Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

Electrodynamics

Dr. M. J. Perry1

Michælmas 1997

1 A
LT
EXed by Paul Metcalfe – comments and corrections to pdm23@cam.ac.uk.
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

T HANKS FOR YOUR PATIENCE

T O C LAIRE
 Contents
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

Introduction v

1 Point of departure 1
1.1 Maxwell’s Equations . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Electrostatics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2.1 Coulomb’s Law . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2.2 Multipole expansion . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Special Relativity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2 Relativistic Electromagnetism 7
2.0.1 Relativistic motion in constant electric field . . . . . . . . . . 8
2.1 Transformation of Fµν . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 Lorentz invariant scalars . . . . . . . . . . . . . . . . . . . . . . . . 9
2.3 Tensorial form of Maxwell’s equations . . . . . . . . . . . . . . . . . 9
2.3.1 Potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.4 Least action principles . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.4.1 Particle motion . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.4.2 Field action . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

3 Energy - Momentum Tensor 13

3.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.1.1 Conservation of energy-momentum . . . . . . . . . . . . . . 14
3.2 Plane waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.3 Radiation pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

4 Solving Maxwell’s Equations 17

4.1 A Green’s Function . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
4.2 The field of a moving charge . . . . . . . . . . . . . . . . . . . . . . 19
4.2.1 Radiation reaction . . . . . . . . . . . . . . . . . . . . . . . 21
4.3 Oscillating Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

5 Quantum mechanical effects 25

5.1 Minimal coupling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
5.2 Conduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
5.2.1 Meissner effect . . . . . . . . . . . . . . . . . . . . . . . . . 27
5.3 Superconducting flux quantisation . . . . . . . . . . . . . . . . . . . 28
5.4 Magnetic monopoles . . . . . . . . . . . . . . . . . . . . . . . . . . 29

iii
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

6 Born-Infeld Theory

iv
CONTENTS
31
 Introduction
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

These notes are based on the course “Electrodynamics” given by Dr. M. J. Perry in
Cambridge in the Michælmas Term 1997. These typeset notes have been produced
mainly for my own benefit but seem to be officially supported. The recommended
books for this course are discussed in the bibliography.
A word or two about the philosophy of these notes seem in order. They are based in
content on the lectures given, but I have felt free to expand and contract various details,
as well as to clarify explanations and improve the narrative flow. Errors in content are
(hopefully) mine and mine alone but I accept no responsibility for your use of these
notes.
Other sets of notes are available for different courses. At the time of typing, these
courses were:
Probability Discrete Mathematics
Analysis Further Analysis
Quantum Mechanics Fluid Dynamics 1
Quadratic Mathematics Geometry
Dynamics of D.E.’s Foundations of QM
Electrodynamics Methods of Math. Phys
Fluid Dynamics 2

They may be downloaded from

http://pdm23.trin.cam.ac.uk/˜pdm23/maths/ or
http://www.damtp.cam.ac.uk/

or you can email me on pdm23@cam.ac.uk to get a copy of the sets you require.
Even if you download them please email me to let me know, so that I can keep you up
to date with the errata and new note sets. The other people who have contributed time
and effort to these note sets are:
Richard Cameron Analysis Hugh Osborn Proofreading
Claire Gough Proofreading Malcolm Perry Accomodation
Kate Metcalfe Probability

Although these notes are free of charge anyone who wishes to express their thanks
could send a couple of bottles of interesting beer to Y1 Burrell’s Field, Grange Road.
Paul Metcalfe
4th December 1997

v
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

vi
INTRODUCTION
 Chapter 1
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

Point of departure

This is a review of terminology and results from Special Relativity and Electromag-
netism (possibly rewritten in a more grown-up way).

1.1 Maxwell’s Equations

These are :

ρ
div E =
ǫ0
curl E = −Ḃ
div B = 0
curl B = µ0 j + µ0 ǫ0 Ė.

ρ is the charge density. ǫ0 is called the permittivity of free space. It is not a funda-
mental constant but merely determines units. Similarly, µ0 is the permeability of free
space and merely determines units. µ0 and ǫ0 satisfy µ0 ǫ0 = c−2 , where c is the speed
of light (and a fundamental constant). In familiar units c ≈ 2.997 × 108 ms−1 , but we
will choose units such that c = 1.1 Dimensional analysis can replace c in any derived
formulae.

1.2 Electrostatics
This is the case where there is no current and a time independent charge distribution.
Then Maxwell’s equations reduce to div E = ǫρ0 and curl E = 0. We will assume
B = 0, but it does not affect the equations for the electric field.
q1 1
The electric field due to a point charge q1 is E = 4πǫ 0 r
2 r̂. To measure the electric

field we can take a charge q2 and measure the force F = q2 E on it.

1 Despite the fact that the Schedules mandate SI units. Exam questions will be set such that c = 1.

1
 2 CHAPTER 1. POINT OF DEPARTURE

1.2.1 Coulomb’s Law

The force between two point charges is
q1 q2 1
F=
4πǫ0 r2
directed on the line between the centres. It is repulsive for two like charges.
A point charge can be regarded as a charge distribution which is a delta function:
ρ = q1 δ(r). To find the electric field due to a distribution of charges we can use linear
superposition to find E everywhere. As curl E = 0 we can introduce the electrostatic
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

potential φ such that E = −∇φ. Then ∇2 φ = −ρ ǫ0 . We can solve this using a Green’s
function, that is a function G(r, r ) such that ∇r G(r, r′ ) = δ(r, r′ ). We can see that
′ 2
1 1
G(r, r′ ) = − 4π |r−r′ | . Then

1
Z
φ(r) = − d3 r′ G(r, r′ )ρ(r′ ).
ǫ0
Proof. Firstly, we see that
1
Z
∇2r φ = − d3 r′ ∇2r G(r, r′ )ρ(r′ )
ǫ0
1
Z
=− d3 r′ δ(r, r′ )ρ(r′ )
ǫ0
−ρ(r)
= .
ǫ0
Then we merely note that solutions to Poisson’s equation are unique.

1.2.2 Multipole expansion

Suppose we have a charge distribution in a region B as shown.

The multipole expansion of the potential is what happens to a general expression

1
for φ if |r| ≫ |r′ |. We expand |r−r′ | using the binomial expansion.

1 1

′
= (r2 − 2ri ri′ + r′2 )− 2
|r − r |
¶− 12
r′2
µ
1 2ri ri
= 1− 2 + 2
r r r
à !
2
1 ri ri′ 1 r′2 3 (ri ri′ )
= 1+ 2 − + + ... .
r r 2 r2 2 r4

We substitute into the general expression for φ to obtain

ri r′
µ ¶
1 1 ¡
Z
d3 r′ ρ(r′ ) 1 + 2 i + 4 3ri rj ri′ rj′ − r2 δij ri′ rj′ + . . . .
¢
φ(r) =
4πǫ0 r B r 2r
 1.3. SPECIAL RELATIVITY 3

This is an expansion of φ in inverse powers of r. The term in r−l is referred to as

the 2l -pole term.
Q
When l = 0 we have a monopole. If Q = d3 r′ ρ(r′ ) then φ = 4πǫ
R
0r
, which is
usually called the Coulomb term.
When l = 1 we have a dipole. Let di = d3 r′ ρ(r′ )ri′ (the dipole moment). Then
R

φ = 4πǫ10 r3 ri di .
When l = 2 we have a quadrupole and the contribution to φ is 4πǫ10 r5 ri rj Qij ,
where Qij is the quadrupole moment and Qij = 12 d3 r′ ρ(r′ ) 3ri′ rj′ − δij r′2 . Qij
R ¡ ¢

is a symmetric tracefree tensor, as

Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

1
Z
d3 r′ ρ(r′ ) 3ri′ rj′ δij − δij δij r′2 = 0.
¡ ¢
Qij δij =
2

It has 5 independent components. In general the r−l term has 2l + 1 independent com-
ponents. When l = 3 we have the octopole moment and when l = 4 the hexadecapole
moment, but these become increasingly cumbersome.

