A Practical Introduction to

Differential Forms

Alexia E. Schulz
and

William C. Schulz

August 12, 2013

Transgalactic Publishing Company

Flagstaff, Vienna, Cosmopolis
ii

c 2012 by Alexia E. Schulz and William C. Schulz

Contents

1 Introduction and Basic Applications 1

1.1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Some Conventions . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Some Formulas to Recall . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Coordinate systems . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.5 The Algebra of Differential Forms . . . . . . . . . . . . . . . . . . 6
1.6 The Operator d . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.7 Orientation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.8 Differential Forms and Vectors . . . . . . . . . . . . . . . . . . . 12
1.9 grad, curl and div . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.10 The Poincaré Lemma and it’s Converse . . . . . . . . . . . . . . 14
1.11 Boundaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.12 Integrals of Forms . . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.13 Variable Changes . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.14 Surface integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.15 The Generalized Stokes Theorem . . . . . . . . . . . . . . . . . . 23
1.16 Curvilinear Coordinates I: preliminary formulas . . . . . . . . . . 26
1.17 Curvilinear Coordinates II: the calculations . . . . . . . . . . . . 30
1.18 Surfaces and Manifolds . . . . . . . . . . . . . . . . . . . . . . . . 36
1.19 The Dualizing Operator ∗ . . . . . . . . . . . . . . . . . . . . . . 40
1.20 The Codifferential δ . . . . . . . . . . . . . . . . . . . . . . . . . 47
1.21 The Laplacian . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
1.22 Maxwell’s Equations in 3-space . . . . . . . . . . . . . . . . . . . 56
1.23 Indefinite Inner Product Spaces . . . . . . . . . . . . . . . . . . . 61
1.24 Maxwell’s equations in Space-Time . . . . . . . . . . . . . . . . . 65
1.25 Energy Flow and Energy Density . . . . . . . . . . . . . . . . . . 71

2 Mathematical Theory 73
2.1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . 74
2.2 Manifolds Mappings and Pullbacks . . . . . . . . . . . . . . . . . 74
2.3 Proofs of Stokes Theorem . . . . . . . . . . . . . . . . . . . . . . 74
2.4 Proofs of the Converse of the Poincare Lemma . . . . . . . . . . 74
2.5 Permutations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
2.6 The operator Φ . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76

iii
iv CONTENTS
Chapter 1

Introduction and Basic

Applications

1
2 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

1.1 INTRODUCTION

These notes began life as an introduction to differential forms for a mathematical

physics class and they still retain some of that flavor. Thus the material is
introduced in a rather formal manner and the mathematical complexities are
put off to later sections. We have tried to write so that those whose primary
interest is in the applications of differential forms can avoid the theoretical
material provided they are willing to accept the formulas that are derived in
the mathematical sections, which are clearly marked as such. Those who wish
may read the mathematical sections as they occur, or later, or indeed may put
them off to a more convenient time, perhaps in a future life, without loss to the
continuity of the applied thread. Anyway, such is my hope. But we want to also
emphasize that those who wish will find all the mathematical details available, at
a level of rigor usual to the better mathematical physics books. The treatment
is mostly local, and what little manifold theory is needed is quietly developed
as we go. We have tried to introduce abstract material in circumstances where
it is useful to do so and we have also tried to avoid introducing a lot of abstract
mathematical material all at one time.
The two areas most completely addressed in these notes, besides the foun-
dational material, are coordinate changes and Maxwell’s equations since we
feel that these illustrate the power of differential forms quite well. We treat
Maxwell’s equations in both three and four dimensions in separate sections. We
will also look at a few other things.
Notation has been carefully chosen to be consistent with standard tensor
notation to facilitate comparison with such treatments, and to facilitate learning
basic differential geometry.
The treatment of Maxwell’s equations requires the derivation of the potential
equations. Although not strictly necessary, we have introduced the codifferential
δ and the Laplace operator △dδ +δd since this is the natural route using modern
mathematics. For example we point out that the condition of Lorenz can be
expressed instantly and easily in terms of the codifferntial in four dimensions.
And as long as we have it available we can look at a couple of other applications
of the Laplace operator on forms.
A justified criticism of these notes might be that many things are done
twice, which is not efficient. We have sacrificed efficiency for convenience to the
reader who may wish to deal with only one particular thing, and so would like a
relatively complete treatment in the section without having to read five others.
Similarly, many formulas are repeated at the beginning of sections where they
are used, rather than referred to in previous sections. The increase in paper is
rather small, and for those getting it electronically there is no waste at all. It
is difficult for a mathematician to resist the call of generality but since one of
us is a physicist the brakes have been applied, and we hope that the product is
a reasonable compromise between the siren song of mathematics and the needs
of practical physics.
1.2. SOME CONVENTIONS 3

1.2 Some Conventions

Here we will introduce some conventions that will be used throughout these
notes. The letter A will be used for a region of 2-dimensional space, for example
the unit disk consisting of points whose distance from the origin is less than or
equal to 1. It’s boundary would be the unit circle consisting of points whose
distance from the origin is exactly 1. We will use the symbol ∂ to indicate the
boundary. Thus if A is the unit disk A = {x ∈ R2 | |x| ≤ 1} then the boundary
of A is ∂A = {x ∈ R2 | |x| = 1} which is the unit circle. Notice carefully the
difference between the terms DISK and CIRCLE. (DISK and CIRCLE are often
confused in common speech.)
The letter M will be used for a (solid) region of 3 dimensional space, for
example the unit ball, M = {x ∈ R3 | |x| ≤ 1} whose boundary is the unit
sphere ∂M = {x ∈ R3 | |x| = 1}. (The terms BALL and SPHERE are often
confused in common speech, particularly in cases like a beach ball or a basketball
since they are filled with air.)
The letter S will be used for a (2 dimensional) surface in three dimensional
space, for example the upper half of the unit sphere. The boundary of this S
would be a circle in the x, y plane.
If we do not wish to specify dimension, we will use the letter K. The use of
K indicates that the formula will work in any dimension, and this usually means
any dimension, not just 1, 2 or 3 dimensional space. Naturally ∂K means the
boundary of K.
The ball and sphere have analogs in every dimension. It is customary to
refer to the ball in Rn is the n-ball and its boundary as the (n − 1)-sphere.
For example, the unit disk is the 2-ball and its boundary, the unit circle, is
the 1-sphere. Note that the m-sphere lives in Rm+1 . It is called the m-sphere
because it requires m variables to describe it, like latitude and longitude on the
2-sphere.
Also useful to know are the terms open and closed. This is a tricky topo-
logical concept, so we will treat it only intuitively. K is closed if it includes
its boundary. Thus the unit disk and unit ball are closed. If we remove the
boundary ∂K from K the resulting set K ◦ is called open. Thus for the unit ball
in R3 we have

M = {x ∈ R3 | |x| ≤ 1} closed 3-ball

M◦ = {x ∈ R3 | |x| < 1} open 3-ball
∂M = {x ∈ R3 | |x| = 1} 2-sphere

We want to give a real world example here but remember it must be inex-
act since real world objects are granular (atomic) in constitution, so can only
approximate the perfect mathematical objects. Some people prefer to eat the
closed peach (with fuzzy skin), some people prefer the open peach (fuzzy skin
removed, peach◦ ) and the boundary of the peach, ∂peach, is the fuzzy skin.
Perhaps this will help you remember. Deeper knowledge of these matters can
be found in the wonderful book [2] and also [4].
4 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

For functions we will use a slightly augmented variant of the physics conven-
tion. When we write f : S → R we mean a function whose input is a point p ∈ S
and whose output is a real number. This is theoretically useful but not suitable
for calculation. When we wish to calculate, we need to introduce coordinates. If
we are dealing with the upper half of the unit sphere (set of points in R3 whose
distance from the origin is exactly one and for which z ≥ 0) then we might write
f (x, y) if we choose to represent points in the x, y coordinate system. Notice,
and this is an important point, that the coordinate x takes as input p ∈ S and
outputs a real number, it’s x coordinate. Hence the coordinates x and y are
functions just like f . If S is the upper half of the unit sphere in R3 then x
and y are not really good coordinates. It would be be better to use longitude
and colatitude for my coordinates and then we would write f (φ, θ). 1 Note
use of the same letter f no matter what the coordinate system, because the f
represents a quantity in physics, whereas in math it represents a functional re-
lationship and we would not use the same letter for different coordinates. Note
also that f (.5, .5) is ambiguous in physics unless you have already specified the
coordinate system. Not so with the math convention.
Finally, we will almost always use the letters f, g, h for functions on A, M, S, K.
Mostly these will occur in coordinate form, for example f (x, y, z) for a function
on M .

1.3 Some Formulas to Recall

You are all familiar with the dx, dy, dz which occur in the derivative notation
dy
dx and the integral notation
Z
f (x, y) dxdy
M
Z
f (x, y, z) dxdydz
M

and you recall the Green, divergence and Stokes theorems, which I list here for
convenience:
Green’s theorem
Z Z
∂g ∂f
f (x, y) dx + g(x, y) dy = − dxdy
∂A A ∂x ∂y

The divergence theorem or Gauss’s theorem

Z Z
∂f ∂g ∂h
f (x, y, z) dydz + g(x, y, z) dzdx + h(x, y, z) dxdy = + + dxdydz
∂M M ∂x ∂y ∂z
1 BEWARE. φ is longitude in physics but colatitude in mathematics. θ is colatitude in

physics but longitude in math.

1.3. SOME FORMULAS TO RECALL 5

The Classical Stokes’ theorem

Z
f (x, y, z) dx + g(x, y, z) dy + h(x, y, z) dz
∂S
Z      
∂h ∂g ∂f ∂h ∂g ∂f
= − dydz + − dzdx + − dxdy
M ∂y ∂z ∂z ∂x ∂x ∂y
You might be more familiar with the last two in the vector forms
Z Z
v · dS = div v dV
∂M M

and Z Z
v · dℓ = curl v · dS
∂S S
There are some conventions on integrals that we will mention now. In for-
mer
R R Rtimes when integrating over a three dimensional object we would write
M div v dV This is now completely antiquated, and we will not do it.
On the other hand, there is a convention that when integrating around curves
or surfaces that have no boundary we put a small circle on the integral, so that
we write I Z
v · dS for v · dS
∂M ∂M
Since this is favored by the physics community we will mostly use it. Notice that
if a geometric object is the boundary of something, then it itself has no boundary,
and so we will use the circled integral almost exclusively with boundaries.
For our purposes we will define a differential form to be an object like

f (x, y) dx f (x, y, z) dydz f (x, y, z) dxdydz

which we find as integrands in the written out forms of the Green, divergence
and Stokes theorem above. If ω is a sum of such objects it turns out that the
three theorems collapse to one mighty theorem, called the generalized Stokes
theorem, which is valid for all dimensions:
I Z
ω= dω
∂S S

To use this theorem and for other purposes it is only necessary to

1. Learn the algebra that the dx, dy, dz satisfy which is almost the same as
ordinary algebra with one important exception.
2. Learn the rule for the operator d which is almost trivial.
Once these are learned differential forms can be manipulated easily and with
confidence. It is also useful to learn how various things that happen in vector
analysis can be mimicked by differential forms, and we will do this, naively at
first and then in much more detail.
6 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

If you are concerned about what differential forms ARE, the answer is a
little tricky and we am going to put it off for the moment. Later we will
discuss the surprisingly dull answer to this question. Incidentally, the difficulty
in explaining what they really are is one reason they have not become more
common in elementary textbooks despite their extreme usefulness.
Just to give a tiny hint of the geometrical interpretation of differential forms.
A two form measures the density of lines of force of a field, as introduced by
James Faraday a century and a half ago. For more on this subject see [1] or [8].
We will discuss it a bit more when we have more equipment.

1.4 Coordinate systems

Our first job is to talk a little about coordinates. You already know most of this
so we can do it quickly. As already specified, we will use the notation A for a
finite region in R2 (which is the standard notation for ordinary two dimensional
real space). We will use M for a three dimensional finite region of R3 and S
for a curved surface in R3 . Standard coordinates for A would be x, y, but we
might want to use polar coordinates r, θ or even more general coordinates u, v.
The important thing for A is that there be two coordinates and that they be
”independent”, so that v is not a function of u.
In the case of M , a three dimensional region of R3 , we will need three
coordinates x, y, z or r, θ, φ or more generally u, v, w to describe it.
Since S is a surface in R3 , it will be described by two coordinates. In
elementary cases this will often be x, y but in more complicated situations it is
often wise to taylor2 the coordinates to the geometric object, for example if one
is working with the Unit Sphere in R3 then the appropriate coordinates would
be θ, φ (longitude and colatitude). Proper choice of coordinates can make a
nasty problem much more pleasant.
It is important to be able to move from one coordinate system for a geometric
situation to a different coordinate system and we will discuss this later.

1.5 The Algebra of Differential Forms

We now begin our discussion of the algebra of differential forms. The type of
multiplication we will use is called exterior multiplication. The term goes back
to Grassmann and he invented it to contrast with interior multiplication, which
is what we now refer to as the inner product. The multiplication symbol used
is ∧ as in dx ∧ dy but this is very often omitted. We will use it for a while,
and then omit it when we get bored with it, but return to using it whenever we
think it clarifies matters.
Let us start with a function in f (x, y, z) on R3 . You already know how to

2 bad pun
1.5. THE ALGEBRA OF DIFFERENTIAL FORMS 7

form df :
∂f ∂f ∂f
df = dx + dy + dz
∂x ∂y ∂z
Recalling that, like f , the coordinate x is also a function on R3 the previous
formula writes the differential of f in terms of the differentials of the three special
functions x, y, z. So we note that there is not much difference between df and
dx; they are the same kind of object. All objects of this type (differentials of
functions) are collected together in the set

Λ1 (R3 )

and are called 1-forms.

If we multiply a couple of these together we get objects like

f dx ∧ g dy = f g dx ∧ dy ∈ Λ2 (R3 )

Note that functions f commute with dx: f dx = dx f ; see below. Linear combi-
nations of such objects are called 2-forms. And of course there are 3-forms

f dx ∧ dy ∧ dz ∈ Λ3 (R3 )

To complete the system we will place the functions in the basement of the
building: f ∈ Λ0 (R3 ). It is customary to omit the wedge when multiplying by
a function; we write
f dx ∧ dy for f ∧ dx ∧ dy
There is no significance to this; it is just convention.
The algebra of these objects is just like ordinary algebra except for the
changes caused by the rule

dg ∧ df = − df ∧ dg (anti-commutativity)

for the 1-forms df and dg. An algebra satisfying this rule is called an exte-
rior or Grassmann algebra. This algebra was invented by Hermann Guenther
Grassmann about 1840 in an attempt to find an algebra mirroring elementary
geometry. It is sufficient to postulate this rule for the coordinate differentials
only,
dy ∧ dx = − dx ∧ dy etc.
since the general rule will follow by linearity.
Thus the exterior algebra is not commutative. Our rule is often called anti-
commutative and is the simplest generalization of commutative, but it has many
consequences which seem strange to the beginner. For example, if we substitute
f for g in the rule dg ∧ df = − df ∧ dg we get

df ∧ df = − df ∧ df
8 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

so moving the right side to the left side by the usual algebraic processes which
are all valid here we get

df ∧ df + df ∧ df = 0
2 df ∧ df = 0
df ∧ df = 0

Thus the product of a one form with itself is 0, which is very important. Let’s
look at another example

(f dx + g dy) ∧ (f dx + g dy) = f f dx ∧ dx + f g dx ∧ dy + gf dy ∧ dx + gg dy ∧ dy
= 0 + f g(dx ∧ dy + dy ∧ dx) + 0
= fg 0 = 0

as promised.
We also see from this that there are no four forms in 3-space, since if we
multiply four of the objects dx, dy, dz together there will be a repetition which
will kill the form:

dx ∧ dy ∧ dz ∧ dx = − dx ∧ dy ∧ dx ∧ dz = dx ∧ dx ∧ dy ∧ dz = 0 ∧ dy ∧ dz = 0

In general, for a space K of dimension n there will be forms ω ∈ Λj (K) for

j = 0, . . . , n. It is not true in general that for r-forms ω with r ≥ 2 that
ω ∧ ω = 0 although this is fortuitously true for dimensions n ≤ 3. There is an
example in dimension 4 where ω ∧ ω 6= 0 in the problems.
Now let us start to deal a little more abstractly, so we will use a region K
of dimension n, and consider ω ∈ Λj (K) and η ∈ Λk (K). Then a little thought
shows that
η ∧ ω = (−1)jk ω ∧ η
For example, with ω = dx ∈ Λ1 (R3 ) and η = dy ∧ dz ∈ Λ2 (R3 ) we have

(dy ∧ dz) ∧ dx = dy ∧ dz ∧ dx
= − dy ∧ dx ∧ dz
= dx ∧ dy ∧ dz

and if you look at how this special case works you will see why the general case
works.
Note that nothing prevents us from mixing things up as in

2 dx + 3 dx ∧ dy

but such things do not occur in practise. Forms where each term has the same
number of differentials (forms of the same degree) are called homogeneous, and
we almost always use homogeneous expressions.
1.6. THE OPERATOR D 9

1.6 The Operator d

Our next project is the differential operator d, which we introduce according to
the following four rules, where K is a space with coordinates u1 , . . . , un :
d is the unique operator that satisfies the following laws

1. d is a linear operator

2. On the functions Λ0 (K), df is given by the advanced calculus formula

∂f ∂f
df = du1 + . . . + dun
∂u1 ∂un

3. if ω ∈ Λj (K) and η ∈ Λk (K) then (Leibniz’s Rule)

d(ω ∧ η) = dω ∧ η + (−1)j ω ∧ dη

4. dd = 0 (and thus ddf = 0 and ddui = 0)

We have used coordinates u1 , . . . , un rather than x1 , . . . , xn to emphasize

that these coordinates need not be rectangular or even orthogonal.
Let’s look at some examples in 3-space of how these rules make everything
work. First we examine d(f dx). Since f ∈ λ0 (R3 ) we have, using rule 3,

d(f dx) = df ∧ dx + (−1)0 f ∧ ddx = df ∧ dx + 0

we also used rule 4, ddx = 0, in the last equation. This derives the practical
rule in a special case, and the general case (see problems) will be the same.

d(f dxi1 ∧ . . . ∧ dxik ) = df ∧ dxi1 ∧ . . . ∧ dxik

This is the practical rule for d and the one you will use for almost everything
you do, so learn it well.
Now let ω = f dx and η = g dy. Then we have

d(f dx ∧ g dy) = d(f dx) ∧ g dy + (−1)1 f dx ∧ d(g dy)

= (df ∧ dx) ∧ g dy − f dx ∧ (dg ∧ dy)
= df g ∧ dx ∧ dy + f dg ∧ dx ∧ dy
= (df g + f dg) ∧ dx ∧ dy
= d(f g) dx ∧ dy

just as we expected from the practical rule since f dx ∧ g dy = f g dx ∧ dy.

