Statmech Notes

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 54

Lecture Notes: Statistical Mechanics

S Chaturvedi
April 19, 2016

Contents
1 Specification of a state in classical mechanics 3

2 Specification of a state in quantum mechanics 4

3 Composite systems in classical mechanics 6

4 Composite systems in quantum mechanics 7


4.1 Entanglement . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
4.2 Indistinguishability: Maxwell Boltzmann, Bose and Fermi statis-
tics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
4.3 Fock description . . . . . . . . . . . . . . . . . . . . . . . . . . 10

5 Review of Thermodynamics 14
5.1 Scaling properties . . . . . . . . . . . . . . . . . . . . . . . . . 16
5.2 Maxwell relations . . . . . . . . . . . . . . . . . . . . . . . . . 18
5.3 Scope of thermodynamics . . . . . . . . . . . . . . . . . . . . 18

6 Classical Statistical Mechanics 19


6.1 Ergodic Hypothesis . . . . . . . . . . . . . . . . . . . . . . . . 19
6.2 Microcanonical Ensemble . . . . . . . . . . . . . . . . . . . . . 19
6.3 Canonical ensemble . . . . . . . . . . . . . . . . . . . . . . . . 20
6.4 Grand canonical ensemble . . . . . . . . . . . . . . . . . . . . 20
6.5 Partition functions . . . . . . . . . . . . . . . . . . . . . . . . 21
6.6 Indistinguishability in the classical context . . . . . . . . . . . 21
6.7 Relation to thermodynamic potentials . . . . . . . . . . . . . . 23
6.8 Transition to Quantum Statistics . . . . . . . . . . . . . . . . 25
6.9 Grand or Landau potential for 1 M-B, Bose and Fermi statistics 28

7 Ideal Maxwell-Boltzmann gas : Classical 31


7.1 Microcanonical partition function and Entropy . . . . . . . . . 31
7.2 Canonical partition function and the Helmholtz free energy . . 32
7.3 Grand canonical partition function and the grand potential . . 33
15 Specific heat of solids 45
15.1 Einstein Model . . . . . . . . . . . . . . . . . . . . . . . . . . 45
15.2 Debye Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

16 Paramagnetism 47

17 Appendix A: Tensor products of Hilbert spaces 49

18 Appendix B: Jacobian identities 50

19 Appendix C: Gaussian integrals, Surface area and Volume of


a sphere in n-dimensions, Stirling formula 51

2
1 Specification of a state in classical mechan-
ics
Classical mechanics began with Newton. Newtonian mechanics deal with
particle motion. For a single particle of mass m to solve Newton’s equations
of motion we need to know its initial position q and its initial velocity q̇.
We may thus say that the state of a system is specified by the pair (q, q̇).
Given the initial state, the states at later times are completely determined
by the Newton’s equations of motion. Newtonian mechanics at the hands of
Lagrange and Hamilton eventually eventually emerged as Classical mechan-
ics, a self contained discipline with a rich mathematical structure capable of
handling dynamics of both particle and fields. In the lagrangian formulation
the state of a particle is specified, as in Newtonian mechanics, by its position
and velocity and the dynamics is determined by the lagrangian equations of
motion  
d ∂L ∂L
− =0
dt ∂ q̇ ∂q
where L(q, q̇) is referred to as the Lagrangian of the system. In the Hamilto-
nian formulation on the other hand the state of a particle is specified by its
position q and its conjugate momentum p. The beauty of the Hamiltonian
formulation is that given the configuration space Q, the manifold of values
which the coordinate can take, it defines the corresponding momenta and
hence the corresponding phase space through a well defined mathematical
construction. Thus for instance, for a particle whose confuaration space Q
is the entire real line R, the momenta p also take values in R and the phase
space can be pictured as a two dimensional plane called the phase plane.

3
A (pure) state of a particle is specified by the pair (q, p) i.e. by a point in
the phase plane.
Given the state of the system at an initial time its state at later times is
completely determined by the Hamilton’s equations of motion:

q̇ = {q, H(q, p)} ; ṗ = {p, H(q, p)}

where H(q, p) denotes the system Hamiltonian and {, } the Poisson bracket
defined, for any pair of phase space functions f (q, p) and g(q, p), as follows:
∂f ∂g ∂f ∂g
{f (q, p), g(p, q}} = −
∂q ∂p ∂p ∂q
As is well known they satisfy antisymmetry, linearity, Leibniz rule and
the Jacobi identity.The Poisson brackets between q and p are referred to as
the fundamental Poisson brackets and for them we have:

{q, p} = 1, {q, q} = {p, p} = 0

More generally a (mixed) state of a classical system can be defined by spec-


ifying a probability distribution ρ(q, p) over the phase space:
Z ∞ Z ∞
ρ(q, p) : pointwise positive; dq dpρ(q, p) = 1
−∞ −∞

Its evolution is governed by

ρ̇(q, p) = {ρ(q, p), H(q, p)}

2 Specification of a state in quantum mechan-


ics
Quantum description of a physical system proceeds by associating an ap-
propriate complex Hilbert space H with the system. Recall that a complex
Hilbert space is a vector space over the complex field equipped with a scalar
product.The Hilbert space could be of finite or infinite dimensions. When
finite, we will denote the dimension of H by d. While physical systems such
as a harmonic oscillator, particle in box require infinite dimensional Hilbert
spaces for their description, those such as a two level atom, a spin-1/2 par-
ticle are described by two dimensional Hilbert spaces. A physical system

4
described by a 2(or d) dimensional Hilbert space is generically referred to
as a qubit or (qudit). In the notation due to Dirac, which we would fol-
low, elements of H, often referred to as state vectors or simply vectors, are
generically denoted by |ψi, |φi etc. and the scalar product of |ψi with |φi
by hφ||ψi or more economically as hφ|ψi. (The object hφ| is called the bra
corresponding to the ket |φi and, consistent with the properties of the scalar
product, if |φi = α|µi + β|λi then hφ| = α∗ hµ| + β ∗ hλ|) The space between
| and i could be viewed as a nameplate for the vector where one puts the
labels such as ψ, φ etc so as to distinguish one vector from another. Often
the the labels are chosen in a more meaningful way so that besides playing
their usual role of distinguishing one vector from another they also help us
specify the vectors in a constructive manner. We could thus speak of a vector
|λi ∈ H with the tacit understanding that this vector is an eigenvector of
some specific operator on H corresponding to the eigenvalue λ. The subset of
all normalised vectors |ψi ∈ H such that hψ|ψi = 1 is denoted by B, the ‘unit
ball’ in H. Note that B unlike H is not a vector space: linear combination
of two normalized vectors is not necessarily a normalized vector.
Given any two vectors |ψi and |φi in H, one can in a natural way construct
out of them an operator A = |ψihφ| on H. Its action on any element |χi of
H is given by A|χi = |ψihφ|χii.e = hφ|χi|ψi. The trace of this operator is,
Tr[A] = Tr[|ψihφ|] = hφ|ψi and its adjoint A† is clearly |φihψ|. When |ψi =
|φi, the operator P ≡ |ψihψ| is self adjoint. Further, if |ψi is a normalized
vector i.e. an element of B, we have P 2 = P and Tr[P] = hψ|ψi = 1 i.e. P
is a rank one projection operator on H.
The the normalized vectors |ψi in H ( elements of the unit ball B in H)
are not in one to one correspondence with the physical states of the system
. There are infinitely many normalized vectors {eiφ |ψi} which correspond to
the same physical state. To make this correspondence one to one we pass
from normalized vectors to their projection operators,

physical state ↔ |ψihψ|

and say that the density operator ρ ≡ |ψihψ| describes a pure state of the
physical system. We can equivalently characterize the pure states by the
density operators ρ obeying

ρ† = ρ, ρ2 = ρ, Tr[ρ] = 1.

Note that by construction ρ is a positive operator. ( An operator A on H is

5
said to be positive if hψ|Aψi ≥ 0 for all |ψi ∈ H and use the notation A ≥ 0
for such operators)
Given two pure states ρ1 , ρ2 , their convex sum

ρ = p1 ρ1 + p2 ρ2 ; p1 , p2 ≥ 0, p1 + p2 = 1

is also a trace one positive operator and we say that such a ρ describes a
mixed state, a statistical mixture of the pure states ρ1 and ρ2 . ( It should
be borne in mind that a given mixed state ρ can be viewed as a statistical
mixture of pure states in infinitely many ways )
To put it all together specifying the physical state of a quantum mechan-
ical system means giving a trace one positive ( and hence necessarily self
adjoint ) operator ρ on H

Physical states ↔ ρ ≥ 0, Tr[ρ] = 1

States can be divided into two categories:

% Pure if ρ2 = ρ
ρ
& mixed if ρ2 6= ρ

3 Composite systems in classical mechanics


In classical mechanics, the phase space describing a composite system A + B
consisting of two subsystems A and B is simply the cartesian product of
the phase spaces describing A and B. A general state of the system is then
a probability distribution ρA+B on this cartesian product. Given such a
probability distribution, one can compute the reduced or marginal distri-
butions ρA (ρB ) for A (B) by integrating it over the phase space varibles
associated with B (A). Given a pure state of a composite system, the re-
duced distributions thereof will also correspond to pure states of the sub-
systems - it will never happen that a complete knowledge of the whole does
not imply a complete knowledge of its parts. Thus for two particle system
ρA+B (q1 , p1 , q2 , p2 ) = δ(q1 − q)δ(p1 − p)δ(q2 − q 0 )δ(p2 − p0 ) the two marginals
ρA , ρB would be δ(q1 − q)δ(p1 − p) and δ(q2 − q 0 )δ(p2 − p0 ) also describe pure
states of A and B. This is not so in quantum mechanics as we shall see.

