Maxwells Theory Notes

Maxwell’s Theory of Electrodynamics
From lectures given by

Yaroslav Kurylev
Edited, annotated and expanded by

Adam Townsend & Giancarlo Grasso
19 May 2012
MATH3308 Maxwell’s Theory of Electrodynamics
Figure 1: Our Dear Leader and His Spirit Guide
2
Contents
1 Introduction to the Theory of Distributions 5

1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Definitions and notation . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Functionals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.4 Distributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.5 Support of a Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.6 Differentiation of Distributions . . . . . . . . . . . . . . . . . . . . . . . 22
1.7 Convolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.8 Density . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.9 Integration of Distributions . . . . . . . . . . . . . . . . . . . . . . . . 37
1.10 The Laplace Operator and Green’s Function . . . . . . . . . . . . . . . 42
2 Electrostatics 47
2.1 Introduction to electrostatics . . . . . . . . . . . . . . . . . . . . . . . . 47
2.2 The fundamental equations of electrostatics . . . . . . . . . . . . . . . 48
2.3 Divergence theorem and Stokes’ theorem . . . . . . . . . . . . . . . . . 51
2.4 Finding E and Φ for different ρ . . . . . . . . . . . . . . . . . . . . . . 52
2.5 Dipoles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.6 Conductors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
2.7 Boundary value problems of electrostatics . . . . . . . . . . . . . . . . 64
2.8 Dirichlet boundary value problems for Laplace’s equation . . . . . . . . 67
3 Magnetism 77
3.1 The laws for magnetostatics . . . . . . . . . . . . . . . . . . . . . . . . 77
3.2 The laws of magnetodynamics . . . . . . . . . . . . . . . . . . . . . . . 82
3.3 Maxwell’s equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
3.4 Étude on differential forms . . . . . . . . . . . . . . . . . . . . . . . . . 87
4 Electromagnetic Waves 91
4.1 Electromagnetic energy . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.2 Electromagnetic waves in an homogeneous isotropic medium . . . . . . 94
4.3 Plane waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.4 Harmonic plane waves . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
3
Editors’ note
This set of course notes was released initially in April 2011 for the University Col-
lege London mathematics course MATH3308, Maxwell’s Theory of Electrodynam-
ics, a ten-week course taught by Professor Yaroslav Kurylev between January and
March 2011. The editors were members of this class. These notes were published
to http://adamtownsend.com and if you wish to distribute these notes, please direct
people to the website in order to download the latest version: we expect to be making
small updates or error corrections sporadically. If you spot any errors, you can leave a
comment on the website or send an email to the authors at adam@adamtownsend.com.
We hope you enjoy the notes and find them useful!
AKT GWG
London, April 2011
Acknowledgments
The editors would like to thank the following for their help in the production of this
set of course notes:
• Lisa for catapults and providing essential Russian to English translation;
• Kav for questions;
• The Permutation Boys (Jonathan, James and Charles) for permuting;
• Ewan for patiently being Ian;
• Prof. Kurylev for coffee and many, many invaluable life lessons.
We would also like to thank those who have offered corrections since the original
publication.
4
Chapter 1
Introduction to the Theory of

Distributions
1.1 Introduction
James Clerk Maxwell (1831–1879), pictured on page 2, was a Scottish mathematician

and physicist who is attributed with formulating classical electromagnetic theory, unit-
ing all previously unrelated observations, experiments and equations of electricity,
magnetism and even optics into one consistent theory. Maxwell’s equations demon-
strated that electricity, magnetism and even light are all manifestations of the same
phenomenon, namely the electromagnetic field. By the end of the course we’ll have
built up the mathematical foundation that we need in order to understand this fully
(more so than Maxwell himself!), and we’ll have derived and used Maxwell’s famous
equations. Although they bear his name, like much of mathematics, his equations (in
earlier forms) come from the work of other mathematicians, but they were brought
together for the first time in his paper, On Physical Lines of Force, published between
1861 and 1862∗ .
The first chapter of this course is concerned with putting together a mathematically
sound foundation which we can build Maxwell’s theory of electrodynamics upon. Max-
well’s theory introduces functions that don’t have derivatives in the classical sense,
so we use a relatively new addition to analysis in order to understand this. This
foundation is the Theory of Distributions. Distributions (not related to probability
distributions) are the far-reaching generalisations of the notion of a function and their
theory has become an indispensable tool in modern analysis, particularly in the theory
of partial differential equations. The theory as a whole was put forth by the French†
mathematician Laurent Schwartz in the late 1940s, however, its most important ideas
(of so-called ‘weak solutions’) were introduced back in 1935 by the Soviet mathem-
atician Sergei Sobolev. This course will cover only the very beginnings of this theory.
∗
Paraphrased from Wikipedia
†
Honest!
5
1.2 Definitions and notation
Let me start by reminding you of some definitions and notation:
Definition 1.1 A compact set K in a domain Ω, denoted K ⋐ Ω is a closed, bounded

set.
Notation 1.2 C ∞ (Ω) is the set of continuous functions which can be differentiated
infinitely many times in a domain Ω.
Notation 1.3 C0 (Ω) is the set of continuous functions which are equal to 0 outside
some compact K ⋐ Ω. That is to say, if f ∈ C0 , ∀x ∈ / K, f (x) = 0. This is known as
the set of continuous functions with compact support
Notation 1.4 C0∞ (Ω) is the set of continuous, infinitely differentiable functions with
compact support
Notation 1.5 We use Br (x0 ) to denote the closed ball of radius r centred at a point
x0 in our domain Ω: that is to say,
Br (x0 ) = {x ∈ Ω : |x − x0 | ≤ r}
where | · | is the norm of our domain. For this course we’ll always say that our
domains are subsets of Rn , so we don’t need to worry about special types of norm:
|v| is the standard (Euclidean) length of the vector v, which we work out in the usual
Pythagorean way.∗
Note: we’ll use the convention of writing vectors in boldface (e.g. v) from chapter 2
onwards—the first chapter deals heavily in analysis and since we are being general in
which domain we’re working in (R, R2 or R3 ), the distinction between one-dimensional
scalars and higher-dimensional vectors is unnecessary; most of our results hold in n
dimensions, and excessive boldface would clutter the notes. As such we’ll use standard
italic (e.g. v) unless clarity demands it of us.
1.2.1 The du Bois-Reymond lemma and test functions
We start by proving a very useful lemma, first proved by the German mathematician
Paul du Bois-Reymond.
Lemma 1.6 Let Ω ⊂ Rn be an open domain (bounded or unbounded). Let f (x) ∈

C(Ω). Assume that for any φ ∈ C0∞ (Ω),
Z
f (x)φ(x)dx = 0.
Ω
∗
You may find that other authors use the notations B(x0 , r), B(x0 , r) or Bc (x0 , r) (where the bar
and c denote ‘closed’).
6
Chapter 1. Introduction to the Theory of Distributions
Then f ≡ 0.
Proof: We’ll approach this using contradiction. Say we have x0 ∈ Ω such that
f (x0 ) > 0.
f ∈ C(Ω), i.e.
∀ ε > 0, ∃ δ > 0 s.t. |x − x0 | < δ =⇒ |f (x) − f (x0 )| < ε
Let ε = 21 f (x0 ), i.e. there exists a ball of radius δ, Bδ (x0 ) such that f (x) > 12 f (x0 ) for
all x ∈ Bδ (x0 ).
Choose φ from C0∞ such that:
1. φ(x) ≥ 0 for all x

2. φ(x) = 1 for x ∈ B δ (x0 )
2
3. φ(x) = 0 for x ∈
/ Bδ (x0 ).
Then look at Z Z
f (x)φ(x)dx = f (x)φ(x)dx
Ω Bδ (x0 )
since φ is zero outside of this ball. But

Z Z
f (x)φ(x)dx ≥ f (x)φ(x)dx
Bδ (x0 ) B δ (x0 )
2
since all the f (x)φ(x) terms are positive. Then

Z Z
f (x)φ(x)dx = f (x)dx
B δ (x0 ) B δ (x0 )
2 2
since φ(x) = 1 by condition 2 above. Then

Z
1
f (x)dx ≥ f (x0 ) · Vol(B δ (x0 )) > 0
2 2
B δ (x0 )
2
by our choice of ε above.
But our initial assumption was that

Z
f (x)φ(x)dx = 0
Ω
which is a contradiction.
The crucial result underlining the proof of this lemma is the following:
7
Proposition 1.7 Let K1 and K2 be two compacts in Rn such that K1 ⋐ K2 . Then

there exists φ ∈ C0∞ (Rn ) such that
1. 0 ≤ φ(x) ≤ 1 on Rn ;
2. φ(x) = 1 on K1 ;
3. φ(x) = 0 on Rn \ K2 .
Rn
φ=0
K2
K1
φ=1
Figure 1.1: Proposition 1.7 tells us this exists
We can take from the lemma a simple corollary:
Corollary 1.8 Let f, g ∈ C(Ω). Then if ∀φ ∈ C0∞ (Ω),

Z Z
f (x)φ(x)dx = g(x)φ(x)dx
Ω Ω
then f = g.
Proof: Subtracting one side from the other gives

Z Z
f (x)φ(x)dx − g(x)φ(x)dx = 0
Ω Ω
Z
(f (x) − g(x))φ(x)dx = 0
Ω
Then f (x) − g(x) = 0 by Lemma 1.6, i.e. f (x) = g(x).
To understand the importance of this result, remember that normally to test whether
two functions f and g are equal, we have to compare their values at all points x ∈ Ω.
By the Du Bois-Reymond lemma, however, this is equivalent to the fact that integrals
of f and g with all φ ∈ C0∞ (Ω) are equal. Thus, functions are uniquely characterised by
their integral with C0∞ (Ω) functions. It is, of course, crucial that we look at the values
of integrals of f with all C0∞ (Ω) functions, i.e. to test f with all C0∞ (Ω) functions
which are, therefore, called test functions.
8
Definition 1.9 C0∞ (Ω) is called the space of test functions and is often denoted
D(Ω)
1.2.2 Topological vector spaces and multi-indices
Definition 1.10 D(Ω) is a topological vector space, i.e.
1. It is a linear vector space, i.e. if f, g ∈ D(Ω) and λ, µ ∈ R,
h(x) = (λf + µg)(x) = λf (x) + µg(x) ∈ D(Ω) x ∈ Ω
and h(x) = 0 for x ∈

/ Kf ∪ Kg , i.e. x not in the compact support of either.
2. There is a notion of convergence, i.e. there is a meaning to fk −−−→ f . This is

k→∞
defined shortly.
For the next few definitions we need to introduce a further few convenient definitions:
Definition 1.11 A multi-index α is an n-dimensional vector, where n = dim(Ω),

with non-negative integer components:
α = (α1 , α2 , . . . , αn ), αi = 0, 1, . . . i = 1, . . . , n
Multi-indices are convenient when we want to write high-order partial derivatives,

namely
∂ |α| φ
∂ α φ(x) = , j = 1, . . . , n, |α| = α1 + · · · + αn .
∂xα1 1 · · · ∂xαnn
1.2.3 Null set and support
Definition 1.12 The null set of a function f , N (f ), is the maximal open set in Ω
where f ≡ 0. In other words, x0 ∈ N (f ) if it has a neighbourhood x0 ∈ U ⊂ Ω with
f (x) = 0 when x ∈ U and if any other open set A holds this property, then A ⊂ N (f )
Definition 1.13 The support of a function f , supp(f ) is the complement of N (f )

in Ω, i.e.
supp(f ) = Ω \ N (f )
Note that by this definition, supp(f ) is always closed in Ω.
9
Example 1.14 Let Ω = R and

0 if x ≤ 0
f (x) =
x if x > 0
Then N (f ) = (−∞, 0) and supp(f ) = [0, ∞).
f (x) 6
-
x
Figure 1.2: Look, a picture of the function!
1.2.4 Convergence
Definition 1.15 A function φk converges to φ in D(Ω) if the following two conditions

are satisfied:
1. For any multi-index α and any x ∈ Ω,
∂ α φp (x) → ∂ α φ(x), as p → ∞
i.e. all derivatives converge
2. There is a compact K ⋐ Ω such that
supp(φp ), supp(φ) ⊂ K, p = 1, 2, . . .
Problem 1.16 For any φ ∈ D(Ω) and any multi-index β, show that
∂ β φ ∈ D(Ω)
Proof:
N (∂ β φ) ⊃ N (φ) =⇒ supp(∂ β φ) ⊂ supp(φ) ⋐ Ω
10
supp φp supp φ
Figure 1.3: Definition 1.15(2) in picture form
Problem 1.17 If ζ ∈ C ∞ (Ω), then for any φ ∈ D(Ω), show
ψ = ζφ ∈ D(Ω)
Proof:
supp(ψ) ⊂ supp(φ) ⋐ Ω
Problem 1.18 Let φp → φ in D(Ω) and let β = (β1 , . . . , βn ) be some fixed multi-
index. Show that
∂ β φp → ∂ β φ in D(Ω).
Solution This just follows from condition 1 of the definition of φp converging to

φ. X
Problem 1.19 ∗ Let ζ ∈ C ∞ (Ω), i.e. ζ is an infinitely differentiable function in Ω.

Show that, for φp → φ in D(Ω),
ζφp → ζφ in D(Ω).
Solution Want to show that if ψp = ζφp, ψ = ζφ, that
1. ∀β, ∂ β ψp → ∂ β ψ
2. ∃K s.t. supp(ψp ), supp(ψ) ⊂ K
So let’s do it.
11
1.
∂ β ψp = ∂ β (ζφp )
X β
= ∂ β−γ ζ (∂ γ φp )
γ
γ:γ≤β
which is the sum of C ∞ terms. So as p → ∞, since φp → φ in D(Ω),

X β
→ ∂ β−γ ζ (∂ γ φ)
γ
γ:γ≤β
= ∂ β (ζφ).
2. N (ψ) ⊃ N (φ) =⇒ supp(ψ) ⊂ supp(φ) ⋐ Ω since φp → φ in D(Ω).

N (ψp ) ⊃ N (φp ) =⇒ supp(ψp ) ⊂ supp(φp ) ⋐ Ω.
And so ∃K ⋐ Ω such that supp(ψ), supp(ψp ) ∈ K.
Problem 1.20 If φp → φ, ψp → ψ in D(Ω) and λ, µ ∈ R, show that
λφp + µψp → λφ + µψ in D(Ω)
Problem 1.21 If φp → φ, ψp → ψ in D(Ω) and λp → λ, µp → µ in R, show that
λp φp + µp ψp → λφ + µψ in D(Ω)
1.3 Functionals
Now we move on to functionals:
Definition 1.22 A function F : V → R, mapping from a vector space onto the

reals, is called a functional.
Definition 1.23 A continuous, linear functional F has the properties,
1. Linearity:
F (λv + µw) = λF (v) + µF (w); v, w ∈ V ; λ, µ ∈ R (1.1)
2. Continuity:
F (vp ) → F (v) as vp → v (1.2)
12
Problem 1.24 1. Let F be a linear functional on a topological vector space V .

Show that F (0) = 0, where 0 is the zero vector in V .
2. Assume, in addition, that F satisfies
F (vp ) → 0, as vp → 0
Show that then F is continuous.
1.3.1 Dual space
Definition 1.25 The space of continuous linear functionals on V is called the dual
space to V and is denoteda by V ′ .
The dual space is itself a topological vector space if, for λ, µ ∈ R and F, G ∈ V ′ we
define λF + µG by
(λF + µG)(v) = λF (v) + µG(v), v∈V (1.3)
and Fp → F if, for any v ∈ V ,
Fp (v) → F (v), as p → ∞. (1.4)
Problem 1.26 Show that H = λF + µG defined by (1.3) is a continuous linear

functional on V , i.e. H satisfies (1.1), (1.2).
Proof: We need to check that this object is linear and continuous:
1. Linearity:
(λF + µG)(αv + βw) = λF (αv + βw) + µG(αv + βw)
and each F, G are linear functionals, therefore
= λ(αF (v) + βF (w)) + µ(αG(v) + βG(w))
= α[λF (v) + µG(v)] + β[λF (w) + µG(w)]
= α(λF + µG)(v) + β(λF + µG)(w)
2. Continuity:
(λF + µG)(vp ) =λF (vp ) + µG(vp )
−−−→λF (v) + µG(v)
p→∞
=(λF + µG)(v)
†
Other, if not most, authors denote the dual space to V by V ∗
13
Problem 1.27 ∗ Show that if Fp → F and Gp → G, where Fp , F, Gp , G ∈ V ′ , then
λFp + µGp → λF + µG.
Solution Again, just like above,
(λFp + µGp )(v) = λFp (v) + µGp (v)
and, by the information given in the question,
−−−→ λF (v) + µG(v)

p→∞
= (λF + µG)(v)
Problem 1.28 If Fp → F , Gp → G, λp → λ, µp → µ, show that
λp Fp + µp Gp → λF + µG
Solution We want to show then that
(λp Fp + µp Gp )(v) → (λF + µG)(v)
So here we go, quite succinctly.
(λp Fp + µp Gp )(v) = λp Fp (v) + µp Gp (v)

