Chapter 4

PHONONS
1. Solids in the adiabatic approximation
Every solid consists of electrons with mass m
e
and charge e and
nuclei with mass M
n
and charge Z
n
e. The interaction between electrons
and nuclei are purely of electromagnetic origin (as are the interactions
among the electrons and nuclei themselves). By far the most important
contribution to the interaction is due to the Coulomb force, but in some
cases other contributions, e.g., the spin-orbit coupling, also have to be
taken into account. When we, for simplicity, ignore such relativistic
eects, the Hamiltonian for the coupled electronic-nuclear system is
H
solid
=

n
P
n
2
2M
n
+

i
p
i
2
2m
e
+
e
2
4
0

i<j
1
|r
i
r
j
|
+
e
2
4
0

l<n
Z
l
Z
n
|R
l
R
n
|

e
2
4
0

i,n
Z
n
|r
i
R
n
|
, (4.1)
where the position and momentum of the nuclei is denoted by capital R
and P, and position and momentum of the electrons is denoted by r and
p. Virtually the whole of solid state physics is contained in this Hamilto-
nian. If we were to have an exact procedure to nd all its eigenstates and
eigenenergies, theoretical solid state physics would become an obsolete
art. However, the Hamiltonian above, even for a small piece of material,
already consists of a set of about 10
23
coupled dierential equations. We
are at a hopeless loss when trying to solve such an astronomical number
of equations exactly (or numerically exactly).
Still, by just staring at the Hamiltonian above, we can come to an
important conclusion. Very elementary arguments tell us that in general
electrons move fast and that nuclei are slow. So, let us stare at the
formidable Hamiltonian (4.1) and ask ourselves the right questions. One
such a question is: what are the characteristic length-scale and energy
scales that rule this set of equations? The constants that appear in
the Hamiltonian are Plancks constant h, the electron mass m
e
, nuclear
47
48
mass M and the elementary charge quantum e. By taking a proper
combination of these natural constants we can dene a length-scale.
This length-scale, the Bohr radius, is the characteristic length-scale
of the Hamiltonian. It can easily be found by dimensional analysis.
Plancks constant h carries dimensions of Js = Nms, the unit of mass
is kg, that of charge C (Coulomb) and 4
0
, the permittivity of the
vacuum, carries the unit C
2
/(Nm
2
). Thus the quantity a
0
=
4
0
h
2
mee
2
carries the unit of length and is the typical length-scale of the Hamil-
tonian it is called the Bohr radius (in SI-units). Using the values
h = 1.0546 10
34
Js, m
e
= 9.1095 10
31
kg, e = 1.6022 10
19
C
and (4
0
)
1
= 8.9875 10
9
Nm
2
/C
2
, we nd that the Bohr radius is
a
0
= 0.529 10
10
m ( 1/2

A). Indeed, in a solid the typical distance
between atoms is a couple of Bohr radii.
There is also a characteristic energy-scale. This so-called atomic unit
for energy is the Rydberg, where Ry =
e
4
m
2h
2
1
(4
0
)
2
= 13.6eV. The Hamil-
tonian (4.1) expressed in atomic units length in terms of the Bohr radius
and energy in Rydbergs (thus replacing r and R by a
0
r and a
0
R) is
H
solid
=

n
m
e
M
n

2
n

2
i
+ 2

i<j
1
|r
i
r
j
|
+ 2

l<n
Z
l
Z
N
|R
l
R
n
|
2

i,n
Z
n
|r
i
R
n
|
. (4.2)
The only parameters that we cannot eliminate by scale transforma-
tions are the mass ratio of the electrons and the nuclei
m
Mn
and the nu-
clear charge Z
n
. It is fascinating to see that there are so few parameters
that give rise to a wealth of solids with very diverse properties.
The inuence of atomic number Z
n
on the properties of solids is, as in
atoms, bizarre (but regular). In both situations the underlying reason
is the same. The shell structure of atoms that depends on the nuclear
charge has strong repercussions for chemistry and, as we will see later,
on solid state physics.
The consequences of the mass ratio m
e
/M
n
are easier to understand,
because this ratio is almost always very small, typically of the order of
10
3
10
6
. Therefore its qualitative eect does not depend much on
its exact value. Obviously, the mass ratio is a very natural expansion
parameter, which leads to a major simplication of the theory: we can
treat the rst term in the Hamiltonian (4.2) in perturbation theory. So
we divide the Hamiltonian in an unperturbed part
H
adb
=

2
i
+ 2

i<j
1
|r
i
r
j
|
+ 2

l<n
Z
l
Z
N
|R
l
R
n
|
2

i,n
Z
n
|r
i
R
n
|
,
Phonons 49
and a perturbing part, which is the kinetic energy of the nuclei:
T =

n
m
e
M
n

2
n
. (4.3)
In the unperturbed, adiabatic Hamiltonian H
adb
the nuclear coordinates
only appear as a set of xed parameters: nuclear positions are con-
stants of motion of H
adb
. The rst two terms in adiabatic Hamiltoni-
ans describe the dynamics of the electronic degrees of freedom. The
two terms correspond to the electronic kinetic energy and the electron-
electron Coulomb interaction, respectively. The third term represents
the Coulomb interactions between nuclei. As the nuclear positions enter
the adiabatic Hamiltonian only as parameters, this term in the Hamil-
tonian is static: it is just a number. The last term in H
adb
species the
Coulomb interaction between electrons and nuclei. This term is nothing
but the static potential landscape generated by the nuclei, in which the
electrons live.
This observation has two important consequences. Firstly, it makes
clear why a solid can have a spatial, or crystal, structure at all: in zeroth
order approximation in the mass ratio m
e
/M
n
we have H
solid
= H
adb
and thus the nuclear coordinates are constants of motion also of H
solid
.
Secondly, it provides us with a relatively simple prescription of how to
nd that structure: we have to minimize the ground state energy for
the electrons with respect to the nuclear positions. So we conclude that
matter has a structure due to the smallness of the ratio of electronic
and nuclear masses. This structure can in principle be determined (to
zeroth order in the mass ratio) by solving the electronic problem for xed
nuclear positions (many times) and minimizing the groundstate energy
with respect to the nuclear positions.
As a next step we want to establish what the inuence of a nite
kinetic energy of the nuclei is (in rst order) on the dynamics of the
nuclei in the lattice. Thus we begin to study the phonons, which are the
elementary excitations of the nuclear subsystem.
We start by determining the characteristic velocity of nuclear motion
with the help of a simple, classical argument. We consider the forces
that act on the electrons and the nuclei. As the system is at rest the
total force must be zero: the force that the electrons and nuclei exert on
each other must be of opposite sign and equal strength, so M