6
4 Composite systems in quantum mechanics
In quantum mechanics, the Hilbert space HA+B describing a composite sys-
tem A + B consisting of two subsystems A and B is obtained by taking
the tensor product of the Hilbert spaces HA and HB describing A and B
respectively:
HA+B = HA ⊗ HB

4.1 Entanglement
Vectors |Ψi in HA ⊗ HB can be divided into two categories:

% separable if |Ψi = |ψi ⊗ |φi,


|Ψi
& nonseparable or entangled otherwise

Given a pure state |Ψi of the composite system A + B one can pass from it to
the reduced states ρA (ρB ) of the subsystem A (B) following a well defined
procedure called partial tracing : ρA = TrB [|ΨihΨ|] ( ρB = TrA [|ΨihΨ|]).
The surprising thing about entangled vectors |Ψi is that even though |Ψi
describes a pure state of the composite system, the corresponding reduced
density operators ρA and ρB do not describe pure states as was first noted
by Schrödinger. In other words, in sharp contrast to the situation in the
classical case, one finds that in quantum mechanics a complete knowledge
of the composite system does not necessarily imply complete knowledge of
its parts. This feature of quantum mechanics plays an important role in the
field of Quantum Information Theory.

4.2 Indistinguishability: Maxwell Boltzmann, Bose and


Fermi statistics
Consider a system consisting of N kinematically identical particles each of
which can exist in M energy states |ii, i = 1, · · · , M corresponding to energies
1 , 2 , · · · M . Let H denote the corresponding single particle Hamiltonian.
The Hilbert space HN describing this composite system then consists of an
N -fold tensor product of the single particle Hilbert space H:

HN = H ⊗ H ⊗ · · · ⊗ H

7
and is of dimension M N . The Hamiltonian for an assembly of N such non-
interacting particles would have the structure:
HN = H(1) + H(2) · · · + H(N )
where, H(1) denotes H ⊗ I ⊗ I · · · ⊗ I and so on.
Let us choose {|ii, i = 1, · · · M } as a basis for each H then the set of M N
states |i1 i⊗|i2 i⊗· · · |iN i, i1 , i2 , · · · , iN = 1, 2, · · · , M serve as a basis for HN .
We can decompose this set of M N states by grouping together states which
have the same number of 1’s, 2’s · · · etc. regardless of their location in the
product. Each such group is characterized by a composition of N i.e. by a set
of occupation numbers n ≡ (n1 , n2 , · · · , nM ), adding up-to N . Elementary
combinatorial considerations tell us that the number of states f (n1 , · · · , nM )
in each such group is given by f (n1 , · · · , nM ) = N !/n1 ! · · · nM !. As with any
sequence of functions labelled by a discrete index ( like Hermite polynomials
, Legendre polynomials etc.), it proves convenient to define the ‘generating
function’ of f (n1 , · · · , nM ) as follows

(I)
X N!
ZN (x) = xn1 1 xn2 2 · · · xnMM ,
ni n1 ! · · · nM !
Σni =N

(It is called a generating function on account of the fact that when expanded
in powers of x1 , · · · , xM the coefficient of xn1 1 · · · xnMM yields f (n1 , · · · , nM ).
We will see later with the identifications x ≡ x1 , · · · , xM ; xi ≡ exp(−i /kB T )
this function yields the canonical partition function of the uncorrected Maxwell-
Boltzmann statistics or infinite statistics.) The sum on the RHS in the equa-
tion above can be carried out using the multinomial theorem to o obtain
(I)
ZN (x) = (x1 + · · · + xM )N ,
(MB) (I)
For later use we also define ZN (x) obtained by dividing ZN (x) by N !:
1 (I)
(MB)
ZN Z (x)
(x) =
N! N
which, as the notation suggests, will be relevant in the context of Maxwell-
Boltzmann statistics.
We now come to the notion of indistinguishabilty (in the permutation
group sense).
The Hilbert space Hphy describing identical and indistinguishable parti-
cles is constructed out of HN by

8
• admitting only those operators on H which are permutation symmetric
i.e. those operators, like HN , which treat all the factors in the tensor
product democratically.

• identifying those states in H which have the same expectation values


for all permutation symmetric operators.
This brings the permutation group into play and entails decomposing HN
into subspaces each of a specific symmetry type and then carrying out the
identification of states which are indistinguishable in the sense above. ( For
more details, if interested, see S. Chaturvedi Phys. Rev E 54, 1378 (1996)).
If we further restrict ourselves only to subspaces of specific symmetry type
i.e. to states which are symmetric ( or antisymmetric ) under permutations
we are led to Bose (Fermi) statistics. Construction of a basis for this subspace
(B) (F )
HN (HN ) can be accomplished by symmetrizing (anti symmetrizing ) the
basis states |i1 i · · · ⊗ |iM i. For a given set of occupation numbers (n1 , ·, nM )
there will only be one state obtained by taking symmetric linear combination
of the basis states in that set and we can denote it by |n1 , · · · , nM iB . There
will be no restrictions on the occupation numbers ni . Again for a given set of
occupation numbers (n1 , · · · , nM ) there will only be one state antisymmetric
state, now denoted by |n1 , · · · , nM iF , constructed by taking suitable linear
combination of the basis states in that set, however this time there will be
restriction on ni that they can be either 0 or 1. These two choices lead to
Bose and Fermi statistics respectively. Thus

f (B) (n1 , · · · , nM ) = 1, n1 , +n2 + · · · + nM = N (Bose)


f (F) (n1 , · · · , nM ) = 1, n1 , +n2 + · · · + nM = N ; ni = 0, 1 (Fermi)

The number of basis states in the first case would be


 
M +N −1
 
N

which is obtained by finding the number of ways in which N balls can be put
into M boxes. The number of basis states in the second case turns out to be
 
M
 
N

9
This is obtained by finding the number of arrangements of N 1’s and M − N
0’s. Thus
   
M +N −1 M
(B) (F)
dimHN =   ; dimHN = 
N N

For the generating functions in the two cases we obtain


(B)
X
ZN (x) = xn1 1 xn2 2 · · · xnMM ,
ni
Σni =N

(F)
X
ZN (x) = xn1 1 xn2 2 · · · xnMM ,
ni
Σni =N ,ni =0,1

Finally, anticipating later use, for the quantities ZN (x) above ( indexed
by a discrete label N ) we may define a ‘grand’ generating function Z(X) as
follows: ∞
X
Z(X) = tN ZN (x); Xi = txi
N =0

to obtain in each case


∞  
(I)
X
N (I) 1
Z (X) = t ZN (x) = P ,
N =0
1− i Xi
X∞ M
Y
(MB) N (MB)
Z (X) = t ZN (x) = eXi ,
N =0 i=1
∞ M  
(B)
X
N (B)
Y 1
Z (X) = t ZN (x) = ,
N =0 i=1
1 − Xi
X∞ M
Y
(F) N (F)
Z (X) = t ZN (x) = (1 + Xi ) ,
N =0 i=1

4.3 Fock description


Fock description stipulates

• a vacuum or zero particle state |0 >

10
• states describing multiparticle states with n1 particles in state 1, n2
particles in state 2 etc are built by the action of monomials in operators
a†i , called the creation operators, involving n1 operators a†1 , n2 operators
a†2 etc on the state |0 >.
• the opertors ai called the annihilation operators satisfy ai |0 >= 0
• suitable commutation relations between the operators are assumed to
take into account the appropriate statistics and to ensure that the
action of ai (a†i ) on a state with ni particles in state i leads to state
with (ni − 1) ((ni + 1)) particles in the state i. This has the advantage
that the same framework allows to treat 1 particle, 2 particle, 3 particle
etc systems without much change.

