Chapter 7

Equilibrium statistical physics

7.1 Introduction
The movement and the internal excitation states of atoms and molecules constitute the
microscopic basis of thermodynamics. It is the task of statistical physics to connect the
microscopic theory, viz the classical and quantum mechanical equations of motion, with
the macroscopic laws of thermodynamics.
Thermodynamic limit The number of particles in a macroscopic system is of the order
of the Avogadro constant Na ∼ 6 · 1023 and hence huge. An exact solution of 1023 coupled
equations of motion will hence not be possible, neither analytically,∗ nor numerically.†
In statistical physics we will be dealing therefore with probability distribution functions
describing averaged quantities and their fluctuations. Intensive quantities like the free
energy per volume, F/V will then have a well defined thermodynamic limit

F
lim . (7.1)
N →∞ V
The existence of a well defined thermodynamic √ limit (7.1) rests on the law of large num-
bers, which implies that fluctuations scale as 1/ N . The statistics of intensive quantities
reduce hence to an evaluation of their mean for N → ∞.
Branches of statistical physics. We can distinguish the following branches of statistical
physics.

• Classical statistical physics. The microscopic equations of motion of the particles are
given by classical mechanics. Classical statistical physics is incomplete in the sense
that additional assumptions are needed for a rigorous derivation of thermodynamics.

• Quantum statistical physics. The microscopic equations of quantum mechanics pro-

vide the complete and self-contained basis of statistical physics once the statistical
ensemble is defined.
∗
In classical mechanics one can solve only the two-body problem analytically, the Kepler problem,
but not the problem of three or more interacting celestial bodies.
†
A computer made of 1023 atoms will not be able to simulate 1023 differential equations.

75
76 CHAPTER 7. EQUILIBRIUM STATISTICAL PHYSICS

• Statistical physics in equilibrium. The macroscopic variables are time-independent

and the macroscopic world can be described in terms of microscopic average values,
distribution functions or probabilities. The derived relations between the macro-
scopic quantities correspond then to the laws of thermodynamics.
• Statistical physics in non-equilibrium. The macroscopic variables are time-dependent.
Advanced approximate methods are needed in this case for a microscopic treatment.
In this course, we will restrict ourselves to classical and quantum statistical physics in
equilibrium.

7.2 Statistical mechanics fundamentals

Our objective is to describe the macroscopic properties of N � 1 particles in thermody-
namic equilibrium.
Conserved quantities. The properties of a system in thermodynamic equilibrium do
not change with time. This implies, as discussed further in Sect. 7.2.1, that there exist
conserved quantities like energy and momentum. Conservation of energy and particle
number will be of particular importance for the definition of statistical ensembles.
Additivity. Macroscopic extensive variables must add when a thermodynamic system
is partitioned into subsystems. The total energy E of
a system subdivided into subsystems with energies Ei
should hence obey
�
E = Ei
i

with negligible boundary effects. Additivity is present

when the correlations between the distinct subsystems
vanish in the thermodynamic limit. In this course we
will not discuss the derivation of additivity, which is
based microscopically on finite correlation lengths.
Energy vs. internal energy. With energy being both conserved and additive we can
identify it with the internal energy U = U (S, V, N ). The total differential (5.14) of the
entropy S(U, V, N ) = S(E, V, N ) then allows us to evaluate with
� � � � � �
1 ∂S P ∂S µ ∂S
= , = , − = (7.2)
T ∂E V,N T ∂V E,N T ∂N E,V
in particular the temperature T . Above relation can be hence viewed as one of the possible
definition of the temperature.
Defining the entropy. The relations (7.2) can be evaluated once the entropy S and the
energy E have been computed for a systems of N particles confined to a volume V . This
is not a problem for the energy, for which well-established microscopic expressions are at
hand both for classical and for quantum mechanics. In order to proceed with statistical
mechanics a microscopic definition of the entropy is hence necessary.
7.2. STATISTICAL MECHANICS FUNDAMENTALS 77

