Advanced Electrodynamics Lecture Notes

Wouter van Burik (4453839)

November 18, 2020

1
Contents
1 Introductory Lecture 6
1.1 Lecture 1a: Recap . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2 Forces and fields . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Charges and currents . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3.1 Maxwell equations . . . . . . . . . . . . . . . . . . . . . . 7
1.3.2 Gauss’ law . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3.3 Gauss’ law for magnetism . . . . . . . . . . . . . . . . . . 8
1.3.4 Faraday’s law . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.3.5 Ampère-Maxwell law . . . . . . . . . . . . . . . . . . . . . 9
1.3.6 Maxwell and charge conservation . . . . . . . . . . . . . . 9
1.3.7 Material media . . . . . . . . . . . . . . . . . . . . . . . . 10
1.4 Lecture 1b: Mathematics . . . . . . . . . . . . . . . . . . . . . . 10
1.4.1 Vector calculus: Cartesian, cylindrical and spherical co-
ordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.4.2 Vector calculus: Gradient, divergence and curl . . . . . . 13
1.4.3 Vector calculus: Gauss’ theorem . . . . . . . . . . . . . . 14
1.4.4 Index gymnastics . . . . . . . . . . . . . . . . . . . . . . . 14
1.4.5 Einstein summation convention . . . . . . . . . . . . . . . 14
1.4.6 Kronecker-delta and Levi-Civita symbol . . . . . . . . . . 15
1.4.7 Vector operator identities . . . . . . . . . . . . . . . . . . 15
1.4.8 Dirac delta-function . . . . . . . . . . . . . . . . . . . . . 16

2 Electrostatics 18
2.1 The electric field and the electrostatic potential . . . . . . . . . . 18
2.2 Multipole expansion . . . . . . . . . . . . . . . . . . . . . . . . . 18

3 Magnetostatics 22
3.1 Biot-Savart law . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.2 Vector potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.3 Magnetic dipole moment . . . . . . . . . . . . . . . . . . . . . . . 23
3.4 Magnetic media . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

4 Electromagnetic forces and conservation laws 25

4.1 Electromagnetic forces . . . . . . . . . . . . . . . . . . . . . . . . 25
4.2 Conservation of energy . . . . . . . . . . . . . . . . . . . . . . . . 25
4.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.3.1 Ohmian dissipation DC-current . . . . . . . . . . . . . . . 25
4.3.2 Skin effect and Ohmian dissipation AC-current . . . . . . 26

5 Waves in vacuum 28
5.1 Wave equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
5.2 Plane waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
5.3 Wave packets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
5.3.1 Free-space diffraction . . . . . . . . . . . . . . . . . . . . . 31

2
6 Waves in simple matter I 33
6.1 Plane waves in matter . . . . . . . . . . . . . . . . . . . . . . . . 34
6.1.1 Dispersion . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
6.1.2 Energy balance . . . . . . . . . . . . . . . . . . . . . . . . 35
6.2 Reflection and transmission at a plane boundary . . . . . . . . . 35
6.2.1 Basic approach . . . . . . . . . . . . . . . . . . . . . . . . 35
6.2.2 Snell’s law . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
6.2.3 Fresnel equations . . . . . . . . . . . . . . . . . . . . . . . 36
6.2.4 Polarization by reflection . . . . . . . . . . . . . . . . . . 36
6.2.5 Total internal reflection . . . . . . . . . . . . . . . . . . . 37
6.2.6 Evanescent waves . . . . . . . . . . . . . . . . . . . . . . . 38

7 Waves in simple matter II 39

7.1 Radiation pressure . . . . . . . . . . . . . . . . . . . . . . . . . . 39
7.2 Layered matter . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
7.2.1 Fabry-Perot geometry . . . . . . . . . . . . . . . . . . . . 39
7.2.2 Stokes Relations . . . . . . . . . . . . . . . . . . . . . . . 40
7.2.3 Matrix formalisms . . . . . . . . . . . . . . . . . . . . . . 40
7.3 Conducting matter . . . . . . . . . . . . . . . . . . . . . . . . . . 41
7.3.1 Monochromatic plane waves . . . . . . . . . . . . . . . . . 42
7.3.2 Skin depth . . . . . . . . . . . . . . . . . . . . . . . . . . 42
7.3.3 Reflection from good conductor . . . . . . . . . . . . . . . 43

8 Waves in dispersive matter 44

8.1 Frequency dispersion . . . . . . . . . . . . . . . . . . . . . . . . . 44
8.2 Response of a medium . . . . . . . . . . . . . . . . . . . . . . . . 44
8.2.1 Causal response of medium: time domain . . . . . . . . . 44
8.2.2 Response of a medium: frequency domain . . . . . . . . . 45
8.3 Equivalence of descriptions . . . . . . . . . . . . . . . . . . . . . 45
8.4 Classical models . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
8.4.1 Drude model . . . . . . . . . . . . . . . . . . . . . . . . . 47
8.4.2 Lorentz Model . . . . . . . . . . . . . . . . . . . . . . . . 48
8.5 Wave packets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
8.5.1 Group velocity dispersion . . . . . . . . . . . . . . . . . . 50
8.6 Causality and consequences . . . . . . . . . . . . . . . . . . . . . 50
8.6.1 Basic equation of causality . . . . . . . . . . . . . . . . . 50
8.6.2 Kramers-Kronig relations . . . . . . . . . . . . . . . . . . 51

9 Retardation and Radiation 52

9.1 Difference with previous lectures . . . . . . . . . . . . . . . . . . 52
9.2 Inhomogeneous wave equations . . . . . . . . . . . . . . . . . . . 52
9.3 Potentials of inhomogeneous wave equations . . . . . . . . . . . . 53
9.4 Gauge conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
9.5 Coulomb gauge . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
9.6 Lorenz gauge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
9.7 Retardation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

3
 9.8 Advanced and retarded waves . . . . . . . . . . . . . . . . . . . . 54
9.9 Retarded solution . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
9.10 Advanced solution . . . . . . . . . . . . . . . . . . . . . . . . . . 55
9.11 Physical relevance of the two solutions . . . . . . . . . . . . . . . 55
9.12 Retarded harmonic waves . . . . . . . . . . . . . . . . . . . . . . 55
9.13 Time-dependent electric dipole . . . . . . . . . . . . . . . . . . . 56
9.14 Shape of fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

10 Lienard-Wiechert potentials 58
10.1 Oscillating electric dipole v  c . . . . . . . . . . . . . . . . . . . 58
10.2 Fields of the harmonic oscillating electric dipole . . . . . . . . . . 60
10.3 Liénard-Wiechert potentials . . . . . . . . . . . . . . . . . . . . . 61

11 Synchrotron Radiation 66
11.1 Liénard-Wiechert potentials recap . . . . . . . . . . . . . . . . . 66
11.2 Fields of a point charge in uniform motion . . . . . . . . . . . . . 66
11.3 Synchrotron Radiation . . . . . . . . . . . . . . . . . . . . . . . . 69
11.3.1 Synchrotron radiation amplification via Lorentz contrac-
tion of the undulator . . . . . . . . . . . . . . . . . . . . . 70
11.3.2 Synchrotron radiation amplification via Doppler shift . . 70
11.3.3 Synchrotron radiation amplification via transverse nar-
rowing of of emission cone . . . . . . . . . . . . . . . . . . 71
11.4 Alternative picture: Fourier transform of ultrashort pulse . . . . 71
11.5 Angular emission and power of electromagnetic radiation . . . . 73
11.5.1 Case A: Dipole radiation of linear movement with v  c . 73
11.5.2 Case B: Dipole radiation of linear movement with v ≈ c . 73
11.5.3 Case C: Dipole radiation of circular movement with v ≈ c 74

12 Synchrotron Radiation and Cherenkov Radiation 75

12.1 Angular emission and power of electromagnetic radiation . . . . 75
12.2 Synchrotron Radiation: frequency spectrum . . . . . . . . . . . . 75
12.3 Finding the frequency spectrum in six steps . . . . . . . . . . . . 77
12.3.1 Step 1 Change of variable . . . . . . . . . . . . . . . . . . 77
12.3.2 Step 2 ’Golden Rule’ . . . . . . . . . . . . . . . . . . . . . 78
12.3.3 Step 3 Consider a relativistic electron in a circular orbit . 78
12.3.4 Step 4 Two small-angle approximations . . . . . . . . . . 79
12.3.5 Step 5 Taylor expansion of advanced/observer time t in
terms of retarded/particle time tr . . . . . . . . . . . . . 79
12.3.6 Step 6 Solve to get the electric field Ex (ω) . . . . . . . . . 79
12.3.7 Comments on frequency spectrum . . . . . . . . . . . . . 80
12.4 Take home messages . . . . . . . . . . . . . . . . . . . . . . . . . 80
12.5 Cherenkov radiation . . . . . . . . . . . . . . . . . . . . . . . . . 81
12.5.1 Radiation of a particle moving slower than the speed of
light . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
12.5.2 Radiation of a particle moving faster than the speed of light 87

4
13 Relativistic motion in EM-fields 89
13.1 Proper time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
13.2 Relativistic motion of particles in electric fields . . . . . . . . . . 90
13.2.1 Linear acceleration in constant E-field with v k E . . . . . 90
13.2.2 Relativistic motion of particles in electric fields with v ⊥ E 93
13.3 4-vectors and relativistic mechanics . . . . . . . . . . . . . . . . . 94
13.3.1 Galilean transformation . . . . . . . . . . . . . . . . . . . 94
13.3.2 Lorentz transformation . . . . . . . . . . . . . . . . . . . 95
13.4 Four-vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
13.4.1 Velocity four-vector . . . . . . . . . . . . . . . . . . . . . 98
13.4.2 Potential field four vectors . . . . . . . . . . . . . . . . . . 98
13.4.3 Four-vectors: derivatives and the Lorenz Gauge . . . . . . 98
13.4.4 Four vectors: Charge conservation . . . . . . . . . . . . . 99
13.4.5 Four-vectors: Maxwell equations (in terms of potential
fields) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
13.5 Electromagnetic field tensor . . . . . . . . . . . . . . . . . . . . . 100
13.6 Take home messages . . . . . . . . . . . . . . . . . . . . . . . . . 102

14 Relativistic equation of motion 103

14.1 Four-vectors: derivatives, Lorenz gauge and charge conservation. 103
14.2 Relativistic form of Maxwell (in terms of potential fields) . . . . 104
14.3 Electromagnetic field tensor . . . . . . . . . . . . . . . . . . . . . 104
14.4 Particle in an E × B field . . . . . . . . . . . . . . . . . . . . . . 106
14.5 Particle in a B-field . . . . . . . . . . . . . . . . . . . . . . . . . 106
14.6 Relativistic motion of particles in electric fields . . . . . . . . . . 107
14.7 Relativistic equation of motion . . . . . . . . . . . . . . . . . . . 108
14.8 Solving relativistic motion of particles in electric fields using rel-
ativistic notation . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

5
1 Introductory Lecture
1.1 Lecture 1a: Recap
So far, our experience with describing electromagnetism has come down to four
topics:
1. Forces and fields
2. Charges and currents
3. The four Maxwell equations
4. Interaction with matter

1.2 Forces and fields

The force on a charge particle is given by:
F = q(E + v × B (1)
Where the electric field is E and gives an electric force (Coulomb force) on a
charge, and a magnetic force (Lorentz force) on a current.

1.3 Charges and currents

We described charges and currents in the discrete limit as a collection of indi-
vidual charges or moving charges. So a charge density is given by:
N
X
ρ(r, t) = qk δ(r − rk ) (2)
k=1
A current density is given by:
N
X
j(r, t) = qk vk δ(r − rk (3)
k=1

The total charge in a volume is given by:

Z
Q= d3 rρ(r) (4)
V

Note that the volume element d3 r = dxdydz. The current through a surface is:
Z
dQ
I= = dS · j (5)
dt S

This allows us to write a conservation of charge law. We know that the decrease
of charge in a volume is given by:
Z Z
dQ d ∂ρ
− =− d rρ = − d3 r
3
(6)
dt dt ∂t

6
The decrease of charge must equal the flow of current through a surface around
that same volume: I Z
I= dS · j = d3 r∇ · j (7)
S V
Where we used Gauss’s theorem to rewrite the surface integral to a volume
integral. Now we can compare the two equations:
Z Z
∂ρ
− d3 r = d3 r∇ · j (8)
∂t V
Z Z
∂ρ
d3 r + d3 r∇ · j = 0 (9)
∂t V
For this to be true for any arbitrary volume V , this must also be satisfied locally:
∂ρ
+∇·j=0 (10)
∂t
We call this equation the continuity equation. We now define different charge
densities for different geometries. For a charge in a volume, we say that the
charge is given by: Z
Q= d3 rρ(r) (11)
V
A surface charge density gives the following total surface charge:
Z
Q= dSσ(rS ) (12)
S

Finally, the 1-dimensional case of a charge on a line:

Z
Q= dlλ(rL ) (13)
L

1.3.1 Maxwell equations

The four Maxwell equations are:

∇ · E = ρ0 Gauss’ law

∇·B=0 also Gauss’ law
(14)
∇ × E = − ∂B ∂t Faraday’s law
∇ × B = µ0 j + c12 ∂E
∂t Ampère-Maxwell law

The speed of light c is here given by:

1
c= √ (15)
0 µ0

Gauss’ first and second law are homogeneous, meaning that having E and B
equal zero is a valid solution. However, Faraday’s law and Ampère-Maxwell’s
law are inhomogeneous because of the sources ρ and j.

7
1.3.2 Gauss’ law
Gauss’s law in differential form is:
ρ
∇·E= (16)
0
In integral form it’s written as:
I
Q
φE = dS · E = (17)
S 0

The flux φE is here the electric flux through a closed surface S. This means that
charges generate electric fields.

1.3.3 Gauss’ law for magnetism

In differential form, Gauss’ law for magnetism is given by:

∇·B=0 (18)

Gauss’ law in integral form is:

I
φB = dS · B = 0 (19)
S

The flux in this equation is the magnetic flux through a closed surface S. In
other words, there’s no magnetic charge, and magnetic field lines are always
closed loops.

1.3.4 Faraday’s law

Faraday’s law in differential form is:
∂B
∇×E=− (20)
∂t
We can rewrite this using Stokes’ theorem to:
I Z 
d dφB
V = dl · E = − dS · B = − (21)
C dt S dt
I
dφB
V = dl · E = − (22)
C dt
In other words, a changing magnetic field generates an electric field. This is
what creates a voltage in a coil when you move a magnet through that coil.

8
1.3.5 Ampère-Maxwell law
The Ampère-Maxwell law in differential form is:
1 ∂E
∇ × B = µ0 j + (23)
c2 ∂t
In integral form, it’s:
I Z 
d
dl · B = µ0 I + µ0 0 dS · E (24)
C dt S

That last term is the ’displacement current’ through a surface S. The Ampère-
Maxwell law tells you the converse of Faraday’s law: magnetic fields are gener-
ated by currents and by changing electric fields!

1.3.6 Maxwell and charge conservation

We can show that the Maxwell equations satisfy charge conservation. The
continuity equation is:
∂ρ
+∇·j=0 (25)
∂t
We use Gauss’ law for the first term and Ampère-Maxwell for the second term:
∂ρ d ∂∇ · E
= 0 ∇ · E = 0
∂t dt  ∂t 
1 1 ∂E
∇·j= ∇· ∇×B− 2
µ0 c ∂t

So the continuity equation says that:

 
∂ρ d ∂∇ · E 1 1 ∂E
+ ∇ · j = 0 ∇ · E = 0 + ∇· ∇×B− 2 (26)
∂t dt ∂t µ0 c ∂t

We know that for any vector field F:

∇ · (∇ × F) = 0 (27)

So we can say that:

1
∇·∇×B=0 (28)
µ0
So we’re left with:
∂∇ · E 1 1 ∂E
0 − ∇· 2 (29)
∂t µ0 c ∂t
Spatial derivatives and time derivatives commute, so we can pull the ∇ in front
of the time derivative and we said previously that:
1 1
c= √ → µ0 = 2 (30)
0 µ0 c

9
And so we find that:
∂E 0 µ0 ∂E ∂E ∂E
0 ∇ · − ∇· = 0 ∇ · − 0 ∇ · =0 (31)
∂t µ0 ∂t ∂t ∂t
So Maxwell’s equation satisfy charge conservation. Note that without the dis-
placement current term, this wouldn’t be true!

1.3.7 Material media

we distinguish in media between free and bound charges:

ρ(r, t) = ρf (r, t) − ∇ · P(r, t) (32)

Similarly, we can distinguish between free and bound currents:

∂P(r, t)
j(r, t) = jf (r, t) + ∇ × M(r, t) + (33)
∂t
The ’bound’ charge and current are expressed through the polarization P and
magnetization M. The polarization is the electric dipole moment per unit vol-
ume, and the magnetization is the magnetic dipole moment per volume. We use
these equations to define new fields, so called auxiliary fields. The ’new’ electric
and magnetic fields are given by:
D(r, t) = 0 (r, t) + P(r, t)
(34)
H(r,t) = µ−1
0 B(r, t) − M(r, t)

This new electric field D is sometimes called the dielectric displacement field.
The field H is sometimes called the magnetic field, and in this context, B
becomes the ’magnetic induction field’. This changes Maxwell’s equations to:
∇ · D = ρf ∇·B=0
(35)
∇ × E = − ∂B
∂t ∇ × H = jf + ∂D
∂t

1.4 Lecture 1b: Mathematics

1.4.1 Vector calculus: Cartesian, cylindrical and spherical coordi-
nates
We can define a vector quantity in three ways, that is, in three coordinate
systems. The first is the Cartesian representation:

V = Vx x̂ + Vy ŷ + Vz ẑ (36)

The second is the cylindrical representation:

V = Vρ ρ̂ + Vφ φ̂ + Vz ẑ (37)

And the third representation is spherical:

V = V r̂ + Vθ θ̂ + Vφ φ̂ (38)

10
The unit vectors in the spherical representation are found via simple geometry:

~r = xx̂ + y ŷ + z ẑ = r cos φ sin θx̂ + r sin φ sin θŷ + r cos θẑ

~r = r(cos φ sin θx̂ + sin φ cos θŷ + cos θẑ) (39)
~r = rr̂(φ, θ)

x = r cos φ sin θ
y = r sin φ sin θ (40)
z = r cos θ
If we have some vector ~r that depends on some parameter u, such that we have
~r(u), we can define a vector pointing in the direction of increasing u as follows:

∂~r
~eu = (41)
∂u
Which looks a lot like a gradient. We can turn this into a unit vector by dividing
it by its magnitude:
~eu
êu = (42)
|eu |
So as an example, say we want to define the unit vector êx . Then we’d write:

∂~r ∂
êx = = (xx̂ + y ŷ + z ẑ) = x̂ (43)
∂x ∂x
So what if we now try taking the derivative of r̂ with respect to r?

∂~r
êr = = r̂(θ, φ) (44)
∂r
This derivative is simply:

∂~r ∂ ∂
= r(cos φ sin θx̂ + sin φ sin θŷ + cos θẑ)~r = rr̂(φ, θ) = r̂(φ, θ) (45)
∂r ∂r ∂r
êr = cos φ sin θx̂ + sin φ sin θŷ + cos θẑ (46)
Now if we want to have the unit vector in the θ direction, we write:

∂~r ∂ ∂
eθ = = (rr̂(θ, φ)) = r [r̂(θ, φ)] = (47)
∂θ ∂θ ∂θ
Taking the derivative of ~r with respect to θ gives:
∂
~eθ = r (cos φ sin θx̂ + sin φ sin θŷ + cos θẑ) (48)
∂θ
~eθ = r(cos φ cos θx̂ + sin φ cos θŷ − sin θẑ) (49)

11
To convert this to a unit vector, we need to divide this by its magnitude. The
magnitude of this is going to be:
q
|r| = r2 cos2 φ cos2 θ + sin2 φ cos2 θ + sin2 θ


q
= r2 cos2 θ cos2 φ + sin2 φ + sin2 θ



q
= r2 cos2 θ (1) + sin2 θ


q (50)
= r2 cos2 θ + sin2 θ


p
= r2 (1)
√
= r2
=r

Ergo:
~eθ ~eθ
êθ = = = cos φ sin θx̂ + sin φ sin θŷ + cos θẑ (51)
|eθ | r
And so we’ve found the unit vector. We can now do the same to find the unit
vector êφ :

∂~r ∂
êφ = = r (r̂(θ, φ)) = r(− sin φ sin θx̂ + cos φ sin θŷ) (52)
∂φ ∂φ
Its magnitude is:
q
r2 sin2 φ sin2 θ + cos2 φ sin2 θ


|eφ | =
q
|eφ | = r2 sin2 φ(sin2 θ + cos2 φ)


q
|eφ | = r2 sin2 φ(1)

(53)
q
|eφ | = r2 sin2 φ

|eφ | = r sin θ

This we recognize from the scaling factors for a gradient in spherical coordinates:
∂f 1 ∂f 1 ∂f
∇f = r̂ + θ̂ + φ̂ (54)
∂r r ∂θ sin φ ∂r
And here we have the three unit vectors in spherical coordinates:

êr = cos φ sin θx̂ + sin φ sin θŷ + cos θẑ (55)

êθ = cos θ cos φx̂ + cos φ sin φŷ − sin θẑ (56)
êφ = sin φx̂ + cos φŷ (57)

12
The inverse relationship is:

x̂ = r̂ sin θ cos φ + θ̂ cos θ cos φ − φ̂ sin φ (58)

ŷ = r̂ sin θ sin φ + θ̂ cos sin φ + φ̂ cos φ (59)

ẑ = r̂ cos θ − θ̂ sin θ (60)
In a similar manner we can derive the unit vectors in a cylindrical system:

ρ̂ = x̂ cos φ + ŷ sin φ (61)

φ̂ = −x̂ sin φ + ŷ cos φ (62)

The inverse is given by:
x̂ = ρ̂ cos φ − φ̂ sin φ (63)
ŷ = ρ̂ sin φ + φ̂ cos φ (64)

1.4.2 Vector calculus: Gradient, divergence and curl

We define now the gradient or ”nabla” operator as a spatial derivative. This
operator’s formulation depends on the coordinate system you use. In Cartesian
coordinate systems, it’s given by:
∂ ∂ ∂
∇ = x̂ + ŷ + ẑ (65)
∂x ∂y ∂z
In cylindrical coordinates it’s given by:
∂ 1 ∂ ∂
∇ = ρ̂ + φ̂ + ẑ (66)
∂ρ ρ ∂φ ∂z
And in spherical coordinates:

θ̂ ∂ φ̂ ∂
∇ = r̂ + (67)
r ∂θ r sin θ ∂φ
A gradient raises a tensor’s order by 1. The divergence, or ’div’, lowers the
tensor order by 1. In other words, it turns a matrix into a vector and a vector
into a scalar. In Cartesian coordinates it’s given by:
∂Vx ∂Vy ∂Vz
∇·V= + + (68)
∂x ∂y ∂z
In cylindrical coordinates, it’s given by:
1 ∂(ρVρ 1 ∂Vφ ∂Vz
∇·V= + + (69)
ρ ∂ρ ρ ∂φ ∂z
Finally, in spherical coordinates, the divergence is:
1 ∂(r2 Vr ) 1 ∂(sin θVθ ) 1 ∂Vφ
∇·V= + + (70)
r2 ∂r r sin θ ∂θ r sin θ ∂φ

13
1.4.3 Vector calculus: Gauss’ theorem
Given a volume V , bounded by a surface S with surface element dS = dS n̂:
Z Z
3
d r∇ · F = dS · F (71)
V S

Finally, we define the curl-operator, which maintains tensor-order. So the curl

of a second order tensor is a second order tensor, the curl of a vector is a vector
and so on. In Cartesian components, it’s given by:
     
∂Vz ∂Vy ∂Vx ∂Vz ∂Vy ∂Vx
∇×V= − x̂ + − ŷ + − ẑ (72)
∂y ∂z ∂z ∂x ∂x ∂y
In cylindrical coordinates it’s given by:
     
1 ∂Vz ∂Vφ ∂Vρ ∂Vz 1 ∂(ρVφ ∂Vρ
∇×V= − ρ̂ + − φ̂ + − ẑ (73)
ρ ∂φ ∂z ∂z ∂ρ ρ ∂ρ ∂φ
Finally, in spherical coordinates it’s given by:
     
1 ∂(sin θVφ ) ∂Vθ 1 1 ∂Vr ∂(rVφ ) 1 ∂(rVθ ) ∂Vr
∇×V = − r̂+ − θ̂+ − φ̂
r sin θ ∂ρ ∂φ r sin θ ∂φ ∂r r ∂r ∂θ
(74)
We use this to define Stokes’ theorem:
Z I
dS · ∇ × F = dl · F (75)
S C

The Laplacian is the divergence of the gradient.