−−−→ λF (v) + µG(v)
p→∞
= (λF + µG)(v)
Example 1.29 Let V = R2 , the space of 2-dimensional vectors. Recall that vp → v

iff kvp − vk → 0. Then V ′ is actually V itself and we define, for w ∈ V ′ (which is
again just a 2-vector) and v ∈ V ,
w(v) = w · v = w1 v1 + w2 v2
i.e. the scalar product of w and v.
1.4 Distributions
Let us return to the topological vector space D(Ω) = C0∞ (Ω). Then it has a dual space
of its functionals.
14
Definition 1.30 The topological vector space D ′ (Ω) of continuous linear functionals
on D(Ω) is called the space of distributions on Ω.
Let me just remind you of a definition and result from real analysis (MATH7102):
Definition 1.31 We say that φp converges to φ (φp −−−→ φ) uniformly if:

p→∞
∀ε > 0 ∃P (ε) s.t. p > P (ε) =⇒ |φp (x) − φ(x)| < ε ∀x ∈ Ω
Lemma 1.32 Let φp ∈ C(Ω) and φp (x) → φ(x) for any x ∈ Ω. Then for any
compact K ⋐ Ω, φp → φ uniformly.
What are some examples of distributions?
Proposition 1.33 Take any f ∈ C(Ω). For any φ ∈ D(Ω), consider

Z
Ff (φ) = f (x)φ(x)dx.
Ω
Then Ff is a distribution (a continuous linear functional on D(Ω)) called a distribution

associated with the function f . That is to say, the integrals form a linear continuous
functional in D(Ω).
Proof:
1. Linearity:
Z
Ff (λφ + µψ) = f (x) (λφ(x) + µψ(x)) dx
Ω
Z Z
=λ f (x)φ(x)dx + µ f (x)ψ(x)dx
Ω Ω
= λFf (φ) + µFf (ψ) (1.5)
i.e Ff satisfies (1.1).
2. Continuity:
Let φp → φ in D(Ω).
Then we’re told in definition 1.15 that for any multi-index α,
∂ α φp (x) → ∂ α φ(x) x ∈ Ω
So take α = (0, . . . , 0), then obviously since ∂ 0 φ = φ, we get
φp (x) → φ(x), x∈Ω
=⇒ f (x)φp (x) → f (x)φ(x)
R R
But we can’t say that f (x)φp (x) dx → f (x)φ(x) dx because this simply isn’t
true.
15
Counterexample: Take the interval (0, 1) and

p if x ∈ (0, p1 )
φp (x) =
0 if x ∈ [ p1 , 1)
R1
Then ∀x ∈ (0, 1), φp (x) → 0 but 0 φp (x)dx = 1.
We need the extra condition that φp → φ uniformly.
Recall that the definition of convergence (1.15) says that there is a compact
K ⋐ Ω such that
supp(φp ), supp(φ) ⊂ K, p = 1, 2, . . .
Then by Lemma 1.32 above, we have all we need and

Z Z
Ff (φp ) = f (x)φp (x)dx → f (x)φ(x)dx = Ff (φ). (1.6)
K K
This completes the proof.
So in summary, every f ∈ C(Ω) defines a distribution Ff ∈ D ′ (Ω), where

Z
Ff (φ) = f (x)φ(x)dx
Ω
Moreover, the Du Bois-Reymond lemma says that to two different functions f1 and
f2 , we associate different Ff1 and Ff2 .
So we say that C is embedded in D ′ , C(Ω) ⊂ D ′ (Ω), and this embedding is continuous.
Previously we defined uniform convergence of a function, let us state what it means

for a function to converge pointwise:
Definition 1.34 A function fp −−−→ f pointwise in C(Ω) if

p→∞
fp (x) → f (x) ∀x ∈ Ω
Problem 1.35 ∗ If fp , f ∈ C(Ω) and fp → f pointwise, show that
Ffp → Ff in D ′ (Ω)
Proof: Ffp → Ff in D ′ (Ω) means that

Z Z
′
∀φ ∈ D (Ω), Ffp (φ) → Ff (φ) ⇐⇒ fp (x)φ(x)dx → f (x)φ(x)dx
Ω Ω
Z Z
⇐⇒ fp (x)φ(x)dx → f (x)φ(x)dx
K K
16
where K = supp(φ).
Since fp φ → f φ pointwise, and fp φ and f φ are continuous, then fp φ → f φ uniformly

in K.
1.4.1 Distributions of different forms
This leads us to the next question: are there any distributions F ∈ D ′ (Ω) which are
not of the form Ff for some f ∈ C(Ω)?
Definition 1.36 Let y ∈ Ω. Define the Dirac delta functional,
δy (φ) := φ(y). (1.7)
Problem 1.37 Show that the Dirac delta functional is a distribution.
Proof:
1. Linearity:
δy (λφ + µψ) = (λφ + µψ)(y)

= λφ(y) + µψ(y)
= λδy (φ) + µδy (ψ)
2. Continuity: if φp → φ in D, then, in particular, φp (y) → φ(y), so that
δy (φp ) = φp (y) → φ(y) = δy (φ)
Problem 1.38 Show that there is no f ∈ C(Ω) such that δy = Ff .
Solution Without loss of generality, take y = 0. Assume δ = Ff with f ∈ C(Ω) and

take a function φε ∈ D(Ω) such that


0 ≤ φε (x) ≤ 1 in Ω
φε (x) = 1 in Bε


φε (x) = 0 outside B2ε
17
δ(φε ) = φε (0) = 1 so

Z

1 = δ(φε ) = Ff (φε ) = f (x)φε (x) dx

Ω
Z

= f (x)φε (x) dx

B
Z2ε
≤ |f (x)| |φε (x)| dx
B2ε
Z
≤ |f (x)| dx
B2ε
Let f ∈ C(Ω). Since f is continuous, it is bounded in any compact, i.e.
|f (x)| ≤ C in B1
(where we’ve picked the radius of the ball to be 1 fairly arbitrarily). Therefore
Z
1≤C dx
B2ε
= C · (vol. of ball in n-dim. case) · (2ε)n

→ 0 as ε → 0
which is a contradiction. X
Now for more examples of distributions:
Example 1.39 Let Ω = R3 . Consider the plane, say x3 = a. Let h(x1 , x2 ) be a

continuous function of two variables, i.e. h ∈ C(R2 ). Define
Z
Fa,h (φ) = h(x1 , x2 )φ(x1 , x2 , a) dx1 dx2 ,
i.e. the integral over the plane x3 = a:

Z
= h(x)φ|R dA
plane
x3 =a
Then Fa,h ∈ D ′(R3 ) (we could denote this by hδa (x3 )).
Example 1.40 More generally, let S ⊂ Ω ⊂ R3 be a surface and h a continuous

function on S. Define Z
FS,h (φ) = hφ dAS ,
S
where dAS is the area element on S, known as the surface integral from MATH1402.
18
Example 1.41 Let now, L be a curve in Ω ⊂ R2 or Ω ⊂ R3 and h be some continuous

function on L. We define Z
FL,h (φ) = hφ dℓ
L
where dℓ is the length element on L.
Problem 1.42 Show that Fa,h ∈ D ′ (R3 ), as well as FS,h , FL,f ∈ D ′ (Ω).
1.5 Support of a Distribution
We know what supp(f ) is when f ∈ C(Ω): this is the closure of the set of points where
f (x) 6= 0. This definition uses the values of the function at all points. Can we find
supp(f ) looking only at Z
Ff (φ) = f (x)φ(x)dx?
Ω
To this end, observe that if supp(φ) ⊂ N (f ), then
Ff (φ) = 0.
Moreover, by the Du Bois-Reymond lemma, if U ⊂ Ω is open and, for any φ ∈ D(Ω)

with supp(φ) ⊂ U, Z
Ff (φ) = f (x)φ(x)dx = 0,
Ω
then f = 0 in U, i.e. U ⊂ N (f ). This makes it possible to find N (f ) and, therefore,

supp(f ), looking only at Ff .
This also motivates the following definition:
Definition 1.43 Let F ∈ D ′ (Ω). We say that F = 0 in an open U ⊂ Ω (we can

denote this F |U ) if
F (φ) = 0, φ ∈ D(Ω), supp(φ) ⊂ U. (1.8)
This is a very natural definition: that, the distribution F is zero in U if, for whichever
function φ ∈ D = C0∞ we put into it, the result is zero so long as the support of φ (i.e.
the non-zero bits) is contained entirely in U.
Definition 1.44 The null-set N (F ) is the largest open set U in Ω such that (1.8) is
valid when supp(φ) ⊂ U (i.e. the largest open set where F = 0).
19
Definition 1.45 The support supp(F ) is the closed set
supp(F ) = Ω \ N (F ).
The consistency of this definition is based on the fact (which I shan’t prove) that
F = 0 in U1 and F = 0 in U2 =⇒ F = 0 in U1 ∪ U2 . (1.9)
Problem 1.46 Show that supp(δy ) = {y}
Proof: If supp(φ) ⊂ Ω \ {y}, then φ(y) = 0. Therefore,
δy (φ) = φ(y) = 0.
Thus
(Ω \ {y}) ⊂ N (δy ).
Now let’s assume that y ∈ N (δy ). Then N (δy ) = Ω since we’ve just added {y} to
(Ω \ {y}), and
N (δy ) = Ω ⇐⇒ δy (φ) = 0 ∀φ ∈ D(Ω).
But Proposition 1.7 says that there exists φ ∈ D(Ω) such that φ(y) = 1. But then
δy (φ) = φ(y) = 1, which is clearly a contradiction.
Hence y ∈
/ N (δy ).
Problem 1.47 Suppose S ⊂ Ω and h is a continuous function on S. Show that

supp(FS,h ) ⊂ S.
Proof: Z
FS,h (φ) = h(x)φ|S (x) dA
S
Look at the supports:

Ω \ S ⊂ N (FS,h)
since F is integrating over an area S (if you take away S you’re left with nothing).
Hence, taking complements,
S ⊃ supp(FS,h )
Actually, supp(FS,h ) = S ∩ supp(h).
Problem 1.48 ∗ Show that supp(FL,h ) ⊂ L.
20
Proof: Z
FL,h (φ) = h(x)φ|L (x) dℓ
L
Look at the supports:

Ω \ L ⊂ N (FL,h )
since F is integrating over a line L (if you take away L you’re left with nothing).
Hence, taking complements,
L ⊃ supp(FL,h )
1.5.1 Distributions with compact support (E ′ )
The subclass of distributions with compact support is denoted by
E ′ (Ω) ⊂ D ′ (Ω).
Definition 1.49 F ∈ E ′ (Ω) if supp(F ) is compact in Ω.
The importance of E ′ (Ω) is that we can unambiguously define F (ψ) for any ψ ∈ C ∞ (Ω)
when F ∈ E ′ (Ω).
Indeed, let supp(F ) ⊂ K1 ⊂ K2 ⋐ Ω, where K1 and K2 satisfy the conditions of

Proposition 1.7.
Let
1 if x ∈ K1
φ(x) ∈ D(Ω) =
0 if x ∈ Ω \ K2
(the existence of such φ is guaranteed by Proposition 1.7).
We now define
F (ψ) := F (φψ). (1.10)
Note that this definition is independent of the choice of φ satisfying
φ(x) = 1 x ∈ supp(F ).
Indeed, if φb is another function with this property, then

b = F ((φ − φ)ψ)
F (φψ) − F (φψ) b = 0,
b
since supp(F ) ∩ supp((φ − φ)ψ) = ∅.
Definition 1.50 We say that Fp → F in E ′(Ω) if Fp → F in D ′ (Ω). That is to

say, Fp (φ) → F (φ) for any φ ∈ D(Ω) and there is a compact K ⋐ Ω such that
supp(Fp ), supp(F ) ⊂ K.
21
1.6 Differentiation of Distributions
Take a function f ∈ C ∞ (Ω). Then, for any multi-index β, define
∂ |β| f
g := ∂ β f =
∂xβ1 1 · · · ∂xβnn
Obviously g ∈ C ∞ (Ω) as well. So

Z
Fg (φ) = g(x)φ(x) dx
Ω
Z
= ∂ β f (x)φ(x) dx
Ω
Z
|β|
= (−1) f (x) ∂ β φ(x) dx (1.11)
| {z }
Ω ∈D(Ω)
|β| β
= (−1) Ff (∂ φ) (1.12)
where line 1.11 has been achieved by integrating by parts |β| times. You can think of
this process like repeating this:
Z

φ(x)∂ f (x) − ∂φ(x) ∂ β−1 f (x) dx
β−1
| {z }
This disappears since Ω
f, φ = 0 at the boundaries
Therefore:
Definition 1.51 For F ∈ D ′(Ω) and a multi-index β, then ∂ β F ∈ D ′ (Ω) is the

distribution of the form
∂ β F (φ) = (−1)|β| F (∂ β φ)
and is called the β-derivative of the distribution F .
Let us just show that this definition is OK.
First, since φ has infinitely many derivatives, then ∂ β φ also has infinitely many deriv-
atives. Moreover, since φ = 0 on the open set N (φ), then (−1)|β| ∂ β φ = 0 on N (∂ β φ),
which is larger than N (φ) and
supp(∂ β φ) ⊂ supp(φ).
This implies that supp(∂ β φ) is a compact in Ω. Therefore, line 1.12 above is well-
defined.
Now, to show that (1.12) is a distribution, we should prove the relations of linearity
(1.5) and continuity (1.6).
22
1. Linearity:
∂ β F (λ1 φ1 + λ2 φ2 ) = (−1)|β| F (∂ β [λ1 φ1 + λ2 φ2 ])

= (−1)|β| F (λ1 ∂ β φ1 + λ2 ∂ β φ2 )
= (−1)|β| λ1 F (∂ β φ1 ) + (−1)|β| λ2 F (∂ β φ2 )
= λ1 ∂ β F (φ1 ) + λ2 ∂ β F (φ2 ).
2. Continuity: by the definition of convergence, if φp → φ, then
∂ β φp → ∂ β φ
Thus,
∂ β F (φp ) = (−1)|β| F (∂ β φp ) → (−1)|β| F (∂ β φ) = ∂ β F (φ).
Thus, for distributions associated with smooth functions, the derivatives of these dis-
tributions are associated with the derivatives of the corresponding functions. This
observation shows that our definition 1.51 is a natural extension of the notion of dif-
ferentiation from functions to distributions.
Problem 1.52 The Heaviside step function, Θ(x) or H(x), for x ∈ R, is given by

0 if x < 0
H(x) = .
1 if x ≥ 0
Show that H ′ = δ, where H represents the distribution derived from the function
H(x).
Proof:
H ′ (φ) = (−1)H(φ′ )
Z ∞
=− φ′ (x) dx
0
= φ(0) [since φ(∞) = 0 because of compact support]
= δ(φ)
Problem 1.53 The sign (or signum) function, sgn(x), for x ∈ R is given by

 −1 if x < 0
0 if x = 0 .

1 if x > 0
Show that sgn′ = 2δ.
23
Proof:
sgn′ (φ) = (−1) sgn(φ′ )

Z ∞
=− sgn(x)φ′ (x) dx
−∞
Z 0 Z ∞
′ ′
=− sgn(x)φ (x) dx + sgn(x)φ (x) dx
−∞ 0
Z 0 Z ∞
′ ′
=− − φ (x) dx + φ (x) dx
−∞ 0

= − − [φ(x)]0−∞ + [φ(x)]∞ 0
= − (− (φ(0) − φ(−∞)) + (φ(∞) − φ(0)))
= 2φ(0) [since φ(∞) = φ(−∞) = 0 because of compact support]
= 2δ(φ)
Problem 1.54 Show that, for any multi-index α and any y ∈ Ω,
∂ α δy (φ) = (−1)|α| (∂ α φ)(y).
Solution
∂ α δy (φ) = (−1)|α| δy (∂ α φ)
= (−1)|α| (∂ α φ)(y)
Note, in a matter unrelated to the problem above, that if F ∈ D ′ is a distribution and

ψ ∈ C ∞ , then
(ψF )(φ) = F (ψφ), ψ ∈ D(Ω)
1.7 Convolution
An important operation acting on distributions is convolution. To define it, we start,

as usual, with continuous functions.
24
Definition 1.55 Let
f ∈ C0 (Rn )
g ∈ C(Rn )
i.e. f has compact support, supp(f ) ⊂ K ⋐ Rn . Then define

Z
h(x) = (f ∗ g)(x) := f (x − y)g(y) dy
n
ZR
= f (y)g(x − y)dy
Rn
= (g ∗ f )(x),
which is called the convolution of f and g. Clearly, f ∗ g ∈ C(Rn ).
Consider the corresponding distribution Fh ,

 
Z Z
Fh (φ) = Ff ∗g (φ) =  f (x − y)g(y)dy  φ(x)dx
Rn n
R 
Z Z
=  f (x − y)φ(x)dx g(y)dy
Rn n
R 
Z Z
(z := x − y) =⇒ =  f (z)φ(z + y)dz  g(y)dy
Rn n
R 
Z Z
(x := z) =⇒ =  f (x)φ(x + y)dx g(y)dy
Rn n
R 
Z Z
(φy (x) := φ(x + y)) =⇒ =  f (x)φy (x)dx g(y)dy
Rn Rn
Z
= Ff (φy ) g(y)dy
| {z }
Rn Φ(y)
= Fg (Φ)
= Fg (Ff (φy ))
Here, for any y ∈ Rn , we denoted by φy the translation of φ,
φy (x) = φ(x + y).
Therefore,
Ff ∗g (φ) = Fg (Ff (φy )) . (1.13)
It turns out that formula (1.13) makes sense for any F ∈ E ′(Rn ), G ∈ D ′ (Rn ).
25
Definition 1.56 For F ∈ E ′(Rn ), G ∈ D ′ (Rn ), we can define the convolution, H =

F ∗ G ∈ D ′ (Ω) by
(F ∗ G)(φ) = G(F (φy )).
We will now show that this definition is (1) well-defined, (2) linear and (3) continuous.
1. Well-defined:
(a) Support:
Since F ∈ E ′(Rn ),
=⇒ supp(F ) ⋐ Rn
i.e. ∃b > 0 such that Rn \ Bb (0) ⊂ N (F ), where Bb (0) is a ball of radius b
centred at 0.
Since φ ∈ D(Rn ),
=⇒ ∃a > 0 such that supp(φy ) ⊂ Ba (y)
i.e. supp(φ) ⊂ Ba (0)

