Electromagnetism and Geometry

电磁与几何

(preliminary draft updated August 2023)

李思 (Si Li)

Tsinghua University
 Thanks very much for your support of this book! It is greatly appreciated if you are willing
to help improve it by sending your comments such as typo, mistake or suggestion at your tea
time. You are welcome to submit your comment via either the website

https://www.wjx.cn/vm/eITIr9H.aspx

or the barcode below

You can also contact me at sili@mail.tsinghua.edu.cn. The draft will be updated on my

homepage: https://sili-math.github.io/. Thank you.

1
 Contents

Preface 5

Chapter 1 Introduction: Maxwell’s Equations 7

1.1 Hodge Star . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.2 Maxwell’s Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3 Conservation Law of Charge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.4 Coulomb’s Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.5 Ampère’s Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.6 Biot-Savart Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.7 Gauss’s Linking Formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.8 Faraday’s Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.9 Ampère-Maxwell Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
1.10 Potential and Gauge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

Chapter 2 Static Electromagnetism 27

2.1 Electric Field and Scalar Potential . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.1.1 Poisson’s and Laplace’s equation . . . . . . . . . . . . . . . . . . . . . . . 27
2.1.2 Point Charge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.1.3 Uniform Ball . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.1.4 Line Charges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.2 Interface Condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.2.1 Infinite Plane Charge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.2.2 Spherical Shell . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.2.3 Interface Condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.2.4 Electric Conductor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.3 Boundary Value Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.3.1 Uniqueness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.3.2 Green’s Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
2.3.3 Method of Image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.4 Electric Dipole and Dielectric Polarisation . . . . . . . . . . . . . . . . . . . . . . 45
2.4.1 Electric Multipole Expansion . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.4.2 The Electric Dipole . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

2
 2.4.3 Dielectric Matter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
2.5 Magnetic Field and Vector Potential . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.5.1 Magnetic Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.5.2 Interface Condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.6 Magnetic Moment and Magnetic Dipole . . . . . . . . . . . . . . . . . . . . . . . 56
2.6.1 Magnetic Moment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
2.6.2 Magnetic Dipole Layers . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
2.7 Linking and Magnetic Helicity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
2.8 Dirac Monopole . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

Chapter 3 Electrodynamics 66
3.1 Force and Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.1.1 Lorentz Force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.1.2 Electromagnetic Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
3.2 Electromagnetic Induction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.2.1 Electromotive Force and Flux Rule . . . . . . . . . . . . . . . . . . . . . . 71
3.2.2 Mutual Inductance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.2.3 Self Inductance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.2.4 Magnetostatic Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
3.3 Electromagnetic Wave . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
3.3.1 Wave Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
3.3.2 Plane Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
3.3.3 Polarization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
3.3.4 Wave Packets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.4 Green’s Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
3.4.1 Wave Equation with Source . . . . . . . . . . . . . . . . . . . . . . . . . . 82
3.4.2 Green’s Function for the Wave Equation . . . . . . . . . . . . . . . . . . . 84
3.4.3 Retarded and Advanced Solutions . . . . . . . . . . . . . . . . . . . . . . 85
3.5 Dipole Radiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.5.1 Spherical Wave . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.5.2 Electric Dipole Radiation . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
3.6 Moving Point Charge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
3.7 Scattering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

Chapter 4 U (1) Gauge Theory 100

4.1 Fiber bundle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
4.1.1 Fiber Bundle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
4.1.2 Vector Bundle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
4.1.3 Principal Bundle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
4.2 U (1)-connection and Parallel Transport . . . . . . . . . . . . . . . . . . . . . . . 111

3
 4.2.1 Vertical Vector Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
4.2.2 Connection 1-form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
4.2.3 Horizontal Vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
4.2.4 Parallel Transport . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
4.3 Curvature and Chern Class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
4.3.1 Curvature 2-form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
4.3.2 Holonomy and Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
4.3.3 Chern Class . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
4.4 Local Gauge and Transition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
4.4.1 Local Gauge 1-form via Trivialization . . . . . . . . . . . . . . . . . . . . 123
4.4.2 Connection via Transition Functions . . . . . . . . . . . . . . . . . . . . . 126
4.5 Gauge Transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
4.5.1 Gauge Transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
4.5.2 Local v.s. Global Gauge Transformations . . . . . . . . . . . . . . . . . . 132
4.6 Maxwell Theory as U (1)-gauge Theory . . . . . . . . . . . . . . . . . . . . . . . . 134
4.6.1 Potential as Gauge 1-form . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
4.6.2 Electromagnetic Field as Curvature . . . . . . . . . . . . . . . . . . . . . 134
4.6.3 Maxwell Action . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
4.6.4 Gauge Principle and Charge Conservation . . . . . . . . . . . . . . . . . . 136
4.6.5 Magnetic Monopole . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
4.7 Associated Bundle and Matter Field . . . . . . . . . . . . . . . . . . . . . . . . . 138
4.7.1 Associated Vector Bundle . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
4.7.2 Hermitian Inner Product . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
4.7.3 Covariant Derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
4.7.4 Matter Wave Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141

Chapter 5 Electromagnetism and Special Relativity 143

5.1 Lorentz Transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
5.1.1 Lorentz Group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
5.1.2 Transformation of Tensor Fields . . . . . . . . . . . . . . . . . . . . . . . 145
5.1.3 Invariance of Inner Contraction . . . . . . . . . . . . . . . . . . . . . . . . 148
5.2 Lorentz Invariance of Maxwell’s Equations . . . . . . . . . . . . . . . . . . . . . . 148
5.2.1 Transformation of Electromagnetic Fields . . . . . . . . . . . . . . . . . . 148
5.2.2 Transformation of Charge-Current Density . . . . . . . . . . . . . . . . . 150
5.2.3 Transformation of Maxwell’s Equations . . . . . . . . . . . . . . . . . . . 150
5.3 Relativistic Lorentz Force Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
5.3.1 Non-relativistic Charged Particle . . . . . . . . . . . . . . . . . . . . . . . 153
5.3.2 Relativistic Charged Particle . . . . . . . . . . . . . . . . . . . . . . . . . 153

Bibliography 157

4
 Preface

In April 2021, Qiuzhen College (求真书院) was newly established at Tsinghua University
under the leadership of Professor Shing-Tung Yau. It homes the distinguished elite mathematics
program in China starting in 2021: the “Yau Mathematical Sciences Leaders Program” (丘成桐
数学科学领军人才培养计划). This program puts strong emphasis on basic sciences related to
mathematics in a broad sense. Though majored in mathematics, students in this program are re-
quired to study fundamental theoretical physics such as classical mechanics, electromagnetism,
quantum mechanics, and statistical mechanics, in order to understand global perspectives of
theoretical sciences. It is an exciting challenge both for students and for instructors.

This preliminary note is written for the course “Electrodynamics” that I lectured at Qi-
uzhen College in the spring semester of 2023. The lecture note consists of two parts. The first
part is to explain key physics ingredients of electromagnetism, such as Maxwell’s equations,
electrostatics, magnetostatics, electromagnetic waves, radiation, scattering, etc. The second
part is of geometric nature, which explains Maxwell theory as U(1)-gauge theory in terms of
fiber bundle theory, as well as its consistency with special relativity. We emphasize on different
faces of concrete examples in order to understand the bridge between physics and mathematics.
This note is in succession to [10] in this series, and assumes basic knowledge on differential
forms and vector fields. Readers can consult [10] for preliminary geometric backgrounds.

I greatly appreciate the “Notes Taker Program” at Qiuzhen College, which has triggered
and supported the production of this note. A preliminary version of this note was carefully
typed by Yang Peng (杨鹏) all the way along the course, and I am extremely grateful to his
great job for Notes Taker. I would also like to thank Zhou Jiawei (周嘉伟)、Dingxu Zhihan (丁
徐祉晗)、Liu Jiuhe (刘九和) and Quan Hanwen (权瀚文) for their help on proofreading of the
first version of this note.

@ Jingzhai (静斋)
June, 2023

5
This is a drawing by my daughter expressing herself with her brother. I found it interest-
ing as it illustrates precisely the essence of electromagnetism on the coupling of electric
and magnetic fields as well as the topological nature of Maxwell’s equations.

6
 Chapter 1 Introduction: Maxwell’s
Equations

In the early 1860’s, James Clerk Maxwell took the work of Faraday and many others, and
summarized into four equations that linked the electric field with the magnetic field. Maxwell’s
equations are nowadays accepted as the basis of all modern theories of electromagnetism.


∇·E ~ = ρ/ε0





∇ · B
~ =0

∇×E
⃗
~ = − ∂B




∂t
 

∇ × B ⃗
~ = µ0 ~j + ε0 ∂ E
∂t
These equations predicted electromagnetic waves travelling at the observed speed of light. This
leads to Maxwell’s speculation that lights are electromagnetic waves, and suddenly brings light,
electricity and magnetism into the same fundamental phenomenon.
Maxwell’s equations in modern geometry take the concise form

dF = 0
d∗ F = J

Here F is a 2-form on spacetime that collects both electric and magnetic fields. J is a 1-form
that represents the electric charge-current. d is the de Rham differential, and d∗ is its adjoint.
This immediately reveals many deep geometric and topological natures of Maxwell theory.
In this chapter, we will review and explain the precise meaning of these equations, preparing
for the journey toward the study of physics and geometry of electromagnetism. The beauty of
this set of equations lies in its the simplicity and nontriviality. For example, it encodes all
together the following fundamental electromagnetic phenomenons

stationary charges =⇒ electric fields

steady moving charges =⇒ magnetic fields
accelerating charges =⇒ electromagnetic wave

These will be discussed and explored in detail in Chapter 2 and Chapter 3. Geometric and
topological aspects of Maxwell’s equations will be explained in Chapter 4 and Chapter 5.
Our flavor will be geometric, and assume basic knowledge on differential forms. Readers
can consult [10, Chapter 3] in this series for preliminary geometric backgrounds.

7
 1.1 Hodge Star

Hodge Star on R3

Let us consider the geometry of R3 with space coordinate (x, y, z). Denote by

∂ ∂ ∂
d3 = dx + dy + dz
∂x ∂y ∂z

for the de Rham differential on R3 for space variables. The Euclidean metric ds2 = dx2 + dy 2 +
dz 2 defines a Hodge star operator on differential forms, which we denote by

∗3 : Ωp (R3 ) −→ Ω3−p (R3 ).

Here Ωp (R3 ) is the space of smooth p-forms on R3 . Explicitly, ∗3 applies to a basis as

∗
1 7−→
3
dx ∧ dy ∧ dz
∗
dx 7−→
3
dy ∧ dz
∗
dy 7−→
3
dz ∧ dx
∗3
dz 7−→ dx ∧ dy
∗3
dx ∧ dy 7−→ dz
∗3
dy ∧ dz 7−→ dx
∗
dz ∧ dx 7−→
3
dy
∗
dx ∧ dy ∧ dz 7−→
3
1

Then ∗3 is defined on all Ω• (R3 ) by C ∞ (R3 )-linear extension over the above basis. For example,

∗3 (f dx + gdy ∧ dz) = f dy ∧ dz + gdx

where f, g ∈ C ∞ (R3 ) are functions on R3 . The above ∗3 has the property that

(∗3 )2 = 1 on all Ω• (R3 ).

Definition 1.1.1. We define the adjoint operator d∗3 of d3 in R3 by

d∗3 = (−1)p ∗3 d3 ∗3 : Ωp (R3 ) −→ Ωp−1 (R3 ).

Let ∇2 be the Laplacian operator defined by

∇2 f := ∂x2 + ∂y2 + ∂z2 f.

∇2 can be extended to Ωp (R3 ) component-wise with respect to the basis as above. For example,

∇2 (f dx + gdy ∧ dz) = (∇2 f )dx + (∇2 g)dy ∧ dz.

We will still denote it by

∇2 : Ω• (R3 ) → Ω• (R3 ).

8
Proposition 1.1.2. d3 d∗3 + d∗3 d3 is related to the Laplacian operator ∇2 by

d3 d∗3 + d∗3 d3 = −∇2 .

Proof: This is a good exercise.

As an example for illustration, let f be a function on R3 . Since f is a 0-form, d∗3 f = 0.

Then
(d3 d∗3 + d∗3 d3 )f = d∗3 d3 f = − ∗3 d3 ∗3 (∂x f dx + ∂y f dy + ∂z f dz)
= − ∗3 d3 (∂x f dy ∧ dz + ∂y f dz ∧ dx + ∂z f dx ∧ dy)
= − ∗3 (∂x2 f + ∂y2 f + ∂z2 f )dx ∧ dy ∧ dz = −(∂x2 + ∂y2 + ∂z2 )f.

Integral in R3

Differential p-forms can be integrated on p-dim oriented spaces. Besides, we will also use
the following notations for various integrals in R3 in this note:

• Volume integral of a function f on a 3-dim region V ⊂ R3 :

ˆ
d3 rf (~r), ~r = (x, y, z)
V

Here d3 r = dxdydz is the standard volume form on R3 .

~ on a 2-dim surface S ⊂ R3 :
• Surface integral of a vector A
ˆ
d~σ · A
~
S

Here d~σ is the vector surface element on S, and

d~σ · A
~ = dσx Ax + dσy Ay + dσz Az = dydzAx + dzdxAy + dxdyAz .

~ on a 1-dim curve C ⊂ R3 :
• Line integral of a vector A
ˆ
d~r · A
~
C

Here d~r is the vector line element on C, and

d~r · A
~ = dxAx + dyAy + dzAz .

´
• We will sometimes write Y f for the integral of a function f over some space Y ⊂ R3 ,
without explicitly expressing the measure. The measure is understood as the canonically
induced one from R3 . When Y is 3-dim, this is the volume integral; when Y is 2-dim, this
is the area integral; when Y is 1-dim, this is the curve integral.

Example 1.1.3. Let S be a surface in R3 and α = fx dx + fy dy + fz dz ∈ Ω1 (R3 ). We associate

~ = (fx , fy , fz ) by collecting components of α. Then the usual surface integral of
a vector field A
~ with respect to the vector surface element d~σ can be expressed via Hodge star as
A
ˆ ˆ
d~σ · A =
~ ∗3 α.
S S

9
Hodge Star on R3,1

Consider now the Minkowski space R3,1 with space-time coordinate (x, y, z, t). We will
simply denote by
∂ ∂ ∂ ∂
d = dx + dy + dz + dt
∂x ∂y ∂z ∂t
for the de Rham differential on R3,1 . The Minkowski metric ds2 = c2 dt2 − dx2 − dy 2 − dz 2
induces a Hodge star operator

∗ : Ωp (R3,1 ) −→ Ω4−p (R3,1 ).

Here c is the speed of light. Explicitly, we have

∗
1 7−→ c dt ∧ dx ∧ dy ∧ dz
∗
c dt 7−→ dx ∧ dy ∧ dz
∗
dx 7−→ c dt ∧ dy ∧ dz
∗
dy 7−→ c dt ∧ dz ∧ dx
∗
dz 7−→ c dt ∧ dx ∧ dy
∗
c dt ∧ dx 7−→ −dy ∧ dz
∗
c dt ∧ dy 7−→ −dz ∧ dx
∗
c dt ∧ dz 7−→ −dx ∧ dy
∗
dx ∧ dy 7−→ c dt ∧ dz
∗
dy ∧ dz 7−→ c dt ∧ dx
∗
dz ∧ dx 7−→ c dt ∧ dy
∗
dx ∧ dy ∧ dz 7−→ c dt
∗
c dt ∧ dx ∧ dy 7−→ dz
∗
c dt ∧ dy ∧ dz 7−→ dx
∗
c dt ∧ dz ∧ dx 7−→ dy
∗
c dt ∧ dx ∧ dy ∧ dz 7−→ −1

It is direct to check that

∗2 = (−1)p+1 on Ωp (R3,1 ).

Definition 1.1.4. We define the adjoint operator d∗ of d on R3,1 by

d∗ = ∗d∗ : Ωp (R3,1 ) −→ Ωp−1 (R3,1 ).

Remark 1.1.5. The sign in defining d∗ via ∗ on R3,1 is different from that on R3 , due to dimension
and signature reason.

Proposition 1.1.6. In the Minkowski spacetime R3,1 , we have

dd∗ + d∗ d = −□,

10
where □ is the d’Alembert operator

1 ∂2 ∂2 ∂2 ∂2 1 ∂2
□= − − − = − ∇2 .
c2 ∂t2 ∂x2 ∂y 2 ∂z 2 c2 ∂t2

Here □ is defined on Ωp (R3 ) component-wise with respect to the basis as above, similar to ∇2 .

Proof: Exercise.

Remark 1.1.7. Here is a useful formula relating geometric operators in R3 and in R3,1 . Let α
be a p-form in R3,1 containing only form indices dx, dy, dz, and |α| = p is the form degree of α.
Then 

∗α = cdt ∧ ∗3 α





∗(α ∧ cdt) = ∗3 α

d∗ α = −d∗3 α





d∗ (α ∧ cdt) = −d∗ α ∧ cdt − (−1)|α| 1 ∂t α
3 c

1.2 Maxwell’s Equations

The modern form of Maxwell’s equations is


 ∇·E~ = ρ/ε0





∇ · B~ =0

∇×E
⃗
~ = − ∂B




∂t
 

∇ × B ⃗
~ = µ0 ~j + ε0 ∂ E
∂t

This set of equations completely describes the dynamics and interactions of electromagnetic
fields. Here

~ = (Ex , Ey , Ez ) is the electric field

E
~ = (Bx , By , Bz ) is the magnetic field
B
ρ is the electric charge density
~j = (jx , jy , jz ) is the electric current density

These quantities in general depend on the space (x, y, z) and time t. The other constants in
Maxwell’s equations are: ε0 the permittivity of free space and µ0 the permeability of free space.
They are related to the speed of light c by the following equation
1
c= √ .
ε0 µ 0

~ and
A particle of charge q moving with velocity ~v in the background of electric field E
~ experiences force
magnetic field B

~ + ~v × B).
F~ = q(E ~

11
This is the Lorentz force law. We can talk about force density f~ per volume. Then the Lorentz
force law becomes
~ + ~j × B.
f = ρE ~

Maxwell’s equations, together with the Lorentz force law, form the foundation of classical elec-
tromagnetism.
Maxwell’s equations have a compact form in geometric terms. We collect electric and
magnetic field into a 2-form F on the space-time R3,1 as

F = (Ex dx + Ey dy + Ez dz) ∧ dt + Bx dy ∧ dz + By dz ∧ dx + Bz dx ∧ dy.

Taking the de Rham differential, we find

dF = dt ∧ [(∂x Ey − ∂y Ex )dx ∧ dy + (∂y Ez − ∂z Ey )dy ∧ dz + (∂z Ex − ∂x Ez )dz ∧ dx]

+ dt ∧ (∂t Bx dy ∧ dz + ∂t By dz ∧ dx + ∂t Bz dx ∧ dy) + (∂x Bx + ∂y By + ∂z Bz )dx ∧ dy ∧ dz.

We observe that two of Maxwell’s equations can be equivalently described as


∇ · B~ =0
⇐⇒ dF = 0.
∇ × E ⃗
~ = − ∂B
∂t

In other words, F is a closed 2-form.

Let us consider the Hodge dual 2-form
1
∗F = (Ex dy ∧ dz + Ey dz ∧ dx + Ez dx ∧ dy) − c(Bx dx + By dy + Bz dz) ∧ dt.
c
We see that ∗ switches electric and magnetic field

∗:E
~ 7−→ −cB ~

~ 7−→ 1 E
∗:B ~
c
It is computed
1
∗d ∗ F = (∂t Ex dx + ∂t Ey dy + ∂t Ez dz) + (∂x Ex + ∂y Ey + ∂z Ez )dt
c2
− (∂x By − ∂y Bx )dz − (∂y Bz − ∂z By )dx − (∂z Bx − ∂x Bz )dy.

Let us define the current 1-form

J = ρ/ε0 dt − µ0 (jx dx + jy dy + jz dz).

Then the other two Maxwell’s equations becomes


∇ · E
~ = ρ/ε0
  ⇐⇒ d∗ F = J.
∇ × B ⃗
~ = µ0 ~j + ε0 ∂ E
∂t

Here d∗ = ∗d∗ is the adjoint of d. We have arrived at the geometric form of Maxwell’s equations

dF = 0
d∗ F = J

12
 1.3 Conservation Law of Charge

Electric Charge

The words “electric” and “electricity” come from the Greek word for “amber”. Electric
charge is an intrinsic property of matter, and all take value in an integral1 multiple of the
elementary charge
e = 1.602176634 × 10−19 C.

A single proton carries electric charge e and a single electron carries electric charge −e.
Since e is practically small, it is natural to consider continuous objects and define “charge
density” ρ(~r, t) per unit volume. Here

~r = (x, y, z)

refers to the space coordinate. The total charge Q in a finite volume V is

ˆ
Q= d3 r ρ
V

where d3 r = dxdydz is the standard volume form on the space.

When dealing with point charges qα located at position ~rα , we can use δ-function and
represent the charge density as
X
ρ= qα δ(~r − ~rα ).
α

qα

There is a similar treatment for line charges (using δ-function supported on the line)

or surface charges (using δ-function supported on the surface)

1
Here we do not consider quarks or quasi-particles.

13
Electric Current

The movement of electric charges constitutes the “electric current”. Such quantity is cap-
tured by a space-vector ~j called current density. In general ~j = ~j(~r, t) depends on the position
and may change with time. For any surface S, the surface integral
ˆ
I= d~σ · ~j
S
counts the charge per unit time passing through S.

~j

If we consider a charge distribution ρ in which the velocity of a small volume at point ~r is

~v = ~v (~r, t), then the current density is
~j = ρ~v .

For example, the current density of a point particle of charge q moving at position ~rα (t) is

~j = q ~r˙α (t) δ(~r − ~rα (t)).

Conservation Law of Charge

Electric charge is conserved in physical processes. Consider the total charge Q contained
in some fixed region V ˆ
Q= d3 r ρ.
V
Its change with time is given by ˆ
dQ ∂ρ
= d3 r .
dt V ∂t
On the other hand, we can compute its change by counting the flow out through its boundary
S = ∂V ˆ ˆ
dQ Gauss s′
=− d~σ · ~j ======= − d3 r ∇ · ~j.
dt S Theorem V

~j
V

14
By comparing the two expressions, we find
ˆ  
∂ρ
3
d r + ∇ · j = 0.
~
V ∂t

This holds for any region V , leading to the following local form of conservation law of charge
∂ρ
+ ∇ · ~j = 0
∂t
This equation is also a consistency equation for Maxwell’s equations. To see this, recall

d∗ F = J

where  
1 1
J= ρdt − (jx dx + jy dy + jz dz) .
ε0 c2
The equation d∗ F = J is equivalent to

d(∗F ) = ∗J.

Since d2 = 0, the consistency of this equation requires

d(∗J) = 0.

Explicitly, we have
1
∗J = (ρdx ∧ dy ∧ dz − jx dt ∧ dy ∧ dz − jy dt ∧ dz ∧ dx − jz dt ∧ dx ∧ dy),
cε0
 
1 ∂ρ
d(∗J) = + ∇ · j dt ∧ dx ∧ dy ∧ dz.
~
cε0 ∂t
Therefore
∂ρ
d(∗J) = 0 ⇐⇒ + ∇ · ~j = 0.
∂t
The conservation law can be also understood via Stokes’ Theorem as follows. Consider the
following region
M = V × [t1 , t2 ] in the spacetime R3,1 .

t t2

V t1

15
The boundary ∂M of M consists of three pieces

V × {t2 }, V × {t1 }, S × [t1 , t2 ].

Applying Stokes’ Theorem,

ˆ ˆ ˆ ˆ ˆ t2 ˆ
0 = cε0 d(∗J) = cε0 ∗J = d rρ(~r, t2 ) −
3 3
d rρ(~r, t1 ) + dt d~σ · ~j.
M ∂M V V t1 S

Taking the infinitesimal form t2 → t1 (or simply taking the derivative of t2 ), we find
ˆ ˆ
∂ρ
d3 r + d~σ · ~j = 0.
V ∂t S

In Section 4.6.4, we will discuss another interpretation of charge conservation from the
point of view of gauge principle.

1.4 Coulomb’s Law

In the early 1770s, Henry Cavendish discovered the dependence of the force between charged
bodies in an unpublished note. It was later published and fully established by Charles-Augustin
de Coulomb and now coined the name “Coulomb’s Law”: the magnitude of the electric force
between two point charges is proportional to the product of the charges and inverse proportional
to the square of their distance. Precisely, if we place two point particles of charge q and q ′ at
position ~r and ~r ′ , then the particle q will experience a force by
1 ~r − ~r ′
F~ = qq ′ .
4πε0 |~r − ~r ′ |3

~r − ~r ′

q′ q

In particular, it is repulsive/attractive if q and q ′ have the same/opposite signs. If we place

several particles q1′ , q2′ , · · · , then the principle of superposition applies and we get

1 X ′ ~r − ~rα′
F~ = qqα .
4πε0 α |~r − ~rα′ |3

This formula naturally generalizes to the case with electric charge distribution ρ
ˆ
1 ~r − ~r ′
F~ = q d3 r′ ρ(~r ′ ) .
4πε0 |~r − ~r ′ |3
We can recast this formula into the form

F~ = q E
~

~ is the electric field produced by the charge distribution

where the vector E
ˆ
1 ~r − ~r ′
~
E(~r ) = d3 r′ ρ(~r ′ ) .
4πε0 |~r − ~r ′ |3

16
 Here now comes the important idea and concept of field. The Coulomb’s law in the form
F~ = q E
~ is a conceptual shift of picture on the nature of force. A particle experiences a force
determined by the local value of the field at the position of the particle. This is different
from the non-local force over large distance. In particular, fields are physical, and can exist
independently of the presence of charged particles.
Now we look into the expression
ˆ ˆ
1 ′ ~ r − ~r ′ 1 1
~
E(~r ) = 3 ′
d r ρ(~r ) ′ 3 =− d3 r′ ρ(~r ′ )∇ .
4πε0 |~r − ~r | 4πε0 |~r − ~r ′ |
 
Here ∇ = ∂x∂ ∂
, ∂y ∂
, ∂z is the gradient operator with respect to the position ~r. Observe that

|⃗
1
r−⃗r ′|
is the Green’s function on R3 , or precisely
1
∇2 = −4πδ(~r − ~r ′ ).
|~r − ~r ′ |
Here ∇2 = ∂x2 + ∂y2 + ∂z2 is the Laplacian operator on R3 . It follows that
ˆ ˆ
~ r) = − 1
∇ · E(~ d3 r′ ρ(~r ′ )∇2
1
=
1
d3 r′ ρ(~r ′ )δ(~r − ~r ′ ) = ρ(~r )/ε0 ,
4πε0 |~r − ~r ′ | ε0
i.e.
∇·E
~ = ρ/ε0 .

We have found that Coulomb’s law gives one of the four Maxwell’s equations.

Gauss’s Law

The equation ∇ · E
~ = ρ/ε0 has the interpretation that the electron field is sourced by the
electric charge. To illustrate this, consider some region V with boundary ∂V = S. We consider
the electric flux through S defined by the surface integral
ˆ
d~σ · E.
~
S

~
E

V S

By Gauss’s Theorem, this is equal to

ˆ ˆ
1 QV
d r∇·E =
3 ~ d3 r ρ =
V ε0 V ε0
´ 3
where QV = V d r ρ is the total charge inside V . This is Gauss’s law. In particular, let us
assume the electric charge is supported in some region N , and consider a surface S surrounding
the entire N as pictured.

17
 V

N S

In other words, N ⊂ V , S = ∂V . Then the surface flux

ˆ ˆ
1
d~σ · E =
~ d3 r ρ
S ε0 N
is always the same: it does not matter what shape the surface S takes, as long as it surrounds
the entire N ˆ ˆ
d~σ · E
~ = d~σ · E.
~
S S′

N S S′

Geometrically, we have
d(∗F ) = ∗J.

Away from the region with electric charge and current,

d(∗F ) = 0.

Therefore Stokes’ Theorem tells that inside those regions

ˆ ˆ ˆ
1 1
∗F = Ex dy ∧ dz + Ey dz ∧ dx + Ez dx ∧ dy = d~σ · E
~
S c S c S
is invariant under continuous deformations of the surface S.

1.5 Ampère’s Law

As we have seen,
charge ρ =⇒ ~
electric field E

18
 In 1800, Alessandro Volta invented the Voltaic pile which produces a steady electric cur-
rent. This can be viewed as an early electric battery, and enabled a rapid many discoveries
in chemistry and electromagnetism. The question of a possible interaction between electricity
and magnetism had arisen soon after the invention of Volta’s pile. In 1820, Hans Christian
Ørsted discovered that a compass needle was deflated from magnetic north by a nearby electric
current, confirming the first connection between electricity and magnetism. This phenomenon
can be summarized as

current j =⇒ ~
magnetic field B

Soon after Ørsted’s discovery, André-Marie Ampère found that two parallel wires carrying
electric currents attract or repel each other. Such mutual action was formulated in mathematics
and led to the important principal: Ampère’s law. In 1827, Ampère published his book Memoir
on the Mathematical Theory of Electrodynamic Phenomena, Uniquely Deduced from Experience,
which coined the name “electrodynamics”.
Let us first consider steady solution of Maxwell’s equation, i.e. all things are independent
of time t. The equations governing the magnetic fields in the steady case are

∇ × B~ = µ0 ~j Ampère’s law
∇ · B
~ =0 Gauss’s law

∂ρ
In the steady situation, ∂t = 0 and hence the charge conservation law becomes

∇ · ~j = 0.

This is compatible with Ampère’s Law since

 
∇· ∇×B
~ =0 ~
for any vector B.

We will next give a geometric description of the above two equations and their solutions.

1.6 Biot-Savart Law

In 1820 after Ørsted’s discovery, Jean-Baptiste Biot and Félix Savart discovered an equation
describing the magnetic field generated by a constant electric current. This is now called Biot-
Savart law, which is a fundamental law in magnetostatics.
Let us come back to the geometry of the following Maxwell’s equation for magnetic fields
in the steady situation (magnetostatics)

∇ × B~ = µ0 ~j
∇ · B
~ =0

19
 ~ = (Bx , By , Bz ) and ~j = (jx , jy , jz ) as 1-forms
~ = (Ex , Ey , Ez ), B
Let us recollect the vectors E
on R3 

E = Ex dx + Ey dy + Ez dz


B = Bx dx + By dy + Bz dz




j = jx dx + jy dy + jz dz

~ corresponds to the 1-form ∗3 d3 B and

Then it is easy to see that the vector ∇ × B

~ = −d∗ B.
∇·B 3

Therefore we have the following geometric form (using (∗3 )2 = 1)

 
∇ × B ~ = µ0 ~j d3 B = µ0 ∗3 j
⇐⇒
∇ · B ~ =0 d∗ B = 0
3

Here recall d3 deontes the de Rham differential on R3 . The consistency condition d23 B = 0 asks
for d3 ∗3 j = 0, or equivalently ∇ · ~j = 0. This is the steady charge conservation.
From d∗3 B = 0 or equivalently d3 ∗3 B = 0, we can write ∗3 B = d3 A for a 1-form

A = Ax dx + Ay dy + Az dz.