R = mr.
This tells us that the acceleration of the nuclei, (and thus their velocity)
is lower than the acceleration (and velocity) of the electrons by a factor
m/M. Again, this is the reason to consider the adiabatic approximation:
at every moment in time the fast electrons adjust adiabatically to the
time dependent potential of by the slowly moving nuclei. Thus at all
50
times the electronic system stays, to good approximation, in the ground
state wave function that is determined by the nuclear positions at that
particular moment in time.
The electrons follow the lattice deformations adiabatically. If the
nucleus moves away from its equilibrium position R
0
, a restoring force
drives it back. In atomic units we have

RM/m (RR
0
) for the cor-
responding equation of motion. The frequency of this harmonic oscillator
is
_
m/M, which sets the energy scale for phonon excitations, which
is typically of the order of room temperature or lower (k
B
T 25 meV ).
Now we wish to calculate the typical amplitude of the nuclear motions
and extend our arguments to the quantum mechanical case. For the
quantum harmonic oscillator we know that its zero point motion related
to the kinetic part
m
2M
P
2
and the potential part of the Hamiltonian
1
2
(RR
0
)
2

1
2
u
2
, where u is the displacement of an atom with respect
to its equilibrium lattice position:
R =
_
(RR
0
)
2
=
_
u
2

_
m
M
_
1/4
, (4.4)
P
_
m
M
_
1/4
and

R
_
m
M
_
3/4
. (4.5)
Note the amplitude R of the zero point motion only scales with
(m/M)
1/4
. This make it understandable that under certain conditions,
quantum uctuations can prevent the formation of a crystal. Indeed
there are two systems that stay liquid even at zero temperatures;
3
He
and
4
He (although not at high pressures).
When a systematic expansion away from the adiabatic limit is per-
formed the expansion parameter is not the ratio m/M, but the much
larger number (m/M)
1/4
(although still a small number in most cases).
When we take into consideration the equations above, this hardly comes
as a surprise.
End 1st half lecture 5
Exercise 4.1
Show that due to the zero point motion of the nuclei, the uctuations in
position, velocity and momentum of nuclei scale with the mass parame-
ters as in equations (4.4) and (4.5).
2. Lattice waves: the discrete elastic line
We now consider a chain that consists of atoms with harmonic inter-
actions. In a cartoon the linear chain looks like a set of masses connected
Phonons 51
M M M M M
a
n
u
n+1
u
Figure 4.1. Linear chain of masses (M) connected by springs. The the distance
between masses labeled by n and m is a|n m|.
with springs, where all masses are equal to M. We label the atoms by
the number n and denote a deviation from the equilibrium position by
u
n
. The kinetic energy of an atom is given by
p
2
n
2M
, and the elastic energy
stored in a spring is B(u
n
u
n+a
)
2
, where B is the spring constant. Its
Hamiltonian is
H =

n
p
2
n
2M
+
B
2
(u
n
u
n+1
)
2
. (4.6)
The corresponding equations of motions are u
n
=
H
pn
and p
n
=
H
un
so that
M u
n
= B(2u
n
u
n+1
u
n1
). (4.7)
A Fourier transform of the displacements (u
n
=

q
u
q
e
iqan
) gives
M u
q
= B(2 e
iqa
e
iqa
)u
q
= 2B(1 cos qa)u
q
, (4.8)
which is the equation of motion of a harmonic oscillator with the fre-
quency
q
=
_
2B(1 cos qa)/M = 2
_
B/M| sin
qa
2
|. The long wave-
length excitations, where q 0, carry an energy
q

_
B/M|qa|.
Next we consider a linear chain with two atoms per unit cell, i.e. with
two masses M
1
and M
2
.
u v v
n n+1 n
u
n+1
M M M M M
1 1 2 2 2
Figure 4.2. Linear chain with two dierent masses M1 and M2 connected by springs.
The the distance between two masses n, m of the same kind is 2a|n m|.
Its Hamiltonian is

n
u
2
n
2M
1
+
v
2
n
2M
2
+
B
2
_
(u
n
v
n
)
2
+ (u
n1
v
n
)
2
_
(4.9)
52

----
2
-

----
2
-
- qa

----
2
-

----
2
qa

----
2
-

----
2
qa

C
Figure 4.3. Phonon dispersion for the homogenous linear chain in the large (A) and
reduced (B) Brioullin-zone. The dispersion for the chain with two dierent masses is
shown in (C).
and the equations of motion are
m
1
u
n
= u
n
v
n
+ u
n
v
n+1
= 2u
n
v
n
v
n+1
(4.10)
m
2
v
n
= (u
n
v
n
) (u
n1
v
n
) = 2v
n
u
n
u
n1
,
where we introduced the rescaled mass m
1
= M
1
/B and similar for m
2
.
The periodicity of the chain is 2a, instead of the a in the chain with
equal masses). After a Fourier transformation we nd
m
1
u
q
= 2u
q
v
q
(1 + e
i2qa
) (4.11)
m
2
v
q
= 2v
q
u
q
(1 + e
i2qa
),
where /2a < q /2a. We now make the Ansatz u
q
= ue
it
and
v
q
= ve
it
, so that
m
1

2
u = 2u v(1 + e
i2qa
) (4.12)
m
2

2
v = 2v u(1 + e
i2qa
)
which is, in matrix form,
_
2 m
1

2
1 + e
i2qa
1 + e
i2qa
2 m
2

2
_ _
u
v
_
= 0.
Phonons 53
This equation is equivalent to demanding that zero is an eigenvalue of
the matrix. This in turn requires that the determinant of the matrix
vanishes:
(2 m
1

2
)(2 m
2

2
) 2(1 + cos 2qa) = 0, (4.13)
which, with the use of 1 + cos 2x = 2 cos
2
x reduces to
m
1
m
2

4
2(m
1
+ m
2
)
2
+ 2(2 2 cos
2
qa) = 0 (4.14)
and has the two solutions
+
(q) and

