LECTURE NOTES ON ELECTRODYNAMICS

PHY 841 – 2017

Scott Pratt
Department of Physics and Astronomy
Michigan State University

These notes are for the one-semester graduate level electrodynamics course taught at Michi-
gan State University. These notes are more terse than a text book, they do cover all the mate-
rial used in PHY 841. They are NOT meant to serve as a replacement for a text. The course
makes use of two textbooks: The Classical Theory of Fields by L.D. Landau and E.M. Lifshitz,
and Classical Electrodynamics by J.D. Jackson. Anybody is welcome to use the notes to their
heart’s content, though the text should be treated with the usual academic respect when it
comes to copying material. If anyone is interested in the LATEX source files, they should con-
tact me ([email protected]). Solutions to the end-of-chapter problems are also provided on the
course web site (http://www.pa.msu.edu/courses/phy841). Please beware that this is a web
manuscript, and is thus alive and subject to change at any time.
PHY 841 CONTENTS

Contents
1 Special Relativity Primer 4
1.1 γ Factors and Such . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 Lorentz Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.3 Invariants and the metric tensor g αβ . . . . . . . . . . . . . . . . . . . . . . . . 9
1.4 Four-Velocities and Momenta . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.5 Examples of Invariants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.6 Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.7 Homework Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2 Dynamics of Relativistic Point Particles 16

2.1 Lagrangian for a Free Relativistic Particle . . . . . . . . . . . . . . . . . . . . . . 16
2.2 Interaction of a Charged Particle with an External Electromagnetic Field . . . . . 19
2.3 Motion in a Constant Magnetic Field . . . . . . . . . . . . . . . . . . . . . . . . 20
2.4 Gauge Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.5 The Electromagnetic Field Tensor . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.6 Homework Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

3 Dynamic Electromagnetic Fields 24

3.1 Lagrangian (Density) for Free Fields: Deriving Maxwell’s Equations . . . . . . . . 24
3.2 Pseudo-Vectors and Pseudo-Scalars . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.3 The Stress-Energy Tensor of the Electromagnetic Field . . . . . . . . . . . . . . . 28
3.4 Hyper-Surfaces and Conservation of Energy, Momentum and Angular Momentum 32
3.5 Homework Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

4 Electrostatics 36
4.1 Gauss’s Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.2 Potential Energy of a Fixed Continuous Charge Distribution . . . . . . . . . . . . 36
4.3 Laplace’s Equations and a Fixed Point Charge . . . . . . . . . . . . . . . . . . . 37
4.4 Laplace’s Equations in Cartesian Coordinates . . . . . . . . . . . . . . . . . . . . 38
4.5 Laplace’s Equations and Solutions in Spherical Coordinates . . . . . . . . . . . . 38
4.6 Solutions to Laplace’s Equations in Cylindrical Coordinates . . . . . . . . . . . . 42
4.7 Boundary Value Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.7.1 Method of Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.7.2 Boundary Value Problems Using Cartesian Solutions to the Laplace Equation 44
4.7.3 Boundary Value Problems Using Spherical Harmonics . . . . . . . . . . . 45

1
PHY 841 CONTENTS

4.7.4 Boundary Value Problems Using Cylindrical Harmonics . . . . . . . . . . 46

4.8 Solving Boundary Value Problems Numerically . . . . . . . . . . . . . . . . . . 47
4.9 Homework Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

5 Multipole Expansions 51
5.1 Expanding Coulomb Potential in Spherical Harmonics . . . . . . . . . . . . . . . 51
5.2 Electric Field of a Dipole . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.3 Energy in arbitrary external field . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.4 Homework Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

6 Magnetostatics 60
6.1 The Biot-Savart Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
6.2 Magnetic Moments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
6.3 Magnetic Forces and Torques . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
6.4 Overlapping Distributions and Hyperfine Splitting . . . . . . . . . . . . . . . . . 64
6.5 Homework Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

7 Electromagnetic Waves 70
7.1 The Wave Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
7.2 Stress-Energy Tensor of Electromagnetic Waves . . . . . . . . . . . . . . . . . . . 71
7.3 The Doppler Effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
7.4 Wave Guides and Cavities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
7.5 Homework Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

8 Radiation 79
8.1 Coupling the Electromagnetic Field to Dynamic Currents and Charges . . . . . . 79
8.2 Radiation from an Accelerating Point Charge . . . . . . . . . . . . . . . . . . . . 81
8.3 Liénard-Wiechert Potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
8.4 Radiation from relativistic particles . . . . . . . . . . . . . . . . . . . . . . . . . 84
8.5 Frequency Dependence of Radiation from Accelerated Charge . . . . . . . . . . . 86
8.6 Radiation from Oscillating Systems with Well-Defined Frequencies . . . . . . . . 87
8.7 Electric and Magnetic Dipole Radiation . . . . . . . . . . . . . . . . . . . . . . . 88
8.8 Homework Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

9 Scattering 93
9.1 Scattering in the Long Wavelength Limit . . . . . . . . . . . . . . . . . . . . . . 93
9.2 Thomson Scattering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
9.3 Compton Scattering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94

2
PHY 841 CONTENTS

9.4 Scattering of Light from Confined Charges . . . . . . . . . . . . . . . . . . . . . 95

9.5 Homework Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

3
PHY 841 1 SPECIAL RELATIVITY PRIMER

1 Special Relativity Primer

Electromagnetism is inherently relativistic. To see this, consider a charged particle moving

through a magnetic field in deep space. The particle undergoes an acceleration proportional
to its velocity because magnetic force, F⃗ = q⃗ v × R,
⃗ depend on velocity. But, who defines the
velocity? If an observer moves with the particle’s velocity, the speed of the particle is zero, and
therefore there is no magnetic force and no acceleration. Clearly, the acceleration cannot exist in
one reference frame and disappear in another. The solution to this paradox is that if one boosts
the observer to the frame of the particle, the boosted observer will then see an electric field. The
fact that velocity boosts mix magnetic and electric field is related to the fact that space and time
get interchanged in boosts due to the special theory of relativity. Thus, magnetism cannot be
understood without relativity. The first part of the course will cover the relativistic formulation
of Maxwell’s Equations. All such material will seem intimidating without a firm grasp of the
principles and formalism of the special theory of relativity in the context of classical physics.
Students should also read through the first chapter of Landau and Lifshitz.

1.1 γ Factors and Such

First, we review the standard arguments for Lorentz length contraction and for time dilation,
i.e., we will demonstrate
p how a meter stick moving with velocity v appears shorter by a factor
1/γ, where γ = 1/ 1 − (v/c)2 , and will also discuss how a moving clock has ticks separated
by extended times, γ.
Both these results stem from the basic postulate of special relativity, that the speed of light is the
same in all reference frames. First, we consider time dilation. Consider two mirrors separated
by distance L0 . To an observer in the frame of the mirror, the time for light to bounce back and
forth from the mirrors is t0 = 2L0 /c. Now let the mirror move perpendicular to the path of the
light with speed v. An observer in the laboratory sees the same light pulse move with speed c,
but that the path is longer than 2L0 because the return point will have moved a distance vtlab ,
making the entire distance equal to
q
L = 2 L20 + (vtlab /2)2 . (1.1)

L0

L0 v

Figure 1.1: The upper clock functions by counting clicks separated in time by t0 = 2L0 /c. If the clock is
moving, but if the speed of light is the same, more time is required for the time between pulses because
the distance traveled is larger.

4
PHY 841 1 SPECIAL RELATIVITY PRIMER

The time for the return in the lab frame is thus

p
2 L20 + (vtlab /2)2
tlab = . (1.2)
c
Replacing L0 with ct0 /2, one can solve for tlab in terms of t0 ,

1
tlab = γt0 , γ ≡ p . (1.3)
1 − (v/c)2

Thus, all moving clocks run slow. The most basic manifestation of this considers the lifetime of
radioactive decays, which are extended by the factor γ. Note that for the future, we will often
work in units where c = 1 to save ink and eyestrain.
Secondly, we consider the same set of mirrors, but instead let the apparatus move parallel to the
path of the light (perpendicular to the plane of the mirrors). Again, the observer moving with
the stick sees the time between pulses as 2L0 /c. The laboratory observer measures,

Llab Llab 2cLlab 2γ 2 Llab

tlab = + = = , (1.4)
c+v c−v c2 − v 2 c
where here we have allowed the length of the meter stick to change from the value observed in
the frame of the stick. Using the fact that tlab = γt0 = 2γL0 /c,

L0 Llab
2γ = 2γ 2 , (1.5)
c c
L0
Llab = .
γ

Thus moving meter sticks appear shorter.

One has to be careful to note that the two simple expressions apply for very specific circum-
stances. The expression for time dilation, tlab = γt0 , only applies when the time separates
two events which occur at the same location in the frame of the moving observer measuring t0 .
Also, the expression for length contraction, Llab = L0 /γ, is applicable only when the distance
between the moving ends are measured simultaneously in time according to the observer mea-
suring L0 . These subtleties are illustrated by the ladder paradox. Imagine a runner moving at
60% of the speed of light carrying a 10 foot ladder and moving through a 10 foot garage that has
doors at both ends. The gamma factor is 1.25. Thus, an observer in the frame of the garage sees
an 8-foot ladder moving through the garage, and could in principle close the garage at both ends
trapping the ladder completely inside. The runner sees an 8-foot garage, and believes there was
a moment when the front of the ladder had gone completely through the garage while the back
of the ladder had not penetrated the front door. The paradoxical question is “Did the ladder
fit?”. The answer has to do with the simultaneity of two events. In this case it could be lights
flashing at the front and back of the ladder. If the runner thinks the lights flash simultaneously,
indeed both lights flash outside the garage and on opposite sides. The observer in the garage
frame agrees with the assessment that the lights flashed outside the garage, but instead thinks
that the light at the back of the ladder blinked first and that the light at the front of the ladder
blinked later. Conversely, if the lights blinked such that the observer in the garage frame thought

5
PHY 841 1 SPECIAL RELATIVITY PRIMER

they were simultaneous, the two lights could have both flashed inside the garage. However, the
runner would have recorded the light at the front of the ladder blinking first.
The ladder paradox underscores the importance of thinking of times and distances as describing
the difference between two events, i.e., times and displacements are always relative to some-
thing. There are other famous paradoxes in relativity that also lead to a better understanding
of the essence of the theory, such as the twin paradox or whether the radius of a rotating wheel
contracts. However, the latter two involve acceleration and are thus related to the general theory
of relativity, which is not considered here, but would be considered in a course on gravity.

1.2 Lorentz Transformations

Space and time are even mixed together in non-relativistic (Newtonian) transformations. For
instance, consider an event that occurs at time t and position x in the laboratory frame (Here we
consider only one spatial dimension). In a moving frame, the event occurs at:

xlab = x0 + vt0 , tlab = t0 , (1.6)

in a Newtonian transformation. For a relativistic transformation, we assume a more general

linear form,

xlab = Ax0 + Bvt0 , (1.7)

tlab = Ct0 + Dvx0 , (1.8)

where A, B, C and D are functions of v 2 . The powers of v are required by parity considerations.
The inverse transformation should look the same, but with v → −v,

x0 = Axlab − Bvtlab , (1.9)

t0 = Ctlab − Dvxlab .

First, we consider the decay of a relativistic particle which passes by the point (x = t = 0) in
both frames. Since the particle does not move in the particle frame (x0 = 0). The transforma-
tions becomes

tlab = Ct0 , (1.10)

xlab = Bvt0 .

The expression for time dilation in the previous section then yields,

C = γ. (1.11)

The decay occurs at the position xlab = vtlab , and since tlab = γt0 , one finds

B = γ. (1.12)

To solve for the other two coefficients, consider the ladder paradox. Assume a light at the back
end of the latter blinks at a time x = t = 0, and at the front end of the ladder a light blinks
simultaneously (in the frame of the ladder) at the space-time point, x0 = L0 , t0 = 0. In the

6
PHY 841 1 SPECIAL RELATIVITY PRIMER

laboratory frame the lights blink at times different by an amount DvL0 . The position at which
the light blinks is
xlab = Llab + vtlab , (1.13)
where Llab is the length of the ladder. Since the apparent length of the ladder is shrunk by a
factor γ, L.lab = x0 /γ, and
xlab = x0 /γ + vtlab , (1.14)
Rearranged,
x0 = γxlab − γvtlab . (1.15)
This gives A = γ. To solve for D take the expressions, Eq.s (1.7,1.9,1.10) in the Lorentz transfor-
mations, and solve for D. One finds D = γ.
The expressions for the coefficients defined in Eq.s (1.7-1.10)) can be summed up as a matrix
equation,

r α = Lαβ r ′β , (1.16)
 
γ γv
L= ,
γv γ

Here, the indices α are either 0 or 1, with “0” referring to time component and “1” referring to
the ‘x” component. If we included y and z, there would be four components representing the
coordinate of an event, t, x, y, z. These four components make up a “four-vector”. The Lorentz
matrix L performs a boost along the x axis and leaves the other two dimensions unchanged. For
our notes we will stick to the convention that greek indices label the components of four-vectors,
while roman indices denote the components of three-vectors, i.e., α = 0, 1, 2, 3 and i = 1, 2, 3..
The choice of making the indices upper vs. lower will be explained later.
For a boost along the x-axis, the 4 × 4 matrix becomes
 
γ γv 0 0
 γv γ 0 0 
L=  0
. (1.17)
0 1 0 
0 0 0 1

For rotations of a coordinate system, all three vectors are transformed according to the same
rotation matrix U , ri′ = Uij rj , regardless of whether the vector ⃗
r refers to a spatial coordinate
or a velocity or momentum. Similarly, for boosts all four-vectors are transformed the same,

r α = Lαβ r β , (1.18)

regardless of whether the quantity r represents the space-time coordinate of an event or a mo-
menta (in which case the zero-th component is the energy).
Due to the identity, γ 2 − γ 2 v 2 = 1, one can express γ and γv as cosh η and sinh η respectively.
The Lorentz matrix then becomes
 
cosh η sinh η 0 0
 sinh η cosh η 0 0 
L= 
, (1.19)
0 0 1 0 
0 0 0 1

7
PHY 841 1 SPECIAL RELATIVITY PRIMER

which illustrates the similarity of the Lorentz transformations to rotations by an imaginary angle.
x·y
Whereas for a rotation, one finds that x1 y1 + x2 y2 + x3 y3 = ⃗ ⃗ is invariant to rotations,
Lorentz transformations leave the quantity,
r · q ≡ r0 q0 − r1 q1 − r2 q2 − r3 q3 (1.20)
unchanged. To see this consider the the boosted quantities,
r ′ = (γr0 − γvr1 , γr1 − γvr0 , r2 , r3 ), (1.21)
q ′ = (γq0 − γvrq1 , γrq1 − γvq0 , q2 , q3 ).
A little algebra shows that
r0′ q0′ − r1′ q1′ − r2′ q2′ − r3′ q3′ = r0 q0 − r1 q1 − r2 q2 − r3 q3 (1.22)
= (γ 2 − γ 2 v 2 )(r0 q0 − r1 q1 ) − r2 q2 − r3 q3
= r · q.
The “dot-product” of four vectors is as a Lorentz invariant, meaning a quantity unchanged by
reference frame. The dot-product is also invariant to rotations in 3−space, and the combinations
of rotations and boosts is known as the Lorentz group. For a group, any combinations of transfor-
mations can be expressed as a single transformation. If one adds parity and time-reversal, it is
known as the full Lorentz group, and if one adds translational symmetry, it becomes the Poincaré
group. Rotations and boosts are not in separable groups, i.e., if one performs two boosts, the re-
sult is a boost plus a rotation. If the boosts had formed a group by themselves, any two boosts
would have been equivalent to a single boost.

EXAMPLE:
Consider two successive boosts along the x direction. The first defined by γv = sinh η1 , and
the second one with γv = sinh η2 . Show that the combination is equivalent to one boost with
γv = sinh(η1 + η2 ).
Writing down the product of the two Lorentz matrices (2-dimensions is sufficient),
  
cosh η2 sinh η2 cosh η1 sinh η1
sinh η2 cosh η2 sinh η1 cosh η1
 
cosh η2 cosh η1 + sinh η2 sinh η1 cosh η2 sinh η1 + sinh η2 cosh η1
=
cosh η2 sinh η1 + sinh η2 cosh η1 cosh η2 cosh η1 + sinh η2 sinh η1
 
cosh(η1 + η2 ) sinh(η1 + η2 )
= ,
sinh(η1 + η2 ) cosh(η1 + η2 )
where double-angle formulas were applied for the last step. For high-energy phe-
nomenology the the quantities η are referred to as “rapidities” when the boosts are
along the beam axis. The simplicity of rapidities comes from the fact that they add
like Newtonian velocities. However, this simple addition only works when the ra-
pidities are defined along a single axis, i.e., if one defines sinh ηx = γvx , · · · , for all
three dimensions, the addition formulas break down due to the non-commutation of
boosts along different directions.

8
PHY 841 1 SPECIAL RELATIVITY PRIMER

1.3 Invariants and the metric tensor g αβ

As shown previously, the dot product of two vectors,

Aα gαβ B β , (1.23)
 
1 0 0 0
 0 −1 0 0 
gαβ ≡ 0 0 −1 0

 (1.24)
0 0 0 −1

is invariant to both rotations and boosts. As a notional trick, one can define four-vectors Aα
(subscript rather than superscript) as

Aα ≡ gαβ Aβ , (1.25)

which means that Aα Aα is an invariant (summation of repeated indices inferred). In all calcu-
lations the summed indices always appear with one superscript (contravariant) and one sub-
scripted (covariant). The fact that multiplying by gαβ simply lowers the index also applies to
tensors, i.e.,
gαβ C βγ = Cαγ . (1.26)
Note that this property means that g αβ is simply the unit matrix. Furthermore, this property also
applies to g αβ ,
g αβ Cβγ = C αγ . (1.27)
Derivatives might at first seem a little backwards. Consider a scalar function ϕ(x), where x is a
four vector. For small changes in x, x
∂ϕ
δϕ = δxµ . (1.28)
∂xµ
Since δϕ is also a scalar, the vector ∂/∂xµ must transform as a covariant vector. This motivates
the notation,
∂
∂µ = . (1.29)
∂xµ
Note that for this convention, the equation of continuity is especially compact. For a four vector
J , where J 0 refers to the charge density and J i signals the current density, ∂ · J = 0 → ∂t J0 +
∇ · J⃗ = 0. Maxwell equations also take on particularly beautiful forms, ∂α F αβ = J β , and
∂α F̃ αβ = 0. Additionally, we point out that we will employ the convention throughout this
course that greek indices refer to all four components, while roman indices suggest only spatial
components. Bold face will refer to the vector components of a three-vector, while four-vectors
will not be put into bold face, i.e., p · x = p0 x0 − p · x.

1.4 Four-Velocities and Momenta

For a particle that moves between two points by a small displacement ∆xα , one can define the
quantity, p
∆τ ≡ ∆xα ∆xα , (1.30)

9
PHY 841 1 SPECIAL RELATIVITY PRIMER

which is an invariant. In the frame of the particle it is easy to see what this quantity represents,
since in that frame the spatial components ∆xi are all zero. Thus ∆τ is the amount a clock,
moving with the particle, has progressed during the differential displacement. Further, one can
define a vector,
∆xα
uα ≡ , (1.31)
∆τ
which is also a four-vector since ∆xα is a four-vector and ∆τ is a Lorentz scalar (or invariant).
The four-vector uα is often referred to as the relativistic velocity, and given that ∆x0 = γ∆τ ,
one can see that
dxi
u0 = γ, ui = γ = γv i . (1.32)
dt
It is easy to see that
uα uα = 1. (1.33)

The momentum is defined by multiplying uα by the particle’s mass m, which is also a scalar.
This then gives,
pα pα = m2 . (1.34)
The zeroth component of the momentum is identified as the energy. This gives the relation that
for a particle at rest,
E = p0 = m. (1.35)
Of course, the more famous relation, E = mc2 , requires keeping track of all the factors of c.

EXAMPLE:
A beam of particles of mass mA is aimed a target with particles of mass mB . What kinetic
energy, K, is required so that one can make a resonance of mass mC .

The solution is based on energy-momentum conservation. One way to move forward

is to calculate the invariant mass in terms of the total momentum and set it to mC .

m2inv = (pA + pB )2 = (K + mA + mB )2 − p
⃗2A
 
= (K + mA + mB )2 − (K + mA )2 − m2A
= m2A + m2B + 2mB (K + mA ) = m2C ,
m2C − (mA + mB )2
K = .
2mB

EXAMPLE:
Consider two particles recorded with momenta pA and pB at space times separated by r α =
xα1 − x2 . In terms of relativistic invariants using pA , pB and r, solve for the impact parameter,
α

i.e., the distance of closest approach as measured by an observer in the center-of-mass frame.

First, we consider two four vectors which will be used as projectors to eliminate the
components of r that, in the center-of-mass frame, are either time-like or along the

10
PHY 841 1 SPECIAL RELATIVITY PRIMER

direction of the relative momentum. First, the time-like vector is the total momen-
tum,
P α = pα α
A + pB ,

which in the c.o.m. frame becomes (EA + EB , 0, 0, 0). One can define a vector,

P ·r
r ′α ≡ r α − P α .
P2
In the center-of-mass frame, r ′ looks exactly like r, except the α = 0 component
vanishes. Next, one defines a vector which in the c.o.m. frame is parallel to the
relative momentum, and is zero for the α = 0 piece. This would be:

P ·q
q ′α = q α − P α A − pB .
, q α = pα α
P2
Again, one can consider the c.o.m. frame, where it is clear that q ′0 = 0 and q ′i =
(piA − piB ). One can then project away the part of r ′ parallel to q, and make a new
vector b,
q′ · r′
bα ≡ r ′α − q ′α ′2
q
Finally, the impact parameter squared is, after some surprisingly painful algebra,

(q · r)2 P 2 + (P · r)2 q 2 − 2(q · r)(P · r)(P · q)

B 2 = −b2 = −r 2 + .
P 2 q 2 − (P · q)2

1.5 Examples of Invariants

Here I discuss several invariants you will encounter at various times throughout this course or
in the literature. First, one can see that d4 x is invariant to boosts by considering the Jacobian of
a Lorentz transformation along the x axis. In that case,
 
cosh η sinh η 0 0
 sinh η cosh η 0 0 
L= 
. (1.36)
0 0 1 0 
0 0 0 1

The Jacobian is simply the determinant of the matrix, J = cosh2 − sinh2 = 1. Thus, d4 x is
invariant. For a boost in an arbitrary direction, one can first to a rotation (d3 x is invariant to
rotation) combined with a boost.
Then d4 p is also invariant as is,
Z Z Z 3
d p
d p δ(p − m )s(p) =
4 2 2
dp0 d p δ(p0 − Ep )s(p) =
3 2 2
s(p)|p0 =Ep . (1.37)
2Ep

11
PHY 841 1 SPECIAL RELATIVITY PRIMER

Here s(p) is some arbitrary scalar function. Since s could be anything, then one can state that
d3 p/Ep is also invariant. This is why spectra in particle physics are expressed as one of the two
identical forms below
Ep dN dN
= . (1.38)
d3 p dϕpt dpt dy
The expression on the right-hand side can be found by seeing that dpz /E = dy, where y is the
rapidity. Since dN is the number of counts in a bin, and since a number of counts is invariant,
you needn’t worry about how dN transforms.
Often, when calculating the number of particles in a box, one writes
Z
d3 xd3 p
N = f (p), (1.39)
(2π)3

where f is the phase space density, or occupation probability. This must be an invariant. For in-
stance, if you had fermions as zero temperature, f would be zero or unity depending on whether
it was inside or outside the Fermi sea. This would not change if viewed from a different reference
frame. Since f is invariant, and dN is invariant, it stands that d3 x d3 p is also invariant. To see
that this is true, one must remember that for this expression one assumes that dx measures the
distance between the boundaries of a cell where both the positions of the boundaries are mea-
sured simultaneously, like measuring the length of a meter stick. If one compares the volume
measured by an observer moving with the velocity of the particle, p/E, and compares to the
length in the lab frame one finds that d3 x shorter by the Lorentz factor, E/m. If one considers
d3 p in the two frames, one sees that d3 p in the lab frame must be larger by a factor of E/m so
that d3 p/E is invariant. Thus, the product d3 pd3 x is invariant.
Another example that comes up often is a collision rate per volume and per time, dNc /d4 x,
which is manifestly invariant. One can also consider the collision rate of particles from specific
regions of momentum space, d3 pa /Ea and d3 p/Eb . The quantity would normally written as

dNc
Ea Eb = fa (pa )fb (pb )Something, (1.40)
d4 x d3 pa d3 pb

where fa and fb are the phase space densities, or occupancies, of the two particles. Occupancies
are also invariant, i.e. if the Fermi sea is full, f = 1 regardless of what frame is considered.
Here, Something has to be an invariant. Since it is invariant one can consider the center-of-
mass frame where the momenta are back to back, pa = −pb . In that frame one knows from
kinematics that
Ea Eb √
Something = σ( s)vrel , (1.41)
(2π)6
where σ is the cross section and the relative velocity is vrel = |p|/Ea + |p|/Eb . One must then
simply write Ea Eb vrel as a Lorentz invariant. To do this,

⃗a Eb − p
Ea Eb vrel = p ⃗b Ea |. (1.42)
√
One can now write Ea = pa · P/ s and replace pa with the spatial components of p′a ≡
pa − P (pa · P )/s to project out the zeroth components of pa . Doing the same for pb one then

12
PHY 841 1 SPECIAL RELATIVITY PRIMER

finds

(Ea Eb vrel )2 = − [(pb · P )(pa − P (pa · P )/s) − (pa · P )(pb − P (pb · P )/s)]2 /s
· · · LOTS of algebra · · · (1.43)
2  
(s − ma + mb )(s + ma − mb ) 2
2 2 2
= s + m4a + m4b − 2m2a m2b − 2sm2a − 2sm2b
4s
(s − m2a + m2b )(s + m2a − m2b )  
= (pa · pb )2 − m2a m2b ,
s
= wx

Here, numerous steps are omitted from the last step as this is related to a homework problem at
the beginning of the last chapter. Putting this all together, one finds
Z
dNc d3 pa d 3 pb √ q
= fa (pa )fb (pb )σ( s) (pa · pb )2 − m2a m2b . (1.44)
d4 x (2π)3 Ea (2π)3 Eb

1.6 Tensors
Tensors are quantities with two or more Lorentz indices. Examples are the stress-energy tensor
and the electromagnetic field tensor. Similarly to how rotations in three-dimensional space affect
three-dimensional tensors,
′
Mim = Uij Mjk U −1 Ukm , (1.45)

a relativistic tensor translates as

M ′αδ = Lαβ Mβγ L(−1)γδ . (1.46)

Again, note that summed indices always involve one covariant and one contravariant index. So
the above equation can be rewritten a number of ways, e.g.