~ = (Ax , Ay , Az ) is called the “vector potential”, and will be playing an impor-

Such a vector A
tant role in electromagnetism. In vector notation, this is equivalent to the equation

~
~ = ∇ × A.
B

The vector potential is not uniquely specified. Indeed, for any function χ, the shift

A 7−→ A + d3 χ = A′

also satisfies
d3 A′ = d3 (A + d3 χ) = d3 A = ∗3 B

since d23 = 0. Such a change of A is called a gauge transformation, which we will explain in
detail in later part of the note.
In turns out that there is a condition to fix this freedom of choice, by asking A to satisfy
the “Coulomb gauge” condition

d∗3 A = 0, or in vector notation, ~ = 0.

∇·A

Let us now assume A satisfies the Coulomb gauge condition. Let us substitute A into

d3 (∗3 d3 A) = µ0 ∗3 j.

This equation is the same as

d∗3 d3 A = µ0 j.

20
Coulomb gauge condition implies
d3 d∗3 A = 0.

Combining the above two and using d3 d∗3 + d∗3 d3 = −∇2 , we find

∇2 A = −µ0 j.

This can be solved using Green’s function

ˆ ~j(~r ′ )
~ r ) = µ0
A(~ d3 r′ .
4π |~r − ~r ′ |

~ r ) indeed satisfies the Coulomb gauge condition using ∇ · ~j = 0

It can be checked that A(~
~ in the presence of steady current ~j by
(Exercise). Now we can write down the magnetic field B
ˆ  
~ r ) = µ0
~ r ) = ∇ × A(~
B(~ 3 ′
d r ∇
1
× ~j(~r ′ ),
4π |~r − ~r ′ |

i.e. ˆ
~ r ) = µ0 ~r − ~r ′
B(~ d3 r′ ~j(~r ′ ) × .
4π |~r − ~r ′ |3
This formula is known as Biot-Savart Law.

1.7 Gauss’s Linking Formula

Consider two closed, non-intersecting curves C1 and C2 in R3 . There is an integer n(C1 , C2 ),
called the linking number of C1 and C2 , which describes how many times one of the curve
winds around the other.

C1 C2 C1 C2

n(C1 , C2 ) = 0 n(C1 , C2 ) = ±1

The sign depends on the orientation

of C1 and C2 .

One way to define the linking number in the current situation is as follows. We first fill C1
by a disk D1 whose boundary is precisely ∂D1 = C1 . Then the linking number

n(C1 , C2 ) = #(D1 ∩ C2 )

counts (with sign) the number of intersection of D1 with C2 .

21
 C2 n(C1 , C2 ) = ±1
C1

You can similarly fill C2 first by a disk and intersect with C1 : the answer is the same.
There is an integral expression for the linking number due to Gauss, which is closely related
to ingredients of electromagnetism. Suppose C2 carries a steady current I. It creates a magnetic
~ in space, which can be computed via Biot-Savart formula
field B
˛
~ r1 ) = µ0 I ~r1 − ~r2
B(~ d~r2 × .
4π C2 |~r1 − ~r2 |3

D1 I
C1 C2

Let us consider the circle integral of B~ along C1

˛ ˛ ˛
µ0 I ~r1 − ~r2
d~r1 · B(~r1 ) =
~ d~r1 · d~r2 × .
C1 4π C1 C2 |~r1 − ~r2 |3

On the other hand, by Stokes’ Theorem and Ampère’s Law

˛ ˆ ˆ
d~r1 · B(~r1 ) =
~ d~σ · (∇ × B) = µ0
~ d~σ · ~j = µ0 I n(C1 , C2 ).
C1 D1 D1

Comparing the above two expressions, we find

˛ ˛ ˛ ˛
1 ~r1 − ~r2 1 (~r1 − ~r2 ) · (d~r1 × d~r2 )
n(C1 , C2 ) = d~r1 · d~r2 × = · .
4π C1 C2 |~r1 − ~r2 |3 4π C1 C2 |~r1 − ~r2 |3

This is precisely the linking number formula of Gauss.

1.8 Faraday’s Law

In 1831, Michael Faraday discovered electromagnetic induction, which shows that the
change of magnetic field produces electric field. This is now called Faraday’s Law of In-
duction. Faraday explained electromagnetic induction using a concept called lines of force,
and essentially proposed the concept of electromagnetic field.
Faraday’s law says

change of magnetic field =⇒ electric field

22
Precisely, this is described by the third Maxwell’s equation
~
∂B
∇×E
~ =− .
∂t
In terms of differential forms as described in Section 1.6, this is written as
∂B
∗3 d3 E = −
∂t
or equivalently
∂
d3 E = − (∗3 B).
∂t
Let us consider a surface S which is bounded by a closed curve C.

~
B

S C

The integral of the magnetic field over the surface S is called the “magnetic flux” through S
ˆ ˆ
Φ= d~σ · B =
~ ∗3 B.
S S

Then we have the integral form of Faraday’s law by

ˆ ˆ   ˆ ˆ
d ∂
− ∗3 B = − ∗3 B = d3 E = E,
dt S S ∂t S C

i.e. ˆ ˆ
d
− d~σ · B =
~ d~r · E.
~
dt S C
This is the form when Faraday discovered the law of induction: changing the magnetic flux
through S will produce a current to flow along C.

1.9 Ampère-Maxwell Law

In Magnerostatics for the steady situation, we have Ampère’s law on the magnetic field
arising from electric current
∇×B
~ = µ0~j.

However, in the dynamical case when all things depend on time, this is not enough. Faraday’s
law says that the change of magnetic field will induce electric field. Parallelly, it is natural to
ask whether the change of electric field will induce magnetic field. This is Maxwell’s addition
to Ampère’s Law, usually called Ampère-Maxwell Law
!
~
~ = µ0 ~j + ε0 ∂ E
∇×B ,
∂t

23
 1
or using the relation µ0 ε0 = c2
,
~
1 ∂E
∇×B
~ = µ0~j +
2
.
c ∂t
The extra term is called “displacement current”, though it is not really a current but
an addition to the current. In terms of differential forms, Ampère-Maxwell Law is
1 ∂
d3 B = µ0 ∗3 j + ∗3 E.
c2 ∂t
The displacement current is in fact necessary for the consistency with charge conservation.
In fact, using (d3 )2 = 0
 
1 ∂ 1 ∂
0 = d3 d3 B = µ0 d3 ∗3 j + 2 d3 ∗3 E = µ0 ∇ · j + 2 ∇ · E dx ∧ dy ∧ dz,
~ ~
c ∂t c ∂t

from which we get

1 ∂
µ0 ∇ · ~j + 2 ∇ · E
~ = 0.
c ∂t
This is precisely the conservation law of charge using ∇ · E
~ = ρ/ε0 .

1.10 Potential and Gauge

We have discussed the full set of Maxwell’s equations


 ∇·E ~ = ρ/ε0





∇ · B~ =0

∇×E
⃗
~ = − ∂B




∂t
 

∇ × B ⃗
~ = µ0 ~j + ε0 ∂ E
∂t

and its geometrical form 

dF = 0
d(∗F ) = ∗J

where F = E ∧ dt + ∗3 B is the electro-magnetic field and J = ρ/ε0 dt − µ0 j is the charge current.

The consistency condition
d(∗J) = 0

is the the conservation law of charge

∂ρ
+ ∇ · ~j = 0.
∂t
Consider the equation
dF = 0.

Since the topology of R3,1 is trivial, HdR

2 (R3,1 ) = 0. Hence we can always find a 1-form A on

R3,1 such that

F = dA.

24
In general when spacetime has nontrivial topology, such A can only be obtained locally and
glued via transformation law. We will discuss this situation in more detail in Section 4.6.
Let us write the 1-form A as
A = −φdt + A

where A = Ax dx + Ay dy + Az dz. Then the relation F = dA becomes

E ∧ dt + ∗3 B = −d3 φ ∧ dt + dt ∧ ∂t A + d3 A.

Comparing the form types of both sides, we find


E = −d3 φ − ∂t A
B = ∗ d A
3 3

~ = (Ax , Ay , Az ), this is
In vector notation, let A

E~ = −∇φ − ∂t A~
B~ =∇×A ~

~ is the vector potential that we have seen in Section 1.6.

φ is called the scalar potential. A
Again, the choice of A is not unique. We can always shift it by

A 7−→ A + dχ

for a function χ = χ(~r, t) on the spacetime. The electric-magnetic field remains the same

F = dA 7−→ d(A + dχ) = dA.

These are gauge transformations, and A is called the gauge field. We will study gauge
theory systematically in Chapter 4.
In components, the gauge transformation is

 ∂χ
φ 7−→ φ −
∂t

A~ 7−→ A
~ + ∇χ

In terms of the gauge field, Maxwell’s equations become a single equation

∗d ∗ dA = J.

To solve this equation, we can choose a gauge condition to fix the gauge degree of freedom.
There are usually two commonly used gauge fixing condition:

~ =0
Coulomb gauge : ∇ · A
~ + 1 ∂φ
Lorenz gauge : ∇·A =0
c2 ∂t
We have seen Coulomb gauge in Section 1.6 on the Biot-Savart Law in magnetostatics. Here
we briefly comment on Lorenz gauge which we will discuss later in details.

25
 The Lorenz gauge can be written as

Lorenz gauge : d ∗ A = 0.

In Lorenz gauge, we have 

d∗ dA = J
d∗ A = 0

So
□A = −(dd∗ + d∗ d)A = −J,

or in components, 

 1 ∂2
∇2 φ − 2 2 φ = −ρ/ε0
c ∂t


2
∇2 A~ − 1 ∂ A ~ = −µ0~j
c2 ∂t2
These are inhomogeneous wave equations that we will study in detail in Section 3.4.
In a region without charge and current, such as vacuum, Maxwell’s equations reduce to

dF = 0
d∗ F = 0

which again leads to the wave equation

□F = −(dd∗ + d∗ d)F = 0.

In components, we have 
 ~
1 ∂2E

 2 2 − ∇2 E ~ =0
c ∂t

 2~
 1 ∂ B − ∇2 B~ =0
c2 ∂t2
which are the standard form of wave equations traveling at the speed of light c. This led
Maxwell to propose that light and radio waves were propagating electro-magnetic waves. In
1887, Heinrich Hertz demonstrated Maxwell’s electromagnetic waves propagating at the same
speed as light. This placed Maxwell’s theory on a firm foundation.
All electromagnetic waves travel at a fixed speed c in the vaccum, independent of any frame
of reference. This looks controversial comparing to classical Newton mechanics. It was studied
by Hendrik Lorentz who was able to derive the Lorentz transformations that preserve the form
of Maxwell’s equations. This subsequently laid the foundation for Einstein’s special relativity.
We will study this in Chapter 5.

26
 Chapter 2 Static Electromagnetism

In this chapter, we discuss static electromagnetism (including electrostatics and magneto-

statics), in which case the electric and magnetic fields do not vary with respect to time.

2.1 Electric Field and Scalar Potential

We still start with the study of electrostatics, which describes the situation of steady charge
distribution and no currents. In electrostatics, we have

~ = ~j = 0, ∂ρ
B = 0.
∂t
The only relevant Maxwell’s equations are

∇ · E
~ = ρ/ε0
∇ × E~ =0

~ is stationary: ⃗
∂E
where the electric field E ∂t = 0.

2.1.1 Poisson’s and Laplace’s equation

In electrostatics, the potentials are given by


φ = φ(~r )
A ~ =0

In terms of scalar potential, we have

~ = −∇φ,
E

which automatically solves the equation ∇ × E

~ = 0. Then the equation ∇ · E
~ = ρ/ε0 becomes

∇2 φ = −ρ/ε0 .

This is Poisson’s equation.

In the region of space that is free of charge, we have the Laplace’s equation

∇2 φ = 0.

27
 When we consider electrostatics problem in the space without boundary, say R3 , we can
solve φ (with appropriate decay condition at infinity) in terms of the Coulomb integral
ˆ
1 ρ(~r ′ )
φ(~r ) = d3 r′ .
4πε0 |~r − ~r ′ |

In practice, we usually need to deal with a finite region V in R3 with boundary ∂V = S.

The electric field could be stimulated by charges inside or outside V , and we can measure them
on the boundary S of V . In this case we need to deal with boundary value problem

∇2 φ = −ρ/ε0 inside V
boundary condition of φ on S = ∂V

S
V

The study of this problem for Poisson’s and Laplace’s equation is the subject of “Potential
theory”. Let us first look at some examples.

2.1.2 Point Charge

Consider a point charge q located at ~r0 . The charge density is

ρ(~r ) = q δ(~r − ~r0 ).

The scalar potential is

ˆ
1 q δ(~r − ~r0 ) q
φ(~r ) = d3 r′ ′ = .
4πε0 |~r − ~r | 4πε0 |~r − ~r0 |

The electric field is

~ = −∇φ = q ~r − ~r0
E ,
4πε0 |~r − ~r0 |3
which is the familiar Coulomb’s law.

28
2.1.3 Uniform Ball

Consider a ball BR of radius R centered at origin, with uniform charge ρ0 per unit volume.
The charge density function in R3 is
ρ = ρ0 χBR .

Here for a subset A ⊂ R3 , χA refers to the characteristic function of A


1 ~r ∈ A
χA (~r ) =
0 ~r ∈
/A

ρ0 R

Outside BR : |~r | > R. The scalar potential is

ˆ
1 ρ0
φ(~r ) = d3 r′ .
4πε0 BR |~r − ~r ′ |

By spherical symmetry, φ = φ(r) is a function of the radius r = |~r| only. Then

~ = −∇φ = −∂r φ ~r .
E
r
~ without explicitly evaluating the above integral. In fact, consider
This is enough to solve φ, E
the following surface integral over the sphere Sr of radius r centered at the origin
ˆ ˆ ˆ
4πR3
d~σ · ∇φ = d r∇ φ = −
3 2
d3 r ρ0 /ε0 = −ρ0 /ε0 Vol(BR ) = − ρ0 /ε0 = −Q/ε0 ,
Sr BR BR 3

where Q = ρ0 4πR3 /3 is the total charge of the ball. On the other hand, using φ = φ(r),
ˆ
d~σ · ∇φ = ∂r φ Area(Sr ) = 4πr2 ∂r φ.
Sr

Comparing the above two expressions, we find

Q
∂r φ = − .
4πε0 r2
The electric field outside the sphere is therefore

~ = −∇φ = Q ~r
E (r > R)
4πε0 r3
which takes the same form as that of a point charge Q.
Inside BR : |~r | < R.

29
 We follow the same strategy via spherical symmetry: φ = φ(r). Then
ˆ ˆ
r3
d~σ · ∇φ = d3 r ∇2 φ = −ρ0 /ε0 Vol(Br ) = − 3 Q/ε0 ,
Sr Br R

and ˆ
~ · ∇φ = 4πr2 ∂r φ.
dS
Sr

Comparing these two expressions, we find

Q r
∂r φ = − .
4πε0 R3
The electric field inside the ball is therefore

~ = −∇φ = Q ~r
E (r < R)
4πε0 R3
which grows linearly inside the ball. We can also describe a continuous solution of φ by
  
 Q 3 r2

 − r<R
4πε0 2R 2R3
φ(r) =


 Q 1 r>R
4πε0 r

φ E

R r R r

2.1.4 Line Charges

We consider a line segment of length 2L with uniform charge λ per unit length, placed
along the z-axis and centered at the origin.
z
y
L

−L

30
It would be convenient to work with cylindrical coordinates


 x = ρ cos θ


y = ρ sin θ




z=z

The charge density is given by

λ δ(x)δ(y)χ|z|≤L .

Due to cylindrical symmetry, the scalar potential is a function of (ρ, z) only and given by
ˆ L "p #
λ 1 λ (z − L)2 + ρ2 − (z − L)
φ(ρ, z) = dz ′ p = ln p .
4πε0 −L ρ2 + (z − z ′ )2 4πε0 (z + L)2 + ρ2 − (z + L)

We can simplify this expression using

p
r± = (z ± L)2 + ρ2

which are the distances from the two segment endpoints to the observation point. They are
related by
2
r+ − r−
2
= 4zL.

z=L r−

0
r+

z = −L

Then  
λ r− − (z − L)
φ= ln .
4πε0 r+ − (z + L)
Observe that
  
r2 − r−
2
(r− + r+ )(r− − r+ ) 1 1
r− −(z−L) = r− +L− + = +r− +L = (r− + r+ ) + L (r− − r+ ) + 1 .
4L 4L 2 2L
Similarly,   
1 1
r+ − (z + L) = (r− + r+ ) − L (r− − r+ ) + 1 .
2 2L
It follows that  
λ r− + r+ + 2L
φ= ln
4πε0 r− + r+ − 2L
is a function of r− + r+ only. This allows us to draw the surface of equipotential, or equivalently

r− + r+ = constant

~ = −∇φ is perpendicular to the equipo-

which is the geometry of an ellipse. The electric field E
tential surface. This can be visualized by

31
 ~
E

φ = const

We consider two limit cases.

p
⃝
1 |z| >> L, r = ρ2 + z 2 >> L.
2L
!
λ 1+ r+ +r− λ 4L 2λL Q
φ= ln ' ' = .
4πε0 1− 2L
r+ +r−
4πε0 r+ + r− 4πε0 r 4πε0 r

Here Q = 2λL is the total charge. So when we observe very far away the segment, it looks like
a point charge as expected.
⃝
2 |z| << L, ρ << L.
r+ + r− 1 p p 
= (L − z)2 + ρ2 + (L + z)2 + ρ2
2L 2L
1 p p 
= (1 − z/L)2 + (ρ/L)2 + (1 + z/L)2 + (ρ/L)2
2
√
Using 1 + x = 1 + 12 x − 18 x2 + O(x3 ), the above is equal to

1 1 1
= 1 + (−2z/L + z 2 /L2 + ρ2 /L2 ) − (−2z/L)2
2 2 8
!
1 1 −3
+ 1 + (2z/L + z /L + ρ /L ) − (−2z/L) + O(L )
2 2 2 2 2
2 8
1
=1 + (ρ/L)2 + O(L−3 ).
2
Therefore
! !
λ
r+ +r−
2L +1 λ 2 + 21 (ρ/L)2 + O(L−3 )
φ= ln r+ +r− = ln 1 −3
4πε0 2L −1 4πε0 2
2 (ρ/L) + O(L )
λ λ
' ln 4 − ln(ρ/L)2
4πε0 4πε0
λ λ
=− ln ρ + ln(2L).
2πε0 2πε0
The second term is a constant, which does not contribute to the electric field. We find

~ = −∇φ ' λ ρ ~
E .
2πε0 ρ2

32
Here the vector ρ
~ has length ρ and points to the radial direction in the xy-plane. Equivalently,
we can view E~ = λ ρ⃗2 as the electric field produced by the infinite line (L → +∞) with
2πε0 ρ
uniform charge per unit length λ.

z-axis

λ
~ = λ ρ ⃗
E 2πε0 ρ2

2.2 Interface Condition

2.2.1 Infinite Plane Charge

We consider the electric field produced by surface charge. We first consider the case of
an infinite plane located at z = 0, carrying uniform charge ξ per unit area. By translation
symmetry, we know that the electric field points along the direction of z-axis, only depend on
z, and with opposite directions above and below the plane:

~ = (0, 0, E = E(z)),
E E(−z) = −E(z).

~
E

z=0

~ In fact, consider a cylinder Y as below. Let A denote

This symmetry is enough to determine E.
the area of the cap
A = Area(S± ) = Area(S0 ).

33
 ~
E

S+ z

Y
S0

S− −z

The charge density is

ρ = ξ δ(z).

Apply Gauss’s law to the cylinder

ˆ ˆ
1
d~σ · E
~ = ρ.
∂Y ε0 Y

This leads to ˆ ˆ ˆ
d~σ · E
~ − ~ = 1
d~σ · E ξ.
S+ S− ε0 S0

So
ξ ξ
E(z)A − E(−z)A = A =⇒ E(z) = for z > 0.
ε0 2ε0
Equivalently, we can write
ξ z
E(z) = .
2ε0 |z|
Up to a constant, the scalar potential is
ξ
φ = φ(z) = − |z|.
2ε0
~ is NOT continuous across the
There is an important observation here: the electric field E
charge plane
ξ
E(z → 0+ ) − E(z → 0− ) = .
ε0
On the other hand, the scalar potential φ is continuous, though its derivative is not. (φ can
also jump in general, such as in Dipole Layer.)
We discuss some generalization of infinite plane charge. Consider a pair of infinite planes
P+ , P− placed at
P± : z = ±l.

P± carries uniform charge ∓ξ per unit area.

34
 ~ =0
E

−ξ
z=l P+

~
E

+ξ
z = −l P−

~ =0
E

By the result from the infinite plane and the superposition law, we find

0 |z| > l
~ =
E
(0, 0, ξ/ε ) |z| < l
0

As another example, we can consider a bit more realistic model of the infinite plane by an
infinite slab of thickness 2l with charge density per unit volume ρ, placed at

|z| ≤ l.

z=l

z = −l

We can think about the effect of each slice of infinite plane and use superposition to sum/inte-
~ = (0, 0, E(z)). When |z| > L, the total effect is the same as an infinite
grate them up. Let E
plane charge with ξ = 2lρ. Therefore



ρl
 z>l
ε0
E(z) = .
 ρl

− z < −l
ε0
Now we consider a point inside the slab, say at z = a where

0 < a < l.

The effect of the slab between a ≤ z ≤ l will cancel that between −l ≤ z ≤ −a.

35
 z=l

z=a

z = −a

z = −l

The slab −a ≤ z ≤ a will contribute

ρa
E(a) = .
ε0
In summary, we find
 ρl

 z>l

 ε0


E(z) = ρz −l ≤ z ≤ l .

 ε0



− ρl z < −l
ε0
In the limit l → 0, we get back to the case of infinite plane.

2.2.2 Spherical Shell

Consider a spherical shell of radius R with uniform surface charge ξ per area, centered at
the origin.

~
E

By spherical symmetry,
~ r ) = A(r) ~r
E(~

for some function A(r) of the radius r.

⃝
1 r > R. We consider the sphere Sr of radius r which surrounds the charged spherical
shell. By Gauss’s law ˆ
d~σ · E
~ = Q/ε0
Sr

36
where Q = 4πR2 ξ is the total charge on the shell. Since
ˆ
Q
d~σ · E
~ = A(r)r Area(Sr ) = A(r) 4πr3 =⇒ A(r) = ,
Sr 4πr3 ε0
i.e.
~ = Q ~r .
E
4πε0 r3
So effectively it feels the same as a point charge Q.
⃝
2 r < R. Again by Gauss’s law and spherical symmetry,
ˆ
d~σ · E
~ = 0 =⇒ A(r) = 0.
Sr

So inside the shell

~ = 0.
E
~ at the surface of the charged shell.
Note that again we have a jump of E

2.2.3 Interface Condition

Now we discuss the relation between the surface charge distribution and the discontinuity
of electric field. Consider a surface S with charge density ξ per area. Let n̂ denote the unit
normal vector on the surface. Let E ~ + denote the limit of the electric field along the n̂ side of
the surface, and E ~ − denote the limit on the other side.

~+
E
n̂

~−
E

Then the interface conditions for electric field along the surface charge distribution is

n̂ · (E
~+ −E
~ − ) = ξ/ε0 jump along normal direction
n̂ × (E~ −E~ )=0 continuous along tangent direction
+ −

We have seen several examples of such phenomenon above. To see how such interface conditions
arise, let us first consider a thin cylinder Y across the surface as shown.

~
E

S+
S0
S−

~
E

37
Apply Gauss’s law ˆ ˆ
~ = 1
d~σ · E ξ.
∂Y ε0 S0
When the cylinder is infinite thin, this leads to
ˆ ˆ
1
(n̂ · E
~ + − n̂ · E
~ −) = ξ.
S0 ε0 S0

This holds for arbitrary S0 , therefore

ξ
n̂ · (E
~+ −E
~ −) = .
ε0
To see the continuity of tangent direction, consider the boundary loop C of a rectangle R
with a length L parallel to the surface and a short length δ across the surface, as shown below.

L
δ
C
R
S

Since ∇ × E
~ = 0, the loop integral
˛ ˆ
d~r · E =
~ d~σ · (∇ × E)
~ = 0.
C R

In the limit δ → 0, this leads to the continuity of the tangent direction of E

~ along S:

n̂ × (E
~+ −E
~ − ) = 0.

There are similar interface conditions for magnetic fields across a surface with current. See
Section 2.5.2.

2.2.4 Electric Conductor

A conductor is a material that contains many free electrons that can move in the mate-
rial but not leave its surface. In “electrostatic” situation, the electrons will move to arrange
themselves to produce zero electric field inside the conductor.

Conductor
ξ

~ =0
E

~ = 0 inside.
E

38
On the other hand, its surface S could have a nontrivial charge density ξ per area due to outside
situation. For example, we can put a conductor inside two parallel plane charge as shown.

+ −

+ − + −
− +
+ − + −

+ −

Let n̂ denote the unit normal vector on S that points outward.

n̂

Conductor

Then we have inside the conductor

~ − = 0.
E
~ + is normal to S. This is easy to understand:
The continuity of tangent direction implies that E
if there is a nontrivial tangent direction, then it will further move electrons on the surface.
Now from the interface condition of the normal direction, we find on the outside surface of the
conductor
~ + = ξ n̂.
E
ε0
~ = 0 inside the conductor, the scalar potential φ is a constant inside. In the outside, we
Since E
have on the surface of the conductor
ξ
∂n̂ φ|S = − .
ε0

2.3 Boundary Value Problem

In electrostatics, we are led to consider the scalar potential φ satisfying the Poisson’s
equation
∇2 φ = −ρ/ε0

subject to certain boundary condition in a region V under consideration. There are usually two
types of boundary conditions on S = ∂V .

39
⃝
1 Dirichlet boundary condition: specify the value of φ on the boundary

φ|S = f.
f
S

⃝
2 Neumann boundary condition: specify the value of the normal derivative on the boundary

∂n̂ φ|S = g.
n̂

2.3.1 Uniqueness

We first show the uniqueness of the solution of the Poisson’s equation

∇2 φ = −ρ/ε0

inside a finite region V subject to either Dirichlet or Neumann boundary conditions.

Suppose we have two solutions, say φ1 and φ2 . Let γ = φ2 − φ1 . Then γ satisfies the
Laplace’s equation
∇2 γ = 0

and either γ|S = 0 (Dirichlet case) or ∂n̂ γ|S = 0 (Neumann case). Consider the volume integral
(we omit the volume measure d3 r on V and also the surface measure on S for simplicity)
ˆ ˆ
|∇γ| =
2
∇ · (γ∇γ) − γ∇2 γ
V ˆV ˆ
= d~σ · (γ∇γ) − γ∇2 γ
ˆS ˆ V

= γ∂n̂ γ − γ∇2 γ.
S V
Either Dirichlet or Neumann condition implies γ∂n̂ γ = 0. Together with the Laplace
equation ∇2 γ = 0, it follows that ˆ
|∇γ|2 = 0.
V
This implies ∇γ = 0, hence γ is a constant. For Dirichlet boundary condition, γ = 0 and the
solution is unique. For Neumann boundary condition, the solution is unique up to an additive
constant.

40
2.3.2 Green’s Function

The solution to the Poisson’s equation in a finite region V with either Dirichlet or Neumann
boundary condition can be obtained by means of Green’s function G(~r, ~r ′ ), which satisfies

∇2′ G(~r, ~r ′ ) = −4πδ(~r − ~r ′ ), ~r, ~r ′ ∈ V.

Here ∇2′ is the Laplacian with respect to ~r ′ . Such a Green’s function can be written as
1
G(~r, ~r ′ ) = + F (~r, ~r ′ )
|~r − ~r ′ |
where F satisfies the Laplace’s equation

∇2′ F (~r, ~r ′ ) = 0 inside V.

Consider φ(~r ) which satisfies the Poisson’s equation

∇2 φ = −ρ/ε0 inside V.

Then for ~r inside V , we have

ˆ ˆ
′ ′ 1
φ(~r ) = 3 ′
d r φ(~r )δ(~r − ~r ) = − d3 r′ φ(~r ′ )∇2′ G(~r, ~r ′ )
4π V
V ˆ ˆ
integration 1 ′ 1
======== 3 ′ ′ ′
d r ∇ φ(~r ) · ∇ G(~r, ~r ) − ′
φ(~r ′ )∂n̂′ G(~r, ~r ′ )
by part 4π V 4π S
ˆ ˆ
1 ′ 1  ′ 
=− 3 ′ 2′
d r ∇ φ(~r )G(~r, ~r ) + ′
∂n̂ φ(~r ′ )G(~r, ~r ′ ) − φ(~r ′ )∂n̂′ G(~r, ~r ′ )
4π V 4π
ˆ ˆ S
1 1  ′ 
= d3 r′ ρ(~r ′ )G(~r, ~r ′ ) + ∂n̂ φ(~r ′ )G(~r, ~r ′ ) − φ(~r ′ )∂n̂′ G(~r, ~r ′ ) .
4πε0 V 4π S
There are two types of Green’s function depending on the type of boundary conditions.
1 Dirichlet Green’s function GD (~r, ~r ′ ).
⃝

∇2′ GD (~r, ~r ′ ) = −4πδ(~r − ~r ′ )
G (~r, ~r ′ ) = 0 for ~r ′ ∈ S
D

Using GD (~r, ~r ′ ), we find

ˆ ˆ
1 ′ 1
φ(~r ) = ρ(~r )GD (~r, ~r ) − ′
φ(~r ′ )∂n̂ ′ GD (~r, ~r ′ )
4πε0 V 4π S

which gives an explicit formula of φ in terms of its boundary value φ|S . Thus this formula is
suitable for Dirichlet boundary value problem.
Remark 2.3.1. GD (~r, ~r ′ ) has the symmetry property in general

GD (~r, ~r ′ ) = GD (~r ′ , ~r ).

2 Neumann Green’s function GN (~r, ~r ′ ).

⃝

∇2′ GN (~r, ~r ′ ) = −4πδ(~r − ~r ′ )
∂ ′ G (~r, ~r ′ ) = −4π/A for ~r ′ ∈ S
n̂ N

41
Here A = Area(S) is the total area of the boundary surface S. Such choice of constant is to be
consistent with the equation ∇2′ GN (~r, ~r ′ ) = −4πδ(~r − ~r ′ ). In fact,
ˆ ˆ ˆ ´
3 ′ 2′ ′ ′ ′ 1
−4π = d r ∇ GN (~r, ~r ) = ∂n̂ GN (~r, ~r ) = (−4π)/A = −4π S
V S S A
ˆ
=⇒ A= 1 = Area(S).
S
Using GN (~r, ~r ′ ),
we find
ˆ ˆ ˆ
1 ′ 1 ′ 1
φ(~r ) = 3 ′ ′
d r ρ(~r )GN (~r, ~r ) + ′ ′
∂ φ(~r )GN (~r, ~r ) − φ(~r ′ )∂n̂ ′ GN (~r, ~r ′ )
4πε0 V 4π S n̂ 4π S
ˆ ˆ ˆ
1 1 1
= φ(~r ′ ) + d3 r′ ρ(~r ′ )GN (~r, ~r ′ ) + ∂ ′ φ(~r ′ )GN (~r, ~r ′ ).
A S 4πε0 V 4π S n̂
which gives an explicit formula of φ in terms of its boundary normal derivative ∂n̂ φ|S and the
´
boundary integral S φ(~r ′ ). Thus this formula is suitable for Neumann boundary value problem.

