Kevin Zhou Physics Olympiad Handouts

are lots of photons, their discreteness doesn’t matter, so the mode can be treated classically,
getting energy kB T by the equipartition theorem.
For modes with high E, we have ⟨n⟩ ≈ e−E/kB T ≪ 1. This also makes sense, because in this
limit there isn’t enough thermal energy to significantly excite the modes.

(c) The logic is exactly the same as part (c) of the previous problem, but instead of saying that
some modes have 1 photon and others have 0 photons, we just assign the proper ⟨n⟩ photons
to every mode. Then
Z  3 Z ∞
1 L
U = ⟨n⟩E dN = 2 ⟨n⟩E 3 dE.
2π ℏc 0

As stated, we multiply by a factor of 2 to account for the two polarizations per mode. We
also change variables from E to ω using E = ℏω. This gives
L3 ℏ ∞ ω3
Z
U= 2 3 dω ℏω/k T
π c 0 e B −1
just as desired.

(d) Now we make the substitution x = ℏω/kB T , where dx = ℏdω/kB T .

L3 ℏ ∞ kB T (kB T x/ℏ)3 (kB T )4 L3 ∞ x3
Z   Z
U= 2 3 dx x
= 2 3 3
dx x .
π c 0 ℏ e −1 π c ℏ 0 e −1
We can rearrange the integral into
Z ∞ Z ∞
x3 1
I= dx x = e−x x3 dx.
0 e −1 0 1 − e−x
Recognizing 1/(1 − e−x ) as a geometric series ∞ −x n
P
n=0 (e ) , we can represent the integral as
Z ∞X ∞
I= x3 e−(n+1)x dx.
0 n=0

We can reindex this since the only instance of n is n + 1, so we can start the summation at
n = 1. Since we can integrate this term by term, we can change the order of the integral and
summation, then integrate by parts to get
∞ Z ∞ ∞ Z ∞  −nx  X ∞ Z ∞  −nx 
X
3 −nx
X
2 e e
I= x e dx = (3x dx) = (6xdx)
0 0 n 0 n2
n=1 n=0 n=0
∞ Z ∞  −nx
 ∞
X e X 6
= (6dx) = = 6ζ(4).
n3 n4
n=0 0 n=0
Putting this into our original expression gets
π 2 kB
4
U= L3 T 4 .

01hˆ
15c3 ℏ3

[5] Problem 25. APhO 2002, problem 1. This useful problem covers the other common example
of a quantized bosonic field. In the above problems, we quantized the electromagnetic field to get
photons. Here, we quantize a displacement field to get phonons.

21
 Kevin Zhou Physics Olympiad Handouts

Idea 7
In problems 23 and 24, we handled a system of bosons (specifically photons) by considering
the modes the photons could occupy, then calculating how many photons were in each mode.
This was the easiest route. If we had instead fixed the number of photons, then counted
the ways they could be distributed among the mode, the combinatorics would have been a
complete nightmare, because multiple photons can occupy the same mode.

Fermions, which obey the Pauli exclusion principle, are simpler, because no two can be in
the same state. For instance, if there are n noninteracting fermions in a system, then the
lowest energy state of the whole system consists of having one fermion occupy the lowest
energy state, the second occupy the second-lowest energy state, and so on. (Accounting for
interactions makes the problem much more complicated, because it means the energy of a
state depends on whether other states are occupied. However, you can explain a surprising
amount while completely neglecting interactions.)

Example 6: Tremaine–Gunn Bound

Suppose all of the dark matter in the galaxy is composed of a single kind of fermionic particle,
of mass m. The escape velocity of the galaxy is of order vesc ∼ 10−3 c, and the dark matter
density near Earth is ρ ∼ 0.3 GeV/(c2 cm3 ). What’s the minimum possible value of m?

Solution
The reason there’s a minimum possible value of m is that, as m gets smaller, we need more
dark matter particles. But the Pauli exclusion principle tells us that if we want to add more
particles, they need to have higher and higher energy, and at some point the particles will
have so much energy they won’t be bound to the galaxy at all.

To get a rough estimate, let’s suppose the galaxy has length scale L, so that we need at
least N ∼ ρL3 /m dark matter particles. They need to have energy less than E0 ∼ m0 vesc 2 .

Plugging this into the final result of example 2 and dropping all numeric factors gives

ρL3 m3 vesc
3 L3
≳
m ℏ3
which yields the bound
1/4
ρ ℏ3

m≲ 3
∼ 10 eV/c2 .
vesc
A few decades ago, neutrinos were leading dark matter candidates, since they are light
fermionic particles that interact very weakly with ordinary matter. But we now know that
the neutrino mass is well below this bound, so that nice idea doesn’t work. On the other
hand, dark matter could still be composed of bosonic particles of much lighter mass.

[2] Problem 26. Consider a system with many noninteracting fermions, and many quantum states.
Each quantum state can be either empty or occupied by a fermion. We want to find the probability
that a given quantum state, of energy E, is occupied.

