Lectures on Theoretical Physics

Linear Response Theory

Peter Hertel

University of Osnabr
uck, Germany

The theory of linear response to perturbations of the equilibrium state, or linear

response theory, is the subject of this series of lectures.
Ordinary matter, if left alone, will sooner or later attain an equilibrium state. This
equilibrium state depends on the temperature of the environment and on external
parameters. External parameters may be the region of space within which a certain
number of particles are confined, mechanical stress, or the strength of an external
electric or magnetic field.
If temperature or the external parameters change slowly enough, the system can
attain the new equilibrium state practically instantaneously, and we speak of a
reversible process. On the other hand, if the external parameters vary so rapidly
that the system has no chance to adapt, it remains away from equilibrium, and we
speak of irreversibility.
The most important application is optics. There is a medium which is exposed to
an electromagnetic wave. The electric field changes so rapidly that matter within
a region of micrometer dimensions cannot react instantaneously, it responds with
retardation. We shall work out the retarded response in linear approximation.
There are quite a few general and important results which hold irrespective of
a particular Hamiltonian, such as the Kramers-Kronig relations, the fluctuation-
dissipation theorem, the second law of thermodynamics, and Onsagers relation.
We discuss various electro- and magnetooptic effects, such as the Pockels effect,
the Faraday effect, the Kerr effect, and the Cotton-Mouton effect.
We also treat spatial dispersion, or optical activity and indicate how the theory is
to be developed further in order to handle the non-linear response as well.

lrt-10.12.01
2 CONTENTS

Contents
1 Maxwell equations 4
1.1 The electromagnetic field . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 Potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Field energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Polarization and magnetization . . . . . . . . . . . . . . . . . . . . 6

2 A simple model 8
2.1 Equation of motion . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2 Greens function . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.3 Susceptibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

3 Thermodynamic equilibrium 11
3.1 Observables and states . . . . . . . . . . . . . . . . . . . . . . . . 11
3.2 The first law of thermodynamics . . . . . . . . . . . . . . . . . . . 11
3.3 Entropy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.4 The second law of thermodynamics . . . . . . . . . . . . . . . . . . 12
3.5 Irreversible processes . . . . . . . . . . . . . . . . . . . . . . . . . 13

4 Perturbing the equilibrium 15

4.1 Time evolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
4.2 Interaction picture . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
4.3 Perturbing the Gibbs state . . . . . . . . . . . . . . . . . . . . . . 16
4.4 Time dependent external parameter . . . . . . . . . . . . . . . . . 17

5 Dielectric susceptibility 18
5.1 Polarization of matter . . . . . . . . . . . . . . . . . . . . . . . . . 18
5.2 Dielectric susceptibility . . . . . . . . . . . . . . . . . . . . . . . . 19
5.3 Susceptibility proper and optical activity . . . . . . . . . . . . . . . 20

6 Dispersion relations 21
6.1 Retarded Green function . . . . . . . . . . . . . . . . . . . . . . . 21
6.2 Kramers-Kronig relations . . . . . . . . . . . . . . . . . . . . . . . 23
6.3 Refraction and absorption . . . . . . . . . . . . . . . . . . . . . . . 23
6.4 Oscillator strength . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

7 Dissipation-fluctuation theorem 26
7.1 The Wiener-Khinchin theorem . . . . . . . . . . . . . . . . . . . . 26
7.2 Kubo-Martin-Schwinger formula . . . . . . . . . . . . . . . . . . . 27
7.3 Response and correlation . . . . . . . . . . . . . . . . . . . . . . . 28
7.4 The Callen-Welton theorem . . . . . . . . . . . . . . . . . . . . . . 29
7.5 Energy dissipation . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
CONTENTS 3

8 Onsager relations 32
8.1 Symmetry of static susceptibilities . . . . . . . . . . . . . . . . . . 32
8.2 Time reversal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
8.3 Onsager theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
8.4 Onsager relation for kinetic coefficients . . . . . . . . . . . . . . . . 35
8.5 Electrical conductivity and Hall effect . . . . . . . . . . . . . . . . 36

9 Electro- and magnetooptic effects 37

9.1 Crystal optics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
9.2 Pockels effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
9.3 Faraday effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
9.4 Kerr effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
9.5 Magneto-electric effect . . . . . . . . . . . . . . . . . . . . . . . . 41
9.6 Cotton-Mouton effect . . . . . . . . . . . . . . . . . . . . . . . . . 41

10 Spatial dispersion 42
10.1 Dispersion relation . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
10.2 Optical activity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

11 Non-linear response 45
11.1 Higher order response . . . . . . . . . . . . . . . . . . . . . . . . . 45
11.2 Susceptibilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
11.3 Second harmonic generation . . . . . . . . . . . . . . . . . . . . . 47

A Causal functions 49

B Crystal symmetry 51
B.1 Spontaneous symmetry breaking . . . . . . . . . . . . . . . . . . . 51
B.2 Symmetry groups . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
B.3 A case study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

C Glossary 55
4 1 MAXWELL EQUATIONS

1 Maxwell equations
In this serious of lectures we will study the interaction of a rapidly oscillating
electromagnetic field with matter. Therefore, a recollection of basic electrody-
namics seems to be appropriate.

1.1 The electromagnetic field

The electromagnetic fields E = E(t, x) and B = B(t, x) are defined by their
action on charged particles. The trajectory t x(t) of a particle with charge
q is a solution of

p = q (E + v B) , (1.1)
p
where v = x and p = mv/ 1 v 2 /c2 . The electromagnetic field is to be
evaluated at the current particle location t, x(t).
The electromagnetic fields act on charged particles, as described by the Lorentz
formula (1.1), and charged particles generate the electromagnetic field. dQ =
dV (t, x) is the amount of electric charge in a small volume element dV at x
at time t. Likewise, dI = dA j(t, x) is the charge current passing the small
area element dA from the back to to the front side.
Maxwells equations read

1 =j
0 E = and B 0 E (1.2)
0

as well as

=0 .
B = 0 and E + B (1.3)

The first group of four equations describe the effect of electric charge and cur-
rent, the second group of likewise four equations say that there is no magnetic
charge (magnetic monopoles).
It is a consequence of Maxwells equations that charge is conserved,

+ j = 0 . (1.4)

1.2 Potentials
Stationary fields decouple.

0 E = and E = 0 (1.5)

describe the electrostatic field,

1
B = j and B = 0 (1.6)
0
1.3 Field energy 5

the magnetostatic field.

The electrostatic field may be derived from a scalar potential,

E = (1.7)

which obeys Poissons equation

0 = . (1.8)

The magnetostatic field can be expressed by

B =A (1.9)

in terms of a vector potential A. Adding the gradient of an arbitrary scalar field

, A 0 = A+, does not change the induction field B. This ambiguity allows
to subject the vector potential to a gauge, e.g. the Coulomb gauge A = 0.
The three components of the vector potential the obey a Poisson equation each,

1
A = j . (1.10)
0

The full set of Maxwell equations are solved by

and B = A .
E = A (1.11)

If we now impose the Lorentz gauge

1
0 + A=0 , (1.12)
0

the following equations result:

1
0 = and A = j . (1.13)
0

The box, or wave operator is 02 where 0 is the partial derivative with

respect to time, divided by c which is defined as c = 1/ 0 0 .

1.3 Field energy

The potential of a point charge q resting at y is the Coulomb potential

q 1
C (x) = . (1.14)
40 |x y|

If charges q1 , q2 , . . . are brought from infinity to their locations at x1 , x2 , . . . the

following work has to be done:
X 1 qb qa
W = . (1.15)
40 |xb xa |
b>a
6 1 MAXWELL EQUATIONS

For a smooth charge distribution this expression may be rewritten into

Z 3 00 3 0 Z
1 d x d x (x 00 )(x 0 ) 1
W = = d3 x (x)(x) . (1.16)
80 |x 00 x 0 | 2

By partial integration we obtain

Z
0 E 2
W = d3 x . (1.17)
2

We interpret 0 E 2 /2 as the energy density of the electric field.

Using Maxwells equation we may prove the following balance equation:

+ S = (1.18)

where
0 E 2 B2 1
= + , S= E B and = j E . (1.19)
2 20 0

As explained before, 0 E 2 /2 is the energy density of the electric field. The

energy density of the magnetic field is B 2 /20 . The energy current density
S = (E B)/0 is also called Poyntings vector.
= j E describes the production of field energy per unit time per unit
volume. Field energy is created if charge runs counter to the electric field
(Bremsstrahlung). Field energy vanishes if charges run with the direction of
the electric field (Ohms law, Joules heat).

1.4 Polarization and magnetization

R R
Q = d3 x (x) is the charge of system. By p = d3 x x (x) we denote its
electric dipole moment. The dipole moment does not depend on the choice of
the coordinate system origin if its charge vanishes.
Denote by P the density of a probes dipole moments, its polarization. For
ordinary matter, which is locally neutral, this is a well defined quantity. One
can show that P is the charge density causing the polarization and that
P contributes to the current density.
We likewise define magnetic dipole moments m and their density M , the mag-
netization. M also contributes to the current density. The charge and
current density should therefore be split into

= P + f and j = P + M + j f . (1.20)

The remainders f and j f are the charge and current density of free, or mobile
charges, as opposed to bound charges.
We introduce auxiliary fields D = 0 E + P (dielectric displacement) and H =
B/0 M (magnetic field strength). They obey the following equations:

D = f and H D
= jf . (1.21)
1.4 Polarization and magnetization 7

=0
B = 0 and E + B (1.22)

remain unchanged.
8 2 A SIMPLE MODEL

2 A simple model
For warming up, we develop a very simple model of the dielectric susceptibility.
Consider an atom which is exposed to an oscillating electric field E = E(t) along
the z-direction. A particular electron of this atom will be forced to oscillate.

2.1 Equation of motion

We assume that the electrons equilibrium position is z = 0. There is an equilib-
rium restoring counter force which, for small deviations, will be proportional to
z. We also allow for friction which we assume to be proportional to the electrons
velocity z.
Therefore, the following equation of motion has to be solved:

e
z + 2 z + 20 z = a(t) where a(t) = E(t) . (2.1)
m

m and e are the electrons mass and charge, m20 is the spring constant of
the linear oscillator, 2m the friction coefficient.

2.2 Greens function

Because there is a linear relationship between position z and acceleration a we
should write
Z
z(t) = ds (t, s) a(s) , (2.2)

where is a Green function.

Because the coefficients of the differential equation (2.1) do not depend on time
t one has (t, s) = (t s). This implies
Z Z
z(t) = ds (t s) a(s) = d ( ) a(t ) . (2.3)

is the time difference between cause (accelerating force, or a) and effect (de-
viation from equilibrium position, z). Causes must be earlier than their effects,
therefore
Z t Z
z(t) = ds (t s) a(s) = d ( ) a(t ) . (2.4)
0

is an age, it cannot be negative.

We differentiate (2.4) with respect to time,
Z t
z(t)
= (0) a(t) + ds (t s) a(s) , (2.5)

and likewise for z(t).

