Sei sulla pagina 1di 116

Solid State Physics 1 notes

Maurizio Scalet

a.y. 2017/2018

1
Contents
1 Einstein and Debye models of the specific heat 5
1.1 The Dulong-Petit Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Einstein Model for the Specific Heat . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Debye Model for the Specific Heat . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2 Drude Theory of Metals 9


2.1 Basic Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 DC Electrical Conductivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.3 Hall Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.4 Thermal Conductivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

3 Sommerfeld Theory of Metals 17


3.1 Ground State of the Electron Gas . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.2 Thermal Properties of the Free Electron Gas . . . . . . . . . . . . . . . . . . . . 20
3.2.1 Fermi-Dirac Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.2.2 Application of the Fermi-Dirac distribution . . . . . . . . . . . . . . . . . 22
3.3 Sommerfeld Theory of Conduction . . . . . . . . . . . . . . . . . . . . . . . . . 24

4 Crystal Lattices 27
4.1 Bravais Lattice . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

5 X-Ray Diffraction 28
5.1 Bragg Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
5.2 Von Laue Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
5.3 Equivalence of the 2 Formulations . . . . . . . . . . . . . . . . . . . . . . . . . . 29
5.4 Geometries Suggested by the Von Laue Condition . . . . . . . . . . . . . . . . . 29
5.4.1 Ewald Construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
5.4.2 The Laue Method, the Rotating Crystal Method and the Powder Method 30

6 X-Rays Interaction with Matter 31


6.1 X-ray Absorption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
6.2 X-ray Scattering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
6.2.1 One Electron . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
6.2.2 One atom . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
6.2.3 One Molecule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
6.2.4 Crystal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
6.2.5 Unit Cell Structure Factor . . . . . . . . . . . . . . . . . . . . . . . . . . 37
6.3 Scattering From a Crystallite . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
6.3.1 Qualitative Origin of Bragg Peaks’ Broadening . . . . . . . . . . . . . . . 39
6.3.2 Scherrer’s Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
6.3.3 Scattered Intensity from Crystallites . . . . . . . . . . . . . . . . . . . . 40
6.4 Powder Diffraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

7 Classification of Bravais Lattices 43


7.1 Classification of Bravais Lattices . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
7.2 Crystallographic Point Groups and Space Groups . . . . . . . . . . . . . . . . . 44
7.3 Point Group Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
7.4 The 230 Space Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
7.5 Examples among the Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

2
8 Diffraction from Non-Crystalline Materials 48
8.1 The Radial Distribution Function . . . . . . . . . . . . . . . . . . . . . . . . . . 48
8.2 The Liquid Structure Factor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
8.2.1 Short Range Order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
8.2.2 Supercooled Liquids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
8.2.3 Small Angle X-ray Scattering . . . . . . . . . . . . . . . . . . . . . . . . 51
8.3 Scattering from a Isolated Particles . . . . . . . . . . . . . . . . . . . . . . . . . 52
8.3.1 Long Wavelength Limit: Guinier Analysis . . . . . . . . . . . . . . . . . 52
8.3.2 Short Wavelength Limit: Porod Analysis . . . . . . . . . . . . . . . . . . 53

9 Electrons in a Periodic Potential 55


9.1 Bloch Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
9.2 Remarks on the Bloch’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 57
9.3 The Fermi Surface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

10 Classification of Solids 59
10.1 Classification of Insulators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
10.2 Ionic Crystals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
10.2.1 Alkali Halides - I-VII Ionic Crystals . . . . . . . . . . . . . . . . . . . . . 60
10.2.2 II-VI Ionic Crystals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
10.2.3 III-V Crystals - Mixed Ionic and Covalent . . . . . . . . . . . . . . . . . 61
10.2.4 Covalent Crystals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
10.2.5 Molecular Crystals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
10.2.6 Metals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

11 Cohesive Energy 64
11.1 Molecular Crystals : the Noble Gases . . . . . . . . . . . . . . . . . . . . . . . . 64
11.1.1 Equilibrium Density . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
11.1.2 Equilibrium Cohesive Energy . . . . . . . . . . . . . . . . . . . . . . . . 66
11.1.3 Equilibrium Bulk Modulus . . . . . . . . . . . . . . . . . . . . . . . . . . 66
11.2 Ionic Crystals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
11.3 Covalent Crystals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
11.4 Free Electron Metals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

12 Lattice Dynamics 71
12.1 1D Two Atoms System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
12.1.1 Classical Statistics of the Harmonic Oscillator . . . . . . . . . . . . . . . 73
12.1.2 Quantum Statistics of the Harmonic Oscillator . . . . . . . . . . . . . . . 73
12.1.3 Small Anharmonicity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
12.2 Multi Atom Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
12.3 Linear Chain of Atoms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
12.4 Crystal Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
12.5 Periodic Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

13 Theory Of Elasticity 90
13.1 Elastic Waves in Crystals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
13.2 Connection with Microscopic Dynamics . . . . . . . . . . . . . . . . . . . . . . . 96

14 Quantum Theory of the Vibrational Dynamics 98


14.1 Crystals in the Quantum Theory . . . . . . . . . . . . . . . . . . . . . . . . . . 101
14.2 Specific Heat of a Crystal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

3
15 Measuring the Phonon Dispersion Relations: Inelastic Scattering 107
15.1 Experimental set-ups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

16 Thermal Expansion 111


16.1 Specific Heat at Low Temperature . . . . . . . . . . . . . . . . . . . . . . . . . . 111
16.2 Thermal Expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
16.3 The Grüneisen Parameter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
16.4 Thermal Conductivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116

4
1 Einstein and Debye models of the specific heat
1.1 The Dulong-Petit Law
From statistical mechanics is known that the specific heat per atom of a mono atomic gas is
Cv 3
= kB (1)
N 2
where kB is the Boltzmann constant. This result is obtained deriving the energy with respect
of temperature. The energy can be computed either through the equipartition theorem or from
explicit statistical mechanics calculations.

Similar calculations can be done in the case of a solid to yield the result that the specific
heat per atom is
Cv
= 3kB (2)
N
Formally this can be obtained considering the atoms to be confined in an harmonic potential
well, either by means of the equipartition theorem (in this case we have 6 degrees of freedom)
or the Boltzmann distribution. These expressions for the specific heat are expressions of the
Dulong-Petit law.

It is important to note that this law is generally true only at high temperature (typically
room temperature is high enough), but it fails when the absolute zero is approached. Most
remarkably, it fails in the case of diamond also at room temperature, but the limiting value of
3kB is approached as the temperature grows.
The first satisfactory explanation of the behaviour of heat capacity at both high and low
temperatures was given by Einstein.

1.2 Einstein Model for the Specific Heat


This models considers the atoms as quantum harmonic oscillators, each oscillating at the same
frequency ω. Therefore each atom has an energy
 
1
En = ~ω n + . (3)
2
Now we are interested in the mean energy for a given temperature T. In order to do this we
start from the canonical partition function:

X 1
Q1D = e−βEn , β=
n=0
kB T
∞ β~ω (4)
X
−β~ω(n+ 12 e− 2
= e =
n=0
1 − e−β~ω
And the mean energy is given by
   
∂F ∂F
E =F −T =F −β (5)
∂T V ∂β V

where
~ω 1
+ log 1 − e−β~ω

F = −kB T log(Q) = −
2 β
 
 1 ~ωe−β~ω (6)
∂F 1
= − 2 log 1 − e−β~ω +
∂β V β β 1 − e−β~ω

5
and therefore the average energy is
1 −β~ω
 ~ωe−β~ω
E = F − log 1 − e + =
β 1 − e−β~ω
| {z }
= ~ω
2
=E0

e−β~ω
 
1
= ~ω + = (7)
2 1 − e−β~ω
 
1 1
= ~ω + β~ω ⇒
2 e −1
1
⇒ nB (ω, T ) ≡ β~ω
e −1
where nB (ω, T ) is the Bose-Einstein population factor.
From this result we can infer that the level corresponding to ω is populated by nB phonons,
each at the same frequency (since they are bosons). The same result could be obtained directly
applying the definition of ensemble average.
If we consider a single atom as a 1D harmonic oscillator we obtain
C ∂hEi ∂hEi eβ~ω
= = −kB β 2 = kB (β~ω)2 β~ω (8)
N ∂T ∂β (e − 1)2
The generalisation to 3D needs to use three parameters, nx , ny and nz , to express the energy
1 1 1
Enx ,ny ,nz = ~ω(nx + )(ny + )(nz + )
X 2 2 2
βEnx ,ny ,nz 3
Q3D = =e = (Q1D ) (9)
nx ,ny ,nz

hE3D i = 3hE1D i
Therefore in the Einstein model we get
C eβ~ω
= 3kB (β~ω)2 . (10)
N (eβ~ω − 1)2
In the limit of high temperature, i.e. kB T  ~ω, this result recovers the Dulong-Petit law. In
the other limit, i.e. in the limit of low temperature, the specific heat becomes:
C
' 3kB (β~ω)2 e−β~ω (11)
N
which goes to zero as the temperature approaches the absolute zero, as the experimental data
show.
From this model we can define the Einstein temperature as
~ω = kB TE (12)
which for diamond has a value approximately of TE ' 1320K. The frequency can be linked to
the bounding strength of the atoms through the Hooke law
r
k
ω= (13)
m
The high temperature for diamond is due to the strong binding present and the low weight of
the atoms.
The limits of the model are due to the assumption that each atom oscillates at the same fre-
quency and, as we have seen this implies an exponential behaviour at low temperatures, whereas
the experimental data show a behaviour proportional to the third power of the temperature.
The next step is to consider quantum harmonic oscillators with a distribution of frequencies.

6
1.3 Debye Model for the Specific Heat
Debye considers collective motion of the atoms. This results in vibrations, that can be described
as particles known as phonons. In a similar fashion to what was done by Planck for the black
body radiation, Debye assigns to the phonons a linear dispersion relation:

ω = v~k (14)

where k is the wave vector and v is the speed of sound in the solid. Instead of having two
polarizations as the photons, the phonons have three: two transverse and one longitudinal. In
a crystal the two modes have different speeds (anisotropy), but we will consider the wave vector
to be a scalar, i.e. the speeds will be equal in each direction.
Since now the frequency is a function of k, the mean energy becomes
 
X 1
hEi = ~ω(k) nB (ω(k), T ) + (15)
2
~k

and the sum is performed on the labels nx , ny and nz of k. Since the number of atoms in the
solid is fixed, the wave vectors k are discrete and finite. This result comes from the Born-Von
Karman boundary conditions when solving for the phonons as free particles.
In the thermodynamic limit the sum becomes an integral
 3 Z
X L
→ d~k (16)

~k

since we are in the isotropic approximation

d~k = 4πk 2 dk (17)

performing a change of variables to get an integral over the frequency ω we get


 
1 2 2 3
d~k = 4πk dk = 4πω dω
2 2
+ = 4πω dω (18)
vL3 vT3 3
vD
where vD is the speed of sound in the material and is called Debye velocity, assuming vL = vT ..
Then the mean energy is
Z ∞
L3
 
1 3
hEi = 3
~ω nB (ω, T ) + 4πω 2 3 dω =
(2π) 0 2 vD
ω2

Z ∞    g(ω) = 6π9N2 nv 3 (19)
1  D
= g(ω)~ω nB (ω, T ) + dω, n = N3
0 2  2 L3

3
6π nvD = ωD

where g(ω) is the density of states, n is the density of atoms, and ωD is the Debye frequency.
Then g(ω)dω is the number of modes with frequency between ω and ω + dω. The computation
of the mean energy is the same as the calculations done by Planck for the photons:
9N ~ ∞ ω3
Z
hEi = 3 dω + int independent of T → x = β~ω →
ωD 0 e−β~ω − 1
Z ∞
9N ~ x3 9N (kB T )4 π 4 (20)
→ 3 −x − 1
=
ωD (β~)4 0 e
| {z } (~ω D ) 3 15

Riemann zeta= π4
function 15

7
To get the heat capacity it just suffices to take the derivative of the mean energy with respect
to the temperature:
∂hEi 36N (πkB )4
C= = (21)
∂T (15~ωD )3
defining the Debye temperature as kB ΘD = ~ωD we can rewrite the heat capacity as

12N kB π 4 T 3
C= (22)
5 Θ3D

This expression has the correct behaviour as seen experimentally, but it has a problem: it does
not reproduce the Dulong-Petit law for high temperatures. This can be fixed imposing a cut-off
frequency ωC when performing the integral. Then the integral of the density of states from
zero to the cut-off frequency must be equal to 3N , since there are three polarizations.
Z ωC Z ωC
9N ω 2
g(ω)dω = 3N ⇒ 3
= 3N ⇒
0 0 ωD
(23)
9N ωC3 ωC3
⇒ 3 = 3N ⇒ 3N 3 = 3N
ωD 3 ωD

Then we get the condition that the cut-off frequency must be exactly the Debye frequency ωD .
Therefore the energy can be computed through
Z ωD  
1
hEi = g(ω)~ω nB + (24)
0 2

At low temperatures the T 3 behaviour is unaffected by the cut-off frequency, whereas at high
temperatures we have
1 kB T
nB = β~ω −→ (25)
e + 1 β→0 ~ω
and therefore the energy becomes
Z ωD
hEi −→ g(ω)kB T dω + const. (26)
β→0 0

and the heat capacity is


C = 3N kB (27)
which once again is the Dulong-Petit law. This model is based on a linear dispersion relation
1
ωD = vD kD ⇒ kD = 6π 2 n 3 (28)

this, together with the ad hoc assumption of the existence of a cut-off frequency, represent
the biggest issue with this model, since many times the dispersion relation is not linear, but
nevertheless this model reproduce quite well the experimental data. For diamond the Debye
temperature is about 1850K, while for lead it is about 400K. The difference is due to the
stronger bindings and lighter atoms in diamond.

8
2 Drude Theory of Metals
When one starts to study the physics of solid matter he is put against a first important classi-
fication: metals and non-metals. The first are characterised by a good electrical and thermal
conductivity, while the latter has much less efficient conductivities. The interest in the study
of the metals comes from their relative abundance, since two thirds of the elements are metals.
Moreover, understanding why metals conduct so well helps in understanding why non-metals
do not. After the discovery of the electron in 1897 by Thomson, many simple models were
made to, at least qualitatively, take into account why metals have their metallic characteristics.
The first model was proposed at the turn to the twentieth century by Drude.

2.1 Basic Assumptions


Drude constructed his theory on the electric and thermal conduction of metals by applying the
kinetic theory of gases to a metal, considering it as a gas of electrons. This theory is grounded
on some assumptions

• The molecules are solid spheres;

• The molecules move in straight lines until they collide with one another;

• The time taken in a single collision is negligible;

• There are no forces between the molecules but those which come into play during the
collisions.

The picture drawn by Drude of a metal shows that each atom has Za electrons, Z of whose
are the conduction electrons and are the members of the electron gas and Za − Z are the core
electrons, which do not participate in the conduction processes. A parameter of key importance
for a gas is its density: the density of the electron gas can be estimated considering that a metal
contains 0.6022 × 1024 atoms per mole and ρAm per cm3 , where ρm is the mass density and A
is the atomic mass of the element. Since each atom contributes with Z electrons, the electron
density is
Zρm  −3 
n = 0.6022 × 1024 cm . (29)
A
Another widely used description of the density of the electron gas is the radius rs of the volume
per electron, considered as a sphere:
1/3
4πrs3

1 3
= ⇒ rs = . (30)
n 3 4πn
2
Typically rs as value of 2 − 3 times the Bohr radius a0 = me ~
2. From these calculations it
can be seen that the density of the electron gas is typically a thousand times greater than
that of a classical gas. Despite this high density and strong electron-electron and electron-ion
interactions, the Drude model treats the electron gas as a classical dilute gas, with some small
modifications. The basic assumptions of Drude are:

I. Independent electron approximation: the electron-electron interactions are neglected;

II. Free electron approximation: the electron-ion interactions are neglected;

III. Collisions are instantaneous events that abruptly change the velocity of an electron;

9
IV. The probability per unit time that an electron experiences a collision is τ1 . τ is typically
known as relaxation time and it is independent of the electron position and velocity;

V. The only way for the electrons to achieve thermal equilibrium is through collisions.

These assumptions have some implications: from the first two we get that in the absence of an
external field the electrons will move in a straight line between collisions, while if an external
field is present they will respond to it according to Newton’s law of motion, neglecting any field
produced by the other electrons and the ions. The first approximation is surprisingly good,
while the second must be abandoned if a qualitative understanding is required.
The third assumption considers collisions with the ions, but as we will see later these colli-
sions are amongst the least important of the scattering events that an electron can experience.
Fortunately this is not important since many results can be obtained just assuming that a
scattering mechanism exist, independently of its nature.
From the fourth hypothesis we conclude that the probability that an electron does not
experience a collision after a time t is a decreasing exponential:
 
dt dP P (t) t
P (t + dt) = P (t) 1 − ⇒ =− ⇒ P (t) = e− τ (31)
τ dt τ

The last hypothesis implies that thermal equilibrium is attained with a simple mechanism:
after each collision the electron emerges with a velocity that is uncorrelated with the velocity
it previously had, but randomly directed and with a speed appropriate to the temperature
prevailing on the site of the collision.

2.2 DC Electrical Conductivity


Ohm’s law tells us that the current flowing through a wire I is proportional to the potential
applied to the wire V :
V = IR (32)
where R is the resistance of the wire and it depends on the geometry of the wire, but not on
either the current or the potential drop. Instead of the resistance of the wire, it is easier to
work with the resistivity ρ, which does not depend on the geometry of the wire, but only on the
~
metal. The resistivity is defined as the constant of proportionality between the electric field E
at a point in the metal and the current density ~j that it induces:
~ = ρ~j.
E (33)

The current density is a vector parallel to the charge flow whose magnitude is the quantity of
charge that crosses a unit area perpendicular to the flow per unit time.
If n electrons are travelling and velocity ~v they will give rise to a current parallel to ~v . In
a time dt those electrons will travel a distance vdt, therefore the number of electrons that will
cross an area A perpendicular to the flow will be n(vdt)A. Since each electron carries a charge
−e, the charge crossing A in the time dt will be −envdtA and hence the current density will
be:
~j = −ne~v . (34)
At any point in the metal the electrons are moving in different directions due to thermal
agitation, thus the current density must be given by the average of the electron velocity. In the
absence of an electric field the average is null due to the randomness of the thermal velocities.
But with an external electric field the electrons acquire a mean velocity directed in the opposite
direction of the field, giving rise to a non null current density. This average velocity can be

10
computed as follows: using Newton’s second law we can calculate the velocity acquired by an
electron after a time t from the last collision as:
~
eEt
~v (t) = − . (35)
m
This acquired velocity is summed to the velocity that the electron got immediately after the
collision ~v0 . Since we assumed that each electron emerges from the collision with a randomly
directed velocity, ~v0 will not contribute to the average electronic velocity, thus we are left with
the average of the velocity acquired from the electric field. The average of t is just the relaxation
time τ , therefore we get:
~
eEτ ne2 τ ~
~vavg = − ⇒ ~j = E (36)
m m
where the fraction is the conductivity σ = ρ1 of the metal.
This gives the linear relation between the current density and the electric field in terms of
all known quantities, except for the relaxation time. Although, inverting the definition of the
conductivity and using the measured resistivities we can estimate the relaxation time:
m
τ= (37)
ρne2

and typically it is of the order of 10−14 ÷ 10−15 s. To know if this is a reasonable estimate we
can look at the mean free path l = v0 τ , where v0 is the average electronic speed. The average
electronic speed can be computed from the equipartition theorem
1 2 3
mv = kB T (38)
2 0 2
and this leads to the result that the speed of the electron is of the order of 107 cm/s at room
temperature and hence to a mean free path of 1 to 10 Å. Since this length is of the order
of the interionic distance, it is compatible to the view of Drude that electrons bump into the
heavy ions. As we will see the velocity estimate is an order of magnitude too small at room
temperature. Moreover, lowering the temperature, the relaxation time can increase of an order
of magnitude, while the velocity is temperature independent which leads to mean free paths of
thousands of Å. This can be done experimentally (indeed mean free paths of centimetres can
be achieved) and it suggests that the electrons do not simply bump off the ions.
Despite these problems of the Drude theory we can still use it to compute quantities that
are independent of the value of the relaxation time. Cases of this kind of particular interest
are those in which a uniform magnetic field is present and that in which the electric field is
uniform but it varies in time. We will focus only on the first case.
First of all we wish to write an equation for the momentum of the single electrons, since
the current density directly depends on it: given the momentum per electron p~(t) at time t, to
calculate the momentum at time t + dt we need to consider that the probability that it does
not undergo a collision after dt is 1 − dt τ
. However, if it does not experience any collision it
will evolve according to the force f~(t), acquiring the additional momentum f~(t)dt + O(dt2 ).
Therefore the total momentum carried by the electrons will be given by the fraction (1 − dt τ
) of
the total electrons that do not undergo any collision times the average momentum per electron
 
dt  
p~(t + dt) = 1 − p~(t) + f~(t) + O(dt2 ) =
τ
(39)
dt ~ 2
= p~(t) − p~(t) + f (t) + O(dt )
τ

11
The correction to this formula from the electrons that experience a collision is of the order dt2 .
To see this we consider the fact that the fraction of the electrons that collide is dt τ
, each of
whose would contribute to the total momentum only to the extent of the momentum acquired
from the force f~(t). This momentum is acquired in times smaller than dt and therefore the
total momentum that these electrons would contribute is dt τ
f~(t)dt and it does not effect the
momentum at the linear order of dt. Therefore we can write
dt
p~(t + dt) − p~(t) = − p~(t) + f~(t)dt + O(dt2 ) (40)
τ
dividing by dt and taking the limit as dt → 0 we get
d~p(t) p~(t) ~
=− + f (t) (41)
dt τ
and this states that the effect of the single electron collisions is to add a frictional damping
term into the equation of motion for the momentum per electron.
In the following we will be applying this equation.

2.3 Hall Effect


Hall wanted to understand the underlying mechanism that leads to a force in a current carrying
wire when put in a magnetic field, suspecting that it was due to moving charges inside the
wire. The presence of the magnetic field should press the current against one side of the wire
increasing the resistance of the wire. Although he was not able to measure this increase in the
resistance, he supposed that the pressing current would lead to a stress state of the wire. It
should appear as a transverse potential, which he was able to measure, and today it is known
as Hall voltage.
A schematic of the experiment can be seen in figure. An electric field Ex applied to a wire
that extends along the x-direction drives a current density jx . A magnetic field H ~ directed
along the positive z-direction is also applied. As a result, the Lorentz force
e ~
F~L = − ~v × H (42)
c
deflects the electrons in the negative y-direction. This leads to an accumulation of electrons in
the proximity of the side of the wire building up an electric field directed along the negative
y-direction such that it opposes the accumulation of charge. When the equilibrium is reached
the transverse field Ey will balance the Lorentz force and the current will flow only in the
x-direction.
As a result there are two quantities of interest. The first is the ratio between the field along
the wire Ex and the current density jx
Ex
ρ(H) = (43)
jx
known as magnetoresistance, which Hall found to be field independent. The other is the mag-
nitude of the transverse field. As it balances the Lorentz force we expect it to be proportional
to both the applied field H and to the current along the wire jx . Therefore we can define a
quantity known as Hall coefficient as
Ey
RH = . (44)
jx H
It should be noted that since the Hall field is in the negative y-direction, RH should be negative.
On the other hand, if the charge carriers would have been positively charged the Hall field would

12
be directed along the opposite direction which would yield a positive Hall coefficient (the current
density would not change sign since positive charges move in the opposite direction of the
negative charges). This is of primary importance since a measurement of the Hall field allows
us to determine the sign of the charge carriers. The data obtained by Hall agreed with the sign
of the electronic charge determined by Thomson. However, for some metals the Hall coefficient
has positive sign, suggesting that the charge carriers are positively charged. This is another
problem that will be solved by a full quantum mechanical treatment.
For the moment we will continue with the Drude model that, although being incapable of
describing the positive Hall coefficient case, shows good agreement with the experiments. The
force acting on each electron is  
~
v
f~ = −e E ~ + ×H ~ (45)
c
and therefore the equation for the momentum per electron becomes
 
d~p ~ p~ ~ p~
= −e E + ×H − (46)
dt mc τ
In the steady state the current is independent of time, and thus px and py satisfy
px
0 = − eEx − ωc py −
τ (47)
py
0 = − eEy + ωc px −
τ
where
eH
ωc = (48)
mc
is called cyclotron frequency.
Multiplying these equation by −neτ /m and introducing the current density components we
obtain
σ0 Ex =ωc τ jy + jx
(49)
σ0 Ey = − ωc τ jx + jy
where σ0 is the Drude model DC conductivity that we already found.
The Hall field is simply computed requiring that the transverse current be null. Therefore
from the second equation  
ωc τ H
Ey = − jx = − jx (50)
σ0 nec
and the Hall coefficient becomes
1
RH = − . (51)
nec
This result implies that the Hall coefficient depends only on the density of the carriers, which we
previously estimated assuming that each atomic valence electron becomes a conduction electron
in the metal. Therefore a measurement of the Hall coefficient allows us to test the validity of
this assumption. But this is easier said than done, since, contrary to what these equations
predict, the Hall coefficient does depend on the magnetic field. Moreover, it depends also on
the temperature and the care with which the sample has been prepared. This is unexpected
since the relaxation time, which strongly depends on the temperature and on the condition
of the sample, does not appear in equation (51). However, at low temperatures and for high
enough fields it is observed that the Hall coefficients approach a limiting value. An important
dependence not predicted by the Drude model can be seen in figure.
Drude explains the observations of Hall that the resistance does not depend on the magnetic
field, since when jy = 0 we have jx = σ0 Ex . However, more careful experiments show that the

13
resistance does depend on the magnetic field. This is another problem of the Drude theory that
can be solved only through a complete quantum treatment of the solids.
Lastly, we observe that the quantity ωc τ is a measure of the magnitude of the magnetic field.
Indeed, when ωc τ is small, we have ~j nearly parallel to E, ~ as in the absence of the magnetic
field. In general the angle φ between the current density and the electric field is given by
tan φ = ωc τ . ωc is known as the cyclotron frequency and is simply the frequency of revolution
of a free electron in the magnetic field H. Thus if ωc τ is small, the electron can complete only a
small fraction of its orbit before colliding with an ion, while if it is large it can complete many
revolutions between collisions. Alternatively we can see that when ωc τ is small the electronic
trajectory is only slightly perturbed by the magnetic field, while when it is large it can have
drastic effects.

2.4 Thermal Conductivity


One of the major successes of the Drude theory was its capability to explain the Wiedemann-
Franz law. This law is an empirical law that states that for metals the ratio between the
thermal and the electrical conductivities is proportional to the temperature, with a constant
of proportionality that does not depend on the metal. This proportionality constant is called
Lorenz number.
To account for this the Drude model assumes that the bulk thermal current is carried by the
conduction electrons. This assumption is fairly reasonable, since it is grounded on the empirical
fact that metals conduct heat much better than insulators. Therefore, this difference must be
due to the fact that the thermal conduction by the ions is much less important than that by
the electrons.
Considering a metal bar with a small temperature difference between its ends, the thermal
current ~j q is given by the Fourier law
~j q = −kth ∇T (52)
and it represents the thermal energy per unit time crossing an unit area perpendicular to the
flow. The proportionality constant kth is known as the thermal conductivity.
In order to concretely compute the thermal current flowing through the metallic bar we need
to exploit the fifth assumption of the Drude model, i.e. each electron emerges from a collision
with a velocity proportional to the mean temperature at the site of the collision. Therefore,
although the mean electronic velocity at a point can vanish, electrons arriving from the high
temperature side will have higher energies than those coming from the low temperature side
leading to a net flow of energy towards the low temperature side.
To get a somewhat quantitative estimate we consider a simplified one dimensional case,
where the electrons can move only on the x-axis. Thus, at a point x half the electrons come
from the high temperature side and half from the low (the high temperature side is on the left
of x). If E(T ) is the thermal energy per electron at equilibrium temperature T, then an electron
that emerges from a collision at point x0 will have a thermal energy E(T (x0 )). Therefore, on
average the electrons coming from the high temperature side will have hadtheir last collision at
point x−vτ and then they carry a thermal energy per electron E T (x−vτ ) . Their contribution
to the thermal current at x will therefore be the number of the electrons per unit volume, n2 ,
times their velocity, v times the energy that they carry
n 
vE T (x − vτ ) . (53)
2
While the electrons coming from the low temperature side will contribute
n 
(−v)E T (x + vτ ) . (54)
2
14
Adding together these two contributions we get
1  
j q = nv E T (x − vτ )) − E(T (x + vτ ) . (55)
2
Provided that the change in temperature over the mean free path l = vτ is small, we can
expand about the point x to obtain
 
q 2 dE dT
j = nv τ − . (56)
dT dx
-this can be generalized to the three dimensional case quite easily: we need to replace v by the
x-component of the velocity ~v and average over every direction. Since hvx2 i = hvy2 i = hvz2 i = 31 v 2
and since ndE/dT = (N/V )dE/dT = (dE/dT )/V = cV is the electronic specific heat, we get

~j q = 1 v 2 τ cv (−∇T ) (57)
3
or
1 1
kth = v 2 τ cv = lvcv (58)
3 3
It should be noted that this argument is quite rough. Indeed it is quite hard to define the
thermal energy per electron with precision, and moreover we replaced carelessly quantities
with their thermal averages. For example we could object that since the energy per electron
depends on the direction from which it is coming, so will their average speed. We shall see later
that this problem is cancelled by another oversight we made. Nevertheless, this estimate of the
thermal conductivity is quite close to the exact solution of more rigorous calculations.
Now that we estimated the thermal conductivity we can test our results with the Wiedemann-
Franz law, calculating
1
kth c mv 2
3 v
= . (59)
σ ne2
At the time, Drude estimated the specific heat and square mean velocity using the ideal gas
laws: (
cv = 32 nkB
1
(60)
2
mv 2 = 32 kB T
which lead to the result  2
kth 3 kB
= T (61)
σ 2 e
which gives the correct behaviour of the Wiedemann-Franz law, giving the Lorenz number as
 2
kth 3 kB
= = 1.11 × 10−8 watt-ohm/K2 (62)
σT 2 e
which is about half the typical value observed experimentally.
Despite its success, this model had a quite puzzling problem: no electronic contribution
to the specific heat remotely comparable with 23 nkB was ever observed. The success of Drude
comes from two errors of about 100 that cancel each other: the electronic specific heat is 100
times smaller than the classical prediction and the electronic speed is 100 times larger.
Before leaving the thermal conduction, we need to correct an oversimplification we made
that obscures an important physical phenomenon:
we calculated the thermal conductivity neglecting the dependence of the velocity on the
temperature of the site where the collision take place, therefore it would appear that we need
to let the electronic velocity depend on the place where the last collision happened.

