Phy 331 Advanced Elsctrodynamics PDF

PHY331: Advanced Electrodynamics & Magnetism
Part I: Electrodynamics
Contents
1 Field theories and vector calculus 3
1.1 Field theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Vector calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Second derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.5 Index notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2 Electromagnetic forces, potentials, Maxwells equations 13
2.1 Electrostatic forces and potentials . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 Magnetostatic forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3 Electrodynamics and Maxwells equations . . . . . . . . . . . . . . . . . . . 16
2.4 Charge conservation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.5 The vector potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3 Electrodynamics with scalar and vector potentials 21
3.1 Scalar and vector potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2 Gauge transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.3 A particle in an electromagnetic eld . . . . . . . . . . . . . . . . . . . . . . 22
4 Electromagnetic waves and Poyntings theorem 25
4.1 Electromagnetic waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.2 Complex elds? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.3 Poyntings theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
5 Momentum of the electromagnetic eld 31
6 Multipole expansion and spherical harmonics 34
6.1 Multipole expansion of a charge density . . . . . . . . . . . . . . . . . . . . 34
6.2 Spherical harmonics and Legendre polynomials . . . . . . . . . . . . . . . 36
6.3 Multipole expansion of a current density . . . . . . . . . . . . . . . . . . . 37
7 Dipole elds and radiation 41
7.1 Electric dipole radiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
7.2 Magnetic dipole radiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
7.3 Larmors formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
PHY331 PART I: ADVANCED ELECTRODYNAMICS LECTURE 0
8 Electrodynamics in macroscopic media 47
8.1 Macroscopic Maxwell equations . . . . . . . . . . . . . . . . . . . . . . . . 47
8.2 Polarization and Displacement elds . . . . . . . . . . . . . . . . . . . . . . 47
8.3 Magnetization and Magnetic induction . . . . . . . . . . . . . . . . . . . . 49
9 Waves in dielectric and conducting media 52
9.1 Waves in dielectric media . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
9.2 Waves in conducting media . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
9.3 Waves in plasmas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
10 Relativistic formulation of electrodynamics 59
10.1 Four-vectors and transformations in Minkowski space . . . . . . . . . . . . 59
10.2 Covariant Maxwell equations . . . . . . . . . . . . . . . . . . . . . . . . . . 62
10.3 Invariant quantities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
A Special coordinates and vector identities 67
A.1 Coordinate systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
A.2 Integral theorems and vector identities . . . . . . . . . . . . . . . . . . . . . 69
A.3 Levi-Civita tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
B Unit systems in electrodynamics 70
Literature
The lecture notes are intended to be mostly self-sufcient, but below is a list of recom-
mended books for this course:
1. R.H. Good, Classical Electromagnetism, Saunders College Publishing (1999). Most
of the material in this course is covered in this book. It is very accessible and
probably should be your rst choice to look something up.
2. J.D. Jackson, Classical Electrodynamics, Wiley (1998). This is the standard work on
classical electrodynamics, and it has everything that is covered in this course. The
level is quite high, but it will answer your questions.
3. R.P. Feynman, Lectures on Physics, Addison-Wesley (1964). Probably the best gen-
eral books on physics ever. The emphasis is on the physical intuition, and it is
written in a very accessible narrative.
4. J. Schwinger et al., Classical Electrodynamics, Westview Press (1998). This is quite a
high-level textbook, containing many topics. It has detailed mathematical deriva-
tions of nearly everything.
After each lecture there is a detailed further reading section that points you to the
relevant chapters in various books.
2
1 Field theories and vector calculus
1.1 Field theory
Michael Faraday
(1791-1867)
Electrodynamics is a theory of elds, and all matter enters
the theory in the form of densities. All modern physical
theories are eld theories, from general relativity to the
quantum elds in the standard model and string theory.
Therefore, apart from learning some important topics in
electromagnetism, in this course you will aquire an under-
standing of modern eld theories without having to deal
with the strangeness of quantum mechanics or the math-
ematical difculty of general relativity. Fields were intro-
duced by Michael Faraday, who came up with lines of
force to describe magnetic phenomena.
You will be familliar with particle theories such as clas-
sical mechanics, where the fundamental object is char-
acterised by a position vector and a momentum vector.
Ignoring the possible internal structure of the particles,
they have six degrees of freedom (three position and three
momentum components). Fields, on the other hand, are
characterised by an innite number of degrees of freedom.
Lets look at some examples:
A vibrating string: Every point x along the string has a displacement r, which is a
degree of freedom. Since there are an innite number of points along the string, the
displacement r(x) is a eld. The argument x denotes a location on a line, so we call the
eld one-dimensional.
Landscape altitude: With every point on a surface (x, y), we can associate a number
that denotes the altitude h. The altitude h(x, y) is a two-dimensional eld. Since the
altitude is a scalar, we call this a scalar eld.
Temperature in a volume: At every point (x, y, z) in the volume we can measure the
temperature T, which gives rise to the three-dimensional scalar eld T(x, y, z).
Mathematically, we denote a eld by F(r, t), where the value of the eld at position
r and time t is given by the quantity F. This quantity can be anything: if F is a scalar, we
speak of a scalar eld, and if F is a vector we speak of a vector eld. In quantum eld
theory, the mathematical object that makes the quantum eld are operators acting on a
vacuum state.
In this course we will be mostly dealing with scalar and vector elds, but occasion-
ally we will encounter tensor elds. A common example of a tensor eld that you may
have encountered is the stress in a material. We will discuss the difference between
tensors and ordinary matrices in a moment.
3
1.2 Vector calculus
In mechanics, a particle does not randomly jump around in phase space, but follows
equations of motion determined by the laws of mechanics and the boundary conditions.
These equations of motion typically involve derivatives, namely the velocity and accel-
eration of the particle. The derivatives tell you how much a quantity changes. Likewise,
elds obey equations of motion, and we need to dene the derivatives of elds.
First, take the scalar eld. The interesting aspect of such a eld is how the values of
the eld change when we move to neighbouring points in space, and in what direction
this change is maximal. For example, in the altitude eld (with constant gravity) this
change determines how a ball would roll on the surface, and for the temperature eld
it determines how the heat ows.
It is easy to see that both a rolling ball and heat ow have a magnitude and a direc-
tion. The measure of change of a scalar must therefore be a vector. Since the change is
dened at every point r (and time t), it is a vector eld. Let the scalar eld be denoted by
f (x, y, z, t). Then the change in the x direction (denoted by

i) is given by
lim
h0
f (x + h, y, z, t) f (x, y, z, t)
h
i =
f (x, y, z, t)
x
i . (1.1)
A
x
(r + l

i)
A
x
(r)
A
y
(r) A
y
A
z
(r)
A
z
(r + l

k)
l
l
l
r
Figure 1: The divergence of a vector eld A
can be found by considering the total ux of
Athrough the faces of a small cube of volume
l
3
at point r = x
i + y
j + z

k.
Similar expressions hold for the change in
the y and z direction, and in general the spa-
tial change of a scalar eld is given by
f
x
i +
f
y
j +
f
z
k = f gradf , (1.2)
called the gradient of f . The nabla or del
symbol is a differential operator, and it is
also a vector:
=

i

x
+

j

y
+

k

z
. (1.3)
Clearly, this makes a vector eld out of a
scalar eld. Note that we do not include
changes over time in the gradient. We rst
study static elds.
When we want to describe the behaviour
of vector elds, there are two main concepts:
the divergence and the curl. The divergence is
a measure of ow into, or out of, a volume
element. Consider a volume element dV =
l
3
at point r = x
i + y
j + z

k and a vector eld A(x, y, z), as shown in gure 1. We
assume that l is very small. The difference of the ux of the eld going into the volume
4
A(x l, y l) A(x + l, y l)
A(x + l, y + l) A(x l, y + l)
y
x x l x + l
y l
y + l
Figure 2: The curl of a vector eld A can be found by considering the change of the vector eld
around a closed loop.
and the ux coming out of the volume in the x direction l
2
A
x
(r +l

i) l
2
A
x
(r) is related
to the change of the component A
x
in the x direction. From the Taylor expansion we get
the approximation (l 1, and ultimately innitesimal)
A
x
(x + l, y) = A
x
(x, y) + l
A
x
x
(x, y) . (1.4)
This leads to
l
2
A
x
(x + l, y, z) l
2
A
x
(x, y, z) = l
3
A
x
x
.
In order to nd the total change in the vector eld we have to add the changes in the y
and z components as well, which then leads to
A
x
x
+
A
y
y
+
A
z
z
= A divA, (1.5)
where we removed the common factor l
3
in order to consider the divergence per unit
volume. By taking l innitesimally small, we can properly dene the divergence (and
later the curl) at a single point. The divergence is a measure of how the vector eld A(r)
spreads out at position r.
The curl of a vector eld is a measure of the vorticity of the eld. In two dimensions
(here meaning the xy plane), consider a four-part innitesimal closed loop starting at
point (x l, y l), going to (x + l, y l), (x + l, y + l) and via (x l, y + l) back to
(x l, y l), as shown in gure 2. The area of the square is 4l
2
. The accumulated
change of the vector eld around this loop is given by the projection of A along the line
elements. We evaluate all four sides of the innitesimal loop in Fig. 2. For example, the
line element that stretches from x l to x + l at y l is oriented in the x direction, and
the corresponding A dl is
_
A
x

i + A
y

j + A
z

k
_
_
2l

i + 0
j + 0

k
_
= 2l A
x
(x, y l) , (1.6)
5
where we evaluate A
x
in the middle of the line element at the point (x, y l). We repeat
the same procedure for the other three line elements. Make sure you get the orientation
right, because the top horizontal part is directed in the x direction. The corresponding
term gets a minus sign. Now add them all up and get A dl around the loop:
_
A dl = 2l A
x
(x, y l) + 2l A
y
(x + l, y) 2l A
x
(x, y + l) 2l A
y
(x l, y) . (1.7)
Just like for the divergence, we can use the rst two terms in Taylors expansion of
A
j
(x + l, y) to evaluate this formally, since l will be again an innitesimal length. For
example:
A
x
(x, y + l) = A
x
(x, y) + l
A
x
y
(x, y) . (1.8)
This leads to
_
A dl = 4l
2
_
A
y
x

A
x
y
_
. (1.9)
For a three-dimensional vector space, we not only have to take the contribution of a loop
in the xy plane, but also in the xz and the yz planes. Since there are three orthogonal
loops, they can be thought of as the components of a vector. For a three-dimensional
vector eld we thus have
i
_
A
z
y

A
y
z
_
+

j
_
A
x
z

A
z
x
_
+

k
_
A
y
x

A
x
y
_
= A curlA, (1.10)
the curl of a vector eld A per unit surface 4l
2
. We again take l 0 to dene the curl of
A(r) at a point r.
In four dimensions (which is relevant when we do relativity) we have loops not only
in the xy, xz, and yz planes, but also in the xt, yt, and zt planes. As you can see, there are
now six loops, which can no longer be regarded as components of a three-dimensional
vector. Therefore the cross product (and hence the curl operator) as a vector is special
to three-dimensional space.
In compact matrix notation, the curl can be written as a determinant
A =
i

j

k
x

y

z
A
x
A
y
A
z
. (1.11)
Sometimes, when there are a lot of partial derivatives in an expression or derivation, it
saves ink and space to write the derivative /x as
x
, etc.
The curl and the divergence are in some sense complementary: the divergence mea-
sures the rate of change of a eld along the direction of the eld, while the curl measures
the behaviour of the transverse eld. If both the curl and the divergence of a vector eld
A are known, and we also x the boundary conditions, then this determines Auniquely.
6
It seems extraordinary that you can determine a eld (which has after all an innite
number of degrees of freedom) with only a couple of equations, so lets prove it. Sup-
pose that we have two vector elds A and B with identical curls and divergences, and
the same boundary conditions (for example, the eld is zero at innity). We will show
that a third eld C = AB must be zero, leading to A = B. First of all, we observe that
C = AB = 0
C = A B = 0 , (1.12)
so C has zero curl and divergence. We can use a vector identity (see Appendix A) to
show that the second derivative of C is also zero:
2
C = ( C) (C) = 0 0 = 0 . (1.13)
This also means that the Laplacian
2
of every independent component of C is zero.
Since the boundary conditions for A and B are the same, the boundary conditions for
C must be zero. So with all that, can C be anything other than zero? Eq. (1.13) does not
permit any local minima or maxima, and it must be zero at the boundary. Therefore,
it has to be zero inside the boundary as well. This proves that C = 0, or A = B.
Therefore, by determining the divergence and curl, the vector eld is completely xed,
up to boundary conditions.
Looking ahead at the next lecture, you now know why there are four Maxwells
equations: two divergences and two curls for the electric and magnetic elds (plus their
time derivatives). These four equations and the boundary conditions completely deter-
mine the elds, as they should.
1.3 Second derivatives
In physics, many properties depend on second derivatives. The most important exam-
ple is Newtons second law F = ma, where a is the acceleration, or the second derivative
of the position of a particle. It is therefore likely that we are going to encounter the sec-
ond derivatives of elds as well. In fact, we are going to encounter them a lot! So what
combinations can we make with the gradient, the divergence, and the curl?
The div and the curl act only on vectors, while the grad acts only on scalars. More-
over, the div produces a scalar, while the grad and the curl produce vectors. If f is
a scalar eld and A is a vector eld, you can convince yourself that the six possible
combinations are
(f ) ( A) (f )
(A) (A)
2
A
Of these, the rst ( (f )) and the last (
2
A) are essentially the same, since in the
latter case the Laplacian acts on each component of A independently, and a component
of a vector is a scalar.
7
Another simplication is that two of the six expressions above are identically zero
for any f or A:
(A) = 0 and (f ) = 0 . (1.14)
Since these identities tend to simplify equations a lot, it is a good idea to learn them by
heart. The remaining derivatives are related by the vector identity
(A) = ( A)
2
A, (1.15)
which is also used very often. In fact, we just used it in Eq. (1.13). All these relations,
and more, can be found in Appendix A.
1.4 Tensors
Lets talk about vectors and tensors. You know that a vector is a quantity with a mag-
nitude and a direction. There is, however, also a more formal denition, which makes
it easier to generalise the concept of a vector to higher rank objects such as matrices (re-
member, a scalar has rank 0, a vector has rank 1, a matrix has rank 2, etc.). These objects
are called tensors. A tensor of rank 0 is just a scalar, and a tensor of rank 1 is just a vector.
A tensor of rank 2 is indeed a matrix, but is every matrix a tensor of rank 2? The answer
is no, because tensors must obey certain transformation properties. The formal denition
of a tensor says that it is an object that transforms in a special way under coordinate
transformations. To see what we mean by this, consider again the case of a vector.
We can write a vector a in Cartesian coordinates as a list of components (a
x
, a
y
, a
z
)
or, in our notation, a
x

i + a
y

j + a
z

k. But the coordinate system is something that we
choose, and has nothing to do with the vector itself. In particular, we can rotate our
coordinates around the z axis over an angle according to
x
= cos x + sin y
y
= sin x + cos y
z
= z . (1.16)
Since the vector a does not change, its description in the new coordinate system must
change, so
a = (a
x
, a
y
, a
z
) = (cos a
x
sin a
y
, sin a
x
+ cos a
y
, a
z
) . (1.17)
You can view this geometrically as follows: If we rotate our coordinate system over
an angle , then in the new coordinate system we must rotate the vector back over an
angle in order to describe the same vector. This means that you cannot choose any
old function of x, y and z as your vector eld, because it must obey this transformation
rule. If a
j
= f (x, y, z), then a
j
= f (x
, y
, z
), that is, the same function f . For example,

you can check using Eq. (1.16) that A = (y, x, 0) is not a proper vector eld.
8
a
a
x
a
x
a
y
a
Figure 3: Rotation of coordinate frame.

