Chapter 1

Introduction to Electrostatics

Electrostatics is the study of time-independent distributions of charges and fields.

1.1 Coulomb’s Law

The foundation of electrostatics is Coulomb’s Law, together with the Super-

position Principle which we will discuss later.

Coulomb’s Law

The force F21 on a particle of charge q2 at r2 due to a particle of charge

q1 at r1 is given by
r̂21
F21 = kq1 q2
|r2 − r1|2
where

• r21 = r2 − r1

• r̂ is a unit vector in the direction of r.

Coulomb’s law is an experimental observation.

1
Chapter 1 2

F 21
q2

r2 F 12

r1 q1

In SI units :

• k = 1/4πǫ0 - the 4π is conventional.

• The charges q1, q2 are measured in Coulombs (C), and defined via the
magnetic effects of currents.

• ǫ0 , the Permitivity of Free Space is also a defined quantity:

ǫ0 = 8.854 187 817 . . . × 10−12 C 2N −1m−2

There are two further observations that we can make:

• The forces on the two charges are equal and opposite, obeying Newton’s
third law: F12 = −F21.

• The force is repulsive (attractive) for like (unlike) charges.

Electric Field: The electric field E at r is defined as the force acting on a unit
test charge at that point. More strictly,
F (r)
E(r) = lim ,
q→0 q

so that the electric field due to the test charge can be ignored.
Chapter 1 3

1.2 The Superposition Principle and Extended Distribu-

tions

In the above we have looked at the fields due to single, isolated point-like
charges. In this section, we will explore the second emperical ingredient necessary
for our understanding of electrostatic fields, the linear superposition principle.

Linear Superposition Principle

The resultant force on a test particle due to several charges is the

vector sum of the forces due to the charges individually.
Example: We have N charges qi (i = 1, . . . , N ), situated at the points ri . The
force on a test particle of charge q at the point r is given by
N
X qi (r − ri )
F (r) = kq 3
i=1 | r − ri |

where k = 1/4πǫ0 in SI units.

Thus the electrostatic field E(r) is
N
X qi(r − ri)
E(r) = k 3
i=1 | r − ri |

1.2.1 Extended Charge Distributions

We will now apply the linear superposition principle to a continuous distribution

of charge.
Consider a continuous distribution of charge density (charge per unit volume)
ρ(r′ ), confined to a volume V .
Chapter 1 4

V In order to use the superposition principle,

r - r’ we will divide the volume V into infinites-
∆V’
simal volume elements ∆V ′ , centred at r′ .
r
The charge occupying the volume element
r’ at r′ is
dq = ρ(r′ )dV ′

Therefore, the electrostatic field at the point r due to the element of charge dq at
r′ is
ρ(r′ ) (r − r′ )∆V ′
∆E(r) = k
| r − r′ |3
where we take ∆E(r) −→ 0 as r −→ ∞. We now use the principle of linear
superposition to write that the resultant field at r as a sum over the elements
∆V ′ in V
X ρ(r ′ ) (r − r ′ )∆V ′
E(r) = k
∆V ′ | r − r′ |3
In the limit that ∆V ′ becomes infinitessimal, we have

Z ρ(r′ )(r − r′ )dV ′

E(r) = k
V | r − r′ |3

Much of the rest of this course is centred on methods for obtaining the electrostatic
field, and we begin with one of the simplest - Gauss’ Law.

1.3 Gauss’ Law

Suppose that the charge density ρ(r) is the sole source of the electrostatic field
E(r). Gauss’ Law relates the flux of E out of a closed surface S bounding a
volume V to the total charge Q contained within V
Chapter 1 5

dS

Gauss’ Law states that:

Z Q
E · dS = 4πkQ = in SI units
S ǫ0
where

• Q = total charge within S

• dS = outward normal to surface, having infinitessimal area dS

Gauss’ Law provides a powerful way to compute the electrostatic field for the
case where there is spherical, or even cylindrical, symmetry. It will also form the
starting point for our derivation of Laplace’s equation later in the course.

1.3.1 Geometrical Interpretation of Gauss’ Law

Consider a point charge q placed at the origin (not necessarily inside V ), and the
electrostatic flux across an area dS.
Chapter 1 6

dS

q
θ
dΩ

Then we have
dS cos θ
E · dS = kq
r2
= kq dΩ

where dΩ is the solid angle subtended by dS at the origin; dΩ is the projection of

the surface element dS onto the unit sphere. Note that S dΩ = 4π where S is a
R

unit sphere, or any closed surface, enclosing the origin.

• If the charge q is outside the volume, then the total flux V E ·dS is zero; the
R

contributions from two elements of surface area produced by the intersection

of a cone with the surface cancel, see below.

• If the charge q is inside the volume, the total flux E · dS = q/ǫ0.

R
V
Chapter 1 7

dS dS

q
dS

Outside Inside

Though this provides an intuitive interpretation of Gauss’ Law, we will now pro-
ceed to a more formal proof.

1.3.2 Proof of Gauss’ Law

We will begin by proving Gauss’ Law for a single, pointlike charge q at the origin.

Gauss’ Law for a Single Charge

Our starting point is once again Coulomb’s Law:

r
E(r) = kq
r3
Lemma: For a single charge at the origin, ∇ · E = 0 for r 6= 0
Proof:
r 1 1
! !
∇· 3 = ∇ 3 · r + (∇ · r) 3
r r ! r
3r 3
= − 5 ·r+ 3 =0 when r 6= 0
r r
Chapter 1 8

Gauss’ Law for a point charge is:


Z 
 4πkq if the surface S encloses the origin
E · dS = 
S  0 otherwise
Proof:
Origin outside V :
E(r) is continuously differentiable, and ∇ · E = 0 everywhere within V . From the
divergence theorem,
Z Z
E · dS = (∇ · E) dV = 0 if origin not within V
S V

Origin inside V :
E(r) is undefined at r = 0. Therefore define V to be the region between the
closed surfaces S ′ and S, where S ′ is a small sphere of radius ǫ centred at the
origin:

S
dS

dS’
O S’
ε V

Now in the region V , ∇ · E = 0. Therefore, by the divergence theorem,

Z Z Z
∇ · E dV = E · dS + E · dS = 0
V S S′

Introduce spherical polar coordinates (r, θ, ψ). Then on the sphere S ′ we have:

dS = −ǫ2 sin θ dθ dψ er ,
Chapter 1 9

where the outward normal for S ′ points towards the origin. Therefore
er
Z Z 2π Z π !
E · dS = kq 2 · (−ǫ2 sin θ dθ dψ er )
S′ 0 0 r r=ǫ
= −4πkq independent of ǫ
We now let ǫ → 0, so that V → total volume within S, and we have
Z q
E · dS = 4πkq = in SI units
S ǫ0
so that the theorem is proved.
If the point charge is at the point r1 , then we have
r − r1
E(r) = kq .
| r − r1 |3
By changing variables to ρ = r − r1 it is easy to show

Z 
 4πkq = q/ǫ0 in SI units if r1 ∈ V
E · dS = 
S  0 otherwise

Gauss’ Law for Distribution of Point Charges

We can extend the proof of Gauss’ Law for a single charge distribution to a set of
N point charges {qi} at {ri } using the linear-superposition principle:
N
X
E(r) = Ei(r)
i=1
where E(r) is the total electrostatic field at the point r, and Ei (r) is the electro-
static field at the point r due to the charge qi at the point ri. Applying Gauss’
Law for point charges proved above, we have

Z 
 4πkqi = qi/ǫ0 in SI units if ri ∈ V
Ei · dS = 
S  0 otherwise
Hence
Z XZ X
E · dS = Ei · dS = 4πk qi
S i S i,ri ∈V
Q
= 4πkQ = (in SI units)
ǫ0
Chapter 1 10

where Q is the sum of the charges contained within the volume V .

Gauss’ Law for Continuous Distribution of Charge

This we prove by exact analogy with derivation of the electrostatic field for a
continuous distribution: we divide up the volume V into elements of volume ∆V ′ ,
centred at r′ , and obtain
Z
ρ(r′ ) ∆V ′
X
E · dS = 4πk
S ∆V ′ ∈V
′
Z
∆V →0
−→ 4πk ρ(r′ ) dV ′ = 4πkQ,
V

where Q is the total charge contained within the volume V .

not included
included
Chapter 1 11

1.3.3 Applications of Gauss’ Law

Gauss’ Law provides a powerful method of determining the electrostatic field

where we have symmetrical or cylindrical symmetry.

Spherical Symmetry

Suppose we have a spherically symmetric distribution of charge - or mass -

ρ = ρ(r), where r = |r|. Then the electrostatic field will depend only on r, and
therefore must be in the radial direction.
Choose a spherical surface S of radius r, centred on the centre of the charge
distribution. Then we have that
Z Z Z
E(r) · dS = E(r) er · dS = E(r) r2 dΩ = 4πE(r) r2.
S S Ω

But by Gauss’ Law, we have

Z
E · dS = 4πkQ(r),
S

where Q(r) = ρ(r′ ) dV is the total charge contained within the sphere of radius
R
V
r.
Thus we have
kQ(r) Q(r)
E(r) = er = er in SI units.
r2 4πǫ0 r2
Note that outside a spherically symmetric charge distribution, the field is the
same as if we had a point-like charge Q(r) at the origin.

Example: Consider a thin spherical shell of charge Q. We can say immediately:

• Outside the shell, the electrostatic field is the same as that of the equivalent
point charge Q at its centre:
kQ
E(r) = er
r2
• Inside the shell, the field is zero.
Chapter 1 12

Cylindrical Symmetry

Suppose we have a infinitely long, cylindrically symmetric distribution of

charge, with the axis of symmetry along the z axis. Introduce cylindrical co-
ordinates (ρ, θ, z). Note: we use θ rather than φ for the axial coordinate, to avoid
confusion with the potential that we will be introducing next.
Consider an element of length L, and radius ρ, containing a charge Q(ρ, L):
S

0 L
Eρ
The field will depend solely on ρ, and therefore must be in the eρ direction,
E(r) = E(ρ)eρ. Applying Gauss’ Law to the cylinder we have
Z
E · dS = 4πkQ(ρ, L)
S

Now on the “end-caps”, z = 0 and z = L, E · dS = 0, and therefore

Z Z Z
E · dS = E(ρ)eρ · dS = E(ρ) dS = E(ρ)2πρL.
S S S

Thus
2kQ(ρ, L) 2Q(ρ, L)
E(ρ) = = in SI units
ρL 4πǫ0 ρL
Example: Infinitely long, thin rod carrying charge λ per unit length. Thus,
Q(ρ, L) = λL and we have
λ
E(ρ) = .
2πǫ0 ρ
Chapter 1 13

We expect the treatment of the rod as infinitely long to be a good approximation

for a rod of finite length providing

w < ρ << l

where w and l and the width and the length of the rod respectively.

1.4 Maxwell’s First Equation (ME1)

Our starting point is Gauss’ Law:

Z Z
E · dS = 4πk ρ(r′ ) dV ′
S V

where ρ(r′ ) is the charge density. By the divergence theorem, we have

Z Z
E · dS = ∇ · E dV ′ ,
S V

and thus Z
{∇ · E − 4πkρ} dV ′ = 0.
V

This applies for any volume V , and therefore the integrand itself must vanish:
ρ
∇ · E = 4πkρ = . (1.1)
ǫ0
This is Maxwell’s First Equation (ME1). ME1 is essentially an expression of
Gauss’ law in differential form.
Chapter 1 14

1.5 The Scalar Potential

ME1 has provided us with a differential equation to describe the electric field,
E(r), but it would be easier were we able to work with a scalar quantity. The
scalar potential provides a means of so doing.

Scalar Potential

• Given a vector field A(r), under what conditions can we write A as the
gradient of a scalar field φ, viz. A(r) = −∇ φ(r), where the minus sign
is conventional?

• What can we say about the uniqueness of φ(r).

Definition: A simply connected region R is a region where every closed curve

in R can be shrunk continuously to a point whilst remaining entirely in R.
Examples:

R R

The region between two cylin-

The inside of a sphere is simply
ders is not simply connected:
connected
it’s doubly connected
Chapter 1 15

1.5.1 Theorems on Scalar Potentials

Let A(r) be a continuously differentiable vector field defined in a simply con-

nected region R. Then the following three statements are equivalent, i.e. any
one implies the other two:-

1. ∇ × A(r) = 0 for all points r ∈ R

I
2. (a) A(r′ ) · dr′ = 0, where C is any closed curve in R
C
Z r
(b) φ(r) ≡ − A(r′ ) · dr′ does not depend on the path between r0 and r.
r0
3. A(r) can be written as the gradient of a scalar potential φ(r)
Zr
A(r) = −∇ φ(r) with φ(r) = − A(r′ ) · dr′
r0
where r0 is some arbitrary fixed point in R.

Proof that (1) implies (2):

Let ∇ × A(r) = 0 in R, and consider any two curves, C1 and C2 from the point
r0 to the point r in R. Introduce the closed curve C = C1 − C2, and let S be a
surface spanning C.

r
C1

r0
C2
Chapter 1 16

Apply Stokes’ theorem:

I Z
′ ′
A(r ) · dr = ∇ × A · dS = 0
C S

since ∇ × A = 0 everywhere. Note that we use r′ as integration variable to

distinguish it from the end-points of C1 and C2 , r0 and r.
Thus we have: I
∇×A= 0⇒ A(r′ ) · dr′ = 0 (1.2)
C
for any curve C in R, and the first part of the proof is done.
For the second part of the proof, we observe
Z Z I
′ ′
′
A(r ) · dr − ′
A(r ) · dr = A(r′ ) · dr′ = 0.
C1 C2 C

Thus the scalar potential φ(r) of the vector field A(r) defined by
Z r
φ(r) = − A(r′ ) · dr′
r0
is independent of the path of integration joining r0 and r.
Proof that (2) implies (3)
Consider two neighbouring points r and r + dr. Define the scalar potential as
before: Z r
φ(r) = − A(r′ ) · dr′
r0

r
dr
r + dr

r0
Chapter 1 17

Now define the quantity δφ(r):

δφ(r) = φ(r + dr) − φ(r)
 
r+dr Z
′
Z r
′
A(r′ ) · dr′ 
 
= − A(r ) · dr + (by definition)
r0 r0
 
Z r+dr ′ ′
Z r
0 ′ ′
 Z
= − A(r ) · dr + A(r ) · dr  (swapped limits on 2nd )
 r0 r
Z r+dr
= − A(r′ ) · dr′ (Integral around closed curve vanishes)
r
" # r +dr
′ ′
= − A(r ) · r (for infinitesimal dr)
r
= A(r) · { − (r + dr) + r}

So δφ(r) = − A(r) · dr (1.3)

To perform the integral, we used path independence and integrated along the
infinitesimal straight line between r and r + dr along which A(r′ ) is constant up
to effects of O(dr).
But, by Taylor’s theorem, we also have
∂φ(r)
δφ(r) = dxi = ∇ φ(r) · dr (1.4)
∂xi
Comparing equations (1.3) and (1.4), we obtain
A(r) = − ∇ φ(r)
Thus we have shown that path independence implies the existence of a scalar
potential φ for the vector field A.
Proof that (3) implies (1)
A = ∇φ ⇒ ∇ × A = ∇ × (∇ φ) ≡ 0
because curl (grad φ) is identically zero (i.e. it’s zero for any scalar field φ).
Chapter 1 18

1.5.2 Terminology

Such a vector field is called

I
• Irrotational: ∇ × A(r) = 0 ⇔ A(r′ ) · dr′ = 0
C
If you look in older textbooks, you will sometimes see rot rather than curl.

• Conservative: e.g. if A = force, then φ is potential energy and total energy

is conserved (see later).

• The field φ(r) is the scalar potential for the vector field A(r).

1.5.3 Uniqueness

φ(r) is uniquely determined up to a constant.

Proof:
Let φ and ψ be scalar potentials obtained by different choices of r0 . Then

∇φ − ∇ψ = A − A = 0

Therefore
∇ (ψ − φ) = 0
Integration of this equation wrt any of x, y, or z gives

ψ − φ = constant

Therefore
ψ = φ + constant
The absolute value of a scalar potential has no meaning, only potential dif-
ferences are significant.
Chapter 1 19

1.5.4 Existence of Scalar Potential for Electrostatic Field

After the digression on subject of the scalar potentials, it is time to show that the
electrostatic field is, indeed, irrotational.
The central result of this chapter was the expression for the electrostatic field due
to a continuous charge distribution
Z ρ(r′ )(r − r′ )dV ′
E(r) = k .
V | r − r′ |3
Thus we have
 ρ(r ′ )(r − r′)  ′
 
Z
∇ × E(r) = ∇× dV
| r − r′ |3 
   
Z 1 1
ρ(r′ ) ∇   × (r − r ′ ) + ′
dV ′
 
= ∇ × (r − r )
V | r − r′ |3 | r − r′ |3 

 −3(r − r ′ )
 
Z
′ ′
dV ′

= ρ(r )  ′ 5
× (r − r ) + 0
|r−r | 

= 0

where the derivatives act only on the unprimed indices.

The electrostatic field E(r) can be written in terms of a

scalar potential E(r) = − ∇φ(r)

1.5.5 Methods for finding Scalar Potentials

We have shown that the scalar potential φ(r) for an irrotational vector field A(r)
can be constructed via
r Z
φ(r) = −A(r′ ) · dr′
r0
for some suitably chosen r0 and any path which joins r0 and r. Sensible choices
for r0 are often r0 = 0 or r0 = ∞.
We have also shown that the line integral is independent of the path of integration
Chapter 1 20

between the endpoints. Therefore, a convenient way of evaluating such integrals is

to integrate along a straight line between the points r0 and r. Choosing r0 = 0,
we can write this integral in parametric form as follows:

r′ = λ r where {0 ≤ λ ≤ 1}

so dr′ = dλ r and therefore

Z λ=1
φ(r) = − A(λ r) · (dλ r)
λ=0

Example:
Let A(r) = (a · r) a where a is a constant vector.
It is easy to show that ∇ × ((a · r) a) = 0. Thus
Z r
φ(r) = − A(r′) · dr′
0
Z r  
= − (a · r′ ) a · dr′
0
Z 1 
= − (a · λ r) a · (dλ r)
0
Z 1
= − (a · r)2 λ dλ
0
1
= − (a · r)2
2
Of course, this is all rather artifical. What we really want to do is to obtain φ
and A from first principles.
Chapter 1 21

1.5.6 Singular Fields

We have seen that, for the case of a point-charge at the origin, the electric field
is singular at r = 0. In such cases, it is not possible to obtain the corresponding
scalar potential at r by integration along a path from the origin. All is not
lost - remember that the starting point for our path is arbitrary, and often it is
convenient to take it at infinity.
Example: Electric field due to point charge at r = 0: E(r) = kqr/r3, so that
E(r = 0) is singular, and hence undefined. As in the proof of Gauss’ law, our
region R must exclude an infinitessimal sphere centred at r = 0.
Here we choose a path from r0 = ∞, yielding
Z r Z 1
φ(r) = − E(r′ ) · dr′ = − E(λr) · dλ r
∞ ∞
dλ r2
Z 1
= −kq
∞ λ2 r 3
1
= kq
r
Thus we have the famous 1/r potential due to a point charge.

Because of the linearity of the gradient operation, we can impose the linear su-
perposition principle on the potential, and hence obtain an expression for the
potential due to an extended charge distribution:
Z
ρ(r′ )dV ′
φ(r) = k (1.5)
V | r − r′ |

1.5.7 Multiply-connected Regions

In this case, ∇× A = 0 does not imply the existence of a scalar potential function.
Example: Work using cylindrical coordinates (ρ, φ, z). A vector field A, with
a
Aρ = Az = 0, Aφ =
ρ
Chapter 1 22

where a is a constant, is defined outside an infinitessimal cylinder about the z-

axis, where Aφ is singular. This region is doubly connected (c.f. example above
where we exclude an infinitessimal sphere).

eθ
eρ

ρ
Excluded
region
θ

Then we have (Exercise!):

• ∇×A=0
I
• A · dr 6= 0 where C is a circular path enclosing the z-axis
C
In this case, the “potential” would depend on the choice of path, and in particular
the winding number - the number of times that a path wraps around the z-axis.
Examples: Vortices in superconductors, Cosmic strings...

1.5.8 Conservative Forces and Physical Interpretation of Potenital

To see how the name conservative field arises, consider a vector field F (r) cor-
responding to the only force acting on some test particle of mass m. The work
done by the force in going around a closed curve C is
I
W = F (r) · dr
C
For a conservative force, ∇ × F = 0, the earlier theorems tell us:
Chapter 1 23

• The total work done by the force in moving the particle around a closed curve
is zero.

• We can write the force in terms of a scalar potential

F (r) = −∇U (r).

where the minus sign is conventional (see later).

We will now show that for a conservative force, the total energy is constant in
time.
Proof
The particle moves under the influence of Newton’s Second Law:

mr̈ = F (r).

Consider a small displacement dr taking time dt along the path followed by the
particle. Then we have

mr̈ · dr = F (r) · dr = −∇U (r) · dr.

Integrating this expression along the path from rA at time t = tA to rB at time

t = tB yields
rB
Z Z r
B
m r̈ · dr = − ∇U (r) · dr. (1.6)
rA rA
We can simplify the left-hand side of equation 1.6 to obtain
ZrB Z t
B
Z t 1 d
B 1
m r̈ · dr = m r̈ · ṙdt = m ṙ2 dt = m[vB2 − vA2 ],
rA tA tA 2 dt 2
where vA and vB are the magnitudes of the velocities at the points labelled by A
and B respectively.
To integrate the right-hand side of equation 1.6, we appeal to Taylor’s theorem
to note that
∂U
∇U (r) · dr = dxi
∂xi
Chapter 1 24

is the change is U when we move from r to r + dr. Thus we have

rB
Z Z r
B
− ∇U (r) · dr = − dU = UA − UB
rA rA
where UA and UB are the values of the potential U at rA and rB , respectively.
Thus we have that
1 2 1
mvA + UA = mvB2 + UB
2 2
• The first term on both sides we recognise as the kinetic energy

• The second term we identify as the potential energy

The Total Energy

1
E = mv 2 + U
2
is conserved, i.e. constant in time.

We have seen that the existence of a scalar potential is associated with the irrota-
tional or conservative nature of a vector field. Where the vector field corresponds
to a force, we have a neat physical motivation for the name: a force is conservative
if the work done in going around a closed path is zero, and if a particle moves
solely under the influence of that force, then the energy is conserved.

Physical Interpretation of φ(r)

In electrostatics, the force F acting on a charge q due to an electrostatic field E

is F (r) = qE(r). Now E(r) = −∇φ(r) so that

F (r) = −∇(qφ(r)).

We have seen that the (conservative) force acting on a particle is minus the gra-
dient of its potential energy: F (r) = −∇U (r).
Chapter 1 25

The potential energy U (r) of a charge q sit-

uated at r in an electrostatic potential φ(r) is

U (r) = qφ(r). (1.7)

1.5.9 Potential Energy of Charge Distribution

For the case where φ vanishes at infinity, the potential U (r) is the work done, W ,
in bringing the charge q from infinity to the point r. We will now consider the
work done in assembling a set of point charges qi at ri , i = 1, . . . , N .
We do this by bringing each charge i in turn, one at a time, to position ri , and
then fixing it in position. The work done in bringing charge i is
qi i−1
X qj
Wi =
4πǫ0 j=1 |ri − rj |
and thus the total work done in assembling the charges is
1 X N i−1
X qi qj
W = = U,
4πǫ0 i=2 j=1 |ri − rj |
where U is the potential energy of the system. We can write this in a more
symmetric form as
1 X N X N qi qj
U=
8πǫ0 i=1 j=1 |ri − rj |
where we do not include the self-energy term, i = j.
We can generalise this to a continuous charge distribution in the usual way, viz
1 Z 3 3 ′ ρ(r)ρ(r′ )
U= d rd r ,
8πǫ0 |r − r′ |
and we now use eqn. 1.5 to write
1Z
U= ρ(r)φ(r)dV, (1.8)
2
analogous to eqn. 1.7.
Chapter 1 26

We can also interpret the potential energy in terms of the electric field, by using
ME1
ǫ0 Z
U = dV ∇ · E(r)φ(r)
2 Z
ǫ0
= − dV E(r) · ∇φ(r) (Integration by parts)
2Z
ǫ0
= dV |E|2 . (1.9)
2
We now identify the integrand as the energy density
ǫ0
u(r) = |E(r)|2 .
2
Chapter 1 27

1.6 Laplace’s and Poisson’s Equation

We are now ready to derive a differential equation for the potential. Our starting
point is Maxwell’s First Equation (ME1), derived earlier:
ρ
∇ · E = 4πkρ = .
ǫ0
We now make use of the irrotational nature of E(r) to write E = −∇φ(r). Thus
ME1 becomes

∇2 φ(r) = −4πkρ(r) = −ρ(r)/ǫ0 in SI units

where ∇2φ(r) ≡ ∇ · (∇φ(r)) ≡ ∂ 2φ(r)/∂x2i .

• This equation is Poisson’s Equation. ρ(r) is the source for the electro-
static potential φ(r).

• If we have that the source ρ(r) ≡ 0 everywhere, then this equation becomes

∇2φ = 0.

This is Laplace’s Equation.

These are two of the most important equations in physics. They, or close variants,
occur in:

• Electromagnetism, as above

• Gravitation, with k → −G, ρ the mass density, and φ the gravitational

potential

• Fluid dynamics, for the irrotational flow of a fluid.

Chapter 1 28

1.6.1 Uniqueness of Solutions of Laplace’s and Poisson’s Equation

Laplace’s and Poisson’s equations are linear, second order, partial differential
equations; to determine a solution we have also to specify boundary conditions.
Example: One-dimensional problem
d2φ(x)
=λ
dx2
for x ∈ [0, L], where λ is a constant. This has solution
1
φ(x) = λx2 + Ax + B
2
where A, B are constants. To determine these constants, we might specify the
values of φ(x = 0) and φ(x = L), i.e. the values on the boundary.

Consider the solution of Poisson’s Equation within a finite volume V , bounded

by a closed surface S. Boundary conditions are classified as:

• Dirichlet boundary conditions, where we require

φ(r) = f (r) on surface S,

i.e. we specify the value of φ(r) on the boundary. Example: Electrostatic

potential inside a conductor, with φ specified on the boundaries.

• Neumann boundary conditions, where we require

∂φ
n · ∇φ(r) = = g(r) on surface S,
∂n
where n is a unit vector normal to the surface S, i.e. we specify the normal
derivative of φ on the boundary.
Chapter 1 29

normal
n. φ
∆

Example: Electrostatic potential inside V S

S, with charge on S specified on the
boundaries.

We will proceed to show that the solutions of Laplace’s and Poisson’s are unique,
up to a constant (Neumann), if subject to either of the above boundary conditions.
We begin with a couple of useful vector identities

Green’s First Identity and Green’s Theorem

We begin with a couple of identities that will be useful both in this proof and
later.
Let ψ1 and ψ2 be two continuously differentiable, arbitrary scalar fields defined
in a volume V bounded by a closed surface S. Introduce the vector field A(r) =
ψ1∇ψ2 . From the divergence theorem, we have
Z Z
∇ · A dV = A · n dS
V S

where n is the unit outward normal to the surface S.

We now apply the vector identity
∇ · A = ψ1 ∇2ψ2 + ∇ψ1 · ∇ψ2 ,
to obtain Z Z
2
(ψ1∇ ψ2 + ∇ψ1 · ∇ψ2 ) dV = ψ1∇ψ2 · n dS. (1.10)
V
This is known as Green’s first identity.
If we write down eqn. 2.5 with ψ1 and ψ2 interchanged, and take the difference of
the two equations, we obtain
Z Z
2 2
(ψ1∇ ψ2 − ψ2 ∇ ψ1 ) dV = (ψ1∇ψ2 − ψ2 ∇ψ1 ) · n dS. (1.11)
V S
Chapter 1 30

This identity is Green’s Theorem.

1.6.2 Proof of Uniqueness of Solutions of Laplace’s and Poisson’s Equa-

tions

We now proceed to the formal proof. Let φ1 (r) and φ2 (r) be solutions of Poisson’s
equation ∇2φi = −ρ/ǫ0 inside a volume V bounded by surface S, satisfying either:

1. Dirichlet boundary conditions

φi (r) = f (r) for r on surface S

2. Neumann boundary conditions

n · ∇φi (r) = g(r) for r on surface S

where f (r) and g(r) are continuous functions defined on the surface S.
Consider the function
ψ(r) = φ1 (r) − φ2 (r).
Then ψ satisfies Laplace’s equation:

∇2ψ(r) = 0 in V

with either

1. ψ(r) = 0 for r on surface S - Dirichlet.

2. n · ∇ψ(r) = 0 for r on surface S - Neumann

We now apply Green’s first identity for the case ψ1 = ψ2 = ψ, and obtain
Z Z
2
|∇ψ| dV = (ψ∇2ψ + |∇ψ|2 ) dV (since ∇2ψ = 0 in V )
V ZV
= ψ∇ψ · n dS (from eqn. 2.5)
S
= 0 (1.12)
Chapter 1 31

since either ψ(r) = 0 or ∇ψ · n = 0 on surface S. Now |∇ψ(r)|2 is positive

indefinite, i.e.
|∇ψ(r)|2 ≥ 0
for all r ∈ V . Therefore, using equation (1.12), we have that ∇ψ(r) = 0 every-
where in V , and thus
ψ(r) = constant
for all r ∈ V .
Thus we have

• Dirichlet Problem: ψ(r) is continuous at surface S, and ψ(r) = 0 on

the surface. Therefore ψ(r) = 0 everywhere, and solution is unique.

• Neumann Problem: ∇ψ(r) · n = 0 on the surface S, and the constant

undetermined. Solution is unique up to an additive constant.

Some observations on the proof:

• We can specify either Dirichlet or Neumann boundary conditions at each

point on the boundary, but not both. To specify both is inconsistest, since
the solution is then overdetermined.

• However, we can specify either Dirichlet or Neumann boundary conditions

on different parts of the surface.

• The uniqueness property means we can use any method we wish to obtain
the solution - if it satisfies the correct boundary conditions, and is a solution
of the equation, then it is the correct solution. A good example: Method of
Images, to be covered in the next chapter.
Chapter 1 32

1.6.3 Uniqueness Theorem in an Infinite Region

We need a slight refinement of the proof if the region is infinite, ie if S contains

a “surface at infinity”.
The two solutions are the same provided they agree to O(1) at infinity.
Proof:
We merely need to show that this is a sufficient condition to ensure that the
surface integral vanishes at infinity.
Consider a sphere, radius r, area S = 4πr2 . Suppose
ψ = φ1 − φ2 = O(1/r) as r → ∞ so that ∇ ψ = O(1/r2)
Z
then ψ ∇ ψ · dS = O(1/r)
S
which vanishes as r → ∞. The remainder of the proof is unchanged.
• If the potential is due to a localised charge distribution, then, by the mul-
tipole theorem, it falls off as least as fast as (1/r) as r → ∞. Hence, the
difference ψ = φ1 − φ2 must also fall off as least as fast as (1/r), and the
uniqueness theorem applies.

• Sometimes a uniform field is specified at infinity. For example, if the uni-

form field E is in the z direction, then
φ(r) = K − E z
where K is a constant. In this case, the uniqueness theorem holds because
the ‘two’ solutions must satisfy the boundary condition
φ(r) + E z → K + O(1/r)
as r → ∞.
The next couple of chapters of this course will be concerned with solving such
boundary-value problems. We will conclude this chapter by discussing the bound-
ary conditions to impose on our solutions, and in particular the boundary condi-
tions at a conductor.
Chapter 1 33

1.7 Boundary Conditions at a Conductor

• In a conductor, electrons are able to move freely so as to set up a charge

distribution.

• In the presence of an external electrostatic field, a charge distribution is gen-

erated under the influence of this field, and itself give rise to an electrostatic
field.

• Once equilibrium is attained (about 10−18 secs. for a good conductor), no

current flows, and thus the electric field E is zero throughout the body of a
conductor.

• If the electric field vanishes in a conductor, the potential must be constant.

This provides the defining property of a conductor, namely that the boundary
of a conductor is an equipotential surface.

On the boundary of a conductor, φ(r) = const.

• Conventionally, we take φ = 0 for an earthed conductor.

• The electric field at the surface of a conductor is normal to the surface; a

tangential field would give rise to a charge flow along the surface.

1.7.1 Surface Charge Density at a Conductor

Within a conductor, the electrostatic field E must be zero. However, the field is
zero because of an induced charge density sufficient to annul the external field.
Now ME1 tells us that ∇ · E = ρ/ǫ0 , where ρ is the charge density. So if E is
zero within the conductor, the charge density must be zero. So where does the
induced charge density reside?
The charge density is confined solely to the surface of the conductor
Chapter 1 34

We can compute this surface charge density using Gauss’ Law.

Area δΑ
E

Consider applying Gauss’ Law to the infinites-

simal “pill-box” of height δh and area δA, as
shown. Within the conductor, E = 0, and at
E=0 the surface of the conductor E is normal to the
surface.
surface charge density
Therefore we have
E.n δA = δA σ/ǫ0
where σ is the surface charge density, and n is the outward normal to the surface
of the conductor.
Thus we have that the surface charge density is proportional to the discontinu-
ity in the normal electrostatic field at the conductor.

E · n = σ/ǫ0

Note: the surface charge density discussed here is different to a sheet of charge
of density σ per unit area discussed earlier in the course. The latter may best be
viewed as a charge distribution in an insulator, i.e. a fixed charge distribution.
Unfortunately, the two terms are often confused in the literature, and indeed
probably in these lectures!

1.7.2 Capacitance and Potential Energy of Conductors

Consider now a set of N isolated conductors, with charge qi , i = 1 . . . N , and with

no external electric field. Then each conductor is an equipotential φi , and the
charges reside on the surface of the conductor.
Chapter 1 35

Thus the potential energy of this system is

1Z 1
U= dV ρ(r)φ(r) = qiφi .
2 2
The potentials φi and the charges qi are not independent. In particular, for a
given set of charges qi the potentials are determined by the solutions of the field
equations. Because of the linearity of the field equations, the relationship between
the φ’s and the q’s must be linear, i.e.
N
X
φi = Pij qj ,
j=1

which in matrix form may be written

~ = P ~q.
φ

We can invert this equation to obtain

N
X
qi = Cij φj (1.13)
j=1

where, formally, C = P −1.

The diagonal elements of this matrix Cii are the capacitances, whilst the off-
diagonal elements Cij , i 6= j are the coefficients of induction. We can use
eqn. 1.13 to write the potential energy of a system of conductors in terms either
of the potentials or charges alone:
1X 1X
U= φi Cij φj ≡ qi Cij−1qj
2 ij 2 ij
Chapter 2

Boundary-Value Problems in
Electrostatics

In this chapter we will examine solutions to Poisson’s and Laplace’s equations

in electrostatics. Before we proceed to a formal solution of Poisson’s equation,
we will look at a few simple solutions. In the next section we will exploit the
uniqueness theorem in a particularly neat way through the Method of Images, but
first, back to Gauss’ Law for a simple example. . .

Example: Charged sphere inside grounded, conducting shell.

A sphere of radius a, carrying a charge Q, is placed inside an grounded, conducting
sphere of radius b (b > a). Find the potential in the region a ≤ r ≤ b.

Φ = 0 on surface

Q
a
O

1
Chapter 2 2

Thus we have to solve Poisson’s equation, subject to the boundary conditions

φ(r) = 0 for r = b. Apply Gauss’ Law to the region a < r < b:
Q
E(r) = er ; a≤r≤b
4πǫ0 r2
for which the potential is
Q
φ= + φ0 ; a≤r≤b
4πǫ0 r
where φ0 is a constant.
The boundary conditions tell us that φ vanishes at r = b. Thus we have
Q 1 1
!
φ= − ; a ≤ r ≤ b.
4πǫ0 r b
Let us check that our solution for φ(r) satisfies Poission’s equation for a ≤ r ≤
b. We are implicitly working in spherical polars (r, θ, ψ), therefore (from your
favourite vector-calculus course, or back of Jackson):
∂ 2φ 
 
1 ∂ 2 ∂φ 1 ∂ ∂φ
! !
∇2φ(r, θ, ψ) = 2

r + 2 2 sin θ sin θ +
r ∂r ∂r r sin θ ∂θ ∂θ ∂ψ 2 
Q 1 ∂ 2 −1
( !)
= r
4πǫ0 r2 ∂r r2
Q 1 ∂  
= − 1 = 0
4πǫ0 r2 ∂r
Hence φ(r) satisfies ∇2 φ(r) = 0 in the charge free region a ≤ r ≤ b, and satisfies
the boundary condition φ(b) = 0 on the surface. Therefore, it is the unique
solution of Poisson’s equation in this region. Of course, due to spherical symmetry,
φ(r) doesn’t depend on θ or ψ, and therefore the calculation of the ∇2φ(r) is
particularly simple.
Finally, let us find the surface charge density on the conductor. At the boundary
of the conductor,
Q
E= er
4πǫ0 b2
Chapter 2 3

Thus the surface charge density is given by

Q
σ=−
4πb2
which is negative, as expected. Indeed the total induced charge on the conductor
is equal and opposite to that of the charge distribution.

Once again, the method was particularly simple in this case because of spherical
symmetry. Similar simplifications occur in the case of cylindrical symmetry.

2.1 Method of Images

The uniqueness property of the solutions of Laplace’s and Poisson’s Equations

leads to a neat method of obtaining their solution in particular geometric cases.

Consider a charge q placed at r1 = hk above q r - r1

P
an infinite grounded conducting plane at
z = 0, as shown on the right. Then on the
r1 r
conducting plane the potential must vanish.

φ=0
Chapter 2 4

Now consider a system with a charge q q r - r1

P
placed at r1, and a charge −q placed at −r1
in the absence of the conducting plane, as
r1 r
shown on the right. The potential φ(r) is
q 1 −q 1
r + r1
φ(r) = + .
4πǫ0 |r − r1| 4πǫ0 |r + r1|
φ=0
At z = 0, the potential vanishes because
here points are equidistant from the posi-
tive and negative charges. Furthermore, in
the upper half plane φ must satisfy Poisson’s
equation for a point charge at r1, since no -q
further changes have been introduced in this
region (the only charge we have introduced
is in the lower half plane).

Thus, by our uniqueness theorem, the potential in the upper half plane is the
same as that of a charge q placed above an grounded sheet at z = 0.
Chapter 2 5

2.1.1 Point Charge near grounded Sphere

φ=0 11
00
11
00 P
Consider a point charge q placed at r
1
0

a distance b from the centre of an

grounded conducting sphere of radius O θ Q
a < b. We will now show that an a q’ q
equivalent problem is to place an im- b’
age charge q ′ = −qa/b as shown.
b
By symmetry, the image charge q ′ must lie along OQ, at a distance b′, say, from
the centre of the sphere. Thus the resultant potential of the image system is
q′ 
 
1  q
φ(r) = + .
4πǫ0  |r − b| |r − b′ | 
We need two equations to determine q ′ and b′ ; we will obtain these by imposing
that φ vanish at the two points where OQ intersects the sphere
q′ 
 
1  q
+ = 0
4πǫ0  b − a a − b′ 
q′ 
 
1  q
+ = 0.
4πǫ0  a + b a + b′ 
Eliminating q ′ , we obtain
a + b′ a+b
=
a − b′ b−a
and hence
b′ = a2 /b.
We can substitute this into either equation to obtain
q ′ = −qa/b.
Finally, let us verify that φ does indeed vanish for all points on the surface of the
sphere. On the surface,
a2 a4
|r − b′ |2 = a2 − 2a cos θ + 2
b b
Chapter 2 6

a2 n 2
= 2 a − 2ab cos θ + b2
o

b
a2
= 2 |r − b|
b
and hence  
1  q qa 1 
φ(r)|r=a = − = 0.
4πǫ0  |r − b| b a/b|r − b| 
Thus we have

1. The image system satisfies the original Poisson’s equation for r ≥ a since the
only additional charge we have introduced is in the region r < a.

2. The potential for the image system satisfies the condition φ = 0 at r = a.

Thus, by the uniqueness theorem, the required potential is

 
1  q qa 1 
φ(r) = − (2.1)
4πǫ0  |r − b| b |r − b′ | 

with b′ = a2 /b.

Induced charge density

In Chapter 1, we showed that the induced charge density on the surface of a

conductor is
σ = ǫ0 E · n = −ǫ0 n · ∇φ
where n is the outward normal to the surface.
From eqn. 2.1, we have
 
1  −q qa 1 2 2

∇φ = (r − b) + (r − a /b b)
4πǫ0  |r − b|3 b |r − a2 /b2b|3 

At the surface, r = a = aer and n = er , yielding


¯ ¯ 
1  −q qa 1 2 2

σ=−  (a − b · e ) + (a − a /b b · e ) .
4π |a − b|3 ¯r b |a − a2 /b2b|3 ¯r 
Chapter 2 7

Using
|a − a2 /b2b|r=a = a/b|a − b|r=a
we find
q a n
2 2
o
σ=− b /a − 1 a.
4π |a − b|3
Note that the surface charge density is not uniform, but that
Z
σ dS = q ′
S

as expected.

2.1.2 Point charge near insulated conducting sphere at potential V

This is a simple modification of the method above. We introduce an additional

image charge q̂ = V a4πǫ0 at the centre of the sphere yielding φ = V at r = a.
Because we have introduced no additional charges in the region r ≥ a, we apply
the uniqueness theorem to say that the resultant potential is
 
1  q qa 1  Va
φ(r) = − + .
4πǫ0  |r − b| b r − a2 /b2b|  r

2.1.3 Point charge near insulated, conducting sphere with total charge
Q

This problem is a slightly more complicated. Our starting point is the point charge
near the grounded conducting sphere, together with the superposition principle.

1. Start with an grounded conducting sphere. We have shown that a total

surface charge q ′ is induced, distributed to balance the electrostatic forces
due to q.

2. Disconnect the sphere from earth, and add a charge Q−q ′ to the sphere. This
charge will be uniformly distributed, since the charge q ′ is already distributed
to balance the forces due to q.
Chapter 2 8

Appealing to the uniqueness theorem, and noting that, once again, no charges
have been introduced in r ≥ a, we have
Q − q′ q′
 
1  q 
φ(r) = + + .
4πǫ0 r 4πǫ0  |r − b| r − a2 /b2b| 
We will now proceed to calculate the Force on the charge q; this is just given by
Coulomb’s law for the forces between q and the two image charges:
1  Q − q′ q′
 
′

F = q 3 b+ (b − b )
4πǫ0 b |b − b′ |3 

qa3 (2b2 − a2 ) 
 
1 qb
= Q−
4πǫ0 b3  b(b2 − a2 )2 
Note that the force is always attractive at sufficiently small distances irrespec-
tive of Q due to the induced surface charge density on the conductor.
Chapter 2 9

2.2 Formal solution of Poisson’s Equation: Preliminaries

We will now proceed to a formal solution using Green functions. First, however,
a mathematical digression. . .

2.2.1 Dirac δ-Function

The Dirac δ-function is defined as follows:

1.

δ(x − a) = 0 if x 6= a.

2.

Z 
 1 if a ∈ R
dx δ(x − a) = 
R  0 otherwise

The delta function is not strictly a function but rather a distribution; it is de-
fined purely through its effect under an integral. It immediately follows from the
definition that Z
dx f (x)δ(x − a) = f (a) (2.2)
if a lies within the region of integration.
The δ-function δ(x − a) may be thought of as the limit of a Gaussian centred at
a in which the width tends to zero whilst the area under the Gaussian remains
unity.

δ(x − a) = lim δǫ (x − a)
ǫ→0
1 − (x−a)2
δǫ (x − a) = √ e ǫ (2.3)
πǫ
R∞
It is easy to see that limǫ→0 δǫ (x) = 0 if x 6= a and −∞ δǫ (x − a) = 1. Let us check
the property (2.2)
Z ∞
lim δǫ(x − a) f (x)
ǫ→0 −∞
Chapter 2 10

1
Z ∞ (x−a)2 1
= lim √ e− ǫ [f (a) + (x − a)f ′(a) + (x − a)2 f ”(a) + ...]
ǫ→0 −∞ πǫ 2
2 4
= lim [f (a) + ǫ f ”(a) + O(ǫ )] = f (a)
ǫ→0

There are some simple relations that follow from the Eq. (2.2)
1. The δ-function is a derivative of a step function θ(x):


 1 x≥0
θ(x) = (2.4)

 0 x<0
Indeed, if f (x) vanishes at infinity
Z ∞ Z ∞ Z ∞
′ ′
f (x) θ (x) = − f (x) θ(x) = − f ′ (x) = f (0)
−∞ −∞ 0
Z ∞
f (x) δ(x) = f (0)
−∞

⇒ δ(x) = θ′ (x) (and δ(x − a) = θ′ (x − a)).

2.
Z Z
′
dx f (x)δ (x − a) = − dx f ′(x)δ(x − a) integ. by parts
= −f ′ (a)

3.
Z XZ 1
dx f (x)δ(g(x)) = dy f (xi(y))δ(y)
i g ′ (xi(y))
X f (xi)
= ′
i |g (xi )|

where y is defined in a small region of each xi.

4. The definition extends naturally to three (or higher) dimensions:

δ(x − X) = δ(x1 − X1 )δ(x2 − X2 )δ(x3 − X3 )

so that 
Z 
1 if X ∈ V
d3 x δ(x − X) = 


V  0 otherwise
Chapter 2 11

Note that it is this last property that defines the multi-dimensional δ-function,
with this simple representation in a Cartesian basis; you have to be a little
careful when working in curvilinear coordinates.
As a simple illustration of the power of the δ-function, let us return to the expres-
sion, eqn. (1.5), for the potential due to a continuous charge distribution
1 Z 3 ′ ρ(x′ )
φ(x) = dx .
4πǫ0 V |x − x′ |
We now introduce the δ-function to enable us to write a set of N discrete charges
qi at xi as a charge distribution

ρ(x′ ) = qi δ (3) (x′ − xi )

X

i
so that
1 Z 3 ′ i qiδ (3) (x′ − xi )
P
φ(x) = dx
4πǫ0 V |x − x′ |
1 X qi
=
4πǫ0 i |x − xi |
which is our familiar expression for the potential due to a set of point charges.

Poisson’s Equation for a Point Charge

It is easy to see that

∇2 (1/r) = 0 r 6= 0.
Furthermore, from our proof of Gauss’ law, we can see that
Z
dV ∇2(1/r) = −4π.