15
When the temperature difference is applied, the electrons will move towards the side with
the lowest temperature with a non-null mean velocity, inducing an electric current. But since
measurements of thermal conductivity are performed under open circuit conditions, no electrical
current can flow. This will lead to an accumulation of charge at one side of the bar, building up
an electric field that will oppose further accumulation of charge. This field will precisely cancel
the effect of the thermal gradient on the mean electronic velocity. When the steady state is
reached there will be no electric current, so our hypothesis that the mean velocity at one point
vanished was correct.
This leads us to an interesting physical effect: a temperature gradient on a long thin bar
is accompanied by an electric field directed opposite to the thermal gradient (Seebeck effect).
Conventionally the field is expressed as
~ = Q∇T
E (63)

and the proportionality constant Q is known as thermopower. To estimate it we note that the
mean electronic velocity at a point x due to the temperature gradient is
1  dv
vQ = v(x − vτ ) − v(x + vτ ) = −τ v =
2 dx
 2 (64)
d v
= −τ
dx 2

Generalizing to three dimensions in a similar fashion to what we have done before we get

τ dv 2
~vQ = − ∇T. (65)
6 dT
The mean velocity due to the electric field is
eτ ~
~vE = − E. (66)
m
To have the cancelling of the velocities, ~vQ + ~vE = 0 we require that

1 d mv 2 cv
Q=− =− (67)
3e dT 2 3ne
which also is independent of the relaxation time. Drude evaluated it applying erroneously the
classical statistical mechanics finding that
kB
Q=− = −0.43 × 10−4 volt/K. (68)
2e
Observed thermopowers at room temperature are of the order of microvolts per kelvin, a factor
100 smaller than what estimated. This is the same error that appeared in the Wiedemann-Franz
law, but now being uncompensated it clearly shows the inadequacy of the classical statistical
mechanics for the description of the electron gas. As it happened for the Hall effect, also in
this case there are some metals for which the thermopower changes sign. The quantum theory
of solids can account for the sign reversal in the thermopower as well.

16
3 Sommerfeld Theory of Metals
Drude model was entirely based upon the classical statistical mechanics and this fact led to its
major failures. Particularly it estimated the velocity of the electrons according to the Maxwell-
Boltzmann distribution. When the quantum mechanics was developed it was discovered that
the electron is a fermion, i.e. a particle subject to the Fermi-Dirac statistics and to Pauli
exclusion principle. Therefore the Maxwell-Boltzmann distribution shall be replaced by the
Fermi-Dirac:
(m/~)3 1
f (~v ) = 3
  (69)
4π exp 1 mv 2 − k T /k T + 1
2 B 0 B

The temperature T0 is to be determined by the normalization condition


Z
n = f (~v )d~v (70)

and is typically in the order of tens of thousands of degrees.


As can be seen in figure, at temperatures of interest (. 103 K) the Maxwell-Boltzmann and
the Fermi-Dirac distributions are definitely different at metallic electronic densities.
In most applications, Sommerfeld’s model is simply the classical theory of Drude where the
Maxwell-Boltzmann distribution is replaced by the Fermi-Dirac. To justify this semi-classical
treatment we must examine the quantum theory of the electron gas. We will start from the
ground state of the gas, i.e. at T = 0.

3.1 Ground State of the Electron Gas


As in Drude’s theory the electrons are supposed to be free and independent. Therefore to study
the ground state of N electrons confined in a volume V it suffices to find the energy levels for a
single electron and then filling them up according to the exclusion principle which allows only
one electron to occupy a particular level.
An electron is described by a wavefunction ψ(~r) and the specification of the orientation of
the spin it possesses. Since the electron is supposed to be free and independent, its wavefunction
will satisfy the free particle Schrödinger equation:
~2 2
− ∇ ψ(~r) = Eψ(~r) (71)
2m
In order to solve this equation we need appropriate boundary conditions, which represent the
confinement of the elctron in the volume V. We assume our volume to be a cube of side L and
that the bulk properties of our solid are not influenced by the surface of the solid. Therefore
we can safely apply the Born-von Karman boundary condition:

ψ(x + L, y, z) = ψ(x, y, z)
ψ(x, y + L, z) = ψ(x, y, z) (72)
ψ(x, y, z + L) = ψ(x, y, z)

thus, solving equation (71), neglecting the boundary condition we obtain


1 ~
ψ~k (~r) = √ eik·~r (73)
V
with energy
~2 k 2
E(~k) = (74)
2m
17
where ~k is a position independent vector. The normalization constant is such that the proba-
bility of finding the electron in the volume V is 1.
The meaning of ~k can be seen noting that the free electron wavefunction is an eigenstate of
the momentum operator:

p̂ = −i~ = −i~∇ ⇒
∂r (75)
∂ i~k·~r ~ i~k·~
r
⇒ −i~ e = ~ke
∂r
and therefore an electron in the level ψ~ (~r) will have a definite momentum proportional to ~k
k

~~k
p~ = ~~k ⇒ ~v = (76)
m
and the energy can be written in the familiar form

p2
E= (77)
2m
~
The vector ~k can also be interpreted as wave vector of the plane wave eik·~r which has a wave-
length

λ= (78)
k
known as the de Broglie wavelength.
Imposing the boundary conditions (72) we obtain a restriction to discrete values of ~k, since
the (72) can be satisfied only if

eikx L = eiky L = eikz L = 1 (79)

which is possible only if


2πnx
kx =
L
2πny
ky = nx , ny , nz integers. (80)
L
2πnz
kz =
L
This implies that in the space spun by the vectors kx , ky and kz the allowed wave vectors are
those whose coordinates are given by integer multiples of 2π/L. This information is usually
exploited when one is interested in the number of allowed wave vectors inside a given volume
in k-space. This is given by the ratio of this volume and the volume per point, which is just
(2π/L)3 . Therefore the region Ω of k-space will contain

Ω ΩV
= (81)
(2π/L)3 8π 3

allowed values of ~k. Equivalently we can say that the density of levels is 8πV 3 .
To build the ground state of the electron gas at T = 0K we simply put each electron into
an allowed level, from the lowest energetic. According to Pauli exclusion principle each level
can be occupied by only one electron, but since the electrons possess a spin that can have two
projections over a definite axis, for each allowed wave vector ~k are associated two electronic
levels. This proceeding is followed until every electron is in an electronic level. The volume
occupied in k.space will be indistinguishable from a sphere when N is sufficiently large. The
radius of this sphere kF is called Fermi wave vector and its volume is 4πkF3 /3. Therefore, using

18
the density of levels, we can express the Fermi wave vector in function of the number N of
electrons, remembering that each k point hosts two electrons:

4πkF3 V kF3
N =2 = V (82)
3 8π 3 3π 2
thus the electronic density can be expressed as

kF3
n= (83)
3π 2
The various quantities used to describe the ground state follow a rather unimaginative nomen-
clature: the sphere of radius kF is the Fermi sphere and its surface is the Fermi surface,
the highest momentum pF = ~kF is the Fermi momentum, the energy of the highest levels
EF = ~2 kF2 /2m is the Fermi energy.
All these quantities can be expressed in terms of the electronic density via equation (83).
For numerical estimates is often useful to express them in terms of the dimensionless parameter
rs /a0 , where rs is the radius of the volume per electron and a0 is the Bohr radius:
3.63 −1
kF = Å
rs /a0
~ 4.20
vF = kF = × 108 cm/s (84)
m rs /a0
e2 50.1
EF = (kF a0 )2 = eV
2a0 (rs /a0 )2

since the Fermi wave vector is of the order of inverse angstroms, the most energetic electrons will
have a de Broglie wavelength of the order of angstroms. The velocity is quite substantial, being
about 1 percent of the speed of light. This result is quite striking, since classically, particles in
the ground state would have zero velocity, but even at room temperature the thermal velocity
is of the order of 107 cm/s. In the first expression of the Fermi energy, the fraction is called
the rydberg (Ry) and it is the ground state binding energy of the hydrogen atom. Typically,
the Fermi energy for metals ranges from 1.5 to 15 electron volts.
To calculate the ground state energy we need to sum all the energies of the occupied one-
electron levels inside the Fermi sphere
X ~2 k 2
E=2 . (85)
k≤k
2m
F

Generally when summing a quite smooth function F (~k) over all allowed values of the wave
vector one wants to express the sum as a limiting integral. To do this we multiply and divide
by the volume per allowed ~k value ∆~k = 8π 3 /V :
X V X ~ ~
F (~k) = 3 F (k)∆k (86)

~k ~k

and taking the limit ∆~k → 0 the sum on the right-hand side approaches the integral F (~k)d~k,
R

provided that F (~k) does not change appreciably over distances of the order 2π/L. Therefore
we can rewrite
F (~k) ~
Z
1 X ~
lim F (k) = dk (87)
V →∞ V 8π 3
~k

19
When applying this formula we are always assuming that V1 ~k F (~k) differs negligibly from its
P
infinite volume limit.
Applying this proceeding to compute the mean energy density we obtain

~k 2 ~ ~k 2 ~2 kF5
Z Z
E 1 1
= 3 dk = 3 4πk 2 dk = (88)
V 4π ~k<~kF 2m 4π ~k<~kF 2m 10mπ 2

from which we can find the energy per electron, dividing this result by the electronic density

EV E ~2 kF5 3π 2 3 ~2 kF2 3
= = 2 3
= = EF (89)
V N N 10mπ kF 10 m 5

This result can be used to define the Fermi temperature equating the Fermi energy to the
thermal energy kB TF , thus
EF 58.2
TF = = × 104 K (90)
kB (rs /a0 )2
Once we got an expression for the energy we can compute some thermodynamics quantities,
such as the pressure and the bulk modulus. The former is given by P = −(∂E/∂V )N , and
since E depends linearly on the Fermi energy, which depends on the square of the Fermi wave
vector which in return depends on V only through n2/3 it follows that
2E
P = . (91)
3V
The latter is given by B = −V (∂P/∂V ), since the energy is proportional to V −2/3 , from the
previous equation we have that P ∝ V −5/3 and therefore
5 10 E 2
B= P = = nEF (92)
3 9 V 3

3.2 Thermal Properties of the Free Electron Gas


When the temperature is greater than the absolute zero we need to examine the excited states
of the electron gas, as well as the ground state. In order to do this we need to use the statistical
mechanics formalism for a system of N particles with the constraint of the Pauli exclusion
principle. This results in the Fermi-Dirac distribution.

3.2.1 Fermi-Dirac Distribution


According to statistical mechanics, the properties of a N-particles system are given averaging
over all stationary states of the system, each with a weight PN (E) given by

e−E/kB T
PN (E) = P −EαN /k T (93)
αe
B

where the sum is performed over all the stationary states α. The denominator is the partition
function, and it s related to the Helmoltz free energy F = U − T S by
N
X
e−Eα /kB T = e−FN /kB T (94)
α

Which allows us to express the statistical weight in a more compact way

PN (E) = e−(E−FN )/kB T (95)

20
Since our system is subject to the exclusion principle, each electron state can be occupied at
most by one electron at a time, therefore is useful to know the probability that a particular
level i is occupied, fiN , when the system is at thermal equilibrium. This probability is just the
sum of the independent probabilities of finding the system in any of those states in which the
i-th level is occupied X
fiN = PN (EαN ). (96)
α

where the sum is performed over the system states in which the i-th level is occupied. An
alternative form can be given in terms of a sum over the system states γ in which the level i is
empty X
fiN = 1 − PN (EγN ) (97)
γ

We observe that the N-electron system can be constructed from a (N+1)-electron system re-
moving one electron that occupies a level i. Therefore the energies of the two systems differ
only by the energy of the i-th level, then
X
fiN = 1 − PN (EαN +1 − Ei ) (98)
α

where the sum is over the (N+1)-electron states with the i-th level occupied. Using the expres-
sion for the statistical weight (95) we can write

PN (EαN +1 − Ei ) = e(Ei −µ)/kB T PN +1 (EαN +1 ) (99)

where µ is the chemical potential which is given by

µ = FN +1 − FN (100)

then we have that


X
fiN = 1 − e(Ei −µ)/kB T PN +1 (EαN +1 ) = 1 − e(Ei −µ)/kB T fiN +1 (101)

with the sum over all the (N+1)-electron states α with an electron in the i-th level. The last
equation is an exact relation between the probabilities of occupation of the level i at temperature
T in an N-electron system and in a (N+1)-electron system. Since we are interested in systems
with 1022 electrons is absurd to think that the addition of one electron could significantly change
the probability for more than few levels. Therefore we can replace fiN +1 with fiN and solve for
fiN :
1
fiN = (102)
1 + e i −µ)/kB T
(E

From now on we can drop the explicit reference to N, which is carried by the chemical potential
anyway. Indeed, since the fi is just the probability for the level i to be occupied, the total
number of electrons is just the sum over all levels of the mean occupation of each level
X X 1
N= fi = (103)
i
1 + e(Ei −µ)/kB T

since typically T and N, or rather the density N/V , are known, this equation allows us to
express µ as function of these quantities, allowing it to be eliminated from every subsequent
formulae.

21
3.2.2 Application of the Fermi-Dirac distribution
Before using the Fermi-Dirac distribution to compute the energy of the electron gas and then
the specific heat, we want to check that it describes correctly our system. In a electron gas
each one-electron level is specified by the wave vector ~k and the spin quantum number s, with
energy given by
~2 k 2
E(~k) = . (104)
2m
At the ground level, the only levels that are occupied are those whose energy satisfy E(~k) ≤ EF ,
so that (
1 E(~k) < EF
f~k,s = (105)
0 E(~k) > EF
Considering the limit T → 0 for the Fermi-Dirac distribution we get
(
1 E(~k) < µ
lim f~k,s = (106)
T →0 0 E(~k) > µ

Therefore, for these equations to be consistent one with the other we must have that

lim µ = EF (107)
T →0

Although for metals the chemical potential maintains its zero temperature value up to room
temperature, it should not be forgot that in precise calculations it is important to keep track
on how much the chemical potential differs from its zero temperature value.
Now we turn our attention to the computation of the energy of the electron gas. In the
independent electron approximation the internal energy is simply given by the sum over the
one electron levels energy times the mean population of the level:
X
U =2 E(~k)f (E(~k)) (108)

where we have introduced the Fermi distribution f (E to put into evidence that f~k depends on
the wave vector only through the electronic energy. Dividing both sides by the volume we can
write the energy density as
Z ~
dk ~
u= E(k)f (E(~k)) (109)
4π 3
If we divide also both sides of (103) by V we obtain an equation for the density

d~k
Z
n= f (E(~k)) (110)
4π 3
These integral can be easily computed using spherical coordinates and changing variables from
k to E. For a general function F (E(~k)):

d~k
Z Z ∞ 2
~k)) = k dk
F (E( F (E(~k)) =
4π 3 0 π2
Z ∞ (
m
q
2mE
(111)
2 2 ~2
E >0
= dEg(E)F (E), g(E) = ~ π
−∞ 0, E <0

Since this integral is the limit form of a sum divided by the volume, this form suggests us that
g(E)dE is the number of levels in the energy range from E to E + dE divided by the volume,

22
i.e. g(E) is the density of levels per unit volume. It can be rewritten in the more transparent
way as   1
3 n E 2
E >0
g(E) = 2 EF EF (112)
0, E <0
A quantity of particular interest is the density of levels around the Fermi energy
mkF
g(EF ) =
~2 π 2 (113)
3 n
=
2 EF
where the two forms are perfectly equivalent. Thus with this notation, the density and the
energy density integrals take the form
Z ∞
n= dEf (E)
−∞
Z ∞ (114)
u= dEEf (E)
−∞

In general these integrals have a rather complex structure, but there exists a systematic
expansion that exploits the fact at almost any temperature of interest in metals, T is much
smaller than the Fermi temperature. Therefore the Fermi-Dirac distribution differs from its
zero temperature form only in a small region about µ of width R ∞ a few kB T , at room temperature
for typical metallic densities. Thus, integrals of the form −∞ H(E)f (E)dE differ from their zero
R EF
temperature value −∞ H(E)f (E)dE in a way determined by the form of H(E) around E = µ.
If H(E does not vary too rapidly in the region of width kBT around µ it can be replace by its
Taylor expansion about E = µ
X dn (E − µ)n


H(E) = H(E) . (115)
dE n
E=µ n!

The result is a series known as the Sommerfeld expansion


∞ µ ∞
d2n−1
Z Z
X
2n
H(E)f (E)dE = H(E)dE + (kB T ) an 2n−1 H(E) (116)
−∞ −∞ n=1
dE E=µ

where the an are dimensionless constants of the order of unity. The functions generally encoun-
dn

tered have major variations on an energy scale of the order of µ and generally dE n H(E) E=µ is

of the order of H(µ)/µn , thus in this case the Sommerfeld expansion can be safely truncated
at the first or rarely at the second order, since successive orders of the expansion are smaller
by O(kB T /µ)2 which is O(10−4 ) at room temperature, then
∞ µ 6
π2 7π 4
Z Z 
kB T
H(E)f (EdE = H(E)dE + (kB T )2 H 0 (µ) + (kB T )4 H 000 (µ) + O (117)
−∞ −∞ 6 360 µ

Applying this expansion to the electronic energy and numeric densities we obtain
Z µ
π2
u= Eg(E)dE + (kB T )2 (µg 0 (µ) + g(µ)) + O(T 4 )
6
Z0 µ 2
(118)
π
n= g(E)dE + (kB T )2 g 0 (µ) + O(T 4 )
0 6

23
The second of these equations implies that µ differs from its zero temperature value by terms
of order T 2 , thus we can write, correctly to order T 2
Z µ Z EF
H(E)dE = H(E)dE + (µ − EF )H(EF ) (119)
0 0

Applying this result to the previous integrals and substituting µ with EF in the terms that
already were in T 2 we find
Z EF
π2 π2
 
2 0
u= Eg(E)dE + EF (µ − EF )g(EF ) + (kB T ) g (EF ) + (kB T )2 g(EF ) + O(T 4 )
0 6 6
Z EF  2

π
n= g(E)dE + (µ − EF )g(EF ) + (kB T )2 g 0 (EF )
0 6
(120)
The temperature independent first term of both equations is nothing more than the values of
u and n in the ground state. Since we are interested in the specific heat at constant density, n
is independent of temperature and thus the second equation reduces to
π2
0 = (µ − EF )g(EF ) + (kB T )2 g 0 (EF ⇒
6
(121)
π2 g 0 (EF )
⇒ µ = EF − (kB T )2
6 g(Ef )
which determines the deviation of the chemical potential from its zero temperature value. Since
for free electrons the density of levels varies as the square root of the energy, we obtain
 2 !
1 πkB T
µ = EF 1 − (122)
3 2EF

which is a shift of order T 2 , as already noted. This result sets the bracket in the expression for
the energy density to zero, thereby we can write
π2
u = u0 + (kB T )2 g(EF ) (123)
6
where u0 is the energy density in the ground state. Therefore the specific heat is
π2 2
 
∂u
cv = = kB T g(EF ) (124)
∂T n 3
or for one electron
π2
 
kB T
cv = nkB (125)
2 EF
Comparing this result with the classical one, cv = 3nkB /2, we note that the effect of the
2
Fermi-Dirac statistics is to depress the classical contribution by a factor π3 (kB T /EF ) which is
temperature dependent and even at room temperatures is only of order 10−2 , which explains
the absence of any observable electronic contribution to the specific heat.

3.3 Sommerfeld Theory of Conduction


In order to describe the conduction in metals we need to specify the velocity distribution for the
electrons. We consider a small volume d~k about a point ~k in k -space. Considering a two-fold
degeneracy due to the spin, the number of levels inside this volume is
V ~
dk. (126)
4π 3
24
The probability for each level to be occupied is given by the Fermi-Dirac distribution, thus the
total number of electrons in the volume element is
V
f (E(~k))d~k
4π 3 (127)
~2 k 2
E(~k) =
2m
~
Since the velocity of a free electron with wave vector ~k is ~v = ~mk , the number of electrons in the
volume element d~v about the point ~v is the same as the number in the volume d~k = (m/~)3~v
about ~k = m~v /~ and consequently the total number of electrons per unit volume of real space
in a velocity space element is
f (~v )d~v (128)
where
(m/~)3 1
f (~k) =  1
 (129)
4π 3 exp ( 2 mv 2 − µ)/kB T + 1
Sommerfeld replaced the Maxwell-Boltzmann velocity distribution with this Fermi-Dirac ve-
locity distribution in the Drude theory. Since the Drude theory is completely classical, the
application of this quantum mechanical distribution function requires some justification. Skip-
ping all the detailed analytical procedure, we just say that the motion of the electron can be
describe classically if one can specify its position and momentum as accurately as needed with-
out violating the uncertainty principle. Since in a metal a typical electron has a momentum
of order ~kF , the uncertainty on the momentum should be small with respect to ~kF for an
accurate classical description. As we saw at the start of the section, kf ∝ r1s , and we obtain
that the uncertainty on the position of the electron should be of order of rs . Since this is of
the order of the interelectronic distance, a classical description is impossible if one wants to
describe an electron localized between interionic dimensions. Since conduction electrons are
displaced all around the volume of the metal, for a macroscopic specimen an accuracy within
10−8 cm is not required, allowing for a classical description.
The Drude model assumes a knowledge of the position of the electrons primarily in two
cases:
I. In the presence of spatially varying electromagnetic fields or temperature gradients.
Therefore one is required to know the position of the electron to within the typical dis-
tance λ over which the fields vary. Generally this is not a problem, since these kind of
fields do not vary on the scale of the angstroms. A particular case in which the classical
description is not applicable is in the presence of short wavelength EM fields, e.g. x-rays,
in which case a quantum mechanical description is needed.
II. There is the implicit assumption of knowing the electron position to within the mean free
path l. Fortunately, as we will see, mean free paths in metals are in the order of 100Å at
room temperature and it increases lowering T .
Thus there is a number of phenomena that can be described classically
Mean Free Path Using vF as a measure of the electronic speed, we can evaluate the mean
free path l = vF τ as
(rs /a0 )2
l= × 92Å (130)
ρµ
Since the resistivity in microohm centimetres,ρµ , is typically between 1 and 100 at room
temperature and rs /a0 ranges between 2 and 6, mean free paths of hundred angstroms
are possible even at room temperature.

25
Thermal Conductivity We estimate the thermal conductivity from
1
k = v 2 τ cv . (131)
3
The correct specific heat is smaller than the Drude guess by a factor kB T /EF and the
correct estimate of v 2 is vF = 2EF /m which is larger than the classical value by a factor
EF /kB T . Inserting these two results in the expression for the thermal conductivity and
eliminating the relaxation time in favour of the electric conductivity yields
 2
k π 2 kB
= = 2.44 × 10−8 Watt-Ohm/K 2 (132)
σT 3 e

which is remarkably close to the Drude’s luckily good value and in excellent agreement
with the experimental data.

26
4 Crystal Lattices
Now we focus our attention on the microscopic structure of crystals. Although generally one
refers to a solid as a crystal when it presents a macroscopic order, with sharp edges between
adjacent faces, this definition lacks rigour and it is not able to account for crystalline solids
which indeed are crystals. Therefore in this section we will define and describe the main
characteristics of crystals.

4.1 Bravais Lattice


A key concept in the description of any crystalline structure is that of Bravais lattice, which
specifies the periodic arrangement of the repeated units of the crystal. The Bravais lattice
summarizes just the geometry of the underlying of the periodic structure, whatever the nature of
the repeated units, were they single atoms or molecules. There exists two equivalent definitions
of Bravais lattice:

I. A Bravais lattice is an infinite array of discrete points with an arrangement and an


orientation that appear the same from any point;

II. A Bravais lattice consists of all points whose position is given by


~ = n1~a1 + n2~a2 + n3~a3
R (133)

where ~a1 , ~a2 and ~a3 are three vectors not all in the same plane and n1 , n2 and n3 are
integers.

The vectors ~ai are called primitive vectors and are said to span the Bravais lattice.
See that these two definitions are equivalent is not immediate, especially understanding that
any lattice satisfying 1 can be generated by an appropriate set of vectors is not obvious. In
figure we can see two examples where the first definition is clearly satisfied and the primitive
vectors are shown. Of particular value is the second figure which is a simple cubic lattice which
is spanned by three mutually perpendicular vectors.
Of particular importance is the fact that both the arrangement and the orientation must be
the same from any point of lattice. As for example in the hexagonal lattice of figure we see that
the array looks the same only if the page is rotated by 180◦ when viewed from neighbouring
points. Therefore the vertices of the honeycomb are not a Bravais lattice.
The fact each point must be equivalent to the others implies that the lattice is infinite in
extent. Real solids are, of course, finite, but for a macroscopic object the vast majority of
points will be deep in the bulk of the solid and therefore would not feel the existence of the
surface. The idealization to infinite Bravais lattice is useful whenever the surface effects are
not of interest. Although if these effects become of interest, the notion of Bravais lattice is still
relevant, but now we must consider the solid to fill only partially the lattice.
Many times, considering the crystal to be finite is of conceptual convenience, just as we
placed the electron gas in a cubic box of side L. Usually one considers the finite region of the
crystal to be that whose points are spanned by the primitive vectors and the integers ni range
from zero to Ni , with N = N1 N2 N3 is the total number of points in the crystal region.

27
5 X-Ray Diffraction
The typical distance between 2 nearby atoms in a crystal is of the order of the Ångstrom,
therefore an electromagnetic probe shall have a wavelength at least this short. This corresponds
to an energy: ~ω = hc λ
' 12.3 × 103 eV . Energies of this magnitude are characteristic x-
ray energies. To describe how the x-rays are scattered by a crystal there are 2 equivalent
formulations: Bragg and von Laue formulations. Both assume the scattering to be elastic.

5.1 Bragg Formulation


The Bragg formulation accounts for the x-ray scattering by regarding the crystal made out of
parallel planes of ions, spaced a distance d. The diffraction is achieved through the conditions:
(
x-rays are specularly reflected by the ions in any one plane;
reflected rays must interfere constructively.
The optical path between the rays scattered by 2 adjoining planes must differ by 2d sin (θ),
where θ is the incidence angle (measured from the plane). In order to have constructive in-
terference between the scattered rays, the optical path difference must be equal to an integer
number of wavelengths, leading to the Bragg condition:

nλ = 2d sin(θ) (134)

n is known as the order of the corresponding reflection. For white radiation, i.e. a beam
containing different wavelengths, many different reflections are observed. Not only there are
different reflections due to higher orders in a given set of lattice planes, but the crystal can be
divided into different sets of planes, and each of them will contribute to the diffraction pattern.

5.2 Von Laue Formulation


In this case there are no particular hypotheses other than the crystal be made up of identical
microscopic objects placed at the sites R of a Bravais lattice, each of which can reradiate the
incident radiation in any direction. Sharp peaks will be seen only in the directions and at
wavelengths for which the radiation scattered by every point interferes constructively.
To find the condition for constructive interference we consider just 2 scatterers separated
by the vector d. Considering an x-ray coming from far away along the direction n̂, with
wavelength λ, and wave vector k = 2πn̂/λ, it will scattered in a direction n̂’ and with wave
vector k0 = 2πn̂0 /λ. There will be constructive interference if the path difference between the
rays scattered by the 2 scatterers is an integral number of wavelengths. From figure it can be
seen that the path difference is:

d cos (θ) + d cos (θ0 ) = d · (n̂ − n̂0 )

and the condition to have constructive interference is thus

d · (n̂ − n̂0 ) = mλ, m = integer ⇒


⇒ d · (k − k0 ) = 2πm

Now, considering many scatterers, each separated from the others by a lattice vector R, the
condition that all the scattered rays interfere constructively is:
equiv 0
R · (k − k0 ) = 2πm −−−→ ei(k−k )·R = 1, for all Bravais vectors R (135)

28
Then, this equation implies the Laue condition: constructive interference will occur provided
that the change in wave vector, K = k − k0 is a vector of the reciprocal lattice. This condition
can alternatively be stated in terms of the incident wave vector alone: since the scattering is
elastic, both the wave vectors have the same magnitude given by

k = |k − K|

. Taking the square leads to


K
k · K̂ =
2
i.e. the component of the wave vector along the direction of the reciprocal lattice vector must
be equal to half the length of K.

5.3 Equivalence of the 2 Formulations


The equivalence between the 2 formulations is easily seen starting assuming that the incident
and scattered wave vectors, k and k’, satisfy the von Laue condition: K = k0 − k. Since the
scattering is elastic, the two wave vectors have the same magnitude and thus they make the
same angle θ with the bisector plane perpendicular to K. Therefore, the scattering can be seen
as a Bragg reflection with an angle θ from the family of planes perpendicular to K. To see that
this leads to the Bragg condition, it suffices to remember that K is an integral multiple of the
shortest reciprocal lattice vector K0 , whose magnitude is 2π/d, where d is the distance between
two successive planes in the family perpendicular to K0 . Thus, on one hand the magnitude of
K is
2πn
K=
d
and on the other
K = 2k sin(θ)
and thus
πn
k sin(θ) = ⇒ 2d sin(θ) = nλ (136)
d

5.4 Geometries Suggested by the Von Laue Condition


Diffraction peaks will be seen if and only if the tip of the incident wave vector lies on a k-space
Bragg plane. Since every set of planes is a discrete set, there will be no diffraction peaks for a
general wave vector k, i.e. for a fixed X-ray wavelength and incident direction relative to the
crystal axes. Thus, in order to see any peak it is necessary to relax the condition of constant
incident k, changing either its magnitude (i.e. the wavelength) or its direction (i.e. the crystal
orientation relative to the beam).