In general, we can write a linear coordinate transformation (such as Eq. (1.16)) as a
matrix equation:
_
_
x
_
_
=
_
_
cos sin 0
sin cos 0
0 0 1
_
_
_
_
x
y
z
_
_
, (1.18)
or, more compactly:
x
i
=
3
k=1
R
ik
x
k
, (1.19)
where we now write the general (non-Cartesian) coordinates as x
1
, x
2
, and x
3
, and R is
an orthogonal real matrix. Note that from this equation we can write
R
ik
=
x
i
x
k
. (1.20)
Our vectora can then be written in component form as
a
i
=
3
k=1
x
i
x
k
a
k
or a
i
=
x
i
x
k
a
k
, (1.21)
where we introduced Einsteins summation convention: When two indices are repeated
(here k), the sum is implied. This saves a lot of writing, but it also hides the complexity
to some extent. Make sure that in the beginning you write every expression with and
without sums, until you develop an intuition for the summation convention.
At last, we are ready to dene the tensor: A tensor of rank 2 is a matrix T
ik
that
transforms under coordinate transformations according to
T
i
k
=
3
i=1
3
k=1
x
i
x
i
x
k
x
k
T
ik
=
x
i
x
i
x
k
x
k
T
ik
. (1.22)
We have chosen to differentiate between coordinate systems by putting the primes on
the components. This is good practice, because the tensor T does not change, only the
9
coordinate systems change. A general tensor of rank n then obeys
T
i
1
...i
n
=
x
i
1
x
i
1
. . .
x
i
n
x
i
n
T
i
1
...i
n
. (1.23)
This includes the regular vector in Eq. (1.21).
We will not make use of tensors very often in this course, but they are indispensible
when talking about the momentum of the electromagnetic eld, and when we give the
relativistic description of Maxwells equations. You should therefore be aware that a
tensor exists, and that it is different from a matrix in its transformation properties.
1.5 Index notation
It is instructive to see how we can write the gradient, the divergence, and the curl in
index notation (using the summation convention). The gradient produces a vector, so
we can write this as
(f )
i
= (gradf )
i
=
i
f . (1.24)
Note that the i
th
component of the eld is given by the derivative
i
, as it should. The
divergence is written as
A = divA =
i
A
i
, (1.25)
where divA does not carry an index because it is a scalar. The repeated index i on the
right hand side is summed over, so it does not show up on the left hand side. The curl,
nally, makes use of the Levi-Civita tensor
i jk
, which returns 1 if the sequence i jk is an
even permutation of the index numbers 1, 2, and 3, and it returns 1 if the sequence i jk
is an odd permutation of 1, 2, and 3. When some indices are repeated (say i = j), then
the Levi-Civita tensor returns 0. Verify that the curl of a vector eld can be written as
(A)
i
= (curlA)
i
=
i jk
j
A
k
. (1.26)
You can always do an immediate check on equations like this, because the unpaired
indices on the left must match the unpaired indices on the right.
Further reading
D. Mowbray, Mathematics for Electromagnetism, PHY205
(http://www.david-mowbray.staff.shef.ac.uk/mathematics for electromagnetism.htm).
H.M. Schey, Div, Grad, Curl, and all that, Norton & Co. (1973).
G. Weinreich, Geometrical Vectors, Chicago Lectures in Physics (1998): An unorthodox but
very insightful treatment of vector calculus.
R.H. Good, Classical Electromagnetism, Saunders College Pub. (1999): Ch. 1, pp 1-32.
R.P. Feynman, Lectures on Physics, volume II, Addison-Wesley (1964): Ch. 2.
W.J. Dufn, Electricity and Magnetism, McGraw-Hill (1990): App. A & B, pp 390-405.
10
B.I. Bleaney & B. Bleaney, Electricity and magnetism, Vol. 1, Oxford University Press (1976):
App. A, pp A1-A12.
J.R. Reitz, F.J. Milford, & R.W. Christy, Foundations of Electromagnetic Theory, fourth edition,
Addison-Wesley (1993): Ch. 1, pp 1-25.
Exercises
1. Let f =
1
x
2
+y
2
+z
2
. Calculate
(a) f ,
(b) f .
2. Let A =
1
2
B(y, x, 0).
(a) Calculate B = A,
(b) Sketch A and B.
3. Prove that (A) = 0 and (f ) = 0.
4. Let Q = (y, x, 0). Can you nd a function T such that Q = T? If T is supposed
to be a temperature eld, what does Q represent? Interpret your result physically.
5. Calculate
_
S
A dS, where S is the surface of the unit sphere centered around the
origin, and A is (Hint: use Gauss theorem):
(a) A = (y, x, 0),
(b) A =
1
3
(x, y, z).
6. Calculate
_
C
A dl, where the closed loop C is the unit circle centered at the ori-
gin in the xy plane. It is traversed in anti-clockwise direction (Hint: use Stokes
theorem):
(a) A = (y, x, 0),
(b) A =
1
3
(x, y, z).
7. Using arrows of the proper magnitude, sketch
(a) (x, y)
(b) (1, 1)/
2
(c) (0, x)
(d) (
y
x
2
+y
2
,
x
x
2
+y
2
)
11
8. An object moves in the xy plane such that its position at time t is
r = (a cos t, b sint)
with a, b, and constants.
(a) How far is the object from the origin at time t?
(b) Find the velocity and the acceleration as a function of time.
(c) Show that the object moves in an elliptical path
_
x
a
_
2
+
_
y
b
_
2
= 1
9. Find a unit vector normal to each of the following surfaces
(a) z = 2 x y
(b) z = x
2
+ y
2
(c) z =
1 x
2
10. Calculate the divergence of F(x, y, z) = ( f (x), f (y), f (2z)) and show that it be-
comes zero at the point (c, c, c/2).
11. Let F(r) = r f (r) and r = (x, y, z)/r. Determine f (r) such that divF = 0.
12. Calculate the curl of the following functions:
(a) (z
2
, x
2
, y
2
)
(b) (3xz, 0, x
2
)
(c) (e
y
, e
z
, e
x
)
(d) (yz, xz, xy)
(e) (yz, xz, 0)
(f) (x, y, x
2
+ y
2
)
13. Show that curl
1
2
Ar = A for r = (x, y, z) and A constant.
14. Verify Stokes theorem when F = (z
2
, y
2
, 0), the closed loop C is given by the
square (0, 0, 0) (0, 0, 1) (1, 0, 1) (1, 0, 0) (0, 0, 0), and S is the surface
of unit cube between (0, 0, 0) and (1, 1, 1) without the side enclosed by C.
15. Verify the vector identities for the rst and second derivatives. Use index notation.
12
+ +
(a) (b)
Figure 4: Electric elds and potentials. (a) The electric eld lines (solid) and the equipotential
surfaces (dashed) due to a point charge. (b) The electric eld lines and equipotential surfaces
due to two opposite charges. The equipotential surfaces are always perpendicular to the eld
lines.
2 Electromagnetic forces, potentials, Maxwells equations
In this lecture we review the laws of electrodynamics as you have learned them previ-
ously, and write them in the form of Maxwells equations. We start with electrostatics
and magnetostatics, and then we include general time-dependent phenomena. We also
introduce the scalar and vector potential.
The theory of electrodynamics is mathematically quite involved, and it will get very
technical at times. It is therefore important to know when we are being mathematically
rigorous, and when we are just putting equations together to t the observed phenom-
ena. First, we postulate the laws. In fact, it was Coulomb, Biot and Savart, Gauss, Fara-
day, etc., who did measurements and formulated their observations in mathematical
form. There is nothing rigorous about that (although the experiments were amazing).
However, when Maxwell put all the laws together in a consistent mathematical frame-
work the rules of the game changed: In order to nd out what are the consequences of
these postulated laws, we have to be mathematically rigorous.
2.1 Electrostatic forces and potentials
We start our journey to the Maxwell equations with the electrostatic force: The electric
force on a particle with charge q is proportional to the eld E at the position of the
particle:
F = qE . (2.1)
The eld E itself must be generated by some charge density , and Gauss law relates
the ux of the electric eld lines through a closed surface to the charge density inside
the surface:
_
S
E dS =
_
V
E dr =
_
V
0
dr =
Q
0
, (2.2)
13
(a) (b) (c)
dl I
I
r r B
B
D
Figure 5: (a) The B eld due to a current I. (b) The same B eld as in (a) but now seen along the
axis of the current, pointing into the page. (c) The magnetic eld lines are always closed, and
therefore B = 0: the ux through the surface of a closed volume D is always zero.
where Q is the total (net) charge enclosed by the surface, and
0
= 8.85 . . . 10
12
Fm
1
is the electric permittivity of free space. This leads to the rst Maxwell equation:
E =

0
(2.3)
It determines the divergence of E.
The static electric eld can be expressed as the gradient of a scalar function , called
the scalar potential. This implies that the curl of E is zero:
E = E = 0 . (2.4)
Combining Eqs. (2.3) and (2.4), we obtain the Poisson equation
2
=
0
, (2.5)
which, in vacuum ( = 0) becomes the Laplace equation
2
= 0 . (2.6)
2.2 Magnetostatic forces
The force of a magnetic eld B on a charged particle is more complicated than the elec-
tric force, because it depends on the velocity v of the particle. The Lorentz force can be
stated as
F = q(E +vB) or dF = dV E + IdlB = (E +JB) dV , (2.7)
where I is a current, dl is a line element, and J is a current density. Magnetostatics means
that the work done by the magnetic part of the force is zero.
14
The Biot-Savart law states that the magnetic eld due to a current density J is given
by
B(r) =

0
4
_
J(r
)(r r
)
[r r
[
3
dr
, (2.8)
where
0
= 410
7
Hm
1
is the magnetic permeability of free space. Look at Fig. 5a
to read this equation: First, forget the integral sign. The B eld due to an innitesimally
small current Idl points into the page and is proportional to sin. This is consistent
with a cross product Idlr. Next, you take the contribution to B(r) of all innitesimal
currents, which gives you the integral. And rather than taking a wire through a current,
we generalize to the current density J(r
). From Figs. 5b and 5c you see that the magnetic

eld lines must be closed. This leads to B = 0, which we will now prove.
Using the vector identity
(ab) = (a) b a (b) (2.9)
we get
B(r) =

0
4
_
_
r r
[r r
[
3
(J(r
)) J(r
)
_
r r
[r r
[
3
__
dr
= 0 . (2.10)
The rst term in the integrand is zero because the curl takes derivatives to x, y, and z,
while the current density J is a function of the variables x
, y
, and z
. The second term

is zero because (r r
)/[r r
[
3
is the gradient of 1/[r r
[, and the curl of a gradient

eld must be zero. This gives us the second Maxwell equation:
B = 0 (2.11)
which is sometimes called Gauss law for magnetism. It determines the divergence of the
magnetic eld.
We nd Amp` eres law by taking the curl of the magnetic eld:
B(r) =

0
4
_
_
J(r
)
r r
[r r
[
3
(J(r
) )
r r
[r r
[
3
_
dr
=
0
J(r) . (2.12)
So for electrostatics and magnetostatics we have
E =

0
, E = 0 , B = 0 , B =
0
J . (2.13)
These are valid only when the E and B elds are constant in time. We consider time-
dependent elds in the next section.
15
2.3 Electrodynamics and Maxwells equations
C
B
I
Figure 6: Lenz law.
So far we have considered only electric and magnetic elds
that do not change in time. Now, lets consider general elec-
tromagnetic elds. Lenz law states that the electromotive
force c on a closed wire C is related to the change of the mag-
netic ux
B
through the loop:
c =
_
C
E dl =
d
B
dt
and
B
=
_
S
B(r, t) dS . (2.14)
Do not confuse
B
with the scalar potential : they are two
completely different things! In general, we can dene the
general (motional) electromotive force as
c =
_
C
F dl
q
=
_
C
E dl +
_
C
vB dl , (2.15)
which has both a purely electric component, as well as a magnetic component due to
vB. This is the motional e.m.f.
For a strictly electric e.m.f., we can use Stokes theorem to write
_
C
E dl =
_
S
(E) dS =
_
S
B(r, t)
t
dS , (2.16)
or
E +
B(r, t)
t
= 0 (2.17)
This is Faradays law, and our third Maxwell equation.
2.4 Charge conservation
Charges and currents inside a volume V bounded by the surface S can be a function of
time:
Q(t) =
_
V
(r, t) dr and I(t) =
_
S
J dS =
_
J dr . (2.18)
Conservation of charge then tells us that the change of the charge in V is related to the
ow of charge through S:
dQ(t)
dt
+ I(t) = 0 or

t
+ J = 0 . (2.19)
As far as we know today, charge conservation is strictly true in Nature.
16
The fourth, and last, Maxwell equation is a modication of Amp` eres law B =
0
J. As it turns out, Amp` eres law, as stated in Eq. (2.12) was wrong! Or at least, it does
not have general applicability. To see this, lets take the divergence of Eq. (2.12):
J =
1
0
(B) = 0 . (2.20)
But charge conservation requires that
J =
t
, (2.21)
James Clerk Maxwell
(1831-1879)
So Amp` eres law in Eq. (2.12) is valid only for static charge
distributions. We can x this by adding the relevant term
to Amp` eres law. Using Gauss law of Eq. (2.3), you see
that this is the time derivative of the E eld. This was
the great insight of James Clerk Maxwell (1831-1879) when
he unied the known laws of electrodynamics in his four
equations. The nal Maxwell equation therefore becomes
B
0
0
E
t
=
0
J (2.22)
and the complete set of (microscopic) Maxwell equations
is
B = 0 and E +
B
t
= 0 (2.23)
E =