(q), with

=
m
1
+ m
2
m
1
m
2
_
1

1
4m
1
m
2
(m
1
+ m
2
)
2
sin
2
qa
_
. (4.15)
When q is small we have
_
1 sin
2
qa 1

2
(qa)
2
, so that

= 2
m
1
+ m
2
m
1
m
2
m
1
m
2
(m
1
+ m
2
)
2
|qa|
2
=
2
m
1
+ m
2
|qa|
2
(4.16)
This is the acoustic phonon branch. In the limit that both masses are
equal we have

=
_
B/M|qa| the result for the homogeneous chain.
More importantly a new phonon mode appears: the optical phonon
branch
+
. The optical mode has a gap: at small q we have
+
=
2B
M
1
+M
2
M
1
M
2
+ O(q
2
). At the edge of the Brillouin zone the dierence in
energy between that acoustical and optical phonon branch is
2
M
1
+ M
2
M
1
M
2
B

1
2M
1
M
2
(M
1
+ M
2
)
2
. (4.17)
Exercise 4.2
Calculate the energy of the (classical) lattice waves for a one dimen-
sional system where the masses are equal, but the spring-constants are
alternating (B
1
and B
2
). Derive the expressions for the dispersion of the
optical and acoustical modes for small wave number q, up to rst order
in q. Sketch the two branches.
1 1 2 2 1
B B B B B
Figure 4.4. Linear chain with two dierent spring constants B1 and B2.
54
3. The elastic line in the continuum limit
When we are interested in the low energy or long wavelength prop-
erties of lattice waves, and we are dealing with length-scales that are
much larger than the inter-atomic spacing, the lattice can be treated as
a continuum. When we want to study how waves break at the beach,
we do not start with investigating the dynamics of water on the atomic
length scale.
We nd the continuum limit by letting the lattice constant go to zero,
while properly re-scaling physical quantities. For the Hamiltonian of the
discrete elastic line, we nd
H =

n
p
2
n
2M
+
Ba
2
2
_
u
n
u
n+a
a
_
2
(4.18)
and thus for the Hamilton density
H/a =

n
p
2
n
2Ma
+
Ba
2
_
u
n
u
n+a
a
_
2
(4.19)
which reduces for a 0 to the continuum Hamiltonian density
H =
1
2

2
+
1
2
T
_
u
x
_
2
, (4.20)
where u(x) is the displacement eld, (x) its conjugate momentum eld
and = M/a is the mass density and T = Ba the tension. The equation
of motion immediately reduces to

2
u
t
2
= T

2
u
x
2
, (4.21)
which is a well-known wave equation with solutions of the form u(x, t) =
u(k, )e
i(kxt)
, where u(k, ) = u
k
is the amplitude of the wave with
wavenumber k and frequency . From the wave equation we see that

2
= T k
2
and thus =
_
T

|k|, which is linear in momentum |k|, as

we expected from the dispersion relation that we derived for the discrete
elastic line in the long wavelength limit.
The partition function Z is easily evaluated, using in the Hamil-
tonian that = u(x, t), which follows directly from the equations
of motion. The eigenmodes with wavenumber k are independent and
thus the partition function can be written as a product of these modes:
Z = Tr e
H
=
k
Z
k
, with
Z
k
= Tr e
T k
2
|u
k
|
2
=
_
du
k
e

2
k
|u
k
|
2
=
1

k
B
T

. (4.22)
Phonons 55
End 2nd half lecture 5
Exercise 4.3
Consider an external force density f(x, t) acting on the elastic line, so
that one has to add to the Hamiltonian density
H
ext
= u(x, t)f(x, t). (4.23)
4.3.0 Show that the response function of the elastic medium is
(q, ) =
1/

2
q
. (4.24)
This wavelength and frequency dependent susceptibility is by denition
(q, ) = u(q, )/f(q, ). Use the equations of motion and functional
derivatives (as in Excercise 3.5).
4.3.1 Dissipation can be introduced by adding a phenomenological term
to the Hamiltonian that describes damping of the lattice waves:
H
dis
= v(x, t), (4.25)
where is a viscosity and v(x, t) = u(x, t) is the velocity of a mass
point. Give an argument why the dissipation has to be proportional to
the gradient of the velocity (and not the velocity itself).
4.3.2 Show that in this case the response function is
(q, ) =
1/

2
q
+ i
, (4.26)
with = q
2
/. Sketch the imaginary part of the response function as
a function of q and .
4. Elasticity in a jellium: plasma oscillations
In the discrete chain we assumed that the forces on a nucleus result
from only the neighboring nuclei. However, in general, interactions can
also be long range, especially in charged systems, where the Coulomb in-
teraction comes into play. System with long range interaction have qual-
itatively dierent behavior. As an example we consider an electron gas in
a charge neutral solid where the total charge of the electrons is compen-
sated by the background change of the nuclei. We assume that nuclear
background charge is uniformly smeared out throughout the volume of
material under consideration, the so-called jellium approximation. In
its groundstate also the electron gas is homogeneously distributed over
the volume. Thus in the equilibrium situation the nuclear background
56
charge precisely compensates the electronic charge everywhere: in its
groundstate the system is both globally and locally charge neutral.
Now we consider variations of the electron density, which are ele-
mentary excitations of the electron gas, as waves in an elastic medium.
The only qualitative dierence with lattice waves as we discussed them
above, is that the interaction between charges, the Coulomb interaction,
is always long ranged. A perturbation in the electron density at some
locus in space causes a change in potential in a region that is far away
from that locus. In this region far away, in turn, the change in potential
causes a change in electron density. This is very dierent for the case of
lattice waves where is is reasonable to assume that the consequence of a
change in position of a certain atom is only a force on its the neighboring
atoms.
First we Fourier transform the Hamiltonian density of the elastic line,
introducing (x) =
1