5.4.1 Ewald Construction


A simple way to visualize the previous discussion is due to Ewald: we draw a sphere in k-space
centred on the tip of the incident wave vector k, starting at the origin, and with radius k. Then,
there will be a wave vector k’ satisfying the von Laue condition if and only if some reciprocal
lattice point lies on the surface of this sphere, other than the origin, in which case there will
be a Bragg reflection.

29
5.4.2 The Laue Method, the Rotating Crystal Method and the Powder Method
However there exists several techniques to assure the observation of some Bragg peaks:
The Laue Method: In order to be sure to get some Bragg peaks, one can scatter from
a single crystal of fixed orientation from a fixed direction n̂ a white X-ray beam containing
wavelengths from λ1 up to λ0 . Then the Ewald sphere would expand to the region contained
between the 2 spheres of radius k0 = 2π/λ0 and k1 = 2π/λ1 and Bragg peaks would be
observed for any reciprocal lattice vector enclosed in this region. This method is best suited
for determining the orientation of a single crystal, whose structure is known, since choosing the
incident direction as one of the symmetry axis of the crystal will reflect the symmetry on the
reflected rays.
The Rotating Crystal Method: In this case monochromatic radiation is used, but the
incident direction is allowed to vary. What it is done in the practice is that the beam is kept
fixed and the crystal is rotated (NB there’s only one axis of rotation). As the crystal rotates,
the reciprocal lattice will rotate of the same angle along the same axis. Thus the Ewald sphere
is fixed in k-space and each reciprocal lattice point moves along a circle. A Bragg peak is
observed for every intersection between these circles and the Ewald sphere.
The Powder Method: This method is equivalent to the rotating crystal, but this time
the rotation axis is varied over all possible directions. This is achieved through a polycrystalline
sample or a powder. Since each single grain has different orientation, the diffraction pattern
is what it would be obtained combining the patterns for each orientation of the crystal. The
Bragg reflections are determined fixing the incident k vector and the reciprocal lattice is allowed
to rotate along all possible angles around the origin, so that each reciprocal lattice vector K
generates a sphere of radius K about the origin. If K < 2k then the corresponding sphere will
intersect the Ewald sphere in a circle. Then, any vector joining a point of this circle with the
tip of k is a vector k’ for which diffraction is observed. Therefore, each reciprocal lattice vector
of length less than 2k will generate a cone of scattered radiation at an angle φ to the forward
direction, where  
φ
K = 2k sin .
2
By measuring the angles φ, the lengths of the reciprocal vectors of lengths less than 2k are
discovered. With this information (plus some facts about the macroscopic symmetry and the
fact that the reciprocal lattice is a Bravais lattice) is possible to reconstruct the reciprocal
lattice.

30
6 X-Rays Interaction with Matter1
The interaction of X-rays with matter consists of 2 processes: absorption and scattering. Both
of these processes posses a characteristic cross section and they can be used to probe several
features of the sample on which the beam shines.

6.1 X-ray Absorption

Figure 1: Photoelectric absorption Figure 2: Fluorescent emission Figure 3: Auger effect

Absorption is observed when the x-rays posses enough energy to expel an electron from an
inner atomic shell leaving an hole in that shell. This process is called photoelectric absorption
and it usually results in fluorescent x-ray emission when the atom relaxes from this excited
state through an electron filling the hole. This consists in the emission of a photon with energy
equal to the energy difference between the shell with the hole and the shell from where the
electron comes. Thus, the monochromatic emitted photon possesses a frequency characteristic
of the nature of the atom, allowing for a non-destructive chemical analysis. Alternatively, this
emitted radiation can knock out another electron from one of the outer shells, resulting in the
Auger electron emission.
The absorption cross section has a particular dependence on the photon energy, as can be
seen in figure 4: as can be seen, below a certain threshold energy, the X-ray photon can expel
only electrons from the outer shells, and the cross section is approximately proportional to E13 ,
where E is the photon’s energy. Once that the photon has an energy equal or greater than the
threshold energy, the so-called K-edge energy, it is able to expel an electron from the innermost
shell and the cross section shows a discontinuous increase of about a decade. Then it continues
to fall as E13 . Studying the fine structure of the absorption around the edge, it is apparent that
it depends on the structure of the material. As can be seen in figure 5, the fine structure can be
studied from the wiggles in the spectrum for a physisorbed 2D lattice of Krypton on graphite.
Another dependence on the photoelectric absorption cross section is on the atomic number Z of
the absorber, which is proportional roughly to Z 4 , and it is exactly this dependence that allows
the use of X-rays as a useful tool for imaging, as for example in the computer axial tomography
(CAT) technique.

6.2 X-ray Scattering


The simplest object off which an X-ray can scatter is an electron. The classical picture of this
process is that the electric field of the incident X-ray exerts a force on the electron, which then
accelerates and radiates the scattered radiation in a spherical wave. The scattered wave has the
same wavelength as the incident X-ray, therefore the scattering is necessarily elastic. In general
this is not true in quantum mechanics, since the incident photon possesses both momentum
~k and energy ~ω. Energy might be transferred to the electron and then the frequency of the
photon changed. This defines the scattering to be inelastic. In both cases it is possible to define
1
This was taken from Jens Als-Nielsen, Des McMorrow ”Elements of Modern X-Ray Physics”

31
Figure 4
Figure 5: Fine structure for physisorbed Kr. χµ is proportional
to the cross section

the transferred momentum as


~Q = ~k − ~k’,
where ~k and ~k’ are the initial and final photon momentum, respectively.
A schematic of a scattering experiment can be seen in figure. The main quantity determined
by such an experiment is the differential cross section, which is defined as:
 
dσ Isc
= (137)
dΩ Φ0 ∆Ω

where Isc is the scattered ray intensity, Φ0 is the incident flux and ∆Ω is the solid angle
subtended by the particle detector.

6.2.1 One Electron


In the case of the scattering of X-rays off an electron the incident flux is proportional to the
incident number density times the speed of light c. The number density is given by the energy
density, proportional to |Ein |2 . This argument is valid also for the scattered intensity, whose
energy density is proportional to |Erad |2 , and must be multiplied by the detector surface, R2 ∆Ω,
where R is the distance from the scattering centre and the detector. Therefore one obtains that

|Erad |2 R2
 

= . (138)
dΩ |Ein |2

In the classical framework, the oscillating electric field of the electromagnetic wave forces the
electron to oscillates, which in returns acts as a source and radiates a spherical electromagnetic
ikR
wave Erad ∝ e R ˆ0 , where ˆ0 is the polarization direction of the wave. To know the exact value
of the field at a point X it would be necessary to solve the Maxwell equations, but here we
will use an heuristic argument in the approximation of far-field limit. We start considering the
observation point to lie on the plane spanned by the polarization vector of the incident wave
and its wave vector, at an angle of π2 − φ from the wave vector. Then, the radiated field is
proportional to the electron charge -e and to the acceleration of the electron as seen from the
observer, evaluated at a time precedent the observation time, since the field propagate at a
finite speed. Therefore it must be:
−e
Erad (R, t) ∝ aX (t0 ) sin(φ), t0 = t − R/c (139)
R

32
where, the term 1/R is needed in order to preserve the conservation of the total energy radiated
and the sine factor accounts for the acceleration as viewed by the observer. To proceed, we
evaluate the acceleration as the force divided by the mass
0
eE0 e−iωt e R e
aX (t0 ) = − = − Ein eiω c = − Ein eikR ,
m m m
where Ein is the field of the incident wave. Hence we obtain

Erad e2 eikR
∝ sin(φ) (140)
Ein m R
If the observation point lies at an arbitrary angle with respect to the polarization of the incident
wave, the factor sin(φ) must be reevaluated. From figure it is easily seen that this factor is the
dot product between the polarization of the incident wave and that of the radiated wave:
π 
ˆ · ˆ0 = cos + φ = − sin(φ)
2
and this writing has the advantage to be valid for any observation angle (this can be seen in
figure 6).

Figure 6

To complete the derivation of the cross section, we must ensure that we have the correct
units. Since the ratio between the electric fields is a pure number, the constant of proportionality
before the spherical wave must have units of length. To find this constant it suffices to remember
e2
that the Coulomb energy at distance r from a charge -e is 4π 0r
, while dimensionally the energy
2
is also equal to mc . Equating these 2 expressions we can obtain the fundamental length scale
of the problem:
e2
r0 = = 2.82 × 10−5 Å (141)
4π0 mc2
This is called the Thomson scattering length or classical radius of the electron. Thus,

Erad (R, t) eikR


= −r0 |ˆ · ˆ0 |,
Ein (t) R

where the factor −r0 |ˆ · ˆ0 | is the scattering amplitude for a free electron. The minus sign means
that there is a π phase shift between the incident and scattered waves (the polarizations are at
an angle of 180◦ from one another). Then, from equation (137) we obtain:


= r02 |ˆ · ˆ|2 (142)
dΩ

33
NB while we obtain the correct form for the magnitude of the scattering amplitude, our treatment is not
capable to give results for the phase shift, which can only be obtained through the rigorous resolution of
Maxwell equations
The polarization factor, P = |ˆ ·ˆ0 |2 , plays a major role in the choice of the optimal geometry
for different experiments with X-rays. For example the synchrotron radiation is always linearly
polarized in the horizontal plane of the synchrotron. This implies that scattering experiments
are best done on the vertical plane since P=1 (as can be inferred from figure 6 when φ = π2 ) in
this case, independent of the scattering angle ϕ = π2 − φ. On the other hand, if the scope of the
experiment is to observe fluorescence from a sample, one will work at best on the horizontal
plane at an angle ϕ = π2 , since P=0. These considerations can be summarised as:

1
 synchrotron radiation on the vertical plane
2
P = |ˆ · ˆ| = cos2 (ϕ) synchrotron radiation on the horizontal plane (143)

1 2
2
(1 + cos (ϕ)) unpolarized source

The total cross section is found integrating the differential cross section over all possible
scattering angles. Exploiting the rotational symmetry of , the mean value of the polarization
factor over a sphere is hP i = 23 and thus the total cross section is
Z
dσ 8π 2
σtot = dΩ = r = 0.665 × 10−24 cm2 = 0.665 barn. (144)
dΩ 3 0

6.2.2 One atom


Now we turn our attention to the case of a single atom with Z electrons. We will describe it in
the pure classical way, and therefore the electron distribution is given by the number density
ρ(~r). The scattered radiation field is the superposition of the scattered radiation from different
volume elements of this charge distribution. To evaluate this superposition it is necessary to
keep track of the phase of the incident and scattered radiation between the element that sits
around the origin and one that sits around the point ~r. For both the incident and scattered
radiation this difference between the two paths is 2π times the distance |~r| divided by the
wavelength, i.e. (
∆φin = ~k · ~r
∆φsc = −k~0 · ~r
and therefore the total phase difference is

∆φ(~r) = (~k − k~0 ) · ~r = Q


~ · ~r (145)

where Q~ is the scattering vector.


Thus, a volume element d~r at ~r will contribute to the scattered field an amount −r0 ρ(~rd~r
~
with a phase factor eiQ·~r , and therefore the total scattering length of the atom is
Z
~ ~
− r0 ρ(~r)eiQ·~r d~r = −r0 f 0 (Q) (146)

~ is known as the atomic form factor.


where f 0 (Q)
~ → 0 all the volume elements scatter in phase and the form factor has its
In the limit of Q
maximum value: f 0 (0) = Z, the number of electrons in the atom. As the scattering vector
increases from 0 to greater values the volume elements scatter out of phase and in the limit
Q~ → ∞ the form factor assumes its minimum value: f 0 (∞) = 0. It should be noted that the
atomic form factor is nothing else but the Fourier transform of the charge density.

34
In reality it is observed that the energy of the X-ray photons modulates the scattering
intensity. This is due to the fact that the K-shell electrons are tightly bound to the nucleus
and if the energy of the photons is much lower than the binding energy, the K-shell electrons’
response will be reduced (in virtue of the binding), whereas we expect the electrons of the less
tightly bound shells to respond more closely to the driving field. Thus we expect the scattering
length to be reduced of some amount, conventionally indicated as f 0 (~ω), i.e. it depends on
the X-ray energy but not on the scattering vector. At energies much greater than the binding
energy the electrons can be treated as free and the correction factor is null. In the intermediate
regime, the factor f 0 shows resonant behaviour in concurrence with the atomic absorption edges.
Moreover, we expect the electrons’ response to alter not only the real part of the scattering
length, but also to have a phase lag which is accounted by an imaginary factor if 00 (~ω), which,
in analogy with the harmonic oscillator, represents the dissipation in the system. Thus the
atomic factor becomes
~ ~ω) = f 0 (Q)
f (Q, ~ + f 0 (~ω) + if 00 (~ω) (147)
and the two correction factors are called dispersion corrections. It should be noted that these
terms assume their extremal values at energies corresponding to absorption edges since these
energies represent resonances in these terms. They are written as functions of the X-ray energy
to emphasize the fact that their behaviour is dominated by the inner shell electrons.

6.2.3 One Molecule


The next step in complexity from the atom is a collection of atoms, i.e. a molecule. As the
atom is characterized by the form factor, so it is possible to define a form factor for a molecule:
indexing the atoms in the molecule with j, we obtain
~ r~j
X
~ =
F mol (Q) ~ iQ·
fj (Q)e (148)
j

To get the scattering length it suffices to multiply the form factor by −r0 . Measuring the form
factor for an high enough number of values of the scattering vector Q ~ it is possible to determine
the positions r~j of the atoms in the molecule. However it is not possible to study the structure
of a single molecule, since its scattering length is not sufficient to produce a measurable signal.
Therefore, to study the molecules’ structure they are arranged in crystals.

6.2.4 Crystal
~j and a unitary cell, which determines which
A crystal is characterized by its Bravais lattice R
atom is associated with which lattice point. For a mono atomic crystal without a basis the
form factor has the form
X ~ R~j
~
F crys (Q) X ~ ~
~ =
F crys (Q) ~ iQ·
fj (Q)e ⇒ = eiQ·Rj ≡ geometrical structure factor
~
f (Q)
j j
(149)
For a Bravais lattice with a basis with N unit cells and M atoms per cell, the positions of the
atoms are given by rl = R ~j + d~m where l = 1, . . . , N × M . It results that the form factor of the
whole crystal factorizes into the product of 2 terms
~ d~m
X X ~ ~
~ =
F crys (Q) fm (Q)e~ iQ· eiQ·Rj (150)
|m {z } | j {z }
unit cell lattice sum
structure factor

35
In applications such as solid state physics one is interested in the structure of the material,
whereas in molecular and protein crystallography the lattice serves only as a signal amplifier
and thus the lattice is of no interest whatsoever.
The terms of the lattice sum are all phases that lie on the unit circle in the complex space
and thus the sum will generally be of the order unity, except when
~ ·R
Q ~ = 2nπ

in which case it becomes of the order N. This implies that the crystal form factor is non
~ is a vector of the reciprocal lattice. This coincides with the von Laue
vanishing if and only if Q
condition in order to have refraction peaks from a crystal.
The lattice sum is therefore of the form
N −1 N −1
X
i2πnx
X
n1 − aN 1 − ei2πxN
SN (x) = e → a = → =
1−a 1 − ei2πx
n=0 n=0 (151)
e−iπxN − eiπxN eiπxN sin(πN x) iπ(N −1)x
= −iπx iπx iπx
= e
e −e e sin(πx)
1D Lattice In one dimension the lattice points are specified by Rn = n a, with n an integer
and a the lattice spacing. For a finite crystal with N atoms the lattice sum is
N −1
sin N Qa

X
iQna 2
SN (Q) = e ⇒ |SN (Q)| = Qa
 .
n=0
sin 2

This results in sharp peaks for


Qa 2π
= nπ ⇒ Q=n
2 a
and thus the scattering vector is a reciprocal lattice vector and therefore the sharp peaks are
generated when the von Laue condition is fulfilled. To study the behaviour near the von Laue
condition we use the small parameter ξ:
2π sin (N πξ)
Q = (n + ξ) ⇒ |SN (ξ)| = −−→ N
a sin(πξ) ξ→0
1
Its width might be estimated setting ξ = 2N :
 
SN ξ = 1 = 1 2N N

π
' '
2N sin 2N π 2

Thus the peak height is N and the full width at half maximum is approximately 1/N , the peak
area is about unity. Therefore, in the limit N → ∞ we an write the modulus of the lattice sum
as
|SN (ξ)| = δ(ξ) (152)
which can be written in the more general way as
2π X
|SN (Q)| = δ(Q − Gn ) (153)
a G
n

where the prefactor arises from from


 
2π δ(ξ)
δ(Q − Gn ) = δ ξ = 2π
a a

36
Since in scattering experiments we are interested in the intensity, which is proportional to the
squared modulus of the lattice sum, from analogous considerations we obtain
2π X
|SN (Q)|2 = N δ(Q − Gn ) (154)
a G
n

2D and 3D Lattice In this case the unit cell is spanned by 2 basis vectors a~1 and A~2 .
We consider the special case where the macroscopic crystal is a parallelepiped, in which case
each column has the same number N1 of cells independent of the rows 1, . . . , N2 . Following the
procedure for the 1D case we obtain
|SN (ξ1 , ξ2 )|2 −−−−−→ N1 N2 δ(ξ1 )δ(ξ2 )
N1 ,N2 1

which, exploiting the deltas as before we can rewrite as


  
~ 2π 2π X ~ X
2
|SN (Q)| = N1 N2 δ(Q − G~n ) = N A ~ − G)
δ(Q ~ (155)
a1 a2
G~n ~
G

where N = N1 × N2 is the total number of cells in the crystal and A is the unit cell area in
the reciprocal lattice. In the general case, one cannot evaluate analytically the sum and then
square to look at the limiting behaviour for a big number of unit cells, but the δ character is
preserved, provided that the number of unit cells is big enough in both directions.
This result is easily generalized to the 3D case. The summation can be carried out analyti-
cally only if the crystal is a parallelepiped, whereas for a general shape it cannot. Nevertheless,
provided that the number of unit cells is big enough, the result is, independently of the actual
crystal shape: X
~ 2 → N νc
|SN (Q)| ~ − G)
δ(Q ~ (156)
~
G
where N is the total number of cells, νc is the volume of the unit cell in reciprocal space, and
~ are the reciprocal lattice points.
G

6.2.5 Unit Cell Structure Factor


We now turn our attention in the evaluation of the unit cell form factor for some examples.
Face Centred Cubic The first example that we will study is that of the fcc structure,
choosing as conventional cell a cubic unit cell since it reflects better the symmetries of the
crystal. with this convention the lattice is a simple cubic with lattice spacing a and a basis of
4 atoms which sit in 

 r~1 = 0
r~ = 1 (a~ + a~ )

2 2 1 2
1


 r~3 = 2 (a~1 + a~3 )
r~4 = 12 (a~2 + a~3 )

where the vectors a~i define the conventional unit cell:



a~1 = aî

a~2 = aĵ

a~3 = ak̂


Thus, the reciprocal lattice is also simple cubic with lattice spacing a
and the reciprocal lattice
vectors are of the form
~ = 2π (h, k, l).
G
a
37
If for simplicity we consider the crystal to be mono atomic, then the atomic form factor can be
factorized and we obtain that the unit cell structure factor is
4
~
X
f cc
Fhkl ~
= f (G) ~
eiG·r~j = f (G)(1 + eiπ(h+k) + eiπ(h+l) + eiπ(k+l) ) =
j=1
( (157)
~ · 4 if h, k, l are all either even or odd
= f (G)
0 otherwise

Therefore we obtain that for certain reflections F f cc is 0 and these reflections are called forbidden
reflections. This effect is due to the fact that we describe the crystal through a conventional
cell.
Diamond Structure There are several ways to consider the structure of diamond, and the
one that we will use is to think about it as an fcc with a 2 atom basis. In this description we
choose the unit cell to be defined by the basis vectors

a
a~1 = 2 (ĵ + k̂)

a~2 = a2 (î + k̂)

a~3 = a2 (î + ĵ)

thus, with this basis vectors the basis atoms sit at


(
d~1 = 0
d~2 = a4 (î + ĵ + k̂)

The reciprocal lattice is a body centred cubic with basis vectors



~ 2π
b1 = a (ĵ + k̂ − î)

b~2 = 2π
a
(î + k̂ − ĵ)
~
 2π
b3 = a (î + ĵ − k̂)

Therefore the unit cell structure factor is


X ~ ~
diam ~ ~ 1 + ei π2 (h+k+l)
eiG·dj = f (G)

Fhkl = f (G) ⇒
j

diam
Fhkl 2
 h + k + l is twice an even number (158)
⇒ = 1±i h + k + l is odd
~
f (G) 
0 h + k + l is twice an odd number

also in this case we see the presence of forbidden reflections.


Zincblende This is a variation of the diamond structure where the two basis atoms are
different one from the other and it is common to the semiconductors (e.g. GaAs, InSb). In the
specific case of GaAs the unit cell structure factor is
π
GaAg
Fhkl ~ + f As (G)e
= f Ga (G) ~ i 2 (h+k+l) (159)

Since the unit cell has different kind of atoms, in this case is possible to see reflections that for
the mono atomic case were forbidden.
NB it is also possible to see scattered radiation for forbidden reflections in the case of multiple scattering
inside the crystal

38
Figure 7

6.3 Scattering From a Crystallite


6.3.1 Qualitative Origin of Bragg Peaks’ Broadening
Here we want to draw a qualitative picture on the breadth of the Bragg peak and where it
originates from. This outlining follows what we have done in the evaluation of the lattice sum
for the 1 dimensional case.
We obtained
N −1 Qa

X sin N
SN (Q) = eiQna ⇒ |SN (Q)| = Qa
2

n=1
sin 2

We are interested in the full width at half maximum ∆Q. This can be done in the following
manner: we consider the point Q + ∆Q 2
which is the point where the half maximum height lies:
  2 2

SN Q + ∆Q =N

2 2

Then, if we study the principal maximum (Q = 0) we have:

sin N ∆Qa
    
4 N ∆Qa N ∆Qa N ∆Qa
= √ ⇒ sin N = √ sin ' √
sin ∆Qa

4 2 4 2 4 2 4

which can be solved numerically to yield


∆Qa 1.39 5.56
= ⇒ ∆Q = (160)
4 N aN
where aN is the size of the sample.

6.3.2 Scherrer’s Formula


It should be noted that destructive interference of the scattered rays is due to the periodic
structure of the crystal, in a fashion similar to the constructive interference. Thus, in an ideal
infinite crystal, if the optical paths between the rays scattered by the first two planes differ
only slightly from an integral number of wavelengths, then the plane that will scatter a ray
totally out of phase with respect to the ray scattered from the first plane will lie deep within
the crystal. Therefore for a real crystal with a finite dimension such a plane could not exist
with the result that these X-rays are not completely suppressed. This results in a broadening
of the reflected beam from small crystals. This means that there will be a small divergence
in the scattered beam, i.e. there will be scattered intensity even in the vicinity of the exact
Bragg angle. The deviations from the Bragg angle will be greater the smaller the dimension of
the crystal, but nevertheless there will be two limits to the angle range for which the intensity

39
will be diverse from 0, θ1 and θ2 . This means that Bragg reflections will be observed for Bragg
angles that go from 2θ2 to 2θ1 .
The width of the scattered intensity ∆(2θB ) is usually given as the width of the peak at
half its height and is called Full-Width at Half-Maximum. ∆(2θB ) is roughly given as one half
the difference between the limits of the range at which the Bragg reflection is observed:
1
∆(2θB ) = (2θ1 − 2θ2 ) = θ1 − θ2
2
The path difference equations between these two angles are similar to the Bragg equation,
but related to the thickness of the entire crystal rather than just to the distance between two
adjacent planes (
2t sin(θ1 ) = (m + 1)λ
2t sin(θ2 ) = (m − 1)λ
where t is the thickness of the crystal and m is the order of the m-th plane that produces
incomplete destructive interference with the ray scattered by the surface plane. Subtracting
these 2 equations we get
   
θ1 + θ2 θ1 − θ2
t(sin(θ1 ) − sin(θ2 )) = λ ⇒ 2t cos sin =λ
2 2

but, since both θ1 and θ2 are close to θB we obtain


θ1 − θ2 λ
2t cos(θB ) = λ ⇒ t= (161)
2 ∆(2θB ) cos(θB ).

A more rigorous treatment results in a correction factor ' 0.88:


0.88λ
t= (162)
∆(2θB ) cos(θB )

which is known as Scherrer’s formula. It should be noted that the numerical factor varies
according to the shape and dimensionality of the crystallites assumed to be in the sample (e.g.
0.88 is the value obtained for 1D crystallites).

6.3.3 Scattered Intensity from Crystallites


We are now interested in the exact evaluation of the integrated intensity of a Bragg reflection,
but to be more precise we will evaluate the differential cross section, that is the quantity that
is directly measured. From what we obtained in the previous sections, the differential cross
section is given by  
dσ ~ 2 N ν ∗ δ(Q
~ − G)
~
= r02 P |F (Q)| c
dΩ
where we suppressed the superscript of the unit cell structure factor and P is the polarization
factor.
An example of the experimental set up can be seen in figure. Generally we assume the
incident beam to be monochromatic and collimated, thus we can expect the scattered beam
to be monochromatic, since the scattering is elastic, but we cannot expect it to be collimated,
since, as we saw, the Bragg peak width is proportional to the inverse of the number of unit cells
N, and for a finite N the peak has a finite width. This means that the Von Laue condition does
not need to be exactly fulfilled to have a measurable intensity. As can be seen in figure, this
is represented with an elliptical contour, and for every Q~ that falls within this contour there

40
will be some measurable intensity. If we assume that the geometry is such as to let all of the
slightly divergent scattered rays to be collected by the detector, we get that we collect only the
rays scattered by a thin sleeve inside the broadened contour. Since we are interested in the
scattering from the whole structure, the small crystal must be rotated a little in order to get
the scattering from the whole surface inside the contour, since we are interested in the Q ~ that
terminates inside the whole contour. In this way the integrated intensity is accumulated.
Thus the rotation of the crystal implies a variation of the angle θ. This results in the need
to integrate the differential cross section both k~0 and θ.
We start integrating over every possible direction of k~0 . This is done integrating over the
unit vector k̂ 0 which results in a 2 dimensional integral. We introduce the vector ~s = k 0 ŝ, where
ŝ is a unit vector. Then the problem is integrating the delta function over dk̂ 0 :
Z Z
δ(Q ~ k̂ = δ(~k − k~0 − G)d
~ − G)d 0 ~ k̂ 0 =
Z Z
2
= 2 2 02
s δ(s − k )ds δ(~k − ~s − G)dŝ
~
k0 (163)
| {z }
=1 , since:
I(k)= x2 δ(x2 −k2 )dx k>0 →
R
dt
→ t=x2 −k2 ⇒ dx =2x →
→ I(k)=x dt |t=0 = k2
2 dx

The trick of using s instead of k’ allowed us to insert the first integral, since in these terms
it is a ”decomposition of the unity”, allowing us to write the 2D integral as a 3 dimensional
integral. This can be seen in a clearer way rearranging the expression above in the following
manner: Z Z
2
δ(~k − k~ − G)d
0 ~ k̂ =
0 δ(s2 − k 2 ) δ(~k − ~s − G)
~ d~s ⇒
k0 | {z } |{z}
2 =s dsdŝ
⇒~s=~k−G
~
Z
~ k̂ 0 = 2 δ (~k − G)
 
⇒ δ(~k − k~0 − Gd ~ · (~k − G)
~ − k2 =
k0
2
= δ(G2 − 2 ~k ·G~ )
↑ k | {z }
k=k 0 =kG sin(θ)

Now we are left to the integration over the angle θ. Since the θ dependence is in the delta
function as a function of θ we can use the identity:
"  #
Z −1
dt
δ(t(θ))dθ = .

t=0

The derivative of the argument of the delta function is:


d(G2 − 2kG sin(θ))
= −2kG cos(θ)

which results into
 
−1 −1
Z
2
δ(G − 2kG sin(θ))dθ = = (164)
2kG cos(θ) t=0
↑ 2k 2 sin(2θ)
t=0⇔G=2k sin(θ)

Putting the results we got together we obtain that the total differential cross section is:
λ3 1
 
dσ 2 ~ 2 ∗2 1 2 ~ 2
= r0 P |F (Q)| N νc = r0 P |F (Q)| N (165)
dΩ int. over k~0 ,θ k 2k 2 sin(2θ) ↑
3
νc sin(2θ)
(2π)
νc = νc∗

41
1
The factor sin(2θ) is known as the Lorentz factor.
Thus, the total integrated intensity is found multiplying the above expression by the incident
flux Φ0 :
3
   
photons photons 2 ~ 2N λ 1
Is c = Φ0 r0 P |F (Q)| (166)
s s · area νc sin(2θ)
This formula applies only for small, but otherwise perfect crystals. Real crystals suffer of ex-
tinctions in the diffracted intensity due to atomic vibrations, the presence of mosaics (domains)
with different orientations, and, in the case of a non sufficiently small crystal, multiple scat-
tering (due to the loss of the kinematic approximation). This last problem is solved through
powder diffraction.