0
and B
0
0
E
t
=
0
J . (2.24)
The top two equations are the homogeneous Maxwell equa-
tions (they are equal to zero), and the bottom two are
the inhomogeneous Maxwell equations (they are equal to a
charge or current density). For every Maxwell equation we have the behaviour of the
elds on the left hand side, and the source terms on the right hand side. This is gener-
ally how eld equations are written. In general relativity, the tensor eld describing the
curvature of space-time is related to the energy-momentum density, which includes all
masses. This way, it is clear how different source terms affect the elds.
2.5 The vector potential
Since the electric eld is the gradient of a scalar eld (the scalar potential), it is natural
to ask whether the magnetic eld is also some function of a potential. As the Lorentz
force shows, the magnetic eld is a good deal more complicated than the electric eld,
17
and this is reected in the magnetic potential. Rather than a scalar potential, we need a
vector potential to make the magnetic eld. The proof is easy: Since B = 0, we can
always write B as the curl of a vector eld A because
B = (A) = 0 , (2.25)
by virtue of a well-known vector identity.
At this point, the vector potential A, as well as the scalar potential , are strictly
mathematical constructs. The physical elds are E and B. However, you can see imme-
diately that whereas E and B have a total of six components (E
x
, E
y
, E
z
, B
x
, B
y
, and B
z
),
the vector and scalar potentials have only four independent components (, A
x
, A
y
,
and A
z
). This means that we can get a more economical description of electrodynamics
phenomena using the scalar and vector potential.
When the current is constant, we can choose a vector potential of the following form:
A(r) =

0
4
_
J(r
)
[r r
[
dr
. (2.26)
We can prove this by taking the curl on both sides
A(r) =

0
4
_
J(r
) dr
=

0
4
_

_
J(r
) dr
_
, (2.27)
with r
= [r r
[. Using the chain rule we can rewrite this as

A(r) =

0
4
_
J(r
) dr
+

0
4
_

_
1
r
_
J(r
) dr
. (2.28)
The rst term is zero because the derivative are to r and J is a function of r
. The second
term can be evaluated using (1/r
) = r
/r
2
. We therefore have
A(r) =

0
4
_
J(r)r
2
dV =

0
4
_
J(r)(r r
)
[r r
[
3
dV = B(r) . (2.29)
Further reading
R.H. Good, Classical Electromagnetism, Saunders College Pub. (1999): Ch. 2-6, pp 33-164.
J.D. Jackson, Classical Electrodynamics, Wiley (1998): Sec. 1.1-1.7 & 5.1-5.4, pp 24-35, 174-
181.
R.P. Feynman, Lectures on Physics, volume II, Addison-Wesley (1964): Ch. 15 & 18.
Addison-Wesley (1993): Ch. 2, 3, 7, 8, 11, & 16, pp 26-96, 162-217, 271-288, 386-411.
C.A. Brau, Modern Problems in Classical Electrodynamics, Oxford University Press (2004):
Sec. 0.1-0.5., pp 1-20.
J. Schwinger, L.L. DeRaad, K.A. Milton, &W. Tsai, Classical Electrodynamics, The Advanced
Book Program, Westview Press (1998): Ch. 1-2, pp 1-20.
18
Exercises
1. A line charge of constant linear charge density Cm
1
extends along the positive
y axis from 0 to L. Find the eld at the position a along the positive x axis.
2. Calculate the pressure (force per area) between two parallel plates that carry op-
posite charge densities .
3. Calculate the magnetic eld at a distance z above the center of a circular loop of
radius r carrying a current I. Evaluate this expression for z = 0, a = 1 cm, and
I = 1 A.
4. Show that charge conservation is implicit in Maxwells equations.
5. Give the expression of the magnetic ux for both B and the vector potential A.
What does the vector potential look like outside an innite solenoid of radius R that
connes a homogeneous eld B in the z direction pointing along the symmetry
axis?
6. Write Maxwells equations in index notation
7. Assessed Homework Exercise: A device that can generate large currents is the
so-called Faraday disk generator (see gure). It consists of a conducting thin disk,
rotating with angular velocity in a homogeneous magnetic eld in the z direc-
tion (B = B

k). The rim of the disk and the axis are connected via a wire with
Ohmic resistance R.
R
B
I
(a) Argue qualitatively that a current will ow in the direction indicated in the
gure. (2 points)
(b) Suppose the resistance is a light bulb. Where does the energy come from that
is dissipated in R? (2 points)
(c) Calculate the (motional) electromotive force between the rim and the center
of the disk. What is the current in the wire? (6 points)
19
(d) Explain why, when the rotation is not driven, the disk will stop spinning.
(Hint: calculate the Lorentz force due to the surface current on the disk. 5
points)
(e) Calculate the torque rF on the disk. (5 points)
8. An electron travels in a circular orbit around a proton that is xed in space with
angular momentum h (the Bohr model). Show that the magnetic eld at the loca-
tion of the proton is given by
B =

0
e h
4ma
3
0
, (2.30)
where e is the elementary charge, m is the mass of the electron, and a
0
is the Bohr
radius. Calculate the numerical value of B.
20
3 Electrodynamics with scalar and vector potentials
3.1 Scalar and vector potential
We can also write Maxwells equations in terms of the scalar and vector potential, but
we have to modify our relation E = : In electrodynamics E ,= 0, so E cant
be a pure gradient. Since we have B = 0, the construction B = A still works.
According to Faradays law
E +
B
t
= E +

t
A = 0 or
_
E +
A
t
_
= 0 . (3.1)
Therefore E +

A is a gradient eld:
E +
A
t
= or E =
A
t
. (3.2)
We should re-derive Poissons equation (2.5) from E = /
0
using this form of E:
2
=

t
( A) (3.3)
Similarly, we can write
B = (A) = ( A)
2
A
=
0
J +
0
t
_
A
t
_
, (3.4)
or
2
A
0
2
A
t
2
=
0
J +
0
t
_
+( A) (3.5)
Eqs. (3.3) and (3.5) are the Maxwell equations in terms of the scalar and vector potential.
This looks rather a lot worse than the original Maxwell equations! Can we simplify these
equations so that they look a bit less complicated? The answer is yes, and involves so-
called gauge transformations.
3.2 Gauge transformations
The electric and magnetic elds can be written as derivative functions of a scalar poten-
tial and a vector potential A. We have already seen that we can add a gradient eld
to the vector potential without affecting the eld equations for B. In general, we can
apply a gauge transformation to the scalar and vector potentials without changing any of
the physical content of the theory.
21
From B = A and () = 0 for all we see that we can always add a
gradient eld to the vector potential. However, we established that the electric eld
E depends on the time derivative of the vector potential. If E is to remain invariant, we
need to add the time derivative of to the scalar potential:
E
=
A
t

t
+
t
= E , (3.6)
since (
t
) =
t
(). It is clear that the full gauge transformation is
(r, t)
(r, t) = (r, t)
(r, t)
t
,
A(r, t) A
(r, t) = A(r, t) +(r, t) . (3.7)

On the one hand, you may think that it is rather inelegant to have non-physical de-
grees of freedom, because it indicates some kind of redundancy in the theory. However,
it turns out that this gauge freedom is extremely useful, because it allows us to simplify
our equations, just by choosing the right gauge .
In the Coulomb gauge (which is sometimes also called the radiation gauge), we set
A = 0. We will encounter this when we discuss electromagnetic waves. Another
useful gauge is the Lorenz gauge, in which we set
A +
0
t
= 0 . (3.8)
This leads to the following Maxwell equations:
_
2
t
2
_
(r, t) =
(r, t)
0
, (3.9)
_
2
t
2
_
A(r, t) =
0
J(r, t) . (3.10)
Gauge transformations are important in eld theories, particularly in modern quantum
eld theories. The differential operator in brackets in Eqs. (3.9) and (3.10) is called the
dAlembertian, after Jean le Rond dAlembert (17171783), and is sometimes denoted by
the symbol . The differential equations (3.9) and (3.10) are called dAlembert equa-
tions.
3.3 A particle in an electromagnetic eld
Often we want to nd the equations of motion for a particle in an electromagnetic eld,
as given by the vector potential. There are several ways of doing this, one of which
involves the Hamiltonian H(r, p), where r and p are the position and momentum of the
particle, respectively. You know the Hamiltonian from quantum mechanics, where it is
22
the energy operator, and it is used in the Schr odinger equation to nd the dynamics of the
wavefunction of a quantum particle. However, the Hamiltonian was rst constructed
for classical mechanics as a regular function of position and momentum, where it also
completely determines the dynamics of a classical particle. The idea behind this is that
the force on an object is the spatial derivative of some potential function. Once the
Hamiltonian is known, the equations of motion become
r
i
t
=
H
p
i
and
p
i
t
=
H
r
i
. (3.11)
These are called Hamiltons equations, and it is clear that the second equation relates the
force (namely the change of momentum) to a spatial derivative of the Hamiltonian.
William Rowan Hamilton
(1805-1865)
The derivation of the Hamiltonian is beyond the scope
of this course, but in vector notation it becomes:
H(r, p) =
1
2m
[p qA(r, t)]
2
q(r, t) . (3.12)
This expression can be further simplied for the specic
problem at hand by choosing the most suitable gauge. In
the weak eld approximation, we set [A[
2
= 0, and the
Hamiltonian is that of a free particle with a coupling term
2q p A.
As mentioned, the Hamiltonian completely deter-
mines the dynamics of a system, and as such is a very use-
ful quantity to know. In quantum mechanics, we replace
the position vector r by the operator
1
r and the momentum
vector p by the operator p. This makes the Hamiltonian an
operator as well. The Schr odinger equation for a particle
in an electromagnetic eld is therefore
i h
d
dt
[) =
_
1
2m
( p qA)
2
q
_
[) . (3.13)
The vector potential A and the scalar potential are classical elds (not operators). The
full theory of the quantized electromagnetic eld is quantum electrodynamics.
Further reading
J.D. Jackson, Classical Electrodynamics, Wiley (1998): Sec. 6.3, pp 240-243.
Sec. 2.4, pp 110-116; this is quite an advanced text.
1
Note that the hat now denotes an operator, rather than a unit vector. In the rest of the lecture notes
the hat denotes unit vectors.
23
Exercises
1. Derive Eqs. (3.3) and (3.5).
2. Construct a homogeneous B eld in the z direction by two vector potentials A
1
and A
2
(with B = A
1
= A
2
), one of which points in the x direction, and
one in the y direction. Find the gauge transformation that connects both poten-
tials.
3. For a charge q in a vector potential A due to some current, show that the charge
receives a momentum kick mv = qA when the current is suddenly switched
off.
4. Find the vector potential for a static charge in the gauge where = 0.
24
4 Electromagnetic waves and Poyntings theorem
In eld theories we can often nd nontrivial solutions to the eld equations (meaning
that the eld is not zero everywhere), even when there are no sources present. For
electrodynamics, in the absence of charges and currents we can still have interesting
effects. So interesting in fact, that it covers a whole discipline in physics, namely optics.
4.1 Electromagnetic waves
We obtain Maxwells equations in vacuum by setting = J = 0 in Eq. (2.23):
B = 0 and E =
B
t
(4.1)
E = 0 and B =
0
0
E
t
. (4.2)
Taking the curl of E and substituting B =
0
t
E we nd
_
E +
B
t
_
= ( E)
2
E +

t
_
0
E
t
_
= 0 , (4.3)
which, with E = 0, leads to the wave equation
_
2
t
2
_
E(r, t) = 0 . (4.4)
This differential equation has the well-know plane wave solutions
E = E
0
e
i(krt)
, (4.5)
where E
0
is a constant vector, k is the wave vector pointing in the propagation direction,
and is the frequency. By virtue of Eq. (4.4), the wave vector and frequency obey the
relation (k
2
= k k):
k
2
2
= 0 . (4.6)
Clearly, in SI units
2
0
has the dimension of inverse velocity-squared, and substitut-
ing the numerical values for
0
and
0
reveals that this velocity is c, the speed of light.
Another great triumph of Maxwells theory was the identication of light as electromag-
netic waves. For the right frequencies, the above equation is therefore the dispersion
relation of light propagating through vacuum:
k
2
=

2
c
2
. (4.7)
2
See Appendix B.
25
x
y
z
Figure 7: An electromagnetic wave with linear polarization. The E eld is pointed in the x
direction, while the B eld points in the y direction. The Poynting vector therefore points in the
z direction, the direction of propagation. The orientation of the E and B elds determine the
polarization of the wave.
So not only does Maxwells theory unify electric and magnetic phenomena, it also en-
compasses the whole of optics. On top of that, it predicts a range of new types of radi-
ation from radio waves and infrared in the long wavelengths, to ultraviolet, X-rays (or
R ontgen rays), and gamma rays in the short wavelength. Prior to Maxwells discovery
of electromagnetic waves, the known types of radiation other than light were infrared,
discovered by William Herschel in 1800, and ultraviolet, discovered in 1801 by Johann
Wilhelm Ritter. After Maxwell, Heinrich Hertz discovered radiowaves and microwaves
in 1887 and 1888, respectively. Wilhelm Conrad R ontgen discovered X-rays in 1895, and
Paul Ulrich Villard discovered gamma rays in 1900. However, it was not until the work
of Ernest Rutherford and Edward Andrade in 1914 that gamma rays were understood
as electromagnetic waves.
Similarly, Maxwells equations in vacuum give rise to a wave equation for the mag-
netic eld, yielding the solution B = B
0
exp[i(k r t)]. Since E = B = 0, the
electric and magnetic elds are perpendicular to the direction of propagation, k. Elec-
tromagnetic waves are thus transverse waves. It is left as a homework question to prove
that E B.
4.2 Complex elds?
You may have noticed that in the previous section the solution to the wave equation
gave us the complex eld of Eq. (4.5). But the electric eld must be a real quantity, oth-
erwise we would get complex forces, momenta, velocities, etc. This is clearly nonsense.
The usual way out of this is that we take the real part of Eq. (4.5) as the physical solution,
and we discard the imaginary part. We calculate things this way because it is easier to
deal with exponentials than with trigonometric functions.
However, it is instructive to delve a little deeper into this: why does the maths allow
26
complex solutions when all the quantities we put in (r, t,
0
, and
0
) are real? Notice
that a different solution of the wave equation is
E = E
0
e
i(krt)
, (4.8)
where the exponential has an overall minus sign. We can understand this as a wave
travelling in the k direction, rather than the +k direction. However, the frequency
must be a positive number, so that leaves us with a wave travelling in the t direction,
or backwards in time! We conclude that Maxwells equations are symmetric under time
reversal, just like Newtons laws and quantum mechanics.
The story is not nished, though, because we can superpose two solutions to a linear
differential equation to obtain a third solution. When we do this we can construct real
solutions for the eld:
E = E
0
e
i(krt)
+E
0
e
i(krt)
= 2E
0
cos(k r t) . (4.9)
More generally, we can take E
0
to be complex, in which case the cosine picks up a phase
shift (check this!). Every real eld can be written as a sum of two complex elds, one
travelling forward in time, and one travelling backward in time.
You may have heard slogans like anti-particles are particles moving backward in
time, and this is where that comes from. In quantum eld theory, the elds are usually
written in terms of the two components where one is the complex (actually, Hermitian)
conjugate of the other. The so-called positive frequency part corresponds to regular parti-
cles, while the negative frequency part corresponds to the anti-particles.
4.3 Poyntings theorem
John Henry Poynting
(1852-1914)
Electromagnetic waves carry energy. For example, the
Earth gets all of its energy from the Sun in the form of elec-
tromagnetic radiation. We therefore want to know what is
the energy density of the elds. We rst determine the en-
ergy density of the electric and magnetic elds separately.
Suppose we have a charge q
1
at position r
1
, and we bring
in a second charge q
2
from innity to the position r
2
. The
work done on the second charge by the eld is
W = q
2
_
r
2
E dl = q
2
1
(r
2
)
=
1
2
[q
1
2
(r
1
) + q
2
1
(r
2
)] , (4.10)
where the last equality follows from the fact that it does
not matter whether we bring q
1
from innity to r
1
before
or after we bring q
2
from innity to r
2
. We can symmetrize
27
c
B
I
Figure 8: Energy in B.
the work function. In general, the work needed to assem-
ble a charge distribution (r) is then
W =
1
2
_
V
(r)(r) dr =
0
2
_
V
2
dr =

