Math C6.

2a, Statistical Mechanics, University of Oxford

Solution Sheet 3 (11/02/12)

Mason A. Porter, Oxford Centre for Industrial and Applied Mathematics

1. Problem 4.2 in Sethna (p. 70). [Liouville versus the damped pendulum]
Note: Although we focus predominantly in statistical mechanics on very high-dimensional systems, many of
the ideas that we study (including Liouvilles theorem) are highly relevant (and can be extremely useful) for
low-dimensional systems as well.

a. In the steps leading from equation (4.5) to equation (4.7), why does Liouvilles theorem not apply to the
damped pendulum? More specifically, what are pp and q
q
?

Solution: We have the Hamiltonian-like equations of motion,

p
q = , p = p K sin q .
M
If it werent for the damping term p in the second equation, then this system would be Hamiltonian.
p
In the proof of Liouvilles theorem, we used p + qq = 0, but clearly this is not the case, because
p q
= , = 0.
p q
Because q is independent of q, that term vanishes. How has this condition been broken? Recall that the
condition that this sum vanishes originally came from the fact that the order of taking partial derivatives
doesnt matter, so
q ? 2 H 2 H ? p
= = = .
q qp pq p
However, this assumes that there exists H for which q = H H
p and p = q . In the present case, such a
function H does not exist.
b. Find an expression for the total derivative d
dt in terms of for the damped pendulum. How does the
probability density vary with time? If we evolve a region of phase space of initial volume A = pq, how
will its volume depend upon time?
Solution: As an aside, we know that the long-time density is
t= (q, p) = (p)(q).
We begin with the definition of the total derivative,
d
= + q + p.
dt t q p
The continuity equation then gives
 
(q) (p) q p
= = q p + .
t q p q p q p
Moving the first two terms from the right to the left allows one to write
 
d q p
= + = .
dt q p
The solution then gives exponential growth:
(t) = 0 et ,
where 0 is the initial density. What does this mean? At any point, the density at that pointas we follow
the points trajectory in phase spaceis increasing. This makes sense, because neighboring trajectories are
getting smushed together with it. Because number is conserved, we see that an initial volume A = pq
will evolve as A(t) = A exp[t].
 2

2. Look up the Fermi-Pasta-Ulam problem and comment briefly on its relation to issues in statistical mechanics.
[Your answer should include some comment on equipartition.] What is paradoxical about this problem?

Solution: This is another one of those questions where the answer can be very long indeed. (There have
been several decades worth of research on this topic!) A short summary for laypeople (no equations except for
1 + 1 = 3 !) of the issues and the history of this problem can be found in a magazine article that I wrote,
which you can download from http://people.maths.ox.ac.uk/~porterm/papers/fpupop_final.pdf. The
article is called Fermi, Pasta, Ulam and the Birth of Experimental Mathematics.
The original FPU experiment is one of the most important research endeavors in math and physics of the
20th century, and it arguably marked the birth of nonlinear science as its own discipline. It is certainly one
of the most important things in the history of statistical mechanics.
The FPU problem is a discretized nonlinear spring, so its a bunch (ideally, a very large number) of nonlinear
springs connected to each other. FPU (and Mary Tsingou) used their mathematical experiment to try to
demonstrate equipartition in this system. They excited one mode of the system and expected the systems
energy to eventually be distributed evenly among all modes (i.e., equipartition). Fortunatelybecause a ton of
awesome discoveries have resulted from what amounts to a historical accidentthat is not what happened, as
there seemed to be a recurrence of the initial condition, in that eventually most of the energy was once again
in the initially-excited mode of the system.
As it turns out, as several decades worth of subsequent research have shown, if FPUT had used an initial
condition with higher-energy, then equipartition would have occurred!
There technically isnt anything paradoxical per se, but there is a wonderful interplay of ordered behavior
and chaotic behavior (of multiple types). And far from challenging the premise of statistical mechanics, which
is what people thought about FPU for several decades, statistical mechanics has come back stronger and more
powerful than ever before. [OK, so that is a rather non-technical way of phrasing it.]
My paper gives several of the technical references if you want to explore any of these issues further. The global
story actually changed while I was writing the paper, so its a really good thing that my coauthors and I took
a couple of years longer than we had promised to do it. :)
3. Problem 5.2 in Sethna (p. 92-93). [Burning Information and Maxwellian Demons]

a. How much work is done by the atom as the piston retracts?

Solution: The work is

Z Z V2
kB T
W = P dV = dV = kB T ln V  .

