Electrodynamics

Babis Anastasiou
Institute for Theoretical Physics, ETH Zurich,
8093 Zurich, Switzerland
E-mail: babis@phys.ethz.ch
May 21, 2021

Abstract
The subject of the course is classical electrodynamics. The follow-
ing topics are discussed:
• Electrostatics
• Magnetostatics
• Laws of Electrodynamics
• Electromagnetic waves
• Retarded Potentials
• Special Relativity
• Radiation
• Electrodynamics in Matter

1
Contents
1 Electromagnetic force 5

2 Electrostatics 8
2.1 Coulomb’s law . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.1.1 Mathematical interlude: the delta-function . . . . . . . 9
2.2 Gauss’ law from Coulomb’s law . . . . . . . . . . . . . . . . . 10
2.2.1 Integral form of Gauss’ law . . . . . . . . . . . . . . . 11
2.2.2 Mathematical interlude: The Laplacian of inverse dis-
tance . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3 Scalar potential . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4 Potential energy of a charge distribution . . . . . . . . . . . . 15
2.4.1 Potential energy of a continuous charge distribution . . 16
2.4.2 Self-energy . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.5 Charged conductors . . . . . . . . . . . . . . . . . . . . . . . . 17

3 Boundary condition problems in electrostatics 22

3.1 Dirichlet and Neumann boundary conditions . . . . . . . . . . 22
3.2 Green’s functions . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.2.1 Dirichlet boundary conditions . . . . . . . . . . . . . . 25
3.2.2 Neumann boundary conditions . . . . . . . . . . . . . . 25
3.3 Explicit solutions of boundary condition problems . . . . . . . 26
3.3.1 Example 1: A conductor filling half of space . . . . . . 26
3.3.2 Example 2: Reverse engineering and the method of
images (mirror charges) . . . . . . . . . . . . . . . . . 29
3.4 Green’s functions from Laplacian eigenfunctions . . . . . . . . 32
3.4.1 Example: all (infinite) space . . . . . . . . . . . . . . . 34
3.4.2 Example: Inside an orthogonal parallelepiped . . . . . 35
3.5 Laplace operator and spherical symmetry . . . . . . . . . . . . 38
3.5.1 Radial differential equation . . . . . . . . . . . . . . . 39
3.5.2 Angular differential equation . . . . . . . . . . . . . . . 39
3.5.3 Expansion of inverse distance in Legendre polynomials 45
3.6 Multipole expansion . . . . . . . . . . . . . . . . . . . . . . . 46

4 Magnetic field 50
4.1 Currents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.2 Magnetic field of steady currents . . . . . . . . . . . . . . . . 52

2
 4.3 Applications of Ampere’s law . . . . . . . . . . . . . . . . . . 53
4.3.1 Long straight wire . . . . . . . . . . . . . . . . . . . . 53
4.3.2 Solenoid . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.4 Vector potential . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.5 Wires of a small thickness . . . . . . . . . . . . . . . . . . . . 57
4.6 Magnetic dipole . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.6.1 Force on a “magnetic dipole” inside a magnetic field . . 63
4.7 Relativity of the electric and magnetic field . . . . . . . . . . . 64

5 Time varying electromagnetic fields 68

5.1 Charge conservation . . . . . . . . . . . . . . . . . . . . . . . 68
5.2 Vector and scalar potential . . . . . . . . . . . . . . . . . . . . 68
5.3 Gauge invariance . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.4 Electromagnetic waves in empty space . . . . . . . . . . . . . 71
5.4.1 Spherical waves . . . . . . . . . . . . . . . . . . . . . . 74

6 General solutions of Maxwell equations with sources 75

6.1 Green’s functions . . . . . . . . . . . . . . . . . . . . . . . . . 75
6.2 Fourier transformations . . . . . . . . . . . . . . . . . . . . . . 78
6.3 Fourier transformations and Green’s functions . . . . . . . . . 79
6.4 Potential of a moving charge with a constant velocity . . . . . 83

7 Special Relativity 85
7.1 Proper time . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
7.2 Subgroups of Lorentz transformations . . . . . . . . . . . . . . 89
7.3 Time dilation . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
7.4 Doppler effect . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
7.5 Particle dynamics . . . . . . . . . . . . . . . . . . . . . . . . . 92
7.6 Energy and momentum . . . . . . . . . . . . . . . . . . . . . . 97
7.7 The inverse of a Lorentz transformation . . . . . . . . . . . . . 98
7.8 Vectors and Tensors . . . . . . . . . . . . . . . . . . . . . . . 99
7.9 Currents and densities . . . . . . . . . . . . . . . . . . . . . . 100
7.10 Energy-Momentum tensor . . . . . . . . . . . . . . . . . . . . 102

8 Relativistic formulation of Electrodynamics 107

8.1 Energy-Momentum Tensor in the presence of an electromag-
netic field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
8.2 Lagrangian formalism of Electrodynamics∗ . . . . . . . . . . . 113

3
 8.3 Maxwell equations as Euler-Lagrange equations∗ . . . . . . . . 114
8.3.1 Gauge fixing . . . . . . . . . . . . . . . . . . . . . . . . 115

9 Radiation from moving charges 116

9.1 The vector potential from a moving charge . . . . . . . . . . . 116
9.1.1 Potential from a moving charge with a constant velocity118
9.2 The electromagnetic field tensor from a moving charge . . . . 120
9.2.1 The electric and magnetic field of a moving charge . . . 122
9.3 Radiation from an accelerated charge in its rest frame: Lar-
mor’s formula . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
9.4 Radiation from an accelerated charge with a relativistic velocity125
9.4.1 Circular motion . . . . . . . . . . . . . . . . . . . . . . 126
9.4.2 Linear accelerators . . . . . . . . . . . . . . . . . . . . 126
9.5 Angular distribution of radiation from a linearly accelerated
relativistic charge . . . . . . . . . . . . . . . . . . . . . . . . . 127

10 Scattering 129
10.1 Thomson scattering . . . . . . . . . . . . . . . . . . . . . . . . 129
10.2 Rayleigh scattering . . . . . . . . . . . . . . . . . . . . . . . . 131

11 Electrodynamics in a medium 133

11.1 Average of the atomic charge density . . . . . . . . . . . . . . 133
11.2 Average of atomic current density . . . . . . . . . . . . . . . . 135
11.3 Maxwell equations in a medium . . . . . . . . . . . . . . . . . 137
11.3.1 The D~ and H ~ field . . . . . . . . . . . . . . . . . . . . 138
11.4 Maxwell equations inside a dielectric material . . . . . . . . . 138
11.5 Waves in a dielectric medium . . . . . . . . . . . . . . . . . . 139
11.6 A model for the dielectric susceptibility χ . . . . . . . . . . . 140
11.7 The complex index of refraction . . . . . . . . . . . . . . . . . 142
11.8 Waves in metals . . . . . . . . . . . . . . . . . . . . . . . . . . 142
11.8.1 Low frequency approximation . . . . . . . . . . . . . . 143
11.8.2 High frequency approximation . . . . . . . . . . . . . . 144
11.9 Reflection and refraction . . . . . . . . . . . . . . . . . . . . . 144
11.9.1 Snell’s law . . . . . . . . . . . . . . . . . . . . . . . . . 146
11.9.2 Polarisation . . . . . . . . . . . . . . . . . . . . . . . . 149
11.10 Cherenkov radiation . . . . . . . . . . . . . . . . . . . . . . . 150

4
1 Electromagnetic force
The electromagnetic force has a simple qualitative description. Electrically
charged matter is divided into two types of positive and negative charges.
Charges of the same type repel each other while charges of different types at-
tract each other. The force among two charges grows when they are brought
closer together.
Matter contains enormous numbers of positive (protons) and negative
(electrons) charges. The forces exerted by these charges are enormous. How-
ever, protons and electrons are found to be very close together in equal
numbers and these forces cancel each other almost entirely.
Exercise: Calculate the force exerted to a single electron from only the pro-
tons in a gram of sugar placed in a distance of one meter. How does this
force compare to the weight of a typical human?
While positive and negative charges are perfectly balanced at a macroscopic
level, charge imbalances are evident at an atomic level. The electric force is
responsible for protons and electrons binding into atoms as well as molecular
chemical bonds.
One realizes easily that the classical qualitative description of the electric
force as an attraction or repulsion of electric charges does not work when
applied to atoms. Protons and electrons should be collapsing on top of each
other due to their attraction. While in orbit, electrons are getting acceler-
ated and according to classical electrodynamics they should lose energy in
the form of radiation, thus falling to lower and lower orbits. To explain the
stability of atoms, we shall need to combine the laws of classical electrody-
namics with the laws of quantum mechanics. The latter, impose a minimum
energy for electrons orbiting around protons which cannot be reduced any
further. In this course, we will not apply electrodynamics to atomic systems.
Nevertheless, there are many interesting macroscopic phenomena which we
can understand within the classical theory, without using quantum laws.
The force acting on an electric charge q at a position ~x depends on the
relative position and relative motion of all other electric charges. We can
sum up the effects of all other charges into two vectors:
~ x), the electric field and
• E(~
~ x), the magnetic field.
• B(~

5
The force is then given by:
 
F~ = q E
~ + ~v × B
~ , (1)

where ~v is the velocity of the charge q. The electric and magnetic fields are
determined from the equations of Maxwell:

~ ·Eρ
~ =
∇ , (2)
0
~
~ ×E
∇ ~ = − ∂B , (3)
∂t
~ ·B
∇ ~ = 0, (4)
~j ~
1 ∂E
~ ×B
∇ ~ = + (5)
c2 0 c2 ∂t.

where ρ(~x) is the electric charge density and ~j is the electric current density.
0 is a constant, the so-called vacuum permittivity, and has the value
A·s
0 = 8.854187817 . . . 10−12 . (6)
Volt · m
c is the speed of light
m
c = 2.99792458 . . . 108 . (7)
s
We remind the definition of the differential operator:
!
~ ≡ ∂ ∂ ∂
∇ , , (8)
∂x1 ∂x2 ∂x3
The inner product of two vectors is:
3
~·B
~ =
X
A Ai Bi (9)
i=1

while the components of the outer product of two vectors are

  3
~×B
~
X
A ≡ ijk Aj Bk (10)
i
j,k=1

In the above, ijk is the fully antisymmetric tensor with 123 = +1.

6
 The solutions of Maxwell equations describe all macroscopic electromag-
netic phenomena. First, we will explore these solutions in mathematically
simple situations, assuming steady currents, fixed charge distributions and
static electric and magnetic fields. In such setups, the electric and magnetic
fields decouple from each other in Maxwell equations. After gaining expe-
rience with solving the equations in electrostatics and magnetostatics, we
will examine phenomena such as induction and electromagnetic radiation,
where the fields vary with time. Finally, we will examine the structure of
the Maxwell equations from a pure aesthetic or better said theoretic point
of view. We will compare the equations in different inertial frames of refer-
ence and will discover how they transform under relativistic transformations.
We will also explore another symmetry of Maxwell equations, the so-called
gauge symmetry. Strikingly, this symmetry governs the laws of physics at
the smallest distances (10−16 m) that we have been able to explore experi-
mentally.

7
2 Electrostatics
Consider fixed charge distributions and static electric fields. In this situation,
the first two of Maxwell equations take the form:

~ ·E ρ
~ =
∇ Gauss Law (11)
0
~ ×E
∇ ~ = 0. (12)

Notice that these equations imply the superposition principle. If two charge
distributions ρ1 and ρ2 the electric fields yield electric fields E~1 and E~2 respec-
tively, a charge distribution ρ = ρ1 +ρ2 produces an electric field E ~ = E~1 + E~2 .

2.1 Coulomb’s law

These equations are equivalent to the very familiar Coulomb’s law. It
states that for two charges q and q1 at positions ~x and y~1 respectively, the
force acting on the charge q is
q q1
F~ = (~x − y~1 ). (13)
4π0 |~x − y~1 |3

According to Coulomb’s law, the electric field due to the charge q1 at a point
~x is
~ = 1
E
q1
(~x − y~1 ). (14)
4π0 |~x − y~1 |3
In order to compute the electric field at a point ~x due to a distribution of N
charges qi at positions y~i , we invoke the superposition principle 1 :
N
~ = 1 X qi
E (~x − y~i ). (15)
4π0 i=1 |~x − y~i |3

It is extremely cumbersome to keep track of the numerous charges which are

contained in very small volumes of matter. It is then appropriate to define
1
We know that the superposition principle is valid down to very small distances. How-
ever, at subatomic distances where quantum physics is also at play, we know that it breaks
down. The electric field of one charge interacts with the electric field of another charge and
the combined electric field is different than the sum. The effect (“light by light scattering”)
is tiny but very well measured in precision experiments.

8
macroscopic continuous charge distributions ρ(~y ). The charge dq in a small
volume d3 ~y is given by
dq = ρ(~y )d3 ~y . (16)
The electric field is then given by an integral over the charge density:

~ = 1 Z 3 ρ(~y )
E d ~y (~x − ~y ). (17)
4π0 |~x − ~y |3

2.1.1 Mathematical interlude: the delta-function

Is there an elegant way to transition from the continuous integral of Eq. 17
to the discrete sum of Eq. 15 in the case of a distribution of point-like well
separated charges? Assume for simplicity the case of only one point-like
charge q at a point ~y . The charge distribution should be zero at every point
in space except ~y . On the contrary, at the point ~y where the charge stands,
the distribution is infinite, since the volume of a single point is zero (ρ = q/0).
Dirac introduced a function, the so-called, δ−function which is:
(
0, x 6= y,
δ(x − y) = (18)
∞, x = y.

In addition, the integrals with a delta function kernel are defined as

Z ∞
dxf (x)δ(x − y) = f (y). (19)
−∞

In many (D) dimensions, the delta-function is defined as

D
Y
δ(~x − ~y ) = δ(xi − yi ). (20)
i=1

With the help of the delta-function, the charge density of a single charge q
at a position ~y is expressed as:

ρ(~x) = qδ(~x − ~y ). (21)

The volume integral over this charge distribution gives correctly:

Z Z
d3~xρ(~x) = d3~xqδ(~x − ~y ) = q. (22)

9
The charge distribution of many point-like charges qi at positions y~i is
X
ρ(~x) = qi δ(~x − y~i ). (23)
i

Substituting the charge density of Eq. 23 into Eq. 17 we recover the expression
of Eq. 15.
Integrating with an infinite kernel may be tricky. A δ−function is a
distribution, mapping test well-behaved functions into real numbers. Their
properties can be derived by a limiting procedure on suitable representation
functions. For example, we can think of a delta function as the limit:

δ(x − y) = lim δa (x − y), (24)

a→0

where  h i
 1, x ∈ y − a1 , y + a1 ,
a
δa (x − y) = h i (25)
 0, x 6∈ y − a1 , y + a1 .
Exercise: What is the meaning of
• the derivative of a δ function?

• δ(g(x))?
Exercise: Prove that the derivative of a step function is a delta function.

2.2 Gauss’ law from Coulomb’s law

In this section we will show that Gauss’ law can be derived from Coulomb’s
law. First we make the mathematical observation that

~ ~x 1 ~x − ~y
∇ =− . (26)
|~x − ~y | |~x − ~y |3
The subscript in the nabla operator denotes that it acts on the components
of the vector ~x, i.e. ∇~ ~x = (∂/∂x1 , ∂/∂x2 , ∂/∂x3 ). One can easily prove the
above performing the differentiations of the left hand side. Therefore, the
electric field is the gradient of a scalar function:
~ = −∇Φ(~
E ~ x) (27)

10
the, so called, scalar potential
1 Z 3 ρ(~y )
Φ(~x) = d ~y . (28)
4π0 |~x − ~y |
We can now compute the divergence of the electric field:
!
~ ·E
~ = −∇2 Φ = 1 Z 3 2 1
∇ d ~y ρ(~y ) −∇~x . (29)
4π0 |~x − ~y |

We remind the definition of the Laplacian operator

3
~ ·∇
2~ =
X ∂2
∇ ≡∇ 2
. (30)
i=1 ∂xi

We will prove shortly that:

1
∇~2x = −4πδ(~x − ~y ). (31)
|~x − ~y |
Thus, we have that
~ ·E
∇ ~ = −∇2 Φ = ρ . (32)
0
This is the differential form of Gauss’ law. Additional Reading: Deriva-
tion of Gauss’ law from Coulomb’s law in Feynman Lectures Vol. 2, 4-5,4-6

2.2.1 Integral form of Gauss’ law

Consider a volume V (S) bounded from a surface S(V ). We can integrate
both sides of Gauss’law in this volume:
Z
~ ~
3 1 Z
d ~x∇ · E = d3~xρ(~x) (33)
V (S) 0 V (S)
The numerator of the rhs is the total charge QS(V ) enclosed in the surface
S(V ). The lhs can be written as a surface integral, by applying the divergence
theorem, which states that the flux of a vector A ~ through a closed surface
S(V ) is equal to the divergence of the vector integrated over the volume V (S)
enclosed by the surface.
Z Z
~ ·A
dS ~= ~ · A.
d3~x ∇ ~ (34)
S(V ) V (S)

11
Combining the above, we obtain Gauss’ law in an integral form, stating that
the flux of the electric field through any closed surface S(V ) is proportional
to the electric charge:
Z
~ ~ 1 Z
dS · E = d3~xρ(~x). (35)
S(V ) 0 V (S)
Given that this is valid for any closed surface, all the steps that we have
made in our derivation are reversible. Therefore, the differential and integral
form of Gauss’ law are equivalent.

2.2.2 Mathematical interlude: The Laplacian of inverse distance

It is now time to prove the identity of Eq. 31 which we used to derive Gauss’
law from Coulomb’s law. For reasons to become soon apparent, we shall as-
sume a generic number N of space dimensions. At the end, we will specialize
to N = 3, but keeping the dimension as a generic parameter will elucidate
mathematically how to proceed in intermediate steps of our derivation. With
a direct differentiation we find that
1 3−N
∇2 = . (36)
|~x − ~y | |~x − ~y |3

In N = 3 and for ~x 6= ~y , we find that the rhs is zero. For ~x = ~y and arbitrary
N the result is infinity. What is the result for N = 3? It will still be an
infinity, as we would expect for a delta function. But let’s resolve carefully
this 00 conundrum.
Consider the integral in N dimensions:
Z
1
IN [f ] ≡ dN ~y f (~y )∇~2x . (37)
|~x − ~y |
f is an arbitrary function. Let’s now split the integration volume into two
regions:

• Vin is a sphere (in N dimensions) drawn around the point ~x with a very
small radius R. Inside the sphere, we can approximate:

f (~y ) ≈ f (~x), ∀ ~y in Vin . (38)

12
 • Vout is the rest of the integration volume, not included in Vin .
We first compute the integral IN [f ] inside Vin . We have
Z
1
IN,in [f ] = dN ~y f (~y )∇~2x
Vin |~x − ~y |
Z
1
= f (~x) dN ~y ∇~2x + O(R)
Vin |~x − ~y |
Z
1
= (3 − N )f (~x) dN ~y + O(R)
Vin |~x − ~y |3
Z R
1
= (3 − N )ΩN f (~x) dr rN −1 3 + O(R)
0 r
N −3 N −3
R −0
= (3 − N )ΩN f (~x) + O(R)
N −3
= −ΩN f (~x)RN −3 + O(R). (39)
ΩN is the solid angle in N dimensions. In the above, we have used spherical
coordinates in a system centered at ~x. For N = 3 and setting the radius of
Vin to zero, R → 0, we obtain:
I3,in [f ] = −4πf (~x). (40)
The integral in the rest of the integration volume is:
Z
1
IN,out [f ] = dN ~y f (~y )∇~2x
Vout |~x − ~y |
Z
f (~y )
= (3 − N ) dN ~y (41)
Vout |~x − ~y |3
The integrand has no singularity, since outside the sphere Vin we always
guarantee that ~x 6= ~y and it is converged for the typical smooth functions
f (~x) that we are interested in. For N = 3, we then have:
I3,out [f ] = 0. (42)
We have then found that
I3 [f ] = I3,in [f ] + I3,out [f ] = −4πf (~x), (43)
or, explicitly, Z
1
d3 ~y f (~y )∇~2x = −4πf (~x). (44)
|~x − ~y |

13
The above is valid for every smooth function f (~x). We have therefore proven
the desired identity:
1
∇~2x = −4πδ(~y − ~x). (45)
|~x − ~y |

2.3 Scalar potential

We have found earlier (Eq. 27) that the electric field can be derived from the
gradient of the effective potential. We conclude immediately that the curl of
a static electric field is zero:

~ ×E
~ = −∇
~ × ∇Φ
~ =
X ∂ 2Φ 1X ∂ 2Φ
∇ ijk = (ijk + ikj ) = 0. (46)
jk ∂xj ∂xk 2 jk ∂xj ∂xk

We have then proven that also the second law of electrostatics,

~ ×E
∇ ~ = 0. (47)

Using Stokes’ theorem,

Z I
~ ·∇
dS ~ ×A
~= d~l · A,
~ (48)
S ∂S

we find that the circulation of the electrostatic field in any closed loop Γ is
zero: I
d~l · E
~ = 0. (49)
Γ
The scalar potential has an immediate physics interpretation. Consider
~ and compute the work needed to transport a test charge
an electric field E
from a position ~xA to a position ~xB . The field exerts a force F~ = q E
~ = −q ∇Φ
~
which we need to counteract with an external force F~ext = −F~ = q ∇Φ ~ while
transferring the charge from the starting to the finishing point over a given
path. The work done is
Z ~xB
WA→B = q d~l · ∇Φ
~ = q (Φ(~xB ) − Φ(~xA )) . (50)
~
xA

Nicely, the work done is independent of the path chosen and depends through
the potential only on the starting and ending points.

14
2.4 Potential energy of a charge distribution
In this section, we will compute the energy stored in a charge distribution.
Assume N charges qi at positions x~i . The potential energy of the system is
equal to the energy required to bring the charges one by one to their positions
from an infinite distance (where the electric field vanishes).
N
X
W = Wi , (51)
i=1

where Wi is the work needed to bring the charge qi at ~xi . It costs no energy
to bring the first charge q1 at ~x1 . Once at ~y1 , q1 creates a potential
1 q1
Φ1 (~x) = (52)
4π0 |~x − ~x1 |
To bring q2 at ~x2 requires work
1 q1 q2
W2 = . (53)
4π0 |~x1 − ~x2 |
Consequently, the new potential is
" #
1 q1 q2
Φ12 = + (54)
4π0 |~x − ~x1 | |~x − ~x2 |
and the work needed to bring the charge q3 at ~x3 is
" #
q3 q1 q2
W3 = + (55)
4π0 |~x3 − ~x1 | |~x3 − ~x2 |
For the i−th charge the work needed is:
qi X qj
Wi = (56)
4π0 j<i |~xi − ~xj |
The potential energy is then
N X
1 X qi qj
W = . (57)
4π0 i=2 j<i |~xi − ~xj |
Alternatively, we can write a more symmetric form
1 X qi qj
W = . (58)
8π0 i6=j |~xi − ~xj |
Additional Reading: Electrostatic energy of a salt crystal in:
Feynman Lectures Vol. 2, 8-3

15
2.4.1 Potential energy of a continuous charge distribution
The expression of Eq. 58 in the continous charge distribution limit becomes:

1 Z 3 3 ρ(~x)ρ(~y )
W = d ~xd ~y . (59)
8π0 |~x − ~y |

This can also be written as

1Z 3
W = d ~x ρ(~x)Φ(~x), (60)
2
where we have recognised in the kernel of the ~x integral the scalar potential.
Using Gauss’ law
ρ
∇2 Φ = − (61)
0
(which in this form is known as the Poisson equation), we have
0 Z 3
W =− d ~x Φ(~x)∇2 Φ(~x). (62)
2
We now use integration by parts identity (exercise)
 ~ 2  ~ 2
       
~ · Φ∇Φ
Φ∇2 Φ = ∇ ~ − ∇Φ ~ · Φ∇Φ
 =∇ ~ − E  . (63)

The first term gives a surface integral (using the divergence theorem) and
vanishes if we take the boundary at infinity, where the electric field vanishes.
We are then left with the result
0 Z 3  2
~
W = d ~x E  . (64)
2
which expresses the electrostatic energy as an integral with a kernel the
squared magnitude of the electric field. We can interpret the positive definite
quantity
0  ~ 2
w = E  (65)
2
as the energy density of the electric field. This is also intuitive, since we
associate more energy with regions of space where the electric field is stronger.