0
2
_
V
dr =

0
2
_
V
E
2
dr .
The energy density of the electric eld in vacuum is therefore given by U
e
=
1
2
0
E
2
,
where E
2
= E E.
The energy density of the magnetic eld is slightly more involved. Since the Lorentz
force always acts perpendicular to the magnetic eld, the work done on a particle by
a static magnetic eld is zero. The work done on a particle therefore occurs when the
magnetic eld is changing, say in a closed loop C with a current I, and it creates an
electric eld, called the electromotive force c. The rate of work needed to keep the
current going can be found by Lenz law:
dW
dt
= cI =
_
C
E I dl =
_
V
E J dV . (4.11)
Now we use that here J = B/
0
and E =
t
B, and
(EB) = B (E) E (B) . (4.12)
This leads to
dW
dt
=
1
0
_
V
E (B) dr =
1
0
_
V
[B (E) (EB)] dr
=
1
0
_
V
B
B
t
dr
_
S
EB
0
dS =
1
0
_
V
B
B
t
dr
=
1
2
0
d
dt
_
V
B
2
dr , (4.13)
where in the second line we have used that we can take the volume to be the entire
space, and the surface integral becomes zero. The magnetic energy density is therefore
U
m
=
1
2
B
2
/
0
and the total energy density of the eld is
U =

0
E
2
2
+
B
2
2
0
. (4.14)
28
Continuing the manipulation of Eq. (4.12), from Maxwells equations we have
E (B) B (E) =
0
J E +

t
_
B
2
2
+
E
2
2c
2
_
. (4.15)
Using the vector identity in Eq. (4.12) we nd
(BE) =
0
J E +
0
t
U , (4.16)
where U = U
e
+U
m
. A little rearrangement of the terms will reveal Poyntings theorem:
U
t
+ S +E J = 0 , (4.17)
where S =
1
0
EB is the Poynting vector, after John Henry Poynting (1852-1914).
To interpret this theorem properly, lets integrate the expression over the volume V
bounded by the surface S (now not at innity):
d
dt
_
V
U dV +
_
S
S n dS +
_
V
J E dV = 0 . (4.18)
We again used Gauss theorem to relate a volume integral to a surface integral. The rst
term is the rate of change of the eld energy, the second term is the rate of the energy
ow through the surface, and the third term is the rate of work done on the charge
density.
Poyntings theorem is a statement of conservation of energy. The energy can leave
the elds in a region of space, but it must then be either transported through the sur-
face (measured by the Poynting vector), or it must be used to do work on the charges
in the volume. When we make the volume innitesimally small, you see that energy
conservation is not only true globally, but it is true locally, at every point in space.
Further reading
R.H. Good, Classical Electromagnetism, Saunders College Publishing (1999): Ch. 14, pp
336-370.
J.D. Jackson, Classical Electrodynamics, Wiley (1998): Sec. 6.8 & 7.1, pp 262-264-566, 295-300.
H.J. Pain, The Physics of Vibrations and Waves, Wiley (1983): Ch. 7, pp 187-218.
Ch. 8, pp 225-257.
sec. 0.5-0.6, pp 19-28.
29
Exercises
1. Show that E B for electromagnetic waves. Argue why the E and B elds must
be out of phase by /2.
2. The Maxwell equations can be written in terms of E, A, and as follows (in this
exercise we set c =
0
=
0
= 1):
E +
A
t
+ = 0 , and
E
t
(A) + J = 0 , and E = .
Furthermore, when a vector eld V is zero at innity, we can separate it into two
components V = V
r
+ V
g
, such that V
r
= 0 and V
g
= 0. V
r
is the
solenoidal part of the eld, while V
g
is the irrotational part.
(a) Show that the three equations above are equivalent to the Maxwell equations.
(b) Show that
A
r
t
+E
r
= 0 , and
A
g
t
+E
g
+ = 0 .
(c) The current density J can also be separated into J
r
and J
g
(J = 0 at innity).
Use this to write down the solenoidal and irrotational parts of the second
Maxwell equation above.
(d) From the above results, derive the wave equation for A
r
(Since J ,= 0, this is
a driven wave equation).
(e) Using the Lorenz gauge, derive the wave equation for A
g
.
(f) Derive the wave equation for the scalar potential .
(g) Finally, derive the wave equations for the scalar potential and the complete
vector potential A directly from the Maxwell equations above and the Lorenz
gauge.
3. The Poynting vector for complex E and B elds is given by S =
1
0
sin Re(E)Re(B),
where is the angle between E and B. For an electromagnetic wave with E(t) =
E
0
e
it
and B(t) = B
0
e
it+i
, show that the time averaged Poynting vector can
be written as
S
av
=
Re(EB
)
2
0
.
30
5 Momentum of the electromagnetic eld
It is not surprising that the electric and magnetic elds carry energy. After all, youre
quite familiar with the notion that it is the potential energy in the elds that make
charges move. The general behaviour of the energy in the elds is described by Poynt-
ings theorem, discussed in the previous lecture.
The electromagnetic eld also has momentum. There are several immediate reasons
why this must be the case:
1. The theory of electromagnetism is a relativistic theory, and we know that energy in
one frame of reference gives rise to momentum in another reference frame. There-
fore, to have energy is to have momentum
3
.
2. Last year you learned about radiation pressure, so you know that electromagnetic
waves can exert a force on objects, and therefore transfer momentum.
3. When two charged particles y close part each other with small relative velocity
(v c), Newtons third law holds, and the momentum of the two particles is
conserved. However, when the relative velocity becomes large, the particles will
each experience a B eld due to the other particles motion. We can congure the
situation such that the total momentum of the two charges is not conserved (see
exercise), and in order to save Newtons third law, the eld must be imbued with
momentum.
Now lets study the properties of the momentum of the electromagnetic eld. Since
momentum is closely related to force, we rst consider the Lorentz force.
We can write the Lorentz force F on a small volume V as
F = (E + JB)V . (5.1)
we now dene the force f on a unit volume exerted by the electromagnetic eld as
f = F/V = E + JB. Since we are interested in the elds, and not the charge or
current densities, we want to eliminate and J from the expression for f. We use the
inhomogeneous Maxwell equations for this:
f =
0
( E)E +
_
1
0
B
0
E
t
_
B. (5.2)
Next, we wish to rewrite the term

EB using the chain rule:
E
t
B =
(EB)
t
E
B
t
. (5.3)
Also, the term (B)B can be rewritten as
1
0
(B)B =
1
0
(B )B
1
2
0
B
2
=
1
0
(B )B +
1
0
( B)B
1
2
0
B
2
, (5.4)
3
Even though the actual value of the momentum in specic frames may be zero.
31
where we used the fact that B = 0, always. We put this term in to make the expres-
sion more symmetric. This leads to a horrible mess:
f =
0
[( E)E + (E )E] +
1
0
[( B)B + (B )B] U
1
c
2
S
t
, (5.5)
where U =
0
E
2
/2 +B
2
/2
0
is the potential energy per unit volume, and S is the Poynt-
ing vector.
By making the following clever substitution, this expression will simplify consider-
ably:
T
i j
=
0
_
E
i
E
j
i j
2
E
2
_
+
1
0
_
B
i
B
j
i j
2
B
2
_
. (5.6)
This is the Maxwell stress tensor, which we will interpret in due course. Since it is a rank
two tensor, for all practical purposes it behaves like a matrix. We are going to take the
divergence of T, which now will yield a vector, rather than a scalar. This comes down
to multiplying a vector with a matrix (where T =
i

i
T
i j
):
T =
0
_
( E)E + (E )E
1
2
E
2
_
+
1
0
_
( B)B + (B )B
1
2
B
2
_
. (5.7)
It is then easy to see that
f = T
1
c
2
S
t
. (5.8)
When we integrate this over the total volume, the total Lorentz force is
F =
_
V
_
T
1
c
2
S
t
_
d =
_
S
T da
d
dt
_
V
S
c
2
d , (5.9)
where da is the surface increment on the boundary surface S of volume V. The rst term
on the right-hand side is the pressure and shear on the volume, while the second term
is the time derivative of the electromagnetic momentum
_
S d/c
2
. The Lorentz force F
is moving around particles, and changing the amount of mechanical energy.
The tension in the elds (or eld lines) also explains why some congurations of
charges and magnets lead to repulsion, and some to attraction. The Maxwell stress
tensor makes precise the notion of tension and pressure in the eld lines.
Finally, now that we have an expression for the momentum of the electromagnetic
eld per unit volume
p =
S
c
2
=
EB
0
c
2
=
0
EB, (5.10)
we can dene the angular momentum per unit volume of the eld:
l = rp =
0
r(EB) . (5.11)
However, we will not investigate this further here.
32
Further reading
R.H. Good, Classical Electromagnetism, Saunders College Pub. (1999): Sec. 14.2, pp 343-347.
J.D. Jackson, Classical Electrodynamics, Wiley (1998): Sec. 6.7, pp 258-262.
Sec. 0.6, pp 24-28.
Book Program, Westview Press (1998): Ch. 3, pp 21-32.
Exercises
1. Write out Maxwells stress tensor in matrix form. What does the trace correspond
to?
2. Two positive charges move with velocities vA and v
B
as indicated in the gure
below. Give a qualitative estimate of the force on charge a and on charge b. Is the
A B
v
A
v
B
Figure 9: Two moving charges.
total momentum conserved in this situation?
3. A photon with energy h crosses a (cticious) surface area A in time t. Calculate
the momentum of the photon given that the momentum per unit volume of the
eld is given by S/c
2
.
4. Write the derivation starting with Eq. (5.1) to Eq. (5.5) in index notation.
33
6 Multipole expansion and spherical harmonics
Charge and current distributions in a small region of space may be very complicated,
and the potentials will generally not be simple. For example, a water molecule is elec-
trically neutral as a whole, but one side of the molecule is slightly positive (+q), while
the other side is slightly negative (q). If the two charges are separated by a distance d,
the potential of each is
+
(r) =
q
4
0
[r[
and
(r) =
q
4
0
[r d[
. (6.1)
Using the superposition principle to nd the scalar potential of the combined charges
, we nd (r d)
(r) =
q[r d[
4
0
[r[[r d[
+
q[r[
4
0
[r[[r d[

qd cos
4
0
r
2
, (6.2)
where measures the angle between d and r. This is the so-called dipole potential. Note
that it is no longer a 1/r potential but falls off faster, with 1/r
2
. If, for some reason
the potentials in Eq. (6.1) had different charges q
1
and q
2
= q
1
, then the scalar
potential would be
(r) =
q
1
4
0
[r[
+
q
1
+
4
0
[r d[

4
0
r
+
q
1
d cos
4
0
r
2
. (6.3)
Now the scalar potential is a superposition of a monopole and a dipole contribution. In
general, the scalar potential of a charge distribution will have a multipole expansion, and
the same will apply to the vector potential. Lets look at this in a bit more detail.
6.1 Multipole expansion of a charge density
Suppose that we have a localized (meaning enclosed in a volume V) static charge den-
sity (r), which gives rise to a scalar potential
(r) =
1
4
0
_
(r
) dr
[r r
[
. (6.4)
Far away from the charge density, the potential looks mainly like that of a point charge,
but with some higher-order corrections. These corrections are the multipoles, and we
can nd them by looking at the Taylor expansion of Eq. (6.4). Let
f (r r
) =
1
[r r
[
=
1
_
(x x
)
2
+ (y y
)
2
+ (z z
)
2
. (6.5)
Now we make a Taylor expansion of f (r r
) around r
= 0:
f (r r
) = f (r)
3
i=1
r
i
f
r
i
=0
+
1
2
3
i, j=1
r
i
r
2
f
r
i
r
j
=0
+ . . . (6.6)
34
d
r
r
r r
O
Figure 10: The eld at position r far away from a charge distribution (r) can be expressed
conveniently in terms of a multipole expansion. O denotes the origin of the coordinate system.
This can be rewritten in compact form as
1
[r r
[
=
1
r
r

1
r
+
1
2
(r
)
2
1
r
+ . . .
=
1
r
+
3
i=1
r
i
r
i
r
3
+
1
2
3
i, j=1
r
i
(3r
i
r
j
r
i j
)r
j
r
5
+ . . . (6.7)
where r =
_
r
2
1
+ r
2
2
+ r
2
3
= [r[ is the magnitude of the vector with components r
i
. After
substituting this into the generic form of the scalar potential in Eq. (6.4), we nd
(r) =
1
4
0
_
(r
) dr
r
+
1
4
0
3
i=1
_
r
i
r
i
r
3
(r
) dr
+
1
8
0
3
i, j=1
_
3r
i
r
j
r
i j
r
5
r
i
r
j
(r
) dr
+ . . . (6.8)
=
1
4
0
_
Q
r
+
3
i=1
Q
i
r
i
r
3
+
1
2
3
i, j=1
r
i
Q
i j
r
j
r
5
+ . . .
_
. (6.9)
The multipole moments of the charge distribution are the total charge Q, the dipole mo-
ment Q
i
, the quadrupole moment Q
i j
, etc. They are dened as follows:
Q =
_
(r
) dr
(6.10)
Q
i
=
_
r
i
(r
) dr
(6.11)
Q
i j
=
_
(3r
i
r
j
r
i j
) (r
) dr
(6.12)
.
.
. (6.13)
35
The quadrupole moment has nine components, but it is easy to see that Q
i j
= Q
ji
, and
3
i=1
Q
ii
=
3
i=1
_
(3r
i
r
i
r
2
)(r) dr =
_
__
3
3
i=1
r
i
r
i
_
3r
2
_
(r) dr = 0 . (6.14)
The quadrupole moment is therefore characterised by ve independent variables.
6.2 Spherical harmonics and Legendre polynomials
It is clear from the previous section that we can continue the multipole expansion to
octupoles and higher, but it is also pretty obvious that the polynomials in the denitions
of Q
i jk...
get quite unwieldy very quickly. Luckily, there is a more systematic approach
based on spherical harmonics. These are functions of the spherical coordinates and of
a vector r, and the function f can then be written as
1
[r r
[
=
l=0
+l
m=l
r
l
r
l+1
_
4
2l + 1
Y
lm
(, )
_
4
2l + 1
Y
lm
(
) (6.15)
with
Y
lm
(, ) =
2l + 1
4
(l + m)!
(l m)!
e
im
sin
m
_
d
d cos
_
lm
(cos
2
1)
l
2
l
l!
. (6.16)
This is still rather complicated, but the advantage is that this is valid for all l. Note also
that the spherical harmonics are complex, but Eq. (6.15) is still real due to the sum over
m. You should think of the Y
lm
as basis functions that can be used to write arbitrary
functions (of and ) as a series, just like any polynomial can be written as a series
n
a
n
x
n
, or a periodic function as a Fourier series. Like any proper set of basis functions,
the Y
lm
obey an orthogonality relation:
_
d Y
lm
(, )Y
l
m
(, )
_

