Kevin Zhou Physics Olympiad Handouts

Thermodynamics I
For an introduction to basic thermodynamics at the right level, see chapter 2 of Wang and Ricardo,
volume 2. For more detail, see chapters 1–9, 19, and 20 of Blundell and Blundell. For interesting
discussion, see chapters I-39 through I-43 of the Feynman lectures. There is a total of 78 points.

1 Ideal Gases and Heat Engines

Questions about ideal gas heat engines are straightforward. They were very common on the USAPhO
a decade ago, though problems today tend to require a deeper understanding of thermodynamics.
Nonetheless, it’s fundamental material that you should know.

Idea 1
The ideal gas law comes in a few common forms,
ρRT ρkB T
P V = nRT = N kB T, P = = .
µ m

Here, Avogadro’s number is NA = N/n, so that R = NA kB , m is the mass of a gas molecule,

and µ = NA m is the molar mass. The first law of thermodynamics says

dU = dW + dQ, dW = −P dV.

The heat capacity at constant volume is defined so that dU = nCV dT at constant volume.
As a result, if we define CP = CV + R, we have

3R/2 monatomic
( 
nCV ∆T at constant volume
Q= CV = 5R/2 diatomic .
nCP ∆T at constant pressure 
3R polyatomic


Using the first law, we can derive the results

Vf
W = −nRT log for isothermal expansion
Vi
and
P V γ = constant for adiabatic expansion, γ = CP /CV ,
both of which you should easily be able to rederive.

Idea 2
For a cyclic process that takes in heat Qin from a hot reservoir at temperature TH and outputs
heat Qout to a cold reservoir at temperature TC , the work done is W = Qin − Qout . The
efficiency η = W/Qin is maximized by the Carnot engine, for which η = 1 − TC /TH .

1
 Kevin Zhou Physics Olympiad Handouts

Remark

The study of thermodynamics arose from efforts in the early 19th century to understand
the efficiency of steam engines. However, Carnot’s ideal reversible engine doesn’t resemble
practical engines, since the isothermal steps take place at zero temperature difference, and
therefore take infinite time. Suppose the isothermal steps take place at temperature T1 < TH
and T2 > TC , and that the rate of heat flow is proportional to the temperature difference.
This engine isn’t reversible, but it does yield a nonzero average power. Whenp T1 and T2 are
set to maximize the average power, it turns out the efficiency is simply 1 − TC /TH , and
this expression more closely matches the efficiencies of real engines.

[2] Problem 1. Derive the Carnot efficiency using the fact that (a) the engine is reversible, so a
complete cycle leaves the entropy of the universe unchanged, or (b) directly from idea 1.

Solution. (a) At the end of a heat engine cycle, the working fluid comes back to its original state,
so its entropy is unchanged. When an engine is reversible, it also leaves the entropy of the
surroundings unchanged, so that the total entropy of the universe doesn’t go up. (If it did go
up, then we wouldn’t be able to reverse the cycle, by the second law of thermodynamics.)
The increase in entropy of the cold reservoir is Qout /TC , while the decrease in entropy of the
hot reservoir is Qin /TH . Setting these equal to each other, we find
Qin − Qout TC
η= =1−
Qin TH
as desired.

(b) There are a lot of ways to do this, but we’ll show one that will be useful later. Note that
during the heating and cooling steps, we have

dQ = dU − dW = nCV dT + P dV.

Using the ideal gas law, we have

1
dT = (P dV + V dP )
nR
and plugging this in gives
CP CV
dQ = P dV + V dP.
R R
Dividing by T and using the ideal gas law again, we have
dQ dV dP d(P V γ )
= nCP + nCV = nCV .
T V P PV γ
Therefore, the quantity ∆Q/T for the heating and cooling steps only depends on the change
of the quantity P V γ . This is equal and opposite for those steps, since P V γ stays the same
in the adiabatic steps, which means Qin /TH = Qout /TC . The derivation then continues as
in part (a). (Of course, what we’ve done here is essentially just deriving the expression for
the entropy, up to constants, without explicitly calling it that. This will be explored in more
detail in T2.)

2
 Kevin Zhou Physics Olympiad Handouts

Remark
The most common mistake students make in this problem set is forgetting to account for the
work done by the atmosphere.

01W
01W
[3] Problem 2. USAPhO 2009, problem A4.

[3] Problem 3. USAPhO 2011, problem A1.

If you want further practice, see USAPhO 1998 A1, 2008 A2, and 2010 A3. This kind of routine
question should be easy. Some competitions try to make them harder by making the cycles more
complicated (some truly crazy ones have been considered in the literature), but this is contrived
and doesn’t really require much insight. Instead, we’ll move on to slightly subtler problems.

Example 1

A cold room is initially at temperature T . The heater is turned on, raising the temperature
to T + ∆T . Assuming the thermal insulation is ideal, at most what fraction of the energy
released by the heater stays in the room?

Solution
As long as the room has any contact with the outside at all, air will leak out to set the
pressure equal to atmospheric pressure. Its volume also stays the same, so by the ideal gas
law, N kB T stays the same, but this is proportional to the internal energy of the air. Thus,
at most 0% of the energy released by the heater stays in the room; the increase in average
energy per molecule is exactly compensated by the decrease in the number of molecules. If
there are thermal losses, the total internal energy of the air in the room actually decreases.

Example 2

A thermally isolated cylinder is divided into two compartments by a thermally conductive

piston. Initially, the piston divides the cylinder into two compartments, A and B, of equal
volume V /2 and temperature T0 . One mole of monatomic gas is in each compartment. An
external agent slowly moves the piston to the side until the volumes are V /3 and 2V /3.
Throughout this process, the temperature remains uniform. What is the final temperature?

Solution
The tricky thing about this problem is that the pressures in the two compartments aren’t equal;
this is possible because the external agent is holding the piston. Instead, the temperatures
are made equal by heat conduction. The work done by the agent is
 
dVA dVB
dW = −pA dVA − pB dVB = −RT + .
VA VB

On the other hand, this is also equal to the increase in energy,

3
dW = dU = nR dT = 3R dT.
2

3
 Kevin Zhou Physics Olympiad Handouts

Combining the two gives a differential equation,

Z Z Z
dT dVA dVB
3 =− −
T VA VB
which means the final temperature Tf obeys

TF 2 4 9 32/3
3 log = − log − log = log , TF = T0 .
T0 3 3 8 2

[3] Problem 4 (EstPhO 2002). In this problem we consider the combustion cycle of a car engine.
Model the engine as a cylinder with a piston on the left and a valve on the right.

The steps of the process are as follows.

1. Gas entry: the piston moves from the rightmost position to the leftmost; fresh air comes in
through the valve and fills the cylinder.

2. Pressure increase: the valve closes, and the piston quickly moves back to the rightmost
position.

3. Work: fuel is injected in the cylinder and is ignited; you may model this process as occurring
instantaneously. Then the gas starts expanding and pushes the piston to the leftmost position.

4. Gas disposal: the valve is opened. The piston is pushed to the right at constant pressure until
it reaches its rightmost position, and the process then repeats.

Neglect friction and heat conduction, suppose the number of fuel molecules is negligible compared
to the number of air molecules, and treat air as a diatomic ideal gas. Let k be the ratio of the
maximum and minimum volumes of the cylinder. Draw the cycle on a P V diagram and find its
efficiency.

Solution. This problem is a bit trickier because it’s less clear how to treat the steps. Of course,
the second step is just an adiabatic compression, but the third is subtle. Since the number of fuel
molecules is negligible, burning the fuel essentially just rapidly deposits energy into the system,
raising its temperature. Thus, the first half of the third step is an isochoric (constant volume)
heating; the second half of the third step is an adiabatic expansion.
Finally, the fourth and first steps should be regarded as one unit. When the valve is opened, the
gas in the cylinder quickly falls to atmospheric pressure, as it freely expands out. Then the piston
moves to the right, doing work Patm ∆V . In the first step, the piston moves to the left, pulling in
fresh air and doing work −Patm ∆V . So the first half of the fourth step has an isochoric pressure
decrease. The second half of the fourth step, and the first step, do no net work, and function solely
to pull in fresh air.
This tells us what’s going on, but where are the heat reservoirs? The heating step occurs when
the fuel is burned, so the burnt fuel itself is effectively the hot reservoir. The gas is cooled by letting

4
 Kevin Zhou Physics Olympiad Handouts

it leave and replacing it with new gas, so in some sense the atmosphere is the cold reservoir. But
unlike the other examples of heat engines above, we use a different set of gas every cycle.
The P V diagram is shown below.

