1

Notes on
Electrodynamics

Daniel F. Styer
2

Notes on Electrodynamics

Daniel F. Styer
Schiffer Professor of Physics, Oberlin College
[email protected]
http://www.oberlin.edu/physics/dstyer

copyright c 27 November 2024 Daniel F. Styer

Cover photo: Light scattering from early morning mist over the Vermilion River in
Wolf Run Nature Preserve, Ohio.

In these notes “Griffiths” means the book David J. Griffiths, Introduction to Elec-
trodynamics, fourth edition (Pearson, Boston, 2013).

The copyright holder grants the freedom to copy, modify, convey, adapt, and/or
redistribute this work under the terms of the Creative Commons Attribution Share
Alike 4.0 International License. A copy of that license is available at
http://creativecommons.org/licenses/by-sa/4.0/legalcode.

You may freely download this book in pdf format from

http://www.oberlin.edu/physics/dstyer/Electrodynamics.

It is formatted to print nicely on either A4 or U.S. Letter paper. The author receives
no monetary gain from your download: it is reward enough for him that you want
to explore this strange and beautiful universe, our home.
Contents

1 Welcome 5

1.1 Electrostatics and Magnetostatics . . . . . . . . . . . . . . . . . . . . . 5

1.2 Electrodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

1.3 Electromagnetic energy . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2 Vector Calculus 10

2.1 What is a vector? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2.2 Geometrical definition of divergence and curl . . . . . . . . . . . . . . 13

2.3 Pictorializing divergence and curl . . . . . . . . . . . . . . . . . . . . . 19

2.4 Vector identities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

2.5 The divergence of the curl is zero . . . . . . . . . . . . . . . . . . . . . 22

3 Conservation of Charge and the Maxwell Equations 23

3.1 Changing electric field makes magnetic field . . . . . . . . . . . . . . . 23

3.2 Do magnetic monopoles make magnetic field? . . . . . . . . . . . . . . 24

4 Energy and Momentum 26

4.1 Conservation of energy . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

4.2 The Maxwell stress tensor . . . . . . . . . . . . . . . . . . . . . . . . . 27

5 Electromagnetic Waves 38

5.1 One-dimensional waves . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

5.2 From Maxwell equations to wave equations . . . . . . . . . . . . . . . 46

5.3 Waves in media . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

3
4 CONTENTS

6 Potentials and Gauges 53

6.1 Math . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

6.2 Potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

6.3 Gauge freedom . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

7 Resultant Potentials and Fields 63

7.1 Single moving point charge . . . . . . . . . . . . . . . . . . . . . . . . 65

7.2 Fields from a good swift kick . . . . . . . . . . . . . . . . . . . . . . . 70

7.3 Radiation simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

7.4 Radiation from an electric dipole . . . . . . . . . . . . . . . . . . . . . 75

8 Relativistic Electrodynamics 83

8.1 Space and time in relativity . . . . . . . . . . . . . . . . . . . . . . . . 83

8.2 Energy and momentum in relativity . . . . . . . . . . . . . . . . . . . 86

8.3 Current density in relativity . . . . . . . . . . . . . . . . . . . . . . . . 86

8.4 Four-tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87

8.5 Magnetism as a relativistic phenomenon . . . . . . . . . . . . . . . . . 89

8.6 Electromagnetic fields in relativity . . . . . . . . . . . . . . . . . . . . 91

8.7 Maxwell equations in relativistic notation . . . . . . . . . . . . . . . . 92

8.8 The stress-energy four-tensor . . . . . . . . . . . . . . . . . . . . . . . 94

8.9 What to do with relativistic electrodynamics . . . . . . . . . . . . . . 97

9 Atmospheric Optics 98

A Pictorializing divergence and curl 104

Chapter 1

Welcome
You have been familiar with electrostatics since, as an infant, you rubbed a balloon
on your hair and then squealed in delight when it stuck to a wall.

1.1 Electrostatics and Magnetostatics

Electrostatics Magnetostatics

~ =
E
1 q
r̂ ~ = µ0 q~v × r̂
B
4π0 r2 4π r2
Coulomb’s Law Biot–Savart Law

~ ·E
∇ ~ = ρ/0 ~ ~
∇
I I ·B =0
E~ · n̂ dA = Qinside /0 B~ · n̂ dA = 0

Gauss’s Law

~ ×E
∇ ~ =0 ~
∇ ~ ~
I I × B = µ0 J
~ · d~` = 0
E ~ · d~` = µ0 Ilinked
B
Ampère’s Law

~ field lines
E B~ field lines
begin and end on charges loop around current
(or infinity) (right-hand-rule)
they never loop they begin or end only at infinity

stationary charge distributions steady currents

5
6 CHAPTER 1. WELCOME

1.2 Electrodynamics
Electrodynamics

1 q
~ = NO! ~ = µNO! v × r̂
0 q~
E r̂ B
4π0 r2 4π r2
Coulomb’s Law Biot–Savart Law

~ ·E
∇ ~ = ρ/0 ~ ~
∇
I I ·B =0
E~ · n̂ dA = Qinside /0 B~ · n̂ dA = 0

Gauss’s Law

~ ~
~ ×E
∇ ~ = − ∂B ~ ×B
∇ ~ = µ0 J~ + µ0 0 ∂ E
∂t ∂t
I I
E~ · d~` = − dΦB ~ · d~` = µ0 Ilinked + µ0 0 dΦE
B
dt dt
Faraday’s Law Ampère’s Law with Maxwell’s term

~ field lines
E B~ field lines
begin and end on charges ~
loop around current and ∂ E/∂t
(or infinity) (right-hand-rule)
~
or they loop around ∂ B/∂t they begin or end only at infinity
(anti-right-hand rule)
1.2. ELECTRODYNAMICS 7
I
Faraday discovered his law E~ · d~` = − dΦB experimentally
dt
Michael Faraday, Experimental Researches in Electricity, 24 November 1831:
Two hundred and three feet of copper wire in one length were coiled round a large
block of wood; other two hundred and three feet of similar wire were interposed as a
spiral between the turns of the first coil, and metallic contact everywhere prevented
by twine. One of these helices was connected with a galvanometer, and the other
with a battery of one hundred pairs of plates four inches square, with double coppers,
and well charged. When the contact was made, there was a sudden and very slight
effect at the galvanometer, and there was also a similar slight effect when the contact
with the battery was broken. But whilst the voltaic current was continuing to pass
through the one helix, no galvanometrical appearances nor any effect like induction
upon the other helix could be perceived, although the active power of the battery was
proved to be great, by its heating the whole of its own helix, and by the brilliancy of
the discharge when made through charcoal.

I
Maxwell discovered his term ~ · d~` = µ0 0 dΦE theoretically
B
dt
µ0 1
In SI units, we think of as “small”, and of as “big”, so µ0 0 is “real
4π 4π0
small”:

µ0 = 1.257 × 10−6 H/m

0 = 8.854 × 10−12 F/m
µ0 0 = 11.13 × 10−18 HF/m2

What is a henry times a farad?

1 ΦB Q
ΦB = Li ∆V = Q so LC = .
C i ∆V
Then, using square brackets to mean “units of” (for example the units of length are
meters, [`] = m):
[ΦB ] [Q] [B][`]2 [Q]
HF = [L][C] = =
[i] [V ] [Q]/[t] [E][`]
But F~ = q(E~ + ~v × B)
~ so [E] = [v][B] and the chain of reasoning continues as

[B][`]2 [t] [Q] [`]2 [t] [`][t] [`][t]

HF = = = = = [t]2 .
[Q] [v][B][`] [v][`] [v] [`]/[t]
(The “electrical like” units cancel out, just as in RC = τ — an ohm times a farad is
a second.) Finally

µ0 0 = 1.113 × 10−17 s2 /m2 .

This effect is so small that it was unlikely ever to be discovered by fiddling with wire
and twine and charcoal.
8 CHAPTER 1. WELCOME

[[Maxwell published his results in a series of papers between 1865 and 1875,
the first direct experimental evidence did not come until 54 years later: R. van
Cauwenberghe, “Vrification expérimentale de léquivalence électromagnétique entre
les courants de déplacement de Maxwell et les courants de conduction” [“Exper-
imental verification of electromagnetic equivalence between Maxwell displacement
currents and conduction currents”] Journal de Physique et le Radium 10 (1929)
303–312. For later work see D.F. Bartlett and T.R. Corle, “Measuring Maxwell’s
displacement current inside a capacitor” Physical Review Letters 55 (1985) 59–62.]]

1.3 Electromagnetic energy

Electrostatic potential energy:

1 X qi qj
UE = . (1.1)
4π0 r i,j
pairs

• The sum is over pairs not over particles. Suppose the particles are 1, 2, and
3. Then there’s the potential energy of interaction of 1 and 2, the potential
energy of interaction of 1 and 3, and the potential energy of interaction of 2
and 3. These sum to the total potential energy. There is no such thing as “the
potential energy of particle 2”, so of course you can’t sum over the potential
energy of 1, of 2, and of 3, because these things don’t exist.

• This potential energy can be positive or negative.

• The derivation starts with the three particles infinitely far away (zero potential
energy by convention). Calculate the work done by the electric force while
bringing in particle 1. (This is of course zero.). Now calculate the work done
by the electric force while bringing in particle 2. (This is the negative of the
potential energy of interaction of 1 and 2.) Finally calculate the work done
by the electric force while bringing in particle 3. (This is the negative of the
potential energy of interaction of 1 and 3 and of 2 and 3.) It is half-way to
a miracle that the final result depends not on the sequence by which particles
brought in, not on the route taken by each particle, not on speed at which each
particle was brought in along that route, but only on the final positions.
1.3. ELECTROMAGNETIC ENERGY 9

Magnetostatic potential energy:

µ0 X qi qj ~vi · ~vj
UB = . (1.2)
4π r i,j
pairs

• The magnetic field does no work. How can there be “magnetostatic potential
energy”?

• The magnetostatic potential energy is not “negative the work done by the
magnetic field while bringing in particle 2”. (This quantity is zero.) Instead
it is the “negative of the work done by the electric field caused by changing
magnetic field while bringing in particle 2”. Magnetostatic potential energy
exists because of Faraday’s Law, a dynamical result!

• It is three-quarters-way to a miracle that the final result depends not on the

route with which each particle comes in, nor on the speed taken by each particle
as it comes in along its route, but only upon the final positions and velocities.

Electrodynamic potential energy:

Z  
0 2 1 2
UED = E (~r) + B (~r) d3 r (1.3)
2 2µ0

• This is always positive, so obviously there is a change in sea-level involved in

deriving this result.

• Equations (1.1) and (1.2) are wrong in electrodynamics.

Chapter 2

Vector Calculus

2.1 What is a vector?

A list of three numbers is a triplet or a 3-tuple; it is not necessarily a three dimensional

vector. For example, the thermodynamic list (p, V, T ) is not a vector. Again, the list

(population of Spain, cost of a cup of coffee in Caracas, height of Kilimanjaro)

is not a vector.

Here is an orthonormal basis:

ê2

ê1

ê3

(Basis vectors don’t have to be orthogonal, they don’t even have to be unit vectors,
but that’s convenient and traditional.)

~r = (r1 , r2 , r3 ) No!
~r = r1 ê1 + r2 ê2 + r3 ê3 Yes!
.
~r = (r1 , r2 , r3 ) Yes!

The last equation is read “The vector ~r is represented by the three numbers (r1 , r2 , r3 )
in the basis {ê1 , ê2 , ê3 }.” or “The name of vector ~r in the basis {ê1 , ê2 , ê3 } is (r1 , r2 , r3 ).”

The name of an object depends upon both the object and the language used:
The name of a horse in the English language is “horse”; the name of a horse in the

10
2.1. WHAT IS A VECTOR? 11

German language is “pferd”; the name of a horse in the Swahili language is “farasi”.
The horse doesn’t change if you use one language rather than another, only its name
changes. Similarly the vector doesn’t change if you represent it in one basis rather
than another, only its name (or “representation”) changes.

Here are two orthonormal bases:

ê02 ê2
ê01
ê1

ê3
ê03

Vector ~r can be written in terms of this basis as

~r = r1 ê1 + r2 ê2 + r3 ê3

= r10 ê01 + r20 ê02 + r30 ê03 (best notation)
= r10 ê01 + r20 ê02 + r30 ê03 (most practical notation)

How do we translate from one “language” (basis) to another? What is our “dic-
tionary”?

r10 = ~r · ê01
= (r1 ê1 + r2 ê2 + r3 ê3 ) · ê01
= r1 ê1 · ê01 + r2 ê2 · ê01 + r3 ê3 · ê01

or for all three components

    
r10 ê1 · ê01 ê2 · ê01 ê3 · ê01 r1
 0  
 r2  =  ê1 · ê02 ê2 · ê02 ê3 · ê02   r2  .
 

r30 ê1 · ê03 ê2 · ê03 ê3 · ê03 r3

In other words
3
X
ri0 = Rij rj where Rij = êj · ê0i .
j=1

The definition of vector is “an entity with components that transform from one basis
to another in this way.” (This is the definition produced by Gregorio Ricci-Curbastro
and Tullio Levi-Civita in the year 1900. Other definitions of “vector” exist, including
the 1888 “vector space” definition of Giuseppe Peano, which was generalized by David
Hilbert and Erhard Schmidt in 1908.)
12 CHAPTER 2. VECTOR CALCULUS

How can we use this defintion?

We can have scalar functions of vectors, f (~r), and vector functions of vectors,
F~ (~r).
~ (~r), with components (“name”)
You know how to take and interpret the gradient ∇f
 
~ . ∂f ∂f ∂f
∇f (~r) = , , .
∂r1 ∂r2 ∂r3
It points in the direction of fastest increase of f (~r), and its magnitude is the slope in
that direction.
There is an obvious way of taking the derivative of a vector function, too. It is
 
∂F1 ∂F2 ∂F3
, , .
∂r1 ∂r2 ∂r3

Here are two orthonormal bases related by a 90◦ rotation in the ê1 -ê2 plane (I
don’t show the third dimension because it doesn’t change (ê03 = ê3 ) and because it’s
hard to draw.)
ê01 ê2

ê02 ê1

If vector ~r is named (r1 , r2 , r3 ) = (x, y, z) in the first basis, it is named (r10 , r20 , r30 ) =
(y, −x, z) in the second basis.
If vector F~ is named (F1 , F2 , F3 ) = (Fx , Fy , Fz ) in the first basis, it is named
(F10 , F20 , F30 ) = (Fy , −Fx , Fz ) in the second basis.
In the first basis, the combination of symbols is
   
∂F1 ∂F2 ∂F3 ∂Fx ∂Fy ∂Fz
, , = , , ≡ (a, b, c).
∂r1 ∂r2 ∂r3 ∂x ∂y ∂z
In the second basis, the combination of symbols is
 0
∂F1 ∂F20 ∂F30
  
∂Fy ∂(−Fx ) ∂Fz
, , = , , = (b, a, c).
∂r10 ∂r20 ∂r30 ∂y ∂(−x) ∂z
The components of a vector would transform from (a, b, c) to (b, −a, c),
This combination of symbols is not a vector, it is an impostor! Like the triplet
(p, V, T ), they are three numbers that do not constitute a vector. Now that we know
how not to take a vector derivative, we investigate how we should do it.
Challenge: Show that the transformation matrix for the change of basis above is
 
0 −1 0
Rij =  1 0 0  .
 

0 0 1
2.2. GEOMETRICAL DEFINITION OF DIVERGENCE AND CURL 13

2.2 Geometrical definition of divergence and curl

Derivative of a single-variable function

The derivative of function f (x) at point x0 is given by

f (x0 + L/2) − f (x0 − L/2)
lim
L→0 L
that is, in involves the function values at the edges of an interval, divided by the
magnitude of that interval.
If you’re a mathematician, you’ll want to prove that the limit exists, and that it is
the same whether L approaches zero from above (through positive numbers) or from
below (through negative numbers). Physicists usually just skip over such questions,
interesting though they may be. Instead we just note that the result has the correct
dimensions for a slope, and that it leads to the indeterminate form 0/0.

Divergence

So how should we define the derivative of a vector function F~ (~r) at point ~r0 ? Here’s
one way. Consider a sequence of volumes V that enclose point ~r0 , but that grow
smaller and smaller.

~r0

A sequence of volumes (shown in cross-section) homing in on point ~r0 .

If the volume V is enclosed by surface S, then

Z
F~ (~r) · n̂ dA
lim S of V (2.1)
volume of V
14 CHAPTER 2. VECTOR CALCULUS

fits the requirements for some sort of derivative: it involves the function values at
the edge of the volume divided by the magnitude of that volume, it has the correct
dimensions, and it leads to the indeterminate form 0/0. Mathematicians will want
to prove that the limit exists, and that it gives the same result regardless of what
sequence of volumes (cubes, spheres, hemispheres, cats, etc.) is used to close in on
~r0 . But we’ll skip over such general questions and ask:
What is the result if the sequence of volumes consists of cubes centered on ~r0 ?

flux through flux through

left face right face
~r0

z
y L

It’s clear from the definition of flux that for a small cube

flux through right face ≈ Fx (evaluated at center of right face)L2

= Fx (x0 + L/2, y0 , z0 )L2

and that this approximation grows better and better as L grows smaller and smaller.
Similarly

flux through left face ≈ −Fx (evaluated at center of left face)L2

= −Fx (x0 − L/2, y0 , z0 )L2 .