Thus we can write formally

 
1 
∇2  ′
= −4πδ (3) (x − x′ )
|x − x |
Chapter 2 12

2.3 Formal Solution of Boundary-Value Problem using Green

Functions

Our starting point is Green’s theorem, eqn. (1.11):

Z Z
3 ′ ′ ′2 ′ ′ ′2 ′
d x (ψ1(x )∇ ψ2(x )−ψ2 (x )∇ ψ1 (x )) = (ψ1(x′ )∇′ ψ2 (x′ )−ψ2(x′ )∇′ ψ1 (x′ ))·n dS.
V S

where the “primed” denotes differentiation with respect to the primed indices.
Let us apply this for the case ψ1 (x′ ) = |x−1x′ | and ψ2 (x′ ) = φ(x′ ) where

∇′2φ(x′ ) = −ρ(x′ )/ǫ0.

and
∇′2ψ1(x′ ) = −4πδ (3) (x − x′ )
yielding
−ρ(x′ ) 
   
Z
3 ′
 1 ′ (3) ′

dx   + φ(x )4πδ (x − x ) =
|x − x′ | ǫ0 
  
Z 1 ′ ′ 1
dS ′ n ·  ′ ′
 
∇ φ(x ) − φ(x )∇  .
|x − x′ | |x − x′ | 
Applying our rule for integrating over δ-functions, we obtain
1 Z 3 ′ ρ(x′ )
φ(x) = dx +
4πǫ0 |x − x′ |
′
  
1 Z 1 ∂φ(x ) ∂ 1
dS ′  ′
 
− φ(x )   . (2.5)
4π |x − x′ | ∂n′ ∂n′ |x − x′ | 
The function 1/|x − x′ | is said to be a Green function for the problem.
The Green function is not unique, and is just a function satisfying

∇′2G(x, x′ ) = −4πδ (3) (x − x′ ).

In general, it has the form

1
G(x, x′ ) = ′
+ F (x, x′ ),
|x − x |
Chapter 2 13

where F (x, x′ ) is a solution of Laplace’s equation

∇′2F (x, x′ ) = 0.
Thus our expression for the potential can be generalised to

1 Z 3 ′
φ(x) = d x G(x, x′ )ρ(x′ ) +
4πǫ0 V
′ ′ 
 
1 Z
∂φ(x ) ∂G(x, x )
dS ′ G(x, x′ ) − φ(x′ )

′
(2.6)
4π S=∂V ∂n ∂n′ 

The utility of this generalisation is the following. In eqn. 2.5, the surface integral
involved both φ(x′ ), and ∂φ(x′ )/∂n′; in general we cannot specify both simultane-
ously at a point on the surface, since the problem is then overdetermined. Thus in
eqn. 2.5 we have an implicit equation for φ(x), with the unknown also appearing
under the integral on the right-hand side. In eqn. 11.79, we can choose G(x, x′ )
so that the surface integral depends only on the proscribed boundary values of φ
(Dirichlet) or ∂φ/∂n′ (Neumann).

2.3.1 Boundary Conditions on Green Functions

We will now consider the boundary conditions we have to impose on the Green
Functions to accomplish the above aim.

Dirchlet Problem

Here the value of φ(x′ ) is specified on the surface, and therefore it is natural to
impose that the Green function GD (x, x′ ) satisfy
GD (x, x′ ) = 0 for x′ on S,
and thus
′
1 Z 3 ′ ′ ′ 1 Z ′ ′ ∂GD (x, x )
φ(x) = d x GD (x, x )ρ(x ) − dS φ(x ) . (2.7)
4πǫ0 V 4π S ∂n′
Thus the surface integral only involves φ(x′ ), and not the unknown ∂φ(x′ )/∂n′.
Chapter 2 14

Neumann Problem

Here it is tempting to construct the Green function GN (x, x′ ) such that

∂GN (x, x′ )
′
= 0 for x′ on S.
∂n
However, recall that the Green function satisfies
Z ∂GN (x, x′ ) Z 3 ′ ′2
′
dS ′
= d x ∇ GN (x, x′ ) = −4π,
S ∂n
and thus ∂GN (x, x′ )/∂n′ cannot vanish everywhere. The simplest solution is to
impose
∂GN (x, x′ ) 4π ′
= − ∀x ∈ S
∂n′ S
where S is the total area of the surface. Thus the solution is
1 Z 3 ′
φ(x) = d x GN (x, x′ )ρ(x′ ) +
4πǫ0 V
′
1 Z ′ ′ ∂φ(x ) 1Z
dS GN (x, x ) ′
+ dS ′ φ(x′ ) (2.8)
4π S ∂n S S

where the final term is just the average value of φ(x′ ) on the surface S. The
inclusion of this term is perhaps not surprising; recall that the solution to the
Neumann problem is unique only up to an additive constant.

2.3.2 Reciprocity relation for GD (x, y)

For the Dirichlet problem, we have GD (x, y) = GD (y, x).

Proof

Apply Green’s theorem for the case ψ1 (x′ ) = GD (x, x′ ), and ψ2 (x′ ) = GD (y, x′ ):
Z
d3x′ (GD (x, x′ )∇′2GD (y, x′ ) − GD (y, x′ )∇′2GD (x, x′ )) =
V Z
dS ′ n · (GD (x, x′ )∇′ GD (y, x′ ) − GD (y, x′ )∇′ GD (x, x′ )).
S
Chapter 2 15

But for the Dirichlet problem GD (x, x′ ) vanishes for all x′ ∈ S, and hence the
right-hand side of the above is zero. Thus we have
Z
d3 x′ GD (x, x′ ){−4πδ (3) (y − x′ )} − GD (y, x′ ){−4πδ (3) (x − x′ )} = 0
n o

and hence

GD (x, y) = GD (y, x)
Chapter 2 16

2.4 Methods of Finding Green Functions

The secret, then, to the solution of boundary value problems is determining the
correct Green function, or equivalently obtaining the function F (x, x′ ). They are
several techniques

1. Make a guess at the form of F (x, x′ ). Here we recall that F is just the
solution of the homogeneous Laplace’s equation ∇′2F (x, x′ ) = 0 inside V , and
therefore is just the solution of the potential for a system of charges external
to V . In particular, for the Dirichlet problem, since GD (x, x′ ) vanishes at
x′ ∈ S, we have that F (x, x′ ) is just that system of charges external to V
that, when combined with a point charge at x, assures that the potential
vanishes on the surface. And finding that system of charges is precisely what
we were doing in the Method of Images. . .

2. Expand the Green function as a series of orthonormal eigenfunctions of the

Laplacian operator. We will be exploring this method later in the chapter.

2.4.1 Dirichlet Green Function for the Sphere

We saw at the beginning of this chapter how to use the method of images to con-
struct the potential φ(x′ ) for a point charge at x outside an grounded conducting
sphere of radius a. In particular, for a charge q = 4πǫ0 , the potential satisfies

∇′2φ(x′ ) = −4πδ (3) (x − x′ )

with φ(x′ ) = 0 for x′ on S. Thus now see that φ(x′ ) is precisely the Green func-
tion GD (x, x′ ) that we need. Note that you have to be careful to distinguish the
variable we are integrating over, x′ , and the variable at which we are evaluating
the potential, x. Perhaps counter-intuitively, it is at the point x that we place
our point charge.
Chapter 2 17

z x
x’
θ From eqn. 2.1, we have that the Green function
γ is
11111111111111111111
00000000000000000000 1 a
00000000000000000000
11111111111111111111
00000000000000000000
11111111111111111111 G(x, x′ ) = − ,
00000000000000000000
11111111111111111111 |x′ − x| x|x − a2 /x2x|
′
00000000000000000000
11111111111111111111
and it is easy to check that, indeed, G(x, x′ ) =
G(x′ , x).

ϕ
We can rewrite this as
 
1 1
G(x, x′ ) =  2
 
−
(x + x′2 − 2xx′ cos γ)1/2 x2x′2/a2 + a2 − 2xx′ cos γ)1/2 
where γ is the angle between x and x′ .
The general solution for the potential is then
1 Z 3 ′ 1 Z ∂G(x, x′ )
φ(x) = d x G(x, x′ )ρ(x′ ) − dS ′ φ(x′) . (2.9)
4πǫ0 V 4π S ∂n′
Thus we need the normal gradient of the Green function to the surface, which
points inward,
∂G ∂G
= −
∂n′ surface ∂x ′ ′
 x =a
2x2a/a2 − 2x cos γ

1 −2a − 2x cos γ 
= −  2 +
2 (x + a2 − 2ax cos γ)3/2 (x2a2 /a4 + a2 − 2ax cos γ)3/2 
x2 − a2
= −
a(x2 + a2 − 2ax cos γ)
Thus we have all the ingredients to solve the Dirichlet problem outside a sphere
of radius a.
Chapter 2 18

2.4.2 Solution of Laplace’s equation outside a sphere comprising two

hemispheres at equal and opposite potentials V

Because the source is zero, we only need the surface term from eqn. 2.9
′
1 Z ′ ′ ∂G(x, x )
φ(x) = − dS φ(x ) .
4π S ∂n′
Now dS ′ = a2 dψd(cos θ′), yielding
1 2 Z 2π ′ ′ ∂G ∂G
(
Z 1 Z 0 )
φ(x) = − a dψ V d(cos θ ) ′ + (−V ) d(cos θ′ ) ′
4π 0 0 ∂n −1 ∂n
2 2

V Z 2π  1
Z
a(x − a )
= dψ ′  d(cos θ′ ) 2 −
4π 0 0 (a + x2 − 2ax cos γ)3/2
2 2

Z 0 a(x − a )
d(cos θ′) 2

.
−1 (a + x2 − 2ax cos γ)3/2 
We can express cos γ in terms of the spherical polar coordinates of x and x′ by
noting that

cos γ = n · n′ = (sin θ cos ψ, sin θ sin ψ, cos θ) · (sin θ′ cos ψ ′ , sin θ′ sin ψ ′ , cos θ′ )
= sin θ sin θ′ cos(ψ − ψ ′ ) + cos θ cos θ′ ,

where n and n′ are unit vectors in the directions of x and x′ respectively. Finally,
we can combine the two integrals through by making the substitution θ′ → π − θ′
and ψ ′ → ψ ′ + π in the second integral, giving


V 2 2
Z 2π
′
Z 1 1
d(cos θ′ )  2

φ(x) = a(x − a ) dψ
4π 0 0 (a + x2 − 2ax cos γ)3/2

1 
− 2
(a + x2 + 2ax cos γ)3/2 

In general, we cannot obtain the solution in closed form; γ is just too complicated
a function of θ′ and ψ ′ . However, we can study the solution in specific cases.
Chapter 2 19

Solution above North Pole

Here θ = 0, so that cos γ = cos θ′, and |x| = z. Thus

 
V 2 2
Z 1  1 1 
φ(z) = a(z − a )2π du  (a2 + z 2 − 2azu)3/2
− .
4π 0 (a2 + z 2 + 2azu)3/2 
The integration can be performed easily, by making the substitution y = a2 + z 2 −
2azu and y = a2 + z 2 + 2azu for the first and second terms respectively, yielding
(z 2 − a2 ) 
 

φ(z)|θ=0 = V 1 − √ 2 .
z z + a2 
Note that for z ≫ a, we have
3V a2
φ(z) ∼ ,
2z 2
and the boundary conditions are trivially satisfied at z = a.

Solution at Large Distances

We can also obtain the solution for x >> a, by means of a Taylor expansion. We
begin by writing

a2 + x2 ± 2ax cos γ = (a2 + x2)(1 ± 2α cos γ)

where
ax
α= ,
a2 + x2
yielding
V a(x2 − a2 ) Z 2π ′ Z 1
 
′
 1 1 
φ(x) = dψ d(cos θ ) − .
4π (a2 + x2)3/2 0 0  (1 − 2α cos γ)3/2 (1 + 2α cos γ)3/2 
We now expand the integrand as a power series in α, yielding

{} = 6α cos γ + 35α3 cos3 γ + O(α5 ).

The integrals for the first two terms in the expansion are perfectly tractable.
Recalling that cos γ = sin θ sin θ′ cos(ψ − ψ ′ ) + cos θ cos θ′ , we find
Chapter 2 20

1.
Z 2π Z 1 Z 2π Z 1
′ ′ ′
dψ d(cos θ ) cos γ = dψ d(cos θ′) cos θ cos θ′ = π cos θ
0 0 0 0

2.
Z 2π Z 1
′
dψ d(cos θ′ ) cos3 γ = π/4 cos θ(3 − cos2 θ)
0 0

and thus
3V a2 x(x2 − a2 ) 35 a2 x2
 
2 4 4
 
φ(x) = cos θ 1 + (3 − cos θ) + O(a /x ) .
2(a2 + x2 )5/2  24 (a2 + x2)2 

Note that we can express this power series as a series in a2 /x2, rather than α,
yielding
3V a2  7a2 5
 
3
!
3 4 4

φ(x, θ, ψ) = cos θ − cos θ − cos θ + O(a /x ) .
x2  12x2 2 2 

and we can verify that this gives the correct expression for θ = 0.
As we go to higher order terms in the expansion, the angular integrals become
increasingly intractable, and this approach fails. However, the eagle-eyed amongst
you may recognise the angular terms as the Legendre polynomials P1 (cos θ) and
P3 (cos θ), and this brings us to the next section.

2.5 Orthogonal Functions

The expansion of the solution of a linear differential equation in terms of orthog-

onal functions is one of the most powerful techniques in mathematical physics.
Consider a set of functions Un (ξ), n = 0, 1, . . ., defined on a ≤ ξ ≤ b.

1. The set {Un(ξ)} is orthonormal iff (if and only if )

Z b
∗
dξ Un (ξ)Um (ξ) = δmn (2.10)
a
Chapter 2 21

2. The set is set to be complete iff

∞
Un (ξ)Un∗(ξ ′ ) = δ(ξ − ξ ′).
X
(2.11)
n=0

The completeness relation is important because it implies that any square-integrable

function f (ξ) defined over the interval a ≤ ξ ≤ b can be expressed as a series in
the orthogonal functions U (ξ). This is easy to see:
Z
f (ξ) = dξ ′ f (ξ ′)δ(ξ − ξ ′ ) (defn. of δ-func.)
Z ∞
dξ ′ f (ξ ′) Un (ξ)Un∗(ξ ′ ) (completeness)
X
=
n=0
∞ Z
dξ ′ f (ξ ′)Un∗(ξ ′ ).
X
= Un (ξ)
n=0
Thus we may write
∞
X
f (ξ) = Un (ξ)an
n=0
where Z
an = dξ ′ f (ξ ′)Un∗(ξ ′).

2.5.1 Fourier Series

One of the best-known cases where we expand in terms of orthogonal functions

is the Fourier expansion. Consider the expansion applied to the interval −a/2 ≤
x ≤ a/2. The set of ortonormal functions is provided by the sines and cosines:
2πmx
q !
Cm(x) = 2/a cos , m = 1, 2, . . .
a !
q 2πmx
Sm (x) = 2/a sin , m = 1, 2, . . .
q
a
C0(x) = 1/a.
It is easy to show that the set Cm (x), Sm(x) forms an orthonormal set of functions,
viz.
Z Z
dx Sm (x)Sn(x) = dx Cm(x)Cn(x) = δmn ,
Z
dxSm(x)Cn(x) = 0.
Chapter 2 22

Later we will prove completeness,

2πmx′  2 X 2πmx′ 
   
1 2X∞ 2πmx
! ∞ 2πmx
!
+ cos cos  + sin sin  = δ(x − x′ )
a a 1 a a a 1 a a
(2.12)
and thus we can write any function f (x) on the interval −a/2 ≤ x ≤ a/2 as
2πmx′  2πmx′ 
    
A0 X∞ 
f (x) = + A cos
 m
 + Bm sin  ,
2 m=1 a a 

where
2 Z a/2 2πmx
!
Am = dx f (x) cos m = 0, 1, 2, . . .
a −a/2 a !
2 Z a/2 2πmx
Bm = dx f (x) sin m = 1, 2, . . .
a −a/2 a
We can combine the sine and cosine terms by noting
1 h ix
e + e−ix
i
cos x =
2
1 h ix
e − e−ix ,
i
sin x =
2i
and introducing a new set of functions
1
Um (x) = √ ei2πmx/a m = 0, ±1, ±2, . . . ,
a
We get an expansion
∞
X
f (x) = Am Um (x),
m=−∞
where
1 Z a/2 ′
Am = √ dx′f (x′)e−2πimx /a .
a −a/2

Proof of completeness
∞ ′
ein(x−x ) = 2πδ(x − x′ )
X

−∞
for x, x′ ∈ [−π, π]:
Chapter 2 23

For simplicity, take x instead of x − x′. We have

∞
X inx
∞
X inx
∞
X −inx 1 e−ix
e = e + e = + = 0
−∞ 0 1 1 − eix 1 − e−ix
if x 6= 0. To check for the δ-function contribution, calculate
∞
Z π X ∞ Z π
einx = einx = 2π
X
−π −∞ −∞ −π

⇒ ∞ −∞ e
inx
= 2πδ(x), Q.E.D.
P

For the interval [−a/2, a/2] we get:

∞ 2π ′
ein a (x−x ) = aδ(x − x′ )
X
(2.13)
−∞

Taking the real part of both sides of this equation we reproduce Eq. (2.12).

An orthonormal set sin πa mx

If we have to expand a function f (x)[0, a] → R which vanishes at the ends of the

interval [0, a] we can use an orthonormal set of sin′ s only: Un(x) = a2 sin πa nx.
q

It is easy to check that

2Za π π
dx sin mx sin nx = δmn (2.14)
a 0 a a
and
2X∞ π π
sin nx sin nx′ = δ(x − x′ ) (2.15)
a 1 a a
(Strictly speaking, in the r.h.s of the eq. (2.15) we get δ(x − x′) − δ(x + x′) but
the last term does not contribute for x, x′ ∈ [0, a]).
Thus, we get an expansion
q ∞
X π
f (x) = 2/a fn sin nx
1 a
q Z a π
fn = 2/a dx f (x) sin nx (2.16)
0 a
Chapter 2 24

2.5.2 Fourier transformation

Suppose we now let a → ∞, so that the discrete sum over m becomes an integral
over a continuous variable k where
2πm
→ k.
a
Then we have
X a Z
→ dk
m 2π
and the discrete coefficients become a continuous function
v
u 2π
u
Am → t
A(k).
a
Thus we may express the Fourier Transforms as
1 Z
f (x) = √ dk A(k)eikx
2π
1 Z
A(k) = √ dx f (x)e−ikx.
2π
Note that the assignment of the coefficients outside the integrals depends on the
convention adopted; in all cases the product is 1/2π.
The orthogonality and completeness relations assume the continuous, and sym-
metric, forms
1 Z∞ ′
dx ei(k−k )x = δ(k − k ′ )
2π −∞
1 Z ′
dk eik(x−x ) = δ(x − x′ )
2π

2.5.3 Sturm-Liouville Equation

How does one obtain a complete set of orthonormal functions? We will now show
that, for a certain class of differential equations, the solutions are orthogonal, for
specific boundary conditions.
Chapter 2 25

The Sturm-Liouville Equation is the differential equation

d2ψλ dp(x) dψλ
p(x) 2 + + q(x)ψλ (x) = −λr(x)ψλ(x)
dx dx dx
which we may write in the more compact form
d dψλ
" #
p(x) + q(x)ψλ = −λr(x)ψλ .
dx dx
Here the parameter λ identifies the solution, and plays the rôle of an eigenvalue,
with ψλ the corresponding eigenvector. In the next couple of lectures we will
encounter several equations of this form - the Legendre and Bessel equations,
and of course you are familiar with the time-independent Schrödinger equation.

2.5.4 Theorem

For the Sturm-Liouville equation, with p, q, r real functions of x, the integral

Z b
∗ ′
(λ − λ ) dxr(x)ψλ∗ (x)ψλ′ (x)
a

is zero provided the following boundary condition is satisfied:

!#b
dψλ′ dψλ∗
"
p(x) ψλ∗ − ψλ′ = 0.
dx dx a

Proof

ψλ and ψλ′ satisfy

d dψλ
" #
p(x) + q(x)ψλ = −λr(x)ψλ (2.17)
dx" dx #
d dψλ′
p(x) + q(x)ψλ′ = −λ′ r(x)ψλ′ , (2.18)
dx dx
respectively. Multiplying eqn. 2.17 by ψλ∗′ and eqn. 2.18 by ψλ∗ and integrating,
we obtain
∗ d dψλ
Z b " # Z b Z b
ψλ′ p(x) + dx ψλ∗′ qψλ = −λ dx ψλ∗′ rψλ
a dx " dx # a a
Z b
d dψλ′ Z b Z b
ψλ∗ p(x) + dx ψλ∗ qψλ′ = −λ′ dx ψλ∗ rψλ′ .
a dx dx a a
Chapter 2 26

Integrating by parts yields

b
dψλ∗′ dψλ Z b ∗ ∗ dψλ
Z b Z " #
− dx p + ψλ′ qψλ = − pψλ′ − λ dx ψλ∗′ rψλ (2.19)
a dx dx a dx a
#b
dψλ∗ dψλ′ Z b ∗ ∗ dψλ′
Z b " Z
− dx p + ψλ qψλ = − pψλ
′ − λ dx ψλ∗ rψλ′
′
(2.20)
a dx dx a dx a
Observing that, since q, p, r are real, the l.h.s. of eqn. 2.19 is the complex conjugate
of the l.h.s. of eqn. 2.20 we can take the difference to obtain
Z
∗ ′
(λ − λ ) dx r(x)ψλ∗ ψλ′ = 0,

providing
!#b
dψλ′ dψλ∗
"
p(x) ψλ∗ − ψλ′ = 0.
dx dx a

Corollaries

1. If r(x) does not change sign in (a, b)

Z b
r(x)|ψλ|2 6= 0
a

and hence λ∗ = λ.

2. For λ′ 6= λ, Z b
dxr(x)ψλ∗ ψλ′ = 0,
a
i.e. the functions ψλ are orthogonal.
Chapter 2 27

2.6 Separation of Variables in Cartesian Coordinates

We will now see how the Sturm-Liouville equation arises in the solution of Laplace’s
equation, and how we can then use the Sturm-Liouville theorem to provide an
orthonormal set of functions. The method we will use will be the separation of
variables. It is best shown by illustration.
Consider the solution of Laplace’s equation in a box 0 ≤ x ≤ a, 0 ≤ y ≤ b,
0 ≤ z ≤ c, with the values of the potential prescribed on the boundary. In
particular, let us consider the case where φ vanishes on the boundary, except on
the plane z = c where φ(x, y, z = c) = V (x, y).
In Cartesian coordinates, the natural coordinate system for the problem, Laplace’s
equation assumes the form
∂2 ∂2 ∂2
φ(x, y, z) + 2 φ(x, y, z) + 2 φ(x, y, z) = 0.
∂x2 ∂y ∂z
We will seek solutions to this equation that are factorisable, i.e.

φ(x, y, z) = X(x)Y (y)Z(z),

and build up our final solution from such factorisable solutions. Substituting this
form into Laplace’s equation, we obtain
d2X(x) d2Y (y) d2Z(z)
Y (y)Z(z) + X(x) X(z) + X(x)Y (y) = 0,
dx2 dy 2 dz 2
which we may write as
1 d2 X 1 d2 Y 1 d2 Z
+ + = 0.
X dx2 Y dy 2 Z dz 2
We have separated the equation into three terms, each dependent on a different
variable. Since the equation holds for all x, y, z, we can say that each term must
separately be constant. Thus
1 ′′
X = C1 (2.21)
X
Chapter 2 28

1 ′′
Y = C2 (2.22)
Y
1 ′′
Z = C3 (2.23)
Z
where C1 + C2 + C3 = 0.
Let us consider eqn. 2.21
d2X(x)
− C1X = 0,
dx2
and choose a trial solution
X(x) = eαx .
Then we have that α2 = C1 .

1. If C1 > 0, α is real, and the trial solution is exponential.

2. If C1 < 0, α is imaginary, and the trial solution is oscillatory.

The boundary conditions require that X vanish at x = 0, a, and this is only

possible for the oscillating solutions. Thus if we choose C1 = −α2, where α real,
the general solution will be of the form

X(x) = A cos αx + B sin αx.

Since X must vanish at x = 0,

X(x) = sin αx.

Furthermore, X also vanishes at x = a, and thus

nπ
α = αn = , n = 1, 2, . . . .
a
Thus we have a set of solutions

Xn (x) = sin αn x.

Eqn. 2.21 is a Sturm-Liouville equation, with p(x) = 1, q(x) = 0, r(x) = 1 and

λ = α2 . It satisfies the conditions required for the Sturm-Liouville theorem, and
Chapter 2 29

hence we immediately know that the functions Xn (x) are orthogonal. We can
treat Y (y) similarly, and obtain
mπ
Ym (y) = sin βm y; βm = , m = 1, 2, . . .
b
Finally, we obtain Z from
Z ′′ 2 2 n2π 2 m2 π 2
= αn + βm = 2 + 2 > 0.
Z a b
In this case, the solution is a real exponential, and imposing the boundary condi-
tion Z(0) = 0 we have
Z(z) = sinh γnmz
where
q
γnm = π n2/a2 + m2 /b2.
Thus the general solution, using the completeness property, is
∞
X
φ(x, y, z) = Anm sin αn x sin βm y sinh γnmz.
m,n=1

We obtain the coefficients Amn by imposing the boundary conditions on the plane
z = c:
∞
X
V (x, y) = Anm sin αn x sin βm y sinh γnmc.
m,n=1

Using the orthonormal property of the basis functions, we have

Z a
nπx Z b mπy
dx sin dy sin V (x, y)
0 a 0 b
X Z a nπx n′ πx Z b mπy m′ πy
= An′ m′ dx sin sin dy sin sin sinh γn′ m′ c
m ,n
′ ′ 0 a a 0 b b
X a b
= An′ m′ δn′ n δm′ m sinh γn′m′ c
n′ ,m′ 2 2
ab
= Anm sinh γnm c
4
Thus we have
Chapter 2 30

4 Z a Z b
Anm = dx dyV (x, y) sin αn x sin βm y
ab sinh γnmc 0 0

2.6.1 Two-dimensional Square Well

φ finite

This is the two-dimensional version of the

above problem. We have a square well, of
width a, with the potential at the bottom con-
strained to be φ(x, 0) = V , and zero potential
φ= V
on the sides, with φ vanishing as y → ∞. We
wish to calculate the potential inside the well.
y=0
x=0 x=a
Laplace’s equation becomes
∂ 2 φ ∂ 2φ
+ =0
∂x2 ∂y 2
subject to the boundary conditions

φ(0, y) = φ(a, y) = 0
φ(x, 0) = V
φ(x, y) → 0 as y → ∞

As before, we look for separable solutions φ(x, y) = X(x)Y (y), yielding

1 d2 X 1 d2 Y
+ = 0,
X dx2 Y dy 2
so that each of the above terms must separately be constant.
Since X(0) = X(a) = 0, the solution for X must be oscillatory,

X ′′ + α2 X = 0
Chapter 2 31

giving X(x) = sin αx. The boundary condition at x = a then yields

nπ
Xn (x) = sin αn x; where αn = a ,n = 1, 2, . . ..

The corresponding function Yn (y) must satisfy

Yn′′ − αn2 Yn = 0

with exponential solutions Yn (y) = exp ±αn y. The boundary condition φ → 0 as

y → ∞ requires that we take the exponentially falling solution, and thus

Yn (y) = e−αn y .

Thus the factorisable solutions are of the form

φn (x, y) = e−αn y sin αn x

so that the general solution is

nπ
Ane−αn y sin αn x;
X
φ(x, y) = αn = .
n a
We determine the coefficients An by imposing the boundary condition at y = 0:
X
V = An sin αn x,
n

and using the orthogonality of the sin functions, we obtain

Z a nπx X Z a nπx n′ πx
V sin = An dx sin sin
0 a n 0 a a
a
= An .
2
The integral is straightforward:
2V Z a nπx
An = dx sin
a 0 " a #
2V a nπx a
= − cos
a nπ a 0
2V
= [1 − (−1)n],
nπ
Chapter 2 32

and thus 

 4V /nπ n odd
An = 
 0 n even
with

4V X 1 −nπy/a nπx
φ(x, y) = e sin .
π n odd n a

For y/a ≪ 1, we can expand this as a series, and we converge to an accurate

solution within a few terms - remember that exponential! But in this case, we
can actually sum the series. We begin by recalling that

eix = cos x + i sin x.

yielding
nπx
sin = ℑeinπx/a .
a
Thus we may write the general solution as
4V X 1 −nπy/a inπx/a
φ(x, y) = e ℑe
π n odd n
4V X 1 (nπi/a)(x+iy)
= ℑe .
π n odd n
We now introduce the variable

Z = e(iπ/a)(x+iy) ,

so that the solution becomes

4V X 1
φ= ℑZ n .
π n odd n
To sum this series, we recall that
Z2 Z3
ln(1 − Z) = −Z − − + ...,
2 3
Z2 Z3
ln(1 + Z) = Z − + + ...,
2 3
Chapter 2 33

and thus
X 1 n 1
Z = − {ln(1 − Z) − ln(1 + Z)}
n odd n 2
1 1+Z
= ln .
2 1−Z
Hence we may write the general solution as
2V 1+Z
φ(x, y) = ℑ ln .
π 1−Z
We will conclude by writing this solution explicitly in terms of x and y. We begin
by noting that Z = |Z| exp iθ where θ is the phase of Z, i.e. tan θ = ℑZ/ℜZ.
Thus
ln Z = ln |Z| + iθ =⇒ ℑ ln Z = θ.
Now,
1+Z (1 + Z)(1 − Z ∗) 1 − |Z|2 + 2iℑZ
= =
1−Z |1 − Z|2 |1 − Z|2
and thus  
1+Z 2ℑZ 
ℑ ln = tan−1  .
1−Z 1 − |Z|2
But we have
πx
ℑZ = e−πy/a sin
a
|Z|2 = e−2πy/a

and thus
−πy/a
sin πx
 
2V −1  2e a 
φ(x, y) = tan −2πy/a
π 1−e
which, after some simplification, becomes

 
2V sin πx/a 
φ(x, y) = tan−1  .
π sinh πy/a
Chapter 2 34

In practice, such problems can be done in a much simpler way, by observing that
the real and imaginary components, u and v respectively, of an analytic complex
function f (z = x + iy) satisfy the two-dimensional Laplace’s equation
∂ 2u ∂ 2u ∂ 2v ∂ 2v
+ = 0; + = 0.
∂x2 ∂y 2 ∂x2 ∂y 2
This is a direct consequence of the Cauchy-Riemann eqnations.

2.6.2 Field and Charge Distribution in Two-dimensional Corners

Consider two conducting planes
meeting at an angle β, with po-
tential V on the planes. The most e
θ
e
ρ
appropriate coordinate system for V P
the problem is that of cylindrical
ρ
polars (s, θ, z), with the z axis along
the line of intersection of the planes. θ
Note that if we consider the problem
sufficiently close to the intersection, β
V
the shape of the surface at larger
distances will be unimportant.
Then Laplace’s equation assumes the form
1∂ ∂φ 1 ∂ 2φ
!
2
∇ φ(s, θ) = s + 2 2
s ∂s ∂s s ∂θ
where we have suppressed the z variable. As before we look for factorizing solu-
tions of the form
φ(s, θ) = R(s)T (θ).
Then we have
s ∂ ∂R 1 ∂ 2T
!
s + = 0.
R ∂s ∂s T ∂θ2
Chapter 2 35

Each term depends on a different variable, and this must hold for all s and z.
Thus each term is separably constant. For the function T (θ), let us take
1 ∂ 2T
2
= −ν 2.
T ∂θ
Since T must attain the same value at θ = 0 and θ = β, the solution must be
oscillatory rather than exponential, and hence ν 2 must be positive. Thus the
solution is 

 Aν cos νθ + Bν sin νθ; ν 6= 0
Tν (θ) = 
 A0 + B0 θ; ν=0
For the radial function, we have
∂ ∂R
!
s s − ν 2R = 0.
∂s ∂s
For ν 6= 0, let us take as trial solution R ∼ sα ,

(α2 − ν 2)sα = 0,

yielding α = ±ν. We need to consider the case ν = 0 separately. Here we have

∂ ∂R
!
s =0
∂s ∂s
with solution
R0 (s) = a0 + b0 ln s.
Thus the general form of Rν is


 aν sν + bν s−ν ; ν > 0
Rν (s) = 
 a0 + b0 ln s; ν=0

so the general solution for the potential has the form

(aν sν +bν s−ν )(Aν cos νθ +Bν sin νθ) (2.24)

X
φ(s, θ) = (a0 +b0 ln s)(A0 +B0θ)+
ν>0
Chapter 2 36

The solution must be valid as s → 0 (note that we are not interested in the
solution for s large), and therefore the terms proportional to ln s and s−ν cannot
contribute. Thus our solution is of the form


 A0 + B0 θ; ν=0
φ(s, θ) = 
 sν (Aν cos νθ + Bν sin νθ); ν > 0
We will now use the boundary conditions on the planes to further constrain the
solution. At θ = 0, β, we have φ = V , independent of s, and therefore we have

Aν = 0
nπ
sin νβ = 0; ν= , n = 1, 2, . . . ,
β
yielding
nπθ
Bn snπ/β sin
X
φ(s, θ) = A0 + B0θ + .
n β
Finally, we impose that the potential be V on the two planes

θ = 0, φ = V =⇒ A0 = V
θ = β, φ = V =⇒ B0 = 0,

and thus our final result is

nπθ
Bn snπ/β sin
X
φ(s, θ) = V + .
n β

As we get closer into the corner, the first term will dominate,
πθ
φ(s, θ) ∼ V + B1 sπ/β sin .
β
Taking the gradient, we obtain
πB1 π/β−1 πθ πB1 π/β−1 πθ
E = −∇φ = − s sin es − s cos eθ
β β β β
and the induced surface charge density is
πB1 ǫ0 π/β−1
σ = ǫ0 [E · n] = − s
β
Chapter 2 37

1. For β < π, we have that E and σ vanish as s → 0.

2. For β > π, E and σ become singular as s → 0.

Thus we see behaviour familiar from our knowledge of “action at points” - the
fields and surface charge densities become singular near sharp edges.
Chapter 3

Boundary-value Problems in Curvilinear

Coordinates

In the previous chapter, we saw how we could look for factorizable solutions to
Laplace’s Equation in Cartesian coordinates, and then construct the solution for
more general boundary values using the completeness property of the such fac-
torisable solutions. In this chapter we will employ analogous methods in spherical
polar and cylindrical coordinate systems. In practice, the coordinate system that
is appropriate depends on the symmetry or geometry of the problem.

3.1 Laplace’s Equation in Spherical Polar Coordinates

We will denote our coordinates by (r, θ, ϕ), in terms of which Laplace’s equation
assumes the form
1 ∂ 2 ∂φ 1 ∂ ∂φ 1 ∂ 2φ
! !
2
∇ φ(s, θ, ϕ) = 2 r + 2 sin θ + 2 2 .
r ∂r ∂r r sin θ ∂θ ∂θ r sin θ ∂ϕ2
We will now seek factorisable solutions of the form
U (r)
φ(r, θ, ϕ) = P (θ)Q(ϕ),
r

1
Chapter 2 2

where the factor of 1/r is conventional. Substituting this into Laplace’s equation,
we have
  
1 d 1 1 dU (r) 
P (θ)Q(ϕ) 2 r2 − 2 U (r) +
r dr r r dr
d2Q(ϕ)
 
U (r)Q(ϕ) 1 d  dP (θ)  U (r)P (θ) 1
+ sin θ + = 0,
r r2 sin θ dθ dθ r r2 sin2 θ dϕ2
yielding
P Q d2 U UQ d dP U P d2 Q
!
+ 3 sin θ + 3 2 = 0,
r dr2 r sin θ dθ dθ r sin θ dϕ2
which we may write as
1 d2 Q 2
 !
2 2 1 d U 1 1 d dP
+ r sin θ + sin θ  = 0. (3.1)
Q dϕ2 U dr2 r2 sin θ P dθ dθ
The first term is a function of ϕ alone, and the remaining term is a function of
(r, θ) alone. Thus they must be separately constant, and we may write
1 d2 Q
2
= −m2 , (3.2)
Q dϕ
where m is a constant. Eqn. 3.2 has solution

Q = e±imϕ .

We now observe that the solution must be periodic, with period 2π, in the az-
imuthal variable ϕ. Thus m must be an integer, and, of course, real. Thus we
may write eqn. 3.1 as
r 2 d2 U 1 1 d dP m2
!
+ sin θ − = 0.
U dr2 sin θ P dθ dθ sin2 θ
We now observe that the first term is purely a function of r, whilst the remaining
terms are purely a function of θ. Thus we may write
r 2 d2 U
= l(l + 1)
U dr2
Chapter 2 3

where l is a constant - we will see the reason for expressing the constant in this
way later. To solve this equation, we will take a trial solution

U (r) = rα ,

yielding
α(α − 1) = l(l + 1)
with solutions α = l + 1, −l. Thus we have

U (r) = Arl+1 + Br−l .

The equation for the polar coordinate θ now assumes the form
m2 
 
1 d dP
!
sin θ + l(l + 1) −
 P = 0.
sin θ dθ dθ sin2 θ
It is convenient to introduce the variable x = cos θ, with −1 ≤ x ≤ 1 and
d d
= − sin θ .
dθ dx
After a little algebra, we have

m2 
 
d 2 dP
" #
(1 − x ) + l(l + 1) −
 P =0
dx dx 1 − x2

This is the Generalised Legendre Equation, and is, once again, an equation
of Sturm-Liouville type, with p(x) = 1 − x2, q(x) = −m2 /(1 − x2), λ = l(l + 1),
and r(x) = 1.
We will now seek solutions of this equation, first for the case m = 0, where the
equation is known as the Ordinary Legendre Equation

d dP
" #
(1 − x2 ) + l(l + 1)P = 0.
dx dx
Chapter 2 4

We begin by noting that the solutions must be both continuous and single-
valued in the region −1 ≤ x ≤ 1, corresponding to 0 ≤ θ ≤ π. We will obtain
the solutions through series substitution, i.e. by trying a solution of the form
∞
cn xγ+n,
X
P =
n=0

from which
dP ∞
cn (γ + n)xγ+n−1
X
=
dx n=0
dP ∞ ∞
(1 − x2) γ+n−1
cn (γ + n)xγ+n+1,
X X
= cn (γ + n)x −
dx n=0 n=0
d ∞ ∞
2 dP
" #
γ+n−2
cn (γ + n)(γ + n + 1)xγ+n.
X X
(1 − x ) = cn (γ + n)(γ + n − 1)x −
dx dx n=0 n=0

Thus Legendre’s equation becomes

∞ ∞
γ+n−2
cn [l(l + 1) − (γ + n)(γ + n + 1)] xγ+n = 0.
X X
cn (γ + n)(γ + n − 1)x +
n=0 n=0

As this equation must be valid ∀x ∈ [−1, 1], we can equate the coefficients of the
powers of x to zero. The leading power of x is xγ−2, and we use this equation, the
indicial equation, to determine γ. Thus

• xγ−2:
c0 γ(γ − 1) = 0 =⇒ γ = 0 or γ = 1

• xγ−1:


 c1 undetermined : γ = 0
c1 (γ + 1)(γ + 1 − 1) = 0 =⇒ 
 c1 = 0 :γ=1

• xγ+n, n ≥ 0:
(γ + n)(γ + n + 1) − l(l + 1)
cn+2 = cn .
(γ + n + 2)(γ + n + 1)
Chapter 2 5

Note that the recursion relation relates only even (odd) polynomials for γ = 0
(γ = 1).
We have already noted that the solution must be valid for x ∈ [−1, 1], and in
particular at the end points x = ±1. Thus the series must be finite at x = ±1.
To explore the convergence properties, we note that
cn+2/cn −→ 1 as n −→ ∞,
and thus the series resembles a geometrical expansion x2n. This diverges at
P

x = ±1 unless the series terminates, i.e. unless cn = 0 for some n. Thus our
requirement for convergence is
(γ + n)(γ + n − 1) − l(l + 1) = 0 for some n.
• γ = 0:
n(n + 1) = l(l + 1) =⇒ n = l

• γ = 1:
(n + 1)(n + 2) = l(l + 1) =⇒ n = l − 1.

Note that in both cases the highest power of x is xl ; the two cases are the same.
We call the corresponding solutions Pl (x) the Legendre Polynomials, and con-
ventially we take Pl (1) = 1. The first few are
P0 (x) = 1
P1 (x) = x
1
P2 (x) = (3x2 − 1)
2
1
P3 (x) = (5x3 − 3x).
2

3.1.1 Rodriques’ Formula and Generating Function

We can write the Legendre polynomials in a more memorable form through Ro-
drigues’ Formula:
1 dl 2
Pl (x) = l l
(x − 1)l .
2 l! dx
Chapter 2 6

An equally useful means of determining the Legendre polynomials is through the

generating function
∞
2 −1/2
Pl (x)tl , |t| < 1.
X
g(t, x) ≡ (1 − 2xt + t ) = (3.3)
l=0

3.1.2 Orthogonality and Normalisation of Legendre Polynomials

Applying our theorem concerning the orthogonality of the solutions of the Sturm-
Liouville equation yields
Z 1 Z 1
′ ′
[l(l − 1) − l (l + 1)] dx Pl (x)Pl′ (x) = 0 =⇒ dx Pl (x)Pl′ (x) = 0, l 6= l′ ,
−1 −1
i.e. the Legendre polynomials are orthogonal. N.B. it is easy to check that our
solutions satisfy the required boundary conditions.
To determine their normalisation, we can use either Rodrigues’ formula, or the
generating function; we use the latter. From eqn. 3.3, we have
Z 1
2
Z 1 1
dx g(t, x) = dx
−1 −1 1 − 2xt + t2
)1
1
(
= − ln(1 − 2xt + t2)
2t −1
2
1 (1 − t)
= − ln
2t (1 + t)2
X∞ 2t2l
= ,
l=0 2l + 1
where we have used the series expansion of ln(1 + t). However, we also have
Z 1 ∞ Z ′
2
dx Pl (x)Pl′ (x)tl+l
X
dx g(t, x) =
−1
l,l′ =0
∞
2l 1
Z
dx Pl (x)2.
X
= t
l=0 −1

Equating the coefficients in these two expansions yields

Z 1 2
dx Pl (x)Pl (x) =
−1 2l + 1
Chapter 2 7

3.1.3 Recurrence Relations

Rodrigues’ formula provides a means to obtain various recurrence relations

between the Legendre Polynomials, for example:

(l + 1)Pl+1(x) − (2l + 1)xPl (x) + lPl−1(x) = 0

d dPl (x)
Pl+1(x) − x − (l + 1)Pl (x) = 0
dx dx
dPl (x)
(x2 − 1) − lxPl (x) + lPl−1(x) = 0
dx
d dPl−1(x)
Pl+1(x) − − (2l + 1)Pl (x) = 0.
dx dx
Such recurrence relations allow us to evaluate many of the integrals we will en-
counter in the problems.

3.1.4 Completeness

Since the Legendre Polynomials form a complete set, we may write any function
f (x), x ∈ [−1, 1] as
∞
X
f (x) = Al Pl (x).
l=0
We obtain the coefficients using the orthogonality relations
Z ∞
X Z 1
dx f (x)Pl(x) = Al dx Pl (x)Pl′ (x)
−1
l′ =0
2
= Al
2l + 1
whence
2l + 1 Z 1
Al = dx f (x)Pl (x)
2 −1

Example

Consider the step-function f (x) defined by



 1 0<x≤1
f (x) = 
 −1 −1 ≤ x < 0
Chapter 2 8

Then we have that

2l + 1 Z 1
Al = dxf (x)Pl(x)
2 (−1
2l + 1 Z 1 Z 0 )
= dx Pl (x) − dx Pl (x)
2 0 −1
2l + 1 Z 1
= dx {Pl (x) − Pl (−x)}.
2 0

Thus we see that Al is non-zero only for l odd:


R1

 (2l + 1) 0 dx Pl (x) : l odd
Al = 
 0 : l even
Now by the last of our recurrence relations
d d
Z 1 ( )
Al = dx Pl+1(x) − Pl−1(x)
0 dx dx
= Pl+1(1) − Pl+1(0) − Pl−1(1) + Pl−1(0)
= Pl−1(0) − Pl+1(0)

where we have used the normalisation condition Pl (1) = 1. But we have (from
Rodrigues’s formula, with a little work)
l/2

 (−1)l/2 (l−1)!!

: l even
2 (l/2)!
Pl (0) =
 0

: l odd
where (l − 1)!! = (l − 1)(l − 3) . . . 3.1. Thus
(−1)(l+1)/2l!! (−1)(l−1)/2(l − 2)!!
Al = − +
2(l+1)/2((l + 1)/2)! 2(l−1)/2((l − 1)/2)!
(−1)(l−1)/2(l − 2)!! l
( )
= (l−1)/2 1+
2 ((l − 1)/2)! l+1
Thus   (l−1)/2
(l−2)!!(2l+1)
 −1 : l odd


2 2( l+1
2 )!
Al =
0 : l even




and we have
3 7 11
f (x) = P1 (x) − P3 (x) + P5 (x) . . .
2 8 16
Chapter 2 9

3.2 Boundary-Value Problems with Azimuthal Symmetry

We may now write our general solution for the boundary-value problem in spher-
ical coordinates with azimuthal symmetry, i.e. no ϕ dependence, as

∞ 
Al rl + Bl r−l−1 Pl (cos θ),
X 
φ(r, θ) =
l=0

where the coefficients Al and Bl are determined from the boundary conditions.

Example:

Consider the case of a sphere, of radius a, with no charge inside but potential
V (θ) specified on the surface.
Since there are no charges inside the sphere, the potential φ inside must be regular
everywhere. Thus Bl = 0 ∀l, and we may write the solution as
∞
Al rl Pl (cos θ).
X
φ(r, θ) =
l=0

Imposing the boundary conditions at r = a yields

∞
Al al Pl (cos θ),
X
V (θ) =
l=0

so that, using the normalisation condition on the Legendre polynomials, we have

2l + 1 Z π
Al = dθ sin θ V (θ)Pl (cos θ).
2al 0
Suppose now that we require the solution outside the sphere. Then the solution
must be finite as r → ∞, and thus
∞
Bl r−l−1Pl (cos θ)
X
φ(r, θ) =
l=0

with
∞
Bl a−l−1Pl (cos θ),
X
V (θ) =
l=0
Chapter 2 10

so that
2l + 1 l+1 Z π
Bl = a dθ sin θ V (θ)Pl (cos θ).
2 0

Let us now go back to the problem in Section 2.4.2:



 V : 0 ≤ θ ≤ π/2
V (θ) = 
 −V : π/2 ≤ θ ≤ π

Then we have
2l + 1 l+1
(Z )
π/2 Z π
Bl = a V Pl (cos θ) sin θdθ − Pl (cos θ) sin θdθ
2 0 −π/2
2l + 1 l+1
(Z )
1 Z 0
= a V dxPl (x) − dxPl (x)
2 0 −1
2l + 1 l+1
(Z )
1
= a V dxf (x)Pl(x)
2 −1

where 

 1 0<x≤1
f (x) = 
 −1 −1 ≤ x < 0
This is just the expression we evaluated in Section 3.1.4, and thus we have:

l−1 (l−2)!!(2l+1)

 V al+1(− 12 ) 2
2( l+1
l odd
2 )!