Example M = 2 N = 2 bosonic system


Basis in S(H ⊗ H) Fock representation
1
|2, 0iB = |1i ⊗ |1i |2, 0iB = √ a†2
1 |0i
2
1
|1, 1iB = √ [|1i ⊗ |2i + |2i ⊗ |1i] |1, 1iB = a†1 a†2 |0 >
2
1
|0, 2iB = |2i ⊗ |2i |0, 2iB = √ a†2
2 |0i
2
[ai , aj ] = [a†i , a†j ] = 0, [ai , a†j ] = δij
Ni ≡ a†i ai ; [Ni , aj ] = −aj δij , [Ni , a†j ] = a†j δij

Example M = 2 N = 2 fermionic system


Basis in A(H ⊗ H) Fock representation
1
|1, 1iF = √ [|1i ⊗ |2i − |2i ⊗ |1i] |1, 1iF = a†1 a†2 |0 >
2
{ai , aj } = {a†i , a†j } = 0, {ai , a†j } = δij
Ni ≡ a†i ai ; [Ni , aj ] = −aj δij , [Ni , a†j ] = a†j δij
In general, for the Fock basis we use the following convention
1 1
|n1 , n2 , · · · , nM iB = √ ··· √ a†n †nM
1 · · · aM |0i
1

n1 ! nM !
†n1 †nM
|n1 , n2 , · · · , nM iF = a1 · · · aM |0i

11
and it can easily be verified that in the bosonic case

ai |n1 , n2 , · · · , nM iB = ni |n1 , · · · , ni − 1, · · · nM iB

a†i |n1 , n2 , · · · , nM iB = ni + 1 |n1 , · · · , ni + 1, · · · nM iB

and in the fermionic case


nk √
Pi−1
ai |n1 , n2 , · · · , nM iB = (−1) k ni |n1 , · · · , ni − 1, · · · nM iF
Pi−1 √
a†i |n1 , n2 , · · · , nM iB = (−1) k nk
ni + 1 |n1 , · · · , ni + 1, · · · nM iF

so that in both the case

Ni |n1 , n2 , · · · , nM iB or F = ni |n1 , n2 , · · · , nM iB or F

The full Fock space has the following structure

F B = F0B ⊕ F1B ⊕ · · · FNB ⊕ · · ·


F F = F0F ⊕ F1F ⊕ · · · FNF ⊕ · · ·

and
X
Basis for each FNB : {|n1 , n2 , · · · , nM iB ; ni = N }
i
X
Basis for each FNF : {|n1 , n2 , · · · , nM iF ; ni = N }
i

The advantages of working with Fock space is obvious : No extra work is


needed in going from N bosons or fermions to N + 1 bosons or fermions.
Further, the Hamiltonian for a system of non interacting Bosons/ Fermions
in this notation is always given by
X
H = Ni i , Ni = a†i ai
i

regardless of the number of particles one is dealing with. To put it differently,


the same Hamiltonian works for all N . For a fixed N all one needs to do is to
restrict oneself to the N particle subspace FN of the Fock space. Further, the
situations where N changes can be dealt with in the same framework owing
to the availability of the creation and annihilation operators. A cautionary

12
note : one comes across operators a, a† in the context of the quantum me-
chanics of systems with single cartesian degree of freedom. There they are
operators associated with the position and momentum of the particle. Here,
in contrast, they are purely formal operators introduced for bookkeeping pur-
poses for dealing with multiparticle bosonic or fermionic systems and are in
no way connected with the position and momentum of the particles. This is
highlighted by the fact that ai , a†i bear state labels i and make no reference
to any specific particle in the N particle system. Also note that having set
up this formalism with the state label i taking discrete values from 1 to M
finite we may permit i to take infinitely many values, and with some care, a
continuous infinity of values.
We have seen above how one can furnish a Fock description for the sym-
metric and antisymmetric subspaces of H⊗N by postulating suitable algebraic
relations between the creation and annihilation operators. One may wonder
if it is possible to provide a Fock description for H⊗N itself. This is indeed
the case and the corresponence is given below :

|i1 i ⊗ |i2 i · · · ⊗ |iN i → a†i1 · · · a†iN |0i

with
ai a†j = δij
Note the absence of any separate algebraic relations between the creation
and annihilation operator themselves. This ensures that the number of inde-
pendent states corresponding to a given set of occupation numbers n1 , · · · nM
is indeed given by
N!
n1 ! · · · nM !
and not 1 as in the case of symmetric and antisymmetric subspaces of H⊗N .
Further, unlike the bosonic or fermionic cases, where the number operators
Ni , obeying the commutation relations [Ni , aj ] = −aj δij ; [Ni , a†j ] = a†j δij ,
have a very simple expression in terms of creation and annihilation operators
viz. a†i ai , here this is not so and instead they are given by
X † † XX † † †
Ni = a†i ai + ak ai ai ak + ak al ai ai al ak + · · ·
k k l

as can easily be verified.

13
5 Review of Thermodynamics
The principal objective of thermodynamics is to describe the state of a macro-
scopic system in equilibrium, in terms of a few macroscopic variables. ( Intu-
itively by equilibrium we mean here that the system has no net temperature
or pressure gradients or particle fluxes)

Internal energy : U, Entropy : S, Pressure : P, Volume :


V, Temperature : T, Number of Particles : N, Chemical potential : µ

Three of these can be chosen independently and can be used to specify the
state of a thermodynamic system. The others are functions thereof. Certain
privilleged choices are dictated by the laws of thermodynamics which describe
the ways in which the thermodynamic state of can be changed from one to
the other. Thermodynamic state of a system can be changed from A to B
by exchange of ‘heat’ or by making the system do ‘work’ by bringing the
system into conact with suitable ‘reservoirs’. Notions of heat and work enter
into the picture only in the conext of processes which change the state of the
system and are not to be taken as attributes of the system itself. Thus one
can not speak of the heat or work contained in a thermodynamic system. In
equilibrium thermodynamics the processes that bring about changes in the
thermodynamic changes are taken to be quasi-static and reversible. The two
laws of thermodynamics for such processes are

dU = ąQ + ąW ; ąW = −P dV + µdN
ąQ = T dS

The quantity µ the way it appears here has the interpretation of the work
done on the system per particle in the process of injecting particles into the
system at a constant volume. Note that two kinds of symbols for infinitesi-
mals d and ą appear in these equations and the distinction between the two
is that while Z B
dU (1)
A
depends only on the end points
Z B Z B
ąQ or ąW
A A

14
depend not only on the end points A and B but also on the path that takes
A to B.
Combining the two laws together one obtains

dU = T dS − P dV + µdN

which tells us that


U = U (S, V, N )
and permits the following identifications:
     
∂U ∂U ∂U
T = ; P =− ; µ=
∂S V,N ∂V S,N ∂N S,V

which express T, P, µ in terms of S, V, N . We can solve U = U (S, V, N ) for


S in terms of U, V, N and obtain

S = S(U, V, N )

Further, from
1 P µ
dS = dU + dV − dN
T T T
one is lead to the following identifications
     
1 ∂S P ∂S µ ∂S
= ; = ; =−
T ∂U V,N T ∂V U,N T ∂N U,V

Rewriting dU = T dS − P dV + µdN variously as

d(U − T S) = −SdT − P dV + µdN


d(U + P V ) = T dS + V dP + µdN
d(U − T S + P V ) = −SdT + V dP + µdN
d(U − T S − µN ) = −SdT − P dV − N dµ

one obtains four more useful ‘thermodynamic potentials’ in addition to U (S, V, N )


and S(U, V, N )

A = A(T, V, N ); A = U − T S Helmholtz free energy


H = H(S, P, N ); H = U + P V Enthalpy
G = G(T, P, N ); G = U − T S + P V = A + P V Gibbs free energy
Φ = Φ(T, V, µ); Φ = U − T S − µN = A − µN Grand or Landau potential

15
Different thermodynamic potentials have their usefulness in different con-
texts. In any case the knowledge of one determines the others. The proce-
dure for passing from one to the other is similar to that in passing from the
Lagrangian to the Hamiltonian in classical mechanics. Thus, for instance,
given U as a function of S, V, N we construct the function A(T, V, N )

A(T, V, N ) = U (S, V, N ) − T S

by solving  
∂U
T =
∂S V,N

for S in terms of T, V, N and using it to eliminate S in U (S, V, N ) − T S in


favour of T, V, N . Conversely, given A as a function of T, V, N we construct
the function U (S, V, N )

U (S, V, N ) = A(T, V, N ) + ST

by solving  
∂A
S=
∂T V,N

for T in terms of S, V, N and using it to eliminate T in A(T, V, N ) + T S in


favour of S, V, N .

5.1 Scaling properties


Consider two systems in thermodynamic equilibrium, both identical in every
way, except that one is a factor λ larger than the other. We say that the
second system is scaled by a factor λ with respect to the first system. We ask:
How are the state variables of the two systems related? Or, in other words,
how do the state variables of a thermodynamic system scale under a scaling
of the system? Clearly, if the system is scaled by a factor λ, it must mean
that U, V , and N are all scaled by such a factor: U → λU, V → λV , and
N → λN . On the other hand, if the scaled system is otherwise identical to the
original system, it must be true that T stays the same: T → T . How about
S, P , and µ? The answer comes from the first law: dU = T dS − P dV + µdN .
If U scales but T does not, it must be true that S scales: S → λS. Similarly,
if both U and V scale, it must be true that P does not: P → P . And finally,
if both U and N scale, it must be true that µ does not: µ → µ. We can

16
therefore separate the thermodynamic variables into two groups. The first
group contains the variables that scale under a scaling of the system; such
variables are called extensive. The second group contains the variables that
do not scale under a scaling of the system; such variables are called intensive.
We have found that U, S, V , and N are extensive variables, and that T , P ,
and µ are intensive variables.
[Caution: The textbook reasoning given above is not quite correct. Given
that U, V, N are extensive, the laws of thermodynamics only require that dS,
change in S, is extensive. This does not imply that S itself is extensive.
While extensivity of S implies extensivity of dS the converse is not true.
Consider the function f (x) = xn + c, for which it is true that under x → λx
df (λx) = λn df (x) but this does imply that f (λx) = λn f (x). Homogeneity
of the differential of a function does not imply homogeneity of the function
itself unless we have other reasons to believe that the constant which spoils
the homogeneity may be set equal to zero. Nevertheless hereafter we will
assume the scaling properties as above and examine its consequences]
The scaling properties of the thermodynamic potentials :

U (λS, λV, λN ) = λU (S, V, N )


A(T, λV, λN ) = λA(T, V, N )
H(λS, P, λN ) = λH(S, P, N )
G(T, P, λN ) = λG(T, P, N )
Φ(T, λV, µ) = λΦ(T, V, µ)

have consequences :
• Homogeneity of U of degree 1 U (λS, λV, λN ) = λU (S, V, N ) upon dif-
ferentiating both sides w.r.t λ and setting λ = 1 yields
     
∂U ∂U ∂U
S +V +N = U (S, V, N )
∂S V,N ∂V S,N ∂N S,V

and hence
1 P µ
U = T S − P V + µN, or S = U+ V − N
T T T
• This in turn leads to

SdT − V dP + N dµ = 0, Gibbs − Duhem relation

17
• These considerations when applied to the Gibbs potential G yield
 
∂G
N = G i.e G = µ(T, P )N
∂N T,P

Thus to determine G we only need to determine the chemical potential


as a function of T, P ( or P = P (T, µ) or T = T (P, µ)).