This also illustrates how rule 2 is a generalization of Leibniz’s formula for the
derivative of a product.
10 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

1.7 Orientation
At this point we must deal with one of the less pleasant aspects of elementary
geometry which is orientation. An orientation, to speak loosely, is a sense of
twist in a space. For example, in R2 our standard sense of twist is counter-
clockwise; we measure angles from th x axis in the direction of the y axis. If we
reverse either axis, we get the opposite twist. If the y axis went down instead
of up then we would measure angles in a clockwise direction and R2 would have
the opposite orientation. If we reverse both axes then the sense of twist returns
to counterclockwise.
In R3 the standard orientation is given by the following rule: if you place
the fingers of your RIGHT hand so that the fingers curl from the x to the y
coordinate axes then your thumb points in the direction of the z axis. This
is called the right hand rule. It has become standard to use this in Calculus
books only since the 1940’s, and the opposite convention is still found in Italian
books, so it is wise to check. To appreciate the subtlety of this concept, think
of trying to communicate it to inhabitants of a planet in another galaxy. Since
nature is almost symmetric in terms of left and right, the only way we know to
clarify this is certain non-symmetric aspects of beta decay. This is referred to
in physics as parity. Hopefully parity properties remain the same from galaxy
to galaxy.
This idea of orientation, though subtle, is strongly coupled with differential
forms and is the reason for the anticommutativity. It also is a prime source of
mistakes, and great care must be taken to keep things in correct order. In R2
(and it’s subspaces) the correct order, which expresses the orientation properly,
is dx ∧ dy. Thus dy ∧ dx is in incorrect order as indicated by the minus sign in

dy ∧ dx = − dx ∧ dy

Recall that when using Green’s theorem

I Z
∂g ∂f
f (x, y) dx + g(x, y) dy = − dxdy
∂A A ∂x ∂y

it is critical that the direction around the boundary of the left integral be coun-
terclockwise. If it is taken clockwise then the two sides of the equation will have
opposite signs. This is again due to the orientation which is built into R2 but
which we seldom notice explicitly. There are similar worries in the use of the
divergence theorem and Stokes theorem.
In applications, the principal place where orientation occurs in Rn is in n-
forms and (n − 1)-forms. We will first tell you the the general formula and
then give you practical methods to make orientation (relatively) easy to deal
with. Let the variables used in Rn be u1 , u2 , . . . , un . (We have switched from
x1 , x2 , . . . , xn to u1 , u2 , . . . , un to emphasize that the variables can be general;
not necessarily rectangular or orthogonal.) The use of superscripts to number
the variables is to conform to tensor analysis standards and we don’t need to go
into the reasons for it here; just do it! And remember u3 is the third variable,
1.7. ORIENTATION 11

not the cube of u. If we choose an order for the variables, which we did by
numbering them, this chooses one of the two orientations. Then

du1 ∧ du2 ∧ . . . ∧ dun is in correct order

du2 ∧ du1 ∧ . . . ∧ dun is in incorrect order

because
du2 ∧ du1 ∧ . . . ∧ dun = − du1 ∧ du2 ∧ . . . ∧ dun
As you can see with a little practice, interchanging any two of the dui reverses
the sign and changes correct to incorrect order or incorrect to correct order. If
you are familiar with permutations, odd permutations of du1 ∧ du2 ∧ . . . ∧ dun
give incorrect order and even permutations give correct order.
That part is easy. The tricky part is the (n − 1)-forms. Here the correct
order is (with dui missing from the list)

(−1)i−1 du1 ∧ . . . ∧ dui−1 ∧ dui+1 ∧ . . . ∧ dun

The reason for this choice is so that

dui ∧ ((−1)i−1 du1 ∧ . . . ∧ dui−1 ∧ dui+1 ) = du1 ∧ . . . ∧ dun

which is correct because the dui must hop over the n−1 elements du1 , . . . , dui−1
in order to get back into correct order and each hop contributes a minus sign.
So much for theory. In R3 correct order is

dx ∧ dy ∧ dz correct order

and for 2-forms we have

dy ∧ dz, − dx ∧ dz, dx ∧ dy correct order

For practical use it is much better to write these in cyclic order.3

dy ∧ dz, dz ∧ dx, dx ∧ dy correct order

and the correct order can be easily remembered by writing

dxdydzdxdydz

and noting that the order of a wedge of two differentials is correct if it occurs
in this list, for example dz ∧ dx is correct but dx ∧ dz is incorrect since dxdz
does not occur in the list. Other incorrects are dy ∧ dx and dz ∧ dy. The use of
differential forms in R3 relies critically on writing things with correct order.
3 Cyclic order is a 3-space concept and does not generalize to n-space at all well.
12 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

1.8 Differential Forms and Vectors

Standard 3-dimensional vector analysis was cobbled together by Josiah Willard
Gibbs in the 1890s using pieces from a variety of sources. While it works well for
many practical purposes it has large deficiencies from a theoretical standpoint.
Those parts of it which concern the dot (scalar) product are fine, but those parts
which concern the cross (vector product v × w) are mathematically clumsy. To
see this, consult any textbook for a proof of the vector triple product law

u × (v × w) = (u · w)v − (u · v)w

It is often said that the cross product cannot be generalized to higher dimensions
but this is not true; what is true is that the analog of the cross product in n
dimensions involves not two but n − 1 vectors. Thus the elementary geometric
applications of the cross product can often be reproduced, but the physical
applications not so much, which is the reason that for relativity (4 dimensions
and space-time rather than just space) we must fall back on Tensor Analysis.
In 3 dimensions there are many formal analogies between differential forms
and vector analysis. That is, differential forms will act like vectors in many ways.
It is a little difficult to find good mathematical reasons for these analogies and
we will not concern ourselves here with what these reasons might be, although
we will return to the matter later. The practical consideration is that things
will work well if we keep in mind the following rules. For 1-forms we have the
correspondences
dx ↔ î dy ↔ ĵ dz ↔ k̂
For 2-forms we have the correspondences

dy ∧ dz ↔ î dz ∧ dx ↔ ĵ dx ∧ dy ↔ k̂

Note that in the second case we have been careful to place the 2-forms in proper
order. This is critical. If you screw this up you will get the wrong sign. Watch
dz ∧ dx term particularly carefully.
With these correspondences kept in mind, we can easily derive many for-
mulas of vector analysis in simple ways. Many things which appear different in
vector analysis can be treated in a unified way with differential forms.

1.9 grad, curl and div

In this section we show how the three vector operators grad, curl and div occur
naturally in the context of differential forms. Since we are in R3 we have 0-forms
(functions), 1-forms, 2-forms and 3-forms. The d operator vanishes on 3-forms
as previously discussed. Hence we look at d on 0-forms, 1-forms and 2-forms.
First 0-forms. Let f be a function (0-form) of the coordinates x, y, z. Then
df is just the old Calculus df :
∂f ∂f ∂f
df = dx + dy + dz
∂x ∂y ∂z
1.9. GRAD, CURL AND DIV 13

and we see here the components of the gradient

 
∂f ∂f ∂f
gradf = , ,
∂x ∂y ∂z
Next we do 1-forms where we omit terms that are 0:

d(f dx + g dy + h dz) = df ∧ dx + dg ∧ dy + dh ∧ dz
 
∂f ∂f ∂f
= dx + dy + dz ∧ dx
∂x ∂y ∂z
 
∂g ∂g ∂g
+ dx + dy + dz ∧ dy
∂x ∂y ∂z
 
∂h ∂h ∂h
+ dx + dy + dz ∧ dz
∂x ∂y ∂z
∂f ∂f
= dy ∧ dx + dz ∧ dx
∂y ∂z
∂g ∂g
+ dx ∧ dy + dz ∧ dy
∂x ∂z
∂h ∂h
+ dx ∧ dz + dy ∧ dz
∂x ∂y
 
∂h ∂g
= − dy ∧ dz
∂y ∂z
 
∂f ∂h
+ − dz ∧ dx
∂z ∂x
 
∂g ∂f
+ − dx ∧ dy
∂x ∂y
Now recall that if v = f i + gj + hk then
 
 i j k 
 ∂
curl v =  ∂x ∂y ∂ ∂ 
∂z 
 f g h 
     
∂h ∂g ∂f ∂h ∂g ∂f
= − i+ − j+ − k
∂y ∂z ∂z ∂x ∂x ∂y
Thus we see we have counterfeited the curl on vectors provided we keep the
2-forms in correct order (so the signs come out right) and we use the correspon-
dence
dy ∧ dz ↔ i dz ∧ dx ↔ j dx ∧ dy ↔ k

Finally, we want to see the differential forms turn up the divergence. Let
once again v = f i + gj + hk and using the above correspondence let us apply d
to the form f dy ∧ dz + g dz ∧ dx + h dx ∧ dy. Then we get

d(f dy ∧ dz + g dz ∧ dx + h dx ∧ dy) = df ∧ dy ∧ dz + dg ∧ dz ∧ dx + dh ∧ dx ∧ dy
14 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

∂f ∂g
= dx ∧ dy ∧ dz + dy ∧ dz ∧ dx
∂x ∂y
∂h
+ dz ∧ dx ∧ dy
∂z
 
∂f ∂g ∂h
= + + dx ∧ dy ∧ dz
∂x ∂y ∂z

and we see div v as the coefficient of dx ∧ dy ∧ dz. Of course dx ∧ dy ∧ dz

corresponds to dV in vector analysis. We will have much more to say about
the relations between these operators and differential forms in the section on
change of coordinates.

1.10 The Poincaré Lemma and it’s Converse

Of the many theorems about differential forms, the three most important are the
converse of the Poincaré lemma, Stokes’ theorem and the Frobenius theorem.
Here we treat the Poincaré lemma; Stokes’s theorem will be treated in the next
section.
The Poincaré lemma was used by Poincaré in his work on the Calculus of
Variations. It is very simple; it says that if ω is a differential form then ddω = 0.
Let us see why this works for functions. We will do it in R3 space but it works
the same in any dimension.
∂f ∂f ∂f
df = dx + dy + dz
∂x ∂y ∂z
 2   2 
∂ f ∂2f ∂ f ∂2f
ddf = − dy ∧ dz + − dz ∧ dx
∂y∂z ∂z∂y ∂z∂x ∂x∂z
 2 
∂ f ∂2f
+ − dx ∧ dy
∂x∂y ∂y∂x
= 0
2 2
∂ f ∂ f
because of the equality of mixed partials: ∂y∂z = ∂z∂y . It is worth noticing that
this is one of the places where we need the functions to have two continuous
derivatives to guarantee this equality.
So much for the easy stuff. The converse of the Poincaré Lemma says that
if ω is a differential form and ω ∈ Λr (K) then there exists a differential form
α ∈ Λr−1 (K) for which dα = ω. Sadly the converse is not true in general. To
be sure it is true, we need to know that K is not too complicated a region.
Specifically we need to know that K is simply connected. This will be a new
concept for most of you. We will try to clarify it with some examples which are
of importance in themselves.
First let us look at the unit disk in R2 which is all the points in R2 whose
distance from the origin is less than or equal to 1. We will call this D0 (1). Let
us draw a curve (say a circle of radius 1/2) then we may ”shrink” the curve in a
continuous manner until is down to just a point. (The technical terminology is
1.10. THE POINCARÉ LEMMA AND IT’S CONVERSE 15

that the curve is homotopic to a point.) A simple closed curve is a continuous

curve with no end points and no intersections with itself. It should be intuitively
clear that any simple closed curve in D0 (1) can be shrunk to a point.
For contrast, consider the annulus consisting of points in R2 that have dis-
tance from the origin less than or equal to 1 and greater than or equal to 1/2.
For this region K some simple closed curves can be shrunk to points and others
(those that go round the central ”hole”) cannot be shrunk to points. As a third
example consider the unit disk with just the center removed. Then, exactly like
the annulus where a whole disk was removed, a curve round the origin cannot
be shrunk to a point, because the point we need, the origin, is not in K.

Def A region K is simply connected if and only if any simple closed curve in K
can be shrunk to a point in K.
A few more examples, this time in R3 . A ball (solid sphere) is simply con-
nected, as is also a ball with the center removed. However, a ball around the
origin with the z axis removed is not simply connected and neither is the whole
of R3 with the z axis removed. The unit sphere (points in R3 at unit distance
from origin) is simply connected. (Note the contrast to the unit circle in R2
which is not simply connected.
These kinds of considerations are called topological, and there is a wonderful
branch of mathematics which studies such things called topology. Due to the
enormous amount of material needed in an engineering curriculum, it is not
customary for topology to be included, but if you wish to learn more we recom-
mend the books [2] and [4], which are specifically written with the engineer and
physicist in mind.
Now that we know some topology, we can state the theorem:
Theorem (Converse of Poincaré Lemma) Let K be a simply connected region
and ω ∈ Λr (K) and dω = 0. Then there is an α ∈ Λr−1 (K) such that dα = ω.
It should be mentioned that the α is highly non-unique; there are many α’s
that will work.
In the problems we will see examples where the region is not simply con-
nected and, though dω = 0 there is no α for which dα = ω.√ Practically speaking,
this usually takes the form of α being multi-valued, like x, and thus not being
a proper function. In this case, we can often manage to get some use out of the
multi-valued function provided we restrict our attention to a simply connected
subregion. The vector potential of the magnetic field of an infinite vertical
current carrying wire is the paradigm example and we will look at it in the
problems.
Next we will derive two popular theorems of vector analysis (in R3 ) from
the converse of the Poincaré lemma.
For the first we again let v = f i + gj + hk and we suppose that curl v = 0.
We then form the corresponding differential 1-form ω = f dx+g dy +h dz. From
the results of the last section, we see that the condition curl v = 0 translates
into the condition dω = 0. Assuming the region K ⊆ R3 is simply connected
16 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

we may conclude that there is a 0-form (function) k for which dk = ω, that is

∂k ∂k ∂k
dk = dx + dy + dz = f dx + g dy + h dz
∂x ∂y ∂z

Translating back into vector notation we have

∂k ∂k ∂k
i+ j+ k = f i + gj + hk
∂x ∂y ∂z
grad f = v

We have thus proved the vector analysis theorem

Corollary Over a simply connected region K ⊆ R3 a curl free vector is a
gradient.
There is a very similar theorem which you will prove in the problems. It
reads
Corollary Over a simply connected region K ⊆ R3 a divergence free vector is
a curl.
More explicitly, if div v = 0 then there is a vector w so that curl w = v.
Although formulas for finding the objects whose existence is asserted exist,
they are of no great practical utility and we will not discuss them.
Other formulas which are easily derived at this point and which you will
derive in the problems are

curl grad k = 0
div curl v = 0

These are both consequences of dd = 0 which you probably already figured out.
Also available are results about path independent line integrals which we will
look at in the problems.

1.11 Boundaries
Before we can deal with Stokes theorem we must talk a bit about boundaries
and their orientation, and also how to compute integrals of differential forms.
The boundary of a area or surface is simply its edge. We will be satisfied
with an intuitive understanding of this and not go into the topological details.
The symbol of the boundary of a region or surface K is ∂K. For example the
boundary of the unit disk in R2 is the unit circle, and the boundary of the upper
half of the unit sphere in R3 is the unit circle in the x, y-plane. The boundary of
the unit ball in R3 is the unit sphere. It is common for there to be no boundary;
the boundary of the unit sphere in R3 is empty, which we write ∂S(0, 1) = ∅
where ∅ is the empty set.
For Stokes theorem it is important that the boundary ∂K be oriented cor-
rectly relative to K itself. We will do this with examples, but the methods are
1.11. BOUNDARIES 17

selected so that they work in any dimension. If K is n-dimensional in n-space,

like A or M then it inherits an orientation from the n-space, that determined
by the order x, y for A or x, y, z for M and analogously for higher dimensions.
To orient the boundary of such regions, which necessarily is described by n − 1
parameters, one uses the exterior normal vector n̂ to K so that n̂ followed by the
parameters has the same orientation as K. (We will clarify this with examples.)
This gives an order to the parameters which is then the order used below in the
integrals. There is no way to get around the fact that this is a bit tricky.
For out first example let A be the unit disk in R2 , with boundary the unit
circle and parameter t for the unit circle. Let r(t) trace out the unit circle.
Then
dr
dt 
T̂ =  dr
 
dt

is the unit tangent vector. There are two directions one may go along a curve,
and the parameterization must be chosen so that n̂, T̂ has the same orientation
as x, y. This amounts to, first, the angle from n̂ to T̂ is a positive right angle,
and second that as t increases we trace the boundary counterclockwise around
A. So this comes down to just going around A counterclockwise.
For M the situation is more difficult to see since the boundary will be two
dimensional. Suppose the parameters are u and v. Let the surface be given
by r(u, v). Then when we increase u and hold v fixed, a curve will be traced
∂r
out in certain direction and it will have tangent vector ru = ∂u pointing in
that direction. Similarly for rv . Then the requirement is that n̂, ru , rv have the
same orientation as x, y, z. If the orientation is opposite, reverse the order of
the parameters u and v.
On a practical note, one usually finds the n̂ for ∂M by forming
ru × rv
n̂ =
|ru × rv |

so that one really has only to check whether this n̂ is pointing in or out of M
and if it is pointing in change the order of u and v. Also, one does not actually
have to calculate ru × rv , only figure out its direction.
As an example, take the upper half of the unit sphere and let us use for
coordinates longitude φ and colatitude θ. A tangent vector to a phi curve goes
to the fight (looking at the sphere from outside) and a tangent vector to the
θ curve goes down. Crossing these gives a vector pointing in so we have the
wrong order: the order should be θ, φ not φ, θ. Any 2-forms that occur should
thus be presented as f (θ, φ) dθ ∧ dφ. This will be critically important when we
form integrals.
Also as a practical matter, for ∂M ⊂ R3 one often uses two of x, y, z as
parameters. If this is the case, the proper orders are dy ∧ dz, dz ∧ dx, dx ∧ dy
as one can determine by using the above rules.
The last case we are going to discuss is S, the surface in R3 . If the surface has
no boundary then the situation is just that of ∂M above (practically speaking)
and we have already covered the situation; an example would be the unit sphere.
18 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

However, if the surface has a boundary, then there is no obvious way to decide
which way the normal should point. Mathematically one just has to choose
oneself either a direction for the normal n̂ or, equivalently, and order for the
coordinates u, v. Physically there may be an obvious way to make the choice.
Make sure that your n̂ is consistent with your choice of u, v so that n̂, ru , rv is
oriented like x, y, z. Once this is settled it only remains to orient the boundary
∂S. Since ∂S is one dimensional, it is only a question of choosing which way to
go around it. Here is the rule.
Walk around the boundary of the surface S with your body pointing the same
way as n̂ and your LEFT hand pointing toward the surface. The direction your
are walking is the correct orientation of the boundary.
This is the orientation you must use for the correct application of Stokes
theorem. If you mess it up the sign will be wrong.
Of course in physics the sign often comes out wrong, and one just reverses
it at the end. Still, it is reassuring when it comes out right without human
intervention.

1.12 Integrals of Forms

This is extremely simple. First, the integrals of forms must be taken over regions
of the same dimension as the degree of the form. Thus

a 1-form must be integrated over a curve C

a 2-form must be integrated over a surface S or A
a 3-form must be integrated over a region M

To calculate the integral we recall that in a multiple integral the order of the
differentials doesn’t matter but for forms the order matters, as the sign will
change if two differentials are interchanged. The calculation of an integral is
reduced to three steps.
1. Rearrange the integral so that the differentials are in correct order.
2. Remove the wedges to get an ordinary multiple integral
3. Calculate the multiple integral by the appropriate iterated integral
This should become clear from an example. We wish to calculate the integral
Z
x dy ∧ dx
A
2
where A is the unit disk in R . Step 1 is to realize that the differentials are not
in the right order. We fix.
Z Z
x dy ∧ dx = − x dx ∧ dy
A A
1.13. VARIABLE CHANGES 19

Now the differentials under the integral are in the correct order. Step 2 is to
throw out the wedges
Z Z
− x dx ∧ dy = − x dxdy
A A

where the last integral is an ordinary double integral from your advanced cal-
culus course. Step 3 is then to go over to the an iterated integral
Z Z 1 Z √
1−x2
− x dxdy = − √ x dxdy
A −1 − 1−x2

Step 4, which we did not mention above, is then to plug it into your fancy cal-
culator and get the answer. Another method is to calculate the double integral
by switching to polar coordinates
Z Z 2π Z 1
− x dxdy = − (r cos θ)r drdθ
A 0 0

from which we see immediately that the result is 0.

This pretty much is the whole story. With a modest amount of care this will
come out right.

1.13 Variable Changes

We begin this section with the following observations. Suppose we are calculat-
ing Z
f (x, y) dx ∧ dy
A
Now suppose we find it convenient to change to polar coordinates or some other
coordinates u, v. Then the old x, y coordinates are functions of the new u, v
coordinates.

x = x(u, v)
y = y(u, v)

Differential Forms allow us to convert the integral immediately. We have

f (x, y) = f (x(u, v), y(u, v)) = f˜(u, v)

   
∂x ∂x ∂y ∂y
dx ∧ dy = du + dv ∧ du + dv
∂u ∂v ∂u ∂v
 
∂x ∂y ∂x ∂y
= − du ∧ dv
∂u ∂v ∂v ∂u
Thus Z Z  
∂x ∂y ∂x ∂y
f (x, y) dx ∧ dy = f˜(u, v) − du ∧ dv
A A ∂u ∂v ∂v ∂u
20 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

Naturally there are generalizations to higher dimensions. Before we deal with

that, though, we should systematize what we have here. The cleverer students,
or those with really good memories, will recognize that expression in brackets
as the determinant of the matrix
 ∂x ∂x 
∂(x, y) ∂u ∂v
= ∂y ∂y
∂(u, v) ∂u ∂v

which is called the Jacobian matrix4 of the variable change. Thus we can write
the above equation, with det(·) signifying the determinant, as
Z Z
∂(x, y)
f (x, y) dx ∧ dy = f˜(u, v) det( )du ∧ dv
A A ∂(u, v)
Z  ∂x ∂x 
 
= f˜(u, v)  ∂u
∂y
∂v 
∂y  du ∧ dv
A ∂u ∂v

Naturally the description of A will be quite different in the x, y and u, v variables.

