The Physics of Fields and Flows

Module PX2131

MO: Dr. Nicolas Peretto

This set of notes has been put together by Dr Nicolas Peretto for the 2nd year module The Physics
of Fields and Flows. It is largely inspired from the hand-written notes of Dr Egor Muljarov who
taught that module for a number of years.
Second edition, August 2021
 Contents

1 Some fundamentals of vector calculus . . . . . . . . . . . . . . . . . . . . . . . . . 7

1.1 Vectors 7
1.1.1 Some basic definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.1.2 The dot product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.1.3 The cross product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.1.4 Scalar triple product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.1.5 Vector triple product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2 Differentiation 10
1.2.1 Derivatives of mono-variable functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2.2 Product rule of differentiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2.3 The chain rule of differentiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2.4 Derivatives of multivariate functions: Partial and total differentiation . . . . . . . . 11
1.3 The del operator: ~∇ 11
1.3.1 Chain rule for gradient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.3.2 Product rule for gradient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.3.3 Product rule for divergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.3.4 Product rule for curl . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.3.5 Successive applications of the del operator . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.4 Integration 14
1.4.1 Curvilinear integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.4.2 Surface integrals in the plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.4.3 Volume integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.5 Coordinate systems and transformations 16
1.5.1 The position vector differential d~r . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.5.2 Differential elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.5.3 Gradient, divergence, curl, and Laplacian . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2 Fundamental theorems of vector calculus . . . . . . . . . . . . . . . . . . . . . 21
2.1 The divergence theorem 21
2.1.1 Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.1.2 Proof for a box . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.1.3 Proof for any arbitrary volume . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.2 Stokes’ theorem 23
2.2.1 Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.2.2 Proof for a rectangle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.2.3 Proof for two joint rectangles and more . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.2.4 Green’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3 The gradient theorem 25
2.3.1 Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.3.2 Proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.4 Potentials and the Helmholtz theorem 26
2.4.1 Scalar potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.4.2 Vector potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.4.3 Helmholtz theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

3 Electrostatics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.1 Gauss’s law 29
3.2 Electrostatic potential 31
3.3 Electrostatic energy 32
3.3.1 Derivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.3.2 A note on self- and interaction energies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.4 Electric dipole 35
3.5 Dielectrics 36
3.6 The origin of Coulomb’s law 39
3.6.1 Green’s function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.6.2 Proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.6.3 Green’s function of Laplace operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

4 Magnetostatics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.1 Ampere’s circuital law 41
4.2 Vector potential and Gauge invariance 42
4.3 General solution of magnetostatics: Biot-Savart’s law. 43
4.4 Lorentz force and the origin of magnetic fields 44
4.4.1 The Lorentz force and some characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.4.2 Special relativity and the origin of magnetic fields . . . . . . . . . . . . . . . . . . . . . . 45
4.5 Magnets and the magnetostatics properties of matter 47

5 Electromagnetism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
5.1 Faraday’s law of induction 51
5.2 Maxwell’s equations 52
5.2.1 The continuity equation of electric charges . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
5.2.2 Maxwell’s correction of Ampere’s law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5.3 Electromagnetic waves inn vacuum 54
5.3.1 Derivation of the electromagnetic wave equations . . . . . . . . . . . . . . . . . . . . . 54
5.3.2 Solution of the wave equation in 1D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
5.3.3 Solutions of the wave equation in 3D and the case of plane waves . . . . . . . . 55
5.3.4 Polarisation of light . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.4 General solution of Maxwell’s equations 57
5.4.1 Lorenz gauge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
5.4.2 D’Alembert formulation of Maxwell equations . . . . . . . . . . . . . . . . . . . . . . . . . 58
5.4.3 Solutions of the 4D Poisson equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
5.5 Generating electromagnetic waves 59
5.6 The energy density of electromagnetic fields 61
5.7 Electromagnetic waves in a medium 61
5.7.1 Solutions of Maxwell equations in a medium, and the refractive index . . . . . . 62
5.7.2 Maxwell’s boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
5.7.3 Reflection and transmission of electromagnetic waves: Normal incidence . . . 64
5.7.4 Reflection and transmission of electromagnetic waves: Non-normal incidence 64

6 Fluid mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
6.1 The continuity equation for mass 67
6.2 Hydrostatics 68
6.3 Stream functions of incompressible fluids 70
6.4 Euler’s equation 71
6.5 Bernouilli’s law 72
6.6 Vortices 73
 1. Some fundamentals of vector calculus

In this chapter we will introduce some elementary notions of vector calculus that are later needed
when discussing the origin of some of the most fundamental equations in physics.

1.1 Vectors
1.1.1 Some basic definitions
In cartesian coordinates, a vector can be expressed as:

~a = ax~i + ay~j + az~k (1.1)

where (~i, ~j,~k) are the three units vectors of the three orthogonal
axes (x, y, z), and (ax , ay , az ) are the three coordinates of vector ~a.
Note that the cartesian unit vectors can also be noted as (~ux ,~uy ,~uz ).
Both notations will be used in the context of this module. The
coordinates of vector ~c defined as:
Figure 1.1: Sketch representing
~c = ~a +~b (1.2) a vector ~a and its three coordi-
nates in the cartesian frame.
are given by:

cx = ax + bx
cy = ay + by (1.3)
cz = az + bz

The length of vector ~a, also called norm or magnitude, is given by:
q
|~a| = a = a2x + a2y + a2z (1.4)
 8 Chapter 1. Some fundamentals of vector calculus

The unit vector ~ua along the axis defined by vector ~a is given by:

~a
~ua = (1.5)
|~a|

An important vector that we will refer to frequently is the so called position vector~r:

~r = x~i + y~j + z~k (1.6)

The position vector starts at the origin of the coordinate system and ends at the point of coordinates
(x, y, z). Its magnitude r is therefore given by:
p
r= x2 + y2 + z2 (1.7)

Often, when one is dealing with a function f of (x, y, z), one can write for simplicity:

f (x, y, z) = f (~r) (1.8)

Note that this is different from:

p
f (r) = f ( x2 + y2 + z2 ) (1.9)

1.1.2 The dot product

The scalar product, also known as the dot product, of two vectors ~a and ~b is defined as:

~a ·~b = ax bx + ay by + az bz (1.10)

The result of the dot product of two vectors is a scalar (i.e. a number).
This definition implies that the dot product is commutative (~a ·~b = ~b ·~a)
and distributive (~a · (~b +~c) = ~a ·~b +~a ·~c). In terms of geometry the dot
product corresponds to the projection of ~a on ~b (or the other way around).
If θ is the angle between ~a and ~b then we have:

~a ·~b = ab cos θ (1.11)

One can show that Eq. (1.10) and (1.11) are indeed equivalent. Let’s Figure 1.2: Sketch repre-
define vector ~c as ~c = ~a −~b. Then the law of cosine (https://en. senting the dot product of
wikipedia.org/wiki/Law_of_cosines) tells us that: two vectors

|~c|2 = |~a|2 + |~b|2 − 2|~a||~b| cos θ (1.12)

In parallel, from the definition given in Eq. (1.10), we also have the following relations:

|~c|2 =~c.~c
= (~a −~b) · (~a −~b)
(1.13)
= ~a ·~a −~a ·~b −~b ·~a +~b ·~b
= |~a|2 + |~b|2 − 2~a ·~b

Combining Eq. (1.12) and the last of (1.13), we obtain:

~a ·~b = |~a||~b| cos θ = ab cos θ (1.14)

 1.1 Vectors 9

1.1.3 The cross product

The vector product, also know as the cross product, of two vectors ~a and ~b is defined as:
   
ax bx
~a ×~b = ay  × by  = (ay bz − az by )~i + (az bx − ax bz )~j + (ax by − ay bx )~k (1.15)
az bz

The result of the cross product of two vectors is a vector. Using

the definitions of the dot product (Eq. 1.10 and 1.11), one can show
that the vector resulting from the cross product of two vectors is
perpendicular to both of them. Equation 1.15 implies that the cross
product is anti-commutative (~a ×~b = −~b ×~a).
Also ~a ×~b = 0 implies that ~a and ~b are co-linear (parallel or anti-
parallel).
In terms of geometry, the magnitude of |~a ×~b| corresponds to the Figure 1.3: Sketch representing
area S of the parallelogram spanned by ~a and ~b: the cross product of two vectors

S = |~a ×~b| = ab sin θ (1.16)

One can show that the two definitions of the cross product (Eq. 1.15 and 1.16) are equivalent. This
can be done by first showing that

|~a ×~b|2 = |~a|2 |~b|2 − (|~a| · |~b|)2 (1.17)

Relation (1.17) can be derived by using the definitions provided by Eq. (1.7) and (1.15). Then by
subbing-in Eq. (1.11) in Eq. (1.17) we obtain the following:

|~a ×~b|2 = |~a|2 |~b|2 (1 − cos2 θ )

(1.18)
= |~a|2 |~b|2 sin2 θ

1.1.4 Scalar triple product

And therefore:

|~a ×~b| = ab sin θ (1.19)

The scalar triple product is defined as:

ax bx cx
~a ·~b ×~c = ay by cy
az bz cz
= ax (by cz − bz cy ) + ay (bz cx − bx cz ) + az (bx cy − by cx )
(1.20)

The result of a scalar triple product is a scalar (number). Note

also that the cross product has "priority" over the dot product in
Eq. (1.20), the expression would not make sense otherwise. In Figure 1.4: Sketch representing
terms of geometry, the scalar triple product corresponds to the the scalar triple product between
volume V of the parallelepiped spanned by vectors ~a, ~b, ~c (see three vectors
Fig. 1.4).
If ~a ·~b ×~c = 0 then ~a, ~b, ~c are co-planar (all three vectors lie in the
same plane).
 10 Chapter 1. Some fundamentals of vector calculus

Some important properties of the scalar triple product are:

~a ·~b ×~c =~c ·~a ×~b = ~b ·~c ×~a (cyclic permutations) (1.21)

and:

~a ·~b ×~c = ~a ×~b ·~c (1.22)

1.1.5 Vector triple product

The vector triple product is defined as ~a × (~b ×~c) and is related to the dot product via:

~a × (~b ×~c) = ~b(~a ·~c) − (~b ·~a)~c (1.23)

Exercise 1.1 Show that Eq. (1.23) is correct by first picking up a frame so that ~a = (ax , ay , az ),
~b = (bx , 0, 0), and ~c = (cx , cy , 0), and then by expanding each term on both sides. 

1.2 Differentiation
1.2.1 Derivatives of mono-variable functions
For a function f with unique variable x, the derivative f 0 is defined as:

df f (x + ∆x) − f (x)
f 0 (x) = = lim (1.24)
dx ∆x→0 ∆x

The infinitesimal quantity d f is known as the differential of f (x).

1.2.2 Product rule of differentiation

For two functions f (x) and g(x), the derivative of f (x)g(x) is:

( f g)0 = f 0 g + f g0 (1.25)

Note that one can use the same product rule for quotient f /h. Indeed, if one sets g = 1/h, then the
chain rule gives us:
 0
f f 0 f h0
= − 2 (1.26)
h h h

One can generalise the product rule as follows:

( f gh)0 = f 0 gh + f g0 h + f gh0 (1.27)

and

( f g)00 = f 00 g + 2 f 0 g0 + f g00 (1.28)

 1.3 The del operator: ~∇ 11

1.2.3 The chain rule of differentiation

If f is a function of a function, e.g. f (g(x)), then we have:

df d f dg
= (1.29)
dx dg dx

The chain rule can be extended to as many arguments as necessary. For instance is x is a function
of t, the chain rule tells us that:
df d f dg dx
= (1.30)
dt dg dx dt
Equation (1.30) can sometimes be written as:
df
= f 0 g0 x0 (1.31)
dt

1.2.4 Derivatives of multivariate functions: Partial and total differentiation

The derivative with respect to one of the variables of a multivariate function is called a partial
derivative. If one has a function f (x, y, z), then the partial derivative of f with respect to x is written:

∂f f (x + ∆x, y, z) − f (x, y, z)
∂x = = lim (1.32)
∂ x ∆x→0 ∆x

In this operation, the two variables y and z are treated as is they were constants.

The total differential of the multivariate function f (x, y, z) is defined as:

∂f ∂f ∂f
df = dx + dy + dz (1.33)
∂x ∂y ∂z

which includes an incremental change of f due to dx, dy and dz, which are independent infinitesimal
changes of x, y, and z. d f therefore to the infinitesimal change of f when infinitesimally changing
x, y and z independently.

1.3 The del operator: ~∇

The del operator, noted ~∇, is a vector and a differential operator at the same time. Often it is noted
∇ and not ~∇. Here we will keep the vector notation. In cartesian coordinates the del operator is
defined as:

~∇ =~i ∂ + ~j ∂ +~k ∂ =~i∂x + ~j∂y +~k∂z (1.34)

∂x ∂y ∂z

There are three key operations using ~∇ that we will use extensively in the rest of the module. These
are:

Gradient : ~∇ f = ∂x f~i + ∂y f ~j + ∂z f~k (1.35)

Divergence : ~∇ · ~F = ∂x Fx + ∂y Fy + ∂z Fz (1.36)
 12 Chapter 1. Some fundamentals of vector calculus

   
∂x Fx
Curl : ~∇ × ~F = ∂y  × Fy  (1.37)
∂z Fz

Here, f (x, y, z) is a scalar field, while ~F(x, y, z) is a vector field, with ~F = Fx (x, y, z)~i + Fy (x, y, z)~j +
Fz (x, y, z)~k.
When using the del operator, you are performing a vector operation and a differential operation at
the same time. Therefore, all rules of vector algebra apply and all rules of differentiations apply.
We note that the total differential as defined in Eq. (1.33) can be simply written as:

d f = ~∇ · d~r (1.38)

where d~r is the infinitesimal variation of the position vector~r defined as:

d~r = dx~i + dy~j + dz~k (1.39)

1.3.1 Chain rule for gradient

The chain rule for gradients is the generalisation of the standard one expressed in Eqs. (1.29) and
(1.30). The only different is that the del operator has three components (∂x , ∂y , ∂z ). For instance we
have:

~∇ f (g(x, y, z)) = d f ~∇g (1.40)

dg
p
Now let’s assume that g is only function of r (= x2 + y2 + z2 ), then one can rewrite Eq. (1.40)
as:

~∇ f (g(r)) = d f ~∇g
dg
d f dg ~
= ∇r (1.41)
dg dr
d f dg ~r
=
dg dr r

1.3.2 Product rule for gradient

The product rule for gradients is very similar to that presented in Eq. (1.25). It is expressed as:

~∇(g f ) = g~∇ f + f ~∇g (1.42)

Note that, to avoid any confusion, we always place the term that does not undergo a differentiation
(i.e. g for the first of the right-hand-side terms and f for the second) before the del operator. We can
also make use of brackets if necessary (i.e. ~∇(g f ) = g~∇( f ) + f ~∇(g)).

1.3.3 Product rule for divergence

The product rule for the divergence of g~F, where g is a function of (x,y,z) and ~F a vector with
coordinates (Fx , Fy , Fz ) each function of (x,y,z), is given by:

~∇ · (g~F) = g(~∇ · ~F) + (~∇g) · ~F (1.43)

 1.3 The del operator: ~∇ 13

This relation can be derived as follows:

~∇ · (g~F) = ∂x (gFx ) + ∂y (gFy ) + ∂z (gFz )

= g∂x Fx + Fx ∂x g + g∂y Fy + Fy ∂y g + g∂z Fz + Fz ∂z g
= g (∂x Fx + ∂y Fy + ∂y Fy ) + Fx ∂x g + Fy ∂y g + Fz ∂z g (1.44)
   
= g(~∇ · ~F) + ~i∂x g + ~j∂y g +~k∂z g · Fx~i + Fy~j + Fz~k

= g(~∇ · ~F) + (~∇g) · ~F

1.3.4 Product rule for curl

Similarly to the product rule for divergence, the product rule for curl is given by:

~∇ × (g~F) = g(~∇ × ~F) + (~∇g) × ~F (1.45)

Exercise 1.2 Using a very similar approach to that used for the derivation of the product rule
of the divergence, derive the proof for Eq. (1.45). 

1.3.5 Successive applications of the del operator

Many fundamental equations in physics involve the successive application of the del operator.
These are:

(a) ~∇ · ~∇ f
(b) ~∇ × ~∇ f
(c) ~∇~∇ · ~F (1.46)
(d) ~∇ · ~∇ × ~F
(e) ~∇ × (~∇ × ~F)

Exercise 1.3 Determine which of these expressions are vectors and which ones are scalars. For
the first four expressions you may place brackets around the relevant terms. 

Equation (1.46a) corresponds to the divergence of the gradient of f . By developing the expression
we obtain:
~∇ · ~∇ f = ∂ 2 f + ∂ 2 f + ∂ 2 f (1.47)
x y z

∂2 f
where ∂x2 f = ∂ 2x
is the second derivative of f with respect to x. Equation 1.47 can be rewritten as:

~∇ · ~∇ f = ∇2 f = ∆ f (1.48)

where ∆ is known as the scalar Laplace operator, or scalar Laplacian.

