Electromagnetism

James Sparks & Erik Panzer

Hilary Term 2024

ρ
∇·E = ∇·B = 0
ϵ0
(Gauss’ law) (no magnetic monopoles)
 
∂B ∂E
∇×E = − ∇ × B = µ0 J + ϵ0
∂t ∂t
(Faraday’s law) (Ampère-Maxwell law)

About these notes

These are lecture notes for the B7.2 Electromagnetism course, which is a third year course in the
mathematics syllabus at the University of Oxford. Starred sections/paragraphs are not examinable,
either because the material is slightly off-syllabus, or because it is more difficult. There are four
problem sheets. Please send any questions/corrections/comments to erik.panzer@maths.ox.ac.uk.
These notes are largely unaltered from the version written by James Sparks for Hilary term
2022, which acknowledged earlier material from Xenia de la Ossa, Fernando Alday, and Paul Tod.

Contents

1 Electrostatics 1
1.1 Point charges and Coulomb’s law . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 The electric field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Gauss’ law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.4 Charge density and Gauss’ law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.5 Electrostatic potential and Poisson’s equation . . . . . . . . . . . . . . . . . . . . . 7
1.6 Conductors and surface charge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.7 Electrostatic energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2 Boundary value problems in electrostatics 14

2.1 Boundary value problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2 Green’s functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

i
 2.3 Method of images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.4 Orthonormal functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.4.1 General theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.4.2 Cartesian coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.4.3 Spherical polar coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.5 * Complex analytic methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

3 Magnetostatics 39
3.1 Electric currents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.2 Continuity equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.3 Lorentz force and the magnetic field . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.4 Biot-Savart law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.5 Magnetic monopoles? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.6 Ampère’s law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.7 Magnetostatic vector potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.8 Multipole expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

4 Macroscopic media 52
4.1 Dielectrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.2 Electric dipoles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.3 Magnetic dipoles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

5 Electrodynamics and Maxwell’s equations 62

5.1 Maxwell’s displacement current . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
5.2 Faraday’s law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5.3 Maxwell’s equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
5.4 Electromagnetic potentials and gauge transformations . . . . . . . . . . . . . . . . 65
5.5 Electromagnetic energy and Poynting’s theorem . . . . . . . . . . . . . . . . . . . . 67
5.6 Time-dependent Green’s function . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.6.1 Moving point charge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5.6.2 Radiation from an accelerated charge . . . . . . . . . . . . . . . . . . . . . 75
5.7 Maxwell’s equations in macroscopic media . . . . . . . . . . . . . . . . . . . . . . . 77

6 Electromagnetic waves 78
6.1 Source-free equations and electromagnetic waves . . . . . . . . . . . . . . . . . . . 78
6.2 Monochromatic plane waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
6.3 Polarization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

ii
 6.4 Reflection and refraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

7 Electromagnetism and Special Relativity 87

A Vector calculus 94
A.1 Vectors in R3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
A.2 Vector operators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
A.3 Integral theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

Reading list

There are many outstanding books on electromagnetism, often going into the subject in much
more depth than we will have time for, especially physical applications. Those who wish to learn
more are encouraged to dip into the following:

• R. P. Feynman, R. B. Leighton, M. Sands, The Feynman Lectures on Physics, Volume 2:

Electromagnetism, Addison-Wesley. Online: https://feynmanlectures.caltech.edu/

• D. J. Griffiths, Introduction to Electrodynamics, Pearson.

• J. D. Jackson, Classical Electrodynamics, John Wiley.

• N. M. J. Woodhouse, Special Relativity, Springer Undergraduate Mathematics.

The Feynman lectures are superb, especially for physical insight and applications. The book by
Griffiths is perhaps closest to this course. The book by Jackson is comprehensive, although is
closer to a graduate level text. The book by Woodhouse is included mainly for those who would
like to pursue the material in section 7 further.

Preamble

In this course we take a first look at the classical theory of electromagnetism. Historically, this
begins with Coulomb’s inverse square law force between stationary point charges, dating from
1785, and culminates (for us, at least) with Maxwell’s formulation of electromagnetism in his 1864
paper, A Dynamical Theory of the Electromagnetic Field. It was in this paper that the electro-
magnetic wave equation was first written down, and in which Maxwell first proposed that “light
is an electromagnetic disturbance propagated through the field according to electromagnetic laws”.
Maxwell’s equations, which appear on the front page of these lecture notes, govern a diverse array
of physical phenomena, and are valid over an enormous range of scales. It is the electromagnetic
force that holds the negatively charged electrons in orbit around the positively charged nucleus of
an atom.1 Interactions between atoms and molecules are also electromagnetic, so that chemical
1
Quantum mechanics also plays an important role.

iii
forces are really electromagnetic forces. The electromagnetic force is then essentially responsible
for almost all physical phenomena encountered in day-to-day experience, with the exception of
gravity: friction, electricity (in homes, laptops, mobile phones, etc), electric motors, permanent
magnets, electromagnets, lightning, electromagnetic radiation (radio waves, microwaves, X-rays,
etc, as well as visible light), . . . it’s all electromagnetism.
Classical electromagnetism is an application of the three-dimensional vector calculus you learned
in Prelims: div, grad, curl, and the Stokes and divergence theorems. Appendix A summarizes the
main definitions and results, and I strongly encourage you to take a look at this at the start of
the course. We’ll then take a usual, fairly historical, route, starting with Coulomb’s law in elec-
trostatics, and eventually building up to Maxwell’s equations on the front page. The disadvantage
of this is that you’ll begin by learning special cases of Maxwell’s equations – having learned one
equation, you will later find that more generally there are other terms in it. On the other hand,
simply starting with Maxwell’s equations and then deriving everything else from them is probably
too abstract, and doesn’t really give a feel for where the equations have come from. My advice
is that after every few lectures you should take another look at the equations on the front page –
each time you should find that you understand better what they mean.

From a long view of the history of mankind – seen from, say, ten thousand years from
now – there can be little doubt that the most significant event of the 19th century will
be judged as Maxwell’s discovery of the laws of electrodynamics – Richard Feynman

iv
1 Electrostatics
1.1 Point charges and Coulomb’s law

It is a fact of nature that elementary particles have a property called electric charge. In SI units2
this is measured in Coulombs C, and the electron and proton carry equal and opposite charges ∓e,
where e ≈ 1.602 × 10−19 C.3 Atoms consist of electrons orbiting a nucleus of protons and neutrons
(with the latter carrying charge 0), and thus all charges in stable matter, made of atoms, arise
from these electron and proton charges.
Electrostatics is the study of charges at rest. We model space by R3 , or a subset thereof, and
represent the position of a stationary point charge q by the position vector r ∈ R3 .

Coulomb’s law Given two charges, q1 , q2 at positions r1 , r2 ∈ R3 , respectively, the first charge
experiences an electrical force F1 due to the second charge given by
1 q1 q2
F1 = (r1 − r2 ) . (1.1)
4πϵ0 |r1 − r2 |3
Note this only makes sense if r1 ̸= r2 , so that the charges are not on top of each other, which we
thus assume. The constant ϵ0 is called the permittivity of free space, which in SI units takes the
value ϵ0 ≃ 8.854 × 10−12 C2 N−1 m−2 . Notice that by symmetry the second charge experiences an
electrical force F2 , due to the first charge, where F2 is given by the right hand side of (1.1) with
the subscripts 1 and 2 swapped. In particular F2 = −F1 , and Newton’s third law is obeyed.

Without loss of generality, we might as well put the second charge at the origin r2 = 0, denote
r1 = r, q2 = q, and equivalently rewrite (1.1) as
1 q1 q
F1 = r̂ , (1.2)
4πϵ0 r2
where r̂ ≡ r/r is a unit vector and r ≡ |r|. This is Coulomb’s law of electrostatics, and is an
experimental fact. Note that:

E1 : The force is proportional to the product of the charges, so that opposite (different
sign ) charges attract, while like (same sign ) charges repel.

E2 : The force acts in the direction of the vector joining the two charges, and is inversely
proportional to the square of the distance of separation.

The above two laws of electrostatics are equivalent to Coulomb’s law. Notice that if (1.2) is the
only force acting on the first charge, by Newton’s second law of motion it will necessarily accelerate
and begin to move, and we are then no longer dealing with statics. We’ll get a feel for electrostatics
problems as the next two sections develop, and look at charges in motion starting in section 3.
The final law of electrostatics says what happens when there are more than two charges:
2
where, for example, distance is measured in metres, time is measured in seconds, force is measured in Newtons.
3
Since 2019, the Coulomb is defined such that the elementary charge is exactly e = 1.602 176 634 · 10−19 C.

1
 E3 : Electrostatic forces obey the Principle of Superposition.

This means that if we have N charges qi at positions ri , i = 1, . . . , N , then an additional charge q

at position r experiences a force
N
X 1 qqi
F = (r − ri ) . (1.3)
4πϵ0 |r − ri |3
i=1

That is, to get the total force on charge q due to all the other charges, we simply add up (superpose)
the Coulomb force (1.1) from each charge qi .

1.2 The electric field

The following looks trivial at first sight, but in fact it’s an ingenious shift of viewpoint:

Definition Given a particular distribution of charges, as above, we define the electric field E =
E(r) to be the force on a unit test charge (i.e. q = 1) placed at position r.

Here the nomenclature “test charge” indicates that the charge is not regarded as part of the
distribution of charges that it is “probing”. The force in (1.3) is thus

F = qE , (1.4)

where by definition
N
1 X qi
E(r) = (r − ri ) (1.5)
4πϵ0 |r − ri |3
i=1

is the electric field produced by the N charges. It is a vector field (here defined on R3 \{r1 , . . . , rN }),
depending on position r.
As we have defined it, the electric field is just a mathematically convenient way of describing the
force a unit test charge would feel if placed in some position in a fixed background of charges. In
fact, the electric field will turn out to be a fundamental object in electromagnetic theory. Notice
that E also satisfies the Principle of Superposition, and that it is measured in N C−1 .

1.3 Gauss’ law

Definition Given a surface Σ ⊂ R3 with outward unit normal vector n, the integral
R
Σ E · n dS
is called the flux of the electric field E through Σ.4

Here one often uses the notation dS for n dS. A central result in electrostatics is:

Theorem 1.1 (Gauss’ law ) For any closed surface Σ = ∂R bounding a region R ⊂ R3 ,
Z N
1 X Q
E · dS = qi = , (1.6)
Σ ϵ0 ϵ0
i=1
4
See appendix A for vector calculus definitions and theorems.

2
where R contains the point charges q1 , . . . , qN , and Q = q1 + · · · + qN is the total charge in R. In
1
words: the flux of the electric field through a closed surface Σ is equal to ϵ0 times the total charge
contained in the region bounded by Σ.

Proof Consider first a point charge q at position r0 . From (1.1), this produces an electric field
q r − r0
E(r) = . (1.7)
4πϵ0 |r − r0 |3

Since ∇ · r = 3 and on R3 \ {r0 } we have ∇ |r − r0 | = (r − r0 )/|r − r0 | (see (A.15)), it follows that

in this domain
 
q 3 3 (r − r0 ) · (r − r0 )
∇·E = − = 0. (1.8)
4πϵ0 |r − r0 |3 |r − r0 |5

The divergence of the electric field E produced by the point charge is thus zero everywhere, except
at the location of the charge itself.
Consider next a sphere S of radius a > 0, centred on the point r0 . Since the outward unit
normal to S is n = (r − r0 )/|r − r0 |, from (1.7) we have
Z Z Z
q 1 q q
E · dS = 2
dS = 2
dS = . (1.9)
S 4πϵ0 S |r − r0 | 4πa ϵ0 S ϵ0

Here we have used the fact that a sphere of radius a has surface area 4πa2 , and |r − r0 | = a on S.

S i bounding ball Bi

charge qi

Σ bounding region R

Figure
SN 1: Region R, with boundary Σ, containing point charges q1 , . . . , qN . We divide R =
i=1 i ∪ U into a small balls Bi around each charge qi , with boundary spheres Si = ∂Bi , and a
B
region U with boundary components Σ and S1 , . . . , SN .

With these results to hand, consider now a region R, with closed boundary Σ = ∂R, and assume
that R contains the charges q1 , . . . , qN at position vectors r1 , . . . , rN ∈ R. Introduce small balls Bi
centred on each charge, such that each Bi contains only the charge qi , and write R = N
S
i=1 Bi ∪ U ,
where the region U then contains no charges – see Figure 1. Each ball Bi has boundary sphere

3
Si , and the outward unit normal to Si = ∂Bi is an inward unit normal to ∂U . The divergence
theorem A.2 applied to U hence gives
Z Z N Z
X
∇ · E dV = E · dS − E · dS . (1.10)
U Σ i=1 Si

The electric field generated by the configuration of charges is given by (1.5), and (1.8) shows that
∇ · E = 0 on R3 \ {r1 , . . . , rN }. Since the region U contains no charges, using (1.10) we have
N Z N Z
Z Z !
X X
E · dS = E · dS − E · dS + E · dS
Σ Σ i=1 Si i=1 Si
Z N Z
X
= ∇ · E dV + E · dS (divergence theorem)
U i=1 Si
N Z N
X X qi
= E · dS = . (1.11)
Si ϵ0
i=1 i=1

Here the last step uses (1.9) for the ith term in the sum (1.5): the remaining terms in the sum
have zero divergence on the ball Bi , and so do not contribute to the integral. ■
Note that this proof starts by analysing a single charge, and the inclusion of multiple charges is
an application of the Principle of Superposition, E3 .

1.4 Charge density and Gauss’ law

For many problems it is not convenient to deal with point charges. If the point charges we have
been discussing are, say, electrons, then a macroscopic object will consist of an absolutely enormous
number of electrons, each with a very tiny charge.5 We thus introduce the concept of charge density
ρ(r), which is a function giving the charge per unit volume. This means that, by definition, the
total charge Q = Q(R) in any region R is
Z
Q = ρ dV . (1.12)
R

We shall generally assume that when describing a smooth three-dimensional distribution of charge,
the function ρ is at least continuous. For the purposes of physical arguments, we shall often think
of the Riemann integral as the limit of a sum (which is what it is). Thus, if δR ⊂ R3 is a small
region centred around a point r ∈ R3 in such a sum, that region contributes a charge ρ(r) δV ,
where δV is the volume of δR.
With this definition, the obvious limit of the sum in (1.5), replacing a point charge q ′ at position
r′ by ρ(r′ ) δV ′ , becomes a volume integral
ρ(r′ )
Z
1
E(r) = (r − r′ ) dV ′ . (1.13)
4πϵ0 r′ ∈R |r − r′ |3
5
A different issue is that, at the microscopic scale, the charge of an electron is not pointlike, but rather is effectively
smeared out into a smooth distribution of charge. In fact in quantum mechanics the precise position of an electron
cannot be measured in principle!

4
Here {r ∈ R3 | ρ(r) ̸= 0} ⊂ R, so that all charge is contained in the (usually bounded) region R.
(1.13) gives the electric field generated by a charge distribution described by the density ρ, where
notice that we have again derived this from Coulomb’s law and the Principle of Superposition.
Similarly, the limit of (1.6) becomes
Z Z
1
E · dS = ρ dV , (1.14)
Σ ϵ0 R
for any region R with boundary ∂R = Σ. Using the divergence theorem we may rewrite this as
Z  
ρ
∇·E− dV = 0 . (1.15)
R ϵ0
Since this holds for all R, we conclude from Lemma A.3 another version of Gauss’ law
ρ
∇·E = . (1.16)
ϵ0

We have derived the first of Maxwell’s equations, on the front page, from the three simple laws of
electrostatics. In fact this equation holds in general, i.e. even when there are magnetic fields and
time dependence. We will directly show that (1.13) satisfies (1.16) in the next subsection.
As already mentioned, when writing a charge density ρ describing a smooth three-dimensional
distribution of charge, we shall generally assume it is at least continuous. However, if (1.16) is to
apply for the density ρ of a point charge q, say at the origin, notice that (1.8) implies that ρ = 0
on R3 \ {0}, but still the integral of ρ over a neighbourhood of the origin is equal to the charge q.
The “function” with this property is q times the Dirac delta function:

Definition The Dirac delta function δ(x) in one dimension is defined by:
(
if 0 ∈ I ⊆ R ,
Z
1
(i) δ(x) = 0 for x ̸= 0 , (ii) δ(x) dx = (1.17)
I 0 otherwise .
One can define this rigorously as a distribution, which is a linear functional on an appropriate space
of test functions. One can also view δ(x) as the limit of a sequence of “bump functions”, supported
around x = 0. For more details the reader might refer to the Part A Integral Transforms course,
but we shall only need an informal understanding of the Dirac delta function for this course. The
following Proposition summarizes some key properties:

Proposition 1.2 The Dirac delta function satisfies:

Z ∞
(i) f (x)δ(x − x′ ) dx = f (x′ ) ,
−∞
1
(ii) δ[a(x − x′ )] = δ(x − x′ ) ,
|a|
n
X δ(x − xi )
(iii) δ(g(x)) = , (1.18)
|g ′ (xi )|
i=1

where (i) f (x) is any continuous function, (ii) a ̸= 0 is a real constant, and (iii) the differentiable
function g(x) has zeros g(xi ) = 0, i = 1, . . . , n, with g ′ (xi ) ̸= 0.

5
The proofs of these essentially follow from the defining property (1.17), where in (iii) one changes
variable y = g(x), dy = g ′ (x) dx in the integral. Notice from (ii) that δ(x − x′ ) = δ(x′ − x).
The Dirac delta function in three dimensions is then defined in Cartesian coordinates r =
(x1 , x2 , x3 ) as the product

δ(r) ≡ δ(x1 ) δ(x2 ) δ(x3 ) . (1.19)

From (1.17) this satisfies

(
if r′ ∈ R ,
Z
1
(i) δ(r − r′ ) = 0 for r ̸= r′ , (ii) δ(r − r′ ) dV = (1.20)
r∈R 0 otherwise .

The density for a point charge q at position r0 is then

ρ(r) = q δ(r − r0 ) . (1.21)

Example As an application of this, starting from (1.16) we may rederive Gauss’ law (1.6) for N
point charges q1 , . . . , qN at positions r1 , . . . , rN inside a region R, with boundary Σ = ∂R:
Z Z Z N N
1 X 1 X
E · dS = ∇ · E dV = qi δ(r − ri ) dV = qi . (1.22)
Σ R r∈R ϵ0 ϵ0
i=1 i=1
PN
Likewise, inserting ρ(r′ ) = i=1 qi δ(r
′ − ri ) into (1.13) gives
Z N N
1 1 ′
X
′ ′ 1 X qi
E(r) = ′ 3
(r − r ) qi δ(r − ri ) dV = (r − ri ) , (1.23)
4πϵ0 r′ ∈R |r − r | 4πϵ0 |r − ri |3
i=1 i=1

which is the point charge formula (1.5). ■

Proposition 1.3 The three-dimensional Dirac delta function may be written as

 
1 1
δ(r − r0 ) = − ∇2 . (1.24)
4π |r − r0 |

Proof This essentially follows from the calculations we have done. On R3 \ {r0 } note that
 
1 (r − r0 ) 4πϵ0
∇ = − = − E(r) , (1.25)
|r − r0 | |r − r0 |3 q

where on the right hand side E(r) is the electric field (1.7) generated by a point charge at posi-
tion r0 . Taking the divergence of (1.25), from (1.8) we immediately deduce that ∇2 (1/|r − r0 |) = 0
for r ̸= r0 , and using the divergence theorem and then (1.9) we have
Z   Z   Z
2 1 1 4πϵ0
∇ dV = ∇ · dS = − E · dS = −4π , (1.26)
r∈B |r − r0 | S=∂B |r − r0 | q S

where B is a ball centred on r0 . These properties establish the identification (1.24). ■

6
1.5 Electrostatic potential and Poisson’s equation

Returning to our point charge q at position r0 , note that on R3 \ {r0 } equation (1.25) implies that
E = −∇ϕ where
q 1
ϕ(r) = . (1.27)
4πϵ0 |r − r0 |
Definition The function ϕ with E = −∇ϕ is called the electrostatic potential.

For a continuous charge density we then have the following:

Theorem 1.4 Consider a continuous charge density ρ(r) with support {r ∈ R3 | ρ(r) ̸= 0} ⊂ R
with R a bounded region, and define
ρ(r′ )
Z
1
ϕ(r) ≡ dV ′ . (1.28)
4πϵ0 r′ ∈R |r − r′ |

Then −∇ϕ = E, with the electric field E given by (1.13), which satisfies Gauss’ law (1.16).
Moreover, ϕ = O(1/r) as r → ∞, so that this electrostatic potential is zero “at infinity”.

Proof The formula (1.13) follows immediately from applying −∇ to (1.28):

Z   
1 1
−∇ϕ(r) = −∇ ρ(r′ ) dV ′
4πϵ0 r′ ∈R |r − r′ |
ρ(r′ )(r − r′ )
Z
1
= dV ′ , (1.29)
4πϵ0 r′ ∈R |r − r′ |3

where in the second line we have used the first equality in (1.25). Instead applying ∇2 to (1.28),
and using Proposition 1.3, we deduce
Z   Z
1 1 ′ ′ 1 ′
∇2 ϕ(r) = ∇2 ρ(r′ ) dV ′


ρ(r ) dV = −4π δ(r − r )
4πϵ0 r′ ∈R |r − r′ | 4πϵ0 r′ ∈R
ρ(r)
= − . (1.30)
ϵ0
It follows that ∇ · E = ρ/ϵ0 , where E = −∇ϕ. Finally, R being bounded means it is contained
inside some closed ball B of radius a, and ρ is continuous on this compact set so it is bounded.
Thus |ρ(r′ )| ≤ M for all r′ ∈ B, and

ρ(r′ )
Z Z
1 ′ 1 M M
|ϕ(r)| = ′
dV ≤ ′
dV ′ ≤ Vol(Ba ) , (1.31)
4πϵ0 r′ ∈B |r − r | 4πϵ0 r′ ∈B |r − r | r−a

where in the last step we have taken r > a, and used the reverse triangle inequality: |r − r′ | ≥
||r| − |r′ || = r − |r′ | ≥ r − a. ■

Corollary 1.5 The electrostatic potential satisfies Poisson’s equation

ρ
∇2 ϕ = − . (1.32)
ϵ0

7
 Since the curl of a gradient is identically zero, we may also deduce from E = −∇ϕ that

∇×E = 0 . (1.33)

Equation (1.33) is another of Maxwell’s equations from the front page, albeit only in the special
case where ∂B/∂t = 0 (the magnetic field is time-independent).

* In fact ∇ × E = 0 implies that the vector field E is the gradient of a function, provided
the domain of definition is simply-connected. Recall the latter means that every closed
loop can be continuously contracted to a point. For example, this is true for R3 or in an
open ball. For non-simply-connected domains, such as R3 minus a line (say, the z-axis), it
is not always possible to write a vector field E with zero curl as a gradient. A systematic
discussion of this is certainly beyond this course. The interested reader can find a proof
for an open ball in appendix B of the book by Woodhouse in the reading list.

Recall from Prelims that forces F which are gradients are called conservative forces. Since
F = q E, we see that the electrostatic force is conservative. The work done against the electrostatic
force in moving a charge q along a curve C is then the line integral
Z Z Z
W = − F · dr = −q E · dr = q ∇ϕ · dr = q [ϕ(r1 ) − ϕ(r0 )] . (1.34)
C C C

Here the curve C begins at r0 and ends at r1 . The work done is of course independent of the
choice of curve connecting the two points, because the force is conservative. Notice that one may
add a constant to ϕ without changing E. It is only the difference in values of ϕ that is physical,
and this is called the voltage. If we fix some arbitrary point r0 and choose ϕ(r0 ) = 0, then ϕ(r)
has the interpretation of work done against the electric field in moving a unit charge from r0 to r.
Note that Theorem 1.4 says that the particular choice for ϕ in (1.28) is zero “at infinity”. From
the usual relation between work and energy, ϕ is also the potential energy per unit charge.

E field
ϕ = constant

Figure 2: The field lines, which represent the direction of the electric field E, and equipotentials
around a positive point charge.

8
Definition Surfaces of constant ϕ are called equipotentials.

Proposition 1.6 The electric field is always normal to an equipotential surface.

Proof To see this, fix a point r and let t be a tangent vector to an equipotential for ϕ(r). By
definition, the derivative of ϕ(r) in a direction tangent to such an equipotential surface is zero, and
hence t · ∇ϕ = 0 at this point. Since this holds for all tangent vectors, this means that ∇ϕ = −E
is normal to a surface of constant ϕ. ■

1.6 Conductors and surface charge

More interesting is when the distribution of charge is not described by a continuous charge density.
We have already encountered point charges. For many problems it is useful to introduce the
concepts of surface charge density σ on a surface S, say for a charge distribution on a thin metal
sheet, and also line charge density λ on a curve C, say for a charge distribution in a thin wire.
These will be taken to be appropriately well-behaved functions on S and C, representing charge
per unit area and charge per unit length, respectively.
In fact the concept of surface charge density doesn’t require a thin metal sheet to be useful, for
the following reason:

Definition An electrical conductor is a material where some of the electrons (“conduction elec-
trons”) are free to move in the presence of an external electric field.

In a static situation, the electric field inside the conducting material must be zero. Why? Because
if it weren’t, then the conduction electrons in the interior would experience a force, and thus move
by Newton’s second law.
Imagine what happens if we now switch on an external electric field: a conduction electron will
move in the opposite direction to the field (because it is negatively charged), until either (a) it
gets to the boundary of the material, or (b) the electric field inside the material has relaxed to its
equilibrium of zero. This way, one ends up with lots of electrons at, or very near, the surface of
the material; their distribution (and the distribution of other immobile charges) throughout the
material produces an electric field which precisely cancels the external field inside the material.
Thus E = 0 inside a conductor. More generally if the conductor is a surface or curve in R3 , then
similarly t · E = 0 for any tangent vector t to the conductor. Since E = −∇ϕ we deduce the
important fact that:

Theorem 1.7 (a “Physics Theorem” ) A conductor in static equilibrium is always an equipoten-

tial for ϕ, i.e. ϕ = constant throughout the conducting material.

Equation (1.16) furthermore implies that ρ = 0 inside a conducting material in equilibrium, and
hence the charge must be described by a surface charge density.

9
 Surface and line charge densities of course contribute to the total charge and electric field via
surface and line integrals, respectively. For example, a surface S with surface charge density σ
gives rise to an electric field
σ(r′ )
Z
1
E(r) = (r − r′ ) dS ′ . (1.35)
4πϵ0 r′ ∈S |r − r′ |3

Notice that for r ∈ R3 \ S and σ smooth the integrand is smooth. However, it turns out that E is
not continuous across S!

Proposition 1.8 For the electric field given by (1.35), generated by a surface charge density σ on
a surface S, the components of E tangent to S are continuous across S, but the normal component
of E is not. Specifically, if n is a unit normal vector field to the surface pointing into what we’ll
call the “+ side”, then
σ
E + · n − E− · n = , (1.36)
ϵ0
at every point on the surface.

Proof To see this, consider a surface S which has a surface charge density σ. Consider the
cylindrical region R on left hand side of Figure 3, of height ε and cross-sectional area δA. Gauss’
law gives
Z
1
E · dS = (total charge in R) . (1.37)
∂R ϵ0

In the limit ε → 0 the left hand side becomes (E+ · n − E− · n) δA for small δA, where E± are
the electric fields on the two sides of S and the unit normal n points into the + side. The right
hand side, on the other hand, tends to σ δA/ϵ0 . Thus there is necessarily a discontinuity in the
component of E normal to S given by (1.36).

area δ A
E+ t curve C
n bounding Σ

height ε height ε

-
E
surface S length δL
region R
Figure 3: The surface S.

10
 Consider, instead, the rectangular loop C on the right hand side of Figure 3, of height ε and
length δL, bounding the rectangular surface Σ. By Stokes’ theorem A.1 we have
Z Z
E · dr = (∇ × E) · dS = 0 , (1.38)
C Σ

where we have used the electrostatic Maxwell equation (1.33) in the second equality. If t denotes
a unit tangent vector along C on the + side, then in the limit ε → 0 we obtain

E+ · t − E− · t δL = 0 ,


(1.39)

for small δL. Thus the components of E tangent to S are continuous across S

E+ · t = E− · t . ■ (1.40)

e3 z

r = (0,0, b )

b y

r' = (a cos θ', a sin θ', 0)

a
θ' x
O

charged wire C

Figure 4: Charged plane circular wire C of radius a, centred on the origin O in the (x, y)-plane at
z = 0.

