ELECTRODYNAMICS

A course for the 2nd year of the Bachelor in Physics

by

Ruth Durrer
Département de Physique Théorique,
Université de Genève

Spring Semester, 2016

Translated from the French version by Manuel Vidal Vielma Blanco

2 Section 0.0

.
Contents

1 Maxwell equations 5

1.1 Microscopic and Macroscopic Maxwell equations and the Lorentz

force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

1.1.1 Macroscopic equations . . . . . . . . . . . . . . . . . . . . . 6

1.1.2 Derivation of the macroscopic equations . . . . . . . . . . . 9

1.1.3 Some simple solutions . . . . . . . . . . . . . . . . . . . . . 11

1.2 Units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

1.3 The Green method . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

1.4 Potentials and Gauge transformations . . . . . . . . . . . . . . . . . 22

1.5 Conservation laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

1.6 The relativistic formalism of electrodynamics . . . . . . . . . . . . . 28

2 Electromagnetic waves 47

2.1 One-dimensional waves . . . . . . . . . . . . . . . . . . . . . . . . . 47

2.2 Three-dimensional waves, electromagnetic waves . . . . . . . . . . . 50

3 Propagation of electromagnetic waves 61

3.1 Introduction, ponderable media . . . . . . . . . . . . . . . . . . . . 61

3.2 Reflection and refraction . . . . . . . . . . . . . . . . . . . . . . . . 63

3.3 Optics of metals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

3
4 Section 0.0

3.4 Dispersion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

3.5 General properties of the dielectric “constant” . . . . . . . . . . . . 78

3.6 Kramers-Kronig dispersion relations . . . . . . . . . . . . . . . . . . 80

3.7 Electromagnetic waves in dispersive media . . . . . . . . . . . . . . 86

3.8 The optical limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

4 Emission of electromagnetic waves 96

4.1 Wave zone, multipole expansion . . . . . . . . . . . . . . . . . . . . 96

4.2 Dipole fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

4.3 The field of a moving point charge . . . . . . . . . . . . . . . . . . . 103

4.4 Cherenkov radiation . . . . . . . . . . . . . . . . . . . . . . . . . . 110

5 Scattering of electromagnetic waves 113

5.1 Thomson scattering . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

5.2 Elastic and inelastic scattering by quasi-free charges . . . . . . . . . 115

5.3 Scattering in gases and liquids . . . . . . . . . . . . . . . . . . . . . 120

5.4 Diffraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125

Chapter 1

Maxwell equations

Preamble

“Höchste Aufgabe der Physiker ist das Aufsuchen jener allgemeinsten, elementaren
Gesetze, aus denen durch reine Deduktion das Weltbild zu gewinnen ist. Zu diesen
elementaren Gesetzen führt kein logischer Weg, sondern nur die auf Einfühlung in
die Erfahrung sich stützende Intuition...”

(A. Einstein, excerpt from his address at the Physical Society, Berlin, for Max
Planck’s 60th birthday)

In English (approximately):

“The supreme task of the physicist is to arrive at those universal elementary laws
from which the cosmos can be built up by pure deduction. There is no logical path
to these laws; only intuition, resting on sympathetic understanding of experience,
can reach them”

I would like you to realise from this quotation that there are no genuine derivations
of Maxwell’s laws: even if driven by some of them, these relations are not totally
implied by any experiment whatsoever. In this course we will in fact postulate
Maxwell’s equations as the point of departure, and we will show thereafter that
these laws describe phenomena very well.

As Einstein puts it, finding such sort of laws is the creative act of the highest rank
of theoretical physics. This has happened no more than five or six times in the
history of physics:

• Gravitation (Newton)

5
6 Section 1.1

• Thermodynamics (Clausius, Carnot)

• Electrodynamics (Maxwell, Faraday)

• Statistical mechanics (Boltzmann, Gibbs, Einstein)

• General relativity (Einstein)

• Quantum mechanics (Einstein, Schrödinger, Planck, Pauli, Heisenberg, Bohr,

Dirac,...)

1.1 Microscopic and Macroscopic Maxwell equa-

tions and the Lorentz force

1.1.1 Macroscopic equations

The Lorentz force determines the action of the electromagnetic field on a point
charge q with velocity v,
 
1
K= q E+ v∧B . (1.1)
c

Here E and B are the electric and the magnetic field and c is the speed of light
(as we shall see later). If we define the charge density (charge per unit volume),
ρ = dq/dV , as well as the current density, J = d(qv)/dV , and use the symbol k to
denote the force per unit volume, we then obtain:

1
k = ρE + J ∧ B. (1.2)
c

The fields E and B themselves are originated by the charge distribution. The
relation between the fields and the charges and currents is determined by Maxwell’s
equations, the homogeneous version of which are the expressions
X ∂B 1 ∂B 2 ∂B 3
divB ≡ ∇ · B ≡ ∂i B i ≡ + + = 0, (1.3)
i
∂x1 ∂x2 ∂x3

which tells us that there are no magnetic charges;

1 1 X 1
rotE + Ḃ ≡ ∇ ∧ E + Ḃ = εiℓm ∂ℓ Em + Ḃi = 0, , (1.4)
c c ℓm
c
Ruth Durrer Electrodynamique Chap. 1 7

∂
that expresses the law of induction, with · ≡ ∂t and
n
sign of the permutation iℓm → 123, if iℓm are all different
εiℓm =
0 otherwise.
The two other equations are, on the one hand, Coulomb’s law,

divD ≡ ∇ · D = 4πρ (1.5)

and, on the other, Ampère’s law with the displacement current,

4π 1
rotH ≡ ∇ ∧ H = J + Ḋ. (1.6)
c c

Here D and H are the electric and magnetic fields in a ponderable medium, and ρ
and J are the macroscopic charge and current densities.

Using Gauss and Stokes’ theorems, Maxwell equations can be rewritten as integral
equations:

• Gauss’ theorem: Z Z
(∇ · W)dv = (W · e) ds (1.7)
V ∂V

Here V is an arbitrary (finite) volume of boundary ∂V . W is a vector field,

also arbitrary, and e is the outward pointing unit normal vector to the surface
element ds.

• Stokes’ theorem: Z I
(∇ ∧ W) · e ds = W · n dl (1.8)
S ∂S

Here S is some (bounded) surface and ∂S is its boundary. W and e are as

before and n is the unit vector parallel to ∂S.

These mathematical theorems allow then to rewrite (1.5) as an integral equation,

Z
(D · e) ds = 4πQ (1.9)
∂V

where Q is the total charge in the interior of the volume V . The flow of D through
the boundary of a volume V determines the total charge in the interior. Similarly,
Eq. (1.3) gives Z
(B · e)ds = 0 . (1.10)
∂V
There are no magnetic charges.
8 Section 1.1

With (1.8) and the second Maxwell equation, (1.4), we find

I Z
d
E · ndl = − (B · e)ds
∂S dt S

which is nothing but Faraday’s law, also known as law of induction: a magnetic
field that changes along a surface delimited by a metallic wire induces a current in
the wire.

The fourth Maxwell equation gives

I Z Z
4π 1d
H · ndl = (J · e)ds + (D · e)ds (1.11)
∂S c S c dt S

The first two terms of this equation are just Ampère’s law. The last term, usually
negligible ( 1c factor), is the famous ’displacement current’.

When, by interpreting the experimental results—particularly those of Faraday—

Maxwell arrived for the first time at these relations, they were all stationary laws,
with the only exception of the one of induction. The displacement current (the
term 1c Ḋ) was yet to appear. Maxwell was well aware of the fact that, without
that term, the equations were not in agreement with the conservation of electric
charge. To get the final form (1.6) he therefore invented, based on complicated
and dubious arguments, the reaction of an ether (carrier of the electromagnetic
field) on the currents, which he named displacement current.

For us, however, the term displacement current is nothing more than a name,
and Ampère’s law with the displacement current is justified by its agreement with
experiments (which, by the way, were not available at Maxwell’s era; in order to
measure the displacement current, very rapidly changing electric fields are needed).

In vacuum, E = D and B = H. For ponderable media one has, in addition, some

phenomenological relations that link E to D and B to H. In their simplest form,
these are linear and isotropic relations with constants ε and µ:

1
D = εE and H = B, (1.12)
µ

where ε and µ are known as the dielectric constant and the magnetic permeability.

In a good conductor one also has

J = σE , σ is the conductivity. (1.13)

Ruth Durrer Electrodynamique Chap. 1 9

1.1.2 Derivation of the macroscopic equations

To remind you the relation between the microscopic and macroscopic fields, we
proceed to derive the equation D = E + 4πP (for further details see Jackson,
§6.7).

The microscopic laws, the authentically fundamental ones, are

∇·b= 0 ∇ ∧ e + 1c ḃ = 0 (homogeneous eqs.) (1.14)

4π
∇ · e = 4πη ∇ ∧ b − 1c ė = j (inhomogeneous eqs.). (1.15)
c
Here (e, b) are the fields and (η, j) are the microscopic charge and current densities,
which vary over length scales down to atomic distances (∼ 10−8 cm) (or less) and
time scales as short as 10−17 s (orbital motion of an electron). The fields E and B
are averaged
E = hei, B = hbi, (1.16)
where we define Z
hai(x) := f (x − y)a(y)d3 y (1.17)
R
with f (y)d3y = 1. Here f is a function that averages over a region of size
R ∼ 10−4 cm. For example

f (x) = 1 e−|x|2 /R2 (Gaussian)

3
(πR2 ) 2
or
 (1.18)
3 , |x| < R

3
f (x) = 4πR (top hat).
 0, |x| > R

For the derivatives of the macroscopic field, one finds the following

Z Z
∂ ∂
∂i Ej = f (x − y)ej (y)d3y = − ( f (x − y))ej (y)d3 y
∂xi ∂y i
Z
∂
= f (x − y) ej (y)d3 y = h∂i ej i.
∂y i
Where the second line is reached after integrating by parts. The derivatives and
the average commute. The macroscopic fields, E and B, satisfy therefore the
homogeneous equations (1.14).
10 Section 1.1

In order to determine the inhomogeneous macroscopic equations, it is necessary

to take the average of both the charge and current densities. The total charge
density can be split up into a density of free charges ηfree , and a density of the
charges of the nuclei and the electrons bound to the molecules, ηbound . We thus
have, hηi = hηfree i + hηbound i, where
XX
ηbound (x) = qℓn δ(x − xℓn ) (1.19)
n ℓn

is the summation over all the molecules (n) and over all the charges (ℓn ) in a
given molecule. Let xn be the center of charge of the molecule n. We then have
xℓn = xn + ∆xℓn , with |∆xℓn | ≪ |xn |. For the average hηbound i we obtain
XX Z
hηbound i(x) = qℓn d3 yf (x − y)δ(y − xn − ∆xℓn )
n ℓn
XX
= qℓn f (x − xn − ∆xℓn ).
n ℓn

Let us now Taylor expand this expression in terms of the small quantity ∆xℓn :
XX
hηbound i(x) = qℓn f (x − xn ) − qℓn ∆xℓn · ∇f (x − xn ) + · · · . (1.20)
n ℓn
P P
With qn := ℓn qℓn and pn := ℓn qℓn ∆xℓn the charge and dipole moment of the
n-th molecule, we find:
X
hηbound i(x) = (qn f (x − xn ) − pn · ∇f (x − xn ) + ...) .
n

The first term is the average of a point charge qn located at xn , and the second
term is the average of a point dipole pn at xn :
X
hηbound i(x) = hηn i
n

hηn i = hqn δ(x − xn )i − ∇ · hpn δ(x − xn )i + ...

Defining the macroscopic charge density as

X
ρ(x) := hηfreei + hqn δ(x − xn )i
n

and the polarisation as X

P(x) := hpn δ(x − xn )i
n

it is found that
hηi = ρ − ∇ · P
Ruth Durrer Electrodynamique Chap. 1 11

If we insert here the microscopic Coulomb equation (the first of equations (1.15)),
we obtain:
∇ · E = 4π(ρ − ∇ · P). (1.21)
Making
D = E + 4πP (1.22)
this gives the macroscopic Coulomb equation (1.5). To the lowest order, P is a
linear function of E, and P is proportional to E if the medium is isotropic. Hence

P = χe E, D = (1 + 4πχe )E = εE. (1.23)

χe is the electric susceptibility and ε is the dielectric constant of the medium. The
quadripolar term that we have neglected by writing ... is indeed almost always
very small.

An analogous procedure leads to the following expression for the average of the
current density

hji = J + c∇ ∧ M + Ṗ + ...

(the rather lengthy derivation of this result can be found in Jackson, §6.7).

Here M is the magnetisation, which is usually proportional to B, M = χm B.

Writing
1
H = B − 4πM = (1 − 4πχm )B = B, (1.24)
µ
we arrive at the macroscopic Ampère’s law, (1.6).

If µ > 1, χm > 0 and M is parallel to B, we speak of paramagnetism. If, instead,

µ < 1, χm < 0 and M is antiparallel to B, we speak of diamagnetism. Typically,
the magnetic susceptibility, χm , is very small, so much as to have |µ − 1| ∼ 10−5 . µ
is known as the magnetic permeability. In the case of ferromagnetism the relation
between B and H is more complicated because it is time-dependent and non-
linear (hysteresis phenomena). (Ferromagnetic materials are paramagnetic, but
they present in addition some spontaneous magnetisation, which means that the
magnetisation M does not vanish when the field B is off).

1.1.3 Some simple solutions

Electrostatics

We consider the case D = E and Ė = Ḃ = 0. Faraday equation (1.4) implies

then ∇ ∧ E = 0, and so there exists a function φ (the electrostatic potential) with
12 Section 1.1

E = −∇φ, such that

3
X
∆φ = −4πρ , where ∆ = ∂i2
i=1

(Poisson equation). The solution of this equation (which decreases for r → ∞ if

the source is contained in a finite volume) is given by

Z Z
3 ′ρ(x′ ) ρ(x′ )(x − x′ )
φ(x) = dx and E= d3 x′ . (1.25)
|x − x′ | |x − x′ |3

Magnetostatics

Once again, we consider D = E and H = B and Ė = Ḃ = 0. From eq. (1.3) we

conclude that B can be written in the form B = ∇ ∧ A. Here A is the magnetic
vector potential. Additionally, we can choose A such that ∇ · A = 0. This choice
is referred to as the Coulomb gauge.

Exercise: Show that for any vector field V

∇ ∧ (∇ ∧ V) = ∇(∇ · V) − ∆V

For Ė = 0 we then obtain, from Ampère’s law, that

4π
−∇ ∧ (∇ ∧ A) = −∇(∇ · A) + ∆A = ∆A = − J
c
The solution of this equation (which also decreases for r → ∞ if the source is
contained in a finite volume) is given by

Z
1 J(x′ )
A(x) = d3 x′ . (1.26)
c |x − x′ |
Equations (1.25) and (1.26) determine the electro- and magneto-static potentials
and, consequently, the fields E and B for a given static distribution of charges and
currents.

Maxwell equations in vacuum

In vacuum, ρ = 0, J = 0, and Maxwell equations reduce to

1
Ė = ∇ ∧ B (1.27)
c
Ruth Durrer Electrodynamique Chap. 1 13

∇·E = 0 (1.28)
1
Ḃ = −∇ ∧ E (1.29)
c
∇ · B = 0. (1.30)

If we differentiate (1.27) with respect to time and introduce (1.29) for Ḃ, we find
1 1
2 Ë = ∇ ∧ Ḃ = −∇ ∧ (∇ ∧ E) = ∆E − ∇(∇ · E}).
| {z
c c
=0

Therefore
1
Ë − ∆E = 0, (1.31)
c2
Taking the derivative with respect to time of (1.29) and replacing Ė by (1.27),
we obtain the same equation (1.31) for B. Note that the sign difference between
(1.27) and (1.29) is responsible for the appearance of the minus sign in (1.31). We
finally arrive at the wave equations

(∆ − 12 ∂t2 )E = 0 
c . (1.32)
(∆ − 12 ∂t2 )B = 0 
c

As we will see in more detail in the chapters to come, the Maxwell equations in
vacuum describe waves that travel at the speed c (see (1.32)).

Exercise: Show that the ansatz

E = E0 f (ct ± n · x)
B = B0 g(ct ± n · x)

with n2 = 1, and E0 , B0 and n constant vectors, solves equations (1.32). Show

that eqs. (1.27) to (1.30) are all also satisfied if and only if f = g, n · E0 = 0 and
B0 = ±n ∧ E0 .

A first consequence is that the speed c at which the fields E and B propagate is
finite and hence there is no action at a distance in electrodynamics. If a charge
located at A moves, a time δt = dc will elapse before this modification is received by
an antenna at B at a distance d. Maxwell predicted (after finding equations (1.32))
that light is an electromagnetic phenomenon, something that has been definitely
verified. Hertz was the first to detect electromagnetic waves experimentally.

As we will see later on, the electromagnetic field carries energy as well as linear
and angular momenta in vacuum, the transport of which takes place at the speed
14 Section 1.2

of light, c, a constant of nature. If Maxwell’s equations are valid in all systems of

reference, then the speed c is the same in all frames. It is precisely this remark
that led Einstein to the theory of special relativity (his original 1905 article was
entitled “Zur Elektrodynamik bewegter Körper”).

Before entering fully into our subject, I would like to discuss once and for all the
issue of units.

1.2 Units

(See Jackson, Appendix)

You have probably noticed that Maxwell’s equations (1.3–1.6), as well as the rela-
tions for the fields H and D, (1.23) and (1.24), are not exactly equal to those you
had already encountered in the first year. The reason for this is that I have adopted
the Gaussian system of units, whereas in your previous year you had expressed
these equations in MKSA units. For fundamental problems and specially later for
quantum electrodynamics, the Gaussian (or Heaviside-Lorentz) units prove to be
better adapted. For engineering problems of a more practical nature, however,
MKSA units turn out to be very well suited. It is therefore convenient to know
both systems (and perhaps even others).

In principle, there is absolute freedom when choosing the units, but it is very
important to be aware of and benefit from the choice of units that is better adapted
to the problem under study. In physics, it makes no sense to state that the outcome
of a measurement of a length is 2. It is imperative to specify if one is speaking of
2 mm, 2 inches, 2 km, 2 light years, etc.

In elementary particle physics, it is customary to set c = 1 and ~ = 1. There then

remains a basic unit that is in general chosen to be the energy (e.g. 1 electron-
Volt, 1eV). From c = 1 it is easy to deduce that time has the same dimension as
length and that mass has the same dimension as energy and momentum. From
i~ = [x, p] = i (see quantum mechanics) it follows that energy, momentum and
mass have all the dimension of inverse length and inverse time.
Ruth Durrer Electrodynamique Chap. 1 15

In electrostatics, the Coulomb force between two charges q and q ′ in rest at a

distance r is given by
qq ′
F = k1 2 . (1.33)
r
The constant k1 must be chosen. The dimension and magnitude of this constant
determines the units of the charges. It is possible and consistent to set k1 = 1.
If one starts from the fundamental dimensions l (length), t (time) and m (mass),
we obtain, in the case where k1 is dimensionless, the following dimension for the
charge:
 1/2


2 1/2 ml 2 m1/2 l3/2
[q] = [F ]l = l = . (1.34)
t2 t

In the system used in particle physics, where t has the same dimension as l and
m has the inverse dimension, the charge is dimensionless. It is a pure number, the
coupling constant, which determines the strength of the electromagnetic interac-
tion:
k1 e2 1
α= = k1 e2 ∼
= is the ‘fine structure constant’
~c 137
where e is the electron charge. In units where k1 = ~ = c = 1, the fine structure
constant is simply α = e2 .
k q 1/2
The electric field is defined by E = F′ = 12 , and thus has the units [E] = m1/2
q r tl
in a system with k1 = 1, something adopted in the Gaussian system. hObserve i that
m mℓ 2
in this case [E2 ] = t2 l , i.e. E2 has the dimensions of energy density, t2 ℓ3 .

Analogously, Ampère’s law determines the force between two stationary currents
I and I ′ at a distance d,
dF k2 II ′
=2 . (1.35)
dl d
The constants k1 and k2 are not independent. Since [I] = [q]/t, it follows that
[k1 /k2 ] = l2 /t2 . The quantity k1 /k2 can be measured and one finds k1 /k2 = c2 ,
where c is the speed of light,
k1
= c2 . (1.36)
k2
The magnetic induction B can be derived from Ampère’s law (1.6). A long and
straight metallic wire carrying a current I induces a B at a distance d given by:
I
B = 2k2 β . (1.37)
d
The constant β determines the difference between the dimensions of E and B.
Equations (1.33) and (1.37) lead to:
   
E k1 qtl q
= (using I = t ).
B l2 k2 βq
16 Section 1.2

2
With k1 = l 2 this gives
k2 t

   
E l
= .
B tβ

If we now write the law of induction in the form ∇ ∧ E + k3 ∂t B = 0, it is clear

that k3 has the same dimension as β −1. Indeed, k3 = β −1 . The simplest way to
verify this is to write Maxwell’s equations with our constants:


∇ · E = 4πk1 ρ 




k 
∇ ∧ B − β∂t E = 4πk2 βJ 
2
k1 . (1.38)
∇ ∧ E + k3 ∂t B = 0 





∇·B = 0 

In a region where ρ = 0 and J = 0, these equations lead to

k2 β 2
∆B − k3 ∂ B = 0. (1.39)
k1 t

kkβ
Hence 3 2 = 12 and, with (1.36), k3 β = 1. (The values of k1 , k2 , β and k3 in
k1 c
different systems are shown in table I).
Ruth Durrer Electrodynamique Chap. 1 17

Magnitudes and Dimensions of the Electromagnetic Constants for Various

Systems of Units

The dimensions are given after the numerical values. The symbol c stands for the velocity
of light in vacuum (c = 2.998 × 1010 cm/sec = 2.998 × 108 m/sec). The first four systems of
units use the centimeter, gram, and second as their fundamental units of length, mass, and
time (l, m, t). The MKSA system uses the meter, kilogram, and second, plus current (I) as
a fourth dimension, with the ampere as unit.
System k1 k2 β k3

Electrostatic
(esu) 1 c−2 (t2 l−2 ) 1 1
Electromagnetic
(emu) c2 (l2 t−2 ) 1 1 1

Gaussian 1 c−2 (t2 l−2 ) c(lt−1 ) c−1 (tl−1 )

1 1
Heaviside-Lorentz (t2 l−2 ) c(lt−1 ) c−1 (tl−1 )
4π 4πc2
1 µ0
Rationalized MKSA = 10−7 c2 ≡ 10−7 1 1
4πε0 4π
(ml3 t−4 I −2 ) (mlt−2 I −2 )

Table I (Table 1 of Jackson, Appendix)

Furthermore, for the relations between the fundamental fields (E, B) and the fields
(D, H) we can still choose some constants

D = ε0 E + λP,
1
H = B − λ′ M.
µ0

There is no point in assigning different dimensions to D and P or to H and M.

Indeed, I do not know of any unit system where λ and λ′ are dimension-full. The
systems with λ = λ′ = 1 are called rational systems. In non-rational systems one
has λ = λ′ = 4π. The constants ε0 and µ0 are the dielectric constant and the
vacuum permeability. In linear and isotropic media one has
1
D = εE, H= B,
µ

The relative permittivity (relative dielectric constant) is εε0 and the relative per-
µ
meability is µ0 .
18 Section 1.2

In table II you can find the values of ε0 and µ0 in different unit systems as well
as the corresponding forms of the macroscopic Maxwell’s equations and of the
Lorentz force. In table III you can find the names and conversion factors between
the rationalised MKSA units and the Gaussian units to be used in this course.
Note that in the Gaussian system of units, E and B have the same dimension, a
fact that will play an important role for the relativistic and explicitly covariant
formulation of the electromagnetic field.
Ruth Durrer Electrodynamique Chap. 1 19
20 Section 1.2

Conversion Table for Given Amounts of a Physical Quantity

The table is arranged so that a given amount of some physical quantity, expressed as so many
MKSA or Gaussian units of that quantity, can be expressed as an equivalent number of units
in the other system. Thus the entries in each row stand for the same amount, expressed in
different units. All factors of 3 (apart from exponents) should, for accurate work, be replaced
by (2.99792456), arising from the numerical value of the velocity of light. For example, in
the row for displacement (D), the entry (12π × 105 ) is actually (2.99792 × 4π × 105 ). Where
a name for a unit has been agreed on or is in common usage, that name is given. Otherwise,
one merely reads so many Gaussian units, or MKSA or SI units.

Physical Quantity Symbol Rationalized MKSA Gaussian

Length l 1 meter (m) 102 centimeters (cm)
Mass m 1 kilogramm (kg) 103 grams (gm)
Time t 1 second (sec) 1 second (sec)
Frequency ν 1 hertz (Hz) 1 hertz (Hz)
Force F 1 newton 105 dynes
Work W
1 joule 107 ergs
Energy U
Power P 1 watt 107 ergs sec−1
Charge q 1 coulomb 3 × 109 statcoulombs
Charge density ρ 1 coul m−3 3 × 103 statcoul cm−3
Current I 1 ampere (amp) 3 × 109 statamperes
Current density J 1 amp m−2 3 × 105 statamp cm−2
1
Electric field E 1 volt m−1 3 × 10
−4
statvolt cm−1
1
Potential Φ, V 1 volt 300 statvolt
Polarization P 1 coul m−2 3 × 105 dipole
moment cm−3
Displacement D 1 coul m−2 12π × 105 statvolt cm−1
(statcoul cm−2 )
Conductivity σ 1 ohm−1 m 9 × 109 sec−1
1 −11
Resistance R 1 ohm 9 × 10 sec cm−1
Capacitance C 1 farad 9 × 1011 cm
Magnetic flux φ, F 1 weber 108 gauss cm2 or
maxwells
Magnetic induction B 1 tesla 104 gauss
Magnetic field H 1 ampere-turn m−1 4π × 10−3 oersted
Magnetization M 1 ampere m−1 10−3 magnetic
moment cm−3
1
Inductance L 1 henry 9 × 10−11

Table III (Table 4 of Jackson, Appendix)

Ruth Durrer Electrodynamique Chap. 1 21

1.3 The Green method

For a given linear differential operator (with constant coefficients) D, a solution G

of the equation
DG = δ
is said to be a Green function for the operator D. Here δ is the Dirac δ distribution
in the corresponding number of dimensions. If G is a Green function for D and ϕ
is a homogeneous solution – that is to say, if Dϕ = 0 – then G′ = G + ϕ is also
a Green function for D. And vice-versa, if G and G′ are Green functions, their
difference is a homogeneous solution. To determine the correct Green function for
a given problem, the boundary conditions must be specified.

We are interested in the wave (or D’Alambert) operator, D = ∂t2 − c2 ∆. We look

then for a solution of the equation

(∂t2 − c2 ∆)G(x, t) = δ 3 (x)δ(t) . (1.40)

We want a solution that decreases for r → ∞ and vanishes for t < 0. The last
condition assures causality: the field should not be present before the charge that
originates it.

To solve (1.40) we first perform a Fourier transform in the spatial coordinates,

Z
Ĝ(k, t) = d3 xei(k·x) G(x, t)
Z
1
G(x, t) = d3 ke−i(k·x) Ĝ(k, t)
(2π)3
R
With δ 3 (x)ei(k·x) = 1, eq. (1.40) becomes

(∂t2 + c2 k2 )Ĝ(k, t) = δ(t) . (1.41)

It is easily verified that (see exercices)

sin(ckt)
ĜR (k, t) = H(t) (1.42)
ck
is a solution of (1.41). Here H(t) is the Heaviside function,

0 if t < 0
H(t) =
1 if t ≥ 0 .

If now we use the fact that the Fourier transform of δ(|x| − R) is 4πR sin(kR)/k,
we obtain
H(t) H(t)
GR (x, t) = 2
δ(ct − |x|) = δ(ct − |x|) . (1.43)
4πc t 4πc|x|
22 Section 1.4

This Green function clearly obeys the required boundary conditions. It is possible
to show that it is the only Green function with that property1 .