T V1

The volume V2 = 2V1 , so we get W = kB T ln(2).

b. Explain how to convert one bit into another.

Solution: Lower the upper piston; Remove the partition; Move both pistons upwards slowly; introduce
partition; lower the lower piston.
(We do this to move things so that the volume remains constant.)
c. What is the change in entropy if the volume doubles?

Solution: With the partition we have Swith = kB T ln V , and without the partition we have Sw/o =
kB T ln(2V ). This gives S = kB ln(2).
The initial information is Si = 0 and the information afterwards is Sf = 2 12 log(1/2) = 1. Hence, the
change in information is Ss = 1. (Can you get this directly from scaling S into units of information.)
d. Demonic States

Solution: The demon still remembers the (now erased) bit of information on the tape. This information is
now garbage, as it is unrelated to anything in the outside world. In this sense, the demon is in an unknown
state, with entropy kB log 2. It would take work kB T log(2) to return the demons memory to a known,
useful state for doing another measurement, just as in part (a). Hence, there is no violation of the second
law.

4. Problem 5.5 in Sethna (p. 94-95).[P-V Diagram] Which of the following are true?
 3

(a) The cycle is reversible: no net entropy is created in the universe.

Solution: False. Leg a has a warm system exchanging heat irreversibly with a cold bath, and leg b has it
touching a hot bath. Hence, a net entropy is generated during a cycle.
(b) The cycle acts as a refrigerator, using work from the piston to draw energy from the cold bath into the hot
bath, cooling the cold bath.

Solution: False. The piston does not cool the cold bath. Only in leg a is there a heat exchange with the
cold bath, and heat is entering the cold bath during that leg. [The net energy E = (3/2)P V goes down in
the piston during that leg.]
(c) The cycle acts as an engine, transferring heat from the hot bath to the cold bath and doing positive net
work on the outside world.
R
Solution: False. The net work done by the piston is the area P dV < 0 swept out by the curve. It is
not an engine, because it absorbs work.
(d) The work done per cycle has magnitude |W | = P0 V0 |4 log 4 3|.
R R
Solution: True. The net work done by the outside world on the piston is [Pc Pb ] dV = (4P0 V0 /V
P0 ) dV = P0 V0 (4[log 4V0 log V0 ]) 3) = P0 V0 (4 log 4 3).
(e) The heat transferred into the cold bath, Qc , has magnitude |Qc | = (9/2)P0 V0 .

Solution: True. In leg a (the only one that matters), the gas in the piston has energy U = (3/2)N kB T =
(3/2)P V . It thus starts with energy (3/2)(4P0 V0 ) = 6P0 V0 and ends with energy (3/2)P0 V0 , so a net
energy (9/2)P0 V0 enters the cold bath.
(f) The heat transferred from the hot bath Qh plus the net work W done by the piston onto the gas equals
the heat Qc transferred into the cold bath.

Solution: True. This follows from energy conservation and the fact that the system returns to its initial
state after a cycle.

5. Problem 5.7 in Sethna (p. 95). [Does Entropy Increase?]   Show for any function f () that ft() =
~q
[f ()~v ] = p (f ()p ) + q (f ()q ), where ~v = ~p,
P is the 6N dimensional velocity in phase space.
R f ()
Hence, show that t d~ q d~
p = 0, assuming that the probability density vanishesR at large momenta and
positions and that f (0) = 0. Then use this to show that the entropy S = kB log is constant in time.

Solution: Let = (q , p , t) be a probability distribution in phase space and (t) = (q (t), p (t), t) be the
distribution along a particular trajectory or streamline, given by q (t) and p (t). [Were thinking here of a
streamline associated with any one DOF, which is why this is written with the suffix rather than considering
the whole vector at once. One can write this out with the whole vector as well. Its a matter of what exposition
one prefers.] Then for any f () satisfying the assumptions described in the problem, we have
f () df ()
f ((q , p , t)) = (q , p , t).
t t d t
Liouvilles theorem applies along the streamline:
d(t) d
(q (t), p (t), t) = (q , p , t) + [(q , p , t)] ~v = 0.
dt dt t
We therefore have
 
f () df ()
= ~v = (f ) ~v = [f~v ] ,
t d
P h p q
i
where the last step holds because ~v = p
+ q
= 0 for a Hamiltonian system. [This follows from
Hamiltons equations of motion (4.6).] Using Gausss law, we get
Z I
f ()
d~q d~
p = f () ~v d~a = 0 ,
t
phase space surface at R
 4

assuming that (~q, p~) = 0 for large p~ and/or ~q (i.e., has finite support). [Note: We apply Gausss law for a
ball of radius R and then we let R so that the integral on the left-hand-side is then over the entire phase
space. Gausss theorem in 6N dimensions works the same way as it does in 3 dimensions.]
Now lets look at the rate of change of the entropy
Z
dS
= kB ( log ) d~q d~
p.
dt t