Bl = 

 0 l even

so that
 3 a2 7 a4 11 a6
 

φ(r, θ) = V  2 P1 (cos θ) − 4 P3 (cos θ) + P5 (cos θ) + . . . . (3.4)
2r 8r 16 r6 

Recall that in Section 2.4.2 we obtained

3V a2  7a2 5 a4 
 
3
!
3
φ(r, θ, ϕ) = cos θ − cos θ − cos θ + O( 4 )
r2  12r2 2 2 r
 3 a2 7 a4
 

= V  2 P1 (cos θ) − 4 P3 (cos θ) + . . . ,
2r 8r
which is precisely the first two terms in the expansion of eqn. 3.4.
Chapter 2 11

The crucial observation in such problems is that the series expansion

Al rl + Bl r−l−1 Pl (cos θ)
X 
φ(r, θ) = (3.5)
l

is unique. Thus it is possible to determine the coefficients Al and Bl from a

knowledge of the solution in some limited domain. As an illustration, we recall
that we obtained a closed solution to the above problem above the North pole,
i.e. at θ = 0:
r2 − a2 
 

φ(z = r, θ = 0) = V 1 − √ 2 .
r r + a2 
We can use the binomial expansion to express this as a series in a/r:
 a2 Γ( 21 )

∞ !2j 
2 2 X a 
φ(z = r, θ = 0) = V  2 − (1 − a /r ) 1 .
r j=1 Γ(j + 1)Γ( 2 − j) r


If we use the property

π
Γ(z)Γ(1 − z) =
sin πz
√
and note that Γ(1/2) = π, we obtain, after a little manipulation (exercise),
∞ 1 1 !2j
V X j−1 (2j − 2 )Γ(j − 2 ) a
φ(r, θ = 0) = √ (−1) .
π j=1 j! r
We now compare this series solution with eqn. 3.5, evaluated at θ = 0, and observe
that only terms with l = 2j − 1 enter, and that
1 1
V j−1 (2j − 2 )Γ(j − 2 ) 2j
B2j−1 = √ (−1) a .
π j!
Let us try the first couple of term
V 0 (3/2)Γ(1/2) 2
j = 1 : B1 = √
π (−1) 1! a = 3V a2 /2
V 1 (5/2)Γ(3/2) 4
j = 2 : B3 = √
π (−1) 2! a = − 87 V a4 ,

and once again we reproduce the expression eqn. 3.4.

Chapter 2 12

3.2.1 Expansion of |x−1x′ |

Let us conclude this section by looking at the expansion of this critical quantity
that occurs in the construction of the Green’s function. We begin by observing
that the result can depend only on r, r′ and γ, the angle between x and x′ . We
may thus simplify the problem by choosing the azimuthal direction (z axis) along
the x′ axis. The problem then displays manifest azimuthal symmetry, and we may
write
1 ∞ 
′ l ′ −l−1
X 
= A l (r )r + B l (r )r Pl (cos γ) (3.6)
|x − x′ | l=0
We now consider the case where x lies parallel to x′ , when cos γ = 1. Then the
l.h.s. of eqn. 3.6 becomes
1 1
= .
|x − x′ | |r − r′ |
There are two cases:
l
r′  ∞ r ′l

1 1 1X ∞
′ X
r>r : = =  =
|r − r′ | r − r′ r l=0 r l=0 r
l+1

1 1 1 X∞ r l X
! ∞ rl
′
r<r : = = ′ =
|r − r′ | r′ − r r l=0 r′ l=0 r
′l+1

Let us introduce r> = max(r, r′) and r< = min(r, r′). Then we may write
1 ∞ rl
X <
′
= l+1
|r − r | l=0 r>
and, comparing with eqn. 3.6, we have

1 ∞ rl
X <
′
= l+1 Pl (cos γ)
|x − x | l=0 r>
Chapter 2 13

3.3 Solution of the Generalised Legendre Equation

Let us now consider the case where we no longer assume azimuthal symmetry.
Then we are concerned with solutions of the Generalised Legendre Equation,
m2 
   
d  2 dP (x) 
(1 − x ) + l(l + 1) −
 P (x) = 0. (3.7)
dx dx 1 − x2
We can obtain a series solution in an analogous way to that of the ordinary
Legendre equation. For solutions to be finite at x = ±1, corresponding to θ = 0, π,
we require that l must be a positive integer or zero, and that m takes the values

m = −l, −l + 1, . . . , l − 1, l.

Recall that we already know that m must be an integer by the requirement that
the azimuthal function Q(ϕ) be single-valued.
For the case where m is positive, we can write the solutions Plm (x) as
dm
Plm (x) m
= (−1) (1 − x ) 2 m/2
Pl (x)
dxm
or for both positive and negative m by adopting Rodrigues’ formula:
(−1)m 2 m/2 d
l+m
Plm (x) = l
(1 − x ) l+m
(x2 − 1)l .
2 l! dx
Note that eqn. 3.7 depends only on m2 . Thus we have that Pl−m (x) must be
proportional to Plm (x), and in fact
(l − m)! m
Pl−m (x) = (−1)m P (x).
(l + m)! l
Eqn. 3.7 is an equation of Sturm-Liouville class, with eigenvalues l(l + 1). We can
apply the orthogonality theorem at fixed m, and we have
Z 1 2 (l + m)!
dxPlm m
′ (x)Pl (x) = δll′ .
−1 2l + 1 (l − m)!
Chapter 2 14

3.4 Spherical Harmonics

We began by looking at separable solutions in spherical polar coordinates, and

writing
1
φ(r, θ, ϕ) = U (r)P (θ)Q(ϕ).
r
It is convenient to combine the angular functions into solutions on the unit sphere:
v
u (l − m)!(2l + 1) m
u
Ylm(θ, ϕ) = Pl (cos θ)eimϕ. (3.8)
u
t
4π(l + m)!
The spherical harmonics (3.8) satisfy the equation

−∇2Ylm (θ, ϕ) = l(l + 1)r2Ylm(θ, ϕ) (3.9)

or, in explicit form

1 ∂ ∂ 1 ∂2
" ! #
− sin θ − 2 2
Ylm(θ, ϕ) = l(l + 1)r2Ylm (θ, ϕ) (3.10)
sin θ ∂θ ∂θ sin θ ∂ϕ
(A person familiar with quantum mechanics may recognize the expression in
square brackets in l.h.s. of this eqn as a square of operator of anglular momentum
L2).
Using our relation between Pl−m (cos θ) and Plm (cos θ) we have

Yl,−m(θ, ϕ) = (−1)mYlm(θ, ϕ)∗

and the normalisation condition is

Z 2π Z π
∗
dϕ dθ sin θYlm (θ, ϕ)Yl′m′ (θ, ϕ) = δll′ δmm′ ,
0 0

i.e. Z
dΩ Ylm(θ, ϕ)Yl∗′m′ (θ, ϕ) = δmm′ δll′ .
For the case m = 0, the solution clearly reduces to the Legendre polynomial, up
to some normalisation:
v
u 2l +1
u
Yl0(θ, ϕ) = t
Pl (cos θ)
4π
Chapter 2 15

3.4.1 Completeness

Any arbitrary function g(θ, ϕ) defined on 0 ≤ θ ≤ π, 0 ≤ ϕ ≤ 2π may be expressed

in terms of Ylm :
∞ X
X l
g(θ, ϕ) = Alm Ylm(θ, ϕ)
l=0 m=−l
where
Z π Z 2π
∗
Alm = dθ sin θ g(θ, ϕ)Ylm (θ, ϕ)
Z0 0
∗
= dΩ Ylm(θ, ϕ)g(θ, ϕ)

3.4.2 General Solution

We can now write the general solution of the Laplace boundary value problem as
∞ X
l h
Alm rl + Blm r−l−1 Ylm (θ, ϕ)
X i
φ(r, θ, ϕ) =
l= m=−l

3.4.3 Addition Theorem for Spherical Harmonics

Consider two vectors x, x′ , with coordinates (r, θ, ϕ) and (r′, θ′, ϕ′) respectively.
Let γ be the angle between x and x′ , so that
x· x′
cos γ = ′
= cos θ cos θ′ + sin θ sin θ′ cos(ϕ − ϕ′).
|x||x |
Then we have

4π X l
∗
Pl (cos γ) = Ylm (θ′, ϕ′ )Ylm(θ, ϕ)
2l + 1 m=−l

This is proved in Jackson, but is more easily proved using group theory. Note
that we can rewrite this in the form
l (l − m)! m
Pl (cos γ) = Pl (cos θ)Pl (cos θ′ ) + 2 Pl (cos θ)Plm (cos θ′ ) cos m(ϕ − ϕ′ )
X

m=1 (l + m)!
Chapter 2 16

Example

An important application is to the expansion of |x−1 x| , discussed in section 3.2.1:

1 ∞ rl
X <
′
= l+1 Pl (cos γ).
|x − x | l=0 r>
Using the addition theorem, we can rewrite this as
∞ X l l
1 X 1 r< ∗ ′ ′
′
= 4π l+1 Ylm (θ , ϕ )Ylm (θ, ϕ).
|x − x | l=0 m=−l 2l + 1 r>

Superficially, this looks like a much more complicated expression, since we have
introduced an additional sum over m. But it is now a sum over terms that factorise
into a function of (θ, ϕ) and a function of (θ′ , ϕ′), and thus much more useful.

3.5 Laplace’s Equation in Cylindrical Polar Coordinates

We will denote the coordinates by (s, ϕ, z)

z

θ z

y
ρ

x
Chapter 2 17

In terms of these coordinates, Laplace’s equation assumes the form

1∂ ∂φ 1 ∂ 2φ ∂ 2φ
!
2
∇ φ(s, ϕ, z) = s + 2 2 + 2 = 0.
s ∂s ∂s s ∂ϕ ∂z
As before, we look for separable solutions of the form

φ(s, ϕ, z) = R(s)T (ϕ)Z(z),

so that Laplace’s equation becomes

1d dR 1 d2 T d2 Z
!
TZ s + RZ 2 2 + RT 2 = 0,
s ds ds s dϕ dz
which we may rewrite as
1 d dR 1 d2 T 1 d2 Z
!
s + 2 + = 0.
Rs ds ds s T dϕ1 Z dz 2
The third term is a function of z alone, whilst the others are a function of s and
ϕ alone. Thus we may write
1 d2 Z
2
= k2
Z dz
where k is a (not necessarily real) constant, with solution

Z(z) = e±kz .

Thus we may now rewrite Laplace’s equation as

s d dR 1 d2 T
!
s + 2
+ k 2 s2 = 0,
R ds ds T dϕ
and so for the angular term we have
1 d2 T
2
= −ν 2
T dϕ
with solution
T (ϕ) = e±iνϕ .
For the solution to be single valued at ϕ = 0 and 2π, ν must be an integer.
Chapter 2 18

Finally, the radial equation is

s d dR
!
s − ν 2 + k 2s2 = 0.
R ds ds
We can eliminate the constant k by the substitution x = ks, yielding
x d dR
!
x − ν 2 + x2 = 0
R dx dx
which we write as

d2 R 1 dR  ν2 
 

+ + 1− 2 R =0
dx2 x dx x

This is the Bessel Equation.

As in the case of the Legendre equation, we find a solution by series substitution
∞
cn xs+n :
X
R(x) = c0 6= 0 (3.11)
n=0
Aside: why do we have to introduce the power xs , rather than just looking for a
solution in terms of a Taylor expansion about x = 0? The reason is that there is
a regular singular point at x = 0, i.e. the coefficents of R and its derivatives in the
Bessel equation vanish at x = 0, and therefore the solution can have a singularity
there. In the case of the Legendre equation, there are regular singular points at
x = ±1.
From eqn. 3.11, we have
dR ∞
cn (γ + n)xγ+n−1
X
=
dx n=0
2 ∞
dR
cn (γ + n)(γ + n − 1)xγ+n−2,
X
2
=
dx n=0
and substituting into the Bessel equation we have
∞ ∞ ∞ ∞
γ+n−2 γ+n−2 γ+n 2 X
cn xγ+n−2 = 0.
X X X
cn (γ+n)(γ+n−1)x + cn (γ+n)x + cn x −ν
n=0 n=0 n=0 n=0

The lowest power of x is xγ−2, and equating the coefficients of this to zero gives
the indicial equation which determines γ.
Chapter 2 19

• xγ−2 :
c0 γ(γ − 1) + c0 γ + ν 2c0 = 0 ⇒ γ = ±ν, since c0 6= 0.

• xγ−1 :

0 = c1 (γ + 1)γ + c1 (γ + 1) − ν 2c1
= c1 (γ 2 + 2γ + 1 − ν 2)
= c1 (2γ + 1) since γ 2 = ν 2
⇒ c1 = 0 since ν is an integer.

• xn+γ , n ≥ 0 :

cn+2 [(γ + n + 2)(γ + n + 1) + (γ + n + 2) − ν 2] + cn = 0

⇒ cn+2[(γ + n + 2)2 − ν 2] = −cn
1
⇒ cn+2 = − cn
(n + 2)(n + 2 + 2γ)
where in the last line we have used γ 2 = ν 2 .

As in the case of Legendre’s equation, the recurrence relation relates either odd or
even values of n. However, we have seen that c1 = 0. Thus cn = 0 for all odd n.
Therefore, let us make the substitution n = 2j, and write the recurrence relation
as
1
c2j+2 = − c2j , j = 0, 1, 2, . . .
4(j + 1)(j + 1 + γ)
c2j+1 = 0.

We can now iterate this recursion relation to obtain

!2j
j 1 Γ(γ + 1)
c2j = (−1) c0 .
2 Γ(j + 1)Γ(γ + j + 1)
Conventionally, we choose
1
c0 = ,
2γ Γ(γ + 1)
Chapter 2 20

so that the solutions may be written

x νX∞ (−1)j x 2j
! !
Jν (x) =
2 j=0 Γ(j + 1)Γ(ν + j + 1) 2
x −ν X∞ (−1)j x 2j
! !
J−ν (x) = .
2 j=0 Γ(j + 1)Γ(j − ν + 1) 2
These are the Bessel Functions of the first kind of order ±ν. Some obser-
vations:
• The series converge for all finite x

• If ν is not an integer, the solutions are linearly independent.

• If ν is an integer, they are linearly dependent, and in particular

J−m(x) = (−1)mJm (x).

Proof: This is a consequence of the properties of the gamma function Γ(z),

which has singularities for z = 0 and for z a negative integer - recall the
earlier relation
π
Γ(z)Γ(1 − z) = .
sin πz
We have
x −m X ∞ (−1)j x 2j
! !
J−m(x) = .
2 j=0 Γ(j + 1)Γ(j − m + 1) 2
Now Γ(j − m + 1) −→ ∞ as argument approaches 0 or a negative integer.
Thus only those terms in the sum for which j − m + 1 ≥ 1 contribute, and
we can write
(−1)j
!−m ∞ !2j
x X x
J−m (x) =
2 j=m Γ(j + 1)Γ(j − m + 1) 2
x −m x 2m X ∞ (−1)l+m x 2l
! ! !
=
2 2 l=0 Γ(l + 1)Γ(l + m + 1) 2
(−1)l
!m
x ∞ x 2l
!
mX
= (−1)
2 l=0 Γ(l + 1)Γ(l + m + 1) 2
m
= (−1) Jm (x)
Chapter 2 21

Because of the linear dependence of J−m (x) on Jm (x), we introduce a second,

linearly independent function
Jν (x) cos νπ − J−ν (x)
Nν (x) =
sin νπ
known as the Neumann Function or the Bessel Function of the second
kind. Conventionally, we choose as our linearly independent functions Jν (x) and
Nν (x) even if ν is not an integer.

Bessel Function of the Third Kind

These are just another pair of linearly independent solutions of the Bessel equa-
tion:

Hν(1) (x) = Jν (x) + iNν (x)

Hν(2) (x) = Jν (x) − iNν (x)

These are also known as Hankel Functions. Their utility is that they have a
more straightforward integral representation than Jν (x) and Nν (x).

3.5.1 Recursion Relations

The sets of solutions of the Bessel equation are collectively known as cylinder
functions, and satisfy recursion relations in the same manner as the Legendre
polynomials, e.g.
2ν
Ων−1(x) + Ων+1(x) = Ων (x)
x
dΩν (x)
Ων−1(x) − Ων+1(x) = 2
dx

3.5.2 Limiting Behaviour of Solutions

In the limit x ≪ 1, we have

!ν
1 x
Jν (x) →
Γ(ν + 1) 2
Chapter 2 22

 2 ln x +
h   i

π γE + . . . ν=0
Nν (x) → 2 ν
 − Γ(ν) 2

π x
ν 6= 0
where ν is real and non-negative, and γE = 0.5772 . . . is the Euler-Mascheroni
constant. Note that, when constructing solutions of the boundary-value problem,
only Jν (x) is regular as x → 0.
In the limit x ≫ 1, ν, we have
v
2 νπ π
u !
u
Jν (x) → t
cos x − −
πx 2 4
v
u 2 νπ π
u !
Nν (x) → t
sin x − − . (3.12)
πx 2 4
The transition between these limiting forms occurs at x ∼ ν.

3.5.3 Roots of the Bessel functions

From the limiting forms 3.12, we see that each Bessel function has an infinite
number of roots, which we denote xνn , n = 1, 2, 3, . . . where
Jν (xνn) = 0, for x = 1, 2, 3, . . ..
In particular, we have
ν = 0 : x0n = 2.405, 5.520, 8.654, . . .
ν = 1 : x1n = 3.832, 7.016, 10.173, . . .
ν = 2 : x2n = 5.136, 8.417, 11.620, . . .

3.5.4 Ortogonality of the Bessel Functions

The roots of the Bessel function Jν (x) are crucial when we consider its orthogonal-
ity properties, which take a rather unexpected form. We introduce the functions
√
sJν (xνns/a), n = 1, 2, 3, . . .
and will now show that, for fixed ν ≥ 0, these functions, identified by n, form an
orthogonal set on 0 ≤ s ≤ a.
Chapter 2 23

Proof

Substitute into the Bessel equation:

x2νn ν 2 
 
1d d
" #
s Jν (xνn s/a) +  − 2 Jν (xνns/a) = 0,
s ds ds a2 s
where we have made the change of variable x → xνn s/a. We now rewrite this as
d dJν ν2 x2
" #
s − Jν = − νn sJν .
ds ds s a2
This is the Sturm-Liouville equation, with

p(x) = s,
q(x) = −ν 2/s,
r(x) = s,
λ = x2νn /a2 .

Thus we have
Z a
(x2νn − x2νn′ ) ds sJν (xνn′ s/a)Jν (xνn s/a) = 0
0

providing
)#a
d d
" (
s Jν xνn′ s/a) Jν (xνn′ s/a) − Jν (xνn s/a) Jν (xνn′ s/a) = 0. (3.13)
ds ds 0

At the upper limit, s = a, this expression vanishes since xνn and xνn′ are roots of
the Bessel function, and at the lower limit, s = 0, the expression vanishes because
of the factor of s. Thus we have
xνn s xνn′ s
Z a ! !
ds sJν Jν = 0, n 6= n′
0 a a
The integral can be evaluated for n′ = n, with the result

xνn s xνn′ s a2
Z a ! !
ds sJν Jν = [Jν+1(xνn)]2 δnn′ .
0 a a 2
Chapter 2 24

3.5.5 Completeness

We now assume that the Bessel functions satisfy the completeness relation, and
therefore we can expand any function on 0 ≤ s ≤ a as
∞
X
f (s) = Aνn Jν (xνns/a)
n=1

where
2 xνn s
Z a !
Aνn = 2 2 ds sf (s)Jν .
a Jν+1(xνn) 0 a
This is a Fourier-Bessel series. This expansion is particularly useful for the
case where f (a) = 0, e.g. the Dirichlet problem, since each term in the expansion
satisfies the boundary conditions. An alternative set of basis functions is provided
by
√ yνn s
!
sJν ,
a
where the yνn are the roots of dJν /dx = 0, because this set still satisfies the
condition of eqn. 3.13. This choice is often more appropriate for the Neumann
problem.

3.5.6 Modified Bessel Functions

Note that if we had chosed a separation constant such that the solution in the
z-variable was
Z(z) = e±ikz ,
then the equation for R(s) would have been
d2R 1 dR  2 ν 2 
 

+ − k + 2 R = 0,
ds2 s ds s
which, after our usual substitution x = ks, becomes
d2 R 1 d2 R  ν2 
 

+ − 1 + 2 R = 0.
dx2 x dx x
Chapter 2 25

with solutions

Iν (x) = i−ν Jν (ix),

π
Kν (x) = iν+1Hν(1) (ix).
2
These, like Iν and Nν , are real functions of a real variable x, with limiting forms:

x≪1

!ν
1 x
Iν (x) →
Γ(ν + 1) 2

x
h   i

 − ln + γE + . . . ν=0
Kν (x) →  2ν
 Γ(ν) 2

ν 6= 0
2 x

x ≫ 1, ν

1 1
" !#
Iν (x) → √ ex 1 + O
2πx x
π −x 1
s " !#
Kν (x) → e 1+O .
2x x
Note again that only Iν (x) is regular as x → 0.
Chapter 2 26

3.6 Boundary-value Problems in Cylindrical Coordinates

Consider the solution of the boundary-value problem in a cylinder of radius a,

and length L, subject to the boundary conditions

φ(s, ϕ, 0) = 0
φ(a, ϕ, z) = 0; 0 ≤ z ≤ L
φ(s, ϕ, L) = V (s, ϕ)
z φ = V (ρ,θ)

a L
φ=0

φ=0
x

We look for separable solutions of the form

φ(s, ϕ, z) = R(s)T (ϕ)Z(z).

The angular factor has the form

Tm (ϕ) = A sin mϕ + B cos mϕ

where m is an integer greater than or equal to zero. The z factor is of the form

Z(z) = sinh kz
Chapter 2 27

where k is the separation constant, and we have imposed the boundary condition
Z(0) = 0. Finally, the radial component is of the form

Rm (s) = Cm Jm (ks) + Dm Nm (ks).

Since there are no charges in the region s ≤ a, the solution must be regular there,
and in particular must be finite at s = 0. Thus we have Dm = 0. Furthermore, R
must vanish at s = a, and thus

Jm (ka) = 0

and hence the values of k are

kmn = xmn /a, n = 1, 2, 3, . . .

where xmn is the n-th root of Jm (x) = 0. Thus our general solution may be written
∞
X B0n
φ(s, ϕ, z) = J0(k0ns) sinh(k0nz)
n=1 2
∞ X
X ∞
+ Jm (kmns) sinh(kmn z) [Amn sin mϕ + Bmn cos mϕ].(3.14)
m=1 n=1

We now impose the boundary condition at z = L:

∞
X B0n
V (s, ϕ) = J0(k0ns) sinh(k0nL)
n=1 2
∞ X
X ∞
+ Jm(kmn s) sinh(kmnL) [Amn sin mϕ + Bmn cos mϕ].
m=1 n=1

This is a Fourier series in ϕ and a Fourier-Bessel series in s. We apply the

orthogonality conditions, e.g., for Amn:
Z a Z 2π
ds s dϕ V (s, ϕ)Jm′ (km′ n′ s) sin m′ ϕ =
0 0
∞ X
X ∞ Z a 
sinh(kmnL) ds sJm (kmn s)Jm′ (km′ n′ s)
m=0 n=1 0
( Z 2π Z 2π )
× Amn dϕ sin mϕ sin m′ ϕ + Bmn dϕ cos mϕ sin m′ ϕ
0 0
 a2
 
∞ X
∞
[Jm+1(xmn)]2δnn′  {πδmm′ }
X 
= sinh(kmnL)Amn 
m=0 n=1 2
Chapter 2 28

and thus
2 Z a Z 2π
Amn = ds s dϕ V (s, ϕ)Jm(kmns) sin mϕ
πa2 sinh(kmnL)[Jm+1(xmn)]2 0 0

2 Z a Z 2π
Bmn = ds s dϕ V (s, ϕ)Jm(kmns) cos mϕ
πa2 sinh(kmnL)[Jm+1(xmn)]2 0 0

This form of the Fourier-Bessel seires is appropriate for problems confined to a

finite region of s. Suppose, however, that we are interested in the solution for all
0 ≤ s ≤ ∞.

Example

Determine φ(s, ϕ, z) in the upper-half plane z ≥ 0, with φ(s, ϕ, 0) = V (s, ϕ), and
φ finite as z → ∞. Then the separable solutions are of the form

e−kz [A sin mϕ + B cos mϕ]Jm(kz)

but there is now no restriction on the value of k other than it be positive (to
ensure that φ is finite as z → ∞). Thus the sum over discrete values of k becomes
an integral over k, and our general solution is
Z ∞ B0(k)
φ(s, ϕ, z) = dk e−kz J0(ks)
0 2
∞ Z ∞
dk e−kz {Am (k) sin mϕ + Bm (k) cos mϕ)} Jm(ks).
X
+
m=1 0

We still have a Fourier series in ϕ, but the Fourier-Bessel series has evolved to a
Bessel transform.
Imposing the boundary conditions at z = 0, we have
Z ∞ B0(k)
V (s, ϕ) = dk J0(ks)
0 2
∞ Z ∞
X
+ dk {Am (k) sin mϕ + Bm (k) cos mϕ} Jm (ks)
m=1 0
Chapter 2 29

and we can invert the Fourier series to obtain

Z ∞ 1 Z 2π
dk Am (k)Jm(ks) = dϕ V (s, ϕ) sin mϕ
0 π 0
Z ∞ 1 Z 2π
dk Bm (k)Jm(ks) = dϕ V (s, ϕ) cos mϕ. (3.15)
0 π 0
We can invert the Hankel transforms on the left-hand side using the complete-
ness relation Z ∞ 1
dx xJm(kx)Jm(k ′ x) = δ(k ′ − k).
0 k
Applying this to the first line of eqn. 3.15, we have
1Z∞ Z 2π Z ∞ Z ∞
ds s dϕ V (s, ϕ) sin mϕJm (k ′s) = ds s dkAm (k)Jm(ks)Jm(k ′s)
π 0 0 0 0
Z ∞ 1
= dk Am(k) δ(k − k ′ )
0 k
1
= ′
Am(k ′ ),
k
and thus we have
kZ∞ Z 2π
Am (k) = ds s dϕ V (s, ϕ) sin mϕJm (ks)
π 0 0
kZ∞ Z 2π
Bm (k) = ds s dϕ V (s, ϕ) cos mϕJm (ks)
π 0 0
Chapter 2 30

3.7 Expansion of Green Functions in terms of Orthogonal

Functions

The solutions found by separation of variables constituted complete sets of or-

thogonal functions satisfying the appropriate boundary conditions. We have
shown that any function, and in particular the Green function, satisfying the same
boundary conditions can be expanded as a series of these orthogonal functions.
We will illustrate the basic principle for our old friend, the Dirichlet Green function
for the sphere in spherical polar coordinates, and then proceed to discuss the
general construction using cylindrical coordinates.

3.7.1 Green function for the Sphere in Spherical Harmonics

Recall that the Green function for the region V satisfies

∇′2G(x, x′ ) = −4πδ(x − x′) for x, x′ ∈ V ,

and has the general form

1
G(x, x′ ) = ′
+ F (x, x′ ),
|x − x |
where F (x, x′ ) is a solution of Laplace’s equation in V , chosen to satisfy the
boundary conditions on G, e.g. for the Dirichlet Green function

G(x, x′ ) = 0 for x ∈ ∂V .

We have already seen the expansion of 1/|x − x′ | in terms of spherical harmonics,

viz.
∞ X l l
1 X 1 r< ∗ ′ ′
′
= 4π l+1 Ylm (θ , ϕ )Ylm (θ, ϕ).
|x − x | l=0 m=−l 2l + 1 r>
Suppose we wish to construct the Dirichlet Green function for the outside of a
sphere of radius a. We use the method of images, and obtain
1 a
G(x, x′ ) = −
|x′ − x| r|x′ − a2 /r2x|
Chapter 2 31

 rl a (a2/r)l 
 
l
∞ X 1
X ∗ ′ ′ <
= 4π Ylm (θ , ϕ )Ylm(θ, ϕ)  l+1 − .
l=0 m=−l 2l + 1 r> r r′l+1 

where we note that, for the image charge, r> = r′ , r< = a2 /r, since the image
charge is always inside the sphere. Thus we have
 l+1 
l 2

∞ X
l 1 r< 
1 a  
G(x, x′ ) = 4π ∗ ′ ′
X  
Y (θ , ϕ )Ylm(θ, ϕ)  l+1 −
1 lm

l=0 m=−l 2l +  r> a rr′ 


We have thus accomplished our goal of expressing the Green function as an ex-
pansion over orthogonal functions. There are some important observations we can
make by looking at the radial part
  
l+1 1 l a2l+1
l 2
r′l+1  r − r < r′

r< 1 a


rl+1 

{ l+1 −  ′ }= 1 a2l+1
.
r> a rr  l+1 r ′l


r − r′l+1 r > r′

• The radial part manifestly vanishes at r = a and r′ = a.

• It is symmetric under r ↔ r′ .

• The solution is a linear combination of the solutions of Laplace’s equation,

regarded as a function of r′ for fixed r, but a different linear combination for
r′ > r and r′ < r. We will see how this property arises below.

3.8 General Solution of Green Function in Cylindrical Po-

lars

The Green function satisfies

∇′2G(x, x′ ) = −4πδ(x − x′ ).

To express the r.h.s. in terms of cylindrical coordinates, we recall that

X 1
δ[g(x)] = δ(xi)
i g ′ (xi)
Chapter 2 32

where xi are the roots of g(x) = 0. Thus, in three dimensions, we have

−1
′ ∂(x, y, z) 1
δ(x − x ) = δ(s − s′ )δ(ϕ − ϕ′)δ(z − z ′ ) = ′
δ(s − s′ )δ(ϕ − ϕ′)δ(z − z ′ ).
∂(s, ϕ, z) s
Hence the Green function satisfies
4π
∇′2G(x − x′ ) = − ′
δ(s − s′ )δ(ϕ − ϕ′ )δ(z − z ′ ) (3.16)
s
where
1 ∂ ′ ∂ 1 ∂2 ∂2
!
′2
∇ = ′ ′ s ′ + ′2 ′2 + ′2 .
s ∂s ∂s s ∂ϕ ∂z
Note that in the following we will treat the unprimed indices as fixed parameters.
We will now specialise to the case where we wish to obtain the Green function in
a volume V encompassing the full angular range 0 ≤ ϕ ≤ 2π. Then any solution
can be expressed as a Fourier series in ϕ′ ,
∞ ′ ′
G(x, x′ ) = G(s, ϕ, z; s′, ϕ′, z ′ ) = Fm′ (s, ϕ, z; s′, z ′ )e−im ϕ .
X

m′ =−∞

Substituting this into eqn. 3.16, we have

∞ 1 ∂ ′ ∂ ′2 1
( " #
−im′ ϕ′ ′ ′
Fm′ (s, ϕ, z; s′, z ′ )
X
e ′ ′
s ′
F m ′ (s, ϕ, z; s , z ) − m
′2
m′ =−∞ s ∂s ∂s s
∂2

4π
+ ′2 Fm (s, ϕ, z; s , z ) = − ′ δ(s − s′ )δ(ϕ − ϕ′ )δ(z − z ′ ).
′ ′

′
∂z s
We now use the orthogonality properties of the exp imϕ to obtain
∞ 1 ∂ ′ ∂ ′2 1
Z 2π ( " #
′ i(m−m′ )ϕ′ ′ ′
Fm′ (s, ϕ, z; s′, z ′ )
X
dϕ e ′ ′
s ′
F m ′ (s, ϕ, z; s , z ) − m
′2
m′ =−∞ 0 s ∂s ∂s s
∂2

′ ′ 4π ′ ′
Z 2π
′
dϕ′ eimϕ δ(ϕ − ϕ′),

+ ′2 Fm′ (s, ϕ, z; s , z ) = − ′ δ(s − s )δ(z − z )
∂z s 0

yielding
1 ∂ ′ ∂Fm m2 ∂ 2 Fm 2
" #

′ ′
s ′
− ′2
F m + ′2
= − ′
δ(s − s′ )δ(z − z ′ )eimϕ .
s ∂s ∂s s ∂z s
Chapter 2 33

Thus we have explicitly exhibited the ϕ dependence, and can write

Fm (s, ϕ, z; s′, z ′ ) = fm (s, z; s′ , z ′ )eimϕ,

where fm obeys the P.D.E.

1 ∂ ′ ∂fm m2 ∂ 2 fm 2
" #

′ ′
s ′
− ′2
f m + ′2
= − ′
δ(s − s′ )δ(z − z ′ ). (3.17)
s ∂s ∂m s ∂z s
Thus from our original P.D.E. in three variables we now have a two-variable
P.D.E..
To proceed further, we must say something about the boundary conditions, or at
the very least specify the volume V . We will assume that is covers −∞ ≤ z ≤ ∞,
and then introduce the Fourier transform (F.T.) of fm , defined by
Z ∞
f˜m (s, z; s′, k) =
′
dz ′ eikz fm (s, z; s′ , z ′ )
−∞
1 Z∞
dke−ikz f˜m (s, z; s′ , k).
′
′ ′
fm (s, z; s , z ) =
2π −∞
We apply the F.T. operator to eqn. 3.17,
m2 ∂ 2 fm 
 
1 ∂ ′ ∂fm 2
Z ∞ " # Z ∞
′
′ ikz ′
dz ′ eikz  s′ ∂s′
s ′
− ′2
f m + ′2
= − ′
δ(s − s′
) dz e δ(z − z ′ ),
∞ ∂s s ∂z  s −∞

yielding
1 ∂  ′ ∂ f˜m  m2 ˜
 
2˜ 2
′ ′
s ′
− ′2
f m − k f m = − ′
δ(s − s′ )eikz
s ∂s ∂s s s
where we have used the well-known properties concerning the F.T. of a derivative.
We have now exhibited the z dependence of the function, and may write
1 ikz
f˜m (s, z; s′, k) = e gm (s, s′; k),
2π
giving
m2
 
1 ∂ ∂gm 4π
" #

′ ′
s ′ − ′2 + k 2 gm = ′ δ(s − s′ ).

s ∂s ∂s s s
Chapter 2 34

This is just a one-dimensional Green function equation, which we may write in

a more familiar form by substituting

x = ks
x′ = ks′ ,

yielding
∂  ′ ∂gm(x, x′)  m2 
   

′
x ′
− x 1 + ′2 gm (x, x′) = −4πδ(x − x′ ).
′
∂x ∂x x
This is just the modified Bessel equation, with inhomogeneous source. As we
noted earlier, the modified Bessel equation (like the Legendre equation) is of
Sturm-Liouville type:
d  ′ dg(x, x′) 
 

′
p(x ) ′
+ q(x′)g(x, x′) = −4πδ(x − x′)
dx dx
with

p(x′) = x′
m2
 

q(x′ ) = −x′ 1 + ′2  .
x
Thus we have finally reduced the problem to the solution of the Green function
for the Sturm-Liouville equation.

3.8.1 Green Function for the Sturm-Liouville Equation

We wish to determine the Green function to the equation

d  ′ dg(x, x′) 
 

′
p(x ) ′
+ q(x′)g(x, x′) = −4πδ(x − x′),
dx dx
defined on the interval x′ ∈ [a, b], with homogeneous boundary conditions at a
and b. Note that we regard x as some arbitrary, fixed parameter.
The Green function must possess the following properties:
Chapter 2 35

1. For x′ 6= x, g(x, x′) satisfies the homogeneous equation, i.e. the Sturm-
Liouville equation with no source on the r.h.s..

2. g(x, x′0 satisfies the homogeneous boundary condition at x′ = a and x′ = b,

e.g. g(x, x′) = 0.

3. g(x, x′) must be continuous at x′ = x. This is subtle; otherwise dg/dx′

would contain a δ-function, and d2g/dx′2 would contain the derivative of a
δ-function at x′ = x, which is more singular than the r.h.s. of the equation.

To see what happens at x′ = x, we integrate the equation from x − ǫ to x + ǫ:

d  ′ dg(x, x′) 
   
Z x+ǫ Z x+ǫ
′ ′ ′
dx′ δ(x − x′ ),
 
dx  ′ p(x ) ′
+ q(x )g(x, x ) = −4π
x−ǫ dx dx x−ǫ

leading to x+ǫ
dg(x, x′) 

Z x+ǫ
′
p(x )
′
+ dx′ q(x′)g(x, x′) = −4π.
dx x−ǫ
x−ǫ

Both q(x′) and g(x, x′) are continuous at x′ = x, and therefore we have
lim Z x+ǫ
ǫ→0 dx′ q(x′)g(x, x′) = 0,
x−ǫ

and we may write

lim ′ x+ǫ
 
′ dg(x, x )
ǫ → 0 p(x ) = −4π.
dx′ x−ǫ
Thus function p(x′) is also continuous at x′ = x, and thus
lim  dg(x, x′ = x + ǫ) dg(x, x′ = x − ǫ) 
 

p(x)× ǫ → 0  − = −4π
dx′ dx′ 

which we write as
dg(x, x′) 
 
4π
 = − ,
dx′ x =x
′ p(x)
i.e. there is a discontinuity in the slope of the Green function of magnitude 4π/p(x)
at x′ = x.
Chapter 2 36

Discontinuity in
slope

a x b
Thus we will write our Green function as

• a ≤ x′ ≤ x:
g(x, x′) = C1(x)y1(x′),
where y1 (x′) is a solution of the homogeneous equation satisfying the appro-
priate boundary condition at x′ = a.

• x ≤ x′ ≤ b:
g(x, x′) = C2(x)y2(x′)
where y2 (x′) is a solution of the homogeneous equation satisfying the appro-
priate boundary condition at x′ = b.

We now impose the conditions on the Green function at x′ = x

• g(x, x′) continuous at x′ = x:

C1(x)y1(x) − C2 (x)y2(x) = 0 (3.18)

• Discontinuity in slope is −4π/p(x):

4π
C2(x)y2′ (x) − C1 (x)y1′ (x) = − (3.19)
p(x)
Chapter 2 37

From eqn. 3.18, we have

C1(x)y1(x)
C2 (x) = .
y2 (x)
Substituting into eqn. 3.19, we find
C1(x)y1(x)y2′ (x) 4π
− c1 (x)y1′ (x) = −
y2 (x) p(x)
4π y2 (x)
⇒ C1 (x) = − ,
p(x) W [y1(x), y2(x)]
where the W is the Wronskian,

W [y1(x), y2(x)] = y1(x)y2′ (x) − y2 (x)y1′ (x).

Note that this method only works if y1 and y2 are linearly independent, since
otherwise the Wronskian vanishes.
Thus our general form for the Green function is
4π y2 (x)y1(x′)

a ≤ x′ ≤ x

−



p(x) W [y1(x), y2(x)]


g(x, x′) = 


4π y2 (x′)y1(x)
x ≤ x′ ≤ b

−



p(x) W [y1(x), y2(x)]



So, as we have already observed for spherical polar coordinates, the Green function
in the regions x′ < x and x′ > x comprises two different, linearly independent
solutions of the homogeneous equation.

3.8.2 Green Function for Modified Bessel Equation

We will now return to the case of the modified Bessel equation with δ-function
source
d  ′ dg(x, x′) 
 

′
p(x ) ′
+ q(x′)g(x, x′) = −4πδ(x − x′).
dx dx
A pair of linearly independent solutions is provided by the modified Bessel func-
tions Im (x) and Km(x). We will now consider the case where we require the
Chapter 2 38

solution over all space, i.e. x ∈ [0, ∞]. The solution must be finite at x = 0, and
thus
y1 (x′) = Im (x′).
If we further require that the solution be finite as x′ → ∞, then we have

y2 (x′) = Km (x′),

which we can see from the limiting behaviour quoted earlier. In this case, the
Wronskian is (see Jackson)
1
W [Im(x), Km(x)] = −
x
and thus our general solution for the Green function is
4π Km(x)Im(x′)

0 ≤ x′ ≤ x

−



x −1/x


gm (x, x′) = 

,
4π Km(x′ )Im(x)
x ≤ x′ ≤ ∞

−



x −1/x



which we may express as

gm (x, x′) = 4πIm (x<)Km(x>)

where x< = min(x, x′) and x> = max(x, x′).

3.8.3 Reconstruction of the Full Green Function

We reconstruct the full Green function in four steps:

1.

f˜m (s, z; s′, k) = gm (s, s′; k)eikz /2π

= 2Im(ks< )Km(ks>)eikz
Chapter 2 39

2.
1 Z∞
dk e−ikz f˜m (s, z; s′ , z ′ )
′
′ ′
fm (s, z; s , z ) =
2π −∞
1Z∞ ′
= dk eik(z−z ) Im(ks< )Km(ks>)
π −∞
3.
1Z∞ ′
Fm (s, ϕ, z; s′, z ′ ) = dk eik(z−z ) Im (ks<)Km (ks>)eimϕ
π −∞

1 X∞
im(ϕ−ϕ′ ) ∞
Z
′
′
4. G(x, x ) = e dk eik(z−z ) Im (ks<)Km(ks> ).
π m=−∞ −∞

Since we have evaluated the Green function with boundary conditions at infin-
ity, this last expression is just the expansion of |x − x′ |−1 in cylindrical polar
coordinates.
Chapter 2 40

3.9 Expansion of Green Function in Spherical Polar Co-

ordinates

This is performed in exactly the same way as for Cylindrical coordinates. In

spherical polars, the Green function satisfiese
4π
∇′2G(x, x′ ) = − ′2 δ(r − r′ )δ(ϕ − ϕ′ )δ(cos θ − cos θ′),
r
where
1 ∂ ∂ 1 ∂ ∂ 1 ∂2
! !
′2 ′2 ′
∇ = ′2 ′ r + ′2 sin θ ′ + ′2 2 ′ ′2 .
r ∂r ∂r′ r sin θ′ ∂θ′ ∂θ r sin θ ∂ϕ
We will consider the case where we require the Green function over the full angular
range 0 ≤ θ′ ≤ π, 0 ≤ ϕ ≤ 2π. Thus we can expand the Green function, as a
function of the primed variables with the unprimed variables fixed, in spherical
harmonics:
G(x, x′ ) = Fl′m′ (r, θ, ϕ; r′)Yl∗′ m′ (θ′, ϕ′).
X

l′ ,m′
Substituting this into the inhomogeneous equation we have
1 ∂ ′2 ∂Fl′ m′
( " #
Yl∗′m′ (θ′, ϕ′)
X
′2 ′
r ′
l′ ,m′ r ∂r ∂r
∗
1 ∂ 2Yl∗′m′ 
 
F l ′ m′  1 ∂ ′ ∂Yl′ m′
!
+ sin θ +
r′2 sin θ′ ∂θ′ ∂θ′ sin2 θ′ ∂ϕ′2 
4π
= − ′2 δ(r − r′ )δ(ϕ − ϕ′)δ(cos θ − cos θ′ ).
r
Now the spherical harmonics are solutions of Laplace’s equation on the unit
sphere, and, from eqn. 3.1, satisfy
1 ∂ 2Yl∗′ m′ 1 ∂ ∗
′ ∂Yl′ m′
!
∗
+ l(l + 1)Yl′m′ + sin θ = 0.
sin2 θ′ ∂ϕ′2 sin θ′ ∂θ′ ∂θ′
Thus our Green function equation becomes
1 ∂ ′2 ∂Fl′ m′ F l ′ m′
( " # )
− ′2 l(l + 1) Yl∗′ m′ (θ′ , ϕ′) =
X
′2 ′
r ′
l′ m′ r ∂r ∂r r
4π
− ′2 δ(r − r′ )δ(ϕ − ϕ′)δ(cos θ − cos θ′ ).
r
Chapter 2 41

∗
We now multiply by Ylm (θ′, ϕ′), and use the orthogonality properties of the spher-
ical harmonics:
1 ∂ ′2 ∂Flm Flm
" #
r − l(l + 1)
r′2 ∂r′ ∂r′ r′2
4π Z
= − ′2 dΩ′ Ylm (θ′, ϕ′)δ(r − r′ )δ(ϕ − ϕ′)δ(cos θ − cos θ′ )
r
4π
= − ′2 Ylm(θ, ϕ)δ(r − r′ ).
r
We may then write
Flm (r, θ, ϕ; r′) = gl (r, r′)Ylm(θ, ϕ)
where gl (r, r′) satisfies
d ′2 d
!

′
r ′
gl (r, r′) − l(l + 1)gl (r, r′) = −4πδ(r − r′ ).
dr dr
This is just the radial part of Laplace’s equation. To proceed further, we must
specifiy boundary conditions.

3.9.1 Dirichlet Green Function between Spheres at r = a and r = b

We require gl (r, r′) subject to the boundary conditions gl (r, a) = gl (r, b) = 0.

1. a ≤ r′ ≤ r: The solution y1(r′ ) of the homogeneous equation must satisfy

y1 (a) = 0. Now the general solution is of the form

y1 (r′) = A1 r′l + B1 r′−l−1,

and thus we have

A1al + B1 a−l−1 = 0 ⇒ B1 = −A1a2l+1

yielding
a2l+1 
 
′ ′l
y1(r ) = A1 r − ′l+1 .

r
Chapter 2 42

2. r ≤ r′ ≤ b: Then the solution y2(r′ ) of the homogeneous equation must

satisfy y2 (b) = 0, yielding
r′l
 
1
y2(r′ ) = B2  − .
r′l+1 b2l+1

We now construct the Wronskian

W [y1(r), y2(r)] = y1 (r)y2′ (r) − y2 (r)y1′ (r)

a2l+1 
 
2l + 1 
= −A1 B2 2 1 − 2l+1 .
r b

Noting that p(r) = r2, we observe that, once again, the Wronskian is independent
of the evaluation point, and we have general solution
a2l+1   1 rl 
    
′l

A1 r − ′l+1 B2 l+1 − 2l+1

 

r r b


; a ≤ r′ ≤ r

−4π


2l+1
 
a



−A1 B2(2l + 1) 1 − 2l+1 




b

gl (r, r′) = 


2l+1
 
′l
 ,
l a 1 r
A1 r − l+1  B2  ′l+1 − 2l+1 




r r b


; a ≤ r′ ≤ r

−4π


2l+1
 
a



−A1 B2(2l + 1) 1 − 2l+1 



b



which we may write in the more compact form

−1 
a2l+1  a2l+1   1 l
  
′ 4π  r l −
r>
gl (r, r ) = 1 − 2l+1 < l+1 l+1 − 2l+1 ,

2l + 1 b r< r> b
and hence
G(x, x′ ) = gl (r, r′)Ylm
∗
(θ′ , ϕ′)Ylm(θ, ϕ).
X
(3.20)
l,m

Note that it is also possible to recover this result using the method of images, but
in this case an infinite number of image charges are required.
Chapter 2 43

Example:

Consider the potential inside an

grounded, conducting sphere of ra- a
dius b, due to a uniform ring of charge
of radius a < b, and total charge Q, b
lying in the plane through the equa-
tor, and centred at the centre of the
sphere.