16
2.4.2 Self-energy
There is a problem with the expression of Eq. 59 when applying to discrete
charge distributions with a density
X
ρ(~x) = qi δ(~x − ~xi ). (66)
i

We obtain a potential energy

N
1 X q i qj
W = (67)
8π0 i,j=1 |~xi − ~xj |

where in the above summation we do not exclude the cases of i = j. These

terms yield an infinity. The problem is not overcome either when we apply
Eq. 64 to discrete cases. For example, if we consider only one point-like
charge q the energy density is

0  ~ 2 q2 1
w= E  = , (68)
2 32π 2 0 r4
which becomes infinite for r = 0. This infinity is not integrable. The total
energy associated with the existence of a single charge is
Z ∞
q Z∞ 1
 
W = 4π r2 dr w = − d = ∞. (69)
0 4π0 0 r
Obviously, our classical electrodynamics results are not valid when we try to
compute the energy associated with point-like particles. The physics laws
need to be modified. The problem persists also in quantum electrodynamics,
although there we have a technical way to remove the singularities due to
the self-energy of point-like charges and absorb then in a renormalisation of
the electric charge unit.
Exercise: Calculate the electrostatic energy of a uniformly charged sphere
in a couple of ways. What happens when you take the limit of a zero radius?
Check out your answer here: Feynman Lectures Vol. 2, 8-1.

2.5 Charged conductors

Conductors are materials which allow electrons to move freely within their
mass. Commonly, conductors are neutral containing the same amount of

17
Figure 1: A conductor charged with some excess charge. The excess charge
distributes itself at the surface. The electric field vanishes in the interior and
the potential is a constant.

positive and negative charges. However, it is possible to charge them bring-

ing some excess ions on it (adding or removing electrons). After we stop
charging the conductor, the excess charges will settle in such a way so that
the electrostatic energy is minimum. This happens when their relative dis-
tances are maximal. Therefore, the excess charges distribute themselves on
the surface of the conductor. Inside a settled charged conductor the electric
field vanishes.
~ inside = 0.
E (70)
If it did not, it would push and set in motion the free electrons in the inside
(which are balanced by positive ions in the crystal structure) contrary to the
requirement that the charges have settled to fix positions. Consequently, the
scalar potential is everywhere the same in the conductor.
~ inside = 0 ; ∇Φ
E ~ = 0 ; Φ = constant. (71)

The electric field is non-zero on the surface of the conductor and it is perpen-
dicular to the surface. Indeed, a parallel component would set in motion the
charges on the surface of the conductor in contrast again of the electrostatic
assumption.
We can compute easily the magnitude of the electric field on the surface
by applying Gauss’s law in the integral form for the surface of Fig. 2. The

18
Figure 2: Using the integral form of Gauss’ law in order to compute the
electric field on the surface of the conductor.

flux of the electric field through the Gauss surface is:

Z
Flux = ~ ·E
dS ~ = E∆S (72)

and the charge enclosed is:

Charge = σ(~x)∆S, (73)

where σ(~x) is the charge surface density (charge per unit surface) of the
conductor. The electric field on the surface is then:
σ(~x)
E= . (74)
0
The energy density on the surface of the conductor is:

0  ~ 2 σ 2 (~x)
w= E  = . (75)
2 20
The excess charges on the surface of a conductor exert a pressure (which
we can calculate). If the charges on the surface of the conductor were not
constrained and were allowed to increase further their separation distances
the electrostatic energy would reduce. Consider the original conductor and

19
Figure 3: Two almost identical conductors, with an infinitesimal ∆x∆S
volume deformation. The conductors have the same excess charge and charge
density. The electrostatic energy density of the two systems is also identical.

an otherwise identical one but a small deformation ∆x∆S in its volume, as

in Fig. 3. The difference in electrostatic energy for the two configurations is
Z Z
σ2
∆W = d3~xw − d3~xw = −∆S∆x . (76)
Vout Vout−∆x∆S 20
The force which is needed to undo such a deformation should provide the
same amount of work:
∆W = F ∆x. (77)
Thus, the pressure on the surface of the conductor is

|F | |∆W | σ2
= = . (78)
∆S ∆x∆S 20

20
END OF WEEK 1

21
3 Boundary condition problems in electro-
statics
We can claim that we have solved an electrostatics physics problem if we are
able to compute the scalar potential. From it, we can deduce the electric
field, the forces on electric charges and everything else we may need to know.
The electrostatic potential is given by an integral over all space:

1 Z Z Z ∞ 3 ρ(~y )
Φ(~x) = d ~y . (79)
4π0 −∞ |~x − ~y |

This equation may be unpractical for computing the potential in a plethora

of situations. It requires an integration over an infinite volume in which
we are supposed to have full knowledge of the charge density. It is more
likely, however, that we are able to have partial information about the charge
density, restricted within the boundaries of a smaller finite volume V (such
as the boundaries of a laboratory or an electric device). Is it possible to
account for the missing information on the charge density outside the small
volume that we control with something equivalent obtained/measured on the
boundaries of our volume?
We have expressed Gauss’ law in a form of a differential equation
ρ
∇2 Φ = − . (80)
0
Instead of using the infinite integration in order to compute the potential
within the volume V , we can solve the differential equation. Differential
equations yield multiple solutions. To obtain a unique solution, we will need
to supply additional information, in lieu of the charge density outside V .
What boundary conditions guarantee a unique solution?

3.1 Dirichlet and Neumann boundary conditions

Assume that we know two different solutions Φ1 and Φ2 of Poisson’s equa-
tion 80 inside a volume V with boundary S(V ). The difference U = Φ1 − Φ2
will satisfy a Laplace equation:

∇2 U = 0 (81)

22
Therefore
U ∇2 U = 0
 ~ 2
 
; ∇ U ∇U − ∇U  = 0
~ ~
Z Z
 ~ 2
 
; 3
d ~x ∇U  = ~ · (U ∇U
d3~x ∇ ~ )
ZV ZV
 ~ 2
 
; d3~x ∇U  = ~ · (U ∇U
dS ~ ) (82)
V S(V )

The first integral is over the entire volume V in which we want to compute
the potential. The second integral is over the surface of the boundary.
Let’s assume that the potential is fully known, or it can be measured, at
the boundary (but not inside the volume):
Φ1 (~x) = Φ2 (~x) = Φ(~x) ∀~x ∈ S(V ). (83)
Then the rhs of Eq. 82 vanishes. The lhs is a volume integrand with a positive
definite integrand. In order for the lhs integral to vanish, the integrand must
vanish at all points inside V . We conclude that
~ = 0 ; ∇(Φ
∇U ~ 1 − Φ2 ) = 0 ; E
~ 1 (~x) = E
~ 2 (~x), ∀~x ∈ V. (84)
We conclude that if we know the scalar potential on the boundary S(V )
(Dirichlet boundary condition) the solution of the Poisson equation is unique
for the electric field inside the volume V . The potential Φ(x) inside the
volume V .
Let us assume now that the information that we have on the boundary
~ = −∇Φ
is the value of the electric field E ~ (Neumann boundary condition),
thus
~ 1 (~x) = ∇Φ
∇Φ ~ 2 (~x) = ∇Φ(~
~ x) ∀~x ∈ S(V ). (85)
As for a Dirichlet boundary condition, we conclude that ∇U ~ = 0 for all
points inside the volume V and the solution for the electric field E~ inside V
is unique.
A unique solution for E ~ inside V is also guaranteed due to Eq. 82 for a
mixed boundary condition, where for part of the surface S(V ) we know the
~
potential Φ and for the rest of the surface we know the electric field E.
We note that while the three types of boundary conditions (Dirichlet,
Neumann, mixed) determine uniquely the solution of Poisson’s equation for
~
the electric field −∇Φ, the scalar potential is determined only up to a phys-
ically unimportant constant.

23
3.2 Green’s functions
From the divergence theorem,
Z Z
3 ~ · F~ =
d ~x ∇ ~ · F~
dS (86)
V S(V )

and by choosing
F~ = φ∇ψ
~ (87)
we obtain Z Z
~ · φ∇ψ
d3~x ∇ ~ = ~ · φ∇ψ
dS ~ (88)
V S(V )

Interchanging φ and ψ in the above and taking the difference, we have:

Z   Z  
~ · φ∇ψ
d3~x ∇ ~ − ψ ∇φ
~ = ~ · φ∇ψ
dS ~ − ψ ∇φ
~ (89)
V S(V )

and performing the differentiations on the lhs we obtain:

Z   Z  
d3~x φ∇2 ψ − ψ∇2 φ = ~ · φ∇ψ
dS ~ − ψ ∇φ
~ (90)
V S(V )

Equation 90 will be our main tool for solving Poisson’s differential equation
with Dirichlet or Neumann boundary conditions.
Let φ in Eq. 90 be a potential φ = Φ(~x) satisfying Poisson’s equation
ρ
∇2 Φ = − .
0
and ψ a, so called, Green’s function G(~x, ~y ) defined to satisfy

∇~2x G(~x, ~y ) = −4πδ(~x − ~y ). (91)

It is of the form,
1
G(~x, ~y ) = + F (~x, ~y ) (92)
|~x − ~y |
where the function F satisfies the Laplace equation:

∇~2x F (~x, ~y ) = 0. (93)

24
 Substituting for φ and ψ into Eq. 90, we obtain:
Z h i Z h i
d3~x Φ(~x)∇~2x G(~x, ~y ) − G(~x, ~y )∇2 Φ(~x) = ~ · Φ(~x)∇
dS ~ ~x G(~x, ~y ) − G(~x, ~y )∇Φ(~
~ x)
V S(V )
" #
Z
ρ(~x) Z h i
; d3~x −4πδ(~x − ~y )Φ(~x) + G(~x, ~y ) = ~ · Φ(~x)∇
dS ~ ~x G(~x, ~y ) − G(~x, ~y )∇Φ(~
~ x) ,
V 0 S(V )

(94)

Performing the integration over the δ−function we finally obtain,

1 Z 3
Φ(~y ) = d ~x G(~x, ~y )ρ(~x)
4π0 V
1 Z h
~ · Φ(~x)∇ ~ ~x G(~x, ~y ) − G(~x, ~y )∇Φ(~
i
~ x) .
− dS (95)
4π S(V )
This expression is valid for every point ~y ∈ V .

3.2.1 Dirichlet boundary conditions

For Dirichlet boundary conditions, we search for a Green’s function GD (~x, ~y )
which vanishes on the surface S(V ):

GD (~x, ~y ) = 0, ∀ ~x on S(V ), (96)

such as that the second term of the surface integral of Eq. 95. Then we have:
1 Z 3
Φ(~y ) = d ~x GD (~x, ~y )ρ(~x)
4π0 V
1 Z ~ · Φ(~x)∇ ~ ~x GD (~x, ~y ).
− dS (97)
4π S(V )

3.2.2 Neumann boundary conditions

With Neumann boundary conditions, one is tempted to use a Green’s func-
tion with
~ x, ~y ) · n̂ = 0,
∇G(~
where n̂ is a unit vector perpendicular to S(V ). However, such a Green’s
function cannot exist. From the definition of a Green’s function we expect:

∇2 G(~x, ~y ) = −4πδ(~x − ~y )

25
 Z Z
; d3~x ∇2 G(~x, ~y ) = −4π d3~x δ(~x − ~y )
ZV V

; ~ · ∇G(~
dS ~ x, ~y ) = −4π 6= 0
S(V )

; ∇G(~
~ x, ~y ) · n̂ 6= 0. (98)

We can choose, however, a Green’s function which has a constant com-

ponent for its gradient in the direction perpendicular to the surface, such
as
~ N (~x, ~y ) = − 4π n̂,
∇G (99)
S
where S is the total surface of the boundary S(V ):
Z
S= ~ · n̂.
dS (100)
S(V )

Then, Eq. 95 yields:

1 Z 3
Φ(~y ) = d ~x GN (~x, ~y )ρ(~x)
4π0 V
1 Z ~ · GN (~x, ~y )∇
~ ~x Φ(~x) + hΦi
+ dS S(V ) , (101)
4π S(V )
where the last term is the average of the potential on the boundary
R
~ · Φn̂
dS
S(V )
hΦiS(V ) = (102)
S
and it is an unimportant constant for physical observables.

3.3 Explicit solutions of boundary condition problems

We have identified required properties of Green’s functions for Dirichlet and
Neumann boundary conditions. Finding these Green’s functions is not easy.
In the following, we will consider some simple examples and develop tech-
niques for such a purpose.

3.3.1 Example 1: A conductor filling half of space

Suppose that we fill half of space with a conducting material as in the fol-
lowing figure:

26
We live in the other half of space where we can have a charge density ρ(~x).
The potential of the conductor is constant throughout its mass. The bound-
ary condition for this problem is of Dirichlet type. We require a Green’s
function which vanishes on the surface S separating the conductor from the
half-space we live in:
G(~x, ~y ) = 0 ∀ ~x on S. (103)
G(~x, ~y ) has the form
1
G(~x, ~y ) = + F (~x, ~y ) (104)
|~x − ~y |
where
∇2 F (~x, ~y ) = 0. (105)

Let’s choose a coordinate system as in the figure above. For each vector
~y = (y1 , y2 , y3 ), we define a dual vector ~y ∗ = (−y1 , y2 , y3 ). We take ~y to
belong to the free-space, which makes ~y ∗ to belong to the space occupied
by the conductor. The two vectors ~y and ~y ∗ are identical for y1 = 0, on the
surface S separating the free space and the conductor. With some guesswork,
we find that
1
F (~x, ~y ) = − (106)
|~x − ~y ∗ |
satisfies all the criteria we have required for our Green’s function. Indeed,
1 1
G(~x, ~y ) = − (107)
|~x − ~y | |~x − ~y ∗ |

27
vanishes on the boundary ~y = (0, y2 , y3 ). In addition,
1
∇2 F (~x, ~y ) = −∇2 = 4πδ(~x − ~y ∗ ) = 0. (108)
|~x − ~y ∗ |
The delta function above is always zero since ~x belongs to the free half-space,
while ~y ∗ belongs always to the half-space occupied by the conductor.
Eq. 97 yields for the potential:
1 Z ρ(~x)
Φ(~y ) = d3~x
4π0 free−space |~x − ~y |
1 Z
ρ(~x)
− d3~x
4π0 free−space |~x − ~y ∗ |
" #
1 Z ~ ~ 1 1
− dS · Φ(~x)∇ − . (109)
4π S |~x − ~y | |~x − ~y ∗ |
We now make the observation that
q
|~x − ~y ∗ | = (x1 + y1 )2 + (x2 − y2 )2 + (x3 − y3 )2 = |~y − ~x∗ | , (110)

where ~x∗ = (−x1 , x2 , x3 ). We can then write:

1 Z ρ(~x)
Φ(~y ) = d3~x
4π0 free−space |~x − ~y |
1 Z
(−ρ(~x))
+ d3~x ∗
4π0 free−space |~x − ~y |
" #
1 Z
~ ~ 1 1
− dS · Φ(~x)∇ − . (111)
4π S |~x − ~y | |~x∗ − ~y |
The potential on the surface S is a constant Φ(~x) = V . It is left as an exercise
to calculate the surface integral in the last term. One finds,
" #
1 Z ~ ~ 1 1
− dS · Φ(~x)∇ − = V. (112)
4π S |~x − ~y | |~x∗ − ~y |
and we have
1 Z ρ(~x)
Φ(~y ) = V + d3~x
4π0 free−space |~x − ~y |
1 Z
(−ρ(~x))
+ d3~x ∗ (113)
4π0 free−space |~x − ~y |

28
Figure 4: The potential for discrete charges qi and an infinite conducting
half-space is the same in the free half-space as the potential of the charges qi
and their mirror charges −qi .

Eq. 113 can be used to calculate the potential for arbitrary charge distribu-
tions for this geometry. If analytic solutions are not possible, the integrals
can be performed with numerical methods.
In the special case of a discrete charge distribution,
X
ρ(~x) = qi δ(~x − ~xi ), (114)
i

where ~xi lie inside the free volume, we obtain

1 X qi (−qi )
Φ(~y ) = + (115)
4π0 i |~y − ~xi | |~y − ~x∗i |

This result is familiar! It is the solution of a different electrostatic problem

than the one that we have considered here. It is the electrostatic potential
due to the charges qi at positions ~xi and −qi charges at reflected (mirrored)
positions ~x∗i . The problem of finding the potential for charges qi at ~x in the
presence of an infinite plane conductor has exactly the same solution in the
free half-volume not occupied by the conductor as the problem of finding the
potential for the charges qi at ~xi and the “mirror charges” −qi at ~x∗i .

3.3.2 Example 2: Reverse engineering and the method of images

(mirror charges)
It is often possible to figure out a clever configuration of charges which has
the same effect within some volume V as boundary conditions on the surface

29
Figure 5: The potential of a positive and a negative charge. There is a sphere
on which the potential is zero.

S(V ) of the volume. As an example, we place a charge +q at a position d~

2
~ The contributions of the
and a charge of opposite − Rd q at a position Rd2 d.
two charges to the scalar potential at a position ~r are of opposite sign:
 
R
q  1 d
Φ(~r) =  −  . (116)
4π0 ~r − d~ ~r − Rd2 d~
2

There is a surface on which the two contributions cancel against each other.
Indeed, Eq. 116 gives
 zero for all points ~
r = R~ on the surface of a sphere
~
with radius R = R:
 
R
~ = q 1 d
Φ(R)   −  
R − d~ R2 ~
4π0 ~ ~−
R d

d2
 
R
q  1 d

=   −h 
i1 
R − d~
4π0  ~
R4 2
~ · d~

R2 + d2
− 2 Rd2 R 2

 
q  1 1 
=   −h 
i1 
R − d~
4π0  ~
~ · d~
 2
d2 + R2 − 2R

30
  
q  1 1 
=  − 
4π0 R
 ~ − d~ R
  ~ − d~
= 0. (117)
Now, consider a different electrostatic problem of a conducting sphere of
a radius R at zero potential and a charge +q at a distance d from the center
of the sphere. The potential inside the sphere conductor is everywhere zero.
What is the potential outside the sphere? The theorem of Eq. 97 tells us
that once we know the potential at the boundary of a volume the solution
of Poisson’s equation for the potential is unique within that volume. In our
case, the volume we are interested in is bounded by the surface of the sphere
of the conductor and the surface of all space at infinity. In both (infinite
and sphere) surfaces the potential is zero. However, recall that we have
identified identical zero potential surfaces with the exact same geometry in
our first problem of the two charges in free space. Therefore, the solution of
the two problems should be identical in the volume enclosed by the surfaces.
Therefore, the solution for the problem with the charge +q at a distance d
from the center of a conductor of radius R is,
  
 R
q  1

−  d  , ∀ ~r : r ≥ R


 4π0 |~r−d~|
 2
r− R2 d~
~

Φ(~r) =  d

(118)




0, ∀ ~r : r < R



Let’s compare this solution with the general solution of Eq. 116. For a charge
density:
1 Z 3 ~
Φ(~r) = d ~x GD (~x, ~r)qδ(~x − d)
4π0 V
1 Z ~ · Φ(~x)∇ ~ ~x GD (~x, ~r),
− dS (119)
4π S(~x=R)~

or, equivalently,
q ~ ~r).
Φ(~r) = GD (d, (120)
4π0
Therefore, the simple solution of the two-charges problem gives us the Green’s
function for all boundary condition problems with a Green’s function. We
have,
R
~ ~r) = 4π0 Φ(~r) =  1  −  d  .
GD (d, (121)
r − d~ ~r − R2 d~
2 
q ~
  
d

31
With the Green’s function and Eq. 97 at hand we can solve any other problem
with Dirichlet boundary conditions on the same geometry. For, example,
one could find the solution for the potential when the conductor is kept at
a potential V 6= 0 and a single charge +q outside the conductor, or the
potential in the presence of a spherical conductor and an arbitrary charge
distribution inside the free volume (exercise: try it for a spherical segment
distribution with the same center as the conductor).

3.4 Green’s functions from Laplacian eigenfunctions

In this section, we will present a systematic way to obtain Green’s functions
for Dirichlet boundary conditions on a surface S(V ) surrounding a volume
V . We remind that a Dirichlet Green’s function vanishes on S(V ).
Consider the eigenfunctions ψn (~x) of the Laplace operator:

∇2 ψn = λn ψn , (122)

and impose on them that they vanish on the boundary:

ψn (~x) = 0, ∀~x ∈ S(V ). (123)

Often, the eigenvalues λn turn out to be discrete. Let us assume that this is
the case for the rest of the analysis. In the continuous cases, our formulae
will need to be modified trivially, changing summations into integrations. In
addition, we will assume that the eigenvalues are not degenerate (all λn are
different from each other).
We now make use of the theorem in Eq. 90. We have
Z h i Z h i
∗
d3~x ψm ∗
∇2 ψn − ψn ∇2 ψm = ~ · ψ ∗ ∇ψ
dS ~ n − ψn ∇ψ
~ ∗ (124)
m m
V S(V )

The rhs of Eq. 124 vanishes since the eigenfunctions ψn vanish on the bound-
ary S(V ). Thus, we obtain:
Z
(λ∗m − λn ) ∗
d3~x ψm (~x)ψn (~x) = 0. (125)
V

For m = n, we have that

Z
(λ∗n − λn ) d3~x |ψn (~x)|2 = 0. (126)
V

32
Since the above integral is positive definite, we conclude that

λ∗n = λn , (127)

stating that the eigenvalues are real. For m 6= n, we conclude that

Z
∗
d3~x ψm (~x)ψn (~x) = 0. (128)
V

Choosing appropriately the normalization of the eigenfunctions, we have the

orthogonality condition:
Z
∗
d3~x ψm (~x)ψn (~x) = δnm (129)
V

It is often the case that the eigenfunctions form a complete basis, meaning
that any other function f (~x) which vanishes on the boundary S(V ) can be
written as a linear superposition of the Laplace eigenfunctions:
X
f (~x) = cn ψn (~x) (130)
n

∗
Multiplying with ψm (~x) and integrating over the volume V we obtain:
Z Z
∗ ∗
d3~x ψm d3~x ψm
X X
(~x)f (~x) = cn (~x)ψn (~x) = cn δnm = cm . (131)
V n V n

Substituting into Eq. 130 we obtain that

Z
d3 ~y f (~y ) ψn∗ (~y )ψn (~x)
X
f (~x) = (132)
n

which leads to the completeness condition:

ψn∗ (~y )ψn (~x) = δ(~x − ~y ).

X
(133)
n

We are now ready to apply the above to a Dirichlet Green’s function

GD (~x, ~y ), which we can write as a linear superposition of Laplace eigenfunc-
tions: X
GD (~x, ~y ) = cn (~y )ψn (~x). (134)
n

33
Applying the Laplace operator on both sides we obtain
; ∇2 GD (~x, ~y ) = cn (~y )∇2 ψn (~x)
X

; −4πδ(~x − ~y ) =
X
cn (~y )λn ψn (~x)
n
; −4π ψn∗ (~y )ψn (~x) =
X X
cn (~y )λn ψn (~x)
n n
4π ∗
; cn (~y ) = − ψ (~y ) (135)
λn n
Knowing the eigenvalues and eigenfunctions of the Laplace operator we can
construct the Green’s function via:
X ψn∗ (~y )ψn (~x)
GD (~x, ~y ) = −4π . (136)
n λn

3.4.1 Example: all (infinite) space

The Green’s function with boundaries at infinity is the inverse of the distance
of the two vectors in the arguments of the function:
1
GD (~x, ~y ) = . (137)
|~x − ~y |
We will derive this result from the eigenfunctions of the Laplacian:
∇2 ψ(~x) = λψ(~x). (138)
The solutions of this equation (eigenfunctions) are:
1 ~
ψ~k (~x) = 3 eik·~x , (139)
(2π) 2

with eigenvalues:
 2
 
λ~k = − ~k  (140)
Exercise: Prove that Z ∞
dxeixa = 2πδ(a). (141)
−∞
Hint: Think what is the Fourier transform of a delta function and the inverse
Fourier transform of 1. These solutions are orthonormal. Indeed,
Z Z
d3~x i~x·(~k0 −~k)
d3~x ψ~k (~x)∗ ψ~k0 (~x) = e = δ(~k 0 − ~k). (142)
(2π)3

34
They are also complete:
Z
3~ ∗
Z
d3~k i~k·(~y−~x)
d k ψ~k (~x) ψ~k (~y ) = e = δ(~x − ~y ). (143)
(2π)3

We have established all conditions required for applying Eq. 136 in order to
calculate the Green’s function from the eigenfunctions of the Laplace opera-
tor. We have
Z
d3~k i~k·(~x−~y)
G(~x, ~y ) = −4π  2 e (144)
− ~k 
 

We can perform the integrations, by writing

 
~k · (~x − ~y ) = ~k  |~x − ~y | cos θ, (145)

using spherical coordinates:

   2
   
d3~k = d ~k  ~k  d cos θ dφ (146)

and Eq. 141. We find:

1
G(~x, ~y ) = , (147)
|~x − ~y |
which is the anticipated result.

3.4.2 Example: Inside an orthogonal parallelepiped

We will now calculate the Green’s function for Dirichlet boundary conditions
on the sides of an orthogonal parallelepiped:

V : x ∈ [0, a], y ∈ [0, b], z ∈ [0, c]. (148)

The eigenfunctions of the Laplace operator which vanish on the boundary of

the orthogonal parallelepiped are:
s !
8 lπx mπy nπx
   
ψlmn = sin sin sin (149)
abc a b c

where
n, l, m = 1, 2, . . .