Thus

flux through right plus left faces ≈ [Fx (x0 + L/2, y0 , z0 ) − Fx (x0 − L/2, y0 , z0 )]L2
 
Fx (x0 + L/2, y0 , z0 ) − Fx (x0 − L/2, y0 , z0 ) 3
= L
L
 
∂Fx
→ (x0 , y0 , z0 ) L3
∂x
where the symbol → means “in the limit as L → 0”.
Parallel reasoning shows that the flux through the back plus front faces is
 
∂Fy
(~r0 ) L3
∂y
while the flux through the top plus bottom faces is
 
∂Fz
(~r0 ) L3 .
∂z
2.2. GEOMETRICAL DEFINITION OF DIVERGENCE AND CURL 15

Finally, the limit presented in definition (2.1) results in

∂Fx ∂Fy ∂Fz
(~r0 ) + (~r0 ) + (~r0 ). (2.2)
∂x ∂y ∂z
Because this derivative is the “flux per volume at a point” we call it the “divergence
at a point”.
Some people like to begin with equation (2.2) and call this the definition of di-
vergence. Then they have a difficult time proving the divergence theorem (or Gauss’s
theorem), namely that if volume V is enclosed by surface S, then
Z Z
~ · F~ (~r) d3 r =
∇ F~ (~r) · n̂ dA.
V S of V
I prefer to begin with the geometrical definition (2.1), and derive expression (2.2) for
the divergence in Cartesian coordinates. In this approach, the divergence theorem
just pops right out of the definition.
You could do the “flux through a shrinking volume” argument for shapes other
than cubes. If you do it for the shape below

you will find the expression for divergence in spherical coordinates. If you do it for
other shapes you will find the expression for divergence in cylindrical coordinates,
or prolate spheroidal coordinates, or confocal paraboloidal coordinates, or any other
kind of coordinates. The idea of “flux per volume at a point” is the same for all
coordinate systems – the shape of the shrinking volume is different for different
coordinate systems.
Summary: You know that the total mass M of an object can be found by
integrating the mass density ρ(~r) over the volume of the object:
Z
M= ρ(~r) d3 r.
V
The divergence plays the role of “flux density” rather than mass density.
16 CHAPTER 2. VECTOR CALCULUS

Curl

Expression (2.1) is not the only possible derivative of a vector function. Consider
a sequence of loops L (all within the plane perpendicular to a unit vector n̂) that
enclose point ~r0 , but that grow smaller and smaller.

n̂

~r0

A sequence of loops (all within the plane perpendicular to n̂) homing in on point ~r0 .
(Direction of loop given through right-hand rule.)

The line integral of F~ (~r) along loop L is called the “circulation of F~ (~r) along L.” If
the loop L embraces a surface S, then
Z
F~ (~r) · d~`
L of S
lim (2.3)
area of S
also fits the requirements for some sort of derivative: it involves the function values
at the edge of the surface divided by the magnitude of that surface, it has the correct
dimensions, and it leads to the indeterminate form 0/0. Mathematicians will want
to prove that the limit exists, and that it gives the same result regardless of what
sequence of shapes (squares, circles, squirrels, etc.) is used to close in on ~r0 . But
we’ll skip over such general questions and ask:
2.2. GEOMETRICAL DEFINITION OF DIVERGENCE AND CURL 17

What is the result if the sequence of loops consists of squares in the x-z plane,
centered on ~r0 ?

circulation due to
left edge circulation due to
right edge
~r0
z
y L

It’s clear from the definition of circulation that for a small square

circulation due to right edge ≈ −Fz (evaluated at center of right edge)L

= −Fz (x0 + L/2, y0 , z0 )L

and that this approximation grows better and better as L grows smaller and smaller.
Similarly

circulation due to left edge ≈ Fz (evaluated at center of left edge)L

= Fz (x0 − L/2, y0 , z0 )L.

Thus

circulation due to right plus left edges ≈ −[Fz (x0 + L/2, y0 , z0 ) − Fz (x0 − L/2, y0 , z0 )]L
 
Fz (x0 + L/2, y0 , z0 ) − Fz (x0 − L/2, y0 , z0 ) 2
= − L
L
 
∂Fz
→ − (x0 , y0 , z0 ) L2
∂x
where the symbol → means “in the limit as L → 0”. Parallel reasoning shows that
the circulation due to the top plus bottom faces is
 
∂Fx
(~r0 ) L2 .
∂z
Finally, the limit presented in definition (2.3) results in
∂Fx ∂Fz
(~r0 ) − (~r0 ). (2.4)
∂z ∂x

You can carry out this “limit of circulation per area” for an infinite number of
planes passing through point ~r0 . Find the plane with the largest “limit of circulation
per area”. The curl of F~ (~r) at ~r0 is a vector with direction perpendicular to this
plane and magnitude equal to that maximum.
18 CHAPTER 2. VECTOR CALCULUS

This process finds the curl, but it requires finding the limit for an infinite number
of planes! It is equivalent, and far more practical, to carry out the process for three
planes only. The process resulting in equation (2.4) finds the “circulation in the plane
perpendicular to y per area at a point” which also equals the “y-component of curl
at a point”.

Parallel considerations for planes perpendicular to x and to z result in the tradi-

tional expression for the curl in Cartesian coordinates, namely
 
~ ~ . ∂Fz ∂Fy ∂Fx ∂Fz ∂Fy ∂Fx
∇ × F (~r0 ) = (~r0 ) − (~r0 ), (~r0 ) − (~r0 ), (~r0 ) − (~r0 ) . (2.5)
∂y ∂z ∂z ∂x ∂x ∂y

Some people like to begin with equation (2.5) and call it the definition of curl.
Then they have a difficult time proving the circulation theorem (or Stokes’s theorem):
If the surface S is bounded by loop L, then
Z Z
~ × F~ (~r)) · n̂ dA =
(∇ F~ (~r) · d~`. (2.6)
S L of S

I prefer to begin with the geometrical definition

Z
F~ (~r) · d~`
~ × F~ (~r0 )) · n̂ = lim
(∇ L of S
, (2.7)
area of S
and derive expression (2.5) for the Cartesian coordinates of the curl. In this approach,
the circulation theorem just pops right out of the definition.

Acknowledgment: In my multivariate calculus course, I learned the “Cartesian

coordinate” definitions of divergence and curl, and these definitions left a bad taste in
my mouth. Why were divergence and curl – particularly curl – defined through such
bazaar combinations of derivatives? Math is supposed to be coordinate-independent:
Why were Cartesian coordinates so special? A one-variable derivative has geometrical
significance as a slope – what was the geometrical significance of divergence and curl?
A related question – why do divergence and curl have such strange names?

I learned the much-more-satisfactory geometric approach to vector derivatives –

the one outlined in this document – from my physics professor Mark Heald. When I
asked him how he had learned it, he told me it was the approach used by his teacher,
Carl Howe, when he took undergraduate electricity and magnetism at Oberlin Col-
lege.
2.3. PICTORIALIZING DIVERGENCE AND CURL 19

2.3 Pictorializing divergence and curl

The picture below [from Edward M. Purcell, Electricity and Magnetism (McGraw-
Hill, 1965) page 72] helped develop my sense of the geometric meaning of divergence
and curl — perhaps it will help you, too. Spend some time puzzling over these figures
before turning to the answers at the back of this book.
20 CHAPTER 2. VECTOR CALCULUS

2.4 Vector identities

I am convinced that every vector calculus identity is not just a jumble of symbols: it
is a statement about geometry.

For example, the Laplacian is often defined, in Cartesian coordinates, as

2 2 2
~ 2 f (x, y, z) = ∇
∇ ~ (x, y, z) = ∂ f + ∂ f + ∂ f .
~ · ∇f
∂x2 ∂y 2 ∂z 2
What makes this particular combination of partial derivatives so special? Why do
we so often encounter this combination and so rarely encounter, say, the combination
∂f ∂f ∂f
+ + ?
∂x ∂y ∂z

I wrote a paper about this question [“The geometrical significance of the Lapla-
cian” American Journal of Physics, 83 (12) 992–997 (December 2015)] in which I
show that an equivalent definition of the Laplacian in dimension d is
 
~ 2 2d
∇ f (~r0 ) = lim [hf ishell − f (~r0 )] .
R→0 R2

James Clerk Maxwell used a property like this to motivate the name “concentration”
~ 2.
for −∇
~ 2 f (~r) = 0 in a region, then within that region there are no maxima, no
If ∇
minima. An example is this two dimensional function

Such a function could be used as a roof to shed rain. And if it were turned upside-
down it would still shed rain!
2.4. VECTOR IDENTITIES 21

I haven’t yet figured out the geometric significance of some vector identities, for
example
~ × (∇
∇ ~ × A)
~ = ∇(
~ ∇~ · A)
~ −∇ ~ · (∇
~ A).
~

Others are subtle. For example. . .

22 CHAPTER 2. VECTOR CALCULUS

2.5 The divergence of the curl is zero

[Approach from Edward M. Purcell, Electricity and Magnetism (McGraw-Hill, 1965)
problem 2.15.]
~ r) is a vector field with continuous derivatives, then
If A(~
~ · (∇
∇ ~ × A(~
~ r)) = 0.

How to prove? You could plug-and-chug in Cartesian coordinates. But it’s easier
and more insightful to do it this way.

P2
P1

Consider (figure on the left) the volume V enclosed by surface S. Apply the
divergence theorem to function F~ (~r) = ∇
~ × A(~
~ r), giving
Z Z
∇~ · (∇
~ × A(~
~ r)) d3 r = ~ × A(~
(∇ ~ r)) · n̂ dA.
V S of V

Now slice out and remove from the surface a tiny sliver (figure on the right).
Technically we’ve altered S, but this tiny alteration will not affect the value of the
surface integral. The edge E of the altered S is the edge of the sliver. Apply the
circulation theorem to F~ (~r) = A(~
~ r), giving
Z Z
(∇~ × A(~
~ r)) · n̂ dA = ~ r) · d~`.
A(~
S E of S
But
Z Z Z Z Z
~ · d~` =
A ~ · d~` +
A ~ · d~` =
A ~ · d~` −
A ~ · d~` = 0.
A
E of S P1 to P2 P2 to P1 P1 to P2 P1 to P2
So for any volume V, Z
~ · (∇
∇ ~ × A(~
~ r)) d3 r = 0.
V
Because this holds for any volume,
~ · (∇
∇ ~ × A(~
~ r)) = 0.

(This is an example of the theorem that “the boundary of a boundary is zero,” as

emphasized by C.W. Misner, K.S. Thorne, and J.A. Wheeler, Gravitation, box 15.1,
pages 365–371.)
Chapter 3

Conservation of Charge and

the Maxwell Equations

Local conservation of charge (charge goes into or out of region through boundary;
doesn’t disappear inside, simultaneously reappear outside):
I
~ r, t) · n̂ dA = − d Qinside (t)
J(~
Z dt Z
∇ ~ r, t) d r = − d
~ · J(~ 3
ρ(~r, t) d3 r
dt
∇ ~ r, t) = − ∂ρ(~r, t)
~ · J(~ (3.1)
∂t
the “continuity equation”. . . equivalent to local conservation of charge.

3.1 Changing electric field makes magnetic field

The “pre-Maxwell equations”

~ ·E
∇ ~ = ρ/0 ~ ·B
∇ ~ =0
~
~ ×E
∇ ~ = − ∂B ~ ×B
∇ ~ = µ0 J~
∂t

Divergence of Faraday’s law

~ · (∇
∇ ~ = − ∂ (∇
~ × E) ~ · B)
~ =0
∂t
Okay!

23
24CHAPTER 3. CONSERVATION OF CHARGE AND THE MAXWELL EQUATIONS

Divergence of Ampere’s law

~ · (∇
∇ ~ × B) ~ · J~ = −µ0 ∂ρ 6= 0
~ = µ0 ∇
∂t
Not okay!

Maxwell asks: How to fix?

!
~
~ · J~ = − ∂ρ = − ∂ (0 ∇
∇ ~ · E)
~ =∇
~ · −0
∂E
∂t ∂t ∂t

So
~ · J~ 6= 0
∇

but !
~
∂E
~ ·
∇ J~ + 0 =0
∂t

Replace J~ in Ampere’s law with

~
∂E
J~ + 0 .
∂t
Now !
~
∂E
~ ×B
∇ ~ = µ0 J~ + 0 .
∂t
The Maxwell term!

So now the Maxwell equations are

~ ·E
∇ ~ = ρ/0 ~ ·B
∇ ~ =0
!
~ ~
∇ ~ = − ∂B
~ ×E ~ ×B
∇ ~ = µ0 J~ + 0
∂E
∂t ∂t

3.2 Do magnetic monopoles make magnetic field?

Can we add magnetic monopoles? First shot:

~ ·E
∇ ~ = ρe /0 ~ ·B
∇ ~ = µ0 ρm
!
~ ~
~ ×E
∇ ~ = − ∂B ~ ×B
∇ ~ = µ0 J~e + 0
∂E
∂t ∂t

with
~ · J~m = − ∂ρm .
∇
∂t
Can’t stop here!

~ · (∇
∇ ~ = − ∂ (∇
~ × E) ~ = − ∂ (µ0 ρm ) = µ0 ∇
~ · B) ~ · J~m
∂t ∂t
3.2. DO MAGNETIC MONOPOLES MAKE MAGNETIC FIELD? 25

Fix in the same way:

!
~
~ · J~m = − ∂ρm = − ∂ ((1/µ0 )∇
∇ ~ · B)
~ =∇
~ · −(1/µ0 )
∂B
∂t ∂t ∂t

So !
~
∂B
~ ·
∇ µ0 J~m + =0
∂t
~
Replace ∂ B/∂t in Faraday’s law with
~
∂B
µ0 J~m + .
∂t
Now the revised Maxwell equations are

~ ·E
∇ ~ = ρe /0 ~ ·B
∇ ~ = µ0 ρm
~ ~
∇ ~ = −µ0 J~m − ∂ B
~ ×E ~ = µ0 J~e + µ0 0 ∂ E
~ ×B
∇
∂t ∂t

I love the symmetry of these equations: the sources of electric field are electric
charge, magnetic current, and changing magnetic field; the sources of magnetic field
are magnetic charge, electric current, and changing electric field. However, nature
seems not to have taken advantage of this possibility.

Whenever people access strange new places (bore holes under the sea, the surface
of the moon) they look for magnetic monopoles. One was found in the strangest
place of all: California. The laboratory of Blas Cabrera at Stanford University found
evidence for a magnetic monopole on the night of 14 February 1982 (the “Valentine’s
Day Monopole”). Since then, nothing. It was probably a fluke, but who knows. . .
Chapter 4

Energy and Momentum

4.1 Conservation of energy

In section 8.1.2 Griffiths considers a collection of moving charged particles some

region V, with surface S. No particles moving through the surface, and there are no
interactions other than electromagnetic interactions. Let KE represent the kinetic
energy of the moving charged particles. At equation (8.9), Griffths proves Poynting’s
theorem, that
Z   I
d(KE) d 0 ~ 1 ~ 1 ~ × B)
~ · n̂ dA = 0.
+ E+ B d3 r + (E
dt dt V 2 2µ0 µ0 S
The integral over volume had previously been interpreted as the electromagnetic
potential energy localized within this region (Griffiths equations (2.45) and (7.35)).
But that was before we had the Maxwell term. Is this interpretation still legitimate?
Yes, because the Maxwell term generates magnetic field, and the magnetic field does
no work.
Comparing the above to the charge conservation equations
I
d
Qinside + J~ · n̂ dA = 0
dt S

and Z I
d
ρ d3 r + J~ · n̂ dA = 0
dt V S
makes it clear that Poynting’s theorem talks about the conservation of energy. The
~ × B)
term (1/µ0 )(E ~ is the current density, not of charge, but of energy. Sure enough
it has the units of watts/m2 .

~ = 1 (E
S ~ × B)
~ = Poynting vector = current density of EM energy
µ0

26
4.2. THE MAXWELL STRESS TENSOR 27

Comments:

• Energy travels from a battery into a resistor, not through the wires, but through
the empty space between wires. [Griffiths example problem 8.1.]

• Circuit simulation program CircuitSurveyor, by Noah Morris. [Noah A. Mor-

ris and Daniel F. Styer, “Visualizing Poynting vector energy flow in electric
circuits” American Journal of Physics 80 (6) June 2012, pages 552–554.]

• If fields carry energy, it’s a good bet they carry momentum as well. Griffiths
equation (8.29) shows that the momentum density is
1 ~
~g = S.
c2

• Feynman disk and angular momentum conservation. [Richard P. Feynman,

Robert B. Leighton, and Matthew Sands, The Feynman Lectures on Physics,
volume II (Addison-Wesley, Reading, MA, 1964) pages 17-6 and 27-11.]

• Thomson’s dipole. [Griffiths problem 8.19.]

4.2 The Maxwell stress tensor

Fields contain energy density that moves around, and that energy flow is described
~ Fields also contain momentum density that moves
through the Poynting vector S.
around. How is that momentum flow described?

We begin with the familiar: the flow of charge density, which is described through
current density — a vector. After this review, we generalize to the flow of momentum
density, which (we will find) is described through the Maxwell stress tensor.
28 CHAPTER 4. ENERGY AND MOMENTUM

Charge transport

Suppose charge flows uniformly in space and time, like a broad steady wind.

at time t1 at time t1 + ∆t

v∆t
Q

A plug of length v∆t, cross-sectional area A⊥ , passes through the imaginary plane
perpendicular to the “wind”. This plug contains charge

Q = ρ|~v |∆tA⊥ ,

so the current density, namely

charge/time
cross-sectional area
has direction parallel to ~v and magnitude

~ = Q/∆t ρ|~v |∆tA⊥ /∆t

|J| = = ρ|~v |,
A⊥ A⊥
whence
J~ = ρ~v .

