Unit 1 Introduction To The Statistical Physics. The Microcanonical Ensemble

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 32

Unit 1

Introduction to the Statistical


Physics. The microcanonical
ensemble.

1.1 Introduction
The aim of Statistical Physics is to derive and understand the behavior of macro-
scopic systems from their microscopic description. There are well-established disci-
plines which provide the governing laws at both scales. For the macroscopic scale
(i.e. for objects much larger than an atom but smaller than the Moon) Thermody-
namics provides a set of basic principles which must be obeyed by any macroscopic
system. These are the Thermodynamics laws, which can be formulated as follows:
• 0th law: this is an statement on equilibrium states, which are macroscopic
stationary states which do not present net flows of energy or matter within
the system or with its surroundings. So, if two systems at different macro-
scopic conditions are placed into thermal contact, they evolve to a common
equilibrium state. A system is obviously in equilibrium with itself (reflexivity
property) and if A is in equilibrium with B, then B is also in equilibrium with
A (symmetric property). The zeroth law states that the equilibrium condition
satisfies also the transitivity property. So, if A is in equilibrium with B, and
B is in equilibrium with C, then A is in equilibrium with C. Thus, the equi-
librium condition is an equivalence relationship, so all the equilibrium states
can be classified into equivalence classes. We can associate to each equivalence
class a labeling number, which is the (empirical) temperature.

1
Introduction to the Statistical Physics. The microcanonical ensemble. 2

• 1st law: it is a generalization of the energy conservation principle in Mechan-


ics. It states that in any process of a macroscopic system between equilibrium
states, the variation of its internal energy (which is an state function, i.e.
depends only on the macroscopic state) is the balance between the heat Q
exchanged with the surroundings and the work W which the system exerts on
the external world
∆U = Q − W, (1.1)
or in differential form
dU = d¯Q − d¯W, (1.2)
where d¯ corresponds to an inexact differential, which does not depend only on
the final and initial states, but also on the followed path.

• 2nd law: the second law has been expressed in many ways (heat cannot flow
from cold to hot objects without any other change, a machine cannot generate
work by draining heat from a single thermal bath,...). However, we will use
Planck’s formulation. First, entropy (a state function as the internal energy)
is defined from the relationship
d¯Qrev
dS = , (1.3)
T
where d¯Qrev is the infinitesimal heat transfer between the system and its sur-
roundings along a reversible path (i.e. quasi-static, so it is so slow that each
step in the path can be considered to be almost in equilibrium, and that it
can be reversed in such a way that both the system and its surroundings are
returned to their original states), and T is the absolute temperature. So in any
process on an isolated system (i.e. a system which does not exchange energy
of any kind or matter with its surroundings) entropy must increase ∆S ≥ 0,
being ∆S = 0 for reversible processes.

• 3rd law: The entropy of a perfect crystal at absolute zero is exactly equal to
zero.

On the other hand, a macroscopic system is constituted by a hugh number N of


microscopic entities (atoms, molecules, ions, electrons, etc.). Typically, this number
is of order of Avogadro’s number NA = 6.023 × 1023 . This particle collection follows,
formally, the laws of the Quantum Mechanics. However, under some circumstances
(high temperature, low to moderate density), Classical Mechanics can describe the
dynamics of the microscopic system. In general, it is assumed that the microscopic
Introduction to the Statistical Physics. The microcanonical ensemble. 3

system is conservative, so its energy is the sum of the kinetic energy K


N
X 1
K= mi vi2 (1.4)
i=1
2

and the potential energy U ≡ U (r1 , . . . , rN ), which can be split into different con-
tributions
XN X
U= U1 (ri ) + U2 (ri , rj ) + . . . , (1.5)
i=1 i<j

where the 1-body term corresponds to the interactions with the surroundings (in-
cluding the container walls), the U2 corresponds to two-body interactions between
particles of the system, and so on. A popular potential used as U2 is the Lennard-
Jones, which it is accurate to describe noble gases.
The connection between both levels of description is far from obvious. First,
macroscopic laws are valid for many systems and virtually independent of the initial
conditions, but the mechanical dynamics is specific to each model and depends
strongly on the initial conditions. Second, the macroscopic state is defined by a
relatively small number of physical quantities (in equilibrium hydrostatic systems, by
the internal energy E, the volume V and the number of particles N , or equivalently
by the temperature T , V and N ; in non-equilibrium states by the hydrodynamic
fields), which must be compared to the values of the positions and velocities of all
the microscopic components of the system (of order of NA ). Finally, note that there
are additional issues we will not consider, as how the time-irreversible nature of the
macroscopic laws can be derived from the time-reversible laws of the (Classical or
Quantum) Mechanics.

1.2 Review of Classical Mechanics


In this Section we will review some of the basic concepts of the Classical Mechanics.
If the system is composed by structureless atoms, we can use Newton’s laws to
determine the dynamical evolution of the system

d2 ri
mi = F i (r1 , . . . , rN ), (1.6)
dt2
where mi and ri are the mass and position vector of the atom i, and F i is the net force
which acts on the particle i. If we provide the positions and velocities v i = dri /dt
at an initial time t = t0 , then we can formally solve the system dynamics for t > t0 .
Introduction to the Statistical Physics. The microcanonical ensemble. 4

However, this is practically impossible for macroscopic systems, with a number of


microscopic components of order of 1023 .
However, there are situations in which the Newtonian framework is not the most
appropriated. For example, consider the case of a macroscopic system of N rigid
diatomic molecules. In Newton’s formulation, we must use Newton’s law for each
atom, but the rigidity constraint means that a constraint forces must by included
between the atoms of the same molecule. An alternative is to use other set of
coordinates to characterize a molecule configuration (for example, the center-of-mass
position vector and two angles associated to the molecular bond orientation). So, we
need 5 coordinates (instead of 6) to fully characterize the molecular configuration.
In general, a system of N particles (in D dimensions) subject to m constraints will
require f = DN − m coordinates to characterize a system configuration. f is the
number of degrees of freedom of the system, and we will denote by {qi , i = 1, . . . , f }
to the set of generalized coordinates which describe each configuration. The set of
possible configurations of the system is the configurational space. Finally, we denote
as {q̇i ≡ dqi /dt, i = 1, . . . , f } as the set of generalized velocities. This is the so-called
Lagrangian framework of the Classical Mechanics.
In order to obtain the system dynamics in the Lagrangian approach, we have to
introduce the concept of Lagrangian. For conservative systems, this is defined as
L = K − U, (1.7)
where K and U are the kinetic and potential energies, respectively, written in terms
of the generalized coordinates and velocities. Thus the evolution equations of the
system are given by the Euler-Lagrange equations
 
∂L d ∂L
− =0 i = 1, . . . , f, (1.8)
∂qi dt ∂ q̇i
which, as Newton’s laws, is a set of second-order differential equations which can be
formally integrated provided the generalized coordinates and velocities at an initial
time are given.
An alternative formulation of the Classical Mechanics is the Hamiltonian frame-
work, which will be more useful for the Statistical Physics. First, the generalized
momenta are defined as
∂L
pi = i = 1, . . . , f. (1.9)
∂ q̇i
Now, the Hamiltonian function H is defined as
f
X
H= pi q̇i − L, (1.10)
i=1
Introduction to the Statistical Physics. The microcanonical ensemble. 5

where all the dependence on q̇i is replaced by the generalized momenta pi . In


particular, for a conservative system H = K + U . Now, the evolution equations of
the system are Hamilton equations:
∂H ∂H
q̇i = , ṗi = − i = 1, . . . , f, (1.11)
∂pi ∂qi
which is a set of 2f first-order differential equations, which can be integrated formally
once the values of the generalized coordinates and momenta are provided for an
initial time. So, the dynamical state of the system is now given by a point in the
2f phase space, with coordinates {q1 , . . . , qf , p1 , . . . , pf }.
To finish this Section, let’s consider how evolves a function defined on the phase
space f = f (q1 (t), . . . , qf (t), p1 (t), . . . , pf (t), t) ≡ f (t). The time variation of this
function has two contributions: first, from the explicit time-dependence of f , and
second, from the time evolution of the phase space coordinates. So
f  
df ∂f X ∂f ∂f
= + q̇i + ṗi . (1.12)
dt ∂t i=1
∂q i ∂p i

By using Hamilton’s equations, this expression reads


f  
df ∂f X ∂f ∂H ∂f ∂H
= + − . (1.13)
dt ∂t i=1
∂q i ∂p i ∂p i ∂q i

We define the Poisson’s bracket of two dynamical variables A and B as


f  
X ∂A ∂B ∂A ∂B
{A, B} = − . (1.14)
i=1
∂q i ∂p i ∂p i ∂q i

Thus
df ∂f
= + {f, H}. (1.15)
dt ∂t
If f = H, then dH/dt = ∂H/∂t, so if the Hamiltonian does not depend explicitly
on time, it is a constant of movement in the system dynamics.