1.4.4 Index gymnastics

We define the components of a vector via index notation as follows:
Ak (76)
For k = x, y, z or k = 1, 2, 3. The components of a matrix are then:
Cij (77)
For i = x, y, z and j = x, y, z. The components of the gradient operator are in
this notation:
∂
∇k = ∂k = (78)
∂rk

1.4.5 Einstein summation convention

The Einstein summation convention says we drop the summation sign whenever
we have repeated indices. For example:

X3



A · B =
Ak B k (79)

k=1

14
And a matrix product becomes:

X3



(AB)ij =
Aik Bkj (80)

k=1
We remove the explicit summation sign for the dummy index k. The dummy
index must appear twice for this, and it must appear in a product or an operator.

1.4.6 Kronecker-delta and Levi-Civita symbol

The Kronecker delta is given by:

1i=j
δij = (81)
6 j
0i=

We use this in our proofs to reduce the number of indices. For example:

Ci Dj ∂i rj = Ci Dj ∂ij = Ci Di (82)

We then define the Levi-Civita symbol to shorten our index notation:

Ci = ijk Aj Bk (83)

This is 0 if two of the indices are equal, +1 for a cyclic permutaiton of xyz and
-1 for a cyclic permutation of yxz. This allows us to write cross-products more
concise:
[V × F]i = ijkVj Fk (84)
[∇ × A]i = ijk∂j Ak (85)
We use the Levi-Cività to make index gymnastics easier and find vector operator
identities. For example:
δij ijk = iik = 0 (86)

1.4.7 Vector operator identities

Using the Levi-Cività symbol and the Kronecker-Delta, we can prove that the
curl of a gradient is always zero in one line:

C = ∇ × ∇f (87)

Writing it out in index notation:

Ci = ijk ∂j ∂k f (88)

We use the anti-cyclic permutation of the Levi-Cività to write that:

Ci = ijk ∂j ∂k f = ijk ∂k ∂j f = −ikj ∂k ∂j f = −ijk ∂j ∂k f (89)

15
In the last step we swapped the dummy indices k and j. It then follows that
Ci = 0, so we obtain that:
∇ × ∇f = 0 (90)
Similarly, we can prove that the divergence of a curl is zero as follows:

C = ∇ · (∇ × F) = ∂i (ijk ∂j Fk ) = ijk ∂i ∂j Fk = −jik ∂j ∂i Fk = −ijk ∂i ∂j Fk = 0

(91)
Where in the second last step we used an anti-cyclic permutation of the Levi-
Cività symbol, and then swapped the dummy indices. This means that:

∇ · (∇ × F) = 0 (92)

For any vector field F. A difficult identity is the product of two Levi-Civita‘s:

ijk ilm = xjk xlm + yjkylm + zjk zlm (93)

xjk xlm = xyz xyz = 1

= xyz xzy = 1
(94)
= xzy xyz = 1
= xzy xzy = 1
Similarly so for the other two terms. This gives us that:

ijk ilm = 1 (95)

If j = l and k = m, and −1 if j = m and k = l. We use the Kronecker delta to

write then:
ijk ilm = δjl δkm − δjm δkl (96)
You can use this to prove the ’double curl’ identity, which is example 1.3 of
Zangwill. The identity is simply:

∇ × (∇ × A) = ∇(∇ · A) − ∇2 A (97)

1.4.8 Dirac delta-function

We define the Dirac-delta function in the physicist way as follows:

δ(x) = 0 for x 6= 0
Z ∞
(98)
dxδ(x) = 1
−∞

By replacing the coordinate x with r we can expand this two 3D. In Cartesian
coordinates, the delta-function is:

δ(r) = δ(x)δ(y)δ(z) (99)

16
Example 1.5 of Zangwill shows how you can use this to find that:
1
∇2 = −4πδ(r) (100)
r
We now define Helmholtz’ theorem. This theorem says that we can split any
vector field into 2 vector fields: a parallel part and a longitudinal part, or one
with no curl and one with no divergence:

C = C⊥ + Ck (101)

Such that:
∇ · C⊥ ∇ × Ck = 0 (102)
We group these together into:

C(r = ∇ × F(r) − ∇Ω(r) (103)

The first term is the transversal part, and the second term is the longitudinal
part. The vector F(r) and the scalar Ω(r) can be obtained form the curl: ∇ × C
and divergence ∇ · C. To show this explicitly, we take the divergence of our
vector field:
∇ · C = ∇ · C⊥ + Ck = −∇2 Ω(r)


(104)
Since the divergence of a curl is always zero, the first term drops out. Similarly,
if we take the curl, we get:

∇ × C = ∇ × C⊥ + Ck = ∇ × (∇ × F(r)) = ∇(∇ · F(r)) − ∇2 F(r) (105)

We can write these then in integral form as follows:

∇0 · C(r0
Z
1
Ω(r) = d3 r
4π |r − r0 |
(106)
∇0 × C(r0
Z
1
F(r) = d3 r
4π |r − r0 |

Provided however, that ∇ · F(r) = 0!

17
2 Electrostatics
2.1 The electric field and the electrostatic potential
In electrostatics we assume that:

j=0 (107)

So no moving charges means no time-dependence, so the Maxwell equations

become:
∇ × E(r = 0 (108)
0 ∇ · E(r) = ρ(r) (109)
The task is to solve for the electric field given a charge density. We use the
Helmholtz theorem to write the electric field in terms of an electrostatic poten-
tial:
E(r = −∇φ(r) (110)
Where the potential is given by:
ρ(r0 )
Z
1
φ(r) = d3 r0 (111)
4π0 |r − r0
The electric field is then:
r − r0
Z
1
E(r) = d3 r0 ρ(r0 )ρ(r0 ) (112)
4π0 |r − r0 |
The electrostatic potential satisfies Poisson’s equation:

∇2 φ(r) (113)

2.2 Multipole expansion

Localized and non-uniform charge distributions can be described by a Taylor
expanded potential field. We expand the spatial dependence as follows:
1 1 1 1 1
= − r0 · ∇ + (r0 · ∇)2 (114)
|r − r0 | r r 2 r
Why does it look like this? Let’s go through it step by step. Let’s we have a
function f that depends only on the distance to the origin:
1 1
f (~r) = =p (115)
|~r| x2 + y 2 + z 2
Its gradient is then:
 
1 x y z ~r
∇f (~r) = ∇ = −p x̂− p ŷ− p ẑ = −
|~r| 2 2
x +y +z 2 2 2
x +y +z 2 2 2
x +y +z 2 |~r|
(116)

18
Now the Taylor expansion. We know that:
1 1 1 1
f (x + c) = f (c) + f 0 (c)(x − c) + f (2) (x − c)2 + f (3) (x − c)3 + ... (117)
0! 1! 2! 3!
1
f (r = r0 ) →
r
1
f 0 (r = r0 )(r − r0 ) → (r − r0 )∇ · (118)
r − r0
 
1 00 0 0 2 1 2 1
f (r = r )(r − r ) → ∇ · (r − r0 )2
2 2 r − r0
Now, if we shift our coordinate by r0 The electrostatic potential is then given
as:
Z  Z   Z  
1 3 0 0 1 3 0 0 0 1 1 3 0 0 1
φ(r) = d r ρ(r ) − d r ρ(r )ri ∇i + d r ρ(ri ) ∇i ∇j
4π0 r r 2 r
(119)
The first term is the electric monopole moment (=total charge):
Z X
Q = d3 rρ(r) = qα (120)
α

The second term is the electric dipole moment p:

Z X
p = d3 rρ(r)r = qα rα (121)
α

If the total charge is zero, the electric dipole moment is independent of the
origin. The third term is the electric quadrupole moment:
Z
1X
Qij = d3 rρ(r)ri rj = qα rαi rαj (122)
2 α

So we can now give a simplified potential as follows:

3ri rj − r2 δij
 
1 Q p·r
φ(r) = + 3 + Qij (123)
4π0 r r r5
Where we used that:  
1 rj
∂j =− 3 (124)
r r
3ri rj − r2 δij
 
1  rj 
∂i ∂j = ∂i − 3 = (125)
r r r3
This offers a far simplified description if the series converges rapidly. The
dipole electric field is given by:
1 3r̂(r̂ · p − p)
E(r) = (126)
4π0 r3

19
For r >> R. Its potential is given by:
1 p · r
φ(r) = (127)
4π0 r3
Must hold for all |r| > 0. There’s a singularity at the origin, which is accounted
for by upgrading the electric field equation to:
 
1 3r̂(r̂ · p − p) 4π
E(r) = − pδ(r − r0 ) (128)
4π0 |r − r0 |3 3
The force on a dipole is given by:

(p · ∇)E(r) (129)

There’s a net force if there’s a gradient in the electric field along the dipole
moment p. The torque on a dipole is given by:

N = p × E(r) (130)

The torque orients the dipole moment p along the electric field E. Dipole energy
is given by:
Ve (r) = −p · E(r) (131)
The force is the gradient of the potential energy:

F = −∇VE = ∇(p · E) (132)

This force expression is the same as the previous force expression. The total
dipole moment of a body is the volume integral of the polarization per unit
volume: Z
d3 rP(r) = p (133)
V
In a dielectric, the total charge density has two contributions:

ρ(r, t) = ρf (r, t) − ∇ · P(r, t) (134)

The first term is the free charge, the second term is the bound charge, which
is the displacement of the electron cloud with respect to its respective nucleus.
The total bound charge is zero:
Z Z
d3 rρP (r) + dSσp (rs ) = 0 (135)
V S

Where:
ρ(r ) = −∇ · P(r)
(136)
σP (rs ) = P(rS ) · n̂(rS )

So the total polarization must be zero. We then define the axuliary field D:

0 ∇ · E(r) = ρ(r) = ρP (r) + ρf (r) = −∇ · P(r + ρf (r (137)

20
 D(r) = 0 E(r + P(r) (138)
Such that:
∇D = ρf (139)
And:
∇×D=∇×P (140)
This last identity holds because the curl of the electric field is zero in the elec-
trostatic case. For a simple dielectric, we assume a linear and isotropic response
to to an electric field:

P = 0 χE = ( − 0 )E = 0 (κ − 1)E (141)

Where  is the permittivity, χ the susceptibility and κ the dielectric constant.

The energy stored in a dielectric is given by:
Z
1
Ue [D] = d3 rE · D (142)
2

21
3 Magnetostatics
Magnetostatics is similar to electrostatics if you replace the word ’charge’ in the
electrostatic context with ’current’. The governing Maxwell equations in the
case of steady currents and fields are:

∇ · B(r) = 0
(143)
∇ × B(r) = µ0 j(r)

3.1 Biot-Savart law

The first law is Gauss’s law of magnetism, the second is Ampère’s law. The goal
is to solve for the magnetic field given a current density. The governing equation
is in this context the Biot-Savart law. To find this, we apply the Helmholtz’
theorem to equations for the magnetic field:
j(r0 )
Z
µ0
B(r) = ∇ × d3 r0 (144)
4π |r − r0 |
We apply the curl operator inside the integral:
j × (r − r0 )(r0 )
Z
µ0
B(r) = d3 r0 (145)
4π |r − r0 |3

In the case of a current through a wire (so when jd3 r = Idl), this becomes:

dl × (r − l0 )(r0 )
Z
µ0 I
B(r) = d3 r0 (146)
4π |r − l|3
Applying this to a current through an infinite wire of current:
µ0 I
B(ρ) = φ̂ (147)
2πρ

3.2 Vector potential

The vector potential is given by:

B(r = ∇ × A(r) (148)

Where the vector potential is given by:

j(r0 )
Z
µ0
A(r) = d3 r 0 (149)
4π |r − r0 |
There’s a freedom in how we define the vector potential. This freedom is called
the gauge freedom. We can add a gradient to our vector potential and still get
the same magnetic field B:

A0 (r) = A(r) + ∇χ(r) (150)

22
We can do this since the curl of a gradient is always zero. We use two gauges:
the Coulomb gauge and the Lorentz gauge. The Coulomb gauge is used most
often to fix the vector potential uniquely:

∇ · A(r) = 0 (151)

3.3 Magnetic dipole moment

We then define a magnetic dipole moment
Z
m = IS = I dS (152)

Or more formally: I
1
m= I r × dl (153)
2 C
Our magnetic dipole moment can also be given as:
Z
1
m= d3 rr × j(r) (154)
2
The vector potential in terms of the magnetic dipole moment is given by:
µ0 m × r
A(r) = (155)
4π r3
The magnetic field in terms of the magnetic dipole moment is:

µ0 r µ0 3r̂(r̂ · m) − m
B(r) = − (m · ∇) 3 = (156)
4π r 4π r3
The force on a magnetic dipole by an external magnetic fiel B is then given by:

F = ∇(m · B) (157)

The torque on a current loop is given by:

N=m×B (158)

Same as in the electric case. Since the magnetic force is a conservative force,
the energy must be given by:

V̂B (r) = −m · B(r) (159)

Again, same as in the electric case. The total magnetic dipole moment of a
body is then defined in terms of the ’magnetic dipole density’, the so-called
magnetization M: Z
m = d3 rM (160)

23
Just like with the electric polarization which was due to free charge and bound
charge, the magnetization is due to bound and free currents. The volume bound
current density is:
jM (r) = ∇ × M(r) (161)
And the surface bound current density is:

KM (rS ) = M(rS ) × n̂(rS ) (162)

The total bound current through any surface is zero:

Z I
I= dS · jM + dl · (KM × n̂ = 0 (163)
S

Inside magnetic media we define free and bound currents:

∇ × B = µ0 (jM + jf ) = µ0 [∇M + jf ] (164)

We define the auxiliary field via:

B(r) = µ0 [H(r) + M(r)] (165)

Ampère’s law in a medium then becomes:

∇ × H = jf (166)

And
∇ · H = −∇ · M = ρ∗ (167)

3.4 Magnetic media

In simple magnetic matter we assume linear and isotropic response from the
magnetization to a field:
M = χm H (168)
Where χm is the magnetic susceptibility. The magnetic field is given by:

B = µ0 (H + M) = µH = µH = κm µ0 H = µ0 (1 + χm )H (169)

Where µ is the permeability and κm is the relative permability. A difference

here with the electric field and the polarization is that here the magnetization
is defined as function of the auxiliary field H instead of B. This is for historical
reasons. For paramagnets χm > 0, for a diamagnetic we have χm < 0 and for a
superconductor we have χm = −1. For simple magnetic media the energy is:
Z
1
UB = d3 rB · H (170)
2

24
4 Electromagnetic forces and conservation laws
4.1 Electromagnetic forces
The total electromagnetic force is the sum of Coulomb force and the Lorentz-
force. If the charged body is a dipole, the total electromagnetic force is given
by:
∂p
F = (p · ∇)E + ×B (171)
∂t
If the induced dipole p is dependent on a time-harmonic field, say p = αE, with
the electric and magnetic field to be proportional to some eiωt , we take the time
average of such a field as follows:

dp∗
 
1 ∗
hFi = Re (p · ∇)E + ×B (172)
2 dt

The work by the electric field E and the magnetic field on charges and currents
respectively is given by:
Z Z
dWmech
= d3 r(ρE + j × B) · v = d3 rj · E (173)
dt V V

4.2 Conservation of energy

The electromagnetic density is given by:
1
uEM = 0 (E · E + c2 B · B) (174)
2
The Poynting vector is the energy current density, and gives the directional flow
of the energy:
1
S= E×B (175)
µ0
We use this to define the energy conservation law:
∂uEM
+ ∇ · S = −j · E (176)
∂t
In simple matter this becomes:
∂uEM
+ ∇ · S = −jf · E (177)
∂t

4.3 Examples
4.3.1 Ohmian dissipation DC-current
Ohm’s law is:
j = σE (178)

25
Inside a wire, the electric field is then:

E = ẑI/πa2 σ (179)

The magnetic field at the surface is:

B = φ̂µ0 I/2πa (180)

The Poynting vector is then:

1 1 1 µ0 I I2
S= E×B= ẑ × φ̂ = − ρ̂ (181)
µ0 µ0 πa2 σ 2πa 2π 2 a3 σ
The resistance of a surface is given by:

R = L/πa2 σ (182)

The energy flow through a surface is then:

Z
L
− dAS · n̂ = I 2 2 = I 2 R (183)
S πa σ

4.3.2 Skin effect and Ohmian dissipation AC-current

We assume a slowly varying electric field. Inside a conductor there’s no free
charge (ρf = 0 and jf = σE. The Maxwell equations inside a conductor are
then:
∇·D = ρ
f ∇·B=0
(184)
∇ × E = − ∂B
∂t ∇ × H = jf + ∂D
∂t

We combine Faraday’s law and the Ampère-Maxwell law to say that:

∂(∇ × B ∂jf ∂E
∇ × (∇ × E) = − = −µ = −µσ (185)
∂t ∂t ∂t
We use the ’double-curl’ equation and ∇ · E = 0:
∂E
∇2 E = ∇(∇ · E) − ∇ × (∇ × E) = µσ (186)
∂t
Then, if the electric field is given by:

E = Re E(z)e−iωt x̂

(187)

The complex amplitude satisfies:

d2 E(z)
= −iµσωE(z) (188)
dz 2
The solution is given by:
E(z) = E0 ei(ki κ)z (189)

26
With: r
µσω 1
k=κ= = (190)
2 δ
Giving an electric field that looks like:

E = E0 e−z/δ cos (ωt − z/δ)x̂ (191)

The term δ here refers to the skin depth, which reflects how deep the electric
field, and with it the current, flows inside the wire. For a frequency of 1015
Hz, which is visible light, this is about 10 nm in copper. For something in the
order of 10 Hz, which is what we use for our house current applications, the
skin depth is about 1 cm. Therefore, you don’t need to make really thick wires,
because the cable won’t carry more current anyway. Instead, you use multiple
thin cables for that purpose.
Now to justify the quasi-static approximation, we look at the current density:

jf = σE0 e−z/δ cos (ωt − z/δ)x̂ (192)

The neglected term in the Ampère-Maxwell term was:

∂D
= −ω E0 e−z/δ sin (ωt − z/δ)x̂ (193)
∂t
This is much smaller than the current density if:
σ 1
ω << ≡ (194)
 τ
That last quantity is the charge relaxation time. That’s much shorter than the
frequency of household electricity.
The magnetic field follows from Faraday’s law:
√
B = 2 cos (ωt − z/δ + π/4) (195)

The time-averaged Poynting vector is:

E02 −2z/δ
hSi = e ẑ (196)
2µδω

27
5 Waves in vacuum
In vacuum, we have no free charge or current or any other sources. Since it’s
√
a vacuum, we’re dealing with the parameters 0 , µ0 and c = 1/ 0 µ0 . The
governing Maxwell equations are:

∇·E=0 ∇·B=0
(197)
∇ × E = − ∂B
∂t ∇ × B = c12 ∂E
∂t

5.1 Wave equation

We take the time derivative of Faraday’s law:
∂ ∂E
∇ × B = c−2 (198)
∂t ∂t
And taking the curl of the the Ampère Maxwell-law:
∂B ∂
∇×∇×E=∇×− = − (∇ × B) (199)
∂t ∂t
We then combine these two to find:
1 ∂2E
∇ × (∇ × E) = − (200)
c2 ∂t2
Working out the left-hand side:

∇ × (∇ × E) = ∇(∇ · E) − ∇2 E (201)

The second term here is zero since we’re in vacuum (cf. the earlier mentioned
Maxwell equations). So we obtain the wave equation:

1 ∂2E
∇2 E − =0 (202)
c2 ∂t2
A similar equation is found for the magnetic field, but with B instead of E.
In Cartesian coordinates, the (x, y, z) components satisfy the same scalar wave
equation. This is not the case for cylindrical and spherical coordinates. We
define our waves then uniquely by choosing a gauge. We first define our fields
in terms of potential functions:
∂A
B=∇×A E = −∇φ − ∂t
(203)

The Lorenz gauge says that:

1 ∂φL
∇ · AL + =0 (204)
c2 ∂t
Both potentials satisfy the same wave equation:
1 ∂ 2 φL
∇2 φ L − =0 (205)
c2 ∂t2

28
 1 ∂ 2 AL
∇ 2 AL − =0 (206)
c2 ∂t2
Simple solutions for the vector potential A can be found, and via the gauge
condition we find the corresponding φL . The solutions for the potentials give
fields that satisfy the Maxwell equations. The fields in terms of the potentials
are:
B=∇×A (207)
∂A
E = −∇φ − (208)
∂t
The Coulomb gauge says similarly that:

∇ · A C = 0 → ∇ 2 φC = 0 (209)

In free space we may take φC = 0, that is, no scalar potential. The vector
potential then has to satisfy two equations:

1 ∂ 2 AC
∇2 A C − =0 (210)
c2 ∂t2
∇ · AC = 0 (211)
The vector field that’s a solution to this gives fields that satisfy the Maxwell
equations.

