You are on page 1of 10

Tutorial 2: Quantum Fermi Gas

Contents
1 A Quantum Gas 1
1.1 The Necessity of Quantum Mechanics . . . . . . . . . . . . . . . . . . . . . 1
1.2 Energy Eigen States . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

2 Density of States 2
2.1 Momentum Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.2 Energy Space (3D) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

3 Exercises in 2D 4
3.1 Chemical Potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3.2 Fermi Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
3.3 Heat Capacity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

4 Pauli Pressure 9

1 A Quantum Gas
1.1 The Necessity of Quantum Mechanics
In Tutorial 1 and the first few lectures the electrons in a metal were treated as an ideal
classical gas. Although it sometimes works, such a treatment leads to results that do
not agree with experimental observations. A prominent example is the heat capacity.
According to classical physics, the heat capacity per particle is a universal constant:
cV /N = 3kB (Dulong-Petit). The experiment on the other hand shows a linear temper-
ature dependence: cV ∼ T . To resolve this contradiction we must treat the electrons
quantum mechanically and respect the fermionic nature of the electrons – even at room
temperature.
Why is quantum mechanics important at such elevated temperatures? Because room
temperature, from the point of view of electrons in a metal, is cold. The energy scale one
must compare the thermal energy kB T to, is the Fermi energy εF . While for a typical
metal εF ≈ 1 eV (and TF = εF /kB ≈ 11600 K), for TRT = 300 K the thermal energy is
kB TRT ≈ 26 meV (∼ 40 times smaller).
We see that room temperature is cold and hence quantum statistics are essential for
obtaining correct results.

1
1.2 Energy Eigen States
We think of a metal in the simplest possible way: N non-interacting electrons confined to
a box of volume L3 (3D infinite potential well) and periodic boundary conditions. Both
N and L can be thought of as very (infinitely) large while the ratio N/L3 , which is the
particle density n, must remain constant. Since the electrons do not interact with each
other (ideal gas), the possible energy levels can be found by solving the single-particle
problem:
~2 k 2 2π
ε(~k) = and ~k = ~n. (1)
2m L
Here k = |~k| and ~n is a vector of integers. We use ε to denote the energy levels (and not
E) to emphasize that these are the single-particle energies. The letter E is reserved for
the total energy of the system which is the sum of all N single-particle energies.
The periodic boundary conditions imply that the particle is not actually confined (as it
would be for closed boundary conditions) but can move around freely – it is a free particle.
The wave function of a single electron is thus equivalent to that of a free particle:
1 i~k·~r
φ~k (~r) = e (2)
L3/2

or a superposition of various ~k’s.


So Eq. (1) denotes the energy levels and ~~k is the particles momentum. Is the momen-
tum quantized? Of course not. Momentum is never quantized. Given a wave function
ψ(x) you can find its momentum distribution via the Fourier transform: ψ(x) → ψ̃(k) →
P (k) = |ψ̃(k)|2 dk. It is a continuous function and hence the momentum is a continu-
ous variable. To resolve the apparent quantization in Eq. (1) we must make the metal
very large (L → ∞). In this case, the difference between two adjacent k-values, namely
2π/L, is negligible and ~k can be thought of as a (almost) continuous variable. To neglect
“boundary effects” we only deal with electrons which are so far from the edge that they
do not feel its influence. The technical term for these electros is “bulk electrons”.

2 Density of States
By density of states (DOS), denoted g(·), we mean how many energy levels are there per
unit energy, g(ε), or per unit momentum, g(k) and per unit volume. The units of, e.g.,
g(ε) in 3D are J−1 m−3 . It is the number of stats per unit energy and per unit volume.

2.1 Momentum Space


Here we calculate g(k) in one, two and three dimensions using intuitive arguments. For
now we will not consider the spin degree of freedom.