We can obtain the Green function by taking the a → 0 limit of eqn. 3.20:
l
(r<r> )l  ∗ ′ ′
 
∞ X
l 4π  r<
′ X
G(x, x ) = l+1 − Ylm (θ , ϕ )Ylm(θ, ϕ).
l=0 m=−l 2l + 1 r> b2l+1
The potential is then given by
′
1 Z′ 3 ′ ′ ′ 1 Z ′ ′ ∂G(x, x )
φ(x) = d x G(x, x )ρ(x ) − dS φ(x ) .
4πǫ0 V 4π S=∂V ∂n′
In our case the surface integral vanishes, because the potential vanishes there.
The (linear) charge density is given by
Q
ρ(x′ ) = 2
δ(r′ − a)δ(cos θ′).
2πa
Exercise: verify that the total charge is indeed Q. Thus the potential is
1 Z Q
φ(x) = dϕ′ d(cos θ′ )dr′ r′2 2
δ(r′ − a)δ(cos θ′ )
4πǫ0 2πa
 rl l
 
∞ l 1 (r r )
∗ < < >
(θ′, ϕ′)Ylm(θ, ϕ)  l+1
X X
× 4π Ylm − 2l+1 
l=0 m=−l 2l + 1 r> b
In this case we have azimuthal symmetry, and the only non-vanishing integrals
arise from the terms with m = 0, for which
v
u 2l+1
u
Yl0(θ, ϕ) = t
Pl (cos θ).
4π
Chapter 2 44

Thus we have
 rl (r<r>)l 
 
1 Z ′ ′2 Q ′ ∞
X <
φ(x) = r r 2 δ(r − a) Pl (0)Pl (cos θ)  l+1 − 2l+1 
4πǫ0 a l=0 r> b
∞ (−1)n(2n − 1)!!  r 2n 2n 
 
Q X < r>
=  r 2n+1
− P2n (cos θ),
4πǫ0 n=0 2nn! > b4n+1 
where we have used

P2n+1(0) = 0
(−1)n(2n + 1)!!
P2n (0) = ,
2nn!
and r< = min(r, a), r> = max(r, a).

3.10 Expansion of Green Function in terms of Eigenfunc-

tions

A closely related method to those discussed above is the expansion of the Green
function in terms of the eigenfunctions of some related problem. Consider the
solution of
∇2 ϕ(x) + [f (x) + λ]ϕ(x) = 0,
in a volume V bounded by a surface S, subject to ϕ satisfying certain homogeneous
boundary conditions for x ∈ S. In general, consistent solutions can be obtained
only for certain (possibly continuous) values of λ, which we will denote λn , the
eigenvalues. The corresponding solutions, the eigenfunctions, we will denote
ϕn(x). The eigenvalue equation is then

∇2ϕn + [f (x) + λn ]ϕn = 0. (3.21)

The eigenfunctions form a complete, orthogonal set of functions (the proof of

orthogonality follows that for the Sturm-Liouville equation), and we will assume
that they are normalised: Z
d3x ϕ∗m ϕn = δmn.
Chapter 2 45

Then any function satisfying the same homogeneous boundary conditions may
be expanded as a series in the eigenfunctions. Consider in particular a Green
function, satisfying

∇′2G(x, x′ ) + [f (x′ ) + λ]G(x, x′ ) = −4πδ(x − x′ ) (3.22)

where λ is, in general, not an eigenvalue. The corresponding eigenfunction expan-

sion is
G(x, x′ ) = an (x)ϕn(x′ ),
X

n
and, inserting in eqn. 11.11.1, we obtain

an (x){∇′2ϕn (x′ ) + f (x′ )ϕn (x′ ) + λϕn (x′ )} = −4πδ(x − x′ ).

X

We now use that ϕn is an eigenfunction of eqn. 3.21 with eigenvalue λn , and obtain

an (x)[λ − λn ]ϕn(x′ ) = −4πδ(x − x′ ).

X

Using the orthonormal property of the eigenfunctions, we obtain

ϕ∗n (x)
an (x) = 4π
λn − λ
and hence

′ X ϕ∗n (x)ϕn(x′ )
G(x, x ) = 4π
n λn − λ

This is often referred to as the spectral representation of the Green function.

Example: Green function in free space

Let us now specialise to Poisson’s equation, i.e. we set f (x) = 0 and λ = 0 in

eqn. 11.11.1. We will begin by considering the solution in free space, for which
the most closely related eigenvalue equation is the wave equation

(∇2 + k 2 )ϕk (x) = 0

Chapter 2 46

where k 2 is the (continuous) eigenvalue, and the corresponding normalised eigen-

function is
1 3/2 ik ·x
!
ϕk (x) = e ,
2π
with normalisation Z
d3 x ϕ∗k′ (x)ϕk (x) = δ(k − k′ ).
Then the expression for the Green function is

eik·(x −x) 1
′ !2
Z
′ 3
G(x, x ) = 4π dk
k2 2π
which we observe may be written

1 Z 3 eik·(x −x)
′
1
= dk .
|x − x′ | 2π 2 k2

Example: Dirichlet Green function inside a rectangular box

We define the surface of the box to be the planes x = 0, a, y = 0, b, and z = 0, c.

The most closely related eigenvalue problem is

∇2ϕ + klmn
2
ϕlmn = 0,

where the eigenvalues and normalised eigenfunctions are

2
m2 n2 
 
2 2l
klmn = π + 2 + 2
a2 b c
v
8 lπx mπy nπz
u
u
ϕlmn(x) = t
sin sin sin .
abc a b c
Thus we can immediately write down the Green function as
lπx mπy nπz lπx′ mπy ′ nπz ′
′ 32 X sin a sin b sin c sin a sin b sin c
G(x, x ) = .
πabc l,m,n l2 m2 n2
+ 2 + 2
a2 b c
Chapter 4

Multipoles and the Electrostatics of

Macroscopic Media

The simplest source for an electrostatic field is a point charge; such a source is
sometimes known as a pole. The arrangement of two point charges, of equal but
opposite sign, is known as a dipole. The concept of a dipole plays a crucial rôle
in electrostatics:

• Even in the case of a neutral atom or molecule, the positive and negative
charges can become separated, e.g. by an applied external electric field. In
that case, the atom or molecule gives rise to an electrostatic field that can
be approximated by a dipole

• The concept of dipoles, and, more generally, multipoles, leads to an im-

portant method for obtaining the electrostatic field and potential far from a
charge distribution, the multipole expansion.

4.1 Introduction and Revision: Electric Dipoles

Consider two charges −q and q at x1 and x2 respectively, and let a be the position
vector of q relative to −q.

1
Chapter 2 2

-q a q

x’
x1
x2 P
x
Let x′ be the mid point of the dipole, so that

x1 = x′ − a/2
x2 = x′ + a/2

Then the potential at the point P is

   
1  q −q  1  1 1
φ(x) = + = q − 
4πǫ0 | x − x2 | | x − x1 | 4πǫ0 | x − x′ − a/2 | | x − x′ + a/2 |
We will now consider the case where the separation between the charges is much
less than the distance of the point P from the charges, i.e. |a| ≪| x − x′ |. Then
we have
−1/2
|a|2


|x − x′ ± a/2|−1 ′ 2
± a · (x − x′ )
 
= |x − x | +
4
−1/2
a · (x − x′ ) |a|2 

′ −1

= |x − x | 1 ± + .
|x − x′ |2 4|x − x′ |2 
Expanding as a series in |a|2 /|x − x′|2 using the binomial expansion, we obtain
a · (x − x′ )  |a|2 
 !   
1
|x − x′ ± a/2|−1 = |x − x′ |−1 1 + −

± +O 
2 |x − x′ |2 |x − x′ |2 
Thus
1 qa · (x − x′ )
φ(x) = .
4πǫ0 |x − x′ |3
Chapter 2 3

We now take the limit |a| → 0, q → ∞, with aq = p fixed and finite. This defines
a simple or ideal dipole and we have

1 p · (x − x′ )
φD (x) =
4πǫ0 |x − x′|3

• p is the vector moment or dipole moment of the dipole.

• φD (p) is the potential at x due to a dipole of moment p at r′ .

We can obtain the electrostatic field due to a dipole by applying E(x) =

−∇φD (x), and obtain
1 3(p1 · x)x − r2p1
E(x) = (4.1)
4πǫ0 r5
for a dipole at the origin.

4.1.1 Dipole in External Electrostatic Field

In this subsection, we will consider not the field due to a dipole, but rather the
energy and forces on a dipole in an external field E(x) = −∇φ(x).

Potential Energy of Dipole in External Electrostatic Field

Recall from Section 3.5 that for a charge q in an electrostatic potential φ(x),
the potential energy is
U (x) = qφ(x)
Let us now apply this to the case of a dipole in an external field; once again, a
is the separation of the charge q from −q.
Chapter 2 4

E(x - a/2) E(x + a/2)

a
-q q
x

The potential energy of the dipole is

UD (x) = (−q)φ(x − a/2) + qφ(x + a/2).

If the separation between the charges is small, we can expand about x to obtain

φ(x ± a/2) =
1 ∂ 1 ai aj ∂ 2
φ(x) ± ai φ(r) + φ(x) + . . .
2 ∂xi 2! 4 ∂xi ∂xj
1
= φ(x) ± a · ∇φ(x) + O(a2 )
2
Thus we have
1 1
UD (x) = q [φ(x) + a · ∇φ(x) − φ(x) + a · ∇φ(x) + O(a3 )]
2 2
2
= q a · ∇φ(x)[1 + O(a )]

Now take the point dipole limit, a → 0, q → ∞, aq = p fixed. Then

UD (x) = p · ∇φ(x)

Aside: why did I take x to be at the mid point of the dipole? Because for a simple
dipole, all the corrections to the formula above involving even derivatives of φ(x)
vanish. It just makes the expansion neater, but of course I could have performed
the expansion about any point between the charges.
Recalling that E(x) = ∇φ(x), we have
Chapter 2 5

UD (x) = −p · E(x)

Note that the potential energy of a dipole is a minimum when E and p are
parallel

Force on Dipole in External Electrostatic Field

We will now consider the force on an electric dipole.

q E(x + a/2)
-q a The force on the dipole is
q
FD (x) = −qE(x − a/2) + qE(x + a/2)
-q E(x - a/2) x
Once again, we can expand about r:
1
E(x ± a/2) = E(x) ± (a · ∇)E + O(a2 ).
2
We thus obtain
FD (x) = q(a · ∇)E(x) = (p · ∇)E(x) (4.2)
Now since E(x) is an electrostatic field, it is irrotational :
∇ × E(x) = 0.
Let c be a constant vector, and let A(x) be an arbitrary vector field. Then we
have the identity
∇(c · A(x)) = c × (∇ × A(x)) + (c · ∇)A(x)
which we can apply to equation (4.2) to obtain
FD (x) = ∇(p · E(x)) = −∇UD (x)
using ∇ × E(x) = 0. Thus the force on a dipole is just minus the gradient of the
potential energy, and furthermore for a uniform external field, independent of x,
the force is zero.
Chapter 2 6

Torque on a Dipole in an External Field

We will now evaluate the torque, or moment of the force, τ on a simple dipole
about its centre. This is just the moment of the forces acting on the two charges
about the centre of the dipole:
1 1
! !
τ = a × (+q) E(x + a/2) + − a × (−q ) E(x − a/2)
2 ! 2
1 
1 1 2

= a × q E(x) + (a · ∇) E(x) + E(x) − (a · ∇) E(x) + O(a )
2 2 2
ie τ = p × E(x) in the point dipole limit

• Note that the torque about some point other than the centre of the dipole
will be different.

• τ = p × E(x) is true for dipoles other than point dipoles if E(x) is constant
over the dipole.

4.1.2 Force between Two Dipoles

Many materials are dipolar; the positive and negative materials are separated.
Here we will consider the force between a dipole p1 at x1 and p2 at x2. The force
F21 on the dipole at x2 due to the electrostatic field E1 produced by the dipole p1
is
· (x2 − x1 ) (x2 − x1 ) − p1 |x2 − x1 |2 
   
3 p
1
F 21(x2) = (p2 ·∇2) E 1 (x2 ) = (p2 ·∇2 ) C  |x2 − x1 |5 

where ∇2 means that we take derivatives with respect to x2 (the position vector
of dipole p2), and we have used eqn. (4.1). As discussed above, we can express
this as
F21(x2) = ∇2 (p2 · E1 (x2)).
Note that the force F12(x1) is equal and opposite to F21(x2).
Chapter 2 7

4.2 Multipole Expansion

In this section, we will see why the concept of dipoles, and more generally multi-
poles, is so important in electrostatics. Consider the case of a charge distribution,
localised to some volume V . For convenience we will take the origin for our vectors
inside V .
x - x’ P
We have that the potential due to the charge
V distribution within V at a point P outside the
x
x’ volume is:
1 Z ρ(x′ ) dV ′
φ(x) =
4πǫ0 V |x − x′ |

For r much larger than the extent of V , i.e. r ≫ r′ for all x′ such that ρ(x′ ) 6= 0,
we can expand the denominator

|x − x′ |−1 = {|r|2 − 2x · x′ + r′2}−1/2

−1/2
x · x′ r′2 

−1

= r 1 − 2 2 + 2 
r r
x · x′
 
1 ′2 2

= 1 + + O(r /r )
r r2 

Thus we have
1 1 x · x′
′
= + 3 + O(r′2/r3).
|x − x | r r
Hence we can write
 
1 Q P · x 1 X 3 xi xj
φ(x) = + 3 + Qij 5 + O(1/r5)
4πǫ0 r r 2 i,j=1 r

where
Z
Q = ρ(x′ )dV ′ is the total charge within V
V
Chapter 2 8
Z
P = ρ(x′ )x′ dV ′ is the dipole moment of the charge about the origin
ZV
Qij = ρ(x′ )(3x′ix′j − r′2 δij )dV ′ is the quadrupole moment of the charge.
V

• We have defined the moments with respect to a particular point, e.g. the
dipole moment is the integral of the displacement x′ times the charge
density ρ(x′ ). In general, the moments depend on the choice of “origin”.
What about the total dipole moment when the total charge is zero?

• At large distances from the charge distribution, only the first few moments
(Q, P , quadrupole moment, . . .) are important.

• For a neutral charge distribution, the leading behaviour is given by the

dipole moment.

Example:

The region inside the sphere: r < a, contains a charge density

ρ(x, y, z) = f z (a2 − r2 )

where f is a constant. Show that at large distances from the origin the potential
due to the charge distribution is given approximately by
2f a7 z
φ(x) =
105ǫ0 r3
Use the multipole expansion in SI units:
1 Q P ·x 1
!!
φ(x) = + + O
4πǫ0 r r3 r3
In spherical polars (r, θ, ϕ),

x = r sin θ cos ϕ ; y = r sin θ sin ϕ ; z = r cos θ

The total charge Q is

Z Z 2π Z π Z a  
Q = ρ(x) dV = f r cos θ (a − r ) r2 sin θ dr dθ dϕ = 0.
2 2
V 0 0 0
Chapter 2 9

The integral vanishes because

Z π Z π 1
cos θ sin θ dθ = sin(2θ) dθ = 0.
0 0 2
The total dipole moment P about the origin is
Z Z
P = x ρ(x) dV = r er ρ(x) dV
V V ¯
Z 2π Z π Z a
= r (sin θ cos ϕ i + sin θ sin ϕ j + cos θ k)
0 0 0  
f r cos θ (a − r ) r2 sin θ dr dθ dϕ.
2 2

The x and y components of the ϕ integral vanish. The z component factorises:

Z 2π Z π
2
Z a 2 2a7
Pz = f dϕ sin θ cos θ dθ r4 (a2 2
− r ) dr = f 2π .
0 0 0 3 35
Putting it all together, we obtain
1 8πa7 f k · x 2f a7 z
φ(x) = = .
4πǫ0 105 r3 105ǫ0 r3

4.2.1 Multipole Expansion using Spherical Harmonics

To proceed further, we go back to our expansion of a pole in spherical harmonics

∞ X l l
1 X 1 r< ∗ ′ ′
′
= 4π l+1 Ylm (θ , ϕ )Ylm (θ, ϕ).
|x − x | l=0 m=−l 2l + 1 r>

We assume that the charge is confined to a sphere of radius a, and take the centre
of the sphere to be the origin for our vectors. Then for the case r > a, we have

r< = r ′
r> = r,

and we have
1 X ∞ X l 1 Ylm(θ, ϕ) Z Z
φ(x) = l+1
dΩ′ dr′ r′2Ylm
∗
(θ′, ϕ′)r′l ρ(x′ ).
ǫ0 l=0 m=−l 2l + 1 r
Chapter 2 10

We now write Z
qlm = dΩ′ dr′r′2 Ylm
∗
(θ′ , ϕ′)r′l ρ(x′ )
so that the expansion may be written

1 X 1 Ylm (θ, ϕ)
φ(x) = qlm .
ǫ0 l,m 2l + 1 rl+1

This is the multipole expansion using spherical harmonics. To make the connec-
tion with our previous expansion, it is useful to consider the few few terms in
Cartesian coordinates
1 Z 3 ′ ′ 1
q00 = √ d x ρ(x ) = Q
4π 4π
v v
u 3 Z u 3
u u
3 ′ ′ ′ ′
q11 = − t
d x ρ(x )(x − iy ) = −t (Px − iPy )
8π 8π
v v
u 3 Z u 3
u u
q10 = t
d3 x′ ρ(x′ )z ′ = t Pz
4π
v
4π v
1 t 15 1u 15
u Z u
3 ′ ′ ′ ′ 2
u
q22 = d x ρ(x )(x − iy ) = t (Q11 − 2iQ12 − Q22)
4 2π 12 2π
v v
u 15 Z 1 u 15
u u
3 ′ ′ ′ ′ ′
q21 = − t
d x ρ(x )z (x − iy ) = − t (Q13 − iQ23)
v
8π 3 8π
v
1ut 5 1u 5
u Z u
q20 = d3x′ ρ(x′ )(3z ′2 − r′2) = t Q33.
2 4π 2 4π
Note that the components for negative m can be trivially obtained using

ql,−m = (−1)mqlm
∗
.

In general, for the l-th multipole moment, there are (l + 1)(l + 2)/2 components
in Cartesian coordinates, while only 2l + 1 components using spherical harmonics.
There is no inconsistency here - the Cartesian tensors are reducible under rota-
tions (i.e. mix with tensors having few indices under rotations) whilst the tensor
moments expressed in spherical harmonics are irreducible (i.e. the qlm for fixed
Chapter 2 11

l mix only amongst themselves under rotations); that is why we remove the trace
in the quadrupole moment Qij , to give us 5 irreducible components.
We can express the electric field components trivially in spherical harmonics. In
particular, the contribution of definite l, m is
1 l+1 1
Er = Ylm(θ, ϕ)qlm l+2
ǫ0 2l + 1 r
1 1 1 ∂
Eθ = − qlm l+2 Ylm(θ, ϕ)
ǫ0 2l + 1 r ∂θ
1 1 1 im
Eϕ = qlm l+2 Ylm(θ, ϕ).
ǫ0 2l + 1 r sin θ
If we now consider the case of a ideal dipole p along the z-axis, then
v
3
u
u
q10 = t
p
4π
q11 = q1,−1 = 0

and we have
v v
1 2u 3 3 p cos θ 2p cos θ
u u
u
Er = t t
3
=
ǫ0 3 4π 4π r 4πǫ0 r3
v v
1 1ut 3 1u 3 p sin θ
u u
Eθ = p 3 t sin θ =
ǫ0 3 4π r 4π 4πǫ0 r3
Eϕ = 0,

which reduces to the expression we derived earlier for an ideal dipole, eqn. 4.1.

4.2.2 Point Dipole vs. Dipole Moment

There is a danger in using the expression for the electrostatic field due to an ideal,
or point dipole. To see this, consider the electrostatic field E(x) due to a localised
charge distribution ρ(x). In partciular, consider the integral of E over some sphere
of radius R, the centre of which we will take as the origin of our vectors.
Chapter 2 12

-
- -

+
+
+ +

We have Z Z Z
3 3 2
d xE = − d x ∇φ = −R dΩφ(x)n
r<R r<R

where n is a unit normal outward from the surface of the sphere, and we have
used the generalisation of the divergence theorem.
Using Coulomb’s law for an extended charge distribution, we may write
Z
3 R2 Z 3 ′ ′
Z
n
d xE = d x ρ(x ) dΩ .
4πǫ0 |x − x′ |
Now we can evaluate the x integration by writing the vector n = sin θ cos ϕi +
sin θ sin ϕj + cos θk, and then expressing these terms in spherical harmonics as
v  
u 8π Y11 (θ, ϕ) + Y1,−1(θ, ϕ) 
u
sin θ cos ϕ = −t 
3 2
v  
u 8π Y11 (θ, ϕ) − Y1,−1(θ, ϕ) 
u
sin θ sin ϕ = −t 
3 2i
v
u 4π
u
cos θ = t
Y10(θ, ϕ)
3
Chapter 2 13

Thus only the l = 1 terms contribute, and using the orthogonality property of
spherical harmonics we have
Z n 4π r< ′
dΩ = n
|x − x′ | 2
3 r>
where n′ is a unit vector in the direction of x′ . Hence we have
Z
3 R2 Z 3 ′ r< 4π ′ ′
d xE = − dx 2 n ρ(x )
4πǫ0 r> 3
R 2 Z 3 ′ r< ′ ′
= − d x 2 n ρ(x ) (4.3)
3ǫ0 r>
where r< = min(r′, R).
We now consider two cases
1. Sphere completely encloses the charge density. Then we have r< = r′ , and
r> = R, and we have, from eqn. 4.3,
Z
P
d3 xE = − , (4.4)
3ǫ0
where P is the electric dipole moment. Note that this expression is indepen-
dent of the size of the sphere, provided it completely encloses the dipole.

2. Charge density completely outside the sphere. Then we have r< = R, r> = r′ ,
and we have
Z
3 R 3 Z 3 ′ n′
d xE = − d x ′2 ρ(x′)
3ǫ0 r
4π 3
= R E(0).
3
Thus the average value of the electric field over a spherical volume containing
no charge is just the value of the field at the centre of the sphere.
Let us now consider the corresponding expression for the integrated E in the case
of an ideal dipole, eqn. 4.1:
Z
3
Z1 3(p · n)n − p
d xE(x) = d3 x
r<R r<R 4πǫ0 r3
= 0 Exercise: let p = pk, and work in spherical polars.
Chapter 2 14

For this to be consistent with eqn. 4.4, our expression for the electrostatic field
due to a dipole at x0 must be modified
 
1  3n(p · n) − p 4π
E(x) = − pδ(x − x0) .
4πǫ0 |x − x0|3 3
This expression only changes the electric field at the position of the dipole, and we
can then, with some care, use the expression as if we were using ideal, or point,
dipoles. The δ-function contains information about the finite distribution of the
charge lost in the multipole expansion.
Chapter 2 15

4.3 Energy of Charge Distribution in External Electro-

static Field

The energy is given by Z

W = d3x ρ(x)φ(x).
We now suppose that φ is slowly varying, so that
1X ∂ 2φ
φ(x) = φ(0) + x · ∇φ + xi xj + ...
2 i,j ∂xi∂xj
1X ∂Ei
= φ(0) − x · E(0) − xi xj .
2 i,j ∂xj
Now in the case of an external electrostatic field, we have ∇ · E = 0, and thus we
may write
 
1X  ∂E
i 1 
φ(x) = φ − x · E − xi xj  − δij ∇ · E 
2 i,j ∂xj 3
1X ∂Ei
= φ−x·E− [3xixj − δij r2 ]
6 i,j ∂xj
where the derivatives are evaluated at 0. Thus we have
Z 1X ∂Ei
W = d3x ρ(x){φ(0) − x · E − [3xixj − δij r2 ] }
6 i,j ∂xj
1X ∂Ei
= φ(0)Q − E(0) · P − Qij (0)
6 i,j ∂xj

4.4 Electrostatics with Ponderable Media

So far we have only considered the case of electrostatics in free space. We will
now consider the case of macroscopic materials in the presence of electric fields.
Such materials are classified according to whether or not electrons, or charges,
can flow over long distances. In the case of conductors, charges can move freely
about the material, and, as we have already seen, generate an induced field that
exactly cancels any applied external field.
Chapter 2 16

In this chapter we consider the case of dielectics. Here the electrons are bound
to atoms, and have only limited freedom to move. The material might have an
inherent dipole moment, or a dipole moment might be generated by the presence
of an external electric field. The crucial property of a dielectric is that

∇ × E = 0.

Thus

• We have a conservative electric force

• We can express the field as the gradient of a potential

In the following, we will assume the applied field induces a dipole moment, but no
higher moments. Now consider the potential at x due to the charge, and dipole
moment, of a volume ∆V at x′ :
1  ρ(x′ )∆V P (x′ ) · (x − x′ )
 

∆φ(x, x′ ) = + ∆V  ,
4πǫ0 |x − x′ | |x − x′ |3
where x is outside the volume ∆V . The dipole moment per unit volume is called
polarization. We now pass to an integral in the usual way, and obtain
1 Z 3 ′  ρ(x′ ) P (x′ ) · (x − x′ ) 
 

φ(x) = dx +
4πǫ0 V |x − x′ | |x − x′ |3
1 Z 3 ′  ρ(x′ )
  
1
= dx ′
+ P (x′ ) · ∇′  ′
 (integ. by parts)
4πǫ0 |x − x | |x − x |
′
1 Z 3 ′ 1 ′ ′ ′ 1 Z ′ P (x ) · n
= dx [ρ(x ) − ∇ · P (x )] + dS
4πǫ0 V |x − x′ | 4πǫ0 S=∂V |x − x′ |
This expression can be rewritten as follows
1 Z 3 ′ ρf (x′ ) + ρb (x′) 1 Z ′
′ σb (x )
φ(x) = dx + dS
4πǫ0 V |x − x′ | 4πǫ0 S=∂V |x − x′ |
where σb ≡ P · n is the surface density of the bound charge, ρb ≡ −∇ · P is the
volume density of the bound charge, and the “old” charge density ρ is called the
free charge density ρf to distinguish from the density of the bound charge.
Chapter 2 17

Thus Maxwell’s equation becomes

1
∇·E = [ρf − ∇ · P ].
ǫ0
which we can write as
∇ · D = ρf
where
D ≡ ǫ0 E + P
is the electric displacement. Note that −∇ · P is the polarisation charge
density.
We now suppose that the media is isotropic, i.e. no preferred direction. Then the
induced dipole moment must be aligned with E, and we set

P = ǫo χe E

where χe is the electric susceptibility. Thus we have

D = ǫ0 E + ǫ0 χe E = ǫE

where ǫ = ǫ0 (1 + χe ). Note that ǫ/ǫ0 is the dielectric constant. Finally, if the

material is uniform, then χe does not depend on position, and we have

D = ǫE, with ∇ · E = ρf /ǫ.

4.4.1 Boundary Conditions at Boundary between Materials

C
We will now consider the boundary conditions at n 21
the boundary between two materials, of permit-
tivities ǫ1 and ǫ2 , and with electric fields E 1, D1 E 2 , D2
and E 2, D2 respectively. E 1 , D1
Chapter 2 18

Tangential condition

We have that ∇ × E = 0, and thus, applying Stoke’s theorem to the closed curve
C shown above, we have Z
E · dl = 0,
C

yielding
|| ||
E1 = E2
which we can express as

(E 2 − E 1) × n21 = 0

where n21 is the normal from 1 to 2.

Normal condition

Applying Gauss’ law to the usual elementary pill-box we have

Z Z
∇·D =ρ ⇒ D · dS = ρf dV

from which we find

(D2 − D1 ) · n21 = σf

where σ is the macroscopic free surface charge density at the interface.

To summarise, at the interface between two dielectrics:

• The tangential component of E is continuous.

• The normal component of D has a discontinuity given by

(D2 − D1 ) · n21 = σf
Chapter 2 19

4.5 Boundary-value Problems with Dielectrics

The method we adopt here essentially follows that of the solution of boundary-
value problems in vacua, with the boundaries given by conducting surfaces. The
method is best illustrated by examples.

Example:

A point charge q in a material of permittivity ǫ1 a distance d from the interface

with a charge-free region of permittivity ǫ2 .
ε2 ε1
z
P
q’ x q

The boundary conditions at the interface z = 0 are

ǫ1 Ez (0+) = ǫ2 Ez (0−) (normal on D)

Ex(0+) = Ex(0−) (tangential )
Ey (0+) = Ey (0−) (tangential ).

In order to determine the potential in the region z > 0, let us try an image charge
q ′ at z = −d. Then the potential at x is
q′
 
1  q
φ(x)|z>0 = + .
4πǫ1 |x − dez | |x + dez |
We know that the potential in the region z < 0 must satisfy Laplace’s equation in
that region, and therefore, in particular, there cannot be any poles in the region
Chapter 2 20

z < 0. Therefore, let us try the potential due to a charge q ′′ at the postion of our
original charge q:
1 q ′′
φ(x)|z<0 = .
4πǫ2 |x − dez |
We now introduce cylindrical polar coordinates, so that
1 q ′′

z<0



4πǫ2 {ρ2 + (z − d)2}1/2




φ(ρ, θ, z) = 
q′
 
 1  q 
+ z>0



4πǫ1  {ρ2 + (z − d)2}1/2 {ρ2 + (z + d)2}1/2 



We have two unknowns, q ′ and q ′′ , which we determine by imposing the boundary

conditions at z = 0. We begin with the tangential condition. We have that
Eρ = −∂φ/∂ρ, and thus
1 q ′′ ρ

z = 0−



4πǫ2 (ρ2 + d2)3/2




Eρ = 
q′ρ
 
 1  qρ 
+ z = 0+



4πǫ1  (ρ2 + d2 )3/2 (ρ2 + d2)3/2 



Thus the tangential boundary condition is

1 1 ′′
[q + q ′ ] = q ⇒ q ′′ ǫ1 = (q + q ′ )ǫ2. (4.5)
4πǫ1 4πǫ2
To impose the normal boundary condition, we note that
−dq ′′

 1
z = 0−



4πǫ2 (ρ2 + d2)3/2



Ez =  ,
 1 d ′
(q − q) z = 0+



4πǫ1 (ρ2 + d2)3/2



from which we find

q ′′ + q ′ = q. (4.6)
We can solve for q ′ and q ′′ from eqns. 4.5 and 4.6, yielding
ǫ2 − ǫ1
!
′
q = − q
ǫ1 + ǫ2
2ǫ2
q ′′ = q.
ǫ1 + ǫ2
Chapter 2 21

Thus we have a solution that satisfies the Laplace’s equation in z < 0, and Pois-
son’s equation in z > 0, and the correct boundary conditions at z = 0. Thus, by
our uniqueness theorem, it is the solution.
To see the form of the field lines we consider two cases, ǫ1 > ǫ2 and ǫ1 < ǫ2 ; in
both cases the field lines for z < 0 are those of a point charge, of magnitude q ′′ ,
at q.

1. ǫ1 > ǫ2 .

Then q ′ is same
sign as q.

2. ǫ2 > ǫ1 .

Then q ′ and q have

different signs.

In order to compute the polarisation (bound) charge density, σpol = −∇ · P , we

observe that P i = ǫ0 χi E i, i = 1, 2, where ǫi = ǫ0 (1 + χi ). Thus we have

P i = (ǫi − ǫ0)E i.

Clearly the polarisation charge density vanishes, except at the point charge q, and
at the interface between the two materials. At the interface, there is a disconti-
Chapter 2 22

nuity in P , and integrating over the discontinuity we obtain

σb = −(P 2 − P 1) · n21, (4.7)

where n21 is the unit normal from region 1 to region 2, and P 2 and P 1 are the
polarisations at z = 0− and z = 0+ respectively. Thus we have
d q ′′
 
 1 d ′ 1 
σb = − (ǫ1 − ǫ0 ) (q − q) − (ǫ2 − ǫ0 )
4πǫ1 (ρ2 + d2)3/2 4πǫ2 (ρ2 + d2)3/2 
d ǫ2 − ǫ0 ′′ ǫ1 − ǫ0 ′
( )
= q − (q − q)
4π(ρ2 + d2 )3/2 ǫ2 ǫ1
dq
= {2(ǫ2 − ǫ0 )ǫ1 + (ǫ1 − ǫ0 )(ǫ1 + ǫ2 + ǫ2 − ǫ1 )}
4π(ρ2 + d2 )3/2(ǫ1 + ǫ2 )ǫ1
q ǫ0 (ǫ2 − ǫ1 ) d
= −
2π ǫ1 (ǫ2 + ǫ1 ) (ρ + d2)3/2
2

Note that in the limit ǫ2 /ǫ1 ≫ 1, the electric field in region z < 0 becomes very
small, and the polarisation charge density approaches the value of the induced
surface charge density for a conductor at z = 0, up to the factor of ǫ0 /ǫ1. In that
sense, the material in z < 0 behaves as a conductor.

Example

E0

Dielectric sphere, radius a, dielectric

a
constant ǫ/ǫ0 , in uniform field along ϕ
z-axis. θ
Chapter 2 23

We will work in spherical polar coordinates, and express our solution as an ex-
pansion in Legendre polynomials:

l
l Al r Pl (cos θ) r<a
 P

φ(r, θ, ϕ) = l −l−1
,
l [Bl r + Cl r ]Pl (cos θ) r>a
 P


where we have noted that the potential must be finite at r = 0.

To determine the coefficients, we impose the boundary conditions. At large dis-
tances, the potential is that for a uniform field along the z axis, and thus our
boundary condition at infinity is
φ(ρ, θ, ϕ) −→ −E0 r cos θ as r −→ ∞
.
We now impose the boundary conditions at the surface of the sphere
Eθ (a−) = Eθ (a+) (tangential condition)
ǫ0 Er (a+) = ǫEr (a−) (normal condition)
The boundary condition at infinity tells us
B1 = −E0
Bl = 0 l 6= 1
To impose the other boundary conditions, we evaluate the components of the
electric field, beginning with Eθ :
l−1 d

X
− A r Pl (cos θ) r<a


l


dθ


Eθ =  X l ,
−l−2 d d
 − Cl r Pl (cos θ) − B1 P1 (cos θ) r > a



dθ dθ

l
From the generalised Rodrigues’ formula, we have
d
Pl1 (x) = (−1)1(1 − x2 )1/2 Pl (x)
dx
d
⇒ Pl1 (cos θ) = − sin θ Pl (cos θ)
d cos θ
d
= Pl (cos θ),
dθ
Chapter 2 24

whence 
Al rl−1Pl1 (cos θ)
X




− r<a
Eθ =  l .
Cl r−l−2Pl1 (cos θ) − B1P11 (cos θ) r > a
X


 −
l
The radial component is straightforward,

Al .l.rl−1Pl (cos θ)
X




− r<a
Er =  l .
Cl (l + 1)r−l−2Pl (cos θ) − B1P1 (cos θ) r > a
X



l

Thus imposing the tangential boundary condition we have

Al al−1Pl1 (cos θ) = Cl a−l−2Pl1 (cos θ) + B1P11 (cos θ).

X X

l l

Using the orthogonality property of the Legendre polynomials, we have, for l 6= 1,

2 (l + 1)! 2 (l + 1)!
Al al−1 = Cl a−l−2
2l + 1 (l − 1)! 2l + 1 (l − 1)!
⇒ Al al−1 = Cl a−l−2
⇒ Al = Cl a−2l−1. (4.8)

For the case l = 1, we have

A1 = C1a−3 − E0. (4.9)

The normal boundary condition yields

 

Cl (l + 1)a−l−2Pl (cos θ) − B1P1 (cos θ) = −ǫ Al l al−1Pl (cos θ).

X  X
ǫ0 
l l

Once again, there are two cases

ǫ0[Cl (l + 1)a−l−2] = −ǫAl lal−1 l 6= 1 (4.10)

ǫ0 [2C1a−3 + E0] = −ǫA1 l=1 (4.11)

Substituting eqn. 4.8 into eqn. 4.10, we find

Al = Cl = 0, l 6= 1.
Chapter 2 25

Finally, from eqns. 4.9 and 4.11 we find

−3E0
A1 =
2 + ǫ/ǫ0
 
ǫ/ǫ0 − 1  a3 E
C1 =  0
2 + ǫ/ǫ0
Thus we have
3

− E0r cos θ r<a



2 + ǫ/ǫ0



φ(r, θ, ϕ) =  
ǫ/ǫ − 1

a3 .
0
−E0r cos θ + cos θ r > a

  

2 + ǫ/ǫ0 r2



• Inside the sphere, the field is parallel to the field at infinity,

3
E in = E ,
2 + ǫ/ǫ0 0
with |E in | < E0 if ǫ > ǫ0 .

• Outside the sphere, the field is equivalent to that of the applied field, together
with that due to a point dipole at the origin, of moment
 
ǫ/ǫ0 − 1  3
p = 4πǫ0  a E0 , (4.12)
2 + ǫ/ǫ0
orientated in the direction of the applied field.

The polarisation P = (ǫ − ǫ0 )E is constant throughout the sphere,

3(ǫ − ǫ0 )
P = E .
2 + ǫ/ǫ0 0
We can evaluate the volume integral of P , to obtain
Z 4 3 3(ǫ − ǫ0 )
dV P = πa E
r<a 3 2 + ǫ/ǫ0 0
 
ǫ/ǫ0 − 1  a3 E ,
= 4πǫ0  0
2 + ǫ/ǫ0
Chapter 2 26

which is just the dipole moment we obtained in eqn. 4.12. Thus the dipole mo-
ment is just the volume integral of the polarisation.

Because P is constant throughout the

sphere, the polarisation charge density E0
−∇ · P vanishes throughout the interior.
However, because of the discontinuity in +
-
P at the surface, we have a surface polar- +
- +
isation charge density, whose magnitude
- +
- +
- +
we can obtain from eqn. 4.7:
- +
- +
σb = P · er (P vanishes outside sphere) - +
- +
 
ǫ/ǫ0 − 1 +
= 3ǫ0   E cos θ
0 -
2 + ǫ/ǫ0 -
Chapter 2 27

4.6 Electrostatic Energy in Dielectric Media

Back in the introduction, we computed the energy of a system of charges in free

space:
1Z 3
W = d x ρ(x)φ(x). (4.13)
2
We obtained this expression by assembling the charges, one-by-one, from infinity
under the potential of the charges already assembled. In the case of dielectrics,
work is done not only in assembling the charges, but also in polarising the medium.
To see how to perform the calculation in this case, consider the change in energy
due to a macroscopic charge density δρ(x),
Z
δW = d3 x δρf (x)φ(x).

We now use recall that ∇ · D = ρf , enabling us to write ∇ · δD = δρf . Thus we

have
Z
δW = d3x ∇ · δDφ(x)
Z
= d3xE · δD,

where we have integrated by parts, assuming that the charge is localised so that
the surface term vanishes. Thus the total energy in constructing the system is
Z Z D
3
W = dx E · δD.
0

We now make the critical assumption of a linear, isotropic constitutive relation

between E and D,
D(x) = ǫ(x)E(x).
Then we have E · δD = 21 δ(E · D), and thus
Z
3
Z D 1
W = dx δ(E · D)
0 2
yielding
1Z 3
W = d x E · D.
2
Chapter 2 28

We can recover our expression eqn. 4.13 either by the substitution E = ∇φ and
using ∇ · D = ρf , or by noting the linear relation between φ and ρ. The crucial
observation is that the expression eqn. 4.13 is valid only if the relation between D
and E is linear.

4.6.1 Energy of Dielectric in an Electric Field with Fixed Charges

As an important application of this formula, we will consider the case of a dielec-

tric medium introduced into an electric field E 0 (x) arising from a fixed charge
distribution ρf = ρ0 (x). Initially, the energy of the system is
1Z 3
W0 = d x E 0 · D0
2
with D0 = ǫ0 E 0 ; here ǫ0 is the initial permittivity of the dielectric, not necessarily
the permittivity of free space.
We now introduce the medium, of volume V1 , with permittivity


 ǫ1 (x) x ∈ V1
ǫ(x) =  ,
 ǫ0 (x) x 6∈ V1
noting that the charge distribution is unaltered. Then the new energy is
1Z 3
W1 = d x E(x) · D(x)
2
and the change in energy is
1Z 3 1Z 3
W = d xE · D − d x E 0 · D0 .
2 2
With a little juggling, we can write this as
1Z 3 1Z 3
W = d x (E · D0 − E 0 · D) + d x (E + E 0 ) · (D − D0 ).
2 2
To evaluate the second term, we note that both ∇ × E = 0 and ∇ × E 0 = 0, and
thus we may write E + E 0 = −∇Φ(x). Hence the second integral may be written
1Z 3 1Z 3
I=− d x ∇Φ · (D − D0 ) = d x Φ∇ · (D − D0 )
2 2
Chapter 2 29

where we assume the integrand falls off sufficiently rapidly at infinity.

Now ∇ · (D − D0 ) = ρf (x) − ρ0f (x) = 0, since we required that the free charge
distribution be unaltered by the introduction of the dielectric. Thus the integral
vanishes, and we have
1Z 3
W = d x (E · D 0 − E 0 · D).
2
We now spilt the region of integration into V1 and the remainder,
1Z 1Z
W = d3x (E · D 0 − E 0 · D) + d3 x (E · D0 − E 0 · D).
2 x∈V1 2 x6 ∈V1

For x 6∈ V1 we have D 0 = ǫ0 E 0 and D = ǫ0E, and the integrand vanishes, so that

1Z 3
W = d x (ǫ0E · E 0 − ǫ1 E 0 · E)
2 V1
1Z 3
= − d x (ǫ1 − ǫ0 )E · E 0.
2 V1
We now specialise to the case where the original dielectric is indeed the vacuum,
and ǫ0 the permittivity of free space, and write

(ǫ1 − ǫ0 )E = P ,

yielding
1Z 3
W =− d x P · E 0.
2 V1
1
We can interpret w = − P · E 0 as the energy density of the dielectric. The
2
expression can be likened to that for the energy of a dipole distribution derived
at the end of Section 4.3. There we were considering a permanent dipole, whilst
here energy is expended in polarizing the dielectric, and this is reflected in the
factor of 1/2.
Note that the energy tends to decrease if the dielectric moves to a region of
increasing E 0 , providing ǫ1 > ǫ0 . Since the charges are held fixed, the total
energy is conserved, and we can interpret the change in field energy W due to
Chapter 2 30

a displacement of the dielectric body ξ as producing a corresponding change in

potential energy, and hence a force on the body of magnitude
∂W
!
F =− ,
∂ξ Q

where the subscript Q denotes at fixed charge.

4.6.2 Energy of Dielectric Body at Fixed Potentials

We will conclude this section by considering the contrasting case where we intro-
duce a dielectric body into a system where the potentials, rather than charges,
are kept fixed. A paradigm is the introduction of a dielectric between the plates
of a capacitor connected to a battery, and hence at a fixed potential difference.

Fixed
Potentials

Dielectric

In this case, charges can flow to or from the conducting plates as the dielectric
is introduced to maintain the potentials, and hence the total energy can change.
Again, we will assume that the media are linear.
It is sufficient to consider small changes to the potential δφ and to the charge
distribution δφ, for which the change in energy δW , from eqn. 4.13, is
1Z 3
δW = d x (ρf δφ + φδρf ).
2
For the case of linear media, these two terms are equal if the dielectric properties
are unaltered. However, in the case where the dielectric properties are altered
during the change, ǫ(x) → ǫ(x) + δǫ(x), this is no longer true, because of a
Chapter 2 31

polarisation charge density generated in the dielectric. We have already considered

this problem for fixed charges, δρf = 0. In order to compute the change of energy
at fixed potentials, we study the problem in two stages;

1. The battery is disconnected, so that the distribution of charges is fixed,

δρf = 0, and the dielectric is introduced. Then there is a change in potential
δφ1 , and the corresponding change in energy is
1Z 3 1Z
δW1 = d x ρf δφ1 = − (ǫ1 − ǫ0 )E · E 0 ,
2 2
using the result of the previous subsection.

2. We now reconnect the battery. The potential on the conductors, where the
only macroscopic charges reside, must regain its original value, i.e. δφ2 =
−δφ1 , and there is a corresponding change in charge density δρ2f , yielding
1Z 3
δW2 = d x (ρf δφ2 + φ2 δρ2f ).
2
In this step, the dielectric properties are unaltered and the two terms are
equal, so we have
Z
δW2 = d3 x ρf δφ2
Z
= − d3x ρf δφ1
= −2 δW1

Thus the total energy change

δW = δW1 + δW2 = −δW1 ,

which we write as
δWV = −δWQ ,
i.e. the change in energy at fixed potential is minus the change in energy at fixed
charges. In this case, if a dielectric with ǫ1 > ǫ0 moves into a region at fixed
Chapter 2 32

potentials, the energy increases, and a mechanical force

∂W
!
Fξ = +
∂ξ V

acts on the body.

Chapter 5

Magnetostatics

5.1 Introduction

The crucial difference between electric and magnetic phenomena is the absence
of isolated magnetic charges, or magnetic monopoles. Here the basic building
blocks are magnetic dipoles. For a magnetic field, or flux density, B, the torque
τ acting on a dipole of moment µ is

τ = µ × B.

The other concept we need in the study of magnetostatics is the electric current
J, defined as the flow of charge per unit time per unit area, with normal in the
direction of J.
dI dQ ˆ
J= = J
da⊥ da⊥ dt

5.1.1 Current Conservation

Current conservation is represented by the continuity equation

∂ρ
+ ∇ · J = 0,
∂t
where ρ is the charge density. This statement just states that the rate of change
of charge in any volume V is (minus) the flux of charge across the surface of V ,

1
Chapter 2 2

as you can see by applying the divergence theorem:

dQ I dQ I I Z
= da⊥ = Jda⊥ = J · da = d3x∇ · J
dt da⊥ dt
For steady currents we are considering in this chapter

∇ · J = 0.

5.2 Biot-Savart Law

P
x
This describes the element of magnetic field B I
at some point x due to an element of current
flow Idl at x′ .
x’
dl

dl × (x − x′ )
dB = kI
|x − x′ |3
where, in SI units,

• I is the current (Ampères),

• dl is an element of length in the direction of the current flow,

• k = µ0 /4π, where µ0 is the permeability of free space.

For a point charge q moving with velocity v, we can replace Idl by qv, and we
have
µ0 qv × (x − x′ )
B= ,
4π |x − x′ |3
providing v is constant, and small compared to the velocity of light.
Chapter 2 3

We can apply the superposition principle to the magnetic field, and obtain for a
general current density
µ0 Z 3 ′ J(x′ ) × (x − x′ )
B(x) = dx .
4π |x − x′ |3

Example

Consider the magnetic field due to straight dl

wire carrying current I. Then the field a dis-
I
tance R from the wire is tangential, and can
be written
µ0 I Z ∞ dl R
B = √ eθ
4π −∞ (l2 + R2) l2 + R2 B
=
µ0 I Z
R
dl
eθ
R
4π (l + R2 )3/2
2

µ0 I Z π/2 R sec2 θdθ µ0 I

= R eθ = eθ
4π −π/2 sec3 θ 2πR

5.2.1 Force on a Current in Presence of Magnetic Field

The element of force on a current element Idl at x in a magnetic field B(x) is

dF = Idl × B.