Thus if |y| > b + a,
supp(F ) ∩ supp(φy ) = ∅
=⇒ F (φy ) = 0
Denote Φ(y) = F (φy ). Then supp(Φ) ⊂ Bb+a (0).
(b) Is Φ(y) ∈ C ∞ (Rn )?
We will show that Φ is infinitely differentiable by proving that
∂ α Φ = F (∂ α φy ),
using induction on |α|.
Assume
∂ α Φ = F (∂ α φy ), |α| ≤ m
and let us show that
∂ β Φ = F (∂ β φy ), |β| = m + 1.
Let β = α + ei , where
ei = (0, . . . , 0, 1, 0, . . . , 0)
with the 1 in the ith place.
By our inductive hypothesis,
∂i ∂ α Φ = ∂i F (∂ α φy )
but we need to prove that the right hand side exists. In order to do this,
we need to show that:
ψs −−→ ψ in D(Ω)
s→0
26
Where
∂ α φy+sei − ∂ α φy
ψs (y) = , ψ(y) = ∂ β φy
s
So we need, for any multi-index γ:
∂ γ ψs −−→ ∂ γ ψ
s→0
For |γ| = 0, we have:
∂ α φy+sei − ∂ α φy ∂ α φ(x + y + sei ) − ∂ α φ(x + y)

ψs (y) = =
s s
−−→∂ α+ei φ(x + y)
s→0
=∂ β φy = ψ(y)
Now, for an arbitrary multi-index γ, we want to show that:

α α

γ γ ∂ φ(x + y + sei ) − ∂ φ(x + y)
∂ ψs = ∂ −−→ ∂ γ+β φ(x + y) = ∂ γ ψ
s s→0
Well,
∂ γ+α φ(x + y + sei ) − ∂ γ+α φ(x + y)

= ∂i ∂ γ+α φ(x + y)
s
= ∂ γ+α+ei φ(x + y)
= ∂ γ+β φ(x + y)
So ψs −−→ ψ in D(Ω), which means that F (ψs ) −−→ F (ψ), now:

s→0 s→0
F (∂ α φy+sei ) − F (∂ α φy )
∂ β Φ = ∂i F (∂ α φy ) = lim
s→0
α y+sesi
∂ φ − ∂ α φy
= lim F
s→0 s
= lim F (ψs )
s→0
= F (ψ) = F (∂ β φy )
So then, by induction, for any multi-index α
∂ α F (φy ) = F (∂ α φy ) (1.14)
Therefore F (φy ) ∈ C ∞ (Rn ). Thus G(F (φy )) is well defined.
2. Linearity: Clearly, the functional
φ 7→ G (F (φy ))
is linear.
27
3. Continuity: This is also a pain, but it’s the same type of pain.
Let φk → φ in D. We need
(F ∗ G)φk → (F ∗ G)(φ)
i.e.
G(F (φyk )) −−−→ G(F (φy )).
k→∞
It is sufficient to show that
F (φyk ) = Φk (y) → Φ(y) = F (φy ) in D.
So let’s check support and differentials
(a) Check support: Let us show that supp(Φk ), supp(Φ) lie in the same ball.
As φk → φ in D, ∃a > 0 such that supp(φk ), supp(φ) ⊂ Ba .
Then supp(Φk ), supp(Φ) ⊂ Bb+a (where supp(F ) ⊂ Bb ).
(b) Show ∂ α Φk (y) → ∂ α Φ(y)
i.e. show F (∂ α φyk ) → F (∂ α φy )
Since φk → φ in D(Rn ), this implies that φyk → φy in D(Rn ).
Then by Proposition 1.7, ∂ α φyk → ∂ α φy in D(Ω) ∀α.
Therefore F (∂ α φyk ) → F (∂ α φy )
Congratulations!!! You did it!
Proposition 1.57 Convolution is a continuous map from E ′(Ω) × D ′ (Ω) to D ′ (Ω),

i.e. if Fp → F in E ′ (Ω) and Gp → G in D ′ (Ω), then
Fp ∗ Gp → F ∗ G.
We won’t prove this as it’s too difficult.
Problem 1.58 ∗ If Fp → F in E ′ (Ω) and G ∈ D ′ (Ω), show that
Fp ∗ G → F ∗ G
Solution
(Fp ∗ G)(φ) = G(Fp (φy ))
G is continuous, and Fp → F in E ′ (Ω), so Fp (φy ) → F (φy ), and hence
−−−→ G(F (φy ))

p→∞
= (F ∗ G)(φ)
28
Lemma 1.59
F ∗ G = G ∗ F. (1.15)
Remark 1.60 We prove (1.15) later, although it’s obvious for distributions from the
functions: if F = Ff , G = Fg ,
Z
G ∗ F = Fh h(x) = g(x − y)f (y) dy
Z
F ∗ G = Fbh b
h(x) = f (x − y)g(y) dy
which are clearly equal.
Although the convolution of distribution is usually defined only for Ω = Rn , in some

cases it may be defined for general Ω. In particular, when 0 ∈ Ω, F ∗ G is defined for
F = ∂ α δ, where α is arbitrary.
Problem 1.61 Show that

(∂ α δ) ∗ G = ∂ α G.
Solution
[(∂ α δ) ∗ G](φ) = G[(∂ α δ)(φy )]

= G[(−1)|α| δ(∂ α φy )]
= (−1)|α| G[δ(∂ α φy )]
= (−1)|α| G[∂ α φ]
= (∂ α G)(φ)
Now we will show that we can move about derivatives:
Proposition 1.62 Let F ∗ G ∈ D ′ (Ω) be the convolution of two distributions. Then,

for any multi-index α,
∂ α (F ∗ G) = (∂ α F ) ∗ G = F ∗ (∂ α G).
Proof: By the definition of differentiation (definition 1.51),
∂ α (F ∗ G)(φ) = (−1)|α| (F ∗ G)(∂ α φ)

= (−1)|α| G(F (∂ α φy ))
by the definition of convolution (definition 1.56).
To proceed, notice that

∂ α φ(x + y) ∂ α φ(x + y)
∂ α φy (x) = =
∂xα ∂y α
29
Therefore

α y ∂ α φy
F (∂ φ ) = F
∂y α
∂α
= (F (φy ))
∂y α
by equation 1.14. So, proceeding,
∂ α (F ∗ G)(φ) = (−1)|α| (F ∗ G)(∂xα φ)

= (−1)|α| G(F (∂xα φy ))
= (−1)|α| G(F (∂yα φy ))
= (−1)|α| G(∂yα F (φy ))
= ∂yα G(F (φy ))
= (F ∗ ∂ α G)(φ)
Note the importance of which variable we differentiate with respect to. And since
F ∗ G = G ∗ F , swapping F and G holds.
Corollary 1.63
∂ β+γ (F ∗ G) = ∂ β (F ) ∗ ∂ γ (G)
1.8 Density
Now for a little notation
Notation 1.64
F : C(Ω) 7→ D ′ (Ω) F (f ) = Ff
As we know F (C(Ω)) 6= D ′ (Ω). However, we have the following density result
Theorem 1.65
cl (F (C0(Ω))) = D ′ (Ω),
i.e. for any F ∈ D ′ (Ω), Ω ⊂ Rn , there is a sequence fp ∈ D(Ω) such that
Ffp −−−→ F in D ′ (Ω) (1.16)

p→∞
Sketch of proof: We will prove this theorem in 2 parts:
1. We will show that the distributions with compact support, E ′(Ω) are dense in
the space of all distributions D ′ (Ω), i.e. we show that cl(E ′ (Ω)) = D ′ (Ω).
30
2. We will show that the space of distributions of compact support due to continu-
ous functions with compact support is dense in the space of distributions with
compact support, i.e. we show that cl (F (C0(Ω))) = E ′(Ω). We will do this in
three steps:
(a) For F , a distribution with compact support, we will construct a sequence
of distributions converging to F in D ′ (Ω).
(b) Then we will show that in fact these distributions all have compact support,
lying in the same compact, so that they converge to F in E ′ (Ω).
(c) Finally we will show that they are in fact all distributions due to continuous
functions with compact support.
Once we have proven these facts we see that F (C0 (Ω)) is dense in E ′ (Ω) which is, in
turn, dense in D ′ (Ω). Recall, by the nature of the closure operation, this then means
that cl(F (C0(Ω))) = D ′ (Ω). (Since D ′ (Ω) = cl(E ′ (Ω)) = cl(cl(C0 (Ω))) = cl(C0 (Ω)) as
the closure of a closed set is itself).
1. We will show that any distribution can be approximated by distributions with

compact support, i.e.
∀F ∈ D ′ (Ω), ∃Fp ∈ E ′ (Ω) s.t. Fp → F
First, let Kp , p = 1, 2, . . . , be a sequence of compacts exhausting Ω, i.e.

∞
[
Kp ⋐ Ω, Kp ⊂ int(Kp+1 ), Kp = Ω.
p=1
In fact we can define them as

1
Kp = x ∈ Ω : |x| ≤ p, dist(x, ∂Ω) ≥
p
Now let χp ∈ D(Ω) be functions with the properties described by Proposition

1.7, where K1 = Kp and K2 = Kp+1 , i.e.

1 if x ∈ Kp
χp (x) =
0 if x ∈ Ω \ Kp+1
Let F ∈ D ′ (Ω), then define Fp by:
Fp = χp F
Fp (φ) = χp F (φ) = F (χp φ).

Note that
supp(Fp ) ⊂ Kp+1. (1.17)
Indeed, if φ ∈ D(Ω) and supp(φ) ∩ Kp+1 = ∅, then
Fp (φ) = χp F (φ) = F ( χp φ ) = 0.
|{z}
0
31
Also, Fp → F . To prove this, we want to show that
F (χp φ) = Fp (φ) −−−→ F (φ) ∀φ ∈ D(Ω)

p→∞
which clearly it does. Indeed, for large p, K ⊂ Kp

Proof:
dist(K, ∂Ω) = d > 0
since K ⋐ Ω. As K is a compact, K ⊂ BR , for some radius R > 0.
Hence for p > max(R, d1 ), K ⊂ Kp .
2. (a) Any distribution with compact support can be approximated by distribu-
tions of functions with compact support, i.e.
∀F ∈ E ′(Ω), ∃fp ∈ D(Ω) s.t. Ffp → F ∈ E ′ (Ω)
Let us introduce a function χ(x) ∈ D(Rn ) such that

i. supp(χ) ⊂ B1 (0)

1 if x ∈ B1/2 (0)
ii. χ(x) =
0 if x ∈ Ω \ B1 (0)
Z
iii. χ(x)dx = 1
Rn
iv. χ = χ(|x|) (i.e. χ is a radial function)
The existence of such χ follows easily from proposition 1.7.
Now consider, for sufficiently small ε > 0, a function χε (x) ∈ D(Rn ) such
that
i. supp(χε ) ⊂ Bε (0)

ε 1 if x ∈ Bε/2 (0)
ii. χ (x) =
0 if x ∈ Ω \ Bε (0)
Z
iii. χε (x)dx = 1
Rn
1 x
iv. χε (x) = χ
εn ε
Now define
F ε := χε ∗ F ∈ E ′ (Ω) (1.18)
then for φ ∈ D(Ω), by the definition of convolution,
F ε (φ) = (χε ∗ F )(φ) = F (χε (φy ))

= Fχε (φy )
Z
:= χε (x)φ(x + y) dx ∈ D(Rn )
Rn
and we’ll show now that
χε (φy ) → φ(y) ∈ D(Ω).
32
First note that

Z
ε y
χ (φ ) = χε (x)φ(x + y) dx
|x|≤ε
Z
= χε (x) [φ(y) + φ(x + y) − φ(y)] dx
|x|≤ε
Z Z
ε
= φ(y) χ (x) dx + χε (x) [φ(x + y) − φ(y)] dx
|x|≤ε |x|≤ε
Z
= φ(y) + χε (x) [φ(x + y) − φ(y)] dx
|x|≤ε
Z
As χε (x) dx = 1.
|x|≤ε
So we can write
Z
ε y
χ (φ ) = φ(y) + χε (x)[φ(x + y) − φ(y)] dx
|x|≤ε
Observe that
sup |φ(x + y) − φ(y)| −−→ 0 ∀y ∈ Ω

|x|≤ε ε→0
i.e. pointwise in Ω, which means that φ(x + y) − φ(y) −−→ 0 uniformly on

ε→0
any K ⋐ Ω. Hence
Z
ǫ ε y
ψ (y) = χ (φ ) = φ(y) + χε (x) [φ(x + y) − φ(y)] dx
| {z }
|x|≤ε →0 ∀y∈Ω
−−→ φ(y) uniformly on any compact

ε→0
i.e.
ψ ǫ (y) = χε (φy ) → φ(y) ∀y ∈ Ω
and uniformly on any compact.
And the same goes for the derivative. If we replace χε (φy ) by ∂ α χε (φy ), we
use the same trick and
∂ α χε (φy ) → ∂ α φ(y) ∀y ∈ Ω, α
This means that χε (φy ) = ψ ε → φ in E(Ω) = C ∞ (Ω).
So (χε ∗ F )(φ) = F (χε (φy )) −−→ F (φ) since F ∈ E ′(Ω) and χε (φy ) → φ in
ε→0
E(Ω)
So χε ∗ F → F in D ′ (Ω).
33
(b) Now we show that χε ∗ F ∈ E ′ (Ω), i.e. We show that χε ∗ F has compact
support. First, we find supp(χε (φy )).
Suppose d(y, supp(φ)) > ε then d(x + y, supp(φ)) > 0 for |x| ≤ ε, which
means that x + y ∈
/ supp(φ), so:
Z
ε y
χ (φ ) = χε (x)φ(x + y) dx = 0 ∀y s.t. d(y, supp(φ)) > ε
|x|≤ε
So, supp(χε (φy )) ⊂ {y : d(y, supp(φ)) ≤ ε}, i.e. supp(χε (φy )) lies in an
ε-vicinity of supp(φ).
Now we claim that supp(χε ∗ F ) ⊂ {x : d(x, supp(F )) ≤ ε} = supp(F )ε , i.e

supp(χε ∗ F ) lies in an ε-vicinity of supp(F ).
Let φ be such that:

supp(φ) ∩ supp(F )ε = ∅
Then:
d(supp(φ), supp(F )) > ε
i.e. the ε-vicinity of supp(φ) lies outside of supp(F ). Thus:
supp(χε (φy )) ∩ supp(F ) = ∅
So:
(χε ∗ F )(φ) = F (χε (φy )) = 0 ∀φ s.t. supp(φ) ∩ supp(F )ε = ∅
Which means that:

supp(χε ∗ F ) ⊂ supp(F )ε
Now take ε0 = 21 d(supp(F ), ∂Ω) so that supp(F )ε0 is compact in Ω. Then:
∀ε ≤ ε0 supp(χε ∗ F ) ⊂ supp(F )ε ⊂ supp(F )ε0 ⋐ Ω
So that supp(χε ∗ F ) ⋐ Ω which means that:
χε ∗ F ∈ E ′ (Ω)
(c) Now we show that, for ε < 12 d(supp(F ), ∂Ω), χε ∗ F ∈ C0∞ (Ω)
In other words, there are functions hε ∈ C0∞ (Ω) s.t. χε ∗ F = Fhε
34
We approach this problem as we always do; in order to construct hε we first

look at the case when F = Ff , then:
(χε ∗ Ff )(φ) = Ff (χε (φy ))

Z Z
ε
= f (y) χ (x)φ(x + y)dx dy
Z Z
ε
= f (y)χ (z − y)φ(z)dz dy
| {z }
z=x+y,dz=dx
Z Z
ε
= f (y)χ (z − y)φ(z)dy dz
Z Z
ε
= φ(z) χ (z − y)f (y)dy dz
= Fhε (φ)
Z
Where: h (z) = f (y)χε(y − z)dy
ε
= Ff (χε,−z )
Where we have denoted χε,−z (x) = χε (x − z). Now we generalise this to

the case when F is any distribution, by trying to show that:
If hε = F (χε,−z ) ∈ C0∞ (Ω), then χε ∗ F = Fhε
Now observe that:

Z Z
ε y ε
χ (φ ) = χ (x)φ(x + y)dx = χε (z − y)φ(z)dz
X
= lim χε (zi − y)φ(zi )δ n
δ→0
i
:= lim Rδε (y)
δ→0
Also that:
Z Z
∂yα χε (φy ) = ∂yα ε
χ (z − y)φ(z)dz = ∂yα χε (z − y)φ(z)dz
X
= lim ∂yα χε (zi − y)φ(zi )δ n
δ→0
i
X
= lim ∂yα χε (zi − y)φ(zi )δ n
δ→0
i
α ε
= lim ∂y Rδ (y)
δ→0
35
Thus
χε (φy ) = lim Rδε in D(Ω)

δ→0
=⇒ (χ ∗ F )(φ) = F (χε (φy ))
ε
= lim F (Rδε )
δ→0
!
X
= lim F χε (zi − y)φ(zi )δ n
δ→0
i
!
X
= lim F (χε,−zi ) φ(zi )δ n
δ→0
i
Z
= F (χε,−z )φ(z)dz
Z
= hε (z)φ(z)dz hε (z) = F (χε,−z ) ∈ C0∞ (Ω)
= Fhε (φ)
These three parts prove that cl(F (C0∞ (Ω)) = E ′(Ω)
So we see that cl(F (C0∞ (Ω)) = D ′ (Ω) since cl(E ′ (Ω)) = D ′(Ω).
Problem 1.66 ∗ Prove that χε −−→ δ (where χε represents the distribution from
ε→0
the function χε defined in part 2a above).
Solution
Z
Fχε (φ) = χε (x)φ(x) dx ∈ D(R2 )
R2
and since χε (x) = 0 outside of Bε ,