1.3 Special Relativity

Special relativity has two postulates:

1. The laws of nature are the same in any inertial frame.

2. The speed of light is independent of the speed of its source.

This leads us to consider Minkowski space, viz. xµ = (t, x) = (t, xi ). µ runs from
0 to 3 and i runs from 1 to 3. These are inertial co-ordinates. If a particle is at rest at
x = 0 at t = 0 then in remains at rest at x = 0 for all time. t is then the proper time
for that particle — what a clock sitting on the particle would measure.
We can relate the physics in one inertial frame to another by Lorentz transforma-
tions. Suppose that one has a second frame moving with velocity v in the x direction
relative to the first frame. Then we have new inertial co-ordinates

t′ = γ(v)(t − vx)
x′ = γ(v)(x − vt)
y′ = y
z ′ = z,

1
where γ(v) = √1−v 2
. This can be written x′µ = Λµ ν xν , where Λµ ν is the matrix
form of the Lorentz transformation, in this case:
 
γ −γv 0 0
−γv γ 0 0
 
 0 0 1 0
0 0 0 1
To find the Lorentz transformation of arbitrary motion described by a unit vector
n then we consider R−1 (n)ΛR(n), where R(n) is a rotation to move n into the x
 4 CHAPTER 1. POINT OF DEPARTURE

direction.  
1 0 0 0
0 
R=
0

R3 
0
with R3 an ordinary spatial rotation. We can make Lorentz transformations look like
rotations — if we put γ(v) = cosh φ, then vγv = sinh φ, which implies that v =
tanh φ. This is sometimes called a hyperbolic rotation.
We define a distance ds2 = −dt2 + dx2 + dy 2 + dz 2 between two infinitesimally
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

separated points. This can be shown to be invariant under Lorentz transforms. If dt = 0

then it reduces to the ordinary Euclidean metric on R3 .
If ds2 > 0 we say the two points are spacelike separated, if ds2 < 0 they are
timelike separated and if ds2 = 0 they are null, or lightlike separated. The interior of
the light cone has ds2 < 0 and the exterior ds2 > 0.
Proper distance is defined for spacelike separated events to be ds and proper time
dτ for timelike separated events by dτ 2 = −ds2 . The invariance of dτ means that the
time seen by a clock sitting on some object can be computed in any inertial frame.
It is easier to write ds2 = ηµν dxµ dxν , where ηµν is called the metric tensor and is
 
−1 0 0 0
0 1 0 0
 .
0 0 1 0
0 0 0 1

We also have the inverse metric η µν defined by η µν ηνρ = δρµ . The matrix form of
η µν is (obviously?)
 
1 0 0 0
0 −1 0 0
0 0 −1 0  .
 

0 0 0 −1
Given two vectors xµ and y µ we want a scalar product that should be invariant
under Lorentz transformations, and S = ηµν xµ y ν will do the trick. This gives us the
idea of defining covectors by xν = ηνµ xµ and then S = xν y ν . Then there is the
inverse operation η µν xν = η µν ηνρ xρ = xµ .
We ask ourselves how a covector transforms, and we obtain xν 7→ x′ν = Λν µ xµ ,
where Λν µ = ηνρ Λρ σ η σµ , or in matrix form
 
γ γv 0 0
γv
 γ 0 0
0 0 1 0
0 0 0 1

for the transform we looked at earlier. We see that (Λµ ν )−1 = Λν µ and

xµ y µ 7→ x′µ y ′µ = Λµ ρ xρ Λµ σ y σ
= Λµ ρ Λµ σ xρ y σ
= δσρ xρ y σ
= xσ y σ ,
 1.3. SPECIAL RELATIVITY 5

which is, on the whole, a good thing. This is analogous to rotations in R3 preserv-
ing the metric, the rotations being classified by RT R = 1 and Rij δjk Rkl = δil ex-
cluding reflections. A Lorentz transform preserves the Minkowski space metric and
Λσ µ Λτ ν ηµν = ηστ . To see the equivalence multiply both sides by η τ λ to get (eventu-
ally) δσλ = δσλ .
The Lorentz transforms are defined by Λµ ν Λσ τ ηντ = ηµσ (the group SO(3, 1))
with no spatial reflections and preservation of time, giving the Lorentz group.
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

6
CHAPTER 1. POINT OF DEPARTURE
 Chapter 2
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

The relativistic theory of

electromagnetism

We start with the Lorentz force law, F = eE + ev ∧ B and seek to generalise it.
Non-relativistically we have

d2 xi i
m = e (E + v ∧ B) ,
dt2
and we already know the equation of motion for a free relativistic particle,

d2 xµ
m = 0.
dτ 2
We also recall the non-relativistic velocity 4-vector,

dxµ
= uµ = (1, u),
dt

and we know that dτ 2 is Lorentz invariant and hence −1 = uµ uν ηµν . We guess a force
law
d2 xµ µ dx
ν
m = eF ν .
dτ 2 dτ
We see immediately that by the quotient theorem, F µ ν must be a tensor. We also
know that this equation must be true in any inertial frame, and so is always true. We
d d
take the non-relativistic case, where dτ = dt to find what F µ ν must be.

dt dx dy dz
eEx + e(vy Bz − vz By ) = eF 1 0 + eF 1 1 + eF 1 2 + eF 1 3
dτ dτ dτ dτ

We thus identify F01 = Ex , F 1 1 = 0, F 1 2 = Bz and F 1 3 = −By . We repeat this

process with y and z, and lower the index, giving us
 
Ex 0 Bz −By 
ηµλ F λ ν = Fµν =  .
 Ey −Bz 0 Bx 
Ez By −Bx 0

7
 8 CHAPTER 2. RELATIVISTIC ELECTROMAGNETISM

We note that the spatial part of this is antisymmetric, and since we treat space and time
equivalently, we can finally define
 
0 −Ex −Ey −Ez
 Ex 0 Bz −By 
Fµν =  Ey −Bz
.
0 Bx 
Ez By −Bx 0

This is the electromagnetic field (strength) tensor, or Maxwell tensor. The only puzzle
d2 t
is what the time component of the relativistic Lorentz equation represents. It is m dτ 2 =
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

0 dxi µ
eF i dτ . We note that u = γ(1, v) and that γ is a kind of relativistic energy, giving
d
dτ (mγ) = ev.E, which we know as “the rate of change of energy equals the rate of
doing work by the electric field”.