We can also interpret Green’s function G(~r, ~r ′ ) as an electrostatic problem as follows. Let
us consider the following potential function φ⃗r
1
φ⃗r (~r ′ ) := G(~r, ~r ′ ).
4πε0
Then it satisfies the equation
1
∇2′ φ⃗r (~r ′ ) = −
δ(~r − ~r ′ )
ε0
which represents a potential for a unit point charge located at ~r inside V . Let us write
1 1
φ⃗r (~r ′ ) = + ξ⃗r (~r ′ )
4πε0 |~r − ~r ′ |

for some ξ⃗r (~r ′ ). Then ξ⃗r (~r ′ ) satisfies the equation

∇2′ ξ⃗r (~r ′ ) = 0.

It represents the potential due to certain external charge distribution outside V chosen in such
a way that the combined potential with the internal point charge at ~r has the specific boundary
condition on ∂V .
Remark 2.3.2. Green’s function on Rn . A strategy to solve the Poisson’s equation on Rn is
to find a fundamental solution of
∇2 Φ(~r ) = −δ(~r ).

Suppose Φ = Φ(r) depends only on r = |~r |. Then

n−1 ′ n−1
∇2 Φ(r) = Φ′′ (r) + Φ (r) = 0 ⇐⇒ (ln(Φ′ (r)))′ = − .
r r
The solution of ∇2 Φ(r) = 0 on Rn − {0} is

c1 ln r + c2 n=2
Φ(r) =
c r−n+2 + c n≥3
1 2

42
where ci ’s are constants. We choose appropriate constants such that



1
− ln r n=2
2π
Φ(r) =

 1
 n≥3
n(n − 2)α(n)rn−2

π n/2
Here α(n) = is the volume of the n-dimensional unit ball, and
Γ(1 + n/2)

(n/2)! n even
Γ(1 + n/2) = √
n!!2−(n+1)/2 π n odd

Then we show ∇2 Φ = −δ(~r ) in the sense of distribution. This means for each smooth
function g : Rn → R with compact support, we have
ˆ
Φ · ∇2 g = −g(0).
Rn

Indeed,
ˆ ˆ ˆ
Φ·∇ g =
2
Φ·∇ g+2
Φ · ∇2 g
Rn B(0,ε) B(0,ε)c
ˆ ˆ ˆ ˆ
Integration
======== Φ · ∇2 g + ∇2 Φ · g − ∂⃗n Φ · g + Φ · ∂⃗n g
by part B(0,ε) B(0,ε)c ∂(B(0,ε)c ) ∂(B(0,ε)c )
| {z´ } | {z } | ´
{z } | {z
´
}
≤c||∇g||L∞ B(0,ε) rΦ =0 ∂(B(0,ε)c ) g ≤||∂⃗n g||L∞ ∂(B(0,ε)c ) Φ
= Vol(∂(B(0,ε)c ))
ε→0
==== 0 + 0 − g(0) + 0 = −g(0).

Here B(0, ε) is the ball of radius ε centered at 0, and B(0, ε)c = Rn − B(0, ε). It follows that
ˆ
1
φ(~r ) = d~r ′ Φ(~r ′ )ρ(~r − ~r ′ )
ε0 Rn
is a solution to the Poisson’s equation

∇2 φ = −ρ/ε0 .

2.3.3 Method of Image

Let us first consider an example where the region is

V = {z ≥ 0}.

We would like to figure out GD (~r, ~r ′ ). Let us put a point charge q at

~r+ = (x0 , y0 , z0 > 0)

and assume that S = {z = 0} is a conducting plane (so the scalar potential is constant on S).

43
 V
q z0

z=0

−q
−z0

We look for a scalar potential φ⃗r+ such that


 δ(~r − ~r+ )
∇2 φ⃗r+ (~r ) = −
ε0


φ⃗r+ (~r ) = 0 when ~r ∈ S
This can be solved by replacing the conductor by a fictitious “image” point charge −q at
the point ~r− = (x0 , y0 , −z0 ). Then
q q
φ⃗r+ (~r ) = −
4πε0 |~r − ~r+ | 4πε0 |~r − ~r− |
satisfies the above Dirichlet boundary value problem in V , hence is the unique solution. As a
consequence, the Dirichlet Green’s function is
1 1
GD (~r, ~r ′ ) = −
|~r − ~r ′ | |~r − σ(~r ′ )|
where σ : (x, y, z) 7→ (x, y, −z) is the mirror transformation. Using |~r − σ(~r ′ )| = |σ(~r ) − ~r ′ |, we
see explicitly the symmetry property

GD (~r, ~r ′ ) = GD (~r ′ , ~r ).

Here we can think about the mirror charge σ(~r ′ ) as an external charge outside V (the F (~r, ~r ′ )
term as discussed in Section 2.3.2).
As another example, we consider a point charge q outside a conducting sphere of radius R.
We set the potential on the conducting sphere to be zero. Finding the potential of this problem
is the same as describing the Dirichlet Green’s function on V = {|~r | ≥ R}, S = {|~r | = R}.
q ~r+

By the same image method, we can guess the result by

 
q 1 R/|~r+ |
φ⃗r+ (~r ) = − .
4πε0 |~r − ~r+ | |~r − σR (~r+ )|
R2
Here σR : ~r 7→ ~r is the mirror image of ~r with respect to the sphere of radius R.
|~r |2

44
 ~r
R

σR (~r )

It is clear that
q
∇2 φ⃗r+ (~r ) = − δ(~r − ~r+ ) inside V.
4πε0
On the other hand, let n̂ = ~r/|~r | and n̂+ = ~r+ /|~r+ | denote the unit direction of ~r , ~r+ and let
r = |~r |, r+ = |~r+ |. Then
 
q 1 R
φ⃗r+ (~r ) = − .
4πε0 |~r − ~r+ | |rr+ n̂ − R2 n̂+ |
It is clear from this expression that

φ⃗r+ (~r ) = φ⃗r (~r+ ).

It follows that
φ⃗r+ (~r )|⃗r ∈S = φ⃗r (~r+ )|⃗r ∈S = 0.

Thus the Dirichlet Green’s function on V is

1 R/|~r ′ |
GD (~r, ~r ′ ) = −
|~r − ~r ′ | |~r − σR (~r ′ )|
which is indeed symmetric: GD (~r, ~r ′ ) = GD (~r ′ , ~r ).

2.4 Electric Dipole and Dielectric Polarisation

2.4.1 Electric Multipole Expansion

Assume we have a charge distribution ρ in a finite region V . It generates a potential via

ˆ
1 ρ(~r ′ )
φ(~r ) = d3 r′ .
4πε0 V |~r − ~r ′ |
We consider an approximation to φ(~r ) when ~r lies far from the region V . This is so-called
electric multipole expansion. Here we assume the region V is inside a sphere of radius R,
´
then by “far”, we mean r >> R. This implies r′ << r in the domain of integration V d3 r′ .

~r
V

45
 In the region r′ << r, we have valid Taylor series expansion
 
1 1 ′ 1
′ = − ~r · ∇ + ··· .
|~r − ~r | r r

Let us write it in the following form

1 1 1
=p = q

|~r − ~r ′ | 2 ′ ′
r − 2~r · ~r + (r ) 2 ′ r′ 2
r 1 − 2 (r̂ · r̂ ′ ) rr + r

⃗
r
where r̂ = r is the unit vector in the direction of ~r.
Recall the series expansion
X ∞
1
√ = tl Pl (x) for |x| ≤ 1, 0 < t < 1.
1 − 2xt + t2 l=0

Here Pl (x)’s are the so-called Legendre polynomials. Explicitly, they are given by

1 dl 2
Pl (x) = (x − 1)l .
2l l! dxl
The first a few terms are

P0 (x) = 1, P1 (x) = x,
1 1
P2 (x) = (3x2 − 1), P3 (x) = (5x3 − 3x),
2 2
···

These Legendre polynomials are orthogonal in the sense

ˆ 1
2
dx Pl (x)Pm (x) = δlm
−1 2l + 1

and complete in the sense

∞ 
X 
1
l+ Pl (x)Pl (x′ ) = δ(x − x′ ).
2
l=0

In terms of Legendre polynomials,

∞  ′ l
1 1X r
′ = Pl (r̂ · r̂ ′ ),
|~r − ~r | r r
l=0

from which we find the multipole expansion

ˆ ∞  ′ l
X
1 r
φ(~r ) = 3 ′
d r Pl (r̂ · r̂ ′ )ρ(~r ′ )
4πε0 r V r
l=0
ˆ "  ′ 2  2 ! #
1 3 ′ ′ ~r · ~r ′ ′ r 1 ~r · ~r ′ ′
= d r ρ(~r ) + 2 ρ(~r ) + 3 − 1 ρ(~r ) + · · ·
4πε0 r V r r 2 rr′
 
1  Q P · ~r
~ X 3ri rj − r δij
2
= + 3 + Qij + ···
4πε0 r r r5
i,j

46
where ˆ
Q= d3 r′ ρ(~r ′ )
V
is the total electric charge, and ˆ
P~ = d3 r′ ρ(~r ′ )~r ′
V
is the electric dipole moment, and
ˆ
1
Qij = d3 r′ ρ(~r ′ )ri′ rj′
2 V

is the electric quadrupole where we write the vector ~r ′ as ~r ′ = (r1′ , r2′ , r3′ ).
For electrically neutral object (Q = 0), the second term describes the leading electrostatic
potential at far distance
1 P~ · ~r
φ(~r ) ' , r >> 0.
4πε0 r3
Then the leading approximation of the electric field is

~ = −∇φ(~r ) ' 1 3(r̂ · P~ )r̂ − P~

E , r >> 0.
4πε0 r3

2.4.2 The Electric Dipole

Consider two point charges +q and −q at a distance d apart. The total charge is zero. The
dipole moment is
~
P~ = q d.

+q

d~
−q

The “electric dipole” is the limit d → 0 and q → +∞ such that P~ is fixed. In this limit,
the potential is
1 P~ · ~r
φ(~r ) =
4πε0 r3
where higher terms in the multipole expansion vanish in the limit. The electric field is

~ r) = 1 3(r̂ · P~ )r̂ − P~
E(~ .
4πε0 r3
This allows us to draw the electric field of the dipole as

+ P~
d→0

−
dipole electric field

47
2.4.3 Dielectric Matter

A dielectric is an electrical insulator that can be polarized by an applied electric field. In

contrast to conductors, dielectrics do not have charges that are free to move around, and are
~ ext will cause dielectric polarisation
typically neutral. Instead, an applied external electric field E
~ int to reduce the overall field within the dielectric. Unlike
that creates an internal electric field E
the conductor, the total macroscopic field is non-zero both inside and outside the dielectric:

~ tot = E
E ~ int 6= 0
~ ext + E inside dielectric.

Let us consider a simple model of a lattice of neutral atoms.

Each nucleus has charge +q at the center surrounded by a spherical cloud of electrons of charge
−q such that the total effect is neutral. Let us now apply an external electric field, and the
atoms get polarized

−+

~
E

with a dipole moment. This can be approximated by electric dipoles. The polarisation P~ (~r )
is the dipole moment density per unit volume. The potential arising from the polarisation is
given by integrating the dipole potential
ˆ ~ r ′ ) · (~r − ~r ′ ) ˆ  
1 3 ′ P (~ 1 3 ′ ~ ′ ′ 1
φ(~r ) = d r = d r P (~r ) · ∇
4πε0 V |~r − ~r ′ |3 4πε0 V |~r − ~r ′ |
ˆ ˆ
1 P~ (~r ′ ) 1 ′ ~ r ′)
3 ′ ∇ · P (~
= d~σ · − d r .
4πε0 ∂V |~r − ~r ′ | 4πε0 V |~r − ~r ′ |
The first surface integral is the potential from the surface charge density

P~ · n̂

where n̂ is the unit normal vector on ∂V .

The second term is the potential from the charge distribution inside the matter

ρbound (~r ) = −∇ · P~ (~r ).

This is called the bound charge due to polarisation. Assume there are also some charges
inside the matter that are free to move, which do not arise form polarisation. Let us call this

48
extra charge ρfree . Then total potential consists of that arising from the polarisation and that
produced from the free charges. Thus the total electric field is

~ = 1 ρ = 1 (ρfree + ρbound ) = 1 (ρfree − ∇ · P~ ).

∇·E
ε0 ε0 ε0
Define the electric displacement
~ = ε0 E
D ~ + P~ .

Then it obeys
∇·D
~ = ρfree .

It turns out that P~ is proportional to E

~ for most materials, which are called linear dielectrics.
Let us write
P~ = ε0 χe E
~

where χe > 0 is called the electric susceptibility. Then

~ = ε0 (1 + χe )E
D ~ = εE
~

where ε > ε0 is called the permittivity of the material. εr = ε/ε0 = 1+χe is called the relative
permittivity. Then
∇·E~ = ρfree
ε
takes the same form as that in the vacuum. The polarisation has simply the effect of replacing
ε0 by ε.

Example 2.4.1. Consider a dielectric sphere of radius R, with a point charge Q at the center.

Then the free charge is

ρfree = Qδ(~r ).

Gauss’s law inside the dielectric

∇·D
~ = ρfree

49
implies (as we have seen before)

~ = Q
D ~r (r < R)
4πr3
The electric field is then
E ~ = 1 Q ~r
~ = D/ε (r < R)
εr 4πε0 r3
This behaves like a charge Q/εr placed at the origin. This says that the bound charge gathers
at the origin, screening the original charge.
Outside the matter, we have

~ = Q
E ~r (r > R)
4πε0 r3
which does not depend on the polarization. On the surface, we have

~+ = Q ~− = Q
E n̂, E n̂.
4πε0 R2 4πεR2

n̂
~+
E
~−
E

From the interface condition, we have a surface charge density

~+ −E
E ~ − = ξ n̂,
ε0
 
Q 1
ξ= 1− .
4πR2 εr
The total surface charge is therefore Q − Q/εr . This is precisely the opposite of the bound
charge at the origin, as expected.

2.5 Magnetic Field and Vector Potential

We move on to study magnetostatics. This is the case when we have a steady current ~j.
As we have seen in the discussion of Maxwell’s equation:

charge =⇒ electric field

current =⇒ magnetic field

Recall the conservation law of charge

∂ρ
+ ∇ · ~j = 0.
∂t

50
In this steady situation when things do not depend on time, we have the steady-current condition

∇ · ~j = 0.

We will focus on the situation ρ = 0 in the study of magnetostatics in this chapter.

2.5.1 Magnetic Field

The Ampère-Maxwell law

!
~
∇×B
~ = µ0 ~j + ε0 ∂ E
∂t

is reduced to the Ampère’s law in the steady case

∇×B
~ = µ0~j.

The equations for magnetic fields generated by a steady current are


∇ × B ~ = µ0~j Ampère’s law
∇ · B~ =0 Gauss’s law
2 = 0)
The Gauss’s law implies that we can express (in the region with HdR

~
~ =∇×A
B

~ called the vector potential. This representation is not unique: the

in terms of a vector A,
~ remains the same under the gauge transformation
magnetic field B

A ~ ′=A
~ 7−→ A ~ + ∇χ

for a function χ = χ(~r ). In terms of the vector potential, the Ampère’s law becomes

~ ) = µ0~j.
∇ × (∇ × A

A direct calculation shows that this is equivalent to

~ + ∇(∇ · A
−∇2 A ~ ) = µ0~j.

Remark 2.5.1. One quick way to see this is to use differential forms. In fact, let us identify the
~ with a 1-form
vector A
A = Ax dx + Ay dy + Az dz.
~ corresponds to the 1-form ∗3 d3 A. Thus the vector ∇ × (∇ × A
The vector ∇ × A ~ ) corresponds
to
∗3 d3 ∗3 dA = d∗3 d3 A.

Using d3 d∗3 + d∗3 d3 = −∇2 , we find

d∗3 d3 A = −∇2 A − d3 d∗3 A = −∇2 A + d3 (∇ · A

~ )

~ + ∇(∇ · A
which corresponds to the vector −∇2 A ~ ).

51
 Due to the gauge degrees of freedom, we can choose some gauge fixing condition to simplify
the problem. A natural choice is the

Coulomb gauge : ~ = 0.
∇·A

~ = f 6= 0. Consider
This can also be achieved for domains in question. In fact, assume ∇ · A
the gauge transformation
~ ′=A
A ~ + ∇χ.

~ ′ = 0 is to solve
Solving the Coulomb gauge condition ∇ · A

∇2 χ = −f

which is the Poisson’s equation. With appropriate boundary condition, the solution exists and
is unique.
~ = 0. Then the
Now let us assume the vector potential satisfies the Coulomb gauge: ∇ · A
Ampère’s law in such gauge becomes

~ = −µ0~j
∇2 A

which becomes the vector version of the Poisson’s equation. Our experience in electrostatics
allows us to write down the solution in the space
ˆ ~j(~r ′ )
~ r ) = µ0
A(~ d3 r′ .
4π |~r − ~r ′ |

We can check that the Coulomb gauge is satisfied when the current ~j is suitably located in some
region V . Then
ˆ   ˆ  
µ0 1 µ0 1
~ r) =
∇ · A(~ d3 r′ ~j(~r ′ ) · ∇ = − d3 ′~ ′
r j(~
r ) · ∇ ′
4π |~r − ~r ′ | 4π |~r − ~r ′ |
ˆ  
µ0 1
= d3 r′ ∇′ · ~j(~r ′ ) = 0.
4π |~r − ~r ′ |

Here we have used integration by part and the steady current condition ∇ · ~j = 0.
As a result, the magnetic field can be written as
ˆ
µ0 ~r − ~r ′
~
B(~r ) = ∇ × A(~r ) =
~ d3 r′ ~j(~r ′ ) × .
4π |~r − ~r ′ |3

This formula is the Biot-Savart law.

There is also an integral form of Ampère’s law. Consider a surface S with boundary curve
C = ∂S. Consider the curve integral
˛ ˆ ˆ
d~r · B =
~ d~σ · (∇ × B) = µ0 d~σ · ~j = µ0 I
~
C S S
´
where I = S d~σ · ~j is the total current passing through S.

52
 ~j

Therefore we find ˛
d~r · B
~ = µ0 I.
C
Example 2.5.2 (Infinite Straight Wire). Consider an infinite straight wire carrying a steady
current I. Assume the wire is placed along the z-axis. Symmetry leads us to consider cylindrical
polar coordinate
(ρ, θ, z)
p
where ρ = x2 + y 2 . The current is
~j = I δ(x)δ(y) ẑ.

Here ẑ is the unit vector along the z-direction.

~
B

θ̂

By transformation symmetry, B ~ = B(ρ,

~ θ) does not depend on z. Using Biot-Savart law,
ˆ ∞ ˆ ∞
~ µ0 Iρ l→ρl µ0 I dl
B(ρ, θ) = dl 2 θ̂ === = θ̂ .
4π −∞ (ρ + l2 )3/2 4πρ −∞ (1 + l2 )3/2
´∞
Here θ̂ is the unit vector along the θ-direction. The last integral can be performed −∞ (1+ldl2 )3/2 =
2. Then
~ µ0 I
B(ρ, θ) = θ̂.
2πρ

~r − ~r ′

53
 We can check this result by considering a loop integral along the circle Cρ of radius ρ in
the xy-plane centered at the origin.

Cρ

˛ ˆ 2π
µ0 I
d~r · B(ρ,
~ θ) = ρdθ = µ0 I
Cρ 0 2πρ
as expected.

Example 2.5.3 (Infinite Plane Current). Consider the surface of xy-plane carrying a constant
~ per unit length. Assume K
surface current density K ~ = kx̂ lies in the x-direction.

y
~
K

Based on our experience from the infinite line wire, and the translation symmetry, we see
~ is oriented along the direction of −ŷ when z > 0, along the direction of ŷ when z < 0,
that B
and its magnitude only depends on z. Let us write

~ = −B(z)ŷ,
B

where B(−z) = −B(z). Now we consider Ampère’s law in the rectangle as illustrated.

C
L

z
y
z

x
−z
˛
d~r · B
~ = B(z)L − B(−z)L = 2B(z)L.
C
The total current through the surface is KL. It follows that 2B(z)L = µ0 KL, i.e.

 µ0 K
− ŷ if z > 0
~ =
B 2

 µ0 K ŷ if z < 0
2

54
Note that the magnitude of the magnetic field is constant, which is the same situation as in
electrostatics. The magnetic field is not continuous and exhibits a jump across a surface current.
Let
~ ± = limit of the magnetic field along the ±n̂ side of the surface.
B

Here is the example n̂ = ẑ. Then

n̂ × (B
~+−B
~ − ) = µ0 K.
~

2.5.2 Interface Condition

A similar argument as in our discussion of electric field interface condition in Section 2.2.3
leads to the following interface condition for magnetic field across a surface with current density
K~ and normal vector n̂.


n̂ · (B
~+−B~ −) = 0 continuous along normal direction
n̂ × (B~+−B~ − ) = µ0 K
~ jump along tangent direction
We leave the details to the reader.
Example. Solenoid
A solenoid consists of a surface current travelling around the cylinder. We consider an
infinite cylinder with surface density in the rotating direction as in the picture.

~
K

Let
(ρ, θ, z)

be the cylinder polar coordinate. Then

~ = K θ̂
K

~ points in the z-direction and is a function of ρ only

along the cylinder. By symmetry, B

~ = B(ρ)ẑ.
B

Away from the cylinder, we have ~j = 0 and

dB
∇×B
~ =0 =⇒ =0 =⇒ B is const.
dρ

55
 ~ = 0 outside the cylinder.
Outside the cylinder: B(+∞) = 0 =⇒ B
Inside the cylinder: Consider the surface

~
B

˛
d~r · B
~ = B(ρ)L, I = KL =⇒ B(ρ) = µ0 K, ~ = µ0 K ẑ.
B
C

2.6 Magnetic Moment and Magnetic Dipole

2.6.1 Magnetic Moment

Assume we have a current distribution ~j localized in a finite region V . It generates a vector

potential via
ˆ ~j(~r ′ )
~ r ) = µ0
A(~ d3 r′ .
4π V |~r − ~r ′ |
~ when ~r lies far away
In parallel with the electrostatic case, we consider an approximation to A
from the region V , called the magnetic multipole expansion. The idea is the same as the
electrostatic case. Consider
∞  ′ l
1 1X r
= Pl (r̂ · r̂ ′ ), r >> 0.
|~r − ~r ′ | r r
l=0

Here r̂ = ⃗rr , and Pl ’s are Legendre polynomials. Plug this into the vector potential, we find
ˆ X∞  ′ l
~ r ) = µ0
A(~ d3 r′
r
Pl (r̂ · r̂ ′ )~j(~r ′ )
4πr V r
l=0
ˆ  
µ0 ~j(~r ′ ) 1 ~r · ~r ′
= d3 r′ + ~j(~r ′ ) 3 + ···
|{z} .
4π V | {z r} | {z r } multipole
monopole term dipole term

The monopole term vanishes. For example, let us consider the x-component. Denote
′
~r = (x′ , y ′ , z ′ ), ~j = (jx , jy , jz ) in components. We have
ˆ ˆ
3 ′ ′
d r jx (~r ) = d3 r′ ∇′ · (x′~j ) − x′ ∇′ · ~j = 0
V V

since ~j is localized inside V and ∇′ · ~j = 0.

Therefore the leading contribution comes from the dipole term
ˆ
µ0
d3 r′ (~r · ~r ′ )~j(~r ′ ).
4πr3 V

56
Using jx = ∇′ · (x′~j(~r ′ )) as above, we have
ˆ ˆ ˆ ˆ
3 ′ ′ ′ ′ ′
d r (~r · ~r )jx (~r ) = d r (~r · ~r )∇ · (x j ) = − d r ∇ (~r · ~r ) · x j = − d3 r′ (~r · ~j )x′ .
3 ′ ′ ′~ 3 ′ ′ ′~
V V V V

Similar formula holds for y, z components, so we get

ˆ ˆ
d r (~r · ~r )j(~r ) = − d3 r′ (~r · ~j )~r ′ .
3 ′ ′ ′
V V

Now using the vector identity ~a × (~b × ~c ) = (~a · ~c )~b − (~a · ~b )~c,

(~r ′ × ~j ) × ~r = (~r · ~r ′ )~j − (~r · ~j )~r ′ ,

we find
ˆ ˆ h i
1
d3 r′ (~r · ~r ′ )~j(~r ′ ) = d3 r′ (~r · ~r ′ )~j(~r ′ ) − (~r · ~j )~r ′
2 V
V
 ˆ 
1 3 ′ ′ ′
= d r ~r × j(~r ) × ~r.
~
2 V

Define the magnetic (dipole) moment by

ˆ
1
m
~ = d3 r′ ~r ′ × ~j(~r ′ ).
2 V
Then the magnetic dipole approximation is
 
~ × ~r
~ r ) ≈ µ0 m µ0 1
A(~ 3
=− m ~ ×∇ .
4π r 4π r

The corresponding magnetic field is (when r >> 0)

  
~ µ0 1
B =∇ × A ' − ∇ × m
~ ~ ×∇
4π r
   
µ0 2 1 µ0 1
=− m∇
~ + ~ · ∇)∇
(m
4π r 4π r
 
µ0 m~ · ~r
=−∇ .
4π r3

Comparing with the electrostatic dipole approximation

1 p~ · ~r ~ = −∇φ,
φ(~r ) = , E
4πε0 r3
we see that the magnetic dipole approximation takes the same form as that of electric dipole

~ = µ0 3(r̂ · m)r̂
B
~ −m ~
, r >> 0.
4π r 3

Example 2.6.1 (Magnetic Moment and Angular Momentum). Consider the current generated
by the motion of a number of charged particles with charges qi , masses Mi , positions ~ri and
velocities ~vi = ~r˙i . The current density is
X
~j(~r ) = qi~vi δ(~r − ~ri ).
i

57
The magnetic (dipole) moment is
ˆ
1 1X 1 X qi L
~i
m
~ = d3 r′ ~r ′ × ~j(~r ′ ) = qi~ri × ~vi =
2 2 2 Mi
i i

~ i = Mi~ri × ~vi is the angular momentum of the i-th particle. Assume all the particles in
where L
motion have the same charge/mass radio qi /Mi = q/M . Then
q ~
L m
~ =
2M
~ =P L
is proportional to the total angular momentum L ~
i i.

~i
L

This classical result is very close to the quantum case describing magnet. The quantum
~ leading to an intrinsic
particles (like electron) carry the so-called “spin” angular momentum S,
spin magnetic moment
q ~
m S.
~g =g
2M
The number g is called the g-factor, and contains important information about the quantum
physics.

Example 2.6.2 (Current Loop). Consider a loop C carrying a steady current I.

~n

I
S

The magnetic moment is

ˆ ˛
1 I
m
~C = d r ~r × ~j(~r ) =
3
~r × d~r.
2 2 C

Assume the circuit C bounds a surface S. Let ξ~ be an arbitrary constant vector. Then
˛ ˛ ˆ
~ I ~ I ~ I
ξ·m ~C = ξ · (~r × d~r ) = d~r · (ξ × ~r ) = d~σ · ∇ × (ξ~ × ~r )
2 C 2 C 2 S
ˆ h i ˆ
I ~ ~ ~
= d~σ · ξ ∇ · ~r − (ξ · ∇)~r = I d~σ · ξ.
2 S S

58
Since ξ~ is arbitrary, we find ˆ
m
~C =I d~σ
S
which holds for any surface S with boundary C.
Now assume C is planer, say lies on a plane with normal vector ~n oriented as indicated.
Then ˆ
d~σ = A~n
S
where A is the area circumscribed by C. In this case

m
~ C = IA~n.

Now let us consider C to be a circle of radius R.

C I

It generates a vector potential by

˛ ˛ " ∞  ′ l
#
d~r ′ X
~ r ) = µ0 I
A(~ =
µ0 I
d~r ′ r ′
Pl (r̂ · r̂ )
4π C |~r − ~r ′ | 4πr C r
l=0
 
X  ′ l
˛
~ C × ~r µ0 I
µ0 m r
= + d~r ′  Pl (r̂ · r̂ ′ ) .
4π r3 4πr C r
l≥2

~ C = IAn̂ = πIR2 n̂. Let us now consider the limit R → 0, I → ∞, while leaving
Here m
a = IA = πIR2 fixed. In the limit,
˛
d~r ′ (~r ′ ) ∼ Rl−1
l
I
C

hence all higher multipoles go to zero. In this way, we obtain a magnetic dipole with magnetic
moment
m
~ = an̂.

This is similar to the limit approach to electric dipole.

n̂

very small loop

59
2.6.2 Magnetic Dipole Layers

We consider a model for a surface S carrying a continuous distribution of magnetic dipoles.

This is called a magnetic dipole layer. Let us assume that the dipoles are oriented normal
to the surface, with the magnetic moment distribution given by

dm
~ = Id~σ .

Let C be the boundary of S.

dm
~

Question: What is the magnetic field produced by such a magnetic dipole layer?
Let us first consider an intuitive approach. We can think of a magnetic dipole as a small
current loop

and decompose the surface S into small current loops

The currents inside S will cancel each other, leading to the effect of a total current I circling
along C. As a result, we should expect that the magnetic field should be the same as that
produced by current I circling along the loop C.
Let us now confirm this intuitive result. By the magnetic dipole formula, we have
ˆ ˆ  
~ µ0 ~r − ~r ′ µ0 ′ 1
A(~r ) = ~ ×
dm = Id~σ × ∇ .
4π S |~r − ~r ′ |3 4π S |~r − ~r ′ |
Here the position on the surface S is parametrized by ~r ′ . Let us choose again an arbitrary
~ then
constant vector ξ,
ˆ    ˆ
~ ~ µ0 I ~ ′ 1 µ0 I ′ ξ~
ξ · A(~r ) = ξ · d~σ × ∇ = d~
σ · ∇ ×
4π S |~r − ~r ′ | 4π S |~r − ~r ′ |
˛  ˛ 
µ0 I ξ~ ~ · µ0 I d~r ′
= d~r ′ · = ξ .
4π C |~r − ~r ′ | 4π C |~r − ~r ′ |

60
 ~ hence
This holds for any vector ξ,
˛
~ r ) = µ0 I d~r ′
A(~
4π C |~r − ~r ′ |

which is indeed the vector potential of a loop C carrying a current I.

2.7 Linking and Magnetic Helicity

We discuss some basic idea about topological aspects of magnetic field. The static Maxwell’s
equations 
∇ · B
~ =0 Gauss’s law
∇ × B~ = µ0~j Ampère’s law

have the integral form ˆ


 d~σ · B
~ =0

˛∂V
ˆ


 d~r · B = µ0 d~σ · ~j = µ0 I
~
∂S S

~
B ~j

∂V V
∂S S

Here V is a region and S is a surface in R3 . ∂V and ∂S are their boundaries.

