Statistical Mechanics – Phys504 – Fall 2006 Anthony J.

Leggett
Lecture #3 Department of Physics, UIUC

Thermodynamic equilibrium
P
Let’s consider a situation where the “Universe”, i.e. system plus its environment E,
possesses a total energy1 E0 and has come to complete thermodynamic equilibrium.2 What
P
now will be the energy ε of the system ? Consider a small range ∆ε surrounding ε. The
weight of the range ε, ε + ∆ε in the total density of states available to the universe will
P
be proportional to the numbers of states of in this range × the number of states of
E. The former is by definition Ws (ε)∆ε, and because of energy conservation the latter is
WEnv (Eo − ε)∆(Eo − ε) ≡ Wenv (E0 − ε)∆ε. Thus the total number of states available to
P P
+E with the energy of in the specified range is

N (ε) = Ws (ε)Wenv (Eo − ε)(∆ε)2

This function is very sharply peaked as a f (ε) around its maximum, so we can say that the
equilibrium (or overwhelmingly probable) value of ε is very close to the maximum, i.e. ε is
determined (since ∆ε is fixed) by

Ws (ε)Wenv (Eo − ε) = max .

Let us take the logarithm and differentiate with respect to ε;

d d d
ln Ws (ε) = − ln Wenv (Eo − ε) ≡ + ln W (εenv )
dε dε dεenv
(εenv ≡ Eo − ε). But apart from the constant kB , ln W (ε) is just S(ε), so the condition for
equilibrium is
dS dSenv
=
dε dεenv
But in thermodynamics, the condition for two bodies which are in thermal contact (i.e. can
exchange energy freely) to be in thermal equilibrium is that their temperatures are equal,
thus we define

T ≡ (dS/dε)−1 (not T = dS/dε, cf. below)

1
Or an energy in a small rangeP[Eo , Eo + ∆E]; this does not affect the outcome. See problem.
2
For the moment we suppose and E can exchange energy but not particles or volume.

1
 This immediately explains the so-called “zeroth low of thermodynamics”: two bodies
which are each in thermal equilibrium with a third are automatically, if placed in contact,
P
in thermal eqm. with one another. Also, let’s consider the case that initially and E
are not in equilibrium, so that the quantity N (ε) is not a maximum. According to the
considerations of l.2, it will tend to a maximum, which is equivalent to the statement that
the total entropy, kB (ln Ws (ε) + ln Wenv (Eo − ε)) will increase:

d
(S(ε) + Senv (εenv )) > 0
dt

Writing dS/dt = dS/dε · dε/dt and dεenv /dt = −dε/dt and using the above definition of
temperature, this gives  
dε 1 1
− >0
dt T Tenv
i.e. heat flows into (out of) the system if it is colder (hotter) than the environment. (Note
that had we defined T = dS/dε rather than (dS/dε)−1 , we would have got the opposite
result).
Suppose the system is just slightly out of equilibrium with the environment:

(T = Tenv + δT, δT → 0).

This means we are close to the maximum of total entropy; the deviation of T ≡ ∂S/∂ε from
its eqm. value −∂Senv /∂ε is (by definition!) prop. to δT , so the deviation of the entropy
of the Universe from its maximum value is of order (δT )2 . Thus to order δT

∆S = −∆Senv

i.e. the process is reversible by changing the sign of δT . Under these conditions we have

∆E = T ∆S ≡ d̄Q (d̄Qis by definition energy input at constant N, V )

which is a special case of the first law of thermodynamics (see below).

What can we say about the case of finite disequilibrium? It is convenient to redefine ε as
the difference between the value of the system energy and its value in thermal equilibrium
with the environment at temperature3 Tenv . Consider first the entropy of the Universe,

3
Note that because the environment is by construction “very large”, its temperature is negligibly changed
when energy is transferred to or from the system (cf. below).