5.2 Maxwell relations


These arise from the assumption of continuity of the partial derivatives of
the thermodynamic potenials. Thus applying this to U (S, V, N ) one has
  !   !
∂ ∂U ∂ ∂U
=
∂V ∂S V,N ∂S ∂V S,N
S,N V,N

yielding    
∂T ∂P
=−
∂V S,N ∂S V,N

5.3 Scope of thermodynamics


Much of the work in thermodynamics was devoted to deriving relations be-
tween experimentally measurable quantities such as
∂S

• Heat capacities : CX = T ∂T X,N
; X = V, P

• Compressibility : κX = − V1 ∂V

∂P X,N
; X = T, S (isothermal, isentropic)

• Coefficient of thermal expansion: αP = V1 ∂V



∂T P,N

∂T

• Joule Thomson coefficient : ∂P H

Some examples are


CP κT
• CV
= κS

T V α2p
• CP − CV = κT

T V α2P
• κT − κS = CP

18
∂T
= (T αP − 1) CVP

• ∂P H

To be able to compute these quantities for a specific system at hand one


needs to pass to Statistical Mechanics.

6 Classical Statistical Mechanics


6.1 Ergodic Hypothesis
Consider a system consisting of N particles occupying a volume V . Let Γ
denote the associated phase space. For any smooth function f (X) of the 6N
phase space variables collectively denoted by X there is a ρ(X) characterizing
the equilbrium state such that

1 T
Z lim
Z
6N
hf i ≡ d ρ(X)f (X) =T → ∞ dtf (X)
Γ T 0
for almost all starting points X(0). Any probability distribution has an
underlying ensemble of identical copies the system. The ensemble in the
present context is referred to as a Gibbs ensemble.
What are the possible candidates for the ρ(X)? Evidently once the system
is in ρ(X), it should stay in that state consistent with our intuitive picture of
equilibrium. It must therefore be a function of H(X) alone ( assuming that
there are no other constants of motion). The question now is what function
of H(X)?

6.2 Microcanonical Ensemble


Here we consider a system which is completely isolated from its surroundings–
it is closed and isoenergetic. A possible candidate for ρ(X) is ρmicro (X) where
Z 6N
δ[HN (X) − E] d X
ρmicro (X) = ; Σ(E) = 3N
δ[HN (X) − E]
Σ(E) Γ h

(Here we use the symbol E to denote the energy of the system instead of U
to allow for the interpretation of U as the average of E over ρ. Of course in
the present case, owing to the assumed form of ρ, E = U . Further, isolation
of the system assumes that V, N are held fixed.). The quantity h here is
introduced for dimensional reasons.

19
Often, in order to ensure the smoothness of the thermodynamic quantities
computed from this ensemble it becomes necessary to smoothen the infinite
spike in the delta function by a rectangle with a finite width ∆. Mathemat-
ically this amounts to replacing δ(H(X) − E) by Θ(E + ∆ − H)Θ(H − E)
where Θ(x) denotes the Heaviside step function.

6.3 Canonical ensemble


We next consider the system in contact with a heat bath at a temperature
T with which the system can exchange heat. The form of ρ(X) appropriate
to this situation is :
e−βHN (X)
ρcan (X, β) = R d6N X −βH(X)
Γ h3N
e
( This form arises from the requirement that PA+B (EA +EB ) = PA (EA )PB (EB )).
The constant β which at this stage can only be a characteristic of the heat
bath would be identified later with 1/kB T . Note that
Z
−βHN (X)
e = dEP (E)δ[HN (X) − E]; P (E) = e−βE

so that ρcan can be seen as arising from ρmicro by weighting the latter by the
probability that the system has energy E

6.4 Grand canonical ensemble


Next we consider an open isothermal–a system which can exchange both
heat and particles with its surroundings. In the spirit of the passage from
the microcanonical ensemble to the canonical ensemble a possible candidate
for ρ in this case is
eβµN −βHN (X)
ρgr (X, β, µ) = P∞ R 6N
N =0 eβµN Γ dh3NX e−βH(X)
We will see a little later that the form proposed here is consistent with the
expectation that the probability PN that there are N particles in the system
has the form of a Poisson distribution :
N
N −N
PN = e
N!

20
6.5 Partition functions
Each of the three ensembles discussed above give rise to respective ‘partition
functions’:
Z 6N
d X
Micro canonical partition function : Σ(E, V, N ) = 3N
δ[HN (X) − E]
Γ h
Z 6N
d X −βH(X)
Canonical partition function : ZN (T, V, N ) = 3N
e
Γ h
∞ Z 6N
X
βµN d X −βH(X)
Grand Canonical partition function : Z(T, V, µ) = e 3N
e
N =0 Γ h

The three partition functions are related to each other as follows:



X
Z(T, V, µ) = eβµN ZN (T, V, N )
N =0
Z
Z(T, V, N ) = dE e−βE Σ(E, V, N )

permitting one to go from one to the other.


Further, from the arguments of the three partition functions the following
associations with the thermodynamic potentials are evident:

{Σ(E, V, N ), Entropy}
{ZN (T, V, N ), Helmholtz free energy }
{Z(T, V, µ), Grand or Landau potential}

Exact relationships will be given later.

6.6 Indistinguishability in the classical context


This entails making the following replacement
Z 6N Z 6N
d X 1 d X
3N

Γ h N ! Γ h3N
in the formulae above. When this is done and one says that one is dealing
with Maxwell-Boltzmann statistics. The idea behind this notion of indistin-
guishability is that in a gas consisting of N particles the configuration in

21
which particle 1 has position and momentum (q1 , p1 ) particle 2 has position
and momentum (q2 , p2 ) · · · is indistinguishable from the one in which which
particle 1 has position and momentum (q2 , p2 ) particle 1 has position and
momentum (q1 , p1 ) · · · and should therefore be identified with the former.
This accounts for the factor (1/N !) above.
For a classical gas consisting of identical and indistinguishable non-interacting
particles, making this replacement in the expression for ρ for the grand canon-
ical ensemble one finds that the expression for the probability that the system
has N particles:
Z 6N
1 d X
PN = ρgr (X, β, µ)
N ! Γ h3N
after using
N
d6N X −βHN (X) d6 X −βH1 (X)
Z Z
e = e
Γ h3N h3
and identifying N appropriately indeed has the form of a Poisson distribution.
[ Why Poisson? Regard the exchange of particles between the system and
the bath as a chemical reaction

SB

in which particles of type S belonging to the system get converted to those of


type B belonging to the bath at a rate k1 and those of type B get converted
to S at a rate k2 Assuming that the number of particles of type B, also
denoted by B, does not change ( consistent with the model for a particle
reservoir) one can write a simple stochastic model for the probability P (N, t)
that the system has N at time t:

P (N, t + ∆t) = k1 ∆t(N + 1)P (N + 1, t) + k2 B∆tP (N − 1, t)


+ (1 − k1 N ∆t − k2 B∆t)P (N, t)
dP (n, t)
⇒ = k1 [(N + 1)P (N + 1, t) − N P (N, t)]
dt
+ k2 B[P (N − 1, t) − P (N, t)]

In the steady state, one can easily check that P (N, ∞) is a Poisson distribu-
tion with N = k2 B/k1 .]