Z
= χε (x)φ(x) dx
Bε
Z
= χε (x)φ(x) dx + φ(0) − φ(0)
Bε
R ε
R
and since φ(0) = φ(0) Bε
χ dx = Bε
χε φ(0) dx,
Z
= χε (x) [φ(x) − φ(0)] dx + φ(0)
Bε
and as ε → 0, the integral → 0, leaving us with
→ φ(0)
Hence χε → δ. X
36
Now we return to a lemma from earlier:
Lemma 1.59
F ∗ G = G ∗ F. (1.15)
Proof: Let Ffp → F , where fp ∈ D(Ω) and Ggp → G, where gp ∈ C(Ω). We know,
by remark 1.60, that:
Ffp ∗ Ggp = Ggp ∗ Ffp
and that, by proposition 1.57,
Ffp ∗ Ggp −−−→ F ∗ G

p→∞
so
F ∗ G = lim (Ffp ∗ Ggp ) = lim (Ggp ∗ Ffp ) = G ∗ F.

p→∞ p→∞
1.9 Integration of Distributions
We complete this section with the notion of integration of distributions with respect
to some parameter τ = (τ1 , . . . , τn ) ∈ A ⊂ Rn . Let Fτ ∈ D ′ (Ω) be continuous with
respect to τ , i.e. Fτ → Fτ 0 , in D ′ (Ω), when τ → τ 0 . Recall that this means that,
Fτ (φ) → Fτ 0 (φ), as τ → τ 0 ; φ ∈ D(Ω).
Then, if g ∈ C0 (A), we define a distribution Gg ∈ D ′ (Ω) by

Z
Gg = g(τ )Fτ dτ .
A
Z
Gg (φ) = g(τ )Fτ (φ)dτ . (1.19)
A
We can check that if

Fτ = Ff (τ )
where f (τ ) = f (x, τ ) ∈ C(Ω × A)
then
Gg = Fh
Z
h(x) = g(τ )f (x, τ ) dτ .
A
37
Anyway, now let’s consider A = Ω = Rn .
As a particular case, consider the operation of translation. As usual, we start with

f ∈ C(Ω), which we then continue by 0 on the whole Rn . Let y ∈ Rn . By translation
of f by y we understand the operation
Ty : f 7→ fy , fy (x) = f (x − y).
Let us consider the corresponding distribution Ty Ff = Ffy ,

Z
Ty Ff (φ) = f (x − y)φ(x) dx
ZΩ
= f (x − y)φ(x) dx
n
ZR
= f (x)φ(x + y) dx
Rn
= Ff (φy ). (1.20)
We would like to use formula (1.20) to define the translation of distributions. However,
in general, φy ∈
/ D(Ω) but just in C ∞ (Ω). Thus, assuming that F ∈ E ′(Ω) we find the
following definition
Definition 1.67 Let F ∈ E ′ (Ω), y ∈ Rn . Then,
Ty F (φ) = F (φy ), φ ∈ D(Ω), (1.21)
and we continue it by 0 outside Ω.
There are a few different notations for translation, one of Prof. Kurylev’s favourites is
Ty F = F (· − y), which the editors wish he wouldn’t use because it sucks. Another is
Ty F = Fy .
Example 1.68
Ty (δ) = δy
Proof:
(Ty δ)(φ) = δ(φy )

= φy (0)
= φ(y)
= δy (φ)
Observe that the operation of translation Ty F is a distribution in D ′ (Rn ), which de-

pends continuously on y:
Ty F → Ty0 F as y → y0 .
38
Is it true that
(Ty (F ))(φ) = F (φy ) −
→ (Ty0 F )(φ) = F (φy0 )?
?
Yes it is because if y → y0 ,
φy = φ(x + y) → φ(x + y0 ) = φy0 in D(Rn )
And clearly
∂ α φ(x + y) −−−→ ∂ α φ(x + y0 ) ∀α ∀x ∈ Rn
y→y0
since the supports lie in the same compact (see the compacts of φy and φy0 moving
towards each other as y → y0 ).
Thus, if g(y) ∈ C0 (Rn ), we can define using equation 1.19

Z
Gg = g(y)(Ty F ) dy
Rn
Z
Gg (φ) = g(y)F (φy ) dy
Rn
Problem 1.69 Show that Gg = Fg ∗ F
Problem 1.70 ∗ Let f (x) = 43 (1 − x2 ) for |x| ≤ 1, f (x) = 0 for |x| > 1, x ∈ R.
Denote fn (x) = nf (nx), n = 1, 2, . . . and let Fn ∈ D ′ (R) be a distribution given by
Z ∞
Fn (φ) = fn (x)φ(x) dx, φ ∈ C0∞ (R).
−∞
Show that Fn → δ0 , as n → ∞, in D ′ (R).
Solution
Z ∞
Fn (φ) = fn (x)φ(x) dx
−∞
and
1 because
f (x) has compact support on [−1, 1], f (nx) has compact support on
1
− n , n , and fn (x) = nf (nx),
Z 1/n
= fn (x)φ(x) dx.
−1/n
39
1 1
Now, by the mean value theorem,
since fn (x) does not change sign on − n , n , there
exists a point x0 ∈ − n1 , n1 such that
Z 1/n
= φ(x0 ) fn (x) dx
−1/n
Z1/n
3
= φ(x0 ) n (1 − x2 n2 ) dx
−1/n 4
1/n
3n x3 n2
= φ(x0 ) x−
4 3 −1/n
2

3n 1 n 3n 1 n2
= φ(x0 ) − + −
4 n 3n3 4 n 3n3

3n 2
= φ(x0 ) 2 ·
4 3n
= φ(x0 )

and since x0 ∈ − n1 , n1 , as n → ∞, x0 → 0, so
Fn (φ) → φ(0)
i.e.
Fn → δ0
X
Problem 1.71 ∗ Let

− 32 n3 x if |x| < 1
n
gn (x) = 1 .
0 if |x| ≥ n
Show that Gn → δ0′ , as n → ∞, in D ′ (R).
Solution Note that

d
gn (x) = fn (x)
dx
using the definition of f from Problem 1.70

d 3n 2 2
1
= 1−n x on |x| <
dx 4 n
3n 2
= − 2n x
4
3 3 1
=− n x on |x| <
2 n
′
and since Fn → δ0 , we conclude that Gn → δ0 , so long as I can show that ∂Hn → ∂H
for some distribution H. And I can do this by pointing out that
∂Hn (φ) = Hn (−∂φ)
−−−→ H(−∂φ)
n→∞
= ∂H(φ)
40
since Hn = Fn and H = δ0 . X
Problem 1.72 Let f ∈ C(Rn ) satisfy

Z Z
f (x)dx = 1, |f (x)|dx < ∞.
Rn Rn
Let
fp (x) = pn f (px), p = 1, 2, . . . .
Show that fp → δ, as p → ∞, in D ′ (Rn ). (here fp stands for the distribution Ffp )
Solution
Z
n
Ffp (φ) = p f (px)φ(x) dx
Rn
Letting y = px, then dy = pn dx, since x is an n-dimensional vector

Z
y
= f (y)φ dy
p
Rn
Z Z
y
= f (y)φ(0) dy + f (y) φ − φ(0) dy
p
Rn Rn | {z }
→0 pointwise
Z
= φ(0) f (y) dy
Rn
| {z }
1 by definition
= φ(0)
Now, pointwise convergence is OK but we really want uniform convergence for rigour,
i.e. we want to show

Z
y
∀ε ∃P (ε) s.t. p > P (ε) =⇒ f (y) φ − φ(0) dy < ε
n p
R
The trick we’ll employ is that Z Z Z

= +
Rn BA Rn \BA
Let’s deal with the area outside the ball of radius A first. Using part of the question
setup, Z Z
|f (x)|dx = C < ∞ =⇒ ∀δ ∃A(δ) s.t. |f (x)|dx < δ
Rn Rn \BA
41
So if we define
M = max |φ(y)|
and take
ε
δ=
4M
then we have
Z Z

f (y) φ y − φ(0) dy ≤ 2M |f (y)|dy
p
Rn \BA Rn \BA
ε
≤
2
Now, inside the ball of radius A,

y A
≤ −−−→ 0
p p p→∞
which we can write analysis-style as

y ε
φ − φ(0) ≤ if p ≥ P (ε)
p 2C
ε
(picking 2C
for use later!) Therefore
Z Z
y ε ε

|f (y)| φ
− φ(0) dy ≤ |f (y)|dy =
p 2C 2
BA Rn
1.10 The Laplace Operator and Green’s Function
Let us have φα ∈ C ∞ (Ω) and F ∈ D ′ (Ω), and

X
DF = φα (x)∂ α F.
|α|≤m
Can we solve the equation DF = δy ?
If we can, then we have a distribution Gy ∈ D ′ (Ω) which we call Green’s function.

Typically Gy depends continuously on y.
Recall that the Laplacian, ∇2 , in Rn is given by

n
X
2 ∂2φ
∇ φ= .
i=1
∂x2i
42
Definition 1.73 A function (or, more generally, a distribution), Φy ∈ D ′ (Rn ) is

called Green’s function (or a fundamental solution) for the Laplacian, ∇2 , at the point
y ∈ Rn , if
−∇2 Φy = δy .
In the study of electromagnetism, the most important case is n = 3 (and, sometimes,

n = 1 or n = 2). So, let us look for Green’s function in R3 .
Lemma 1.74 The function

1
Φy (x) = (1.22)
4π|x − y|
is the fundamental solution for ∇2 in R3 .
Before proving the lemma, let us make some remarks. The notation Φy (x) for the
Green’s function is typical in physics, in mathematics we tend to use Gy (x). Note also
that this fundamental solution is a distribution associated with the function (1.22).
/ C(R3 ), as it has a singularity at x = y, however, Φy (x) ∈ L1loc (R3 ).
Note that Φy (x) ∈
Strictly speaking we should write, instead of Φy , FΦy but from now on we will identify
f as representing Ff and just write
f (φ) = Ff (φ).
Proof: We need to show that
− (∇2 Φy )(φ) = δy (φ) = φ(y). (1.23)
By definition
(∇2 Φy )(φ) = Φy (∇2 φ)

Z
∇2 φ(x)
= dx.
4π|x − y|
R3
Denote by Bε (y) a ball of radius ε centred at y and observe that

Z Z
∇2 φ(x) ∇2 φ(x)
dx = lim dx. (1.24)
4π|x − y| ε→0 4π|x − y|
R3 R3 \Bε (y)
1
Since 4π|x−y| is smooth in R3 \ Bε (y) we can integrate by parts in the above integral.
Recall the second of Green’s identities from MATH1402,
Z Z
2 2
∂f ∂g
g∇ f − f ∇ g dV = g−f dS, (1.25)
∂n ∂n
V S
where S is the surface surrounding V and n is the exterior unit normal to S.
43
We take
V = R3 \ Bε (y)
f =φ
1
g= .
4π|x − y|
Note that since φ ∈ D(R3 ), we can integrate over the infinite domain R3 \ Bε (y). Note
also that on the sphere Sε (y), the normal vector n exterior to V looks into Bε (y). At
last, note that
2 2 1
∇ g=∇ = 0. (1.26)
4π|x − y|
So Z Z
2 2
∂f ∂g
g∇ f − f ∇ g dV = g−f dS,
∂n ∂n
V S
becomes
Z Z
∇2 φ ∂φ 1 ∂ 1
dV = −φ dS (1.27)
4π|x − y| ∂n 4π|x − y| ∂n 4π|x − y|
R3 \Bε (y) Sε
Now, plugging this into equation 1.24 gives us

Z Z
∇2 φ(x) ∂φ 1 ∂ 1
lim dx = lim −φ dS
ε→0 4π|x − y| ε→0 ∂n 4π|x − y| ∂n 4π|x − y|
R3 \Bε (y) Sε
Z
1 ∂φ 1 ∂ 1
= lim −φ dS
4π ε→0 ∂n |x − y| ∂n |x − y|
Sε
Now recall that y is fixed —this is really important because it means we can do the
following substitutions
z := x − y
Then convert to spherical coordinates
(z = ρ, θ, ϕ)
So φ(x) becomes a function of φ(y + ρω), where ω is a unit vector made up of θ and
ϕ. In fact,
ω = (sin θ cos ϕ, sin θ sin ϕ, cos θ)
Observe that
∂ ∂
=−
∂n ∂ρ
Giving us Z
1 ∂φ 1 ∂ 1
lim − +φ dS
4π ε→0 ∂ρ ρ ∂ρ ρ
ρ=ε
44
Z
1 ∂φ 1 1
= lim − − φ 2 dS
4π ε→0 ∂ρ ε ε
ρ=ε
2
and dS = ε sin θ dθ dϕ which gives us that
Z
2 1 ∂φ 1 1
(∇ Φy )(φ) = − lim + φ 2 ε2 sin θ dθ dϕ
4π ε→0 ∂ρ ε ε
ρ=ε
Z ZZ
and of course = .
ρ=ε 0≤θ≤π
0≤ϕ≤2π
Now note that ZZ

∂φ 1 2
ε sin θ dθ dϕ −−→ 0
∂ρ ε ε→0
∂φ
since ∂ρ
is bounded, and from what’s left,
ZZ ZZ
1 1
− φ(y + εω) sin θ dθ dϕ −−→ − φ(y) sin θ dθ dϕ
4π ε→0 4π
θ,ϕ θ,ϕ
| {z }
vol. of sphere
of radius 1 = 4π
= −φ(y).
So we have shown that

2 1
−∇ = δy .
4π|x − y|
We want to let y = 0:
2 1
−∇ = δ0 ?
4π|x|
Well,
2 1
−∇ Ty = Ty δ0 ,
4π|x|
and noting that Ty ◦ δ α = δ α ◦ Ty ,

1
Ty −∇2 = Ty δ0 = δy ,
4π|x|
So it’s sufficient to prove that

2 1
−∇ = δ0
4π|x|
and use commutativity to solve

2 1
−∇ = δy
4π|x − y|
45
1.10.1 Motivation
Why is knowledge of Green’s function such a useful thing?
Say we want to solve a differential equation with constant coefficients. Let cα be

constants and D be the differential operator:
X
D= cα ∂ α .
|α|≤m
Suppose we wish to solve the differential equation:
DF = H
where H is a known function (or distribution) and F is our unknown. Now suppose
we know G such that DG = δ, i.e. we know the Green’s function for our differential
operator D. Now construct F = G ∗ H so that:
DF = D(G ∗ H)
X
= cα ∂ α (G ∗ H).
|α|≤m
By Proposition 1.62 and linearity

 
X
= cα ∂ α G  ∗ H
|α|≤m
=δ∗H
= H.
So we see that knowledge of the Green’s function for a differential operator, allows us
to solve a differential equation with any right hand side, by use of the comparatively
simple operation of convolution.
For example if you ever wanted to solve −∇2 F = g in Ω ⊂ R3 , then

1
F = ∗ g.
4π|x|
Observe that if g ∈ C0 (R3 ),

Z
1 g(y)
F (x) = dy.
4π |x − y|
R3
46
Chapter 2
Electrostatics
Now finally we can apply this mathematics. In this chapter we apply it to electrostatics
and in the following chapter, we’ll apply it to magnetism (which is very similar), and
in doing so, derive Maxwell’s famous equations.
2.1 Introduction to electrostatics
2.1.1 Coulomb’s law

F2←1
*
x2
q2
x1
q1

F1←2
Figure 2.1: Two charged particles exerting a force on each other
If we have two charged particles of charges q1 , q2 at points x1 , x2 ∈ R3 , then Coulomb’s

law tells us that
x1 − x2
F1←2 = kq1 q2 (2.1)
|x1 − x2 |3
1
where k is a constant to do with our choice of units (typically 4πε0
).
How do they exert a force on each other? Scientists used to think there was an ether
which gave a transfer material for charge to go through. This was later rubbished, of
course, and it turns out that when you place a charged particle in space, it propagates
a magnetic field.
If we place a charge q2 in an electric field E(x), the force exerted on the particle is
Fq2 (x) = q2 E(x).
47
Combining this with Coulomb’s law (2.1) above,
x1 − x2 (x\1 − x2 )
E(x) = kq1 = kq1
|x1 − x2 |3 |x1 − x2 |2
Electric fields add linearly.
2.1.2 Electric potential
If we have an electric field E, then we define the (scalar) electric potential Φ by
E(x) = −∇Φ(x)
(note that ∇×E = 0 and the domain is simply connected, so by what we learnt in
MATH1402, E has a potential Φ.)
Obviously Φ is defined up to a constant, so it has no physical meaning, but the

difference of potentials does, of course.
e along the
So, taking a path ℓ between our two charges at x1 and x2 , at any point x
path,
F(e
x) = qE(e
x).
e to x
The work done then getting from x e + δe x) · de
x is then dW = F(e x, so
Z
Work = W = x) · de
F(e x
ℓ
Z
=q x) · de
E(e x
ℓ
Z
= −q ∇Φ(e
x) · de
x
ℓ
= −q [Φ(x2 ) − Φ(x1 )]
= q [Φ(x1 ) − Φ(x2 )]
The convention is that Φ = 0 at infinity.
2.2 The fundamental equations of electrostatics
We are now going to derive the fundamental equations of electrostatics, first in the case
of discrete point charges, and second in the case of a continuous charge distribution.
48
Chapter 2. Electrostatics
2.2.1 The discrete case
Definition 2.1 If we have a group of electric charges qi at positions yi , for i =

1, . . . , p, then we say that the electric potential at the point x is
p
X qi
t
Φ (x) =
i=1
4πε0 |x − yi |
(the t is for total!)
Definition 2.2 The electric field created by these charges is defined as
Et (x) = −∇Φt (x)
Definition 2.3 The charge density distribution created by these charges is defined
as p
X
ρ= qi δyi
i=1
Lemma 2.4 If Φ is Green’s function and ρ is the discrete distribution as described

above, then the electrostatic potential due to the distribution ρ of charges is
ρ
Φt = Φ ∗
ε0
Proof:
p
ρ 1 1 X
Φ∗ = ∗ qi δyi
ε0 4π|x| ε0 i=1
p
1 X 1
= ∗ qi δyi
ε0 i=1 4π|x|
| {z }
What is this?
Recall that, for a distribution H,
(δz ∗ H)(φ) := H(δz φy )

= H(φy (z))
= H(φz) (since φy (z) = φz (y))
= (Tz H)(φ)
so δz ∗ H = Tz H.
So
1 qi
∗ qi δyi =
4π|x| 4π|x − yi |
49
and we get
p
X qi
= Φt (x)
i=1
4πε0 |x − yi |
2.2.2 The continuous case
Now we are given a distribution ρ ∈ E ′ (R3 ). So what is Φρ , the electrostatic potential

due to ρ?
Well, we want to use

p
X qi
i=1
4πε0 |x − yi |
but we want to make it continuous.
So consider that inside a small cube centred at xi in R3 , the approximate charge is
qi = ρ(xi ) δx δy δz.
Now take the limit as δx, δy, δz → 0.