Equation (1.46b) can be rewritten as:
   
∂x ∂x f
~∇ × ~∇ f = ∂y  × ∂y f  = (∂y ∂z f − ∂z ∂y f )~i + (∂z ∂x f − ∂x ∂z f ) ~j + (∂x ∂y f − ∂y ∂x f )~k
∂z ∂z f
=0
 14 Chapter 1. Some fundamentals of vector calculus

(1.49)

This is true as long as the second derivatives of f are continuous in which case ∂x ∂y = ∂y ∂x (and
same for all other combinations). Equation (1.49) tells us that the curl of a gradient is always zero.
Vector fields ~F that satisfy the following relation:

~∇ × ~F = 0 (1.50)

are said to be irrotational vector fields. With that definition, all gradient vector fields (i.e. those
such that ~F = ~∇ f ) are irrotational vector fields.

Equation (1.46d) is a triple scalar product (see Eq. (1.20), and performing a similar calculation as
in Eq. (1.49) one can show that:
~∇ · ~∇ × ~F = 0 (1.51)

This means that the divergence of a curl always vanishes. Vector fields ~F that satisfy the following
relation:

~∇ · ~F = 0 (1.52)

are said to be solenoidal vector fields.

Exercise 1.4 Show that Equation (1.51) is verified. 

Finally, Equation (1.46e) and c are related via the triple vector product relation (Eq. 1.23) by:
   
~∇ × (~∇ × ~F) = ~∇ ~∇ · ~F − ~∇ · ~∇ ~F (1.53)

Note that the vector ~F from the second term on the right-hand-side has been moved to the right.
This is allowed since ~∇ · ~∇ is a scalar and α ~F = ~Fα where α is a scalar (i.e. number). Again the
brackets in the right hand side are not necessary since there only one way the calculation can be
made so that it makes sense. Equation (1.53) can be rewritten as:
 
~∇ × (~∇ × ~F) = ~∇ ~∇ · ~F − ∆~F (1.54)

Note that here the laplacian is a vector laplacian, i.e. it applies to a vector field. It is obtained
by performing the following operation: ~∇ · ~∇. It is different to the scalar laplacian, introduced in
Eq. (1.48, that applies to a scalar field. While the symbols are the same, one can know which one
we are using by looking at what they are applied to.

1.4 Integration
1.4.1 Curvilinear integration
Curvilinear integrals are a generalisation of 1D definite integrals, i.e. ab f (x)dx. They are of the
R

following form:
Z
I= f (~r)dr (1.55)
C

where C is the line over which the integral is made,~r is the position vector, and dr is the infinitesimal
element along the curve C:
p
dr = dx2 + dy2 + dz2 (1.56)
 1.4 Integration 15

the norm of the vector d~r:

d~r = dx~i + dy~j + dz~k (1.57)

For line integrals,~r can always be expressed as a function of one variable t, so that~r(t = a) and
~r(t = b) mark the ends points of the curve C. This means that we can express the integral as:
Z b
d~r
I= f (~r(t)) dt (1.58)
a dt
Here is an example in two dimensions (in the (x,y) plane). We want
to integrate the function f (x, y) = xy along a closed counter-clock wise
triangular contour (see Fig. 1.5). The integral can be split into three parts
I = IAB + IBC + ICA . Along the path AB y = 0 everywhere, therefore
IAB = 0. Along the path BC, we √ have~r = x~√
i + y~j = x~i + (1 − x)~j, which
d~r
means that we also have dx = 1 + 1 = 2, therefore:
Z r~C
IBC = f (~r)dr
r~B
Z 0 √ Figure 1.5: Sketch
= x(1 − x) 2dx representing a triangular
1
√ Z0 √ Z0 2 counter-clock wise closed
= 2 xdx − 2 x dx (1.59) contour going through
1 1
√ x2 x3 0
  points ABC
= 2 −
2 3 1
√
2
=−
6
Now, along √the CA path x = 0 everywhere so ICA = 0. So in the end we
have I = − 62

1.4.2 Surface integrals in the plane

Surface integration is the 2D extension of the line integral presented
above. Here we will focus on the special case where the surface to
integrate is in the plane which we will chose to be (x,y). Such an integral
is expressed as:
x
I= f (x, y)dxdy (1.60)
S

where S is the surface over which the integral is performed and f the function to integrate. In the
simple case where the limits of S are constants (i.e. S is a rectangle with x = [a, b] and y = [c, d])
then Eq. (1.60) becomes:
Z b Z d  Z d Z b 
I= f (x, y)dy dx = f (x, y)dx dy (1.61)
a c c a

In the more general case where S is not a rectangle, Eq. (1.60) becomes (see Fig. 1.6):
Z b Z g(x)  Z d Z n(y) 
I= f (x, y)dy dx = f (x, y)dx dy (1.62)
a h(x) c m(y)
 16 Chapter 1. Some fundamentals of vector calculus

In Eq. (1.63), the terms in between brackets have to be calculated first.

They represent the equivalent of a line integral assuming x (or y depend-
ing on which of the two forms you choose) is constant over the column
(or row) as drawn in the figure. Then the sum of these line integrals give
you the surface integral and this is what the external integral is doing.
In practical terms, the only difficulty is to determine the limits of the
surface, h(x) and g(x) if integrating over x last, or m(y) and n(y) if you
are integrating over y last (the green and pink lines in Fig. 1.63). Cases
you will have to deal with will have simple geometries. Let’s take again
the example of the surface, S, covered by the triangle from Fig. 1.5 and
the function f (x, y) = xy. Then the surface integral of f over S is:
Z 1 Z 1−x 
I= xydy dx
0 0 Figure 1.6: Sketch repre-
Z 1
(1 − x)2 senting how the surface of
= x dx integration of a double in-
0 2
1 (1.63) tegral can be split.
x2
2x3 x4

1
= − +
2 2 3 4 0
1
=
24

Exercise 1.5 Show that by integrating over y last leads to the same
result. 

1.4.3 Volume integrals

Volume integrals are triple integrals, integral that are performed over a
given volume. The principle for integration is identical to that presented
for the surface integrals. Triple integrals are noted as follows:
y
I= f (x, y, z)dxdydz (1.64)
V

Similarly to the surface integrals, the difficulty is to determine the

functional shape of the functions that limit the volume of integration.
Exercise 1.6 Figure 1.7 shows the volume of a wedge of length of 1 in
all three directions. Compute the volume integral of f (x, y) = xy over
that volume. 

As we will see next, in the vast majority of cases that you will see in the
context of this module, volume integrals can be simplified is choosing
the correct system of coordinates.
Figure 1.7: Sketch repre-
1.5 Coordinate systems and transformations senting how the volume of
1.5.1 The position vector differential d~r wedge can be split for inte-
gration
There are multiple ways one can assign coordinates to a point in space.
The one that you are most familiar with is the cartesian coordinate
system, for which any point in 3D space is given 3 coordinates (x, y, z)
with ~r = x~i + y~j + z~k and ~i · ~j = 0, ~i ·~k = 0, and ~j ·~k = 0, and finally
1.5 Coordinate systems and transformations 17

~i ·~i = 1, ~j · ~j = 1, ~k ·~k = 1. The last 6 equations show that (~i, ~j,~k) are the
orthogonal unit vectors of the cartesian frame.
But this is not the unique way one can identify the location of a point in
space, the other two you are familiar with being cylindrical coordinates and spherical coordinates. .
Figure 1.8 shows a representation of these 3 coordinate systems for a point A. In these three frames
point A has coordinates (z, y, z), (ρ, θ , z), (r, θ , ϕ), respectively. Note that polar coordinates, which
you have also come across in the past, are the 2D version of the cylindrical coordinates for which
there is no z coordinates. The position vector~r is express as:

cartesian :~r = x~i + y~j + z~k

cylindrical :~r = ρ~uρ + z~uz (1.65)
spherical :~r = r~ur

Exercise 1.7 Express the cartesian coordinates (x, y, z) as a function of the cylindrical coordi-
nates (ρ, θ , z), and then as a function of the spherical coordinates (r, θ , ϕ). 

In order now to derive other key quantities in all three coordinate systems,
we first define the general relation for curvilinear orthogonal coordinate
systems (i.e. the general term used for the systems mentioned above)
of the position vector differential d~r:

d~r = ∂u (~r)du + ∂v (~r)dv + ∂w (~r)dw (1.66)

where (u, v, w) are the coordinates of point A. Equation (1.66) is the

vector counterpart of the definition of the total differential of scalar field
seen in Eq. (1.33). Each of the three terms in Eq. (1.66) have vector
units that are tangent along the lines of increasing u, v, w (just think of
a 1D curve y(x), the term dy/dx corresponds to the tangent of y at each
x points). Let’s call these unit vectors (~uu ,~uv ,~uw ) then Eq. (1.66) can be
rewritten as:

d~r = |∂u (~r)|du~uu + |∂v (~r)|dv~uv + |∂w (~r)|dw~uw (1.67)

with:

∂u (~r)
~uu =
|∂u (~r)| Figure 1.8: Sketches rep-
∂v (~r) resenting the definition of
~uv = (1.68)
|∂v (~r)| the coordinates of a point A
∂w (~r) in the cartesian top), cylin-
~uw = drical (middle), and spher-
|∂w (~r)|
ical (bottom) frames.
Note that this definition of unit vectors is in agreement with that given
in Eq. (1.5).
For the cylindrical case, Eq. (1.67) can rewritten as:

d~r = |∂ρ (~r)|dρ~uρ + |∂θ (~r)|dθ~uθ + |∂z (~r)|dz~uz

(1.69)
= dρ~uρ + ρdθ~uθ + dz~uz
 18 Chapter 1. Some fundamentals of vector calculus

Exercise 1.8 Show that Eq. (1.69) is verified (hint: you first need to answer Exercise 1.7). 

1.5.2 Differential elements

With the definition of d~r in hand, one can determine the general 3D relation of the line element dr:
√ q
dr ≡ d~r · d~r = (∂u (~r)du)2 + (∂v (~r)dv)2 + (∂w (~r)dw)2 (1.70)

The general expression for the surface elements dS are:

In the (u, v) plane : dSu,v = |∂u (~r)||∂v (~r)|dudv

In the (u, w) plane : dSu,w = |∂u (~r)||∂w (~r)|dudw (1.71)
In the (v, w) plane : dSv,w = |∂v (~r)||∂w (~r)|dvdw

Finally, the general expression for the volume element dV is:

dV ≡ |∂u (~r)||∂v (~r)||∂w (~r)|dudvdw (1.72)

These line, surface, and volume elements are those that need to be used when doing line, surface,
and volume integrals in a given coordinate frame.
In cylindrical coordinates, we therefore have the following expressions:
q
dr = (dρ)2 + ρ 2 (dθ )2 + (dz)2
dSρ,θ = ρdρdθ
dSρ,z = dρdz (1.73)
dSθ ,z = ρdθ dz
dV = ρdρdθ dz

Exercise 1.9 Compute all elements (dr, the three dS, and dV) in spherical coordinates. 

1.5.3 Gradient, divergence, curl, and Laplacian

In the context of this module, we will to use the del operator expressed in different coordinate
systems. We therefore need its expression in both cylindrical and spherical coordinates (we have so
far defined it only in cartesian coordinates - see Section 1.3).
The total differential of a scalar field f can be expressed as a function of the gradient of this field
(see Eq. 1.38) as:

d f = ~∇ f · d~r (1.74)

Expending d f using our set of general curvilinear coordinates (u, v, w) we also have:

d f = ∂u f du + ∂v f dv + ∂w f dw (1.75)

Subbing in Eq. (1.67) into Eq. (1.74) and identifying each with those of Eq. (1.75) we obtain the
following three relations:

∂u f du = ∇u f |∂u (~r)|du
∂v f dv = ∇v f |∂v (~r)|dv (1.76)
∂w f dw = ∇w f |∂w (~r)|dw
1.5 Coordinate systems and transformations 19

where ∇u , ∇v , ∇w are the 3 coordinates of the ~∇ operator in the corresponding frame (for instance
in cartesian coordinates, we have ∇x = ∂x , ∇y = ∂y , ∇z = ∂z ). From Eq. (1.76) we find that the ~∇
operator can be expressed as:

~∇ = ~uu ~uv ~uw

∂u + ∂v + ∂w (1.77)
|∂u (~r)| |∂v (~r)| |∂w (~r)|
So in cylindrical coordinates, the del operator is expressed as:
~∇cyl = ~uρ ∂ρ + ~uθ ∂θ +~uz ∂z (1.78)
ρ

Exercise 1.10 Find the expression of the del operator in spherical coordinates. 

Now that we know how to express the gradient of a scalar field, what about the expression of the
divergence of a vector field ~∇ · ~F? Using the general definition of the 3D del operator given in
Eq. (1.77), we have:
 
~∇ · ~F = ~uu ~uv ~uw
∂u + ∂v + ∂w · (Fu~uu + Fv~uv + Fw~uw ) (1.79)
|∂u (~r)| |∂v (~r)| |∂w (~r)|
In cartesian the coordinates, the unit vectors ~i, ~j,~k are constant, i.e. they do not rotate, this means
that, for instance, ∂x~i = 0. This is not necessarily the case for the other two coordinate systems we
have seen. For the divergence this means that while in cartesian coordinates we have~i∂x (Fx~i) = ∂x Fx ,
for any other non constant unit vector we have:
~uu ∂u Fu Fu~uu
∂u (Fu~uu ) = + ∂u (~uu ) (1.80)
|∂u (~r)| |∂u (~r)| |∂u (~r)|
In order to express to find an expression for the last term ∂u (~uu ) one needs first to write the unit
vector ~uu as a function of the cartesian unit vectors. These have already been defined in Eq. (1.68).

Exercise 1.11 Derive the expression of ~∇cyl · ~

F (hint: First find the expression of ~uρ ,~uθ ,~uz as a
function of ~i, ~j,~k). 

One can see that the derivation of the divergence in the general case is quite complicated. It becomes
even more complicated when deriving the curl and Laplacian operators. So in the following we only
give the expressions of each operators in cylindrical and spherical coordinates without derivations.

~∇cyl · ~F = 1 ∂ρ (ρFρ ) + 1 ∂θ Fθ + ∂z Fz
ρ ρ
~∇sph · ~F = 1 ∂r (r2 Fr ) + 1 ∂ϕ (Fϕ sin ϕ) + 1 ∂θ (Fθ )
r2 r sin ϕ r sin ϕ
 
~∇cyl × ~F = 1 ∂θ Fz − ∂z Fθ ~uρ + ∂z Fρ − ∂ρ Fz ~uθ + 1 ∂ρ (ρFθ ) − ∂θ Fρ ~uz



ρ ρ
 
~∇sph × ~F = 1 1 1 1



∂ϕ (Fθ sin ϕ) − ∂θ Fϕ ~ur + ∂θ Fr − ∂r (rFθ ) ~uϕ + ∂r (rFϕ ) − ∂ϕ Fr ~uθ
r sin ϕ r sin ϕ r
1
1
∆cyl f = ∂ρ ρ∂ρ f + 2 ∂θ2 f + ∂z2 f
ρ ρ
1 1 1
∆sph f = 2 ∂r r2 ∂r f + 2 ∂ϕ sin ϕ∂ϕ f + 2 2 ∂θ2 f



r r sin ϕ r sin ϕ
(1.81)
 2. Fundamental theorems of vector calculus

2.1 The divergence theorem

2.1.1 Theorem
The divergence theorem (also known as Gauss’s or Ostrogradsky’s the-
orem) is one that links the divergence of a vector field (~∇ · ~F) to its flux
(~F · ~S) such that we have the following relation:

Theorem 2.1.1
y {
~∇ · ~FdV = ~F · d~S (2.1)
V S
Figure 2.1: Sketch rep-
where V is any volume, dV is the volume element (which takes different resenting the volume V ,
forms dependingvon the coordinate system used - in cartesian coordinates and corresponding bound-
dV = dxdydz), S is the surface integral across the closed boundary ary surface S over which
one may apply Gauss’s the-
surface S of V , and d~S is the vector surface element whose direction is
orem on vecor field ~F. We
normal to the surface (d~S = dS~n where ~n is the unit vector normal to the
have also represented a
surface).
surface element vector d~S
While this theorem might seem, at first sight, complicated, it is actually
at one particular location
relatively intuitive to understand its meaning. The left-hand side of
on the surface S.
Eq. (2.1) corresponds to the sum of the divergence of the field ~F, i.e.
how much does the field "get stuck" or "run away" from that particular
region of space. The right-hand-side corresponds to the flux of that
field through the boundary surface of the volume we consider. A large
inward flux means that the flux is getting stuck in that particular region
of space while an outward flux means that it is running away. So one
can intuitively see that the two terms of Eq. (2.1) are indeed about the same thing.