The first, second, third, fourth bullet steps correspond to AB, BC, CD+DE, and EB+BA
respectively. Since BC and DE are adiabats (PE VBγ = PD VAγ and PC VAγ = PB VBγ ), the net work is

PD VA − PE VB PB VB − PC VA
I
W = p dV = + .
γ−1 γ−1
The heat from the fuel, Q, can be found with the internal energy change from C to D:
CV (PD − PC )VA
Q = CV n(TD − TC ) = VA (PD − PC ) = .
R γ−1

Thus the efficiency can be found with ϵ = W/Q and PE = PD k −γ , PB = PC k −γ ,

W (PD − PC )VA − (PE − PB )VB

ϵ= = = 1 − k 1−γ .
Q (PD − PC )VA

For diatomic gas, γ = 7/5, so ϵ = 1 − 1/k 2/5 .

[3] Problem 5 (IZhO 2022). A quasistatic process is carried out with one mole of an ideal monatomic
gas, as a result of which its initial volume V0 = 1 m3 increases four times, and the initial pressure
P0 = 105 Pa decreases two times. For each small section of the quasistatic process, the ratio of work
to the change in internal energy is the same. Find the total work done by the gas in this process.

Solution. This problem is good practice for working with the laws of thermodynamics directly.
Let η = dW/dU be the constant ratio of work to internal energy change. By combining the results
dW = P dV , dU = (3/2)R dT , and P V = RT , and following essentially the same derivation as that
for an adiabatic process, we find  
dP 2 dV
− = 1−
P 3η V
which implies
P ∝ V (2/3η)−1 .

5
 Kevin Zhou Physics Olympiad Handouts

In other words, this is like an adiabatic process, but with a different effective value of γ. Using the
given initial and final conditions, we have η = 4/3, so that P ∝ V −1/2 . Thus,
Z Z 4V0  1/2
V0
W = P dV = P0 dV = 2P0 V0 = 2 × 105 J.
V0 V

This problem might look contrived, but “polytropic” processes where P V β is constant, for a general
value of β, are commonly considered in engineering thermodynamics.
In physics we often assume processes are adiabatic, β = γ, but in real life nothing is ever an ideal
adiabatic process. Instead, engineers parametrize this by allowing β to be general, and measuring
its value. As a simple concrete example, if the chamber containing the gas also contains some dirt,
in thermal equilibrium of the gas, that dirt contributes to the system’s heat capacities CV and CP .

01^‚
It therefore shifts the effective value of γ away from its ideal gas value.

[3] Problem 6. USAPhO 2018, problem A3.

2 Dynamic Ideal Gases

Idea 3
Problems involving ideal gases can be mechanics questions. For example, the first law of
thermodynamics becomes conservation of energy, where the energy includes the internal
energy of the gas in addition to the usual kinetic and potential energy. You may also have
to use the principles of hydrostatic equilibrium and Bernoulli’s principle from M7.

Example 3

A space station is a large cylinder of radius R0 filled with air molecules of mass m. The
cylinder spins about its axis at an angular velocity ω, and the air rotates along with it. If
the temperature T is constant inside the station, what is the ratio of the air pressure at the
center of the station to the pressure at the rim?

Solution
If we work in a frame rotating with the space station, this is a fluid statics problem. By
considering force balance on a thin parcel of air of radial thickness dr and area A, we have

A dP = ρgeff A dr

where geff = ω 2 r is the centrifugal acceleration. Applying the ideal gas law,
dP mgeff
= dr
P kB T
which integrates to give
P (r = 0) 2 2
= e−mω r /2kB T .
P (r = R0 )

[5] Problem 7. In this problem we’ll make a simple model for the atmosphere.

6
Kevin Zhou Physics Olympiad Handouts

(a) Assume the atmosphere to be an ideal gas at constant temperature T in mechanical equilibrium,
with gas molecules of mass m. Show that the pressure depends on height as
P (h) = P0 e−mgh/kB T
by demanding that small parcels of gas be in mechanical equilibrium.
(b) The assumption of constant temperature is not very accurate. Sunlight warms air near the
ground, causing large parcels of it to slowly rise; simultaneously other parcels of air slowly
fall. This results in a well-mixed atmosphere and, since heat conduction in air is poor, the
rising and falling processes are approximately adiabatic, not isothermal. Assuming the air
molecules are diatomic with mass m, show that the temperature varies linearly with height.
Does the atmosphere get colder or hotter with increasing height?
(c) Estimate the rate of temperature change with height numerically; is your result reasonable?
(d) Now ignoring the mixing effect of the sun, argue that an atmosphere with a temperature
gradient or larger or smaller magnitude than the result you found in part (c) will be unstable
or stable against spontaneous convection, respectively. (Hint: see idea 4.)
(e) ⋆ More generally, one might wonder how the total energy of the atmosphere, summed over
all molecules, is divided into kinetic (i.e. thermal) and potential (i.e. gravitational) energy.
Show that for any configuration in mechanical equilibrium (i.e. not necessarily adiabatic or
isothermal), Egrav /Ekin has the same value, and find this value.
When “thermal inversion” occurs, the temperature gradient has the opposite sign to the natural
one you found in part (b), causing the atmosphere to be very stable against convection. Such
events can cause very high air pollution in cities, since the pollutants can’t escape. For more about
atmospheric physics, see chapter 37 of Blundell.
Solution. (a) The ideal gas law becomes P V = N kB T , and the density is ρ = N m/V , so
P m = ρkB T . By considering forces on a parcel of height dh and cross sectional area A, we see
A dP = −ρgA dh
which implies
dP mg
=− dh.
P kB T
Integrating yields the desired result.
(b) Since the expansion is adiabatic, T γ P 1−γ is constant, so T ∝ P 1−1/γ . This implies
dT dP
= (1 − 1/γ).
T P
Since the motion of the parcels of air is slow, hydrostatic equilibrium remains approximately
true. Using the equation derived in part (a) gives
dT mg
=− (1 − 1/γ) dh
T kB T
which simply rearranges to
dT mg
=− (1 − 1/γ)
dh kB
which is a linear decrease as desired.

7
Kevin Zhou Physics Olympiad Handouts

(c) We have that

dT mg 2 2 (30 g/mol)(9.8 m/s2 )

=− ≈− ≈ −10 K/km
dh kB 7 7 (8.314 kgm2/s−2 K−2 mol−1 )

which is reasonable. One reason it’s a bit unrealistically high is because air typically contains
water vapor, which increases the heat capacity.

(d) Suppose the temperature gradient is smaller in magnitude than the gradient we found above.
Consider a packet of air that is perturbed and moves upward. As it moves upward, it expands
adiabatically, lowering its temperature; since the existing temperature gradient is less than
in the well-mixed adiabatic atmosphere, the packet will end up colder than its surroundings.
However, it is also at the same pressure because mechanical equilibrium is quickly reached,
so since P ∝ ρT , the density is higher and it falls back down. Hence the situation is stable.
The same reasoning occurs in reverse for a larger temperature gradient.

(e) In mechanical equilibrium, the pressure differential should balance the gravitational pull.

dP = −ρg dh
R
The thermal energy for an ideal Rgas is Ekin = βkB T dN where β = CV /R. The potential
energy of the molecule is Egrav = gh dM , where N represents number of molecules and M as
the cumulative mass of the atmosphere (so dM = m dN , and dM = ρA dh). The differential
forms are
dEkin = βkB T dM/m, dEg = gh dM.
Since ρ = mP/kB T from ideal gas law and substituting in dM = ρA dh, we have

dEkin = βP A dh dEg = ρghA dh

We can find the total kinetic energy by integrating by parts. Let the atmosphere end at height
H (where we could also take H → ∞).
Z H  Z H 
Ekin = βA P dh = β AP (H)H − 0 − hA dP .
0 0

Since at the top of the atmosphere, it supports nothing, P (H) = 0. We also have dP = −ρg dh
for mechanical equilibrium:
Z H
Ekin = β ρghA dh = βEgrav .
0

Thus, we conclude that

Egrav R
= .
Ekin CV
For monatomic gases, this ratio is 2/3, while for diatomic gases like air, it’s 2/5.
By the way, this result is actually a corollary of the virial theorem, which we met in M6. The
virial theorem states that for particles interacting by a power law potential V ∝ rn in long-
term mechanical equilibrium, we have ⟨K⟩ = n2 ⟨V ⟩ on average. In this case, the particles are
the gas molecules and the Earth, which interact with n = 1, so that we expect ⟨V ⟩/⟨K⟩ = 2.

8
 Kevin Zhou Physics Olympiad Handouts

Here, ⟨V ⟩ corresponds to what we’ve called Egrav , but we need to be careful with ⟨K⟩. The
proof of the virial theorem doesn’t account for the energy of any degrees of freedom that aren’t
acted on by the force, which here includes the internal degrees of freedom of the gas, and
the horizontal translational degrees of freedom. We should therefore single out the vertical
translational kinetic energy, ⟨K⟩ = Ekin (R/2CV ), giving the expected result. (We also should
add on the vertical kinetic energy of the Earth, but this is negligible.) Thus, you could also

01mƒ
have solved this problem in one step using the virial theorem, which illustrates its power.