M ′αδ = Lαβ Mβγ L(−1)γδ . (1.47)

An example of a fourth-rank tensor is the anti-symmetric tensor,


 1, for even permutations of αβγδ from 0123,
ϵαβγδ = −1, for odd permutations of αβγδ from 0123, (1.48)

0, if any index repeats.

With these definitions,

ϵ0123 = ϵ1230 = ϵ2301 = ϵ3012 = ϵ0231 = ϵ2310 (1.49)

= ϵ3102 = ϵ1023 = ϵ0312 = ϵ3120 = ϵ1203 = ϵ2031 = 1,
ϵ1023 = ϵ0231 = ϵ2310 = ϵ2310 = ϵ0132 = ϵ1320
= ϵ3201 = ϵ2013 = ϵ0213 = ϵ2130 = ϵ1302 = ϵ3021 = −1,
ϵαβγδ = 0 otherwise.

13
PHY 841 1 SPECIAL RELATIVITY PRIMER

1.7 Homework Problems

1. Suppose you are doing a fixed-target experiment at the LHC. The protons have a beam
kinetic energy of 7 TeV. (The proton mass is 938.28 MeV/c2 ). If the experiment were redone
with a collider built with an equivalent center-of-mass energy, what would the kinetic
energy of each beam be?
2. Find the equivalent fixed beam energy for a fixed target to have the same center-of-mass
energy as the collider experiment at the LHC.
3. Consider a 1+1 dimension vector, (E, p), where m2 ≡ E 2 − p2 . Consider the transformed
vector, p′α = Lαβ pβ , where L is defined according to Eq. (1.19). Show that m′2 ≡ E ′2 −
p′2 = m2 .
4. Consider two particles with four momenta pa and pb . Particle a is recorded at the space
time point ra = (0, 0, 0, 0) and particle b is recorded at rb = r. For an observer moving
with particle a find the time at which particle b passes at the point of closest approach.
Express your answer in terms of Lorentz invariants, i.e., dot products involving pa , pb and
r.
5. Consider two particles of mass ma and mb with four momenta pa and pb . Show that the
relative velocity, |⃗
va − ⃗
vb |, according to an observer in the center-of-mass frame is:

s2 /4  2 
2
vrel = s + m4a + m4b − 2sm2a − 2sm2b − 2m2a m2b .
(pa · P )2 (pb · P )2
Here, P is the total momentum and s = P 2 . FYI: The answer √ is sometimes called the
triangle function. If you have a triangle with sides of length ( s, ma , mb ) and solve for
the area of the triangle you get a similar function aside from the prefactor. This term for
the relative velocity appears often, e.g. you divide by this factor to convert rates from
Feynman diagrams into cross sections.
6. The Lorentz transformation is a tensor, Lαβ , which transforms some four vector pα ob-
served by an observer moving with four velocity uα to a vector p′α as determined by an
observer moving with four-velocity u′α .

Lαβ pβ = p′α .

Since L is a tensor it must be of the form,

Lαβ = Auα u′β + Bu′α uβ + Cuα uβ + Du′α u′β + Eg αβ , (1.50)

where A − E are scalar functions of u and u′ . Since u2 = u′2 = 1, the only scalar function
available is u · u′ . Consider the transformation from the rest frame u = (1, 0, 0, 0) to the
frame u′ = (γ, γv, 0, 0). You know that the Lorentz transformation is:
 
γ −γv 0 0
 −γv γ 0 0 
Lαβ =   0
.
0 1 0 
0 0 0 1

14
PHY 841 1 SPECIAL RELATIVITY PRIMER

Further, u · u′ = γ. From the form in Eq. (1.50),

 
Aγ + Bγ + C + Dγ 2 + E −Aγv − Dγ 2 v 0 0
 Bγv + Dγ 2 v −Dγ 2 v 2 + E 0 0 
Lαβ = 
.

0 0 E 0
0 0 0 E

(a) Solve for the coefficients A through E in terms of γ = u · u′ .

(b) Show that for the four vector u′ ,

Lαβ u′β = uα .

(c) Show that

Lαβ (u, u′ )Lβγ (u′ , u) = g αγ .

15
PHY 841 2 DYNAMICS OF RELATIVISTIC POINT PARTICLES

2 Dynamics of Relativistic Point Particles

We won’t discuss the dynamics of individual charged particles very much in this course, but
it is good to review the least-action principle, and the example for relativistic particles. In the
next chapter we will apply these principles to field equations, deriving Maxwell’s equations,
so it will be helpful to review the connection between least action and Lagrange’s equations,
with an emphasis on how symmetries lead to conservation laws. In this chapter we consider the
interaction with an external electromagnetic field. This chapter closely follows the approach in
Landau and Lifshitz.

2.1 Lagrangian for a Free Relativistic Particle

The action, S, is related to a Lagrangian by
Z tb
S= dt L(⃗ r˙ (t), t).
r (t), ⃗ (2.1)
ta

Usually, there L has no explicit dependence on t and only depends on ⃗ r˙ . In that case
r and ⃗
minimizing S leads to Lagrange’s equations,
Z tb X  ∂L ∂L

δS = dt δ ṙi + δri (2.2)
ta i
∂ ṙi ∂ri
Z tb X  d ∂L ∂L

= dt − + δri
ta i
dt ∂ ṙ i ∂r i

= 0.

The equivalence must be true for any δri at any time, thus one derives the usual Lagrange
equations of motion,
d ∂L ∂L
δri = . (2.3)
dt ∂ ṙi ∂ri
If there are symmetries, in that L does not depend on some coordinate, a conservation law
ensues. For instance, if L is independent of ϕ (but not ϕ̇) the rotational symmetry gives,

d ∂L
= 0, (2.4)
dt ∂ ϕ̇

and pϕ = ∂L/∂ ϕ̇ is conserved. More generally, consider any small change xi → x + ϵi (x, t).
For translational invariance ϵ is independent of x and δxi = ϵi (t), and setting δS = 0, one
finds

δS = 0 (2.5)
Z tb
d ∂L
= − dt ϵi (t).
ta dt ∂ ẋi

16
PHY 841 2 DYNAMICS OF RELATIVISTIC POINT PARTICLES

Thus ∂L/ẋi is conserved, and is the momentum in the i direction. For rotational invariance
about an axis Ω̂, δxi = ϵijk xj Ω̂k δϕ(t),
δS = 0 (2.6)
Z tb
d ∂L
= − dt ϵijk (t)xj (t)Ω̂k δϕ(t).
ta dt ∂ ẋi
This must be true for any small angle δϕ, and the conserved quantity is known as the angular
momentum vector about the Ω̂ axis.
∂L
LΩ = − ϵijk xj (t)Ω̂k (2.7)
∂ ẋi
r×p
= (⃗ ⃗) · Ω̂.
⃗ are conserved.
If there exists rotational invariance about any axis then all three components of L
The connection between symmetries and conservation laws is as important a concept as any
in all of physics. For example, in quantum field theory, the arbitrary phase of a complex field
operator is related to conservation of electric charge.
Conservation of energy comes from Lagrange’s equations. Defining πi ≡ ∂L/∂ q̇i , where qi are
the generalized coordinates,
d ∂L ∂L
(πi q̇i − L) = π̇i q̇i + πi q̈i − q̈ − q̇i (2.8)
dt ∂ q̇ ∂qi
= 0.
Thus, the Hamiltonian H = πi q̇i − L is conserved. Note this was contingent on the lack of
explicit time dependence in L. Thus, invariance under translation in time is associated with
energy conservation.
Now, we turn back to the problem at hand, the relativistic motion of a free-streaming particle.
The action is a Lorentz-invariant, which greatly constrains what form it can have. For free parti-
cles it can only depend on velocity so a good guess for the form is
s  2
Z tb q Z tb
d⃗r
S = −m dt − (d⃗
2 r ) = −m
2
dt 1 − . (2.9)
ta ta dt
The choice of the mass, −m, for the preceding factor is motivated by having the non-relativistic
expansion have a term mv 2 /2. The equations of motion become
d ∂L m
= √ ṙi (2.10)
dt ∂ ṙi 1 − v2
∂L
= = 0. (2.11)
∂ri
So the conserved momentum is the usual mγ⃗
v . One can also calculate the energy,
m
H = πi q̇i − L = mγv 2 − (2.12)
γ
= mγ(v + (1 − v ) = mγ.
2 2

17
PHY 841 2 DYNAMICS OF RELATIVISTIC POINT PARTICLES

Of course, if you included the factors of c we would get E = mc2 for v = 0.

Boost symmetries are a bit more difficult to consider due to the mixing of time and position. The
best way to consider these is to define t = r0 and ⃗ r in one frame, then consider the boost of
small velocities or rotations

r ′α = r α + δΩαβ (t)rβ . (2.13)

Here, the tensor Ωαβ is any anti-symmetric tensor. The three 0i elements represent boosts, while
the three non-zero ij elements represent rotations.
Now, we rewrite the Lagrangian and action as
s 2  2
Z tb
dt′ r′
d⃗
S = −m dt − . (2.14)
ta dt dt

For δv = 0 the coordinates are t′ = t, ri′ = ri , which gives the usual answer. For small δ⃗
v,
Z tb
∂L d αβ


δS = dt δΩ r β (2.15)
t ∂ ṙ α dt
Z atb      
d ∂L ∂L
= dt − α
rβ + α
ṙβ δΩαβ
t dt ∂ ṙ ∂ ṙ
Z atb  
d
= dt − (πα rβ ) + πα ṙβ δΩαβ
ta dt
= 0.

The second term disappears because δΩαβ is anti-symmetric and π is parallel to ṙ. This must be
zero for any contribution from given components δΩαβ and δΩβα = −δΩαβ ,

d

r α pβ − pα r β = 0, (2.16)
dt
for any choice of αβ. For αβ both with space-like indices, these components correspond to
angular momenta.

M αβ = r α pβ − pα r β , (2.17)
M 12 = Lz , M 23 = Lx , M 31 = Ly .

For the components with α = 0,

M 01 = tpx − xE, M 02 = tpy − yE, M 03 = tpz − zE. (2.18)

Each of the six elements represent conserved quantities. For the M 0i elements the quantities rep-
resent the fact that the center-of-mass moves with fixed velocity. If there were several particles
the quantities would be
X
pa t − ⃗
(⃗ r Ea ) = constant. (2.19)
a

18
PHY 841 2 DYNAMICS OF RELATIVISTIC POINT PARTICLES

P
Dividing the equation by the conserved energy, Etot = a Ea , one gets
P  P
ap
⃗a a Ea ⃗
ra
t − = constant. (2.20)
Etot Etot
Thus the center of mass is defined as an average sum over the positions weighted by their en-
⃗tot /Etot . Of course, all this is postulated on the
ergies, and it moves with constant velocity P
Lagrangian not having interactions that might destroy the Lorentz invariance assumed above.

2.2 Interaction of a Charged Particle with an External Electromagnetic Field

Here, the external electromagnetic field is a four-vector Aµ (r). The zeroth component is the
electric potential ϕ and the spatial components are the usual vector potential. To make a Lorentz-
invariant action that has a contribution that looks like the usual potential energy, eϕ(r), we
consider
Z
S = dτ (−m − eu · A) (2.21)
s   ′ 2
Z Z  
dt′ 2 d⃗r
= −m dt − − e dt A0 − ⃗ v·A .⃗
dt dt
Here, we have used the fact that u0 dτ = γdτ = dt and u v dt. The conjugate momenta is
⃗ dτ = ⃗
v · A term,
changed by the appearance of the velocity in the ⃗ ⃗

⃗
π = mγ⃗ ⃗
v + eA. (2.22)

Lagrange’s equations then become

d
(mγvi + eAi ) = −e∂i A0 + e⃗
v ∂i · A
⃗ (2.23)
dt
d
mγvi = −e∂t Ai − e∂i A0 − e⃗
v · ∇Ai + e⃗
v ∂i · A.
⃗
dt
v · ∇. Next, the last two terms can be
Here, we have made use of the fact that d/dt = ∂t + ⃗
manipulated with the vector identity,
⃗ × (B
A ⃗ × C)
⃗ = B( ⃗ · C)
⃗ A ⃗ − C(
⃗ A ⃗ · B),
⃗ (2.24)
d
(mγvi ) = −e∂i A0 − e∂t Ai + e⃗v × (∇ × A).
⃗
dt
Thus, the electric and magnetic fields are
⃗ = −∇A0 − ∂t A,
E ⃗ (2.25)
⃗ = ∇ × A,
B ⃗

and the equations of motion are

d
(mγvi ) = eE v × B.
⃗ + e⃗ ⃗ (2.26)
dt

19
PHY 841 2 DYNAMICS OF RELATIVISTIC POINT PARTICLES

The electric potential is the zeroth component of the four-vector potential. It is NOT a scalar.
Since the four components of A mix during a boost, magnetic and electric fields also mix. Con-
sider a frame where there is a constant electric field in the ẑ direction due to a potential,

A0 = −zE, A
⃗ = 0. (2.27)

If one boosts in the x direction by a velocity v,

A′0 = γ(−zE), (2.28)

A′x = γv(−zE).

Using the fact that z = z ′ for a boost along the x axis, the ensuing electric and magnetic fields
are
⃗ ′ = γE ẑ,
E (2.29)
⃗′ = ∇ × A
B ⃗ ′ = −vE ŷ.

2.3 Motion in a Constant Magnetic Field

Let’s consider the gauge where

⃗ = xB ŷ,
A (2.30)
⃗ = B ẑ. Beginning with the action, we solve Lagrange’s equations
which gives B
Z n p o
S = − dt m 1 − v 2 − exBvy , (2.31)
py = mγvy + eBx, = constant because there is no y dependence,
px = mγvx ,
d
px = eBvy . (2.32)
dt
Now, let’s show that is indeed circular motion. We need to show that these equations (clockwise
circular motion) are satisfied by

x = x0 + R cos(ωt − ϕ), (2.33)

y = y0 − R sin(ωt − ϕ),
vx = −ωR sin(ωt − ϕ),
vy = −ωR cos(ωt − ϕ).

From the equation for dpx /dt one sees that

eB
ω = , (2.34)
mγ
which is the usual expression for the cyclotron frequency except for the 1/γ factor. Note this is
why cyclotrons can’t work for relativistic energies.

20
PHY 841 2 DYNAMICS OF RELATIVISTIC POINT PARTICLES

Next, we must show that py is indeed a constant.

py = −mγV cos(ωt − ϕ) + eBx, (2.35)
= −mγωR cos(ωt − ϕ) + eB(x0 + R cos(ωt − ϕ))
= eBx0 . ✓
This emphasizes that p ⃗ So having
⃗ in not the velocity multiplied by the mass, but differs by eA.
py being a constant does not mean that the y-component of the velocity is fixed.

2.4 Gauge Transformations

Consider the following change to the vector potential,
A′µ = Aµ + ∂ µ Λ(t, x, y, z). (2.36)
Here Λ is a arbitrary scalar function. The electric and magnetic fields become (with the use of
Eq. (2.40))
Ei′ = Ei − ∂i ∂t Λ + ∂t ∂i Λ = Ei , (2.37)
Bi′ = Bi − ϵijk ∂j ∂k Λ = Bi . (2.38)
Thus, E and B are unchanged by the function Λ, even though A ⃗ is changed. This is known
⃗ are indeed physical when one considers quantum
as gauge invariance. In nature, the fields A
mechanics, e.g. the Aharonov-Bohm effect. However, even then gauge invariance is still true the
scalar function Λ is arbitrary.

2.5 The Electromagnetic Field Tensor

One can define a second-rank anti-symmetric tensor using the vector potential,
F αβ = ∂ α Aβ − ∂ β Aα . (2.39)
Using the definitions,
⃗ = −∂t A
E ⃗ − ∇A0 , (2.40)
⃗ = ∇ × A,
B ⃗

one can consider F αβ component-by-component and find

   
0 Ex Ey Ez 0 −Ex −Ey −Ez
 −Ex 0 −Bz By   Ex 0 −Bz By 
Fαβ =   −Ey Bz
 , F αβ =   . (2.41)
0 −Bx   E y Bz 0 −Bx 
−Ez −By Bx 0 Ez −By Bx 0
The equations of motion then take the form (see HW problem),
d
m uα = eF αβ uβ , (2.42)
dτ
p
where dτ = dt2 − (d⃗
r )2 , is the differential time step in the frame of the particle, i.e. its proper
time.

21
PHY 841 2 DYNAMICS OF RELATIVISTIC POINT PARTICLES

2.6 Homework Problems

1. Consider a system of free particles a with the Lagrangian set as
s
X  dt′ 2  d⃗ 
ra′ 2
L = −ma a
− , (2.43)
a
dt dt

with t′a = t before transforming. Now consider a translation in time,

t′a = t + ϵ(t), ra′ = ⃗

⃗ ra .

Calculate δS and express it so that δS is proportional to ϵ(t), not ϵ̇. Show that this quickly
gives energy conservation.

2. Consider a particle of charge e and mass m moving in a constant electric field in the x
direction, A0 = −eEx. The particle’s initial momentum is p
⃗(t = 0) = py ŷ.

(a) Solving Lagrange’s equation, find px (t).

q
(b) Using the fact that vx = px / m2 + p2x + p2y , find x(t).
q
(c) Using the fact that vy = py / m2 + p2x + p2y , find y(t).
(d) Find the trajectory, x(y).
(e) Take the limit py /m << 1 and find x(y) again. Solve non-relativistically and com-
pare.

3. Consider a particle of charge e and mass m moving in a constant magnetic field in the z
direction. We will consider the gauge where,

Ay = xB/2, (2.44)
Ax = −yB/2, (2.45)
Bρϕ̂
⃗ =
A . (2.46)
2
p
Here, ρ = x2 + y 2 and ϕ̂ = ŷ cos ϕ − x̂ sin ϕ is one of the three unit vectors in the
cylindrical coordinate basis, ẑ, ρ̂, and ϕ̂.
⃗ =
(a) Find the scalar function Λ that transforms the choice used in the earlier section, A
xB ŷ to the vector potential used here.
∂L
(b) Find pϕ = ∂ ϕ̇
.
(c) Setting ż = 0, consider motion in the x − y plane. Beginning with (d/dt)πϕ = 0,
show that the solution with fixed radius (ρ̇ = 0) will work if the ϕ̇ is constant and is
related to the cyclotron frequency as shown in the previous subsection,

eB
ϕ̇ = .
mγ

22
PHY 841 2 DYNAMICS OF RELATIVISTIC POINT PARTICLES

r ) = −F x. This simply changes

4. Consider a particle of mass m moving in a scalar field, Φ(⃗
the local mass to m = m0 − F x, and the Lagrangian is
p
L = −(m0 − F x) 1 − v 2 .

The mass changes with time as m = m0 − F x, and dm/dt = −vF , where v will be the
velocity in the x direction. For the questions below, assume there is no movement in the y
or z directions.

(a) Using Lagrange’s equations, show

dv F
= (1 − v 2 ) .
dt m(t)

(b) Consider a particle at rest at x = y = z = t = 0. Show that the solutions to the

above equations are

m = m0 cos(at), a = F/m0
v = sin(at),
1
x = (1 − cos(at)).
a

(c) At what time, tmax , does the mass become zero?

(d) What is x(tmax )?
(e) What is v(tmax )?
(f) What is ux = γv at tmax ?

5. It is easy to derive the spatial components of Eq. (2.42),

d
pi = qF iα uα ,
dτ

by expressing F αβ in terms of E⃗ and B⃗ using Eq. (2.41), then comparing the Eq. (2.26). It
is less obvious to see how one obtains the equation for d/dτ p0 . Given that p0 = mγ, one
can quickly see that
d 0 d
p = e⃗ v· p
⃗.
dτ dτ
Beginning with the two expressions above, show that

d
p0 = qF 0β uβ .
dτ
Hint, you will need to use the fact that F αβ is anti-symmetric.

23
PHY 841 3 DYNAMIC ELECTROMAGNETIC FIELDS

3 Dynamic Electromagnetic Fields

So far we have discussed the motion of particles in a field, but have ignored how the fields might
change in time. To do so, we need to consider all three parts of the action: the action of a free
particle, Sm , the action involving the interaction of matter with the field Sf m and, new for this
chapter, the action of the field, Sf . We will show how this action leads to Maxwell’s equations.

3.1 Lagrangian (Density) for Free Fields: Deriving Maxwell’s Equations

In the last chapter we considered the motion of a particle in a field, Aα (r), where the field was
given as a function of time. First we will derive Maxwell’s equations, which describe how the
field responds to the current and how it evolves. For the moment, we ignore the external currents
and look at Sf ,
Z
Sf = d4 rL(Aµ , ∂ν Aµ ), (3.1)

where A and ∂ν A are functions of R the four-position r. Here L is not actually the Lagrangian,
3
it is the Lagrangian density, and d rL is the Lagrangian. However, people typically refer to
it as the Lagrangian anyway. A difference between the Lagrangian for a point particle is that
L is a function of all four derivatives, ∂t Aµ , ∂x Aµ , ∂y Aµ and ∂z Aµ . Again, we start with the
condition of minimizing the action,
Z  
4 ν
∂L ν
∂L
δS = d r (∂µ δA ) + (δA ) (3.2)
∂∂µ Aν ∂Aν
Z  
∂L ∂L
= d r δA −∂µ
4 ν
+ .
∂∂µ Aν ∂Aν

Minimizing the action gives Lagrange’s field equations,

∂L ∂L
∂µ = . (3.3)
∂∂µ Aν ∂Aν

Again, this assumes that L has no explicit dependence on r, as it only depends on r through its
dependence on A and derivatives of A. The only visual difference between the usual Lagrangian
equations and what we see here is that d/dt is replaced by ∂µ .
Our next step is to write the Lagrangian density for free fields. The form must be Lorentz invari-
ant, and must ultimately lead to the usual expressions for the energy density. Another criteria
is that it is gauge-invariant, i.e. that it should depend only on F µν , and not A. The choice that
works is
1 1
L = − F µν Fµν = F µν Fνµ . (3.4)
16π 16π

24
PHY 841 3 DYNAMIC ELECTROMAGNETIC FIELDS

The last step used the fact that F µν = ∂ µ Aν − ∂ ν Aµ is anti-symmetric. From Lagrange’s field
equations, Eq. (3.3),

∂L 1 ∂
∂µ = ∂µ (∂ α Aβ − ∂ β Aα )(∂α Aβ − ∂β Aα ) (3.5)
∂∂µ Aν 16π ∂∂µ Aν
1
= ∂µ (F µν ) = 0.
4π

If one adds the interaction between the fields and particles,

Z
Sf m = d4 r J · A, (3.6)

Lagrange’s field equations then become

1
∂µ (F µν ) = J ν . (3.7)
4π
Here, the current J α is the current density which is a four-vector. For charged particles moving
in a small volume Ω,
1 X uα
a
Jα = qa . (3.8)
Ω a∈Ω γa

For J 0 this u0 /γ = 1 and one sees that J 0 is the charge density. For the spatial components,
u
⃗ /γ = ⃗ v and one sees that J i is the current density. One can also check to see that J α (r) is a
four-vector by writing it as
XZ
α
J (r) = dτa δ 4 (ra − r)qa uα
a, (3.9)
a
X qa uα
= δ 3 (r⃗a (r0 ) − ⃗
r) a
.
a
γ

If one averages this over some small volume Ω by integrating over the volume and dividing by
Ω, one obtains Eq. (3.8). Thus, Eq.s (3.8) and (3.9) are equivalent expressions of the four-current.
⃗ and B,
Using Eq. (2.41) to express F µν in terms of E ⃗ these four equations (one equation for each
value of ν) then become

∇·E ⃗ = 4πJ 0 , (3.10)

(∇ × B)
⃗ − ∂t E
⃗ = 4π J.
⃗

The zeroth component of the current is the charge density, so this are simply a statement of two
of Maxwell’s equations. To obtain the other two equations we consider the tensor

1
F̃ µν = eµναβ Fαβ , (3.11)
2

25
PHY 841 3 DYNAMIC ELECTROMAGNETIC FIELDS

with F̃ known as the dual electromagnetic tensor. One can them see that
1
∂µ F̃ µν = − ϵνµαβ ∂µ (∂α Aβ − ∂β Aα ) (3.12)
2
= 0.