L
_
p
k
e
ikx
dk and u
x
=
1

L
_
u
k
e
ikx
dk we obtain
H =
_
dkH
k
, with H
k
=
1
2
p
k
p
k
+
1
2
Tk
2
u
k
u
k
(4.27)
and thus the equation of motion
u
k
=
T

k
2
u
k

k
=

|k|. (4.28)
We now consider a jellium where the restoring force to density varia-
tions is due to long-range Coulomb interactions in addition to the elastic
tension T. In the groundstate the jellium is globally and locally charge
neutral. The density of electrons is n|e| and the nuclei give rise to a
rigid background charge density n|e|; equal and opposite to the aver-
age density of electrons. A density oscillation of the electrons causes the
charge in some region to accumulate and in another region to decrease.
The induced change in charge density (x) at position x is directly
proportional to the local compression/dilation at this position in space:
(x) = n|e|
u
x
x
. (4.29)
The charge density (x) is subject to the local potential (x), and we
have to add a term to the Hamiltonian density
H
coul
=
1
2
(x)(x). (4.30)
The factor 1/2 enters because the electrostatic term is just the self-
energy of the electron gas. The electrostatic potential (x) in turn, is
Phonons 57
k

ne

4
-------

k

T
-----

Figure 4.5. Dispersion of plasma oscillation (full line) and phonon (dashed line).
The plasma gap is indicated.
determined by the Poisson equation

2
(x) = 4(x), (4.31)
which, introducing the Fourier transform of the potential, is

2
(x) =

2
1

L
_

k
e
ikx
dk =
1

L
_
k
2

k
e
ikx
dk. (4.32)
The accumulated charge density (x) is directly related to the Fourier
transform of the displacement-eld as
(x) = n|e|
u(x)
x
= n|e|
1

L
_
iku
k
e
ikx
dk (4.33)
so that we nd from the Poisson equation the relation between the
Fourier coecients of the potential and displacement eld

k
=
4in|e|
k
u
k
. (4.34)
Thus the Fourier components of Coulomb term in the Hamiltonian re-
duce to
H
coul
k
=
n|e|ik
2

k
u
k
=
4n
2
e
2
2
u
k
u
k
(4.35)
and the Hamiltonian density for elastic waves in a jellium nally reduces
to
H
jellium
k
=
1
2
p
k
p
k
+
4n
2
e
2
+ Tk
2
2
u
k
u
k
(4.36)
58
so that the energy of an electron density oscillation with wave vector k
is

k
=

4n
2
e
2
+ Tk
2

. (4.37)
When k 0 the excitation energy stays nite: there is a gap in
the excitation spectrum. The gap is due to the long range character
of Coulomb interactions. The excitation of the electron gas that we de-
scribed here is the so-called plasma oscillation: it is a longitudinal charge
density oscillation. Plasma oscillations are encountered in virtually all
solid state systems, with energies varying from about 3 to 20 eV .
5. Plasma oscillations in lower dimensions
In the previous section we used in the Poisson equation

2
= 4
the derivative

2
=

2
r
2
. Note that this is only justied for one-dimensional
systems. In this case the potential generated by a charge q at the origin
is determined by

2
(r) = 4q
r,0
, which has the solution (r) r
(plus some constant). But what does it mean that our system is one-
dimensional? If we consider an electron gas conned to a one dimen-
sional geometry, the electric eld and potentials of the electrons on the
line extend into the whole of the three-dimensional space, see Fig. 4.6.
Thus in order to calculate the potential of, say, one electron on the line
e e e e e e e e
Figure 4.6. Schematic representation of a one-dimensional electron gas in a three-
dimensional space.
we have to solve the 3D Laplace equation where the radial part is
1
r
2

r
_
r
2

r
_
(r) = 4e
r,0
, (4.38)
with the well known solution (r) 1/r, plus some constant determined
by the boundary conditions, as usual. (The relation

2 1
|r|
= 4(r),
can easily be derived, taking a proper limiting procedure for r 0).
Clearly the plasma oscillations that we discussed before are not related
to the oscillations of an electron gas in a one-dimensional connement.
Phonons 59
z
x
y
Figure 4.7. Plasma oscillation in a homogenous three-dimensional system: sheets of
charge are oscillating.
The proper interpretation of the use of the one dimensional Laplacian

2
r
2
is that we considered a system with only one spatial degree of freedom:
the potential is assumed only to depend on one space coordinate r (or,
equivalently x); the potential does not depend on the other two space
coordinates (that always exist in our 3D world) say, y and z. The one-
dimensional geometry that we considered therefore looks like a capacitor,
where the capacitor plates have an innite extend, see gure 4.7.
It is important to note that if we would have considered a an electron
gas in a one-dimensional connement the Coulomb interaction would
still be long-ranged (decaying with inverse distance), but not long-range
enough to cause a gap in the plasma oscillation spectrum. We elaborate
on this point by considering the dispersion of plasma oscillations in a
two-dimensional system. The one-dimensional case we will leave as an
exercise. Thus the excitations we are looking for are those of a two
dimensional sheet of electrons (with a compensating background charge),
see gure 4.8.
z
x
y
Figure 4.8. Plasma oscillation in a two-dimensional electron gas.
Now the potential depends on the spatial coordinates x and z, as
we investigate oscillations that are translational invariant along the y-
axis. A general displacement in the plane is now a given by the vector
u = (u
x
, u
y
). This implies that the strain eld has four components:
ux
x
,
ux
y
,
uy
x
,
uy
y
. When e.g.
ux
x
is non-zero, the atoms are displaced along
(longitudinal to) the x-axis. For non-zero
uy
x
, the displacements are
perpendicular (transversal) to the x-axis: We consider the longitudinal
60
k
k
z
x
y
Figure 4.9. Plasma oscillation in a stack of two-dimensional planes. Only charge
oscillations within the plane are allowed.
wave, with (x, y) = ne
ux
x
so that the compression/expansion of the
charge density takes place along the x-direction; along the y-direction
there is no charge displacement at all. Thus the potential that is due
to the charge displacement is translational invariant in the spatial y-
direction and thus only depends on the spatial x and z-coordinates:
(x, y, z) = (x, z).
We now consider the Fourier components of the displacement eld
u
x
u
kx
e
ikxx
and electrical potential (x, z)
kx,kz
e
i(kxx+kzz)
, where
k
x
/k
z
is the momentum parallel/perpendicular to the plane, respectively.
Let us explicitly keep the k
z
momentum dependence of the potential: and
thus implicitly assume that we have some kind of translational invariance
of the potential along the z-axis, giving rise to a dependence of the
plasma oscillations on the wave-vector parallel to plane. At the same
time we do not allow the possibility of the displacement of charge along
the z-direction (otherwise we would have to take the dependence of u on
k
z
into account). Such a constrained type of translational invariance is
present in real materials when the system consists of a set of planes and
no possibility for charge transfer between the planes, see gure 4.9 This
is to good approximation the case for the High-T
c
superconductors. In
this case we have for the Fourier components of the potential
_