The solution to the problem (∂t2 − c2 ∆)φ(x, t) = f (x, t) is then (see exercises)
Z Z
3 ′ ′ ′ ′ ′ ′ 1 f (x′ , t − |x − x′ |)
φ(x, t) = d x dt GR (x−x , t−t )f (x , t ) = d3 x′ . (1.44)
4π |x − x′ |

1.4 Potentials and Gauge transformations

(See Jackson §6.5)

You already know from magnetostatics that the equation ∇·B = 0 is automatically
satisfied if one sets
B = ∇ ∧ A, (1.45)
since ∇ · (∇ ∧ A) ≡ 0 for any vector A. One can also show that any vector field v
in three dimensions can be split up into an irrotational part and a rotational part
(spin-0 component and spin-1 component),

v = −∇φ + ∇ ∧ a . (1.46)

φ and a can be chosen to be solutions of the following equations


∆φ = −∇ · v 
. (1.47)
∆a = −∇ ∧ v 

Notee that a and φ are not unique unless ∇ · a = 0 and some boundary conditions
are imposed.

The law of induction,

1 1
∇ ∧ E + ∂t B = ∇ ∧ (E + ∂t A) = 0
c c
implies the existence of a scalar φ such that
1
E + ∂t A = −∇φ
c
1
From this one might be led to think that this one is the only Green function of any relevance
in physics. The situation, however, is a bit more intricate in quantum mechanics. The most
important Green function in quantum mechanics is Feynman’s one, which is a superposition
of the retarded Green function, GR , and the advanced Green function, GA , the latter being
obtainable from the former under the replacement of t − |x| with t + |x|.
Ruth Durrer Electrodynamique Chap. 1 23

or
1
E = −∇φ − ∂t A. (1.48)
c
With this ansatz, the homogeneous Maxwell equations are automatically satisfied.
In terms of our electromagnetic potentials (φ, A), the inhomogeneous Maxwell
equations (for the case ǫ = µ = 1) give
1
(∇ · E) = −∆φ − ∇ · Ȧ = 4πρ (1.49)
c
and
1 1 1 4π
(∇ ∧ B − ∂t E) = ∇(∇ · A) − ∆A + ∇φ̇ + 2 Ä = J. (1.50)
c c c c
The potentials φ and A are not uniquely fixed by conditions (1.45) and (1.48). If
we perform the transformation

A → A + ∇χ, (1.51)

the field B remains the same. If we farther replace φ with

1
φ → φ − χ̇, (1.52)
c
the field E is also invariant. Transformations (1.51) and (1.52) are known as gauge
transformations. Under a gauge transformation
1 1 1
(∇ · A + φ̇) → ∆χ − 2 ∂t2 χ + ∇ · A + φ̇
c c c
1 1
= (∆ − 2 ∂t2 )χ + ∇ · A + φ̇.
c c
Thus, if one finds a scalar field χ such that
1 2 1
(∆ − 2 ∂t )χ = −(∇ · A + φ̇), (1.53)
c c
then the transformed potentials,
1
à = A + ∇χ and φ̃ = φ − χ̇,
c
satisfy the equation
1
∇ · Ã + φ̃˙ = 0, (1.54)
c
which is known as the Lorentz gauge condition2 .
2
Since the wave equation for a given source always has solution (see eq. (1.44)), we can always
solve (1.53), at least locally.
24 Section 1.5

In the Lorentz gauge (1.54), Maxwell’s equations have a particularly simple form
because they are completely decoupled (we omit the tilde ∼):

1 2
(∆ − ∂ )φ = −4πρ (1.55)
c2 t
1 4π
(∆ − 2 ∂t2 )A = − J. (1.56)
c c

Equations (1.55) and (1.56) together with condition (1.54) are equivalent to Maxwell’s
equations (four 1st order equations have been transformed into two 2nd order equa-
tions).

From (1.43), equations (1.55) and (1.56) are solved by

Z Z
3 ′ ′ ′ ′ ′ ρ (x′ , t − |x − x′ |/c) 3 ′
φ(x, t) = d x dtGR (x − x , t − t )ρ(x , t ) = d x (1.57)
|x − x′ |

and
Z Z
3 ′ ′ ′ 1
′ ′ J (x′ , t − |x − x′ |/c) 3 ′
A(x, t) = d x dtGR (x − x , t − t )J(x , t ) = d x.
c |x − x′ |
(1.58)

1.5 Conservation laws

(See Jackson, §6.8)

Charge conservation

Maxwell’s equations imply

 
1 c
ρ̇ + ∇ · J = ∇ · Ḋ + J = ∇ · (∇ ∧ H) = 0. (1.59)
4π 4π

For an open set G ⊂ R3 we then have

Z Z
d 3
0 = (ρ̇ + ∇ · J)d x = Q(G) + J · e ds, (1.60)
G dt ∂G

which corresponds to the conservation of charge.

Ruth Durrer Electrodynamique Chap. 1 25

Energy conservation

By using
1
∇ ∧ E + Ḃ = 0 and
c
1 4π
∇ ∧ H − Ḋ = J,
c c
one can find
4π 1 1
E · (∇ ∧ H) − H · (∇ ∧ E) = E · J + E · Ḋ + H · Ḃ. (1.61)
c c c
But, in agreement with a simple identity, one has −∇ · (A ∧ B) = A · (∇ ∧ B) − B ·
(∇ ∧ A) and so the left-hand side of (1.61) corresponds to −∇ · (E ∧ H). Equation
(1.61) leads finally to
1
(E · Ḋ + H · Ḃ) + ∇ · S = −J · E, (1.62)
4π
where
c
S := E∧H (1.63)
4π
is the Poynting vector. The relation (1.62) holds in general. Let us now assume
that there is an instantaneous and linear relation between H and B and between
D and E such as, for instance, D = εE with ε time-independent (that is the case
for microscopic fields, where E = D and H = B). In that case we have that
E · Ḋ = 21 (E · D)· and H · Ḃ = 12 (H · B)· , and (1.62) can be written in the form

u̇ + ∇ · S = −J · E (1.64)

where
1
u := (E · D + H · B) .
8π
If we integrate this identity over an open set G ⊂ R3 , we obtain
Z Z Z
d 3
ud x + S · e ds = − J · E d3x. (1.65)
dt G ∂G G

The last term is minus the work done on the charges by the field. If the region G
and the electromagnetic fields are such that the surface integral can be neglected,
a decrease of the integral of u corresponds then to the work performed by the field.
This justifies the interpretation of u as the energy density of the electromagnetic
field.

The first term of the left-hand side of equation (1.64) is then the time derivative
of the electromagnetic energy, and the term to the right is the work per unit time
26 Section 1.5

performed by the electromagnetic field on the currents taken with the minus sign.
If J is parallel to E, the field loses energy because it performs work on the currents.
S must then be interpreted as the energy flux: equation (1.65) means therefore that
the energy of the electric field in G changes depending on the energy flux towards
the outside of G and on the work performed on the currents.

If the nonlinear effects are relevant, (1.65) is no longer valid for the macroscopic
fields and one has to turn to equation (1.62).

Conservation of momentum

The force density applied on the charges and currents by the electromagnetic field
is
1
k = ρE + J ∧ B.
c
We get rid of ρ and J by means of Maxwell’s equations (Coulomb and Ampère’s
laws) to obtain
1
4πk = E(∇ · D) + (∇ ∧ H − Ḋ) ∧ B.
c
To this equation we can add the vanishing term
1
0 = H(∇ · B) − D ∧ (∇ ∧ E + Ḃ)
c
which gives
1
4πk = [E(∇ · D) − D ∧ (∇ ∧ E)] + [H(∇ · B) − B ∧ (∇ ∧ H)] − ∂t (D ∧ B).
c
Let us first consider the macroscopic theory, where
D ≡ E et H ≡ B.
With Π := 12 S = 4πc1 (E ∧ B) we find that
c
1
Π̇ = −k + [E(∇ · E) − E ∧ (∇ ∧ E) + B(∇ · B) − B ∧ (∇ ∧ B)] . (1.66)
4π
Let us compute the i-th component of the expression
P for E inside [ ]. In what fol-
lows, we won’t write the summation symbol but will use instead the convention
according to which a sum from 1 to 3 is implied for any repeated indices in a term
(Einstein’s convention).
For example:
3
X
∇·A = ∂i Ai ≡ ∂i Ai
i=1
3
X
(∇ ∧ A)i = εijℓ ∂j Aℓ ≡ εijℓ ∂j Aℓ .
j,ℓ=1
Ruth Durrer Electrodynamique Chap. 1 27

We will also use the identities (the proof is left as an exercise)

εiℓm = εℓmi = εmiℓ ,

εiℓm εijk = δℓj δmk − δℓk δjm .

With this we have that

[E(∇ · E) − E ∧ (∇ ∧ E)]i = Ei ∂j Ej − εijk Ej εkℓm ∂ℓ Em

| {z }
(δℓi δjm −δim δjℓ )Ej

= Ei ∂j Ej − Ej ∂i Ej + Ej ∂j Ei
1
= ∂j [Ei Ej − δij E2 ].
2
Since the expression for B is completely analogous, we arrive at the relation
1
[E · (∇ · E) − E ∧ (∇ ∧ E) + B(∇ · B) − B ∧ (∇ ∧ B)]i =
4π
1 1 1
∂j (Ei Ej − δij E2 + Bi Bj − δij B2 ) = −∂j Tij
4π 2 2
with
1 1
Tij := [ δij (E2 + B2 ) − Ei Ej − Bi Bj ]. (1.67)
4π 2
Finally, equation (1.66) reduces to

Π̇i + ∂j Tij = −ki .

R
Using the fact that k = d P(mec) , one finds
G dt
Z  Z
3 d  (field) (mec)
(Π̇ + k)i d x = Pi + Pi = Tik ek ds. (1.68)
G dt ∂G

If this expression isR interpreted as the equation of conservation of momentum in an

open set G, then G Π d3 x represents the momentum of the electromagnetic field.
(field)
In the right-hand side, Tik ek is the flow per unit area of the field’s force, and Pi
is the i-th component of the momentum of the electromagnetic field in G. From
(1.63),
1 1
S = E ∧ H. (1.69)
c2 4πc
In a linear and isotropic medium where D = εE and H = µ 1 B with ε = const. and
µ = const., one arrives at equation (1.68) with Π and Tij given by

1 εµ
Π= D∧B = 2S
4πc c
28 Section 1.6

and
1 1
Tij = [ δij (E · D + B · H) − Ei Dj − Hi Bj ].
4π 2
If ε and µ are position-dependent (but time-independent!) the supplementary
1 E2 ∇ε and k = 1 H2 ∇µ must be added.
forces kE = − 8π M 8π
The conservation of angular momentum can be discussed along similar lines:

Z
d (field)
(L + L(mec) )i = − Mij ej ds
dt ∂G

or
∂t (L(field) + L(mec) )i = −∂j Mji . (1.70)
If the electric and magnetic susceptibilities can be neglected, ǫ ≡ µ ≡ 1, one finds
(1.70) with

(field) 1
Li = (x ∧ (E ∧ B))i
4πc
= (x ∧ Π)i
L(mec) = x ∧ p
Z
(·)
Li = Li d3 x
G

Mij = εjℓmTiℓ xm .

The proof of equation (1.70) is left as an exercise. Mij is a rank-2 (pseudo) tensor
that describes the flow of angular momentum of the electromagnetic field.

1.6 The relativistic formalism of electrodynam-

ics

(Jackson, chap. 11)

Lorentz transformations

The fundamental principle of special relativity is the equivalence of all laws of

physics in every inertial reference frame. A particle that does not experience any
force moves with constant velocity with respect to an inertial frame. (This is
Ruth Durrer Electrodynamique Chap. 1 29

often used as the definition of inertial frames. However, in order to establish the
definition unambiguously, one would have to know a priori all the forces, which are
in general defined as via their acceleration of a particle as perceived by an inertial
observer...) I do not know of any fully satisfactory definition of inertial frame, but
it is clear that if one inertial frame is given, all the others are obtained through
boosts (see below), rotations, reflections and time reversal transformations. (In
reality, not all physics laws are invariant under the latter discrete transformations.
The weak interactions, responsible for the β–decay of unstable isotopes, do not
possess the reflection and time reversal invariances!)

If Σ is an inertial reference frame, then Σ′ , which moves with constant speed along
a straight line with respect to Σ (boost) or has an arbitrary but fixed rotation
with respect to Σ, is also an inertial frame. A physicist in the reference frame
Σ obtains the same results as her colleague in Σ′ when they perform the same
experiment. This result was originally formulated by Galilei (principle of Galilei),
who postulated that the laws of mechanics are invariant under the transformation:

Σ → Σ′
     
t t′ t
→ = (Galilei transformation). (1.71)
x x′ x − tv

Here Σ and Σ′ are two inertial frames that coincide at t = 0, and Σ′ moves with
velocity v with respect to Σ. If a particle has the velocity u = dx/dt in Σ, then it
has the velocity u′ = dx′ /dt′ = u − v in Σ′ .

According to Maxwell’s equations, the invariance of the laws of physics implies that
the speed of light is the same in every inertial frame. It is precisely this condition
that Einstein wanted to implement with new transformations different from (1.71).
New transformations are certainly necessary since we want c′ = c 6= c − v.

A simple but significant consequence of the constancy of the speed of light is the
relativeness of simultaneity. To illustrate this, let us consider a lighting flash that
takes place halfway between two screens fixed on either side. For an observer at
rest with respect to the experiment, both screens will light up at the same time.
An observer moving to the right perceives the right screen as moving towards the
source of the flash and so, according to her, the photons must travel a shorter path
to arrive at the right screen than to do so at the left one. Since the speed of light
is the same for her, the right screen lights up before the left screen, and therefore
the events are not seen as simultaneous in her reference frame. This indicates that
the new transformations will transform not only x but also the time.

We now want to determine how the coordinates t and x are modified under a
change of the frame of reference. For this we just use the fact that the speed of
light c which comes out from Maxwell’s equations is the same in every frame. The
30 Section 1.6

equation (ct)2 − x2 = 0 is then the same independently of the frame of reference.

This equation is also invariant if one of the frames is dilated with respect to the
other, t′ = αt, x′ = αx by an arbitrary constant α ∈ R. But we know that
the physics is not invariant under dilations (for example, the spectral lines of the
hydrogen atom are the same in every inertial reference frame). To get rid of these
dilations that are in disagreement with phenomena, we assume that (ct)2 − x2 is
frame-independent for any value of this invariant3 . We also require transformation
to be linear and we consider Σ′ to be moving with speed v = const. in the direction
x = x1 . We assume

   ′   
t t α(t − c−1 δx)
 x   x′   γ(x − cβt) 
 → ′ =  (1.72)
 y   y   y 
z z′ z

and we demand that c2 t2 − x2 = c2 t′ 2 − x′ 2 . If we insert the expressions of (1.72)

in (ct)2 − x2 , we find
c2 t2 − x2 = c2 α2 t2 − 2cαδtx + α2 δ 2 x2 − γ 2 x2 + 2cγβxt − c2 γ 2 β 2 t2 ∀x, t .
Requiring that the coefficients of the quadratic expressions on both sides be the
same, we obtain α = γ, β = δ and
1
γ=p . (1.73)
1 − β2
The dimensionless, discontinuous function β(v) remains. At low speeds, v ≪ c, we
want to restore the result of Galilei, which implies β = v/c + O((v/c)2 ). A boost
of velocity v in direction x then transforms

   0     
ct x x′ 0 x0
 x   x1   ′1   x1 
    →  x′ 2  = Λ  ,
 y  ≡  x2   x   x2 
(1.74)
z x3 x′ 3 x3
with  
γ −γβ 0 0
 −γβ γ 0 0 
Λ = (Λµ ν ) = 
 0
. (1.75)
0 1 0 
0 0 0 1
3
It is a mathematical exercise to show that the transformations that preserve (ct)2 − x2 = 0
are precisely those that keep (ct)2 − x2 = s2 invariant for any value s and dilations. If only
linear transformations are considered, the exercise is simple. It is striking that the same can be
said even if non-linear (but differentiable) transformations are considered; nevertheless, the proof
becomes much more difficult. This is Alexandrov’s theorem (1975) (See: W. Benz, ’Geometrische
Transformationen’, BI Wissenschaftsverlag, 1992).
Ruth Durrer Electrodynamique Chap. 1 31

The inverse boost is the one with velocity −v, and it is easy to verify that the
matrix Λ−1 (β) = Λ(−β). Therefore β(−v) = −β(v) and β cannot have terms
quadratic in v/c. It is an experimental fact that β has no corrections, i.e.

β = v/c .

We define the four-vector

x = (xµ ) = (ct, x), x0 = ct, xi = xi for i = 1, 2, 3

and the ‘metric’

 
−1
 1 
g ≡ (gµν ) ≡ 

 = (gµν )−1 ≡ (g µν ) ≡ g −1 .
 (1.76)
1
1

In this course we have chosen the metric signature (−, +, +, +), and we denote
with (g µν ) the inverse of the metric.

One therefore has that (ct)2 −x2 = −xµ gµν xν is independent of the inertial frame4 .

xµ gµν y ν = [(x + y)µ gµν (x + y)ν − (x − y)µ gµν (x − y)ν ]/4 ,

it follows that xµ gµν y ν is frame-invariant for every couple of four-vectors (xµ ), (y ν ).

For a boost,
(x′ )µ = Λµν xν ,
this implies that Λαν xν gαβ Λβµ y µ = xν gνµ y µ for any four-vectors (xµ ), (y ν ) ∈ R4 ,
and so
Λαν gαβ Λβµ = gνµ . (1.77)
In matrix notation (1.77) translates into ΛT gΛ = g. Apart from the boost in
direction x1 , it is clear that rotations and reflections, with
 
1 0
Λ(R) ≡
0 R

where RRT = 1, i.e, R ∈ O(3), also satisfy condition (1.77).

A boost in an arbitrary direction can be decomposed into a rotation followed by a

boost in direction x1 and the inverse rotation. It is possible to show that a matrix
Λ that satisfies ΛT gΛ = g with det(Λ) = 1 and Λ00 ≥ 1 can be decomposed into a
boost and a rotation.
4
A sum from 0 to 3 must be performed over any Greek index appearing twice.
32 Section 1.6

For all four-vector (v µ ) that transforms as v µ → Λµν v ν , eq (1.77) implies that

v 2 ≡ gµν v µ v ν is invariant under Lorentz transformations. We require vµ ≡ gµν v ν ,
such that v 2 = vµ v µ is invariant under Lorentz transformations.

Examples
(v µ ) = (xµ ) = (ct, x). So v 2 = −c2 t2 + x2 = constant is the equation of motion of
a light ray. That it is independent of the reference system is the basis of special
relativity.
(v µ ) = (pµ ) = (ε/c, p). So −ε2 /c2 + p2 = −m2 = constant. Here ε is the energy of
the particle (see the Mechanics I course). (pµ ) = ( εc , p) is the energy-momentum
four-vector of the particle.

(pµ ) = (muµ ) = (mγc, mγv)

p2 = −(p0 )2 + p2 = −m2 c2 ;

here m denotes the invaiant rest mass of the particle.

Multiplying eq. (1.77) by (Λ−1 )µσ from the right and by g ρν from the left one finds

(Λ−1 )ρσ = g ρβ Λν β gνσ . (1.78)

In matrix notation this is Λ−1 = g −1 ΛT g. With eq. (1.78) we find that

ρ
vµ′ = gµν Λν σ v σ = gµν Λν σ g σρ vρ = ((Λ−1 )T )µ vρ .

Frequently we denote
ρ
gµν Λν σ g σρ ≡ Λµ ρ ≡ ((Λ−1 )T )µ , such that vµ′ = Λµ ν vν . (1.79)

The four-vectors (v µ ) are called contravariant vectors; they transform with Λ =

(Λµ ν ) under Lorentz transformations. In contrast, the four-vectors (vµ ) are called
covariant vectors, and they transform with (Λ−1 )T = (Λµ ν ) under Lorentz trans-
formations.

We consider not only four-vectors but also tensors, for example Tµν , such that
Tµν v µ w ν is an scalar, hence invariant under Lorentz transformations for arbitrary
four-vectors (v µ ) and (w ν ). With v ′ µ = Λµ ν v ν and similarly for w ν we have
′
Tµν Λµ α v α Λν β w β = Tµν v µ w ν therefore ′
Tµν Λµ α Λν β = Tµν . (1.80)

Multiplying by (Λ−1 )α ρ (Λ−1 )β σ using (1.79), we find

′
Tρσ = Λρ α Λσ β Tαβ . (1.81)

For a four-tensor of rank n we define the general way to ”raise and lower indices”:

Tµ1µ2 ···µn = gµ1 ν T νµ2 ···µn ,

Ruth Durrer Electrodynamique Chap. 1 33

which implies
T µ1 ···µn = g µ1 ν Tν µ2 ···µn ,
and analogously for the other indices. For a four-tensor every lower index trans-
forms then with (Λµ ν ) while upper indices transform with (Λµ ν ) under a Lorentz
transformation.

If we have an equation of the form

T µ1 µ2 ···µn = S µ1 µ2 ···µn

then this equation remains valid in all frames. On the other hand, if we want an
equation to keep the same form in any frame, then this equation must be written
in terms of four-tensors. According to special relativity, the fundamental laws of
nature are valid in any frame and so they must be expressed in terms of four-tensors
(this is often called covariantly).

Proper time We consider a particle or an object that moves with speed v in Σ.

In a time interval dt the particle moves a distance dx = vdt. The invariance of the
speed of light implies that

c2 2
c2 dτ 2 = ds2 = c2 dt2 − dx2 = c2 (1 − β 2 )dt2 = dt , β 2 = v2 /c2
γ2

is the same in all frames. In the reference system in which the particle is (momen-
tarily) at rest this gives dτ 2 = c−2 ds2 = dt′ 2 , and so one calls τ the proper time of
the particle. The proper time is independent of the frame, it is a Lorentz scalar.
In a generic frame dt = γdτ . The time that elapses between two proper times τ1
and τ2 is then Z Z
t2 τ2
t2 − t1 = dt = γdτ ≥ τ2 − τ1 . (1.82)
t1 τ1

So a watch that moves is slower than a watch at rest. This fact has been experi-
mentally tested not only with elementary particles whose lifetime becomes longer
in accelerators, but also with very precise watches in aircrafts.

Addition of velocities We consider once again the system Σ′ that moves with
speed v in direction x1 with respect to Σ. For coordinate differentials we then have

   
dx0 γv (dx′ 0 + βdx′ 1 )
 dx   γv (dx′ 1 + βdx′ 0 )
1  1
 2 =  γv = p . (1.83)
 dx   dx′ 2  1 − v 2 /c2
dx3 dx′ 3
34 Section 1.6

Let us consider a particle that moves with speed u′ = dx′ /dt′ = cdx′ /dx′ 0 in Σ′ .
We assume that the part of the velocity that is normal to x′1 is parallel to y ′ = x′ 2 .
Its velocity in Σ is then

1dx1 dx′ 1 + βdx′ 0 u′ 1 + cβ u′k + v u′k + v

uk ≡ u = = c ′0 = = =
dt dx + βdx′ 1 1 + βu′ 1 /c 1 + (vu′k )/c2 1 + (v · u′ )/c2
dx2 c dx′ 2 1 u′⊥ 1 u′⊥
u⊥ ≡ u2 = = = = .
dt γv dx′ 0 + βdx′ 1 γv 1 + (vu′k)/c2 γv 1 + (v · u′ )/c2
′
This expression does
not look symmetrical in v and u . But it is easy to show that
v·u′
γu = γv γu′ 1 + c2 and so the above equations become

γu uk = γv γu′ (u′k + v)
γu u⊥ = γu′ u′⊥ ,

which is perfectly symmetric in v and u′ .

Formulated as an exercise, the following part is addressed to the students of theo-

retical physics:

Definition: The set of matrices Λ ∈ R4×4 |ΛgΛT = g ≡ O(1, 3) ≡ L, with g =
diag(−1, 1, 1, 1) is the Lorentz group.
Show the following statements.

• ΛT gΛ = g if Λ is a rotation or a boost of direction x1 and if Λ = T =

diag(−1, 1, 1, 1) or Λ = P = diag(1, −1, −1, −1). (i.e. rotations, boosts of
direction x1 , T (time reversal) and P (parity) are all elements of L.)

• L is a genuine group in the mathematical sense of the term.

• For Λ ∈ O(1, 3), | det Λ| = 1 and |Λ00| > 1.

• O(1, 3) ⊂ R4×4 ∼
= R16 is composed by four disconnected components:

L↑+ = {Λ ∈ O(1, 3)| det Λ = 1, Λ00 > 1}

L↑− = {Λ ∈ O(1, 3)| det Λ = −1, Λ00 > 1}
L↓− = {Λ ∈ O(1, 3)| det Λ = −1, Λ00 6 −1}
L↓+ = {Λ ∈ O(1, 3)| det Λ = 1, Λ00 6 −1}.
Ruth Durrer Electrodynamique Chap. 1 35

• In a neighbourhood of identity, the group O(1, 3) is parametrised by six

parameters, i.e the Lorentz group is 6-dimensional.

Boost and rotations compose a group of matrices referred to as L↑+ , the proper
Lorentz group. In terms of time inversion, T , and parity, P . The complete Lorentz
group is given by
L = L↑+ ⊕ P L↑+ ⊕ T L↑+ ⊕ P T L↑+
= L↑+ ⊕ L↑− ⊕ L↓− ⊕ L↓+
Here ± accounts for the sign of the determinant, and ↑ or ↓ denotes Λ0 0 > 1 or
Λ00 6 −1, respectively.

Covariant electrodynamics

We then want to find a covariant (or relativistic) formulation of the laws of elec-
trodynamics.

To find a relativistic formulation of the conservation of charge we first write (j µ ) =

(cρ, J). Considering that (xµ ) = (ct, x), then it is clear that the partial derivatives
(∂µ ) ≡ ( 1c ∂t , ∂i ) also constitute a four-vector. With these definitions, the equation
for the conservation of charge, eq. (1.59), can be rewritten as
∂µ j µ = 0 (ρ̇ + ∇ · J = 0). (1.84)

Since 0 is a Lorentz-invariant scalar, this implies that (j µ ) is also a four-vector.

We would like to find a covariant form for Maxwell’s equations in E and B. With
that in mind, let us first consider the static case where ρ is constant and J vanishes.
We will see that requesting covariance will completely determine the full Maxwell
equations. The equations of electrostatics are
∇ ∧ E = 0 and ∇ · E = 4πρ.
To promote Coulomb’s equation to a covariant equation one has to notice that ρ
is the 0 -component of the four-vector j µ . The desired equation must then be of
the form

4π µ
−(∂ · F )µ = j .
c
Here ∂ · F is the divergence of a rank-2 tensor field that is linear in E. (The sign
− is simply a convention.) We therefore have
4π µ
−∂ν F νµ = j . (1.85)
c
36 Section 1.6

In addition, we want to make sure that charge is conserved, ∂µ j µ = 0, which implies

that
∂µ ∂ν F µν = 0.
This drives us to postulate an antisymmetric tensor F µν ,

F µν = −F νµ .

(We could add to Fµν a symmetric part Sµν that satisfies ∂µ ∂ν S µν = 0, but S µν
is entirely decoupled from the antisymmetric part and from the source (j µ ). It is
therefore consistent with Maxwell’s equations to fix Sµν ≡ 0, which we do also for
reasons of economy.)

A four dimensional antisymmetric tensor has 6 independent components. Because

of (1.85) for µ = 0, the components F i0 are to be identified with the electric field.
We then write  
0 E1 E2 E3
 −E1 0 B3 −B2 
(F µν ) = 
 −E2 −B3
, (1.86)
0 B1 
−E3 B2 −B1 0
where the interpretation of the components Bi is not clear a priori. This translates
into the relations 
F i0 = −E i = −F 0i
. (1.87)
F ij = Fij = εijℓ Bℓ = −Fji
4π 0
With this, −∂µ F µ0 = c
j is equivalent to Coulomb’s law.