X ρ(xi ) δx δy δz Z
1 ρ(y)
lim
= dy
δx
i
4πε0 |x − xi | 4πε0 |x − y|
δy →0
δz R3
1 ρ
= ∗
4π|x| ε0
ρ
=Φ∗
ε0
ρ
=Φ
So we get the same result as we did in the discrete case in Lemma 2.4, i.e.
ρ
Φρ = Φ ∗
ε0
2.2.3 The equations
Take the Laplacian of our potential distribution Φρ :

ρ
−∇2 Φρ = −∇2 Φ ∗
ε0
ρ
(by definition of Φ) = δ ∗
ε0
ρ
=
ε0
50
And recall that the definition of the electric field due to a particle distribution ρ is
Eρ = −∇Φρ
which gives us. . .
Definition 2.5 The fundamental equations of electrostatics are

ρ
1. ∇ · Eρ = (Gauss’ law in differential form)
ε0
2. ∇ × Eρ = 0
2.3 Divergence theorem and Stokes’ theorem
Problem 2.6 If F ∈ D ′ (Ω), where Ω ⊂ R3 , then

∂F ∂F ∂F
, , = ∇F ∈ D ′ (Ω)3 .
∂x ∂y ∂z
Show that
∇ × ∇F = 0 and ∇ · (∇F ) = ∇2 F
Taking the fundamental equations of electrostatics, let us put these into the divergence
theorem from MATH1402, for a domain V with boundary S:
Z Z
E · n dS = ∇ · Eρ dV
ρ
S V
ρ
But ∇ · Eρ = ε0
=⇒
Z
1 Q(V )
= ρ dV = (Gauss’ law in integral form)
ε0 ε0
V
where Q(V ) is the total charge in V . This is known as Gauss’ law in integral form.
Now what about putting Eρ into Stokes’ law? Recall that Stokes’ law, for some vector
F, is Z I
(∇ × F) · n dS = F · dr
S C
ρ
Substituting F = E gives us
Z I
ρ
0= (∇ × E ) ·n dS = Eρ · dr
| {z }
S 0 C
but remember that Stokes’ law is only valid for simply connected domains.
51
2.4 Finding E and Φ for different ρ
Problem 2.7 Suppose that charge distribution is a function of radius alone, i.e.
ρ = ρ(r) r 2 = x2 + y 2 + z 2 = |x|2 .
1. Find Eρ (x).
2. Then using part 1, find Φρ (x) in terms of ρ(r).
Solution 1. Because of spherical symmetry,

e r
E = E(r) = E(r)b
where b
r is a unit vector in the radial direction.
By Gauss’ law in integral form, we have
ZZ
1
E · n dS = Q(Br )
ε0
Sr
Z
1
= ρ dV
ε0
Br
Z Z
1 r ′ ′2 ′
= ρ(r )r dr dω
ε0 0
| {z } S2
:=R(r) | {z }
4π
4π
= R(r)
ε0
where ω = sin θ dθ dϕ.
But we also have
ZZ ZZ
E · n dS = E · r dS
Sr Sr
ZZ
e
= E(r) dS
Sr
e 2
= 4πr E(r)
Therefore
e R(r)
E(r) = .
ε0 r 2
e → 0 as r → 0 since R(r) ∼ r 3 .
Note that E
2. Now to find Φρ . We know by definition that
−∇Φ(r) = E
52
Therefore, setting up Φ so that it is 0 at ∞,

Z r
Φ(r) = − e ′ ) dr ′
E(r
Z ∞∞
= e ′ ) dr ′
E(r
Zr ∞
R(r ′ ) ′
= dr
r r ′2
We could be done at this point but it would be nice to integrate by parts since
R itself is an integral. So, letting our ‘first’ be R(r) and our ‘second’ be 1/r 2 ,
∞ Z ∞
−1 1
= R(r) + R′ (r ′ ) ′ dr ′
r r r r
Z ∞
R(r)
= + ρ(r ′ )r ′ dr ′
r r
Problem 2.8 ∗ Let ρ = ρ(r) in cylindrical coordinates (r, ϕ, z), i.e. r 2 = x2 + y 2 .

Find Eρ (using Gauss’ law).
Solution e r, so
By symmetry, Eρ = E(r) = E(r)b
e r
∇ · Eρ = ∇ · E(r)b
1 d e
= r E(r)
r dr
ρ
=
ε0
And so
Z r
e − E(0)
e 1
E(r) = reρ(e
r ) de
r
| {z } rε0 0
0
Z r
e 1
=⇒ E(r) = reρ(e
r ) de
r
rε0 0
e r.
and Eρ = E(r)b X
2.5 Dipoles
We now introduce the idea of dipoles.
53
Definition 2.9 If we have a charge distribution
ρ = p · ∇y δy = −p · ∇x δy p = pu
where u is a unit vector, then p · ∇δy is called the dipole of size p in the direction u
at the point y.
Recall for maths’ sake that

∂δy ∂δy ∂δy
p · ∇δy = p u1 + u2 + u3
∂x ∂y ∂z
∂δy
=p
∂u
p
Consider placing a charge s
at the point y + su, and a charge − ps at the point y. This
is, physically, a dipole.
Then the electrostatic potential Φs = Φsp,y due to the two point charges
p p
δy+su − δy
s s
is
s p 1 1
Φ = −
sε0 4π|x − (y + su)| 4π|x − y|
Now take s → 0:
p ∂
lim Φs (x) = Φy (x)
s→0 ε0 ∂uy
p ∂
=− Φy (x)
ε0 ∂ux
Now in order to find Φdp,y , we first look at:

1
∇
4π|x|
Taking each component of ∇ in turn:

!
∂ 1 1 1
p = − · 2x · 2
∂x x2 + y 2 + z 2 2 (x + y + z 2 )3/2
2
x
=− 3
|x|
!
∂ 1 y
p =− 3
∂y x2 + y 2 + z 2 |x|
!
∂ 1 z
p =− 3
∂z x2 + y 2 + z 2 |x|
54
hence

1 x
∇ =−
4π|x| 4π|x|3
But we know that for a dipole (d), of size p in direction u sitting on the point y,
1
Φdp,y = Φ ∗ (p · ∇y δy )
ε0
1
= Φ ∗ (−p · ∇x δy )
ε0
1
= − p · ∇x (Φ ∗ δy )
ε0
1
= − (p · ∇Φ) ∗ δy
ε0

1 1
=− p·∇ ∗ δy
ε0 4π|x|

1 x
=− p· − ∗ δy
ε0 4π|x|3

1 x
= p· ∗ δy
4πε0 |x|3
1 p · (x − y)
=
4πε0 |x − y|3
Since convolution with δy is the same as translation by −y
And so we get:
Definition 2.10 The potential of a dipole of size p in the direction u at a point y is
1 p · (x − y)
Φdp,y =
4πε0 |x − y|3
Now to find the corresponding electric field. Edp,y .
We know
Edp,y = −∇Φdp,y
so let’s work it out.
Let y = 0 for ease. Then
1 (p1 x1 + p2 x2 + p3 x3 )
Φdp,0 =
4πε0 |x|3
55
and working out Edp,0 component-wise, we get
Edp,y = −∇Φdp,y
so
∂Φdp,0 1 p1 3p1 x21
=− − .
∂x1 4πε0 |x|3 |x|5
Continuing this process we end up with

1 3(p · x x−p
b)b
Edp,0 =
4πε0 |x|3
2.5.1 Jumps
Recall that for a surface S and h ∈ C0 (S),

Z
Fh,S (φ) = h(x)φ(x) dSx
S
We want to find, for the h-surface charge distribution, Φh,S and Eh,S .
Lemma 2.11 Z
1 1 h(y)
Φh,S (x) := Φ ∗ Fh,S = dSy
ε0 4πε0 |x − y|
S
Proof: We have
1
Φh,S (φ) = Φ ∗ Fh,S (φy )
ε
Z0
Fh,S (φy ) = h(x)φ(x + y)dSx ∈ D(R3 )
S
(where y is fixed in dSx ). Therefore

 
Z Z
1 1 
Φh,S (φ) = h(x)φ(x + y)dSx  dy
4πε0 |y|
R3 S
 
Z Z
1 φ(x + y) 
= h(x)  dy dSx
4πε0 |y|
S 3
R 
Z Z
1 φ(z)
= h(x)  dz dSx
4πε0 |x − z|
S R3
 
Z Z
1 h(x)
= φ(z)  dSx  dz
4πε0 |x − z|
R3 S
56
Hence
Z
1 h(y)
Φh,S (x) = dSy
4πε0 |x − y|
S
Note that Φh,S is continuous over S. However, Eh,S has a jump discontinuity across
S.
h(y) 1
To find E, we need to find ∇Φ, and |x−y|
becomes like ∼ r2
r dr and we have problems.
But this is how we get around it:
Imagine we have a surface S upon which we have a charge density h. Take a cylinder
of circular area ω and of height 2ε. Push this through the surface at a point so that
it’s perfectly half-way through the surface. So, denoting the cylinder C,
C = ω × (−ε, ε)n
Gauss’ law in integral form, applied to this cylinder, tells us that

Z
Q(C)
(Eh,S · ν)dS =
ε0
∂C
where ν is normal to ∂C
Z
1
= h dS (2.2)
ε0
ω
because the only charge in the cylinder is in the disc of area ω.
Now, let’s take a look at the left hand side.

Z Z Z Z
= + +
ε
∂Cω Sε
ωε+ ωε−
where
ωε± = {ω ± εn} the top and bottom of the cylinder

Sε = {∂ω × (−ε, ε)n} the side surface of the cylinder
Let’s take a look at these three integrals as we take ε → 0

Z
E · ν dS → 0
Sε
57
since Area(Sε ) → 0
Z Z
E · ν dS = E(x + εn) · n dSx
ωε+ ω
Z
−−−→ lim E(x + εn) · n dSx
ε→0+ ε→0
ω
Z
= E+ (x) · n dSx
ω
and similarly
Z Z
E · ν dS = E(x − εn) · (−n) dSx
ωε− ω
Z
−−−→ − E− (x) · n dSx
ε→0−
ω
Therefore, Z Z

lim Eh,S · ν dS = E+ (x) − E− (x) · n dS
ε→0
ε
∂Cω ω
So taking that and comparing it with the right-hand side of equation 2.2, we get
+ h(x)
E (x) − E− (x) · n = (2.3)
ε0
h
i.e. the normal component of Eh,S has a jump across S of size .
ε0
2.5.2 Tangential continuity
We are now going to show that despite the normal component of the electric field
having a jump, the tangential component is continuous.
To do this, let’s take a line ℓ from point x0 to x1 upon a surface S. Draw similar lines
at a distance ε above and below ℓ. Now join lines at the sides connecting them up,
forming a loop. Suppose we decide on an anticlockwise direction round this loop. So
the four lines, mathematically speaking, are
Lε = x0 × (ε, −ε)n (left)

Rε = x1 × (−ε, ε)n (right)
ℓ−
ε = ℓ × −εn (bottom)
ℓ+
ε = ℓ × εn (top)
58
Now, I
E · dr = 0 (2.4)
Cεℓ
and, using a similar trick to last time,

I Z Z Z Z
= + + +
Cεℓ ℓ−
ε ℓ+
ε
Lε Rε
Notice that  
Z Z
lim  + =0
ε→0
Lε Rε
since the length is 2ε → 0. And for the other two,

 
Z Z Z
  −
lim  +  = E (x) − E+ (x) · dr
ε→0 ℓ
ℓ−
ε ℓ+
ε
=0 (2.5)
by the initial observation in equation 2.4.
Say that dr = ṙ dt, where ṙ is a tangential vector to S, which we’ll define as t.
Change ℓ through any x ∈ S (i.e. rotate ℓ) to have all possible tangential directions.
Then 2.5 means that +
E (x) − E− (x) · t = 0
for any tangential t, i.e. the tangential part of the electric field is continuous!
2.5.3 Brief recap of definitions
If we have a dipole p · ∇δy at y,

1 p · (x − y)
Φp,y (x) =
4πε0 |x − y|3
and across the whole surface,
Z
1 p · (x − y)
Φdp,S (x) = dSy
4πε0 |x − y|3
S
Problem 2.12 ∗ Let S = {z = 0}, the xy-plane. Assume the magnitude of the
dipole is constant p = p0 .
1. Find Φ(x), the corresponding electrostatic potential of this surface.
2. Find the jump of Φ(x) across z = 0.
59
Solution From the above,

Z
1 p · (x − y)
Φ(x) = dSy
4πε0 |x − y|3
z=0
Now, letting    
x1 y1
x = x2 
 y = y2 

x3 y3
and using the fact that p = p0b
z,
Z
1 p0 · (x3 − y3 )
= dSy
4πε0 |x − y|3
z=0
And since on the surface z = 0, y3 = 0,

Z
1 p0 x3
= dSy
4πε0 |x − y|3
z=0
Z
p0 x3 1
= dSy
4πε0 |x − y|3
z=0
Z
p0 x3 1
= dSy
4πε0 [(x1 − y1 )2 + (x2 − y2 )2 + (x3 )2 ]3/2
z=0
Z Z
p0 x3 ∞ ∞ 1
= dy1 dy2
4πε0 −∞ −∞ [(x1 − y1 ) + (x2 − y2 )2 + (x3 )2 ]3/2
2
Using the substitutions
y1′ = y1 − x1 y2′ = y2 − x2 dy1′ = dy1 dy2′ = dy2
Z ∞ Z ∞
p0 x3 1
= dy1′ dy2′
4πε0 −∞
′
−∞ [(y1 )2 + (y2′ )2 + (x3 )2 ]3/2
Then converting to polar coordinates
r 2 = (y1′ )2 + (y2′ )2
Z Z
p0 x3 2π ∞ 1
= r dr dθ
4πε0 0 0 [r 2 + x3 2 ]3/2
Z ∞
p0 x3 r
= 2π dr
4πε0 0 [r 2 + x3 2 ]3/2
" #∞
p0 x3 −1
=
2ε0 (r + x3 2 )1/2
2
0
60
p0 x3 1
=
2ε0 |x3 |
p0
= sgn(x3 )
2ε0
p0
and the jump, therefore, is . X
ε0
2.6 Conductors
A conductor is a very good man metal. Assume we have a conductor in a domain D,

placed into an electrostatic field Eext .
All the positively charged particles want to move along Eext , but they’re too heavy so
they don’t. All the negatively charged particles want to move along −Eext under the
action of the electric field, and they can because they’re small. (Hand-waving much?)
Eventually then, inside this conductor, the electric field is 0 for our study of elec-
trostatics, since all the negatively charged particles at the boundary somehow compete
with Eext , creating E = 0.
So electrons move to the boundary on one side.
To put it another way: in conductors, elementary charges can move freely. As a result,
there will be redistribution of these elementary charges inside D and the total electric
field becomes zero in D.
As Etotal = E + Eext = 0 in D, =⇒ Φ = const inside D.
By the Poisson equation,

ρ
−∇2 Φ =
ε0
i.e. ρ = 0 inside D.
Therefore the charges are concentrated on S = ∂D, i.e. we have a surface charge FS,σ ,
where σ is the charge distribution over the surface.
Using E := Etotal from now on, recall that for jumps of the electric field across a
charged surface, 
 + σ(y)
E (y) − E− (y) · ny =
ε
 [E+ (y) − E− (y)] · t = 0 0
for any tangential vector t. But E− = 0 since E = 0 inside D. Therefore

1
E+ (y) = σ(y) ny .
ε0
Also,
Φ+ (y) = Φ− (y)
61
because Φ is continuous across the boundary, for y on the surface. And since Φ is
constant on the surface,

