Physics 127b: Statistical Mechanics

Lecture 2: Dense Gas and the Liquid State

Mayer Cluster Expansion
This is a method to calculate the higher order terms in the virial expansion. It introduces some general
features of perturbation theory in many body systems, for example the use of diagrams to represent terms in
the series, and the topology of diagrams to understand which ones have the correct extensive behavior that
are therefore are the ones to be retained. The Mayer cluster expansion also illustrates the intricacy of such
methods, and I will only sketch the results. Pathria has a fairly terse discussion. The best reference for a
sufficiently detailed account to be understandable, but not so dry as to be indigestible, is chapter 4 in States
of Matter by Goodstein. Other references are chapter 12 of Statistical Mechanics by Mayer and Mayer (on
reserve), and a pedagogical paper by Salpeter in Annals of Physics 5, 185 (1958)—the first part of this paper
is available as a handout on the website.
The configuration integral is written
Z Z P
Z Z
−β i<j u(|rEi −Erj |).
ZN = . . . d r e 3N
= . . . 5i<j (1 + fij ) d 3N r (1)

with fij = e−βu(rij ) − 1 with the idea that the effect of fij is small, either because the potential is small, or
because a large potential only acts over a small volume, as in the previous lecture. In this case we can expand
the product in ZN  
Z Z X X
  3N
ZN = ... 1 + fij + fij fkl + · · · 
  d r, (2)
i<j i<j,k<l
i,j 6 =k,l

where you can see what terms are in the multiple sums by thinking of all, unrepeated, interactions (e.g. 1 − 2
and 1 − 3 is included, but 1 − 2 and 1 − 2 is not). We then want to take logs to get thermodynamic quantities,
e.g. in the canonical ensemble  
1 1
A = −kB T ln ZN , (3)
N ! λ3N
and in the microcanonical ensemble
"∞ #
PV X 1  z N
= ln Q = ln ZN . (4)
kB T N =0
N ! λ3

Note that the latter expression gives a convenient derivation of the virial expansion, since small density
corresponds to small z. We will have to expand ln(1 + a1 z + a2 z2 · · · ) in small z and then relate ρ to z via
z ∂ ln Q
ρ= . (5)
V ∂z
The key idea is to represent the terms diagrammatically. This is generally a very useful approach in
complicated perturbation expansions. Often simplifications of the results can be expressed in terms of the
topology of the diagrams, such as the linked cluster expansion below. In the Mayer cluster expansion, each
particle index is denoted by a dot or small circle, and each fij by a link between the dots i and j (see Fig.
(1)). Each diagram corresponds to a contribution to the expansion of ZN (with each term in the sums in Eq.
(2) considered as a separate contribution), and there is an algebraic expression that can be associated with
each diagram, as listed in the caption. The configuration integral ZN is given by the sum of all distinct N-dot

1
 (a) 1 3 (b) 1 3 (e) 1 3 5 7
=
2 4 2 4 2 4 6 8

(c) 1 3 (d) 1 3 1
x 3 x 5 x 6 7 x 8
2 4 2 4 2 4

R
Figure
R 1: Some
R diagrams in the R Mayer cluster expansion at fourth order representing: (a) 1 d 12 r; (b)
f12 d 12 r; (c) f34 d 12 r; and (d) f12 f13 d 12 r (with d 12 r meaning d 3 r1 d 3 r2 d 3 r3 d 3 r4 . Note that (b) and (c)
lead to the same values of the integral, and so are equivalent (but must be counted separately). Panel (e)
shows an eighth R order diagram demonstrating
R R the factorization
R property:R the value represented is the product
of the terms f12 f14 d 3 r1 d 3 r2 d 3 r4 , d 3 r3 , d 3 r5 , f67 d 3 r6 d 3 r7 , and d 3 r8 .