35
and eigenvalues:
l2 m2 n2
!
2
λlmn = −π + + 2 (150)
a2 b2 c
You can check that they are orthonormal and satisfy the completeness con-
dition (bf exercise). The Green’s function is then an infinite series:
∞
X ψlmn (~x)ψlmn (~y )
G(~x, ~y ) = −4π . (151)
l,m,n=1 λlmn

36
END OF WEEK 2

37
3.5 Laplace operator and spherical symmetry
Let us now consider boundary condition problems with spherical symmetry.
To find the eigenfunctions of the Laplace operator,

∇2 ψ = λψ, (152)

it is best to work in spherical coordinates (r, θ, φ). The Laplace operator

takes the form
1 ∂2 Â
∇2 = 2
r + 2, (153)
r ∂r r
where  a differential operator acting only on the angles θ, φ:
1 ∂ ∂ 1 ∂2
 = sin θ + . (154)
sin θ ∂θ ∂θ sin2 θ ∂ 2 φ
Setting x = cos θ, the angular differential operator takes the form:
∂ ∂ 1 ∂2
 = (1 − x2 ) + . (155)
∂x ∂x 1 − x2 ∂ 2 φ
We can solve this differential equation by means of the method os separation
of variables. We seek factorizable solutions:
R(r)
ψ(r, θ, φ) = Y (θ, φ), (156)
r
where the radial and angular parts are factorized. Substituting in Eq. 152,
we obtain:
2
" #
1 ∂ R 1
r2 2
− λ = − ÂY. (157)
R ∂r Y
Since the rhs depends only on angular coordinates and the lhs depends only
on the radial coordinate, for them to be equal they must be both the same
constant l(l + 1) (written in such a way for future convenience). We thus
have two separate differential equations:
1 ∂ 2 R l(l + 1)
− −λ=0 (158)
R ∂r2 r2
and
ÂY = −l(l + 1)Y. (159)
In the following we will solve the above differential equations separately.

38
3.5.1 Radial differential equation
We will defer the general solution of the radial differential equation for later
(exercise classes). Here, we consider the special case of a zero eigenvalue:
λ = 0. Notice that this corresponds to solving ∇2 Ψ = 0, which is to calculate
the potential in empty space. The radial equation becomes:

1 ∂ 2R l(l + 1)
2
= (160)
R ∂r r2
For
R = ra ,
we obtain:
a(a − 1) = l(l + 1), (161)
with solutions:
a = −l, 1 + l. (162)
Thus, the general solution of the Laplace equation in spherical coordinates
is:
X1 
Ψ(r, θ, φ) = Al r−l + Bl rl+1 Yl (θ, φ). (163)
l r

where the sum could represent an integral if l turns out to be a continuous

constant (it will not!).

3.5.2 Angular differential equation

Eq. 159 can also be solved with the method of separation of variables. We
write:
Y (θ, φ) = Θ(θ)Φ(φ). (164)
Then, it becomes:

1 − x2 ∂ 1 ∂ 2Φ
" #
∂
(1 − x2 ) + l(l + 1) Θ = − . (165)
Θ ∂x ∂x Φ ∂φ2

As before, the two sides must be equal to a constant, which we call m2 . This
leads to the two differential equations:

∂ 2Φ
2
= −m2 Φ (166)
∂φ

39
and
m2
" #
∂ ∂
(1 − x2 ) + l(l + 1) − Θ = 0. (167)
∂x ∂x 1 − x2
The differential equation on the azimuthal variable can be solved easily,
yielding:
Φ(φ) = eimφ . (168)
Demanding that this function is single-valued,

Φ(φ) = Φ(φ + 2π), (169)

we obtain that m must be an integer,

m = 0, ±1, ±2, . . . (170)

We will find solutions for the differential equation (Eq. 167) on the polar
angle θ for m = 0 first. Int this special case it becomes:
" #
∂ ∂
(1 − x2 ) + l(l + 1) Θ = 0. (171)
∂x ∂x
Consider a polynomial of degree l:

pl (x) = (x2 − 1)l (172)

Let’s introduce an abbreviation for the differential operator:

d dm
D≡ , Dm ≡ .
dx dxm
The product rule for multiple derivatives is:
m
m!
Dm (f g) = Dk f Dm−k g.
X
(173)
k=0 k!(m − k)!

We start with the identity:

(x2 − 1)Dpl (x) = 2lxpl (x) (174)

Applying the differential operator Dl+1 on the above and using the product
rule of Eq. 173, we arrive at the identity
   
D(1 − x2 )D Dl pl (x) + l(l + 1) Dl pl (x) = 0. (175)

40
Therefore, the polynomials pl (x) are solutions of the differential equation 171.
These polynomials, normalized as:

1 dl 2
Pl (x) = l l
(x − 1)l (176)
2 l! dx
are known as the Legendre polynomials. The first few of them are:

P0 (x) = 1
P1 (x) = x
P2 (x) = 3/2 x2 − 1/2
P3 (x) = 5/2 x3 − 3/2 x
35 4 15 2
P4 (x) = x − x + 3/8
8 4
... (177)

The Legendre polynomials can be cast as the sum:

l
!2
1 X l
Pl (x) = l (x − 1)l−k (x + 1)k (178)
2 k=0 k

Notice that they are normalized so that

Pl (1) = 1. (179)

Legendre polynomials vanish when integrated with any other polynomial

of a lesser degree in the range [−1, 1]:
Z 1
dxxk Pl (x) = 0, ∀ k = 0, 1, . . . (l − 1). (180)
−1

Indeed,
Z 1 Z 1
dxxk Dl (x2 − 1)l = dxxk DDl−1 (x2 − 1)l
−1 −1
Z 1
1

k l−1 2
=x D (x − 1)l  −k dxxk−1 Dl−1 (x2 − 1)l (181)
−1 −1

The first term vanishes at the boundary, since

Dl−1 (x2 − 1)l ∝ (x2 − 1)

41
We can perform integration by parts k < l times where we always end up
with vanishing boundary terms proportional to (x2 − 1). It is then obvious
that two different Legendre polynomials are necessarily orthogonal:
Z 1
dxPl (x)Pm (x) = 0, ∀ l 6= m. (182)
−1

With the above procedure, it is also easy to prove that:

Z 1
l!2 2l+1
dxxl Pl (x) = (183)
−1 (1 + 2l)!

To calculate the normalization of a Legendre polynomial, we first note that

the highest order term is

(2l)! l
Pl (x) = l 2
x + O(xl−1 ) (184)
2 l!
and therefore
Z 1
(2l)! Z 1 2
dxPl (x)2 = l 2
dxxl Pl (x) = . (185)
−1 2 l! −1 1 + 2l
We then write the orthogonality relation
Z 1
2
dxPl (x)Pm (x) = δlm . (186)
−1 1 + 2l
Legendre polynomials form a basis for all continuous functions f (x) in
[−1, 1].
∞
X
f (x) = cn Pn (x), (187)
n=0

where the coefficients are found as:

Z 1 ∞
X Z 1
dxPm (x)f (x) = cn dxPn (x)Pm (x)
−1 n=0 −1
Z 1
2
; dxPm (x)f (x) = cm , (188)
−1 1 + 2m
and finally,
2n + 1 Z 1
cn = dxPn (x)f (x). (189)
2 −1

42
Inserting this coefficient into Eq. 187, we find:
Z 1 ∞
X 2n + 1
f (x) = dyf (y) Pn (x)Pn (y) (190)
−1 n=0 2
which gives the completeness relation:
∞
X 2n + 1
Pn (x)Pn (y) = δ(x − y). (191)
n=0 2
We now turn to the more complicated differential equation 167 for arbi-
trary values of m. Finite solutions in the range: x ∈ [−1, 1] are:
m ∂m
Plm (x) = (−1)m (1 − x2 ) 2 Pl (x) (192)
∂xm
or,
(−1)m 2 m ∂ l+m 2
Plm (x) = (1 − x ) 2 (x − 1)l . (193)
2l l! ∂xl+m
where the constant m must be an integer:
m = −l, −l + 1, . . . 0, . . . l − 1, l. (194)
The polynomials Plm (x) are the associated Legendre polynomials. Polyno-
mials with negative values of the integer m are related to polynomials with
positive values of the integer m via:
(l − m)! m
Pl−m = (−1)m P (x). (195)
(l + m)! l
They satisfy the orthogonality relation
Z 1
2(m + l)!
dxPkm (x)Plm (x) = δkl (196)
−1 (2l + 1)(l − m)!
exercise: Prove that
Z 1
Plm (x)Pln (x) (l + m)!
dx 2
= δm,n , if m 6= 0. (197)
−1 1−x m(l − m)!
Putting together the polar and azimuthal solutions, we find that the
functions: v
u (2l + 1)(l − m)!
u
Ylm (θ, φ) = t eimφ Plm (cos θ) (198)
4π(l + m)!

43
are partial solutions of Eq 159. These functions are called spherical har-
monics and appear in almost every physics problem with spherical symme-
try.
Spherical harmonics are orthogonal. The orthogonality condition is that:
Z 2π Z π
dφ dθ sin θYl∗0 m0 (θ, φ)Ylm
∗
(θ, φ) = δl0 l δm0 m (199)
0 0

and it follows directly from the separate orthogonality conditions of the as-
sociated Legendre polynomials (Eq. 196) and the azimuthal solutions eimφ :
Z 2π
dφ −imφ im0 φ
e e = δm0 m . (200)
0 2π
Every function of the polar and azimuthal angles can be written as a
linear superposition of spherical harmonics:
∞ X
X l
f (θ, φ) = clm Ylm (θ, φ) (201)
l=0 m=−l

where the coefficients are:

Z 2π Z π
∗
clm = dφ dθ sin θYlm (θ, φ)f (θ, φ). (202)
0 0

Inserting this expression back to Eq. 201 we obtain the completeness identity
for spherical harmonics:
∞ X
l
∗
(θ0 , φ0 )Ylm (θ, φ) = δ(φ0 − φ)δ(cos θ0 − cos θ).
X
Ylm (203)
l=0 m=−l

We note that for m = 0, the spherical harmonics collapse to the Legendre

polynomials: s
1 + 2l
Yl0 (θ, φ) = Pl (cos θ). (204)
4π
and are independent of the angle φ. In addition, spherical harmonics with
m 6= 0 vanish at θ = 0. To be convinced about it, recall the factor (1 −
x2 )m/2 ∼ sinm θ in the definition of the associated Legendre polynomials
Plm (cos θ).
Finally, we note that spherical harmonics with same “polar” integer l and
opposite “azimuthal” integer m are related by:
∗
Yl,−m (θ, φ) = (−1)m Ylm (θ, φ). (205)

44
 exercise: Find the eigenfunctions of the Laplace operator in terms of
spherical harmonics
∇2 Ψλ (r, θ, φ) = λΨλ (r, θ, φ) (206)
with the boundary conditions Ψλ (0, θ, φ) = Ψλ (a, θ, φ) = 0.

3.5.3 Expansion of inverse distance in Legendre polynomials

Consider two vectors:
~rL , ~rS , rL > rS .
Their distance is:
h i1
|~rL − ~rS | = rL2 + rs2 − 2rL rs cos θ 2
(207)
where θ is the angle of the two vectors. Any function of cos θ can be expanded
as a series in Legendre polynomials. In particular, for the inverse of a distance
we have ∞
1 X
= cl Pl (cos θ). (208)
|~rL − ~rS | l=0
To compute the coefficients, we consider the special case of the two vectors
being parallel θ = 0. Recall that Legendre polynomials are normalized to be
one for a unit argument:
Pl (cos 0) = Pl (1) = 1.
Then
∞ ∞
1 rSl rSl
;
X X
cl = = l+1 c l = l+1 . (209)
l=0 rL − rS l=0 rL rL
We then have the result
∞
1 X rSl
= Pl (cos θ). (210)
|~rL − ~rS | l=0 rLl+1
Similarly, we can derive the more general result:
∞
1 X (a, l) rSl
a = Pl (cos θ), (211)
|~rL − ~rS | l=0 l! rLl+a
where the Pochhammer symbol is defined as:
Γ(a + l)
(a, l) = . (212)
Γ(a)

45
3.6 Multipole expansion
Consider a charge distribution ρ(~x) which occupies a small volume V 0 . We
are interested in the potential that this distribution creates at a distance ~r
outside the region of the charge distribution. Intuitively, we expect that if
we are sufficiently far from the charge distribution the potential will resem-
ble a Coulomb potential. This is correct to a first approximation, but not
exact; there should be corrections which become more important the closer
we approach to the charges of the distribution.
The potential is given by

1 Z ρ(~x)
Φ(~r) = d3~x . (213)
4π0 V 0 |~x − ~r|

Since r > x, we can expand the inverse distance in the integrand in xr ,

obtaining:
∞
1 X 1 Z
Φ(~r) = d3~xρ(~x)xl Pl (cos γ), (214)
4π0 l=0 rl+1 V 0
where γ is the angle formed by the two vector ~x and ~r. We write the two
vectors as:
~x ≡ (x, θx , φx ), ~r ≡ (r, θ, φ). (215)
It is an easy geometry exercise to prove that:
~x · ~r
cos γ = = cos θ cos θx + sin θ sin θx cos(φ − φx ). (216)
xr
We would like to perform the d3~x = x2 dxdΩx integration in spherical coor-
dinates. However, the expression for cos γ in terms of θx , φx seems overly
complicated. To our rescue comes the following identity, called the “addition
theorem”:
l
4π X
Pl (cos γ) = Y ∗ (θx , φx )Ylm (θ, φ). (217)
1 + 2l m=−l lm
It is a consequence of the fact that the lhs is a function of angular co-ordinates
and it can therefore be expressed as a linear superposition of spherical har-
monics. It is also manifest that the expression is symmetric under the ex-
change: (θ, φ) ↔ (θx , φx ). We will not prove this theorem here, since it can
be proven easier in a future course of Quantum Mechanics. Substituting into

46
Eq. 214, we obtain:
∞ l
1 X 1 1 X l Ylm (θ, φ)
Z 
2 ∗
Φ(~r) = x dxdΩx Ylm (θx , φx )ρ(~x)x ,
0 l=0 1 + 2l rl+1 m=−l V 0 r1+l
(218)
or, in a more compact form:
∞ l
1 X 1 1 X qlm Ylm (θ, φ)
Φ(~r) = l+1
, (219)
0 l=0 1 + 2l r m=−l r1+l

where we have defined the quantities:

Z
∗
qlm = d3~x Ylm (θx , φx )ρ(~x)xl (220)
V0

which are called multipole moments. The moments characterise the geome-
try of the charge distribution. For example,
Z
1 Z 3 Q
q00 = d3~xρ(~x)Y00 (θ, φ) = √ d ~xρ(~x) = √ (221)
4π 4π
and it is proportional to the total charge in the distribution. For the higher
moments, we find: s
3
q11 = − (p1 − ip2 ), (222)
8π
s
3
q00 = (p3 ), (223)
4π
with Z
p~ = (p1 , p2 , p3 ) = d3~x ~xρ(~x) (224)
the dipole moment. For the next moments:
s
1 15
q22 = (Q11 − 2iQ12 − Q22 ), (225)
12 2π
s
1 15
q21 = − (Q13 − iQ23 ), (226)
3 8π
s
1 5
q21 = Q33 (227)
2 4π

47
where Z
Qij = d3~x (xi xj − x2 δij )ρ(~x) (228)
the quadrupole tensor, etc. The most important moments are the first ones,
since the higher moments are suppressed by powers of 1/r in their contri-
bution to the potential. A measurement of the angular distribution of the
potential due to some unknown charge distribution can be used for the ex-
traction of the multipole moments of the distribution. This is our best way
to learn something about the geometrical characteristics of the distribution.

48
END OF WEEK 3

49
4 Magnetic field
The force acting on a charge q which moves with velocity ~v inside an elec-
~ B)
tromagnetic field (E, ~ is:
 
F~ = q E
~ + ~v × B
~ . (229)

The direction of the electric component of the force is along or against the
~ The magnetic component of the force is at
direction of the electric field E.
right angles with both the magnetic field and the direction of motion of the
charge.

4.1 Currents
Macroscopically, it is easy to observe the magnetic force acting on a large
~ Currents
number of charges which move together inside a magnetic field B.
can be materialized inside conductors which allow electrons to move freely
within their body.

Let us consider a volume element dV = dxdS with a charge density

∆Q0
ρ=
dV
at an initial time t0 . ∆Q0 is the total charge contained inside dV at t0 .
Assume that all elementary charges are identical (q). Then

∆Q0 = N0 q, (230)

where N0 is their number inside dV at t0 . Let’s assume that all charges are
moving with a velocity ~v (which we have aligned with the x−axis).

50
In a small time dt a number of charges will move outside the volume dV .
This number is:
~
qNescaping = ρ(~v dt) · dS. (231)
~ per
We define the current density ~j as the charge transversing a surface dS
unit time and surface. Specifically:

qNescaping = J~ · dSdt
~ (232)

By comparing the last two equations we conclude that:

J~ = ρ~v . (233)

After time dt, the remaining charge inside the volume dV is

∆Qt0 +dt = ∆Q0 − qNescaping

∆Qt0 +dt − ∆Q0
; = −J~ · dS
~
dt
∂ρ
; dV = −J~ · dS.
~ (234)
∂t
Eq. 234 states that the change in the charge density during dt is equal with
~ Integrating over the
the flux of the current density through the surface dS.
surface S(V ) of an arbitrary volume V , we have:
Z
∂ρ Z
dV = − J~ · dS
~
V ∂t S(V )

3 ∂ρ
Z Z
; d ~x = − ∇ ~ · J~
V ∂t V
" #
Z
∂ρ ~ ~
; 3
d ~x + ∇ · J = 0. (235)
V ∂t

51
Eq. 235 must hold for any volume V . In order for this to happen, the
integrand must vanish. Therefore, we arrive to the continuity equation:
∂ρ ~ ~
+∇·J =0 (236)
∂t
The above equation is a consequence of charge conservation.

4.2 Magnetic field of steady currents

We will now study a very common situation where the charge density is
constant:
∂ρ
= 0.
∂t
This can be materialized in simple circuits, where charge flows in and out (at
the same rate) of any volume element. The divergence of the electric current
density is therefore:
~ · J~ = 0.
∇ (237)
In such situations, the electric and magnetic fields are constant in time. The
magnetic field is determined via two of Maxwell equations, which become:

∇ ~ =0
~ ·B (238)

and
~
~ ×B
c2 ∇ ~ = J. (239)
0
We can derive an integrated form for the above. From Eq. 238, we obtain
that: Z Z
d3~x ∇ ~ =0;
~ ·B ~ · dS
B ~ = 0. (240)
V S(V )

Therefore, the flux of the magnetic field out of any volume is zero. Recall
than in electrostatics,

~ = Qinside V .
Z
~ · dS
E
S(V ) 0
The analogous Eq. 240 states that there is no magnetic charge to be found
anywhere.
Integrating Eq. 239 over an open surface S(Γ) bounded by a closed loop
Γ,

52
we obtain:

J~ · dS
~
Z R
 
2 ~· ∇
~ ×B
~ = S(Γ)
c dS
S(Γ) 0
J~ · dS
~
I R
~ · d~l = S(Γ)
; c2 B . (241)
Γ 0
The rhs is proportional to the current intensity passing through the loop Γ:
Z
Ithrough Γ ≡ J~ · dS.
~ (242)
S(Γ)

and denotes the total charge passing through the closed loop Γ per unit time.
We then arrive at:
I
~ · d~l = through Γ ,
I
B (243)
Γ 0 c2
which is known as Ampere’s law.

4.3 Applications of Ampere’s law

For problems with a manifest symmetry, Ampere’s law can be sufficient to
determine the magnetic field. Let’s review here two very well known appli-
cations.

4.3.1 Long straight wire

Consider a long straight wire with a current I flowing through it.

53
Close to the wire, in a distance r which is much smaller than the length of
the wire, the magnetic field is by symmetry the same at all positions of the
same distance r. Applying Ampere’s law for a loop which is a circle of radius
r around the wire we obtain:
I
I
~ · d~l =
B
0 c2
I
; B2πr =
0 c2
1 2I
;B = . (244)
4π0 c2 r

4.3.2 Solenoid
Consider a long solenoid with a current I through it.

Very long solenoids have a homogeneous magnetic field inside them and a
zero magnetic field outside them, to a good approximation. These two as-
sumptions allow us to estimate the magnetic field inside the solenoid.

Let’s apply Ampere’s law for the loop Γ in the picture above. The total
current through Γ is the number Nin Γ of spires in it times the current I
through one spire:
Ithrough Γ = INin Γ (245)
Then, from Ampere’s law:
INin Γ
BL = (246)
0 c2

54
which gives the magnetic field:
nI
B= . (247)
0 c2
n is the number of spires per unit length in the solenoid.

4.4 Vector potential

Let’s examine further the basic equations of magnetostatics (Eqs 238-239).
Eq. 238 is automatically satisfied if we cast the magnetic field as the curl of
~
a vector A:
~ = ∇ × A.
B ~ (248)
We introduce the abbreviation:
∂
∂i ≡ (249)
∂xi
and Einstein’s summation convention:
3
~ · B.
~
X
Ai Bi ≡ Ai Bi = A (250)
i=1

Then
~ · (∇
∇ ~ × A)
~ = ∂i (∇
~ × A)
~ i = ∂i (ijk ∂j Ak ) = ijk ∂i ∂j Ak = 0. (251)

The tensor ijk is fully antisymmetric, while the second derivative is symmet-
ric and therefore their contraction vanishes.
From Eq. 239 we obtain:

~ i = Ji
~ × B)
(∇
0 c2
Ji
; ijk ∂j Bk =
0 c2
Ji
; ijk ∂j klm ∂l Am =
0 c2
Ji
; ijk klm ∂j ∂l Am = . (252)
0 c2

55
The contraction of two epsilon antisymmetric tensors yields:
 
δ δ 
ijk klm = il im
 
= δil δjm − δim δjl (253)
δjl δjm
 
 

Thus, we have:
Ji
(δil δjm − δim δjl ) ∂j ∂l Am =
0 c2
Ji
∂i ∂j Aj − ∂j2 Ai = . (254)
0 c2
In full vector form:
~
~ − ∇(
∇2 A ~ ∇ ~ =− J .
~ · A) (255)
c2 0
This is a second order differential equation for the components of the vector
potential. Had it not been for the second term in the lhs, this equation
would become our familiar Poisson differential equation from electrostatics.
In fact, we can choose to eliminate this term by exploiting a property of the
vector potential, known as gauge invariance. Consider two vector potential
functions A~ and A~ 0 which differ from each other by the gradient of a scalar
function f :
~0 = A
A ~ + ∇f
~ (256)
These two vector potentials are physically equivalent and give rise to the
same magnetic field
~ ×A
∇ ~0 = ∇
~ ×A ~ = B.
~ (257)
Indeed:
(∇ ~ × ∇f
~ )i = ijk ∂j ∂k f = 0. (258)
~ 0 satisfies the same differential equation 255. Indeed,
It is easy to verify that A
under a “gauge transformation”:
~→A
A ~0 = A
~ + ∇f,
~ (259)
we have
~ 0 − ∇(
∇2 A ~ ∇
~ ·A
~ 0)
= ∇ 2 (A~ + ∇f
~ ) − ∇[~ ∇
~ · (A
~ + ∇f
~ )]
~ − ∇(
= ∇2 A ~ ∇~ · A)
~
J~
=− 2 . (260)
c 0

56
We therefore have the liberty to add any gradient to the vector potential
without changing the physics. This freedom that we have is called a gauge
choice. We can exercise this choice to simplify the differential equation for
the potential. For example, assume a potential A~ 0 with

~ ·A
∇ ~ 0 6= 0. (261)

Now, let us find a solution of the Poisson equation:

~ ·A
∇2 f = −∇ ~ 0. (262)

The solution is our familiar:

1 Z 3 (∇ ~ ·A ~ 0 )(~x)
f (~r) = d ~x (263)
4π |~r − ~x|

Then for
~ ·A
1 ~ Z 3 (∇ ~ 0 )(~x)
~ ~ 0 ~ ~ 0
A = A + ∇f = A + ∇ d ~x
4π |~r − ~x|
we obtain
~ ·A
∇ ~ = 0. (264)
The “gauge” for which the above happens is called the “Coulomb” gauge. In
that gauge, the vector potential satisfies a Poisson equation:

~=− J~
∇2 A . (265)
c2 0
which can be solved with the same techniques as we have developed in elec-
trostatics. If we know all currents in all space, we can write the solution
as:
1 Z 3 J(~ ~ y)
~
A(~x) = d ~
y . (266)
4π0 c2 |~x − ~y |

4.5 Wires of a small thickness

A common situation is the calculation of the vector potential A ~ and the
~ =∇
corresponding magnetic field B ~ ×A ~ due to electric currents circulating
in thin wires.