By making B = (B1 , B2 , B3 ), equation (1.85) for µ = i gives

4π i
∂0 F 0i + ∂ℓ F ℓi = − j
c
1 i 4π
Ė + ∂ℓ εℓij Bj = − ji
c | {z } c
−εiℓj ∂ℓ Bj
| {z }
−(∇∧B)i

1 4π
− Ė + ∇ ∧ B = J,
c c
that is nothing but Ampère’s law if we interpret B as the induction field. The
existence of an induction field B is then a pure consequence of the covariance of
electrodynamics! This is a clear example of the power of symmetries in physics: the
requirement that equations are to be the same in any inertial frame, i.e. that they
are to be covariant, leads us from electrostatics to the dynamic Maxwell equations
and implies the existence of magnetic induction. The tensor F µν is called the
electromagnetic field tensor or the ’Faraday tensor’.
(Historically, of course, this is not how things happened!)
Ruth Durrer Electrodynamique Chap. 1 37

To find the generalisation of the homogeneous static equation, ∇ ∧ E = 0, we first

define the dual tensor of F µν ,
1
∗F µν := ǫµναβ Fαβ , (1.88)
2
where
n
ǫµναβ
= sgn(0, 1, 2, 3 7→ µναβ) if µ, ν, α, β are all different
0 otherwise.
 
0 B1 B2 B3
 −B1 0 −E3 E2 
(∗F µν ) = 
 −B2 E3
. (1.89)
0 −E1 
−B3 −E2 E1 0
From this representation it follows that F µν → ∗F µν if (E, B) → (B, −E).

We therefore have
1
(∗F )ij = (ǫijℓ0 Fℓ0 + ǫij0ℓ F0ℓ ) = −εijℓ Eℓ
2
and 0 = (∇ ∧ E)i = εijℓ ∂j Eℓ is equivalent to

0 = (∇ ∧ E)i = −∂j (∗F )ij .

The covariant generalisation of this equation is, evidently,

∂µ (∗F )νµ = 0, (1.90)

which is equivalent to the homogeneous Maxwell equations (exercise). In conclu-
sion: Maxwell’s equations are compatible with special relativity and their explicitly
covariant form is

∂ν (∗F νµ ) = 0 (homogeneous equations),

∂ν F νµ = − 4π
c j
µ
(inhomogeneous equations).

The homogeneous equations, (1.90), are equivalent to

Fαβ,ν + Fβν,α + Fνα,β = 0

(show it as an exercise). In 4 dimensions this implies (Poincaré’s Lemma) the

existence of a potential Aµ such that

Fαβ = ∂α Aβ − ∂β Aα . (1.91)
38 Section 1.6

We make (Aµ ) = (φ, A), which gives

1
Ei = Fi0 = −∂i φ − ∂t Ai
c
1 ij 1
Bℓ = εℓ Fij = εℓ ij (∂i Aj − ∂j Ai )
2 2
= (∇ ∧ A)ℓ .

The ansatz (1.91) automatically satisfies the homogeneous Maxwell’s equations.

The inhomogeneous equations become
4π
∂ µ Fµν = ∂ µ ∂µ Aν − ∂ µ ∂ν Aµ = − jν
c
4π
⊓Aν − ∂ µ ∂ν Aµ
−⊔ = − jν . (1.92)
c
Here we have used
1 2
⊓ = −∂µ ∂ µ =
⊔ ∂ − ∆,
c2 t
the d’Alembertian or wave operator. A gauge transformation is now a transforma-
tion of the form:
Aµ → Aµ + ∂µ χ
and the Lorentz gauge condition is simply (cf. eq. (1.54))

∂µ Aµ = 0. (1.93)

Within the Lorentz gauge, the inhomogeneous equations (1.92) give

4π µ
⊓Aµ =
⊔ j . (1.94)
c
In vacuum, j µ = 0, we have that (within the Lorentz gauge)

⊓Fµν = ⊔
⊔ ⊓(∂µ Aν − ∂ν Aµ ) = ∂µ ⊔
⊓Aν − ∂ν ⊔
⊓Aµ = 0.

Given that Fµν is independent of the gauge, this result is valid in any gauge what-
soever. Indeed, without having to turn to Aµ , we have that

0 = ∂ α (∂α Fµν + ∂µ Fνα + ∂ν Fαµ )

⊓Fµν + ∂µ ∂ α Fνα +∂ν ∂ α Fαµ .
= −⊔
| {z } | {z }
0 0

Then
⊓Fµν = 0 (for j µ = 0).
⊔ (1.95)
In vacuum, every component of the electromagnetic field satisfies the wave equa-
tion.
Ruth Durrer Electrodynamique Chap. 1 39

Lorentz transformations of the electromagnetic field

F µν is a second rank tensor. Under a Lorentz transformation Λ, it then transforms

as
(F µν )′ = Λµα Λνβ F αβ . (1.96)
We first consider a rotation,
 
1 | 0
− + − − −
 
(Λ ν ) = 
µ
 | ,
 (1.97)
0 | R 
|

with R · RT = 1I and det R = 1 (R ∈ SO(3)). Since E and B are three-dimensional

vectors, (1.96) must imply (E i )′ = Ri j E j , and similarly for B. We want to verify
this:
(E i )′ = (−F i0 )′ = −Λi α Λ0β F αβ = −Λi j F j0 = Ri j E j ;

F ij = εijℓ B ℓ ⇒ εkij F ij = εkij εijℓ B ℓ = 2B k ,

1 k 1
(B k )′ = ε ij (F ij )′ = εkij Λi α Λj β F αβ
2 2
1 k i j mn 1 k i j mn ℓ
= ε R R F = ε ij R m R n ε ℓ B .
2 ij m n 2
Let us now use the fact that εmnℓ is a tensor invariant under rotations:

Rpk Rj m Ri n εkmn = εpji

⇒ Rpℓ εpji = Rjm Ri n Rpk Rpℓ εkmn = Rj m Ri n εℓ mn .

For the last equality we have made use of the identity R · RT = 1I, which is
equivalent to Rpk Rpℓ = δkℓ .

Hence Ri m Rjn εℓ mn = Rpℓ εpij . Therefore

1 k p ij ℓ
(B k )′ = ε R ε B = Rkℓ B ℓ .
2 | ij {zℓ p }
2δpk Rpℓ

We have consequently found that B and E transform as three-dimensional vectors

under rotations, and this agrees with the behaviour expected for electrodynamics
in its non-covariant formulation.
40 Section 1.6

Let us now study the behaviour of E and B under a boost of speed v in the
direction x1 . This Lorentz transformation is given by
 
γ −γβ 0 0
 −γβ γ 0 0
(Λµν ) = 
 0
, (1.98)
0 1 0
0 0 0 1
p
where β = v/c and γ = 1/ 1 − β 2 .

(F µν )′ = Λµα Λνβ F αβ
(E1 )′ = (−F 10 )′ = −Λ10 Λ01 F 01 − Λ11 Λ00 F 10
(E1 )′ = (Λ11 Λ00 − Λ10 Λ01 ) E1 = E1 (1.99)
| {z }
γ 2 (1−β 2 )

(E2 )′ = (−F 20 )′ = −Λ22 (Λ00 F 20 + Λ01 F 21 )

(E2 )′ = γE2 − γβB3 (1.100)

and, likewise,
(E3 )′ = γE3 + γβB2 (1.101)
1 1 1
(B1 )′ = (εij1 F ij )′ = εij1 Λi α Λj β F αβ = (ε231 F 23 + ε321 F 32 ) = B1
2 2 2
B1′ = B1 (1.102)
1
(B2 )′ = (εij2 F ij )′ = (ε132 F 13 )′ = ε132 Λ1α Λ3β F αβ = ε132 (Λ10 F 03 + Λ11 F 13 )
2
B2′ = γB2 + γβE3 (1.103)
and similarly
B3′ = γB3 − γβE2 . (1.104)

Example: electromagnetic field of a point charge in uniform

motion

We consider a charge that travels with speed v in the direction x1 in the reference
frame K. In the frame K ′ , which moves with speed v in the direction x1 , the
charge is at rest. In K ′ it then produces the field

ex′
E′(x′ , t′ ) = , B′(x′ , t′ ) = 0.
|x′ |3
Ruth Durrer Electrodynamique Chap. 1 41

Using the expressions of the transformation of (xµ ) under a boost of speed −v in

the direction x1 , we have that

ct = γ(ct′ + βx′1 ), ct′ = γ(ct − βx1 ) (1.105)

x1 = γ(x′1 + βt′ ), x′1 = γ(x1 − βct) (1.106)
x2 = x′2 (1.107)
x3 = x′3 (1.108)

and applying equations (1.100) to (1.104) we find

eγ(x1 − vt)
E1 (x, t) = E1′ (x′ , t′ ) =   23
γ 2 (x1 − vt)2 + (x2 )2 + (x3 )2 
| {z }
b2

eγ(x1 − vt)
=   3 , b2 = (x2 )2 + (x3 )2 (1.109)
γ 2 (x1 − vt)2 + b2 2
eγx2
E2 (x, t) =  3 , (1.110)
2 1 2 2 2
γ (x − vt) + b
eγx3
E3 (x, t) =  3 . (1.111)
γ 2 (x1 − vt)2 + b2 2

At time t = 0, we obtain

e(1 − β 2 )x
E(x, t = 0) =  3 , r = |x|. (1.112)
r 2 − β 2 b2 2

With sin ϑ = rb this gives

e(1 − β 2 )
|E|(x, t = 0) = 3 . (1.113)
r 2 (1 − β 2 sin ϑ) 2

For a fixed value of r, |E| attains its maximum value in the direction ϑ = π
2 (i.e.
when b = r), which corresponds to the normal plane to the direction of motion.
|E| minimizes in the direction ±x1 , ϑ = 0 and π:

|E| = 2 e π
ϑ= 2  
r (1 − β 2 )1/2 . (1.114)
e(1 − β 2 ) 

|E| = ϑ = 0, π
r2
42 Section 1.6

Figure 1.1: Surfaces |E| = constant for a point charge at rest (dashed line) and
for a point charge moving at a constant velocity v (solid line).

The surfaces |E| = const. are contracted in the direction of motion and dilated in
the orthogonal directions (see figure 1.1).

For the magnetic field we have

B1 = 0, B2 = −βγE3′ = −βE3 , B3 = βγE2′ = βE2 ,
or, equivalently,
B = β ∧ E. (1.115)
A moving charge produces a magnetic field perpendicular to v of magnitude pro-
portional to vc .

The relativistic Lorentz force

In order to complete the relativistic formalism, we look for the relativistic form of
the Lorentz force equation. For particles of charge q moving at low speed v ≪ c
we have the nonrelativistic limit
 
dv i i 1 i
m = q E + (v ∧ B) . (1.116)
dt c

From relativistic mechanics, you know the four-velocity

X
(uµ ) = (cγ, vγ) , u2 = −(u0 )2 + (ui )2 = −c2 , p = mu .
i
Ruth Durrer Electrodynamique Chap. 1 43

µ
But, due to the fact that ct is the 0-component of a four-vector, du is not a four-
dt
vector. We have to replace dt with the particle’s proper time, dτ = dt/γ which is,
as we have seen, a Lorentz scalar. In terms of proper time we have uµ = dxµ /dτ .
With dt = γdτ (1.116) becomes

dui q
m = F iα uα .
dτ c
The four-dimensional formulation of this equation is
dpµ duµ q
=m = F µν uν . (1.117)
dτ dτ c
The zeroth component of (1.117) describes the change of energy ε of a particle in
a magnetic field:
dp0 q 1
= F 0i vi γ = qγ E · v
dτ c c
dε
ε = p0 c, = qγE · v. (1.118)
dτ
The energy of the particle changes due to the work done by the electric field.

The energy-momentum tensor of the electromagnetic field

We define yet another second rank tensor, the energy-momentum tensor of the
electromagnetic field Tµν , given by

1 2
T00 = 8π (E + B ) = u 2 



−1
Ti0 = 4π (E ∧ B)i = −cΠi = T0i . (1.119)
  

Tij = 4π 1 1 δ (E 2 + B 2 ) − E E − B B = T  
2 ij i j i j ji

Exercise: Show that the expression given for the components of Tµν in eq. (1.119)
is equivalent to  
1 α 1 αβ
Tµν = Fµα Fν − gµν Fαβ F . (1.120)
4π 4

The indices of Tµν are, as usual, raised and lowered with the Lorentz metric:

Tµν = Tµ α gαν , (gαν ) = diag(−1, 1, 1, 1) (1.121)

T µν = g µα Tαν . (1.122)
44 Section 1.6

In order to show that Tµν is a true four-tensor it is necessary to show, as for the
electromagnetic field tensor, F µν , that under a Lorentz transformation (Λµν ) one
has
(T µν )′ = Λµα Λνβ T αβ . (1.123)
We first consider a rotation R,
 
1 | 0
− + − − −
 
(Λ µ ) = (Λν ) = 
ν µ
 | ,
 (1.124)
0 | R 
|

with R · RT = 1I. We know that E and B transform, under rotations, the way
three-dimensional vectors do,

E ′i = Ri j E j , B ′i = Ri j B j ,

so

E′2 = E ′i E ′i = Rji E j Rℓi E ℓ

= (RT · R)jℓ E j E ℓ = δjℓ E j E ℓ = E j E j = E2
1 (E2 + B2 ) is then invariant under rotations. For the
and similarly for B. T00 = 8π
components Ti 0 , we use

(E′ ∧ B′ )q = εqℓm E ′ℓ B ′m = εqℓm Rℓ k Rmj E k B j . (1.125)

As for the tensor F µν , the invariance of the tensor εijl implies

Rqi εiℓm = Rkℓ Rnm εqkn .

With (1.125) this gives

(E′ ∧ B′ )q = Rqi εiℓm E ℓ B m = Rqi (E ∧ B)i .

We have therefore shown that (E ∧ B) transforms as a vector. We now proceed to

verify (1.123). For the “00 ” component we have
′
T00 = Λ0µ Λ0ν Tµν = T00 .

Which is as expected because

1 1
(E′2 + B′2 ) = (E2 + B2 ).
8π 8π
For the “i0 ” components we make use of the fact that
−1 j −1 ′
Ti0′ = Λi µ Λ0ν Tµν = Λi j Tj0 = Ri (E ∧ B)j = (E ∧ B′)i .
4π 4π
Ruth Durrer Electrodynamique Chap. 1 45

Correspondingly, for the “ij ” components we have:

Tij′ = Λi µ Λj ν Tµν = Λi k Λj ℓ Tkℓ

= Ri k Rj ℓ Tkℓ
 
k ℓ 1 2 2
= Ri Rj δkℓ (E + B ) − Ek Eℓ − Bk Bℓ
2

but Ri k Ek = Ei′ , Ri k Bk = Bi′ , E2 = E′2 , B2 = B′2 and

Ri k Rj ℓ δkℓ = Ri k (RT )kj = δij . We then finally have

1
Tij′ = δij (E′2 + B′2 ) − Ei′ Ej′ − Bi′ Bj′ .
2

Thus we have verified that under a rotation Λµν (R), given by (1.124), Tµν transforms
according to (1.123).

We now show that Tµν also transforms as a second rank tensor under a boost of
speed v in the direction x1 . Let again β and γ be given by β = vc and γ = p 1 ;
1 − β2
 
γ −γβ 0 0
 −γβ γ 0 0
(Λµν ) = 
 0

0 1 0
0 0 0 1

(If χ := arcthβ, it follows that γ = cosh χ and βγ = sinh χ, where χ is the speed
of the boost). Let us consider first T00 :

(T 00 )′ = Λ0α Λ0β T αβ = γ 2 T 00 − 2γ 2 βT 01 + γ 2 β 2 T 11
1 2 2
= [γ (E + B 2 ) − 4γ 2 β(E2 B3 − B2 E3 ) + γ 2 β 2 (E 2 + B 2
8π
−2(E12 + B12 ))]
1
1 z 2 }| 2{ 2
= [γ (1 − β )(E1 + B12 ) + γ 2 (1 + β 2 )(E22 + E32 + B22 + B32 )
8π
−4γ 2 β(E2 B3 − B2 E3 )]
1  2
= (E1 + B12 ) + γ 2 (E2 − βB3 )2 + γ 2 (E3 + βB2 )2 + γ 2 (B2 + βE3 )2
8π

+γ 2 (B3 − βE2 )2
1

= E′2 + B′2 .
8π
46 Section 1.6

The last equality follows from equations (1.100) to (1.104). For T 01 , we obtain

(T 01 )′ = Λ0α Λ1β T αβ
1/8π(E 2 +B 2 ) 1/4π(1/2(E2 +B2 )−E12 −B12 )
z}|{ z}|{
= γ 2 T 01 − βγ 2 ( T 00 + T 11 ) + γ 2 β 2 T 10
1 2
= [γ (1 + β 2 )(E2 B3 − B2 E3 ) − βγ 2 (E22 + E32 + B22 + B32 )]
4π
1 2
= [γ (E2 − βB3 )(B3 − βE2 ) − γ 2 (B2 + βE3 )(E3 + βB2 )]
4π
1 1
= (E2′ B3′ − B3′ E2′ ) = (E′ ∧ B′ )1
4π 4π
and similarly for the other components.
We illustrate this by computing (T 23 )′
−1
(T 23 )′ = Λ2α Λ3β T αβ = T 23 = (E2 E3 + B2 B3 )
4π
−1 2
= [γ (E2 − βB3 )(E3 + βB2 ) + γ 2 (B2 + βE3 )(B3 − βE2 )]
4π
−1 ′ ′
= (E2 E3 + B2′ B3′ ).
4π
The computation of the remaining components is left as an exercise.

The conservation of energy and momentum can now be simply written as

1 1
∂µ T µν = −k ν , with k 0 = J · E, k i = (ρE + J ∧ B)i . (1.126)
c c
k 0 is the work done on the charges and k is the Lorentz force density. It is easy to
see that kµ = Fµν j ν , and therefore (kµ ) is a four-vector.

We have discussed the behaviour of Tµν and F µν (second rank tensors) under
rotations around, and boosts in the direction x1 . A boost in an arbitrary direction
n is always the product of a rotation R1 that rotates n in the direction x1 , of a
boost Λ(β, e1 ) in direction x1 and, finally, of the inverse rotation, R1−1 :

Λ(β, n) = R1−1 Λ(β, e1 )R1 .

It then follows that Tµν transforms correctly under a boost in an arbitrary direction.
Chapter 2

Electromagnetic waves

2.1 One-dimensional waves

In order to restrict ourselves to a concrete example, let us consider a string in a

two-dimensional plane with one of its points fixed:

ϕ0
ϕ(x)
x

We look for an equation that describes the wave-like motion of the string, i.e. an
equation for the acceleration d2 ϕ/dt2 at every point x, taking into account the
tension forces. We make:
X∞
∂2ϕ ∂n
(x, t) = α n ϕ(x, t).
∂t2 n=0
∂xn

• Since the choice of the position ϕ0 is arbitrary, the force is independent of

the amplitude ϕ and, consequently, α0 = 0.

• If ϕ(x) describes a straight line, no tensions are applied, the force does not
depend on the slope, and therefore α1 = 0.

47
48 Section 2.1

• The lowest-order term is α2 : the curvature implies a tension.

For many applications, α2 turns out to be the dominant term and we then obtain
(α2 ≡ v 2 ):
∂2ϕ 2
2∂ ϕ
= v . (2.1)
∂t2 ∂x2
Eq. (2.1) is the equation of a one-dimensional wave. It can also be used to describe
the motion of a spring or the air compressions in a tube, as well as the physics of
many other one-dimensional problems.

The general solution of (2.1) which satisfies the initial conditions

ϕ(x, t = 0) = ϕ0 (x) and ϕ̇(x, t = 0) = ϕ1 (x)
is
ϕ(x, t) = f (x + vt) + g(x − vt) (2.2)
such that
Z x Z x
1 1 ′ ′ 1 1
f (x) = [ϕ0 (x) + ϕ1 (x )dx ] and g(x) = [ϕ0 (x) − ϕ1 (x′ )dx′ ] .
2 v 2 v
The function f describes the part of the wave that moves towards the −x direction,
the “left mover”, and the function g describes the “right mover”. v is the speed of
propagation of the wave. In all physically realistic cases (with finite energy), the
functions f and g can be written as Fourier integrals,
Z
f (x + vt) = dkA(k) exp[i(kx + ωt)] , (2.3)

with ω = vk. The function |A(k)|2 is the spectrum of the wave f .

Normal modes (stationary waves)

The time dependence of the normal modes of a wave go as cos(ωt+β). We therefore

look for a solution to (2.1) of the form
ϕ(x, t) = A(x) cos(ωt + β).
Using (2.1) it is found that
d2 A
−ω 2 A(x) = v 2 ,
dx2
whose solution is
A(x) = A0 cos(kx + α) (2.4)
k2v2 = ω2. (2.5)
Ruth Durrer Electrodynamique Chap. 2 49

We then have:
ϕ(x, t) = A0 cos(ωt + β) cos(kx + α)
A0
= [cos(ωt + kx + γ1 ) + cos(ωt − kx + γ2 )] , (2.6)
2
with γ1 = β + α and γ2 = β − α.

Equation (2.5) is the dispersion law for the wave equation (2.1). A dispersion
law is a relation between ω and k. We will encounter other dispersion relations
that are more complicated than (2.5). The quantity ω is the angular frequency (or
pulsatance), whose dimension is [1/s]. k is the wavenumber, of dimension [1/cm].
A0 is the amplitude; its dimension depends on the physical phenomenon under
consideration. For the string, for example, it corresponds to [cm]. γ1 are γ2 phases
determined by the initial conditions. T = 2π/ω is the wave’s period, the time
needed for one complete oscillation, and ν = 1/T is the frequency of the wave.
The distance λ = 2π/k is the wavelength, the length of one complete oscillation.
From (2.5) we have
λν = v. (2.7)
Considering that the dispersion relation (2.5) has the solutions k = ±ω/v, it is
possible to add to (2.6) a term B0 [cos(ωt − kx + γ1 ) + cos(ωt + kx + γ2 )]. This
leads to the general solution
ϕ(x, t) = A cos(ωt + kx + δ1 ) + B cos(ωt − kx + δ2 )
= A1 cos(ωt+kx) + A2 sin(ωt+kx) + B1 cos(ωt−kx) + B2 sin(ωt−kx).
with
A1 = A cos δ1 , A2 = −A sin δ1 ,
B1 = B cos δ2 , B2 = −B sin δ2 .
This solution can also be written as follows
 
ϕ(x, t) = Re C1 ei(ωt+kx) + C2 ei(ωt−kx) (2.8)
where C1 = A(cos δ1 +i sin δ1 ) and C2 = B(cos δ2 +i sin δ2 ). The first term describes
a wave that propagates towards the left (“left mover”) and the second term repre-
sents a wave propagating towards the right (“right mover”). The position xc of the
crest of an oscillation is given by ωt + kxc + δ1 = 2πN, so dxc /dt = −ω/k = −v for
the first term; and ωt − kxc + δ2 , dxc /dt = ω/k = v for the second. It is because
of this that v is known as the phase velocity of the wave.

We now consider a string with fixed ends: ϕ(0, t) = ϕ(L, t) = 0. For t = 0 and
x = 0 this gives C2 = −C1 . For x = L and t arbitrary, we have


Re C1 eiωt eikL − e−ikL = 0 and so 2iC1 sin(kL) = 0. (2.9)
50 Section 2.2

This implies that sin(kL) = 0, or, in other words, that k can only take discrete
values
jπ 2L
kj = , λj = , j ∈ N. (2.10)
L j
From this it follows that λ0 = ∞. The angular frequency that corresponds to kj is
given by ωj = vkj . The general solution of our problem with boundary conditions
ϕ(0, t) = ϕ(L, t) = 0 is finally given by
"∞ #
X

ϕ(x, t) = Re Cj eiωj t eikj x − e−ikj x
j=1
∞
X
= (Aj cos ωj t + Bj sin ωj t) sin kj x (2.11)
j=1

with Aj = −2Im [Cj ] et Bj = −2Re [Cj ]. A situation like this is said to be of the
discrete spectrum type. The function A defined in equation (2.3) is zero except for
the discrete values k = jπ/L.

Exercise: Determine the coefficients Cj for a given solution ϕ(x, t) .

Remark: For the string, ϕ is an amplitude in a direction transverse to the di-

rection of propagation. A wave like this is called a transverse wave. For a spring
that compresses and stretches, the oscillating magnitude (the number of coils per
centimetre) is parallel to the direction of propagation. In this case we speak of a
longitudinal wave, of which sound waves constitute another example.

2.2 Three-dimensional waves, electromagnetic waves

To study three-dimensional waves, let us forget the case of the string and take
instead, for the moment, a compression wave in a gas (an acoustic wave). The
quantity ϕ is no longer the position of the string, but the density of a gas. For a
wave that propagates in an arbitrary direction n ∈ R3 (n2 = 1) we then have
   
ϕ = Re Cei(ωt−kn·x) = Re Cei(ωt−k·x) (2.12)

where k = kn is called the wave vector. By means of simple inspection, one can
verify that ϕ satisfies the equation

∂t2 ϕ = v 2 (∂x2 + ∂y2 + ∂z2 )ϕ

with v 2 = ω 2 /k2 , which can be rewritten as

∂t2 − v 2 △ ϕ = 0. (2.13)
Ruth Durrer Electrodynamique Chap. 2 51

Equation (2.13) is the equation of a three-dimensional wave. v is the phase velocity

of the wave. There is also something known as the group velocity of the wave:

∂ω
vg = .
∂k
In the examples so far considered, vg = v because the relation between ω and k
has always been linear. We will arrive at situations where this will no longer be
the case.

Remark: There are also two-dimensional waves. Examples of this are the waves
on the surface of water and the acoustic waves of a metallic disc.

Let us turn to the subject we are interested on: electromagnetic waves. We have
already seen [equation (1.32)] that the electric and magnetic fields in vacuum
satisfy the wave equations:
)
(∂t2 − c2 △) E = 0
(2.14)
(∂t2 − c2 △) B = 0

If we first consider a plane wave-like solution:

 
E(x, t) = Re E0 ei(k·x−ωt)
 
B(x, t) = Re B0 ei(q·x−σt) .

then equation (2.14) demands ω 2 = c2 k 2 and σ 2 = c2 q 2 . Hence c is the phase (and

group) velocity of the wave. The equations ∇ · E = 0 and ∇ · B = 0 imply

k · E = q · B = 0.

Electromagnetic waves are then transverse waves. In addition, the law of induction
gives  
0 = −∂t E + c∇ ∧ B = Re iωE0ei(k·x−ωt) + ciq ∧ B0 ei(q·x−σt) .
For this equation to be satisfied in x = 0, say, for every time t, one must have
σ = ω. What is more, if it is to be satisfied in any point x of space, then the
equality q = k is needed. Maxwell’s equations then imply for the fields that
 
E(x, t) = Re E0 ei(k·x−ωt)
  (2.15)
B(x, t) = Re B0 ei(k·x−ωt)

with
k · E0 = k · B0 = 0
(2.16)
ωE0 + ck ∧ B0 = 0
52 Section 2.2

or, using k2 = ω 2/c2 and k̂ = k/k,

E0 = −k̂ ∧ B0 . (2.17)
Likewise, Ampère’s law, ∂t B − c∇ ∧ E = 0, gives
B0 = k̂ ∧ E0
(something that follows also from (2.17) after multiplying by k̂∧...).

The vectors k, E0 , and B0 are therefore mutually orthogonal, and E20 = B20 . E0
and B0 are, in general, complex vectors (E0 ∈ C3 , B0 ∈ C3 ), and the equations we
have derived are valid for both their real and imaginary parts (why are they valid
for the imaginary part too?!).