6.4 Powder Diffraction


An ideal crystalline powder is made up of thousands of randomly oriented crystallites. If
we focus our attention to a particular reciprocal lattice vector, Ghkl , specified by the Miller
indexes (h, k, l ), we see that in the powder sample the directions of this vector are isotropically
distributed over a sphere, as seen in figure. Given a particular incident wave vector ~k, only a
small fraction of the crystallites will be oriented in a way such to allow for Bragg reflection.
In the figure they are represented by the circle obtained by the cross cut of the sphere by the
plane orthogonal to ~k. Therefore the scattered wave vectors k~0 are evenly distributed on a
cone which has the incident wave vector has axis and apex half angle 2θ. This cone is called
Debye-Scherrer cone.
We now want to know how the measured intensity from powder diffraction is related to
the structure factor. For a certain (h, k, l ) the number of grains oriented to reflect is pro-
portional to the circumference of the Debye-Scherrer cone’s base. The circumference is given
π

by: Ghkl sin 2 − θ = Ghkl cos(θ). Due to the randomness of the orientation of the grains,
permutations of the indexes (h, k, l ) may have the same sphere, and therefore we introduce
the multiplicity of the reflection, mhkl . For example the reflection (h, 0, 0 ) from a simple
cubic lattice has mh00 = 6, since the (±h, 0, 0), (0, ±h, 0) and (0, 0, ±h) reflections have the
same Bragg angle 2θ. Thus, the intensity at the detector will be proportional to mhkl cos(θ).
Changing the Ghkl and hence the angle 2θ will change the fraction of the base circle seen by the
detector. Independent of Ghkl the circumference is 2πk sin(2θ). Since the detector covers the

solid angle δ, the fraction of the circle that it sees is: 2πk sin(2θ) . Finally, the intensity scattered
1
from a single crystallite will be proportional to the Lorentz factor, which is sin(2θ) . Thus, the
observed intensity will be given by:

cos(θ) 1
Isc ∝ Lpowder = mhkl = mhkl
sin(2θ) sin(2θ) 2 sin(θ) sin(2θ)

The structure factors squared can be evaluated on a relative scale from the observed intensities.
If for example we need to evaluate the ratio of the squared structure factors between the (1,
1, 1 ) and the (2, 0, 0 ) reflections of a fcc crystal, we will obtain, taking into account also the
polarization factor P, which depends on the scattering angle, that the ratio of the intensities is:

I111 |F111 |2 Lpwdr (θ111 ) P (cos(2θ111 ))


=
I200 |F200 |2 Lpwdr (θ200 ) P (cos(2θ200 ))

42
7 Classification of Bravais Lattices
In the preceding sections we described and exploited only the translational symmetries of
Bravais lattices. Although translational symmetries are the most important for the general
theory of solids, it is clear from several examples that Bravais lattices fall into categories
of symmetries other than translational. The systematization of these properties is subject
of crystallography. Here we shall only outline the fundamental concepts of crystallographic
classifications since in most applications what matters are the features of particular cases rather
than a general theory.

7.1 Classification of Bravais Lattices


As far as the symmetries concern, a Bravais lattice is characterized by the specification of all
the rigid operations (i.e. operations that preserve the distances between lattice points) that
take the lattice into itself. This operations include all the translations through lattice vectors,
rotations, reflections and inversions. This set of operations is called the space group of the
Bravais lattice.
Any symmetry of a Bravais lattice can be made up of a translation TR~ through a lattice
vector R~ and a rigid operation that leaves at least one point fixed. This can be seen in the
following way: let us consider the symmetry operation S that leaves no point fixed. Suppose it
takes the origin O~ to the point R.~ If we consider S composed with a translation through −R ~
we obtain the composite operation T−R~ S which is also a symmetry of the lattice but leaves the
origin fixed. Thus the composite operation belongs to the space group.If we want to recover
the symmetry operation S, it suffices to apply a translation through R ~ after the composite
operation, therefore S can be compounded out of T−R~ S, which leaves a point fixed, and T−R~
which is a pure translation. Thus the full symmetry group of a Bravais lattice contains only
the operations of the form:

I. Translations through Bravais lattice vectors;

II. Operations that leave a certain point fixed;

III. Operations compounded out of operations of the preceding kinds.

Operations of the (2) kind belong to the point group of the Bravais lattice.
If one focuses only on the point groups, it turns out that there are only 7 distinct point
groups that a Bravais lattice can have. Depending on which of these groups is the group of the
underlying Bravais lattice the crystal structure belongs to one of the seven crystal systems.
If the full symmetry of the Bravais lattice is considered, it turns out that there are 14
distinct space groups that a Bravais lattice can have. Thus, from a symmetry point of view
there are only fourteen different kind of Bravais lattice.
The seven crystal systems and the Bravais lattices belonging to each are described in the
following. In parentheses is given the number of Bravais lattices for each system:
Cubic (3) This system contains the Bravais lattices that own the same symmetry prop-
erties of a cube. These three lattices are the simple cubic, body-centred cubic and face-centred
cubic.
Tetragonal (2) The cubic symmetry can be reduced by pulling on two opposite sides
generating a rectangular prism with square base. The symmetry group of such an object is the
tetragonal group. Stretching the simple cubic generates the simple tetragonal Bravais lattice,
which is generated by three mutually orthogonal primitive vectors, only two of which are of the
same length (the third axis is called c-axis). In the same fashion, from either the bcc or the fcc
only one more Bravais lattice is constructed, i.e. the centred tetragonal. The fact that both the

43
fcc and bcc generate the same Bravais lattice of the tetragonal system can be inferred from the
picture.
Orthorhombic (4) Keeping on reducing the symmetries of the cube, the tetragonal struc-
ture can be deformed to obtain a structure with three mutually orthogonal primitive vectors,
each with unequal lengths. The orthorhombic group is the symmetry group of this object.
Stretching the simple tetragonal along an a-axis we obtain the simple orthorhombic Bravais
lattice, whereas stretching it along a square diagonal we produce the base-centred orthorhom-
bic. By stretching the centred tetragonal along parallel lines as in figure(figura precedente,
primo pannello) or along the parallel lines in figure(figura precedente, secondo pannello) we
obtain the body-centred orthorhombic and the face-centred orthorhombic, respectively.
Monoclinic (2) Distorting the rectangular faces perpendicular to the c-axis of the or-
thorhombic into general parallelograms results into the monoclinic group. From the simple
orthorhombic we obtain the simple monoclinic Bravais lattice (distorting the base-centred or-
thorhombic results in the same space group). By distorting either the face-centred or body-
centred orthorhombic we obtain the centred monoclinic Bravais lattice. These groups have no
more symmetries than those required by the fact that they can be generated by three prim-
itive vectors, one of whose is perpendicular to the plane of the other two. These two groups
correspond to the two tetragonal.
Triclinic (1) Tilting the c-axis so that it is no more perpendicular to the other two
completes the destruction of the cubic symmetries, obtaining the triclinic Bravais lattice. This
object has the only request that pairs of opposite faces are parallel. This group is the group
of least symmetry, however it is not the group of an object without any symmetry, since every
Bravais lattice is invariant under an inversion in a lattice point (this is the only symmetry
operation in the triclinic group).
Trigonal (1) The trigonal point group describes the symmetry of the object obtained by
stretching a cube along the body diagonal. The lattice produced by deforming any of the cubic
lattices is the rhombohedral Bravais lattice. It is generated by three primitive vectors of the
same length that make equal angle with one another.
Hexagonal (1) This point group is the symmetry group of a right prism with a regular
hexagon as base. The simple hexagonal Bravais lattice is the only lattice with this point group.

7.2 Crystallographic Point Groups and Space Groups


We now want to describe the results of a similar analysis applied to general crystal structures
(i.e. structures whose basis does not have spherical symmetry, as in the Bravais lattices case).
The structure is obtained by translating an arbitrary object through Bravais lattice vectors.
Therefore the symmetry of the structure will depend both on the symmetry of the lattice and
that of the object. Since these objects do not have the maximum symmetry any longer, the
number of space groups is greatly increased: it turns out that there are 230 space groups. Also
the possible point groups for the general crystal structure have been enumerated. There are 32
crystallographic point groups.
Each of the 32 crystallographic point group can be constructed out of the 7 Bravais lattice
point groups by systematically reducing the symmetry of the objects considered by these groups.
These new 25 groups are associated to one of the 7 crystal symmetry according to the simple
rule: a group is associated to a particular crystal system until its symmetry is reduced to the
point that all of the remaining symmetry operations belong also to a system of less symmetry.
Thus the crystal system of a crystallographic point group is that of less symmetry of the Bravais
lattice point groups containing every symmetry operation of the crystallographic group.
In the crystallographic point groups there may be the following symmetry operations:

I. Rotations through integral multiples of n
about some axis The axis is called

44
n-fold rotation axis. It can be shown that a Bravais lattice point group can contain only
2-, 3-, 4- and 6-fold axes, and therefore also the crystallographic point groups contain
only these axes.

II. Rotation-Reflections Even though sometimes a rotation through 2π n


is not a symmetry
element, such a rotation followed by a reflection about a plane orthogonal to the rotation
axis might be a symmetry. The axis is then called n-fold rotation-reflection axis.

III. Rotation-Inversions In a similar fashion, a rotation through 2π n


followed by an inversion
about a point of the rotation axis can be a symmetry element, even though the rotation
itself is not. Then the axis is called an n-fold rotation-inversion axis.

IV. Reflections A reflection takes every point into its mirror image in a plane, known as a
mirror plane.

V. Inversions An inversion keeps fixed only one point. If this point is taken as origin, then
every point ~r is brought into the point −~r.

7.3 Point Group Nomenclature


There are two widely used nomenclatures: the Schönflies and the international.
Schönflies notation for the noncubic crystallographic point groups
Cn : These groups contain only an n-fold rotation axis.

Cnv : Other than the n-fold rotation axis, there is a mirror plane that contains the rotational
axis, plus as many mirror planes as the existence of the n-fold axis requires.

Cnh : In addition to the n-fold axis there is a single mirror plane orthogonal to the axis.

Sn : These groups contain an n-fold rotation-reflection axis.

Dn : In addition to the n-fold rotation axis there is a 2-fold axis orthogonal to the n-fold axis,
plus every other 2-fold axis required by the existence of the n-fold axis.

Dnh : These groups contain all the elements of Dn plus a mirror plane orthogonal to the n-fold
axis (these are the most symmetrical groups).

Dnd : They contain the elements of Dn plus the mirror planes that contain the n-fold axis that
bisect the angles between the 2-fold axes.
International notation for the noncubic crystallographic point groups
Three categories are the same of three of the Schönflies categories:
n : It is the same as Cn .

nmm : It is the same as the Cnv . The two m’s refer to different mirror planes in the case
of a 2j-fold axis. Indeed a 2j-fold axis takes a vertical mirror plane into j mirror planes
containing the axis, but in addition there are automatically other j planes that bisect the
angles between adjacent planes of the first set. However, in the case of a (2j+1)-fold axis,
the vertical plane is taken into (2j+1 ) mirror planes, and therefore the C3v is called 3m.

n22 : It is the same as Dn . The discussion is the same as for the nmm, but now we have
orthogonal 2-fold axes instead of vertical mirror planes.
The other international categories are:

45
n/m : It is the same as the Cnh , but the C3h is put into another category. Note that the C1h
becomes just m instead of 1/m.

n̄ : This indicates a group with an n-fold rotation inversion axis. It contains the C3h disguised
as a 6̄. It contains also the S4 as 4̄. However, S6 becomes 3̄ and S2 becomes 1̄, by virtue
of the difference between rotation-reflection and rotation-inversion axes.
n 2 2
mmm
, or n/mmm : This is just Dnh , with the exception of D3h , which falls into another
category. It should be noted that 2/mmm is further abbreviated in mmm. This notation
would suggest an n-fold axis with an orthogonal mirror plane plus two sets of 2-fold axes,
each with its own perpendicular mirror planes.

n̄2m : It is just Dnd with the addition of D3h , denoted as 6̄2m. The name is intended to
suggest an n-fold rotation-inversion axis qith a perpendicular 2-fold axis and a vertical
mirror plane. Also in this case the n = 3 is exceptional, denoted as 3 m2 or simply 3̄m

Nomenclature for the cubic crystallographic point groups


Schöflies nomenclature:

Oh : It is the full symmetry group of the cube (the O comes from octahedron) plus the improper
operations that the horizontal reflection plane (h) admits (the improper operations are
those that take a right-handed object into a left-handed one, as for example an odd
number of reflections or inversions).

O : It is the cubic group without the improper operations.

Td : Full tetragonal symmetry group with improper operations.

T : Tetragonal group excluding the improper operations.

Th : Tetragonal with the adding of the inversion.

International nomenclature:
The cubic groups are distinguished from the other crystallographic point groups by having
the second number equal to 3, referring to the 3-fold axis present in the cubic group.

7.4 The 230 Space Groups


We now focus a little bit on the number of crystallographic space groups. Indeed it is greater
than one can have thought. We can construct a crystal structure with a different space group
from each crystal system by placing an object with the symmetries of each point group of the
system into each Bravais lattice of the system. However in this way we find only 61 space
groups. Five more can be found noting that placing an object of trigonal symmetry into an
hexagonal Bravais lattice yields a space group not yet enumerated. Another 7 are obtained
from the case where the object with a given symmetry can be oriented in more than one way
in the Bravais lattice. These 73 space groups are called symmorphic.
The great majority of the space groups are nonsymmorphic, meaning that they contain
operations that cannot be simply made up of translations through Bravais lattice vectors and
point group operations. For the existence of these different operations there must be some spe-
cial relation between the dimensions of the basis and that of the lattice. When the dimensions
are suitably matched, two new operations arise:
Screw axes A crystal structure with a screw axis is brought into itself by a translation through
a vector not in the Bravais lattice followed by a rotation about the same axis.

46
Glide planes If a crystal structure has a glide plane it is brought into itself by a translation
through a vector not in the Bravais lattice followed by a reflection in a plane containing
that vector.

The hexagonal close-packed structure offers example of both type of operation as shown in
figure.

7.5 Examples among the Elements


Over 70 percent of the elements fall into the fcc, bcc, hexagonal close-packed or diamond
crystal structures, while the remaining are scattered among several crystal structures, most
with polyatomic primitive cells that can be quite complex from time to time.

47
8 Diffraction from Non-Crystalline Materials
In opposition to a crystalline material, whose atoms sit at the points of a regular array, the
atomic distribution in a non-crystalline material has a degree of randomness. This results in
the fact that the structural order of such a material, if any, can be deduced only in a statistical
sense. As strange as it may sound, in disordered systems like liquids and glasses there may
exist well defined, short-range structural correlations that can be studied with great details via
x-ray scattering techniques.

8.1 The Radial Distribution Function


In non-crystalline materials the atoms change position over a wide range of time scales, from
the nanoseconds in the case of liquids, to several millennia for glasses. But, since the probes
used are x-rays, which are a fast probe, we are justified to consider the atoms to be still
(i.e. we consider an instantaneous snapshot of the structure). Then, our task is to provide
a statistical description of the structure frozen in time and to perform an average over every
possible configuration. The first objective can be achieved defining the radial distribution
function, g(r).
The radial distribution is evaluated starting choosing an atom as origin (this choice is
arbitrary, since in an experiment we get an average over many snapshots). If we consider a 2D
system, we can define the radial density as:

N (r)
ρ(r) =
2πrdr
where N (r ) is the number of atoms inside the annulus of radius r and width dr, and 2πrdr is
the surface of the annulus. Then the radial distribution function is defined as:
ρ(r)
g(r) =
ρa
where ρa is the average areal number density. In the 3D case, instead of an annulus we shall
consider a spherical shell of thickness dr and the radial distribution function is:
 
1 N (r) N
g(r) = , ρ at = .
ρat 4πr2 dr V

Calculating g(r) by taking a series of concentric circles in both the crystalline and non-
crystalline cases we see that the differences become evident when the spacing between the
circles decreases. In the crystalline case, the distribution function is characterized by a series
of sharp, non.overlapping peaks that extend to indefinite big r, whereas in the non-crystalline
case broad peaks are shown that decrease as r increases to quickly reach unit amplitude, since
ρ(r) becomes equal to the average ρa . Although this is just a toy model that should not be
taken too seriously, it can reproduce quite well some experimental features of non-crystalline
materials, e.g. the first peak in g(r) is a measure of the minimal distance of approach between
two atoms (in liquids two atoms or molecules are forbidden to occupy the same volume and in
glasses there is a minimum average distance for rigid chemical bonds).

48
8.2 The Liquid Structure Factor
For simplicity we consider the mono-atomic or mono-molecular system case. The scattered
intensity, in suitable dimensionless units can be written as
X ~ X ~
X ~
~ = f 2 (Q)
I(Q) ~ eiQ·~rn ~
e−iQ·~rm = f 2 (Q) eiQ·(~rn −~rm ) =
↓ n m n,m
monoatomic
system
(167)
 
~
XX
~ 
=f 2 (Q)  N + eiQ·(~rn −~rm ) 


from the sum n n6=m
forn=m

Next we want to replace the sum for n 6= m by an integral over the volume. Moreover, since
the X-ray scattering is due to fluctuations in the electron density from its mean, we add and
subtract a term proportional to the average density ρat . Then we get:
~
=ISRO (Q)
z }| {
XZ ~ rn −~
~ = N f 2 (Q)
I(Q) ~ + f 2 (Q)
~ (ρn (~rnm ) − ρat ) e iQ·(~ rm )
dVm +
n V
XZ (168)
2 ~ ~ rn −~
iQ·(~ rm )
+ ρat f (Q) e dVm
n V
| {z }
~
=ISAXS (Q)

where ρn (~rnm )dVm is the number of atoms in dVm located at ~rn − ~rm relative to the atom in r~n .

The term labelled ISAXS (Q) ~ is important only when Q ~ → 0, i.e. for small angles (since
~
Q ∝ sin(θ)), as for any finite Q it oscillates rapidly averaging to zero. This small angles X-ray
scattering (SAXS) regime turns out to give important informations on the size and morphology
~ → 0 corresponds in real space to long distances).
of large scale structures (since the limit Q
On the other hand, as already said, ρ(~r) tends to its average value, ρat , after only few
inter-atomic spacings. Therefore the second term in ISRO (Q) ~ is important only when Q ~ is big,
or the distance is small. Thus it is sensitive to the short range order (SRO)and it contains the
structural information relevant to inter-atomic distances.

8.2.1 Short Range Order


We now turn our focus onto the short range term. Since in experiments we take several
snapshots, we first take an average over every possible choice of the origin: hρn (~rnm )i → ρ(~r).
This allows us to write
Z
~ = N f (Q)
2 ~ + N f (Q)
2 ~ ~
ISRO (Q) (ρ(~r) − ρat ) eiQ·~r dV
V

49
next we suppose our system to be isotropic, which is a reasonable assumption for a liquid or
glass, allowing us to replace ρ(~r) → ρ(r). We then evaluate the integral:
Z Z ∞ Z π Z 2π
~ r
iQ·~
(ρ(r) − ρat ) e dV = (ρ(r) − ρat )eiQr cos θ r2 sin θdφdθdr =
V 0 0 0
Z ∞ Z π Z −iQr
2 iQr cos θ −1 x
= 2π (ρ(r) − ρat )r e sin θdθ dr → I = e dx =
0 0 iQr iQr
| {z }
=I
 
1 x
iQr sin(Qr)
= e =2 →
iQr −iQr Qr
 Z ∞  (169)
2 ~ 2 sin(Qr)
→ ISRO (~r) = f (Q)N 1 + (ρ(r) − ρat )4πr dr ⇒
0 Qr
~
⇒ S(Q) ~ ≡ ISRO (Q) ≡ Liquid Structure Factor
~
N f 2 (Q)


4π ∞
Z
=1+ (ρ(r) − ρat )r sin(Qr)dr
Q 0
It is useful to consider the limiting forms of the liquid structure factor. In the limit Q → ∞ it is
easily seen that S(Q)~ = 1 and thus for short real space distances the scattering is independent
of any effect due to the correlations between particles. In the limit of long wavelengths, Q → 0
the integrand becomes: Z ∞
2
(ρ(r) − ρat ) |4πr
{z dr}
0
=dV
sin(Qr)
as Q
→ r. This means that the structure factor depends directly on the density fluctua-
Q→0
tions in the sample.
These fluctuations are particularly strong when the material has a large compressibility,
since if it is easy to change its density for some external force then thermal density fluctuations
can do the same. At the critical point of a fluid, the compressibility diverges and the fluctuations
extend to macroscopic distances. Indeed in this regime S(0) can become so large that the sample
is opaque to visible light and to X-rays as well. This phenomenon is called critical opalescence.
1 ∂ρ

The isothermal compressibility is defined as kT = ρ ∂P T . The equation of state for an
ideal gas is P = ρat kB T and the compressibility is kT = ρat k1B T , which varies smoothly with
density. For an interacting gas, as described by Van der Waals, the compressibility diverges at
a critical point in P − T space. It could be shown explicitly that S(0) = ρat kT kB T , which is
quantitatively in agreement with the expectation that for an ideal gas S(Q) = 1 for all Q.
The result obtained in (169) can be seen as the sine Fourier transform of the deviation of
the density from its average. This can be made clearer writing:
Z ∞
Q [S(Q) − 1] = H(r) sin(Qr)dr
0

with H(r) = 4πr(ρ(r) − ρat ) = 4πrρat (g(r) − 1). Then from the definition of inverse Fourier
transform,we obtain:
2 ∞
Z
H(r) = Q[S(Q) − 1] sin(Qr)dQ
π 0
which can be rearranged in order to yield:
Z ∞
1
g(r) = 1 + 2 Q[S(Q) − 1] sin(Qr)dQ (170)
2π rρat 0

50
Thus, the radial distribution function for a liquid or a glass can be directly obtained from the
measured structure factor. It should be noted that (169) was derived with the implicit assump-
tion of elastic scattering. While this is a reasonable assumption for general rigid structures as
glasses, it fails completely when considering liquids. However not everything is lost, since the
mean energy of an X-ray (' 10keV ) is much higher than the energy of the excitations of the
liquid molecules (' 10meV for diffusion, acoustic modes etc.) that the loss of energy of the
X-ray is negligible (but this does not mean it cannot be detected, as the evolution of X-ray
spectrometer allows them to resolve 1 part in 108 , and indeed phonon dispersion relations are
routinely measured with such instruments).
These results are easily generalised to the case of a multi-component system considering a
radial distribution function gij (r) that describe the correlations between atoms of type i and j.
This leads to the so-called partial liquid structure factor Sij (Q).

8.2.2 Supercooled Liquids


X-ray diffraction can be used to study the structure of supercooled liquids. These are liquids
that pertain their state even under the melting point temperature. This was explained by
Frank in 1952. Taking for example materials that when solid have the face centred cubic or
the hexagonal closed packed structure we have that each atom of these solids has 12 nearest
neighbours. It is reasonable to think that the liquid near the melting point would form clusters
with similar coordination. Frank’s deduction was that these clusters would have a third kind of
structure with 12 nearest neighbours, namely the icosahedron. This would imply that in order
to achieve the solidification of the liquid an energy barrier must be overcome. Even though
the icosahedron is a preferred structure for an isolated cluster, it cannot be pertained by every
dominion inside the liquid since it cannot completely fill a 3D space without overlapping. These
structures can be studied through X-ray diffraction.

8.2.3 Small Angle X-ray Scattering


We now turn our focus on the SAXS term present in equation for the scattered intensity. This
term can be recast in the form:
XZ ~ rn −~
X ~ Z ~
ISAXS (Q) = f 2
ρat eiQ·(~ rm )
dVm = f 2
eiQ·~rn ρat e−iQ·~rm dVm
n V n V

proceeding in a similar fashion to how we derived the short range order and small angle x-ray
scattering terms, we want to replace the sum on the right hand side with an integral, obtaining:
Z Z
~ rn ~
ISAXS (Q) = f 2
ρat e iQ·~
dVn ρat e−iQ·~rm dVm
V V

which for sufficient averaging, as usually happens in small angle scattering, becomes:
Z 2
i ~
Q·~
r

ISAXS (Q) = ρsl e dV (171)
V

where we have defined ρsl ≡ f ρat , which when multiplied by r0 gives the scattering length
density. Although this equation has the same form as the one derived for the atomic structure
factor, this one describes the scattering from objects bigger than the typical inter-atomic dis-
tance. This confines the scattering to small angles and allows the analysis of the scattering to
be carried out with a number of simplifying assumptions.

51
8.3 Scattering from a Isolated Particles
The simplest case is that of a dilute solution, where the inter-particle correlations can be
neglected and it is assumed the the particles are identical. Assuming ρsl,p to be the scattering
length density of each particle and ρsl,0 to be the scattering length density of the solvent, the
scattered intensity from a particle is given by:
Z 2 Z 2

~ ~
I1SAXS (Q) = (ρsl,p − ρsl,0 )2 eiQ·~r dVp = ∆ρ2 eiQ·~r dVp

Vp Vp

where Vp is the volume of the particle. Introducing the single particle form factor
Z
~ 1 ~
F(Q) = eiQ·~r dVp
Vp Vp

we obtain
I1SAXS (Q) ~ 2
~ = ∆ρ2 Vp2 |F(Q)| (172)
The form factor depends on the morphology of the particle and it can be evaluated analytically
only in few simple cases, such as a sphere of radius R:
Z R Z 2π Z π Z R
~ = 1 iQr cos θ 2 4π sin(Qr) 2
F(Q) e r sin θdθdφdr = r dr =
Vp 0 0 0 Vp 0 Qr
sin(QR) − QR cos(QR) J1 (QR)
=3 3 3
=3
QR QR
where J1 (x) is the Bessel function of the first kind. In figure can be seen how the form factor
changes with the particle’s dimension.
For Q = 0 ⇒ |F(Q)| ~ 2 = 1 and thus the intensity is I SAXS (Q) ~ = ∆ρ2 V 2 , which is the
1 p
expected result as in the forward direction all the electrons radiate in phase and thus the
intensity is proportional to the excess number of electrons squared (in the case of a single
particle embedded in a material).
It is highly useful to study the limiting cases for the form factor.

8.3.1 Long Wavelength Limit: Guinier Analysis


In the long wavelength limit, QR → 0 expanding the trigonometric functions yields:
(QR)3 (QR)5 (QR)2 (QR)4
  
3
F(Q) ∼ QR − + + . . . − QR 1 − + + ...
(QR)3 6 120 2 24
    
3 1 1 3 1 1 5
= − + (QR) + − (QR)
(QR)3 6 2 120 24
(QR)2
=1−
10
Therefore the intensity scattered in the long wavelength limit becomes
2
(QR)2 (QR)2
  
SAXS 2 2 2 2
I1 (Q) ∼ ∆ρ Vp 1 − ∼ ∆ρ Vp 1 − →
10 5 (173)
(QR)2
x
→ e ∼1+x → ∆ρ 2
Vp2 e− 5


The exponential formulation is useful as when log I1SAXS (Q) is plotted as function of Q2 the
graph is a straight line with slope equal to − R5 which enables to derive the dimension of the

52
sphere. This analysis is valid also for particles of general morphology, but instead of the sphere
radius there is the radius of gyration Rg . It is defined as the root mean square distance from
the particle centre of gravity: R
ρ (~r)r2 dr
Vp sl,p
2
Rg = R
ρ (~r)dr
Vp sl,p

It is straightforward to see that for a uniform sphere it is Rg2 = 35 R2 and thus the scattered
intensity is
(QRg )2
I1SAXS (Q) ∼ ∆ρ2 Vp2 e− 3

which allows to directly obtain Rg from the Guinier analysis. Although this result has been
obtained for the particular case of a sphere it could be shown that it as general validity.

8.3.2 Short Wavelength Limit: Porod Analysis


In the limit of short wavelength, QR  1, but with wavelength greater than inter-particle
spacings, the form factor becomes
   
sin(QR) cos(QR) cos(QR)
F(Q) = 3 − ∼3 −
(QR)3 (QR)2 (QR)2

In this regime the squared cosine oscillates quickly with mean value 12 . Then the intensity can
be written as:
hcos(QR)i2 1 1
I1SAXS (Q) = 9∆ρ2 Vp2 4
= 9∆ρ2 Vp2 (174)
(QR) 2 (QR)4
2
Using the relation between the volume and the surface of a sphere Vp2 = 43 R3 = 4π 9
R4 Sp we
find
2π∆ρ2
I1SAXS (Q) = Sp
Q4
Thus, in the Porod regime, the SAXS intensity is directly proportional to the surface of the
sphere and inverse proportional to the fourth power of Q. On a log-log plot (inserisci figura), it
can be seen that the mean of the intensity falls of with a gradient of -4, as represented by the
dashed line.
Until now we have considered a three dimensional object, a sphere, so one could be interested
in how the form factor varies with the dimensionality of the scattering particle. This dependence
is suggested by the defining equation for the form factor, since it involves an integral over the
particle’s volume. Depending on the dimensionality, the volume element inside the integral
varies: ∝ r2 for the sphere (3D), ∝ r for a disc (2D) and constant for a thin rod (1D). The
form factors for these cases are showed in figure. The comparison is facilitated by plotting the
form factors in function of the product of Q by the radius of gyration Rg . It can be noted
that the differences are more pronounced in the Porod regime, where we can see that the form
factors change proportionality: |F(Q)|2 ∝ Q−4 , Q−2 and Q−1 in d = 3 , 2 and 1 dimensions,
respectively.
It is important to note that the apparent dimensionality inferred from the experiments can
change changing the range of wave vectors considered. This because different wave vectors
probe different real space length scales and some objects may change appearance at different
length scales.

We now briefly describe the scattering from a dense solution of particles. In this case the
inter-particles correlations cannot be neglected, and in fashion similar to what we did in the

53
case of liquids they are taken into account by a structure factor S(Q). Therefore the intensity
takes the form
~ 2 S(Q)
I1SAXS (Q) = ∆ρ2 Vp2 |F(Q)|

54
9 Electrons in a Periodic Potential
Since the solids are characterised by a periodic structure where the ion are located, we are led
~ = U (~r) for all Bravais
to consider the problem of an electron in a periodic potential U (~r + R)
lattice vectors R.~ In principle this is a many-electron problem. since the full Hamiltonian
would depend both on the potential between the periodic array of ions and an electron and the
pair potential between electrons. In order to have a solvable problem we make the independent
electron approximation, i.e. we represent all these interactions by an effective one-electron
potential U (~r). The particular form of this potential is not what we are concerned right now.
To derive important features of the problem at hand it suffices to know that it satisfies the
periodic condition. Therefore we are interested in examining the general properties of the
Schrödinger equation for a single electron:

~2 2
 
Hψ = − ∇ + U (~r) ψ = Eψ (175)
2m

It should be noted that the free electron Schrödinger equation is a special case of eq.(175),
zero potential being a particular case of periodic potential. Independent electrons that obeys
Schrödinger equation with a periodic potential are called Bloch electrons.
The stationary states in the presence of a periodic potential posses a general property that
is consequence of the Bloch theorem.