2.0.1 Relativistic motion in constant electric field

We consider a constant electric field in the x direction. We have a particle with charge
e which starts at rest at the origin. The non-relativistic case is mẍ = eE, which gives
eE 2
x = 2m t and ẋ = eE m t, which eventually exceeds 1!
Relativistically mÿ = mz̈ = 0, which are trivial. We also have the equations
In the relativistic
mẍ = eF x 0 ṫ = eE ṫ and mẗ = eE ẋ.
case, f˙ ≡ df
dτ . We integrate these once and use the initial conditions to get mṫ = eEx + C! and
mẋ = eEt. (Note that we set τ = 0 at t = 0.) Integrating again we get
eEτ eEτ mC1
x(τ ) = A sinh + B cosh −
m m eE
eEτ eEτ
t(τ ) = A cosh + B sinh + C̃.
m m
We have the boundary conditions x = ẋ = 0 at τ = 0 and t = 0 at τ = 0. There
is a temptation to put ṫ = 0 at τ = 0 — but this is inconsistent, as −1 = uµ uµ =
(−ṫ2 + ẋ2 ). Thus we put ṫ = 1 at τ = 0 — which we could have guessed, we know
that at rest, co-ordinate time is the same as proper time. Finally, we get
µ ¶
m eEτ
x(τ ) = cosh −1
eE m
m eEτ
t(τ ) = sinh .
eE m
To find the velocity we can write x(t) by eliminating τ
õ ¶1 !
m e2 E 2 t2 2
x(t) = 1+ −1 .
eE m2

We can now find the velocity

dx eE t
= q .
dt m 1 + e2 E 2 t2
m

For small t we have the reassuring dx eEt

dt = m , but in the large t case we find only
that v → 1 from below, and is always less than 1.
 2.1. TRANSFORMATION OF Fµν 9

2.1 Transformation of Fµν

We Lorentz transform (in the x direction with velocity v) the tensor Fµν to see how the
electric and magnetic fields change under Lorentz transformations.
′
We know that Fµν 7→ Fµν = Λµ ρ Λν σ Fρσ , and we just perform these sums to get:

Ex′ = Ex
Ey′ = γ(Ey − vBz )
Ez′ = γ(Ez + vBy )
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

Bx′ = Bx
By′ = γ(By + vEz )
Bz′ = γ(Bz − vEy ).

These are radically different from what we would expect if there were two electric
and magnetic 4-vectors.

2.2 Lorentz invariant scalars

We know that Fµν and ηµν are Lorentz invariant, and we can derive some Lorentz
scalars from them. The most obvious one is Fµν η µν , but as F is antisymmetric
¡ η2 symmetric
and this evaluates to zero. A more useful Lorentz scalar is Fµν F µν =
2
¢
2 B −E .
We can get another Lorentz scalar by introducing the alternating tensor, which is
defined as

1
 if µνρσ is an even permutation of 0123
µνρσ
ε = −1 if µνρσ is an odd permutation of 0123

0 otherwise.


1 µνρσ
We can now define the dual field strength tensor, Gµν = 2ε Fρσ . We can
evaluate this  
0 −Bx −By −Bz
Bx 0 −Ez Ey 
Gµν =  .
By Ez 0 −Ex 
Bz −Ey Ex 0

The dual tensor G can be found from F by E 7→ B and B 7→ −E, which

is sometimes called a duality rotation. We can now define a further Lorentz scalar
Fµν Gµν = −4 E.B.

2.3 Tensorial form of Maxwell’s equations

∂ ∂
∂i ≡ ∂xi and ∂µ ≡ ∂xµ
We start from Maxwell’s equations to see what they turn into in tensor notation. We
take them in an slightly unusual order, div E = µ0 ρ and curl B = µ0 j + Ė and seek
to write them as tensor equations.
 10 CHAPTER 2. RELATIVISTIC ELECTROMAGNETISM

1
We note that Ei = Fi0 = −F 0i and Bi = jk
2 ǫijk F . The divergence equation
becomes ∂i Fi0 = µ0 ρ. The curl equation is
1 1
ǫijk ∂j ǫklm F lm = (∂j Fij − ∂j Fji )
2 2
= ∂j Fij = µ0 ji + ∂0 Fi0 .

We hope that this is consistent with a tensor equation ∂µ F µν = X ν . If we study

this we see that this works if X µ = −µ0 (ρ, j). Thus two of the Maxwell equations
become
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

∂µ F µν = −µ0 j ν .
where j µ = (ρ, j) is the 4-vector (electric) current. This illustrates that a moving
charge and a current are just the same thing as j ν is a four-vector and so is consistent
with Lorentz transforms.
Incidentally, we have not lost conservation of charge as 0 ≡ ∂ν ∂µ F µν = −µ0 ∂ν j ν .
We now go after the next two Maxwell equations, div B = 0 and curl E = −Ḃ.
The first is easy; it gives ∂i ǫijk Fjk = 0. We guess that this is a component of
εµνρσ ∂ν Fρσ = 0, and indeed if we evaluate the spatial components we reproduce
the last Maxwell equation.
We can rewrite this in terms of the dual field-strength tensor to get ∂µ Gµν = 0 —
this says that there is no magnetic current. It turns out to be more useful to explicitly
antisymmetrize our equation to get

∂ν Fρσ + ∂ρ Fσν + ∂σ Fνρ = 0.

2.3.1 Potentials
In the non-relativistic case we know that E and B can be derived from potentials:

E = − grad φ − Ȧ
B = curl A.

It turns out that we can come up with an electromagnetic 4-vector potential, Aµ =

(φ, A) such that Fµν = ∂µ Aν − ∂ν Aµ (simply expand this to see).
We know that A is not unique in non-relativistic electromagnetism; we can add
on the gradient of any scalar function — called a gauge transformation. Similarly, we
see that the 4-vector potential Aµ is unique up to ∂µ Λ for any scalar function Λ. This
means we can try to impose extra conditions on Aµ which (partially) prevents gauge
transformations (called gauge fixing). The point is to ensure that a given Fµν is the
product of a moderately unique Aµ .1
A useful covariant gauge is to impose ∂µ Aµ = 0. Thus if we have Aµ such that
Fµν = ∂µ Aν − ∂ν Aµ and we set A′µ = Aµ + ∂ µ Λ, then we can have 0 = ∂µ A′µ =
∂µ Aµ + ∂µ ∂ µ Λ. In principle we can solve ∂µ ∂ µ Λ = −∂µ Aµ , so this gauge condition
is possible.
If we consider this a little more, we see that Aµ is still non-unique, but only up
to a solution of the wave equation, ∂µ ∂ µ Λ = 0. Solutions of the wave equation are a
µ
combination of plane waves, eıkµ x , where k µ = (ω, k) and ω = |k|. k µ is the wave
4-vector or the momentum 4-vector.
1 What “moderately unique” means has yet to be defined.
 2.4. LEAST ACTION PRINCIPLES 11

If we wish to find F µν given a collection of charges and currents we solve the

equation ∂µ ∂ µ Aν − ∂µ ∂ ν Aµ = −µ0 j ν . This is where our gauge condition comes in
useful, we get ∂µ ∂ µ Aν = −µ0 j ν , or 2Aν = −µ0 j ν , where ∂µ ∂ µ ≡ 2 (pronounced
“box”). We can now see that ∂ν Aν = 0 is sensible — it is compatible with current
conservation.

2.4 Least action principles

2.4.1 Particle motion
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

For a single free particle in special relativity we have the action

Z
p
I = dτ m −ẋµ ẋν ηµν .

d
¡ ∂L ¢ ∂L
The usual Euler-Lagrange equations dτ ∂ ẋµ − ∂xµ = 0 give −mẍµ = 0.
µ
For a particle with charge e in a potential A we can generalise this to
Z
I = dτ m −ẋµ ẋν ηµν − eAµ ẋµ .
p

But can this possibly be gauge invariant — the quantity Aµ appears explicitly? But
we see a current resulting from the motion of j µ = eẋµ . So suppose we make a small
gauge transformation such that δAµ = ∂µ Λ. Then
Z Z
δ dτ j µ Aµ = dτ j µ ∂µ Λ and integrating by parts we get
Z
= − dτ ∂µ j µ Λ = 0 by conservation of charge.

This is all somewhat academic if varying xµ does not give us the Lorentz force law.
Using the Euler-Lagrange equations we get
à !
d −mẋν ηµν d
0= p − (eAµ ) + eẋν ∂µ Aν
dτ µ ν
−ẋ ẋ ηµν dτ

τ is the proper time so ẋµ ẋν ηµν = −1 on the worldline

d
= (−mẋµ ) − eẋν ∂ν Aµ + eẋν ∂µ Aν
dτ
= −mẍν + eẋν Fµν .