2.3 Susceptibility 9

Inserting these expressions into (2.1) results in

+ 2 + 20 = 0 (2.6)

and

(0) = 0 and (0) = 1 . (2.7)

t
This differential equation with constant coefficients is solved by e which
implies 2 + 2 + 20 = 0, i.e. = i where
q
= 20 2 . (2.8)

We assume weak damping, /0 < 1. The two fundamental solutions are to be

superimposed such that the initial condition (2.7) are satisfied. We obtain

1
( ) = e sin . (2.9)

The original equation of motion has thus been solved for an arbitrary oscillating
electric field strength E = E(t).

2.3 Susceptibility
p = ez is the dipole moment of the electron under consideration. If there are
N of them per unit volume, the polarization is P = N p. We just have obtained
the result
Z
N e2
P (t) = d ( ) E(t ) . (2.10)
m 0

Let us Fourier decompose the electric field strength:

Z
d it
E(t) = e E() . (2.11)
2

We work out
Z
d it
P (t) = 0 e () E() (2.12)
2

where
Z
N e2 i N e2 1
() = d e ( ) = . (2.13)
m0 0 m0 ( + + i)( i)

The Fourier component P (), which is declared in analogy to (2.11), is propor-

tional to the Fourier component E() at the same angular frequency :

P () = 0 () E()
. (2.14)
10 2 A SIMPLE MODEL

= () is called the susceptibility of the material under study. It is a function

of angular frequency. Note that we may rewrite (2.13) as

N e2 1
() = 2 2
. (2.15)
m0 0 2i

(2.13) is an oversimplification because not all electrons have the same resonance
frequency 0 and the same damping constant ; in fact, one must sum over such
terms. However, (2.15) or a sum over such terms reflect the essential features
quite well:

The static value (0) is always positive and real.

() 0 with .

The imaginary part of the susceptibilty is always non-negative.

The imaginary part of the susceptibility is large close to a resonance

0 .

Dispersion is unavoidable, susceptibilities must be different for different

angular frequencies.
 11

3 Thermodynamic equilibrium
In this serious of lectures we will study the interaction of a rapidly oscillating
electromagnetic field with matter. Therefore, a recollection of basic statistical
thermodynamics seems to be appropriate.

3.1 Observables and states

We describe the system quantum-mechanically, i. e. by an appropriate Hilbert
space H. Observables M , or measurable quantities, are represented by self-
adjoint linear operators mapping the Hilbert space into itself, M : H H.
As is well known, the eigenvectors of self-adjoint operators form a complete
system of normalized and mutually orthogonal vectors. The eigenspaces of an
observable M describe the alternatives, the eigenvalues the measured values of
these alternatives.
Pure states of the system are represented by wave functions, or vectors of H.
For a large system, the notion of a pure state is an oversimplification. In
fact, there is a complete system of normalized and mutually orthogonal vectors
1 , 2 , . . . and a set of probabilities w1 , w2 , . . .. The wj are the probabilities
that the system will be found to be in the pure state j .
There is a unique linear operator W such that the j are its eigenvectors and
the wj it eigenvalues,

W j = wj j . (3.1)

This probability operator1 W describes the state of the system. It is character-

ized by

0 W 1 and tr W = 1 . (3.2)

Here 0 and 1 stand for the zero and the unity

P operator, and tr denotes the
trace. (3.2) boils down to 0 wj 1 and wj = 1.
The expectation value hM i of the observable M , while the system is in state
W , is given by
X
hM i = wj (j , M j ) = tr W M . (3.3)
j

3.2 The first law of thermodynamics

The energy observable H (the Hamiltonian) is of particular interest because it
governs the time development of the system. Its expectation value is convention-
ally denoted by U , the internal energy. It is a convention that thermodynamic
systems are at rest, hence energy is internal energy.
1
also density matrix
12 3 THERMODYNAMIC EQUILIBRIUM

The internal energy of a system may change because the state changes or be-
cause the Hamiltonian changes,

U = tr W H + tr W H . (3.4)

The first contribution Q = tr W H is heat, the second contribution A =

tr W H is work2 . The Hamiltonian usually depends on external parameters
which may change. An external electric field is a typical example. The state-
ment U = Q + A is the first law of thermodynamics: energy may be trans-
ferred to a system as heat or as work.

3.3 Entropy
A mixed state W is a mixture of pure states. The pure states j are contained
in the mixed state W with probabilities wj . Two mixed states W1 and W2 may
be further mixed to become W = 1 W1 + 2 W2 where 0 1 , 2 1 and
1 + 2 = 1. It is a simple exercise to show that 0 W 1 holds as well as
tr W = 1.
We require a measure to answer the question: how much mixed is a state. This
measure should vanish if W represents a pure state, and it should increase upon
mixing. Here we just report the result of a lengthy discussion:
X 1
S(W ) = kB wj ln = kB tr W ln W (3.5)
wj
j

is a measure of the degree of mixture. S(W ) is the entropy of the mixed state
W . The Boltzmann constant kB shows up for purely historical reasons.
If all probabilities wj equal zero, up to one, which must be unity, than the
entropy vanishes. The entropy of a pure state vanishes. It can also be shown
that

1 S(W1 ) + 2 S(W2 ) S(1 W1 + 2 W2 ) (3.6)

is true for states W1 , W2 and weights 1 , 2 . Mixing increases entropy.

3.4 The second law of thermodynamics

Assume a system which is well isolated from its environment. The internal
energy within the system remains constant. Nevertheless, the environment will
influence the time evolution of the system. The second law of thermodynamics
states that the interactions between the system and its partially chaotic envi-
ronment will increase the amount of chaos within the system which is measured
by the entropy of its state. Put otherwise, the entropy of the systems state
increases in the course of time.
2
German Arbeit. The symbol W (for work) is already in use (for the probability operator,
or state)
3.5 Irreversible processes 13

It increases until it has reached a maximum. The state with maximal entropy
does not change any more, it describes the equilibrium of the system with
respect to it environment. Let us denote this equilibrium state by G. Denote
by

S = {W : H H | 0 W 1, tr W = 1} (3.7)

the set of states. The equilibrium state, or Gibbs state, is characterized by

S(G) = max S(W ) , (3.8)

W S

which amounts to

S(G + W ) = S(G) for tr W = 0 and tr W H = 0 (3.9)

for small deviations W . G + W must be a state, and U = tr (G + W )H

should remain constant.
The solution of this problem reads

(F H)/kB T
G= e , (3.10)

where F and T are Lagrange multipliers. F , the free energy, is a number. H is

the Hamiltonian of the system, and T the temperature of the Gibbs state.
The free energy is calculated according to

H/kB T
F = kB T ln tr e (3.11)

such that tr G = 1 holds. The temperature T is determined by solving

H/kB T
tr H e
U= (3.12)
H/kB T
tr e
for T . One can show that the right hand side of (3.10) increases with the
temperature T which guarantees a unique solution. Adding energy to a system,
by heat or work, will make it warmer.

3.5 Irreversible processes

We denote the external parameters summarily by , i. e. H = H(). The
free energy depends on the temperature and on the external parameters, F =
F (T, ). One can easily show that
X
dF = SdT Vr dr (3.13)

holds true where

S = S(G) (3.14)
14 3 THERMODYNAMIC EQUILIBRIUM

is the entropy of the Gibbs state. The generalized forces Vr are given by

H()
Vr = tr G . (3.15)
r

We speak of a process if the temperature T = Tt of the systems environment

and the external parameters r = r,t change in the course of time. The
process is reversible provided the state Wt of the system is always very close to
the corresponding equilibrium state,

Wt ' G(Tt , t ) . (3.16)

If the external parameters change too rapidly, the system will remain away
from equilibrium, and we speak of irreversible processes. With light, the typical
spatial dimensions of material points are micrometers, and the corresponding
period is 3 1014 s. This time is much too short for achieving equilibrium. The
interaction of light with matter is an irreversible process. We will describe in
the following sections how to cope with this problem.
 15

4 Perturbing the equilibrium

In this section we will describe how the state of a system changes with time.
We concentrate on small perturbations of the equilibrium state.

4.1 Time evolution

So far we have spoken of preparing a state W and immediately measuring an
observable M . Let us now discuss the situation that we wait for a certain time
t between preparing W and measuring M .
Let us first discuss the Heisenberg picture of time evolution. Waiting the time
span t and then measuring M is a new observable Mt . Mt must be self-adjoint.
Its eigenvalues, the possible outcomes of a measurement, are independent on
the waiting time span t. Therefore,

Mt = Ut M Ut (4.1)

holds where Ut is a unitary operator.

Because of

Ut1 +t2 = Ut2 Ut1 (4.2)

we conclude that U depends exponentially on t,

~i tH
Ut = e (4.3)

where H is a self-adjoint operator, the Hamiltonian, or the energy. Note that

(4.3) results in

d i
Mt = [ H , Mt ] , (4.4)
dt ~
the Heisenberg equation of motion. [A, B] = AB BA is the commutator of A
with B.
Another aspect is the Schr odinger picture. Conceptually, preparing W and
waiting a certain time t is preparing a state Wt . Because of

tr W Mt = tr W Ut M Ut = tr Ut W Ut M (4.5)

we have to write

Wt = Ut W Ut . (4.6)

We have made use of tr AB = tr BA.

(4.6) results in the Schr
odinger equation for mixed states, namely

d i
Wt = [ Wt , H ] . (4.7)
dt ~
16 4 PERTURBING THE EQUILIBRIUM

4.2 Interaction picture

Very often the Hamiltonian may be split into a manageable, time independent
part H and a small perturbation V = Vt which may depend on time. Just think
of H as the Hamiltonian of matter and Vt describing the perturbation by the
electric field of a light wave.
It is then useful to work in the Heisenberg picture with respect to H only.
In the Schr odinger picture we have to solve
d i
W t = [ W t , H + Vt ] . (4.8)
dt ~
With
A(t) = Ut AUt (4.9)
where the time translation operator Ut is defined by (4.3) we arrive at

d i i
Wt (t) = Ut [ H , Wt ] + [ W t , H + V t ] U t (4.10)
dt ~ ~
or
d i
Wt (t) = [ Wt (t) , Vt (t) ] . (4.11)
dt ~
This equation of motion is driven by the perturbation only. If there were no
V = Vt , then Wt (t) would be constant. It is to be expected that a small
perturbation will cause the state Wt (t) to change only slowly.

