential given by
𝑟0 12 𝑟0 6 2
𝑈(𝑟)~ ( ) − ( )
𝑟 𝑟
or by imposing an infinite potential as given below
∞, 𝑟 < 𝑟0
𝑓(𝑥) = { 𝑟0 6 3
−𝑈0 ( ) , 𝑟 ≥ 𝑟0
𝑟
The Hamiltonian of the gas is calculated as
𝑁
𝑝𝑖2
𝐻=∑ + ∑ 𝑈(𝑟𝑖𝑗 ) 4
2𝑚
𝑖=1 𝑖>𝑗

where 𝑟𝑖𝑗 = |𝑟⃗𝑖 − ⃗𝑟𝑗 | is the distance between the atoms of the gas. The partition function for the
system using the Hamiltonian can be simplified to
1
𝑍(𝑁, 𝑉, 𝑇) = ∫ ∏ 𝑑 3 𝑟𝑖 𝑒 −𝛽 ∑𝑗<𝑘 𝑈(𝑟𝑗𝑘 ) 5
𝑁! λ3N
𝑖

where λ is the thermal wavelength.

We can now define a quantity known as Meyer function which is given by

𝑓𝑖𝑗 = 𝑓(𝑟) = 𝑒 −𝛽𝑈(𝑟) − 1 6

Thus, the partition function can be rewritten and expanded in terms of Meyer function as

1 7
𝑍(𝑁, 𝑉, 𝑇) = ∫ ∏ 𝑑3 𝑟𝑖 (1 + ∑ 𝑓𝑗𝑘 + ∑ 𝑓𝑗𝑘 𝑓𝑙𝑚 + ⋯)
𝑁! λ3N
𝑖 𝑗>𝑘 𝑗>𝑘,𝑙>𝑚
In order to track each of the terms in the sum given in the expression above, a methodology using
diagrams of bonds (known as graphs) among the particles is used. Each Meyer functional term is
denoted using these graphs under the following rules
- Each ‘N’ number of atoms are drawn
- Lines are used to connect the pair of atoms that form the subscript pairs of the Meyer
functions used in the partition function formula.
- Each possible graph only appears once in each summation term of the partition function.

The integral over graph 𝐺 is denoted by 𝑊[𝐺] and the partition function can therefore be written
as
1
𝑍(𝑁, 𝑉, 𝑇) = ∑ 𝑊[𝐺] 8
𝑁! λ3N
𝐺

Almost all cases of graphs that can be drawn will have particles or atoms that are disconnected.
For example, the graph for a single 𝑓𝑖𝑗 𝑓𝑙𝑚 have two pair of atoms connected or have one common
atom for each of the two connections and the rest of the atoms sitting in solitary.

3 4 3 4
𝑓12 𝑓34 𝑓12 𝑓23

1 2 1 2

Figure 1: Graphs of pair of atoms connected for 𝑓𝑖𝑗 𝑓𝑙𝑚

Each of the disconnected components of graphs is called clusters. For example, in case when N=5
we can have many number of clusters, one pair of which is shown in Figure 1 below.

1 5

2 3 4

Figure 2: An example of many graph clusters when N=5

The first cluster in Figure 2 is a 2-cluster (particles 1,2) and the other one is a 3-cluster (particles
3,4,5). In general, if a cluster contains 𝑙 particles it is called a 𝑙-cluster and a graph G can be split
into 𝑚𝑙 𝑙-clusters. Thus, in terms of the number of clusters we can write the following constraint
which will be used later in the analysis
𝑁

∑ 𝑚𝑙 𝑙 = 𝑁 9
𝑙=1

For the analysis we don’t have to draw clusters for all N ~1023 particles. We just need to focus
particular 𝑙-clusters and organize the expansion in a way such that (𝑙+1)-clusters are less
important than 𝑙-clusters. For example, take the case of 3-cluster and there can be 4 different
ways of 3-cluster as given in Figure 3.
 3 3 3 3

1 2 1 2 1 2 1 2
Figure 3: Different types of 3-cluster
These 3-cluster appears in graphs along with many other clusters formed by the remaining N-3
elements. These clusters can be resolved in the partition function, one of the terms of Z must
contain

𝑈3 ≝ ∫ 𝑑3 𝑟1 𝑑3 𝑟2 𝑑3 𝑟3

( )
𝑈3 includes terms 𝑓 2 and 𝑓 3 and the partition function contains factors of all other 𝑙-
clusters.Therefore we can write in general for a 𝑙-cluster
𝑙