22
 Kevin Zhou Physics Olympiad Handouts

(a) To put a fermion in this state, we need to remove a fermion from some other state. Suppose
the energy released by doing this, suitably averaged, is µ. (This is the chemical potential, and
it depends on the temperature, the number of fermions, and the number of states and their
energies.) Using the Boltzmann distribution, show that the probability of occupancy is
1
⟨n⟩ = .
e(E−µ)/kB T + 1
This is the Fermi–Dirac distribution.

(b) Sketch ⟨n⟩ as a function of E for small but nonzero temperature, as well as the limit attained
for zero temperature.

Solution. (a) The two possibilities are being occupied and not occupied, and the latter comes
with a Boltzmann factor of e−(E−µ)/kB T . Thus,

e−(E−µ)/kB T 1
⟨n⟩ = −(E−µ)/k T
= (E−µ)/k
.
1+e B 1+e BT

(b) The graph looks like this for µ = 0:

For a nonzero value of µ, the graph should look like this:

As T → 0, the form of ⟨n⟩ will start to look like a step function, θ(µ − E). This simply means
that the fermions fill up the lowest energy states first, to minimize their total energy. The
chemical potential is set by how many fermions there are in total.

[3] Problem 27. In this problem we’ll consider the energy of the conducting electrons in a solid at low
temperatures. Model a solid as a cubical box of side length L with periodic boundary conditions.

(a) Find the number of quantum states with energy at most EF , making sure to account for the
two spin states of the electron.

23
 Kevin Zhou Physics Olympiad Handouts

(b) Suppose there are N electrons in total. They will fill all of the energy levels up to µ = EF ,
where EF is called the Fermi energy. Show that
2/3
h2

3N
EF = .
2m 8πV

(c) A sodium crystal has one conduction electron per atom. The density and molar mass are

ρ = 0.971 × 103 kg/m3 , M = 0.023 kg/mol.

Find N/V and EF , and use this to evaluate the typical speed v of an electron.

Solution. (a) In a cubical box with period boundary conditions, the wavenumbers satisfy ki =
2π
L n√ i where ni can also be a negative integer. As seen in the example, the number of states is
4
3 π( 2mEF )3 (2πℏ/L)−3 , but we multiply that by 2 due to the two spin states of the electron.
 −3
8 2πℏ
N = π(2mEF )3/2 .
3 L

(b) In a cube, V = L3 . Rearranging the above equation gets

2/3
h2

3N
EF = .
2m 8πV

(c) If NA is Avogadro’s number, then the mass of each atom is m = M/NA , which gives an atom
density of ρ/m. Since there’s one conduction atom per atom, we have N/V = ρ/m = NA ρ/M ,
which gives us N/V = 2.54 × 1028 m−3 . Putting this into our previous formula gives EF =
5.05 × 10−19 J. The typical speed vF satisfies
1
EF ∼ mvF2
2
which gives vF ∼ 106 m/s, or on the order of 1% of the speed of light! This verifies a statement
made in E4 about electrical conduction in solids.

[3] Problem 28 (MIT). [A] This is an advanced problem that is only placed here because the final
result is neat. An integer N can be partitioned by writing it as a sum of positive integers, and the
partition function p(N ) is the number of unique ways this can be done. For example,

4=1+1+1+1=1+1+2=2+2=1+3=4

which implies p(4) = 5. Counting the number of partitions of an integer is a very hard combinatorics
problem, but we can get an estimate for large N using string theory.

(a) Consider an ideal string with hard boundary conditions and fundamental frequency ω. Show
that the number of distinct quantum states with energy N ℏω is p(N ).

(b) Now suppose the string is at temperature T , where T is chosen so that the expected energy
is N ℏω. In the thermodynamic limit N ≫ 1, find a relation between N and T . You may use
the result ζ(2) = π 2 /6.

24
Kevin Zhou Physics Olympiad Handouts

(c) By approximating the entropy as S ≈ kB log p(N ), show that

ℏω d log p(N )
= .
kT dN
Combine this with your previous result to find an estimate for p(N ).

To check your answer, the celebrated Hardy–Ramanujan formula (which is more accurate than the
very rough treatment we give above) is
r !
1 2N
p(N ) ∼ √ exp π .
4 3N 3

While we only considered a simple nonrelativistic string here, calculations of this sort show up in
the thermodynamics of string theory. For further discussion, see chapter 22 of Zwiebach.

Solution. (a) The modes of the string have frequencies mω, which means that upon quantization,
each quantum in mode m will have energy mℏω. Thus, putting in quanta to reach total energy
N ℏω is the same as building a partition of N out of integers m. For example, 4 = 1 + 1 + 1 + 1
corresponds to putting 4 quanta in the fundamental mode, while 4 = 4 corresponds to putting
1 quantum in the fourth harmonic.