1D The vector ~k = kx x̂ has only one dimension and the momentum ~kx ∈ [−∞, ∞] can
take any value. The total number of states up to some value k = |~k| is therefore Ns = 2k,
one k for the positive side and one for the negative side – so a total of 2k. The number
of states dNs in the unit momentum-volume dk is

dNs = Ns (k + dk) − Ns (k) = 2dk (3)

2
and hence the DOS, so the number of states per dk, is
dNs
g1D (k) = = 2. (4)
dk
It turns out that g1D (k) is a pure number (dimensionless). Does this make sense? Let’s
recap: g(k) is the number of states per unit k and per unit volume. “Per unit k” has
dimensions of 1/k, so length. “Per unit volume” (in 1D) has dimensions of inverse length.
Hence, g1D (k) is dimensionless.

2D The momentum now depends p on two values: kx and ky . Since the energy only cares
about the absolute value |k| = kx2 + ky2 it is convenient to think of the problem in polar
coordinates. The number of states between k = 0 and some value of k is just the area of
a circle with radius k: Ns = πk 2 , and hence the DOS is
dNs
g2D (k) = = 2πk. (5)
dk

3D To add one more dimension we go from polar to spherical coordinates, i.e. Ns =


4πk 3 /3 is given by the volume of a sphere. Taking the derivative with respect to k results
in the DOS:
g3D (k) = 4πk 2 . (6)

d Dimensions The generalization to d dimensions is straight forward: Ns is given by


the volume of a (hyper-)sphere in d dimensions and the DOS by its (d − 1) dimensional
surface area. This simple form is due to the relation ε ∼ k 2 in Eq. (1). For a more
complicated dispersion relation it becomes a little more involved.

Fermi Level The N electrons populate the (discrete, but almost continuous) levels
ε(~k), i.e. each electron is in one out of the (infinitely many) possible states. In the
ground state, so for T = 0, all electrons will go to the lowest possible state. Due to their
fermionic nature there can be only one electron (two, if one considers the spin degree
of freedom) per level and hence they will pile up from the bottom. The k-value of the
highest populated state is called the Fermi level kF . It may also be expressed in terms of
energy: εF = ~2 kF2 /2m.

2.2 Energy Space (3D)


After having computed g(k) we now wish to find g(ε). The total energy of the system is
X
E= ε(~k)fF D (~k), (7)
~k

where
1
fF D (~k) = (8)
eβ(ε−µ)
+1
is the Fermi-Dirac distribution function and β = 1/kB T . Remember that for T = 0, fF D
is a step function: (
1 k < kF
fF D (~k) = , T =0 (9)
0 k > kF

3
as was discussed above. Since the number of electrons in the system N → ∞ is very large,
the total energy is very large. Moreover, measuring it is nearly impossible because one
must make a simultaneous measurement of the entire system before the particles move
around. A much better (local) quantity is the energy density u = E/L3 :

1 X ~
u= ε(k)fF D (~k). (10)
L3
~k

Next, because ~k is a (almost) continuous variable, we replace the sum with an integral.
Doing so we must introduce an infinitesimal volume in momentum space, d3 k, without
changing the dimensions of the expression. The correct way to do so is:
ˆ  3
X L
→ d3 k. (11)

~k

The factor of 2π/L can be traced back to the 2π/L in Eq. (1). After canceling the volume
L3 the energy density is ˆ
d3 k ~
u= ε(k)fF D (~k). (12)
(2π)3
~
´ 3 of k´= 2|k| only
Now remember that both ε(k) and fF D (k) are functions ´ and therefore
´ 2 the
angular integral in Eq. (12) is simple to compute: d k = k dk dΩ = 4π k dk. In
addition, lets express u as an integral over energies, not momentum. Using Eq. (1) we
find r
2mε m ~2
k2 = 2 and dk = 2 dε (13)
~ ~ 2mε
and hence
ˆ ˆ r
1 m 2mε
u = 2 dkk 2 · ε(k)fF D (k) = dε 2 2 · ε · fF D (ε). (14)
2π 2π ~ ~2
´
Since the energy density is given by u = dεg(ε)εfF D (ε), we recognize
r
m 2mε
g(ε) = 2 2 . (15)
2π ~ ~2
At this point it is a good idea to check that the obtained expression indeed has units of
J−1 m−3 .
The connection between g(ε) and g(k), for d space-dimensions, is

g(k)dk
g(ε)dε = (16)
(2π)d

which also comprises an alternative way for calculating g(ε), provided g(k) is known.