1.3 Microscopic and macroscopic descriptions


As we mentioned in the Introduction, the macroscopic state is defined by a relatively
small number of physical macroscopic quantities, but microscopically an astronomic-
large number of initial conditions are required to follow the system evolution. Fur-
thermore, the macroscopic state is virtually independent of the initial conditions,
Introduction to the Statistical Physics. The microcanonical ensemble. 6

but the microscopic dynamics is specific to each model and depends strongly on
the initial conditions. In order to reconcile both description, the Statistical Physics
postulates a probabilistic approach. It is obvious that the macroscopic state (or
macrostate) cannot specify the microscopic state (or microstate) in which the sys-
tem is at certain instant. In another words, there are many possible microstates
compatible with a macrostate, so at a given time, the system may be in any of these
microstates. Now, if we follow the evolution of many identical, independent systems
in the same macrostate (maybe prepared from different initial conditions), a certain
microstate will appear with a frequency which is given by an a priori probability
density in the phase space
ρ(q1 , . . . , qf , p1 , . . . , pf , t) ≡ ρ(q, p, t). (1.16)
The set of microstates compatible with a given macrostate, together with the cor-
responding probability density ρ(q, p, t), defines an ensemble. If the macrostate cor-
responds to an equilibrium thermodynamic state, we will denote them as a Gibbs
ensemble.
Once we have introduced the concept of ensemble of microstates associated to a
macrostate, we must establish the connection between the macroscopic and micro-
scopic properties. Let’s suposse that a macroscopic property Amacro has a micro-
scopic analogue A(q, p) (such as the internal energy U and the Hamiltonian H(q, p)).
The connection between both properties is postulated in the Statistical Physics to be
probabilistic. So, the first postulate of the Statistical Physics establishes that
the macroscopic property Amacro is an ensemble-average or mean of the microscopic
property A(q, p)
Z
Amacro (t) = A(t) = dqdpA(q, p)ρ(q, p, t), (1.17)

where dqdp ≡ dq1 . . . dqf dp1 . . . dpf . Note that this postulate is only valid for macro-
scopic properties which have a microscopic analogue. It is not valid for properties
without a microscopic analogue, such as the entropy or free energies. For these
quantities, we will see how they will be defined within the Statistical Physics.
As we saw in the previous Unit, the mean is only one of the quantities which
define a probability distribution. In general, the values of A(t) depart from its mean
value (see Fig. 1.1). These deviations are called fluctuations of the quantity A. In
order to estimate de typical size of the fluctuations of A, we can use its standard
deviation σA . In general, a instantaneous value of A in a typical microstate is in the
range [A(t) − rσA (t), A(t) + rσA (t)], where r is a number of order of 1. If σA /A  1,
the difference between A(t) and A(t) is macroscopically negligible. In this situation,
A represents a well-behaved macroscopic law, otherwise it does not correspond to
any physical law, since there will be no certainty on the value that A may have.
Introduction to the Statistical Physics. The microcanonical ensemble. 7

A(t)
A(t)-σΑ(t)
A(t)+σΑ(t)
A(t)

Figure 1.1: Macroscopic description A(t) and a realization A(t), with the confidence
ranges A(t) ± σA (t).

1.4 The Liouville equation


As the probability density ρ is a function on the phase space, its evolution is deter-
mined by the equation
dρ ∂ρ
= + {ρ, H}. (1.18)
dt ∂t
In order to evaluate this derivative, we will make an analogy with the fluid dynamics.
Probability is carried out by the phase space points in their time evolution, so
probability is neither created nor destroyed. Let’s suppose that we have a (fixed)
phase space volume Γ. The probability stored in Γ at the instant t is
Z
dqdpρ(q, p, t), (1.19)
Γ

where (q, p) ≡ (q1 , . . . , qf , p1 , . . . , pf ), and dqdp ≡ [q1 , q1 + dq1 ] × [qf , qf + dqf ] ×


[p1 , p1 + dp1 ] × [pf , pf + dpf ]. The time variation of this probability is due to the
probability flux through the boundary of Γ, ∂Γ
Z
ρv · nds, (1.20)
∂Γ
Introduction to the Statistical Physics. The microcanonical ensemble. 8

where v = (q̇1 , . . . , q̇f , ṗ1 , . . . , ṗf ), and n is the outward normal to the elementary
hypersurface ds on ∂Γ. Thus
Z Z
d
dqdpρ(q, p, t) + ρv · nds = 0. (1.21)
dt Γ ∂Γ

As Γ is independent of time, and by using the divergence theorem, we can rewrite


this expression as Z  
∂ρ
dqdp + ∇ · (ρv) = 0. (1.22)
Γ ∂t
Since Γ is arbitrary, the integrand must vanish everywhere. In coordinates
f  
∂ρ X ∂(ρq̇i ) ∂(ρṗi )
+ + = 0. (1.23)
∂t i=1
∂qi ∂pi

Expanding the derivatives on the second term in the right-hand side, we obtain
f   f  
∂ρ X ∂ q̇i ∂ ṗi X ∂ρ ∂ρ
+ρ + + q̇i + ṗi = 0. (1.24)
∂t i=1
∂qi ∂pi i=1
∂qi ∂pi

From Hamilton’s equations (ṗi = −∂H/∂qi , q̇i = ∂H/∂pi ), we get that

∂ ṗi ∂ 2H ∂ q̇i ∂ 2H
=− , = . (1.25)
∂pi ∂pi ∂qi ∂qi ∂qi ∂pi
As the crossed derivatives of H are equal, the second term in the left-hand side
vanishes, and the third one reduces to the Poisson’s bracket {ρ, H}. Thus, we get
that
dρ ∂ρ
= + {ρ, H} = 0, (1.26)
dt ∂t
which is the Liouville’s equation. Now let’s suppose that we have the infinitesimal
volume dq(t)dp(t), which corresponds to the set of phase space points (q(t), p(t))
which at t = 0 were at dq(0)dp(0). As the probability is carried out by the phase
space points, the probability stored at dq(0)dp(0) is conserved along the time evo-
lution
ρ(q(0), p(0), 0)dq(0)dp(0) = ρ(q(t), p(t), t)dq(t)dp(t). (1.27)
Since dρ/dt = 0, ρ(q(0), p(0), 0) = ρ(q(t), p(t), t), so the previous expression leads to
the conservation of the phase space volume

dq(0)dp(0) = dq(t)dp(t). (1.28)


Introduction to the Statistical Physics. The microcanonical ensemble. 9

This expression is consistent with the fact that the time evolution is a canonical
transformation, which is known to conserve the phase space volume.
For equilibrium states, which are stationary, i.e. independent of time, ∂ρ/∂t = 0.
Consequently, Liouville’s equation states that ρ is a constant of movement. One
option is that ρ is a function of other constants of movement, such as the energy:
ρ(q, p) = ρ[H(q, p), ξ1 (q, p), ξ2 (q, p), . . .], (1.29)
where ξi are other constants of movement, i.e. {ξi , H} = 0. However, not all the
constant of movements are equally important. For example, in a free system (i.e. a
system which does not have any interaction or constraint with the surroundings),
there are seven fundamental constants of movement, which are energy, total momen-
tum and total angular momentum, associated to the symmetries of time translation,
spacial translation and rotations, respectively, associated to time invariance and the
homogeneity and isotropy of the space. These properties are extensive. However,
the presence of confining walls breaks the symmetry of the system with respect to
spacial translations and rotations, but keeping the conservation of energy. This im-
plies that energy will play an important role to determine ρ. As we will see, all the
probability densities associated to equilibrium Gibbs functions will be functions of
the Hamiltonian of the system.