Polarisation, Stokes parameters

Let us decompose E0 = A1 − iA2 , A1,2 ∈ R3 . In a fixed position x (x ≡ 0, say),

we have
E(t) = A1 cos ωt + A2 sin ωt. (2.18)

a) Linear polarisation

If A1 k A2 , then only the amplitude of the electromagnetic field, and not its
direction, changes in time (E(t) k E(t′ )). The same can be stated for B. Such an
electromagnetic wave is said to be ’linearly polarized’.

b) Circular polarisation

Consider now the opposite case, A1 ⊥ A2 , but with the further condition A21 =
A22 = A2 . We choose a coordinate system in which A1 points in the direction x
and A2 points in the direction y. Our wave then propagates in the direction z (k
is parallel to the z direction). Thus, from (2.18), we have (at a fixed position x)
Ex = A cos ωt, Ey = A sin ωt, Ez = 0,
i.e., E sweeps out a circle of radius A in the (x, y)-plane.
If (A1 , A2 , k) form a system of positive helicity ((A1 ∧
A2 ) · k > 0), the sweep goes counterclockwise, i.e. in the y
trigonometric direction (from the x axis to the y axis). In
optics, such a wave is called a left wave. On the other hand,
if (A1 , A2 , k) form a negative helix ((A1 ∧ A2 ) · k < 0), we ωt
A

x
have a right wave. (It rotates clockwise for an observer that
receives the wave.) Because the direction of E determines z
that of B, B0 = k̂ ∧ E0 , the polarisation of B is the same
as the one of E.
Ruth Durrer Electrodynamique Chap. 2 53

c) General case

Let (ǫ1 , ǫ2 , k̂) be an orthonormal system of positive helicity. The general form of
E is
 
E(x, t) = Re (E1 ǫ1 + E2 ǫ2 )ei(k·x−ωt) (2.19)
 
= Re (E+ ǫ+ + E− ǫ− )ei(k·x−ωt) (2.20)
1 1
where ǫ± = √ (ǫ1 ± iǫ2 ) , E± = √ (E1 ∓ iE2 ) .
2 2
Equations (2.19) and (2.20) are the explicit decompositions into linearly polarised
(E1 = 0 or E2 = 0) and circularly polarised (E− = 0, left wave; E+ = 0, right
wave) waves.

We now show that in the general case E describes an ellipse and we determine its
direction and axes lengths. Let us choose ǫ1 = ex and ǫ2 = ey . The X and Y
components of E(t) = Xex + Y ey (at the fixed position x = 0) are then given by
1

X = E1 e−iωt + E1∗ e+iωt
2
1

Y = E2 e−iωt + E2∗ e+iωt
2
1

X ± iY = (E1 ± iE2 )e−iωt + (E1∗ ± iE2∗ )eiωt
2
1

= √ E∓ e−iωt + E±∗ eiωt .
2
If we rotate counter clockwise the system (x, y) by an angle α, we obtain the
components
(ξ, η) = (X cos α − Y sin α, X sin α + Y cos α).
Then ξ ± iη = (X ± iY )e±iα , and similarly for the new E±′ ;
1 ∗
(ξ ± iη) = √ (E∓′ e−iωt + E±′ eiωt )
2
with E±′ = e∓iα E± . So E+′ /E−′ = (E+ /E− )e−2iα . We can choose α such that
ρ = E+′ /E−′ becomes real and positive. The angle α and the radius ρ are then
determined by
E+
= ρe2iα .
E−
√
Let E−′ be given by E−′ = 2Beiδ with B, δ ∈ R, B > 0 the polar decomposition
of the complex number E−′ . We have
ξ + iη = Beiδ e−iωt + ρBe−iδ eiωt
ξ − iη = ρBeiδ e−iωt + Be−iδ eiωt
54 Section 2.2

and, consequently,

ξ = B(1 + ρ) cos(ωt − δ)
η = B(1 − ρ) sin(ωt − δ)
ξ2 η2
+ = 1.
B 2 (1 + ρ)2 B 2 (1 − ρ)2
This means that E(t) describes an ellipse with semi-axes B(1 + ρ) and B|(1 − ρ)|
rotated an angle α with respect to ǫ1 and ǫ2 . Their respective lengths are
|E− | |E+ |
B(1 ± ρ) = √ 1 ± .
2 |E− |
If ρ = 1, this ellipse is reduced to a line in the ξ direction. In that case the
polarisation is linear and has direction ξ. If, in contrast, ρ = 0, the polarisation is
circular.

Let us make, as before, ǫ1 = ex , ǫ2 = ey and ǫ± = √1 (ǫ1 ± iǫ2 ). Additionally, we

2
define ǫ3 = ǫ1 ∧ ǫ2 ≡ k̂. It is not difficult to verify that

ǫ∗± ǫ∓ = 0  


ǫ± ǫ3 = ǫ± ǫ3 = 0 
∗
. (2.21)
ǫ∗± ǫ± = 1  


± ǫ∗ = ǫ  ∓

According to what we have shown, the polarisation state of a wave is established

if we can write it in the form (2.19) (or 2.20)) with known amplitudes E1 , E2 (or
E+ , E− ).
 We now ask the opposite question: Given a wave in the form (2.15),
i(k·x−ωt)
E = Re E0 e , how can we determine the polarisation state? In this regard,
a useful formalism was proposed by Stokes: the Stokes parameters.

These are four parameters determined by the measure of intensity and two rela-
tively easy polarisation measures (linear and circular). Let us make

Ẽ = (E1 ǫ1 + E2 ǫ2 )ei(k·x−ωt) = E0 ei(k·x−ωt)

E = Re[Ẽ]
E1 = a1 eiδ1 , E2 = a2 eiδ2
E+ = a+ eiδ+ , E− = a− eiδ−
1
a+ eiδ+ = √ (a1 eiδ1 − ia2 eiδ2 )
2
1
a− eiδ− = √ (a1 eiδ1 + ia2 eiδ2 ).
2
Ruth Durrer Electrodynamique Chap. 2 55

We define

I = |ǫ1 · Ẽ|2 + |ǫ2 · Ẽ|2 = a21 + a22 = |E0|2 





Q = |ǫ1 · Ẽ|2 − |ǫ2 · Ẽ|2 = a21 − a22 
h i . (2.22)
U = 2Re (ǫ1 · Ẽ)∗ (ǫ2 · Ẽ) = 2a1 a2 cos(δ2 − δ1 ) 



h i 

∗
V = 2Im (ǫ1 · Ẽ) (ǫ2 · Ẽ) = 2a1 a2 sin(δ2 − δ1 ) 

This four parameters are not independent. Clearly, I 2 = Q2 + U 2 + V 2 . The Stokes

parameters can also be written in the base of circular polarisations, ǫ± :

∗ 2 ∗ 2 2
I = |ǫ+ · Ẽ| + |ǫ− · Ẽ| = a+ + a−2 

h i 



Q = 2Re (ǫ+ · Ẽ) (ǫ− · Ẽ) = 2a+ a− cos(δ− − δ+ ) 
∗ ∗ ∗
h i . (2.23)
U = 2Im (ǫ∗+ · Ẽ)∗ (ǫ∗− · Ẽ) = 2a+ a− sin(δ− − δ+ ) 



∗ 2 ∗ 2 2 2


V = |ǫ+ · Ẽ| − |ǫ− · Ẽ| = a+ − a−

I = 2h|E|2i measures the “intensity” |E|2 of the wave averaged over a period.
Since E2 = B2 , we have I = hE2 + B2 i, where h. . .i denotes average over a
period T = 2π/ω. The energy density averaged over this interval of time is then
u = I/(8π).

For a linearly polarised wave (δ1 = δ2 ), V = 0, whereas for a circularly polarised

wave (a1 = a2 ) one has Q = 0. In general, Q measures the difference of intensity
in directions ǫ1 and ǫ2 , and V measures the difference of intensity of helicities +
and −.

Exercise: Show that under a rotation of ǫ1 and ǫ2 by an angle α around k̂, the
Stokes parameters transform as
I → I
Q → Q cos(2α) + U sin(2α)
U → U cos(2α) − Q sin(2α)
V → V .

Doppler effect, aberration

For a plane wave, the electromagnetic field is of the form

µ)
F̃µν (x) = fµν e−i(kµ x , Fµν (x) = Re[F̃µν (x)]
56 Section 2.2

where we define k µ = (ω/c, k). Let Fµν be specified in an inertial frame Σ. In a

different inertial frame Σ′ , related to Σ through the Lorentz transformation Λµα ,
it becomes
′ µ
F̃µν (x′ ) = Λαµ Λβν F̃αβ (x) = Λαµ Λβν fαβ e−i(kµ x ) .
For k ′µ = Λµν k ν one has (kµ xµ ) = (kµ′ x′µ ), and so
′ ′ ′µ
Fµν (x′ ) = fµν
′
e−i(kµ x ) , (2.24)
′
with fµν = Λαµ Λβν fαβ . Equation (2.24) shows that (k µ ) is really a four-vector and
that a plane wave that propagates with the four-vector (k µ ) in Σ does the same
in Σ′ with four-vector (k ′µ ) (and is also a plane wave in this transformed system).
We thus have k′ · x′ − ω ′ t′ = k · x − ωt.

We consider a light source in an inertial system Σ and an observer in the system

Σ′ that moves with speed v along the z direction:

We have, β = v/c, z, z’
 
γ 0 0 −βγ
 0 1 0 0 
Λµα = 
 0
 v ϑ, ϑ’ k, k’
0 1 0 
−βγ 0 0 γ
z ′ = γ(z − vt), x′ = x, y′ = y y, y’
ϕ, ϕ’
β
t′ = γ(t − z) x, x’
c
and

kx′ = kx , ky′ = ky 



β . (2.25)
kz′ = γ(kz − c ω)




ω′ = γ(ω − vkz )

Since |k| = ω/c, |k ′ | = ω ′ /c, it follows that

ω ′ = γω(1 − β cos ϑ) . (2.26)

This is the formula of the relativistic Doppler effect.

We also want to determine the direction and amplitude of the transformed wave
vector, k′ . To determine ϑ′ we use
kz′ kz′
cot ϑ′ = 1 = 1 .
(kx′2 + ky′2 ) 2 (kx2 + ky2 ) 2
Ruth Durrer Electrodynamique Chap. 2 57

With (2.25), this gives

ω
γ(kz − β ) k
cot ϑ′ = c = γ cot ϑ − γβ
2 2 21 1
(kx + ky ) (kx + ky2 ) 2
2

 
1 cos ϑ − β
cot ϑ′ = γ cot ϑ − β =γ
sin ϑ sin ϑ
1 sin ϑ
tan ϑ′ = . (2.27)
γ cos ϑ − β
This equation describes the relativistic aberration. Already in the base of symmetry
arguments one sees that the angle ϕ is not modified, and since kx = kx′ , ky = ky′ ,
we have
ϕ′ = ϕ. (2.28)
Relation (2.27) is equivalent to (exercise)


sin ϑ′ = γ1 sin ϑ 


1 − β cos ϑ 


cos ϑ − β (2.29)
cos ϑ′ =
r− β cos ϑ
1 



′ 1+β 

tan ϑ2 = tan ϑ2 .

1−β
Equation (2.26) is the formula for the relativistic Doppler effect. Equations (2.28)
and (2.29) describe the aberration.

For ϑ = 0 (k parallel to v), the longitudinal Doppler effect, one has

 1  1
′ (1 − β)2 2 1−β 2
ω = γω(1 − β) = ω =ω
1 − β2 1+β
or, with λ = c/ν = 2πc/ω s
1+β
λ′ = λ. (2.30)
1−β
We define the redshift Z through the relation
λ′ ∆λ
1+Z = =1+ , ∆λ = λ′ − λ (2.31)
λ λ
 1
1+β 2
= , (2.32)
1−β
for small redshifts, Z ≪ 1, we have Z ≃ β = v/c, the same formula obtained
when analysing the non-relativistic case. The largest redshift observed for distant
galaxies is Z ≈ 10. To which speed does this value correspond?
58 Section 2.2

For ϑ = π/2, transversal Doppler effect, (2.26) implies


ω ′ = γω = √ ω 2 
1−β
p (2.33)

1+Z = 1 − β 2;

for small redshifts, Z ≪ 1, this gives Z ≃ − 12 β 2 . The transversal effect is a purely

relativistic effect and is much harder to perceive for low speeds.

Equations (2.28) and (2.29) determine the aberration of light. Let, for example,
Σ be the rest frame of a star and let ϑ be the angle of incidence with respect to
the Earth’s surface. An observer on Earth moves with velocity v parallel to the
planet’s surface. For ϑ = π/2 (vertical incidence) we have
1
tan ϑ′ = ±
βγ
and for the aberration angle, α = |π/2 − ϑ′ |,

tan α = βγ. (2.34)

In the non-relativistic limit one obtains tan α ≃ β = v/c, which is corrected by

the γ factor.

Already in the base of symmetry arguments it is clear that there is no aberration if

ϑ = 0 or π. For the Earth’s speed, β ≈ 10−4 , the relativistic correction in (2.34) is
not measurable [O(10−8)]. However, for the case of emission of elementary particles
in relativistic motion, this factor γ becomes relevant (for example for the position
of counters in the decay process π 0 → γ + γ for fast π 0 ).

Apparent superluminal velocity

In the ’70s, some radio astronomers discovered that the composing elements of
certain “quasars” move away at speeds four to six times the speed of light c. The
same effect was observed in the nineties in galactic objects. We will show with an
example that the effects of retardation can simulate superluminal relative velocities
of moving objects. To simplify our notation, we remove the third dimension, the
axis z, which does not play a role in our analysis. We consider a disk of gas
centred, at t = 0, in the (x, y) plane of system Σ, and travelling at speed v along
−x direction:
In system Σ′ , which moves at speed v along the −x direction (comoving with
the disk), the disk emits a light flash at time t′ = 0. The observer located at a
distance D receives first the light emitted from point ϕ = 0. He then observes
two rays coming from positions ±ϕ that move at first away from each other (for
Ruth Durrer Electrodynamique Chap. 2 59

R
ϕ x
−ϕ
D

0 < |ϕ| < π/2) and then come closer together (for π/2 < |ϕ| < π). We will see
that (for ϕ small enough) those two rays appear to be moving at superluminal
speed.

With respect to Σ′ the ray event has the coordinates

t′ = 0, x′ = R cos ϕ, y ′ = R sin ϕ (−π < ϕ 6 π).

In system Σ these coordinates correspond to

v ′ γv
t = γ(t′ −
2 x ) = − 2 R cos ϕ
c c
′ ′
x = γ(x − vt ) = γR cos ϕ
y = y ′ = R sin ϕ.

The instant of flashing, t, depends on the position, i.e, on ϕ. The time ta of arrival
at distance D from the center x = y = 0 is given by
 1
c(ta − t) = (D − γR cos ϕ)2 + R2 sin2 ϕ 2 .

Since R ≪ D,
 
γv 1 γv R2
cta = − R cos ϕ + [. . .] 2 ≈ − R cos ϕ + D − γR cos ϕ + O . (2.35)
c c D

The distance between the rays is 2y = 2R sin ϕ. With (2.35) one finds
"  2 # 12
p cta − D
y(ta ) = R2 (1 − cos2 ϕ) = R 1 −
Rγ(1 + β)
  21
2 21
−β
= R − (D − cta ) .
1+β
60 Section 2.2

The relative speed is then

dy 11 1−β
2 = 2× 2c (D − cta ) = 2cγ(1 − β) cot ϕ
dta 2y | {z } 1 + β
| {z } γR(1+β) cos ϕ
1
R sin ϕ
s
1−β
= 2c cot ϕ + O(R/D) . (2.36)
1+β

This is proportional to cot ϕ and diverges for ϕ → 0, even for low speeds. For an
observer, this gives the impression of an object that explodes into two fragments
that move away at superluminal speed (as long as ϕ is small enough).
Chapter 3

Propagation of electromagnetic
waves

3.1 Introduction, ponderable media

We consider a medium with dielectric constant ε and permeability µ, but with

neither currents nor charges; ε are µ assumed to be independent of both position
and time. We have
1
D = εE H = B,
µ
and
1
∇·B = 0 ∇ ∧ E + ∂t B = 0
c
µε
∇·E = 0 ∇ ∧ B − ∂t E = 0,
c
and therefore  
1 2
∆ − 2 ∂t E = 0
v
  (3.1)
1 2
∆ − 2 ∂t B = 0
v
with
c
v=√ . (3.2)
µε
v is the speed of light in the medium. For plane wave-like solutions we obtain

E(x, t) = Re[E0 ei(k·x−ωt) ]

(3.3)
B(x, t) = Re[B0 ei(k·x−ωt) ],

61
62 Section 3.2

with E0 = A1 + iA2 , E0 ∈ C3 , A1 , A2 ∈ R3 , and the dispersion relation

ω2 µε
k2 = 2
= 2 ω2.
v c
Equations ∇ · E = ∇ · B = 0 lead to
k·E=k·B=0 ⇒ k · E0 = k · B0 = 0.
The law of induction gives
ω
k ∧ E0 − B0 = 0.
c
√
Using ω = v|k| = c/ µε |k| this gives
√ √
B0 = µε k̂ ∧ E0 , B= µε k̂ ∧ E. (3.4)
The energy flux is given by the Poynting vector,
r
c c c ε
S= [E ∧ H] = E∧B= E ∧ (k̂ ∧ E).
4π 4πµ 4π µ
Let us consider S at point x = 0; we have
E = Re[(A1 + iA2 )e−iωt ] = A1 cos ωt + A2 sin ωt
and
n
√
E∧B = µε A1 ∧ (k̂ ∧ A1 ) cos2 ωt + A2 ∧ (k̂ ∧ A2 ) sin2 ωt
h i o
+ A1 ∧ (k̂ ∧ A2 ) + A2 ∧ (k̂ ∧ A1 ) sin ωt cos ωt
√

= µε A21 cos2 ωt + A22 sin2 ωt + 2A1 · A2 sin ωt cos ωt k̂
where we have used the relations a ∧ (b ∧ c) = (a · c) b − (a · b) c and A1 · k =
A2 · k = 0. The time average of the Poynting vector over a period T = 2π/ω is
then r r
1 c ε 2 2 c ε
hSi = (A1 + A2 )k̂ = |E0 |2 k̂. (3.5)
2 4π µ 8π µ
In the same way, one finds for the energy density
 
1 2 1 2
u = εE + B
8π µ
 
1 2 1 2 ε
hui = ε|E0| + |B0 | = |E0 |2 (3.6)
16π µ 8π
where we have used (3.4). The time average of |S| is called the intensity I of the
wave: r r r
c ε ∗ c µ ∗ 1
I= Ẽ · Ẽ = H̃ · H̃ = c hui = vhui; (3.7)
8π µ 8π ε µε
and has the units of energy/(area × time), [erg/(cm2 s)].
Ruth Durrer Electrodynamique Chap. 3 63

3.2 Reflection and refraction

As before, we consider electromagnetic waves in homogeneous and isotropic pon-

derable media but we now add interfaces where µ or ε change their value discontin-
uously. In the absence of charges ρ and currents J, we analyse what happens when
an electromagnetic wave crosses the border between two ponderable media with
different dielectric constants ε1 and ε2 and different permeabilities µ1 and µ2 . We
consider the following situation with the incident and reflected plane waves and a
third plane wave that has crossed the bordering surface F between the media:

y
(k, ω) (k’’, ω’’)

(ε1 , µ1 )
α α’
F
x
(ε2 , µ2 )

β
(k’, ω’)

boundary

S
V

R R
Coulomb’s law, ∂V (D · e)ds = 0, and ∂V (B · e)ds = 0, integrated over the
boundary surface ∂V of the volume as indicated in the figure, imply that D⊥ = εE⊥
Hand B⊥ = µH⊥−1areRcontinuous in the interface. Similarly, Ampère’s law gives
∂S
H · ndl = c ∂t S D · eds → 0 for the limit where the thickness of the surface
S indicated in the figure tendsH to 0. Last but not least, the law of induction,
∇ ∧ E + c−1 ∂t B = 0, implies ∂S E · ndl → 0. From the two last equations we
conclude that Ek , Hk are continuous in the interface.

For a continuous component we have (A 6= 0):

′′ ·x−ω ′′ t) ′ ′
Aei(k·x−ωt) + A′′ ei(k = A′ ei(k ·x−ω t) , ∀ x ∈ F.

Denoting by k̃, k̃′ and k̃′′ the projections of k, k′ and k′′ on the plane F , the last
equation can be equivalently written as:
′′ ·x−ω ′′ t) ′ ′
Aei(k̃·x−ωt) + A′′ ei(k̃ = A′ ei(k̃ ·x−ω t) , ∀ x ∈ R3 .

For x = 0, this gives ω = ω ′′ = ω ′ , whereas for t = 0 we obtain k̃ = k̃′ = k̃′′ .

Hence, the projections of the three vectors on the plane F are equal to each other.
The three vectors k, k′ , and k′′ thus live in a single common plane (the incidence
64 Section 3.2

plane),
k, k′ , k′′ ∈ Rk̃ ⊕ Rey .
Considering that
ω ω ω
|k̃| = sin α, |k̃′ | = sin β and |k̃′′ | = sin α′ ,
v1 v2 v1
the fact that all the three vectors are equal leads then to
1 1
α = α′ and
sin α = sin β.
v1 v2
√ √
Let us now go a step further and use v1 = c/ ε1 µ1 and v2 = c/ ε2 µ2 , which gives
√ √
ε1 µ1 sin α = ε2 µ2 sin β,
or, more insightfully,
n1 sin α = n2 sin β (Snell’s law ), (3.8)
where we have introduced the refractive index:
√
n := εµ (Maxwell’s relation). (3.9)
As you can see, the requirement of continuity of a component has enabled us
to find the reflection and refractive angles. However, this is not all we can do;
electrodynamics enables us to compute also the intensities and polarisations. To
this end, we distinguish two different cases:

(i ) E = (0, 0, Ez ) is orthogonal to the plane where the vectors k,, k′ , k′′ lye
[(x, y) plane in the figure, or plane of incidence].
The directions k̂, k̂′ , k̂′′ are given by
k̂ = (sin α, − cos α, 0)
k̂′′ = (sin α, cos α, 0)
k̂′ = (sin β, − cos β, 0).
p
Additionally, H = ε1 /µ1 k̂ ∧ E gives
r r
ε1 ε1
Hx = k̂y Ez , Hy = − k̂x Ez , etc.,
µ1 µ1
which leads to:
“Ez ” “Hx ”
Ez −(ε1 /µ1 )1/2 Ez cos α
Ez′ −(ε2 /µ2)1/2 Ez′ cos β
Ez′′ +(ε1/µ1 )1/2 Ez′′ cos α
Ruth Durrer Electrodynamique Chap. 3 65

Because Ez and Hx are continuous in the interface, we have Ez + Ez′′ = Ez′ and
Hx + Hx′′ = Hx′ . Thus,

Ez + Ez′′ = Ez′
  21
ε2 µ1 cos β
Ez − Ez′′ = Ez′
ε1 µ2 cos α
 1 !
Ez′ ε2 µ1 2 cos β
Ez = 1+
2 ε1 µ2 cos α
 1 !
Ez′ ε2 µ1 2 cos β
Ez′′ = 1− .
2 ε1 µ2 cos α

(ii ) H is orthogonal to the plane of incidence.

In this case we find (simply by substituting E → H, ε → µ and µ → ε)

  21 !
Hz′ µ2 ε1 cos β
Hz = 1+
2 µ1 ε2 cos α
  12 !
Hz′ µ2 ε1 cos β
Hz′′ = 1− .
2 µ1 ε2 cos α

In the important case in which µ1 = µ2 = 1, one has

  21
ε1 n1 sin β
= = .
ε2 n2 sin α

All this leads to the following relations for the cases (i ) and (ii ):

(i ) E normal to the plane of incidence:


! 

2 sin β cos α 

Ez′ = Ez sin α cos β
= 2E z 

1 + sin β cos α sin(α + β)
(3.10)
  

sin β cos α sin α cos β sin(β − α)  

Ez′′ = Ez 1− = Ez . 

sin(α + β) sin β cos α sin(β + α)
66 Section 3.2

(ii ) H normal to the plane of incidence:


! 

2 sin 2α 

Hz′ = Hz = 2Hz 

sin β cos β sin 2α + sin 2β 

1+ sin α cos α 



  

sin 2α sin β cos β 



Hz′′ = Hz 1− 

sin 2α + sin 2β sin α cos α 

| {z }
sin 2β
1− sin (3.11)
2α 





sin 2α − sin 2β 

= Hz 

sin 2α + sin 2β 






sin α cos α − sin β cos β tan(α − β)  

= Hz = Hz . 

sin α cos α + sin β cos β tan(α + β)
To arrive at the last equation we have used the trigonometric identities sin 2α =
tan α tan α±tan β
2 1+tan 2 α and tan(α ± β) = 1∓tan α tan β .

Relations (3.10) and (3.11) are the Fresnel equations (which were derived from
Maxwell’s theory by Lorentz for the first time in 1875).

Remarks:

• We have not used the conditions εE⊥ = D⊥ and µH⊥ = B⊥ , which should
also be continuous at the interfaces, but it is easy to verify that they also
hold.
• In case (i ), there is a reflected ray (if n1 6= n2 ) for all values 0 < α < π/2,
whereas in case (ii ) H′′ and, consequently, the intensity of the reflected ray
vanish for α = αB , defined through αB + βB = π2 . For αB + βB = π2 ,
sin βB = cos αB , and
n2 sin αB sin αB
= = .
n1 sin βB cos αB
The angle αB is known as Brewster’s angle:
n2
tan αB = . (3.12)
n1
If unpolarised light reach, with an angle αB , the surface that divides two
media, the reflected ray is then linearly polarised (its electric field is normal
to the plane of incidence). When the medium (1) is the vacuum (ε1 = 1),
the microscopic explanation of this phenomenon goes as follows: the reflected
ray is created by the oscillations of the dipoles (atoms, molecules) present
in medium (2), parallel to the transmitted ray. If the electric field E thus
produced is parallel to k′′ , the intensity of the reflected wave (which must,
when propagating in direction k′′ , be normal to k′′ ) vanishes.
Ruth Durrer Electrodynamique Chap. 3 67

αB
(1) α B + βB =
π
2
E
(2)
βB

Total internal reflection

This phenomenon occurs when n1 > n2 . In this case the equation sin β =
(n1 /n2 ) sin α has no real solutions for β if α > αT , where
n2
sin αT = .
n1
Since sin β must be real, only the complex values of β that are of the form β =
π/2 + iγ are allowed. Thus
π π
eiβ − e−iβ ei 2 −γ − e−i 2 +γ e−γ + eγ
sin β = = = = cosh γ
2i 2i 2
n1
cosh γ = sin α. (3.13)
n2
The absolute value of γ can then be determined with eq. (3.13). The wave vector
of the transmitted ray is now complex:
n2 ω n2 ω
k′ = (sin β, − cos β, 0) = (cosh γ, i sinh γ, 0).
c c
The phase of the transmitted wave is
′ in2 n2
ei(k ·x−ωt) = eω( c
x cosh γ− c
y sinh γ−it)
. (3.14)
For the amplitude of this wave to decrease when y → −∞, it is necessary to set γ
as a negative quantity. The refracted wave (3.14) propagates then in direction ex ,
along the limit surface, and it is attenuated exponentially for y < 0 (evanescent
wave).

For case (i ) we find

Ez′′ sin( π2 − α + iγ) cos α cosh γ + i sin α sinh γ
= π = .
Ez sin( 2 + α + iγ) cos α cosh γ − i sin α sinh γ
Since this is a ratio between two complex numbers conjugates of each other, we
find that the intensities are equal:
I ′′ E ′′ E ′′ ⋆
= z z⋆ = 1. (3.15)
I Ez Ez
68 Section 3.2

The same is true for case (ii ).