2
x
2
+

2
z
2
_
(x, z) (k
2
x
+ k
2
z
)
kx,kz
e
i(kxx+kzz)
(4.39)
and for the Fourier components of the gradient of the displacement eld
are
u
x
x
ik
x
u
kx
e
i(kxx+kzz)
(4.40)
so that we nd for the Fourier coecients of the potential from the
Poisson equation

kx,kz
= 4ne
ik
x
k
2
x
+k
2
z
u
kx
. (4.41)
Phonons 61
k

- 0
Figure 4.10. Dispersion of plasma oscillations in a stack of planes. The momentum
in the plane is k

and perpendicular to the plane k

.
Note that this expression reduces to the one-dimensional solution of the
Poisson equation under the constraint that the potential is independent
of the spatial z-coordinate, so that k
z
= 0. Such a constraint would
obviously lower the symmetry of the problem from 2D to 1D.
Using this relation in the Coulomb part of the Hamiltonian for the
two-dimensional jellium we nd
H
Coul
k
=
1
2
4n
2
e
2
k
2
x
k
2
x
+ k
2
z
u
kx
u
kx
, (4.42)
which gives the dispersion relation for two-dimensional plasma oscilla-
tions as
(k

, k) =

_
Tk
2

+
4n
2
e
2

k
2

k
2

+ k
2

, (4.43)
where k

= k
x
is the momentum parallel to the plane (in plane) and
k

= k
z
the momentum perpendicular to the plane (out of plane). The
dispersion relation gives gapless modes when k

0. Thus even for

a set of planes, where there is a dependence of the potential on the z-
coordinate (but no charge displacement along the z-direction) and the
system is eectively two-dimensional (or three-dimensional but transla-
tional invariant in one direction) the Coulomb potential is eectively not
long range enough to create a gap in the plasma oscillations spectrum
for (very) long wavelength excitations.
End 1st half lecture 6
Exercise 4.4
Show that in a single two dimensional sheet of electrons, the plasma
frequency for small momentum k is proportional to

k.
62
Hint: Use that the potential at z = 0 is
_

kx,kz
dk
z

_
k
2
x
k
2
x
+k
2
z
dk
z
and
_
dx
X
=
2

arctan
2ax+b

when > 0 and where X = ax

2
+ bx + c;
= 4ac b
2
.
6. Elastic medium in d > 1
In the previous section we saw that in the 2D system longitudinal
and transversal stress elds appear. The longitudinal waves are related
to local compression and expansion of the solid, and are, in fact not
characteristic of solids at all: after all air supports sound-waves.
Transversal waves are more characteristic of a solid as these waves are
due to the fact that a solid opposes shear strains (in contrast to a liquid
or a gas). Solids are characterized by shear rigidity and this is why your
toes hurts when you kick it. Transversal waves are also well known from
the theory of electromagnetism. In this case photons are the transversal
waves (of which there are two kinds, the left and right circularly polarized
photon). In fact there is a deep and interesting relation between the
theory of (quantum) elasticity and (quantum) electrodynamics.
We consider a solid that is elastically isotropic. We look for invariants
1
that are quadratic in the strain components
u
x
, where u

= u
x
, u
y
or u
z
and x

= x, y, or z. We abbreviate this strain component as

. One such invariant is (

u)
2
=

, where summation
over repeated indices is implied. Another invariant is |

u|
2
. The
deformation associated with this invariant is a rotation, not a strain, so
we do not include this term into the elastic energy. The other invariant
is

= (

)
2
. So we obtain a Hamiltonian density:
H =
1
2

+
1
2

+
1
2

(4.44)
where the last term in the strain tensor gives rise to shear rigidity. Let
us consider the Fourier components of u

(x), the terms u

(k)e
ikx
. We
have

(k)e
ikx
= k

(k)e
ikx
. Using this we nd the Hamiltonian
in Fourier space
H =
1
2

+

2
k

+

2
k
2

, (4.45)
1
An invariant in the present context is a quantitity that is translational invariant, i.e. does
not depend on the absolute position in the lattice, or equivalently, does not depend on the
choice of the origin for the axes.
Phonons 63
x
y
du
x
----------
dx
,0
du
x
----------
dy
=0
x
y
du
x
----------
dx
=0
du
x
----------
dy
,0
Figure 4.11. Longitudinal (left) and transversal wave (right).
where the labels k of the displacement and conjugate momentum elds
are suppressed. The equations of motion are

=
H
u

= k

+ k
2

and u

=
H

, (4.46)
which results in the wave equation
u

= k

+ k
2

(4.47)
for each component the vector eld u = (u
x
, u
y
, u
z
). In vector notation
the wave equation can be written a more compact form:

2
u
t
2
=

u) +

2
u. (4.48)
We are now in the position to break up the displacement u into a
longitudinal and (two) transversal components. For the longitudinal
displacement the direction of propagation of the displacement eld is
parallel to the direction of displacement, so that

u
L
= 0. For
the transversal waves we have

u
T
= 0. We use the vector identity

2
u =

(

u)

(

u) and get the wave equation for the

longitudinal waves

2
u
L
t
2
= ( + )

2
u
L
(4.49)
and for the transversal waves

2
u
T
t
2
=

2
u
T
. (4.50)
It follow immediately that the longitudinal and transversal wave veloci-
ties (sound velocities) are
v
L
=

and v
T
=

. (4.51)
64
The longitudinal propagation velocity is therefore always larger than
the transversal velocity in an isotropic elastic medium. The appearance
of transverse elastic waves is formally identical to the appearance of
transverse electromagnetic waves in the context of electromagnetism:
the photon (of either one of the two polarizations). We can identify u
T
with the vector potential A and

u
T
with

A = B, the magnetic
eld. The mass density translates into the permittivity of the vacuum

0
and the transversal wave velocity v
T
=
_
/ into the velocity of light
c = 1/

0
.
This analogy also explains the absence of longitudinal photon modes
in electromagnetism: we can always choose the gauge