1.5 The microcanonical ensemble


Now we consider that the macrostate corresponds to an equilibrium state of an
isolated system, i.e. that does not exchange either matter or energy with its sur-
roundings. Under these circumstances, the macroscopic energy E, the number of
particles N , and all other external extensive parameters Xα are constant (in an
hydrostastic system, X ≡ V , where V is the volume of the system). The Gibbs
ensemble associated to this state is the so-called microcanonical ensemble. The
microstates compatible with the macrostate are those where the microscopic energy
is E (although it is also possible to allow energies between E and E + ∆E, where
∆E  E), and the same values of N and Xα as the macrostate. In order to fully
characterize the ensemble, we have to define the probability distribution ρ for the
microstates. Although it would be desirable to obtain the expression of ρ from the
underlying mechanical description (this is the goal of the Ergodic Theory), we will
proceed in a pragmatic way, by postulating the expression of ρ. The probability dis-
tribution of an equilibrium state is stationary, and thus it is a constant of movement,
as we showed in the previous Section. Then, the second postulate of the Statistical
Physics, or the equal a priori probability postulate, can be stated as follows:
Introduction to the Statistical Physics. The microcanonical ensemble. 10

Second postulate: All the microstates compatible with the macrostate


of an isolated system have the same probability.

Note that this equality is restricted to the microstates which have energy E (or
with energies in [E, ∆E], with ∆E  E), being the probability zero otherwise.
The idea behind this postulate is that the system, in its time evolution, visits the
neighbourhood of any phase space point (except for a null-measure set of initial
conditions), in such a way that the time average of a property along its evolution
is equal to its average on the H(q, p) = E hypersurface. Mechanical systems which
satisfy this condition are called ergodic, and then ρ(q, p) can be identified as the
fraction of time the system visits the neighbourhood of (q, p). However, in most
cases it is impossible to show this property (or it can be shown not to be ergodic,
such as in ideal gases or systems of independent harmonic or weakly anharmonic
oscillators).
Let’s obtain the explicit expression of ρ. We will first consider that the energy
spectrum of the system is discrete (this is typical of its quantum mechanical de-
scription, but it can also be described at a classical level). Let’s call {Ej } the set
of energy levels, each of them with a degeneracy gj . We may assume that ∆E = 0,
so the possible energies of the macroscopic system would be any of the values of
Ej . Thus, the probability of a microstate of energy E = Ej would be pj = 1/gj .
However, this definition is not very convenient, since the dependence of pj on E
will be very “noisy”: it jumps from a finite to a null value as E deviates slightly
from any value of Ej , and in addition the values of gj can differ significantly even
for close energy levels. In order to avoid this, we define ρ in coarse-grained way, by
allowing for microstates with energies between the macroscopic energy Emacro and
Emacro + ∆E for the equilibrium macrostate of the isolated system, provided that
∆E  Emacro , but being much larger than the typical gaps in the energy spectrum.
Then,
1
pj = E ∈ [Emacro , Emacro + ∆E] (1.30)

= 0 otherwise. (1.31)
P
The normalization condition j pj leads to the following expression for Ω
X
Ω= gj , (1.32)
j:Emacro ≤Ej ≤Emacro +∆E

which is the number of microstates with energies between Emacro and Emacro + ∆E.
Note that it is an adimensional number.
Introduction to the Statistical Physics. The microcanonical ensemble. 11

Now we turn to the continuum classical case. In this case, an state would be that
the microstate is in the infinitesimal hypercube of the phase space dqdp ≡ [q1 , q1 +
dq1 ]×. . .×[qf , qf +dqf ]×[p1 , p1 +dp1 ]×. . .×[pf , pf +dpf ]. The probability of being in
this microstate will be ρ(q, p)dqdp, where we denote (q, p) as (q1 , . . . , qf , p1 , . . . , pf ).
By the second postulate, we assume that all the microstates of energy between
Emacro and Emacro + ∆E is the same. So, we can define:
ρ = Cθ(Emacro + ∆E − H(q, p))θ(H(q, p) − Emacro ), (1.33)
where θ(x) is the Heaviside’s step function, which is R1 if the argument is positive,
and 0 otherwise. From the normalization condition dqdpρ = 1 we get that the
constant C is
1
C=R . (1.34)
Emacro ≤H(q,p)≤Emacro +∆E
dqdp
We define the phase volume Γ(E) as:
Z
Γ(E) = dqdp, (1.35)
E0 ≤H(q,p)≤E

where E0 is the minimum value of H(q, p), i.e. the energy of its ground state. Thus
1
C= . (1.36)
Γ(Emacro + ∆E) − Γ(Emacro )
Usually, Γ(E) is a well-behaved function, so if ∆E  Emacro

∂Γ(E)
Γ(Emacro + ∆E) − Γ(Emacro ) ≈ ∆E ≡ Ω(Emacro )∆E, (1.37)
∂E

E=Emacro

where Ω = ∂Γ/∂E. Thus


1 θ(Emacro + ∆E − H(q, p))θ(H(q, p) − Emacro )
ρ(q, p) = . (1.38)
Ω(Emacro ) ∆E
In Classical Statistical Physics, ∆E → 0, so finally we obtain that
1
ρ(q, p) = δ(Emacro − H(q, p)), (1.39)
Ω(Emacro )
where δ(x) is the Dirac’s delta. From this expression, and taking into account the
normalization condition, we can define Ω as
Z
Ω(Emacro ) = dqdpδ(Emacro − H(q, p)). (1.40)
Introduction to the Statistical Physics. The microcanonical ensemble. 12

By comparison with the discrete energy spectrum case, Ω(E) is also known as the
number of microstates, although it is its energy density (the true analogue would be
Ω(E)∆E).
Two remarks are pertinent at this point. First, note that, unlike the discrete
energy spectrum case, Ω∆E is dimensional. This could be corrected if we multiply
the integrals over the phase space by a factor h−f , where h is a constant with units
of angular momentum. This is equivalent to identify the microstates as cells of size
hf which mesh the phase space. At the classical level, h is an arbitrary constant,
but we will see in next Units that h must be the Planck’s constant in order to match
with the underlying quantum formulation. Thus, we redefine Γ, Ω and ρ as
Z Z
1 ∂Γ 1
Γ(E) = f dqdp , Ω(E) = = f dqdpδ(E − H(q, p))
h E0 ≤H(q,p)≤E ∂E h
1
ρ(q, p) = f δ(E − H(q, p)). (1.41)
h Ω(E)
Secondly, in our derivation we assume that all the particles are distinct or, if they
are identical, we assume that they are distinguishable. This means that, if the
particles are labeled for a certain time, this labelling can be followed through the
time evolution. We will see that this is not appropriate for some cases, so we will
have to further correct the expressions of Γ and Ω for systems of indistinguishable
particles.
Now we are going to consider two simple examples where to evaluate these ex-
amples.
• The monodimensional harmonic oscillator. In this case f = 1, so strictly
speaking it is not a macroscopic system. However, we will solve this case
because this result will be useful to get the thermodynamic properties of a
macroscopic large number of harmonic oscillators. The Hamiltonian of the
system is
p2 K
H(q, p) = + q2, (1.42)
2m 2
so the value of E0 = 0. Thus
Z
Γ(E) = dqdp. (1.43)
0≤H(q,p)≤E

This is a conservative system, so the energy is conserved along the time evo-
lution. This means that the trajectory in phase space of the system is in the
following curve
p2 K p2 K
E ≡ 0 + q02 = + q2, (1.44)
2m 2 2m 2
Introduction to the Statistical Physics. The microcanonical ensemble. 13

1/2
(2mE)

1/2
(2E/k)
hΓ(Ε)

Figure 1.2: Phase volume Γ(E) of a monodimensional harmonic oscillator.

or equivalently
p2 q2
1= + , (1.45)
2mE 2E/K
which is the equation of an ellipse. So, the phase volume will be proportional
to the area of the ellipse associated to the largest energy E (see Fig. 1.2)
The area of the ellipse is πab, where a and b are the major an minor ellipse
semiaxes, respectively. So

1 √
r r
2E E m E
Γ(E) = π 2mE = 2π = , (1.46)
h K h K hν

where ν is the oscillator frequency. Thus, Ω = ∂Γ/∂E = 1/(hν).