4.3 Perturbing the Gibbs state

Assume that we perturb the Gibbs state:
(F H)/kB T
Wt G = e and Vt 0 for t . (4.12)
Because the Gibbs state is stationary with respect to H, i. e. G(t) = G, we
may also write
Wt (t) G for t . (4.13)
By combining the initial condition (4.13) with (4.10) we arrive at the following
integral equation:
Z t
i
Wt (t) = G + ds [ Ws (s) , Vs (s) ] . (4.14)
~
This integral equation gives rise to a perturbation expansion with respect to V .
In lowest order we have to set Wt (t) = G + . . .. In first order we arrive at
Z t
i
Wt (t) = G + ds [ G , Vs (s) ] + . . . . (4.15)
~
Inserting the first order expression (4.15) into (4.14) will result in the second
order approximation, and so forth.
4.4 Time dependent external parameter 17

4.4 Time dependent external parameter

We now specialize to
X
Vt = r (t) Vr . (4.16)
r

r = r (t) is an external parameter, Vr the corresponding generalized force,

an observable. We may rewrite the linear response (4.15) of the system to a
perturbation of the equilibrium state into the following form:
Z t X i
Wt (t) = G ds r (s) [ G , Vr (s) ] + . . . , (4.17)
r
~

or
Z t X i
Wt = G ds r (s) [ G , Vr (s t) ] + . . . . (4.18)
r
~

The expectation value of an observable M is

Z X
tr Wt M = tr G M + d r (t ) (M Vr ; ) + . . . (4.19)
0 r

where
i
(M Vr ; ) = tr G [ M ( ) , Vr ] . (4.20)
~

This is a remarkable result. The expectation value of an observable M is its

equilibrium expectation value plus an additional contribution. The addition
depends on past perturbations only. It is proportional to the external param-
eters r . The Green function = (M Vr ; ) depends on the age of the
perturbation and is linear in the perturbing generalized forces Vr . In fact, the
commutator [M (t), Vr (s)] will vanish unless Vr (s) and M (t) affect each other.
What is most important: the Green function is an expectation value of the
unperturbed state G. Loosely speaking, the Gibbs state G knows already how
it will react on perturbations.
18 5 DIELECTRIC SUSCEPTIBILITY

5 Dielectric susceptibility
We now specialize to ordinary matter in an oscillating electric field. We may
write
X p2 1 X qb qa
a
H= + (5.1)
a
2ma 40 |xb xa |
b>a

for the Hamiltonian of matter. Charged particles, nuclei and electrons, interact
via Coulomb forces. a, b . . . label the particles, ma and qa are the respective
masses and charges. xa and pa denote the particle positions and momenta,
they are observables. This is no treatise on many particle physics, and (5.1)
serves only to introduce notation. In particular, spin and relativistic effects are
missing.

5.1 Polarization of matter

If there is an electromagnetic field, the interaction with matter can be classified
by a multi-pole expansion. For our purpose, the electric dipole approximation
is sufficient.
We define by
X
P (x) = qa xa 3 (x xa ) (5.2)
a

the density of dipole moments, or polarization as a field of observables. The

interaction of the external electromagnetic field with matter is described in
electric dipole approximation by
Z
Vt = d3 y E(t, y) P (y) . (5.3)

We assume a dielectric material, in contrast with a ferroelectric material. With-

out electric field there is no polarization, tr GP (x) = 0. By inserting (5.3) into
(4.19) and (4.20) we arrive at the following expression for the time-dependent
polarization:
Z Z
Pi (t, x) = tr Wt Pi (x) = d d3 y ij (, x, y) Ej (t , y) . (5.4)
0

Here and later we rely on Einsteins summation convention: a sum over the
doubly occurring index j is silently understood.
The tensor of Green functions3 is given by

i
ij (, x, y) = tr G [ Pi (, x) , Pj (0, y) ] . (5.5)
~
3
ij (, x, y) = (Pi (x)Pj (y); ) in the notation of section 4
5.2 Dielectric susceptibility 19

Recall the definition of the polarization Pi (x) and its time translation

~i tH
Pi (t, x) = Ut Pi (x) Ut where Ut = e . (5.6)

Also recall the definition of the equilibrium, or Gibbs state

(F H)/kB T
G= e . (5.7)

H in (5.6) and (5.7) is the Hamiltonian (5.1) of unperturbed matter.

We assume that the Gibbs state is not only invariant with respect to time
translations, but also with respect to spatial translations. One may then write

i
ij (, ) = tr G [ Pi (, ) , Pj (0, 0) ] (5.8)
~

instead of (5.5) and

Z Z
Pi (t, x) = tr Wt Pi (x) = d d3 ij (, ) Ej (t , x ) (5.9)
0

instead of (5.4).

5.2 Dielectric susceptibility

Let us Fourier decompose the external electric field:
Z Z
d it d3 q iq x
Ei (t, x) = e e Ei (, q) . (5.10)
2 (2)3

We may likewise decompose the polarisation4 Pi (t, x).

The well-know convolution theorem of Fourier theory allows to write

j (, q)
Pi (, q) = 0 ij (, q) E (5.11)

where
Z Z
1 i iq
ij (, q) = d e d3 e ij (, ) (5.12)
0 0

is the tensor of dielectric susceptibility. It is a simple exercise to verify that the

susceptibilities are dimensionless. If we could solve all equations which we may
write down, the susceptibility of any material can be calculated. It depends
on frequency, on the wave vector, on temperature, and on all other parameters
which characterize the equilibrium state.
4
We use the same word for the observable Pi = Pi (x) and its expectation value in the time
dependent state Wt , i.e. Pi (t, x) = tr Wt Pi (x).
20 5 DIELECTRIC SUSCEPTIBILITY

5.3 Susceptibility proper and optical activity

Since the velocity of sound, which sets the time to distance scale in solids, is
so much smaller than the velocity of light, the susceptibility varies only weakly
with the wave number. We may expand

ij (, q) = ij () + ijk () qk + . . . . (5.13)

In most optical applications only ij () is required. The frequency dependent

tensor ij () is likewise called the susceptibility of the material in question.
The next term describes optical activity. Higher order expansion coefficients
are seldom encountered. Optical activity is deferred to a later section. In
the following we are concerned entirely with the ordinary frequency dependent
susceptibility tensor ij (),
Z Z
1 i i
ij () = d e d3 tr G [ Pi (, ) , Pj (0, 0) ] . (5.14)
0 0 ~
 21

6 Dispersion relations
We will now exploit the fact that the response to a perturbation is retarded.
The generalized susceptibility is given by
Z
i
(AB; ) = d e (AB; ) , (6.1)
0

where the response function is

i
(AB; ) = tr G [ A( ) , B(0) ] . (6.2)
~
It weighs the impact of a perturbation caused by B on A after a delay .
See appendix A for a summary on causal functions.

6.1 Retarded Green function

In fact, the generalized susceptibility is the Fourier transform of a product,
Z
i
(AB; ) = d e ( ) (AB; ) , (6.3)

where = (x) is the Heaviside5 jump function, a distribution.

The Fourier transform of a product is the convolution of the respective Fourier
transforms,
Z
du
(AB; ) = ( u) (AB; u) . (6.4)
2

The Fourier transform of the Heaviside function is

Z
(i )t i
() = dt e = . (6.5)
0 + i

is a positive, yet arbitrarily small number. Therefore,

Z
1 du
(AB; ) = (AB; u) . (6.6)
2i u i

(6.6) says that the pole at u = is to be avoided by running in the lower

complex plane.
Now, the Gibbs state is stationary, therefore

i i
h [ A( ) , B(0) ]i = h [ A(0) , B( ) ]i (6.7)
~ ~
holds true which implies

(AB; ) = (BA; ) (6.8)

5
(x) = 0 for x < 0 and (x) = 1 for x > 0
22 6 DISPERSION RELATIONS

or

(AB; ) = (BA; ) = (BA; ) . (6.9)

The last conclusion relies on the fact that response functions (AB; ) are real.
With this information we may work out

Z
1 du
(BA; ) = (AB, u) . (6.10)
2i u + i

Note the similarity with (6.6). The only difference is how to evade the singu-
larity u = . The difference of (6.6) and (6.10) is a ring integral around the
singularity, it results in

(AB; ) (BA; ) = (AB; ) . (6.11)

We define the refractive part of the susceptibility by

(AB; ) + (BA; )
0 (AB; ) = (6.12)
2

and the absorptive part by

(AB; ) (BA; )
00 (AB; ) = . (6.13)
2i

Both are Hermitian in the sense that interchanging A and B as well as complex
conjugating it leave the expressions unchanged.
Inserting (AB; ) = 2i 00 (AB; ) into the sum of (6.6) and (6.10) we arrive
at
Z
0 1 du
(AB; ) = Pr 00 (AB; u) . (6.14)
u

The principal value of the integral is the mean of avoiding the singularity via
the upper and the lower complex u-plane. It may also be defined by

Z Z Z
du du
Pr f (u) = + f (u) . (6.15)
u + u

(6.15) is the prototype of a dispersion relation. In the following subsection we

will specialize to optics thereby justifying the terminology: refractive, absorp-
tive, dispersion.
6.2 Kramers-Kronig relations 23

6.2 Kramers-Kronig relations

Recall the definition of the dielectric susceptibility tensor
Z Z
1 i i
ij () = d e d3 tr G [ Pi (, ) , Pj (0, 0) ] . (6.16)
0 0 ~

It fits into the general scheme,

Z
ij () = d3 (Pi ()Pj (0); ) . (6.17)

Response functions (AB; ) and their susceptibilities (AB; ) are linear in

both arguments, A and B, and the same can be said of the dispersion relation
(6.16). It can be summed over A and remains true.
With

ij () + ji ()
ij0 () = (6.18)
2

and

ij () ji ()
ij00 () = (6.19)
2i

we may write
Z
1 du
ij0 () = Pr 00 (u) . (6.20)
u ij

This particular form of a dispersion relation is known as the Kramers-Kronig

relation. It first of all explains why refraction depends on frequency. A beam
of natural light consists of many colors, and a glass prism will lead to disper-
sion. The Kramers-Kronig relation also tells that there is no refraction without
absorption, although at different frequencies. It remains to show that the ab-
sorptive part ij00 () is a non-negative Hermitian tensor.

6.3 Refraction and absorption

We discuss Maxwells equations for purely periodic fields in absence of charges
and currents:

H = i0 E and E = i0 H . (6.21)

The remaining divergence equations are automatically fulfilled.

The permittivity tensor of (6.21) is

ij () = ij + ij () . (6.22)
24 6 DISPERSION RELATIONS

We simplify the discussion and discuss an isotropic material, ij () = ij ().

We intend to show that the real part of the susceptibility is responsible for
refraction while the imaginary part causes absorption.
Consider a damped plane wave traveling in z direction:

it ink0 z z/2
E1 = E e e e ; E 2 = 0 ; E3 = 0 . (6.23)

The wave number is q = nk0 where k0 = /c. n is the refractive index, the
absorption constant. Inserting (6.23) into (6.21) shows that we have guessed
correctly, provided

0 00 2 n
= + i = n + i = n2 + i + ... (6.24)
2k0 k0

holds true. Thus we have shown that

n = 0 (6.25)

is the refractive index and

k0 00
= (6.26)
n
the absorption constant. Recall that the energy current density S is quadratic
in the electromagnetic field strengths. Therefore, is indeed the power decay
constant of radiation. Proving 00 0 is close to proving the second law of
thermodynamics. We will address this challenge in the following section.
The susceptibility is the Fourier transform of a real function, therefore () =
(). Consequently, the real part is an even, the imaginary part an odd
function of angular frequency. Averaging 0 () and 0 () and taking into
account that u 00 (u)/(u2 2 ) is an even function of u we arrive at
Z
2 u 00 (u)
0 () = Pr du . (6.27)
0 u2 2

6.4 Oscillator strength

We have derived in section 2 an expression for the susceptibility of elastically
bound, weakly damped electrons. If there are N such electrons per unit volume
resonating at frequency u, the refractive part of the susceptibility is

N e2 1
0 () = . (6.28)
m0 u 2
2

This explains why the Kramers-Kronig relation is often written in the following
form:
Z
2 e2 f (u)
n () = 1 + Pr du 2 . (6.29)
m0 0 u 2
6.4 Oscillator strength 25

f = f () is the oscillator strength describing the distribution of resonance

frequencies.
Z
N= du f (u) (6.30)
0

is the spatial density of elastically bound electrons. The oscillator strength of

(6.29) is given by

2 mc0
f () = n()() . (6.31)
e2

Note that the static dielectric constant is

N e2 1
(0) = 1 + (6.32)
m0 2

where
is an average resonance frequency defined by
Z
f (u) N
du 2
= 2 . (6.33)
0 u

The integral converges because f () is proportional to 0 () 2 .