2
which is a sum of the entropies of the system and the environment, and expand it in ε:
   
eqm ∂S ∂Senv
Suniv = Suniv + ε− ε + S (2) + Senv
(2)
∂ε eqm ∂εenv
eqm
= Suniv + S (2) + Senv
(2)

where S (2) denotes all the higher-order terms in the expansion. Now, since the environment
(2)
is “very large”, Senv is vanishingly small (e.g. the second-order term is ∂ 2 Senv /∂ε2env ≡
2
(∂/∂εenv ) (1/Tenv ) = −1/Tenv cenv , and cenv → ∞). Thus,

eqn
Suniv = Suniv + S (2)

and since we know that Suniv is a maximum at equilibrium, it follows that S (2) < 0.
Now expand S itself in ε:
 
∂S ε
S(ε) = Seqn + εeqm + S (2) (ε) ≡ Seqm + + S2 (ε)
∂ε T

Suppose the system goes from a nonzero value of ε to equilibrium (ε = 0). The change
in energy ∆E is then −ε, and we know that S2 (ε) < 0, so

∆E ∆E d̄Q
∆S ≡ Seqm − S(ε) = − S2 (ε) ≥ ≡
Tenv Tenv Tenv

i.e. the increase in entropy is always ≥ the heat input/Tenv . From the above argument it
also follows that ∆Suniv is of second order in the deviation from eqm., δT .

The Helmholtz free energy

In general, when a system is in thermal isolation from any environment, it will tend
to maximize its entropy at constant energy. On the other hand, if it is in thermal contact
with an environment (at constant V and N ) then as we have seen it is the total entropy of
P
the Universe, ( +E), which is maximized, and this leads to the equality of temperature:
T ≡ (∂S/∂ε)−1 = Tenv . In fact, we already saw that for such a system

∆S ≥ ∆E/Tenv

the equality being reached when T = Tenv . Consequently, the quantity

F (Tenv ) ≡ E − Tenv S (*)

3
will decrease until it reaches its minimum value in the equilibrium state. Thus, for a system
in thermal equilibrium, at constant V and N , with an environment at temperature T , we
have the general principle
F (T ) ≡ E − T S = min.

Evidently for a system whose environment is at zero temperature, this leads to E = min.
as we expect. We will see below that (*) is a special case of a class of minimum principles.

Adiabatic processes

So far, we have supposed that all the parameters which control he Hamiltonian of the
P
system (volume, particle number, etc.) are fixed. (so that the only interactions of with E
is by energy exchange at the microlevel). Let us now for the moment thermally isolate the
system, and suppose that the Hamiltonian depends on some parameter λ which we allow
to vary slowly in time. (An example might be a slow change in one of the dimensions of
the system). The energy of the system will then evidently also be a function of time. What
about the entropy?
It is actually easiest to consider this question in a QM picture. Suppose that we have a
system subject to a Hamiltonian which for t < 0 is constant and for > 0 is a function of
some continuously varying control parameter λ(t):

Ĥ = Ĥo + ∆Ĥ[λ(t)] ∆Ĥ = 0, t < 0

Suppose that at time 0 the system is known to be in a particular eigenstate |ψn i of Ĥo
with eigenvalue En . For t > 0 both the eigenfunctions and the eigenvalues of Ĥ will be
time-dependent. However, provided that λ(t) is indeed continuously varying, we can keep
track of each of the eigenvalues and eigenfunctions Em (t), |ψm (t)i. The quantum adiabatic
theorem then states that provided λ(t) is sufficiently slowly varying, the system will, with
probability approaching 1, stay in its original state n, i.e. in the state |ψn (t)i with energy
En (t). The criterion for λ to be “sufficiently slowly” varying is, crudely speaking, that for
all states m 6= n !2
  (t)
∂ Em − En (t)
∆Ĥ(λ(t))  (‡)
∂t mn ~
At first sight this criterion looks very difficult to fulfill in a many-body system since as we
have seen the energy levels of such a system are densely spaced. However, it often turns