Example (Line charge density) Consider a curve C with line charge density λ. This generates an
electric field
λ(r′ ) dr′
Z
1 ′
E(r) = (r − r ) ds′ , (1.41)
4πϵ0 r′ ∈C |r − r′ |3 ds′

where the curve C ⊂ R3 is parametrized by r′ = r′ (s′ ), with s′ ∈ [s0 , s1 ] ⊂ R. For example, consider
a plane circular wire of radius a, centred on the origin in the (x, y)-plane at z = 0, carrying a total
charge Q that is uniformly distributed around the wire – see Figure 4. A point on the wire is
r′ (θ′ ) = (a cos θ′ , a sin θ′ , 0), parametrized by θ′ ∈ [0, 2π], and so |dr′ /dθ′ | = a. The uniform line
charge density is then λ = Q/2πa, and the electric field (1.41) is

Q r − r′ (θ′ )
Z 2π
1
E(r) = a dθ′ . (1.42)
4πϵ0 θ′ =0 2πa |r − r′ (θ′ )|3

11
It is difficult to evaluate this integral at a general point r = (x, y, z), so let us look at the point
r = (0, 0, b) on the z-axis. In this case notice that |r − r′ (θ′ )| = |(0 − a cos θ′ , 0 − a sin θ′ , b )| =
√
a2 + b2 is independent of θ′ , and (1.42) easily integrates to give
Z 2π
1 Q 1 ′ ′

′ Q b
E((0, 0, b)) = −a cos θ , −a sin θ , b dθ = e3 ,(1.43)
4πϵ0 2π θ′ =0 (a2 + b2 )3/2 4πϵ0 (a2 + b2 )3/2

where e3 is a unit vector pointing along the z-axis. Here the integrals over cos θ′ and sin θ′ in the
x and y components of (1.43) give zero, but we could also have anticipated this by the symmetry
of the problem: the z-axis is an axis of symmetry, and on this locus the electric field then also
necessarily points along the z-axis. ■

Figure 5: Electric field lines around the plane circular wire, given by (1.42).

1.7 Electrostatic energy

In this subsection we derive a formula for the energy of an electrostatic configuration as an integral
of a local energy density. We shall return to this subject again in section 5.5.
We begin with a point charge q1 at r1 . This generates a potential
1 q1
ϕ(1) (r) = . (1.44)
4πϵ0 |r − r1 |

Consider now moving a charge q2 from infinity to the point r2 . From (1.34), the work done against
the electric field in doing this is
1 q2 q1
W2 = q2 ϕ(1) (r2 ) = . (1.45)
4πϵ0 |r2 − r1 |

12
Next move in another charge q3 from infinity to the point r3 . We must now do work against the
electric fields of both q1 and q2 . By the Principle of Superposition, this work done is
 
1 q3 q1 q3 q2
W3 = + . (1.46)
4πϵ0 |r3 − r1 | |r3 − r2 |
The total work done so far is thus W2 + W3 . We may continue this process and inductively deduce
that the total work done in assembling charges q1 , . . . , qN at r1 , . . . , rN is
N N
1 X X qi qj 1 1 X X qi qj
W = = · , (1.47)
4πϵ0 |ri − rj | 2 4πϵ0 |ri − rj |
i=1 j<i i=1 j̸=i

where notice that qi qj /|ri − rj | is symmetric under swapping i and j.

We next rewrite (1.47) as
N
1X
W = qi ϕ i , (1.48)
2
i=1

where we have defined

1 X qj
ϕi ≡ . (1.49)
4πϵ0 |ri − rj |
j̸=i

This is simply the electrostatic potential produced by all but the ith charge, evaluated at position
ri . In the usual continuum limit, (1.48) becomes
Z
1
W = ρ ϕ dV , (1.50)
2 R
where ϕ(r) is given by (1.28). Now, using Gauss’ law (1.16) we may write
ρ
ϕ = ϕ ∇ · E = ∇ · (ϕ E) − ∇ϕ · E = ∇ · (ϕ E) + E · E , (1.51)
ϵ0
where in the last step we used E = −∇ϕ. Inserting this into (1.50), we have
Z Z 
ϵ0
W = ϕ E · dS + E · E dV , (1.52)
2 Σ=∂R R

where we have used the divergence theorem on the first term. Taking R to be a very large ball of
radius r, enclosing all charge, the surface Σ is a sphere. From Theorem 1.4 we have ϕ = O(1/r)
as r → ∞, and one can check that this surface term is zero in the limit that the ball becomes
infinitely large. We hence deduce the elegant formula
Z
ϵ0
W = E · E dV . (1.53)
2 R3
Definition When the integral (1.53) exists the electrostatic configuration is said to have finite
energy W . The formula (1.53) suggests that the energy is stored in a local energy density
ϵ0 ϵ0
E ≡ E·E = |E|2 . (1.54)
2 2

13
2 Boundary value problems in electrostatics
2.1 Boundary value problems

In the previous section we have seen that electrostatics reduces to solving the Poisson equation
ρ
∇2 ϕ = − . (2.1)
ϵ0
For problems where we are simply given a charge distribution ρ everywhere in space, the solution
to (2.1) for ϕ is given by the integral formula (1.28) in Theorem 1.4. We have then effectively
solved this class of electrostatics problems.
In this section we consider a different class of problems. We still want to solve the Poisson
equation (2.1), but now in a region of space R with boundary ∂R = Σ, where we specify boundary
conditions for the fields on Σ. From a mathematical perspective, given a PDE such as (2.1) it is
natural to ask what types of boundary condition we can impose. In this section we consider:

(i) Dirichlet boundary conditions, where ϕ is specified on Σ,

(ii) Neumann boundary conditions, where E · n is specified on Σ, where n is the outward unit
normal to Σ = ∂R. Here the normal component of the electric field E · n on Σ is
∂ϕ
E · n = −n · ∇ϕ ≡ − , (2.2)
∂n
which is minus the normal derivative of ϕ.

From a mathematical perspective it makes sense to solve (2.1) only on the domain R ⊆ R3 , ignoring
the exterior (complement) of R. However, in physical problems there should be something that is
effectively imposing these boundary conditions, and indeed then something outside the domain R.
Here are two classes of examples:

(i) Consider a charge distribution described by a density ρ in the interior of a region R, where
∂R = Σ and this boundary is surrounded by a thin conducting material – see Figure 6. As
explained in section 1.6, the conduction electrons will distribute themselves in such a way
to give ϕ = constant inside the thin layer of conducting material, but also induce a surface
charge density σ on Σ, that we do not know a priori. However, given the solution ϕ to the
Dirichlet problem, on Σ we may identify

∂ϕ
σ = ϵ0 . (2.3)
∂n Σ

This follows from (1.36) in Proposition 1.8, where inside the thin conducting material we
have E+ = 0, and we have then used (2.2) to write E− · n = −∂ϕ/∂n, where E− is the
electric field just inside the boundary Σ. Of course, the solution will determine E = −∇ϕ in
the whole interior of R.

14
 thin conducting material
with ϕ = constant inside
the thin layer

ρ specified in region R

boundary ∂ R = Σ

induced surface charge

density σ = ε 0 ∂∂nϕ
Σ
Figure 6: The Dirichlet problem with ϕ = constant on the boundary ∂R = Σ of a region R
containing a charge density ρ. Physically this boundary condition is enforced by wrapping a thin
conducting material around Σ, with ϕ = constant inside this thin layer.

(ii) Consider the same set up, but where now Σ is replaced by a thin layer of insulating material,
on which we place a specific surface charge density σ. By definition the insulator doesn’t
allow these charges to move. Assuming that E+ = 0 in the exterior of R, the same equation
(2.3) still holds, but where we now interpret this as fixing the normal derivative of ϕ on Σ.

As we explain at the end of this subsection, the electric field will also be zero outside the thin
conducting material, provided there are no charges outside of R and the conductor is grounded,
meaning the potential ϕ on Σ is the same as that “at infinity”. We shall see various physical
examples later in this section, but for now we focus mainly on developing the general mathematical
theory. We begin with the following:

Proposition 2.1 (Green’s identities ) For any closed surface Σ bounding a region R, and for all
suitably differentiable functions u, v in R, we have
Z Z
2

∂v
(a) u ∇ v + ∇u · ∇v dV = u dS ,
R Σ ∂n
Z Z  
2 2

∂v ∂u
(b) u ∇ v − v ∇ u dV = u −v dS . (2.4)
R Σ ∂n ∂n

Proof For (a) apply the divergence theorem A.2 to f = u ∇v, where then ∇ · f = u ∇2 v + ∇u · ∇v.
Then for (b) simply take (a) and subtract the same equation with u and v interchanged. ■

Theorem 2.2 The Poisson equation with either Dirichlet or Neumann boundary condition has a
unique solution for ϕ, in the latter case up to an unphysical additive constant.

Proof Let ϕ1 , ϕ2 be two solutions to (2.1) in the interior of a region R with closed boundary
surface Σ = ∂R, with the same boundary conditions on Σ. Define ψ ≡ ϕ1 − ϕ2 , so that the

15
Laplace equation
 
2 2 2 ρ ρ
∇ ψ = ∇ ϕ1 − ∇ ϕ2 = − − − = 0, (2.5)
ϵ0 ϵ0

holds inside R, with boundary conditions either (i) ψ = 0 on Σ, or (ii) ∂ψ/∂n = 0 on Σ. Now
apply Green’s first identity (a) in Proposition 2.1, with u = v = ψ, to deduce
Z Z
∂ψ
ψ ∇2 ψ + ∇ψ · ∇ψ dV =


ψ dS = 0 . (2.6)
R Σ ∂n

Since ∇2 ψ = 0 this gives R |∇ψ|2 dV = 0. Notice that ∇ψ needs to be differentiable in order

R

for the Laplace equation to make sense, so the integrand |∇ψ|2 is manifestly non-negative and
continuous. We deduce that ∇ψ = 0 and hence ψ = constant in R, so that ϕ1 = ϕ2 + constant.
Recall that the electrostatic potential is only determined up to such an additive constant, which
drops out of the electric field E = −∇ϕ. However, in the Dirichlet case ϕ1 = ϕ2 on the boundary
then forces this to hold everywhere in R, with ψ ≡ 0. ■

In principle we could consider mixed boundary conditions, writing Σ = ΣD ∪ ΣN as a disjoint

union, with ϕ specified on ΣD , and ∂ϕ/∂n specified on ΣN . The uniqueness proof in Theorem 2.2
still goes through, where on the right hand side of (2.6) either ψ = 0 or ∂ψ/∂n = 0 on the
respective components of Σ. The following is an important physical application of Theorem 2.2:

Theorem 2.3 Consider a region R with zero charge density ρ inside R, which is bounded by a
closed surface Σ = ∂R consisting of a thin electrical conductor. Then the electric field is zero
inside R.

Proof From Theorem 1.7, the surface Σ is an equipotential for ϕ. On the other hand, since ρ = 0
inside R then ϕ = constant inside R solves the Poisson equation with Dirichlet boundary condition.
Theorem 2.2 implies this is the unique solution, and we deduce that E = −∇ϕ = 0 inside R. ■

This is sometimes referred to as a Faraday cage: the thin conductor “shields” its interior from any
external electric field. This is why it’s hard to get a mobile phone signal inside a building where
the walls have been reinforced with a steel mesh – that phone signal is an electromagnetic wave.
Finally, recall that in Figure 6 we had a thin conducting material wrapped around Σ. Assuming
there is no charge in Rc ≡ R3 \ R, then the charge is confined to a bounded region in R3 , and
Theorem 1.4 says that ϕ = O(1/r) at infinity. By definition, the conductor is grounded if ϕ = 0
on Σ, so that the potential is the same as that at infinity. We may then consider the Dirichlet
problem for Rc , which has two boundary components: Σ and a sphere of very large radius r → ∞.
Since ϕ = 0 on this boundary, then ϕ = 0 inside Rc solves the Poisson equation with Dirichlet
boundary condition, and it follows that E = 0 on Rc = R3 \ R.

16
2.2 Green’s functions

To investigate these boundary value problems further, we next introduce:

Definition A Green’s function is a function G(r, r′ ) satisfying

∇′2 G(r, r′ ) = −4π δ(r − r′ ) , (2.7)

where r, r′ ∈ R. Notice we have written the derivative with respect to the second, primed coordi-
nates r′ in G(r, r′ ), and recall δ(r − r′ ) = δ(r′ − r).

From equation (1.24), we can write down the particular solution to (2.7)
1
G(r, r′ ) = . (2.8)
|r − r′ |

Physically, this is 4πϵ0 times the electrostatic potential generated by a unit charge at position r′ .
On the other hand, this solution is not unique: the general solution to (2.7) is
1
G(r, r′ ) = + F (r, r′ ) , (2.9)
|r − r′ |

where F (r, r′ ) satisfies the Laplace equation in R:

∇′2 F (r, r′ ) = 0 . (2.10)

To see how this helps us solve the Poisson equation (2.1), take Green’s second identity (b)
in Proposition 2.1, with u = ϕ(r′ ) and v = G(r, r′ ), where the integrals are over the primed
coordinates r′ . Then
Z  ′ ′
Z 
′ ∂G(r, r ) ′ ∂ϕ(r )
 ′ ′2 ′ ′ ′2 ′ ′
dS ′ ,(2.11)

ϕ(r ) ∇ G(r, r ) − G(r, r ) ∇ ϕ(r ) dV = ϕ(r ) ′
− G(r, r ) ′
R Σ ∂n ∂n

where this holds for all r ∈ R. Using the definition (2.7) and the Poisson equation (2.1), the left
hand side of (2.11) is
Z
1
−4π ϕ(r) + G(r, r′ ) ρ(r′ ) dV ′ , (2.12)
ϵ0 R

so that rearranging (2.11) we have proven

Proposition 2.4 The electrostatic potential inside the region R may be expressed as
′ ′
Z Z  
1 ′ ′ ′ 1 ′ ∂ϕ(r ) ′ ∂G(r, r )
ϕ(r) = G(r, r ) ρ(r ) dV + G(r, r ) − ϕ(r ) dS ′ . (2.13)
4πϵ0 R 4π Σ=∂R ∂n′ ∂n′

This is a key formula, so let us make some remarks:

• The Green’s function G(r, r′ ) in (2.13) is any solution to (2.7). In particular, we may choose
any solution F to the Laplace equation (2.10) in constructing G(r, r′ ) via (2.9).

17
 • The density ρ in R is specified as part of the data of the problem, while (i) in the Dirichlet
problem it is ϕ that is specified on the boundary, while (ii) in the Neumann problem it is
∂ϕ/∂n that is specified. However, both terms appear in the boundary intergal in (2.13).

• We immediately recognize the solution (1.28) as a special case of (2.13), where we take the
Green’s function (2.8). In (1.28) we took the charge density ρ to be supported in a bounded
region of space, and ϕ(r) given by (1.28) then solves the Poisson equation in the whole of
R3 . The boundary terms in (2.13) are then effectively evaluated at infinity, where both the
Green’s function (2.8) and the potential ϕ are O(1/r) and hence tend to zero.

Let us analyse (2.13) for the two boundary value problems in more detail:

(i) Dirichlet When ϕ is specified on Σ, it is convenient to choose the Green’s function to also
satisfy the Dirichlet boundary condition (denoted with a subscript D on GD )

GD (r, r′ ) = 0 , for all r′ ∈ Σ , r ∈ R . (2.14)

That is, the Green’s function, viewed as a function of the second variable r′ , for any fixed r,
is zero on Σ = ∂R. Given such a function, (2.13) becomes
∂GD (r, r′ )
Z Z
1 ′ ′ ′ 1
ϕ(r) = GD (r, r ) ρ(r ) dV − ϕ(r′ ) dS ′ . (2.15)
4πϵ0 R 4π Σ ∂n′
Notice we have effectively reduced the problem to constructing the Green’s function solution
G = GD to (2.7), with Dirichlet boundary condition (2.14). This depends only on the region
R, with boundary Σ = ∂R. Assuming we know GD , then given any charge density ρ and
any prescribed ϕ on Σ, the (unique by Theorem 2.2) solution to the Poisson equation with
this Dirichlet boundary data is (2.15).

(ii) Neumann In this case we begin by using the divergence theorem to show that for r ∈ R
Z Z Z
′ ′ ′2 ′ ′
−4π = −4π δ(r − r ) dV = ∇ G(r, r ) dV = ∇′ · ∇′ G(r, r′ ) dV ′
R R R
∂G(r, r′ )
Z Z
′ ′ ′
= ∇ G(r, r ) · dS = ′
dS ′ . (2.16)
Σ Σ ∂n
Thus we cannot simply set ∂G(r, r′ )/∂n′ to zero on Σ, analogously to (2.14). Instead a
convenient condition to impose is
∂GN (r, r′ ) 4π
′
= − , for all r′ ∈ Σ , r ∈ R , (2.17)
∂n A
R
where A = ΣdS is the area of Σ. This is consistent with (2.16), and (2.13) becomes
∂ϕ(r′ )
Z Z
1 ′ ′ ′ 1
ϕ(r) = GN (r, r ) ρ(r ) dV + GN (r, r′ ) dS ′ + ⟨ ϕ ⟩ , (2.18)
4πϵ0 R 4π Σ ∂n′
where we have defined
Z
1
⟨ϕ⟩ ≡ ϕ(r′ ) dS ′ , (2.19)
A Σ

18
 which is an average of ϕ, per unit area, on the boundary. Notice that in the Neumann
problem we are free to shift ϕ → ϕ + constant, without changing the boundary condition,
and using this we may set ⟨ ϕ ⟩ to any value (for example, ⟨ ϕ ⟩ = 0 is natural), thus fixing
this non-uniqueness. Having found the Green’s function GN satisfying (2.17), again given
any charge density ρ and any prescribed ∂ϕ/∂n on Σ, the solution to the Poisson equation
with this Neumann boundary data is (2.18).

In summary, we have reduced both problems to finding Green’s functions (2.7) satisfying either
the Dirichlet (2.14) or Neumann (2.17) boundary conditions. It is a general theorem in analysis
that such Green’s functions do in fact exist, assuming the boundary Σ = ∂R is suitably well-
behaved (for example, smooth is sufficient). Indeed, notice from (2.9) that finding such Green’s
functions is equivalent to solving the Laplace equation for F (r, r′ ), with appropriate boundary
conditions. However, finding such functions explicitly is in general very hard. In the remainder of
this section we will study methods that allow us to solve a variety of boundary value electrostatics
problems, and find the Green’s functions for the associated domains R. We focus mainly on the
Dirichlet problem, where we note the following result:

Proposition 2.5 The Dirichlet Green’s function GD , solving (2.7) subject to the boundary con-
dition (2.14), is symmetric; that is, GD (r1 , r2 ) = GD (r2 , r1 ) holds for all r1 , r2 ∈ R.

Proof We again make use of Green’s second identity (b) in Proposition 2.1, this time with u =
G(r1 , r′ ) and v = G(r2 , r′ ), where the integrals are over the primed coordinates r′ . We deduce
Z
GD (r1 , r′ ) ∇′2 GD (r2 , r′ ) − GD (r2 , r′ ) ∇′2 GD (r1 , r′ ) dV ′
 
4π [GD (r2 , r1 ) − GD (r1 , r2 )] =
ZR  ′ ′

′ ∂GD (r2 , r ) ′ ∂GD (r1 , r )
= GD (r1 , r ) ′
− GD (r2 , r ) ′
dS ′ ,
Σ ∂n ∂n
= 0, (2.20)

where the boundary term is immediately zero due to the Dirichlet condition (2.14). ■

Remark Notice the Green’s function G(r, r′ ) = 1/|r − r′ | in (2.8) is the Dirichlet Green’s function
on R3 that is zero on the sphere at infinity, and is indeed symmetric.

2.3 Method of images

Consider the Dirichlet problem in a region R ⊂ R3 with boundary ∂R = Σ, where for simplicity
we begin with a point charge distribution in R. Thus r1 , . . . , rN ∈ R are the locations of point
charges q1 , . . . , qN , and from (1.27) we know these generate an electrostatic potential
N
1 X qi
ϕ(r)point charges ≡ . (2.21)
4πϵ0 |r − ri |
i=1

19
 PN
This solves the Poisson equation with density ρ(r) = i=1 qi δ(r − ri ) everywhere in R3 , and hence
in R ⊂ R3 , but it will not in general satisfy the prescribed Dirichlet boundary condition for ϕ on
Σ, which a priori is arbitrary.
We have already seen a hint of how to fix this in the last subsection: as in (2.9), write

ϕ(r) = ϕ(r)point charges + F (r) . (2.22)

If F (r) satisfies the Laplace equation in R, then ϕ satisfies the same Poisson equation in R.

The method of images Take F (r) to be the electrostatic potential generated by a set of charges
q1⋆ , . . . , qM
⋆ that lie outside the domain R, i.e. set

M
1 X qi⋆
F (r) = ≡ ϕ(r)image charges , (2.23)
4πϵ0 |r − r⋆i |
i=1

with r⋆i ∈ R3 \ R, so that ∇2 F = 0 holds inside R. The idea is to choose these charges and their
locations so that (2.22) satisfies the prescribed boundary condition for ϕ on Σ = ∂R.

image charges qi★

point charges qi

ϕ specified on ∂ R = Σ induced ϕ on ∂ R = Σ

Figure 7: On the left hand side: the original Dirichlet problem, with point charges q1 , . . . , qN inside
R, and ϕ specified on ∂R = Σ. Replace with the right hand side: image charges q1⋆ , . . . , qM ⋆ are
3
added outside R, so that the solution to the Poisson equation on R induces the given ϕ on Σ.

The additional charges q1⋆ , . . . , qM

⋆ are known as image charges (see the example below). They

aren’t part of the original problem, since they lie outside the domain R. Adding them to an
enlarged domain is just a mathematical trick to solve the original boundary value problem on R.
Of course, there is an art to this: how do we know where to put these image charges? There is no
simple answer in general, but in certain problems it is clear from the geometry.
That is the mathematical problem, but what about in physical problems? Recall that in sec-
tion 2.1 we described how the Dirichlet problem arises naturally when the boundary ∂R = Σ is
surrounded by a conducting material. This forces ϕ to be constant on Σ, and if the conductor is

20
grounded then in fact ϕ |R3 \R = 0, as explained at the end of section 2.1. As in Figure 6, this
will result in a discontinuity of the normal derivative ∂ϕ/∂n across Σ, which we then interpret
physically in terms of a surface charge density on Σ, due to conduction electrons. The method of
images gives us a mathematical way to solve this physics problem: forget about the conductor and
the fact that we want ϕ = 0 in the region external to R, and instead introduce fictitious image
charges in this region, as in (2.23), so that the potential in (2.22) has ϕ = 0 on Σ. Once we have
this solution in R3 , which by construction solves the correct Poisson equation inside R with the
correct boundary condition ϕ |Σ = 0, now simply set ϕ = 0 also outside R to obtain the solution
to the original physics problem.

Example (Infinite conducting plane) Consider an infinite conducting plane at {z = 0} ⊂ R3 , and

place a point charge q at position vector r0 = (x0 , y0 , z0 ), with z0 > 0. We want to solve the
Poisson equation for ϕ in the region R ≡ {z ≥ 0}, namely
q
∇2 ϕ(r) = − δ(r − r0 ) , for z > 0 , (2.24)
ϵ0
with the boundary condition that ϕ |Σ = 0 on the conducting plane boundary Σ ≡ {z = 0}.
As in (2.21) we can write down
1 q
ϕ(r)point charge ≡ . (2.25)
4πϵ0 |r − r0 |
This solves the Poisson equation in R, but it does not satisfy the boundary condition that ϕ |Σ = 0.
The problem is that for a point charge source the electric field points radially outwards, as in
Figure 2, and the electric field lines then do not generically cross Σ = {z = 0} perpendicularly.
However, we can solve this problem with the method of images: place an image charge q ⋆ ≡ −q
at the mirror image point to r0

r⋆0 ≡ (x0 , y0 , −z0 ) , (2.26)

across the plane Σ. The electric field generated by this charge has the same magnitude as the
original charge, but points in the opposite direction. If we superpose these electric fields generated
by opposite charges at mirror image points across Σ, the net electric field tangent to Σ will be zero
– see Figure 8. To see this explicitly, following (2.22), (2.23), put
1 q 1 q
ϕ(r) = ϕ(r)point charge + ϕ(r)image charge = − . (2.27)
4πϵ0 |r − r0 | 4πϵ0 |r − r⋆0 |
This satisfies:

(i) ϕ |Σ = 0, where restricting to Σ means setting r = (x, y, 0). (Geometrically, this is because
any point on Σ is equidistant from r0 and its mirror image point r⋆0 through Σ.)

(ii) The Poisson equation in R3 :

q q
∇2 ϕ(r) = − δ(r − r0 ) + δ(r − r⋆0 ) . (2.28)
ϵ0 ϵ0

21
In particular this satisfies the correct Poisson equation inside R = {z ≥ 0}, given in (2.24),
precisely because the image charge lies outside of R. Having found this solution in R with the
correct boundary condition on Σ, the solution to our original physics problem is given by taking
ϕ(r) to be (2.27) for z > 0, and ϕ ≡ 0 for z ≤ 0.

Figure 8: Electric field lines for E = −∇ϕ, with ϕ(r) given by (2.27). The original charge q
is shown in black, and is taken to be positive, while the image charge q ⋆ = −q is red. Note E
is perpendicular to the (x, y)-plane Σ = {z = 0}, shown in blue. The solution to the Dirichlet
problem for z ≥ 0 simply discards the red part of the figure, beneath the conducting plane Σ.

Having found ϕ, we can now compute the induced surface charge density σ on the conducting
plane Σ. This is given by (2.3), where the outward unit normal to Σ is n = −e3 . For simplicity
we set r0 = (0, 0, z0 ), and compute (careful with the signs)
" #
∂ϕ q ∂ 1 1
σ = − ϵ0 = − p −p
∂z Σ 4π ∂z x2 + y 2 + (z − z0 )2 x2 + y 2 + (z + z0 )2 z=0
qz0 1
= − . (2.29)
2π (x + y + z02 )3/2
2 2

Notice this is rotationally symmetric about the origin {x = y = 0}, where σ takes its maximum
value, and tends to zero at infinity. The total charge Q induced on the conducting plane is
qz0 ∞
Z Z Z ∞
1
Q ≡ σ dS = − dx dy = −q . (2.30)
Σ 2π x=−∞ y=−∞ (x + y + z02 )3/2
2 2

Notice this is the same as the image charge q ⋆ = −q.

Using the above results, we may now also construct the Dirichlet Green’s function GD in the
region R = {z ≥ 0}. Recall from section 2.2 this satisfies
(
∇′2 GD (r, r′ ) = −4π δ(r − r′ ) , r′ , r ∈ R ,
(2.31)
GD (r, r′ ) = 0 , r′ ∈ Σ , r ∈ R .