5.2 Plane waves

The most important class of solutions of the wave equation is the plane wave
solution. All other waves can be construed by a superposition of plane waves.
An arbitrary solution w that satisfies the wave equation consists of two functions
with similar arguments:
1 ∂2w
∇2 w − 2 2 = 0 (212)
c ∂t
w(z, t) = g(z − ct) + f(z + ct) (213)
The first term propagates to the right and the second propagates to the left,
both with a speed c. Here we considered only waves that depend on z, that is,
they propagate in a single direction. The two Maxwell equation said that:
∂Ez
∇·E→ =0
∂z (214)
1 ∂Ez
∇×B= 2 =0
c ∂t
That is, the electric field component Ez is constant in z, t. In other words, since
the wave propagates in the z direction, there’s no longitudinal component of
the electric field, so the electric field consists of two plane waves perpendicular
to ẑ.
Applying the same logic to the magnetic field shows that the magnetic field must

29
also be transverse to the propagation direction. If we generalize the propagation
direction to k̂, we obtain that:

E(r, t) = E⊥ (k · r − ckt)
1 (215)
B(r, t) = k̂ × E(r, t)
c
This last equation is of particular importance and is worth remembering by
heart. From this we come at the conclusion that:

|E| = c|B| (216)

The velocity with which this propagates is:

dr d
vp = = (r0 + ctk̂) = ck̂ (217)
dt dt
We deal now with monochromatic plane waves, which have single angular fre-
quency ω. It’s easier to work with the complex exponential than it is to work
with sines and cosines, so we give our electric and magnetic fields a complex
amplitude, of which the real part yields the physical field. So you add an imag-
inary part which is discarded later on. So our transversal electric and magnetic
field look like:

E(r, t) = Ẽ exp i(k · r − ωt)

(218)
B(r, t) = B̃ exp i(k · r − ωt)

5.3 Wave packets

Monochromatic plane waves are unbounded, and therefore unphysical. We ob-
tain physical waves via a superposition of monochromatic plane waves with
different k vectors:
 Z 
1 3
E(r, t) = Re dk Ẽ(k exp i(k · r − ωt) (219)
(2π)3

This is a wave packet. We can use potentials to represent free space electro-
magnetic fields, meaning we only need scalar wave packets:
Z
1
u(r, t) = dk 3 û(k) exp i(k · r − ω(k)t) (220)
(2π)3

So we sum over all different k-vectors to find our wave packet. Later on, with
waves in dispersive matter, we will sum over frequencies to find a single wave
vector k. Suppose now that we have a scalar wave packet u(r, t) and we want
to know the amplitude weighing factor û(k). Consider it’s shape first at t = 0:
Z
1
u(r, 0) = dk 3 û(k) exp i(k · r) (221)
(2π)3

30
This is an inverse Fourier transformation, so the forward Fourier transformation
yields û(k). Z
u(k) = dk 3 û(r, 0) exp −i(k · r) (222)

This will yield a big wave carrying small waves. The big wave is called the
envelope, and it’s carrying the carrier wave, commonly referred to as the carrier.
We are limited here in our representations by a fundamental limit of wave-
packets. We have that:
∆x∆kx = 2 (223)
This means that wide functions in the space domain yield narrow functions in
the wavenumber domain, and narrow functions in the space domain yield wide
functions in the wavenumber domain. This is a general property of the Fourier
transformation and is called complementarity.

The relation between ω and k is called the dispersion relation. Generally an

intricate function, but in a small domain around k0 , we can use a first-order
Taylor approximation:

∂ω  
ω(k) ≈ ω(k0 ) + (k − k0 )i = ω0 + ∇k ω(k)k=k0 · (k − k0 ) (224)
∂ki k=k0


The group velocity is defined as:


vg = ∇k ω(k)k=k0 (225)

The carrier wave travels with the phase velocity vp = (ω0 /k0 )k̂ = ck̂. The
envelope function travels with group velocity vg . In vacuum, the phase velocity
and the group velocity are the same, but this isn’t necessarily true in matter.

5.3.1 Free-space diffraction

Diffraction is the spreading of waves during propagation. This can be due to
waves encountering objects, or due to a transverse limitation of wave packets.
The latter holds also for a vacuum. The cause of this is the combination of k
vectors with components transverse to k0 , that is, the higher order terms in
the Taylor expansion. These facts cause the wave package to spread out in the
tangential direction. This is free-space diffraction. Complementarity dictates
that free-space diffraction is also more severe for wave packages with smaller
tangential size. Mathematically, this means that when we divide our wave
packet û via the Helmholtz theorem into a tangential and parallel component,
we find propagating waves and evanescent waves. The latter rapidly decay in
tangential direction.
We finally define different basic types of waves. In the Transverse Electric and
magnetic (TEM) mode, both the electric field and the magnetic field (which are
always perpendicular to one another in free space) are transverse to the direction
of travel. In the transverse electric mode (TE) there is no electric field in the

31
direction of propagation. There is only a magnetic field along the direction of
propagation. In the transverse magnetic mode (TM) there is no magnetic field
in the direction of propagation. These are sometimes called E modes because
there’s only an electric field along the direction of propagation.

32
6 Waves in simple matter I
We assume here monochromatic waves and simple matter where we have con-
stant  and σ. Note that having σ = 0 implies no frequency dispersion. The
Maxwell equations in matter are:

∇ · D = ρf ∇·B=0
(226)
∇ · E = − ∂B
∂t ∇ · H = jf + ∂D
∂t

The constitutive equations are:

D = E
(227)
B = µH

With definitions:
p √
n= µ/µ0 0 = c µr r
p (228)
Z = µ/

We choose the electric field E and magnetic auxiliary field H as the fundamental
field quantitities. The reasons for this are:
1. Polarization and magnetization are expressed in terms of E and H:

P = ( − 0 )E
(229)
M = (µ/µ0 − 1)H

2. Matching conditions (i.e., boundary conditions) of E and H at interfaces

are similar.

3. Poynting in vector in matter is most easily written in terms of E and H:

S=E·H (230)

At an interface between two media, we have that:

n̂ · [D1 − D2 ] = 0 n̂ · [B1 − B2 ] = 0
(231)
n̂ × [E1 − E2 ] = 0 n̂ × [H1 − H2 ] = 0

Where the surface normal vector n̂ points from the surface ’inwards’, i.e., into
medium 2. These boundary conditions tell you that the components of E and
H parallel to the interface are continuous, and that the components of D and
B perpendicular to the interface are continuous. Furthermore, we assume that
ρf = 0 and that jf = 0. The Maxwell equations in a simple medium without
sources then becomes:
∇·E=0 ∇·B=0
(232)
∇ · E = − ∂H
∂t ∇ · H =  ∂E
∂t

33
6.1 Plane waves in matter
We consider monochromatic plane waves:

E(r, t) = Eei(k·r−ωt)
(233)
H(r, t) = Hei(k·r−ωt)

We actually mean here that you should take the real part of these electric
field for the actual physical fields. However, since we don’t take the real part
except at the end of an equation (which may or may not return a real quantity
anyway), we just let the electric and magnetic field be represented by these
complex quantities. Substituting these equations into the Maxwell equations
gives:
k·E=0 k·H=0
(234)
k × E = ωµH k × H = −ωE
The wave impedance is given by:

Z = E/H (235)

6.1.1 Dispersion
A dispersion relation gives us the link between ω and k. For simple media with
σ = 0 we find that:

k × (k × E) = ωµk × H = ωµ · ωE = −ω 2 µE (236)

Because:
k × (k × E) = k(E · k) − E(k · k) (237)
The first term we obtain is zero because the electric field is transversal, so it’s
perpendicular to the wave vector k. We are now left with:

k × (k × E) = −E(k · k) (238)

But it’s also true that:

r
2 1 c
k · k = µω → ω(k) = k= k (239)
µ n
So we obtain that:
k × (k × E) = −ω 2 µE (240)
The resulting phase velocity for a wave propagating in the k̂ direction is:
ω c
vp = k̂ = k̂ (241)
k n
Same goes for other quantities. In a medium we see that the speed of the wave
c in vacuum is replaced by c/n.

34
6.1.2 Energy balance
If  or µ have an imaginary part, there will be losses. The simplest formulation
is  = 0 + i00 . There’s no free current, so the Poynting theorem says:
Z 0 Z Z
d  µ0
− |E|2 + |H|2 d3 r = (E · H) · dS + ω00 |E|2 d3 r (242)
dt V 2 2 S V

The first integral is the stored energy in the medium, the second the energy
outflow rate and the third the energy loss rate.

6.2 Reflection and transmission at a plane boundary

At a sharp flat boundary at z = 0 there’s a jump in  and µ. We can distinguish
three waves:
1. Incident wave
2. Reflected wave
3. Transmitted or refracted wave
The problem now is to find the reflected and transmitted wave, given the incident
wave and the material properties.

6.2.1 Basic approach

You solve a reflection/transmission problem by following these steps:
1. Make an assumption (’ansatz’) about the reflected and transmitted waves
2. Determine the total fields for z < 0 and z > 0
3. Write down the matching conditions (boundary conditions) for the field
components at the boundary
4. Apply the boundary conditions to find:
• Direction of the emerging waves
• Relations between the amplitudes of the waves
5. Solve the amplitude relations to find the reflection and transmission coef-
ficients.

6.2.2 Snell’s law

The wave number k = |k| in both media is known, so:
ω
kI = kR = k1 = n1
c (243)
ω
kT = k2 = n2
c

35
We can also always orient the coordinate axes such that:

kIy = kRy = kT y = 0 (244)

Since kIx = kRx = kT x we find Snell’s laws:

k1 sin θI = kI sin θR → θI = θR
(245)
k1 sin θI = k2 sin θT → n1 sin θI = n2 sin θT

These are a direct consequence of the continuity of the tangential components

of E and H.
For an arbitrary polarization of the incident wave it’s very difficult to find
relations between amplitudes of the electric field and amplitudes of the magnetic
field. To make things simpler, we decompose the total field in two standard
polarization states: ’p’or TM polarization, and ’s’ or TE polarization. In TM
polarization the magnetic field is parallel to the surface (but perpendicular to
the plane of incidence), and in TE polarization the electric field is parallel to
the surface (but perpendicular to the plane of incidence).

6.2.3 Fresnel equations

The Fresnel reflection and transmission coefficents for p-polarization are:
 
ER Z1 cos θ1 − Z2 cos θ2
rp = =
EI p Z1 cos θ1 + Z2 cos θ2
  (246)
ET 2Z2 cos θ1
rp = =
EI p Z1 cos θ1 + Z2 cos θ2

For s-polarization we find:

 
ER Z2 cos θ1 − Z1 cos θ2
rs = =
EI s Z2 cos θ1 + Z1 cos θ2
  (247)
ET 2Z2 cos θ1
rs = =
EI s Z2 cos θ1 + Z1 cos θ2

6.2.4 Polarization by reflection

Take an incident wave with both p and s polarization. When rp or rs becomes
zero, the reflected wave has the other polarization, so you obtain polarization by
reflection. The physical origin of this lies with the polarization of the medium
P and magnetization M, which produce electromagnetic fields that interfere
destructively with the incident wave in medium 2 and produce the right reflected
wave in medium 1 and the right transmitted wave in medium 2. In the case of
no reflection, we obtain that:

rp = 0 → Z1 cos θ1 = Z2 cos θ2 (248)

36
Combined with Snell’s law

n1 sin θ1 = n2 sin θ2 (249)

this gives:
2 2 µ1 − 1 µ2
tan2 θB = → θB = θ1 (250)
1 2 µ2 − 1 µ1
For non-magnetic media (µ1 = µ2 = µ0 ) we find:
2 n2
tan2 θB = → θB = tan−1 (251)
1 n1
This is the Brewster angle (hence the B in the subscript). The reflected with is
then entirely s-polarized, the p-polarization has been killed.

In the case of no s-reflection, we have that:

rs = 0 → Z1 cos θ2 = Z2 cos θ1 (252)

Combining this with Snell’s law gives:

µ2 1 µ2 − 2 µ1
tan2 θE = (253)
µ1 2 µ2 − 1 µ1
For non magnetic media we find:

tan2 θB = −1 → θE doesn’t exist (254)

The reflected wave can never be entirely p-polarized.

6.2.5 Total internal reflection

Consider two media with n1 > n2 , so θ2 > θ1 . Then θ2 = π/2 for θ1 = θC =
sin−1 (n2 /n1 ), the so called critical angle. We now ask: what happens for an
angle θ1 > θC ? Assuming no propagation in the y-direction, we obtain that:
ω
k2x = k1x = k1 sin θ1 = n1 sin θ1 (255)
c
So:
ω
2
k2z = k22 −k2x
2
= ( )2 (n22 −n21 sin2 θ1 ) = (ωn1 /c)2 (sin2 θC −sin2 θ1 ) < 0 for θ1 > θC
c
(256)
2
Because k2z < 0 for θ1 > θC , we find that k2z will be imaginary:
ω
q
k2z = iκ = i n1 sin2 θ1 − sin2 θC (257)
c
The transmitted electric field is then:

ET (r, t) = ET exp (−κz) exp (i(k1 sin θ1 x − ωt) (258)

37
The first term is a decay in the z-direction, while the complex ecponential
shows propagation in the x-direction. This is an interfacial wave because it
only exists near the interface. It has a phase speed vp1 =< vp < vp2 and has
a longitudinal component. This wave is non-uniform and evanescent. Non-
uniform plane waves do not have the same amplitude in planes perpendicular
to the propagation direction, but decrease exponentially in some direction.

6.2.6 Evanescent waves

The general form of an evanescent wave is:

E(r, t) = E exp i(q · r − ωt) exp (−κ · r) (259)

The first term shows propagation in direction q and a decay in κ. The wave
vector is k = q + iκ. The dispersion equation is:

k · k = µω 2 → q 2 − κ2 = µω 2 and q · κ = 0 (260)

Take care: E · k = 0 etcetera no longer implies geometrical orthogonality!

38
7 Waves in simple matter II
7.1 Radiation pressure
Objects reflect or absorb an electromagnetic wave, transferring linear momen-
tum to from the wave to the object. This causes radiation pressure. The
mechanical force density is given by:
∂gEM
f=− +∇·T (261)
∂t
Where gEM = D × B is the field momentum density and T the electromagnetic
stress tensor.
The total force on a volume V and surface S is the integral:
Z Z
3 ∂gEM
F=− d r + dS n̂ · T (262)
V ∂t S

We asume a normally incident, time harmonic, plane wave, so the average force
on an interface is: Z
hFj i = dAnk hTkj i (263)
S
Since we’re usually dealing with infinite surfaces, the force would be infinitely
large if we’d integrate over the entire surface. We therefore much rather prefer
to talk about the radiation pressure:
Fz 2I0 n − 1
hPrad h= = I0 /c(1 + R − nT ) = −2 (264)
A c n+1

7.2 Layered matter

In layered matter, we assume the media has two or more parallel interfaces,
and that each layer is homogeneous. The first and last ’layer’ are homogeneous
halfspaces.

7.2.1 Fabry-Perot geometry

In a Fabry-Perot geometry, lights hits an interface. A part is transmitted and
gets t in front of it, part is reflected and gets r in front of it. The refracted light
then travels through the medium, until it hits the next interface. There, it can
reflect, or it be transmitted. The transmission coefficient here is not t, but since
we’re dealing medium 2 to 1, it’s now t0 . The wave that is then reflected will
first travel down back to the first interface, where it can reflect/transmit again,
and then, with now some additional phase with respect to the first transmitted
wave, depending on the thickness of the material. This will continue but finally
converge to:
tt0
ET = EI (tt0 +tr0 r0 t0 exp (i∆φ)+t(r0 r0 )2 t0 exp (2i∆φ)+...) = EI = TEI
1− (r0 )2exp (i∆φ)
(265)

39
The phase difference between successive transmissions is given by:

∆φ = 2kl − k0 a (266)

Where the thickness of the material is d and l = cosd θ and a = s sin θ0 =

2d tan θ sin θ0 , with θ0 the angle of incidence on the medium and θ the angle
with which the wave refracted into the wave. Using all this, the phase difference
becomes:
2nωd 2n0 ωd sin θ 2nωd cos θ
∆φ = 2kl − k0 a = − sin θ0 = (267)
c cos θ c cos θ c

7.2.2 Stokes Relations

The reflection into a wave is given by:

1=t+r (268)

Its time reverse version is:

t=1−r (269)
Which we decompose into:

t = 1 − r → (t + tr0 = tt0 ) + (r + rr = rt) (270)

Now we recognize the Stokes relations:

r2 + tt0 = 1 tr0 + rt = 0 (271)

Now we come up with Airy’s formula. The Stokes relations imply that:

tt0 = 1 − r2 = 1 − R ; r = −r0 (272)

Now we write the energy transmission factor as:

 −1
4R δφ
2
|T| = 1 + sin2 (273)
(1 − R)2 2

This is Airy’s formula. We see that transmission for δφ/2 = mπ is independent

of R.

7.2.3 Matrix formalisms

In a multilayer structure you have many waves, so we use a more systemic
approach. We have two options:
• Transfer matrix approach
• Scattering matrix approach

40
We assume N + 1 layers with index of refraction nj , wave impedance Zj and
thickness dj . We assume waves propagating normal to the interfaces:
−
Ej = E+
j (z) + Ej (z) = Ej (z)x̂ (274)
−
Hj = H+
j (z) + Hj (z) = Hj (z)x̂ (275)
We define wave amplitude of waves in layer j at boundary j. At boundary j we
get:
Ej = Ej+ + Ej− (276)
Hj = Hj+ + Hj− = Ej+ /Zj − Ej− /Zj (277)
The phase change when passing over layer j is:
ω
φj = kj dj = nj dj (278)
c
This gives us the transfer matrix of a single matrix:
    
Ej−1 cos φj −iZj sin φj Ej
= (279)
Hj−1 −iZj sin φj cos φj Hj
Doing so for repeated layer means taking the product of the incoming wave with
a matrix for each layer. This gives the set of equations:
 −1 
 NY  
1+R cos φj −iZj sin φj T
= (280)
(1 − R)/Z0 −iZj sin φj cos φj T/ZN
j=1

Where E0 = 1 + R, EN = T and H0 = (1 − R)/Z0 and HN = T/ZN . This

approach can be extended to waves that are not normally incident and to media
with attenuation. However, for media with attenuation, the matrices become
numerically unstable and produce unreliable results. Therefore, it’s better to
use the scattering matrix formalism.

7.3 Conducting matter

We assume no free charge, and a medium governed by , µ, σ. The Maxwell
equations in this context are:
∇·E=0 ∇·H=0
(281)
∇ × E = −µ ∂H
∂t ∇ × H = σE +  ∂E
∂t

The wave equation is:

∂2
 
∂ E
∇2 − µσ − µ 2 =0 (282)
∂t ∂t H
The rate of electromagnetic energy lsos is given by:
Z
dW
=σ |E|2 d3 r (283)
dt V

41
7.3.1 Monochromatic plane waves
We again assume plane waves, and now substitution into the Maxwell equations
give:
k·E=0 k·H=0
(284)
k × E = ωµH k × H = −ω( + iσ/ω)E
Where we can also say that:

 + iσ/ω = ˆ(ω) = 0 + i00 (285)

The dispersion relation is now:

ω2
k · k = ˆ(ω)µω 2 = n̂2 (ω) (286)
c2
The wave vector is k + q + iκ. The dispersion relationship is then:

q 2 − κ2 = µω 2 (287)

And
2q · κ = σµω (288)
The book actually has a wrong equation for this new dispersion relation. We
assume that q and κ are parallel (so same k̂), then we obtain that:

k = n̂ ωc k̂ k̂ · E ẐH = k̂ × E (289)

Where in √the last equation we assume transverse waves. Recall that the impedance
Ẑ here is µˆ. Writing out the electric field now gives an ordinary time-varying
complex part and a real decaying exponential in the direction k̂:
ω ω
E(r, t) = E exp (i(n0 k̂ · r − ωt)) exp (−n00 k̂ · r (290)
c c
Equations for n0 and n00 from dispersion equation are:

n02 − n002 = µc2 (291)

And
2n0 n00 = σµc2 ω (292)
Again, the dispersion relation here is correct, and the one in the book is wrong.

7.3.2 Skin depth

For a good conductor we have that σ/ω >> 1, so n0 ≈ n00 , and you find that:
r
µσ 1 c
n0 ≈ n00 ≈ c = >> 1 (293)
2ω δ(ω) ω

42
Or r
2
δ(ω) = << 1 (294)
µσω
This is the skin depth, showing that the electric field doesn’t penetrate deeply
into a conductor. The impedance is given by:
r r r
µ µω µω
Ẑ = ≈ = exp (−iπ/4) (295)
 + iσ/ω iσ σ

The square root is very small, and the complex exponential indicates that the
magnetic field lags behind the electric field by 45 degree. Now, using the fact
that n02 >> n1 , and putting that into Snell’s law, we see that the refractive angle
must be approximately zero. This is why in high voltage lines, they actually
use multiple lines rather than one thick cable - the current doesn’t penetrate
deeply into the material anyway, so making the wire thicker doesn’t help.

7.3.3 Reflection from good conductor

Just use Fresnel expressions with complex impedance:
p
Ẑ = µ/ˆ  = µc/n̂ (296)

43
8 Waves in dispersive matter
There are two types of dispersion: frequency or chromatic dispersion, and modal
or wavegudie dispersion. Chromatic dispersion is the Pink Floyd style white
light in a prism comes out as a rainbow dispersion, whereas modal dispersion
means that one color travels faster through a medium than the other, meaning
it typically undergoes less reflection (since the index of refraction is lower, the
’curving’ is less, so it has shorter optial path length through a medium.

8.1 Frequency dispersion

In simple conductive matter, the electric permittivity of a medium is dependent
on the frequency and is a complex number:
σ
ˆ(ω) =  + i (297)
ω
Previously we said that wave propagation was dependent on frequency, though
the electric permittivity , the magnetic permeability µ and the conductivity σ
were not. In dispersive matter, , µ, σ are frequency dependent. The reason for
this is that polarization, magnetization and conduction cannot respond imme-
diately on external changes of the electromagnetic field. Moreover, the response
of a medium cannot depend on future moments, meaning these quantities are
causal. Since these properties cannot respond immediately, they respond more
readily to slowly varying quantities than to fast varying quantities. This be-
haviour what we aim to capture. We describe it using only media which have a
linear response.