22
6.7 Relation to thermodynamic potentials
• Microcanonical Ensemble

S = kB ln Σ

• Canonical Ensemble
∂ lnZN
U =−
∂β
S = kB (ln ZN + βU )
 
∂ lnZN
P = kB T
∂V
A = −kB T ln ZN

For a non interacting monoatonic gas of N identical particles


 3N/2
1 2πm
ZN =
N ! βh2

comparison of U = − ∂ lnZN
∂β
and U = (3/2)kB T leads to the identifica-
tion of β with 1/kB T

• Grand canonical ensemble



U − µN = − (lnZ)
∂β

βN = (lnZ)
∂µ
S = kB [lnZ + βU − βµN ]

P = kB T (lnZ)
∂V
Φ = −kB T lnZ

We now give the derivations of these results: In the microcanonical case the
result given above follows from maximization of entropy
Σ
X
S = −kB pi lnpi
i

23
Clearly it S attains its maximum value when pi = 1/Σ and its value then is
given by :
S = kB ln Σ
In the canonical case the first follows from the definition of ZN . To derive
the second and third we proceed as follows. Since ZN is a function of β and
V
∂lnZN ∂lnZN
d(lnZN ) = dβ + dV
∂β ∂V
∂lnZN
= −U dβ + dV ⇒
∂V
∂lnZN
d(lnZN + βU ) = βdU + dV
∂V
∂lnZN
kB T d(lnZN + βU ) = dU + dV
∂V
which on comparison with T dS = dU + P dV gives the second and the third
relations. The fourth follows from substituting for S in A = U − T S
In the grand canonical case the first two follow from the definition of Z.
Here since Z is a function of β, V, µ we have
∂lnZ ∂lnZ ∂lnZ
d(lnZ) = dβ + dV + dµ
∂β ∂V ∂µ
∂lnZ
= (−U + µN )dβ + dV + βN dµ ⇒
∂V
∂lnZ
d(lnZ + βU − µN ) = βdU + dV − µβdN
∂V
∂lnZ
kB T d(lnZN + βU − µN ) = dU + dV − µdN
∂V
which on comparison with T dS = dU + P dV − µdN yield the third and the
fourth relations. The fifth follow from substituting for S in Φ = U −T S −µN .
We are thus led to the anticipated relations between the three partition
functions and the three thermodynamical potentials alluded to earlier :

S = kB lnΣ
A = −kB T lnZN
Φ = −kB T lnZ

24
6.8 Transition to Quantum Statistics

ρcan →ρ̂can = ZN−1 e−β Ĥ


ZN → ZN = Tr(e−β Ĥ )
ρgr →ρ̂gr = Z −1 e−β(Ĥ−µN̂ )

X
Z→Z= eβµN ZN
N =0

= Tr(e−β(Ĥ−µN̂ )

where Ĥ and N̂ respectively denote the Hamiltonian operator for the system
and the number operator.
With this one has now a complete framework for computing the ther-
modynamics of a specific classical or quantum system. It simply entails
computing the appropriate partition function. Note that the space on which
the operators act and correspondingly the meaning of trace operation must
be interpreted in the appropriate context. For instance, when one writes ρ̂gr
as Z −1 e−β(Ĥ−µN̂ ) with Z = Tr(e−β(Ĥ−µN̂ ) it is assumed that one is working
in the Fock description and the trace operation is to be carried out in that
context as would be clear from the discussion below:
• N identical non interacting particles : First quantized description
Here

Hilbert space HN : H⊗N


Hamiltonian HN : H ⊗ I · · · ⊗ I + · · · + I ⊗ I · · · ⊗ H
Basis {|i1 i ⊗ · · · ⊗ |iN i; i1 , i2 · · · iN = 1, · · · M }; H|ii = i |ii

Given this
M
X M
X
−βHN
ZNI = Tre = ··· , hi1 | ⊗ · · · ⊗ hiN |e−βHN |i1 i ⊗ · · · ⊗ |iN i
i1 =1 iN =1

Using

HN |i1 i ⊗ · · · ⊗ iN i = (n1 1 + n2 2 + · · · + nM M )|i1 i ⊗ · · · ⊗ iN i

25
where ni is the number of times i occurs in |i1 i ⊗ · · · ⊗ iN i and the fact
that for a given set n1 , · · · , nM there are N !/(n1 !n2 ! · · · nM !) states of type
|i1 i ⊗ · · · ⊗ |iN i we obtain
ZN = Tre−βHN
X N!
= e−β(n1 1 +···+nM M )
ni
n 1 ! · · · nM !
P
i ni =N

We have already encountered this sum and with the identification xi = e−βi
we get
ZNI = (e−β1 + · · · + e−βM )N
This may be written as
M
X
ZNI =z ; N
z= e−βi = Tre−βH
i=1

where z is referred to in the literature as the ‘single particle’ partition func-


tion. Note that this is similar in structure to the classical case where the
classical canonical partition function for a non interacting system has the
form Z 6
I N d X −βH(X)
ZN = z ; z = e
h3
Dividing the two partition functions by N ! leads respectively to the canonical
partition functions appropriate to Maxwell-Boltzmann statistics.
M
zN X
ZNM B = ; z= e−βi = Tre−βH
N! i=1
N
d6 X −βH(X)
Z
z
ZNM B = ; z= e
N! h3

• N non interacting particles obeying Bose statistics : First quantized de-


scription
Here
Hilbert space HNB
: S(H⊗N )
Hamiltonian HN : H ⊗ I · · · ⊗ I + · · · + I ⊗ I · · · ⊗ H
Basis {S(|i1 i ⊗ · · · ⊗ |iN i); i1 , i2 · · · iN = 1, · · · M }; H|ii = i |ii

26
Here recognising the fact there is only one state corresponding to a given set
of occupation numbers one obtains

ZNB = Tre−βHN
X
= e−β(n1 1 +···+nM M )
P ni
i ni =N

• N non interacting particles obeying Fermi statistics : First quantized


description Here

Hilbert space HNF


: A(H⊗N )
Hamiltonian HN : H ⊗ I · · · ⊗ I + · · · + I ⊗ I · · · ⊗ H
Basis {A(|i1 i ⊗ · · · ⊗ |iN i); i1 , i2 · · · iN = 1, · · · M }; H|ii = i |ii

Again, as in the Bose case recognising the fact there is only one state corre-
sponding to a given set of occupation numbers one obtains

ZNF = Tre−βHN
X
= e−β(n1 1 +···+nM M )
ni ,ni =0,1
P
i ni =N

• N identical non interacting particles : Fock description Here

Hilbert space : FNI


X
Hamiltonian H : i Ni ; (2)
i

Basis {a†i1 a†i2 †


· · · aiN |0i; i1 , i2
· · · iN = 1, · · · M }
X X X † † †
Ni = a†i ai + a†k a†i ai ak + ak al ai ai al ak + · · · ; ai a†j = δij
k k l

Of course, by construction, computing the trace in ZNI = Tre−βH we get the


same expression as above.
•N particles obeying Bose statistics : Fock description

27
Here

Hilbert space : FNB


X
Hamiltonian H : i Ni ; (3)
i
1 X
Basis { √ √ a†n
1
1
· · · · · · a †nM
M |0i; ni = N }
n1 ! · · · nM ! i

Ni = a†i ai ; [ai , aj ] = [a†i , a†j ] = 0; [ai , a†j ] = δij

• N particles obeying Fermi statistics : Fock description


Here

Hilbert space : FNF


X
Hamiltonian H : i Ni ; (4)
i
1 X
Basis { √ √ a†n
1
1
· · · · · · a †nM
M |0i; ni = N, ni = 0, 1}
n1 ! · · · nM ! i

Ni = a†i ai ; {ai , aj } = {a†i , a†j } = 0; {ai , a†j } = δij

Of course, by construction, computing the trace in ZN = Tre−βH in both


the cases we get the same expressions as above for ZNB , ZNF .

6.9 Grand or Landau potential for M-B, Bose and Fermi


statistics
The grand canonical partition function for non interacting systems obeying
the three statistics are respectively given by

X M
Y
(MB)
Z (MB) (X) = tN ZN (x) = eXi ,
N =0 i=1
∞ M  
(B)
X
N (B)
Y 1
Z (X) = t ZN (x) = ,
N =0 i=1
1 − Xi
X∞ M
Y
(F) N (F)
Z (X) = t ZN (x) = (1 + Xi ) ,
N =0 i=1

28
where Xi = e−β(i −µ) .
The grand potential Φ in each case is given by Φ = −kB T lnZ which in
turn permits us to compute all the thermodynamic properties of the system
at hand.
From the definition of the of Φ it follows that the mean number ni of
particles in the single particle state i is given by
∂ ∂lnZ
ni = Φ = −kB T
∂i ∂i
For the three statistics this gives
Maxwell-Boltzmann : ni = e−β(i −µ)
1
Bose : ni = β(i −µ)
e −1
1
Fermi : ni = β(i −µ)
e +1
We may rewrite the grand potentials in the three cases as
X
Maxwell-Boltzmann : ΦM B = −kB T ni
i
X
Bose : ΦB = −kB T ln (1 + ni )
i
X
Fermi : ΦF = kB T ln(1 − ni )
i

For the mean number of particles N we have


X
N= ni
i

It proves convenient to rewrite the relations above as


Z
MB
Φ = −kB T dD()n()
Z
B
Φ = −kB T dD()ln(1 + n())
Z
F
Φ = kB T dD() log(1 − n())
Z
N = dD()n()

29
P
with D() = δ( − i ) and

n() = e−β(−µ) Maxwell-Boltzmann


1
n() = β(−µ) Bose
e −1
1
n() = β(−µ) Fermi
e +1
With D() interpreted as density of states, these formulae provide smooth
passage to situations where the single particle energy take continuous values.
For free particles with energy-momentum relation  = p2 /2m the expres-
sion for D() is given by
3/2


2m
D() = 2πV 
h2

Note that in the Fermi case n() as a function of , has the property that
as T → 0 it equals 1 for  < µ and equals 0 for  > µ. Fermi energy is defined
as the value of  above which n() is 0.