The Biot-Savart law gives a formula of B ~ from the current ~j. One interesting consequence
is the Gauss’s linking number formula that we have seen before. Let us briefly recall here.
Consider a circle C2 carrying a steady current I. It generates a magnetic field as illustrated

~
B

C2
I

Consider another circle C1 that links with C2 as before. The total current through S is
C1
C2

S
I

61
n(C1 , C2 )I where n(C1 , C2 ) is the linking number of C1 and C2 . Then
˛
d~r1 · B(~
~ r1 ) = µ0 n(C1 , C2 )I
C1

together with the Biot-Savart law

˛
~ r1 ) = µ0 I ~r1 − ~r2
B(~ d~r2 ×
4π C2 |~r1 − ~r2 |3
leads to the Gauss’s integral formula
˛ ˛
1 ~r1 − ~r2
n(C1 , C2 ) = d~r1 · d~r2 × .
4π C1 C2 |~r1 − ~r2 |3
There is a way to capture the topological complexity of magnetic field based on similar
idea, called the magnetic helicity. It is defined by
ˆ
H= ~ · B.
d3 r A ~
V

In terms of differential forms, this is (Exercise)

ˆ
H= A ∧ dA
V

where A = Ax dx + Ay dy + Az dz. For those who are familiar with geometry, you realize that
this is the same form of abelian Chern-Simons action.
Let us first consider the gauge invariance of the magnetic helicity. Under the gauge trans-
formation
A 7−→ A + dχ,

the magnetic helicity changes by

ˆ ˆ ˆ ˆ
dχ ∧ dA = χdA = χ∗3 B = χ (n̂ · B)
~
V ∂V ∂V ∂V

where n̂ is the normal vector on the surface ∂V . We see that in the case when ∂V is a magnetic
surface (n̂ · B
~ = 0), the magnetic helicity will be invariant under gauge transformations.
Let us use Biot-Savart law to analyze the magnetic helicity. We assume the current ~j lies
inside V and ∂V is a magnetic surface. Then the vector potential is given by
ˆ ~j(~r ′ )
µ0
~
A(~r ) = d3 r′ .
4π V |~r − ~r ′ |

Using Ampère’s law ∇ × B~ = µ0~j, this is

ˆ ~ r ′) ˆ
~ 1 ′
3 ′ ∇ × B(~ 1 r − ~r ′ ) × B(~
3 ′ (~
~ r ′)
A(~r ) = d r = − d r .
4π V |~r − ~r ′ | 4π V |~r − ~r ′ |3
Then the magnetic helicity becomes
 
ˆ ˆ ˆ ~ r ′ ) × (~r − ~r ′ )
~ r ) · B(~
B(~
H= d3 r A ~ = 1
~ ·B d3 r d3 r′ .
V 4π V V |~r − ~r ′ |3
Comparing with the Gauss’s linking formula, we see that the magnetic helicity can be intuitively
interpreted as the averaged linking over all pairs of magnetic field lines.

62
 2.8 Dirac Monopole
The Gauss’s law of magnetic field can be written as

d3 ∗3 B = 0

where B = Bx dx + By dy + Bz dz. Here recall d3 denotes the de Rham differential on R3 . In

other words, ∗3 B is a closed 2-form, a fact which inherits many topological natures. When the
region has trivial topology with H 2 = 0, we can always find a vector potential A~ such that
dR

∗3 B = d3 A

where A = Ax dx + Ay dy + Az dz. However, when we work with a region V that has nontrivial
topology, we may not be able to find such an A globally, although locally it always exists
(Poincaré’s Lemma). We will systematically study this later in Chapter 4 in terms of fiber
bundle. Here we illustrate some basic feature via the example of Dirac monopole.
The Dirac monopole has magnetic field

~ = g ~r
B
4π r3
where g is the magnetic charge. It satisfies

∇·B
~ = gδ(r)

hence describing a magnetic monopole of charge g at the origin. Although we have not observed
magnetic monopole in the lab yet, this is still an interesting theoretical model. We can compare
the above formula with the point electric charge

~ = q ~r
E .
4πε0 r3

Let us consider the region of the complement of the origin in R3 :

V = R3 − {0}.

We have
g
∗3 B = (xdy ∧ dz + ydz ∧ dx + zdx ∧ dy)
4πr3
which defines a closed 2-form on V . However, there does not exist globally a 1-form A on V
such that
∗3 B = d3 A.

Indeed, assume this equation holds. Consider the surface S of the unit sphere. On one hand,
we have ˆ ˆ ˆ
∗3 B = d~σ · B
~ = d3 r∇ · B
~ = g.
S S
On the other hand, ˆ ˆ
dA = A = 0.
S ∂S

63
This is a contradiction.
In fact, V has the homotopy type of S 2 , and

2
HdR (V ) ' HdR
2
(S 2 ) ' R.

The 2-form ∗3 B defines a nontrivial element in HdR

2 (V ). Instead, it is possible to write ∗ B as
3

dA locally. For example, let us consider

U+ = R3 \ {(0, 0, z) | z ≤ 0}
U− = R3 \ {(0, 0, z) | z ≥ 0}

z≥0

z≤0

U+ U−

It is clear that 
U+ ∪ U− = V = R3 − {0}
U ∩ U = R3 − {z-axis}
+ −

Consider the following 1-form on each patch

g
A± = (−ydx + xdy) on U± .
4πr(z ± r)
In spherical coordinate (r, θ, ϕ),


x = r sin θ cos ϕ


y = r sin θ sin ϕ




z = r cos ϕ

(r, θ, ϕ)
r
θ

y
ϕ

We can write
g
A± = ± (1 ∓ cos θ)dϕ.
4π

64
Note that ∗3 B in spherical coordinate is
g
∗3 B = sin θdθdϕ
4π
where sin θdθdϕ is the standard area form on S 2 . It can now be easily checked that

∗3 B = dA± valid on U± .

On the intersection U+ ∩ U− , where both A+ and A− are defined, we have

d(A+ − A− ) = 0

so A+ − A− is a closed 1-form. Explicitly, we have

g
A+ − A− = dϕ
2π
which defines a nontrivial element of

1
HdR (U+ ∩ U− ) = HdR
1
(R3 − {z-axis}) ' HdR
1
(S 1 ) = R.

S1

U+ ∩ U− ' R2 − {0} ' S 1

Indeed, consider the unit circle S 1 on the xy-plane. Then

ˆ ˆ
g
A+ − A− = dϕ = g 6= 0.
S1 2π S 1

An important consequence of this calculation is the Dirac quantization condition. The

combination eA plays the role of a connection 1-form for a U (1)-bundle (see Section 4.6). Then
the 2-form
1 1
edA = e∗3 B
2π 2π
is the corresponding first Chern class hence an integral form. In particular, the integration
ˆ
1 ge
e∗3 B = ∈Z
2π S 2 2π

must be an integer (Theorem 4.3.6). This gives the Dirac quantization condition

ge ∈ 2πZ.

65
 Chapter 3 Electrodynamics

In this chapter, we study dynamical aspects of electromagnetism where the electric and
magnetic fields will evolve with time in general. We will see how electromagnetic waves arise
from solving Maxwell’s equations, and how they are produced by accelerating charges.

3.1 Force and Energy

3.1.1 Lorentz Force

~ and
A particle of charge q moving with velocity ~v in the background of electric field E
~ experiences force via the Lorentz force law
magnetic field B

~ + ~v × B).
F~ = q(E ~

Example 3.1.1 (Force between steady charges). Consider two charge distributions ρ1 and ρ2
in the space.

ρ1 ρ2

Then electric field produced by ρ2 is

ˆ
1 ~r − ~r ′
~ 2 (~r ) =
E d3 r′ ρ2 (~r ′ ) .
4πε0 |~r − ~r ′ |3

The electric force that ρ2 exerts on ρ1 is then given by

ˆ ˆ
1 ~r − ~r ′
F~1 =
e
d r d3 r′ ρ1 (~r )ρ2 (~r ′ )
3
.
4πε0 |~r − ~r ′ |3

By symmetry, the force that ρ1 exerts on ρ2 is

F~2e = −F~1e .

In particular, F~1e = F~2e = 0 when ρ1 = ρ2 . Thus the net force of a charge distribution on itself
is zero.

Example 3.1.2 (Force between steady currents). Consider two steady current ~j1 and ~j2 in the
space.

66
 ~j1 ~j2

The magnetic field produced by ~j2 is

ˆ
µ0 ~r − ~r ′
~
B2 (~r ) = d3 r′ ~j2 (~r ′ ) × .
4π |~r − ~r ′ |3

The magnetic force that ~j2 exerts on ~j1 is given by

ˆ ˆ ˆ  ′ 
µ 0 ′ ′ ~
r − ~
r
F~1 = d r ~j1 (~r ) × B
m 3 ~ 2 (~r ) = d r d r ~j1 (~r ) × ~j2 (~r ) ×
3 3
4π |~r − ~r ′ |3
   
ˆ ˆ ~j1 (~r ) · (~r − ~r ′ ) ~j2 (~r ′ ) − ~j1 (~r ) · ~j2 (~r ′ ) (~r − ~r ′ )
µ0 3 3 ′
= d r d r .
4π |~r − ~r ′ |3

In the first term, using ∇ · ~j1 = 0,

ˆ ˆ !
~r − ~r ′ ~j1 (~r )
d r j1 (~r ) ·
3 ~
= − d3 r ∇ · .
|~r − ~r ′ |3 |~r − ~r ′ |

If ~j1 (~r ) → 0 faster than 1

r as r → ∞, then the divergence theorem implies that this integral
vanishes. We assume this is the case. Then we find
ˆ ˆ   ~r − ~r ′
µ0
F1 = −
~ m
d r d3 r′ ~j1 (~r ) · ~j2 (~r ′ )
3
.
4π |~r − ~r ′ |3

By symmetry, the force that ~j1 exerts on ~j2 is

F~2m = −F~1m .

Again, F~1m = F~2m = 0 when ~j1 = ~j2 . Thus the net force of a current distribution on itself is
zero.
Comparing with the electric force case, we find that F~1m takes the same form as that of
F~2m , except with an extra minus sign. This sign difference says that parallel currents attract
and anti-parallel currents repel.

~j1 ~j2 ~j1 ~j2

attract repel

67
Example 3.1.3 (Helical motion in magnetic field). Consider a uniform constant magnetic field

~ = (0, 0, B)
B

pointing in the z-direction with magnitude B. We consider the motion of a particle of charge
q, mass m. The equation of motion is

m~r¨ = q~r˙ × B.
~

In components, this reads



ẍ = ωc ẏ

 qB
ÿ = −ωc ẋ where ωc =

 m


z̈ = 0

This can be solved by



x(t) = x0 + R cos(ωc t − θ0 )


y(t) = y0 − R sin(ωc t − θ0 )




z(t) = z0 + vz t
Here x0 , y0 , z0 , R, θ0 , vz are constants. The particle executes circular motion at the cyclotron
frequency ωc in the xy-direction, and moves with constant speed in the z-direction.
y

vz = 0

3.1.2 Electromagnetic Energy

We explore the energy stored in electromagnetic fields. Suppose we have some charge and
~ Here we consider
~ and magnetic field B.
current configuration which produces electric field E
the dynamical case and things will vary with time t. Then we need the full Maxwell’s equations.
Consider the infinitesimal work dW done by the electromagnetic forces on these charges in
the infitinesimal time dt. According to Lorentz force law, the work on a charge q is
   
F~ · d~r = q E~ + ~r˙ × B
~ · ~r˙ dt = q E~ · ~v dt.

This can be generalized to the case of charge and current densities, by replacing

q −→ ρd3 r, ρ~v −→ ~j.

68
Therefore the total work done in the region V is
ˆ
dW ~ · ~j.
= d3 r E
dt V

Using Ampère-Maxwell law

!
~ ~
~ = µ0~j + 1 ∂ E = µ0 ~j + ε0 ∂ E
∇×B 2
c ∂t ∂t

and Faraday’s law

~
~ = − ∂B ,
∇×E
∂t
we find
~
~ · ~j = 1 E
E ~ · (∇ × B)
~ − ε0 E~ · dE
µ0 dt
h i ~
=
1
∇ · (B
~ × E)
~ +B ~ · (∇ × E) ~ · dE
~ − ε0 E
µ0 dt
~ ~
~ · dE − 1 B
= −ε0 E ~ · ∂ B − 1 ∇ · (E
~ × B).
~
dt µ0 ∂t µ0
It follows that ˆ   ˆ
dW d 1 1 ~2 1
=− ε0 E + B −
~ 2
d~σ · (E
~ × B).
~
dt dt V 2 µ0 µ0 ∂V
This formula has the following interpretation: the quantity
ˆ
1 ~2 + 1 B~2
U= ε0 E
2 V µ0
is the total energy of electromagnetic field stored in the region V . Define the

Poynting vector : ~= 1E
S ~ ×B
~
µ0
which describes the energy flow. Then
ˆ ˆ
1
d~σ · (E
~ × B) =
~ d~σ · S
~
µ0 ∂V ∂V

describes the energy that flows out through ∂V .

~
S

Then the formula ˆ

dW dU
=− − d~σ · S
~
dt dt ∂V

69
gives the expected interpretation: the work done by the electromagnetic force is equal to the
decrease in the total energy in the region less the energy that flowed out of the region. Let
 
1 ~2 + 1 B~2
u= ε0 E
2 µ0

denote the energy density. We have

~ · ~j = − ∂u − ∇ · S.
E ~
∂t
In the region of empty space where no work is done, the above equation becomes
∂u
+∇·S
~ = 0.
∂t
∂ρ
Comparing with the charge conservation ∂t + ∇ · ~j = 0, the above equation can be viewed as
the local form of the conservation of electromagnetic energy.

Example 3.1.4. Consider a uniformly charged spherical shell of total charge Q and radius R.

R
Q

The electric field is 

 Q ~r
 3
r>R
~ = 4πε0 r
E


0 r<R
Therefore the total energy is
ˆ ˆ  2 ˆ ∞
ε0 3 ~ = ε0
~ ·E 3 Q 1 Q2 1 Q2
U= d rE d r = dr (4πr2 ) = .
2 2 |⃗
r |≥R 4πε0 r4 32π 2 ε0 R r4 8πε0 R

Example 3.1.5 (Electromagnetic wave). Consider the following solution of Maxwell’s equations
in the vacuum: 

E~ = cos(~k · ~r − ωt)E
~0 ~k
where k̂ = .

B~ = 1 cos(~k · ~r − ωt)k̂ × E
~0 |~k|
c
~ 0 , ~k, ω are constants and satisfy
Here E

c2~k · ~k = ω 2 , ~k · E
~ 0 = 0.

This describes an electromagnetic wave that we will discuss in detail in Section 3.3. This
wave travels in the direction of ~k, while the electric field and magnetic field are oscillating in
orthogonal directions.

70
 ~k
k̂
~
B
~
E

~
B
~
E

The Poynting vector is

2
~= 1E
S ~ = E0 cos2 (~k · ~r − ωt)k̂
~ ×B where E0 = |E
~ 0 |.
µ0 cµ0
We see that the energy is indeed transporting in the direction of the wave propagation k̂. The
energy density is  
1 1 ~2
u= ε0 E + B = ε0 E20 cos2 (~k · ~r − ωt).
~ 2
2 µ0
We see that
~ = cu k̂
S

which simply says that the wave is transporting energy in the speed of light.

3.2 Electromagnetic Induction

We explore a bit about Faraday’s law
~
~ + ∂B = 0
∇×E
∂t
which connects the production of electric field with the change of magnetic field. This phe-
nomenon is also called Faraday’s Law of Induction.

3.2.1 Electromotive Force and Flux Rule

Let C be a conducting circuit. The electromotive force (“emf” for short) is the accu-
mulated tangential force per unit charge throughout the circuit. Equivalently, emf is the work
done on a unit charge moving around the whole circuit.
In a circuit context, the emf will drive the charges in the wire and form a current. In many
circumstances, the resulting current I is proportional to the applied emf E, which is called
Ohm’s Law:
E = IR.

The constant of proportionality R is called the resistance.

Let ~v denote the velocity of a unit charge on C. Using the Lorentz force law, the tangential
force per unit charge integrated along the circuit C is
˛
E= d~l · (E
~ + ~v × B
~ ).
C

71
Note that if the circuit C is static in the space, then ~v is always pointed along the direction of
the circuit, hence d~l · (~v × B
~ ) = 0. In this case, we have
˛
E= d~l · E
~ if C is at rest.
C

In other words, the second term in E comes from the effect of moving circuit. The precise
relationship is described by the “Flux Rule” that we now derive.
Assume C bounds a surface S. The magnetic flux through S is defined to be
ˆ
Φ= d~σ · B.
~
S

~
B

Let us now consider the change of Φ with respect to time. Assume the circuit C is moving with
velocity ~vC (~r ) at each point ~r of C. Let us write
ˆ
Φ= ∗3 B
S

where B = Bx dx + By dy + Bz dz. Then

ˆ ˆ
d ∂
Φ= (∗3 B) + ι⃗vC (∗3 B).
dt S ∂t C

Here ι⃗vC is the contraction with respect to the vector field ~vC . Expanding the notation, we find
(Exercise) ˆ ˆ
ι⃗vC (∗3 B) = − d~l · (~vC × B
~ ).
C C
This can be also understood from the picture: the total charge of the flux is computed by the
“volume” formed by d~l, ~vC , B.
~ We know that the “volume” is computed by the determinant,
hence the expression d~l · (~vC × B
~ ).

~vC ~
B

d~l

72
 It follows that ˆ ˆ
dΦ ∂B~
= d~σ · − d~l · (~vC × B) ~
dt Sˆ ∂t C ˆ
Faraday’s
======= − d~σ · (∇ × E)
~ − d~l · (~vC × B)
~
law
ˆS ˆ C

=− ~
dl · E −
~ d~l · (~vC × B).~
C C
Now the velocity ~v differs from ~vC by a tangential vector along C, hence
ˆ ˆ
~
dl · (~vC × B) =
~ d~l · (~v × B).
~
C C

We have now arrived at

dΦ
Flux Rule : E=−
dt
the change of the magnetic flux is the emf.
To understand the minus sign, imagine that the emf will produce a current which itself will
induce a magnetic field. The minus sign says that the induced magnetic field will always be in
the direction that opposes the change. This is called Lenz’s Law. As a result, the Faraday
law of induction is similar to the “inertial” phenomenon in mechanics. A conducting circuit
prefers to maintain a constant flux through it. The change of the flux will result in a responding
current in such a direction to oppose the change.

~
B

~ increases
assume B emf induced current and magnetic field

Example 3.2.1 (Jumping Ring). Place a metal ring on top of a solenoidal coil around an iron
core. Let us now switch on the current, which will produce a magnetic field. This will induce
a current in the ring in the opposite direction. Since opposite current repel, the ring will jump
out.

73
 ring

solenoid

3.2.2 Mutual Inductance

Consider two loops C1 , C2 of wire at rest. If we run a steady current I1 around C1 , it will
~ 1 and hence a magnetic flux Φ2 through the loop C2 . The flux Φ2 is
produce a magnetic field B
proportional to the current I1
Φ2 = M21 I1

where the constant M21 is called the mutual inductance of C1 and C2 .

S2 C2

~1
B

C1
I1

In Coulomb gauge, the vector potential A ~ 1 produced by I1 is

˛
~ 1 (~r ) = µ0 I1 d~r ′
A .
4π C1 |~r − ~r ′ |

Therefore
ˆ ˆ ˛ ˛ ˛
~ 1) = ~ 1 (~r ) = µ0 I1 d~r · d~r ′
Φ2 = d~σ · B
~1 = d~σ · (∇ × A d~r · A .
S2 S2 C2 4π C2 C1 |~r − ~r ′ |

It follows that ˛ ˛
µ0 d~r · d~r ′
M21 =
4π C2 C1 |~r − ~r ′ |

74
which is a purely geometric quantity. This expression is invariant under the switch of C1 and
C2 , hence
call
M21 = M12 ==== M.

As a consequence, if we vary the current in C1 , Faraday’s law of induction says this will
vary the flow through C2 and produce a current in C2 . Isn’t this amazing!

3.2.3 Self Inductance

A changing current also induces an emf in the source loop itself. The flux is proportional
to the current
Φ = LI

where L is called the self induction of the loop. The induced emf is then given by
dI
E = −L .
dt
Example 3.2.2 (Inductance of the Solenoid). Consider a solenoid wrapped by a wire which
carries current I and winds n times per unit length. Assume the solenoid has length l and
√
cross-sectional area A, and l >> A. This is approximately the infinite cylinder example with
horizontal surface current density N I as we discussed before. The magnetic field inside the
solenoid is
~ = µ0 N I ẑ.
B

~
B

The wire winds N l times, hence the total flux is

Φ = (µ0 N IA)N l = µ0 IN 2 Al = µ0 IN 2 V

where V = Al is the volume of the solenoid. The self-inductance is therefore L = µ0 N 2 V .

75
3.2.4 Magnetostatic Energy

Let us try to increase the current of a circuit C to achieve the current I. The work done
in this process can be viewed as the energy stored in the circuit. At time t, the induced emf is
dI
E = −L .
dt
Therefore the rate of electrical work by induced forces is
dW dI
= EI = −LI ,
dt dt
so
1
W = − LI 2 .
2
The total energy stored is U = −W = 12 LI 2 . It is illustrating to compare this with the
mechanical case

Particle Circuit
mv (momentum) LI
1 2 1 2
2 mv (kinetic energy) 2 LI

To see this is the correct magnetostatic energy, we can write it as

ˆ ˛
1 1 1 1 ~
U = I(LI) = IΦ = I d~σ · B = I
~ d~l · A.
2 2 2 S 2 C

Let us rewrite this into the form of current density by replacing Id~l → ~j d3 r
ˆ
1 ~
U= d3 r ~j · A
2
In the magnetostatic case, this is
ˆ ˆ ˆ
1 ~ 1 ~ 1
U= d r (∇ × B) · A =
3 ~ d r (∇ × A) · B =
3 ~ ~ ·B
d3 r B ~
2µ0 2µ0 2µ0
~ = 0) that we find previously in Section 3.1.2.
which is the same electromagnetic energy (when E

Example 3.2.3 (Solenoid). We consider the solenoid as described in the previous example.
The energy density of the solenoid is

1 ~ ~ µ0 N 2 I 2
B·B= .
2µ0 2
The total stored energy is
µ0 N 2 I 2
U= V.
2
Comparing with the formula U = 12 LI 2 , we find

L = µ0 N 2 V

which coincides with the previous calculation of self inductance.

76
 3.3 Electromagnetic Wave
We study Maxwell’s equations in the absence of sources, which gives rise to solutions
by electromagnetic waves propagating in spacetime. In particular, this leads to Maxwell’s
speculation of light as electromagnetic wave, and brings light, electricity and magnetism into
the same fundamental phenomenon.

3.3.1 Wave Equation

Maxwell’s Equations in the vacuum without sources are



∇·E ~ =0





∇ · B
~ =0

∇×E ~ = − ∂B⃗




∂t

∇ × B ⃗
~ = µ0 ε 0 ∂ E
∂t

Here µ0 ε0 = 1/c2 . In terms of the electromagnetic 2-form

F = E ∧ dt + ∗3 B = E ∧ dt + ∗(B ∧ cdt)
= (Ex dx + Ey dy + Ez dz) ∧ dt + (Bx dy ∧ dz + By dz ∧ dx + Bz dx ∧ dy),

the above equations have the geometric form


dF = 0
d∗ F = 0

Here d∗ = ∗d∗ is the adjoint of d in R3,1 . Let

1 ∂2 1 ∂2 ∂2 ∂2 ∂2
□= − ∇ 2
= − − −
c2 ∂t2 c2 ∂t2 ∂x2 ∂y 2 ∂z 2
be the d’Alembert operator. Using the relation

dd∗ + d∗ d = −□,

we find that F satisfies

□F = 0.

In components, this is 
 ~
1 ∂2E

 2 2 − ∇2 E ~ =0
c ∂t

 2~
 1 ∂ B − ∇2 B ~ =0
c2 ∂t2
which are the vector valued wave equations travelling at the speed of light c.
Since Maxwell’s Equations in the vacuum are linear, it turns out to be convenient to express
solutions via complex valued functions while electromagnetic fields are obtained by taking their
real part. We will assume this implicitly and will not distinguish real and complex valued
solutions when it is clear from the context.

77
3.3.2 Plane Waves

Let ~k = (kx , ky , kz ) ∈ R3 be a constant vector in R3 . A standard plane wave solving the

scalar wave equation
□ϕ = 0

propagating along ~k is given by the form

⃗
ϕ = ei(k·⃗r−ωt) .

Plug into the wave equation, we find

□ϕ = 0 ⇐⇒ ω 2 = c2 k 2 where k = |~k|.

The two cases ω = ±ck corresponding to two waves propagating in opposite directions along ~k.
We will take
ω = ck

in the following discussion. The other case can be equivalently described by the propagating
vector −~k.
To see the precise meaning of ~k and ω, let us first consider standing at a fixed point ~r in the
⃗
space. Then the wave ei(k·⃗r−ωt) at the point ~r will oscillate ω
2π times per unit time. Similarly, if
k
we look at the wave at an instant time t, we find that the wave will oscillate 2π times per unit
length along the direction ~k.

ω k
2π times per unit time 2π times per unit length

ω k
So 2π / 2π = ω/k is the velocity of the wave.

Remark 3.3.1. The ratio ω/k is called the “phase velocity” of the wave. For electromagnetic
waves in the vacuum, the phase velocity is always the speed of light c for waves of any frequency:
ω/k = c. This is called “dispersionless” waves. If we consider waves in dispersive matter, we
will find the phase velocity depending on frequency

ω(k) = v(k) · k.

Such equation is generally called the dispersion relation.

Let us now describe an electromagnetic plane wave given by


E ~ =E ~ 0 ei(⃗k·⃗r−ωt)
where ω = ck.
B ~ =B ~ ei(⃗k·⃗r−ωt)
0

78
 ~ 0 are complex valued constant vectors. Physical solutions are their real parts as
~ 0, B
Here E
remarked above. Since E ~ 0 are constants, they solve the wave equation
~ 0, B

□E
~ = □B
~ = 0.

On the other hand, Maxwell’s equations say more than wave equations and will put further
constraints. The divergence free relations

∇·E
~ = 0, ∇·B
~ =0

is equivalent to
~ 0 · ~k = 0,
E ~ 0 · ~k = 0.
B
~ are both perpendicular to the propagating direction ~k. The curl
~ and B
This means that E
relations
~
∂B 1 ∂E~
∇×E
~ =− , ∇×B
~ =
∂t c2 ∂t
is equivalent to
~k × E ~ 0,
~ 0 = ωB ~0 = −ωE
~k × B ~ 0.
c2
~ 0 and
~ 0 is also perpendicular to B
This means that E

~k
~ 0 = k̂ × E
cB ~0 where k̂ = .
k
The above relations can be summarized by the geometric figure

~k

|E
~ 0 | = c|B
~ 0|
~0
cB

~0
E

3.3.3 Polarization

Let us consider electromagnetic waves in the propagating direction ~k. Let us fix two unit
vectors ê1 , ê2 ∈ R3 such that

k̂

ê1 · ê2 = 0, ê1 × ê2 = k̂.

ê2

ê1

79
Let us express
~ 0 = E1 ê1 + E2 ê2
E
where we emphasise that E1 , E2 are complex numbers. Then
~ = (E1 ê1 + E2 ê2 ) ei(⃗k·⃗r−ωt) .
E
~ can be obtained from E
The magnetic field B ~ by

B~ = 1 k̂ × E~ = 1 (−E2 ê1 + E1 ê2 ) ei(⃗k·⃗r−ωt) .

c c
Let us focus on the electric field. The physical field is given by the real part
h i
⃗
Re (E1 ê1 + E2 ê2 ) ei(k·⃗r−ωt) .

Let us write
E1 = A1 eiθ1 , E2 = A2 eiθ2 , Ai ≥ 0.
The behavior of the electric/magnetic field can be described via the following cases:
⃝
1 Linear Polarization: eiθ2 = ±eiθ1 or θ2 − θ1 ∈ Zπ. In this case we have
h i
~ = Re ~uei(⃗k·⃗r−ωt+θ1 ) = ~u cos(~k · ~r − ωt + θ1 )
E

where ~u is the fixed vector ~u = A1 ê1 ± A2 ê2 . The electromagnetic fields are oscillating along
fixed directions. The wave at an instant time looks like
~k
k̂
~
B
~
E

k̂ × ~u
~u

⃝
2 Circular Polarization: E2 = ±iE1 . In this case we have
h i  
~ ± = Re (ê1 ± iê2 )A1 ei(⃗k·⃗r−ωt+θ1 ) = A1 cos(~k · ~r − ωt + θ1 )ê1 ∓ sin(~k · ~r − ωt + θ1 )ê2 .
E
~ ± has a constant magnitude, but rotates in a
At a fixed position ~r in the space, the vector E
circle at a frequency ω. E ~ + is rotating counterclockwise, while E
~ − is rotating clockwise, as
shown below

~+
E ~−
E
ê2 ê2

ê1 ê1

Left circular polarization Right circular polarization

80
The wave at an instant time looks like (for the right circular polarization case)

~k

⃝
3 Elliptical Polarization: general E1 and E2 .
h i
⃗
Re (E1 ê1 + E2 ê2 )ei(k·⃗r−ωt) = A1 cos(~k · ~r − ωt + θ1 )ê1 + A2 cos(~k · ~r − ωt + θ2 )ê2 .

It swaps out an ellipse at any point in the space.

ê2 ~ ê2
cB
~
E

ê1 ê1

3.3.4 Wave Packets

In the above, we have focused on the case of a plane wave with fixed ~k. Such a wave
⃗
ei(k·⃗r−ωt) spreads out the full space and can not exist in nature. In general, since the wave
equation is linear, a localized wave can be obtained by superposing plane waves (like Fourier
modes) with different wave vectors. A general electromagnetic wave packet is given by
ˆ
1 ⃗
~
E= 3
d3 k ~E(~k )ei(k·⃗r−ωt)
(2π)
ˆ  
B~ = 1 d3 1
k k̂ × ~E(~k ) ei(⃗k·⃗r−ωt)
(2π)3 c

or more precisely their real part. The vector function ~E(~k ) plays the role of Fourier coefficients.
In the vacuum which is dispersionless, we always have ω = ck. In dispersive matters, we have
a more complicated dispersion relation between ω and k.
Recall that the energy density of an electromagnetic field is
   
1 ~ 2 + 1 |Re B|
~ 2 = ε0 |Re E|
u= ε0 |Re E| ~ 2 + |Re cB|
~ 2
2 µ0 2
and the total energy U is
ˆ  
ε0
U= d3 r |Re E|
~ 2 + |Re cB|
~ 2 .
2

81
In the vacuum case (ω = ck), we calculate

~ 2 = 1 (E
|Re E| ~ ·E
~ +E~∗ ·E ~ ∗ + 2E ~ ∗ ) = 1 (E
~ ·E ~ ·E ~ +E ~ ·E ~ ∗ ) + c.c.
4 ˆ ˆ 4
1 1 1  
= d 3
k d3~ ′ ~ ~
k E(k ) · ~E(~k ′ )ei(⃗k+⃗k′ )·⃗r−ic(k+k′ )t + ~E(~k ) · ~E∗ (~k ′ )ei(⃗k−⃗k′ )·⃗r−ic(k−k′ )t + c.c.
4 (2π)3 (2π)3
Here c.c. means the complex conjugate of the expression before it. Using the identity
ˆ
1
d3 r ei⃗p·⃗r = δ(~
p ),
(2π)3
we find
ˆ ˆ  
~ 2=1 1
d3 r |Re E| d3 k ~E(~k ) · ~E∗ (~k ) + ~E(~k ) · ~E(−~k )e−2ickt + c.c.
4 (2π)3
Similarly, the contribution from the magnetic field is
ˆ ˆ h        i
~ 2=1 1
d3 r |Re cB| d3
k k̂ × ~E(~k ) · k̂ × ~E∗ (~k ) + k̂ × ~E(~k ) · −k̂ × ~E(−~k ) e−2ickt + c.c.
4 (2π)3
ˆ  
1 1 3 ~E(~k ) · ~E∗ (~k ) − ~E(~k ) · ~E(−~k )e−2ickt + c.c.
= d k
4 (2π)3
Adding the above two contributions, we find the following total energy formula for electromag-
netic wave packet in a region of vacuum
ˆ
ε0 1
U= d3 k |~E(~k )|2 .
2 (2π)3

3.4 Green’s Functions

3.4.1 Wave Equation with Source

We now consider solving the full Maxwell’s equations with specified distributions of charge
and current: 

∇·E ~ = ρ/ε0





∇ · B
~ =0

∇×E
⃗
~ = − ∂B




∂t
 

∇ × B ⃗
~ = µ0 ~j + ε0 ∂ E
∂t

In terms of the electromagnetic 2-form F and the charge current 1-form J

F = E ∧ dt + ∗3 B, J = ρ/ε0 dt − µ0 j,

Maxwell’s equations take the geometric form


dF = 0
d∗ F = J

Here d∗ = ∗d∗ is the adjoint of d in R3,1 . Recall

d2 = 0, (d∗ )2 = 0, dd∗ + d∗ d = −□.