Equating to zero comes from the fact that two derivatives, ∂µ ∂α or ∂µ ∂β , are contracted through
the Levi-Civita tensor, which being anti-symmetric must cancel the contribution from the deriva-
tives. Using Eq. (2.41) one can see that
 
0 −Bx −By −Bz
 Bx 0 −Ez Ey 
F̃ αβ =   By E z
. (3.13)
0 −Ex 
Bz −Ey Ex 0

⃗ and B,
Expressing Eq. (3.12) in terms of E ⃗

∇·B
⃗ = 0, (3.14)
⃗ +∇×E
∂t B ⃗ = 0.

Summarizing, Maxwell’s equations in covariant notation are

∂α F αβ = 4πJ β , (3.15)
∂α F̃ αβ = 0.

For a point charge within a sphere of radius R, Gauss’ law says that
Z I
3
4π d r J + 0
dA ⃗ · E.
⃗ (3.16)

Using symmetry, E ⃗ = E r̂, and combined with the recognition that integrating the charge den-
sity inside the sphere gives the charge
Z
Q = d3 rJ 0 , (3.17)

one finds

4πQ = 4πR2 E, (3.18)

Q
E = .
R2
Compared to the expressions of Coulomb’s law to which one is usually accustomed, there is
no 4πϵ0 in the denominator or constant k in the denominator. This is due to the way in which
charge is defined. For example, if you read texts in nuclear physics you will typically find the
Coulomb energy for a proton near a nucleus of atomic number Z written as

Ze2
PE = . (3.19)
r

26
PHY 841 3 DYNAMIC ELECTROMAGNETIC FIELDS

If one wishes the energy to be in MeV, and the radius r to be in Fermi (aka femtometers), one
uses the fact that
ℏc
e2 = , (3.20)
137.036
where e2 /ℏc = 1/137.036 is known as the fine structure constant, which is dimensionless, and
ℏc = 197.327 MeV fm. In other fields one might more typically see a more arbitrary definition of
charge, e.g. Coulombs. In such cases, Coulomb’s law must then be modified by some prefactor,
e.g. 1/4πϵ0 .

3.2 Pseudo-Vectors and Pseudo-Scalars

The vector potential Aα is a four-vector, and the field tensor F αβ is a second-rank tensor. The
⃗ = −∇A0 + ∂t A
electric field E ⃗ is a 3-vector, but one that transforms as part of a tensor if the
transformation involves a boost. The magnetic field B ⃗ = ∇×A ⃗ is a pseudo-vector. The "pseudo"
comes from the fact that its definition, Bi = ϵijk ∂j Ak , involves two-vectors, so even though it
has only one vector index, it does not switch sign under parity (x → −x, y → −y, z → −z).
Tensors can behave rather strangely under parity because some of the elements don’t change
under parity while others do. One can define the pseudo-scalar

F αβ F̃αβ = −4E
⃗ · B.
⃗ (3.21)

Like scalars, this is manifestly invariant under Lorentz transformation, however it is odd under
parity. This can also be written as
1
F αβ F̃αβ = ϵαβγδ Fαβ Fγδ (3.22)
2
1
= ϵαβγδ (∂α Aβ )(∂γ Aδ ).
2
Each term has a product of four four-vector components, one being a zeroth component and the
other three being spatial. Thus, this quantity is odd under parity.
Finally, another obvious example of a scalar, which is even under parity, is

F αβ Fαβ = −2(|E|
⃗ 2 − |B|
⃗ 2 ). (3.23)
⃗ and B
Thus, although E ⃗ mix under boosts, the difference of their magnitudes remains fixed.
The sign of a pseudo-vector or pseudo-scalar changes if one changes from a right-handed to
a left-handed coordinate system. This is because ϵijk was arbitrarily defined so that ϵxyz was
positive. Even though magnetic forces feature pseudo-vectors, the interaction conserves parity,
i.e. it does not matter whether you used a right-handed or left-handed coordinate system. This
is because the force, which is something you can observe, behaves as ∇ × B, ⃗ Fi = qϵijk vj Bk .
Thus, in considering a force, the Levi-Civita symbol appears twice, once in defining B⃗ and once
in defining the force. Thus, one would see the same effect using a left-handed coordinate system.
The weak interaction does not conserve parity. If one aligns a nucleus so that its angular mo-
mentum, which is a pseudo-vector, points along the positive ẑ, and if that nucleus undergoes

27
PHY 841 3 DYNAMIC ELECTROMAGNETIC FIELDS

a weak-interaction decay (beta decay), the electrons are emitted with strong preference parallel
to the angular momentum. This violates parity because the direction of the electrons is real, it
makes a difference if the electrons go up vs. down, but the direction of the angular momen-
tum (or the magnetic field used to polarize the atoms) is a pseudo-vector and dependent on the
right-handed vs. left-handed choice of coordinate system.

3.3 The Stress-Energy Tensor of the Electromagnetic Field

Just like the energy, H = π q̇ − L, is conserved for a usual Lagrangian, one can define a sim-
ilar quantity for the Lagrangian density. However, this quantity will be a second-rank tensor
and will express local conservation of both energy and momentum. Just to reduce the index
overload, we consider a function L(ϕ, ∂α ϕ). We can replace ϕ with Aµ , but essentially the
proof will be the same aside from summing over each of the four fields Aµ . First, we define the
stress-energy tensor T αβ ,

T αβ = π α ∂ β ϕ − g αβ L, (3.24)
∂L
πα ≡
∂(∂α ϕ)
When α = β = 0, this looks like the usual definition of energy, except this is the Lagrangian
density so T 00 has dimensions of energy per length cubed. To see how this is related to a con-
served quantity, consider
∂L ∂L
∂α T αβ = (∂α π α )(∂ β ϕ) + π α (∂α ∂ β ϕ) − ∂βϕ − ∂α ∂ β ϕ. (3.25)
∂ϕ ∂(∂α ϕ)
The first and third terms vanish via Lagrange’s field equations, Eq. (3.3), and the second and
fourth terms cancel via the definition of the conjugate momentum, π α , in Eq. (3.24). Thus,

∂α T αβ = 0. (3.26)

Any function J µ that satisfies the equation of continuity ∂µ J µ = 0 implies that J is a conserved
current density. For such a four-vector, the zeroth component is the charge density and the three
spatial components J i are current densities. The conservation ensues because
Z Z
∂t d rJ0 = − d3 r∇ · J.
3 ⃗ (3.27)
R
The latter terms vanishes because the currents vanish at infinity so the net charge Q = d3 rJ0
is conserved. The conservation is local because if one considers a small volume, the change in
the charge equals the flux of current through the boundary, also seen via the divergence theorem
(aka Gauss’ law). The “charge” needn’t be electric charge but any conserved quantity.
In our case, for each value of β in T αβ one has a conserved four-current. Thus, T 00 , T 01 , T 02 and
T 03 represent densities of conserved quantities. Those quantities are the energy and momentum
densities. The quantities T 10 , T 20 and T 30 represent the flux of energy density, i.e.
I I
∂t d r T00 = − dA
3 ⃗ · J,
⃗ Jα = T α0 . (3.28)

28
PHY 841 3 DYNAMIC ELECTROMAGNETIC FIELDS

The three quantities T 0i represent momentum densities. It is not obvious, but the stress-energy
tensor is symmetric, and the flux of the energy, T i0 equals the momentum density, T 0i . To prove
the symmetry once must consider the effect of an asymmetric term (further ahead). In this case
the one finds an infinite angular acceleration, and angular momentum is not conserved.
The other nine components of the stress-energy tensor represent the flux of momentum. For
example, the quantity T xy dSy represents the rate momentum Px flows through a surface ele-
ment of area dSy which points in the y direction. This would be a shear. In hydrodynamics, this
vanishes and in the frame of the fluid (where T 0i = 0) T ij = P δij , where P is the pressure.
For field equations, the spatial components T ij simply represent momentum fluxes and can be
quite complicated, thought the tensor does have to remain symmetric.
Now, to return to the specific case of the electromagnetic field. In that case, the quantities π α
and T αβ discussed above are functions of four fields and to express the conjugate momenta or
the stress-energy tensor, one must simply extend the above relations to sums over all four fields.
There are four conjugate momenta for each field, and because Aα is effectively four fields, there
are four conjugate momenta for each α and the momenta are represented by a four-by-four
tensor.
∂L
π αβ = , (3.29)
∂(∂α Aβ )
T αγ = π αβ ∂ γ Aβ − g αγ L.

Using L = − 16π
1
F αβ Fαβ ,

−1
π αβ = F αβ , (3.30)
4π
−1 1
T αγ = F αβ ∂ γ Aβ + g αγ F µν Fµν .
4π 16π
However, as previously stated the stress-energy tensor should be symmetric. To make it sym-
metric, we add a derivative of the form ∂µ Gµ , which by Gauss’s law will integrate to zero if
one considers all space. Thus if we add the term, (1/4π)∂β (F αβ Aγ ), the stress-energy tensor
becomes
−1 1 1
T αγ = F αβ ∂ γ Aβ + g αγ F µν Fµν + ∂β (F αβ Aγ ) (3.31)
4π 16π 4π
1 1 αγ µν 1
= F αβ Fβ γ + g F Fµν + (∂β F αβ )Aγ .
4π 16π 4π
The last term vanishes for free fields because ∂α F αβ = 0 when there are no currents present, so
1 1
T αγ = F αβ Fβ γ +g αγ F µν Fµν . (3.32)
4π 16π
In addition to being symmetric, the tensor is traceless, T αα = 0. Expressing the components in

29
PHY 841 3 DYNAMIC ELECTROMAGNETIC FIELDS

⃗ and B,
terms of E ⃗
1
T 00 = (E 2 + B 2 ), (3.33)
8π
1
T 0i = ϵijk Ej Bk ,
4π
1
T ij = −T i j = (δij (E 2 + B 2 ) − 2Ei Ej − 2Bi Bj ).
8π
⃗×
The momentum density, or energy flux is T 0i or writing the three components as a vector, (E
⃗
B)/4π. This is known as the Poynting vector.
To explain why the stress-energy tensor must be symmetric, consider an infinitesimal cube of
dimension a × a × a. Consider rotation about the z axis. The shear forces, Ti̸=j , contribute
to the angular momentum. The forces on the four sides due to Txy and Tyx are Tyx a2 x̂ on the
upper face, −Tyx a2 x̂ on the lower face, a−Txy a2 ŷ on the right face and Txy a2 ŷ on the left-side
face. The net torque is thus
τ = a3 (Tyx − Txy ). (3.34)
However, the moment of inertia scales as a5 , so the angular acceleration would scale as 1/a2 →
∞, unless the tensor is anti-symmetric. To understand why T0i = Ti0 , one can consider a boost.
Under boosts, any such asymmetry would translate into an asymmetry in the ij components.
One can also show explicitly that the stress-energy tensor is conserved in the presence of in-
teractions with currents. Beginning with the definition for the field contribution, T (f ) , in Eq.
(3.32),
1 αγ β 1 αβ µν
T (f )αβ = F Fγ + g F Fµν , (3.35)
4π 16π
1 1 αγ 1 αγ β
∂α T (f )αβ = (∂α F αγ )Fγ β + F ∂α Fγ β + F ∂ Fαγ ,
4π  4π 8π 
F αγ 1 β 1 β
γ
= J Fγ + β
∂α ∂γ A − ∂α ∂ Aγ + ∂ ∂α Aγ − ∂ ∂γ Aα
β β
4π 2 2
αγ  
F 1 1
= J γ Fγ β + ∂α ∂γ Aβ − ∂ β ∂α Aγ − ∂ β ∂γ Aα
4π 2 2
The terms inside the parenthesis vanish because they are explicitly symmetric in the αγ indices,
which when contracted with the anti-symmetric tensor F αγ , must vanish. Thus,
∂α T (f )αβ = Jγ F γβ . (3.36)
Next, we consider the contribution to the stress energy tensor from the matter contribution. To
do this we consider matter to have a mass density
X
µ(⃗r) = ma δ(⃗r−⃗ ra ), (3.37)
a

which given that the total mass is conserved, for free particles, one can write a conserved mass
current,
J 0 = µ(⃗
r ), (3.38)
X
J⃗α (r) = r−⃗
ma δ(⃗ ra (r0 ))uα (r)/u0 (r).
a

30
PHY 841 3 DYNAMIC ELECTROMAGNETIC FIELDS

To see that J is indeed a four-vector, one could write

X Z
α
J (r) = ma dτa δ(r0 − ta )δ(⃗
r−⃗
ra (ta ))uα
a (3.39)
a
X
= r−⃗
ma δ(⃗ ra (r0 ))uα
a /γa
a
X
= r−⃗
ma δ(⃗ ra (r0 ))uα (r)/u0 (r).
a

Here, the four-velocity dependence on ⃗ r is some complicated function that has the correct veloc-
ity for each particle a when r = ra . The mass current is conserved, ∂ · J = 0. The stress energy
tensor can be written as
T (m)αβ (r) = J α (r)uβ (r), (3.40)
∂α T (m)αβ = (∂ · J )uβ + (J · ∂)uβ (r).
The first term vanishes due to current conservation, and the second term becomes
X
∂α T (m)αβ = ra (t) − ⃗
ma δ(⃗ r )(1/u0a )(ua · ∂)uβa . (3.41)
a

The derivative ua · ∂ is a Lorentz invariant, so it can be considered in the rest frame. In that
frame it becomes d/dτa , where τa is the proper time for the particle a. From the equations of
motion for a particle of charge qa in an electromagnetic field, Eq. (2.42),
X
∂α T (m)αβ = ma δ(⃗ra (t) − ⃗
r )qa F βγ ua,γ /(u0a ma ) (3.42)
a
X
= ra (t) − ⃗
qa δ(⃗ r )F βγ ua,γ /u0a
a
= F βγ
Jγ = −Jγ F γβ .
Thus, using Eq. (3.36), the sum of the field and matter contribution vanishes,


∂α T (f )αβ + T (m)αβ = 0. (3.43)
Conspicuous by its absence is the part of the action that represents the coupling between the
current and the field, −J · A. Indeed this would contribute a third portion of the stress-energy
tensor. However, when we wrote down the field-part of the contribution there was a step where
a term J α Aγ was discarded due to being in a field-free region. If included, it would cancel the
contributions to the energy density from the J · A term. The “field” energy effectively accounted
for the J · A part of the Lagrangian by ignoring the last term in Eq. (3.32). The energy density,
T00 is thus
1
T 00 = ⃗ 2 + |B|
(|E| ⃗ 2 ) + T (m)00 , (3.44)
8π
where the matter contribution is the kinetic energy of the particles. For a static charge distribu-
tion, one expected the energy to have terms of the form,
X q a qb
∆P E = . (3.45)
a<b
|⃗
ra − ⃗rb |

31
PHY 841 3 DYNAMIC ELECTROMAGNETIC FIELDS

The mystery of the missing interaction comes from showing that the change of the electric field
energy due to bringing the two charges from infinity to their current locations is precisely the
expected value. To see this, we can consider the field energy of two charges qa and qb brought
to within a relative distance R. The field energy is
Z
1
U (f )
= d3 r (E ⃗a + E⃗ b )2 , (3.46)
8π
where E ⃗ a,b are the electric fields due to the two charges. We are interested to the change in U (f )
due to moving the charges, and can thus worry only about the term from E ⃗a · E
⃗ b,
Z
1
∆U (f )
= d3 r E⃗a · E⃗b (3.47)
4π
Z
qa q b r · (⃗
⃗ r − R)
⃗
= d3 r
4π r 3 |⃗
r − R|
⃗ 3
Z !
qa q b 1 1
= d3 r ∇ · ∇
4π r |⃗
r − R|⃗
Z !
qa qb 1 1
= − d3 r ∇2
4π r |⃗
r − R| ⃗
Z
qa q b 1
= d3 r 4πδ(⃗ r − R)⃗ (3.48)
4π r
qq qb
= .
R
For an array of charges, the potential energy is a then the sum over the potential energy of each
pair, as expected.
One can also quickly show the usual description of the energy being related to the charge density
convoluted with the the electric potential,
Z
1
U (f )
= ⃗ ·E
d3 r E ⃗ (3.49)
8π Z
1
= d3 r(−∇A0 (⃗ r )) · E
⃗
8π Z
1
= d3 rA0 (⃗r )∇ · E⃗
8πZ
1
= d3 rA0 (⃗
r )J0 (⃗
r ).
2
The factor of 1/2 accounts for double counting the contributions from pairs of charges.

3.4 Hyper-Surfaces and Conservation of Energy, Momentum and Angular

Momentum
The energy and momentum, P α , in a three-dimensional hyper-surface element Ωγ is
I
α
PΩ = dΩγ T γα . (3.50)
Ω

32
PHY 841 3 DYNAMIC ELECTROMAGNETIC FIELDS

Here Ωγ is a a region of four-space. If γ = 0 this corresponds to a region with fixed time, and
is thus a volume, whereas if γ = 1, 2, 3 would correspond to a surface area multiplied by some
duration in time. More formally,
Z
Ωγ = d4 r∂γ C(r). (3.51)

Here, C is some function that is unity in some region and zero outside. For instance, C =
Θ(r0 − t) would correspond to a hyper-surface (volume in this case) at fixed time t. Only the
γ = 0 component would then be non-zero. If one chose C = Θ(rx − x), the hypersurface
would be an area for fixed x multiplied by the the entire time. The vector P α is the energy or
momentum that traverses the hyper-surface Ω. As an example, consider the function
C(r) = Θ(r0 − t0 )Θ(tf − r0 )Θ(R2 − rx2 − ry2 − rz2 ), t f > t0 . (3.52)
This is a sphere that appears at r0 = t0 and disappears at r0 = tf . For the contribution
from r0 = t0 , one has a contribution dP α = dΩ0 T 0α = d3 r T 0α (⃗ r , t0 ), which is the en-
ergy/momentum that appears in the volume at that time. For t0 < r0 < tf the differential
contributions come as dP α = dSi dt T iα (|⃗ r | = R, t), and represent/energy and momen-
tum that flows in/out of the sphere during that time. Finally, the contribution for r0 = tf is
dP α = −d3 rT 0α (⃗r , tf ) is the loss of the remaining energy/momentum once the sphere disap-
pears. The sum of these components must be zero, which can be seen by the divergence theorem
in four dimensions,
I Z
αβ
dΩα T = d4 r(∂α C(r))T αβ (3.53)
Z
= − d4 rC(r)∂α T αβ = 0.

Thus, stress energy tensor element T αβ represents the flow of momentum P α through the hyper-
surface element dΩβ . Similarly the current J β represents the flow of charge through the hyper-
surface element dΩβ .
This also works for angular momentum. As stated earlier the angular momentum is Lαβ =
r α P β − P α r β . The angular momentum flux tensor will will require an additional component to
represent the angular momentum that travels through a hyper-surface element dΩα . Thus, we
define
M αβγ = r α T βγ − r β T αγ , (3.54)
dLαβ = M αβγ dΩγ.
The choice for M is motivated by the fact that if Ω is purely time-like, dΩ0 = d3 r, then M αβ0
indeed looks like the angular momentum density. One can also check this further by testing
whether ∂γ T αβγ = 0, which should be true for local conservation of angular momentum,
∂γ M αβγ = ∂γ (r α T βγ − r β T αγ ) (3.55)
= (∂γ r α )T βγ − (∂γ r β )T αγ
= gγα T βγ − gγ β T αγ
= T βα − T αβ = 0.
In fact, if one believes in angular momentum conservation, this is proof that the stress-energy
tensor is symmetric.

33
PHY 841 3 DYNAMIC ELECTROMAGNETIC FIELDS

3.5 Homework Problems

1. Consider a sphere of radius R and charge Q, where the charge is spread uniformly through-
out the sphere.

(a) Find the strength of the electric field as a function of r.

(b) Find the electric potential as a function of r.
(c) Find the potential energy required to move the charges to their positions,
Z
1
PE = d3 r ρ(r)V (r).
2

(d) Find the energy contained in the electric fields.

⃗ and B,
2. Beginning with F αβ written in term of E ⃗ restate

∂α F αβ = 4πJ β
∂α F̃ αβ = 0

as

∇·E
⃗ = 4πρ,
∇×B⃗ = ⃗ + 4π J,
∂t E ⃗
∇·B⃗ = 0,
∇×E⃗ = −∂t B.
⃗

3. For a charge-free region, J α = 0

⃗ and show the speed of
(a) Use Maxwell’s equations to write a wave equation for E,
propagation is unity (c).
(b) For a wave traveling in the ẑ direction with the electric field in the ±x̂ direction, write
a solution for the propagating plane wave for both E(⃗ ⃗ r , t) and B(⃗
⃗ r , t).

4. First calculate ℏc in standard mks units. Then, using the fact that the charge on an electron
is 1.602 × 10−19 Coulombs, find the constant k in mks units used in Coulomb’s law,
P E = kq 2 /r. Use the fact that P E = e2 /r, where e2 = ℏc/137.036.

5. Consider two very large parallel capacitor plates of area A, carrying charge densities σ
and −σ, and oriented perpendicular to the z axis. The plates are initially at a very small
separation at t = 0, but are pulled apart, moving with constant non-relativistic velocities
v/2 and −v/2.

(a) What is the electric field between the plates?

(b) Find all four non-zero elements of the stress-energy tensor (Txx , Tyy , Tzz and T00 ).
Check that the stress-energy tensor is traceless.

34
PHY 841 3 DYNAMIC ELECTROMAGNETIC FIELDS

(c) In hydrodynamics, the work done by an expanding a gas is P dV . Here, because the
expansion is along the z axis the work is Tzz dV . What is the power required to pull
the plates apart at these velocities?
(d) What is energy density of the field between the plates?
(e) What is the rate (energy per time) at which the field energy between the plates in-
creases due to the growing volume?

35
PHY 841 4 ELECTROSTATICS

4 Electrostatics

Here, we consider the electric field of fixed charge distributions. All currents are set to zero, so
there is only electric field, and all time derivatives in Maxwell’s equations are neglected.

4.1 Gauss’s Law

Beginning with

∇·E
⃗ = 4πρ, (4.1)

one can integrate over a volume, then use the divergence theorem (also known as Gauss’s theo-
rem) to find Gauss’s law
I I
d r ∇ · E(⃗
3 ⃗ r ) = 4π d3 r ρ(⃗ r ), (4.2)
I
dA⃗·E ⃗ = 4πQ. (4.3)

For a point charge, define the volume as a sphere of radius r surrounding the charge Q, one then
quickly finds Coulomb’s law.

4πr 2 Er = 4πQ, (4.4)

Q
Er = 2 . (4.5)
r

4.2 Potential Energy of a Fixed Continuous Charge Distribution

For a continuous charge density ρ(⃗
r ), the potential energy required to bring the last bit of charge
δQ to a position r is

δP E = Φ(⃗
r )δQ, Φ = A0 , (4.6)
Z
δQ
= d3 r ′ ρ(⃗
r′ ) .
|⃗
r−⃗ r′ |

It would beRtempting to write the entire potential energy as a sum over all δQ, or as a separate
integral of d3 r ρ(⃗ r ), but that would lead to double counting. The double counting would
come from considering the effect of brining a differential charge δQ = d3 rρ(⃗ r ) towards each
′ 3 ′ ′
differential charge δQ = d r ρ(⃗ r ), and the opposite. Thus, one introduces a factor of 1/2
when writing the entire potential energy,
Z
1 3 3 ′
ρ(⃗ r′ )
r )ρ(⃗
PE = d rd r (4.7)
2 |⃗
r−⃗
r′ |
Z
1
= d3 r ρ(⃗
r )Φ(⃗
r ). (4.8)
2

36
PHY 841 4 ELECTROSTATICS

Example 4.1:
Find the net potential energy for a charge Q uniformly spread out in a sphere of radius R.
First, find the potential Φ(r). For r > R, it is easy, Φ = Q/r. For r < R, you need to first find
the electric field. Beginning with Gauss’s law,

Qr 3 /R3 Qr
E = = ,
r2 R3
Z R
Q
Φ = + drE(r)
R r
Q 1 Q
= + (R2 − r 2 )
R 2 R3
3Q Qr 2
= − .
2R 2R3
Next, integrate over the charge density, 3Q/4πR3 , to get the potential energy,
Z R  
1 3Q 3Q Qr 2
PE = 2
4πr dr −
2 4πR3 0 2R 2R3
3Q2
= .
5R

4.3 Laplace’s Equations and a Fixed Point Charge

First, we consider expressions, mainly for the electric potential Φ = A0 , where nothing changes
with time. In that case, Maxwell’s equation all terms with ∂t Something are set to zero,

∇·E⃗ = 4πρ, (4.9)

E⃗ = −∇Φ,
∇2 Φ = −4πρ.