57
The integral of Eq. 266

1 Z 3 0 J(~ ~ x0 )
~ x) =
A(~ d ~
x . (267)
4π0 c2 |~x − ~x0 |
receives zero contributions from points outside the wires, where the current
density is zero. We therefore have

1 Z ~ x0 )
~ x) = 3 0 J(~
A(~ d ~
x . (268)
4π0 c2 in wires |~x − ~x0 |
Let’s assume that the thickness of the wires is small and that the density
~ x0 ) of the current as well as the distance |~x − ~x0 | are practically constant
J(~
on a cross-section S~ of the wire. We can then integrate over the transverse
~ obtaining:
directions dS,

1 I ~ x0 )
I(~
~ x) =
A(~ dl(~
x 0
) . (269)
4π0 c2 wire |~x − ~x0 |

In the above, l(~x0 ) parameterises the curve of the wire. I~ is a vector with a
magnitude equal to the current intensity
Z
I(~x) = ~ · J(~
dS ~ x) (270)

and direction the one of the current. We can now define a vector d~l which
is tangential to the loop. Usually, we take d~l(~x) to circulate anti-clockwise.
~ x) at a point ~x is either parallel or anti-parallel
The current intensity vector I(~
to d~l(~x):
dˆl(~x) = ±I(~
ˆ x). (271)
Assuming an anti-clockwise steady current we have:
I I 1
~ r) =
A(~ 2
d~l(~x) (272)
4π0 c wire |~r − ~x|

58
For a clock-wise current one should remember to include a minus sign or to
change the direction convention for d~l.
Let us now see an application of the above result. Consider a very long
wire along the x−axis with current intensity I.

Since the current flows only on the x−axis the vector potential in the y, z
axes is zero:
Ay = Az = 0. (273)
The x−component of the vector potential is:
I Z∆ dx
Ax = 2
√
4π0 c −∆ r2 + x2
q 
r2
I 1+ ∆2
+1 I h 2 2
i
= log  q  ≈ − log r + log(4∆ ) (274)
4π0 c2 1+ r2
−1 4π0 c2
∆2

where we have taken the limit ∆ → ∞. Notice that the expression has
a logarithmic singularity in this limit. However, this is not of any physical
consequence. The magnetic field can be computed from the curl of the vector
potential, where this infinite constant drops out upon differentiation. We
have:
Bi = ijk ∂j Ak , (275)
which gives for the components:

Bx = 0 (276)

59
 I z
By = − (277)
2π0 c r2
2

I y
Bz = + (278)
2π0 c r2
2

(279)
The magnitude of the magnetic field is
q I 1
B= Bx2 + By2 + Bz2 = , (280)
2π0 c2 r
which is in agreement with the result obtained from Ampere’s law.

4.6 Magnetic dipole

Consider a steady current J~ circulating in a small region of space. We are
interested in computing the vector potential and the magnetic field in a
position ~r which is far from the current.
The vector potential is given by Eq. 266:
1 Z 3 J(~ ~ x)
~ r) =
A(~ d ~x (281)
4π0 c 2 |~r − ~x|
We could expand the inverse distance in the ratio xr by means of spherical
harmonics. Since, here, we are interested only in the leading terms of the
expansion we do it with a simpler method by means of a Taylor series:
x2
!
1 1 1 ~x · ~r
= 1 = + + O (282)
|~r − ~x| [r2 + x2 − 2~x · ~r] 2 r r3 r3
The first term of the rhs in Eq. 282 gives a zero contribution to the vector
potential of Eq. 281. Indeed, total derivatives of a localized current den-
sity integrated over all infinite volume should vanish (using the divergence
theorem) since the current density vanishes at infinity. We therefore have:
Z Z Z
0 = d3~x ∂i (xk Ji ) = d3~x [δik Ji + xk ∂i Ji ] = d3~x Jk
Z
; d3~x J~ = 0. (283)

In the above, we have used that

~ · J~ = ∂i Ji = − ∂ρ = 0.
∇
∂t
60
Similarly, we can prove that:
Z
0 = d3~x ∂i (xk Ji xl )
Z
; d3~x [Jk xl + Jl xk ] = 0. (284)

Now we are ready to calculate the contribution of the second term in the
inverse distance expansion to the vector potential:
1 Z
Ai ≈ 2 3
d3~x Ji xj rj
4π0 c r
rj Ji xj + Jj xi Ji xj − Jj xi
Z  
3
= d ~
x +
4π0 c2 r3 2 2
1 Z
= d3~x [Ji (xj rj ) − (Jj rj )xi ] (285)
8π0 c2 r3
In vector notation, we have

~≈ 1 Z
3
h
~ x · ~r) − (J~ · ~r)~x
i
A d ~
x J(~ (286)
8π0 c2 r3
The bracket in the integral is a double cross-product:
   
~r × ~x × J~ = ~x ~r · J~ − J~ (~r · ~x) . (287)

Therefore,
~= 1 m ~ × ~r
A , (288)
4π0 c r3
2

where the vector m~ depends on the geometric characteristics of the distribu-

tion and it is known as the magnetic moment:
1Z 3 ~
µ
~= d ~x (~x × J). (289)
2
As an example, consider the special case of a steady current circulating anti-
clockwise in a wire which lays on a plane. The magnetic moment becomes:
I
I
µ
~= ~x × d~l, (290)
2
where I is the intensity of the current. Notice that integrand is a surface
element

61
 ~ = 1 ~x × d~l.
dS (291)
2
Therefore, the magnetic moment is simply:

m ~
~ = I S, (292)

where Z
~=
S ~
dS,
the total area of the loop.
As a second example, consider the current created by a single charge q
and mass M moving along a closed loop. The current density is:

J~ = ρ~v = qδ(~x − ~r(t))~v . (293)

where
d~r
~v =
dt
the velocity of the charge. The magnetic moment is
1Z 3 q q ~
m
~ = d ~x ~x × J~ = ~r × ~v = L, (294)
2 2 2M
where
~ = ~r × (M~v )
L (295)
the angular momentum of the charge.

62
 Having computed the vector potential of a “magnetic dipole”, we can
compute the magnetic field as:
~ × ~r
~ ×m ~1 .
 
~ =∇
B ~ ×A
~=∇ = − ~ × m
∇ ~ × ∇ (296)
r3 r
Performing the differentiations and using the identity
1
∇2 = −4πδ(~r) = 0 ∀ r > 0, (297)
r
we find:
~ = m

~ ∇ ~ − 3r̂ (r̂ · m)
~ 1 = −m
 ~
B ~ ·∇ 3
. (298)
r r

4.6.1 Force on a “magnetic dipole” inside a magnetic field

~ The
Let’s place a “magnetic dipole” inside a homogeneous magnetic field B.
force acting on it is
Z Z
F~ = ~ x) =
d3 ~xρ(~x)~v × B(~ ~ x) × B(~
d3~x J(~ ~ x) (299)

If the current is contained in a small region of space, we can assume that the
magnetic field varies slowly within this region and approximate it with the
first few terms of a Taylor expansion:
 
~ x) ≈ B(0)
B(~ ~ ~ B(0)
+ ~x · ∇ ~ + ... (300)

The first term vanishes upon integration. The second term yields (exercise)

F~ ≈ ∇(m ~
~ · B). (301)

The potential energy associated with this force is

F~ = −∇U,
~ (302)

where
U = −m ~
~ · B. (303)
exercise: Prove that the torque is
Z
τ= ~x × dF~ = m ~
~ × B(0). (304)

63
4.7 Relativity of the electric and magnetic field
Consider a wire on the x−axis producing a magnetic field
1 I
B= (305)
2π0 c2 r
at a distance r due to a current with an intensity I. Let’s assume that
the induction electrons inside the wire have a velocity ~v . We now imagine
another electron outside the wire to move parallel to the wire at a distance
r from it with the same velocity ~v .

The force acting on the electron outside the wire is

1 qvI
F = qvB = . (306)
2π0 c2 r
The intensity I of the current is:

I = Sρ− v, (307)

where S is the cross-section of the wire and ρ− is the electron charge density
inside the wire. The force is then:
qS ρ− v 2
F = . (308)
2π0 r c2
We now change reference frame and choose one which moves with a rela-
tive velocity ~v along the wire. In the new frame the electron outside is static
and has zero velocity:
v 0 = 0. (309)
In this frame, the magnetic component of the force must vanish! This is a
paradox. It appears that the charge is accelerated in one frame and it is free

64
in another. The paradox is resolved due to Einstein’s special relativity and
the development of an equivalent electric force in the new frame.
In the original frame, the positive charge density inside the wire cancels
the negative charge density of the free electrons:

ρ = ρ+ + ρ− = 0. (310)

The total charge density in the moving frame is

ρ0 = ρ0+ + ρ0− . (311)

It turns out that it is not zero!

Consider N charges inside a volume V = SL which are at rest in a certain
reference frame.

The total charge is

qN = ρrest V = ρrest LS. (312)
In a reference frame where the charges are

moving with velocity ~v the same number of charges is included in the same
piece of material, which, however will appear contracted due to the laws of
special relativity:
s
v2
N q = ρmoving SLmoving = ρmoving SL 1 − . (313)
c2

65
We can therefore obtain a relation for charge densities in two different frames,
where in one the charges appear at rest and a frame where the charges appear
moving:
ρrest
ρmoving = q 2
. (314)
1 − vc2
For the electron and positive charge densities inside the wire we obtain:
ρ+
ρ0+ = q (315)
v2
1− c2

and s
ρ0− v2
ρ− = q ; ρ0− = ρ− 1 − . (316)
1− v2 c2
c2

Given that ρ+ = −ρ− we have:

v2
0 c2
ρ = −ρ− q (317)
v2
1− c2

This charge distribution spreads over an infinitely long wire and creates an
electric field:
1 ρ0
E0 = S . (318)
2π0 r
The force is:
v2
0 0 qS ρ− c2 F
F = qE = − q =q (319)
2π0 r 1 − 2 v 2 2
1 − v2
c c

Notice that these two forces cause the same physical effect in their re-
spective reference frames. Both are vertical to the wire and point towards it.
The momentum change in the vertical direction for the charge q is identical:
s
v 2 ∆t0
∆pT = F ∆t = F 0 1 − q = F 0 ∆t0 = ∆p0T . (320)
c2 1 − v22
c

From this discussion, we conclude that the magnetic and electric fields
can transform to each other by changing reference frame. In view of this
observation it is desired to find a unified description of the two field as com-
ponents of a unique electromagnetic field.

66
END of WEEK 4

67
5 Time varying electromagnetic fields
It is now time to consider the full Maxwell equations.

∇ ~ = ρ
~ ·E (321)
0
~ ·B
∇ ~ =0 (322)
~
∇~ ×E~ = − ∂B (323)
∂t
~ ~
~ ×B
c2 ∇ ~ = j + ∂E (324)
0 ∂t

5.1 Charge conservation

These equations imply charge conservation. Taking the divergence of Eq. 324
we obtain:
~ · J~ + ∂ 0 ∇ ~ · J~ + ∂ρ .
   
~ · ∇
c2 0 ∇ ~ ×B
~ =∇ ~ ;0=∇
~ ·E (325)
∂t ∂t
which is the continuity equation derived from charge conservation. Integrat-
ing over all space,
" #
~ · J~ + ∂ρ = ∂Quniverse
Z
3
0= d ~x ∇ (326)
∂t ∂t
we obtain that the total charge in the universe is a constant.

5.2 Vector and scalar potential

It is now time to find the properties of the scalar and vector potentials in the
general case of time-dependent electromagnetic fields. The equations

ρ ~=− J~
∇2 φ = − , ∇2 A (327)
0 c2 0
are not general and they are correct only for static fields.
The equation ∇~ ·B~ = 0 is generally valid and it can be satisfied by
introducing a vector potential as before:
~ =∇
B ~ × A.
~ (328)

68
Maxwell Eq. 323 becomes
 
~
∇ ~ =−∂B;∇
~ ×E ~ + ∂ A  = 0.
~ × E (329)
∂t ∂t

This is satisfied if we introduce a scalar potential

~
~ = − ∂ A − ∇φ.
E ~ (330)
∂t
Indeed,
~ × ∇φ
∇ ~ = ijk ∂j ∂k φ = 0. (331)
Substituting Eq. 330 and Eq. 328 into the Maxwell equation 324, we
obtain:

1 ∂2 ~
" # " #
− ∇ 2 ~
A + ~ ∇
∇ ~ + 1 ∂φ = J .
~ ·A (332)
c2 ∂t2 c2 ∂t c2 0

In this derivation, we have used the identity

 
~ × ∇
∇ ~ ×A
~ = ∇(
~ ∇~ · A)
~ − ∇2 A.
~ (333)

With the same substitutions, Maxwell equation 321 becomes

∂ ~ ~ ρ
∇2 φ − ∇·A= . (334)
∂t 0
Equivalently,

1 ∂2
" # !
∂ ~ ~ 1 ∂φ ρ
2 2
− ∇2 φ − ∇·A+ 2 = . (335)
c ∂t ∂t c ∂t 0

The differential operator on the left of Eqs 332-335 is known as the D’ Alem-
bert or “box” operator:
1 ∂2
2 ≡ 2 2 − ∇2 (336)
c ∂t

69
5.3 Gauge invariance
The electric and magnetic field remain invariant under a simultaneous tran-
formation, known as gauge transformations, of the scalar and vector poten-
tial:
∂
φ → φ0 = φ − f, (337)
∂t
A~→A ~0 = A
~ + ∇f,
~ (338)
where
f = f (~x, t)
an arbitrary function of space-time. Indeed:
~ 0 = −∂t A
E ~ 0 − ∇φ
~ 0 = −∂t A
~ − ∂t ∇f
~ − ∇φ
~ + ∇∂
~ tf
~ − ∇φ
= −∂t A ~ = E.~ (339)

For the magnetic field,

~0 = ∇
B ~ ×A
~0 = ∇
~ ×A
~+∇
~ × ∇f
~ =∇
~ ×A
~ = B.
~ (340)

Equations 332-335 also remain invariant under gauge transformations. This

can be verified easily observing that the combination:

~ 0 + 1 ∂t φ0 = ∇ ~ · f + 1 ∂t (φ − ∂t f )
 
~ ·A
∇ ~ · A~+∇
c2 c2
= ∇ ~ + 1 ∂t φ − 2f.
~ ·A (341)
c2
Given our freedom to modify our vector and scalar potential by choosing
the function f (~x, t), we have an opportunity to make a choice which simplifies
their differential equations. For example, we can choose a gauge f so that
the potentials A ~ L , φL in this gauge satisfy

∇ ~ L + 1 ∂t φL = 0.
~ ·A (342)
c2
This is called the “Lorentz gauge”. The subscript L is to remind us that this
equation is valid only for this specific gauge. Maxwell equations give for the
potentials in the Lorentz gauge the differential equations:
ρ
2φL = (343)
0

70
and
J~
2A
~L = . (344)
c2 0

exercise: Show that the D’Alembert operator is not invariant under Galileian
transformations of Newtonian mechanics but it is invariant under the Lorentz
transformations of special relativity.

5.4 Electromagnetic waves in empty space

Maxwell equations admit non-zero solutions for the electric and magnetic
field in places where there exist no currents and charges. This is of course
not a surprise since already in magnetostatics and electrostatics we have
found that the scalar and vector potentials extend over infinite distances
from their sources. However, now we are set to find more exciting solutions
of Maxwell equations which correspond to ripples of electromagnetic energy
propagating through empty space with a constant speed, the speed of light.
In empty space and in the Lorentz gauge, Maxwell equations take the
form:
2A ~ = 2φ = 0. (345)
The electric and magnetic fields are given by:
~ = −∇φ
E ~ − ∂t A,
~ ~ =∇
B ~ × A.
~ (346)

Acting on the above equations with the D’Alembert operator, we obtain:

2E
~ = 2B
~ = 0. (347)

In empty space, all fields satisfy the same equation:

2f = 0, ~ E,
f ∈ {φ, A, ~ B}.
~ (348)

A solution of this equation is:

f (~x, t) = f (η̂ · ~x − ct) (349)

with
η̂ 2 = 1. (350)

71
In this solution, f depends on a single combination

u = η̂ · ~x − ct. (351)

We can verify easily that f (u) is indeed a solution. We have

∂t f (u) = (∂t u)∂u f (u) = −c∂u f. (352)

For the second time derivative, we find:

∂t2 = c2 ∂u2 f. (353)

Similarly,
~ (u) = (∇u)∂
∇f ~ u f = η̂∂u f, (354)
and
∇2 f (u) = η̂ 2 ∂u2 f = ∂u2 f. (355)
Therefore
1 2
 
∂ − ∇2 f = ∂u2 f − ∂u2 f = 0.
c2 t
The solution f (η̂ · ~x − ct) is a travelling wave with a speed equal to the
speed of light c, moving along the direction of the unit vector η̂.

The greatest success of Maxwell’s theory of electrodynamics has been to

realize that electromagnetic energy (light) can propagate as electromagnetic
waves.

72
 For the electric and magnetic field, we write:
~ = êE(η̂ · ~x − ct),
E (356)

~ = b̂B(η̂ · ~x − ct),
B (357)
Then
~ ·E
0=∇ ~ = ∂i êi E(η̂k xk − ct) = êi η̂i ∂u E = 0 (358)
which leads to
ê · η̂ = 0. (359)
Similarly, from
~ ·B
∇ ~ =0

we conclude that also

b̂ · η̂ = 0. (360)
Therefore, the electric and magnetic field in an electromagnetic wave propa-
gating in empty space are vectors which are transverse to the direction ~η of
the propagation of the wave:
~ B
E, ~ ⊥ η̂

We now turn to the remaining Maxwell equations:

~ ×E
∇ ~ ; (η̂ × ~e) ∂u E = c∂u B b̂.
~ = −∂t B (361)

The magnetic field is therefore perpendicular to both the electric field and
the direction of the wave propagation.

For the magnitudes of the electric and magnetic fields we have the equation:

∂u E = c∂u B (362)

73
For dynamical fields (time-varying with no constant component) the electric
and magnetic fields are proportional to each other

E = cB. (363)

The last of Maxwell equations

~ ×B
c2 ∇ ~ = ∂t E
~ (364)

yields an equivalent identity to Eq. 361.

5.4.1 Spherical waves

In the last section we found plane wave solutions for Maxwell equations in
empty space. These are not the only wave solutions that there exist. One
can find a diversity of solutions depending on the sources and boundary
conditions that generate them, For example, if we are able to change the
charge and current density at one point in the entire space only, we will
generate an electromagnetic wave with spherical symmetry, i.e. no preferred
direction.

Maxwell equations give

2f = 0, ~ B,
∀f ∈ {A, ~ E}.
~ (365)

A wave solution with spherical symmetry is

f (~r, t) = f (r, t) (366)

and is is independent of the azimuthal and polar angles φ, θ. In spherical

coordinates, we have
1 2 1
2f (r, t) = 2
∂t f − ∂r2 (rf ) = 0, (367)
c r
74
which gives:
1 2
 
2
∂t − ∂r2 (rf ) = 0. (368)
c
A solution of this equation is:

rf = A(r − ct) + B(r + ct) (369)

and, equivalently,

A(r − ct) B(r + ct)

f (r, t) = + . (370)
r r
The first term corresponds to a spherical wave which propagates outwards.
The second term is a spherical wave that propagates inwards. It is very
hard to realize boundary conditions where a wave can surge at a periphery
simultaneously at a time t0 and allow this disturbance to propagate towards
a center. Typically, outwards propagation is realistic, where a disturbance
takes place at t0 at one and only point.

6 General solutions of Maxwell equations with

sources
We will now develop tools for finding solutions of Maxwell equations in the
~ ρ 6= 0). We have seen that in Lorentz gauge Maxwell
presence of sources (J,
equations for the scalar and vector potentials become:

ρ J~
2φ = , 2A
~= . (371)
0 c2 0

6.1 Green’s functions

To solve the above differential equations, we will use the technique of Green’s
functions. We seek functions

G (~r, t; ~r0 , t0 ) (372)

which satisfy:
2~r,t G(~r, t; ~r0 , t0 ) = δ(~r − ~r0 )δ(t − t0 ). (373)

75
Then a solution of Maxwell’s equation for the scalar potential is:
Z ∞
ρ(~r0 , t0 )
φ(~r, t) = d3~r0 dt0 G(~r, t; ~r0 , t0 ) (374)
−∞ 0
This is easy to verify by acting with the D’Alembert operator on both sides
of Eq. 374. To the above expression, we can add solutions φfree (~r, t) in free
space, as the ones we have found in the last chapter for electromagnetic
waves, which satisfy
2φfree (~r, t) = 0 (375)
Our general solution is then:
Z ∞
3 0 0 ρ(~r0 , t0 )
0 0
φ(~r, t) = φfree (~r, t) + d ~r dt G(~r, t; ~r , t ) . (376)
−∞ 0
The solution is fixed by boundary conditions.
Equivalently, the general solution for the vector potential is
Z ∞ ~ r0 , t0 )
J(~
~ r, t) = A
A(~ ~ free (~r, t) + d3~r0 dt0 G(~r, t; ~r0 , t0 ) . (377)
−∞ c2 0

76
END WEEK 5

77
6.2 Fourier transformations
In this section, we would like to recall the salient properties of Fourier trans-
formations. As an exercise, please prove that
Z ∞
dwe−iwt = 2πδ(t). (378)
−∞

(Hint: Use that

∂t Θ(t) = δ(t)
and that the integral
1 Z ∞
e−iwt
−lim+ dw
2πi →0 −∞ w + i
is a representation of the heavyside Θ function.) For a general smooth func-
tion f (t) we have that
Z ∞ Z ∞
1 Z∞ 0
f (t) = dt0 δ(t − t0 )f (t0 ) = dt0 dke−ik(t−t ) f (t0 ) (379)
−∞ −∞ 2π −∞
which leads to
Z ∞
dk Z ∞
dt0 0
f (t) = √ e−ikt · √ eikt f (t0 ) (380)
−∞ 2π −∞ 2π
We define the function:
Z ∞
dt0 0
f˜(k) = √ eikt f (t0 ) (381)
−∞ 2π
Then, we have Z ∞
dk
f˜(t) = √ e−ikt f˜(k) (382)
−∞ 2π
Eqs 381-382 define the Fourier transform of the function f and its inverse.
It is often beneficial to work with f˜(k) rather that f (t) in solving differential
equations. We will make a good use of Fourier transforms in finding Green’s
functions.