4
out that for such pairs of levels the LHS of (‡) is also very small, so that the condition is
nevertheless fulfilled. A process taking place under the condition (‡) is called adiabatic (in
the quantum context).
Suppose now we start with our thermodynamic system in a (pseudo-) microcanonical
ensemble, so that formally the DM is

ρ̂ = 1̂/D · Θ (Eo , E, Eo + ∆E)

i.e. the system is with equal probability in any of the energy eigenstates lying in ∆E. The
entropy is just kB lnD. It is clear that an evolution which is adiabatic in the quantum sense
cannot change those probabilities, and since the system is thermally isolated there are no
other effects. Consequently, if we define

adiabatic process (SM ) ≡ adiabatic (QM) and thermally isolated

then in an adiabatic (SM) process the entropy of the system cannot change.
Thus, the dependence of any thermodynamic quantity on λ in an adiabatic process must be
just the derivative at constant S, and in particular
   
∂E ∂E
=
∂λ ad ∂λ S

Note that according to the above definition every adiabatic process must be reversible
(since the entropy of the system and hence of the Universe is unchanged) but not every
reversible process is adiabatic (e.g. isothermal heat transfer is not). C.f. LL§11 (p. 37 in
1958 edn.). Also note that in an adiabatic process the temperature of the system in general
changes.
Let’s consider the quantity (∂E/∂λ)s . We know that dS = 0 in an adiabatic process, so
     
∂S ∂S ∂S
dλ + dE ≡ dλ + T −1 dE = 0
∂λ E ∂E λ ∂λ E

(∂E/∂λ)s = −T ∂S


so ∂λ E
.

We can define Xλ ≡ −(∂E/∂λ)s as the generalized “force” exerted on the body by the
source of λ.
Consider a system in contact with an environment, and suppose that the total “λ” is
conserved: λP + λE = const. (Example: λ = volume, particle number, magnetization

5
(under appropriate circumstances)). Suppose that λ can be freely exchanged between the
system and its environment, whereas energy may or may not be exchangeable. What is the
condition for equilibrium?
Suppose we first assume that the system is thermally isolated, so that no heat can be
P
transferred between and E. Then by an argument exactly similar to the one at the
beginning of this lecture, we must adjust the λ’s so as to maximize the total entropy of the
Universe, and this leads to    
∂S ∂Senv
=
∂λ ∂λenv
in other words
(env)
Xλ X
= λ
T Tenv
In general, therefore, the “generalized forces” are not equal. However, if in addition we
now allow transfer of energy between the two systems, then we have in eqm. as above
(independently of λ) T = Tenv , and so

(env)
Xλ = Xλ

The best-known cases of “λ” are the volume and the particle number. Recalling that
Xλ is −(∂E/∂λ)S , we see that in the first case the ”generalized force” is −(∂E/∂V )S,N ≡ P ,
the pressure, and in the second it is (−(∂E/∂N )S,V ) ≡ −µ, the negative of the chemical
potential. If then we exclude for the moment other λ’s (corresponding to magnetization) we
can write for any infinitesimal reversible change
     
∂E ∂E ∂E
dE = dS + dV + dN
∂S V,N ∂V S,N ∂N S,V

≡ T dS − P dV + µ dN

which is just the first law of thermodynamics. (“thermodynamic identity”)

Thermodynamic potentials

Let us consider a system which possibly starts out of equilibrium with its environment,
but eventually comes into equilibrium with it under various possible conditions (isother-
P
mal/adiabatic, isochoric/isobaric, etc.). For any given conditions of contact between and
E, we can ask (a) what thermodynamic property of the system has to be minimized in

6
equilibrium? (b) if we start from a nonequilibrium state, how much work can we get out
of the system during the transition to equilibrium? The answer to both questions is given
by the relevant (free) energy. At first sight one would think that there should be 8 different
P
free energies, corresponding to the 8 different conditions of equilibrium between and E,
but we shall see that in effect there are only 4. In the following To and Po represent the
temperature and pressure of the environment, which can be regarded as constant during any
process occurring in the system.
Consider the energy balance of the system, for the moment treating N as constant. We
have for any process, reversible or irreversible, the relation