8.2 Response of a medium

8.2.1 Causal response of medium: time domain
An example of causality: the current density due to an electric field. This is
simply Ohm’s law: Z t
j(r, t) = σ(t − t0 )E(r, t0 )dt (298)
∞
’ Since the current density j cannot depend on fields E in the future, we set:

σ(τ ) = 0 (299)

for τ < 0. This allows us to describe j with the convolution integral:

Z ∞
j(r, t) = σ(t − t0 )E(r, t0 )dt0 (300)
−∞

44
8.2.2 Response of a medium: frequency domain
We just described the response of a medium by looking at its time behaviour.
However, we can look at the frequency response by using the Fourier transform.
The Fourier transform of a function f is given by:
Z ∞
fˆ(ω) = f (t) exp (iωt)dt (301)
−∞

The inverse Fourier transform is:

Z ∞
1
f (t) = fˆ(ω) exp (−iωt)dω (302)
2π −∞

A Fourier transformation turns a convolution into a multiplication, so our cur-

rent density becomes: convolution integral:
Z ∞
j(r, t) = σ(t − t0 )E(r, t0 )dt0 → ĵ(r, ω) = σ̂(ω)Ê(r, ω) (303)
−∞

The conductivity in the Fourier domain σ̂(ω) is here written as a complex quan-
tity because it usually is. However, in the time domain the conductivity is a
real quantity. This has interesting consequences for the conductivity in Fourier-
domain. So in the time domain we have a real quantity:

σ(t) = σ ∗ (t) (304)

So in the Fourier domain this becomes:

Z ∞
1
σ(t) = σ̂(ω) exp (−iωt)dω
2π −∞
Z ∞
1
= σ̂ ∗ (ω) exp (−iωt)dω (305)
2π −∞
Z ∞
1
= σ̂ ∗ (−ω) exp (−iωt)dω
2π −∞

We have that σ̂(ω) = σ̂ ∗ (−ω). The real part then is an even quantity:

σ 0 (ω) = σ 0 (−ω) (306)

The imaginary part is odd:

σ 00 (ω) = −σ 00 (−ω) (307)

8.3 Equivalence of descriptions

Up to now, we described charges as either bound or free. Free charges cause
conduction, whereas bound charges are responsible for polarization and magne-
tization. For static fields, it’s easy to distinguish the long-distance displacement

45
of ’free’ charge from the short-distance displacement of ’polarization’ charge.
This distinction becomes blurred when E(r, t) is time-harmonic because charge
oscillates back and forth in both cases. At high frequency, particularly, there’s
simply no way to distinguish a time-harmonic conduction current with density
j = σE from a time harmonic polarization current with density j = ∂P/∂t.
So we say that the medium response is always due to the localized motion of
charge, bound or free. So we previously said that:

P̂(r, ω) = 0 χ̂(ω)Ê(r, ω) (308)

A bound charge (polarization) current is then given by:

∂P
j= (309)
∂t
But it’s now indistinguishable from the free charge current, given by:

j = σE (310)

So we can equate these two. The conductivity of a medium is then described

by:
σ̂(ω)Ê = −iω P̂ = −ω0 χ̂e (ω)Ê (311)
So we now have a relation between σ̂(ω) and χ̂(ω). Then, because:

D̂(r, ω) = 0 Ê(r, ω) + P̂(r, ω) = ˆ(ω)Ê(r, ω) (312)

We can expand the polarization in terms of the electric field, so this becomes:

D̂(r, ω) = 0 Ê(r, ω) + P̂(r, ω)

= 0 (1 + χ̂e (ω))Ê (313)
 
σ̂(ω)
= 0 + i Ê = ˆ(ω)Ê
ω
It’s a matter of taste whether one uses a dielectric function description or a
conductivity description for time-dependent electromagnetic problems in mat-
ter. Polarization and magnetization is similarly ambiguous. Polarization and
magnetization both produce currents, and they’re indistinguishable:
∂P
j= +∇×M (314)
∂t
Substitute this in the equation for vacuum:
∂E
∇ × B = µ0 j + 0 µ0 (315)
∂t
For a linear medium we have that:
 
1 + χm
B = µ0 M (316)
χm

46
 1
E= P (317)
0 χ e
We collect the terms to obtain that:
1   dP
∇ × M = 1 χ1 (318)
χm e dt
We see that we can write the medium reaction j entirely in terms of polarization
or magnetization.

8.4 Classical models

The properties of the electromagnetic medium are determined by the way the
fields interact with charged particles in the medium. We consider therefore two
classical (i.e., non-quantum) models:
1. Drude model
2. Lorentz model
We assume here that the particles have low speed (non-relativistic) and there’s
no strong external magnetic field, meaning we don’t have to consider the mag-
netic force.

8.4.1 Drude model

The basic idea behind the drude model is to view a medium as free charge
carriers with a mean time τ between collisions (which cause momentum loss).
The equation of motion here is:
dv mv
m = qE − (319)
dt τ
In time-harmonic cases we have:

v(t) = v̂ exp (−iωt) (320)

So the equation of motion becomes:

mv̂(ω)
− iωmv = qE − (321)
τ
Rewriting this a little bit, we get:

(q/m)Ê(ω)
v̂(ω) = (322)
1/τ − iω
A current density is simply the charge of each individual current, times the
number of charges, times the velocity. It’s no different in the Fourier-domain:

nq 2 τ Ê(ω) σ0
ˆ(ω) = nqv̂(ω) = = σ̂(ω)Ê(ω) → σ̂(ω) = (323)
m 1 − iωτ 1 − iωτ

47
Where we identify the following constant:
nq 2 τ
σ0 = (324)
m
We substitute this into our equation for the electric permittivity:
σ̂(ω)
ˆ(ω) = 0 + i (325)
ω
This gives the Drude dielectric function:
! !
ˆ(ω) ωp2 τ 2 ωp2 τ 1
= 1− +i (326)
0 1 + ω2 τ 2 ω 1 + ω2 τ 2
Where we identify the plasma frequency:
nq 2
ωp2 = (327)
0 m
For metals, we have 1/τ  ωp . In the high frequency limit, we have ωτ  1,
so Drude dielectric function:
ˆ(ω) ωp2
≈1− 2 (328)
0 ω

8.4.2 Lorentz Model

The basic idea behind the Lorentz model is an electron bound to a fixed nucleus
by a spring with damping. The equation of motion is:
d2 r dr
m + mΓ + mω02 r = −eE (329)
dt2 dt
Where mΓ is the damping coefficient and mω0 is a spring constant. For a
time-harmonic case, the solution to this equation of motion is:
−e/m
r̂(ω) = Ê(ω) (330)
ω02 − iωΓ − ω 2
This has a following corresponding polarization:
ne2 /m
P̂(ω) = np̂(ω) = −enr̂(ω) = Ê(ω) (331)
ω02 − iωΓ − ω 2
Again, we substitute this into our equation for the electric permittivity:
σ̂(ω)
ˆ(ω) = 0 + i (332)
ω
So the Drude dielectric function is:
ˆ(ω) ωp2
=1+ 2 (333)
0 ω0 − ω 2 − iωΓ
We see that there’s a resonance peak in this function. However, there can be
more of these proccesses working in parallel (so multiple damping coefficients
and spring constants), so in practice you have more resonance peaks.

48
8.5 Wave packets
We now describe groups of wave as wave packets. In a vacuum, a wave packet
is described as follows:
Z
1
u(r, t) = dk 3 û(k exp{i[k · r − iω(k)t]} (334)
(2π)3
That is, a wave packet of plane waves with different wave vectors. These spread
out transversally because not all wave vectors are in the same direction. This
is how we previously described them. Now, in dispersive matter, we have that:
Z ∞
E(z, t) = dωA(ω) exp{i[k(ω)z − ω(k)t]} (335)
0

That is, a sum of plane waves with different frequencies. These spread out
longitudinally because not all phase speeds are the same. If we only need k(ω)
in a small domain around ω0 , we can use a Taylor expansion for a simplified
description:
∂k  1 ∂ 2 k 
k(ω) = k(ω0 ) + Ω + Ω2 + ... (336)
∂ω ω=ω0 2 ∂ω 2 ω=ω0
The zeroeth order term is k0 , the first order term is k00 , the second order term
is k000 and so on. Substitution of this into our wave packet equation yields:
Z
1
E(z, t) = exp i[k0 z − ω0 ] dΩA(ω0 + Ω) exp i[Ωk00 + Ω2 k000 + ...)z − iΩt]
2
(337)
We see a carrier wave, given by:
exp i[k0 z − ω0 ] (338)
And an envelope function A(z, t), given by:
Z
1
A(z, t) = dΩA(ω0 + Ω) exp i[Ωk00 + Ω2 k000 + ...)z − iΩt] (339)
2
The envelope function satisfies the following differential equation:
1 ∂ 2 A(z, t)
 
∂ 0 ∂
i + k0 A(z, t) = k000 (340)
∂z ∂t 2 ∂t2
The speed at which the envelope function travels is called the group velocity.
The group velocity enters our analysis because:
 −1
∂ω  ∂ω  ∂ω  1
vg = → vg = → = (341)
∂k k=k0 ∂k ω=ω0 ∂k k=k0 vg
If we neglect the second-order Taylor term, we get:
 
∂ 1 ∂
+ A(z, t) = 0 (342)
∂z vg ∂t

49
Which means that A(z, t) = Ã(z − vg t). We put this in front of our wave and
obtain:
E(z, t) = exp i[k0 z − ω0 ]Ã(z − vg t) (343)
Now we can have two cases. If n(ω) is constant, we have vg = vp , i.e., the group
velocity is the same as the phase velocity. If n(ω) is not constant, we have that:
 
∂ω ∂ ck c
vg = = = dn
< vp (344)
∂k ∂k n n + ω dω

For normal dispersion.

8.5.1 Group velocity dispersion

If the second and higher order Taylor terms may not be neglected, the envelope
function will change its shape during propagation. This is group velocity disper-
sion. The local frequency of the wave varies as a function of position (or time).
The components with longer wavelengths appear at the front of a pulse, while
the shorter wavelengths appear at the back of the pulse. As the pulse passes
by, you get the low frequencies first, and it quickly rises up - a phenomenon
described as ’chirp’.

8.6 Causality and consequences

All function that respond to a stimulus should be causal. The response function
is given by χ(t). The mathematical consequences of this are found for this are
seen in the Fourier-domain χ̂(ω). The real part χ̂(ω)0 and imaginary part χ̂(ω)
are directly related. Also, note that χ̂(ω) is analytic if Im[ω] > 0. The physical
consequence of this is that frequency dispersion and dissipation are intrinsically
linked. Furthermore, no electromagnetic signal can travel faster than the speed
of light.

8.6.1 Basic equation of causality

Causality requires that:
χ(t) = 0 (345)
For t < 0. We should therefore have that:
1
χ(t) = H(t)χ(t) = (1 + sgn(t))χ(t) (346)
2
Where H(t) is the Heaviside step function and sgn(t) is the sign function. The
Heaviside function is 0 for t < 0 and 1 for t > 0. The sign-function is −1 for
and +1 for the same time domains respectively.

50
8.6.2 Kramers-Kronig relations
Because of causality, we can relate the real and imaginary part of the response
function χ̂0 (ω) and χ̂00 (ω). Recall the Fourier transform pair:
Z ∞
χ̂(ω) = χ(t) exp (iωt)dt (347)
−∞

The inverse Fourier transform is:

Z ∞
1
χ(t) = χ̂(ω) exp (−iωt)dω (348)
2π −∞

We use χ(t) = 21 [1sgn (t)]χ(t) in the forward Fourier transform. This gives us:

1 ∞ 1 ∞
Z Z
χ̂(ω) = χ(t) exp (iωt)dt + sgn(t)χ(t) exp (iωt)dt (349)
2 −∞ 2 −∞

The first term will be simply the Fourier transform of the function:

1 ∞
Z
1
χ̂(ω) = χ̂(ω) + sgn(t)χ(t) exp (iωt)dt (350)
2 2 −∞

We subtract half the Fourier transform:

1 ∞
Z
1
χ̂(ω) − χ̂(ω) = sgn(t)χ(t) exp (iωt)dt (351)
2 2 −∞

1 ∞
Z
1
χ̂(ω) = sgn(t)χ(t) exp (iωt)dt (352)
2 2 −∞
Z ∞
χ̂(ω) = sgn(t)χ(t) exp (iωt)dt (353)
−∞

We describe the response function χ(t) via the inverse Fourier transform, using
the dummy variable s:
Z ∞ Z ∞ 
1
χ̂(ω) = χ̂(s) exp (−ist)ds sgn(t) exp (iωt)dt (354)
−∞ 2π −∞

51
9 Retardation and Radiation
We start with the following notions:
• Retardation
• Radiation
• Source
Retardation means that the observer at time t sees the field that is emitted by
the source at time t − R/c. Radiation means that fields can propagate away
from their source, carrying energy (which can travel to infinity). A source is
understood to be a prescribed distribution of electric charge and/or electric
current that causes an electromagnetic field.

9.1 Difference with previous lectures

In previous lectures, waves were generated outside the domain of interest. There
was no prescribed charge and current density, which gave us homogeneous wave
equations. Now the waves will be generated by sources, which will lead to
prescribed charge and current density. This will give us inhomogeneous wave
equations.

9.2 Inhomogeneous wave equations

The four homogeneous Maxwell equations were:

∇·E=0 ∇·B=0
(355)
∇ × E = − ∂B
∂t ∇ × B = + c12 ∂E
∂t

The four inhomogeneous Maxwell equations we will use now are as follows:

∇ · E = ρ0 ∇·B=0
(356)
∇ × E = − ∂B∂t ∇ × B = µ0 j + c12 ∂E
∂t

To derive the wave equations, we take the same path. We take the curl of one
curl equation, and subtract the time derivative of another curl equation. The
result is the following two inhomogeneous wave equations:
1 ∂2E 1 ∂j 1 ∂2B
∇2 E − c2 ∂t2 = 0 ∇ρ + mu0 ∂t ∇2 B − c2 ∂t2 = −µ0 ∇ × j (357)

These wave equations were previously homogeneous. Now they are inhomoge-
neous. These wave equations aren’t suited for our purposes for two reasons:
1. Inhomogeneous wave equations are harder to solve
2. Solutions need not satisfy original Maxwell equations
The use of potentials will solve this.

52
9.3 Potentials of inhomogeneous wave equations
We demand that ∇ · B = 0. This happens automatically if we take B = ∇ × A,
with A the vector potential. Next, this implies that:
∂B ∂A
∇×E=− = −∇ × (358)
∂t ∂t
Therefore, we always want to satisfy that:
 
∂A
∇× E+ =0 (359)
∂t

This happens automatically if we take

∂A
E+ = −∇φ (360)
∂t
With φ being the scalar potential. Unfortunately, the equations for φ and A we
obtain this way, are not nice.
∂ ρ
∇2 φ + (∇ · A) = −
∂t 0
(361)
1 ∂2A
 
1 ∂φ
∇2 A − 2 2 − ∇ ∇ · A + 2 = −µ0 j
c ∂t c ∂t

9.4 Gauge conditions

The equations for the potentials aren’t nice, but we fortunately do have some
freedom. Nothing will change if we do the following:
1. Add gradient of an arbitrary function Λ to A

2. Subtract time derivative of Λ from φ

We don’t have to define the gauge function Λ explicitly.

9.5 Coulomb gauge

We can choose:
∇·A=0 (362)
This is the Coulomb gauge. This gives:
ρ
∇2 φ = − (363)
0
This is Poisson equation. We also obtain:

1 ∂2A
 
2 1 ∂φ
∇ A− 2 = −µ0 j − ∇ (364)
c ∂t c2 ∂t

53
9.6 Lorenz gauge
Another choice is to take:
1 ∂φ
∇·A+ =0 (365)
c2 ∂t
This is the Lorenz gauge. This gives:
1 ∂φ ρ
∇2 φ − =−
c2 ∂t2 0
(366)
1 ∂A
∇2 A − 2 2 = −µ0 j
c ∂t
This is what we will use to solve radiation problems.

9.7 Retardation
There are two ways to find the retarded potentials. The first way is that of the
book, which is very complicated. We will follow a simpler derivation with the
same result. We have to solve scalar wave equations of the form:

1 ∂2
 
∇2 − 2 2 ψ(r, t) = −f (r, t) (367)
c ∂

We define the Green function by:

1 ∂2
 
∇2 − 2 2 G(r, t|r0 , t0 ) = −δ(r − r0 )δ(t − t0 ) (368)
c ∂

Here r and t are the observer coordinates, and r0 and t0 are the source
coordinates. There are two solutions for the Green’s function:
1
G± (r, t|r0 , t0 ) = δ(t − t0 ± |r − r0 /c) (369)
4π|r − r0 |

That is, there’s an outward travelling wave and in inward travelling wave. The
latter is unphysical and therefore discarded. The physical interpretation of G−
is to imagine a wave emitted at time t0 by the source at r0 , which reaches
the observer at position r at time t = t0 + |r − r0 /c. This wave is retarded
with respect to the source, i.e., it lags behind what is happening now. The
physical interpretation of G+ is a wave absorbed at time t0 reaches observer at
t = t0 − |r − r0 |/c. This is advanced with respect to the source.

9.8 Advanced and retarded waves

Assume we know G. The source term may be written as:
Z t2 Z
0
f (r, t) = dt d3 r0 δ(r − r0 )δ(t − t0 )f (r0 , t0 ) (370)
t1 V

54
This is a weighted sum of delta functions. The source is entirely contained by
the volume V and is only active for the time t1 ≤ t0 ≤ t2 . Because the wave
equation is linear, we may apply superposition:
Z t2 Z
ψ(r, t) = dt0 d3 r0 G(r − r0 , t − t0 )f (r0 , t0 ) (371)
t1 V

This is a weighted sum of Green’s functions. There are two solutions for ψ. The
retarded solution and the advanced solution.

9.9 Retarded solution

Z t2 Z
ψret (r, t) = dt0 d3 r0 G− (r, t|r0 , t0 )f (r0 , t0 ) + ψin (r, t)
t1 V
0
(372)
, t − |r − r0 |/c)
Z
3 0 f (r
= d r + ψin (r, t)
v |r − r0 |
Where ψin (r, t) is a solution of the homogeneous differential equation of ψ
and depends on the boundary condition at t1 . If f (r, t) = 0 for t < t1 then
ψret (r, t) = 0 for t < t1 (that is, we have causality), then ψin (r, t) = 0.

9.10 Advanced solution

Z t2 Z
0
ψadv (r, t) = dt d3 r0 G+ (r, t|r0 , t0 )f (r0 , t0 ) + ψout (r, t)
t1 V
0
(373)
, t + |r − r0 |/c)
Z
3 0 f (r
= d r + ψout (r, t)
V |r − r0 |
Where ψout (r, t) is a solution of the homogeneous differential equation of ψ
and depends on the boundary condition at t1 . If f (r, t) = 0 for t > t2 then
ψadv (r, t) = 0 for t > t2 (that is, we have anti-causality), then ψout (r, t) = 0.

9.11 Physical relevance of the two solutions

In a bounded domain, both the retarded and the advanced solution are phys-
ically relevant. In an infinite domain, only the retarded solution is physically
relevant. An example where both are relevant is a layered sphere: a source in-
side a layered sphere will emit, but each wave will also be partially be reflected
from each wall. Outside this sphere, the advanced solution is no longer relevant,
only the retarded solution is.

9.12 Retarded harmonic waves

We can describe the retarded waves as retarded harmonic waves. We introduce
the complex notation:

f (r, t) = f (r|ω) exp (−iωt) (374)

55
This directly yields the solution:
exp (ik|r − r0 |)
Z
Ψ(r|ω) = dr3 0|
f (r0 |ω) (375)
V 4π|r − r
You obtain the same result via the Fourier domain form:
Z
Ψ(r|ω) = dr3 G(r − r0 |ω)f (r0 |ω) (376)
V

The Fourier transform of the retarded Green’s function is:

exp(ikr)
G− (r, ω) = (377)
4πr
The physical G that you obtain should satisfy the Sommerfeld radiation condi-
tion:  
∂
lim r − ik G = 0 (378)
r→∞ ∂r

9.13 Time-dependent electric dipole

We apply this to a time dependent electric point dipole p(t) in the origin with
charge and current density:
ρ(r, t) = −p(t) · ∇δ(r)
(379)
j(r, t) = ṗ(t)δ(r)
This gives the vector potential:
µ0 j(r0 , t − |r − r0 |/c
Z
1
A(r, t) = d3 r0
4π V |r − r0 |
ṗ(t − |r − r0 |/c 0
Z
µ0
= d3 r0 δ(r ) (380)
4π V |r − r0 |
µ0 ṗ(t − r/c)
=
4π r
Now we’ve found the vector potential, we find the scalar potential as well. We
do so via the Lorenz gauge:
Z
1 ∂φ
∇·A+ 2 = 0 → φ = −c2 ∇ · Adt (381)
c ∂t
The divergence of the vector potential is
 
µ0 1
∇·A= ∇ · ṗ(t − r/c)
4π r
  
µ0 1 1
= ∇ · ṗ(t − r/c) + ṗ(t − r/c) · ∇ (382)
4π r r
 
µ0 p̈(t − r/c) · r ṗ(t − r/c) · r
=− +
4π cr2 r3

56
The resulting scalar potential is
 
2 µ0 ṗ(t − r/c) · r p(t − r/c) · r
φ(r, t) = c +
4π cr2 r3
  (383)
1 ṗ(t − r/c) · r p(t − r/c) · r
= +
4π0 cr2 r3
We define p(t − r/c) = pret . The resulting electric and magnetic fields are then:
 
µ0 ṗret p̈ret
B(r, t) = − r̂ × +
4π r2 cr
 
1 3r̂(r̂ · pret ) − pret 3r̂(r̂ · ṗret ) − ṗret r̂(r̂ · p̈ret ) − p̈ret
E(r, t) = + +
4π0 r3 cr2 c2 r
(384)
We distinguish between these three terms by looking at the different depen-
dences on r:
• r−3 : near field, has no average energy flow
• r−2 : induction field, has no average energy flow
• r−1 : far field or radiation field, has average energy flow
An easy way to see why the first two have no average energy flow is to remem-
ber that the Poynting vector is proportional to the square of the electric field.
Squaring each term and multiplying them with πr2 reveals that the first two
will go to zero for r → ∞, but the latter will not.

9.14 Shape of fields

In the far field, we can approximate that r ≈ r0 . We then obtain the famil-
iar Biot-Savart law for the magnetic field. The electric field follows from the
orthogonality principle of electromagnetic waves:
µ0 j(r0 , t − |r − r0 |/c
Z
1
Arad (r, t) = d3 r 0
4π V |r − r0 |
Z (385)
µ0
≈ d3 r0 j(r0 , t − |r − r0 |/c + r̂/c)
4πr V
The magnetic radiation field is then:
1 ∂Arad (r, t)
Brad (r, t) = − r̂ × (386)
c ∂t
And the electric field is
 
∂Arad (r, t)
Erad (r, t) = −r̂ × cBrad (r, t) = r̂ r̂ × (387)
∂t
In the far field, E and B only depend on ∂Arad (r, t)/∂t.

57
 Figure 1: Shape of the different contributions.

10 Lienard-Wiechert potentials
The Lienard-Wiechert potentials are general solutions to the Maxwell equations
taking retardation into account. They are as follows:
   
1 q 1 q
φ(r, t) = = (388)
4π0 R(t)(g(t) t=tret 4π0 R − β~ · R ret
   
µ0 qv(t) µ0 qv
A(r, t) = = (389)
4π R(t)(g(t) t=tret 4π R − β~ · R ret
We will derive them in this section. They will show us how our sources ρ and j
lead to potentials. We first do a small recap of the oscillating dipole.