Thus the force on a closed loop of current l1 due to magnetic field from closed
loop l2 is
 
µ0 I I dl × (x − x ) 
2 1 2
F 12 = I1 I2 dl1 ×  3
4π |x1 − x2 | 

µ0 I I dl × [dl × (x − x )]
1 2 1 2
= I1 I2 .
4π |x1 − x2 |3
Chapter 2 4

We can put this expression in a more symmetric form by writing

dl1 × (dl2 × x12) = (dl1 · x12)dl2 − (dl1 · dl2)x12 ,

yielding  
µ0 I1I2 I I  dl1 · dl2 dl1 · x12 
F 12 = − x + dl . (5.1)
4π  |x12|3 12 2
|x12|3 
We will now show that the second term vanishes. Consider the integration around
loop 1, for fixed x2. Then under a change x1 → x1 + dl1, we have

x12 → x12 + dl1 .

Now consider the change in 1/|x12|:

 
1  1 1
δ = −
|x12 | |x12 + dl1| |x12 |
 
1  x12 · dl1 
= 1 − − 1
|x12|  |x12|2 

x · dl
= − 12 3 1 .
|x12|
Thus the integrand in the second terms of eqn. 5.1 is an exact differential, and
therefore the integrand around the closed loop vanishes, and we have
µ0 I1I2 Z dl1 · dl2
F 12 = x .
4π |x12|3 12
Now Newton’s third law is satisfied explicity, and we have

F 12 = −F 21 .

For a general current density J(x) in a magnetic field B(x), we have

Z
F = d3x J(x) × B(x)
Z
τ = d3x x × (J × B).
Chapter 2 5

5.3 Laws of Magnetostatics in Differential Form

In analogy with electrostatics, our starting point is the expression for B due to
general current density
µ0 Z 3 ′ J(x′ ) × (x − x′ )
B(x) = dx .
4π |x − x′ |3
We begin by recalling that ∇ × (ϕa) = ∇ϕ × a, where a is a constant vector.
Thus
   
′ 1 1  × J(x′ )
∇x × J(x ) ′
 = ∇ 
x ′
|x − x | |x − x |
(x − x′ )
= − ′ 3
× J(x′ )
|x − x |
Thus we can write

µ0 Z 1
B(x) = ∇ × d3 x ′ ′
J(x′ ) (5.2)
4π |x − x |

From eqn. 5.2, we immediately see that

∇ · B = 0. (5.3)

This is another of Maxwell’s equations, and is just another statement that you
cannot have isolated magnetic charges, and that the total flux of B through any
closed surface vanishes Z
dS · B = 0.
S=∂V
To obtain another differential equation, we evaluate ∇ × B. We begin by recalling
the vector identity
∇ × (∇ × A) = ∇(∇ · A) − ∇2 A,
so that
J(x′ )
 
µ0 Z 3 ′ µ0 Z 3 ′ ′ 2 1 
∇×B = ∇ d x ∇x · − d x J(x )∇x .
4π |x − x′ | 4π |x − x′ |
Chapter 2 6

Now
J(x′ )
 
1 
∇x · ′
= −J(x′ ) · ∇x′ 
|x − x | |x − x′ |
 
2 1 
∇x ′
= −4πδ(x − x′ ),
|x − x |
and thus
 
µ0 Z 3 ′ ′ 1  Z
∇ × B = − ∇ d x J(x ) · ∇x′ 
′
+ µ0 d3x′ J(x′)δ(x − x′ )
4π |x − x |
µ0 Z 3 ′ 1
= ∇ dx ∇ ′ · J + µ0 J(x)
4π |x − x′ | x
For, for magnetostatics, we have ∇ · J = 0, and thus

∇ × B = µ0 J (5.4)

This is the second fundamental differential equation. We can apply Stoke’s theo-
rem to a closed curve C spanned by a surface S to obtain
Z Z Z
∇ × B · dS = B · dl = µ0 J · dS.
Chapter 2 7

5.4 Vector Potential

For static fields, the governing equations of magnetostatics are

∇·B = 0
∇ × B = µ0 J

For the case J ≡ 0, we have ∇ × B = 0, and we can introduce a magnetic scalar

potential φM .
Much more interesting is the general case J 6= 0. We can show that if ∇ · B = 0
is a star-shaped region,1 then a vector potential A can be found such that

B = ∇ × A.

In the case where B is the magnetic field, we call A the magnetic vector po-
tential.

5.4.1 Uniqueness of A and Gauge Transformations

If A is a solution of B = ∇ × A, then A′ = A + ∇f , where f is an arbitrary,

continuously differentiable scalar field, is also a solution, because

∇ × (∇f ) = 0.

Transformation of this form are called Gauge Transformations; we say that B

is invariant under gauge transformations. To simplify calculations, we often make
a specific choice of gauge.

Examples

1. We could require A1(x) = 0 ∀x.

1
A star shaped region is one in which there exists a point which can be connected to every other point by a
straight line
Chapter 2 8

2. We could require
∇ · A = 0 ∀x.
This is the Coulomb Gauge.
Choosing, or fixing, the gauge reduces the number of degrees of freedom, clear in
example (1) above. All the fundamental forces of nature are described by Gauge
Theories, having the property of a gauge, or local, symmetry.

5.4.2 Solutions for the Vector Potential in Free Space

We will specify that we work in the Coulomb gauge, ∇ · A = 0. Then the second
of our governing equation becomes
∇ × B = ∇ × (∇ × A) = ∇(∇ · A) − ∇2 A = µ0 J
and thus
∇2A = −µ0 J.
This is just Poisson’s equation, applied to each of the Cartesian components of
A, and from our investigation of electrostatics has the solution
µ0 Z 3 ′ J(x′ )
A(x) = dx . (5.5)
4π |x − x′ |

Example

Potential due to a wire loop of radius a, carrying current I.

z

y
ϕ

I
x
Chapter 2 9

The current is purely in the azimuthal direction, and in spherical polars, we can
write the current density as

′ ′ δ(r′ − a)
Jϕ = I sin θ δ(cos θ ) .
a
You should convince yourself that this expression is correct. W.l.o.g. we will
consider the case where the observation point is in the x − z plane, so that, in
Cartesian coordinates, the current density is

J = −Jϕ sin ϕ′i + Jϕ cos ϕ′ j.

Thus the vector potential, from eqn. 5.5, is given by

µ0 Z 1
A(x) = dΩ′ r′2dr′ {−Jϕ sin ϕ′ i + Jϕ cos ϕ′j} × .
4π |x − x′ |
The x component of A will vanish, since the expansion of 1/|x − x′ | is symmetric
under ϕ′ ↔ −ϕ′ . Thus the only non-vanishing component of A is in the y-
direction, which coincides with eϕ . Thus we have
′
µ0 Z ′ ′ ′2 ′ ′ δ(r − a) 1
Aϕ = I dΩ dr r sin θ δ(cos θ ) cos ϕ′ .
4π a |x − x′ |
Performing the integrations over r′ and θ′ yields
µ0 Ia Z 2π ′
Aϕ = dϕ cos ϕ′ {a2 + r2 − 2ar sin θ cos ϕ′ }−1/2.
4π 0
This is an eliptic integral, and its expression in eliptic functions is not particularly
illuminating. Instead, we will perform an expansion in spherical harmonics:
µ0 I Z δ(r′ − a) iϕ′
Aϕ = ℜ dΩ′ dr′ r′2 sin θ′ δ(cos θ′ ) e
4π a
l
X 1 r< ∗ ′ ′
×4π l+1 Ylm (θ , ϕ )Ylm(θ, ϕ),
l,m 2l + 1 r>

where we write
′
cos ϕ′ = ℜeiϕ .
Chapter 2 10

Performing the delta-function integrations, we arrive at

l
r< 1 Z ′ iϕ′ ∗
e Ylm(π/2, ϕ′).
X
Aϕ = µ0 Iaℜ l+1 Y lm (θ, 0) dϕ
l,m r> 2l + 1
We now use the orthogonality properties of the functions exp imϕ to write (you
see why we expressed cos ϕ′ this way. . . ):

Z
′

2πYl1(π/2, 0) m = 1
dϕ′ eiϕ Ylm
∗
(π/2, ϕ′) = 

,
 0 otherwise
and thus
l ∞
r< X 1
Aϕ = 2πµ0 Ia l+1 Y l1 (θ, 0)Y l1 (π/2, 0) .
l=1 r> 2l + 1
Now we have that
v
u (l − 1)!(2l + 1) 1
u
Yl1(π/2, 0) = Pl (0)
u
t
4π(l + 1)!
which vanishes if l is even, since Pl1 (0) has the opposite parity to Pl (0). The
explicit evaluation of these integrals is performed in Jackson, so I leave it for
you to look them up there. However, the important feature is that even when
we have azimuthal symmetry, the vector potential and magnetic fields involve
the Pl1 Legendre polynomials; this reflects the vector nature of the source in
magnetostatics, as opposed to the scalar nature of the source in electrostatics.

5.5 Magnetic Field Far from Current Distribution

Consider a localized current distribution J(x′ ) , and the magnetic vector potential
produced at a point P (x) where |x| ≫ |x′ |. Then we can write
1 1 x · x′
= + + ...,
|x − x′ | |x| |x|3
so that, in the Coulomb gauge
 
µ0  1 Z 3 ′ ′ x Z 3 ′ ′ ′

Ai (x) = d x J i (x ) + · d x J i (x )x + . . . (5.6)
4π  |x| |x|3 
Chapter 2 11

To study this expression further, we begin by proving a small identity. Recall

that for magnetostatics, ∇ · J ≡ 0, and thus for any two scalar functions f (x′ )
and g(x′ ) we have
Z Z
3 ′ ′ ′ ′
0=− d x f (x )g(x )∇ · J = d3x′ ∇′ [f (x′ )g(x′ )] · J(x′ )

where in the second step we have integrated by parts, using the fact that the
surface integral vanishes for a localised current distribution. Thus we have
Z
d3x′ [f J · ∇′ g + gJ · ∇′ f ] = 0 (5.7)

We now consider the first term in eqn. 5.6. Applying eqn. 5.7 for the case f (x′ ) =
1, g(x′ ) = x′i, we have
Z
d3x′ [Jj δij + x′iJj · 0] = 0
Z
⇒ d3x′ Ji = 0.

Thus the first term vanishes. This is just a further restatement that there is no
“monopole” contribution to the multipole expansion for magnetic fields.
We now applying the identity to the case f = x′i , and g = x′j . Then from eqn. 5.7,
we have
Z
3 ′ ′ ∂x′j ∂x′i
dx [xiJk ′ + x′j Jk ] = 0
∂xk ∂x′k
Z
⇒ 3 ′
d x [x′iJj + x′j Ji ] = 0.

Thus, going back to eqn. 5.6, we may write

µ0 1 Z
Ai (x) = 3
x j d3x′ Ji x′j
4π |x|
1 µ0 1 Z
= − 3
x j d3x′ [x′iJj − x′j Ji ].
2 4π |x|
Chapter 2 12

Levi-Civita Tensor

To take the discussion further, we recall the definition of the Levi-Civita tensor





0 if any two if i, j, k are equal

ǫijk =  1 if (ijk) is an even permutation of (123)


 −1 if (ijk) is an odd permutation of (123)


This tensor is isotropic, and totally anti-symmetric. In particular, we have

A × B|i = ǫijk Aj Bk .

There is the following well-known and easily shown identity

ǫijk ǫilm = δjl δkm − δjm δkl ,

which we will now use to write

x′iJj − x′j Ji = (δil δjm − δimδjl )x′l Jm

= ǫkij ǫklm x′l Jm
= ǫijk (x′ × J)k .

Thus we have

1 µ0 1  Z
3 ′ ′

Ai (x) = − x × d x x × J
2 4π |x|3 i

The vector
1Z 3 ′ ′
m= d x x ×J
2
is the magnetic moment, whilst
1
µ = x′ × J
2
is the magnetic moment density. Thus we can write
Chapter 2 13

µ0 1
A(x) = m×x
4π |x|3
This is the lowest, non-vanishing term in the multipole expansion of the magnetic
vector potential for a localised current density. Applying B = ∇ × A, we have
µ0  3(x · m)x − r2m 
 

B= ,
4π r5
exactly analogous to the electrostatic field due to a point dipole.2

Example

For the case of a current confined to a loop, we have

II
m= x × dl.
2
Furthermore, if we have a planar loop, x × dl is
normal to the plane of the loop, and we have
1 1
x × dl = x dl sin ξ
2 2 da
111
000
000
111
= da 000
111
000
111
000
111
000
111
000
111
000
111
0000000000000
1111111111111
0001111111111
1110000000000
00000000000000000
11111111111111111
0000000000000
1111111111111
000
1110000000000
1111111111
00000000000000000
11111111111111111
0000000000000
1111111111111
000
1110000000000
1111111111
00000000000000000
11111111111111111
0000000000000
1111111111111
0000000000
1111111111
0001111111111
111
00000000000000000
11111111111111111
0000000000000
1111111111111
0000000000dl
so that 000
111
00000000000000000
11111111111111111
0000000000000
1111111111111
0001111111111
1110000000000
00000000000000000
11111111111111111
0000000000000
1111111111111
000
1110000000000
1111111111
00000000000000000
11111111111111111
0000000000000
1111111111111
0000000000
1111111111
00000000000000000
11111111111111111
0000000000000
1111111111111
0000000000
1111111111
00000000000000000
11111111111111111
0000000000000
1111111111111
0000000000
1111111111
00000000000000000
11111111111111111
0000000000000
1111111111111
0000000000
1111111111
00000000000000000
11111111111111111
m = IAn 0000000000000
1111111111111
0000000000
1111111111
1111111
0000000
00000000000000000
11111111111111111
00000000000000000
11111111111111111
00000000000000000
11111111111111111
00000000000000000
11111111111111111
ξ
00000000000000000
11111111111111111
00000000000000000
11111111111111111
00000000000000000
11111111111111111
00000000000000000
11111111111111111
00000000000000000
11111111111111111
where n is a normal to the plane of the loop,
and A is the total area of the loop.

Example

We conclude this section by considering the case where the current distribution
arises from the motion of a number of charged point-like particles:
X
J= qi vi δ(x − xi ),
i
2
It is possible to introduce a vector potential to describe electric dipole fields
Chapter 2 14

where vi is the velocity of the ith particle, which we assume is much less than the
velocity of light.
Then we have
1X
m= qi x i × v i .
2 i
Now the orbital angular momentum of a particle is given by

Li = Mi xi × v i ,

where Mi is the mass of the ith particle. Thus we may write

X qi
m= L.
i 2Mi i
In the case where all the particles have equal mass, we see that the magnetic
moment is proportional to the total angular momentum.
Chapter 2 15

5.6 Magnetostatics of Matter

5.6.1 Torques and forces on magnetic dipoles

First, consider a magnetic dipole in the uniform magnetic field B. Let us visualize
magnetic dipole m as a wire loop with area a carrying current I such as m = Ia.
The total force acting on the loop is zero:
I I
F = I dl × B = − IB × dl = 0

The torque acting on the loop is m × B:

I I I I I
2
N = x′ × dF = x′ × (Idx′ × B) = I dx′ (x′ · B) − BI dx′ = I dx′ (x′ · B)

It is easy to prove that for an arbitrary constant vector a

I
1 I
dx′ (x′ · a) = − a × (x′ × dx′ ) (5.8)
2
Indeed,
I I
a × (x × dx ) = [x′ (a · dx′ ) − dx′ (a · x′ )]
′ ′
I I I
x′ (a · dx′ ) = [d(x′ (a · x′ )) − dx′ (a · x′ )] = − dx′ (a · x′ )

and therefore a × (x′ × dx′ ) = −2 dx′ (a · x′ ). Taking a = B we get

H H

I I
′ II ′
′
N = − B × (x × dx ) = ( x × dx′ ) × B = m × B
2 2
so the torque in a uniform external field is a cross product of the magnetic moment
and the field.
Let us now consider a small dipole in the non-uniform external field (the size of
the dipole ≪ characteristic size of the field). The formula for the torque remains
the same: N = m × B where the magnetic field should be taken at the position
of the dipole. However, the total force is no longer zero.
I
F = I dl × B 6= 0
Chapter 2 16

Since our dipole is small we can expand B(x′ ) in powers of x′ . For simplicity,
suppose that the dipole is located at the origin. We get

B(x′ ) = B(0) + (x′ · ∇)B(0) + ...

and therefore (dx′ ≡ dl′ )

I I I
′2
F = I dl × B(0) + I dl × (x · ∇)B + O(x ) = I dx′ (x′ · ∇) × B
′ ′ ′

Next we use formula (5.8) with a = ∇ and obtain

I
′ II ′
′
I dx (x · ∇) = (x × dx′ ) × ∇ = m × ∇
2
so finally

F = (m × ∇) × B = ∇(m · B) − m(∇ · B) = ∇(m · B)

because ∇ · B = 0.
Since F = −∇U we see that the potential energy of a (small) magnetic dipole in
the external magnetic field is

U = −m·B

(similarly to U = −p · E for the electric dipole).

5.6.2 Maxwell equations in matter

We could, in principle, attempt to describe the magnetostatics of a material in

terms of the microscopic, or “vacuum”, fields. As in the case of electrostatics, this
approach is neither feasible nor desirable. At the microscopic level, the individual
atoms have magnetic moments and eddy currents are generated that we cannot
account for exactly. Rather, we discuss macroscopic quantities, including that part
of the magnetic field arising from these microscopic currents. In the following, we
will use the subscript micro to denote microscopic properties, with the remaining
variables denoting macroscopic quantities.
Chapter 2 17

At the microscopic level, we have ∇ · B micro = 0. We can average this to obtain

∇·B =0
and hence we know that we can write the macroscopic magnetic field in terms of
a vector potential
B = ∇ × A.
Suppose now that we have a collection of atoms of various types i, with magneti-
sation mi . Then the macroscopic magnetisation
X
M= Ni hmi i,
i
where Ni is the numer of atoms of type i/unit volume, and hmii is their aver-
age magnetic moment. Note the M is analogous to the polarisation density of
electrostatics.
We will now consider the contribution to the vector potential at x due to an
infinitessimal volume ∆V at x′ . There are two contributions
µ0 J f (x′ ) µ0 1
∆A(x) = ′
∆V + ′
M × (x − x′ )∆V,
4π | x − x | 4π | x − x |
where the first term arises from the “free” macroscopic current densities and the
second is due to the macroscopic magnetisation described above. We now sum
over the volume elements ∆V and get
µ0 Z 3 ′ J f (x′ ) µ0 Z 3 ′ 1
A(x) = dx ′
+ d x ′
M × (x − x′ ).
4π | x − x | 4π |x−x |
There is a way to rewrite the second term in a more illuminating way. First, note
that
′
 
3 ′ M × (x − x ) 1
Z Z
3 ′ ′
dx = d x M × ∇ 
| x − x′ |3 | x − x′ |
From the formula ∇ × (f M) = f (∇ × M) − M × (∇f ) (see the cover of Jackson)
for f = |x−1x′ | we obtain

M (x′ ) 
 
Z
3 ′ ′ ′ 1 Z
3 ′ ′
d x (∇ × M (x ) + dx ∇ ×  ,
| x − x′ | | x − x′ |
Chapter 2 18

Using the divergence theorem for vector fields (again, see the cover)
Z Z
3 ′ ′ ′
d x (∇ × A(x ) = n × A dS (5.9)
V S=∂V

the second term can be rewritten as a surface integral

µ0 Z M (x′ ) × n
dS
4π S=∂V |x − x′ |
Finally, we get
Z M × (x − x′ )
3 ′ µ0 Z 3 ′ Jb µ0 Z K b (x′ )
dx = dx + dS
| x − x′ |3 4π | x − x′ | 4π S=∂V |x − x′ |
where Jb ≡ ∇ × M is called a bound volume current density and Kb ≡ M × n a
bound surface current density.
If we take the surface to be an infinitely large sphere and assume that ∼ K
vanishes at infinity ,we get
µ0 Z dV ′ n ′ ′ ′
o
A(x) = J (x ) + ∇ × M .
4π | x − x′ | f
Comparing with the fundamental equation of magnetostatics in vacua,

∇ × B = µ0 J f ,

we have
∇ × B = µ0 {J f + ∇ × M }.
It is now conventional to introduce the magnetic field H, where
1
H = B − M.
µ0
In the context of media, the field B is known as the magnetic induction or
magnetic flux density. In terms of H and B, the fundamental equations of
magnetostatics in matter are

∇·B = 0
∇ × H = Jf
Chapter 2 19

Note that H is analogous to D in electrostatics; E and B are the fundamental

fields, whilst H and D depend on the medium.

5.6.3 Constitutive relation

In the case of (isotropic) diamagnetic and paramagnetic materials, where the

magnetic moment arises solely from the applied magnetic field, there is a simple
linear relation between H and B

M = χm H,

where χm is the magnetic susceptibility. Then we may write

1
H= (B − µ0 χm H)
µ0
yielding
B = µH
where µ ≡ µ0 (1 + χm ) is the magnetic permeability.
For ferromagnets, the corresponding relation is non-linear and exhibits hysterisis,
i.e. the material retains a memory of its preparation.
B(M)
Remnant
magnetisation

B = F(H)
Chapter 2 20

5.6.4 Boundary Conditions at Surface Between Media

We will now obtain boundary conditions for the normal and tangential compo-
nents of the field at the boundary between two materials. Note that the followign
discussion is independent of whether or not there is a linear relation between the
H and B.

δΑ
n

Normal Condition

Apply Gauss’ Law to the pillbox shown

Z Z
0= dV ∇ · B = B · n dS = (B 2 − B 1 ) · n δA

where n is a unit normal from medium 1 to medium 2, and δA is the surface area
of the pillbox. Thus we have

B⊥ ⊥
1 = B2
Chapter 2 21

Tangential Condition

We now apply Stoke’s theorem to get the boundary conditions on the tangential
components: I Z Z
H · dl = (∇ × H) · da = J f · da,
c S S

where S is a surface spanning C.

Thus we have the tangential boundary condition

H 2 −H 1 = K ×n ⇒ n×(H 2 −H 1) = K

where K is the surface current density.

5.7 Methods of Solving Boundary Value Problems

We will now look at various methods of solving boundary value problems be-
tween different media. The method depends on nature of the constitutive relation
between B and H, and on whether there is non-zero current density.

5.7.1 Vector Potential

The magnetic field is always solenoidal, and therefore we can essentially always
introduce a vector potential A such that B = ∇ × A.
The dynamical information for the magnetostatics of media is provided by the
equation
∇ × H = Jf.
We will now specialise to the case where we have a linear constitutive relation,
B = µH, enabling us to write
∇×A
" #
∇× = Jf.
µ
Chapter 2 22

This can be written

∇2 A − ∇[∇ · A] = −µJ f ,
which in Coulomb gauge (∇ · A = 0) becomes

∇2 A = −µJ f .

This is analogous to the case discussed in Section 5.4.2, and the solution is that
of eqn. 5.5, with µ0 replaced by µ.

5.7.2 Solution when Jf ≡ 0

In this case we have ∇ × H = 0, and therefore we may admit introduce a scalar

potential φM such that
H = −∇φM .
Once again, we will consider linear media, so that B = µH. Then we find that
the scalar potential satisfies Laplace’s equation

∇2φM = 0,

where we assume that µ is piecewise constant, i.e has a constant value in each of
the different media we are considering.

5.7.3 Hard Ferromagnetic

In the case of a hard ferromagnet, we have J f ≡ 0, and the magnetisation is non-

zero, and essentially independent of the magnetic field H providing it is sufficiently
small.
Chapter 2 23

M constant

Since J f ≡ 0, we can solve this problem using either a scalar or a vector potential.

Solution using Scalar Potential

The governing equations are

∇·B = 0 (5.10)
∇×H = 0 (5.11)
1
H = B−M (5.12)
µ0
We will introduce a scalar potential for the magnetic field,

H = −∇φM .

Then from eqns. (5.10) and (5.12), we have

∇2φM = −ρM ,

where
ρM = −∇ · M.
In the case where there are no boundaries, this equation has the solution
1 Z 3 ′ ρM
φM = dx
4π |x − x′ |
Chapter 2 24
 
1 Z 3 ′ ′ 1 
= − dx∇ ′
M (x′ ) (integration by parts)
4π |x − x |
1 Z
1
= − ∇ · d3 x ′ ′
M (x′ ).
4π |x − x |
Note that if we are far away from a non-zero M , i.e. f ≫ r′ , then we have
1 1 Z 3 ′ 1
!
φM ≃− ∇ · d x M (x′ ) = m · x.
4π r 4πr3
where Z
m= d3x′ M (x′ ).
Suppose now that we had a hard ferrormagnet confined to a volume V , with
surface S. Then there is a contribution arising from the discontinuity in M at the
surface, which we can express as a surface magnetisation density,

σM = n · M ,

and apply Gauss’ Law to obtain its contribution

1 Z 3 ′ ∇′ · M(x′ ) 1 I σM
φM =− dx + dS . (5.13)
4π V |x − x′ | 4π S |x − x′ |
Note that for a uniform magnetisation, the bulk volume integral vanishes, and
the only contribution arises from the surface term.

Solution using Vector Potential

We now write B = ∇ × A, so that we have

1
H= ∇ × A − M.
µ0
Thus eqn. (5.11) becomes
1
0= ∇×H = ∇ × (∇ × A) − ∇ × M .
µ0
Chapter 2 25

Introducing an effective magnetisation current

JM = ∇ × M,
we have, in Coulomb gauge,
∇2A = −µ0 J M .
Thus again each component of A satisfies Poisson’s equation, with solution
µ0 Z 3 ′ J M
A= dx .
4π |x − x′ |
In the case where there is a sharp boundary between two media, we again have a
surface contribution which we treat as for the case of a scalar potential, yielding
µ0 Z 3 ′ ∇′ × M µ0 I M(x′ ) × n′
A(x) = dx + dS .
4π V |x − x′ | 4π S |x − x′ |

Example: uniformly magnetised sphere in a vacuum

ez
M

Consider a sphere of radius a, with uniform magnetisation M = M0 ez . We will

consider the solution using a scalar potential.
Since the magnetisation is constant throughout the body of the sphere, only the
surface integral contributes in eqn. (5.13), and we have
′ ′
1 I ′ n · M (x )
φM = dS
4π S |x − x′ |
M0 a2 Z cos θ′
= dΩ′ .
4π |x − x′ |
Chapter 2 26

To proceed further, we expand |x−1x′ | in terms of spherical harmonics

∞ X l l
1 X 1 r< ∗ ′ ′
′
= 4π l+1 Ylm (θ , ϕ )Ylm (θ, ϕ).
|x − x | l=0 m=−l 2l + 1 r>
√
Noting that cos θ′ = P1 (cos θ′ ) = 4πY10(θ′, ϕ′ ), and using orthogonality, we can
write
1 r<
φM (r, θ) = M0 a2 2 cos θ,
3 r>
where r<{>} = min{max}(r, a).
Inside the sphere, we have r< = r and r> = a. Thus
1 1
φM = M0 r cos θ = M0 z.
3 3
Thus we have 
Hin = −∇φM = − 13 M 

,
Bin = µ0 (H + M ) = 32 µ0 M 


and we have that H (B) is anti-parallel (parallel) to M .

Outside the sphere,
1 a3
φM = M0 2 cos θ.
3 r
Since M is uniform inside the sphere, we can associate this with the potential due
to a magnetic dipole of moment
4πa3
m= M.
3
Both the magnetic induction and the magnetic field are parallel to the magneti-
sation
2 a3 1
Bout = µ0 Hout = −µ0 ∇φM = M0 µ0 3 (cos θer + sin θeθ )
3 r 2

Sphere in External Field

Suppose now we add a uniform magnetic induction B0 = µ0 H0 . Then by the

principle of linear superposition, the resulting field inside the sphere is just the
Chapter 2 27

sum of the two solutions

2µ0
Bin = B0 + M (5.14)
3
1 1
Hin = B0 − M (5.15)
µ0 3
Suppose now that the field is not permanently magnetised, but rather has a linear
relation between B and H,
Bin = µHin
Then M is also linearly related, and from eqns. (5.14) and (5.15) we have
3 µ − µ0
!
M= B0 .
µ0 µ + 2µ0
For the case of ferromagnets described earlier, we do not have such a linear rela-
tion; indeed we have non-zero M for zero applied magnetic field. We can obtain
one relation between Bin and Hin by eliminating M in eqns. (5.14) and (5.15),
whilst obtaining another from the hysterisis curve.

Example: spherical shell in uniform field

Consider a shell of permeability µ in a vacuum, as shown below.

B0 = µ0 H0
a

b
Chapter 2 28

Since the current density is zero, we can once again write H = −∇φM . Further-
more, B = µH, and thus ∇ · H = 0 so that the scalar potential satisfies

∇2φM = 0,

subject to the boundary conditions at r = a and r = b. We are now experts at

writing down the solution in terms of Legendre polynomials.
∞
X αl
φM = −H0 r cos θ + P (cos θ)
l+1 l
r>b
l=0 r
∞ " γl
#
X l
φM = βl r + l+1 Pl (cos θ) a<r<b
l=0 r
∞
δl rl Pl (cos θ)
X
φM = r<a
l=0

where we have imposed that there be a uniform field at infinity for the case r > b,
and that the solution is regular as r → 0.
We now impose the boundary conditions at the interfaces r = a and r = b

B ⊥ is continuous
H k is continuous

which become:
∂φM ∂φM
(b+) = (b−)
∂θ ∂θ
∂φM ∂φM
µ0 (b+) = µ (b−)
∂r ∂r
∂φM ∂φM
(a+) = (a− )
∂θ ∂θ
∂φM ∂φM
µ0 (a−) = µ (a+)
∂r ∂r
We now use these equations to determine the coefficients αl , βl , γl , noting that
∂
Pl (cos θ) = Pl1 (cos θ).
∂θ
Chapter 2 29

All the coefficients vanish for l > 1 (exercise), and we have (see Jackson)
(2µ′ + 1)(µ′ − 1)
 

α1 =  (b3 − a3 )H
0

2a 3
(2µ′ + 1)(µ′ + 2) − b3 (µ′ − 1)2
′
 
9µ
δ1 = − ′ ′ 2a3 ′ 2
H ,
0 (5.16)
(2µ + 1)(µ + 2) − b3 (µ − 1)
where µ′ = µ/µ0 .
For r > b, we have the uniform field together with a dipole of moment α1 , parallel
to H1:
α1
φM = −H0 r cos θ + 2 cos θ.
r
For r < a, there is a uniform magnetic field parallel to H0, of magnitude −δ1 :

φM = −(−δ1)r cos θ.

From eqn. (5.16), we see that δ1 ≃ 1/µ′ as µ′ → ∞: the effect of a shell of high
permeability is to shield the interior from the magnetic field.
Chapter 6

Time-dependent Phenomena and

Maxwell’s Equations

So far we have studied static (time-independent) behaviour of electric and mag-

netic fields. The governing equations are

∇·D = ρ
∇×E = 0 (6.1)

and

∇×H = J
∇ · B = 0. (6.2)

Electric and magnetic phenomena are completely separate, except for the fact
that current density is associated with the motion of charges.

6.1 Faraday’s Law of Magnetic Induction

Faraday (1831) observed that a current could be induced in a closed loop of wire
by varying the flux of magnetic field through a surface spanning the loop.

1
Chapter 2 2

B C

We define the flux φ of the magnetic field through the loop by

Z
φ= B · dS,
S

where S is any surface spanning C.

N.B. Since ∇ · B = 0, φ is independent of the precise surface.
The electromotive force, or voltage, across the curve C is
I
E= E · dl.
C

Then Faraday’s law, in integral form, may be written

dφ
E = −k ,
dt

where, in SI units, k = 1. Note that the sign here is a consequence of Lenz’s law:
the induced current is in such a direction as to oppose the change of flux producing
it. You could argue that the whole application of electricity in the modern world
dates rests on Faraday’s law; the observation that a changing magnetic field can
produce an electric current.
Chapter 2 3

We can can generalise this integral equation as applying to any closed curve in
space, spanned by a surface,
I
d Z
E · dl = − B · dS.
C dt S
We now apply Stokes’ theorem to the l.h.s.,
I Z
E · dl = (∇ × E) · dS.
S

Specialising to the case where both C and S are fixed in time, we have
d Z Z
∂B
B · dS = · dS,
dt S S ∂t

and thus Z Z ∂B
(∇ × E) · dS = − · dS,
S S ∂t
yielding
Z
∂B
(∇ × E + ) · dS = 0.
S ∂t
Since both C, S are arbitrary, we obtain the differential form of Faradays law,

∂B
∇×E+ =0
∂t

This equation replaces the second equation in eqn. (6.1).

The eqns. (6.1) and (6.2) reveal an immediate inconsistency when applied to time-
dependent phenomena. Let us apply the divergence theorem to the first equation
in eqn. (6.2),
∇ · (∇ × H) = ∇ · J.
The l.h.s. is identically zero, whilst the r.h.s. vanishes only for time-independent
problems; in general, we have the continuity equation
∂ρ
∇·J =− .
∂t
Chapter 2 4

To see how to resolve this inconsistency, let us return to Coulomb’s law

∇ · D = ρ,

and substitute into the continuity equation, to obtain

∂D
∇·J +∇· = 0.
∂t
We can make Ampere’s law (∇ × H = J) consistent with the continuity equation
simply by modifying through the substitution
∂D
J →J+
∂t
giving
∂D
∇×H =J + .
∂t

6.2 Maxwell’s Equations

With this final modification of Ampere’s law, and Faraday’s law, we have the
completed the construction of Maxwell’s equations

∇ · D = ρ (ME1) Coulomb’s Law

∂B
∇×E+ = 0 (ME2) Faraday’s Law
∂t
∂D
∇×H = J + (ME3) Ampere’s Law + Maxwell
∂t
∇ · B = 0 (ME4)

The unification of electrical and magnetic phenomena through these equations

represents the crowning achievement of classical, 19th. century physics. The ad-
dition of the electric displacement to the r.h.s. of Ampere’s law was essential to
showing that the solutions admit wave propagation at the speed of light.
Chapter 2 5

6.3 Vector and Scalar Potentials

Maxwell’s equations comprise a set of coupled, first-order PDE’s. In particularly

simple cses, they can be solved directly, but in the case of both electrostatics
and magnetostatics we have seen the efficacy of introducing vector and scalar
potentials. We will now do likewise for the time-dependent case.
We introduce potentials so that the two homogenous equations (Faraday’s law
and the solenoidal condition) are satisfied automatically. Since

∇·B =0

we have seen we can introduce a vector potential A such that

B = ∇ × A.

Substituting into Faraday’s law (ME2), we obtain

∂
∇×E+ [∇ × A] = 0
∂t
∂A
" #
=⇒ ∇ × E + = 0.
∂t
We can now introduce a scalar potential φ such that
∂A
E+ = −∇φ.
∂t
Thus the electric and magnetic fields can be written

B = ∇×A (6.3)
∂A
E = −∇φ − (6.4)
∂t

and ME2 and ME4 are automatically satisfied.

Chapter 2 6

The two remaining equations (ME1 and ME3) determine the dynamical be-
haviour, i.e. the dependence of A and φ on t and x. To solve them, we need
some constitutive relation between (D, H) and (E, B). We will initially restrict
ourselves to the case of the vacuum, where we have

D = ǫ0 E
1
H = B.
µ0
Coulomb’s law, ME1, is thus
∇ · E = ρ/ǫ0
whilst Ampère’s law, ME3, is
1 ∂E
∇ × B = J + ǫ0 .
µ0 ∂t
Thus, in terms of the potential (φ, A), ME1 becomes

∂
∇2 φ + (∇ · A) = −ρ/ǫ0 (6.5)
∂t

Substituting for the potential in ME3, we have

∂φ ∂ 2A 
 
1 
∇ × (∇ × A) = J + ǫ0 −∇ − 2
µ0 ∂t ∂t
2 
 
∂φ ∂ A
=⇒ ∇[∇ · A] − ∇2A = µ0 J + µ0 ǫ0 −∇

− 2 .
∂t ∂t
We now write ǫ0 µ0 = 1/c2 (we of course all no what c will be!), and write

1 ∂ 2A 1 ∂φ
" #
2
∇ A− 2 2 −∇ ∇·A+ 2 = −µ0 J (6.6)
c ∂t c ∂t

Thus we have derived two, coupled second-order PDE’s that are, with the def-
initions of the potentials in eqns (6.3) and (6.4), equivalent to the original four
Maxwell equations.
Chapter 2 7

6.4 Gauge Transformations Revisited

Is it possible to decouple these two equations? One way to do this is through

a clever choice of gauge transformation. A gauge transformation exploits the
redundant degrees of freedom in the problem to simplify the problem.
Recall that the physical fields are not (A, φ), but rather (B, E). A gauge transfor-
mation is a transformation of the (A, φ) that leaves the physics unaltered. In this
section, we will derive gauge transformations for the complete Maxwell equations.
We have already encoutered gauge transformations in the context of magnetostat-
ics; the substition
A −→ A′ = A + ∇Λ
leaves B = ∇ × A invariant. In this case, however, E also depends on A, and the
above transformation will change E unless we make a suitable change φ −→ φ′ .
In terms of the transformed potentials (A′ , φ′ ), we have
∂A′
′
E = −∇φ −
∂t
∂
= ∇φ′ − [A + ∇Λ] .
∂t
But we have
∂A
E = ∇φ − ,
∂t
and thus equating the two expressions gives
∂Λ
" #
′
∇ φ + −φ = 0
∂t
∂Λ
=⇒ φ′ = φ −
∂t
where we have noted that the potential is only defined up to an additive constant.
Thus the gauge transformation of Maxwell’s equations takes the form
Chapter 2 8

A −→ A′ = A + ∇Λ (6.7)
∂Λ
φ −→ φ′ = φ − (6.8)
∂t

We will now discuss some particular choice of gauges.

6.4.1 Lorentz Condition

Suppose we can find a gauge transformation such that

1 ∂φ
∇·A+ = 0. (6.9)
c2 ∂t

This is known as the Lorentz condition, and the dynamical equations assume
the form
2 1 ∂ 2φ
∇ φ − 2 2 = −ρ/ǫ0 (6.10)
c ∂t
2 1 ∂ 2A
∇ A − 2 2 = −µ0 J. (6.11)
c ∂t
The A and φ fields have become decoupled, and the simplified equations are just
the wave equations, with a inhomogeneous source. But is it actually possible
to find a gauge transformation that satisfies eqn. (6.9)?
Let (A, φ) be potentials satisfying eqns. (6.6) and (6.5), and let Λ be a gauge
transformation such that the transformed fields satisfy eqn. (6.9). Then we have
1 ∂φ′
′
∇·A + 2 = 0
c ∂t
1  ∂φ ∂ 2Λ 

2
=⇒ ∇ · A + ∇ Λ + 2 − 2 = 0
c ∂t ∂t
Thus we need to find Λ satisfying
1 ∂ 2Λ 1 ∂φ
" #
2
∇ Λ− 2 2 =− ∇·A+ 2 . (6.12)
c ∂t c ∂t
Chapter 2 9

Note that the Lorentz condition does not specify a gauge uniquely. Let (A, φ)
satisfy the Lorentz condition. Now consider the transformation

A −→ A′ = A + ∇Λ
∂Λ
φ −→ φ′ = φ − .
∂t
Then the Lorentz condition transforms as
1 ∂φ ′ 1 ∂φ′ 2 1 ∂ 2Λ
∇·A+ 2 −→ ∇ · A + 2 =∇ Λ− 2 2
c ∂t c ∂t c ∂t
Thus the new gauge also satisfies the Lorentz condition, providing
1 ∂ 2Λ
∇2 Λ − = 0.
c2 ∂t2
The Lorentz gauge is important because:

• The wave equation is manifest explicitly,

• (A, φ) are treated on an equal footing and, when we discuss Special Relativity,
we will see that the Lorentz condition is Lorentz covariant, i.e. independent
of the choice of coordinate system.

6.4.2 Coulomb Gauge

We have introduced this gauge,

∇·A
in the discussion of magnetostatics. It is not manifestly Lorentz covariant, but
has the property that the scalar potential satisfies Poisson’s equation (Coulomb’s
law!),
∇2φ = −ρ/ǫ0 ,
with solution
1 Z 3 ′ ρ(x′ , t)
φ(x, t) = dx . (6.13)
4πǫ0 |x − x′ |
Chapter 2 10

The vector potential satisfies the inhomogeneous wave equation

2 1 ∂ 2A 1 ∂φ
∇ A − 2 2 = −µ0 J + 2 ∇ . (6.14)
c ∂t c ∂t
Note that the scalar potential φ(x, t) is the instantaneous Coulomb potential
due to a charge density ρ(x, t), i.e. we do not take account of “causality” through
the use of a retarded potential.
The equation for the vector potential contains a gradient operator, ∇∂φ/∂t arising
from the solution of Poisson’s equation for the scalar potential, and this term is
irrotational,
∂φ
" #
∇× ∇ = 0.
∂t
It would be useful to completely decouple the equations governing the vector and
scalar potentials, as in the case of the Lorentz gauge. To accomplish this, we will
separate the current into an irrotational, or longitudinal, piece and a solenoidal,
or transverse, piece,
J = Jl + Jt (6.15)
with

∇ × Jl = 0
∇ · Jt = 0.

We can always perform this separation, as will now be demonstrated. At first, we

do it for the Fourier transforms

J(k) = d3xe−ik·x J(x)

Z
(6.16)
ki kj ki kj
J(k) = J t (k) + J l (k), Jit (k) = (δij − )Jj (k), Jil (k) = Jj (k)
k2 k2
Going back to the coordinate space we obtain
Z d3 k ik ·x ki kj Z 3 −ik·y
Jit(x) = 3
e (δ ij − 2 ) d yJ(y)e
8π k
Chapter 2 11

d3k ik·(x−y ) ki kj ∂ ∂ Z d3 k
eik·(x−y )
Z Z Z
3 2 3
= d yJ(y) 3
e (δij − 2 ) = ( − δij ∇ ) d yJ(y) 3
8π k ∂xi ∂xj 8π
∂ ∂ Z J(y)
=( − δij ∇2 ) d3 y
∂xi ∂xj 4π|x − y|
and similarly
∂ ∂ Z J(y)
Jil (x) =− d3 y
∂xi ∂xj 4π|x − y|
Using the formula  
1 
∇2  = −4πδ(x − y)
|x − y|
it is easy to check the self-consistency Jit (x) + Jil (x) = Ji (x).
Thus we have performed the decomposition of eqn. (6.15) with
1 Z 3 ′ ∇′ · J(x′ )
Jl = − ∇ d x (6.17)
4π |x − x′ |
′
 
1 Z J(x )
Jt = ∇ × ∇ × d3 x′  (6.18)
4π |x − x′ |
Now, from the continuity equation, we have
∂ρ
∇ · Jl + = 0.
∂t
and substituting in eqn. (6.17) we obtain
1 Z 3 ′ 1 ∂ρ
Jl = ∇ dx .
4π |x − x′ | ∂t
We now identify the r.h.s. of this equation with our expression for the scalar
potential of eqn. (6.13) and observe that
∂φ
Jl = ǫ0 ∇
∂t
1 ∂φ
=⇒ µ0 Jl = 2 ∇ ,
c ∂t
Chapter 2 12

where we have used µ0 ǫ0 = 1/c2. Thus, returning to the equation for the vector
potential, eqn. (6.14), we find
1 ∂ 2A
∇2 A − = −µ0 Jt . (6.19)
c2 ∂t2
Only the transverse part of the current is a source for A. Thus this gauge is also
known as the transverse or radiative gauge, and once again we have decoupled
the scalar and vector potentials.
Chapter 2 13

6.5 Green Function for the Wave Equation

In both the Lorentz and Coulomb gauges, we have reduced the problem of finding
the potentials to the solution of the wave equation
1 ∂ 2ψ
∇2 ψ − = −4πf (x, t), (6.20)
c2 ∂t2
where f is some known source, and c, as we have intimated earlier, is the velocity
of wave propagation.
Such a hyperbolic equation, like the elliptic equations encountered in electrostat-
ics, can be solved by means of Green functions. In particular, we will find the
Green function G(x, t; x′ , t′) satsifying
1 ∂2 
 
∇2 − G(x, t; x′ , t′ ) = −4πδ(x − x′ )δ(t − t′ ). (6.21)
2
c ∂t2

The solution to the inhomogeneous wave equation, eqn. (6.20), for a general source
is then Z
ψ(x, t) = ψ0(x, t) + d3x′ dt′ G(x, t; x′ , t′)f (x′ , t′ )
where ψ0 is a solution of the homogeneous equation. Note that this is essen-
tially an initial-value problem, rather than the boundary-value problem encoun-
tered with elliptic equations.
To obtain the Green function, we take the Fourier transform with respect to t:
1 Z
G(x, t; x′ , t′) = dωe−iωtg(x, ω; x′ , t′ )
Z2π
′ ′
g(x, ω; x , t ) = dt eiωt G(x, t; x′ , t′ )

Then taking the F.T. of eqn. (6.21), we find

ω2 
 
∇2 + ′ ′ ′ iωt′
g(x, ω; x , t ) = −4πδ(x − x )e .
c2
We now introduce the spatial Fourier transform,

d3 xe−iq·xg(x, ω; x′ , t′ ),
Z
′ ′
g̃(q, ω; x , t ) =
Chapter 2 14

yielding

(−q 2 + k 2)g̃(q, ω; x′ , t′) = −4πe−iq ·x eiωt

′ ′

e−iq ·x eiωt
′
′ ′
=⇒ g̃(q, ω; x , t ) = 4π 2
q − k2
where k ≡ ω/c is the wave number. We can invert this expression to obtain

4π iωt′ Z 3 eiq ·(x−x )

′
′ ′
g(x, ω; x , t ) = e dq 2 .
(2π)3 q − k2
In order to exhibit the behaviour of this integral, we consider a oordinate system
in which the z-axis is aligned with x − x′ , and let θ be the angle between q and
x − x′ . Thus

eiq|x−x | cos θ
′
′ ′ 4π iωt′ Z ∞ 2 2π
Z Z 1
g(x, ω; x , t ) = e dq q dψ d(cos θ) 2
(2π)3 0 0 −1 q − k2
q 2  eiq|x−x | e−iq|x−x | 
 ′ ′ 
4π iωt′ Z ∞
= e dq 2 −
(2π)2 0 q − k 2  iq|x − x′| iq|x − x′ | 
′
4π eiωt Z ∞ dq q iq|x−x′ |
= e
(2π)2 i|x − x′ | −∞ q 2 − k 2
The integrand has poles at q = ±k, and therefore we have to specify how to treat
the poles in order to evaluate the integrals. We will do this by displacing the poles
off the real axis as follows:
(±) ′ ′ 4π eiωt Z ∞ dq q iq|x−x′ |
g (x, ω; x , t ) = e ,
(2π)2 i|x − x′ | −∞ q 2 − k 2 ∓ iη
where η is small. We now write

q 2 − k 2 ∓ iη = (q − k ∓ iǫ)(q + k ± iǫ),

with ǫ = η/2k (η, k > 0).