2.1.2 Proof for a box

 22 Chapter 2. Fundamental theorems of vector calculus

First we consider a rectangular parallelepiped with sides a, b, c. We then

evaluate the right-hand-side of Eq. (2.1) by splitting the total flux ϕ of
~F into 3 parts:
{
ϕ= ~F · d~S = ϕx + ϕy + ϕz (2.2)
S

where ϕα is the flux through the sides of the box which are perpendicular
to axis α (α =x,y, or z). Consider that the flux ϕx : Figure 2.2: Example of a
x x box with the two surfaces
ϕx = ~ ~
F · dS + ~ ~
F · dS (2.3) S1 and S2 perpendicular to
S1 S2 the x axis hatched in green
and purple, with normal
Note that, in principle every surface as two normal unit vectors oriented vectors ~n1 and ~n2
in opposite direction. By convention, for closed surfaces, the normal
unit vectors point outwards. This is the reason why that in Fig. 2.2 ~n1
and ~n2 point outward from the parallelepiped. For the surface S1 we have:

d~S = dS~n1 = dxdy~i (2.4)

while for surface S2 we have:

d~S = dS~n2 = −dxdy~i (2.5)

So the flux ϕx is:

Z c Z b 
ϕx = [Fx (a, y, z) − Fx (0, y, z)] dy dz (2.6)
0 0

On the other hand, we also have the following general calculus relation:
Z b
f 0 (x)dx = f (b) − f (a) (2.7)
a

which in the context of Eq. (2.6) leads to:

Z a
Fx (a, y, z) − Fx (0, y, z) = ∂x (Fx )dx (2.8)
0

which when subbed in Eq. (2.6) leads to:

Z c Z b Z a  
ϕx = ∂x (Fx )dx dy dz (2.9)
0 0 0

Now we can repeat the same calculations for the other two directions (y and z) in order to obtain ϕy
and ϕz , and adding them all three up we obtain:
Z c Z b Z a  
ϕ= (∂x (Fx ) + ∂y (Fy ) + ∂z (Fz )) dx dy dz
0 0 0
y (2.10)
= ~∇ · ~FdV
V

which proves the divergence theorem.

2.1.3 Proof for any arbitrary volume

 2.2 Stokes’ theorem 23

Let’s assume now that instead of one box, we have two boxes connected
on one of their sides (see Fig. 2.3). For such a configuration we have:
y
~∇ · ~FdV = ϕ1
V
1
y (2.11)
~∇ · ~FdV = ϕ2
Figure 2.3: Example of
V2
two boxes connected by
and because the two volumes V1 and V2 are independent from each other, one of their surfaces (green
we have: hatched)
y y y
~∇ · ~FdV + ~∇ · ~FdV = ~∇ · ~FdV
V1 V2 V (2.12)
= ϕ1 + ϕ2

where V = V1 +V2 . However, it is not obvious that the flux through the
boundary of V is equal to the to ϕ1 + ϕ2 as S1 and S2 are not independent, i.e. they share one side.
It is clear however that we have:

ϕ = (ϕ1 + ϕ12 ) + (ϕ2 − ϕ21 ) (2.13)

where we have subtracted from ϕ1 the flux through the interface in the direction of ~n1 , called ϕ12 ,
and from ϕ2 the flux ϕ21 through the same interface in the opposite direction ~n2 = −~n1 . Obviously
we have ϕ12 = −ϕ21 , and therefore ϕ = ϕ1 + ϕ2 , which, combined with Eq. (2.12), proves The
divergence theorem for two boxes.
One can see now that one can add as many boxes, as small that we need, to fill up any arbitrary
volume, for which then the divergence theorem will also be valid. Here we assumed that the volume
of the ramps needed to fully complete the arbitrary volume is negligible compared to the volume
within the boxes, which is true in the limit where the number of boxes tends towards infinity.

2.2 Stokes’ theorem

2.2.1 Theorem
Stokes’ theorem is one that links the surface integral of the curl of a
vector field to its line integral along the boundary curve of the same
surface:
Theorem 2.2.1
x I
~∇ × ~F · d~S = ~F · d~r (2.14)
C
S
Figure 2.4: Illustration
Here the path C is the boundary curve of the surface S (see Fig. 2.4), and
of the surface S of integra-
is positively oriented, i.e. if you were to walk along that curve with your
tion and its corresponding
head pointing in the same direction as the normal vectors to the surface
boundary path C
~n, then the surface would be on your left.
Similarly to the divergence theorem, it might seems at first that Stokes’
theorem is complicated, but again, it can be intuitively understood. The
left-hand part of Eq. (2.14) corresponds to the flux of the curl of ~F
through the surface S. Since the curl describes the "rotation ability"
of a vector field, this left-hand-side term is a quantitative measure of
the total rotation ability of the vector field along the surface S. But the right-hand-side term of
 24 Chapter 2. Fundamental theorems of vector calculus

Eq (2.14) tells us that this total rotation ability of the field along S is equal to its circulation along
the boundary curve of S. This can easily be understood when looking at Fig. (2.6) where we realise
that individual rotating cells cancel each others except at the boundary of the surface. Figure (2.6)
makes the basis for the proof of Stokes’ theorem.

Exercise 2.1 Using the divergence theorem show that the left-hand-side of Stokes theorem
does not depend on the shape of the surface that is considered. 

2.2.2 Proof for a rectangle

Let’s assume that we have a 3D vector field such that ~F = Fx~i+Fy~j +Fz~k,
and let’s consider the rectangle of Fig. 2.5. Then the line integral along
the boundary curve C can be split into four sections so that we have:
I Z a Z b
~F · d~r = [Fx (x, 0, z) − Fx (x, b, z)] dx + [Fy (a, y, z) − Fy (0, y, z)] dy
C 0 0
Z a Z b
 
Figure 2.5: Illustration
= (−∂y Fx + ∂x Fy ) dy dx
0 0 for the one rectangle proof
x of Stokes’ theorem. Sec-
= ~∇ × ~F · d~S
tions I and III, and sec-
S
tions II and IV correspond
(2.15)
to the first and second
which proves Stokes’ theoremRfor a rectangle. Equation (2.15) makes use terms, respectively, of the
of the fact that f (a)− f (0) = a ∂ ( f )dx and that ~∇× ~F ·~k = ∂ F −∂ F first line of Eq. (2.15
0 x x y y x

2.2.3 Proof for two joint rectangles and more

Figure 2.6 shows a configuration whereby two rectangles, of respective
surface S1 and S2 , are joint by one side. In that particular configuration,
the flux of the curl of ~F through the total surface S = S1 + S2 is just the
sum of the two:
x x x
~∇ × ~F · d~S = ~∇ × ~F · d~S + ~∇ × ~F · d~S (2.16)
S S1 S2

Now, what aboutHthe circulation along the boundary curves? Do we

have the relation C ~F · d~r = C1 ~F · d~r + C2 ~F · d~r ? The answer has to
H H

be positive for Stokes’ theorem to be valid. And the reason for that can
be seen by decomposing the line integral as follows:
I I I
~F · d~r = ~F · d~r + ~F · d~r (2.17)
C1 C1 :A→B C1 :B→A

Similarly, we also have:

I I I
~F · d~r = ~F · d~r + ~F · d~r (2.18)
C2 C2 :A→B C2 :B→A

Looking at Figure 2.6 one can see that we have:

I I
~F · d~r = − ~F · d~r (2.19)
C1 :A→B C2 :B→A
 2.3 The gradient theorem 25

In other words, joined sections in the line integralHcancel eachHother.

Which means that in the end we indeed have C ~F · d~r = C1 ~F ·
d~r + C2 ~F · d~r. So this proves Stokes’ theorem for two joint rectan-
H

gles.

Now one can easily imagine that one can pave any surface with small
enough rectangles so that Stokes’ theorem can apply to any surfaces.
All joint sections will cancel out when doing the line integral while the
surface integral of each small rectangles will add up. Note as well that Figure 2.6: Illustration
for triangles, which might be necessary to completely fill in the surface, for the two rectangle proof
the line integral cancels out when the vector field is constant (i.e. when of Stokes’ theorem.
the surface of the triangle is small enough).

Exercise 2.2 Show that the line integral along the boundary of a triangle is always zero if ~
F is
uniform. 

2.2.4 Green’s theorem

Green’s theorem is Stokes’ theorem in the plane. It gives the relationship between the curl of a 2D
vector field ~F = (Fx , Fy ) across a surface S and its anti-clockwise circulation along the boundary of
that surface C.
Theorem 2.2.2
x I
(∂x Fy − ∂y Fx ) dS = ~F · d~r (2.20)
C
S

Note that if the circulation is computed in the clockwise direction then one needs to add a minus sign
to the right-hand side of Eq. (2.2.2). We also note that this equation is very similar to Eq. (2.14), as
mentioned above, and its derivation is identical. Finally, you will very often find Green’s theorem
written as:

x I
(∂x Q − ∂y P) dS = (Pdx + Qdy) (2.21)
C
S

One realises that this is identical to Eq. (2.2.2) once we set ~F(x, y) = P(x, y)~i + Q(x, y)~j, or again
Fx = P and Fy = Q. This theorem is most often used to compute complex surface or line integrals

Exercise 2.3 Verify Green’s theorem for P = y3 , Q = −x3 and a circular contour C given by
x2 + y2 = 1. 

2.3 The gradient theorem

2.3.1 Theorem
The gradient theorem has been eluded to in Sec. 3.2.2 when deriving the proof of Stokes’ theorem
for a rectangle. The gradient theorem, also known as the fundamental theorem of calculus for line
integrals, links the line integral of the gradient of a function to its end values. Its mathematical
expression is:
 26 Chapter 2. Fundamental theorems of vector calculus

Theorem 2.3.1
Z B
~∇ f · d~r = f (A) − f (B) (2.22)
A

This theorem means that for any irrotational vector field the line integral does not depend on the
path over which the integral is performed.

Exercise 2.4 Why am I talking about irrotational field in the sentence above? 

2.3.2 Proof
Let’s consider the total differential of f :

d f = ∂x ( f )dx + ∂y ( f )dy + ∂z ( f )dz = ~∇ f · d~r (2.23)

This means that we also have:

Z B Z B
~∇ f · d~r = d f = f (B) − f (A) (2.24)
A A

Note that here, A and B are points in space, meaning that f (A) = f (xA , yA , zA ) = f (~rA ).

2.4 Potentials and the Helmholtz theorem

2.4.1 Scalar potential
An irrotational vector field ~F is defined by the following relation:
~∇ × ~F = 0 (2.25)

As per Eq. (1.49), any such vector field is also a gradient field where:
~F = ~∇ϕ (2.26)

The scalar field ϕ is known at the scalar potential of the vector field ~F. Any gradient (or irrotational,
or conservative) vector field possesses a scalar potential.

2.4.2 Vector potential

Any solenoidal vector field ~F is defined by the following relation:
~∇ · ~F = 0 (2.27)

As per Eq. (1.51), any such vector field can also be expressed as:
~F = ~∇ × ~A (2.28)

The vector field ~A is known at the vector potential of the vector field ~F. Any solenoidal vector field
possesses a vector potential.

2.4.3 Helmholtz theorem

As we have already seen, a solenoidal field cannot be irrotational, and an irrotational field cannot
be solenoidal. This can be understood as the consequence of the orthogonality of the curl and the
gradient. In fact, it can be shown that any vector field can be described as the sum of the curl of a
vector field and the gradient of a scalar field so that we have:
2.4 Potentials and the Helmholtz theorem 27

Theorem 2.4.1

~F = −~∇ϕ + ~∇ × ~A (2.29)

This is known as the Helmholtz theorem. Obviously for irrotational field ~A = 0, and for solenoidal
field we have ϕ = 0. The proof of the Helmoltz theorem goes beyond the scope of this lecture but
for those who are interested you can read Section 1.16. of the Mathematical methods for physicists
book (G.B. Arfken, H. J. Weber - available online).
 3. Electrostatics

Electrostatics is the study of electric fields and charge spatial distributions when those do not vary
in time (i.e. they are not moving).

3.1 Gauss’s law

The derivation of Gauss’s law starts from Coulomb’s law that states that the electric field ~E generated
by a point charge q is given by:

~E(~r) = q ~r
(3.1)
4πε0 r3

where ε0 is the vacuum permittivity. In the context of this lecture, we will take Coulomb’s law as
an empirical law derived from decades of experiments started at the end of the 18th century. This
law is now known to be precise down to better than 10−15 . For an ensemble of n point charges,
the resulting electric field at any point in space is just the superposition of those created by the
individual charges:

n
~E(~r) = ∑ qi ~r −~ri (3.2)
ri |3
i=1 4πε0 |~r −~

where the ~ri are the locations of each individual charges. For a continuous distribution of charges,
the electric field at position~r is written:

1 y ~r − ~r0
~E(~r) = ρ(~r0 )dV 0 (3.3)
4πε0 0 |~r − ~r0 |3
V
30 Chapter 3. Electrostatics

where ρ is the charge density (i.e. electric charge per unit volume), V 0 is the charged volume. We
also note that we have the following relation:
 
~r ~ 1
3
= −∇ (3.4)
r r
which implies that Eq. (3.3) can be rewritten as:

y
!
~E(~r) = − 1 ~∇r 1
ρ(~r0 )dV 0 (3.5)
4πε0 0
V
|~r − ~r0 |

Note that the notation ~∇r is there to emphasise that the del operator operates on r and not r0 . Now
if one takes the divergence of the electric field, we get:
y
!
~∇ · ~E = − 1 1
∆ ρ(~r0 )dV 0
4πε0 0 ~ 0
|~r − r |
V
1
Z
= ρ(~r0 )δ (~r − ~r0 )dV 0 (3.6)
ε0 V 0
ρ(~r)
=
ε0
Here, we used the fact that ∆ 1
= −4πδ (~r − ~r0 ). This can be shown as follows, where we take
|~r−~r0 |
the case ~r0 = 0 for simplicity. First, one needs to notice that ∆ 1r = 0 everywhere (i.e. use the
definition of the Laplacian in Eq. (1.81) except for~r = 0 where the function is undefined. Then,
according to the divergence theorem, if we integrate this function over a sphere of volume V and
surface boundary S, we obtain:
y y 1
~∇ · ~r dV = − ∆ dV
r3 r
V V
{ ~r (3.7)
= ~
· dS
r3
S

Because we are integrating over a sphere, we have d~S = dS~ur = r2 sin θ dθ dϕ~ur , and ~r = r~ur by
definition. So the last term of Eq. (3.7) becomes:
x ~r Z π Z 2π
· d~S = sin θ dθ dϕ = 4π (3.8)
r3 0 0
S
This implies that we have the following relation:
y 1
∆ dV = −4π (3.9)
r
V

And, as we already said that ∆ 1r = 0 everywhere except for r = 0, then we must have the following
relation:
 
1
∆ = −4πδ (r) (3.10)
r
+∞ R +∞ R
where δ (r) is the Dirac delta function is defined as −∞ δ (x)dx = 1 and −∞ f (x)δ (x − a)dx = f (a)
(i.e. the Dirac delta function can also be loosely defined as δ (x − a) = 0 everywhere except at x = a
where δ (x − a) = +∞).
Let’s now rewrite the last line of Eq. (3.6):
 3.2 Electrostatic potential 31

Theorem 3.1.1

~∇ · ~E = ρ(~r) (3.11)
ε0
This is the differential form of Gauss’s law, it also corresponds to the first of the four Maxwell’s
equations.

The integral form of Gauss’s theorem is obtained by first integrating Eq. (3.11) over volume V:
y y ρ
~∇ · ~EdV = dV
ε0
V V
y (3.12)
~∇ · ~EdV = Qenc
ε0
V

where Qenc is the total electric charge enclosed within the volume V . And finally, by using, again,
the divergence theorem we obtain the integral form of Gauss’s law:

Theorem 3.1.2
{
~E · d~S = Qenc (3.13)
ε0
S

where S is the boundary surface of volume V .

Gauss’s law shows that positive electric charges are the sources of electric fields (the electric flux is
positive), while negative charges are the sinks (the electric flux is negative).

3.2 Electrostatic potential

In Eq. (3.5) we have been able to describe the electric field as a function of a gradient. This has
been possible because of the 1/r2 dependence of the electric field. Let’s here recall that equation:
y
!
1 ~∇r 1
~E(~r) = − ρ(~r0 )dV 0 (3.14)
4πε0 0 |~
r − ~
r 0 |
V

Here, ~∇r can be removed from the integral as it is independent of ~r0 , which means that it can be
rewritten as:
y
!
1 ~∇r 1
~E(~r) = − ρ(~r0 )dV 0 (3.15)
4πε0 |~
r − ~
r 0|
V0

This shows that the electric field is an irrotational (i.e. conservative) vector field, and therefore:
~∇ × ~E = 0 (3.16)

The electric potential ϕ is then defined by:

~E = −~∇ϕ (3.17)

It then follows that the electric potential is given by:

1 y 1
ϕ(~r) = ρ(~r0 )dV 0 (3.18)
4πε0 0 |~r − ~r0 |
V
 32 Chapter 3. Electrostatics

This is known as the Coulomb potential. Note that for a single point charge it simplifies to :

1 q
ϕ(r) = (3.19)
4πε0 r

The electric potential also satisfies another important equation, known as Poisson’s equation. In
order to derive it, we first combine the differential form of Gauss’s law, and the gradient notation of
the electric field:

~∇ · ~E = −~∇ · ~∇ϕ = −∆ϕ (3.20)

which then combined with the right hand side of Eq. (3.11) leads to Poisson’s equation:

Theorem 3.2.1
ρ
∆ϕ = − (3.21)
ε0

whose general solution is given by Eq. (3.18). In the special case where there isn’t any charge,
ρ = 0, Poisson’s equation becomes Laplace’s equation:

Theorem 3.2.2

∆ϕ = 0 (3.22)

3.3 Electrostatic energy

3.3.1 Derivation
Let’s imagine an empty space. Then, let’s bring a point charge q1
from infinity to position ~r1 (Fig. 3.1a). The work done W1 during
that operation is zero since, in the absence of any other charge,
there is no force that is being applied to q1 : W1 = 0. Now let’s
bring a 2nd charge q2 from infinity to position ~r2 (Fig. 3.1b). The
work done W2 . also know as the interaction energy between the
two charges, is:

Figure 3.1: Illustration for the

Z ~r2
work done when no charge is al-
W2 = −q2 ~1 · d~r
E
−∞ ready present (a) and when one
Z ~r2 charge is already present (b).
= q2 ~∇ϕ1 · d~r
−∞
Z ~r2
(3.23)
= q2 dϕ1
−∞
= q2 ϕ1 (~r2 )
q2 q1
=
4πε0 |~r2 − ~r1 |

where E~1 is the the electric field generated by charge q1 and ϕ1

the corresponding Coulomb potential ϕ1 (~r) = 4πε0q|~r−~ 1
r1 | (which
vanishes at infinity). Note that the initial minus sign in Eq. (3.23) comes about because one needs
to oppose the electric force in its opposite direction to bring the charge to ~r2 . And then it disappears
3.3 Electrostatic energy 33

because we have ~E = −~∇ϕ. Now, if we keep bringing point charges to the system, the work done
for bringing charge q j from infinity to position ~r j is:

W j = q j ϕ ( j−1) (~r j ) (3.24)

where the cumulated potential ϕ ( j−1) of the j − 1 charges already present is given by:

j−1 j−1
qi
ϕ ( j−1) (~r) = ∑ ϕi (~r) = ∑ 4πε0 |~r −~ri | (3.25)
i=1 i=1

which means that W j is given by:

j−1
qi
Wj = q j ∑ (3.26)
i=1 4πε0 |~
r j −~ri |

Finally, the work done to bring N charges is:

N N j−1
1 q j qi
Wtot = ∑ Wj = ∑ ∑ |~r j −~ri |
j=2 4πε0 j=2 i=1

1 1 N N qi q j
= ∑ ∑ |~ri −~r j | (3.27)
2 4πε0 i=1 j=1
j6=i
N
1
= ∑ qi ϕ(i)
2 i=1

where we have:
N
1 qj
ϕ(i) = ∑ |~ri −~r j | (3.28)
4πε0 j=1
j6=i

Note that ϕi , ϕ (i) , and ϕ(i) are all related but different: ϕi is the potential generated by charge qi ;
ϕ (i) is the potential generated by the first i charges while building up the charge distribution; ϕ(i) is
the potential experienced by the ith charge due to all the other (N − 1) charges in the distribution.