The presence of the wave in medium (2) can be verified experimentally. If one
picks a thin sheet of material (2) and continues in the other side with medium (1)
as shown below,

α > αT
n1 (1)

n2 d (2)

n1 (1)

one finds that the wave is being attenuated by a factor:

n2 ω
− dd
e− c
d sinh γ
= e; 2

c
d2 (ω) := is the length of the attenuation.
n2 ω sinh γ

Intensity

The intensity is given by equation Eq. (3.7), I = |S| = (c/8π)n|E|2. To study the
intensity of waves, let us consider first two extreme cases.

(a) Grazing incidence, α = π/2

From (3.8)
n1
sin β =
n2
sin(α + β) = cos β , sin(β − α) = − cos β.

From (3.10) and (3.11), we therefore have for cases (i ) and (ii )

I ′′
= 1, I ′ = 0. (3.16)
I
So, in the grazing incidence case, the reflection is always total (the mirror image
of mountains at the opposite shore in a lake, or the mirror image of the setting
sun in the sea are examples of this).

(b) Perpendicular incidence, α ≈ 0

Ruth Durrer Electrodynamique Chap. 3 69

In this case, sin α ≈ α and sin β ≈ β, and so n1 α ≈ n2 β. For case (i ), equation

(3.10) gives

Ez′′ n1 n2 (β − α) n1 − n2 I ′′ (n1 − n2 )2
≈ = ; = . (3.17)
Ez n1 n2 (β + α) n1 + n2 I (n1 + n2 )2

Analogously
2  2
Ez′ 2β 4n21
≈ ≈
Ez α+β (n1 + n2 )2
c 2
I′ = n2 Ez′
8π
c
I = n1 Ez2
8π
and then
I′ 4n1 n2
= . (3.18)
I (n1 + n2 )2
The same is obtained for case (ii ).

The reflexion and transmission coefficients are defined by the ratios

ey · S′′ E ′′2 I ′′ ey · S′ n2 cos βEz′2 I ′ cos β

R= = z2 = , T = = = .
ey · S Ez I ey · S n1 cos αEz2 I cos α

In the cases of grazing and perpendicular incidences, these coefficients do not

depend on the polarisation; they are equal for the two linear polarisations (i ) and
(ii ). This is no longer the case in the most general situation. Indeed, in general,
because the reflected intensity depends on the polarisation, the reflected ray is
partially polarised even if the incident ray is not (see figure and exercises).

Clearly, for all values of the incidence angle α, R+T = 1 in the light of energy con-
servation. The behaviour of the transmitted and reflected intensities as a function
of the incidence angle is illustrated in figure 3.1.

Exercise: Show that R + T = 1 in all cases.

3.3 Optics of metals

In this paragraph we consider good conductors (such as metals). Because the elec-
tromagnetic field performs work on the currents, its magnitude decreases quickly
by producing heat if the conductivity σ is high (dissipation). This is why metals
are not transparent but are good reflective materials.
70 Section 3.3

Figure 3.1: The transmission and reflection coefficients are represented for n1 < n2
(top) and for n1 > n2 (bottom left, case (ii ) in dashed line). The angles αB and
αT for n1 > n2 are also indicated (bottom right).

To make our discussion quantitatively concrete, let us make µ = 1 and let us

neglect the dispersion of ε and σ – i.e., the dependence of ε and σ on the frequency.
(This is not an accurate approximation for visible electromagnetic waves, but it is
approximately valid in the infrared regime). In any case, we can assume consider
a wave of a given frequency (monochromatic wave), E ∝ e−iωt . We use Ohm’s law,
which is valid for good conductors:

J = σE.

Maxwell’s equations give

1
∇·H = 0 ∇ ∧ E + Ḣ = 0
c
4π ε 4π
∇·E = ρ ∇ ∧ H − Ė = σE.
ε c c
Ruth Durrer Electrodynamique Chap. 3 71

Taking the divergence of the last equation we obtain

ε 4πσ
∇ · Ė + ∇ · E = 0,
c c
which leads to
4πσ
ρ̇ + ρ = 0,
ε
an expression that can be also deduced from charge conservation ρ̇ + ∇ · J = 0. Its
solution is
t ε
ρ = ρ0 e− τ , τ = , (3.19)
4πσ
where τ is the relaxation time. For high values of the conductivity, τ is a very
short period of time, the density of interior charges vanishes quickly, and we recover
∇ · E = 0, which will be used from now on.

The difference of the curl of the law of induction and the time derivative of
Ampère’s law leads to:  ε 2 4πσ
∆ − 2 ∂t E = 2 Ė. (3.20)
c c
The right-hand side of the above equation acts as an attenuation term. In the
monochromatic case, E and H ∝ e−iωt , eq. (3.20) gives
 
ω 2 ε′
∆+ 2 E = 0. (3.21)
c
with
4πσ √
ε′ = ε + i , (nold = ε). (3.22)
ω
We define 
n′2 = ε′ , n′ = n(1 + iκ) 



Re(n′ 2 ) = ε = n2 (1 − κ2 ) = n2old . (3.23)


4πσ 

Im(n′ 2 ) = = 2n2 κ
ω

We have in addition that

iω
∇ · E = ∇ · H = 0 and H = ∇ ∧ E. (3.24)
c
Equations (3.21) and (3.24) are formally identical to Maxwell’s equations for insu-
lators, except for the fact that here ε′ , the dielectric constant, is complex. Thus,
the solutions for insulators are valid in this case. For a plane wave the wave vector
is now a complex quantity:
ω 2 ′ ω 2 ′2
k2 = ε = 2n , (3.25)
c2 c
ω ′
k = n k̂.
c
72 Section 3.3

The electric field is given by

κnω nω
E = E0 ei(k·x−ωt) = E0 e− c
(k̂·x)
ei( c
k̂·x−ωt)
.

The attenuation of a plane wave is then measured by the attenuation constant:

κnω
χ=2 , I = I0 e−χk̂·x . (3.26)
c
d = 1/χ is the penetration depth. For copper, for example, one has:

c 2π
ω
= λ0 (vacuum) 1 Å= 10−8 cm 1 µ = 10−4 cm 1 cm
d 0.18 × 10−4 cm 0.18 × 10−2 cm 0.18 cm

An electromagnetic wave cannot penetrate through metal. It is because of this

that automobiles can actually protect you from lightings! If the conductivity is
very high, 4πσ/ω ≫ ε, we find, from eq. (3.23), that
r
2πσ
κ ≈ 1 and n ≈ .
ω

Reflection of light off a metallic surface

Given that the equations for conductors are formally identical to those for insula-
tors, we can, by substituting the dielectric constant ε by the complex constant ε′
and n by n′ , repeat the arguments that led us to the reflection and refraction equa-
tions, and in particular we can reproduce Fresnel equations. We should, however,
take care of the interpretation and physical meaning of our results!

From Snell’s law we have

sin α = n′ sin β.
n′ and hence β are complex numbers and this has some interesting consequences.
First, remember that n′ 2 = ε + i4πσ/ω is very large for good conductors (because
σ is very large) and so:
q r
sin2 α
cos β = 1 − sin2 β = 1− ≈ 1.
n′2
For a refracted wave we obtain
ω ′ ′
k′ = n k̂ , k̂′ = (sin β, − cos β, 0) ≈ (sin β, −1, 0),
c
Ruth Durrer Electrodynamique Chap. 3 73

which can be rewritten as

 
′ ′ sin α
k = k0 n , −1, 0
n′
where k0 = ω/c is the wave number in vacuum. With n′ = n(1 + iκ) this gives

′
ei(k ·x−ωt) = exp{i[k0 (x sin α − ny) − ωt] + nκk0 y}. (3.27)
The transmitted wave is then exponentially attenuated for y < 0. Fresnel equations
give in case (i ) (eq. (3.10))
2 sin β cos α 2 sin β cos α
Ez′ = Ez = Ez
sin(α + β) sin β cos α + sin α cos β
(2/n′ ) sin α cos α 2 cos α
≈ Ez ′
≈ Ez (|n′ | ≫ 1). (3.28)
sin α cos α/n + sin α n′
√
For Hx′ we find (from Hx′ = − ε′ Ez′ cos β) that
Hx′ ≈ −2Ez cos α. (3.29)
The component Hx′ remains then finite in y = 0 in the limit |n′ | → ∞, (σ → ∞).
On the other hand, it is easy to verify that the component Hy′ decreases. The
refracted wave does not penetrate through the material and so it is not of our
interest. The reflected wave is more interesting: since β is complex, a phase
difference between the incident and the reflected wave appears (see exercise).

Intensity relations. As in (3.17), we restrict ourselves to the case of normal inci-

dence (α ≈ 0), with n1 = 1:
2 2
I ′′ E ′′ n′ − 1 (n − 1)2 + n2 κ2
= z = =
I Ez n′ + 1 (n + 1)2 + n2 κ2
and then r r
I ′′ 4n 2 ω c
1− = 2 2 2
≈ ≈2 =2 (3.30)
I (n + 1) + n κ n 2πσ λσ
where we have used the fact that ωσ is very large and so κ2 ≈ 1, n ≈ 2πσ/ω ≫ 1.
It is for this reason that metals are ideal reflectors. For copper, for example, one
finds, for λ = 12µ (= 12000 Å, infrared),
I ′′
1− = 1.6 × 10−2
I exp

I ′′
1− = 1.4 × 10−2 .
I (3.30)

Exercise: At which value of its wavelength does metal with conductivity σ lose
its good reflectiveness properties? What happens at this point?
74 Section 3.4

3.4 Dispersion

Until now we have described the properties of materials for which ε, µ and σ are
constants. This approximation is often inadequate, specially for electromagnetic
waves, which represent fast changing fields: a certain time will elapse before ε,
µ and σ react to the variations of the field. An everyday life example of this is
the decomposition of white light by a prism, a phenomenon non compatible with
a constant (frequency-independent) index of refraction. For monochromatic light
(composed by a single constant frequency ω) the results of the previous section
remain valid.

To study the effects resulting from the dependence of ε and µ on ω, we first discuss
a simple model.

Simple model for ε(ω)

Let us assume that the motion of an electron in matter can be seen as a damped
harmonic oscillator,

m[ẍ + γ ẋ + ω02 x] = −eE(x, t). (3.31)

The left-hand side of the above equation is the most general linear approximation,
and so it is always valid for sufficiently small deviations x from equilibrium. For
small amplitude oscillations x, the field E can be evaluated at the electron’s average
position, x = 0. Let us consider a harmonically oscillating field, E ∝ e−iωt . The
same ansatz for x(t) leads to

mx(−ω 2 − iωγ + ω02 ) = −eE

and so the contribution of this electron to the dipole moment is given by

e2 2
p = −ex = (ω − ω 2 − iωγ)−1 E.
m 0
Let us consider a substance composed by N molecules per unit volume with Z elec-
trons each. In each molecule we have fj electrons of natural frequency P
ωj (binding
energy = ~ωj ) and friction coefficient (damping) γj . It follows that j fj = Z,
and the polarisation per unit volume is

Ne2 X
P= fj (ωj2 − ω 2 − iωγj )−1 E = χe E. (3.32)
m j
Ruth Durrer Electrodynamique Chap. 3 75

Thus, we obtain the dielectric constant

4πNe2 X
ε(ω) = 1 + 4πχe = 1 + fj (ωj2 − ω 2 − iωγj )−1 . (3.33)
m j

Clearly, the constants fj , ωj and γj are determined only after having a quantitative
description of the substance. Being incapable of computing these constants in
detail, we limit ourselves to the discussion of some general properties.

Anomalous dispersion and resonant absorption

The constants γj are in general very small with respect to the resonant frequencies,
γj /ωj ≪ 1, and so ε(ω) is almost entirely real (its imaginary component becomes
very small) for most frequencies. For ω < ωj , the factor (ωj2 − ω 2 ) is positive
whereas for ω > ωj it is negative (see Jackson §7).

At low values of the frequency, ω < minj {ωj }, all the terms in the summation
(3.33) are positive, and therefore

ε(ω) > 1 at low frequency.

The higher the frequency, the larger is the number of terms in the summation that
are negative, making the summation become eventually negative:

ε(ω) < 1 at high frequency

The behaviour is very interesting for values close to a resonant frequency ωj : the
real part of the denominator vanishes and the term becomes large and imaginary.
(see figure 3.2).
76 Section 3.4

Figure 3.2: The real (solid line) and imaginary (dashed line) parts of the dielectric
constant as a function of frequency.

If Re[ε(ω)] increases for increasing values of the frequency ω we then speak of

normal dispersion. The opposite case, that in which Re[ε(ω)] vanishes as the fre-
quency increases, goes by the name of anomalous dispersion. Close to a resonant
frequency we then experience the phenomenon of anomalous dispersion. Corre-
spondingly, Im[ε(ω)] conduces to resonant absorption. As we have already seen,
this is easily expressed in terms of the real and imaginary parts of the wave number:

k = β + iα

ω2 
β − α = 2 Re(ε) 
2 2
ω2
c2 k 2 = 2 ε.
ω  c
2βα = 2 Im(ε) 
c
If the absorption is not so high (α ≪ β), one finds that

1 Im(ε) p ω
α≈ β; β= Re(ε) .
2 Re(ε) c

The intensity of the wave decreases with the distance d as

ω Im(ε)
− Im(ε) βd − d√
I = I0 e−2αd = I0 e ( Re(ε) ) = I0 e c Re(ε) .
Ruth Durrer Electrodynamique Chap. 3 77

Electric conductivity at low frequency

In the limit of low frequencies, ω → 0, the behaviour of ε(ω) is qualitatively

different if a fraction f0 of the electrons in each molecule are free, i.e., if they
possess the resonant frequency ω0 = 0. In this case, which corresponds precisely
to metals, the dielectric constant at low frequency has a non-negligible imaginary
part:
4πNe2 f0
ε(ω) = ε0 (ω) + i , (3.34)
mω(γ0 − iω)
where ε0 (ω) accounts for the contribution of bound electrons to the dielectric
constant. If we compare this result with our discussion of metals in the previous
section, it is clear that the conductivity is given by

f0 Ne2
σ= . (3.35)
m(γ0 − iω)

This corresponds roughly to Drude’s conductivity model (1900). Nf0 is the number
of free electrons per unit volume; γ0 is the friction coefficient that depends on the
collision processes of the free electrons with themselves and with the ions of the
metal. For copper, for instance, one has

N ≈ 8 × 1022 atomes/cm3
σ ≈ 5 × 1017 sec−1 at low frequency,

and consequently, for ω < 1011 sec−1 (microwaves)

γ0
≈ γ0 ≈ 3 × 1013 sec−1 .
f0

High frequency limit, plasma frequency

In the domain where ω ≫ maxj {ωj }, equation (3.33) takes the form:

ωp2
ε(ω) = 1 − 2 (3.36)
ω
with
4πNZe2
ωp2 = . (3.37)
m
The value ωp2 , which depends uniquely on the electronic density NZ, is the mate-
rial’s plasma frequency. The dispersion relation reduces in this situation to
q
ck = ω 2 − ωp2 or ω 2 = ωp2 + c2 k 2 . (3.38)
78 Section 3.5

In dielectric media, the limit (3.36) holds only for values ω ≫ ωp . The dielectric
constant is then very close to, but always smaller than 1. It converges to 1 for
ω → ∞. In other situations, such as in our atmosphere’s ionosphere or in the
interior of stars (as well as in certain laboratory plasmas), the electrons are all free
and friction is then negligible. In this last case, the relation (3.36) is valid even for
frequencies ω ≪ ωp .

For ω < ωp the wave number k becomes purely imaginary. This means that any
incident electromagnetic wave will be completely reflected by such a plasma. The
intensity decreases exponentially at the interior of the plasma with an attenuation
constant given by
2ωp
αplasma ≈ for ω ≪ ωp .
c
At very high frequencies the metallic dielectric constant (3.34) is of the form
ωp2
ε(ω) = ε0 (ω) − 2 , ω ≫ γ0
ω
2
4πNf0 e
with ωp2 = .
m
For ω ≪ ωp this gives a negative dielectric constant and light does not penetrate
through the material. But for sufficiently high values of the frequency, ε(ω) > 0 and
the metal becomes transparent! This happens typically in the regime of ultraviolet
waves (a phenomenon known as the ultraviolet transparency of metals).

3.5 General properties of the dielectric “constant”

We proceed under the hypothesis that the polarisation density P depends linearly
on E (linear response) (something untrue in the case of very strong fields). The
most general relation between P and E is then of the form:
Z
Pi (x, t) = d3 x′ dt′ Kij (x, x′ ; t, t′ )Ej (x′ , t′ ).

If the properties of the medium are time-independent, Kij does not depend on
the difference t − t′ . Moreover, if the medium is homogeneous and isotropic, Kij
becomes of the form Kij (x, x′ ; t − t′ ) = δij K(x − x′ ; t − t′ ). Furthermore, the
spatial dependence is usually local,
K(x − x′ ; t − t′ ) = δ(x − x′ ) χ(t − t′ )
Z
P(x, t) = dt′ χ(t − t′ ) E(x, t′ ), i.e. (3.39)

P = χ ∗ E. (3.40)
Ruth Durrer Electrodynamique Chap. 3 79

Here “∗” represents the convolution with respect to time1 .

We restrict ourselves to the case µ = 1, H = B, and we consider dielectrics only.

We perform the Fourier transform with respect to time:

Z ∞
E(x, t) = 2π1
dω Ê(x, ω) e−iωt
−∞
Z ∞
Ê(x, ω) = dt E(x, t) eiωt .
−∞

Let us assume that χ(t) ∈ L1 (R), i.e.

Z
|χ(t)| dt < ∞.

We now use the mathematical result that tells us that for f , g ∈ L1 (R), one has fˆ,
ĝ ∈ C0 (R) ⊂ L∞ (R) ⊂ L1 (R) and f[
∗ g = fˆĝ. For the Fourier transforms we find
therefore the relations
P̂(x, ω) = χ̂(ω) Ê(x, ω) (3.41)
D̂(x, ω) = ε̂(ω) Ê(x, ω) (3.42)
ε̂(ω) = 1 + 4π χ̂(ω). (3.43)
In principle, how ε̂(ω) depends on the frequency is something determined by quan-
tum mechanics. We have discussed in the previous section a simple and classical
model. We now want to discuss some general conditions that one can deduce and
that hold even when the quantum effects are not negligible.

(a) Dissipativity: Matter polarisation absorbs energy; the work performed on the
medium is positive. In a medium where µ = 1, J = ρ = 0, with
1 1 d 2
(E · Ḋ + H · Ḃ) = (E + H2 ) + E · Ṗ,
4π 8π dt
energy conservation gives
1 d 2
∇·S+ (E + H2 ) = −E · Ṗ
8π dt
where we have used D = E + 4πP. Hence, the work performed on the medium by
the electromagnetic field is
Z
dW
= d3 x E · Ṗ,
dt
Z
W = d3 x dt E · Ṗ > 0. (3.44)

1
R
The convolution of two functions, f, g ∈ L1 (R) is defined by (f ∗ g)(t) = f (t − t′ )g(t′ )dt′
One can easily show that f ∗ g = g ∗ f ; (f ∗ g)′ = f ′ ∗ g = f ∗ g ′ .
80 Section 3.6

R R R R ⋆
Parseval’s equation ( |f |2 dt = |fˆ|2 dω) leads to f ⋆ g dt = 2π
1
fˆ ĝ dω.
Z
1
0<W = d3 x dω Ê⋆ (x, ω)(−iω)P̂(x, ω)
2π
Z ∞ Z
1
= dω (−iω) d3 x Ê⋆ (x, ω) · Ê(x, ω) χ̂(ω).
2π −∞

Since χ(t) is real, χ∗ (ω) = χ(−ω) and similarly for E. Thus,

Z Z ∞
3
2πW = dx dω [−iω χ̂(ω) + iω χ̂(−ω)] |E(x, ω)|2
0
Z Z ∞
= d3 x dω 2ωIm [χ̂(ω)] |E(x, ω)|2
0
Z Z ∞
= d3 x dω 2ωIm [ε̂(ω)] |E(x, ω)|2 > 0.
0

As this has to be true for any value of the electromagnetic field E(x, ω), we con-
clude that
Im [ε̂(ω)]
Im [χ̂(ω)] = > 0 ∀ ω > 0. (3.45)
4π
As we had already seen in our simple model, the imaginary part of ε̂ determines
the dissipation (absorption). Any non-stationary process occurring in a realistic
medium is irreversible at some point. This is why there are always losses and
dissipation, that is to say, Im [ε̂(ω)] > 0 for all ω > 0. The set of frequencies for
which Im [ε̂(ω)] is very small constitute the domain of transparency of the medium.

(b) Causality: The polarisation P(x, t) cannot depend on the future value the
electromagnetic field. Consequently,

χ(t) = 0 for t < 0. (3.46)

3.6 Kramers-Kronig dispersion relations

(Jackson, 7.10)

From the causality condition (χ(t) = 0 for t < 0) we deduce here some properties
of the Fourier transform of χ(t),
Z ∞
χ̂(ω) = dt χ(t) eiωt . (3.47)
0
R∞
In this equation, we consider ω a complex variable and we assume that 0 |χ(t)| dt <
∞. In this case, (3.47) is well defined for all ω ∈ C with Im(ω) ≥ 0 because
Ruth Durrer Electrodynamique Chap. 3 81

|eiωt | = e−Im(ω)t ≤ 1 for t ≥ 0. For Im(ω) > 0, the function in (3.47) is analytic
(holomorphic). Besides, χ̂(ω) is continuous and bounded in the plane Im(ω) > 0
and χ̂(ω) → 0 for |ω| → ∞ uniformly in all directions of the complex plane where
Im(ω) > 0. (This is an elementary consequence of the Riemann-Lebesgue lemma:2
χ ∈ L1 (R) ⇒ χ̂ ∈ C0 (R) and for all Im(ω) > 0 the convergence is even faster.)

Let us summarize the properties of the susceptibility χ̂(ω):

(i ) χ̂(ω) is analytic in the region of the complex plane where

Im(ω) > 0.
(ii ) χ̂ is continuous and bounded in the Im(ω) > 0 half-plane
and |χ̂(ω)| → 0 for |ω| → ∞ uniformly in all directions
for which Im(ω) > 0. (3.48)
(iii ) χ̂⋆ (ω) = χ̂(−ω ⋆ ).
(iv ) Im [χ̂(ω)] > 0 for 0 < ω < ∞.
R∞
(v ) χ̂(0) = 0 χ(t) dt > 0.

The last property follows from the fact that the static polarisation “costs” a certain
amount of work. For E(x, t) = H(t)E(x), H being the Heavyside function, that
work is given by (3.44)

Z Z P(∞) Z ∞ Z Z
3 3 2
dx E(x) · dP = dt χ(t) d x E (x) = χ̂(0) d3 x E2 (x) > 0.
0 0

The Kramers-Kronig relations link the real and imaginary parts of χ̂(ω). In order
to derive those relations we need a number of results of the theory of complex
functions that you have already worked out in your math courses (I hope):

Cauchy theorem — Let f : D → C, D ⊂ C be an analytic (holomorphic) function

and let Γ : [0, 1] → D be a closed path whose interior belongs to D. For ω0 ∈ D,
we have

Z
f (ω) dω  0 if ω0 lies outside Γ
=
Γ ω − ω0  2πif (ω ) if ω lies inside Γ.
0 0

We consider the following path

2
See compl. de math.
82 Section 3.6

Im(ω)

R
ρ
Re(ω)
ω0

According to the Cauchy theorem,

Z
χ̂(ω) dω
= 0. (3.49)
Γ ω − ω0
In the limit ρ → 0, the contribution of the semicircle ω = ω0 + ρeiφ , π ≥ φ ≥ 0
gives Z Z 0
χ̂(ω) dω
lim = i lim χ̂(ω)dφ = −iπ χ̂(ω0 ). (3.50)
ρ→0 y ω − ω0 ρ→0 π

On the other hand, the principal value of integral (3.49) is defined by

Z ω0 −ρ Z ∞  Z ∞
χ̂(ω) dω χ̂(ω) dω χ̂(ω) dω
lim + ≡ P.V. .
ρ→0 −∞ ω − ω0 ω0 +ρ ω − ω0 −∞ ω − ω0

In the limit R → ∞ the circle at infinity does not contribute because of property
(ii ) of (3.48) and because of the exponential decreasing for Im(ω) > 0. (3.49) then
gives Z ∞
χ̂(ω) dω
P.V. − iπ χ̂(ω0 ) = 0.
−∞ ω − ω0
The real and imaginary parts of this equation are the Kramers-Kronig relations:
Z ∞ 
1 Im [χ̂(ω)] 

Re [χ̂(ω0 )] = P.V. dω  
π −∞ ω − ω 0
Z ∞ . (3.51)
1 Re [χ̂(ω)] 

Im [χ̂(ω0 )] = − P.V. 
dω 
π −∞ ω − ω0

From property (iii ) we conclude (for ω ∈ R)

Re [χ̂(ω)] = Re [χ̂(−ω)]
Im [χ̂(ω)] = −Im [χ̂(−ω)] .
Ruth Durrer Electrodynamique Chap. 3 83

We can go further and write (3.51) in the form

Z ∞ 
2 ωIm [χ̂(ω)] 

Re [χ̂(ω0 )] = P.V. dω 

π 0 ω 2 − ω02 (3.52)
Z ∞
2ω0 Re [χ̂(ω)] 

Im [χ̂(ω0 )] = − P.V. dω . 

π 0 ω 2 − ω02

The first equation implies for the dielectric function ε̂(ω) = 1 + 4π χ̂(ω) that:
Z ∞
2 ωIm [ε̂(ω)]
Re [ε̂(ω0 )] = 1 + P.V. dω. (3.53)
π 0 ω 2 − ω02

Equations (3.51) to (3.53) are the Kramers-Kronig dispersion relations. They show
that the absorption properties (Im [χ̂]) together with causality determine entirely
the dispersion Re [χ̂] and vice-versa. These relations also play an important role in
particle physics: diffusion and absorption are intimately related via the causality
condition. Equations (3.52) were first derived by Kramers (1927) and p Kronig
(1926), whom wrote them as relations for the refractive index n(ω) = ε̂(ω).
(Relation (3.39) can also be justified in the case in which χ(t) is a distribution.)

Theorem — The properties (3.48) of the susceptibility lead to the following results:

• χ̂(ω) does not take real values in the half-plane Im(ω) > 0 apart from the
imaginary axis.

• Along the imaginary axis χ̂(ω) decreases monotonically from χ0 ≡ χ̂(0) > 0
to lims→∞ χ̂(is) = 0.

In particular, χ̂ has no zeros in the upper half-plane.

Proof — We use a result of the theory of complex functions: let D ⊂ C be an open

set and f : D → C a meromorphic function (analytic in all its domain except in
a set of isolated points zi in which the function diverges as (z − zi )−ni , zi being
then called a pole of order ni ). Let Γ be the boundary of a compact subset of D,
K ⊂ D, such that f has no poles on Γ and f (z) 6= a ∀z ∈ Γ. In this situation it is
possible to show that Z
1 f ′ (z) dz
=Z−P
2πi Γ f (z) − a
where Z is the number of zeros (multiplied by their order) and P is the number of
poles (multiplied by their order) of f − a in K. (This result is a simple corollary
of the residue theorem.)

We apply this result to the function χ̂(ω) by choosing Γ as shown below in the
figure:
84 Section 3.6

Im(ω)

R
Re(ω)

In the upper half-plane χ̂ has no poles. Hence

Z ′
1 χ̂ (ω) dω
= Z.
2πi Γ χ̂(ω) − a

Let Γ′ be the image of Γ under the application ω 7→ χ̂(ω). We take the limit
R → ∞. Since Im [χ̂(ω)] > 0 for 0 < ω < ∞ [property(iv )] and thus Im [χ̂(ω)] < 0
for −∞ < ω < 0, Γ′ = χ̂(Γ) intersects the real axis only at χ0 = χ̂(0) and
0 = χ̂(∞). Γ′ looks then as follows

Im(χ)
Γ’

a Re(χ)
0 χ0

We conclude that

Z ′ Z  1 if 0 < a < χ
1 χ̂ (ω) dω 1 dχ̂ 0
Z= = =
2πi Γ χ̂(ω) − a 2πi Γ′ χ̂ − a  0 if a ∈ R\[0, χ0 ].