(a) (b) (c)

Figure 2: Linked clusters: (a) and (b) show the only irreducible, second and third order diagrams; (c) is a
reducible third order diagram.

graphs. Note that in the counting, relabelled, similar-looking graphs such as in Fig. (1b,c) are counted as
distinct, although they will give the same expressions once the integrals are done. Since the integrals over
the coordinates i, j for dots linked by fij gain contributions only from small rij , we think of the linked dots
as particle clusters—there is a contribution to the configuration integral from all the various clusters that can
form in the gas. Note the factorization property of unlinked clusters, Fig. (1e). Furthermore, each cluster has
one free integration variable, giving a factor of V , with the remaining integrations over difference coordinates
involving f giving factors of atomic volumes, so that the volume dependence of the whole graph is V L with
L the number of linked clusters.
The diagrams are useful even for direct evaluation of low order expansions: as an example you will
investigate the direct expansion up to third order in the homework. However the full power of the method
comes because the complications introduced by the logarithm in forming the thermodynamic quantities can
actually be eliminated by a general result that is most easily stated in terms of the diagrams, and in fact
the expansions for the thermodynamic quantities are actually easier than for ZN . This is the linked cluster
expansion.
The linked cluster results take the form, for example for the grand canonical potential,
X
ln Q = (counting factor) × (cluster integral) (6)
linked
clusters

where the summation is restricted to linked cluster, where a linked cluster is one with at least one fij line

2
connecting each part of the diagram. The counting factor takes into account topologically similar diagrams
that give the same integral, and numerical factors depending on the order of the term. Notice that we would
guess that only linked clusters occur in the thermodynamic quantities, since other terms do not give the right
extensive scaling with V . Finding out what the counting factor is, however, involves considerable careful
work, which the references describe.
From this the virial expansion can be formed, and introduces a new topological property (irreducibility)
of retained diagrams: " #
X∞
l l
P = ρkB T 1 − βl ρ (7)
l=1
l + 1
with
1
βl = × (sum over all irreducible l + 1 clusters), (8)
l!
where an irreducible cluster is one that cannot be cut into two pieces by cutting at a single dot. Reducible
diagrams factor into simpler ones, since the integrals only depend on difference coordinates, e.g. for the
diagram in Fig. (2c)
Z Z Z Z
f (r12 )f (r13 )dr12 dr13 = f (r12 )dr12 f (r13 )dr13 . (9)

(The particular form of Eq. (7) comes from developing the expansion for ln γN with γN = ZN −1 /ZN .) The
l = 1 term in the virial expansion (Fig. (2a)) is what we have found before; the l = 2 term in Eq. (7) depicted
in Fig. (2b) is
Z
2ρ 2 1
l=2: − (eβu(r12 ) − 1)(eβu(r13 ) − 1)(eβu(r23 ) − 1)d 3 r12 d 3 r23 (10)
3 2!

The Liquid State

The classical liquid state is one of the most difficult states of matter to understand quantitatively, since there
are no small parameters to use as a crutch (temperature, kinetic, and potential energies are all comparable).
Quantum liquids are actually easier to understand, based on expansions in the number of thermal excitations
at low temperatures (and the fact that the noninteracting gases already have interesting properties). A low
order truncation of the virial expansion is not much use in understanding the strongly interacting classical
liquid. There are two general classes of methods that are used to go beyond this. Neither of these methods are
controlled approximations in general (nor are they particularly imaginative), but rather represent a last-ditch
hope in the absence of really good methods. In some cases, e.g. with a long range Coulomb interaction, these
methods may become exact in some limits.
It is hard to say much about this topic without spending more time on it than the successes warrant. I
will just mention a few of the ideas. The best discussion I know of is §4.5 of Goodstein’s States of Matter.
If we focus on the deriving the equation of state, the calculation can be reduced to finding the radial
distribution function gN (r). Remember that for the nearly ideal gas the radial distribution function is
gN (r) ' e−βu(r) . (11)
We can phrase the calculation in terms of calculating W (r) defined by the correction to this
gN (r) = e−βu(r) eW (r) (12)
or
ln gN (r) + βu(r) = W (r). (13)
Usually, approximate expressions for W (r) are derived as functions or functionals of gN (r) leaving algebraic
or integral equations to be solved for gN .