Let’s now consider an isolated charge whose distribution is given by its charge density ρc and
let’s assume that this charge is under the influence of the electric potential ϕ. Then, following
Eq. (3.23), the interaction energy is given by:
y
Eint = ρc ϕdV (3.29)
V

If, instead of considering an isolated charge, we now consider a complete charge distribution
characterised by the charge density ρ, then, by analogy to Eq. (3.27), the electrostatic energy
becomes:

1y
Wtot = ρϕdV (3.30)
2
V
 34 Chapter 3. Electrostatics

where ϕ is given by Eq. (3.18). The factor 12 comes from the fact that the integral counts every
interaction between each ρdV charges twice. From the differential form of Gauss’s law (Eq. 3.11),
Eq. (3.30) can be written as:
ε0 y ~ ~
Wtot = ϕ ∇ · EdV (3.31)
2
V

Now using the product rule for the divergence (Eq. (1.43)), we obtain:

ε0 y ~ ~ y
" #
2
Wtot = ∇ · (Eϕ)dV + E dV (3.32)
2
V V

where we also made used of the fact that ~E = −~∇ϕ. Now if we apply the divergence theorem to
the first term of the right hand side of that equation we get:

ε0 { ~ ~ y 2
" #
Wtot = ϕ E · dS + E dV (3.33)
2
S V

where S is the boundary surface of volume V . Now let’s assume that the volume V is a sphere, and
let’s take the limit in which the radius goes to infinity, i.e. the sphere has a radius much larger than
that of the charge distribution responsible for generating the electric field. We know that in such a
case, the electric field will fall off as 1/r2 at infinity, and the potential as 1/r. However, the surface
S will increase as r2 . Hence, it becomes clear that in the limit of r → ∞, the surface integral in
Eq. (3.33) will tend towards 1/r, which means that it will tend towards zero. However, the volume
integral will always include the part of the region where the charges are located, and therefore even
in the limit r → ∞, it will not tend towards zero. Therefore, Eq. (3.33) becomes:

ε0 y 2
Wtot = E dV (3.34)
2
V

This also means that the energy density U of an electric field is:

ε0 2
U= E (3.35)
2

3.3.2 A note on self- and interaction energies

When inspecting Eq. (3.27) and (3.34), one realises that both equations are inconsistent with each
other. Indeed, one can see that the former may take negative values (take the case of two charges
of opposite sign) while the latter is always positive (the integral of a positive quantity is always
positive). The discrepancy comes from the incorrect assumption that Eq. (3.30) is analogous to the
last line of Eq. (3.27). The reason why these two expressions are in fact different is because we
explicitly removed the i = j terms in the latter, i.e. the interaction of a charge with its own electric
field. Let’s assume we have a system of two charges, then Eq. (3.34) can be rewritten as:
ε0 y 2 ε0 y ~ ~2 |2 dV
Wtot = E dV = |E1 + E (3.36)
2 2
V V

where E ~1 and E
~2 are the electric field generated by the two charges. If we keep expanding we
obtain:
ε0 y 2 ε0 y 2 y
~1 · E
~2 dV
Wtot = E1 dV + E2 dV + ε0 E (3.37)
2 2
V V V
 3.4 Electric dipole 35

The first two terms are the self-energy terms, i.e. the energy it takes to build up the charges
themselves, while the last one is the interaction energy term, i.e. the energy it takes to bring one
charge from infinity to its position while interacting with the electric field of the other charge, and is
equal to Eq. (3.29) (this can be derived in the exact same way as the derivation for the electrostatic
energy in the previous section).
The interaction energy for point charges is what we have first derived in the previous section.
But what about the self-energy? Can we derive some expression for it? For this, suppose that we
have a charge Q uniformly distributed within a sphere of radius a. Let’s imagine that we build up
this charge as a succession of thin spherical layers of infinitesimal thickness dr. At each stage,
we bring a small amount of charge dq from infinity and spread it within the thin layer from r and
r + dr. We continue this process until the final radius of the sphere is a. If q(r) is the charge in the
sphere when it has reached radius r, then the work done in bringing a charge dq to it is:

1 q(r)dq
dW = (3.38)
4πε0 r
If the consider that the charge density ρ within the sphere is constant, then:
4
q(r) = πr3 ρ (3.39)
3
and

dq = 4πr2 ρdr (3.40)

Therefore we get:
4π 2 4
dW = ρ r dr (3.41)
3ε0
The total work needed to build up the sphere from nothing to radius a is:
Z a
4π 2 4π 2 5
W= ρ r4 dr = ρ a (3.42)
3ε0 0 15ε0

which can be written in terms of the total charge Q = 34 πa3 ρ:

3 Q2
W= (3.43)
20πε0 a

This energy is the self-energy of a uniform density charge of radius a. We see that for point charges,
a → 0 and the self-energy tends towards infinity. This shows the limitation of the point charge
model which is only valid for r >> a, which is valid in many situations since a ∼ 10−15 m for
electrons and protons.

3.4 Electric dipole

An electric dipole consists of a system of two equal charges q of opposite sign (i.e. net charge is
zero) separated by a distance a. The dipole moment d~ is defined as:

d~ = q~a (3.44)

where ~a is directed from the negative to the positive charge. Dipoles can be used as models for
some molecules for which the spatial distribution of electrons is asymmetric, i.e. one side of the
 36 Chapter 3. Electrostatics

molecule is more negatively charged, while the other is more positively charged. The electric field
generated by a dipole vanishes at large distances, r >> a, but it is not exactly zero. To be zero, the
two charges would have to be located exactly at the same place. One can compute the residual
potential (and electric field) generated by a dipole in the far field limit r >> a.
First, let’s look at expanding the following term:
 
2 2 2 2 2 ~r ·~a
|~r −~a| = r − 2~r ·~a + a ' r − 2~r ·~a = r 1 − 2 2 (3.45)
r

This leads to the following relation:

~r ·~a −1/2 1
   
1 1 1 ~r ·~a
=p ' 1−2 2 ' 1+ 2 (3.46)
|~r −~a| |~r −~a|2 r r r r

where the last part of the equation comes from the 1st order Taylor expansion of (1+x)−1/2 = 1− 12 x.
Now, one can write the total potential of a dipole for two point charges:
 
q 1 1
ϕ(~r) = − (3.47)
4πε0 |~r −~a| r

which, according to the superposition principle, is just the sum of the potential of each individual
charges (the same principle applies to the electric fields of each individual charges). Now using the
approximation shown in Eq. (3.46), the potential for a dipole in the limit r >> a becomes:

1 ~r · d~
 
1 ~ ~ 1
ϕ(~r) ' =− d ·∇ (3.48)
4πε0 r3 4πε0 r

The corresponding electric field is then given by:

~ r − r2 d~
1 3(~r · d)~
~E(~r) = −~∇ϕ ' (3.49)
4πε0 r5

Note that the last term has been derived using the product rule, and then taking advantage of the
fact that d~ is a constant vector that can be chosen to be along one of the cartesian coordinate axes.
One can see that the electric field and potential of a dipole at position~r only depends, in the far field
approximation, on d. ~ This means that one gets the same field and potential for (q,~a), and (2q, 1~a).
2
One can then introduce the concept of point (or ideal) dipole in the limit ~a → 0 q → ∞ for a finite d. ~

3.5 Dielectrics
A dielectric is an electrical insulator that gets polarised (i.e. an electric dipole is created) in the
presence of an electric field. In a dielectric the charges that make up the insulator are bound (as
opposed to a conductor in which electrons are mostly free). When an external electric field E ~0
exists, an induced electric dipole is generated as the result of a small shift between the negative
and the positive charge distribution of each atom/molecule. Within the insulator itself the induced
dipole of the atoms/molecules compensate each other, except as the surfaces of the insulator. As a
result of this, an induced electric field is generated within the dielectric with the opposite direction
of E~0 , effectively decreasing the electric field within the insulator, and also modifying the electric
field outside the dielectric according to the dipole approximation.
Now let’s consider the potential at location~r of one polarised dielectric molecule whose average
position is at ~r j = 0. Its total net (or free) charge and dipole moment are q j and d~ j . Because any
3.5 Dielectrics 37

measurements of the potential and electric field are here considered to be on a macroscopic scale
(i.e. on a scale much larger than the size of a molecule) we can consider that we are dealing with an
ideal dipole. The potential can then be written as:

1 x ρ j (~r0 )
 
0 1 qj ~ ~ 1
ϕ j (~r) = dV = −dj ·∇ (3.50)
4πε0 |~r − ~r0 | 4πε0 r r

where ρ j is the microscopic charge distribution within the molecule, and:

y
qj = ρ j (~r0 )dV 0 (3.51)

and
y
d~ j = ~r0 ρ j (~r0 )dV 0 (3.52)

In the example of the electric dipole presented in the previous section we had q j = 0, and the only
contribution to the potential was that of the ideal dipole. However, for ions for instance we have
q j 6= 0 and therefore it exists a standard 1/r contribution (i.e. also known as monopole) to the
potential from the net charge of the ion.
Now let’s sum the contribution from all polarised molecules, taking into account their different
average positions ~r j :
 
1 qj ~ ~ 1
ϕ(~r) = ∑ ϕ j (~r −~r j ) = −dj ·∇
j 4πε0 ∑
j |~r −~r j | |~r −~r j |
(3.53)
1 y ρ f (~r0 ) ~ ~0 ~
!
1 0
= − P(r ) · ∇r dV
4πε0 |~r − ~r0 | |~r − ~r0 |

Two things to notice here. First, ρ f (~r) is the free charge density, and ~P(~r) is the macroscopic
polarisation, which corresponds to the dipole moment density of the dielectric (or dipole moment
per unit volume). Second, we have the ~∇r operator which is the standard del operator, the subscript
r is there just emphasise that it is acting on~r and not on ~r0 . However, one can use the product rule
to show that we have:

~∇r 1 1
= −~∇r0 (3.54)
|~r − ~r0 | |~r − ~r0 |

Equation (3.53) becomes

1 y
!
ρ f (~r0 ) ~ ~0 ~ 1
ϕ(~r) = + P(r ) · ∇r0 dV 0 (3.55)
4πε0 |~r − ~r0 | |~r − ~r0 |

Now using the product rule of the divergence we have:

! !
~P(~r0 ) 1 1 ~ ~ ~0
~∇r0 · = ~P(~r0 ) · ~∇r0 + ∇r0 · P(r ) (3.56)
|~r − ~r0 | |~r − ~r0 | |~r − ~r0 |

and Eq. (3.55) becomes:

1 y
" #!
ρ f (~r0 ) 1 ~ ~ ~0 ~ ~P(~r0 )
ϕ(~r) = − ∇r0 · P(r ) + ∇r0 · dV 0 (3.57)
4πε0 |~r − ~r0 | |~r − ~r0 | |~r − ~r0 |
38 Chapter 3. Electrostatics

We can now use the divergence theorem to transform the third term to the integral over the surface
of the dielectric, which results in:
h i
y ~ 0 ) + ρ (~ 0)
1 ρ f (r ind r 1 { σind (~r)dS
ϕ(~r) = dV 0 + (3.58)
4πε0
V
|~r − ~r0 | 4πε0
S
|~r − ~r0 |

where the induced volume charge density ρind is defined as:

ρind (~r) = −~∇ · ~P(~r) (3.59)

and the induced surface charge density σind is defined as:

σind (~r) = ~P(~r) ·~n (3.60)

where ~n is the unit vector of the infinitesimal element d~S = dS~n. Equation (3.58) has a volume term
and a surface term. When evaluating the potential outside the dielectric, the surface term is zero.
i.e. there can’t be any flux of the polarisation vector since there isn’t any polarisation vector outside
the dielectric itself, and the potential becomes:
h i
y ρ f (~
r 0 ) + ρ (~
ind r 0)
1
ϕ(~r)out = dV 0 (3.61)
4πε0
V
|~r − ~r0 |

In the case of a dielectric, the differential form of Gauss’s law becomes:

~∇ · ~E = ρ = ρ f + ρind (3.62)
ε0 ε0

One can see that given the definition of the induced charge density, it becomes handy to define an
auxiliary displacement field ~D, such that:

~∇ · ~D(~r) = ρ f (~r) (3.63)

With such definition we have the following relation:

~D = ε0 ~E + ~P (3.64)

Now, for relatively weak electric fields it is reasonable to assume (and it is verified experimentally)
that ~P is proportional to ~E, and for isotropic dielectrics we have:

~P = ε0 χe ~E (3.65)

where χe is the electric susceptibility of the medium. This implies that we have the following
relations:

~D = ε0 ~E + ε0 χe ~E = ε0 (1 + χe )~E = ε0 ε ~E (3.66)

where ε = 1 + χe is called the dielectric constant, or relative permittivity. The dielectric constant in
vacuum is therefore ε = 1, and ε > 1 in dielectrics, and ε = ∞ in metal (i.e. no static electric field
in metals - electrons rearrange themselves until they feel no force, until they become static, i.e. no
electric field).
 3.6 The origin of Coulomb’s law 39

3.6 The origin of Coulomb’s law

We started off this chapter on electrostatics by using the empirical Coulomb’s law on point charges.
However, one can wonder why this law (along with Newton’s law of gravity) is as it is and does not
scale as any other power law of r. The reason for this can be found in Green’s functions.

3.6.1 Green’s function

d2
Let Ô(x) be a 1D differential operator, for example Ô(x) = dx2
. The Green’s function G(x, x0 ) of
the operator Ô(x) is defined as:

Ô(x)G(x, x0 ) = δ (x − x0 ) (3.67)

where δ is the Dirac delta function. Then for any "source" function g(x), the equation

Ô(x) f (x) = g(x) (3.68)

has the following solution:

Z +∞
f (x) = G(x, x0 )g(x0 )dx0 (3.69)
−∞

3.6.2 Proof
If one substitutes Eq. (3.69) into Eq. (3.68) we obtain:
Z +∞
Ô(x) f (x) = Ô(x) G(x, x0 )g(x0 )dx0
−∞
Z +∞
= Ô(x)G(x, x0 )g(x0 )dx0
−∞ (3.70)
Z +∞
= δ (x − x0 )g(x0 )dx0
−∞
= g(x)

In 3D, the Green’s function is a function of 6 variables and becomes:

Ô(~r)G(~r,~r0 ) = δ (~r − ~r0 ) (3.71)

and

y
+∞
f (~r) = G(~r,~r0 )g(~r0 )dV 0 (3.72)
−∞

is the solution of

Ô(~r) f (~r) = g(~r) (3.73)

3.6.3 Green’s function of Laplace operator

As it can be seen in Eq. (3.10), we have already found the Green’s function of Laplace operator,
which is :
1
G(~r,~r0 ) = − (3.74)
4π(~r − ~r0 )
40 Chapter 3. Electrostatics

This implies that, according to Eq. (3.72) the solution to Laplace equation ∆ϕ = − ερ0 is given by

y
+∞
1 ρ(~r0 ) 0
ϕ(~r) = dV (3.75)
−∞ 4π(~r − ~r0 ) ε0

which is nothing else bu Coulomb’s potential. Therefore the origin of the 1/r relation in Coulomb’s
potential is encoded in the symmetry of space in relation to the Laplace operator and any physical
quantity that satisfies Laplace equation must follow a 1/r power-law, and nothing else.
 4. Magnetostatics

Magnetostatics studies systems of stationary (i.e time-independent) currents.