78
6.3 Fourier transformations and Green’s functions
We seek Green’s functions which are

G(~x, ~x0 ; t, t0 ) = G(~x − ~x0 , t − t0 ), (383)

anticipating that they depend on space-time differences rather than values

of absolute positions and time. Writing ~x − ~x0 = ∆~x and t − t0 = ∆t, the
Green’s function must satisfy:

2G(∆~x, ∆t) = δ(∆~x)δ(∆t). (384)

We substitute in the above the Fourier transform:

Z
d3~kdE −i(Ec∆t−~k·∆~x) ~
G(∆~x, ∆t) = e G̃(k, E). (385)
(2π)4

We then find:
Z
d3~kdE −i(Ect−~k·~x) h 2
i Z 3~
~k 2 G̃(~k, E) = c d kdE e−i(Ect−~k·~x)
e −E + (386)
(2π)4 (2π)4

We then find for the Fourier transform of the Green’s function:

c
G̃(~k, E) = − . (387)
E2 − ~k 2
This solution is a bit naive and it does not impose any physical boundary
condition to the Green’s function. In physical situation, we require that
the effect to the vector and scalar potentials φ(~x, t), A(~ ~ x, t) from the sources
ρ(~x0 , t0 ), ~j(~x0 , t0 ) takes place for times t > t0 . In other words, if we switch on a
source at a given prime t0 we expect physically that this source will cause an
effect to the potentials only at a later time and not earlier. Physical Green’s
function must then satisfy the boundary condition

G(∆~x, ∆t) = 0, for ∆t < 0. (388)

It is possible to implement the above by modifying the Fourier transformation

of the Green’s function. We write
Z
d3~kdE −i((E+iδ)c∆t−~k·∆~x) ~
G(∆~x, ∆t) = e G̃(k, E) (389)
(2π)4

79
where δ is a infinitesimally small positive variable. This Fourier transfor-
mation differs from the one of Eq. 385 in that the “energy” E integration
variable has a small positive imaginary part in the exponent. For ∆t → −∞
the Green’s function vanishes:

G(∆~x, ∆t) ∝ e∆tδ → e−∞δ → 0, (390)

which is in accordance with the physical requirement for the Green’s function
to vanish before the sources are turned on. Up to corrections of O(δ), we
now find for the Fourier transform of the Green’s function:
−c c 1 1
 
G̃(~k, E) = = − (391)
2 ~
(E + iδ) − k 2 2k E + k + iδ E − k + iδ
 
where k ≡ ~k . We can now compute the Green’s function as

c Z d3~kdE −i((E+iδ)c∆t−~k·∆~x) 1 1
 
G(∆~x, ∆t) = e −
2(2π)4 k E + k + iδ E − k + iδ
(392)
To perform the integrations we use spherical coordinates:

d3~k = k 2 dkdφd cos θ

~k · ∆~x = k∆x cos θ. (393)

After the angular φ, θ integrations we arrive at:

ic Z h
−ik∆x +ik∆x
i
G(∆~x, ∆t) = dk e − e
2(2π)3 ∆x
1 1
Z  
−i(E+iδ)c∆t
dE e − . (394)
E + k + iδ E − k + iδ
We will do the “energy” dE integration using Cauchy’s theorem. First,
we promote E to a complex variable:

E = Er + iEi .

The integration takes place over the horizontal axis of the (Er , Ei ) plane.
The time dependent phase factor behaves as:


 ∆t < 0, Ei → +∞,
e−ic∆t(Er +iEi ) = e−ic∆tEr eEi c∆t → 0 for or (395)
∆t > 0, Ei → −∞.



80
Therefore, we can close the contour of integration on the upper (Er , Ei )
half-plane for ∆t < 0 and on the lower half-plane for ∆t > 0. The upper
half-plane contains no poles and therefore the integral is zero for negative
∆t. We then have
G(∆~x, ∆t) = 0, if ∆t < 0. (396)
This is in agreement with what we have required for our Green’s function, i.e.
to vanish for times earlier than the times for which the sources are turned on.
The integral for ∆t > 0 does not vanish, since the lower half-plane contains
two poles at E = ±k − iδ. With residues theorem, we obtain:
Z ∞
c n
G(∆~x, ∆t > 0) = dk e−ik(c∆t−∆x) + e+ik(c∆t−∆x)
2(2π)2 ∆x 0
o
−e−ik(c∆t+∆x) − e+ik(c∆t+∆x)
Z ∞
c n
−ik(c∆t−∆x) +ik(c∆t+∆x)
o
= dk e − e
2(2π)2 ∆x −∞
(397)

Finally, the last integral yields δ−functions, according to Eq. 378. We then
find
c
G(∆~x, ∆t > 0) = [δ(c∆t − ∆x)) − δ(c∆t + ∆x))] (398)
4π |~x − ~x0 |

Since ∆t and ∆x are both positive, the argument of the second delta function
never vanishes,
δ(c∆t + ∆x)) = 0, for ∆x, ∆t > 0. (399)
Thus the Green’s function becomes
|~x − ~x0 |
!
1
G(~x − ~x0 , t − t0 ) = δ t − t0
− Θ(t > t0 ), (400)
4π |~x − ~x0 | c

where the Θ-function is +1 for t > t0 and 0 otherwise.

Substituting into Eq. 376 and assuming no prior electromagnetic fields
~ free in the absence of sources, we find
φfree , A
x0 |
 
|~
x−~
1 Z δ t − t0 − c
φ(~x, t) = d3~x0 dt0 ρ(~x0 , t0 ) Θ(t > t0 ) (401)
4π0 |~x − ~x0 |

81
Performing the t0 integration yields:
0
 
1 Z
3 0
ρ ~x0 , t − |~x−~
c
x|

φ(~x, t) = d ~x (402)
4π0 |~x − ~x0 |
We observe that for time-independent charge densities:
ρ(~x0 , ~t0 ) = ρ(~x0 )
Eq. 402 reproduces our known result from electrostatics. In the limit of an
infinite speed of light c → ∞, the potential becomes once again the known
potential from electrostatics:
1 Z 3 0 ρ (~x0 , t)
lim φ(~x, t) = d ~x . (403)
c→∞ 4π0 |~x − ~x0 |
If c were infinite, we would need simply Coulomb’s law in order to compute
the potential φ(~x, t) at a time t, integrating over the contributions of the
charge density ρ(~x, t) as it appears at the same time t. For a finite value of
c, the potential Φ(~x, t) is due to the charge density ρ(~x0 , t0 ) as it appears at
x0 |
an earlier time t − |~x−~
c
. We interpret the time delay (retardation)
|~x − ~x0 |
δt =
c
as the time required for the action of the source to propagate with the speed
of light, in the form of an electromagnetic wave, up to the point ~x where we
measure the potential.
~ x, t). We find:
Similarly, we can calculate the vector potential A(~
x0 |
 
1 Z J~ ~x0 , t − |~x−~
c
~ x, t) =
A(~ 3
d ~x0
(404)
4π0 c2 |~x − ~x0 |
The Green’s function of Eq. 400 is known as the retarded Green’s function
due to its property to account for the retardation due to the propagation of
the electromagnetic signals from one point to another with a finite speed.
It is useful to derive an alternative form for the Green’s function. We first
prove that for real a, b
δ(a2 − |b|2 )Θ(a > 0) = δ((a − |b|)(a + |b|))Θ(a > 0)
1
= [δ(a − |b|) + δ(a + |b|)] Θ(a > 0)
2|b|
1
= δ(a − |b|)Θ(a > 0) (405)
2|b|

82
 |~
x−~ x0 |
Applying the above for a = t − t0 and b = c
we can cast the retarded
Green’s function as:
|~x − ~x0 |2
!
0 0 1 0 2
G(~x − ~x , t − t ) = δ (t − t ) − 2
Θ(t > t0 ). (406)
2π c

6.4 Potential of a moving charge with a constant ve-

locity
As a first application of Eqs 402-404 we will compute the scalar and vector
potential of a point-like charge q, moving with a velocity ~v . The charge
density corresponding to the moving charge is

ρ(~x0 , t0 ) = qδ(~x0 − ~v t0 ). (407)

Using the above charge density and the retarded Green’s function in the form
of Eq. 406 we obtain for the scalar potential:

|~x − ~x0 |2
!
q Z 3 0 0 0 0 0 2
Φ(~x, t) = d ~x dt δ(~x − ~v t )δ (t − t ) − Θ(t > t0 )
2π0 c2
|~x − ~v t0 |2
!
q Z 0 0 2
= dt δ (t − t ) − 2
Θ(t > t0 ). (408)
2π0 c

We now need to find the zeros of the argument of the delta function. We can
write:
|~x − ~v t0 |2 (xk − vt0 )2 + x2⊥
0 = (t − t0 )2 − = t02
− 2t0
t + t2
− (409)
c2 c2
where we have decomposed
~x = ~xk + ~x⊥ (410)
into its parallel and perpendicular components to the velocity ~v . We now
define the “boosted” variables:
xk v
   
xb = γ xk − vt , tb = γ t − , (411)
c2
where
1
γ=q (412)
v2
1− c2

83
and we introduce the quantity
τ 2 = c2 t2 − (x2k + x2⊥ ) = c2 t2b − (x2b + x2⊥ ). (413)
Eq. 409 can then be cast in the form:
t02 t0 τ 2
− 2 + 2 = 0. (414)
γ2 γ c
The discriminant is
4(t2b − τ 2 /c2 ) 4rb2
∆= = , (415)
γ2 γ 2 c2
where we have defined the “boosted” distance
rb2 = x2b + x2⊥ . (416)
The solutions are:
rb
 
t± = γ tb ± . (417)
c
The delta function becomes:
|~x − ~v t0 |2
!
0 2 cγ
δ (t − t ) − 2
Θ(t > t0 ) = Θ(t > t0 )δ(t0 − t− ). (418)
c 2rb
It is now trivial to perform the t0 integration in Eq. 408, obtaining:
q γ
Φ(~x, t) = . (419)
4π0 rb
Explicitly,
q 1 1
Φ(~x, t) = q 1 . (420)
4π0 1 − v2 
 2 2
c2 xk −vt
+ x2⊥ 
 q  
 2
1− v2
c

Following the same steps, we find that the vector potential of a moving
charge is (exercise):
~ x, t) = ~v Φ(~x, t).
A(~ (421)
c2
In this derivation of the scalar and vector potential for a moving charge,
we see the “rise” of special relativity. In a reference frame where the charge
is at rest, the scalar and vector potentials are:
q 1 ~ x, t)

Φ(~x, t)|rest = i1 , A(~ = 0. (422)
4π0 2
h
2 2 rest
x k + x⊥

84
Comparing with the corresponding expressions in a frame where the charge is
moving, we obtain an excellent hint that space co-ordinates transform under
boosts according to Lorentz transformations,
 
xk → γ xk − vt , x⊥ → x⊥ , (423)
and not Galileian transformations.
How about the emergence of the factor γ and of a non-vanishing vector
potential in the frame where the charge is moving? These are hints that the
vector and a scalar potential are components of a single relativistic “four-
vector”: !
Φ µ
~ ≡A .
A

7 Special Relativity
Special relativity is based on the assumption that the laws of nature are the
same for inertial observers where their co-ordinates are related via Lorentz
transformations:
xµ → xµ0 = Λµν xν + ρµ . (424)
where
x0
 
 x1 
xµ =  (425)
 
x2

 
x3
a “four-vector” comprising space-time coordinates with x0 = ct, ρµ is a con-
stant four-vector and Λµν satisfies:
Λµρ Λνσ gµν = gρσ . (426)
The 4 × 4 matrix gµν is the so-called metric, defined as:


 +1, µ = ν = 0,
gµν = −1, µ = ν = 1, 2, 3 (427)
µ 6= ν


0,
In the above we have used Einstein’s summation convention. For example,
one would write explicitly
Λµν xν = Λµ0 x0 + Λµ1 x1 + Λµ2 x2 + Λµ3 x3 . (428)
This is a convention that we will use extensively from now on.

85
7.1 Proper time
Lorentz transformations leave invariant “proper-time” intervals. These are
defined as:
dτ 2 ≡ c2 dt2 − d~x2 = gµν dxµ dxν . (429)
Indeed, in a different reference frame we have from Eq. 424:
dxµ0 = Λµν dxν . (430)
A proper-time interval in the new frame is
dτ 02 = gµν dx0µ dx0ν
 
= gµν Λµρ dxρ (Λνσ dxσ )
 
= gµν Λµρ Λνσ dxρ dxσ
= gρσ dxρ dxσ = dτ 2 . (431)
As a consequence of the invariance of proper-time intervals, the speed of light
is the same in all inertial frames. Indeed, for light we have:
 
 d~
 x

  = c ; dτ 2 = c2 dt2 − d~x2 = 0 (432)
 dt 

In a new frame,
0
 
 d~
 x 
dτ = dτ = 0 ;
20 2
 0 =c (433)
 dt 

Lorentz transformations are the only non-singular transformations which

preserve proper-time intervals:
0
dτ 2 = dτ 2
; gρσ dxρ dxσ = gµν dx0µ dx0ν
∂xµ ∂xν ρ σ
; gρσ dxρ dxσ = gµν dx dx , (434)
∂xρ ∂xσ
concluding that:
∂xµ ∂xν
gρσ = gµν . (435)
∂xρ ∂xσ
Differentiating with dx , we obtain:
∂ 2 x0µ ∂x0ν ∂ 2 x0µ ∂x0ν
" #
0 = gµν + . (436)
∂x ∂xρ ∂xσ ∂x ∂xσ ∂xρ

86
To this, we add the same equation with  ↔ ρ and subtract the same equation
with  ↔ σ. We obtain:
∂ 2 x0µ ∂x0ν
0 = 2gµν (437)
∂x ∂xρ ∂xσ
Assuming that the transformation xµ → x0µ is a well behaved differentiable
function and that the inverse of the transformation also exists,

∂x0µ ∂xσ
= δµν , (438)
∂xσ ∂x0ν
we obtain that
∂ 2 x0µ
= 0. (439)
∂x ∂xρ
Therefore, the transformation xµ → x0µ ought to be linear:

x0µ = Λµ ν + ρµ . (440)

87
END OF WEEK 6

88
7.2 Subgroups of Lorentz transformations
The set of all Lorentz transformations

x0µ = Λµν xν + ρµ . gµν Λµρ Λνσ = gρσ (441)

form a group (exercise: prove it), which is known as the group of inhomo-
geneous Lorentz group or the Poincare’ group. The subset of transformations
with ρµ = 0 is known as the homogeneous Lorentz group.
From
gµν Λµρ Λνσ = gρσ
and for ρ = σ = 0, we have:
3
(Λ00 )2 − (Λi0 )2 = 1 ; (Λ00 )2 ≥ 1.
X
(442)
i=1

Also, in matrix form the definition of the Lorentz transformation becomes:

g = ΛT gΛ ; det g = det(ΛT gΛ) ; (det Λ)2 = 1. (443)

The subgroup of transformations with

det Λ = 1, Λ00 ≥ 1,

which contains the unity 1 = δνµ , is known as the proper group of Lorentz
transformations. All other transformations are known as improper Lorentz
transformations. It is impossible with a continuous change of parameters to
change
det Λ = 1 → det Λ = −1 or Λ00 ≥ 1 → Λ00 ≤ −1.
Improper Lorentz transformations involve either space-reflection (det Λ =
−1, Λ00 ≥ 1) or time-inversion (det Λ = 1, Λ00 ≤ −1) or both (det Λ =
−1, Λ00 ≤ −1).
Proper homogeneous or inhomogeneous Lorentz transformations have a
further subgroup: the group of rotations,

Λ00 = 1, Λi0 = Λ0i = 0, Λij = Rij , (444)

with
det R = 1, RT R = 1. (445)

89
Thus, for rotations and translations (xµ → x0µ = xµ + ρµ ) Lorentz transfor-
mations are no different than Galilei transformations.
A difference with Galilei transformations arises in boosts. Assume a refer-
ence frame O in which a certain particle appears at rest, and O0 a reference
frame where the particle appears to move with a velocity ~v . Space-time
intervals in the two frames are related via
dx0µ = Λµν dxν = Λµ0 cdt, (446)
given that d~x = 0 in the frame O. For µ = 0, this equation gives
dt0 = Λ00 dt. (447)
For µ = i = 1, 2, 3 we have:
dx0i = Λi0 cdt (448)
Dividing the two, we have
dx0i Λi0 vi 0
vi ≡ = c ; Λ i
0 = Λ . (449)
dt0 Λ00 c 0
From
gµν Λµρ Λνσ = gρσ
and for ρ = σ = 0, we have:
(Λ00 )2 − (Λi0 )2 = 1
!− 1
~v 2 2
; Λ00 =γ = 1− 2 . (450)
c
and thus
vi
Λi0 = γ. (451)
c
The remaining components are not determined uniquely by knowing the ve-
locity ~v of the particle. Indeed, two Lorentz transformations
Λµν and Λµρ Rρν
where R is a rotation, boost a particle to the same velocity. For coordinate
systems O and O0 with parallel axes we find that (exercise)
vivj
Λij = δji + 2 (γ − 1) (452)
v
and
vj
Λ0j = γ . (453)
c
90
7.3 Time dilation
Consider an inertial observer O which looks as a clock at rest. Two ticks of
the clock correspond to a space-time interval

d~x = 0, dt = ∆t. (454)

The proper time interval is

1
dτ = (c2 dt2 − d~x2 ) 2 = c∆t. (455)

A second observer sees the clock with velocity ~v . Two ticks of the clock
define a space-time interval

dt0 = ∆t0 , d~x0 = ~v dt0 . (456)

The proper-time interval in the new frame is:

v
0 2
u   s
1 u  d~
 x  ~v 2
dτ 0 = (c2 dt02 − d~x02 ) = c∆t
2
0t
1 −  0  = c∆t0 1 − . (457)
 cdt c2

The proper-time is invariant under the change of inertial reference frames.

Thus we conclude that
∆t
∆t0 = q 2
= γ∆t (458)
1 − ~vc2

7.4 Doppler effect

Take our clock to be a source of light with a frequency
2π
ω= .
∆t
For an observer where the light-source is moving with velocity ~v this time
interval is measured to be
dt0 = γ∆t.
In the same period, the distance of the observer from the light source increases
by
vr dt0

91
where vr is the component of the velocity of the light-source along the direc-
tion of sight of the observer. The time elapsing between the reception of two
successive light wave-fronts from the observer is

cdt0 = cdt0 + vr dt0 . (459)

The frequency measured by the observer is

q
2
0 2π 1 − vc2
ω = = ω. (460)
dt0 1 + vcr

If the light-source is moving along the line of sight, vr = v, we have

v
u 1 − vc
u
ω0 = t
v ω. (461)
1+ c

If the light-source moves away from the observer, vr > 0, the frequency
decreases and the light appears to be more red (red shift). If the source
moves towards the observer, the frequency increases (violet shift).
Exercise: Calculate the angle of the direction of motion of the light-source
with respect to the line of sight of the observer for which there is no shift in
the frequency.
For an application of the Doppler effect in cosmology, read about Hubble’s law.

7.5 Particle dynamics

How can we compute the force of a particle which moves with a relativistic
velocity ~v ? We should expect that our classical formulae from Newtonian
mechanics need to be modified. Nevertheless, Newtonian expressions for the
force should be valid if a particle is at rest. We can always change reference
frame with Lorentz transformations to bring a particle at rest and calculate
the change in its velocity for a small time interval using Newtonian mechanics.
However, we will need to perform these changes of reference frame at every
small increase of the velocity of the particle during its acceleration due to
the force.
In a more elegant solution to the problem, we define a relativistic force
acting on a particle as
d 2 xµ
f µ = mc2 2 , (462)
dτ
92
where m is the mass of a particle 2 . If the particle is at rest, the proper-time
interval dτ coincides with the common time-interval dt

dτ = cdt.

Therefore, in the rest frame of the particle, the“space”-components of the

force four-vector become

i d2 xi i
frest =m 2
= FNewton , for i = 1, 23, (463)
dt

where F~Newton is the force-vector as we know it from Newtonian mechanics.

The “time” component of the force four-vector vanishes:

0 d2 t
frest = mc = 0. (464)
dt2
Under a Lorentz transformation, f µ transforms as
2 0µ
d 2 xµ 2d x
2 µ ν µ
2 d (Λ ν x + ρ )
2 ν
2d x
f µ = mc2 → mc = mc = Λµ
ν mc (465)
dτ 2 dτ 2 dτ 2 dτ 2
Therefore,
f 0µ = Λµν f ν . (466)
The components of f µ transform under Lorentz transformations in exactly
the same way as the components of space-time coordinates. It is therefore a
four-vector as well.
For a specific transformation from the rest frame of a particle to a frame
where the particle moves with a velocity ~v , we have

f µ = Λµν (~v )frest

ν
. (467)

where, we have found that,

!− 1
~v 2 2 vi
Λ00 (~v )
=γ = 1− 2 , Λi0 (~v ) = Λ0i (~v ) = γ .
c c
vivj
Λij (~v ) = δji + 2 (γ − 1) (468)
v
2
With mass, we mean the mass of a particle as it is measured in its rest-frame. We will
refrain from using the “relativistic”, velocity dependent, mass.

93
Therefore, the force on a moving particle is:
 
~v F~Newton · ~v
f~ = F~Newton + (γ − 1) , (469)
v2
and
~v · F~Newton ~v
f0 = γ = · f~. (470)
c c
In Newtonian mechanics, if the force F~ is given, we can compute the
trajectory ~x(t) by solving the second order differential equation:

d2~x F~ (~x, t)
= . (471)
dt2 m
Similarly, in special relativity, when the relativistic force f µ is known, the
differential equation 462 can, in principle, be solved to give the space-time
coordinates as a function of the proper time τ :

xµ = xµ (τ ). (472)

To calculate the trajectory, we then need to calculate the proper-time in

terms of the time coordinate by inverting

x0 = x0 (τ ) ; τ = τ (x0 ), (473)

which we can use to cast the space components directly as functions of the
time-coordinate.
We should not forget a second constrain that must be satisfied for our
solutions xµ (τ ), namely
dxµ dxν
Ω ≡ gµν = 1. (474)
dτ dτ
We have for the derivative of Ω with respect to proper-time:

dΩ d2 xµ dxν 2 µ dx
ν
= 2gµν 2 = g µν f . (475)
dτ dτ dτ mc2 dτ
The rhs is a Lorentz invariant quantity. In a new frame,
dx0ν   (Λν dxσ )   dxσ dxσ
gµν f 0µ = gµν Λµρ f ρ σ
= gµν Λµρ Λνσ f ρ = gρσ f ρ .
dτ dτ dτ dτ
94
We are therefore allowed to compute dΩdτ
in any reference frame we wish. Let
us choose the rest frame of the particle, where

xµ = (ct, ~0), f µ = (0, F~Newton ).

We obtain:
0
!
dΩ 2 0 dx
= f − f~ · ~x = 0. (476)
dτ mc2 dτ
Therefore, the quantity Ω is always a constant:

Ω(τ ) = constant. (477)

If for some initial value τ0 we choose the constant to be one, we will have

Ω(τ ) = Ω(τ0 ) = 1, ∀ τ. (478)

Exercise: Calculate the trajectory of a particle on which the four-vector

force exerted is f µ = 0.

95
 END OF WEEK 7
(before Easter: only two hours)

96
7.6 Energy and momentum
We can define a relativistic four-vector analogue of momentum as
dxµ
pµ = mc (479)
dτ
We have that
 !2  21 #1
~v 2
"

2 2 2
1
2 d~x 2
cdt
dτ = c dt − d~x = cdt 1 −  = cdt 1 − = . (480)
cdt c2 γ

Thus, for the time-component (µ = 0) of the four-momentum we have

dx0
p0 = mc = mγc. (481)
dτ
For the space-components (µ = i = 1, 2, 3) we have
dxi dxi
pi = mc = mγ = mγv i . (482)
dτ dt
For small velocities, we can expand the factor γ as
#− 1
v2 1 v2 v4
" !
2
γ = 1− 2 ≈1+ 2 +O 4 . (483)
c 2c c
Therefore, for small velocities the space-components of the four-momentum
become the classical momentum,
pi ≈ mv i + . . . , (484)
while the time-component becomes
1
p0 ≈ mc + mv 2 + . . . (485)
2c
In the second term of the above expansion we recognize the kinetic energy
1
2
mv 2 of the particle. We then identify the relativistic energy of a particle
with
E = cp0 = mγc2 . (486)
Eliminating the velocity ~v from Eqs 482-486, we obtain the relation:
q
E= p~2 c2 + m2 c4 (487)

97
7.7 The inverse of a Lorentz transformation
Recall the metric matrix

gµν = diag(1, −1, −1, −1) (488)

We define an inverse
g µν : g µν gνρ = δ µν , (489)
where δ µν is the Kronecker delta. It is easy to verify that the inverse of the
metric is the metric itself:

g µν = gµν = diag(1, −1, −1, −1). (490)

Now consider a Lorentz transformation Λµν , which satisfies the identity:

Λµρ Λνσ gµν = gρσ . (491)

We can prove that the matrix

Λµν ≡ gµρ g νσ Λρσ (492)

is the inverse of Λµν . Indeed

Λµλ Λµν = gµρ g νσ Λρσ Λµλ = gσλ g νσ = δ λν . (493)

If Λµν is a velocity ~v boost transformation of Eq. 468, then

!− 1
~v 2 2 vi
Λ00 (~v )
=γ = 1− 2 , Λi0 (~v ) = Λ0i (~v ) = −γ .
c c
vivj
Λij (~v ) = δij + 2 (γ − 1) (494)
v
We, therefore have that the inverse of a boost is

Λµν (~v ) = Λµν (−~v ), (495)

as we also expect physically.