We first consider the case of g (+) , which has a pole in the upper half plane at
q = k + iǫ, and in the lower half plane at q = −k − iǫ.
Chapter 2 15

k+i ε

-k-i ε
We can complete the contour in the upper-half plane, where the contribution from
the semi-circle at infinity vanishes, and obtain
1
eiωt +ik|x−x | .
′ ′
g (+) (x, ω; x′ , t′) = ′
|x − x |
Similarly, in the case of g (−) , we have a pole in the upper half plane at q = −k +iǫ,
and performing the contour integration we obtain,
1 iωt′ ±ik|x−x′ |
g (±) (x, ω; x′ , t′) = e .
|x − x′ |
We now invert the temporal Fourier Transform
(±) ′ 1 Z
′ 1 iωt′ ±k|x−x′ |
G (x, t; x , t ) = dω e−iωt e .
2π |x − x′ |
The ω integration is straightforward, and we find

1 1
" #
(±) ′ ′ ′
G (x, t; x , t ) = ′
δ (t − t) ± |x − x′ | (6.22)
|x − x | c

The Green function G(+) is known as the retarded Green function, because a
change at time t arises from an effect at an earlier time
1
t′ = t − |x − x′ |.
c
Chapter 2 16

It manifestly exhibits causality. G(−) is known as the advanced Green function.

We now construct the complete solutions as follows:

1. Retarded Solution. We imagine that, as t −→ −∞, we have a wave ψin (x, t)

satisfying the homogeneous equation. The source f (x, t) then turns on, and
the complete solution is
Z
ψ(x, t) = ψin (x, t) + d3 x′ dt G(+) (x, t; x′ , t′ )f (x′ , t′ ).

The use of the retarded Green function ensures that the observer only feels
the effect of the source after it is turned on.

2. Advanced Solution Here we measure a wave ψout (x, t) as t −→ ∞,

Z
ψ(x, t) = ψout (x, t) + d3x′ dt′ G(−) (x, t; x′ , t′ )f (x′ , t′ ).

The use of G(−) means that, once the source ceases, the effects from the
source are no longer felt, or more precisely they are contained within ψout .

Case 1 above is the more commonly encountered, for example in the case ψin ≡ 0
so that there is no wave in the distant past, and a source f (x, t) switches on
at some time. Then inserting our explicit expression for the Green function, we
obtain
f (x′ , t′ret)
Z
ψ(x, t) = d3 x ′
|x − x′ |
where the subscript ret denotes that the function f is evaluate at time
1
t′ret = t − |x − x′|.
c

6.6 Conservation of Energy and Momentum and Poynting

Vector

In this section, we will derive laws expressing conservation of energy and momen-
tum for electric and magnetic fields.
Chapter 2 17

The force acting on a particle carrying charge q, and moving with velocity v is

F = q(E + v × B).

The work done/unit time, or rate of change of mechanical energy, is then

d
Emech = v · [q(E + v × B)]
dt
= qv · E,

since the second term vanishes. Thus generalising to a current density J we

have
d Z
Emech = d3 x J · E. (6.23)
dt
We will now relate the rate of change of mechanical energy to the change of energy
in the electric and magnetic fields. The starting point is Maxwell-Ampère’s law
(ME3), which gives
∂D
Z Z " #
3 3
d xJ · E = d xE · ∇ ×H − .
V ∂t
We can use the vector identity ∇ · (E × H) = H · ∇ × E − E · (∇ × H) to write
∂D
Z Z ( )
d3 x J · E = d3x H · (∇ × E) − ∇ · [E × H] − E · .
V ∂t
Identifying the l.h.s. of this equation with the rate of change of mechanical energy
in eqn. (6.23), and using Faraday’s law (ME2) on the r.h.s., we obtain
d ∂B ∂D
Z ( )
Emech = − d3x H · + ∇ · (E × H) + E · .
dt ∂t ∂t
We will now assume that the medium is linear, allowing us to write
∂B 1∂
H· = [H · B]
∂t 2 ∂t
∂D 1∂
E· = (E · D),
∂t 2 ∂t
and thus
d ∂ 1
Z ( " #)
Emech = − d3 x ∇ · (E × H) + (H · B + E · D)
dt ∂t 2
Chapter 2 18

We have already, in Chapter 4.6, interpreted 12 ǫ0 |E|2 ≡ 12 E · D as the energy

density of an electric field. Likewise we will identify 21 H · B as the magnetic
energy density and hence their sum
1
u = (H · B + E · D) (6.24)
2
as the electromagnetic energy density. With this identification, we now have
Poynting’s Theorem expressing conservation of energy

∂u
Z Z " #
3 3
− d x J ·E = dx + ∇ · (E × H) (6.25)
V V ∂t

Since this applies for any volume V , we have a differential energy continuity
equation

∂u
+ ∇ · (E × H) = J · E (6.26)
∂t

The vector

S = E×H

is the Poynting Vector. It only enters through a divergence in the above expres-
sions but, when we come to consider its properties under Lorentz transformations
later in the course, we will discover that it is essentially unique.
We can reduce the integral over the Poynting vector in eqn. (6.25) to a surface
integral using the divergence theory. Thus we can interpret the Poynting vector
as the energy flux across a surface, and the Poynting theorem in essence says:
“The rate of change of electromagnetic energy in a volume together with energy
flux across the boundary is equal to minus the total work done by sources within
the volume”.
Chapter 2 19

6.6.1 Energy Conservation in terms of the Fundamental Microscopic

Fields

The field energy density of eqn. (6.24) contains not only the fundamental fields,
but also the “derived” fields H and D. Thus they include contributions associ-
ated with the polarization and magnetisation of the medium which are in essence
mechanical, and should be associated with the J · E term.
Let Emech be the mechanical energy in some fixed volume V . We have seen
that the work done per unit time per unit volume J · E is the rate of increase of
mechanical energy,
dEmech Z 3
= d x J · E.
dt V

In the case of a vacuum, we have

Z 1Z 3
d3 x u = d x(H · B + E · D)
V 2 Z
ǫ0
= d3 x(E 2 + c2 B 2 )
2 V
= Efield

where now we have expressed the field energy solely in terms of the fundamental
fields. It is this expression that is more naturally associated with the field energy,
and Poynting’s theorem reads
d I
(Emech + Efield ) = − dA · S (6.27)
dt
Chapter 2 20

6.6.2 Conservation of Linear Momentum

Again we work with the microscopic fields. The force on a particle of charge q
is
F = q(E + v × B).
Thus Newton’s second law may be expressed as
d Z
P mech = d3x [ρE + J × B]
dt
where P mech is the total momentum of the particles in a volume V . To evaluate
this expression, we once again use Coulomb’s law (ME1) and Ampère’s law (ME3),
yielding for the integrand
1 ∂E
" #
ρE + J × B = ǫ0 E(∇ · E) − B × ∇ × B − ǫ0 .
µ0 ∂t
We now use
∂ ∂E ∂B
E×B = ×B+E×
∂t ∂t ∂t
∇·B = 0

to write

ρE + J × B =
∂B ∂
ǫ0 [E(∇ · E) + c2 B(∇ · B) − c2 B × (∇ × B) + E × − (E × B)].
∂t ∂t
We now use Faraday’s law (ME2) to write
d d Z
P mech + ǫ0 d3x E × B =
dt Z dt V
ǫ0 d3x [E∇ · E + c2 B∇ · B − E × (∇ × E) − c2 B × (∇ × B)], (6.28)

where we assume that the volume V is fixed. The second term on the l.h.s. we
associate with the momentum carried by the field
Z
P field = ǫ0 d3x E × B, (6.29)
Chapter 2 21

which we can rewrite as

Z
1 Z
P field = ‘ d3 x
2
E × H = d3 x g, (6.30)
c
where g is the electromagnetic momentum density given, up to a constant,
by the Poynting Vector,
1
g = 2 S. (6.31)
c
To proceed further, let us consider the r.h.s. of the momentum conservation law,
eqn. (6.28). Using index notation, we may write
∂Ej ∂Em
[E(∇ · E) − E × (∇ × E)]i = Ei − ǫijk ǫklm Ej
∂xj ∂xl
∂Ej ∂Ej ∂Ei
= Ei − Ej + Ej
∂xj ∂xi ∂xj
∂ 1
= [EiEj − E 2δij ].
∂xj 2
What we have done is to write the electric part of the integrand as a derivative. We
may treat the magnetic term similarly, and now introduce the Maxwell Stress
Tensor
1
" #
2
Tij = ǫ0 EiEj + c Bi Bj − (E 2 + c2 B 2)δij (6.32)
2
Note that this tensor is symmetric.
We can thus write the momentum conservation law as
d Z
∂Tij
[P mech + P field ] = d3 x
dt V ∂xj
which, after applying the divergenceb theorem, becomes
d I
[P mech + P field ] = dA Tij nj (6.33)
dt S=∂V

where n is the outward normal to the surface enclosing V .

Note that Tij nj is the flow of momentum per unit area across surface S into the
volume V , i.e. it is the force per unit area acting on the combined system of
particles and fields within volume V .
Chapter 7

Plane Electromagnetic Waves and Wave

Propagation

We begin by the considering the propagation of waves in a non-conducting medium.

Thus J ≡ 0, we assume ρ ≡ 0 and Maxwell’s equations reduce to

∇·B = 0
∂B
∇×E+ = 0
∂t
∇·D = 0
∂D
∇×H − = 0.
∂t
In the case of plane waves, it is sufficient to consider those propagating with
a definite frequency ω, and hence time dependence exp −iωt; essentially this is
equivalent to taking the Fourier Transform. We have a set of linear, homogeneous
equations and hence all fields have the same harmonic behaviour. Thus we may
write Maxwell’s equations as

∇·B = 0
∇·D = 0
∇ × E − iωB = 0
∇ × H + iωD = 0.

1
Chapter 2 2

We will now specialise to the case of a linear constitutive relation between the
fields: D = ǫE and B = µH. We will also assume ǫ, µ are real. Note that later
we will consider the complex case; taking them to be real corresponds to there
being no energy losses. Then the last two equations of eqn. (7.1) become

∇ × E − iωB = 0
∇ × B + iωǫµE = 0,

which, with the remaing two Maxwell equations, yield

∇2 E + ω 2 ǫµE = 0
∇2B + ω 2 ǫµB = 0 (7.1)

These are known as the Helmholtz wave equations. As is well known, they
support the plane-wave solutions
   
E E0  ik ·x−iωt



 =
 e , (7.2)
B B0
√
where k = ω µǫ, and
√
v = ω/k = 1/ µǫ
is the phase velocity.
We now recall the velocity of light in a vacuum is given by
√
c = 1/ µ0 ǫ0 .

Thus we can write

v = c/n
where v
µǫ
u
n= (7.3)
u
t
µ0 ǫ0
is the index of refraction. It is usually a function of the frequency, e.g. a prism,
and therefore the phase velocity is likewise frequency dependent - hence the name.
Chapter 2 3

7.1 Propagation of Monochromatic Plane Wave

We will now consider in greater detail a monochromatic plane wave of frequency

ω, propagating in the direction n with wave number k. Note that complex n
corresponds to dissipation. We have seen that the solution of the Helmholtz
equations are

E(x, t) = E 0 eikn·x−iωt
B(x, t) = B eikn·x−iωt
0 (7.4)

with
k 2 = µǫω 2 .
Thisis actually shorthand for

E(x, t) = ℜ E0eikn·x−iωt .
n o

The imaginary part contains no physical information. It is important to remember

this when considering quantities that are quadratic or higher in the fields, such
as the energy density.

7.1.1 Energy Density for Monochromatic Plane Wave

Recall the expression for the energy density

1 1 2
" #
2
u= ǫE + B .
2 µ
The real parts of the fields B and E must be taken before evaluating the quadratic
terms. In the case of the time-averaged energy density, we have the particularly
simple result
1 1
" #
hui = ǫE · E ∗ + B · B ∗ (7.5)
4 µ
where we use h. . .i to denote that the time average has been taken, and the
additional factor of one half arises from the observation

hcos2 ωti = 1/2.

Chapter 2 4

Likewise, the time-averaged Poynting vector is

1 ǫv
hSi = E × H ∗ = E · E ∗n. (7.6)
2 2
This quantity is called the intensity of the wave.

7.2 Polarisation of a Monochromatic Plane Wave

Applying ∇ · B = 0 and ∇ · E = 0 to the solutions of eqn. (7.4), we find

n · B0 = 0
n · E 0 = 0. (7.7)

Thus both E and B are perpendicular to the direction of propagation. We say

they are transverse wave.
We now apply the remaining Maxwell equations

∇ × E − iωB = 0
∇ × B + iωµǫE = 0,

to yield
√
B0 = µǫn × E 0 . (7.8)
√
Setting c = 1/ µǫ to be the volocity of light in the medium, we see that both cB
and E have the same magnitude.
N.B. Had we chosen to work with H, rather than B, then we would have
n × E0
H0 =
Z
q
where Z = µ/ǫ is the impedence
We will now specialise to the case where n is indeed real. Then B 0 is perpendicular
to E 0 , and has the same phase.
Chapter 2 5

ε2

ε1
The vectors E, B and n form an orthogonal triad, and it is usual to introduce three
mutually-orthogonal basis vectors ǫ1 , ǫ2 and n and to write the electromagnetic
field as

E1(x, t) = ǫ1 E1ei(k ·x−ωt) ; cB1 = ǫ2 E1 ei(k ·x−ωt)

E2(x, t) = ǫ2 E2ei(k ·x−ωt) ; cB2 = −ǫ1 E2 ei(k ·x−ωt) (7.9)

Note that E1 and E2 can be complex to account for a phase shift between the two
plane waves.
The general solution for the wave equation is

E(x, t) = (ǫ1 E1 + ǫ2 E2 )ei(k·x−ωt).

Chapter 2 6

Linear Polarization
ε2

θ
ε1
If E1 and E2 have the same phase we talk about a linearly polarized wave;
the direction of the E field is constant, with the angle given by
θ = tan−1 E2/E1.

Elliptical and Circular Polarization

If E1 and E2 have different phases, we say the wave is elliptically polarized.

The direction of E is no longer constant.
A special case is that of circularly polarized waves. Here E1 and E2 have the
same magnitude, but differ by a phase of ±π/2. Thus we can write
E(x, t) = E0(ǫ1 ± iǫ2 )ei(k·x−ωt)

where E0 is real. W.l.o.g., we tak ǫ1 and ǫ2 in the x and y directions respectively.

Thus taking the real (physical) part, we find
Ex = E0 cos(kz − ωt) = E0 cos(ωt − kz)
Ey = ∓E0 sin(kz − ωt) = ±E0 sin(ωt − kz).
At fixed z, this is just the equation of a circle.
Chapter 2 7

θ=ω t

k
The different signs correspond to rotating to the left or rotating to the right; these
are more commonly known as positive and negative helicities.
Since it is possible to use any two mutually orthogonal vectors as polarization
vectors, it is usually for circularly polarized waves to introduce
1
ǫ± = √ (ǫ1 ± iǫ2 ) (7.10)
2
with the properties

ǫ± ∗ · ǫ± = 1, ǫ± ∗ · ǫ∓ = 0, ǫ± ∗ · n = 0,

so that a general plane-wave solution is

E(x, t) = (E+ǫ+ + E− ǫ− )ei(k·x−ωt) .

An important question is, given an electric field E(x, t), how can one determine
its polarization properties; one way of specifying the relative importance of the
different components is through the Stokes Parameters. This is described in
Jackson 7.2.
Chapter 2 8

7.3 Reflection and Refraction at Plane Interface between

Dielectrics

The laws describing the behaviour of a wave at the interface between two media
are well known:

1. Angle of reflection = Angle of incidence

2. sin θi/ sin θt = n′/n (Snells’s law) where n′, n are the refractive indices of the
final and initial media respectively.

These are simple kinematic laws; we would like to determine dynamic properties
- intensities and phase changes.

θT kT

kr
ki
θi θr

We begin by writing

Incident wave: Ei = E0i ei(ki ·x−ωt)

√ 1
Bi = µǫ ki × Ei
ki

Reflected wave: Er = E0r ei(kr ·x−ωt)

√ 1
Br = µǫ kr × Er
kr
Chapter 2 9

Refracted wave: ET = E0T ei(kT ·x−ωt)

q 1
BT = µ′ ǫ′ kT × ET
kT
where ki2 = kr2 = µǫω 2 and kt2 = µ′ ǫ′ ω 2.

Boundary Conditions at Interface

We first observe that the boundary conditions must be satisfied ∀x, y at all times
t. Thus all fields must have the same phase factor at z = 0. N.B.: We have
implicitly assumed this in saying that the frequency in z > 0 must be the same
as that in z < 0.
Thus ki · x = kr · x = kT · x at z = 0. The k’s lie in a plane - plane of incidence.
From the figure, we see that

ki · x = |x||ki | cos(π/2 + θi ) = −|x||ki | sin θi

kr · x = |x||kr | cos(π/2 + θr ) = −|x||kr | sin θr

and thus we have

θi = θr

Similarly,

|ki | sin θi = |kT | sin θT

=⇒ µǫ sin θi = µ′ ǫ′ sin θT .

and thus

v
sin θi u ′ ′
uµ ǫ n′
=t =
sin θT µǫ n
Chapter 2 10

Thus both laws are purely kinematic properties.

The boundary conditions themselves are
Ek is continuous
Hk is continuous

D⊥ is continuous
B⊥ is continuous
Applying to the fields at the interface, we have
(E0i + E0R − E0T ) × n = 0 (7.11)
1 1
" #
i r T
(ki × E0 + kr × E0 ) − ′ kT × E0 × n = 0 (7.12)
µ µ
ǫ(E0i + E0r ) − ǫ′ E0T · n = 0
h i
(7.13)
ki × E0i + kT × E0r − kT × E0T · n = 0.
h i
(7.14)
We we now consider two cases; where the electric polarization vector is normal to
plane of incidence and where it is parallel to plane of incidence.

7.3.1 Normal to Plane of Incidence

BT
n ET
θT kT

ex
kr
ki
Bi θi θr Br

Ei Er

The z axis is normal to the interface, and we choose the x axis to be in the plane of
incidence as shown. Thus the electric field is along the y axis. The first boundary
Chapter 2 11

condition eqn. (7.11) yields

E0i + E0r − E0T = 0. (7.15)

We now turn to the second second boundary condition eqn. (7.12). The first term
yields
1h 1h i
ki × E0i × n = E0(n · ki ) − ki (n · E0i )
i i

µ µ
v
1 i uǫ
u
= E0|ki | cos θi = ω t cos θi E0i .
µ µ
We treat the other two terms similarly, and we find
v
v u ′
uǫ uǫ
u
(E0i − r
E0 )ω t cos θi − ωt ′ cos θT E0T = 0,
µ µ
yielding v
v u ′
uǫ uǫ
u 
cos θi t E0i − E0r − cos θT E0T = 0

t (7.16)
µ µ′
The remaining boundary conditions yield no new information, so combining eqns. (7.15)
and (7.16) we find
r
1 − ǫǫµµ′ cos θT
′
E0r cos θi 1 − µµ′ tan
tan θi
θT
i = = µ tan θi (7.17)
r
E0 ǫ ′ µ cos θ
1 + ǫµ′ cos θTi 1 + µ′ tan θT
E0T 2 2
= = µ tan θi (7.18)
E0i
r
1 + ǫǫµµ′ cos
′ θT 1 + µ′ tan θT
cos θi

For visible light, we can usually put µ = µ′ , giving

E0r sin(θT − θi )
i =
E0 sin(θi + θT )
T
E0 2 sin θT cos θi
i = .
E0 sin(θT + θi )
This is just Fresnel’s formula for light polarized perpendicular to plane of inci-
dence.
Chapter 2 12

7.3.2 Electric Field in Plane of Incidence

Here we use boundary conditions eqns (7.11) and (7.13) to yield

r
1 − ǫµ cos θT
′
E0r ǫ′ µ cos θi 1 − ǫǫ′ tan
tan θi
θT
= = (7.19)
i ǫ
1 + ǫ′ tan θθTi
tan
r
E0 ǫµ ′ cos θ
1 + ǫ′ µ cos θTi
r
ǫµ′
E0T 2 ǫ′ µ 2 nn ǫǫ′
′

= = (7.20)
E0i 1 + ǫǫ′ tan
tan θi
r
ǫµ′ cos θT
1 + ǫ′ µ cos θi θT

If µ = µ′ , then ǫ/ǫ′ = sin2 θT / sin2 θi = n2 /n′2, and we have

E0r tan(θi − θT )
=
E0i tan(θi + θT )
E0T 2 sin θT cos θi
=
E0i sin(θi + θT ) cos(θi − θT )

Incident Wave Normal to Interface

In this particular case, we find

r
1 − ǫǫµµ′
′
E0r n − n′
= −→ if µ = µ′
E0i
r
1 + ǫǫµµ′
′
n + n′
E0T 2 2
= −→ if µ = µ′
E0i
r
1 + ǫǫµµ′
′
1 + n′/n

In this formula we assume that the directions of E0r and E0i are the same. (contrary
to Eq. (7.42) from Jackson where the directions of E0r and E0i are assumed to be
opposite).
Thus we see that, if both refractive indices are equal
E0r = 0
E0T = E0r
Chapter 2 13

as expected. If the second media is a conductor, n′ −→ 0 , so all of the wave is

reflected, with
E0r = −E0i (7.21)

7.4 Brewster’s Angle and Total Internal Reflection

7.4.1 Brewster’s Angle

In the case of polarization in the plane of incidence, we have

E0r
tan θi
1 − ǫǫ′ tan θT
i = ǫ tan θi .
E0 1 + ǫ′ tan θT

There is an angle for which no wave is reflected, given by

ǫ tan θi
= 1.
ǫ′ tan θT
Setting µ = µ′ = 1, so that ǫ/ǫ′ = n2 /n′2, we find
′
 
−1  n 
θi = tan . (7.22)
n
This is Brewster’s Angle. If we have a plane wave of mixed polarization incident
at this angle, the reflected radiation only has a polarization component perpen-
dicular to the plane of incidence. It is a simple way to produce plane-polarized
light.
Chapter 2 14

7.4.2 Total Internal Reflection

θT kT

kr
ki
θi θr

i0

If light passes from a medium of higher optical density to one of lower optical
density, the angle of refraction is greater than the angle of incidence.
Hence there is a θi for which θT = π/2, given by

sin θi = sin i0 = n′ /n (7.23)

From Snell’s law, we have in general

v
n2
q u
cos θT = 1 − sin2 θT = sin2 θi
u
t
1− ′2
v
n
u !2
u
sin θi
= t1 − .
u
sin i0
For θi > i0, cos θT becomes purely imaginary. Thus the refractive wave has a
phase factor

eikT ·x = ei(kT x sin θT +kT z cos θT )

√
ikT x(n/n′ ) sin θi −kT z (sin θi / sin i0 )2 −1
= e e .

We see that the refracted wave propagates parallel to the surface, and is ex-
ponentially attenuated with increasing z. The attenuation occurs over only a
few wavelengths unless θi ≈ i0 .
Chapter 2 15

Note that the time-averaged energy flux across the interface is

1 h
hS · ni = ℜ n · (ET × HT∗ ) .
i

2
Now HT = (kT × ET )/µ′ ω, and thus

n · (ET × HT∗ ) = n · ET × (kT × ET∗ ) /µ′ ω

h i

= |ET |2 n · kT /µ′ ω,

whence
1 h
ℜ |ET |2 n · kT /µ′ ω
i
hS · ni =
2
1 h
= ℜ |ET |2 kT cos θT /µ′ ω
i

2
= 0,

since cos θT is purely imaginary; there is no time-averaged energy flux across the
interface.
The principle of total internal reflection has many applications, most notably
in fibre-optic cables. The analysis presented here assumes, of course, that the
material is wide compared to the wave length of light.
Chapter 2 16

7.5 Dispersion

So far, we have been investigating the propagation of waves of a fixed frequency.

The wave number is related by

k 2 = µǫω 2 .

Suppose now we consider a wave having a spread of frequencies. In general,

the values of µ and, in particular, ǫ are frequency dependent, and thus different
frequencies have different propagation properties. This is call dispersion

7.5.1 Simple Model for Dispersion

Consider an electron of mass m and charge −e, bound to a (fixed) nucleus by a

harmonic potential with resonant frequency ω0 , and a damping term with damping
constant γ. In the absence of an external electric field, the electron will undergo
damped simple-harmonic motion about an equlibrium.
111
000
000
111
000
111
000
111
x e-

E (t)
Nucleus
We now apply an external electromagnetic field (E, B). Then the force on the
electron is
F (t) = −e(E(t) + v × B(t)).
Chapter 2 17

Providing the velocity is small compared to that of light, the magnetic force will
be negligible; recall that c|B| ≈ |E|. Thus the equation of motion of the electron
is
d2 d
m( 2 x + γ x + ω02 x) = −eE(t).
dt dt
The dipole moment of the system is just p = −ex. We now assume that the
external field has frequency ω, so that the time dependence is

E = E0e−iωt.

Thus the displacement will have the same frequency dependence, and we have an
equation of motion
m(−ω 2 − iωγ + ω02 )x = E0 ,
yielding a dipole moment
e2 2
p = (ω0 − ω 2 − iωγ)−1.
m
We now consider the case of N atoms/unit volume, each having Z electrons of
which fj electrons have resonant frequency ωj . We will take this as a model for a
linear medium, in which the polarization P arises solely from the applied external
field. Thus, recalling that
P = ǫ0 χe E
we find
ǫ(ω) N e2 X
= 1 + χe = 1 + fj (ωj2 − ω 2 − iωγj )−1.
ǫ0 ǫ0 m j
with fj = Z. We will rewrite this expression as
P
j

ǫ N e2 X (ωj2 − ω 2) + iωγj
=1+ fj . (7.24)
ǫ0 ǫ0 m j (ωj2 − ω 2 )2 + ω 2γj2
We have thus seen how even a simple model gives a frequency-dependent permit-
tivity.
Chapter 2 18

7.5.2 Permittivity in Resonance Region

In general, we can assume that the damping factor γ is small. From the form of
eqn. (7.24), it is clear that at very log frequencies, the susceptibility is positive and
the relative permittivity greater than one. As successive resonant frequencies are
passed, more negative terms contribute and eventually the relative permittivity
is less than one.
Particularly interesting is the behaviour in the neighbourhood of a resonance.

Re ε

Im ε
ω
ωj

Here the real part of ǫ(ω) is peaked around ωj , and furthermore displays anoma-
loous dispersion in which light of higher frequency is less refracted than light of
lower frequency.
The presence of an appreciable imaginary part of ǫ(ω) near ω = ωj represents
absorption; energy dissipated in the medium. To see how this arises, consider a
wave propagating in the z-direction. We will write the wave number as

k = β + iα/2; amplitude ≈ e−αz/2.

Chapter 2 19

Thus α clearly represents absorption of the wave. Setting µ = µ0 , and recalling

√
k 2 = µǫω, we have
√
(β 2 − α2/4) + iαβ = ( µ0 ǫ0 )2ω 2 ǫ/ǫ0

which gives 
2 2 ω2
β − α /4 = c2 ℜ ǫ/ǫ0


ω2
.
αβ = c2 ℑ ǫ/ǫ0



Note that if α ≪ β, we have

ℑ ǫ(ω)
α = β
ℜ ǫ(ω)
where
ωq
β = ℜ ǫ/ǫ0
c

7.5.3 Low Frequency Behaviour and Electrical Conductivity

In a conductor, some of the electrons can move freely. Thus there are some
electrons with resonant frequency ω0 = 0, whose contribution to the permittivity
is
N e2 f0
ǫ(ω) = ǫ̃(ω) + i ,
mω(γ0 − iω)
where ǫ̃ represents the background permittivity coming from all the other modes.
We see from this that ǫ(ω) is singular as ω −→ 0, and we will now relate this
property to electrical conductivity.
Our starting point is the Maxwell-Ampère law (ME3):
∂D
∇×H =J + . (7.25)
∂t
We will now impose that J and E are related through Ohm’s law

J = σE

where σ is the conductivity. If we assume the usual frequency behaviour exp −iωt,
and assume the background permittivity is a constant ǫb = ǫ̃(ω), eqn. (7.25)
Chapter 2 20

becomes
σ
" #
∇ × H = −iω ǫb + i E. (7.26)
ω
An alternative procedure is to ascribe all properties, including current flow, to
the dielectric properties of the medium. In that case we have
N e2 f0 
 

∇ × H = −iωD = −iω ǫb + E. (7.27)

mω(γ − iω)
Comparing eqns. (7.26) and (7.27), we find
N e2 f0
iσ/ω = i
mω(γ0 − iω)
i.e.
N e2 f0
σ= . (7.28)
m(γ0 − iω)

Note that we can rewrite this expression as

σ0
σ=
1 − iωτ
where
N f0 e2
σ0 = ,
mγ0
and τ = γ −1. Essentially, we have

• N f0 is number of free electrons per unit volume.

• γ0 /f0 is damping constant, determined experimentally.

For good conductors γ0/f0 ≃ 4 × 1013 s−1. If we assume f0 ≃ 1, then ωτ is small

rights up to the microwave region ω ≃ 1011 s−1 ; σ is real.
Note that if ωγ0 ≫ 1, then σ is purely imaginary, and we have a phase shift
between E and J.
Chapter 2 21

7.6 High-Frequency Behaviour and Plasma Frequency

Suppost that ω is much larger than the highest resonance frequency. Then we
have
N e2 X (ωj2 − ω 2 ) + iωγj
ǫ/ǫ0 = 1+ fj
ǫ0 m j (ωj2 − ω 2 )2 + ω 2 γj2
ω/ωj ≫1 N e2 X
−→ 1− fj ω 2 /ω 4
ǫ0 m j
= 1 − ωP2 /ω 2 (7.29)

where
N Ze2
ωP2 = (7.30)
ǫ0 m
is the plasma frequency , so called because all the electrons essentially behave
as if free.
Recalling that v
√ 1u
uǫ
k= µǫω = t ω,
c ǫ0
where c is the velocity of light in vacua, we have
q
ck = ω 2 − ωP2

whence
ω(k)2 = ωP2 + c2 k 2 . (7.31)
Such an expression, describing the relationship between wave number and fre-
quency, is known as a dispersion relation . Similar expressions occur in many
places in physics, including special relativity and sound propagation.
In a typical dielectric, when ω 2 ≫ ωP2 , the dielectric constant is slightly less than,
but close to, unity.
In a true plasma, such as the ionosphere, all the electrons are essentially free, and
the expression eqn. (7.29) is valid for a range of frequencies, including ω < ωP .
The wave number k is purely imaginary for frequencies less than the plasma
Chapter 2 22

frequency. Thus a wave incident on a plasma are attenuated in the direction of

propagation, with intensity
√
−2 ωP2 −ω 2 z/c ω→0 −2ωP z/c
I∝e −→ e .

7.6.1 Model of Wave Propagation in the Atmosphere

The above plasma model for the ionosphere is modified considerably through the
presence of the earth’s magnetic field. In the model we now construct, we assume
propagation parallel to the earth’s field B0 . We assume that there is a force acting
on the charges due to a propagating electric field, but that the only magnetic force
is that arising from the earth’s field; recall once again that c|B| ≃ |E|.
Thus the equation of motion for an electron of charge −e and mass m is
d2 x
m 2 = −ev × B0 − eE.
dt
Once again, we consider a monochromatic plane wave with time dependence

e−iωt.

It is convenient to consider the case of circularly polarized waves, for which we

introduce the complex polarization vectors
1
ǫ± = √ (ǫ1 ± iǫ2 )
2
ǫ3 = k̂ (Normal in direction of k).

Thus we have
x = x+ǫ+ + x−ǫ− + x3ǫ3 ,
so that the equation of motion becomes
d2 x + d2 x − d2 x 3 dx+ dx− dx3
m[ 2
ǫ+ + 2
ǫ− + 2
ǫ3 ]−eB 0 ǫ3 ×[ ǫ+ + ǫ− + ǫ3 ] = −e[E+ǫ+ +E− ǫ− ]e−iωt.
dt dt dt dt dt dt
First, it is easy to see that since ǫ3 ×ǫ3 = 0, the motion along the Z direction is free:
x3 = x30 + v3t. Since the forces acting in XY plane are periodic, the motion of the
charges in the XY plane will be periodic, too: x+ (t) = x+ e−iωt, x− (t) = x− e−iωt.
Chapter 2 23

Now
1
ǫ3 × ǫ+ = √ (ǫ3 × ǫ1 + iǫ3 × ǫ2 )
2
i
= − √ (ǫ1 + iǫ2),
2
yielding

ǫ3 × ǫ+ = −iǫ+
ǫ3 × ǫ− = iǫ− (7.32)

and thus

−ω 2m[x+ ǫ+ + x−ǫ− ] + iωeB0 [−ix+ǫ+ + ix− ǫ− ] = −e[E+ǫ+ + E−ǫ− ].

Looking at the individual components, we find

−ω 2 mx+ + ωeB0 x+ = −eE+

−ω 2 mx− − ωeB0 x− = −eE−

which we may write

e
x± = E± .
mω(ω ± eB0 /m)
We now introduce
ωB ≡ eB0 /m,
the frequency of precession of a charged particle in a magnetic field.
Recalling that p = −ex, we have a dipole moment of the particle
−e2
p± = E± .
mω(ω ± ωB )
Thus, recalling the expression for the plasma frequency eqn (7.30), the polariz-
ability may be written
ωP2
P± = −ǫ0 E±
ω(ω ± ωB )
whence
Chapter 2 24

ωP2
ǫ± /ǫ0 = 1 − (7.33)
ω(ω ± ωB )

ε-

ω/ωB

ε+
ε-

Thus, in this highly simplified model, we see that the permittivity depends on
the polarization of the incident wave. Indeed, for certain ranges of ω we find that
the permittivity can be negative, and hence one or both polarizations no longer
propagate.

7.7 Superposition of Waves and Group Velocity

So far we have considered monochromatic waves, but have seen that, if the medium
is dispersive, different frequencies will travel with different velocities. In the sec-
tion, we will describe how, for a general plane wave, the rate of energy flow is in
general different from the phase velocity, or velocity of propagation of a particular
Chapter 2 25

frequency component. To simplify the discussion, we will consider the problem in

one dimension.
We will write a general wave in terms of its physical components. The dispersive
properties are encompassed in the dispersion relation

ω ≡ ω(k)

where ω(−k) = ω(k). The general solution is then

Z ∞dk
u(x, t) = A(k)eikx−iω(k)t,
−∞ 2π

where the amplitudes A(k) are given by

Z ∞
A(k) = dxu(x, 0)e−ikx.
∞

For a monochromatic wave, of wave number k0, we have

u(x, 0) = eik0 x

yielding
A(k) = 2πδ(k − k0).
In practice we virtually never deal with pure monochromatic plane waves of fixed
frequency k0, but rather with pulses, centred about a frequency k0. In particular,
we will consider the propagation of a Gaussian wave packet, of width ∆x, centred
at x = 0. Then !1/4
1 −x2 /4∆x2 ik0 x
u(x, 0) = e e .
2π∆x2
Chapter 2 26

∆x

This satisfies
Z
2
hx i = dx |u(x, 0)|2x2
1/2 Z
1
!
−x2 /2∆x2 2
= dxe x
2π∆x2
!1/2
1 d Z
2 2
= 2
(−2) 2
dx e−x /2∆x
2π∆x d(1/∆x )
= ∆x2,

showing that the width is indeed ∆x.

The amplitudes of the various componenents are given by
!1/2 Z
1 2
/4∆x2 i(k0 −k)x
A(k) = dx e−x e
2π∆x2
2 2
≃ e−(k0 −k) /4(1/2∆x)
. (exercise)

By analogy with the width of the wave packet, we see that the amplitude A(k) is
centred at k = k0, with width
1
∆k = .
2∆x
Chapter 2 27

In fact, more generally we have

∆x∆k ≥ 1/2 (7.34)

Thus we have the important observation that a short pulse, even of “fixed” fre-
quency k0 , contains a spread of monochromatic components. This expression, of
course, is more familiar from Heisenberg’s Uncertainty Principle.

7.7.1 Group Velocity

To see how this spread of frequencies effects the propagation of a wave, we con-
sider the simple case of two monochromatic waves, of the same amplitude and
of neighbouring frequencies (k1, ω1) and (k2, ω2), where k1 , k2 ∼ k0 . Then the
resulting “wave packet” propagates as

U (x, t) = A ei(k1 x−ω1 t) + ei(k2 x−ω2 t)

h i

= Aei[(k1 +k2 )x/2−(ω1 +ω2)t/2] ei[(k1 −k2 )x/2−(ω1−ω2 )t/2] + ei[(k2 +k1 )x/2−(ω2+ω1 )t/2]
n o

k1 − k2 ω1 − ω2 i[(k1 +k2 )x/2−(ω1 +ω2)t/2]

" #
= 2A cos x− t e
2 2
We have written the wave as a slowly moving amplitude factor with velocity
ω1 − ω2 dω
vg = −→ as k2 → k1 , (7.35)
k1 − k2 dk k0

known as the group velocity , and a rapidly moving “phase” with velocity
ω1 + ω2 ω
vp −→ = as k2 → k1 . (7.36)
k1 + k2 k
Since the energy density is associated with the amplitude of the wave, we see
that, in this approximation, energy is transmitted with the group velocity, given
by eqn. (7.35) with k0 the central value of the wave number.
We now recall the relationship between ω and k
ck
ω= , (7.37)
n(k)
Chapter 2 28

where n(k) is the index of refraction, and c is the velocity of light in a vacuum.
The phase velocity can then be written
ω(k) c
vp = = . (7.38)
k nk
This can be either less than or greater than the speed of light; for most media at
optical frequencies, n(k) > 1. We can rewrite the group velocity using eqn. (7.37),
regarding k = k(ω), and find
dn dk
n(ω) + ω = c
dω dω
dω c
=⇒ vg = = dn .
dk k0 n(ω) + ω dω
Providing dn/dω > 0, we have vg < c. However, if dn/dω < 0 (anomalous
refraction), vg can be greater than c.

7.7.2 Propagation of a Gaussian wave packet in the dispersive medium

First, let us recall the propagation of a Gaussian pulse in a linear medium without
dispersion
1 1/4 (x − vt)2
( )
u0(x, t) = ( 2 ) exp − + ik0(x − vt) (7.39)
πL 2L2
√
where L = ∆x 2 is the width of the Gaussian wave packet.
Suppose at t = 0 we switch on the dispersion so that ω = ω(k) (some non-linear
function). What will happen with the pulse?
Starting from t = 0, the solution of the wave eqn is
dkZ
u(x, t) = ℜ A(k) e−iω(k)t+ikx
2π
A solution of the second-order duifferential eqn is specified if we know both u(x, 0)
and u̇(x, 0). It is easy to prove that
Z i
A(k) = dx e−ikx[u(x, 0) + u̇(x, 0)] (7.40)
ωk
Chapter 2 29

The initial condition should be taken from the form of a non-dispersive Gaussian
pulse (7.39) at t = 0:
1 1/4 x2
! ( )
u(x, 0) = u0(x, t) = exp − 2 + ik0 x
πL2 2L
!1/4
1 ivx x2
! ( )
iu̇(x, 0) = iu̇0(x, t) = k0 v + 2 exp − 2 + ik0 x
πL2 L 2L
From Eq. (7.40) we obtain
kv (k − k0 )2L2
" # ( )
2 1/4
 
A(k) = 4πL 1+ exp − (7.41)
ωk 2
A typical behavior of ω(k) is given by eq. (7.31). For simplicity, we will consider
an approximate model of the behavior of frequency in the vicinity of ω0 in the
form
a2 k 2 
 

ω(k) = ω0 1 + (7.42)
2
where ω0 = vk0 is the center of our Gaussian wave packet.
We obtain
dk kω0 (k−k0 )2 L2
!
2 1/4
Z
u(x, t) = ℜ(4πL ) 1+ e− 2 e−iωk t+ikx
2π k 0 ωk
kω0
The term k0 ωk is approximately 1 in the vicinity of k0 so

2(4πL2)1/4 −iω0t(1+ a2k02 )+ik0x (x − ω0 a2 k0t)2

( )
u(x, t) = ℜ √ 2 e 2 exp − 2 (7.43)
L + iω0 a2 t 2L2(1 + iω0 aL2t )
The peak of the pulse (7.43) is located at x = ω0 a2 k0 t ⇒ it moves with the it
group velocity ω0a2 k0 = ∂ω k
∂k k=k
.
0

The wave packet spreads as it moves:

v
√ u
u
t 2 a4 ω02 t2
2∆x(t) ≡ L(t) = L +
L2
(for a proof, see Jackson). This is a general feature of non-linear Gaussian wave
q
packets: for the same reason (ωk = (m2 c4 /h̄2 ) + k 2) wave packets corresponding
to relativistic particles broaden with time.
Chapter 2 30

7.8 Causality between E and D and Kramers-Kronig Re-

lations

When ǫ(ω) is frequency dependent, there is a non-local temporal relation between

D and E. To exhibit this, we write D and E in terms of their temporal Fourier
components
Z ∞dω
D(x, t) = D̃(x, ω)e−iωt.
−∞ 2π
For a linear medium
D̃(x, ω) = ǫ(ω)Ẽ(x, ω),
and thus Z ∞ dω
D(x, t) = ǫ(ω)Ẽ(x, ω)e−iωt.
−∞ 2π
We now use Z ∞
′
Ẽ(x, ω) = dt′ E(x, t′)e+iωt .
−∞
to write
1 Z∞ Z ∞
′
D(x, t) = dω ǫ(ω)e−iωt dt′ eiωt E(x, t′ ).
2π −∞ −∞
To display the non-locality, we write

ǫ(ω) = ǫ0 [(ǫ(ω)/ǫ0 − 1) + 1] = ǫ0 [χe + 1]

and thus
1 Z∞
( )
′
D(x, t) = ǫ0 E(x, t) + dωdt′ eiω(t −t) χe (ω)E(x, t′) .
2π −∞
By a change of variable τ = t − t′ , we can rewrite this as
 Z ∞ 
D(x, t) = ǫ0 E(x, t) + dτ G(τ )E(x, t − τ ) (7.44)
∞

where
1 Z
G(τ ) = dω χe (ω)e−iωτ . (7.45)
2π
We have essentially just used the convolution theorem of Fourier transforms, and
have exhibited the non-local connection between D and E.
Chapter 2 31

To explore the nature of this connection, we consider a simple one-resonance

model for χe (ω)
ωP2
χe (ω) = 2 ,
ω0 − ω 2 − iγω
where ωP is the plasma frequency. This has poles in the l.h.p. at
γ
ω = −i ± ν0
2
where
ν02 = ω02 − γ 2/4.

-i γ/2 − ν -i γ/2 + ν

To evaluate G(τ ) we use contour integration, noting that there are two cases
1. τ < 0: circle at ∞ vanishes in lower half plane.

2. τ > 0: circle at ∞ vanishes in upper half plane.

Thus G(τ ) vanishes for τ < 0. By the residue theorem
ωP2 X
G(τ > 0) = × 2πi ×
2π residues
Chapter 2 32

sin ν0 τ
= ωP2 e−γτ /2 ,
ν0
and thus
sin ν0τ
G(τ ) = ωP2 e−γτ /2 θ(τ ). (7.46)
ν0
We can make two observations

• There is an oscillatory frequency ≈ ω0 .

• The damping factor 1/γ is that of the oscillators.

Thus non-locality is confined to a region τ ≈ γ −1.

7.8.1 Causality

Because G(τ ) vanishes for τ < 0, D only depends on the values of E at earlier
times, i.e.  Z ∞ 
D(x, t) = ǫ0 E(x, t) + dτ G(τ )E(x, t − τ ) .
0
We can thus write the dielectric constant as
Z ∞
ǫ(ω)/ǫ0 = 1 + dτ G(τ )eiωτ .
0

Since G(τ ) is real, we have

ǫ(−ω) = ǫ∗ (ω ∗).
Furthermore, if G(τ ) is finite ∀τ , ǫ(ω)/ǫ0 is analytic in the upper half plane, since
integral is convergent there. We can therefore apply Cauchy’s theorem for any z
in the upper half plane
′
1 I ′ ǫ(ω )/ǫ0 − 1
ǫ(z)/ǫ0 = 1 + dω .
2πi ω′ − z
If we assume that ǫ falls off as fast as 1/ω 2, the contribution from the semi-circle
at infinity vanishes, and we have
′
1 Z∞ ′ ǫ(ω )/ǫ0 − 1
ǫ(z)/ǫ0 = 1 + dω .
2πi −∞ ω′ − z
Chapter 2 33

We now consider a pole just above the ω-axis, by writing z = ω + iδ. Then
1 1
!

′
= P ′
+ iπδ(ω ′ − ω)
ω − ω − iδ ω −ω
whence
1 Z∞ ǫ(ω ′ )/ǫ0 − 1
ǫ(ω)/ǫ0 = 1 + P dω ′ .
πi −∞ ω′ − ω
Thus taking the real and imaginary parts, we find
1 Z∞ ℑǫ(ω ′ )/ǫ0
ℜǫ/ǫ0 = 1 + P dω ′
π −∞ ω′ − ω
′
1 Z∞ ′ ℜǫ(ω )/ǫ0 − 1
ℑǫ/ǫ0 = − P dω (7.47)
π −∞ ω′ − ω
These are the Kronig-Kramer relations; they relate absorption (imaginary
part of ǫ) to dispersion (real part of ǫ) through analyticity.
Chapter 8

Wave Guides and Cavities

In this chapter we will consider propagation of waves in hollow, metal wave guides
and cavities.

• wave guide: ends are open

• cavity: ends are closed

8.1 Boundary Conditions at Surface of Conductor

Recall that at the boundary between two media, 1 and 2, we have

(H2 − H1 ) × n = K
(B2 − B1) · n = 0
(D2 − D1 ) · n = σ
(E2 − E1 ) × n = 0.

Inside a conductor, the electrons are completely free, with infinitely fast response,
such that B = E = 0.
Thus our boundary conditions just below the conducting surface reduce to

H ×n = K
B·n = 0

1
Chapter 2 2

D·n = σ
E × n = 0.

Thus just outside the surface of the conductor, we have that

• B is tangential to the surface.

• E is normal to the surface.

The case where we do not have a perfect conductor is discussed in detail in

Jackson, chapter 8.1. Note that in these cases we have energy losses associated
with the absorption at the boundary surface.

8.2 Propagation of Monochromatic Wave

We consider the propagation of monochromatic waves in a hollow cylinder, of

arbitrary cross section, which we take to be uniform along, say, the z-direction.
We assume a harmonic time dependence e−iωt , so that Maxwells equations become

∇ × E = iωB
∇·B = 0
∇ × B = −iµǫωE
∇·E = 0

Thus, in the usual way, these equations reduce to

 

E 
(∇2 + µǫω 2 ) 
 
=0 (8.1)
 B 


Because of the cylindrical symmetry in the problem, we expect to find waves

travelling in the positive or negative direction, or standing waves. Therefore we
look for solutions of the form
  
E(x, t)  
E(x, y) 
e±ikz−iωt.
  