4.1 Ampere’s circuital law

The same way that the existence of electric fields and Coulomb’s electric force have been exper-
imentally demonstrated in the presence of static charges, the existence of magnetic fields and
Ampere’s law have also been experimentally demonstrated (∼ 150 years later) in the presence, this
time, of an electric current.
In the case where an electric current runs along a straight wire aligned with z axis of a cylindrical
coordinate frame, then Ampere showed that the magnetic field ~B at position~r can be expressed as:

~B(~r) = µ0 I u~ϕ (4.1)

2πρ

where µ0 is the permeability of vacuum, and ρ is here the radial coordinate of the cylindrical
coordinate system (i.e. not a density). Note that here I can be either positive or negative depending
if it runs in the positive or negative direction along the z axis.
The morphology of the magnetic field lines as expressed by Eq. (4.1) is one of circular loops
centred around the main axis of the wire carrying the current. An important consequence of such
morphology is that magnetic fields are solenoidal vector fields:

~∇ · ~B = 0 (4.2)

Exercise 4.1 Using Eq. (4.1) as a starting point, show that Eq. (4.2) is verified. 

The direction of circulation of ~B can be determined with the right hand rule: if the thumb of the
right hand gives the direction of the current, then the circulation of ~B is given by the other fingers.
Ampere quickly realised that Eq. (4.1) can be expressed in a more general and concise way, known
now as Ampere’s circuital law (or Ampere’s law for short):
 42 Chapter 4. Magnetostatics

Theorem 4.1.1
I
~B · d~r = µ0 I (4.3)
C

This equation states that the circulation of the magnetic field along a closed loop is proportional to
the total current I passing through that loop. Like Gauss’s law of electrostatics, Ampere’s law can
be very useful in some symmetric situations to derive the magnitude of ~B.
Let’s now define the current density J~ = ρ~v as the amount of electric charges that flows per unit
surface and unit time, where ρ is the charge density and ~v is the velocity of the infinitesimal charge
ρdV . This definition of current density implies that the electric current I is the flux of current
density running through a given surface:
Z
I= J~ · d~S (4.4)
S

Now if one combines this new definition of the electric current with Eq. (4.3) we obtain:
I Z
~B · d~r = µ0 J~ · d~S (4.5)
C S

where C is here the boundary line of surface S. If one now applies Stokes’ theorem to the left hand
side of that equation we get:
Z Z
~∇ × ~B · d~S = µ0 J~ · d~S (4.6)
S S

And because this equation has to be true for any surface S, one can derive the differential form of
Ampere’s circuital law:

Theorem 4.1.2

~∇ × ~B = µ0 J~ (4.7)

At this stage, it is worth noting that the following set of four equations:

~∇ · ~E = ρ
ε0
~∇ × ~E = 0
(4.8)
~∇ × ~B = µ0 J~
~∇ · ~B = 0

are known as the four Maxwell’s equations for the electrostatics and magnetostatics. These four
equations fully describe the whole of electrostatics and magnetostatics. And we have derived them
starting from two experimental results: Coulomb’s force and Ampere’s law. We will see later how
these are modified when the the electric and magnetic fields are not time-independent anymore,
which is Maxwell’s main contribution to electromagnetism.

4.2 Vector potential and Gauge invariance

As we have seen above, magnetic fields are solenoidal vector fields such that:

~∇ · ~B = 0 (4.9)
 4.3 General solution of magnetostatics: Biot-Savart’s law. 43

As we have seen in Sec. 2.4, this means that we can write the ~B field as the curl of another vector ~A
as:

~B = ~∇ × ~A (4.10)

as ~∇ × ~∇ = 0. The vector field ~A is known as the magnetic vector potential. It is not immediately
clear why this is useful, but as we will see, this equation will allows us to derive magnetic fields in
complex settings.
One key property of the vector potential ~A is that is not uniquely defined. Indeed, if, in Eq. (4.9),
one replaces ~A by A ~1 = ~A + ~∇ f , one sees that we have ~B = ~∇ × A ~1 = ~∇ × ~A as ~∇ × ~∇ f = 0.
Therefore, the magnetic vector potential is known up to a gradient of an arbitrary function. This
freedom in choosing the vector potential is known as gauge invariance. Note that the electric
potential is also gauge invariant, but the gauge is not the gradient of a function but just a constant
(~E = −~∇ϕ = −~∇(ϕ + c) where c is the constant).
The fact that one can choose whatever gauge we want provides an opportunity to choose one that
simplifies calculations. One such gauge is Coulomb’s gauge that requires that:

~∇ · ~A = 0 (4.11)

This gauge imposes that the magnetic vector potential is, along with the magnetic field itself, a
solenoidal field. The mathematical implication for the choice of the function f is given by the
following relation:

~∇ · A
~1 = ~∇ · ~A + ~∇ · ~∇ f = ∆ f = 0 (4.12)

In other words, Coulomb’s gauge imposes that f is a solution of Laplace’s equation.

Exercise 4.2 Show that using Coulomb’s gauge, a constant magnetic field ~
B can be represented
(still not uniquely) as:

~
~A(~r) = B ×~r (4.13)
2


4.3 General solution of magnetostatics: Biot-Savart’s law.

We said in the previous section that the choice of Coulomb’s gauge would make our life easier. The
reason for this becomes obvious once we plug in Eq. (4.10) in the differential form of Ampere’s
law:

~∇ × ~B = ~∇ × (~∇ × (~A + ~∇ f ))

= ~∇ × (~∇ × ~A)
(4.14)
= ~∇(~∇ · ~A) − ∆~A
= −∆~A

Using now the right hand side of the differential form of Ampere’s law, we obtain:

∆~A = −µ0 J~ (4.15)

 44 Chapter 4. Magnetostatics

In cartesian coordinates, this is equivalent to:

∆Ax = −µ0 Jx
∆Ay = −µ0 Jy (4.16)
∆Az = −µ0 Jz
This is just three times Poisson’s equation seen in Eq. (3.21) for which the general solution is given
by Eq. (3.18). By analogy, the general solution of Eq. (4.15) is therefore:
y J(~ ~r0 )
~A(~r) = µ0 dV 0 (4.17)
4π |~r − ~r0 |
The magnetic field is then obtained via:
y
!
~B(~r) = ~∇ × ~A(~r) = − µ0 ~ ~r0 ) × ~∇r 1
J( dV 0 (4.18)
4π |~r − ~r0 |

Here, we used the product rule for the curl, and made advantage of the fact that ~∇r × J(
~ ~r0 ) = 0. We
can further develop the equation above as follows:
y ~0
~B(~r) = µ0 ~ ~r0 ) × ~r − r dV 0
J( (4.19)
4π |~r − ~r0 |3
This is very similar to the general solution of electrostatics (Eq. 3.3). It provides the magnetic field
~B for any steady current configuration. Equation (4.19) is known as Biot-Savart’s law.
A more widely known form of Biot-Savart’s law is obtained when replacing the distributed
current by that of an idealised zero thickness wire: JdV~ = Id~l where d~l is the line element if a
steady current I. Substituting we obtain:
d~l × (~r − ~r0 )
Z
~B(~r) = µ0 I(~r0 ) (4.20)
4π |~r − ~r0 |3
or again, in its differential form:

µ0 ~0 d~l × (~r − ~r0 )

d ~B(~r) = I(r ) (4.21)
4π |~r − ~r0 |3
The corresponding vector potential is:
µ0 I(~r0 ) d~l
d~A(~r) = (4.22)
4π |~r − ~r0 |

4.4 Lorentz force and the origin of magnetic fields

4.4.1 The Lorentz force and some characteristics
Since the end of the 18th century, and Ampere’s experiment, it was known that moving charges in
the vicinity of a magnetic field was feeling a magnetic force. It is only in the late 19th century with
J.C. Maxwell, M. Faraday, J.J Thomson and H. Lorentz that a modern formulation of that force
came to life. Lorentz force is usually formulated with a combination of its electric part F ~e and its
magnetic part F~m , such that we have:

~F = F
~e + F~m = q~E + q~v × ~B (4.23)
 4.4 Lorentz force and the origin of magnetic fields 45

It is worth noting that a moving charged particle with velocity ~v per-

pendicular to the direction of a uniform magnetic field ~B will follow a
circular trajectory, called Larmor orbit (see Fig. 4.1), whose radius is
obtained by equating the centripetal force to the magnetic force:

mv
R= (4.24)
qB

where m and q are the mass and charge, respectively of the particle. If Figure 4.1: Illustration of
the particle has also a non-zero velocity in the direction of the magnetic Larmor orbit and radius
field, then it will follow a spiral trajectory along the magnetic field lines.
One interesting property of the magnetic force is that it does not produce
any work. Indeed we have:
Z B Z B Z B
Wm = q ~v × ~B · d~r = −q ~B ·~v × d~r = −q ~B ·~v ×~vdt = 0 (4.25)
A A A

where we used ~v = d~r/dt and ~v ×~v = 0. One can easily understand that the magnetic force does
not produce any work once we realise that its direction is always perpendicular to the direction
of motion, and therefore F~m · d~r = 0. Similarly, the magnetic force leaves the kinetic energy of a
particle unchanged:

dEk d m~v2 d~v

= = m~v · = m~v ·~v × ~B = m~v ×~v · ~B = 0 (4.26)
dt dt 2 dt
In other words, the magnetic force does not change the magnitude of the velocity of a particle, only
its direction, which does not matter in the calculation of the kinetic energy. Overall, the magnetic
force does not seem to behave like any other forces we have came across. So let us try to understand
a bit better why it is there in the first place.

4.4.2 Special relativity and the origin of magnetic fields

At the beginning of this magnetostatics chapter, we accepted the fact
that electric current generates magnetic fields, as demonstrated by, e.g.,
Ampere’s experiments. Now let’s consider the following situation. A
negative charge (q < 0) is moving with a velocity v parallel to a wire. In
this wire, electrons, with density ρ− , are flowing in the same direction
as our charge and with the same velocity creating a current I within the
wire (see Fig. 4.2). The wire is electrically neutral (ρ+ = −ρ− ), so that
Fe = 0, and the resulting Lorentz force is purely magnetic: Figure 4.2: Illustration of
set up where a charge is
µ0 I µ0 qρ+ Av2 moving in the parallel di-
Ftot = Fe + Fm = Fm = qvB = qv = (4.27) rection to a current within
2πR 2πR
a wire, and at the same
where we took the B-field from Eq. (4.1) and evaluated the current as speed.
I = AJ = Aρ+ v using J = ρ+ v. The Lorentz force is directed towards the
wire, so that the observer will see that the distance between the charge
and the wire decreases with time. This is what we have in the usual
frame S (see Fig. 4.2).

Now we consider the same system in frame S0 moving with velocity

v (relative to S) in the same direction as our charge (see Fig. 4.2) so that
our charge and electrons are not moving in that frame, but the ions of
46 Chapter 4. Magnetostatics

the of the wire move in the opposite direction. No matter how large the
current is, and how big the B-field is, since our charge is not moving,
the magnetic part of the Lorentz force acting on the charge in frame S0
is zero: Fm0 = 0. However, the laws of physics have to remain the same
independently of the inertial frame of reference, and if the charge is
moving towards the wire in the S frame, it has to move towards the wire
in the S0 frame too! So what is the force that pushes the charge towards
the wire in the S0 frame? If Fm0 = 0 it can only mean that Fe0 6= 0. But
where does this electric force come from? The wire was neutral in the
S frame. Has that changed?
The solution to this problem is provided by special relativity. In fact,
a well known issue of Maxwell equations when they were first derived
was that they would have a different solution depending on the frame
of reference one would choose, which goes against the principle that all
physical laws should be the same in all frames. Again, this issue was
solved by A. Einstein and his theory of special relativity. In that theory,
a piece of wire of length L in the S frame, is being shrink in the S0 frame
following:
r
0 v2
L = L 1− 2 (4.28)
c
However, the total charge Q+ of the ions of that piece of wire does not
change and the cross area does not change either since it is perpendicular
to the motion of the frame. Therefore we have:

Q+ = ρ+ LA = ρ+0 L0 A (4.29)

so that the ion charge density is higher in frame S0 "

L ρ+
ρ+0 = ρ+ 0
=q > ρ+ (4.30)
L 2
1 − vc2

For the same reason the charge density of electron decreases in the S0
frame as the electrons are still in it, while they move in S, so we have:

L ρ−0
ρ− = ρ−0 = > ρ−0 (4.31)
L0
q
2
1 − vc2

And so, while the wire was neutral in the S frame, in frame S0 the wire
turns out to be positively charged:

ρ+ v2 v2
∆ρ 0 = ρ+0 − ρ−0 = q 2
' 2 ρ+ > 0 (4.32)
v2 c c
1 − c2

where in the last equation we assumed that v << c which is the non-
relativistic case. Now we can evaluate the electric force in the S0 frame:

λ qρ+ Av2 µ0 qρ+ Av2

Fe0 = qE 0 = q = = (4.33)
2πε0 R 2πε0 c2 R 2πR
 4.5 Magnets and the magnetostatics properties of matter 47

where we used λ = ∆ρ 0 A and µ0 = ε 1c2 , which has been experimentally

0
verified. The beauty of that derivation is that the electric force in the
S0 frame turns out to be identical to the magnetic force in the S frame.
Therefore one can see that magnetic fields (and the magnetic force)
is a relativistic correction of the electric force. Magnetic fields find
their origin in the necessity that physical laws remain the same in all
inertial frames, and is a direct manifestation of Einstein theory of special
relativity. It also shows how much tangled electric and magnetic fields
are.

4.5 Magnets and the magnetostatics properties of matter

We have seen above that for a magnetic force to exist, a charged particle
needs to be in motion. One can therefore wonder why certain, apparently
inert, solid materials such as magnets attract each other. The answer is
found in the microscopic structure of matter, and electrons in particular.
Indeed, by spinning and orbiting around the atom’ nucleus, electrons
create microscopic electric currents. Bathed within a magnetic field,
these electrons can experience magnetic forces.
The magnetic field of a small bar magnet can be assimilated to
that of a small current loop. As we have seen in one of the videos,
the morphology of the magnetic field of such a small current loop is
the same as that of the electric field generated by an electric dipole.
Therefore we often refer to such a small current loop at a magnetic
dipole. In a magnet, all the microscopic currents generated by electrons
run in the same directions, or said in a different way, the magnetic dipole
moment of all electrons are aligned, generating a macroscopic magnetic
Figure 4.3: Illustration
dipole. The magnetic dipole moment ~m of a current loop is defined as
of current loop of vector
(see Fig. 4.3): ~
area A and through which
~m = I~A (4.34) passes a current I. These
two quantities allows us to
where I is the current in the loop, A is the area of the loop, and is oriented define the magnetic dipole
so that with I they satisfy the right-hand rule. moment ~m = I~A. ~r is the
Now let’s imagine that we put such a current loop within a uniform position vector as used in
magnetic field. All the electrons within the loop will feel a magnetic the calculations of the mag-
force according to Lorentz force. The magnetic force d F~m felt by a netic torque.
small section dl of the wire carrying a linear charge density λ can be
expressed as a function of the current going through it (see Fig. 4.4):
d F~m = λ dl~v × ~B
(4.35)
= Id~l × ~B
Now the corresponding torque d τ~m felt by the the small section of the loop is:
d τ~m =~r × d F~m =~r × (Id~l × ~B) (4.36)
where~r is the position vector whose origin is at the centre of the loop and whose magnitude is the
radius of the loop. Therefore the total torque that is being applied on the loop is:
I
τ~m = I ~r × d~l × ~B (4.37)
48 Chapter 4. Magnetostatics

Here we recognise the vector triple product. So we have:

I I
τ~m = I (~B ·~r)d~l − (~r · d~l)~B (4.38)

The second integral is zero since, by definition,~r and d~l are perpendic-
ular to each other. Using the product rule for the curl along with Stokes’
theorem, we can show that:
I x Figure 4.4: Illustration of
f d~l = − ~∇ f × d~S (4.39) current loop and forces ap-
plied to it when interacting
So this means that the torque becomes:
with a magnetic field.
x
τ~m = −I ~∇(~B ·~r) × d~S (4.40)

where the surface integral is over the loop. And because ~B is a constant we finally obtain:
x
τ~m = I d~S × ~B = I~A × ~B (4.41)

The last part can then be expressed as a function fo the magnetic dipole moment:

τ~m = ~m × ~B (4.42)

So we can see that the magnetic field will generate a torque on the
current loop until they become aligned. The work done by the loop Wm
when going from an initial position where ~m and ~B are at an angle θ to
alignment (θ = 0) is equal to:
Z
Wm = F~m · d~l
Z 0
π
=− Fm cos( − θ )rdθ
θ 2
Z 0 (4.43)
=− Fm sin(θ )rdθ Figure 4.5: Illustration of
θ
Z 0 two current loops, each ex-
=− τm dθ pose to the magnetic field
θ of each other (only the
The potential
R0
energy is always the opposite of the work done: Um = one from current loop I1 is
−Wm = θ τm dθ . By developing further this expression of the potential shown here as green field
energy, we get: lines). One can see that
across the size of current
Um = −~m · B~ (4.44) loop I2 the magnetic field is
non-uniform and will gen-
This expression is identical to the potential energy of an electric erate a force on loop 2.
dipole once the magnetic dipole moment and magnetic filed are replaced
by the electricR dipole moment and the electric field. Now, because we
have Um = − F~m · d~r the total force exerted on the loop is obtained by:

F~m = −~∇Um = ~∇(~m · ~B) (4.45)

In the case where ~m and ~B have been aligned, we have:

F~m = m~∇B (4.46)

4.5 Magnets and the magnetostatics properties of matter 49

So we can see that, if the field is uniform, there will not be any force exerted onto the loop. But
if the field is not uniform, the loop will experience a force, and this is the reason why when
approaching the two poles of two magnets close to each other they are either attracted to each other
(opposite pole polarity) or pushed away from each other (same pole polarity). This can also be
simply understood with the sketch shown in Fig. 4.5.
We have said at the beginning of this section that a small magnet could be modelled with a
small current loop. The reason why this is the case is that in magnets the microscopic current
generated by the orbital motion and spin of electrons are mostly all aligned, i.e. their magnetic
dipole moments all point on the same direction. However, most material do not behave like that
and have their atomic magnetic dipole randomly oriented. One can define the magnetisation vector
~ as the net magnetic dipole moment per unit volume:
M
N
~i
~ = ∑i=1 m
M (4.47)
V
where i is the sum over all atomic dipoles within the volume V .
For most material M ~ = 0. However, if an external magnetic field ~B0 = µ0 H
~ is applied then the
~
atomic dipoles will tend to align or anti-align with H. The resulting magnetic field is then:

~B = ~B0 + ~BM = µ0 (H
~ + M)
~ (4.48)

In weak fields, magnetisation is proportional to external fields and we have:

~ = χm H
M ~ (4.49)

where χm is the magnetic susceptibility. Combining we get:

~B = µ0 (χm + 1)H
~ (4.50)

We also note that, in the absence of magnetisation, the differential form of Ampere’s law can be
written as:

~∇ × H
~ = J~f (4.51)

where J~f is the volume density of free current. We obviously notice here the similarities with the
electric polarisation of dielectric materials and its impact on the electric field. The following fields
~P, ~D, ~E, are the electric equivalent of M,
~ H,
~ ~B.
Materials are defined according to their magnetic susceptibility: i. χm < 0: materials are
diamagnetic (such as water) and magnetic fields are reduced; ii. χm > 0: materials are paramagnetic
and the magnetic field is enhanced (a bit); iii. χm >> 0: materials are ferromagnetic (i.e. such as
magnets) an magnetic fields are enhanced (a lot!). The latter show a non-linear relationship between
magnetisation and external magnetic field strength. When completely relaxed, in the absence of
external field, M ~ = 0. However, after experiencing an external magnetic field, a ferromagnetic
material will not return directly to M ~ = 0 even after the external field has been switched off. The
material becomes therefore a magnet.
 5. Electromagnetism

So far, we have only considered static electric and magnetic fields separately. Now, we are going
to consider their time variations. As we have already realised, magnetic and electric forces are
intimately related, and this becomes even more obvious when time variations are taken into account.
In fact, magnetic and electric forces are two different aspects of a unique force, the electromagnetic
force. We will see in this chapter that time variations of electric and magnetic fields necessarily
involves the existence of electromagnetic waves.

5.1 Faraday’s law of induction

Let’s perform the following experiment (see Fig. 5.1). Consider a rect-
angular loop of which one side is a conductor. This conductor is moving
with a constant velocity v. A uniform constant magnetic field is perpen-
dicular to the plane of the loop. Because the conductor is moving the
charges within it are feeling the Lorentz force, aligned with the conduc-
tor. An electric field and an electric current are therefore created within
the conductor as the conductor becomes polarised. Measurements of the Figure 5.1: Experiment il-
created electric field show that we have EL = −BLv, i.e. the circulation lustrating Faraday’s law of
of the electric field along the conductor is proportional to the change of induction. The change of
magnetic flux through the rectangular loop. Mathematically this relation magnetic flux through the
is written as: moving rectangular frame
is directly linked to the cir-
I culation of the electric field
ε= ~E · d~r = − ∂ φB (5.1) along it.
∂t
s
where φB = ~B · d~S is the magnetic flux, and ε is known as the electric-
motive force, even though this is not a force but the work done by the electric field on the loop,
which also corresponds to the electric potential (or voltage for an electric circuit). The negative sign
in the right-hand side of the equation emphasises the fact that any induced current will be such that
it will try to oppose the change of magnetic flux. This latter principle is known as the Lenz’s law,
 52 Chapter 5. Electromagnetism

while Eq. (5.1) is Faraday’s law of induction. Applying Stokes theorem to Eq. (5.1) one obtains the
differential form of Faraday’s law:

~
~∇ × ~E = − ∂ B (5.2)
∂t

This equation represents a major change with respect to the electrostatic case where ~∇ × ~E = 0.
Equation (5.2) shows that when magnetic fields are not static anymore, then electric fields are not
irrotational. Faraday’s shows, once more, that magnetic forces and electric forces are in fact the two
facets of the same electromagnetic force, where depending on the measurement reference frame, an
electric field can morph into a magnetic field, and vice versa. We can also note here that, in the
experiment depicted above the right hand-side of Eq. (5.2) is zero. So how can both Eq. (5.1) and
(5.2) be verified? The answer is, in fact, found in Lorentz force. Indeed, if the loop C continuously
changes in size (as it is the case in our experiment), then one can show that we have:

∂ x ~ ~ x ∂ ~B ~
I
B · dS = · d S − ~v × ~B · d~r (5.3)
∂t ∂t C
S S

While going from Eq (5.1) to (5.2) we simply ignored the second term of the right hand side, which
we can do only if the loop does not change. Therefore, for a loop that does change in size and a
constant magnetic field, Eq (5.2) becomes:
~∇ × ~E = ~∇ × (~v × ~B) (5.4)

Electric currents induced by the change of magnetic flux have a large range of applications in every
day life (e.g. generator, motor). In practice these devices often use a permanent magnet moving
in and out of a conducting loop. In this case, the induced current is the result of the change of
magnetic flux that is operated when changing the field lines goign through the loop.

5.2 Maxwell’s equations

Until J.C. Maxwell’s work, the known laws of electricity and magnetism were the following ones:
~∇ · ~E = ρ , ~∇ × ~E = − ∂ ~B , ~∇ · ~B = 0, ~∇ × ~B = µ0 J.
~ While the first three turn to be always true (i.e.
ε0 ∂t
always verified by experiments), Maxwell noticed an issue with the last one. In order to realise
what the problem is we must first talk about the continuity equation of charges.

5.2.1 The continuity equation of electric charges

The so-called continuity equation is one that appears in many different areas of physics, including
electromagnetism and fluid mechanics, and it directly arises from the divergence theorem.

As we have seen previously the flux of current density running through a given surface is related to
the electric current via:
Z
I= J~ · d~S (5.5)
S

Now, if we consider a volume V and its boundary surface S then the time variation of the amount of
electric charge within V is equal to the net flux through S. In other words we have:

I Z
J~ · d~S = −∂t ρdV (5.6)
S V
 5.2 Maxwell’s equations 53

where ρ is the volume density of electric charges. If one now applies the divergence theorem to the
right hand side of that equation, we obtain:

Z Z
~∇ · JdV
~ = −∂t ρdV (5.7)
V V

And because this relation is true for any volume V , the integrands from each side must be equal:

~∇ · J~ + ∂t ρ = 0 (5.8)

Equation (5.8) is known as the continuity equation for electric charges. It expresses the conservation
of electric charges.

5.2.2 Maxwell’s correction of Ampere’s law

Going back to Ampere’s law now: ~∇ × ~B = µ0 J.
~ As noticed by Maxwell, if one takes the divergence
of it we obtain:

~∇ · (~∇ × ~B) = ~∇ · (µ0 J)

~ (5.9)

Because the divergence of a curl is always equal to zero, it means that Ampere’s law implies that
~∇ · J~ = 0. However, as stated by the continuity equation, charges can move around, pile up in some
places, and get rarer in others. Therefore, Ampere’s law cannot be correct. By symmetry with
Faraday’s law of induction, Maxwell decided to add a term to Ampere’s law, therefore becoming:

~
~∇ × ~B = µ0 J~ + µ0 ε0 ∂ E (5.10)
∂t

Now, if one takes the divergence of the corrected version of Ampere’s law we get:

~∇ · J~ + ε0 ∂ (~∇ · ~E) = 0 (5.11)

∂t

But we know that ~∇ · ~E = ερ0 . We we plug that in Eq. (5.11) we go back to the continuity
equation. In other words, Maxwell’s correction to Ampere’s law ensures charge conservation.
Beautiful. And with that the four Maxwell’s equations are determined:

~∇ · ~E = ρ
ε0
~
~∇ × ~E = − ∂ B
∂t (5.12)
~∇ · ~B = 0.
~
~∇ × ~B = µ0 J~ + 1 ∂ E
2
c ∂t

where we used the fact that the speed of light c is related to ε0 and µ0 via:

1
c2 = (5.13)
µ0 ε0
 54 Chapter 5. Electromagnetism

5.3 Electromagnetic waves inn vacuum

In the absence of charges and currents (ρ = 0, J~ = 0), Maxwell’s equations reduce to

~∇ · ~E = 0
~
~∇ × ~E = − ∂ B
∂t (5.14)
~∇ · ~B = 0
~
~∇ × ~B = 1 ∂ E
c2 ∂t

5.3.1 Derivation of the electromagnetic wave equations

Using the two equations that involve both ~E and ~B, one can get sub-in for ~E. For instance, let’s
take the time derivative of Faraday’s law of induction, and the curl of what is left of the corrected
version of Ampere’s law, we get:

∂ ~ ~ ∂ 2~B ~
~∇ × (~∇ × ~B) = 1 ~∇ × ∂ E
(∇ × E) = − 2 and (5.15)
∂t ∂t c2 ∂t
Because for all "nice" functions (i.e. twice differentiable) one can interchange partial derivatives,
we can combine these two equation to get:
2~
~∇ × (~∇ × ~B) = − 1 ∂ B (5.16)
c2 ∂t 2
Now one can use the triple vector product identity applied to the del operator (see Chapter 1) so
that we have:
~∇ × (~∇ × ~B) = ~∇(~∇ · ~B) − (~∇ · ~∇)~B (5.17)

Since one of Maxwell’s equation is that the divergence of ~B is zero we get:

~∇ × (~∇ × ~B) = −∆~B (5.18)

Which leads to the following equation:

1 ∂ 2~B
∆~B = 2 2 (5.19)
c ∂t

Note that by doing the same calculations, we can get the equivalent for the electric field:

1 ∂ 2 ~E
∆~E = 2 2 (5.20)
c ∂t

Equations (5.19) and (5.20) are both wave equations.

5.3.2 Solution of the wave equation in 1D

2~
In one dimension, we have ∆~E = ∂ E2 , which means that Eq. (5.20) becomes:
∂x

∂ 2E 1 ∂ 2E
2
= 2 2 (5.21)
∂x c ∂t
 5.3 Electromagnetic waves inn vacuum 55

where the vectors have switched to scalars since there is only one dimension. The general solution
to equation (5.21) is:

E(x,t) = f (kx − ωt) + g(kx + ωt) (5.22)

where f and g are arbitrary functions, k and ω are constants that are not independent. These are
known as the wave number and angular frequency, respectively.
Let’s prove now that Eq. (5.22) is indeed solution to the wave equation. To do this, let’s
substitute it to Eq. (5.21). Let’s first take the spatial derivative:

∂E
= k f 0 + kg0
∂x
(5.23)
∂ 2E
= k2 f 00 + k2 g00 = k2 ( f 00 + g00 )
∂ x2
Let’s now have a look at the time derivative part:

∂E
= −ω f 0 − ωg0
∂t
(5.24)
∂ 2E
= ω 2 f 00 + ω 2 g00 = ω 2 ( f 00 + g00 )
∂ x2
If we now substitute these last two relations into the wave equation (Eq 5.21) we get:

ω 2 00
k2 ( f 00 + g00 ) = ( f + g00 ) (5.25)
c2
So we can see that indeed, Eq. (5.22) is solution to the wave equation, provided that we have the
following relation:

ω 2 = k 2 c2 or ω = ±kc (5.26)

The relation between angular frequency and wave number is known as the dispersion relation.
It is worth noting at this point that Eq. (5.22) is stating that any function can be solution to
the wave equation. But you might then wonder why when studying waves you have always been
looking at sin and cosin functions. Well, the reason is that any random function can be written as
the sum of sin and cosin functions. This is Fourier’s theorem. So a cosin wave is just the elementary
brick of any wave, of any shape.

5.3.3 Solutions of the wave equation in 3D and the case of plane waves
A general 3D solution of the wave equation is given by the superposition of 1D solutions propagating
in any possible direction. Such a solution is given by:
 
~0 f (~k ·~r − ωt) + g(~k ·~r + ωt)
~E = E (5.27)

where f and g are any function, and E ~0 is any constant vector. Using the same technic as that
used above, one can show that, indeed, Eq. (5.27) is solution to Eq. (5.20). In fact Eq. (5.27)
describes what we call a plane wave propagating in the direction of ~k, also known as the wave
vector. The reason why this solution is a plane wave is because at any given moment in time
~k ·~r + ωt = constant is the equation of the plane perpendicular to ~k that contains the position vector
~r, which means that for all points lying in that plane the electric field is constant.
56 Chapter 5. Electromagnetism

Plane waves with a single frequency, also known as monochromatic plane wave, solutions to
Maxwell’s equation are given by:

~E = E~0 exp[i(~k ·~r − ωt)]

(5.28)
~0 exp[i(~k ·~r − ωt)]
~B = B

where E~0 and B~0 are some constant vectors that are not independent to each other. Indeed, let’s
substitute Eq. (5.28) into Maxwell’s equation. Equation ~∇ · ~E = 0 gives:

~∇ · ~E = E
~0 · ~∇ exp[i(~k ·~r − ωt)]
=E~0 · exp[i(~k ·~r − ωt)]i~k
(5.29)
= i~k · ~E
=0

which implies that ~k is orthogonal to ~E. If one then apply the same method to ~∇ · ~B = 0 then we get:

i~k · ~B = 0 (5.30)

which implies that ~k is orthogonal to ~B too. Now if one takes the two Maxwell’s equations left we
can show that:

i~k × ~E = iω ~B
iω (5.31)
i~k × ~B = − 2 ~E
c

These last two equations imply that ~E and ~B are orthogonal to each other. Indeed, if one just take
the first one:

i~k × ~E · ~E = iω ~B · ~E
(5.32)
=0

since ~k × ~E is, by definition, perpendicular to ~E. So, in the end, all three vectors ~k, ~E, and ~B are
orthogonal to each other (see Fig. 5.2).
Now, if one uses Eq. (5.31), and substitute ~B (or ~E) into one another,
2
we end up with the same dispersion relation as for the 1D case: ω 2 = kc2 .
Finally, using that dispersion back into Eq. (5.31) we can show that:

|~E| = c|~B| (5.33)

However, if one goes back to the definition of the energy density of Figure 5.2: Sketch illus-
electric field (Eq. 3.35) then we have: trating how the electric and
magnetic fields of an elec-
ε0 ~ 2 c2 ε0 1 ~ 2 tromagnetic wave are or-
|E| = |~B|2 = |B| (5.34) ganised
2 2 2µ0

The last term is the energy density of magnetic fields. Electric fields and
magnetic fields contribute equally to the electromagnetic energy of a wave, they are equal partners.
 5.4 General solution of Maxwell’s equations 57

5.3.4 Polarisation of light

Let’s assume that we have an electromagnetic wave for which ~E is parallel to x, ~B is parallel to y,
and ~k is parallel to z. Now if we take the real part of the wave described by Eq (5.28), i.e. the only
part that we can measure, then we have:
~E = E0 cos(kz − ωt)~i
(5.35)
~B = B0 cos(kz − ωt)~j

Here, ~E (and ~B) is linearly polarised: The field is always parallel to the same axis, the x-axis in
this case (and y-axis for ~B). Now what happens is we superpose to electric waves with no phase
difference between the two (i.e. they reach their maxima and minima for the same z value), but one
aligned with the x-axis and the other one aligned with the y-axis .Then, the resulting field will also
be linearly polarised, but the electric field will be aligned with an axis which neither x or y.
Now, let’s again superpose two electric waves, one aligned with the x-axis, one aligned with
the y-axis and |E ~1 | = |E
~2 |, but with a phase difference of π/2 (i.e. when one reaches its maximum,
the other one is at zero). In that case, the wave becomes circularly polarised, i.e. as z increases, the
direction of ~E rotates in the x − y plane.
Finally, if we do the same experiment as above but we take |E ~1 | = ~2 |, then we end up with an
6 |E
elliptically polarised wave.

5.4 General solution of Maxwell’s equations

5.4.1 Lorenz gauge
For the general case, i.e. where currents and charges are present, the relevant set of Maxwell’s
equations are given by Eq. (5.12). As we have already seen in the magnetostatics chapter of these
notes, ~∇ · ~B = 0 implies that one can represent the magnetic field by a vector potential ~A such that:
~B = ~∇ × ~A (5.36)

However, because of Faraday’s law of induction, ~∇ × ~E 6= 0, and therefore, unlike in the electrostat-
ics case, one cannot write anymore that ~E = −~∇ϕ. Instead, one can write that:
~E = −~∇ϕ + C(~
~ r,t) (5.37)

translating the fact that there is, in the general case, a non-potential component to the electric field.
By substituting Eq (5.36) and (5.37) into Faraday’s law, we get:
~∇ × (−~∇ϕ + C)
~ = −∂t (~∇ × ~A)
~∇ × C
~ = −∂t (~∇ × ~A) (5.38)
~ = −∂t ~A
C

So that the electric field becomes:

~E = −~∇ϕ − ∂t ~A (5.39)

And therefore, Gauss’s law can be written as:

ρ
−(∆ϕ + ~∇ · ∂t ~A) = (5.40)
ε0

Now, remember that the magnetic field is gauge invariant (see Section 4.2), which means that ~A is
known up to the gradient of any function, giving us the opportunity to choose a gauge that simplifies
 58 Chapter 5. Electromagnetism

the calculations. In the magnetostatics case, we chose Coulomb’s gauge defined by ~∇ · ~A = 0. Here,
we choose Lorenz gauge defined by:

~∇ · ~A = − 1 ∂t ϕ (5.41)
c2

Lorenz gauge is a generalisation of Coulomb’s gauge.