Hence, every value a ∈ [0, χ0 ] is reached exactly once by the function χ̂. But
since [χ0 , 0] is the image of the imaginary half-plane iR+ , there exists x ∈ R+
such that χ̂(ix) = a. Therefore χ̂ (which is real on the imaginary axis) decreases
monotonically along the imaginary axis. The real values of χ̂ are just the images
of iR+ (on the upper half-plane).

This theorem allows us to unambiguously define the refractive index as

p
n(ω) = ε̂(ω). (3.54)
Ruth Durrer Electrodynamique Chap. 3 85

For z ≡ 4π χ̂(ω) we then have

√
n(z) = 1 + z.

We choose the root by limiting the z plane from −1 to −∞

Im(z)
0<ω<∞

Re[n(z)] > 0 4πΓ’

Im[n(z)] > 0
n(0) = 1 Re(z)
−1 z(∞) z(0)
Re[n(z)] > 0
Im[n(z)] < 0

−∞ < ω < 0

In the reduced plane, D = C\{z ∈ R, z 6 −1} we uniquely define the root by

n(0) = 1. This gives Re(n) > 0 and Im(n) > 0 for Im(z) > 0, Re(n) > 0 and
Im(n) 6 0 for Im(z) 6 0.

If we choose the path Γ in the plane ω as in the theorem, we find its image in the
plane z = 4π χ̂(ω). The upper circle |ω| = R is applied on 0 in the limit R → ∞.
The imaginary axis, iR+ is applied on the real interval [0, z(0)]. The image of the
upper half-plane Im(ω) > 0 is the interior of the path 4πΓ′ .

From these facts we can draw the following conclusions:

(i ) n(ω) is analytic for Im(ω) > 0.

(ii ) n(ω) is continuous and bounded for Im(ω) > 0 and |n(ω) −
1| → 0 for |ω| → ∞ uniformly in all the directions for which
Im(ω) > 0.
(3.55)
(iii ) n⋆ (ω) = n(−ω ⋆ ).
(iv ) 1 6 n(0) < ∞.
(v ) Im [n(ω)] > 0 for ω ∈ R+ and Im [n(ω)] < 0 for ω < 0
Re [n(ω)] > 0 for all ω with Im(ω) > 0.

Along the same lines as those leading to (3.52), we can derive a dispersion relation
for n:
Z ∞ ′
2 ω Im [n(ω ′ )] ′
Re [n(ω)] = 1 + P.V. dω , (3.56)
π 0 ω ′2 − ω 2
86 Section 3.7

Z ∞
2ω Re [n(ω ′ ) − 1] ′
Im [n(ω)] = − P.V. dω . (3.57)
π 0 ω ′2 − ω 2

3.7 Electromagnetic waves in dispersive media

(Jackson, 7.11)

We first consider a homogeneous and isotropic medium with a dynamic dielectric

function ε(ω), with µ = 1 and with neither charges nor currents (ρ = J = 0). We
decompose the fields E(x, t) and B(x, t) into modes of frequency ω, Ê(x, ω) and
B̂(x, ω) (the Fourier transforms of E and B with respect to time). Ê and B̂ are
complex but they satisfy the reality condition Ê⋆ (ω) = Ê(−ω) (and similarly for
B̂).

In the Maxwell equations for Ê and B̂ the derivatives are replaced by factors −iω,
i.e.
iω
∇ · Ê = 0 ∇ ∧ Ê − B̂ = 0
c
iω
∇ · B̂ = 0 ∇ ∧ B̂ + ε(ω)Ê = 0.
c
As before, these equations lead to the wave equations for Ê and B̂:



∆ + k 2 Ê = ∆ + k 2 B̂ = 0 (3.58)

with
ω2 2
k2 = n (ω). (3.59)
c2
Given that ε = n2 , then k 2 is also complex. Equations (3.58) have as solutions the
following plane waves:
Ê(x, ω) = E0 (ω) eik·x,
p
with k = (ω/c)n(ω)k̂, k̂ ∈ R3 , |k̂| = 1 and n(ω) = ε(ω) where this square root
is defined as in the previous paragraph:

Re [n(ω)] > 0 and Im [n(ω)] > 0 for ω > 0

Re [n(ω)] > 0 and Im [n(ω)] < 0 for ω < 0.
Z
1
dω E0 (ω) ei( c Re[n(ω)]k̂·x−ωt) e− c Im[n(ω)]k̂·x .
ω ω
E(x, t) =
2π
Since ω Im [n(ω)] is always positive, the last factor always represents an attenuation
in the direction in which the wave propagates. As in the optics of metals, we
thus obtain an exponentially damped wave that propagates in direction k̂. The
wavelength that corresponds to a given frequency ω is determined by Re [n(ω)]
Ruth Durrer Electrodynamique Chap. 3 87

(dispersion). The attenuation is determined by Im [n(ω)] (damping). If Re [n(ω)]

is an increasing function, the dispersion is said to be “normal”. Accordingly, if
Re [n(ω)] is a decreasing function, we speak of “anomalous” dispersion.

Several different velocities play a role in the description of the propagation of

electromagnetic waves in dispersive media.

Phase velocity

As we have seen, the crest (surface of constant phase) of a wave of the form
E(x, t) = E0 exp {i(k · x − ωt)} with k = (ω/c) n(ω)k̂ propagates in the direction
k̂ with the velocity
ω c
vphase = = .
|Re[k]| Re[n(ω)]
For a refractive index Re[n(ω)] < 1, the phase
p velocity is greater than c (example:
high frequency plasmas, ω > ωp , or ck = ω 2 − ωp2 ).

Group velocity vg ≡ ∂ω/∂k

We consider ω as a function of k = |k|: ω = ω(k). We study, for simplicity, a scalar

wave (the E1 component of the electric field, say). In particular, we consider a
wave packet of the form

φ(x, t) = Re [ψ(x, t)]

Z
1
ψ(x, t) = d3 k a(k) ei(k·x−ωt) ,
(2π)3

where a(k) is focused around a value k0 , i.e. a(k) is different from zero only for
values of k sufficiently close to k0 . The energy density is proportional to φ2 . We
define the energy centre, hx(t)i, through
R 3
d x x φ2 (x, t)
hx(t)i ≡ R 3 2 .
d x φ (x, t)

φ = 21 (ψ+ψ ⋆ ) and so φ2 = 14 ψ 2 + 14 (ψ ⋆ )2 + 12 ψψ ⋆ . The first two terms oscillate rapidly

at frequencies close to 2ω(k0) (Zitterbewegung, “trembling motion”) whereas the
term |ψ|2 varies slowly with respect to time. We neglect the first two terms, which
vanish in time averages taken over an interval much longer than the wave’s period
T0 ≈ 2π/ω(k0). The oscillating contributions taken away, we are left with:
R 3
d x x |ψ(x, t)|2
hx(t)i = R 3 .
d x |ψ(x, t)|2
88 Section 3.7

To compute the denominator we use

Z Z
3 2 1 ′ ′
d x xj |ψ(x, t)| = d3 x d3 k d3 k ′ xj ei(k−k )·x a(k)a⋆ (k′ ) e−i(ω−ω )t .
(2π) 6 | {z }
′ )·x
−i dkd ei(k−k
j

Integration by parts conduces to

Z  
1 3 3 3 ′ i(k−k′ )·x ⋆ ′ ∂a ′
hxj (t)i = 6
d xd kd k e a (k )i e−i(ω−ω )t +
(2π) ∂kj
.Z
∂ω −i(ω−ω′ )t
+t a⋆ (k′ )a(k) e d3 x |ψ(x, t)|2 .
∂kj
3
RThe3 integration over
R 3d x gives the distribution (2π)3 δ(k − k′ ) and so we find, with
2 1 2
d x |ψ| = (2π)3 d k |a| , that:
R ∂a
R 3 ∂ω 2
d3 k a⋆ (k)i ∂k dk |a|
hxj (t)i = R j
+ t R ∂kj .
d3 k |a|2 d3 k |a|2

For the speed of propagation of the centre of energy, we obtain

d ∂ω
hx(t)i = h i
dt ∂k
R 3 ∂ω 2
∂ω d k |a|
h i ≡ R 3 ∂k 2 ,
∂k d k |a|

the average of the group velocity, vg , defined by vg = ∂ω∂k

. In the isotropic case,
dω dω dv
ω = ω(|k|), vg = k̂ dk , vg = |vg | = dk or, with ω = vphase k, vg = vphase + k phase
dk
.
ck
The relation ω = Re[n(ω)] gives

dω 1 c
vg = = = . (3.60)
dk dk/dω Re(n) + ω d[Re(n)]
dω
 
d[Re(n)]
For normal dispersion dω
>0 and Re(n) > 1 that leads to

vg < vphase < c.

d[Re(n)]
For anomalous dispersion, dω
can become negative. In this scenario, it is
possible that vg ≫ vphase and, moreover, that vg > c. Large values of d[Re(n)]
dω
traduce to a fast variation of ω as a function of k. In this case, the oscillating
terms in the expression for hx(t)i cannot be neglected and the center of energy
moves in a very complicated way!
Ruth Durrer Electrodynamique Chap. 3 89

Speed of propagation

Here we show that the true speed of propagation is always 6 c, even if vg and
vphase can be greater than c. This is a consequence of the analytic properties of
n(ω) (which are themselves a consequence of causality).

We consider a linearly polarised wave that represents the superposition of plane

waves of direction e = (1, 0, 0), such that the vector E has non-zero component
only along y, E(x, t) = E(x, t):

E = (0, E(x, t), 0)

Z ∞
E(x, t) = 1
2π
dω A(ω) ei(kx−ωt)
−∞
Z ∞
A(ω) = dt E(0, t)eiωt
−∞

We assume that this wave reaches x = 0 at t = 0, and not before, i.e.:

E(0, t) = 0 for t < 0,

hence Z ∞
A(ω) = dt E(0, t)eiωt .
0

We now show that in this case E(x, t) = 0 for t < xc , which means that vsignal 6 c).
2
For simplicity, we assume in our proof that |E(0, t)|, | dtd E(0, t)| and | dtd 2 E(0, t)|
are integrable over 0 6 t < ∞. This is the case if, for example, E(0, t) describes a
pulse of finite duration. Under these considerations, A(ω) is analytic for Im(ω) > 0,
and bounded and continues for Im(ω) > 0. Since the time derivative of E(0, t)
corresponds to a multiplication of A by a factor −iω, the same properties hold for
ωA and ω 2A. Since A and ω 2 A are bounded for Im(ω) > 0, there exists a constant
a > 0 such that
a
|A(ω)| 6 for Im(ω) > 0.
1 + |ω|2
iωn(ω)
n(ω) is analytic for Im(ω) > 0 and so ei(kx−ωt) = e( c
x−iωt)
is also analytic for
Im(ω) > 0.
x x
ei(kx−ωt) = e−Im(ω)( c −t) e−Im[ω(n(ω)−1)] c
Im [ω(n(ω) − 1)] = Im(ω)(Re(n) − 1) + Re(ω)Im(n).

The last term is always non-negative, according to property (v ) of (3.55). We thus

find
a h x x i
i(kx−ωt)
A(ω) e 6 exp −Im(ω) − t + (Re [n(ω)] − 1) .
1 + ω2 c c
90 Section 3.8

For |ω| → ∞, Re [n(ω)] → 1. If xc − t > 0 the inequality written below is satisfied

for sufficiently large values of |ω|:
a
A(ω) ei(kx−ωt) 6 e−αIm(ω)
1 + |ω|2
x
with α = c
− t + (Re [n(ω)] − 1) xc > 0. If we substitute the integral
Z ∞
1
E(x, t) = dω A(ω) ei(kx−ωt)
2π −∞
by the integral along the path ΓR ,
Im(ω)

ΓR
R

Re(ω)
0
In the limit R → ∞, the large circle does not contribute and
Z
1 x
E(x, t) = lim dω A(ω) ei(kx−ωt) for − t > 0.
2π R→∞ ΓR c
But A(ω) ei(kx−ωt) is analytic in the upper half-plane and so, according to Cauchy’s
theorem, the integral of this function along the path ΓR vanishes, which gives
x
E(x, t) = 0 for − t > 0.
c
One can show by the same reasoning that for a wave such that E(x, t) = 0 for
|x| > L at t < 0, E(y, t) = 0 for |y| > L + ct, independently of the specific form
of the dispersion law.

Even if the phase and group velocities can be greater than c, an electromagnetic
signal does not propagate at a speed higher than c.

3.8 The optical limit

The wavelength of visible light is within the range (4 − 7) × 10−5 cm. This is,
from a macroscopic viewpoint, a very small length, and the limit λ → 0 is often
a good approximation for the propagation of light. In this limit one obtains the
geometrical optics (ray optics), which we derive in this section. The following
considerations are completely equivalent to those that lead to classical mechanics
as a limit of quantum mechanics.
Ruth Durrer Electrodynamique Chap. 3 91

The Eikonal equation

In addition to the wavelength λ, we consider a macroscopic distance L along which

the amplitude and polarisation of the wave change in a significant manner. Geo-
metrical optics is a good approximation if

λ ≪ L.

We analyse the case of an isotropic and insulating medium that is overall inhomoge-
neous with respect to the scale L. We neglect the imaginary part of ε (absorption).
A quasi-plane wave is an electromagnetic field of the form

E(x, t) = E0 (x, t) eiφ(x, t) 
, (3.61)
B(x, t) = B (x, t) eiφ(x, t) 
0

where E0 , B0 , ∇φ and ∂t φ vary much more slowly than φ. For an authentic plane
wave, all those magnitudes are constant: E0 and B0 are constants; φ = k · x − ωt
and therefore ∇φ = k and ∂t φ = ω.
 
−1 ∂i E0j ∂i B0j ∂j ∂i φ ∂i ∂t φ
L = max ∂i ε, ∂i µ, , , , , .
E0j B0j ∂i φ ∂t φ
Maxwell’s equations (in the absence of charges and currents) are

1
∇·B = 0 ∇ ∧ E + Ḃ = 0
c
1
∇·D = 0 ∇ ∧ H − Ḋ = 0.
c
We insert in the above relations the fields (3.61) and neglect all the derivatives of
ε, µ, E0 and B0 . This leads to

1 
∇φ · B0 = 0 ∇φ ∧ E0 + φ̇ B0 = 0  
c . (3.62)
µε 

∇φ · E0 = 0 ∇φ ∧ B0 − φ̇ E0 = 0 
c
We conclude that E0 , ∇φ and B0 form an orthogonal system in the trigonometric
sense. From (3.62) we obtain −B0 = φ̇c (∇φ ∧ E0 ) and so
 
c µε
∇φ ∧ ∇φ ∧ E0 = − φ̇ E0 ,
φ̇ c
µε
−∇φ ∧ (∇φ ∧ E0 ) = 2 φ̇2 E0 ,
c
µε
E0 (∇φ)2 = 2 φ̇2 E0
c
92 Section 3.8

where
µε 2
(∇φ)2 = φ̇ (3.63)
c2
(3.63) is the eikonal equation (φ is the eikonal of geometrical optics).

In most applications, ε and µ are time-independent. In this case, the wave has a
constant frequency,
φ(x, t) = χ(x) − ωt.
With n2 = εµ, the eikonal equation leads then to
n2 (ω)ω 2
(∇χ)2 = . (3.64)
c2
This is the core equation of geometrical optics. The surfaces defined by χ(x) =
constant are the constant-phase surfaces or wavefronts.

Light rays

The time averages of the electric and magnetic energy densities are
ε
huei = E0 · E⋆0
16π
1
hum i = B0 · B⋆0 .
16πµ
With (3.62) one finds [by using B⋆0 · (∇χ ∧ E0 ) = E0 · (B⋆0 ∧ ∇χ)]
c/ω
hue i = E0 · (B⋆0 ∧ ∇χ) = hum i. (3.65)
16πµ
For the time average of the Poynting vector we obtain
c
hSi = Re [E0 ∧ B⋆0 ]
8πµ
c2
= Re [E0 ∧ (∇χ ∧ E⋆0 )]
8πµω
c2
= (E0 · E⋆0 )∇χ
8πµω
2c2
= hue i∇χ
n2 ω
or, with hui ≡ huei + hum i = 2huei,
c ∇χ
hSi = hui .
n nω/c
Ruth Durrer Electrodynamique Chap. 3 93

But from the eikonal equation (3.64)

∇χ
ŝ ≡ (3.66)
(nω/c)
is a unit vector and
hSi = vphase huiŝ (3.67)
where vphase = nc is the phase velocity in the medium. The time average of the
Poynting vector has the same direction as ŝ and its amplitude is the average energy
density multiplied by the phase velocity.

The rays of light are integral curves of the vectorial field ŝ; they are therefore
trajectories orthogonal to the surfaces χ = constant. These integral curves satisfy
the equation
dx   nω dx  
= s x(s) or = ∇χ x(s) . (3.68)
ds c ds
Since (dx/ds)2 = ŝ2 = 1, s is the arclength parameter. We can derive, from (3.68)
and (3.64), a differential equation for x(s) containing only n(x), the refractive
index determined by the properties of the medium:
 
d dxj c d h  i c dxi
n = ∂j χ x(s) = · ∂i ∂j χ =
ds ds ω ds ω ds
(c/ω)2 (c/ω)2 1
= ∂i χ · ∂i ∂j χ = ∂j [∇χ]2 = ∂j n2 =
n 2n 2n
= ∂j n.

We have then found the following equation for the rays:

 
d dx(s)
n = ∇n. (3.69)
ds ds

Fermat’s principle

Fermat’s principle states that the light that arrives at point P2 coming from point
P1 takes the path along which the integral
Z P2
n ds (3.70)
P1

attains its minimum value as compared to the values it would have along neigh-
bouring paths with the same ending points. Let us consider a path γ neighbour of
the path R taken by light:
94 Section 3.8

γ P2

A x(s)

R
P1

R
Since (ω/c) nŝ = ∇χ, ∇∧(nŝ) = (c/ω)∇∧(∇χ) = 0. So we have A
∇∧(nŝ) da = 0.
But, after Stokes’ theorem,
Z Z Z
∇ ∧ (nŝ) da = nŝ · dl − n ds = 0.
A γ R
R R
With γ
nŝ · dl 6 γ
n dl we then obtain
Z Z
n ds 6 n dl.
R γ

In the exercises you will show that (3.69) is nothing but the Euler-Lagrange equa-
tion for the Lagrangian given by (3.70)

Hamiltonian formulation

We now show that the eikonal equation, (∇χ)2 = n2 ω 2 /c2 , can be interpreted as a
Hamilton-Jacobi equation with S ∗ = χ−ωt. (See the analytical mechanics course.)

This means that the function

H(k, x) = ω
where k = ∇χ is a Hamilton function and H is determined by the solution of
k2 = n2 ω 2 /c2 (3.71)
for the variable ω.

We want to show that the eikonal equation implies the canonical equations for
(k, x). With this in mind, we first consider a focused (tenuously-varying) wave
group with wave vector k. For the wave’s centre of energy, x(t), the vector ẋ(t) is
the group velocity,
∂ω ∂H
ẋ(t) = = ; (3.72)
∂k ∂k
which constitutes the first canonical equation. To arrive at the second canonical
equation we use the fact that k = ∇χ and so
d ∂2χ ∂ 2 χ ∂H
k̇j = ∂j χ = j i ẋi = j i . (3.73)
dt ∂x ∂x ∂x ∂x ∂ki
Ruth Durrer Electrodynamique Chap. 3 95

2
We still need to prove that ∂x∂j ∂x
χ ∂H ∂H
i ∂k = − ∂xj . To arrive at this result, we differen-
i
tiate eq. (3.71) with respect to kj and xi . This gives

1 ∂
∂H
2k j = 2
n2 (ω)ω 2 (3.74)
c ∂ω ∂kj
2nω 2 ∂n 1 ∂
∂H
0 = 2 j
+ 2 n2 (ω)ω 2 . (3.75)
c ∂x c ∂ω ∂xj
Here we have used the fact that (k, x) are independent variables and ω = H(k, x).
We now differentiate the eikonal equation with respect to xj . (Note that in the
eikonal equation ω is just a parameter and χ and n are not functions of x.):

∂χ ∂ 2 χ 2nω 2 ∂n
2 = .
∂xj ∂xi ∂xj c2 ∂xj
With eq. (3.75) this gives

∂χ ∂ 2 χ 1 ∂ 2


2 ∂H
2 = − n (ω)ω .
∂xj ∂xi ∂xj c2 ∂ω ∂xj
∂χ
Substituting ∂xj
by kj we find, with (3.74), that

1 ∂ 2


2 ∂H ∂ χ
2
1 ∂
∂H
2
n (ω)ω i j
=− 2 n2 (ω)ω 2 .
c ∂ω ∂kj ∂x ∂x c ∂ω ∂xj
∂(n2 (ω)ω 2 )
Under the hypothesis that ∂ω
6= 0, this implies

∂ 2 χ ∂H ∂H
i j
= − j,
∂x ∂x ∂kj ∂x

which is precisely what we wanted to show. With (3.73) we thus arrive at the
second canonical equation,
∂H
k̇j = − j . (3.76)
∂x
For x(t) and k(t) we then have the canonical equations

∂H ∂H
ẋj = , k̇j = − .
∂kj ∂xj

Exercise: By using (3.74) and (3.75) one can re-derive equation (3.69) of light
rays.
Chapter 4

Emission of electromagnetic waves

So far we have studied the propagation of electromagnetic waves without being

concerned about how they are originated. However, there is something we already
know from Maxwell’s equations: the electromagnetic fields are produced by charges
and currents. How the electromagnetic waves are created by charges and currents
that vary in time is precisely the subject of this chapter.

4.1 Wave zone, multipole expansion

In the Lorentz gauge (∂µ Aµ = 0), Maxwell’s equations (with µ = ε = 1) take the
form [(1.55) and (1.56)]:

4π
⊓φ = 4πρ and ⊔
⊔ ⊓A = J,
c
with the retarded solution (see chap. I)
Z
ρ (x′ , t − |x − x′ |/c) 3 ′
φ(x, t) = dx (4.1)
|x − x′ |

and Z
1 J (x′ , t − |x − x′ |/c) 3 ′
A(x, t) = d x. (4.2)
c |x − x′ |
These expressions represent retarded potentials, which are the unique solutions of
equations (1.55) and (1.56) and satisfy the condition that

φ(x, t) = 0, A(x, t) = 0 if
ρ(x′ , t′ ) = 0, J(x′ , t′ ) = 0 ∀ x′ ∈ R3 with |x − x′ | < c(t − t′ ) .

96
Ruth Durrer Electrodynamique Chap. 4 97

Consider a finite region of size d that contains charges and currents. Let us deter-
mine the energy emitted by this region (which we will call the source) at a large
distance:

Source x

x’

We place the origin of the coordinate system at a point of the source. Since
|x| ≫ |x′ |, we can perform the expansion (n = x/|x|, r = |x|)


|x − x′ | = r − n · x′ + O x′2 /r . (4.3)

The dominant terms in (4.1) and (4.2) decrease as 1/r. In order to get them, we
can replace the denominator with 1/r. For the retarded time we use eq. (4.3):
|x − x′ | r n · x′

tret = t − =t− + + O x′2 /rc . (4.4)
c c c
The last term, O(x′2 /rc), is negligible only if the temporal variation of ρ and J
during the interval d2 /rc is small (d is the diameter of the surface). In the case of
harmonic oscillations (∼ eiωt ), this translates into the condition
d2 1
≪ .
rc ω
In the general case, ω represents a typical frequency of the source. The wave zone
is defined by the conditions
d2 ω
r ≫ d and r ≫ . (4.5)
c
Within the wave zone we can then approximate the electromagnetic potentials by
Z  
1 ′ r n · x′
φ(x, t) ≃ ρ x, t− + d3 x′ (4.6)
r c c
Z  
1 ′ r n · x′
A(x, t) ≃ J x, t− + d3 x′ . (4.7)
rc c c
Let us now compute the contributions O(1/r) of the electromagnetic field, keeping
in mind that in the limit r → ∞ the higher order terms do not contribute to the
energy emission in the cone of given aperture angle ∆Ω:
Z    
n ′ r n · x′ 3 ′ 1
−∇φ(x, t) = ρ̇ x , t − + d x +O .
rc c c r2
98 Section 4.1

The continuity equation gives [t′ ≡ t − r/c + n · x′ /c]

 
′ ′ ′ ′ ′ ′ ′ r n · x′ 1
ρ̇(x , t ) = −∇ · J(x , t )|t′ = −∇ · J x , t − + + n · J̇(x′ , t′ )|t ,
c c t c
where (. . .)|t′ implies that the derivative is taken at t′ constant. Gauss’s theorem
gives Z   Z
′ ′ r n · x′ 3 ′
∇ ·J x, t− + dx = J · n′ dΩ = 0 .
c c S∞

Neglecting terms of order 1/r 2 we obtain

Z  
1 ′ r n · x′
−∇φ(x, t) = 2 n n · J̇ x , t − + d3 x′ ; (4.8)
rc c c
and Z  
1 1 ′ r n · x′
− Ȧ(x, t) = − 2 J̇ x , t − + d3 x′ .
c rc c c
Which leads finally to
  Z  Z 
1 1 ′ ′ 3 ′ ′ 3
E(x, t) = −∇φ − Ȧ = 2 n · n · J̇(x , t ) d x − (n · n) J̇(x , t ) d x
c rc
 Z   
1 ′ r n · x′ 3 ′
= n ∧ n ∧ J̇ x , t − + dx , (4.9)
rc2 c c
where the relation a ∧ (b ∧ c) = b(a · c) − c(a · b) has been used. For B, one
concludes from B = ∇ ∧ A that:
Z  
1 ′ r n · x′
B(x, t) = − 2 n ∧ J̇ x , t − + d3 x′ + O(1/r 2) ,
rc c c
and so
B = − 1c n ∧ Ȧ
(4.10)
E = B ∧ n = 1c n ∧ (n ∧ Ȧ)
and Z  
1 ′ r n · x′
Ȧ(x, t) = J̇ x , t − + d3 x′ . (4.11)
rc c c

We see then that a current J that is time-independent does not emit electromag-
netic waves!

First, let us consider the monochromatic case (harmonic oscillation):

 
J(x, t) = Re J(x)e−iωt
Z
J̃(k) = J(x)e−ik·x d3 x.
Ruth Durrer Electrodynamique Chap. 4 99

We have
Z   
1 ′ −iω(t−r/c+n·x′ /c) 3 ′ e−iω(t−r/c)
A(x, t) = Re J(x )e d x = Re J̃(k)
rc rc
 
with k = (ω/c)n. This gives A(x, t) = Re ǫ(k)ei(k·x−ωt) with ǫ(k) = J̃(k)/rc
and so  
1 iω i(k·x−ωt)
B = − n ∧ Ȧ = Re n ∧ ǫ(k)e , E = B ∧ n.
c c
The time average of the Poynting vector is
c c  ω 2
hSi = hE ∧ Bi = n |n ∧ ǫ(k)|2 . (4.12)
4π 8π c
To arrive at this result we have used the fact that hE ∧Bi = h(B ∧n) ∧Bi = nhB 2 i
and for A(t) = Re [Aeiωt ], hA(t)2 i = |A|2 /2.