𝑈𝑙 = ∫ ∏ 𝑑3 𝑟𝑖 ∑ 𝐺 10
𝑖=1 𝐺∈{𝑙−𝑐𝑙𝑢𝑠𝑡𝑒𝑟}

Each of the graph should only be counted once in the sum and the number of graphs with unique
𝑙-clusters is given by
𝑁!
∏𝑙(𝑙!)𝑚𝑙 11

Thus, the sum over the graphs G is given by

𝑚
𝑁! 𝑈𝑙 𝑙
∑ 𝑊[𝐺] = ∑ ∏ 12
(𝑙!)𝑚𝑙 𝑚𝑙 !
𝐺 {𝑚𝑙 } 𝑙

The resulting partition function is given by

𝑚
1 𝑁! 𝑈𝑙 𝑙
𝑍(𝑁, 𝑉, 𝑇) = ∑ ∏ 13
𝑁! λ3N (𝑙!)𝑚𝑙 𝑚𝑙 !
{𝑚𝑙 } 𝑙

Since the number of clusters and particles are constrained by Eq 9 it is easier to consider partition
function based on the grand canonical ensemble to find the sum above. Since the particle number
is not fixed we can sum over any number of clusters. Grand canonical ensemble is given by
 𝒵(𝜇, 𝑉, 𝑇) = ∑ 𝑒 𝛽𝜇𝑁 𝑍(𝑁, 𝑉, 𝑇) 14
𝑁

Fugacity is defined as 𝕫 = 𝑒 𝛽𝜇 and thus grand canonical partition function can be reduced to
∞ ∞
𝑉 𝑉
𝒵(𝜇, 𝑉, 𝑇) = ∏ 𝑒𝑥𝑝 ( 3 𝑏𝑙 𝕫𝑙 ) = 𝑒𝑥𝑝 ( 3 ∑ 𝑏𝑙 𝕫𝑙 ) 15
𝜆 𝜆
𝑙=1 𝑙=1
where 𝑏𝑙 is given by

𝜆3 𝑈𝑙
𝑏𝑙 = 16
𝑉 𝑙! 𝜆3𝑙

We can see that sum over all pictures got reduced to exponential of sum over all clusters. Using
the grand canonical partition function, we can compute the pressure of the given system.
∞
𝑝𝑉 𝑉
= 𝑙𝑜𝑔 𝒵 = 3 ∑ 𝑏𝑙 𝕫𝑙 17
𝑘𝐵 𝑇 𝜆
𝑙=1
Particle density is calculated as
∞
𝑁 𝕫 𝜕 1
= 𝑙𝑜𝑔 𝒵 = 3 ∑ 𝑙𝑏𝑙 𝕫𝑙 18
𝑉 𝑉 𝜕𝕫 𝜆
𝑙=1
Diving Eq 17 by Eq 18 gives the equation of state

𝑝𝑉 ∑∞𝑙=1 𝑏𝑙 𝕫
𝑙
= ∞ 19
𝑁𝑘𝐵 𝑇 ∑𝑙=1 𝑙𝑏𝑙 𝕫𝑙

Equating Eq 19 with Eq 1(virial expansion of equation of state and 𝑏1 = 𝐵1 = 1) we can deduce

∞
𝐵2 𝐵3
∑ 𝑏𝑙 𝕫𝑙 = [1 + 3
(𝕫 + 2𝑏2 𝕫2 + 3𝑏3 𝕫3 + ⋯ ) + 6 (𝕫 + 2𝑏2 𝕫2 + 3𝑏3 𝕫3 + ⋯ )2 + ⋯ ]
𝜆 𝜆 20
𝑙=1
2 3
× [𝕫 + 2𝑏2 𝕫 + 3𝑏3 𝕫 + ⋯ ]
Expanding the left side of Eq 20
𝐵2 4𝑏2 𝐵2 𝐵3 3 21
𝕫 + 𝑏2 𝕫2 + 𝑏3 𝕫3 + ⋯ = 𝕫 + ( 3
+ 2𝑏2 ) 𝕫2 + (3𝑏3 + + 3) 𝕫 + ⋯
𝜆 𝜆3 𝜆

Equating similar terms (powers of 𝕫) and using the definitions of 𝑈𝑙 and 𝑏𝑙 from Eq 10 and Eq 16
we get the second and third virial expansion coefficients as
𝑈2 1
𝐵2 = −𝜆3 𝑏2 = − =− ∫ 𝑑3 𝑟𝑓(𝑟)
2𝑉 2𝑉 22
𝐵3 = −𝜆6 (4𝑏22 − 2𝑏3 )

Thus, we can derive using cluster expansion the values of virial expansion coefficients of the
equation of state. By equating similar terms in Eq 21 we can find whichever virial expansion
coefficients as required.