22
But this is solved by taking GD to be 4πϵ0 /q times the solution we have already obtained for ϕ,
where r′ = r0 is the location of the point charge. We can hence simply write down the solution
1 1
GD (r, r′ ) = − , (2.32)
|r − r′ | |r − r′⋆ |
where r′ ≡ (x′ , y ′ , z ′ ), and r′⋆ ≡ (x′ , y ′ , −z ′ ) is the reflection of r′ across Σ = {z = 0}. Comparing
to (2.9), the second term on the right hand side may be identified with F (r, r′ ), which satisfies the
Laplace equation in R, as r′⋆ lies outside of R.
Using (2.15) we may then write down the potential ϕ(r) for any distribution of charge ρ(r) in
the region R:
Z
1
ϕ(r) = GD (r, r′ ) ρ(r′ ) dV ′ , (2.33)
4πϵ0 R

where we have used the boundary condition ϕ |Σ = 0, so that the boundary term in (2.15) is zero.
For example, for the original problem of a point charge q at position r0 we have ρ(r) = q δ(r − r0 ),
q
and (2.33) gives ϕ(r) = 4πϵ0 G(r, r0 ), which is the solution (2.27). For a general charge distribution
ρ in R we may write (2.33) as
ρ(r′ ) ρ(r′ )
Z Z
1 1
ϕ(r) = ′
dV ′ − dV ′ . (2.34)
4πϵ0 R |r − r | 4πϵ0 R |r − r′⋆ |
The first term is the usual solution (1.28) for the electrostatic potential generated by the charge
distribution ρ in R, while the second term is the electrostatic potential generated by the image of
minus this charge distribution, after reflection across Σ. ■

Example (Conducting sphere) Consider a point charge q placed at a point r0 outside a grounded
conducting sphere of radius a, centred on the origin. From Theorem 2.3, the electrostatic potential
is zero on and inside the sphere, so that ϕ(r) = 0 for all r with |r| ≤ a. Write r0 = r0 r̂0 , where r̂0
is a unit vector pointing from the origin towards the charge q, with that latter a distance r0 > a
from the origin.
It is less clear how to use the method of images in this case, but the simplest possibility would
be to use a single image charge q ⋆ at a point inside the sphere. By symmetry this image charge
must lie on the line joining the origin to the original charge q, so we write the ansatz
q⋆
 
1 q
ϕ(r) = + , (2.35)
4πϵ0 |r − r0 | |r − r⋆0 |
where r⋆0 = r0⋆ r̂0 , with 0 ≤ r0⋆ < a. We will have solved the problem if we can show there are q ⋆
and r0⋆ such that ϕ(r) given by (2.35) satisfies ϕ(r) |Σ = 0, where Σ = {|r| = a}. We obtain two
equations by imposing ϕ(r) = 0 on (2.35), where r = ±a r̂0 , namely
q q⋆ q q⋆
0 = + , 0 = + , (2.36)
r0 − a a − r0⋆ a + r0 a + r0⋆
respectively, where note that 0 < r0⋆ < a < r0 . These are easily solved to give
a a a2
q⋆ = − q, r0⋆ = a = . (2.37)
r0 r0 r0

23
 z

r0 charge q

a charge q★
y
O

sphere Σ
x

Figure 9: Conducting sphere Σ of radius a, centred on the origin O, with charge q at position r0
outside the sphere, and image charge q ⋆ lying on the same line through O, but inside the sphere.

Notice that r0 and its image point r⋆0 are then geometrically inverse points for the sphere Σ. That
is, they lie on the same straight line through the origin of the sphere, with r0 ·r⋆0 = a2 . Substituting
(2.37) back into (2.35), and writing r = r r̂ with r̂ a unit vector, we have
 
q  1 a 1
ϕ(r) = p − q  , (2.38)
4πϵ0 r2 + r02 − 2rr0 r̂ · r̂0 r0 r2 + ( a2 )2 − 2 a2 r r̂ · r̂ 0
r0 r0

It is then straightforward to see that putting r = a on the right hand side of (2.38) indeed gives
zero, for all directions r̂. Thus (2.38) is the required solution for ϕ, for r ∈ R ≡ {|r| ≥ a}.
Since r̂ points away from the origin, it is the inward unit normal to the region R = {|r| ≥ a},
and thus (2.3) reads
∂ϕ q r02 − a2
σ = −ϵ0 r̂ · ∇ϕ |Σ = −ϵ0 = − , (2.39)
∂r r=a 4π a a2 + r2 − 2ar0 r̂ · r̂0
3/2
0

the last expression being obtained after a short computation using (2.38). For fixed r0 , without
loss of generality we may take this to point along the z-axis, so that r̂ · r̂0 = cos θ, with θ the usual
spherical polar coordinate giving the angle between r̂ and r̂0 . The total charge induced on the
spherical conductor is then
Z π Z " #π
2π
r02 − a2
Z
q
Q ≡ σ dS = σ(θ) a2 sin θ dθ dφ = 2π · p
Σ θ=0 φ=0 4π r0 a2 + r02 − 2ar0 cos θ θ=0
a
= − q, (2.40)
r0
which again is the same as the image charge q ⋆ = − ra0 q.

Finally, we can write down the Dirichlet Green’s function GD (r, r′ ) in the region R = {|r| ≥ a}.
The discussion is analogous to that for the previous example, with this being 4πϵ0 /q times the

24
Figure 10: Electric field lines for E = −∇ϕ, with ϕ(r) given by (2.38). The original charge q is
shown in black, and is taken to be positive, while the negative image charge q ⋆ is red. These are
located at inverse points for the sphere Σ, shown in blue. Note E is perpendicular to Σ.

solution (2.27) for ϕ, where r′ = r0 is the location of the point charge. Writing r′ = r′ r̂′ , we can
identify r0 = r′ and recall that r0⋆ = a2 /r0 = a2 /r′ , so that r′⋆ = r0⋆ r̂′ = (a2 /r′2 ) r′ . From (2.35) we
then write down6
1 a 1
GD (r, r′ ) = ′
− ′ . (2.41)
|r − r | r |r − (a/r′ )2 r′ |

Consider the Dirichlet boundary problem where we fix ϕ(r) |Σ = V (θ, φ) to be a given function on
the sphere Σ of radius a, and for simplicity suppose that there is no charge distribution, so that
ρ ≡ 0. Only the second term in (2.15) now contributes, and recalling that our unit vectors r̂, r̂′
point into R, we compute

∂GD (r, r′ ) ∂GD (r, r′ ) r2 − a2

= − = − . (2.42)
∂n′ r′ ∈Σ ∂r′ r′ =a a (r2 + a2 − 2ar r̂ · r̂′ )3/2
Here the last equality follows from the same calculation as (2.39). The solution for the potential
(2.15) is then
π 2π
a(r2 − a2 )
Z Z
1
ϕ(r) = V (θ′ , φ′ ) sin θ′ dθ′ dφ′ . (2.43)
4π θ′ =0 φ′ =0 (r2 + a2 − 2ar r̂ · r̂′ )3/2
Here to perform the integral one should also write r̂′ = (sin θ′ cos φ′ , sin θ′ sin φ′ , cos θ′ ) in spherical
polars. Notice that the normal derivative of the Green’s function (2.42) is acting as an integral
kernel in (2.43), effectively “propagating” the prescribed electrostatic potential ϕ(r) |Σ = V (θ, φ)
into the region outside this sphere. Physically we are holding the surface of the sphere Σ at fixed
6
Notice this is symmetric GD (r, r′ ) = GD (r′ , r), as must be the case by Proposition 2.5: to see this for the second
term write |r′ r − (a2 /r′ ) r′ |2 = r′2 r2 + a4 − 2a2 r · r′ .

25
voltage V (θ, φ), and (2.43) is the resulting electrostatic potential outside the sphere. In this case
Σ couldn’t be a conductor for non-constant V , but it could be an insulator. ■

2.4 Orthonormal functions

The electrostatic Maxwell equations (1.16), (1.33), or equivalently Poisson’s equation (2.1), are
linear – a manifestation of the Principle of Superposition. Solutions may then be expanded in
terms of a convenient basis of solutions. This is a powerful technique, where the particular basis
of functions is chosen according to the symmetries of the problem.

2.4.1 General theory

Definition A set of complex-valued functions un : [a, b] → C, defined on the interval [a, b] ⊂ R

and labelled by a countable index n ∈ I (usually I = N or I = Z), is said to be orthonormal if
Z b
um (x) un (x) dx = δmn , ∀m, n ∈ I . (2.44)
a

Suppose that a function f : [a, b] → C may be expanded as a uniformly convergent series

X
f (x) = cn un (x) , (2.45)
n∈I

for coefficients cn ∈ C. Then the coefficients may be determined via

Z b X Z b
um (x) f (x) dx = cn um (x) un (x) dx = cm , (2.46)
a n∈I a

where we have used (2.45) and uniform convergence in the first equality, and (2.44) in the second
equality. Substituting (2.46) back into (2.45) then gives
!
X XZ b Z b X
f (x) = cn un (x) = un (x′ ) f (x′ ) un (x) dx′ = f (x′ ) un (x′ ) un (x) dx′ .(2.47)
n∈I n∈I a a n∈I

If this is to hold for any function, then from the definition of the Dirac delta function we identify
X
un (x′ ) un (x) = δ(x′ − x) . (2.48)
n∈I

Definition The set of orthonormal functions {un (x)}n∈I is said to be complete if (2.48) holds.
Rb
Notice then that f (x) = a f (x′ ) δ(x′ − x) dx′ , and substituting (2.48) and reading equation (2.47)
from right to left says that f (x) may be expanded as in (2.45).

Example (Fourier sine series) Consider a function f : [0, a] → R with f (0) = f (a) = 0. Then we
may expand f in a Fourier sine series 7 , with complete set of orthonormal functions
r
2  πn 
un (x) ≡ sin x , n∈N. (2.49)
a a
7
Particularly familiar to those who took Part A Quantum Theory.

26
That is, f (x) may be expanded as in (2.45) with I = N, so
r ∞
2 X  πn 
f (x) = cn sin x , (2.50)
a a
n=1

where the coefficients are given by (2.46)

Z a r Z a
2  πn 
cn = un (x) f (x) dx = sin x f (x) dx . (2.51)
0 a 0 a
The series ∞
P
n=1 cn un (x) converges absolutely and uniformly to f (x), provided the latter is con-
tinuously differentiable. More generally, for any function that is square-integrable, meaning
Ra 2
0 |f (x)| dx < ∞, the Fourier series converges almost everywhere to f . This general subject
is put on a rigorous footing in functional analysis. We also have
N N
X 2X  πn   πn 
N →∞
δN (x, x′ ) ≡ un (x′ ) un (x) = sin x′ sin x −→ δ(x′ − x) , (2.52)
a a a
n=1 n=1

(understood in terms of distributions). The function δN (x, x′ ) is plotted in Figure 11. ■

100

80

60

40

20

0.2 0.4 0.6 0.8 1.0

-20

Figure 11: The function δN (x, x′ ) in (2.52), with x′ = 1

2 the midpoint of [0, 1], and N = 100.

Example (Complex exponential Fourier series) More generally, from Prelims you know that a
function on [− L2 , L2 ] may be expanded in terms of Fourier modes involving both cos 2πn


L x and
sin 2πn L L


L x . The Fourier sine series arises in the special case that f : [− 2 , 2 ] → R is an odd
function, and in the previous example L = 2a and we then restrict this function to x ≥ 0. It is
sometimes convenient to rewrite the sine and cosine functions in terms of complex exponentials,
leading to a basis of functions
1 2πn
un (x) ≡ √ ei( L )x , n∈Z. (2.53)
L
Notice that now n ∈ I = Z, and it is straightforward to check the orthonormal property (2.44).
The usual Fourier expansion may be written as
∞ ∞
1 X 2πn
cn ei( L )x =
X
f (x) = √ cn un (x) , (2.54)
L n=−∞ n=−∞

27
as in (2.45), where the coefficients are
Z L/2 Z L/2
1 −i( 2πn
L )
x
cn = √ e f (x) dx = un (x) f (x) dx . ■ (2.55)
L −L/2 −L/2

The last example has a natural “continuum limit”, where formally we take the size of the interval
L → ∞. In this case the discrete set of orthonormal functions {un (x)}n∈Z becomes a continuous
set of functions {uk (x)}k∈R , and correspondingly sums such as (2.45) are replaced by integrals,
and the Kronecker delta symbol in the orthonormality relation (2.44) is replaced by a Dirac delta
function. The main example is:

Example (Exponential Fourier expansion/Fourier transform) Analogously to (2.53), consider

1
uk (x) ≡ √ eikx , (2.56)
2π
where now the index k ∈ R is a continuous variable, rather than discrete. A general integrable
function on R may be expanded as
Z ∞
1
f (x) = √ C(k) eikx dk , (2.57)
2π −∞

which is usually called the exponential Fourier expansion. The “coefficients” C(k), which are now
complex-valued functions on R, are given by (cf. (2.46))
Z ∞ Z ∞
1
C(k) = √ e−ikx f (x) dx = uk (x) f (x) dx , (2.58)
2π −∞ −∞

which is also called the Fourier transform of f (x). The orthonormality condition (2.44) now reads
Z ∞ Z ∞
1 ′
uk′ (x) uk (x) dx = ei(k−k )x dx = δ(k − k ′ ) , (2.59)
−∞ 2π −∞

while the completeness relation (2.48) takes the similar-looking form

Z ∞ Z ∞
1 ′
uk (x′ ) uk (x) dk = eik(x−x ) dk = δ(x − x′ ) . (2.60)
−∞ 2π −∞

These give useful representations of the Dirac delta function. ■

Remark Note that if two expansions are equal to each other, we may identify their coefficients.
R∞ R∞
For example, if −∞ C(k) uk (x) dk = −∞ D(k) uk (x) dk then using (2.59)
Z ∞ Z ∞  Z ∞ Z ∞ 
C(k) uk (x) dk uk′ (x) dx = D(k) uk (x) dk uk′ (x) dx
x=−∞ k=−∞ x=−∞ k=−∞
Z ∞ Z ∞
′
=⇒ C(k) δ(k − k ) dk = D(k) δ(k − k ′ ) dk =⇒ C(k ′ ) = D(k ′ ) . (2.61)
−∞ −∞

28
2.4.2 Cartesian coordinates

Consider the Laplace equation in Cartesian coordinates

∂2ϕ ∂2ϕ ∂2ϕ

∇2 ϕ = + 2 + 2 = 0. (2.62)
∂x2 ∂y ∂z
This is of course Poisson’s equation (2.1) with zero charge density ρ ≡ 0. We may seek separable
solutions by substituting ϕ(x, y, z) = X(x) Y (y) Z(z) into (2.62), and dividing through by ϕ, giving

1 d2 X 1 d2 Y 1 d2 Z
+ + = 0. (2.63)
X dx2 Y dy 2 Z dz 2
The usual separation of variable argument implies that each term is separately constant, so that

1 d2 X 1 d2 Y 1 d2 Z
= −α2 , = −β 2 , = α2 + β 2 . (2.64)
X dx2 Y dy 2 Z dx2

Here a priori the constants α2 , β 2 are any real numbers, so that α, β can be real or purely imaginary,
although in the example below we will take α, β ∈ R. The two solutions to the first ODE in (2.64)
are X(x) = A± e±iαx , with A± integration constants, and linearity of the original Laplace equation
(2.62) implies that any linear combination of
√
α2 +β 2 z
ϕ(x, y, z) = e±iαx e±iβy e± (2.65)

is a solution. Imposing boundary conditions on ϕ will then restrict the possible values of α, β, and
also the coefficients in the series.

Example (Electrostatic potential for a rectangular box) Consider a rectangular box R = {(x, y, z) |
x ∈ [0, a] , y ∈ [0, b] , z ∈ [0, c]}, and consider the Dirichlet problem in which ϕ = 0 on all boundary
surfaces except the face {z = c}, where we impose ϕ(x, y, c) = V (x, y), with charge density ρ ≡ 0.8
The resulting Laplace equation in R has separable solutions which are linear combinations of
(2.65). Let us look at a solution for fixed α, β, and impose the boundary conditions

ϕ |x=0 = 0 , ϕ |y=0 = 0 , ϕ |z=0 = 0 . (2.66)

Writing e±iαx = cos αx ± i sin αx, (2.66) sets the coefficient of the cosine term to zero, so that the
solutions satisfying (2.66) take the form
p
sin αx sin βy sinh( α2 + β 2 z) . (2.67)

On the other hand, the boundary conditions ϕ |x=a = 0, ϕ |y=b = 0 then imply that sin αa = 0 =
sin βb, which implies
mπ nπ
α = , β = , m, n ∈ N . (2.68)
a b
8
cf. the discussion of the Dirichlet Green’s function outside a sphere, after equation (2.41).

29
 z

ϕ (x,y,c) = V(x,y)

c a

x b

Figure 12: Rectangular box R = {(x, y, z) | x ∈ [0, a] , y ∈ [0, b] , z ∈ [0, c]}, with boundary
condition that ϕ is zero on all boundary faces except {z = c}, where ϕ(x, y, c) = V (x, y).

Notice here that we have taken m, n > 0 without loss of generality, since sine is an odd function.
We may then write
∞
r
m2 n2
X  mπ   nπ   
ϕ(x, y, z) = cm,n sin x sin y sinh π + 2 z , (2.69)
a b a2 b
m,n=1

with coefficients cm,n . From section 2.4.1, we recognize this as a Fourier sine series in the x and y
variables.
Finally, we just need to impose the boundary condition ϕ |z=c (x, y) = V (x, y). Setting z = c in
(2.69) gives the Fourier sine series expansion
∞
"  r 2 #
X m n2  mπ   nπ 
ϕ(x, y, c) = V (x, y) = cm,n sinh π + c sin x sin y . (2.70)
a2 b2 a b
m,n=1

We can then read off the coefficients, in square brackets, from (2.50), (2.51), which gives
Z a Z b
4  mπ   nπ 
cm,n =  q sin x sin y V (x, y) dx dy . ■ (2.71)
a b

2 n2
ab sinh π m a2 + b2 c x=0 y=0

We know from the general theory in section 2.2 that the above problem must also have a solution
in terms of the Dirichlet Green’s function GD (r, r′ ). Recall this satisfies (2.7) for r, r′ ∈ R inside
the rectangular box, with Dirichlet boundary (2.14) that GD (r, r′ ) |r′ ∈Σ = 0, where Σ = ∂R is the
boundary of the box. The solution for the electrostatic potential is then (2.15). Since here ρ ≡ 0
and ϕ |Σ is zero except on the face ϕ |z=c = V , this solution reads

∂GD (r, r′ )
Z
1
ϕ(r) = − V (x′ , y ′ ) dS ′ . (2.72)
4π z ′ =c ∂n′
The last example suggests making an orthonormal series expansion for GD . This Green’s function
is then effectively constructed from the solution to the following physical problem:

30
Example (Point charge inside a rectangular box) Consider a point charge q inside a rectangular
box that consists of a grounded conductor. Thus ϕ |Σ = 0, so that ϕ is zero on all six faces, and
q
∇2 ϕ(r) = − δ(r − r0 ) , (2.73)
ϵ0
for r = (x, y, z) inside the box R, where r0 = (x0 , y0 , z0 ) ∈ R is the location of the point charge.
Following the last example, we can solve this via separation of variables. Due to the boundary
conditions we can expand each variable as a Fourier sine series, leading to
r ∞  
8 X  mπ   nπ  ℓπ
ϕ(r) = cm,n,ℓ sin x sin y sin z , (2.74)
abc a b c
m,n,ℓ=1

with coefficients cm,n,ℓ . Substituting (2.74) into (2.73) then gives for the left hand side
r ∞
m2 π 2 n2 π 2 ℓ2 π 2
    
2 8 X  mπ   nπ  ℓπ
∇ ϕ = cm,n,ℓ − 2 − 2 − 2 sin x sin y sin z .(2.75)
abc a b c a b c
m,n,ℓ=1

On the other hand, from (2.52) we can write the Dirac delta function in one dimension as δ(x−x0 ) =
2 P∞ mπ mπ



a m=1 sin a x 0 sin a x , and then (1.19) gives the right hand side of (2.73) as
∞
"
q q 8 X  mπ   mπ   nπ   nπ 
− δ(r − r0 ) = − sin x0 sin x sin y0 sin y
ϵ0 ϵ0 abc a a b b
m,n,ℓ=1
   #
ℓπ ℓπ
× sin z0 sin z . (2.76)
c c

Equating (2.76) with (2.75), the coefficients of sin mπ nπ ℓπ

a x sin b y sin c z must be equal, giving
8 sin mπ x0 sin nπ ℓπ
r



q
a b y0 sin  c z0
cm,n,ℓ = . (2.77)
ϵ0 abc m2 π 2
+ n2 π 2
+ ℓ2 π 2
a2 b 2 c 2

Substituting this into (2.74) then gives the solution, expressed as a Fourier sine series. As for the
examples in section 2.3, the Dirichlet Green’s function GD (r, r′ ) for this problem is then simply
(4πϵ0 /q) ϕ(r), with ϕ(r) given by (2.74), (2.77), and where we identify r0 = r′ . ■

In the above examples we have expanded the functions using the orthonormal basis of Fourier
sine modes (2.49). This was a convenient basis to use due to the rectangular boundary conditions,
which took a simple form after separating variables in Cartesian coordinates. However, in other
problems it is convenient to use a different orthonormal basis.

Example (Fourier transform of the Dirichlet Greens function on R3 ) Consider the Dirichlet
Green’s function on R3 given by GD (r, r′ ) = 1/|r − r′ |. This satisfies (2.7), where from (2.60)
and setting r = (x1 , x2 , x3 ), r′ = (x′1 , x′2 , x′3 ) we may write

δ(r − r′ ) ≡ δ(x1 − x′1 ) δ(x2 − x′2 ) δ(x3 − x′3 )

Z ∞ Z ∞ Z ∞
1 ′ ′ ′
= 3
dk1 dk2 dk3 eik1 (x1 −x1 ) eik2 (x2 −x2 ) eik3 (x3 −x3 )
(2π) −∞ −∞ −∞
Z
1 ik·(r−r′ ) 3
= e d k, (2.78)
(2π)3 k∈R3

31
where k = (k1 , k2 , k3 ), and d3 k ≡ dk1 dk2 dk3 . We may then similarly expand the Green’s function
as in (2.57), which using the compact notation on the last line of (2.78) reads
Z
1 1 ′
′
= 3/2
C(k) eik·(r−r ) d3 k , (2.79)
|r − r | (2π) k∈R3

with coefficient function C(k), which is a function on R3 . Noting that, analogously to the compu-
tation in (2.75), we have
∂2 ∂2 ∂2
   
ik·(r−r′ ) ′ ′ ′
∇ ′2
e = ′2 + ′2 + ′2 ei[k1 (x1 −x1 )+k2 (x2 −x2 )+k3 (x3 −x3 )]
∂x1 ∂x2 ∂x3
2 2 2

ik·(r−r′ ) ′
= −k1 − k2 − k3 e = −|k|2 eik·(r−r ) , (2.80)

applying ∇′2 to (2.79) gives

  Z
′2 1 1 ′
∇ ′
= − 3/2
C(k) |k|2 eik·(r−r ) d3 k . (2.81)
|r − r | (2π) k∈R3

Since ∇′2 (1/|r − r′ |) = −4π δ(r − r′ ), equating (2.81) with −4π times (2.78) leads to the identifi-
cation of coefficient functions (cf. the remark at the end of section 2.4.1)
4π 1
C(k) = . (2.82)
(2π)3/2 |k|2
Thus the Dirichlet Green’s function on R3 is
Z
1 1 1 ik·(r−r′ ) 3
GD (r, r′ ) = ′
= e d k. (2.83)
|r − r | 2π 2 k∈R3 |k|2
We shall return to this formula in section 5.6. ■

2.4.3 Spherical polar coordinates

For problems with spherical symmetry, it is often more convenient to use a complete set of or-
thonormal functions adapted to that symmetry.
The Laplacian in spherical polar coordinates (r, θ, φ) is
1 ∂ 2 (rϕ) 1 ∂2ϕ 1 ∂ 2 (rϕ)
   
2 1 1 ∂ ∂ϕ 1
∇ ϕ = + sin θ + 2 ≡ + 2 ∇2θ,φ ϕ , (2.84)
r ∂r2 r2 sin θ ∂θ ∂θ sin θ ∂φ2 r ∂r 2 r
where the second equality defines the angular Laplacian ∇2θ,φ ϕ as the term in square brackets.
Recall here that r ≥ 0, θ ∈ [0, π], φ ∈ [0, 2π), with the latter being periodically identified.
Following the start of section 2.4.2, we may again use separation of variables, but this time we
write ϕ(r, θ, φ) = R(r) Y (θ, φ). Substituting this into (2.84) and dividing through by ϕ/r2 =
R(r) Y (θ, φ)/r2 , the Laplace equation reads
r d2 (rR) 1 2
+ ∇ Y = 0. (2.85)
R dr 2 Y θ,φ
Both terms must be constant, so that
d2 R(r)
2
(rR(r)) = λ , ∇2θ,φ Y (θ, φ) = −λ Y (θ, φ) , where λ = constant . (2.86)
dr r

32
 Let us focus first on the second equation in (2.86), for the angular coordinates. We may
also separate variables here, writing Y (θ, φ) = P (θ) Φ(φ). Dividing through by Y / sin2 θ =
P (θ) Φ(φ)/ sin2 θ, this equation then reads
1 d2 Φ
   
sin θ d dP 2
sin θ + λ sin θ + = 0. (2.87)
P dθ dθ Φ dφ2
Again, both terms must be constant, with the second leading to the equation Φ′′ (φ) = −c Φ(φ),
with c constant. Since φ is a periodic coordinate, in order to be single-valued we must have
Φ(φ) = Φ(φ + 2π) for all φ, and this fixes c = m2 with m ∈ Z an integer, with solutions
Φ(φ) = e±imφ . Substituting this back into (2.87) leads to a second order ODE for P (θ), depending
on the constants m2 and λ ∈ R, called the general or associated Legendre equation. The solutions to
this equation are typically singular in the limits θ → 0, π. To find smooth solutions Y (θ, φ) = Y (r̂)
on the unit sphere S2 ≡ {r = 1} =⊂ R3 , we need a condition on λ:9

Theorem 2.6 In the space of smooth functions Y : S2 −→ C, the eigenvalue equation

∇2θ,φ Y (θ, φ) = −λ Y (θ, φ) (2.88)

has non-trivial solutions only for λ of the form λ = ℓ(ℓ+1) with some non-negative integer ℓ ∈ Z≥0 .
For such ℓ, the eigenspace has dimension 2ℓ + 1 and is spanned by separable solutions of the form
Y (θ, φ) = P (θ)eimφ with m ∈ {−ℓ, −ℓ + 1, . . . , ℓ − 1, ℓ}.

We can write down such a basis of Eigenfunctions explicitly:

Definition The spherical harmonics Yℓ,m : S2 −→ C are defined for integers ℓ ≥ m ≥ 0 by

s
2ℓ + 1 (ℓ − m)!
Yℓ,m (θ, φ) = · (−1)m · Pℓ,m (cos θ) · eimφ (2.89)
4π (ℓ + m)!

and Yℓ,−m = (−1)m Yℓ,m . Here, Pℓ,m are the associated Legendre functions, defined by
1 dℓ 2
Pℓ,0 (x) =(x − 1)ℓ and
2ℓ · ℓ! dxℓ (2.90)
dm
Pℓ,m (x) = (1 − x2 )m/2 m Pℓ,0 (x).
dx
The polynomials Pℓ (x) = Pℓ,0 (x) are known as Legendre polynomials.

Theorem 2.7 The spherical harmonics {Yℓ,m }ℓ≥|m|≥0 are a complete set of orthonormal Eigen-
functions on the unit sphere S2 = {r = 1} ⊂ R3 , for the operator ∇2θ,φ . They satisfy:

(i) Yℓ,−m (θ, φ) = (−1)m Yℓ,m (θ, φ) ,

Z π Z 2π
(ii) Yℓ,m (θ, φ) Yℓ′ ,m′ (θ, φ) sin θ dθ dφ = δℓ,ℓ′ δm,m′ (orthonormal) ,
θ=0 φ=0
∞ X
ℓ (2.91)
X 1
(iii) Yℓ,m (θ′ , φ′ ) Y ℓ,m (θ, φ) = δ(φ′ − φ) δ(θ′ − θ) (completeness) ,
sin θ
ℓ=0 m=−ℓ
(iv) ∇2θ,φ Yℓ,m = −ℓ(ℓ + 1)Yℓ,m .
9
We state this here without proof. This material will be familar to those who took Part A Quantum Theory.

33
In particular, any square-integrable function f (θ, φ) on the sphere S2 can be uniquely written as
∞ X
X ℓ
f (θ, φ) = cℓ,m Yℓ,m (θ, φ) , (2.92)
ℓ=0 m=−ℓ

where the coefficients cℓ,m ∈ C are constants.

Compare the orthonormal and completeness conditions in (2.91) to the one-dimensional conditions
in equations (2.44) and (2.48), respectively. The first few spherical harmonics are
r
1
Y0,0 = ,
4π
r r
3 3
Y1,0 = cos θ , Y1,1 = − sin θ eiφ , (2.93)
4π 8π
r r r
1 5 2 15 iφ 1 15
sin2 θ e2iφ .