If we now go over to unoriented integrals, that is ordinary double integrals, we
will have, assuming that x, y and u, v give the same orientation to A,
Z Z
∂(x, y)
f (x, y) dxdy = f˜(u, v) det( )dudv
A A ∂(u, v)
Since x, y and u, v give the same orientation to A, the determinant in the integral
will have a positive value. If the orientations are reversed, it will all be taken
care of in the oriented integrals but in the ordinary integrals we need to make
the determinant positive by putting in an absolute value, so we have
Z Z
∂(x, y)
f (x, y) dxdy = f˜(u, v) |det( )| dudv
A A ∂(u, v)
This is the famous change of variable formula. In most Calculus books they do
not prove it because the proof without differential forms is somewhat difficult.
Using differential forms it is quite easy, as you have seen. The reason for this
is that differential forms keep control of the orientation, whereas old techniques
have to deal with it in an ad hoc and confusing way.
Everything we have done here in 2 dimensions will work perfectly well in
n-dimensions. Enthusiasts may wish to work it out for 3 dimensions. You will
then see the connection between differential forms and determinants. In fact,we
can assert, with certainty of contradiction from some member of the audience,
that determinants are important because they are the coefficients in Grassmann
Algebra calculations. We will see more of this.

1.14 Surface integrals

It is convenient to work out the connection between surface integrals in vector
form and differential forms at this point, so we have it available for Stokes
4 properly pronounced Yacobian, although we know it’s hopeless
1.14. SURFACE INTEGRALS 21

theorem. We will work it out for surfaces in R3 but the techniques applied will
work just as well for (n − 1)-surfaces in Rn . This is one of the places in the
theory which, while not difficult, is a little tricky.
While it may not be possible to cover the surface with a single coordinate
patch, it is always possible to break up the integral into integrals over each
patch and add them. Hence we only need to work out the formula for a single
coordinate patch.
The standard thing to have to work out for surface integrals is
Z
P dy ∧ dz + Q dz ∧ dx + R dy ∧ dz
S

If we let v = P î + Q ĵ + R k̂ then the above is a disguised form of our old friend

from advanced calculus Z Z
v · n̂ dS = v · dS
S S
First we need to get some idea of what dS should be. We will be using the
variables u, v as coordinates on the surface. If a normal is already available be
sure to pick the order of the variables so that n̂, du, dv has the same orientation
as x, y, z. The surface will then be described by r(u, v). If we think of moving
from a point p in the u and v directions we will have vectors
∂r
r(u + ∆u, v) − r(u, v) ≈ ∆u
∂u
∂r
r(u, v + ∆v) − r(u, v) ≈ ∆v
∂v

and so the bit of oriented area can be obtained from the cross product
 
∂r ∂r
∆S = × ∆u∆v
∂u ∂v
Now remembering that
r(u, v) = x(u, v) î + y(u, v) ĵ + z(u, v) k̂
we have
∂r ∂x ∂y ∂z
= î + ĵ + k̂
∂u ∂u ∂u ∂u
∂r ∂x ∂y ∂z
= î + ĵ + k̂
∂v ∂v
 ∂v ∂v   
∂r ∂r ∂y ∂z ∂z ∂y ∂z ∂x ∂x ∂z
× = − î + − ĵ
∂u ∂v ∂u ∂v ∂u ∂v ∂u ∂v ∂u ∂v
 
∂x ∂y ∂y ∂x
+ − k̂
∂u ∂v ∂u ∂v
∂(y, z) ∂(z, x) ∂(x, y)
= î + ĵ + k̂
∂(u, v) ∂(u, v) ∂(u, v)
22 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

Going over from ∆ to d we have at last

 
∂r ∂r
dS = × dudv
∂u ∂v
 
∂(y, z) ∂(z, x) ∂(x, y)
= î + ĵ + k̂ dudv
∂(u, v) ∂(u, v) ∂(u, v)

Now we also have

∂r ∂r
∂u × ∂v
n̂ = ∂r ∂r
| ∂u × ∂v |

Finally we have
Z
P dy ∧ dz + P dz ∧ dx + P dx ∧ dy
S
Z
∂(y, z) ∂(z, x) ∂(x, y)
= P du ∧ dv + Q du ∧ dv + R du ∧ dv
S ∂(u, v) ∂(u, v) ∂(u, v)
Z  
∂(y, z) ∂(z, x) ∂(x, y)
= (P i + Q j + R k) · î + ĵ + k̂ du ∧ dv
S ∂(u, v) ∂(u, v) ∂(u, v)
Z  
∂(y, z) ∂(z, x) ∂(x, y)
= (P i + Q j + R k) · î + ĵ + k̂ dudv
∂(u, v) ∂(u, v) ∂(u, v)
ZS
= v · dS
S

In the next to the last step we switched from an oriented to an unoriented

integral since everything was carefully set up to be correctly oriented.
We can sweat a little more out of this. The scalar element of area is of course
the absolute value of dS so that
s 2  2  2
∂(y, z) ∂(z, x) ∂(x, y)
dS = |dS| = + + dudv
∂(u, v) ∂(u, v) ∂(u, v)

This is itself a very important formula. Also since

∂r ∂r
∂u × ∂v
n̂ = ∂r ∂r
| ∂u × ∂v |

we have  
∂r ∂r  ∂r ∂r 
dS = × dudv = n̂  ×
∂u ∂v  ∂u ∂v  dudv = n̂ dS

we can write Z Z
v · dS = v · n̂ dS
S S

It would be wise to note that the integrals we have been working with in
this section are often called flux integrals. In contrast, there is another kind of
1.15. THE GENERALIZED STOKES THEOREM 23

surface integral which has little to do with what we are doing. This kind of
integral is not vectorial in nature. They look like
Z
f (x, y, z) dS
S

A problem that utilized this kind of integral would be finding the center of mass
of a hemispherical shell. We have developed the necessary tools to evaluate
these integrals, although for us this is a side issue and we only mention it since
students should be aware of the two distinct kinds of surface integral.
To evaluate this kind of integral, choose parameter u, v for the surface, (or
a portion of the surface,) and then use the above above formula for dS to get
Z
f (x, y, z) dS
S
s 2  2  2
Z
∂(y, z) ∂(z, x) ∂(x, y)
= f (x(u, v), y(u, v), z(u, v)) + + dudv
∂(u, v) ∂(u, v) ∂(u, v)
These kinds of integrals are very sensitive to the choice of parameters, and an
integral that might be extremely difficult with parameters x, y (so that z =
z(x, y)) might be quite easy with, say, θ, φ as parameters.

1.15 The Generalized Stokes Theorem

The generalized Stokes theorem is one of the triumphs of elementary mathemat-
ics, subsuming as it does the fundamental theorem of Calculus, Green’s theorem,
the divergence theorem (= Gauss’s theorem), and the ordinary Stokes’s theo-
rem. Moreover, it is dimension independent; it works as well in 26 dimensions
as in 2 or 3.
Theorem (Generalized Stokes Theorem)
Let K be a oriented subset of Rd of dimension n and let ∂K be the properly
oriented boundary of K. Let ω be an (n-1)-form. Then
Z I
dω = ω
K ∂K

One of the important qualities of this theorem is that it is very easy to

remember. Hence it can be used to write down the Green, divergence and
Stokes’ theorems even if you don’t precisely remember where the signs go; it
figures that out for you. We will now look at the Green, divergence and Stokes’
theorems one at a time and see how they fall out of the generalized Stokes
theorem. At the end, for fun, we will come back and look at the fundamental
theorem of Calculus, which is tricky because it is too simple.
Green’s theorem Let A ⊂ R2 be a region with boundary ∂A oriented coun-
terclockwise. Then
I Z
∂Q ∂P
P (x, y) dx + Q(x, y) dy = − dxdy
∂A K ∂x ∂y
24 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

Proof Let ω = P (x, y) dx + Q(x, y) dy. Applying the general Stokes’ theorem
we have
I Z
P (x, y) dx + Q(x, y) dy = ω
∂A
I∂A
= dω
ZK
= d(P (x, y) dx + Q(x, y) dy)
K
Z  
∂P ∂Q
= dy ∧ dx + dx ∧ dy
K ∂y ∂x
Z  
∂Q ∂P
= − dx ∧ dy
K ∂x ∂y
Z  
∂Q ∂P
= − dxdy
K ∂x ∂y

where we used dy ∧ dx = − dx ∧ dy and in the last line we used the rule from
converting from oriented integrals to ordinary double integrals.
Notice that it was not necessary for us to remember where the sign goes in
Green’s theorem; our methodology automatically puts it in the correct place.
We will do Stokes’ theorem next since it looks so similar in many ways to
Green’s theorem.
Stokes’ Theorem Let S be a surface in R3 with definite choice of normal n̂
and correctly oriented boundary ∂S. Let v = P î + Q ĵ + R k̂ Then
I
P dx + Q dy + R dz =
∂S
Z   Z   Z  
∂R ∂Q ∂P ∂R ∂Q ∂P
= − dydz + − dzdx + − dxdy
S ∂y ∂z S ∂z ∂x S ∂x ∂y
or in vector notation I Z
v · dl = curl v · n̂ dS
∂S S

Recall that the boundary of S is oriented so that when you walk around the
boundary with your body in the direction of n̂ and your left hand reaching
in toward the surface then you are walking in the direction of the orientation.
(This can be reformulated in terms of fingers and thumb but we find this less
confusing.)
Proof: Let ω be the 1-form
ω = P dx + Q dy + R dz
Then applying Stokes’ theorem
I
P dx + Q dy + R dz =
∂S
1.15. THE GENERALIZED STOKES THEOREM 25
I
= ω
∂S
Z
= dω
ZS
= d(P dx + Q dy + R dz)
ZS   Z   Z  
∂R ∂Q ∂P ∂R ∂Q ∂P
= − dy ∧ dz + − dz ∧ dx + − dx ∧ dy
S ∂y ∂z S ∂z ∂x S ∂x ∂y
Z   Z   Z  
∂R ∂Q ∂P ∂R ∂Q ∂P
= − dydz + − dzdx + − dxdy
∂y ∂z ∂z ∂x ∂x ∂y
ZS S S

= curl v · dS
S

where we have used the rules governing the transition from oriented to unori-
ented (ordinary) integrals. Recall that in unoriented integrals the order of the
differentials does not matter, but we have left them in cyclic form as a matter
of good practise. The transition to vector form was discussed in the previous
section.
Now to the divergence theorem. This is quite easy.
The Divergence theorem (Gauss’s theorem) Let M be a region of R3 and
∂M the surface which is its correctly oriented boundary. Then
I
P dydz + Qdzdx + P dxdy =
∂M
Z  
∂P ∂Q ∂R
= + + dxdydz
M ∂x ∂y ∂z

or in vector notation
I Z
v · n̂ dS = div vdxdydz
∂M M

Proof Let ω be the 2-form

ω = P dy ∧ dz
Then applying the generalized Stokes’ theorem
I I
P dxdy = P dx ∧ dy
∂M
I∂M
= ω
Z∂M
= dω
M
Z
= d(P dy ∧ dz)
M
26 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS
Z
∂P
= dx ∧ dy ∧ dz
dx
ZM
∂P
= dxdydz
M dx

The other two terms are handled similarly. Notice that we have proved slightly
more than the theorem states; we have proved the analog of the theorem for
each term. This can occasionally be useful.

1.16 Curvilinear Coordinates I: preliminary for-

mulas
In this section we will derive formulas for the gradient, curl, divergence and
Laplacian in curvilinear coordinates. This is not nearly as straightforward as
one might expect, but you will not see these troubles because of good organiza-
tion. The central trick in this development is to express each of the differential
operators in terms of operators that are manifestly coordinate independent.
Each of these operators has a coordinate formulation and when they are put
together the desired expression arises.
Unfortunately the derivation of some of the basic formulas is a little more
conceptual and difficult than most of the material in these notes. To spare
those who are interested only in the results, we have put these derivations into
Chapter 2 (Mathematical Theory). There is also a lower level introduction to
the ∗ operator in the section after the Curvilinear coordinate sections.
To express grad, divergence and curl invariantly we need two functions,
Φ : V → V ∗ and ∗ : Λr (V ) → λn−r (V ). Both of these functions are independent
of the choice of the coordinate system, as can be seen in Chapter 2. Although we
are just using these formulas to derive the formulas for curvilinear coordinates,
they have much wider application in modern mathematics, and effort put into
learning them may well pay off in other circumstances. And you will be glad to
know that in our circumstances there is not much difficulty.
We will write the formulas we need in n-space because it is no more difficult
than writing them in 3-space, and indeed some things are actually clearer. Also
it is important to know which things work generally and which things, like curl,
work only in three space.
Let u1 , u2 , . . . , un be curvilinear coordinates. Then we have a position vector
r and its derivatives which are more or less the base vectors. They require
normalization later, but it would be a big mistake to normalize now.

r = r(u1 , . . . , un )
∂r
ei =
∂ui
gij = (ei , ej ) = ei · ej
(g ij ) = (gij )−1
1.16. CURVILINEAR COORDINATES I: PRELIMINARY FORMULAS 27

We are going to mostly use the parentheses notation (ei , ej ) instead of ei · ej

because it keeps things grouped together better. It is just notation. The gij are
called the metric coefficients and allow one to work with distance an angles in
the coordinates u1 , u2 , . . . , un . We think of vectors as being based at particular
points which are determined by r. This is an elementary form of the concept
tangent space which we are not going to use.
Any vector v based at r(u1 , . . . , un ) can be expressed in terms of the basis
vectors e1 , . . . , en which are also thought of as being based at r. Thus
n
X
v= v i ei = v i ei (sum sign ommited)
i=1

The sum sign is often omitted when the same letter appears both up and down;
since invented by Einstein it is probably a good idea, but you do need to keep
it in mind. This simplifies the notation considerably.
Next we need a function where you input the vector v and it outputs the ith
coordinate v i . You have known this function for years, but never knew what it
was. It is dui . Thus
dui (v) = v i
This is indeed what dui really is, but with the current organization of the math
curriculum it is not convenient to explain this at the entry level. However, now
you know! Incidentally, it is possible to make this consistent with the idea that
dui is a little bit of ui . It works like this. Nearby points are connected by
vectors which are very short. If v is such a short vector, then v i = dui (v) is
also small. So if we abbreviate dui (v) by dui it will behave as a small number,
sort of. It’s surprising that this works as well as it does.
Since dui : V → R (and is linear) it is a linear functional and thus dui ∈ V ∗ ,
where V ∗ is the vector space of linear functionals on V. A wedge product of
three dui , for example du1 ∧ du3 ∧ du5 would then be in Λ3 (V ∗ ).
The du1 , . . . , dun are the dual basis to e1 , . . . , en and form a basis for V ∗ .
Now we discuss Φ : V → V ∗ and Φ−1 : V ∗ → V . We have

Φ(v) = Φ(v i ei ) = gij v i duj

Φ−1 (ω) = Φ−1 (λi dui ) = g ij ωi ej

Here ω is an arbitrary element of V ∗ and the λi are its coordinates in the basis
du1 , . . . , dun .
Recall that the gij give an inner (dot) product for V and in Chapter two we
show that this may be ”lifted” to V ∗ . If

λ = λi dui λ∈ V∗ λi ∈ R
µ = µi dui µ∈V∗ µi ∈ R

then
(λ, µ) = g ij λi µj
28 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

just like
(u, v) = gij ui v j

but using the inverse matrix.

Now we digest the information on the ∗ operator, which is derived in subse-
quent section and in Chapter 2.
X √
∗dui = (−1)k−1 g ik g du1 ∧ . . . ∧ duk−1 ∧ duk+1 ∧ . . . ∧ dun
k

∗∗ω = (−1)r(n−r) ω for ω ∈ Λr (V ∗ )

√
Ω0 = g du1 ∧ . . . ∧ dun
∗Ω0 = 1
∗1 = Ω0

If the coordinates are orthogonal then things simplify. Setting

 2
hi for i = j
(ei , ej ) = gij = h2i δij =
0 for i 6= j
 1
1 h2i
for i = j
(dui , duj ) = g ij = δij =
h2i 0 for i 6= j
√
g = h1 h2 . . . hn

We know that a basis for Λr (V ∗ ) is given by

dui1 ∧ dui2 ∧ . . . ∧ duir where i1 < i2 < · · · < ir

Let {k1 , . . . , kn−r } = {1, 2, . . . , n} − {i1, . . . , ir } so that we can form the permu-
tation   
1 . . . r  r + 1 . . . n
i1 . . . ir  k1 . . . kn−r

of {1, 2, . . . , n}. The sign of this permutation is (−1)s where s is the number
of interchanges necessary to rearrange the bottom line into the top line (or vice
versa). The vertical line is just for the convenience of the reader; it divides the
first part of the permutation from the second. Then

hk1 · · · hkn−r k1
∗(dui1 ∧ dui2 ∧ . . . ∧ duir ) = (−1)s du ∧ duk2 ∧ . . . ∧ dukn−r
hi1 · · · hir

as we show in the Chapter 2.

Now it is time for two examples. The results will just be listed but it is
easy to verify they are correct; mostly they can be done in ones head. We will
now begin to omit the wedges between differentials; the reader may assume that
if differentials are being multiplied that wedges have been omitted for ease of
reading.
1.16. CURVILINEAR COORDINATES I: PRELIMINARY FORMULAS 29

Example 1. n = 3 and coordinates are x, y, z.

r = (x, y, z)
∂r
e1 = = (1, 0, 0)
∂x
∂r
e2 = = (0, 1, 0)
∂y
∂r
e3 = = (0, 0, 1)
∂z 
1 if i = j
gij = ei · ej = (ei , ej ) = δij =
0 6 j
if i =
Thus we have orthogonal coordinates and also h2i = gii = 1. Thus our formulas
give
∗dx = dy ∧ dz ∗dy = (−1)dx ∧ dz ∗dz = dx ∧ dy
∗dy ∧ dz = dx ∗dx ∧ dz = (−1)dy ∗dx ∧ dy = dz
Naturally in practise we prefer dz ∧ dx (cyclic order) to (−1)dx ∧ dz.
Example 2. n = 3 and coordinates are ρ, θ, φ.
Here θ (physics standard) is the angle off the z-axis.
r = (ρ sin θ cos φ, ρ sin θ sin φ, ρ cos θ)
∂r
e1 = = (sin θ cos φ, sin θ sin φ, cos θ) h21 = g11 = 1
∂ρ
∂r
e2 = = (ρ cos θ cos φ, ρ cos θ sin φ, −ρ sin θ) h22 = g22 = ρ2
∂θ
∂r
e3 = = (−ρ sin θ sin φ, ρ sin θ cos φ, 0) h23 = g33 = ρ2 sin2 θ
∂φ

g = det(gij = h21 h22 h23

√
g = h1 h2 h3 = ρ2 sin θ
√
Ω0 = gdρdθdφ = ρ2 sin θ dρdθdφ
h2 h3
∗dρ = dθdφ = ρ2 sin θ dθdφ
h1
h1 h3 h1 h3
∗dθ = − dρdφ = dφdρ = sin θ dφdρ
h2 h2
h1 h2 1
∗dφ = dρdθ = dρdθ
h3 sin θ
1
∗(dθdφ) = dρ
ρ2 sin θ
1
∗(dφdρ) = dθ
sin θ
∗(dρdθ) = sin θ dφ
For the last three equations use ∗ ∗ ω = (−1)r(n−r) ω
30 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

1.17 Curvilinear Coordinates II: the calculations

Now we must consider the strategy of how we will find the formulas for grad, div
and curl in curvilinear coordinates. This is quite simple once you know how.
We look at the formulas in x, y, z coordinates and figure how to express the
x, y, z expressions in terms of the operators Φ and ∗. Since these are coordinate
independent, we can use the formulas we find in any coordinate system. The
rest is just relatively easy calculations.
We will start with curl. Recall that the input and output of curl are vectors
and that if we set ω = v 1 dx + v 2 dy + v 3 dz then
 3   1   2 
∂v ∂v 2 ∂v ∂v 3 ∂v ∂v 1
dω = − dydz + − dzdx + − dxdy
∂y ∂z ∂z ∂x ∂x ∂y

which looks a lot like curl. Now if we are starting with a vector v = v 1 e1 +
v 2 e2 + v 3 e3 we can get to ω by using Φ. Indeed, using the formulas from the
last section,