This is the Lorentz force law. One can use the action as a quick way of finding the
motion of a particle. For a constant electric field in the x direction we get Ax = −Et
and all other components of Aµ are zero. To get the motion of the particle we vary the
action Z p
I = dτ m ṫ2 − ẋ2 + eEtẋ.

The Euler-Lagrange equations give the same differential equations for the motion
of the particle as before, but more easily.
 12 CHAPTER 2. RELATIVISTIC ELECTROMAGNETISM

2.4.2 Field action

We also want to find a Lagrangian L that reproduces Maxwell’s equations, that is if we
take Z
I = d4 x L,

δI = 0 under variations of something or other must reproduce Maxwell’s equations.

L must be a Lorentz scalar, built out of Fµν (or Aµ ) and it must be gauge invariant.
The Maxwell equations involve first derivatives of Fµν (or second derivatives of Aµ ).
So the only real possibilities are varying Aµ and L must be quadratic in Fµν . εµνρσ Fρσ
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

gives nothing. The usual choice is

1 µν
L=− F Fµν + jµ Aµ .
4µ0
This is gauge invariant. If δAµ = ∂µ Λ,R then if we assume Fµν = ∂µ Aν − ∂ν Aµ
we get δFµν = ∂µ ∂ν Λ − ∂ν ∂µ Λ = 0. The jν Aν part gives 0 as before.

µ ¶
1
Z
4 µν µ
I=− d x Fµν F − jµ A and
4µ0
µ ¶
1
Z
4 µν µν µ
δI = − d x (δFµν F + Fµν δF ) − j δAµ
4µ0
µ ¶
1
Z
4 µν ν
=− d x δFµν F − j δAν
2µ0
µ ¶
1
Z
4 µν ν
=− d x (∂µ δAν − ∂ν δAµ ) F − j δAν
2µ0
µ ¶
1
Z
4 µν ν
=− d x (∂µ δAν ) F − δAν j
µ0
Z ³ ´
= d4 x δAν µ10 ∂µ F µν + j ν .

We perform the last line by integrating by parts and assuming that boundary condi-
tions are all zero. As we can arbitrarily vary Aµ we must have µ10 ∂µ F µν = −j ν . The
other Maxwell equation is automatic as we have assumed that Fµν is derived from a
potential Aµ . This least action principle requires that there is no magnetic current.
 Chapter 3
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

Energy - Momentum Tensor

3.1 Definition
We seek a relativistic form of the field energy. We define the (stress-)energy tensor

1 ¡ µσ ν
T µν = F F σ − 41 η µν F ρσ Fρσ .
¢
µ0

00
1
¡ 2note in2 ¢passing that this is a symmetric tensor. Note that the T component is
We
2µ0 E + B , which reproduces the non-relativistic energy density. To interpret the
rest of the components of T µν we recall Poynting’s theorem.
Poynting’s Theorem. Let D be a region in space. Then the rate of change of
energy in D is

1 ∂ 1
Z Z
E2 + B2 dV = E.Ė + B.Ḃ dV
2µ0 ∂t D µ0 D
1
Z
= E. (curl B − µ0 j) − B. curl E dV
µ0 D
1
Z Z
=− j.E dV + E. curl B − B. curl E dV
D µ0 D
Z Z
=− j.E dV − N.dS,
D ∂D

where we have introduced the Poynting vector N = µ10 E ∧ B. The Poynting vector is
the energy flux.
By performing the sum we find that T 0k = N k . We find that
 
Energy Energy flux
 density 
T µν =  .

Energy

 flux “Stress” 

We want to evaluate the “stress” part of this tensor.

13
 14 CHAPTER 3. ENERGY - MOMENTUM TENSOR

1
Fiµ Fj µ − 41 ηij F ρσ Fρσ
¡ ¢
Tij = µ0
1
Fi0 Fj 0 + Fik Fj k − 41 δij −2E 2 + 2B 2
¡ ¡ ¢¢
= µ0
1
−Ei Ej + ǫikl Bl ǫjkm Bm + 12 δij E 2 − 21 δij B 2
¡ ¢
= µ0
1
−Ei Ej + 21 δij E 2 − Bi Bj + 21 δij B 2 .
¡ ¢
= µ0

This is the Maxwell stress tensor, and can be thought of as the pressure of the
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

electromagnetic field.

3.1.1 Conservation of energy-momentum

We compute the divergence of T µν .

1
∂ν T µν = F σ − 14 η µν F ρσ Fρσ
µσ ν
¡ ¢
µ0 ∂ν F
1 µσ
) F ν σ + F µσ (∂ν F ν σ ) − 21 (∂ µ F ρσ ) Fρσ
¡ ¢
= µ0 (∂ν F
−F µσ jσ + µ10 Fρσ − 12 ∂ µ F ρσ + ∂ρ F µσ
¡ ¢
=
= −F µσ jσ − 2µ1 0 Fρσ (∂ µ F ρσ + ∂ σ F µρ + ∂ ρ F σµ )
= −F µσ jσ by Maxwell’s equations.

Thus in the absence of charges/currents, the energy-momentum tensor is conserved.

We can evaluate the right hand side of this equation to get

∂ν T µν = (−j.E, ρE + j ∧ B) .

The time component of this is the work done by the electromagnetic field and the
spatial components give the electric force on a current j due to B and on a charge
density ρ due to E.

3.2 Plane waves

This has the equation 2Aµ = 0 in the gauge ∂µ Aµ = 0, which has solutions

Aµ = Aǫµ exp (ıkσ xσ ) ,

Note that this is a complex solution, so when we work with E and B we must take
the real part of the field, which corresponds to the imaginary part of Aµ .
ǫµ is the polarisation vector, and kσ = (−ω, k) is the wave 4-vector. ω is the
angular frequency, k is the wave vector and |k| = ω. A is the amplitude. Imposing the
gauge condition requires ǫµ k µ = 0, so we have the transversality of the wave.
This does not completely specify the gauge. If we let δAµ = ∂µ Λ with Λ =
−ıCeık.x then ǫµ 7→ ǫµ + Ckµ . Since kµ k µ = 0 this preserves the gauge condition.
This freedom is usually exploited to put A0 = 0. In this case,

Aµ = (0, Aǫeık.x ) = (0, Aǫeı(k.x−ωt) ),

where ǫ is a spatial vector. The gauge condition gives k.ǫ = 0.

 3.3. RADIATION PRESSURE 15

Finding E and B from Aµ is easy:

Ei = −F0i = −∂0 Ai + ∂i A0 = ıωAi .

and
Bi = 21 ǫijk Fjk = ǫijk ∂j Ak = −ık ∧ A.
The physical fields correspond to the real parts of these quantities.

3.3 Radiation pressure

Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

Suppose that we have a situation as drawn above, with a plane wave propagating in
the z direction. Then the electric and magnetic fields are in the y and x directions
respectively, with Ey = ωA sin ω(t − z) and Bx = ωA cos ω(t − z).
The rate of flow of momentum per unit area is |N|, where N is the Poynting vector
and this has a time average 2µ1 0 ω 2 A2 . This is coincidentally the same as the energy
density. We also evaluate the stress-energy tensor

Tij = µ10 12 δij E 2 + B 2 − Ei Ej − Bi Bj .