[3] Problem 8. USAPhO 1997, problem B2.

Remark
A Foehn is a hot, dry wind that comes down from a mountain range. This occurs in three
steps. First, warm air rises adiabatically up the opposite side of the mountain range. As the
air rises, it cools, causing the water vapor to condense and fall as rain. The now dry air then
falls adiabatically down the mountain range. Since the heat capacity is now lower, the falling
air heats up more than the rising air cooled down, becoming hot and dry at the bottom.

Idea 4
Consider an ideal gas in a container. In simple heat engine problems, we assume the gas stays
in equilibrium, meaning that it has a single, well-defined pressure and temperature through-
out. But in almost all real-world applications, there will be some deviations from equilibrium.

For example, suppose you started to heat the bottom of the container. Then the gas would
no longer be in thermal equilibrium, because it doesn’t have a uniform temperature, and
if the heating is sufficiently sudden, it wouldn’t be in mechanical equilibrium, because
it wouldn’t have a uniform pressure either. For a human-scale container, mechanical
equilibrium is usually reestablished quickly, by a readjustment of the density. Thermal
equilibrium is reestablished on a longer timescale, as energy spreads out through heat transfer.

In this simple example, we were able to talk about the temperature of individual parts of
the gas, even though the gas as a whole wasn’t in thermal equilibrium. That’s because each
piece of the gas is in thermal equilibrium with itself, so temperature can be defined locally.
In more violent situations, even that might not be possible.

In general, conservation laws are quite useful for nonequilibrium problems, because following
the detailed dynamics may be impossible.

Example 4

A thermally insulated chamber contains a vacuum; it is connected to the outside by a small

valve. The valve is opened until the air inside the chamber reaches atmospheric pressure, then
closed. The temperature of the air outside the chamber is T0 . Treating the air as diatomic,
what is the temperature T of the air inside the chamber?

9
 Kevin Zhou Physics Olympiad Handouts

Solution
Let the chamber have a volume V , and let the atmospheric pressure be p0 . As our system,
consider the set of all air that eventually makes it inside the chamber, and suppose this air
has volume V0 before it enters the chamber. The work done on this air by the entire rest of
the atmosphere, as it enters the chamber, is p0 V0 . The final internal energy of the air is
5 7
E = nRT0 + p0 V0 = nRT0 .
2 2
On the other hand, we also have E = nCV T = (5/2)nRT , which gives
7
T = T0 .
5
At that point, the flow stops because the pressure is equalized, even though the temperature
isn’t. This is an example of mechanical equilibrium being attained before thermal equilibrium.
(In the long run, the temperature will equality too, by heat transfer through the walls.)

Where does the extra thermal energy of the gas come from? As we just said, it’s from the air
behind it pushing it into the chamber. But on a slightly deeper level, the energy is ultimately
gravitational: the entire atmosphere shrinks down toward the Earth a bit once the volume
V0 of air is removed from it, and this decrease in gravitational potential energy is exactly the
same as the increase in thermal energy of this system.

[2] Problem 9. Consider two cylinders A and B of equal volume V , connected by a thin valve. The
cylinders are thermally insulated from the environment, but conduct heat well between each other.
Cylinder A is equipped with a piston that can compress the gas inside. Initially, the valve is sealed,
cylinder A contains an ideal monatomic gas at temperature T , and cylinder B contains a vacuum.
Now suppose the valve is opened, and the piston is slowly pushed inward so that the pressure in
cylinder A remains constant, until it cannot go any further. What is the final temperature of the
gas, and what is the final volume of cylinder A?

Solution. Suppose the pressure in cylinder A is always P . Suppose the final volume of A is V ′ ,
and suppose the final temperature is T ′ . Then, the energy of the system before the compression
is Ei = 32 n0 RT where n0 is the total number of moles of gas. The energy after the compression is
Ef = 32 n0 RT ′ . We have that Ef − Ei = P (V − V ′ ). The piston stops once the pressure in B is
equal to P , so P V ′ = nRT ′ and P V = (n0 − n)RT ′ , so P (V + V ′ ) = n0 RT ′ . Therefore,
3
2n0 RT = 2P V = P (V − V ′ ) + P (V + V ′ ) = n0 R(T ′ − T ) + n0 RT ′ ,
2
so
3 7
2T = (T ′ − T ) + T ′ =⇒ 4T = 3T ′ − 3T + 2T ′ =⇒ T ′ = T .
2 5
We see that
3 4
2P V ′ = P (V + V ′ ) − P (V − V ′ ) = n0 RT ′ − n0 R(T ′ − T ) = n0 RT,
2 5
′
so 2 VV n0 RT = 54 n0 RT , so V ′ = 2V /5 .

10
 Kevin Zhou Physics Olympiad Handouts

[4] Problem 10. Consider a cylinder of gas with cross-sectional area A and volume V . Assume all
surfaces are frictionless and thermally insulating. A piston of mass m is placed snugly on top, and
the entire setup is inside an atmosphere with pressure Patm .

(a) First suppose the system is in equilibrium, so that the pressure of the gas inside is Patm +
mg/A. The piston is then given a slight downward displacement. Find the frequency of small
oscillations by assuming the ideal gas law always holds for the gas as a whole. This setup is
known as the Ruchardt experiment, and can be used to determine γ.

(b) Under what circumstances is the result of part (a) a good approximation?

(c) Now suppose that instead, the piston is initially suspended from a thread carrying tension
mg, so that the pressure of the gas is just Patm . Suddenly, the thread is cut. The piston falls
down the cylinder and bounces up and down several times before eventually coming to rest.
Explain why the equation P V γ = const cannot be used to determine the final state.

(d) Assuming the final downward displacement of the piston is d, find the net work done on the
gas, assuming that all of the energy dissipated goes into the gas.

(e) Find d and the final temperature of the gas, assuming the gas is monatomic.

Solution. (a) When the mass has a displacement of x, let the gas pressure be Px . Then

P V γ = Px (V − Ax)γ

so  γ  
V Ax
Px = P ≈p 1+γ .
V − Ax V
Thus, the force on the mass is
Ax
F = −(Px − P )A ≈ −P γ A
V
so r
γP A2 γP A2
ẍ ≈ − x, ω= .
mV mV
(b) First, we’ve treated the gas as always having the pressure of a static ideal gas. That means
the piston needs to move slowly enough for the gas to have time to adjust to this pressure,
i.e. the piston should always be moving much slower than the speed of sound in the gas. This
happens automatically, as long as the amplitude is small enough.
Second, as mentioned in M6, even in the case where the piston is moving arbitrarily slowly, it
has effective extra inertia because it needs to move the gas in front of it out of the way. That
is, in the language of M4, the Lagrangian for the system should contain both the kinetic
energy of the piston and the gas, and the latter contributes an effective extra inertia, lowering
the frequency. This extra term is negligible as long as the density of the gas is much lower
than the density of the piston.
For reasonable experimental setups, both of these conditions are easily satisfied. In real life,
the hardest part of getting this to work is probably making the piston oscillate with low
friction, while still being airtight.

11
 Kevin Zhou Physics Olympiad Handouts

(c) It depends on how you derive P V γ = const. One way to derive it, which you saw in problem 1,
is to write down basic results like dW = −P dV , dU = nCV dT , and P V = nRT , and combine
them. What could possibly go wrong with that? The problem is that during this process, the
gas does not even have a uniform pressure or temperature; there is no such thing as a single
P or T .
Another way to derive it, as we’ll show in T2, is to argue that the entropy is constant. (Then
we can ignore all the complicated stuff that happens in the middle, because entropy is a state
function.) The entropy turns out to be a function of P V γ , which implies P V γ is constant. The
reason this argument fails is because entropy isn’t constant. As the piston’s motion damps
out, kinetic energy is dissipated to heat, increasing the entropy.

(d) The work done on the gas is

W = mgd + Patm Ad
where we counted the work done by the weight of the piston, and the atmospheric work.
How does this energy get into the gas? As the piston bounces up and down, it creates sound
waves in the gas. These eventually fall apart and turn into ordinary thermal energy, i.e. random
motion of the gas molecules; this is what damps the motion of the piston even though it doesn’t
have any friction. However, sound waves are excited in the atmosphere during this processes
as well, and this also carries energy away. We expect most of the energy to end up in the gas
if the pressure variations in the gas are larger, e.g. with a very heavy weight. We assume that
all of it ends up in the gas just for simplicity, since there’s no reasonable way to compute the
relative contributions.