The average energy emitted per unit time within the solid angle dΩ in direction n
is
dP 1
(n) = (hSi · n)r 2 = 3
|n ∧ J̃(k)|2 ω 2.
dΩ 8πc

Multipole expansion

(Jackson, §§9.2 and 9.3)

We assume here that eq. (4.5) holds as well as the condition

λ ≫ d ⇒ T = λ/c ≫ d/c. (4.13)
Under these circumstances, the temporal variation of ρ and J is again small in the
time interval d/c within which light to traverses the source. This means that in
the interior of the source, the retardation is a small effect. We can then expand
  
′ r n · x′ r  n · x′  ′ r
J x, t− + = J x′ , t − + J̇ x , t − + O(d2 /λ2 ). (4.14)
c c c c c
The first term gives the following contribution to A [see eq. (4.7)]: (Given that
the
R extension of Rthe source is finite,
R surface terms vanish and
xℓ (∇ · J) = − (∇xℓ ) · J = − Jℓ .)
Z  Z
1 ′ r 3 ′ 1 ′
h
′

′ r i 3 ′
J x, t− dx = − x ∇ ·J x, t− dx
rc c rc c
Z 
1 r 3 ′
= x′ ρ̇ x′ , t − dx
rc c
1  r
= Ṗ t − ,
rc c
100 Section 4.1

R
where P(t) = x ρ(x, t)d3 x is the dipole moment of the charge distribution ρ.
The second term of (4.14) gives to the component Aj the following contribution:
Z Z Z 
1 ′ ˙ 3 ′ 1 ′ ˙ ′ ˙ 3 ′ ′ ˙ ′ ˙ 3 ′
nℓ xℓ Jj d x = nℓ (xℓ Jj + xj Jℓ )d x + (xℓ Jj − xj Jℓ )d x
rc2 2rc2
1 ¨ 1
= 2
Iℓj nℓ + µ̇ℓj nℓ (4.15)
2rc rc
with
Z h 
1 r  r i 3 ′
µ̇ℓj = x′ℓ J˙j x′ , t − − x′j J˙ℓ x′ , t − dx
2c c c
Z  Z
r ′ ′ 3 ′
I˙ℓj = ′
ρ̇ x , t − x x d x = − (∇′ · J) x′ℓ x′j d3 x′
c ℓ j
Z Z
′ ′ 3 ′
= Jm ∂m (xℓ xj )d x = (Jℓ x′j + Jj x′ℓ )d3 x′ .

We have then found (neglecting higher order terms)

1  r 1  r 1 ¨  r
Aj (x, t) = Ṗj t − + µ̇ℓj t − nℓ + Iℓj t − nℓ . (4.16)
rc c rc c 2rc2 c
If we define the magnetic dipole as
Z
1
µ= (x ∧ J)d3 x,
2c
the dipole contributions [the first two terms of(4.16)] give
dipole 1  r 1  r
A = Ṗ t − − n ∧ µ̇ t − . (4.17)
rc c rc c
In the quadrupole contribution, I¨ℓj , we can remove the trace because it ends up
being a term parallel to n in A, and so it does not contribute to the fields E and
B (see 4.10). We make
1
Qℓj := Iℓj − δℓj Imm .
3
The quadrupole term then gives
1  r 1  
Aquadrupole
m (x, t) = Q̈mℓ t − nℓ ≡ Q̈n . (4.18)
2rc2 c 2rc2 m

This contribution is in general lower with respect to the electric dipole by a factor
≈ dω/c ≈ d/λ. Denoting with Qn the vector whose components are Qmℓ nℓ , we
find for the fields [see eqs. (4.10)]:
 
1 1 1 ...
B = − c n ∧ Ȧ = P̈ ∧ n + (µ̈ ∧ n) ∧ n + Qn ∧ n
rc2 2c
  (4.19)
1 1  ... 
E =B∧n = (P̈ ∧ n) ∧ n + (n ∧ µ̈) + Qn ∧ n ∧ n .
rc2 2c
Ruth Durrer Electrodynamique Chap. 4 101

The moments P, µ and Q must be evaluated at the retarded time t − r/c. The
magnetic dipole radiation is obtained from the electric dipole radiation by replacing
E with −B, B with E and P with µ.

Emitted power

As before, the Poynting vector is (E = B ∧ n , B = n ∧ E)

c c c
S= E∧B = |E|2n = |B|2 n;
4π 4π 4π
and so the emitted power (= energy per unit time) per unit solid angle in a given
direction n is
dP 1
= (S · n)r 2 = |[. . .]|2 . (4.20)
dΩ 4πc3
where [. . .] corresponds to the terms in brackets in (4.19). If the electric dipole
dominates, we have
 
dP 1 2 1
= | P̈ ∧ n| = |P̈|2 sin2 ϑ, (4.21)
dΩ el. dip. 4πc3 4πc3

where ϑ is the angle between P̈ and n. After integrating over the angles dΩ =
sin ϑdϑdϕ, we obtain the total power (energy emitted per unit time)
2
Pel. dip. = |P̈|2 . (4.22)
3c3
Similarly, one finds for the magnetic dipole
2
Pmagn. dip. = |µ̈|2 . (4.23)
3c3
A simple calculation leads to the following result for the electric quadrupole radi-
ation (see exercise):
 
dP 1 h ... ... ... ... i
= Qkℓ Qkm nℓ nm − Qkℓ Qsm ns nℓ nk nm ;
dΩ el. quad. 16πc5

and, integrating over the angles,

1 ... ... 1 ... 2
Pel. quad. = 5
Qkℓ Qkℓ ≡ |Q| . (4.24)
20c 20c5
If one integrates (4.20) over the angles, all the mixed terms disappear and one
obtains the sum of (4.22), (4.23) and (4.24):
2 2 2 2 1 ... 2
P = | P̈| + | µ̈| + |Q| . (4.25)
3c3 3c3 20c5
102 Section 4.2

4.2 Dipole fields

(See Jackson 9.2,9.3)

Let us consider a small source of diameter d; we assume that

r ≫ d et λ ≫ d (4.26)
without making any hypothesis on the relation between λ and r as in (4.5) and
(4.13). The second among conditions (4.26) coincides with (4.13) and therefore we
can expand the retardation in the source as
    
′ |x − x′ | ′ r n · x′ r  n · x′  ′ r
ρ x , t− ≈ ρ x , t− + ≈ ρ x′ , t− + ρ̇ x , t− .
c c c c c c
To first order in d/r and d/λ ≈ (d/c)∂t , we obtain for the integrand of (4.1):
ρ (x′ , t − |x − x′ |/c) 1  ′ r  n · x′  ′ r  n · x′  ′ r
≈ ρ x , t − + ρ x , t − + ρ̇ x , t − .
|x − x′ | r c r2 c rc c
(4.27)
The second term of (4.27) is new. It does not contribute to the emission of radia-
tion. At large distances this and the third term are smaller than the first one by
a factor of d/r and of d/λ, respectively.

If Q is the total charge and P is, as before, the dipole moment, we find the following
for the scalar potential [with (4.1) and (4.27)]
Q 1  r 1  r
φ(x, t) = + 2 n · P t − + n · Ṗ t − . (4.28)
r r c rc c

x
λ λ

Curves r 2 hE2 (x)i = cst represented in the (x, y) plane for an electric dipole oriented in
the y direction.
Ruth Durrer Electrodynamique Chap. 4 103

The vector potential is obtained by replacing in (4.27) ρ with −J/c and by using
(4.15)
1  r 1  r 1  r
A(x, t) = Ṗ t − − 2n ∧ µ t − − n ∧ µ̇ t − . (4.29)
rc c r c rc c
(The same tricks that led to (4.15) have been applied to obtain (4.29)).

We now compute the fields E and B in the case where Q = 0 and µ is negligible.
Such a configuration is called a Hertzian dipole. Defining the Hertz vector,
1  r
Z= P t− (4.30)
r c
we find, for a Hertzian dipole,
1
φ(x, t) = −∇ · Z and A(x, t) = Ż. (4.31)
c
One can now determine the fields B = ∇ ∧ A and E = −∇φ − 1c Ȧ. After a brief
calculation one finds:
 
1 1
B = −n ∧ P̈ + 2 Ṗ
rc2 r c
1 h i 1 h i 1
E = ( P̈ · n)n − P̈ + 3( Ṗ · n)n − Ṗ + 3 [3(P · n)n − P] (. 4.32)
rc2 r2 c r
For all P, Ṗ and P̈ must be evaluated at the retarded time t − r/c. The last
term is the field of a static electric dipole that you know from electrostatics. If we
consider a harmonic time dependence (∼ eiωt ), we have
1 ω 1
Ṗ ≈ |P| ≈ |P|.
c c λ
The three terms of the electric field (4.32) are then the quotients
1 1 1
2
: 2 : 3.
rλ r λ r
In the near zone, r ≪ λ, the last term dominates and |B| ≪ |E|: this is the field of
an static dipole. In the induction zone, r ≈ λ, B is of the same order of magnitude
as E. In this zone B is generated by the displacement current Ė. In the wave zone,
r ≫ λ, the terms of order 1/r of the previous paragraph dominate (see figure).

4.3 The field of a moving point charge

(See Jackson Chap. 14)

104 Section 4.3

Here we determine the field of a moving point charge. The result is important
for particle accelerators and astrophysics (synchrotron radiation). We consider a
point charge e that travels along the path z(t). The charge and current densities
are given by 
(3)
ρ(x, t) = eδ (x − z(t)) 
. (4.33)
J(x, , t) = e ż(t)δ (3) (x − z(t)) 

Since its sources are distributions, the potentials (4.1) and (4.2) are also distribu-
tions and one can prove mathematically and rigorously that the convolutions (4.1)
and (4.2) exist. The following is a formal calculation.
Z
δ (t − |x − x′ |/c − t′ ) (3) ′
φ(x, t) = e ′
δ (x − z(t′ )) d3 x′ dt′
|x − x |
Z
δ (t − |x − z(t′ )|/c − t′ ) ′
= e dt , (4.34)
|x − z(t′ )|

and similarly
Z
e ż(t′ )
A(x, t) = δ (t − |x − z(t′ )|/c − t′ ) dt′ . (4.35)
c |x − z(t′ )|

The function f (t′ ) = t′ + |x − z(t′ )|/c − t has just one zero, tret (x, t), a fact that
follows from the figure below:

t’
(x, t)
c(t − tret ) = |x − z(tret )| So
(z(t’), t’) z(t) intersects the light cone
of (x, t) only once. Oth-
tret erwise the particle would
(z(tret ), t ret ) travel at a speed larger than
c!
x
x
Using the relation (see math. methods II)
Z
g(y0 )
g(y)δ(f (y))dy =
|f ′(y0 )|

for a function with a unique zero y0 . When |ż| < c, the derivative

żret · (x − zret )
f ′ (t′ )|t′ =tret = 1 − > 0,
c|x − zret |
Ruth Durrer Electrodynamique Chap. 4 105

and the function f is then monotone, which proves mathematically that it has only
one zero. For the potentials we thus obtain
e 1
φ(x, t) = żret ·(x−zret )
;
|x − zret | 1 −
c|x−zret |

and so
e
φ(x, t) = 1 (4.36)
|x − zret | − ż
c ret
· (x − zret )
eżret
A(x, t) = . (4.37)
c|x − zret | − żret · (x − zret )

Here zret designates z(tret ) where tret is implicitly determined by f (tret ) = 0, i.e.
by the equation
1
tret = t − |x − z(tret )|.
c
(4.36) and (4.37) are the Liénard and Wiechert potentials.

One could now compute the fields E and B from (4.36) and (4.37), but it is easier
to use the original expressions, (4.34) and (4.35). We make

ż x − z(t′ )
β= , R = |x − z(t′ )| and n = .
c R
For f (t′ ) = t′ − t + R(x, t′ )/c, we use the fact that |β| < 1, |n| = 1, and so
df
κ := = 1 − n · β > 0.
dt′
With these definitions we have
  Z  
φ(x, t) 1 1
=e ′ δ (f (t′ )) dt′ . (4.38)
A(x, t) β(t ) R(t′ )

The right-hand side of (4.38) depends on x only through R and on t only through
∂
f . Using ∇ = n ∂R , we then find
Z     
1 ′ R (β − n) ′ ′ R
E(x, t) = e nδ t + − t + δ t + − t dt′
R2 c Rc c
Z     
1 ′ R 1 ′ ′ R
B(x, t) = e (n ∧ β) − 2 δ t + − t + δ t + − t dt′ .
R c Rc c

Using the relations

1 1 1
δ (f (t′ )) = δ(t′ − tret ) = δ(t′ − tret ) = δ(t′ − tret )
|f ′ (tret )| 1−n·β κ
106 Section 4.3

Z Z
′ ′ ′ ′ dτ dτ dτ
δ (f (t ))g(t )dt = δ ′ (τ )g((t′ (τ )) , dt′ = = ′
|{z} f ′ (t′ (τ )) dτ /dt′ f
τ
Z    
g (t′ (τ ))
d d g (t′ (τ ))
= − δ(τ ) dτ = −
f ′ (t′ (τ ))
dτ dτ f ′ (t′ (τ )) τ =0
     
1 d g(t′ ) 1 d g(t′)
= − ′ ′ ′ =− ,
f (t ) dt f ′ (t′ ) t′ =tret κ(t′ ) dt′ κ(t′ ) t′ =tret

we find
  
1 1 d n−β
E = e n+
κR2 κc dt′ κR
 
ret
 (4.39)
1 1 d β∧n
B = e β∧n+ .
κR2 κc dt′ κR ret

To evaluate the derivatives with respect to t′ we use

dR
= −ż · n = −cβ · n,
dt′
and
1 d β n dR 1 1
′
n=− − ′
= [−β + n(n · β)] = n ∧ (n ∧ β).
c dt R cR dt R R
This conduces to
 
   
 n n d 1 1 1 d β 
E(x, t) = e  2 + + 2 2 n ∧ (n ∧ β) − .
κR ′
cκ dt κR κR | {z } cκ dt κR
′

−β +n(n·β )

Using
n κn n − n(n · β)
2
= 2 2 = ,
κR κ R κ2 R2
the sum of the first and third terms can be written more simply as:
    
n−β n d 1 1 d β
E(x, t) = e + − . (4.40)
κ2 R2 cκ dt′ κR cκ dt′ κR ret

Along similar lines, one can find B:

   
β 1 d β
B(x, t) = e + ∧n ; (4.41)
κ2 R2 cκ dt′ κR ret

so
B = n ∧ E. (4.42)
Ruth Durrer Electrodynamique Chap. 4 107

Writing β̇ = dβ/dt′ , we have

1 d κ dR 1 d
(κR) = − R ′ (n · β)
c dt′ c dt′
| {z } | dt{z
c }
−κβ ·n ˙ 2
Rn·β /c+[−β +(n·β )2 ]

R
= −n · β + (n · β)2 − (n · β̇) + β 2 − (n · β)2
c
R
= β 2 − (n · β) − (n · β̇).
c
The substitution of this result in (4.40) gives after a brief calculation:
    
(1 − β 2 ) e 1
E(x, t) = e (n − β) 3 2 + n ∧ (n − β) ∧ β̇ . (4.43)
κ R ret c κ3 R ret

The fields E and B are now explicitly determined by (4.43) and (4.42).

The first term of (4.43) is a deformed Coulomb field, a purely kinematic effect.
This field is obtained through the Lorentz transformation of a Coulomb field and
it decreases as 1/R2 . The second term, which contains a factor β̇, is new. It
decreases as 1/R and its contribution to the Poynting vector gives the emission
(which decreases as 1/R2 and so (S · n)R2 remains finite for R → ∞). The
contribution to S · n that dominates at a very large distance from the particle is
then
   2
2 c 2 2 e2 1
(S · n)R = |E| R = n ∧ (n − β) ∧ β̇ . (4.44)
4π 4πc κ6 ret

The vector inside | . . . | indicates the direction of the field E (at large distance).

We consider first the small velocity limit, |β| ≪ 1. We then have n − β ≈ n and
κ ≈ 1. For the emitted power in a solid angle dΩ, we obtain
dP e2
= (S · n)R2 = |n ∧ (n ∧ β̇)|2
dΩ 4πc
e2 2 2
= v̇ sin ϑ (|β| ≪ 1). (4.45)
4πc3
The total power is then the integral of (4.45) over the angles dΩ = sin ϑdϑdϕ,
2e2 2
P = |v̇| . (4.46)
3c3
This is the Larmor formula.

For the discussion of the general case (v . c), one should observe that (4.44) gives
the energy flow per unit time in the direction n. This radiation was emitted at
108 Section 4.3

time tret = t − R(tret )/c. If we consider the radiation emitted during the time
interval T1 6 tret 6 T2 , we obtain for the energy received by the observer
Z T2 +R(T2 )/c Z T2
2 dt
W = (S · n)R dt = (S · n)R2 dtret .
T1 +R(T1 )/c T1 dtret
The angular power emitted is then
dP dt
(tret ) = (S · n)R2 ;
dΩ dtret
with t = tret + R(tret )/c, we obtain
dt
=1−β·n=κ
dtret
and using (4.44) we arrive at the expression
dP e2 1  2
= n ∧ (n − β) ∧ β̇ . (4.47)
dΩ 4πc (1 − n · β)5
It is possible (but a bit annoying) to integrate this relation over the direction, an
operation that conduces to the result
2e2 6 h 2 2
i
P = γ |β̇| − (β ∧ β̇)
3c
1
γ = p . (4.48)
1 − β2
This is the relativistic Larmor formula. It is this effect that produces the enormous
losses present in accelerators at CERN, for example.

The simplest example that one could consider is the rectilinear motion in which β̇
is parallel to β. In this scenario, cos ϑ = β · n is constant and β ∧ β̇ = 0. Equation
(4.47) then gives
dP e2 2 sin2 ϑ
= | v̇| . (4.49)
dΩ 4πc3 (1 − |β| cos ϑ)5
Because of the power 5 at the denominator, the radiation is highly focused in the
forward direction for |β| close to 1.

β→1

β=0
ϑ max
v
Ruth Durrer Electrodynamique Chap. 4 109

The angle for which the radiation maximizes is given by (β ≡ |β|)

 p 
1 2
1
cos ϑmax = 1 + 15β − 1 ϑmax ≈ for β → 1.
3β 2γ
For small angles, ϑ ≪ 1, one obtains for β → 1:
dP 8 e2 2 8 (γϑ)2
≈ |v̇| γ . (4.50)
dΩ π c3 [1 + (γϑ)2 ]5
For the power, the integral of (4.49) gives
2e2 2 6
P = |v̇| γ (rectilinear motion). (4.51)
3c3
z
As a second example, we consider a particle
β n
in circular motion, i.e. with β̇ ⊥ β. Let β
be parallel to the axis z and β̇ parallel to x ϑ
as shown on the right.
Equation (4.47) gives (after a trigonometric β
exercise) ϕ y

x
 
dP e2 1 sin2 ϑ cos2 ϕ
= 3
|v̇|2 1− . (4.52)
dΩ 4πc (1 − β cos ϑ)3 γ 2 (1 − β cos ϑ)2
For β → 1, we obtain once again radiation focused around the forward direction
(see figure). For small angles, ϑ ≪ 1, and high speeds, β ≈ 1, we obtain
 
dP 2e2 2 γ6 4γ 2 ϑ2 cos2 ϕ
≈ 3 |v̇| 1− . (4.53)
dΩ πc (1 + γ 2 ϑ2 )3 (1 + γ 2 ϑ2 )2
For the total power eq. (4.48) gives
2e2 2 4
P = |v̇| γ (circular motion). (4.54)
3c3
For circular motion of radius ρ and of angular frequency ω, one has |v̇| = v 2 /ρ = vω
and so
2e2 c 4 4 2e2 2 2 4
P = β γ = ω β γ . (4.55)
3 ρ2 3c
The factor γ 4 leads the huge synchrotron losses.

Numerical example: For the LEP at CERN, ρ ≈ 5 km. The final energy of an
electron was about
Eél. 5
Eél. ≈ 100 GeV ⇒ γLEP = 2 ≈ 2 × 10 .
mc
|{z}
0.5 MeV
110 Section 4.4

Therefore v ≈ c ≈ 3×1010 cm/s and β ≈ 1. The frequency is ω = v/ρ ≈ 6×104 s−1 .

The electron charge can be expressed as e2 = me re c2 ≈ 0.5 MeV × 2.8 × 10−13 cm,
which gives
2e2
≈ 3 × 10−24 MeV s.
3c
For the total power (4.55) then gives
P = 3 × 10−24 MeV s × (6 × 104 s−1 )2 × (2 × 105 )4 ≈ 2 × 107 MeV/s
≈ 3 × 10−6 J/s per electron!
(1 J = 6.242 × 1012 MeV.) Since γ = E/mc2 , for a fixed energy, the losses are
less important if the accelerated particle is heavier. This is why the particles that
are accelerated in the new machine, the LHC (that reaches energies of 7TeV), are
protons (mp c2 ≃ 938MeV) and not electrons. With this, γLHC ≃ 7.5 × 103 is
around 27 times smaller than γLEP and the losses are reduced by a factor 5 × 105
per particle.

4.4 Cherenkov radiation

(Jackson §13.5)

A free particle in vacuum (non accelerated) does not radiate (β̇ = 0). However,
a particle of constant velocity inside a medium does radiate if its speed is higher
than the speed of light in that medium. This form of radiation was discovered by
Cherenkov in 1934 and was explained theoretically by Frank and Tamm in 1937.
(In 1958, these three physicists obtained the Nobel prize for these works,) The
Cherenkov radiation is of practical importance, particularly for the identification
and counting of elementary particles of high energy (Cherenkov detector).

Here we restrict ourselves to the elementary aspects of the theory of Frank and
Tamm (for an extensive discussion, see Jackson). We consider an homogeneous and
isotropic medium with µ = 1. We neglect the dispersion, ε(ω) = ε = constant.
Consider a small source in rectilinear motion at constant speed v in the direction
x. The charge and current densities are:
ρ(x, t) = ρ0 (x − vt)
J(x, t) = ρ0 (x − vt)v, v = (v, 0, 0).
Maxwell equations become
1
∇ · B = 0, ∇ ∧ E + ∂t B = 0
c (4.56)
4π 4π ε
∇·E = ρ, ∇∧B= J + Ė.
ε c c
Ruth Durrer Electrodynamique Chap. 4 111

With B = ∇ ∧ A and E = −∇φ − 1c Ȧ, we obtain from (4.56)

1 4π ε ε 4π
∆φ + ∇ · Ȧ = − ρ et ∆A − 2 Ä − ∇(∇ · A + φ̇) = − J.
c ε c c c

We use the “Lorentz gauge”, ∇ · A + εc φ̇ = 0, which gives

 ε 2 4π  ε 2 4π
∆− 2
∂t φ = − ρ; ∆ − 2
∂t A = − J. (4.57)
c ε c c
√
The phase velocity is then c/n = c/ ε where n is the index of refraction. Since the
sources are function of the variables x − vt, y and z, we propose the same Ansatz
for φ and A:
φ, A ∼ f (x − vt, y, z).
The derivatives with respect to time are ∂t f = −v∂x f and ∂t2 f = v 2 ∂x2 f . With
Jy = Jz = 0, we obtain  
n2 v 2 2
∆ − 2 ∂x Ay,z = 0.
c
|x|→∞
Looking for solutions that decrease at infinity, |A(x)| → 0, we make Ay = Az =
0. For Ax and φ we have
  
n2 v 2 2 2 2 4π
1− 2 ∂x + ∂y + ∂z Ax = − ρv (4.58)
c c
  
n2 v 2 2 2 2 4π
1− 2 ∂x + ∂y + ∂z φ = − 2 ρ. (4.59)
c n
2
The Lorentz condition, ∂x Ax − nc v∂x φ = 0 and (4.58) are satisfied with (4.59) and

n2
Ax = vφ. (4.60)
c
Equation (4.59) remains. With β := nv/c = v/vphase , we write it as
  4π
(1 − β 2 )∂x2 + ∂y2 + ∂z2 φ = − 2 ρ. (4.61)
n
For β < 1, this is an elliptic equation. Its solution is a Coulomb potential con-
tracted in the direction x. At β = 1, the nature of the equation changes and it
becomes hyperbolic for β > 1. This last case is the one we want to examine.
p
When making the variable transformation x → τ = (−x+vt)/ β 2 − 1;
Let β > 1. p
∂x = −1/( β 2 − 1)∂τ , the equation (4.61) becomes

4π  p 2 
(∂y2 + ∂z2 − ∂τ2 )φ = − 2 ρ0 −τ β − 1, y, z . (4.62)
n
112 Section 4.4

This is the two-dimensional wave equation. In the mathematical complements, we

have determined its retarded Green function which satisfies (∂y2 + ∂z2 − ∂τ2 )G = δ
and G = 0 for τ < 0:
(
√ 1 for τ 2 > y 2 + z 2 , τ > 0
2π τ 2 −y 2 −z 2
G(y, z, τ ) =
0 otherwise.

For ρ0 (x) = eδ(x), we have

4π  p 2  4πe  p 2 
− ρ0 −τ β − 1, y, z = − δ τ β − 1 δ(y)δ(z)
n2 n2
4πe
= − p δ(τ )δ(y)δ(z).
n2 β 2 − 1

If we convolve with G, this gives

4πe
φ=− p G,
n2 β2 − 1

that is to say

 2e 1

 p

 n2
(x − vt)2 − (β 2 − 1)(y 2 + z 2 )
φ(x, y, z, t) = if (x − vt)2 > (β 2 − 1)(y 2 + z 2 ) (4.63)





0 otherwise,
Ax = nβφ, Ay = 0, Az = 0. (4.64)

We then obtain a discontinuity surface where

p φ diverges. It forms a cone, the Mach
cone, with an aperture angle tan α = 1/ β 2 − 1

c/n α v
v x

1 c/n phase velocity of the waves

sin α = = = .
β v speed of the particle
As for supersonic motion, α is called the Mach angle. In the exterior of a Mach cone
the fields vanish and on the cone they diverge. This is not a physical singularity.
It disappears if one takes into account the dispersion. In vacuum, this last remark
does not apply. (A further indication that v cannot be greater than c!)
Chapter 5

Scattering of electromagnetic
waves

If an electromagnetic wave enters a medium, the charges in the medium are accel-
erated by the field of the wave. This acceleration creates new waves, the scattered
waves. This process is at the origin of the redness of the sunset and the blueness
of the sky, for example.

5.1 Thomson scattering

(Jackson §14.7)

If a plane wave reaches a free point charge, the latter is accelerated. The particle
then absorbs part of the energy of the incident wave and emits a new wave. We
want to compute the cross section of this process:

dσ (emitted energy)/(time×angle)
(n) = (5.1)
dΩ (incident energy)/(time×area)

is the differential cross section in the direction n.

Z
dσ
σ= dΩ
dΩ

is the total cross section. Its dimension is cm2 . We now consider a plane wave
scattered by a non-relativistic particle of mass m and charge e. The electric field
of the incident plane wave (in complex representation) is given by

E(x, t) = E0 ǫei(k·x−ωt)

113
114 Section 5.1

where |E0 | is the amplitude, ǫ is the polarisation vector and ω is the angular
frequency. Let z(t) be the position of the particle at time t. We have

mz̈(t) = eE(z, t) = eE0 ǫei(k·z(t)−ωt) .