A = 0 (the
transverse gauge), where longitudinal modes are absent, which corre-
sponds to an elastic theory with

u = 0, so that u = u
T
and u
L
= 0.
If we can get rid of longitudinal photons by choosing a certain gauge,
these modes should be absent in general if we are working in a gauge
invariant theory (like electromagnetism). When gauge invariance is bro-
ken (e.g. in a superconductor), the longitudinal mode cannot be gauged
away and becomes physically relevant.
Exercise 4.5
Show that the equation
u

= k

+ k
2

can be rewritten in vector rotation as

2
u
t
2
=

u) +

2
u.
7. Thermodynamics of the lattice waves
In this section we consider the lattice contribution to the specic
heat. The most important message is that one needs quantum statistics
to explain the experimentally observed low temperature behavior of the
specic heat in solids. This was a strong indication in favor of quantum
theory in its early days besides Plancks radiation law.
We consider the contribution to the specic heat of of a classical
lattice wave with wave vector k. The amplitude of the wave is u
k
and
its energy
k
. The contribution to the partition function of modes with
wave vector k, but dierent amplitudes is
Z
k
=
_

du
k
e

2
k
u
2
k
=

2
k
, (4.52)
Phonons 65
which is just a Gaussian integral of equation 4.22. The internal energy
is by denition
E =
1
Z
Z

=
lnZ

(4.53)
and the specic heat at xed volume is
c
V
=
E
T
=
2
E

. (4.54)
The contribution to the specic heat of the wave with wave vector k is
easily evaluated, introducing x =

2
k

, we obtain
c
V
=
2

2
ln Z

2
=
x
2
2

2
ln x
x
2
=
1
2
. (4.55)
Actually what we did here is derive the equipartition theorem: the con-
tribution to the internal energy of a particular classical degree of free-
dom (in this case a lattice wave with wave vector k) is k
B
T/2. Note the
specic heat in this case does not depend on temperature nor the dis-
persion relation of the classical wave. The number of degrees of freedom
is nothing but the number of possible values of k, which is equal to the
number of lattice sites N. Actually the waves have another degree of
freedom: their polarization and parity. In a cubic 3D system there are
six directions (and 3 dierent polarizations), so that the specic heat of
a classical crystal is
c
V
= 3Nk
B
Dulong Petit. (4.56)
This result is very robust and known as the Dulong-Petit law. Exper-
imentally, however, the specic heat strongly depends on temperature.
At high temperatures the Dulong-Petit law is obeyed, but at low T the
specic heat vanishes. To explain this one needs quantum mechanics.
Suppose now that our lattice waves are bosons (which they are) and
that we have just a single, non-dispersive lattice mode with energy
0
the Einstein model. The total energy is now
E =

k
n
k

k
=
3N
0
e

0
1
(4.57)
and from equation 4.54 the specic heat is
c
v
= 3N
2

2
0
e

0
(e

0
1)
2
Einstein model, (4.58)
66
T
c
V
Dulong - Petit
e
-T e
Figure 4.12. Specic heat for the Einstein model.
which indeed for high temperatures ( 0) reproduces the Dulong-Petit
law, see Fig. 4.12. But at low temperatures c
V
now vanishes exponen-
tially
c
V
3N
_

E
T
_
2
e

E
/T
T 0, (4.59)
where we introduce the Einstein temperature k
B

E
= h
0
.
But besides lattice waves there are also other degrees of freedom in a
solid: electrons. When one supposes these obey Bose statistics as well,
they would contribute to the Dulong-Petit law. So the high T specic
heat of solids would depend strongly on the number of electrons per until
cell. Experimentally this is certainly not the case. The reason is that
electrons obey Fermi statistics. The Pauli principle for fermions does
not allow electrons that are deep in the Fermi sea to contribute to any
physical property at moderate (low) temperatures. The 1s electron for
Cu for instance, is about 6 10
3
eV lower in energy than the 3d electrons
that form the valence- and conduction band. For the 1s electron to
contribute to the specic heat we thus need a temperature in the order
of 10
7
K, which is about the temperature of the sun.
End 2nd half lecture 6
Exercise 4.6 Debije model.
4.6.0 Calculate the specic heat for a system with acoustical phonons,

k
= c|k| assuming quantum (Bose) statistics.
Remember that there is a natural cut-o wave vector k
D

1
a
, where a
is the lattice spacing, when doing integrations/summations. Write the
specic heat in terms of the Debije temperature
D
, where k
B

D
= ck
D
.
4.6.1 Show that at high temperatures (T
D
) the Dulong-Petit law is
recovered and for low temperatures
c
v

_
T

D
_
3
.
Phonons 67
Thus low temperature specic heat measurements pick up gapless phonon
modes and dierentiate between the Debije and Einstein model for phonons.
8. Quantized harmonic chain
We will quantize the harmonic chain with the Hamiltonian
H =

n
p
2
n
2M
+
B
2
(u
n
u
n+1
)
2
, (4.60)
where p
n
is the momentum and u
n
the deviation from equilibrium of
the mass M at position n. Quantum mechanics requires that [p
n
, u
n
] =
ih
n,m
. The Fourier transformation
p
k
=
1

n
e
ikrn
p
n
and u
k
=
1

n
e
ikrn
u
n
. (4.61)
of the Hamiltonian leads to
H =

k
p

k
p
k
2M
+ B[1 cos(ka)]u

k
u
k
(4.62)
and [p
k
, u
k
] = ih
k,

k
.
Remember that from the appendix we have for a harmonic oscillator
the Hamiltonian
H =
p
2
2M
+
B
2
u
2
the quantized oscillation amplitude and momentum
u =
_
1
2mh
(b
+
+ b) and p = i

mh
2
(b
+
b)
so that the Hamiltonian reduces to
H = h(b
+
b + 1/2), with =
_
B/M, (4.63)
where n
k
= b
+
k
b
k
is the number operator related to the operator b
+
k
that
creates a boson with wave vector k.
So in the quantum chain we for each wave vector k a quantized har-
monic oscillator equation with energy
k
= 2
_
B/M sin|
ka
2
|. The only
dierence with the classical chain is that now modes are quantized:
E
k
= h
k
(n
k
+
1
2
). From this we easily nd the partition function
of the harmonic crystal
Z =