0
sin d
_
2
0
d Y
lm
(, )Y
l
m
(, ) =
ll

mm
, (6.17)
where we introduced the integration over the sold angle d.
We can now dene the Legendre polynomials as
P
l
(w) =
_
d
dw
_
l
(w
2
1)
l
2
l
l!
, (6.18)
with normalisation P
l
(1) = 1. In the spherical harmonics, we have set w = cos . The
Legendre polynomials also obey an orthogonality relation
_
1
1
dw P
l
(w)P
l
(w) =
2
2l + 1

ll
. (6.19)
36
Figure 11: The uctuations of the cosmic microwave background radiation as measured by
WMAP. On the right is the plot of the amplitude of Y
lm
for different values of l. The plot is going
all the way up to l = 1 000.
The Legendre polynomials are closely related not only to the multipole moments
Q
i jk...
, but also to the quantum mechanical wavefunction of a particle in a central poten-
tial. The index l is then the quantum number for the orbital angular momentum of the
particle, and m is the quantum number of its z-component. The most famous example
of this is the wavefunction of the electron in a hydrogen atom. The more general spher-
ical harmonics also have many applications outside electrodynamics. One example is
the description of the temperature uctuations of the cosmic microwave background.
In gure 11 the relative strength of the harmonics are plotted against l.
6.3 Multipole expansion of a current density
In order to nd the magnetic multipole expansion due to complicated stationary current
desities, we use the time-independent form of the Maxwell equation (3.5) in either the
Lorentz gauge or the Coulomb gauge (that makes no difference here):
2
A(r) =
0
J(r) . (6.20)
The three components of this vector equation are independent of each other, so they are
effectively three scalar Poisson equations. If the source J vanishes at innity (in other
words, it is localized), then we can write this as
A(r) =

0
4
_
J(r
) dr
[r r
[
. (6.21)
If we assume that the total charge in the system does not change (d/dt = 0), then
we can use the fact that J = 0. We can again insert the Taylor expansion for the
37
function 1/[r r
[ given by Eq. (6.6), which leads to

A
i
(r) =

0
4
_
1
r
_
J
i
(r
) dr
+
1
r
3
3
j=1
_
r
j
r
j
J
i
(r
) dr
+ O(r
5
)
_
. (6.22)
The rst term in square brackets is the magnetic monopole term, which we expect to
vanish for magnetostatics, and the second term is the magnetic dipole moment.
To show that the magnetic monopole contribution vanishes for all J(r), note that
_
J
i
(r
) dr
=
3
j=1
_
J
j
(r
)
r
i
r
j
dr
. (6.23)
Integration by parts of the right-hand side of this expression gives
_
J
i
(r
) dr
=
3
j=1
_
J
j
(r
)r
i
d
2
r
k,=j
j
=
r
j
=
j=1
_
r
i
J
j
(r
)
r
j
dr
. (6.24)
The rst termof the right-hand side is zero because the current density at innity is zero:
we consider a localized current density. The differential in the second term (together with
the sum) is J, which, we already determined, is zero.
The magnetic dipole moment is therefore the lowest order term in Eq. (6.22). Lets
separate it into the symmetric and anti-symmetric parts:
r
j
J
i
=
1
2
_
r
j
J
i
+ r
i
J
j
_
+
1
2
_
r
j
J
i
r
i
J
j
_
. (6.25)
The integral over the symmetric part is zero. You can see this by integration by parts:
3
j=1
r
j
_
_
r
j
J
i
+ r
i
J
j
_
dr
=
3
j,k=1
r
j
_
_
r
j
r
i
r
k
J
k
+ r
i
r
j
r
k
J
j
_
dr
=
3
j,k=1
r
j
_
_
r
j
r
i
J
k
d
2
r
l,=k
k
=
r
k
=
_
r
i
J
j
(r
)
r
j
dr
_
= 0 . (6.26)
In the rst line we again used the Kronecker delta representation r
i
/r
j
=
i j
, and in
the second line we used the same tricks to show that the monopole moment is zero.
The lowest order moment of the vector potential is therefore
A
i
(r) =

0
8r
3
3
j=1
_
r
j
_
r
j
J
i
(r
) r
i
J
j
(r
)
_
dr
. (6.27)
It is very tempting to write the antisymmetric part r
j
J
i
r
i
J
j
as a cross product. This
leads to the denition of the magnetic dipole moment of the current distribution
m =
1
2
_
rJ(r) dr . (6.28)
38
However, we cant just plug the expression for m in Eq. (6.28) into Eq. (6.27), because
the components do not match up: The cross product (r
j
J
i
r
i
J
j
) in Eq. (6.27) makes use
of component i, which according to the equation should contribute to A
i
. This is not
possible unless another cross product is involved. If we call
1
2
_
(r
j
J
i
r
i
J
j
)dr
= m
k
with k ,= i, j, then we have the equation
A
i
(r) =

0
4r
3
3
j=1(,=i)
k,=j,i
r
j
m
k
. (6.29)
You see that the A
i
component depends on terms r
j
and m
k
, with neither j nor k equal
to i. This is crying out for another cross product, and it will involve m and r. So lets see
if we can massage mr such that we obtain Eq. (6.27):
(mr)
i
=
3
j=1
1
2
_
_
r
j
r
j
J
i
r
j
r
i
J
j
_
dr
. (6.30)
The integrand involves the antisymmetric part of r
j
J
i
. However, since we have just
proved that the integration over the symmetric part is zero, we can add this to the
integrand without affecting the integral. But adding the symmetric and anti-symmetric
part of r
j
J
i
is just r
j
J
i
, and the cross product becomes
(mr)
i
=
3
j=1
1
2
_
r
j
r
j
J
i
(r
) dr
. (6.31)
This leads immediately to the vector potential of a magnetic dipole:
A(r) =

0
4
mr
r
3
. (6.32)
Further reading
J.D. Jackson, Classical Electrodynamics, Wiley (1998): Sec 4.1 & 5.6, pp 145-150, 184-188.
Sec. 1.4, 2.3 & 4.2, pp 11-14, 38-43, 101-107.
Addison-Wesley (1993): Sec. 2.9, pp 46-48.
39
Exercises
1. Let the charge distribution (r, t) consist of two static charges (r, t) = q(r)
q(r d). Calculate the total charge Q, the dipole moment Q
i
, and the quadrupole
moment Q
i j
by evaluating Eqs. (6.10), (6.11), and (6.12). How does the position of
the origin affect the outcome?
2. A sphere of radius R rotates with angular velocity around a symmetry axis,
and carries a uniformly distributed surface charge Q. Give the charge and current
densities and J, and verify that J = 0.
3. The magnetic moment of an electron is one Bohr magneton,
B
= e h/2m. Suppose
we model this as a small ring current of radius r, which can be considered the size
of the electron. What is the smallest possible size of the electron according to this
model? Experimentally, the electron behaves as if it is a point particle. Is this a
problem for this model?
4. Verify Eqs. (6.7) and (6.8).
40
7 Dipole elds and radiation
7.1 Electric dipole radiation
s
+q
q
r
r
Figure 12: The Hertzian dipole:
Radiation from an electric dipole
current.
So far we have looked at properties of electromagnetic
waves (radiation) without asking how they are pro-
duced. Since radiation is an electromagnetic effect,
we expect that it is created by electric charges. How-
ever, both static and uniformly moving charges cannot
produce radiation (a uniformly moving charge is the
same as a static charge in a different frame) because
there is no characteristic time scale
1
in the phys-
ical system. We therefore need to look at accelerating
charges, and to this end we construct an electric dipole
in which the charge q oscillates between two points
separated by a distance s (see gure 12). The charge
at the top of the dipole can be written as Q = q e
it
,
and we have a dipole p = sq e
it
in the z direction.
This is equivalent to a sinusoidal current I
0
e
it
.
We now calculate the scalar and vector potentials
in spherical coordinates for this situation. At some
distance r the scalar potential due to the top of the
dipole is given by
(r, t
) =
q e
it
4
0
r
=
q e
i(tr/c)
4
0
r
=
q e
it+ikr
4
0
r
=
R
(r, t) , (7.1)
where t
= t r/c is the retarded time: It takes a while for the change in the potential at
r = 0 to propagate to r. To take this into account, we introduced the retarded potential
R
(r, t), and we used the dispersion relation for radiation in free space k = /c.
Next, we add the two retarded potentials of the two charges
(+)
R
and
()
R
=
(+)
R
in the dipole. They almost cancel, but for the small separation s in the z direction, which
gives rise to r in the r direction:
R
(r, t) =
(+)
R
+
()
R
=
(+)
R
(r, t)
(+)
R
(r +r, t) r
(+)
R
r
(7.2)
Substituting Eq. (7.1) and r = s cos we obtain
R
(r, t) = s cos
r
_
q e
i(krt)
4
0
r
_
=
qs e
i(krt)
k cos
4
0
r
_
1
kr
i
_
41
=
p
R
k cos
4
0
r
_
1
kr
i
_
, (7.3)
where p
R
is the retarded dipole moment.
The vector potential of this oscillating dipole in spherical coordinates is
A
R
(r, t) =

0
4
_
J
R
(r, t)
r
dV =
1
4
0
c
2
r
I
0
e
i(krt)
s =
1
4
0
c
2
r
I
R
s , (7.4)
where s is the dipole vector s(cos r sin

), parallel to the z axis, and I
R
is the re-
tarded current in the dipole. Using the relations Q = q e
it
and I = dQ/dt, we can
write the dipole moment p
R
in the scalar potential in terms of I
R
:
p
R
=
is
I
R
. (7.5)
We therefore have
R
(r, t) =
I
R
s cos
4
0
cr
_
1 +
i
kr
_
and A
R
(r, t) =
I
R
s(cos r sin

)
4
0
c
2
r
. (7.6)
The electric eld E is given by E =
R

A
R
. Using the derivatives in spherical
coordinates we nd
E
r
(r, t) =
R
r
(

A
R
)
r
=
kI
R
s cos
4
0
cr
_
2
kr
+
2i
(kr)
2
_
,
E
(r, t) =
1
r
(

A
R
)
=
kI
R
s sin
4
0
cr
_
1
kr
+
i
(kr)
2
i
_
. (7.7)
We already know that radiation carries energy, so the total energy ux through a closed
surface around the source is constant. If we take the surface to be that of a sphere
with radius r, the energy ux 4r
2
([E[
2
/2 +[B[
2
/2
0
) must be constant for all r. The
radiating part of the E and B elds are therefore proporional to 1/r. Looking at Eq. (7.7),
we see that the only contribution to the radiation is due to the i term in E
. Similarly,
the only contributing term of the B eld is (see exercise)
B
(r, t) = i
kI
R
s sin
4
0
c
2
r
=
E
(r, t)
c
, (7.8)
where E
(r, t) now denotes the radiating part of the E eld. Poyntings vector is given
by
S
e
=
ReE
ReB
0
= Re
EB
0
=
1
16
2
0
c
_
ksI sin
r
_
2
r . (7.9)
Note that we can calculate the Poynting vector by taking the complex conjugate of the B
eld and take the real part of S afterards. Again, it is often much easier to calculate with
complex phase factors of the form e
i
than with the trigonometric counterparts sin,
cos and tan.
42
E
B
Figure 13: The directions of the E and B elds of a spherical wave radiated by a dipole in the
centre. After half the oscillation period the elds point in the opposite direction.
7.2 Magnetic dipole radiation
s
s
I
r
Figure 14: Radiation from a mag-
netic dipole current.
Now suppose we have a small current going in a
square loop of area s
2
, leading to a magnetic dipole
moment m = s
2
I
0
e
it
. The current is changing in
time in a periodic way. Again, we can dene the re-
tarded magnetic dipole by substituting t t r/c. If
the current has a low impedance the scalar potential is
zero, while the vector potential is
A
R
(r, t) =

0
m
R
k sin
4r
_
1
kr
+ i
_

. (7.10)
We can again eld the radiating parts of the E and B
elds via E =

A and B = A:
E
=

0
m
R
k
2
c sin
4r
= cB
, (7.11)
and the Poynting vector is
S
m
= Re
EB
2
0
=
1
16
2
0
c
_
(ks)
2
I sin
r
_
2
r . (7.12)
The difference between S
e
and S
m
is a factor (ks)
2
.
7.3 Larmors formula
Clearly, radiation is generated by accelerated charges and changing currents. In the
previous sections we have looked at periodically changing charges and currents, but
43
in many situations the charges and currents change in an aperiodic way. For example,
when a charge is kicked, it will emit a burst of radiation. We will now consider the
question of how much power P is radiated when a charge is accelerated by an amount
a. This will lead to Larmors formula.
Joseph Larmor
(1857-1942)
Suppose a particle with charge q follows a trajectory
r
q
(t). We can formally associate a dipole moment p to this
particle:
p(t) = q r
q
(t) . (7.13)
The current due to the motion of the charge is clearly re-
lated to the velocity v
q
, so we can write
I =
dp
dt
= q
dr
q
dt
= q v
q
(t) . (7.14)
The acceleration a
q
of the particle is proportional to the
second derivative of the dipole moment:
d
2
p
dt
2
= q a
q
(t) . (7.15)
Next, we calculate the vector potential and the magnetic eld B of the moving charge.
First, we note that the current I is the spatial integration over the current density, so
we have
A(r, t) =

0
4r
_
J(r
, t r/c) dr
=

0
4r
I
R
=

0
4r
dp
dt
(t r/c) , (7.16)
evaluated at the retarded time, because the eld needs time to propagate from the charge
to the point where the vector potential is calculated.
The magnetic eld is found using B = A. We need to keep in mind two things:
(1) We can ignore all the contributions that do not scale like 1/r, because they do not
propagate. (2) We must remember that the time variable depends on r because we
consider the retarded time. The magnetic eld is then
B(r, t) = A(r, t) =

0
4
_
_
1
r
_
dp
dt
+
1
r
dp
dt
_
=

0
4r
dp
dt
, (7.17)
where the last equality holds because (1/r) leads to a non-propagating contribution
to the eld. The calculation of the curl of dp/dt is a bit tricky, so well do it here in some
detail. In particular, it is convenient to use index notation:
_
dp
dt
_
i
=
i jk
j
_
dp
dt
_
k
. (7.18)
Using the chain rule, we can write (t = t
q
r/c with )
j
=
t
x
j
d
dt
=
1
c
x
j
r
d
dt
, (7.19)
44
which allows us to rewrite the cross product:
_
dp
dt
_
i
=
1
c
i jk
x
j
r
_
d
2
p
dt
2
_
k
. (7.20)
In vector notation, this becomes
dp
dt
=
1
cr
r
d
2
p
dt
2
=
1
c
r
d
2
p
dt
2
=
q
c
ra
q
, (7.21)
and therefore the B eld can be written as
B(r, t) =