Φ(v) = gij v i duj

= δij v i duj
= v 1 dx + v 2 dy + v 3 dz
= ω

Then dω gives us the above 2-form. However, we cannot use Φ−1 to descend
again to vectors because Φ−1 eats only 1-forms. However ∗ gets us from 2-forms
to 1-forms, and we have
 3   1   2 
∂v ∂v 2 ∂v ∂v 3 ∂v ∂v 1
∗dω = − dx + − dy + − dz
∂y ∂z ∂z ∂x ∂x ∂y
 3   1   2 
∂v ∂v 2 ∂v ∂v 3 ∂v ∂v 1
Φ−1 (∗dω) = − ~e1 + − ~e2 + − ~e3
∂y ∂z ∂z ∂x ∂x ∂y
Φ−1 (∗dΦ(v)) = curl v

Thus we see, using ◦ for function composition,

curl = Φ−1 ◦ ∗ ◦ d ◦ Φ

Since Φ, ∗, and d are all independent of the choice of coordinate system, so is

curl= Φ−1 ◦ ∗ ◦ d ◦ Φ. Thus we can use this formula for curl in any coordinate
system; we just have to be able to calculate the functions Φ, ∗, and d in the
new coordinates and this is trivial. We will now use our method to get formulas
for curl in any orthogonal coordinate system. We could do it in an arbitrary
coordinate system also, but this would be beyond the needs of most engineers
and scientists, and would also come out messy.
We will find the formula for curl in the orthogonal coordinate system u1 , u2 , u3 .
Although we will eventually have to change over to physical coordinates, we will
1.17. CURVILINEAR COORDINATES II: THE CALCULATIONS 31

begin with v expressed in the natural coordinates system

∂r
ei =
∂ui
v = v 1~e1 + v 2~e2 + v 3~e3

Since the coordinate system is orthogonal, we have gij = 0 for i 6= j and gii = h2i .
Thus we have

v = v 1~e1 + v 2~e2 + v 3~e3

Φ(v) = gij v i duj = h21 v 1 du1 + h22 v 2 du2 + h23 v 3 du3
 
∂(h23 v 3 ) ∂(h22 v 2 )
dΦ(v) = − du2 du3 + etc.
∂u2 ∂u3
 
h1 ∂(h23 v 3 ) ∂(h22 v 2 )
∗dΦ(v) = − du1 + etc.
h2 h3 ∂u2 ∂u3
 
1 h1 ∂(h23 v 3 ) ∂(h22 v 2 )
curl v = Φ−1 ∗ dΦ(v) = − e1 + etc.
h21 h2 h3 ∂u2 ∂u3
 
1 ∂(h23 v 3 ) ∂(h22 v 2 )
= − e1 + etc.
h1 h2 h3 ∂u2 ∂u3

This is correct but it is in terms of the natural basis e1 , e2 , e3 whereas in physics

it is usual to express the vector in terms of the normalized vectors
1 1 1
ê1 = e1 , ê2 = e2 , ê3 = e3
h1 h2 h3
Then we have

v = v 1 e1 + v 1 e2 + v 1 e3
= v 1 h1 ê1 + v 2 h2 ê2 + v 3 h3 ê3
= ṽ 1 ê1 + ṽ 2 ê2 + ṽ 3 ê3

where
ṽ1 = h1 v 1 ṽ2 = h2 v 2 ṽ3 = h3 v 3
In terms of the ṽ i and the êi the formula for curl becomes
 
1 h ∂(h23 v 3 ) ∂(h22 v 2 )
curl v = − h1 ê1 + etc.
h1 h2 h3 ∂u2 ∂u3
 
1 h ∂(h3 ṽ 3 ) ∂(h2 ṽ 2 )
= − h1 ê1 + etc.
h1 h2 h3 ∂u2 ∂u3
 
 h1 ê1 h2 ê2 h3 ê3 
1  ∂ 
=  ∂ ∂ 
h1 h2 h3  ∂u 1 ∂u 2 ∂u 3 
1 1 1

h1 ṽ h1 ṽ h1 ṽ

This is the formula you find in reference books.

32 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

As an example, let us write the formula for Spherical Coordinates using the
material for Spherical coordinates at the end of the previous section. Recall
that h1 = 1, h2 = ρ, h3 = ρ sin θ. Then
 
 ê1 ρ ê2 ρ sin θ ê3 
1  ∂ 
curl v =  ∂ ∂ 
ρ2 sin θ  ∂ρ1 ∂θ
2
∂φ 
3 
ṽ ρṽ ρ sin θṽ

While the curl will only work in 3 dimensions, we can get formulas for grad
and div in n dimensions with no extra effort at all. This is important since
we might need either of them in 2 as well as 3 dimensions, and maybe even 4
dimensions for relativity. So we shall do them all at once.
There is no difficulty at all with grad. It inputs a function and outputs a
vector so the immediate candidate for grad is Φ−1 ◦ d. This obviously works in
rectangular coordinates since then Φ(ei ) = dui . and thus
∂f
grad f = Φ−1 ( dui ) (sum on i understood)
∂ui
Xn
∂f
= e
i i
i=1
∂u
The last is our old friend the advanced calculus gradient. Notice that we could
not use the summation convention here since both the i’s in the last term count
as low indices. (Recall that high index in a denominator counts as low.) Watch
carefully in the next developments to see why this has happened.
Next recall that for general (not necessarily orthogonal) coordinates we have
Φ(ei ) = gij duj Φ−1 (dui ) = g ij ej
and so
grad f Φ−1 (df )
=
∂f
= Φ−1 ( i dui )
∂u
∂f ij
= g ej
∂ui
which is nice and simple. For orthogonal coordinates we have

ij 0 if i 6= j
g = 1
= 1 if i = j
gii h2i

and thus, and thus with natural basis vectors ei and physical (normalized) basis
vectors êi = h1i ei we have
Xn
1 ∂f
grad f = 2 ∂ui ei
h
i=1 i
Xn
1 ∂f
= êi
h
i=1 i
∂ui
1.17. CURVILINEAR COORDINATES II: THE CALCULATIONS 33

The second formula, with physical basis vectors êi , is what one generally sees
in reference books.
Our next task is the divergence. This is harder, since we must use the ∗
operator, but we can break it down into easy steps. The first observation we
make is important for many purposes. Let
n
X
ω= (−1)i−1 v i du1 ∧ · · · ∧ dui−1 ∧ dui+1 ∧ · · · ∧ dun
i=1

It is customary to write this as

n
X
ω= (−1)i−1 v i du1 ∧ · · · ∧ dûi ∧ · · · ∧ dun
i=1

where the hat on the dûi indicates that it is NOT THERE. This is just notation;
it tends to make everything look cleaner, but you must keep your wits about
you and watch for it. Now when we find dω it comes out very nice:
n
X ∂v i j
dω = (−1)i−1 du ∧ du1 ∧ · · · ∧ dûi ∧ · · · ∧ dun
i,j=1
∂uj

Now note that if j 6= i then the differential contain a repetition and thus give
0. Hence only the terms with j = i need be retained and we have
n
X ∂v i j
dω = (−1)i−1 du ∧ du1 ∧ · · · ∧ dûi ∧ · · · ∧ dun
i,j=1
∂uj
X n
∂v i i
= (−1)i−1 du ∧ du1 ∧ · · · ∧ dûi ∧ · · · ∧ dun
i=1
∂ui
Xn
∂v i 1
= i
du ∧ · · · ∧ dui ∧ · · · ∧ dun
i=1
∂u
n
!
X ∂v i
= i
du1 ∧ · · · ∧ dun
i=1
∂u

where in one step the dui had to hop over i − 1 other differentials to find its
proper slot. You will notice has this has the look of divergence about it.
To utilize this formula we must get from our vector to an (n − 1)-form. We
can get from the vector to a 1-form using Φ and then to an (n − 1)-form using
∗. From there, d will take us to an n-form and then another ∗ will get us back
to a 0-form, or scalar, so ∗ ◦ d ◦ ∗ ◦ Φ will get us from a vector to a scalar as we
wish for divergence. Thus ∗ ◦ d ◦ ∗ ◦ Φ is a good candidate for divergence. Let’s
try it in rectangular coordinates. In rectangular coordinates the natural basis
e1 , . . . en is just i, j, k when n = 3

v = v i ei
34 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

Φ(v) = v i dui
Xn
∗Φ(v) = v i (−1)i−1 du1 ∧ · · · ∧ dûi ∧ · · · ∧ dun
i=1
n
!
X ∂v i
d ∗ Φ(v) = du1 ∧ · · · ∧ dun
i=1
∂ui

as we saw above. Finally

Xn
∂v i
∗ d ∗ Φ(v) = = div v
i=1
∂ui

which shows us that indeed the invariant form of div is ∗ ◦ d ◦ ∗ ◦ Φ. If you are
worried about the expressions for ∗, they follow immediately from the general
expressions below.
Before we launch into the final calculation for div, let’s recall the formulas
for the ∗ operator. For ω = ωi dui we have
n
X √
∗ω = (−1)k−1 g ik g ωi du1 ∧ . . . ∧ dûk ∧ . . . ∧ dun
i,k=1

where, as before, the hat on dûj indicates the term is NOT THERE. This is
derived in the Chapter 2. Also derived there are
√
Ω0 = g du1 ∧ . . . ∧ dun
∗ Ω0 = 1
∗ 1 = Ω0

We also require the formula in general coordinates for Φ.

Φ(v) = Φ(v i ei ) = gij v i duj

Now we are ready for the final calculation for div.

div v = ∗ d ∗ Φ(v)
= ∗ d ∗ (gij v i duj )
n
!
X
k−1 jk √ i 1 k n
= ∗d (−1) g g (gij v )du ∧ . . . ∧ dû ∧ . . . ∧ du
k=1

Since g jk gij = δik , the only non-zero terms are when k = i, to the above simplifies
to
n
!
X
i−1 √ i 1 i n
div v = ∗ d (−1) g v du ∧ . . . ∧ dû ∧ . . . ∧ du
i=1
1.17. CURVILINEAR COORDINATES II: THE CALCULATIONS 35
 
n
X √ i
∂( g v )
= ∗ (−1)i−1 j
duj ∧ du1 ∧ . . . ∧ dûi ∧ . . . ∧ dun 
i,j=1
∂u
n √ !
X ∂( g v i ) 1 i n
= ∗ ∧ du ∧ . . . ∧ du ∧ . . . ∧ du
i=1
∂ui
n √ !
1 X ∂( g v i ) √
= ∗ √ g du1 ∧ . . . ∧ dun
g i=1 ∂ui
n √ !
1 X ∂( g v i )
= √
g i=1 ∂ui

and this is the formula for div when v is expressed in natural coordinates v =
v i ei . For orthogonal coordinates we have as usual ei = hi êi and
n
X
v = v i ei
i=1
n
X
= v i hi êi
i=1
n
X
= ṽ i êi
i=1

where ṽ i = hi v i are the physical coordinates with respect to the orthonormal

basis êi . Now using
√
g = h1 h2 · · · hn
ṽ i = hi v i no sum on i

we have
Xn
1 ∂
div v = (h1 h2 · · · hn v i )
h1 h2 · · · hn i=1 ∂ui
Xn
1 ∂
= (h1 · · · hi−1 hi+1 · · · hn ṽ i )
h1 h2 · · · hn i=1 ∂ui

Lastly we want to give a formula for the Laplacian. In physics the formula
for the Laplacian in rectangular coordinates is
Xn
∂f
△f =
p
i
i=1
∂u

The normal notation for the Laplacian is △f . However, in recent years in math-
ematics there has been a big tendency to use the symbol △f for the negative
36 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

of the sum above. There are very good reasons for this5 . So to make sure there
will be no confusion, we will use the symbol △ p f for the Laplacian as custom-

ary in physics. It will be awhile before we need to deal with the mathematical
Laplacian.
To find the Laplacian of functions in any coordinate system is now absurdly
easy, because we have
△p f = div grad f

Thus we need only recall the two previously derived formulas for grad and div

∂f ij
grad f = g ej
∂ui
n √
1 X ∂( g v j )
div v = √
g j=1 ∂uj

Thus it comes down to substituting

∂f ij
vj = g
∂ui
into the formula for div v to get
n  
1 X ∂ √ ij ∂f
△
p f = √ g g
g i,j=1 ∂uj ∂ui

One sees immediately that in orthogonal coordinates we have

Xn  
1 ∂ ∂f
△f =
p h1 · · · hi−1 hi+1 · · · hn i
h1 · · · hn i=1 ∂ui ∂u

1.18 Surfaces and Manifolds

Up to this point we have confined ourselves mostly to Rn but now it is time
to take a more general viewpoint, especially since it will require no additional
effort. We have occasionally talked about two dimensional surfaces in R3 but
there is nothing special about that configuration. A convenient place to start
our discussion is m−dimensional subspaces in Rn . The word surface is usually
used for a 2-dimensional object; the corresponding word for an m−dimensional
subspace of Rn in embedded manifold. (If m = n−1 then the embedded manifold
is often called a hypersurface. This is a special case where things are generally
nicer than they are in general.) The embedded in embedded manifold refers to the
surrounding n−dimensional space. There are manifolds which are not embed-
ded, (for example, the 4-dimensional space of the Universe of General Relativity
5 For example, the mathematical Laplacian has positive or 0 eigenvalues; the physics one

has negative or 0 eigenvalues

1.18. SURFACES AND MANIFOLDS 37

is usually thought of as not embedded in a higher dimensional space,) and we

will briefly discuss this later and also in the second chapter. The advantage of
embedding is that a) it seems familiar and b) it is easy to define tangent vectors.
Our discussion here will be more or less intuitive, and we will not discuss
various pathologies that can occur to make life difficult, as these are not usual
in ordinary physics. Also, we will imitate some of the constructions that are
used in the more general case, of nonembedded manifolds in order to make the
transition easier.
First, it is important that the the surface neither cross itself, nor approach
itself. The first is easy to avoid; the second is more tricky. Recall that a ball in
Rn is the set of points lying within a certain distance of the central point. More
explicitly Br (p) = {x ∈ Rn  |x − p| < r}. Also recall that a homeomorphism is
a one to one onto map from one set to another which is continuous and has a
continuous inverse. Now let p be a point on the manifold. We require that there
is an r small enough so that the the intersection of Br (p) with the manifold be
homeomorphic to a disk in Rm . Put another way, every point p on the manifold
has a neighborhood that looks like a disk in Rm (and the neighborhood is found
by intersecting the manifold with a Br (p)). That is why the manifold has m
dimensions.
Now we need to superimpose coordinates on this situation. That’s easy; the
disk in Rm has coordinates u1 , . . . , un and if q on the manifold corresponds to
y in the disk which has coordinates u1 , . . . , um , which we can then use for p.
Home free? By no means. The most important thing comes next. Suppose p̃
is in the neighborhood of p (as above) and on the manifold. Then p̃ also has
a neighborhood defined by Bs (p̃) and a homomorphism of its own to a disk
in Rm , which will give the neighborhood of p̃ it’s own set of coordinates, say
ũ1 , . . . , ũm . Then, each point in the overlap of Br (p) and Bs (p̃) will have two
sets of coordinates u1 , . . . , um and ũ1 , . . . , ũm . Each set will be a function of the
other:

ũi = ũi (u1 , . . . , um )

ui = ui (ũ1 , . . . , ũm )

The critical thing here is that these functions have many derivatives. Many
means, for physics purposes, at least three continuous derivatives. The manifold
is then called a C3 -manifold. In mathematics it is customary to avoid the boring
details and require the functions to have infinitely many continuous derivatives
so it is C∞ -manifold. This seemingly boring requirement is the key to the whole
subject; it is what makes the manifold smooth like a ball rather than unsmooth
like a cube. It takes some contemplation to see this but it’s true.
One more comment; in classical tensor analysis they are continually talking
about coordinate changes as if they made some big difference. They do; the
talk of coordinate changes in tensor analysis is just how that subject deals with
the material in the previous paragraph6.
6 This is not a well known fact even among the tensor users. Reveal it only to trusted

friends
38 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

There is a little more to the story, because, after all, we have to put the
manifold inside Rn and this requires a little more to be put into the definition.
The requirement is easy to state. Since the manifold is inside Rn and p will
thus have coordinates x1 , . . . , xn . These will be functions of p’s disk coordinates
u1 , . . . , um ; that is

xi = xi (u1 , . . . , um ) i = 1, . . . , n

The requirement is that the n × m matrix

 ∂x1 ∂x1 
∂u21 ··· ∂um
 ∂x
··· ∂x2 
 ∂u1 ∂um 
 .. .. .. 
 . . . 
n
∂x ∂xn
∂u1 ··· ∂um

have rank m. Since m ≤ n, m is the largest rank it could have, so we often say
the matrix has maximal rank. The reason this is important is that we could set
up a coordinate system around any p for Rn that used u1 , . . . , um for the first
m coordinates and then n − m other coordinates to finish the job, but we do
not need to pursue this.
We should also mention that while in theory we have set the system up to
use disks in Rm to make coordinates, you can actually do it with any regions
you like rather than disks. We only did it with disks to make it precise and
visual. Things are rather looser in practise.
The test to see that this condition is satisfied is to check that all the m × m
minors of the matrix are non-zero. It is now time for an example. We will
use (as usual) the sphere in 3-space, which illustrates the concepts nicely. The
coordinates will be θ, φ where φ is the longitude and θ is the angle off the z-axis7.
The R3 coordinates of the point with coordinates θ, φ are

x = sin θ cos φ
y = sin θ sin φ
z = cos θ
R = (sin θ cos φ, sin θ sin φ, cos θ)

The natural basis of the tangent space is

∂R
e1 = = (cos θ cos φ, cos θ sin φ, − sin θ)
∂θ
∂R
e2 = = (− sin θ sin φ, sin θ cos φ, 0)
∂φ
e1 × e2
n̂ = = (sin2 θ cos φ, sin2 θ sin φ, sin θ cos θ)/ sin θ
|e1 × e2 |
= (sin θ cos φ, sin θ sin φ, cos θ)
7 Mathematicians beware; math convention is opposite!
1.18. SURFACES AND MANIFOLDS 39

The matrix discussed above is then

 
cos θ cos φ − sin θ sin φ
 cos θ sin φ sin θ cos φ 
− sin θ 0

The minors are

 
cos θ sin φ sin θ cos φ
Row 23 = = sin2 θ cos φ
− sin θ 0
 
cos θ cos φ − sin θ sin φ
Row 13 = = − sin2 θ sin φ
− sin θ 0
 
cos θ cos φ − sin θ sin φ
Row 12 = = sin θ cos θ
cos θ sin φ sin θ cos φ

There is a lot that can be learned from this example. Notice first that the minors
are all 0 at the poles where θ = 0 or θ = π. This is reflected also in e2 which is
0 at the poles and thus not much of a basis vector for the tangent space. Note
that at the poles φ is not defined either. Note that in n̂ = (e1 × e2 )/|e1 × e2 |
both numerator and denominator are 0, so n̂ is not well defined though it seems
to be after the 0’s disappear because of the cancellation of sin θ.
The proper response to this is to select a second coordinate system, perhaps
one with east and west poles, to cover the two offending points with proper
coordinates. Needless to say, this is hardly ever done in this and many other
cases, because with a little sympathetic treatment it is possible to get around
the troubles at the poles. However, it is really important to notice such bad
points and to make sure whatever you are doing makes sense at such bad points.
Often it does, sometimes it doesn’t. Beware.
Now on to other things. Notice in the example how easy it was to get the
basis of tangent vectors e1 , e2 , one for each coordinate, at each point of the
manifold. The vector space spanned by e1 , e2 is called the Tangent plane, and
is denoted by Tp (S) (where S is the sphere). In the more general case of an
embedded m-dimensional manifold S in n-dimensional space, we will have m
tangent vectors at each point, e1 , . . . , em , and they will be a basis for an m-
dimensional space Tp (S) called the Tangent Space to the manifold S at p. The
condition on the minors guarantees that the ei will be linearly independent.
1 m
Consider an m-dimensional manifold S with coordinates Pu i , . . . , u and basis
i
vectors ei = ∂R/∂u . Now let v ∈ Tp (S). Then v = v ei and we need a
function whose input is v and whose output is v i . This function is dui . Although
the idea is new to you, this is what dui actually is. Thus the defining equation
is
dui (v) = v i
(We have seen this before; I’m just reminding you in this new context.) If
you are familiar with the the space of linear functionals of a vector space, then
du1 , . . . , dum form a basis for this space of linear functionals (called the dual
space and denoted by Tp∗ (S).
40 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

When you follow this path for awhile, you become aware of what a small role
the embedding space Rn plays in the theory. Riemann also wondered about this,
and eventually showed how to get around it. Let us ask what the important
contributions of the embedding space Rn are. There are two most important
contributions. First, the embedding space provides an inner product for the
Tp (S) which live inside it. Second, it provides us with a convenient way (visible
in the sphere example) of finding tangent vectors and thus the tangent space
Tp (S). The first is a critical idea; the second just requires technical trickery. For
the first, Riemann proposed attaching an inner product in the form of an m × m
positive definite matrix to each point of S. In doing this Riemann invented
the Riemannian Manifold. For the second, the tangent vectors, Tensor Analysis
defined a vector as an object that changes in certain ways under coordinate
changes. This works but is a bit unsatisfying. The more modern way is to define
tangent vectors as directional derivatives on the manifold. We will look at this
further in chapter 2 where we will define differentiable manifolds, although most
of the work has already been done in this section. For the remainder of this
chapter we will stay with embedded manifolds.
We must also mention that the theory we develop subsequently must be
modified if the manifold is not orientable. Examples of non-orientable manifolds
are the Möbius strip and the Klein bottle. The problem is that when a right
handed orthonormal basis of the tangent space is slid around the median line
of the Möbius strip it returns as a left handed orthonormal basis. The easiest
definition of orientable is that an n-dimensional manifold K (embedded or not)
is orientable if an only if it has a never vanishing form ω ∈ Λn (K). Such a form
is called a topform. From it one can always manufacture a volume form if K
has an inner product (e.g. is a Riemannian Manifold). We will not consider
non-orientable manifolds in what follows and some of the material will not work
on them. See Chapter 2 for more discussion of this matter.