3 Exercises in 2D
In the following we will solve three examples in a 2D electron gas. As we shall see,
the DOS g2D (ε) turns out to be energy independent and hence many thermodynamic

4
quantities can be computed exactly even for T 6= 0. Other dimensions do not have this
convenient property and need a more complicated treatment (Tutorial 3).
First we find the DOS. Lets apply Eq. (16) to the 2D DOS in Eq. (5) to find g(ε) of
a 2D electron gas:
g(k) dk
g(ε) = . (17)
(2π)2 dε
First we plug in r r
dk d 2mε m
= = (18)
dε dε ~2 2~2 ε
p
and Eq. (5) for g(k). Then we must replace k → ε by means of k = 2mε/~2 . One
obtains r r
2πk dk 1 2mε m m
g(ε) = = = (19)
(2π)2 dε 2π ~2 2~2 ε 2π~2
which, as advertised, is energy independent.

3.1 Chemical Potential


Statement of the Problem Using the DOS, show that the chemical potential of a 2D
electron gas is given by
 
2π~2 n/mkB T
µ(T ) = kB T ln e −1 . (20)

Setting Up the Integral The chemical potential is always found from normalization,
i.e. from the fact that there are N electrons in the total area L2 . As mentioned, these
quantities are not very useful so we will later express our solution in terms of the ratio
n = N/L2 .
The DOS is the number of electrons per unit energy and per unit area. Hence we must
integrate g(ε) over the entire space and entire energy range to obtain N , while respecting
the population distribution fF D (ε):
ˆ ∞
2
N =L dεg(ε)fF D (ε). (21)
0

Since for the 2D case g(ε) = const. we must solve


ˆ ∞
N m 1
2
= 2
dε β(ε−µ) (22)
L 2π~ 0 e +1
from which we will be able to eliminate µ.

Solving the Integral We substitute x = β(ε − µ), such that dε = kB T dx and the
integration limits are −βµ → ∞:
ˆ
N mkB T ∞ 1
2
= 2
dx x . (23)
L 2π~ −βµ e +1

Then the substitution u = ex , so dx = e−x du = du/u, is made:


ˆ
N mkB T ∞ du 1
= . (24)
L2 2π~2 e−βµ u u + 1

5
The integral can now be solved by means of partial fraction decomposition, i.e. the
integrand is spit into two via
1 1 1 1
= − (25)
uu+1 u u+1
and each part is solved individually. When plugging in the infinite upper limit, though,
one must pay attention. We hence first solve the integrals, put them together and then
plug in the limits. Using
ˆ ˆ
du du
= ln u and = ln (u + 1) (26)
u u+1
we get ∞
N mkB T
= [ln u − ln (u + 1)] . (27)
L2 2π~2 | {z } e−βµ
ln(u/(u+1))

It is now easy to see that the expression vanishes for u = ∞ [ln(1) = 0]. Hence, using
n = N/L2 , one obtains
2π~2 n e−βµ −βµ −βµ

= − ln −βµ =− ln e + ln e + 1 . (28)
mkB T e + 1 | {z }
βµ

Chemical Potential In order to eliminate µ we move βµ to the lhs and exponentiate.


Then we eliminate e−βµ and take the logarithm on both sides. Dividing by β finally results
in the desired expression for µ. Lets go:
2π~2 n
= ln e−βµ + 1

− βµ
mkB T
2π~2 n −βµ
 
exp e = e−βµ + 1
mkB T
2π~2 n
   
−βµ
e exp −1 = 1 (29)
mkB T
2π~2 n
 
exp −1 = eβµ
mkB T
2π~2 n
   
ln exp −1 = βµ
mkB T
The final solution is thus
2π~2 n
   
µ = kB T ln exp −1 (30)
mkB T
which coincides with Eq. (20).
Here we were able to obtain an analytical expression for µ(T ). This is not possible for
any other dimension d 6= 2, because only for d = 2, g(ε) = const. and the computation
reduces to integrating over fF D (ε).

3.2 Fermi Energy


Statement of the Problem Show that the Fermi energy is given by εF = 2π~2 n/m,
where n = N/L2 is the electron density. Show that in the low temperature limit kB T 
εF , which is applicable to room temperature, one obtains µ = εF from Eq. (20).