If there is no charge density one is left with Laplace’s equation,

∇2 Φ = 0. (4.10)

This is applicable for any region with no charge density, and if charges exist outside the region,
one must solve Laplace’s equations with boundary conditions.
The most obvious example of a field free region, with a charge confined outside, is that of a point
charge Q at the origin. In that case ρ = 0 for r > ϵ. Gauss’s law, combined with symmetry
gives

4πR2 |E|
⃗ = 4πQ, (4.11)
Q
E⃗ = r̂.
r2

37
PHY 841 4 ELECTROSTATICS

The potential must then be

Q
Φ = . (4.12)
r
From this constraint, the identity
 
1
∇ 2
= −4πδ(⃗
r) (4.13)
r

becomes manifest. Thus, the function Q/r is a solution to Laplace’s equation in the region r > ϵ
satisfying the boundary condition that the electric flux entering the region is 4πQ.

4.4 Laplace’s Equations in Cartesian Coordinates

In Cartesian coordinates separation of variables assumes that Φ is a product of three pieces,

Φ(x, y, z) = ψx (x)ψy (y)ψz (z) (4.14)

∂x2 ψ(x) = −kx2 ψ(x),
∂y2 ψ(y) = −ky2 ψ(y),
∂z2 ψ(z) = −kz2 ψ(z),
∇2 Φ = −(kx2 + ky2 + kz2 )Φ = 0,
kx2 + ky2 + kz2 = 0.

With these eigenvalues,

ψx (x) = Ax eikx x + Bx e−ikx x , (4.15)

ψy (y) = Ax eiky y + Bx e−iky y ,
ψz (z) = Az eikz z + Bz e−ikz z ,

with Ai and Bi being arbitrary constants chosen to fit the boundary conditions. Because kx2 +
ky2 + kz2 = 0, at least one of the wave numbers must be complex.

4.5 Laplace’s Equations and Solutions in Spherical Coordinates

To obtain Laplace’s equation in spherical coordinates, we first write the gradient operator in
those coordinates,
dr dθ dϕ
∇ = r̂ ∂r + θ̂ ∂θ + ϕ̂ ∂ϕ , (4.16)
dℓr dℓθ dℓθ
where the small steps dℓi represent a Cartesian coordinate system in those directions. The ge-
ometry,

dℓr = dr, dℓθ = rdθ, dℓϕ = r sin θ, (4.17)

38
PHY 841 4 ELECTROSTATICS

gives
1 1
∇ = r̂∂r + θ̂ ∂θ + ϕ̂ ∂ϕ . (4.18)
r r sin θ
Next, before finding the Laplacian, ∇ · ∇, we realize that the unit vectors themselves depend
on the angle. Thus, when one takes the second round of derivatives w.r.t. θ and ϕ, one must
remember that θ̂ and ϕ̂ depend on angle. For small changes in the angles

r̂ = r̂0 + θ̂0 δθ + ϕ̂0 sin θδϕ, (4.19)

θ̂ = θ̂0 − r̂0 δθ + ϕ̂0 cos θδϕ,
ϕ̂ = ϕ̂0 − r̂0 sin θδϕ − θ̂0 cos θδϕ.

Although we will set δθ and δϕ to zero, that will only be after taking the second divergence.
The gradient operator is then
 
1 cos θ
∇ = r̂0 ∂r − δθ ∂θ − δϕ ∂ϕ (4.20)
r r sin θ
 
1 cos θ
+θ̂0 ∂θ + δθ∂r − δϕ ∂θ
r r sin θ
 
1 1
+ϕ̂0 ∂ϕ + δϕ sin θ∂r + δϕ cos θ ∂θ .
r sin θ r
We will set δθ = δϕ = 0 after taking the second round of derivatives, so we need only worry
about the term with δθ term in the part proportional θ̂ and the δϕ term in the part proportional
to ϕ̂ because ∂θ δθ = ∂ϕ δϕ = 1 while all others are zero. Thus,
 
1 2 1 1 1 1
∇ · ∇ = ∂r + 2 ∂θ + 2
2 2
∂ + ∂r + sin θ∂r + cos θ ∂θ (4.21)
r r sin2 θ ϕ r r sin θ r
2 1 cos θ 1
= ∂r2 + ∂r + 2 ∂θ2 + 2 ∂θ + 2 ∂2,
r r r sin θ r sin2 θ ϕ
1 1 1
∇2 = 2 ∂r (r 2 ∂r ) + 2 ∂θ (sin θ∂θ ) + 2 2
2 ∂ϕ . (4.22)
r r sin θ r sin θ

Next, to solve ∇2 Φ = 0, we first write Φ as a product of three pieces,

Φ(r, θ, ϕ) = R(r)P (θ)Q(ϕ). (4.23)

If one assumes that each function satisfies the following differential equations,

∂ϕ2 Qm (ϕ) = −m2 ϕ, (4.24)

1
∂r (r 2 ∂r R) = λR,
r2
1 m2
∂θ (sin θPλm (θ)) − Pλm (θ) = −λPλm (θ)
sin θ sin2 θ
Laplace’s equations will be satisfied. Because the function must be periodic, m is an integer. The
solutions R(r) behave as r n and λ = n(n + 1). For positive values, n = ℓ, one finds the same

39
PHY 841 4 ELECTROSTATICS

λ as for n = −ℓ − 1. Thus, we switch labels from λ to ℓ, with λ = ℓ(ℓ + 1), and one can write
the general radial solutions as

Rℓ (r) = Ar ℓ + Br −ℓ−1 , (4.25)

1
∂θ (sin θ∂θ Pℓm (θ)) = (−ℓ(ℓ + 1) + m2 )Pℓm (θ).
sin θ
The functions Pℓ,m (θ) are known as associated Legendre polynomials, and the products

Yℓ,m (θ, ϕ) = Pℓ,m (θ)eimϕ , (4.26)

1 1
∂θ (sin θYℓ,m (θ)) + ∂ϕ2 Yℓ,m (θ, ϕ) = ℓ(ℓ + 1)Yℓ,m (θ, ϕ).
sin θ sin2 θ
are referred to as spherical harmonics. The functions are orthonormal,
Z
dϕ d cos θ Yℓ,m (θ, ϕ)Yℓ′ ,m′ (θ, ϕ) = δℓℓ′ δmm′ . (4.27)

Recurrence relations allow one to generate solutions for Yℓm for a given ℓ and m from solutions
for lower ℓ or m. The operators, L+ and L− change Pℓ,m to Pℓ,m±1 and are known as raising
and lowering operators,

L± = −e±iϕ (±i∂θ − cot θ∂ϕ ) , (4.28)

L± Pℓm±1 (θ, ϕ) = [ℓ(ℓ + 1) − m(m ± 1)] Pℓm (θϕ).

Checking this relation is a homework problem. One can see that because both ℓ and m are
integers that |m| ≤ ℓ, i.e. there 2ℓ + 1 values of m for each ℓ, from m = −ℓ to m = ℓ. Simple
expansions provide the form Pℓm=0 . Beginning with the definition of Legendre polynomials,

1
Pℓ (cos θ) ≡ √ Pℓ,m=0 (θ, ϕ), (4.29)
2ℓ + 1
one can express
 
∂x (1 − x2 )∂x Pℓ (x) = −ℓ(ℓ + 1)Pℓ (x), (4.30)
ℓ  2
1 X ℓ!
Pℓ (x = cos θ) = n (x − 1)ℓ−k (x + 1)k ,
2 k=0 (ℓ − k)!k!
s
4π
Pℓ (cos θ) = Yℓm=0 (θ).
2ℓ + 1

One can then use the raising and lowering operators to find expressions for Yℓm for any m.

40
PHY 841 4 ELECTROSTATICS

Various Yℓm (θ, ϕ):

1
Y0,0 = √ , (4.31)
4π
r
3
Y1,0 = cos θ,
4π
r
3
Y1,±1 = ∓ sin θei±ϕ ,
r 8π
5
Y2,0 = (3 cos2 θ − 1),
16π
r
15
Y2,±1 = ∓ sin θ cos θe±iϕ ,
r 8π
15
Y2,±2 = sin2 θe±2iϕ .
32π
Various Pℓ (x):

P0 (x) = 1, (4.32)
P1 (x) = x,
1
P2 (x) = (3x2 − 1),
2
1
P3 (x) = (5x3 − 3x),
2
1
P4 (x) = (35x4 − 30x2 + 3).
8
For negative m, one can use the identity,
∗
Yℓ−m (θ, ϕ) = (−1)m Yℓm (θ, ϕ). (4.33)

Legendre polynomials satisfy a number of identities,

Pℓ (x = 1) = 1, (4.34)
Z 1
2
dx Pℓ (x)Pℓ′ (x) = δℓℓ′ ,
−1 2ℓ + 1
X
(2ℓ + 1)Pℓ (x)Pℓ (x′ ) = 2δ(x − x′ ),
ℓ
d
(2ℓ + 1)Pℓ (x) = [Pℓ+1 (x) − Pℓ−1 (x)] ,
dx
(ℓ + 1)Pℓ+1 (x) = (2ℓ + 1)xPℓ (x) − ℓPℓ−1 (x),
1 dℓ
Pℓ (x) = ℓ (x2 − 1)ℓ (Rodriguez formula),
2 ℓ! dxℓ
1 X
√ = Pℓ (x)tℓ (generating function).
1 − 2xt + t 2
ℓ

41
PHY 841 4 ELECTROSTATICS

4.6 Solutions to Laplace’s Equations in Cylindrical Coordinates

p
In cylindrical coordinates, ρ ≡ x2 = y 2 , ϕ ≡ tan−1 y/x and z, Laplace’s equation becomes
 
1 1 2
∇ Φ =
2 2
∂ρ (ρ∂ρ )Φ + 2 ∂ϕ + ∂z Φ = 0, (4.35)
ρ ρ
Φ(ρ, ϕ, z) = R(ρ)Q(ϕ)Z(z).

First, choose Q and Z to be eigenstate of the corresponding part of the Laplacian,

∂z2 Z(z) = kz2 Z(z), (4.36)

∂ϕ2 Q(ϕ) = −m2 Q(ϕ).

With these choices,

Z(z) = Aekz z + Be−kz z , (4.37)

Q(ϕ) = eimϕ .

The equation for R(ρ), which depends on m and kz so it is labeled Rm (kz , ρ), is then
 
1 m2
∂ρ (ρ∂ρ ) − 2 + kz Rm (kz , ρ) = 0,
2
(4.38)
ρ ρ
   
1 m2
∂ρ + ∂ρ Rm (kz , ρ) + kz − 2 Rm (kz , ρ) = 0.
2 2
ρ ρ
If one makes the change of variables, and considers for R(x = kz ρ), the differential equation
becomes
   
1 m2
∂x Rm (x) + ∂x Rm (x) + 1 − 2 Rm (x) = 0.
2
(4.39)
x x
For the solution to have a form
X
∞
α
Rm (x) = x aj x2 , (4.40)
j=0

the lowest power of in the expansion, α, must either be |m| or −|m|. This can be seen by letting
k → 0 in Eq. (4.38), so that in Eq. (4.39) one can simplify the equation by replacing (1 − m2 /x2 )
with 1/x2 (see the end-of-chapter problem).
The solutions with each power are known as Jm (x) and Nm (x), with it being understood that
the index is positive. The general solution has arbitrary constants A and B which will be deter-
mined by boundary conditions,

Rm (x) = AJm (x) + BNm (x). (4.41)

As stated above, for small x and m > 0,

Jm (x) ∼ xm + · · · , (4.42)
Nm (x) ∼ x−m + · · · ,

42
PHY 841 4 ELECTROSTATICS

where the functions J and N are known as Bessel functions or Neumann functions respectively.
The number m is known as the order of the equation. From the differential, one can see quickly
derive the recursion relation for the coefficients am in Eq. (4.40),
1
a2j = − a2j−2 , (4.43)
4j(j + α)
with all the odd coefficients vanishing.
The large and small x expansions are:
 m
1 x
x ≪ 1, Jm (x) → , (4.44)
Γ(m + 1) 2
 2

ln x
+ 0.5772 · · · , m=0
Nm (x) → Γ(m) 2
m
π 2 ,
− π x
, ̸ 0.
m=
and the expansion for large x are:
r  
2 mπ π
x ≫ 1, m Jm (x) → cos x − − , (4.45)
πx 2 4
r  
2 mπ π
Nm (x) → sin x − − .
πx 2 4
Each of these expressions assumes m ≥ 0, and the constant 0.57772 is Euler’s constant.

4.7 Boundary Value Problems

Boundary value problems involve finding solutions for Laplace’s equations that satisfy the B.C.
for some region of space. The B.C. must be satisfied at all boundaries of the space. Often, the
boundaries are either a conductor, constant potential, or at infinity, with the potential either
vanishing or behaving with a known manner, e.g. becoming a constant electric field. If the
potential is defined at the boundary, this is known as a Dirichlet problem. Another option would
be to define the electric field or charge density at a boundary, or one can have some mixture.

4.7.1 Method of Images

This method can be applied in certain situation where charges are in the presence of conducting
surfaces that divide space into separate regions. Conducting surfaces have equipotential, and if
one is considering a given sub-space (e.g. the region above an infinite conducting plane), one can
attempt to consider how one could mimic the effect of the conducting surface by placing charges
in the other spaces (e.g. below the plane). The typical example is to consider a charge +Q above
an infinite grounded conducting plane defined by z = 0 at a postition x = y = 0, z = a. For
the z > 0 region, one could consider an imaginary charge −Q at a position x = y = 0, z = −a.
Clearly, this imaginary charge would lead to Φ(x, y, z = 0+) = 0. Thus, the overall potential
for z > 0 is
Q Q
Φ(x, y, z > 0) = − . (4.46)
|⃗
r − aẑ| |⃗
r + aẑ|

43
PHY 841 4 ELECTROSTATICS

For z < 0 the conducting plane shields the effect of the charge in the upper half plane, and
Φ(x, y, z < 0) = 0.
Charges are attracted to their images, thus a charge is attracted to a nearby conductor. This is
often relevant to accelerator design.

Example 4.2:
Consider a point charge +Q outside a grounded conducting sphere. The sphere has radius R
and is centered at the origin, and the point charge is at position aẑ.
Solution:
We will consider an image charge Qi place inside the sphere at position xi = yi = 0, zi = ai ,
so that the sum of the two potentials cancel at r = R,
Q Qi
Φ(r = R, θ) = q +q (4.47)
R2 sin2 θ + (a − R cos θ)2 R2 sin2 θ + (ai − R cos θ)2
Q Qi
= p + p
R 1 + a2 /R2 − 2(a/R) cos θ R 1 + a2i /R2 − 2(ai /R) cos θ

In order for the potential to vanish for all cos θ,

Q Qi
p = −p , (4.48)
2
1 + a /R 2 1 + a2i /R2
2a/R 2ai /R
= . (4.49)
1 + a2 /R2 1 + a2i /R2

The latter expression becomes a quadratic equation with two solutions, ai = a and ai = R2 /a.
The first solution is obvious – the image charge sits right on top of the of the real charge, but is
in the same region, so we neglect it. The second solution is the one we desire, and solving for
Qi ,
s
1 + a2i /R2
Qi = −Q (4.50)
1 + a2 /R2
s
1 + R2 /a2
= −Q
1 + a2 /R2
QR
= − .
a
The potential is zero inside the sphere.

4.7.2 Boundary Value Problems Using Cartesian Solutions to the Laplace Equation

Example 4.3:

44
PHY 841 4 ELECTROSTATICS

a) Consider an infinite plane defined by z = 0, where the potential has the form

Φ(x, y, z = 0) = V0 cos(qx). (4.51)

Assuming the remainder of the volume is vacuum, find the potential for all x, y, z ̸= 0.
Solution:
To match the B.C. at z = 0, and to decay to zero for large z, the solution is

Φ(x, y, z) = V0 cos(qx)e−q|z| . (4.52)

4.7.3 Boundary Value Problems Using Spherical Harmonics

Laplace’s equation is applicable in any charge-free region, but doesn’t mean it doesn’t apply
in problems with charge densities. You simply only use Laplace’s equation in the charge-free
part of the volume. There are a few, with emphasis on few, boundary-value problems one
can easily perform using the spherical harmonics mentioned before. The most common non-
trivial example is a conducting sphere in a constant electric field, which we work out below.

Example 4.4:
⃗ =
A conductor of radius R is placed at the origin in a field which previously was uniform, E
E0 ẑ.

1. Find the potential for r > R. This region has no charges, hence it satisfies Laplace’s
equations. For the solutions that behave as r ℓ , only the ℓ = 1 solution can appear, because
otherwise the solution would not look like a constant field at large r. Because a conductor’s
potential has to be constant at r = R, only the ℓ = 1 term of the 1/r ℓ+1 solutions can be
non-zero, because they need to cancel the r ℓ solutions to for all θ. Thus, the solution must
be of the form,
 
B
Φ = cos θ −E0 r + 2 , r > R,
r
B
E0 R = ,
R2
B = E0 R3 ,
 
R3
Φ(r > R) = −E0 cos θ r − 2 .
r

One can always add an arbitrary constant to the potential, which would correspond to the
ℓ = 0 term.

45
PHY 841 4 ELECTROSTATICS

2. Find the electric field for all r.

⃗ = −∇Φ
E
1
= E0 ẑ − r̂∂r Φ − θ̂ ∂θ Φ
 r   
2E0 R3 E0 R 3
= E0 ẑ + r̂ cos θ + θ̂ sin θ
r3 r3
   
2R3 R3
= r̂E0 cos θ 1 + 3 + θ̂E0 sin θ −1 .
r r3
The electric field must be perpendicular to a conducting surface, because it is at constant
potential, which is indeed satisfied by seeing that the θ̂ component vanishes at r = R.
3. Find the charge density per unit area on the surface of the sphere. Because there is no
electric field inside the conductor, one can consider a small area dA on the sphere at angle
θ. Gauss’s law relates the radial part of the electric field to the charge density σ,

Er dA = 4πσdA,
3E0 cos θ = 4πσ,
3E0
σ = cos θ.
4π

4.7.4 Boundary Value Problems Using Cylindrical Harmonics

This is very similar in spirit to the spherical case.

Example 4.5:
Consider a long conducting cylinder of radius R that is positioned so that its axis is perpendic-
⃗ = E0 x̂. Assuming the axis of the cylinder is defined
ular to an initially uniform electric field, E
by z = 0, find the electric potential and field at all points outside the conductor.
Solution:
The initial field is
Φ = −E0 ρ cos ϕ. (4.53)
Since we only wish to add corrections with the correct ϕ dependence, we assume the answer to
be of the form
Φ(ρ, ϕ) = −E0 ρ cos ϕ + AJ1 (kz ρ) cos(ϕ) + BN1 (kz ρ) cos ϕ, kz → 0. (4.54)
Because kz → 0, we need only consider the first order term when expanding in x. In that case,
N1 expands as 1/ρ and J1 expands as ρ. We can neglect the term that expands as ρ because it
was already included by choosing the long-range field. Thus
C
Φ(ρ, ϕ) = −E0 ρ cos ϕ + cos ϕ, kz → 0. (4.55)
ρ

46
PHY 841 4 ELECTROSTATICS

By inspection, to cancel the potential at ρ = R, C = R2 E0 ,

 
R2
Φ(ρ, ϕ) = −E0 ρ + E0 cos ϕ, (4.56)
ρ
E0 xR2
= −E0 x +
ρ2
E0 R2 E0 x2 R2
Ex = E0 − + ,
ρ2 ρ4
E0 xyR2
Ey = .
ρ4

4.8 Solving Boundary Value Problems Numerically

Typically, boundary value problems are multidimensional second-order differential equations
(Laplace’s equation). Thus, they can be difficult to solve given that you need to know two layers
of BC to integrate forward to the third layer, whereas boundary conditions are typically provided
only for one layer. One strategy for solving such problems is to look for iterative solutions.
Assume you have a guess of the solution, Φ0 (x, y, z), where x, y and z are discrete values.
For example you might divide the x range, from xmin to xmax into N + 1 values of size dx =
(xmax − xmin )/N . In the discretized space, Laplace’s equation becomes,
∂x2 Φ + ∂y2 Φ + ∂z2 Φ = 0, (4.57)
 
1 Φ(x + dx, y, z) − Φ(x, y, z) Φ(x, y, z) − Φ(x − dx, y, z)
−
dx dx dx
 
1 Φ(x, y + dy, z) − Φ(x, y, z) Φ(x, y, z) − Φ(x, y − dy, z)
+ −
dy dy dy
 
1 Φ(x, y, z + dz) − Φ(x, y, z) Φ(x, y, z) − Φ(x, y, z − dz)
+ −
dz dz dz
1
(Φ(x + dx, y, z) − 2Φ(x, y, z) + Φ(x − dx, y, z))
dx2
1
+ (Φ(x, y + dy, z) − 2Φ(x, y, z) + Φ(x, y − dy, z))
dy 2
1
(Φ(x, y, z + dz) − 2Φ(x, y, z) + Φ(x, y, z − dz)) = 0.
dz 2
One can solve for Φ(x, y, z) in terms of its neighbors,
1 (dx)2 (dy)2 (dz)2
Φ(x, y, z) = (4.58)
2 (dx)2 (dy)2 + (dy)2 (dz)2 + (dx)2 (dz)2

1
· [Φ(x + dx, y, z) + Φ(x − dx, y, z)]
(dx)2
1
+ [Φ(x, y + dy, z) + Φ(x, y − dy, z)]
(dy)2

1
+ [Φ(x, y, z + dz) + Φ(x, y, z − dz)] .
(dz)2

47
PHY 841 4 ELECTROSTATICS

If dx = dy = dz, this is simply an average of the six neighbors.

To perform the iterative solution, one finds new values Φ(x, y, z) for each x, y, z from Eq.
(4.58) using Φ0 (x, y, z) on the right-hand side of the equation. If the solution is correct, Φ will
not differ from Φ0 . However, if Φ0 is incorrect, it will differ. If the iterative procedure indeed
is convergent, the new value of Φ will be closer to the answer than Φ0 , and if one repeats the
procedure numerous times, replacing Φ0 with the new answer at each iteration, Φ will approach
the correct answer.

4.9 Homework Problems

1. Consider two charge densities,

3Q1
ρ1 (⃗
r) = Θ(R1 − |⃗
r |),
4πR13
3Q2
ρ2 (⃗
r) = Θ(R2 − |⃗
r − ax̂|).
4πR23

(a) Find the potential energy of the charge distribution when a → ∞.

(b) Find the change in the potential energy for moving from a = ∞ to a > R1 + R2 but
finite.

2. Consider two concentric conducting spherical shells of radius R and R + a. The charges
on the spheres are Q and −Q.

(a) Calculate the capacitance of the spheres, C = Q/V .

(b) For a potential V , find the electric field as a function of r for all r.
(c) For a potential V calculate the net energy stored in the electric field.
(d) Compare this to CV 2 /2.

3. Show that

Yℓ,ℓ (θ, ϕ) = cℓ eiℓϕ sinℓ θ,

s
 ℓ
(−1) (2ℓ + 1)(2ℓ)!
cℓ = ℓ
.
2 ℓ! 4π

is a solution to
1 1
∂θ (sin θ∂θ Y (θ, ϕ) + 2 ∂ϕ2 Y (θ, ϕ) = −ℓ(ℓ + 1)Y (θ, ϕ),
sin θ sin θ
and that it has unit normalization.

4. Show that the expansion in Eq. (4.30) for Legendre polynomials is indeed a solution for
 
(1 − x2 )∂x2 − 2x∂x + ℓ(ℓ + 1) Pℓ (x) = 0.