98
7.8 Vectors and Tensors
It is now time to give officially a definition for vectors in special relativity.
We call any set of four components which transform according to the rule:

V µ → V 0µ = Λµν V ν (496)

a contravariant vector. Contravariant vectors transform in the same way as

space-time coordinates xµ do under homogeneous Lorentz transformations.
Not all vectors transform as contravariant vectors. Consider the derivative
∂
∂xµ
. Under a Lorentz transformation, it transforms as:
∂ ∂ ∂xρ ∂
→ = . (497)
∂xµ ∂x0µ ∂x0µ ∂xρ
We have that
∂xρ ∂x0µ ∂xρ
! ! !
= δ ρν ; Λµν = δ ρν . (498)
∂x0µ ∂xν ∂x0µ
 
∂xρ
Therefore, ∂x0µ
is the inverse of a Lorentz transformation Λµν :

∂xν
= Λµν . (499)
∂x0µ
Substituting into Eq. 497, we find:
∂ ∂ ρ ∂
→ = Λµ . (500)
∂xµ ∂x0µ ∂xρ
We found that the derivative does not transform according to the Lorentz
transformation but according to its inverse. All vectors which transform with
the inverse Lorentz transformation:

Uµ = Λµν Uν , (501)

are called covariant vectors.

For every contravariant vector U µ there is a dual vector

Uµ = gµν U ν . (502)

We can invert the above equation multiplying with g ρµ ,

g ρµ Uµ = g ρµ gµν U ν = δ ρν U ν = U ρ . (503)

99
The dual vector Uµ is a covariant vector. Indeed, under a Lorentz transfor-
mation we have

Uµ → Uµ0 = gµν U 0ν = gµν Λνρ U ρ = gµν Λνρ g ρσ Uσ = Λµσ Uσ . (504)

The scalar product of a contravariant and a covariant vector

A · B ≡ Aµ Bµ = Aµ B µ = gµν Aµ B ν = g µν Aµ Bν (505)

is invariant under Lorentz transformations. Indeed,

A · B → A0 · B 0 = A0µ Bµ0 = Λµρ Aρ Λµσ Bσ = δ σρ Aρ Bσ = Aρ Bρ = A · B. (506)

Let us define for short:

∂
∂µ ≡ , (507)
∂xµ
and the dual contravariant vector:
∂
∂µ = = g µν ∂ν . (508)
∂xµ
The D’ Alembert operator is the scalar product:
1 ∂2
2 ≡ ∂ 2 ≡ ∂µ ∂ µ = − ∇2 . (509)
c2 ∂t2
Due to it being a scalar product, the D’Alembert operator is invariant under
Lorentz transformations.
Finally, we define a tensor with multiple “up” and/or “down” indices to
be an object
Tνµ11νµ22...... (510)
which transfroms as:

Tνµ11νµ22...... → Λµρ11 Λµρ22 . . . Λνσ11 Λνσ22 . . . Tσρ11σρ22...

...
(511)

7.9 Currents and densities

Consider a set of particles {n} with charged qn at positions ~rn (t). The charge
and current density are
X
ρ(~x, t) = qn δ(~x − ~rn (t)), (512)
n

100
 ~j(~x, t) =
X d~rn (t) X d~ x
qn δ(~x − ~rn (t)) = qn δ(~x − ~rn (t)), (513)
n dt n dt
In the above, we can set ~rn = ~x due to the delta function which vanishes for
~x 6= ~rn . Recall now that for the vector xµ = (ct, ~x), we have

dxµ
!
d d~x
= (ct, ~x) = c, . (514)
dt dt dt

We can then combine the charge and current densities into one object:
 
j µ ≡ cρ, ~j (515)

with
dxµ
j µ (~x, t) =
X
qn δ(~x − ~rn (t)), (516)
n dt
We can now cast j µ in a form which manifestly shows that it is a con-
travariant four-vector. First, we define a delta function in four dimensions
as
1
δ (xµ − y µ ) = δ(x0 − y 0 )δ(~x − ~y ) = δ(tx − ty )δ(~x − ~y ). (517)
c
Notice that the δ−function of a four-vector is a scalar. Under a Lorentz
transformation,

δ(U ν )
δ (U µ ) → δ (U 0µ ) = δ (Λµν U ν ) = = δ(U ν ). (518)
|det Λ|

With the use of the delta-function, we can write the current-density four-
vector of Eq. 516 as an integral
µ
Z
0 dx 0
Z
dxµ
µ
dt0 δ(xµ − rnµ (t)),
X X
j (~x, t) = qn dt δ(~x −~rn (t))δ(t − t) = c qn
n dt0 n dt0
(519)
where
xµ ≡ (ct0 , ~x), rnµ (t) ≡ (ct, ~rn (t)).
Now, we change integration variables from t0 → τ . Recall that
dτ
dt0 = γ. (520)
c

101
Then, we have
Z
dxµ
j µ (~x, t) = c δ(xµ − rnµ (τ )),
X
qn dτ (521)
n dτ
µ
which shows manifestly that j µ transforms as dx dτ
and it is therefore a con-
travariant four-vector.
We have already shown that charge-conservation implies the continuity
equation
∂ρ ~ ~
+ ∇ · j = 0. (522)
∂t
In relativistic notation, the continuity equation takes the more elegant form:
∂µ j µ = 0 (523)

7.10 Energy-Momentum tensor

Consider a collection of particles {n} at positions ~rn (t). The energy density
is: X
energy density = En (t)δ(~x − ~rn (t)) (524)
n
Changes in the energy density result to a “energy-current density”:
X d~rn
energy current density = En (t) δ(~x − ~rn (t)). (525)
n dt
As with the charge density and its current-density of the last section, we can
combine the two together into a single object:
X drnν
En (t) δ(~x − ~rn (t)). (526)
n dt
Similarly to the energy, the density/current-density for the components of
the three-dimensional momentum are:
drν
pin (t) n δ(~x − ~rn (t)), i = 1, 2, 3.
X
(527)
n dt
Collectively, we can form the so called “energy-momentum tensor” which
encompasses the density and current-density for all components of the four-
momentum:
X drnν
T µν ≡ pµn δ(~x − ~rn (t)) (528)
n dt

102
or, in the equivalent form:
XZ dxνn
T µν = dτ pµn δ(xρ − rnρ (τ )). (529)
n dτ
From the last equation, we can see manifestly that this object is justifiably
called a tensor since it transforms as the product of two four-vectors under
Lorentz transformations:
T 0µν = Λµρ Λνσ T ρσ (530)
The energy momentum tensor is symmetric:
T µν = T νµ . (531)
To verify this, we recall that
drnν drν drν
pνn = mn = mn γ n = En n . (532)
dτ dt dt
The energy momentum tensor takes then the explicitly symmetric form:
X pµ ν
n pn
T µν ≡ δ(~x − ~rn (t)). (533)
n En
For the charge-density four-vector, we have found a continuity identity
∂µ j µ = 0.
This was a consequence of the conservation of charge. If the total energy and
momentum of the system of particles is conserved, we anticipate a similar
continuity identity for the energy-momentum tensor:
∂ν T µν = 0.
We have:
drni ∂
∂i T µi = pµn
X
δ (~x − ~rn )
n dt ∂xi
X drni ∂
= − pµn δ (~x − ~rn )
n dt ∂rni
∂
pµn δ (~x − ~rn )
X
= −
n ∂t
∂ X ∂pµ n
pµn δ (~x − ~rn ) +
X
= − δ (~x − ~rn )
∂t n n ∂t
∂ X ∂pµ n
= − T µ0 + δ (~x − ~rn ) . (534)
∂t n ∂t

103
In the above we have used that
Z ∞ Z ∞
∂x δ(x − y) = 2π∂x dwe−iw(x−y) = −2iπ dw we−iw(x−y)
−∞ −∞
Z ∞
−iw(x−y)
= −2π∂y dwe = −∂x δ(x − y). (535)
−∞

We have therefore arrived to the equation

∂ν T µν = Gµ , (536)
where
∂pµn X ∂τ
Gµ = fnµ (t)δ (~x − ~rn )
X
δ (~x − ~rn ) = (537)
n ∂t n ∂t
is the “density of force”.
For free particles, where the energy and momentum of all particles sepa-
rately is conserved pµn = constant, the energy-momentum tensor satisfies the
continuity equation:
∂ν T µν = 0. (538)
The energy momentum tensor is also conserved if the particles interact only
at the points where they collide with each other. In that case, the force
density is
∂pµn
Gµ =
X
δ (~x − ~rn )
n ∂t
X d X µ
= δ(~x − ~xcoll (t)) p (t). (539)
coll. dt n∈coll. n
We have grouped the sum over all particles contributing to the force density
according to the collision points that they meet at. In each collision point,
the sum of the momenta of the colliding particles is conserved
d X µ
p (t) = 0 ; ∂ν T µν = 0. (540)
dt n∈coll. n
If the continuity equation is satisfied, then
0 = ∂ν T µν
; 0 = ∂0 TZ µ0 + ∂i T µi Z
; 0 = ∂0 d3~xT µ0 + d3~x∂i T µi
Z
0 = ∂0 d3~xT µ0 (541)

104
which means that the vector
Z
µ
P ≡ d3~xT µ0 = constant (542)

is conserved. We find that the conserved vector is the sum of all the total
four-momentum of the particles in the distribution:
XZ
µ
d3~xpµn δ(~x − ~rn (t)) = pµn .
X
P = (543)
n n

105
END of WEEK 8

106
8 Relativistic formulation of Electrodynam-
ics
From now on we will set 0 = c = 1. It will be easy to restore the full depen-
dence on these constants with dimensional analysis when necessary. Maxwell
equations are

~ ·E
∇ ~ = ρ (544)
~
~ ×E
∇ ~ + ∂B = 0 (545)
∂t
~ ~
∇·B =0 (546)
~
~ ×B
∇ ~ − ∂ E = ~j (547)
∂t
(548)

We construct an “electromagnetic field tensor” F µν from the components of

~ ≡ (E 1 , E 2 , E 3 ) and magnetic field B
the electric E ~ ≡ (B 1 , B 2 , B 3 ) as:

F 0i = −E i , F ij = −ijk B k , F µν = −F νµ . (549)

(we use 123 = +1)

Explicitly,
0 −E 1 −E 2 −E 3
 
 E1 0 −B 3 B 2 
F µν =  (550)
 
2 3 1 
 E B 0 −B 
3 2 1
E −B B 0
Conversely, If the tensor F µν is given, we can obtain the magnetic field via
(exercise):
1
B i = − ijk F jk . (551)
2

107
Exercise: What is the covariant tensor Fµν in matrix form?
Exercise: Write the matrix:

F̃µν = µνρσ F ρσ . (552)

Exercise: Prove that:

1 ~2 − B
E ~2
− Fµν F µν = , ~ · B.
µνρσ F µν F ρσ = 8E ~ (553)
4 2
As we have already noted, the charge and current densities form a four-
vector:
j µ = (ρ, ~j).
We can observe that two of Maxwell equations can be combined into one:

∂ν F νµ = j µ . (554)

Indeed, for µ = 0, we have

∂0 F 00 + ∂i F i0 = j 0 ; ∇
~ ·E
~ = ρ.

For µ = i = 1, 2, 3 we have
~
∂0 F 0i + ∂j F ji = j i ; −∂0 E i + ijk ∂j B k = j i ; ∇ ~ − ∂ E = ~j.
~ ×B
∂t
The remaining two Maxwell equations tell us, as we have found earlier, that
we can derive the electric and magnetic fields by means of the scalar and
~ via
vector potential φ, A,
~
E ~ − ∂A ,
~ = −∇φ ~ =∇
B ~ × A.
~ (555)
∂t
We can combine the scalar and vector potential into a single four-vector:
~ = (φ, A1 , A2 , A3 ).
Aµ ≡ (φ, A) (556)

Then, the above equations take the elegant form:

F µν = ∂ µ Aν − ∂ ν Aµ . (557)

108
Indeed, for µ = 0, ν = i = 1, 2, 3 we have
~
~ = ∂ A + ∇φ
F 0i = ∂ 0 Ai − ∂ i A0 ; −E ~
∂t
For µ = i, ν = j, i, j = 1, 2, 3 we have
1
F ij = ∂ i Aj − ∂ j Ai ; − ijk F ij = ijk ∂i Aj ; B
~ =∇
~ × A.
~
2
It is now straightforward to prove the following identities (exercise):

µνρσ ∂ν Fρσ = 0 (558)

∂µ Fνρ + ∂ν Fρµ + ∂ρ Fµν = 0. (559)

Substituting Eq. 557 into Eq. 554 we find

∂ 2 Aν − ∂ ν (∂µ Aµ ) = j ν . (560)

Explicitly, for ν = 0 and ν = i = 1, 2, 3 we recover the known differential

equations for the scalar and vector potentials respectively:
!
∂ ~ ~ ∂φ
2φ − ∇·A+ = ρ, (561)
∂t ∂t

" #
2A
~+∇
~ ∇ ~ + ∂φ = J.
~ ·A ~ (562)
∂t
The property of gauge invariance becomes more elegant as well in rela-
tivistic notation. The gauge transformations of the vector and scalar poten-
tials which leave Maxwell equations invariant are written as:

Aµ → A0µ = Aµ + ∂µ χ, (563)

where χ is a scalar function. Indeed, under a gauge transformation the

electromagnetic field tensor remains invariant:

Fµν → ∂µ (Aν + ∂ν χ) − ∂ν (Aµ + ∂µ χ) = ∂µ Aν − ∂ν Aµ = Fµν . (564)

The Lorentz gauge-fixing condition

∇ ~ + ∂φ = 0
~ ·A (565)
∂t
109
can be written simply as
∂µ Aµ = 0. (566)
In the Lorentz gauge, Maxwell equations for the four-vector potential
become:
∂ 2 Aµ = j µ . (567)
Earlier, we have found that the vector and scalar potential can be computed
via Z ∞
φ(~r, t) = d3~r0 dt0 G (~r − ~r0 ; t − t0 ) ρ(~r0 , t0 ) (568)
−∞
and Z ∞
~ r, t) =
A(~ d3~r0 dt0 G (~r − ~r0 ; t − t0 ) ~j(~r0 , t0 ) (569)
−∞
where the Green’s function can be written in the form:
1  2 
G(∆~r, ∆t) = δ ∆t − ∆~r2 Θ(∆t > 0). (570)
2π
In relativistic notation, the solutions for the four-vector potential take the
form:
1 Z 4 0 ν 0µ  0µ   
Aν (xµ ) = d x j (x )δ (x − xµ )2 Θ x0 > x00 . (571)
2π
The electromagnetic force acting on a particle with a charge q is:
dxν
f µ = qF µν . (572)
dτ
Indeed, in the rest frame of the particle dτ = dt, d~x = 0, we get
µ
frest = qF µ0 ; frest
0
= 0, f~rest = q E.
~ (573)
Since our expression for the electromagnetic force is correct in one reference
frame, it should hold in every inertial reference frame due to it being written
as an equation of four-vectors. In a frame where the charge is moving with
a velocity ~v , the three-dimensional force is:
dx0 dxj  
f i = qF i0 − qF ij = qγ E i + ijk B k v j
 dτ  dτ
~
; f = qγ E + ~v × B .
~ ~ (574)
Recall that
d~p d~p
f~ = =γ . (575)
dτ dt
We therefore have recovered our familiar expression for Lorentz’ force:
d~p 
~ + ~v × B

~ .
=q E (576)
dt
110
8.1 Energy-Momentum Tensor in the presence of an
electromagnetic field
Consider a number of charges qn which interact via the electromagnetic field.
The energy-momentum tensor is not conserved:
∂ν T µν = Gµ (577)
where the force density is given by
∂τ µ
Gµ =
X
fn (t)δ (~x − ~rn )
n ∂t
X ∂τ ν
µ drn
= qn F ν δ (~x − ~rn )
n ∂t dτ
X drnν
= F µν qn δ (~x − ~rn )
n dt
= F µν jν . (578)
Consider the tensor
µν 1
Tem ≡ F µρ F ρν + g µν Fρσ F ρσ . (579)
4
This tensor is explicitly
• symmetric
• gauge-invariant.
The components of the tensor are (exercise):
E~2 + B~2
00
Tem = , Tem 0i i0
= Tem ~ × B)
= (E ~ i (580)
2
Exercise: Find the remaining components.
We find
g µν
 
µν µρ ν ρσ
∂ν Tem = ∂ν F Fρ + Fρσ F
4
1
= F µρ ∂ν Fρν + (∂ ν F µρ ) Fρν + Fρσ ∂ µ F ρσ
2
1 1
= −F µρ jρ + (∂ ν F µρ − ∂ ρ F µν ) Fρν + Fρσ ∂ µ F ρσ
2 2
1
= −F µρ jρ + (∂ σ F µρ + ∂ µ F ρσ + ∂ ρ F σµ ) Fρσ (581)
2
111
which, due to Eq. 559, yields:
µν
∂ν Tem = −F µν jν . (582)
µν
Tem is purely electromagnetic. While neither T µν nor Tem
µν
satisfy a continuity
equation, but their sum
X pµ ν
n pn g µν
µν µν µν µρ
Θ ≡T + Tem = δ(~x − ~rn (t)) + F Fρν + Fρσ F ρσ . (583)
n En 4

does:
∂ν Θµν = 0. (584)
From the continuity equation, we obtain that there is a conserved four-vector:
Z Z 
µ 3 µ0
∂t P ≡ ∂t d ~xΘ =− d3~x∂i Θ0i = − Θ0i  = 0. (585)

∞

The conserved vector is:

Z Z
µ 3 µ0
pµn d3~xTem
µ0
X
P = d ~xΘ = + . (586)
n

The four-momentum of the charges

µ
pµn
X
Pcharges = (587)
n

is not conserved on its own. Some momentum

Z
µ
Pem = d3~xT µ0 (588)

is also carried by the electromagnetic field itself. This is not conserved either.
Momentum can be exchanged between the charges and the field, however this
is done in such a way so that the total momentum is always the same:
µ
P µ = Pcharges µ
+ Pem = constant. (589)

The time-component of the four-vector is the total energy. The energy stored
in the electromagnetic field is:
Z
Eem = d3~xTem
00
. (590)

112
Therefore, the energy density w of the electromagnetic field is:
~2 + B
E ~2
00
w = Tem = , EM field energy density. (591)
2
~ of the electromagnetic
Similarly, we find that the three-momentum density S
field is:
~ = T 0i = T i0 = (E
S ~ × B)
~ i EM field momentum density. (592)
em em

The vector
~≡E
S ~ × B,
~ (593)
is known as the Poynting vector.
Setting µ = 0 in Eq. 582 we have

∂0 T 00 + ∂i T i0 = −F 0i ji , (594)

and equivalently,
∂w ~ ~ ~ · ~j.
+ ∇ · S = −E (595)
∂t

8.2 Lagrangian formalism of Electrodynamics∗

In classical mechanics, we can obtain equations of motion by requiring that
an action remains stationary under small variations of the physical degrees
of freedom around their physical values. This is a convenient form to cast
laws of physics, since the action is a scalar quantity and it is guaranteed to
remain invariant at all reference frames. We have

δS = 0 (596)

where for a system with a finite number of degrees of freedom,

Z
S= Ldt (597)

and the Lagrangian is X

L= Li (598)
i

Li corresponds (for typical systems) to the difference of the kinetic and po-
tential energy.

113
 For the electromagnetic field, energy is stored in a continuous space.
Then, we should introduce a Lagrangian density L where
Li → d3~xL. (599)
The action becomes a four-dimensional integral over the Lagrangian density:
Z Z Z
3
S= dt d ~xL = d4 xL. (600)

The principle of minimum action reads:

Z
δ d4 xL = 0. (601)

8.3 Maxwell equations as Euler-Lagrange equations∗

For the electromagnetic field, we assume a Lagrangian density which is a
function of the four-vector potential Aµ and its first derivatives 3 :
L = L(Aµ , ∂ ν Aµ ). (602)
Then
( )
Z
δS δS
δS = 0 ; 4
dx µ
δAµ (x) + ν µ δ∂ ν Aµ (x) = 0 (603)
δA (x) δ∂ A (x)
Recall that
δf (x)
= δ(x − y) (604)
δf (y)
and
δG(f (x)) ∂G
= δ(x − y). (605)
δf (y) ∂f (x)
Thus,
0 = δS " #
∂L
Z
∂L
;0 = dx µ
4
δAµ (x) + ν µ
δ(∂ ν Aµ (x))
∂A ∂(∂ A )
" #
Z
∂L ∂L
;0 = 4
dx µ
δA (x) + ν µ
∂ (δA (x)) (606)
∂Aµ ∂(∂ ν Aµ )
3
We do not need that the Lagrangian density depends on second or higher order
derivatives since Maxwell equations are second order differential equations. From our
experience with classical mechanics, we know that higher derivatives in L lead to third or
higher order differential equations as equations of motion.

114
Performing integration by parts and dropping the total derivative term, we
obtain: " #
Z
4 ∂L ν ∂L
dx −∂ δAµ (x) = 0. (607)
∂Aµ ∂(∂ ν Aµ )
The above must be satisfied for arbitrary small variations δAµ (x). This can
happen if:
∂L ∂L
µ
− ∂ν = 0. (608)
∂A ∂(∂ ν Aµ )
Consider now the Lagrangian:
1
L = − Fµν F µν − Jν Aν (609)
4
Euler-Lagrange equations (Eq 608) give (exercise) Maxwell equations:

∂µ F µν = J ν . (610)

8.3.1 Gauge fixing

We can include a gauge choice within the Lagrangian formalism for the elec-
tromagnetic field. Consider the Lagrangian density
1 ξ
L = − Fµν F µν − Jν Aν − (∂µ Aµ )2 (611)
4 2
Euler-Lagrange equations yield:

∂ 2 Aµ − (1 − ξ)∂ µ (∂ · A) = J µ . (612)

Choosing ξ = 1 produces Maxwell equations in the Lorentz gauge.

If we promote ξ into a field, the corresponding Euler-Lagrange equation
yields:
∂L ∂L
− ∂ν = 0 ; (∂ · A)2 = 0 ; ∂ · A = 0, (613)
∂ξ ∂(∂ ν ξ)
which is the Lorentz gauge-fixing condition.

115
9 Radiation from moving charges
In this section we will study the electromagnetic field which is created by
moving charges.