9.1 Bloch Theorem


~2 ~ = U (~r)
The eigenstates ψ of the one-electron Hamiltonian − 2m ∇2 + U (~r), where U (~r + R)
with R~ a vector of the Bravais lattice, can be chosen to have the form of a plane wave times a
function with the same periodicity as the potential:
~
ψn~k (~r) = eik·~r un~k (~r), ~ = u ~ (~r)
un~k (~r + R) (176)
nk

This result can be recast in the equivalent form

~ = ei~k·~r ψ(~r)
ψ(~r + R) (177)

This theorem can be proven in the following way:


For each Bravais lattice vector R ~ we can define a translation operator T ~ which, when acting
R
on a function f (~r) shifts its argument by R. ~ These operators posses a number of properties:

~
TR~ f (~r) = f (~r + R)

TR~ H = HTR~

TR~ TR~ 0 = TR~ 0 TR~ = TR+

~ R~0

the second is consequence of the periodicity of the Hamiltonian and the third is directly obtained
applying the operators.
These properties result in the fact that the eigenstates of H can be chosen to be simultane-
ously eigenstates of every TR~ : (
Hψ = Eψ
~
TR~ ψ = c(R)ψ
As consequence of the last property of the translation operators, its eigenvalues must obey:
~ R
c(R)c( ~ 0 ) = c(R
~ +R
~ 0)

55
Now, let ~ai be three primitive vectors for the underlying Bravais lattice. We can always write
c(~ai ) = ei2πxi
with a suitable choice for the xi . In general the c’s can be complex. We now consider the case
~ = n1~a1 + n2~a2 + n3~a3 . Then we obtain that
R
~ = c(~a1 )n1 c(~a2 )n2 c(~a3 )n3
c(R)
which is equivalent to
~ ~
~ = eik·R
c(R)
where ~k = x1~b1 + x2~b2 + x3~b3 and the ~bi are the primitive vectors of the reciprocal lattice that
satisfy the property ~ai · ~bj = 2πδij. Therefore we have shown that the eigenstates ψ of H can
be chosen so that they satisfy:
TR~ ψ = ψ(~r + R) ~ = ei~k·R~ ψ
~ = c(R)ψ (178)
that is precisely the Bloch’s theorem.
By imposing periodic boundary conditions it can be showed that the ~k are real. We must
use a generalization of the Born-Von Karman boundary conditions used in the Sommerfeld
theory of free electrons, since unless the Bravais lattice is cubic and L is an integral multiple of
the lattice constant a, it is not convenient to continue to work with a cubical volume of side L.
Instead is more convenient to work with a volume commensurate to the volume of the primitive
cell, thus the periodic boundary conditions are generalised in the following manner:
ψ(~r + Ni~ai ) = ψ(~r) i = 1, 2, 3
1
where the ~ai are the primitive vectors, and Ni are integers of the order of N 3 , with N = N1 N2 N3
is the total number of primitive cells in the crystal. When we adopt this boundary conditions
we are always supposing that the bulk properties of the solid will not depend on the choice of
boundary condition, which can therefore be chosen by analytical convenience.
Applying Bloch’s theorem to the boundary conditions we find:
~
ψn~k (~r + Ni~ai ) = eiNi k·~ai ψn~k , i = 1, 2, 3 ⇒
~
⇒ eiNi k·~ai = 1 i = 1, 2, 3
when ~k is a vector of the reciprocal lattice the last equation requires that
mi
e2πiNi xi = 1 ⇒ xi = , mi integral
Ni
Thus, the general form for the allowed Bloch wave vectors is
3
~k =
X mi ~bi
i=1
Ni

From this it follows that the volume ∆~k of k-space per allowed wave vector ~k is
!
~b1 ~b2 ~b3 1 ~ ~
∆~k = · × = b1 · (b2 × ~b3 )
N1 N2 N3 N ↓
volume of a reciprocal
lattice primitive cell

This result states that the number of allowed wave vectors in a primitive cell of the reciprocal
lattice is equal to the number of primitive cells in the crystal.
3
The volume of a reciprocal lattice primitive cell is (2π)
v
V
where v = N is the volume of the
direct lattice primitive cell. Then the volume per wave vector can be written in the form
(2π)3
∆~k =
V
56
9.2 Remarks on the Bloch’s Theorem
I. Bloch’s theorem introduces a wave vector ~k that plays the same fundamental role for
the electronic motion in a periodic potential as the free electron wave vector plays in the
Sommerfeld theory. However, while in the free electron case, ~k is directly tied to the
momentum of the electron, ~k = ~p~ , in the Bloch case it is not directly proportional to the
momentum. This is due to the fact that with a periodic potential the Hamiltonian does not
posses full translational invariance, thus its eigenstates cannot be simultaneous eigenstates
of the momentum operator. This is confirmed applying the momentum operator to a
Bloch wave vector  
~
−i~∇ψn~k = −i~∇ eik·~r un~k (~r)
~
= ~~kψn~k − i~eik·~r ∇un~k (~r)
In the Bloch case ~k is known as the crystal momentum of the electron to emphasize its
similarity with the momentum, but it must kept in mind that ~~k is not a momentum.
II. ~k can always be confined to the first Brillouin zone. This because any ~k 0 not in the first
Brillouin zone can be written as
~k 0 = ~k + G
~
~ R~
where ~k lies in the first zone and G~ is a reciprocal lattice vector. Since eiG· = 1 for any
reciprocal lattice vector, if the Bloch theorem holds for ~k 0 it holds also for ~k.
III. The index n appears on the Bloch eigenstates because for a given ~k there are many
solutionsto the Schrödinger equation. If we look for all the solutions to Schrödinger
equation that have the Bloch form with ~k fixed we find
~
ψ(~r) = eik·~r u(~r)
~2 2
 
~ ~
Hψ = − ∇ + U (~r) eik·~r u(~r) = Eeik·~r u(~r) →
2m

~ ~
→ −∇2 (eik·~r u(~r)) = −∇ · ∇(eik·~r u(~r))
~ ~
= −∇(i~keik·~r u(~r) + eik·~r ∇u(~r))
 
2 i~k·~
r ~ i~k·~
r ~ i~k·~
r i~k·~
r 2
= − −k e u(~r) + ik · e ∇u(~r) + ik · e ∇u(~r) + e ∇ u(~r)
 
~
= eik·~r k 2 u(~r) − 2i~k · ∇u(~r) − ∇2 u(~r) ⇒
 2
⇒ −i∇ + ~k = −∇2 + k 2 + 2i~k · ∇ →
~2 
 2 
i~k·~ ~
→ Hψ~k =  r
− −i∇ + ~k + U (~r) u~k = E eik·~
r
 
e 
u~k
2m
(179)
~
Since the wave function is periodic, u(~r + R) = u(~r), the problem we have to solve is
equivalent to an Hermitian eigenvalue problem restricted to a single primitive cell of the
crystal, which suggests, in analogy with the free electron enclosed in a box, that the
solutions are discrete and labelled with the band index n (i.e. we find an infinite family
of solutions with discretely spaced eigenvalues labelled with n). It should be noted that in
the eigenvalue problem at hand, the wave vector ~k is just a parameter of the Hamiltonian.
Therefore we expect each of the energy levels, for given wave vector, to vary continuously
as ~k varies. In this way we arrive at a description of the energy levels of an electron in a
periodic potential in terms of a family of continuous functions En (~k).

57
IV. Although the full set of levels can be described with ~k restricted to a single primitive cell,
often it is useful to let the wave vector range through all of k-space, even though this
gives a redundant description. Since the set of eigenfunctions and energy levels for two
wave vectors differing only by a reciprocal lattice vector must be the same we can assign
the index n in such a way that for a given n the eigenstates and eigenvalues are periodic
functions of ~k
ψn,~k+G~ = ψn,~k
(180)
En,~k+G~ = En,~k
This leads to a description of the energy levels of an electron in a periodic potential
in terms of a family of continuous functions En,~k . The information that these functions
contain is referred as the band structure of the solid. For a given n, the electronic levels
described by En,~k is called energy band. It should be noted that as these functions are
periodic and continuous they must have an upper and lower bound, so that every level
specified by a specific function lies within these limits.

9.3 The Fermi Surface


The Fermi surface in the case of Bloch electrons differs quite a bit from that of free electrons:
one-electron levels are labelled with two quantum number, n and ~k, En (~k) does not have the
simple explicit free electron form and ~k must be confined into a single primitive cell in order
to count each level just once. Other than these differences, the Fermi surface is constructed in
a similar fashion. When the lowest energy levels are filled by a specified number of electrons,
two distinct configurations can arise:

I. A certain number of bands is completely filled and the others are left empty. The energy
difference between the highest occupied level and the lowest unoccupied is called band
gap. If this band gap is much greater than kB T , with T near room temperature, then
the solid is an insulator, if the band gap is of the order of kB T the solid is known as an
intrinsic semiconductor.
Since each energy level can accommodate two electrons, a band gap can arise (though it
need not) only if the number of electrons per primitive cell is even.

II. A number of bands is partially filled. When this happens, the energy of the highest
occupied level, the Fermi energy EF lies within the energy range of at least one band. For
each partially occupied band there will be a surface separating the occupied levels from
the unoccupied ones. The set of all such surfaces is known as the Fermi surface and is a
generalization to Bloch electrons of the Fermi sphere. The single surfaces lying within a
band are called branches of the Fermi surface. If a Fermi surface exists, than the solid
has metallic properties. Analytically, a branch of the Fermi surface is characterized by

En (~k) = EF

i.e. the Fermi surface is a constant energy surface in k-space.

58
10 Classification of Solids
We previously described solids with categories characterized by a particular crystal structure.
This description does not allow to infer important structural aspects that affect the physical
properties of the solid. Now we want to give a different, less rigorous description of the solids,
based on the configuration of the valence electrons.
The most important distinction that can be made by the distribution of the valence electrons
is that between metals and insulators. As we already saw, these two categories are discriminated
by the presence or absence of partially filled energy bands. This is a rigorous criterion for ideal
crystals at zero kelvin, provided that the independent electron hypothesis holds. Thus the
basis for these two categories is the electronic distribution in wave vector space, rather than in
real space. Indeed in real space only qualitative considerations can be made on the electronic
distribution, observing that in metals the electrons are not as concentrated around the ion cores
as in the insulators, as can be seen in figure.
Deducting the properties of solid neon and solid sodium would require the following chain
of thought: appreciable wave function overlapping suggests the presence of broad bands, which
leads to the possibility of band overlap and therefore metallic properties. Willing or not one is
lead back to k-space, where the only really satisfactory criterion can be given.

10.1 Classification of Insulators


Although the distinction between metals and insulators is based on the electronic distribution
in k-space, it is of great utility to further discriminate different categories within the insulators.
These distinctions are drawn on basis of the electronic spatial distribution. The categories that
can be defined are:

Covalent Crystals These crystals have an electronic spatial distribution non dissimilar to
that of metals, but with no partial filled bands in k-space. However, they are not likely to
have an almost uniform density of electrons throughout the crystal, nut it is more likely
that the interstitial electronic density will be localized on certain well defined directions,
leading to what in chemist is known as bond. This interstitial electronic density is what
distinguish covalent crystals from the other categories of insulators.

Molecular Crystals These are the crystals showed by solid noble gases, except for He. Since
they posses completely filled shells that are only slightly perturbed in the solid, there is
very little electronic density in the interstitial space, since every electron is well localized
around its parent ion. Band structure theory is hard to apply on these solids since every
electron can be considered to be a core electron.

Ionic Crystals Ionic crystals are compounds made up of a metallic and non metallic element.
As molecular crystals, also in this case the electronic density is concentrated around the
ion cores. Although they present this similarity with molecular crystals, a huge difference
comes from the fact that in this case not every electron is in the neighbourhood of their
parent ion, but they can stray so far to be closely bound to the constituent of the other
type. Indeed ionic crystals can be regarded as molecular crystals whose constituents are
the ions, whose charge distribution is only slightly perturbed in the solid. However, since
they are charged ions, the electrostatic forces have an high influence in the determination
of the properties of the ionic crystals.

Now we will describe in a somewhat more detailed manner each basic type of solid, empha-
sizing the different kind of models used in their description and also looking at the continuity
that prevails between the categories.

59
10.2 Ionic Crystals
The most naive model considers the ions as impenetrable spheres that are held together by the
electrostatic force between opposed charges. The impenetrability is due to the Pauli exclusion
principle and the electronic configurations of the ions. Indeed, the ions are in a stable closed
shell configuration, which implies that when two ions are brought so close together that their
electronic clouds overlap, the exclusion principle requires that the exceeding electronic density
introduced in the ion neighbourhood by the other ion be accommodated in unoccupied levels.
However, since the ions are in the stable closed shell configurations, the first available levels
are much higher in energy than the last occupied. As a result it costs much energy to overlap
the charge distributions, i.e. a strong repulsive force ensues between the ions, whenever they
get too close to each other.
For our scopes it suffices to consider the ions as impenetrable spheres, i.e. the the potential
that represents the repulsion is infinite within a certain distance and zero beyond that distance
(in more detailed calculations a more sophisticated potential is required).

10.2.1 Alkali Halides - I-VII Ionic Crystals


The ideal ionic crystal is almost perfectly realized by the alkali halides. These crystals’ structure
at normal pressure is cubic. The cation is one of the alkali metals (Li+ , Na+ , K+ , Rb+ or Cs+ )
and the anion is one of the halogens (F− , Cl− , Br− or I− ). Under normal conditions they all
crystallise in the sodium chloride structure except for CsCl, CsBr and CsI which posses the
cesium chloride structure.
The basic entities of these structures are ions instead of atoms because it is more favourable
in energetic terms. Taking as example RbBr, we have that to ionise the rubidium it costs
4.2 eV , while the bromine can attract an electron which has a binding energy of 3.5 eV . It
might appear that bringing close the atoms would cost 0.7 eV less than ionising the atoms
and then bringing them close together. This is true when the ions are very far apart, since
when they are brought together the energy of the pair is lowered by the attractive electrostatic
force. At the inter-ionic distance observed for crystalline RbBr (r = 3.4 Å), a pair of ions has
2
an additional Coulomb energy of − er = −4.2 eV which more than compensate for the 0.7 eV
favouring the atoms over the ions. The idea of an alkali halide as a set of spherical ions packed
together is confirmed by the electronic charge distribution observed through X-ray diffraction.

The values of the conventional cubic cell side s given by x-ray diffraction measurement are
in good agreement with the naive model of impenetrable spheres of definite radius r, known as
the ionic radius. Let d be the distance between

the centres of neighbouring ions, so that d = a2
for the sodium chloride structure and d = a 23 the the cesium chloride structure. In most cases
the nearest neighbour distance dXY can be given within an accuracy of 2% by dXY = rX + rY .
The only exceptions being LiCl, LiBr and LiI where the radius is shorter by 6,7 and 8 percent
respectively, and NaBr and NaI where the radius is short by 3 and 4 percent.
Leaving aside the exceptions, we obtain that the rigid sphere approximation is in good
agreement with the observed data. However the choice of ionic radii is not unique, since we
can increase one and decrease the other by the same amount keeping fixed d. Our ansatz
on the nearest neighbour distance to be given by the sum of the ionic radii assumes that the
ions touch themselves. This is the case when the radius r> of the larger ion is not too much
greater than that of the smaller ion, r< . If the difference is too large, the smaller ion may not
touch the greater at all. Thus the relation that gives the nearest neighbour
√ > distance must be
corrected, obtaining, in the case of sodium chloride structure, d = 2r . The critical radius
ratio occurs when one of the larger ions touches both the smaller nearest neighbouring ion and
the successive nearest neighbour large ion. Therefore for the sodium chloride structure, the

60
critical ratio satisfies
r> 1 √
<
=√ = 2 + 1 = 2.41
r 2−1
From the data, we see that this ratio is exceeded√only by the lithium halides, hence the nearest
neighbour distance should be compared with 2r> instead of r> + r< . Indeed it fits the
observed data within the 2% accuracy that the other formula give for the other crystals. A
similar calculation for the cesium chloride structure yields
r> 1 √
= ( 3 + 1) = 1.37
r< 2
Neither of the crystals with this structure exceeds this ration, and the radial sums are n good
agreement with observed data.

10.2.2 II-VI Ionic Crystals


Double ionised elements can also form ionic crystal. With the exceptions of the beryllium
compounds and MgTe, these assume the sodium chloride structure. For calcium, strontium
and barium salts, and MgO the radial sums agree within a few percent with the measured
nearest neighbour
√ distance. For MgS and MgSe, the critical ratio is exceeded, and d = a/2
agrees with 2r> to within 3%. However, there is no satisfactory agreement with the radial
sum and the beryllium compounds and MgTe. BeS, BeSe and BeTe crystallise in the √ zincblende
>
structure and the other two in the wurtzite structure. The critical ratio is rr< = 2 + 6 = 4.45,
which is exceeded by all the beryllium compounds √ >with the zincblende structure. Then the
observed data for d should be compared with 6r /2, but the agreement is relatively poor
compared to what we find for the crystals with the sodium chloride structure. This is due to
the relatively high energy required to ionise the beryllium, thus it costs more energy to produce
widely separated ions as opposed to atoms. Moreover, since beryllium is quite small it cannot
take full advantage of the crystal structures with high coordination number to compensate
with the Coulomb energy. The anions, in this case, would repel due to their charge overlapping
before coming close enough to the beryllium ion. These observations suggest that with beryllium
compounds we are straying away from the pure ionic crystals. It turns out that tetrahedrally
coordinated structures tend to be covalently bounded. Then the tetrahedrally coordinated
II-VI compounds are more covalent than ionic in character.

10.2.3 III-V Crystals - Mixed Ionic and Covalent


Crystals made by these categories of elements are still less ionic in character. Almost every
of them assumes the zincblende structure characteristics of covalent crystals. Most behave as
semiconductors rather than insulators, thus their band gaps are quite small. This indicates that
their ionic nature is weak and electrons are not strongly localized. Usually they are described
as primarily covalent with some residual concentration of excess charge about the ion cores.

10.3 Covalent Crystals


The most striking difference between ionic and covalent crystals is the difference in the electron
density change along a nearest neighbour line: in ionic crystals it drops below 0.1 electrons
per cubic angstrom in NaCl, whereas it is always greater than 5 electron per cubic angstrom
in diamond. All the elements of the column IV crystallise in the tetrahedrally coordinated
diamond structure. In chemical terms it is said that each atom partakes in four covalent bonds
by sharing an electron with each of its nearest neighbours. Although the bond is electrostatic
in nature, it is a picture quite different from that of the ionic crystals.

61
The covalent crystals are not as good insulators as ionic crystals, due to the delocalization
of the charge in the covalent bond. The semiconductors are all covalent crystals, sometimes
with a little touch of ionic character.

10.4 Molecular Crystals


The best examples of molecular crystals come from the noble gases (except helium), which
crystallise in mono-atomic fcc Bravais lattice. The electronic configuration for each atom is of
the stable closed shell type and it is little perturbed in the solid. The solid is held together by
the Van der Waals or fluctuating dipole forces. The physical picture of this force can be drawn
in the following way:
we consider two atoms (1 and 2) a distance r one from the other. Although the charge
distribution of a single noble gas atom is spherically symmetric, at any moment there may be
a net dipole moment (with null time average). The instantaneous dipole moment of atom 1 is
p~1 and its electric field is ∝ p1 /r3 at a distance r from the atom. This will induce an electric
dipole on atom 2, which is proportional to the field
αp1
p2 = αE '
r3
where α is the polarizability of the atom. The energy of interaction between two dipole is
proportional to the product of their dipole moments divided by the cube of the distance between
them, the energy lowering will be of the order

p1 p2 αp21
' 6
r3 r
Since this quantity depends on the dipole moment squared, its time average does not need to
vanish. Since the force falls off rapidly with distance it is really weak and hence the low melting
and boiling points of the condensed noble gases.
A more accurate treatment would need to include the fluctuating dipolar interactions be-
tween groups of three or more atoms. These fall off more rapidly than the Van der Waals force,
but are also important when accurately describing the solid.

Elements of column V, VI and VII (except for metallic polonium and the semimetals anti-
mony and bismuth) participate to varying degrees of both molecular and covalent character.

10.5 Metals
Moving to the left from column IV we find the family of metals. The covalent bond expands
until there is an appreciable charge density throughout the interstitial regions and, in k-space,
there is a significant band overlapping.
The prime examples of metallic crystals are the alkali metals. These can be accurately
described by the Sommerfeld free electron model, in which the valence electrons are depicted
as completely separated from their ion cores to form a nearly uniform gas.
Aspects of molecular and covalent bonding can be found in the metals, especially in the
noble metals. This because in noble metals the filled d-shells are not very tightly bound and
therefore are subject to considerable distortion in the metal.
If we try to compare the ionic radii of the metals as inferred by the ionic crystals in which
they participate to the nearest neighbour distance in the metal, we don’t find any particular
correlation between these two quantities. This is consistent with the fact that quantities such
as the alkali metal compressibility are of the order of their electron gas values (the ions are

62
truly small entities embedded in a sea of electrons). On the other hand, the filled d-shells of the
noble metals play a far more important role in determining the properties of the solid: indeed,
this is reflected by the fact that the nearest neighbour half distances are not too much larger
than the ionic radii. In both ionic crystals and metals, size is determined by the d-shells.

63
11 Cohesive Energy
The cohesive energy of a solid id the energy required to disassemble it into its constituent part,
i.e. its binding energy. This of course depends on what is taken to be the constituents of the
solid. One can consider them to be the single atoms, but in the case of solid nitrogen it is
more convenient to define them to be the nitrogen molecules. Knowing the binding energy of
the single molecule and how many molecules there are in the solid it is easy to convert one
definition to the other. In the case of ionic crystals we shall speak of the ions as the constituents
and the two definitions are linked knowing the first ionisation energy and the electron affinity
of both ions.

In what follows we will discuss some facts about cohesive energy at zero temperature. These
energies will be calculated for an external imposed lattice constant, thus we will consider solids
under an external pressure. By calculating the rate of change of the cohesive energy with lattice
constant we will be able to find the equilibrium lattice constant as that requiring zero external
pressure for maintenance.
The ion cores will be treated as classical particles which can be perfectly localized with zero
kinetic energy. This is profoundly incorrect since it violates the uncertainty principle. Indeed if
an ion core can be localized in a region of linear dimension ∆x, the uncertainty on its momentum
is at least ~/∆x. Therefore it will have a kinetic energy of the order ~2 /(M (∆x)2 ), known as
the zero point kinetic energy, whose contribution must be taken into account. Moreover, since
the ion cores are not perfectly localized on the crystal lattice sites, deviations in their potential
energy must be allowed for. In successive sections we will treat more accurately the problem of
lattice vibrations. For the moment we are satisfied noting that the lighter the ionic mass, the
larger the zero point kinetic energy.
We will develop a crude theory for the molecular and ionic crystals which is quite satisfactory.
However, for covalent crystals and metals even a crude theory is hard to build. This because
the electronic distribution in the solid, for both cases, is very different from that of the isolated
atoms or ions.
For simplicity we will focus on cubic crystals with cubic cell side a. In doing so we are ignor-
ing every other geometrical effect that can affect the energy. We shall also consider only uniform
compression, which preserves the cubic symmetry of the crystal (not because the physics of gen-
eral deformations is different, but because the geometrical aspects can be complicated.

11.1 Molecular Crystals : the Noble Gases


We consider only the simplest molecular crystals, i.e. those formed by noble gases (except
helium, since quantum effects don’t allow it to crystallise). As already previously said, noble
gases atoms are only slightly distorted in the solid. Such small distortion can be described as
an oscillating dipole interaction represented by the potential ∝ r−6 . This weak interaction is
what holds the solid together.
When the atoms approach one another too closely, the repulsion of the ion core comes into
play and it is what determines the equilibrium size of crystal. At short distances this repulsion
is much stronger than the attraction, and it falls off quicker. therefore it is custom to represent
it as a power law. Generally it is chosen the 12 power, and the resulting potential is
A B
φ(r) = − 6
+ 12 ⇒
r r ( 1/6
σ= B
  
σ 12  σ 6 A
⇒ φ(r) = 4 − , A2
r r  = 4B

64
and it is known as the Lennard-Jones potential. The power on the repulsive part is chosen
to be 12 only for an analytical convenience and the requirement that it must be greater than
6. With this choice, by suitable choices of σ and , the gaseous properties of Ne, Ar, Kr and
Xe at low densities can be well reproduced. It should be emphasized that the precise form of
the Lennard-Jones potential should not be taken too seriously, since it is nothing more than a
simple way of taking into account some observed properties:

I. The potential is attractive and proportional to r−6 at large distances.

II. There is a strong repulsion at small separations.

III. The parameters  and σ measure the strength of the attraction and the radius of the
repulsive core as fitted from the experimental data in the gaseous state.

We note, moreover, that  is of the order of 0.01 eV , consistent with the weak binding in solid
noble gases.

We now want to fit some observed properties of solid noble gases in term of the parameters
 and σ. We treat the solid as a set of classical particles, localized with negligible kinetic energy
at the points of the observed fcc Bravais lattice.
To compute the total potential energy we first note that the energy of interaction of the
atom at the origin with every other is
X
Eint = ~
φ(R)
~ =~0
R6

Summing the interaction energy for each atom of the crystal we obtain a multiplicative factor
N , the total number of atoms. This result is twice the total potential energy since each pair of
atoms has been counted two times, thus the total potential energy is
NX ~
Eptot = φ(R)
2
~ =~0
R6

and the energy per single particle is


1X ~
u= φ(R)
2
~ =~0
R6

Both sums are over all non zero Bravais lattice vectors. Writing each Bravais lattice vector as a
dimensionless number, α(R)~ times the nearest neighbour separation r, the energy per particle
can be written as   σ 12  σ 6 
u = 2 A12 − A6
r r
X 1 (181)
An =
~ n
α(R)
~ =~0
R6

The constants An depend on the crystal structure and on the number n. Since α(R) ~ = 1 for any
lattice vector joining nearest neighbouring atoms, as n grows large, An approaches the number
of nearest neighbours. When n diminishes, the next nearest neighbours start to contribute and
An increases. For n = 12, An is given to an accuracy of a tenth of percent by the contributions
of the nearest, next-nearest and third-nearest neighbours of the origin. For the fcc we have
A6 = 14.45 and A12 = 12.13.

65
11.1.1 Equilibrium Density
The nearest-neighbour separation at equilibrium r0 , and hence the density, is found minimizing
the potential energy with respect to r:
 1
∂u th 2A12 6
=0 ⇒ r0 = σ = 1.09σ
∂r r0 A6
This theoretical value is in good agreement with that found experimentally. This agreement
deteriorates as the mass of the considered atom becomes lighter (the experimental value ex-
ceeding the theoretical one). This can be understood as an effect due to the neglecting of
the zero-point kinetic energy. This energy grows bigger tighter the volume where the atom is
confined, thus it acts as repulsive force, increasing the lattice constant.

11.1.2 Equilibrium Cohesive Energy


By substituting the equilibrium nearest neighbour separation into the energy per particle we
find the equilibrium cohesive energy
A26
 
th A6
u0 = 2 A12 2 − A6
4A12 2A12
2
A
= − 6 = −8.6
2A12
Comparing this value with that found experimentally, good agreement is again found, but |uth 0 |
increasingly exceeds |uexp
0 | as the atomic mass decreases. Again this makes sense as an effect of
the neglected zero-point kinetic energy, since we are neglecting a positive term that decreases
the binding.

11.1.3 Equilibrium Bulk Modulus


The bulk modulus is defined as  
∂P
B = −V
∂V T
dU
and can be calculated in terms of  and σ. As the pressure at T = 0 is given by P = − dV ,
where U is the total energy, we can write B in terms of the energy per particle u and volume
V
per particle v = N as  2   
∂ U ∂ ∂u
B= V =v
∂V 2 T ∂v ∂v
3
The volume per particle in the fcc lattice is v = ar , where the side a of the conventional cubic

cell is related to the nearest neighbour distance r by a = 2r. Therefore we get

r3 ∂ 2 ∂
v=√ , = 2
2 ∂v 3r ∂r
then the bulk modulus can be written as
√  
2 ∂ 1 ∂
B= r (u)
9 ∂r r2 ∂r
If we consider the equilibrium distance r0 , we have that ∂u
∂r
= 0 and the bulk modulus reduces
to: √ 2  5
th 2 ∂ u 4 A6 2 75
B0 = 2
= 3 A12 = 3 (182)
9 ∂r r=r0 σ A12 σ

66
Also in this case, when comparing the theoretical data with the experiments, one finds good
agreement with the heavier elements, while for the lighter ones there are big discrepancies, due
to the neglected zero point motion.

11.2 Ionic Crystals


The simplest model that we can construct uses the same hypothesis used in the case of molecular
crystals: the cohesive energy is entirely given by the potential energy of classical particles
localized at the equilibrium positions. Since in ionic crystals the particles are charged ions,
the largest contribution in the interaction energy comes from the the inter-ionic Coulomb
interaction, that varies as the inverse first power of the inter ionic distance.
To determine equilibrium lattice parameters we must take into account the short range
core-core repulsion due to the Pauli principle. Therefore the energy per ion pair is

u(r) = ucore (r) + ucoul (r)

where r is the nearest neighbour distance.


The calculation of the Coulomb term is not as straightforward as it was the attractive force
for the molecular crystals, because the Coulomb potential has a long range. If we consider the
~ for the anions, with a second fcc shifted
sodium chloride structure, which is an fcc at sites R
~ a
by a translation d through 2 along a cube side. Measuring again the inter-ionic distances in
terms of the nearest neighbour distance r = a2 :
(
~ = α(R)r
|R| ~
~ = α(R
~ + d|
|R ~
~ + d)r

we can write the total potential energy for a single ion as


 !
2
e 1 X 1 1
−  + − 
r α(d) ~ ~ ~
α(R + d) α(R)~
~ =~0
R6

If there are N ions in the crystal, the total potential energy is obtained multiplying the previous
expression by N2 . The energy per ion pair is obtained dividing the total energy by the number
of pairs, N2 :  !
2
e 1 X 1 1
ucoul (R) = −  + −  (183)
r α(d) ~ ~ ~
α(R + d) α(R) ~
~ =~0
R6

However, 1r falls off slowly with distance, and (183) is not a well defined sum, since its value
depends on the order in which the summation is performed. This is not just a mathematical
detail, but it has a physical meaning: when constructing the infinite crystal we can always
arrange the ions in order to have arbitrary surface charge distributions and/or dipolar layers at
all stages. This allows to make the energy per ion pair to approach any desired value. This can
be solved constructing the crystal in such a way that there are no appreciable contributions to
the energy from the charges on the surface. There are several methods that can guarantee this.
The result of all such calculations is that the electrostatic interaction per ion pair has the
form:
e2
ucoul (r) = −α (184)
r
where α, known as the Madelung constant, depends only on the crystal structure.