∆E = ∆Q − Po ∆Venv + ∆W = To ∆Senv − Po ∆Venv − ∆W.

where ∆Q is the heat input from the environment and ∆W is the (non-P dV ) work done,
on a third system distinct from the environment, e.g. by magnetization, or . . . (we do not
need to be explicit). Now evidently ∆V = −∆Venv , and from the fact that the entropy of
the Universe can only increase, ∆S ≥ −∆Senv . Hence

∆W ≤ −(∆E − To ∆S + Po ∆V ) ≡ ∆Wmax

∆Wmax is the maximum work obtainable from the system as it comes into equilibrium with
the environment. Note that this maximum is obtained when the inequality ∆S ≥ −∆Senv is
an equality, i.e. when the process is reversible. Since ∆W cannot be negative (equilibration
is a spontaneous process!) it follows that the quantity ∆E − To ∆S + Po ∆V is ≤ 0, i.e. if
we define a quantity K such that

∆K ≡ ∆E − To ∆S + Po ∆V

then (a) in the process of attaining equilibrium, dK/dt ≤ 0, and (b) ∆K is the maximum
work obtainable.
Let’s consider some special cases

Case I. ∆S = ∆V = 0 (adiabatic isochoric process). Clearly in this case K = E ≡ E(S, V )

i.e. the relevant energy is just the total internal energy of the body. This is natural since
no heat is absorbed and no external P dV work is done.

7
Case II. ∆S = 0 but ∆V 6= 0: rather, P = Po (adiabatic, isobaric process). Now ∆K =
∆E + P ∆V , or since P is constant, ∆(E + P V ). Hence in this case

K = E + P V ≡ H(S, P )

the so-called enthalpy or heat function.

Case III. ∆V = 0 but ∆S 6= 0, rather T = To (isothermal isochoric process). Now

∆V = ∆E − T ∆S, or since T is constant, ∆(E − T S). Thus,

K = E − T S ≡ F (T, V )

the Helmholtz free energy.

Finally,

Case IV. Neither ∆S or ∆V are zero, but T = To and P = Po (isothermal isobaric process):
Now ∆V = ∆E − T ∆S + P ∆V , so

K = E − T S + P V ≡ G(T, P )

the Gibbs free energy.

The reason for expressing the four free energies as functions of four different pairs of the
variables (S, V : T, P ) (since only 2 are independent, we could of course in principle write
G as a f (T, V )!) is that their differentials then come out in simple form: e.g.

dG = dE − T ds − SdT + P dV + V dP

but using the 1st law, dE − T ds + P dV = 0, so

dG = −SdT + V dP (etc.)

From these differentials we can obtain a large number of Maxwell relations, e.g. from the
above and    2   2   
∂S ∂ G ∂ G ∂V
≡− ≡− =−
∂P T ∂P ∂T ∂T ∂P ∂T P
Finally, we need to return to the role of total particle number. When we include this,
the thermodynamic identity is generalized to (above)

dE = T dS − P dV + µdN

8
where µ is the chemical potential. So in analogy to the way we have treated the entropy
and volume, for the case of osmotic contact with the environment, let us say also isothermal
and isobonic, it would be natural to introduce a quantity

Φ ≡ E − T S + P V − µN ≡ G − µN, µ ≡ (∂E/∂N )S,V ≡ (∂G/∂N )P,T

which means µ ≡ (∂G/∂N )P,T = f (P T ) ≡ G/N . Thus, Φ = 0 identically! It further

follows that, for example, the quantity which we would want to introduce for a process at
constant T, V and µ, i.e. F −µN , is just −P V and does not contain any explicit reference to
E. This quantity, and the other two (corresponding to S, V, µ and S, P, µ) are occasionally
introduced but do not seem to have standard names. (cf. KIUH § 1.14).