Suppose you had a current detector shaped like a tennis racquet, which measured
the current (charge per time) passing through the webbing of the racquet. If the
webbing were oriented parallel to the charge flow, then the detector would read
zero current. You you changed its orientation to the plane facing the flow, then
the detector reading would be a maximum. This is how you could measure the
current density: twist your racquet detector this way and that until its reading is a
maximum, then the current density vector is oriented perpendicular to the webbing,
and has magnitude of the current reading divided by the area of the webbing.

While this technique works, it’s inefficient. You have to take many readings,
homing in on a maximum. You have to twist your body like a contortionist in order
to check out various orientations. Isn’t there an easier way?
4.2. THE MAXWELL STRESS TENSOR 29

There is. In fact, you can find the current density vector by taking just three read-
ings, one with the racquet webbing held perpendicular to x̂, one with the racquet
webbing held perpendicular to ŷ, and one with the racquet webbing held perpendic-
ular to ẑ. The first such experiment is illustrated below:

Charge/time
passing through
this area perpendicular
to flow

y Charge/time θ
passing through
this vertical
area
x

Suppose the marked-off part of the vertical plane has area Ax . Then the marked-
off part of the perpendicular-to-flow plane has area Ax cos θ. The two areas are equal
in charge/time passing through, namely

~ x cos θ = Jx Ax = J~ · x̂ Ax .
|J|A

(In general, the current flowing through area A perpendicular to n̂ is J~ · n̂ A.) Thus,
if you hold the racquet detector with webbing perpendicular to x̂, find the current
passing through, and divide that current by the area, you will find the x-component of
current density, which we call Jx . Similarly you can measure the y and z components,
and with just these three measurements you will find the current density vector
.
J~ = (Jx , Jy , Jz ).

Notice that we never experimentally measure a vector: all our measurements are
of components of a vector projected on various unit vectors. By putting together
a series of such measurements (either through three components projected onto the
basis vectors, or through homing in on the maximum), we can uncover the vector.

I hope that nothing I’ve said in this section is new to you. I just wanted to remind
you of how current flow is defined and what it means to be a vector.
30 CHAPTER 4. ENERGY AND MOMENTUM

Momentum transport

Suppose momentum flows uniformly in space and time, like a broad steady wind. If
this momentum were carried by matter, like a literal wind or rain, then the motion
of the air or raindrops (~v ) would be parallel to the momentum being transported (~
p).
But the case of electromagnetic fields is less familiar, so the diagram below admits
that possibility that the momentum might not be parallel to the direction the field
is moving.

at time t1 at time t1 + ∆t

v∆t
p

The reasoning proceeds exactly as it did for charge transport, except that charge
density ρ is replaced by momentum density ~g . A plug of length v∆t, cross-sectional
area A⊥ , passes through the imaginary plane perpendicular to the “wind”. This plug
contains momentum
p~ = ~g |~v |∆tA⊥ ,
so the current density of x-momentum, namely
x-momentum/time
cross-sectional area
has direction parallel to ~v and magnitude
px /∆t gx |~v |∆tA⊥ /∆t
= = gx |~v |,
A⊥ A⊥
whence the current density of x-momentum is

gx~v .

Of course, you can do the same calculation for y-momentum density and z-momentum
density. In total, the momentum current density is

~g~v .

What is this thing we get by multiplying two vectors in this way? It’s not a
scalar, which would result from a dot product, and it’s not a vector, which would
4.2. THE MAXWELL STRESS TENSOR 31

result from a cross product. This is called an “outer product” or a “tensor product”
and the result is called a tensor. James Clerk Maxwell recognized the importance
of this particular combination, except that he defined it with a negative sign. The
“Maxwell stress tensor” is defined through
↔
T = −~g~v .

The physical meaning of the Maxwell stress tensor is exactly as described above.
You could image a momentum detector shaped like a tennis racquet, and everything
we said above about charge current density measurements (charge current density
is a vector) would apply to momentum current density measurements (momentum
current density is a tensor). In particular, if you look at components in some given
basis you’ll find
     
Txx Txy Txz gx gx vx gx vy gx vz
↔.   
T =  Tyx Tyy Tyz  = −  gy  vx vy vz = −  gy vx gy vy gy vz  .
    

Tzx Tzy Tzz gz gz vx gz vy gz vz

Thus, for example, −Txz represents the current of x-momentum passing through a
plane perpendicular to z.

As with vectors, we never experimentally measure tensors: all our measurements

are of components of tensors. And as with vectors, this doesn’t mean that tensors
are useless or unnatural. It just means that they’re less familiar.

Exercise: Suppose the material is indeed, say, wind, so that the momentum
transported is always in the direction of the motion of the material flow. Is the
tensor then represented by a diagonal matrix? Symmetric? Antisymmetric? What
would the tensor components look like if the basis were oriented so that the x direction
points along the direction of the wind?

Answer: In this case the tensor would be represented by a symmetric matrix. If

the basis were oriented as described, all matrix elements would vanish but for the
Txx element.
32 CHAPTER 4. ENERGY AND MOMENTUM

Force

We’ve considered the flow of momentum, carried in a field, across a plane. But there’s
another possibility. Remember that momentum is conserved only in the absence of
external force. We could start with nothing, and end up with a plug of momentum,
by exerting a force.

at time t1 at time t1 + ∆t

v∆t
p

In this figure, think of the lower left portion as occupied by some charges and
currents. After some time has passed, there is a plug of fields on the upper right.
The force exerted by these charges and currents on the field, in order to produce this
momentum, is
p~ ~g |~v |∆tA⊥ ↔
F~ = = = ~g~v · n̂ A⊥ = − T ·n̂ A⊥ .
∆t ∆t
By Newton’s third law, the force on the charges and currents by the field is thus
↔
F~ = T ·n̂ A⊥ .

If the surface is not uniform, then the total force on the charges and currents is
found by integrating over the surface
Z
↔
F~total = T ·n̂ dA
S

Cute trick! Instead of integrating the force density over the volume of the charges
and currents, you only need to surface integrate the Maxwell stress tensor over the
surface.

One of the traditional difficult problems of physics is “the problem of the damn
rotating cylinder, one-third full of dielectric [fluid], in a magnetic field”. [Leon Led-
erman, “An open letter” Physics Today 47 (3) March 1994, 9–10]. I am not going
to assign this problem to you. But when, eventually, it is assigned to you, solve the
problem using the Maxwell stress tensor. (Also, check your answer by investigating
what happens when the dielectric constant  equals 0 . That’s the mistake I made.)
4.2. THE MAXWELL STRESS TENSOR 33

Connection to Griffiths

The motivation of the Maxwell stress tensor given here differs dramatically from
the one given by Griffiths in section 8.2.2. I think that this way carries a lot more
insight, which is why I use it. But Griffiths’s approach has two advantages. First,
my way holds only for static situations. (I said “steady wind” in the first sentence.)
For a non-static situation, the total force on the charges and currents is (Griffiths
equation 8.20) Z Z
~
↔ d ~ d3 r,
Ftotal = T ·n̂ dA − 0 µ0 S
S dt V
~ is the Poynting vector.
where S

Griffiths’s second advantage is that in my approach, everything depends on “the

velocity of the fields”. A nice, compact concept, but how are you supposed to cal-
culate it?! Griffiths gives a formula (equation 8.17) for the stress tensor in terms of
fields alone:
~ − 1 E2 I + 1 B
↔  ↔  ↔
T = 0 E ~E ~B~ − 1 B2 I ,
2 µ0 2

↔
where I represents the identity tensor, with components δij . This formula makes it
clear that the stress tensor for electromagnetic field, like the stress tensor for wind
or rain or fluid flow, is symmetric.

Field stress

If there’s a volume of charges and currents and fields bounded by surface S, the net
force on this volume is
Z
~
↔ d
Ftotal = T ·n̂ dA = (~ pmatter + p~fields ).
S dt
If there is no matter within the volume, only fields, then the net force must vanish.

Think about a region of uniform electric field (and any field can be made uniform
by thinking of a small enough region). Orient the coordinate axes so that the x axis
~ In this basis E .
~ =
is parallel to E. (E, 0, 0) so
   
E2 0 0 E2 0 0
↔ h ↔i .
T = 0 E ~E
~ − 1 E 2 I = 0 
 0 0 0  − 12  0 E 2 0 
   
2
0 0 0 0 0 E2

or finally  
1 0 0
↔.
1 2
T = 2 0 E  0 −1 0 . (4.1)


0 0 −1
34 CHAPTER 4. ENERGY AND MOMENTUM

Now think of a little cube, length L on a side, within the electric field.

The force on the right face of this cube (where n̂ = x̂) is

    
1 0 0 1 1
↔
2 . 2 2 2 2  . 2 2
T ·x̂ L = 21 0 E L  0 −1 0   0  = 12 0 E L  0  = 12 0 E L x̂.
 

0 0 −1 0 0

Similar calculations can be performed on the other cube faces. You need to keep
your signs straight. The resulting forces are shown below.

As promised, the net force cancels to zero. The pair of forces acting parallel to the
electric field are tugging the cube apart; the pairs of forces acting perpendicular to
the electric field are squishing their two sides together. My teacher Mark Heald liked
to say that “electric field lies behave like furry rubber bands”: the tugs parallel to the
field lines act like a taut rubber band, but adjacent field lines don’t get too close to
each other because the squishing, like fur on the bands, prevents them from getting
too close. They want to be as short as possible (rubbery) but don’t want to be near
each other (furry).

The same calculation can of course be performed for magnetic fields, with the
same result except that 0 is replaced with 1/µ0 .

Why do the field lines for electric dipoles and magnetic dipoles look so much alike?
Not because the sources are similar. . . they’re not (one source is a pair of charges and
the other a loop of current). Not because Coulomb’s law and the Biot-Savart law are
similar. . . they’re not (one has cross products and the other doesn’t). It’s because
the stress tensor for E ~ and for B
~ are similar.
4.2. THE MAXWELL STRESS TENSOR 35

Generalizations

The flow of a scalar (like charge density ρ) is described through a vector (like current
~
density J).

The flow of a vector (like momentum density ~g ) is described through a tensor

↔
(like the negative of the Maxwell stress tensor − T ).

What mathematical tool would one use to describe the flow of a tensor?

I ask this question not to make your brain hurt, but to open your mind to more
and richer possibilities. The tensor that we’ve discussed, namely the Maxwell stress
tensor, is an example of a “rank-2 tensor”. In three dimensions, a rank-2 tensor can
be described using 9 projections, called components, which are conveniently presented
in a 3 × 3 matrix. The flow of a rank-2 tensor is described through a “rank-3 tensor”.
In three dimensions, a rank-3 tensor can be described using 27 components, and
there’s no real convenient way to present one on flat paper. (Sometimes I use a stack
of three index cards, on each of which I write a 3 × 3 matrix. But even this is not
really effective.) From this point of view, a vector is a rank-1 tensor and a scalar is
a rank-0 tensor. In general, a rank-r tensor in d dimensions is specified through dr
components.

People get upset by the word tensor, but they shouldn’t:

The motion of a point is described by a vector

The motion of a vector is described by a (rank-2) tensor
The motion of a (rank-2) tensor is described by a rank-3 tensor

and so forth. It’s nothing more than that.

36 CHAPTER 4. ENERGY AND MOMENTUM

Appendix: Definition of tensor through components

Back on page 11 I defined a vector as “an entity with components xi that transform
from one basis to another through
3
X
x0i = Rik xk . ”
k=1

Can I generalize this definition to a tensor?

I can. We’ve seen how to write a tensor through an outer product as, for example,
↔
T = −~g~v .

In terms of components, this means

Tij = −gi vj .

Because the components of ~g and of ~v transform as

3
X 3
X
gi0 = Rik gk and vj0 = Rj` v` ,
k=1 `=1

↔
the components of a rank-2 tensor like T transform as
3 X
X 3
Tij0 = Rik Rj` Tk` .
k=1 `=1

Generalizing, the components xijk of a rank-3 tensor transform as

X 3
3 X
3 X
x0ijk = Ri` Rjm Rkn x`mn .
`=1 m=1 n=1

and so forth.

Appendix: What is stress?

How did the entity we’re talking about get the name “stress” tensor?

The word “stress”, in physics, means internal force per area. Grab a pencil by the
eraser end and hold it horizontally. In your mind, think of a “cut plane” separating
the pencil into the half closer to the eraser and the half closer to the tip. (This plane
is not artificial at all: you know that if you select a plane through an eraser it will
hit mostly empty space.) The tip half is being pulled down by gravity. Why doesn’t
it accelerate downward? Because there’s an internal upward force exerted through
the cut plane by the eraser half on the tip half. (This is the sum of all the forces
4.2. THE MAXWELL STRESS TENSOR 37

by atoms in the eraser half on atoms in the tip half.) With a finger on your other
hand, press the pencil tip downward, while keeping it horizontal. Now this internal
upward force has increased. The internal force per area across a cut plane is called
the “stress”.

If the stress is always perpendicular to the cut plane, as it is in a static fluid, then
this stress is called “pressure”. Pressure is a form of stress that can be represented
through a scalar. But internal forces in a solid, or in a moving fluid, are not always
pressure stresses: They will have a direction that depends on the orientation of the
cut plane, and they must be represented through a tensor.

If you’re building a bridge, or building, or railroad locomotive, it’s essential to keep

track of the stresses in your structure, and to know at what stress your material will
fracture. This was a big task for physics in the nineteenth century, and when James
Clerk Maxwell attended classes at the Universities of Edinburgh and of Cambridge
(1847–1856) he must have taken courses on this topic.

When Maxwell researched electromagnetism, it was only natural that he would

apply what he knew about mechanical stresses to those researches. He considered
all electromagnetic phenomena, not just light, to reflect forces and displacements
within the “luminiferous ether”. When Maxwell needed to keep track of internal
forces within the ether, he used what we now call the “Maxwell stress tensor”.

I have read part of Maxwell’s 1873 Treatise on Electricity and Magnetism. It is

remarkable not only for its insight, but for how differently Maxwell framed electro-
dynamics from how we frame it today.
Chapter 5

Electromagnetic Waves

5.1 One-dimensional waves

A string is stretched taut. Call position along the straight string x. Now the string
is pulled away from straightness in some fashion: the displacement from straightness
is called η(x, t). How does this displacement change with time?

To high accuracy, this displacement obeys “the wave equation”

∂2η 1 ∂2η
2
= 2 2. (5.1)
∂x c ∂t
(The wave equation does not describe exactly the motion of a disturbed spring.
In fact, the wave equation is not exactly correct even for electromagnetic waves in
vacuum, because of quantum mechanical effects. But it is highly accurate for both
situations.) I won’t derive the wave equation for a taut string (see W.C. Elmore and
M.A. Heald, Physics of Waves (1969) section 1.1), but instead focus on solving it.

I genuinely enjoy solving ordinary differential equations, but partial differential

equations tend to make my stomach churn. Can we make any progress on this
equation? Jean-Baptiste le Rond d’Alembert led the way. Inspired by the algebraic
factorization a2 − b2 = (a − b)(a + b), he decided to write the wave equation as

∂2η 1 ∂2η
  
∂ 1 ∂ ∂ 1 ∂
0= − = − + η(x, t).
∂x2 c2 ∂t2 ∂x c ∂t ∂x c ∂t
This suggests a change of variable to

u = x − ct
v = x + ct

38
5.1. ONE-DIMENSIONAL WAVES 39

or, what is the same thing,

1
x = (v + u)
2
1
t = (v − u).
2c
In terms of these variables, the equation above is
∂ 2 η(u, v)
0=4
∂u ∂v
and the solution of this equation is clear from inspection: If f ( ) and g( ) are arbitrary
one-variable functions, then the general solution is

η(u, v) = f (u) + g(v).

In short, d’Alembert’s solution shows that the general solution to the wave equa-
tion (5.1) above is
η(x, t) = f (x − ct) + g(x + ct). (5.2)
Wow, that was easy. Three cheers for d’Alembert!
Character of d’Alembert’s solution. But after the cheering stops we need to
dissect this general solution to see what it’s telling us about nature. Begin with the
case g(v) = 0. If f (u) looks like

f(u)

then η(x, t) = f (x − ct) looks like

η(x,t), t = 0

x
x0
η(x,t), t > 0

x
x0

This is a shape-preserving pulse wave moving to the right at speed c!

40 CHAPTER 5. ELECTROMAGNETIC WAVES

Meanwhile, what does this wave look like as a function of t? Suppose I stand at
point x0 and let the wave wash over me. The function η(x0 , t) looks like the mirror
image of the above.

η(x0,t)

Move on to the case f (u) = 0, so η(x, t) = g(x − ct). This is a shape-preserving

pulse wave moving to the left at speed c:

η(x,t0)

η(x0,t)

t
5.1. ONE-DIMENSIONAL WAVES 41

What if neither f (u) nor g(v) vanish? This is called “superposition”! In the
example below, a big semicircular wave moves right, a small one moves left. When
they cross over each other, the two waves add. Then each continues independently
on its own way as if they had never known each other: “two ships that pass at night”.

beginning x

middle x

end x

The same holds if the small wave moving left happens to have a downward rather
than an upward displacement.

beginning x

middle x

end x
42 CHAPTER 5. ELECTROMAGNETIC WAVES

What if the two waves are the same size? In this case their displacements cancel
out completely as they pass over each other.

beginning x

middle x

end x

It would be very funny if this were done as a lecture demonstration and you
happened to walk into class late just at the time marked “middle”. You would see a
straight string with no displacement at all, then two semicircular waves would pop
into being on the straight string, the “up” wave moving right and the “down” wave
moving left! This is a puzzle. How can a straight, unstretched spring just pop two
semicircular waves into existence? Where does the energy for those two waves come
from?