10.1 Oscillating electric dipole v  c

Note: This section includes a recap of 20.1-20.7.
We will revisit the case of the oscillating electric dipole. The vector potential is
as follows:
j(r0 , tr = t − |r − r0 |/c)
Z
µ0
A(r, t) = dr0 (390)
4π |r − r0 |
We do a multipole expansion for the field far away, so r  r0 .

1 1 r · r0
0
= + 3 + ... (391)
r−r r r

58
The first term is the monopole term and the second is the dipole term. If the
total charge is zero (as it is for a dipole), then the monopole term will drop
out. An additional assumption we make is tha of a low velocity, so v  c. The
dipole moment is given by:
Z
p(tr ) = ρ(r0 , tr )r0 d3 r0 (392)
V

Its time derivative is then proportional to the current.

dp(tr ) dz(tr )
=q = qv(tr ) = j(tr ) (393)
dtr dtr
Note that since v  c, the term dq/dtr is negligible. If the velocity was similar
to the speed of light (next chapter), then this term will not be negligible. Note
too, that since the dipole is fixed in space, so is the current. We show this via
insertion of a delta-function:
j(tr ) = ṗ(tr )δ(r0 ) (394)
We insert this in the equation for the vector potential, which simplifies the
integral. The vector potential via the dipole approximation then becomes:
µ0
A(r, t) = ṗ(tr = t − r/c) (395)
4πr
The magnetic field is, as always, the curl of the vector potential. There are two
terms here with spatial dependency: the time derivative of the dipole moment
and the 1/r. We start with the latter. Since there’s only one space dependency,
this spatial derivative becomes:
 
∂ 1 z
=− 3 (396)
∂z r r
For the spatial derivative of the dipole moment, We will use the product rule of
differentiation, which says that:
∂ ∂tr
ṗ = p̈ (397)
∂x ∂x
B = ∇ × A = Bφ êφ (398)
 
µ0 µ0 ṗ(t − r/c) p̈(t − r/c)
B=∇× ṗ(tr = t − r/c) = − sin θ + êφ (399)
4πr 4π r2 cr
The first term that appears is the Coulomb-field or near-field. The second term
is the ”radiation-field” or far-field. To see why one contributes to radiation,
recall that the Poynting vector is proportional to the square of the field. The
square of the first term yields a dependency on r4 and the second one a depen-
dency on r2 . Multiply both with πr2 and let r → ∞. The first term vanishes
while the second term remains. This makes sense also physically - the first
term is proportional to velocity, which, in Newtonian sense, is no different than
standing still and therefore can be said to have no ’energy’. The second term is
proportional to acceleration, which does give a net energy.

59
10.2 Fields of the harmonic oscillating electric dipole
We now assume a harmonic oscillating electric dipole. Its dipole moment is of
the following form:
p(tr ) = (0, 0, p0 cos ωtr ) (400)
The vector potential is, by virtue of the equation we had found before, then as
follows:
µ0 ωp0 sin (ω(t − r/c))
A(r, t) = ẑ (401)
4π r
Since we’re now dealing with a wave, it’s worth to recall the following much-used
wave parameters:
2π ω
k= = (402)
λ c
In the far field (where we’re dealing with radiation), we have that r  λ. The
magnetic field, as found via the curl, then becomes:
µ0 ω 2 p0 cos (ωt − kr)
Brad (r, t) = − sin θ êφ (403)
4πc r
The electric field of this dipole is found also in the familiar way:
∂A(r, t)
E(r, t) = −∇(r, t) − (404)
∂t
We also use that
∂(r, t)
= −c2 ∇ · A(r, t) (405)
∂t
The electric field is then:
µ0 ω 2 p0 cos (ωt − kr)
Erad (r, t) = − sin θ êθ (406)
4πc r
Note that this is in-phase with the magnetic field B(r, t). Also note that:

|E| = c|B| (407)

E ⊥ B ⊥ r = rn̂ (408)
Where n̂ = r̂. So using this, we find that:
r̂ × E
B= (409)
c
This allows for some simplifications for finding the Poynting vector and the
emitted power. The energy flow of the Poynting vector is as follows:
1
S(r, t) = E(r, t) × B(r, t) = c0 |E(r, t)|2 r̂ (410)
µ0
If you calculate this for the electric dipole radiation, you obtain:
µ0 ω 4 p20 cos2 (ωt − kr)
S= sin2 θr̂ (411)
16π 2 c r2

60
The power per solid angle is given by:
˙ = Sr2 dΩr̂
dP = SdA (412)

dP µ0 ω 4 p2 (tr ) sin2 θ
(t) = (413)
dΩ 16π 2 c
The total power emitted is the surface integral of the Poynting vector:
Z Z Z 2π Z π
dP dP
P = S · dA = dΩ = sin θdθdφ (414)
surface Ω=4π dΩ 0 0 dΩ

µ0 ω 4 p2 (tr )
P (t) = (415)
6πc
This last equation for the power is known as the Larmor equation. So to recap,
the power solid angle is:

dP µ0 ω 4 p2 (tr ) sin2 θ
(t) = (416)
dΩ 16π 2 c
For a harmonic function, the average is simply half this. So the time average of
power per solid angle is:

dP µ0 ω 4 p2 (tr ) sin2 θ
= (417)
dΩ av 32π 2 c
The total power is:
µ0 ω 4 p2 (tr )
P (t) = (418)
6πc
The time average of the total power is:
µ0 ω 4 p2 (tr )
Pav (t) = (419)
12πc
The power emitted by the radiation fields had a dependency of 1/r. We assumed
that we were far away such that we could do the multipole expansion with
r  r0 . We also assumed far field r  λ and a low velocity v  c. Now, what
will happen if we let v approach c?

10.3 Liénard-Wiechert potentials

We start off by stating that the retarded time (or ’particle time) is given by:
R(tret )
tret = t − (420)
c
We start by assuming there’s a single charge q moving along some path R. The
retarded electromagnetic potentials in the Lorenz gauge potential are given by:
ρ(r0 , t − |r − r0 |/c
Z
1
φ(r, t) = d3 r 0 (421)
4π0 |r − r0 |

61
 j(r0 , t − |r − r0 |/c
Z
µ0
A(r, t) d3 r0 (422)
4π |r − r0 |
The sources are described by:

ρ(r, t) = qδ(r − r0 (t)) (423)

j(r, t) = qv(t)δ(r − r0 (t)) (424)

The length of R(t) and the corresponding unit vector (from the particle at
location r0 to the observation point r). These are as follows:

R(t) = r − r0 (t) = R(t)n̂(t) (425)

We want to calculate the fields, but substitution of these sources into the po-
tentials gives us difficult integrals. We therefore introduce the dummy variable
t0 and a delta function to enforce the time-retardation of the charge density:

ρ(r0 , t0
Z Z
1
φ(r, t) = d3 r0 dt0 δ(t0 − t + |r − r0 |/c (426)
4π0 |r − r0 |

We insert the charge density and perform the space integral to obtain:

δ(t0 − t + R(t0 )/c)

Z
q
φ(r, t) = dt0 (427)
4π0 R(t0 )

Calculating this integral isn’t as straightforward as it may seem. To calculate

it anyway, we use two properties of the delta function. These are described on
page 12 in section 1.5.1 of Zangwill, but is cited here for completeness’ sake.
We know that:
1
δ(ax) = δ(x) (428)
|a|
As long as a 6= 0. A delta function ’filters’ a test function as follows:
Z ∞
dxf (x)δ(x − x0 ) = f (x0 ) (429)
−∞

We combine these two to obtain:

Z ∞ 
d 0 df 
dxf (x) δ(x − x ) = − (430)
−∞ dx dx x=x0

Let’s apply this to what we know. So we know that:

1
δ(f (t0 )) = δ(t0 − t00 ) (431)
|f 0 (t00 )|

In our case, our function f is given by:

f (t0 ) = t0 − t + R(t0 )/c (432)

62
 t0 = t − R(t0 )/c (433)
Where f (t00 ) = 0 and f 0 (t00 ) 6= 0. We name the time derivative of f (t0 ) the
function g(t0 ), such that g(t0 ) = f 0 (t0 )(6= 0). The integral then reverts to:

d 0
g(t0 ) = f 0 (t0 ) = (t − t + R(t0 )/c) (434)
dt
We can do this explicitly. We obtain:
d 0 1 d p ~ 0 ) · n̂(t0 ) > 0
g(t0 ) = 0
[t − t + R(t0 )/c] = 1 + R(t0 ) · R(t0 ) = 1 − β(t
dt c dt0
(435)
So the Liénard-Wiechert scalar potential is given by:
   
1 q 1 q
φ(r, t) = = (436)
4π0 R(t)g(t) t=tret 4π0 R − β~ · R ~ ret

The vector potential is evaluated in a similar fashion and becomes:

   
µ0 qv(t) µ0 qv
A(r, t) = = (437)
4π R(t)(g(t) t=tret 4π R − β~ · R ret

These are the Liénard-Wiechert potentials we started with. We find the electric
field of the particle in the familiar fashion:
∂A
E = −∇φ − (438)
∂t
Note that this is what the observer sees, so the time derivative is here with
respect to observer time t - something we have to tackle in a bit. First, we write
out the field explicitly
0 0 0 0 0
0 δ(t − t + R(t )/c) 0 v(t )δ(t − t + R(t )/c
Z Z
q µ0 q ∂
E(r, t) = − ∇ dt − dt
4π0 R(t0 ) 4π ∂t R(t0 )
(439)
We evaluate this using our knowledge of Using our knowledge of derivatives:

∇R = n̂ ∇(1/R) = −R/R3 (440)

So the electric field is given by

" #
q

n̂

q d n̂ − β~
E(r, t) = + (441)
4π0 gR2 ret 4π0 dt gcR
ret

Where g = 1 − β~ · n̂. The time derivative here is in terms of observer time,

and as such, not particulary useful (though it is exact!). We want to have it
in terms of the retarded/particle time tr . We recall the relation between the

63
retarded time and observer time, as well as the expression for the vector from
the particle (at r0 ) towards the observation point r:

R(tret )
tret = t − (442)
c

R(t) = r − r0 (t) = R(t)n̂(t) (443)

We found previously that:

d 0 1 d p ~ 0 ) · n̂(t0 ) > 0
g(t0 ) = 0
[t − t + R(t0 )/c] = 1 + R(t0 ) · R(t0 ) = 1 − β(t
dt c dt0
(444)
Such that:
g = 1 − β~ · n̂ (445)
Comparing these two, we see that the time derivative we see that the time
derivative of the position function becomes:
dR
= −n̂ · cβ~ (446)
dt
Using this, we can write that:
dt 1 dRret h i
=1+ = 1 − β~ · n̂ = gret (447)
dtret c dtr et ret

Secondly, we directly differentiate n̂ = R/R directly and use dR ~

dt = −n̂ · cβ to
deduce that:
dn̂ c ~
= n̂ × (n̂ × β) (448)
dt R
We return to the electric field we had just found, and replace d/dt by d/dtret
using the relations we just found. All variables are now retarded, and carrying
out the indicated derivatives produces:
" #
q

n̂

q d n̂ − β ~
E(r, t) = + (449)
4π0 gR2 ret 4π0 dt gcR
ret
"! #
q

n̂

d n̂ − β~
E(r, t) = + (450)
4π0 gR2 ret dt gcR
ret
" ! #
q n̂ 1 dn̂ dβ ~ n̂ − β~ dg

dR
E(r, t) = + − − 2 2 R+g (451)
4π0 gR2 cgR dt dt cg R dt dt
ret
˙
Apart from the dimensionless acceleration β~ = dβ/dt,
~ the time derivatives in
this expression are given by the substitutions we just found. Doing so, we obtain

64
the Liénard-Wiechert electric field
 o
~ × β~˙
n
~ 2
q  (n̂ − β)(1 − β ) n̂ × (n̂ − β)
E(r, t) = + (452)

4π0 g 3 R2 cg 3 R
 
ret

We see two terms appear. The first term is dependent on the velocity, and
is therefore called the ’velocity field ’ term, with a dependence of R2 . Again,
recall that in the case of a Poynting vector, this decreases too rapidly if it’s to
contribute anything - it remains ”attached” to the source charge in the sense
of the near field. The second term is dependent on the acceleration, and is
therefore called the ’acceleration field’. It has 1/R dependence, so this term will
contribute to a net energy flow. Stating them explicitly:

~
q (n̂ − β)(1 − β2)
Ev =
4π0 g 3 R2
(453)
~ × β~˙
n o
q n̂ × (n̂ − β)
Ea =
4π0 cg 3 R

Note that every parameter of the source is evaluated at the retarded time (a.k.a.
the particle’s own time), rather than the observer time! We can go through the
same steps to find the magnetic field. We use B = ∇ × A. The magnetic field
then becomes:
" ˙ ˙ #
µ0 q (v × n̂)(1 − β 2 ) (β~ × n̂)(β~ · n̂) + g β~ × n̂
B= + = Bv + Ba (454)
4π g 3 R2 g3 R
ret

Like the electric field, this decomposes naturally in a velocity magnetic field and
an acceleration magnetic field. From this, we derive the following important
results:
cB = n̂ret × E (455)
B·E=0
(456)
B · n̂ret = 0
This is always and everywhere true for one moving point charge. Lastly,
note that:
Ev · n̂ret 6= 0
(457)
Ea · n̂ret = 0

65
11 Synchrotron Radiation
We will make a short recap of the Liénard-Wiechert potentials and the E and
B fields. We will then move on to describe the fields of a point charge moving
at constant velocity and dive into synchrotron radiation.

11.1 Liénard-Wiechert potentials recap

The relation between the retarded time and observer time was:
R(tret )
tret = t − (458)
c
The scalar and vector potential in the Lorenz gauge, accounting for retardation,
were:    
1 q 1 q
φ(r, t) = = (459)
4π0 R(t)g(t) t=tret 4π0 R − β~ · R
~ ret
   
µ0 qv(t) µ0 qv
A(r, t) = = (460)
4π R(t)(g(t) t=tret 4π R − β~ · R ret
We found the relation:
dt 1 dRret h i
=1+ = 1 − β~ · n̂ = gret (461)
dtret c dtr et ret

The electric and magnetic fields were:

 o
~ × β~˙
n
~ 2
q  (n̂ − β)(1 − β ) n̂ × (n̂ − β)
E(r, t) = + = Ev + Ea (462)

4π0 g 3 R2 cg 3 R
 
ret
" ˙ ˙ #
µ0 q (v × n̂)(1 − β 2 ) (β~ × n̂)(β~ · n̂) + g β~ × n̂
B= + = Bv + Ba (463)
4π g 3 R2 g3 R
ret

11.2 Fields of a point charge in uniform motion

The electromagnetic fields produced by a point charge with constant velocity
were calculated by solving the inhomogeneous wave equation. This did not
produce radiation, but the phenomenon of retardation did not play an obvious
role in this calculation. We shed light on this issue by transforming the retarded-
˙
time formulae into observer-time formulae for the special case when β~ = 0. To
recall, the Lienard-Wiechert fields were as follows:
 i
~ × β~˙
h
~ 2
q  (n̂ − β)(1 − β ) n̂ × (n̂ − β)
E= +  = Ev + Ea (464)

4π0 g 3 R2 cg 3 R

ret

66
Figure 2: A point charge q with uniform velocity v = cβ.~ The location of the
charge at the retarded time is labeled A. The observation point is labeled C.

Where h i
gret = 1 − β~ · n̂ (465)
ret

cBv = n̂ret × Ev and cBa = n̂ret × Ea (466)

These are the ones we’re going to transform to observer-time for the case that
˙
β~ = 0. The point charge q in the figure moves uniformly with velocity v = cβ. ~
The balck dot labels the position of the charge at the time t, when the fields are
osberved at the point C. The point labeled A is the position of the charge at
˙
the retarded time tret . We begin with the electric field and, because β~ = 0, only
the first (velocity) term in the electric field equation requires our attention:
" #
q ~
(n̂ − β)(1 − β2)
Ev = (467)
4π0 g 3 R2
ret

Our first task to prove that:

R
[n̂ − β]ret = (468)
Rret
To rewrite this, we use the implicit equation for the retarded time:
R(tr )
tr = t − (469)
c
We use this to compute the vector ∆s which points from the retarded-time
position of the charge to the observation-time position in the figure:
~ ret
∆s = v(t − tr ) = βR (470)

The geometry of the figure shows that:

~ ret + R = Rret = Rret n̂
βR (471)

That is, the vector sum of the particle moving between t and tr , and its position
at observation time t to the observer, is the same as the vector from the retarded

67
position to the observer directly. Combining our expression for ∆s with this
vector-sum equation gives:

R = Rret n̂ − βRret (472)

R
= n̂ − β (473)
Rret
Which was the first thing we wanted to prove. Inserting this proven identity
into the equation for the velocity field gives:

q R(1 − β 2 )
Ev = (474)
4π0 [g 3 R3 ]ret

To make further progress, we note that the distance AB = βR ~ ret · n̂ret is the
~
projection of the vector βRret onto the direction n̂, where we note that the total
distance AC is Rret . In that case, the definition of gret from (23.27) tells us
that the distance
~ = [gR]ret
BC = Rret (1 − n̂ · β) (475)
since h i
gret = 1 − β~ · n̂ (476)
ret

The distance from point B, to the black dot (let’s call it q) is given through
simple geometry:
Bq = βRret sin α (477)
We use this with the Pythagorean theorem that:
2
BC + β 2 Rret
2
sin2 α = R2 (478)

Imagine now that if you were to drop straight down from C such that you get
~ Let’s call this the distance D. Then, say you are on
a line perpendicular to β.
the β-axis, say at A. Then that distance D would be equal to:

Rret sin α = D (479)

But if you were at q, that distance D would be:

R sin θ (480)

So, in other words, we have that:

Rret sin α = R sin θ (481)

Now we’re going to combine all this. Our three equations are:
h i
gret = 1 − β~ · n̂ (482)
ret

2
BC + β 2 Rret
2
sin2 α = R2 (483)

68
Figure 3: Electric field lines for a point charge in uniform motion with v/c =
0.95. For a slow-moving particle, all the field lines would point equally in all
direction radially away from the particle. This ’compression’ for a fast moving
particle are described as being no longer ’Coulomb’.

Rret sin α = R sin θ (484)

So we say that:
1
[gR] = R(1 − β 2 sin2 θ) 2 (485)
We substitute this into our velocity field to obtain:

q R(1 − β 2 )
Ev = (486)
4π0 R3 (1 − β 2 sin2 θ) 23

And so we’ve transformed the retarded-time formula for the velocity field to the
observer-time. To find the magnetic velocity field, we use our known equation:
 
n̂ 1 ~ R v
Bv = × Ev = β+ × Ev = 2 × Ev (487)
c c Rret c

Note: the field of the moving point charge is expressed using the ’actual time’,
observer time if you will, t. You can see the electric field lines of a fast-moving
particle in figure 3.

11.3 Synchrotron Radiation

Synchrotron radiation is the radiation of an electron in a circular motion. Syn-
chrotrons are the type of particle accelerators used to obtain this radiation.
The one in Grenoble has a radius of approximately 100 m, and the velocity of
the electrons there is approximately that of the speed of light. A typical naive
calculation would therefore say that the radiation emitted would by:
2πR 1
v≈c→T ≈ →f = ≈ 106 Hz (488)
c T
Megahertz frequencies are radiowaves, so this doesn’t seem particulary useful.
Nonetheless, the Grenoble synchrotron is a brilliant X-ray source which can

69
reach frequencies up to 1019 Hz. As it turns out, for such large frequencies the
frequency emitted is approximately

fsynchr. ≈ γ 3 f0 ≈ 1018 Hz (489)

where
1
γ=q ≈ 104 (490)
v 2


1− c

Relativity is responsible here for three effects. The first is that of Lorentz
contraction.

11.3.1 Synchrotron radiation amplification via Lorentz contraction

of the undulator
The first amplification effect is that of Lorentz contraction. The electron beam
moves through a magnetic field produced by permanent magnets, called an
undulator. The reason for this is because electrons moving through the periodic
magnet structure are forced to undergo oscillations and thus to radiate energy.
The observed magnet period in the electron rest-frame becomes contracted. The
the distance between two magnets in the Lab-frame is L, which in the electron
rest-frame becomes:
L0 = L/γ (491)
The spatial period of the undulator magnets in the Lab-frame is λu . In the
electron rest-frame this becomes:

λu → λu /γ (492)

The frequency emitted approximately therefore:

1
f0 ≈ ≈ γf0 (493)
λ0

11.3.2 Synchrotron radiation amplification via Doppler shift

The second effect is that of the Doppler shift. This shifts the frequency of the
radiation emitted in the longitudinal direction. This is proportional to

f 0 ∝ γf (1 + cos θ0 ) ≈ 2γf (494)

Combining the Lorentz contraction and the Doppler shift, we find that the
amplification so far is
f 00 ≈ γf 0 ≈ γ 2 f0 (495)

70
Figure 4: The angular distribution of the radiation cone of an electron with
β~ ⊥ cis amplified in the forward direction

11.3.3 Synchrotron radiation amplification via transverse narrowing

of of emission cone
The third relativistic effect is another consequence of the Doppler-effect. It’s the
transverse spatial narrowing of emission to a narrow cone, as seen in figure 4.
The angle of emission θ is proportional to 1/γ. An electron emits only towards
the observer for a short while, so only in a short pulse. The amplification of the
frequency then becomes:
000
fcrit ≈ γf 00 ≈ γ 3 f0 (496)
Wavelength, speed and frequency are related as follows:
c
λ= → f = c/λ (497)
f
The energy of emitted photons is

E = hf (498)

The first amplification keeps us in the radiowave area. The second amplification
effect takes us into visible light regions, and the third amplification takes us into
X-ray energies of approximately 1019 Hz.

