Central University of Rajasthan (CURAJ)

Department of Physics

I. M. S. - M. Sc. - M. Sc. Ed.

SEMESTER January - April, 2019
STATISTICAL MECHANICS KPN
26 January 2019

Partition Function : Sum of Weights

Prologue
Physicists have a name for an un-normalized probability.
They call it weight.
The sum of the weights gives the normalization : the weight divided by the normalization gives
the probability.
Thus the sum of weights is normalization; and another name for the normalization is Partition
Function; this is what a physicist says; and he says so for good reasons, see below.
Consider the micro states of a closed system. Label them with counting numbers i = 1, 2, · · · .
Attach a weight Wi to the micro state i.
X
Q= Wi ,
i

is the normalization. The probability of a micro state j is given by

Wj
Pj = .
Q
Boltzmann discovered that
Wi = exp(−βEi ).
This discovery was made in the context of finding a reasonable mathematical expression for the
distribution of speeds of molecules in an equilibrium ideal gas. We call this Maxwell-Boltzmann
distribution of molecular speeds. Boltzmann extracted a generalized principle from the Maxwell
Boltzmann distribution of molecular velocities. I shall tell you of this interesting episode in one of
the lectures; remind me.
However, we have ”derived” the Boltzmann weight in an entirely different manner altogether,
following the method given in the book of Balescu1 . We had considered the system, and the
surroundings; we said these two together constitute an isolated system; we called the isolated
system, the universe; we invoked the principle of ergodicity : all micro states of the universe are
equally probable; We took the relation proposed by Boltzmann, between entropy and the number of
micro states, Taylor expanded the entropy upto first order derivative and extracted the Boltzmann
weight. If you do not recall what I taught, refer to my notes2 .
The taake-home message is that the weight of a micro state of a closed system, at a given
temperature3 T depends only on its energy. Of course the energy of a micro state shall in turn
depend on other properties like volume4 V and the number of particles N of the system. In other
words, E ≡ E(V, N, · · · ). Hence the sum of weights will depend on V , and N , in addition to
T.
Thus we have X
Q(T, V, N ) = exp[−βEi (V, N ).
i

The normalization Q is called the partition function. It is a function of T , V , and N .

The individual terms in the sum that denotes formally Q(T, V, N ) tell us how the total prob-
ability is partitioned amongst the micro states.
1R Balescu, Equilibrium and Non-Equilibrium Statistical Mechanics, Wiley (1975).
2K P N Murthy, Lecture Notes on Statistical Mechaics (2019).
3 T = 1
kB β
4 note : the energy eigenvalue depends of the boundary conditions.
Partition Function : Isolated System
Consider an equilibrium system of N non interacting point particles, confined to a volume V . A
micro state of the system is defined by

ξ~q = (q1 , q2 , q3 , q4 · · · , q3N ) = { q3(n−1)+i : i = 1, 2, 3 ; n = 1, 2, · · · N }

ξ~p = (p1 , p2 , p3 , p4 , · · · , p3N ) = { p3(n−1)+i : i = 1, 2, 3 ; n = 1, 2, · · · N }

• ξ~q is 3N dimensional;

• ξ~p is 3N dimensional; and

~ obtained juxtaposing the 3N components of ξ~q with the 3N components of ξ~p ,
• ξ,

ξ~ = (q1 , q2 , · · · q3N , p1 , p2 , · · · , p3N )

ξ~ is 6N dimensional.
Consider a 6N dimensional hypercube √ of volume. We take the infinitesimal volume of a 6N
dimensional phase space hyper cube as h borrowing from quantum mechanics5 . We imagine
that these tiny cubes fill up the entire 6N dimensional phase space. Each cube is a possible micro
state of the ideal gas system under considerations. In this fashion we are able to express the phase
space volume occupied the system in terms of integers. In other words we are able to ”count the
phase space volume”.
If the point in the 6N dimensional phase space is such that

ξ~q ∈ Vq

and
ξ~p ∈ Vp ,

P Vq is the space occupied by the container and

then, we attach a weight unity to the micro state.
Vp is a 3N dimensional sphere described by i p2i = 2mE.
If the spatial location of the micro state does not fall with in the container or if its momentum
location does not fall within the 3N dimensional sphere then we attach a weight zero to the micro
state.
We carry out this exercise of attaching weight to to all the micro states in the 6N dimensional
phase space and obtain formally,
Z
b
Ω(E, V, N ) = sum of weights = ~ W (ξ)
dξ ~ =
{6N dimensional phase space}

Z +∞ Z +∞ Z +∞ Z +∞ Z +∞ Z +∞
dq1 dq2 · · · dq3N dp1 dp2 · · · ~
dp3N W (ξ)
−∞ −∞ −∞ −∞ −∞ −∞

Z Z Z Z 3N
!
VN +∞ +∞ +∞ +∞ X √ 2
= dp1 dp2 dp3 · · · dp3N Θ p2
i − 2mE
h3N −∞ −∞ −∞ −∞ i=1

···
cdots
V N (2πmE)3N/2
=  
h3N Γ 3N + 1
2

b
Thus Ω(E, V, N ) is the number of microstates of a system of N identical, non - interacting, point particles confineed
to a container of volume V and with a total fixed energy E. We can call it as partition function, more precisely as
micro canonical partition function : The probability is equally partitioned amongst the micro states of an isolated
system.
5 Quantum Mechanics prescribes that ∆x∆p ≥ ~/2π as the minimum uncertainly volume. In the classical

counting of the number of micro states, the term h3N is put in by hand. However, when we do quantum counting
of micro states, the term h3N comes in the expression naturally. I will tell you, how to do the quantum counting of
micro states in one of the lectures. We will find that both classical and counting counting lead to the same results.