30
7 Ideal Maxwell-Boltzmann gas : Classical
7.1 Microcanonical partition function and Entropy

N
d6N X X p2
Z
i
Σ= ; H(X) =
U ≤H(X)≤U +∆ h3N N ! i=1
2m
VN
Z
= 3N p2
dp3N ;
h N! U ≤
PN i
i=1 2m ≤ U +∆

V N √ Z
= ( 2m)3N dp3N ;
h3N N ! U ≤
PN
i=1 p2i ≤ U +∆

V √ N
= 3N ( 2m)3N × Volume of the shell between 3N - dimensional spheres of radii
√h N ! √
U and U + ∆
VN √ √
= 3N ( 2m)3N × surface area of a 3N dimensional sphere of radius U ×
h N! √ √ √
width of the shell = U + ∆ − U ≈ ∆/(2 U )

VN √ 3N 3N π 3N/2 ( U )3N −1 ∆
= 3N ( 2m) × 3N
 × √
h N! Γ 2 +1 2 U
3N/2
2mπU V 2/3
 
3N ∆ 1
=
2U N !Γ 3N

2
+1 h2
" 3/2 #N
3N ∆ 1 2mπU
= V
2U N !Γ 3N
 2
2
+ 1 h

Hence, on using Stirling approximation and retaining terms of order N one


obtains the following expression for the entropy in terms of its proper vari-
ables U, V, N :

S = kB ln Σ
" 3/2 #
2mπU 3 3N 3
= N kB {ln 2
V − ln + − lnN + 1}
h 2 2 2
" 3/2 #
4mπ V U 3/2 5
= N kB {ln 2 5/2
V + } (Sackur-Tetrode relation)
3h N 2

31
Using  
1 ∂S
=
T ∂U V,N

one obtains
3
U = N kB T.
2
Using this in A = U − T S to express the RHS in terms of its proper variables
T, V, N for the Helmholtz free energy A one obtains
"  3/2 #
2πmkB T V
A = −N kB T ln 2
+1
h N

7.2 Canonical partition function and the Helmholtz


free energy

N
d6N X −βH(X) X p2
Z
i
ZN = 3N
e H(X) =
h N! i=1
2m
N
VN
 Z 
3 −βp2 /2m
= d pe
h3N N !
VN
= 3N
(2πmkB T )3N/2
h N!
" 3/2 #N
1 2πmkB T
= V
N! h2

and hence using Stirling approximation


"  3/2 #
2πmkB T V
A = −kB T ln ZN = −N kB T ln +1
h2 N

as before. Using  
∂A
µ=
∂N T,V

one obtains " 3/2 #


2πmkB T
N = eβµ V
h2

32
Using this in Φ = A−µN to express the RHS in terms of the proper variables
T, V, µ for the grand potential Φ one obtains
" 3/2 #
2πmkB T
Φ = −kB T eβµ V
h2

7.3 Grand canonical partition function and the grand


potential

X
Z= eβµN ZN
N =0
" 3/2 #!
2πmkB T
= exp eβµ V
h2

and hence

Φ = −kB T ln Z
" 3/2 #
2πmkB T
= −kB T eβµ V
h2

as before.

8 Assembly of one dimensional oscillators :


Classical
Here we would treat the particles as distinguishable by virtue of the fact that
each oscillator has a distinct centre of oscillation.

33
8.1 Microcanonical partition function and Entropy

N
d2N X p2i
Z  
X 1
Σ= ; H(X) = + mω 2 qi2
U ≤H(X)≤U +∆ hN i=1
2m 2
Z
1
= N p2
(dqdp)N ;
h U ≤
PN i 1 2 2
i=1 2m + 2 mω qi ≤ U +∆
!N Z
1 √
r
2
= ( 2m)N (dqdp)N ;
hN mω 2 U ≤
PN 2 2
i=1 pi +qi ≤ U +∆
2
= ( )N × Volume of the shell between 2N -dimensional spheres of radii
√ hω √
U and U + ∆
 N
2 √
= × surface area of a 2N dimensional sphere of radius U ×

√ √ √
width of the shell = U + ∆ − U ≈ ∆/(2 U )
 N √
2 2N π N ( U )2N −1 ∆
= × × √
hω Γ (N + 1) 2 U
 N
N ∆ 1 2πU
=
U N ! hω

Hence, on using Stirling approximation and retaining terms of order N one


obtains the following expression for the entropy in terms of its ptoper vari-
ables U, V, N :

S = kB ln Σ
 
U
= N kB {ln + 1}

Using  
1 ∂S
=
T ∂U V,N

one obtains
U = N kB T.

34
Using this in A = U −T S to express the RHS in terms of the proper variables
T, V, N appropriate to the Helmholtz free energy A, one obtains
 

A = N kB T ln
kB T

8.2 Canonical partition function and the Helmholtz


free energy

N  2
d2N X −βH(X)
Z 
X pi 1 2
ZN = e H(X) = + mωi
hN i=1
2m 2
Z N
1 p2
−β( 2m + 12 mω 2 )
= N dqdpe
h
r r N
1 π2m 2π
= N
h β βmω 2
 N
kB T
=

and hence

A = −kB T ln ZN
 

= N kB T ln
kB T

as before.

9 Assembly of one dimensional oscillators :


Quantum
9.1 Microcanonical partition function and Entropy
In this case
N
U− ~ω = M ~ω, M = integer
2

35
and computing Σ reduces to finding the number of ways in which M quanta
(or balls) can be distributed among N oscillators (boxes). We have already
solved this combinatorial problem and the result is
 
M +N −1
Σ=
M

Hence using Stirling approximation


       
U 1 U 1 U 1 U 1
S = N kB + ln + − − ln −
N ~ω 2 N ~ω 2 N ~ω 2 N ~ω 2

For T using T1 = ∂U∂S



V,N
one obtains
 
1 1
U = N ~ω + ~ω/k T
2 (e B − 1)

and hence  
β~ω −β~ω
S = N kB − ln (1 − e )
(eβ~ω − 1)
As before one can compute A from S using A = U − T S and the result is
 
1 1 −β~ω
A = N ~ω + ln(1 − e )
2 β

9.2 Canonical partition function and the Helmholtz


free energy


!N
X
ZN = e−β~ω(i+1/2)
i=0
N
e−β~ω/2

=
1 − e−β~ω

from which A can directly be computed through A = −kB T ln ZN to obtain


the same expression as above.

36
10 Assembly of two level atoms : Quantum
10.1 Microcanonical partition function and Entropy
Consider a system of N two atoms each capable of having the values ± for
its energy. The atoms are imagined to be located at distinct lattice sites. Let
N± denote the number of atoms in the energy state ±. Then clearly

N+ + N− = N ; U = (N+ − N−

. The microcanonical partition function and the entropy for this system is
given by
N!
Σ= , S = kB ln Σ
N+ !N− !
. For large N , using Stirling approximation one obtains
        
1 U U 1 U U
S = N kB ln2 − 1+ ln 1 + − 1− ln 1 −
2 N N 2 N N

from which one can deduce that

U = −N  tanh(β)

10.2 Canonical partition function


In this case
ZN = z N ; z = (e−β + eβ ) = (2coshβ)

Using U = − ∂β lnZN one obtains the same expression for U as above.

37
11 Ideal Bose gas
• Grand potential:

Φ = −kB T ln Z
Z ∞
= kB T dD()ln(1 − e−β(−µ) )
0
3/2 Z ∞


2m
= kB T 2πgV 2
d  ln(1 − e−β(−µ) )
h 0
 3/2 Z ∞ 
2m
5/2
= (kB T ) 2πgV dyy ln(1 − Z e ) ; Z = eβµ
1/2 −y
h2 0
3/2 
2 ∞ y 3/2
 Z 
2m 2  3/2 −y ∞
5/2
y ln(1 − Z e ) 0 −

= (kB T ) 2πgV dy 1 y
h2 3 3 0 ( Z e − 1)
 3/2
2 2m
= − (kB T )5/2 2πgV Γ(5/2)B5/2 (Z )
3 h2
 3/2
5/2 2mπ
= −gV (kB T ) B5/2 (Z )
h2

where ∞
y r−1
Z
1
Br (Z ) ≡ dy
Γ(r) 0 ( Z1 ey − 1)

• Pressure P :
∂Φ
P =−
∂V
 3/2
5/2 2mπ
= g(kB T ) B5/2 (Z )
h2

38
• Internal energy U :
Z ∞
1
U= dD()
0 eβ(−µ) −1
3/2 Z ∞
3/2

2m
= 2πgV d
h2 eβ(−µ) − 1
0
3/2

2m
= (kB T )5/2 2πgV Γ(5/2)B5/2 (Z )
h2
 3/2
3 5/2 2mπ
= V g(kB T ) B5/2 (Z )
2 h2