5.4.2 D’Alembert formulation of Maxwell equations

Taking the time derivative of Lorenz gauge and substituting it in Eq. (5.40), we get:

ρ
2ϕ = − (5.42)
ε0

where:
1 ∂2
2 = ∆− (5.43)
c2 ∂ 2t
is the d’Alembert operator. Now if we substitute Eq (5.36) and (5.37) into the corrected version of
Ampere’s law, we obtain:

2~A = −µ0 J~ (5.44)

Equations (5.42) and (5.44) are equivalent to the four Maxwell’s equations. In essence they
represent 4D (3 space, 1 time) Poisson equations. However, they reduce the number of components
from 6 (3 for ~E, 3 for ~B) to 4 (3 for ~A, 1 for ϕ), and also fully separate the problems for ~A and ϕ
(i.e. one can solve them separately).
In the special case where there is no charge and no current (considered in the previous section),
Eqs. (5.42) and (5.44) becomes:

2ϕ = 0
(5.45)
2~A = 0

which is the equivalent the more usual:

2~E = 0
(5.46)
2~B = 0

In the special case of stationary currents and charges, Eqs. (5.42) and (5.44) become the Poisson
equations of magnetostatics and electrostatics:
ρ
∆ϕ = −
ε0 (5.47)
~
∆A = −µ0 J~

5.4.3 Solutions of the 4D Poisson equations

Equations (5.42) and (5.44) have explicit solutions similar to those derived for the electrostatics
and magnetostatics cases:

|~r−~r0 |
1 y ρ(~r0 ,t − c ) 0
ϕ(~r,t) = dV (5.48)
4πε0 |~r − ~r0 |
 5.5 Generating electromagnetic waves 59

and

µ0
~ ~r0 ,t − |~r−~r0 | )
y J(
~A(~r,t) = c
dV 0 (5.49)
4π ~ 0
|~r − r |

Equation (5.48) can be proven by substituting it into Eq. (5.42) and using the fact that ∆ 1r =
−4πδ (~r). Equation (5.49) can be proven in a very similar way, by just taking each component of ~A
separately.
The main difference between the solutions presented in Eqs. (5.48) and (5.49) with the static
analogues is the retardation effect, manifesting itself in the time argument:

|~r − ~r0 |
T =t− (5.50)
c
~0
Indeed, |~r−cr | is the time it takes for the signal to travel from the source at point ~r0 to the observer at
point~r. This signal is delivered by electromagnetic waves which travel with the speed of light c. In
other words, it is only at time t the observer at point~r can see the result of what was happening at
point ~r0 at an earlier time T .

5.5 Generating electromagnetic waves

As we have seen above, the general solution of Maxwell’s equations provided in Eq. (5.49) allows
us to determine the magnetic field part of the electromagnetic field. In the following we are going
to see how one can link this solution and the generation of electromagnetic waves. Note that we
will be focussing on ~B since it only depends on ~A while ~E depends on both ~A and ϕ.
Let’s consider the simplest case of an electric charge q oscillating in time about the origin. Its
coordinate is given by:

~r = ~a cos(ωt) (5.51)

where ~a is a constant vector, and ω is the angular frequency of the oscillations. For a static point
charge located at position~r, the charge density can be written as ρ(~r) = qδ (~r), while for a charge
moving according to Eq. (5.51) it is:

ρ(~r,t) = qδ (~r −~a cos(ωt)) (5.52)

The corresponding electric current density is:

J~ = ρ(~r,t)~v(t) (5.53)

where:

d~r
~v(t) = = −ω~a sin(ωt) (5.54)
dt

is the charge velocity. Using Eqs (5.53) and (5.54) in the general solution of ~A (Eq. 5.49), we get:

y ρ(~r0 , T )~v(T )
~A(~r,t) = µ0 dV 0 (5.55)
4π ~
|~r − r |0
60 Chapter 5. Electromagnetism

If now we use the far-field approximation r  a then we also have r  r0 which means that we can
use the following approximations: 1~0 ' 1r and T ' t − cr . Equation (5.55) can then be rewritten
|~r−r |
as:
y
~A(~r,t) = µ0~v(T ) qδ (~r −~a cos(ωt))dV 0
4πr (5.56)
µ0~v(T )q
=
4πr

Now we can find the magnetic field:

~B = ~∇ × ~A
 
~ µ0~v(T )q
= ∇×
4πr (5.57)
 
µ0 qω ~ sin(ωT )
= ~a × ∇
4π r

Noting that ~∇T = − 1c ~∇r = − ~ucr we obtain the following:

 
~∇ sin(ωT ) = − ω cos(ωT )~ur − 1 sin(ωT )~ur (5.58)
r cr r2

Because the second term above is λ /r times smaller than the first one, where λ is the light
wavelength, it can be neglected. This is obviously only valid in the situation where r  r, i.e. we
measure the field at distances much larger than the wavelength, which is a similar approximation to
the far field approximation already made. Finally, if ~a is along the y-axis,~r is the x − y plane, then
the magnetic field can be expressed as:

2
~B ' µ0 dω sin θ cos(kr − ωt)~uz (5.59)
4πc r

where d = qa is the electric dipole moment magnitude, θ is the angle between ~a and ~r with
~ur ×~a = a sin θ~uz , and k = ω/c is the wave number. The solution for the magnetic field presented
in Eq. (5.59) is that of any linear antenna.
There are a few important points to make from Eq. (5.59)
(see also Fig 5.4). First, B = 0 for θ = 0 and θ = π, i.e. in
the direction of the charge oscillations ~a, and B is maximum
at θ = ±π/2, i.e. on the x-axis perpendicular to ~a. So the
emission of a linear antenna is very much directional. Another
important point is that the magnetic field scales as 1/r, which
is expected as the energy ∝ B2 as to decrease as 1/r2 . Also,
for large distances, plane waves are generated as the direction Figure 5.3: Sketch illustrating the
of propagation and the position vector are parallel, i.e B ∝ propagation pattern of the magnetic
cos(~k ·~r − ωt). Finally, integrating over all directions, one can field part fo the electromagnetic wave
obtain the total radiation power of an antenna: generated by a charge that oscillates
along the y axis.
d2ω 4
P= < cos2 (ωt + ϕ) > (5.60)
6πε0 c3

where the time averaging < cos2 (ωt + ϕ) >= 1/2

 5.6 The energy density of electromagnetic fields 61

5.6 The energy density of electromagnetic fields

Let’s take the corrected version of Ampere’s law and multiply by ~E, we therefore get:

~
~E · ~∇ × ~B = µ0 ~E · J~ + 1 ~E · ∂ E (5.61)
c2 ∂t

Now let’s take Faraday’s law of induction and multiply it by ~B, we get:

~
~B · ~∇ × ~E = −~B · ∂ B (5.62)
∂t

Now if we subtract the two equations above we obtain:

2 2
~E · ~∇ × ~B − ~B · ~∇ × ~E = µ0 ~E · J~ + 1 ∂ E + 1 ∂ B (5.63)
2c2 ∂t 2 ∂t

In free space we have ~B = µ0 H

~ so the previous equation can be written as:

2 2
~E · ~∇ × H ~ · ~∇ × ~E = ~E · J~ + ε0 ∂ E + µ0 ∂ H
~ −H (5.64)
2 ∂t 2 ∂t

The left hand side term is equal to −~∇ · (~E × H),

~ which means that we can rewrite the equation
above as:
2 2
~ = ~E · J~ + ε0 ∂ E + µ0 ∂ H
−~∇ · (~E × H) (5.65)
2 ∂t 2 ∂t

Or again:
 
~∇ · ~S + ~E · J~ + ∂ ε0 ~ 2 1 ~2
E + B =0 (5.66)
∂t 2 2µ0

where:

~S = ~E × H
~ (5.67)

is known as the Poynting vector, and Eq (5.66) is known as the Poynting theorem. This theorem is
the electromagnetic energy conservation law, and ~S has the meaning of the electromagnetic energy
flux density. Being integrated over a given surface, it tells us that is the energy flux through that
surface. Physically, it indicates in which direction the electromagnetic energy flows and quantifies
that energy.

5.7 Electromagnetic waves in a medium

We have seen, in the electrostatics and magnetostatics cases, that the presence of materials (di-
electrics, conductors, dia/para/ferromagnetic) influence the electric and magnetic fields via the
generations of induced charges and currents causing polarisation ~P and magnetidation M.
~ Now we
are going to evaluate the influence of these materials on the progagation of electromagnetic waves.
 62 Chapter 5. Electromagnetism

5.7.1 Solutions of Maxwell equations in a medium, and the refractive index

First, let’s consider a medium where these is no free charges and no free currents, only induced
ones are present. In this case, Maxwell equations become:
~∇ · ~D =0

~∇ × ~E ∂ ~B
=−
∂t (5.68)
~
~∇ × H ~ =− D
∂
∂t
~∇ · ~B =0
with a linear relation between fields (for weak electric and magnetic fields):
~D = εε0 ~E
(5.69)
~B = µ µ0 H
~

If one considers a monochromatic wave (i.e. with a fixed and unique frequency), the solution of
Equations (5.75) are similar to those derived in the vacuum and we have:
~   ~ 
E E0
~D  D ~ 
  =  0 ei(~k·~r−ωt) (5.70)
 ~B   B ~0 
~
H ~0
H
If one then substitutes Eq. (5.70) into Equations (5.75), we obtain (similarly to the vacuum
case) that:
~k · ~D = 0
~k × ~E = ω ~B
(5.71)
~k × H ~ = −ω ~D
~k · ~B = 0

And therefore, we see, as in the vacuum case, that ~E, ~B, and ~k are all perpendicular to each other,
with the addition that ~E is parallel to ~D and that ~B is parallel to H.
~
Again, similarly to the vacuum case, one can show that the dispersion relation for a wave in a
medium with no free charges and free currents is:
c2 2
ω2 = k = v2 k2 (5.72)
µε
Here, v = √c is the speed of light in a medium and:
µε
√
n= µε (5.73)
is the refractive index of the medium. For example, glass has ε = 2.25 and µ = 1, so that n = 1.5.
The speed of light in glass is 1.5 times slower than in vacuum. Note as well that the wavelength
also changes:
2π 2π c λvac
λmed = = = (5.74)
k ω n n
so the wavelength is also reduced by the same factor.

5.7.2 Maxwell’s boundary conditions

5.7 Electromagnetic waves in a medium 63

Maxwell’s boundary conditions follow from Maxwell’s equation in a

medium:
~∇ · ~D =ρ f

~∇ × ~E ∂ ~B
=−
∂t (5.75) Figure 5.4: Even if the
~
~∇ × H ~ = − D + J~f
∂ boundary between two me-
∂t dia is not flat, one can al-
~∇ · ~B =0 ways consider a piece of it
and are useful when fields are found in different regions separately. They which is small enough so
state that the tangent components of ~E and H, ~ as well as the normal that the boundary can be
components of ~B and ~D (relative to the boundary between medium 1 considered flat, and there-
and medium 2) are continuous. These conditions are: fore define a local normal
and tangent component of
E~1t = E~2t
the field with respect to that
H~1t = H~2t boundary
(5.76)
D~1n = D~2n
B~1n = B~2n
To prove the condition on ~D, one needs to integrate the differential form
of Gauss’s law over the volume of infinitely thin (∆z → 0) cuboid with
finite area A, and apply the divergence theorem (see Fig. 5.5):
{ y
~D · d~S = ρ f dV (5.77)
s V

The volume integral vanishes, because V = A∆z → 0, but ρ f is finite. Regarding the flux of ~D, we
have:
{
~D · d~S = A(D1~· n~1 + D
~ 2 · n~2 ) = 0 (5.78)
s
since the contribution of the 4 thin sides of teh cuboid vanishes due to
the fact that ∆z → 0. Since n~1 = −~ n2 we end up with D~1n = D~2n which
are the normal components of the vector D ~ 1 and D
~ 2 . The boundary
~
condition for B is obtained in the exact same way.
Now regarding the boundary condition on H, ~ we need to integrate
the differential form of the corrected version of Faraday’s equation over
the area of an infinitely thin rectangle, ∆ → 0 so that we get (see Fig.
5.6):
Figure 5.5: Sketch illus-
x x ∂ ~D
!
~∇ × H
~ · d~S = trating the configuration of
+ J~f · d~S (5.79)
∂t the cuboid taken to com-
S S
pute the boundary condi-
~
The right hand side vanishes as ∂∂tD and J~f are finite, but the rectangle tion for ~D.
area L∆z → 0 as ∆z → 0. Now one can apply Stokes theorem to the left
hand side:
I
~ · d~r = −H
H ~ 1 ·~i + H
~ 2 ·~i = 0 (5.80)

since the two sides of length ∆z → 0. This therefore leads to H~1t = H~2t for the tangent components
~ 1 and H
of H ~ 2 . Then the boundary condition for ~E is determined in the exact same way.
 64 Chapter 5. Electromagnetism

5.7.3 Reflection and transmission of electromagnetic waves: Normal

incidence
Consider a flat boundary at z = 0 between two media with refractive indices
n1 and n2 . Let ~E be directed along x (see Fig 5.7). The magnitude of the
electric field of incoming (i), reflected (r), and transmitted (t) wave is given
by:

Ei (z,t) = ei(k1 z−ωt)

Er (z,t) = rei(−k1 z−ωt) (5.81)
i(k2 z−ωt) Figure 5.6: Sketch il-
Et (z,t) = te
lustrating the config-
where r and t are the amplitudes of the reflected and transmitted wave, uration of the rectan-
respectively. The amplitude of the incoming wave is set to 1 for convenience gle taken to compute
(although it can be any). In order to determine r and t we need to apply the boundary condi-
Maxwell’s boundary conditions (Eqs 5.76). ~
tion for H.
Since in our case the field is in the x direction, and travels in the z direction
(see Fig 5.7), there isn’t any normal component, so we only use the first
two boundary conditions. The first one tells us that the electric field is
continuous across the boundary, while the second tells us that it is its spatial
derivative that is continuous. Indeed, if we take Maxwells second equation
(see Eqs. 5.75), we have:
~∇ × ~E = ∂z E~j = −∂t ~B = iω µ µ0 H
~ (5.82)
~ implies, in this case, the continuity of ∂z E.
Therefore the continuity of H
Applying these two boundary conditions to Eqs (5.87), we get:

1+r = t
(5.83)
k1 (1 − r) = k2t

since E1 = Ei + Er and E2 = Et . By combining the two

expressions above together we find:

k1 − k2 n1 − n2
r= = (5.84)
k1 + k2 n1 + n2

and

2k1 2n1
t= = (5.85)
k1 + k2 n1 + n2
Figure 5.7: Sketch illustrating the configu-
where we used k = ω
So, these equation show that: ration for the normal refraction case.
c n.

n2  n1 r → +1
(5.86)
n2  n1 r → −1

These mean that the wave is 100% reflected, with the same phase (n2  n1 ) or with a phase shift of
π (n2  n1 ) at z = 0. Also we see that r = 0 when n2 = n1 , which is when there is effectively no
boundary since the media have the same refractive index.

5.7.4 Reflection and transmission of electromagnetic waves: Non-normal incidence

5.7 Electromagnetic waves in a medium 65

Let’s first treat the case of S-polarisation, also known as

transverse-electric (TE) polarisation, which is when the
electric field ~E is parallel to the interface. Let ~E be along
the y-axis, so that we have (see Fig. 5.8):

Ei (x, z,t) = ei(kx x+k1 z−ωt)

Er (x, z,t) = rei(kx x−k1 z−ωt) (5.87)
Et (x, z,t) = tei(kx x+k2 z−ωt)

essentially adding a eikx x factor to all fields. This factor is Figure 5.8: Sketch illustrating the config-
the consequence of the phase continuity at z = 0. Indeed, uration for the non-normal S-polarisation
the phase of all three waves have to be equal in order case.
to be physical, there is nothing that would justify the
phase fo the wave to suddenly jumps to a different value.
Therefore we have, at z = 0:
~ki ·~r = ~kr ·~r = ~kt ·~r
(5.88)
kix x + kiy y = krx x + kry y = ktx x + kty y
and since this has to be true for any set of coordinates we have:

kix = krx = ktx = kx

(5.89)
kiy = kry = kty = ky

In the example we are considering the field has no y component, so only kx appears in our set of
electric fields. From the light dispersion relation we have:
ω
|~ki | = |~kr | = n1
c (5.90)
~ ω
|kt | = n2
c
which means that we have:
ω ω
kx = n1 sin θ1 = n2 sin θ2 (5.91)
c c
where θ1 and θ2 are the incident and refracted angles, respectively. So previous equation can reduce
to:

n1 sin θ1 = n2 sin θ2 (5.92)

which is known as Snell’s law, which is a direct consequence of phase continuity at the interface.
Now, let’s apply Maxwell’s boundary conditions which give that, as in the Normal case, E and
∂z E are continuous. Applying the exact same method we obtain the same relation between r, t, and
k1 and k2 , but this time k1 = n1 cos θ1 , and k2 = n2 cos θ2 (i.e. the components of ~ki and ~kr along
the z axis). which means that we end up with:

k1 − k2 n1 cos θ1 − n2 cos θ2
r= = (5.93)
k1 + k2 n1 cos θ1 + n2 cos θ2

and
2k1 2n1 cos θ1
t= = (5.94)
k1 + k2 n1 cos θ1 + n2 cos θ2
66 Chapter 5. Electromagnetism

Using Snell’s law one can substitute θ2 and we get, for the reflectivity, :

 q 2
1 − ( nn12 sin θ1 )2
n1 cos θ1 − n2
R = r2 =  q  (5.95)
n1 2
n1 cos θ1 + n2 1 − ( n2 sin θ1 )

which is Fresnel’s equation for S-polarisation.