(We have assumed that v ≪ c and so we can neglect the Lorentz force, ec v ∧ B.)
In order to compute the energy emitted per unit time using the Larmor formula
e2
(4.45), dP
dΩ
= 4πc 2 2 2
3 v̇ sin ϑ, we need the time average hv̇ i. Under the hypothesis

that the particle does not move too much during a period as compared to the wave
length (which is equivalent to the condition v ≪ c), we obtain

1  e 2
hv̇2 i = h|Re(z̈)|2 i = |E0 |2 .
2 m
Using the Larmor formula this gives
 2
dP c e2
(n) = |E0 |2 sin2 ϑ
dΩ 8π mc2

where ϑ is the angle between the direction of observation n and the direction of
acceleration z̈ = v̇. The flux of incident energy (energy per unit time and unit area)
c
for a plane wave is the average amplitude of the Poynting vector, |S| = 8π |E0 |2 .
This gives
 2 2
dσ e
(n) = 2
sin2 ϑ, (5.2)
dΩ mc
which is the Thomson cross section. For a wave coming from direction z (k̂ = ez )
and with a polarisation ǫ in the (x, y) plane that forms an angle ψ with the x axis,
we have (see figure) n = (sin θ cos ϕ, sin θ sin ϕ, cos θ), ǫ = (cos ψ, sin ψ, 0), and so
z
cos ϑ = n · ǫ
k n
= sin θ(cos ϕ cos ψ + sin ϕ sin ψ) θ
ϑ
= sin θ cos(ϕ − ψ) ϕ
cos2 ϑ = sin2 θ cos2 (ϕ − ψ). ψ y

x ∈

For unpolarised
R 2π incident radiation, we can average (integrate) over ψ by using the
identity 2π 0 cos (ϕ − ψ)dψ = 12 . We obtain
1 2

 2    2
dσ e2 1 e2 1
= 1 − sin2 θ = (1 + cos2 θ) (5.3)
dΩ mc2 2 mc2 2

where θ is now the angle between n and k, i.e. between the direction of the incident
wave and the direction of the scattered wave. The total Thomson scattering cross
Ruth Durrer Electrodynamique Chap. 5 115

section is
Z  2 2
dσ 8π e
σT = dΩ = = 0.665 × 10−24 cm2 = 0.665 barn for an electron.
dΩ 3 mc2
(5.4)
e2 −13
The magnitude re = mc2 = 2.82 ×10 cm is called the “classical electron radius”.
In quantum electrodynamics one can show that the equation (5.3) is valid only at
low frequencies, ~ω ≪ mc2 .

5.2 Elastic and inelastic scattering by quasi-free

charges

We study here in a more general way an incident wave on a system of quasi-free

charges {ej } located at positions {xj }. As a concrete example, one can think of
the scattering of X rays by the electrons in a solid. The energy of a photon in the
X ray is much higher than the binding energy of the electrons in the solid, which
is why the latter can be considered as free. As in the previous section, the incident
wave is given by
E(x, t) = E0 ǫei(k·x−ωt) .
According to (4.43), the radiation of the sources {ej } is (vj ≪ c)

1 X ej h i
Es (x, t) = n ∧ (n ∧ β̇ j ) . (5.5)
c j Rj ret

The acceleration is given by

1 ej
β̇ j = v̇j = E0 ǫei(k·xj −ωt) . (5.6)
c mj c

This leads to
X e2j 1
Es (x, t) = E0 [n ∧ (n ∧ ǫ)] exp[ik · xj − iωtret ] (5.7)
j
mj c2 Rj

Rj
with tret = t − c
. With enough precision (|xj | ≪ |x|), we have
x
Rj = |x − xj | ≈ |x| − n · xj
x O Rj
n = , r = |x|.
r
xj
116 Section 5.2

We then obtain

e−iω(t− c ) X e2j −iq·xj

r

Es (x, t) = E0 [n ∧ (n ∧ ǫ)] 2
e (5.8)
r j
mj c

where q = (ω/c)n − k. As in the previous paragraph, we find for the cross section
2
dσ r 2 |Es |2 X e2j
= = e−iq·xj sin2 ϑ, (5.9)
dΩ |E0 |2 j
mj c 2

where ϑ is the angle between n and ǫ. This equation is valid only if the electrons
are quasi free, i.e. if the frequency of the incident wave is much higher than the
frequencies of the atomic transitions. To use (5.9), we yet have to take the average
of the positions xj .

Coherent and incoherent scattering

We discuss here the equation (5.9). The considerations to follow are also relevant
for other situations beyond ours, and they can be applied to any scattering process,
in particular to the scattering of a beam of particles, because matter particles also
possess a wave-like nature. The cross section (5.9) depends markedly on |q|. Let
a = h|xj |i be the dimension of the system of particles. The cross section will be very
different for qa ≪ 1 and qa ≫ 1. Defining the scattering angle θ, cos θ = (k · n)/k,
and use ω/c = k we obtain
ω 2 θ
q2 = n − k = 2k 2 (1 − cos θ) = 4k 2 sin2
c 2
θ
q = 2k sin .
2
If ka ≪ 1, qa ≪ 1 for every angle θ. In the opposite case in which ka ≫ 1, qa is
small (qa ≪ 1) only if θ ≪ θc = 1/ka and qa ≫ 1 for large angles.

For qa ≪ 1, all the exponentials e−iq·xj are very close to 1 and (5.9) gives
2
dσ X e2j
≈ sin2 ϑ. (5.10)
dΩ qa≪1 j
mj c2

For an atom with Z electrons, this leads to

 2
dσ 2 e2
≈Z sin2 ϑ. (5.11)
dΩ qa≪1 mc2
Ruth Durrer Electrodynamique Chap. 5 117

The Z electrons behave in a coherent manner, as a particle of radius R = Zre =

Ze2 /(mc2 ). The cross section is Z 2 times the cross section of a single electron!

In the opposite limit, qa ≫ 1, the exponentials in (5.9) oscillate rapidly and

have very different phases. The mixed terms vanish and only the diagonal terms
contribute to the summation. Thus,

dσ X  e2j 2
≈ 2
sin2 ϑ. (5.12)
dΩ qa≫1 j
mj c

For Z identical particles of mass m and change e, instead of (5.11), this leads to
 2 2
dσ e
≈Z sin2 ϑ. (5.13)
dΩ qa≫1 mc2

In this case, the contributions of the particles superpose in an incoherent way. In

quantum mechanics, one calculates (Thomas-Fermi model) a ≈ 1.4a0 Z −1/3 where
a0 = ~2 /(me2 ) = re /α2 ≃ 5.3 × 10−9 cm is the Bohr radius of the hydrogen atom
and α = e2 /(~c) ≈ 1/137 is the fine structure constant.

We now consider in more detail the situation of identical particles, ej ≡ e and

mj ≡ m ∀j. The charges considered can be the electrons of an atom of total
charge Z, for example. In this case, (5.9) reduces to
 2 Z
X
2
dσ e2 2 −iq·xj
= sin ϑ e . (5.14)
dΩ mc2 j=1

We take the statistical average of the last factor. Let W (x1 , . . . , xZ ) be the dis-
tribution of the probability of presence of the electrons. We define the form factor
Z X
Z 2

F 2 (q) = e−iq·xj W (x1 , . . . , xZ )d3Z x. (5.15)

j=1

We have therefore  
dσ dσ
= F 2 (q). (5.16)
dΩ dΩ Thomson
R
From (5.15), one has (W is a probability, so W (x1 , . . . , xZ )d3Z x = 1)

F 2 (0) = Z 2 . (5.17)

We split (5.15) into the diagonal and non-diagonal terms:

Z 2
X X
−iq·xj
e =Z+ eiq·(xj −xm ) .
j=1 j6=m
118 Section 5.2

Therefore
XZ
2
F (q) = Z + eiq·(xj −xm ) W (x1 , . . . , xZ )d3Z x.
j6=m

We call Z Y
Wmj (xm , xj ) := W (x1 , . . . , xZ ) d3 xk (5.18)
k6=j,m

the distribution of probability of presence of the electrons m and j. We define

additionally P (q) through
XZ
Z(Z − 1)P (q) = eiq·(xj −xm ) Wmj (xm , xj )d3 xm d3 xj . (5.19)
j6=m

One then has

F 2 (q) = Z + Z(Z − 1)P (q).
Evidently P (0) = 1, which agrees with (5.17).
The integral in (5.19) is proportional to the Fourier transform of Wmj (xm , xj )
at the position (q, −q) ∈ R6 . The Riemann-Lebesgue Lemma tells us that the
cmj ∈ L∞ ,
Fourier transform of a continuous function decreases to 0 at infinity: W
cmj (q, −q) = 0. It follows that
so lim|q|→∞ W

lim P(q) = 0
|q|→∞

lim F 2 (q) = Z.
|q|→∞

For large values of the momentum, the cross section is then constant and its value
is Z times the cross section of a single particle. This is known as the “deeply
inelastic” regime.

The comparison between the elastic and the total cross sections allows to determine
the number of particles in the scattering target,
dσ
dΩ qa≪1
Z= dσ
. (5.20)
dΩ qa≫1

This is why the measurements of the cross section of nucleons were interpreted as
evidence of the composition of the nucleon by three “partons” (quarks).

This means that with scattering experiments one can study the structure of an
object. Using the fact that Z = F 2 (0)/F 2(∞) one can in principle determine
whether an object is composed or not. For this, one needs waves with a wave
number k > 1/a, if a is the dimension of the object (i.e., for the study of an object
of size a, “photons” of energy E = ~ω = ~kc > ~c/a are required.)
Ruth Durrer Electrodynamique Chap. 5 119

Elastic scattering

The elastic part of the scattered wave is the one with the same frequency ω as the
incident wave. The incident wave P depends on time as eiωt . The scattered wave
depends additionally on the factor j e−iq·xj , which depends on time through the
positions xj . This is why it contains some other frequencies apart from ω. We
obtain Pthe part of the field with frequency ω by taking the time average of the
factor j e−iq·xj . With (5.14) we obtain
    X
2
dσ dσ
= e−iq·xj . (5.21)
dΩ elast. dΩ Thomson j

We substitute the time average by the spacial average:

X Z Z
X Z
X
−iq·xj 3Z
e−iq·xj = e W (x1 , . . . , xZ )d x = Fj⋆ (q).
j j=1 j=1

Here Fj (q) is the Fourier transform of the probability density of a single electron,
Z Y
Wj (xj ) = W (x1 , . . . , xZ ) d3 xk ,
k6=j

and its Fourier transform or ’form’ factor is defined by

Z
Fj (q) = Wj (x)eiq·x d3 x . (5.22)

With (5.21) this gives

   
dσ dσ 2
= Felast. (q) , (5.23)
dΩ elast. dΩ Thomson

Z 2
X
2
Felast. = Fj (q) . (5.24)
j=1

The “elastic form factor” can be interpreted as follows: the density of charge of
the Z electrons is
XZ
ρ(x) = e δ(x − xj )
j=1

and its statistical average is

XZ X
hρi(x) = e δ(x − xj )W (x1 , . . . , xZ )d3Z x = e Wj (x) , (5.25)
j j
120 Section 5.3

From
P (5.22), the Fourier transform of the charge distribution is then hρ̂i(q) =
e j Fj (q), and so
e2 Felast.
2
(q) = |hρ̂i(q)|2 . (5.26)
The elastic form factor is the square of the absolute value of the Fourier transform
of the average charge density (ignoring the extra e2 factor). In the forward scatter
direction (θ = 0, hence q = 0), one has from (5.22) Felast. (0) = Z: the forward
scattering is coherent
     
dσ 2 dσ dσ
(θ = 0) = Z = (θ = 0). (5.27)
dΩ élast. dΩ Thomson dΩ total

From the Riemann-Lebesgue Lemma,

lim Felast. (q) = 0.

|q|→∞

Qualitatively this conduces to the situation sketched in figure 5.1.

Z2

F 2 (q)
Z 2
Felast.(q)
0
0
q

Figure 5.1: The qualitative behaviour of the elastic and total form factors as a
function of q = |kn − k| = 2k sin(θ/2).

5.3 Scattering in gases and liquids

(Jackson §9.7)

Here we analyse the following situation: we look upon an homogeneous and

isotropic dielectric medium with a dielectric “constant” ε0 (ω) that does not de-
pend on the position x. We consider a wave that propagates in this ponderable
medium and that finds in its way a precisely localised inhomogeneity of the dielec-
tric. A scattered wave is emitted from this inhomogeneity. We want to determine
the energy emitted and compute the cross section of this process. As important
applications let us remark:
Ruth Durrer Electrodynamique Chap. 5 121

i ) The scattering of light by particles of dust.

ii ) The scattering of light by the inhomogeneities of a gas (Lorentz theory).

Generalities

Let E(x, ω) and B(x, ω) be the Fourier transforms:

Z
E(x, ω) = E(x, t)eiωt dt
Z
B(x, ω) = B(x, t)eiωt dt.

The Maxwell equations in the absence of charges and currents and without mag-
netisation (µ = 1) are

iω 
∇ · (εE) = 0 ∇∧E− B = 0  
c . (5.28)
iω 

∇·B = 0 ∇ ∧ B + εE = 0 
c
In our situation we have

ε(x, ω) = ε0 (ω) + ε1 (x, ω) = 1 + 4π(χ0 + χ1 ) (5.29)

where ε1 is a localised contribution to the dielectric constant. We represent the

fields B and E by the potentials:

iω
B = ∇ ∧ A, E=− A − ∇ϕ
c
with the gauge conditions
iω
∇·A− ε0 ϕ = 0. (5.30)
c
Maxwell’s equations lead to
 
iω iω iω iω
∇ ∧ (∇ ∧ A) = − εE = − ε0 − A − ∇ϕ − ε1 E
c c c c |{z}
4πP1

Using our gauge condition this gives

 
ω2 4πiω
∆ + 2 ε0 (ω) A = P1 . (5.31)
c c
122 Section 5.3

The equation for ϕ is obtained by taking the divergence of (5.31). With (5.30) this
gives  
ω2 4π
∆ + 2 ε0 (ω) ϕ = ∇ · P1 . (5.32)
c ε0
If we make
ic
Z= A, (5.33)
ωε0
then we have ϕ = −∇ · Z and
4π
−(∆ + k 2 )Z = P1 , (5.34)
ε0
with
ω 2n20 (ω)
k2 = , n20 (ω) = ε0 (ω).
c2
The fields B and E are easy to find from Z:
iω
B=− ε0 ∇ ∧ Z, E = k 2 Z + ∇(∇ · Z). (5.35)
c
ε1
Observe that in the right-hand side of (5.34), P1 = 4π E is a function of Z. The
Green function of the Helmholtz operator, −(∆ + k 2 ),
1 eikr
G(x) = . (5.36)
4π r
That is to say, −(∆ + k 2 )G(x) = δ 3 (x) and so for
Z
ϕs (x) = G ∗ s(x) ≡ d3 x′ G(x − x′ )s(x′ ) (5.37)

one has
Z Z
2 3 ′ 2 ′ ′
−(∆+k )ϕs (x) = − d x (∆x +k )G(x−x )s(x ) = d3 x′ δ(x−x′ )s(x′ ) = s(x) .
(5.38)
2
The general solution of the equation −(∆ + k )ϕ(x) = s(x) is then ϕ = ϕ0 + ϕs
where ϕ0 is a homogeneous solution, (∆ + k 2 )ϕ0 = 0.

This allows us to transform (5.34) into the following integral equation:

Z ik|x−x′|
(0) 1 e
Z(x) = Z + P1 (x′ )d3 x′ . (5.39)
ε0 |x − x′ |
Here Z(0) is a homogeneous solution of (5.34) [i.e. −(∆+k 2 )Z(0) = 0], and describes
the incident wave. The second term describes the scattered wave. We denote it by
Z(d) . At large distances |x| = R, we approximate Z(d) by
Z
(d) eikR ′ ′ eikR
Z = P1 (x′ )e−ik ·x d3 x′ = P̃1 (k′ ) (5.40)
ε0 R ε0 R
Ruth Durrer Electrodynamique Chap. 5 123

with n = R/R and k′ = kn. Here P̃1 is the Fourier transform of P1 ,

Z
′ ′
P̃1 (k ) = P1 (x′ )e−ik ·x d3 x′ .
′
(5.41)



Using the relations ∇ eikR /R ≈ ikneikR /R + O(R−2 ), et ∇ ∧ eikR V/R ≈
ikn ∧ VeikR /R + O(R−2 ) for an arbitrary vector V, we obtain the scattered fields

(d) k 2 eikR ′ 

B = n ∧ P̃1 (k ) 
n0 R . (5.42)
k 2 eikR   1


E (d)
= ′
n ∧ P̃1 (k ) ∧ n = (d)
B ∧n 
n20 R n0
We now consider the fields E(x, ω) and B(x, ω) as the amplitudes of waves of a
given frequency ω. Adding the factor e−iωt , we find with tret := t − R/( nc0 )

2
k 
B(d) = n ∧ P̃1 (k′ , tret ) 

Rn0 . (5.43)
1 (d) 

E (d)
= B ∧n 
n0
We want to determine the intensity of the scattered wave polarised according to
ǫ′ , with ǫ′ · k′ = 0. This gives
1 k2 ′
ǫ′ · E(d) = ǫ · P̃1 . (5.44)
R n20
The Poynting vector of the part of the wave with polarisation ǫ′ is
 2 2
c ′ (d) 2 1 c k
S= n0 ǫ · E n = 2 n0 2
|ǫ′ · P̃1 |2 n.
4π R 4π n0
The intensity of the scattered wave is then (time average taken!)
dI c k4 D ′ 2
E n0 4 D ′ 2
E
(n) = |ǫ · P̃1 | = ω |ǫ · P̃1 | . (5.45)
dΩ 4π n30 4πc3

Scattering at long wavelengths

Let us assume that the wavelength λ is much larger than the size of the inho-
mogeneity. In this situation we can replace with 1 the exponential factor in the
definition of P̃1 . This is the dipole approximation,
dI n0 4 D ′ 2
E
= ω |ǫ · P 1 | (5.46)
dΩ 4πc3
Z
P1 = P1 (x′ ) d3 x′ .
124 Section 5.3

The intensity is then proportional to k 4 ∝ λ−4 .

Consequences: Blue light is much more scattered than red light. This is why the
sky appears blue and the setting sun, that has lost all its compo-
nents of wavelength below red, appears red.

Born approximation

It is an accurate enough approximation to substitute in (5.41)

ε1 (x′ ) ε1 (x′ )
P1 (x′ ) = E(x′ ) ≈ E0 (x′ ) (5.47)
4π 4π
′
where E0 (x′ ) is the incident field. Taking E0 (x′ ) = E0 ǫei(k·x −ωt) , a plane wave
linearly polarised in the direction ǫ (k · ǫ = 0), we obtain
Z
′ ε1 (x′ ) ′ ′ ′ E0
P̃1 (k ) = E0 ǫei(k·x −ωt) e−ik ·x d3 x′ = ǫ ε̃1 (k′ − k)e−iωt , (5.48)
4π 4π
and for the cross section of emission with polarisation ǫ′ :
   
dσ dI 1 1  ω 4 ′ 2
= c 2
= 2
|ǫ · ǫ|2 |ε̃1 (k′ − k)| .
dΩ Born dΩ 4π n0 |E0 | (4π) c
In the dipole approximation, k − k′ | ≪ 1/d, this gives
  Z
1  ω 4 ′
2
dσ 2 ′ 3 ′
= |ǫ · ǫ| ε1 (x ) d x . (5.49)
dΩ dip. 16π 2 c

After integrating over all the directions ǫ (which gives a factor 4π/3) and adding
over the two final polarisations (factor 2), we obtain the total cross section (within
the dipole approximation):
Z
1  ω 4
2
′ 3 ′
σ= ε1 (x )d x . (5.50)
6π c

We now apply this general result to a gas with small statistical fluctuations of
density. For a gas with NV atoms in a small volume V (V 1/3 ≪ λ), one has
NV
ε(ω) − 1 ≈ 4πα(ω)
V
where α(ω) is the molecular polarisability. For a fluctuation ∆NV of the number
of atoms, this gives (from 5.50) a cross section
1  4
2 ω
σ≈ (4πα) h(∆NV )2 i. (5.51)
6π c
Ruth Durrer Electrodynamique Chap. 5 125

This scattering due to the inhomogeneities in a gas is known as the Rayleigh

scattering. For a random distribution of particles (uncorrelated), the statistical
average, h(∆NV )2 i is proportional to V and σ/V = S, the scattering coefficient, is
independent of the volume. Einstein used this result to determine the Avogadro
number (see A. Einstein, Collected Papers Vol 3, p. 287, 1910).

5.4 Diffraction

(Jackson §9.8 ff)

Any deviation of a ray of light from its optical path is called diffraction. In the ge-
ometrical optic limit, an illuminated body creates a precise shadow. At sufficiently
large wavelengths, or for sufficiently small bodies, this approximation is no longer
valid. The diffraction phenomena are entirely explained by the wave-like nature of
light, but in this case the boundary conditions are not trivial. It is necessary in this
case to solve Maxwell’s equations for some given sources and boundary conditions
at the screens and at infinity.

Here we will discuss only the Kirchhoff approximation. Additionally, we make the
following (non-essential) simplifications:

i ) We make use of a scalar wave equation

 
1 2
∆ − 2 ∂t u = 0, (5.52)
c
that is assumed to be valid everywhere outside the source and the screens.
(Every component of the electric and magnetic fields satisfy this equation).
We also assume that the dispersive effects of the medium are negligible:
ε = µ = 1.
ii ) We consider monochromatic light,
u = u(x)e−iωt .
Equation (5.52) then becomes
ω
(∆ + k 2 )u = 0 with k = . (5.53)
c

Kirchhoff approximation

The 2nd Green formula is (D, an open set, u, v ∈ C 2 (D))

Z Z  
3 ∂u ∂v
(v∆u − u∆v)d x = v −u dσ.
D ∂D ∂n ∂n
126 Section 5.4

(This is a simple consequence of Gauss’s law:

Z Z
∇·W = W · ndσ with Wi = v∂i u − u∂i v. )
D ∂D

If we apply this formula to a solution u of equation (5.53) and for

1 eik|x−y|
v(x) = G(x − y) = − ,
4π |x − y|
(∆x + k 2 )G(x − y) = δ(x − y) ,

the Green function of the Helmholtz operator, ∆ + k 2 , we obtain

Z   Z
∂v ∂u  2 
u −v dσ = vk u + uδ(x − y) − k 2 vu d3 x
∂D ∂n ∂n D

u(y) y ∈ D
= (5.54)
0 y∈/D
∂u
Should we know the boundary conditions u(x) and ∂n
(x), x ∈ ∂D, we could then
find the solution u(y) in D.
E
In diffraction problems, one determines the influ- F
ence of non-transparent screens (E) and of aper- E
S D
tures (slits, F ) on the propagation of light.
F
E
The Kirchhoff approximation consists in assuming that

∂u 
u = 0, =0 on the screens E 

∂n , (5.55)
∂u ∂u0 

u = u0 , = in the slits F 
∂n ∂n
where u0 = Aeikr1 /r1 is the solution without screens (r1 is the distance to the
source). In the Kirchhoff approximation, we obtain
Z  
∂v ∂u0
u(y) = u0 −v dσ . (5.56)
F ∂n ∂n

If u = u0 and ∂n∂u
= ∂u
∂n
0
on both, the screens and the slits, we should find the
solution u0 [according to (5.54)]. Therefore (5.56) is equivalent to
Z  
∂v ∂u0
u(y) = u0 (y) − u0 −v dσ. (5.57)
E ∂n ∂n
This approximation is useful if the apertures F are relatively large relative to the
wavelength λ. From the theory of partial differential equations, it is known that
Ruth Durrer Electrodynamique Chap. 5 127

in general it is not possible to solve (5.53) by specifying the two conditions, u and
∂u
∂n
, at the boundaries: the solution u(x) found with (5.56) and (5.57) does not
satisfy in general the conditions (5.55). The mathematical contradiction of giving
∂u
both, u and ∂n at the boundaries can be eliminated if we choose the correct Green
function satisfying the correct boundary conditions
Dirichlet: GD = 0 on the screen and apertures, (5.58)
∂GN
Neumann: = 0 on the screens and apertures. (5.59)
∂n
In addition, we demand that G decreases at infinity, as the emitted wave:
  ikr
∂G r→∞ r→∞ e
r − ikG → 0, G → .
∂n r
With (5.58) and (5.59), we obtain
Z
∂GD
u(x) = u(x′ ) (x, x′ )dσ(x′ ) (5.60)
E∪F ∂n
Z
∂u ′
u(x) = − (x )GN (x, x′ )dσ(x′ ). (5.61)
E∪F ∂n
This approximation in not contradictory mathematically speaking. For a plane
screen, the Green functions GN and GD are easy to determine. One has
 ′′ 
′ 1 eikr eikr
GD,N (x, x ) = − ∓ ′′ . (5.62)
4π r r
The sign − corresponds to GD and the sign + corresponds to GN , and r = |x − x′ |
and r ′′ = |x′′ − x′ |. Here x′′ is the mirror position of x with respect to the screen
(see figure)

D
x’’ x

r
L r’
θ’ θ
x’
n

For x′ ∈ E ∪ F , we have r = r ′′ and

 
∂GD k eikr i n·r
= 1+
∂n 2πi r kr r
128 Section 5.4

where r = x − x′ . In the Kirchhoff approximation we then obtain

Z  
k eikr i n·r
u(x) = 1+ u0 (x′ )dσ. (5.63)
2πi F r kr r
′
Neglecting the term 1/r 2 , this gives for u0 = Aeikr /r ′ (r ′ is the distance between
x′ and the source)
Z ikr ikr′
kA e e
u(x) = cos θ dσ. (5.64)
2πi F r r ′
From the approximative Neumann formula, we obtain the same result with cos θ
replaced with cos θ′ and from the original formula (5.56) we obtain (5.64) with
cos θ replaced with 12 (cos θ + cos θ′ ). Next, we substitute cos θ by an average value
cos ϑ and 1/rr ′ by an average value 1/RR′ :
Z
k cos ϑ ′
u(x) = A eik(r+r ) dσ
2πi RR′ F

We place the split in a coordinate system in the (x, y) plane with O in the aperture
z x = (x, y, z)

R
r
y
(ξ, η, 0)
ecran
R’ r’
x

L (x’, y’, z’)

r 2 = (x − ξ)2 + (y − η)2 + z 2
r ′2 = (x′ − ξ)2 + (y ′ − η)2 + z ′2
R2 = x2 + y 2 + z 2 , R′2 = x′2 + y ′2 + z ′2
xξ + yη ξ 2 + η 2 (xξ + yη)2
r = R− + + + ...
R 2R 2R3
x′ ξ + y ′ η ξ 2 + η 2 (x′ ξ + y ′η)2
r′ = R′ − + + + ...
R′ 2R′ 2R′3

This gives
′ Z
k eik(R+R )
u(x) = A cos ϑ eikφ(ξ, η) dξdη (5.65)
2πi RR′ F
Ruth Durrer Electrodynamique Chap. 5 129

with
xξ + yη ξ 2 + η 2 (xξ + yη)2
φ(ξ, η) = − + + + ...
R 2R 2R3
x′ ξ + y ′η ξ 2 + η 2 (x′ ξ + y ′η)2
− + + + ...
R′ 2R′ 2R′2

Frauenhofer diffraction

We now assume that R and R′ are much larger than the dimension d of the aper-
ture, so that we can neglect the terms of order d/R and d/R′ . With the directional
cosines,
x y x′ y′
α = , β = ; α′ = ′ , β ′ = ′ ,
R R R R
and defining η
a = α + α′ , b = β + β ′ ,

we obtain B
Z ξ
const.
u(x) = exp(−ik(aξ + bη))dξdη.
λ F
A
For an aperture of rectangular shape, this integral gives an intensity
 2  2
2const sin kaA sin kbB
I(x) = |u(x)| = 2
(4AB)2 .
λ kaA kbB

For a source at position (α′, β ′ ) to be distinguished from a source at position

(α′′ , β ′′ ), it is necessary that the first zeros in the intensity of the image through
the “aperture” of size AB be well distinct. That is to say

π λ
α′ − α′′ > =
kA 2A
π λ
β ′ − β ′′ > = .
kB 2B
This diffraction limit is important in astronomy. A telescope of diameter D that
observes light of wavelength λ is able to distinguish two stars only if their angular
separation δ is large enough:
λ
δ> .
D
130 Section 5.4

Exercise: What is the diffraction limit of an optical telescope with a mirror of

4 m (take red light)?

Answer: 0.036 arc seconds .

THE END