• The monoatomic 3D ideal gas of N identical distinguishable parti-


cles. For this system, the Hamiltonian is
N
X p2i
H(q, p) = . (1.47)
i=1
2m

In principle, we should include a one-body potential term which accounts for


the interactions of the walls of the container with the gas, but this potential
usually only acts on particles close to the wall, vanishing inside the volume.
Introduction to the Statistical Physics. The microcanonical ensemble. 14

So, we will consider that its only effect will be to restrict the available positions
of the particles of the gas. Then
Z
1
Γ(E) = 3N p2
dr1 . . . drN dp1 . . . dpN . (1.48)
h 0≤ N
P i
i=1 2m ≤E

Note that the number of degrees of freedom is f = 3N , and that E0 = 0. As


the Hamiltonian does not R depend on the particles positions, we can make the
integrations on ri . As dr = V , with V being the gas volume, then
VN
Z
Γ(E) = 3N dp1,x dp1,y dp1,z . . . dpN,x dpN,y dpN,z .
h 0≤ N
P 2 2 2
i=1 (pi,x +pi,y +pi,z )≤2mE
(1.49)
The
√ integral is the volume of a 3N -dimensional hypersphere of radius R =
2mE. Let’s evaluate the volume v of an n-dimensional hypersphere of radius
R. This can be done in hyperspherical coordinates as
Z R
Rn
Z
n−1
v= r dr dSn = Sn , (1.50)
0 Sn n
where dSn is the infinitesimal solid angle associated to the angular coordinates,
and Sn is the total solid angle. In order to evaluate Sn , we use a trick. First,
we evaluate the following integral in Cartesian coordinates
Z ∞ Z ∞ Z ∞ n  Z ∞ n
−x21 −...−x2n −x2 −x2
dx1 . . . e = e dx = 2 e dx . (1.51)
−∞ −∞ −∞ 0

The same integral, in hyperspherical coordinates, and taking into account that
x21 + . . . x2n = r2 , is Z ∞
2
Sn rn−1 e−r dr. (1.52)
0
So  R n
∞ −x2
2 0 e dx
Sn = R ∞ n−1 −x2 . (1.53)
0
x e dx
The integrals in the previous expression can be represented by the gamma
function Γ(x) (not to be confused with the phase volume), which is defined as
Z ∞
Γ(x) = duux−1 e−u . (1.54)
0

Some properties of this function are



Γ(x + 1) = xΓ(x) , Γ(1) = 1 , Γ(1/2) = π , Γ(n + 1) = n! (n = 0, 1, . . .).
(1.55)
Introduction to the Statistical Physics. The microcanonical ensemble. 15

So, let’s consider the integral in the denominator of Eq. (1.53). Note that the
integral in the numerator is just a particular case of this integral, with n = 1. By
making the change of variable u = x2 , then
Z ∞
1 ∞ n −1 −u
Z
n−1 −x2 1 n
x e dx = u 2 e du ≡ Γ . (1.56)
0 2 0 2 2
Thus n n
Γ 12

π2
Sn = 1 n  = 1 n  . (1.57)
2
Γ 2 2
Γ 2
So, the volume of the n-dimensional hypersphere of radius R will be
n n
π 2 Rn π 2 Rn
v= n n =  n
. (1.58)
2
Γ 2
Γ 2
+ 1

Finally, the phase volume Ec. (1.49) is


3N
V N (2πmE) 2
Γ(E) = 3N , (1.59)
h Γ 3N 2
+ 1

where again we remark that the Γ function on the right-hand side must not be
confused with the phase volume. The number of microstates Ω(E) is
3N 3N
∂Γ V N (2πmE) 2 2
Γ(E)
Ω(E) = = 3N  = . (1.60)
∂E h EΓ 3N
2
E

So, for E/N . ∆E  E


Ω(E)∆E 3∆E
= E ∼ 1. (1.61)
Γ(E) 2N
Consequently !
3∆E
ln(Ω(E)∆E) = ln Γ(E) + ln E
. (1.62)
2N
Since Γ(E) ∼ E 3N/2 , ln Γ is of order of Avogadro’s number, which is much larger
than ln(N ∆E/E) ∼ 1, so

ln(Ω(E)∆E) ≈ ln Γ(E). (1.63)

This expression is true even if we assume that ∆E ∼ E, since then ln(N ∆E/E) ∼
ln N  N . This is a quite general result, since for most systems Γ(E) ∼ E νf , where
ν is a number of order of 1 and f is the number of degrees of freedom of the system.
Introduction to the Statistical Physics. The microcanonical ensemble. 16

1.6 Connection with the Thermodynamics


1.6.1 Microscopic interpretation of the heat and work
Let’s consider two systems A and A0 , each of them initially in equilibrium. We
place these systems in contact, so they can interact and, consequently, exchange
energy between them, but their union is isolated. In this way, A0 is the environment
for A and viceversa. After some time, A ∪ A0 will reach a new thermodynamic
equilibrium. The energy exchange in the equilibration process can be of two kinds:
heat and work. We will say that the systems are in pure thermal contact if they can
exchange energy, but the external extensive parameters of both A and A0 are kept
constant. During the equilibration process, the average energy of A and A0 , which
0 0
we will denote by E and E , respectively, will change by amounts ∆E and ∆E .
0
Since the system A ∪ A0 is isolated, E + E is constant, so
0
∆E + ∆E = 0. (1.64)

We will define the heat Q as the energy absorbed by system A, so Q = ∆E. Note
that if Q < 0, then the system loses energy, so we denote by −Q the heat given by
A. With this notation, Eq. (1.64) reads

Q + Q0 = 0 → Q = −Q0 , (1.65)

so heat is transferred from one system to the other when these are in pure thermal
contact.
Now, let’s suposse that any equilibrium state of A is compatible with any equilib-
rium state of A0 when placed in contact, provided their external extensive parameters
are kept constant. These systems are said to be thermally isolated. An example of
this situation is when A and A0 are separated by an adiabatic wall. Under these
circumstances, the only possible exchange of energy between A and A0 comes from
work, which involves changes in the external extensive parameters of A and A0 .
Then, A and A0 are said to be in mechanical interaction. Again, as A ∪ A0 is iso-
0
lated, ∆E + ∆E = 0 in the equilibration process. Now, we will define the work
W done by the system as W = −∆E. Alternatively, we can define −W = ∆E as
the work that the environment does on the system. With this notation, Eq. (1.64)
reads
−W − W 0 = 0 → W = −W 0 , (1.66)
so the work done by A on A0 is the work that the environment of A0 (i.e. A) does
on A0 .
Introduction to the Statistical Physics. The microcanonical ensemble. 17

General interactions involve both thermal and mechanical energy exchanges


 
∆E = ∆E thermal + ∆E mechanical , (1.67)

which, with the previous definitions, can be written as

∆E = Q − W. (1.68)

This is the first law statement, in which the internal energy is identified as the
averaged energy of the system. In differential form

dE = d¯Q − d¯W. (1.69)