7.2.1 Classical mechanics

The state of a particle at any instant of time is specified within classical mechanics by its
position �q and momentum p�. A N -particle system has therefore 6N degrees of freedom
specified by � �
�qi , p�i , i = 1, . . . , N . (7.3)
Hamiltonian. The motion of the particles are governed by the canonical equations
∂H ∂H
p�˙i = − , �q˙i = , H = H(�q1 , . . . , �qN , p�1 , . . . , p�N ) , (7.4)
∂�qi ∂�pi
where
�N
p� 2i 1�
H = + Φ(�ri − �rj ) (7.5)
i=1
2m 2 i�=j
is the Hamiltonian of the system. In (7.5) we have used the real-space �ri coordinates for
the generalized coordinates �qi . The interaction between the particles is given in (7.5) by
the interaction potential Φ(�ri − �rj ).
Noether’s theorem. A corollary of Emmy Noether’s theorem is that energy is conserved
when the Hamiltonian is independent of time. For a thermodynamic setting, where par-
ticles are confined to a volume V , one needs to add a one-particle potential
�
0 (�r ∈ V )
ΦV (�r) = (7.6)
∞ (�r �∈ V )
to the Hamiltonian (7.5). Energy is therefore not guaranteed to be conserved for a volume
change, which corresponds to a time-dependent confining potential ΦV and therefore also
to a time-dependent Hamiltonian.
Ideal gas limit. The ideal gas limit corresponds to a negligible inter-particle interaction
Φ(Δ�r) → 0 in (7.5). For most real gases the density of molecules is small enough that
this condition is approximately fulfilled, with the residual contribution of Φ(Δ�r) giving
rise to corrections, as captured e.g. by the van der Waals theory discussed in Sect. 6.4.

7.2.2 Microstates
The complete description of an isolated system with N particles is given by the generalized
coordinates
q = (q1 , . . . , q3N ), p = (p1 , . . . , p3N ) , (7.7)
which span the phase space. The generalized coordinates (�qi , p�i ) of the individual particles
are lumped together as components in (7.7). Alternatively on may consider as in (7.3)
directly the set of all individual (�qi , p�i ).‡
Microstates. In dynamical system theory one is interested in evaluating the trajectories
in the phase space, viz the time evolution (�qj (t), p�j (t)) for j = 1, . . . , N . In contrast one
is interested in statistical mechanics in the statistical properties of the system. One then
defines a given state (7.7) in phase state as a microstate.
‡
Eq. (7.7) is sometimes called the Γ-representation of phase space and Eq. (7.3) the µ-representation.
78 CHAPTER 7. EQUILIBRIUM STATISTICAL PHYSICS

Energy surface. Energy is a conserved

quantity if H does not depend explicitly on
time. The trajectories are then confined to
the energy surface.
As an examples consider the one-dimensio-
nal harmonic oscillator (N = 1) with the
Hamiltonian
p2 1
H = + mω 2 q 2 = E .
2m 2
The trajectory of this system in the Γ-space is given by the ellipse E = const.

The energy surface is on the other side discontinuous for a particle in a potential box with
ideal reflecting walls, as specified by p� ↔ (−�p) and (7.6).

7.2.3 Probability distribution functions

Integrating over the microstates (7.7) corresponds to the differential

d3N q d3N p = dq1 , . . . , dq3N , dp1 , . . . , dp3N .

Probability distribution function. We define the distribution function ρ(q, p) (also

called probability density), with
�
ρ(q, p) d3N q d3N p = 1, ρ(q, p) ≥ 0 , (7.8)

which gives the probability to find the system in the state (q, p) = (q1 , . . . , q3N , p1 , . . . , p3N ).
Ideal-gas limit. The constituent particles do not interact in the ideal-gas limit. Defining
with �
f (�p, �q ) d�q d�p = 1, f (�q , p�) ≥ 0

the probability that a given particle has coordinates �q and momentum p� leads then to
N
�
ρ(q, p) = f (�pi , �qi ) , (7.9)
i=1
7.3. LIOUVILLE THEOREM 79

where i runs now over all particles. This relation just states that the probability of a
combination of independent events (like throwing a dice several times) is just the product
of the individual probabilities.
Expectation values. We use in our notation E for the total energy of the system and
� 2 � 2
E p� p�
= f (�q , p�) d�q d�p = f (�p) d�p (7.10)
N 2m 2m
for the energy per particle, where we have used in the last step that f (�q , p�) = f (�p) when
the system is uniform. In this case the properties do not depend on where they are
measured. Note that only the kinetic energy did enter (7.10), as we are working in the
ideal-gas limit.

7.3 Liouville theorem

Let us consider a system with the distribution function as defined by (7.8). We are now
interested in the time dependence, viz in ρ = ρ(q, p, t),
Phase space flow. The velocity �vps in phase space (PS),
d� �
�vps = q, p = (q̇1 , q̇2 , . . . , q̇3N , ṗ1 , . . . , ṗ3N ) , (7.11)
dt
is called the flow. Its divergence
3N �
� �
∂ q̇j ∂ ṗj
∇ · �vps = + , ∇ · �vps ≡ 0 , (7.12)
j=1
∂qj ∂pj

vanishes generically for Hamiltonian systems governed by (7.4), since.