1.19 The Dualizing Operator ∗

One cannot get very far in differential forms without the dualizing operator
∗ : Λr (R) → Λn−r (R). This operator reflects analytically certain geometric
properties studied in higher geometry courses called duality properties. Sadly
we cannot take the time to talk about this geometry, which would take many
pages to deal with and anyway is not a matter of great practical importance for
most of physics. In this section we will give a brief introduction to ∗ which will
suffice for most of the more elementary applications, as for example Maxwell’s
equations which we present in a later section.
In most elementary applications of ∗ we deal with the cases

r=0 ∗ : Λ0 (R) → Λn (R)

r=n ∗ : Λn (R) → Λ0 (R)
1.19. THE DUALIZING OPERATOR ∗ 41

and
r=1 ∗ : Λ1 (R) → Λn−1 (R)
r =n−1 ∗ : Λn−1 (R) → Λ1 (R)
In each case it is only necessary to derive one of the two formulas; the other is
then obtained from
∗ ∗ ω = (−1)r(n−r) for ω ∈ Λr (R)
To define ∗ it is necessary first to put an inner product on each Λr . This is
easily done; We define the inner product with
ω = ωi dui and η = ηi dui
by
(ω, η) = g ij ωi ηj
We can now put an inner product on each Λr as follows
 
 (ω1 , η1 ) . . . (ω1 , ηr ) 
 
 .. .. .. 
(ω1 ∧ . . . ∧ ωr , η1 ∧ . . . ∧ ηr ) =  . . . 
 
 (ωr , η1 ) . . . (ωn , ηr ) 

and extending by linearity. This defines an inner product on Λr for 1 ≤ r ≤ n,

but we also need it on Λ0 which is just the scalars. For these we define
(r, s) = rs for r, s scalers
For completeness, we remark that if ω ∈ Λr and η ∈ Λs and r 6= s, 0 ≤ r, s ≤ n,
then we put
(ω, η) = 0
Our most important use of this formula is the following case where r = n:
 
 (ω1 , η1 ) . . . (ω1 , ηn ) 
 
 .. .. .. 
(ω1 ∧ . . . ∧ ωn , η1 ∧ . . . ∧ ηn ) =  . . . 
 
 (ωn , η1 ) . . . (ωn , ηn ) 

Noting that the above formula gives us

(dur , dus ) = (δri dui , δsj duj ) = g ij δri δsj = g rs
we have with the inner product in Λn
 
 (du1 , du1 ) ... 
(du1 , dun )
 
 .. .. ..
(du1 ∧ . . . ∧ dun , du1 ∧ . . . ∧ dun ) =  . . .
 
 n 1 n n 
(du , du ) . . . (du , du )
 
 g 11 . . . g 1n 


=  .. .. ..  = det (g i,j ) = 1
 . . .  g
 g n1 . . . g nn 
42 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

Thus
√ √
( g du1 ∧ . . . ∧ dun , gdu1 ∧ . . . ∧ dun ) = 1
We will define
def √
Ω0 = g du1 ∧ . . . ∧ dun
and refer to it as the normalized topform. The normalized topform is unique up
to sign; it will change sign if two variables interchange their numbers. Choice of
a sign is the same as choice of an orientation. The reason it is essentially unique
is that Λn is one-dimensional, so there are only two elements of size 1 and they
are negatives of one another.
If M is a n-dimensional region than the volume of M is
Z
vol(M ) = Ω0
M

There are of course subtleties here; for example the sign could come out wrong.
And of course you have to prove this works using whatever your definition of
volume is. We will ignore these problems. Also we mention that if f is the
density of something, electrical charge for example, or gas, then the amount of
the stuff will be Z
Amount of stuff in M = f Ω0
M

Next we want the official definition of ∗ which is at last possible since we

have the inner product on Λr . For any α ∈ Λn−r we will always have

ω ∧ α = k Ω0

for some constant k. Then for η ∈ Λr we define ∗η as that unique element of

Λn−r for which the constant k is (ω, η), which comes down to

ω ∧ ∗η = (ω, η) Ω0

This is the most important equation involving ∗; if you remember this you
can derive everything else. The existence and uniqueness of ∗η are derived in
Chapter 2, but it is not a difficult matter; it comes down to the representation
of a linear functional in an inner product space.
Some equations readily follow from the basic equation. For example, we
have

ω ∧ ∗η = (ω, η)Ω0 = (η ω)Ω0 = η ∧ ∗ω

ω ∧ ∗η = η ∧ ∗ω

which has important uses, for example in the next section on the codifferential.
Other important equations like

∗ ∗ ω = (−1)r(n−r)ω for ω ∈ Λr
1.19. THE DUALIZING OPERATOR ∗ 43

do not follow so readily from the basic equation. We will have more to say about
this equation later.
Next, using the basic equation ω ∧ ∗η = (ω, η) Ω0 we want to derive the
formulas for ∗ for general coordinates and 1-forms. Recall that a hat over a
term means that it is missing. We have {du1 , . . . , dun } is a basis for Λ1 . A
basis for Λn−1 is
ci ∧ . . . ∧ dun
du1 ∧ . . . ∧ du 1≤i≤n
Thus, since dui ∈ Λ1 , we have ∗dui ∈ Λn−1 and then we can express ∗dui as
n
X
∗dui = dj ∧ . . . ∧ dun
(−1)j−1 aj du1 ∧ . . . ∧ du
j=1

where the a are functions of u1 , . . . , un which we must determine and the

j

(−1)j−1 is inserted for convenience. We now compute both sides of the basic
equation and setting them equal will give us the aj . The basic equation is
duk ∧ ∗dui = (duk , dui )Ω0 . Computing the left side of the basic equation we
have
 
n
X
duk ∧ ∗dui = duk ∧  (−1)j−1 aj du1 ∧ . . . ∧ du dj ∧ . . . ∧ dun 
j=1
n
X
= dj ∧ . . . ∧ dun
(−1)j−1 aj duk ∧ du1 ∧ . . . ∧ du
j=1

All terms on the right side in the sum will be 0 except the one where j = k
because if j 6= k there will be repetition in the differentials killing the term.
Thus
duk ∧ ∗dui = ak du1 ∧ . . . ∧ duk ∧ . . . ∧ dun
Notice how the (−1)j−1 was used to return duk to its proper place in the product
of differentials. This is a good trick to remember, though it is never important,
just convenient. Next we compute the right side of the of the basic equation.
We have
√
(duk , dui )Ω0 = g ki g du1 ∧ . . . ∧ dun
Comparing the two expressions we see that
√
ak = g ki g
and thus
n
X
∗dui = dk ∧ . . . ∧ dun
(−1)k−1 ak du1 ∧ . . . ∧ du
k=1
n
X √ dk ∧ . . . ∧ dun
= (−1)k−1 g ki gdu1 ∧ . . . ∧ du
k=1
44 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

This equation
Pn k−1 ki √ dk ∧ . . . ∧ dun
∗dui = k=1 (−1) g gdu1 ∧ . . . ∧ du

is of fundamental importance. You have already seen it used in the derivation

of the formulas for curvilinear coordinates.
dk ∧ . . . ∧ dun in the same way but
It is possible to determine ∗du1 ∧ . . . ∧ du
that method leads into a forest of determinants. Instead, we will use the formula
∗ ∗ ω = (−1)r(n−r) ω with r = 1 and some trickery involving the inverse matrices
(g ji ) and (gik ).
n
X √ dj ∧ . . . ∧ dun
∗dui = (−1)j−1 g ji g du1 ∧ . . . ∧ du
j=1
n
X Xn
√ dj ∧ . . . ∧ dun
gik ∗dui = (−1)j−1 g ji gik g du1 ∧ . . . ∧ du
i=1 i,j=1
 
n
X n
X √ dj ∧ . . . ∧ dun
= (−1)j−1  g ji gik  g du1 ∧ . . . ∧ du
j=1 j=1
n
X √ dj ∧ . . . ∧ dun
= (−1)j−1 δkj g du1 ∧ . . . ∧ du
j=1
√ dk ∧ . . . ∧ dun
= (−1)k−1 g du1 ∧ . . . ∧ du

Hence, starring both sides, we have

n
X
k−1 √ 1 dk ∧ . . . ∧ dun
(−1) g ∗ du ∧ . . . ∧ du = gik ∗ ∗ dui
i=1
n
X
= gik (−1)1(n−1) dui
i=1
n
X
dk ∧ . . . ∧ dun
∗du1 ∧ . . . ∧ du = (−1)−k+1 (−1)n−1 gik dui
i=1

giving us our final answer

dk ∧ . . . ∧ dun = (−1)n−k Pn gik dui

∗du1 ∧ . . . ∧ du i=1

This is not quite sufficient for our needs. We also need formulas for ∗ when
r = 0 and when r = n. These are easy. Recall that Λ0 is just the scalars R and
that the inner product is just the ordinary product in R and that a basis for R
is just the single scalar 1. Then

∗1 = Ω0
1.19. THE DUALIZING OPERATOR ∗ 45

since
1 ∧ ∗1 = 1 ∧ Ω0 = Ω0 = (1, 1)Ω0

as required. Then
∗Ω0 = ∗ ∗ 1 = (−1)0(n−0) 1 = 1

We have determined the formulas in generalized coordinates for 0-forms, 1-

forms, (n − 1)-forms and n-forms. It is possible to work out similar formulas
for r-forms, but these are not as important in applied mathematics and to
derive them we would have to introduce some equipment involving minors of
determinants. Hence we will put this off to Chapter 2 and concentrate here
on the more practically oriented formulas for orthogonal coordinates which are
relatively easy to derive. Recall that for orthogonal coordinates we have for the
position vector R = R(u1 , . . . , un )

∂R
ej =
∂ui
hi = (ei , ei )
1
êi = ei
hi
1
(dui , duj ) = g ii =
h2i
(hi dui , hi duj ) = 1
Ei = hi dui

so that {h1 du1 , . . . , hn dun } form an orthonormal set. For convenience let us
set Ei = hi dui . Then (Ei , Ej ) = δij . Now from the basic equation ω ∧ ∗η =
(ω, η)Ω0 we see immediately that

∗(Ei1 ∧ · · · ∧ Eir ) = sgn(σ)(Ej1 ∧ · · · ∧ Ejn−r )

where {j1 , . . . , jn−r } = {1, . . . , n} − {i1 , . . . , ir } and σ is the permutation

  
1 . . . r  r + 1 ... r
σ=
i1 . . . ir  j1 . . . jn−r

Here the convention is 1 ≤ i1 < . . . < ir ≤ n and 1 ≤ j1 < . . . < jn−r ≤ n but in
fact the formula will work just as well with the i’s and j’s in any order. Indeed
it suffices to check this for ω running through a basis {Ek1 ∧ · · · ∧ Ekr ) 1 ≤
i1 , . . . , ≤ kr } of Λr .

(Ek1 ∧ · · · ∧ Ekr ) ∧ ∗(Ei1 ∧ · · · ∧ Eir ) = Ek1 ∧ · · · ∧ Ekr ∧ sgn(σ)(Ej1 ∧ · · · ∧ Ejn−r )

Now if {k1 , . . . , kr } 6= {i1 , . . . , ir } there will be repetitions on the right side and
the result will be 0. Hence we will take {k1 , . . . , kr } = {i1 , . . . , ir } and since
46 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

both are in increasing order we have iℓ = kℓ and

(Ei1 ∧ · · · ∧ Eir ) ∧ ∗(Ei1 ∧ · · · ∧ Eir ) = Ei1 ∧ · · · ∧ Eir ∧ sgn(σ)(Ej1 ∧ · · · ∧ Ejn−r )

= E1 ∧ · · · ∧ En
= (h1 du1 ) ∧ · · · ∧ (hn dun )
= h1 · · · hn du1 ∧ · · · ∧ dun
√
= g du1 ∧ · · · ∧ dun
= Ω0

On the other hand

 
(Ek1 ∧ · · · ∧ Ekr , Ei1 ∧ · · · ∧ Eir )Ω0 = det (Ekℓ , Eim ) Ω0

Now if {k1 , . . . , kr } =
6 {i1 , . . . , ir } then there will be a row in the determinant
which is entirely 0. Hence we take {k1 , . . . , kr } = {i1 , . . . , ir } and since both
are in increasing order we have iℓ = kℓ , the determinant has 1’s on the main
diagonal and 0’s elsewhere, so the determinant is 1 and the result is

(Ei1 ∧ · · · ∧ Eir , Ei1 ∧ · · · ∧ Eir )Ω0 = Ω0

Hence if {k1 , . . . , kr } =
6 {i1 , . . . , ir } both sides of the fundamental equation are
0 and if {k1 , . . . , kr } = {i1 , . . . , ir } then both sides are Ω0 , proving that

∗(Ei1 ∧ · · · ∧ Eir ) = sgn(σ)(Ej1 ∧ · · · ∧ Ejn−r )

Replacing the Ei by hi dui we have

∗(hi1 dui1 ∧ · · · ∧ hir duir ) = sgn(σ)(hj1 duj1 ∧ · · · ∧ hjn−r dujn−r )

and from this

hj1 · · · hjn−r
∗(dui1 ∧ · · · ∧ duir ) = sgn(σ) (duj1 ∧ · · · ∧ dujn−r )
hi1 · · · hir

where σ is the permutation

  
1 . . . r  r + 1 ... n
σ=
i1 . . . ir  j1 . . . jn−r

This gives us a formula valid for any r

This is also a convenient moment to prove the formula

∗ ∗ ω = (−1)r(n−r) ω
1.20. THE CODIFFERENTIAL δ 47

We can do this using the basis {E1 , . . . , En }. We define the reverse σ̃ of σ to

be, with the above σ,
  
1 . . . n − r  n − r + 1 . . . n
σ̃ =
j1 . . . jn−r  i1 . . . ir

Now if we move the r i’s each past the n − r j’s there will be a total of r(n − r)
hops to get from σ̃ back to σ. Hence if it takes s hops to return σ to the identity,
sgn(σ) = (−1)s and

sgn(σ̃) = (−1)s+r(n−r) = (−1)r(n−r) sgn(σ)

sgn(σ)sgn(σ̃) = (−1)r(n−r)

Then, with ω = Ei1 ∧ · · · ∧ Eir we have

∗∗ω = ∗ ∗ Ei1 ∧ · · · ∧ Eir

= sgn(σ) ∗ Ej1 ∧ · · · ∧ Ejn−r
= sgn(σ)sgn(σ̃)Ei1 ∧ · · · ∧ Eir
= (−1)r(n−r) Ei1 ∧ · · · ∧ Eir
= (−1)r(n−r) ω

Since this is true for the elements of a basis, it will be true for all ω by linearity.

1.20 The Codifferential δ

To deal with the Laplacian for forms, we need the codifferential δ : Λr → Λr−1 .
We will use this in our treatment of Maxwell’s equations. For example, the
condition of Lorenz is naturally expressed in terms of the codifferential.
In contrast to previous constructions, δ depends for its very definition on
the presence of an inner product for the forms on a manifold M . Using the ∗ we
can easily define such an inner product. Let ω, η ∈ Λr . Then the inner product
is defined by either of the following
Z Z
((ω, η)) = ω ∧ ∗η = (ω, η)Ω0
M M

Some points to remember. The inner product (ω, η) is an inner product that
lives in Tp (M ) for each p ∈ M . It and ∗ are strictly local, being algebraic con-
structions on each tangent space Tp (M ) separately. On the other hand, ((ω, η))
depends on integrating the information given at each p over the manifold. It is
thus a global object. Second, although we express it in our formulas for a single
coordinate system, in many cases it will be necessary to cut M into separate
pieces each of which lives inside the domain of a coordinate system, and to use
that coordinate system when integrating over that piece. Sometimes, as for the
sphere, we can use a single coordinate system even though that system is bad
at specific points, like the North and South pole. This is best considered dumb
48 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

luck, and one should always be careful when doing it. But it does work often in
applications as long as nothing is discontinuous at the bad points.
Now that we have the inner product on forms, we can consider ((ω, dη))
where ω ∈ Λr and η ∈ Λr−1 . We will consider two possible scenarios
a The manifold M has no boundary; ∂M = ∅ Such a manifold is often referred
to as closed. The idea ”no boundary” includes ”no boundary at infinity”
so that in this case the manifold M is compact.
b Either ω or η vanishes off a compact subset M0 of M , so the integral will
always be finite.
With one of these situations in place, we can do an integration by parts in
the way you have seen done in differential equations, and come up with formal
adjoint δ for d. After all this preamble, the actual calculation is quite short.
We need to recall that, with ω ∈ Λr−1 and η ∈ Λr

ω ∧ ∗η = η ∧ ∗ω
d(ω ∧ ∗η) = dω ∧ ∗η + (−1)r−1 ω ∧ d ∗ η

Also note because of the above assumptions on ω and η we have by Stokes

theorem Z Z
d(ω ∧ η) = ω∧η =0
K ∂K
We can now derive the formula for the (formal) adjoint of d.
Z
((dω, η)) = dω ∧ ∗η
ZK Z
r−1
= d(ω ∧ η) − (−1) ω∧d∗η
K K
Z
= 0 + (−1)r ω∧d∗η
K
Z
= (−1)r (−1)(n−r+1)(r−1) ω∧∗∗d∗η
K

Now we must simplify the exponent. Recall that

k2 ≡ k (mod 2)
−k ≡ k
2k ≡ 0
k(k − 1) ≡ 0

Thus

r + (n − r + 1)(r − 1) ≡ r + (n − r)(r − 1) + r − 1 (mod 2)

1.20. THE CODIFFERENTIAL δ 49

≡ (n − r)(r − 1) − 1
≡ n(r − 1) − r(r − 1) − 1
≡ n(r − 1) − 1 + 2
≡ n(r − 1) + 1

Hence
Z
n(r−1)+1
((dω, η)) = (−1) ω∧∗∗d∗η
K
Z  
= ω ∧ ∗ (−1)n(r−1)+1 ∗ d ∗ η
K

Thus if we set
δ : Λr → Λr−1
defined by
δη = (−1)n(r−1)+1 ∗ d ∗ η η ∈ Λr

we have
((dω, η)) = ((ω, δη)) ω ∈ Λr−1 , η ∈ Λr

as we wanted. Naturally δ is only a formal adjoint to d because we have not

considered any boundary conditions, but this is not unusual in mathematical
physics.
It is worth noticing that the formula for δ simplifies if we consider the cases
of even and odd dimensional spaces separately. We have

n odd: δω = (−1)r ∗ d ∗ ω

n even: δω = − ∗ d ∗ ω
There are some useful identities connecting ∗, d, and δ which are conse-
quences of ∗ being almost an involution, which we now derive. First we have

∗ δω = (−1)n(r−1)+1 ∗ ∗ d ∗ ω ω ∈ Λr
= (−1)n(r−1)+1 (−1)(n−r+1)(r−1) d ∗ ω ω ∈ Λr
= (−1)r d ∗ ω ω ∈ Λr

since

n(r − 1) + 1 + (n − r + 1)(r − 1) ≡ (n + n − r + 1)(r − 1) + 1 mod 2

≡ −(r − 1)(r − 1) + 1
≡ (r − 1)2 + 1
≡ (r − 1) + 1
≡ r
50 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

Applying our identity to ∗ω we have

∗δ ∗ω = (−1)n−r d ∗ ∗ ω
= (−1)n−r+r(n−r) d ω
= (−1)(n−r)(r+1) d ω
∗∗ δ∗ω = (−1)(n−r)(r+1) ∗ d ω
(n−r−1)(r+1)
(−1) δ∗ω = (−1)(n−r)(r+1) ∗ d ω
δ∗ω = (−1)(n−r)(r+1)+(n−r−1)(r+1) ∗ d ω
δ∗ω = (−1)(r+1)(n−r+n−r−1) ∗ d ω
δ∗ω = (−1)r+1 ∗ d ω

The r in these formulas always refers to the degree of ω, and this must be
carefully remembered when applying the formulas. It is easy to make mistakes
applying these formulas.