0
4r
dp
dt
=

0
4cr
r
d
2
p
dt
2
=
q
0
4cr
ra
q
, (7.22)
evaluated at time t, and due to an accelerated charge at the retarded time t r/c.
The radiating part of the electric eld E must be perpendicular to the radiating part
of the magnetic eld that we just calculated, and they differ by a factor c. Keeping the
correct handedness, the radiating E eld can be written as a cross product between B
and r:
E(r, t) = c B r =
q
0
4r
_
ra
q
_
r . (7.23)
Now that we know the E and B elds, we can calculate the Poynting vector S, which
gives the energy ux through a surface. If we take the surface to be a sphere with radius
R surrounding the accelerating charge, we can calculate the total radiated power
P =
_
S
S

R R
2
d, (7.24)
where d is a solid angle.
The Poynting vector is straightforward to calculate:
S =
EB
0
=

0
q
2
16
2
cr
2
__
ra
q
_
r
_
ra
q
_
. (7.25)
Using the vector identity A(BC) = B(A C) C(A B) we rewrite this as
S =
q
2
16
2
0
c
3
r
2
_
a
q
r(a
q
r)
_
ra
q
_
=
q
2
16
2
0
c
3
r
2
_
r
_
[a
q
[
2
(a
q
r)
2
_
a
q
_
a
q
r a
q
r
_
_
=
q
2
16
2
0
c
3
r
2
_
a
2
q
(a
q
r)
2
_
r
=
q
2
a
2
q
sin
2
16
2
0
c
3
r
2
r , (7.26)
45
where we dened a
q
= [a
q
[ and we oriented our coordinate system such that the accel-
eration is along the z direction. The inner product a
q
r is therefore equal to a
q
cos .
The radiated power through a sphere surrounding the accelerating charge is now
straightforward to calculate:
P =
_
S
S

R R
2
d =
_
S
q
2
a
2
q
sin
2
16
2
0
c
3
R
2
R
2
d =
q
2
a
2
q
16
2
0
c
3
_

0
_
2
0
sin
2
d d. (7.27)
The integral over sin
2
is equal to 8/3, so we arrive at
P =
1
4
0
2q
2
a
2
q
3c
3
. (7.28)
This is the celebrated Larmor formula, derived by Joseph Larmor in 1897.
Further reading
R.H. Good, Classical Electromagnetism, Saunders College Publishing (1999): Sec. 15.1 &
15.2, pp 371-379.
J.D. Jackson, Classical Electrodynamics, Wiley (1998): Sec. 9.1-9.3 pp 407-416.
Exercises
1. Atoms can have both an electric and a magnetic dipole that produces radiation.
Which of these typically two produces stronger radiation?
2. Consider again the Bohr model, in which an electron orbits a proton. The size of
the orbit is given by the Bohr radius (5.29 10
11
m), and the angular momentum
of the ground state is h = 1.05 10
34
Js. According to classical electrodynamics,
how strong is the radiated power? Give an order-of-magnitude estimate of the
lifetime of the atom.
46
8 Electrodynamics in macroscopic media
8.1 Macroscopic Maxwell equations
So far, we have considered the elds of (moving) charges and the interactions between
the charges. In principle, this covers all classical electromagnetic phenomena, since all
matter is made up of charged particles that interact with each other. However, if we
are dealing with 10
23
atoms we dont want to solve the Maxwell equations for every
individual atom. Especially when the time and length scales relevant to our problem
are large compared to that of the atoms, we much rather average over the atomic elds
and have effective elds in the media.
The electric and magnetic elds in Maxwells equations are now average elds E)
and B), and the charge and current densities ) and J) now consist of free charges
and currents as well as bound charges and currents. Maxwells equations in macroscopic
media then become
B) = 0 and E) +
B)
t
= 0 (8.1)
E) =
)
0
and B)
0
0
E)
t
=
0
J) . (8.2)
The charge and current densities are
) =
f
+
b
and J) = J
f
+J
b
(8.3)
where the subscripts f and b denote free and bound, respectively.
The elds in the materials are no longer the bare E and B elds due to freely moving
charges and currents (as was the case for the microscopic Maxwell equations), but they
are modied by the presence of the bound charges and currents. We want to relate
the modied elds to material properties (the bound charges and currents) and modify
Maxwells equations such that they involve the macroscopic elds and the free charges
and currents.
8.2 Polarization and Displacement elds
First, we consider a material that has bound charges. These are, for example, electrons
that cannot move freely through the material, but that are bound to the parent molecule.
When an external electric eld is applied, the negative electrons move with respect to
the positive nuclei, creating a small dipole moment. This dipole moment is aligned
against the external eld, because the external eld pulls the charges such that the local
eld becomes smaller. All these little dipole moments create a macroscopic eld P,
which is related to the bound charge density via
b
= P . (8.4)
47
+
+
+
+
+
+ + +
+ + +
+
+ + +
+ +
+ + +
+
+
+
+ + +
+ +
+
+ + +
+ +
+ +
Uniform polarization Nonuniform polarization
Figure 15: Polarization in a medium
This is in some sense the denition of P. We must now relate the P to the dipole moments
of the molecules.
Let V be the volume of the medium, and the dipole density in the i direction is
b
r
i
.
Then
_
V
r
i
b
dr =
3
j=1
_
V
r
i
P
j
r
j
dr =
3
j=1
_
S
r
i
P
j
d
2
r
k,=j
r
j
onS
r
j
onS
+
_
V
P
i
dr , (8.5)
where the rst equality comes from Eq. (8.4), and the second comes from partial inter-
gration. The surface term is zero when we take the surface just outside the medium. In
vector notation, we therefore have
_
V
r
b
dr =
_
V
Pdr =

V
p
n
, (8.6)
where the last term is the sum of all microscopic dipole moments in the volume V.
Outside V we have P = 0.
Now that we know what P means, we can rewrite the Maxwell equation E) =
)/
0
using Eqs. (8.3) and (8.4)as
E) =
f
+
b
0
=
0
P . (8.7)
Considering we want the macroscopic eld in terms of the free charges, we can dene
the macroscopic eld as the displacement eld D:
D =
0
E) +P (8.8)
All the material properties are determinedby P. The corresponding macroscopic Maxwell
equation is then
D =
f
(8.9)
Often the polarization eld is directly proportional to E, and we write P =
0
E,
where is the susceptibility of the medium. The displacement eld then becomes
D =
0
(1 +)E
0
r
E E . (8.10)
48
Assuming for the moment that the permeability
0
does not change, one can see imme-
diately that the wave equation yields a propagation velocity c
2
/
r
. In ordinary materi-
als, we therefore have
r
1. As we will see in the next lecture, this is directly related
to the index of refraction.
Some materials have a non-isotropic response to the electric eld, for example
P
i
=
0
3
j=1
i j
E
j
, (8.11)
where the polarization in the i direction depends in some way on the E eld in the x,
y, and z direction, while the polarization in the k ,= i direction depends on the same
external E eld in a different way. The susceptibility is therefore in general a tensor.
Clearly, such a medium has an intrinsic orientation. When the medium is transparent, a
medium like this gives rise to birefringence, where light with different polarizations (not
to be confused with P) is bent differently when it enters the medium.
Other materials have a polarization eld P that depends on the external eld E in a
nonlinear way:
P
i
=
0
_
(0)
i
+
3
j=1
(1)
i j
E
j
+
3
j,k=1
(2)
i jk
E
j
E
k
+
3
j,k,l=1
(3)
i jkl
E
j
E
k
E
l
+ . . .
_
. (8.12)
The susceptibility has different orders
(n)
i...l
corresponding to the n
th
power in the exter-
nal eld. In practice, we consider only the effects up to third order, since higher order
effects tend to be extremely weak. The
(2)
term is responsible for frequency doubling
materials, among other things, and the
(3)
term can be used to induce optical phase
shifts that are proportional to the intensity of a second eld. It is also called the Kerr
nonlinearity.
8.3 Magnetization and Magnetic induction
Now we will construct the macroscopic magnetic eld using the free and bound cur-
rents. Apart from the free currents in the material, there may be bound currents that
are conned to the molecules. In quantum mechanics these bound currents can be the
spins of the electrons and nuclei, but lets keep things classical for now.
The bound currents may be due to a change of bound charges, which can be written
as P/t. Alternatively, they may correspond to little current loops, which give rise to a
microscopic magnetic dipole moment. Under the inuence of an external eld B, these
little magnets can line up to form a macroscopic eld that we shall denote by M. We
then have the following expression for the bound current:
J
b
=
P
t
+M (8.13)
49
Uniform magnetization Nonuniform magnetization
Figure 16: Magnetization in a medium
The divergence of the bound charge is
J
b
=
P
t
+ (M) =
b
t
= 0 , (8.14)
since the bound charge is conserved.
Now lets consider the second inhomogeneous Maxwell equation
B)
0
0
E)
t
= B)
0
t
_
D
0
_
=
0
J) =
0
_
J
b
+J
f
_
. (8.15)
This leads to
B)
0
D
t
+
0
P
t
=
0
P
t
+
0
M +
0
J
f
. (8.16)
We can now dene the magnetic eld H in terms of the magnetic induction B and the
macroscopic magnetization M as
H =
1
0
B) M, (8.17)
yielding the Maxwell equation
H
D
t
= J
f
(8.18)
The homogeneous Maxwell equations remain unchanged, since here E and B are
fundamental (that is, no material properties are present in these equations). Also, for the
macroscopic case it is understood that all elds are averages, and we drop the brackets
.). The macroscopic Maxwell equations then become
B = 0 and E +
B
t
= 0 (8.19)
D =
f
and H
D
t
= J
f
. (8.20)
50
Further reading
R.H. Good, Classical Electromagnetism, Saunders College Publishing (1999): Ch. 9 & 10, pp
204-277.
J.D. Jackson, Classical Electrodynamics, Wiley (1998): Sec. 4.3-4.6 & 5.8, pp 151-165, 191-194.
W.J. Dufn, Electricity and Magnetism, McGraw-Hill (1990): Ch. 11, 12, pp 282-335.
Addison-Wesley (1993): Ch. 4 & 9, pp 97-126, 218-255.
Sec. 6.1, pp 282-293.
Exercises
1. A slab of linear dielectric material is placed inside a capacitor. The dielectric does
not ll the space inside the capacitor completely. Determine the eld lines of E, P,
and D between the capacitor plates.
2. A slab of linear dielectric material is placed inside a capacitor. The dielectric slab
and the plates of the capacitor are square with side w, and the thickness of the slab
is s. The plates are held at constant potential V.
w
x
s
F
V
(a) Write down the energy in a capacitor.
(b) How does the charge on the plates change? Does this change the energy of
the capacitor?
(c) Calculate the force on the dielectric slab.
3. A one-dimensional dielectric has a nonlinear polarization eld given by
P(t) =
(1)
E(t) +
(2)
E(t)
2
.
If the incoming wave is described by E(t) = E
0
cos t show that the dielectric
gives rise to a wave with double the frequency. What happens when the incoming
wave is a superposition of two frequencies
1
and
2
? (Hint: write the waves in
exponential notation.)
51
9 Waves in dielectric and conducting media
From our everyday experience it is clear that electromagnetic radiation (in particular
light) is able to propagate through some media, but not through others. Light has no
problem travelling through air and glass, but when it hits a conductor, most of it will
be reected. In this lecture we investigate why this is, and what are the characteristic
features of different media.
Georg Ohm (1789-1854)
When waves penetrate a medium, the molecules in the
medium respond to the electric eld and the magnetic
induction to form polarization and magnetization elds.
Lets assume that the medium is electrically neutral, so
that there are no free charges ( = 0), but there may be
electrons in a conduction band that can produce free cur-
rents J. These electrons presumably respond directly to
the electric eld, such that the free current is propotional
to E:
J =E (Ohms law). (9.1)
This is Ohms law, and the constant of proportionality is
the conductance. This law has a wide range of applicability,
but you should be aware that there are many interesting
situations where Ohms law does not hold.
There are essentially three regimes of interest for : (1)
is small, and the medium is a dielectric; (2) is large, and the medium is a conductor;
or (3) is imaginary, and the medium is a plasma. We will derive the index of refraction
for dielectrics, the skin depth for conductors, and the plasma frequency for plasmas. We
will also derive the reectivity of these three types of materials.
9.1 Waves in dielectric media
First, we rederive the wave equations from the Maxwell equations, but this time we
keep the J term. We nd the following differential equations for E and B:
2
E
2
E
t
2
=
J
t
, (9.2)
2
B
2
B
t
2
= J , (9.3)
where =
r
0
and =
r
0
. Using Ohms law, this leads to
2
E
2
E
t
2

E
t
= 0 , (9.4)
2
B
2
B
t
2

B
t
= 0 . (9.5)
52
Air Dielectric Air Dielectric
(a) (b)
E elds B elds
Figure 17: Waves incident on a dielectric. The waves propagate in the dielectric. (a) The electric
eld is mostly transmitted, and the reected wave experiences a phase shift. (b) The magnetic
eld is mostly transmitted, and the reected wave does not experience a phase shift.
We substitute again the plane wave solutions for waves in the positive z direction E =
E
0
e
i(kzt)
and B = B
0
e
i(kzt)
, and we immediately nd that
k =
_
2
+ i = k
0
r
_
1 +
i
, (9.6)
where k
0
= /c. You can see that the wave vector k is complex!
Now we can look at the three different regimes for . If the conductance is small
(/ 1), the medium is a dielectric and the imaginary root is approximately equal
to 1. The wave vector is real, so the waves propagate freely. However, the value of the
wave vector is modied by the factor
r
. It turns out that this is indeed the index of
refraction, at least for apolar materials
4
.
The reectivity of a dielectric can be calculated as well. For ordinary dielectrics
where D = E and B = H, the ratio between the electric eld and the magnetic
induction is given by the velocity of the wave in the medium:
E
B
=
c
r
. (9.7)
In air, this leads to approximately E = cB, and the (average) Poynting vector in a
medium is then
S
av
= n
0
cE
2
rms
, (9.8)
where n is the index of refraction, and E
2
rms
is the average strength to the electric eld
squared.
4
In apolar media the distribution of the electrons in the molecules is fast enough to follow the oscillat-
ing elds. In polar materials, however, the response of the medium is slow since the polar molecules have
to re-orient themselves. This is the principle behind the microwave: The water molecules try to realign
themselves with an oscillating microwave eld, and as a consequence the water gets heated.
53
In order to calculate the reectivity, we need to gure out what the boundary con-
ditions are for the different elds at the separation of the medium and the air. We will
consider the difference between the elds just outside the medium and the eld just
inside the medium. From D = we nd that
D
n
= D
1,n
D
2,n
=
f
, (9.9)
where the subscript n denotes the normal component with respect to the surface, and
the 1 and 2 denote the eld outside and inside the medium, respectively.
f
is the surface
charge. From B = 0 we nd that
B
n
= B
1,n
B
2,n
= 0 . (9.10)
Air Medium
l l
z
Figure 18: Boundary condi-
tions on a dielectric surface.
The boundary condition for E
t
is more complicated.
From E =
B we have (see Fig. 18)