48
PHY 841 4 ELECTROSTATICS

5. Show that the solution to Laplace’s equation in cylindrical coordinates when kz = 0,

 
1 m2
∂ρ + ∂ρ R(ρ) − 2 R(ρ) = 0,
2
ρ ρ
becomes 
Aρ−m + Bρm , m ̸= 0,
Rm (ρ) = .
C ln(ρ), m=0

6. Consider the function Φ = (x2 + y 2 ) = r 2 sin2 θ.

(a) Calculate ∇2 Φ in Cartesian coordinates.
(b) Repeat in spherical coordinates
7. Consider a cavity that extends from x = −a to x = a and from y = 0 to y = ∞, i.e.
it is infinite in the y direction. Assume it is infinite in both directions in the z direction.
Along the y = 0 boundary, the surface is an insulator kept at a uniform potential V0 ,
while the boundaries along the x = ±a surfaces are grounded. Note this means that at
the corner of the boundaries the potential is discontinuous, thus one might need to imagine
an infinitesimal insulator at the intersection of the boundaries.
(a) Write down general solutions for the system that exponentially die for large y, and
that satisfy the B.C. that Φ(x = −a, y, z) = Φ(x = a, y, z) = 0. For the moment,
ignore the B.C. at the y = 0 surface.
(b) Find the sum of such solutions from (a) that satisfies the B.C. that V (x, y = 0, z) =
V0 .
8. Consider the solution to a point charge outside a conducting sphere of radius R performed
with images from the notes. Consider two point charges, one with charge Q at −aẑ, and
a second with charge −Q at aẑ. As a → ∞, the sphere sees a constant electric field,
E⃗ = 2Q/a2 ẑ. Find the electric potential from the two point charges and from the two
images in the limit that a and Q both go to infinity in such a way that 2Q/a2 = E0 .
Compare your solution to that you get from using spherical harmonics.
9. Consider two infinite conducting planes at z = a and z = −a. A point charge Q is
replaced at the origin. Find a set of image charges that satisfy the B.C. for −a < z < a.
10. Using the solution for a conducting cylinder in a constant field from the notes, show that
the electric field is perpendicular to the surface at ρ = R.
11. Consider a sphere of radius R centered at the origin, The surface of the potential is V (cos θ).
(a) In spherical coordinates, using the azimuthal symmetry, the potential at r = R can
be written as X
Φ(r = R, cos θ) = aℓ Pℓ (cos θ).
ℓ
Find Cℓ in the expression for aℓ of the form,
Z 1
aℓ = Cℓ dx Φ(r = R, x)Pℓ (x).
−1

49
PHY 841 4 ELECTROSTATICS

Here are some identities you might find useful:

Pℓ (x = 1) = 1,
X
(2ℓ + 1)Pℓ (x)Pℓ (x′ ) = 2δ(x − x′ ),
ℓ
Z 1
2
dx Pℓ (x)Pℓ′ (x) = δℓℓ′ ,
−1 2ℓ + 1
X
(2ℓ + 1)Pℓ (x)Pℓ (x′ ) = 2δ(x − x′ ),
ℓ
d
(2ℓ + 1)Pℓ (x) = [Pℓ+1 (x) − Pℓ−1 (x)] ,
dx
(ℓ + 1)Pℓ+1 (x) = (2ℓ + 1)xPℓ (x) − ℓPℓ−1 (x),
1 dℓ
Pℓ (x) = ℓ ℓ
(x2 − 1)ℓ (Rodriguez formula).
2 ℓ! dx
(b) Find aℓ for all ℓ for the potential

Φ(r = R, cos θ) = V0 cos(2θ).

Assuming the inside of the sphere is empty, write the potential Φ(⃗
r ) for all ⃗
r.

12. Like the previous problem, but with the potential


V0 , cos θ > 0
Φ(r = R, cos θ) =
−V0 cos θ < 0

(a) Using the identities from the previous problem, show that for this potential

(2ℓ + 1)
aℓ = V0 Pℓ−1 (x = 0) .
(ℓ + 1)

(b) Again, using the identities above, show that

ℓ
Pℓ+1 (x = 0) = − Pℓ−1 (x = 0),
(ℓ + 1)
(ℓ − 2)
Pℓ−1 (x = 0) = − Pℓ−3 (x = 0).
(ℓ − 1)

(c) Putting these together, show that

(2ℓ + 1)(ℓ − 2)
aℓ = −aℓ−2 ,
(ℓ + 1)(2ℓ − 3)
a1 = 3V0 /2, a(even) = 0.

(d) To test your answer, write a short program to calculate Φ(r = R) and see whether it
matches the expectation.

50
PHY 841 5 MULTIPOLE EXPANSIONS

5 Multipole Expansions

Here, we consider fields due to compact figurations of static charges when viewed from far
away. The fields are dominated by the lowest non-zero moment of the charge distributions, e.g.
monopole, dipole, quadrupole, etc.

5.1 Expanding Coulomb Potential in Spherical Harmonics

Consider a point charge Q at position aẑ. We wish to express the Coulomb potential as an
expansion in 1/r, where if a = 0 only the first term survives.

Q Q
= √ (5.1)
|⃗
r − aẑ| r 2 − 2ar cos θ + a2
Q 1
= p .
r 1 − 2(a/r) cos θ + (a/r)2

We now treat ϵ = 2(a/r) cos θ − (a/r)2 as a small number

Q Q 1
= √ (5.2)
|⃗
r − aẑ| r 1−ϵ
Q

= 1 + ϵ/2 + 3ϵ2 /8 + 15ϵ3 /48 · · · .
r
Organizing terms by powers of a/r,

Q Q 1
= 1 + (a/r) cos θ + (a/r)2 (3 cos2 θ − 1) (5.3)
|⃗
r − aẑ| r 2

1
+(a/r) (5 cos θ − 3 cos θ) + · · · .
3 3
2

One can see that this matches up with the expansion for the Legendre polynomials,
 ℓ
Q QX a
= Pℓ (cos θ). (5.4)
|⃗
r − aẑ| r ℓ
r

The above expression assumed the charge was along the z axis. One could prove this expression
term-by-term, or simply apply the generating function for Legendre polynomials (copied from
Eq. (4.34)),

1 X
√ = Pℓ (x)tℓ . (5.5)
1 − 2xt + t2 ℓ

From one perspective, this generating function defines the Legendre polynomials, and other
properties such as recurrence relations derive from the generating function.

51
PHY 841 5 MULTIPOLE EXPANSIONS

For our multipole expansion, we wish to find an expression for the field at some angle θ, ϕ when
the charge is at angle θ ′ , ϕ′ . The addition theorem for spherical harmonics,

4π X
ℓ
∗
Pℓ (cos γ) = Yℓm (θ ′ , ϕ′ )Yℓ,m (θ, ϕ), (5.6)
2ℓ + 1 m=−ℓ

plays a critical role in going forward with the multipole expansion. The angle γ is the angle
between the diretions θ, ϕ and θ ′ , ϕ′ . Here is the proof, along with explanations of each step,
(1) Pℓ (cos γ) = Pℓ (cos γ)Pℓ (1) (5.7)
4π
(2) = Y ∗ (0, 0)Yℓ0 (γ, 0)
(2ℓ + 1) ℓ0
4π
(3) = ⟨θ = 0, ϕ = 0|ℓ, 0⟩⟨ℓ, 0|γ, 0⟩
(2ℓ + 1)
4π X
(4) = ⟨θ = 0, ϕ = 0|ℓ, m⟩⟨ℓ, m|γ, 0⟩
(2ℓ + 1) m
4π X
(5) = ⟨θ = 0, ϕ = 0|R−1 R|ℓ, m⟩⟨ℓ, m|R−1 R|γ, 0⟩
(2ℓ + 1) m
4π X
(6) = ⟨θ = 0, ϕ = 0|R−1 |ℓ, m⟩⟨ℓ, m|R|γ, 0⟩
(2ℓ + 1) m
4π X
(7) = ⟨θ ′ , ϕ′ |ℓ, m⟩⟨ℓ, m|ϕ, θ⟩
(2ℓ + 1) m
4π X
∗
(8) = Yℓm (θ ′ , ϕ′ )Yℓm (θ, ϕ). (5.8)
(2ℓ + 1) m

1. Use the fact that Pℓ (cos θ = 1) = 1.

p
2. Use the fact that Pℓ (cos θ) = 4π/(2ℓ + 1)Yℓ0 (θ, ϕ).

3. Writing the Yℓm in bra-ket notation so that completeness is more apparent.

4. Yℓm (θ = 0) = 0 for all m ̸= 0, so extra parts of sum can be added without changing
result.

5. R are rotations that move γ, 0 to θ, ϕ and 0, 0 to θ ′ , ϕ′ .

6. Rotations don’t affect completeness |m⟩⟨m|.

7. Using definitions of rotation, note γ is angle between θ ′ , ϕ′ and θ, ϕ.

8. Leaving bra-ket notation.

The potential depends on the relative angle γ, so we can rewrite it as

 ℓ
Q 4πQ X 1 a ∗
= Yℓm (θ ′ , ϕ′ )Yℓm (θ, ϕ), (5.9)
|⃗
r −⃗
a| r ℓm
(2ℓ + 1) r

52
PHY 841 5 MULTIPOLE EXPANSIONS

where the potential is evaluated at r, θ, ϕ due to a charge at a, θ ′ , ϕ′ .

Equation (5.9) has profound implications due to the fact that the same ℓm combination appears
in both spherical harmonics. If one expresses the angular distribution of the potential at some
distance r outside the charge distribution, the moments ℓ, m will depend only the moments of
the local charge distribution with the same ℓ and m. To see this, we consider an array of charges,
represented by a charge density ρ(⃗ r ). Eq. (5.9) then gives
Z X  ℓ
4π 3 ′ ′
1 r′ ∗
Φ(⃗
r) = d r ρ(⃗
r) Yℓm (θ ′ , ϕ′ )Yℓm (θ, ϕ), (5.10)
r ℓm
(2ℓ + 1) r

r and θ ′ , ϕ′ describe the direction of r ′ . For a fixed mag-

where θ, ϕ describe the direction of ⃗
nitude r, the θ, ϕ dependence of Φ(⃗ r ) can be described by coefficients, Φℓm (r). Using the
completeness relation for spherical harmonics,
Z
dΩ Yℓ′ m′ (Ω)Yℓm (Ω) = δℓℓ′ δmm′ , (5.11)

one can see that the coefficients Φℓm can be consistently defined as
X 1
Φ(r, θ, ϕ) = Φℓm (r)Yℓm (θ, ϕ), (5.12)
(2ℓ + 1)
Z ℓm

∗
Φℓm (r) = (2ℓ + 1) dΩ Yℓ,m (θ, ϕ)Φ(r, θ, ϕ).

Thus, Eq. (5.9) becomes

Z Z X  r ′ ℓ
′

∗
4π 3 ′ ′
Φℓm (r) = dΩ Yℓ,m (θ, ϕ) d r ρ(⃗
r) Yℓ∗′ m′ (θ ′ , ϕ′ )Yℓ′ m′ (θ, ϕ) (5.13)
r ℓ ′ m′
r
4π
= qℓm ,
r ℓ+1
where qℓm are the multipole moments of the charge distribution,
Z
qℓm = d3 r ′ r ′ℓ ρ(⃗
r ′ )Yℓm
∗
(θ ′ , ϕ′ ). (5.14)

The potential can then be written as a sum over harmonic with coefficients given by qℓm ,
X 4π Yℓm (θ, ϕ)
Φ(⃗
r) = qℓm . (5.15)
ℓm
(2ℓ + 1) r ℓ+1

The moments, qℓm have dimension charge multiplied by length to the ℓth power, and the effect
of higher moments falls off as 1/r ℓ+1 .
The lowest multipole moments can be re-expressed using the forms for the spherical harmonics

53
PHY 841 5 MULTIPOLE EXPANSIONS

in Eq.s (4.31),
Z
1 1
q00 = √ r ) = √ q,
d3 r ρ(⃗ (5.16)
4π 4π
r Z r
3 3
q11 = − 3
d r ρ(⃗ r )(x − iy) = − (px − ipy ),
8π 8π
r Z r
3 3
q10 = − d3 r ρ(⃗r )z = − pz ,
4π 4π
r Z r
15 15
q22 = 3
d r ρ(⃗ r )(x − iy) =
2
(Q11 − 2iQ12 − Q22 ),
32π 288π
r Z r
15 15
q21 = − d3 r ρ(⃗r )(x − iy)z = − (Q13 − iQ23 ),
8π 72π
r Z r
5 5
q20 = 3
d r ρ(⃗ r )(3z − r ) =
2 2
Q33 .
16π 16π
Here q is the net charge,
Z
q = d3 r ρ(⃗
r ), (5.17)

pi are the dipole moments,

Z
pi = d3 r ρ(⃗
r )ri , (5.18)

and Qij are the quadrupole moments,

Z
Qij = r )(3ri rj − r 2 δij ).
d3 r ρ(⃗ (5.19)

Example 5.1:
Consider the charge density
r ) = Axe−(x
2 +y 2 +z 2 )/2R2
ρ(⃗ .
Find all the multipole moments, qℓm for ℓ ≤ 2.
Solution:
First find the net charge q, the dipole moments pi and the quadrupole tensor Qij . The net charge
q = 0 because the charge density is odd in x. Again, by symmetry, the only non-zero dipole
moment is px ,
Z
px = A d3 r x2 e−(x +y +z )/2R
2 2 2 2

= (2πR2 )3/2 AR2 .

The quadrupole tensor elements, Qij , are all zero because they all involve an overall even power
of coordinates. Some involve odd powers of x which would permit a non-zero result for the

54
PHY 841 5 MULTIPOLE EXPANSIONS

integral over x, but they then involve odd powers of y or z, which would then be zero. From
Eq.s (5.16), the only non-zero multipole moments are q11 and q1−1 ,
√
q11 = − 3πAR7/2 ,
∗
√
q1−1 = −q11 = 3πAR7/2 .

The sign in the last term was found by using Eq. (4.33).

5.2 Electric Field of a Dipole

Here we consider a dipole moment pz = pẑ, due to a small charge distribution. By small, we
assume that r is much larger than any separation of charges. Only the ℓ = 1, m = 0 multipole
is not zero. From Eq. (5.16)
r
3
q10 = p . (5.20)
4π
The potential for a dipole, from Eq. (5.15) is
r
3 4π
Φ(⃗r) = p Y10 (θ, ϕ) (5.21)
4π 3r 2
p cos θ z
= 2
= p 3.
r r
The electric field for various components is

p 3pz 2
Ez = − + ,
r3 r5
3pxz 3pyz
Ex = , Ey = ,
r5 r5
2p cos θ
Er = ,
r3
p sin θ Electric field lines as described with the ẑ pointing
Eθ = , to the right.
r3
1 ⃗·⃗
p r
⃗ = −
E p
⃗+3 ⃗
r.
3
r r5
In a constant electric field, Φ = −Ez, the interaction energy of the field with the dipole is
Z
U = d3 r ρ(⃗
r )(−Ez) (5.22)

= −pz |E|
⃗ = −⃗
p · E.
⃗ (5.23)

For two dipoles, with dipole moments p

⃗a and p
⃗b , the interaction energy between the two can be

55
PHY 841 5 MULTIPOLE EXPANSIONS

found by considering the interaction energy of dipole b due to the field generated by dipole a,

U = −⃗ pb · E⃗ a, (5.24)
1 pb · ⃗
(⃗ r)
⃗b = − p
E ⃗ b + ⃗
r, ⃗r≡⃗ ra − ⃗
rb ,
r3 r5
⃗a · p
p ⃗b pa · ⃗
(⃗ r )(⃗pb · ⃗
r)
U = − .
r3 r5

In an extremely hot system, the average energy between two dipoles vanishes because on aver-
age the directions of the dipoles are random and unaligned. However, if the system is at a finite
temperature T , a dipole moment is induced because the energy is lower if the dipole points
parallel to the field. Finding the value of the average induced moment depends on how many
orientation states exist for the dipole, which is a matter for quantum mechanics. Classically, one
would average over all directions of the dipole,
R1
d cos θ p cos θepE cos θ/T
−1
⟨pz ⟩ = R1 (5.25)
−1
d cos θ epE cos θ/T
p T
= − .
tanh(pE/T ) E

For high temperatures, one can expand in pE/T and find

Ep2
⟨pz ⟩ = .
3T
The average interaction energy with an electric field is then

E 2 p2
⟨U ⟩ = −⟨⃗
p · E⟩
⃗ =− , (5.26)
3T
which is attractive. If the dipole in question, #1, is due to another dipole, #2, its interaction
energy depends only ⟨E 2 ⟩ due to the dipole #2. If the two dipoles are separated by a distance r,
and if the second dipole is basically randomly oriented (again, consistent with the high T limit),
the average squared field is
Z
1 1  
⟨E ⟩ =
2
d cos θ Er2 (r, θ) + Eθ2 (r, θ) (5.27)
2 −1
Z
p2 1
= d cos θ (4 cos2 θ + sin2 θ)
r6 0
2p2
= .
r6
The average interaction energy is then

4p4
⟨U ⟩ = − . (5.28)
3r 6 T

56
PHY 841 5 MULTIPOLE EXPANSIONS

An extra factor of two was added to account for the non-random part of the orientation of dipole
#2 due to the random orientation with #1. This should be the dominant source of the potential
between two dipoles at large distances in the limit that the temperature is high enough that the
dipoles are for the most part randomly distributed. At short distance, atoms or molecules begin
to repel one another once the electronic wave functions attempt to overlap one another. These
forces typically have forms that die off faster than 1/r 6 , perhaps exponentially or perhaps as a
higher power than 6. Indeed, the 1/r 6 attractive force is known as the Van der Waals force and
has had great phenomenological success for a variety of phenomena. A common potential is the
6-12 potential which behaves as A/r 12 − B/r 6 .

5.3 Energy in arbitrary external field

As stated earlier, the energy of a charge in an external potential Φ, or that of a dipole in an
external electric field is

U(monopole) = QΦ0 , (5.29)

U(dipole) = −⃗
p·E
⃗ 0. (5.30)

The subscript “0” emphasizes that this is an external field, not the field from the charge or the
dipole itself. This can be extended to arbitrary multipoles in an external field,
Z
U = d3 r Φ(⃗r )ρ(⃗
r ), (5.31)
X xnx y ny z nz
Φ(⃗
r) = ∂xnx ∂yny ∂znz Φ ,
nx ny nz
nx !ny !nz ! r=0

X 1
U = Mnx ny nz ∂xnx ∂yny ∂znz Φ ,
nx ny nz
nx !ny !nz ! r=0
Z
Mn x n y n z = d3 r ρ(⃗
r )xnx y ny z nz .

The moments of the charge distribution,

M000 = q, (5.32)
M100 = px , M010 = py , M001 = pz ,
M200 = R11 , M110 = R12 , M101 = R13 ,
M110 = R21 , M020 = R22 , M011 = R23 ,
M101 = R , M011 = R32 , M002 = R33 ,
Z 31
Rij ≡ d3 r ρ(⃗
r )ri rj . (5.33)

57
PHY 841 5 MULTIPOLE EXPANSIONS

With these definitions,

1
U = qΦ0 − p
⃗·E
⃗0 − Rij ∂i E0j , (5.34)
2
1
= qΦ0 − p⃗·E ⃗ 0 − (Qij + R̄δij )∂i E0j ,
Z 6
R̄ ≡ d3 r r 2 ρ(⃗
r ),
Z
Qij ≡ d3 r (3ri rj − r 2 δij )ρ(⃗
r ).

where the fields and their derivatives are evaluated at r = 0. The term with R̄ does not con-
tribute to the energy because ∇ · E
⃗ 0 = 0 (the external field is that part of the field not affected
by local charge density), thus

1
U = qΦ0 − p
⃗·E
⃗0 − Qij ∂i E0j . (5.35)
6
Here, the quadrupole moment tensor Qij is defined as it was previously in this chapter, Eq.
(5.19).

5.4 Homework Problems

1. Consider four charges +q, +q, −q, −q at the Cartesian positions (a, 0, 0), (0, a, 0), (−a, 0, 0),
(0, −a, 0) respectively.

(a) Find the dipole moments pi and the quadrupole tensor Qij for all i and j.
(b) Find all qℓm for all ℓ ≤ 2.

2. Consider a charge distribution with q21 = q2−1 =some imaginary number iQa2 , see defini-
tions in Eq. (5.16). Draw a figure where you place a minimum number of discrete charges
that reproduces the given q21 and q2−1 , while having all other qij = 0 for ℓ ≤ 2.

(a) Provide the positions and find the individual charges, all of which are ±q, in terms of
Q. Only place charges on a lattice where the step size is a, i.e. at positions ia, ja, ka,
where i, j, k are integers.
(b) In terms of the magnitude of the individual charges, q, and the lattice spacings a, find
the potential as a function of r, θ and ϕ.

3. Consider a simple model of an atom being a particle of charge e that moves in a three-
dimensional harmonic oscillator with effective spring constant k. A constant electric field
E0 is added.

(a) What is the magnitude of the induced dipole moment p?

(b) What is the change in total energy of the charge due to the introduction of the field?
Give answer in terms of p and E0 .

58
PHY 841 5 MULTIPOLE EXPANSIONS

r ′ = aẑ.
4. Consider a point charge q at ⃗

(a) Find the moments, qℓm defined in Eq. (5.14), for all ℓ and m, defining the moments
around the origin
(b) Show that the potential calculated with qℓm using Eq. (5.15) matches Eq. (5.9) for the
case where the charge is along the z axis.
(c) Show that for the case where Φ(⃗
r ) is evaluated with ⃗
r lying along the z-axis that the
sum becomes q/(r − a).

5. Any function that can be written as a sum over Cartesian polynomials of order ≤ ℓ, i.e.,
X
F (x, y, z) = Aℓx ℓy ℓz xℓx y ℓy z ℓz ,
ℓx +ℓy +ℓz ≤ℓ

can be expressed as a sum of spherical harmonics with order ≤ ℓ,

X
F = Aℓ′ m′ (r)Yℓ′ m′ (θ, ϕ).
ℓ′ ,m′ ,ℓ′ ≤ℓ

Using this fact prove that the multipole moments of order ≤ ℓ, for the case when all mo-
ments qℓ′ m′ vanish for ℓ′ < ℓ, are unaffected by a translation of the origin, and that the
higher moments, > ℓ, are affected by this change of the coordinate system. This means
that the dominant multipole is unaffected by a translation of the coordinate system.
Hint: Using the definition of the moments, Eq. (5.14), replace ρ(⃗ r ) with ρ(⃗
r+⃗ a), then
express the new charge density as a Taylor expansion in powers of a.

59
PHY 841 6 MAGNETOSTATICS

6 Magnetostatics

Here, we consider magnetic fields from steady currents, and set the stage for fields from dynamic
sources.

6.1 The Biot-Savart Law

In Chapter 2 we derived Maxwell’s equations from Lagrangians written in terms of the vector
potential, and for electrostatics the Maxwell equation that was ∇2 Φ = −∇ · E ⃗ = −4πρ.
This then led to Coulomb’s law. Here, we show one can similarly relate ∇2 A ⃗ = 4π J⃗, so that
each component of A ⃗ is related to the corresponding component in J⃗ in the same manner that
Φ = A0 is related to ρ = J0 .
⃗ = 0),
To this we begin with the Maxwell relation in a static system (∂t E

∇×∇×A ⃗ = 4π J,
⃗ (6.1)
∇2 A
⃗ − ∇(∇ · A)
⃗ = −4π J. ⃗

a × (⃗b × ⃗
This last step is a simple vector identity, ⃗ a·⃗
c) = ⃗b(⃗ c) − (⃗a · ⃗b)⃗ a = ⃗b = ∇ and
c, with ⃗
⃗ ⃗
c = A.
Our difficulty going forward is the term ∇(∇ · A). ⃗ However, one can eliminate that term by
choosing a convenient gauge. As show earlier, one can add a term to the vector potential, A ⃗′ =
A⃗ + ∇Λ, without changing any physics, because E ⃗ and B⃗ are unchanged. This works for any
function Λ(x, y, z). We are in need of a new vector potential A ⃗ ′ where ∇ · A ⃗ ′ = 0,
 
∇· A⃗ + ∇Λ = 0, (6.2)
∇2 Λ = −∇ · A.
⃗

If there were some charge density, ρ = ∇ · A, ⃗ then finding Λ would be equivalent to finding
the potential corresponding to the static charge density. Of course, there is such a potential, thus
there is always a gauge such that ∇ · A⃗ = 0. Assuming one is in that gauge,

∇2 A
⃗ = −4π J.
⃗ (6.3)

Of course, due to gauge invariance there are many vector potentials one could choose, but they
all lead to the same electromagnetic fields. Thus, we are free to choose this gauge.
Equation (6.3) is actually three separate equations. Each one is effectively a Poisson’s equation
where the components Ji play the role of charge densities. This is equivalent to Poisson’s equa-
tion for the electric potential, ∇2 Φ = −4πρ, so one can use the same results as before. Mainly,
Z
1
⃗ r) =
A(⃗ d3 r ′ J⃗(⃗
r′ ) . (6.4)
|⃗
r−⃗ r′ |

60
PHY 841 6 MAGNETOSTATICS

The magnetic field is then

⃗ r) = ∇ × A
B(⃗ ⃗ (6.5)
Z
r ′ ) × (⃗
J⃗(⃗ r−⃗ r′ )
= d3 r ′ ,
|⃗r−⃗ r ′ |3

which is the Biot-Savart Law.