I’ll resolve this puzzle on the next page. But think about it for a few moments
before turning the page, both to make sure you understand why it’s puzzling and to
see if you can’t resolve the puzzle yourself.
5.1. ONE-DIMENSIONAL WAVES 43

You know from introductory mechanics that to specify the state of a particle you
must specify both its position and its velocity. The wave pictures on the previous
pages show only the positions of the particles that make up the string, and not their
velocities, so they don’t specify the state. (There are black velocity arrows, but they
signify the velocity of the waveform, not the velocity of the particles on the string.)

Paint a green dot on a single bit of string, and think about the motion of that dot
as the rightward moving, “up” wave washes over it. That dot moves first up, then
down. When the leftward moving, “down” wave washes over the green dot, it moves
first down, then up. For the “beginning” and “end” situations, when the two waves
are well-separated, the velocities of representative dots are shown using red arrows
in the figure below.

beginning x

middle x

end x

Now think about what happens to a single string element at the “middle” situ-
ation. The total displacement of a dot on the string is the sum of the displacement
due to the two superposing waves. And the same is true of the velocities. But at the
“middle” time, when the string element displacements sum they cancel out to zero,
while when the string element velocities sum they actually increase.
44 CHAPTER 5. ELECTROMAGNETIC WAVES

beginning x

middle x

end x

When you walked late into class, you saw the string at an instant, as in a snapshot,
and of course a snapshot can’t show the velocities. The situation at the middle is
indeed a straight, unstretched string, but it’s not a straight, unstretched spring at
rest. It’s the motion of the string (invisible in the snapshot) that enables it to pop
two semicircular waves into existence.

Challenge: Can you show that if the waveform is y(x, t) = f (x − ct), and if
df (x)
f 0 (x) = ,
dx
then the velocity of the string element at (x, t) is −cf 0 (x − ct)?

Sine waves. Suppose f (u) = A sin(ku) so that the wave is

f (x − ct) = A sin(kx − ωt).

Here kc = ω (“Kansas cows eat wheat”). Also the wavelength is λ = 2π/k and
the period is T = 2π/ω. This is a solution to the wave equation, but it doesn’t
correspond to any physical wave, because the function sin(kx − ωt) extends over all
space and all time. This wave started infinitely far in the past and will keep going
for ever and ever, amen. Real waves are finite in space (waves on the ocean end when
they hit the beach) and of course finite in time.1 However, the pure sine wave can
1 Following songwriters Nickolas Ashford and Valerie Simpson, (“Ain’t No Mountain High

Enough,” 1966, sung most famously by Marvin Gaye and Tammi Terrell) I like to say that there
“ain’t no ocean wide enough, ain’t no string long enough” to carry a pure sine wave.
5.1. ONE-DIMENSIONAL WAVES 45

be an accurate approximation to a wave that lasts for many periods. (And because
the period of light is so small, most light waves fit this description.)
Superposition of sine waves. Suppose the two waves superposing are not
semicircular pulses, but instead sine waves:

f (x − ct) = A sin(kx − ωt)

g(x + ct) = A sin(kx + ωt).

Then the total wave is

η(x, t) = A sin(kx − ωt) + A sin(kx + ωt). (5.3)

Okay, so that’s the sum, but how can we understand the character of η(x, t)? If you
knew the trigonometric sum and difference formulas, you might be able to make some
progress. But I forgot those formulas the minute I left high school (if not before).
Instead, I like to perform trigonometric manipulations using complex arithmetic and
Euler’s formula that, for θ real,

eiθ = cos(θ) + i sin(θ).

First, establish two consequences of Euler’s formula:

sin(θ) = =m eiθ and cos(θ) = 12 (eiθ + e−iθ ).



Now go at it:

η(x, t) = A sin(kx − ωt) + A sin(kx + ωt) (5.4)

n o
= A =m ei(kx−ωt) + ei(kx+ωt)
= A =m eikx [e−iωt + eiωt ]


= A =m eikx [2 cos(ωt)]


= 2A cos(ωt) =m eikx


= 2A cos(ωt) sin(kx). (5.5)

In other words, y(x, t) is just a sine function of x, but with amplitude that varies
with time: The amplitude is 2A at t = 0, then diminishes to 0 at t = 21 π/ω, becomes
−2A at t = π/ω, 0 again at t = 32 π/ω, and returns to 2A at t = 2π/ω.

η(x,t)

x
46 CHAPTER 5. ELECTROMAGNETIC WAVES

I don’t know about you, but I never would have guessed that this behavior is hidden
within the equation (5.4). This does not look like “two ships that pass at night”, but
it is! These are called “standing waves”.
It’s tempting to overgeneralize d’Alembert’s solution. It seems so natural:
The general solution of a second-order linear ODE has two adjustable parameters,
which are set by the initial or boundary conditions. A natural generalization is “A
second-order linear PDE has two adjustable functions, which are set by the initial or
boundary conditions.” This is wrong.
For example a wave η(x, y, t) moving in two dimensions (like the motion of a
taut drumhead, or ripples on the surface of a puddle) satisfies, to high accuracy, the
two-dimensional wave equation
∂2η ∂2η 1 ∂2η
+ = ,
∂x2 ∂y 2 c2 ∂t2
1 ∂2η
∇2 η = .
c2 ∂t2
What is the general solution to this equation? One is tempted to guess

f1 (x − ct) + g1 (x + ct) + f2 (y − ct) + g2 (y + ct). (5.6)

This is indeed a solution, as you can readily check, but it’s not the most general
solution. Drop a pebble into the puddle: a circular wave emerges. This circular wave
is not shape-preserving: it gets smaller as it spreads from the pebble. This waveform
is not of the form (5.6), so form (5.6) can’t be the most general solution.

5.2 From Maxwell equations to wave equations

In a region with no charges or currents, the Maxwell equations are

~ ·E
∇ ~ =0 ~ ·B
∇ ~ =0
~ ~
∇ ~ = − ∂B
~ ×E ~ = µ0 0 ∂ E
~ ×B
∇
∂t ∂t
An equation like Faraday’s law on the bottom left,
~
~ = − ∂B ,
~ ×E
∇
∂t
doesn’t tell you much, because you don’t know what B ~ is! But, because of the
Ampere-Maxwell law on the bottom right, you do know what ∇ ~ ×B~ is. So let’s take
the curl of Faraday’s law:
!
∂ ~ ∂ ~
∂E ∂2E ~
~ ~ ~
∇×∇×E =− ∇×B =− ~ µ0 0 = −µ0 0 2 .
∂t ∂t ∂t ∂t
5.3. WAVES IN MEDIA 47

The right hand side looks like it belongs in a wave equation, but the left hand side
looks dramatically foreign. Remember the vector calculus identity

~ ×∇
∇ ~ × F~ = ∇(
~ ∇~ · F~ ) − ∇
~ 2 F~ .

(Normally the Laplacian ∇~ 2 is applied to scalar functions. Here we apply it to each

of the three Cartesian components in turn.) For our case ∇ ~ ·E
~ = 0, so

2~
∇ ~ = µ0 0 ∂ E .
~ 2E (5.7)
∂t2
This is the wave equation with
1
c= √ . (5.8)
µ0 0

You can apply exactly the same reasoning to the magnetic field to find
2~
∇ ~ = µ0 0 ∂ B .
~ 2B (5.9)
∂t2

So much fuss is made of these two wave equations that you might forget that the
Maxwell equations are more fundamental. The two wave equations can be derived
from the Maxwell equations but not vice versa. That’s because the Maxwell equations
contain more information than the two wave equations. For example, Griffiths shows
(section 9.2.2) that monochromatic plane electromagnetic waves are traverse waves:
~ and B
If the direction of wave propagation is k̂, then E ~ are perpendicular to k̂.
~ ~
Furthermore E is perpendicular to B. Still further, at any one place and time the
~ = c|B|.
magnitudes of the fields are related through |E| ~ None of these facts can be
derived from the two wave equations by themselves.
Finally, the analysis of this section shows that electromagnetic waves exist, but
doesn’t show how to use charges and currents to get them started. That’s the task
of chapter 7, “Resultant Potentials and Fields”.

5.3 Waves in media

Electromagnetic waves in vacuum: Start with the wave equation

∂2η 1 ∂2η
= ,
∂x2 c2 ∂t2
this generates the “shape preserving” solutions

η(x, t) = f (x − ct) + g(x + ct)

which specializes to

η(x, t) = Aei(kx−ωt) with kc = ω.

48 CHAPTER 5. ELECTROMAGNETIC WAVES

(There is an understood “text” behind this equation: Take the real or the imaginary
part. No wave lasts forever, but this applies approximately to any wavepacket long
compared to λ.)

In a medium, the process goes almost the other way around: Think of waves like

η(x, t) = Aei(kx−ωt) .

This sine wave moves with speed cm (ω) = ω/k: In an infinite (or very long) sine
wave, the crests move at speed cm (ω).

[[You will ask: Why does wave speed depend upon ω alone? The answer is, it
might not. In an inhomogeneous medium, cm will depend upon location. In a non-
linear medium, cm will depend upon amplitude A. But it’s an experimental result
that for many uniform media, cm is a function of ω alone.]]

There is not, and cannot be, an equation like

∂2η 1 ∂2η
= !
∂x2 c2m (ω) ∂t2

One often sees experimental results like

cm

0 ω
0

It makes sense that at high frequencies the speed of light in medium cm approaches
the speed of light in vacuum c: When the jiggling is very fast the atoms in the
medium cannot react to the light, so the light passes through as if the atoms were
not there.

Okay, so this is how infinitely long sine waves behave. But what about real,
finite waves? In this case you have to start with the initial finite wave, Fourier
decompose it into sine components, move each component forward at its own speed
cm (ω), then Fourier recompose it to find the wave at the final time. [Very much
like quantum mechanics, where, to solve time evolution problems, you start with the
5.3. WAVES IN MEDIA 49

initial wavefunction, expand in energy eigenstates (with eigenvalue En ), propagate

each energy eigenstate forward in time through its phase factor e−(i/h̄)En t , then
recombine them.]
So, to solve time evolution problems, you must know the details of η(x, t0 ) and
of cm (ω). The wave η(x, t) might spread out, compact in, change shape, split in
two, etc. Lots of strange behaviors are possible, showing the many, various, and
diverse phenomena of physics. As such, there is no single “wave speed”. In his book
Wave Propagation and Group Velocity, Léon Brillouin discusses phase velocity, group
velocity, velocity of energy transport, signal velocity, and forerunner velocity.
But most of the time the packet spreads out or “disperses”. Graphs like the
previous one are called “dispersion curves”. For example

cm
"normal dispersion"

"anomalous dispersion"

0 ω
0

What?! Speeds cm greater than c?!?!

Yes. This is the speed of the crests in an infinite sine wave. When you watch the
wave come in, it’s been known for a long time when the crests are going to arrive.
An infinite sine wave in fact transmits no information whatsoever!
By the way, anomalous dispersion is in fact more common than normal dispersion.
Phase velocity and group velocity. To get an idea of the variety of behaviors
possible through this decompose-recompose technique, consider a simple initial wave,
composed of only two Fourier components:
n o
η1 (x, t) = A cos(k1 x − ω1 t) = A <e ei(k1 x−ω1 t)
n o
η2 (x, t) = A cos(k2 x − ω2 t) = A <e ei(k2 x−ω2 t) . (5.10)

If the two components are close in frequency, it makes sense to define the average
wavenumber k0 and the average frequency ω0 such that

k1 = k0 − 21 ∆k ω1 = ω0 − 21 ∆ω
k2 = k0 + 12 ∆k ω2 = ω0 + 21 ∆ω. (5.11)
50 CHAPTER 5. ELECTROMAGNETIC WAVES

In terms of these new variables,

n 1
o
η1 (x, t) = A <e ei(k0 x−ω0 t)− 2 i(∆k x−∆ω t)
n 1
o
η2 (x, t) = A <e ei(k0 x−ω0 t)+ 2 i(∆k x−∆ω t) . (5.12)

The total wave is then

n h 1 1
io
η1 + η2 = A <e ei(k0 x−ω0 t) e− 2 i(∆k x−∆ω t) − e+ 2 i(∆k x−∆ω t)
n
o
= A <e ei(k0 x−ω0 t) 2 cos 21 (∆k x − ∆ω t)


= 2A cos (k0 x − ω0 t) cos 12 ∆k x − 12 ∆ω t ,

(5.13)

the product of two cosine waves!

So far we have not made use of our idea that ω1 and ω2 are close. Adopting
this idea, we see that ∆k  k0 , so ∆k corresponds to a long wavelength, so the
right-hand factor in equation (5.13) varies slowly compared to the left-hand factor.
We can plot the sum wave η1 + η2 using the “envelope trick”. In blue I plot the
slowly varying envelope 2A cos 12 ∆k x − 12 ∆ω t . In gray I plot the negative of this

envelope. In brown I plot the sum wave η1 + η2 :

The envelope wave — the humps — has a wavenumber of 21 ∆k and a frequency

1
of 2 ∆ω,
so it moves at the so-called “group velocity”
1
2 ∆ω dω
vg = 1 ≈ , (5.14)
2 ∆k
dk
whereas the short-wavelength wave within this envelope — the wiggles — moves at
the so-called “phase velocity” (Griffiths calls it “wave velocity”)
ω0
vp = . (5.15)
k0
This argument is based on a simplification (only two Fourier components) but it
contains the seeds of general truths:
5.3. WAVES IN MEDIA 51

group velocity: speed of hump

phase velocity: speed of wiggles within hump
usually information and energy moves at the group velocity, not always!
usually vg < c, not always!
when vg > c, the hump breaks up too fast to carry information

This graph shows the curve of ω(k) for a hypothetical medium. For k large,
wavelength and period small, this medium behaves like vacuum, where ω = ck.

slope = c

slope = vg
ω2 slope = vp
ω1

0 k
k1 k2
0

I show the wave numbers k1 and k2 that appear in equations (5.10). We’ve discussed
what happens when these are the sole Fourier components, but I think you can see
that if you had a spread of wave numbers the qualitative behavior would be much
the same.
Wave speed faster than infinity! Here’s an example of fantastic wave behavior
told to me by Steven Wong.2
Everyone knows that a light wave in a medium might make that medium more
absorptive or, in lay terms, darker. This is why people lie in the sun to get a
tan. Steven developed a material that would switch from pure transparent to pure
absorptive in about half a nanosecond, the time required for light to travel about
10 centimeters.
Steven flashed a light pulse two meters long (6 nanoseconds worth) into a wafer
of his material. The first 10 centimeters of this light pulse made it through, but the
2 S.Wong and D. Styer, “Answer to Question #52: Group Velocity and Energy Propagation,”
Am. J. Phys. 66, 659–661 (1998).
52 CHAPTER 5. ELECTROMAGNETIC WAVES

rest was blocked. The entire emerging pulse departed while 1.9 meters of incoming
pulse had yet to encounter the medium. The “center of the light pulse”, emerged
from the wafer before the “center of the light pulse” had reached the wafer!

But information in this case is carried by the forerunner that makes it through,
not by the mathematical “center of the light pulse”. The information travels safely
at or below light speed.

The wide variety of wave behavior. We have only scratched the surface.
If you examine quantal waves, deep-water waves, shallow-water waves, rogue waves,
seismic waves, and electromagnetic waves within waveguides, or within plasmas, or
within magnetic materials, you will encounter behaviors3 beyond your wildest imag-
ination.

You will understand rainbows, sun dogs, rings around the moon, the green flash
the precedes sunrise, and mirages. You will understand the erie sounds made by
frozen lakes. You will see with a new eye boat wakes and ocean breakers. Both your
everyday life and your scientific life will be enriched.

3 See,
for example, Anatoly Patsyk, Uri Sivan, Mordechai Segev, and Miguel A. Bandres, “Ob-
servation of branched flow of light” Nature 583 (2 July 2020) 60–65.
Chapter 6

Potentials and Gauges

6.1 Math

This chapter relies on two mathematical theorems:

Vanishing curl. If a vector function F~ (~r) has vanishing curl everywhere, then
~ F~ (~r) = 0
that function is the gradient of some scalar function f (~r). In other words ∇×
implies F~ (~r) = ∇f
~ (~r).

Vanishing divergence. If a vector function F~ (~r) has vanishing divergence ev-

~ r). In other
erywhere, then that function is the curl of some other vector function A(~
~ ~ ~ ~ ~
words ∇ · F (~r) = 0 implies F (~r) = ∇ × A(~r).

These theorems hold for any reasonable function. If you are interested in unrea-
sonable functions, look up “Helmholtz decomposition”.