82
The consistency condition for d∗ F = J requires

d∗ J = 0.

Since d∗ J = − c21ε0 ∂t ρ − µ0 ∇ · ~j = −µ0 (∂t ρ + ∇ · ~j ), the consistency condition is precisely the

charge conservation as we discussed in Section 1.3

∂t ρ + ∇ · ~j = 0.

The equation dF = 0 can be solved by introducing the potential 1-form A on R3,1 such
that
F = dA.

Expressing in components,
A = −φdt + A

where φ is the scalar potential and A = Ax dx + Ay dy + Az dz is the vector potential. Now

in general, φ and A depends on the position ~r and time t. The equation F = dA reads in
components by 
E ~
~ = −∇φ − ∂t A
B~ =∇×A ~

The choice of A is not unique, and we have the following gauge transformation leaving F
invariant
A 7−→ A + dχ

for any function χ on R3,1 . We need a gauge fixing condition to specify the solution of A as
discussed in Section 1.10. We will focus on the Lorenz gauge here:
1 ∂φ
Lorenz gauge: d∗ A = 0 ⇐⇒ ~ = 0.
+∇·A
c2 ∂t
In terms of the potential 1-form A, the other half of Maxwell’s Equations d∗ F = J becomes

d∗ dA = J.

In Lorenz gauge, 
d∗ dA = J Maxwell’s Equations
d∗ A = 0 Lorenz gauge
from which we find
□A = −J.

In components, this becomes 

□φ = ρ/ε0
□A
~ = µ ~j 0

These are inhomogeneous wave equations with source terms.

83
3.4.2 Green’s Function for the Wave Equation

To solve the inhomogeneous wave equation, we follow the same strategy as before by con-
structing the inverse “ □1 ”, which is called the Green’s function for the wave equation.
By definition, the Green’s function G(~r, t; ~r ′ , t′ ) for the wave equation is the wave produced
at (~r ′ , t′ ) by a unit point source at (~r, t). It satisfies

□′ G(~r, t; ~r ′ , t′ ) = δ(~r − ~r ′ )δ(t − t′ ).

Here □′ = 1 ∂2
c2 ∂t′2
− ∇2′ acts on the variable (~r ′ , t′ ).
Let us first explain how to solve inhomogeneous wave equation using the Green’s function,
and come back to the construction of the Green’s function later. Consider a function ϕ(~r, t)
satisfying the inhomogeneous wave equation

□ϕ = f.

To find the expression of ϕ in terms of f , we apply the Green’s function in the region V ∈ R3
and time interval [t1 , t2 ]:
ˆ t2 ˆ
ϕ(~r, t) = d3 r′ ϕ(~r ′ , t′ )δ(~r − ~r ′ )δ(t − t′ )
dt ′
t1 V
ˆ t2 ˆ  
′ ′ 1 ∂2
= dt′ 3 ′
d r ϕ(~r , t ) 2 ′2 − ∇ 2′
G(~r, t; ~r ′ , t′ ).
t1 V c ∂t
For the first term, we use integration by part and find
ˆ t2 ˆ t2  2   
′ ∂2 ′ ∂ ∂G ∂ϕ t2
dt ϕ ′2 G = dt ϕ G+ ϕ ′ −G ′  .
t1 ∂t t1 ∂t′2 ∂t ∂t t1

For the second term, we have as before

ˆ ˆ ˆ
3 ′ 3 ′



2′
d r ϕ∇ G = d r ∇ ϕ G+
2′
ϕ∂n̂′ G − G∂n̂′ ϕ
V V ∂V

where n̂ is the unit normal vector on the surface ∂V .

Combining the above two equations and using □ϕ = f , we find
ˆ t2 ˆ
ϕ(~r, t) = dt ′
d3 r′ G(~r, t; ~r ′ , t′ )f (~r ′ , t′ )
t1 V
ˆ ˆ
t2  
+ dt′ G(~r, t; ~r ′ , t′ )∂n̂′ ϕ(~r ′ , t′ ) − ϕ(~r ′ , t′ )∂n̂′ G(~r, t; ~r ′ , t′ )
t1 ∂V
ˆ ′
1 
3 ′ ′ ′ ′ ′ ′ ′ ′ ′ 
 t =t2
+ 2 d r ϕ(~r , t )∂t′ G(~r, t; ~r , t ) − G(~r, t; ~r , t )∂t′ ϕ(~r , t )  .
c V t′ =t1

The first term is the superposition of unit point source by the source function f . The
second term is the boundary term in the spatial variables. The third term is the boundary term
in the time variable.
Let us now construct the Green’s function in R3,1 . By symmetry considerations, we seek a
Green’s function of the form

G(~r, t; ~r ′ , t′ ) = G(~r − ~r ′ , t − t′ )

84
and the function G(~r, t) satisfying
 
1 ∂2
− ∇ G(~r, t) = δ(~r )δ(t).
2
c2 ∂t2
Furthermore, it is natural to assume G(~r, t) is spherically symmetric, and hence G(~r, t) = G(r, t)
only depends on the radius r = |~r | and time t. Then
 
1 ∂ 2 ∂G ∂ 2 G 2 ∂G 1 ∂2
∇ G= 2
2
r = + = (rG).
r ∂r ∂r ∂r2 r ∂r r ∂r2
We are reduced to solve  
1 1 ∂2 ∂2
− (rG) = δ(~r )δ(t).
r c2 ∂t2 ∂r2
At r > 0, we have the one dimensional wave equation
    
1 ∂2 ∂2 1∂ ∂ 1∂ ∂
− (rG) = − + (rG) = 0
c2 ∂t2 ∂r2 c ∂t ∂r c ∂t ∂r
which has two independent solutions of the form
1
G(∓) (r, t) = g± (t ± r/c)
r
for functions g± to be determined from the behavior at r = 0. Using
 
2 1
∇ = −4πδ(~r ),
r
we have the following singular behavior at r = 0
 
1 ∂2
− ∇ G(∓) (r, t) = 4πδ(~r )g± (t ± r/c).
2
c2 ∂t2
Comparing with the equation for Green’s function, we find
1
g± (u) = δ(u).
4π
We conclude that
1
δ(t ± r/c).
G(∓) (~r, t) =
4πr
Therefore we find two solutions of Green’s function
1
G(∓) (~r, t; ~r ′ , t′ ) = δ(t − t′ ± |~r − ~r ′ |/c).
4π|~r − ~r ′ |

3.4.3 Retarded and Advanced Solutions

Let us plug these Green’s functions into the above expression of ϕ satisfying the inhomo-
geneous wave equation
□ϕ = f.

We assume a spatially localized source and such that the spatial boundary term is absent. Thus
ˆ t2 ˆ
ϕ(~r, t) = dt ′
d3 r′ G(~r, t; ~r ′ , t′ )f (~r ′ , t′ )
t1 V
ˆ ′
1 3 ′
 ′ ′ ′ ′ ′ ′ ′ ′ 
 t =t2
+ 2 d r ϕ(~r , t )∂t G(~r, t; ~r , t ) − G(~r, t; ~r , t )∂t ϕ(~r , t ) 
′ ′ .
c V t′ =t1

85
There are two cases:
⃝
1 Retarded Solution: This is to use
1
G(+) (~r, t; ~r ′ , t′ ) = δ(t − t′ − |~r − ~r ′ |/c).
4π|~r − ~r ′ |
Note that in the expression for ϕ(~r, t), we have

t1 ≤ t ≤ t2

so G(+) is only nonzero at the t′ = t1 boundary. The retarded solution is therefore given by
ˆ
1 f (~r ′ , t − |~r − ~r ′ |/c)
ϕret (~r, t) = d3 r′ + ϕin (~r, t)
4π |~r − ~r ′ |
where ϕin (~r, t) collects the boundary term at t′ = t1 .
The term involving the source f has the following interpretation: ϕret (~r, t) is contributed
from source located at ~r ′ at time t − |~r − ~r ′ |/c. Such time delay reflects the propagating wave
signaled at the speed of light c.
Assume t1 is at the time before the source appears. Then at t ≤ t1 , the source integral
does not contribute and we have

ϕret (~r, t) = ϕin (~r, t), t ≤ t1 .

So physically, ϕin (~r, t) can be viewed as an “incoming wave” solution of the homogeneous wave
equation at the initial time t1 before the effect of the source. This is also consistent with the
fact that a general solution of the inhomogeneous wave equation is given by the sum of a special
solution (here is the source term) and a solution of the homogeneous wave equation (here is the
incoming wave).
⃝
2 Advanced Solution: This is to use
1
G(−) (~r, t; ~r ′ , t′ ) = δ(t − t′ + |~r − ~r ′ |/c).
4π|~r − ~r ′ |
The discussion is similar and we find the advanced solution
ˆ
1 f (~r ′ , t + |~r − ~r ′ |/c)
ϕadv (~r, t) = d3 r ′ + ϕout (~r, t).
4π |~r − ~r ′ |
The advanced solution at (~r, t) takes the effect of the source located at ~r at a future time
t + |~r − ~r ′ |/c. And ϕout is the “outgoing wave” describing the situation at time t2 after the
effect of the source.
Mathematically, both the retarded and the advanced waves are solutions. Physically, we
will pick the retarded one which reflects the causal structure.

Solving Maxwell’s Equations

We now use the retarded Green’s function to write down a solution of the inhomogeneous
wave equations 
□φ = ρ/ε0
□A
~ = µ ~j 0

86
It reads  ˆ

 1 ρ(~r ′ , t − |~r − ~r ′ |/c)

φ(~r, t) = d3 r′
4πε0 |~r − ~r ′ |
 ˆ ~ ′ ′

 ~ r, t) = µ0 d3 r′ j (~r , t − |~r − ~r |/c)
A(~
4π |~r − ~r ′ |
Let us check the Lorenz gauge condition
1 ∂ ~ = 0.
φ+∇·A
c2 ∂t
Let us write the above formula via Green’s function
 ˆ
 1
φ(~r, t) = d3 r′ dt′ G(+) (~r, t; ~r ′ , t′ )ρ(~r ′ , t′ )
ε0 ˆ
 ~ r, t) = µ0 d3 r′ dt′ G(+) (~r, t; ~r ′ , t′ )~j (~r ′ , t′ )
A(~

and using ε0 µ0 = 1/c2 , we find

ˆ h i
1 ∂ ~ 3 ′ ′ ′ ′ ′ ′ ′ ′ ~ ′ ′
φ + ∇ · A = µ 0 d r dt ∂ t G (+)
(~
r , t; ~
r , t )ρ(~
r , t ) + ∇ · G (+)
(~
r , t; ~
r , t ) j (~
r , t )
c2 ∂t
ˆ h i
= −µ0 d3 r′ dt′ ∂t′ G(+) (~r, t; ~r ′ , t′ )ρ(~r ′ , t′ ) + ∇′ · G(+) (~r, t; ~r ′ , t′ )~j (~r ′ , t′ )
ˆ  
= µ0 d3 r′ dt′ G(+) (~r, t; ~r ′ , t′ ) ∂t′ ρ + ∇′ · ~j = 0

via charge conservation: ∂t ρ + ∇ · ~j = 0.

3.5 Dipole Radiation

We have seen from electrostatics and magnetostatics that

stationary charges =⇒ electric field

steady current =⇒ magnetic field
(charges moving in a constant speed)

We have also learned that Maxwell’s equations admit solutions by electromagnetic waves. We
now explain
accelerating charges =⇒ radiation

In other words, accelerating charges will generate propagating electromagnetic waves.

3.5.1 Spherical Wave

Let us first describe waves which are spreading out from some center (say the origin) and
are spherically symmetric. In other words, we consider a solution ϕ of

□ϕ = 0

87
and such that ϕ = ϕ(r, t) where r = |~r |. We have seen that for spherically symmetric functions,
the wave equation becomes  
1 1 ∂2 ∂2
− (rϕ) = 0.
r c2 ∂t2 ∂r2
A general retarded solution is given by
f (t − r/c)
ϕ(r, t) = .
4πr
It represents a spherical wave travelling outward from the origin. Strictly speaking, ϕ solves the
wave equation outside the origin but would have singularity at the origin due to the presence
of source. In fact, from the general solution of inhomogeneous wave equation with source S
ˆ
1 S(~r ′ , t − |~r − ~r ′ |/c)
ϕ(r, t) = d3 r′ ,
4π |~r − ~r ′ |
we see that the source is located at the origin given by

S(~r, t) = δ(~r )f (t)

so in fact ϕ solves
□ϕ = δ(~r )f (t).

Another thing worth mentioning is that the amplitude of ϕ decays in proportion to 1/r as
the wave propagates. This is consistent with energy conservation. As the wave propagates, the
total energy flux over the sphere of radius r must be the same. The area of the sphere is 4πr2 ,
and the energy density depends on the square of the wave amplitude, so the amplitude of the
wave must decrease as 1/r.

3.5.2 Electric Dipole Radiation

Consider the following retarded solution

ˆ
1 f (~r ′ , t − |~r − ~r ′ |/c)
ϕ(~r, t) = d3 r′ .
4π |~r − ~r ′ |
Let us assume the source f is localized in some region V , and we look ϕ at a distance far away
from the region. Then we have the multipole expansion

1 1 ~r · ~r ′
= + 3 + ··· for r >> 0.
|~r − ~r ′ | r r
We also assume that the source does not vary too fast: the motion of charges and currents are
non-relativistic, and they do not change very much over the time that it takes light to cross the
r′
region V . This in particular says that the operation c ∂t would produce something very small.
′ r′
r·⃗
Then we can use |~r − ~r | = r − ⃗
r + · · · to expand
 
′ ′ ′ ~r · ~r ′
f (~r , t − |~r − ~r |/c) = f ~r , t − r/c + + ···
rc
~r · ~r ′
= f (~r ′ , t − r/c) + ∂t f (~r ′ , t − r/c) + · · · .
rc

88
 Therefore ϕ is approximated by
ˆ   
1 3 ′ ~r · ~r ′ ′ ~r · ~r ′ ′
ϕ(~r, t) ' d r 1 + 2 + ··· f (~r , t − r/c) + ∂t f (~r , t − r/c) + · · ·
4πr r rc
ˆ  
1 3 ′ ′ ~r · ~r ′ ′ ~r · ~r ′ ′
= d r f (~r , t − r/c) + ∂t f (~r , t − r/c) + 2 f (~r , t − r/c) + · · · .
4πr rc r

Let us apply this to Maxwell’s equations with charge distribution ρ and current distribution ~j.
The vector potential
ˆ ~j (~r ′ , t − |~r − ~r ′ |/c)
~ r, t) = µ0
A(~ d3 r′
4π |~r − ~r ′ |
has the leading approximation (under the same assumption as above)
ˆ
~ r, t) ' µ0
A(~ d3 r′ ~j (~r ′ , t − r/c).
4πr
Consider the x-component for example, we have
   
jx (~r ′ , t′ ) = ∇′ · x′ ~j (~r ′ , t′ ) − x′ ∇′ · ~j (~r ′ , t′ ) = ∇′ · x′ ~j (~r ′ , t′ ) + x′ ∂t′ ρ(~r ′ , t′ ).

Therefore ˆ ˆ
d3 r′ jx (~r ′ , t − r/c) = d3 r′ x′ ∂t ρ(~r ′ , t − r/c).

The y and z components are similar. Thus

ˆ ˆ
µ0 µ0 d µ0 ˙
d3 r′ ~j ′ (~r ′ , t − r/c) = d3 r′ ρ(~r ′ , t − r/c)~r ′ = p~ (t − r/c)
4πr 4πr dt 4πr
´
where p~ (t) = d3 r′ ρ(~r ′ , t)~r ′ is the electric dipole moment. So the leading approximation of
the vector potential is
~ r, t) ' µ0 p~˙ (t − r/c).
A(~
4πr
This is called the electric dipole approximation.
We can compute the approximated magnetic field by

~ ' − µ0 r̂ × p~˙ (t − r/c) − µ0 r̂ × p~¨ (t − r/c)

~ =∇×A
B
4πr2 4πrc
where r̂ = ⃗rr . The second term is the leading contribution far away, so

~ ' − µ0 r̂ × p~¨ (t − r/c).

B
4πrc
We next compute the approximated electric field. The scalar potential
ˆ
1 ρ(~r ′ , t − |~r − ~r ′ |/c)
φ(~r, t) = d3 r′
4πε0 |~r − ~r ′ |
ˆ  
1 3 ′ ′ ~r · ~r ′ ′ ~r · ~r ′ ′
= d r ρ(~r , t − r/c) + ∂t ρ(~r , t − r/c) + 2 ρ(~r , t − r/c) + · · ·
4πε0 r rc r
 
1 r̂ r̂ · p~ (t − r/c)
= Q + · p~˙ (t − r/c) + + ··· .
4πε0 r c r

89
Here Q is the total charge. The leading approximation of the electric field is therefore (keeping
only the 1/r-order term)

~
~ = −∇φ − ∂t A
E
r̂ r µ0 ¨
' ¨
· p~ (t − r/c)∇ − p~ (t − r/c)
4πε0 rc c 4πr
µ0 h  i
= r̂ · p~¨ r̂ − p~¨ (r̂ · r̂)
4πr  
µ0
= r̂ × r̂ × p~¨ (t − r/c) .
4πr
As a consistency check, we can also use Maxwell’s equations. The approximated electric field
~ can be computed via that of B
E ~ by

∂~ ...

~ ' µ0 r̂ × r̂ × p~ (t − r/c) ,
E = c2 ∇ × B
∂t 4πr
so
 
~ ' µ0 r̂ × r̂ × p~¨ (t − r/c) .
E
4πr
Let us denote this leading electric dipole approximation by


B~ ED = − µ0 r̂ × p~¨ (t − r/c)
4πrc
 µ  
E~ ED = 0 r̂ × r̂ × p~¨ (t − r/c)
4πr
They are both spherical waves and related by

~ ED = −cr̂ × B
E ~ ED .

~ ED are perpendicular to the propagating direction and decays as

~ ED and B
In particular, both E
1/r.

r̂

~ ED
E
~ ED
cB

The power radiated from the source can be computed by the Poynting vector

~= 1E
S ~ ' 1E
~ ×B ~ ED × B
~ ED = µ0
|r̂ × p~¨ |2 r̂
µ0 µ0 16π 2 r2 c
which points in the same direction as ~r. We conclude that oscillating dipole is emitting spherical
electromagnetic waves that transporting power radially.
We can also study higher order approximations. For example. the next order are magnetic
dipole and electric quadrupole radiations. We will not discuss them in this note.

90
 Let us assume the dipole is oscillating along the z-direction:

p~ = p(t)ẑ.

In spherical coordinate (r, θ, φ),

|r̂ × ẑ| = sin θ.

dipole
θ ~r

Then
µ0
~=
S |p̈|2 sin2 θr̂
16π 2 r2 c
which is largest in the direction perpendicular to the dipole (when θ = π/2) and smallest in the
direction parallel to the dipole (when θ = 0, π).
The total radiated power is computed by
ˆ ˆ 2π ˆ π
µ0 µ0 2
P = d~σ · S =
~
2
|p̈| 2
dφ dθ sin3 θ = |p̈| .
S2 16π c 0 0 6πc
Example 3.5.1. Consider a particle of charge Q oscillating in the z-direction with frequency
ω and amplitude d. The electric dipole moment is

p~ = p cos(ωt)ẑ where p = Qd.

−d

Then
p~¨ = −ω 2 p cos(ωt)ẑ

and the total radiated power is

µ0 p2 ω 4
P (t) = cos2 (ωt).
6πc
The time-averaged power is
ˆ 2π/ω
1 µ0 p2 ω 4
hP i = P (t)dt = .
2π/ω 0 12πc

This is called the Larmor’s formula.

91
 3.6 Moving Point Charge
Let us consider a point particle with charge q moving in the trajectory

~ = position of q at time t.
ξ(t)

~ is general, the point charge will accelerate along the way and radiate electromagnetic
Since ξ(t)
waves. We would like to calculate the electromagnetic fields produced by such a moving point
charge.
The charge and current densities are
  

ρ(~r, t) = q δ 3 ~r − ξ(t)~
 

~j (~r, t) = q~v (t) δ 3 ~r − ξ(t)
~

~˙ is the velocity of the particle.

where ~v (t) = ξ(t)
The scalar and vector potentials are then given by
  

 ˆ δ 3 ~ r ′
− ~ − |~r − ~r ′ |/c)
ξ(t

 q
φ(~r, t) =
 d3 r′
4πε0 |~r − ~r ′ |
 

 ˆ − |~ − ′
|/c) 3 ~ − ~ − |~r − ~r ′ |/c)

 ~
v (t r ~
r δ r ξ(t
~ r, t) = qµ0 d3 r′
A(~

4π |~r − ~r ′ |
Let us first consider the scalar potential φ(~r, t). To deal with the δ-function, let us first rewrite
 

ˆ ˆ δ 3 ~ r ′
− ~ ′ ) δ t′ − (t − |~r − ~r ′ |/c)
ξ(t
q
φ(~r, t) = dt′ d3 r′
4πε0 |~r − ~r ′ |
 
ˆ δ t ′ − t + |~ r − ~ ′ )|/c
ξ(t
q
= dt′ .
4πε0 ~ ′ )|
|~r − ξ(t
Recall we have the following δ-function relation
X δ(x − xi )
δ(f (x)) =
xi
|f ′ (xi )|

where the sum is over all roots f (xi ) = 0. In fact, this is about change of variable formula:
ˆ X
g(x)δ(f (x))df = g(xi )
xi

and ˆ ˆ
g(x)δ(f (x))df = g(x)|f ′ (x)|δ(f (x))dx.

Comparing the above two expressions, we get the above δ-function relation.
Let us now apply this to δ(f (t′ )) where

f (t′ ) = t′ − t + |~r − ξ(t

~ ′ )|/c.

A root of f (t′ ) describes a time t′ such that

~ ′ )| = c(t − t′ ).
|~r − ξ(t

92
 ~r
c(t − t′ )

t′
ξ~
t

~ ′ ) is at some time t′ earlier than t

Such a root is called a retarded time, where the point ξ(t
such that the light sent at this point will travel to ~r at time t. Such retarded time is unique.
In fact, assume we have two retarded times t′1 and t2 both solving f (t′1 ) = f (t′2 ) = 0.

~r
c(t − t′1 )

c(t − t′2 )

t′1
ξ~
t′2 t

Assume t′1 < t′2 ≤ t. Then the length of the curve

 !

length ξ~  ≥ c(t − t′1 ) − c(t − t′2 ) = c(t′2 − t′1 ).
′ ′
[t1 ,t2 ]

We conclude that the speed of the particle ≥ c. Since no charged particle can travel at the speed
of light, this is a contradiction. The existence of retarded time can be achieved by chasing back
along the trajectory for long enough until |~r − ξ(t~ ′ )| = c(t − t′ ) is satisfied since the charged
particle travels relatively smaller than the speed of light.
Let us denote tret for the the retarded time solving the equation

~ ret )|/c = 0
tret − t + |~r − ξ(t

with the understanding that the dependence of tret on (~r, t) is implicit when it is clear from the
context. To simplify notation, let us also write

~
R(t) ~
= ~r − ξ(t), R(t) = |R(t)|,
~ ~
n̂R = R/R, ~ = ~v (t)/c,
β(t) ~
β(t) = |β(t)|.

~
R(t) ~ of the particle to the point ~r of the field.
represents the relative vector from the point ξ(t)
Then
1
δ(f (t′ )) = δ(t′ − tret ).
~
1 − R(tret ) · β(tret )/R(tret )
~
Plugging this into the potential, we find
 
1 q
φ(~r, t) = .
~ · β~ ret
4πε0 R − R

93
Here [−]ret means the time variable is evaluated at tret . By a similar argument, we find the
vector potential  
~ µ0 q~v
A(~r, t) = .
4π R − R~ · β~ ret
The above two expressions are called the “Liénard-Wiechert potentials” for a moving point
charge.

Example 3.6.1 (Point charge of constant velocity). Consider

~ = ~v t
ξ(t) with ~v constant.

The retarded time is computed via

|~r − ~v tret | = c(t − tret ).

Geometrically, let C represent the point at time t, and B represent the point at time tret .

D h

θ
B C ~v
tret t

Then
|AB| = |~r − ~v tret |, |BC| = v(t − tret ), where v = |~v |.

The above equation becomes

|AB| c 1
= = .
|BC| v β
It is not hard to see that
s  
p |BC| 2
~ ret ) · β~ = |AB| − |BD| = |AD| = |AC|2 − |CD|2 =
R(tret ) − R(t |AC|2 − h ·
|AB|
q
= |AC| 1 − β 2 sin2 θ.

Therefore 


1
p
q

φ(~r, t) = 4πε0 R 1 − β 2 sin2 θ

 ~ r, t) = µ0 p ~v

A(~ 4π R 1 − β 2 sin2 θ
~ = ~r −~v t is the relative vector from the present position of the particle to the field point,
Here R
and θ is the angle between R ~ and ~v . Note that for nonrelativistic velocities (v << c),

1 q
φ(~r, t) ' .
4πε0 R

94
 We now move on to compute the electromagnetic fields

E ~ = −∇φ − ∂t A~
B ~
~ =∇×A

Since the retarded time tret depends on ~r and t implicitly, we work directly with the expression
 ˆ
 q δ (t′ − t + R(t′ )/c)

φ(~r, t) = 4πε dt′
0 R(t′ )
ˆ ′ ′ ′

 ~ r, t) = qµ0 dt′ ~v (t )δ (t − t + R(t )/c)
A(~
4π R(t′ )
Using
 
′

′ ∂ ′ ′


∇R = R/R
~ = n̂R , ∇δ t − t + R(t )/c = ∇(R/c) − δ t − t + R(t )/c ,
∂t
we find
~ = −∇φ − ∂t A
E ~
ˆ  
q ′ ∇R(t )
′
′ ′

∇R(t′ ) ∂ ′ ′


= dt δ t − t + R(t )/c + δ t − t + R(t )/c
4πε0 R(t′ )2 cR(t′ ) ∂t
ˆ
q ∂ ~v (t′ )δ (t′ − t + R(t′ )/c)
− dt′
4πε0 ∂t c2 R(t′ )
" # " #
q ~
R q ∂ R/R ~ − ~v /c
= +
4πε0 R3 g 4πε0 c ∂t Rg
ret ret
  " #
q n̂R q ∂ n̂R − β~
= + .
4πε0 R2 g ret 4πε0 ∂t Rgc
ret
⃗ v
where g = 1 − n̂R · β~ = 1 − R·⃗
Rc . To compute t-derivative term, we use

tret − t + R(tret )/c = 0

to find
dt 1 ∂R
=1+ (tret ) = g(tret ).
dtret c ∂t
From R ~
~ = ~r − ξ(t), we have

∂R~ ∂R
= −~v = −cβ,~ ~
= −cn̂R · β,
∂t ∂t
! ! !
∂ n̂R 1 ∂R~ ~
R ∂ ~
R 1 ∂ ~
R ∂ ~
R c   
= − 3 R ~· = − n̂R · n̂R = n̂R · β~ n̂R − β~
∂t R ∂t R ∂t R ∂t ∂t R
c   c  
= n̂R × n̂R × β~ = ~ − c gn̂R ,
n̂R − β
R R R
∂Rg ∂   ~ ~
= R−R ~ · β~ = −cn̂R · β~ + cβ 2 − R ~ · dβ = −cn̂R · β~ + cβ 2 − R n̂R · dβ .
∂t ∂t dt dt
This enables us to compute
" # " # " !# " !#
∂ n̂R − β~ dtret ∂ n̂R − β~ 1 ∂ n̂R − β~ 1 ∂ n̂R − β~
= = =
∂t Rgc dt ∂tret Rgc g (tret ) ∂t Rgc g ∂t Rgc
ret ret ret ret

95
and
   
q  n̂R
∂
∂t n̂R − β~ ~
n̂R − β ∂
~ =
E + − 2 3 (Rg)
4πε0 R2 g Rg 2 c R g c ∂t
ret
" !#
q n̂R n̂R − β ~ n̂R dβ/dt n̂R − β~
~ dβ~
= + − 2 − − 2 3 −cn̂R · β~ + cβ − R
2 ~·
4πε0 R2 g R2 g 2 R g Rg 2 c R g c dt
ret
" ! #
n̂R − β~   ~ ~ ~
=
q
g + n̂R · β~ − β 2 + n̂R − β n̂R · dβ − dβ/dt
4πε0 R g 2 3 Rg 3 c dt Rg 2 c
ret
    ⃗

~ ~ dβ
q  n̂R − β (1 − β ) n̂R × n̂R − β × dt 
2
= + .
4πε0 R2 g 3 Rg 3 c
ret

That is,     
⃗
dβ
~ (1 − β 2 )
n̂R − β n̂R × n̂R − β~ ×
~ = q  dt

E + .
4πε0 R2 g 3 Rg 3 c
ret
A similar computation leads to
~ = 1 [n̂R ] × E.
B ~
ret
c
These are called Liénard-Wiechert electric and magnetic fields.
The above expression decomposes the electric and magnetic field into a velocity field and
~ =E
an acceleration field: E ~v +E~ a where
   

 n̂ − ~ (1 − β 2 )
β

E~ q  R


 v = 4πε

 0 R2 g 3
   ret 
⃗ 

 × − ~ × dβ

~ q  R n̂ n̂ R β

Ea =
dt


 4πε0 Rg c3
ret

~ v only depends on the velocity β~ and decays as 1/R2 in space. This field and its energy is
E
~ a contains β~˙ which is about the acceleration. It decays as 1/R in
attached to the particle. E
space, and therefore contains surface energy that propagates to infinity. This is the radiation
field which contain energy that radiates from the particle to the space faraway.

~ = ~v t.
Example 3.6.2. We again look at a point charge of constant velocity with ξ(t)

D h

θ
B C ~v
tret t

96
The particle is not accelerating and
     
n̂ R − ~ (1 − β 2 )
β ~ − Rβ~ (1 − β 2 )
R
~ =E
E ~v = q   =
q   .
4πε0 R2 g 3 4πε0 R3 g 3
ret ret

We have seen in the previous example that

1/2
[Rg]ret = R 1 − β 2 sin2 θ ,
h i −−→ −−→ −→ ~
R ~
~ − Rβ = BA − BC = CA = R.
ret
It follows that
~ = q 1 − v 2 /c2 ~
R
E

4πε0 1 − v 2 sin2 θ /c2 3/2 R3
which remains in the radial direction, but is stronger in the direction perpendicular to the
moving direction and weaker in the direction parallel to the moving direction.
For example, assume the particle is moving along x-direction with

~ = (vt, 0, 0).
ξ(t)

~ = (Ex , Ey , Ez )
Then the above formula leads to the explicit expression in components E

 q γ(x − vt)
Ex =


 4πε0 (γ 2 (x − vt)2 + y 2 + z 2 )3/2



q γy
Ey =

 4πε0 (γ 2 (x − vt)2 + y 2 + z 2 )3/2



 q γz

Ez =
4πε0 (γ (x − vt) + y 2 + z 2 )3/2
2 2

where γ = √ 1
. This expression has a very suggestive relativistic meaning. Indeed, we
1−v 2 /c2
will show in Example 5.2.5 that this formula can be simply obtained from the Lorentz boost
along x-direction of the fields of a stationary point charge.