9.1 The vector potential from a moving charge

A moving charge q has a current density
!
µ d~r
j (x) = qδ(~x − ~r(t)), q δ(~x − ~r(t)) (614)
dt
In an explicitly covariant form, we have found:
Z
µ
j =q dτ v µ δ ((x − r(τ ))ρ ) , (615)

where τ is the proper time and

drµ
vµ ≡ , (616)
dτ
the four-velocity of the charge.
The solution of Maxwell equations for the vector potential at a position
µ
x is: Z
Aµ (x) = d4 x0 Gret (x − x0 )j µ (x0 ), (617)
with the “retarded” Green’s function given by
1
Gret (x − x0 ) = δ((x − x0 )2 )Θ(x0 > x00 ). (618)
2π
Substituting the expression of the Green’s function into the potential integral
solution and performing the d4 x0 integration, we obtain:
q Z
Aµ (x) = dτ v µ (τ )δ((x − r(τ ))2 )Θ(x0 > r0 (τ )) (619)
2π
To perform the τ integration, we need to solve for the contraint imposed by
the delta function:
f (τ ) = (x − r(τ ))2 = 0 (620)
with the additional constraint

x0 > r0 (τ ). (621)

116
We have

0 = (x − r(τ ))2 = (x0 − r0 (τ ))2 − |~x − ~r(τ )|2

; x0 − r0 (τ ) = ± |~x − ~r(τ )| = ±R(τ ), (622)

where R(τ ) is the distance of the charge from the observation point and it
is a positive quantity. Due to the constraint x0 > r0 (τ ), we keep only the
solution originating from:

x0 = r0 (τ ) + |~x − ~r(τ )| (623)

Once we know the roots τi of Eq. 620, f (τi ) = 0, which satisfy Eq. 621,
we can use that
X δ(τ − τi )
δ [f (τ )] =  
 df  . (624)
i  dτ 
τ =τi

For relativistic particles with velocity v < c = 1 we find only one such
solution τ0 . We will demonstrate how this is done geometrically for a charge
moving with a constant velocity later. Let us now proceed to perform the τ
integration assuming that this solution τ0 has been found.
The derivative is
df (τ ) d d
= (x − r(τ ))2 = [gµν (x − r(τ ))µ (x − r(τ ))ν ]
dτ dτ dτ
ν
µ d ν µ dr (τ )
= 2gµν (x − r(τ )) (x − r(τ )) = −2gµν (x − r(τ ))
dτ dτ
= −2gµν (x − r(τ ))µ v ν (τ ) = −2v · (x − r(τ )). (625)

The vector potential becomes:

q Z δ(τ − τ0 )v µ (τ )
Aµ (x) = dτ (626)
4π |(x − r(τ )) · v(τ )|
As it can be seen in the rest frame of the charge, the quantity (x − r(τ )) · v(τ )
is positive definite and we can therefore drop the absolute value from the
denominator. Performing the τ integration is trivial, yielding

q vµ 
Aµ (x) = . (627)

4π v · (x − r(τ )) retarded


The subscript denotes that all quantities in this expression (distance and
velocity of the charge) must be computed at a retarded proper time τ = τ0

117
and not at the current time of the measurement. The expression of Eq. 627
is known as the Lienard-Wiechert potential. Notice that for a charge in its
rest frame, v µ = (1, ~0) it reproduces our known Coulomb potential:

q (1, ~0) q (1, ~0)

Aµ (x) = = (628)
4π x0 − r0 4π |~x − ~r|

9.1.1 Potential from a moving charge with a constant velocity

Let us now see how we can use Eq. 627 in order to recover the results of
Section 6.4 for the scalar and vector potential due to a charged particle
moving with a constant velocity ~v . At a time t, we assume the particle to be
at a position P which is at a distance

(P S) = |~v |t,

from the origin S. We are interested in knowing the potential Aµ (~x, t) at a

position O(~x) at the time t.

The electromagnetic wave which is measured at (~x, t) has been emitted at

an earlier time r0 when the charge was at a retarded position PR (~v r0 ). The
time needed for the signal to travel is

t − r0 (629)

118
and (in units of c = 1) the distance covered is
 
(Pr O) = R = ~x − ~r0  = t − r0 . (630)
 

In the same time, the charge advanced by a distance:

(PR P ) = |~v |(t − t0 ) = |~v |R. (631)
Let Q be the projection of the position P of the charge on the trajectory of
the light signal. Then,
(PR Q) = |~v |R cos θ = R~v · n̂, (632)
where n̂ points to the direction of travel of the light signal. Then,
(QO) = R(1 − ~v · n̂). (633)
Now consider the projection M of the observation point O on the trajectory
line of the charge. From Pythagoras’ theorem we have:
(P Q)2 + (QM )2 = (P M )2 + (M O)2 = r2 . (634)
The vertical distance of the charge and the observation point is:
(M O) ≡ r⊥ , (635)
and the parallel distance is
(M P ) ≡ rk = xk − vt. (636)
From Eq. 632 and a bit of trigonometry, we obtain (exercise)
 !2 1
√ xk − vt 2
2

R(1 − ~n · ~v ) = 1 − v2  √ + r⊥ (637)
1 − v2
In the lhs, we recognise the scalar product:
(x − r) · v|retarded = R(1, n̂) · γ(1, ~v ) = γR(1 − ~v · n̂). (638)
Thus, the potential is:
q vµ q γ(1, ~v ) q γ(1, ~v )
Aµ = = = 1 (639)
4π (x − r) · v 4π γR(1 − ~v · n̂) 4π xk −vt 2

2
2
√
1−v 2
+ r⊥

This is in agreement with the result found in Section 6.4.

119
9.2 The electromagnetic field tensor from a moving
charge
The electromagnetic field tensor
F µν = ∂ µ Aν − ∂ ν Aµ (640)
for a moving charge can be derived by differentiating the expression for the
Lienard-Wiechert potential. Due to the requirement of calculating the terms
of Eq. 627 at retarded positions, it is easier to work with the integral of
Eq. 619 and perform the integration after we have carried out the differenti-
ations. We define the four-vector:
Rµ ≡ xµ − rµ (τ ). (641)
The potential is then:
ν q Z
A (x) = dτ v ν (τ )δ(R2 )Θ(R0 > 0) (642)
2π
Its derivative is
µ ν q Z h i
∂ A = dτ v ν (τ )∂ µ δ(R2 )Θ(R0 > 0)
2π Z
q h i
= dτ v ν (τ ) Θ(R0 > 0)∂ µ δ(R2 ) + δ(R2 )∂ µ Θ(R0 > 0)
2π Z
q h i
= dτ v ν (τ ) Θ(R0 > 0)∂ µ δ(R2 ) + δ(R2 )δ(R0 )g µ0 . (643)
2π
The second term in the bracket vanishes,
δ(R2 )δ(R0 ) = δ((R0 )2 − R ~ 2 )δ(R0 ) = δ(−R
~ 2 )δ(R0 ) = 0, (644)
unless we are interested in calculating the electromagnetic field exactly at
a space-time point occupied by the electric charge q. For all other points,
where in fact classical electrodynamics should be valid, we have
µ ν q Z
∂ A = dτ v ν (τ )Θ(R0 > 0)∂ µ δ(R2 ). (645)
2π
The derivative on the delta-function is:
∂δ(F )
∂ µ δ(F ) = ∂ µ δ(R2 ) = ∂ µ F
∂F
∂ µ F ∂δ(F ) ∂ µ R2 ∂δ(R2 )
= ∂F = ∂R2
∂τ
∂τ ∂τ
∂τ
2Rµ ∂δ(R2 ) −Rµ ∂δ(R2 )
= = (646)
2R · ∂R
∂τ
∂τ R · v(τ ) ∂τ

120
Thus, the derivative of the potential becomes:

µ ν q Z Rµ v ν ∂δ(R2 )
∂ A =− dτ Θ(R0 > 0) . (647)
2π R · v ∂τ
The field tensor is then

F µν = ∂ µ Aν − ∂ ν Aµ
q Z Rµ v ν − Rν v µ ∂δ(R2 )
= − dτ Θ(R0 > 0)
2π R·v ∂τ
µ ν ν µ
q Z ∂ R v − R v
 
0 2
= − dτ Θ(R > 0) δ(R )
2π ∂τ R·v
q Z ∂ Rµ v ν − Rν v µ
 
+ dτ δ(R2 ) Θ(R0 > 0) (648)
2π ∂τ R·v
The first term requires the evaluation of the integrand in the boundaries of
integration and it is zero; δ(R2 ) has a solution inside the integration region
and not at the edges. Also, the term which is produced by differentiating
Θ(R0 ) vanishes for the same reason as we have seen earlier. Thus, the field
tensor becomes:
q Z ∂ Rµ v ν − Rν v µ
 
µν 2 0
F =+ dτ δ(R )Θ(R > 0) . (649)
2π ∂τ R·v
We can now perform the τ integration in exactly the same manner as in
Section 9.1. We obtain:
h i
∂ Rµ v ν −Rν v µ
q

∂τ R·v
F µν

=+  (650)
4π R·v 

retarded

Performing the differentiations, and using that:

drµ drµ dτ 2
vµ v µ = = 2 = 1, (651)
dτ dτ dτ
as well the definition of the acceleration four-vector:
dv µ
aµ ≡ (652)
dτ
we find that 
q Rµ Qν − Rν Qµ 
F µν = (653)
(R · v)3

4π 
retarded

121
where
Qν = v ν + aν R · v − v ν R · a. (654)
The acceleration four-vector is:
dv ν d
aν = = γ(1, ~v )
dτ dτ !
dγ d~v
= (1, ~v ) + γ 0,
dτ dτ
1 dγ ν
= v + aµT , (655)
γ dτ
with !
d~v ·
 
aµT = γ 0, =γ 2
0, ~v (656)
dτ
and
· d~v
~v ≡ (657)
dt
Question: Is aµT a four-vector?
Substituting into Eq. 654, we find that

Qν = v ν + aνT R · v − v ν R · aT . (658)

9.2.1 The electric and magnetic field of a moving charge

We can read the components of the electric and magnetic field from the
electromagnetic field tensor of Eq. 653. Let’s now write the distance vector
Rµ as
~
Rµ = |R|(1, n̂). (659)
Then, the components of the electric field are

i i0 q Ri Q0 − R0 Qi 
E =F =
4π (R · v)3 retarded


~ Q0 n̂ − Q
q|R| ~ 
;E
~ =  (660)
4π (R · v)3 
retarded

The components of the magnetic field are


1 1 q Rj Qk − Rk Qj 
B i = − ijk F jk = − ijk
(R · v)3

2 2 4π 
retarded

122
 
~ ijk n̂i Qk 
q|R|

= − 
4π (R · v)3 
 retarded
~ ~ 
~ = − q|R| n̂ × Q 
;B . (661)
4π (R · v)3 
retarded

We have therefore found explicitly that

~ = n̂ × E.
B ~ (662)

To compute the magnitude of the electric field, we need the products:

~
R · v = |R|γ(1 − n̂ · ~v ), (663)

and ·
~ 2 ~v ·n̂.
R · aT = −|R|γ (664)
Finally, we arrive at the result (exercise):

(1 − ~v 2 )
( )
q 1 ·

~ =
E (n̂ − ~v ) + n̂ × (n̂ − ~v )× ~v

4π(1 − n̂ · ~v )3 ~ 2 ~

|R| |R| 
retarded
(665)

9.3 Radiation from an accelerated charge in its rest

frame: Larmor’s formula
For a charge q which is accelerated but it is momentarily at rest, the electric
field is: ( )
q n̂ 1 ·

~ =
E + n̂ × n̂× ~v  (666)
4π |R| ~ 2 ~
|R| 
retarded

To study the flow of energy (radiation) we compute the Poynting vector:

~ 2 n̂ − (E
   
~=E
S ~ ×B
~ =E
~ × n̂ × E
~ = E ~ · n̂)E
~ (667)

~ we have:
Expanding in 1/|R|,

q2
!
· 2 1
 
~=
S n̂

n̂ × (n̂× ~
v )  +O . (668)
~ 2
16π 2 |R|
 ~ 3
|R|

123
Equivalently,
q2
!
· 1
~ = n̂
S 2
| ~v | sin Θ + O2
. (669)
~ 2
16π 2 |R| ~ 3
|R|
·
where Θ is the angle between n̂ and ~v :
· ·
~v ·n̂ = | ~v | cos Θ. (670)

The power (energy per unit time) dP emitted through a segment dA ~ of a

closed surface around the retarded position of the charge q is given by
dW ~·S
~
dP ≡ = dA (671)
dt
~ centered around the retarded
For a segment of a sphere with radius |R|
position of the charge, we have:
~ = n̂|R|
dA ~ 2 dΩ (672)

and therefore
q2 · 2 2
!
dP 1
= | ~v | sin Θ + O (673)
dΩ 16π 2 ~
|R|
The radiation power, i.e. the power which is radiated at infinitely large
distances, per solid angle dΩ is:

dPrad. q2 · 2 2
= | ~v | sin Θ (674)
dΩ 16π 2
The total power radiated at all solid angles surrounding the retarded
position of the charge is:
Z
dPrad.
Prad. = dΩ (675)
dΩ
Writing
dΩ = sin ΘdΘdφ (676)
and performing the angular integrations we obtain:

q2 2 · 2
Prad. = | ~v | . (677)
4π 3

124
Restoring the 0 , c constants, we have:
·
q 2 2 | ~v |2
Prad. = . (678)
4π0 3 c
This is known as Larmor’s formula. It gives the total power of radiation of an
accelerated charge when it has zero velocity. However, it can also be used as
a good approximation for the radiation power emitted by accelerated charges
with a small (non-relativistic) velocity |~v |  c.

9.4 Radiation from an accelerated charge with a rela-

tivistic velocity
The power of radiation
Energy
Power =
Time
is invariant under Lorentz transformations. If a charge +q is accelerated
and moving with a velocity ~v , we can make a Lorentz transformation to its
rest frame and use Larmor’s formula. Alternatively, we can write Larmor’s
formula in a manifestly Lorentz invariant form expressing 3-dimensional vec-
tors in the rest frame with the corresponding four-vectors. Larmor’s formula
reads: !2
q2 2 · 2 q2 2 d~p
Prad. = | ~v | = . (679)
4π 3 4π 3m2 dt
The relativistic force in the rest-frame is:
dpµ
!
d~p
= 0, (680)
dτ dt
and !2
dpµ dpµ d~p
=− . (681)
dτ dτ dt
We can then write:
q 2 2 dpµ dpµ
Prad. = − . (682)
4π 3m2 dτ dτ
In a reference frame where the particle moves with velocity ~v the above
expression gives:
q 2 2 6 · 2 
"  #
· 2

Prad. = γ | ~v | − ~v × ~v  . (683)
4π 3

125
9.4.1 Circular motion
Let us examine the case of a charge moving in a circular motion. The accel-
eration in that case is perpendicular to the velocity of the charge:
·
~v ⊥~v . (684)

Eq. 683 gives that

q2 2 4 · 2
Prad. = γ | ~v | . (685)
4π 3
A charged particle in a cyclical orbit will radiate constantly. This observa-
tion tells us that a planetary-type model for atoms does not work. According
to classical electrodynamics, electrons orbiting a nucleus would lose energy
from emitting radiation continuously falling to lower and lower orbits and
eventually collapsing on the nucleus. This does not happen due to quantum
mechanics and the uncertainty principle. The electrons of an atom occupy
discrete energy levels. In the ground state, which is the state of minimum
allowed energy, they never radiate.
We also notice in Eq. 685 that the radiated energy

Prad. ∼ γ 4

is larger for particles with larger velocity (larger γ).

In a circular collider, such as the Large Hadron Collider, one has to com-
pensate for the radiation loss (synchrotron radiation) due to the acceleration
of the charged particles which keeps them in a circular orbit.

9.4.2 Linear accelerators

Let us now assume that a charge is accelerated in a straight line:
·
~v k~v .

Eq. 683 gives that the radiated power is:

q2 2 6 · 2
Prad. = γ | ~v | . (686)
4π 3
For a relativistic charge the energy loss is

Prad. ∼ γ 6

126
9.5 Angular distribution of radiation from a linearly
accelerated relativistic charge
In this section, we will study the angular dependence of radiation from a
charge which is accelerated in a straight line. The electric field of an accel-
erated charge is:
 

q 1 ·
 1 
  
~ =
E n̂ × (n̂ − ~v )× ~v + O   2  (687)
3 ~
4π(1 − n̂ · ~v ) |R| ~ 
R  
retarded

and for ·
~v k~v ,
it becomes
 

q 1 ·
 1 
  
~ =
E n̂ × n̂× ~
v + O   2  (688)
~
4π(1 − n̂ · ~v )3 |R| ~ 
R  
retarded

The Poynting vector is:

· 2
 

n̂ × (n̂× ~
v )
q2
!
~ = n̂
 1
S +O . (689)
~ 2 (1 − n̂ · ~v )6
16π 2 |R| ~ 3
|R|

~
The power of radiation through a solid angle dΩ at a retarded distance |R|
is:
dPrad. q 2  · 2 sin2 Θ
 
= ~v (690)
dΩ 16π 2   (1 − v cos Θ)6
where Θ is the angle between the direction of the emitted radiation and the
velocity (or, equivalently, the acceleration) of the particle.
Let us now examine the limit where the velocity of the particle is very
close to the speed of light:
v ≈ c = 1.
In that limit, the denominator of Eq. 690 becomes large for a collinear emis-
sion of radiation to the direction of motion:
1 v≈1 1 Θ→0
6
−→ −→ ∞.
(1 − v cos Θ) (1 − cos Θ)6

127
Therefore, radiation tends to be collinear. To study better this double limit,
Θ → 0, v → 1, we observe that (exercise):
1
1 − v cos Θ ≈ 2
(1 + γ 2 Θ2 ) (691)
2γ
and
q2 (γΘ)2
 · 2
dPrad.  
≈ ~
v 
 . (692)
dΩ 16π 2 (1 + (γΘ)2 )6
The distribution (exercise: plot it) vanishes for small and large values of
γΘ. The maximum of the distribution is for values of
1
γΘ = √ = 0.4472 . . .
5
Therefore, we conclude that the radiation is emitted within a characteristic
angle:
1
Θ∼ . (693)
γ

128
10 Scattering
Consider an electric charge +q which is stricken by an incident electromag-
netic field with a certain frequency ω. An electromagnetic force is then
exerted on the charge which accelerates it. Consequently, the electric charge
will emit radiation. In this Section, we will study the characteristics of the
scattered radiation by the accelerated charge. We will distinguish two cases.
In one (Thomson scattering), the charge is free as it happens inside a con-
ductor or plasma. In the second (Rayleigh scattering), the charge is bound
inside an atom or a molecule.

10.1 Thomson scattering

Consider a free charge q in its rest frame ~v = 0 (or, with a very small non-
relativistic velocity v ≈ 0) stricken by an electromagnetic field:

~ = êE<ei(ωt−~k·~x)
E (694)

The force acting on the charge, placed at the position ~x = 0, is

F~ = q E
~ (695)

and the acceleration of the charge will be:

· F~ qE qE
~v = = ê <eiωt = ê cos(ωt) (696)
m m m
Due to its acceleration, the charge emits radiation with a power angular
distribution given by Larmor’s formula:
dPrad. q2 · 2 2 q4
= ~
v sin Θ = sin2 ΘE 2 cos2 (ωt). (697)
dΩ 16π 2 16π 2 m2
where Θ is the angle between the direction of the acceleration ê and the
direction n̂ of the emitted radiation. Averaging over time, we have
1ZT 1 2π
dt cos2 (ωt) = , T = . (698)
T 0 2 ω
and thus
q4
* +
dPrad.
= E 2 sin2 Θ. (699)
dΩ 32π 2 m2

129
The average incoming flux of radiation is the time-average of the incoming
Poynting vector:
D E E2
hSi = E 2 cos2 (ωt) = . (700)
2
The differential cross-section is defined as the ratio of the outgoing radi-
ation power per unit solid angle divided by the incoming flux of energy:
D E
dPrad.
dσ dΩ q4
≡ = sin2 Θ (701)
dΩ hSi 16π 2 m2
It has units of a surface. It is experimentally better to find the cross-section
distribution on the angle θ formed by the propagation direction k̂ of the
incident electromagnetic field and the direction of the outgoing radiation.

We can write:
sin2 Θ = 1 − sin2 θ cos2 (φ − ψ). (702)
Averaging over the azimuthal angle ψ of the polarisation vector, we have
1 Z 2π 1
dψ cos2 (φ − ψ) = . (703)
2π 0 2
Thus, the differential cross-section for unpolarised incoming photon beam is:
dσ q 4 1 + cos2 θ
= (704)
dΩ 16π 2 m2 2
130
The total cross-section is obtained by integrating the differential cross-section
over all solid angles:
Z
dσ q 4 8π 8π 2
σ= dΩ = = r (705)
dΩ 2 2
16π m 3 3 q
where
q2
rq ≡ (706)
4π0 mc2
For an electron, the Thomson radius is:

re ≈ 2.8210−13 cm. (707)

Notice that the cross-section is independent of the frequency of the incoming

electromagnetic wave. Our result for the cross-section receives significant
quantum corrections for frequencies h̄ω ∼ mc2 .

10.2 Rayleigh scattering

We now examine the scattering of an electromagnetic wave on an electric
charge which is bound in an atom or molecule. Then, the equation of motion
of the charge is given by solving
qE ·· ·
=x +γ x +ω02 x (708)
m
with a solution:
qE~
m
~x = . (709)
ω02 − ω2 + iγω
The acceleration is then:
··
~
· ·· qE ~
qE −ω 2
m
~v =~x= = (710)
ω02 − ω2 + iγω m ω02 − ω 2 + iγω
Unlike Thomson scattering where the electron is free, the acceleration is now
frequency dependent. We can write
· · −ω 2
~v =~v Thomson , (711)
ω02 − ω 2 + iγω

131
where
· qE~
~v Thomson = , (712)
m
the acceleration of the charge if it were free. The calculation of the cross-
section proceeds identically as in Thomson scattering. We find

ω4
σ = σThomson (713)
(ω 2 − ω02 )2 + γ 2 ω 2

with
8π 2
σThomson ≡ r . (714)
3 q
For ω  ω0 , the process is known as Rayleigh scattering and the cross-
section has a characteristic ω 4 frequency dependence:

ω4
σRayleigh = σThomson (715)
ω04
High frequencies are scattered more than lower frequencies. Rayleigh scat-
tering takes place in the atmosphere. Sunlight contains all visible light fre-
quencies, however the blue light (highest visible frequency) scatters the most.
This explains why the light we see away from the sun line of sight is more
blue while it is more yellow or red (during sunset/sunrise) when looking in
the direction of the sun. (see A video on blue sky )

132
11 Electrodynamics in a medium
In a macroscopic amount of matter we can identify two types of motions of
charged particles:
• fast currents within small distances, due to the motions of charges in
atoms and molecules,
• slow currents, which extend at distances much larger than the size of
atoms due to free electrons or ions.
In our macroscopic measurements of the currents and the electromagnetic
field, we are sensitive only to the slow variations and insensitive to the fast
ones taking place at the atomic level. We can then average over the currents
within the atoms. We can write
D E
µ µ
j µ ≈ jslow + jatomic (716)
(717)
, where hXi is the macroscopic average of the quantity X. It is not clear how
one should average over the atomic distances. However, we can parametrize
our ignorance by means of macroscopic properties of the material which re-
flect its atomic structure and we can measure experimentally.
We introduce an averaging of a function over some distances via:
Z
hF (~x, t)i = d3 ~y f (~y )F (~x − ~y , t). (718)

The weighting factor f (~y ) is unknown, but let’s assume that it is a well-
behaved smooth positive function with unit norm (for a probabilitic inter-
pretation). Z
d3 ~y f (~y ) = 1. (719)
The function f (~y ) should be peaked at ~y = 0 and fall off to zero very fast for
distances |~y | > ratom .