n
k
e
h
k
(n
k
+1/2)
=

k
e
h
k
/2
1 e
h
k
(4.64)
68
and the free energy
F = T ln Z =
1
2

k
h
k
+ T

k
ln(1 e
h
k
)
= T

k
ln[2 sinh
h
k
2T
] (4.65)
where we used
ln(1 e
x
) = ln
_
2e
x/2
e
x/2
e
x/2
2
_
=
x
2
+ ln
_
2 sinh
x
2
_
. (4.66)
9. Symmetry breaking in the quantum chain
In the previous section we started with quantizing the deviations from
the equilibrium position, not the positions of the masses themselves. In
other words, we assumed that the translation symmetry of the system
was broken. Now our aim is to nd the elementary excitations of the
lattice without a priori breaking translation symmetry.
The way we go about this is to quantize momentum and position
from the very beginning, to introduce the Bogoliubov transformations
and then to extract the thin spectrum that is related to the dynamics
of the order state a whole.
We start with a Hamiltonian where potential between the masses is
harmonic, centered around a certain length a, see Fig 4.13,
H
0
=

i,j
p
2
i
2M
+
B
2
( x
i
x
j
a)
2
. (4.67)
The product terms involving powers of a can easily be extracted. We
use that

i,j
( x
i
x
j
) = 0 (when the sum is over all i.j) and nd that
the Hamiltonian is, apart from a constant,
H =

i
p
2
i
2M
+
B
2
( x
i
x
j
)
2
. (4.68)
We can already learn something here. Apparently the Hamiltonian for
the excitations of the system phonons is invariant for a uniform change
in the average distance between masses. Thus the excitation energies
are independent of the volume in the system, and the other way around:
by exciting harmonic phonons one does not change the volume of the
system. At higher temperatures more and more phonons get excited, but
as long as the system is harmonic the volume does not change. Thermal
Phonons 69
a
x
1
- x
2
V x
1
- x
2

Figure 4.13. Harmonic potential centered around a between two atoms.

expansion of a solid is must therefore caused by anharmonic eects such
as phonon-phonon interactions.
The momentum and position operators can be written in terms of
bosonic operators as
p
i
= iC

h
2
(b
+
i
b
i
); x =
1
C

h
2
(b
+
i
+ b
i
), (4.69)
so that the commutation relation [ p, x] = ih is fullled. C is a constant
that we are free to choose to our convenience (but is non-zero). We plug
these boson operators into the Hamiltonian and obtain
H =
h
2

i,j

C
2
2M
(b
+
i
b
i
)
2
+
B
2C
2
[(b
+
i
+ b
i
) (b
+
j
+ b
j
)]
2
=
h
2

i,j
_

C
2
2M
(b
+
i
b
i
)
2
+
B
C
2
(b
+
i
+ b
i
)
2

B
C
2
(b
+
i
+ b
i
)(b
+
j
+ b
j
)
_
. (4.70)
We choose C such that the terms with b
+
i
b
+
i
and b
i
b
i
cancel:
C
2
2M
=

B
C
2
C
2
=

2MB and the Hamiltonian reduces to

H =
h
4

2B
M

i,j
2(b
+
i
b
i
+ b
i
b
+
i
) (b
+
i
+ b
i
)(b
+
j
+ b
j
). (4.71)
A Fourier transform leads to (i and j nearest neighbors)
H =
h
2

2B
M

k
[2 cos(ka)] b
+
k
b
k

cos(ka)
2
(b
+
k
b
+
k
+ b
k
b
k
) + 1.
This Hamiltonian is still not diagonal as the last, anomalous terms
b
+
b
+
+ bb, create and annihilate two bosons at the same time. We can
70
get rid of these terms in the Hamiltonian by applying the Bogoliubov
transformation (see appendix).
End of 1st half lecture 7
We thus have
A
k
= (2 cos ka) and B
k
= cos ka, (4.72)
so that
k
=
_
A
2
k
B
2
k
= 2

1 cos ka = 2

2 sin |ka/2|. In the end we

obtain the Hamiltonian in terms of transformed boson operators
H = h

B
M

k
2 sin |ka/2|
_

+
k

k
+
1
2
_
+

2
2

2
4
(2 cos k)
= 2 h

B
M

k
sin|ka/2|
_
n
k
+
1
2
_
, (4.73)
since

k
cos k =
N
2
_

dk cos k = 0.
This is exactly the result that we obtained from quantizing the Fourier
transformed Hamiltonian for the displacements, as we did in the previous
section. However, the Bogoliubov transformation has the advantage that
it brings to the fore a rather subtle point. When k 0 the excitation
energy
k
0 and the two parameters in the Bogoliubov transforma-
tion diverge: u
k
and v
k
. Therefore precisely at k = 0 the
canonical transformation is not well dened. Therefore we have to in-
vestigate the bosonic terms in the Hamiltonian with k = 0 separately.
The zero momentum Hamiltonian is
H
k=0
=

2B
M
h
4
(2b
+
0
b
0
(b
+
0
b
+
0
+ b
0
b
0
) + 2)
=

2B
M
h
4
[(b
+
0
b
0
)
2
+ 1] (4.74)
where (b
+
0
b
0
)
2
=
2
hC
2
p

0
p
0
=
2
h

2MB
p

0
p
0
so that
H
k=0
=
1
2M
p

0
p
0
, with p
0
=
1

i
p
i
=
p
tot

N
, (4.75)
where p
tot
is the total momentum of the entire chain, or, equivalently,
its center of mass momentum. So the Hamiltonian for k = 0 is the thin
spectrum of the chain, apart from a constant. The symmetry breaking
Phonons 71
eld h couples to the center of mass of the chain x
tot
=

i
x
i
, so that
the Hamiltonian
H
thin
=
p
2
tot
2MN
+ hx
2
tot
(4.76)
describes the ordered state and its dynamics as whole.
By not quantizing the deviations from equilibrium position, but the
positions themselves, the thin spectrum emerges. Is this the only dier-
ence? Actually the there is another subtlety. In one of the next sections
we will show that in the one-dimensional system zero point uctuations
in the position of the masses destroy any ordering even at zero temper-
ature. A one dimensional lattice does simply not exist, and the concept
of equilibrium positions is not well-dened: a 1D solid is more like a
liquid. This is clear from the fact that in this case X-ray scattering does
not show Bragg reections, as we will see shortly.
So there is no long-range order and no breaking of translation symme-
try within the system (but there is spontaneous breaking of translation
symmetry of the system as a whole!). In spite of this lack of order, we
managed to nd all energies and wavefunctions of the excitations of this
1D system: the bosonic modes in equation 4.73 that loosely refered to
as phonons.
Exercise 4.7
Show that a Fourier transformation of
H =