11.4 Alternative picture: Fourier transform of ultrashort

pulse
We can also find the frequencies emitted via the Fourier transform of an ul-
trashort pulse. A particle emits in the same ’cone’ for a small amount of time
∆t, much like a sweeping searchlight as seen in figure 5. We know that the
angular width of the beam is approximately ∆θ ≈ 1/γ (homework assignment),
so we’re looking for the that it takes the particle to ’sweep’ that cone. The
time that it takes to do so is the time ∆t. The ’thickness’ of this pulse is the
difference between the distance a particle travels and the distance light travels
in this time. We call this thickness L, the distance the particle travels d and
the distance light travels D.
The distance D light travels in a short time ∆t is D = c∆t. We can rewrite
the time as follows:
d 2ρ · 1/γ 2ρ
D = c∆t = c =c = (499)
v v γβ

71
Figure 5: The electron moves in a circular direction (the small circle in the
center) and projects its energy forward, leading to a coiling trail of energy
emitted radially outwards. The ’thickness’ of the beam is L distribution of the
radiation cone of an electron with β~ ⊥ cis amplified in the forward direction

Which we can say since we know that:

2ρ
d = v∆t = (500)
γ

Where ρ is the radius of the synchrotron. The angular width of the pulse is 1/γ,
so the arc that is sweeped in this time is θ(rad)ρ = γρ . The width is defined
here is the distance from the center to the maximum ’sweep’, so the total is 2
times this distance. The thickness L of our beam is then the difference between
D and d:
2ρ ρ
L=D−d= − (501)
γβ γ
The total time duration of this light pulse is then this distance divided by the
speed of light:
L 1 2ρ (1 − β) ρ
Tpulse = ≈ ≈ 3 (502)
c c γ β cγ
And so the frequency of our pulse is approximately:
1 c 3
ωc ≈ ≈ γ (503)
Tpulse ρ

Where used the fact that ∆ωTpulse ≈ 1. Turning it around, the duration of our
pulse is: The duration of our ultrashort pulse is about:
ρ 1
Tpulse ≈ ≈ (504)
cγ 3 ω0 γ 3

72
From this we see that the X-ray frequency of our synchrotron are tunable by
increasing the velocity. We’ll find in the next section that, if the power of the
electron for low velocities is given by PLarmor , the power for relativistic particles
will be P = γ 4 PLarmor . So synchrotrons X-rays are tunable X-ray energies with
a small beam spot and high power.

11.5 Angular emission and power of electromagnetic ra-

diation
11.5.1 Case A: Dipole radiation of linear movement with v  c
~˙ The dipole is defined as:
Let’s look at an oscillating point particle with β~ k β.

p = qd (505)
˙
p̈ = qa = qc~β (506)
~v
β~ = (507)
c
Its electric field is then given by:

~˙
" #
q n̂ × (n̂ × β)
E(r, t) = (508)
4π0 c R0

The Poynting vector is:

S(r, t) = c0 |E(r, t)|2 n̂ (509)
The power emitted per solid angle is:

dP ~˙ 2 sin2 θ
q 2 |β|
= R02 (S · n̂) = (510)
dΩ 16π 2 0 c
The total power emitted is:

~˙ 2
q 2 |β|
PLarmor = (511)
6π0 c
This is Larmor equation.

11.5.2 Case B: Dipole radiation of linear movement with v ≈ c

Using that
1
γ=q (512)
v 2


1− c

The electric field now becomes:

~˙
" #
q n̂ × (n̂ × β)
E(r, t) = (513)
4π0 c R0 (1 − β~ · n̂)3
ret

73
The power emitted per solid angle is also slightly tweaked:

dP ~˙ 2
q 2 |β| sin2 θ
= R02 (S · n̂) = (514)
16π 2 0 c 1 − vc cos5 θ


dΩ

Where θ = 0 is the forward direction. Note that the denominator goes to zero
for the forward direction and v c. The angle at which maximum energy is
projected is given by:
1
θmax = (515)
2γ
If v → c.

11.5.3 Case C: Dipole radiation of circular movement with v ≈ c

~˙ The electric field is given by:
We now have that β~ ⊥ β.
 
~ × β~˙

q  n̂ × (n̂ − β)
E(r, t) = (516)

4π0 c R0 (1 − β~ · n̂)3
 
ret

The power emitted per solid angle is:

~˙ 2
" #
dP 2 q 2 |β| 1 sin2 θ cos2 φ
= R0 (S · n̂) =
1− 2 (517)
16π 2 0 c 1 − v cos θ 3 γ 1 − vc cos2 θ


dΩ
c

The denominator of the term between the brackets goes to 0 for θ → 0 (forward
direction) and v/c 1, so the term between the brackets becomes unity. The max-
imal angle of emitted power is therefore concentrated in the forward direction,
or, in other words, emission is in a narrow cone in the forward direction. The
power emitted in the backward hemisphere virtually disappears when β → 1.
The frequency spectrum we find this way applies to a single pulse of synchrotron
radiation. Such a pulse is produced whenever a relativistic charge traverses even
a very small segment of circular arc. For a fixed observer, a relativistic charge
moving in a circular orbit produces a sequence of these pulses separated by the
period T of the orbit. However, a fixed observer will not only see such this fun-
damental frequency, but also its higher harmonics, leading to a broad frequency
spectrum.

74
12 Synchrotron Radiation and Cherenkov Radi-
ation
Synchrotron radiation is encountered in cases where the velocity of the particle
becomes relativistic. the impact of this on the particle’s radiation is that the
emission becomes focussed in a narrow cone in the forward direction. The power
is then also amplified:

PLarmor → P = γ 4 PLarmor (518)

12.1 Angular emission and power of electromagnetic ra-

diation
For a circular motion in the relativistic limit (v → c), we have that γ  1. If
the particle is undergoing circlular motion, we have that β~ ⊥ β. ~˙ The radiation
field of the electric field is:
" ˙#
q n̂ × (n̂ − β~ × β~
E(r, t) = (519)
4π0 c R(1 − β~ · n̂)3
ret

The power emitted per solid angle is given by:

dP ~˙ 2
q 2 |β| 1

sin2 θ cos2 φ

=
1− 2 (520)
dΩ 16π 2 0 c 1 − v cos θ 3 γ (1 − vc cos θ)2
c

Note that the denominator goes to 0 for θ → 0 (forward direction) and vc →

1 for relativistic velocities. For an angle θm ax, the denominator resolves to
1
θm ax = 2γ . Calculating it explicitly gives a total dependence of γ 4 , with the
total power becoming:
~˙ 2
q 2 |β|
P = γ4 (521)
6π0 c

12.2 Synchrotron Radiation: frequency spectrum

Were it not for this amplification of γ 4 , a synchrotron would produce mi-
crowaves. A synchtron also allows for harmonics of the fundamental wavelength,
so we obtain a generation of many higher harmonics. We assume the radiation
to last very shortly.
A particle emits in the same ’cone’ for a small amount of time ∆t, much
like a sweeping searchlight. We know that the angular width of the beam is
approximately ∆θ ≈ 1/γ (homework assignment), so we’re looking for the that
it takes the particle to ’sweep’ that cone. The time that it takes to do so is the
time ∆t. The ’thickness’ of this pulse is the difference between the distance a
particle travels and the distance light travels in this time. We call this thickness
L, the distance the particle travels d and the distance light travels D.

75
 The distance D light travels in a short time ∆t is D = c∆t. We can rewrite
the time as follows:
d 2ρ · 1/γ 2ρ
D = c∆t = c =c = (522)
v v γβ
Which we can say since we know that:
2ρ
d = v∆t = (523)
γ

Where ρ is the radius of the synchrotron. The angular width of the pulse is 1/γ,
so the arc that is sweeped in this time is θ(rad)ρ = γρ . The width is defined
here is the distance from the center to the maximum ’sweep’, so the total is 2
times this distance. The thickness L of our beam is then the difference between
D and d:
2ρ ρ
L=D−d= − (524)
γβ γ
The total time duration of this light pulse is then this distance divided by the
speed of light:
L 1 2ρ (1 − β) ρ
Tpulse = ≈ ≈ 3 (525)
c c γ β cγ
And so the frequency of our pulse is approximately:
1 c 3
ωc ≈ ≈ γ (526)
Tpulse ρ

Where used the fact that ∆ωTpulse ≈ 1. Turning it around, the duration of our
pulse is: The duration of our ultrashort pulse is about:
ρ 1
Tpulse ≈ ≈ (527)
cγ 3 ω0 γ 3
From this we see that the X-ray frequency of our synchrotron are tunable by
increasing the velocity. Since the pulse is narrow in the time-domain, it’s broad
in the frequency domain. The ’pulses’ come in at regular interval, so the Fourier
transform will be peaks at the many harmonics of the fundamental frequency.
Synchrotron radiation is therefore tunable X-ray energy, of which the beam spot
can be very small but with high power (in jargon that ’s called high brilliance).
Such small wavelengths make it suitable for the imaging of molecules such as
enzymes.

76
12.3 Finding the frequency spectrum in six steps
The Fourier transform of our radiation field is given by:
 
Z ∞ n̂ × n̂ − β~ × β~˙
 
q iωt
E(ω) =  e dt (528)
 
4π0 c −∞ ~
 3
R(1 − n̂ · β)
ret

This function gives us the frequency spectrum of the radiated power. To find
it, we use six steps.
1. Fourier transform of E-field (Liénard-Wiechert), where we use dtret instead
of dt.
2. Golden rule synchrotron radiation. Take n̂ constant, which is allowed
because the energy is emitted in a very small amount of time.
3. Electron in a circular orbit (so express orbit in tret .

4. Two small angle approximations

5. Taylor expansion to express t in temrs of tret (in eiωt ).
6. E-field in terms of tabulated integral, which is an Airy function

We go through each step.

12.3.1 Step 1 Change of variable

The radiation part of the electric field is as follows:
" ˙#
q n̂ × (n̂ − β~ × β~
E(r, t) = (529)
4π0 c R(1 − β~ · n̂)3
ret

The Fourier transform of the electric field is as follows:

Z ∞
E(ω) = dtE(t) exp (iωt) (530)
∞

This integral is easier if you integrate using the particle’s own time tret :

R(tret )
tret = t − (531)
c
dt 1 dRret
=1+ = [1 − β~ · n̂]ret (532)
dtret c dtret
So the Fourier transform becomes:
Z ∞
E(ω) = E(t) exp (iωt)[1 − β~ · n̂]ret dtret (533)
∞

77
12.3.2 Step 2 ’Golden Rule’
Writing out the Fourier transform, we obtain:
Z ∞" ˙#
q n̂ × (n̂ − β~ × β~
E(ω) = exp (iωt)[1 − β~ · n̂]ret dtret (534)
4π0 c ∞ R(1 − β~ · n̂)3 ret

~ × β~˙
" #
∞
n̂ × (n̂ − β
Z
q
= exp (iωt)dtret (535)
4π0 c ∞ R(1 − β~ · n̂)2 ret
Now we use the ’Golden rule’ of synchrotron radiation, which says that we can
take n̂ constant (and taken at the ’flash point’). We integrate by parts and
obtain: Z ∞h
qiω ~
i
E(ω) = n̂ × (n̂ × β) eiωt dtr (536)
4π0 cR −∞ ret

12.3.3 Step 3 Consider a relativistic electron in a circular orbit

By considering a relativistic electron in a circular orbit, we can rewrite the
double cross product. We know that:
v cβ
ω0 = = (537)
ρ ρ

Assume that at tr = 0 the particle is at the origin r(tr = 0) = (0, 0, 0). The
position vector then becomes:

r(tr ) = ρ (1 − cos ω0 tr )) x̂ + ρ (sin ω0 tr ) ẑ (538)

And
~ r ) = ~v (tr ) = 1 d~r(tr ) = ρω0 (sin (ω0 tr )x̂ + cos (ω0 tr )ẑ)
β(t (539)
c c tr c
The Golden rule formulated mathematically says that:
!
~
R
n̂ = = (0, sin θ, cos θ) (540)
R
tr =0

The double cross product then becomes:

h i
~
n̂ × (n̂ × β) = β(− sin (ω0 tr ), sin θ cos θ cos (ω0 tr ), − sin2 θ cos (ω0 tr ))
ret
(541)
Note that: h i
~
n̂ × (n̂ × β) = β(− sin (ω0 tr ), 0, 0) (542)
ret
For θ = 0.

78
12.3.4 Step 4 Two small-angle approximations
We’re gonna take two small angle approximations. Recall that:
v cβ
ω0 = = (543)
ρ ρ
So we make the following two separate small angle approximations:
• γ  1 → ω0 tr very small
• θ ≈ 1/γ
The double cross product then becomes:
h i
~
n̂ × (n̂ × β) = β(− sin (ω0 tr ), sin θ cos θ cos (ω0 tr ), − sin2 θ cos (ω0 tr ))
ret
(544)
~ = β(−ω0 tr , θ, 0)
n̂ × (n̂ × β) (545)

12.3.5 Step 5 Taylor expansion of advanced/observer time t in terms

of retarded/particle time tr
p
R(tr ) = r2 − 2ρr cos θ sin (ω0 tr ) + 2ρ2 [1 − cos ωtr ] ≈ r − ρ cos θ sin (ω0 tr )
(546)
We look in the radiation zone where r  ρ. For the case that θ = 0 we have
that:
t − r/c = tr − (r/c) sin (ω0 tr ) (547)
And using the fact that
v cβ
ω0 = = (548)
ρ ρ
We obtain that
r 1 1
t≈+ tr + ω02 t3r (549)
c 2γ 2 6
For the general case, where θ can have any value, we obtain that:
r (1 + γ 2 θ2 1
t≈ + tr + ω02 t3r (550)
c 2γ 2 6

12.3.6 Step 6 Solve to get the electric field Ex (ω)

We solve for the electric field Ex (ω) (so for the case θ = 0). Our simplified
Fourier transform was:
Z ∞h
qiω ~
i
E(ω) = n̂ × (n̂ × β) eiωt dtr (551)
4π0 cR −∞ ret

And we know we can approximate t in tr as follows:

r (1 + γ 2 θ2 1
t≈ + tr + ω02 t3r = r/c + a1 tr + a2 t3r (552)
c 2γ 2 6

79
The coefficients a1 + a2 form the frequency ω in
3
eiωt = ei(b1 tr +b2 tr ) = cos (b1 tr + b2 t3r ) + i sin (b1 tr + b2 t3r ) (553)

To transform back, we use the substitution:

 13
ω c2

u= tr (554)
2 ρ2

The Fourier transform of the electric field then becomes:

 2  23 Z (" 2 # )
qωω0 2ρ 3ω 3 u3
Ex (ω) = u sin u + du (555)
4π0 cR ωc2 4ωc 3

This integral is a variation on the tabulated integral of the Airy function:

1 ∞ u3
Z  
0 dAi(x)
Ai (x) = =− u sin xu + du (556)
dx π 0 3

12.3.7 Comments on frequency spectrum

The frequency spectrum of synchrotron radiation was first derived by Julian
Schwinger. His solution yields an expression in terms of modified Bessel funtions,
like in Zangwill. The ’brilliance’ spectrum, that is, the amount of power emitted
per solid angle per frequency for the plane θ = 0 is given as:
2
d2
  
ω 2 ω
∝ K2/3 (557)
dωdΩ ωc 2ωc

with ωc = 32 γ 3 ω0 . Since the photon energy is given by E = h̄ω, we see that the
synchrotron emits more of some photons than it does of others.

12.4 Take home messages

The Liénard-Wiechert potentials give the velocity and acceleration fields of E
and B. Synchrotron radiation is produced by a relativistic point charge in
circular motion (v ⊥ v̇ and v ≈ c). The angular power emission is concentrated
in a very narrow cone, oriented in the forward direction. This power is enhanced
and has a broad frequency spectrum with many higher harmonics.

80
12.5 Cherenkov radiation
For Cherenkov radiation we look at particles that travels faster than the phase
speed of light in a non-vacuum medium. For example, water has a refractive
index of n = 1, 33, so c/n ≈ 0, 75c < v < c. The characteristic blue light that is
emitted by Cherenkov radiation isn’t produced by the particle itself, but by the
polarization wave that it induces in the medium the particle travels through.
To see why, we take a few steps back.
The electromagnetic fields produced by a point charge with constant velocity
were calculated by solving the inhomogeneous wave equation. This did not
produce radiation, but the phenomenon of retardation did not play an obvious
role in this calculation. We shed light on this issue by transforming the retarded-
˙
time formulae into observer-time formulae for the special case when β~ = 0. To
recall, the Lienard-Wiechert fields were as follows:
 i
~ × β~˙
h
~ 2
q  (n̂ − β)(1 − β ) n̂ × (n̂ − β)
E= +  = Ev + Ea (558)

4π0 g 3 R2 cg 3 R

ret

Where h i
gret = 1 − β~ · n̂ (559)
ret

cBv = n̂ret × Ev and cBa = n̂ret × Ea (560)

These are the ones we’re going to transform to observer-time for the case that

Figure 6: A point charge q with uniform velocity v = cβ.~ The location of the
charge at the retarded time is labeled A. The observation point is labeled C.

˙
β~ = 0. The point charge q in the figure moves uniformly with velocity v = cβ. ~
The balck dot labels the position of the charge at the time t, when the fields are
osberved at the point C. The point labeled A is the position of the charge at
˙
the retarded time tret . We begin with the electric field and, because β~ = 0, only
the first (velocity) term in the electric field equation requires our attention:
" #
q ~
(n̂ − β)(1 − β2)
Ev = (561)
4π0 g 3 R2
ret

81
Our first task to prove that:
R
[n̂ − β]ret = (562)
Rret
To rewrite this, we use the implicit equation for the retarded time:
R(tr )
tr = t − (563)
c
We use this to compute the vector ∆s which points from the retarded-time
position of the charge to the observation-time position in the figure:
~ ret
∆s = v(t − tr ) = βR (564)

The geometry of the figure shows that:

~ ret + R = Rret = Rret n̂
βR (565)

That is, the vector sum of the particle moving between t and tr , and its position
at observation time t to the observer, is the same as the vector from the retarded
position to the observer directly. Combining our expression for ∆s with this
vector-sum equation gives:

R = Rret n̂ − βRret (566)

R
= n̂ − β (567)
Rret
Which was the first thing we wanted to prove. Inserting this proven identity
into the equation for the velocity field gives:

q R(1 − β 2 )
Ev = (568)
4π0 [g 3 R3 ]ret

To make further progress, we note that the distance AB = βR ~ ret · n̂ret is the
~
projection of the vector βRret onto the direction n̂, where we note that the total
distance AC is Rret . In that case, the definition of gret from (23.27) tells us
that the distance
~ = [gR]ret
BC = Rret (1 − n̂ · β) (569)
since h i
gret = 1 − β~ · n̂ (570)
ret

The distance from point B, to the black dot (let’s call it q) is given through
simple geometry:
Bq = βRret sin α (571)
We use this with the Pythagorean theorem that:
2
BC + β 2 Rret
2
sin2 α = R2 (572)

82
Imagine now that if you were to drop straight down from C such that you get
~ Let’s call this the distance D. Then, say you are on
a line perpendicular to β.
the β-axis, say at A. Then that distance D would be equal to:

Rret sin α = D (573)

But if you were at q, that distance D would be:

R sin θ (574)

So, in other words, we have that:

Rret sin α = R sin θ (575)

Now we’re going to combine all this. Our three equations are:
h i
gret = 1 − β~ · n̂ (576)
ret

2
BC + β 2 Rret
2
sin2 α = R2 (577)
Rret sin α = R sin θ (578)
So we say that:
1
[gR] = R(1 − β 2 sin2 θ) 2 (579)
We substitute this into our velocity field to obtain:

q R(1 − β 2 )
Ev = (580)
4π0 R3 (1 − β 2 sin2 θ) 23

And so we’ve transformed the retarded-time formula for the velocity field to the
observer-time. To find the magnetic velocity field, we use our known equation:
 
n̂ 1 ~ R v
Bv = × Ev = β+ × Ev = 2 × Ev (581)
c c Rret c
We will need the electric field expression later on, when describing Cherenkov
radiation.
It should be noted that, microscopically speaking, the particle doesn’t emit
radiation itself. The actual source of the characteristic blue light is the time-
dependent polarization of the medium induced by the motion of the particle.
On the other hand, a particle moving at any speed through a medium induces
transient accelerations of the particles of the medium. Therefore, subtle inter-
ference must be at work to restrict the generation of Cherenkov radiation to
situations where the particle speed exceeds the phase velocity of light in the
medium. Nonetheless, it’s convenient from a macroscopic point of view to re-
gard the moving charge as the source of the radiation. This makes the following
elementary discussion sufficient to reveal its basic characteristics and practical
importance.

83
 First off, we need to update our potentials. To describe Cherenkov radiation,
we use retarded potentials, but now we replace the physical constants:

c → cn 0 →  µ0 → µ (582)

We convert to these new quantities via simple substitution wherever possible,

and otherwise via the following relation:

cn = c/n β~ = ~v cn (583)

Therefore, the modified Liénard-Wiechert potentials for our Cherenkov problem

are:
 
1 q
φ(r, t) =
4π R − β~n · R
~ ret
  (584)
µ qv
A(r, t) =
4π R − β~n · R ret

[]

Figure 7: The Cherenkov effect for a charged particle (black dot) moving at
constant speed. (a) spherical waves are emitted at previous positions of the
particle (small white circles) and expand at speed c/n. (b) When the particle
speed v > c/n, two information-collecting shells collapse onto every observation
point (star) inside the Mach cone at the same observation time.

[]

Figure 8: The present-time position (black dot) and retarded time position
(small white circle) of a charge moving with constant velocity v. R and Rret
point to the observation point.