3.7 Scattering
We briefly discuss the basic idea behind the scattering of electromagnetic waves. Its physics
contains several steps:

Incident electromagnetic wave hits the particle

↓
Particle oscillates and accelerates
↓
Radiate scattering electromagnetic wave
The ratio
scattered power
σ=
incident power per unit area
is called the cross-section for scattering.

97
Thomson Scattering
~ 0 cos(~k · ~r − ωt) interacting with a free particle of
~ =E
Consider an incoming plane wave E
mass m and charge q. The equation of motion is
 
m~r¨ = q E~ + ~r˙ × B
~ .

When the particle speed is non-relativistic, |~r˙ |/c << 1,

˙ 
 ~r 
~ =  × cB
|~r˙ × B| ~  << |cB|
~ = |E|,
~
c 

so we can neglect the magnetic Lorentz force. In the non-relativistic limit, we can also assume the
oscillation of the particle is small comparing with the wave length. Under these simplification,
we have
m~r¨ = q E
~ 0 cos ωt

which can be solved by

~0
qE
~r = − cos ωt.
mω 2
We use the electric dipole approximation and the Larmor’s formula as discussed in Section
3.5.2 to compute the time-averaged radiation power
h  i2
qE0
µ0 q mω 2 ω4 µ0 q 4 E20
hPrad i = = .
12πc 12πm2 c
On the other hand, we have computed the Poynting vector for the plane wave (Example 3.1.5)
2  
~inc = E0 cos2 ~k · ~r − ωt k̂
S
cµ0

whose time average over a single period 2π/ω is

D E E20
Sbinc = .
2cµ0
The cross-section is given by

hPrad i µ2 q 4
σ=D E = 0 2.
Sbinc 6πm

Note that the cross-section of Thomson scattering does not depend on the frequency ω. All
wave lengths of light are scattered equally.

Rayleigh Scattering

Rayleigh scattering describes the scattering of electromagnetic waves on a neutral molecule

or a small dielectric object. The object exhibits electric dipole polarization under the effect of
the electric field
P~ = αE.
~

98
 ~ 0 cos(~k · ~r − ωt) as
~ =E
We assume linear polarization so α is a constant. For the plane wave E
above, Larmor’s formula gives the time-averaged radiation power

µ0 α2 E20 ω 4
hPrad i = .
12πc
The cross-section is given by
hPrad i µ2 α 2 ω 4
σ=D E= 0
Sbinc 6π

which is stronger for high frequencies or short wave lengths. This explains blue sky in the day
time and red sky at sunset since
λblue < λred .

During the day, we look away from the sun and see light that has scattered by the atmosphere.
At sunset, we look directly at the sun and see light that remains from the scattering.

99
 Chapter 4 U (1) Gauge Theory

There are four fundamental interactions in nature: the weak nuclear force, the strong
nuclear force, the electromagnetic force, and the gravity. One landmark of modern physics is
to realise that all these interactions have a common feature in terms of gauge principle.
The notion of gauge transformation and gauge invariance was introduced by Hermann
Weyl in his attempt to unify gravitation and electromagnetism via a geometric framework.
Weyl emphasized the role of gauge invariance as a symmetric principle, and his proposal forms
the foundation of what is now known as gauge theory.
The modern aspect of gauge theory has rich content both in mathematics and physics. We
are not intended to present a full overview of this beautiful theory. Instead, we aim to put
hands on one basic example, the electromagnetism, as a U (1)-gauge theory from the modern
geometric perspective. We assume basic knowledge on the notion of manifolds, and readers
can consult [10, Chapter 3] in this series for preliminary geometric backgrounds. We will give
a self-contained discussion on the geometry of fiber bundles in order to understand the bridge
between the physical content in the first half of this note and the mathematical content that
constitutes the modern framework. We will briefly mention and comment on generalizations,
such as non-abelian gauge theory or Yang-Mills theory, at certain steps along the way.

4.1 Fiber bundle

4.1.1 Fiber Bundle

The notion of fiber bundle describes a family of geometric object (called the fiber) varying
with respect to a parameter space (called the base). Precisely, it consists of the following
geometric data:

• a manifold B, called the base manifold

• a manifold F , called the fiber

• a manifold E, called the total space

• a Lie group G, called the structure group.

Their relationships are described by the following:

⃝
1 a surjective map π : E → B which is locally trivial with fiber F . This means that there
exists a covering {Uα } of B such that π −1 (Uα ) is diffeomorphic to the product Uα × F via

100
 φα
π −1 (Uα ) Uα × F

π πα
Uα
and this diagram is commutative, i.e. πα ◦ ϕα = π. Here πα : Uα × F → Uα is the projection to
the Uα -factor. This property says that for each x ∈ Uα , ϕα defines a diffeomorphism
∼
ϕα : π −1 (x) → F

so every preimage π −1 (x) of a point x ∈ B is diffeomorphic to F . The total space looks like

π −1 (x) ' F

F F

π π

B x

so π : E → B can be viewed as a family of the fiber manifold F varying over the base B. The
locally trivial condition says that locally it can be parameterized as a trivial family. However,
globally E may not be the same as B × F : nontrivial topological phenomenon can arise.
⃝
2 G is equipped with an action on F

ρ : G × F −→ F

which we denote simply by g · u ∈ F for the action of g ∈ G on u ∈ F . Each g gives a way to

identify F with itself. Equivalently, we have a group homomorphism

ρ : G −→ Diff(F )

where Diff(F ) denotes the diffeomorphism group of F .

⃝
3 We demand compatibility between ⃝ 1 and ⃝.
2 Precisely, let

ϕα : π −1 (Uα ) −→ Uα × F
ϕ : π −1 (U ) −→ U × F
β β β

be two local trivializations as in ⃝.

1 Then on the intersection Uαβ := Uα ∩ Uβ , we have
φα φβ
Uαβ × F π −1 (Uαβ ) Uαβ × F

Uαβ

101
By definition, the composition

ϕβα := ϕβ ◦ ϕα −1 : Uαβ × F −→ Uαβ × F

is a fiberwise diffeomorphism, i.e. for each x ∈ Uαβ , ϕβα maps {x} × F to {x} × F and defines
a diffeomorphism of F under the canonical identification {x} × F ' F . Equivalently, we can
write
ϕβα : Uα ∩ Uβ −→ Diff(F ).

Such ϕβα is called the translation function, which characterizes the difference of identifying
the fiber π −1 (x) ' F under two trivializations. In fact, for x ∈ Uαβ , we have
ϕβα

ϕα ϕβ

ϕα ϕβ F
F
F π −1 (x) F

ϕβα

Then we require that each ϕβα (x) is realized by a group action by g ∈ G on F :

ϕβα (x) · u = g(x) · u ∀u ∈ F

for some g(x) ∈ G. We say the transformation has the structure captured by the Lie group G,
and this is why G is called the structure group. We will simply write ϕβα (x) ∈ G. Then the
transformation function is required to be expressed as

ϕβα : Uα ∩ Uβ −→ G

with the understanding that its actual transformation on the fiber F is realized via the G-action
ρ
Uα ∩ Uβ −→ G −→ Diff(F ).

This requirement clearly puts further constraint on the trivializations {Uα , ϕα }, and in this case
we say the fiber bundle has structure group G.
We usually denote a fiber bundle by

F E
π

B
and specify the structure group G in the context when we need to.

102
Definition 4.1.1. π : E → B is called a trivial bundle if there is a global trivialization
≃
E B×F

B
so we can identify E as the product B × F .

Theorem 4.1.2. If B is contractible, then any fiber bundle E over B is a trivial bundle.

In particular, any fiber bundle over Rn is trivial.

j
Definition 4.1.3. A (smooth) section of the fiber bundle π : E → B over a subspace U ,→ B
is a (smooth) map s : U → E such that the following diagram is commutative

E
s π i.e. j = π ◦ s.
j
U B
In other words, a section s over U assigns every point x ∈ U an element s(x) of the fiber
π −1 (x). We will mainly consider smooth sections without further specification in this note.

We denote
Γ(U, E) = {sections over U}.

Elements of Γ(B, E) are also called global sections.

Example 4.1.4. If E = B × F is trivial, then

Γ(B, E) = Map(B, F )

can be identified with smooth maps from B to F .

s : B −→ E = B × F
b 7−→ (b, f (b))

for f : B → F .

Example 4.1.5 (Möbius strip). Möbius strip is obtained by gluing the two ends of a strip with
a half-twist, as shown below

103
 M

glue two ends oppositely

as the arrows suggest π

S1

Mathematically, the Möbius strip is described by the quotient

M = (I × I) / ∼ = {(t, x) | t, x ∈ I = [0, 1]}/ ∼

where the equivalence relation is to identify

(0, x) ∼ (1, 1 − x) for x ∈ I = [0, 1].

It can be viewed as a fiber bundle over S 1

π : M −→ S 1
(t, x) 7−→ t

where we identify S 1 = I/{0, 1}.

It is not hard to see that for any small open interval U ⊂ S 1 , π −1 (U) can be trivialized as
U × I by “straightening”. We leave the details to the reader. So
I M
π

S1
is a fiber bundle with fiber I = [0, 1].
By construction, it is clear that global sections of this fiber bundle can be identified with

Γ(S 1 , M ) = {f : [0, 1] → [0, 1] | f (1) = 1 − f (0)}.

The Möbius strip is a nontrivial fiber bundle. To see this, assume M ' S 1 × I is trivial,
then we can find two global sections f1 , f2 such that their images in M do not intersect. On
the other hand, as described above, f1 and f2 can be identified with two functions

fi : [0, 1] −→ [0, 1]

104
such that fi (1) = 1 − fi (0). Assume f2 (0) > f1 (0). Then f2 (1) < f1 (1). By Intermediate Value
Theorem, there must be some point t such that

f2 (t) = f1 (t).

So these two sections must intersect. Contradiction.

f2
f1
t

This proves that π : M → S 1 is nontrivial.

Example 4.1.6 (Hopf fibration). Consider the map

π : S 3 −→ S 2

defined as follows. Let us use complex numbers to identify

S 3 = {(z1 , z2 ) ∈ C2 | |z1 |2 + |z2 |2 = 1}

S 2 = CP 1 = C2 − {0} / ∼

A point of S 2 = CP 1 is described by homogeneous coordinates [z1 , z2 ], where we identify

[λz1 , λz2 ] = [z1 , z2 ] for all λ ∈ C∗ . Then the map π is expressed as

π : S 3 −→ S 2
(z1 , z2 ) 7−→ [z1 , z2 ]

For each point p ∈ S 2 , we have π −1 (p) ' S 1 . For example, consider the north pole [1, 0], then

π −1 ([1, 0]) = {(z, 0) | z ∈ C, |z|2 = 1}.

It can be checked that

S1 S3
π

S2

is a fiber bundle with fiber S 1 . This is called the “Hopf fibration”. This is a nontrivial fibration
since S 3 and S 2 × S 1 are topologically different. For example,

π1 (S 3 ) = 1
π (S 2 × S 1 ) = Z
1

i.e. any loop in S 3 can be shrinked continuously to a point, but a loop wrapping along the S 1
factor of S 2 × S 1 can not.

105
4.1.2 Vector Bundle

Definition 4.1.7. A real vector bundle, or simply vector bundle, is a fiber bundle E with
fiber F = Rm and structure group G ⊂ GLm (R). The integer m is called the rank of the vector
bundle, and we denote
rank(E) = m.

Similarly, a complex vector bundle of rank m is a fiber bundle with fiber Cm and structure
group G ⊂ GLm (C).

One important fact about vector bundle is that each fiber π −1 (x) for x ∈ B is a linear
vector space, so we can add elements of π −1 (x) and multiply by a scalar. In fact, let {Uα , ϕα }
be a trivialization of the vector bundle E. Assume x ∈ Uα . Then we can use ϕα to identify

ϕα : π −1 (x) −→ Rm .

This allows us to define the linear structure on π −1 (x) by

λ1 s1 + λ2 s2 := ϕ−1
α (λ1 ϕα (s1 ) + λ2 ϕα (s2 ))

where λi ∈ R, s1 ∈ π −1 (x). It does not depend on the choice of the local trivialization. Let
x ∈ Uβ and
ϕβ : π −1 (x) −→ Rm

be another identification of the fiber π −1 (x) via a different trivialization ϕβ . Then the transition

φ−1
Rm π −1 (x)
α

φβ
g

Rm

is a linear map g ∈ GLm (R). Therefore

g (λ1 ϕα (s1 ) + λ2 ϕα (s2 )) = λ1 g(ϕα (s1 )) + λ2 g(ϕα (s2 )),

i.e.
−1
ϕ−1
α (λ1 ϕα (s1 ) + λ2 ϕα (s2 )) = ϕβ (λ1 ϕβ (s1 ) + λ2 ϕβ (s2 )) .

This implies the linear operation λ1 ϕα (s1 ) + λ2 ϕα (s2 ) is intrinsically defined on the fiber of a
vector bundle, and it does not depend on the choice of the local trivialization.
Similarly, let U ⊂ B be an open subset. Then the space of sections of the vector bundle on
U has a structure of C ∞ (U )-module:

Γ(U, E) is a C ∞ (U ) − module.

For any two sections s1 , s2 ∈ Γ(U, E) and any two functions f1 , f2 ∈ C ∞ (U ), we have

s = f1 s1 + f2 s2 ∈ Γ(U, E)

106
by defining the value of s(x) at each x ∈ U via

s(x) = f1 (x)s1 (x) + f2 (x)s2 (x).

In particular, global sections Γ(B, E) is a C ∞ (B)-module.

s multiply fs

by a function f

π
Definition 4.1.8. Let E → B be a vector bundle of rank m, and U ⊂ B be an open subset. A
set of m sections {s1 , · · · , sm } of E over U is said to be a frame of E over U if

{s1 (x), · · · , sm (x)} form a basis of π −1 (x)

for each x ∈ U.

Assume {s1 , · · · , sm } is a frame of E over U. Then it allows us to define a local trivialization

of E over U
ϕ : π −1 (U) −→ U × Rm

by defining for any point e ∈ E, π(e) = x ∈ U, via

ϕ(e) = {x} × (a1 , · · · , am )

where e = a1 s1 (x) + · · · + am sm (x), ai ∈ R. Conversely, any local trivialization of E over U

gives rise to a frame of E over U, essentially by the same formula above. Therefore a frame of
E may not exist globally on B (unless E is trivial), but always exist locally.

Example 4.1.9 (Tangent bundle). Let X be a smooth manifold of dim = n. We can define
the tangent bundle which has rank n as follows. Set-theoretically, T X is the union
[
TX = Tp X
p∈X

where Tp X is the space of tangent vectors at p ∈ X.

107
 Tp X

Local trivializations of T X can be constructed with the help of local coordinates. Let

x1 , · · ·
, xn be a local coordinate system on an open subset U ⊂ X. Then we have a frame of
T X over U by the local vector fields
∂ ∂
1
,··· , n on U.
∂x ∂x
A section V of T X over U is the same as a vector field on U, which can be expanded via the
frame as
X
n
∂
V = V i (x) , x ∈ U.
∂xi
i=1

V (x)
X


If y 1 , · · · , y n is another choice of local coordinates on U′ , then we have the coordinate
transformation
(x1 , · · · , xn ) 7−→ (y 1 , · · · , y n ) on U ∩ U′ .

It gives rise to a linear transformation of the frame via the chain rule

∂ X ∂y j ∂
n
= .
∂xi ∂xi ∂y j
j=1
 
∂y j
In particular, the matrix ∂xi
is precisely the transition map between two local trivializations
of T X. We can expand the same vector field V in both coordinates on the intersection U ∩ U′
X
n
∂  i
V = V i (x) in x coordinates
∂xi
i=1

X
n
∂  j
V = Ve j (y) j in y coordinates
∂y
j=1

108
Then their coefficients are related via the transformation rule of the frame by
X
n
∂y j
Ve j (y(x)) = V i (x) .
∂xi
i=1

Example 4.1.10 (T S 2 ). Consider the unit sphere S 2 ⊂ R3 . Let p ∈ S 2 . Then Tp S 2 can be

identified with vectors in R3 orthogonal to p.

p Tp S 2

The bundle space of T S 2 can be identified as

n o
(~r, ξ~ ) ∈ T S 2 = (~r, ξ~ ) ∈ R3 × R3 | |~r | = 1, ~r · ξ~ = 0

~r ∈ S2

T S 2 is a nontrivial vector bundle: there can not exist a global frame on S 2 . In fact, Hairy Ball
Theorem says that any global vector field on S 2 must vanish at some point, thus can not be
part of a frame everywhere on S 2 .

4.1.3 Principal Bundle

Definition 4.1.11. A principal G-bundle is a fiber bundle with fiber F = G being a Lie
group G. The structure group is a subgroup of G and its action on F is given by the left
multiplication
G × F (= G) −→ F (= G)
(g, h) 7−→ g · h

We denote a principal G-bundle P over B as

G P
π

The fact that G admits both left and right G-action allows us to define a fiberwise right G-action
on P
P × G −→ P
(p, g) 7−→ p · g

109
as follows. Let b ∈ B and ϕα be a local trivialization. Then ϕα gives a diffeomorphism

∼
ϕα : π −1 (b) −
→ G.

Then for any point p ∈ π −1 (b), we define

p · g := ϕ−1
α (ϕα (p) · g)

under the above identification. This does not depend on the choice of the local trivialization.
Assume we have another trivialization ϕβ which gives a different diffeomorphism

∼
ϕβ : π −1 (b) −
→ G.

Then the transition function

φ−1 φα
ϕαβ : G −−→ π −1 (b) −−→ G
β

is a left multiplication by some element t ∈ G

ϕαβ (g) = t · g.

Since left multiplications commute with right multiplications,

t·(−)
G G
φβ φα

π −1 (b)

ϕαβ (ϕβ (p) · g) = t · (ϕβ (p) · g) = (t · ϕβ (p)) · g = ϕαβ (ϕβ (p)) · g = ϕα (p) · g,

which implies
−1
ϕ−1
α (ϕα (p) · g) = ϕβ (ϕβ (p) · g).

This says that the right action p · g is well-defined and does not depend on the choice of local
trivializations.
Note that the right G-action on itself is transitive and free. Therefore each fiber π −1 (b) is
a single G-orbit. The fibration
π : P −→ B

can be viewed as the quotient map by the right G-action, and

B = P /G

can be identified as the orbit space.

Example 4.1.12 (Hopf fibration). The Hopf fibration

S1 S3

S2

110
can be realized as a principal S 1 -bundle. Let us identify
n o
S 1 = eiθ = {λ ∈ C| | λ| = 1} ,

S 3 = (z1 , z2 ) ∈ C2 | |z1 |2 + |z2 |2 = 1 .

We define a right S 1 -action (Since S 1 is an abelian group, left and right actions are essentially
the same) on S 3
S 3 × S 1 −→ S 3
 
(z1 , z2 ), e iθ
7−→ (z1 eiθ , z2 eiθ )

This action is free. The orbit space

S 3 /S 1 ' CP 1 = S 2

is precisely CP 1 = S 2 . This tells that the Hopf fibration is a principal S 1 -bundle.

To emphasize the role of a group, we will use U (1) for the group of unitary complex numbers

U (1) = {z ∈ C | |z| = 1} .

This is an abelian Lie group. As we will see, electromagnetism is about the geometry of U (1)-
principal bundles, hence usually called abelian gauge theory. In general, Yang-Mills theory
generalizes Maxwell theory to principal G-bundles, hence is about non-abelian gauge theory.

4.2 U (1)-connection and Parallel Transport

We discuss the notion of connection on principal U (1)-bundles. Such construction exists
on any principal G-bundles, with a bit more care on the non-abelian nature of G. We work on
U (1)-bundles toward explaining Maxwell theory, and illustrate the basic geometric ideas.
Let π : P → B be a principal U (1)-bundle

U (1) P

B
Geometrically, this is a family of circles

111
 We already know that points of a fiber of P are related to each other via the (right)
U (1)-action. The essential geometric idea underlying the notion of connection is to be able
to transport points of a fiber of P to another fiber along a path in B, so different fibers can
communicate and compare with each other.

transporting the fiber

along the curve γ

B
x0 γ x1

In other words, we need a notion to help lifting a path from the base B to P horizontally.

4.2.1 Vertical Vector Field

The U (1)-action on P defines a vector field on P along the fiber direction via its infinitesimal
transformation. Let us parametrize U (1) by
n o
U (1) = eiθ .

Given a point x ∈ P and eiθ ∈ U (1), the transformed point will be simply denoted by
x · eiθ ∈ P . This defines a curve γx

γx : (−ε, ε) −→ P
θ 7−→ x · eiθ

with γx (0) = x. Its derivative at θ = 0 gives a tangent vector


′ d 
γx (0) = γx (θ) ∈ Tx P.
dθ θ=0

This construction applies to all points of P , and gives rise to a vector field on P , denoted by
Vθ . Explicitly, 
d 
Vθ (x) := (x · eiθ ) for x ∈ P.
dθ θ=0
The vector field Vθ points along the fiber direction of P , or in formula this means

π∗ (Vθ ) = 0.

112
 Vθ

Moreover, if you follow the flow of the vector field Vθ , then it will circle around the fiber and
come back to the start point after time 2π.
Remark 4.2.1. In general for a principal G-bundle P , every element of the Lie algebra g of G
leads to a vector field on P via the infinitesimal right G-action. In this case we have a map

g −→ Vect(P ).

When G = U (1), g = R, it gives a single vector field Vθ as discussed above.

4.2.2 Connection 1-form

Definition 4.2.2. A connection on a principal U (1)-bundle π : P → B is a 1-form A ∈ Ω1 (P )

on P satisfying the following
⃝
1 ιVθ A = 1 as a function on P . Here ιVθ is the interior product (contraction) with respect
to the vector field Vθ .
⃝
2 A is invariant under the U (1)-action on P .

Condition ⃝
2 says that for any eiθ ∈ U (1), let us define a diffeomorphism

f : P −→ P
x 7−→ x · eiθ

Then f ∗ A = A. Equivalently, this is

LVθ A = 0

where LVθ is the Lie derivative with respect to the vector field Vθ .

Example 4.2.3. Assume P = B × U (1) is trivial. Let xi denote local coordinates on B and
θ denote the angle coordinate on U (1) as above. Then
∂
Vθ = .
∂θ
An arbitrary 1-form A can be expressed in coordinates by
X
A = Aθ (x, θ)dθ + Ai (x, θ)dxi .
i

113
We have 
ιV A = Aθ
θ

L A = (∂ A )dθ + P (∂ A )dxi
Vθ θ θ i θ i

Then A defines a connection if

⃝
1 Aθ = 1,
⃝
2 ∂θ Aθ = ∂θ Ai = 0,
i.e. A is of the form
X
A = dθ + Ai (x)dxi .
i

For a general U (1)-bundle, we can describe the 1-form A in terms of a local trivialization
φ
π −1 (Uα ) Uα × U (1)

Uα

Then a local coordinate system xi on Uα and the angle coordinate θ on U (1) define a local
coordinate system on π −1 (Uα ) via the map ϕ. In such coordinates, the connection A will again
take the form locally
X
(ϕ−1 )∗ A = dθ + Ai (x)dxi
i

for some functions Ai (x) on Uα .

4.2.3 Horizontal Vector

A connection 1-form A allows us to lift a tangent vector from the base manifold to the
total space “Horizontally”.

Definition 4.2.4. Let A be a connection on the principal U (1)-bundle π : P → B. Let q ∈ P .

A tangent vector v ∈ Tq P is called a Horizontal vector if

ιv A = 0.

The space of Horizontal vectors at q will be denoted by

Hq = {v ∈ Tq P | ιv A = 0} .

We emphasize that Hq depends on the choice of A.

Clearly we have a direct sum decomposition

Tq P = Hq ⊕ RVθ .

114
 Vθ

q Hq

x Tx B


In local coordinates xi , θ from a local trivialization as above, with A expressed as
X
dθ + Ai (x)dxi ,
i

the horizontal vectors are spanned by

 
∂ ∂
− Ai (x) .
∂xi ∂θ
It is clear that the push-forward

π∗ : Hq −→ Tx B, where x = π(q)

is a vector space isomorphism. In coordinates, it sends

 
∂ ∂ ∂
π∗ i
− Ai (x) = .
∂x ∂θ ∂xi

Definition 4.2.5. Let q ∈ P , x = π(q) ∈ B. We define the horizontal lift of a tangent vector
v ∈ Tx B at q to be the tangent vector ve ∈ Hq such that

π∗ (e
v ) = v.

ve
q

x v

115
4.2.4 Parallel Transport

Proposition 4.2.6. Let γ : [0, 1] → B be a smooth curve on B from x0 = γ(0) to x1 = γ(1).

Let q0 ∈ π −1 (x0 ) be any chosen point. Then there exists a unique curve

e : [0, 1] −→ E
γ

e′ (t) is the horizontal lift of

e(0) = q0 and such that the tangent vector γ
with initial condition γ
the tangent vector γ ′ (t) for any t ∈ [0, 1].

q0 e
γ
q1

x0 γ x1 B

e is obtained by following the direction of the lifting of that of γ. In

Intuitively, the curve γ
local coordinates as above, if γ(t) is described by

xi (t) = γ i (t),

e(t) is given by
then γ


e(t) = xi (t) = γ i (t), θ(t)
γ

where θ(t) solves the equation (Exercise: show this)

dθ X  dγ i 
=− Ai (γ(t)).
dt dt
i

The existence and uniqueness follow from the standard theory in ordinary differential equations.

Definition 4.2.7. Given a curve γ : [0, 1] → B, with x0 = γ(0) and x1 = γ(1), we define the
parallel transport
Tγ : π −1 (x0 ) −→ π −1 (x1 )

by
e(1)
Tγ (q0 ) = γ

e is the horizontal lift of γ with initial condition γ

where γ e(0) = q0 as in the previous proposition.

Proposition 4.2.8. Tγ : π −1 (x0 ) → π −1 (x1 ) is U (1)-equivariant, i.e.

Tγ (q0 · g) = Tγ (q0 ) · g, ∀q0 ∈ π −1 (x0 ), g ∈ U (1).

116
Proof: The fact that A is U (1)-invariant implies that horizontal vectors are preserved under the
e is a horizontal lift of γ. Then for any g ∈ U (1), the new curve
U (1)-action. Assume γ

eg (t) = γ
γ e(t) · g

e(0) = q0 , then the initial point of γ

is also a horizontal lifting of γ. If γ eg is

eg (0) = γ
γ e(0) · g = q0 · g.

It follows that
Tγ (q0 · g) = γ
eg (1) = γ
e(1) · g = Tγ (q0 ) · g.

eg
γ
q0 e
γ
q1

x0 x1 B
γ

Now we consider the case when γ is a loop, i.e. γ(0) = γ(1) = x0 . Then for any q0 ∈
π −1 (x0 ), we have Tγ (q0 ) ∈ π −1 (x0 ). Therefore

Tγ (q0 ) = q0 · g

for a unique g ∈ U (1). This g is called the holonomy of γ and q0 , denoted by

Holγ (q0 ).
q0 · g

q0

γ
x0

117
 In the U (1)-case which is an abelian group, Holγ (q0 ) does not depend on the choice of q0
(Exercise: show this) and we will simply write Holγ ∈ U (1).

4.3 Curvature and Chern Class

Let A be a connection 1-form on the principal U (1)-bundle π : P → B. We have seen that
for any loop γ : [0, 1] → B with γ(0) = γ(1) = x0 , it defines a holonomy

Holγ ∈ U (1)

describing the parallel transport action from the fiber π −1 (x0 ) to itself.

γ
x0 Holγ ∈ U (1)

The nontriviality of such holonomies indicate nontriviality of the U (1)-bundle via certain
twist or curving. We will make this precise in this section. The relevant geometric notion is
called the curvature.

4.3.1 Curvature 2-form

Definition 4.3.1. Let π : P → B be a principal U (1)-bundle, and A be a connection 1-form.

We define its curvature by the 2-form

FA = dA.

Since A is a 1-form on the total space P , the curvature FA is firstly defined as a 2-form on
P . However, FA is can be viewed in fact as a 2-form on B, which does not depend on the fiber
direction. To see this, let us choose a local trivialization over an open U ⊂ B

π −1 (U) U × U (1)

U
with local coordinates {xi } on the base U and angle coordinate θ on the fiber U (1). The
connection A will take the form
X
A = dθ + Ai (x)dxi .
i

The curvature is then given by

X 1X
F = dA = ∂i Aj dxi ∧ dxj = (∂i Aj − ∂j Ai )dxi ∧ dxj
2
i,j i,j

118
which is clearly a 2-form on the base.
Note that this expression is independent of the choice of local trivialization. In fact, suppose
we have a different local trivialization

π −1 (U)

φ
U × U (1) U × U (1)

∈
∈
e
(x, eiθ ) (x, eiθ )

Then the transition map

e
(x, eiθ ) 7−→ (x̃ = x, eiθ )

is given by a family of U (1) -action on the fiber

e
eiθ = eiϕ(x) eiθ

for some map

eiϕ(x) : U −→ U (1).

In other words, the coordinate transformation is


xi = x̃i
θ = θe + φ(x) mod 2π

e
for some function φ(x). Then in (x̃, θ)-coordinate,
X X
A = dθ + Ai (x)dxi = dθe + dφ(x) + Ai (x)dxi .
i i

The curvature
!
X X
F = dA = d dθe + dφ(x) + Ai (x)dx i
= (∂i Aj ) dxi ∧ dxj
i i,j

is independent on how to choose the fiber coordinate as expected.

We will treat the curvature F = dA as a 2-form on B in the subsequent discussions, despite
the fact that it is defined in terms of data on P . A more precise statement is that F equals to
the pull-back of a 2-form on B via the map π. In fact, the pull-back of forms

π ∗ : Ω• (B) → Ω• (P )

for a fiber bundle is an injective map, and identifies forms on B as a subspace of forms on
P (such forms are called basic forms in fiber bundle geometry). Then the claim is that the
curvature F lies in the image of this map. This explains the above local calculation.

119
Remark 4.3.2. The curvature takes the simple form F = dA since U (1) is an abelian Lie group.
In general, a connection on a principal G-bundle is a g-valued 1-form A ∈ Ω1 (P, g). The
curvature will take the form
1
F = dA + [A, A]
2
where [−, −] denotes the Lie bracket on the Lie algebra g of G. In the abelian U (1)-case,
[−, −] = 0. In the G-bundle case, such defined curvature form will again descend to the base B.

4.3.2 Holonomy and Curvature

Now we explain the relation between the holonomy and curvature. Let γ be a loop on B
which bounds a surface Σ. Here γ could be piece-wise smooth, and Σ may not be a disk but
could have non-trivial topology.

γ
Σ

γ = ∂Σ
x0

Theorem 4.3.3. The holonomy of γ is given by the integration of curvature via

´
Holγ = e−i Σ F
∈ U (1).