67
One-dimensional case We consider a straight line of alternated charged ions, each sepa-
rated by a distance r. The pair energy for this configuration is:

e2
 
coul 1 1 1
u (r) = −2 1 − + − + ... →
r 2 3 4
x2 x3 x4
→ ln(x + 1) = x − + − + ... →
2 3 4
e2 e2
→ = −2 ln(2) ' −1.39
r r
Confronting the values of the Madelung constant for different structures we note that it is
an increasing function of the coordination number.

Equation (184) is to be expected to overestimate the strength of the binding, since it neglects
the positive contribution given by the short-range core-core repulsion. Nonetheless we can
expect this contribution to be small, as the repulsion is a fast varying function of the inter ionic
separation. Representing it as an inverse power law we get

e2 C
u(r) = −α + m (185)
r r
Thus, the equilibrium separation r0 can be found minimizing the pair potential energy, obtain-
ing:
mC
r0m−1 = 2 (186)

In the noble gas case we could use this to determine the equilibrium distance, but not in this
case since we lack of an independent measurement of C. Instead we can use this equation to
determine C from the experimental data of the equilibrium distance:

αe2 r0 m − 1
C=
m
which, once substituted in the pair energy gives:

αe2 m − 1
uth
0 = u(r0 ) = − (187)
r0 m
For the noble gases we chose m = 12 for analytical convenience and because it reproduces quite
well the experimental data, but for alkali halides we don’t have any particular reason to make
such a choice. We may just determine the exponent by fitting the experimental data. It is not
advisable to use the pair energy to determine m as it is a slowly varying function of m, and
small errors in the energy would result in big uncertainties on m. A better procedure is to
find a way to independently determine the exponent and then see if the theoretical value of the
energy agrees with the experimental data. This is provided by the experimentally measured
bulk moduli. Given B0 and r0 the equilibrium bulk modulus and nearest neighbour separation,
respectively, m has the value:
18B0 r3
m = 1 + coul 0 (188)
|u (r0 )|
The value of m vary from 6 to 10. The correction to the purely electrostatic contributions
greatly improve the agreement of the theoretical value with the experimental one (except for
lithium halides and sodium iodide).
This model could be further improved by some better approximations: the core-core re-
pulsion could be represented by an exponential, the fluctuating dipole force between ion cores

68
could be taken into account and the zero point vibrations of the lattice should be allowed for.

Much of the success found by the models built so far is due to the fact that in these solids
the valence electron configuration is not greatly distorted from what it is in the isolated atoms
and ions. This is true no more in covalent crystals and metals, which are characterized by
an electronic distribution that does not resemble that of the free atoms or ions. Therefore to
calculate the cohesive energy it is not sufficient to compute the classical potential energy of a
set of weakly deformed atoms arranged in the appropriate crystal structure. Therefore a theory
of cohesive energy of covalent crystals and metals needs to contain a computation of the band
structure. For this reason there is no model of comparable simplicity to that of the model we
saw. Here we shall limit ourselves to a qualitative treatment about covalent crystals and crude
estimates for metals, based on the free electron picture.

11.3 Covalent Crystals


To accurately describe the cohesion of a covalent insulator it should be needed a theory similar
to that of chemical bonding in molecule, a subject beyond this course. Here we limit ourselves
to note that the way electrostatic forces conspire to keep the solid together is much more subtle
than the electrostatic attraction between point ions, or even the fluctuating dipole interaction.
For example let us consider diamond. Suppose we have a bunch of carbon atoms and placed
them onto the sites of a diamond lattice, but with a separation big enough that the energy of
the collection is just the sum of the energies of the isolated atoms (i.e. the cohesive energy is
zero). Cohesion will result if the energy is lowered once the separation of the atoms is lowered
to its observed value. In doing so we have that the atomic wave functions centred on different
sites start to overlap. In the case of noble gases or ions, whose outermost atomic shells are
completely filled, this overlap results in a strong repulsion that raises the energy from that of
the isolated atoms (as we already saw this is due to Pauli exclusion principle). If the outermost
electronic shells are only partially filled, then the electrons in the outer shells can rearrange
themselves when the wave functions of neighbouring atoms start to overlap, since levels of
comparable energy in the same shell are available.
It turns out that under these circumstances the total electronic energy is lowered, with
the electrons forming levels that are not localized on a single ion core. The qualitative reason
for this is that the less localized the electron is the lower its momentum from the uncertainty
principle and therefore the lower its kinetic energy. To this must be added a change in the
potential energy in the less localized levels, with the net result that the energy is lowered.

11.4 Free Electron Metals


In this case we can compare the solid not to a set of atoms but rather with a free electron gas.
To use this to get a theory on cohesion we must add to the electron gas kinetic energy the total
electrostatic potential energy.
We treat the ions in the alkali metal as point charges localized at the sites of a bcc lattice
and the electrons as a uniform compensating background of negative charge. By techniques
similar to that used in computing the electrostatic energy for ionic crystals, we get the energy
per atom of this configuration, which results:
24.35
ucoul = − eV /atom
rs /a0

where rs is the radius of the Wigner-Seitz sphere (i.e. the volume per electron: 4πrs3 /3) and a0
is the Bohr radius. This attractive electrostatic energy must be balanced against the electronic

69
kinetic energy per atom. Since in alkali metals there is one free electron per atom we have:
3 30.1
ukin = EF = eV /atom.
5 (rs /a0 )2

To be more accurate we should replace this last equation by the complete ground state energy
per electron of a uniform electron gas at the density 1/(4πrs3 /3), but this calculation is quite
difficult and not much useful given the crudeness of the model we are using. Nonetheless, here
we include the exchange correction (antisymmetry of wave function):
12.5
uex = − eV /atom
rs /a0

It should be noted that the exchange correction has the same density dependence as the average
electrostatic energy, and it is about half its size. This suggests the importance of electron-
electron interaction in metallic cohesion and the consequently difficulty that a cohesion theory
must deal with.
Adding these three contributions together we find
30.1 36.8
u= 2
− eV /atom. (189)
(rs /a0 ) rs /a0

Minimizing this with respect to rs yields:


rs
= 1.6.
a0
The observed values for this ratio range from 2 to 6 in alkali metals. The failure of this formula
even to come close to the observed data is symptom of the difficulty in coming to terms with
metallic cohesion with any simple picture. A striking feature is that it predicts the same value
of rs /a0 for every alkali metal. This would not be corrected by any more accurate evaluation
of the total electron gas energy, since it would yield a constant value for each alkali metal.
Then there must be some other length scale that must be introduced to distinguish the various
metals. Evidently this comes from the non negligible radii of the ion cores. Even though the
point ion approximation is not absurd since they occupy a quite small fraction of the total
volume it ignores at least two important effects: if the ion cores have non zero radius, the
electron gas is prevented from entering the fraction of the volume occupied by the ion cores.
This leads to a greater density of the electrons and therefore a greater kinetic energy. Moreover,
as the conduction electrons are excluded from the ion cores’ volume, they cannot get arbitrarily
close to the positive ions as the picture underlying the attractive electrostatic energy assumes.
Therefore we should expect the electrostatic energy to be less negative than we have estimated.
Both these effects should have as results that the equilibrium value of rs /a0 increases with
increasing ion core radius, as experimentally observed. Evidently, such a calculation, in order
to be moderately accurate, requires good estimates of both the conduction electrons wave
functions and the crystal potential appearing in the one electron Schrödinger equation.

70
12 Lattice Dynamics
The static lattice picture must be deserted in favour of a dynamical one. Indeed at any tem-
perature greater than the absolute zero every atom in the system will undergo vibrations. In
order to save the picture obtained from x-ray scattering, which shows the presence of fixed
scattering centres, we assume the vibration amplitude of each atom to be smaller than half the
nearest neighbour distance. Another approximation that we are willing to do is the adiabatic
approximation: since the nuclei are much heavier than the electrons, the two motions can be
decoupled, so that they are described by two independent Schrödinger equations. A further
approximation is that of separation of variables, i.e. the total wave function of an atom is
written as the nuclear wave function times the electronic wave function. Therefore, since we
are interested in the motion of the nuclei, we can focus only on their motion and solve their
equation of motion.

12.1 1D Two Atoms System


We start by considering two atoms whose average equilibrium distance is r0 . We then make an
expansion of the potential
0 0  0 1 1
) (r − r0 ) + V 00 (r0 )(r − r0 )2 + V 000 (r0 )(r − r0 )3 + . . . →
*
V
V (r) =  (r
0) + V (r0:
2 6
1 00 2 1 000 3 (190)
→ x = r − r0 →= V (0)x + V (0)x + . . .
|2 {z } |6 {z }
≡harmonic term ≡anharmonic terms

From this expansion we can define the force constants as


(
K2 = V 00 (0) ≡ harmonic force constant
(191)
Kn = V (n) ≡ anharmonic force constants → typically k3 < 0

In the harmonic approximation the average of the displacement x is null: hxi = 0, i.e. there is
no thermal expansion. Although if we consider some anharmonicity, this is no longer true, we
lose the symmetry and hxi =6 0.
We will now study the two atoms in the harmonic approximation. Their positions are:
r1 (t) = 0 + x1 (t)
(192)
r2 (t) = r0 + x2 (t)
Therefore the equations of motion are
(
m1 ẍ1 = K0 (x2 − x1 )
−→ x = x2 − x1 ⇒ ẍ = ẍ2 − ẍ1 ⇒
m2 ẍ2 = −K0 (x2 − x1 )
 
K0 K0 K0
⇒ ẍ = − + x = − x , µ ≡ reduced mass ⇒
m2 m1 µ (193)
K0
⇒ x = ue−iωt , ω 2 = ⇒
µ
⇒ x(t) = Re(ue−iωt ) = A cos(ωt + ϕ)

u=Ae−iϕ

The total energy of the system is given by


p2 1 1
H= + K0 x2 = µω 2 A2 = E (194)
2µ 2 ↑ 2
p=µẋ

71
and we have that (
hxi = 0
2 (195)
hx2 i = A2 = E
µω 2

We can treat the same problem in the quantum case, thus we study the quantum har-
monic oscillator. Given x̂ the position operator and p̂ the momentum operator, the canonical
commutation rule holds:
[x̂, p̂] = i~. (196)
The Hamiltonian of the harmonic oscillator is
p̂2 1
H0 = + mω02 x2 (197)
2m 2
We can redefine the position and momentum operators to ease the notation
(
x̂ = mω
p 0
x
1
~

p̂ = √m~ω0 p
(198)
Ĥ0 1
⇒ [x̂, p̂] = i −→ Ĥ = = (x̂ + p̂)
~ω 2
Now we can define two additional operators
( (
â+ = √12 (x̂ − ip̂) x̂ = √1 (â+ + â)
1
⇒ 2
, [â, â+ ] = 1 (199)
â = √2 (x̂ + ip̂) p̂ = √1 (â+ − â)
2

It follows that
1 1 1 1
â+ â = (x̂2 + p̂2 + ix̂p̂ − ip̂x̂) = (x̂2 + p̂2 − 1) ⇒ Ĥ = â+ â + = ââ+ − (200)
2 | {z } 2 2 2
=−1

where N = â+ â is called the number operator. The eigenstates and eigenvalues of this operator
are
N |φν i = ν |φν i , ν ∈ N (201)
The eigenstates can be rewritten in the more transparent form

|φn i = |ni . (202)

The energy of the n-th harmonic oscillator is En = ~ω(n + 21 ). The action of the operators â+
and â on the number operator eigenstates is

â+ |ni = n + 1 |n + 1i −→ â+ ≡ creation operator
√ (203)
â |ni = n |n − 1i −→ â ≡ destruction operator

Evaluating now the mean value of the displacement and of its square we obtain
r
~
hxi = hn|â+ + â|ni = 0
2mω0
(204)
~ ~ 1 En
hx2 i = hn|(â+ + â)2 |ni = (n + ) =
2mω0 mω0 2 mω02

which is the same result as the classical harmonic oscillator.

72
12.1.1 Classical Statistics of the Harmonic Oscillator
We now take a look to the classical statistics of a one dimensional harmonic oscillator. The
average of the energy of the oscillator is hEi = kB T , as can be easily computed from the
partition function of the canonical ensemble:

e−βE(x,p)
ZZ
ρ(x, p) = R −βE(x,p) ⇒ hAi = A(x, p)ρ(x, p)dxdp (205)
e dxdp
p 2
Since in the expression of the energy the variables are not mixed E = 2m +V (x) we can integrate
the partition function aver the momenta to obtain a partition function of the positions
R −β  p2 

e 2m dp e−βV (x)
Z

ρ(x) = ρ(x, p)dp = R p2 R . (206)
e−β dp e−βV (x) dx
2m


Therefore the averages of the position and of its square are (in the harmonic approximation):
Kx2
xe−β
R
2 dx
hxi = R 2 =0
−β Kx
e 2 dx
2 (207)
2 −β Kx
R
xe dx 2 1
hx2 i = R Kx2
=
e−β 2 dx βK

This last result could also be obtained from the equipartition theorem:
 
1 1 1
Kx = kB T ⇒ hx2 i =
2


V (x) = (208)
2 2 βK

12.1.2 Quantum Statistics of the Harmonic Oscillator


In the quantum case the partition function is

1 X e−β 2
w = e−βH , Z = T r(w) = −βEn
e = (209)
Z n
1 − e−β~ω

then it follows that the energy is given by


 
X
−βEn 1 1
hHi = T r(Hw) = En e = ~ω β~ω
+ (210)
n
e −1 2

where
1
= nω (T ) (211)
eβ~ω
−1
is the Bose-Einstein population factor. We know that in quantum mechanics the Hamiltonian
can be written as  
+ 1
H = ~ω â a + (212)
2
Then we see that
hâ+ ai = nω (T ). (213)

73
12.1.3 Small Anharmonicity
We now wonder what happens when a small anharmonic term is added to the potential:
K0 2
V (u) = u + K3 u3 , K3 < 0 (214)
2
Then the classical equation of motion reads
dV K0 K 3 2
µü = − = −K0 u − 3K3 u2 ⇒ ü = −ω02 u − 3 u →
du µ K0
(215)
K3
→ S=3 → ü = −ω02 u − Sω02 u2
K0
we make the ansatz that the solution has the form

u = u0 + A cos(ωt) (216)

and we require that S · A  1. Plugging this solution into the equation of motion
(
u̇ = −ωA sin(ωt)

ü = −ω 2 A cos(ωt)

⇒ −ω 2 A cos(ωt) = −ω02 u0 − ω02 A cos(ωt) − Sω02 u20 + A2 cos2 (ωt) + 2Au0 cos(ωt)


2 2 2 2 2 2 2 2 2

⇒ Sω0 A cos (ωt) + cos(ωt) (ω0 − ω )A + 2ω0 SAu0 + ω0 u0 + Sω0 u0 = 0 −→
(
hcos(ωt)i = 0 2 SA2
−→ −→ Su 0 + u0 + =0 ⇒
hcos2 (ωt)i = 12 2

√ ( 2
−1 ± 1 − 2S 2 A2 −1 ± (1 − S 2 A2 ) − SA 2
⇒ u0 = ' =
2S 2S − S1
(217)
Plugging these solutions in the bracket that multiplies the cosine in the previous equation and
setting it to zero we obtain two expressions for the frequency
2S 2 A2
I. ω 2 = ω02 (1 − 2
) = ω02 (1 − S 2 A2 ) > 0
2S
II. ω 2 = ω02 (1 − S
) = −ω02 < 0

Since the second solution leads to a negative frequency, only the first is physically acceptable.
Thus
SA2 3 K3 2
u0 = − =− A > 0 , K3 < 0
2 2 K0
  (218)
2 2 K3
⇒ ω = ω0 1 + 6 u0
K0

12.2 Multi Atom Systems


We now consider a system with N atoms. We define the position of the l-th atom as R ~ l (t)
l
and its equilibrium position as ~x = hR ~ (t)i. Therefore the excursion of the atom from its
l

equilibrium position is
~ l (t) − ~xl .
~ul (t) = R (219)

74
We make the assumption that
h(~ul (t))2 i  d2nn (220)
where dnn is the nearest neighbour distance. The Hamiltonian of this system is
(
l 2
1 X (Pα ) n
~ l (t)
o α = x, y, z
H= +Φ R , (221)
2 l,α Ml l = 1, 2, . . . , N

Classically we have
∂Φ
fαl (t) = −
∂Rαl
(222)
d2 Rαl (t) l l
n
~ 1 (t), . . . , R
~ N (t)
o
⇒ Ml = f α (t) = f α R
dt2
this equation is not solvable analytically in the general case, but it must be solved through
numerical simulations. Since an exact solution is not feasible, we need to make some approxi-
mations: we make an expansion of the potential

n o n o 1 X X ∂ 2 Φ 0
Φ R ~l =Φ R ~l + 0 ulα ulβ + third order + . . . (223)
2 l l

eq 0
∂uα ∂uβ
l,α l ,β eq

We can choose n o
Φ R~l = 0 (224)

eq

Therefore the potential can be expressed as


n o 1 X X 0
Φ R ~l = Φαβ (ll0 )ulα ulβ (225)
2 l,α l0 ,β

where we defined
2
∂ Φ
Φαβ (ll0 ) ≡ (226)

l l0
∂uα ∂uβ
eq

which are (3N )2 quantities called force constants that represent the force on the l-th atom in
direction α due to an unitary displacement of the l0 -th atom along direction β.
The Hamiltonian in the harmonic approximation has the form

1 X (Pαl )2 1 X X 0
H= l
+ Φαβ (ll0 )ulα ulβ . (227)
2 l M 2 l,α l0 ,β

The force constants posses the following properties

I. Φαβ (ll0 ) = Φβα (ll0 ) −→ symmetry


0
P P
II. l l0 Φαβ (ll ) = 0 −→ translation of the solid does not change the potential:
XX
∆Φ(~v ) = Φαβ (ll0 )~vα~vβ = 0 , ∀~v ≡ translation vector (228)
l,α l0 ,β

75
0
Φαβ (ll0 ) = 0:
P P
III. l Φαβ (ll ) = l0

0 0
1 XX 0
∂ulα ulβ ∂ulα ulβ 0
fγi=− Φαβ (ll ) i
→ i
= ulα δl0 i δγβ + ulβ δli δγα →
2 l,α l0 β ∂uγ ∂uγ
!
1 X X 0
X
→ fγi = − Φαγ (li)ulα + Φγβ (il0 )ulβ = − Φαγ (li)ulα
2 lα l0 ,β
using
lα symmetry
l l
performing a translation of ~v : ~u → ~u + ~v , ∀l
X X
⇒ Φαγ (li)ulα = Φαγ (li)(ulα + vα ) , ∀~v ⇔ → the force does not
change for a translation
l,α l,α
!
X X X
⇔ Φαγ (li)vα = 0 ⇒ Φαγ (li) vα = 0
l,α α l
| {z }
=0

Now we want to diagonalize the harmonic Hamiltonian (feasible since it is positive definite
and quadratic in both the atomic momenta and positions)
d2 ulα (t) X 0
Ml 2
= f l
α (t) = − Φαβ (ll0 )ulβ 3N equations (229)
dt l0 β

To solve we make the ansatz that the solution is written in terms of the harmonic oscillator
Qλ (t) of frequency ωλ :
1
ulα (t) = √ elα (λ)Qλ (t) , Q̈λ (t) = −ωλ2 Qλ (t) (230)
Ml
Therefore 0
p
2 l
X el (λ)
0 β
Ml ωl eα (λ)
Qλ (t) = −

Φαβ (ll ) √  λ (t) ⇒
Q 
l0 ,β
Ml

Φαβ (ll0 ) (231)


 
0
X
⇒ elβ (λ) ωλ2 δll0 δαβ − √ =0
l0 ,β ↓ ↓ Ml Ml0
eigenvectors diagonal matrix ↓
(eigenvalues) 3N ×3N matrix
0
Which is a set of 3N linear algebraic equations in the 3N dependent variables elβ (λ).
To solve this eigenvalue problem we impose
Φαβ (ll0 )
 
2
det ωλ δαβ δll0 − √ = 0 −→ Secular equation (232)
Ml Ml0
The roots of the secular equation are the 3N eigenvalues ωλ2 , that we assume to be positive.
Indeed, if any of them was negative the system would be unstable. The eigenvectors of the
problem are ~el (λ), λ = 1, . . . , 3N .
The general solution to the equation of motion is any linear combination of these eqigen-
vectors, so that:
1 X
~ul (t) = √ ~e l (λ) · Qλ (t) −→ normal mode expansion (233)
Ml λ ↓ ↓
determines the normal
spatial movement coordinate
of the atom

Since the Hamiltonian is positive definite and symmetric we have that {~e l (λ)} is orthonormal
and complete:

76
l l 0
P
I. l,α eα (λ)eα (λ ) = δll0

l l0
P
II. λ eα (λ)eα0 = δαα0 δll0 → closure relation

Therefore
1X 2
H= {Q̇λ (t) + ωλ2 Q2λ (t)} (234)
2 λ
is a sum of 3N independent oscillators. To show this we rewrite the kinetic energy in terms of
the normal modes expansion:
2
1 X Pαl (t) 1X 2 1 X X l
Ekin = = Ml u̇lα (t) = e (λ)Q̇λ (t)elα (λ0 )Q̇λ0 (t) =
2 l,α Ml 2 l,α 2 l,α λλ0 α
(235)
1X 2
Q̇ (t).
2 λ λ

Keeping in mind that


X Φαβ (ll0 )elβ0 (λ)
elα (λ)ωλ2 = √ (236)
0
l ,β
Ml Ml0
we can write the potential energy in the normal modes expansion:
0
1 X 0 l l0 1 X X Φαβ (ll0 )elα (λ)elβ (λ0 )Qλ (t)Qλ0 (t)
Epot = Φαβ (ll )uα (t)uβ (t) = √ =
2 lα,l0 β 2 lα,l0 β λλ0 Ml Ml0
(237)
1 XX l 0 2 l 1X 2 2
= eα (λ )ωλ eα (λ)Qλ (t)Qλ0 (t) = ω Q (t)
2 l,α λλ0 ↑ 2 λ λ λ
orthonormality

Thus we can write


X
H= Hλ (238)
λ
1 2 2 2
 1 2
Pλ (t) + ωλ2 Q2λ (t)

Hλ = Q̇λ (t) + ωλ Qλ (t) = (239)
2 ↑ 2
Pλ (t)≡Q̇λ
generalized
momenta

Therefore (
~u l (t) = √1 ~e l (λ)Qλ (t)
P
√Ml
Pλ (240)
p~ l (t) = Ml λ~
e l
(λ)Pλ (t)

77
If we consider just two atoms in one dimension our formulas become
Φ(1, 2) = Φ(2, 1)
X X
Φαβ (ll0 ) = Φαβ (ll0 ) = 0
l l0
Φ(11) = −Φ(12) = Φ(22)
K0 1
V (u1 , u2 ) = (u − u2 )2 ⇒
( 2
2V
Φ(12) = ∂u∂1 ∂u 2 = −K0
⇒ ∂2V
Φ(11) = ∂(u1 )2 = K0
 K0 K0   
0 −√ 1
 1
Φαβ (ll ) M 1 M M e 2 e
⇒√ −→  1 2
 2 =ω ⇒
 
Ml Ml0 K 0 K 0 e e2
−√
M1 M2 M2

ω1 = r 0
⇒ K0 M1 M2
ω2 = , µ≡
µ M1 + M2
We now study the two normal modes that we found, starting from ω = 0:
K0 1 Ko
e −√ e2 = 0 ⇒
M1 M1 M2
r
M2 1
⇒ e2 = e ⇒ (241)
M1
e1 e2
⇒ √ =√ → pure translation
M1 M2
Normalization yields:
  r r
1 2 M2 M1
1 µ
(e ) 1+ =1 ⇒ e = =
M1 M1 + M2 M2
r (242)
µ
e2 =
M1
r
K0
Studying now the other solution, ω = :
µ
r
1 1 1 M1 1
e +√ e2 = 0 ⇒ e2 = − e (243)
M2 M1 M2 M2
i.e. the displacements are in opposed directions. Therefore we have
M1 e1 M2 e2
√ +√ =0 (244)
M1 M2
which implies that the centre of mass is fixed for this mode.
  r r
1 2 M 1 1 M2 µ
⇒ (e ) 1 + = 1 ⇒ |e | = =
M2 M1 M2 M1
 r
µ (245)
e1 (λ = 2) = −


⇒ r M1
µ
e2 (λ = 2) =


M2

78
Then √   1  
1 1
√ M1 u e (λ = 1) e (λ = 2) Q 1 (t)
= 2 (246)
M2 u2 e (λ = 1) e2 (λ = 2) Q2 (t)
| {z }
≡P

The eigenstates matrix P is orthogonal, so that P −1 = P T , therefore


 
Q1 (t) p
= P T { Mi ei } =
Q2 (t)
r r 
µ µ
 1  √  √  (247)
e (λ = 1) e1 (λ = 2) √ M 1 u 1  M2 M M1 u1
= r µ r µ  √ 1

= 2  
e (λ = 1) e2 (λ = 2) M2 u2 M2 u2

M1 M2
r 
µ
Which represents a rotation by an angle θ = arcsin . Then we have
M1
µ M1 − µ M1 µ
cos2 θ = 1 − = = = ⇒
M1 M1 M1 + M2 M2
  (248)
cos θ sin θ
⇒ PT =
− sin θ cos θ
r r
M1 µ 1 M2 µ 2 M1 u1 + M2 u2
Q1 (t) = u + u = √ ← center of mass coordinates
M2 M1 M1 + M2
√ 1 2
 (249)
Q2 (t) = µ −u (t) + u (t)
Qλ (t) = Re{eλ eiωλ t } = Aλ cos(ωλ t + φ)
If we had N atoms in 3D:
X el (λ
√α Qλ (t) ⇒
ulα (t) =
Ml
√  λ  
M1 u1α (t) e1x (λ) e1y (λ) e1z (λ) Q1 (t)
⇒ ..   .. .. ..   .. 
= .

√ . N . .  .  (250)

N N N
MN uα (t) ex (λ) ey (λ) ez (λ) QN (t)
| {z }
≡P
X
elα (λ)elα (λ0 ) = δλλ0
l,α

0
P is orthogonal and we have the completeness relation λ elα (λ)elβ (λ) = δαβ δll0 . Then
P

  √ 
Q1 (t) M1 u1 (t)
 ..  T  ..
 . =P   ⇒

√ .
QN (t) MN uN (t)
X p X el (λ) (251)
α
⇒ Qλ (t) = elα (λ) Ml ulα (t) −→ ulα (t) = √ Qλ −→
l,α λ
Ml
XX
−→ = elα (λ)elα (λ0 )Qλ (t) = Qλ (t)
λ0 l,α

This is hard to solve since we have a 3N × 3N force constants matrix. In the case of a crystal
things are easier. Until now we did not make any hypothesis on the structure. We can write

79
our Hamiltonian as function of harmonic Hamiltonians:
N
X
Ĥ = Ĥλ
λ=1
1 2 (252)
Pλ + ωλ2 Q2λ

Ĥλ =
2
Pλ = Q̇λ

In the classical description we have


1
hHλ i = 2 · kB T = kB T
2
kB T (253)
hωλ2 Q2λ i = kB T ⇒ hQ2λ i =
ωλ2
hQ̇2λ i = kB T

In the steady state we have


hQ2λ (t)i = hQ2λ (0)i (254)
We will see the quantum description later with some experiments.

12.3 Linear Chain of Atoms


We consider to have N atoms of mass M in one-dimension with an interaction that is not limited
at nearest neighbours, in the harmonic approximation:

I. φ(l, l0 ) = φ(l0 , l)

φ(l, l0 ) = φ(l, l0 ) = 0
P P
II. l0 l

III. φ(l, l0 ) = φ(a, l − l0 ) ≡ φ(l − l0 ) ←− from translational invariance of the lattice

Therefore
d2 u
ul (t) = xl eiωt ⇒ = −ω 2 ul (255)
dt2
and the equation of motion is
l0
 P 0
2 l
M ω u (t) = 0 φ(l − l)x

0 l
X
M ül (t) = − φ(ll0 )ul ⇒ ∞
0
2 l+1
φ(l0 − l − 1)xl = φ(m − l)xm+1 , m = l0 − 1
P P
l0 ω M u =

l0 =−∞ m
(256)
These two equations can be written as operator equations
(
P xl = 0
(257)
P xl+1 = 0

Thus we want to find two solutions that satisfy these equations. We have

xl+1 = kxl = eika xl (258)

80
which is an expression of the Bloch theorem. Starting from 0

x1 = eika x0 ⇒ xl = eikal x0 (259)

Then we have that k must be real, otherwise the solution will diverge at ±∞. Therefore we
have plane waves. Implementing the periodic boundary conditions:

atom N + 1 = atom 1 (260)

we have
xN +1 = eika(N +1) x0 = eika x0
2πm (261)
⇒ kaN = 2πm ⇒ k =
aN
thus m and m + N have the same solution. Therefore we can confine the index m:
N N π π
− <m≤ ⇒ − <k≤ (262)
2 2 a a
i.e. K is confined in the First Brillouin Zone. This comes from the fact that periodicity in the
direct space implies a periodicity in the solution in the reciprocal space.
0
X
⇒ M ω 2 (k)eikal = φ(l0 − l)eikal ⇒
l0

1 X 0
⇒ 2
ω (k) = φ(l0 − l)eika(l −l) =
M l0 =−∞

1 X
= φ(l)eika(l) =
M l=−∞ (263)
 
∞ ∞
1 X X
= φ(l)eikal + φ(−l) e−ikal  =

2M l=−∞

| {z }
l=−∞
≡φ(l)

1 X
= φ(l) cos(kal)
M l=−∞

We have some general properties


I. ω 2 (k) = ω 2 (−k)
II.
0 00
2 2
2 2

ω (k) ' ω (0) + (ω (k)) + (ω (k)) + ...
k1
k=0 k=0

2 1 X
ω (0) = φ(l) = 0
M l=−∞ (264)

0
2

(ω (k)) =0
k=0
⇒ ω (k) ' const k 2 + · · · = vk
2
k1

where v is the speed of sound. This is the dispersion relation in the case of long wavelength
(i.e. k small). k is the wavevector

xl = eikal x0 , if k = , λ = al ⇒ xl = x0 (265)
λ
81
We now consider the interaction to be only between nearest neighbours (see figure 8), with
the atoms linked together by springs of elastic constant k0 :
k0 l
V (ul , ul+1 ) = (u − ul+1 )2 (266)
2
Then
X k0
φ({ul }) = V (ul , ul+1 ) = (ul − ul+1 )2 + (ul−1 )2

l
2
X (267)
⇒ φ(l) = 0 = φ(−1) + φ(0) + φ(1)
l

By symmetry we have that


φ(−1) = φ(1) (268)
Then
(
φ(0) φ(0) = 2k0
2φ(1) + φ(0) = 0 ⇒ φ(1) = φ(−1) = − ⇒ (269)
2 φ(1) = φ(−1) = −k0

1  
2 1 X 1 φ(0) φ(0)
ω (k) = φ(l) cos(kal) = − cos(−ka) + φ(0) − cos(ka)
M l=−1 M 2 2
 
φ(0) φ(0) ka
= (2 + 2 cos(ka)) → cos(2θ) = 1 − 2 sin2 (θ) → ω 2 (k) = 2 sin2 ⇒
2M M 2
r  
2φ(0) ka
⇒ ω(k) = sin
M 2
q (270)
2φ(0)
where M
is the maximum frequency. A plot of the dispersion relation can be seen in figure
9.