11.1 Average of the atomic charge density

0
The charge density jatomic = ρatomic corresponds to the charge density of
the charges inside molecules/atoms. Characteristically, the charges are con-
strained within regions of a small atomic size. We write
X
ρatomic = ρ(n) , (720)
n∈ atoms

133
where ρ(n) is the charge density of the n−th atom/molecule. Let’s assume
that the n−th atom/molecule consists of j charges qj :
X
ρ(n) = qj δ(~x − ~xn − ~xj ), (721)
j∈ (n)

where ~xn is the position of the atom/molecule and ~xj is the position of the
charge with respect to the “centre” of the molecule. The average is:
D E X
ρ(n) = hqj δ(~x − ~xn − ~xj )i
j∈(n)
Z
d3 ~y f (~y )δ(~x − ~xn − ~xj − ~y )
X
= qj
j∈(n)
X
= qj f (~x − ~xn − ~xj ). (722)
j∈(n)

We can now perform a Taylor expansion in

|~xj |
, (723)
|~x − ~xn |
the atomic scale over the macroscopic scale of observation. We obtain:
D E X h i
ρ(n) = ~ (~x − ~xn ) + . . .
qj f (~x − ~xn ) − (qj ~xj ) · ∇f
j∈(n)
   
D E
; ρ(n) ~ (~x − ~xn ) + . . .
X X
=  qj  f (~x − ~xn ) −  qj ~xj  · ∇f
j∈(n) j∈(n)

(724)

In the rhs, we recognise the total charge

X
qn = qj (725)
j∈(n)

and the dipole moment X

p~n = qj ~xj (726)
j∈(n)

of the n−th molecule. Thus:

D E
~ (~x − ~xn ) + . . .
ρ(n) = qn f (~x − ~xn ) − p~n · ∇f (727)

134
The gradient operator ∇ ~ differentiates with respect to the components of ~x
(it does not act on ~xn or ~xj We can then write the above in the form
D E
~ (~pn · f (~x − ~xn )) + . . .
ρ(n) = qn f (~x − ~xn ) − ∇ (728)

Summing over all atoms/molecules, we find that the average charge den-
sity in atoms/molecules can be written as:
XD E D E
hρatomic i ≡ ~ · P~ + . . .
ρ(n) = ρeff/atom − ∇ (729)
n

where D E X
ρeff/atom = qn f (~x − ~xn ) (730)
n∈atoms

is an effective charge distribution computed by considering atoms being

P
point-like with a total charge qn j∈(n) qj , and

P~ ≡
X
p~n · f (~x − ~xn ) (731)
n∈atoms

is the so called “polarization of the medium”. The medium polarization P~

sums together the average dipole moment of all atoms and it is the leading
contribution due to the non point-like geometry of the atoms.

11.2 Average of atomic current density

We will restrict our computation to currents with motions which are not
relativistic.
The current density in an atom/molecule can be written as
~j(n) (~x, t) =
X
qk (~vn + ~vk )δ(~x − ~xn − ~xk ). (732)
k∈n

where the velocity of the k-th charge in the n-th atom being as expressed as
the sum ~vn + ~vk of the velocity of the atom vn as a whole and vk the relative
velocity of the charge to the center of the atom 4 .
For the average, we have:
D E
~jn
X
= qk (~vn + ~vk )f (~x − ~xn − ~xk ) (733)
k
4
Relativistically, the velocity of the charge is not a simple sum as we have assumed
here.

135
For what it follows, it is important to remember that |~xk |  |~xn | and that
the velocity of the atom as a whole system is much smaller than the relative
velocities of its constituent charges |~vn |  |~vk |. Once again, we expand
~ (~x − ~xn ) + . . .
f (~x − ~xn − ~xk ) ≈ f (~x − ~xn ) − ~xk · ∇f (734)
This yields
D E
~jn
X
= qk~vk f (~x − ~xn )
k
~ (~x − ~xn )
X X
+ qk~vn f (~x − ~xn ) − qk~vk ~xk · ∇f
k k
 
+O x2k , xk vn , vn2 (735)
The term in the first line is parametrically the largest, as it depends on large
velocities within an atom. The terms in the second line are smaller as they
are suppressed by either a small distance within an atom xk or a slow velocity
of a whole atom vn . Finally, we ommit higher order terms suppressed by the
product of two such small parameters.
We now rewrite
!
X d X X d
qk~vk f (~x − ~xn ) = qk ~xk f (~x − ~xn ) − qk ~xk f (~x − ~xn )
k dt k k dt
d X 
~ (~x − ~xn )

= (~pn f (~x − ~xn )) + qk ~xk ~vn · ∇f
dt k
d
≈ (~pn f (~x − ~xn )) (736)
dt
Also we can write
X qk X qk
~ (~x − ~xn ) = ~ (~x − ~xn ) − ~ (~x − ~xn )
X
qk~vk ~xk · ∇f ~vk ~xk · ∇f ~xk ~vk · ∇f
k k 2 k 2
!
d X qk ~ (~x − ~xn )
+ ~xk ~xk · ∇f
dt k 2
X qk X qk
≈ ~ (~x − ~xn ) −
~vk ~xk · ∇f ~ (~x − ~xn )
~xk ~vk · ∇f
k 2 k 2
~
≈ −∇ × (m ~ n f (~x − ~xn )) (737)
where we have used that
   
~a × ~b × ~c = ~a ~b · ~c − ~b (~a · ~c) (738)

136
and recognised the magnetic moment of an atom,
X qj
m
~n= (~xj × ~vj ). (739)
j 2

Summing the averaged contributions from all atoms, we have

~
~j(n) = ~jeff/atomic + dP + ∇
D E XD E
~jatomic = ~ ×M
~, (740)
dt
where
~ =
X
M ~ n f (~x − ~xn ))
(m (741)
n
is the magnetization of the material and
~jeff/atomic =
X
qn~vn f (~x − ~xn ) (742)
n

11.3 Maxwell equations in a medium

We can now approximate the charge and current density in the medium by
µ µ µ µ
j µ = jfree + jatomic ≈ jfree + hjatomic i (743)
By direct substitution into Maxwell’sequations
µ
∂µ F µν ≈ jfree ν
+ hJatomic i, (744)
we obtain
~ ·B
∇ ~ = 0 (745)
~
∇ ~ = − ∂B ,
~ ×E (746)
∂t
 
~
∇ ~ + P  = ρeff
~ · E (747)
0 0
where
ρeff = ρfree + ρef f /atomic (748)
and
   
~ ~ ~
~ − M  = Jfree + 1 ∂ E
~ × B
∇ ~ + P . (749)
c2 0 c2 0 c2 ∂t 0
with
~jeff = ~jfree + ~jeff/atomic (750)

137
11.3.1 ~ and H
The D ~ field

Historically, Maxwell equations in matter have been often presented in terms

of the combinations:
D~ ≡ 0 E
~ + P~ (751)
and
~
~ ≡B
H ~− M . (752)
c2 0
In this notation, Maxwell equations in the medium take the form:
~ ·D
∇ ~ = ρ (753)
~
∇ ~ = − ∂B
~ ×E (754)
∂t
~ ~
∇·B = 0 (755)
~ ~
∇ ~ = J + 1 ∂D .
~ ×H (756)
c2 0 0 c2 ∂t

11.4 Maxwell equations inside a dielectric material

Assume a dielectric material, such as air or water, with no magnetisation
M~ = 0 but with the ability to acquire a polarisation P~ . Maxwell equations
take the form:
 
~
~ + P = ρ
~ · E
∇
0 0
~
~ = − ∂B
~ ×E
∇
∂t
~ ~
∇·B = 0 (757)
 
~ ~
~ = J + 1 ∂ E
~ ×B
∇ ~ + P (758)
0 c 2 2
c ∂t 0

To solve these equations, we need information about the polarisability P~

of the medium. P~ and E ~ are not truly independent. In fact, the larger
~ the more it will stretch the atoms and molecules in the
the electric field E
medium inducing a larger polarisation P~ . Ignoring non-linear effects, we can

138
write a phenomenological proportionality relation of the electric field and the
polarisability,
P~ = χ0 E.
~ (759)
where χ is known as the “electric susceptibility” of the dielectric medium.
Maxwell equations then become:

~ ·E
~ = ρ
∇
0 (1 + χ)
~
~ = − ∂B
~ ×E
∇
∂t
~ ~
∇·B = 0 (760)
~ ~
∇ ~ = J + 1 + χ ∂E .
~ ×B (761)
0 c2 c2 ∂t
Defining
c
cm = √ (762)
1+χ
and
 = (1 + χ)0 , (763)
Maxwell equations take the same form as in the vacuum,

~ = ρ
~ ·E
∇

~
~ = − ∂B
~ ×E
∇
∂t
~ ~
∇·B = 0 (764)
~ ~
∇ ~ = J + 1 ∂E .
~ ×B (765)
c2m c2m ∂t

with the electric permittivity and the speed of light constants replaced by

0 → , c → cm . (766)

11.5 Waves in a dielectric medium

Obviously, the solutions that we have found for Maxwell equations in the
vacuum are also solutions of Maxwell equations in a dielectric, as long as

139
we make the replacement of Eq. 766. Away from macroscopic charges and
currents, we have (as in the vacuum) that the electric and magnetic field
satisfy differential equations of the type:

1 ∂2
" #
~ x, t) = 0,
− ∇2 E(~ (767)
2
cm ∂t 2

which admit solutions

E ~ 0 E i(ωt−~k·~x) .
~ =E (768)
This represents a wave with frequency

ω = 2πf (769)

and phase-velocity
ω
vphase =   . (770)
~
k
Substituting the solution into the differential equation, we obtain

ω2 ω2
k2 = = (1 + χ) (771)
c2m c2

For the phase-velocity, we have

c
uphase = (772)
n
where q
n= 1+χ (773)
is the so-called refraction index and characterises the dielectric material.

11.6 A model for the dielectric susceptibility χ

To make further progress in understanding the properties of electromagnetic
waves inside dielectrics, we need to calculate χ = n2 − 1 which characterises
the material. Equivalently, we need to calculate the polarisability P~ of the
~
material when we subject it in an electric field E.
In a dielectric, positive and negative charges are bound together inside
atoms or molecules. To a good approximation, we can view the dielectric as a
collection of dipoles each with a positive and a negative charge (+q, −q). The

140
charges in a dipole are bound together due to their electric attraction but
they cannot come to arbitrarily close distances due to the laws of quantum
mechanics. Making a full account of the quantum effects which are at play
is beyond our scope. However, we can approximate phenomenologically the
resulting binding force as a harmonic oscillator. If a charge separates away
from the dipole, a force will pull it back and if it moves closer to the other
charge a force will push it away. As we have seen, electric charges which
accelerate radiate and lose energy. To account for this effect, we can also
include a friction or drag force. Let’s assume that the electric field is in the
x̂ direction. The force acting on a charge +q of a dipole is then
·· ·
F = qE = m(x +γ x +ω02 x). (774)
For a time varying electric field
E = E0 eiωt , (775)
we find solutions of the form:
x = x0 eiωt , (776)
with
qE0
m
x0 = . (777)
ω02 − ω2 + iωγ
Therefore,
q
~x = m ~
E, (778)
ω02 − ω2 + iωγ
and the electric dipole moment is
q2
p~ = q~x = m ~
E. (779)
ω02 − ω2 +
iωγ
Assuming a constant density of +q charges N , the polarisability of the di-
electric is:
N q2 1
P~ = N p~ = N q~x = 2
~
E. (780)
m ω0 − ω 2 + iωγ
The dielectric susceptibility is then
N q2 1
χ = n2 − 1 = 2
. (781)
m0 ω0 − ω 2 + iωγ
Equivalently, the square of the index of refraction is
N q2 1
n2 = 1 + 2
. (782)
m0 ω0 − ω 2 + iωγ

141
11.7 The complex index of refraction
We have found that the index of refraction is a complex number:

n = nR − inI (783)

A plane wave propagating in the dielectric is:

nR
E ~ 0 eiω[t−nk̂·~x] = E
~ =E ~ 0 eiω[t− c
k̂·~
x] e−ω ncI k̂·~x . (784)

The amplitude of the plane wave is

    nI
~  ~  −ω k̂·~
x
E  = E 0 e c (785)

and it decreases (for the physical case nI > 0) as the wave penetrates further
inside the material.

11.8 Waves in metals

Metals are conductors permitting electrons to move freely at large distances.
We can compute their index of refraction as a special limit of Eq. 782 by
setting
ω0 = 0.
This corresponds to a zero force for binding the charges to a fixed position
as it was the case for dielectrics. The index of refraction is then:
N q2 1
n2 = 1 + 2
. (786)
m0 −ω + iωγ
The density N can be obtained from macroscopic properties of the metal.
The constant γ is an intrinsic parameter of our model. To make contact with
reality and test our theory of wave propagation in metals, we must find a
way to relate γ to an experimentally measured quantity.
In our model, the equation of motion for the electrons in the metal can
be obtained by solving the differential equation:
·· ·
qE = m(x +γ x) (787)

which is a first order differential equation for the velocity v of the electrons:
·
qE = m(v +γv) (788)

142
with a general solution:
~
qE
~v = + ~v0 e−γt (789)
mγ
The second term is exponentially decreasing with time. Therefore, the elec-
trons after some time drift in the metal with a practically constant velocity:
~
qE
~vdrift = . (790)
mγ
This corresponds to a current density

N q2 ~
J~ = N q~vdrift = E. (791)
mγ
Our model has lead us to Ohm’s law:

J~ = σ E,
~ (792)

where σ is the conductivity of the metal.

Exercise: What is the relation of σ and the resistance R in

V = IR?

We therefore obtain:
N q2
γ= (793)
mσ
To get a physical picture for the propagation of waves in metals, let us look
at the low and high frequency limits.

11.8.1 Low frequency approximation

In the low frequency limit, ω → 0, we will assume that ωγ  ω 2 . Eq. 786
gives:
σ
n2 = −i . (794)
0 ω
The refraction index is then
s
σ
n= (1 − i). (795)
20 ω

143
The index of refraction has a large imaginary part (as large as the real part).
The amplitude of a wave reduces exponentially with the distance x as it
propagates inside the conductor:
   
~  ~  − xδ
E  = E e (796)

where the penetration distance is:

s
20 c2
δ= . (797)
σω
ω
For copper and a microwave frequency ν = 2π
= 104 MHz the penetration
length is
δ ≈ 6.710−5 cm (798)
(see Feynman Lectures Vol. 2, 32-7. ).

11.8.2 High frequency approximation

In the high frequency limit, ω 2  ωγ, the refraction index becomes:

N q2 1 ωP2
n2 ≈ 1 − = 1 − , (799)
m0 ω 2 ω2
where
N q2
ωP2 = . (800)
m0
For frequencies which are smaller than the “plasma frequency” ω < ωP , the
refraction index is imaginary and the wave is damped after some distance
inside the metal. For frequencies larger than the plasma frequency, ω >
ωP the refraction index is real and the metal becomes transparent to the
electromagnetic wave.
Exercise: Calculate the wavelengths λP = 2πc ωP
for Li, N a, K, Rb. (see
Feynman Lectures Vol. 2, 32-7. )

11.9 Reflection and refraction

Consider two materials with refraction indices n1 and n2 separated by a
boundary surface on the y − z plane as in the picture below:

144
Maxwell equations should govern both regions, as well as the “intermediate”
boundary region. Let’s assume no macroscopic charges and currents in the
two media. The first Maxwell equation is

∇ ~ =−1∇
~ ·E ~ · P~ . (801)
0
Written explicitly in terms of components, we have:
1
∂x Ex + ∂y Ey + ∂z Ez = − (∂x Px + ∂y Py + ∂z Pz ) (802)
0
Derivatives with respect to the x−variable in the boundary regions compare
the fields on the two sides of the boundary. We anticipate them to be larger
than derivatives with respect to y or z or t, which compare the fields on the
same side of the boundary or at different times. If this is the case, for the
boundary region, the above Maxwell equation becomes:
1
∂x Ex = − ∂x Px (803)
0
or, equivalently,
Ex2 − Ex1 1 Px2 − Px1
=− (804)
∆x 0 ∆x
which yields our first boundary condition:
0 Ex2 + Px2 = 0 Ex1 + Px1 . (805)

145
 Let us now work similarly with another Maxwell equation:
~
∇ ~ = − ∂B .
~ ×E (806)
∂t
Decomposing in components, the above encompasses three equations. For
example,
∂Ez ∂Ey ∂Bx
− =− (807)
∂y ∂z ∂t
∂
contains no large ∂x
derivatives and it yields no boundary condition. How-
ever,
∂Ex ∂Ez ∂By
− =− (808)
∂z ∂x ∂t
yields
∂Ez
= 0 ; Ez1 = Ez2 . (809)
∂x
Similarly, we obtain
Ey1 = Ey2 (810)
From the remaining Maxwell equations, we obtain (exercise) the boundary
condition:
B~2 = B~ 1. (811)
In summary, we have obtained the following boundary conditions for the
electromagnetic field:
B~1 = B~2 (812)
E~ 1,k = E
~ 2,k (813)
~ 1 + P~1 )⊥ = (0 E
(0 E ~ 2 + P~2 )⊥ (814)
where ⊥ denotes the component perpendicular to the boundary plane and k
the components parallel to it.

11.9.1 Snell’s law

Let us now assume that an electromagnetic plane-wave (E ~I, B
~ I ) approaches
from medium n1 the boundary. Experience tells us that there will be a
reflected (E ~ R, B
~ R ) and a transmitted electromagnetic field (E
~T , B~ T ). The
electric field and magnetic fields on the two sides of the boundary will be:
~1 = E
E ~I + E
~ R, ~2 = E
E ~T , (815)

146
and
~1 = B
B ~I + B
~ R, ~2 = B
B ~T , (816)
with
~ I = êI EI ei(ωI t−~kI ·~x) ,
E (817)
~ R = êR ER ei(ωR t−~kR ·~x) ,
E (818)
~ T = êT ET ei(ωT t−~kT ·~x)
E (819)

and

~ ~ ~ ~ ~ ~
~ I = kI × E I , B
B ~ R = kR × ER , B ~ T = kT × ET . (820)
ωI ωR ωT
The electric and magnetic fields are perpendicular to the direction of propa-
gation:
~kI · êI = ~kR · êR = ~kT · êT = 0. (821)
For the magnitudes of the ~k wave-vectors we have:
kI kR n1 kT n2
= = , = . (822)
ωI ωR c ωT c
From the boundary condition,
~ 1,k = E
E ~ 2,k ,

we have that on the boundary

~x = (0, y, z) (823)

the electric fields satisfy

~ ~ ~
êI,k EI ei(ωI t−kI,k ·~x) + êR,k ER ei(ωR t−kR,k ·~x) = êT,k ET ei(ωT t−kT,k ·~x) , (824)

where ~kk = (0, ky , kz ). Let us choose to evaluate Eq. 824 at

~x = (0, 0, 0). (825)

Then, we obtain:

êI,k EI eiωI t + êR,k ER eiωR t = êT,k ET eiωT t (826)

147
Let’s assume that we arrange for the above equation to be satisfied at t = 0.
If the frequencies ωI , ωR , ωT are different from each other, then the three
terms of the equation will change differently at an arbitrary later time t > 0
and the equation will not be satisfied any longer. We thus conclude that
the frequencies of the transmitted and reflected light are the same as the
frequency of the incident light:

ωI = ωR = ωT = ω. (827)

Similarly, to satisfy Eq. 824 at t = 0, ~x = ((), 0, z) and t = 0, ~x = ((), y, 0)

yields that
~kI,k = ~kR,k = ~kT,k = ~kk . (828)
From the above, we conclude that the incident, the reflected and the trans-
mitted waves propagate on the same plane.

Assuming that the refraction indices n1 , n2 are real, Eq. 828 yields a relation
for the incident, reflected and transmitted wave angles:

kI sin θI = kR sin θR = kT sin θT . (829)

Recalling Eq. 822, we obtain:

n1 sin θI = n1 sin θR = n2 sin θT . (830)

148
We conclude that the incident and reflected wave angles are equal,
θI = θR (831)
and that the transmitted wave angle is:
n1
sin θT = sin θI . (832)
n2
Eq. 832 is the familiar law of Snell in optics.

11.9.2 Polarisation
Let us align the y−axis so that
~kk = kparallel ŷ. (833)
The electric field and the magnetic fields are perpendicular to the wave-vector
~k. We distinguish two possibilities:
~ I k ~z and B
• E ~ I lies on the x − y plane

~ I k ~z and E
• B ~ I lies on the x − y plane

All other polarisations can be calculated as a superposition of these cases.

For, ~eI = ŷ, Eq. 824 gives that
EI + ER = ET . (834)
Also, from
B ~ 2 ; BI,x + BRx = BT,x ; kI,x EI + kR,x ER = kT,x ET
~1 = B (835)
The two equations yield for the reflected and transmitted waves:
2kI,x
ET = EI , (836)
kI,x + kT,x
and
kI,x − kTx
ER = EI , (837)
kI,x + kT,x
where we have used that
kR,x = −kI,x .
Exercise: Calculate ET , ER in terms of θI , θT .
Exercise: Calculate ET , ER for the polarisation B~ k ẑ.

149
11.10 Cherenkov radiation
In a dielectric medium, like air of water, Maxwell equations take approxi-
mately the same form as in the vacuum where we substitute
c
c → cm = √ , 0 →  = 0 (1 + χ). (838)
1+χ
Therefore, the solutions of Maxwell equations for the four-vector potential
Aµ in the vacuum are also solutions of Maxwell-equations in a dielectric as
long as we adjust appropriately the constants c, 0 .
In the vacuum, all particles propagate with a velocity smaller than the
speed of light. However, in a medium electrically charged particles can have
a velocity which is bigger than the speed of light in that medium:

cm < v < c.

Our solutions of Maxwell equations have assumed so far that v < cm . It

is interesting to see what is Aµ (x) in the case that some currents j µ contain
charged particles with
v > cm .
Consider a single charge q travelling with a uniform velocity v > cm . In units
of cm = 1,  = 1, the solution of Maxwell equations takes the form:
Z
Aµ (x) = d4 x0 G(x − x0 )J µ (x0 ) (839)

where
1  
G(x − x0 ) = δ (x − x0 )2 Θ(x0 − x00 ) (840)
2π
and
dx0µ
J (x ) = q 0 δ(~x0 − ~v t0 ).
µ 0
(841)
dt
Identifying
xµ ≡ (t, ~x), x0µ ≡ (t0 , ~x0 ),
we have
(q, q~v ) Z 0  2

Aµ (x) = dt δ (t − t0 )2 − |~x − ~v t0 | Θ(t > t0 ). (842)
2π
Let us set
~ ≡ ~x − ~v t,
R

150
and
δt ≡ t − t0 .
As usual, we will now find the roots of the equation which requires the
argument of the delta function to vanish:
2
0 = (t − t0 )2 − |~x − ~v t0 |
 2
~
= δt2 − R + ~v δt


= δt2 (1 − v 2 ) − 2Rv cos θδt − R2 , (843)

~ and ~v :
where θ is the angle formed by R
~ · ~v = Rv cos θ.
R
The discriminant of the binomial in Eq. 843 is
1
 
2 2 2 2 2 2 2
∆ = 4R v cos θ + 4(1 − v )R = 4R v 2
− sin2 θ . (844)
v
For v < 1, the discriminant is positive and the delta-constraint has always a
real solution. In our case v > 1, we have a solution only when
1 1
− < sin θ < (845)
v v

Outside this interval, there are no solutions for the delta-function constraint
and the potential is zero. The solutions of the binomial equation are:
 s 
Rv  1
δt = 2 − cos θ ± − sin2 θ (846)
v −1 v2

151
For v > 1, the square root in the bracket is smaller in magnitude than the
first term − cos θ in the bracket. The Theta-function constraint Θ(x0 >
x00 )requires that δt > 0. This can only occur for
π
cos θ < 0 ; θ > (847)
2
In summary, the vector potential is non-zero only in a cone trailing the
charged particle, defined by the angle
π 1 1
θ> , − < sin θ < (848)
2 v v

(see Visualisation of Cherenkov radiation) Computing the value of the po-

tential inside the cone is straightforward, yielding:

(q, q~v )
q
µ
A = 1 − v 2 sin2 θ (849)
2πR
The gradient of the vector potential is very large as we move from the inner
side of the cone, where the potential is non-zero, to the outside of the cone
where the potential is zero and gives value to a large value of the electro-
magnetic field across the cone boundary. This is the Cherenkov radiation
associated with the superluminal motion of charged particles in a dielectric
material.
Recently, a lot of excitement was created due to a mistaken experimental
measurement of the OPERA collaboration which gave a velocity for neutrinos

152
bigger than the speed of light in the vacuum travelling from CERN in Geneva
to their neutrino detector in Gran Sasso in Italy. Theorists raised skepticism
about this measurement which was not fitting expectations of Cherenkov
radiation phenomena. Although neutrinos do not have electric charge, they
do have “weak-force” charge and they are expected to slow down as they
travel by emitting Cherenkov “weak-radiation”. The neutrinos detected in
Gran Sasso were measured to be more energetic, as if they were not losing
energy during their travel in contradiction to expectations. The measurement
turned out to be wrong and when repeated correctly the velocity of the
neutrinos was consistent with v < c. (see superluminal neutrinos )

153
References
[1] Feynman Lectures in Physics, Vol.2

[2] J.D Jackson, Classical Electrodynamics

[3] The Classical Theory of Fields, Fourth Edition: Volume 2, by L D

Landau (Author), E.M. Lifshitz

[4] S. Weinberg, Gravitation and Cosmology

154