Up to now, the equilibration process for the system A (and its environment
0
A ) is arbitrary. However, it is convenient to consider quasi-static processes which,
as we mentioned in the Introduction, are slow enough that each step in the path
can be considered to be almost in equilibrium. By slow enough we mean that the
perturbations which drive the process are so small that the typical relaxation times
are smaller than the time scale associated to the perturbation. Note that the larger
the perturbation is, the longer the relaxation time will be. Quasi-static processes
are idealizations of this situation, but they are convenient tools for our theoretical
arguments.
Let’s call X = {X1 , . . . , Xn } to the set of external extensive properties of the
system, such as volume, etc. The Hamiltonian of the system depends on these
quantities, so strictly speaking, H = H(q, p, X). Now, let’s vary a parameter Xα
quasi-statically for a thermally isolated sytem. So, the energy of any microstate of
the system will change as
∂H
dH = dXα . (1.70)
∂Xα
Note that if the system is initially isolated, so all the microstates correspond to the
same energy, this may not be true along the quasi-static process, since the variation
of H on X depends on the microstate. From Classical Mechanics, we define the
generalized force Yα conjugated to Xα as
∂H
Yα = − . (1.71)
∂Xα
Now we turn to the macroscopic system. The infinitesimal variation of the internal
energy, dE, is identified as the ensemble-average of dH, dH, which is

∂H
dE = dH = dXα = −Y α dXα . (1.72)
∂Xα
Introduction to the Statistical Physics. The microcanonical ensemble. 18

As the system is thermally isolated, dE = −d¯W , so

d¯W = Y α dXα , (1.73)

where Y α is the macroscopic generalized force. This result is not restricted to the
microcanonical ensemble as we only made use of the equilibrium condition (not the
specific macroscopy system constraints), so it is valid for any equilibrium Gibbs
ensemble. An example is the pressure p, which is the generalized force associated
to the system volume. Note that, as the process is quasi-stationary, the value of Y α
coincides with the external value.
If different external fields are changed simultaneously along the quasi-stationary
process, then Eq. (1.73) generalizes to
X
d¯W = Y α dXα . (1.74)
α

1.6.2 Reversibility and irreversibility


As we saw previously, the macrostate of an isolated system is determined by the
macroscopic energy E and the set of external extensive parameters Xα . The latter
can be regarded as constraints imposed on the system. Let’s see which effect has
the relaxation of these contraints. As an example, consider a system of volume 2V ,
divided by an adiabatic wall into two regions of volume V , one of them empty and
other filled with an ideal gas. The equilibrium state of this system will be that
of an isolated ideal gas within the filled region of volume V , being empty on the
other volume. Let’s call Ωi the number of microstates associated to this equilibrium
state. Now, the wall is moved out, so the gas spreads over the total volume of the
system until reaches the equilibrium (in our case, the associated to the gas occupying
the whole system of volume 2V ). The number of microstates at the final state, Ωf ,
increases since, in addition to the microstates compatible with the initial equilibrium
macrostate, we have additional microstates where there are particles in the originally
empty region. So, the elimination of the constraint leads to the relationship

Ωf (E) ≥ Ωi (E) (1.75)

for any value of E, and consequently

Γf (E) ≥ Γi (E). (1.76)

Suppose that, after equilibration, the wall is replaced. The replacement of the
constraint reduces again the number of microstates, since now the number of par-
ticles and energies on each region are the same as in the microstate in which the
Introduction to the Statistical Physics. The microcanonical ensemble. 19

system was just before the introduction of the wall. However, as the initial system
was equilibrated, the macroscopic energy and number of particles on each region are
E/2 and N/2, with fluctuations much smaller than these values. Thus, we will see
that ln Ωf ≈ ln Ωi , where the corrections are much smaller than ln Ωi . In practice,
this is equivalent to require that Ωf = Ωi . In order to obtain a macrostate such as
the original one, where the gas occupies only one region, we need to do a work to
compress the system, which is against the system isolation condition.
From these arguments, we can classify the processes for isolated systems in two
classes:

• Reversible processes: the number of microstates at the initial and final


equilibrium states is the same Ωi = Ωf (in the sense mentioned above). The
system can return to its initial state without any energy exchange. This means
that the system must be in equilibrium along the process, so it is quasi-static.

• Irreversible process: Ωf > Ωi , and the system cannot return to its initial
condition unless energy is supplied to the system. The gas expansion outlined
above is an example of irreversible process. Although typically the system
is out-of-equilibrium along the process, under some circumstances it can be
slow enough to be considered quasi-static. For example, we can consider the
irreversible expansion of the gas performed by a sequencial removal of a set
of adiabatic walls close each other (so the removal can be regarded as a small
perturbation) in the empty region, so slow that, after the removal of a wall,
the gas is let to equilibrate.

These results show that there must be a connection between the number of
microstates and the entropy, which can also only increase in irreversible processes
or stay constant for reversible processes in an isolated system. We will see below
how this connection is done.
We note that these arguments are only valid in isolated systems. On the other
hand, the time arrow which irreversibility introduces is a macroscopic concept. How
to reconcile this fact with the reversibility of the underlying microscopic mechanical
description, which is reversible in time, is an issue which is beyond the scope of this
course.

1.6.3 Entropy and temperature


In order to make clear the connection between the number of microstates and en-
tropy, we will first consider how the phase volume changes along a quasi-static
Introduction to the Statistical Physics. The microcanonical ensemble. 20

adiabatic process (i.e. the system interacts with its surroundings via pure mechan-
ical interactions). If we consider an infinitesimal portion of the process, dΓ can be
written as
∂Γ X ∂Γ
dΓ = dE + dXα . (1.77)
∂E α
∂X α

We know that Ω(E) = ∂Γ/∂E. On the other hand, the derivative of Γ with respect
to Xα is given by
Z
∂Γ 1 ∂
= f dqdp θ(E − H(q, p, X))θ(H(q, p, X) − E0 (X)), (1.78)
∂Xα h ∂Xα
where E0 is in general a function of X. The derivative of the Heaviside’s step
function with respect to its argument is the Dirac’s delta dθ(x)/dx = δ(x). So

∂θ(E − H(q, p, X)) ∂θ(E − H(q, p, X)) ∂H


=
∂Xα ∂H ∂Xα
∂θ(E − H(q, p, X)) ∂H
= −
∂E ∂Xα
∂H
= −δ(E − H(q, p, X)) . (1.79)
∂Xα
Analogously
 
∂θ(H(q, p, X) − E0 ) ∂θ(H(q, p, X) − E0 ) ∂H ∂θ(H(q, p, X) − E0 ) ∂E0
= +
∂Xα ∂H ∂Xα ∂E0 ∂Xα
 
∂θ(H(q, p, X) − E0 ) ∂E0 ∂H
= −
∂E0 ∂Xα ∂Xα

= −δ(H(q, p, X) − E0 ) (E0 − H(q, p, X)). (1.80)
∂Xα

Taking into account that θ(E − E0 ) = 1, we obtain that


Z  
∂Γ 1 ∂H
= dqdp − δ(E − H(q, p, X))
∂Xα hf ∂Xα
Z
1 ∂
− f dqdp (E0 − H(q, p, X))δ(H(q, p, X) − E0 ). (1.81)
h ∂Xα
The second term in the right-hind side of the previous equation vanishes, since the
integral is over the phase space region where H = E0 , so the integrand is zero. On
the other hand, from the definition of the average in the microcanonical ensemble
Introduction to the Statistical Physics. The microcanonical ensemble. 21

and of the macroscopic generalized force, the first term is Y α Ω(E, X). Now, the
total variation of the phase volume along the quasi-static adiabatic process is
" #
X
dΓ(E, X) = Ω(E, X) dE + Y α dXα , (1.82)
α

which vanishes from the definition of the work in quasi-static adiabatic processes.
As a consequence, Γ(E, X) is invariant along these processes, as it happens with the
entropy.
Now, let’s consider a general quasi-static process. The variation of the phase
volume along an infinitesimal portion of the process is still given by Eq. (1.82), but
this expression does not vanishes in general. If we divide Eq. (1.82) by Γ, we obtain
" #
dΓ Ω(E, X) X
d ln Γ(E, X) = = dE + Y α dXα
Γ Γ(E, X)
  " # α  " #
1 ∂Γ X ∂ ln Γ X
= dE + Y α dXα = dE + Y α dXα ,(1.83)
Γ ∂E X α
∂E X α

where the subscripts on the partial derivative


P means the variables which are kept
constant. So, as d¯Q = dE + d¯W = dE + α Y α dXα , then
 