Proof: We triangulate Σ into small pieces as indicated

Σα

For each small Σα , let γα = ∂Σα be its boundary loop. For any path in the interior, it is part
of two different small loops with opposite direction. The corresponding parallel transport will
cancel each other. It follows that
Y
Holγ = Holγα .
α

120
On the other hand, ˆ Xˆ
F = F.
Σ α Σα

Therefore we only need to prove

´
Holγα = e−i Σα F

for each small Σα . Thus we assume Σ is small enough and lies inside an open U ⊂ B with a
local trivialization
π −1 (U) U × U (1)

U
 i
Let x denote local coordinates on U, and θ denote the angle coordinate on the fiber. The
connection 1-form is of the form
X
dθ + α where α= Ai (x)dxi .
i
 
Let e is parametrized by xi (t), θ(t)
xi (t) parametrize the curve γ in B. Its lifting γ
where θ(t) satisfies the flow equation (Horizontal condition)

dθ X dxi (t)
+ Ai (x(t)) =0
dt dt
i

which is solved in integral form as

ˆ t
θ(t) = θ(0) − γ ∗ α, where γ : [0, 1] → U.
0

By definition, the holonomy of γ is

´1 ´ ´ ´
γ∗α
Holγ = ei(θ(1)−θ(0)) = e−i 0 = e−i γ α
= e−i Σ dα
= e−i Σ F
.

This proves the theorem.

4.3.3 Chern Class

The curvature 2-form F is clearly a closed form on B:

dF = 0.

It may not be an exact form on B (although it is written as F = dA, but A is not a 1-form
on B but a 1-form on P ). Thus F defines a de Rham cohomology class on B which captures
topological information of the bundle.
Let us first understand how the curvature depends on the choice of the connection. Assume
e
A, A are two connection 1-forms. Let

e−A
α=A

121
which is a 1-form on P . We claim that α is a pull-back of a 1-form form B. In fact, in local
coordinates of a local trivialization, we can write
X
A = dθ + Ai (x)dxi
i
X
e = dθ +
A ei (x)dxi
A
i
 
e−A = P A
Then α = A ei (x) − Ai (x) dxi which clearly depends only on the base. So we will
i
write
α ∈ Ω1 (B).

Remark 4.3.4. The above computation says that the space of connections on P is an affine space

{connections on P} = A0 + Ω1 (B)

for any specific chosen connection A0 .

Now the curvatures of different connections are related by

FAe = FA + dα,

i.e. differs by an exact 1-form on B. In particular, the de Rham cohomology class

[F ] ∈ H 2 (B)

depends only on the bundle P , but not on the choice of the connection.

Definition 4.3.5. The (first) Chern class of the U (1)-principal bundle π : P → B is the de
Rham cohomology class  
1
c1 (P ) := F ∈ H 2 (B).
2π
One important property of c1 (P ) is that it is an integral class. More precisely, we have

Theorem 4.3.6. Let Σ be any closed surface on B without boundary. Then

ˆ ˆ
1
c1 (P ) = F ∈Z
Σ 2π σ
is an integer.

Proof: We can treat Σ as having the boundary with a trivial constant loop γ. Clearly we have
Holγ = 1. By Theorem 4.3.3
´ ˆ
−i
e Σ F
= Holγ = 1 =⇒ F ∈ 2πZ.
Σ

122
 Σ

x0
γ: constant loop mapped to x0

Example 4.3.7. If the U (1)-bundle P = B × U (1) is trivial, then

c1 (P ) = 0 ∈ H 2 (B).

In fact, we can choose the connection globally expressed as

A = dθ.

Then F = dA = 0.

Conversely, if we find some closed surface Σ without boundary such that

ˆ
F 6= 0,
Σ

then such bundle can not be trivial. Dirac monopole is such an example (see Example 4.4.5).

4.4 Local Gauge and Transition

We give a concrete description of a connection 1-form in terms of local trivializations.

4.4.1 Local Gauge 1-form via Trivialization

Let
U (1) P
π

B
be a principal U (1)-bundle. We first give an equivalent description of local trivializations in
terms of local sections.

Proposition 4.4.1. Let U ⊂ B. Then there is a one-to-one correspondence

1:1
{local trivializations of P over U} {local sections of P over U}
∈

ϕ : π −1 (U ) → U × U (1) σ ∈ Γ(U, P )

123
Proof: Let σ ∈ Γ(U, P ). Then it defines a diffeomorphism

ϕ−1 : U × U (1) −→ π −1 (U)

(x, g) 7−→ σ(x) · g

Its inverse defines a local trivialization ϕ.

Conversely, given a local trivialization ϕ, it defines a local section σ by

σ(x) = ϕ−1 (x, 1)

where 1 ∈ U (1) is the identity element. This establishes the correspondence.

σ
ϕ S1
1

U
π

Let A be a connection 1-form on P . We aim at an explicit description of A locally. Let

ϕ : π −1 (U) −→ U × U (1)

be a trivialization over an open U ⊂ B. It corresponds to a section σ ∈ Γ(U, P ) via

σ(x) = ϕ−1 (x, 1)


as described in the previous proposition. Let xi be local coordinates on U, and θ be the angle
coordinate on U(1). As we have described before, the connection A in the local trivialization
will take the form
X
dθ + Ai (x)dxi .
i

Precisely, this means that

!
X
A|π−1 (U) = ϕ∗ dθ + Ai (x)dxi
i

for some Ai (x). To describe the meaning of Ai (x), consider the section σ : U → P . Let

Aσ := σ ∗ A

which is a 1-form on U. To compute Aσ , consider the composition

φ
U −→ π −1 (U) −→ U × U (1).
σ

124
We have
! !
X ∗
X
∗ ∗ ∗
Aσ = σ A = σ ϕ dθ + Ai (x)dx i
= (ϕ ◦ σ) dθ + Ai (x)dx i
.
i i

Since
ϕ ◦ σ : U −→ U × U (1)
x 7−→ (x, 1)
It follows that
X
Aσ = Ai (x)dxi .
i

In other words, the information of Ai (x) in the chosen local trivialization is precisely the pull-
back of A to the base via the corresponding local section.
We next describe how Aσ depends on the choice of local trivializations. Let

e : π −1 (U) → U × U (1)
ϕ

be another local trivialization on U, which corresponds to another section σ

e ∈ Γ(U, P ). Then
it determines a map
eiα(x) : U −→ U (1)

such that
e(x) = σ(x) · eiα(x) ,
σ ∀x ∈ U.

It relates the two trivializations by

e(x) · e−iα(x) g.
σ(x) · g = σ

π −1 (U)
φ e
φ

e −1
φ◦φ
U × U (1) U × U (1)
∈
∈

(x, g) (x, e−iα(x) g)

Let 
Aσ = σ ∗ A = P Ai (x)dxi
i
Aeσ = σ P ei (x)dxi
e∗ A = i A
be local descriptions of A with respect to the corresponding trivialization. By construction,
under the diffeomorphism
e −1
φ◦φ
U × U (1) −−−−→ U × U (1),

we have !

X X
−1 ∗ ei (x)dxi
e◦ϕ
ϕ dθ + A = dθ + Ai (x)dxi .
i i

125
On the other hand, since
 
e ◦ ϕ−1 : (x, eiθ ) 7−→ x, e−iα(x) eiθ ,
ϕ

we have explicitly
!

∗ X X X
e ◦ ϕ−1
ϕ dθ + ei (x)dxi
A = d (θ − α(x)) + ei (x)dxi = dθ − dα(x) +
A ei (x)dxi .
A
i i i

Thus we find
X X
ei (x)dxi =
A Ai (x)dxi + dα(x).
i i
To summarize, we have proved the following

Proposition 4.4.2. Let A be a connection 1-form on P . Let σ, σ

e ∈ Γ(U, P ) be two sections
over an open U ⊂ B, which are related by

e = σ · eiα
σ

for some eiα : U → U (1). Then the two local descriptions Aσ = σ ∗ A and Aσe = σ
e∗ A are related
by
Aσe = Aσ + dα.

Definition 4.4.3. Aσ is called the local gauge 1-form of A with respect to the trivialization
σ. The above change of Aσ via different local trivializations is called “local gauge transfor-
mations”.

4.4.2 Connection via Transition Functions

Let {Uα } be an open cover of B, with local trivializations

ϕα : π −1 (Uα ) −→ Uα × U (1)

and transition function

ϕβα : Uα ∩ Uβ −→ U (1)

such that

π −1 (Uαβ )
φα φβ

(Uα ∩ Uβ ) × U (1) (Uα ∩ Uβ ) × U (1)

∈

(x, g) (x, ϕβα (x) · g)

The collection of transition functions {ϕαβ } clearly satisfies the following

1 ϕαβ = ϕ−1
⃝ βα ,
⃝
2 ϕαγ (x)ϕγβ (x)ϕβα (x) = 1, ∀x ∈ Uα ∩ Uβ ∩ Uγ .
Condition ⃝
2 is also called the cocycle condition.

126
 (x, g) = (x, ϕαγ (x)ϕγβ (x)ϕβα (x)g)

∈
Uαβγ × U (1)
φα

π −1 (Uαβγ )
φβ φγ

Uαβγ × U (1) Uαβγ × U (1)

∈ 3
(x, ϕβα (x)g) (x, ϕγβ (x)ϕβα (x)g)

Here Uαβγ = Uα ∩ Uβ ∩ Uγ .
In fact, any collection of maps

ϕβα : Uα ∩ Uβ −→ U (1)

satisfying condition ⃝
1 and ⃝
2 defines a principal U (1)-bundle P by the quotient
!,
a
P = Uα × U (1) ∼
α

where the equivalence relation ∼ identifies

(x, g) ∼ (x, ϕβα (x)g)

∈

Uα × U (1) Uβ × U (1)

for x ∈ Uα ∩ Uβ . We leave the details to the reader to check that this is indeed a principal
U (1)-bundle.
Now let us reformulate the data of local trivializations {ϕα } and transition functions {ϕαβ }.
By the proposition above, each ϕα corresponds to a local section σα : U → π −1 (U) by

σα (x) = ϕ−1
α (x, 1).

Then for x ∈ Uα ∩ Uβ , we have

−1
σβ (x) = ϕ−1 −1 −1
α ◦ ϕα ◦ ϕβ (x, 1) = ϕα (x, ϕαβ (x)) = ϕα (x, 1)ϕαβ (x) = σα (x)ϕαβ (x),

i.e. we have
σβ = σα ϕαβ on Uα ∩ Uβ .

Let us fix a trivialization of P described by local sections σα : Uα → P on each Uα and

transition functions {ϕαβ : Uα ∩ Uβ → U (1)} as above. Let A be a connection 1-form on P . We
can use σα to define a local gauge 1-form Aα on Uα via pull-back

Aα := σα∗ (A) ∈ Ω1 (Uα ).

127
 On the intersection Uα ∩ Uβ , we have

σβ = σα ϕαβ on Uα ∩ Uβ .

By Proposition 4.4.2, we have

1
Aβ = Aa + ϕ−1 dϕαβ (∗)
i αβ
Equivalently, if we write ϕαβ = eiχαβ , then

Aβ = Aα + dχαβ .

Note that the curvature is FA = dAα , and it glues Fα = Fβ on Uα ∩ Uβ to define a

global 2-form on B. This is consistent with our previous discussion. Conversely, local 1-forms

Aα ∈ Ω1 (U) which are related by equation (∗) on intersections give rise to a connection 1-form
A on P , by running back the above process. We have now proved the following
Proposition 4.4.4. Let π : P → B be a principal U (1)-bundle with chosen local trivializations
{σα ∈ Γ(Uα , P )} on an open cover {Uα } and transition functions {ϕαβ : Uα ∩ Uβ → U (1)} such
that σβ = σα ϕαβ on Uα ∩ Uβ . Then there is a one-to-one correspondence
 
A collection of 1-form Aα on each Uα such that
{Connection 1-form A on P } 1:1 1
 Aβ = Aα + ϕ−1 dϕαβ on each Uα ∩ Uβ 
i αβ
∈

A {Aα = σα∗ A}
Example 4.4.5 (Dirac Monopole). Consider the unit sphere S 2 ⊂ R3 with spherical coordinates
(θ, ϕ).
z

(θ, ϕ)

y
ϕ

We consider a cover of S 2 by
U+ = S 2 − {(0, 0, −1)}
U− = S 2 − {(0, 0, 1)}

U+ U−

128
Each U± is contractible, hence the fiber bundle can be trivialized on U± .
Consider the following 1-form
n
A± = ± (1 ∓ cos θ) dϕ on U± .
2
Here n is a constant to be determined. On U+ ∩ U− ,

A+ − A− = ndϕ.

If we ask {A+ , A− } to define a connection 1-form of some principal U (1)-bundle, then the
transition function on U+ ∩ U− is

U+ ∩ U− −→ U (1)
(θ, ϕ) 7−→ einφ

Since ϕ is defined modulo 2π, this map is well-defined if and only if n ∈ Z is an integer.
In other words, for each integer n ∈ Z, the transition function

eiχ : U+ ∩ U− −→ U (1)
(θ, ϕ) 7−→ einφ

defines a principal U (1)-bundle Pn on S 2 , and the collection {A± = ± n2 (1 ∓ cos θ) dϕ} defines
a connection on Pn . Then the curvature form is
n
F = dA+ (= dA− ) = sin θ dθ ∧ dϕ.
2
It is direct to compute ˆ
1
F = n.
2π S2
Hence the U (1)-bundle Pn is nontrivial for n 6= 0. Moreover, these U (1)-bundle {Pn }’s are all
topologically different since they have different Chern classes (Theorem 4.3.6).

4.5 Gauge Transformation

We describe the group of gauge transformations on principal U (1)-bundles. Such transfor-
mations give rise to equivalent data both in physics and in geometry.

4.5.1 Gauge Transformation

Let π : P → B be a principal U (1)-bundle. We define an automorphism of P to be a

diffeomorphism
f : P −→ P

such that
⃝
1 f is fiberwise, i.e. π ◦ f = π

129
 f
P P
π π
B
⃝
2 f is U (1)-equivariant, i.e.

f (q · g) = f (q) · g, ∀q ∈ P, g ∈ U (1).

We denote Aut(P ) to be the group of all automorphisms of the principal U (1)-bundle P .

For each point x ∈ B, f defines a U (1)-equivariant map

f : π −1 (x) −→ π −1 (x).

Since U (1) is abelian, this is simply given by a right action by an element of U (1). It follows
that such f can be equivalently described by

fe : B −→ U (1)

such that
f (q) = q · fe(x), ∀q ∈ π −1 (x).

Thus we have proved

Proposition 4.5.1. There is a group isomorphism

Aut(P ) = Map(B, U (1)).

Here given fe1 , fe2 ∈ Map(B, U (1)), their group product is

fe1 · fe2 (x) := fe1 (x) · fe2 (x), ∀x ∈ B.

Let A be a connection 1-form on P . Let f ∈ Aut(P ) be an automorphism of P . We can

use f to pull-back A to get another 1-form

f ∗ A ∈ Ω1 (P ).

We claim that f ∗ A is also a connection. In fact, since f is U (1)-equivariant, for any q ∈ P ,

 
∂    ∂ 

 
 f (qeiθ ) =  f (q)eiθ ,
∂θ  ∂θ 
θ=0 θ=0

which says
f∗ Vθ = Vθ .

It follows that
1 ιVθ f ∗ A = f ∗ (ιVθ A) = f ∗ (1) = 1
⃝
2 LVθ f ∗ A = f ∗ (LVθ A) = f ∗ (0) = 0.
⃝

130
 We can also describe this explicitly in local coordinates via a local trivializations, with base

coordinates xi and U (1) angle coordinate θ. Then A can be written as
X
A = dθ + Ai (x)dxi .
i

The automorphism f can be expressed by a map

fe : B −→ U (1)

which in local coordinates is

fe(x) = eiϕ(x) .

Then
f (x, eiθ ) = (x, eiθ eiϕ(x) ),

so
X X
f ∗ A = d(θ + φ(x)) + Ai (x)dxi = dθ + Ai (x)dxi + dφ(x)
i i

which still takes the form of a connection. This proves the claim that f ∗ A is also a connection.

Proposition 4.5.2. Let A be a connection 1-form. Let f ∈ Aut(P ) which corresponds to

fe = eiϕ ∈ Map(B, U (1)) as above. Then

f ∗ A = A + π ∗ dφ.

Proof: This follows from the above local computation.

Remark 4.5.3. φ is defined modulo 2π, but dφ is a well-defined 1-form.

Definition 4.5.4. Let Conn(P ) denote the space of connections on P . The natural action

Aut(P ) ↷ Conn(P )

is called global gauge transformations, or simply gauge transformations.

Proposition 4.5.5. The curvature 2-form of the principal U (1)-bundle P is invariant under
gauge transformations.

Proof: Let A ∈ Conn(P ) and f ∈ Aut(P ). Then

Ff ∗ A = d(f ∗ A) = dA + d(dφ) = dA = FA .

Remark 4.5.6. In general for principal G-bundle, the curvature 2-form will transform via group
conjugation under gauge transformations. When G = U (1) is abelian, group conjugation is
trivial.

131
 Therefore the curvature of U (1)-bundle can be viewed as a map

F : Conn(P )/ Aut(P ) −→ Ω2 (B).

e = A + π ∗ α. The
In summary, any two different choices of connection differ by a 1-form on B: A
curvatures differ by the exact form dα

FAe = FA + dα.

Gauge transformations lead to change of the connection by α = dφ, hence leaving F invariant.

4.5.2 Local v.s. Global Gauge Transformations

The formula of local and global gauge transformations look very similar. We clarify these
two notions to avoid possible confusions. In short,

• Local gauge transformation is about the different expression of the same connection via
different choices of local trivializations.

• Global gauge transformation is about the change from one connection to another, so
linking different connections.

However, these two notions are closely related when the bundle P is trivial, where we
have a trivialization over the full base B. Indeed, assume now P is trivial, then we have an
identification (Proposition 4.4.1)

{trivialization of P over B} = Γ(B, P ).

Let σ1 , σ2 ∈ Γ(B, P ) be two trivializations. Then there exists a unique map

fe : B −→ U (1)

such that
σ2 (x) = σ1 (x)fe(x), ∀x ∈ B.

Let f ∈ Aut(P ) be the automorphism corresponding to fe.

The above relations are equivalently described as

σ2 = f ◦ σ1 .
f
P P

σ1 σ2
B
In particular, Aut(P ) acts on Γ(B, P ) freely and transitively in the case when P is trivial.
Let A be a connection on a trivial U (1)-bundle P over B. Let σ ∈ Γ(B, P ) be a global
section which gives a trivialization of P . Let f ∈ Aut(P ) be an automorphism of P which
corresponds to a map
fe : B −→ U (1)
x 7−→ eiϕ(x)

132
⃝
1 f gives a (global) gauge transformation
e = f ∗ A = A + π ∗ dφ.
A 7−→ A

In the local form described by the same trivialization σ, we have


Aσ = σ ∗ A
A eσ = σ ∗ (f ∗ A) = Aσ + dφ

⃝
2 f gives a different trivialization by the section

e =f ◦σ
σ ∈ Γ(B, P ).

e give different local descriptions of the same connection A by

These two sections σ and σ

Aσ = σ ∗ A
A = σ e∗ A = A + dφ
e
σ σ

1 and ⃝
Although the transformation formula in ⃝ 2 look similarly, they are different in nature.
⃝
1 is about gauge transformations, which changes one connection to another connection. ⃝
2 is
about local gauge transformation, where the connection is fixed but expressed on the base with
different trivializations. In the case when P is trivial, we have a (noncanonical) identification

Γ(B, P ) ' Aut(P ) if P is trivial.

In general, they are different and we hope it will not cause further confusion.
Now we discuss the case when P is nontrivial in general. Let us fix an open cover {Uα } of
B and local trivialization on each Uα by σα ∈ Γ(Uα , P ). Let A be a connection 1-form on P .
Then we can express A locally via the collection {σα } by

{Aα = σα∗ A} .

Let f ∈ Aut(P ) be an automorphism of P , which corresponds to a map

fe = eiϕ : B −→ U (1).

The gauge transformation of A via f is

A 7−→ f ∗ A.

We can express the gauge transformation locally via the fixed trivialization {σα }

{Aα } 7−→ {(f ∗ A)α } .

Then locally on each Uα , we have

(f ∗ A)α = (f ◦ σα )∗ A = Aα + dφ.

Thus a gauge transformation is equivalently described with respect to the fixed trivialization
{σα } by a transformation of collections
f
{Aα } 7−→ {Aα + dφ} , where dφ is the same expression in all α

for fe = eiϕ as above. Again, the gauge transformation preserves the curvature form
f
{Fα = dAα } 7−→ {(f ∗ F )α = dAα + d(dφ) = dAα } .

133
 4.6 Maxwell Theory as U (1)-gauge Theory
We are now ready to explain electromagnetism in terms of U (1)-gauge theory.

4.6.1 Potential as Gauge 1-form

Let us now consider the base manifold being the spacetime

B = R3,1 .

Let P be a principal U (1)-bundle on R3,1 . Since R3,1 is contractible, P is a trivial bundle. We

will fix once for all a global section σ of P that gives a global trivialization

∼
P −→ R3,1 × U (1).

Any connection 1-form A on P therefore corresponds equivalently to a gauge 1-form A on B

via
Aσ = σ ∗ A.

Since σ will be fixed, we will simply write the gauge 1-form as A = Aσ . In spacetime coordinates
{x, y, z, t} on R3,1 ,
A = At dt + Ax dx + Ay dy + Az dz.

The gauge 1-form is identified with potentials in electromagnetism as


φ = −At scalar potential
A~ = (Ax , Ay , Az ) vector potential

Let eiχ : R3,1 → U (1). It gives a gauge transformation of the connection 1-form, hence the
(local) gauge 1-form, via
A 7−→ A + dχ.

This is the same gauge transformation that we have encountered in electromagnetism.

4.6.2 Electromagnetic Field as Curvature

The curvature 2-form of the connection

F = dA

collects the components of electric and magnetic fields. It takes the form that we have seen

F = E ∧ dt + ∗3 B = (Ex dx + Ey dy + Ez dz) ∧ dt + (Bx dy ∧ dz + By dz ∧ dx + Bz dx ∧ dy) .

The curvature of the U (1)-connection is invariant under gauge transformations.

134
4.6.3 Maxwell Action

Let J denote the electric charge-current 1-form

J = ρ/ε0 dt − µ0 (jx dx + jy dy + jz dz).

Then Maxwell’s equations are 

dF = 0
d∗ F = J

Here d∗ = ∗d∗ is the adjoint of d. These equations can be derived from the action principle as
follows.

Definition 4.6.1. The Maxwell action SM [A] is a functional on Conn(P ) defined by

ˆ  
1 1
SM [A] = F ∧ ∗F + ∗J ∧ A .
µ0 R3,1 2

If we write it as ˆ
SM [A] = L (A) dx ∧ dy ∧ dz ∧ cdt,
R3,1
where L (A) is the Lagrangian density, then a direct computation gives
ε0 ~ ~ 1 ~ ~  
~ · ~j .
L = E ·E− B · B + −ρφ + A
2 2µ0
Now we consider the variation of SM [A] under an arbitrary variation of the connection

A −→ A + δA

where δA is a 1-form on R3,1 . Under this variation,

ˆ
1 1 1
δSM = (δF ) ∧ ∗F + F ∧ ∗δF + ∗J ∧ δA
µ0 R3,1 2 2
ˆ
1
= (δF ∧ ∗F − δA ∧ ∗J)
µ0 R3,1
ˆ
1
= (dδA ∧ ∗F − δA ∧ ∗J)
µ0 R3,1
ˆ
1
= δA ∧ (d (∗F ) − ∗J) .
µ0 R3,1
An extremal point of SM corresponds to a gauge 1-form A such that

δSM = 0 for arbitrary δA.

This is the same as asking

d(∗F ) = ∗J

which is equivalent to d∗ F = J. On the other hand, the equation dF = 0 is already captured

by the use of potential. We conclude that

Solution of Maxwell’s equations ⇐⇒ Critical point of SM

135
4.6.4 Gauge Principle and Charge Conservation

Gauge principle asks for invariance of physical quantities under gauge transformations. Let
us analyze the Maxwell action under gauge transformations.
Consider a gauge transformation described by

eiχ : R3,1 −→ U (1).

χ can be viewed as a function on R3,1 defined modulo 2π. Since the topology of R3,1 is trivial,
we can actually lift χ to be a single-valued function on R3,1 , and we assume this.
Under the gauge transformation generated by eiχ , the gauge 1-form transforms as

A 7−→ A + dχ

and the curvature F is invariant. It follows that the Maxwell action transforms as
ˆ ˆ
1 1
SM [A + dχ] = SM [A] + ∗J ∧ dχ = SM [A] + d (∗J) ∧ χ.
µ0 R3,1 µ0 R3,1
If we require that SM should be invariant under arbitrary gauge transformations, then we need

d(∗J) = 0.

This is precisely the charge conservation. In other words, charge conservation can be viewed as
a direct consequence of gauge principle. Assume charge conservation, then the Maxwell action
defines a functional
SM : Conn(P )/ Aut(P ) −→ R.

4.6.5 Magnetic Monopole

Maxwell’s equations 
dF = 0
d(∗F ) = ∗J

is based on the assumption (in known experiments) that magnetic monopole does not exist. In
theory, the full Maxwell’s equations are

dF = ∗Jm
d(∗F ) = ∗J
e

where Je is the electric charge-current 1-form, and Jm is the magnetic charge-current 1-form.
It exhibits the full electromagnetic duality

F ←→ ∗F
Jm ←→ Je

The case without magnetic monopole corresponds to Jm = 0. In general, suppose Jm 6= 0 but

is supported on a subspace M ⊂ R3,1 . Thus

Jm = 0 on R3,1 − M

136
and on the region R3,1 − M we have

dF = 0
on R3,1 − M.
d(∗F ) = ∗J
e

Therefore it corresponds to a U (1)-gauge theory on R3,1 −M . However, the topology of R3,1 −M

may no longer be trivial, and there could exist nontrivial principal U (1)-bundle on R3,1 − M .
In this case the connection can not be described by a single gauge 1-form, but need a collection
of gauge 1-forms on a covering of R3,1 − M .
We describe the example of Dirac magnetic monopole to illustrate this. Consider a magnetic
monopole with magnetic charge n sitting at rest at the origin of R3 . Its trajectory in spacetime
R3,1 is
M = {0} × Rt ⊂ R3 × Rt

where Rt denotes the real line parameterized by time t. In this case we are led to consider
U (1)-gauge theory on


R3,1 − M = R3 − {0} × Rt .

We use spherical coordinates (r, θ, ϕ) to parametrize points in the space.

z

(r, θ, ϕ)
r
θ

y
ϕ

Then R3,1 is parametrized by (r, θ, ϕ, t) and R3,1 − M corresponds to the locus where r > 0.
Now we choose a covering of R3,1 − M by


U+ = R3 \ {(0, 0, z) | z ≤ 0} × Rt


U− = R3 \ {(0, 0, z) | z ≥ 0} × Rt

z≥0

z≤0

U+ U−

137
Each U± is contractible and so a principal U (1)-bundle can be trivialized on the cover {U+ , U− }.
The local gauge 1-form of the Dirac magnetic monopole is described in the local trivialization
by
n n
A± = (−ydx + xdy) = ± (1 ∓ cos θ)dϕ on U± .
2r(z ± r) 2
On the intersection U+ ∩ U− , we have

A+ − A− = ndϕ on U+ ∩ U−

which corresponds to the transition function (assuming n ∈ Z)

einφ : U+ ∩ U− −→ U (1).

This transition function defines a principal U (1)-bundle Pn on R3,1 − M (see Example 4.4.5).
The curvature form is
n
F = dA+ (= dA− ) = sin θdθ ∧ dϕ.
2
Let Sr2 denote a sphere of radius r in R3 . Then
ˆ ˆ
1
c1 (Pn ) = F = n ∈ Z.
Sr2 2π Sr2

This describes a magnetic monopole of magnetic charge n. In fact

ˆ ˆ ˆ
1 1 1
F = ∗3 B = d~σ · B
~
2π S 2 2π S 2 2π S 2

which describes the magnetic flux over S 2 . It computes the total magnetic charge surrounded
by the sphere Sr2 .

4.7 Associated Bundle and Matter Field

At the end, we describe how electromagnetic field is coupled with matter field.

4.7.1 Associated Vector Bundle

Let us start with a general discussion of the construction of vector bundles associated to a
principal G-bundle. Let

G P
π

B
be a principal G-bundle. Let
ρ : G −→ End(V )

be a representation of G on a vector space V of dimension m. We consider the quotient space

P ×ρ V := P × V / ∼

138
where the equivalence relation ∼ is

(p · g, v) ∼ (p, g · v), ∀p ∈ P, v ∈ V, g ∈ G.

Here p · g is the right g-action on P , and g · v is the left g-action on V with respect to the
representation ρ
g · v := ρ(g)(v).

It carries a natural projection map

P ×ρ V 3 (p, v)
πρ

B 3 π(p)

Proposition 4.7.1. πρ : P ×ρ V → B defines a vector bundle of rank = dim V .

Proof: Let
φ
π −1 (U) U×G

U
be a local trivialization of P over an open U ⊂ B. Then it induces

π −1 (U) ×G V (U × G) ×G V U×V

U
which defines a local trivialization of P ×G V over U with fiber = V . It can be checked that the
transition functions are linear transformations on V , with the help of the representation map
ρ : G → End(V ).

Definition 4.7.2. P ×ρ V is called the associated vector bundle of the principal G-bundle
P with respect to the G-representation ρ on V .

In the following discussion, we will be mainly discussing principal U (1)-bundles. Irreducible

unitary representation of U (1) is always on C, with the representation ρn classified by

ρn : eiθ 7−→ einθ , n ∈ Z.

The associated vector bundle is a complex vector bundle of rank = 1, i.e. a complex line bundle.

4.7.2 Hermitian Inner Product

Let us fix a principal U (1)-bundle π : P → B and the U (1)-representation ρn on C as

above. Let
πρ
Ln = P ×ρn C −→ B

denote the associated complex line bundle.

139
 For any point x ∈ B, the fiber πρ−1
n
(x) is a one dimensional complex vector space. We can
define a Hermitian inner product on each finer πρ−1
n
(x) as follows. Let v ∈ πρ−1
n
(x), which is
represented by a pair (p, z) for p ∈ π −1 (x), z ∈ C. Then we define its norm by

|v|2 := |z|2 .

This is independent of the choice of the representative. In fact, another representative of v is

pe−iθ , ρn (eiθ )z = (pe−iθ , einθ z). Then
 
 inθ 2
e z  = |z|2

which has the same value.

In general, for a section ψ ∈ Γ(B, Ln ), we can define its norm pointwise

|ψ|2 ∈ C∞ (B).

As we will discuss below, the wave function of particle in quantum mechanics will be a section
of a complex line bundle L as above, and |ψ|2 plays the role of probability distribution of the
quantum particle.