Φ+ (y) = Φ− (y)y∈S = const.
Now, suppose we have a large conductor D3 with a big hole cut out of it. Inside
the cavity are two small conductors D1 , D2 which do not touch. Call the cavity Ω =
R3 \ (D1 ∪ D2 ∪ D3 ). And let Si be the boundary of Di .
For x ∈ Ω,
Φ|S1 = c1
Φ|S2 = c2
Φ|S3 = c3
Now assume we divide D1 into two by a thin insulator running somewhere through
it. Then the electrostatic potential in the two halves are different. Calling the two
boundaries S1± ,
Φ|S1+ = c+
1
Φ|S1− = c−
1
which are different. Keep on dividing D1 into smaller and smaller pieces, each giving
a different value of Φ, i.e. we obtain
Φ|∂Ω = φ,
an arbitrary continuous function, where φ tells you the charge at a point.
Therefore, when considering Φ(x), x ∈ Ω, we can often assume that we know, for a
given φ,
Φ|∂Ω = φ (Dirichlet boundary condition)
but we also know that Φ in Ω satisfies
ρ
− ∇2 Φ = . (Poisson’s equation)
ε0
The combination of these two equations is called in mathematics, the Dirichlet bound-
ary value problem for Poisson’s equation. How original.
62
Problem 2.13 Let us have a paraboloid sitting centrally on top of the xy-plane,
and a potential

sin x if z ≥ x2 + y 2
Φ(x) =
sin x + ez (z − x2 − y 2) if z < x2 + y 2
1. Find the charge density ρ inside the paraboloid.
2. Find the charge density ρ outside the paraboloid.
3. Find the surface charge density σ on the paraboloid z − (x2 + y 2 ) = 0.
Recall that if your surface is given by the equation
S = {f (x, y, z) = 0}
then
∇f
n=
|∇f |
Solution 1. If we ever want to find out ρ then we want to be using the equation
ρ
−∇2 Φ = .
ε0
Inside the paraboloid, z ≥ x2 + y 2 and so Φ = sin x, hence
−∇2 Φ = −(− sin x)
and so
ρ = ε0 sin x.
2. Outside the paraboloid, Φ = sin x + ez (z − x2 − y 2 ), so

−∇2 Φ = − − sin x + ez (z − x2 − y 2 − 2)
and so
ρ = ε0 sin x − ez (z − x2 − y 2 − 2)
3. By equation (2.3), for a surface charge distribution σ,

σ
(E+ − E− ) · n =
ε0
and of course we know E+ and E− since we know Φ+ and Φ− , if we denote the
inside as ‘+’ and the outside as ‘−’.
E+ − E− = −∇Φ+ + ∇Φ−

= −∇ Φ+ − Φ−

= −∇ −ez (z − x2 − y 2 )
 
ez (−2x)
= ez (−2y) 
z 2 2
e (z − x − y + 1)
63
Now what is n? As we’re reminded, if the surface is given by the equation

f (x, y, z) = 0, which in this case is
f (x, y, z) = z − x2 − y 2 = 0
then  
−2x
∇f 1 −2y 
n= =p
|∇f | 1 + 4(x2 + y 2) 1
Calculating,
   
ez (−2x) −2x
1   · −2y 
(E+ − E− ) · n = p ez (−2y)
2 2
1 + 4(x + y ) ez (z − x2 − y 2 + 1) 1
z 2 2 2 2
e [4x + 4y + z − x − y + 1]
= p
1 + 4(x2 + y 2 )
So
σ = ε0 (E+ − E− ) · n
ε0 ez [4x2 + 4y 2 + z − x2 − y 2 + 1]
= p
1 + 4(x2 + y 2)
2 +y 2
ε0 ex [4x2 + 4y 2 + x2 + y 2 − x2 − y 2 + 1]
σ(x, y) = p
1 + 4(x2 + y 2 )
2 2
ε0 ex +y [4x2 + 4y 2 + 1]
= p
1 + 4(x2 + y 2 )
and that’s the answer. A bit ugly but there you go. Not everything in mathem-
atics is beautiful.
2.7 Boundary value problems of electrostatics
Say we have a domain (cavity) Ω ⊂ R3 , and we have the following three conditions:
 ρ
 2
−∇ Φ = ε
 in Ω
(O) Φ|∂Ω = φ (given φ)


Φ(x) → 0 if |x| → ∞
The third condition is strictly not necessary in simple cases, so we “have it with a
pinch of salt”.
64
Then this problem actually reduces to the following case. Let Φ1 , Φ2 solve problems
(A) and (B) below: 
 2 ρ
−∇ Φ1 = ε in Ω

(A) Φ1 |∂Ω = 0


Φ (x) → 0 if |x| → ∞
1

2

−∇ Φ2 = 0 in Ω
(B) Φ2 |∂Ω = φ (given φ)


Φ2 (x) → 0 if |x| → ∞
Then Φ = Φ1 + Φ2 solves (O). Nice, huh?
Notice that (B).1 is Laplace’s equation: Φ2 is a harmonic function.
Now, let’s take a particular case when ρ = δ. So we’re looking for Green’s function
(where we use G notation as not to confuse ourselves!) corresponding to (A). It is a
distribution (actually a discontinuous function) in D ′ (Ω)
G(x, y) y ∈ Ω fixed
So we have 
2

−∇ G(x, y) = δy (x)
(A′ ) G(x, y)|x∈∂Ω = 0


G(x, y) → 0 if |x| → ∞
If Ω = R3 then
1
G0 =
4π|x − y|
where we use the convention of using the subscript 0 to denote free space, i.e. R3 .
Let
Ψy (x) = G(x, y) − G0 (x, y).
Then
−∇2 Ψy = −∇2 x G − (−∇2 x G0 )
(where y is constant)
= δy (x) − δy (x)
=0
So −∇2 Ψy = 0 in Ω.
Meanwhile, let’s have a look what happens on the boundary.
Ψy (x)|x∈∂Ω = G(x, y)|x∈∂Ω − G0 (x, y)|x∈∂Ω (2.6)
65
but G on the boundary is 0

1
=− (2.7)
4π|x − y| x∈∂Ω
So as x → ∞, Ψy → 0.
So we’ve reduce the problem (A′ ) to




 −∇2 Ψy = 0 in Ω
 1
′
(B ) Ψy (x)|x∈∂Ω = −
 4π|x − y|

 x∈∂Ω
Ψ(x, y) → 0 if |x| → ∞
Now what about solving Φ for any ρ ∈ C0 (Ω)? Let

Z
1
Φ= G(x, y)ρ(y) dy
ε0
Z
2 1
−∇ x (Φ) = −∇2 x G(x, y)ρ(y) dy
ε0
Z
1
= δ(x − y)ρ(y) dy
ε0
1
= ρ(x)
ε0
So we’ve proved the following lemma!
Lemma 2.14 Let G(x, y) be known, x, y ∈ Ω. Then, for any ρ ∈ C0 (Ω),

Z
ρ 1
Φ (x) = G(x, y)ρ(y) dy
ε0
Ω
solves 
−∇2 Φρ = 1 ρ
ε0
 Φρ | = 0
∂Ω
2.7.1 To summarise
We had 
 2 ρ
−∇ Φ = ε

 Φ|∂Ω = 0

Φ(x) → 0 as |x| → ∞
Which we reduced to finding the Green’s function

2

−∇ x G = δy y ∈ Ω
G|∂Ω = 0


G(x) → 0 as |x| → ∞
66
and then since G(x, y) = G0 (x, y) + Ψy (x), it reduces to find Ψ such that
 2
−∇ x Ψy (x) = 0


y∈Ω
1
Ψy (x)|x∈∂Ω = −

 4π|x − y|

Ψy (x) → 0 as |x| → ∞
and this last set of equations dictate the special Dirichlet boundary value problems for
Laplace’s equation.
2.8 Dirichlet boundary value problems for Laplace’s

equation
Now we’ll use u instead of Φ so as not to imply R3 , since this result holds for any Rn .
We have the system 
 2
−∇ u = 0
u|x∈∂Ω = φ ∈ C(∂Ω)


u(x) → 0 as |x| → ∞
Recalling that ∇2 u = 0 in Ω ⇐⇒ u(x) is a harmonic function in Ω.
Problem 2.15 ∗ Assume that u ∈ C 2 (Ω) is harmonic. Show that, in 3D (i.e.

Ω ⊂ R3 ), Z
∂u
dS = 0
∂n
∂Ω
(hint: use Green’s identities)
Solution Green’s identity in 3D is, for Ω ⊂ R3 ,

Z I
2 2 ∂g ∂f
[f ∇ g − g∇ f ] dV = f −g dS
∂n ∂n
Ω ∂Ω
Letting f = 1, g = u, Z Z
2 ∂u
∇ u dV = dS
∂n
Ω ∂Ω
And u is harmonic ⇐⇒ ∇2 u = 0, so
Z
∂u
=⇒ dS = 0
∂n
∂Ω
which is what we were looking for. X
67
Problem 2.16 Assume that u ∈ C 2 (S) is harmonic. Show that, in 2D (i.e. S ⊂ R2 ),

Z
∂u
dℓ = 0
∂n
∂S
Solution Again, we need to use Green’s identity, but we first need to find Green’s
identity in 2D. Let me write down the identity in 3D again:
Z I
2 2 ∂g ∂f
[f ∇ g − g∇ f ] dV = f −g dS
∂n ∂n
Ω ∂Ω
Now let
Ω = S × {−h, h} f = f (x, y) g = g(x, y)
i.e., turn Ω into a cylinder of height 2h. What does this mean for our integrals?
I Z Z
= 2h + 2
∂Ω ∂S S
I Z
= 2h
Ω S
So Green’s theorem becomes

Z Z Z
2 2 ∂g ∂f ∂g ∂f
2h [f ∇ g − g∇ f ] dS = 2h f −g dℓ + 2 f −g dS
∂n ∂n ∂z ∂z
S ∂S S
where we’ve made the observation that n = z on the top and bottom of the cylinder.
∂g
This means that ∂z = ∂f
∂z
= 0, since f and g are functions of x, y alone.
Z
∂g ∂f
= 2h f −g dℓ
∂n ∂n
∂S
Z Z
2 2 ∂g ∂f
=⇒ [f ∇ g − g∇ f ] dS = f −g dℓ
∂n ∂n
S ∂S
which gives us Green’s identity for 2 dimensions.
Now, to solve the problem, just like last time we let f = 1 and g = u to give
Z Z
2 ∂u
0 = ∇ u dS = dℓ
∂n
S ∂S
once again. X
Let us introduce a theorem to help us do more stuff.
68
Theorem 2.17 (The Mean Value Theorem)
Let u be harmonic in Ω ⊂ R3 (i.e. ∇2 u = 0), and BR (x0 ) ⊂ Ω. Then in the 3D case,

Z
1
u(x0 ) = u(x) dSx
4πR2
∂BR (x0 )
=SR (x0 )
and in the 2D case, Z

1
u(x0 ) = u(x) dℓx .
2πR
CR (x0 )
Proof: (The 3D case)

=0
Z z}|{ −1

0= ∇2 u dx
4π|x − x0 |
BR (x0 )
which, by Green’s formula,

δ(x−x0 )
Z z }| { Z
−1 ∂u −1
= u ∇2 dx + dS
4π|x − x0 | ∂n 4π|x − x0 |
BR (x0 ) SR (x0 )
| {z }
u(x0 )
Z
∂ 1
+ u dS
∂n 4π|x − x0 |
SR (x0 )
Z Z
1 ∂u ∂ 1
= u(x0 ) − dS + u dS
4πR ∂n ∂n 4π|x − x0 |
SR (x0 ) SR (x0 )
| {z }
=0 by problem
Z
1 ∂ 1
= u(x0 ) + u(x) dS
4π ∂n |x − x0 |
SR (x0 )
Z
1 ∂ 1
= u(x0 ) + u(x) dS
4π ∂r r r=|x−x0 |
SR (x0 )
Z
1 1
0 = u(x0 ) + u(x) − 2 dS
4π r
SR (x0 )
and, by rearranging
Z
1 1
=⇒ u(x0 ) = u(x) 2 dS
4π R
Z
1
= u(x) dS
4πR2
69
Problem 2.18 ∗ Prove the theorem for the 2D case.
Solution The fundamental solution to the equation ∇2 G = δ is:

1
G(x) = ln |x|
2π
So we shall use Green’s identity in 2D as proved in problem 2.16. Once again, that is
Z Z
2 2 ∂g ∂f
[f ∇ g − g∇ f ] dS = f −g dℓ
∂n ∂n
S ∂S
Letting g = u, f = G,
=0
Z z}|{ ln |x − x |
0
0= ∇2 u dx
2π
SR (x0 )
δ(x−x0 )
z }|
Z { Z
2 ln |x − x0 | ∂u ln |x − x0 |
= u∇ dx + dℓ
2π ∂n 2π
SR ∂S
| {z }
u(x0 )
Z
ln |x − x0 |
∂
− u dℓ
∂n2π
∂S
Z Z
1 ∂u ∂ ln |x − x0 |
= u(x0 ) − ln(R) dℓ − u dℓ
2π ∂n ∂n 2π
∂S ∂S
| {z }
=0 by prob 2.16
Z
1
u(x)
= u(x0 ) − dℓ
|x − x0 |
2π
∂S
Z
1
= u(x0 ) − u(x) dℓ
2πR
∂S
and so, by rearranging

Z
1
=⇒ u(x0 ) = u(x) dℓ
2πR
CR (x0 )
which proves the theorem. X
Problem 2.19 Prove this version of the Mean Value Theorem: Let u be harmonic
in Ω ⊂ R3 (i.e. ∇2 u = 0), and BR (x0 ) ⊂ Ω. Then
Z
1
u(x0 ) = 4π 3 u(x) dV
3
R
BR (x0 )
70
Solution If you consider integrating over the volume as being equivalent to integ-
rating over all the surfaces of balls with radius 0 through to R, we have
 
Z Z R Z
 
u(x) dV =  u(x) dS  dr
r=0
BR (x0 ) Sr (x0 )
Z R
= 4πr 2 u(x0 ) dr
r=0
R3
= 4πu(x0 )
Z3
1
=⇒ u(x0 ) = 4π 3 u(x) dV
3
R
BR (x0 )
as desired. X
Time for another theorem.
Theorem 2.20 (The Maximum Principle) Let us have a bounded domain D ⊂

R3 . On this domain, let us have a function u ∈ C 2 (D int )∩C(D) which has the property
∇2 u = 0. That is to say, let u be harmonic and D-bounded.
If we let M = max u(x), then

x∈∂D
u(x) ≤ M ∀x ∈ D
Moreover, if D is connected, and there exists a point x0 ∈ D int such that u(x0 ) = M,
then
u(x) ≡ M ∀x ∈ D.
Proof: Assume that there exists a point x0 ∈ D int such that

u(x0 ) = max u(x) := A
x∈D
Let d = dist(x0 , ∂D) > 0. Now consider a ball of this radius at the point x0 , i.e.
Bd (x0 ) ⊂ D. By Problem 2.19,
Z
1
A = u(x0 ) = 4π 3 u(x) dx.
3
d
Bd
By the definition of A, we have that u(x) ≤ A in Bd . Therefore

Z Z
1 1
4π 3 u(x) dx ≤ A · 4π 3 dx = A
3
d 3
d
Bd B
| {zd }
vol
= vol =1
If u(y) < A for some y ∈ Bd (and therefore nearby), then

Z
1
4π 3 u(x) dx < A
3
d
Bd
71
Therefore, assuming that there exists a point y ∈ Bd such that u(y) < A we get
Z
1
A = u(x0 ) = 4π 3 u(x) dx < A
3
d
Bd
which is clearly a contradiction.
Therefore u(x) ≡ A inside Bd . Now for a bit of tidying up to complete the proof of
the first part.
As d = dist(x0 , ∂B), this implies that Sd ∩ ∂D 6= ∅. Therefore there exists a point

z0 ∈ ∂D such that u(z0) = A = maxx∈D u(x).
Therefore
M = max u(z) = max u(x) = A
z∈∂D x∈D
which proves the first part. (part 1)
Now, let’s again have

u(x0 ) = max u(x) = A
x∈D
Take any y ∈ D and connect x0 and y by a curve γ lying in D int . Then let
ρ = dist(γ, ∂D) > 0.
Assume y is not in the ball, since this would makes things trivial: y ∈ Bd =⇒ u(y) =
A.
Let x1 be the last point on γ and Sd . Then
u(x1 ) = A = max u(x).

x∈D
So then take a ball of radius ρ (so as it doesn’t touch the boundary) centred at x1 ,
Bρ (x1 ). The same arguments show that u(x) = A in Bρ (x1 ).
We continue this process until y ∈ Bρ (xn ). We need only a finite number of steps as
γ has finite length. This proves the second part which proves the whole thing.
Problem 2.21 Prove, either by rewriting the above proof, or as a corollary, the
Minimum Principle Theorem, i.e.
min u(z) = min u(x)

z∈∂D x∈D
Solution It falls trivially out of the Maximum Principle Theorem if we say
e(x) = −u(x).
u
There is, however, a stronger form of the theorem, but it needs these definitions first:
72
Definition 2.22 A point x0 is a local maximum of u(x) if there exists Bε (x0 ) such
that
u(x0 ) = sup u(x)
x∈Bε (x0 )
Definition 2.23 A point x0 is a local minimum of u(x) if there exists Bε (x0 ) such
that
u(x0 ) = inf u(x)
x∈Bε (x0 )
And the strong version of the theorem is:
Theorem 2.24 (Strong version of the theorem) If D is connected and u(x) 6≡

const and ∇2 u = 0 then u has no local maxima and minima inside D int .
We shan’t prove this theorem.
2.8.1 Now, gentlemen, we move to physics
Let’s have a domain D with no charges inside it. So
∇2 Φ = 0 in D
Introduce a point charge e into the domain. Then eΦ is the electrostatic (potential)
energy of the charge e.
Now recall that if we have Φ, then we also have E such that E = −∇Φ.
Then there is a force F = eE. This electrostatic force tries to decrease the electrostatic
energy of e. And since, by problem 2.21, min eΦ is achieved on ∂D, the charge ends
up on the boundary.
2.8.2 Unbounded domains
Now we shall take a look at what happens in unbounded domains. Let’s have

 2
∇ u = 0 in D
u|x∈∂Ω = φ ∈ C(∂Ω)


u(x) → 0 as |x| → ∞
We shall consider the unbounded domain as the limit of the following set. Let
DR = D ∩ BR .
73
By the maximum principle theorem,

u(x)|x∈DR ≤ max u(z).
z∈∂DR
Now what is the boundary of DR ?