For very high frequencies (X rays) the refractive index is given by

N e2 1
n2 () = 1 . (6.34)
m0 2

A refractive index less than 1 means that the phase velocity /k = c/n is larger
than c. This does not imply that X ray signals may travel faster than light.
Wave packets travel with group velocity v = d/dk which differs from phase
velocity. The speed of a signal is yet another story.
If the angular frequency lies outside an absorption band, f () = 0, the
principal value operator Pr in (6.29) may be omitted, and we deduce
Z
dn2 2 e2 f (u)
= du 0 . (6.35)
d m0 0 (u2 2 )2

The permittivity, or the refractive index, grows with angular frequency. This
behaviour of dispersion is normal. Abnormal dispersionthe refractive index
decreases with increasing frequencyis possible only within absorption bands.
If f () > 0, the above argument does not hold true. Abnormal dispersion
comes necessarily with large absorption.
26 7 DISSIPATION-FLUCTUATION THEOREM

7 Dissipation-fluctuation theorem
It is well known that things become simpler if one concentrates on the essentials.
We shall therefore discuss the relation between dissipation and fluctuation on a
very abstract level. The interaction of a light wave with matter will then serve
as an example.
We discuss a system which, if unperturbed, is described by its Hamiltonian H.
The systems equilibrium state is G exp (H) where = 1/kB T is short for
the inverse temperature (up to the Boltzmann constant kB ). The expectation
value of an observable M in the equilibrium state is denoted by hM i = tr GM
throughout.
If the equilibrium is disturbed by a time dependent contribution

Ht = H (t)V , (7.1)

the linear response, as felt by an observable M , is given by

Z
hM it = hM i d (t ) (M V ; ) . (7.2)
0

There is an intimate relation between response functions (AB; ) and corre-

lation functions K(AB; ) which is the subject of this section.

7.1 The Wiener-Khinchin theorem

If A denotes an observable of the system under discussion,

~i tH
A(t) = Ut AUt where Ut = e (7.3)

is considered to be a process. Note that the equilibrium values are constant,

hA(t)i = tr GUt AUt = tr Ut GUt A = tr GA = hA(0)i . (7.4)

However, there will be fluctuations. We describe them by the time correlation

function

hA(t + )B(t) + B(t)A(t + )i

K(AB; ) = hAihBi . (7.5)
2

Because we have chosen the symmetrized product, the correlations function of

two observables A and B will always be real. Note that the time argument t
is absent on the left hand side of (7.5). K(AB; 0) is the correlation proper.
K(AA; ) is a time auto-correlation function.
Let us insert the Fourier decomposed processes,
Z
d i(t + )
A(t + ) = hAi + e A() (7.6)
2
7.2 Kubo-Martin-Schwinger formula 27

and
Z
d 0 i 0 t 0
B(t) = hBi + e B ( ) . (7.7)
2

i( 0 )t
There will be a factor e leading to a dependency on t unless

B
A() ( 0 ) + B
( 0 )A()

= 2( 0 )S() . (7.8)
2

In a stationary state, like in the Gibbs state, the Fourier components of fluctu-
ations with different frequencies are not correlated. Correlated fluctuations of
different frequencies would cause beating.
The frequency dependent function S = S(AB; ) in (7.8) is called a spectral
density. With it we may write
Z
d i
K(AB; ) = e S(AB; ) . (7.9)
2

S(AA; ), the spectral density of an auto-correlation function K(AA; ) is never

negative,
Z
d i
K(AA; ) = e S(AA; ) where S(AA; ) 0 . (7.10)
2

This finding is known as the Wiener-Khintchin theorem.

7.2 Kubo-Martin-Schwinger formula

Let us define

~i zH i
~ zH
A(z) = e Ae (7.11)

for z C.
We now exploit the fact that both the Gibbs state and the time translation
operators are exponential functions of the energy:

H H H H
A(z) e = e e A(z) e , (7.12)

i.e. A(z)G = GA(z i~). We multiply from the right by B and apply the
trace operator. The result

hBA(z)i = hA(z i~)Bi (7.13)

is the famous Kubo-Martin-Schwinger, or KMS formula.

28 7 DISSIPATION-FLUCTUATION THEOREM

7.3 Response and correlation

Note the formal similarity between the response function

i
(AB; ) = h [ A( ) , B ]i (7.14)
~

and the correlation function

hA( )B + BA( )i
K(AB; ) = hAihBi . (7.15)
2

The difference is essentially the difference between a commutator and an anti-

commutator.
We therefore investigate the product,

( ) = hA( )Bi hAihBi . (7.16)

Its Fourier transform is

Z
i
() = d e ( ) . (7.17)

Let us now discuss the function

Z
d iz
f (z) = e () . (7.18)
2

It can be shown that this function is analytic within a large enough strip of
the complex z-plane around the real axis. On the real axis we clearly have
f ( ) = ( ). Indeed, function f is the analytic continuation into the complex
plane of function .
With (7.11) we may work out

g(z) = hA(z)Bi hAihBi (7.19)

which is analytic in a sufficiently broad strip around the real axis. On the real
axis we have g( ) = f ( ). Therefore, g(z) and f (z) coincide everywhere.
Applying the KMS formula yields

hBA( )i hAihBi = f ( i~) . (7.20)

One should compare this with

hA( )Bi hAihBi = f ( ) . (7.21)

7.4 The Callen-Welton theorem 29

7.4 The Callen-Welton theorem

The response function can now be written as

i
(AB; ) = {f ( ) f ( i~)} . (7.22)
~

For the time correlation function we obtain

1
K(AB; ) = {f ( ) + f ( i~)} . (7.23)
2

We insert the representation (7.18) of f ,

Z
i d i ~
(AB; ) = e () 1 e , (7.24)
~ 2
Z
1 d i ~
K(AB; ) = e () 1 + e . (7.25)
2 2

One can read off immediately the Fourier transforms, and eliminating results
in

i ~
(AB; ) = 2 S(AB; ) tanh . (7.26)
~ 2

We are more interested in the (generalized) susceptibility

Z
i
(AB; ) = d e (AB; ) . (7.27)
0

Recall the result

(AB; ) (BA; ) = (AB; ) (7.28)

of section 6 where A was Pi and B = Pj .

Combining (7.26) and (7.28) results in

(AB; ) (BA; ) 1 ~
= S(AB; ) tanh . (7.29)
2i ~ 2

This is the famous fluctuation-dissipation theorem of Callen and Welton.

7.5 Energy dissipation

H
Let us discuss a process Wt with the initial condition W = G e .
This process is driven by the time dependent Hamiltonian
X
Ht = H r (t)Vr , (7.30)
r
30 7 DISSIPATION-FLUCTUATION THEOREM

and we assume r (t) 0 for |t| . During the small time span dt the
following amount of work
X X
dA = dr hVr it = dt r (t)hVr it (7.31)

is spent on the system, where hVr it = tr Wt Vr . This follows from the first law
of thermodynamics. The total work spent on the system is
XZ
A= dt r (t)hVr it . (7.32)
r

We assume that the external parameters are always small, so that the linear
approximation to the response is sufficient:
XZ
hVr it = hVr i + d rs ( ) s (t ) . (7.33)
s 0

The response functions are

i
rs ( ) = (Vr Vs ; ) = h [ Vr ( ) , Vs (0) ]i . (7.34)
~

The constant terms hVr i do not contribute, and we may write

XZ Z
A= dt d r (t)s (t ) ( ) rs ( ) . (7.35)
rs

This is an expression of type

Z Z Z
0 00 0 0 00 00 d
dt dt f (t ) g(t t ) h(t ) = f () g() h() , (7.36)
2

and we may write

X Z d
A= r () rs ()
i s () . (7.37)
rs
2

Note that
Z
i
rs () = d e ( )rs ( ) (7.38)

are generalized susceptibilities.

Complex conjugating (7.37) and interchanging the summation indices r, s will
not change the value of A,
X Z d
A= s () ()
i () , (7.39)
sr r
sr
2
7.5 Energy dissipation 31

and averaging (7.37) and (7.39) results in

X Z d
A= r () rs () sr ()
s () . (7.40)
rs
2 2i

With the dissipation-fluctuation theorem (7.29) we finally arrive at

Z X
2 d r () Srs ()
s ()
A= q(~) (7.41)
2 2 rs

where
2 x
q(x) = tanh (7.42)
x 2

is a quantum mechanical correction. By the Wiener-Khintchin theorem the

matrix Srs = Srs () of spectral densities is non-negative.
We conclude that it is impossible to perturb an equilibrium state in such a way
that the system extracts heat from its environment and delivers work: A 0.
This statement is an alternative formulation of the second law of thermody-
namics. We have proven it here for small perturbations of the equilibrium by
rapidly oscillating external parameters.
A more prosaic formulation reads: the absorptive part of the susceptibility truly
causes absorption of energy.
32 8 ONSAGER RELATIONS

8 Onsager relations
If the equilibrium is perturbed by more than one driving force, the generalized
susceptibilities rs () form a matrix. We have shown in section 7 that the
imaginary, or absorptive part rs ()sr () is non-negative and proportional
to the matrix of spectral densities. This was the subject of the dissipation-
fluctuation theorem.
In this section we discuss symmetry properties of the susceptibility matrix.

8.1 Symmetry of static susceptibilities

Let us first discuss the symmetry properties for adiabatic processes. Let =
1 , 2 , . . . denote external parameters or deviations thereof such that
X
H() = H r Vr (8.1)
r

is the relevant Hamiltonian. If the external parameters change slowly, the state
W of the system will always be the equilibrium state to the current external
parameters . The free energy is given by

1 H()
F (, ) = ln tr e . (8.2)

With h. . .i as expectation value in the Gibbs state to H(0) and h. . .i to H()

we find
X
hVr i
hVr i = hVr i + s . (8.3)
s
s =0

Comparing with (7.33) we conclude

Z
2 F (, )
rs (0) = d rs ( ) = . (8.4)
0 s r

It follows immediately that

rs (0) = sr (0) (8.5)

holds true. Can this symmetry be generalized to fast processes? Almost, as we

shall see.