(e) By conservation of energy,

3
W = nR∆T.
2
On the other hand, by the ideal gas law,
∆(P V ) = nR∆T
which means we have
3  mg  
mgd + Patm Ad = Patm + (V − Ad) − P V .
2 A
Solving for d, we have
3 mgV
d= .
5 A(mg + Patm A)
Interestingly, this approaches only 3/5 of the total height when mg goes to infinity.
[3] Problem 11 (Russia 2008). A cylinder with a metal bottom and insulating walls is underneath a
thin massive metal piston located at a height h, which is much smaller than the cylinder diameter.
A resistor of resistance r is placed inside and connected to an electric circuit with an emf E.

12
 Kevin Zhou Physics Olympiad Handouts

The circuit is connected to the piston and cylinder bottom with light flexible wires. Initially, the
switch is open, the cylinder is filled with helium at a pressure p ≫ ϵ0 E 2 /h2 , which you can treat as
a monatomic ideal gas with a dielectric constant of 1. The system is thermally insulated, placed in
vacuum, and at thermal and mechanical equilibrium. Then the switch K is closed. Find the height
H of the piston after a long time.

Solution. Several things happen at once. Energy is dissipated in the resistor, causing the gas to
warm up and increase in pressure. At the same time, charge accumulates on the top and bottom
plates, which form a parallel plate capacitor, causing them to attract each other. And when the
piston moves, energy is exchanged between the gas and the capacitor. Keeping track of the detailed
time evolution of the gas, piston, and RC circuit would be very complicated, but since we only care
about the final state, we can use energy conservation instead.
The total energy added to the system is the work done by the battery, Eq, where q is the final
charge on the capacitor plates, so energy conservation gives

3 3 q2
p1 V1 + mgh + Eq = p2 V2 + mgH + .
2 2 2C
We also know that in the final state,
ϵ0 AE
q = CE =
H
where A is the surface area of the top and bottom plates. The attractive force between the plates is

q2 ϵ0 AE 2
F = =
2ϵ0 A 2H 2
which means the final pressure is

F ϵ0 E 2
p2 = p1 + = p1 + .
A 2H 2
Of course, we also have V1 = Ah, V2 = AH, p1 = p, and force balance in the initial state implies
mgh = p1 V1 . Carefully plugging all of this in and writing everything in terms of H/h and the small
ratio ϵ0 E 2 /h2 p yields the result
 2
H H ϵ0 E 2
− =−
h h 10h2 p
and solving the quadratic gives
s !
1 1 ϵ0 E 2 ϵ0 E 2
H=h + − ≈h− .
2 4 10h2 p 10hp

[4] Problem 12 (Cahn). A long, cylindrical tank of length L and radius R is placed on a carriage
that can slide without friction on rails. The mass of the empty tank and carriage is M . Initially,
the tank is filled with an ideal gas of total mass m ≪ M at pressure P0 and temperature T0 . The
left end of the tank is heated to a fixed temperature T1 , while the right end of the tank has its
temperature fixed at T0 , where T1 − T0 ≪ T0 .
In this problem, you need only work to first order in (T1 −T0 )/T0 . Suppose that the temperatures
have been maintained for long enough for the gas to enter a steady state.

(a) Argue that the temperature T (x) of the gas in the tank is a linear function of position.

13
Kevin Zhou Physics Olympiad Handouts

(b) Find the density of the gas in the tank as a function of position.

(c) Find the distance the carriage has moved.

(d) In order for the carriage to have moved, a horizontal force had to have acted on it. Where
did this force come from?

Solution. (a) At steady state, the temperature of the gas at each location will be constant. Thus
considering a thin cross section of thickness dx perpendicular to the cylinder’s axis, the power
going through it is P = κAdT /dx where κ and A = πR2 are thermal conductivity and cross
sectional area, respectively. In steady state, no net heat will enter/exit the gas, so P is
constant, and with T1 − T0 ≪ T0 , κ is constant. Thus dT /dx is constant (giving a linear
T (x) = T1 − x(T1 − T0 )/L) with x = 0 and x = L being the left and right ends respectively.

(b) The pressure is constant since horizontal pressure gradients would cause movements, and
ρ(x) = RTµP(x) , and a Taylor expansion gives us
 
µP x(T1 − T0 )
ρ(x) ≈ 1+ .
RT1 LT1

Thus density is a linear function too, so the average density is ρ0 = ρ(L/2). Thus ρ(L/2)πR2 L =
m. From our ρ(x) equation and taking first order approximations (uniform temperature), we
get that
dρ µP T1 − T0 ρ0 (T1 − T0 )
= ≈ .
dx RT1 LT1 T0 L
Thus  
m T1 − T0 x − L/2
ρ(x) = 1+
πR2 L T0 L

(c) The center of mass of the gas is at position

L
1 L
Z  
T1 − T0 x − L/2
Z
1 2
xcm = xρ(x)πR dx = 1+ xdx
m 0 L 0 T0 L
 
1 T1 − T0
xcm = L +
2 12T0
Thus it was displaced to a distance ∆x = L(T1 − T0 )/(12T0 ) to the right with respect to the
carriage.
Since there’s no net force on the system, the center of mass of the entire system must stay
stationary. Thus the displacement of the carriage D satisfies

M D + m(D + ∆x) = 0
 
m T1 − T0
D = −L
M +m 12T0
The lowest order approximation yields
m T1 − T0
D = −L .
M 12T0

14
 Kevin Zhou Physics Olympiad Handouts

(d) When a gas molecule bounces off a hotter wall, it picks up kinetic energy in the collision;
in other words, it bounces off faster than it came in. (This is the microscopic way heat is
transferred through conduction.) That means that the pressure the gas molecules exert on a
hot wall is actually greater than the pressure of the gas itself. (Similarly, the pressure on a
cold wall is lower.) So even though the pressure of the gas was initially uniform, an unbalanced

01@
force was exerted on the walls, pushing the entire carriages in the direction of the hot wall.

[4] Problem 13. APhO 2010, problem 3B. A mathematical problem on a collapsing bubble.

Remark
Students often get stuck on problem 14, which is about a chimney above a furnace, because
they forget that the base of the furnace is open to the air, and so its pressure is equal to the
atmospheric pressure. Indeed, in real life it is very hard to produce air pressures substantially
above atmospheric pressure. You need to either tightly seal a container (which applies to the
engines of the problems above, or to pressure cookers), or make the air move very quickly
(which occurs in jet engines, covered in T3, or in specialized “blast” furnaces).

[4] Problem 14. 01@ IPhO 2010, problem 2. A neat, tricky problem about how chimneys work.

3 Statistical Mechanics
There are fundamentally two approaches to describing systems of many interacting particles: bottom-
up and top-down. In the top-down approach of thermodynamics, we try to roughly describe the
behavior of the whole system in terms of a few macroscopically measurable observables, such as
pressure and temperature, and hope this is enough information to extract what we want. In
the bottom-up approach, we start by analyzing the behavior of individual molecules, governed by
Newtonian mechanics. Of course, we can’t do this exactly, but it turns out to be possible to make
probabilistic statements about individual molecules. This is the approach of statistical mechanics.

Idea 5: Boltzmann Distribution

The probability distribution for the states of a particle in a system of temperature T is
proportional to e−E/kB T . Specifically:

• For quantum systems, where the energy levels are discrete, the probability of being in a
state n with energy En is proportional to e−En /kB T .

• For a single classical particle, the state is instead specified by (x, p), the position and the
momentum, and the probability density in this space, called phase space, is proportional
to e−E(x,p)/kB T .

It isn’t possible to derive the Boltzmann distribution from anything we’ve already covered,
but you’ll see in T2 how it emerges from a simpler postulate.

Example 5: Isothermal Atmosphere

Do problem 7 using statistical mechanics.

15
 Kevin Zhou Physics Olympiad Handouts

Solution
The energy is
p2
E(x, p) = mgz + .
2m
The probability distribution for height z is found by integrating over all the other quantities,
Z Z Z Z Z Z
−E(x,p)/kB T −mgz/kB T 2
p(z) ∝ dx dy d p e 3
=e dx dy d3 p e−p /2m .

However, the remaining integrals are just constants independent of z, so we just get

p(z) ∝ e−mgz/kB T .

Since the particles are assumed independent (since we have an ideal gas), the probability for
a particle to be at a point is proportional to the density of gas at that point. We see the
density falls exponentially with height, so by the ideal gas law, the pressure does too.

[1] Problem 15. Do example 3 using statistical mechanics.

Solution. Working again in the rotating frame of reference, integrating the centrifugal force gives
a potential energy U (r) = −mω 2 r2 /2. Therefore, the density of gas is
2 r 2 /2k
ρ(r) ∝ p(r) ∝ emω BT
.

At fixed temperature, the pressure is proportional to the density by the ideal gas law, so

P (r = 0) 2 2
= e−mω r /2kB T .
P (r = R0 )

exactly as found earlier.

[3] Problem 16. Some basic computations for ideal gases.