Figure 8 Figure 9

82
Figure 10

The density of vibrational states is


N
X
ρ(ω) = δ(ω − ωλ )
λ=1
2πm N N
k= , − +1<m≤
Na 2 2
Na
ρ(ω)dω ≡ # of states between ω and ω + dω = 2dn = 2 dk , k > 0 ⇒
  2π (271)
Na 1 dω ka a
⇒ ρ(ω) = dω
→ = ωM cos →
π dk dk 2 2
1
2N 1 2N ωM 2N 1
→ ρ(ω) = ka
= s = p
πωM cos 2 π 
ω
 2 π 2
ωM − ω 2
1−
ωM
which must be normalized Z ωM
ρ(ω)dω = N (272)
0
and has a behaviour for small frequencies
ρ(ω) ' ω d−1 d ≡ dimension (273)
ω→0

as can be seen in figure 10

12.4 Crystal Dynamics


We have
~xl , l = 1, . . . , N Bravais Lattice Points
~xχ , χ = 1, . . . , r basis
(274)

⇒ ~xlχ = ~xl + ~xχ


where ~xlχ represents the average position of the atoms
~ lχ (t)i = ~xlχ , R
hR ~ lχ (t) = ~xlχ + ~ulχ (t) (275)
where ~u lχ is the displacement vector.
The total number of atoms is given by rN , with N = N1 N2 N3 ≡ total # of cells. The
momentum of an atom is
p~ lχ (t) = Mχ~u˙ lχ (t) (276)
Then the Hamiltonian reads
2
1 X p~ lχ (t) 1 XX 0 0
H= + φαβ (l0 χ0 , lχ)~uαlχ~uβl χ (277)
2 lχ 2Mχ 2 l0 χ0 α lχβ

83
In crystals we have that

φαβ (lχ, l0 χ0 ) = φαβ (0χ, (l − l0 )χ0 ) (278)

and with a monatomic Bravais lattice


0 0
φαβ (l, l0 ) = φαβ (l0 , l) ⇔ φαβ (0, ~x l − ~x l ) = φαβ (0, −(~x l − ~x l )) (279)
| {z } | {z }
=φαβ (l0 −l) =φαβ (l−l0 )

This last relation can be seen from the potential


1 XX l X
Epot = uα (t) φαβ (l0 − l)ulβ (t)
2 ll0 α β
| {z }
x l0 ) ⇒
x l ,~
=U (~
0 0
(280)
⇒ U (~x l , ~x l ) = U (−~x l , −~x l ) ⇒
X X
⇒ ulα (t) φαβ (−l0 + l)ulβ (t) ⇒ φαβ (l − l0 ) = φαβ (l0 − l)
α β

We obtained plane wave solutions


1 X ~ l
ulχ
α (t) = p ~eχα (~kj)eik·~x Qkj (t) (281)
N Mχ ~
kj

if λ −→ ~kj then we have


~el (λ) ~eχ (~kj) i~k·~x l
p = pα e (282)
Mχ N Mχ
To prove this we use the following ansatz:
1 ~ l
uαlχ (t) = p uχα (~k)eik·~x eiωt

⇒ ~u¨ lχ (t) = −ω 2~u lχ (t)

X 0 0
⇒ −Mχ ω 2 uαlχ (t) = − φαβ (lχ, l0 χ0 )ulβ χ ⇒ (283)
ll0 β
X φαβ (0χ, (l − l0 )χ0 l0 χ0 ~ l ~ l
⇒ p uβ (~k)eik~x = ω 2 uχα (~k)eik~x −→ m = l − l0 −→
lχ0 β
Mχ Mχ0
X φαβ (0χ, mχ0 ) ~ m χ0
−→ p eik~x uβ (~k) = ω 2 uχα (~k)
mχ0 β
Mχ Mχ0

We went from 3rN equations to a 3r × 3r matrix, called dynamical matrix:

χχ 0 X φαβ (0χ, lχ0 ) ~ l


Dαβ = p eik~x
l
M M
χ χ 0
(284)
~ ~k) = ω 2 u(~k)
⇒ Du(

Then we obtain 3r solutions for each value of ~k:

ω 2 = ωj2 (~k) , j = 1, . . . , 3r (285)

84
The dynamical matrix is hermitian
 ∗
χχ0 ~ χχ0
Dαβ (k) = Dβα (−~k) (286)

therefore ω 2 (~k) are real and symmetrical ωj2 (~k) = ωj2 (−~k). Moreover we assume that they are
positive, which adds a constraint on the force constants.
ω = ωj (~k) are the 3r vibrational branches of the crystal.

~ e(~k) = ω 2~e(~k)
D~
P
 e∗χ ~ χ~ 0
α (kj)eα (kj ) = δjj 0 ←− orthonormality
χα
P e∗χ (~kj)eχ0 (~kj) = δ δ 0 ←− closure relation
j α β αβ χχ
X χχ0 0
(287)
→ ωj2 (−~k)eχα (−~k) = Dαβ (−~k)eχβ (−~kj) ⇒
αβ
χχ0 ∗χ0
X
⇒ Dαβ (~k)eβ (−~kj) = e∗χ ~ 2 ~
α (−kj)ωj (−k)
αβ

which states that ωj2 (−~k) follows the same secular equation as ωj2 (~k). Another important
property of the eigenvectors is that

eα∗χ (−~kj) = eχα (~k, j) (288)

Now we discuss the properties of the dynamical matrix in the case of a monatomic crystal
X φαβ (0, l) ~ l
Dαβ (~k) = eik~x
l
M
φαβ (0, l) = φαβ (l)
X (289)
φαβ (l) = 0 ⇒
l
X
⇒ φαβ (l) = −φ(0)
l6=0

Then
1 X  l
~

Dαβ (~k) = φαβ (l) eik~x − 1 (290)
M l6=0

Since in a monatomic lattice the spatial parity of the force constants is even: φαβ (l) = φαβ (−l),
we have that
!
X ~ l 1 X ~ l
X ~ l
X  
φαβ (l)eik~x = φαβ (l)eik~x + φαβ (−l)e−ik~x = φαβ (l) cos ~k~xl (291)
l6=0
2 l6=0 l6=0 l6=0

Thus we have
!
1 X     2 X ~k~xl
Dαβ (~k) = φαβ (l) cos ~k~xl − 1 = − φαβ (l) sin2 (292)
M l6=0 M l6=0 2

Thus we have three branches ω = ωj (~k) , j = 1, 2, 3.


In the limit of small ~k we have
k2 X
lim Dαβ (~k) = − φαβ (~xl )(k̂ · ~xl )2 ⇒ ωj (~k)~k→0 = vj (~k)~k (293)
~k→0 2M l6=0

85
Figure 11

This shows that in the long wavelength limit the dispersion relation vanishes linearly with
~k which is typical of the acoustic modes in the elastic theory of continuous media (see next
chapter).
We now consider another classical example, i.e. the linear chain with two masses per Bravais
lattice point and only nearest neighbour interactions
1 X
(u1l − u2l )2 + (u1l − u2l−1 )2 + (u2l − u1l+1 )2

φ({u}) = k0 (294)
2 l

we have that 2
φ(01, 01) = ∂∂uφ2 = 2k0 φ(02, 02) = 2k0 φ(01, 02) = −k0
0 (295)
φ(02, 11) = −k0 φ(01, −12) = −k0 φ(02, 01) = −k0
Thus
0
X φ(0χ, lχ0 ) ~ l
Dχχ (~k) = p eik~x ⇒
l
M M
χ χ 0

φ(01, 01) 2k0


⇒ D11 (k) = =
M1 M1
φ(01, 02) φ(01, 12) −ika k0 (296)
⇒ D12 (k) = √ + √ e =√ (1 + e−ika )
M1 M2 M1 M2 M1 M2
k 0
⇒ D21 (k) = D12∗ = √ (1 + eika )
M1 M2
2k0
⇒ D22 (k) =
M2
where a is the spacing between nearest neighbours.
Now we want to diagnolize D. We start evaluating the determinant:
k0 k0
ω 4 − 2(k1 + k2 )ω 2 + k1 k2 (2 − 2 cos(ka)) = 0 , k1 = , k2 = ⇒
M1 M2
s   (297)
2 ka
⇒ ω12 = k1 + k2 ± (k1 + k2 )2 − 4k1 k2 sin2
2

Then the max value for the first eigenvalue is


p
(ω12 )max = k1 + k2 − (k1 + k2 )2 − 4k1 k2 = 2k2 , k1 > k2 (298)

while in the small k limit it has the form


s  2 s
2 2
ka k1 k2 (ka)2
lim ω1 = k1 + k2 − (k1 + k2 ) − 4k1 k2 = k1 + k2 − (k1 + k2 ) 1 − '
k→0 2 (k1 + k2 )2
!
k1 k2 (ka)2 k1 k2 (ka)2
' k1 + k2 − (k1 + k2 ) 1 − = −→ 0
2(k1 + k2 )2 2(k1 + k2 ) k→0
(299)

86
Figure 12

therefore the speed of sound is r


k1 k2 a
v= √ (300)
k1 + k2 2
The extremal values of the second eigenvalue are
p
(ω22 )min = k1 + k2 + (k1 + k2 )2 − 4k1 k2 = 2k1
(301)
(ω22 )max = k1 + k2 + k1 + k2 = 2(k1 + k2 )

Since it does not vanish linearly with k this is called optical branch.
We now look at the eigenvectors’ behaviour for k → 0:

I. ω1 −→ 0
k→0

e1 (0, 1) e2 (0, 2)
  1 
2k
√ 1 2 k 1 k2 e (k = 0, j = 1)
= 0 ⇒ √ = √ (302)
2 k1 k2 2k2 e2 (k = 0, j = 2) M1 M2

i.e. in the acoustic branch the atoms move together (in phase).
r
2k0
II. ω2 −→
k→0 M1 + M2
⇒ M1 e1 (0, 1) + M2 e2 (0, 2) = 0 (303)
i.e. the centre of mass is fixed. This gives the name optical branch since we have a
time-varying dipole moment that can be coupled through a laser.
π
While at the boundary of the Brillouin zone, k = , we have
a
π
ω12 = 2k2 ⇒ e1 ( , 1) = 0 −→ only M2 moves
a (304)
2 2 π
ω2 = 2k1 ⇒ e ( , 2) = 0 −→ only M1 moves
a
A plot of the two dispersion relations can be seen in figure 12.
If M = M1 = M2 :   
2 2k0 ka
ω12 = 1 ± cos (305)
M 2
We can use the reduced Brillouin scheme, supposing that we fold the extended scheme inside
the first Brillouin zone, as can be seen in figure 13.

87
Figure 13

We now proceed with a brief summary of the main ideas of crystal dynamics in the harmonic
approximation. We have N unit cells, each containing r atoms. Thus the solution to the
equation of motion has the form
X
ulχ (t) = ulχ (~k, j)Q~kj (0)eiωt (306)
~k,j

where
~eχ (~kj) i~k~x l
ulχ (~k, j) = p e (307)
N Mχ
Thus we have 3r branches (→ j = 1, . . . , 3r), 3 of whose are acoustic (2 transverse acoustic
and 1 longitudinal acoustic), while the remaining 3r − 3 are optic branches (transverse optic
and longitudinal optic)
As a solution of the problem we find a maximum frequency at which the crystal can vibrate
(same condition that was applied by Debye).

12.5 Periodic Boundary Conditions


With this kind of boundary conditions we have discrete acceptable wavevectors that have the
form
~k = l1 ~b1 + l2 ~b2 + l3 ~b3 (308)
N1 N2 N3
where ~bi are the reciprocal space unit vectors, Ni the number of unit cells along direction i and
li = 0, ±1, . . . , ±Ni an integer.
The number of ~k points compatible with these conditions is infinite, but not every one of
them is independent from the others. Indeed if a pair of ~k differs only by a reciprocal lattice
vector ~τ , they are the same
~
1, ∀R
~k 0 = ~k + ~τ i~k0 R
~ i~kR
~ i~ ~
τ
>

R
⇒ e =e e
 (309)

So that we now restrict ourselves to the first Brillouin zone

χχ0 ~
X φαβ (0χ, lχ0 ) ~ l
Dαβ (k) = p eik~x , ~
~xl ≡ R
l
M M
χ χ 0
(310)
χχ0 χχ0 xl χχ0
Dαβ (~k + ~τ ) = Dαβ (~k)e i~
τ~
= Dαβ (~k)

88
thus the dynamical matrix is invariant under a reciprocal lattice translation. This invariance
is immediately reflected on eigenvectors, eigenvalues and normal coordinates:

~e l (~k, j) = ~e l (~k + ~τ , j)
ωj (~k) = ωj (~k + ~τ ) (311)
Q~kj (t) = Q~k+~τ ,j (t)

These hold for any reciprocal lattice vector ~τ . Therefore the wavevectors that give rise to
different solutions are those N which belong to the first Brillouin zone. It is worthwhile to
note that this implies that there are 3rN independent solutions, which is exactly the number
of degrees of freedom of the system.

89
13 Theory Of Elasticity
In the theory of elasticity solids are considered as continuous media neglecting the microscopic
structure of the solid. The deformation of the solid is described through the deformation
vectorial field ~u(~r). This field describes how the point ~r changes after the deformation to the
point ~r0 .
We consider two points that before the deformation are a distance dxi apart, while after
the deformation they are separated by dx0i = dxi + dui . Then we have for the solid
X X X ∂ui
dl02 = dx02
i = (dxi + dui )2 → dui = dxk →
i i k
∂x k
!
X X ∂ui X ∂ui ∂ui
→ dx2i + 2 dxi dxk + dxk dxl =
i k
∂x k
k,l
∂x k ∂x l
X ∂ui X ∂ui ∂ui (312)
= dl2 + 2 dxi dxk + dxk dxl =
i,k
∂x k
ikl
∂x k ∂xl
| {z } | {z }
∂uk P ∂ul ∂ul
= ikl ∂xk ∂xi dxk dxi
P
= ik ∂xi dxi dxk
X
= dl2 + 2 uik dxi dxk
ik

Where uik is the strain tensor


!
1 ∂ui ∂uk X ∂ul ∂ul
uik = + + (313)
2 ∂xk ∂xi l
∂xk ∂xi

which is symmetric. It can be diagonalised therefore at any given point the axes can be chosen
in a way that only the diagonal elements are non null. The principal values of the strain tensor
are u(1) , u(2) and u(3) . Therefore
X
dl02 = (δik + 2uik )dxi dxk = (1 + 2u(1) )dx21 + (1 + 2u(2) )dx22 + (1 + 2u(3) )dx23 . (314)
i,k

Therefore the length dxi along the principal axis i becomes


p
dx0i = 1 + 2u(i) dxi (315)

which implies that


dx0i − dxi p
= 1 + 2u(i) − 1. (316)
dxi
We now make the approximation of small relative deformation
dx0i − dxi
1 (317)
dxi
In the general case a small relative deformation implies a small strain. Then we can neglect
the second derivative terms in the strain tensor, which now reads
 
1 ∂ui ∂uk
uik = + (318)
2 ∂xk ∂xi
Then we have
dx0i − dxi p
= 1 + 2u(i) − 1 ' u(i) (319)
dxi

90
Now we look at what happens to the volume element

dxi = (1 + u(i) )dxi


dV 0 = dx01 dx02 dx03
(320)
dV = dx1 dx2 dx3
⇒ dV 0 = (1 + u(1) )(1 + u(2) )(1 + u(3) )dV = (1 + u(1) + u(2) + u(3) )dV
P
We note that there is the sum of the of the principal values of the strain tensor, i uii which
is well known to be invariant and it is equal in any coordinate system. Moreover we note that
it is equal to the relative change in volume
dV 0 − dV X
= Uii . (321)
dV i

When the body is not deformed it is in mechanical equilibrium, while when the deformation
acts it leads to internal stresses given by forces between molecules or atoms. This forces are of
short range (since we are considering the body on a macroscopic scale we can take their range
to be null). The forces acting deep in the bulk of the solid balance themselves, except for the
forces on the surface of the solid (an exception is for pyroelectric or piezoelectric materials, but
we will not consider them), as can be seen in the following:
Z X ∂σik
Fi dV = , Fi = , σik ≡ stress tensor
V k
∂x k
X Z ∂σik XI (322)
= dV = σik dfk
k V ∂x k
k S
P
where k σik dfk is the i-th component of the force acting on the surface dfk . It could be proven
that the stress tensor is symmetrical.
We need also to use the thermodynamics:
X
dE = T dS − P dV = T dS + σik duik (323)
ik

for a unit volume element in the non deformed body. The last term is the work in terms of the
strain tensor.
We consider a uniform compression σik = −P δik :
X
dE = T dS − P duii (324)
i

P dV 0 − dV
where i uii = is the change in volume of the body. The stress tensor can be derived
dV
from the energy at constant entropy
 
∂E
σik = , (325)
∂uik S
or we can derive it from the free energy at constant temperature
 
X ∂F
F = E − T S ⇒ dF = −SdT + σik duik ⇒ σik = . (326)
ik
∂uik T

In order to be able to apply the thermodynamic formulae for the general case we need to
have an expression of the free energy in terms of the stress tensor components. This is easily

91
done knowing that the deformations are small and expanding the free energy in terms of the
strain tensor components uik . We will consider only isotropic bodies:
in the undeformed state we have
uik = σik = 0 (327)
since
∂F
σik = =0 (328)
∂uik
it follows that there is no linear term in the power expansion. Next, since F is a scalar, each
term in its expansion should be a scalar also. Then we can define two scalars of second order
in the strain: the squared sum of the diagonal components and the sum of the squares of all
the components. Therefore we have, up to second order:
!2
1 X X
F = F0 + λ uii + µ u2ik (329)
2 i ik
where λ and µ are called Lamé coefficients. We have the following properties
P
I. if i uii = 0, then there is no volume change during the deformation and we have a pure
shear.
II. if uik = cδik then we have only a change in volume and not in shape, thus we have an
hydrostatic compression
These conditions can be summed up decomposing the strain as follows:
1 X 1 X
uik = uik − δik ull + δik Ull (330)
3 3
| {z l } | {zl }
pure shear hydrostatic compression

As general expression for the free energy of a deformed body it is convenient to replace
the previous expression with one exploiting the preceding decomposition of the strain. The F
becomes: !2 !2
X 1 X 1 X
F =µ uik − δik ull + K ull (331)
ik
3 l
2 l

where K ≡ 32 µ + λ is the bulk modulus, or the modulus of the hydrostatic compression and µ
is the shear modulus, or the modulus of rigidity.
Since at equilibrium F is at a minimum, and at equilibrium uik = 0, we have that F has a
minimum for uik = 0. Then we have that µ > 0, K > 0:
!
∂F X 1 X
σik = = Kδik ull + 2µ uik − δik ull (332)
∂uik l
3 l

which determines the stress tensor in terms of the strain tensor for an isotropic body. We can
easily invert this expression. We start evaluating the sum of the diagonal terms:
P
X X X σii
σii = 3K ull ⇒ ull = i . (333)
i l l
3K

substituting this result in the previous expression we obtain:


σ − ik − 13 δik l σll
P
X σll
uik = δik + (334)
l
9K 2µ

92
which gives the strain in function of the stress.
If we have an hydrostatic compression, i.e. a compression equal in every direction and
orthogonal to every surface
 
X P 1 1 X 1 ∂V
σik = −P δik ⇒ uii = − ⇒ =− uii = − (335)
i
K K P i V ∂P T

where the last equality comes from the thermodynamics definition of compressibility, which is
the inverse of the bulk modulus.
The previous expression which gives the strain in function of the stress is the Hooke’s law,
i.e. the strain is linear in the stress, which holds for small variations of volume.
We now take into consideration a crystal subjected to a small deformation. The free energy
must be quadratic in the strain:
1X
F = λiklm uik ulm (336)
2 iklm

where λiklm is a rank 4 tensor called Elastic Modulus Tensor. Since the strains do not change
inverting their indexes, we have that the tensor follows

λiklm = λkilm = λkiml = λikml = λlmik . (337)

Moreover we can write


λiklm = λαβ (338)
where

i, k, l, m = 1, 2, 3
α, β = 1, . . . , 6 = xx, yy, zz, xy, yz, zx

Then λαβ = λβα is a 6 × 6 matrix with 21 independent components. Greater the symmetry
of the crystal cell, smaller the number of independent components (e.g. the triclinic has 21,
while the simple cubic has 3). These components are the elastic stiffness constants (or elastic
moduli).
If we are in adiabatic conditions we have
1X
E= λiklm uik ulm (339)
2 iklm

where λiklm is the adiabatic elastic tensor.


The inverse of the matrix λαβ , S = {λαβ }−1 , is the elastic compliance matrix whose com-
ponents are the elastic constants.
Then the stress tensor is
 
∂F X
σik = = λiklm ulm . (340)
∂uik T lm

If we perform a change of coordinates, xi → x0i we have


1X 1X
E= λiklm uik ulm = λiklm u0ik u0lm (341)
2 iklm 2 iklm
where
∂u0i ∂u0k
 
1
u0ik = + (342)
2 ∂x0k ∂x0i

93
the coefficient of the elastic moduli tensor with an odd repetition of an index are null so that
we are left with

λ11 = λxxxx = λyyyy = λzzzz


λ12 = λxxyy = λyyzz = λxxzz
λ44 = λxyxy = λxzxz = λyzyz

In the case of an isotropic body (i.e. the most symmetric) the body itself does not change
for a rotation. The energy density is
!2
λ X X
E= uii + µ u2ik =
2 i ik
λ 2
uxx + u2yy + u2zz + 2uxx uyy + 2uxx uzz + 2uyy uzz +

= (343)
2
+ µ u2xx + u2yy + u2zz + 2u2xy + 2u2yz + 2u2xz =

 
λ
+ µ u2xx + u2yy + u2zz + λ (uxx uyy + uxx uzz + uyy uzz ) + 2µ u2xy + u2yz + u2xz
 
=
2

then for isotropic bodies we can define



λ = λ12

µ = λ44 (344)

λ11 = λ + 2µ = λ12 + 2λ44

From the last equation we can obtain the condition of isotropy

λ11 − λ12 − 2λ44 = 0 (345)

A polycrystal, which is made up with many microcrystallites each oriented in a different direc-
tion) closely follows the isotropy condition.

13.1 Elastic Waves in Crystals


We assume to be in adiabatic conditions, thus we have the adiabatic constant moduli. From
the equations of motion
X ∂σik
ρüi = Fi =
K
∂xk
X
σik = λiklm ulm ⇒
lm
∂σik X ∂ulm

1 ∂ul ∂um
 (346)
⇒ = λiklm −→ ulm = + −→
∂xk lm
∂x k 2 ∂x m ∂x l
 2
∂ 2 um

X 1 ∂ ul
−→ = λiklm +
lm
2 ∂x k ∂xm ∂xk ∂xl

where the two terms in the last bracket are equal when we put l = m, then we have
X ∂ 2 ui
ρüi = λiklm (347)
klm
∂xk ∂xl

94
Writing the solution as
~
ui (t) = U0i ei(k~r−ωt) (348)
we have X
−ω 2 ρu0i = − λiklm kk kl u0m ⇒
klm
X X
! (349)
2
⇒ λiklm kk kl − ρω δim u0m = 0
m lm
which is an eigenvalues problem. We have a symmetric tensor:
X
Tim = λiklm kk kl , Tim = Tmi . (350)
kl

Our problem involves a 3 × 3 matrix, therefore we have three eigenvalues ω = ωj (~k). Their
dependence is only on the direction of k rather than from its modulus, due to the double k in
ωj
the equation (k̂).
k
We can define the group velocity as
~v = ∇ω(~k) ⇒ ~v = ~v (k̂) (351)
we have that it depends only on the direction of k.
For a cubic crystal X
Tim = λiklm kk kl
kl
Txx = λ11 kx2 + λ44 (ky2 + kz2 ) (352)
Tyy = λ11 ky2 + λ44 (kx2 + kz2 )
Txy = (λ12 + λ44 )kx ky
while for an isotropic body we have the isotropic condition
λ11 = λ + 2µ (353)
thus
(λ + 2µ)kx2 + µ(ky2 + kz2 )
 
(λ + µ)kx ky (λ + µ)kx kz
 (λ + µ)ky kx (λ + 2µ)ky2 + µ(kx2 + kz2 ) (λ + µ)ky kz  (354)
2 2 2
(λ + µ)kz kx (λ + µ)kz ky (λ + 2µ)kz + µ(kx + ky )
With ~k = k(1, 0, 0) we have
 
(λ + 2µ)k 2 0 0
 0 µk 2 0  (355)
0 0 µk 2
and we have three solutions, two of whose are degenerate:
r
λ + 2µ
I. ωL = which is a longitudinal wave (i.e. with the same polarization as k) and
µ
the group velocity is
s s
λ + 2µ 2 k + 34 µ
vL = →k =λ+ µ → (356)
ρ 3 ρ
3
where k + µ is the longitudinal modulus.
4
µ
r
II. ωT = k which are two degenerate solutions representing two strictly transverse waves
ρ
(i.e. orthogonal to ~k).

95
13.2 Connection with Microscopic Dynamics
We consider a monatomic Bravais lattice in the harmonic approximation:
1X 0
UHA = φαβ (l − l0 )ulα ulβ (357)
2 lαl0 β

which, thanks to the properties of the force constants can be rewritten as


1 X X l0 0
uHA = − (uα − ulα )φαβ (l − l0 )(ulβ − ulβ ) =
4 lα l0 β
(358)
1 XX l l 0 0 1 X X l0 l 0
=− (uα uβ + uαl ulβ )φαβ (l − l0 ) + (uα uβ + ulα ulβ )φαβ (l − l0 )
4 lα l0 β 4 lα l0 β

where the first term on the right hand side vanishes since l φαβ (l − l0 ) = 0. If now we consider
P
displacements that that vary little from cell to cell, we can introduce a continuous function of
the coordinate ~u(~r) which in every Bravais lattice points coincides with ~u l = ~u(~x l ). Then we
can make a Taylor expansion
0
 0 
~u(~x l ) = ~u(~x l ) + ~x l − ~x l ∇~u(~r) ~r=~x l
∂ul (359)
  
l0
X 0
(xlγ − xlγ ) αl
l

where ~x − ~x ∇~u(~r) ~r=~x l =
α γ
∂uγ

Then the potential can be written

1 X X X l0 l ∂uα
l
0
X 0
l l
∂ulβ
UHA =− (x − xγ ) l φαβ (l − l) (xδ − xδ ) l (360)
4 lα l0 β γ γ ∂xγ δ
∂xδ

In the infinite crystal approximation we have


X 0 X 0 X 0 0 0
(xγl − xlγ )φαβ (l0 − l) (xlδ − xlδ ) = xlγ φαβ (~x l )xlδ = −2Eγαδβ (361)
l0 δ l0

which is symmetric in α, β and in δ, γ, so it has 36 independent components. Therefore

1 X X ∂ulα ∂ulβ
UHA = Eγαδβ
2 l αβδγ ∂xlγ ∂xlδ
(362)
1
Z X ∂uα (~r) ∂ulβ
= d~r Ēγαδβ
2 V αβδγ
∂xlγ ∂xδ (~r)

1
Where Ēγαδβ = Eγαδβ where v is the volume of the primitive cell. This result is similar to
v
what we wrote in the elasticity theory
1X
E= λiklm uik ulm (363)
2 iklm
 
1 ∂uγ ∂uα
If uγα = 2 ∂xα
+ ∂xγ
, then:
Z
1 X
UHA = d~r uγα λγαδβ uδβ (364)
2 V γαδβ

96
where
1 
λγαδβ = Ēγαδβ + Ēαγδβ + Ēγαβδ + Ēαγβδ (365)
4
which is symmetric for γ → α, δ → β and γα → δβ. Finally we have
1 X l
Ēγαδβ = − xγ φαβ (~xl )xlδ (366)
2v l

Then the dynamical matrix of a monatomic Bravais lattice is


!
2 X ~k~x l
Dαβ (~k) = − φαβ (~x l ) sin2 −→ k → 0 −→
M l 2
1 X
=− φαβ (~xl )(~k~x l )2 =
2M l
(367)
1 XX
=− φαβ (~x l )xlγ xlδ kγ kδ =
2M l γδ
v X
= Ēγαδβ kγ kδ
M γδ

97
14 Quantum Theory of the Vibrational Dynamics
In the quantum theory we have the Hamiltonian
2
1 X p̂lα 1 XX 0
Ĥ = + φαβ (ll0 )ûlα ûlβ (368)
2 l,α Ml 2 lα l0 β

where ûlα is the displacement operator and p̂lα is the momentum operator, which obey to the
canonical commutation relation: h i
0
ûlα , p̂lβ = i~δαβ δll0 (369)
In order to find the normal coordinates we expand the displacements and the momenta:
1 X l
~uˆl (t) = √ ~e (λ)Q̂λ (t)
Ml λ
p X (370)
p~ˆl (t) = Ml ~el (λ)P̂λ (t)
λ

from which we obtain the normal coordinates (from closure relation and orthonormality)
X p
Q̂λ = elα (λ) Ml ûlα

X 1 (371)
P̂λ = elα (λ) √ p̂lα

Ml

which obey the commutation relation


h i
Q̂λ , P̂λ0 = i~δλλ0 (372)

Then the Hamiltonian can be written as


X 1 2 
Ĥ = Ĥλ , Ĥλ = P̂λ + ωλ2 Q̂2λ (373)
λ
2

We now introduce the creation and annihilation operators:


r r
ωλ 1
â+
λ = Q̂λ − i P̂λ
2~ 2~ωλ
r r (374)
ωλ 1
âλ = Q̂λ + i P̂λ
2~ 2~ωλ
which obey
âλ , â+
 
λ0 = δλλ0
(375)
[âλ , âλ0 ] = â+ +
 
λ , âλ0 = 0

This is the formalism of a bosonic field.