3
Infinite summation
If the low order Mayer (virial) expansion is of no use, we might try summing an infinite number of diagrams.
The diagrammatic expansion for W is given by
Z Z Y
W (r) βu(r) V2
e =e gN (r) = ··· (1 + fij )d 3 r3 . . . d 3 rN (14)
ZN i<j
i,j 6 =1,2

where the product on the right hand side does not include (1 + f12 ) which gives the term eβu(r) on the right
hand side. The diagrammatic expansion is similar to the Mayer expansion, except there are now two fixed
points rE1 , rE2 that we do not integrate over and provide “anchors” for the diagrams. It is not possible to sum
all the diagrams, so various infinite classes are summed, chosen by some simplicity of topological form,
or intuition of important physical content. The diagrams left out are in principle just as large as the ones
retained, and so getting a good answer (for a particular result) is more an art than a science. Two examples
are the “hypernetted chain” and the “Percus-Yevick” approximations (see Goodstein §4.5b for more details).
The details are quite intricate, and I leave you to consult that reference if you are interested.

Truncated Hierarchies
In this method a differential equation for gN (r) is derived. This equation depends on the next higher order
density correlation functions. These in turn can be related to higher ones, and an exact hierarchy of equations
can be derived. This is then truncated at some order, by making a simple factorization approximation of a
high order correlation function, i.e. supposing that it factorizes into products of lower order ones. The hope
is that high-order correlation functions do not have much physical content. Unfortunately, in most cases it
seems that they do contain important physics, but this is now obscured in the complexity.
We define an effective “potential of mean force” φN (cf. the nearly ideal gas expression) in terms of the
exact (unknown) gN
φN (r) = −kB T ln gN (r) (15)
and try to calculate the effective mean force

∂φN (Er1 , rE2 ) kB T ∂gN (Er1 , rE2 )

− = . (16)
∂ rE1 gN ∂ rE1

(Although φ and g only depend on the difference rE1 − rE2 the notation in terms of individual position is
convenient here.) This can be interpreted as the force on the particle 1 at rE1 holding particle 2 fixed at rE2 , but
averaging over all positions of the other particles in the potentials of these two and their own interactions.
Using R 3N P
V 2 d r i6=j δ(Er − rEi )δ(Er 0 − rEj )e−βU
gN (Er , rE0 ) = R (17)
N (N − 1) d 3N r e−βU
gives R
∂φN (Er1 , rE2 ) d 3 r3 . . . d 3 rN ∂U/∂ rE1 e−βU
− = R . (18)
∂ rE1 d 3 r3 . . . d 3 rN e−βU
The potential energy derivative ∂U/∂ rE1 contains the term ∂u(r12 )/∂ rE1 involving the fixed particle and N − 2
terms of the form ∂u(r1j )/∂ rE1 with j 6= 1, 2. Equation (18) can then be written
R 3
∂φN (r12 ) ∂u(r12 ) d r3 gN(3) (Er12 , rE13 ) ∂u(r13 )/∂ rE1
= +ρ (19)
∂ rE1 ∂ rE1 gN (Er12 )

4
where gN(3) is the three-particle distribution function
* +
V2 X
gN(3) (Er , rE0 ) = 0
δ(Er + rEi − rEj )δ(Er + rEi − rEk ) (20)
N (N − 1)(N − 2) i6 =j 6 =k

which can be used to define a third order effective potential φN(3) = −kB T ln gN(3) . This procedure can be
repeated to define the hierarchy. We can truncate the hierarchy at second order by making the superposition
or factorization approximation
φN(3) (Er12 , rE13 ) ' φN (Er12 ) + φN (Er23 ) + φN (Er13 ) (21)
gN(3) (Er12 , rE13 ) ' gN (Er12 )gN (Er23 )gN (Er13 ) (22)
to give Z
∂φN (r12 ) ∂u(r12 ) ∂u(r13 )
= + ρ d 3 r3 gN (r23 )gN (r13 ) . (23)
∂ rE1 ∂ rE1 ∂ rE1
This equation is usually expressed in the integrated form. First introduce the function
Z r
du(r 0 )
E(r) = gN (r 0 )dr 0 . (24)
∞ dr 0