6
Intuitive Thinking The Fermi energy εF is the highest populated state at T = 0. This
means that the number of electrons in the system (N ) is equal to the number of states
below εF . Expressed as an equation this is
ˆ εF
2
N =L g(ε)dε. (31)
0

Note that this coincides with Eq. (21) for T = 0.


´
Computation Since g(ε) = const. in the 2D case the integral is simply dε = εF and
hence
N m
2
= εF . (32)
L 2π~2
Eliminating εF gives
2π~2 n
εF = (33)
m
which is what we set out to show.

Low Temperature Limit Plugging the found expression for εF into Eq. (20) allows
us to write the chemical potential as

µ(T ) = kB T ln eβεF − 1 .

(34)

In the limit kB T  εF , so βεF  1, the 1 is negligible and we find

µ(T → 0) = kB T ln eβεF = εF .

(35)

3.3 Heat Capacity


Statement of the Problem Find the heat capacity (per unit area)

1 ∂E ∂u
cV = 2
= (36)
L ∂T ∂T
in the limit kB T  εF ≈ µ.

Setting Up the Integral In Sec. 2.2 we saw that the energy density (units of J/m2 )
is given by ˆ ˆ ∞
m ε
u = dεg(ε)εfF D (ε) = 2
dε β(ε−µ) . (37)
2π~ 0 e +1
This integral does not have an analytical solution but lets worry about that later. For
now we take the derivative with respect to T :
  ˆ ∞
∂u ∂β ∂u 1 m ∂ ε
cV = = = − dε (38)
∂T ∂T ∂β kB T 2 2π~2 0 ∂β eβ(ε−µ) + 1

resulting in
ˆ ∞
1 m ε(ε − µ)eβ(ε−µ)
cV = dε 2 . (39)
kB T 2 2π~2 0 [eβ(ε−µ) + 1]

7
If we write the denominator as
 
2 β(ε − µ)
 β(ε−µ) 2 β(ε−µ)
 β(ε−µ)/2 −β(ε−µ)/2 2
 β(ε−µ)
e +1 =e e +e = 4e cosh (40)
2
the exponent cancels and the integral becomes
ˆ ∞
1 m ε(ε − µ)
cV = 2 2
dε 2 1
 . (41)
kB T 8π~ 0 cosh 2 β(ε − µ)

Next we make the substitution x = β(ε − µ)/2. Using ε = 2xkB T + µ and dε = 2kB T dx
we get ˆ
mkB ∞ x (2xkB T + µ)
cV = dx . (42)
2π~2 −βµ/2 cosh2 x
This is as far as mathematical tricks can take us.

Low Temperature Limit In order to actually solve the integral we must make an
approximation. In virtue of kB T  εF ≈ µ we take the lower integration limit to −∞.
We then have two integrals from −∞ to ∞. The second one, the one over x/ cosh2 x,
vanishes because the integrand is an odd function and we are left with
ˆ
mkB2
T ∞ x2
cV = dx . (43)
π~2 −∞ cosh2 x
At this point we consult a list of known integrals and find
ˆ ∞
x2 π2
dx = , (44)
−∞ cosh2 x 6

a well-known fact1 . The heat capacity is therefore given by


2
2πmkB 4π 2 kB
2
n
cV = 2
T = T, (45)
3~ 3εF
where we used Eq. (33) in the second equality to express cV in terms of εF .

Summary The take-away message is that, for T → 0, cV ∼ T - a direct contradiction


to the Dulong-Petit (DP) law which states that cV = 3N kB = const. for any material
and any temperature. The truth is that DP is correct in the classical limit, i.e. when the
fermions behave like distinguishable particles. At low temperatures though their mutual
behavior is governed by the Fermi-Dirac statistics defined by fF D (ε).
In fact, it is the Fermi-Dirac statistics that causes the linear dependence cV ∼ T . The
heat capacity is a measure for the number of degrees of freedom that can absorb heat
(or energy) added to the system. In the case of identical fermions, a particle with some
energy can only absorb heat if the state at “its energy + the added heat” is unoccupied.
According to fF D (ε) this is only true for a narrow band of width ∼ T around εF . Hence
the number of electrons that can absorb heat is ∼ T and cV , which is a measure for this
number, behaves like cV ∼ T .
1
What if I don’t know this? Do I have to know integrals of this kind by heart? Of course not! The
important thing is that you realize the integral is dimensionless. It is a pure number and does not affect
the temperature-dependence of cV (or any other parameter-dependence).