¡ ¡ ¢ ¢

This is clearly diagonal, and evaluating the diagonal components we get hTxx i = 0,
hTyy i = 0 and hTzz i = 2µ1 0 ω 2 A2 . There is a pressure due to the wave, but importantly
it is not isotropic.
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

16
CHAPTER 3. ENERGY - MOMENTUM TENSOR
 Chapter 4
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

Solving Maxwell’s Equations

4.1 A Green’s Function

We hope to find a general expression for Aµ given a time dependent distribution of
charges and currents. We will work in the ∂µ Aµ = 0 gauge, and so we have to solve
the equation 2Aµ = −µ0 j µ . We proceed naively and see what happens.
We hope to find a Green’s function G(x, x′ ) such that 2G = δ (4) (x, x′ ) and so
Z
Aµ (x) = −µ0 d4 x′ G(x, x′ )j µ (x′ )

is a solution of 2Aµ = −µ0 j µ . One problem is that 2 is a hyperbolic operator so there

exist non-trivial solutions to 2φ = 0 with φ → 0 at infinity.
The four dimension Fourier transform is defined by
Z
ˆ
f (k) = d4 x f (x)e−ık.x .

The minus sign in the exponential is not arbitrary. If f is a plane wave f (x) ∼ eıp.x
then fˆ(k) = (2π)4 δ (4) (p − k), which is what we want.
We will solve 2G(x, x′ ) = δ (4) (x, x′ ) using the Fourier transform. Ĝ(k, x′ ) =
−2 −ık.x
−k e and so defining zµ = xµ − x′µ = (z 0 , z) we find
0 0
1 eık.z e−ık z
Z
′ 3 0
G(x, x ) = d kdk
(2π)4 k 02 − k2

and note that if we perform the k0 integral we see that the integrand is singular at
k0 = ± |k|. We thus need to choose on which contour to perform the integral.

If we consider the retarded Green’s function Gret , which we get by integrating along
Γ1 , we see that for z 0 < 0, G(x, x′ ) = 0 as we can close the contour in the upper half
plane and apply Cauchy’s theorem. For z 0 > 0 we have to close the contour in the

17
 18 CHAPTER 4. SOLVING MAXWELL’S EQUATIONS

lower half plane. In doing this we pick up two poles at ± |k| and can apply the residue
theorem.
The advanced Green’s function Gadv is obtained by integrating along Γ2 . In this
case G is only non-zero for z 0 > 0.
The retarded Green’s function agrees with intuitive ideas of causality so we use
that. All we have to do now is evaluate it.

0 0
1 eık.z e−ık z
Z
′ 0 3 0
Gret (x, x ) = θ(z ) d kdk
(2π)4 k 02 − k2
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

We close the contour clockwise so we get {−2πı residues} for the k 0 integral, thus
P

à 0 0
!
2πıθ(z 0 ) e−ı|k|z − eı|k|z
Z
′ 3 ık.z
Gret (x, x ) = − d ke .
(2π)4 2 |k|

We convert this into spherical polars in k-space: kx = k sin θ cos φ, ky = k sin θ sin φ
and kz = k cos θ and so

ıθ(z 0 ) eıkz cos θ ³ −ıkz0

Z ´
′ 2 ıkz 0
Gret (x, x ) = − k dk sin θdθ dφ e − e
16π 3 k
ıθ(z 0 ) eıkz cos θ ³ −ıkz0
Z ´
ıkz 0
=− kdk sin θdθ e − e
8π 2 k
θ(z 0 ) ∞
Z
£ ıkz cos θ ¤θ=π ³ −ıkz0 ıkz 0
´
= dk e e − e .
8π 2 z 0 θ=0

The integrand is even in k, so

∞
θ(z 0 )
Z
¢³ 0 0
´
Gret (x, x′ ) = dk e−ıkz − eıkz e−ıkz − eıkz .
¡
16π 2 z −∞
R∞
Recall that −∞ dkeıkx = 2πδ(x). Thus the integral is a combination of four delta
functions, but the step function θ kills two of them off and we get

θ(z 0 )
Gret (x, x′ ) = − δ(z − z 0 ).
4πz
θ(−z 0 ) 0
For the record, Gadv = 4πz δ(z + z ). We can make our result for Gret look more
covariant by recalling that
X δ(x − ai )
δ(f (x)) =
i
|f ′ (ai )|
2 2
where f (ai ) = 0. Then as δ (3) (z 2 ) = δ(|z| − z 0 ) (z is a four-vector) we have
1 ³ (3) ´
δ (3) (z 2 ) = δ (|z| − z 0 ) + δ (3) (|z| + z 0 )
2 |z|
and as the step function removes one of these delta functions we get
1
Gret (x, x′ ) = − θ(z 0 )δ (3) (z 2 ), z µ = xµ − x′µ .
2π
 4.2. THE FIELD OF A MOVING CHARGE 19

Now suppose we wish to evaluate Aµ (x) for some current distribution j µ as shown.
We get
Z
µ0
Aµ (x) = d4 x′ j µ (x′ )δ (3) (x − x′ )2 θ(x0 − x′0 ).
¡ ¢
2π
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

This comes about because we chose Gret . The advanced Green’s function gives the
reverse. Thus Gret is consistent with our ideas of causality. Other choices of G are
not. This choice goes beyond local physics. It is presumably solved by appealing to
cosmology or quantum theory.
We also note that the only contributions to Aµ (x) come from points x′ such that
(x − x′ )2 = 0 — that is only when x and x′ can be joined by a light ray pointing
towards the future of x′ .

4.2 The field of a moving charge

Suppose we have a moving charge, with (non-relativistically) ρ = eδ 3 (x − (t)) and
therefore j = e dy 3 µ
dt δ (x − (t)). In the relativistic case we replace y(t) with y (τ ) and
µ µ 3 i i
get j = eu δ (x − y (t)). We can use a trick to make this look more covariant,
Z
j = e dτ uµ δ 4 (xν − y ν (τ )).
µ

Then
dy µ 4 ν
Z
eµ0
Aµ = d4 x′ dτ δ (x − x′ )2 θ(x0 − x′0 ) δ (x − y ν (τ )),
¡ ¢
2π dτ
which has the effect of integrating over the backward light cone. This can be evaluated
by carrying out the x′ integral first. Note that

¢ δ |x − y| − (x0 − y 0 ) + δ |x − y| + (x0 − y 0 )
¡ ¢ ¡ ¢
2
¡
δ (x − y) = ν
−2(x − y)ν dy
dτ

The second delta function does not contribute (because we are using the retarded
Green’s function) and so

dy µ δ |x − y| − (x0 − y 0 )
Z ¡ ¢
µ eµ0
A = dτ
2π dτ −2(x − y)ν dy
dτ
ν

dy µ
¯
eµ0 dτ
¯
=
¯
ν
4π dy (x − y)ν ¯
¯
dτ τ =τ0
 20 CHAPTER 4. SOLVING MAXWELL’S EQUATIONS

where τ0 is the value of τ on the world-line where the past light cone of x intersects
the world line of the particle. This is usually referred to as evaluating at some instant
of retarded time. These are the Lienard-Wiechert potentials, and are painful to use in
arbitrary relativistic motion.
The result
dy µ
Z
eµ0
µ
dτ δ (x − y)2 θ(x0 − y 0 )
¡ ¢
A =
2π dτ
is useful for relativistic motion. To evaluate the fields we need to calculate terms of the
form
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

dy µ ν ¡
Z
eµ0
∂ ν Aµ = dτ θ(x0 − y 0 ) ∂ δ (x − y)2
¢
2π dτ
dy µ xν − y ν (τ ) d ¡
Z
eµ0
dτ θ(x0 − y 0 (τ )) δ (x − y)2
¢
= dy ρ
2π dτ dτ (x − y)ρ dτ

To evaluate this we integrate by parts. We take x0 6= y 0 and thus remove points on

the world-line of the particle from consideration. The field is not well defined there.
Thus
dy ν dy µ
" #
µν eµ0 d (x − y)µ dτ − (x − y)ν dτ
F =− dy ρ
.
2π dτ dτ (x − y)ρ τ =τ 0

We write (x − y)µ = (+R, Rn) where R is the spatial distance |x − y| and n is

a unit vector. The plus sign on R comes from the retarded Green’s function. We also
need the velocity v = (γ, γv) where v = dydt . After evaluating this we get
· ¸
eµ0 n−v n ∧ {(n − v) ∧ v̇}
E= +
2π γ 2 (1 − n.v)3 R2 (1 − n.v)3 R retarded time
B = n ∧ E.