=
B(x, t) 
  B(x, y) 

Chapter 2 3

Note: this does not mean that the propagation vector is in the z direction as such.
We now write
∇2 = ∇2T + ∇2z
where
∂2 ∂2
∇2T = +
∂x2 ∂y 2
2 ∂2
∇z = .
∂z 2
Then our wave equation eqn. (8.1) reduces to

[∇2T + (µǫω 2 − k 2)]E = 0 (8.2)

and similarly for B.

We now write E and B in terms of components parallel and transverse to z, i.e.
E = E z + E T etc., and show that it is only necessary to solve for the longitudinal
components Ez and Bz .
We start with two of Maxwell’s eqns

∇ × E = iωB
∇ × B = −iµǫωE. (8.3)

Writing the first of these in terms of longitudinal and transverse components, we

have
(∇T + ∇z ) × (E T + E z ) = iω(B T + B z ).
If we now consider the transverse and longitudinal components, we find

∇T × E T = iωB z (8.4)
∇T × E z + ∇z × E T = iωB T . (8.5)

From the second of these, we find

iω∇z × B T = ∇z × [∇T × E z + ∇z × E T ]
= ∇T [∇z · E z ] − ∇2z E T
Chapter 2 4

THen, using the z-dependence of E, B, we find

iω∇z × B T = ∇T [∇z · E z ] + k 2E T . (8.6)

To proceed further, we use the second equation of (8.3), which becomes

∇T × B T = −iµǫωE z
∇T × B z + ∇z × B T = −iµǫωE T .

Substituting in eqn. (8.6), we find

iω[−iµǫωE T − ∇T × B z ] = k 2E T + ∇T [∇z · E z ],

yielding

E T = (µǫω 2 − k 2)−1[∇T (∇z · E z ) − iωez × ∇T Bz ]

H T = (µǫω 2 − k 2)−1[∇T (∇z · H z ) + iǫωez × ∇T Ez ] (8.7)

with an analogous equation for B T .

Thus we can see that we have expressed the transverse components entirely in
terms of longitudinal componenets.

8.3 Classification of Modes

We have now shown that the propagation of the waves can be solved solely by
solving the two-dimensional wave equation
 

Ez (x, y) 
(∇2T + µǫω 2 − k 2) 
 
= 0, (8.8)
 Bz (x, y) 


subject to suitable boundary conditions. In the case of perfectly conducting walls

S, the boundary conditions are

n × E|S = 0
n · B|S = 0.
Chapter 2 5

It can be shown that these boundary conditions are equivalent to

Ez = 0 (8.9)
∂Bz
= 0. (8.10)
∂n
Thus we are in principle simultaneously solving two boundary-value equations
subject to each of the above conditions. However, in general the eigenvalue equa-
tion (8.2) will have different eigenvalues for the two different sets of boundary
conditions. Hence we cannot satisfy both simultaneously unless one is trivial.
Thus we classify the solutions as

Transverse Magnetic (TM)

Here Bz = 0 everywhere, and Ez = 0 on boundary. The differential equation

(8.8)a with the above Dirichlet boundary condition determines Ez in the wave
guide. If we know Ez , the transverse fields can be obtained from Eq. (8.7):
ik ~ iǫω ~ T Ez
ET = ∇T Ez , HT = ê3 × ∇ (8.11)
γ2 γ2

Transverse Electric (TE)

∂Bz
Ez = 0 everywhere, and ∂n = 0 on boundary. Hre we must solve the Eq. (8.8)b
with Neumann boundary condition. The transverse fields are
iµω ~ T Hz , HT = ik ∇ ~ T Hz
ET = − 2 ê3 × ∇ (8.12)
γ γ2
Finally, we must consider

Transverse Electric Magnetic (TEM)

Here we have Bz = Ez = 0 everywhere, so that the only non-trivial components

are those in the transverse direction. Then Maxwell’s equations reduce to

∇T × E TEM = 0
∇z × E TEM = iωB TEM.
Chapter 2 6

In addition, we have
∇T · E TEM = 0.
Combining the first and third of these equations, we find

∇2T E TEM = 0,

and comparing with the wave equation (8.2), we find

k 2 = µǫω 2 .

This is just the infinite-medium value. Similarly, we find

√
B TEM = ± µǫeZ × E TEM .

Thus we essentially have plane-wave propagation.

We see that E TEM obeys Laplace’s equation. Furthermore, the walls of the wave
guide are an equipotential. Thus the only solution inside a single, hollow perfect
conductor is the trivial one.

TEM modes cannot propagate inside a single conductor

They can, however, propagate inside a coaxial cable.

8.4 Modes of a Waveguide

We begin by discussing TM modes, for which we write

Ez = φ(x, y)e±ikz−iωt.

Then ψ satisfies
(∇2T + µǫω 2 − k 2 )ψ = 0,
subject to φ = 0 on the boundary.
We now introduce
γ 2 = µǫω 2 − k 2 ,
Chapter 2 7

so that our eigenvalue equation becomes

(∇2T + γ 2)φ = 0.

In general, the boundary equations require that γ 2 be positive, yielding a discrete

set of eigenvalues {γλ }, with corresponding wave number

kλ2 = µǫω 2 − γλ2 . (8.13)

If kλ2 > 0, kλ is real, and the propagation is oscillatory. If it is negative, the wave
number is imaginary and the wave will no propagate.
We define the cut-off frequency ωλ by
γλ
ωλ = √ (8.14)
µǫ
where

• ω < ωλ : wave cannot propagate

• ω > ωλ : wave can propagate

Finally, it is worth noting that the group velocity of the wave in the wave guide
is always smaller than the speed of light. We first note that we may write
√ q
kλ = µǫ ω 2 − ωλ2 .

We recall that the phase velocity

vp = ω/k
1 1
= √ q
µǫ 1 − ωλ2 /ω 2
c
= q
1 − ωλ2 /ω 2
which is always larger than the velocity of light, and diverges as ω → ωλ .
In contrast, the group velocity
!−1
dk q
vg = = c 1 − ωλ2 /ω 2 ,
dω
Chapter 2 8

which is always smaller than the infinite-space velocity of light, and vanishes as
ω → ωλ . In this limit the wave no longer propagates. Note that

vpvg = c2 .

8.5 Modes of a Rectangular Waveguide

a
For the sake of illustration, we will consider the case of TE modes. In Cartesian
coordinates, we have to solve the eigenvalue equation
∂2 ∂2
 

2
+ 2 + γ 2 ψ = 0
∂x ∂y
subject to
∂ψ(0, y) ∂ψ(a, y)
= = 0,
∂x ∂x
∂ψ(x, 0) ∂ψ(x, b)
= = 0.
∂y ∂y
This clearly has eigenfunctions for Hz
mπx mπy
! !
ψmn (x, y) = H0 cos cos
a b
Chapter 2 9

with eigenvalues
2
n2 
 
2 2 m
γmn =π + 2 .
a2 b
We denote the modes TEm,n . The lowest non-trivial mode is TE1,0 if a > b, with
cut-off frequency given by
2
γ10 = π 2 /a2 .
For this mode, for wave propagating in the positive direction, we have
πx ik1,0 z−iωt
!
Hz = H0 cos e .
a
We can obtain the transverse components of the field from eqn. (8.12)
ika πx ikz−iωt
!
HT = − H0 sin e ex
π a!
iωaµ πx ikz−iωt
ET = H0 sin e ey ,
π a
with k = k1,0.
The analysis of TM modes proceeds likewise. However, here the lowest propa-
gating mode is TM1,1, with a higher cut-off frequency. Wave guides are often
constructed such that TE1,0 is the only propagating mode. Recalling that
q
kλ = µǫ (ω 2 − ωλ2 )
√
we can show kλ / µǫω as follows:
Chapter 2 10

ω
TE1,0 TE0,1 TE1,1
TM1,1
ω
8.5.1 Energy Flux along Waveguide

The time-averaged energy flux is given by the real part of the Poynting Vector
1
S= E × H ∗.
2
Let us evaluate this for TE modes
1 1
S = E × H ∗ = (E T × H ∗T − Hz∗ ê3 × E T ).
2 2
Since Hz = ψ(x, y)e−iwt+ikz we get from Eq. (8.12)
ωkµ iωµ ωkµ ωµ
S= 4
∇T Hz ×(ê3×∇T Hz∗ )− 2 Hz∗ ê3 ×(ê3×∇T Hz ) = 4
ê3 |∇t ψ|2 −i 2 ψ ∗ ∇T ψ
2γ γ 2γ γ
Taking the real part, we get
ωkµ
ℜS = 4
|∇T ψ|2 ê3 .
2γ
Chapter 2 11

This is in the z-direction, and we see that energy propagation is along the waveg-
uide.
Similarly, for the TM wave Ez = φ(x, y)e−iωt+ikz one obtains
ωkǫ
ℜS = 4 |∇T φ|2 ê3 .
2γ
The total power transmitted by the TE wave is
Z ωkµ Z
P = ℜ S · ez dA = 4
dA (∇T ψ)∗ · (∇T ψ).
A 2γ
where A is a cross-section through the wave guide. Recalling Green’s identity, we
have Z I
∂ψ
∗ 2
(ψ ∇T ψ + ∇T ψ · ∇T ψ) dA = ψ ∗ dl.
∗
C ∂n
∂ψ
Because of the boundary conditions, either ∂n or φ (for the TM mode) vanish on
the surface. Thus
ωkµ Z ∗ 2
P = − 4 ψ ∇T ψ dA
2γ A
ωkµ 2 Z
= 4
γ |ψ|2 dA,
2γ A

using wave equation

(∇2T + γ 2)ψ = 0.
Thus we have 1/2
ω 2 ωλ2  Z ∗
! 
µ
P = √ 1− 2 ψ ψ dA, (8.15)
2 µǫ ωλ ω A

√
r
ω2
where we represented k as ω µǫ 1 − ωλ2 and γ 2 as µǫωλ2 ).
Similarly, for the TM modes we get
1/2
ωλ2 
!2 
µ ω Z
P = √ 1 − 2 φ∗ φ dA, (8.16)
2 µǫ ωλ ω A

From Chapter 7, we have that the field energy per unit length is given by
1Z 1Z
hU i = [ǫE · E ∗ + µH · H ∗ ] dA = [ǫE T · E ∗T + µH T · H ∗T + µHz · Hz∗ ] dA
4 4
µ 2 2
Z
2 2 µ Z
= 4
(µǫω + k ) [|∇T ψ| + µ|ψ| ] dA = 2
(µǫω + k + γ ) |ψ|2 dA
2 2 2
4γ 4γ
Chapter 2 12

where we have used the fact that |∇T ψ|2 = γ 2 |ψ|2 since ∇2T ψ = −γ 2ψ. Finally,
R R

we obtain
µ2 ǫω 2 Z 2 µ ω2 Z
hU i = 2
|ψ| dA = 2 |ψ|2 dA.
2γ 2 ωλ
Using eqns. (8.15) and (8.5.1), we find
1/2
ω2

1
P/U = √ 1 − λ2  ≡ vg (8.17)
µǫ ω

Thus we see that the energy propagates with the group velocity. N.B. you
should convince yourself that this expression has the correct dimension.
For the TM wave, we get
µ ω2 Z
hU i = 2 |φ|2 dA.
2 ωλ
yielding the same result (8.17) for group velocity.

8.6 Boundary Conditions at Surface of Good Conductor

At surface of infinitely good conductor, we have

n·B = 0
n×E = 0
n·D = Σ
n×H = K (8.18)

where Σ is the surface charge density.

Chapter 2 13

conductor

H, E
ξ

Hc , E c
n

In the case of a conductor of conductivity σ, we have

J = σE.
We cannot have a surface current, since that would imply an infinite tangential
E. Instead, we have
n × (H − Hc ) = 0.
where we use the subscript c to denote fields inside the conductor. (As σ → ∞,
we recover our surface current as a volume current over the thin layer close to the
boundary).
We obtain the results for finite conductivity by successive approximation. We
assume that initially E is perpendicular, and H parallel, to the surface just out-
side the conductor. Then H c |surface ≃ Hk , and Maxwell’s equations within the
conductor become
1 ∂H c
∇ × Ec + = 0
µc ∂t
∂Dc
∇ × Hc = J +
∂t
. If we assume harmonic time depedence, these reduce to
i
Hc = − ∇ × Ec
µc ω
Chapter 2 14

∇ × Hc = σEc − iωǫEc .

Thus if σ is sufficiently large, these reduce to

i
Hc = − ∇ × Ec
µc ω
1
Ec = ∇ × Hc .
σ
We now assume all variation to be normal to the surface. (Spatial variation of the
fields on the normal direction is much more rapid than in the parallel direction
so we can neglect ∇k in comparison to ∇T ). Then we have

∂
∇ = −n
∂ξ
and our equations become
i ∂Ec
Hc = n×
µc ω ∂ξ
∂ Hc
Ec = − σ1 n × ∂ξ .

We immediately see that n · Hc = 0, consistent with our boundary assumptions.

Furthermore, combining these two equations we obtain
i ∂ 2 Hc
Hc = − n × [n × ],
µc ωσ ∂ξ 2
yielding
∂2 2i
2
Hc + 2 Hc = 0,
∂ξ δ
where
2 1/2
!
δ≡ ,
µc ωσ
is the skin depth. Thus, combining this we the condition n · Hc = 0, we find

Hc = Hk e(i−1)ξ/δ . (8.19)
Chapter 2 15

Thus the magnetic field is tangential and falls off exponentially as we go into
the conductor. We can differentiate this, to obtain
µω
s
Ec = (1 − i)(n × Hk )e−ξ/δ eiξ/δ . (8.20)
2σ
Thus Ec is also tangential to the surface, but of much smaller magnitude.
We now go back to our boundary condition
n × (E − Ec) = 0.
Since E c has a small tangential component, so does E just outside the conductor.
µω
s
Ek = (1 − i)(n × Hk ).
2σ
Thus there is a non-zero component of the Poynting vector into the conductor,
and hence a net flow of energy, given by
dP 1
h i = ℜ [E × H ∗ ] · (−n)
da 2
µc ωδ
= |H k |2
4
1
= |H k |2.
2σδ
It can be demonstrated that this power is dissipated into heat as ohmic losses in
the skin of the conductor.
Applying this to our wave guide, we see that we have an energy loss/unit length
given by
dP 1 I 2 1 I
= − dl |H k | = − dl |n × H|2
dz 2σδ C 2σδ C
∂φ 2

1
1 ω
!2 I 

 2 2 (TM)
= dl  µ 1ωλ ∂n ω2  2
2
ωλ 2
2σδ ωλ C 

µǫω 2 1 − ω 2 |n × ∇T ψ| + ω 2 |ψ|
λ
(TE)
λ

8.7 Resonant Cavities

A resonant cavity differs from a wave guide in being closed. Thus, rather than
having wave propagation, we have standing waves.
Chapter 2 16

As before, we can have both TM and TE fields. However, now the z-dependence
is of the form, for the case of TM modes,

Ez = φ(x, y)[A sin kz + B cos kz]ez

Hz = 0

Then the transverse part of the wave is

1
ET = [∇ (∇ · Ez ) − iωez × ∇T Bz ]
γ2 T z
k
= 2 ∇T φ(x, y)[A cos kz − B sin kz].
γ
Now the boundary condition ET = 0 at z = 0, z = d yields A = 0, k = pπ/d and
thus
pπ
Ez = φ(x, y) cos (8.21)
d
pπ pπz
ET = − 2 sin ∇ φ. (8.22)
dγ d T
We can obtain HT similarly, yielding
iǫω pπz
HT = cos ez × ∇T φ. (8.23)
γ2 d
A corresponding analysis for the TE modes yields

Hz = ψ(x, y)(A sin kz + B cos kz)

Chapter 2 17

so ET = − iωµ
γ2
(A sin kz + B cos kz)ez × ∇T ψ. From the boundary conditions
E T |z=0,d = 0 we get
pπz
Hz = ψ(x, y) sin
d
iωµ pπz
ET = − 2 sin ez × ∇T ψ
γ d
pπ pπz
HT = cos ∇ ψ. (8.24)
dγ 2 d T
The function ψ(x, y) now satisfies the wave equation
 !2 
pπ
∇2T ψ 2
+ [µǫω −  ψ =0
d
where
p2 π 2
γ 2 = µǫω 2 − .
d2
We can solve this eigenvalue problem as for propatation along a wave guide, but
now the eigenvalues γλ determine not the cut-off frequencies but the allowed
frequencies:
2 2
 
2 1 γ 2 +
p π
ωλp = (8.25)
µǫ λ

d2
Chapter 2 18

Example: cylindrical cavity, radius R

z

R y

We work in cylindrical polar coords ψ(s, ϕ). Because of cylindrical symmetry, we

seek separable solutions to the two-dimensional wave equation of the form

ψ(s, ϕ) = ψ(s)e±imϕ

where m = 0, 1, 2, . . .. Then we have

∂2 m2 
 
1∂ 2
 + + γ − ψ(s) = 0.
∂s2 s ∂s s2
This is just Bessel’s equation (see last semester), with solution

ψ(s, ϕ) = Jm(γmns)e±imϕ .

In the case of a TM mode, where ψ(s, ϕ) = 0 at s = R, we have

γmn R = xmn,
Chapter 2 19

where xmn is the nth root of Jm(x) = 0. Thus the resonant requencies are given
by
2 2
 
2 1 x
 mn +
pπ
ωmnp = 2 2
 (TM mode). (8.26)
µǫ R d
The solution for TE modes is similar and the resonant frequencies are given by
′
2
p2 π 2 
 
2 1  xmn
ωmnp = + 2 (TE mode), (8.27)
µǫ R2 d

where x′mn is now the nth root of Jm

′
(x) = 0.
Note that for TM modes we have p = 0, 1, 2, . . . whilst for TE modes we have p =
1, 2, 3, . . .. Furthermore, the smallest x′mn < min(xmn), and thus for sufficiently
large d the dominant mode is
TE1,1,1.
We can compute the energy loss is a resonant cavity in a similar manner to that
for a wave guide.
Chapter 9

Radiating Systems

In this chapter, we will study radiation of varying current distributions. We will

begin by working in Lorentz gauge, where the equation for the vector potential is
2 1 ∂ 2A
∇ A − 2 2 = −µ0 J.
c ∂t
From Chapter 6, we recall that this has the retarded solution
µ0 Z 3 ′ ′
A= d x dt J(x′ , t′) × G(+) (x, t; x′ , t′ ),
4π
where
(+) ′ ′ 1 ′ |x − x′ |
G (x, t; x , t ) = δ(t − t + ).
|x − x′ | c
We now consider the case where the fields arise from a current with harmonic
time variation
J(x, t) = J(x)e−iωt
More general time dependence can be studied simply by taking the Fourier trans-
form. The potential corresponding to this current is then
µ0 Z 3 ′ ′ ′
A(x, t) = d x dt J(x′ )e−iωt G(+) (x, t; x′ , t′)
4π
= A(x)e−iωt,

with
µ0 Z 3 ′ 1 ik|x−x′ |
A(x) = d x J(x′ ) e .
4π |x − x′ |
1
Chapter 2 2

where k ≡ ω/c is the wave number.

d

We will now consider the form of the field a distance r away from a localised,
time-varying source of extent d. We begin by introducing the wavelength
2π 2πc
λ= ≡ ,
k ω
where λ ≫ d.
We now consider the form of the potential in three different regions:

1. d ≪ r ≪ λ - the near zone

Then exp ik|x − x′ | ∼ exp 2πir/λ ∼ 1, and we have
µ0 Z 3 ′ 1
A(x) ≃ d x J(x′ ) .
4π |x − x′ |
The field is of the familiar form which we can expand as a series in, say,
Legendre polynomials.

2. r ≫ λ ≫ d - the radiation zone

The the exponent is rapidly oscillating, and we can write
′
|x − x′ | = [x2 − 2x · x′ + x 2 ]1/2
′ 2
 
|x |
≃ r − n · x′ + O  .
r
Chapter 2 3

Thus, to leading order in 1/r we have

µ0 eikr
kr→∞
Z
′
A(x) −→ d3x′ J(x′ )e−ikn·x (9.1)
4π r
where n is a unit vector in the radial direction. Thus we have an outgoing
spherical wave. We can compute the magnetic and electric fields through
1
H = ∇×A (9.2)
µ0
iZ0
E = ∇×H
k
qµ
which also fall off as 1/r, corresponding to radiation. (Hereafter Z0 ≡ 0
ǫ0 =
µ0 c).
Since kn · x′ ≪ 1 - recall that d ≪ λ - we can expand the exponent in
eqn. (9.1) yielding
µ0 eikr X (−ik)n Z 3 ′
A(x) ≃ d x J(x′ )(n · x′ )n . (9.3)
4π r n n!
Successive terms are O((kd)n), which dies off with increasing n.

3. r ∼ λ Here we need to expand the solution in terms of the vector multipole

expansion, discussed in detail in Jackson, 9.6.

An analogous analysis for the scalar potential yields

Z
3 ′
Z
′ρ(x′ , t′ ) ′ |x − x′ |
φ(x, t) = dx dt δ(t + − t).
|x − x′ | c
Keeping the leading term yields
1
φ(x, t) ≃ q(t′ = t − r/c).
r
where q is the toal charge of the source. If the source is localised, and isolated,
no charge can flow in and out, and thus the total charge is constant in time- the
monopole part of the potential is static, i.e. has no time dependence.
Chapter 2 4

9.1 Electric Dipole Fields

If we keep only the leading term in eqn. (9.3), we have

µ0 eikr Z
A(x) = d3x′ J(x′ ). (9.4)
4π r
In fact, as discussed in Jackson, this is the leading l = 0 term in the vector
multipole expansion of the vector potential, and thus valid everywhere outside the
source as part of the multipole expansion. We will now show that this corresponds
to a dipole term. We begin by recalling the continuity equation
∂ρ
+∇·J =0
∂t
which with our assumed time dependence becomes

−iωρ + ∇ · J = 0.

We now use integration by parts to write

Z Z Z
3 ′ 3 ′ ′ ′
dxJ= d x (J · ∇ )x = − d3x′ x′ (∇′ · J)
Z
= −iω d3x′ x′ ρ(x′ ) = − iωp

enabling the potential to be expressed as

iµ0 ω eikr
A(x) = − p
4π r
where Z
p≡ d3x′ x′ ρ(x′ )
is the electric dipole moment.
The magnetic and electric fields are simply obtained from eqn. (9.2).
1 ck 2 eikr 1
H = ∇×A = (n × p) (1 − ) (9.5)
µ0 4π r ikr
iZ0 1 2 eikr 1 ik
E = ∇×H = [k (n × p) × n + (3(np)n − p)( 3 − 2 )eikr ]
k 4πǫ0 r r r
Chapter 2 5

In the spherical coordinates it takes the form (n × p = −p sin θφ̂)

pck 2 eikr 1
H = − sin θ (1 − )ϕ̂ (9.6)
4π r ikr
ikr
p 2e 1 ik
E = [−k sin θθ̂ + (2 cos θr̂ + sin θθ̂)( 3 − 2 )eikr ]
4πǫ0 r r r
It is interesting to examine their limiting forms

• Radiation Zone: r ≫ λ ≫ d:

ck 2 eikr ω2p eikr

H = (n × p) = − ϕ̂ sin θ
4π r 4πc r
ikr
µ0 e
E = Z0H × n = − ω 2 p sin θ θ̂
4π r
Both these field manifest clearly the characteristic properties of radiation:

– The fields fall off as 1/r.

– The electric and magnetic fields are normal to the direction of propaga-
tion n.

• Near Zone: λ ≫ r ≫ d:
Here the leading behaviour of the fields is given by
1 1
E = [3n(n · p) − p] 3
4πǫ0 r
1 i k
H = (n × p) 2 .
4πǫ0 Z0 r
Thus at very short distances, there is essentially an electric dipole field with
time dependence exp −iωt, and a magnetic field suppressed by kr/Z0 that
vanishes as k → 0

In order to show that this solution does indeed correspond to radiation, we will
look at the time-averaged power flux in the radiation zone. This, of course, is
Chapter 2 6

just given by the Poynting Vector, and we have

dP 1
= r2Re [n · E × H ∗ ]
dΩ 2
c2 Z0 4
= k |(n × p) × n|2 (9.7)
32π 2
There is a net flux of power away from the charge distribution, independent of r -
radiation. For the case where all components of p have the same phase, we have
the characteristic expression for dipole radiation,
dP c2 Z0 4 2 2
= k |p| sin θ
dΩ 32π 2

θ=0

The total power transmitted is just obtained by integrating eqn. (9.7) over the
unit sphere, and is independent of the phases of p:

c2 Z0 k 4 2
P = |p| .
12π

Centre-fed Linear Antenna

Once again we assume that the dimensions of the antenna are much less that the
wavelength. The antenna consists of two conductors of length d/2, along the z
Chapter 2 7

axis. The linear current density in the wires is

 
2|z| 
I(z) = I0 1 −
d
where we again suppress the time dependence.

ez
θ

This current flow gives rise to a line charge density Λ through the continuity
equation
∂I
iωΛ(z) = .
∂z
yielding
2iI0
Λ(z) = sgn(z).
ωd
This charge density has a non-zero dipole moment
Z d/2
2iI0
p = dz z ez
−d/2 ωd
iI0d
= ez .
2ω
N.B. if we had current flowing in opposite directions in the two arms of the
antenna, there would have been no dipole radiation term.
Thus, from eqn. (9.7), we see that this apparatus gives dipole radiation, with
power distribution
dP Z0 I02
= 2
(kd)2 sin2 θ
dΩ 128π
Chapter 2 8

Z0I02 (kd)2
P = . (9.8)
48π
If we identify the power radiated with energy dissipation through an effective
resistance, the coefficient of I02 /2 is eqn. (9.8) is the radiation resistance - the
factor of 2 arises from time-averaging, in the usual way.

9.2 Dipole Fields Revisited

In this section we’ll derive the formulas for the dipole radiation again - this time
without Fourier transformation dωe−iωt implied.
R

The general formulas for vector and scalar potentials due to an arbitrary source
are:
1 Z 3 ′ ρ(x′ , tr )
φ(x, t) = dx
4πǫ0 |x − x′ |
µ0 Z 3 ′ J(x′ , tr )
A(x, t) = dx (9.9)
4π |x − x′ |
|x−x′ |
where tr = t − c is the retarded time.
To study the behavior of these expressions in the radiation zone |x| ≫ |x′ |, we
choose the origin somewhere inside the radiating body and expand the denomi-
nators in a usual way:
1 1 n̂ · x′
= (1 − + ...) (9.10)
|x − x′ | r r
where r ≡ |x| and n̂ ≡ r̂ is the propagation vector for our would-be shperical
′
wave. We need also to expand the retarded time in powers of rr :
|x − x′ | r n̂ · x′
tr = t − ≃t− +
c c c
so that
′ ′ n̂ · x′
ρ(x , tr ) = ρ(x , t0 ) + ρ̇(x′ , t0 ) + ... (9.11)
c
Chapter 2 9

where t0 ≡ t− rc is the retarded time for our origin. The parameter of the expansion
(9.11) is λd ≪ 1 (see previous Section). Indeed, ρ̇ ∼ ωcharρ where ωchar are the
characteristic frequencies of the emitted radiation, hence dcρρ̇ ∼ dω d
c = λ ≪ 1. )
Substituting the expansions (9.10) and (9.11) in the expression (9.9), one obtains:
1 Z 3 ′ ′ n̂ · x′ ′ n̂ · x′
φ(x, t) = d x [ρ(x , t0 ) + ρ̇(x , t0 )](1 − + ...)
4πǫ0 r c r
Q n̂ · p(t0) n̂ · ṗ(t0)
= + + + ...
4πǫ0 r 4πǫ0 r2 4πǫ0 rc
For the vector potential in Eq. (9.9), the first term in the expansions (9.10) and
(9.11) is sufficient:
µ0 Z 3 ′ J(x′ , tr ) µ0 Z 3 ′
A(x, t) = dx ′
≃ d x J(x′ , t0)
4π |x − x | 4πr
In the previous Section, we demonstrated that
Z
d3x′ J(x′ , t) = ṗ(t)

so the dipole potentials in the radiation zone take the form

1 Z 3 ′ n̂ · x′ n̂ · x′
φ(x, t) = d x [ρ(x′ , t0 ) + ρ̇(x′ , t0 )](1 − + ...)
4πǫ0 r c r
Q n̂ · p(t0) n̂ · ṗ(t0)
= + +
4πǫ0 r 4πǫ0 r2 4πǫ0 rc
′
µ0 Z
J(x , tr ) µ0 ṗ(t0)
A(x, t) = d3 x ′ ≃ (9.12)
4π |x − x′ | 4πr
Next we calculate the electric and magnetic field in the radiation zone. Discarding
terms ∼ r12 , one obtains after some algebra (note that ∇f (t0) = f˙(t0)∇t0 and
∇t0 = − n̂c :
n̂
∇φ(x, t) = − (n̂ · p̈(t0))
4πǫ0 rc2
∂ µ0 p̈(t0 ) µ0
A(x, t) = , ∇×A=− n × p̈(t0)
∂t 4πr 4πrc
Chapter 2 10

Thus, the dipole fields in the radiation zone are

µ0 µ0
E(x, t) = [n̂(n̂ · p̈(t0 )) − p̈(t0 )] = n̂ × (n̂ × p̈(t0))
4πr 4πr
µ0 n̂
B(x, t) = − p̈(t0) = E(x, t) (9.13)
4πcr c
If we choose the frame with OZ axis collinear to p̈(t0 ), the fields take the form
µ0 p̈(t0 ) sin θ µ0 p̈(t0) sin θ
E(r, θ, ϕ) = θ̂, B(r, θ, ϕ) = ϕ̂, (9.14)
4π r 4πc r
The Poynting vector is then
2
1 µ0 2 sin θ
S = E×B = (p̈(t0 )) n̂
µ0 16π 2 c r2
⇒ the total radiated power takes the form
Z µ0
P = S · n̂dA = (p̈(t0))2 (9.15)
6πc
For a single point charge q p(t) = qx(t) so we get the Larmor formula
µ0 q 2a2
P = (9.16)
6πc
Later, we will reobtain Larmor formula using the Lenard-Wiechert potentials of
the moving point charge.

9.3 Magnetic dipole and Electric Quadrupole Radiation

The next term in the multiple expansion is

µ0 eikr 1 Z !
A(x) = − ik d3 x′ J(x′ )n · x′ ,
4π r r
where the additional term is to ensure the expansion is valid at all distances. To
exhibit the form of this potential, we express the integrand as pieces symmetric
and anti-symmetric in J and x′ , by writing
1 1
(n · x′ )J = [(n · x′ )J + (n · J)x′ ] + (x′ × J) × n. (9.17)
2 2
Chapter 2 11

We now introduce the magnetisation density

1
M = x × J.
2
Then the second term gives rise to a vector potential
ikµ0 eikx 1
!
A(x) = 1− n × m, (9.18)
4π r ikr
where m is the magnetic dipole moment.
As an example of magnetic dipole radiation, consider the circular loop of radius
b with current
I(t) = I cos ωt = ℜIe−iωt
The magnetic dipole moment of this loop oscillates in time as

m(t) = m cos ωt = ℜπb2 Ie−iωt

Let us calculate the magnetic vector potential due to this setup. W.l.o.g. we
can assume that the point x lies in the XZ plane. The general formula for the
magnetic vector potential has the form
′
µ0 I ′ e−iωtr
A(x, t) = dl ′
êφ′ Ie−iωt (9.19)
4π |x − x |
x·x′ r̂·x′
Expanding tr′ ≃ t0 − c (where t0 = t − rc ) and |x−1x′ | ≃ 1r (1 + r ) we get

µ0 bI eikr Z 2π ′ b ′
A(x) = dφ (−ê1 sin φ′ + ê2 cos φ′ )(1 + sin θ cos φ′ )e−ikb sin θ cos φ
4π r 0 r
Since kb = 2π λb ≪ 1 we can expand the exponential in the r.h.s. of this equation
and get
µ0 bI eikr Z 2π ′ b
A(x) = dφ (−ê1 sin φ′ + ê2 cos φ′ )(1 + sin θ cos φ′ − ikb sin θ cos φ′ )
4π r 0 r
Performing integration over φ′ we obtain
ikµ0 Ib2 1 ikr
A(x) = ê2 (1 − )e sin θ (9.20)
4r ikr
Chapter 2 12

For our setup ê2 = êφ so the final result for the vector potential takes the form
ikµ0 m̂ 1 ikr
A(x) = êφ (1 − )e sin θ (9.21)
4πr ikr
which coincides with Eq. (9.18).
Let us find now electric and magnetic fields of the magnetic dipole radiation.
Taking the curl of Eq. (9.18), we find
eikr
 
1  2 1 ik ikr 
!
H= k (n × m) × n + [3n(n · m) − m] − e . (9.22)
4π  r r3 r2
The field H due to the magnetic dipole is of the same form as the field E due to
the electric dipole (see Eq. (9.5)). Similarly we have
Z0 2 eikr 1
!
E = − k (n × m) 1− , (9.23)
4π r ikr
so that the electric field due to a magnetic dipole is of the same form as the
magnetic field due to an electric dipole:
ǫ0 m µ0 m
Hmag.dipole = Eel.dipole, Emag.dipole = Hel.dipole ,
p p
Since the radiated power is proportional to n · (E × H),

mag.dipole m2 el.dipole µ0 m2 ω 4
Prad = P = (9.24)
p2 c2 rad 12πc3
In order to get an estimate of the relative strength of the electric and magnetic
dipole radiation, consider a physical dipole made from two charges q and −q
separated by distance d which rotate with angular velocity ω around the center
of the dipole. The magnetic moment of this system can be approximated by an
d2 ω
oscillating current I = Tq = 2πq
ω
so we get an oscillating magnetic moment m = 8
.
The ratio of powers for this example is
Pmag ω 2 d2 v2
∼ = (9.25)
Prmel 4c2 c2
Chapter 2 13

where v is the linear velocity of the rotating charges. We see that for charges
moving with non-relativistic velocities the electric dipole radiation is the most
important part while the magnetic dipole radiation is of the size of the relativistic
corrections.
The interesting part is the quadrupole moment, obtained from the symmetric
part of eqn. (9.17). We use
1Z 3 ′ ′ ′ iω Z 3 ′ ′
d x {(n · x )J + (n · J)x } = − d x ρx (n · x′ ),
2 2
using the same tricks we encountered earlier, and write
µ0 ck eikr 1
!Z
A(x) = − 1− d3 x′ρ(x′ )x′ (n · x′ ). (9.26)
8π r ikr
In the limit r ≫ λ, we find
H = ikn × A/µ0
E = ikZ0(n × A) × n/µ0 . (9.27)
If we now recall our expression for the quadrupole moment
Z
Qαβ = d3xρ(x)(3xαxβ − r2 δαβ )
then we find that H can be written
ick 3 eikr
H=− n × Q(n)
24π r
where Q(n) is defined by
X
Qα = Qαβ nβ .
β
The power dissipation is
dP c2 Z0 6
= 2
k |[n × Q(n)] × n|2 .
dΩ 1152π
We encountered a simple model of a quadrupole moment in the multipole expan-
sion last term:
Q33 = Q0
1
Q11 = Q22 = − Q0, (9.28)
2
Chapter 2 14

which is clearly traceless. Then the angular power distribution is

dP c2 Z0 k 6 2 2
= 2
Q0 sin θ cos2 θ, (9.29)
dΩ 512π
and the total power radiated is
c2 Z0 k 6 Q20
P = . (9.30)
960π
For quadrupole radiation, we have a four-lobe pattern of power distribution

The complete description requires the full multipole expansion which is beyond
what I am going to do in this course.

9.4 Radiation from a moving point charge

9.4.1 Lenard-Wiechert Potentials

Consider a point charge moving along the trajectory r = w(t).

~ What are the
electric and magnetic fields due to this charge?
Chapter 2 15

As usually, it is convenient to start with the potentials. In the Lorentz gauge

1 Z 3 ′ ρ(r′ ) ′ |r − r′ |
φ(r, t) = dx δ(t − t + )
4πǫ0 |r − r′ | c
µ0 Z 3 ′ J(r′ ) ′ |r − r′ |
A(r, t) = dx δ(t − t + ) (9.31)
4π |r − r′ | c
For a point charge

ρ(r, t) = qδ(r − w(t)),

~ J(r, t) = qv(t)δ(r − w(t))
~

At first, let us find the scalar potential

q Z 3 ′Z ′ δ(r′ − w(t)) ~ ′ |r − r′ |
φ(r, t) = d x dt δ(t − t + )
4πǫ0 |r − r′ | c
|r −r′ |
q Z ′ δ(t′ − t + c ) Z ′ 1 δ(t′ − tr )
= dt = dt
4πǫ0 ~ ′ )|
|r − w(t ∂
(t′ − t + |r −w(t
~ ′ )|
) ~ ′ )|
|r − w(t
∂t ′ c t′ =tr
q Z ′ δ(t′ − tr ) c
= dt 1 =
4πǫ0 ~ ′ )| − c v(t′ ) · (r − w(t
|r − w(t ~ ′ ) c|r − w(t
~ r )| − v(tr ) · (r − w(t
~ r ))
∂
where v(t) ≡ ∂t w(t)
~ is the velocity of the particle and tr is the solution of the
equation c(t − tr ) = |r − w(t
~ r )| = 0.
Similarly,
µ0 q c
A(r, t) = v(tr )
4π c|r − w(t
~ r )| − v(tr ) · (r − w(t
~ r ))
The potentials

q c
φ(r, t) =
4πǫ0 c|r − w(t
~ r )| − v(tr ) · (r − w(t
~ r ))
v
A(r, t) = 2 V (r, t) (9.32)
c
are called the Lenard-Wiechert potentials for a point charge. The corresponding
electric and magnetic fields are (see Jackson or Griffiths)
Chapter 2 16

q ς
E(r, t) = 3
[~u(c2 − v 2 ) + ~ς × (~u × a)]
4πǫ0 (~ς · ~u)
ςˆ
B(r, t) = × E(r, t) (9.33)
c
~ς
where v = v(tr ), a = a(tr ), ~ς ≡ r − w(t
~ r ), and ~u ≡ cˆ
ς − v(tr ) (as usually, ςˆ ≡ |~ς| ).

9.4.2 Power radiated by a point charge

The electric and magnetic fields due to a point charge moving along an arbitrary
trajectory w(t)
~ are given by Eq. (9.33)
q ~u(c2 − v 2) q ςˆ × (~u × ~a)
E(r, t) = +
4πǫ0 ς 2 (ˆ ς · ~u)3 4πǫ0 ς ς · ~u)3
(ˆ
~ς
B(r, t) = × E(r, t). (9.34)
c
where ~ς = ~r − w(t
~ r ), ~u = cˆς − ~v , and tr is defined as a solution to the equation
c(t − tr ) = ς. As usuallly, velocity and acceleration in Eq. (9.34) are taken at
t = tr . The first term (∼ ~u) is called the velocity field and the second (∼ ~a) is
called the acceleration or the radiation field.
The Poynting vector is
~= 1E
S ~ ×B~ = 1 E ~ × (ˆ ~ = 1 [E 2ςˆ − (ˆ
ς × E) ~ E]
ς · E) ~ (9.35)
µ0 µ0 c µ0 c
Some of the energy is radiation; another part is just a field energy carried along
by the particle as it moves. To calculate the power radiated by the particle at
time t∗ , we draw a large sphere with radius ς = R, wait for t − t∗ = Rc , and
integrate Poynting vector over the surface. Since the velocity field is ∼ 1/R2
the corresponding Prad is ∼ R2 R14 = R12 so it does not contribute to the radiated
power at large R. The power due to the acceleration field (∼ 1/R) is finite:
Prad ∼ R2 R12 = 1. We get

~ rad(~r, t) = q ςˆ × (~u × ~a)

E
ς · ~u)3
4πǫ0 ς (ˆ
Chapter 2 17

→ ~ς × E ~rad = ςˆ E 2 .
~ rad(r, t) = 0 ⇒ S (9.36)
µ0 c rad
For simplicity, consider the charge which is instantaneously at rest at t = t∗ . Since
~v(t∗ ) = 0, ~u(t∗) = cˆ
ς so the Eq. (9.36) reduces to

~ ςˆ µ0 q 2 2 2 µ0 q 2 a2 sin2 θ
Srad = ( ) [a − (ˆ
ς · ~a) ] = ςˆ (9.37)
µ0 c 4πR 16π 2 c R2
The total power is given by the following Larmor formula
I 2 2 Z
~ = µ0 q a
~rad · da sin2 θ 2 µ0 q 2 a2
Prad = S R sin θdθdφ = (9.38)
S 16π 2 cR2 R2 6πc
which we have already obtained using the electric dipole radiation, see the Eq.
(9.16).
We have derived the Larmor formula under the assumption that v = 0 but one can
demonstrate that it holds true as long as v ≪ c. In the general case of arbitrary
velocity, the radiation is given by the Lienard formula
µ0 q 2 γ 6 2 (~v · ~a)2
Prad = (a − ) (9.39)
6πc c2
r
v2
where γ ≡ 1/ 1 − c2 .

9.4.3 Electromagnetic fields due to a point charge moving with con-

stant velocity.
Potentials

For a point charge moving with constant velocity v the trajectory is w

~ = tv so
the retarded time is

c(t − tr ) = |r − tr v| ⇒ r2 − 2tr r · v + v 2t2r = c2 (t2 − 2ttr + t2r )

q
c2 t − r · v − (c2 t − r · v)2 − (c2 − v 2 )(c2t2 − r2 )
⇒ tr =
c2 − v 2
Chapter 2 18

The Lenard-Wiechert potentials (9.32) take the form

qc 1
φ(r, t) = (9.40)
4πǫ0 c|r − tr v| − v · (r − tr )
qc qc −1/2
= 2 2 2
= [(c2t − vr · v)2 − (c2 − v 2)(c2t2 − r2 )]
4πǫ0 [c t − (c − v )tr − r · v] 4πǫ0
and
v µ0 qv 2
A(r, t) = 2
φ(r, t) = [c t − (c2 − v 2 )tr − r · v]−1
c 4π
µ0 qcv 2 −1/2
= [(c t − vr · v)2 − (c2 − v 2 )(c2t2 − r2)] (9.41)
4π
Let us demonstrate that φ(r, t) can be rewritten as
q v 2 2 −1/2
φ(r, t) = (1 − 2 sin θ) (9.42)
4πǫ0 R c
where R = r − tv and θ is the angle between R and v. (R is the distance to the
position of the moving charge at the time of measurement of the fields).
We have

(c2t − v · r)2 − (c2 − v 2)(c2t2 − r2 ) = [c2 t − v · (R + tv)]2 − (c2 − v 2)[c2t2 − (R + tv)2 ]

= [(c2 − v 2 )t − v · R]2 − (c2 − v 2)[(c2 − v 2 )t2 − 2tv · R − R2 )2 = (c2 − v 2)R2 + (v · R)2

and therefore
v
v2 2
u
q u
(c2t − v · r)2 − (c2 − v 2)(c2t2 − r2 ) = Rc 1 −
t
sin θ (9.43)
c2

Fields

For w
~ = tv (and a = 0) the electric field in Eq. (9.33) reduces to (recall ~u ≡
cˆ
ς − v(tr ) = cˆ
ς −v )
q ς~u 2 2 q(c2 − v 2) c~ς − ςv
E(r, t) = (c − v ) = (9.44)
4πǫ0 (~ς · ~u)3 4πǫ0 (cς − v · ~ς)3

and B(r, t) = ~ςc × E(r, t).

Chapter 2 19

It is easy to see that (~ς ≡ r − w(t

~ r ) = r − vtr = R + v(t − tr ))

c~ς − vς = c(r − tr v) − |r − tr v|v = c(r − tr v) − c(t − tr )v = (r − tv)c = cR

Similarly, we get

cς − v · ~ς = c2 (t − tr ) − v · [R + (t − tr )v] = (c2 − v 2 )(t − tr ) − v · R

and

[cς − v · ~ς]2 = [(c2 − v 2)(t − tr ) − v · R]2

= (c2 − v 2)2(t − tr )2 − 2(c2 − v 2 )(t − tr )v · R + (v · R)2
2 2 2 2 2 2 v2 2
= R (c − v ) + (v · R) = R c (1 − 2 sin θ) (9.45)
c
so the electic and magnetic fields (9.46) take the form
qc ς~u
E(r, t) = 3
(c2 − v 2) (9.46)
4πǫ0 (~ς · ~u)
2
q(c2 − v 2 ) cR q R̂ 1 − vc2
= =
4πǫ0 (R2 c2 − R2v 2 sin2 θ)3/2 4πǫ0 R2 (1 − v22 sin2 θ)3/2
c
ςˆ 1
B(r, t) = = × E(r, t) = 2 v × E(r, t)
c c
Let us demonstrate that the fields (9.46) are Lorentz transforms of the usual
q R̂
Coulomb field of a point charge (E(r, t) = 4πǫ0 R2 , B = 0).
Chapter 11

Special Theory of Relativity and

Covariant Electrodynamics

Central to Newtonian Mechanics is the concept of an inertial frame; a frame in

which a body, acted on by no external forces, moves with a constant velocity. A
transformation between two inertial frames is known as a Galilean Transfor-
mation.
Aside: a practical definition of an inertial frame is one moving with constant
velocity relative to the distant stars (Mach’s principle).

11.0.4 Galilean Transformations

Consider two inertial frames K, K ′, moving with a relative constant velocity v.

The coordinates in the two frames are related by

t′ = t
x′ = x − vt (11.1)

Now consider the interactions of an ensemble of N particles at positions xi; i = 1, . . . , N,

acting solely under the influence of a central potential Vij (|xi − xj |). Then the
eqn. of motion of particle i in K is
dv i X
mi = − ∇xi Vij (|xi − xj |).
dt j

1
Chapter 2 2

Suppose now that we look at the equation of motion in K ′ . Then we have v′i =
vi − v, and
dv
mi i = − ∇x′i Vij (|x′i − x′j |).
X

dt j
Now under eqn. (11.1),
∂ ∂
′ =
∂xi ∂xi
and we have
|x′i − x′j | = |xi − xj |,
and we see that the eqn. of motion in K ′ is of exactly the same form as that
in K - we say that classical Newtonian mechanics transforms covariantly under
Galilean Transformations.