8.2 Time reversal

To each linear operator A we assign a time-reversed operator A? by

(A) = A? (8.6)
8.2 Time reversal 33

where = (x1 , x2 , . . .) is an arbitrary wave function depending on the positions

of the particles. Let us ignore spin at this point. As usual, denotes complex
conjugation.
Note that the mapping A A? is anti-linear in the sense that A + B map to
A? + B ? , but zA becomes z A?
A position operator x amounts to the multiplication of the wave function by
its argument, which is real. Therefore, x? = x. Momentum operators are
represented by p = i~, therefore p? = p. This is the reason why the time-
reversal is sometimes called momentum reversal. By the way, angular momenta
change their sign at timer reversal as well.
If is an eigenfunction of A with eigenvalue a, then is an eigenfunction of A?
with eigenvalue a . Hence, if M is an observable, so is M ? , and if W represents
a mixed state, so does W ? . Observables are characterized by real eigenvalues.
States also have real eigenvalues which are probabilities. It is a simple exercise
to prove the expectation value of an observable in a state coincides with the
expectation value of the time reversed observable in the time reversed state,

tr W M = tr W ? M ? . (8.7)

An observable is of parity (with respect to time reversal) if M ? = M holds

true. Because of M ? ? = M only = 1 are possible. Position operators have
even, momentum or angular momentum operators odd parity.
Ordinarily, the Hamiltonian has even parity. However, if there is P an external
there is a contribution i B
quasi-static magnetic induction field B, i , the
magnetic dipole interaction. Since the magnetic momentum is proportional to
an angular momentum, which has odd time reversal parity, we conclude

? = H(B)
H(B) . (8.8)

The external magnetic induction field has to be reversed in order to guarantee

even parity of the Hamiltonian.
Now, the Gibbs state depends, among other external parameters, upon the
We conclude
quasi-static magnetic induction applied to the system, G = G(B).

? = G(B)
G(B) . (8.9)

The time translation operator is

= e ~i tH(B)

Ut (B) , (8.10)

and we easily work out

? = Ut (B)
Ut (B) . (8.11)

This explains why we speak of time reversal. The time reversed time translation
operator Ut? translates by the negative time span, Ut . However, B
B has
to be observed.
34 8 ONSAGER RELATIONS

8.3 Onsager theorem

Let us come back to the original situation that there is a Hamiltonian with a
time-dependent perturbation,
X

Ht = H(B) s (t)Vs . (8.12)
s

For the expectation values of the perturbing observables we have found

Z

tr Wt Vr = tr G(B)Vr + )s (t ) ,
d rs (B; (8.13)
0

where

) = tr G(B)
i r U (B)
, Vs ] .
rs (B; [ U (B)V (8.14)
~
The latter expression may be rewritten as

? i [ U (B)V
tr G(B) , Vs ]?
r U (B) (8.15)
~
which becomes

i r? U (B)
, Vs? ] .
tr G(B) [ U (B)V (8.16)
~
We assume that the Vr observables have parities r with respect to time reversal.
Thus we may write

i [ U (B)V
r s tr G(B) r U (B)
, Vs ] (8.17)
~
holds. By shifting the entire commutator in time by (which is allowed because
G is stationary) we arrive at

i [ Vr (B)
r s tr G(B) , U Vs U (B)
] , (8.18)
~
and reverting the commutator finally yields

) = r s sr (B;
rs (B; ) . (8.19)

This symmetry is maintained if (8.19) is Fourier transformed over positive time

spans resulting in a symmetry relation for the (generalized) susceptibilities:

) = r s sr (B;
rs (B; ) . (8.20)

(8.20) is an Onsager relation. It applies in particular to the dielectric suscepti-

bility where r = s = 1.
By the way, (8.20) and (8.5) do not contradict each other. r 6= s will imply
rs (0; 0) = 0.
8.4 Onsager relation for kinetic coefficients 35

8.4 Onsager relation for kinetic coefficients

Let us define the flux of V by

i
J = V = [ H , V ] . (8.21)
~

In the case of electric polarization Pr (x),

X
Jr = qa x a 3 (x xa ) (8.22)
a

are the three components of the electric current density.

If V has parity with respect to time reversal, then J has parity .
Because the Gibbs state is stationary, the equilibrium expectation value of fluxes
vanishes. The expectation value J(t) = tr Wt J in the perturbed state is also
called a flux. Fluxes exist because the equilibrium is perturbed.
With

i
(Jr Vs ; ) = h [ Vr ( ) , Vs ]i = sr ( ) (8.23)
~

we conclude

XZ
Jr (t) = d rs ( )s (t ) . (8.24)
s 0

Fourier transforming results in

X
Jr () = s () ,
rs () (8.25)
s

where
Z
i
rs () = d e rs ( ) (8.26)
0

are kinetic coefficients. Differentiating (8.19) and inserting into (8.26) leads to

) = r s sr (B;
rs (B; ) . (8.27)

This symmetry relation for kinetic coefficients was derived already in 1931 by
Onsager, although for stationary currents only ( = 0). His reasoning was
different and, from todays point of view, questionable. However, Onsager also
considered other causes of irreversible effects like gradients of temperature and
chemical potential.
36 8 ONSAGER RELATIONS

8.5 Electrical conductivity and Hall effect

If the equilibrium state of matter is disturbed by an electric field, then
X
Ji () = )E
ij (B; j () . (8.28)
j

This is Ohms law. ij is the conductivity tensor which may depend on an ex-
ternal magnetic field. For a sufficiently weak magnetic field and for an isotropic
material we may expand according to
X
) = () ij + h()
ij (B; k + . . . .
ijk B (8.29)
k

The first term describes electric conduction in the absence of a magnetic field.
The second term must be antisymmetric in i and j because it is linear in
There is a current contribution perpendicular to the driving electric field
B.
strength E and perpendicular to the quasi-static induction B.
 37

9 Electro- and magnetooptic effects

Let us recall the expression
Z Z
1 i i
ij () = d e d3 tr G [ Pi (, ) , Pj (0, 0) ] (9.1)
0 0 ~

for the dielectric susceptibility tensor.

X
Pi (, ) = qa xa 3 (xa ) (9.2)
a

is the polarization at location , translated by a time span . The suscepti-

bilities are integrals over expectation values in the Gibbs, or equilibrium state.
Therefore, they depend on all parameters which describe the thermal equilib-
rium state: composition of matter, temperature, external quasi-static electric
and magnetic fields, stress, and so forth.
We will concentrate on quasi-static external electric and magnetic fields E and
B, respectively.
A systematic power series expansion up to second order reads

ij (E, B) = 00
ij (9.3)
+ 10
ijk Ek (9.4)
+ 01
ijk Bk (9.5)
+ 20
ijk` Ek E` (9.6)
+ 11
ijk` Ek B` (9.7)
+ 02
ijk` Bk B` (9.8)
+ ...

Recall that we speak of the Hermitian refractive part,

ij (E, B) = ji (E, B) , (9.9)

and that Onsagers relation demands

ij (E, B) = ji (E, B) . (9.10)

The same is true for the permittivity tensor ij = ij + ij .

9.1 Crystal optics

We assume a real symmetric, but otherwise arbitrary permittivity tensor. The
medium is assumed to be non-magnetic. Let us look for plane wave solutions:

it i n k0 w x
F (t, x) = f e e . (9.11)
38 9 ELECTRO- AND MAGNETOOPTIC EFFECTS

F is any vector field. w, a unit vector, defines the direction of propagation, k0

is short for /c, and n denotes the refractive index.
Maxwells equations read

cb = n w ( e) ; e = n w (cb) . (9.12)

Note that the remaining equations w ( e) = 0 as well as w b = 0 are

automatically fulfilled: the dielectric displacement D and the induction field B
are divergence free. c is the speed of light in vacuum.
For a prescribed direction w of propagation, (8.2) may be understood as an
eigenvalue problem. w . . . is a linear operator described by the matrix

0 w3 w2

W = w3 0 w1 . (9.13)
w2 w1 0

Eliminating the induction result in

1 W 2 e = n2 e . (9.14)

Because of W w = 0, there is one unphysical solution, namely e = w, with

n2 = 0. The remaining two eigenvectors describe the polarization of allowed
propagation modes, the eigenvalues n2 defining the respective refractive in-
dices.

Optically isotropic media The three eigenvalues of the susceptibility coin-

cide, 00
ij = ij . Amorphous substances, such as glass, and crystals with cubic
symmetry, such as NaCl, are examples. All transversely polarized plain waves

are eigenmodes, the refractive index being n = 1 + .

Optically uniaxial crystals If two eigenvalues of 00 ij coincide, but differ

from the third, one has a preferred axis and, orthogonal to it, a preferred plane.
If the polarization vector lies in the preferred plane, the beam propagates with
the ordinary main refractive index6 no , the square root of the doubly occurring
eigenvalue. If the beam is polarized in the preferred axis, one speaks of the
extraordinary main refractive index ne . The preferred axis is an optical axis in
the following sense: if light propagates along it, the refractive index does not
depend on polarization.
If a light beam enters an optically uniaxial crystal at an arbitrary angle, it
splits into an ordinary7 and an extraordinary beam which travel with different
refractive indices. One therefore speaks of double refraction, or birefringence.
LiNbO3 is a well-known and extensively studied birefringent crystal.
6
The refractive index n is a property of the wave. With k0 = /c and w as propagation
direction, the wave vector is q = nk0 w.
7
ordinary, because it propagates in the plane of incidence spanned by the incoming beam
and the surface normal
9.2 Pockels effect 39

Optically biaxial crystals If the symmetry of a crystal is low enough, all

three eigenvalues of the susceptibilty tensor 00 ij may be different. One can,
however, demonstrate that there are two optical axes such that beams, when
traveling along it, have a polarization independent refractive index. Therefore,
crystals with three different susceptibilty eigenvalues are called optically biaxial.
They are birefringent as well. An incident beam splits into two beams which
propagate with different refractive indices. However, in the general case, none
of them is ordinary in the sense that it remains in the plane of incidence.

9.2 Pockels effect

The addition 10ijk Ek is linear in the strength of an external electric field. (9.9)
and (9.10) require 10
ijk to be a real third-rank tensor which has to be symmetric
in the first two indices.
Note that space inversion t t and x x is a symmetry of Maxwells
equations if E E and B B. Put otherwise, the electric field strength
is a polar, the magnetic field strength an axial vector. If the Gibbs state is
invariant with respect to space inversion, 10ijk will acquire a minus sign, it must
coincide with its negative, hence vanish. We conclude that a third rank tensor
is possible only for crystals which do not have an inversion center. If there is
a location (at x = 0) such that x x is a symmetry, then 10 ijk necessarily
vanishes.
Lithium niobate8 LiNbO3 may serve as an example. There is an xy-plane with
a 120 symmetry and mirror symmetry x x. The c-axis, orthogonal to
this plane, has a preferred direction. After all, LiNbO3 is ferro-electric. Conse-
quently, there are four different third-rank tensors which fulfill all requirements.
We have deferred a detailed discussion to appendix B.
The Pockels effect, the dependence of the refractive index on the first power of
a quasi-static electric field strength, allows to switch and modulate light.