∂ ln Γ
d ln Γ(E, X) = d¯Q. (1.84)
∂E X

As the left-hand side of this equation is an exact differential, (∂ ln Γ/dE)X is an


integrating factor for d¯Q. Actually, comparing this expression with the definition of
the entropy dS = d¯Q/T , it is reasonable to define the entropy S as

S = kB ln Γ, (1.85)

so the (absolute) temperature T is


   −1
∂ ln Γ
T = kB . (1.86)
∂E X
From these expressions, we see that entropy and temperature are intrisically macro-
scopic properties which cannot be associated to a given microstate but to the whole
Gibbs ensemble. However, temperature has another definitions (for example, from
the average of the kinetic energy) which have a microscopic analogue. Note that,
as usually Γ ∼ E νf , which is an increasing function, the temperature is generally
Introduction to the Statistical Physics. The microcanonical ensemble. 22

positive. However, there are situation where Γ is a decreasing function of E (for


example, if the allowed energies of the system are bounded from above) which can
have negative absolute energies. The constant kB is determined by the units of the
temperature, and it is known as the Boltzmann constant. For simplicity, we define
the parameter β as  
1 ∂ ln Γ
β= = . (1.87)
kB T ∂E X
Now we can check different thermodynamic relationships. First, the temperature
and entropy definitions Eqs. (1.86) and (1.85) lead to this well-known expression
 
∂E
T = . (1.88)
∂S X
On the other hand, as d ln Γ can be written as
  X  ∂ ln Γ 
∂ ln Γ
d ln Γ = dE + dXα . (1.89)
∂E X α
∂X α E,X β6=α

Comparison with Eq. (1.83) leads to the following expression


   
∂ ln Γ ∂ ln Γ
=Yα = βY α , (1.90)
∂Xα E,Xβ6=α ∂E X
or equivalently  
∂S Yα
= . (1.91)
∂Xα E,Xβ6=α T
Finally, we mention that the entropy has an alternative expression in terms of Ω.
As we showed in the derivation of the phase volume for an ideal gas, most systems
satisfy that ln Ω∆E ≈ ln Γ, so entropy can be defined alternatively as
S = kB ln(Ω∆E), (1.92)
where in many cases the term ∆E is dropped. So, temperature and generalized
forces can be obtained by previous formulae, just substituting Γ by Ω. On the other
hand, the entropy for systems with a discrete energy spectrum follow an expression
closer to Eq. (1.92)
S = kB ln Ω (1.93)
where we recall that Ω is the number of microstates with energy between E and
E + ∆E As for the continuum case, sometimes it is possible to assume that ∆E =
0 (for example, when the degeneracy of the energy levels gj can be analytically
continued to a smooth function of the energy by interpolation).
Introduction to the Statistical Physics. The microcanonical ensemble. 23

1.6.4 Additivity of the entropy


One of the properties of the thermodynamic entropy is that it is an extensive prop-
erty, that is, if a macroscopic system in equibrium is considered to be the union of
two subsystems, the entropy is additive, i.e. the entropy S of the whole system is
the sum of the entropies S1 and S2 of the two subsystems, S = S1 + S2 . We will
check if this property is also obeyed by the entropy defined in the previous Section.
Let’s first consider two systems A1 and A2 which are in pure thermal contact
(i.e. separated by a diathermal wall), such that A1 ∪ A2 is isolated. The external
extensive properties of each system are kept constant, which in addition are closed
(i.e. there is no matter exchange between them). We denote by (q, p) the generalized
set of coordinates and momenta of system A1 , and (Q, P ) to the corresponding set
of system A2 . Any other properties will be labeled by 1 or 2 if they correspond to
system A1 and A2 , respectively.
The Hamiltonian H of the system A1 ∪ A2 can be written as
H(q, Q, p, P ) = H1 (q, p) + H2 (Q, P ) + H12 (q, Q, p, P ), (1.94)
where H12 is the interaction energy between A1 and A2 . This term is needed for
the energy exchange between systems, but usually is much smaller than H1 and H2 ,
so it can be neglected. Note that the energies E1 and E2 are proportional to the
volumes of A1 and A2 , but for short-ranged interactions H12 scales like the surface
of the contact boundary between A1 and A2 . As A1 ∪ A2 is isolated, the equilibrium
probability distribution is written as
1
ρ(q, Q, p, P ) = δ(E − H1 (q, p) − H2 (Q, P )), (1.95)
hf Ω(E)
where the total number of degrees of freedom f = f1 + f2 . From this expression,
the marginal probability distribution of the A1 phase space coordinates is
Z
1 1
ρ1 (q, p) = f dQdP δ(E − H1 (q, p) − H2 (Q, P )) = f1 Ω2 (E − H1 (q, p)),
h Ω(E) h Ω(E)
(1.96)
where Ω2 is the number of microstates of system A2 . Note that this probability
density is not the microcanonical one. The probability density for E1 reads
Z
ω(E1 ) = dqdpδ(H1 (q, p) − E1 )ρ1 (q, p)
Ω2 (E − E1 ) 1
Z
= dqdpδ(E1 − H1 (q, p))
Ω(E) hf1
Ω1 (E1 )Ω2 (E − E1 )
= . (1.97)
Ω(E)
Introduction to the Statistical Physics. The microcanonical ensemble. 24

From the normalization condition of ω(E) we obtain the composition rule for the
number of microstates
Z E−(E0 )2
Ω(E) = dE1 Ω1 (E1 )Ω2 (E − E1 ). (1.98)
(E0 )1

In general, Ω1 and Ω2 are both rapidly increasing functions of energy (for example,
Ω ∼ E νf ), so their product will have a very sharp maximum which gives the most
probable value of E1 , Ẽ1 (see Fig. 1.3). Thus, the integral in Eq. (1.98) can be
approximated as
Ω(E) ≈ Ω1 (Ẽ1 )Ω2 (E − Ẽ1 )nσE1 , (1.99)
q
2
where n is of order of unity and σE1 = E12 − E1 . Normally σE1  Ẽ1 , so taking
logarithms

ln Ω(E) ≈ ln Ω1 (Ẽ1 )+ln Ω2 (E − Ẽ1 )+ln(nσE1 ) ≈ ln Ω1 (Ẽ1 )+ln Ω2 (E − Ẽ1 ) (1.100)

since ln Ω1 ∼ f1 ln Ẽ1  ln(nσE1 ). The most probable value of E1 is determined


from the condition

∂ ln ω(E1 )
∂ ln Ω1 (E1 )
∂ ln Ω2 (E − E1 )
= +
∂E1 ∂E1 ∂E1


E1 =Ẽ1
E1 =Ẽ1 E1 =Ẽ1

∂ ln Ω1 (E1 )
∂ ln Ω2 (E2 )

= − = 0. (1.101)
∂E1 ∂E2


E1 =Ẽ1 E2 =E−Ẽ1

From the definition of the absolute temperature and the fact that ln Γ ≈ ln Ω, this
expression is equivalent to
T1 = T2 . (1.102)
Now, we can set the additivity of entropy. Recall that S = kB ln Ω, so we can define
the entropy of the combined isolated system as

S(E) = S1 (E1 ) + S2 (E2 ), (1.103)

with E = E1 + E2 as isolation imposes. This is a generalization of the expression


obtained above (valid, for example, when the systems are separated by an adiabatic
wall), although some caution must be taken: entropy, as defined here, cannot be
used if the system is not composed by subsystems which are in equilibrium. The
system A1 ∪ A2 will be in thermal equilibrium if E1 = Ẽ1 which maximizes ω(E1 ),
or equivalently S1 (E1 ) + S2 (E2 ). As we showed before, this condition is equivalent
to the equality of temperatures of both systems.
Introduction to the Statistical Physics. The microcanonical ensemble. 25

Ω1(Ε1)
Ω2(Ε−Ε1)
Ω1(Ε1)Ω2(Ε−Ε1)


σΕ
1

*
E1 E1

Figure 1.3: Number of microstates of a combined system.