53
54 CHAPTER 6. POTENTIALS AND GAUGES

6.2 Potentials
Electrostatics Magnetostatics

~ ·E
∇ ~ = ρ/0 ~ ·B
∇ ~ =0
~ ×E
∇ ~ =0 ~ ×B
∇ ~ = µ0 J~

~ ×E
On the electrostatic side, start with ∇ ~ = 0. This means that E
~ = −∇V
~ , so
Gauss’s law becomes Poisson’s equation

~ 2 V = −ρ/0 .
∇

It is much easier to work with electrostatic potential than with electric field. For
example the potential due to source charge q at the origin is
1 q
V (r) = .
4π0 r
We don’t have to mess with vectors, this is a scalar!
~ ·B
On the magnetostatic side, start with ∇ ~ = 0. This means that B
~ =∇
~ × A,
~ so
Ampere’s law becomes

~ ×∇
∇ ~ ×A ~ = µ0 J~
~ 2A
∇ ~ − ∇(
~ ∇~ · A)
~ = −µ0 J~

which is almost Poisson’s equation. It is somewhat easier to work with this vector
potential than with magnetic field. For example the vector potential due to source
charge q moving with velocity ~v at the origin is

~ µ0 q~v
A(r) = .
4π r
This is still a vector, but at least it doesn’t involve a cross product.
6.2. POTENTIALS 55

Electrodynamics

~ ·E
∇ ~ = ρ/0 ~ ·B
∇ ~ =0
~ ~
~ ×E
∇ ~ = − ∂B ~ = µ0 J~ + µ0 0 ∂ E
~ ×B
∇
∂t ∂t

~ ·B
In this case, start with ∇ ~ = 0. This still means that

~ r, t) = ∇
B(~ ~ × A(~
~ r, t), (6.1)

so Faraday’s law becomes

~
∂B
~ ×E
∇ ~ = −
∂t
~ ×E~ ∂ ~ ~
∇ = − ∇×A
! ∂t
~
~ ×
∇ ~ + ∂A
E = 0
∂t
~
~ + ∂A
E ~
= −∇V
∂t
or
~
E(~ ~ (~r, t) − ∂ A(~r, t) .
~ r, t) = −∇V (6.2)
∂t
What happens with the other two Maxwell equations? Gauss’s law becomes

~ ·E
∇ ~ = ρ/0
!
∂A~
~ ·
∇ ~ −
−∇V = ρ/0
∂t
~ 2V + ∂ ∇
∇ ~ ·A
~ = −ρ/0 (6.3)
∂t
which is a little more involved than Poisson’s equation.

Meanwhile the Ampere-Maxwell equation becomes

∂E~
~ ×B
∇ ~ = µ0 J~ + µ0 0
∂t !
∂ ∂ ~
A
~ ×∇
∇ ~ ×A
~ = µ0 J~ + µ0 0 ~ −
−∇V
∂t ∂t
2~
~ ∇
∇( ~ · A)
~ −∇~ 2A ~ ∂V + µ0 0 ∂ A = µ0 J~
~ + µ0 0 ∇
∂t ∂t2
2~
 
~ 2~ ∂ A ~ ~ ~ ∂V ~
∇ A − µ0 0 2 − ∇ ∇ · A + µ0 0 = −µ0 J. (6.4)
∂t ∂t

Notice that if magnetic monopoles exist, none of these steps would be possible.
56 CHAPTER 6. POTENTIALS AND GAUGES

Thinking mathematically, we have gone from a description of electromagnetism

~ B)
in terms of (E, ~ (six components) to a description in terms of (V, A)
~ (four compo-
nents). There are eight component Maxwell equations. (Gauss’s law is one equation,
Faraday’s law is a vector equation whence three components, that’s four, and then
there are four on the magnetism side as well.) There are four component equations
(equations (6.3) and (6.4)) for the scalar and vector potentials.

Maxwell approach: 6 functions determined by 8 equations

Potential approach: 4 functions determined by 4 equations

Thinking physically, it’s pretty simple to have an idea of scalar potential —

indeed “voltage” is known to little kids who don’t know about electrostatic field and
who couldn’t line integrate to save their lives. But the idea of vector potential is
more iffy. James Clerk Maxwell was interested in cases far from electrostatics, where
the primary source of E~ was not −∇V~ but instead

~
~ = − ∂A .
E
∂t
He compared that equation to
d~
p
F~ =
dt
and called A~ “electromagnetic momentum”. (He was not someone to let a minus sign
get in his way.) This is not momentum in the sense of
Z
1 ~ d3 r,
S
c2
but momentum in the sense of “oomph”. (When George H.W. Bush was running for
president in 1980, he used the term “momentum” in this sense when he said of his
campaign’s “oomph” that “what we will have is momentum. We will look forward
to big mo”.)
6.2. POTENTIALS 57

When Maxwell gave names to electromagnetic quantities, he listed them in order

of importance. This is how magnetic field got its strange name B.~ And what did
Maxwell consider the most important of all quantities? That’s right, what we today
~
call vector potential, A.

James Clerk Maxwell, A Treatise on Electricity and Magnetism,

volume II (Clarendon Press, Oxford, 1873) pages 236–237.

(You see from this table that it is also Maxwell who gave the name J~ to current
density.)
~ B,
Maxwell, in fact, thought of E, ~ A,
~ and V as fields of comparable footing. It was
Oliver Heaviside who, in 1888, made the distinction between fields and potentials that
we know and love today. [The self-educated Heaviside also invented coaxial cable, the
concept of impedance, the use of complex numbers to represent sinusoidal current, the
Heaviside step function and, remarkably, the Lorentz force law (F~ = q~v × B)!
~ In 1902
Heaviside surmised the existence of the ionosphere to explain Guglielmo Marconi’s
1901 observation that radio waves can propagate over the horizon. In Britain the
58 CHAPTER 6. POTENTIALS AND GAUGES

ionosphere is sometimes called the “Heaviside layer”. See Bruce J. Hunt, “Oliver
Heaviside: A first-rate oddity” Physics Today 65 (November 2012) pages 48–54.]

Heaviside made the distinction between fields and potentials when he realized
that he could change A ~ and V without changing E ~ and B,
~ and hence without chang-
ing the force experienced by any particle. He asked this question: I know that in
electrostatics I can add a constant to the potential V (~r) and obtain a potential just
as good as the original. Can I do a similar thing in electrodynamics? This is, if I
~ r, t) and V (~r, t) that produce particular fields E(~
have potentials A(~ ~ r, t) and B(~
~ r, t),
then can I find different potentials

~ 0 (~r, t) = A(~
A ~ r, t) + α
~ (~r, t).

and
V 0 (~r, t) = V (~r, t) + β(~r, t)
that produce the same fields?
~ =∇
If these are to generate the same B ~ × A,
~ then ∇
~ ×A
~0 = ∇
~ × A,
~ so

~ ×α
∇ ~ (~r, t) = 0

which happens whenever

~ r, t).
~ (~r, t) = ∇λ(~
α

Meanwhile, if these two different potentials are to generate the same

~
~ − ∂A
~ = −∇V
E
∂t
they will have to have

~ − ∂~α
−∇β = 0
∂t
~
~ + ∂ ∇λ
∇β = 0
 ∂t
~ β+ ∂λ
∇ = 0
∂t
∂λ(~r, t)
β(~r, t) + = k(t).
∂t
That is, if the new potentials are

~ 0 (~r, t) = A(~
~ r, t) + ∇λ(~
~ r, t) ∂λ(~r, t)
A and V 0 (~r, t) = V (~r, t) − + k(t),
∂t
~ and B
then the E ~ generated will be exactly the same for both potentials. In fact, we
can replace λ(~r, t) with the different function
Z
λ0 (~r, t) = λ(~r, t) − k(t) dt
6.3. GAUGE FREEDOM 59

and make the same transformation.

Thus, finally, for any function λ(~r, t), the potentials

~ 0 (~r, t) = A(~
~ r, t) + ∇λ(~
~ r, t) ∂λ(~r, t)
A and V 0 (~r, t) = V (~r, t) − (6.5)
∂t
generate the same E ~ and B~ fields. Changing from one set of potentials to another is
called a “gauge transformation”. The word “gauge” means “a standard or scale of
measurement”. For example, the choice of whether to measure elevations of building
floors relative to sea level or relative to the elevation of the basement is a choice of
gauge.
[[I am not certain, but I suspect that the term “gauge” follows from the nineteenth
century fascination with railroad technology. In this context “gauge” refers to the
separation between the two rails. Today, most railroads are “standard gauge” (4 feet,
8 12 inches or 1435 mm) or, for railroads in mountainous regions, “narrow gauge”. But
formerly each railroad company used its own gauge. If you search the Internet for
“track gauge” you will uncover information about the 1841–1892 “Gauge War” (also
called the “Battle of the Gauges”).]]
~ r, t)) to
So you see that in fact we don’t need four functions (V (~r, t) and A(~
describe electromagnetism in a particular situation. Because we can select a gauge
at will, there are only three independent functions. This is great power! But with
great power comes great responsibility. How will we use our newfound power?

6.3 Gauge freedom

Weyl gauge
~ r, t)) for our
By hook or by crook, we have found a set of potentials (V (~r, t), A(~
situation. Now set
∂λ(~r, t)
= V (~r, t)
∂t Z
λ(~r, t) = V (~r, t) dt.

The new potentials obtained through the gauge transformation equations (6.5) are
∂λ(~r, t)
V 0 (~r, t) = V (~r, t) − =0
∂t Z
~ 0 (~r, t)
A ~ ~ ~ ~
= A(~r, t) + ∇λ(~r, t) = A(~r, t) + ∇ V (~r, t) dt.

In this gauge — called the Weyl gauge — the scalar potential is zero everywhere and
there are only three component functions to find. Through a similar argument you
could instead set Ax = 0 everywhere, or Ay = 0, or Aθ = 0.
60 CHAPTER 6. POTENTIALS AND GAUGES

The Weyl gauge is amusing when applied to any electrostatic situation, where
~ ~ r) = 0. In the Weyl gauge in this situation the potentials are
E(~r) is given and B(~

V (~r, t) = 0 and ~ r, t) = −E(~

A(~ ~ r)t, (6.6)

as you can readily check through equations (6.2) and (6.1):

~ r, t)
∂ A(~ ~
~ r, t)
E(~ = ~ (~r, t) −
−∇V ~ − ∂(−E(~r)t) = E(~
= −∇0 ~ r)
∂t ∂t
~ r, t)
B(~ = ~ × A(~
∇ ~ r, t) = −t∇~ × E(~
~ r) = 0.

I imagine walking into the office of the president of Eveready Battery Company:
“I purchased a battery advertised as 9 volts. But the voltage is zero everywhere!
Furthermore [pointing to the minus sign in equation (6.6)] the vector potential has
been going down ever since I bought it!”

Coulomb gauge

Well, the Weyl gauge is great for jokes, but can we do something more profitable
with our new power? Recall from equations (6.3) and (6.4) that the equations for
the potentials are

~ 2V + ∂ ∇
∇ ~ ·A
~ = −ρ/0 , (6.7)
∂t
2~
 
∇ ~ − µ0 0 ∂ A − ∇
~ 2A ~ ∇ ~ ·A~ + µ0 0 ∂V ~
= −µ0 J. (6.8)
∂t2 ∂t
When we wrote down the first equation we said it was “a little more involved than
Poisson’s equation”, because of that ∇ ~ ·A~ piece. Is there some gauge in which we
can get rid of it and turn it exactly into Poisson’s equation?
~ r, t))
Suppose that, by hook or by crook, we have found a set of potentials (V (~r, t), A(~
for our situation. Now we look for a new set of potentials for which ∇ ~ ·A ~ 0 = 0.

~ 0 (~r, t) = A(~
A ~ r, t) + ∇λ(~
~ r, t)
~ ·A
∇ ~ 0 (~r, t) = ∇
~ · A(~
~ r, t) + ∇
~ 2 λ(~r, t)

We desire for the left hand side to vanish, and it will do so when
~ 2 λ(~r, t) = −∇
∇ ~ · A(~
~ r, t).

~ · A(~
To find λ(~r, t), we have to solve Poisson’s equation using, as a “source”, −∇ ~ r, t).
This might be hard to do, but we have a theorem guaranteeing that such a solution
exists.
In this new gauge (dropping the primes),
~ · A(~
∇ ~ r, t) = 0
6.3. GAUGE FREEDOM 61

so the scalar potential is generated exactly by Poisson’s equation of electrostatics

~ 2 V (~r, t) = −ρ(~r, t)/0 .

∇

But from electrostatics we know the solution for the potential:

ρ(~r 0 , t) 3 0
Z
1
V (~r, t) = d r.
4π0 r

~r field point

~r

~r 0
origin

On the other hand, the vector potential is particularly difficult to find in this gauge!

One peculiarity of this gauge demands investigation. If a source charge moves at

0
~r , the potential V (~r) at the field point ~r changes. . . instantly! The field point might
be light-years away, but the potential changes the instant that the source charge
moves. How can this be?

When the source charge moves, both the scalar potential and the vector potential
at the field point change instantly, but they change in such a way that the result-
ing electric and magnetic fields at the field point do not change instantly. They
don’t change until the information reaches the field point, traveling at light speed.
(This claim will be proven on page 64.) This underscores the role of potentials as
mathematical tools that help enormously in making calculations but that are not
themselves physically observable. (You know from way back in introductory circuits
that a voltmeter doesn’t measure absolute potential, it can measure only potential
~ r, t) propagate instantly, but all
differences.) In the Coulomb gauge V (~r, t) and A(~
~ r, t) and B(~
gauges produce the same E(~ ~ r, t), and these fields propagate at light speed.

[[This is my favorite analogy for wavefunction in quantum mechanics. When a

measurement is made the wavefunction “collapses” instantaneously, but that’s al-
lowed because wavefunction is not itself physically observable: it is nothing more
than a computational tool.]]

This gauge is often used in quantum electrodynamics, and it’s used far from
charges and currents so that V (~r, t) = 0.
62 CHAPTER 6. POTENTIALS AND GAUGES

It is called the “Coulomb gauge” despite the facts that (i) Charles-Augustin
de Coulomb knew nothing about it (my limited historical research suggests it was
introduced by J. Willard Gibbs) and that (ii) Coulomb’s law
ρ(~r 0 , t) 3 0
Z
~ 1
E(~r, t) = r̂ d r
4π0 r2
doesn’t hold!!

Lorenz gauge

Let’s look again at equations (6.7) and (6.8). Instead of trying to make the first
equation simple at the expense of the second, can we make the second simple?
Yes we can, by selecting a gauge in which
~ ·A
∇ ~ + µ0 0 ∂V = 0.
∂t
(You can find the appropriate λ(~r, t) yourself, if you wish.) In this gauge equa-
tions (6.7) and (6.8) become
2
~ 2 V − µ0 0 ∂ V
∇ = −ρ/0 , (6.9)
∂t2
2~
~ 2A
∇ ~ − µ0 0 ∂ A ~
= −µ0 J. (6.10)
∂t2
This pair has a very pleasing symmetry between V and A. ~ Furthermore (and in
contrast to the general equations (6.3) and (6.4)) the equation for V does not involve
~ and vice versa. In our attempt to make the second equation simple, we have in
A,
fact made both of them simple!
Recall that 2
∇~ 2 f (~r, t) − 1 ∂ f (~r, t) = 0
c2 ∂t2
was called “the wave equation”. Equations like
2
∇~ 2 f (~r, t) − 1 ∂ f (~r, t) = ρ(~r, t)
c2 ∂t2
are called “inhomogenious wave equations” or “wave equations with the source ρ(~r, t)”.
This gauge was invented by the Dane Ludvig V. Lorenz and is called the Lorenz
gauge. Unfortunately Lorenz’s last name is so similar to that of the famed Dutch
theorist Hendrik A. Lorentz that for many years it was misnamed the “Lorentz
gauge”. (This tendency to name things, not after the inventor, but after someone
already famous is called “the Matthew effect” from the Biblical passage Matthew
25:29: “For to every one who has will more be given, and he will have abundance;
but from him who has not, even what he has will be taken away.”) You should not
use this misnomer. I’m telling you because you might look up the gauge in an older
book and find the wrong name.
This gauge is so convenient that for the rest of the course we will use it exclusively.
Chapter 7

Resultant Potentials and

Fields

Statics: You have a bunch of charges ρ(~r 0 ) and currents J(~ ~ r 0 ). What are the
resulting potentials at point ~r ? (Define ~r = ~r − ~r 0 .) You know the answer:

Governing equation Resulting solution

ρ(~r 0 ) 3 0
Z
~ 2 V (~r) = −ρ(~r)/0 1
∇ V (~r) = d r
4π0Z r
~ r 0)
~ 2 A(~
∇ ~ r) = −µ0 J(~
~ r) ~ r) = µ0
A(~
J(~
d3 r 0
4π r

~ r 0 , t) are chang-
Dynamics: What if these source charges ρ(~r 0 , t) and currents J(~
ing? A good guess is that the potentials are retarded. . . that is the effect of a charge
or current propagates away from that source at the speed of light c. Thus the poten-
tial at point ~r at time t will depend not upon the sources at ~r 0 now, but upon the
sources that were there some time ago, at the “retarded” time
r
tr = t − .
c
If that guess holds, then the answer would be:

Governing equation Resulting solution

2
ρ(~r 0 , tr ) 3 0
Z
~ 2 V (~r, t) − 1 ∂ V (~r, t) = −ρ(~r, t)/0
∇ V (~r, t) =
1
d r
c2 ∂2t 4π0Z r
2~ ~ r , tr )
0
~ 2 A(~
~ r, t) − 1 ∂ A(~r, t) ~ r, t) ~ r, t) = µ0 J(~
∇ = −µ0 J(~ A(~ d3 r 0
c2 2
∂ t 4π r

Does our guess hold true? Griffiths proves (pages 445–446) that it does.