Y2,0 = 3 cos θ − 1 , Y2,1 = − sin θ cos θ e , Y2,2 =
2 4π 8π 4 2π
Notice we have only listed the functions with m ≥ 0, with those for m < 0 determined by
property (i) in (2.91).
We now return to the first, radial equation in (2.86), which reads
d2 R(r)
2
(rR(r)) = ℓ(ℓ + 1) . (2.94)
dr r
We may try to solve this by setting R = rα . Substituting this into (2.94) gives

α(α + 1) rα−1 = ℓ(ℓ + 1) rα−1 =⇒ α(α + 1) = ℓ(ℓ + 1) , (2.95)

which has solutions α = ℓ, α = −(ℓ + 1). The general solution is hence R(r) = A rℓ + B r−(ℓ+1) ,
with A and B integration constants.
Putting everything together, a general solution to the Laplace equation ∇2 ϕ = 0 may be ex-
panded in spherical harmonics as
∞ X
X ℓ  
ϕ(r) = Aℓ,m rℓ + Bℓ,m r−(ℓ+1) Yℓ,m (θ, φ) , (2.96)
ℓ=0 m=−ℓ

with constants Aℓ,m , Bℓ,m .

Example (Electrostatic potential inside a sphere, with Dirichlet boundary condition) Consider
the Dirichlet boundary problem inside a sphere of radius a, with ϕ(a, θ, φ) = V (θ, φ) prescribed
on the boundary, and zero charge density ρ ≡ 0.
We thus want to solve the Laplace equation for r < a, with the above boundary condition.
General solutions to the Laplace equation may be expanded as in (2.96), but notice that we should
set the constants Bℓ,m = 0, as the negative powers r−(ℓ+1) all diverge at the centre of the sphere
r = 0. Imposing the boundary condition at r = a then sets
∞ X
X ℓ
ϕ(a, θ, φ) = V (θ, φ) = aℓ Aℓ,m Yℓ,m (θ, φ) . (2.97)
ℓ=0 m=−ℓ

34
Comparing to (2.92), this is simply the spherical harmonic expansion of the function V (θ, φ). We
may compute the coefficients via
Z π Z 2π
Yℓ,m (θ, φ) V (θ, φ) sin θ dθ dφ
θ=0 φ=0
∞ ℓ Z π Z 2π
ℓ′
X X
= a Aℓ′ ,m′ Yℓ,m (θ, φ) Yℓ′ ,m′ (θ, φ) sin θ dθ dφ
ℓ′ =0 m′ =−ℓ′ θ=0 φ=0

= aℓ Aℓ,m , (2.98)

where in the second line we have substituted the expansion (2.97), and the last line uses or-
thonormality in (2.91). Notice here that the factor of 1/ sin θ on the right hand side of (iii) in
(2.91) cancels the sin θ factor in the area element sin θ dθ dφ for the sphere (see also the following
example). Substituting for Aℓ,m given by (2.98) into (2.96) then gives the solution. ■

Finally, comparing to the last example in section 2.3, we examine the following:

Example (Dirichlet Green’s function outside a sphere, in spherical harmonics) Let R = {r ≥ a}

be the region outside a sphere Σ = {r = a} of radius a, centred on the origin, and consider the
Dirichlet Green’s function GD (r, r′ ) in R, which recall satisfies

∇′2 GD (r, r′ ) = −4π δ(r − r′ ) , and GD (r, r′ ) = GD (r′ , r) , ∀ r, r′ ∈ R ,

GD (r, r′ ) = 0 , ∀ r ∈ R, r′ ∈ Σ . (2.99)

Focusing first on the dependence on r, which we write in spherical polars (r, θ, φ), we may expand
in spherical harmonics as in (2.92) by writing
∞ X
X ℓ
GD (r, r′ ) = cℓ,m (r, r′ ) Yℓ,m (θ, φ) . (2.100)
ℓ=0 m=−ℓ

Note the coefficients cℓ,m (r, r′ ) depend on the scalar distance r to the origin, and the vector
position r′ . Acting on (2.100) with the Laplacian in spherical coordinates (2.84) gives
∞ X
ℓ
1 d (r cℓ,m (r, r′ ))
 2 
X 1
∇2 GD (r, r′ ) = − ℓ(ℓ + 1) cℓ,m (r, r ′
) Yℓ,m (θ, φ) , (2.101)
r dr2 r2
ℓ=0 m=−ℓ

where we have used ∇2θ,φ Yℓ,m (θ, φ) = −ℓ(ℓ + 1) Yℓ,m (θ, φ). The expression (2.101) should equal
−4π δ(r − r′ ), where notice that it doesn’t matter whether we act with ∇2 or ∇′2 , due to the
symmetry GD (r, r′ ) = GD (r′ , r).

Lemma 2.8 The Dirac delta function in spherical polar coordinates is

1
δ(r − r′ ) = δ(r − r′ ) δ(θ − θ′ ) δ(φ − φ′ ) . (2.102)
r2 sin θ

35
Proof Notice the denominator on the right hand side is precisely the Jacobian in going from
Cartesian coordinates to spherical polar coordinates, and indeed this is also the origin of the
1/ sin θ term on the right hand side of (iii) in equation (2.91). Starting with the defining property
of the Dirac delta function we have
Z Z ∞ Z π Z 2π
f (r′ ) = f (r) δ(r − r′ ) dx1 dx2 dx3 = f (r) δ(r − r′ ) r2 sin θ dr dθ dφ (2.103)
 
R3 r=0 θ=0 φ=0

In order for the right hand side to give f (r′ ), thus picking out r = r′ , θ = θ′ , φ = φ′ in the integral,
we precisely have to make the identification (2.102). ■

The completeness relation (iii) in (2.91) for the spherical harmonics in turn allows us to write
∞ X
ℓ
1 X
δ(r − r′ ) = δ(r − r ′
) Yℓ,m (θ′ , φ′ ) Yℓ,m (θ, φ) . (2.104)
r2
ℓ=0 m=−ℓ

Equating (2.101) with −4π times (2.104) then allows us to equate the coefficients

1 d2 (r cℓ,m (r, r′ )) 1 4π
2
− 2 ℓ(ℓ + 1) cℓ,m (r, r′ ) = − 2 δ(r − r′ ) Yℓ,m (θ′ , φ′ ) . (2.105)
r dr r r
In particular, focusing on the r′ dependence we may read off

cℓ,m (r, r′ ) = Aℓ,m (r, r′ ) Yℓ,m (θ′ , φ′ ) . (2.106)

The Green’s function (2.100) now reads

∞ X
X ℓ
GD (r, r′ ) = Aℓ,m (r, r′ ) Yℓ,m (θ′ , φ′ ) Yℓ,m (θ, φ) , (2.107)
ℓ=0 m=−ℓ

where from (2.105) the coefficient functions Aℓ,m (r, r′ ) satisfy

1 d2 (r Aℓ,m (r, r′ )) 1 4π
2
− 2 ℓ(ℓ + 1) Aℓ,m (r, r′ ) = − 2 δ(r − r′ ) . (2.108)
r dr r r

It remains to solve this second order ODE. When r ̸= r′ , so either r < r′ or r > r′ , then (2.108)
reduces to
d2 (r Aℓ,m (r, r′ )) 1
2
= ℓ(ℓ + 1) Aℓ,m (r, r′ ) . (2.109)
dr r
We already solved this equation in (2.94), where the two independent solutions are proportional
to rℓ and r−(ℓ+1) . We may thus write down
(
′ Aℓ (r′ ) rℓ + Bℓ (r′ ) r−(ℓ+1) , a ≤ r < r′ ,
Aℓ,m (r, r ) = (2.110)
Cℓ (r′ ) rℓ + Dℓ (r′ ) r−(ℓ+1) , r > r′ ≥ a ,

anticipating that the remaining coefficients will depend on ℓ, but not on m. In order for the Green’s
function to be bounded as r → ∞, notice we must set Cℓ (r′ ) ≡ 0. Next we impose the Dirichlet

36
boundary condition at r = a (or equivalently r′ = a due to the symmetry property on the first line
of (2.99)). This sets

0 = Aℓ,m (a, r′ ) = Aℓ (r′ ) aℓ + Bℓ (r′ ) a−(ℓ+1) =⇒ Bℓ (r′ ) = −a2ℓ+1 Aℓ (r′ ) . (2.111)

Then (2.110) reads

 
a2ℓ+1
(
Aℓ (r′ ) rℓ − , a ≤ r < r′ ,
Aℓ,m (r, r′ ) = rℓ+1 (2.112)
Dℓ (r′ ) rℓ+1
1
, r > r′ ≥ a ,

Next we impose the symmetry property Aℓ,m (r, r′ ) = Aℓ,m (r′ , r).10 From (2.112), this relates

a2ℓ+1 Dℓ (r′ )
 
1
Dℓ (r′ ) = Aℓ (r) r′ℓ − ′ℓ+1 =⇒  = Aℓ (r) rℓ+1 = Kℓ , (2.113)
rℓ+1

r r′ℓ − a2ℓ+1
r′ℓ+1

where in the last step notice both sides must be a constant Kℓ . Substituting into (2.112) gives

Kℓ rℓ
   a 2ℓ+1 
 ′ℓ+1 1 − , a ≤ r < r′ ,


′ r r
Aℓ,m (r, r ) = (2.114)
Kℓ r′ℓ
  a 2ℓ+1 
 ℓ+1 1 − ′

 , ′
r>r ≥a,
r r

which is indeed symmetric under r ↔ r′ , and continuous at r = r′ .

It remains only to determine the constants Kℓ , where notice that we haven’t yet used the
normalization of the Dirac delta function on the right hand side of (2.108). Multiplying this
equation by r and integrating from r = r′ − ε to r = r′ + ε implies
 r′ +ε
d 4π
r Aℓ,m (r, r′ )


lim = − . (2.115)
ε→0 dr r′ −ε r′

Notice that the second term on the left hand side of (2.108) doesn’t contribute in this limit, as
Aℓ,m (r, r′ ) is continuous at r = r′ . Note also that in the upper limit r = r′ + ε > r′ , while in the
lower limit r = r′ − ε < r′ . Using (2.114) thus gives
( )
d r′ℓ d rℓ+1
   a 2ℓ+1     a 2ℓ+1 
4π
− ′ = Kℓ lim 1− ′ − 1− ,
r ε→0 dr rℓ r r=r′ +ε dr r′ℓ+1 r r=r′ −ε
( 2ℓ+1 )
r′ℓ (r′ − ε)ℓ (r′ − ε)ℓ
  a 2ℓ+1  
a
= Kℓ lim − ℓ ′ 1− ′ − (ℓ + 1) −ℓ
ε→0 (r + ε)ℓ+1 r r′ℓ+1 r′ℓ+1 r′ − ε
2ℓ + 1
= −Kℓ , (2.116)
r′
so that
4π
Kℓ = . (2.117)
2ℓ + 1
10
This symmetry property was derived in Proposition 2.5, while here it is convenient to impose this to simplify
finding the Green’s function.

37
 We can now write down the final form of the Dirichlet Green’s function outside the sphere.
Combining (2.107), (2.114) and (2.117) gives
∞ X
X ℓ
GD (r, r′ ) = Aℓ,m (r, r′ ) Yℓ,m (θ′ , φ′ ) Yℓ,m (θ, φ) , (2.118)
ℓ=0 m=−ℓ

Here we may write the compact expression

"  2ℓ+1 #
r ℓ
4π a
Aℓ,m (r, r′ ) = <
ℓ+1
1− , (2.119)
2ℓ + 1 r> r<

where we have defined

(
r> ≡ r′ , r< ≡ r , r < r′ ,
(2.120)
r> ≡ r , r< ≡ r ′ , r > r′ .

That is, r< is the smaller of r or r′ , while r> is the larger of r or r′ . ■

2.5 * Complex analytic methods

One can also solve certain boundary value problems in electrostatics using complex analysis. For
example, for problems that have translational symmetry in the z-axis direction, so that ϕ = ϕ(x, y)
depends only on the x and y coordinates, the Laplace equation in three dimensions (2.62) effectively
reduces to the Laplace equation in two dimensions:

∂2ϕ ∂2ϕ
0 = ∇2 ϕ = + 2 . (2.121)
∂x2 ∂y
Introduce the complex coordinate

z ≡ x + iy , (2.122)

thus identifying R2 ∼
= C with the complex plane. Recall that a function f is said to be holomorphic
in a domain U ⊆ C if it is complex differentiable at every point of U . We write the real and
imaginary parts of f as u(x, y) ≡ Re f (z), v(x, y) ≡ Im f (z), viewed as functions on R2 ∼
= C. Then
a result in complex analysis shows that both u and v are harmonic functions, i.e. ϕ = u and ϕ = v
both satisfy (2.121).

Example Taking f (z) = z2 , we have u + iv = (x + iy)2 = x2 − y 2 + i(2xy), and hence both

ϕ = u = x2 − y 2 and ϕ = v = 2xy are harmonic.

This example is of course particularly simple, but it immediately solves an interesting electrostatics
problem: notice that the equipotentials for ϕ(x, y) ≡ xy are rectangular hyperbolae, xy = constant.
It follows that ϕ solves the Dirichlet problem for the first quadrant R ≡ {x, y ≥ 0}, bounded
by the positive x-axis and positive y-axis, with ϕ |∂R = 0. Physically, this then models the
electrostatic potential outside the right-angled corner of a conductor! Further examples, that also
have interesting physical applications, may be found in the Feynman lectures.

38
3 Magnetostatics
3.1 Electric currents

So far we have been dealing with stationary charges. In this subsection we consider how to describe
charges in motion.
Recall that in an electrical conductor there are electrons (the “conduction electrons”) which
are free to move when an external electric field is applied. Although these electrons move around
fairly randomly, with typically large velocities, in the presence of a macroscopic electric field there
is an induced average drift velocity v = v(r). This is the average velocity of a particle at position
r. In fact, we might as well simply ignore the random motion, and regard the electrons as moving
through the material with velocity vector field v(r).

δS
v δt

Figure 13: The current flow through a surface element δΣ of area δS.

Definition Given a distribution of charge with density ρ and velocity vector field v, the electric
current density J is defined as

J ≡ ρv . (3.1)

To interpret this, imagine a small surface δΣ of area δS at position r, as shown in Figure 13. Recall
that we define δS = n δS, where n is the unit normal vector to the surface δΣ. The volume of the
oblique cylinder in Figure 13 is v δt · δS, which thus contains the charge ρ v δt · δS = J · δS δt. In
the time δt this is the total charge passing through δΣ. Thus J is a vector field in the direction of
the flow, and its magnitude is the amount of charge flowing per unit time per unit perpendicular
cross-section to the flow.

Definition The electric current I = I(Σ) through a surface Σ is defined to be

Z
I ≡ J · dS . (3.2)
Σ

This is the rate of flow of charge through Σ. The units of electric current are C s−1 , which is also
called the Ampère A.

39
3.2 Continuity equation

An important property of electric charge is that it is conserved, i.e. it is neither created nor
destroyed. There is a differential equation that expresses this experimental fact called the continuity
equation.
Suppose that Σ is a closed surface bounding a region R, so ∂R = Σ. From the discussion of
current density J in the previous subsection, we see that the rate of flow of electric charge passing
out of Σ is given by the current (3.2) through Σ. On the other hand, the total charge in R is
Z
Q = ρ dV . (3.3)
R

If electric charge is conserved, then the rate of charge passing out of Σ must equal minus the rate
of change of Q:
Z Z
dQ ∂ρ
J · dS = − = − dV . (3.4)
Σ dt R ∂t

Here we have allowed time dependence in ρ = ρ(r, t). Using the divergence theorem A.2 this
becomes
Z  
∂ρ
+ ∇ · J dV = 0 , (3.5)
R ∂t

which holds for all R, and thus

∂ρ
+ ∇·J = 0 . (3.6)
∂t

This is the continuity equation.

In magnetostatics we shall impose ∂ρ/∂t = 0, and thus

∇·J = 0 . (3.7)

Definition Currents J satisfying (3.7) are called steady currents.

3.3 Lorentz force and the magnetic field

The force on a point charge q at rest in an electric field E is simply F = q E. We used this to
define E in fact. When the charge is moving the force law is more complicated. From experiments
one finds that if q at position r is moving with velocity u = dr/dt it experiences a force

F = q E(r) + q u × B(r) . (3.8)

Here B = B(r) is a vector field, called the magnetic field, and we may similarly regard the Lorentz
force F in (3.8) as defining B. The magnetic field is measured in SI units in Teslas, which is the
same as N s m−1 C−1 .

40
 Since (3.8) may look peculiar at first sight, it is worthwhile discussing it a little further. The
magnetic component may be written as

Fmag = q u × B . (3.9)

In experiments, the magnetic force on q is found to be proportional to q, proportional to the

magnitude |u| of u, and is perpendicular to u. Note this latter point means that the magnetic force
does no work on the charge. One also finds that the magnetic force at each point is perpendicular
to a particular fixed direction at that point, and is also proportional to the sine of the angle
between u and this fixed direction. The vector field that describes this direction is called the
magnetic field B, and the above, rather complicated, experimental observations are summarized
by the simple formula (3.9).
In practice (3.8) was deduced not from moving test charges, but rather from currents in test
wires. A current of course consists of moving charges, and (3.8) was deduced from the forces on
these test wires.

3.4 Biot-Savart law

If electric charges produce the electric field, what produces the magnetic field? The answer is that
electric currents produce magnetic fields! Note carefully the distinction here: currents produce
magnetic fields, but, by the Lorentz force law just discussed, magnetic fields exert a force on
moving charges, and hence currents.
The usual discussion of this involves currents in wires, since this is what Ampère actually did in
1820. One has a wire with steady current I flowing through it. Here the latter is defined in terms
of the current density via (3.2), where Σ is any cross-section of the wire. This is independent of the
choice of cross-section, and thus makes sense, because of the steady current condition (3.7).11 One
finds that another wire, the test wire, with current I ′ experiences a force. This force is conveniently
summarized by introducing the concept of a magnetic field: the first wire produces a magnetic
field, which generates a force on the second wire via the Lorentz force (3.8) acting on the charges
that make up the current I ′ .
Rather than describe this in detail, we shall instead simply note that if currents produce mag-
netic fields, then fundamentally it is charges in motion that produce magnetic fields. One may
summarize this by an analogous formula to the Coulomb formula (1.7) for the electric field due to
a point charge:

Biot-Savart law A charge q at position r0 moving with velocity v produces a magnetic field
µ0 q v × (r − r0 )
B(r) = . (3.10)
4π |r − r0 |3
11
To see this, use the divergence theorem for a cylindrical region bounded by any two cross-sections and the surface
of the wire.

41
The constant µ0 is called permeability of free space. It is approximately µ0 ≈ 1.257 · 10−6 N A−2 .12
Compare (3.10) to (1.7).

As in electrostatics we also have: the magnetic field obeys the Principle of Superposition.
Using the above laws of magnetostatics we may compute the magnetic field due to the steady
current I in a wire C by summing contributions of the form (3.10):
dr′ × (r − r′ )
Z
µ0 I
B(r) = . (3.11)
4π C |r − r′ |3
To derive this, imagine dividing the wire into segments, with the segment δr′ at position r′ . Suppose
this segment contains a charge q(r′ ) with velocity vector v(r′ ) – notice here that v(r′ ) points in
the same direction as δr′ . From (3.10) this segment contributes
µ0 q v × (r − r′ )
δB(r) = (3.12)
4π |r − r′ |3
to the magnetic field. Now by definition of the current I we have I δr′ = J · δS δr′ = ρ v · δS δr′ ,
where we may take δΣ of area δS to be a perpendicular cross-section of the wire. But the total
charge q in the cylinder of cross-sectional area δS and length |δr′ | is q = ρ δS |δr′ |. We thus deduce
that I δr′ = q v and hence that (3.11) holds.

Remark The formula (3.11) for the magnetic field produced by the steady current in a wire is
also often called the Biot-Savart law.

Example (Magnetic field produced by a current in a long straight wire) As an example of (3.11),
let us compute the magnetic field due to a steady current I in an infinitely long straight wire.
Place the wire along the z-axis, and let P be a point in the (x, y)-plane at distance s from the
origin O, as in Figure 14. This is the point at which we compute the magnetic field B(r), so that
−−→ −−→
OP = r. Let Q be a point on the wire at distance z ′ from the origin, and write OQ = r′ . Then in
−−→
the integral Biot-Savart formula (3.11) we have OQ = r′ = e3 z ′ , where e3 is a unit vector in the
−−→
z-direction, i.e. along the wire, and so dr′ = e3 dz ′ . Notice that r − r′ = QP .
−−→
Next notice that the vector dr′ × (r − r′ ) points in a direction that is independent of OQ = r′ :
it is always tangent to the circle of radius s in the (x, y)-plane. To evaluate the integral in (3.11)
−−→
we thus simply have to compute the magnitude |dr′ × (r − r′ )| = dz ′ |QP | sin θ = s dz ′ , where θ is
−−→ −−→ −−→ √
the angle between OQ and QP , as shown. Since also |r − r′ | = |QP | = s2 + z ′2 , from (3.11) we
compute the magnitude B(s) ≡ |B| of the magnetic field to be
Z ∞
µ0 I s
B(s) = dz ′
4π −∞ (s + z ′2 )3/2
2

µ0 I
= , (3.13)
2πs
12
The recommended 2018 CODATA value is µ0 = (1.256 637 062 12±0.000 000 000 19)·10−6 N A−2 , so it is measured
to about 10 digits. Until 2018, the value was exactly µ0 = 4π · 10−7 N A−2 = 1.256 637 061 43 . . . · 10−6 N A−2 , by the
old definition of the unit Ampère in terms of the force between two parallel wires.

42
 I

z'

O s B
P

Figure 14: Computing the magnetic field around a straight wire carrying a steady current I.

where the integral may be evaluated by setting z ′ = s tan ξ. The magnetic field is thus tangent
to circles in the plane perpendicular to the wire, with a magnitude which decreases inversely with
the perpendicular distance to the wire in this plane. ■

Figure 15: Magnetic field lines around the infinite straight wire. You might have seen a demon-
stration of this with iron filings, where the latter line up with the direction of the magnetic field.

43
3.5 Magnetic monopoles?

Let us return to the point charge q at position vector r0 with velocity v, generating the magnetic
field (3.10). We might as well put the charge at the origin, so r0 = 0. Then note that since
r/r3 = −∇ (1/r) we have
 
v×r
∇· = v · [∇ × ∇ (1/r)] = 0 . (3.14)
r3
Here we have used the identity (A.12) in the first equality, and the fact that the curl of a gradient
is zero in the second equality. Thus we have shown that ∇ · B = 0, except at the origin r = 0.
However, unlike the case of the electric field and Gauss’ law, the integral of the magnetic field
around the point charge is zero. To see this, let Σ be a sphere of radius a centred on the charge,
so that the outward unit normal is n = r/r, and compute
Z    
v×r
Z
µ0 q r
B · dS = 3
· dS = 0 , (3.15)
Σ 4π Σ r r
since v × r is perpendicular to r. By the divergence theorem, it follows that
Z
∇ · B dV = 0 , (3.16)
R

for any region R, and thus

∇·B = 0 . (3.17)

This is another of Maxwell’s equations on the front cover. It says that there are no magnetic
monopoles (i.e. magnetic point charges) that generate magnetic fields, analogous to the way that
electric charges generate electric fields. Instead magnetic fields are produced by electric currents.
Although we have only deduced (3.17) above for the magnetic field of a moving point charge, the
general case follows from the Principle of Superposition.
You might wonder what produces the magnetic field in permanent magnets, such as bar magnets.
Where is the electric current? We discuss this in section 4.3.

* Mathematically, it is certainly possible to allow for magnetic monopoles and magnetic

currents in Maxwell’s equations. In fact the equations then become completely symmetric
under the interchange of E with −c B, c B with E, and corresponding interchanges of
electric with magnetic charge densities and currents. Here c2 ≡ 1/ϵ0 µ0 . There are also
theoretical reasons for introducing magnetic monopoles. For example, the quantization of
electric charge – that all electric charges are an integer multiple of some fixed fundamental
quantity of charge – may be understood in quantum mechanics using magnetic monopoles.
This is a beautiful argument due to Dirac. However, no magnetic monopoles have ever
been observed in nature. If they existed, (3.17) would need correcting.

3.6 Ampère’s law

There is just one more static Maxwell equation to discuss, namely the equation involving the curl
of B. This is Ampère’s law. In many treatments of magnetostatics, this is often described as an

44
additional experimental result, and/or is derived for special symmetric configurations, having first
solved for B using the Biot-Savart law (3.10) or (3.11). However, it is possible to derive Ampère’s
law directly from (3.10), as we now show.
We first use (3.10) and the Principle of Superposition to write B generated by a current density
J as a volume integral
J(r′ ) × (r − r′ )
Z
µ0
B(r) = dV ′ . (3.18)
4π r′ ∈R |r − r′ |3

This follows directly from the definition J = ρ v in (3.1) and taking the limit of a sum of terms of
the form (3.10). R ⊂ R3 is by definition a region containing the set of points with J ̸= 0. We next
define the vector field
J(r′ )
Z
µ0
A(r) ≡ dV ′ . (3.19)
4π R |r − r′ |
One then computes (cf. the proof of Theorem 1.4)

Ji (r′ )
Z
∂Ai µ0
(r) = − (xj − x′j ) dV ′ , (3.20)
∂xj 4π R |r − r′ |3

where r = (x1 , x2 , x3 ), r′ = (x′1 , x′2 , x′3 ). From the first line of (A.9) we have
3 3
X ∂A X ∂Ai
∇×A ≡ ej × =− (ei × ej ) , (3.21)
∂xj ∂xj
j=1 i,j=1

so that comparing (3.20) with (3.18) we see that

B(r) = ∇ × A(r) . (3.22)

Note that (3.22) immediately implies ∇ · B = 0, as the divergence of a curl is zero.

Lemma 3.1 For steady currents J, supported inside a bounded region R, the vector field A defined
by (3.19) satisfies ∇ · A = 0.

Proof Using
   
1 ′ 1
∇ = −∇ , r ̸= r′ , (3.23)
|r − r′ | |r − r′ |

where ∇′ denotes derivative with respect to r′ , from (3.19) we compute

Z   Z  
µ0 ′ 1 ′ µ0 ′ ′ 1
∇·A = J(r ) · ∇ dV = − J(r ) · ∇ dV ′
4π R |r − r′ | 4π R |r − r′ |
J(r′ )
Z   Z
µ0 ′ ′ µ0 1
∇′ · J(r′ ) dV ′ .


= − ∇ · ′
dV + ′
(3.24)
4π R |r − r | 4π R |r − r |

The second term on the right hand side of this equation is zero for steady currents, satisfying (3.7).
Moreover, we may use the divergence theorem on the first term to obtain a surface integral on ∂R.
But by assumption J vanishes on this boundary, so this term is also zero. ■

45
 We next take the curl of (3.22). Using the identity

∇ × (∇ × A) = ∇ (∇ · A) − ∇2 A , (3.25)

which holds for any vector field A, together with Lemma 3.1, we deduce that

∇ × B = −∇2 A . (3.26)

On the other hand, from (3.19)

Z   Z
µ0 ′ 1 ′ µ0
2 2
J(r′ ) −4π δ(r − r′ ) dV ′ = −µ0 J(r) , (3.27)


∇ A = J(r ) ∇ ′
dV =
4π R |r − r | 4π R
where we have used Proposition 1.3 in the middle step. We hence deduce

∇ × B = µ0 J . (3.28)

This is Ampère’s law for magnetostatics. It is the final Maxwell equation on the front page, albeit
in the special case where the electric field is independent of time, so ∂E/∂t = 0. Notice this
equation is consistent with the steady current assumption (3.7).
We may equivalently rewrite (3.28) using Stokes’ theorem as

Ampère’s law For any simple closed curve C = ∂Σ bounding a surface Σ

Z Z
B · dr = µ0 J · dS = µ0 I , (3.29)
C=∂Σ Σ

where I is the current through Σ.

Example Notice that integrating the magnetic field B given by (3.13) around a circle C in the
(x, y)-plane of radius s, centred on the z-axis, indeed gives µ0 I. ■

3.7 Magnetostatic vector potential

Definition In magnetostatics the magnetic vector potential is a vector field A such that the
magnetic field B is given by

B = ∇×A . (3.30)

In fact we have already introduced such a vector potential in equations (3.19) and (3.22). It is
analogous to the electrostatic potential ϕ in electrostatics.

* Notice that (3.30) is a sufficient condition for the Maxwell equation (3.17) to hold, since
the divergence of a curl is zero. It is also necessary if we work in a domain with simple
enough topology, such as R3 or an open ball. An example of a domain where not every
vector field B with zero divergence may be written as a curl is R3 \ {point}. Compare this
to the corresponding starred paragraph in section 1.5. Again, a proof for an open ball is
contained in appendix B of the book by Woodhouse.