Therefore we can write the Hamiltonian in terms of these new operators:
 
+ 1
Ĥλ = ~ωλ âλ âλ + (376)
2

98
where â+
λ âλ = n̂λ is the population operator. We can also rewrite the normal coordinates as
r
~
â+

Q̂λ = λ + âλ
2ωλ | {z }
≡Âλ ≡field operator (377)
r
~ωλ + 
P̂λ = i âλ − âλ
2
Since in our system we have N atoms, we get 3N normal modes in the harmonic approxima-
tion. We suppose that each mode is independent, therefore we have an ideal gas of elementary
excitations, called Bose-Einstein quasi-particles. At temperature T, the population of the mode
λ is given by the Bose-Einstein population factor:
1
nλ (T ) = ~ωλ (378)
e KB T
−1
Then, the expectation value of the Hamiltonian is
 
X 1
hĤi = hĤλ i , hĤλ i = ~ωλ nλ (T ) + (379)
λ
2

The state of the quasi-particles is



|ψn i = ψ{n1 ,...,nλ ,... } (380)

where ni is the quasi-particle with λ = i.


In quantum mechanics the time evolution of an operator is given by

∂ Ô(t) h i
i~ = Ô(t), Ĥ (381)
∂t

The value at time t = 0 of the normal coordinate Q̂λ is


~
hQ̂2λ (0)i = h(â+ +
λ + âλ )(âλ + âλ )i =
2ωλ
~
hâ+ + + +

= λ (0)âλ (0)i + hâλ (0)âλ (0)i + hâλ (0)âλ (0)i + hâλ (0)âλ (0)i = (382)
2ωλ
~ωλ (hnλ i + 12 )
=
ωλ2

Where hâ+ +
λ (0)âλ (0)i is a correlation function, since it is the average of a product. We now will
give some theorems that are useful for our treatment

I. Theorem 1: The population operator n̂λ is time independent. Indeed


h i
n̂λ , Ĥλ = 0 (383)

II. Theorem 2: The creation and annihilation operators are time dependent

â+ +
λ (t) = âλ (0)e
iωλ t
(384)
âλ (t) = âλ (0)e−iωλ t

99
To prove it we expand the commutator
h i
â+ + + + + + +
    + 
λ , Ĥλ = ~ωλ âλ âλ âλ − âλ âλ âλ = → âλ , âλ = 1 → = âλ âλ , âλ ~ωλ ⇒

∂â+ (385)
⇒ i~ λ = −~ωλ â+ λ ⇒
∂t
⇒ â+ +
λ (t) = âλ (0)e
iωλ t

and analogously can be proven for the annihilation operator âλ .


III. Theorem 3: The cross correlation functions are given by
(
hâ+
λ (t)âλ0 (0)i = hnλ ie
iωλ t
δλλ0
+
(386)
hâλ (t)âλ0 (0)i = (hnλ i + 1) e−iωλ t δλλ0

The proof is:


hâ+ +
λ (t)âλ0 (0)i = hâλ (0)âλ0 (0)ie
iωλt

 
−β Ĥ + 0
Tr e âλ âλ P −βEn + 0
+ n hψn |e âλ âλ |ψn i
→ hâλ (0)âλ0 (0)i =   = P −βEn
Tr e−β Ĥ n hψn |e |ψn i
(387)

hψ

P
n |ψn0 i = δnn0

 n |ψn i hψn | = 1
hψn | e−β Ĥ = e−βEn hψn |

We start by considering the numerator, using the notation used previously: n = {n1 , n2 , . . . , nλ , . . . }
with ni equal to the number of quasi-particles with λ = i.
X X
e−βEn hψn |â+ λ |ψn0 i hψn0 |âλ0 |ψn i →
n n 0 (388)

+
p 0
→ âλ ψ{n01 ,...,n0λ ,... } = nλ + 1 ψ{n01 ,...,n0λ +1,... }

i.e. we increased n0λ by one. Thus the matrix element hψn |â+
λ |ψn0 i is non null if and only
if
n1 = n01
n2 = n02
...
nλ = n0λ + 1
nλ+1 = n0λ+1
...
Repeating this same procedure with the annihilation operator we obtain the destruction
of one quasi particle:
p
âλ0 |ψn i = n0λ ψ{n1 ,..., nλ0 −1,... }
(
nΛ = n0Λ , ∀Λ 6= λ0 (389)
⇒ hψn0 |âλ0 |ψn i =
6 0 if and only if 0
n λ0 − 1 = n λ0

The two conditions so obtained can be satisfied at the same time if and only if λ0 = λ.
Thus we have
hâ+
λ âλ0 i = δλλ0 hn̂λ i (390)

100
1
where hn̂λ i = comes from a direct evaluation of the series. Then, for the other
eβ~ωλ
−1
equation of the theorem, we have

hâλ â+ +
λ0 i = hâλ0 âλ i + δλλ0 (391)

Lastly it suffices to add the time dependence to the results obtained to prove the theorem.

IV. Theorem 4: The autocorrelation functions are identically null:

hâ+ +
λ0 âλ i = hâλ0 âλ i = 0 (392)

i.e. it is impossible to have a non null matrix element since the same operator is acting
twice.

V. Theorem 5: The field operator correlation function is given by:

 = â+ + iωλ t
+ (nλ (T ) + 1)e−iωλ t

λ + âλ ⇒ hÂλ (t)Âλ0 (0)i = δλλ0 nλ (T )e (393)

which is proven through the previous theorems.

Therefore, using these theorems we have

~ ~ ~ωλ (nλ (T ) + 21 )
hQ2λ (0)i = hÂ+ (0)Â λ i = (2n λ (T ) + 1) = (394)
2ωλ λ 2ωλ ωλ2

~
Where is the amplitude of the quantum oscillator. In the classic limit
2ωλ
T  1 ⇒ KB T  ~ωλ

we have
1 KB T
nλ (T ) ' =
β~ωλ ~ωλ
(395)
~ KB T KB T
⇒ hQ2λ i ' =
ωλ ~ωλ ωλ2

14.1 Crystals in the Quantum Theory


Essentially the formalism is the same, but we replace λ with ~k and branch index j. The full
process is more tedious since the eigenvectors are complex:
1 X ~ l
~u lχ (t) = p ~e l (~kj)eik~x Q~kj (t). (396)
N Mχ ~
kj

The eigenvectors satisfy orthonormality, closure and reality condition:


(
~e ∗ (~kj) = ~e(−~kj)
⇒ reality condition (397)
Q~kj (t) = Q∗−~kj (t)

Then the Hamiltonian can be written as


X1 
H= Q̇~kj Q̇~∗kj + ωj2 (~k)Q~kj Q~∗kj (398)
2
~kj

101
We have also the crystallographic sum:
X ~ l
eik~x = N ∆(~k) −→ lattice sum
l
X (399)
∆(~k) = δ~k,0 + δ~k~τh
h

i.e. ∆(~k) is diverse from zero only when ~k is a reciprocal lattice vector.
Then we can write the Hamiltonian as the sum of 3N harmonic oscillators (if the conditions
stated above are satisfied). We introduce the quantization

ulχ lχ
α (t) → ûα (t)
(400)
plχ lχ
α (t) → p̂α (t)

together with the canonical commutation relation


h i
l 0 χ0
ûlχ
α , p̂ β = δαβ δll0 δχχ0 i~ (401)

Then the normal coordinates are


1 Xp ~ l
Q̂~kj (t) = √ M χeχα (~kj)eik~x ûlχ
α (t)
N lχα
(402)
1 X eχα (~kj) −i~k~xl lχ
P̂~kj (t) = √ p e p̂α (t)
N lχα Mχ

which obey the following commutation rules


h i
Q̂~∗kj , P̂k~0 j 0 = i~δjj 0 ∆(~k − k~0 )
h i (403)
Q̂~kj , P̂k~∗0 j 0 = i~δjj 0 ∆(~k − k~0 )

The ∆(~k−k~0 ) condition states that the wave vectors are either equal or they differ by a reciprocal
lattice vector.
Then the Hamiltonian, with the canonical quantization for crystals, is
1 X ∗ 2 ~ ∗

H= P̂~kj P̂~kj + ωj (k)Q̂~kj Q̂~kj (404)
2
~kj

In the classical case, from the Hamilton equations:

Q̈~kj (t) = −ωj2 (~k)Q~kj (t)


~ ~
⇒ Q~kj (t) = C1 eiωj (k)t + C2 e−iωj (k)t (405)
 
~ ~
⇒ P~kj (t) = iωj (~k) C1 eiωj (k)t − C2 e−iωj (k)t

where the constants C1 and C2 are determined from the initial conditions:
!
1 P~kj (0)
C1 = Q~kj (0) − i
Q~kj (0) = C1 + C2 2 ωj (~k)
⇒ (406)
P~ (0) = iωj (~k)(C1 − C2 )
!
kj 1 P~kj (0)
C2 = Q~kj (0) + i
2 ωj (~k)

102
Going back to the quantum case
X 1 
Ĥ = H~kj , H~kj = ∗
P̂~kj P̂~kj + ωj2 (~k)Q̂~kj Q̂~∗kj (407)
2
~kj

Taking inspiration from the solution of the general case, we can define the creation and anni-
hilation operators in order to use the second quantization formalism
s !
~ 1 P̂~kj (0)
â+
−~kj
= Q̂~kj (0) − i (408)
~
2ωj (k) 2 ωj (~k)

where the reality condition of the coordinates has been used to write ~k instead of −~k. Taking
the complex conjugate of this expression we get
s ! s !
~ 1 P̂−~kj (0) ~ 1 P̂~kj (0)
â−~kj = Q̂−~kj (0) + i ⇒ â~kj = Q̂~kj (0) + i (409)
2ωj∗ (~k) 2 ωj∗ (~k) 2ωj (~k) 2 ωj (~k)

Thus inversion of these expressions leads us to the normal coordinates


s
~  + 
Q̂~kj = â ~ + â~kj
2ωj (~k) −kj
s (410)
~ωj (~k)  + 
P̂~kj = i â−~kj − â~kj
2
These equations are true as functions of time since the creation and annihilation operators have
a time evolution:
~
â~+
kj
(t) = â~+
kj
(0)eiωj (k)t
~
(411)
â~kj (t) = â~kj (0)e−iωj (k)t

From now on the treatment is exactly the same as when we had λ instead of ~kj, starting
from the commutation relations:
h i
â~kj , â+k~0 j 0
= δjj 0 ∆(~k − k~0 )
h i h i (412)
â~+
kj
, â +
k~0 j 0
= â~kj , â k~0 j 0 = 0

The difference with the general case is that the Bose-Einstein quasi particles in this case
are plane waves called ”phonons”. These phonons have:

Energy of the phonon: E~kj = ~ωj (~k)


Polarization: ~~kj = {~e χ (~kj)}
Quasi-momentum: p~~kj = ~~k → conserved, except for a factor of
reciprocal lattice vector
Thus  
X
~ + 1
Ĥ = ~ωj (k) â~kj â~kj + (413)
2
~kj

and the average energy is  


X 1
hĤi = ~ωj (~k) n(~kj, T ) + (414)
2
~kj

103
14.2 Specific Heat of a Crystal
We now can see the specific heat in the general case:

U = Ueq + hĤi ⇒
 
1 ∂U 1 ∂  X (415)
⇒CV = = ~ωj (~k)n(~kj, T )
V ∂T V ∂T
~kj

thus we have to perform the derivative of the Bose population factor. We now consider the
density of normal modes Q(ωj (~k)):

1 X
CV = Q(ωj (~k))
V
~kj
(416)
~ ∂  ~ ~

Q(ωj (k)) = ~ωj (k)n(kj, T )
∂T

since Q(ωj (~k)) is a function only of the frequency, we can replace the sum with an integral. We
introduce the density of states
g(ω)dω (417)
which is the number of states in the range ω, ω + dω divided by the crystal’s volume. We also
(2π)3
need to remember that every k point has a volume . Thus
V

1 X X V
Z X Z Q(ωj (~k))
CV = ~
Q(ωj (k)) −→ → ~
dk −→ d~k =
V (2π)3 (2π) 3
~kj k j
(418)
Z X δ(ω − ωj (~k))
Z
= g(ω)Q(ω)dω , g(ω) = 3
d~k
j
(2π)
P
In the general case of a disordered solid we would have g(ω) = λ δ(ω − ωλ ). The density of
states must be normalized:
Z∞
3rN
g(ω)dω = (419)
V
0

where rN is the number of atoms in the crystal.


Sometimes it is better to write the density of states as follows:
XZ 1 1
g(ω) = dσ (420)
j
(2π) |∇ωj (~k)|
3

where the integral is performed over the surface of the first Brillouin zone, where ω = ωj (~k).
Thus we have (
X Z d~k 1, if ω < ωj (~k) < ω + dω
g(ω)dω = 3
(421)
j
(2π) 0, otherwise
Now we consider the distance of an element dσ of S(ω) from S(ω + dω), with ω and ω + dω
in the first Brillouin zone (see figure 14): δk(~k) is the infinitesimal distance between S(ω) and

104
Figure 14 Figure 15

S(ω + dω) measured along a direction orthogonal to S(ω) in the point ~k. Then
(
X Z d~k 1, if ω < ωj (~k) < ω + dω
=
j
(2π)3 0, otherwise
XZ dσδk(~k)
= 3
→ dω = |∇ωj (~k)|δk → (422)
j S(ωj (~
k)=ω (2π)
XZ dσ 1
→ g(ω) =
S(ω=ωj (~k) (2π) |∇ωj (~
3
j k)|

The group velocity of the phonons is

∇ωj (~k) (423)

Since we are in a crystal, there must be points in k-space where the group velocity vanishes
due to the periodicity of the dispersion relation. This fact leads to divergences in g(ω) called
Van Hove singularities, as can be seen in figure 15. Typically the first peak corresponds to the
boundary of the transverse acoustic mode.
With this formalism we can directly go to the Debye or Einstein models: Debye approximates
to only acoustic phonons (see figure 16). Thus instead of the first Brillouin zone we have a
sphere of radius ωD = kD vD in the reciprocal space. The speed of sound considered is
3 1 2
3
= 3 + 3 (424)
vD vL vT
Then, the density of states is
 2
Z ~
dk 3
Z kD  3 ω , ω < ωD
gD (ω) = 3 3
δ(ω − vD k) = 2 k 2 δ(ω − vD k)dk = 2π 2 vD
2
(425)
(2π) 2π 0
0 , ω > ωD

which is normalized through Z ωD


3N
g(ω)dω = . (426)
0 V

Figure 16

105
Figure 17

For the Einstein model we have

d~k
Z
N
gE (ω) = 3
δ(ω − ωE ) = δ(ω − ωE ) (427)
(2π) V

where ωE is constant. Therefore it is approximating an optic mode. A paragon between the


two models can be seen in figure 17.

106
Figure 18

15 Measuring the Phonon Dispersion Relations: Inelas-


tic Scattering
In order to probe crystal excitations we need inelastic scattering. We have

∆E = Ei − Ef
(428)
~ = ~ki − ~kf
Q |~ki | =
6 |~kf |

The typical energy of phonons is 10 ÷ 100 meV at the boundary of the first Brillouin zone.
The historical probes in this case are neutrons, but now also x-rays are used.

I. Neutrons scattering
~2 k 2
En = (429)
2Mn
For thermal neutrons En ' 25 meV , so that the wavevector at the boundary of the
−1
Brillouin zone is k ' 1Å . Since the energy of thermal neutrons is quite similar to the
energy of the phonons, it is relatively easy to measure the phonons’ dispersion relations
with them.

II. x-ray scattering


hc 12.4 KeV
E= ' (430)
λ λ(Å)
So for λ = 1Åwe have an energy of E ' 12.4 KeV . Ashcroft-Mermin say that this
kind of experiment is not feasible, but with time the technology advanced such that it
made possible to perform such experiments. The difficulty of such experiments is the
high resolution needed: if we suppose to want to resolve the phonons dispersion up to
∆E = 1 meV and we have x-rays of energy E = 20 KeV we need a resolution

∆E 10−6
' = 5 × 10−8 . (431)
E 20

With elastic scattering we do not excite any phonon, but we see Bragg peaks, which give
us informations on the structure of the crystal. While in the case of one phonon scattering we
have:
∆E = Ei − Ef = ±~ωj (~k)
(432)
Q~ = ~k + ~τ

where ~k is the momentum of the phonon and ~τ is a wavevector of the reciprocal lattice which
comes out since neutrons typically probe higher Brillouin zones than the first.
This type of scattering can happen through two processes:

I. Stokes process:
Ef < Ei ⇒ ∆E > 0 (433)

107
then the system acquires energy from the beam, which excites a phonon. This process is
also possible at 0K since its probability is proportional to
!
0 , T = 0K
 ~kj,
n( T :

) + 1 × phonon energy (434)

Figure 19

∆E = ~ωj (~k)
(435)
~ = ~ki − ~kf = ~k + ~τ
Q

II. Antistokes process


Ef > Ei ⇒ ∆E < 0 (436)
in this process a phonon is destructed and absorbed by the beam. It is impossible at the
absolute zero.

Figure 20

∆E = −~ωj (~k)
(437)
~ = ~ki − ~kf = −~k + ~τ
Q

Looking at the intensity profile

Figure 21

We see that the peaks are proportional to the Bose-Einstein population factor. Lowering
the temperature it becomes harder to perform such experiments since the peaks are less visible.
Taking a look at the kinematics of the process we have
 2 !
Q~ = ~ki − k~f ⇒ Q2 = ki2 + kf2 − 2~ki~kf = ki2 1 + kf − 2 cos(2θ)
kf
(438)
ki ki

108
Figure 22

Figure 23

~2 k 2
Then for neutron scattering, E =
2Mn
 
∆E
∆E = Ei − Ef ⇒ Ef = Ei − ∆E = Ei 1 −
Ei
s   r (439)
Q ∆E ∆E
= 1+ 1− − 2 cos(2θ) 1 −
ki Ei Ei

so we have a dependence on both θ and ∆E


Q
While in the case of x-ray scattering, since ∆E  Ei we can neglect the terms in ∆E in
ki
so that |~ki | ' |~kf |. Then we have  

Q = 2ki sin (440)
2
which depends only on the angle θ.
As can be seen in figure 23, different kinds of experiments probe different zones of the
Brillouin zone: Raman and Brillouin scattering probe the center of the zone (optical and
acoustic branches, respectively), while if we want to follow the dispersion curve we either use
neutrons or x-rays.

15.1 Experimental set-ups


The Brillouin spectrometer uses a laser with wavelength around the visible λ = 250 nm ÷ 1 µm,
which has frequencies of tens of GHz. The scattered light is collected through a Fabry-Perot
interferometer which allows only a narrow frequency band to pass and go to the analyser. In
Raman spectroscopy the experimental apparatus is almost the same, with the only change

Figure 24

109
that instead of a Fabry-Perot interferometer there is a diffraction grating, which has a wider
frequency window.
The maximum value if the exchanged momentum that we can see is

Qmax = . (441)
λ
With crystals the Brillouin zone can be sampled with neutrons looking at the Bragg peaks
that result in higher order Brillouin zones, while with disordered systems there are regions
unaccessible to current techniques, since there is no translational symmetry.

Applications of Inelastic X-Ray Scattering


see lecture slides

110
16 Thermal Expansion
Before starting with the theory of thermal expansion we take a look to the specific heat at low
temperature

16.1 Specific Heat at Low Temperature


The specific heat for an harmonic crystal, as we already saw, is given by
 
1 ∂U 1 ∂ X   X
CV = = ~ωj (~k)n ωj (~k), T = CVj (~k) (442)
V ∂T V V ∂T
~kj ~kj

Then we have
~ ∂
~ωj (~k) −eβ~ωj (k) ∂T β~ωj (~k)
CVj (~k) = 2
V

eβ~ωj (~k) − 1
!2 (443)
~
KB ~ωj (~k) eβ~ωj (k)
= 2
V KB T

β~ω (~k)
e j −1
which in the low temperature limit can be written as
!2
K B ~ω J (~
k) ~ω (~
− j
k)

CVj ' e KB T (444)


V KB T

We can write the sum over ~k as an integral


XZ d~k  j ~ 
CV = V CV (k) (445)
j
(2π)3

where the integral is performed over the first Brillouin zone, since it is where the physics of the
crystal is restricted to.
In order to obtain an expression for the specific heat in function of the temperature at low
temperatures we need to make some assumptions: we neglect the optical modes, so we have
only acoustic modes of the form ωj (~k) ' vj (k̂)k. Then
3 Z
∂ X 1  
~k)n ωj (~k), T d~k
CV ' ~ωj ( (446)
∂T j=1 (2π)3

Making the substitution


d~k = k 2 dkdΩ
(447)
x = β~ωj (~k) = β~vj (~k)k

111
we obtain
3
dΩ ∞ 2 x 1
Z Z
X ∂ 1
CV = k dk =
j=1
∂T 2π 2 4π 2 0 β ex − 1
3 Z ∞
∂ 1 (KB T )4 x3
Z
X dΩ 1
= dx −→
j=1
∂T 2π 2 ~3 4π 2 vj3 (~k) 0 ex − 1
3 Z
1 X 1 dΩ 1 (448)
−→ 3 = this is the mean of the sound velocity −→
v 3 2 ~
2π vj (k)
3
j=1
Z ∞ 3
∂ (KB T )4 3 x3 2π 2

KB T
−→ CV = dx = KB
∂T (~v)3 2π 2 0 ex − 1 3 v~
| {z }
π4
=
15
Therefore we obtained the same expression that was obtained in the Debye model, but here
the sound speed has a different form.

16.2 Thermal Expansion


This is an effect due to anharmonic terms, since in the harmonic treatment the atomic average
position does not change with the temperature.
We start from searching an equation of state P = P (T, V ) for the crystal:
F = U − TS
(  
dU = T dS − P dV ∂F ∂
⇒ P =− =− (U − T S)

dF = −SdT − P dV ∂V T ∂V (449)
T
*0
    Z  
∂S 1 ∂U 1 ∂U
= ⇒ S(T ) − 
S(0)
 = dT 0
∂T V T ∂T 0 T 0 ∂T 0 V
Then, these equations allow us to write
Z T
1 ∂U (T 0 , V )
   
∂ 0
P =− U (T, V ) − T 0
dT . (450)
∂V 0 T ∂T 0 V
For a crystal the energy can be written, in the small oscillations approximation, as
1X X  
U = Ueq + ~ωj (~k) + ~ωj (~k)n ωj (~k), T (451)
2
~kj ~kj

Thus we have Z T
1 X ∂n
T 0
~ωj (~k) 0 dT 0 (452)
0 T ∂T
~kj

using the following substitutions


~ωj (~k)
x=
KB T 0
1 KB
0
= x
T ~ωj (~k) (453)
∂ ∂
0
ndT 0 = ñ(x)dx
∂T ∂x
~ωj (~k)
a=
KB T

112
we can write
Z a
a
 Z ∞ 
X ∂ 1 X x dx
KB T x dx = KB T + =
∞ ∂x ex − 1 ex − 1 ∞ a ex − 1
~kj ~kj
 Z ∞ 
X a dx
= KB T + =
ea − 1 a ex − 1
~kj
  ∞ 
X a x

= KB T a
+ log(e − 1) − x =
e −1 a
~kj
 
X a a
= KB T − log(e − 1) + a =
ea − 1
~kj
X  ~ω (~k)  
j
 
= ~ωj (~k)n ωj (~k), T − KB T log e KB T − 1 + ~ωj (~k)
~kj
(454)
Then, inserting this result into the expression for the pressure, we get
 
  ~ω (~k)  
∂  1 X
~ ∂ X j
~
P =− Ueq + ~ωj (k) −
 KB T log e KB T
− 1 − ~ωj (k) (455)
∂V 2 ∂V
~kj ~kj

after some calculation we arrive at


 
∂  1 X X   ∂  
P =− Ueq + ~ωj (~k) − n ωj (~k), T ~ωj (~k) (456)
∂V 2 ∂V
~kj ~kj

At the absolute zero, the pressure is just the derivative of the ground state energy with re-

spect to the volume. At a temperature greater than zero, the pressure gains a term: n ∂V ~ωj (~k) .
In the harmonic approximation this added term is null, since the dispersion relation does not
depend on the volume:
1 XX l 0
U = Ueq + uα φαβ (l0 − l)ulβ (457)
2 αl βl0
| {z }
monatomic lattice
supposing to have an homogeneous expansion
~x l → ~x˜ l = (1 + )~x l (458)
then the new volume is (1 + )3 times the initial volume
~ l (t) = ~x l + ~u l (t)
R
(459)
~˜ l (t) = ~x˜ l + ~u˜ l (t)
R
Then we have that the relation between the displacements in the new configuration and those
in the original crystal structure is
~u l = ~x˜ l − ~x l + ~u˜ l (t) = ~x l + ~u˜ l (t) (460)
Then the energy can be written as
XX 0
 0 0

xlα + ũlα φαβ (xl − xl ) xlβ + ũlβ =

U = Ueq +
αl βl0
1 1 XX l (461)
0 0
XX
= Ueq + 2 xlα φαβ (l0 − l)xlβ + ũα φαβ (l0 − l)ũlβ
2 ll0 αβ
2 ll0 αβ

113
where the first two terms represent the new equilibrium energy, while the last term has the same
dynamics as ~uα , then it has the same solutions and therefore no dependence on the volume.
Since the pressure has no dependence on the temperature the isobaric expansion is

( ∂P )V
 
∂V
= − ∂T ∂P
 =0 (462)
∂T P ∂V T

in the harmonic approximation.


The coefficient of thermal expansion is
     
1 ∂l 1 ∂V 1 ∂P
α= = = (463)
l ∂T P 3V ∂T P 3B ∂T V
∂P
where B = −V ( ∂V )T is the bulk modulus. In the harmonic approximation α = 0.
Another anomaly of the harmonic approximation is the difference between the specific heats
at constant pressure and constant volume:
 2
T ∂P∂T V
C P = CV − ∂P
 (464)
V ∂V T

which in the harmonic approximation results


∂P

CP ∂V S
CP = CV ⇒ = ∂P
 = 1 ⇒ χT = χS (465)
CV ∂V T

i.e. the isothermal compressibility is equal to the adiabatic compressibility. In reality they are
 
1 1 ∂V
χT = =−
B V ∂P T
  (466)
1 ∂V
χS = −
V ∂P S

16.3 The Grüneisen Parameter


Thus to describe the thermal expansion we must go beyond the harmonic approximation. We
will still consider a gas of phonons and we will treat the anharmonicities as perturbations. We
start from
 
∂  1 X
~
X 
~
 ∂ 
~

P =− Ueq + ~ωj (k) −
 n ωj (k), T ~ωj (k) (467)
∂V 2 ∂V
~kj ~kj

Then !
1 X ∂  ~  ∂~ωj (~k)
α= n ωj (k), T − (468)
3B ∂T ∂V
~kj

In this expression we see a small similarity to the specific heat at low temperatures
 
X j X ~ωj (~k) ∂n ω j (~
k), T
CV (~k) = (469)
V ∂T
~kj ~kj

114
We now define:  
V ∂ωj (~k) ∂ log ωj (~k)
γ~kj ≡ − =− (470)
ωj (~k) ∂V ∂ log(V )

which is the Grüneisen parameter of the mode ~kj. The overall Grüneisen parameter is defined
as P j
~kj γ~kj CV
γ≡ P j (471)
~kj CV

It follows that
!
1 X ~ωj (~k) ∂  ~  V ∂ωj (~k) γCV
α= n ωj (k, T − = . (472)
3B V ∂T ωj (~k) ∂V 3B
~kj | {z }
j ~ | {z }
=CV (k)
=γ~kj

The reason to write the expansion coefficient in this way is that all simplest models lead
to the same Grüneisen parameter for every mode, which therefore becomes just an universal
multiplicative factor.
In the model of Debye we have
∂ log(ωD )
γ = γ~kj = − . (473)
∂ log(V )
In theories with γ constant we have that
γCV
α= ⇒ α(T ) ' CV (T ) (474)
3B
Therefore we have (
constant high T
α∼ (475)
T 3 low T
Now we want to prove that these limits are true also in the general case:
P j
~kj γ~kj CV
γ≡ P j (476)
~kj CV

if the γ~kj are not the same the Grüneisen parameter depends on the temperature. At high
temperature
KB
CVj (~k) ∼ (477)
V
then X 1
γ∼ γ~kj (478)
3N
~kj

which is constant with temperature. At low temperature, an argument similar to that of the
specific heat leads to P3 R γ~
j=1 dΩ v3kj~
j (k)
γ ∼ P3 R (479)
j=1 dΩ v31(~k)
j

which is constant in temperature. Therefore the coefficient of thermal expansion is proportional


to the specific heat (the bulk modulus has only a mild dependence on the temperature, so its
effect is small), as we found in the Debye model.

115
Figure 25

16.4 Thermal Conductivity


The results from the Drude model are
1 1
KT = vCV l = v 2 CV τ , l ≡ vτ (480)
3 3
We can use the same argument for insulators treating the phonons as localized particles. In
the harmonic approximation the mean free path is infinite, but there exist some limits on it:

I. Lattice defects;

II. Finite dimension of the crystal;

III. Anharmonicities

These anharmonicities present themselves as collisions between phonons. This results in a


peculiar temperature dependence of the thermal conductivity.

This Is The End, My Only Friend, The End, I’ll Never


Look At Your Eyes Again

116

Potrebbero piacerti anche