63
64 CHAPTER 7. RESULTANT POTENTIALS AND FIELDS

There are two equivalent ways to look at these retarded solutions. The first
(“source centric”) way is that the sources make V and A ~ and those effects propagate
0
outward from the sources at point ~r at light speed. The second (“field point centric”)
way is to start at the field point ~r. To calculate the field there, think of a spherical
shell centered on ~r, expanding at light speed but going backwards in time. When
this shell touches a source, that’s the source affecting the field point at time t. The
first approach focuses on a spherical shell expanding from each source point going
forward in time, the second focuses on a spherical shell expanding from each field
point going backward in time.
You might think this is an obvious generalization, but (i) it holds only for poten-
tials in the Lorenz gauge and (ii) it doesn’t hold for fields. The fields are not simply
retarded. In fact, to find the fields you need to use equations (6.1) and (6.2), namely
~
~ r, t) = −∇V
E(~ ~ (~r, t) − ∂ A(~r, t) and B(~ ~ r, t) = ∇ ~ × A(~ ~ r, t).
∂t
This is more difficult to do than you might think (Griffiths pages 449–450), but the
answers are
Z " ~˙ r 0 , tr )
#
~ 1 ρ(~r 0 , tr ) ρ̇(~r 0 , tr ) J(~
E(~r, t) = r̂ + r̂ − d3 r0 (7.1)
4π0 r2 cr c2 r
Z "~ 0 ~˙ r 0 , tr )
#
~ r, t) = µ 0 J(~
r , t r ) J(~
B(~ + × r̂ d3 r0 . (7.2)
4π r2 cr
These are the Jefimenko equations (after Oleg Dmitrovich Jefimenko of West Virginia
University in Morgantown).
It is clear that these fields are “retarded” — that is the electromagnetic effects of
a source propagate away from that source at light speed: no faster, no slower. These
same fields would be derived from any gauge, but it’s easiest to find them in the
Lorenz gauge. This is the source of my claim on page 61 that in the Coulomb gauge
the potentials propagate instantaneously, but the fields do not.
These equations exhibit surprising, perhaps shocking, behavior. From way back
at the Coulomb and Biot-Savart laws, we’ve grown used to the idea that electric
and magnetic fields fall off like 1/(distance)2 . And sure enough two terms in the
˙
Jefimenko equations do just that. If the situation is static (ρ̇ = 0, J~ = 0) these are
the only terms that exist. But if the situation is dynamic, then there are three terms
˙
that fall off much more slowly, like 1/(distance). These terms resulting from ρ̇ and J~
are the EM fields involved in light and radio propagation. Because they fall off like
1/(distance), not 1/(distance)2 , they are much more effective in spreading over large
distances. This is why we use electrodynamics, not electrostatics, to send messages
over long distances (e.g., via radio).
Scientists who work in detecting gravity waves (LIGO, VIRGO, NANOGRAV,
LISA) like to boast about gravity wave observatories: “Light falls off like 1/r2 , but
7.1. SINGLE MOVING POINT CHARGE 65

gravitational field from gravity waves falls off like 1/r, so we can see farther into
space than optical observatories can.” Well in fact, the electric field due to light
radiation and the gravitational field due to gravitational radiation both fall off like
1/r. The difference is that optical observatories measure the energy carried by light,
whereas gravity wave observatories measure the field carried by the gravity wave. If
we detected the E ~ field associated with light, rather than the energy associated with
light, we would find that it falls off like 1/r too.

7.1 Single moving point charge

A single point particle with charge q travels along a trajectory. What potentials and
fields does it produce?

~r field point

~r

~r 0
origin

The open red dot represents the location of the charged particle now. More important
is the solid red dot which represents the location of the charged particle back at the
retarded time,
r
tr = t − .
c
The retarded potential is
ρ(~r 0 , tr ) 3 0
Z
1
V (~r, t) = d r
4π0 r
and I’d think that this would evaluate to
1 q
.
4π0 r
I’m wrong, for a subtle reason involving the spherical shell expanding as it moves
backward in time.
66 CHAPTER 7. RESULTANT POTENTIALS AND FIELDS

To make things easier we consider, first treat the apparent volume of a truck
cargo box, not of a spherical shell expanding backwards in time.

L'
L

where truck used to be

Jessica stands on the sidewalk as a truck travels down the road at speed v. The light
from the front of the cargo box takes some time to reach Jessica’s eye, but the light
from the back takes still more time. Thus she sees the cargo box as longer than it
really is. How much longer? If L is the true length of the cargo box and L0 the
apparent length, then the truck moves a distance L0 − L while the light moves a
distance L:
L0 − L L0
=
v c
L
L0 =
1 − v/c
L
L0 =
1 − r̂ · ~v /c
true volume of cargo box
apparent volume of cargo box =
1 − r̂ · ~v /c
d3 r
d3 r 0 =
1 − r̂ · ~v /c
Thus
ρ(~r 0 , tr )
Z
1 1 q 1
V (~r, t) = d3 r0 = (7.3)
4π0 r 4π0 r 1 − r̂ · ~v /c
where all the quantities on the right are evaluated at the retarded time tr . Similarly
for the vector potential
~ r, t) = µ0 q~v
A(~
1
(7.4)
4π r 1 − r̂ · ~v /c
where again all the quantities on the right are evaluated at the retarded time. These
two equations are the Liénard–Wiechert potentials.

[[The French mining engineer Alfred-Marie Liénard derived these potentials in

1898. The German geophysicist Emil Wiechert derived them independently in 1900.
Liénard was not related to Philipp Lenard, who won the Nobel Prize for his work with
electron beams and who was appointed “Chief of Aryan Physics” by Adolf Hitler.]]
7.1. SINGLE MOVING POINT CHARGE 67

~u
~v field point

cr̂

origin

The fields associated with these potentials are, where ~u = cr̂ − ~v ,

~ r, t) q r
E(~ = [(c2 − v 2 )~u + ~r × (~u × ~a)] (7.5)
4π0 (~r · ~u)3
~ r, t) 1 ~ r, t)
B(~ = ~r × E(~ (7.6)
c
all quantities on the right of the first equation being evaluated at the retarded time.
(I noted back on page 6 that the familiar Coulomb and Biot-Savart laws do not hold
in electrodynamics. What holds instead? These two equations.)

The challenging thing about these two formulas is not in their derivation — which
is straightforward if laborious — but in making sense of the result. What do these
formulas tell us about nature?

First, make sure that they give the proper electrostatic result. If ~a = 0 and ~v = 0,
then ~u = cr̂ so

~ r, t) q r q 1
E(~ = [(c2 )cr̂ ] = r̂
4π0 (~r · (cr̂ )) 3 4π0 r 2
~ r, t)
B(~ = 0

So far so good.
~ (za) (~r, t) that
I like to break equation (7.5) down into two parts: the electric field E
would be at point ~r if there were zero acceleration, and the electric field E ~ (a) (~r, t)
due to acceleration:

~ r, t)
E(~ = E~ (za) (~r, t) + E
~ (a) (~r, t) (7.7)
~ (za) (~r, t) = q r
E (c2 − v 2 )~u (7.8)
4π0 (~r · ~u)3
~ (za) × ~a)
~ (a) (~r, t) = ~r × (E
E . (7.9)
(c − v 2 )
2
68 CHAPTER 7. RESULTANT POTENTIALS AND FIELDS

Griffiths investigates the constant velocity result E~ (za) in his example 10.4. He
~ is the vector from the location where the charge is now (not its retarded
finds that if R
location) to the field point, then

~ (za) = q 1 − (v/c)2 R̂
E . (7.10)
4π0 [1 − (v sin θ/c)2 ]3/2 R2

~v field point
source now, in absence θ
of acceleration ~
R

~v
source at retarded time

What does this equation tell us? One surprising feature is that the field is radial
from the source’s current location, not from its retarded location. Once you’re over
this surprise, the equation looks remarkably like Coulomb’s law. There is an obvious
and necessary cylindrical symmetry. There is a less obvious front-back reflection
symmetry: Because sin(θ) = sin(180◦ − θ), the field of a particle with velocity −~v is
exactly the same as that of a particle with velocity +~v .

The field immediately in front of the particle (θ = 0) is weaker than the Coulomb
field by a factor of 1 − (v/c)2 . As θ increases the field becomes stronger until, at
p
θ = 90◦ , the field is stronger than the Coulomb field by a factor of 1/ 1 − (v/c)2 .
As θ increases still further, the field declines until, immediately behind the particle,
it is equal to the magnitude immediately in front.
7.1. SINGLE MOVING POINT CHARGE 69

All this information is summarized in a field line diagram that I call “cat whiskers”:

I think of it this way: Gauss’s Law says that the number of electric field lines launched
by a charge q is always the same, regardless of velocity. When the velocity is zero,
the field lines necessarily spread out uniformly. When the velocity increases, they
drift away from the bow and wake of the particle toward “midships” (left and right
of the velocity).
70 CHAPTER 7. RESULTANT POTENTIALS AND FIELDS

So now we have a global picture for E ~ (za) . Can we use this, accompanied by
equation (7.9), to obtain a global picture for E ~ (a) ? In general no. The problem is
that at each field point we would need a different set of cat whiskers. For example in
the figure below, if we wanted to find the field at point 1, far from the source charge
trajectory, we would need to use the cat whiskers from some time in the distant past.
But at point 2, close to the source charge trajectory, we would need to use the cat
whiskers from some time in the recent past. Every point needs to employ a different
set of cat whiskers, and the task quickly degenerates into a morass for all but the
simplest situations.

field point 2

field point 1

Before giving up completely, I do want to look at one qualitative feature. The

cat whiskers equation (7.10) shows the electric field due to a constant velocity
charge falling off like 1/(distance)2 . But the equation for E ~ (a) from an accelerat-
ing charge (7.9) multiples this E ~ (za) by a distance ~r in the numerator. Admittedly
the distance R in the denominator is not exactly the same as the distance r in the
2

numerator, but this at least reinforces our arguments on page 64 that the electric
field due to acceleration falls off like 1/(distance).

7.2 Fields from a good swift kick

The task of finding the electric field everywhere “quickly degenerates into a morass
for all but the simplest situations.” All right then, let’s not give up, let’s look at a
simple situation.
7.2. FIELDS FROM A GOOD SWIFT KICK 71

A charge is stationary until given a good swift kick that accelerates it instantly
to velocity ~v . Then it maintains that velocity without change for a time T . All
points outside of sphere of radius cT haven’t yet gotten the message that the charge
has gotten the kick. The electric field outside that sphere is just the Coulomb field
pointing at where the charge had been before the kick.

cT
72 CHAPTER 7. RESULTANT POTENTIALS AND FIELDS

What about points inside the sphere? It’s resonable to suppose that the field
lines inside are cat whiskers.
7.2. FIELDS FROM A GOOD SWIFT KICK 73

How do the field lines inside match up with those outside? They can only connect
through field lines on the surface of the expanding sphere.
74 CHAPTER 7. RESULTANT POTENTIALS AND FIELDS

Can we make these observations quantitative? Look at equation (7.9),

~ (za) × ~a)
~ (a) (~r, t) = ~r × (E
E .
c − v2
2

~ (za) to the field point, and ~r the

Here ~a and v refer to the retarded source point, E
vector from the retarded source point to the field point.

On the outer edge of the expanding shell, ~r = ~r, v = 0, and

~ (za) = q r̂
E .
4π0 r2
Thus
~ (a) = q 1
E ~r × (r̂ × ~a).
4π0 c r2
2

The direction is tangential to the expanding shell,

~ (a)
E
~a θ ~r

and the magnitude is

~ (a) | = q a sin θ
|E .
4π0 c2 r
Everything about this formula makes sense. The prefactor q/4π0 is expected for an
electric field. It has the 1/r character that we’ve been harping on for radiation fields.
It is directly proportional to a sin θ, which is the amount of acceleration perpendicular
to ~r. (If you peer down −~r from the field point toward the accelerating charge, you
won’t see the part of ~a parallel to ~r. . . you only see the perpendicular part, with
magnitude a sin θ). And if the equation with all those characteristics is to have the
proper dimensions, there must be a constant factor with dimensions 1/(speed)2 .

This electric field is perpendicular to the direction of propagation. It’s light!

7.3. RADIATION SIMULATIONS 75

7.3 Radiation simulations

I have solved enough electrostatics problems that I have a “gut” feel for them and
I can usually guess the qualitative character of electrostatic field before I solve the
problem mathematically. I’m usually but not always right. When I’m not right I
have fun figuring out why my gut feel is wrong, and that figuring helps refine my
feel.
But I have a harder time getting a gut feel for radiation problems. My best
approach is to look at an arbitrary trajectory as a sum of good swift kicks, but I
really need a computer simulation to get a good qualitative picture. I recommend
you spend time looking at these two simulations:

1. The PhET simulation “Radiating Charge” at

https://phet.colorado.edu/en/simulation/legacy/radiating-charge.

(Try giving a stationary charge a good swift kick to check out the conclusions
of the previous section.)

2. Wolfgang Christian’s more limited simulation “Accelerated Charge Radiation”

available through ComPADRE at

https://www.compadre.org/osp/EJSS/4126/154.htm.

3. Bala Juluir’s simulation of dipole radiation at

https://balajuluri.com/radiation-from-dipole.html.

(Note that the electric field lines always loop! We are a long distance from
electrostatics.)

What do I take away from these simulations? That electric field lines act like
furry rubber bands.

7.4 Radiation from an electric dipole

Why would anyone be interested in the artificial situation of a pure dipole oscillating
with a single frequency? Surely any charge configuration encountered in nature will
be more complicated than two equal point charges of opposite sign! The reason is
that for many charge configurations the dominant portion of the radiated EM field
is dipolar. (Furthermore, once we understand the math of finding the radiation from
a dipole, it’s (relatively) easy to go back and redo the calculation for quadrupoles,
octupoles, etc.)
76 CHAPTER 7. RESULTANT POTENTIALS AND FIELDS

An oscillating electric dipole points always along the same axis and has dipole
moment p0 cos(ωt). Call the axis z and locate the coordinate origin at the center of
the dipole.

At equation (11.20), Griffiths finds1 that at detection point ~r the current of elec-
tromagnetic energy (power radiated/cross-sectional area) produced by this dipole
is 2 4
~ r, t) = µ0 p0 ω sin2 θ cos2 [ω(t − r/c)]r̂.
S(~
16π 2 cr2
I don’t want to derive this result; instead I want to dissect it to uncover what it tells
us about nature.

The radiation is outbound (in the direction r̂) and has frequency ω. As time
proceeds, this energy current varies from 0 to some maximum and then back to zero
and so forth. This is expected behavior and (for almost all frequencies) it will be
hard to follow those rapid variations. Instead, focus on the time-averaged2 energy
current of
µ0 p20 ω 4
hSi = sin2 θ.
32π 2 cr2
(1) Dependence on distance r. It makes perfect sense that the power should
decrease like 1/r2 : For radiated fields the electric field behaves like 1/r, the magnetic
field behaves like 1/r, so the Poynting vector S ~ = (1/µ0 )E ~ ×B ~ should behave like
1/r2 .

Furthermore, the 1/r2 behavior is expected from energy conservation. Imagine

two spherical shells surrounding the dipole. In a steady state the total power passing
1 After
making three reasonable approximations saying that the detection point is far from a
small dipole.
2 The time average of cos(ωt + φ) over a period is zero. The time average of cos2 (ωt + φ) over

a period is 21 . I know I always say (quoting from the syllabus) “In writing your solutions, do not
just write down the final answer. Show your reasoning and your intermediate steps. Describe (in
words) the thought that went into your work as well as describing (in equations) the mathematical
manipulations involved.” But by this point in your education you are suppose to know these facts
without having to derive them.
7.4. RADIATION FROM AN ELECTRIC DIPOLE 77

through the inner shell must equal the total power passing through the outer shell.
Thus the power per area must obey
total power 1
S= ∼ 2.
4πr2 r

(2) Dependence on angle θ.

If you view the dipole from the right, you see a tall dipole of height p0 . But if you
view it from above, you look down and see but a single point! If you view it looking
down the axis at angle θ, you see an intermediate dipole of height p0 sin θ. So it
makes sense that the electric field would vary as the dipole height, E (a) ∼ p0 sin θ,
and that the power would vary as S ∼ p20 sin2 θ.

(3) Dependence on frequency ω. It certainly makes sense that higher frequen-

cies, hence higher accelerations, would radiate more power. But can we understand
why the power goes up like ω 4 rather than some other power?

We can. Simple harmonic motion has

x(t) = x0 cos(ωt)
v(t) = −ωx0 sin(ωt)
a(t) = −ω 2 x0 cos(ωt)

But we know (from equation 7.9) that the radiated electric field is proportional to
acceleration a. Hence the power, proportional to electric field squared, is proportional
to ω 4 .

This is a big difference. Even within the narrow band of visible light, in going
from violet light (λ = 380 nm) to red light (λ = 700 nm), ω decreases by a factor of
380/700 = 0.54 so ω 4 decreases by a factor of 0.087.