The first term in the expression for E is just the Coulomb field (put v = 0 to
see this). The second term only appears if v̇ 6= 0 — it depends on the acceleration.
Then E ∼ accelerationR and B ∼ accelerationR and are perpendicular. Thus the
2
Poynting vector is N ∼ acceleration
R2 . Thus the energy flux out of a large radius sphere
2
∼ acceleration — accelerating particles radiate energy.
For a non-relativistic particle it is somewhat easier. We use the Lienard-Wiechert
potentials
dy µ
¯
µ eµ0 dτ
¯
A =
¯
ν
4π dy (x − y)ν ¯
¯
dτ retarded time
µ µ ′
and put x = (t, x), y = (t , y) with x − y = Rn where n is a unit vector. For
µ
non-relativistic motion dy
dτ = (1, v) and
¯
µ eµ0 (1, v) ¯¯
A = .
4π |x − y| ¯at t′ = t − R

It is straightforward to calculate E and B from B = curl A and E = − grad A0 −

Ȧ. We first evaluate B, and note that we are only interested in the O(R−1 ) terms —
to get the radiation at infinity. Thus
eµ0 eµ0
B∼− n ∧ v̇ and E∼− n ∧ [nv̇] .
4πR 4πR
 4.3. OSCILLATING FIELDS 21

e2 µ0 2
The Poynting vector N = 1
µ0 E ∧ B evaluates as N = 16π 2 R2 (ÿ) sin2 θ n.

The radiation is mainly perpendicular to the direction of the acceleration and is

axisymmetric about that axis. v̇ determines the time dependence of the radiation and
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

thus the frequency can be found by Fourier transforming v̇.

R
The power radiated (or the total flux of radiation) is N.dS over a sphere at in-
finity which we assume is at a large distance from the particle (the celestial sphere).
Converting to polars we get

e2 µ0 e2 µ0
Z
2 2
flux of radiation = (ÿ) sin3 θ dθdφ = (ÿ) .
16π 2 6π

This is Larmor’s formula.

4.2.1 Radiation reaction

We consider a particle with mass m and charge e moving under an external force Fext .
Assume (naively) Newton’s Law mÿ = Fext and dot this with ẏ and integrate to get
that the change in kinetic energy equals the work done by the applied force. But we
know that this is not true — there are radiative losses at infinity. We therefore guess
another force FR and propose mÿ = Fext + FR . Dotting this with ẏ and integrating
we see that FR .ẏ is the radiative energy loss. Using Larmor’s formula and assuming
2 ...
that there is no acceleration at the endpoints of the motion we get FR = e6πµ0 y and
derive the Abraham-Lorentz equation

e2 µ0 ...
µ ¶
m ÿ − y = Fext .
6πm

This is very odd and leads to embarrassing difficulties. To solve such an equation
three initial conditions are needed, position, velocity and acceleration. But if we take
t 2
Fext = 0 we see that the solutions are A + Bt + Ce τ , where τ is the timescale e6πµ µ0
.
This exponential runaway solution is presumed to be unphysical.
If Fext is a delta function and we assume the initial conditions x = ẋ = 0 we
have to adjust ẍ to suppress the runaway solution. We see that the particle (if charged)
must start accelerating before the force is applied. This acausal behaviour is called pre-
acceleration and is governed by the timescale τ , which for an electron is approximately
6 × 10−24 s, into the realm of quantum mechanical effects.

4.3 Oscillating Fields

Assume that j µ (x, t) = j µ (x)eıωt with j µ non-zero only in some domain D.
 22 CHAPTER 4. SOLVING MAXWELL’S EQUATIONS

Using our result for Aµ we get

1
Z
µ0 ıωt
e−ıω|x−x | .
′
Aµ (x, t) = e d3 x′ j µ (x′ ) ′
4π |x − x |
1
If R ≫ d we can expand |x−x′ | in the usual way as

! 1
2 −2
Ã
1 1 2x.x′ |x′ |
′
= 1− + 2
|x − x | R R2 R
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

à !
2 2
1 x.x′ 1 |x′ | 3 (x.x′ )
= 1+ 2 − + ... .
R R 2 R2 2 R4

The O(R−2 ) and lower terms do not contribute to the radiation and will be omitted.
We get Z
µ0 ıωt
d3 x′ j µ (x′ )e−ıω|x−x | .
′
Aµ (x, t) = e
4πR

We can perform a similar expansion on e−ıω|x−x | and finally get

′

Z
µ0 ıω(t−R) x.x′
µ
A (x, t) = e d3 x′ j µ (x′ )eıω R
4πR

provided R ≫ λ, the wavelength. Thus the expansion we have derived is valid when
R ≫ d, λ. This is called the radiation zone.
Thus at large distances the system appears to be a source of spherical waves. To
proceed further we can expand out the phase factor in powers of ω. We get
" #
′ 2
x.x′ 2 (x.x )
Z
µ µ0 ıω(t−R) 3 ′ µ ′
A (x, t) = e d x j (x ) 1 + ıω −ω + ... .
4πR R 2R2

In the radiation zone when ωd ≫ 1 these terms are successively smaller.

Recall that j µ = (ρ, j). Then
µ0 ıω(t−R) h ıω i
A0 (x, t) = e Q + x.p + . . .
4πR R
where p is the electric dipole moment of the system. Note that Q = 0 as the total
charge cannot depend on time. For the vector potential,
·Z Z ¸
µ0 ıω(t−R) ıω
Ai (x, t) = e d3 x′ ji (x′ ) + xj d3 x′ x′j ji (x′ ) + . . .
4πR R

We can simplify this by noting (integrate by parts) that

Z Z
d3 x′ ji (x′ ) = − d3 x′ x′i ∂j jj (x′ )

and applying the continuity equation, which in this case is ıωρ + div j = 0. Thus
Z
d3 x′ ji (x′ ) = ıωp
 4.3. OSCILLATING FIELDS 23

and we get
ıωµ0 ıω(t−R) p.x ıωµ0 ıω(t−R)
A0 = e and A= e p.
4πR2 R 4πR
We can now calculate E and B as
ω 2 µ0 ıω(t−R) ¡ 2 ¢
E= 3
e R p − (x.p)x
4πR
ω 2 µ0 ıω(t−R)
= e x ∧ (x ∧ p) and
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

4πR3
ω 2 µ0 ıω(t−R)
B= e x ∧ p.
4πR2
The time averaged Poynting vector thus points radially outwards and has magnitude
4 2
0 ω |p| µ0 ω 4 |p|2
N = µ32π 2
2 R2 sin θ and the average power radiated is therefore 12π .
The scattered light has ω 4 dependence times the spectrum of the light. Thus blue
light is scattered preferentially to red and the sky appears blue. This also explains the
red sun at sunset; since there is more scattering when the angle of the sun is low and
the blue light is scattered more.
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