Figure 12.5(a) shows a point charge (black dot) moving at constant speed
v > c/n through a dielectric medium. The small open circles represent the

84
position of the charge at earlier, equally spaced moments in time. The large
circles indicate the outer limit of the spherical wave fronts emitted by the particle
at those times. The geometry shows that the expanding radiation front is the
surface of a cone which is tangent to all spherical fronts and which has its apex
at the position of the charge. The phrase ”Mach cone” is often used because the
Cherenkov wave front is analogous to the conical wave front formed behind an
airplane when it flies at supersonic speeds. The geometry shows that the cone
angle θc is
c/n
sin θc = (585)
v
Putting it analytically, we have two solutions for the retarded time tret (given
r and t) inside the cone
R(tret )
tret − t = (586)
cn
and we have no solutions outside the cone. Figure 12.5 (b) shows an observation
point (star) inside the Mach cone of a charge moving at constant velocity. Also
shown are two views of an information-collecitng shell which collapses at speed
cn and arrives at the star at time t. The moving charge enters the volume
enclosed by the shell at time t1 < t and exits that volume at time t2 where
t > t2 > t1 . Both of these are legitimate ”retarded times” when v > cn . By
contrast, the trajectory r0 (t) (particle path) of the charge never enters or exits
the volume when the observation point lies outside the Mach cone. There is
no retarded time and the field is zero at such points. To be more quantitative,
we refer to figure 12.5 and note that the square of Rret = |R + v(t − tret | is a
quadratic equation for t − tret with the solutions:
p
−v · R ± (v · R)2 − (v 2 − c2n )R2
t − tret = ≥0 (587)
v 2 − c2n
If v > cn , the positivity condition on the far right of this equation imposes two
conditions:
v · R < 0 (v · R)2 > (v 2 − c2n )R2 (588)
With the definition for the Mach cone angle θc , these two demands respectively
imply that solutions for the time exist if two following two are true as well:
π
θ> 2 and sin θ ≤ sin θC = 1/βn (589)

These define the volume inside the Mach cone in figure 12.5. Our velocity
electric field in terms of observer time t is as follows
q R(1 − β 2 )
Ev = (590)
4π0 R3 (1 − β 2 sin2 θ) 23

We want to know the radiation power emitted at a certain frequency:

d2 P
(ω) ∝ |E(r, ω)|2 (591)
dΩdω

85
We know only the radiation field emits power, so we take the Fourier transform
of that to find the power per frequency:
 i
~ × β~˙
h
Z ∞ n̂ × (n̂ − β)
q iωt
E(ω) =  e dtr (592)
 
4π0 c −∞ ~ 2

R(1 − n̂ · β)
ret

Integrate this by parts to obtain:

Z ∞
qiω h
~
i
E(ω) = n̂ × (n̂ × β) eiωt dtr (593)
4π0 cR −∞ ret

This looks weird, since it only depends on velocity, not on acceleration. Then
again, our particle was moving at a constant velocity, so any acceleration terms
have to drop out. It’s now worth remembering that physically, it’s not our par-
ticle that’s emitting the Cherenkov radiation, but the EM-fields that it induces
in its medium for the same reason we previously said it didn’t contribute to
radiation. Nonetheless, pretending as if it’s the particle itself that emits the ra-
diation provides us with reasonable insight. We see that the Fourier transform
has an exponential in terms of observer time t, but is evaluated for retarded
time tr . so we have to replace one with the other:
R(tr ) n̂ · (~r − ~r0 (tr )) n̂ · ~r n̂ · ~r0 (tr )
t = tr − = tr + → t = tr + − (594)
c c c c/n
This may seem long and tedious, and it is. We substitute this into our Fourier
transform. The second term here only depends on space and not on time, so
that part of the exponential can be put in front of the integral. Doing so gives
us:
qiω iωr/c ∞ h
Z i h
n̂·~
r0 (tr )
i
~ iω tr − c/n
E(ω) = e n̂ × (n̂ × β) e dtr (595)
4π0 cR −∞ ret

We use that ~r0 (tr ) = ~v tr to rewrite this to:

qiω iωr/c ∞ h
Z
n̂·~
v tr
i
E(ω) = e ~
n̂ × (n̂ × β) eiω[tr − c/n ] dtr (596)
4π0 cR −∞ ret

12.5.1 Radiation of a particle moving slower than the speed of light

Let’s take a step back from this and assume a particle moving with constant
v < cn = c/n, but we stil have that ~r0 (tr ) = ~v tr . The power emitted per
frequency per solid angle is given by:
Z ∞ 2
d2 P q2 ω2 
~ v ·n̂/(c/n))
iωtr (1−~

= 2
 (n̂ × (n̂ × β))e dtr  (597)
dΩdω 16π 0 (c/n) −∞


This integral oscillates for values of ω. This may seem intimidating, but recall
that the integral of a complex exponential traversed around the clock will simply

86
return zero. In other words:
Z ∞
eiωt dt = 2πδ(ω) (598)
−∞

This yields only a non-zero contribution for ω = 0, in which case you just
calculate the circumference of a circle (hence the 2π). In other words, there’s
no radiation!

12.5.2 Radiation of a particle moving faster than the speed of light

Let’s now assume we have a charged particle moving with constant v > cn =
c/n. The radiation then becomes:
Z ∞ 2
d2 P q2 ω2 
~ v ·n̂/(c/n))
iωtr (1−~

=  (n̂ × (n̂ × β))e dtr  (599)
dΩdω 16π 2 0 (c/n)  −∞

That exponential is now more interesting. The Mach cone angle was given by:

c/n
sin θc = (600)
v
So we now have that:
~v · n̂
1− =0 (601)
c/n
This is always true when you’re on the Mach cone, which is the ’volume’ of
radiation. This integral therefore does not oscillate and could give a non-zero
contribution for ω 6= 0! In other words, now we do have radiation from our
velocity field. To evaluate the power emitted we look at what has been emitted
in a finite slab:
Z r 2
d2 P q2 ω2  v ·n̂/(c/n))
iωtr (1−~

= 2
 v cos θe dtr  (602)
dΩdω 16π 0 (c/n) −r


Evaluating this integral gives something horrible:

2
d2 P q 2 ω 2 v 2 cos2 θ 2 2

sin (ωτ (1 − v sin θ/(c/n))/2)
= ω τ ωtr (603)
dΩdω 16π 2 0 c2 (c/n) ωτ (1 − v sin θ/(c/n))/2

If you look at the term between the square brackets, you see that it’s actually
a sinc-function:  2
sin x
f (x) = (604)
x
The peaks are at sin θc = c/n
v . Now what makes Cherenkov radiation blue lies
in the cosine squared term before the square brackets.
2
d2 P

c
∝ cos2 θ ≈ 1 − sin2 θc = 1 − (605)
dΩdω n(ω)c

87
This is true because:
c/n(ω)
sin θC = (606)
v
The refractive index is dependent on the frequency of the light (that is, its
color). Waters refractive index has a complicated form with some absorption
bands (as seen in figure 12.5.2), but in the visible region it has a nice increasing
function. As a consequence, the Cherenkov radiation intensity increases with
frequency, making blue light the strongest form of radiation. The take home

[]

Figure 9: Refractive index of water. Note the increasing value of the refractive
index in the transparent region.

messages of Cherenkov radiation are as follows:

• Cherenkov radiation is radiation of a charged point particle moving faster

than the speed of light in the medium v > c/n.
• The angle of the Cherenkov angle/cone, the counterpart to the Mach cone
of supersonic particles, is dependent on the emitted frequency
• Emitted power is in a broad frequency spectrum, but the refractive index
of water increases with frequency in the visible region before approaching
an absorption band.

88
13 Relativistic motion in EM-fields
13.1 Proper time

[]

Figure 10: A Minkowski diagram where past and future light cones separate
space-time into space-like intervals and time-like intervals with respect to an
event O at the origin. The dashed curve is the ”world line” of a particle moving
with non-uniform velocity.

A relativistic (or Lorentz) invariant quantity takes the same numerical value
in every inertial frame. Invariants play a special role in relativity, beginning
with Einstein’s postualte that the speed of light is a relativistic invariant, and
so is electric charge. Another relativistic quantity is the interval. The square of
the interval between two events is defined as:
(∆s)2 = (∆x)2 + (∆y)2 + (∆z)2 − (c∆t)2 (607)
p
This interval combines a distance in space d = (∆x)2 + (∆y)2 + (∆z)2 , with
a distance (or lapse) in time, ∆t, into a single quantity. Like d, the interval is
invariant to rotations and translations in space.

The proper time is an invariant measure of the motion of a particle along

its trajectory in space-time. A definition for proper time follows naturally if we
define the ”world line” of a particle as the locus of points in space-time which
describes the trajectory in question. The dashed curve in 13.1 is a typical world
line for a particle with non-uniform velocity u(t) = dr/dt. All world lines lie
inside the light cone because the particle speed is always less than the speed of
light. Focus now on the interval between two points on the world line which lie
infinitesimally close to each other. This is the time-like quantity:
u2 (t)
 
2 2 2 2
(ds) = (dr) − (cdt) = −(cdt) 1 − 2 (608)
c

89
Dividing this by the speed of light produces another invariant. This leads us to
define a differential element of the invariant proper time in an inertial frame K
as r r
(ds)2 u2 (t) dt
dτ = − 2 = 1 − 2 dt = (609)
c c γ(u)
The invariance of dτ means that this expression has the same numerical value
in any other inertil frame K 0 where u0 6= u, t0 6= t and dt0 6= dt. This fact will
provide a natural way to define Lorentz invariant time derivatives.

13.2 Relativistic motion of particles in electric fields

13.2.1 Linear acceleration in constant E-field with v k E
Consider a particle initially at rest, but subject to a static electric field, also
pointing in the x̂-direction. We’re interested in finding its velocity u as a function
of time t. We’d start with the classical approach. The electric field is given by:

E = E0 x̂ (610)

The force on the particle is given by:

dp
F= = qE0 x̂ (611)
dt
In the relativistic approach however, the force is defined as:

p = γ(u)m0 u → qE0 = a0 (612)

The quantity a0 is the acceleration the particle would experience in the classical
case. In the relativistic case, the acceleration is found as follows:
d dγ du
(γ(u)u) = a0 → u + γ(u) (613)
dt dt dt
The factor γ(u) depends on the time via its velocity u, though not explicitly so.
To circumvent this, we use the partial derivative instead:
dγ du du u/c2 du 1 du
u + γ(u) = 3/2
u+ q = a0 (614)
dt dt dt 2
1 − vc2

dt 1− u2 dt
c2

We can add the two fractions together by rewriting the second term:
u2 u2
1 1− c2 1− c2
q · u2
=
32 (615)
1− u2 1− c2 1− u2
c2 c2

So putting them together gives:

2
u2 /c2 du 1 − uc2 du 1 du
+ 3 = 3 = a0 (616)
v 2 3/2 dt 2
2 dt 2
2 dt

u u
1 − c2 1 − c2 1 − c2

90
Figure 11: Velocity of a particle that’s initially at rest in a static electric field.

Figure 12: Momentum of a particle that’s initially at rest in a static electric

field.

This can be split into:

1
=
32 du = a0 dt (617)
u2
1− c2
We integrate this to obtain:
u
q = a0 t (618)
u2
1− c2

We want to have a function for the velocity, so with some algebra we rewrite
this to:
a0 t
u(t) = q
2 (619)
1 + ac0 t

91
Figure 13: Energy of a particle that’s initially at rest in a static electric field.

In classical mechanics, the momentum is defined as:

dx
p=m (620)
dt
Now we look at momentum. Momentum is the time derivative of the force, and
the force particle experiences is the Lorentz-Coulomb force:

FL = q(E + u · B) (621)

The time derivative of momentum is the force, so:

dp
= qE0 x̂ (622)
dt
d qE0
(γ(u)u) = = a0 (623)
dt m0
The time derivative we see here is the one we had just calculated. So using our
previous result, we obtain that:

p(t) = γ(u)m0 u(t) = m0 a0 t (624)

The relativistic total energy of a particle is given as:

s  2
2 a0 t
E(t) = γ(u)m0 c = 1 + · m0 c2 (625)
c

The kinetic energy is the difference between the total energy and the rest mass
energy:
Ek = γ(u)m0 c2 − m0 c2 (626)

92
13.2.2 Relativistic motion of particles in electric fields with v ⊥ E
We now assume that the particle has an initial velocity in the y-direction, but
then encounters a static electric field in the x-direction. So let’s say the particle
has an initial velocity
u0 = u0 ŷ (627)
And we want to know u(t), when it encounters a field of which we can say that
u0 ⊥ E (still the same field, so E = E0 x̂ and u = uŷ. All physics we just used is
still true - the Coulomb force still pushes the particle to the right, so we obtain
that:
a0 t
ux = q
2 (628)
1 + ac0 t
The total velocity of the particle is added together via the Pythagorean theorem:
q
u = u2x + u2y (629)

There’s ofcourse a maximum velocity, which is the speed of light. There-

fore, as the total velocity approaches the speed of light, another velocity must
decrease. So classically speaking, where v  c, we find:

ux (t) = a0 t
uy (t) = u0

In the relativistic limit, where v ≈ c, we have that:

d
[γ(u)ux (t)] = a0
dt
d
[γ(u)uy (t)] = 0
dt
So the x-component of the velocity will continue to grow, but that can only be
the case if the velocity in the y direction shrinks. So even though the acceleration
in the y-direction is zero, it will nonetheless be ’contracted’ and eventually,
vanish. The velocities are:
a0 t/γ(u0 )
ux (t) = r  2 (630)
a0 t
1 + γ(u 0 )c

and
u0
uy (t) = r  2 (631)
a0 t
1+ γ(u0 )c

The momentum in the y-direction is non-zero, and will remain non-zero, even
though the velocity does decrease. The reason for that is because there’s no
force acting in the y-direction.

93
Figure 14: Velocity in both cardinal direction of a particle that initially trav-
els in the x-direction, but then encounters at t = 0 an electric field oriented
transversally in the y-direction.

13.3 4-vectors and relativistic mechanics

13.3.1 Galilean transformation
Let’s assume there’s a particle in frame K (inertial frame 1) that is moving in
the z-direction with velocity v, so v = vẑ. A force on such a particle is then
described in Newtonian mechanics via
d2 r
F = ma = m (632)
dt2
Now, let’s assume a particle in frame K’(inertial frame 2). This inertial frame
is moving with velocity v = vẑ. The position of the particle r0 is then described
as follows:
r0 = r − vt (633)
We can accompany for the moving of the frame by using the Galilean transfor-
mation:

x0 = x
y0 = y
z 0 = z − vt
t0 = t

The Newtonian force is then given by:

d2 r0
F0 = m =F (634)
d(t0 )2

So we see that the force is invariant under transformation of inertial frames.

However, the Maxwell equations are not invariant if we use this transformation,
because the Lorentz force depends on the coordinate system because of the
velocity dependence.

94
13.3.2 Lorentz transformation
Consider a particle in stationary frame K (inertial frame 1) that is moving with
velocity v = vẑ. It’s position is described by:

r(x, y, z, t) (635)

Now let’s consider the same particle, but in a moving frame K’ (inertial frame
2). We now subject our coordinate system to the Lorentz transformation, which
is as follows:

x0 = x
y0 = y
0
z = γ(v)(z − vt)
0
t = γ(v)(t − vz/c2 )

With
1
γ(v) = q (636)
v2
1− c2

The Maxwell equations are invariant in this transformation (though the fields
you see are different), but the Newtonian force formulation is not invariant under
a Lorentz transformation:
d2 r0
F0 = m 0 2 6= F (637)
d(t )
Einstein formulated in 1905 that the Maxwell equations M1 to M4 are valid in
any moving coordinate system. The consequence of this are as follows:
1. The velocity of light is the same in any moving coordinate system.
2. The two inertial frames we just treated are mathematically different, but
physically they must be the same
3. Therefore, Newton’s law is wrong: F 6= ma.
4. Instead we use relativistic mechanics, so that both inertial frames do be-
come one and the same
Let’s detour for a moment for a small historical note, so we can see what brought
this change in vision about. Let’s consider case A, a particle at rest. Here we
have that:

ρ(r, t) = ρ0 (r)
j(r, t) = 0

Using the Maxwell equations, we then obtain the following fields:

E(r)
B(r, t) = 0

95
This is an ordinary Coulombic field. Let’s now consider case B, where we look
at a particle moving with constant velocity. This gives us a complication solu-
tion with a non-zero current, a non-zero electric field and a non-zero magnetic
field. This doesn’t follow from Coulomb’s law. Everybody realized that case B
looks awfully different from A. So too, does the Lorentz force. So instead of
using the Maxwell equations to derive the fields in both cases, we use Lorentz
transformations and Four-vectors.

13.4 Four-vectors
We’re now going to approach the fields using our relativistic theories. This
means we’re going to use Minkowski space, where we treat time and space
equivalently (the former becomes ’space-time’) and describe them using 4D vec-
tors, so called four-vectors. Let’s again consider the case of a particle moving
with constant velocity v = vẑ. Its coordinates in stationary frame K are given
by (r, t).

If we now consider the moving frame K’, which is moving with velocity v = vẑ,
the particle appears at rest. Its coordinates are described by (r0 , t0 ). We want
to transform our coordinates from the stationary frame K (r, t) to the moving
frame K’ (r0 , t0 ). We use the Lorentz transformation for this:

x0 = x
y0 = y
z 0 = γ(v)(z − vt)
t0 = γ(v)(t − vz/c2 )

With
1
γ(v) = q (638)
v2
1− c2

Except now we’re going to Minkowski space. Our position vector for a moving
particle in a stationary frame now becomes

rµ = (r, ict) (639)

We transform it to a stationary particle in a moving frame described by:

rµ0 = (r0 , ict0 ) (640)

The Lorentz transformation still uses:

1
γ(v) = q (641)
v2
1− c2

96
Furthermore, we let β = v/c. The transformation is then carried out via the
following tensor:
 0    
x 1 0 0 0 x
 y 0  0 1 0 0  y
 0 = =  (642)
 z  0 0 γ iβγ   z 
0
ict 0 0 −iβγ γ ict

The magnitude of the four-vector is a conserved quantity. The magnitude of

the four-vector is defined as:
X
rµ · rµ = x2 + y 2 + z 2 + (ict)2 = x2 + y 2 + z 2 − (ct)2 (643)
µ

Note that in Minkowski space, we do not take the complex conjugate of the
vector for our definition of the inner product, but leave the vector as it is. The
transformed four-vector is:
x0 = x
y0 = y
0 (644)
z = γz − βγct
ict0 = −iβγz + γict

Its magnitude is then:

X
rµ0 · rµ0 = x02 + y 02 + z 02 − (ct0 )2 (645)
µ

Substituting the identities for the coordinates, we obtain that:

= x2 + y 2 + (γz − βγct)2 + (−iβγz + γict)2 (646)

That the x0 and y 0 are conserved, can be seen by simple comparison. To see
that the transformed four-vectors magnitude is still the same as the four-vector
in the first reference frame, we write out the squares and use the fact that
1 − β 2 = γ −2 :

(γz − βγct)2 + (−iβγz + γict)2 = γ 2 z 2 − 2βγczt + β 2 γ 2 c2 t2 + −β 2 γ 2 z 2 + 2βγ 2 czt − γ 2 c2 t2

= γ 2 z 2 + β 2 γ 2 c2 t2 − β 2 γ 2 z 2 − γ 2 c2 t2
= γ 2 z 2 + β 2 (ct)2 − β 2 z 2 − (ct)2

= γ 2 z 2 1 − β 2 + (ct)2 β 2 − 1

= z 2 γ 2 1 − β 2 + γ 2 (ct)2 β 2 − 1

= z 2 γ 2 1 − β 2 − γ 2 (ct)2 1 − β 2

= z 2 − (ct)2
(647)

And so the magnitude of the vector rµ is preserved

97
13.4.1 Velocity four-vector
It might be tempting to define the velocity four vector in much the same manner
as we had defined the position-four vector, so as uµ = (~u, ict). This however,
is wrong, because the time derivative of position in the particle’s own time will
give us a factor of γ(u). So the correct velocity four-vector is uµ = γ(~u, ict).
The magnitude of the velocity four vector is:
X u · u − c2
uµ · uµ = u · u + U42 = = −c2 (648)
µ
1 − u2 /c2

This is sensible because we can always find an inertial frame where u=0.

13.4.2 Potential field four vectors

The vector and scalar potential are grouped together in a four vector as well.
We define the four-potential as:
 
φ
Aµ = A, i (649)
c

If this is the potential of a particle moving with constant velocity v = vẑ in

frame K, then we can also look at the particle at rest in frame K’, where the
frame itself is moving along with the particle. The transformation is identical
to the the one for position:
 0    
Ax 1 0 0 0 Ax
 A0y  0 1 0 0   Ay 
 0= =  (650)
 Az  0 0 γ iβγ  Az 
0
i φc 0 0 −iβγ γ i φc

13.4.3 Four-vectors: derivatives and the Lorenz Gauge

The Lorenz gauge is a relativistic gauge and is given by:
1 ∂φ
∇·A+ =0 (651)
c2 ∂t
To solve this four our four-vectors, we need a ”four-derivative”. This is given
by:  
∂
∂µ = ∇, (652)
∂(ict)
with µ = 1, 2, 3, 4. It’s then easy to verify that:
4
X 1 ∂φ
∂ µ Aµ = ∇ · A + =0 (653)
µ=1
c2 ∂t

98
Ergo, the relativistic formulation of the Lorenz gauge is as follows:
4
X
∂µ Aµ = 0 (654)
µ=1

The Lorenz gauge is valid in all inertial frames, which is why we’ll use it for our
relativistic purposes.

13.4.4 Four vectors: Charge conservation

Charge conservation in classical electromagnetism is defined as:
∂ρ
∇·j+ =0 (655)
∂t
In relativistic mechanics, we use the ’four-derivative’ and ’four-current’:

jµ = (j, icρ) (656)

Combining this with the ’four-derivative’:

∂µ = 0 (657)

We formulate our charge conservation as follows:

4
X ∂ρ
∂µ jµ = ∇ · j + (658)
µ=1
∂t

Or, more concisely:

4
X
∂ µ jµ = 0 (659)
µ=1

Charge conservation is valid in all inertial frames.

13.4.5 Four-vectors: Maxwell equations (in terms of potential fields)

The Maxwell equations in terms of the potential fields are:

1 ∂2
 
2
∇ − 2 2 A(r, t) = −µ0 j(r, t) (660)
c ∂t

1 ∂2
 
∇2 − 2 2 φ(r, t) = −ρ(r, t)/0 (661)
c ∂t
This is equal to the following relativistic formulation:

1 ∂2
 
∇2 − 2 2 Aµ (r, t) = −µ0 j(r, t) (662)
c ∂t

99
It’s then easy to verify that:
4
X 1 ∂2
∂α ∂α = ∇ · ∇ − (663)
α=1
c2 ∂t2

This allows then for the formulation of the Maxwell equations in one single line:
4
!
X
∂α ∂α Aµ = −µ0 jµ (664)
α=1

This formulation of the Maxwell equations is valid in all inertial frames.