2
Partition Function : Closed System
To each micro state ξ~i we attach a Boltzmann weight Wi = exp[−βE(ξ
~i )]. The sum of the weights is given by
Z
Q(T , V, N ) = sum of weights = ~ W (ξ)
dξ ~ =
{6N dimensional phase space}

Z +∞ Z +∞ Z +∞ Z +∞ Z +∞ Z +∞
dq1 dq2 · · · dq3N dp1 dp2 · · · ~
dp3N W (ξ)
−∞ −∞ −∞ −∞ −∞ −∞

Z Z Z Z 3N
!
VN +∞ +∞ +∞ +∞ X p2
i
= dp1 dp2 dp3 · · · dp3N exp −β
h3N −∞ −∞ −∞ −∞ i=1
2m

Z  3N
VN +∞ p2
dp exp −β
h3N −∞ 2m

VN
=
Λ3N
where
h
Λ= √
2πmkB T

b
Ω(E, V, N ) → Q(T, V, N )
Let g(E, V, N ) denote the density of (energy) states. g(E, V, N )dE is the number of micro states with energy in
an infinitesimal interval dE around E.
I told you that a partition function can be considered as a transform6 . Accordingly we proceed as follows.
b
∂ Ω(E, V, N )
g(E, V, N ) =
∂E

Z ∞
Q(β, V, N ) = dE exp(−βE)g(E, V, N )
0

We have,

V N (2πm)3N/2 3N
g(E, V, N ) = E (3N/2)−1
h3N Γ((3N/2)
+ 1) 2

V N (2πm)3N/2 (3N/2)−1
= E
h3N Γ(3N/2)

Z
V N (2πm)3N/2 ∞
Q(β, V, N ) = dE E (3N/2)−1 exp(−βE)
h3N Γ(3N/2) 0

Carry out the above integral, I leave this to you as an exercisee, and show,
 
V 3N
Q(T , V, N ) = 3
.
Λ
Notice that the above is the same result that we derived in the last section by a different route.

End Note
We include a multiplicative factor of 1/N !, as suggested by Boltzmann to resolve Gibbs’ paradox and restore the
extensive property of entropy. I have told you of this. You can also see my notes for details of Gibbs paradox, and
Boltzmann’s ”indistinguishable particles”.
6 Fourier transform : Z +∞
φ(k) = F [f (x)] = dx exp(−ikx)f (x),
−∞
where F is the Fourier transform operator. φ(k) is called the Fourier transform of f (x).
Laplace transform : Z ∞
φ(s) = L[f (t)] = dt exp(−st)f (t)
0
where L denotes the Laplace transform operator; etc.

3
 We have also expressed the partition function in terms of thermal or quantum wave length Λ. This length
corresponds approximately to the de Broglie wavelength of a particle with energy equal to kB T .
Λ is a good parameter to discuss at what densities and temperatures quantum effect will start dominating and
dictate the properties of the particle. To see this let r denote the average distance separating two particles in the
system. It is clear if Λ << r then we can expect the system to obey classical Maxwell-Boltzmann statistics. The
N particle wave function can be written as the product of N single particle wave functions. If the single particle
wave function is extremely localized we can say classical behaviour obtains. We can write this condition as,
Λ << r
Λ3 << r3
3
Λ << 1/ρ
where rho is density of particles : number of particles per unit volume. We can write the above inequality as
ρΛ3 << 1,
whence classical behaviour is expected to prevail.
We can say that the quation ρΛ3 = 1 sets the crossover temperature T ⋆ for a given density or the crossover
density ρ⋆ for a given temperature.
We have
h3
ρ √
3 = 1
2πmkB T

Therefore
(2πmkB T )3/2
ρ⋆ =
h3

ρ2/3 h2
T⋆ =
2πmkB

Problems
(1) Consider Hydrogen gas.
(i) Calculate T ⋆ at normal density.
(ii) Calculate ρ⋆ at room temperature.
Show that the properties of Hydrogen gas at normal temperature and pressure can be studied employing
classical statistical mechanics.
(2) Carry out the calculations given in Problem (1) for liquid Helium.
(3) Consider electron gas. Show that we need quantum statistics (Fermi-Dirac statisticS) to study electon
behaviour at room temperature.
Hint : If required consult Kersen Huang, Introduction to Statistical Physics, Second Edition, CRC Press (2010)
p.196; Table 14.1

System ρ per cm3

Hydrogen gas 2 × 1019

Liquid Helium 2 × 1022

electrons in metals 1022