∂DR = (∂D ∩ BR ) ∪ (SR ∩ D)
So

u(x)|x∈DR ≤ max max u(z), max u(x)
z∈∂D∩BR x∈SR ∩D

≤ max max u(z), max u(x)
z∈∂D x∈SR ∩D
since the first term is now the maximum of a smaller set. Note that the first term
then does not depend on R. And what happens as R → ∞? The second term goes to
0, since u → 0 as x → ∞.

=⇒ u(x)|x∈D ≤ max max u(z), 0 .
z∈∂D
But we can do the same thing using the minimum principle theorem, which gives us

u(x)|x∈D ≥ min min u(z), 0 .
z∈∂D
So in the case when D = R3 , then ∂D = ∅, which implies that
u(x)|x∈R3 = 0.
2.8.3 Uniqueness of solutions for Dirichlet boundary condi-

tions in bounded domains
Let us have a bounded domain D once again. Then the Dirichlet boundary conditions
are (
∇2 u = 0 in D
u|x∈∂Ω = φ ∈ C(∂Ω)
Proposition 2.25 For any bounded D any any φ ∈ C(∂D), there exists a unique
uφ ∈ C 2 (D int ) ∩ C(D) which solves the Dirichlet boundary conditions.
We can’t prove the existence: this requires potential theory for Dirichlet problems for
Laplace’s equation. But we can prove uniqueness:
Proof: Uniqueness Assume we have two functions uφ1 , uφ2 which solve the Dirichlet
boundary conditions. Let
u = uφ1 − uφ2 .
74
Then (
∇2 u = 0 inD
u|x∈∂Ω = 0
By the maximum principle, u(x) ≤ 0 for x ∈ D.
By the minimum principle, u(x) ≥ 0 for x ∈ D. Therefore
u ≡ 0 =⇒ uφ1 = uφ2
which is what we want.
Now what happened to distributions, you ask. Let’s look at Dφ′ (D) ⊂ D ′ (D), i.e. those
distributions which have a ‘meaningful’ restriction on ∂D which is equal to φ.
If u ∈ Dφ′ (D) is harmonic, then ∇2 u = 0, i.e. u ∈ C ∞ (D int ).
75
76
Chapter 3
Magnetism
3.1 The laws for magnetostatics
The Danish physicist Hans Christian Ørsted noticed during a lecture in 1820 that if
we have an electric current flowing through a circuit, it affects a magnetic dipole, i.e.
there is a force which acts on magnets.
So let us have two electric circuits with two currents I1 and I2 . Turn on I1 , then turn
on I2 . Then there is a force applied to the first circuit.
Taking a small part of each circuit,
I1 dl1 I2 dl2
and defining x12 as the vector distance from piece 2 to piece 1, they found the relation
for the force by 2 on 1 as
µ0 (I1 dl1 ) × (I2 dl2 × x12 )

F1←2 =
4π |x12 |3
3.1.1 Gauss’s law for magnetism
Definition 3.1 The Biot-Savart law is an equation in electromagnetism that de-

scribes the magnetic field B generated by an electric current. (In English literature,
the term ‘magnetic field’ is commonplace, foreign literature uses the phrase ‘magnetic
flux density’.) It’s given by
µ0 I2 dl2 × (x − x2 )
B=
4π |x − x2 |3
77
Definition 3.2 Suppose we have a flow of charges J(y). Then

Z
µ0 J(y) × (x − y)
B(x) = dVy
4π |x − y|3
R3
where we can say

J(y) = ẏ(s)δI2 ,C
where
C = {y = y(s)}, the curve around the circuit with current I2

ẏ(s) is tangent to the curve
δI2 ,C is like δh,L in Problem 1.42
I2 is constant
so
ẏ(s) ds = dl
Then
B(x) = J ∗× ∇Φ
(this is vector-valid convolution, i.e. it’s valid for each component of Ji .)
Let us have an electric field E(x). If we have a charge q, in electrostatics,
F(x) = qE(x).
If ρ(x) is an electric charge distribution (i.e. we don’t have point charges but a con-
tinuous distribution instead), then
Z
total
F = ρ(x)E(x) dV.
D
e
Now suppose we have a body with some J(x) inside. Then
Z
Ftotal e × B(x) dV
mag = J(x)
D
which is Ampère’s law of magnetic forces. (Ampère was a big boy: he had lots of laws)
But first we need to review some vector identities.

∇×(af (x)) = −a×∇f
∇×∇×F = ∇(∇ · F) − ∇2 F
78
Chapter 3. Magnetism
Solution In fact, this is just a rearrangement of the definition

F1ofthe vector Laplacian
2
∇ F. But you can show it explicitly in 3D by letting F = F2 , if you like wasting
F3
your time. X

∇ · (af (x)) = a · ∇f

Z Z
F · ∇f dV = − (∇ · F)f dV
(hint: use the divergence theorem, and that ∇ · (Ff ) = F · ∇f + (∇ · F)f ).
Now we’re going to take B(x) from Definition 3.2 and manipulate it. First, note that
x−y 1
− 3 = ∇
|x − y| |x − y|
and that J(y) is constant with respect to x. Let J(y) ∈ D(R3 ). Then
Z
µ0 J(y) × (x − y)
B(x) = dVy
4π |x − y|3
R3
Z
µ0 −1
= J(y) × ∇ dVy
4π |x − y|
R3
and by Problem 3.3,

Z
µ0 J(y)
= ∇× dVy (3.1)
4π |x − y|
3
 R 
Z
µ0 J(y)
= ∇×  dVy 
4π |x − y|
R3
so B is the curl of some magnetic field. So it follows, since div(curl(·)) = 0,
∇·B=0
Definition 3.7 Gauss’s law for magnetism is
∇·B=0
79
3.1.2 Ampère’s law
But what is ∇×B? Combining the identity from Problem 3.4 with equation line (3.1)
above,
Z
µ0 J(y)
∇×B = ∇x ×∇x dVy
4π |x − y|
R3
Z Z
µ0 J(y) µ0 2 J(y)
= ∇x ∇ x · dVy − ∇x dVy
4π |x − y| 4π |x − y|
R3 R3
| {z }| {z }
B A
Looking at the terms A and B in turn,

Z
µ0 2 1
A=− J(y)∇ x dVy
4π |x − y|
3
ZR
= µ0 J(y)δx−y dVy
R3
= µ0 J(x)
and
Z
J(y)
B= ∇ x ∇x · dVy
|x − y|
R3
 
Z
J(y)
= ∇x  ∇x · dVy 
|x − y|
R3
which, by Problem 3.5 is

Z
1
= ∇x J(y) · ∇x dVy
|x − y|
R3
Z
1
= −∇x J(y) · ∇y dVy
|x − y|
R3
which, by Problem 3.6 is

 
Z
1
= ∇x  (∇ · J
| {z } )(y) · dVy  (3.2)
|x − y|
R3 0
claiming that in magneto-electrostatics, ∇ · J = 0, (we prove this ahead in Lemma 3.9)
=0
which gives us
∇×B = µ0 J(x)
80
Definition 3.8 Ampère’s law is
∇×B = µ0 J(x)
Now we prove the claim:
Lemma 3.9 In magneto-electrostatics,

∇·J=0
Proof: This claim really comes from the philosophy of life: nothing comes from
nothing. We don’t want deus ex machina here, else we should shut our doors and go
to church.
Take a region V and look for the total charge Q in the region
Z
Q(t) = ρ(x, t) dVx .
V
We believe that charge doesn’t come from nothing. If the charge changes, then there
is some charge coming in and out of the region: i.e. if Q as a function of time changes,
there should be an in/outflux of charge in the region. Mathematically, if ∂Q∂t
6= 0, there
should be a current flux through S = ∂V . But this current flux is
Z
J(x, t) · n dS.
S
So the continuity law is essentially, before we tidy it up,

Z Z
∂ρ
(x, t) dV = − J(t) · n dS
∂t
V S
where we have a minus sign since if ρ increases, J should be pointing inwards. Using
the divergence theorem on the right-hand side,
Z Z
∂ρ
(x, t) dV = − ∇ · J(t) dV
∂t
V V
and since V is arbitrary,

∂ρ
= −∇ · J(x, t)
∂t
which of course we can rewrite in the more usual form,
∂ρ
+∇·J=0 (Continuity equation)
∂t
∂ρ
Now since we’re talking about statics at the moment, ∂t
= 0, so we get
∇·J=0
which is what we claimed.
So we’ve derived the four laws of electromagnetostatics. In differential form, they are
81
ρ
1. ∇ · E = Gauss’s law (Def n. 2.5)
ε0
2. ∇×E = 0 (Def n. 2.5)
3. ∇×B = µ0 J Ampère’s law (Def n. 3.8)
4. ∇ · B = 0 Gauss’s law for magnetism (Def n. 3.7)
In integral form, they are

Z Z
1
1. E · n dS = ρ dV
ε0
S=∂V V
I
2. E · dr = 0
C
I Z
3. B · dr = µ0 J · n dS
C=∂S S
Z
4. B · n dS = 0
S
3.2 The laws of magnetodynamics
3.2.1 Maxwell–Ampère law
In the dynamical case, ∇ · J 6= 0 like we used in Lemma 3.9. Instead,

∂ρ
∇·J =− .
∂t
What does this mean for our equations? Recall in equation (3.2) we had
Z
µ0 ∂ ρ(y, t)
∇×B = − ∇x dVy + µ0 J(x).
4π ∂t |x − y|
But what is Z
1 ρ(y, t)
dVy ?
4πε0 |x − y|
What is that? What is that? It’s Φρ (x, t)! So
∂
∇×B = ε0 µ0 (−∇x Φρ ) + µ0 J(x)
∂t
∂
= ε0 µ0 E(x, t) + µ0 J(x)
∂t
82
and so by rearranging we get

∂
∇×(µ−1
0 B) = (ε0 E) + J (Maxwell-Ampère eqn in vacuum)
∂t
What are µ0 , ε0 ? In the vacuum (or any isotropic material), they are just some con-
stants. But if we go to the brick, we have to replace them by matrices ε, µ. These
Figure 3.1: ε and µ are no longer constants if we go to the brick
matrices are different for each material, and they depend on x and t. Pretty cool,
huh?
We can define two new things to make our equations a bit nicer.
Definition 3.10 The magnetic H field, or magnetic field strength is defined as
H := µ−1 B
Definition 3.11 The electric displacement in a vacuum is
D := εE
So the Maxwell–Ampère equation becomes...
Definition 3.12 The Maxwell–Ampère equation is

∂D
∇×H = +J
∂t
and this is for any material.
3.2.2 Maxwell–Faraday equation
How does the statics equation ∇×E = 0 change for dynamics? In the 1830s, Michael
Faraday (of 1991 £20 note fame) carried out his famous experiment. He took a closed
83
(i.e. circular) piece of wire and he produced a magnetic flux through it by placing
another wire with a current flowing through it next to it (by the Biot-Savart law). If
he changed the magnetic field, there appeared a current in the wire. We change the
flux, which creates a force through it, proportional to the flux. What produces forces?
Electric fields! So he concluded that when you change an electric field in time, you
change a magnetic field. After all these fantastic experiments he found that
Definition 3.13 The Maxwell–Faraday equation is

∂B
∇×E = −
∂t
Now, gentlemen and ladies, we shall show that all of Faraday’s work (and the two
knighthoods which he declined!) was unnecessary since we shall derive the Maxwell–
Faraday law from Ampère’s law alone.
Let us have a circuit C as given in Figure 3.2. The top bar (U) is moving upwards at
y
6 6
v
h(t)
q U
? St = area C 6
- -
ℓ x
Figure 3.2: The circuit we’re looking at
b where B0 is the magnetic field of the Earth. Take a

a speed v = h′ (t). Let B = B0 k,
charge q along the bar U at height h(t).
J = qv = qvbj
Then by Ampère’s law of forces,
F = qv×B = qvB0bi.
But we know that F = qE, so
E = v×B = vB0bi.
Now let’s have a look at the integral around the circuit.
I Z
E · dr = vB0bi · dr
C(t) U
= −vB0 ℓ (3.3)
84
and
Z Z
B · n dS = b dS
B·k
St St
Z
= b·k
B0 k b dS
St
Z
= B0 dS
St
= B0 area(St )
= B0 ℓh(t) (3.4)
Z
∂ d
=⇒ B · n dS = B0 ℓh(t)
∂t dt
St
= B0 ℓv (3.5)
And by equating (3.3) and (3.5),

I Z
∂
− E · dr = B · n dS
∂t
C(t) St
or putting in the dependencies,

I Z
∂
− E(x, t) · dr = B(t) · n dS. (3.6)
∂t
C(t) St
Now here comes the trick. Say the change is not because of area, but instead because
of changing flux. By Stokes’ law, the left hand side of equation (3.6) is equal to
I Z
− E(x, t) · dr = (−∇×E) · n dS
C(t) S
and because S is not dependent on t, the right hand side of equation (3.6) is equal to
Z Z
∂ ∂B
B(t) · n dS = · n dS
∂t ∂t
St St
i.e. Z Z
∂B
(−∇×E) · n dS = · n dS
∂t
S St
And by the lemma we’re about to prove, since S is arbitrary, it’s safe to say that
∂B
∇×E = − .
∂t
85
Lemma 3.14 The step above is valid.
Proof: Basically this step is valid because it’s a version of the DuBois-Reymond
lemma. We start with Z
∂B
∇×E + · n dS = 0 (3.7)
∂t
and say that this is
∂B
⇐⇒ F := ∇×E + = 0.
∂t
Why? Assume it is not zero at some point. Then there exists a point x0 such that
F(x0 ) 6= 0.
Take a plane Π perpendicular to F through the point x0 , and call the disc of radius ε
in Π, Sε . We know that, by equation (3.7),
Z
F·nb dS = 0.
Sε
But by our clever choice of plane,
F(x0 )
b=
n
|F(x0 )|
And so
F·n
b |x=x0 = |F(x0 )| > 0.
But F is continuous on Π, so
1
F·n
b > |F(x0 )|
2
if ε is sufficiently small. Thus
Z
1
b dS > |F(x0 )| area(Sε )
F·n
Sε 2
>0
which is a contradiction. Hence F(x) = 0 ∀x.
So we’ve proved the Maxwell–Faraday equation another way.
3.3 Maxwell’s equations
And so we have derived Maxwell’s equations in any material. Recalling that, for
matrices ε and µ,
D = εE and B = µH,
In differential form:
86
1. ∇ · D = ρ Gauss’s law (Def n. 2.5)

∂B
2. ∇×E = − Maxwell–Faraday equation (Def n. 3.13)
∂t
3. ∇ · B = 0 Gauss’s law for magnetism (Def n. 3.7)
∂D
4. ∇×H = +J Maxwell–Ampère equation (Def n. 3.12)
∂t
In integral form:
Z Z
1. D · n dS = ρ dV
S V
Z Z
∂B
2. E · dr = − · n dS
∂t
C S
Z
3. B · n dS = 0
S
Z Z
∂D
4. H · dr = + J · n dS
∂t
C S
3.4 Étude on differential forms
Coordinates do not come from God. What we want to be able to say is that Maxwell’s
equations (particularly in integral form) are valid for any set of coordinates that we
choose to integrate along. So let’s have a brief discussion about differential forms.
Z Z
E · dr = E1 dx1 + E2 dx2 + E3 dx3
C C
Let x1 = x1 (t), x2 = x2 (t), x3 = x3 (t), then

Z b
∂x1 ∂x2 ∂x3
= E1 (x(t)) dt + E2 (x(t)) dt + E3 (x(t)) dt
a ∂t ∂t ∂t
Z b
∂x1 ∂x2 ∂x3
= E1 (x(t)) + E2 (x(t)) + E3 (x(t)) dt
a ∂t ∂t ∂t
So what if xi = xi (y k ), where i, k = 1, 2, 3, i.e. our x-coordinates are in terms of

y-coordinates, e.g. polars in Cartesians. Then
∂x1 1 ∂x1 2 ∂x1 3
dx1 = dy + 2 dy + 3 dy
∂y 1 ∂y ∂y
87
and similarly for dx2 and dx3 so in this case,

Z Z
E · dr = E1 dx1 + E2 dx2 + E3 dx3
C C
Z
∂x1 ∂x2 ∂x3
= E1 1 + E2 1 + E3 1 dy 1
∂y ∂y ∂y
Z 1 2

∂x ∂x ∂x3
+ E1 2 + E2 2 + E3 2 dy 2
∂y ∂y ∂y
Z 1 2

∂x ∂x ∂x3
+ E1 3 + E2 3 + E3 3 dy 3
∂y ∂y ∂y
Z
= E b · dy
b is E in y-coordinates, i.e.
where E
X ∂xi 3
b1 = ∂x · E =
E Ei
∂y 1 i=1
∂y 1
or generally,
3
X
bj = ∂xi
E Ei
i=1
∂y j
so we can interchange
3
X 3
X
Ei (x)dx ←→ i bj (y)dy j
E
i=1 j=1
i.e. Maxwell’s equations are valid for any choice of coordinates. This means that we
can apply them, with suitable considerations of course, in any manifold.
3.4.1 Vectors and covectors
Velocity is a real vector.
i dxi
v =
dt
dxi (y)
=
dt
X3
∂xi ∂y j
=
j=1
∂y j ∂t
3
X ∂xi
= vbj .
j=1
∂y j
88
b where we’re summing over i. So