46
 The vector field A in (3.30) is far from unique: since the curl of a gradient is zero, we may add
∇ψ to A, for any function ψ, without changing B:

A → A
b ≡ A + ∇ψ . (3.31)

That is, B = ∇ × A = ∇ × A.
b

Definition The transformation (3.31) is called a gauge transformation of A.

We may fix this gauge freedom by imposing additional conditions on A. For example, suppose we
have chosen a particular A satisfying (3.30). Then if ψ is a solution of the Poisson equation

∇2 ψ = −∇ · A , (3.32)

it follows that A
b in (3.31) satisfies

∇·A
b = 0. (3.33)

Definition The condition (3.33) on the magnetostatic vector potential is called the Lorenz gauge.13

From Theorem 2.2 we know that solutions ψ to the Poisson equation (3.32) are unique, for fixed
boundary conditions.
Many equations simplify with this gauge choice for A. For example, Ampère’s law (3.28) is

µ0 J = ∇ × (∇ × A) = ∇ (∇ · A) − ∇2 A , (3.34)

so that in Lorenz gauge ∇ · A = 0 this becomes

∇2 A = −µ0 J . (3.35)

Compare with Poisson’s equation (1.32) in electrostatics. Notice that Lemma 3.1 shows that the
vector potential A given by (3.19) is in Lorenz gauge.

* Gauge invariance and the vector potential A play a fundamental role in more advanced
formulations of electromagnetism. The magnetic potential also plays an essential physical
role in the quantum theory of electromagnetism.

3.8 Multipole expansion

In electrostatics and magnetostatics we have now derived the similar formulae

ρ(r′ ) J(r′ )
Z Z
1 ′ µ0
ϕ(r) = dV , A(r) = dV ′ . (3.36)
4πϵ0 r′ ∈R |r − r′ | 4π r′ ∈R |r − r′ |

In both cases R ⊂ R3 is a bounded region, and the first formula gives the electrostatic potential
ϕ generated by a charge density ρ supported inside R, while the second formula gives the mag-
netostatic vector potential A generated by a steady current density J supported inside R. The
13
Mr Lorenz and Mr Lorentz were two different people.

47
electric and magnetic fields are obtained from these via E = −∇ϕ, B = ∇ × A, respectively. In
this subsection we want to examine what these fields look like far away from the localized source
region R. Here “far away” means that the observation point P , with position vector r measured
from the origin O, is at a distance that is large compared to the size of R – see Figure 16.14

region R P

r
r'

O
Figure 16: Observing the fields, generated by sources supported in a region R, at a point P far
away from R.

Our starting point is the Taylor expansion

1 1 1 1
= q = + 3 r · r′ + O(1/r3 ) . (3.37)
|r − r′ | r 1− 2
r · r′ + 1
r′ · r′ r r
r2 r2
√
Here as usual r ≡ |r| = r · r, and we have expanded for r ≫ r′ . In particular the Cauchy-Schwarz
inequality gives |r · r′ | ≤ |r| |r′ | = r r′ , so that the second term on the right hand side of (3.37) is
O(1/r2 ). Applying (3.37) to the electrostatic potential in (3.36) immediately gives
Q 1 1 p·r
ϕ(r) = + + O(1/r3 ) , (3.38)
4πϵ0 r 4πϵ0 r3
where we have defined
Z Z
′ ′
Q ≡ ρ(r ) dV , p ≡ r′ ρ(r′ ) dV ′ . (3.39)
r′ ∈R r′ ∈R

Here Q is simply the total charge, and (3.38) says that far away from the localized charge distri-
bution, the electrostatic potential to leading order looks like that generated by a point charge Q
(at the origin r = 0). This is, of course, entirely sensible.

Definition The vector p defined in (3.39) is called the electric dipole moment.

This governs the next-to-leading order term in the general expansion (3.38), and is the leading
order term when the total charge Q = 0. On the first problem sheet you will have computed the
electrostatic potential of an electric dipole. This is a configuration of two point charges ±q, with
14
The methods developed in this subsection apply also to other areas of theoretical physics, perhaps most notably
to gravitational waves in General Relativity, that might have been generated by a distant collision and subsequent
merger of two black holes.

48
separation vector d. Defining p ≡ q d, then in the limit that q → ∞, d ≡ |d| → 0, with q d = |p|
held fixed and the charges at the origin, you showed that the resulting electrostatic potential is
1 p·r
ϕ(r)dipole = . (3.40)
4πϵ0 r3
This is precisely the second term on the right hand side of (3.38). It arises when there is asymmetry
in the charge distribution around the origin.

Remark One might be tempted to define a centre of charge for a general charge distribution
as d ≡ p/Q, with Q and p defined in (3.39), in precise analogy with centre of mass for a mass
distribution. However, charge distributions are often neutral, with Q = 0, and the electric dipole
moment then governs the leading order behaviour of the resulting electric field. For example, a
neutral molecule, such as a water molecule, will involve shared electrons between atoms. These
are not distributed uniformly around the molecule, which thus behaves like a tiny electric dipole.
The behaviour of a large number of small electric dipoles is discussed in section 4.

Using (3.37) we may perform a similar expansion of the magnetostatic vector potential in (3.36):
 Z Z 
µ0 1 ′ ′ 1 ′ ′ ′
A(r) = J(r ) dV + 3 (r · r ) J(r ) dV + · · · . (3.41)
4π r R r R

The following result is useful for manipulating this expression:

Lemma 3.2 For any vector field J satisfying ∇ · J = 0 in a region R, with J being zero on the
closed boundary Σ = ∂R of R, and any functions f , g, we have
Z
(f J · ∇g + g J · ∇f ) dV = 0 . (3.42)
R

Proof Using the product rule ∇ · (f g J) = f g ∇ · J + J · ∇(f g) we may write the first term in
(3.42) as
Z Z Z Z
f J · ∇g dV = ∇ · (f g J ) dV − f g ∇ · J dV − g J · ∇f dV
R ZR Z R RZ

= f g J · dS − 0 − g J · ∇f dV = − g J · ∇f dV . (3.43)
Σ R R

Here in the second equality we have used the divergence theorem for the first term, and ∇ · J = 0
for the second term, and the last equality follows since J is assumed to be zero on Σ = ∂R. ■

Corollary 3.3 For a steady current density J supported inside a region R

Z
J(r) dV = 0 . (3.44)
R
R
Proof Applying Lemma 3.2 with f ≡ 1 and g = xi , with r = (x1 , x2 , x3 ), gives R J · ∇xi dV =
R
R Ji dV = 0, where in the first equality note (∇xi )j = δij . ■

49
It follows that the first term on the right hand side of the expansion (3.41) is always zero! This is
ultimately a reflection of the fact there are no magnetic point charges, analogous to electric point
charges in electrostatics. For the second term in (3.41) we may again use Lemma 3.2, this time
with f = x′i , g = x′j , and we take the integration to be over the primed variable r′ . From (3.42)
we then deduce
Z Z
 ′ ′ ′ ′ ′ ′ ′ ′ ′
 ′
xi Jj (r′ ) + x′j Ji (r′ ) dV ′ ,
 
0 = xi J(r ) · ∇ xj + xj J(r ) · ∇ xi dV = (3.45)
R R

and hence, for any vector f , we have

Z 3
X Z  3
X Z 
′ ′ ′
(f · r ) J(r ) dV = fi x′i Jj (r′ ) dV ′ ej = − fi x′j ′
Ji (r ) dV ′
ej
R i,j=1 R i,j=1 R
Z
= (f · J(r′ )) r′ dV ′ . (3.46)
R

Combined with the vector triple product identity (A.6), we can thus write
Z Z 
′ ′ ′ 1 ′ ′ ′
(f · r ) J(r ) dV = − f × r × J(r ) dV . (3.47)
R 2 R

Definition The magnetic dipole moment generated by a steady current density J in R is

Z
1
m ≡ r′ × J(r′ ) dV ′ . (3.48)
2 r′ ∈R

This is analogous to the definition of electric dipole moment we gave earlier, and combining (3.47)
for f = r with (3.41) we have proven

Proposition 3.4 The magnetostatic vector potential generated by a steady current density J in a
localized region R has an expansion with leading order term
µ0 m × r
A(r) = + O(1/r3 ) , (3.49)
4π r3
as r → ∞, where m is the magnetic dipole moment (3.48) generated by the current density.

The correction terms O(1/r3 ) in (3.49) are called higher moments of J, starting with the quadrupole
moment, and can be computed with more effort.

Definition Analogously to (3.40), we may define the magnetic dipole vector potential to be
µ0 m × r
A(r)dipole ≡ . (3.50)
4π r3

This gives the leading order term in the expansion (3.49). The corresponding magnetic field is
µ0  r µ0 h r  r i
B(r)dipole = ∇ × A(r)dipole = ∇× m× 3 = m∇ · 3 − m · ∇ 3
 4π  r 4π r r
µ0 m 3(m · r) r
= − 3 + , (3.51)
4π r r5

50
Figure 17: Magnetic field lines for the magnetic dipole (3.51), where we have aligned the magnetic
dipole moment m pointing upwards along the z-axis direction. The magnetic field lines of a bar
magnet are similar, where the bar magnet is aligned vertically along the z-axis, with north pole
at the top and south pole at the bottom.

where r ̸= 0, and on the first line we have used the identity (A.11). Notice here that ∇ · (r/r3 ) = 0
for r ̸= 0, as in our computation of (1.8). The field lines for B(r)dipole are shown in Figure 17.

To summarize, we have shown that, far away from any localized steady current source, the
magnetic field generated will take the form (3.51), to leading order, and thus resemble Figure 17.
Notice that in this figure m points upwards along the z-axis, and the corresponding magnetic field
lines come out of the top of the magnetic dipole, which is called the north pole, and go into the
bottom of the magnetic dipole, which is called the south pole.

51
4 Macroscopic media
4.1 Dielectrics

In order to write down the electrostatic or magnetostatic Maxwell equations, we need to know the
charge density ρ or current density J precisely, in principle everywhere in space. Except for certain
idealized situations, such as for point charges in vacuum, this is usually not possible in practice.
For instance, suppose we wish to study the effects of electromagnetism in water. In a cubic
centimetre of water there are around 1022 water molecules. As described in section 3.8, a water
molecule is neutral, having total charge zero, but this then behaves like a tiny electric dipole: the
negative electron and positive proton charges are not uniformy distributed in the molecule, and
this generates an electric dipole moment p, with dipole field (3.40). One can view such an electric

E
- -
+

+
+

+
+

-
+

-
-

-
-

+
+

-
+
-

+
-

+ -

- +
+ - +

-
-

+
+

+
+
-
-

+
-

-
+
+

-
- +

- - +
-
+

+
-

-
-

-
+
-

+
+
+

-
+
+ +
-

- +

(a) Unpolarized. (b) Medium polarized by external E.

Figure 18: The dipoles in a dielectric medium.

dipole as in Figure 18, with a positive end + and a negative end −, with p pointing from the
negative end towards the positive end.
A dielectric medium, such as water, by definition contains a very large number of these dipoles.
With no external electric field applied, the dipoles are aligned somewhat randomly, as in Figure 18a,
due to their random thermal motion. Such a configuration is said to be unpolarized, with no net
macroscopic electric field produced by the dipoles. However, now consider the effect of turning on
an external electric field E: the electric dipoles will align with E, as in Figure 18b. In doing so the
large number of aligned dipoles will superpose to generate their own macroscopic electric field. In
Figure 18b notice that the negative ends of the dipoles are on the left, while the positive ends are
on the right, creating an effective surface charge density on the left and right ends of the box. An
electric field generated by a positive/negative charge points away from/towards it, respectively, so
the electric field generated by the dipoles is in the opposite direction to E, reducing the overall
electric field.

4.2 Electric dipoles

We may describe this more quantitatively by examining electric dipoles in a little more detail.
Recall that an electric dipole is realized by starting with two point charges ±q, with separation

52
vector d, and taking the limit in which the distance d ≡ |d| between the charges tends to zero, the
positive charge q is taken to infinity, with the electric dipole moment p ≡ q d held fixed.
If we place such an electric dipole in an external electric field E, what force does it experience?
The total force on the pair of point charges is (see Figure 19)

F = q E r + d2 − q E r − d2

= q E(r) + d2 · ∇ E(r) + O(d2 ) − q E(r) − d

· ∇ E(r) + O(d2 )

 

2

= q (d · ∇) E(r) + O(d2 ) −→ (p · ∇) E .
 
(4.1)

Here we have placed the charge +q at position vector r + d/2 and the charge −q at position
vector r − d/2, and then used Taylor’s theorem and taken the point dipole limit. This force is
conservative, with potential energy function Vdipole given by

F = −∇Vdipole , Vdipole ≡ −p · E = p · ∇ϕ . (4.2)

To see this, we may compute

−∇Vdipole = −∇ [(p · ∇)ϕ] = (p · ∇)(−∇ϕ) = (p · ∇) E . (4.3)

Here we have used the fact that p is constant to pass the gradient ∇ through the directional
derivative p · ∇. The potential energy Vdipole ≡ −p · E is minimized when p points in the same
direction as E, and maximized when p points in the opposite direction to E.

force
+
q
p

d /2
electric field

r E
- d /2

-q O

-
force

Figure 19: The electrical forces on a dipole consisting of a positive point charge q at position vector
r + d/2, and a point charge −q at position vector r − d/2.

There is also a torque on the dipole, causing it to rotate. The total torque about the point r is
     
d d d d
τ = × qE r+ − × −q E r − → p×E . (4.4)
2 2 2 2

53
The torque is perpendicular to the plane defined by the electric dipole moment p and external
electric field E, and gives a rotational force about this axis, as shown in Figure 19.15 In equilibrium
the torque on a dipole is by definition zero, but that is the case if and only if p is aligned with E,
so their cross product is zero, precisely as in Figure 18b. (Notice that if p points in the opposite
direction to E this gives an unstable equilibrium, being a maximum of the potential energy Vdipole ).
An electric dipole of course also generates its own electric field. Recall that the electrostatic
potential generated by a single electric dipole moment p at the origin is
 
1 p·r 1 1
ϕ(r)dipole = 3
= − p·∇ . (4.5)
4πϵ0 r 4πϵ0 r
A dielectric medium by definition consists of a very large number of electric dipoles. By the
Principle of Superposition, dipole moments pi at position vectors ri will generate a potential
N
1 X pi · (r − ri ) P(r′ ) · (r − r′ )
Z
1
ϕ(r)dipoles = −→ dV ′ . (4.6)
4πϵ0 |r − ri |3 4πϵ0 r′ ∈R |r − r′ |3
i=1

Here we have taken the usual continuum limit on the right hand side, where by definition P(r′ ) δV ′
is the electric dipole moment in a small volume δV ′ , centred at position r′ .

Definition The vector field P is called the electric polarization density, describing the distribution
of dipoles in the dielectric medium.

On the other hand, as on the right hand side of (4.5) we may then calculate
Z   Z  
1 ′ 1 ′ 1 ′ ′ 1
ϕ(r)dipoles = − P(r ) · ∇ dV = P(r ) · ∇ dV ′
4πϵ0 r′ ∈R |r − r′ | 4πϵ0 r′ ∈R |r − r′ |
P(r′ ) ∇′ · P(r′ )
Z   Z
1 1
= ∇′ · ′
dV ′
− dV ′
4πϵ0 r′ ∈R |r − r | 4πϵ0 r′ ∈R |r − r′ |
(−∇′ · P(r′ ))
Z
1
= dV ′ . (4.7)
4πϵ0 r′ ∈R |r − r′ |

Here we have used (3.23) on the first line, and in the last step have used the divergence theorem,
with P by definition being zero on, and outside, the boundary of the region R in which the dipoles
are supported. Comparing to (1.28) we may define

ρbound (r) ≡ −∇ · P(r) . (4.8)

This is the effective charge density that produces the electrostatic potential (4.7) due to dipoles. It
is called ρbound (r) as it is effectively generated by charges that are bonded to their overall neutral
molecules. Notice that the definition (4.8) looks like Gauss’ law (1.16), with −P/ϵ0 playing the
role of the electric field.
What is Gauss’ law in this set-up? We may divide the total charge density ρ into two types: (i)
the bound charge density ρbound in (4.8), that generates an electric field due to the polarization of
15
Recall that an applied torque is equal to the rate of change of angular momentum.

54
the dielectric medium, and (ii) an “ordinary” charge density, which we call ρfree . The latter consists
of any charge added to the interior or exterior of the dielectric medium, where we effectively ignore
the charges that make up the dipoles in the material (those are accounted for in ρbound ). Then
Gauss’ law (1.16) reads
ρ 1 1 1
∇·E = = (ρfree + ρbound ) = ρfree − ∇ · P . (4.9)
ϵ0 ϵ0 ϵ0 ϵ0
On the other hand, our discussion of Figure 18 led us to conclude that the polarization density P
is everywhere aligned with the electric field E . That is, P(r) × E(r) = 0, so there is no torque on
the dipoles at position r. We thus write
 
1 ϵ
P ≡ χe E ≡ −1 E . (4.10)
ϵ0 ϵ0
Definition ϵ in (4.10) is called the permittivity of the dielectric medium, with χe ≡ (ϵ/ϵ0 − 1)
called the electric susceptibility, which measures the degree of response of the dielectric to an
applied electric field.

Here in general ϵ may depend on: (i) position r if the dielectric medium is not uniform (ϵ may
depend on local temperature, pressure, etc), (ii) time t if the medium is not static, and (iii) the
frequency, magnitude and direction of the electric field. However, in many practical situations
ϵ may be treated as approximately constant within the medium. For the vacuum ϵ/ϵ0 = 1, so
that P = 0 and there is no induced polarization, while for other media ϵ/ϵ0 > 1, e.g. for air
ϵ/ϵ0 ≃ 1.0005, for water ϵ/ϵ0 ≃ 1.77.
Substituting (4.10) into (4.9), we have
   
1 1 ϵ
ρfree = ∇ · E + P = ∇ · E , (4.11)
ϵ0 ϵ0 ϵ0
and the electrostatic Maxwell equations become simply16

∇ · (ϵ E) = ρfree , ∇×E = 0 . (4.12)

Notice that all we have done is effectively replace ϵ0 → ϵ! When ϵ is constant the resulting
Maxwell equations are mathematically identical to the electrostatic Maxwell equations we studied
in sections 1 and 2, and may be solved with exactly the same methods. Note here that since
ϵ/ϵ0 > 1, the electric field E generated by the free charges ρfree in a dielectric medium is smaller
than it would have been without the dielectric present, by a factor of ϵ0 /ϵ.
When we have boundaries between materials with different values of ϵ, for example, air and
water, the following version of Proposition 1.8 applies (we shall need this result in section 6):

Proposition 4.1 Consider a surface S with (free ) surface charge density σ, that is the boundary
between two dielectric materials with different permittivities ϵ± . Then

ϵ+ E+ · n − ϵ− E− · n = σ , (4.13)
16
The quantity D ≡ ϵ E in (4.11) is called the electric displacement.

55
relates the electric fields E± on the two sides of S, with n the unit normal pointing into the “+
side”. On the other hand, the components of E tangent to S are continuous across S.

Proof The proof is almost identical to that for Proposition 1.8, where instead in the first step we
integrate Gauss’ law ∇ · (ϵ E) = ρfree over the cylindrical region R. The divergence theorem then
gives terms ϵ± E± integrated through the top and bottom surfaces of this cylinder, respectively.
The equation ∇ × E = 0 is the same as before, and so proof for the components of E tangent to S
is identical to that for Proposition 1.8. ■

4.3 Magnetic dipoles

In section 3 we explained that electric currents generate magnetic fields, via the Biot-Savart law
(3.10), and conversely magnetic fields exert a force on electric currents via the magnetic component
of the Lorentz force (3.8). In fact these two statements essentially summarize magnetostatics. But
you might then ask: where is the electric current that generates the magnetic field of a permanent
magnet? Similarly, in Figure 15 we mentioned that you might have seen iron filings lining up with
the magnetic field lines around a current carrying wire: but why do they line up? If the magnetic
force in (3.8) is responsible for this, where is the current in an iron filing, or in the needle of a
compass that aligns with the Earth’s magnetic field?!

* Remark on spin Perhaps surprisingly, the answers to these questions involve quantum me-
chanics in an essential way, even though the effects are macroscopic. In some magnetic materials,
the macroscopic magnetic field is indeed produced by the alignment of tiny atomic currents gener-
ated by the electrons in the material. However, in materials such as iron it is the alignment of the
“spins” of (unpaired) electrons in the atoms that is responsible for producing the magnetic field.
If you took the Part A Quantum Theory course, you will have encountered the fact that a single
electron has an intrinsic angular momentum, called its spin. A proper discussion of this quantum
mechanical notion is beyond our course here, but we can understand to some extent why “spin”
angular momentum might generate a magnetic field, via the following classical argument.
First recall the definition (3.48) of the magnetic dipole moment:
Z
1
m ≡ r′ × J(r′ ) dV ′ . (4.14)
2 r′ ∈R

Here J is a steady current density supported in a region R, and the magnetic dipole moment deter-
mines the leading order behaviour of the vector potential A(r) in the expansion (3.49). Recalling
also that J = ρ v, for a charge density ρ with velocity vector v, notice that the expression (4.14)
is very similar to that for angular momentum. More precisely, for a mass distribution with density
ρ(r)mass supported in a region R, the angular momentum (about the origin O) is
Z
L = r′ × ρmass v dV ′ . (4.15)
r′ ∈R

56
Here ρ(r′ )mass δV ′ is the mass of a small volume δV ′ centred at r′ . Let us assume that the charge
and mass densities of some matter are proportional to each other, so ρ = 2γ ρmass with γ a constant;
for example, this is the case if the matter is all made of the same elementary particles. Then
Z Z
1 ′ ′
m = r × ρ v dV = γ r′ × ρmass v dV ′ = γ L . (4.16)
2 r′ ∈R r′ ∈R

The magnetic dipole moment and angular momentum of the matter are hence proportional, with
gyromagnetic ratio γ satisfying
Z Z
Q
Q = ρ dV = 2γ ρmass dV = 2M γ ⇒ γ = , (4.17)
R R 2M
where M is the total mass. It is then the angular momentum L of the charge distribution that is
effectively generating the magnetic dipole field in (3.51), with m = γ L.
We might then crudely imagine an electron as a ball of charge that is spinning about some
axis, with the resulting angular momentum of this body being its “spin” angular momentum. The
rotating charge gives an electric current, which in turn generates a magnetic dipole moment m.
This classical picture is not really a correct description of an electron, but it is nevertheless true
that a single electron behaves like a little magnetic dipole, with the magnetic dipole moment
aligned with the direction of its spin. In a ferromagnetic material, such as iron, the spins of the
electrons can all be aligned, and superposing all the magnetic dipole fields of the electrons leads
to the macroscopic magnetic field in your fridge magnets. [end of remark on spin]

As in the previous subsection, we can make all of this more quantitative by studying magnetic
dipoles in more detail. An electric dipole can be constructed from two point charges ±q, in a limit
where the charges coalesce. There is a similar construction for a magnetic dipole, although it is a
little more fiddly as magnetic fields are generated from currents, not point charges. We start from
the general formula (3.19) for the vector potential:
J(r′ )
Z
µ0
A(r) = dV ′ . (4.18)
4π R |r − r′ |
Here J δV ′ = ρ v δV ′ = q v, where q is the charge in the small volume δV ′ , centred at position r′ .
On the other hand, precisely as in our derivation of the integral Biot-Savart law formula (3.11),
we may identify I δr′ = q v for a current I flowing through a loop C, with segment δr′ . Thus such
a loop of current generates a magnetic vector potential
dr′
Z
µ0 I
A(r) = . (4.19)
4π C |r − r′ |
To construct a point magnetic dipole at the origin, we take C to be a small circle, where we
choose our coordinate axes so that this lies in the (x, y)-plane, with centre at the origin. Then
r′ = (a cos φ′ , a sin φ′ , 0) parametrizes C, with a > 0 the radius of this circle, and the Taylor
expansion (3.37) gives
1 r · r′
Z  
µ0 I ′
A(r) = dr + 3 + O(|r′ |2 ) . (4.20)
4π C r r

57
We will eventually take |r′ | = a → 0. The first term in the expansion (4.20) is zero, as the
fundamental theorem of calculus gives C dr′ ≡ (dr′ /dφ′ ) dφ′ = 0, since the circle C is a closed
R R

loop. Thus with dr′ = (−a sin φ′ , a cos φ′ , 0) dφ′ , r = (x, y, z) we may evaluate (4.20) explicitly as
Z 2π
µ0 I
(−a sin φ′ , a cos φ′ , 0 ) x a cos φ′ + y a sin φ′ + O(a2 ) dφ′
 
A(r) = 3
4πr 0
µ0 I  2
πa e3 × r + O(a3 ) .

= 3
(4.21)
4πr
Comparing to the magnetic dipole vector potential (3.50), we thus define

m ≡ Iπa2 e = I · area(C) e , (4.22)

where area(C) = πa2 is the area enclosed by C, and e is a unit vector perpendicular to this surface,
which here is e = e3 because of how we aligned our coordinate axes. By analogy with the electric
dipole, we then take I → ∞ and a → 0, holding m fixed. In this limit the O(a3 ) terms in (4.21)
do not contribute, and the vector potential (4.21) is exactly the dipole vector potential (3.50).
If we place such a magnetic dipole in an external magnetic field B, what force does it experience?
The dipole is generated by the circular current I in C, and it is the magnetic component q v × B of
the Lorentz force (3.8) that acts on the moving charges in this current. Recalling that q v = I δr′
for an element of current, we sum these forces to obtain
Z
F = I dr′ × B(r′ )
C
Z 2π
(−a sin φ′ , a cos φ′ , 0 ) × B(0) + ∂x B(0) a cos φ′ + ∂y B(0) a sin φ′ + O(a2 ) dφ′
 
= I
0
= I πa2 [e2 × ∂x B(0) − e1 × ∂y B(0)] + O(a3 )
= I πa2 [∂x B3 (0) e1 + ∂y B3 (0) e2 − (∂x B1 (0) + ∂y B2 (0)) e3 ] + O(a3 )


= I πa2 [∇ (B3 (0)) − (∇ · B(0)) e3 ] + O(a3 ) .


(4.23)

Here in the second line we have Taylor expanded B about the origin, where the dipole is, and have
then proceeded to evaluate the integral and cross products explicitly. Taking the point magnetic
dipole limit, where m = Iπa2 e3 , then gives

F = ∇ (m · B) − (∇ · B) m = ∇ (m · B) , (4.24)

where we have used the Maxwell equation ∇ · B = 0. This force is conservative, with potential

F = −∇ Vdipole , Vdipole ≡ −m · B . (4.25)

Remarkably, this is exactly the same as for the force on an electric dipole in (4.2), where we replace
electric dipole moment p by magnetic dipole moment m, and electric field E by magnetic field B!
The torque about the origin is
Z Z
′ ′ ′
dr′ r′ · B(r′ ) ,
   
τ = r × I dr × B(r ) = I (4.26)
C C

58
where we have used the vector triple product identity (A.6), together with the fact that r′ is
orthogonal to dr′ for the circle C. We may similarly compute this to obtain
Z 2π
(−a sin φ′ , a cos φ′ , 0 ) B1 (0) a cos φ′ + B2 (0) a sin φ′ + O(a2 ) dφ′
 
τ = I
0 2
= I πa (−B2 (0) , B1 (0) , 0 ) + O(a3 ) → m × B .

(4.27)

Again, (4.27) is the same as the torque (4.4) on an electric dipole, but replacing p → m, E → B.
Magnetic dipoles in an external magnetic field thus behave in exactly the same way as electric
dipoles behave in an external electric field. In particular, magnetic dipoles will tend to align
everywhere with B. This explains why iron filings align in an external magnetic field: an iron
filing behaves as a magnetic dipole, due the alignment of electron spins within it. But actually
you might have noticed that even more is true: the electric and magnetic fields produced by point
dipoles at the origin are respectively (for r ̸= 0)
   
1 p 3(p · r) r µ0 m 3(m · r) r
Edipole = − 3+ , Bdipole = − 3 + . (4.28)
4πϵ0 r r5 4π r r5

They take exactly the same form!