∂ q̇j ∂ ∂H ∂ ṗj ∂ ∂H
= , = − .
∂qj ∂qj ∂pj ∂pj ∂pj ∂qj
Any vector field with a vanishing divergence, like the phase space velocity �vps , is volume-
conserving. Repeating the arguments leads to Liouville’s theorem (see below).
80 CHAPTER 7. EQUILIBRIUM STATISTICAL PHYSICS

Phase space current. The flow (7.11) defines also the phase-space current density �ps ,
viz
�ps = ρ(q, p, t)�vps .
Remember that the electrical current is defined in analogy as the product of charge,
particle density and velocity.
Continuity equation in integral form. Consider a
region G in phase space with surface S. The number
of trajectories that cross the surface S per unit time
is given by
�
� ·�ps ,
dS � = dS · �n ,
dS
S(G)

where �n is a vector orthonormal to the surface S. The

flux of trajectories through the surface S(G) obeys
� �
� ∂
dS ·�ps = − d3N q d3N p ρ(q, p, t) , (7.13)
S(G) ∂t G

as there are no sources or sinks for the trajectories. Eq. (7.13) is just the continuity
equation in integral form.
Gauss’ law. By making use of the Gauss law we can replace the integral over the surface
in (7.13) by a volume integral,
� � �
3N 3N ∂ � ·�ps
d qd p ρ+∇ = 0 , (7.14)
G ∂t

with � �
∂ ∂ ∂ ∂
∇ ≡ ,..., , ,...,
∂q1 ∂q3N ∂p1 ∂p3N
and � � � �
∇ ·�ps = ρ ∇ · �vps + �vps · ∇ ρ ≡ �vps · ∇ ρ , (7.15)
where we have used (7.12), namely that the divergence ∇ · �vps of the flow vanishes for
Hamiltonian systems.
Liouville equation. With the continuity equation in integral form (7.14) being valid for
all G we obtain

∂ � � d
ρ(q, p, t) + �vps · ∇ ρ(q, p, t) = 0, ρ(q, p, t) = 0 , (7.16)
∂t dt

where we have used (7.15) for the divergence of �ps and that the left-hand side of the first
term in (7.16) corresponds to the total differential of ρ(q, p, t) with respect to the time.
Eq. (7.16) is denoted Liouville theorem.
7.4. RELAXING TOWARDS THERMAL EQUILIBRIUM 81

Conservation of phase space volume. The

Liouville equation (7.16) tells us that the total
time derivative dρ/dt of the density ρ(q, p, t) of
trajectories vanishes. It implies that

ρ (q(t), p(t), t) ≡ ρ (q(0), p(0), 0)

is a constant of motion. Phase space volume is

hence conserved. This result holds whenever the
divergence ∇·�vps of the flow vanishes, see (7.12).

7.4 Relaxing towards thermal equilibrium

One mol of an ideal gas occupies a molar volume of

P0 V 0 J
V0 = 2.24 · 104 cm3 , R = = 8.3 ,
T0 mol K

where we have used T0 = 273.15 K and P0 = 1 bar, corresponding to 105 Pa = 105 N/m2 .
The density of the gas is then

Na
n = = 2.7 · 1019 atoms/cm3 , Na = 6.022 · 1023 .
V0

Scattering cross section. The thermal equilibrium is reached in a gas because of

particle collisions, viz due to the inter-particle interaction Φ(Δ�r) in (7.5). The scattering
cross section
σ = πr02 , r0 ≈ 2 · 10−8 cm
is for short-ranged inter-particle interactions Φ(Δ�r) of the order of the area of the particles.
Note that the Bohr radius is 5.29 · 10−11 m = 0.529 ◦ A.
Mean free path. We define the mean free path λ
as the average distance traveled by a particle between
two successive collisions. Consider a cylinder with a
cross section A containing a gas of density n.
The length of the cylinder is of the mean-free path
λ when A is approximately equal to the cumulative
cross-sections of the contained atoms, viz when

1
A ≈ ����
Aλn σ, λ = ∼ 10−5 cm .
nσ
N

The average distance between two collisions is hence much larger than the typical size of
a molecule, λ � r0 , which implies that the molecules in real gases interact only weakly.
82 CHAPTER 7. EQUILIBRIUM STATISTICAL PHYSICS

Average velocity. We did evaluate in Sect. 3.4.1 the internal energy per particle of
an ideal gas as 3kB T /2. Then, we can obtain the average velocity v of the constituent
particles as
�
1 2 3 3kB T
mv = kB T, v =
2 2 m

The mean velocity is at room temperature (T = 300 K) of the order of v ≈ 105 cm/s, viz
of the order of one km/s.
Relaxation time. The average time between successive collisions τc ,

λ
τc = ≈ 10−10 s ,
v

corresponds then to the time the gas needs to relax to thermal equilibrium. The relaxation
time τc is essentially instantaneous with regard to the time scales macroscopic objects need
to move.