9.3 Faraday effect

Faraday has discovered that a magnetic field may affect the propagation of
light. The contribution 01 ijk Bk to the susceptibility must be Hermitian and
symmetric if B is reversed. Consequently, 01 ijk is to be purely imaginary and
antisymmetric in i and j.
To be specific, let us discuss yttrium iron garnet (YIG), a complicated artificial
crystal which is transparent in the micrometer wavelength region and ferri-
magnetic at the same time. There is a contribution

01
ijk Bk = iKijk Mk (9.15)

to the susceptibility. Because YIG is ferri-magnetic, it is customary to refer to

the magnetization M instead of B.
8
at room temperature of 3m-symmetry
40 9 ELECTRO- AND MAGNETOOPTIC EFFECTS

Let us discuss a situation where a wave propagates along the direction of mag-
netization, the z-axis, say. With k0 = /c and k = nk0 (0, 0, 1) we have to
solve
k k E = k02 E (9.16)
where

iKM 0

=
iKM 0
. (9.17)
0 0

With

1 1
1 1
L =
e i and e
= i (9.18)
2 2
R
0 0

we have made out solutions which describe left and right hand circularly polar-
ized light,
it inL,R k0 z
eL,R e
E L,R = E e . (9.19)
Their refractive indices differ:
n2L = + KM and n2R = KM . (9.20)
If a linearly polarized plane wave enters a crystal with linear magnetooptic
effect, it splits into left and right hand circularly polarized waves. These waves
travel with slightly different phase velocity. When exiting the crystal, the two
waves recombine again to a linearly polarized wave. However, the direction of
polarization is rotated by an angle `, where ` is the path length through the
magnetooptic medium. The specific Faraday rotation constant is given by
2 KM
= , (9.21)
2n

where n = is the average refractive index.
The Faraday effect distinguishes between forward and backward propagation of
light and allows to build an optical isolator.

9.4 Kerr effect

Unlike the Pockels effect, which is linear in the external electric field strength
and requires a medium without inversion center, the Kerr effect is usually much
weaker, because of second order. It is a property even of isotropic media since
20
ijk` ij k` (9.22)
fulfills all symmetry requirements: real, symmetric in the first pair of indices,
symmetric in the second pair of indices. (9.18) amounts to a shift of the refrac-
tive index by a term proportional to |E|2 .
9.5 Magneto-electric effect 41

9.5 Magneto-electric effect

There is no accepted name for a contribution 11 ijk` Ek B` to the susceptibility
tensor. One may speak of a linear dependence of the Pockels effect on an
external magnetic field. One may likewise say that the Faraday effect depends
linearly on an external electric field. The 11
ijk` must be purely imaginary and
antisymmetric in the first pair of indices. The crystal should be ferro- or ferri-
magnetic, transparent, and must not have an inversion center.

9.6 Cotton-Mouton effect

The Cotton-Mouton effect, a permittivity shift which is of second order in
the magnetic field, is rather weak for para-magnetic material. However, if it
occurs in a ferri- or ferro-magnetic substance, it may be strong. A contribution
02 2
ijk` ij k` will change by a term which is proportional to |M | and will
escape attention. The Cotton-Mouton effect is reciprocal, unlike the Faraday
effect it does not distinguish between forward and backward propagation.
42 10 SPATIAL DISPERSION

10 Spatial dispersion
Recall the definition of the susceptibility tensor:
Z Z
1 i iq
ij (, q) = d e d3 e ij (, ) (10.1)
0 0

where
i
ij (, ) = tr G [ Pi (, ) , Pj (0, 0) ] . (10.2)
~
The polarizationas an operatoris defined by
X
Pi (x) = qa xi 3 (xa x) . (10.3)
a

The sum extends over all charged particles, and xa is particles a position,
likewise an observable. The time shift is accomplished by the unperturbed
Hamiltonian H. The entire Hamiltonian is
Z
Ht = H d3 xPj (x)Ej (t, x) (10.4)

where the external electric field may depend on time.

The response functions ij appear in
Z Z
tr Wt Pi (x) = d d3 ij (, ) Ej (t , x ) (10.5)
0

where W = Wt is the disturbed state, and the right hand side refers to the
lowest order approximation9 .
Fourier transforming E = E(t, x) and the left hand side of (10.5) results in

j (, q) .
Pi (, q) = ij (, q) E (10.6)

10.1 Dispersion relation

So far we have argued that the dependence of ij (, q) on the wave vector
q is so weak that the value at q = 0 suffices. Interactions in a solid spread
out with the speed of sound while the electromagnetic field is governed by the
speed of light, and both differ by orders of magnitude. Still, there are certain
effects where spatial dispersionthe dependency of the susceptibility on the
wave vectorhas to be taken into account.
The arguments and q for E = E(,
q) are not independent. After all, the
electric field has to obey Maxwells equations which, for this purpose, read
2
j =
(q 2 ij qi qj )E j .
(ij + ij )E (10.7)
c
9
Ferro-electric materials contain an additional static contribution Pi = tr GPi
10.2 Optical activity 43

Recall that both E j and ij in (10.7) depend on and q. The wave equation
may be viewed as an eigenvalue equation: q given, find an such that (10.7)

is satisfied for a non-trivial eigenvector E.
The relation between wave vectors q and angular frequency is a dispersion
relation. Since enters (10.7) twice, as a factor on the right hand side and as
an argument of the susceptibility, the eigenvalue problem is non-linear.
This complication is seldom realized because the approximation ij (, q)
ij (, 0) is used. The angular frequency then may be considered as an indepen-
dent variable, and the refractive index squared10 appears to be the eigenvalue
to be determined.

10.2 Optical activity

To take spatial dispersion into account, let us expand the susceptibility as

ij (, q) = ij () + oa
ijk qk + . . . . (10.8)

The additional contribution oa oa

ij = ijk qk causes optical activity by traditional
terminology.
It is not difficult to show that invariance with respect to time reversal11 results
in

ij (, ) = ji (, ) , (10.9)

which, in turn, implies

ij (, q) = ji (, q) . (10.10)

We conclude that oaijk is purely imaginary and antisymmetric in the first two
indices, because the susceptibilty must be Hermitian and because of (10.10).
We write12

oa
ij = iijk gk with gk = Gk` q` . (10.11)

g is the gyration vector which depends linearly on the wave vector.

With respect to space inversion, Gk` must be a rank 2 pseudo-tensor because
ijk is a rank 3 pseudo-tensor while the susceptibility and the wave vector are
normal tensors of rank 2 and 1, respectively.
Only crystals with a built-in screw sense can be optically active. Quartz is an
example, or a solution of natural dextrose. Note that the expression

fe
ij = iKijk Mk , (10.12)
10
Recall |q| = n /c where n is the refractive index.
11
in the absence of magnetic fields
12
ijk is the total antisymmetric Levi-Civita symbol, a rank 3 pseudo-tensor with respect
to space inversion
44 10 SPATIAL DISPERSION

which causes the Faraday effect, has the same structure. In full generality,
also the Faraday effect should be described by a gyration vector which depends
linearly on the magnetization vector. However, since the magnetization is a
pseudo-vector, the tensor G in gk = Gk` M` is an ordinary rank 2 tensor, such
as k` .
Optical activity results in the rotation of the polarization vector, either to the
left or to the right13 . Also K in (10.12) can be positive or negative. The princi-
pal difference between the Faraday effect and optical activity is the following. If
a wave passes the medium twice, forward and backward, the Faraday rotations
add, the rotation of the polarization vector, as caused by optical activity, is
reverted. Optical activity is a reciprocal effect.
The fact that all plants produce glucose in its optically right active form (dex-
trose) is very astonishing. Artificial glucose consists of equal amounts of left and
right optically active molecules. One molecule is the mirror image of the other
form, and no chemical mechanism is known which prefers right-handedness over
left-handedness. May it be that all plants stem from one and the same mother
plant?
As for quartz, both forms occur in nature. The crystal growth mechanism is
such that entire crystals are optically either left or right active (although there
are twins).

13
Right is clockwise as seen by an observer facing the light source.
 45

11 Non-linear response
Denote by H the Hamiltonian of the unperturbed system. The time translation
operator associated with it is

~i tH
Ut = e , (11.1)

as derived in section 4. A time translated operator A is denoted by

A(t) = Ut AUt . (11.2)

We want to solve the following Schr

odinger equation for the systems state Wt :

d i
W t = [ W t , H + Vt ] . (11.3)
dt ~

Vt is a possibly explicitly time dependent perturbation of the system. Before

the perturbations has been switched on, the system was in an equilibrium state,

(F H)
Wt G = e for t . (11.4)

Recall that F is the systems free energy which guarantees tr G = 1, and

= 1/kB T where T is the temperature of the Gibbs state.
As explained in detail in section 4, (11.3) is best solved in the interaction picture,
with states and observables time translated according to (11.2). One arrives at
the following integral equation which combines the differential equation (11.3)
and the initial condition (11.4):
Z t
i
Wt (t) = G + ds [ Ws (s) , Vs (s) ] . (11.5)
~

A sensible result: the state now, at time t, depends on previous states and on
previous perturbations only, and on the initial state. Small perturbations will
cause only small deviations from the initial state.

11.1 Higher order response

The zeroth order approximation is Wt (t) = G+. . .. The first order contribution
is obtained by inserting the lowest order into (11.5):
Z t
i
ds [ G , Vs (s) ] + . . . . (11.6)
~

The second order addition is

Z t Z s
i i
ds du [ [ G , Vu (u) ] , Vs (s) ] + . . . . (11.7)
~ ~
46 11 NON-LINEAR RESPONSE

The expectation value of an observable M is tr GM (t) = tr GM in lowest

approximation. The next term (linear response) is
Z t
i
ds tr G [ Vs (s) , M (t) ] . (11.8)
~

The quadratic response may be written as

Z t Z s
i i
ds du tr G [ Vu (u) , [ Vs (s) , M (t) ] ] . (11.9)
~ ~

Let us now specialize to

Vt = (t)V . (11.10)

The linear response contribution (11.8) is

Z
d1 (t 1 ) (1) (1 ) (11.11)
0

with
i
(1) (1 ) = tr G [ M , V (1 ) ] . (11.12)
~

(11.9) reads
Z Z
d1 d2 (t 1 )(t 1 2 ) (2) (1 , 2 ) (11.13)
0 0

with
i i
(2) (1 , 2 ) = tr G [ [ M , V (1 ) ] , V (1 2 ) ] . (11.14)
~ ~

Note that also the second order response function (2) (1 , 2 ) is an expectation
value in the unperturbed Gibbs state.

11.2 Susceptibilities
Let us Fourier transform m(t) = tr Wt M tr GM and (t). We arrive at the
following expression,
Z
(1) du (2) u) + . . . , (11.15)
m()
= () () + (u, u) (u) (
2

with
Z
(1) i1 1
(1 ) = d1 e (1) (1 ) , (11.16)
0
11.3 Second harmonic generation 47

Z Z
(2) i(1 + 2 )1 i2 2
(1 , 2 ) = d1 e d2 e (2) (1 , 2 ) , (11.17)
0 0

and so forth.
For the interaction of the electromagnetic field with matter, in electric dipole
approximation, we replace (11.10) by
Z
Vt = d3 x Ei (t, x)Pi (x) . (11.18)

E(t, x) is a time dependent external electric field, P (x) the polarization at x.

As usual, we use the same symbol for the polarization (as an observable) and
its expectation value. It Fourier transform is
Z
j () + du (2) (u, u) E
Pi () = (1) () E j (u) E
k ( u) + . . . (11.19)
2 ijk

The expression
Z Z
(1) i1 1 i
ij (1 ) = d1 e d3 h [ Pi (0, 0) , Pj (1 , ) ]i (11.20)
0 ~

for the linear response susceptibility has been thoroughly studied in previous
sections.
The quadratic response is described by

(2)
ijk (1 , 2 ) =
Z Z Z Z
i(1 + 2 )1 i2 2 3
d1 e d2 e d d3 (11.21)
0 0

i i
h [ [ Pi (0, 0) , Pj (1 , ) ] , Pk (1 2 , ) ]i .
~ ~

(11.21) is already a rather complicated expression. What is the counterpart

to the Kramers-Kronig relation? What is the counterpart to the fluctuation-
dissipation theorem? Does the second law of thermodynamics hold up to second
order response theory? Are there symmetries with respect to frequencies and
tensorial indices? Non-linear response theory is not yet well investigated, and
we shall stop at this point.