11.0.5 Maxwellian Mechanics under Galilean Transformations

We have seen that electric and magnetic propagation in a vacuum satisfies the
wave equation
2
 
∇2 −
1 ∂  ψ(x, y, z; t) = 0. (11.2)
c2 ∂t2
Let us now consider the transformation of this equation under eqn. (11.1). We
have
∂ ∂x′j ∂ ∂t′ ∂
= +
∂xi ∂xi ∂x′j ∂xi ∂t′
∂ ∂
= δij ′ + 0 = ′
∂xj ∂xi
∂ ∂x′j ∂ ∂t′ ∂
= +
∂t ∂t ∂x′j ∂t ∂t′
∂ ∂
= −vi ′ + ′ .
∂xi ∂t
Thus the wave equation (11.2) becomes
1 ∂ ∂
" ! !#
′2
∇ − 2 ′
− v · ∇′ ′
− v · ∇′ ψ = 0
c ∂t ∂t
Chapter 2 3

1 ∂2
 
′2 2 ′ ∂ 1
i.e. ∇ −
2 ′2
+ 2
v · ∇ ′
− 2
(v · ∇′ )(v · ∇′ ) ψ = 0 (11.3)
c ∂t c ∂t c
This equation is clearly of a different from to equation (11.2). The wave equation
does not transform covariantly under Galilean Transformations. For sound waves
there is no problem; they propagate in a medium, and it is natural to formulate
the wave equation in a frame in which the medium is at rest. Thus the natural
question arose - Is there a frame in which the “ether” is at rest”?. Of course, we
all know the answer (Michelson-Morley) that the velocity of light is the same in
all frames, and the resolution of this nasty transformation property is the Special
Theory of Relativity.

11.1 Postulates of Special Theory of Relativity

1. The same laws of nature hold in all systems moving uniformly with respect
to one another.

2. The velocity of light has the same value in all systems moving uniformly
with respect to each other, independent of velocity of observer relative to the
source.

11.2 Lorentz Transformations and Kinematic Results of

Special Relativity

We will now derive the relationship between coordinates in two frames K, K ′

moving with constant velocity v relative to one another. W.l.o.g. we will let the
origin of the coordinates coincide at t = t′ = 0.
We suppose that a flashlight is rapidly switched on and off at the origin at t =
t′ = 0. Then, by postulate 2, observers in both K and K ′ see a spherical shell of
radiation expanding with the velocity of light c. The wavefront satisfies
In K: c2 t2 − (x2 + y 2 + z 2 ) = 0
Chapter 2 4

In K ′: c2 t′2 − (x′2 + y ′2 + z ′2 ) = 0

Thus we see that, under such a transformation, the quantity c2 t2 −(x2 +y 2 +z 2 ) = 0

remains invariant. The emission of the light, and its subsequent absorption at
some later times, are each events. We have considered the case where the events
are separated by something travelling at the speed of light. More generally, we
have

∆s2 = c2 t2 − (x2 + y 2 + z 2 ) (11.4)

is invariant under transformations between inertial frames. This is the interval

between the two events.
To consider the form of the transformations satisfying eqn. (11.4), we will spe-
cialise to the case where the axes in K, K ′ are parallel, and the frames are moving
with a relative velocity v = ve3 . Because the transformations must reduce to
Galilean transformations in the limit of small relative velocities, we need consider
only the linear relations
t′ = a1 t + b1z
z ′ = a2 t + b2z
x′ = x
y′ = y (11.5)
The transverse dimensions do not change (see the gedanken experiment of Taylor
and Wheeler discussed in Griffiths textbook).
Because the frames are moving with relative velocity v, we have that the event
z ′ = 0 corresponds to z = vt, yielding
a2 = −vb2.
We now impose invariance of ∆s2 :
c2 t2 − (x2 + y 2 + z 2 ) = c2 (a1 t + b1 x)2 − (a2t + b2x)2 − y 2 − z 2 ,
Chapter 2 5

which we can expand as

c2 t2 [1 − a21 + a22 /c2] − z 2 [1 + b21c2 − b22 ] + 2zt[a2 b2 − c2 a1 b1 ] = 0.
This is true ∀x, t, so equating the coefficients to zero yields
a21 − a22 /c2 = 1
b22 − c2 b21 = 1
a2 b2 = c2 a1 b1 .
Solving these simultaneous equations we find
v
ct′ = γ[ct − z]
c
v
z ′ = γ[z − ct]
c
′
x = x
y′ = y
where
1
γ= q. (11.6)
1 − v 2/c2
We can write this in an axis-independent form as

ct′ = γ(ct − βxk) 




x′k = γ(xk − βct) 
(11.7)

x′ ⊥ = x⊥




where
β = v/c
γ = (1 − β 2)−1/2
x·v
xk = . (11.8)
|v|
In vector form, this is
ct′ = γ(ct − β · x)
γ−1
x′ = x + (β · x)β − γβct. (11.9)
β2
Chapter 2 6

Alternative Parametrisation

Introduce β = tanh ζ, so that γ = cosh ζ. Then, for frames moving parallel to the
x axis, we have

ct′ = ct cosh ζ − z sinh ζ

z ′ = z cosh ζ − ct sinh ζ, (11.10)

which has the form of a “rotation” on the complex angle φ = iζ

Chapter 2 7

11.3 Kinematical Properties of Lorentz Transformations

Given two events (ct1 , x1 ) and (ct2 , x2), Lorentz transformations leave the interval

∆s2 = c2 (t2 − t1 )2 − (x2 − x1)2

invariant. Thus we can classify the interval by the sign of δa2 , as follows

• ∆s2 < 0. This is timelike separation. We have c|t2 − t1 | > |x2 − x1 |, so that
the two points can communicate by a signal travelling at less than the speed
of light, and indeed a frame can be chosen such that |x2 − x1| = 0.

• ∆s2 = 0. This is lightlike separation. We have c|t2 − t1 | = |x2 − x1 |, so that

the two points can only be connected by a signal travelling at the speed of
light.

• ∆s2 < 0. This is spacelike separation, with c|t2 − t1 | < |x1 − x2 |. The two
space-time points cannot communicate, and indeed a frame exists in which
t1 = t2 .

11.3.1 Light Cone

Points that can be connected with the space-time origin by a light signal are said
to lie on the light cone.
Chapter 2 8

ct

Points within the light cone can be causally connected with the origin, whilst
those outside cannot. The forward (ct > 0) and backward (t < 0) cones define
absolute future and absolute past, and the ordering is preserved under Lorentz
transformations.

11.3.2 Simultaneity, Length Contraction and Time Dilation

Consider a rocket moving with constant velocity v along the x direction relative
to the lab frame K. Let us denote the rest frame of the rocket by K ′. We assume
that the axes of the frames are parallel, and the origins coincide at t = 0.
On the side of a rocket is a meter rule. We also have, in the lab. frame, a high
density of observers, each with a very accurate clock synchronised in the frame
K.
Chapter 2 9

y y’

v
111111111
000000000

11111111
00000000 x
00000000
11111111
00000000
11111111 x’

z z’

Simultaneity

At time t, an observer in the lab frame, co-incident with one end of the meter
rod, records his position (ct, x1 ), and an observer coincident with the other end
does likewise (ct, x2 ). Thus (ct, x1) and (ct, x2) denote two events, which are
simultaneous in the lab. frame.
In the rocket rest frame K ′ we have

ct′1 = γ(ct − βx1)

x′1 = γ(x1 − βct)

ct′2 = γ(ct − βx2)

x′2 = γ(x2 − βct) (11.11)

We immediately see that t′1 = t′2 iff x1 = x2; in general the points as not simula-
taneous in the rocket rest frame.
Chapter 2 10

Length Contraction

In the rocket frame, our meter rule has length x′2 −x′1 . However, from eqn. (11.11),
we see that in the laboratory frame the length is given by

x′1 − x′2 = γ(x1 − x2),

i.e.

x′1 − x′2
x1 − x2 =
γ

Since γ ≥ 1, we have that lengths are contracted

Time Dilation

We now imagine that the clocks in K, K ′ are synchronised at t1 = t′1 = 0 as the

rocket passes origin in frame K. An observer at some point x in K records the
time t2 at which rocket passes x, and an observer in K ′ records time t′2 at which
he passes the laboratory observer. The rocket observer is always at x′2 = 0, so we
have

0 = γ(x − βct2 )
=⇒ x = βct2

From the third eqn. of (11.11), we have

ct′2 = γ(ct2 − βx) = γ(ct2 − β 2ct2 )

t2
t′2 = .
γ

Thus we see that time is dilated.

Chapter 2 11

11.4 Proper Time

We now generalize the discussion to the case where the rocket is moving with a
velocity v(t) along some path relative to the lab frame K. We will now introduce
K ′ as the instantaneous rest frame of the rocket.
Consider two closely separated points on the trajectory, with coordinates in the
two frames {(ct, x), (c[t+dt], x+dx)} and {(ct′ , x′ ), (c[t′+dt′ ], x′ +dx′ )} respectively.
The interval between the points is the invariant, and we have
c2 dt′2 − dx′2 = c2 dt2 − dx2 .
But dx′ = 0 in k ′ , and furthermore dx2 = v2 dt2 , and thus
q
cdt′ = cdt 1 − β(t)2,
where
v(t)
. β(t) =
c
Then the elapsed time in the rocket between two events is
Z t
2
q
t′2 − t′1 = dt 1 − β(t)2 < t2 − t1 .
t1
The proper time τ is the elapsed time in the frame in with the object is at rest.
Thus
cdτ = ds
where ds is the interval introduced earlier. In this case we have
q
dτ = dt 1 − β(t)2. (11.12)
Note that proper time can only be defined for time-like quantities.

11.5 Addition of Velocities

Suppose now that a projectile is fired with velocity u′ from the rocket, relative to
the rocket. Then the co-ordinates of the projectile in K ′ satisfies
dx′
u′ = ′ .
dt
Chapter 2 12

while in K we have
dx
u= .
dt
Using the inverse Lorentz transform we have

xk = γv [x′k + βct′ ]
dx′k dt′ ′
 
dxk dt
=⇒ uk ≡ = γv  ′ + βc 
dt dt dt dt
′ ′
dxk
 
dt
= γv  ′ + βc ,
dt dt
where we use k to denote the component along v. We also have

ct = γ[ct′ + βx′k]
dt′ ′
 
′ dt 
=⇒ c = γc c + βuk

dt dt
dt′ c
=⇒ =
dt γv [c + βu′k]
Combinding these two results, we find
u′k + v
uk = (11.13)
1 + βu′k/c
Similarly
dx⊥ dx′ dt′
u⊥ = = ⊥′ · ,
dt dt dt
yielding
u′⊥
u⊥ = . (11.14)
γ(1 + βu′k/c)
In vector notation, this becomes
u′k + v
uk =
1 + v · u′ /c2
u′⊥
u⊥ = (11.15)
γ(1 + v · u′ /c2)
As expected, this reduces to the Galilean result u = u′ + v for the case u′, v ≪ c.
Chapter 2 13

11.6 Special Relativity and Four Vectors

We can formulate this picture in a much more convenient fashion through the
introduction of four vectors. To see how these work, let us return briefly to
Galilean transformations, and rotations in Euclidean space.

11.6.1 Vectors, Tensors and Rotations in R3

Consider two co-ordinate systems P , P ′ whose origins coincide, but which are
related by rotation through an angle θ.

z, z’
1
0
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
0
1
00000000000
11111111111
0
1
00000000000
11111111111
y’
0
1
00000000000
11111111111
0
1
00000000000
11111111111
0000000000000000
1111111111111111
0
1
00000000000
11111111111
0000000000000000
1111111111111111
0
1
y
00000000000
11111111111
0000000000000000
1111111111111111
0
1
00000000000
11111111111
0000000000000000
1111111111111111
0
1
00000000000
11111111111
0000000000000000
1111111111111111
0
1
00000000000
11111111111
0000000000000000
1111111111111111
0
1
00000000000
11111111111
0000000000000000
1111111111111111
0
1
00000000000
11111111111
0000000000000000
1111111111111111
0
1
00000000000
11111111111
0000000000000000
1111111111111111
0
1
00000000000
11111111111
0000000000000000
1111111111111111
0
1
00000000000
11111111111
00000000000000000
11111111111111111
0000000000000000
1111111111111111
0
1
00000000000
11111111111
00000000000000000
11111111111111111
0000000000000000
1111111111111111
0
1
00000000000
11111111111
00000000000000000
11111111111111111
0000000000000000
1111111111111111
0
1
00000000000
11111111111
00000000000000000
11111111111111111
0000000000000000
1111111111111111
0
1
00000000000
11111111111
00000000000000000
11111111111111111
0000000000000000
1111111111111111
0
1 00000
11111
00000000000
11111111111
00000000000000000
11111111111111111
0000000000000000
1111111111111111
0
1 00000
11111
00000000000
11111111111
00000000000000000
11111111111111111
0000000000000000
1111111111111111
0
1 00000
11111
00000000000000000
11111111111111111
0000000000000000
1111111111111111
111111111111111111
000000000000000000
00000
11111
00000
11111
00000
11111
x
00000
11111
00000
11111
00000
11111
00000
11111
θ
00000
11111
00000
11111

The coordinates of a point in the two systems are related through

x′i = Rji xj , (11.16)

where R is a rotation matrix. You will note I have put the indices upstairs on
the vectors - I will return to this later. For the specific case of a rotation through
Chapter 2 14

θ about he z axis, the rotation matrix is

 

cos θ sin θ 0 
R=
 

 − sin θ cos θ 0 

 
0 0 1
Quantities that transform as

′i ∂x′i j
A = Rji Aj = A (11.17)
∂xj
are called vectors.
A simple example of a vector is dx, which transforms as
∂x′i j
′i
dx = dx ,
∂xj

Scalars

A scalar is a quantity which transforms as f ′ = f .

Co-vectors or Forms

Let us now consider how the gradient of a function transforms:

∂f ∂xj ∂f
∇′if = ′i = ′i j .
∂x ∂x ∂x
This is an example of the transformation property
∂xj
Bi′ = Bj , (11.18)
∂x′i
which is different to that of eqn. (11.17). Quantities that transform in this way
are known as covectors or forms, and we put their indices downstairs.
Summarising, we have
′i 
Vector: A′i = ∂x
∂xj A
j 




Scalar: f ′ = f 
(11.19)
∂xj

Covector: Bi′ = ∂x

′i Bj


Chapter 2 15

Finally, we have that a tensor is an object that transforms as a vector on each

upstairs index, and a covector on each downstairs index.
′ ′
′ ′ ∂x′i ∂x′j ∂xk ∂xl i′ j ′ ...
Ck′i′ lj′ ...... = . . . . . . C k ′ l ′ ...
∂xi ∂xj ∂x′k ∂x′l
′ ′

11.6.2 Metric Tensor

The length of a vector is a bilinear, and independent of the choice of frame.

Define the inner product of two vectors by

X · Y = gij X i Y j .

The tensor gij must be isotropic. There is only one isotropic rank-two tensor:

gij = δij .

We call gij the metric tensor .

We can use the metric tensor to raise or lower indices:

Xi = gij X j
X · Y = X i Yi = XiY i .

We only have the luxury of indentifying vectors with covectors in Cartesian co-
ordinates in Euclidean space, where the components of the two are numerically
equal.

Example

Show that in spherical polars

gij = diag(1, r2, r2 sin2 θ).

Chapter 2 16

11.6.3 Minkowski Space-Time

We will now apply the above ideas to Lorentz transformations of four-dimensional

space-time. We will introduce “ct” as the coordinate x0, and write a contravariant
four vector as
xµ ≡ (ct, x, y, z) = (x0, x1, x2, x3) (11.20)
The “length” of the vector is the interval left invariant under Lorentz transfor-
mations. More generally, we define the inner product of two vectors by

x · y = gµν xµ y ν , (11.21)

and we immediately see that

gµν = diag(1, −1, −1, −1) (11.22)

• Note that it is conventional to use Greek Letters for the components of a

four-vector. Four vectors are not underlined or printed in bold.

• In some areas of physics, time is introduced as the fourth component of the

vector. Furthermore, the metric can be defined such that the spatial compo-
nents are positive, and the temporal component negative. The convention I
am using is probably the most widely used, and essentially universal amongst
particle physicists.

• The summation convention is as follows - A index can appear no more than

twice. Any index appearing twice must have one upper index and one lower
index, and that index is summed over.

The covariant four vector or form can be obtained as before by using the
raising and lowering properties of the metric tensor

xµ = gµν xν .

In our example we have that xµ = (ct, −x, −y, −z) - the components of a covector
are numerically different to those of the vector.
Chapter 2 17

11.6.4 Lorentz Transformations and Four Vectors

Let us return to our two frames K and K ′. The relation between vectors in the
two frames is given by
′µ ∂x′µ ν
x = ν
x = Lµν xν (11.23)
∂x
Let us assumed a similar transformation law for the covectors

x′µ = Lµν xν .

Since xµ xµ is invariant we have

x′µ x′µ = Lµν Lµσ xν xσ ,

and since this is true for all vectors, we have

Lµν Lµσ = δνσ (11.24)

where 

1 if ν = σ
δν σ = 

(11.25)
 0 if ν 6= σ
Note that
∂xσ
Lµσ = , (11.26)
∂x′µ
the characteristic transformation property of a form.
Thus the various quantities we will encouter in the remainder of this course are

• Contravariant Vectors:
A′µ = Lµν Aν

• Covariant Vectors:
Bµ′ = Lµν Bν

• Tensors:
′ ′
C ′µ νρ′... ρ σ
′ ′
µ ν µν...
σ ′ ... = L µ L ν . . . Lρ′ Lσ . . . C ρσ...
Chapter 2 18

• Scalars:
A · B = Aµ B µ = gµν AµB ν

Finally we have the relation

gµν g νσ = δµσ .

11.6.5 Derivatives

As we have noted earlier, these transform as covectors

∂ ∂
!
∂α = = ,∇
∂xα ∂x0
∂ ∂
!
∂α = = , −∇ . (11.27)
∂xα ∂x0
Suppose now that we have a four vector Aµ . Then
α ∂A0
α
∂ Aα = ∂α A = + ∇ · A. (11.28)
∂x0
The Laplacian is defined by
∂2
2 = ∂α ∂ = 0 2 − ∇ 2 .
α
(11.29)
∂x

11.7 Relativistic Dynamics

In our introduction to Lorentz Transformations, we found that the canonical def-

inition of the velocity did not transform as a vector under a Lorentz tranforma-
tion, eqn. (11.15). Is it possible to find a definition of a velocity that does indeed
transform covariantly under Lorentz transformations, yet reduces to a Galilean
transformation for v ≪ c?
In order to construct a four velocity, we need to take the derivative with respect
to a Lorentz Scalar that can play the role of time. Such a scalar is provided by
the Proper Time dτ , defined by

c2 dτ 2 = ds2 ,
Chapter 2 19

where ds is the Lorentz-invariant interval. The proper time is clearly a scalar,

and therefore a natural definition of the four velocity is

α dxα
v = (11.30)
dτ

Recalling that the proper time is related to the lab. time by

q
dτ = dt 1 − β(t)2

we have
1 d
vα = √ (ct, x) = γ(c, v),
1 − β 2 dt
yielding
v α = (γc, γv), (11.31)
whose spatial components clearly reduce to our familiar definition of velocity in
the non-relativistic (NR) limit.

11.7.1 Four Momentum

The definition of a Lorentz-covariant four momentum is now straightforward:

pµ = mv µ = (mγc, mγv), (11.32)

where m is a Lorentz scalar that we will call the rest mass.

The spatial components of pµ clearly reduce to our usual definition of momentum.
To interpret the temporal component, we will look at its NR limit:
1 1
( )
0 2 2 −1/2
mc2 + mv 2 + O(v 4/c2 ) .
n o
p = mγc = mc 1 − v /c =
c 2
The second term in braces is clearly the kinetic energy. The first term we identify
as the rest energy, and write
p0 = E/c
where E is the energy. Thus the four momentum contains both the energy and
the three momentum.
Chapter 2 20

The “length” of pµ is a Lorentz scalar

pµ pµ = m2 γ 2c2 − m2 γ 2v 2 = m2 γ 2c2 1 − v 2 /c2

h i

= m2 γ 2c2 γ −2 = m2 c2 .

Thus we have

pµ pµ = p2 = m2 c2 (11.33)

confirming that the rest mass is a (frame-independent) scalar.

Finally, if we now go back and write eqn. (11.33) in terms of our old-fashioned
three vectors we have
1 2
2
E − p2 = m2 c2
c
=⇒ E 2 = m2 c4 + c2 p2 . (11.34)

For a particle at rest, we have perhaps the most famous equation in physics.
The use of four-vectors is essential to solve problems in special (and general. . . )
relativity. Whilst simple kinematical problems can be solved using three vectors,
it is very clumsy indeed.

11.8 Covariant Formulation of Maxwell’s Equation

Before considering Maxwell’s equations in totality, we will return to the charge

conservation.

11.8.1 Continuity Equation and Four Current

Charge conservation is expressed through the continuity equation

∂ρ
+ ∇ · J = 0. (11.35)
∂t
Chapter 2 21

We can write this in a more manifestly covariant form as

1∂
(ρc) + ∇ · J = 0.
c ∂t
It is therefore tempting to try to introduce a four-current

J µ = (ρc, J) (11.36)

in terms of which eqn. (11.35) can be formally written

∂µ J µ = 0.

However, it remains to be shown that the J µ thus constructed does indeed trans-
form as a four vector.
Consider J µ defined through eqn. (11.36) under a transformation to a frame K ′
moving with velocity v along the x axis. Then, if J µ we indeed a four vector we
would have
v
" #
′
ρ c = γ ρc − Jx
c
′
Jx = γ [Jx − vρ]
Jy′ = Jy
Jz′ = Jz .

In the NR limit 
′
J = J − ρv 

,
ρ′ = ρ 


as expected.
Consider now the case Jx = 0. Then we have

Jx′= −γvρ 

′
.
ρ = γρ 


The second equation would appear to violate charge conservation. However, let
us consider what happens to a volume element under this transformation. In the
frame K, we have
dV = dx dy dz.
Chapter 2 22

However

dx = γ(dx′ + v dt′ )
v
dt = γ(dt′ + 2 dx′ )
c
′
dy = dy
dz = dz ′ .

Thus for measurements made at the same time (dt′ = 0)

dV = dx dy dz = γdx′dy ′ dz ′ = γdV ′ ,

and the total charge in dV ′ is

ρ′ dV ′ = ρ′ γ −1dV = γργ −1dV = ρ dV

Thus both the charge densities and volumes are not separately conserved under
this Lorentz transformation, but the chage itself is.
There is much experimental evidence that ρ′ = γρ, and we will postulate that
J µ in eqn. (11.36) is indeed a four vector, and that

∂µ J µ = 0 (11.37)

11.8.2 Units

At this point, Jackson changes from SI to Gaussian units - the aim being to avoid
carrying superfluous factors of c. In my youth I did everything in SI units, and
then in units in which c ≡ 1 (a huge simplification!). But to avoid confusion (!),
I will also make the switch so as to be in keeping with Jackson.

Gaussian Units

∇ · D = 4πρ (11.38)
Chapter 2 23

4π 1 ∂D
∇×H = J+ (11.39)
c c ∂t
1 ∂B
∇×E + = 0 (11.40)
c ∂t
∇·B = 0 (11.41)

D = ǫE = E + 4πP (11.42)
H = B/µ = B − 4πM (11.43)

You will notice that in these units ∂/∂t has an associated factor of 1/c, correspond-
ing to our definition of a four vector. Also, ǫ and µ are the relative permittivity
and permeability respectively.

11.8.3 Potentials as Four Vectors

We introduce vector and scalar potentials so as to satisfy the homogeneous Maxwell

equations

B = ∇×A
1 ∂A
E = −∇φ − (11.44)
c ∂t
In a vacuum (ǫ = µ = 1), the inhomogeneous equations become:
1 ∂∇ · A
∇2 φ +
= −4πρ
c ∂t
1 ∂ 2A 1 ∂φ 4π
" #
2
∇ A− 2 2 −∇ ∇·A+ = − J.
c ∂t c ∂t c
In the Lorentz gauge, we have
1 ∂φ
∇·A+ = 0,
c ∂t
and the dynamical equations become
2 1 ∂ 2φ
∇ φ − 2 2 = −4πρ
c ∂t
1 ∂ 2A 4π
∇2 A − 2 2 = − J. (11.45)
c ∂t c
Chapter 2 24

We now recognise the operator on the l.h.s. of these equations as the four-dimensional
Laplacian introduced in eqn. (11.29), and the r.h.s. as the temporal and spatial
components of the current J µ of eqn. (11.36). We will therefore introduce a four-
vector potential
Aµ = (φ, A), (11.46)
so that both equations in (11.45) can be unified in the manifestly covariant form
4π µ
2Aµ = J . (11.47)
c
Furthermore, the Lorentz gauge condition is also manifestly covariant:

∂ µAµ = 0. (11.48)
Chapter 2 25

11.8.4 Field-Strength Tensor

In order to formulate the full Maxwell’s equations in covariant form, we need

to return to the relation between the fields (E, B) and the potentials (φ, A) of
eqn. (11.44). We need to find a covariant relation between electric and magnetic
fields, and the four vector Aµ , and indeed express the fields themselves in covariant
form. Let us write out a couple of these components explicitly
∂Az ∂Ay ∂A3 ∂A2 ∂A2 ∂A3
Bx = − = − = −
∂y ∂z ∂x2 ∂x3 ∂x3 ∂x2
∂φ 1 ∂Ax ∂A0 ∂A1 ∂A0 ∂A1
Ex = − − =− 1 − = −
∂x c ∂t ∂x ∂x0 ∂x1 ∂x0

N.B.: I am using a slightly confusing notation: E i to denote the i component

of a three vector, where we do not need to distinguish between covariant and
contravariant vectors. The equivalent four-vector components are given by

Ei = Ei
Ei = −E i .

We can see that (E, B) are related to a second-rank tensor, and there are six
independent components of the two fields.
For a general second-rank tensor T µν , we can write
µν
T µν = Tsym
µν
+ Tanti−sym .

The symmetric part has ten components, but the anti-symmetric part has the six
independent components that we could associate with fields E and B. Thus we
introduce the anti-symmetric Maxwell Field-Strength Tensor

F µν = ∂ µAν − ∂ ν Aµ (11.49)
Chapter 2 26

Writing out the components of F µν explicitly, we have

 
 0 −Ex −Ey −Ez 
 

Ex 0 −Bz By 
F µν = . (11.50)
 

Ey Bz 0 −Bx
 
 
 
 
Ez −By Bx 0

We see that E and B are not components of four vectors, but rather of an anti-
symmetric, second-rank tensor. Note that we can lower the indices in the usual
way
Fµν = gµα gνβ F αβ ,
so that the components corresponding to E change sign, whilst those correspond-
ing to B are unaltered.
Finally, we will introduce the dual field-strength tensor. But as a precursor we
will return to the Levi-Civita tensor.

Levi-Civita Tensor

This is the four-dimensional version of the ǫijk encountered in 3-D Euclidean space.
It is defined by





1 if µ, ν, ρ, σ is an even perm of 0, 1, 2, 3
ǫµνρσ = 

−1 if µ, ν, ρ, σ is an odd perm of 0, 1, 2, 3 (11.51)


0 if any two indices are equal



Lowering the indices in the usual way, we immediately see that

ǫµνρσ = −ǫµνρσ .

Note the very useful relation

ǫαβµν ǫαβρσ = −2 δ µρ δ νσ − δσµ δρν .

 
(11.52)
Chapter 2 27

11.8.5 Dual Field-Strength Tensor

The dual field-strength tensor is defined by

1
F̃ µν = ǫµνρσ Fρσ . (11.53)
2
The elements of F̃ µν are related to those of F µν through the substitution

E −→ B
B −→ −E,

so that  
 0 −Bx −By −Bz 
 

Bx 0 Ez −Ey 
F̃ µν = .
 

By −Ez 0 Ex 
 

 
 
Bz Ey −Ex 0
Thus F̃ µν reverses the roles of the electric and magnetic fields.
Finally, using eqn. (11.52), we have
µν 1
F̃˜ = ǫµνρσ F̃ρσ
2
1
= ǫµνρσ ǫµνλτ F λτ
4
= −F µν (11.54)

11.8.6 Maxwell’s Equations

Let us return to Maxwell’s equation in a vacuum

∇ · E = 4πρ (11.55)
1 ∂B
∇×E + = 0 (11.56)
c ∂t
4π 1 ∂E
∇×B = J+ (11.57)
c c ∂t
∇ · B = 0. (11.58)
Chapter 2 28

These are all first-order differential equations expressed in terms of E and B. Thus
we might suspect that the covariant form of Maxwell’s equations will contain terms
of the form
∂µ Fνρ .
Looking at eqn. (11.55), we see that it may be written
∂ i J0
E = 4π .
∂xi c
i i0 00
Recalling that E = F , and noting that F vanishes, we can rewrite (11.55) as
4π 0
∂µ F µ0 = J . (11.59)
c
Turning now to the second inhomogeneous equation, eqn. (11.57), we see that it
may be written
∂ 4π i 1 ∂ i
ǫ0ijk j F̃ k0 = J + E,
∂x c c ∂t
where we use B k = F̃ k0 . To put this equation in a form analogous to eqn. (11.59),
we perform a clever piece of manipulation:
ǫ0ijk F̃ k0 = ǫk0ij F̃k0
1
= ǫµνij F̃µν
2
ij
= F̃˜
= −F ij
where in the second line we have used that one of µ or ν must be the temporal
component, and the other a spatial component. Thus eqn. (11.57) can be written
∂ 4π i 1 ∂ i0
− j F ij = J + F
∂x c c ∂t
∂ ji ∂ 0i 4π i
=⇒ F + F = J (11.60)
∂xj ∂x0 c
4π i
=⇒ ∂µ F µi = J. (11.61)
c
Thus we see that both the inhomogeneous Maxwell equations can be written in
the unified form
Chapter 2 29

4π ν
∂µ F µν = J (11.62)
c

Turning now to the homogeneous equations, we see that eqn (11.58) can be written
∂ i0
F̃ = 0
∂xj
=⇒ ∂µ F̃ µν = 0.

Eqn. (11.56) takes the form

∂ k0 1 ∂ i0
ǫ0ijk F + F̃ = 0
∂xj c ∂t
∂ ∂
=⇒ j F̃ ij + 0 F̃ i0 = 0
∂x ∂x
=⇒ ∂µ F̃ µi = 0.

Thus the two homogeneous Maxwell equations can be written in the unified form

∂µ F̃ µν = 0 (11.63)

Eqns. (11.62) and (11.63) constitute the covariant formulation of Maxwell’s equa-
tions.
Note that we can rewrite eqn. (11.63) as
1 µνρσ
∂µ ǫ Fρσ = 0
2
=⇒ ǫµνρσ ∂µ Fρσ = 0,

which we can express as

∂µ Fρσ + ∂ρFσµ + ∂σ Fµρ = 0. (11.64)

This is known as the Jacobi Identity.

Chapter 2 30

11.9 Energy and Momentum Law

The Lorentz force law in Gaussian units is

dp 1
( )
=q E+ v×B .
dt c
In order to write this in a covariant form, we introduce the proper time

dτ = γ −1dt,

and write
dpi dpi dτ 1 dpi
= = .
dt dτ dt γ dτ
Thus the force law may be expressed as
dpi 1
( )
= γq E i + ǫ0ijk v j B k .
dτ c
We now introduce the four-velocity V µ = (γc, γv), yielding
dpi q
= {V0F i0 + ǫ0ijk V j B k }
dτ c
q
= {V0F i0 + ǫ0ijk Vj F̃k0 }
c
q
= {V0F i0 + F ij Vj } (using eqn. (11.54)).
c
Thus the Lorentz force law becomes
dpi q
= Vµ F iµ . (11.65)
dτ c
The analogous equation for the energy is
d mech dt d mech
E = E = qE · v.
dt dτ dτ
Thus we have
dE mech
= γqF i0v i
dτ
= qF 0i Vi ,
Chapter 2 31

yielding
E mech  q
 
d  = Vµ F 0µ.
dτ c c
Identifying E mech/c with the component p0 , we see that both this equation and
the Lorentz law, eqn. (11.65), can be expressed as

dpµ q
= Vν F µν (11.66)
dτ c

and Newton’s second law is in a manifestly covariant form.

Lorentz Invariants

There are two invariants we can construct from the field-strength tensor

1.

Fµν F µν = F0iF 0i + Fij F ij + Fi0F i0

= −2(E 2 − B 2)

Thus
1
E 2 − B 2 = Fµν F µν (11.67)
2
is a Lorentz Scalar.
2.

ǫµνρσ Fµν Fρσ = 2Fµν F̃ µν = −8E · B.

Thus
−4E · B = Fµν F̃ µν (11.68)
is a Lorentz Scalar

This are the only Lorentz invariants.

Chapter 2 32

11.10 Transformation Properties of EM Field

Since F µν is a second-rank tensor, we can immediately say it transforms according

to
′µν ∂x′µ αβ ∂x′ν
F = F ,
∂xα ∂xβ
which we can write as
F ′ = ΛF ΛT , (11.69)
where
∂x′µ
= Λµν .
∂xν
Specifically, let us consider a boost from K to K ′ where K ′ has velocity v in
x-direction w.r.t. K, and origins coincide at t = t′ = 0. Then
 
 γ −γβ 0 0 
 

−γβ γ 0 0 
Λ= ,
 

0 0 1 0
 
 
 
 
0 0 0 1

where β = v/c and γ = (1 − β 2)−1/2. Using this expression in eqn. (11.69), we

find
 
 0 −E1 −γ(E2 − βB3 ) −γ(E3 + βB2 ) 
 

E1 0 −γ(B3 − βE2) γ(B2 + βE3) 
F′ =
 
 
γ(E2 − βB3 ) γ(B3 − βE2) 0 −B1 
 

 
 
γ(E3 + βB2 ) γ(B2 + βE3) B1 0.

Writing out the individual vector components, we find


E1′ = E1 ; B1′ = B1 




E2′ = γ(E2 − βB3 ) ; B2′ = γ(B2 + βE3) 
(11.70)

E3′ B3′

= γ(E3 + βB2 ) ; = γ(B3 − βE2)


Chapter 2 33

We can express this in (three) vector form as

′ γ2
E = γ[E + β × B] − β(β · E)
γ+1
γ2
B ′ = γ(B − β × E) − β(β · B), (11.71)
γ+1
where β = v/c. Thus the E and B fields mix under a Lorentz transformation.
Chapter 2 34

11.10.1 Electric and magnetic fields due to relativistically moving

point change.

Consider a charge q moving along a line at velocity (in K) v = ve1 . The charge
is at rest in the frame K ′.
y y’

1
0 1
0
0
1 0
1
0000000
1111111
0
1 1111111
0000000
1
0 0000000
1111111
P 0
1
0
1
1
0
0000000
1111111
1111111
0000000
0000000
1111111
0000000
1111111
v
111111111
000000000
0
1 1111111
0000000
0
1
1 0000000
1111111
0
0
1 0000000
1111111
1
0 1111111
0000000
0
1 0000000
1111111
b 0
1
1
0
0
1
1
0
1111111
0000000
0000000
1111111
0000000
1111111
1111111
0000000 θ
0
1
0
1
1
0
r
0000000
1111111
1111111
0000000
0000000
1111111
0
1 0000000
1111111
1111111
1
0 0000000
0
1
0
1
0000000
1111111
1111111
0000000
1111111
0000000
1111111
0000000
x
x’

z z’
At t = t′ = 0, the origins of the two frames co-incide. We have an observer P at
impact parameter b (i.e. distance of closest approach) as shown above.
We will begin by looking at electric and magnetic fields at point P in frame K ′
at time t′ .
P has coordinates

x′ = −vt′
y′ = b
z ′ = 0.

Thus, from Coulomb’s law

E1′ = −qvt′ /r′3 ; E2′ = qb/r′3 ; E3′ = 0

B1′ = 0 ; B2′ = 0 ; B3′ = 0.
Chapter 2 35

In order to express this in terms of coordinates in K, we note that r′2 = b2 + v 2 t′2 .

But we have
ct′ = γ(ct − βx) = γct.
Thus
r′2 = b2 + v 2γ 2t2
and we have
qγvt
E1′ = −
(b2 + v 2 γ 2t2 )3/2
q
E2′ = − 2
(b + v γ 2t2 )3/2
2

E3′ = 0.
We now use our transformation laws eqn. (11.70) to write
qγvt
E1 = E1′ = − 2
(b + v 2 γ 2t2 )3/2
γqb
E2 = γE2′ = 2
(b + v 2γ 2t2 )3/2
E3 = γE3′ = 0

B1 = 0; B2 = γB2′ = 0
B3 = γβE2′ = βE2
Thus in the laboratory frame we see a magnetic induction.
Note that in the limit v −→ c, we have β −→ 1 and the magnetic induction equals
the transverse electric field. In the Galilean limit v −→ 0,
v γqb vqb
B3 = −→
c (b2 + v 2 γ 2t2 )3/2 c(b2 + v 2 t2 )3/2
qv ×r
=⇒ B ∼
c r3
where we have used vb = vr sin θ, which we observe is just the Biot-Savart Law.
Finally, let us look at the field lines. We have that
E2 b
=− ,
E1 vt
Chapter 2 36

so that the electric field is still a central field in the frame K. If we now look at
the magnitude of the field, however, we find
γq
|E| = 2 2 2 2 3/2
(b2 + v 2t2 )1/2.
(b + v γ t )
Setting b = r sin θ, vt = −r cos θ, we have
γqr γq
|E| = 3 2 2 3/2
∼ 2
(1 + γ cos2 θ)3/2.
r (1 + γ cos θ) r
So the lines of force, whilst central, are no longer isotropic - they are predomi-
nantly transverse in strength.

11.11 Plane Electromagnetic Radiation and Doppler Shift

Let us look at the propagation of a plane wave in vacuum. Our starting point is
the Jacobi identity eqn. (11.64). Applying ∂ α we find

∂ α ∂α Fβγ + ∂β ∂ α Fγα + ∂γ ∂ α Fαβ = 0. (11.72)

In the absence of sources,

4π
∂ µFµν = Jν = 0.
c
Thus the last two terms on the r.h.s. of eqn. (11.72) vanish, and we have the plane
e.m. waves satisfy

2Fµν = 0 (11.73)

In complete analogy to the three-dimensional NR formulations, we note that this

admits the solution
α
Fµν = fµν eikα x (11.74)
where
kα k α = k 2 = 0. (11.75)
Chapter 2 37

Writing k α = (ω/c, k), we see that eqn. (11.75) is just

k 2 = ω 2 /c2

which is our usual relation between wave number and frequency.

We will now look at the transformation properties of the solution. We will let the
solution in frame K be
Fµν = fµν eik·x
whilst that in K ′ be
′ ′
′ ′ ik ·x
Fµν = fµν e .
The solutions in the two frames are related by
′
Fµν = Λµρ Λνσ Fρσ .

This can be satisfied ∀x iff

k ′ · x′ = k · x
showing that k and k ′ are indeed four vectors. Because of this, we know that kµ
and kµ′ are related by

kk′ = γ[kk − βk0]

k0′ = γ[k0 − βkk]
k ′⊥ = k ⊥ (11.76)

Introducing θ as the angle between k and v, we can use the second eqn. of (11.76)
to compute the Doppler shift:
ω′ ω v ω v ω
" # " #
=γ − cos θ|k| = γ − cos θ .
c c c c c c
Thus we have the Doppler Shift formula

ω ′ = γω(1 − β cos θ) (11.77)

where β = v/c. This is modified from the usual Galilean formula through the
factor of γ.
Chapter 2 38

11.11.1 Aberation

This is the change in direction of a wave vector between the two frames.

k k’
θ θ’
v
v
We can calculate this from
′ |k ′⊥ | |k ⊥|
tan θ = ′ = ′ .
kk kk

By our Lorentz transformation formula

ω ω ω
" # " #
kk′ = γ kk − β =γ cos θ − β
c c c
ω
= γ (cos θ − β)
c
Also we have
!2 !2
ω ω
k 2⊥ = k02 − kk2 = 2
(1 − cos θ) = sin θ ,
c c
and thus
ω
|k ⊥ | = sin θ.
c
Thus, taking the ratio, we find

sin θ
tan θ′ = (11.78)
γ(cos θ − β)
Summary of First Three Chapters

Introduction

There are two founding principles of electrostatics:

• Coulomb’s Law:
1 r̂21
F21 = q1 q2
4πǫ0 |r2 − r1 |2
• Principle of Linear Superposition
The resultant force on a test particle due to several charges is the
vector sum of the forces due to the charges individually.
In the case of a continuous charge distribution, these two principles yield
1 Z ρ(r′ )(r − r′ )dV ′
E(r) = .
4πǫ0 V | r − r′ |3

Gauss’ Law

If charge density ρ(r) is the sole source of the electrostatic field E(r), the flux of
E out of a closed surface S bounding a volume V is given by
Z
Q
E · dS = 4πCQ = in SI units
S ǫ0
where Q = total charge within S. This can be expressed in differential form as
Maxwell’s First Equation (ME1)

∇ · E = ρ/ǫ0 .

39
Chapter 2 40

Scalar Potential

It is easy to show ∇ × E = 0, and for such fields

E(x) = −∇φ(x),

where φ is the Scalar Potential.

Laplace’s and Poisson’s Equation

ME1 can be expressed in terms of potential as

∇2 φ(x) = −ρ(x)/ǫ0

Uniqueness Theorem

The solution φ(x) of Poisson’s or Laplace’s equation inside a vol-

ume V bounded by surface S satisfying either Dirichlet or Neumann
boundary conditions is unique (Dirichlet) or unique up to a constant
(Neumann).

This is perhaps the most important theorem in the course, which we will use
implicitly in much of the following.

Boundary conditions at Surface of Conductor

Throughout the body of a conductor, E vanishes and at the surface

φ(x) = constant

so that E is normal to the surface. The surface charge density σ is related to

the discontinuity in the normal electrostatic field

E · n = σ/ǫ0.
Chapter 2 41

Boundary-Value Problems in Electrstatics

We explored variety of ways of obtaining unique solution.

• Method of Images
Here we introduce an image charge outside the region we are seeking a solu-
tion such that image system satisfies boundary conditions of original problem.
Two crucial things to check

– Only additional charges are introduced outside the region so that Pois-
son’s equation is unaltered.
– Image system satisfies correct boundary conditions.

Then by uniqueness theorem, image potential is the unique solution in

region of interest.

• Solution in terms of Green Functions.

The Green function G(x, x′ ) for Laplaces or Poisson’s equation in a volume
V bounded by surface S satisfies

∇′2G(x, x′ ) = −4πδ (3) (x − x′ )

subject to

G(x, x′ ) = 0 for x′ ∈ S - Dirichlet

∂G(x, x′ ) 4π
′
= − for x′ ∈ S - Neumann.
∂n S
Poisson’s equation has formal solution,,
1 Z 3 ′
φ(x) = d x G(x, x′ )ρ(x′ ) +
4πǫ0 V 
′ ′ 

1 Z ∂φ(x ) ∂G(x, x )
dS ′ G(x, x′ ) − φ(x′ )


4π S=∂V ∂n ′ ∂n′ 
Green functions can be obtained using, e.g., method of images, or in terms
of orthonormal eigenfunctions.
Chapter 2 42

• Expansion of Solution in Orthogonal Functions.

Equations of Sturm-Liouville type have a set of solutions which are orthog-
onal and complete, subject to certain specific boundary conditions. Then any
square-integrable function can be expanded in terms of these eigenfunctions.

• Separation of Variables in Cartesian Coordinates

This provides a powerful method of seeking solutions of Laplace’s equation,
and indeed often generates a set of useful basis functions.
Writing φ(x, y, z) = X(x)Y (y)Z(z), Laplace’s equation can be written
1 d2 X 1 d2 Y 1 d2 Z
+ + = 0,
X dx2 Y dy 2 Z dz 2
each term of which is separately constant. There are eleven coordinate sys-
tems admitting such a separation!
Chapter 2 43

Boundary-value Problems in Curvilinear Coordinates

Here we seek factorisable solutions in spherical polars and in cylindrical polars

Spherical Polars, Azimuthal Symmetry

We seek solution of form

U (r)
φ(r, θ, ψ) = P (θ)Q(ψ).
r
• In case of azimuthal symmetry (no ψ dependence), the general solution
assumes the form
∞ 
Al rl + Bl r−l−1 Pl (cos θ),
X 
φ(r, θ) =
l=0

where Pl (cos θ) are the Legendre Polynomials, satisfying Legendre’s

equation
d 2 dP
" #
(1 − x ) + l(l + 1)P = 0.
dx dx
• Pl (cos θ) form a complete set of orthogonal functions. Note that they are not
orthonormal.

• Important expansion for construction of Green functions

1 ∞ rl
X <
′
= l+1 Pl (cos γ)
|x − x | l=0 r>
where r> = max(r, r′) and r< = min(r, r′).

Spherical Harmonics

Where there is no azimuthal symmetry, we have the Generalized Legendre

equation
2
   
d  dP (x) m
(1 − x2)  + l(l + 1) −
2
 P (x) = 0.
dx dx 1−x
Chapter 2 44

• Convenient to combine ψ and θ functions into solutions on unit sphere de-

scribed by Spherical Harmonics Ylm (θ, ψ), with general solution
∞ X
l h
Alm rl + Blm r−l−1 Ylm(θ, ψ)
X i
φ(r, θ, ψ) =
l= m=−l

• Important, and very useful, result:

∞ X l l
1 X 1 r< ∗ ′ ′
′
= 4π l+1 Ylm (θ , ψ )Ylm (θ, ψ),
|x − x | l=0 m=−l 2l + 1 r>

where we have factorised the (θ, ψ) and (θ′ , ψ ′ ) behaviour.

Laplace Equation in Cylindrical Polars

Here we seek factorisable solutions of form

φ(ρ, θ, z) = R(ρ)T (θ)Z(z).

• Gives rise to Bessel’s equation

d2 R 1 dR  ν2 
 

+ + 1− 2 R =0
dx2 x dx x

• Solutions are the Bessel Functions Jν (x) and Nν (x), which are linearly
independent. A further set is provided by Hankel Functions.

Expansion of Green Function in terms of Orthogonal Functions

• Because any function can be expanded in set of orthogonal functions, we can

expand Green function.

• Green Function for Sturm-Liouville Equation

d  ′ dg(x, x′) 
 

′
p(x ) ′
+ q(x′)g(x, x′) = −4πδ(x − x′ ),
dx dx
defined on the interval x′ ∈ [a, b], with homogeneous boundary conditions at
a and b.
Chapter 2 45

• Green function
4π y2 (x)y1(x′)

a ≤ x′ ≤ x

−



p(x) W [y1(x), y2(x)]


g(x, x′) = 


4π y2 (x′)y1(x)
x ≤ x′ ≤ b

−



p(x) W [y1(x), y2(x)]



where
W [y1(x), y2(x)] = y1(x)y2′ (x) − y2 (x)y1′ (x).
is the Wronskian, and y1 and y2 are solutions to the homogeneous equation

• Spectral Representation Expand Green function in terms of the eigen-

functions of some related problem. Consider the solution of

∇2ψ(x) + [f (x) + λ]ψ(x) = 0,

in volume V subject to ψ satisfying certain homogeneous boundary conditions

for x ∈ S. Find set of eigenfunctions ψn , with eigenvalues λn .
Green function satisfies

∇′2G(x, x′ ) + [f (x′ ) + λ]G(x, x′ ) = −4πδ(x − x′ )

where λ is, in general, not an eigenvalue, and we can write

′ X ψn∗ (x)ψn(x′ )
G(x, x ) = 4π
n λn − λ