(a) For an ideal gas in a box, show that the probability distribution of speeds obeys
2 /2k T
p(v) ∝ v 2 e−mv B

at any point in the box, regardless of the shape of the box.

(b) Compute the most probable speed, i.e. the location of the peak of this probability distribution.

(c) Show that the average kinetic energy is ⟨mv 2 /2⟩ = 3kB T /2. This is a special case of the
equipartition theorem, shown below. (Hint: you will have to do a somewhat tricky integral.
See the example below and the examples in P1 for guidance.)

Solution. (a) Note that the probability that a particle has velocity (vx , vy , vz ) is given by
− 2km T (vx2 +vy2 +vz2 )
f (vx , vy , vz )dvx dvy dvz where f (vx , vy , vz ) ∝ e B . Then, we see that
− 2km T v 2
p(v)dv ∝ (4πv 2 dv)e B ,
2 /2k T
so p(v) ∝ v 2 e−mv B .

16
Kevin Zhou Physics Olympiad Handouts

(b) We set p′ (v) = 0, so

r
−mv 2 /2k −mv 2 /2k 2kB T
v 2 (−mv/kB T )e BT
= −2ve BT
=⇒ v = .
m

(c) Using the result of part (a),

R∞ 2 R∞ 2
1
mv 2 v 2 e−mv /2kB T dv kB T 0 x4 e−x /2 dx


2 0 R2∞
⟨mv /2⟩ = 2 −mv 2 /2kB T dv
= R∞
0 v e
2 0 x2 e−x2 /2 dx

where we nondimensionalized the integral. To evaluate it, note that by integration by parts,
Z ∞ Z ∞
−x2 /2 2
3
(x )(xe dx) = 3 x2 e−x /2 dx.
0 0

Then the ratio of integrals is just 3, giving

3
⟨mv 2 /2⟩ = kB T
2
as desired.

Remark: Deriving the Maxwell Velocity Distribution

Statistical mechanics implies that the velocity distribution in an ideal gas is

2 /k
p(v) ∝ e−mv BT

which is a three-dimensional Gaussian. This result was first derived by Maxwell, long before
statistical mechanics was understood, using an ingenious argument.

Suppose the ideal gas is inside a rectangular box, so that collisions with its left and right sides
determine vx , the front and back sides determine vy , and the top and bottom sides determine
vz . The distributions of velocities in each direction should therefore be independent, and
identical by rotational symmetry, so that we can write

p(v) = f (vx )f (vy )f (vz )

for some function f . Moreover, by rotational symmetry, p(v) can only depend on v 2 . Taking
the logarithm of both sides and defining g = log f , we have

log p = g(vx ) + g(vy ) + g(vz )

and the right-hand side only depends on v 2 . This is only possible if g(x) = −αx2 for a
2
constant α, which yields p(v) ∝ e−αv . The value of α can be determined by demanding the
pressure match the ideal gas law (see example 7).

Modern statistical mechanics gives us another slick way to derive the distribution. The idea
is that in thermodynamic equilibrium, time reversal symmetry implies every reaction is in
“detailed balance”. Specifically, suppose that when two gas molecules collide elastically, with
velocities (v1i , v2i ), they exit with velocities (v1f , v2f ). Then the rate at which this happens

17
Kevin Zhou Physics Olympiad Handouts

must be equal to the rate at which gas molecules with velocities (v1f , v2f ) collide, which
causes them to exit with velocities (v1i , v2i ). (Otherwise, you could distinguish a video of
the molecules’ dynamics from the same video played in reverse.) Therefore,

p(v1i )p(v2i ) |v1i − v2i | = p(v1f )p(v2f ) |v1f − v2f |

where the relative velocity factor is present because a larger relative velocity gives more
opportunities for collision. However, for elastic collisions, the initial and final relative
velocities are the same, so this factor just cancels out.

Now, by rotational symmetry we have p(v) = f (v 2 ) for some function f , so

2 2 2 2
f (v1i )f (v2i ) = f (v1f )f (v2f )
2 + v 2 = v 2 + v 2 . Then the above equation
Furthermore, by energy conservation we have v1i 2i 1f 2f
2
can only be satisfied if f is an exponential, f (x) ∝ e−αx , which again gives p(v) ∝ e−αv .

Idea 6: Equipartition Theorem

Suppose some degree of freedom in phase space contributes to the energy by a power law,

E ∝ pn or H ∝ xn .

Then in thermal equilibrium, there is an average energy of kB T /n in this degree of freedom.

The proof of this is just a straightforward generalization of the reasoning in problem 16(c).
Note that the proof of the equipartition theorem is entirely classical, i.e. it holds when a
system is adequately described by classical mechanics. It does a good job of describing the
heat capacities of simple gases near room temperature, as you can see here.

Here are a few important examples of the equipartition theorem.

• A mass on a spring in one dimension has
p2 kx2
E= + .
2m 2
This involves two degrees of freedom in phase space, each with n = 2, so the average
energy is 2(kB T /2) = kB T .

• In problem 7 you considered a system with a linear potential energy, E(h) = mgh. This
is a degree of freedom with n = 1, and accordingly, the average potential energy of each
molecule is kB T .

• An atom in a solid can be thought of as attached to its neighbors by springs. Since the
atom is in three dimensions, this yields three copies of the first example, and hence an
average thermal energy of 3kB T . This implies that the specific heat per atom in a solid
is 3kB (or equivalently 3R per mole), which is called the Dulong–Petit law.
Incidentally, because systems in thermal equilibrium are also in long-term mechanical equilib-
rium, the virial theorem from M6 applies, and tells us the ratio between the average kinetic

18
 Kevin Zhou Physics Olympiad Handouts

and potential energy; of course, it matches the result of the equipartition theorem. You’ll
investigate some more key examples in the problems below.

Example 6

A free particle in one dimension, H = p2 /2m, has energy levels En = n2 E0 for n = 0, 1, 2, . . .,

as we will show in X1. Show that the average energy at high temperatures is consistent with
the equipartition theorem.

Solution
2 E /k T
The probability of being in energy level n is proportional to e−n 0 B . Therefore, by
normalizing the probability distribution, the probability itself is
2
e−n E0 /kB T
pn = P∞ −m2 E0 /kB T
.
m=0 e

Therefore, the average value of the energy is

∞ P∞ 2 2
X
2 n E0 e−n E0 /kB T
⟨E⟩ = n E0 pn = n=0
P∞ −n2 E /k T .
n=0 e
0 B
n=0

At high temperatures, this quantum result should reduce to the classical result of the equipar-
tition theorem. Note that at such temperatures, the typical values of n will be very high.
Therefore, we can treat the sums over n as continuous integrals,
R∞ 2 −n2 E0 /kB T
R ∞ 2 −x2
0 R dn n E0 e x e dx
⟨E⟩ ≈ ∞ −n 2 E /k T = 0R ∞ −x2 kB T
0 dn e 0 e dx
0 B

p
where we switched to the dimensionless variable x = n2 E0 /kB T . Evaluating either of these
integrals is tricky, but we can relate them using integration by parts,
Z ∞ Z ∞
1 ∞ −x2
Z
2 2 x
x2 e−x dx = (2xe−x dx) = e dx.
0 0 2 2 0

Therefore, the ratio of integrals is 1/2, giving an average energy of kB T /2 as expected.

[3] Problem 17. The energy of a photon is E = pc.

(a) Treating the photon classically, compute the average energy of a photon moving in one
dimension, assuming it obeys the Boltzmann distribution with temperature T . Check that
the result obeys the equipartition theorem.

(b) Generalize this result to three dimensions to compute γ for a 3D photon gas.

Solution. (a) Using the Boltzmann distribution and E = pc,

P (p) ∝ e−pc/kB T .

19
 Kevin Zhou Physics Olympiad Handouts

Then the expected energy is

R∞ −pc/kB T dp
R ∞ −x
(pc)e xe dx
⟨E⟩ = ⟨pc⟩ = 0R ∞ −pc/k T = kB T R0 ∞ −x .
0 e dp 0 e dx
B

Note that the numerator can be related to the denominator using integration by parts,
Z ∞ Z ∞
−x
xe dx = e−x dx.
0 0

Then the fraction is just 1, so

⟨E⟩ = kB T.
Evidently, every degree of freedom where the energy is linear in the momentum contributes
energy kB T . This is just as we expect by the equipartition theorem, since there’s one degree
of freedom with n = 1.

(b) The reasoning is very similar, but now

P (p) ∝ e−|p|c/kB T .

The expected energy is then, using spherical coordinates,

R∞ −pc/kB T (4πp2 ) dp
R ∞ 3 −x
0R (pc)e x e dx
⟨E⟩ = ⟨pc⟩ = ∞ −pc/kB T 2
= kB T R0∞ 2 −x .
0 e (4πp ) dp 0 x e dx

Again the numerator can be related to the denominator using integration by parts,
Z ∞ Z ∞
3 −x
x e dx = 3 x2 e−x dx.
0 0

Then the fraction is 3, so ⟨pc⟩ = 3kB T , which is just what we expected since there are three
linear degrees of freedom. Hence we have γ = 4/3.