24
CHAPTER 4. SOLVING MAXWELL’S EQUATIONS
 Chapter 5
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

Quantum mechanical effects

5.1 Minimal coupling

Consider a particle with charge e, worldline xµ (τ ) in an electromagnetic field with
potential Aµ . Recall we obtained an action
Z ³ p ´ Z
I = − dτ m −ẋ2 − eAµ ẋµ = dτ L(x, ẋ),

where the minus sign is inessential; it just normalizes things nicely. The momentum
π µ conjugate to xµ is ∂∂L µ µ
ẋµ = mẋ +eA , consisting of the mechanical momentum and
a modification due to the electromagnetic field. The Hamiltonian H(x, π) = π µ ẋµ −
2
L = (π−eA)
m .
The replacement of p with π = p + eA is usually termed “minimal coupling” and
corresponds to classical electrodynamics.
In quantum mechanics the momentum p is replaced with a momentum operator p̂,
and we assume that the momentum operator for charged particles is modified “like the
classical momentum”, that is p 7→ π = p̂ + eA 7→ −ı~∇ + eA.
p2
The Schrödinger equation for a static field (0, A) is 2m ψ = Eψ, which according
(−ı~∇+eA)
to minimal coupling, and turning p into an operator, is 2m ψ = Eψ.
Since gauge transformations are not supposed to have any physical effect solu-
tions of the Schrödinger equation in one gauge must be solutions in another gauge. If
we start with the universal combination (−ı~∇ + eA) ψ, on sending A 7→ A + ∇Λ
the universal combination becomes (−ı~∇ + eA + e∇Λ) ψ ′ . This must be invariant
ıeΛ
(up to a phase factor), and so if ψ ′ = ψe− ~ we get (−ı~∇ + eA + e∇Λ) ψ ′ =
− ıeΛ
e ~ (−ı~∇ − e∇Λ + eA + e∇Λ) ψ and the universal combination is invariant (up
to a phase factor). Phase should not be too disturbing; the matrix element
Z Z
ıeΛ ıeΛ
d3 x ψ1∗ Ôψ2 7→ d3 x ψ1∗ e ~ Ôψ2 e− ~

and under all normal circumstances the phase factors cancel; the matrix element is
invariant.
This minimal coupling means that the vector potential can give rise to observable
physical effects. One which you may have met before is the Aharonov - Bohm effect.

25
 26 CHAPTER 5. QUANTUM MECHANICAL EFFECTS

Consider the long, thin solenoid shown, with B 6= 0 inside and B = 0 outside.
In classical mechanics, charged particles with be unaffected since B = 0 outside the
solenoid.
In quantum mechanics; consider eigenstates of π; states ψ with (−ı~∇ + eA) ψ =
πψ. If the phases of the waves on the two paths differ then there will be destructive
interference.
Now suppose that a neutral particle has a wavefunction ³ψ0 (x). For a charged
´ par-
ıe x
R
ticle the corresponding wavefunction is ψ(x) = ψ0 (x) exp − ~ x0 A.dl . Thus the
phase factor (the difference in phase) between the two paths is
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

ıe ıe
H R
e− ~ A.dl
= e− ~ curl A.dS

ıe
R
= e− ~ B.dS

ıe
= e− ~ (flux) .

By appropriate choice of the flux Φ we can get as much or as little interference as

we want. If ~e Φ = π then there is completely destructive interference; if ~e Φ = 2π then
the interference is completely constructive and the solenoid is undetectable. In general
if Φ = 2πn~e the solenoid is unobservable. This is an inherently quantum mechanical
effect.
One might think that B = 0 outside the solenoid implies that A = 0 outside the
solenoid. This is true only if the region is simply connected — which it isn’t. We can
make a gauge transformation to put A = 0 at a point but because the region is not
simply connected we cannot do this everywhere.
This was experimentally verified in the 1960’s.

5.2 Conduction
An ordinary conductor looks something like a regular lattice of atoms, with the valence
electrons forming an electron gas throughout the material.

An applied E field moves the gas, but electrons collide with atoms and stop. Sup-
pose they move with an average velocity v. Then the current density is the charge on an
electron × the number density ×v. The mean free path only depends on the geometry,
so the current density is σE, with σ the conductivity.
Superconductivity is very different. It was first discovered by Kammerlingh-Onnes
in 1905; he noticed that when some metals are cooled to ≈ 4K the electric conductivity
became infinite. Nowadays superconductivity is observed in certain materials up to
about liquid nitrogen temperatures, ≈ 100K.
The fundamental description of superconductivity is due to Bardeen, Cooper and
Schreiffer and is in detail beyond this course. The result is that the current is an inher-
ently quantum mechanical effect in which bound states of pairs of electrons behave as
bosons rather than as fermions. They have a charge −2e and an effective mass of m
(say).
 5.2. CONDUCTION 27

We will examine the Landau-Ginzburg theory. Suppose the charge carriers have a
wavefunction χ = Reıφ . We can then interpret the probability current as the flux of
these particles. We can evaluate

~ ¡ ∗ ∗ ¢ ~
jprob = χ ∇χ − (∇χ) χ = R2 ∇φ.
2ım m
We interpret R2 as a number density ns and so we guess an electric current j =
q~
m ∇φ. However this is not gauge invariant and as the electric current must stay the
same under gauge transformations we fix up the equation to get the result (which can
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

be derived from the BCS theory)

q~ns ³ q ´
js = ∇φ − A .
m ~

5.2.1 Meissner effect

Since div B = 0 lines of B cannot end. However if one takes a material in a mag-
netic field and cools it to its superconducting temperature one observes a change in the
magnetic field.

We are led to guess that B = 0 inside a superconductor. The ªabove expression for
js and the Maxwell equations give curl B = µ0 q~n q
©
m
s
∇φ − ~ A . Taking the curl of
this we get a differential equation for B:

µ0 q 2 ns
∇2 B = B.
m

2 2
µ0 q ns
In the region shown this simplifies to ∂∂zB
2 = m B and so we find that B =
q
2
µ0 q ns
B0 exp − m z, taking the negative root since the energy must be bounded.
q
B decays exponentially away from the surface on a distance scale ns qm2 µ0 , which
is of the order of atomic size. Thus in practice we have B = 0 inside a supercon-
ductor and this is a better definition of a superconductor than saying that it has infinite
conductivity.
A is not necessarily 0, but in order to get a superconducting current we must have
ns 6= 0. Landau and Ginzburg tried to construct an analog of the Schrödinger equation
which gave this result.
It is easier (as always) to start from an action:
 28 CHAPTER 5. QUANTUM MECHANICAL EFFECTS

Z
I= d3 x [kinetic energy − potential energy]

~2 ∗ 2
Z · ¸
= d3 x − ψ ∇ ψ − V ψ∗ ψ
2m
· ¸
1
Z
∗
= d3 x (−ı~∇ψ) (−ı~∇ψ) + V ψ ∗ ψ
2m
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

incorporating the magnetic field via minimal coupling

Z
∗
d3 x (−ı~∇ψ − qAψ) (−ı~∇ψ − qAψ) + V ψ ∗ ψ .
£ ¤
=

This is gauge invariant.

It cannot depend on where we are in the superconductor and so V is constant. We
get the familiar Schrödinger equation which has the obvious solution ψ = 0 and no
other solution independent of x.
4
Landau and Ginzburg proposed the addition of a term 21 b |ψ| to this action to get

~2 ¯¯³ ıq ´ ¯¯2 1
Z
2 4
I= d3 x ¯ ∇ − A ψ ¯ + V |ψ| + b |ψ| .
4m ~ 2

This action can be derived from BCS theory and gives a nonlinear analog of the
Schrödinger equation:
1 2 2
− (−ı~ − qA) ψ + V ψ + b |ψ| ψ = 0.
4m
∗¢ 2 2
q~
ψ ∇ψ − ψ (∇ψ) − 2q
¡ ∗
The current js = 2ım m A |ψ| .
We get a non-vanishing spatially independent solution of this “Schrödinger equa-
tion” when V < 0 and b > 0. This occurs when the temperature T is less than some
critical temperature Tc ; normal matter has V > 0.
BCS theory gives b > 0 and V = V0 (T − Tc ).