n
p
2
n
2M
+
B
2
(u
n
u
n+1
)
2
with [p
n
, u
m
] = ih
n,m
(4.77)
results in
H =

k
p

k
p
k
2M
+ B(1 cos ka)u

k
u
k
with [p
k
, u
k
] = ih
k,

k
(4.78)
Exercise 4.8
Show that one can nd the coecients u = cosh and v = sinh of
the Bogoliubov transformation from the condition that the groundstate
energy is minimal.
Hint: collect the terms without operators in the transform of Ab
+
b =
B
2
(b
+
b
+
+ bb) and minimize with respect to .
Exercise 4.9
Show that the excitation energy =

A
2
B
2
of the Bogoliubov
Hamiltonian can be obtained from the equations of motion of the oper-
ators b
+
and b (
h
i
b
+
t
= [H, b
+
] and
h
i
b
t
= [H, b] ).
72
u
1
- u
2
Vu
1
-u
2

Figure 4.14. General, anharmonic potential V between two atoms. The displacement
from their equilibrium poistion is u1 u2.
Exercise 4.10
Show that the total energy of the harmonic lattice is
E(T) =

k
h
k
2
cth
h
k
2T
. (4.79)
Exercise 4.11
Show that translation symmetry in the one-dimensional quantum chain
is spontaneously broken.
10. Anharmonicity and thermal expansion
In the previous section we used an expansion of the potential between
nuclei for small displacements, i.e. for small deviations of the masses
from their equilibrium positions. We assumed a harmonic potential be-
tween the atoms: masses with coordinate x
1
and x
2
are subject to the
potential V(x
1
x
2
) = V
0
+
B
2
(u
1
u
2
)
2
, where x
1
= x
0
1
+ u
1
and
x
2
= x
0
2
+ u
2
. Of course also higher order, anharmonic terms exist.
Expanding the potential gives
V = V
0
+
B
2!
(u
1
u
2
)
2
+

3!
(u
1
u
2
)
3
+

4!
(u
1
u
2
)
4
+ ... (4.80)
where
B =

2
V(x)
x
2

x
0
1
x
0
2
; =

3
V(x)
x
3

x
0
1
x
0
2
, and
=

4
V(x)
x
4

x
0
1
x
0
2
. (4.81)
The term with coecient gives rise to the cubic anharmonicity,
quartic and so on. Usually the coecient < 0 and the potential looks
as shown in Fig. 4.14.
The anharmonic terms have important consequences. They lead to
a phonon-phonon interaction, as now terms with three and four boson
Phonons 73
k 2 / q
k 2 +q /
k
k

k+q
q
Figure 4.15. Phonon-phonon scattering processes due to anharmonicity of the lattice.
On the left a phonon decays into two other phonons, conserving momentum. The
righthand side corresponds to two phonons scattering o eachother.
operators appear in the Hamiltonian. These terms lead schematically
to the scattering processes shown in left panel of Fig. 4.16, where a
phonon with wave vector k decays into two phonons, one with momen-
tum k/2 + q and one with momentum k/2 q, as the total momentum is
conserved in the scattering process. The quartic term leads to diagrams
of the type shown in Fig. 4.16 on the right. The scattering probability
is given by and , respectively. Such phonon-phonon interactions are
important for thermal conductivity and essential for thermal expansion.
Thermal expansion. One can relate the coecient of thermal expan-
sion to the phonon contribution to the specic heat. To do so, we start
by considering the free energy of the harmonic crystal (equation 4.65)
F = T

k
ln[2 sinh

k
2T
]. (4.82)
In the anharmonic crystal the phonon frequencies
k
depend on the
specic volume V : we have
k
(V ). This can easily be understood when
we when realize that in the anharmonic crystal the curvature of the
potential V(x) between atoms,

2
V
x
2
, which determines the phonon fre-
quencies, depend on x (and therefore on the volume) in the anharmonic
crystal. Usually the dependence of phonon energy on volume is cast in
the relation
V

d
dV
=
d ln
d ln V
= , (4.83)
which is the Gr uneisen approximation and is the Gr uneisen constant
(which typically is 1 2 for ordinary crystals). The free energy
depends on a volume change V as
F(V ) =
1
2
_
V
V
_
2
+ T

k
ln[2 sinh

k
(V )
2T
]. (4.84)
The rst term is the elastic energy associated with the volume change,
where is the compressibility of the solid. Dierentiating the free energy
74
x
Ix
<x>=0
x
Ix
<x>,0
Figure 4.16. Left: phonon wavefunctions when the potential is harmonic: the average
displacement x is zero for all modes. Right: the same wavefunctions when the
potential in anharmonic: the average displacement increases with phonon energy.
Therefore anharmonicity gives rise to thermal expansion.
with respect to V and using the Gr uneisen approximation, we obtain
the following relation between the volume change and internal energy
E of the solid
1

V
V
=

k
2
cth

k
2T
= E(T), (4.85)
where we use equation 4.79. The thermal volume expansion coecient
is by denition =
1
V
V
T
, and the linear thermal expansion coecient
is =
1
l
l
T
, so that in cubic crystals the relation =
1
l
3
l
3
T
=
3
l
l
T
= 3
holds. With can now relate the volume expansion coecient to the
specic heat:
=
1
V
(V )
T
=
E(T)
T
= c
V
=
c
V
B
, (4.86)
where B
1

is the bulk modulus and c

V
the specic heat. Thus, we
obtain the so-called Gr uneisen equation
=
c
V
3B
. (4.87)
Why is anharmonicity necessary for thermal expansion? The physics
of the explanation is illustrated in gure 4.16. For the harmonic oscillator
the average position of all the wavefunctions is the center of the potential
well. If at higher temperatures phonons with higher and higher energy
are excited, the average distance between the atoms stays the same in
the harmonic crystal. In the anharmonic crystal the average distance
between the atoms is given by the average over all occupied excited
quantum levels in the asymmetric potential well. In this case the average
distance between atoms increases with temperature.
End of 2nd half lecture 7