Note the expression assumes natural units, if the currents are in MKSA units, where Amperes
(Coulombs/s) are the unit of current, the expression picks up an additional factor,
Z
µ0 r ′ ) × (⃗
J⃗(⃗ r′ )
r−⃗
⃗ (MKSA) (⃗
B r) = d3 r ′ , (6.6)
4π |⃗r−⃗
r ′ |3
µ0 = 4π10−7 Tm/A.

Example 6.1:
A 5 Ampere current travels through a thin wire in a circular loop of radius R = 10 cm. Find the
strength of the magnetic field in the center of the loop.
Solution:
The current density element Jd ⃗ 3 r ′ becomes Idℓ
⃗ for a thin wire. Each element Idℓ
⃗ provides a
differential component to the magnetic field that is perpendicular to the plane of the loop (we’ll
call it ẑ). Integrating around the loop,
Z
µ0 I ẑ
⃗
B(MKSA) (⃗ r = 0) = dℓ 2 , (6.7)
4π R
µ0 I
|B
⃗ (MKSA) (⃗ r = 0)| = = π · 10−5 T. (6.8)
2R

6.2 Magnetic Moments

If one is outside a localized current distribution, the vector potential falls off according to the
same mathematics as used for multipole moments of the electric potential from the previous
chapter. Because the net current density is zero for a confined static current,
Z
d3 r Ji (⃗
r ) = 0, (6.9)

the “monopole” charge for each component i vanishes. Whereas p ⃗ referred to the dipole moment
of a charge distribution, here we will worry about equivalent of a dipole moment vector for each
component Ji . Thus, we consider the moment
Z
πij = d3 r ′ Ji (⃗
r ′ )rj′ . (6.10)

61
PHY 841 6 MAGNETOSTATICS

Far away, the dipole contribution to the ith component of the vector potential is
πij rj
Ai (⃗r) = , (6.11)
r3
where we have simply translated from the result for the electric potential for a dipole, Φ =
⃗·⃗
p r /r 3 . Next, one can find the magnetic field,
Bi = ϵijk ∂j Ak (6.12)
 
πkj πkℓ rℓ rj
= ϵijk −3 .
r3 r5
Using the identity,
ϵijk ϵkℓn = δiℓ δjn − δin δjℓ , (6.13)
One can see that
1 1
ϵijk ϵkℓn πℓn = (πij − πji ) (6.14)
2 2
= πij ,
where the last step used the fact that πij is anti-symmetric (which follows from current conser-
vation, see end-of-chapter problem). Defining the magnetic moment m, ⃗
1 1
mi ≡ − ϵijk πjk = ϵijk πkj , (6.15)
2
Z 2
1
m
⃗ = r × J.
d3 r ⃗ ⃗
2
one can express πij in terms of m,
⃗
πij = −ϵijk mk . (6.16)
⃗ in Eq. (6.12),
Inserting these into the expression for B
2mi 3
Bi = + ϵijk ϵkℓn mn rj rℓ (6.17)
r3 r5
2mi 3
= + 5 (δiℓ δjn − δin δℓj ) mn rj rℓ
r3 r
mi 3ri
= − 3 + 5 (m ⃗ ·⃗ r)
r r
m
⃗ 3⃗r
⃗ = − +
B ⃗ ·⃗
(m r ).
r3 r5
For a current loop with a thin wire, the magnetic moments have a simple geometric expression.
For a thin wire one can
d3 r = dℓd2 r⊥ , (6.18)
Z
dℓ d2 r⊥ J⃗(⃗ ⃗
r ) = Idℓ,
Z
1
m
⃗ = r × I⃗
dℓ ⃗
2
Z
I
= r × d⃗
⃗ ℓ.
2

62
PHY 841 6 MAGNETOSTATICS

If the current lies in a plane, |m|

⃗ = Ia, where a is the area of the loop, and the direction of m
⃗ is
perpendicular to the plane. Here, dℓ is along the direction of the wire and ⃗ r⊥ is perpendicular
to the wire.

6.3 Magnetic Forces and Torques

The magnetic force q⃗ v×B ⃗ leads to torques on small current loops. In a magnetic field B,
⃗ the
net force acting on some set of currents J⃗ is
Z
⃗
F = d3 r J⃗ × B,
⃗ (6.19)

because J⃗d3 r is the differential contribution to q⃗ ⃗ factors out

v . For a constant magnetic field B
from the integral and one can see that the net force acting on a static current distribution is zero
R
because d3 r J⃗ = 0 (See end-of-chapter problem). Nonetheless, there is a net torque,
Z
⃗τ = d3 r ′ ⃗ r ′ ) × B),
r × (J⃗(⃗ ⃗ (6.20)
Z Z
τi = Bk d3 r ′ rk′ Ji (⃗ r ′ ) − Bi d3 r ′ rk′ Jk (⃗
r ′ ).

Now one can use that fact that

Z
πik ≡ r′ )
d3 r ′ ri′ Bk (⃗ (6.21)

is anti-symmetric to throw away the second term, and rewrite the first term,
Z
1  
τi = Bk d3 r ′ rk′ Ji (⃗
r ′ ) − ri′ Jk (⃗
r′ ) (6.22)
2 
Z
1 3 ′ ′ ′
⃗
τ = d r ⃗r × J (⃗
⃗ r) ×B ⃗
2
= m ⃗ × B.⃗

The work done by the torque in rotating an object is τ dθ, so the potential energy of the dipole is

U = −m
⃗ · B,
⃗ (6.23)

and magnetic dipoles prefer to be aligned parallel to a magnetic field.

For two physically separated dipoles, m
⃗ a and m ⃗ b , one can then calculate the energy associated
with their spin configuration using Eq. (6.17) which gives the magnetic field due to dipole a that
would be experienced by dipole b,
 
m⃗a 3⃗
r
U = −m ⃗ b · − 3 + 5 (m ⃗ a·⃗r) (6.24)
r r
⃗a·m
m ⃗b 3⃗
r
= − ⃗ a·⃗
(m ⃗ b·⃗
r )(m r ).
r3 r5

63
PHY 841 6 MAGNETOSTATICS

6.4 Overlapping Distributions and Hyperfine Splitting

Thus far, we have considered only the interaction between to distributions separated by a finite
amount. Sometimes, e.g. electrons interacting with a nucleus to produce the hyper-fine inter-
action, the two distributions lay atop one another. Here, we consider the electric and magnetic
field integrated over volumes that include all charges. This does not provide the entire distribu-
tion of the field, but for energies such as −⃗
p·E⃗ or −m ⃗ · B,
⃗ it will give the interaction energies
for a smooth distribution of electric or magnetic dipole densities within a volume. This mainly
comes into play in the interior of atoms. For examples, the smooth distribution of electron den-
sity, and therefore the magnetic-dipole density, are well known from solving the Schrödinger
equation. This interacts with the magnetic field due to the nuclear magnetic dipole. Because the
electron density is smooth, one needs only integrate the magnetic dipole density of the electrons
multiplied by the magnetic field due to the nuclear magnetic dipole.
To calculate the magnetic interaction of a nuclear magnetic moment, µ ⃗ N , with the intrinsic spin
of electrons electrons in an atom, ρe , one would do the integral
Z
U = −µe d3 r [ρe,↑ (⃗ r ) − ρe,↓ (⃗ ⃗ z (⃗
r )] B r ), (6.25)

where Bz would be the field due to the nuclear magnetic moment. The magnetic moment of an
electron is
eℏ
µe = ge , (6.26)
2me
where ge ≈ 2.0 is the g−factor for electrons, and the ↑ and ↓ symbols denote the densities of
spin-up or spin-down electrons respectively. The difficulty with performing the integral in Eq.
(6.25) is that we know how to calculate the magnetic field due to a magnetic moment at distances
outside the nucleus, but within the current it is more difficulty. This is a small volume, but the
fields are strong. Also due to the smallness, the density of electrons can be taken as constant
within it, which means we need only the integrated value of Bz . The energy within this small
spherical volume of radius we denote as
Z
UΩ ≈ −µe (ρe,↑ − ρe,↓ )(r = 0) d3 r Bz (⃗
r ). (6.27)
r<R

This approximation is excellent because the radius of a nucleus is ∼ 10−4 that of the electron
cloud. Our goal is then to calculate
Z
⃗
BV = ⃗ r ),
d3 r B(⃗ (6.28)
r<R

due to the nuclear magnetic moment

eℏ
µ
⃗ N = gN ẑ. (6.29)
2MN

We will consider both the cases of the electric and magnetic fields. First, we find an expression
⃗ V and B
for E ⃗ V for the case where all charges and currents are within the volume, then move to

64
PHY 841 6 MAGNETOSTATICS

the case where all charges and currents are within the volume. The electric potential in a volume
with no charges can be expanded in spherical harmonics,
X
Φ(⃗ r ) = Φ(⃗ r = 0) − E(r
⃗ = 0) · ⃗ r+ Aℓm Yℓm (θ, ϕ)r ℓ . (6.30)
ℓ>2,m

Calculating the integrated electric field, only the ℓ = 1 term contributes and one finds,
Z
EV = −
⃗ d3 r ∇Φ (6.31)
r<R
3
4πR
= ⃗ = 0),
E(r
3
because all other terms will be constructed from Yℓ,m s with ℓ > 0 so they vanish. For example,
if one applies the gradient operator, which transforms as an ℓ = 1 object, to r ℓ Yℓ,m , the resulting
pieces will consist of terms that behave as Yℓ−1,m′ and Yℓ+1,m , which will all integrate to zero
for ℓ ≥ 2. Thus, if there are no charges within a spherical volume, the integrated electric field is
the the electric field at the center multiplied by the volume. This means that the average electric
field within a spherical volume with no charges is the electric field at the center.
Similarly, one can do the same for the magnetic field. Again, one can make an expansion for the
vector potential components in the Coulomb gauge, A, ⃗ in spherical harmonics because ∇2 A
⃗ =
0 in a region with no currents,
1 X
A(⃗ ⃗ r = 0) − ⃗
⃗ r ) = A(⃗ r × B(r
⃗ = 0) + Aℓm Yℓm (θ, ϕ)r ℓ . (6.32)
2 ℓ>2,m

When calculating
Z
⃗V
B = d3 r ∇ × A
⃗ (6.33)
r<R
4πR3
= ⃗ = 0).
B(r
3
Thus the average magnetic field in a volume with no currents is the magnetic field at the center.
Next, we calcuate the average field in a spherical volume that contains all the charges or currents.
First, we consider the electric field
Z
⃗
EV = ⃗ r)
d3 r E(⃗ (6.34)
Z
r<R

= − d3 r ∇Φ
Z
= −R 2
dΩ r̂Φ(R, Ω).

All the charges are inside R, so one can expand Φ in multipoles. Using the fact that the unit
vector in the outward direction is

r̂ = sin θ cos ϕx̂ + sin θ sin ϕŷ + cos θ ẑ, (6.35)

65
PHY 841 6 MAGNETOSTATICS

one can see that r̂ only has ℓ = 1 terms of Φ contribute when the angular dependencies are
expressed in Yℓm s. Thus, only the ℓ = 1 parts of Φ contribute, which are the dipole pieces.
Z
⃗·⃗
p r
EV = −R
⃗ 2
dΩ 3 r̂, (6.36)
Z r
pj r j
EV,i = −R2 dΩ 3 r̂i ,
r
4π
E⃗V = − p ⃗.
3
To convince yourself the last step is correct, define the z axis to be along p
⃗, then only the ẑ part
of r̂ contributes.
Next, one can calculate the corresponding quantity for magnetic fields. Similar as before,
Z
⃗
BV = ⃗ r)
d3 r B(⃗ (6.37)
r<R
Z
= ϵijk d3 r ∂j Ak
Z
2
= R ϵijk dΩ r̂j Ak (R, Ω).

One can expand Ak in the same way Φ was expanded. Again, only the ℓ = 1 terms contribute.
That term is given in Eq. (6.11),
Z
π r
B⃗ V = R ϵijk dΩ r̂j kℓ ℓ
2
(6.38)
3
Zr
rj rℓ
= −R2 ϵijk ϵkℓn mn dΩ 4
r
4π
= − ϵijk ϵkjn mn
3
8π
= mi ,
3
8π
B⃗V = m.
⃗
3
Summarizing these results for the electric and magnetic fields,
Z 
3 ⃗ −4π⃗ p/3, all charges within sphere
d r E(⃗ r) = ⃗ (6.39)
r<R E(r = 0)V, all charges outside sphere
4π 3
⃗ r) ≈ E
E(⃗ r) −
⃗ multipole (⃗ p⃗δ (r),
Z  3
3 ⃗ 8π m/3,
⃗ all currents within sphere
d r B(⃗ r) = ⃗
r<R B(r = 0)V, all currents outside sphere,
Here, V is the volume of the sphere. Putting these results together, we can express the magnetic
field due to a magnetic moment as
8π
⃗ r) ≈ B
B(⃗ ⃗ multipole (⃗
r) + ⃗ 3 (r).
mδ (6.40)
3

66
PHY 841 6 MAGNETOSTATICS

electron −2.00231930436182 ± 0.00000000000052

muon −2.0023318418 ± 0.0000000013
proton 5.585694702 ± 0.000000017
neutron −3.82608545 ± 0.00000090

Table 1: g factors for various particles. Even the neutral neutron has a magnetic moment because it is
constructed of charged up and down quarks.

These expressions assume the electric and magnetic dipoles are confined to a spatially small re-
gion, with “small” being relative to the distributions with which they interact. The construction
clearly provides the correct field outside the volume of the dipole, and in the neighborhood of
the dipole it integrates to the correct value. This is sufficient for the purposes of calculating the
hyperfine interaction, between the magnetic field of the dipole and the smooth electron cloud.
The hyperfine interaction involves the coupling of the magnetic moments of the electron, mu
⃗ e
with the magnetic fields arising from the nuclear magnetic moment, µ
⃗N.

µN · µ
(⃗ ⃗ e) µN · ⃗
3(⃗ µe · ⃗
r )(⃗ r) 8π µN · L)
(⃗ ⃗
U = − − µN · µ
(⃗ r) − e
⃗ e )δ 3 (⃗ . (6.41)
r3 r5 3 mr 3
The last term arises because if the electron is moving with finite angular momentum, it generates
a magnetic field which then couples with the nuclear magnetic moment. This term only comes
into play if the net orbital angular momentum of the electrons is non-zero.
Magnetic moments have dimensions of charge multiplied by velocity. For a classical particle of
charge e moving in a circle of radius R with velocity v, the current and magnetic moment would
be
e
I = , (6.42)
2πR/v
IR ev
µ(class) = =
2 2
eL⃗
µ
⃗ = .
2m
⃗ this no longer holds, and we use a factor g to describe the ratio of µ to the
For intrinsic spin, S,
classical expectation,

⃗
eS ⃗  eℏ 
S
µ
⃗ = g =g . (6.43)
2m ℏ 2m

The last factor, eℏ/2m (or eℏ/2mc) depending on the units of magnetic field) is known as a
Bohr magneton. The g-factor is nearly 2.0 for point-particle fermions (e.g. electrons), differing
from 2 due to higher-order perturbative corrections of order e2 . Protons are not point particles
and the g-factor depends on details of the spin and angular momentum configuration of the
constituent quarks.

67
PHY 841 6 MAGNETOSTATICS

6.5 Homework Problems

1. Prove the following using current conservation (∇ · J⃗ = 0 for static systems):
(a) Beginning with Z
d3 r (∇ · J⃗(⃗
r ))ri = 0,

show that Z
d3 r Ji (⃗
r ) = 0.

(b) Beginning with Z

d3 r (∇ · J⃗(⃗
r ))ri rj = 0,

show that πij is antisymmetric, where

Z
πij ≡ d3 r Ji (⃗
r )rj .

2. Consider two current loops, each moving in the z = 0 plane. Each loop has current I and
radius a. The first loop is centered at x = −a, y = 0 and is circulating clockwise, and the
second loop is centered at x =, y = 0 and is circulating counter-clockwise.
(a) Calculate the analog of the quadrupole moment for calculating the ith component of
the vector potential,
Z
Qikℓ = r )(3rk rℓ − r 2 δkℓ ).
d3 r Ji (⃗

r ) far away, to order 1/r 3 . Use relations from the pre-

(b) Find the vector potential Ai (⃗
vious chapter, where you merely add an additional index to account for the fact you
are dealing with the vector potential. You can leave your answer in terms of Qikℓ , but
clarify which components of Ai are zero, and how each component depends explicitly
on the angles θ and ϕ.
3. A particle of mass m and charge e moves in a circular orbit of radius R with angular
momentum L. What is the magnitude of the magnetic field at the origin? Note this was
used in writing the expression in Eq. (6.41) for the hyper-fine coupling.
4. Using Eq. (6.41), consider the hyper-fine energies for the 1s levels of a hydrogen atom.
There are two levels because the electron and proton spin can couple to either zero or
unity.
(a) Which terms in Eq. (6.41) contribute to the energy?
(b) Find the difference between the energy levels. Express your answer numerically in
eV, using the electron mass and charge, and using the fact that the hydrogen atom
electron density for the 1s state behaves as
r ) ∼ e−2r/a0 ,
ρ(⃗
a0 = 0.529 Å.

68
PHY 841 6 MAGNETOSTATICS

(c) What would the wavelength of light be for a transition between the states.

5. Consider a parallel-plate capacitor where the area of the plates is A and the small separa-
tion is a. The charge on the plates are ±Q.

(a) What is the dipole moment of the capacitor if the plates are aligned in the z direction?
(b) What is the electric field in the capacitor?
(c) Determine whether Eq. (6.36) is satisfied in this case where the spherical volume en-
gulfs the entire capacitor and one assumes the electric field is assigned to the interior
of the capacitor.

69
PHY 841 7 ELECTROMAGNETIC WAVES

7 Electromagnetic Waves

In the last several sections we have considered static systems, where we could neglect all the
∂t · · · terms in Maxwell’s equations. In this chapter we consider the propagation of waves, and
wave equations clearly require the ∂t .... terms.

7.1 The Wave Equation

Leaving off the current and charge density terms in Maxwell’s equations,

∇·E ⃗ = 0, (7.1)
∇·B ⃗ = 0,
(∇ × B)
⃗ − ∂t E
⃗ = 0,
⃗ +∇×E
∂t B ⃗ = 0.

One can add the curl of the third equation to the time derivative of the fourth to obtain

∇ × (∇ × B)
⃗ + ∂ 2B
t
⃗ = 0, (7.2)
−∇2 B
⃗ + ∇(∇ · B)
⃗ + ∂ 2B
t
⃗ = 0,
∇2 B⃗ = ∂ 2 B,
⃗
t
⃗ = 0.
2
∂ B

This represents three separate wave equations, each with a wave velocity of unity. We wrote
Maxwell’s equations in units where the speed of light is unity, otherwise the time derivatives
would change, ∂t → (1/c)∂t . Similarly, one can subtract the curl of the fourth equation above
from the time derivative of the second and obtain

∇2 E
⃗ = ∂ 2 E,
t
⃗ (7.3)
⃗ = 0.
∂ 2E

Equations (7.2) and (7.3) appear to represent six wave equations. For propagation along in the
direction of a wavenumber ⃗ k axis, the solutions have the forms

r , t) = ai eik·⃗r−iωt ,
⃗
Ei (⃗ (7.4)
i⃗ r −iωt
k·⃗
Bi (⃗
r , t) = bi e ,
ω = |⃗k|.

To discern the direction of propagation, one can ask for what δ⃗

r would one considers a wave
packet with combinations of different wave numbers centered around k. The peak of the packet
is defined by that point in coordinate space where the various wave numbers contribute with a

70
PHY 841 7 ELECTROMAGNETIC WAVES

steady phase, i.e.

 
k·⃗
∂ki i⃗ r − iω(⃗
k)t = 0, (7.5)
∂ω(⃗
k)
ri −
⃗ t = 0,
∂ki
⃗
k
⃗
ri = t.
|⃗
k|

The last step made use of the fact that ω = |⃗ k|. If one were in a medium, the group velocity,
dω/d⃗ k, could differ from the speed of light, a subject for the next course.
The six solutions in Eq. (7.4) are not independent. In order to satisfy the connection between E ⃗
and B⃗ in Maxwell’s equations in Eq.s (7.1), once given E, ⃗ one can find B, ⃗ and vice versa. Also,
in order to satisfy ∇ · E
⃗ = 0 and ∇ · B ⃗ = 0, the wave equations must satisfy

a·⃗
⃗ k = 0, (7.6)
⃗b · ⃗
k = 0.

Thus, the six solutions for a given wave number become two. The amplitudes ⃗ a and ⃗b must be
normal to the wave number ⃗ k and additionally, the third and fourth Maxwell’s equations require
that

k ×⃗
⃗ a = ω⃗b, (7.7)
k × ⃗b = −ω⃗
⃗ a.

There are two independent components of ⃗ a given ⃗

k. This defines the polarization. Once given
⃗ ⃗ k. A solution with ax ̸= 0 and
a and ⃗
a, b is also determined. The vector b will be normal to both ⃗
⃗
ay = 0 is considered linearly polarized in the x direction, whereas the solutions with ax = 0
is linearly polarized in the y direction. One can consider ay = iax and find a solution where
the real part of Ex is a cosine wave and the real part of Ey is a sine wave. These are circularly
polarized solutions.

7.2 Stress-Energy Tensor of Electromagnetic Waves

The zero-zero component of the stress-energy tensor is the energy density. Using the solutions
for the waves in Eq. (7.4) and the stress-energy tensor for the electromagnetic field from Eq.

71
PHY 841 7 ELECTROMAGNETIC WAVES

(3.33),
1  
T00 = |E|
⃗ 2 + |B|
⃗ 2 (7.8)
8π
a2i + b2i |⃗
a|2
= = k·⃗
cos2 (⃗ r − ωt),
8π 4π
E j Bk
T0i = ϵijk
4π
|⃗
a|2
= k̂i k·⃗
cos2 (⃗ r − ωt),
4π
1
T ij = −T i j = (δij (E 2 + B 2 ) − 2Ei Ej − 2Bi Bj )
8π
1  2
= |⃗
a| δij − ai aj − bi bj cos2 (⃗
k·⃗
r − ωt).
4π

r − ωt) arise from using only the real part of ei(k·⃗r−ωt) and assuming ⃗
⃗
k·⃗
The factors cos2 (⃗ a is
real (equivalent to saying the polarization is linear). Otherwise, phases, e.g. cos(k · ⃗
⃗ r − ωt + ϕ)
would enter.
For a wave moving along the z axis, with polarization components ax and ay ,

a2x + a2y
T 00
= cos2 (kz − ωt), (7.9)
4π
a2x + a2y
T 0z = cos2 (kz − ωt) = T 00 ,
4π
ax + a2y
2
T zz = cos2 (kz − ωt) = T 00 ,
4π
T xx = T yy = T i̸=j = 0.

For the last expression, we used the fact that bx = −ay and by = ax , which comes from
Eq. (7.7). These results are expected because T0z is both the momentum density and the flux
of energy. The energy flux is energy density multiplied by velocity, so one expects it to equal
T00 because the velocity in unity. Further, Tzz is the flux of momentum, momentum density
multiplied by velocity, so it equal to T0z = T00 .

7.3 The Doppler Effect

An observer sees light from a source with wave number ⃗ k. Its frequency is |⃗
k|. If the source
is moving with velocity ⃗
v , one can find the frequency of the emission according to an observer
moving with the source, ωs , by treating the frequency and wave number as as four vector,

k = (ω, ⃗
k). (7.10)

The frequency in the source frame is then found by the Lorentz transformation,

ωs = γ(ω − ⃗
v ·⃗
k) = ωγ(1 − v cos θ), (7.11)

72
PHY 841 7 ELECTROMAGNETIC WAVES

v and ⃗
where θ is the angle between ⃗ k. If the source is moving directly toward the observer,
s
1−v
ωs = ω . (7.12)
1+v

If the source is moving away from the observer, the sign of v shifts, and the observed frequency
is less than the source frequency (red shift). If the source were to move perpendicular to the
observer, one would see the frequency shift by a factor γ, i.e. the red shift would be smaller but
would not vanish.