(b) By borrowing some results from previous problems, we have

Z ∞
X m m dm
N= mℏω/kT
≈ mℏω/kT
.
m
e −1 0 e −1

We have already done a similar integral in problem 24, and using the same trick of expanding
the denominator in a Taylor series and integrating term by term gives

π 2 kT 2
 
N≈ .
6 ℏω

(c) Using the definition of temperature,

dE ℏω dN
T = = .
dS k d log p(N )
Rearranging slightly gives the desired result. Eliminating T using the above result,
d log p π
≈√ .
dN 6N
Separating and integrating,
r
Z N
π 2N
log p(N ) ∼ √ dN = π
0 6N 3
which agrees with the Hardy–Ramanujan formula. (Of course, this only gets the leading term
right, because we made approximations. The most severe approximation we did was taking
S ≈ kB log p(N ). In reality, the energy at a given temperature is not fixed, but rather has
fluctuations, so we should have instead summed p(n) for a range of n near N .)

25
Kevin Zhou Physics Olympiad Handouts

Example 7: Casimir Forces

Consider an infinite uniform string, on which waves propagate with speed v. The string is
held fixed by pins at two points separated by a distance L. When the string is in its ground
state, what is the resulting force between the pins?

Solution
Of course, in classical mechanics the ground state would just be y(x) = 0, and the force would
just be the tension T . But there is an additional quantum mechanical contribution, which
arises because each of the standing waves between the pins, with frequencies ωn = nπv/L,
carries a so-called “zero point” energy ℏωn .

As usual, force can be found by differentiating energy, F = −dE/dx. By summing up the

zero point energy in all of the standing waves, we naively get
∞ ∞
X ℏωn πℏv X X
E= = n = E0 n=∞
n
2 2L
n=1 n=1

which is rather unhelpful. This result is off for two reasons. First, even when there aren’t
any pins, the string still has standing waves, and these waves also have a naively infinite
energy density. When we move the pins a bit, we change both the zero point energy outside
the pins and inside, by infinite amounts, but the net change is finite, giving a finite force.

Quantitatively, the energy in between the pins due to the standing waves is
∞
X
Eplate = E0 n=∞
n=1

and the energy we would have had there if the pins didn’t exist is the “continuous” sum,
Z ∞
Evac = E0 x dx = ∞.
0

The difference should be finite, but we can’t just subtract infinity with infinity, which brings
us to the second problem: none of these quantities are actually infinite. Any real string
will have a finite maximum oscillation frequency – for instance, the wavelength certainly
can’t get smaller than the atomic spacing. Alternatively, even if we had an idealized string
where Evac was actually infinite, no real pin can perfectly block waves of all frequencies. For
sufficiently high frequencies the waves won’t be affected by the pins, so that the sum in
Eplate eventually behaves like the integral in Evac , leaving a finite difference between the two.

In other words, the difference between Eplate and Evac in reality comes from only low n and
x, Therefore, let’s “regulate” the two expressions above so that they’re unchanged in this
regime, but match each other at high n and x. The simplest way to do this is to take
∞ Z ∞
X
−ϵn E0
Eplate = E0 ne , Evac = E0 xe−ϵx dx =
0 ϵ2
n=1

26
Kevin Zhou Physics Olympiad Handouts

for small ϵ. To handle the sum, let α = e−ϵ , so that

Eplate /E0 = α + 2α2 + 3α3 + . . . .

Now we use the usual trick for arithmetic-geometric series. Note that

αEplate /E0 = α2 + 2α3 + 3α4 + . . . .

Subtracting, we find
α
(1 − α)Eplate /E0 = α + α2 + α3 + . . . = .
1−α
We thus conclude that
e−ϵ 1
Eplate = −ϵ 2
E0 = Evac − E0 + O(ϵ)
(1 − e ) 12

where we used a result from P1. Finally, when we take ϵ to zero, the difference is simply
1
E = Eplate − Evac = − E0 .
12
Differentiating gives the force,
πℏv
F =
24L2
which turns out to be attractive. Not only is this finite, it’s right! Experiments have
measured this “Casimir force” precisely for light between two conductors, where v = c, and
confirmed the expected results.

You’re probably suspicious about this derivation because it depends on the arbitrary choice
of an exponential suppression. What if the sums and integrals were regulated at high n and
x in a different way? Shouldn’t the answer depend on the details of the string and pin?
2
Remarkably, the answer is no: the regulator doesn’t matter. If you try others, such as e−ϵn
or 1/nϵ , you’ll get the same result; you can find a general proof in chapter 15 of Schwartz’s
Quantum Field Theory and the Standard Model. The reason is that the effect comes from
physics at low frequencies, so it doesn’t matter how you regulate the high frequencies.

It is for precisely this reason that you will sometimes see the mysterious equation
1
1 + 2 + 3 + ... = − .
12
It’s not really true. Instead, what it physically means is that the difference between the
regulated sum and integral is −1/12 for any reasonable regulator.

27