[3] Problem 18. A one-dimensional quantum harmonic oscillator has energy levels En = nE0 for
n ≥ 0, as you will show in X1.

(a) Compute the average energy of the system at temperature T .

(b) Show that at high temperatures, the average energy obeys the equipartition theorem.

(c) Show that the energy at low temperatures is instead exponentially suppressed. This is why
some modes are said to “freeze out”, so they do not contribute to CV .

The phenomenon of “freezing out” was one of the greatest puzzles of classical physics in the 19th
century, though nobody anticipated the resolution would be as strange as quantum mechanics.

Solution. (a) The average energy is

X
nE0 e−nE0 /kB T
n≥0
⟨E⟩ = X .
e−nE0 /kB T
n≥0

20
 Kevin Zhou Physics Olympiad Handouts

Let X 1
f (x) = e−nx = ,
1 − e−x
n≥0
so
X e−x
ne−nx = −f ′ (x) = .
(1 − e−x )2
n≥0
Thus,
e−E0 /kB T
⟨E⟩ = E0 .
1 − e−E0 /kB T

(b) Note that if kB T ≫ E0 , then e−E0 /kB T ≈ 1 − E0 /kB T , so plugging into the formula, we get
that ⟨E⟩ ≈ kB T , as desired.
(c) For low temperatures, we have e−E0 /kB T ≪ 1, which gives ⟨E⟩ ≈ E0 e−E0 /kB T .
[3] Problem 19. The Boltzmann distribution can be normalized to a probability distribution by
dividing by the so-called partition function Z = n e−En /kB T .
P

(a) For a quantum harmonic oscillator, compute the probability of occupancy of the ground state
(i.e. the lowest energy state) at temperature T .
(b) Do the same for the hydrogen atom, where En = −E0 /n2 for n ≥ 1. You should get a rather
strange answer. Does it make physical sense?
Solution. (a) The probability is
e−(0)E0 /kB T 1
P0 = =
Z Z
where the partition function is
1
Z=
1− e−E0 /kB T
by summing a geometric series, so the answer is
P0 = 1 − e−E0 /kB T .

(b) The problem here is that there are an infinite number of energy levels that have similar energy.
So the sum used to define Z diverges, which means that the probability of occupying any
given state is zero! This seems to be a paradox.
The resolution is that this really is the correct answer, if you have true thermal equilibrium in
an infinite volume. The electron simply has a huge volume it could wander around to, so as
the volume goes to infinity, that means the chance it’ll be near the proton falls to zero. This
can happen even if the electron begins in the ground state and the temperature is extremely
low; it just takes a long time. After some number of aeons, the random thermal radiation at
some moment will just happen to be enough to pop the electron free, and then it’ll wander
away, probably never to return.
So the answer is formally true, but it’s not realistic, because in everyday life and in physics
labs we perform experiments with finite volumes and finite times. This can be accounted for
by hand by just cutting off the sum at some high n, say
10 6
X
Z= e−En /kB T .
n=0

21
 Kevin Zhou Physics Olympiad Handouts

Such a cutoff corresponds to only allowing atomic orbitals that fit in a typical room, which is
reasonable, and using this value of Z gives reasonable results, as would any sufficiently high
but finite cutoff. For more on this point, see this paper.
[3] Problem 20. To model paramagnetism, consider a set of N independent magnetic dipoles in a
vertical magnetic field B and temperature T . Each dipole has two possible quantum states, spin
up and spin down, with energies µ · B and magnetic dipole moment µz = ±eℏ/2m.

(a) Defining the magnetization M as the total magnetic moment, find the average magnetization
of the magnet as a function of T .

(b) Show that M ∝ 1/T in the limit of high T . This result is called Curie’s law.

Solution. (a) Since the dipoles are completely independent, we can treat them separately, giving
eℏB eℏB
eℏ 2mkB T eℏ − 2mkB T
2m e − 2m e N eℏ eℏB
M =N eℏB eℏB = tanh .
e 2mkB T
+e
− 2mk
B T 2m 2mkB T

(b) Note that tanh x ≈ x for x ≪ 1, so M ∝ 1/T in limit of high T as desired.

Remark
In principle, all of our results about statistical mechanics can be derived by following the
microscopic dynamics, e.g. the collisions of atoms with each other for a gas. It’s just very
hard to do so in practice.

However, you can still get a bit of insight in simple cases. For example, consider a
one-dimensional ideal gas in a uniform gravitational field, E = p2 /2m + mgh. Since the
kinetic energy is a quadratic degree of freedom, and the potential energy is a linear degree of
freedom, the average potential energy must be twice the average kinetic energy. Furthermore,
this should be true no matter how rarely collisions happen; less frequent collisions simply
mean it takes longer to get to thermal equilibrium, without changing what that equilibrium is.

Therefore, we have the concrete, classical mechanical prediction that a mass bouncing
elastically on the ground, suffering no collisions at all, will have an average potential energy
equal to twice its average kinetic energy, which you can verify directly from Newton’s laws.

Here’s a more subtle puzzle. In our analysis of the isothermal atmosphere, we found that
the distribution of speeds is the same everywhere. But if you look at any one particle, it
will clearly pick up speed as it falls down. How is this self-consistent? Wouldn’t this tend to
increase the temperature at lower heights?

Again, it’s easiest to think about this situation in the limit of a very thin gas, where collisions
are rare. Indeed, each particle that starts at the top of the atmosphere will pick up a lot
of speed as it falls down. But why are there any particles at the top at all? Because at
the bottom of the atmosphere, which is much more dense, particles occasionally get a lot
of energy through collisions, which propels them to the top. In equilibrium, the particles
going down do arrive at the bottom with unusually high speed, but they’re balanced out

22
 Kevin Zhou Physics Olympiad Handouts

by the unusually fast particles going the other way. Keeping track of the details here can
be quite complex, especially when collisions are frequent, but miraculously the Boltzmann
distribution takes care of it all for us!

4 Kinetic Theory
Idea 7
Kinetic theory is the branch of statistical mechanics that focuses on the motion of individual
molecules in a gas. It can get quite mathematically involved, since one must consider the
effects of molecular collisions, which is why the subject is usually reserved for graduate
school. However, in some simple situations, collisions can be neglected, effectively making
the molecules independent.

Example 7: Ideal Gas Law

Derive the ideal gas law using kinetic theory.

Solution
Without loss of generality, we consider a cubical container of gas of side length L and N gas
molecules. Consider the pressure exerted on a wall aligned with the yz plane. A given gas
molecule will hit this wall with period 2L/vx , transferring a momentum of 2px . Then the
pressure on the wall is
F 1 N ⟨px vx ⟩
P = = 2 .
A L L
This can be written as
N N ⟨p · v⟩ N N kB T
P = ⟨px vx ⟩ = = ⟨mv 2 ⟩ =
V V 3 3V V
where we used rotational symmetry and the equipartition theorem.

Of course, this calculation was doable because we neglected interactions between different
gas molecules. However, this doesn’t matter as much as one might think. We assumed
molecules can cross the whole container without colliding, but if they do, the derivation still
basically works because they just transfer their momentum to another molecule. As long as
the molecules are moving freely most of the time, the answer above is approximately right.

In T2, we’ll see how an attractive interaction lowers the pressure. In T3, we’ll see that when
the gas gets very dense, this leads to a dramatic effect: the gas condenses into a liquid.

[2] Problem 21. Use kinetic theory to find the pressure of a photon gas (i.e. an ideal gas of massless
particles, which satisfy E = pc) in terms of its total internal energy U and volume V . Use this
result to find γ for a photon gas, where γ is defined so that P V γ is constant in an adiabatic process.

23
Kevin Zhou Physics Olympiad Handouts

Solution. By the exact same reasoning as in the example,

N 1N
P = ⟨px vx ⟩ = ⟨p · v⟩.
V 3V
However, for photons we have pv = pc = E, so
1N U
P = ⟨E⟩ = .
3V 3V
In other words, the pressure is always one third of the energy density. Notice that the number of
photons cancelled out, as it must have because it wasn’t specified anywhere in the problem. As
you’ll see in T2, the number of photons is determined by the volume and temperature; it isn’t an
independent parameter like it is for the ideal gas. As you heat or cool a photon gas, photons can
be freely emitted and absorbed by the walls, changing the total number to reach equilibrium.
Starting from U = 3P V and taking the differential of both sides, in an adiabatic process we have

dU = d̄W = −P dV, dU = 3(P dV + V dP ).

Simplifying gives
4 dV dP
− =
3 V P
which is equivalent to P V 4/3 being constant, so γ = 4/3.