_
E dl =
_
B
t
da , (9.11)
where the contour is taken as a rectangle perpendicular
to the surface and sticking half-way into the surface. The
parallel sides of the loop have length l, and the sides per-
pendicular to the surface have length z. We therefore
have
l(E
1,t
E
2,t
) = lz
B
t
= 0 , (9.12)
where the last equality follows from taking the limit z
0. We therefore have the third boundary condition
E
t
= E
1,t
E
2,t
= 0 . (9.13)
Finally, we establish the boundary condition for the transverse magnetic eld H
t
.
There may be innite bound surface current densities J
b
, so we use H = J +

D,
which ignores J
b
. The free surface current will be denoted by K
f
, and the last boundary
condition reads
H
t
= H
1,t
H
2,t
= K
f
. (9.14)
Now we can state the relations between the incident (I), reected (R), and transmitted
elds (T):
E
I
E
R
= E
T
and B
I
+ B
R
=
B
T
r
. (9.15)
Note that the reected B eld picks up a relative minus sign. This is because the Poynt-
ing vector must point in the opposite way to the incident wave, and from Eq. (9.7) we
know that the magnetic induction must gain in relative strength in the medium.
54
Now assume that the medium is such that
r
= 1. We then have
B =
nk
E , (9.16)
which leads to
E
I
+ E
R
= nE
T
. (9.17)
Together with E
T
= E
I
E
R
, this yields the ratio
E
R
E
I
=
n 1
n + 1
. (9.18)
The reectivity R of a surface can be dened as the ratio of the reective and incident
Poynting vectors. Since the Poynting vector is proportional to E
2
, the reectivity of a
dielectric medium becomes
R =
_
n 1
n + 1
_
2
. (9.19)
For glass, the index of refraction is n = 1.6, so the normal reection is about ten percent.
9.2 Waves in conducting media
In a conducting medium is very large, and we can ignore the +1 term in Eq. (9.6). The
wave vector k becomes
k = k
0
i
r
0
=
_
2
(1 + i)
_
1
+
i
_
. (9.20)
Since the wave propagates in the positive z direction, we choose the positive sign. We
dened the skin depth as
=
. (9.21)
When we substitute this value for k into the solution for the E eld we obtain
E(z, t) = E
0
exp
_
i
_
z
t
__
exp
_
_
, (9.22)
that is, there is a propagating term and an exponentially decaying term, which falls off
to e
1
when the wave has reached the skin depth.
The magnetic eld is given by B = kE/, where now k is complex. With the time
dependence made explicit, we nd that
B =
_
E
0
e
i(z/t+/4)
e
z/
. (9.23)
55
Air Metal Air Metal
(a) (b)
E elds B elds
Figure 19: Waves incident on a metal. (a) The electric eld is mostly reected with a phase shift.
(b) The magnetic eld is mostly reected, and does not have a phase shift. There is no propaga-
tion of the wave in the metal, and the exponential fall-off of the amplitudes is characterized by
the skin depth.
Note the /4 phase lag of B with respect to E. Inside the metal we have
B
0
=
_
E
0
e
i/4
(9.24)
and the Poyning vector is
S
av
=
1
2
_

2
E
2
0
e
2z/

k . (9.25)
This shows that the power falls off twice as fast as the elds.
The reectivity of metals is high, and consequently the transmittance T is low (R +
T = 1 from energy conservation). Using again B = kE/, and the boundary conditions
of the previous section, we once more use
E
I
E
R
= E
T
and B
I
+ B
R
=
B
T
r
, (9.26)
and nd (n is a large complex number)
E
T
E
I
=
2
1 + n

2
n
, (9.27)
where n is now a complex index of refraction
n =
i
r
0
. (9.28)
56
In addition, we can solve Eq. (9.26) for the magnetic induction, which yields
B
T
B
I
=
2
1 + 1/n
2 . (9.29)
The ratio of the Poynting vectors for the transmitted and incident waves is therefore
S
T
S
I
= Re
4
n
. (9.30)
This leads to the transmittance
T =
_
8
0
. (9.31)
Usually, the transmitted waves will be absorbed completely unless the metal is suf-
ciently thin. When such as in a superconductor, the transmittance drops to zero
and everything is reected. That is why high quality cavities are made with supercon-
ducting mirrors.
9.3 Waves in plasmas
Finally, consider a plasma in which the electrons are disassociated from the atomic nu-
clei. When an electric eld is applied the electrons accelerate. The current density J
therefore lags the electric eld, and the phase happens to be e
i/2
= i. The nuclei are
much heavier than the electrons, so they do not accelerate as fast and make a negligable
contribution to the conductivity. We can nd the velocity v of the electrons by integrat-
ing F = ma = eE, where all vector quantities are in the z direction. Using the electron
density Ne, the current density is
J = Nev = Ne
ieE
m
, (9.32)
and the conductivity is
=
J
E
=
ie
2
N
m
. (9.33)
We substitute this into our expression for the complex wave vector k
k = k
0
1
e
2
N
2
m
0
= k
0
_
1
_
_
2
, (9.34)
where we dened the plasma frequency
p
=
e
2
N
m
0
. (9.35)
57
Clearly, when the frequency of the wave is larger than the plasma frequency, k is real
and the plasma acts as a dielectric. When < )p the plasma acts as a conductor.
The ionosphere of the Earth is a plasma, and consequently for low enough frequen-
cies (long wavelengths) it acts as a mirror. Since the earth is also a conductor, low
frequency radio waves (AM) can bounce between the ionosphere and the Earth to reach
much further than high frequency (FM) radiowaves. The plasma frequency of the iono-
sphere is about 3 MHz.
Further reading
390-412.
J.D. Jackson, Classical Electrodynamics, Wiley (2003): Sec. 7.1-7.6, pp 295-319.
H.J. Pain, The Physics of Vibrations and Waves, Wiley (1983): Ch. 7, pp 187-218.
Exercises
1. Calculate the skin depth of copper for 60 Hz, 10 kHz, and 10 MHz. The conduc-
tivity of copper is 5.8 10
7
1
m.
2. Calculate the speed of a 1 MHz wave in copper and in glass.
58
10 Relativistic formulation of electrodynamics
Albert Einstein
(1879-1955)
One of the remarkable things about the Maxwell equa-
tions is that they give rise to a correct description of all
sorts of radiation, most notably light. In particular, it pre-
dicts the correct velocity of light in vacuum via the rela-
tion c
2
= (
0
0
)
1
, as we have seen in Lecture 5. Ex-
perimentally it was found by Michelson and Morley in
1887 that the velocity of light is independent of the veloc-
ity of the source. This caused Lorentz quite a headache,
which ultimately resulted in the Lorentz transformations
for moving bodies: Objects that move with respect to an
inertial observer are seen to experience a length contrac-
tion in the direction of their motion, and their clocks seem
to run slower.
There are similar rules for the electromagnetic eld,
meaning that what one inertial observer calls the E eld,
a second observer may describe in terms of both E and B
elds. We therefore want to know two things: (1) How
do the elds transform from one coordinate system to
another, and (2) What do Maxwells equations look like
when written in covariant form, that is, independent of the coordinate system? We will
rst revise the basics of special relativity and Minkowski space, and then we will con-
struct Maxwells equations from the vector and scalar potentials, and the elds. This
leads naturally to the transformation laws for the E and B elds. We end the lecture
with a look at some invariant properties. This lecture will rely heavily on the transfor-
mation properties of vectors and tensors, and everything will be in index notation.
10.1 Four-vectors and transformations in Minkowski space
Remember that covariant descriptions take place in four-dimensional Minkowski space.
There is the position four-vector x
, where = 0, 1, 2, 3, r = (x
1
, x
2
, x
3
) or r = x
i +
y
j + z

k, and x
0
= ct, and the momentum four-vector p
with p = (p
1
, p
2
, p
3
) and p
0
=
U/c. Minkowski space is a strange place in that it has a non-trivial metric: Ordinarily, if
you want to nd the length of a (short) interval, you add all the components squared,
according to Pythagoras:
ds
2
= dx
2
+ dy
2
+ dz
2
. (10.1)
However, in relativity the length changes in different reference frames, and we need
to include the change in time coordinate as well. It is tempting to just add c
2
dt
2
to
Eq. (10.1), but that would be wrong! The correct distance between two events is
ds
2
= c
2
dt
2
+ dx
2
+ dy
2
+ dz
2
, (10.2)
59
with a minus sign in front of c
2
dt
2
! This is the denition of the distance in Minkowski
space, and we can rewrite this as a vector equation (and also using Einsteins summation
convention):
ds
2
= (ct, x, y, z)
_
_
_
_
1 0 0 0
0 1 0 0
0 0 1 0
0 0 0 1
_
_
_
_
_
_
_
_
ct
x
y
z
_
_
_
_
x
= x
= x
, (10.3)
where the four-vector x
can be written as (ct, x, y, z). Notice that the matrix g
has a
1 on the diagonal, which is responsible for the minus sign in the metric. You also
see that there is a difference between upper and lower indices. The four-vectors with
upper indices are called contravariant, and the four-vectors with lower indices are called
covariant. You can verify from Eq. (10.3) that the covariant four-vector has a relative
minus sign in the rst component:
x
= (ct, x, y, z) . (10.4)
We are cheating a bit here, because the x
on the left is a component, while the quantity

on the right is a proper vector. However, this expression is to emphasize the effect of the
vertical position of the index; it is not a proper equation. As you can see from Eq. (10.3),
the metric raises or lowers the index, and g
= g
in the case of special relativity.

We now amend the Einstein summation convention: the sum is implied over two re-
peating indices, one of which is upper, and the other is lower (convince yourself that this
distinction is important). Carrying out the sum is called contraction. We can construct
invariant quantities (scalars) by contracting contravariant four-vectors with covariant
four-vectors. In general we write the components of a four-vector a
as (a
0
, a
1
, a
2
, a
3
),
and we have:
a
= a
=

,
a
= a
0
b
0
+
3
j=1
a
j
b
j
= a
0
b
0
+a b , (10.5)
where greek indices sum over all four components, while roman indices sum only over
the spatial part
5
.
As an example, lets look at the four-momentum of a particle: p = (U/c, p
1
, p
2
, p
3
),
and p
2
= p
can be written as p
2
= U
2
/c
2
+[p[
2
. Since this is true for all inertial
frames, it is true for the frame where the particle is at rest, so [p[
2
= 0. Using U = mc
2
with m the rest mass of the particle, we nd p
2
= m
2
c
2
(yes, in Minkowski space
5
You should be aware of the fact that the metric g
can also be written as a diagonal matrix with

three 1 entries for the spatial part and one +1 for the temporal part. This does not lead to observable
differences in the theory, so it is a convention. Unfortunately, both conventions are used regularly, so
make sure you know what the metric is before you copy a relativistic formula from a book!
60
a square can be negative without complex numbers!), and the energy of a particle is
therefore
U =
_
m
2
c
4
+[p[
2
c
2
. (10.6)
This is one of the most important formulas in physics.
Hendrik Antoon Lorentz
(1853-1928)
Any four-vector a
is transformed into a
due to a
Lorentz transformation
via
a
. (10.7)
Note that the primed frame of reference is indicated by a
primed index
. The Lorentz transformation involves a

contraction over the old (unprimed) coordinates in order
to remove them from the equation. In general, a tensor
transforms as
T
. (10.8)
Every component is transformed with a separate Lorentz
transformation
. The actual form of the Lorentz trans-

formation, for example for a boost in the z direction, is
=
_
_
_
_
0 0
0 1 0 0
0 0 1 0
0 0
_
_
_
_
, (10.9)
with = 1/
_
1
2
and = v/c. In component notation the Lorentz transformation
x
becomes
x
0
= x
0
x
3
=
1
_
1 v
2
/c
2
_
ct
vz
c
_
,
x
1
= x
1
= x ,
x
2
= x
2
= y ,
x
3
= x
0
+x
3
=
1
_
1 v
2
/c
2
(z vt) . (10.10)
Note that the Lorentz transformation is a proper coordinate transformation:
=
x
. (10.11)
Using these rules it is straghtforward (but somewhat lengthy) to nd the transformation
rule for any tensor.
61
10.2 Covariant Maxwell equations
Just like the position and momentum four-vectors, we would like to construct elec-
tric and magnetic four-vectors. Unfortunately, it is not that simple: There are six eld
components (three for E and three for B), and we cannot force them into four-vectors.
On the other hand, we can combine the vector potential and the scalar potential into a
fourvector A
:
A
(x) =
_
(x)
c
, A(x)
_
, (10.12)
where x is again the position four-vector. The second four-vector that we can construct
straight away is the current density j
:
j
(x) =
_
c(x), J(x)
_
. (10.13)
Since the Maxwell equations are differential equations, we also need to construct a
four-vector out of the differential operators. Again, they come in two variations, covari-
ant
and contravariant
=

x
=
_
1
c
t
,
_
and
=

x
=
_
1
c
t
,
_
. (10.14)
These operators transform just like ordinary vectors (but note the vertical position of
the indices). We can construct an invariant with these operators if we contract them:
=
2
x
+
2
y
+
2
z
1
c
2
2
t
= . (10.15)
This is the dAlembertian, which we rst encountered in lecture 3.
The continuity equation (conservation of charge) then becomes particularly com-
pact:
= 0 . (10.16)
Compare this with Eq. (2.19); the two equations say exactly the same thing! The Maxwell
equations in Eqs. (3.9) can be written as
=
0
j
(10.17)
Often people say that this encompasses all of electrodynamics. However, you should
remember that we also need to know how charges respond to the elds, that is, we
need to know the Lorentz force. Since this force depends explicitly on the velocity of
the particle, it is not in covariant form. We will consider the covariant Lorentz force later.
We also see explicitly that the Maxwell equations are already relativistically correct: We
did not change anything in Eqs. (3.9), we just rewrote everything. The compactness of
Eq. (10.17) indicates that Lorentz invariance is a symmetry of electrodynamics (The more
symmetric the object, the more compact we can make its description: For example,
62
compare the description of a perfect sphere with the description of a sponge). Lorentz
invariance is an important symmetry in quantum eld theory and the standard model.
Maxwell thought that the electric elds were carried by some substance that per-
vades all space, called the ther. Such a substance denes a natural reference frame,
which would break Lorentz invariance. It was Einstein who disposed of the ther and
made all inertial reference frames equivalent. Thus he uncovered that electrodynamics
was a relativistic theory all along. In fact, his famous paper of 1905 on special relativity
was titled On the electrodynamics of moving bodies.
In addition to the Maxwell equations in Eq. (10.17), we need to specify a particular
gauge for the four-vector potential A
if we actually want to calculate anything. The

relativistically invariant gauge is the Lorenz gauge
= 0 . (10.18)
Nowthat we have constructed the four-vector potential for the electromagnetic eld,
we can ask what the E and B elds look like in a relativistic setting. We can use the
relations
E =
A
t
and B = A. (10.19)
If we write out a few components we get, for example,
E
x
=
x