11.3 Second harmonic generation

(2)
ijk (1 , 2 ) is a proper tensor of rank three, symmetric in the second and
third index. It vanishes if the Gibbs state is invariant with respect to space
inversion. Only crystals without inversion center will respond in second order
to perturbations by an electric field14 .
14 (2)
ijk (0, ) is the same as 20
ijk () which describes the Pockels effect.
48 11 NON-LINEAR RESPONSE

Let us discuss lithium niobate, LiNbO3 . We choose a coordinate system such

that the z-axis coincides with the crystallographic c-axis. See appendix B for
details. Assume a light wave traveling with wave number q in x-direction, being
polarized in y-direction. The relation between angular frequency and wave
number is

q = no () , (11.22)
c

where no is the ordinary refractive index. The plane wave is described by

it iqx
E1 = E3 = 0 and E2 (t, x) = A e e . (11.23)

(2) (2)
122 and 222 vanish, therefore the second order polarization response is

(2) 2i 2iqx
P1 = P2 = 0 and P3 (t, x) = A2 322 e e . (11.24)

In general, this does not excite a light wave because frequency and wave number
do not fit. Only if the phase matching condition

2
2q = ne (2) (11.25)
c

holds, (11.24) will excite an electromagnetic plane wave. Note that it propagates
with the extraordinary refractive index.
(11.22) and (11.25) amount to

no () = ne (2) . (11.26)

The ordinary refractive index of lithium niobate is always larger then the ex-
traordinary. Since both increasein the infrared or visiblewith frequency,
condition (11.26) can be met, for a certain angular frequency . This frequency
to be doubled depends on all quantities which affect the refractive indices, such
as temperature, composition of the material etc.
Note that the intensity of frequency doubled light grows quadratically with the
intensity of incident light. In particular, a resonator which confines -light and
is leaky for 2-light, allows high conversion rates.
Frequency doubling or tripling is technically important because cheap semi-
conductor lasers emit light of low frequency, while many optical applications
demand short wavelengths.
Frequency doubling is well known in music. Musical instruments are highly non-
linear. A harmonic tone (of one frequency only) always excites other tones, such
as the octave, or the second harmonic. Therefore, frequency doubling is also
known as second harmonic generation, or SHG.
 49

A Causal functions
Assume
Z
d it
f (t) = e g() (A.1)
2

where g = g() is analytic. It is a simple exercise to prove

f (t) = 0 for t < 0 (A.2)

provided g = g() is holomorphic15 in the upper plane Im > 0. The converse

is also true. If f = f (t) is a causal function, as characterized by (A.2), its
Fourier transform g = g() is holomorphic in the upper half plane.
Hence, f (t) = (t)f (t) holds true which may be expressed as
Z
du g(u)
g() = , (A.3)
2 i( u)

where the limit 0 < 0 is understood. The Fourier transform of a product

is the convolution of the respective Fourier transforms.
(A.3) says that the pole at u = has to be avoided in the lower plane. Adding
and subtracting an integration path which avoids the singularity u = in the
upper half plane results in
Z
du g(u)
g() = i Pr . (A.4)
u
R
Pr denotes the principal value integral, the average of avoiding the singularity
in the lower and the upper half plane. The dispersion relation
Z
0 du g 00 (u)
g () = Pr (A.5)
u

is a simple consequence, where g 0 is the real and g 00 the imaginary part of g. The
Fourier transform of a causal function obeys a Kramer-Kronig like dispersion
relation.
So far we never required f = f (t) to be real. If f is a causal function, then if
is causal as well. Therefore,
Z
00 du g 0 (u)
g () = Pr (A.6)
u
is an alternative formulation.
If f is also real, then its Fourier transform must obey g () = g(). g 0 = g 0 ()
is an even and g 00 = g 00 () an odd function. This then implies
Z
0 du 2u g 00 (u)
g () = Pr . (A.7)
0 u2 2
15
analytic and free of singularities
50 A CAUSAL FUNCTIONS

Now, only positive frequencies need to be considered. The counterpart to (A.6)

is
Z
du g 0 (u)
g 00 () = 2 Pr . (A.8)
0 2 u2

Let us study an example.

The Fourier transform of

t
f (t) = (t) e (A.9)

is
1
g() = . (A.10)
i

Indeed, this Fourier transform has a singularity at = i in the lower half

plane (we assume > 0), and it respects g() = g (). The real and
imaginary parts read

g 0 () = , g 00 () = 2 , (A.11)
2 + 2 + 2

they are even and odd functions of , respectively.

Z
du 1 1 1
Pr = (A.12)
u + i u i

is easily established, and the real part and imaginary part of this identity are
the dispersion relations (A.5) and (A.6).
By the way, (A.9) is the Greens function for a relaxation process.

p + p = F (t) (A.13)

is solved by
Z

p(t) = d e F (t ) . (A.14)
0
 51

B Crystal symmetry
The electromagnetic interaction, which governs the field of solid state physics,
is invariant with respect to translations in space and time, rotations, space
inversion, time reversal, and boosts. Let us concentrate on a body at rest, hence
the boosts need not be considered any more. We have already investigated the
consequences of invariance with respect to time reversal, so let us disregard
this aspect here as well. We remain with spatial translations, rotations, and
inversion. We first discuss why a symmetric Hamiltonian may give rise to an
unsymmetric equilibrium state. This phenomenon, the spontaneous breakdown
of symmetry, is widespread, in particular, when a system is very large, like a
carbon hydrogen molecule or an ideally infinite crystal.

B.1 Spontaneous symmetry breaking

The Hamiltonian H of ordinary matter is invariant with respect to transla-
tion, rotation, and inversion. Consequently, the equilibrium, or Gibbs state,
which is a function of the Hamiltonian, should also be invariant with respect to
translations, rotations, and inversion. This is obviously wrong. Crystals have
preferred directions (crystal axes), and they need not have an inversion center.
In general, the symmetry of a theory does not imply the symmetry of the
ground state of a particular system. Symmetry may be broken spontaneously.
Just think about the planetary system. Although there is rotational invariance,
the planets do not move on spheres (how should they?) or circles. Rotational
symmetry merely says that a planetary system which is tilted by a certain angle
would be possible as well.
However, quantum mechanics teaches something else. The hydrogen atom, for
example, is not a minute planetary system. All possible electron orbits interfere
in such a way that the ground state is truly spherically symmetric.
Another, less trivial example is the ground state of the ammonia molecule
NH3 . The three protons form an equilateral triangle. The nitrogen ion is
either above or below. In fact, the ground state is a symmetrical superposition
of above/below, and its energy is lower, by a tiny amount, than the energy
of the anti-symmetrical superposition. The energy difference ~0 defines the
microwave frequency standard (f0 = 23.87012 MHz).
The larger the molecule, the less probable is a transition between a state and its
mirror-image. The right-handed version of a glucose molecule may, in principle,
become a left handed version. The true ground state is a symmetric combination
of both, but since so many nuclei would have to change position simultaneously,
the transition time exceeds the age of the universe by orders of magnitude.
Hence, if produced by a right handed plant, a dextrose molecule will stay in a
state which defies the principle of inversion symmetry.
The same applies to crystals. For instance, if LiNbO3 is grown, an electric
current defines a preferred direction. Once the crystal cools down, it keeps its
c-axis orientation. Not even the largest external electric field may revert the
crystals polarization.
52 B CRYSTAL SYMMETRY

B.2 Symmetry groups

One has to distinguish between translations and point transformations. A crys-
tal lattice consists of translated copies of a unit cell. A point symmetry element
sends each ion of the cell to another position such that the transformed cell looks
the same. The point symmetry elements can be an inversion center, a mirror
plane, a 1, 2, 3, 4, or 6-fold rotation axis, or a 1, 2, 3, 4, or 6-fold rotation axis
with inversion. These point symmetry elements form groups, altogether 32,
each describing a crystal symmetry class. There is an internationally accepted
short hand notation system. For example, 3m says that there is a three-fold
rotation axis, without inversion, and a mirror plane such that the rotation axis
lies in that plane. 3m would indicate an additional reflection symmetry with
respect to a plane orthogonal to the rotation axis. Consult the standard work
by Nye16 for details.

B.3 A case study

Let us discuss lithium niobate LiNbO3 , a non-centric crystal of class 3m. This
symmetry class allows proper tensors of rank 3.
The symmetry group consists of the identity I, a reflection

1 0 0

=
0 1 0
, (B.1)
0 0 1

and a 120 rotation

p p
1/4 3/4 0
p p
R=
3/4 1/4 0 . (B.2)
0 0 1

The entire group 3m is made up of {I, , R, R1 , R, R }, because of R2 =

R1 , 2 = I and R1 = R . The multiplication table is

I R R1 R R
I I R R1 R R
I R R R R1
R R R R1 I R (B.3)
R1 R1 R I R R
R R R1 R I R
R R R R R1 I
16
J. F. Nye, Physical Properties of Crystals, their Representation by Tensors and Matrices;
Oxford University Press
B.3 A case study 53

We denote by c the polar three-fold symmetry axis (a vector of unit length)

and by u
, v
, w
three unit vectors in a plane orthogonal to it, with angles of
120 between them. We identify u =x and choose y orthogonal to it. This
means
r r r r
1 3 1 3
v
= x
+ y and w = x
y
. (B.4)
4 4 4 4

x
x should be a symmetry as well as a permutation of u
, v
, and w.
c c

must not be a symmetry. After all, we deal with 3m, not 3m. We look for a
third rank tensor dijk = dikj which would describe the Pockels effect or second
harmonic generation.

(1)
Dijk = ci cj ck (B.5)

fulfills all requirements.

u
j u
k + vj vk + w
j w
k is a symmetric second rank tensor with three-fold rotation
symmetry. A short calculation shows that it is proportional to x i x
j + yi yj .
Consequently,

(2)
Dijk = ci (
xj x
k + yj yk ) (B.6)

is another admissible rank 3 tensor. A third possibility is

(3)
Dijk = x
i (
cj x
k + x
j ck ) + yi (
cj yk + yj ck ) . (B.7)

Another tensor, namely u

i u
j u
k + vi vj vk + w
i w
j w
k , proportional to

x
i ( k yj yk ) yi (
xj x xj yk + yj x
k ) , (B.8)

is not acceptable, because it is antisymmetric with respect to x x. How-

ever, we only need to replace u by c u , and the same for v
and w which
amounts to interchanging x and y . Instead of (B.8) we now obtain the fourth
tensor
(4)
yj yk x
Dijk = yi ( k ) x
j x i (
xj yk + yj x
k ) . (B.9)

Not that the first three tensors are invariant with respect to x
y
, so c
u

instead of u
etc. does not produce new tensors. For the point group 3m there
are just four linearly independent tensors of rank 3.
The most general rank 3 tensor with 3m symmetry may we written as

4
X (r)
dijk = dr Dijk (B.10)
r=1

with four different invariants dr which depend on the effect and the material
under study.
54 B CRYSTAL SYMMETRY

We list the non-vanishing tensor elements:

d1 = d333 , (B.11)
d2 = d311 = d322 , (B.12)
d3 = d131 = d113 = d223 = d232 , (B.13)
d4 = d222 = d211 = d112 = d121 . (B.14)

A word of caution. Some authors represent a symmetric matrix by a six com-

ponent vector:

T1 T 6 T 5

Tij =
T 6 T 2 T 4
.
(B.15)
T5 T4 T3

When summing over indices, they count the off-diagonal index pairs, such as
6 = (1, 2) = (2, 1), only once. Consequently you will find factors 2 in some such
tables. The rank 3 tensor dijk is represented by a 3 6 matrix di` where i runs
from 1 to 3 and ` from 1 to 6. In our case, (B.14) might read d16 = 2d22 .
Nye17 follows this convention.