Example 8

An astronaut of mass M stranded in outer space makes a primitive rocket as follows. They
take a piece of dry ice of mass m ≪ M and molar mass µ and put it in a long, thin glass held
at temperature T . The dry ice sublimates, and the astronaut points the opening of the glass
in the opposite of the direction they want to go. Estimate, within an order of magnitude,
how fast the astronaut is going after all the dry ice is gone.

Solution
Let m′ be the mass of a molecule of dry ice. By the equipartition theorem, the molecules
have speed s
r r
kT N kT RT
v∼ ′
= ′
= .
m Nm µ
The molecules exit the glass traveling more or less in the same direction, so by momentum
conservation the final speed of the astronaut is on the order of
s
m m RT
v= .
M M µ

We have implicitly assumed that the glass is long enough that the molecules get into thermal
equilibrium with the glass before they leave it. For a short glass, the answer would instead
depend on the typical speeds of molecules at the moment they sublimate.

24
 Kevin Zhou Physics Olympiad Handouts

Example 9

A container of volume V is filled with a gas of uniform temperature T , and placed in a

vacuum. If a small hole is punched in the container and gas slowly leaks out, how does the
temperature in the container change over time?

Solution
The mean free path λ is the average distance a gas molecule moves before it hits another
one. Let d be the width of the hole. There are two important limiting cases.

If λ ≪ d, then molecules can’t make it out of the hole without undergoing many collisions.
Thus, if the gas has some average flow velocity, its molecules will be carried along with it. We
can thus treat the gas like a continuous fluid, and the logic of example 4 applies. The gas inside
the container does P dV work on the gas exiting and hence loses energy. Thus, the remaining
gas cools down, so the temperature of the container decreases over time. (Doing this problem
quantitatively requires the analogue of Bernoulli’s principle for gases, which we derive in T3.)

If d ≪ λ, then molecules fly out of the hole without encountering any others at all; this is the
regime of “effusion”. Therefore, we shouldn’t think in terms of an average gas pressure or
flow velocity, but rather just consider each molecule on its own. An individual molecule has
a greater chance of escaping through the hole if it’s moving faster. Thus, the molecules that
come out will on average be higher energy than the ones that stay in the container. So just
as in the opposite limit, the remaining gas cools down. (This limit is treated quantitatively
in problem 2.24 of Wang and Ricardo, volume 2.)

Remark
The previous example implies that a gas of uniform temperature can turn into a hot gas
and a cold gas without needing any external work, which seems to directly contradict the
second law of thermodynamics! To phrase it another way, you can use the free temperature
difference to run a heat engine, which seemingly allows perpetual motion.

Of course, the resolution is that the total entropy of the gas is still going up. The kinetic
energy is getting less spread out, but the positions of the atoms are getting more spread out,
since some can now be outside the container. Thus, the “perpetual motion” only works as
long as the gas is still leaking out of the container; it stops once the density becomes uniform.
The lesson here is that, while most setups extract work from a temperature difference, more
generally you can extract it from any potential increase in entropy; in this case, it comes
from an initial difference in densities.

[2] Problem 22 (Kalda). Natural uranium consists of mainly two isotopes, 238 U and 235 U, and the
relative concentration of the latter is 0.7%. Uranium is enriched by a multi-stage process, where
at each stage, evaporated UF6 is led through a porous wall. The porous wall is a thin film with
microscopic holes, much smaller than the mean free path of the molecules, but larger than the
dimension of the molecules. How many stages are needed to increase the 235 U content to 1.4%?
The molar mass of fluoride is 19 g/mol.

25
 Kevin Zhou Physics Olympiad Handouts

Solution. Say the two different versions of UF6 have molecular masses M1 , M2 . The key idea here
is that the average kinetic energy of each one is the same (namely 32 kB T ), so 12 M1 v12 = 12 M2 v22 . This
p
means
p that v1 /v2 = M2 /M1 , so the rate that M1 leaves divided by the rate that M2 leaves is
M2 /M1 . Thus, the rate at which 238 leaves divided by the rate that 235 leaves is
p
r = (235 + 6 · 19)/(238 + 6 · 19) = 0.9957 . . . .
Throughout the process, almost all of the uranium is 238 U, so the proportion of it falls by approxi-
mately r per stage. Therefore, the number of stages N obeys
0.9957N ≈ 1/2
which has solution N ≈ 160.
This problem assumed a large mean free path, in which case the molecules escaped by effusion.
In the opposite limit, where the mean free path is small, we would instead describe the gas using
fluid flow equations, such as Bernoulli’s principle. However, in that limit the two isotopes would
just get carried along with the flow in the same way, leading to no enrichment.
[2] Problem 23 (Cahn). A small vessel with a tiny hole of diameter d is placed inside a chamber,
where the pressure is so low that the mean free path is λ ≫ d. The temperature of the gas in the
chamber is T0 , and the pressure is P0 . The temperature in the vessel is kept at a constant T1 = 4T0 .

What is the pressure inside the vessel when steady state is reached?
Solution. Since the chamber is much larger than the vessel, gas entering the vessel and leaving
from the vessel won’t affect the pressure or temperature of the chamber. Now, it is very easy to fall
into the following trap. One may argue that steady state is reached when the pressures in the vessel
and chamber are the same. After all, isn’t that what happens when two gases come in equilibrium?
The problem is that we are in the effusive regime λ ≫ d. Thus, if a particle is near the hole
and heading towards it, it’s likely to just pass through without hitting anything at all; it doesn’t
“know” that it’s entered a region of higher pressure, so there’s no reason for the pressures to equalize.
Instead, equilibrium just occurs whenever the rate of particles going from the chamber to the vessel
is the same as the rate of particles going from the vessel to the chamber. This rate is proportional
to the number density times the average speed, so in equilibrium,
n0 ⟨v0 ⟩ = n1 ⟨v1 ⟩.
√
We also know that ⟨vi ⟩ ∝ Ti by equipartition. Therefore, n1 = n0 /2, and the ideal gas law says
p ∝ nT , so p1 = 2p0 .
If you think the step where we asserted ni ⟨vi ⟩ matched on both sides was not rigorous enough,
here’s a more formal derivation. (The result of this derivation will also be used in T2.)

26
 Kevin Zhou Physics Olympiad Handouts

Lemma. Given a box with an (ideal) gas of particle mass m, temperature T , and number density
n (number of particles per unit volume), the number of particles leaving a small hole of area A per
unit time is r
nA nA 8kB T
⟨v⟩ =
4 4 πm
√
assuming that A ≪ λ.

Proof. Let f (v) be the distribution of velocities of the particles. What this means is that the
probability that a particle has velocity in [vx , vx + dvx ] × [vy , vy + dvy ] × [vz , vz + dvz ] is
q 
f vx2 + vy2 + vz2 dvx dvy dvz .

Set up spherical coordinates with origin at the hole. We will now count the number of particles that
hit the hole in a time dt using a funny double counting argument, where we start by counting the
number of particles that hit the hole with a certain velocity and then integrate over all velocities.
We will start by counting the number of particles that move with speed v (technically speed
in [v, v + dv], but from now on we’ll be lazy about this) and spherical coordinate angles (θ, ϕ).
Here θ = 0 means pointing toward the hole, and θ = π/2 is parallel to the plane of the hole (the
spherical coordinates for the velocity are flipped compared to those for space, since the θ = 0 rays
are anti-parallel). In a given volume dV , the number of particles with this velocity is just

(ndV ) · f (v) · v 2 sin θ dv dθ dϕ.

For this given velocity, the volume in space that will allow such particles to hit the hole is a tilted
cone object with base A, slant θ, slant height vdt, and aligned in the proper ϕ direction. In particular,
its volume is A(vdt) cos θ, so the number of particles with velocity (v, θ, ϕ) hitting the hole in time
dt is
(nAdt) · f (v) · v 3 sin θ cos θ dv dθ dϕ.
Thus, the rate of particles leaving is
Z ∞ Z π/2 Z 2π Z ∞
3
α = nA v f (v) dv sin θ cos θ dθ dϕ = πnA v 3 f (v) dv.
0 0 0 0

On the other hand, note that the average speed is

Z ∞ Z π Z 2π Z ∞
2
⟨v⟩ = v · f (v) · v sin θ dv dθ dϕ = 4π v 3 f (v) dv,
0 0 0 0

which tells us that

nA
⟨v⟩, α=
4
as desired. Note that this derivation didn’t depend on the particular form of f (v). For instance, it
would work just as well for a relativistic particle, or even photons, as we’ll see in T2.

[5] Problem 24. In this problem, we consider a simple kinetic theory model for “transport” in a
sparse gas. Two parallel plates are at a distance L from each other. The space between the plates is
filled with a gas of density ρ, with molecule of mass m. Assume that the gas density is low enough
to neglect collisions between gas molecules, and that all radiation effects can be neglected.