For P-polarisation, or transverse-magnetic (TM) po-
larisation, ~B is parallel to the interface (see Fig. 5.9).
Snell’s law remain the same, and the boundary conditions
1
are the continuity of B and 0 epsilon ∂z B. These are derived
in a very similar way as we have done for E in the case
of S-polarisation. The relations between r and t become:

1+r = t
k1 (1 − r) k2t (5.96)
=
ε1 ε2
Figure 5.9: Sketch illustrating the config-
which leads to: uration for the non-normal P-polarisation
case.
k1 /ε1 − k2 /ε2 n2 cos θ1 − n1 cos θ2
r= = (5.97)
k1 /ε1 + k2 /ε2 n2 cos θ1 + n1 cos θ2

and finally, Fresnel’s equation for P-polarisation is:

 q 2
1 − ( nn12 sin θ1 )2
n2 cos θ1 − n1
R = r2 =  q  (5.98)
n2 cos θ1 + n1 1 − ( nn12 sin θ1 )2

At Brewster’s angle (see Fig. 5.10) the reflectivity R in P-

polarisation vanished. This allows, in particular, to filter out unpo-
larised light (leaving in reflection only the S-polarised component).

Figure 5.10: Sketch show-

ing the relectivity R as a func-
tion of the incident angle. We
see that R goes to zero for P-
polarisation. The corresponding
angle is known as Brewster’s an-
gle.
 6. Fluid mechanics

In this last chapter, we are going to introduce the topic of fluid mechanics. While this is an entirely
different physics topic, fluid mechanics deals with vector fields in a very similar way to what we
have been doing until now with electromagnetism. In fact, in certain conditions, fluid mechanics is
extremely similar to magnetostatics, as we are going to discover.

6.1 The continuity equation for mass

In Section 5.2.1, we have shown that for electric charges to be conserved, the electric charge density
ρe and the current volume density J~ have to be linked by the following continuity equation:
∂t ρe + ~∇ · J~ = 0 (6.1)
Now we are going to show that one gets a similar equation when it comes to mass conservation.
Indeed, let’s define the mass of matter within a given volume by:
y
m= ρ(~r,t)dV (6.2)
V

where ρ is the mass density. The rate of mass change within the same volume is therefore given by:

y
∂t m = ∂t (ρ(~r,t)) dV (6.3)
V

For the mass to be conserved, the rate of mass change within V must be equal to the opposite of the
mass flux in/out of V. Remember, a positive flux is one where the quantity under consideration is
escaping the volume (this is a direct consequence of the convention regarding the normal vectors to
closed surfaces pointing outwards). Therefore, if the mass within the volume increases over time,
for instance, then it means that the mass flux must be negative. So to be equal, one needs to take
the negative of it. This mass flux is given by:
{
φm = ρ(~r,t)~v(~r,t) · d~S (6.4)
S
 68 Chapter 6. Fluid mechanics

where S is the surface enclosing the volume V , and ~v is the fluid velocity field. So we have:
y {
∂t (ρ(~r,t)) dV = − ρ(~r,t)~v(~r,t) · d~S (6.5)
V S

Using the divergence theorem we get:

y y
∂t (ρ(~r,t)) dV = − ~∇ · (ρ(~r,t)~v(~r,t))dV (6.6)
V V

And this this has to be valid for any volume, we get (dropping the position and time variable for
clarity):

∂t ρ + ~∇ · (ρ~v) = 0 (6.7)
This is the continuity equation for a fluid, or also known as the mass conservation equation. One
can see that this is identical to the continuity equation for electric charges. In fact, a continuity
equation for all conserved quantities can be derived.
A special, but important case is that of incompressible fluids. These are fluids for which ρ(~r,t)
is constant. Let’s expand the right-hand-side term of the continuity equation. We get, using the
chain rule:
~∇ · (ρ~v) = ρ~∇ ·~v +~v · ~∇ρ (6.8)
For an incompressible fluid ~∇ρ = 0and ∂t ρ = 0, therefore the continuity equation transforms into:

~∇ ·~v = 0 (6.9)
The velocity field of incompressible fluids is solenoidal.

6.2 Hydrostatics
Hydrostatics is the study of fluids at rest: All time derivatives are
equal to zero. We will here focus on how forces applied to a piece
of fluid balance out to maintain that fluid in a static equilibrium.
One key quantity that characterise the state of a fluid is its pressure.
The general definition of the pressure p is:

d ~F = −pd~S (6.10)

where d~S is an infinitesimal surface element, and d ~F is the force

acting on this surface element. The negative sign comes from the Figure 6.1: Sketch illustrat-
fact that the direction of the force applied to the surface of the ing the direction of the pressure
~
volume is in the opposite direction to d~S which is always outwards forces d F and the normal sur-
(see Fig. 6.1). Pressure is a scalar quantity (same as, for instance, face vector, i.e. always in oppo-
temperature), and can be defined as at every position within a fluid site directions.
element.
Consider now a reservoir filled with water (which to a good approximation, is incompressible).
Select an infinitely small cylindrical volume of height dz0 and based area dS (see Fig. 6.2). Then
the force on the bottom side of the cylinder due to water pressure is p(z0 )dS. The force on top side
of the cylinder due to water pressure is p(z0 + dz0 )dS which we can also express as (p + d p)dS. In
the case of a static fluid, forces balance out, and obtain:
pdS = (p + d p)dS + dmg (6.11)
6.2 Hydrostatics 69

where dm is the mass of the cylinder, and g the gravitational accel-

eration. Here we consider that the gravitational force is uniform
throughout the infinitely small cylindrical volume. We can rewrite
the mass as:

dm = ρdV = ρdSdz0 (6.12)

Within the infinitely small volume the density can be considered

uniform (even more so for an incompressible fluid such as water),
an we get:
Figure 6.2: Sketch illustrating
d p = −ρgdz 0
(6.13) the situation described in the text
where we consider a small cylin-
Now let’s integrate this between z0 = z and z0 = 0 (z0 = 0 correspond drical volume within water reser-
to the surface fo the water reservoir), we get: voir.

Z 0
p(z) − p(0) = − ρgdz0
z (6.14)
p(z) = p(0) − ρgz

The pressure increases linearly with depth (−z). Here, one could get the density out of the integral
because of the incompressibility of water.
In another example, we can consider the pressure balance in the atmosphere. The same balance
of forces as in Eq. (6.13) applies. However, one difference here is that the gas in the atmosphere
is compressible, and its density is a function of altitude. So one cannot get the density out of the
integral as we did above with water. Instead, the gas in the atmosphere is well approximated by an
ideal gas for which the relation between pressure, density and temperature is well known:

NkB T
p=
V
p = nkB T (6.15)
ρkB T
p=
mmol
where N is the number of molecules within volume V , kB is the Boltzmann’s constant, T the
temperature, n = VN the number density (the number of molecules per unit volume), and mmol = ρn
is the average mass of a molecule. Substituting Eq. (6.15) within Eq. (6.13) we get:
mmol g
dp = − pdz0 (6.16)
kB T

Now let’s integrate again between z0 = z and z0 = 0 using variable separation:

Z p(z)
dp mmol g z 0
Z
=− dz
p(0) p kB T 0
 
p(z) mmol g
ln =− z (6.17)
p(0) kB T
 
mmol g
p(z) = p(0) exp − z
kB T

It is interesting to note here that mmol gz is the gravitational potential energy of a molecule at altitude
z, an dkB T is the approximate average energy of a molecule in thermal equilibrium. It is important
 70 Chapter 6. Fluid mechanics

to note here that in the derivation of Eq. (6.17) we assumed that the temperature was not a function
of altitude (i.e. we got it out of the integral), while we all know that temperature does vary with
altitude. A full solution would require solving thermal processes in the atmosphere in order to get
an expression for the temperature independently of pressure and density.
Finally, it is important to stress that Eq. (6.13) is a very general expression that is at work in
many different situation. For instant, hydrostatic equilibrium is what hold stars together, preventing
them from collapsing. In the case of stars, equation 6.13 is expressed as:
dp Gm
= −ρ 2 (6.18)
dr r
where r is the radial coordinate, m is the mass within radius r and G is the gravitational constant.
This expression is basically identical to Eq. (6.13), once we realised that Gm
r2
is the gravitational
acceleration generated by the mass m.

6.3 Stream functions of incompressible fluids

We have seen that, for an incompressible fluid, the mass continuity equation reduces to:
~∇ ·~v = 0 (6.19)

Similarly to the magnetostatic case, any solenoidal field can be written as the curl of a vector
potential vector field ~A(~r,t) such that:

~v = ~∇ × ~A (6.20)

However, for a 2-dimensional flow, the vector potential ~A(~r,t) is represented by a scalar function
ψ(~r,t) known as the stream function. The reason for this, is due to the fact that the curl of a 2D
vector field, let’s say in the xy plane, is a vector perpendicular to that plane, ie along the z axis, and
therefore a scalar quantity is enough to represent such a field, ~A = (0, 0, ψ).
This means that the velocity field of a 2D incompressible flow can be rewritten as:

v(~r,t) = ~∇ × ~A = (∂y ψ)~i − (∂x ψ)~j (6.21)

Or, said differently, the stream function of a 2D incompressible flow lying in the xy plane must
satisfy the conditions:

vx = ∂y ψ
vy = −∂x ψ (6.22)
vz = 0

The reason why stream functions are useful is because they give the directions of the flows. Indeed,
lines of constant ψ are tangent to the velocity field ~v. Let’s show that this is indeed the case. The
normal unit vector to the stream function ψ is defined as:
~∇ψ
~n = (6.23)
|~∇ψ|
Now if ψ is tangent to ~v then ~n is perpendicular to ~v and ~n ·~v = 0. For a 2D incompressible flow we
have:
~v = ~∇ × ~A0
~v = ~∇ × (ψ~k) (6.24)
~v = ~∇(ψ) ×~k
 6.4 Euler’s equation 71

which means that we have:

~∇ψ · (~∇ψ) ×~k)
~n ·~v = =0 (6.25)
|~∇ψ|

since ~∇ψ ×~k is, by definition of the cross product, perpendicular to ~∇ψ. So lines ψ = constant do
give, indeed, the direction of the flow velocity and are known as streamlines.
Another important fluid quantity is vorticity, which is defined as:

~Ω = ~∇ ×~v (6.26)

Irrotational fields, have, by definition, zero vorticity, and for these one can define a scalar potential,
as in electrostatics, such that:

~v = −~∇ϕ (6.27)

Lines of equipotential, i.e. ϕ = constant, are, by definition perpendicular to the velocity field and
the lines of constant stream function.

6.4 Euler’s equation

Let’s now look at the piece of fluid that isn’t static, and let’s try to establish an equation that describe
the fluid’s motion. To achieve this, let’s consider an arbitrary volume V (see Fig. 6.1), and let’s
apply Newton’s second law to it. We have:
y y {
dm~a = dm~g + d ~F (6.28)
V V V

where d ~F is the force due to the pressure acting on the small element of surface d~S part of the
boundary surface of V :
{ { y
~F = d ~F = − pd~S = − ~∇pdV (6.29)
S S V

where the last equality is obtained using a corollary of the divergence theorem (derived by replacing
the vector field by a the product of a scalar field and a non-zero constant vector). Substituting this
last equation into Newton’s second law above, we obtain:
y y y
dm~a = dm~g − ~∇pdV (6.30)
V V V

Since this has to be valid for any volume, and nothing that dm = ρdV , we obtain:

ρ~a = ρ~g − ~∇p (6.31)

Now, the acceleration ~a can be written as the total time derivative of ~v(~r,t) (since during the time
interval dt a fluid particle will have also changed position from~r to~r + d~r). This can be written as:

d~v dx dy dz
~a = = ∂t~v + ∂x~v + ∂y~v + ∂z~v
dt dt dt dt (6.32)
= ∂t~v + (~v · ~∇)~v
 72 Chapter 6. Fluid mechanics

Substituting Eqs (6.32) in Eq. (6.31), we obtain Euler’s equation of fluid motion:

~∇p
∂t~v = −(~v · ~∇)~v − ~∇φg − (6.33)
ρ

where we expressed the gravitational acceleration ~g as a function of the gravitational potential φg :

~g = −~∇φg . Euler’s equation describes the motion of an ideal fluid, and, even though it has a term
containing the fluid vorticity (~Ω = ~∇ ×~v), the latter is normally zero, reflecting the fact that an ideal
(i.e. inviscid) fluid has no vortices.
In order to obtain, and describe properly vortices, one needs to introduce terms in Eq. (6.33)
containing viscosity η. In real fluids, η 6= 0. We will here admit that the simplest term that involves
viscosity is η∆~v. This is some sort of frictional term where friction between layers of fluids
generates transverse motions. Adding this term leads to the simplest form of the Navier-Stokes
equation:

~∇p
∂t~v = −(~v · ~∇)~v − ~∇φg − + η∆~v (6.34)
ρ

Let’s have a look at the meaning of each term here. On the left, we have the local acceleration of
the fluid. On the right hand side the first term correspond to the velocity advection, i.e. the change
of the fluid particle velocity as the result as it changes position along the flow. Then we have the
gravitational acceleration term, followed by the acceleration due to the pressure gradients, and
finally the viscosity acceleration that will tend to spread inhomogeneities of velocities.

6.5 Bernouilli’s law

Let’s first rewrite Euler’s equation in a slightly different form using the identity relation for the
triple vector product:
1
(~v · ~∇)~v = ~∇v2 −~v × (~∇ ×~v) (6.35)
2
which leads to an alternative expression of Euler’s equation:

1 ~∇p
∂t~v = − ~∇v2 +~v × (~∇ ×~v) − ~∇φg − (6.36)
2 ρ

Let’s now apply the following limitations. First, the flow is steady (∂t~v = 0), the fluid is incompress-
ible (ρ = constant), and the velocity field is irrotational (~∇ ×~v = 0). With these three limitations,
Euler’s equation becomes:
 2 
~∇ v + p + φg = 0 (6.37)
2 ρ
Note that if we replace the irrotational condition by only looking at the flow along a streamline (i.e.
project along a streamline using ~v·), we obtain the same condition:
  2 
~ ~ v p
~v · ~v × (∇ ×~v) − ∇ + + φg =0
2 ρ
 2  (6.38)
~ v p
~v · ∇ + + φg = 0
2 ρ
 6.6 Vortices 73

In both cases, we end up with Bernouilli’s law:

v2 p
+ + φg = constant (6.39)
2 ρ

Bernoulli’s law reflects energy conservation (again). If one multiply every term by the density
ρ, we realise that the first term in the kinetic energy per unit volume of the flow, the third is the
gravitational potential energy per unit volume of the flow, while the second term term corresponds
to the work done by the fluid due to the pressure.

6.6 Vortices
We have previously seen that the vorticity of a fluid ~Ω is defined as:
~Ω = ~∇ ×~v (6.40)

If vorticity is zero, then the fluid does not contain any forced vortices, but it may, however, contain
free vortices. A forced vortex is caused by external forces on the fluid, e.g. rotating a container
holding the fluid.
Let’s assume that we do have a rotating bucket of water (see
Fig. 6.3). The tangential fluid velocity ~v is linearly proportional to
the radial distance r (i.e. solid body rotation):

~
~v = rω (6.41)

where ω~ is the angular velocity. From Bernouilli’s law, for all

points ate the surface of the water (experiencing the same pressure
and the same gravitational acceleration), we have:

v2
+z =C (6.42) Figure 6.3: Sketch illustrating
2g the situation described in the
where C is a constant. If we substitute the expression of the velocity, text where we consider rotating
we obtain: bucket of water

ω 2 r2
z =C− (6.43)
2g
This is the equation of the water surface profile, which has the form of a parabola.
Free vortices are formed, for instance, when water flows out of a hole i the base of a container
such as a bath plug hole (see Fig. 6.4). No external force is required to create the rotation. The
tangential fluid velocity is inversely proportional to the radial distance r:
k
~v = ~uϕ (6.44)
r
where k is a constant. One can verify that the curl of such velocity field is, indeed, zero.
If one considers Bernouilli’s law for the surface of a water bath escaping through the plug hole,
we get the same equation as for the forced vortex case (Eq. 6.42). However, the velocity now has a
different relationship with respect to r, and by substitution we get:

k2
z =C− (6.45)
2r2 g
74 Chapter 6. Fluid mechanics

This is the equation of a hyperbolic curve.

Figure 6.4: Sketch illustrating

the situation described in the text
where water of a bath is freely
escaping a plug hole.