Finally, we mention that these results are valid for combined systems in which
particles in A1 are distinguishable from those in A2 and that the interaction between
A1 and A2 is weak (which is equivalent to the condition H12  H1 ). In particular,
the composition rule Eq. (1.98) is verified under these circumstances.
Now let’s suppose that A1 and A2 can also interact mechanically. To illustrate
this, we will consider that the volume V is the only external parameter for each
system. As the combined system is isolated, and V is an extensive property, V =
V1 + V2 . We still assume that both A1 and A2 are closed. Now, the probability
density of the combined system is
1
ρ(q, Q, p, P ) = f δ(E − H1 (q, p, V1 ) − H2 (Q, P, V − V1 )). (1.104)
h Ω(E, V )
Integrating out the phase space coordinates of A2 , we get that
1
ρ1 (q, p) = f1 Ω2 (E − H1 (q, p, V1 ), V − V1 ) (1.105)
h Ω(E, V )
and the energy and volume probability density ω reads
Ω1 (E1 , V1 )Ω2 (E2 , V2 )
ω(E1 , V1 ) = . (1.106)
Ω(E, V )
The composition rule in this case is
Z V Z E−(E0 )2
Ω(E, V ) = dV1 Ω1 (E1 , V1 )Ω2 (E − E1 , V − V1 ). (1.107)
0 (E0 )1
Introduction to the Statistical Physics. The microcanonical ensemble. 26

As in the case of pure thermal interaction, Ω is a rapidly increasing function on both


E1 and V1 (in the ideal gas case, Ω ∼ V N ). So, Eq. (1.107) can be approximated as
ln Ω(E, V ) ≈ ln Ω1 (Ẽ1 , Ṽ1 ) + ln Ω2 (E − Ẽ1 , V − Ṽ1 ), (1.108)
where Ẽ1 and V˜1 are the values of E1 and V1 which maximizes ln ω, that is

∂ ln ω(E1 , V1 ) ∂ ln Ω1 (E1 , V1 ) ∂ ln Ω2 (E2 , V2 )
= − =0
∂E1 ∂E1 ∂E2


E1 =Ẽ1 ,V1 =Ṽ1 E1 =Ẽ1 ,V1 =Ṽ1 E2 =E−Ẽ1 ,V2 =V −Ṽ1
(1.109)
and

∂ ln ω(E1 , V1 ) ∂ ln Ω1 (E1 , V1 ) ∂ ln Ω2 (E2 , V2 )
= − = 0.
∂V1 ∂V1 ∂V2


E1 =Ẽ1 ,V1 =Ṽ1 E1 =Ẽ1 ,V1 =Ṽ1 E2 =E−Ẽ1 ,V2 =V −Ṽ1
(1.110)
As above, Eq. (1.109) reduces to T1 = T2 . On the other hand, ∂ ln Ω/∂V = βp,
where the pressure p is the generalized force conjugated to V . Thus, Eq. (1.110)
reduces to
(βp)1 = (βp)2 . (1.111)
Both results can be combined as
T1 = T2 , p1 = p2 . (1.112)
So, two closed systems in thermodynamic equilibrium must have the same tem-
perature, which assures the thermal equilibrium, and pressure (in general all the
generalized forces), which assures the mechanical equilibrium. Again, we can write
the total entropy of the combined system as S = S1 (E1 , V1 ) + S2 (E2 , V2 ), and the
thermodynamic equilibrium condition implies that E1 and V1 must have the values
which maximize the entropy of the combined system.

1.6.5 The ideal gas and the Gibbs’s paradox


Now we will apply the formalism outlined above to the three-dimensional ideal gas of
N particles. We obtained previously that the phase volume of the ideal gas enclosed
on a volume V is 3N
V N (2πmE) 2
Γ(E) = 3N , (1.113)
h Γ 3N 2
+ 1
from which
 
3N 3N
ln Γ(E) = N ln V + ln E − log Γ + 1 + N σ̃, (1.114)
2 2
Introduction to the Statistical Physics. The microcanonical ensemble. 27

where σ is defined as  
3 2πm
σ̃ = ln . (1.115)
2 h2
So,
∂ ln Γ 3N
β= = , (1.116)
∂E 2E
or equivalently
kB T
E = 3N , (1.117)
2
which is a particular case of the equipartition theorem (which we will see in next
Unit), since each degree of freedom carries an energy kB T /2. On the other hand,
the pressure p is obtained as
1 ∂ ln Γ N kB T
p= = . (1.118)
β ∂V V

If we define the mole number n = N/NA , where NA is Avogadro’s number, then the
equation of state Eq. (1.118) can be expressed as the well-known ideal gas equation
of state
nRT
p= , (1.119)
V
where R = NA kB is the constant of the ideal gases. So, in the International System
of Units, kB = R/NA = 1.38 × 10−23 J/K.
The entropy of the system is obtained from Eq. (1.114) as
   
3 3N
S = N kB ln V + ln E + σ̃ − kB ln Γ +1 . (1.120)
2 2

The last term can be simplified by Stirling’s formula. From properties of the gamma
function, Γ(1 + x) = xΓ(x). We can repeat the application this property, so Γ(1 +
x) = x(x − 1)(x − 2) . . . x0 Γ(x0 ), where 0 < x0 ≤ 1. If we take logarithms

ln Γ(1 + x) = ln x + ln(x − 1) + . . . + ln x0 + ln Γ(x0 ). (1.121)

If x is large, we can replace the sum on the right-hand side by an integral


Z x
ln Γ(1 + x) ≈ dt ln t + ln Γ(x0 ) = x ln x − x − x0 ln x0 + x0 + ln Γ(x0 ) ≈ x ln x − x,
x0
(1.122)
Introduction to the Statistical Physics. The microcanonical ensemble. 28

which is Stirling’s formula. So, Eq. (1.120) can be written as


   
3 3N kB 3N 3N kB 3 2E
S = N kB ln V + ln E + σ̃ − ln + = N kB ln V + ln +σ ,
2 2 2 2 2 3N kB
(1.123)
where σ = σ̃ + (3/2)(1 + ln kB ). Now, by using Eq. (1.117), we get that
 
3
S = N kB ln T + ln V + σ . (1.124)
2
Now we consider the situation discussed in the section on reversibility and irre-
versibility. Let’s consider we have α ideal gases enclosed in regions on volume V
which are in thermodynamic equilibrium, so all of them will have the same tem-
perature and numerical density N/V (to ensure that pressure is the same). So, the
entropy of the combined system is
α  
X 3
Si = Si = αN kB ln T + ln V + σ . (1.125)
i=1
2

Now, we remove the walls which separate the systems. The macroscopic situation
has not changed, but if we evaluate the entropy of the gas of αN particles enclosed
in a volume αV at the temperature T , we obtain that
 
3
Sf = αN kB ln T + ln V + ln α + σ = Si + α ln αN kB (1.126)
2
instead the expected value Sf = Si from additivity. This is the Gibbs’s paradox.
Note that the situation is different if each compartment is filled initially by a different
gas (still with the same temperature and number density): after the removal of the
walls, the combined system will evolve towards an equilibrium state where each
component is evenly distributed over the whole system. This is an irreversible
process, where the entropy change is given by the mixing entropy α ln αN kB . So
there is something fundamentally different when we mix the same gas. A way to
solve the Gibbs’s paradox is modifying the definition of the entropy of the ideal gas
to include an additional factor −N kB ln N + N kB C, where C is a constant, so now
the corrected entropy
 