(a) The magnetic field generated by a small cir- (b) The electric field generated by nearby point
cular current loop lying in a horizonal plane per- charges lying on a vertical axis, with the positive
pendicular to the page. charge above the negative charge.

Figure 20: Comparing dipoles: the field lines look identical far from the centre, as in (4.28), but
notice they point upwards in the middle of Figure 20a, and downwards in the middle of Figure 20b.

This might lead you to suspect there is more than just an analogy going on here: does this
mean that a point magnetic dipole can also be constructed from two point magnetic charges? To
some extent the answer is yes, at least mathematically, but conceptually this is wrong: as far
as we know, point magnetic charges don’t exist. Nevertheless, some textbooks introduce point

59
magnetic charges for precisely this purpose, and indeed many physicists will then use this model
when thinking about the behaviour of magnetic fields. For example, does the north pole of one
magnetic dipole attract or repel the north pole of another magnetic dipole? The answer can be
determined using (4.25) and (4.27), and knowing the form of the dipole magnetic field in Figure 17;
but viewing the north poles as positive point magnetic charges makes it immediately clear they
repel, which is correct! More fundamentally though, in (4.28) have taken an idealized point dipole
limit: the field lines near the “core” of a finite sized current loop and pair of nearby point charges
±q actually point in opposite directions – see Figure 20.
The effective magnetostatic Maxwell equations in a material can be derived in a precisely anal-
ogous way to those for electrostatics in a dielectric medium. A large number of magnetic dipole
moments mi at positions ri will generate a vector potential
N
µ0 X mi × (r − ri ) M(r′ ) × (r − r′ )
Z
µ0
A(r)dipoles = → dV ′ , (4.29)
4π |r − ri |3 4π R |r − r′ |3
i=1

with continuum limit precisely as in (4.6), so that M(r′ ) δV ′ is the magnetic dipole moment in a
small volume δV ′ , centred at position r′ .

Definition The vector field M is called the magnetization density.

A similar computation to (4.7) then gives

Z  ′
∇ × M(r′ ) M(r′ )
Z    
µ0 ′ ′ 1 ′ µ0 ′
A(r)dipoles = M(r ) × ∇ dV = −∇ × dV ′
4π R |r − r′ | 4π R |r − r′ | |r − r′ |
∇′ × M(r′ ) M(r′ )
Z Z
µ0 ′ µ0
= dV + × dS′
4π R |r − r′ | 4π ∂R |r − r′ |
∇′ × M(r′ )
Z
µ0
= dV ′ . (4.30)
4π R |r − r′ |
Here in the second line we have used a corollary of the divergence theorem to write the volume
integral of a curl as a boundary integral of a cross product, and in the last step we have assumed
that M is zero on the boundary of R, the region containing the magnetic dipoles. Comparing
(4.30) to (3.19) we may define

JM ≡ ∇ × M . (4.31)

The subscript M here denotes these are effective magnetizing currents, that generate the vector
potential (4.30) due to magnetic dipoles in the material.
We may then divide the electric current in Ampère’s law (3.28) into a free current density Jfree
and the magnetizing current JM in (4.31). Thus

∇ × B = µ0 J = µ0 (Jfree + JM ) = µ0 Jfree + µ0 ∇ × M . (4.32)

On the other hand, the magnetization density M will align everywhere with the magnetic field B,
so that the cross product M(r) × B(r) = 0, and there is no torque on the magnetic dipoles. Thus
 
χm 1 1
M ≡ B ≡ − B. (4.33)
µ µ0 µ

60
Definition µ is called the permeability, with χm ≡ (µ/µ0 − 1) the magnetic susceptibility.

µ is not constant in general, but for uniform materials it is approximately constant. For the vacuum
µ/µ0 = 1, for air µ/µ0 ≃ 1.00000037, for water µ/µ0 ≃ 0.999992, while for iron µ/µ0 ≃ 200, 000!
Substituting (4.33) into (4.32), we have
 
1
µ0 Jfree = ∇ × (B − µ0 M) = µ0 ∇ × B . (4.34)
µ
which leads to the effective Maxwell equations
 
1
∇·B = 0 , ∇× B = Jfree . (4.35)
µ
Notice that the magnetic field generated by Jfree in a magnetic material is µ/µ0 times the field that
would be generated without the magnetic material present. This is e.g. a little larger for air, since
χm > 0 (called paramagnetism), but a little smaller for water, since χm < 0 (called diamagnetism).
1
Definition The quantity H ≡ µ B in (4.35) is called the magnetic field strength.

Magnetism is more complicated than electric polarization, for a number of reasons. First, the
alignment of magnetic dipoles in an external magnetic field that we have described is more specif-
ically called paramagnetism. It usually results in a small positive χm > 0, with the aligned dipoles
effectively increasing slightly the overall magnetic field. However, there are also materials, such as
water, with a small but negative χm < 0. This diamagnetism is not due to the alignment of dipoles,
but rather an applied magnetic field can result in a change in electric currents in the medium (at
the atomic scale, by changing electron orbits), which in turn generates a magnetic field in the op-
posite direction. These two effects in general compete, and which is dominant depends on precise
atomic/molecular structure. Finally, ferromagnetic materials, such as iron, become magnetized
under even a small applied magnetic field, and moreover then remain magnetized. Here the align-
ment of (spin) magnetic dipoles in one region influences the alignment in neighbouring regions –
our discussion ignored dipole-dipole interactions, which in ferromagnetic materials are important.
Finally, analogously to Proposition 4.1 we have:

Proposition 4.2 Consider a surface S that is the boundary between two materials with different
permeabilities µ± . Then the components of B normal to S are continuous across S, so

B+ − B− · n = 0 ,


(4.36)
1
while the components of µ B tangent to S are continuous across S, so
 
1 + 1 −
B − −B ·t = 0 . (4.37)
µ+ µ
Here n is the unit normal to S, pointing into the “+ side”, and t is any tangent vector to S.

Proof Following the proof of Proposition 1.8, (4.36) follows from integrating ∇ · B = 0 over the
cylindrical region R, while (4.37) follows from integrating ∇ × ( µ1 B) = 0 over the surface Σ. ■

61
5 Electrodynamics and Maxwell’s equations
5.1 Maxwell’s displacement current

Let’s go back to Ampère’s law (3.29) in magnetostatics

Z Z
B · dr = µ0 J · dS . (5.1)
C=∂Σ Σ

Here C = ∂Σ is a simple closed curve bounding a surface Σ. Of course, one may use any such
surface spanning C on the right hand side. If we pick a different surface Σ′ , with C = ∂Σ′ , then
Z Z
0 = J · dS − J · dS
ZΣ Σ′

= J · dS . (5.2)
S

Here S is the closed surface obtained by gluing Σ and Σ′ together along C. Thus the flux of J
through any closed surface is zero. We may see this in a different way if we assume that S = ∂R
bounds a region R, since then
Z Z
J · dS = ∇ · J dV = 0 , (5.3)
S R

and in the last step we have used the steady current condition (3.7).
But in general, (3.7) should be replaced by the continuity equation (3.6). The above calculation
then changes as follows:
Z Z Z Z Z Z
∂ρ ∂
J · dS − J · dS = J · dS = ∇ · J dV = − dV = −ϵ0 (∇ · E) dV
Σ Σ′ S R ∂t ∂t
Z Z R Z R
∂E ∂E ∂E
= −ϵ0 · dS = −ϵ0 · dS + ϵ0 · dS . (5.4)
S=∂R ∂t Σ ∂t Σ′ ∂t

Here in the first line we have used the divergence theorem in the second equality, the continuity
equation (3.6) in the third equality, and Gauss’ law (1.16) in the final equality. Notice we now
regard E = E(r, t) as a vector field depending on time. In the second line of (5.4) we have then
again used the divergence theorem. We have thus shown that
Z   Z  
∂E ∂E
J + ϵ0 · dS = J + ϵ0 · dS (5.5)
Σ ∂t Σ′ ∂t

for any two surfaces Σ, Σ′ spanning C, and thus suggests replacing Ampère’s law (3.28) by
 
∂E
∇ × B = µ0 J + ϵ0 . (5.6)
∂t

This is indeed the correct time-dependent Maxwell equation on the front page. The additional
term ∂E/∂t is called the displacement current, and the above argument is due to Maxwell. It says
that a time-dependent electric field also produces a magnetic field.

62
5.2 Faraday’s law

The electrostatic equation (1.33) is also modified in the time-dependent case. We can motivate
how precisely by the following argument. Consider the electromagnetic field generated by a set of
charges all moving with constant velocity v. The charges generate both an E and a B field, the
latter since the charges are in motion. However, consider instead an observer who is also moving
at the same constant velocity v. For this observer, the charges are at rest, and thus he/she will
measure only an electric field E′ from the Lorentz force law (3.8) on one of the charges! Indeed,
when we wrote down the Lorentz force law (3.8) and Biot-Savart law (3.10), we didn’t specify what
inertial reference frame we should use to measure the velocities, and this should have worried you
at the time!
Since (or assuming) the two observers above must be measuring the same force on a given
charge, we conclude that

E′ = E + v × B . (5.7)

Now since the field is electrostatic for the moving observer,

0 = ∇ × E′ = ∇ × E + ∇ × (v × B)
= ∇ × E + v (∇ · B) − (v · ∇) B = ∇ × E − (v · ∇) B . (5.8)

Here in the second line we have used the identity (A.11), and in the last step we have used ∇·B = 0.
Now, for the original observer the charges are all moving with velocity v, so the magnetic field at
position r + vτ and time t + τ is the same as that at position r and time t:

B(r + vτ, t + τ ) = B(r, t) . (5.9)

This equation holds for all τ . Dividing through by τ and then taking the limit τ → 0 leads to the
partial differential equation
∂B
(v · ∇) B + = 0. (5.10)
∂t
Substituting this into the right hand side of (5.8) then gives

∂B
∇×E = − . (5.11)
∂t
This is Faraday’s law, and is another of Maxwell’s equations. The above argument raises issues
about what happens in general to our equations when we change to a moving frame. A systematic
study of this leads to Einstein’s theory of Special Relativity, which we shall comment on in section 7.
As usual, the equation (5.11) may be expressed as an integral equation as

Faraday’s law For any simple closed curve C = ∂Σ bounding a fixed surface Σ
Z Z
d
E · dr = − B · dS . (5.12)
C=∂Σ dt Σ

63
This says that a time-dependent magnetic field produces an electric field. For example, if one
moves a bar magnet through a loop of conducting wire C, the resulting electric field from (5.11)
induces a current in the wire via the Lorentz force. This is what Faraday did, in fact, in 1831.
R
Definition The integral Σ B · dS is called the magnetic flux through Σ.

* The current in the wire then itself produces a magnetic field of course, via Ampère’s
law. However, the signs are such that this magnetic field is in the opposite direction to
the change in the magnetic field that created it. This is called Lenz’s law, and a similar
effect is what leads to diamagnetism, mentioned at the end of section 4. The whole setup
may be summarized as follows:
Faraday Lorentz Ampère
changing B −→ E −→ current −→ B. (5.13)

5.3 Maxwell’s equations

We now summarize the full set of Maxwell equations.

There are two scalar equations, namely Gauss’ law (1.16) from electrostatics, and the equation
(3.17) from magnetostatics that expresses the absence of magnetic monopoles:

ρ
∇·E = , (5.14)
ϵ0

∇·B = 0 . (5.15)

Although we discussed these only in the time-independent case, they are in fact true in general.
The are also two vector equations, namely Faraday’s law (5.11) and Maxwell’s modification (5.6)
of Ampère’s law (3.28) from magnetostatics:

∂B
∇×E = − , (5.16)
∂t
 
∂E
∇ × B = µ0 J + ϵ0 . (5.17)
∂t

Together with the Lorentz force law

F = q (E + u × B) , (5.18)

which governs the mechanics, this is all of electromagnetism. Everything else we have discussed
may in fact be derived from these equations.
Maxwell’s equations, for given ρ and J, are 8 equations for 6 unknowns. There must therefore
be two consistency conditions. To see what these are, we first compute
∂
(∇ · B) = −∇ · (∇ × E) = 0 , (5.19)
∂t

64
where we have used (5.16). This is clearly consistent with (5.15). We get something non-trivial by
instead taking the divergence of (5.17), which gives
 
1 ∂
0 = ∇· ∇ × B = ∇ · J + ϵ0 (∇ · E)
µ0 ∂t
∂ρ
= ∇·J + , (5.20)
∂t
where we have used (5.14) in the last step. Thus the continuity equation arises as a consistency
condition for Maxwell’s equations: if ρ and J do not satisfy (5.20), there is no solution to Maxwell’s
equations for this choice of charge density and current. Alternatively, we may regard this as saying
that Maxwell’s equations imply that charge is conserved.

5.4 Electromagnetic potentials and gauge transformations

In the general time-dependent case one can introduce electromagnetic potentials in a similar way
to the static cases. We work in a suitable domain in R3 , such as R3 itself or an open ball therein,
as discussed in previous sections. Since B has zero divergence (5.15), we may again introduce a
vector potential

B = ∇×A , (5.21)

where now A = A(r, t). It follows from Faraday’s law that

 
∂B ∂A
0 = ∇×E + = ∇× E + . (5.22)
∂t ∂t

Thus we may introduce a scalar potential ϕ = ϕ(r, t) via

∂A
E+ = −∇ϕ . (5.23)
∂t
Thus

B = ∇×A , (5.24)

∂A
E = −∇ϕ − . (5.25)
∂t

Note that, by construction, with (5.24) and (5.25) the Maxwell equations (5.15) and (5.16) are
automatically satisfied.

Definition Generalizing the discussion in section 3.7, we define the gauge transformations
∂ψ
A → A
b ≡ A + ∇ψ , ϕ → ϕb ≡ ϕ − , (5.26)
∂t
which leave (5.24) and (5.25) invariant.

65
Again, one may fix this non-uniqueness of A and ϕ by imposing certain gauge choices. Suppose
we have chosen a particular A and ϕ satisfying (5.24), (5.25), and let ψ = ψ(r, t) be a solution to
the following wave equation with source
1 ∂2ψ 1 ∂ϕ
− ∇2 ψ = 2 +∇·A , (5.27)
c2 ∂t2 c ∂t
where we have defined
1
c2 ≡ . (5.28)
ϵ0 µ0
Compare (5.27) to the analogous time-independent Poisson equation (3.32) in magnetostatics.
Then from (5.26) we compute

1 ∂ ϕb 2
+ ∇ · b = 1 ∂ϕ − 1 ∂ ψ + ∇ · A + ∇2 ψ = 0 .
A (5.29)
c2 ∂t c2 ∂t c2 ∂t2
Definition The Lorenz gauge (cf. (3.33)) for A, ϕ is the condition
1 ∂ϕ
+ ∇·A = 0 . (5.30)
c2 ∂t
In Lorenz gauge Gauss’ law (5.14) becomes
1 ∂2ϕ
 
ρ ∂A
= ∇ · E = ∇ · −∇ϕ − = −∇2 ϕ + 2 2 , (5.31)
ϵ0 ∂t c ∂t
while the Ampère-Maxwell equation (5.17) becomes
∂2A
  
2 ∂
∇ × B = ∇ (∇ · A) − ∇ A = µ0 J − ϵ0 ∇ϕ + . (5.32)
∂t ∂t2

In Lorenz gauge ∇ (∇ · A) = − c12 ∂t

∂ ∂
∇ϕ = −ϵ0 µ0 ∂t ∇ϕ, which cancels against the same term on the
right hand side of (5.32), giving
1 ∂2A
−∇2 A + = µ0 J . (5.33)
c2 ∂t2
We may summarize this as follows:

Theorem 5.1 In Lorenz gauge Maxwell’s equations reduce to the wave equations with sources
ρ
□ϕ = − , (5.34)
ϵ0

□ A = −µ0 J . (5.35)

Here we have defined the d’Alembertian operator (some references have the opposite overall sign )
1 ∂2
□ ≡ − + ∇2 . (5.36)
c2 ∂t2
It will turn out that the wave speed c, defined in terms of the permittivity and permeability of free
space via (5.28), is the speed of light in vacuum, as we discuss in detail in section 6. The Green’s
function method for solving these wave equations with sources is discussed in section 5.6.

66
5.5 Electromagnetic energy and Poynting’s theorem

Recall that in section 1.7 we derived a formula for the electrostatic energy density Eelectric =
ϵ0 |E|2 / 2 in terms of the electric field E. The electrostatic energy of a given configuration is
the integral of this density over space (1.53). One can motivate the similar formula Emagnetic =
|B|2 / 2µ0 in magnetostatics, although we won’t elaborate on this here. This leads to the following:

Definition The electromagnetic energy density is

 
1 2 1 2 ϵ0
|E|2 + c2 |B|2 .


E ≡ ϵ0 |E| + |B| = (5.37)
2 µ0 2

We then have

Theorem 5.2 (Poynting’s Theorem ) The electromagnetic energy density satisfies

∂E
+ ∇ · P = −E · J , (5.38)
∂t

where we have defined the Poynting vector P to be

1
P ≡ E×B . (5.39)
µ0

Proof Taking the partial derivative of (5.37) with respect to time we compute
∂E ∂E 1 ∂B
= ϵ0 E · + B·
∂t ∂t µ0 ∂t
1 1
= E · (∇ × B − µ0 J) − B · (∇ × E)
µ0 µ0
 
1
= −∇ · E×B − E·J . (5.40)
µ0

Here after the first step we have used the Maxwell equations (5.17) and (5.16), respectively. The
last step uses the identity (A.12). ■

Notice that, in the absence of a source current, J = 0, (5.38) takes the form of a continuity
equation, analogous to the continuity equation (3.6) that expresses conservation of charge. It is
thus natural to interpret (5.38) as a conservation of energy equation, and so identify the Poynting
vector P as some kind of rate of energy flow density. One can indeed justify this by examining
the above quantities in various physical applications, and we shall look at the particular case of
electromagnetic waves in section 6.
Integrating (5.38) over a region R with boundary Σ, using the divergence theorem we obtain
Z Z Z
d
E dV = − P · dS − E · J dV . (5.41)
dt R Σ R

Given our discussion of E, the left hand side is the rate of increase of energy in R. The first term
on the right hand side is the rate of energy flow into the region R. When J = 0, this is precisely

67
analogous to our discussion of charge conservation in section 3.2. The final term on the right hand
side of (5.41) is interpreted as (minus) the rate of work by the field on the sources. To see this,
remember that the force on a charge q moving at velocity v is F = q (E + v × B). This force does
work at a rate given by F · v = q E · v. Recalling the definition (3.1) of J = ρ v, we see that the
force does work on the charge q = ρ δV in a small volume δV at a rate F · v = E · J δV . The final
term in (5.41) is thus expressing the rate of conversion of electromagnetic energy into mechanical
energy, acting on the sources.

5.6 Time-dependent Green’s function

The form of the time-dependent Maxwell equations (5.34), (5.35) in Lorenz gauge motivates study-
ing the wave equation with source

□ ψ = −4π f (r, t) , (5.42)

for arbitrary source function f (r, t), where recall □ is defined in (5.36). Following section 2.2:

Definition A (time-dependent) Green’s function is a function G(r, t ; r′ , t′ ) satisfying

□ G(r, t ; r′ , t′ ) = −4π δ(t − t′ ) δ(r − r′ ) . (5.43)

Given such a function, a solution to (5.42) is

Z Z ∞
ψ(r, t) = G(r, t ; r′ , t′ ) f (r′ , t′ ) dV ′ dt′ , (5.44)
r′ ∈R3 t′ =−∞

as one sees by applying □ to the right hand side. As we saw (and also exploited) in section 2, Green’s
functions are not unique: we may add to G any solution F = F (r, t ; r′ , t′ ) to the homogeneous wave
equation □ F = 0, and this will be another solution to (5.43). Uniqueness follows after imposing
appropriate boundary conditions, which are ultimately determined by the precise physical setup.
To gain some insight into the time-dependent problem, let us recall the electrostatics Green’s
function equation

∇2 G(r, r′ ) = −4π δ(r − r′ ) . (5.45)

The unique solution to this equation that is zero “at infinity” in R3 is G(r, r′ ) = 1/|r − r′ |.
Physically, this Green’s function is 4πϵ0 times the electrostatic potential generated by a unit
charge at position r′ . Our aim in this subsection is to find the analogous time-dependent solution
to (5.43). Physically, from the Maxwell equation (5.34) this should be 4πϵ0 times the electric
potential ϕ(r, t) generated by a unit charge that appears at position r′ at an instant of time t′ .
Before this localized disturbance happens at time t′ we assume there are no electromagnetic fields
excited, and hence the Green’s function G(r, t ; r′ , t′ ) = 0 for t < t′ . Since □ is a wave operator,
we can guess that the disturbance at position r′ and time t′ should lead to a wave that propagates
spherically outwards from the source point r′ , at speed c.

68
 To work out the details, we make use of the exponential Fourier expansion/Fourier transform of
section 2.4, with complete set of functions (2.56). We already computed the electrostatic Green’s
function G(r, r′ ) = 1/|r − r′ | in this expansion in equation (2.83). Following (2.78), we first write
Z Z ∞
′ ′ 1 ′ ′
δ(t − t ) δ(r − r ) = 4
e−iω(t−t ) eik·(r−r ) d3 k dω , (5.46)
(2π) k∈R3 ω=−∞
which expresses the completeness relation for the functions (2.56). Notice that compared to (2.78)
there is additional integral over an angular frequency variable ω, which leads to the time depen-
dence. We correspondingly write
Z Z ∞
′ ′ 1 ′ ′
G(r, t ; r , t ) = g(k, ω) e−iω(t−t ) eik·(r−r ) d3 k dω , (5.47)
(2π)2 k∈R3 ω=−∞

where we have tacitly assumed that the solution we want depends only on r − r′ (as for the
electrostatic Green’s function 1/|r − r′ |) and t − t′ . The function g(k, ω) is the Fourier transform
2
∂
of G. Applying □ = − c12 ∂t 2
2 + ∇ to (5.47), a similar calculation to (2.80) gives
 2 

−iω(t−t′ ) ik·(r−r′ )
 ω ′ ′
□ e e = − |k| e−iω(t−t ) eik·(r−r ) .
2
(5.48)
c2
Equating □ G with −4π times (5.46), we can read off the Fourier coefficient function
4π 1
g(k, ω) = . (5.49)
(2π)2 |k|2 − ω 2 /c2
Substituting this back into (5.47) gives
Z Z ∞
4π 1 ′ ′
G(r, t ; r′ , t′ ) = 4 2 2 2
e−iω(t−t ) eik·(r−r ) d3 k dω . (5.50)
(2π) k∈R3 ω=−∞ |k| − ω /c
This is analogous to the way we derived (2.83), although there is now an additional integral over ω.
Looking more closely at the latter integral
Z ∞
1 ′
I = 2 2 2
e−iω(t−t ) dω , (5.51)
ω=−∞ |k| − ω /c

we see that there are two simple poles at ω = ±c |k|. Because of this, as written the integral (5.51)
is not actually well-defined, and this is related to the already-mentioned fact that Green’s function
solutions to (5.43) are not unique.
We may define and then evaluate the integral (5.51) more carefully using a complex contour
method. We set
1
ω ≡ Re z , a ≡ −(t − t′ ) , f (z) ≡ , (5.52)
|k|2 − z 2 /c2
and consider the contour integral
Z
I(Γ) ≡ f (z) eiaz dz . (5.53)
Γ

This is to be regarded as a definition of the ill-defined integral in (5.51). The contour Γ should
include the real axis, so as to reproduce the real integral I in (5.51), although as mentioned there
are simple poles at z = ±c |k|: these must be avoided in order for the integral to be well-defined.
We also recall the following:

69
Lemma 5.3 (Jordan’s Lemma ) Let f (z) be a meromorphic function on the complex plane, and
upper
suppose that f (z) → 0 as |z| = R → ∞. Let γR (θ) = R eiθ for θ ∈ [0, π] be the semi-circular
lower (θ) = R eiθ for θ ∈ [π, 2π] be the semi-circular
contour of radius R in the upper half plane, and γR
contour of radius R in the lower half plane. Then
(
limR→∞ γ upper f (z) eiaz dz = 0 ,
R
for a > 0 ,
RR (5.54)
limR→∞ γ lower f (z) eiaz dz = 0 , for a < 0 .
R

-R z = - c |k| O z = c |k| R

Figure 21: The contour Γ. For t < t′ we close in the upper half plane (shown in green/small
dashes), while for t > t′ we close in the lower half plane (shown in blue/longer dashes). In both
cases we indent the contour above the simple poles at z = ±c |k|, so that the upper green contour
contains no singularities, while the lower blue contour contains the two simple poles.

The correct contour Γ for the Green’s function we want is shown in Figure 21. Notice

(i) We choose to indent Γ above both simple poles at z = ±c |k|.

(ii) For a = −(t − t′ ) > 0, or equivalently t < t′ , we close the contour in the upper half plane,
while for a = −(t − t′ ) < 0, or equivalently t > t′ , we close the contour in the lower half plane.

Point (ii) here allows us to apply Jordan’s Lemma 5.3 to conclude that the semi-circular part of
the contour does not contribute to the integral I(Γ) in the limit R → ∞. The choice of indentation

70
in point (i) then means that the upper contour for t < t′ contains no singularities, and the Residue
Theorem immediately gives I(Γ) = 0. With this definition of the ω integral in (5.51), and hence
(5.50), we deduce that

G(r, t ; r′ , t′ ) = 0 for t < t′ . (5.55)

This was a physical requirement we mentioned earlier: before the localized disturbance at time t′
we assume there are no electromagnetic fields excited.

Definition Green’s functions satisfying (5.55) are called retarded Green’s functions.

Remark If we had chosen to indent below both simple poles, we would instead obtain a Green’s
function which is zero for t > t′ (called an advanced Green’s function, which is also useful).

Using the lower contour in Figure 21, for t > t′ the Residue Theorem instead gives
" ′
! ′
!#
e−i(t−t )z e−i(t−t )z
I(Γ) = −2πi Residue , z = c |k| + Residue , z = −c |k|
|k|2 − z 2 /c2 |k|2 − z 2 /c2
" #
e−ic |k|(t−t′ ) e ic |k|(t−t′ )
= 2πi c2 −
2c |k| 2c |k|
c
sin c |k|(t − t′ ) .
 
= 2π (5.56)
|k|
Here in the first line note that the overall minus sign is because the lower blue contour Γ runs
clockwise in the plane, so that −Γ is positively oriented. Inserting (5.56) for the integral I back
into Green’s function (5.50) hence gives
Z  
′ ′ 4π c  ′
 ik·(r−r′ ) 3
G(r, t ; r , t ) = 2π sin c |k|(t − t ) e d k
(2π)4 k∈R3 |k|
Z ∞
 π
Z Z 2π
c 1 2 ′ ′
eik|r−r | cos θ sin θ dk dθ dφ .