See how we’re summing over j, in contrast to E,
−1
∂x
b=
v v
∂y
T
Eb = ∂x E
∂y
So when you’re looking at a point, v is a vector, and E is a covector.
Generally, E, H are one-forms, and B, D are two-forms.
89
90
Chapter 4
Electromagnetic Waves
Solutions to Maxwell’s equations in the absence of sources and sinks, i.e. with ρ = 0,
Jext = 0, are called electromagnetic waves.
Definition 4.1 Ohm’s law, as generalised by Gustav Kirchhoff, says that
J = Jext + σE
where σ is the conductivity matrix, a function of x and t. σ has to be positive definite.
In our study of (the very beginnings!) of the theory of electromagnetic waves we will
also assume that
σ = 0,
i.e. the conductivity is 0.
We will also assume that ε = ε(x), µ = µ(x), i.e. the electric and magnetic permitiv-
ities are time-independent. These reduce Maxwell’s equations to
1. ∇ · D = 0
∂B
2. ∇×E = −
∂t
3. ∇ · B = 0
∂D
4. ∇×H =
∂t
91
4.1 Electromagnetic energy
Electromagnetic waves carry with them electromagnetic energy. The energy density
E(x, t) of this energy is given by
1
E(x, t) = [D · E + B · H]
2
1
= [(εE) · E + (µH) · H] . (4.1)
2
Now we look at how E changes with time t in some domain V ⊂ R3 ,

Z Z
d ∂E(x, t)
E(x, t) dV = dV
dt ∂t
V V
Z
∂ 1
= [(εE) · E + (µH) · H] dV
∂t 2
V
∂
∂
and using ∂t
[(εE) · E] = 2 ∂t
(εE) · E ,
Z
∂E ∂H
= ε ·E+ µ · H dV
∂t ∂t
V
Z
∂D ∂B
= ·E+ · H dV. (4.2)
∂t ∂t
V
Using what’s left of Maxwell’s equations, rewrite (4.2) as

Z
= [(∇×H) · E − H · (∇×E)] dV
V
Z
=− ∇ · (E×H) dV, (4.3)
V
where we use that −∇ · (a × b) = (∇×b) · a − b · (∇ × a) for any vector fields a, b.
−∇ · (a × b) = (∇×b) · a − b · (∇ × a)
a1 b1
Solution Let a = a2
a3
and b = b2 , and just do the computation. Boring but
b3
trivial. X
Let’s now introduce the Poynting vector field.
92
Chapter 4. Electromagnetic Waves
Definition 4.3 The Poynting vector field S, named after its inventor John Henry
Poynting, who was a professor at the University of Birmingham from 1880, is defined
as
S = E×H.
With this definition, (4.3) gives rise to

Z Z
∂E(x, t)
dV = − ∇ · S dV
∂t
V V
and then using the divergence theorem

Z
=− S · n dA, (4.4)
∂V
where dA is an area element and n is the exterior unit normal to ∂V .
Equation (4.4) implies that the Poynting vector S describes the flow of electromagnetic
energy. Rewritten in the differential form it gives rise to the following continuity
equation
∂E(x, t)
+ ∇ · S(x, t) = 0,
∂t
cf. the charge continuity equation ∂ρ
∂t
+ ∇ · J = 0.

∂E(x, t)
+ ∇ · S(x, t) = 0.
∂t
Solution
∇ · S = ∇ · (E×H)
= (∇×E) · H − E · (∇×H)

∂B ∂D
= − ·H−E·
∂t ∂t

∂B ∂D
=− ·H+E·
∂t ∂t
and
∂E ∂ 1
= [D · E + B · H]
∂t ∂t 2
∂ 1
= [(εE) · E + (µH) · H]
∂t 2
∂ ∂
= (εE) · E + (µH) · H
∂t ∂t
∂D ∂B
= ·E+ ·H
∂t ∂t
∂E
=⇒ ∂t
+ ∇ · S = 0, as desired. X
93
4.2 Electromagnetic waves in an homogeneous iso-

tropic medium
In the following we assume that the electric and magnetic permittivities ε and µ are
isotropic and homogeneous (i.e. independent of x),
εij (x) = εδij µij (x) = µδij ,
An example of such a medium is the vacuum, where ε = ε0 , µ = µ0 . An example of
one that is not is the brick that we introduced in section 3.2.1.
Rewriting Maxwell’s equations, taking into account the above,
1. ∇ · D = 0
∂H
2. ∇×E = −µ
∂t
3. ∇ · B = 0
∂E
4. ∇×H = ε
∂t
Differentiating equations 2 and 4 with respect to t, we get

∂E ∂2H
∇× = −µ 2
∂t ∂t
2
∂H ∂ E
∇× =ε 2.
∂t ∂t
∂E ∂H
Using Maxwell’s equations as rewritten above, for , ,
∂t ∂t
we get
∂2H
∇× (∇×H) = −εµ (4.5)
∂t2
∂2E
∇× (∇×E) = −εµ 2 . (4.6)
∂t
Recall that, by the definition of the vector Laplacian ∇2 H,
∇× (∇×H) = ∇ (∇ · H) − ∇2 H,
where
∇2 H = (∇2 H1 , ∇2 H2 , ∇2 H3 ).
Then using equation 3 in Maxwell’s equations, ∇ · H = 0, so we get

∇× (∇×H) = −∇2 H
and (4.5) becomes
∂2H 1 2
2
= ∇ H.
∂t εµ
Similarly we can also derive
∂2E 1 2
2
= ∇ E.
∂t εµ
94
Problem 4.5 Derive

∂2E 1 2
2
= ∇ E.
∂t εµ
Component wise, this is equivalent to six scalar wave equations for the components
Ei (x, t), Hi (x, t), where i = 1, 2, 3,
∂ 2 Ei (x, t)
− c2 ∇2 Ei (x, t) = 0
∂t2
∂ 2 Hi (x, t)
− c2 ∇2 Hi (x, t) = 0, (4.7)
∂t2
where
1
c= √ (4.8)
εµ
is the speed of electromagnetic waves (the speed of light) in the medium. (Note that
in vacuum, c0 = √ε10 µ0 ).
4.3 Plane waves
An important special case of electromagnetic waves are the plane waves. In this case,
E and H depend, in addition to t, only on one space variable xe = x · b e, where
b e = bi,
e = (e1 , e2 , e3 ) is a unit vector. Since we’re in isotropic material, we can take b
and xe = x1 . Explicitly,
E = E(x1 , t) H = H(x1 , t) (4.9)
Clearly, in this case, E and H remain the same, at any fixed time t, in any plane
orthogonal to bi, which explains the name “plane waves”. Then, the 3D wave equations
(4.7) become 1D wave equations (which we studied in MATH1302∗ ),
∂ 2 Ei (x1 , t) 2
2 ∂ Ei (x1 , t)
− c =0
∂t2 ∂x21
∂ 2 Hi (x1 , t) 2
2 ∂ Hi (x1 , t)
− c =0 (4.10)
∂t2 ∂x21
We also do know the general solution in this case,
Ei (x, t) = Ei+ (x1 + ct) + Ei− (x1 − ct)

Hi (x, t) = Hi+ (x1 + ct) + Hi− (x1 − ct). (4.11)
i.e. in each case we have the superposition of two waves, one going to the right and
one going to the left.
∗
LOL
95
Problem 4.6 Show that if E = E(x1 , t), H = H(x1 , t), then
Ei (x, t) = Ei+ (x1 + ct) + Ei− (x1 − ct)

Hi (x, t) = Hi+ (x1 + ct) + Hi− (x1 − ct)
for i = 1, 2, 3 and some functions Ei± (x1 ), Hi± (x1 ).
Solution Look at the wave equation

∂2f
= c2 ∇2 f,
∂t2
assuming that f = f (x1 , t).
Use the substitution

ζ = x1 − ct η = x1 + ct
Then
∂f ∂f ∂x1 ∂f ∂t
= +
∂η ∂x1 ∂η ∂t ∂η
∂f 1 ∂f
= +
∂x1 c ∂t
:= F (x1 , t)
2

∂ f ∂ ∂f 1 ∂f
= +
∂ζ∂η ∂ζ ∂x1 c ∂t
∂F ∂x1 ∂F ∂t
= +
∂x ∂ζ ∂t ∂ζ
12 2
∂ f 1 ∂2f ∂ f 1 ∂2f 1
= 2
+ + + 2
· −
∂x1 c ∂x1 ∂t ∂x1 ∂t c ∂t c
2 2
∂ f 1∂ f
= 2
− 2 2
∂x1 c ∂t
∂2f 1
= 2
− 2 (c2 ∇2 f )
∂x1 c
∂2f ∂2f
= −
∂x21 ∂x21
= 0.
So
∂F
=0
∂ζ
∂f
since F = ∂η
. Therefore F is an arbitrary function of η: F = F (η). That is to say
∂f
= F (η),
∂η
an arbitrary function of η. So
f = α(η) + β(ζ)
96
where Z η
α(η) = F (e
η) de
η
0
hence
f = α(x1 − ct) + β(x1 + ct)
X
This formula shows that, indeed, the waves propagate with speed c in the bi-direction,
or, more precisely, E(x1 − ct), H(x1 − ct) is the wave propagating in the bi-direction
while E(x1 + ct), H(x1 + ct) is the wave propagating in the −bi-direction. Note that
to satisfy Maxwell’s equations, each of these two waves should satisfy these equations
inpedendently.
Using the original Maxwell equations, rather than the 1D wave equation (4.7), we can
further analyse E, H in this case. However, we will concentrate on one special type of
plane waves.
4.4 Harmonic plane waves
An electromagnetic wave is harmonic if its dependence on t is periodic and of the form

e±iωt , where ω is the frequency of oscillation. Thus, a harmonic electromagnetic wave
has the form
E(x, t) = E+ (x)eiωt + E− (x)e−iωt ,

H(x, t) = H+ (x)eiωt + H− (x)e−iωt .
If the electromagnetic wave is harmonic and plane, then the above equation, together
with (4.11), implies that
E(x, t) = E+ ei(kx1 +ωt) + E− ei(kx1 −ωt) + complex conjugate

H(x, t) = H+ ei(kx1 +ωt) + H− ei(kx1 −ωt) + complex conjugate, (4.12)
ω
where E± , H± are (complex) constant vectors and k
= c.
Again, to satisfy Maxwell’s equations, each of the terms in these two sums of four
terms should satisfy them. Thus, it is sufficient to concentrate on, say, E− ei(kx1 −ωt)
and H− ei(kx1 −ωt) . We’ll drop the minus sign to signal our generalisation.
Recall the divergence pair of Maxwell’s equations: ∇ · B = ∇ · D = 0. Obviously then

∇ · H = ∇ · E = 0 so

0 = ∇ · Eei(kx1 −ωt)
= (E · bi)ikei(kx1 −ωt)
=⇒ 0 = (E · bi)
97
and similarly
=⇒ 0 = (H · bi) (4.13)
i.e. E, H lie in the plane orthogonal to the direction of the wave propagation bi.
Next, the Maxwell-Faraday equation tells that

∂B ∂H
∇×E = − = −µ
∂t ∂t
and (4.12) tells us that
∂H
−µ = ickµH
∂t
And we can work out that
∇×E = ik(bi×E)

∇×E = ik(bi×E)
Solution Recall that from equation (4.9), E = E(x1 , t). So

bi bj b
k

∇×E = ∂x1 ∂
∂x
∂ ∂
∂x

E (x , t) E (x2 , t) E (x3 , t)
1 1 2 1 3 1
∂E2 b ∂E3 b
= k− j
∂x1 ∂x1
which by equation (4.12),
b − E3bj)
= ik(E2 k
and on the other side,

bi bj b
k

ik(bi×E) = ik 1 0 0

E1 (x1 , t) E2 (x1 , t) E3 (x1 , t)
= ik(E2 kb − E3bj)
which is clearly equivalent. X
so this implies
1 b
H= (i × E) (4.14)
cµ
r
ε b
= (i × E)
µ
1
= (bi×E)
Z
98
where r
µ
Z=
ε
is called the electromagnetic impedance.
Similarly, the Maxwell-Ampère equation implies that
E = −Zbi × H. (4.15)
Note that (4.15) follows from (4.14) if we apply bi× to both parts.
Let us analyse these equations. In principle, E is a complex vector. By shifting t, we

can assume
E = E1 u1 + E2 eiα u2 ,
where E1 , E2 are positive and u1 , u2 , bi form a right-hand triple of orthonormal vectors.
Then,
ZH = E1 u2 − E2 eiα u1 .
This means that the real parts of E and H are orthogonal, and the same is true for
their imaginary parts.
The last remark is about polarisation. When α ∈ {0, ±π}, then the direction of
E exp i(kx1 − ωt) does not depend on (x, t): we have linearly polarised waves since
eiα = ±1. However, for other α, if we fix any x, the direction of physical electric field
(which is the real part of E exp i(kx1 − ωt)) will rotate with period T = 2π
ω
.
π
For example, if E1 = E2 and α = 2
Ephysical (x, t) = E1 (cos(kx1 − ωt)u1 − sin(kx1 − ωt)u2 ) .
which gives you an ellipse.
Problem 4.8 Let α 6= π2 , u2 6= 0. So eiα is not real. Find the formula of Ephysical
and show that we get a slanted ellipse.
This problem is left as an exercise for the reader!
And that concludes our study of Maxwell’s Theory of Electrodynamics. The exam in
2011, for which these notes were given, featured two questions from chapter 1, two
questions from chapters 2 and 3, and one question from chapter 4. Please report any
errors you have noticed in the document to the editors, whose details can be found on
page 4. Chocolate, beer and any other form of appreciation will be warmly received.
谢谢
99
Figure 4.1: Class of 2011
100

Maxwells Theory Notes

Uploaded by

Copyright:

Available Formats

Maxwells Theory Notes

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Maxwells Theory Notes

Uploaded by

Copyright:

Available Formats

Maxwell’s Theory of Electrodynamics

From lectures given by

Edited, annotated and expanded by

Figure 1: Our Dear Leader and His Spirit Guide

1 Introduction to the Theory of Distributions 5

• Lisa for catapults and providing essential Russian to English translation;

• Kav for questions;

• The Permutation Boys (Jonathan, James and Charles) for permuting;

• Ewan for patiently being Ian;

Introduction to the Theory of

James Clerk Maxwell (1831–1879), pictured on page 2, was a Scottish mathematician

1.2 Definitions and notation

Let me start by reminding you of some definitions and notation:

Definition 1.1 A compact set K in a domain Ω, denoted K ⋐ Ω is a closed, bounded

1.2.1 The du Bois-Reymond lemma and test functions

Lemma 1.6 Let Ω ⊂ Rn be an open domain (bounded or unbounded). Let f (x) ∈

∀ ε > 0, ∃ δ > 0 s.t. |x − x0 | < δ =⇒ |f (x) − f (x0 )| < ε

Choose φ from C0∞ such that:

1. φ(x) ≥ 0 for all x

since φ is zero outside of this ball. But

since all the f (x)φ(x) terms are positive. Then

since φ(x) = 1 by condition 2 above. Then

by our choice of ε above.

But our initial assumption was that

Proposition 1.7 Let K1 and K2 be two compacts in Rn such that K1 ⋐ K2 . Then

Figure 1.1: Proposition 1.7 tells us this exists

We can take from the lemma a simple corollary:

Corollary 1.8 Let f, g ∈ C(Ω). Then if ∀φ ∈ C0∞ (Ω),

Proof: Subtracting one side from the other gives

Then f (x) − g(x) = 0 by Lemma 1.6, i.e. f (x) = g(x).

1.2.2 Topological vector spaces and multi-indices

Definition 1.10 D(Ω) is a topological vector space, i.e.

1. It is a linear vector space, i.e. if f, g ∈ D(Ω) and λ, µ ∈ R,

h(x) = (λf + µg)(x) = λf (x) + µg(x) ∈ D(Ω) x ∈ Ω

and h(x) = 0 for x ∈

2. There is a notion of convergence, i.e. there is a meaning to fk −−−→ f . This is

Definition 1.11 A multi-index α is an n-dimensional vector, where n = dim(Ω),

Multi-indices are convenient when we want to write high-order partial derivatives,

1.2.3 Null set and support

Definition 1.13 The support of a function f , supp(f ) is the complement of N (f )

Example 1.14 Let Ω = R and

Then N (f ) = (−∞, 0) and supp(f ) = [0, ∞).

Figure 1.2: Look, a picture of the function!

Definition 1.15 A function φk converges to φ in D(Ω) if the following two conditions

1. For any multi-index α and any x ∈ Ω,

i.e. all derivatives converge

2. There is a compact K ⋐ Ω such that

Figure 1.3: Definition 1.15(2) in picture form

Problem 1.17 If ζ ∈ C ∞ (Ω), then for any φ ∈ D(Ω), show

Solution This just follows from condition 1 of the definition of φp converging to

Problem 1.19 ∗ Let ζ ∈ C ∞ (Ω), i.e. ζ is an infinitely differentiable function in Ω.

Solution Want to show that if ψp = ζφp, ψ = ζφ, that

2. ∃K s.t. supp(ψp ), supp(ψ) ⊂ K

which is the sum of C ∞ terms. So as p → ∞, since φp → φ in D(Ω),

2. N (ψ) ⊃ N (φ) =⇒ supp(ψ) ⊂ supp(φ) ⋐ Ω since φp → φ in D(Ω).

Problem 1.20 If φp → φ, ψp → ψ in D(Ω) and λ, µ ∈ R, show that

λφp + µψp → λφ + µψ in D(Ω)

Problem 1.21 If φp → φ, ψp → ψ in D(Ω) and λp → λ, µp → µ in R, show that

Now we move on to functionals: