Chapter 6

Lattice vibrations - Phonons

6.1 Phonons in mono-atomic crystals
How do we treat the motion of the ions within a crystal structure? Or in other words, how will the motion of
a bound ion inuence its immediate neighbours and indeed beyond? This is most easily visualised in a simple
cubic lattice along a direction of high symmetry - for example the [100] direction where we shall regard the
displacement of lattice planes :
u
n 1
u
n
u
n+1
u
n+2
u
n+3
n 1 n n +1 n +2 n +3
Figure 6.1: Schematic illustration of displacements of planes along one spatial direction. In the two orthogonal
directions translational invariance is assumed.
The above schematic can easily be understood as a longitudinal wave, but clearly transverse excitations can also
be envisaged, as shown below.
K
Figure 6.2: Schematic illustration of transverse displacements.
How can we calculate the dynamics? Assume elastic forces are linear and given by Hookes law -

F = kx,
where the vector x is the dierence of the displacements. The force resulting from the displacements u
n
, u
m
is
26
CHAPTER 6. LATTICE VIBRATIONS - PHONONS 27
then given by:

F
nm
= C
nm
(u
n
u
m
) (6.1)
For simplicity we shall only consider nearest neighbours, so we nd the total force acting upon an atom within
a plane:

F
n
= C (u
n+1
u
n
) + C (u
n1
u
n
) (6.2)
u
n 1
u
n
u
n+1
a a
Figure 6.3: The displacements for a periodic conguration of lattice planes is shown.
and C is assumed linear. Newtons second law then yields,
M
d
2
u
n
dt
2
= C [u
n+1
u
n1
2u
n
] (6.3)
the equation of motion for a plane (M is the mass of a atom in the plane). Assume all displacements have the
same frequency of dependence (normal modes) - exp(it):
M
2
u
n
= C [u
n+1
+ u
n1
2u
n
] (6.4)
This dierential equation can be solved by assuming a wavelike solution with wave vector

k and with an additional

phase factor which scales linearly with the plane position, i.e.
u
n
= uexp(inka) (6.5)
Here we are considering a simple cubic lattice with lattice constant a. This assumption simply means that a
xed phase relationship exists between any two neighbouring planes. Thus we nd:

2
Muexp(inka) = Cu[exp(i(n + 1)ka) + exp(i(n 1)ka) 2 exp(inka)] (6.6)
Which may be simplied by using the exponential denition of the cosine - exp(ix) + exp(ix) = 2 cos (x),

2
M = 2C [1 cos (ka)] (6.7)

2
=
4C
M
sin
ka
2
2
. (6.8)
Where we used the trigonometric relation cos (2) = cos
2
sin
2
= 1 2 sin
2
2 sin = 1 cos (2). This
is the dispersion relation for an elastic wave of wave vector k and of frequency .
CHAPTER 6. LATTICE VIBRATIONS - PHONONS 28
Figure 6.4: Illustration of the dispersion relation Equation 6.8, the maxima lies at (4C/M)
1/2
but has been normalised
in the above schematic. /a k /a
What range of k makes physical sense? The wave is only dened for a particular lattice plane located at nka.
The phase dierence between two plans can be written as
u
n+1
u
m
=
uexp(i(n + 1)ka)
uexp(inka)
= exp(ika) (6.9)
as expected.
A sensible range for the argument (ratio is periodic, 2i) is merely ka and hence /a k /a; i.e.
we restrict ourselves to solutions within the rst Brillouin zone of our reciprocal space, dened by the Wigner-
Seitz cell of our reciprocal lattice.
-2a -a 0
2a a
Figure 6.5: The above two waves clearly dier by a extra wavelength between adjacent planes.
The above two oscillations show the displacements for the lattice planes (separated by a) - no additional infor-
mation is to be gained by considering the smaller (or larger k). The wave vector

k outside of the rst Brillouin
zone can always be transformed back into the rst Brillouin zone by the addition of an appropriate reciprocal
lattice vector.

=

k +

K = exp(ink

a) = exp(inka) exp(in

Ka)
. .
=1
1
(6.10)
1
see chapter 3 - this is simply the denition of the reciprocal lattice considered earlier.
CHAPTER 6. LATTICE VIBRATIONS - PHONONS 29
where a is a lattice vector. The solution is a travelling wave everywhere except for k = /a. There the phase
factor between adjacent planes is exp(ika) = exp(i) = 1, therefore the motion of neighbouring planes is in
exact antiphase and we have a standing wave solution.
6.1.1 Velocity of solutions
The phase velocity is, as always, dened as
v
p
=

k
(6.11)
but we are more interested in the motion of the wave packet, i.e. in the group velocity
v
g
=

k
= grad

k
((

k) (6.12)
We now calculate the group velocity from Equation 6.8:
v
g
=
_
4C
M
_1
2
a
2
cos
_
ka
2
_
_
0 < k

a
_
v
g
=
_
Ca
2
M
_1
2
cos
_
ka
2
_
(6.13)
v
g,max
/a
direction of motion
reversed
zero, standing
wave solution
Figure 6.6: Illustration of the group velocity calculated in Equation 6.13. Near extrema the group velocity is roughly
speaking independent of the wave vector

k.
For small wave vectors

k we arrive at the long wavelength limit:

2
=
_
2C
M
_
[1 cos (ka)]
_
2C
M
__
1
_
1
1
2
k
2
a
2
+ ...
__

_
2C
M
1
2
k
2
a
2
_
=
_
C
M
_1
2
ka (6.14)
The velocity of sound is independent of frequency, this is obvious in the sense that when k goes to zero then ka
goes to zero, which means our crystal looks like a continuous medium.
How many modes are there? Every atom within the plane has essentially three degrees of freedom. There
are three modes of vibration - 2 transverse modes plus one longitudinal mode, and almost certainly they have
dierent C !
6.2 Phonons in diatomic crystals
So far we have considered a simple cubic structure (a lattice point is one atom). What happens when our
primitive unit cell contains more than one atom? If there are p atoms and N unit cells, then there are 3pN
CHAPTER 6. LATTICE VIBRATIONS - PHONONS 30
degrees of freedom. 3N degrees of freedom are contained in the modes we have just discussed and (3p 3)N in
some other modes. What is the nature of these additional modes?
Consider, for example, a diatomic chain of atoms with masses m
1
and m
2
. Clearly associated with each lattice
plane there is an atom of mass m
1
and an atom with mass m
2
.
m
1 m
1
m
2
m
2 m
2
s 1 s s s + 1 s + 1
a
Figure 6.7: Representation of a chain of atoms consisting of two dierent types of atoms with masses m
1
and m
2
.
However the displacements of the atoms from their equilibrium positions are, of course, considered separately.
Thus we nd two coupled dierential equations for the dynamics of the two atomic species:
m
1
d
2
n
s
dt
2
= C [v
s
+ v
s1
2u
s
] (6.15)
m
2
d
2
v
s
dt
2
= C [u
s+1
+ u
s
2v
s
]
where we assume, as before, that the wavemotion may be written as:
u
s
= u exp(iska) exp(it) (6.16)
v
s
= v exp(iska) exp(it). (6.17)
Inserting this into equation Equation 6.15 we nd:

2
m
1
u = Cv [1 + exp(ika)] 2Cu (6.18)

2
m
2
v = Cu[exp(ika) + 1] 2Cv. (6.19)
We nd the solutions by considering the following determinant (and requiring it to be 0):

2C m
1

2
C(1 + exp(ika))
C(1 + exp(ika)) 2C m
2

= 0 (6.20)
4C
2
2C(m
1
+ m
2
)
2
+ m
1
m
2

4
C
2
(1 + exp(ika))(1 + exp(ika)) = 0 (6.21)
The last term can be written as 2 + 2 cos (ka):
m
1
m
2

4
2C(m
1
+ m
2
)
2
+ 2C
2
(1 cos (ka)) = 0 (6.22)
This is a quadratic equation in
2
. So we can solve this equation for
2
and then take the square root for
= (k). Consider initially the limiting behaviour for k 0. As before we can expand cos (ka) 1 1/2k
2
a
2
:

4
m
1
m
2
2C(m
1
+ m
2
)
2
+ C
2
k
2
a
2
= 0 (6.23)

2
=
2C(m
1
+ m
2
)
_
4C
2
(m
1
m
2
)
2
4m
1
m
2
C
2
k
2
a
2
_1
2
2m
1
m
2

k (6.24)
One possible solution (at k = 0) is:

2
= 2C
_
m
1
+ m
2
m
1
m
2
_
= 2C
_
1
m
1
+
1
m
2
_
optical branch (6.25)
CHAPTER 6. LATTICE VIBRATIONS - PHONONS 31
The other possible root yields

2
=
1
2
k
2
a
2
C
(m
1
+ m
2
)
acoustic branch. (6.26)
and is also found for the mono-atomic chain.
What happens at the zone boundary at ka = ?
m
1
m
2

4
2C(m
1
+ m
2
)
2
+ 4C
2
= 0 (6.27)

4
2C
_
1
m
1
+
1
m
2
_

2
+
4C
2
m
1
m
2
= 0 (6.28)

2
=
2C
m
1
or
2
=
2C
m
2
(6.29)
If m
1
is larger than m
2
we can conclude
2
1
<
2
2
:
optical branch
acoustic branch
/a
2C
1
m1
+
1
m2
1
2
2C
m
2
1
2
2C
m
1
1
2
k
Figure 6.8: Dispersion relations for the diatomic chain.
In long wavelength limit (k 0) the acoustic and optical mode dier with regard to the direction of motion of
the atomic species as shown below:
Figure 6.9: For the acoustic mode, neighbouring atoms of diering species are displaced in the same direction. For the
optical mode the converse is true.
For the optical mode the behaviour can be mathematically derived by substituting
2
into Equation 6.18 or
CHAPTER 6. LATTICE VIBRATIONS - PHONONS 32
Equation 6.19 and solving for
u
v
=
m
2
m
1
. (6.30)
For optical modes the centre of mass is xed and the atoms oscillate against each other.
For 2C/m
1
<
2
< 2C/m
2
there are no solutions for the problem, k must be complex and therefore decay-
ing in space.
6.3 Phonon momentum
A lattice vibration is characterised by its frequency and its wave vector

k. What is the physical meaning of

k?
Is the wave vector

k related to the momentum ? No it is not, because the motion we are describing with

k is a
motion of the relative coordinates. The physical momentum is given by the following relation:
p = M
d
dt

u
s
with u
s
= uexp(iska) (6.31)
= M
du
dt

s
exp(iska) (6.32)
For N atoms the sum extends from s = 0 to s = N 1.

s
x
s
=
1 x
N
1 x
= M
du
dt
1 exp(iNka)
1 exp(ika)
(6.33)
The allowed values of k
2
are k = 2m/Na or k = m/N 2/a, where 2/Na is the spacing in k space as
before.
M
du
dt

1 exp(i2m)
1 exp(i2m/N)
= 0 (6.34)
We therefore conclude that phonons carry no real physical momentum but they do behave (in their interactions
with other particles) as if they carried a denite momentum

k. This quantity is known as the crystal momentum.

For the elastic scattering of photons we have

=

k +

K (6.35)
where

k is the initial state,

k

the nal state and |

k| = |

| and

K is a reciprocal lattice vector. We see that the
elastic scattering of a photon implicitly yields momentum conservation only if the crystal absorbs the momentum

K (recoil of the crystal).

General principal: The total wave vector is conserved in a periodic lattice but we can always add an ad-
ditional reciprocal lattice vector such that

k
ph
lies in the rst Brillouin zone:

k
ph
=

k +

K (6.36)
For example the above equation describes the inelastic scattering of a photon with the creation of a phonon
with crystal momentum

k
ph
. This is called phonon emission.
The process of phonon absorption may be written as,

=

k +

K +

k
ph
. (6.37)
The measurement of phonon dispersion relations (

k) require the simultaneous determination of phonon frequency

and its momentum

k
ph
, which can, by an appropriate choice

K (reciprocal lattice vector), always be placed into
the rst Brillouin zone.
As before, we write for phonon emission/absorption (+/):

k +

K =

k

k
ph
(6.38)
2
This is an applications of the BvK boundary conditions as considered earlier for a system with length L = Na: k = 2/L
mk = 2m/L with L = Na.
CHAPTER 6. LATTICE VIBRATIONS - PHONONS 33
In this case we need to determine the vectors

k and

k

, therefore we require not just the directions of

k and

k

but
also the diering magnitudes of

k and

. An apparatus which performs this analysis is known as a spectrometer.

The above discussion can be immediately extended to the case of inelastic neutron scattering. In this case
energy conservation then yields the frequency of the phonon:

2
k
2
2m
n
=

2
k
2
2m
n
(6.39)
where we have assumed that the energy of the phonon with frequency can be described, as for photons, by the
simple relation
= . (6.40)
Why this is so is considered in the next section?
6.4 Phonon modes and the Planck distribution function
The photons are the quanta of energy associated with a mode of frequency f = /2. Each photon has the
energy . The energy in a particular mode is given by (n+1/2), with /2 the zero point energy of the mode.
What are the modes? In electromagnetism they can be regarded as the modes compatible with the boundary
condition (E = 0) at the surface of some cavity (This picture leads to Plancks law for black-body radiation).
For the lattice we argue a mode is a lattice vibration of any frequency . How can n and the energy in the
mode be seen from a classical point of view to be related to each other? For a standing wave solution we have:
u = u
0
cos (ka) cos (t) (6.41)
For the energy of a harmonic oscillator we know that the energy oscillates between potential and kinetic energy:
<
1
2
p.E. >=<
1
2
k.E > (6.42)
Thus the time average of the kinetic energy is 1/2 the total energy. The kinetic energy can be written as
1/2
_
u
t
_
2
, with the mass density. We now consider a crystal with volume V and integrating over the volume
of the crystal and averaging over space
1
4
V
2
u
0
sin
2
(t) (6.43)
and now averaging over time yields:
1
8
V
2
u
2
0
=
1
2
_
n +
1
2
_
(6.44)
This equation is the connection between n the number of quanta in the mode and the classical amplitude.
Now we wish to consider the thermal properties of these phonon modes. The energies are as for a simple
harmonic oscillator:

n
=
_
n +
1
2
_
= s for simplicity (6.45)
Where n is the quantum number for the harmonics oscillator, but here we talk of the number of phonons (pho-
tons) in the mode. This picture has a physical basis in the sense that the motion of a particle in a potential is
localised about x
0
- 1/2 m
2
(x x
0
)
2
whereas we are dealing with an excitation over the whole crystal.
To proceed further we need to consider the thermal average occupancy of these phonon modes.
Z is the so-called partition function, which we require to normalise the thermal occupation probabilities and is
equal to the sum over all Boltzmann factors.
Z =

n=0
exp(n/k
B
T) =

x
n
geometric series for x < 1 (6.46)
CHAPTER 6. LATTICE VIBRATIONS - PHONONS 34
=
1
1 exp(/k
B
T)
(6.47)
Hence the probability of nding n phonons in the mode is given by:
P(n) =
exp(n/k
B
T)
Z
(6.48)
What is the thermal average occupancy < n > ?
< n >=

n=0
nP(n) = Z
1

nexp(n/k
B
T) (6.49)
Setting /k
B
T = x we have

nexp(nx) =
d
dx

exp(nx) =
1
1 exp(x)
=
exp(x)
[1 exp(x)]
2
(6.50)
< n >=
exp(x)
1 exp(x)
. (6.51)
or using our physical notation we have
< n >=
1
exp (/k
B
T)1
Planck distribution function. (6.52)
The Planck distribution function is valid for photons, phonons and other comparable bosons. But how do we
calculate the total energy in the lattice vibrations of the crystal? This can be done as follows:
< >= < s >=

exp(/k
B
T) 1
(6.53)
gives the energy associated with the thermal average occupancy of a single mode. We now have to sum over all
allowed frequencies i.e. over all allowed wave vectors

k
ph
and indeed polarisations:
U =

k
ph

k,p
exp(

k,p
) 1
(6.54)
Clearly this sum depends critically upon the dispersion of the crystal. It can be most readily evaluated by
introducing (as in the free electron model) the concept of the phononic density of states D
p
() whereby:
D
p
()d = the number of phonon modes of polarisation p between and + d
U =

p
_
d D
p
()

exp (/k
B
T)
(6.55)
But what is the density of states in practice? Let us consider initially a 1D chain of atoms where the position of
the rst and the last atom is assumed xed.
0 1 2 3 4 5 6
L
a
7
lenght L
separation a
N+1 particles
L=Na
Figure 6.10: Illustration of the phononic density of states.
CHAPTER 6. LATTICE VIBRATIONS - PHONONS 35
Clearly we may write a standing wave solution for the amplitudes of the oscillations.
u
n
= u
0
exp(i

k,p
t) sin(nk
ph
a) (6.56)
In order for the solutions to vanish at n = 0 and n = N, i.e. at the ends of the string, we can only allow the
wave vector

k
ph
to have the following values:
k =

L
,
2
L
,
3
L
...
(N 1)
L
(6.57)
0 is clearly physically not allowed because in this case u
n
would be zero for all n, as is the case for N/L because
sin(n N/La) = sin(n) = 0 n. The number of allowed k-modes is equal to the number of atoms (lattice
places) which are free to move.
Note the dierence between the xed boundary conditions and the Born- von Karman boundary conditions.
(as considered for free electron motion)
Fixed: Mode spacing

L
; only k > allowed
B.v.K.: mode spacing
2
L
; but the allowed values for k are 0,
2
L
,
4
L
...
But the total number of modes is the same for B.v.K. boundary conditions and for xed boundary conditions.
For N atoms and their lattice vibrations the total number of modes is 3N (= 2 transverse + 1 longitudinal
polarisation).
What we now have to calculate is the phononic density of states D(k) or,
D(k)dk = D()d D() = D(k)
dk
d
_
=
L

dk
d
_
. (6.58)
For a one dimensional system the above formulation is valid, but in a real crystal we expect lattice vibrations in
all directions in real space. Hence we need to consider a three dimensional picture of lattice vibrations.
In order to calculate the density of states in 3D we consider, as before for free electrons a box of volume V = L
3
and periodic Born - von Karmann boundary conditions.
exp(i

kr) = exp(ik
x
x + ik
y
y + ik
z
z) = exp(ik
x
(x + L)) + ik
y
(y + L) + ik
z
(z + L) (6.59)
allowed values for k
x
, k
y
, k
z
= 0,
2
L
,
4
L
; ...
2n
L
n N (6.60)
The volume of a single state in k-space is (2/L)
3
. The number of states within radius k can be calculated as
follows:
=
4
3
k
3
/
_
2
L
_
3
N = V
1
6
2
k
3
(6.61)
for each polarisation, i.e. for two transverse and for one longitudinal polarisation.
3
Thus for each polarisation
we may be write:
D() = V
1
2
2
k
2
_
k

_
(6.62)
Where the last term has to be found from the concrete form of the dispersion.
6.4.1 The Debye approximation
At low enough temperatures we can write = v
s
k, i.e. only low energy acoustic phonons will be excited and
hence we can safely approximate our (k) through a linear, constant speed of sound.

k
= v
s
and D() =
V
2
2
2
v
2
s
. (6.63)
3
To account for the spin degeneracy for free electrons, we multiply by 2 - we cannot do something similar here for the various
polarisations are not necessarily degenerate -
L
(

k) =
T
(

k).
CHAPTER 6. LATTICE VIBRATIONS - PHONONS 36
How many modes does the acoustic branch contain? As before we expect 3N degrees of freedom per atom. For
one atom per primitive cell and for one polarisation we therefore require N modes:
N = V
1
6
2


3
D
v
3
s
(6.64)
where we have dened the so-called Debye-frequency and Debye temprature
k
D
=

D
v
=
_
6
2 N
V
_
1/3
. (6.65)
Only k-vectors smaller than k
D
are allowed! We now can calculate the total energy:
U
tot
=

k
ph

k,p
exp(/k
B
T) 1
(6.66)
U =

p
_
dD() < n() >
=

p
_

p
0
_
V
2
2
2
v
3
__

exp(/K
B
T) 1
_
(6.67)
For simplicity we assume v
s,L
= v
s,T
= v and therefore summing over polarisations simply yields a factor of 3:
U =
3V
2
2
v
3
_

D
0

3
exp(/k
B
T) 1
d (6.68)
We now substitute /k
B
T = x d = k
B
T/dx and x
D
=
D
/k
B
T = /T where is the so-called Debye
temperature,
=

k
B
=
v
k
B
_
6
2
N
v
_
1/3
(6.69)
The substitution yields:
U =
3V k
4
B
T
4
2
2
v
3

3

_

D
0
dx
x
3
e
x
1
= 9Nk
B
T
_
T

_
3
_

D
0
x
3
e
x
1
dx (6.70)
Note that this integral is not independent of temperature since the temperature is implicitly contained in the
upper limit of the integral. Now we may derive the specic heat from C
V
=
U
T
. It follows from Equation 6.70
C
V
= 9Nk
B
_
T

_
3
_
x
D
0
dx
x
4
e
x
(e
x
1)
2
(6.71)
At very low temperatures x
D
= /k
B
T :
_

0
dx
x
3
e
x
1
=
_

0
dx x
3

s=1
exp(sx) = 6

1
1
s
4
=

4
15
(6.72)
U
tot
= 3
4
Nk
B
T/5
3
(6.73)
C
V
= 12
4
Nk
B
(T)
3
/5 234Nk
B
_
T

_
3
(6.74)
We thus conclude that we have found the source of the T
3
term in our empirical heat capacity.
C
V
= aT + bT
3
(6.75)
CHAPTER 6. LATTICE VIBRATIONS - PHONONS 37
This law is found for metals at low temperatures; empirically godd agreement is found for temperatures T < /50.
C
V
T
T
2
a
slope b
Figure 6.11: For the practical, independent determination of the two important contributions to the specic heat C
V
/T
is plotted against T
2
. The intercept a corresponds to the electric contribution, while the slope b yields the phonon
contribution.
Typical values of the Debye temperature vary from 200 to 2000 K.
There is one problem however, C
V
T
3
cannot be vaild at high temperatures. Clearly in this limit the integral
of Equation 6.71 has a upper limit which tends to zero. As expected the specic heat saturates at a value of 3N.
Einstein model of the density of states. N oscillators with the same frequency
D() = N(
0
) (6.76)
U = N < XXX > =
N
exp(/k
B
T) 1
(6.77)
C
V
=
U
T
= Nk
B
_

k
B
T
_
2
exp (/k
B
T)
exp (/k
B
T)1
(6.78)