= · k sin ck(t − t ) (5.57)
2π 2 k=0 k θ=0 φ=0

Here in the second line we have introduced spherical polar coordinates for k ∈ R3 , with k ≡ |k|,
θ being the angle between the vectors k and r − r′ , and d3 k = k 2 sin θ dk dθ dφ. Integrating first
over the angular variables θ, φ immediately gives
Z ∞  π
′ ′ c  ′
 1 ik|r−r′ | cos θ
G(r, t ; r , t ) = k sin ck(t − t ) 2π − e dk
2π 2 k=0 ik|r − r′ | 0
Z ∞
c 1
sin ck(t − t′ ) 2 sin k|r − r′ | dk
 

= ′
π |r − r | k=0
Z ∞
c 1  ′
 ′


= sin ck(t − t ) sin k|r − r | dk
π |r − r′ | k=−∞
Z ∞ 
c 1 ick(t−t′ ) −ick(t−t′ )

ik|r−r′ | −ik|r−r′ |

= − e − e e − e dk
4π |r − r′ | k=−∞
Z ∞ h
c 1 ick(t−t′ +|r−r′ |/c) ick(t−t′ −|r−r′ |/c)
i
= − e −e dk . (5.58)
2π |r − r′ | k=−∞

71
Here in the third line we have used the fact that the integrand is an even function of k, and in
the fourth line we have written sine in terms of exponentials. In the final step the four terms
one obtains in multiplying out the brackets in the penultimate line are seen to be pairwise equal
on replacing k →7 −k. Finally, recall that the completeness relation (2.60) allows us to identify
∞ ′
δ(x − x′ ) = 2π k=−∞ eik(x−x ) dk. Changing the integration variable k ≡ k̃/c in the last line of
1
R

(5.58), and using the completeness relation to integrate over k̃, gives

|r − r′ | |r − r′ |
    
′ ′ 1 ′ ′
G(r, t ; r , t ) = −δ t − t + +δ t−t − . (5.59)
|r − r′ | c c

The first Dirac delta function term is zero, since |r − r′ |/c ≥ 0 and notice here we are assuming
t > t′ , so that the argument is strictly positive. We have thus proven:

Proposition 5.4 The retarded Green’s function is

|r − r′ |
 
′ ′ 1 ′
G(r, t ; r , t ) = δ t−t − . (5.60)
|r − r′ | c

Notice here that when t < t′ the argument of the Dirac delta function is strictly negative, so that
the Green’s function is indeed zero. In fact G(r, t ; r′ , t′ ) = 0 unless c(t − t′ ) = |r − r′ | ≥ 0, which
is the equation for a sphere of radius c(t − t′ ), centred on the point r′ . Indeed, t − t′ = |r − r′ |/c is
the time taken for a disturbance at the point r′ to reach the point r, travelling radially outwards
with constant speed c. This solution hence has all the properties we were looking for: it gives a
wave that originates from a localized electromagnetic disturbance at position r′ and time t′ , that
expands spherically with speed c from this source.
Having obtained the correct Green’s function, using (5.44) we can now simply write down the
solution to the original wave equation with source (5.42):
Z Z ∞
ψ(r, t) = G(r, t ; r′ , t′ ) f (r′ , t′ ) dV ′ dt′ ,
′
r ∈R 3 ′
t =−∞
Z ∞
|r − r′ |
Z  
1 ′
= ′|
δ t−t − f (r′ , t′ ) dV ′ dt′ ,
′
r ∈R 3 ′
t =−∞ |r − r c
|r − r′ |
Z  
1 ′
= ′
f r ,t − dV ′ , (5.61)
r′ ∈R3 |r − r | c

where in the last step we have integrated over t′ . We may hence also write down the electromagnetic
potentials that solve the general time-dependent Maxwell equations (5.34), (5.35):

|r−r′ |
 
1
Z ρ r′ , t − c
ϕ(r, t) = ′
dV ′ , (5.62)
4πϵ0 r′ ∈R3 |r − r|
|r−r′ |
 
µ0
Z J r′ , t − c
A(r, t) = ′
dV ′ , (5.63)
4π r′ ∈R3 |r − r|

Remark One can verify that these potentials satisfy the Lorenz gauge condition (5.30).

72
 Compare (5.62), (5.63) to the static case equations in (3.36). The only difference is the time
dependence, where notice that the fields at time t are determined by integrating the charge density
ρ and current density J at the retarded time
|r − r′ |
t′ = tretarded ≡ t − . (5.64)
c
As already remarked, |r − r′ |/c is the time it takes for the Green’s function wave to propagate
from the source point r′ to the observation point r. We shall see in section 6 that c is the speed
of light in vacuum, and formulas (5.62), (5.63) say there is no instantaneous action at a distance
in electromagnetism, but rather there is a delay between a change in the sources and the resulting
change in the fields at time t, given by (5.64). We now have the tools to address the following:

5.6.1 Moving point charge

Consider a point charge q moving on an arbitrary trajectory r = r0 (t), with velocity v = dr0 (t)/dt.
If the charge were stationary, we could use Coulomb’s law (1.7) to deduce the electric field E
produced. But Coulomb’s law led to the electrostatic Maxwell equations, while in a general time-
dependent setting we have Faraday’s law ∇ × E = −∂B/∂t. Indeed, because the charge is moving
we expect it to produce a magnetic field B via the Biot-Savart law (3.10). But the latter led to
Ampère’s law, which in general has the Maxwell displacement current ∂E/∂t on the right hand
side of (5.6). In fact neither Coulomb’s law nor the Biot-Savart law are correct in this setting,
because of the complicated way that the fields feed back into each other in Maxwell’s equations.
But we have solved the general time-dependent equations via (5.62), (5.63), and so may use these
to solve this problem.
The charge density and current density of the point charge are by definition
dr0 (t)
ρ(r, t) = q δ[r − r0 (t)] , J(r, t) = q v(t) δ[r − r0 (t)] , where v(t) = . (5.65)
dt
From (5.62) we thus compute
Z Z ∞
q 1 ′ ′
′ ′
dV ′ dt′


ϕ(r, t) = δ[r − r 0 (t )] δ t − t − |r − r |/c
4πϵ0 r′ ∈R3 t′ =−∞ |r − r′ |
Z ∞
q 1
= δ[t′ − (t − |r − r0 (t′ )|/c)] dt′ . (5.66)
4πϵ0 t′ =−∞ |r − r0 (t′ )|

Here in the first line we have imposed the retarded time condition (5.64) by (re)inserting a Dirac
delta function and integral over t′ , and in the second line we have integrated over r′ . Notice that
the Dirac delta function on the second line is a non-trivial function of t′ , for which we shall need
property (iii) in equation (1.18) of Proposition 1.2.
We next define

R(t) ≡ r − r0 (t) , R(t) ≡ |R(t)| , (5.67)

73
where R(t) is the position vector of the observation point r, relative to the position vector r0 (t)
of the point charge at time t, and compute
dR d p 1 dr0 (t) R
= (r − r0 (t)) · (r − r0 (t)) = − (r − r0 (t)) · = − ·v . (5.68)
dt dt R dt R
The Dirac delta function on the second line of (5.66) sets t′ = t − R(t′ )/c, where R(t′ ) is a known
function of t′ , as the trajectory r0 (t′ ) is given, and using property (iii) in Proposition 1.2 and (5.68)
we obtain
q 1 1
ϕ(r, t) = . (5.69)
4πϵ0 R(t′ ) 1 − 1 R(t′ )
c R(t′ ) · v(t′ )
such that t′ = t−R(t′ )/c
t′
We can see the usual Coulomb field q/(4πϵ0 R) generated by a point charge q, but there is an
extra term multiplying this, together with the complication of determining the retarded time t′ ,
which is determined implicitly from t′ = t − R(t′ )/c. Notice that R(t′ )/R(t′ ) is a unit vector, while
|v(t′ )/c| ≪ 1 for speeds small compared to c, so as long as the particle is moving slowly compared
to c, the multiplicative correction term to the Coulomb potential in (5.69) is approximately 1.
One can write down a similar general formula for A using (5.63): Since J = vρ, we just get an
extra factor of v(t′ ), concretely
ϕ(r, t) ′ qµ0 v(t′ )
A(r, t) = v(t ) = . (5.70)
c2 4π R(t′ ) − R(t′ ) · v(t′ )/c t′ such that t′ = t−R(t′ )/c

The equations (5.69) and (5.70) are known as Liénard–Wiechert potentials.

Let us now specialize to the case where the particle moves with constant velocity v = v e1
along the x-axis direction, starting at the origin at time t = 0, so that r0 (t) = vt e1 . We
shall study this example again at the end of section 7, from a different point of view. We have
R(t) = (x − vt) e1 + y e2 + z e3 and R(t) = (x − vt)2 + y 2 + z 2 , and the retarded time t′ solves
p

c2 (t − t′ )2 = R(t′ )2 = (x − vt′ )2 + y 2 + z 2
v 2 ′2 x2 + y 2 + z 2
   v 
⇒ 1 − 2 t − 2t′ t − 2 x + t2 − = 0. (5.71)
c c c2
Introducing
1
γ = γ(v) ≡ p , (5.72)
1 − v 2 /c2
the quadratic equation (5.71) in t′ has solution
 
v 1p 2
t′ = γ 2 t − 2 x − γ (x − vt)2 + y 2 + z 2 . (5.73)
c γc
Here we have chosen the root with t − t′ > 0. Those who have studied Special Relativity will
recognize the Lorentz factor γ making a remarkable appearance, as well as other features of Lorentz
transformations! Examining the potential ϕ(r, t) in (5.69), some algebra gives
1 v v
R(t′ ) − R(t′ ) · v(t′ ) = (x − vt′ ) + y 2 + z 2 − (x − vt′ ) = c(t − t′ ) − (x − vt′ )
p
c   c c
1 ′ v 1p 2
= c t − 2t − 2x = γ (x − vt)2 + y 2 + z 2 , (5.74)
γ c γ

74
where the last step uses (5.73). The potential ϕ(r, t) in (5.69) thus simplifies to
qγ 1
ϕ(r, t) = p . (5.75)
4πϵ0 γ (x − vt)2 + y 2 + z 2
2

This is the Coulomb potential one might naively guess is generated by a point charge q at position
r0 (t) = vt e1 , up to the relativistic factors of γ. The vector potential in this case is given by the
similar formula
qγ 1 ϕ(r, t)
A(r, t) = 2
v = v, (5.76)
c2
p
4πϵ0 c 2 2 2
γ (x − vt) + y + z 2

where from (5.28) note 1/(ϵ0 c2 ) = µ0 .

5.6.2 Radiation from an accelerated charge

To avoid confusion with derivatives, let tr (previously t′ ) denote the solution to the implicit equation
c(t − tr ) = R where R = |R| and R = r − r0 (tr ). With the notations β = v0 (tr )/c and n =
R(tr )/R(tr ), we can write the Liénard–Wiechert potentials of a moving point charge as
q 1 q β
ϕ(r, t) = and A(r, t) = .
4πϵ0 R − R · β 4πϵ0 c R − R · β
To compute the fields E and B, we need to know the implicit dependence of tr = tr (r, t) on r and
t. Therefore, consider the defining relation t = tr + |r − r0 (tr )|/c:

• Taking the tr -derivative, we find

∂t R ∂  
=1+ · r − r0 (tr ) = 1 − n · β.
∂tr cR ∂tr
| {z }
−cβ

• Taking the gradient, we find (using the chain rule)

1 n
0 = ∇t = ∇tr + (R · ∇) (r − r0 (tr )) = + (∇tr ) (1 − n · β) .
cR | {z } c
R−(∇tr )R·∂r0 (tr )/∂tr

In summary, we have obtained that

∂tr 1 n/c
= and ∇tr = − . (5.77)
∂t 1−n·β 1−n·β
Using these, we can compute the electromagnetic fields. For −∇ϕ, we need
1 1
−∇ = 2 ∇ (R − (r − r0 (tr )) · β)
R−R·β R (1 − n · β)2 | {z }
−β+(∇tr ) ∂t∂ (c(t−tr )−(r−r0 (tr ))·β)
r

β n
1 − β 2 + R · β ′ /c


=− 2 2
+ 2 3
(5.78)
R (1 − n · β) R (1 − n · β)

75
where we set β ′ = v0′ (tr )/c to the accelartion of the charge, at the retarded time, divided by c.
We also write β = |β| = |v0 (tr )|/c. To compute ∂A/∂t, we obtain
β′
 
∂ β ∂tr β ∂
= − 2
(c(t − tr ) − (r − r0 (tr )) · β)
∂t R − R · β ∂t R − R · β (R − R · β) ∂t
β′ βc
= 2
−
R(1 − n · β) (R − R · β)2
 
β ∂tr ∂
− 2
(c(t − tr ) − (r − r0 (tr )) · β) . (5.79)
(R − R · β) ∂t ∂tr
| {z }
−c(1−β 2 )−R·β ′

Combining the two calculations above, we obtain for E = −∇ϕ − ∂A/∂t the result
β′
 
q n−β 2 ′


E(r, t) = 1 − β + R · β /c − . (5.80)
4πϵ0 R2 (1 − n · β)3 cR(1 − n · β)2
Similarly we can compute the magnetic flux density B. The result can be stated as
1
B(r, t) = n × E. (5.81)
c
Note that B is orthogonal to E and n. Organizing the expression for E according to the R-scaling,
we find two contributions:
1 q(n − β)(1 − β 2 ) 1 q
(n − β)(n · β ′ ) − β ′ (1 − n · β) .


E= 2 3
+ 3
(5.82)
R 4πϵ0 (1 − n · β) R 4πϵ0 c(1 − n · β) | {z }
n×((n−β)×β ′ )

The first contribution ∝ 1/R2 is Coulomb-like, but the second contribution ∝ 1/R falls off only
linearly and thus has longer range. Note that this long-range component is only present when the
charge accelerates (β ′ ̸= 0), hence we do not see it for a charge with constant velocity.
We conclude: An accelerated electric charge emits electromagnetic radiation! This is the princi-
ple behind antennas, synchrotrons, and free electron lasers.
For the Poynting vector P = E × B/µ0 we find
n×E |E|2 n − E(E · n)
P =E× =
cµ0 cµ0
|n × (n − β) × β ′ |2


q 2
= n + O(1/R3 ). (5.83)
16π 2 ϵ0 cR2 (1 − n · β)6
This energy flux density is directed along n, that is, radiation is being emitted by the moving
charge. Integrating P over a sphere, once finds for the total emitted power
q2 4
 Z
dW ∂t
γ |β ′ |2 + γ 2 (β · β ′ )2 ,


= P · dS = (5.84)
dtr ∂tr 6πϵ0 c
p
where γ = 1/ 1 − β 2 . If one applies this formula to the model of a hydrogen atom, with the
electron moving in a circular orbit (note β · β ′ = 0)
     
cos ωt − sin ωt 2 − cos ωt
r ω
0  r0 ω
r0 (t) = r0  sin ωt  , β(t) = cos ωt  , β ′ (t) =  − sin ωt  ,
c c
0 0 0

76
one finds that the entire binding energy of an electron in the ground state would be radiated away
in ≈ 10−11 seconds. This contradicts the existence of hydrogen and suggests that the classical
mechanical model of the atom is too simplistic. This puzzle is resolved in quantum mechanics.

5.7 Maxwell’s equations in macroscopic media

In section 4 we derived the electrostatic Maxwell equations (4.12) and magnetostatic Maxwell
equations (4.35) in macroscopic media. One can generalize this discussion to include time depen-
dence, in essentially the same way as we have done for the vacuum Maxwell equations in this
section. Rather than repeat the arguments, we here simply present the the final complete set of
macroscopic Maxwell equations (or Maxwell equations in matter ):

∂B
∇ · (ϵ E) = ρfree , ∇×E=− ,
∂t
  (5.85)
1 ∂(ϵ E)
∇ · B = 0, ∇× B = Jfree + .
µ ∂t

These are similar in form to the equations originally introduced and studied by Maxwell. Al-
though we have motivated them as an effective set of equations, approximately valid in certain
media where ϵ, µ characterize the media’s electromagnetic properties, notice that setting ϵ = ϵ0 ,
µ = µ0 gives the microscopic Maxwell equations on the front page, which are regarded as funda-
mental, and exact.

77
6 Electromagnetic waves
6.1 Source-free equations and electromagnetic waves

We begin by writing down Maxwell’s equations in vacuum, with no electric charge or current:

∇·E = 0 , ∇·B = 0 , (6.1)

∂B 1 ∂E
∇×E + = 0, ∇×B − 2 = 0, (6.2)
∂t c ∂t
where as in (5.28) we have defined
r
1
c ≡ . (6.3)
ϵ0 µ0
We have already seen that c is a speed ; for example, the speed of the spherical wavefront that
propagates out from the source point in the Green’s function (5.60). The great insight of Maxwell
was to realise that this is the speed of light in vacuum.
Taking the curl of the first equation in (6.2) we have
∂B ∂
0 = ∇ (∇ · E) − ∇2 E + ∇ × = −∇2 E + (∇ × B)
∂t ∂t
1 ∂2E
= −∇2 E + . (6.4)
c2 ∂t2
Here after the first step we have used the first equation in (6.1), and in the last step we have used
the second equation in (6.2). It follows that each component of E satisfies the wave equation

□u = 0 , (6.5)

1 ∂2
where u = u(r, t), and as in (5.36) the d’Alembertian operator is □ ≡ c2 ∂t2
−∇2 . You can similarly
check that B also satisfies □ B = 0.
The equation (6.5) governs the propagation of waves of speed c in three-dimensional space. It
is the natural generalization of the one-dimensional wave equation
1 ∂2u ∂2u
− = 0, (6.6)
c2 ∂t2 ∂x2
which you met in Prelims. Recall that this has particular solutions of the form u± (x, t) = f (x∓ct),
where f is any function which is twice differentiable. In this case, the waves look like the graph
of f travelling at constant speed c in the direction of increasing/decreasing x, respectively. The
general (d’Alembert) solution to (6.6) is u(x, t) = f (x − ct) + g(x + ct), as shown in the Prelims
course on Fourier Series and PDEs .
The above generalizes naturally to the three-dimensional equation (6.5), by writing

u(r, t) = f (e · r − ct) , (6.7)

where e is a fixed unit vector, |e|2 = 1. Indeed, using the chain rule we compute ∇2 u = e · e f ′′ ,
∂ 2 u/∂t2 = c2 f ′′ , so that (6.7) solves (6.5) for any twice differentiable function f of one variable.

78
Definition Solutions to the wave equation (6.5) of the form (6.7) are called plane-fronted waves.

The terminology here is justified by noting that at any constant time, u is constant on the planes
{e · r = constant} orthogonal to e. As time t increases, these plane wavefronts propagate in the
direction of e at speed c. However, unlike the one-dimensional equation, we cannot write the
general solution to (6.5) as a sum of two plane-fronted waves travelling in opposite directions.

6.2 Monochromatic plane waves

An important special class of plane-fronted waves (6.7) are given by the complex harmonic waves

u(r, t) = α ei(k·r−ωt) , (6.8)

where α is a complex constant, ω > 0 is the constant frequency of the wave, and k is the constant
wave vector. To relate to (6.7), note that
ω
k · r − ωt = (e · r − ct) , (6.9)
c
provided we identify
ω
k = ke , where k = . (6.10)
c
Here k ≡ |k| is called the wave number. Thus (6.8) solves the wave equation (6.5) provided (6.10)
holds, which is simply the equation
ω ω 2π
speed c = = · = frequency × wavelength . (6.11)
k 2π k
The harmonic waves (6.8) are of course complex, although notice that the real and imaginary
parts separately solve the wave equation. One can then take linear combinations of these real sine
and cosine solutions. Note also that the complex harmonic wave (6.8) is simply the product of
exponential Fourier modes (2.56) in each variable x, y, z, t. In fact it is a result of Fourier analysis
that every solution to the wave equation (6.5) is a linear combination (in general involving an
integral) of these harmonic waves, as (6.8) form a complete set of orthonormal functions, (5.46).
Since the components of E and B satisfy (6.5), it is natural to look for solutions of the complex
harmonic wave form

EC (r, t) ≡ E0 ei(k·r−ωt) , BC (r, t) ≡ B0 ei(k·r−ωt) , (6.12)

where E0 and B0 are constant complex vectors. Here and in the following we understand these
expressions to mean that we take the real part of the complex exponential to obtain the real
electromagnetic field, so E ≡ Re (EC ), B ≡ Re (BC ). The expressions (6.12) of course satisfy the
wave equation, but we must ensure that we satisfy all of the Maxwell equations in vacuum (6.1),
(6.2). Since ∇ · EC = E0 · ∇ ei(k·r−ωt) = i k · E0 ei(k·r−ωt) , the two equations in (6.1) immediately
give

k · E0 = 0 = k · B0 . (6.13)

79
The first equation in (6.2) reads
∂BC
0 = ∇ × EC + = (ik × E0 − iω B0 ) ei(k·r−ωt) , (6.14)
∂t
allowing us to read off
1 1
B0 = k × E0 = e × E 0 , (6.15)
ω c
where in the second equality we have used (6.10). One can then verify that the second equation
2
in (6.2) is automatically satisfied, using k × ω1 k × E0 + cω2 E0 = ω1 (k · E0 ) k − kω E0 + cω2 E0 = 0,

where the first equality uses the vector triple product (A.6), and in the second we have used (6.13)
and (6.10). To summarize, we have shown

Proposition 6.1 The monochromatic electromagnetic plane wave, given by

1
EC (r, t) = E0 ei(k·r−ωt) , BC (r, t) = k × EC (r, t) , (6.16)
ω
solves the vacuum Maxwell equations (6.1), (6.2), provided |k| = ω/c, and k · E0 = 0.

Notice that the solution is specified by the angular frequency ω, the direction of propagation
e ≡ k/|k|, and with the electric field direction specified by a constant vector E0 that is orthogonal
to the direction of propagation. In fact both E and B are orthogonal to the direction of propagation
(6.13), which is known as a transverse wave. One can contrast this with, e.g. sound waves, which
are longitudinal waves, with the wave motion aligned with the direction of propagation. E and B
are also orthogonal to each other. Fourier analysis implies that the general vacuum solution is a
combination of these monochromatic plane waves.

6.3 Polarization

So far we have written complex solutions to the vacuum Maxwell equations, but as mentioned the
actual electric and magnetic fields are given by the real (or imaginary) parts of the monochromatic
plane waves in (6.16). Focusing on the electric field, we thus have

E(r, t) = Re (EC (r, t)) = α cos(k · r − ωt) − β sin(k · r − ωt) , (6.17)

where we have defined α ≡ Re E0 , β ≡ Im E0 , which are real vectors, orthogonal to the direction
of propagation k. The magnetic field is
1 1 1
B(r, t) = k × E(r, t) = k × α cos(k · r − ωt) − k × β sin(k · r − ωt) . (6.18)
ω ω ω
If we fix a particular point in space, say the origin r = 0, then the electric field (6.17) is

E(0, t) = α cos ωt + β sin ωt . (6.19)

As t varies, this sweeps out an ellipse in the plane spanned by α and β (similar remarks apply
to the B-field in (6.18)). As a simple example, taking α = α e1 , β = β e2 where e1 , e2 are

80
orthonormal vectors, so that e ≡ k/|k| = e3 is the direction of propagation, then the 1 and 2
components of E in (6.19) are E1 = α cos ωt, E2 = β sin ωt, and thus

E12 E22
+ = 1. (6.20)
α2 β2
This is an ellipse, with semi-major(minor) axis length α, semi-minor(major) axis length β, centred
on the origin.

E
cB

cB
ωt
E

(a) Linear polarization. (b) Circular polarization (right-handed).

Figure 22: Polarizations of monochromatic electromagnetic plane waves, viewed from the direction
of propagation.

There are two special choices of α and β, which have names:

(i) Linear polarization: If α is proportional to β, so that the ellipse degenerates to a line,

then the monochromatic plane wave is said to be linearly polarized. In this case, E and B
oscillate in two fixed orthogonal directions – see Figure 22a. For example, taking β = 0 the
electric and magnetic fields at the origin are
1
E = α e1 cos ωt , B = α e2 cos ωt . (6.21)
c

(ii) Circular polarization: If α · β = 0 (as in the example (6.20) above) and also |α| = |β|,
so that the ellipse is a circle, then the monochromatic plane wave is said to be circularly
polarized. In this case, E and B rotate at constant angular velocity about the direction of
propagation – see Figure 22b. A circularly polarized wave is said to be right-handed or left-
handed, depending on whether |α|2 e = α × β or |α|2 e = −α × β, respectively. With your
thumb aligned with the direction of propagation e, the direction of the electric and magnetic
fields is given by the curl of your fingers on your right and left hands, respectively.

In general, electromagnetic waves are combinations of waves with different polarizations.

81
 Finally, notice the formula (6.15) gives
1
B = e×E =⇒ c|B| = |E| , (6.22)
c
and thus the energy density (5.37) of a monochromatic plane wave is
ϵ0
|E|2 + c2 |B|2 = ϵ0 |E|2 .


E ≡ (6.23)
2
The Poynting vector (5.39) is
 
1 1 1 1
P ≡ E×B = E× e×E = |E|2 e = c E e , (6.24)
µ0 µ0 c µ0 c

which is in the direction of propagation of the wave, with magnitude |P| = c E. Thus electromag-
netic waves carry energy, a fact which anyone who has made a mircowave pot noodle can confirm.

6.4 Reflection and refraction

Maxwell postulated that the electromagnetic waves we have been discussing describe light. If
that’s the case, then all observed (classical) properties of light must be a consequence of Maxwell’s
equations. In this section we prove the laws of optics. Here a ray of light that hits the boundary
between two transparent materials, such as air and water, is divided into a reflected ray and a
refracted ray. These obey (see Figure 23):

(i) Law of reflection: the reflected ray lies in the plane of incidence, with the angle of incidence
θ equal to the angle of reflection θ′′ .

(ii) Law of refraction: the refracted ray lies in the plane of incidence, with the angle of incidence
θ and angle of refraction θ′ related by Snell’s law

n sin θ = n′ sin θ′ , (6.25)

where the materials have a refractive index n, n′ , respectively.

For example, nair ≃ 1, while nwater ≃ 1.3. In fact we will show that
r
ϵµ
n = , (6.26)
ϵ0 µ0

where ϵ, µ are the permittivity and permeability of the medium. Snell’s law (6.25) rearranges to
n
sin θ′ = sin θ , (6.27)
n′
so that when moving from a material with a lower n to a higher n′ (such as from air to water), the
angle of refraction θ′ is smaller than the angle of incidence θ, causing the light to bend towards the
normal direction to the boundary (as in Figure 23). On the other hand, moving from a material
with a higher n to a lower n′ (such as from water to air), there is a critical angle of incidence

82
 z
incident reflected

k k''
θ θ''
ε ,μ x

ε' , μ'

θ' k'

refracted

Figure 23: A light ray striking the boundary between two materials. The incident light ray,
refracted light ray and reflected light ray travel in the directions k, k′ , k′′ , respectively, with angle
of incidence θ, angle of refraction θ′ , and angle of reflection θ′′ .

θc ≡ arcsin (n′ /n), where for θ > θc the right hand side of equation (6.27) is larger than 1, and
hence there is no solution for θ′ . There is then no refracted light ray, a phenomenon called total
internal reflection.
To derive these laws from Maxwell’s equations, we model the incident light ray as a monochro-
matic plane wave, travelling in the direction k, with electric field

Eincident = E0 ei(k·r−ωt) . (6.28)

1
The corresponding magnetic field is B = ω k × E, where it is convenient to work with complex
plane waves and drop the subscript C from EC and BC . Without loss of generality, we take the
plane that divides the two materials to be the (x, y)-plane {z = 0}, and take
k
e ≡ = ( sin θ , 0 , − cos θ ) , (6.29)
|k|
so that the plane of incidence, spanned by the normal e3 to {z = 0} and incident direction k, is
the (x, z)-plane {y = 0}, as in Figure 23. Notice then that the electric field (6.28) is independent
of the y coordinate. Necessarily k · E0 = 0, and we take

E0 = E0 e2 , (6.30)

so that Eincident points into the page in Figure 23.

Similarly including the reflected and refracted waves, we may write
( ′′ ′′
Eincident + Ereflected = E0 ei(k·r−ωt) + E′′0 ei(k ·r−ω t) , z > 0,
E = ′ ′ (6.31)
Erefracted = E′0 ei(k ·r−ω t) , z < 0.

83
The Maxwell equations in a general macroscopic medium are (5.85), and simply involve replacing
ϵ0 → ϵ, µ0 → µ. It follows from (6.3) that in a general macroscopic medium the wave speed
propagation is not the speed of light c in vacuum, but rather
r r r r
1 ϵ0 µ0 ′ 1 ϵ0 µ0
v = = c, v = ′ ′
= c. (6.32)
ϵµ ϵµ ϵµ ϵ′ µ′
Thus (6.10) becomes
ω ω ′′ ω′
|k| = , |k′′ | = , |k′ | = . (6.33)
v v v′
Given the incident ray, it remains to then determine the directions of k′ , k′′ , which will give the
laws of refraction and reflection, and also ω ′ , ω ′′ , E′0 , E′′0 . This comes from analysing the boundary
conditions at {z = 0}.
Proposition 4.1 and Proposition 4.2 say17 :
1
(i) The components of E and µB tangent to {z = 0} are continuous across this surface.

(ii) The components of ϵ E and B normal to {z = 0} are continuous across this surface.