13.5 Electromagnetic field tensor

Now that we have a relativistic formulation of the Maxwell equations in terms of
the potentials, we can also define our electric and magnetic fields. The electric
field is defined classically as:
∂A(r, t)
E(r, t) = −∇φ(r, t) − (665)
∂t
B(r, t) = ∇ × A(r, t) (666)
To define each component of the electric field, we need to have derivatives and
potentials. For example, the x̂-component in the relativistic notation is:
∂φ ∂Ax
Ex = − − = i∂1 (cA4 ) − icδ4 A1 (667)
∂x ∂t
Similarly, the magnetic field’s component in the x̂ direction is given by:
∂Az ∂Ay
Bx = − = ∂2 A3 − ∂3 A2 (668)
∂y ∂z
There’s a pattern here in the relativistic notation of the electric field and mag-
netic field. We group them together in the electromagnetic field tensor:

Fαβ (r, t) = ∂α Aβ (r, t) − ∂β Aα (r, t) (669)

This is a 4x4 antisymmetric matrix. Antisymmetric means that that if you

transpose the matrix, you end up with the same matrix but times −1. Writing
out the electromagnetic field tensor:
 
0 Bz −By −iEx /c
 −Bz 0 Bx −iEy /c
Fαβ =   (670)
 By −Bx 0 −iEz /c
iEx /c iEy /c iEz /c 0
Important to remember: the electric and magnetic fields are elements of a 4x4
tensor. They are not the elements of a four-vector. There’s no ’electric four-
vector’, no ’magnetic four-vector’ or ’electromagnetic four-vector’. We’re dealing

100
here with a second order tensor.
This is a second order tensor and is dependent on the inertial frame we choose.
Because the tensor is second order, we need to multiply it with two transfor-
mation matrices if we want the tensor in a different inertial frame. We do so as
follows:
0
Fαβ = L(v)Fαβ LT (v) (671)
Using the antisymmetric property of the transformation matrix, we can also
write this also as:
0
Fαβ = L(v)Fαβ LT (−v) (672)
The transformation matrix L(v) for the aforementioned cases where in frame 1
the particle is moving with v = vẑ, where as in frame 2 the frame is moving
with v = vẑ. The transformation from frame 1 to frame 2 can be written out
in linear algebraic formulation as follows:
 
1 0 0 0
0 1 0 0 
L(T )(v) = 
0 0
 (673)
γ iβγ 
0 0 −iβγ γ
Let’s do an example for transforming an electromagnetic field tensor from one
stationary frame to a moving frame, where the moving frame is moving with
v = vẑ. The EM-tensor for a magnetic field that’s pointing in the ẑ, so B = B0 ẑ,
in a not-moving frame, is given as:
 
0 B0 0 0
−B0 0 0 0
Fαβ =  0
 (674)
0 0 0
0 0 0 0
We transform to the moving frame as follows:
0
Fαβ = L(v)Fαβ L(−v) (675)
     
1 0 0 0 0 B0 0 0 1 0 0 0 0 B0 0 0
 −B0 0 0 0 0 1  = −B0 0
0 1 0 0    0 0   0 0
 
0 0 γ iβγ   0 0 0 0 0 0 γ −iβγ   0 0 0 0
0 0 −iβγ γ 0 0 0 0 0 0 iβγ γ 0 0 0 0
(676)
Therefore, B0 = B0 ẑ is unchanged for this case. Let’s flip the magnetic such
that we now have B = B0 x̂.
0
Fαβ = L(v)Fαβ L(−v) (677)
     
1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
0
 1 0 0 
0 0 B0 0 
0 1 0 0  0
= 0 γB0 −iβγB0 

0 0 γ iβγ  0 −B0 0 0 0 0 γ −iβγ  0 −γB0 0 0 
0 0 −iβγ γ 0 0 0 0 0 0 iβγ γ 0 iβγB0 0 0
(678)

101
The magnetic field that we now have is B0 = γB0 x̂. This is an enhanced
magnetic field. Furthermore, we also see that in this inertial frame there’s a
non-zero electric field E0 = γβB0 ŷ.

13.6 Take home messages

The relativistic equation of motion has the expression for momentum changed:

p = γ(u)m0 u(t) (679)

The Lorentz force then becomes:

dp
= FL = qE(r, t) + qu(t) × B(r, t) (680)
dt
Where
1
γ(u) = q (681)
v2
1− c2

We also introduced the Lorentz transformations and the use of space-time

coordinates, giving us the quantities of four-vector coordinates. We found
momentum-energy via the four-momentum vector and defined our potential
fields (vector-scalar) via four-vector potentials. We also found a Lorentz trans-
formation invariant formulation of the Maxwell equations, and saw that the
electromagnetic field was part of a 4x4 electromagnetic field tensor.

102
14 Relativistic equation of motion
NOTE: the first subsections here are simply recaps of the previous lecture. They
are included here for completeness’ sake. NOTE 2: the lecture ends with a 30
minute discussion on the limits of classical electrodynamics. This isn’t part of
the final exam, so it’s omitted here.

14.1 Four-vectors: derivatives, Lorenz gauge and charge

conservation.
We use the Lorenz gauge, which says that:
1 ∂φ
∇·A+ =0 (682)
c2 ∂t
We define the four-derivative:
∂
∂µ = (∇, (683)
∂(ict)
Where µ = 1, 2, 3, 4. It’s then easy to verify that:
4
X 1 ∂φ
∂ µ Aµ = ∇ · A + =0 (684)
µ=1
c2 ∂t

This is the relativistic notation of the Lorenz gauge and it has the nice property
that it’s valid in all inertial frames. This is not true of the Coulomb gauge.
Charge conservation in classical electromagnetism is defined as:
∂ρ
∇·j+ =0 (685)
∂t
In relativistic mechanics, we use the ’four-derivative’ and ’four-current’:

jµ = (j, icρ) (686)

Combining this with the ’four-derivative’:

∂µ = 0 (687)

We formulate our charge conservation as follows:

4
X ∂ρ
∂µ jµ = ∇ · j + (688)
µ=1
∂t

Or, more concisely:

4
X
∂ µ jµ = 0 (689)
µ=1

Charge conservation is valid in all inertial frames.

103
14.2 Relativistic form of Maxwell (in terms of potential
fields)
The Maxwell equations in terms of the potential fields are:

1 ∂2
 
∇2 − 2 2 A(r, t) = −µ0 j(r, t) (690)
c ∂t

1 ∂2
 
∇2 − 2 2 φ(r, t) = −ρ(r, t)/0 (691)
c ∂t
This is equal to the following relativistic formulation:

1 ∂2
 
2
∇ − 2 2 Aµ (r, t) = −µ0 j(r, t) (692)
c ∂t

It’s then easy to verify that:

4
X 1 ∂2
∂α ∂α = ∇ · ∇ − (693)
α=1
c2 ∂t2

This allows then for the formulation of the Maxwell equations in one single line:
4
!
X
∂α ∂α Aµ = −µ0 jµ (694)
α=1

This formulation of the Maxwell equations is valid in all inertial frames.

14.3 Electromagnetic field tensor

Now that we have a relativistic formulation of the Maxwell equations in terms of
the potentials, we can also define our electric and magnetic fields. The electric
field is defined classically as:

∂A(r, t)
E(r, t) = −∇φ(r, t) − (695)
∂t
B(r, t) = ∇ × A(r, t) (696)
To define each component of the electric field, we need to have derivatives and
potentials. For example, the x̂-component in the relativistic notation is:
∂φ ∂Ax
Ex = − − = i∂1 (cA4 ) − icδ4 A1 (697)
∂x ∂t
Similarly, the magnetic field’s component in the x̂ direction is given by:
∂Az ∂Ay
Bx = − = ∂2 A3 − ∂3 A2 (698)
∂y ∂z

104
There’s a pattern here in the relativistic notation of the electric field and mag-
netic field. We group them together in the electromagnetic field tensor:
Fαβ (r, t) = ∂α Aβ (r, t) − ∂β Aα (r, t) (699)
This is a 4x4 antisymmetric matrix. Antisymmetric means that that if you
transpose the matrix, you end up with the same matrix but times −1. Writing
out the electromagnetic field tensor:
 
0 Bz −By −iEx /c
 −Bz 0 Bx −iEy /c
Fαβ =   (700)
 By −Bx 0 −iEz /c
iEx /c iEy /c iEz /c 0
Important to remember: the electric and magnetic fields are elements of a 4x4
tensor. They are not the elements of a four-vector. There’s no ’electric four-
vector’, no ’magnetic four-vector’ or ’electromagnetic four-vector’. We’re dealing
here with a second order tensor.
This is a second order tensor and is dependent on the inertial frame we choose.
Because the tensor is second order, we need to multiply it with two transfor-
mation matrices if we want the tensor in a different inertial frame. We do so as
follows:
0
Fαβ = L(v)Fαβ LT (v) (701)
Using the antisymmetric property of the transformation matrix, we can also
write this also as:
0
Fαβ = L(v)Fαβ LT (−v) (702)
The transformation matrix L(v) for the aforementioned cases where in frame 1
the particle is moving with v = vẑ, where as in frame 2 the frame is moving
with v = vẑ. The transformation from frame 1 to frame 2 can be written out
in linear algebraic formulation as follows:
 
1 0 0 0
0 1 0 0 
L(T )(v) = 
0 0
 (703)
γ iβγ 
0 0 −iβγ γ
Let’s do an example for transforming an electromagnetic field tensor from one
stationary frame to a moving frame, where the moving frame is moving with
v = vẑ. The EM-tensor for a magnetic field that’s pointing in the ẑ, so B = B0 ẑ,
in a not-moving frame, is given as:
 
0 B0 0 0
−B0 0 0 0
Fαβ =  0
 (704)
0 0 0
0 0 0 0
We transform to the moving frame as follows:
0
Fαβ = L(v)Fαβ L(−v) (705)

105
     
1 0 0 0 0 B0 0 0 1 0 0 0 0 B0 0 0
0 1 0 0  −B0 0 0 0 0 1 0 0  −B0 0 0 0
   = 
0 0 γ iβγ   0 0 0 0 0 0 −iβγ   0
γ 0 0 0
0 0 −iβγ γ 0 0 0 0 0 0 iβγγ 0 0 0 0
(706)
Therefore, B0 = B0 ẑ is unchanged for this case. Let’s flip the magnetic such
that we now have B = B0 x̂.
0
Fαβ = L(v)Fαβ L(−v) (707)
     
1 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0
0
 1 0 0 
0 0 B0 0

0 1 0 0  0
= 0 γB0 −iβγB0 

0 0 γ iβγ  0 −B0 0 0 0 0 γ −iβγ  0−γB0 0 0 
0 0 −iβγ γ 0 0 0 0 0 0 iβγ γ 0iβγB0 0 0
(708)
The magnetic field that we now have is B0 = γB0 x̂. This is an enhanced
magnetic field. Furthermore, we also see that in this inertial frame there’s a
non-zero electric field E0 = γβB0 ŷ.

14.4 Particle in an E × B field

Let the electric field point in positive x̂ and the magnetic field point in ŷ, so
that the two fields are perpendicular. The charged particle is initially at rest
in the origin. The electric field draws the particle up, which creates a current.
This current is then subject to a Lorentz force, which pushes it initially in the
positive ẑ direction. The particle then returns to a ’height’ of x = 0, after
which the cycle starts anew. This is a cycloic trajectory. The particle moves
on a circle, whose center (R, 0, vt) travels in the ẑ-direction at a constant speed
v = ωR. The cycloid trajectory is as follows:
E E E
z(t) = ωB(ωt − sin ωt) = B t − ωB sin ωt
E (709)
x(t) = ωB (1 − cos ωt)

The velocity of the particle is a constant:

|E|
v= (710)
|B|

The frequency of this circular movement is:

qB
ω= (711)
m

14.5 Particle in a B-field

An observer O sees the magnetic field

B = B0 ŷ (712)

106
And no electric field. The particle has velocity u0 = −vẑ. This will lead to a
movement in a circle with frequency:
qB
ω= (713)
m
An observer O’ is moving with velocity u = −vẑ, so he sees an initial velocity
for the particle of u00 (t0 = 0). The magnetic and electric field he sees are now:
B0 = γB0 ŷ
(714)
E0 = γvB0 x̂
In this case, there is an electric field, so this observer sees a cycloid motion as
we had shown in the previous subsection. This particle is moving, again, with
the velocity:
|E|
v= (715)
|B|
Its frequency in the frame of observer O0 is now:
qB 0
ω0 = = γ(v)ω (716)
m

14.6 Relativistic motion of particles in electric fields

Let’s now assume a static electric field and a particle initially at rest (u0 = 0).
The electric field is E = E0 ẑ. The force on the particle is
FL = qE (717)
The force is equal to the change in momentum, so we can also say that:
dp d
= qE0 ẑ → [γ(u)m0 u] = qE0 (718)
dt dt
So the momentum is given by:
p = γ(u)m0 u (719)
The velocity of the particle is given by:
a0 t
uz (t) = q (720)
a0 t 2


1+ c

The Lorentz prefactor γ is given by

s  2
a0 t
γ(u) = 1+ (721)
c
Where the initial acceleration is given by:
qE0
a0 = (722)
m0

107
Close inspection shows two intuitive results. For low velocities γ ≈ 1, so we
obtain that u = a0 t, as we’d expect from Newtonian mechanics. However, as
the velocity becomes relativistic, we’d expect the velocity to ’flatten out’ as it
approaches the velocity of light. The γ factor does exactly this.

14.7 Relativistic equation of motion

We want to do these calculations we just did in a more general way using our
four-vectors. To do so, we first define the Lorentz force and equation of motion
as follows:
dp
= q(E + u × B) = FL (723)
dt
The four-momentum is given by:
 
p(t)
pα (t) = m0 Uα (t) = (724)
iε/c
Where ε is the total energy. The equation of motion now becomes:
   
d d p(t) F
pα (t) = = FLL·u (725)
dt dt iε/c i c
We rewrite the right hand side to:
P
q β Fαβ Uβ (t)
 
FL
= (726)
i FLc·u γ(u)
So the equation of motion now looks like this:
d X
γ(u)m0 Uα (t) = q Fαβ Uβ (t) (727)
dt
β

The left hand side has a four-vector in it, whereas the right hand side has a
matrix times a vector in it. So in vector notation this becomes:
   
U1 (t) U1 (t)
d U2 (t) = qF U2 (t)
 
γ(u)m0  (728)
dt U3 (t)
   U3 (t)
U4 (t) U4 (t)
With F being the electromagnetic field tensor
 
0 Bz −By −iEx /c
 −Bz 0 Bx −iEy /c
F=  (729)
 By −Bx 0 −iEz /c
iEx /c iEy /c iEz /c 0
and    
U1 (t) ux (t)
U2 (t) uy (t)
U3 (t) = γ(u) uz (t) (730)
   

U4 (t) ic
being the Four-velocity. This is all we need to solve these problems.

108
14.8 Solving relativistic motion of particles in electric fields
using relativistic notation
Let’s apply this to the static electric field pointing in ẑ with the particle initially
at rest. We’ve already found the equation of motion classically, so let’s now use
the Einstein way. We simply fill in what we know in the relativistic notation of
the equation of motion:
    
U1 (t) 0 0 0 0 U1 (t)
d U2 (t)  = q 0 0
 0 0   U2 (t)
 
γ(u)m0  (731)
dt U3 (t) 0 0 0 −iE0 /c U3 (t)
U4 (t) 0 0 iE0 /c 0 U4 (t)
Calculating this isn’t a straightforward differential equation because we have
that γ(u) in front of our four-velocity, because we’re in the frame K, where
we sit still and watch the particle move away. It can be solved this way, but
this gives a set of coupled differential equations, something both humans and
computers have difficulty dealing with. Fortunatuly, there’s another way, and
that’s by transforming to the particle’s own time. Since the particle is moving,
it has a different time than us observers. So we assume there’s some relation
t(τ ) between our time t and the particle’s own time τ . To transform from the
stationary frame to the moving frame we would normally use a Lorentz trans-
formation, which will give us t(τ ). We find this relationship via a differential
equation for the fourth component of the Four-Velocity (the time-component).
From time dilation, the differentials in coordinate time t and proper time τ are
related by
dt = γ(u)dτ (732)
The time derivative in our frame is rewritten via the product rule to the particle’s
frame as follows
d dτ d 1 d
= = (733)
dt dt dτ γ(u) dτ
This allows us to rewrite the time derivative here in terms of τ , which we want
because it allows us to get rid of the γ(u). The equation of motion then becomes:
    
U1 (τ ) 0 0 0 0 U1 (τ )
d U2 (τ ) = q 0 0
  0 0   U2 (τ )
 
m0 (734)
dτ  U3 (τ )  0 0 0 −iE 0 /c  U 3 (τ )
U4 (τ ) 0 0 iE0 /c 0 U4 (τ )
    
U1 (τ ) 0 0 0 0 U1 (τ )
U2 (τ ) = qE0 0 0 0 0  U2 (τ )
d     
(735)
dτ U3 (τ )
  m0 c 0 0 0 −i
   U3 (τ )
U4 (τ ) 0 0 i 0 U4 (τ )
If we take a step back and disregard the fact that the quantities here are matrices
and vectors, we see a differential equation of the form of:

f 0 (τ ) = af (τ ) (736)

109
Its solution is simply:
f (τ ) = f (0)eaτ (737)
It’s no different now, except now we have a matrix as our factor a
    
U1 (τ ) 0
qE0 τ 0
0
0
0
0
0
0 
U1 (0)
U2 (τ ) m0 c 0 0 0 −i U2 (0)
 =e 0 0 i 0   (738)
U3 (τ ) U3 (0)
U4 (τ ) U4 (0)
With    
U1 (0) 0
U2 (0)  0 
U3 (0) =  0  (739)
   

U4 (0) ic
This works - take the derivative of the exponential, and the exponent itself
drops down in front, giving back our differential equation. However, what do
we mean with a matrix in an exponent? Before we tackle that question, we
realize that the first and second component are zero anyway, because there’s no
movement before or after in that direction. This would’ve been different if the
particle didn’t move in only one direction! So U1 (τ ) = U2 (τ ) = 0. This helps
us because now we don’t have to work with this 4x4 matrix, but can confine us
to the lower-right 2x2 matrix, that is, we only have to look at U3 (τ ) and U4 (τ ).
Focussing on that, the differential equation becomes:
    
U3 (τ ) qE0 τ 0 −i
U3 (0)
=e m0 c i 0
(740)
U4 (τ ) U4 (0)
We will now tackle the matrix in the exponent using the Taylor expansion of
an exponential. The Taylor expansion for an exponential is:
∞
X 1 n n
eαx = α x (741)
n=0
n!

Because the matrix is very simple of form and only a 2x2, we can easily calculate
its higher powers. We recognize that:
qE0 τ
α= (742)
m0 c
And the variable we expand is:
 
0 −i
x= (743)
i 0
Taking this to the power 0 will return the unity matrix. This to the power 1
will return the normal matrix, and taking its square returns the unity matrix
again.     
0 −i 0 −i 1 0
= (744)
i 0 i 0 0 1

110
This pattern repeats itself. All odd powers are the matrix itself, all its even
powers are the unity matrix:
         
α 0 −i 1 2 1 1 0 1 4 1
 
0
α i
−i
1 0 0 −i 0
e 0
= + + α + + α +...
0 1 1 i 0 2 0 1 6 i 0 1
24 0
(745)
We can sum this pretty easily. If you look at the top-left component, you have
1 4
1 + 0 + 21 α2 + 0 + 24 α .... You only get the even terms here - in fact, close
inspection reveals that this is the Taylor expansion of the cosh function (not a
cosine - than you’d have a plus-minus-plus pattern of your power, now you only
have positive powers). So we get:
1 1
1 + 0 + α2 + 0 + α4 ... = cosh α (746)
2 24
Summing the top right elements, we obtain:
1
− i(α + α3 + ...) = −i sinh α (747)
6
We get the sine hyperbolic function (times −i)! Repeating for the bottom left,
we get:
1
i(α + α3 + ...) = i sinh α (748)
6
And the bottom right
1 1
1 + α2 + α4 ... = cosh α (749)
2 24
So summing the powers of each element gives us a trigonometric function, mean-
ing we can say that:
   
α i
0 −i
cosh α −i sinh α
e 0
= (750)
i sinh α cosh α

Previously, we’d found that

     
U3 (τ ) qE0 τ 0 −i
U3 (0)
=e m0 c i 0
(751)
U4 (τ ) U4 (0)

which now becomes

!
cosh qE 0τ
−i sinh qE 0τ
  
U3 (τ ) m0 c m0 c U3 (0)
= (752)
U4 (τ ) i sinh qE 0τ
m0 c cosh qE 0τ
m0 c
U4 (0)

The complete 4x4 tensor would be

1 0 0 0
    
U1 (τ ) U1 (0)
U2 (τ ) 0 1 0 0  U2 (0)
U3 (τ ) = 0 0 cosh qE 0τ
−i sinh qE 0τ   (753)
    
m0 c m0 c U3 (0)
U4 (τ ) 0 0 i sinh qE 0τ
m0 c cosh qE 0τ
m0 c
U4 (0)

111
With    
U1 (0) 0
U2 (0)  0 
 =  (754)
U3 (0)  0 
U4 (0) ic
So putting it all again together, for completeness’ sake:

1 0 0 0
    
U1 (τ ) 0
U2 (τ ) 0 1 0 0 0
U3 (τ ) = 0 cosh qE 0τ
−i sinh qE 0τ   (755)
     
0 m0 c m0 c 0
U4 (τ ) 0 0 i sinh qE 0τ
m0 c cosh qE 0τ
m0 c
ic

Let’s interpret this physically. You start with the initial state, which is the four-
vector on the right, you multiply it with the matrix and you have the state of the
particle at any time τ . The only ’interesting’ components of the Four-Velocity
that change are U3 and U4 .

U1 (τ ) = 0
U2 (τ ) = 0
qE0 τ (756)
U3 (τ ) = c sinh
m0 c
qE0 τ
U4 (τ ) = ic cosh (= icγ (u(τ ))
m0 c
The only thing that now remains is for us to go back to our observer frame K
where we work with time t, not with τ . Fortunately, we now have the explicit
relation for the time between our time t and particle time τ ! The time relation
is given by the time component (i.e., the fourth component) of the the Four-
Velocity. Recall that we had stated in the previous lecture that
”...the correct velocity four-vector is uµ = γ(~u, ict). The magnitude of the
velocity four vector is:
X u · u − c2
uµ · uµ = u · u + U42 = = −c2 ” (757)
µ
1 − u2 /c2

Or, in vector notation:

   
U1 (τ ) ux (τ )
U2 (τ ) uy (τ )
U3 (τ ) = γ(u) uz (τ ) (758)
   

U4 (τ ) ic

So the time is simply the fourth component of our Four-Velocity:

qE0 τ
U4 (τ ) = icγ(u(t)) = ic cosh (759)
m0 c

112
 qE0 τ
γ(u(t)) = cosh (760)
m0 c
Now that we have an explicit relation for the time component, we recall that
we had
dt
= γ (u(τ )) (761)
dτ
Filling in what we know, we get:
dt qE0 τ
= cosh (762)
dτ m0 c
This is a simple separable differential equation:
qE0 τ
dt = cosh dτ (763)
m0 c
Integrate left and right:
m0 c qE0 τ
t= sinh (764)
qE0 m0 c
and
qE0 τ qE0 a0 t
sinh = t(τ ) = (765)
m0 c m0 c c
Looking now at the transformed z-coordinate, we have that
U3 (τ ) = γ(u)uz (τ ) (766)
We want to have an explicit relationship for uz (τ ), so we divide by γ(u):

U3 (τ ) c sinh qE 0τ
m0 c c sinh qE0τ
m0 c
uz (τ ) = = = (767)
cosh qE 0τ
q
γ(u) m0 c 1 + sinh2 qE 0τ
m0 c

where we used
cosh2 x − sinh2 x = 1 (768)
That last rewrite is useful, because we the time relation we found was that the
time t was proportional to the sine hyperbolic of τ . This allows us to easily
rewrite uz (τ ) to uz (t) by simple substitution:
qE0
m0 t a0 t
uz (t) = s  qE0 2 = q
2 (769)
m0 t 1 + ac0 t
1+ c

Where a0 = qE m0 . Now we’ve found our velocity in ’real time’ t. This is exactly
0

the same equation of motion has we had found before! So we’ve shown that
this method works, without any solving of coupled differential equations. It’s
all pretty straightforward, but comes at a price: you have to convert the time
t to τ , solve the problem and then convert τ back to t. This was a pretty
simple example that you could’ve solved quicker via the ’normal’ method, but
this method scales really well: you can add all kinds of magnetic and electric
fields, varying in time and space and this will still work.

113