7.4 Wave Guides and Cavities

Here we consider oscillating solutions at fixed frequency. We consider the simplest case, that of
a vacuum surrounded by a perfect conductor. For non-perfect conductors the boundary condi-
tions are more complicated due to the penetration of the fields into the media. Such penetration
also leads to damping of waves. Because this course explicitly ignores electromagnetism in me-
dia, our coverage of wave guides will be brief.
The boundary conditions for an electric field at a conductor are simple, E ⃗ × n̂ = 0 at the
surface. Any non-zero transverse field would cause an infinite current along the surface. The
conductor has no problem with the perpendicular component being non-zero. From Gauss’s
law, one can see that the surface charge density, σ = E ⃗ · n̂/4π, can maintain the field being
zero inside the conductor. For waves the fields are changing in time, and Maxwell’s equation
∇×B ⃗ = ∂t E ⃗ demands that the presence of an oscillating electric field be accompanied by
an oscillating magnetic field. Also, any oscillating magnetic field would be accompanied by
an electric field due to the Maxwell’s equation, ∇ × E ⃗ = −∂t B. ⃗ This leads to the demand the
B⃗ = 0 inside the conductor – except for static fields, which are not relevant for waves. Whereas a
charge density can screen the perpendicular component of E, ⃗ there is no such thing as magnetic
charge density, ∇ · B ⃗ = 0, so the boundary condition for magnetic field is B ⃗ · n̂ = 0. One
can then consider the transverse magnetic field at the surface. One can draw a loop drawn just
⃗ then cuts into the surface, return in the opposite direction
outside the surface for a distance L,
−L ⃗ inside the surface, then close the loop. If there is magnetic field along L ⃗ outside the loop,
one can still have no magnetic field inside the conductor if there is a surface current ⃗js . Thus,
the boundary conditions for a perfect conductor are:

n̂ × E ⃗ = 0, (7.13)
n̂ · B
⃗ = 0,
n̂ · E
⃗ = 4πσ,
n̂ × B ⃗ = 4π⃗js .

First we consider a rectangular wave guide, constrained in the x and y directions

0 < x < Lx , (7.14)

0 < y < Ly .

73
PHY 841 7 ELECTROMAGNETIC WAVES

The equations of motion for a wave, rewriting Eq.s (7.2) and (7.3),

∇2 E
⃗ = ∂ 2 E,
t
⃗ (7.15)
∇2 B
⃗ = ∂ 2 B,
t
⃗

and viewing the boundary conditions suggest solutions of the form for the electric field,

r , t) = E0x e−iωt+ikz z cos(qx x) sin(qy y),

Ex (⃗ (7.16)
r , t) = E0y e−iωt+ikz z cos(qy y) sin(qx x),
Ey (⃗
r , t) = E0z e−iωt+ikz z sin(qy y) sin(qx x).
Ez (⃗

with

ω 2 = qx2 + qy2 + kz2 ,

nx π ny π
qx = , qy = ,
Lx Ly

nx and ny being any four integers. In principal, one could have imagined using different num-
bers ni for Ex , Ey and Ez if they somehow added up to the same frequency due to a fortuitous
choice of Lx and Ly – we return to that further below. One can also write down solutions for the
magnetic field wave equations,

r , t) = B0x e−iωt+ikz z sin(qx x) cos(qy y),

Bx (⃗ (7.17)
r , t) = B0y e−iωt+ikz z sin(qy y) cos(qx x),
By (⃗
r , t) = B0z e−iωt+ikz z cos(qy y) cos(qx x).
Bz (⃗

By inspection one can see that these six components listed above have the same frequency and
satisfy the boundary conditions independently, i.e. any choice of the six amplitudes E0i and B0i
would work. However, one must also satisfy the Maxwell’s equations that link E ⃗ and B,
⃗

⃗ = −∇ × E,
∂t B ⃗ (7.18)
⃗ = ∇ × B.
∂t E ⃗

This leads to the conditions

−iωB0x = ikz E0y + qy E0z , (7.19)

−iωB0y = −ikz E0x − qx E0z ,
−iωB0z = qx E0y − qy E0x ,
iωE0x = ikz B0y + qy B0z ,
iωE0y = −ikz B0x − qx B0z ,
iωE0z = qx B0y − qy B0x .

It might seem worrisome that we have six unknown amplitudes and six conditions, as we cer-
tainly expect more than one solution, not to mention an arbitary multiplicative constant. How-
ever, these conditions are not independent.

74
PHY 841 7 ELECTROMAGNETIC WAVES

In fact, there are two independent sets of solutions, each with an arbitrary overall amplitude.
The first set of solutions is referred to as transverse magnetic (TM) and the second as transverse
electric (TE). For TM solutions, we set B0z = 0 and for TE solutions we set E0z = 0. For the TM
case, we solve for the remaining amplitudes in terms of E0z . The algebra gives

qx k z
E0x = −i E0z , (7.20)
ω 2 − kz2
qy k z
E0y = −i E0z ,
ω 2 − kz2
kz qy kz qy
B0x = −i E 0z − i E0z ,
ω ω 2 − kz2 ω
kz q x kz qx
B0y = i E 0z + i E0z .
ω ω 2 − kz2 ω

Similarly, one can solve for TE amplitudes in terms of B0z ,

kz qy k z qy
E0x = −i B0z − i B0z , (7.21)
ω ω2 − kz2 ω
kz qx kz qx
E0y = i B0z + i B0z ,
ω ω 2 − kz 2 ω
qx kz
B0x = −i 2 B0z ,
ω − kz2
qy kz
B0y = −i 2 B0z .
ω − kz2

The group velocity is not the speed of light. From the dispersion relation (ω vs kz ) one can find
the group velocity,

ω 2 = kz2 + qx2 + qy2 , (7.22)

dω kz kz
vg = = = q .
dkz ω 2 2 2
k z + qx + q y

The group
q velocity is less than the speed of light, and takes the form of a massive particle with
m = qx2 + qy2 .
The procedure can be followed for any cross-sectional shape, assuming the wave-guide is trans-
lationally invariant along the z axis. One can always divide the solutions into TM and TE modes.
For the TM modes, one can solve the boundary conditions first for the function ψ(x, y), which
gives Ez (x, y) by the relation,

Ez = ψ(x, y)e−iωt+ikz z . (7.23)

One can solve the equations for ψ from the differential equation,

−(∂x2 + ∂y2 )ψ = −(ω 2 − kz2 )ψ, (7.24)

75
PHY 841 7 ELECTROMAGNETIC WAVES

with the boundary conditions

ψ(x, y)|S = 0. (7.25)
This boundary condition forces Ez to be zero at the surface. Once one has solved the boundary
condition, the transverse components of the electric and magnetic fields can be found via,
ikz
⃗ t (x, y) =
E e−iωt+ikz z ∇t ψ(x, y), (7.26)
(ω 2 − kz2 )
 
ω
⃗ t (x, y) =
B ẑ × E
⃗ t.
kz
The expression for E ⃗ t manifestly points into the surface because ψ is constant along the surface.
⃗ t , which is normal to both E
Then, B ⃗ t and ẑ must lie parallel to the surface. Our next task is to
show that these expressions solve each of the four Maxwell’s equations. For example,
∇×B
⃗ =? ∂t E,
⃗ (7.27)
   
ω
∇ × (ẑ × E ⃗ t + ψe−iωt+ikz z ẑ ,
⃗ t ) =? −iω E (7.28)
kz
1 1  
−iωt+ikz z
ẑ(∇ · E
⃗ t) − ∂z Et =? −i Et + ψe
⃗ ⃗ ẑ , (7.29)
kz kz
∇2t ψ
=? −ψ. ✓ (7.30)
(ω 2 − kz2 )
One can check the other Maxwell equations (See HW problems).
For TE modes, a similar procedure can be applied. In that case
Bz = ψ(x, y)e−iωt+ikz z , (7.31)
with ψ(x, y) still satisfying Eq. (7.24), but with the boundary conditions
(n̂ · ∇t )ψ(x, y)|S = 0. (7.32)
The transverse fields are then given by,
ikz
⃗ t (x, y) =
B e−iωt+ikz z ∇t ψ(x, y), (7.33)
(ω 2 − kz2 )
 
ω
⃗ t (x, y) = −
E ẑ × B
⃗ t.
kz

7.5 Homework Problems

1. Consider the Lagrangian density
1 µν λ
L = F Fνµ − (∂ · A)2 .
4 2
R 4
Here, the action is S = (1/4π) d x L. The extra term here (proportional to λ) is known
as the gauge-fixing term, which is a misnomer. This term explicitly destroys gauge invari-
ance, so after one calculates quantities, one must actually use the Lorentz gauge, ∇ · A = 0
to recover the correct result.

76
PHY 841 7 ELECTROMAGNETIC WAVES

(a) Consider the Feynmann gauge (not actually a gauge, but just a choice for λ), where
λ = 1. Show that the Lagrangian, after integrating the action by parts, becomes

1
L = − ∂ α Aβ ∂α Aβ .
2

(b) Solve for the equations of motion in the Feynmann gauge. Then, setting ∂·A = 0, and
using Eq.s (2.40), show that the equations of motion become the Maxwell’s equations
for free space,

∇·E
⃗ = 0, (7.34)
∇×B⃗ = −∂t E.
⃗

(c) (Extra Brownie Points) Write down the stress-energy tensor in the Feynmann gauge.
Then, set ∂ · A = 0, show that T00 = (|E|
⃗ 2 + |B|
⃗ 2 )/2. You may also need to apply
equations of motion and use some messy vector identities.

2. Consider solutions for electro magnetic waves of frequency ω moving in the ±z directions
which are linearly polarized in the x direction. Assume the incoming wave has a form
E0 x̂e−iωt+ikz /2.

(a) Find the linear combination of such waves where the electric field vanishes at z = 0,
⃗ and B
i.e. reflecting off a conducting plane. Express your answer for both E ⃗ as a real
field in terms of sines and cosines.
(b) Find the elements of the stress-energy tensor as a function of z and t.
(c) Show that the stress-energy tensor is traceless, T i i = 0.

3. Consider a plane wave moving in the z direction according to Eq. (7.4) with ay = iax and
ax real. Taking the real part of the solution, solve for the direction of ⃗
a as a function of time
and position.

4. Consider a simple model of the universe where the expansion velocity for cosmological
purposes is ⃗ v = ⃗r /t. This corresponds to a “flat” universe with gravitational effects ig-
nored. All matter starts at a point (the origin) and there is no acceleration for any fluid
element. Observer A is moving with the source, and she records light being emitted at a
time τ0 = 105 years after the birth of the universe, according to a clock in her pocket. A
second observer, B, records the light moving past at a time τ = 1.4 × 1014 years after
the beginning of the universe according to a watch in his pocket. Both A and B are at rest
relative to the neighboring expanding matter. If observer A records the frequency of the
emitted light as being f0 , find the frequency f of the recorded light according to observer
B.
Some Help: the time measured by the co-moving observer, τ , is related to the time mea-
sured by a different observer with velocity v by the relations:

t p p p
τ = = t 1 − v 2 = t 1 − r 2 /t2 = t2 − r 2 .
γ

77
PHY 841 7 ELECTROMAGNETIC WAVES

5. Show that there is no solution to the conditions for the rectangular wave-guide amplitudes
in Eq. (7.19) when both E0z and B0z are set to zero. This demonstrates that there are no
solutions other than the TE and TM solutions.

6. Show that Eq.s (7.23-7.26) satisfy the Maxwell relation ∇ × E

⃗ = −∂t B.
⃗

7. Consider a cylindrical wave-guide of radius R. Consider the lowest frequency TM solution

to the generating function ψ satisfying the differential equation

∇2t ψ(ρ, ϕ) = −α2 ψ(ρ, ϕ), α2 = ω 2 − kz2 ,

where kz is the wavenumber for longitudinal motion.

(a) Find a solution for ψ in polar coordinates. Express answer in terms of a0 , the first
zero of the Bessel function J0 .
(b) Find expressions for the electric and magnetic fields.
(c) What is the group velocity of a wave with momentum kz .

8. Consider two infinite parallel plates with the plane of the plates being along the x direction
and the separation being Lx , i.e. a rectangular wave guide with Ly = ∞. Consider a
wave moving in the z direction with wave number kz . Using the method of generating
functions,

(a) Solve for the lowest frequency TM wave. Find expressions for the fields and the group
velocity.
(b) Solve for the lowest frequency TE wave. Again find expressions for the fields and the
group velocity.

78
PHY 841 8 RADIATION

8 Radiation

Here we discuss classical radiation. The source of such radiation is the J · A terms in the La-
grangian, and the “classical” assumption is that the current does not change due to the radiated
photon. In contrast, quantum emission involves changing discrete levels. For instance one could
fall from a p-state to an s-state, with the frequency determined by the change in energy levels.
For classical emission the frequency is a property of how charge moves within the source, and it
is assumed that there is no feedback from the radiation that alters the current. Additionally, the
classical assumption ignores the fact that light of a given frequency carries quantized amount of
energy, but this latter part of the assumption is irrelevant if one records many photons.

8.1 Coupling the Electromagnetic Field to Dynamic Currents and Charges

To incorporate dynamics into the generation of electric and magnetic fields, we consider Maxwell’s
equations for electric and magnetic fields, written in term of the four-vector potential Aµ ,

∂α F αβ = 4πJ β , (8.1)
∂α (∂ α Aβ − ∂ β Aα ) = 4πJ β ,
∂ 2 Aβ − ∂ β (∂ · A) = 4πJ β .

Similar as to our derivations for magneto-statics, this last expressions could be treated as four
separate Poisson’s equations if the second term of the bottom expression would disappear.
Again, we argue that this term can be ignored because one can choose a gauge that explicitly
cancels it, and given that physical results cannot depend on the gauge, that term cannot matter.
The choice of Λ,

Aα α α
(new) = A(old) + ∂ Λ, (8.2)

must be able to enforce ∂ · A(new) = 0.

∂ · A(new) = ∂ · A(old) + ∂ 2 Λ = 0, (8.3)

∂ 2 Λ = −∂ · A(old) .

Just as before, for any scalar function ∇ · A(old) , one can find a solution to Poisson’s equation for
Λ where ∂ · A(old) serves as a source. Physics doesn’t depend on Λ, so we choose Λ to satisfy
the gauge constraint, and are thus left with simpler equations. This is know as the Lorentz
gauge, and aside from being convenient, has an attractiveness due to its invariance to boosts or
rotations,

∂ · A = 0. (8.4)

We are thus left with four independent equations,

∂ 2 Aα = 4πJ α . (8.5)

When the right-hand side was set to zero this is known as Laplace’s equation, when the right-
hand side is non-zero, and independent of A, it is Poisson’s equation. The right-hand side can

79
PHY 841 8 RADIATION

be thought of as a source term. If there are no fields at t → −∞, any subsequent appearance
of fields is due to the source term. The equation is linear in A, so one can write the solutions as
a sum (integral) over solutions from each differential contribution on the right-hand side. If the
right-hand side were,
∂ 2 G(x) = δ 4 (x), (8.6)
and if the solution G> satisfied the boundary conditions of being zero for all negative times, one
can apply this solution to write solutions for any form on the right-hand side. For an arbitrary
source function, S(x), where one is searching for a solution F (x),
∂ 2 F (x) = S(x), (8.7)
one could write the solution for F as
Z
F (x) = d4 x′ S(x′ )GR (x − x′ ), (8.8)
x′0 <x0

which effectively is taking a linear combination of solutions for each differential source element
to generate a solution for a continuous source. For the case here, the source function for solving
for the evolution of Aα (r) is 4πJ α (r ′ ), and
Z
α
A (x) = 4π d4 x′ J α (x′ )GR (x − x′ ). (8.9)
x′0 <x0

The function GR (r), the retarded Green’s Function, is independent of the source, and applies
to any Poisson equation. The retarded solution is the solution to the differential equation in Eq.
(8.6) that vanishes for negative times.
The Green’s function is found by solving Eq. (8.6). By inspection, one can see that
Z ⃗
−1 3
eiωt−ik·⃗r
G(t, ⃗
r) = dωd k 2 (8.10)
(2π)4 ω − k2
is a solution to Eq. (8.6). However, this solution does not satisfy the boundary conditions that it
vanish for all negative times. Altering the solution to satisfy the boundary condition,
Z ⃗
−1 eiωt−ik·⃗r
GR (t, ⃗
r) = dωd k3
, ϵ→0+. (8.11)
(2π)4 (ω − k − iϵ)(ω + k − iϵ)
Performing the integral over ω by contour integration,
Z
−i 1  ikt−i⃗k·⃗r −ikt−i⃗

GR (t, ⃗
r) = d 3
k e − e k·⃗
r
Θ(t) (8.12)
(2π)3 2k
Z
1 1
d3 k sin(kt)e−ik·⃗r Θ(t).
⃗
= 3
(2π) k
⃗
r and ⃗
Next, choosing θ as the angle between ⃗ k, writing eik·⃗r = eikr cos θ , and integrating over
cos θ then k,
Z ∞
1
GR (t, ⃗
r) = dk sin(kt) sin(kr)Θ(t) (8.13)
4π 2 r −∞
1
= δ(t − r).
4πr

80
PHY 841 8 RADIATION

This allows the vector potential to be written in terms of an integral driven by currents at previ-
ous times,
Z
1
α
A (x) = d4 x′ J α (x′ )δ(x0 − x′0 − |⃗
x−⃗x′ |). (8.14)
|⃗
x−⃗ x′ |
The physical interpretation of this is clear, the vector potential is driven by the configuration of
the currents along the light-front. For the static case, J α does not depend on time, one quickly
recovers the usual expressions given in the previous sections (See HW problem).

8.2 Radiation from an Accelerating Point Charge

Radiative energy flux is given by the Poynting vector, S ⃗ × B/4π,
⃗ = E ⃗ and because for static
⃗ and B
cases E ⃗ fall off at least as quickly as 1/r , the flux must fall off faster than 1/r 2 and thus
2

there is no radiated energy. However, this changes for the case of accelerated charges because
the current density depends on time. . We consider the equation for vector potential, Eq. (8.14),
and consider a moving point charge of charge e. In that case,

d4 x′ J α (x′ ) = euα dτ, (8.15)

when the differential covers the particle’s position. Here, dτ is the differential time as measured
in the frame of the charge. To verify this, consider the frame of the particle, u = (1, 0, 0, 0). In
that frame J 0 d3 x = e, Jd⃗ 3 x = 0, which is the correct answer, and because the Lorentz indices
match, d4 x is a scalar, it must be correct in all frames.
The vector potential is then
Z
α
A (x) = 2e dτ uα (τ )δ[(x − r(τ ))2 ]Θ(x0 − r0 ), (8.16)

where r α (τ ) is the trajectory of the particle. The extra factor of 2 and stepfunction, and the
missing 1/|x − ⃗ r| arise from the fact that the delta function is now a function of the invariant
distance squared. Given that (d/dx)δ(f (x)) = δ(f (x))/f ′ , the expressions are equivalent.
Because the new delta function would also contribute when r0 > x0 , the step function is added.
To find the electro-magnetic fields,
Z
∂ A = 2e dτ uβ ∂ α δ[(x − r(τ ))2 ]
α β
(8.17)
Z
∂δ[(x − r(τ ))2 ] ∂τ
= 2e dτ uβ ∂ α (x − r(τ ))2
∂τ ∂(x − r(τ ))2
Z
(x − r)α ∂δ[(x − r(τ ))2 ]
= −2e dτ u β
(r − x) · u ∂τ
Z  β 
u (τ )(r(τ ) − x)α
= 2e dτ δ[(x − r(τ )) ]∂τ 2
.
u(τ ) · (r(τ ) − x)

Here, we have made use of the fact that uα = dr α /dτ . We are only interested in the fields
far away, x0 ≈ |⃗
x| >> ri . Thus, we need only worry about derivatives of u above because

81
PHY 841 8 RADIATION

derivatives of x would result in additional factors of 1/x. Taking the derivatives, then throwing
away all terms that fall off too quickly,
Z  α β 
x a xα uβ
∂ A (x) ≈ 2e dτ δ[(x − r(τ )) ]
α β 2
a −
β
(a · x) , (8.18)
(u · x) (u · x)2
d β
aβ ≡ u .
dτ
The zeroth component a can be rewritten (see HW problem) as a0 = u ⃗ ·⃗
a/u0 . One can now
again use the chain rule for delta functions to obtain
 
e uβ (a · x)
α β
∂ A (x) = x α
a −
β
. (8.19)
(u · x)2 (u · x)

We are interested in radiation, which implies the long-distance limit where x → ∞, so we

replaced the ≈ sign in Eq. (8.18) with an equal sign and replaced u · (x − r) with u · x.
The electric and magnetic fields can now be calculated,
    
e ui (a · x) u0 (a · x)
i
E = x a −
0 i
−x a −
i 0
. (8.20)
(u · x)2 (u · x) (u · x)

With a significant amount of algebraic effort, one can express this in the form,
 
 n̂ × [(n̂ − β)⃗ × β] ⃗˙ 
⃗ = e
E . (8.21)
 (1 − β x| 
⃗ · n̂)3 |⃗

Here, n̂ is a unit vector in the direction of ⃗ ⃗=u

x, and β ⃗ /u0 = ⃗v /c is the non-relativistic velocity.
Again, with a bit of effort, one can see that the magnetic field is given by

⃗ = n̂ × E.
B ⃗ (8.22)

The direction of E⃗ defines the polarization. For small velocities that direction can be found by
⃗˙ then projecting out the part of the vector along ⃗
taking the direction of β, x. The small velocity
limit, β << 1, also makes it easy to calculate the net power. The power per area is given by the
magnitude of the Poynting vector. For β << 1,
1
|S|
⃗ = |E|
⃗ 2 (8.23)
4π
e2 ⃗˙ 2 sin2 θ,
= |β|
x|2
4π|⃗

x|2 d cos θ, the net

⃗ Integrating over the area, dA = 2π|⃗
where θ is the angle between n̂ and β.
power is

2e2 ⃗˙ 2
P = |β| , (8.24)
3c

82
PHY 841 8 RADIATION

which is known as Larmor’s formula. The factor 1/c was added should you need to calculate in
units where c ̸= 1.
For relativistic motion, the expression is more complicated. In Eq. (8.21) one cannot ignore the
⃗ in the numerator, but more dramatically, the denominator,
β

1
, (8.25)
(1 − β
⃗ · n̂)3

diverges as β → 1 when the emission direction ⃗ ⃗ This is especially

x is parallel to the velocity β.
true for circular motion of electrons in a high energy accelerator. Due to their light mass β is
very nearly unity. The acceleration is inward, toward the center of the motion, so one can use
Eq. (8.21) to see that the emitted light is linearly polarized with E ⃗ perpendicular to both the
acceleration and the direction of the light, which tends to be parallel to the velocity. Thus, that
light is polarized perpendicular to the plane of the motion.

8.3 Liénard-Wiechert Potentials

The expressions for the fields due to currents in the previous section assumed that the observer
was far away, in both distance and time, from the fields. A more general expression for the
vector potential was given in Eq. (8.16),
Z
A (x) = 2e dτ uα (τ )δ[(x − r(τ ))2 ]Θ(x0 − r0 ),
α
(8.26)
euα (τ )
= .
u · (x − r(τ ))

Here r(τ ) is the position of the particle at a time when a light-pulse from the particle’s trajectory
would reach space-time point x. Effectively, one would follow the trajectory until one found
such a point. Because the particle moves slower than the speed of light, there can be only one
such point. The factor 1/(u · (x − r)) can be factored into

u · (x − r) = u0 (x0 − r0 ) (1 − n̂ · ⃗
v) , (8.27)
α
e(u (τ )/u0 )
Aα (x) = ,
1 − n̂ · ⃗
v
x−⃗
where n̂ is the unit vector parallel to ⃗ r . As was seen in the last section, in the non-relativistic
limit this factor simply provides the inverse distance from the point of emission to the observer.
However, relativistically the additional factor (1− n̂·⃗v )−1 amplifies the response of the potential
to the charge when the velocity approaches the speed of light and is pointed toward the observer.
In a simulation of relativistic charged particles interacting electromagnetically, Liénard-Wiechert
potentials offer one way (probably not the most computationally efficient way) of including re-
tardation effects. Rather than assuming instantaneous interactions, where the potentials depend
on the current positions and currents of the particles, one would calculate the contribution to the
Aα (x) for some point x by using the positions and velocities of each charge at the appropriate
retarded time.