7.4.1 Ergodicity timescales

A point
(q, p) = (q1 , . . . , q3N , p1 , . . . , p3N )
in phase space corresponds to a state of the N -particle system. It follows as a function
of time a trajectory on the energy surface. After a certain time, which should be long
enough compared to the collision time τc ≈ 10−10 s (for an ideal gas at a temperature of
T = 300 K and a pressure of P = 1 bar), the trajectory has lost essentially all information
about its initial state. Averaging further over time then corresponds to a sampling of the
thermal state.
Ergodicity. A dynamical system is said to be ergodic when it eventually explores the
entire available phase space.

ERGODICITY Given sufficient time, the representative point of

an isolated system will come arbitrarily close to
any given point of the energy surface.

A system needs to be ergodic in order to correspond to a uniquely defined thermal state.

Its time-averaged properties would otherwise depend on the initial conditions.
Liouville theorem. Liouville theorem (7.16) states that phase space volume is invariant
with respect to Hamiltonian flows like (7.4). With the energy surface having a finite
volume this implies that the flow visits eventually every point in phase space in the sense
that is will approach it arbitrarily close, time given. The involved recurrence times may
however be exceedingly long.§
§
A famous example is the teacup experiment. Try to stare at your teacup and to wait for a drop of
tea to be pushed (by coordinated movements of the water molecules) out into the open air.
7.4. RELAXING TOWARDS THERMAL EQUILIBRIUM 83

Chaotic states. The continuous scattering of the particle results in unpredictable and
hence chaotic motions. The loss in predictability results from the fact that small variations
in the distance of near miss collisions may result in large changes of the trajectories of
the two colliding particles. Systems showing deterministic chaos sample phase space
effectively.

7.4.2 Time- vs. statistical averaging

Classical experiment, measuring thermodynamic quantities like the temperature, typically
need a certain time. They hence perform averages over time intervals that are small on
the macroscopic scale but large compared to the collision time τc .

POSTULATE In statistical mechanics it is assumed that time averages can

be replaced by averages over a suitably chosen collection of
systems called statistical ensemble.

Observables. To see what this means let us consider an observable O = O(q, p),
O(q, p) = O(q1 , . . . , q3N , p1 , . . . , p3N ) ,
which is classical just a function on phase space. A well known observable is the kinetic
energy
�N
p� 2i
Ekin = .
i=1
2m
Measurements. A measuring process provides via
�
1 τ
Ō(t0 ) = dt O(q(t0 + t), p(t0 + t)) (7.17)
τ 0
a time average Ō of the observable. t0 is here initial time and τ the duration of the
experiment.
Ergodic hypothesis. It is impossible to calculate Ō(t0 ) from the the solution of the
Hamiltonian equations for N = 1023 molecules. For an chaotic and hence ergodic system
we can however assume that the system samples the chaotic attractor within the duration
τ of the measurement. It is then possible to replace the time average (7.17) with
� �
�O� ≡ d q d3N p ρ(q, p) O(q, p) ,
3N
(7.18)

viz by the corresponding expectation value over the phase-space density ρ(q, p).

STATISTICAL ENSEMBLE A statistical ensemble is defined as the infinite

collection of identical copies of the system char-
acterized by a distribution function ρ(q, p, t).

Interpretation. The time average Ō of a function O = O(q, p) on phase space and the
ensemble average �O�, as defined respectively by (7.17) and (7.18), coincide for systems
in thermal equilibrium.
84 CHAPTER 7. EQUILIBRIUM STATISTICAL PHYSICS

7.5 Postulate of equal a priori probabilities

In order to connect the microscopic equations of motion to the laws of thermodynamics
one needs to define in addition either the temperature T or the entropy S, as done in the
following chapters. A key step in this direction is to understand the structure of ρ(q, p).
Constant energy. The total energy E is defined in the ideal gas limit as
N
1 � � �2
E = p�j (7.19)
2m j=1

This equation defines the surface of constant energy as a hyper-sphere in phase space. The
question regards then the form the density of states ρ(q, p) takes on this hypersphere.
Constant velocities. It could be, as a matter of principle, that
3N
� �
ρ(q, p) = δ(pi − p̄), p̄ = 2mE/(3N ) . (7.20)
i=1

In this case all particles would have identical and constant momenta and hence velocities.
Eq. (7.20) is however not what is observed.
All allowed microstates are equally likely. The postulate of equal a priori probabili-
ties states that all allowed microstates are equally likely. Eq. (7.20) is hence as wrong as
it gets, with ρ(q, p) being on the contrary constant on the hypersphere defined by (7.19).
The postulate of equal a priori probabilities can only be motivated but not proven. We
will discuss it further in the next chapter. It has the stance of a law of nature.

All laws of nature are postulates in the sense that they

need to be verified by experiments. They cannot be
derived.