Taking f () = log , we see that f () 0 as 0. The above result therefore applies:

dS
= 0.
dt
The entropy is a constant in time.
6. Problem 5.12 in Sethna (p. 98-99). [Rubber Band]

Note 1: This problem meshes very well with one of the main learning outcomes of Math C6.2a. Namely,
given an ensemble that is defined in the problem, work out the details with the usual types of considerations.
(In this example, as mentioned several times below, you should compare the presented situation with an ideal
gas in a container with the volume V held constant, and you should compare increasing/decreasing V versus
increasing/decreasing the length L below.)

Note 2: The length L is being measured from the reference x-axis location, as indicated in the figure in Sethnas
book.

a. Find an exact formula for the entropy of this system in terms of d, N , and L.

Solution: We count the states. Suppose that there are n+ rightward links and n leftward lefts, so
n+ + n = N is the total number of links and (n+ n )d = L is the total length. This gives n+ =
N/2 + L/(2d), and the number of configurations in this constant-length ensemble is then
 
N N! N!
= = = .
n+ n+ !n ! [N/2 + L/(2d)]! [N/2 L/(2d)]!

Taking the logarithm gives

S = kB log = kB [log N ! log([N/2 + L/(2d)]!) log([N/2 L/(2d)]!)] .

b. Find an expression for the force F exerted by the bath on the molecule in terms of the bath entropy.
Using the fact that the length L must maximize the entropy of the universe, write a general expression for
F in terms of the internal entropy S of the molecule.

Solution: We have the relationship dEbath = (F )dL. Thus,

1 Sbath 1 Sbath
= = .
T Ebath F L
However, we know that L maximizes Suniv , so
Suniv Sbath Sband
= + = 0.
L L L
Consequently,
Sband
F =T .
L

c. Use the model of the molecule from part (a), the general law of part (b), and Stirlings formula to write
the force law F (L) for our molecule for large lengths N . What is the spring constant K in Hookes law
F = KL for our molecule when L is small?
 5

Solution: Recall the result of (a) and apply Stirlings formula:

S kB {N log N [N/2 + L/(2d)] log [N/2 + L/(2d)] [N/2 L/(2d)] log [N/2 L/(2d)]}
4N 2 d2
    
kB Nd + L
= N d log L log .
2d (N d + L)(N d L) Nd L

We now differentiate to get

4N 2 d2
      
kB Nd + L Nd + L
F =T Nd log L log log
2d L (N d + L)(N d L) L Nd L Nd L
 
kB T Nd + L
= log .
2d Nd L

[Note that I am writing the above line as an equality because the calculations after Stirlings approximation
have been exact ones.]
Finally Taylor expand L  N d to get
 
kB T L kB T
F (L) log 1 + 2 L = KL ,
2d Nd N d2
kB T
where K = N d2 .
d. If we increase the temperature of our rubber band while it is under tension, will it expand or contract?

Solution: If we increase the temperature, the rubber band will contract. The lower-entropy stretched-out
states become more expensive. One can try to understand this as follows:
L  L  F 
 =  .
T F F T T L


This gives a Maxwell relation

F  S  F KL
 =  = =
T L L T T T
and the spring constant
L  1
 = .
F T K
Combining these results gives
L  L
 = < 0,
T F T
so the rubber band contracts with increasing temperature and constant force.
e. True or False questions.

Solution: In order, the answers are FFTFTT. The first four are more or less exclusive (though this isnt
quite true... have a think about that). When we stretch the rubber band, the configurational entropy
decreases, so the vibrational entropy increases, which in turn increases the temperature. Inversely, relaxing
the band increases configurational entropy, thereby decreasing vibrational entropy and temperature.
We have seen that work on the system (i.e., changing the length) causes a temperature change. However,
notice that this works oppositely from what we saw in the ideal gas, where compression cools and expansion
heats. This is because the rubbers force is inwards, wheras pressure is outward. That is, dWgas = P dV
but dWband = F dL.