1.21 The Laplacian

One of the most important operators in mathematical physics is the Laplacian.
We have derived formulas for the Laplacian on functions but using the codif-
ferential it is possible to give formulas for the Laplacian on differential forms
also. This is important, for example, in Electromagnetic theory where we will
need the Laplacian of the one form which expresses the vector potential of the
magnetic field.
The Laplacian
Xn
∂2f
△p f =

i=1
∂xi2

has a long history in mathematical physics and almost as long a history in

pure mathematics. New vistas for the Laplacian were opened by Beltrami in
the 1860’s when he defined it on a surface. Eventually it was found to have
an unexpected geometric meaning when defined on differential forms. We will
have the briefest of looks at this much later in the book. In earlier ages the
Laplacian was defined by use of Tensors, but in the 20th Century it was realized
it could be defined in terms of the ∗ operator and d, which made for a more
elegant exposition, though problems remain in the calculation of formulas. The
mathematical definition on forms is now
Def △ = dδ + δd
on any Riemannian or pseudo-Riemannian manifold. Note that we must have
an inner product on the surface or manifold in order to define ∗; no inner
product, no Laplacian! The pseudo in pseudo-Riemannian means that the inner
product must be non-degenerate but need not be positive definite as is the case in
Relativity. We will go into this is some detail in the section of Electromagnetics
in four-dimensional space-time.
1.21. THE LAPLACIAN 51

Note also the extremely important point that by it’s definition the Laplacian
is independent of the coordinate system. It is even independent of the choice of
orientation, since reversing the orientation reverses the sign of each of the two
∗’s found in each term of it’s definition and since there are two minus signs each
term will come out the same.
Most annoyingly, the mathematical operator △ does not quite coincide with
the operator △ p used in classical mathematical physics and we must be a bit

careful about this; the relationship is △ p = −△. In order to keep it absolutely

clear which Laplacian we are using I have invented the symbol △ p for the familiar

Physics Laplacian.
There are a couple of things that can be said in favor of the mathematics
notation. First, the eigenvalues λ of △ are positive (or 0) while those of △ p

are negative (or 0). Connected with this is the fact that ((△ω, ω)) ≥ 0 while
((△p ω, ω)) ≤ 0. The wave equations are

1 ∂2u 1 ∂ 2u
+ △ω = 0 −△
p ω = 0
c2 ∂t2 c2 ∂t2
and the Schrödinger equations

∂Ψ ~2 ∂Ψ ~2 p
i~ = △Ψ+VΨ i~ =− △Ψ + V Ψ
∂t 2m ∂t 2m
and Poisson’s equation for the potential of charge distribution
ρ ρ
△φ = △
p φ = −
ǫ ǫ
with the math forms perhaps marginally more elegant. There may be other
reasons too for suspecting that Laplace chose the wrong sign. All we can do at
the present time is to be careful about which one we are using.
Notice that d : Λr → Λr+1 and δ : Λr → Λr−1 so that the combined
△ : Λr → Λr . Notice also that the definition of the Laplacian defines △ on
forms of all degrees, not just on functions. This is very important.
Another point of interest is that △ is a sort of square. Recall that d2 = dd = 0
and δ 2 = δδ = 0. Thus we have

(d + δ)2 = (d + δ)(d + δ) = dd + dδ + δd + δδ
= dδ + δd = △

However, this is not as interesting as it looks because

d + δ : Λr → Λr+1 ⊕ Λr−1

and it is hard to see how to get practical use out of that.

Before going on it will be instructive to calculate a couple of Laplacians from
scratch. Although we could make use of formulas derived in previous sections,
I think it is interesting to see the whole calculation in one place. For functions
52 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

f ∈ Λ0 we have δf = 0 (since ∗f ∈ Λn and thus d ∗ f = 0) which simplifies the

calculation greatly. Thus we have

△f = (dδ + δd)f = δdf

= (−1)n(1+1)+1 ∗ d ∗ (df )
n
!
X ∂f i
= −∗d∗ dx
i=1
∂xi
n
!
X ∂f
= −∗d (−1) i−1
dx1 ∧ . . . ∧ dxi−1 ∧ dxi+1 ∧ . . . ∧ dxn
i=1
∂xi
 
n
X 2
∂ f
= −∗ (−1)i−1 dxj ∧ dx1 ∧ . . . ∧ dxi−1 ∧ dxi+1 ∧ . . . ∧ dxn 
i,j=1
∂xi ∂xj

The only non zero term (due to repetitions) is when j = i, hence

n
!
X ∂2f
△f = −∗ (−1)i−1
dxi ∧ dx1 ∧ . . . ∧ dxi−1 ∧ dxi+1 ∧ . . . ∧ dxn
i=1
∂xi2
n
!
X ∂2f
= −∗ dx1 ∧ . . . ∧ dxi−1 ∧ dxi ∧ dxi+1 ∧ . . . ∧ dxn
i=1
∂xi2
Xn
∂2f
= −
i=1
∂xi2

which is just what we were expecting. A very similar calculation will get △f
for general coordinates, a result which we already have found by other methods,
and which you can do in the problems.
For r = 1 the computation is a lot more complicated and we will only do it
for n = 3, which avoids unpleasantness and suffices for our applications. (The
big advantage of doing it for n = 3 is that we can exploit cyclic order.) We will
first compute δdω, then compute dδω, and then put them together to get △ω.
Although there are some interesting features, you might want to just skim this
calculation.

ω = ω1 dx + ω2 dy + ω3 dz
     
∂ω3 ∂ω2 ∂ω1 ∂ω3 ∂ω2 ∂ω1
dω = − dy ∧ dz + − dz ∧ dx + − dx ∧ dy
∂y ∂z ∂z ∂x ∂x ∂y
     
∂ω3 ∂ω2 ∂ω1 ∂ω3 ∂ω2 ∂ω1
∗ dω = − dx + − dy + − dz
∂y ∂z ∂z ∂x ∂x ∂y
    
∂ ∂ω2 ∂ω1 ∂ ∂ω1 ∂ω3
d ∗ dω = − − − dy ∧ dz
∂y ∂x ∂y ∂z ∂z ∂x
    
∂ ∂ω3 ∂ω2 ∂ ∂ω2 ∂ω1
+ − − − dz ∧ dx
∂z ∂y ∂z ∂x ∂x ∂y
1.21. THE LAPLACIAN 53
    
∂ ∂ω1 ∂ω3 ∂ ∂ω3 ∂ω2
+ − − − dx ∧ dy
∂x ∂z ∂x ∂y ∂y ∂z
 2 
∂ ω2 ∂ 2 ω1 ∂ 2 ω1 ∂ 2 ω3
∗ d ∗ dω = − − + dx
∂y∂x ∂y 2 ∂z 2 ∂z∂x
 2 
∂ ω3 ∂ 2 ω2 ∂ 2 ω2 ∂ 2 ω1
+ − − + dx
∂z∂y ∂z 2 ∂x2 ∂x∂y
 2 
∂ ω1 ∂ 2 ω3 ∂ 2 ω3 ∂ 2 ω2
+ − − + dx
∂x∂y ∂x2 ∂y 2 ∂y∂z

OK, ∗d ∗ d is done. Now we need d ∗ d∗, and then we can put them together.
This one is easier.

ω = ω1 dx + ω2 dy + ω3 dz
∗ω = ω1 dy ∧ dz + ω2 dz ∧ dx + ω3 dx ∧ dy
 
∂ω1 ∂ω2 ∂ω3
d∗ ω = + + dx ∧ dy ∧ dz
∂x ∂y ∂z
 
∂ω1 ∂ω2 ∂ω3
∗d∗ ω = + +
∂x ∂y ∂z
 2 2
  2 
∂ ω1 ∂ ω2 ∂ 2 ω3 ∂ ω1 ∂ 2 ω2 ∂ 2 ω3
d∗ d∗ ω = + + dx + + + dy
∂x2 ∂x∂y ∂x∂z ∂y∂x ∂y 2 ∂y∂z
 2 
∂ ω1 ∂ 2 ω2 ∂ 2 ω3
+ + + dz
∂z∂x ∂z∂y ∂z 2

δdω = ((−1)3(2+1)+1 ∗ d∗)dω = ∗d ∗ dω

dδω = d((−1)3(1+1)+1 ∗ d∗)ω = −d ∗ d ∗ ω
△ω = (δd + dδ)ω
= ∗d ∗ dω − d ∗ d ∗ ω
 2   2 
∂ ω1 ∂ 2 ω1 ∂ 2 ω1 ∂ ω2 ∂ 2 ω2 ∂ 2 ω2
= − + + dx − + + dy
∂x2 ∂y 2 ∂z 2 ∂x2 ∂y 2 ∂z 2
 2 
∂ ω3 ∂ 2 ω3 ∂ 2 ω3
− + + dz
∂x2 ∂y 2 ∂z 2

This is certainly an interesting result. We denote the Laplacian on functions

momentarily by △0 : Λ0 → Λ0 then the result can be written

△ ω = (△0 ω1 )dx + (△0 ω2 )dy + (△0 ω3 )dz

How general is this result? It certainly cannot work for general coordinates, but
it does work for rectangular coordinates on Λr (Rn ). Unfortunately we do not
know any better way to prove this than by brute force. For our applications
what we have done suffices.
54 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

There is a vector form of the above calculation based on a well known formula
in three dimensions

curl curl(v) = grad div(v) − ▽2 (v)

which we rewrite as

△
p (v) = grad div(v) − curl curl(v)

where in rectangular coordinates

▽2 (v) = (▽2 v 1 ) î + (▽2 v 2 ) ĵ + (▽2 v 3 ) k̂

In general coordinates and three dimensions we can use the above formula as a
definition of △
p (v) since we have formulas for grad div(v) and curl curl(v). If

you examine the differential form calculation above carefully, you will be able
to see the dδω corresponds to grad div(v) and δdω corresponds to curl curl(v).
Soon we will show that ∗ △ ω = △ ∗ ω in all circumstances. From this we
can easily derive that in Λ2 (R3 )

△(ω 1 dydz + ω 2 dzdx + ω 3 dxdy) = △0 (ω 1 ) dydz + △0 (ω 2 ) dzdx + △0 (ω 3 ) dxdy

and in Λ3 (R3 )
△(f dxdydz) = △0 (f ) dxdydz
which tells us all we need to know about Λr (R3 ).
Now let’s develop a bit more theory. First, recall that the global inner
product (on all of K) is
Z Z
((ω, η)) = (ω, η)Ω0 = ω ∧ ∗η ω, η ∈ Λr (K)
K K

provided the integral is finite, which it will be if the support


supp(ω) = (closure of {x ∈ K ω(x) 6= 0})

of ω or the support of η is compact. (If K itself is compact this condition

is automatically valid. Compact for our purpose means closed and bounded.)
We are assuming here that all functions and coefficients of forms have enough
continuous derivatives to make the calculations make sense. In general three is
enough8
Recall from the chapter on the codifferential that

((dω, η)) = ((ω, δη)) ω ∈ Λr , η ∈ Λr+1

8 In mathematics these objects are usually taken to have infinitely many derivatives, so as

to avoid thinking about such things.

1.21. THE LAPLACIAN 55

From this we have

((△ω, η)) = ((dδ ω, η) + ((δd ω, η))

= ((δω, δ η) + ((d ω, dη))

This shows that if ω has coefficients with two continuous derivatives that

((△ω, ω)) = 0 if and only if δω = 0 and dω = 0

and this is also clearly equivalent to △ω = 0. This leads to the definition

Def ω (with at least one continuous derivative) is Harmonic if and only if
δω = 0 and dω = 0.
and then to the
Theorem If ω has at least two continuous derivatives then ω is Harmonic if
and only if △ω = 0.
We prefer to use δω = 0 and dω = 0 as the definition of Harmonic and
△ω = 0 as a consequence because of the number of derivatives necessary for
these two conditions. The definition as it stands is competing with the old
definition △ω = 0; bet on the former.
Next, from the previous equation we have

((△ω, η)) = ((δω, δη)) + ((dω, dη))

= ((ω, dδη)) + ((ω, δdη))
= ((ω, △η))

where we are still assuming that supp(ω) or supp(η) is compact. The equation

((△ω, η)) = ((ω, △η))

with the condition on the supports means that △ is a formally self adjoint
operator. It is standard in elementary mathematical physics to leave things here.
To get real self adjointness requires consideration of the boundary conditions
for the operator, and to deal properly with this requires functional analysis and
Sobolev spaces, which is a can of worms we wish to keep in the can here.
Our final Laplacian duty is the equation ∗ △ ω = △ ∗ ω which we can do with
a lovely little calculation. Recall from the chapter on the codifferential that for
ω ∈ Λr (K)

d∗ω = (−1)r ∗ δω
δ∗ω = (−1)r+1 ∗ dω

Then we have

△∗ω = (dδ + δd) ∗ ω

56 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

= d(δ ∗ ω) + δ(d ∗ ω)
= d((−1)r+1 ∗ dω) + δ((−1)r ∗ δω)
= (−1)r+1 (d ∗ dω) + (−1)r (δ ∗ δω)
= (−1)r+1 (−1)r+1 (∗δdω) + (−1)r (−1)r (∗dδω)
= ∗(δdω + dδω)
= ∗△ω
Notice that in going from step 4 to step 5 that dω ∈ Λr+1 and δω ∈ Λr−1 which
is what makes the signs come out right.

1.22 Maxwell’s Equations in 3-space

In this section we will show how the standard vector analysis treatment of
Maxwell’s equations can be recast in the notation of differential forms. Notation
has been chosen to make it easy to switch over to a tensor treatment, which we
provide in an addendum to the section.
The prerequisites for this section include the ∗ operator, the codifferential
which in three dimensions is δ = ∗d∗ and the Laplacian △ = δd + dδ (math
form) and △ p = −△ (physics form).

A standard form of Maxwell’s equations using vector formulation is

div D = ρ curl E = − 1c ∂B
∂t D = ǫE
1 ∂D 1
div B = 0 curl H = c ∂t + c j B = µH
The form of these equations indicates how they should be put into differential
forms. The divergence is mirrored by d on 2-forms and the curl by d on 1-forms.
This suggests that D and B should be 2 forms and E and H should be 1-forms.
We also note that this suggests the e.g. D and E are not trivial variants of one
another, which was the attitude expressed by Maxwell and Faraday. We set
D = D1 dydz + D2 dzdx + D3 dzdx E = E1 dx + E2 dy + E3 dz
B = B 1 dydz + B 2 dzdx + B 3 dzdx H = H1 dx + H2 dy + H3 dz
Since j is a current density (to be integrated over a surface) it should be a two
form
j = j 1 dydz + j 2 dzdx + j 3 dzdx
and since ρ is a charge density it should be integrated over a region and thus
should be a three form, for which we will use the letter P, an upper case Greek
ρ.
P = ρ Ω0
In rectangular coordinates P = ρ dxdydz. The matter equations can also be
written as forms, for which we require the ∗ operator:
D = ǫ∗E
B = µ∗H
1.22. MAXWELL’S EQUATIONS IN 3-SPACE 57

and then the differential equations become

1 ∂B
dD = P dE = −
c ∂t
1 ∂D 1
dB = 0 dH = + j
c ∂t c
The equations of most significance derived from these are the equation of con-
tinuity and the potential equations. The equation of continuity is easy:

1 ∂D 1
dH = + j
c ∂t c
1 ∂ 1
0 = ddH = dD + dj
c ∂t c
1 ∂ 1
0 = P + dj
c ∂t c
∂P
0 = + dj
∂t

This is the form that the equation of continuity takes in our treatment. We note
however that if we just consider the coefficient of dxdydz from the equation it
becomes
∂ρ
0= + div j
∂t
Now we start on the potential equations.

dB = 0 =⇒ B = dA

(by the converse of the Poincaré lemma), where A is a 1-form. A is not uniquely
determined; we can add dG to A for any G ∈ Λ0 (that is, a function), since

d(A + dg) = dA + ddG = dA + 0 = B

We can use this G to modify A and this is called changing the gauge. We will
return to this matter later on. Next we have
1 ∂
dE = − B
c ∂t  
1 ∂ 1 ∂A
= − dA = − d
c ∂t c ∂t
 
1 ∂A
d E+ = 0
c ∂t
1 ∂A
Since E + c ∂t is a 1-form, there must be a 0-form −φ for which

1 ∂A
E+ = −dφ
c ∂t
58 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

(The minus sign is historical and of no theoretical importance.) φ is called the

scalar potential. Often the letter U is used for it instead of φ. A is called the
vector potential.
If A is changed to A′ = A + dG there will be a corresponding change in φ
which we now determine.
1 ∂A
E+ = −dφ
c ∂t
1 ∂A′
E+ = −dφ′
c ∂t
1 ∂A 1 ∂
E+ + dG = −dφ′
c ∂t c ∂t
1 ∂
−dφ + dG = −dφ′
c ∂t
1 ∂
d(φ′ − φ + G) = 0
c ∂t
1 ∂
φ′ − φ + G = C
c ∂t
where C is some constant. Thus
1 ∂
φ′ = φ − G+C
c ∂t
Digesting, we have

A′ = A + dG
′ 1 ∂
φ = φ− G
c ∂t
where we have set C = 0 as customary.
Our next job is to derive the potential equations. However, there is so much
slop in A that we cannot reasonably expect nice equations without putting some
extra conditions on A. We could simply pull the condition out of the air, but it
will be more fun to see it appear in context.
Recall that the physics Laplacian △ p is the negative of the mathematical

Laplacian △ = dδ + δd where δ in three dimensions is

δω = (−1)r ∗ d ∗ ω ω ∈ Λr

Now we have

△A = (δd + dδ)A = δdA + dδA

= δB + dδA
= (−1)2 ∗ d ∗ B + dδA
= ∗d(µH) + dδA = µ ∗ dH + dδA
 
1 ∂D 1
= µ∗ + j + dδA
c ∂t c
1.22. MAXWELL’S EQUATIONS IN 3-SPACE 59
 
1 ∂(∗D) 1
= µ + ∗ j + dδA
c ∂t c
 
1 ∂(ǫE) 1
= µ + ∗ j + dδA
c ∂t c
µ ǫµ ∂E
= ∗j+ + dδA
c c ∂t  
µ ǫµ ∂ 1 ∂A
= ∗j+ −dφ − + dδA
c c ∂t c ∂t
 
µ ǫµ ∂ 2 A ǫµ ∂φ
= ∗j− 2 + d δA −
c c ∂t2 c ∂t

Rearranging and replacing △A by −△ p A we have

 
ǫµ ∂ 2 A µ ǫµ ∂φ
− △A =
p ∗ j + d δA −
c2 ∂t2 c c ∂t

This would be the familiar wave equation for A

ǫµ ∂ 2 A µ
A = 2 2
−△p A = ∗j
c ∂t c
 
with velocity √cǫµ if not for the term d δA − ǫµ ∂φ
c ∂t . Hence, using the slop in
A, we will set this to 0.