Focusing first on the electric field, the two tangent directions are e1 and e2 , so we may write down

(Eincident + Ereflected ) · ei = Erefracted · ei , i = 1, 2 ,

z=0 z=0

ϵ (Eincident + Ereflected ) · e3 = ϵ′ Erefracted · e3 . (6.34)

z=0 z=0

More explicitly, these read

′′ ·(x,y,0)−ω ′′ t) ′ ′
(E′′0 · e1 ) ei(k = (E′0 · e1 ) ei(k ·(x,y,0)−ω t) ,
′′ ·(x,y,0)−ω ′′ t) ′ ′
E0 ei(|k| x sin θ−ωt) + (E′′0 · e2 ) ei(k = (E′0 · e2 ) ei(k ·(x,y,0)−ω t) ,
′′ ·(x,y,0)−ω ′′ t) ′ ′
ϵ (E′′0 · e3 ) ei(k = ϵ′ (E′0 · e3 ) ei(k ·(x,y,0)−ω t) , (6.35)

where we have used the form of E0 and k in (6.30) and (6.29), respectively. These equations hold for
all x, y and t, which are clearly quite strong conditions! In particular, notice that the exponential
Fourier modes eikx are linearly independent functions of x, for different k. So for example the
′
functional equation α eikx + γ = β eik x , with coefficients α, β, γ ̸= 0, implies k = k ′ = 0, α + γ = β.
Unpacking (6.35) is a little fiddly. We look first at the middle equation. Notice we cannot have
both (E′′0 · e2 ) and (E′0 · e2 ) equal to zero, otherwise E0 = 0 and there is no incident ray. Suppose
these coefficients are both non-zero.18 Then from the remark above about linear independence,
looking at the t dependence of the middle equation in (6.35) immediately gives

ω = ω ′ = ω ′′ , (6.36)
17
The alert reader will notice these were derived for statics. However, one can verify that the additional time-
dependent terms in the surface integral over the rectangular surface Σ do not contribute on taking ε → 0.
18
If instead say (E′′0 · e2 ) ̸= 0 but (E′0 · e2 ) = 0, then if there is a refracted ray one of (E′0 · e1 ) or (E′0 · e3 ) must be
non-zero. Then just add a multiple of the first or last equation in (6.35) to the middle equation, respectively, and
the reasoning below proceeds in the same way.

84
so that reflected and refracted frequencies are the same as the incident frequency. Since the term
proportional to E0 is independent of y, and non-zero, looking at the y dependence implies that
the other two terms in the middle equation of (6.35) are also independent of y. That is,

e2 · k′ = e2 · k′′ = 0 . (6.37)

But this says that the reflected and refracted rays lie in the plane of incidence, namely the (x, z)-
plane. As in Figure 23, we may hence write
k′ k′′
= ( sin θ′ , 0 , − cos θ′ ) , = ( sin θ′′ , 0 , cos θ′′ ) . (6.38)
|k′ | |k′′ |

The x dependence in the middle equation of (6.35) then gives

|k| sin θ = |k′ | sin θ′ = |k′′ | sin θ′′ . (6.39)

But substituting (6.36) into (6.33), this implies |k| = |k′′ | and v|k| = v ′ |k′ | and hence
1 1
θ = θ′′ , sin θ = ′ sin θ′ . (6.40)
v v
We have thus proven the law of reflection, and the law of refraction (6.25), with the dimensionless
refractive index given by (6.26), which using (6.32) may also be written as n = c/v.
To finish solving the problem we must also impose the boundary conditions on the B-field at
{z = 0}, although in the following we just outline the steps. These boundary conditions read
1 1
k × E0 + k′′ × E′′0 · ei = ′ k′ × E′0 · ei ,



i = 1, 2 ,
µ µ
k × E0 + k′′ × E′′0 · e3 = k′ × E′0 · e3 .



(6.41)

One can verify that (6.35), (6.41) are solved by taking

E′0 = E0′ e2 , E′′0 = E0′′ e2 , (6.42)

so that the incident, reflected and refracted electromagnetic fields all have the same polarization
(linear, in the y-axis direction), where equation (6.35) then imposes only

E0 + E0′′ = E0′ , (6.43)

while after a little work one checks that (6.41) imposes only
s
ϵ′ ′
r
ϵ
E0 − E0′′ cos θ = E cos θ′ ,


(6.44)
µ µ′ 0

where we have used (6.32), (6.33). Assuming µ = µ′ , which is approximately the case for air and
water, one can solve (6.43), (6.44) using (6.40) to find

2 cos θ sin θ′ sin(θ − θ′ )

E0′ = E0 , E0′′ = − E0 . (6.45)
sin(θ + θ′ ) sin(θ + θ′ )

85
We have thus determined completely the reflected and refracted waves, in terms of the incident
wave, in the case when the wave is polarized orthogonal to the incidence plane.
For polarization parallel to the incidence plane,

cos θ′
     
cos θ − cos θ
E0 = E0  0  E′0 = E0′  0  E′′0 = E0′′  0 
sin θ sin θ′ sin θ
      (6.46)
0 0 0
kE0   k ′ E0′   kE0′′  
B0 = −1 B′0 = −1 B′′0 = −1
ω ω ω
0 0 0

solving the boundary conditions similarly as above leads to the solution (for µ = µ′ )

sin θ′ cos θ sin(θ − θ′ ) cos(θ + θ′ )

E0′ = 2E0 , E0′′ = 2E0 . (6.47)
sin(θ + θ′ ) cos(θ − θ′ ) sin(θ + θ′ ) cos(θ − θ′ )

Note that when θ + θ′ = π/2, then E0′′ = 0, that is there is no reflection! This angle θ is called
Brewster angle. It can be used to produce linearly polarized light: shine light with arbitrary
polarizations under the Brewster angle θ′ on the material; then the reflected beam is linearly
polarized (orthogonally to the plane of incidence).

86
7 Electromagnetism and Special Relativity

The theory of electromagnetism developed in the 19th century was extraordinarily successful,
unifying the previously unrelated phenomena of electricity and magnetism into a single theory.
For example, it explained Faraday’s law, where a time-dependent magnetic field produces an
electric field, which in turn led to the development of electric motors, transformers, etc. As
we saw in the last section, the theory also interprets visible light, along with the rest of the
electromagnetic spectrum (X-rays, microwaves, radio waves, etc), as a wave propagating through
√
this electromagnetic field. Maxwell identified c = 1/ ϵ0 µ0 with the speed of light in vacuum. But
that also led to a problem: speed relative to what?
Suppose that Louisa is on a train that moves in a straight line with constant speed v relative to
Franklin, who is at rest in the train station. Louisa rolls a marble along the aisle of the train, in
the direction of its motion, with speed u. This means that in Louisa’s inertial reference frame S ′ ,
fixed relative to the train, the marble moves with speed u. In Frankin’s inertial reference frame
S, fixed relative to the Earth’s surface, what is the observed speed of the marble? It’s certainly
greater than u, due to the train’s speed v > 0. If you asked a random person in the street, they
would almost certainly say Franklin sees the marble moving with speed u + v. This is intuitively
obvious, and wrong. It turns out it’s only approximately true, for speeds u, v ≪ c.
Rather than experiment with marbles, suppose that Louisa and Franklin instead measure the
electrostatic force between electric charges, and the magnetostatic force between current carrying
wires, and from Maxwell’s equations thus measure ϵ0 , µ0 . Going back to Galileo, we have:

Postulate 1 The laws of physics are the same in all inertial reference frames.

By this principle, Louisa and Franklin should measure the same values for ϵ0 , µ0 in their two
reference frames, namely those quoted earlier in these lecture notes. But according to Maxwell
√
they will then both observe light to be propagating at the same speed c = 1/ ϵ0 µ0 . Light
is clearly not like marbles: it’s always moving at the same speed, no matter how your inertial
reference frame is moving relative to it. If we believe that Postulate 1 (the Principle of Relativity)
applies to electromagnetism, we are led to:

Postulate 2 The speed of light in vacuum is the same in all inertial reference frames.

These two postulates directly led to Einstein’s 1905 theory of Special Relativity. It supersedes the
Galilean view of space and time, although reduces to it in the limit of small speeds v ≪ c.19
Let us examine the consequences of this a little further. Introduce time and space coordinates
t, x, y, z for Franklin’s reference frame S, and t′ , x′ , y ′ , z ′ for Louisa’s reference frame S ′ . Suppose
19
It is worth remarking that before 1905 physicists, including both Maxwell and Einstein, had instead postulated
that Maxwell’s equations are only valid in a unique universal rest frame. This was supposed to be filled with
something called aether, through which light propagated. However, a famous 1887 experiment by Michelson and
Morely provided strong evidence that Postulate 2 is correct.

87
 z z'

y y'

x x'
O O'
frame S frame S ' v

x = vt
Figure 24: Reference frame S with coordinates t, x, y, z, and reference frame S ′ with coordinates
t′ , x′ , y ′ , z ′ . The origins O, O′ coincide at times t = 0 = t′ , with S ′ moving in the x-axis direction,
relative to S, with speed v. The origin O′ is thus at position x = vt, y = z = 0 in the frame S.

as above that Louisa’s reference frame has speed v relative to Franklin’s, moving in the x-axis
direction. Suppose furthermore that their origins O, O′ coincide at times t = 0 = t′ . At this
moment, a flash of light is emitted from the common origins, expanding as a spherical wave
(precisely as in the retarded Green’s function (5.60)). According to Postulate 2, the speed of this
wave is c in both reference frames, so in particular Franklin will see the wave obey

ct = |r| ⇒ −c2 t2 + x2 + y 2 + z 2 = 0 , (7.1)

p
in his frame S. Here ct is the distance travelled by light in time t, while |r| ≡ x2 + y 2 + z 2 is the
distance of the point r = (x, y, z) from the origin O. But similarly Louisa will see the wave obey

ct′ = |r′ | ⇒ −c2 t′2 + x′2 + y ′2 + z ′2 = 0 , (7.2)

in her frame S ′ .
The issue now is how these coordinates are related to each other. Galileo would say

Galilean transformation : t′ = t , x′ = x − vt , y′ = y , z′ = z . (7.3)

This is a particular case of the more general set of Galilean transformations with

t′ = t − t0 , r′ = R r − r0 − vt . (7.4)

Here t0 is a constant, that is zero if the two observers synchronize their clocks; r0 is a constant
vector, that is zero if the two observers fix a common origin at time t = 0; R is a 3 × 3 orthogonal
matrix (i.e. a rotation and potentially also reflection of the spatial directions); and v is a constant
velocity. The set of transformations (7.4) form a group, called the Galilean group. They map
inertial reference frames to inertial reference frames, in particular meaning they map uniform
motion (i.e. with constant velocity) in one reference from to uniform motion in the other frame.

88
Postulate 1 says that the law physics are the same in any inertial reference frame, and indeed
Newton’s laws of motion are invariant under Galilean transformations.
This Galilean view of space and time was the standard lore before 1905, but it is not consistent
with electromagnetism. Specifically, if you substitute the Galilean transformation (7.3) into (7.2),
you obtain −c2 t2 + (x − vt)2 + y 2 + z 2 = 0, which is not the spherical wavefront (7.1) seen in
Franklin’s frame. It should be clear why: according to Galileo, a ray of light travelling at speed
u = c in the positive x-axis direction in Louisa’s frame S ′ has x′ = ut, which in Franklin’s frame
S is x = x′ + vt = (u + v)t, and thus has speed u + v = c + v ̸= c. Galilean transformations are
inconsistent with Postulate 2.
The transformations (7.4) are linear maps from (ct, x, y, z) ∈ R4 to (ct′ , x′ , y ′ , z ′ ) ∈ R4 . Here
we have multiplied the time coordinate by c so that ct also has dimensions of length. Notice the
maps are linear because we want to map uniform motion (which traces out straight lines in R4 ) to
uniform motion. We may rewrite (7.4) as
 ′       
ct 1 0 0 0 ct ct0 ct
 x′   − v1  x  x0  x
  =  vc    +   ≡ G   + constant . (7.5)
 y′   − 2 R y  y0  y
c
′ v3
z −c z z0 z
Here the 3 × 3 orthogonal matrix R fills the lower right hand block of the 4 × 4 matrix G.
The linear transformation that maps (ct, x, y, z) to (ct′ , x′ , y ′ , z ′ ) that is consistent with (7.1)
and (7.2) is
ct − v x x − v · ct
Lorentz transformation : ct′ = q c , x′ = q c , y′ = y , z ′ = z . (7.6)
2 2
1 − vc2 1 − vc2

Specifically, one can easily verify that −c2 t′2 + x′2 + y ′2 + z ′2 = −c2 t2 + x2 + y 2 + z 2 under this
transformation. Notice that (7.6) approximately reduces to (7.3) for speeds v ≪ c. Since (7.6) is
linear, we may write it similarly to (7.5)
 ′
− vc γ
    
ct γ 0 0 ct ct
 x′  v
  = − c γ γ 0 0
 x x
  ≡ L  , (7.7)
 y′   0 0 1 0y y
z′ 0 0 0 1 z z
where we have introduced
1
γ = γ(v) ≡ p . (7.8)
1 − v 2 /c2
The Galilean transformation (7.3) has x′ = x − vt, while the Lorentz transformation has x′ =
γ(x − vt), and moreover treats the time and space directions symmetrically. Notice also that a
marble moving with speed u along the positive x-axis direction in Louisa’s frame S ′ has x′ = ut′ ,
which in terms of x and t is the equation
 v  (u + v)t
γ(x − vt) = uγ t − 2 x ⇒ x = , (7.9)
c 1 + uv/c2

89
so that the speed as seen in Franklin’s frame S is (u+v)/(1+uv/c2 ). This approximately reduces to
u + v, for uv ≪ c2 . On the other hand, for u = c the speed in the frame S is (c + v)/(1 + cv/c2 ) = c.
Lorentz discovered these transformations by studying Maxwell’s equations, realizing they were
not invariant under Galilean transformations. Indeed, we noted this already in section 5.2 when
motivating Faraday’s law. For example, the Biot-Savart law says that moving charges generate
magnetic fields, but moving relative to which reference frame? Einstein showed that the same
transformations follow directly from Postulates 1 and 2, independently of Maxwell’s equations.
The most striking feature of (7.6) is that the time coordinates in the two inertial frames are not
the same, due to the factor of γ. Consider a clock at rest in Franklin’s frame S. The location of
the clock on two different ticks is the same, so ∆x = 0, and (7.6) gives

∆t′ = γ ∆t . (7.10)

Here ∆t is the time interval between ticks of the clock, as seen in Franklin’s frame S, while ∆t′ is
the time interval between ticks of the clock, as seen in Louisa’s frame S ′ . Since γ > 1 for v ̸= 0,
∆t′ > ∆t. In other words, Louisa sees the time between ticks of Franklin’s clock taking longer
than the time ∆t. His clock seems to be running slow. The fact that ∆t′ = ∆t in Galileo’s view
of space and time was always a (tacit) assumption, and it is not compatible with Postulate 2.
The Lorentz transformations may be characterized mathematically as follows. We first assemble
⃗ ∈ R4 , writing X
the time and space coordinates into a four-vector X ⃗ = (ct, x, y, z)T . We may then
write a general Lorentz transformation, as in (7.7), as

⃗′ = LX
X ⃗ , (7.11)

with L a linear map on spacetime R4 . We then define the 4 × 4 diagonal matrix

 
−1 0 0 0
 0 1 0 0
η ≡  0 0 1 0 ,
 (7.12)
0 0 0 1

(the Minkowski metric tensor ), and note that we may write

⃗Tη X
−c2 t2 + x2 + y 2 + z 2 = X ⃗ . (7.13)

Lorentz transformations preserve this quadratic form, meaning

⃗ ′T η X
X ⃗′ ≡ X
⃗ T LT η L X
⃗ = X
⃗Tη X
⃗ ⃗ ∈ R4 ,
holds for all X (7.14)

or in other words −c2 t′2 + x′2 + y ′2 + z ′2 = −c2 t2 + x2 + y 2 + z 2 . This in turn implies

LT η L = η . (7.15)

90
This is the defining property of a Lorentz transformation L. One can compare to the 3 × 3
orthogonal transformation R, which by definition satisfies RT R = 13×3 , and preserves Euclidean
distance so

r′T r′ ≡ rT RT R r = rT r holds for all r ∈ R3 . (7.16)

Indeed, rotations are contained within the Lorentz transformations as

 
1 0 0 0
 0 
L =   0
 , (7.17)
R 
0

just as they are contained within the Galilean transformations (7.5). To summarize, Lorentz
transformations preserve the Lorentzian square distance −c2 t2 + x2 + y 2 + z 2 in R4 , which unlike
a usual distance can be positive, negative, or zero.
After this brief detour into Special Relativity and Lorentz transformations, we return to dis-
cuss electromagnetism. We have already noted that Maxwell’s equations are not invariant under
Galilean transformations. Under a 3 × 3 rotation R, a vector such as E or B would rotate as a
vector, and we have seen that Lorentz transformations naturally act on four-vectors, rather than
three-vectors, but contain rotations as a special case. The correct Lorentz transformations of
electromagnetism are most easily stated by first recalling that
∂A
E = −∇ϕ − , B = ∇×A , (7.18)
∂t
1
in terms of the potentials ϕ and A. Recalling also that c E and B have the same dimensions, it is
natural to put ϕ and A into the four-vector
 T  T
⃗ ≡ ϕ ϕ
A , A1 , A2 , A3 = ,A , (7.19)
c c

and similarly define the four-current

J⃗ ≡ (c ρ , J) , (7.20)

⃗ ′ and current J⃗′ in the

in terms of the charge density ρ and current density J. The four-vector A
frame S ′ are then simply

⃗′ = L A
A ⃗, J⃗′ = L J⃗ , (7.21)

where L is the Lorentz transformation from S to S ′ . In Lorenz gauge we can write the Maxwell
equations (5.34), (5.35) as the single four-vector equation

⃗ = −µ0 J⃗ .
□A (7.22)

91
Moreover, notice that
3
1 ∂2 X ∂2
□ ≡ − + = ∂⃗ T η ∂⃗ , (7.23)
c2 ∂t2 ∂x2i
i=1

where ∂⃗ ≡ ( 1c ∂t , ∂x , ∂y , ∂z )T is the gradient operator on spacetime. The d’Alembertian □ is thus the

natural analogue of the Laplacian in spacetime, and is invariant under Lorentz transformations.
It follows that Maxwell’s equations (7.22) take the same form in both reference frames, with both
sides transforming as a Lorentz four-vector.

Example We reconsider the example at the end of section 5.6, namely a point charge q moving
with constant velocity v = v e1 in the reference frame S. We computed the potentials ϕ and A in
equations (5.75) and (5.76), using the general solution to the time-dependent Maxwell equations
(5.62) and (5.63), respectively. The point charge hence generates both an E and a B field in the
frame S. On the other hand, in the frame S ′ this charge is at rest. Taking this to be the origin
r′ = 0 in S ′ , the laws of statics imply that the charge generates the potentials
q 1
ϕ′ (r′ ) = , A′ = 0 , (7.24)
4πϵ0 r′
in the frame S ′ .
⃗′ = ( ϕ′ ′ T ⃗ = ( ϕ , A)T in the
The four-vector A c ,A ) in the frame S ′ is related the four-vector A c
frame S via the Lorentz transformation (7.21). Using (7.7) this reads
ϕ′ ϕ v v ϕ
= γ − γ A1 , A′1 = − γ + γ A1 , A′2 = A2 , A′3 = A3 . (7.25)
c c c c c
The Lorentz transformed postion vector is

r′ = γ(x − vt) e1 + y e2 + z e3 , r′2 = γ 2 (x − vt)2 + y 2 + z 2 . (7.26)

ϕ
Since from (7.24) A′ = 0, one immediately solves the last three equations in (7.25) to find A = c2
v.
Substituting this into the first equation in (7.25) then gives
ϕ′ v2 ϕ
 
1 ϕ
= γ 1− 2 = , (7.27)
c c c γ c
and hence from (7.24) we deduce
qγ 1 ϕ(r, t)
ϕ(r, t) = , A = v, (7.28)
c2
p
4πϵ0 γ (x − vt)2 + y 2 + z 2
2

in precise agreement with (5.75), (5.76)! We have here derived these formulae from Coulomb’s law
(7.24) in the frame S ′ , together with a Lorentz transformation.
The electric and magnetic fields in the frame S may be computed from these potentials using
the usual formulae (7.18). Indeed, combining the latter with the Lorentz transformation of the
potentials (7.25) leads to the transformations

E1′ = E1 , E2′ = γ (E2 − vB3 ) , E3′ = γ (E3 + vB2 ) ,

 v   v 
B1′ = B1 , B2′ = γ B2 + 2 E3 , B3′ = γ B3 − 2 E2 . (7.29)
c c

92
After a computation, in our current example with potentials (7.28) one finds
q γ 1
E(r, t) = (r − vt) , B(r, t) = v×E . (7.30)
4πϵ0 [γ (x − vt) + x2 + z 2 ]3/2
2 2 c2

If we denote R ≡ r − vt to be the postion vector of the observation point r, relative to the position
vector vt of the point charge q, we may write these as
qγ 1 µ0 qγ v × R
E = R, B = . (7.31)
4πϵ0 γ 2 R2 + R2 + R2 3/2 4π γ 2 R2 + R2 + R2 3/2
1 2 3 1 2 3

When v ≪ c we may approximate γ ≃ 1, and the equation for E is Coulomb’s law (1.7), while the
equation for B is the Biot-Savart law (3.10)! In particular, notice that we have effectively derived
the Biot-Savart law from Coulomb’s law, using only a Lorentz transformation!
There is of course much more to say about Special Relativity than the comments we have
made in this section, but having derived magnetostatics from electrostatics and the structure of
spacetime, we conclude here. ■

93
A Vector calculus

The following is a summary of some results from the Prelims Multivariable Calculus course. As
in the main text, all functions and vector fields are assumed to be sufficiently well-behaved in
order for formulae to make sense. For example, one might take everything to be smooth (partial
derivatives to all orders exist). Similar remarks apply to (the parametrizations of) curves and
surfaces in R3 .

A.1 Vectors in R3

We work in R3 , or a domain therein, in Cartesian coordinates. If e1 = (1, 0, 0), e2 = (0, 1, 0),

e3 = (0, 0, 1) denote the standard orthonormal basis vectors, then a position vector is
3
X
r = xi ei , (A.1)
i=1

where x1 = x, x2 = y, x3 = z are the Cartesian coordinates in this basis. We denote the Euclidean
length of r by
q
|r| = r = x21 + x22 + x23 , (A.2)

so that r̂ ≡ r/r is a unit vector for r ̸= 0. A vector field f = f (r) may be written in this basis as
3
X
f (r) = fi (r) ei . (A.3)
i=1

The scalar product of two vectors a, b is denoted by

3
X
a·b ≡ ai bi , (A.4)
i=1

while their vector cross product is the vector

a × b ≡ (a2 b3 − a3 b2 ) e1 + (a3 b1 − a1 b3 ) e2 + (a1 b2 − a2 b1 ) e3 . (A.5)

The vector triple product identity reads

a × (b × c) = (a · c) b − (a · b) c , (A.6)

which holds for all vectors a, b, c.

A.2 Vector operators

The gradient of a function ψ = ψ(r) is the vector field

3
X ∂ψ
grad ψ = ∇ψ ≡ ei . (A.7)
∂xi
i=1

94
The divergence of a vector field f = f (r) is the function (scalar field)
3 3
X ∂f X ∂fi
div f = ∇ · f ≡ ei · = , (A.8)
∂xi ∂xi
i=1 i=1

while the curl is the vector field

3
X ∂f
curl f = ∇ × f ≡ ei ×
∂xi
i=1
     
∂f3 ∂f2 ∂f1 ∂f3 ∂f2 ∂f1
= − e1 + − e2 + − e3 . (A.9)
∂x2 ∂x3 ∂x3 ∂x1 ∂x1 ∂x2
Two important identities are

∇ × (∇ψ) = 0 , ∇ · (∇ × f ) = 0 , (A.10)

or in words: the curl of a gradient is zero, and the divergence of a curl is zero. Two more identities
we shall need are

∇ × (a × b) = a (∇ · b) − b (∇ · a) + (b · ∇) a − (a · ∇) b , (A.11)
∇ · (a × b) = b · (∇ × a) − a · (∇ × b) . (A.12)

The second order operator ∇2 , defined by

3
X ∂2ψ
∇2 ψ ≡ ∇ · (∇ψ) = , (A.13)
i=1
∂x2i

is called the Laplacian. We shall also use the identity

∇ × (∇ × f ) = ∇ (∇ · f ) − ∇2 f . (A.14)

Notice from the definitions (A.2) and (A.7) that

3 3
X ∂r X xi r
∇r = ei = ei = = r̂ , (A.15)
∂xi r r
i=1 i=1

and, by translating r → r − r′ , more generally ∇ |r − r′ | = (r − r′ )/|r − r′ |.

A.3 Integral theorems

Definition (Line integral) Let C be a curve in R3 , parametrized by r : [t0 , t1 ] → R3 , or r(t) for

short. Then the line integral of a scalar field ψ and vector field f along C are respectively
Z Z t1 Z Z t1
dr(t) dr(t)
ψ ds ≡ ψ(r(t)) dt , f · dr ≡ f (r(t)) · dt . (A.16)
C t0 dt C t0 dt
Note that τ (t) ≡ dr/dt is the tangent vector to the curve – a vector field defined on C. The values
of these integrals are independent of the choice of oriented parametrization (proof uses the chain
rule).
A curve is simple if r : [t0 , t1 ] → R3 is injective (then C is non-self-intersecting), and is closed if
r(t0 ) = r(t1 ) (then C forms a loop).

95
Definition (Surface integral) Let Σ be a surface in R3 , parametrized by r(u, v), with (u, v) ∈ D ⊆
R2 . The unit normal n to the surface is
tu × tv
n ≡ , (A.17)
|tu × tv |
where
∂r ∂r
tu ≡ , tv ≡ (A.18)
∂u ∂v
are two tangent vectors to the surface. These are all vector fields defined on Σ. The surface integral
of a function ψ over Σ is
Z ZZ
∂r ∂r
ψ dS ≡ ψ(r(u, v)) × du dv . (A.19)
Σ D ∂u ∂v
The sign of n in (A.17) is not in general independent of the choice of parametrization. Typically,
the whole of a surface cannot be parametrized by a single domain D; rather, one needs to cover
Σ with several parametrizations using domains DI ⊆ R2 , where I labels the domain. The surface
integral (A.19) is then defined in the obvious way, as a sum of integrals over DI ⊆ R2 . However,
in doing this it might not be possible to define a continuous n over the whole of Σ (an example
being a Möbius strip).

Definition (Orientations) A surface Σ is orientable if there is a choice of continuous unit normal

vector field n on Σ. If an orientable Σ has boundary ∂Σ, a simple closed curve, then the normal
n induces an orientation of ∂Σ: we require that τ × n points away from Σ, where τ denotes the
oriented tangent vector to ∂Σ – see Figure 25.

The point here is that the choice of direction τ along the curve ∂Σ in turn fixes the choice of sign
when integrating over ∂Σ.

n
∂Σ
Σ
τ

τ n
Figure 25: Surface Σ with unit normal n, and boundary ∂Σ with oriented tangent vector τ . The
direction of τ is fixed by requiring τ × n to point away from Σ. (Right hand rule for the cross
product: τ points along the index finger, n along the middle finger, and τ × n along the thumb.)

With this in hand, we may now state

Theorem A.1 (Stokes ) Let Σ be an orientable surface in R3 , with unit normal vector n and
boundary curve ∂Σ. If f is a vector field then
Z Z
(∇ × f ) · n dS = f · dr . (A.20)
Σ ∂Σ

96
Definition (Volume integral) The integral of a function ψ in a (bounded) region R in R3 is
Z ZZZ
ψ dV ≡ ψ(r) dx1 dx2 dx3 . (A.21)
R R

Theorem A.2 (Divergence ) Let R be a bounded region in R3 with boundary surface ∂R. If f is
a vector field then
Z Z
∇ · f dV = f · n dS , (A.22)
R ∂R

where n is the outward unit normal vector to ∂R.

Note that the surface Σ in Stokes’ theorem has a boundary ∂Σ, whereas the surface ∂R in the
divergence theorem does not (it is itself the boundary of the region R).
Finally, we will need the following result proved in the Prelims Multivariable Calculus course:

Lemma A.3 If f is a continuous function such that

Z
f dV = 0 , (A.23)
R

for all bounded regions R, then f ≡ 0.

The proof of this is standard: if f is non-zero at a point r0 ∈ R3 , say f (r0 ) > 0 is positive, then
by continuity f is also positive in a small neighbourhood around r0 . But then the intergal of f
over that neighbourhood will be positive, contradicting (A.23).

97