5.3 Superconducting flux quantisation

Consider a ring of superconducting material as shown.

In the material B = 0 and j = 0. Since j ∝ ∇ψ − ~q A we must have A = ~q ∇φ

¡ ¢

inside the ring.

The magnetic flux through the loop is
Z I
B.dS = A.dl.
shaded surface boundary loop
 5.4. MAGNETIC MONOPOLES 29

Evaluating this inside the superconductor we get ~q [φ]. As the wavefunction must
be single valued this must be n 2π~q and since the charge carriers are electron pairs then
the flux is quantised in units of π~
|e| .
If we make a current I flow on the surface of the superconductor then as the flux
through the loop is the inductance times the current, and so the flux is quantised we see
that the current is quantised.

5.4 Magnetic monopoles

Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

µ0
Suppose that a B field 4π P rr̂2 is possible, by analogy with the Coulomb field in elec-
trostatics.
Using Gauss’ Law we have

1 1
Z Z
P = B.dS = div B dV.
µ0 closed surface µ0

Thus if div B = 0 everywhere then P = 0 and magnetic charges cannot arise.

Thus Maxwell’s equations must be modified in order to get this field.
A suitable vector potential A is Aφ = µ4π 0P
(1 + cos θ) (in spherical polars). We
µ0 (1+cos θ)
have |A| = 4π r sin θ . There is a difficulty at θ = 0 for all r.
This singularity on the North axis is called the Dirac string. It can be moved about
by gauge transformations; if we have A 7→ A + ∇ −µ2π0 P φ we can put the Dirac string
onto the South axis.
Since (by axiom) the observable physics should not depend on the gauge used the
string singularity should be unobservable.
We showed earlierHthat the phase difference between two paths going in front of /
behind the string is ~e A.dl and this must be an integer multiple of 2π. Evaluating
the integral gives the Dirac quantisation condition P = n 2π~
µ0 e .
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

30
CHAPTER 5. QUANTUM MECHANICAL EFFECTS
 Chapter 6
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

Born-Infeld Theory
This material is starred and
Recall that Maxwell’s theory is (in the absence of currents) governed by an action was included as a fill-in lec-
1
Z
1 ture.
I= d4 x − Fµν F µν ,
µ0 4

giving the two Maxwell equations ∂µ Fρσ + ∂ρ Fσµ ∂σ Fµρ = 0 and ∂µ F µν = 0.

There is a hidden duality symmetry under Fµν 7→ 21 εµνρσ F ρσ of both the action and
the equations of motion.
Q 1
Recall also that electric charges have a radial component Er = 4πǫ 0 r
2 and that
1 2
the energy density in the electric field is 2 ǫ0 |E| . We can see that the energy density
blows up at the origin and also that the total energy in the electric field is infinite.
We also propose a similiar magnetic monopole field Br = P4π µ0 1
r 2 ; the energy in
this magnetic field is also infinite.
The Born-Infeld theory emerges from string theory. It depends on a parameter b
with the dimensions of length. We take a new action;
½ ¾
1
Z q
4
d x 1 − |det ηµν + bFµν |
µ0 b2

and we suppose that Fµν = ∂µ Aν − ∂ν Aµ .

Since we have that, up to a Lorentz transform,
 
0 λ1 0 0
−λ1 0 0 0
Fµν = F̃µν =   
0 0 0 λ2 
0 0 −λ2 0

we can see that det ηµν + bF̃µν = det ηµν + bFµν and so we can evaluate the action as
( r )
1 1 1
Z
4 2
d x 1− 1 − b2 Fµν F µν − b4 (εµνρσ Fµν Fρσ ) .
µ0 b2 2 16

The limit b → 0 (clearly) gives the Maxwell action. Since we are assuming that
Fµν is derived from a potential we still have the equation ∂µ Fνρ + ∂ν Fρµ + ∂ρ Fµν = 0
and the other equation is ∂µ G µν = 0, the difference being that the equation for G µν is

31
 32 CHAPTER 6. BORN-INFELD THEORY

a horrible mess:
b2 λτ
µν F µν − 4 F εµνλτ Fαβ Fγδ εαβγδ
G =q
b2 ξζ b4 2
1− 2 Fξζ F − 16 (εξζχ̟ F ξζ F χ̟ )

(you have no idea how difficult it was to find that many different Greek letters).
We obtain ½ ¾
1 1
Tµν = Gµ λ Fνλ + ηµν L
µ0 4
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

n o
where L is the Lagrangian; L = b12 1 − |det ηµν + bFµν | .
p

There is a symmetry in these equations under Fµν 7→ 12 εµνρσ G ρσ and Gµν 7→

− 21 εµνρσ F ρσ :
the Lagrangian and equations of motions are invariant. This has the
effect of swapping E and B.
The analog of the electric field of a point charge is an E field which is purely radial,
defined by A0 = φ and Er = −∇r φ. In this case the action reduces to
1
Z h p i
2 2 φ2
r dr sin θdθdφ 1 − 1 − b r
µ0 b2
and variation of this yields
r 2 φr
p = const = a.
1 − b2 φ2r
Solving this for φr = −Er yields
a
Er = − √
r + a2 b2
4

and so as r → ∞, Er ∼ − ra2 . Thus if we wish to reproduce the Maxwell field for

Q
large distances a = − 4πǫ 0
. Thus as r → 0 we see that Er → 1b .
The energy density in the electric field is
s 
2 2
½ ¾
1 1 1  b Q 
√ − 1 = 1 + − 1
µ0 b2 1 − b2 E2 µ0 b2  16π 2 ǫ0 r4 

which is singular (but integrably so) at r = 0. Performing the integral to find the
total energy in the electric field we obtain
µ ¶ 23
4π 1 Q 1
√ ,
3 b 34 )2
1
Γ( 4 ǫ02
which is noticably finite.
This theory also has magnetic monopoles; an easy way is to see that the theory is
invariant under swapping E and B. The energy in a magnetic monopole field is
µ ¶ 32
4π 1 P √
√ µ0
3 b Γ( 34 )2 4
µ0 P
if Br ∼ 4πr 2 as r → ∞.
µ0 P
In fact Br = 4πr 2 which although it looks singular is perfectly reasonable.
 References
Copyright © 2004 University of Cambridge. Not to be quoted or reproduced without permission.

◦ J.D. Jackson, Classical Electrodynamics, Second ed., Wiley, 1975.

This is probably the best book for the course. It is very complete and reasonably easy
to read. It has two major flaws; its price and its use of CGS units. Every college library
should have at least one copy.

◦ Feynman, Leighton, Sands, The Feynman Lectures on Physics Vol. 2, Addison-

Wesley, 1964.
The Feynman lectures are always good reading. However Feynman II, the book on
electromagnetism, is not really suitable as a textbook for this course. It is not complete
enough although it is good for introductory reading and revision of earlier work. It’s
also fun...

◦ Landau and Lifschitz, The Classical Theory of Fields, Fourth ed., Butterworth-
Heinemann, 1975.
This is a good reference but isn’t so good to learn by. It also uses CGS units.

The sign conventions for the tensor Fµν are variable and it is
essential to check which convention any given book is using.

There appears to be a gap in the market at about the level of this course. Most books
on electromagnetism seem to be written either for a first course or for a postgraduate
course. If you find a good one I haven’t mentioned please send me a brief review and
I will append it to this bibliography if I think it is suitable.

33