• Mean Number N :
Z ∞
1
N= dD()
0 eβ(−µ) −1
3/2 Z ∞
1/2

2m
= 2πgV d
h2 eβ(−µ) − 1
0
3/2

2m
= (kB T )3/2 2πgV Γ(3/2)B3/2 (Z )
h2
 3/2
3/2 2mπ
= V g(kB T ) B3/2 (Z )
h2

From the expressions above one finds that

2 PV B5/2 (Z )
P V = U, =
3 N kB T B3/2 (Z )

12 Photons
4πgV 2
For Photons µ = 0 and D() =  so that
(hc)3

39
• Grand potential:

Φ = −kB T ln Z
Z ∞
= kB T dD()ln(1 − e−β )
0
4πgV ∞
Z
= kB T 3
d2 ln(1 − e−β( )
(hc) 0
Z ∞ 
4 4πgV 2 −y
= (kB T ) dyy ln(1 − e ) ;
(hc)3 0
1 ∞ y3
 Z 
4 4πgV 1 3 −y ∞

= (kB T ) y ln(1 − e ) 0 − dy y
(hc)3 3 3 0 (e − 1)
1 4πgV
= − (kB T )4 Γ(4)B4 (1)
3 (hc)3
8πg
= −V (kB T )4 B4 (1)
(hc)3

• Pressure:
∂Φ
P =−
∂V
8πg
= (kB T )4 B4 (1)
(hc)3

• Internal energy:
Z ∞

U= D() β
0 e −1
Z ∞
4πgV 3
= d
(hc)3 0 eβ − 1
4πgV
= (kB T )4 Γ(4)B4 (1)
(hc)3
24πgV
= (kB T )4 B4 (1)
(hc)3

40
• Number density :
Z ∞
N 1
= dD() β
V 0 e −1
Z ∞
4πg 2
= d
(hc)3 0 eβ − 1
4πg
= (kB T )3 Γ(3)B3 (1)
(hc)3
8πg
= (kB T )3 B3 (1)
(hc)3
From the expressions above one finds that
1 PV B4 (1)
P V = U, =
3 N kB T B3 (1)
Note that

X Zs
Br (Z ) =
s=1
sr

X 1
Br (1) = = ζ(r)
s=1
sr

• Planck Distribution:
Z ∞
U 4πg 3
Energy density :u = = d
V (hc)3 0 eβ − 1
du 4πg 3
=
d (hc)3 eβ − 1
du g h ω3
=
dω 4 (πc)3 eβ~ω − 1

13 Ideal Fermi gas


All the formulae derived above for the Bose case hold with the replacement
Br (Z ) → Fr (Z ) where
Z ∞
1 y r−1
Fr (Z ) ≡ dy 1 y
Γ(r) 0 ( Z e + 1)
Thus

41
• Grand potential:

Φ = −kB T ln Z
 3/2
5/2 2mπ
= −gV (kB T ) F5/2 (Z )
h2

• Pressure P :
∂Φ
P =−
∂V
 3/2
5/2 2mπ
= g(kB T ) F5/2 (Z )
h2

• Internal energy U :
Z ∞
1
U= dD()
0 eβ(+µ) + 1
 3/2
3 5/2 2mπ
= V g(kB T ) F5/2 (Z )
2 h2

• Mean Number N :
Z ∞
1
N= dD()
0 eβ(−µ) +1
 3/2
2mπ
= V g(kB T )3/2 F3/2 (Z )
h2

From the expressions above one finds that


2 PV F5/2 (Z )
P V = U, =
3 N kB T F3/2 (Z )

Note that

X Zs
Fr (Z ) = (−1)s+1 r
s=1
s
 
1
Fr (1) = 1 − r−1 ζ(r)
2

42
13.1 Behaviour of a Fermi gas at low temperatures
Using the Sommerfeld expansion valid near T << TF
Z ∞
f ()
I[f ] ≡ d β(−µ)
e +1
Z0 µ
π2
= df () + (kB T )2 f 0 (µ) + · · ·
0 6
one finds that
" 2 #
π 2 kB T

F = µ 1 + + ···
12 µ
" 2 #
π 2 kB T

µ = F 1 − + ···
12 F
" 2 #
5π 2 kB T

3
U = N F 1 + + ···
5 12 F
" 2 #
5π 2 kB T

2
P V = N F 1 + + ···
5 12 F
π2
 
T
S = N kB + ···
2 TF

14 Heteronuclear diatomic molecules : Rota-


tional partition function
Here we consider a gas of heteronuclear diatomic molecules and compute
the contribution to the specific heat arising from the rotational degrees of
freedom. Each molecule is modelled after a rigid rotator– the single particle
Hamiltonian is taken to be
L2
H= : L : orbital angular momentum operator, I moment of inertia
2I
We know from quantum mechanics that the eigenvalues of the operator L2
are ~2 `(` + 1) and further that each eigenvalue is (2` + 1) degenerate. Hence

43
for N such molecules we have
"∞ #
N
X
−ell(`+1)(Θr /T ) ~2
ZN = z = (2` + 1)e ; Θr =
`=0
2IkB

Defining

X
fr (x) = (2` + 1)e−ell(`+1)x
`=0
one has
f1 (Θr /T )
U = N kB Θr
f0 (Θr /T )
 2
Θr f0 (Θr /T )f2 (Θr /T ) − f12 (Θr /T )
CV = N k B
T f02 (Θr /T )
At low temperatures (T << Θr ), approximating z by the first two terms :
z = 1 + 3−2Θr /T one obtains
 2
−2Θr /T Θr
U = 6N kB Θr /T e , ⇒ Cv == 12N kB e−2Θr /T
T
In the limit of high temperatures (T >> Θr ) defining x(l) = −`(` + 1)Θr /T
rewriting z as

T X
z= ∆x(`)e−x(`) ; ∆x(`) = x(` + 1) − x(`)
Θr `=0

and noticing that in the limiting case considered x(`) becomes a continuous
variable, one obtains the result :
T
z=
Θr
and hence
U = N K B T ; CV = N k B
as expected classically.
Corrections to the classical results may be derived using the Euler-Mclaurin
summation formula
n−1 Z n
X 1 1 1 h 000 000
i
f` = d`f (`)+ [f (0) − f (n)]− [f 0 (0) − f 0 (n)]+ f (0) − f (n) +· · ·
`=0 0 2 12 720

44
Use of this formula yields
T 1 1 Θr
z= + + + ···
Θr 3 15 T
and hence "  2 #
1 Θr
CV = N k B 1+ + ···
45 T

15 Specific heat of solids


15.1 Einstein Model
Einstein modelled a solid consisting of N atoms as N independent three
dimensional isotropic quantum harmonic oscillators, all oscillating with a
common frequency ω. We have already learnt that for a one dimensional
quantum harmonic oscillator

(β~ω)2 eβ~ω
 
1 1
U = Nω + β~ω ; CV = N kB β~ω
2 e −1 (e − 1)2

Hence, for N independent three dimensional isotropic oscillators


2
eΘE /T

ΘE ~ω
CV = 3N kB ; θE ≡ (Einstein Temperature)
T (eΘE /T − 1)2 kB

In the limit of high temperatures (θE << T )

CV → 3N kB

as expected. However, in the low temperature limit T << ΘE ,


 2
ΘE
CV ' 3N kB e−θE /T
T

which disagrees with the experimental results suggesting that CV goes to


zero as a power of T rarther than exponentially as suggested by Einstein’s
model.

45
15.2 Debye Model
Debye model treats the solid as a continuum and focusses on the possible
standing wave patterns to describe the collective behaviour. In a solid in the
shape of a cube of side length L, standing waves have the structure

A sin(kx x) sin(y y) sin(kz z) sin(ωt)

where ω = vk and to ensure that the amplitude goes to zero at the boundary
surfaces, k = (kx , ky , kz ) takes on discrete values

(nx π/L, ny π/L, nz π/L), nx , ny , nz = positive integers.

(Restriction to positive integers arises because of the fact that ±kx , ±ky , ±kz
all describe the same standing wave.) In a solid, two types of waves are
possible, longitudinal (or sound waves) with velocity vl and two transverse
waves (shear waves) with velocity vt . Now

Number of standing waves with the wave vector lying between k and k + dk
1 V
= 4πk 2 dk/(π/L)3 = 3 4πk 2 dk
8 8π
and hence

Number of standing waves with angular frequency the lying between ω and ω + dω
V
= 3 4πω 2 dω
8v
Including the contributions from the three wave types, one has for, D(ω),
the density of states
12πV 2 3 1 2
D(ω) = ω ; ≡ 3+ 3
8π 3 v 3 v 3 vl vt
The discreteness of the solid is brought into play by requiring that
Z ωD dω
12πV ω 3
3N = D(ω) = 3 3 D
0 8π v 3
Using this we may rewrite D(ω) as
9N 2
D(ω) = 3
ω
ωD

46
Hence
ωD
(β~ω)2 eβ~ω
Z
CV = kB dωD(ω) β~ω
0 (e − 1)2
Z β~ωD
9N x4 ex
= kB dx ;
(β~ωD )3 0 (ex − 1)2
3 x u4 eu
Z
= 3N kB D(ΘD /T ); D(x) = 3 du u Debye function
x 0 (e − 1)2
where ΘD = ~ωD /kB is called the Debye temperature. From this expression
one obtains
CV → 3N kB
for θD << T as expected and
2
12π 4

T
CV = N kB
5 θD
for ΘD >> T in agreement with experimental results.