8
a Maxwell-Boltzmann c
𝑘𝐵 𝑇 (boson) (fermion)
𝑣Ԧ

size of gas

b cooling d
Bose-Einstein Fermi-Dirac

Figure 1: Pauli Pressure. a. At high temperatures, where the atoms have a well defined
position and momentum, the gas is described by the Maxwell-Boltzmann distribution. b.
When cooled, the quantum nature of the particles becomes important. c. Lithium atoms
in a harmonic trap. At elevated temperatures the bosons and fermions behave the same.
d. Cooling causes the bosons to collapse towards the center of the trap. For fermions the
collapse is forbidden by the Pauli exclusion principle.

This line of reasoning can also be applied to bosons. In this case the number of
degrees of freedom that can absorb heat is equal the number of momentum axes (the
dimension of the system), because when heat is absorbed it must become kinetic energy.
E.g. in 3D, there are 3 momentum axes [since ~k = (kx , ky , kz )] and cV ∼ p3 . If the
bosons are massive√ (i.e. they have mass) their energy is proportional to ∼ p2 and hence
kB T ∼ p2 → p ∼ T . Finally we get cV ∼ T 3/2 . If the bosons are massless (phonon) the
energy-momentum relation is linear (kB T ∼ p) and we obtain cV ∼ T 3 .

4 Pauli Pressure
Here the difference between Bose-Einstein and Fermi-Dirac statistics is explored by means
of a beautiful experiment. It is important to stress that in the following, as during
the entire course, both the bosons and the fermions are non-interacting, i.e. all the
observations are a direct consequence of the quantum statistics of the particles and have
nothing to do with forces exerted by one particle onto another.
The experiment is done with so-called cold atoms. This research field specializes in
trapping and cooling2 atoms inside optical and magnetic traps. Being cold, the de Broglie
wavelength is relatively large3 and the quantum nature of the atoms becomes important.
2
When I say cooling I mean slowing down. In other words, cold atoms have almost no kinetic energy.
While room temperature atoms fly around at several 100 m/s, cold atoms are typically slower than 1
cm/s.
3
You might know that when the de Broglie wavelength is larger than the interparticle distance, bon-
sons condense into a Bose-Einstein condensate. This is not the case discussed here. In the following
experiments the de Broglie wavelength is large enough for quantum mechanics to be important but still
substantially smaller than the interparticle distance.

9
Typically, a cold atomic gas is trapped inside a harmonic potential. In Fig. 1(a) the
classical situation is shown. The particles move around having a well-defined position
and momentum at each point in time. The distribution of the various positions and
momenta is described by Maxwell-Boltzmann statistics. In particular, the density profile
(distribution of positions) is a Gaussian whose full-width at half-maximum is given by
the temperature.
Now we lower the temperature so that quantum mechanics becomes important. The
particles no longer have a well-defined position and momentum but populate discrete en-
ergy levels (quantum harmonic oscillator). As shown in Fig. 1(b) the bosonic or fermionic
nature leads to a very different behavior. The bosons more and more populate the
lower energy levels and thus the size of the atomic cloud corresponds to the tempera-
ture. Fermions, on the other hand, cannot populate an already full energy level and are
hence also found at energies higher than kB T . This leads to a much larger atomic cloud
which no longer depends on the temperature but instead on the total number of atoms.
Figs. 1(c) and (d) show the experiment conducted in the group of Randall G. Hulet at
Rice University in Houston, Texas4 . Two isotopes of Lithium, bosonic 7 Li and fermionic
6
Li, were cooled and trapped inside a magnetic trap. One clearly sees [Fig. 1(c)] that for
elevated temperatures there is no difference between the bosons and fermions. However,
as the temperature is lowered [Fig. 1(d)] the bosonic gas shrinks while the fermionic gas
remains at constant size.

4
Science 291, 2570 (2001)

10

You might also like