27
Kevin Zhou Physics Olympiad Handouts

(a) Suppose the two plates are held at temperatures T + ∆T and T , where ∆T ≪ T . Assume
that when gas molecules bounce from a plate, they instantly obtain the plate’s temperature.
Estimate the heat flux per area P/A (in units of W/m2 ) between the plates. (This determines
the thermal conductivity, to be defined in T2.)

(b) Suppose the top plate moves with a transverse speed v∥ parallel to itself, while the bottom
plate is still, and mv∥2 ≪ kB T . Assume that when gas molecules bounce from a plate, they
obtain that plate’s transverse speed. Estimate the force per area F/A on each plate. (This
determines the viscosity, as defined in M7.)

Both the thermal conductivity and viscosity are associated with the transport of conserved quantities
(energy and transverse momentum, respectively) through the gas. Now, above we have neglected
collisions entirely, which is unrealistic for most gases unless they are extremely sparse. However,
it’s possible to get some rough estimates for what happens in the presence of collisions.

(c) Let σ be the cross-sectional area of a gas molecule. Estimate the mean free path d of a gas
molecule.

(d) Our previous analysis held when L ≪ d. Now suppose, more realistically, that L ≫ d.
Estimate P/A and F/A in the steady state. (Hint: consider a set of imaginary planes that
are spaced by a distance d, and consider the heat/momentum flow between pairs of adjacent
planes; for such pairs, collisions can be neglected.)

(e) Suppose a gas molecule begins at the top plate, and both plates have temperature T . Again
assuming that L ≫ d, estimate the typical time it takes before the gas molecule hits the
bottom plate. (Hint: model the position of the particle as a sum of random steps, and recall
the facts about variance in P2.)

For a clear discussion of the ideas of this problem, see sections 7.3 and 7.4 of The Art of Insight.
p
Solution. (a) The gas molecules have speed v ∼ kB T /m, and they hit the plates every time
t ∼ L/v. Each time a plate is hit, an energy ∆E ∼ kB ∆T is deposited or taken away. So the
average energy flow rate for one atom is
r
∆E kB ∆T kB T
Patom = ∼ .
t L m
The number of atoms per unit area of the plate is Lρ/m, which gives
r
P ρ kB T
∼ kB ∆T .
A m m

(b) This is rather similar, except that instead of an energy ∆E deposited every impact, a momen-
tum ∆p∥ = mv∥ is deposited per impact. Then by the exact same logic,
r
F ρ kB T
∼ mv∥ .
A m m
Notice that the result of this problem depends explicitly on the molecular mass m. This is
very important, because in the early days of kinetic theory, the quantity m canceled out of
almost every calculation. This led people to believe that atoms were unobservable in principle;

28
 Kevin Zhou Physics Olympiad Handouts

they were just a fictitious calculational tool used to approximate continuous matter. One of
Einstein’s most famous discoveries was showing that macroscopically observable properties of
Brownian motion did depend on m, hence giving a way to indirectly observe the reality of
atoms; you will do this calculation in XR.

(c) If a molecule moves a distance x, it’ll sweep out a volume of σx. The number density is ρ/m,
so the molecule should expect ρσx/m collisions. The mean free path d is when you would
expect 1 collision:
m
d= .
σρ

(d) The results for when collisions can be neglected,

r r
P ρ kB T F ρ kB T
∼ kB δT ∼ mδv
A m m A m m
can still give us the heat/momentum flow between adjacent planes separated by distance d
(so a total of L/d of these planes). Thus ∆T = Ld δT and v = Ld δv. This transfer will be the
same throughout in steady state, and plugging in the previously obtained d = m/σρ yields
r r
P kB ∆T kB T F mv kB T
∼ ∼ .
A σL m A σL m
Remarkably, the density ρ has completely dropped out! This was one of the most surprising
early predictions of kinetic theory, which was confirmed by Maxwell. For higher densities,
more molecules are available to carry energy or momentum, but they can’t go as far before
colliding.
p
(e) The molecule will be moving at velocity v ∼ kB T /m, so it takes time t ∼ d/v to go between
planes of separation d. However, after this point the molecule’s velocity will be randomized,
so its next step might go in any direction. This is a description of a random walk.
Each step of time t changes the displacement from the original plate by approximately ±d,
and the steps are independent. Then the standard deviation of the displacement adds in
quadrature, as covered in P2. The typical time to hit is when we get a standard deviation of
order L,
 2
L2
r
L d m
thit ∼ ∼ .
d v d kB T

[5] Problem 25. 01hˆ EuPhO 2017, problem 2. A nice problem using many of the above ideas.

Solution. See the official solution here.

Example 10

A box of mass M contains an ideal gas consisting of many particles, with total mass Mg .
Find ⟨Vx2 ⟩, the average mean-square velocity of the box along the x-axis.

29
Kevin Zhou Physics Olympiad Handouts

Solution
This problem illustrates how kinetic theory can get very subtle, even when we don’t have to
keep track of the detailed collision dynamics. First, note that if each gas molecule has mass
m, then each one has ⟨vx2 ⟩ = kB T /m by equipartition. Now there are two simple arguments
that give different answers.

1. If we apply equipartition of energy to the box itself, we get ⟨Vx2 ⟩ = kB T /M .

2. The total momentum of the box and gas together is zero, so

N
X
M Vx = −m vx,i .
i=1

On the other hand, by the rules of error propagation introduced in P1,

* N
!2 +
X kB T
m vx,i = m2 N = kB T Mg
m
i=1

which implies
Mg
⟨Vx2 ⟩ = kB T.
M2
What’s going on? It actually turns out that both of these arguments are wrong in general,
but each one is correct in different limiting cases.

The problem with the first argument is that the derivation of equipartition of energy assumes
that each degree of freedom is independent of the others, i.e. it ignores the fact that when
the box picks up momentum, the gas must have a compensating opposite total momentum.
The problem with the second argument is that it assumes the velocities of the molecules are
independent of each other, which is also not quite true.

Note that the energy and momentum of the box are related by E = P 2 /2M . Now, if the
box is heavy, M ≫ Mg , it can store lots of momentum even when it has very little energy,
and conversely when it has a decent amount of energy it must have an enormous momentum.
The latter implies that the first argument fails. But the second argument works, because the
box serves as a “sink” for momentum. For example, the gas molecules could all be moving
to the right, and that would be easily compensated by the box moving slightly to the left.

If the box is light, M ≪ Mg , then it can have lots of energy even having negligible momentum.
In this case, the first argument works, because the box doesn’t ever have enough momentum
to substantially affect the gas’s behavior. But the second argument fails, because if the box
can’t carry much momentum, then the gas molecules’ momenta must almost entirely can-
cel out among themselves by momentum conservation, meaning that they are not independent.

The general solution can be found with a simple trick. The problem with the equipartition
argument is essentially that we treat the box velocity Vx as independent of the center of mass

30
Kevin Zhou Physics Olympiad Handouts

velocity of the gas, v x . We can therefore switch to the variables

Mg v x + M V x
vCM = , vx,rel = Vx − v x .
Mg + M

These two new variables are independent, because the momentum conservation constraint
just says the former is always equal to zero. Thus, we can safely apply equipartition of energy
to vrel . Recalling the reduced mass idea from M6, we have
1 2 M Mg
E ⊃ µvx,rel , µ= .
2 M + Mg

By equipartition we have
2 kB T
⟨vx,rel ⟩=
µ
and by momentum conservation we conclude
 2
Mg kB T Mg
⟨Vx2 ⟩ = 2
⟨vx,rel ⟩=
M + Mg M M + Mg

which reduces to the two answers found above in the appropriate limits. This problem turns
out to be relevant to astrophysics, where it is a toy model for the dynamics of a supermassive
black hole in a galaxy. For a complete analysis which starts from the postulates of kinetic
theory, see this paper.

Remark
Above, we discussed the distinction between “bottom-up” and “top-down” approaches, but
there are also bridges between the subjects. For example, suppose you had a large volume of
gas, where the mean free path is much smaller than the container’s size. Here, the gas can
come into equilibrium locally, at which point we may describe its state with a temperature
field T (x, t), along with other fields, such as pressure, density, and velocity. This leads to
the subject of hydrodynamics, which can be derived from kinetic theory. We started on this
subject in M7, but there we neglected any thermal properties; we will return to it in T3.

On the other hand, suppose an entire macroscopic system is already in thermal equilibrium.
The system is homogeneous, so there’s no need for hydrodynamics, but it can still be difficult
to infer the macroscopic behavior of the system. For example, if you had water molecules
at a given temperature and pressure, it’s still hard to calculate what phase they’re in! In
physics, our best tool for this kind of problem is the renormalization group, which bridges
the gap by considering a series of “coarse-graining” operations that gradually zoom out. This
is a deep subject, typically reserved for graduate courses.

31