83
PHY 841 8 RADIATION

8.4 Radiation from relativistic particles

The velocity of a proton in the LHC is 99.9999991% the speed of light. Electrons are lighter, so
even though the energy of the LEP ring of the LHC is smaller than that of the LHC, the speed of
an electron in LEP is 99.9999999988% the speed of light. In such cases the relativistic factors are
surprisingly large and non-trivial.
Using the results of Eq.s (8.21) and (8.22), the power per solid angle from an accelerating charge
is
dP e2 ⃗˙ 2 .
= |(n̂ − β)
⃗ × β)| (8.28)
dΩ 4π(1 − β
⃗ · n̂)6

This represents the energy emitted into the solid angle during a time dx0 . During that time
interval dx0 , the time interval of the particle’s trajectory that contributed to the emission is
dr0
dr0 = dx0 , (8.29)
dx0
(r − x)2 = 0,
2(r0 − x0 )(dr0 − dx0 ) − 2(⃗
r−⃗ x) · d⃗
r = 0,
h i
dr0 r0 − x0 − β⃗ · (⃗
r−⃗ x) = dx0 (r0 − x0 ),
dr0 1
= .
dx0 (1 − β
⃗ · n̂)

The energy emitted per time interval of the accelerated charge is thus

dP e2 ⃗˙ 2 .
= |n̂ × [(n̂ − β)
⃗ × β)]| (8.30)
dΩ 4π(1 − β · n̂)
⃗ 5

⃗ and n̂. With a devoted effort, one can integrate the differential
Here, θ is the angle between β
expression for power. The result for the remarkably difficult integral is
2 h i
P = e2 γ 6 β̇ 2 − |β ⃗˙ 2 .
⃗ × β| (8.31)
3c
According to Jackson, this is the Liénard result from 1898, which is remarkable given that special
relativity was not generally explained until 1905.
First, we consider the case where β ⃗˙ are parallel or anti-parallel. This would be the case for
⃗ and β
a linear accelerator or for a charge gradually stopping. In that case

dP e2 ⃗˙ 2 sin2 θ.
= |β| (8.32)
dΩ 4π(1 − β cos θ)5
The unusual angular shape is solely determined by the magnitude of the velocity β. It vanishes
in the forward direction due to the sin2 θ factor, then has a maxima at an angle θmax < π/2. In
the β << 1 limit, the maximum θmax ≈ π/2, and in the ultra-relativistic limit the maximum
θmax ≈ 0 and the emission is strongly forward-peaked. The angular shape is identical for

84
PHY 841 8 RADIATION

3 ×104
β =0.95
β =0.7( ×200)
β =0.3( ×1000)
2 ×104
sin2 θ/(1−cosθ)5
104

-1.0 -0.8 -0.6 -0.4 -0.2 -0.0 0.2 0.4 0.6 0.8 1.0
0

cosθ
Figure 8.1: The angular dependence from Eq. (8.32) is shown for three velocities. As β approaches the
speed of light, the strength of the radiation increases dramatically and the emission becomes strongly
forward-peaked. For the same acceleration, radiation increases by more than three orders of magnitude
as β increases from 0.5 to 0.95.

⃗ not in the direction of

accelerating or decelerating particles, and is peaked in the direction of β,
⃗˙ From Eq. (8.31) one can see,
β.

2e2 β̇ 2
P = γ 6. (8.33)
3c

Another interesting example is for when β ⃗˙ as is the case

⃗ is perpendicular to the acceleration β,
for circular motion. In this case Eq. (8.30) becomes

dP e2

= | ⃗˙ 2 (1 − βnβ )2 − (1 − β 2 )n2 .
β| (8.34)
dΩ 4π(1 − βnβ )5 r

Here, nβ is the component of n parallel to β and nr is the component parallel to β. ⃗˙ As nβ → 1,

nr vanishes and the angular dependence scales as 1/(1 − β)3 . Unlike the form for the case
where the acceleration and velocity were parallel, this does not vanish as θ → 0, and becomes
infinite in the limit that β → 1. For this reason, electron accelerators are excellent candidates for
high-luminosity light sources. From Eq. (8.31) the net power is then

2
P = e2 β̇ 2 γ 4 . (8.35)
3c

⃗˙ by two powers of γ.
⃗ is parallel to β
This differs from the case where β

85
PHY 841 8 RADIATION

8.5 Frequency Dependence of Radiation from Accelerated Charge

For some trajectory, ⃗
r (τ ), one can calculated the contribution to the electric field from Eq. (8.20),
⃗
E(x), and then find the Fourier transform of the electric field, E(⃗ ⃗˜ x, ω). From this, one can
calculate the energy radiated per unit frequency at some large distance R and angle Ω,
Z
dU 1 2
= R ⃗
dt E(R, t)2 (8.36)
dΩ 4π Z
1 ˜ ⃗
= R 2
dtdωdω ′ ⃗
E( R, ω)E⃗˜ ∗ (R,
⃗ ω ′ )ei(ω−ω′ )t
4π(2π)2
Z
1 2
= R 2
dω ⃗˜ R,
E( ⃗ ω) ,
8π 2 Z ∞
1 2
= R 2
dω ⃗˜ R,
E( ⃗ ω)
2
Z4π 0
˜ ⃗ ω) ≡
⃗ R, ⃗ R,⃗ ω).
E( dt eiωt E(

⃗˜
Now, using Eq. (8.20) to express E,
 
Z ˙
e n̂ × [n̂ − ⃗
β] × ⃗
β
⃗˜
E(ω) = dt eiωt   , (8.37)
R (1 − β ⃗ · n̂)3
 ret

Z ˙
e n̂ × [n̂ − ⃗
β] × ⃗
β
dt′ eiω(t +R−n̂·⃗r(t ))  .
′ ′
=
R (1 − β ⃗ · n̂)2

The third step shortened the integral over all frequencies to simply those over positive frequen-
⃗˜
cies by noting that |E(ω)| 2
is the same for ω and −ω. The first expression involves calculating
the velocities at the retarded times but the integral is over the observers time, whereas the sec-
ond expression involves replacing the integral over the observer’s time with an integral over the
retarded time, t′ ,
t = t′ + R − n̂ · ⃗
r (t′ ), R ≡ |⃗
x|, (8.38)
′
dt/dt = 1 − β ⃗ · n̂.
which should be accurate for large R. For the next step we use an identity (See H.W. problem),
" #
n̂ × (n̂ − β̂) × β⃗˙ d n̂ × (n̂ × β)⃗
= . (8.39)
(1 − β⃗·⃗n)2 dt 1−β ⃗ · n̂

⃗˜
Inserting this into Eq. (8.37) allows us to express E(ω) in terms of an integral involving only the
velocity,
t′ =tf
e n̂ × (n̂ × β)
⃗
⃗˜
E(ω) = e iω(t′ +R−n̂·⃗
r (t′ ))
(8.40)
R 1−β ⃗ · n̂
t′ =ti
Z tf h i
e ′ iω(t′ +R−n̂·⃗
r (t′ ))
−i ω dt e n̂ × (n̂ × β) .
⃗
R ti

86
PHY 841 8 RADIATION

One can add a factor e−ϵ|t| to each integrand, which makes it possible to discard the first term.
Effectively this term represents slowing down the currents at large times with infinitesimally
slow accelerations, which doesn’t cause any radiation, but allows one to regulate the integrals.
Thus, for the radiative energy,
Z h i2
dU e2 ω 2 ′ ′
= dt′ eiω(t −n̂·⃗r(t )) n̂ × (n̂ × β)
⃗ . (8.41)
dωdΩ 4π 2

8.6 Radiation from Oscillating Systems with Well-Defined Frequencies

All currents can be expressed in terms of Fourier transforms,
Z
1
α
J (x) = x)e−iωx0 .
dω J˜α (ω, ⃗ (8.42)
2π
Each frequency component contributes to the vector potential. Equation 8.14 becomes
Z
1 ′ x′ )
J˜α (ω, ⃗
α
A (⃗
x, t) = dωd3 x′ dt′ e−iωt δ(t′ + |⃗ x′ | − t)
x−⃗ (8.43)
2π |⃗
x−⃗ x′ |
Z
1 ′ J˜α (ω, ⃗x′ )
= dωd3 x′ eiω(|⃗x−⃗x |−t) .
2π |⃗
x−⃗ x′ |

Any classically radiating system could be treated using the Fourier transform. However, this
does not always simplify the problem. If the J˜ has a range of frequencies, they all contribute
to the vector potential, and the intensities, which require squaring the fields, will likely involve
integrals over two frequencies, dωdω ′ J˜(ω)J˜(ω ′ ) · · · .
However, the expressions simplify significantly when only one well defined frequency enters
the problem. In that case we forego using Fourier transforms, and instead assume that the time
dependence of J (x) factors into a single phase e−iω0 t ,

J α (x) = e−iω0 t j α (⃗
x). (8.44)

This would be reasonable for well-designed antennas, or some other system driven by an os-
cillating term. Some systems have a characteristic frequency, e.g. the orbital frequency in a
synchotron, but the Fourier transform would include many harmonics of that motion due to the
fact that the current of a single electron looks like a series of delta function pulses separated by
the orbital period. Note j0 is determined by ⃗j through current conservation.

x) = ∇ · ⃗j(⃗
iω0 j0 (⃗ x). (8.45)

In this case the solution for A from Eq. (8.14) becomes

Z
−iω0 t ′ j α (⃗x′ )
α
A (⃗ x, t) = e d3 x′ eiω0 (|⃗x−⃗x |) (8.46)
|⃗
x−⃗ x′ |
Z α ′
′ j (⃗ x)
= e−ikt d3 x′ eik(|⃗x−⃗x |) ,
|⃗x−⃗ x′ |

87
PHY 841 8 RADIATION

where k = ω0 /c is the wave number for light with frequency ω0 . Here, we switch the notation
from ω0 to k to emphasize that for low frequencies we can expand about kx′ being a small
number.
We are interested in radiation, r → ∞, so we can expand

|⃗ x′ | ≈ r − n̂ · ⃗
x−⃗ x′ , (8.47)

where once again n̂ points in the direction of ⃗r≡⃗ x. For radiation, we only wish terms that fall
2
of as 1/r, so the Poynting vector falls as 1/r , so
Z
eik(r−t) ′
α
A (⃗
x, t) = d3 x′ j α (⃗
x)e−ikn̂·⃗x . (8.48)
r
′
For slower frequencies one can expand e−in̂·⃗x in powers of k. This expansion converges well if
the period of the oscillating source is much longer than the time it takes light to cross the source.
Z
eik(r−t) X (−ik)n
α
A (⃗
x, t) = d3 x′ j α (⃗
x′ )(n̂ · ⃗
x′ )n . (8.49)
r n
n!

8.7 Electric and Magnetic Dipole Radiation

If one keeps the lowest term in the expansion of Eq. (8.49),
Z
eik(r−t)
⃗ x, t) =
A(⃗ d3 x′ ⃗j(⃗
x′ ), (8.50)
r
Z
eik(r−t)
Ai (⃗
x, t) = d3 x′ jk (⃗
x′ )(∂k x′i )
r
Z
eik(r−t)
= − d3 x′ x′i (∇ · ⃗j)
r
ik(r−t) Z
e
⃗ x, t) =
A(⃗ − d3 x′ ⃗
x(∇ · ⃗j(⃗ x′ ))
r
Z
eik(r−t)
= −ik d3 x′ ⃗x′ j0 (⃗
x′ )
r
ik(r−t)
e
= −ik p
⃗
Z r
p
⃗ = d3 x′ ⃗x′ j0 (⃗
x′ ).

After factoring out the e−iωt factor from ρ(⃗ ⃗ looks like the electric dipole moment. Here,
x, t), p
the second step exploited current conservation. The magnetic and electric fields then become

eik(r−t)
⃗ = k2 (n̂ × p
B ⃗) , (8.51)
r
eik(r−t)
E p − n̂(⃗
⃗ = k2 (⃗ n·p
⃗)) .
r

88
PHY 841 8 RADIATION

The power per solid angle is then

dP 1
= ⃗ ×B
E ⃗ (8.52)
dΩ 8π
1
= k2 |n̂ × p
⃗|2 .
8π
The factor of 1/8π, instead of the usual 1/4π expression for the Poynting vector arises because
one uses only the real part of the fields, and the average of cos2 [k(r − t)] is one half. Finally,
one can integrate the flux to find the net power radiated,
Z
1 2 2
P = k p dΩ sin2 θ (8.53)
8π
k4 2
= |⃗
p| .
3
This is referred to as electric dipole radiation.
To extract the magnetic dipole moment one can look at the next power (n = 1) in the expansion
in Eq. (8.49). This is one higher power in k,
Z
eik(r−t)
A⃗ = (−ik) d3 x′ ⃗j(⃗
x′ )(n̂ · ⃗
x′ ). (8.54)
r
For the next step we make use of a vector identity
1 1 1
⃗ B
A( ⃗ · C)
⃗ = ⃗ B
A( ⃗ · C)
⃗ + ⃗ A
B( ⃗ · C)
⃗ + ⃗ × (A
C ⃗ × B).
⃗ (8.55)
2 2 2
With this identity,
Z Z
3 ′⃗ ′ ′
1
x )(n̂ · ⃗
d x j(⃗ x) = d3 x′ n̂ × [⃗
x′ × ⃗j(⃗ x′ )] (8.56)
2 Z
1 h i
+ d3 x′ ⃗j(⃗x′ )(n̂ · ⃗
x′ ) + ⃗
x′ (n̂ · ⃗j(⃗
x′ ) .
2
As was done for the electric dipole case, the second term can be written in terms of the charge
density by applying current conservation and integrating by parts.
Z Z
3 ′⃗ ′ ′
1
d x j(⃗ x )(n̂ · ⃗
x) = d3 x′ n̂ × [⃗
x′ × ⃗j(⃗ x′ )] (8.57)
2
Z
ik
− d3 x′ ⃗
x′ (n̂ · ⃗
x′ )j0 (⃗
x′ ).
2
The second term involves second moments of ⃗ x′ and leads to electric quadrupole radiation. As
such, it is of a higher power of k. The first term is the magnetic dipole term,

eik(r−t)
⃗
A = (−ik) m̂ × ⃗ n (8.58)
Z r
1
m
⃗ = d 3 x′ ⃗
x′ × ⃗j(⃗
x′ ).
2

89
PHY 841 8 RADIATION

Even though the magnetic dipole contribution as one higher power in k from the expansion in
Eq. (8.49) compared to the electric dipole term, both terms came out linear in k. This is because
the use of current conservation, ikρ = ∇·⃗j, added an additional factor of k to the electric dipole
term. Calculating the electric and magnetic fields,

eik(r−t)
B(⃗ ⃗ −⃗
⃗ r , t) = k2 (m n · m))
n(⃗ ⃗ , (8.59)
r
eik(r−t)
⃗ r , t) = −k2 (⃗
E(⃗ n × m)
⃗ .
r
For the electric dipole case, the polarization is defined by the electric field pointing along the
direction of p⃗ after the n̂ component was projected away. For magnetic dipole radiation, the
direction of the magnetic field in the wave is along the direction of m
⃗ after the n̂ component is
projected away.
Electric dipole radiation can come from having charge oscillating back and forth along a wire. It
also ensues from having a charge move in a circle. One would think that a particle moving in a
circle would have magnetic dipole radiation because ⃗ x′ × J⃗ is non-zero. However, this is con-
stant in time for circular motion, so there is no finite frequency component, and thus radiation
from circular motion proceeds through the electric dipole form.

8.8 Homework Problems

1. Consider Eq. (8.14) in the case where J α has no time dependence. Show that one quickly
obtains the usual expressions for the potentials in the static case.

2. Using the fact that ∇2 (1/r) = −4πδ 3 (⃗

r ),

(a) show that any function f (r − t) satisfies the differential equation,

 
f (r − t)
∂ 2
= 4πf (r − t)δ 3 (⃗r ).
r

(b) Now, let f (r − t) = δ(r − t). Show that this satisfies the equation
 
2
f (r − t)
∂ =0
r
for all t > 0. Also, because r > 0 the function is zero for t < 0.
(c) Show that the form f (r − t) = δ(r − t) satisfies the integral of Eq. (8.7).
Z ϵ    Z
δ(r − t)
dt ∂ 2
= −4π dt δ 4 (x).
−ϵ r

3. Consider a function f (x) that is a super-position of plane waves,

Z
′ ′ ′
f (x) = dk g(k′ )eiω(k )t−ik x+iϕ0 (k ) ,

90
PHY 841 8 RADIATION

where g(k′ ) is a narrow function centered about k, e.g.

1 ′
g(k′ ) = √ e−(k −k)
2 /2a2
,
2πa2
with a << k.
(a) For a given time t find the position x at which the phase [iω(k′ )t − ik′ x + iϕ0 (k′ )]
is steady as a function of k′ at k′ = k, i.e.
d
[iω(k′ )t − ik′ x + iϕ0 ] = 0.
dk′
(b) What are the group velocities for the following cases:
a) massless particle in a vacuum, ω = |k|c
b) massive particles in a vacuum, ℏω = (ℏk)2 /2m
c) plasma oscillation, ω 2 = ωp2 + 3k2 vth
2
.
4. Show that u · a = 0, where u is the four-velocity and a = (d/dτ )u is the acceleration.
⃗ ·⃗
Then show that a0 = u a/u0 .
5. Show that the electric field given in Eq. (8.20) is perpendicular to ⃗
x.
6. Consider Eq. (8.24):
(a) Using the fact that e2 /(ℏc) = α is dimensionless, show that Eq. (8.24) gives dimen-
sions of energy per time.
(b) Suppose you had one Coulomb of charge and dropped it off a building, where it
accelerated downward with g = 9.8 m/s2 . What power (in W) would be radiated
while it fell?
7. The circumference of the LHC is 27 km, and the energy of a proton in the ring is 6.5 TeV.
The beam current of the LHC is 0.58 Amperes. (The mass of a proton is 938.3 MeV).
(a) What is the acceleration of a proton? Assume it moves in perfect circular motion,
though in reality it passes between magnets for parts of its trajectory, so the accelera-
tion falls between magnets and is higher while the proton is inside the dipoles.
(b) What is the power radiated by the proton?
(c) What fraction of the energy is lost during one revolution of the trajectory?
(d) If electrons were accelerated to the same energy, what would the fraction be? (The
mass of an electron is 0.511 MeV)
(e) If electricity can be purchased for 10¢ per kwh, estimate the cost of the LHC due to
radiative energy loss for running one day. Give cost for both protons and electrons (if
they were put in at the same energy).
8. Consider a particle of charge e moving non-relativistically in a synchotron of radius R
with the orbit around the z axis, such that

x = R cos ωt, y = R sin ωt.

91
PHY 841 8 RADIATION

(a) Find Jx (⃗
r , t) as defined in Sec.s 8.6 and 8.7.
(b) Mis-stated in original assignment – ignore
(c) Find px as defined in Sec. 8.7.
(d) Using Eq. (8.53), what is the radiated power? Be sure to include contribution from
both px and py .
(e) Compare to the result for a non-relativistic point particle moving in a circle from Eq.
(8.35).
(f) Why should you not apply Eq. (8.53) in the relativistic case?

92
PHY 841 9 SCATTERING

9 Scattering

Here we discuss scattering of light, i.e. light comes in with one wave vector, then leaves with
another. This includes Thomson scattering, Rayleigh scattering and Compton scattering.

9.1 Scattering in the Long Wavelength Limit

If the object from which light is scattered is small compared to the wavelength of light, it is
convenient to think of the problem in two steps. First, the object is excited by the electromagnetic
wave, inducing oscillating multipole moments of either the current or charge. These objects
then radiate according to the moments. The incoming electromagnetic wave provides a driving
force at the frequency of the light, ω. This incoming wave persists for a significant time, which
means that the response will settle down to the particular solution of the the differential equation
describing the particle with which it interacts. This solution oscillates with the same frequency
as the driving frequency. The radiation from this source then also occurs at ω, which means
the process can be considered as elastic scattering. The approach we consider here is only valid
in the classical limit, i.e. it ignores the momentum and energy of the outgoing particle. For
scatterings the momentum transferred to the object, ∆p, can approach 2ℏω if the scattering is
backward and the object is heavy. The outgoing kinetic energy of the object of mass M is then
∆p2 /2M ∼ ℏ2 ω 2 /2M , which is negligible for for ℏω << M .

9.2 Thomson Scattering

For scattering off a free particle, the ℏω << M limit is known as Thomson scattering. If one
accounts for the energy of the outgoing object, the frequency of the scattered light is reduced
to conserve the energy of the scattered photon and one has Compton scattering. Thomson scat-
tering was first explained by J.J. Thomson, the same scientist who discovered the electron and
established many of the basic precepts of radioactivity. For Thomson scattering, we consider
the electric field acting on a particle at the origin. If the wave length is long compared to the
subsequent motion, we can consider the electric field as depending on time only,

⃗
E(t) = E0 cos ωt. (9.1)

Here, E0 is the amplitude of the electro-magnetic field. If the polarization is in the x direction,
the particle’s subsequent motion is given by

eE0
ẍ = cos ωt, (9.2)
m
eE0
x = − cos ωt,
mω 2

93
PHY 841 9 SCATTERING

where the particle’s mass and charge are m and e. The dipole moment and emitted power, see
Eq. (8.53), are
e2 E0
px (t) = − cos ωt, (9.3)
mω 2
ω4
P = p2x
3
e4 E02
= ,
3m2
which is independent of ω. This can be expressed as a cross section σ by considering the expres-
sion for the scattering rate,
Γ = nσv, (9.4)
where v in this case is the velocity of light, and n is the density of scatterers. The scattering rate
Γ is the ratio of power emitted from the scatters within a volume V , divided by the electromag-
netic energy of the incoming wave in that volume. The emitted power is the power off a single
electron, Eq. (9.3), multiplied by the density of electrons and the overall volume,
 
e4 E02 8π
Γ = n
3m2 E02
8πne4
= ,
3m2
and the cross section is
8πe4
σ = . (9.5)
3m2
The differential cross section for polarized light, can be calculated with the help of Eq. (8.52),
 
dΓ 1 2 8π
= k |n̂ × p⃗| 2
(9.6)
dΩ 8π E02
dσ e4
= 2
(sin2 θ sin2 ϕ + cos2 θ),
dΩ m
where θ is the angle relative to the incoming wave and ϕ is relative to the polarization. Thus the
scattered light prefers to travel transverse to the polarization of the incoming wave.

9.3 Compton Scattering

For Compton scattering, ℏ plays a role and it served as one of the original examples of quantum
phenomena in the early 1920s. The effect was first observed by Arthur Compton with X-rays,
to make ℏω large, and the scattering was off electrons, the lightest charge particle. In that case
energy and momentum conservation required
p
ℏω + m = ℏω ′ + m2 + p2 , (9.7)
ℏω = ℏω ′ cos θ ′ + p cos θp ,
0 = ℏω ′ sin θ ′ + p sin θp .

94
PHY 841 9 SCATTERING

Next, we use these equations to find an expression where p and θp are eliminated. We first
combine the last two equations to obtain
p2 = ℏ2 (ω 2 + ω ′2 − 2ωω ′ cos θ ′ ), (9.8)
which we then insert into the first equation to obtain,
(ℏω + m − ℏω ′ )2 = m2 + ℏ2 ω 2 + ℏ2 ω ′2 − 2ℏ2 ωω ′ cos θ ′ , (9.9)
2mℏω − 2mℏω ′ − 2ℏ2 ωω ′ = −2ℏ2 ωω ′ cos θ ′ ,
1 1 ℏ(1 − cos θ ′ )
− ′ = ,
ω ω m
′
2πℏ(1 − cos θ ′ )
λ −λ = .
m
where λ = 2π/ω is the wave length. This can be written as
∆λ ℏω
= (1 − cos θ ′ ), (9.10)
λ m
to see that for low frequencies or very heavy targets, the wave length does not change, and one
recovers the Thomson limit.

9.4 Scattering of Light from Confined Charges

Most atoms are neutral, but are made of confined charges. These charges also react to the ex-
ternal electric field. A simple example would be a charge in a damped harmonic oscillator with
characteristic frequency ω0 and damping rate Γ. The equations of motion in an oscillating elec-
tric field are
 
eE0
ẍ − Γẋ + ω0 =
2
cos ωt. (9.11)
m
The particular solution is of the form, x = A cos(ωt − ϕ), and the amplitude A determines the
dipole moment, which then determines the power from Eq. (8.53), and then the cross section,
e2 E02 /m2
|A|2 = , (9.12)
(ω 2 − ω02 )2 + Γ2 ω 2
px = e|A| cos(ωt − ϕ),
ω4 2
P = e |A|2
3
e4 E02 ω4
= ,
3m2 (ω 2 − ω02 )2 + Γ2 ω 2
8πe4 ω4
σ = .
3m2 (ω 2 − ω02 )2 + Γ2 ω 2
If the harmonic oscillator is soft, ω0 << ω, one recovers the free particle limit, Thomson scat-
tering. However, if the oscillator is stiff, ω0 >> ω, the scattering scales as ω 4 . This is the case
for visible light, where the energies are one or two eV, but the excitation energies are several eV.
For this reason blue light scatters much more than red light, and when one looks at the sky away
from the sun, it appears blue. This limit is known as Rayleigh scattering.

95
PHY 841 9 SCATTERING

9.5 Homework Problems

1. For Thomson scattering, show that for un-polarized light the angular distribution of the
scattered light ∼ (1 + cos2 θ), where θ is the scattering angle.

2. Consider the limit that Γ → 0 in Eq. (9.12). When ω → ω0 the cross section then diverges.
Does the contribution to the integrated cross section,
Z ωb
I(ωa , ωb ) ≡ dω σ(ω),
ωa

where ωa and ωb confine the integral to the region surrounding ω0 , diverge as well?

96