17
Physical properties of Crystals, loc. cit.
 55

C Glossary
Birefringence In the absence of gyrotropy the refractive part of the suscepti-
bility tensor ij , and with it the permittivity tensor ij = ij + ij are real and
symmetric. There is a Cartesian coordinate system such that the permittivity is
diagonal, ij = (i) ij . A plane wave, linearly polarized
along the ith coordinate
axis, propagates with main refractive index n = (i) . If all three eigenvalues
of the permittivity tensor coincide, the medium is optically isotropic. If two
are equal and differ from the third, there is a preferred axis. Light propagating
along this optical axis may be polarized arbitrarily, but perpendicular to the
optical axis. The corresponding main refractive index is called ordinary. Light
propagating perpendicular to the optical axis is characterized by the main ex-
traordinary refractive index. Crystals with an optical axis, or uniaxial crystals,
are birefringent. If all three eigenvalues of the permittivity tensor differ, one
speaks of a biaxial crystal.

Cotton-Mouton effect A quasi-static external magnetic field, or the satu-

ration magnetization of a ferro- resp. ferri-magnetic crystal, also contributes
in second order to the susceptibility (Cotton-Mouton effect). The suscepti-
bility change cm 02
ij = ijk` Mk M` is not bound to special crystal properties,
02
ijk` ij k` is allowed. Because it is quadratic in the magnetization, the
Cotton-Mouton effect is reciprocal, in contrast to the Faraday effect.

Dissipation-fluctuation theorem If a system is disturbed by a time depen-

dent addition -(t)V to the Hamiltonian, the absorption of energy is described
by the imaginary part of the corresponding susceptibility. Fluctuations of the
equilibrium state are measured by the spectral density. Both functions, the
dissipative part of the susceptibility and the spectral density, are intimately
related, because of the Kubo-Martin-Schwinger identity. In particular, the
Wiener-Khintchin theorem allows the proof of a weak form of the second law
of thermodynamics.

Faraday effect An external quasi-static magnetic field B causes a change of

the susceptibilty by feij = iijk gk , where the gyration vector g (see gyrotropy)
depends linearly on the magnetic field strength, or the saturation magnetiza-
tion. If linearly polarized light passes a medium with Faraday effect, its po-
larization vector is rotated by an angle L, where L is the path length within
the medium and the specific Faraday rotation constant. If the light beam is
reflected and passes the medium a second time, but backward, the polarization
vector is rotated by the same amount, such that both angles add. This property
allows to build an optical isolator.

Gibbs state The equilibrium states of matter G(T, ) exp H()/kB T , or

Gibbs states, depend on temperature T and on external parameters appearing
in the Hamiltonian H. Besides chemical composition these may be mechanical
56 C GLOSSARY

stress, an external electric field, or an external magnetic field. If the exter-

nal parameters or the temperature of the systems environment change slowly
enough, then Wt G(Tt , t ) is a good approximation. The state Wt is always
very close to an equilibrium state, and we speak of a reversible process. If the
external parameters vary rapidly, the system is always away from equilibrium,
and we speak of an irreversible process.

Gyrotropy A gyrotropic contribution to the susceptibility is formally de-

scribed by gy ij = iijk gk . g is the gyrotropy vector. If light propagates in
the direction of g, the eigenmodes are circularly polarized waves with different
refractive indices. When a linearly polarized wave enters a gyroscopic medium,
it is split into circularly polarized waves which propagate with different phase
velocities. If this beam leaves the medium, the circularly polarized waves are
recombined into a linearly polarized wave. The polarization vector of this out-
going wave is rotated by an angle L, where L is the path length within the
gyroscopic medium. Note that the sign of rotation, right (clockwise) or left
(anti-clockwise), is judged by an observer facing the light source. Optical ac-
tivity and the Faraday effect cause gyrotropy.

Hall effect If charged particles within a solid move in a static magnetic field,
the current contains a component which is proportional to the driving elec-
tric field, proportional to the quasi-static induction field, and perpendicular
to both. These properties are a consequence of Onsagers theorem for kinetic
coefficients, such as the tensor of electric conductivity. The sign of the Hall
constant depends on whether electrons or electron wholes dominate the charge
transport mechanism.

Interaction picture The state W is defined by preparing the system un-

der study in a well-described manner. An observable M is a class of equiva-
lent measuring procedures which, for all states, show the same results. Both
are represented, in conventional quantum theory, by self-adjoint linear oper-
ators mapping a Hilbert space into itself. Time enters the game as follows.
After preparing the state W and before measuring M , one may wait for a
time t. This either defines a state Wt (Schrodinger picture) or an observable
Mt (Heisenberg picture). Waiting is described by the unitary waiting operator
Ut = exp (itH/~), an exponential function of time t and an observable H, the
energy, or Hamiltonian.
The interaction picture is in-between. Often the Hamiltonian is the sum of
a manageable part H and a perturbation V . It is advisable to resort to the
Heisenberg picture with H. Then the Schr odinger equation for states is driven
by V only, giving rise to a power series expansion in V .

Kerr effect The second order electro-optic Kerr effect ke 20

ij = ijk` Ek E` is
allowed even for otherwise isotropic media. It is usually much weaker than the
Pockels effect, if present.
 57

Kramers-Kronig relation The Hermitian part ij0 of the susceptibility ten-

sor describes refraction, the anti-Hermitian contribution iij00 absorption. The
refractive part at a certain angular frequency is an integral over all frequencies
u of the absorptive part, weighted by 1/( u). There is no refraction without
absorption, although possibly in another frequency domain. And: refractive
indices depend necessarily on angular frequency. The Kramers-Kronig relation
is a consequence retarded response functions.

Kubo-Martin-Schwinger identity Formally, the Gibbs state is equivalent

to the time translation operator with the inverse temperature as an imaginary
time. This results in hBA(z)i = hA(z i~)Bi for an arbitrary imaginary
time z. A and B are observables, A(t) is the time-translated observable, and
A(z) a proper analytic continuation. The KMS identity is an important step
in deriving the dissipation-fluctuation theorem.

Linear response A system, originally in an equilibrium, or Gibbs state, may

be perturbed by an explicitly time-dependent addition (t) V to the otherwise
constant Hamiltonian H. The equation of motion and the initial condition may
be formulated as an integral equation which can be expanded into a power
series in . The first non-trivial contribution describes the response of the
system to such perturbations which is linear in the driving force = (t).
By construction, the linear response, as described by response functions, is
retarded.

Onsager relations A symmetry relation for the generalized susceptibility

tensor based on the invariance of physical laws with respect to time reversal.
The susceptibility tensor is symmetric provided the direction of an external
magnetic field is reversed. It can be extended to the tensor of kinetic coefficients
describing the linear relationship between generalized fluxes and driving forces.
Onsager relations are relevant for the tensorial properties of various effects, such
as birefringence, the Pockels effect, Kerr effect, Faraday effect, Cotton-Mouton
effect, optical activity, and the Hall effect.

Optical activity The susceptibility depends on angular frequency and on the

wave vector, ij = ij (, q). The dependence on q is weak, and ij () suffices
for most applications. In some cases, however, the first order of an expansion
in q will result in an detectable effect, namely in a rotation of the polarization
vector along the propagation path. This gyrotropic effect is known as optical
activity. It will occur only if the material is left-right hand asymmetric, such
as quartz or a solution of dextrose.

Optical axis A direction such that a beam propagating along it has a re-
fractive index which does not depend on polarization. For optically isotropic
media, any direction is an optical axis. If two eigenvalues of the permittivity
tensor coincide, but differ from the third, there is but one optical axis. If all
three eigenvalues are different, one has two optical axes.
58 C GLOSSARY

Optical isolator A bulk optical isolator consists of a polarizer, a crystal with

Faraday effect rotating the polarization by 45 , and another polarizer being
rotated by 45 with respect to the first. If light which has passed the isolator
is reflected and possibly de-polarized, the polarization of the reflected wave
passing the second polarizer is rotated by another 45 ; it is therefore blocked
by the first polarizer. Optical isolators are required to protect a laser from its
own light. An integrated optical isolator is the goal of intensive research effort.

Pockels effect A linear opto-electric effect pe 10

ij = ijk Ek is possible for
ferro-electric or piezo-electric crystals which possess no inversion symmetry cen-
ter. This Pockels effect allows to switch or modulate light beams. It is usually
much stronger than the Kerr effect.

Refractive index Plane harmonic waves are characterized by an angular

frequency and a wave vector q. With w (a unit vector) as the direction
of propagation and k0 = /c one may write q = n k0 w. This defines the
refractive index n of the wave. If the susceptibilty, consequently the permittivity
tensor ij = ij + ij is real and symmetric (no absorption, no gyrotropy), one
may choose a Cartesian coordinate system such that becomes diagonal. The
square roots of the diagonal entries (eigenvalues) are sometimes called main
refractive indices. They are refractive indices if the wave is polarized along the
corresponding main axis.

Response function If Ht = H (t)V drives a process, than the expecta-

tion value of an observable M at time t is the equilibrium expectation value
plus a retarded integral over the driving force = (t). The Greens, or re-
sponse function ( ) is proportional to the equilibrium expectation value of
the commutator [M ( ), V ]. The Fourier transform over positive ages is a
susceptibility.

Second harmonic generation The quadratic response of a medium to per-

turbations of angular frequency produces a polarization of double frequency.
Only crystals without inversion center show this effect, such as lithium niobate.
Efficient frequency doubling requires phase matching.

Susceptibility If the Gibbs, or equilibrium state is disturbed by a time de-

pendent addition (t)V to the Hamiltonian, the expectation values of ob-
servables M vary with time, at lowest order with the same frequency as the
perturbation. The Fourier transform of hM it hM i is linearly related with
the susceptibility = () being the constant of pro-
the Fourier transform ,
portionality. Susceptibilities are Fourier transforms of corresponding response
functions ( ) over positive ages .

Wiener-Khintchin theorem The process t M (t) in a stationary state

is described by the time auto-correlation function K( ) indicating how much
 59

M ( ) is still correlated with M (0). The Fourier transform of K( ), the spectral

density S(), is never negative. A very short correlation time results in an
almost constant spectral density which qualifies the process t M (t) as white
noise. The Wiener-Khintchin and the dissipation fluctuation theorem allow to
prove the second law of thermodynamics.