ǫµ ∂φ
δA − =0 Condition of Lorenz
c ∂t
Of course, we need to know that adding a suitable dG will force the Condition
of Lorenz to be true. We will look at this later. (Usually this step is neglected!)
It is also useful to decode the condition of Lorenz. We have
ǫµ ∂φ
(−1)1 ∗ d ∗ A − = 0
c ∂t
ǫµ ∂φ
− ∗ d (A1 dydz + A1 dzdx + A1 dxdy) − = 0
  c ∂t
∂A1 ∂A2 ∂A3 ǫµ ∂φ
−∗ + + dxdydz − = 0
∂x ∂y ∂z c ∂t
 
∂A1 ∂A2 ∂A3 ǫµ ∂φ
− + + + = 0
∂x ∂y ∂z c ∂t

We must also have the potential equation for φ. This is derived in a similar
manner, but easier. It is interesting that it throws up the same extra terms which
we eliminate with the Condition of Lorenz. This suggests that the time and
space variables might have more of a connection than we expect á priori. This
was likely one of the things that induced Lorenz develop the Lorenz-Einstein
transformation equations. We imitate the previous calculation, noting that
60 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

δφ = 0 since δ is 0 on Λ0 ,

△φ = (δd + dδ)φ = δdφ + dδφ

 
1 ∂A
= δ −E − +0
c ∂t
1 ∂(δA)
= −δE −
c ∂t    
1 1 ∂ ǫµ ∂φ 1 ∂ ǫµ ∂φ
= −(−1) ∗ d ∗ E − − δA −
c ∂t c ∂t c ∂t c ∂t
2
1 ǫµ ∂ φ
= ∗d D − 2 2
ǫ c ∂t
ǫµ ∂φ
where we have invoked the Condition of Lorenz δA − c ∂t = 0 once again.
Continuing

1 ǫµ ∂ 2 φ
△φ = ∗P − 2 2
ǫ c ∂t
ǫµ ∂ 2 φ 1 1
−△
p φ = ∗P = ∗ (ρΩ0 )
c2 ∂t2 ǫ ǫ
2
ǫµ ∂ φ 1
−△
p φ = ρ
c2 ∂t2 ǫ
which is the familiar wave equation for φ in coordinate independent form (except
for orientation considerations and provided that ρ really is the physical charge
density and has not been modified to fit into some special coordinate system).
Note that we used △ p φ = − △φ again. Note also that the equation is identical in

form to the Dalembertian equation for A. And finally note that the Condition of
Lorenz did not pop up so naturally in this derivation as it did in the calculation
for A.
Our next job is a look at the Condition of Lorenz. A close examination of
the derivation of the wave equation for A will show that if A is a solution of
the equation then the Condition of Lorenz must hold. The question is, can we
always force the Condition of Lorenz to hold by choosing an appropriate G in
1 ∂G
A′ = A + dG φ′ = φ −
c ∂t
The answer, as we will show, is yes. We need

ǫµ ∂φ′
δA′ − =0
c ∂t
How do we find the G?. Substituting into this equation we have
 
ǫµ ∂ 1 ∂G
δ(A + dG) − φ− = 0
c ∂t c ∂t
1.23. INDEFINITE INNER PRODUCT SPACES 61

Since δG = 0, this can be rewritten as

ǫµ ∂ 2 G ǫµ ∂φ
(δd + dδ)G + + δA − = 0
c2 ∂t2 c ∂t  
ǫµ ∂ 2 G ǫµ ∂φ
+ △G = − δA −
c2 ∂t2 c ∂t
 
ǫµ ∂ 2 G ǫµ ∂φ
−△ p G = − δA −
c2 ∂t2 c ∂t

(recalling that △
p = −△) and once again we have a wave equation. We solve for

G, correct the A and φ to A′ and φ′ , and then solve the wave equations for A′
and φ′ . Actually we don’t have to make the correction, since solving the wave
equations for A and φ will work just fine; the A and φ we find will automatically
satisfy the Condition of Lorenz. The importance of the above is to show that
A and φ we seek actually exist.

1.23 Indefinite Inner Product Spaces

This section and the next contain material of far less general usefulness than
the preceding sections. They are directed specifically to Relativity, and if you
are not interested in this subject then you have little need of these sections.
This first of the two contains the modications necessary when the inner
product is not postive definite, so that there are vectors for which (v, v) < 0.
This causes some modifications in the ∗ operator and some changes in sign
in the Laplacian. In this section we will handle the situation in general and
in the next the specific material for Relativity. We will concentrate again on
Maxwell’s equations as an example. Naturally developing the theory again with
the additional complications makes for a high degree of repetition, and we will
be somewhat more terse this time since you have alrady seen the material once.
Let V be an n-dimensional space on which there is an inner product which
is non-degnenerate. Thus the inner product satisfies the following

a. (αu + βv, w) = α(u, w) + β(v, w) Linearity

b. (v, w) = (w, v) Symmetry

c. If (v, w) = 0 for all w then v = 0 Non-degenarcy

Notice the c. replaces the usual condition (v, v) > 0 for v 6= 0. In indeffinite
inner product spaces there will certainly be vectors for which (v, v) = 0. In
Relativity these are called null vectors and in math isotropic vectors.
Using the usual methods of linear algebra and being careful to avoid vectors
v for which (v, v) = 0 one can construct without difficulty an orthonormal basis
e1 , . . . , en for V . Renumbering if necessary they can be arranged in a sequence
62 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

e1 , . . . , en−s , en−s+1 , . . . , en for which

(ei , ei ) = 1 for i = 1, . . . , n − s
(ei , ei ) = −1 for i = n − s + 1, . . . , n

Here, orthonormal means (ei , ei ) = ±1. Those elements of the basis with
(ei , ei ) = +1 will be called positive basis elements and those elements with
(ei , ei ) = −1 will be called negative basis elements. J. J. Sylverster proved early
in the history of linear algebra (1852) that the number s of negative basis ele-
ments does not depend on the choice of basis; all orthonormal bases will have
the same s. This is called Sylvester’s law of inertia. This s will metastisize
through all the formulas.
It is worth mentioning that since we have assumed that the inner product
is non-degenerate there will indeed by n basis vectors in the orthonormal basis.
For the orthonormal basis defined above we get the usual matrix of metric
coefficients gij = (ei , ej ) and it and its inverse will be
 
1 0 ··· 0 0 0 ··· 0
 0 1 ··· 0 0 0 ··· 0 
 
 .. .. .. .. .. .. .. .. 
 . . . . . . . . 
 
 0 0 ··· 1 0 0 ··· 0 
E=



 0 0 · · · 0 −1 0 ··· 0 
 0 0 · · · 0 0 −1 ··· 0 
 
 .. .. .. . .. .. .. .. 
 . . . .. . . . . 
0 0 ··· 0 0 0 · · · −1

with n − s ones and s negative ones.

As far as the exterior or Grassmann product (wedge product) of the vectors
goes, there will be no changes, since this apparatus works independently of the
inner product. We first begin to get effects from the indefiniteness when we
introduce the ∗ operator.
We will begin work in Rm or some n-dimensional submanifold K of it with
coordinates u1 , . . . , un and an indefinite inner product on the cotangent space
Λ1 (K). At each point the cotangent space is spanned by du1 , , . . . , dun . The
topform will be du1 ∧ . . . ∧ dun , but the inner product of this with itself may
well be negative and thus unsuitable for use in finding volumes. If this should
happen, (which it often does in applications) we put in a negative sign to fix
it. To see the sign, we reason as follows. There is a process in linear algebra
by which we can manipulate du1 , , . . . , dun to give us an orthonormal basis
E 1 , . . . , E n which will have the matrix E above for it’s metric. By reverseing
the sign of E n f necessary we can assure that we can assure that du1 , , . . . , dun
and E 1 , . . . , E n have the same orientation. Suppose that
X
Ei = αij duj
1.23. INDEFINITE INNER PRODUCT SPACES 63

We know that det(αij ) > 0 because the orientations match. Then we have


E = (αij )⊺ (dui , duj ) (αij )

and thus



det E = det (αij )⊺ (dui , duj ) (αij ) = det(αij )⊺ det (dui , duj ) det(αij )
 2
(−1)s = det(αij ) det(g ij )
2
Since [det(α)] is positive (−1)s and det(g ij ) have the same sign and we have
(−1)s det(g ij ) > 0. Thus to normalize the topform du1 ∧ . . . ∧ dun it is natural
to take
1
Ω0 = p du1 ∧ . . . ∧ dun
(−1) det(g ij )
s

If we set, as is commonly done, g = det(gij then we have det(g ij = g −1 and the

formula becomes p
Ω0 = (−1)s g du1 ∧ . . . ∧ dun
There is another useful point. Both Ω0 and E 1 ∧ . . . ∧ E n are normalized
topforms in the one-dimensional space Λn . At worst they could be negatives
of one another, but this cannot happen since they have the same orientation.
Hence we have
E 1 ∧ . . . ∧ E n = Ω0

Our next job is to revise the formula for ∗ ∗ ω. We take E 1 , . . . , E n−s be the
normalized positive basis elements, ((E i , E i ) = +1), and E n−s+1 , . . . , E n to be
the normalized negative basis elements, ((E i , E i ) = −1). We now take a typical
basis element E i1 ∧ . . . ∧ E ir with i1 < i2 < · · · < ir . To this basis element
corresponds a permutation
  
1 2 . . . r  r + 1 r + 2 . . . n
σ=
i1 i2 . . . ir  k1 k2 . . . kn−r

with k1 < k2 < · · · < kn−r . We suspect from our previous work that

∗ E i1 ∧ . . . ∧ E ir = a E k1 ∧ . . . ∧ E kn−r

and we must determine a.

Let s1 of E i1 , . . . , E ir have (E i , E i ) = −1
k1 kn−r
Let s2 of E , . . . , E have (E i , E i ) = −1

where of course s1 + s2 = s. Using the definition of ∗ and our guess above about
the form of ∗ E i1 ∧ . . . ∧ E ir we have


E i1 ∧ . . . ∧ E ir ∧ ∗(E i1 ∧ . . . ∧ E ir ) = det (E ij , E iℓ ) Ω0
64 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

E i1 ∧ . . . ∧ E ir ∧ a E k1 ∧ . . . ∧ E in−r = (−1)s1 Ω0
a sgn(σ)E 1 ∧ . . . ∧ E n = (−1)s1 Ω0
a sgn(σ)Ω0 = (−1)s1 Ω0
a sgn(σ) = (−1)s1
a = (−1)s1 sgn(σ)

and thus

∗ E i1 ∧ . . . ∧ E ir = (−1)s1 sgn(σ) E k1 ∧ . . . ∧ E kn−r

Now we are in a position to find ∗ ∗ ω. Let

  
1 2 . . . n − r  n − r + 1 n − r + 2 . . . n
σ̃ =
k1 k2 . . . kn−r  i1 i2 . . . ir

Then, in a similar fashion to above,

∗ E k1 ∧ . . . ∧ E kn−r = (−1)s2 sgn(σ̃) E i1 ∧ . . . ∧ E ir

Hence

∗ ∗ E i1 ∧ . . . ∧ E ir = (−1)s1 sgn sgn(σ) ∗ (E k1 ∧ . . . ∧ E kn−r )

= (−1)s1 sgn(σ) (−1)s2 sgn(σ̃)E i1 ∧ . . . ∧ E ir
= (−1)s (−1)r(n−r) E i1 ∧ . . . ∧ E ir

since s1 + s2 = s and sgn(σ) sgn(σ̃) = (−1)r(n−r) . Since any element ω ∈ Λr

can be expressed in terms of these basis elements E i1 ∧ . . . ∧ E ir we have

∗ ∗ ω = (−1)r(n−r)+s ω for ω ∈ Λr

If we now write this using differentials as the basis we have

∗ dui1 ∧ . . . ∧ duir = (−1)s1 sgn(σ) duk1 ∧ . . . ∧ dukn−r

where the notation is that used above.

Our next job is the codifferential δ. Because the inner product is no longer
positive definite there are some odd effects and operator which corresponds to
the Laplacian is now the D’Albertian. We will return to this later. For the
moment we will be interested in the formula for the codifferential, which we
derive in a similar manner to the positive definite case. Let ω ∈ Λr−1 and
η ∈ Λr ). Recall that

d(ω ∧ ∗η) = dω ∧ ∗η + (−1)r−1 ω ∧ d ∗ η

Then
Z
((dω, η)) = dω ∧ ∗η
K
1.24. MAXWELL’S EQUATIONS IN SPACE-TIME 65
Z Z
= d(ω ∧ ∗η) − (−1)r−1 ω∧d∗η
K K
Z Z
= ω ∧ ∗η − (−1)r−1 ω∧d∗η
∂K K
Z
= 0 + (−1)r ω∧d∗η
K

by Stoke’s theorem, because we are assuming boundary conditions that kill off
the boundary intergral, as is usual in this kind of calculation. Continuing we
have
Z
r (n−r+1)(r−1)+s
((dω, η)) = (−1) (−1) ω∧∗∗d∗η
K
Z
= (−1)n(r−1)+s+1 ω∧∗∗d∗η
K
Z h i
= ω ∧ ∗ (−1)n(r−1)+s+1 ∗ d ∗ η
K
= ((ω, (−1)n(r−1)+s+1 ∗ d ∗ η))

So if we define
δη = (−1)n(r−1)+s+1 ∗ d ∗ η

we have
((dω, η)) = ((ω, δη))

Finally we want to introduce the analog of the Laplacian. Since the situation
is so different from that of positive definite inner products, it seems reasonable
to use a different notation, especially since it (more or less) coincides with that
usual in physics. We will use the notation  ω for the new operator, which is
defined exactly as was the Laplacian in the positive defintate case, namely
Def  ω = (dδ + δd)ω
The notation is meant to suggest a relationship with the classical Dalember-
tian
1 ∂2f ∂2f ∂2f ∂2f
p
f= 2
− 2
− 2 − 2
c ∂t ∂x ∂y ∂z

1.24 Maxwell’s equations in Space-Time

In this section we will use the material in the preceding section to work out
Maxwell’s equations in Space-Time. We will do this twice. First we will work
out the the theory in free space where we assume ǫ = µ = 1 and then we will
work it out in general without the constraings on ǫ and µ. As is the preceding
section this is highly specialized material and needed by a small minority of
readers.
66 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

For Special Relativity, the coordinate system cdt, dx, dy, dz forms an oth-
onormal coordinate system, with that order. Thinking of it as dx0 , dx1 , dx0 , dx1
for convenience, we have

(cdt, cdt) = +1, (dx, dx) = −1, (dy, dy) = −1, (dz, dz) = −1,

and
(dxi , dxj ) = 0 for i 6= j
The matrix for this coordinate system is
 
1 0 0 0

 0 −1 0 0 
ij i j 
(g ) = (dx , dx ) =  
0 0 −1 0 
0 0 0 −1
We can perform the ∗ operation using the following permutations, all of
which are even permutations (have sgn(σ) = +1).
     
0 1 2 3 0 1 2 3 0 1 2 3
0 1 2 3 0 2 3 1 0 3 1 2
     
0 1 2 3 0 1 2 3 0 1 2 3
2 3 0 1 3 1 0 2 1 2 0 3
Using these and the formula

∗ dxi1 ∧ . . . ∧ dxir = (−1)s1 sgn(σ) dxj1 ∧ . . . ∧ dxjr

where  
1 2 ··· r r+1 r + 2 ··· n
σ=
i1 i2 · · · ir j1 j2 · · · jn−r
and s1 is the number of negative basis elements (that is dx, dy, dz) among
dxi1 , . . . , dxir , we have the following formulas

∗1 = cdt ∧ dx ∧ dy ∧ dz
∗ cdt = dx ∧ dy ∧ dz ∗ dx ∧ dx ∧ dx = cdt
∗ dx = cdt ∧ dy ∧ dz ∗ cdt ∧ dy ∧ dz = dx
∗ dy = cdt ∧ dz ∧ dx ∗ cdt ∧ dz ∧ dx = dy
∗ dz = cdt ∧ dx ∧ dy ∗ cdt ∧ dx ∧ dy = dz
∗ cdt ∧ dx = −dy ∧ dz ∗ dy ∧ dz = cdt ∧ dx
∗ cdt ∧ dy = −dz ∧ dx ∗ dz ∧ dx = cdt ∧ dy
∗ cdt ∧ dz = −dx ∧ dy ∗ dy ∧ dz = cdt ∧ dx
∗ cdt ∧ dx ∧ dy ∧ dz = −1
For example, for the 2nd entry we have the permutation
 
0 1 2 3
σ=
0 1 2 3
1.24. MAXWELL’S EQUATIONS IN SPACE-TIME 67

with s1 = 0 and sgn(σ) = +1, whereas for the fourth entry we have
 
0 1 2 3
σ=
2 0 3 1

with s1 = 1 and sgn(σ) = −1, this being the the fourth permatation in the
above list with the second and third entries swapped. For the seventh entry we
have  
0 1 2 3
σ=
0 2 3 1
so s1 = 1 and sgn(σ) = 1, this being the second permutation in the list. The
entries in the second column can be derived from those in the first, but beware
since the permutation is reversed. The fourth row second column has
 
0 1 2 3
σ=
0 3 1 2

and here s1 = 2 and σ is the third entry in the list of permatations so sgn(σ) = 1.
It is far easier to derive the second column using ∗ ∗ ω = (−1)r(4−r)+3 ω.
Next we are interested in the codifferential δ. We have, as usual, with wedges
omitted for ease of reading,

A = −φ cdt + A1 dx + A2 dy + A3 dz

Then

δA = ∗ d ∗ A = ∗ d(−φ dxdydz + A1 cdtdydz + A2 cdtdxdy + A3 cdtdzdx)

 
1 ∂φ ∂A1 ∂A2 ∂A3
= ∗ − + + + cdtdxdydz
c ∂t ∂x ∂y ∂z
1 ∂φ ∂A1 ∂A2 ∂A3
= − − −
c ∂t ∂x ∂y ∂z
Cognoscenti will immediately recognize that the familiar condition of Lorenz,
div A − 1c ∂φ
∂t = 0, is here expressed by δA = 0.
Next we will compute the d’Alembertian  f of a function. We have δf = 0
so

f = (δd + dδ)f = δdf + 0

 
1 ∂f ∂f ∂f ∂f
= ∗d∗ cdt + dx + dy + dz
c ∂t ∂x ∂y ∂z
 
1 ∂f ∂f ∂f ∂f
= ∗d dxdydz + cdtdydz + cdtdzdx + cdtdxdy
c ∂t ∂x ∂y ∂z
 
1 ∂2f ∂ 2f ∂2f ∂2f
= ∗ − − − cdtdxdydz
c2 ∂t2 ∂x2 ∂y 2 ∂z 2
1 ∂2f ∂2f ∂2f ∂2f
= − 2 2 + + +
c ∂t ∂x2 ∂y 2 ∂z 2
68 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

The minus signs come from swapping differentials around and sometimes from
the action of ∗. This is the negative of the classical d’Albertian
1 ∂2f ∂2f ∂2f ∂2f
p f =
 − − −
c2 ∂t2 ∂x2 ∂y 2 ∂z 2
An obnoxious calculation shows that if

A = A0 cdt + A1 dx + A2 dy + A3 dz

then
 A = ( A0 )cdt + ( A1 )dx + ( A2 )dy + ( A3 )dz
We know of no intelligent way to prove this but in view of its importance we
will do the calculation in an appendix to this chapter, which should be skimmed
from a distance.
We now want to apply this equipment to the problem of potential equations
in four dimensional Space-Time. We first consider the case when ǫ = µ = 1
where everything is simple. When these are not 1, there are certain complica-
tions which we will discuss later.
For ease of reference we repeat here Maxwell’s equations.

div D = ρ curl E = − 1c ∂B
∂t D = ǫE
1 ∂D 1
div B = 0 curl H = c ∂t + c j B = µH

along with
1 ∂A
curl A = B dφ = −E −
c ∂t
Let us set as is usual F = dA. We then have

A = −φcdt + A1 dx + A2 dy + A3 dz
     
1 ∂A1 ∂φ 1 ∂A2 ∂φ 1 ∂A3 ∂φ
F = dA = + cdtdx + + cdtdy + + cdtdz
c ∂t ∂x c ∂t ∂y c ∂t ∂z
     
∂A3 ∂A2 ∂A1 ∂A3 ∂A2 ∂A1
+ − dydz + − dzdx + − dxdy
∂y ∂z ∂z ∂x ∂x ∂y
= −E1 cdtdx − E2 cdtdy − E3 cdtdz
+B1 dydz + B2 dzdx + B3 dxdy