4.7.3 Covariant Derivative

We know that we can take derivatives on functions. We would like to extend such notion
of derivative on sections of vector bundles. We will focus on the case of the complex line bundle
Ln arising from the principal U (1)-bundle P and the U (1)-representation ρn on C as above.
We explain that a connection 1-form A on P will allow us to define derivatives of sections of all
the associated bundles.
Precisely, let s ∈ Γ(B, Ln ) denote a section of Ln on B, and V ∈ Vect(B) be a vector field
on B. We will define a notion of covariant derivative of s with respect to the vector field V ,
denoted by ∇V s, which is again a section ∇V s ∈ Γ(B, Ln ) as follows. Let U ⊂ B be an open
subset, with a local trivialization of P defined by a local section

σ ∈ Γ(U, P ).

It allows us to specify a unique representative of s(x) for x ∈ U by

s(x) = (σ(x), f (x)) , x ∈ U.

Here f (x) is a complex valued function on U. This representation clearly depends on the choice
of σ. Now we define the section ∇V s whose value on U is represented by

∇V s = (σ, ∂V f + in ιV (Aσ )f ) .

Here Aσ = σ ∗ A is the local gauge 1-form with respect to the choice σ. Note that since f is a
function, ∂V f is the usual derivative with respect to V . We can write it in a compact form

∇V s = (σ, ιV (df + inAσ f )) ,

140
where df + inAσ f is a 1-form on U.
We check that such ∇V s is well-defined, i.e. does not depend on the choice of σ. Assume
e ∈ Γ(U, P ). Then
we have another section σ

e(x) = σ(x)eiϕ(x) ,
σ x ∈ U,

where
eiϕ(x) : U −→ U (1).

e, we can represent the section s as

Using σ
 
s(x) = σe(x), fe(x) .

σ (x), fe(x)), which should represent the same element s(x), we find
Now since (σ(x), f (x)) ∼ (e

fe(x) = ρn (e−iϕ(x) )f (x) = e−inϕ(x) f (x).

On the other hand, the local gauge 1-form Aσe is related to Aσ by a local transformation

Aσe = Aσ + dφ.

Now let us compute ∇V s with respect to the local section σ

e. It gives
  
e, ιV dfe + inAσe fe .
σ

We need to check that it defines the same section of Ln on U as

(σ, ιV (df + inAσ f )) .

Indeed, from
 
dfe + inAσe fe = d e−inϕ f + in (Aσ + dφ) e−inϕ f = e−inϕ (df + inAσ f ) .

We see that
      
e, ιV dfe + inAσe fe = σ
σ e, e−inϕ ιV (df + inAσ f ) ∼ σee−iϕ , ιV (df + inAσ f )

= (σ, ιV (df + inAσ f )) .

This shows that ∇V s is well-defined and does not depend on the choice of local trivialization.

4.7.4 Matter Wave Function

In our previous study of electromagnetism, the electric and magnetic fields are of central
role, while the scalar and vector potentials are introduced as an auxiliary object to help solve
Maxwell’s equations. However in quantum mechanics, the situation is completely changed and
the use of gauge potential is essential. We briefly mention some key constructions.

141
 As we have discussed before, the Maxwell theory of electromagnetism can be viewed as a
U (1)-gauge theory
connection ←→ potentials
curvature ←→ electromagnetic fields

In quantum mechanics, particles are described by wave functions ψ, where |ψ|2 describes the
probability distribution of the quantum particle. In an electromagnetic background, a particle
of electric charge n will be described by a section

s ∈ Γ(R3,1 , Ln )

of the complex line bundle Ln associated to the principal U (1)-bundle P of Maxwell theory. It
seems unnecessary to introduce U (1)-bundle here since the topology of R3,1 is trivial. However,
when the space-time encodes nontrivial topology, this description will be essential. As we
have seen before, the norm |s|2 is still well-defined, with the physical meaning of probability
amplitude. With respect to a choice of section σ on R3,1 , we can represent the section as

s = (σ, ψ)

for ψ a function on R3,1 . Then

|s(x)|2 = |ψ(x)|2 .

If we choose another section to represent s, then ψ will undertake a phase rotation

ψ −→ einψ

but leaving |ψ|2 invariant.

Now the coupling between electromagnetism and matter field s is through the construction
of covariant derivative associated to a connection 1-form on P . Let us fix σ as before, and let

A = σ ∗ A = −φdt + Ax dx + Ay dy + Az dz

be the corresponding gauge 1-form. Then we have covariant derivatives on s as defined in

Section 4.7.3. For example,
∇x s = (σ, ∂x ψ + inAx ψ) .

Sometimes we simply write it as

∇x ψ = ∂x ψ + inAx ψ.

For the Schrödinger equation governing the wave function ψ, we simply need to replace every
derivative such as ∂x , ∂y , ∂z by ∇x = ∂x + inAx , ∇y = ∂y + inAy , ∇z = ∂z + inAz . For example,
the Hamiltonian H is  2
1 X ∂
H =− + inAi + V (r).
2m ∂xi
i

This explains the basic principle on how gauge field is coupled with matter field.

142
 Chapter 5 Electromagnetism and Special
Relativity

The theory of special relativity originates from electromagnetism. The work of Hendrik
Lorentz through his study of electromagnetism plays a fundamental role. In this chapter, we
will study basic aspects of the coherence of electromagnetism with respect to special relativity.
We will see in particular how Lorentz transformations preserve the form of Maxwell’s equations.

5.1 Lorentz Transformation

5.1.1 Lorentz Group

We will parametrize the spacetime R3,1 via

xµ = (ct, x, y, z) µ = 0, 1, 2, 3.

We consider the following Minkowski distance

X
c2 t2 − x2 − y 2 − z 2 = ηµν xµ xν ,
µ,ν

where  
1 0 0 0
 
0 −1 0 
 0 
η= 
0 0 −1 0 
 
0 0 0 −1

is called the Minkowski metric. The space R3,1 equipped with the Minkowski metric is called
Minkowski spacetime.

Definition 5.1.1. A Lorentz transformation is a linear transformation Λ : R3,1 → R3,1 pre-

serving the Minkowski distance. The Lorentz group is the group of all Lorentz transformations,
denoted by O(3, 1).

In coordinates, if
X
Λ : xµ 7−→ x
eµ = Λ µ ν xν ,
ν

143
then the 4 × 4 matrix Λµ ν gives a Lorentz transformation if it obeys
X
ηµν Λµ ρ Λν σ = ηρσ , ∀ρ, σ = 0, 1, 2, 3
µ,ν

or equivalently in matrix form

ΛT ηΛ = η, ΛT = transpose of Λ.

Example 5.1.2. Let R ∈ O(3) be a 3 × 3 rotation on R3 . Then RT · R = I. It gives a Lorentz

transformation  
1 0 0 0
 
0 
 
Λ= 
0 R 
 
0
which is simply a rotation on (x, y, z) and leaves t invariant. It preserves the Euclidean distance
x2 + y 2 + z 2 , hence preserves the Minkowski distance c2 t2 − x2 − y 2 − z 2 .

Example 5.1.3 (Lorentz Boost). A Lorentz boost along x-direction with velocity v is the linear
transformation 
 ct − xv/c

cet= p

 1 − v 2 /c2



 x − vt

xe= p
 1 − v 2 /c2



ye = y




ze = z

It is direct to check that

(ce
t )2 − x
e2 − ye2 − ze2 = (ct)2 − x2 − y 2 − z 2 .

This corresponds to a Lorentz transformation with matrix form

 
γ −γv/c 0 0
 
−γv/c 0
 γ 0 
Λ= 
 0 0 1 0
 
0 0 0 1

where γ = √ 1
. We can view this as a hyperbolic rotation in the (x0 , x1 )-plane
1−v 2 /c2
 
cosh α − sinh α 0 0
 
− sinh α cosh α 0 0
 
Λ= .
 0 0 1 0
 
0 0 0 1

Similarly, we have Lorentz boosts along y and z directions.

144
 Observe that the relation
ΛT ηΛ = η

implies
det(Λ)2 = 1 =⇒ det(Λ) = ±1.

Definition 5.1.4. The group of proper Lorentz transformations is defined as

SO(3, 1) = {Λ ∈ O(3, 1) | det(Λ) = 1} .

!
1 0
For example, a Lorentz transformation Λ = is a proper Lorentz transformation if
0 R
and only if R ∈ SO(3) is a special orthogonal transformation. Lorentz boosts are always proper
Lorentz transformations.

5.1.2 Transformation of Tensor Fields

Any tensor fields transform naturally under Lorentz transformations. For example, let

Λ : R3,1 −→ R3,1 .
P
Let α = µ αµ (x)dx
µ be a 1-form on R3,1 . Then α is transformed via pull-back by Λ as

α 7−→ Λ∗ (α).

Remark 5.1.5. Strictly speaking, our convention of the transformation α → Λ∗ (α) defines a
right action of Lorentz transformation on differential forms. Indeed, we have

(Λ1 )∗ ((Λ2 )∗ (α)) = (Λ2 ◦ Λ1 )∗ (α).

We could as well define α → (Λ−1 )∗ (α) for a left action. However, since we will not discuss
composition of Lorentz transformations, it does not really matter. We will keep the above
convention, which is convenient for differential forms, to simplify the presentation. It should be
straight-forward to relate different conventions.

In coordinates, if
Λ : R3,1 −→ R3,1
X
xµ 7−→ x
eµ = Λ µ ν xν
ν

then by definition of pull-back

X X X
Λ∗ (α) = αµ (e xµ =
x)de x)Λµ ν dxν =
αµ (e αµ (Λx)Λµ ν dxν .
µ µ,ν µ,ν

Let us denote the transformed form by α̃ = Λ∗ (α)

X X
α= αµ (x)dxµ 7−→ α
e= eµ (x)dxµ .
α
µ µ

145
Expanding the relation α̃ = Λ∗ (α) which says
X X
x)Λµ ν dxν =
αµ (e eµ (x)dxµ ,
α
µ,ν µ

we find the transformation rule in components

X X
eµ (x) =
α x)Λν µ =
αν (e αν (Λx)Λν µ .
ν ν

For simplicity, we will use Einstein summation convention, where repeated upper and lower
indices refer to summation. Then under the Lorentz transform,

xµ 7−→ x
eµ = Λµ ν xν .

A 1-form α = αµ (x)dxµ is transformed to

α 7−→ α
e=α
eµ (x)dxµ ,

where
eµ (x) = αν (e
α x)Λν µ .

Similarly, for a vector field V = V µ (x) ∂x∂ µ , we define its transformation under Λ by

∂ ∂xν ∂

−1 ν ∂
V 7−→ V µ (e
x) = V µ
(e
x ) = V µ
(e
x) Λ µ ∂xν
.
xµ
∂e xµ ∂xν
∂e
Let us denote transformed vector field of V by Ṽ
∂ ∂
V = V µ (x) 7−→ Ve = Ve µ (x) µ .
∂xµ ∂x
Then comparing the above two expressions, we find the transformation rule as

µ
µ
Ve µ (x) = V ν (e
x) Λ−1 ν
= V ν (Λx) Λ−1 ν
.

Remark 5.1.6. In geometric term, this transformation of a vector field is the push-forward

V 7−→ Ve = (Λ−1 )∗ V.

This can be further simplified using the Minkowski metric ηµν . We can use η to turn a
form into a vector field, and vice versa. This is the standard rule to raise or lower tensor indices
via η. For example, we can construct

Vµ = ηµν V ν

which turns a vector field component into a 1-form component. Similarly,

αµ = η µν αν

turns a 1-form component into a vector field component. Here

µν
η µν = η −1

146
is the inverse matrix of η.
   
1 0 0 0 1 0 0 0
   
0 −1 0 0 0 −1 0 0
   
η= , η −1 = .
0 0 −1 0  0 0 −1 0 
   
0 0 0 −1 0 0 0 −1

In general, for an arbitrary tensor field T with components T µ1 ···µk ν1 ···νm , we can raise and
lower their indices by
′ ′ ′ ′
Tµ1 ···µk ν1 ···νm = ηµ1 µ′1 · · · ηµk µ′k η ν1 ν1 · · · η νm νm T µ1 ···µk ν1′ ···νm
′ .

We can also raise or lower part of indices. The formula is similar.

With this notion at hand, we look back at the Lorentz transformation Λ = (Λµ ν )

ΛT ηΛ = η, or ηµν Λµ ρ Λν σ = ηρσ .

This is equivalently described as

Λµ ρ Λµ σ = δ ρ σ .

In other words, the inverse matrix Λ−1 is precisely

µ
Λν µ = Λ−1 ν
.

Warning: Be careful about the position of indices. (Λν µ ) is the inverse matrix of (Λµ ν ).
As a result, we can write the Lorentz transformation of a vector field by

V µ (x) 7−→ Ve µ (x) = V ν (e

x)Λν µ .

For a general tensor field

∂ ∂
T = T µ1 ···µk ν1 ···νm (x) µ
⊗ · · · ⊗ µ ⊗ dxν1 ⊗ · · · ⊗ dxνm ,
∂x 1 ∂x k
it transforms under a Lorentz transformation Λ : x 7→ x
e by

T 7−→ Te

which is described in components via

Teµ1 ···µk ν1 ···νm (x) = T µ1 ···µk ν1′ ···νm

′ ′ ′ ′
x)Λµ′1 µ1 · · · Λµ′k µk Λν1 ν1 · · · Λνm νm .
′ (e

This is the transformation rule for general tensor fields. Note that this is consistent if we raise
or lower some indices of Te (Exercise: Check this).

147
5.1.3 Invariance of Inner Contraction

We can change the type of a tensor field by contracting their indices. For example, consider
a tensor field
∂
T = T µν ⊗ dxν .
∂xµ
Then we can get a scalar (function) by

f (x) = T µ µ (x).

This is compatible with Lorentz transformation. In fact, under Λ : x 7→ x

e, the tensor field T is
e
transformed to T where
Teµ ν (x) = T µ ν ′ (e
′ ′
x)Λµ′ µ Λν ν

and the scalar function f is transformed to f˜ where

fe(x) = f (e
x).

Since
Teµ µ (x) = T µ ν ′ (e
′ ′ ′ ′
x)Λµ′ µ Λν µ = T µ ν ′ (e
x)δµ′ ν = T µ µ (e
x),

we see that fe is again given by

fe = Teµ µ .

This consistency follows essentially from the invariance of Minkowski metric under Lorentz
transformations.

5.2 Lorentz Invariance of Maxwell’s Equations

Now we study the Lorentz transformation of electromagnetic fields and explore the invari-
ance property of Maxwell’s equations.

5.2.1 Transformation of Electromagnetic Fields

We first describe how electromagnetic fields change under Lorentz transformations. Let

Λ : R3,1 −→ R3,1

be a Lorentz transformation which in coordinates is

Λ : xµ 7−→ x
eµ = Λµ ν xν .

The electric and magnetic fields are organized into a 2-form F

F = (Ex dx + Ey dy + Ez dz) ∧ dt + Bx dy ∧ dz + By dz ∧ dx + Bz dx ∧ dy.

Let us write F into tensor notation as

1X
F = Fµν dxµ ∧ dxν , Fµν = −Fνµ .
2 µ,ν

148
The tensor components Fµν are expressed in matrix form
 
0 −Ex /c −Ey /c −Ez /c ← µ = 0
 
E /c B −B 
 x 0 z y  1
Fµν =  
Ey /c −Bz 0 Bx  2
 
Ez /c By −Bx 0 3
↑
ν=0 1 2 3

Under the Lorentz transformation Λ, the tensor field Fµν is transformed to

Feµν (x) = Fρσ (e

x)Λρ µ Λσ ν = Fρσ (Λx)Λρ µ Λσ ν .

This allows us to read off the transformation of electric and magnetic fields via components.

Example 5.2.1 (Lorentz boost). Consider the Lorentz boost Λ in the x-direction
 
γ −γv/c 0 0
 
−γv/c 0 0
 γ 
Λ= 
 0 0 1 0
 
0 0 0 1

where γ = √ 1
. Then
1−v 2 /c2
   
γ −γv/c 0 0 −Ex /c −Ey /c −Ez /c
0 γ −γv/c 0 0
   
−γv/c 0 0  −By   0 0
 γ  Ex /c 0 Bz  −γv/c γ 
Feµν =   
 0 0  
1 0 Ey /c −Bz 0  
Bx   0 0 1 0
 
0 0 0 1 Ez /c By −Bx 0 0 0 0 1
 
0 −Ex /c −γEy /c − γBz v/c −γEz /c + γBy v/c
 
 Ex /c γEy v/c2 + γBz γEz v/c2 − γBy 
 0 
= .
γEy /c + γBz v/c −γEy v/c2 − γBz 0 Bx 
 
γEz /c − γBy v/c −γEz v/c2 + γBy −Bx 0

It reads in components as 

Ee x = Ex





 e y = γ (Ey + vBz )

E



Ee z = γ (Ez − vBy )

 e x = Bx

B




 e y = γ By − vEz /c2

B



B

e z = γ Bz + vEy /c2

where fields on the left hand side are evaluated at point x and fields on the right hand side are
e = Λx.
evaluated at the point x
One important observation is that Lorentz transformation will mix electric and magnetic
fields in general.

149
5.2.2 Transformation of Charge-Current Density

The charge and current densities are organized into a 1-form on R3,1

J = ρ/ε0 dt − µ0 (jx dx + jy dy + jz dz).

Under the Lorentz transformation Λ : R3,1 → R3,1 , it transforms J = Jµ dxµ to Je = Jeµ dxµ by

Je = Λ∗ (J)

or in components,
Jeµ (x) = Jρ (e
x)Λρ µ = Jρ (Λx)Λρ µ .

Example 5.2.2 (Lorentz boost). Under the Lorentz boost

 
γ −γv/c 0 0
 
−γv/c 0 0
 γ 
Λ= ,
 0 0 1 0
 
0 0 0 1

we find 



 ρe = γ ρ + jx v/c2




ejx = γ (jx + ρv)

ejy = jy





ejz = jz

Again, fields on the left hand side are evaluated at point x and fields on the right hand side are
e = Λx.
evaluated at the point x

5.2.3 Transformation of Maxwell’s Equations

The Maxwell’s equations are 

dF = 0
d(?F ) = ?J

To understand the Lorentz transformation of Maxwell’s equations, we need to know how the
Hodge star ? intertwines with the Lorentz transformation. Note that we use the notation ? for
Hodge star in this subsection, in order to distinguish with the ∗-symbol in pull-backs.

Proposition 5.2.3. Let Λ : R3,1 → R3,1 be a proper Lorentz transformation. Then for any
differential form α ∈ Ω• (R3,1 ), we have

Λ∗ (?α) = ?Λ∗ (α).

In other words, the Hodge star ? commutes with the pull-back via Λ.

150
Proof: It is enough to check on the basis

1, dxµ , dxµ ∧ dxν , dxµ ∧ dxν ∧ dxρ , dx0 ∧ dx1 ∧ dx2 ∧ dx3 .

We check one case for α = 1. The others are similar.

Λ∗ (?1) = Λ∗ (dx0 ∧ dx1 ∧ dx2 ∧ dx3 ) = det(Λ)(dx0 ∧ dx1 ∧ dx2 ∧ dx3 )

and
?Λ∗ (1) = ?1 = dx0 ∧ dx1 ∧ dx2 ∧ dx3 .

For a proper Lorentz transformation, det(Λ) = 1. Thus the above two terms are equal.

Remark 5.2.4. For a general Lorentz transformation Λ, we have

Λ∗ (?α) = (det Λ) ? Λ∗ (α).

The proof is similar.

We will now focus on proper Lorentz transformations Λ ∈ SO(3, 1), which preserves the
orientation on R3,1 . Under the Lorentz transformation by Λ, we have

Fe = Λ∗ (F )
Je = Λ∗ (J)

Since the pull-back Λ∗ commutes with d, and also commutes with the Hodge star ? as in
Proposition 5.2.3, we find
 
dF = 0 dFe = 0
pull-back
=====⇒
d(?F ) = ?J via Λ d(?Fe) = ?Je

Thus the transformed electromagnetic field and the transformed charge-current density again
satisfy the Maxwell’s equations! In other words, Lorentz transformations will transform solu-
tions of Maxwell’s equations to solutions of Maxwell’s equations.
Another illustrating way to see this is to consider the Maxwell action
ˆ  
1 1
SM [A, J] = F ∧ ?F + ?J ∧ A .
µ0 R3,1 2
Under the Lorentz transformation Λ, the gauge 1-form A = Aµ dxµ transforms as

eµ (x)dxµ = Λ∗ (A)
A 7−→ A

eµ (x) = Aρ (e
or in components A x)Λρ µ . Then
ˆ  
e e 1 1e e e e
SM [A, J] = F ∧ ?F + ?J ∧ A
µ0 R3,1 2
ˆ  
1 1
= Λ∗ F ∧ ?F + ?J ∧ A
µ0 R3,1 2
ˆ  
1 1
= F ∧ ?F + ?J ∧ A = SM [A, J].
µ0 R3,1 2

151
Here in the second line, we have used the invariance of integral under orientation-preserving
e is a critical point
diffeomorphisms. It follows that A is a critical point of SM [A, J] if and only if A
e J],
of SM [A, e i.e. Lorentz transformations send solutions of Maxwell’s equations to solutions.

Example 5.2.5. Consider the electromagnetic field of a static point particle of charge q at the
origin
~ = q ~r ~ = 0.
E , B
4πε0 r3
Under the Lorentz boost Λ : R3,1 → R3,1
 
γ −γv/c 0 0
 
−γv/c 0 0
 γ 
Λ= 
 0 0 1 0
 
0 0 0 1

with 
 ct − xv/c

cet= p

 1 − v 2 /c2



 x − vt

xe= p
 1 − v 2 /c2



ye = y




ze = z

We find the transformed fields by


 ex = q x e q γ(x − vt)

 E =

 4πε0 re3 4πε0 (γ 2 (x − vt)2 + y 2 + z 2 )3/2





 e y = q γe y q γy

 E =

 4πε0 re3 4πε0 (γ 2 (x − vt)2 + y 2 + z 2 )3/2




Ee z = q γe z
=
q γz
4πε0 re3 4πε0 (γ 2 (x − vt)2 + y 2 + z 2 )3/2



 ex = 0
B






 e y = − q γv ze = q
B
γvz/c2

 4πε0 c2 re3 4πε0 (γ 2 (x − vt)2 + y 2 + z 2 )3/2






Be z = q γv ye = q γvy/c2
4πε0 c2 re3 4πε0 (γ 2 (x − vt)2 + y 2 + z 2 )3/2

On the other hand, the charge-current densities


ρ = qδ(~r )
~j = 0

is transformed to
  

ρe = γqδ ~re = γqδ(γ(x − vt))δ(y)δ(z) = qδ(x − vt)δ(y)δ(z)



ejx = qvδ(x − vt)δ(y)δ(z)




ejy = ejz = 0

152
This describes a point charge moving in the x-direction with constant velocity v. We have re-
covered the Liénard-Wiechert electric and magnetic fields for a point charge of constant velocity
described in Example 5.2.5.

5.3 Relativistic Lorentz Force Law

We discuss relativistic version of Lorentz force law.

5.3.1 Non-relativistic Charged Particle

A particle of charge q moving with velocity ~v in the background of electromagnetic fields

will experience a force via the Lorentz force law
 
F~ = q E~ + ~v × B
~ .

Assume the particle has mass m, and the trajectory is described by ~r (t) = (x(t), y(t), z(t)).
Then the equation of motion is described via Newton’s law by
 
m~r¨ = q E~ + ~r˙ × B
~ .

This equation can be described by a Lagrangian

1
L = m~r˙ 2 − qφ(~r, t) + q A(~
~ r, t) · ~r˙.
2
~ is the vector potential. Then the Euler-Lagrange equation
Here φ is the scalar potential and A
of the action ˆ  
1 ˙2 ~ · ~r˙
S[~r ] = dt m~r − qφ + q A
2
gives precisely (for example, see [10, Example 1.2.3] in this series) the above Lorentz force law
 
δS = 0 =⇒ m~r¨ = q E~ + ~r˙ × B
~ .

5.3.2 Relativistic Charged Particle

Let us now work with a relativistic particle of mass m and charge q, whose trajectory in
the spacetime is described by

γ = (x0 (σ), x1 (σ), x2 (σ), x3 (σ)).

Here σ is a parameter of the curve γ in R3,1 .

The relativistic action for the charged particle in the electromagnetic background is
ˆ ˆ
p
S[γ] = −mc µ ν
ηµν dx dx + q A.
γ γ

Let us explain these two terms.

153
 ⃝
1 In terms of a parameter σ on γ, the first term is expressed explicitly as
ˆ r
dxµ dxν
−mc dσ ηµν .
dσ dσ
It is clear from the expression that this term does not depend on the choice of parameter σ.
e of the curve γ,
For another parameter σ
r s  2 ˆ  r r
ˆ ˆ ˆ
µ
dx dx ν dxµ dxν de
σ  de
σ  dx µ dxν dxµ dxν
dσ ηµν = dσ ηµν = dσ   ηµν = de
σ ηµν .
dσ dσ deσ deσ dσ dσ de
σ de σ de
σ deσ
This also explains why we write the action as
ˆ
p
−mc ηµν dxµ dxν
γ

without using any explicit parameter.

If we use the time t as a parameter, then the curve is parameterized as

γ(t) = (x0 (t), x1 (t), x2 (t), x3 (t))

= (ct, x(t), y(t), z(t))
= (ct, ~r (t))

where ~r (t) = (x(t), y(t), z(t)). Then the action becomes

ˆ q ˆ q
2 ˙
−mc dt c − ~r = −mc 2 2
dt 1 − ~r˙ 2 /c2
ˆ !
1 ~r˙ 2
= −mc2 dt 1 − + ···
2 c2
ˆ
1
= const + m dt ~r˙ 2 + · · ·
2

Here terms in · · · involve higher orders in ~r˙ /c, which become zero in the non-relativistic limit.
It follows that the term ˆ
p
−mc ηµν dxµ dxν
γ
´
becomes the standard kinetic term 1
2m dt ~γ˙ 2 in the non-relativistic limit, up to a constant
which is irrelevant for equation of motion. Note that ηµν dxµ dxν is always nonnegative since
particles can not travel faster than the speed of light.
⃝
2 In the second term,

A = −φdt + Ax dx + Ay dy + Az dz
´
is the gauge 1-form, which can be integrated along a curve γ in R3,1 . This is q γ A.
If we parametrize the curve γ via time t along the curve

γ(t) = (ct, ~r(t)),

then ˆ ˆ   ˆ  
dx dy dz ~ · ~r˙ .
q A=q dt −φ + Ax + Ay + Az = dt −qφ + q A
γ dt dt dt

154
This coincides with the potential term in the case of non-relativistic charged particle.
The above shows that
ˆ ˆ
p
S[γ] = −mc ηµν dxµ dxν + q A
γ γ

is indeed a natural relativistic generation of action functional for charged particle.

Let us now work out the equation of motion for S[γ], which would describe the relativistic
Lorentz force law. Let us parametrize the curve as

γ(σ) = {xµ (σ)}

and the action becomes

ˆ r ˆ
dxµ dxν dxµ
S = −mc dσ ηµν +q dσ Aµ .
dσ dσ dσ
Consider an arbitrary variation
γ −→ γ + δγ.

The variation of the action is

ˆ µ dxν ˆ ˆ
ηµν dδx dδxµ dxν
δS = −mc dσ q dσ dσ
+ q dσ Aµ + q dσ (∂µ Aν )δxµ
ηµν dx
µ dxν dσ dσ
dσ dσ
 
ˆ dxν ˆ
µ d  ηµν dσ
 dxν
= mc dσ δx q + q dσ δxµ (∂µ Aν − ∂ν Aµ )
dσ ηµν dx
µ dxν dσ
dσ dσ
   
ˆ  d dxν ν
mcη dx
= dσ δxµ  q µν dσ  + qFµν .
 dσ µ ν dσ 
η dx dx
µν dσ dσ

The equation of motion is obtained by asking δS = 0 for an arbitrary variation δxµ . This leads
to
d dxν
Pµ = −qFµν
dσ dσ
where Pµ = ηµν P ν and
dxν
P = mc q
ν
. dσ
ρ dxν
ηρν dx
dσ dσ

Note that P µ is invariant under the change of the parametrization σ, by the same reason as
above. It is precisely the relativistic momentum. In fact, let us choose the time t as the
parameter, then  
ẋµ mc m~r˙
P µ = mc p = q ,q 
˙
c − ~r
2 2 ˙ ˙
1 − ~r /c
2 2 1 − ~r /c
2 2

where ~r = (x(t), y(t), z(t)). We see that

P µ Pµ = (P 0 )2 − (P 1 )2 − (P 2 )2 − (P 3 )2 = m2 c2

155
which is the familiar result on relativistic 4-momentum. cP 0 gives the relativistic energy

mc2 1
q = mc2 + m~r˙ 2 + · · ·
2
1 − ~r˙ 2 /c2

r˙
and 12 m~r˙ 2 is the non-relativistic kinetic energy. The non-relativistic limit of √ m⃗
˙
gives m~r˙ ,
r 2 /c2
1−⃗
which is the standard momentum vector.
It is illustrating to unpack the extra equation coming from µ = 0:
dP0 dxν ~ · ~r˙ /c,
= −qF0ν = qE
dt dt
i.e.
d(cP0 ) ~ · ~r˙.
= qE
dt
Since cP0 is the energy, this equation simply tells the change of kinetic energy when work is
done by an electric field.

156
 Bibliography

[1] Berger, Mitchell A., and George B. Field. The topological properties of magnetic helicity.
Journal of Fluid Mechanics 147 (1984): 133-148.

[2] Buchwald, Jed Z. Electrodynamics from Thomson and Maxwell to Hertz. The Oxford Hand-
book of the History of Physics. Oxford University Press; 2013.

[3] Butcher, Ginger. Tour of the electromagnetic spectrum. Government Printing Office, 2016.

[4] Eguchi, Tohru, Peter B. Gilkey, and Andrew J. Hanson. Gravitation, gauge theories and
differential geometry. Physics reports 66.6 (1980): 213-393.

[5] Garrity, Thomas A. Electricity and Magnetism for Mathematicians: A Guided Path from
Maxwell’s Equations to Yang-Mills. Cambridge University Press, 2015.

[6] Göckeler, Meinulf, and Thomas Schücker. Differential geometry, gauge theories, and gravity.
Cambridge University Press, 1989.

[7] Griffiths, David J., and Reed Colleger. Introduction to Electrodynamics. Prentice Hall Up-
per Saddle River. New Jersey 7458 (1999).

[8] Jackson, John David. Classical electrodynamics. John Wiley & Sons, 1999.

[9] Jackson, John David, and Lev Borisovich Okun. Historical roots of gauge invariance. Re-
views of Modern Physics 73.3 (2001): 663.

[10] Li, Si. Classical Mechanics and Geometry. International Press of Boston, 2023.

[11] Maxwell, James Clerk. A treatise on electricity and magnetism. Oxford: Clarendon Press,
1873.

[12] Nakahara, Mikio. Geometry, Topology and Physics. CRC press, 2003.

[13] Nayfeh, Munir H., and Morton K. Brussel. Electricity and magnetism. Courier Dover Pub-
lications, 2015.

[14] O’Raifeartaigh, Lochlainn. The dawning of gauge theory. Vol. 106. Princeton University
Press, 1997.

157
[15] Schwinger, Julian, Lester L. Deraad Jr., Kimball A. Milton, Wu-yang Tsai. Classical elec-
trodynamics. Westview Press, 1998.

[16] Tong, David. Electromagnetism. University of Cambridge Mathematical Tripos. 2015.

[17] Weyl, Hermann. Space–Time–Matter. Dutton, 1922.

[18] Zangwill, Andrew. Modern electrodynamics. Cambridge University Press, 2013.

158