3 V 0
Scorrected = S − N kB ln N + N kB C = N kB ln T + ln + σ , (1.127)
2 N

where σ 0 = σ + C. A way to introduce this correction is by including an (N !)−1


factor in the definitions of Ω, Γ and ρ (note that, for large N , ln N ! ≈ N ln N − N ,
Introduction to the Statistical Physics. The microcanonical ensemble. 29

so C = 1 in the previous expression):


Z Z
0 Ω 1 0 Γ 1
Ω = = f dqdpδ(E − H(q, p)) , Γ = = f dqdp
N! h N! N! h N! E0 ≤H(q,p)≤E
(1.128)
1
ρ0 (q, p) = δ(E − H(q, p)). (1.129)
hf N !Ω0
The underlying concept is the indistinguishability of identical particles: we cannot
say macroscopically which set of particles is in a certain macroscopic volume, we only
can say how many particles we have in that volume. From a fundamental point of
view, the origin of indistinguishability can be traced to the quantum indistinguisha-
bility of identical particles, as we will see in Unit 5. Taking the classical limit of the
underlying quantum ideal gas, in addition to an explanation to the hf factor, it is
shown that the number of microstates, when approximated by the integral on the
phase space, is overcounted by a N ! factor, since each physically-meaning microstate
is characterized by N values of positions occupied by particles and their associated
momenta, regardless which particle is placed in which position. However, there is no
need to consider the quantum indistinguishability, but the weaker concept of indis-
tinguishability mentioned above is enough to explain the introduction of the (N !)−1
factor. In order to show that, let’s consider an isolated ideal system of identical
distinguishable particles with total energy E which can move on a volume V , in
which we consider a control volume V1 . We want to monitor the number of particles
which can be in the control volume. In this sense, the subsystem of particles inside
the control volume can be regarded as an open system, since it exchanges matter
with its surroundings. The probability density of the system is given by expression
(1.95). We will proceed as in the case of two systems at pure thermal contact, but
assuming that A1 is certain set of N1 particles (for example, the particles labeled as
1, 2, . . . , N1 ) enclosed in volume V1 , and A2 the rest of the system particles. As the
system is ideal, we can neglect interactions between particles, so
H(q, Q, p, P ) = HN1 (q, p) + HN −N1 (Q, P ), (1.130)
where (q, p) = (q1 , . . . , qN1 , p1 , . . . , pN1 ) and (Q, P ) = (qN1 +1 , . . . , qN , pN1 +1 , . . . , pN ).
We obtain in a similar way as we did previously that the probability density for E1
in the system of N1 particles, ω {1,...,N1 } (E1 , N1 ), is
Ω1 (E1 , N1 )Ω2 (E − E1 , N − N1 )
ω {1,...,N1 } (E1 , N1 ) = , (1.131)
Ω(E, N )
where Z
1
Ω1 (E1 , N1 ) = f1 dq1 . . . dqN1 dp1 . . . dpN1 δ(E1 − HN1 ) (1.132)
h
Introduction to the Statistical Physics. The microcanonical ensemble. 30

and
Z
1
Ω2 (E − E1 , N − N1 ) = f1 dqN1 +1 . . . dqN dpN1 +1 . . . dpN δ(E − E1 − HN −N1 ).
h
(1.133)
However, in this expression it is clear that we know which particles are in volume V1
(in our case, the first N1 particles). The probability density of having N1 particles
(regardless their identity) with energy E1 is obtained by summing the contributions
of the form (1.131) where the set of N1 particles is replaced by any of the selections
P of N1 different particles from the total number N
X
ω(E1 , N1 ) = ω P (E1 , N1 ), (1.134)
P

where Ω1 and Ω2 are obtained in a similar way as in Eqs. (1.132) and (1.133)
substituting accordingly the labels in the integrals. For example, if P = {2, . . . , N1 +
1}, then
Z
1
Ω1 (E1 , N1 ) = f1 dq2 . . . dqN1 +1 dp2 . . . dpN1 +1 δ(E1 − HN0 1 ), (1.135)
h
with HN0 1 being the new Hamiltonian associated to the particles in P, and
Z
1
Ω2 (E − E1 , N − N1 ) = f1 dq1 dqN1 +2 . . . dqN dp1 dpN1 +2 . . . dpN δ(E − E1 − HN0 −N1 ).
h
(1.136)
However, as the particles are identical, these expressions are exactly the same for all
P. On the other hand, the total number of P is equal to the number of N1 groups
we can form from N numbers, which is given by
N!
#P = . (1.137)
N1 !(N − N1 )!
So
N! Ω1 (E1 , N1 )Ω2 (E − E1 , N − N1 )
ω(E1 , N1 ) =
N1 !(N − N1 )! Ω(E, N )
N ! Ω1 (E1 , N1 ) Ω2 (E − E1 , N − N1 )
= . (1.138)
Ω(E, N ) N1 ! (N − N1 )!
The normalization condition for ω is
X N Z E−(E0 )2
ω(E1 , N1 )dE1 = 1, (1.139)
N1 =0 (E0 )1
Introduction to the Statistical Physics. The microcanonical ensemble. 31

so the composition rule for this situation is


N Z E−(E0 )2
Ω(E, N ) X Ω1 (E1 , N1 ) Ω2 (E − E1 , N − N1 )
= dE1 . (1.140)
N! N =0 (E0 )1
N1 ! (N − N1 )!
1

Again the integrand has a sharp maximum for values of E1 = E1∗ and N1 = N1∗
which are determined by the conditions
   
∂ Ω1 (E1 , N1 ) ∂ Ω(E2 , N2 )
ln − ln =0
∂E1 N1 ! ∂E2 N2 !


E1 =Ẽ1 ,N1 =Ñ1 E2 =E−Ẽ1 ,N2 =N −Ñ1
(1.141)
and
   
∂ Ω1 (E1 , N1 ) ∂ Ω(E2 , N2 )
ln − ln = 0,

∂N1 N1 ! ∂N2 N2 !


E1 =Ẽ1 ,N1 =Ñ1 E2 =E−Ẽ1 ,N2 =N −Ñ1
(1.142)
where we treat N as a continuous variable. These results suggest that the correct
definition of the entropy is
   
Ω(E, N ) Γ(E, N )
S = kB ln ≈ kB ln . (1.143)
N! N!

Then, the thermodynamic conditions (1.141) and (1.142) are equivalent to

β1 = β2 , (βµ)1 = (βµ)2 , (1.144)

or alternatively
T1 = T2 , µ1 = µ2 , (1.145)
where the chemical potential µ is defined as
   
1 ∂ Ω ∂S
µ=− ln = −T . (1.146)
β ∂N N ! E,X ∂N E,X

Again, the additivity of entropy can be shown in a similar way as in the cases we
considered previously. Thus, hereafter we will use the corrected expressions of Ω0 , Γ0
and ρ0 in Eq. (1.129) as Ω, Γ and ρ for systems of identical indistinguishible particles
(dropping the prime). On the other hand, if the system is composed by g sets of
particles which can be distinguished (for example, different chemical species, or even
Introduction to the Statistical Physics. The microcanonical ensemble. 32

different sets of identical particles which are forced to be in disjoint volumes), these
expressions can be easily generalized as
Z Z
1 1
Ω = f Qg dqdpδ(E − H(q, p)) , Γ = f Qg dqdp
h i=1 Ni ! h i=1 Ni ! E0 ≤H(q,p)≤E
(1.147)
1
ρ(q, p) = f Qg δ(E − H(q, p)), (1.148)
h i=1 Ni !Ω

where Ni is the number of particles in the set i. It is easy to show the results
obtained for two closed systems, each of them with fixed number of particles, in
thermal (and mechanical) contact, are still valid with these definitions, regardless if
the particles in each system are indistinguishable or not.

1.7 Bibliography
1. J. Javier Brey Abalo, J. de la Rubia Pacheco y J. de la Rubia Sánchez,
Mecánica Estadı́stica (UNED, 2011).

2. F. Reif., Fundamentals of Statistical and Thermal Physics (Waveland Pr. Inc.,


2008).

3. L. E. Reichl, A modern course in Statistical Physics (John Wiley and Sons,


1997).

You might also like