Equation (23) reduces to

Z
∂φN (r12 ) ∂u(r12 ) ∂E(r13 )
= +ρ d 3 r3 gN (r23 ) . (25)
∂ rE1 ∂ rE1 ∂ rE1
The integral can then be expressed in the from
Z
φN (r12 ) = u(r12 ) + ρ d 3 r3 gN (r23 )E(r13 ) + C. (26)

The integration constant C is evaluated by taking particle 2 to infinity for fixed rE1 , rE3 (φN (r12 ) → 0, u(r12 ) →
0, gN (r23 ) → 1). The expression can then be written in the form known as the Yvon-Born-Green equation
Z
u(r) ρ  
ln gN (r) + =− E(rE − rE0 )[gN (r 0 ) − 1]d 3 r 0 . (27)
kB T kB T
Notice this is an integral equation for gN in the form of Eq. (13).

∗
Quantum Cluster Expansion
This is an advanced topic to show how cluster expansions can be done for a dilute quantum gas. You can
read it or skip it as you like. The more interesting aspects of quantum systems, such as Bose-Einstein
condensation cannot be treated using this formalism, and the methods introduced later in the term are more
useful in general. You can read Pathria §9.4-6 for more details.
The starting point is
ZN (V , T )  z N
X∞ X∞
Q= QN (V , T )zN = (28)
N=0 N =0
N ! λ 3

with z = eβµ the fugacity, and where the second expression just defines the quantity ZN (which in the
classical case is the configuration integral). In the V → ∞ we get
∞
P ln Q 1 X l
= = 3 bl z (29)
kB T V λ l=1

5
with
Z1
b1 = =1 (30)
V
1
b2 = (Z2 − Z12 ) (31)
2!λ3 V
..
. (32)
Note that ZN scales as V N , but the combination Z2 − Z12 appearing in b2 must scale as V (cf. the linked
cluster result), and similarly for higher order terms.
We can also calculate the density ρ
∞
z ∂ ln Q 1 X
ρ= = 3 lbl zl (33)
V ∂z λ l=1

which can be inverted to give a series for z(ρ), and so with Eq. (29) the virial expansion
X ∞
P
= al (T )(ρλ3 )l−1 (34)
ρkB T l=1

with a1 = b1 = 1, a2 = −b2 , a3 = 4b22 − 2b3 ….

The nth term in the virial expansion is given in terms of the partition functions for m particles with m ≤ n.
Classically, it is straightforward to write down expressions for Zn and this is just reproducing the results of
the Mayer expansion. Quantum mechanically, for Zn we have to solve the n-body quantum mechanical
problem, and this can only be done easily up to l = 2, i.e. the first nontrivial virial coefficient.
It is easiest to compare with the noninteracting system (note that even the noninteracting quantum system
has nonzero higher terms in the virial expansion)
1
b2 − b2(0) = (Z2 − Z2(0) ) (35)
2λ3 V
λ3
= (Q2 − Q(0)2 ) (36)
V
λ3 (0)
= T r(e−βH2 − e−βH2 ) (37)
V
with
h̄2 2
H2 = − (∇ + ∇22 ) + u(r12 ). (38)
2m 1
The eigenfunctions of the two body problem can be separated
1 E E
9= ei P ·R/h̄ ψn (Er12 ) (39)
V 1/2
where ψn satisfies  
h̄2
− + ∇rE + u(r) ψn (Er ) = εn ψn (Er ),
2
(40)
2µ
with µ = m/2 the reduced mass, and then
λ3 X −βPj2 /4m X  −βεn (0)

b2 − b2(0) = e e − e−βεn (41)
V j n
X (0)

= 81/2 e−βεn − e−βεn (42)
n

6
where the sum over the center of mass momentum states is just 81/2 V /λ3 . The final sum over the energy
eigenstates can be split into contributions from the bound states and from the continuum (scattering) states
(e.g. see Pathria §9.5)
X Z ∞
(0) −βεB 81/2 X ∂nl (k)
e−β h̄ k /2m
2 2
b2 − b2 = 8 1/2
e + (2l + 1) dk (43)
B
π l 0 ∂k

with ηl the phase shift. The l-sum is over l even for Bosons and l odd for Fermions.