A
x
t
= c
_
1
A
0
0
A
1
_
,
B
x
=
A
z
y

A
y
z
= (
2
A
3
3
A
2
) . (10.20)
Obviously we can only t all six components of the electric and magnetic eld into a rel-
ativistic object if we have something with two indices. We call this the (anti-symmetric)
eld-strength tensor F:
F
=
_
_
_
_
0 E
x
/c E
y
/c E
z
/c
E
x
/c 0 B
z
B
y
E
y
/c B
z
0 B
x
E
z
/c B
y
B
x
0
_
_
_
_
. (10.21)
The relativistic transformation of the elds then becomes
F
. (10.22)
For a boost v in the z direction given by Eq. (10.9) we nd
E
x
=
_
E
x
+ vB
y
_
, E
y
=
_
E
y
vB
x
_
, E
z
= E
z
(10.23)
B
x
=
_
B
x
v
c
2
E
y
_
, B
y
=
_
B
y
+
v
c
2
E
x
_
, B
z
= B
z
. (10.24)
63
+ +
(a) (b)
v
Figure 20: The eld lines of an electric charge. (a) The charge is stationary. (b) The charge is
moving with velocity v.
In terms of the parallel and perpendicular components of the elds with respect to the
direction of motion this becomes
E
[[
= E
[[
E
= E
+ vB
B
[[
= B
[[
B
= B
vE
c
2
. (10.25)
A physical situation with only a static charge distribution and no currents obviously
does not involve B elds. The magnetic elds arise when the charge distribution is
viewed by a moving observer. In particular, the electrostatic force F = qE aquires a term
proportional to vB, so that F
= qE +qvB (verify this!). This is of course the familiar

Lorentz force. In other words, when an observer sees a moving electric charge, the E
eld is not spherically symmetric due to Lorentz contraction in the direction of motion.
As a result, there is a component of the electric force that acts only on moving charges.
This is the magnetic eld. This argument shows that you can think of the magnetic eld
as just the relativistic part of the electric eld of moving charges. Whenever there are
B elds, there must be moving charges, whether it be macroscopic currents or spinning
charges in atoms. The eld lines of a stationary and moving charge are shown in Fig. 20.
Finally the Lorentz force must be formulated in a covariant way. The force experi-
enced by a particle with charge q, as viewed from the lab frame, is proportional to the
velocity, so we need to dene a four-velocity u. We can use the relation
p
mu
= m(c, v) (10.26)
such that u
= c
2
(verify this!). Rather than deriving the relativistic Lorentz force,
we will state it here and conrm that it behaves the way it should. Suppose that is the
proper time of a particle, as recorded by a co-moving clock. The force will cause a mo-
mentum transfer dp
/d. This must be proportional to E + vB. The cross product is

built in the eld strength tensor F
, so we guess that the covariant form of the Lorentz

force is
m
d
2
x
d
2
=
dp
d
= qu
= qF
dx
d
. (10.27)
64
These are the equations of motion for a relativistic particle in an electromagnetic eld.
Lets see what the different components of f look like:
f
x
=
dp
1
d
= qF
1
u
= q
_
c
E
x
c
+ v
x
0 + v
y
B
z
v
z
B
y
_
f
y
=
dp
2
d
= qF
2
u
= q
_
c
E
y
c
v
x
B
z
+ v
y
0 + v
z
B
x
_
f
z
=
dp
3
d
= qF
3
u
= q
_
c
E
x
c
+ v
x
B
y
v
y
B
x
v
z
0
_
(10.28)
These are indeed the three components of the Lorentz force, with an extra factor . This
is correct, because the term dp
/d also has an implicit factor in the relativistic mass.

Remains the question of what is f
t
:
c f
t
= c
dp
0
d
= qcF
0
u
= q
_
v
x
E
x
+ v
y
E
y
+ v
z
E
z
_
= q v E. (10.29)
This is the relativistic work done on the particle by the elds. So you see that it all works
out.
10.3 Invariant quantities
When the elds and potentials change when viewed in different inertial frames, it is
important to know what the invariant quantities of a theory are. These quantities are
the same in all reference frames, so you can evaluate them in whichever frame makes
the calculation easiest. There are many invariants, most notably the magnitude of four-
vectors a
2
= a
. You can use the differential operator as well in the construction of

invariants (see Eq. (10.13) for an important example). A quantity is invariant when it
does not depend on the coordinates, so in index notation the quantity has no indices.
The quantity
1
4
F
is invariant (why?), and can be evaluated as

1
4
F
=
1
4
g
=
B
2
2

E
2
2c
2
. (10.30)
This is the Lagrangian (density) of the electromagnetic eld, which plays a pivotal role
in particle physics.
We can also construct the dual of the eld-strength tensor
T
=
1
2
=
_
_
_
_
0 B
x
B
y
B
z
B
x
0 E
z
/c E
y
/c
B
y
E
z
/c 0 E
x
/c
B
z
E
y
/c E
x
/c 0
_
_
_
_
. (10.31)
which is a pseudo-tensor. Here we used the Levi-Civita (pseudo) tensor
, which
returns 1 if the indices make an even permutation, 1 if they make an odd permutation,
65
PHY331 PART I: ADVANCED ELECTRODYNAMICS LECTURE
and 0 otherwise. The invariant product F
is
F
= 4E B/c . (10.32)
Finally, the microscopic Maxwell equations can be written in terms of the elds as
=
0
j
and
= 0 (10.33)
Further reading
430-490.
J.D. Jackson, Classical Electrodynamics, Wiley (1998): Ch. 11, pp 514-566.
R.P. Feynman, Lectures on Physics, volume II, Addison-Wesley (1964): Ch. 25, 26 & 31.
Ch. 1, pp 29-72.
Exercises
1. Calculate the determinant of the Lorentz transformation
in Eq. (10.9)
2. Calculate the expression for the energy of a relativistic particle as in Eq. (10.6), but
now with the metric
g
=
_
_
_
_
1 0 0 0
0 1 0 0
0 0 1 0
0 0 0 1
_
_
_
_
.
3. Show that g
is a proper transformation that can be written as

g
=
x
.
4. Prove that E B is Lorentz invariant.
5. Show that the phase of an electromagnetic wave is a Lorentz invariant quantity.
What exactly is invariant here?
6. Calculate F
, F
, T
, and verify Eq. (10.32).

66
PHY331 PART I: ADVANCED ELECTRODYNAMICS LECTURE A
A Special coordinates and vector identities
A.1 Coordinate systems
Often it is much more convenient to use cylindrical or spherical coordinates, rather
than cartesian coordinates. The differential operators change accordingly. For cartesian
coordinates:
AB = (A
y
B
z
A
z
B
y
) x + (A
z
B
x
A
x
B
z
) y + (A
x
B
y
A
y
B
x
) z (A.1)
f =
f
x
x +
f
y
y +
f
z
z (A.2)
A =
A
x
x
+
A
y
y
+
A
z
z
(A.3)
A =
_
A
z
y

A
y
z
_
x +
_
A
x
z

A
z
x
_
y +
_
A
y
x

A
x
y
_
z (A.4)
Cylindrical coordinates:
f =
f
+
1
+
f
z
z (A.5)
A =
1
(A
+
1
+
A
z
z
(A.6)
A =
_
1
A
z
z
_
+
_
A
z

A
z
_

+
1
_
(A
_
z (A.7)
2
f =
1
_
+
1
2
f
2
+

2
f
z
2
(A.8)
x
y
z
Figure 21: Cartesian coordinates (x, y, z).
67
x
y
z

Figure 22: Cylindrical coordinates (, , z).
Spherical coordinates:
f =
f
r
r +
1
r
f
+
1
r sin
f
(A.9)
A =
1
r
2
(r
2
A
r
)
r
+
1
r sin
(A
sin)
+
1
r sin
A
(A.10)
A =
1
r sin
_
(A
sin)
_
r (A.11)
+
1
r
_
1
sin
A
r
(rA
)
r
_

(A.12)
+
1
r
_
(rA
)
r

A
r
_

(A.13)
2
f =
1
r
2
r
_
r
2
f
r
_
+
1
r
2
sin
_
sin
f
_
+
1
r
2
sin
2
2
f
2
(A.14)
x
y
z
Figure 23: Spherical coordinates (r, , ).

68
A.2 Integral theorems and vector identities
Having dened the divergence and the curl of vector elds, we can state the important
integral theorems (without proof):
_
V
AdV =
_
S
A dS (Gauss theorem) (A.15)
_
S
(A) dS =
_
C
A dl (Stokes theorem) (A.16)
_
b
a
f dl = f (b) f (a) (A.17)
_
V
_
f
2
g g
2
f
_
dV =
_
S
( f g gf ) dS (Greens theorem) (A.18)
and some important vector identities, such as relations between 3-vectors:
A (BC) = B (CA) = C (AB) (A.19)
A(BC) = (A C)B (A B)C (A.20)
(AB) (CD) = (A C)(B D) (A D)(B C) (A.21)
rst derivatives:
(A) = ( A) +A () (A.22)
(A) = (A) + ()A (A.23)
(A B) = A(B) +B(A) + (A )B + (B )A (A.24)
(AB) = B (A) A (B) (A.25)
(AB) = (B )A(A )B +A( B) B( A) (A.26)
and second derivatives:
(A) = 0 (A.27)
() = 0 (A.28)
(A) = ( A)
2
A (A.29)
A.3 Levi-Civita tensor
Finally, we may sometimes use the so-called Levi-Civita symbol
i jk
, which returns a
value of 1 when (i jk) is an even permutation of the indices, 1 if (i jk) is an odd per-
mutation, and 0 otherwise (for example when two indices are the same). It obeys the
identity
3
k=1
i jk
klm
=
3
k=1
i jk
lmk
=
il
jm
im
jl
, (A.30)
where
i j
is of course the Kronecker delta, which returns 1 if i = j and 0 if i ,= j.
69
PHY331 PART I: ADVANCED ELECTRODYNAMICS LECTURE B
B Unit systems in electrodynamics
Carl Friedrich Gauss
(1777-1855)
Traditionally, electrodynamics is plagued by the use of
many different sets of units. In these lecture notes we
use SI units (Syst` eme International) exclusively. However,
many people also use Gaussian units, so it is worthwhile
to spend some time exploring the difference betwen these
units. In particular, you should be able to read the set of
units used off the Maxwell equations. The rule of thumb
is: if they involve factors of 4, then they most likely use
Gaussian units.
Carl Friedrich Gauss was a German mathematician
who in the 1830s did (among very many other things)
worldwide measurements of the Earths magnetic eld.
Not only are the Gaussian units named after him, the unit
of magnetic induction in Gaussian units is also called the
Gauss (G). In SI units the unit of magnetic induction is
the Tesla (T), after the Serbian inventor Nikola Tesla (1853-
1943).
There is a certain arbitrariness in Maxwells equations,
the Lorentz force, and the continuity equation, in the sense that we can add constants
k
1
, k
2
, . . . to the equations that affect only the units, and not the physics:
k
1
J +

t
= 0 and F = k
2
E + k
3
JB (B.1)
E = k
4
and B = 0 (B.2)
E + k
5
B
t
= 0 and B = k
6
J + k
7
E
t
(B.3)
Not all the constants k
1
to k
7
are independent. Some manipulation of the above equa-
tions will give the identity
k
6
J + k
4
k
7
t
= 0 (B.4)
so we have the restriction that k
6
= k
1
k
4
k
7
. Similarly, by rederiving the wave equations
we nd that k
5
k
7
= c
2
.
More confusion arises when we consider macroscopic media. Gauss law in terms of
the polarization eld becomes
(E + k
4
P) = k
4
free
. (B.5)
which denes the displacement eld
D = k
8
(E + k
4
P) . (B.6)
70
PHY331 PART I: ADVANCED ELECTRODYNAMICS LECTURE B
Similarly, with a bound current density J
bound
= M +

P the Maxwell-Amp` ere law
becomes
(B k
6
M) = k
6
J
free
+ k
7
t
(E + k
4
P) . (B.7)
The magnetic eld is then dened as
H = k
9
(B k
6
M) . (B.8)
The different constants k
1
. . . k
9
in both SI and Gaussian units are given by
k
1
k
2
k
3
k
4
k
5
k
6
k
7
k
8
k
9
SI units 1 1 1
1
0
1
0
c
2
0

1
0
Gaussian units 1 1 c
1
4 c
1
4c
1
c
1
1 1
As you know, in the SI system the fundamental units are meter (m), kilogram (kg),
second (s), and amp` ere (A). The unit of force is the Newton (N), which is of course
kg m/s
2
. Just like the second and the kilogram, the amp` ere is dened operationally.
One amp` ere is the amount of current needed to create a force of 2 10
7
N per meter
length between two innitely long parallel wires separated by a distance of one meter.
In the Gaussian system, the fundamental units of mass, length, and time are the
gram (g), centimeter (cm), and again the second (s). Instead of the amp` ere, the other
fundamental unit is the statcoulomb (statC) for the unit of charge. The statvolt (statV)
is the unit of potential, and obeys 1 statV = 1g cm/statC s
2
.
charge current E B
0

0
SI C = As A N/C = V/m T = N/Am Tm/A C
2
/Nm
2
Gauss statC statA = statC/s statV/cm G(statV/cm)
In Gaussian units the permittivity and the permeability are dimensionless.
Further reading
W.J. Dufn, Electricity and Magnetism, McGraw-Hill (1990): App. C, pp 406-413.
Addison-Wesley (1993): App. III, pp 599-602.
Appendix, pp 581-589.
Book Program, Westview Press (1998): App. A, pp 555-559.
71

Phy 331 Advanced Elsctrodynamics PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Phy 331 Advanced Elsctrodynamics PDF

Uploaded by

Copyright:

Available Formats

PHY331: Advanced Electrodynamics & Magnetism

), that is, the same function f . For example,

Figure 3: Rotation of coordinate frame.

). From Figs. 5b and 5c you see that the magnetic

. The second term

[, and the curl of a gradient

[. Using the chain rule we can rewrite this as

(r, t) = A(r, t) +(r, t) . (3.7)

[ given by Eq. (6.6), which leads to

B we have (see Fig. 18)

can be written as (ct, x, y, z). Notice that the matrix g

on the left is a component, while the quantity

in the case of special relativity.

can also be written as a diagonal matrix with

. The Lorentz transformation involves a

. The actual form of the Lorentz trans-

if we actually want to calculate anything. The

= qE +qvB (verify this!). This is of course the familiar

/d. This must be proportional to E + vB. The cross product is

, so we guess that the covariant form of the Lorentz

/d also has an implicit factor in the relativistic mass.

. You can use the differential operator as well in the construction of

is invariant (why?), and can be evaluated as

is a proper transformation that can be written as

, and verify Eq. (10.32).

Figure 23: Spherical coordinates (r, , ).

You might also like