16 Paramagnetism
The single particle Hamiltonian for an electron in an atom subject to an
external magnetic field B is given by
H = −µ · B
where the dipole operator µ is given by
J e~ 3 s(s + 1) − l(l + 1)
µ = µB g ; µB = , (Bohr Magneton), g = + (Landé factor
~ 2me 2 2j(j + 1)
Without loss of generatity taking the magnetic field to be in the z-direction,
one obtains for the single particle partition function
z = Tr[e−βµB gBJz /~ ]
j
X
= e−βµB gm
m=−j
sinh[(2j + 1)η] gµB B
= , η=
sinh η 2kB T

47
and hence  N
sinh[(2j + 1)η]
ZN =
sinh η
which in turn gives for the magnetization M
1 ∂lnZN
M=
β ∂B
= jgN µB Bj (η);
1
Bj (η) = [(2j + 1) coth[(2j + 1)η] − coth[η]] (Brillouin function)
2j
For η >> 1
Bj (η) = 1
and for η << 1
2
Bj (η) = (j + 1)η
3
As a result For η >> 1
M = jgN µB
and for η << 1
1 µB B
M = j(j + 1)g 2 N µB
3 kB T
In contrast, the classical treatment gives
Z 2π Z 1

z= d(cos θ)eβµB cos θ = sinh βµB
0 −1 βµB

and hence
1 ∂lnZN
M= = N µL(βµB)
β ∂B
where
1
L(x) = cothx − (Langevin function)
x
At high temperatures ( µB << kB T ) one has
1
M = N µ2 βB (Curie Law)
3

48
17 Appendix A: Tensor products of Hilbert
spaces
Let HA and HB denote two Hilbert spaces ( over the same field) of dimensions
m and n. Let ei , i = 1, · · · m be a basis for HA and fi , i = 1, · · · n that
for HB . Through the use of an abstract symbol ⊗ we define mn objects
ei ⊗ fj , i = 1, · · · , m; j = 1, · · · , n arranged in a lexicographical order and
decree them as the basis for a new Hilbert space denoted by HA ⊗ HB of
dimension mn. The abstract ‘multiplication’ of ei ⊗ fj through the symbol
⊗ is assumed to saisfy

(αei + βej ) ⊗ fk = αei ⊗ fk + βej ⊗ fk


(αei ) ⊗ (βfk ) = αβei ⊗ fk
(ei ⊗ fj , ek ⊗ fl ) = (ei , ek ) · (fj , fl )

The last item defines a scalar product on HA ⊗ HB through those on HA


and HB . Further, if ei } and fj } are orthonormal bases for HA and HB then
{ei ⊗ fj } is an orthonormal basis for HA ⊗ HB .
Elements of HA and HB can be divided into two categories : (a) separable
i.e. those which can be written as a tensor product of an element from HA
and an element from HB (b) nonseparable or entangled i.e. those which can
not be written in this way.
Coming to operators on HA ⊗ HB , those which are of the form A ⊗ B
are referred to as local operators and those which can not be expressed in
this way are referred to as non local operators. Of course every operator
can be expressed as a linear combination of local operators much the same
way as every vector in HA ⊗ HB can be expressed as a linear combination of
separable vectors.
An operator A (B) on HA ( HB ) can be extended in a natural way to an
operator A ⊗ I ( I ⊗ B) on HA ⊗ HB . Conversely an operator A ⊗ B can be
‘reduced’ to an operator on HA (HB ) as A · TrB (TrA · B ). This prescription
can be extended to any operator by linearity.
If the operator A on HA is represented by A in the basis {ei } and the
operator B on HB by B in the basis {fj } then the operator A ⊗ B in the
(lexicographically ordered) basis {ei ⊗fj } is represented by the tensor product
A⊗B of the matrices A and B. Recall that the tensor product of two matrices

49
A and B is defined as :
 
A11 B · · A1m B
 · · · · 
A⊗B = 
 · · · · 
Am1 B · · Amm B

The considerations above easily extend to tensor products of three or


more Hilbert spaces.

18 Appendix B: Jacobian identities


[1] Let f and g be functions of two variables x and y and let
     
∂f ∂f
∂(f, g)  ∂x y  ∂y  x 
 
= Det 
 ∂g
∂(x, y) ∂g 
∂x y ∂y x

denote the Jacobian of the transformation. These satisfy (a)

∂(f, g) ∂(g, f ) ∂(f, g)


=− =−
∂(x, y) ∂(x, y) ∂(y, x)

(b)  
∂(f, y) ∂f
=
∂(x, y) ∂x y

[2] If x and y themselves are functions of the variables u and v then using
the chain rule it follows that
(c)
∂(f, g) ∂(x, y) ∂(f, g)
=
∂(x, y) ∂(u, v) ∂(u, v)
and hence
(d)
 −1
∂(f, g) ∂(x, y)
=
∂(x, y) ∂(f, g)

50
[3] From the identities in [1] and [2] one can easily deduce the following
relations:
(a)      
∂f ∂y ∂f
=−
∂y x ∂x f ∂x y

(b)    
∂x ∂f
=1
∂f y ∂x y

(c)      
∂f ∂x ∂y
= −1
∂y x ∂f y ∂x f

19 Appendix C: Gaussian integrals, Surface


area and Volume of a sphere in n-dimensions,
Stirling formula
• Gaussian Integral : one variable
Z ∞ r
−αx2 π
I(α) = dxe = ;α > 0
−∞ α
Proof: Z ∞ Z ∞
2 +y 2 )
2
I (α) = dx dye−α(x
−∞ ∞
Using polar coordintates
Z ∞ Z 2π
2
2
I (α) = rdr dθe−αr
0 0
Z ∞
2
= 2π rdre−αr
0
Z ∞
=π dte−αt ; (r2 = t)
0
π
=
α

51
and hence r
π
I(α) =
α
• Gaussian Integral : n variables
For α1 , α2 , · · · , αn > 0
Z ∞ Z ∞ Z ∞ √
− 21 (α1 x21 +···+αn x2n ) ( 2π)n
dx1 dx2 · · · dxn e =√
−∞ ∞ −∞ α1 α2 · · · αn
Defining  
x1

 x2 

x=
 · ;
 A = Diag (α1 , α2 , · · · , αn )
 · 
xn
the above eqn may be written as
Z ∞ Z ∞ ∞

( 2π)n
Z
− 21 xT Ax
dx1 dx2 · · · e =√
−∞ ∞ −∞ Det A
r 3/2 R ∞ 3/2
=2πgV 2m h2 0
1eβ(−µ) − 1
3/2
= (kB T 5/2 2mh2
Gamma(5/2)B5/2 (Z )

where we have used the fact that Det A = α1 α2 · · · αn


This formula though derived for A diagonal remains true for any positive
symmetric matrix A.
• Area and volume of a sphere in n dimensions
From purely dimensional considerations the volume Vn (R) of a sphere in
n-dimensions is given by
Vn (R) = Cn Rn
Further, if Sn−1 (r) denotes the area of the surface of a sphere of radius r in
n dimensions then
Z R
Vn (R) = drSn−1 (r)dr i.e. dVn (r) = Sn−1 (r)dr
0

52
expressing the fact that the elemental volume of a sphere between r and
r + dr equals the surface area of the sphere of radius r times dr. This implies
that
Sn (r) = nCn rn−1
. Hence Z R
Vn (R) = nCn rn−1 dr
0
Now Z
Vn (R) = dx1 · · · dxn
x21 +···+x2n ≤R2

Using polar coordinates in n-dim


Z R Z
n−1
Vn (R) = r dr dΩn−1
0

which implies Z
dΩn−1 = nCn

Consider the equation


Z ∞ Z ∞
dx1 dx2
−∞ ∞
Z ∞
2 2
dxn e−(x1 +···+xn ) == (π)n
−∞

Using polar coordinates


Z R R
√ n √
Z Z
n−1 −r2 2
drr e dΩn−1 = ( π) i.e nCn drrn−1 e−r = ( π)n
0 0

which in turn yields



√ ( π)n
Cn Γ(n/2 + 1) = ( π)n i.e Cn =
Γ(n/2 + 1)

and hence
√ n √
π ( π)n n−1
Vn (R) = Rn ; Sn−1 (R) = 2 R
Γ(n/2 + 1) Γ(n/2)

53
• Stirling approximation
Z ∞
n! = dttn e−t
Z0 ∞
= dte−φ(t) ; φ(t) = t − nlnt
0

Now
d n d2 n
φ(t) = 1 − ; φ(t) =
dt t dt2 t2
which imply that φ(t) has a minimum at t = n. Expanding φ(t) in a
Taylor series around t = n and retaining only the quadratic terms ( Gaussian
approximation) we have
Z ∞ Z ∞
1 1 2
−φ(n)− 2n (t−n)2 −φ(n)
n! ' dte =e dxe− 2n x
0 −n
 n n Z ∞ 1 2
 n n
for large n → dxe− 2n x = 2πn
e −∞ e

Thus for large n


lnn! ' nln − n
.

54

You might also like