Here is one consequence: On the Moon, the sky is dark. But on Earth, the
sky is bright because the atmosphere scatters sunlight. Sunlight streams through the
atmosphere. The oscillating electric field in that sunlight induces oscillating dipoles
in the nitrogen, oxygen, and other molecules in the atmosphere. (In the absence of
78 CHAPTER 7. RESULTANT POTENTIALS AND FIELDS

light, the nitrogen molecule has zero dipole moment.) Because the sunlight oscillates
at frequencies ranging from fast (violet) to slow (red), the molecular dipoles oscillate
in that same frequency range. The oscillating dipoles themselves radiate, explaining
why the sky on Earth is bright. We’ve just seen that the fast oscillators (short
wavelength, violet) radiate far more than the slow oscillators (long wavelength, red)
so you might expect the sky to be violet. However the sun doesn’t radiate strongly
in the violet (380–450 nm)

and the human eye is not particularly sensitive in the violet

so in fact the sky is blue. Because the blue is removed from sunlight through this
scattering process, the light that remains after passing through a lot of atmosphere,
as during sunrise and sunset, is red.
7.4. RADIATION FROM AN ELECTRIC DIPOLE 79

Challenge: Make reasonable approximations to show that when the sun is directly
overhead, sunlight passes through 10 miles of air to reach our eyes, whereas when
the sun is setting, it passes through 300 miles of air.

This is one of the things I love about physics. We’ll be pushing our way through
a thorny patch of dense mathematics, tangled and rough, then suddenly emerge into
an open grassy meadow to understand something about the everyday world that has
puzzled us since childhood.

But another thing I love about physics is that it not only explains what we already
know, it also brings out things we didn’t know.
80 CHAPTER 7. RESULTANT POTENTIALS AND FIELDS

The light from the sun is randomly polarized, switching back and forth rapidly
from x polarized to y polarized to 17◦ polarized (which can be regarded as a super-
position of x and of y polarized). [Such light is often called “unpolarized”, but it
really deserves the name “randomly polarized”.] At an instant when the light from
the Sun is polarized in-and-out of the page (top of figure), it makes the dipole of a
nitrogen molecule oscillate in-and-out of the page. In turn, that oscillating dipole
radiates in the direction of the person shown.

electric field
SUN
oscillating dipole

strong radiation in this

direction

electric field
SUN
oscillating dipole

no radiation in this
direction

At an instant when the light from the Sun is polarized within the page (bottom of
figure), the dipole of the nitrogen molecule oscillates within the page. That oscillating
dipole radiates, but, as seen above at “(2) Dependence on angle θ”, it sends none of
that resulting radiation in the direction of the person shown.

Look at the blue sky through a Polaroid. (For safety’s sake, don’t look directly at
the sun.) You will find that the sky is unpolarized near the sun, but highly polarized
90◦ from the sun. At sunset on a clear evening, this 90o belt is noticeable directly
above your head if you look up while wearing Polaroid sunglasses.

Because the clear blue sky is polarized but the white clouds are unpolarized,
landscape photographers often use polarizing filters to make white clouds stand out
beautifully in a deep blue sky.
7.4. RADIATION FROM AN ELECTRIC DIPOLE 81

There’s a lot more to say about waves and optics in the everyday world.
On 19 July 2004, I was goatpacking with my brother on Grandfather Mountain in
Saint Joe National Forest, Idaho. We had a beautiful hike up to a meadow. We were
beginning to set up camp at the crest of the meadow when a thunderstorm rolled in.
We scurried to finish setting up the tent. Once the tent was erect, one of the goats
slipped inside, bringing a lot of water and mud with it. By the time we cleaned up
from that goatscapade the cloudburst was almost over. A stunningly beautiful rain-
bow hung over the meadow to the east, and over the woods and mountains beyond.
I grabbed my brother’s camera to take a photo of him standing in the meadow with
rainbow and mountains in the background. I peered through the camera’s viewfinder.
My brother was there, and the meadow and the mountains, but no rainbow! I looked
over the camera instead of through the viewfinder. The rainbow was there! I looked
back through the viewfinder. No rainbow! I remembered that light from a rainbow
is polarized. “Bill, do you have a polarizing filter on this camera?” “I don’t know,”
he replied, “I just put on whatever the salesman recommended.” So I twisted the
camera by 90◦ , and there was the rainbow in the viewfinder. I snapped the photo.

To give you a glimpse of more topics involved, I show one photo that I took while
backpacking on the Na Pali coast of Kauai (part of my effort to go backpacking in
each of the fifty states)
82 CHAPTER 7. RESULTANT POTENTIALS AND FIELDS

and a photo of so-called “supernumerary rainbows” (which I have never personally

witnessed)

If you like these sorts of questions, I recommend these three books: Waves by
Frank S. Crawford, Jr. (1968); Light and Colour in the Open Air by M.G.J. Minnaert
(1940) [also published under titles The Nature of Light and Colour in the Open Air
and Light and Color in the Outdoors]; and Rainbows, Halos, and Glories by Robert
Greenler (1980). (Bob Greenler’s daughter Robin attended Oberlin College, and
Bob is the only person to have served twice as the OC Physics Department’s Hays
lecturer.)
Chapter 8

Relativistic Electrodynamics

8.1 Space and time in relativity

You know the setup: Reference frame F̄ moves uniformly at speed V to the right past
reference frame F. The two reference frames coincide when t = t̄ = 0. Some event
occurs.

y ȳ event

F F̄
x x̄

z z̄

How are the event coordinates (t, x, y, z) in frame F related to the event coordi-
nates (t̄, x̄, ȳ, z̄) in frame F̄? The answer is

t − V x/c2
t̄ = p
1 − (V /c)2
x−Vt
x̄ = p (8.1)
1 − (V /c)2

ȳ = y
z̄ = z

This is called the “Lorentz transformation”.

83
84 CHAPTER 8. RELATIVISTIC ELECTRODYNAMICS

You know that there are some extremely non-obvious consequences of these four
simple equations: time dilation, length contraction, relativity of simultaneity, the
transformation of velocity
vx − V
v̄x = , (8.2)
1 − vx V /c2
the fact that interval
(ct)2 − (x2 + y 2 + z 2 ) (8.3)

is “invariant” — the same in all reference frames, no causal signal can travel faster
than light, no body is infinitely rigid, no body is infinitely strong, the twin para-
dox, the busted bus, the pole in the barn, etc., etc.1 I am willing to discuss these
consequences with you for as long as you are willing to ask questions.

But eventually we will stop taking about consequences and return to the Lorentz
transformation equations themselves. It’s pretty clear that these equations become
more symmetric and easier to remember if you focus on the product ct rather than
on t. The variable ct has the dimensions of length, putting it on an even footing with
x, y, and z. In addition, we will use the common abbreviations
V 1
β= and γ=p , (8.4)
c 1 − (V /c)2

which save a lot of penstrokes. [[Use these abbreviations cautiously, however. I

remember solving a long and intricate relativity problem and being puzzled by the
fact that my solution differed from the solution in the back of the book by a factor
of γ 2 (1 − β 2 ). Try as I might, I just couldn’t find my error.]]

With these cosmetic changes, the Lorentz transformation becomes

ct̄ = γ(ct − βx)

x̄ = γ(x − βct)
(8.5)
ȳ = y
z̄ = z

And this form suggests one last cosmetic change. Call the space-time coordinates of
the event
(x0 , x1 , x2 , x3 ) ≡ (ct, x, y, z) (8.6)

and write
x̄0 = γ(x0 − βx1 )
x̄1 = γ(x1 − βx0 )
(8.7)
x̄2 = x2
x̄3 = x3
1 The
appendix to G.P. Sastry, “Is length contraction really paradoxical?” American Journal
of Physics 55 (October 1987) 943–946 presents an extensive yet necessarily incomplete catalog of
paradoxes in relativity.
8.1. SPACE AND TIME IN RELATIVITY 85

Challenge: Because of the many apparent paradoxes of relativity, it’s not clear
that these equations are consistent. Prove to yourself: (i) If you transform coordi-
nates from frame F to frame F̄ moving at velocity V relative to F, and then from
frame F̄ to frame F moving at velocity −V relative to F̄, you get back to your original
coordinates. (ii) If you transform coordinates from frame F0 to frame F1 moving
at speed V1 relative to F0 , and then transform those coordinates from frame F1 to
frame F2 moving at speed V2 relative to F1 , you get the same result as transforming
directly from frame F0 to frame F2 , moving at speed
V1 + V2
1 + V1 V2 /c2
relative to F0 . These two items are sufficient to prove that the set of Lorentz trans-
formations constitute a mathematical group, and then group theory assures us that
the equations are consistent.

Four-scalars and four-vectors

A four-scalar is the same in all reference frames. Examples are the mass of a particle,
the charge of a particle, and the invariant interval between two events. The time
separation between two events is not a four-scalar, but the time ticked off by a watch
attached to a particle — the so-called “proper time”2 τ — is a four-scalar: different
reference frames disagree on the time elapsed, but all agree that the watch attached
to a particle has ticked off a certain amount of time. (The proper time squared is just
the invariant interval between two events, so of course it is a four-scalar.) According
to time dilation, the relation between lab time and proper time is
dτ
dt = p . (8.8)
1 − (v/c)2

The coordinates of a four-vector transform from frame to frame like

ā0 = γ(a0 − βa1 )

ā1 = γ(a1 − βa0 )
ā2 = a2
ā3 = a3

Or in other words
3
X
āµ = Λµα aα ,
α=0
2 The word “proper” is irritating. Any inertial frame is as good as any other inertial frame, so
why should the time in one of those frames be considered more “proper” than any other time? The
word origin is that the “proper” in “proper time” derives not from the English “proper” meaning
“respectable, genteel”, but from the French “propre” meaning “own”. The particle’s “proper time”
means the particle’s “own time”.
86 CHAPTER 8. RELATIVISTIC ELECTRODYNAMICS

where the matrix Λ with components Λµα (µth row; αth column) is
 
γ −γβ 0 0
 −γβ γ 0 0 
.
 

 0 0 1 0 
0 0 0 1

The most familiar four-vector is the location of an event in space-time r which is

represented in a particular reference frame through

r =. [ct, x, y, z] = [ct, ~r]. (8.9)

(I like to represent four-vectors using open face symbols. It would be nice if this
habit were universal, but it isn’t.)

8.2 Energy and momentum in relativity

Another four-vector is the four-momentum

dr
p = m dτ . (8.10)

In some particular inertial frame

dr dt dr .
p = m dτ =m
dτ dt
=p
m
1 − (v/c)2
[c, vx , vy , vz ] = [E/c, px , py , pz ]. (8.11)

8.3 Current density in relativity

A charged wind moves through the laboratory. A particular dot of charge has velocity
~v . Transform into the reference frame moving along with that dot of charge. (In
general, this will involve a rotation as well as a Lorentz boost — a so-called Poincaré
transformation.) In this so-called comoving (or proper) reference frame erect a tiny
box of volume V0 about the dot. The proper charge density is
Q
ρ0 = (8.12)
V0
and of course in this reference frame the current vanishes.

In the laboratory reference frame the tiny box has contracted volume
p
V = 1 − (v/c)2 V0 ,

so the lab charge density is

ρ0
ρ= p
1 − (v/c)2
8.4. FOUR-TENSORS 87

while the lab current density is

ρ0~v
J~ = ρ~v = p .
1 − (v/c)2

Comparison to the four-momentum

" #
p
.
= [E/c, p~] = p
mc
1 − (v/c)2
,p
m~v
1 − (v/c)2

leaves no doubt that the four-current

" #
J . ~ = p
= [ρc, J]
ρ0 c
1 − (v/c)2
,p
ρ0~v
1 − (v/c)2
(8.13)

transforms as a four-vector.

The continuity equation

~ · J~ = − ∂ρ
∇
∂t
becomes in four-vector notation
3
X ∂J µ
= 0. (8.14)
µ=0
∂xµ

8.4 Four-tensors

Just as we generalized ordinary vectors to tensors (page 36), so we can generalize

four-vectors to four-tensors. A four-tensor has 16 components
 00 01 02 03 
t t t t
 t10 t11 t12 t13 
 
 20 21 22 23 
 t t t t 
t30 t31 t32 t33

that transform from one reference frame to another through

3 X
X 3
t̄ µν = Λµα Λνβ tαβ . (8.15)
α=0 β=0

Normally statements about tensor components, as opposed to statements about

tensors, are frame-dependent. For example if t13 = 7, then in general t̄ 13 6= 7. But
not always. For example the tensor with components δ µν has the same components
in all frames. Another example is that if the tensor components are symmetric
(tµν = tνµ for all µ, ν) in one reference frame then they are symmetric in all reference
frames. Similarly if they are antisymmetric (tµν = −tνµ for all µ, ν).
88 CHAPTER 8. RELATIVISTIC ELECTRODYNAMICS

The proof starts with equation (8.15). Write it down again, except swap the
indices µ and ν:
X3 X 3
νµ
t̄ = Λνα Λµβ tαβ . (8.16)
α=0 β=0

Now, write this second equation again, but swapping the dummy summation indices
α and β:
3 X
X 3
t̄ νµ = Λνβ Λµα tβα . (8.17)
β=0 α=0

Finally, make use of the fact that addition and multiplication of real numbers is
commutative:
3 X
X 3
t̄ νµ = Λµα Λνβ tβα . (8.18)
α=0 β=0

Finally write equations (8.15) and (8.18) right on top of each other:
3 X
X 3
t̄ µν = Λµα Λνβ tαβ , (8.19)
α=0 β=0
3 X
X 3
t̄ νµ = Λµα Λνβ tβα . (8.20)
α=0 β=0

It’s now plain to see that if tαβ = tβα for all α, β on the right, then it’s also true
that t̄ µν = t̄ νµ for all µ, ν on the left. Furthermore, if tαβ = −tβα for all α, β then
t̄ µν = −t̄ νµ for all µ, ν.

Challenge: The proof also tells us that if tµν = 5tνµ for all µ, ν in one reference
frame, that this property holds in all reference frames. Does this mean that we should
talk about a class of “five-fold tensors”?

A general four-tensor has 16 independent components. But a symmetric four-

tensor has 10 independent components. And an antisymmetric four-tensor has four
zero components on the diagonal so it has only 6 independent components.

Two things about this theorem: (i) It works for ordinary tensors in three-dimensions
as well as for four-tensors. (ii) It’s telling us something about geometry. It’s telling
us that the property of symmetry or antisymmetry doesn’t depend merely on coor-
dinates, but on something coordinate independent. I wish I understood what this
something was; I wish I could give you a picture of what a symmetric four-tensor
“looks like”. For that matter I wish I could give you a picture of what a symmetric
three-tensor “looks like”. Or what a four-vector “looks like”. Alas, I have spend
much of my life trying to develop a pictorial sense of tensors and of spacetime, yet I
cannot give you any of these things.
8.5. MAGNETISM AS A RELATIVISTIC PHENOMENON 89

8.5 Magnetism as a relativistic phenomenon

Griffiths section 12.3.1 shows that magnetism can be regarded as a relativistic phe-
nomenon due to electric charges in motion (currents).

You will immediately object: But relativity is important only for objects
moving near light speed! Magnetic effects are produced by currents moving at very
modest speeds.

Indeed. But the electrostatic force is so huge (on a human scale) that even these
tiny relativistic effects are noticeable (on a human scale).

[[Are electrostatic forces huge on a human scale? Whenever I teach Physics 111,
I assign this problem on the first week:

Two students, Ivan and Veronica, stand about 100 feet apart. Ivan weighs
about 200 pounds and Veronica weighs about 100 pounds. Suppose each
student has a 0.01% excess in his or her amount of positive and negative
charge, one student being positive and the other negative. Estimate the
force of attraction between the two.

In my model solution I estimate the charge on the protons in 100 pounds of water
(and people are mostly water) as about 25 × 108 C. Thus

Force at 100 ft ≈ 30 m is
(25 × 104 C)(50 × 104 C)
(9 × 109 Nm2 /C2 ) ≈ 1.2 × 1018 N.
(30 m)2
Wow! This is, of course, a huge force. If Ivan and Veronica really were
attracted to each other with a force like this, they would be pulled toward
each other and both would die in the resulting crash. The moral of the
story is that Ivan and Veronica do not suffer charge excesses of 0.01% . . .
real life charge excesses are much smaller than this.]]
90 CHAPTER 8. RELATIVISTIC ELECTRODYNAMICS

“Derivation” of Maxwell Equations from Coulomb’s law plus relativity.

It was once popular3 to say

Coulomb’s Law of Electrostatics + Relativity =⇒ Maxwell Equations

That’s not exactly true. If it were true, then

Newtons’s Law of Gravity + Relativity =⇒ Maxwell Equations

would hold. And it doesn’t. Such discussions shine light on neglected corners of
electrodynamics, but in the end they are suggestive. They don’t “derive” anything.

Ben Lemberger (in 2014) pointed out that Coulomb’s Law is exactly true in
classical electrostatics (as far as we know). But Newton’s Law of Gravity is not
exactly true — it doesn’t hold in high field situations, and field becomes high as
r → 0.

Total charge. Katie Rigdon (in 2018) asked: Why is it that

total charge = sum of constituent charges

but
total mass 6= sum of constituent masses ?

It is because the source of EM field is charge, a four-scalar, whereas the source of

gravitational field is not mass, a four-scalar, but instead the stress-energy four-tensor,
which involves mass, and energy, and momentum, and the flow of mass, energy, and
momentum. (This also explains why massless photons fall in gravity.)

3 JackR. Tessman, “Maxwell — Out of Newton, Coulomb, and Einstein” American Journal
of Physics 34, 1048–1055 (1966). W.G.V. Rosser, Classical electromagnetism via relativity: An
alternative approach to Maxwell’s equations (Butterworth, 1968). A. Nicolaide, “Derivation of the
electromagnetic field equations by applying Coulomb’s formula and the special theory of relativity”
Archiv für Elektrotechnik 56, 156–160 (August 1974). B. Konorski, “Coulomb’s law and the theory
of relativity — a basis of the entire electrodynamics (Horak’s method)” Przeglad elektrotechniczny
53, 428–432 (1 October 1977) (in Polish).
8.6. ELECTROMAGNETIC FIELDS IN RELATIVITY 91

8.6 Electromagnetic fields in relativity

Griffiths equation (12.109) shows how fields transform from one reference frame to
another:

Ēx = Ex  y − V Bz ) 
Ēy = γ (E Ēz = γ (E
 z + V By ) 
V V
B̄x = Bx B̄y = γ By + 2 Ez B̄z = γ Bz − 2 Ey
c c

~ but on E/c
I like to write these with an emphasis not on E, ~ (which has the same
~ giving
dimensions as B),

Ēx /c = Ex /c Ēy /c = γ (Ey /c − βBz ) Ēz /c = γ (Ez /c + βBy )

B̄x = Bx B̄y = γ (By + βEz /c) B̄z = γ (Bz − βEy /c)

r p
In the cases of four-vectors for spacetime and energy-momentum and current
J
density , we started with a three-vector (~r and p~ and J) ~ and tacked on front a
three-scalar (ct and E/c and ρc) to make a four-vector. This strategy will not work
for E~ and B.
~ Just look at the transformation properties for Ex and Bx !
~ and B
Instead, the trick is to look at E ~ together. There are six components. And
an antisymmetric four-tensor also has six components! Is there any way to shoehorn
the six components of E ~ and B~ into an antisymmetric four-tensor?

Yes! You have to do some fiddling around, but with sufficient fiddling you find
not one but two different ways to perform this shoehorning! The first is the “elec-
F
tromagnetic field four-tensor” , with components
 
0 Ex /c Ey /c Ez /c
 0 Bz −By 
. (8.21)
 

 0 Bx 
0

(I leave blank the elements below the diagonal because they’re just the negatives
G
of the elements above the diagonal.) The second is the “dual four-tensor” , with
components  
0 Bx By Bz
 0 −Ez /c Ey /c 
. (8.22)
 
−Ex /c 

 0
0

Let me inject a few words about general relativity here. (C.W. Misner, K.S. Thorne,
and J.A. Wheeler, Gravitation, pages 220–221.) There is a field four-tensor for grav-
itational field just as there is for electromagnetic field. It is the rank-4 “Riemann
tensor” Rµνσλ . This tensor has 44 = 256 components, but they’re not all inde-
pendent. Because of various symmetries and antisymmetries, there are “only” 20
92 CHAPTER 8. RELATIVISTIC ELECTRODYNAMICS

independent components. (By comparison the electromagnetic field four-tensor has

16 components, of which 6 are independent.)
Just as the 6 independent components of the electromagnetic field four-tensor can
be usefully dissected into two three-vector fields with 3 components each (electric and
magnetic), so the 20 independent components of the Riemann tensor can be usefully
dissected. This dissection produces, among other things, two symmetric, tracefree,
three-tensor fields with 5 independent components each. One of the people who
figured out how to do this is our own Rob Owen, so if you have questions you should
ask him.

8.7 Maxwell equations in relativistic notation

You have to fiddle around a lot to write the Maxwell equations in terms of the field
and dual four-tensors. When you do, you find:

The four source equations The four homogeneous equations

~ ·E
∇ ~ = ρ/0 ~ ·B
∇ ~ =0
~ ~
~ ×B
∇ ~ = µ0 J~ + 1 ∂ E ~ ×E
∇ ~ = − ∂B
c2 ∂t ∂t
are are
3 3
X ∂F µν µ
X ∂Gµν
= µ0 J =0
ν=0
∂xν ν=0
∂xν

All those funny curls and weird derivatives that show up in the traditional form
of the Maxwell equations. . . what were they there for? To make this relativistic result
come out so cleanly!
[[If you don’t like using the dual vector, the bottom right equation can be recast
as
∂Fµν ∂Fνλ ∂Fλµ
+ + = 0.
∂xλ ∂xµ ∂xν
I don’t like this form because: (i) These look like 4 × 4 × 4 = 64 equations but in
fact there are only 4 independent equations. (For example, if µ = 2, ν = 2, and
λ = 2 then this equation reads 0 + 0 + 0 = 0, which is true but not informative.)
(ii) To use it, you have to appreciate the distinction between F µν and Fµν (“covariant
components” vs. “contravariant components”). I have learned this distinction a few
dozen times and each time forgotten it a few weeks later.4 I suspect you’re the same
way. Learn this distinction when you need to use it.]]
4 Some people (including, I’m sorry to say, Griffiths on page 526) refer to a “covariant four-vector”

and a “contravariant four-vector” instead of to the “covariant components” and “contravariant

components” of a single four-vector. This is very bad. An arrow has one set of components in one
8.7. MAXWELL EQUATIONS IN RELATIVISTIC NOTATION 93

So, this is elegant! Physics 111, 311, and 411 is the study of these two equations.
But as Einstein pointed out

“Matters of elegance ought to be left to the tailor and to the cobbler.”5

and

“Since the mathematicians have invaded the theory of relativity, I do not

understand it myself anymore.”6

In fact, it represents more than elegance. If you write the Maxwell equations in
the traditional form shown on page 6, then transform them to the barred frame using
the equations on page 91, and if you also remember to transform all the positions
and times and charge densities and current densities correctly, you will be surprised
to find at the end that all the γs and βs exactly cancel out: the Maxwell equations
in the barred frame are exactly the same as the Maxwell equations on page 6. But
this way of doing the transformation is clearly a slog. In the four-tensor formulation,
it’s obvious that the Maxwell equations have the same form in all reference frames.
Maxwell himself didn’t understand this. He knew that his equations weren’t
Galilean invariant, but he thought that meant his equations applied only in the
reference frame of the ether.
~ r, t) combine
Challenge: The scalar potential V (~r, t) and the vector potential A(~
Ar
to make a four-vector potential ( ) through

A = [V /c, A].
~

Show that the fields are related to the potentials through

∂Aν ∂Aµ
F µν = − .
∂xµ ∂xν
For example, this equation with µ = 0, ν = 1 reads
 
Ex ∂Ax ∂(V /c) 1 ∂Ax ∂V
= − = − − .
c ∂(−ct) ∂x c ∂t ∂x
And now that it’s in this light, the strange combination
~
E ~ − ∂A
~ = −∇V
∂t
really does look sort of like a curl.
coordinate system and a different set of components in a rotated coordinate system. The arrow
(the vector ~r) doesn’t change when the coordinate system is rotated — only the coordinates change
(from (rx , ry , rz ) to (rx0 , ry0 , rz0 )). That’s the whole point of “vector” (see section 2.1, “What is
a vector?”). Similarly for the distinction between convariant and contravariant components: The
change in components does not mean there’s a change in four-vector.
5 Albert Einstein, Relativity: The Special and the General Theory (1916) page v.
6 Statement by Einstein as recalled by Arnold Sommerfeld, “To Albert Einstein’s Seventieth

Birthday” in Albert Einstein, Philosopher–Scientist, edited by Paul A. Schilpp (Library of Living

Philosophers, Evanston, Illinois, 1949) page 102.
94 CHAPTER 8. RELATIVISTIC ELECTRODYNAMICS

8.8 The stress-energy four-tensor

We’ve previously discussed the flow of momentum in space, and our discussion re-
sulted in the Maxwell stress tensor. In relativity we have to discuss the flow of
four-momentum in space-time. The result will be the stress-energy four-tensor.

As a technical appetizer, I just want to mention that the time-space four-vector

expressed in the laboratory frame

r =. [ct, ~r] (8.23)

has time derivative

r
d .
= [c, ~v ]. (8.24)
dt
This is not a four-vector, because we’re taking the derivative with respect to labora-
tory time t rather than with respect to the four-scaler proper time τ . But we’ll soon
find the result useful anyway. Now on to the main course.
8.8. THE STRESS-ENERGY FOUR-TENSOR 95

We build up the stress-energy four-tensor representation in the laboratory frame.

Remember that the negative of the Maxwell stress tensor described the flow of elec-
tromagnetic momentum. We called the momentum density ~g , and the velocity of a
plug of fields ~v , and described that momentum flow through the tensor
current density of . . .
 

 gx~v 
 ← x-momentum
← y-momentum
 
 gy ~v 
gz ~v ← z-momentum

↑↑↑
move in
direction
x, y, z
One lesson of relativity is that we can’t think of momentum in isolation: we have
to consider also the zero component of the energy-momentum four-vector, namely
energy/c. But we’ve already talked about the energy density u, so we fill in the zero
row of this four-tensor as
current density of . . .
 
u~v /c ← energy/c

 gx~v 
 ← x-momentum
←
 
 gy ~v  y-momentum
gz ~v ← z-momentum

↑↑↑
move in
direction
x, y, z
An even earlier lesson of relativity is that we can’t think of space in isolation: we
have to consider the zero component of the time-space four-vector, namely ct. Using
equation (8.24), we extend the bottom three rows of this four-tensor into the zero
column:
current density of . . .
 
u~v /c ← energy/c

 gx c gx~v 
 ← x-momentum
←
 
 gy c gy ~v  y-momentum
gz c gz ~v ← z-momentum

↑ ↑↑↑
move in move in
direction direction
ct x, y, z
96 CHAPTER 8. RELATIVISTIC ELECTRODYNAMICS

Finally we fill in the zero-zero component of the four-tensor

current density of . . .
 
u u~v /c ← energy/c

 gx c gx~v 
 ← x-momentum
←
 
 gy c gy ~v  y-momentum
gz c gz ~v ← z-momentum

↑ ↑↑↑
move in move in
direction direction
ct x, y, z
~ = u~v and
Recognizing that the flow of energy is related to the Poynting vector S
~ 2 , this four-tensor can be written
that the momentum density is ~g = S/c
~
 
u S/c
 S /c gx~v 
 x 
 
 Sy /c gy ~v 
Sz /c gz ~v
↔
or, remembering the Maxwell stress tensor T
~
 
u S/c
 S /c 
 x 
 ↔ 
 Sy /c −T 
Sz /c
and in this form, the four-tensor is clearly symmetric. It is called the “stress-energy
four-tensor”.
The stress-energy four-tensor resolves the conundrum raised in the problem as-
signment concerning the transformation of electromagnetic energy-momentum, but
the concept goes beyond electromagnetism: any flow of energy-momentum has an
associated stress-energy four-tensor. For example, a particle of mass m follows the
p
trajectory ~r(t) in the laboratory. Using ~v = d~r/dt and γ = 1/ 1 − (v/c)2 , I think
you can see for yourself that the stress-energy four-tensor is
γmc2
 
γmc~v
 γmv c γmvx~v 
x
 × δ (3) (~x − ~r(t)).
 

 γmvy c γmvy ~v 
γmvz c γmvz ~v
In Newton’s theory of gravity, mass is the source of gravity. But in Einstein’s
theory of gravity, general relativity, the stress-energy four-tensor is the source of
gravity. The zero-zero component above shows that mass is a source of gravity, but
the other components show that energy and momentum are also sources, and that
the flow of energy and momentum is also a source. This explains how light can be
bent by gravity: it has zero mass but it doesn’t have zero energy or zero momentum.
8.9. WHAT TO DO WITH RELATIVISTIC ELECTRODYNAMICS 97

8.9 What to do with relativistic electrodynamics

We have set up relativistic electrodynamics. The famous graduate-level textbook by

J.D. Jackson, Classical Electrodynamics, uses that setup to solve problems: he poses
a problem, finds the easiest frame in which to solve it, and then transforms back to
the lab frame in order to make testable predictions.
Chapter 9

Atmospheric Optics

We’ve been working at a high level of abstraction and generality. Nothing wrong
with that — it’s been stimulating and enlightening. But we started the course with
electrostatics: rubbing your hair with a balloon and then watching it cling to a wall.
Let’s conclude the course by coming back to everyday phenomena.

I want to think about light scattering from a transparent sphere, at the level of
geometric (ray) optics. Think of a sphere of radius R and index of refraction n (for
water, n = 1.333, for glass, n = 1.52), immersed in air (n = 1.000). Let me remind
you of three facts:

1. If light reflects off an interface, the angle of incidence equals the angle of reflec-
tion: θi = θr .

n1 θi θr

n2

98
 99

2. If light transmits through the interface, the angle of incidence θ1 in medium n1

is related to the angle of refraction θ2 in medium n2 through Snell’s Law

n1 sin θ1 = n2 sin θ2 .

n1 θ1

n2
θ2

3. If light reflects off an interface at Brewster’s angle where

tan θB = n2 /n1 ,

then the reflected light is polarized as shown.

n1

n2
100 CHAPTER 9. ATMOSPHERIC OPTICS

With this background let’s tackle the problem of light scattering from a sphere. A
ray of light enters the sphere a distance b (the “impact parameter”) above the sphere’s
center. (After finding the behavior of a ray entering above the axis indicated by the
dashed line, we will rotate about this axis to find the behavior of any ray.)

θ1
θ2
b θ2
θ2

θ2

θ1

This ray strikes the sphere with angle of incidence θ1 . If you like geometry, you’ll
enjoy proving that
b
sin θ1 = .
R
Some of this light is reflected, but let’s follow the ray that’s transmitted. This ray
strikes the surface of the sphere toward the right of the figure. Some of this light is
transmitted, but let’s follow the ray that’s reflected. This ray strikes the surface of
the sphere toward the bottom of the figure. Some of this light is reflected, but let’s
follow the ray that’s transmitted and leaves the sphere for good.

What is the total angle of bending suffered by this ray? Put on your geometry
hat again. At the first transmission it is bent by angle θ1 − θ2 . At the reflection it is
bent by angle π − 2θ2 . At the second transmission it is bent by angle θ1 − θ2 . The
total bend is thus
Θ = π + 2θ1 − 4θ2 . (9.1)

Remember that θ1 and θ2 are related through Snell’s Law

sin θ1 = n sin θ2 ,

so the bend is of course a function of b.

 101

An equation is not merely a jumble of symbols awaiting numbers to “plug in

and chug through”. An equation is a troubadour singing songs about nature. Let’s
examine this equation’s song. As b varies from 0 to R, other relevant parameters
vary as follows:

b : 0 → R
θ1 : 0 → π/2
sin θ1 : 0 → 1
sin θ2 : 0 → 1/n
θ2 : 0 → asin(1/n)
Θ : π → 2π − 4 asin(1/n)

To help us hear the song, we graph this dependence of Θ upon θ1 . First, find the
slope
dΘ dθ2
=2−4 .
dθ1 dθ1
But taking the derivative of both sides of Snell’s Law
dθ2
cos θ1 = n cos θ2 ,
dθ1
so
dΘ 4 cos θ1
=2− .
dθ1 n cos θ2
Thus at θ1 = 0,
dΘ 4
=2− ,
dθ1 n
a negative number for typical values of n, while at θ1 = π/2,
dΘ
= 2.
dθ1
The function Θ(θ1 ) starts off sloping downward, but ends up sloping upward. It must
have a minimum. For n = 1.333, the graph looks something like this.

0 θ1 π/2

The graph is, of course, flat near the minimum. Most of the outgoing rays will have
this value of Θ — call it the most popular value Θp . (All of the outgoing rays will
have Θ ≥ Θp .)
102 CHAPTER 9. ATMOSPHERIC OPTICS

The location of this popular value is given through

dΘ 4 cos θ1
0= =2− .
dθ1 n cos θ2
That is
4 cos θ1
2 =
n cos θ2
n cos θ2 = 2 cos θ1
n2 cos2 θ2 = 4 cos2 θ1 .

But the square of Snell’s law is

n2 sin2 θ2 = sin2 θ1
n2 (1 − cos2 θ2 ) = 1 − cos2 θ1
n2 − 1 + cos2 θ1 = n2 cos2 θ2 .

We can eliminate θ2 by putting these two together to get

n2 − 1 + cos2 θ1 = 4 cos2 θ1
n2 − 1 = 3 cos2 θ1
cos2 θ1 = (n2 − 1)/3
sin2 θ1 = (4 − n2 )/3. (9.2)

For any particular value of n, we can use this equation to find θ1 , then use Snell’s
law to find θ2 , and finally use the total bend equation (9.1) to find the most popular
total bend Θp .

[[Challenge: Our math shows that Θp depends on index of refraction n but not
on radius R. Can you produce any simple argument showing that Θp should be
independent of R?]]

Executing this program for water (n = 1.333) gives Θp = 138◦ — most of the
rays bend this much, that is they leave the sphere 42◦ from the incoming light beam.
(Because the index of refraction varies with color, this most popular angle will vary
somewhat with color.) Furthermore, in the n = 1.333 case the most popular reflection
angle θ2 is 40◦ , and just by coincidence Brewster’s angle for this reflection is 37◦ !
The exiting light is largely polarized perpendicular to the plane of the paper.

Rotating this construction around the dashed axis we find that the sphere scatters
a cone of light backwards, and this light is polarized parallel to the cone.
 103

What would happen if there were a bunch of water spheres in the sky, perhaps
very small ones, and you looked up at them? Here’s a planar sketch of the situation:

SUN

Figure out what’s happening within this sketch, then rotate about the dashed line of
symmetry to figure out what happens in three-dimensional space. What is the name
of this phenomenon?
Appendix A

Pictorializing divergence and

curl

The picture on the next page is the answer given by Purcell to the challenge poised
on page 19.

104
105