Next we take the ∗ of both sides. We remember that here we have ǫ = 1 and
µ = 1 so that E = D and B = H. Then

F̃ = ∗F = E1 dydz + E2 dzdx + E3 dxdy + B1 cdtdx + B2 cdtdy + B3 cdtdz

= D1 dydz + D2 dzdx + D3 dxdy + H1 cdtdx + H2 cdtdy + H3 cdtdz
     
1 ∂D1 ∂H3 ∂H2 1 ∂D2 ∂H1 ∂H3
d∗F = − − cdtdydz + − − cdtdzdx
c ∂t ∂y ∂z c ∂t ∂z ∂x
    
1 ∂D3 ∂H2 ∂H1 ∂D1 ∂D2 ∂D3
+ − − cdtdxdy + + + dxdydz
c ∂t ∂x ∂y ∂x ∂y ∂z
1.24. MAXWELL’S EQUATIONS IN SPACE-TIME 69

1 1 1
= ρ dxdydz − j1 cdtdydz − j2 cdtdzdx − j3 cdtdxdy
c c c
1 1 1
δF = ∗ d ∗ F = ρ cdt − j1 dx − j2 dy − − j3 dz
c c c

Now the coup; recall that we may take δA = 0 (the condition of Lorenz) so that
we have

 A = (δd + dδ)A = δdA + 0

1 1 1
= δF = ρ cdt − j1 dx − j2 dy − − j3 dz
c c c
Using the fact that in rectangular coordinates we have

 A = −( φ)cdt + ( A1 )dx + ( A2 )dy + ( A3 )dz

we have, with the classical d’Alembertian

1 ∂ 2 Ai ∂ 2 Ai ∂ 2 Ai ∂ 2 Ai
p A =
 i − − − = − Ai
c2 ∂t2 ∂x2 ∂y 2 ∂z 2
we have the potential equations in four space

p φ = ρ


p A =
1
 i ji
c
which is the same result we got in the section on Maxwell’s equations in three
dimensional space.
We now want to extend this result to the case where ǫ or µ are not one.
This is surprisingly difficult. However, we suspect it can be done because the
corresponding equations in three space have the same form for the potential φ
and the vector potential components Ai . Recall that the form of the equation
is
ǫµ ∂ 2 Ai ∂ 2 Ai ∂ 2 Ai ∂ 2 Ai µ
− − − = ji
c2 ∂t2 ∂x2 ∂y 2 ∂z 2 c
This suggests an electromagnetic wave moving at a speed of k = √cǫµ . Now we
must steer the boat, like Odysseus, between Charybdis and Skilla. Charybdis
refers to the fact that for the d’Alembertian to come out with k12 = ǫµ c2 in it,
we are going to have to modify the ∗ operator by changing c to k. Skilla refers
to the fact that Maxwell’s equations have c not k in them. Thus replacing all
the c’s by k’s won’t work; we must steer more subtlely. Nevertheless the fact
that the equations for φ and Ai have the same form in 3-space suggests that it
is possible to navigate successfully, which we now do.
In the following calculations the ∗ operator uses the same equations as before
but the constant c in those equations is replaced by the constant k. Except for
70 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

some trivial algebra the calculation goes as before; the potential form A is, as
√
before, recalling that k = c/ ǫµ,

A = −φ cdt + A1 dx + A2 dx + A3 dx
c
A = − φ kdt + A1 dx + A2 dx + A3 dx
k
√
A = − ǫµ φ kdt + A1 dx + A2 dx + A3 dx

First we go for the codifferential δA,

√
∗A = − ǫµ φ dxdydz + A1 kdtdydz + A2 kdtdzdx + A3 kdtdxdy
 √ 
ǫµ ∂φ ∂A1 ∂A2 ∂A3
d∗ A = − − − − kdtdxdydz
k ∂t ∂x ∂y ∂z
ǫµ ∂φ ∂A1 ∂A2 ∂A3
δA = ∗ d ∗ A = + + + = 0
c ∂t ∂x ∂y ∂z
by the condition of Lorenz, which we are assuming, as before. One of the the
positive aspects of the four dimensional treatment is that the condition of Lorenz
is so simply expressed: δA = 0
Next we have, and note we here use c not k so that we can use Maxwell’s
equations,

F = dA     
1 ∂A1 ∂φ 1 ∂A2 ∂φ 1 ∂A3 ∂φ
= + cdtdx + + cdtdy + + cdtdy
c ∂t ∂x c ∂t ∂y c ∂t ∂z
     
∂A3 ∂A2 ∂A1 ∂A3 ∂A2 ∂A1
+ − dydz + − dzdx + − dxdy
∂y ∂z ∂z ∂x ∂x ∂y
= −E1 cdtdx − E2 cdtdy − E3 cdtdz
+B1 dydz + B2 dzdx + B3 dxdy
√
= ǫµ (−E1 kdtdx − E2 kdtdy − E3 kdtdz)
+B1 dydz + B2 dzdx + B3 dxdy
√
∗F = ǫµ (E1 dydz + E2 dzdx + E3 dxdy)
+B1 kdtdx + B2 kdtdy + B3 kdtdz
√
ǫµ
= (D1 dydz + D2 dzdx + D3 dxdy)
ǫ
µ
+ √ (H1 cdtdx + H2 cdtdy + H3 cdtdz)
ǫµ
r
µ
= F̃
ǫ

where
1.25. ENERGY FLOW AND ENERGY DENSITY 71

r
ǫ
F̃ = ∗F
µ
= H1 cdtdx + H2 cdtdy + H3 cdtdz
+D1 dydz + D2 dzdx + D3 dxdy
Now we find dF̃ ,
     
1 ∂D1 ∂H3 ∂H2 1 ∂D2 ∂H1 ∂H3
dF̃ = − − cdtdydz + − − cdtdzdx
c ∂t ∂y ∂z c ∂t ∂z ∂x
    
1 ∂D3 ∂H2 ∂H1 ∂D1 ∂D2 ∂D3
+ − − cdtdxdy + + + dxdydz
c ∂t ∂x ∂y ∂x ∂y ∂z
1
= ρ dxdydz − (j1 cdtdydz + j2 cdtdzdx + j3 cdtdxdy)
c
√
ǫµ
= ρ dxdydz − (j1 kdtdydz + j2 kdtdzdx + j3 kdtdxdy)
c
Finally, we have
A = (δd + dδ)A
= δdA + 0 (conditon of Lorenz δ A = 0)
= δF r = ∗ d ∗ Fr
µ µ
= ∗d F̃ = ∗ dF̃
ǫ ǫ
r  √ 
µ ǫµ
= ∗ ρ dxdydz − (j1 kdtdydz + j2 kdtdzdx + j3 kdtdxdy)
ǫ c
r  √ 
µ ǫµ
= ρ kdt − (j1 dx + j2 dy + j3 dz)
ǫ c
r  √ 
µ 1 ǫµ
= √ ρ cdt − (j1 dx + j2 dy + j3 dz)
ǫ ǫµ c
ρ µ
= cdt − (j1 dx + j2 dy + j3 dz)
ǫ c
p = −, to
This decodes in the usual way, with 
1 ∂ 2 φ ∂ 2 φ ∂ 2φ ∂ 2 φ ρ
p φ
 = 2 2
− 2
− 2 − 2 =
c ∂t ∂x ∂y ∂z ǫ
1 ∂ 2 Ai ∂ 2 Ai ∂ 2 Ai ∂ 2 Ai µ
p A
 i = 2 2
− 2
− 2
− 2
= ji
c ∂t ∂x ∂y ∂z c

1.25 Energy Flow and Energy Density

The force on a particle of charge e and velocity ~v is
F~ = e(E
~ + ~v × B)
~
72 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS

~ + d~ℓ is
~ to R
Thus the work done in going from R

dW = F~ · d~ℓ = e(E ~ · d~ℓ

~ + ~v × B)
= e(E~ + ~v × B)
~ · ~v dt
~ · ~v dt
= eE

Thus
dW ~ · ~v
= eE
dt
Now let ρ be a continuous charge distribution. Then, as usual, we replace e by
ρ dt and integrate to find dW/dt. We a also need ~j = ρ~v . The we have
Z
dW ~ · ~v
= (ρ dτ )E
dt
ZM
= ~ · (ρ~v ) dτ
E
M
Z
= ~ · ~j dτ
E
M

Rewriting in terms of differential forms we have

Z
dW
= E ∧j
dt M

Now we can eliminate j by using the Maxwell equation

1 ∂D 1
dH = + j
c ∂t c
which is Ampère’s law. Thus
1 1 ∂D
j = − + dH
c Zc ∂t
dW 1
= c E∧ j
dt c
ZM
1 ∂D

= c E∧ − + dH
M c ∂t
For the next bit of the calculation we need the formula
∂D 1 ∂
E∧ = (E ∧ D)
∂t 2 ∂t
We use D = ǫ ∗ E and the general formula for differential forms λ ∧ ∗µ = µ ∧ ∗λ.
We then have
∂D ∂
E∧ = E∧ (ǫ ∗ E)
∂t ∂t
∂
= ǫ E ∧ (∗E)
∂t
1.25. ENERGY FLOW AND ENERGY DENSITY 73

∂E
= ǫE ∧ ∗
∂t
∂E
= ǫ ∧ ∗E
∂t
∂E
= ∧D
∂t
Thus
1 ∂ 1  ∂E ∂D 
(E ∧ D) = ∧D+E∧
2 ∂t 2 ∂t ∂t
1  ∂D ∂D 
= E∧ +E∧
2 ∂t ∂t
∂D
= E∧
∂t
In an exactly similar calculation we have
∂B 1 ∂
∧H = (B ∧ H)
∂t 2 ∂t
Resuming the previous calculation we have
Z
dW 1 ∂D

= c E∧ − + dH
dt c ∂t
ZM
1 ∂D
= c −E ∧ + E ∧ dH
M c ∂t
Z
1 ∂D

= c −E ∧ + dE ∧ H − d(E ∧ H)
c ∂t
ZM
1 ∂D 1 ∂B

= c −E ∧ + − ∧ H − d(E ∧ H)
c ∂t c ∂t
ZM
1 ∂ 1 ∂
= c − (E ∧ D) − (B ∧ H) − d(E ∧ H)
M 2c ∂t 2c ∂t
Z Z
∂ 1 1
= − (E ∧ D) − (B ∧ H) − c E∧H
∂t M 2 2 ∂M

by Stokes’ theorem. If we define the Energy Density u of the field as

1
u = − (E ∧ d + B ∧ H)
2
and the Flux of energy through the surface ∂M as

S = cE ∧H Poyting’s Flux Form

we can interpret the previous equation as the change in the Energy as due to
the decrease in the field energy less the flux of energy through the boundary.
74 CHAPTER 1. INTRODUCTION AND BASIC APPLICATIONS
Chapter 2

Mathematical Theory

75
76 CHAPTER 2. MATHEMATICAL THEORY

2.1 INTRODUCTION
This chapter has two purposes. First we must prove the two major theorems
of elementary differential forms, and second we want to extend the theory with
some deeper methods. For example, we would like to have formulas for the ∗
operator on manifolds, where orthonormal coordinates systems are usually not
possible, so we need some more flexible methods for that.
We also want discuss the mathematical theories that underlie the develop-
ment in the first chapter. Also, we will look at the theory in a more general
setting; that of a differentiable manifold rather than simply n-space. A dif-
ferentiable manifold is like a surface in n-space but with more dimensions and
without the n-space. If the manifold is actually inside an n-space the manifold
is called an embedded manifold. In point of fact, all manifolds can be embed-
ded in an n-space of sufficiently high dimension 1 , but it is no longer customary
to do this. Also it may not be natural to do this; few people think that the
four dimensional Space-Time of our universe is embedded in a five (or more)
dimensional Euclidean space Rn .

2.2 Manifolds Mappings and Pullbacks

In order to have a firm basis on which to proceed it would be wise to spend a bit
of time on some technical material of great importance for modern differential
geometry and everything that uses it. This material is not very deep but requires
a bit of thought. First we want to give a good definition of manifold. Then we
want to discuss mappings f : X → Y between manifolds and how a differential
form on Y may be ”pulled back” to X using f . This material is useful in the
sections that follow, as well as form a great many other things, and should be
studied carefully.

2.3 Proofs of Stokes Theorem

We will present in this chapter two proofs of Stokes theorem. The first is a
low level proof using the familiar methods of Advanced Calculus, as seen in a
third semester Calculus course. The second using more modern mathematical
machinery, in particular partitions of unity, which are now common in advanced
mathematics.

2.4 Proofs of the Converse of the Poincare Lemma

We will present two proofs of the Converse of the Poincare Lemma. The first
uses familiar material from Advanced calculus. The second uses slightly more
advanced material but is very interesting because a borrows the idea of chain
1 This is accomplished by the Whitney or Nash embedding theorems, depending on the

circumstances
2.5. PERMUTATIONS 77

homotopy from Algebraic Topology, although no previous knowledge of this is

necessary.

2.5 Permutations
One cannot get very far using differential forms without the dualizing operator
∗ which we saw used in the previous section for deriving formulas for curvilinear
coordinates. In this section we will give an introduction to the ∗ operator which
will be more of a tour than a treatment; the full treatment will be found in
Chapter 2. Thus this section will present the basic ideas and the formulas the
physicist will find most useful. Formulas for more sophisticated uses will be
found in Chapter 2.
A critical and not particularly well liked part of this is some of the lore of
permutations. A permutation is a rearrangement of the set {1, 2, . . . , n}; for ex-
ample {3, 5, 6, 1, 2, 4} is a rearrangement of {1, 2, 3, 4, 5, 6}. By interchanging two
elements, adjacent or not, a permutation may be brought back to {1, 2, . . . , n}.
There are many ways to do this of course, but it turns out that the number
of interchanges to do this is always the same mod 2. Thus no matter how the
interchanges are done, (−1)s will always have the same value, and that value
is the sign of the permutation. It is convenient for us (though somewhat non-
standard) to write a permutation σ as a function with inputs at the top and
outputs at the bottom. Thus the permutation above would be written
 
1 2 3 4 5 6
σ =
3 5 6 1 2 4
Here are a possible sequence of interchanges to get back to the identity.
   
1 2 3 4 5 6 1 2 3 4 5 6
−→ −→
3 5 4 1 2 6 3 2 4 1 5 6
   
1 2 3 4 5 6 1 2 3 4 5 6
−→
1 2 4 3 5 6 1 2 3 4 5 6
We used four interchanges; s = 4. Clearly this can be done in many ways but
the number of interchanges s will always be an even number, and thus
sgn(σ) = (−1)s = +1
is well defined.
There is one other useful observation. Suppose that a permutation has the
property that it compounded of two increasing subsequences of the numbers
{1, 2, . . . , n}; for example
  
1 2 3 4  5 6 7 8 9
3 5 7 8  1 2 4 6 9
or more generally with i1 < . . . < ir and ir+1 < . . . < in
  
1 . . . r  r + 1 . . . n
σ=
i1 . . . ir  ir+1 . . . in
78 CHAPTER 2. MATHEMATICAL THEORY

Note that the elements after the vertical line are determines by those before it.
Now the interesting thing here is that the sign of such a permutation can be
found as follows. First set
r(r + 1)
Tr = 1 + 2 + · · · + r =
2
Then for such a σ Pr
sgn(σ) = (−1) j=1 ij −Tr

Thus for the example above

sgn(σ) = (−1)3+5+7+8 −T4 = (−1)23−10 = −1

Since this kind of permutation is the most common type in differential forms,
this is quite handy.

2.6 The operator Φ

In this mathematical appendix we provide mathematical justification for many
of the things that were skipped over in the foregoing sections. For example, in
some of the sections we showed how differential forms ”mimicked” the action of
curl on vectors. In this section we show how this can be made mathematically
precise.
There are two coordinate independent operators whose theory we must de-
velop. We do not do this in complete generality, but limit ourself to our particu-
lar needs. Nevertheless, we do enough so that anyone who reads and understands
this appendix will find large areas of differential geometry familiar.
The first operator, which we call Φ, sets up an isomorphism between a vector
space V with an inner product and its dual space, the space of linear functionals
on V , which we call V ∗ . This is quite easy and is merely an invariant form of
the raising and lowering of indices used in Tensor Analysis.
The second operator is the ∗ operator, invented by Heinrich Grassmann and
popularized by William V. D. Hodge. This is a little more difficult and requires
introduction of an inner product on the dual space and also the introduction of
a normalized topform, which we explain below.
Let V be a n-dimensional vector space with an inner product, which we write
(v, w) rather than v · w, and let e1 , e2 , . . . , en be a basis. We set

gij = (ei , ej )
P i
Then v = i v ei for any v ∈ V but, following Einstein, we will omit the
sum sign when an index is repeated, one up one down, and we will write this
v = v i ei . Similarly w = wi ei and we have

(v, w) = (v i ei , wj ej )
= v i wj (ei , ej )
= gij v i wj
2.6. THE OPERATOR Φ 79

This connects (v, w) whose value does not depend on a choice of basis with
quantities gij , v i and wj whose values do depend upon the basis. It is important
to be able to go back and forth between coordinate free notation and indexed
notation where the objects depend upon the choice of basis. Tensor notation is
an attempt to live in both worlds, and it does it quite well, at the expense of
rather complex looking equations.
Because an inner product is non-degenerate, (which means that if (v, w) = 0
for all w ∈ V then v = 0) we must have
det(gij ) 6= 0
This is most important.
A linear functional ℓ is defined by the requirement that
ℓ(αv + βw) = α ℓ(v) + β ℓ(w)
The set of all linear functionals is a vector space and is called V ∗ .
An important theorem is
Theorem Representation of linear functional.
Let ℓ be a linear functional on a vector space with an inner product. There
there exists a unique u ∈ V for which
ℓ(v) = (u, v)
We will prove this shortly. First we give an important example. If v = v i ei
the we can define a linear functional whose value on v is v i . This functional is
called Ei (note upper index). It should be clear that

1 if i = j
Ei (ej ) = δji =
0 if i 6= j
Now let ℓ ∈ V ∗ be a linear functional and let λi = ℓ(ei ). Then I claim that
ℓ = λi Ei . Indeed
λi Ei (v) = λi Ei (v j ej )
= λi v j Ei (ej )
= λi v j δji
= λj v j
= v j ℓ(ej )
= ℓ(v j ej )
= ℓ(v)
Then it is easy to prove that the Ei are linearly independent and thus form a
basis for V ∗ .
Now back to Φ : V → V ∗ . Let ℓ ∈ V ∗ be a linear functional and let it be
represented by u ∈ V . Then the theorem says that for all v ∈ V we have
ℓ(v) = (u, v)
80 CHAPTER 2. MATHEMATICAL THEORY

We define Φ by
Φ(u) = ℓ
Since this has been defined in terms of the inner product, which does not depend
upon the basis, Φ is independent of the basis.
For our next task we need to define

(g ij ) = (gij )−1

When these matrices are written out the inverse relationship becomes

gij g jk = δik
ij
g gjk = δki

Also note that gij = (ei , ej ) = (ej , ei ) = gji the order of the indices in the
above equations is not important.
It is now time to prove the theorem. Since we would like to know the
connection between ℓ and u in coordinates, it makes sense to prove the theorem
using coordinates. Let ℓ = λj Ej and define u by

ui = g ij λj and u = u i ei

We now verify this works:

(u, v) = (ui ei , v k ek )
= gik ui v k
= gik g ij λj v k
= gki g ij λj v k
= δkj λj v k
= λk v k
= λk Ek (v)
= ℓ(v)

Then the formula for Φ in coordinates is

Φ(u) = ℓ if and only if gij uj = λi

and of course

Φ−1 (ℓ) = u if and only if g ij λj = ui

Our next job is to put an inner product on V ∗ and we will do this by exporting
the inner product on V to V ∗ in such a way as to make Φ an isometry. That is
we define, for λ, µ ∈ V ∗ ,
Def
(λ, µ) = (Φ−1 (λ), Φ−1 (µ))
2.6. THE OPERATOR Φ 81

Now to do this is coordinates we have, with λ = λk Ek and µ = µn En

(λ, µ) = gij g ik λk g jm µm
= δjk λk g jm µm
= λj g jm µm
= g jm λj µm

which is the desired formula for the inner product in coordinates.

82 CHAPTER 2. MATHEMATICAL THEORY
Bibliography

[1] Misner, Thorne and Wheeler, GRAVITATION, W.H. Freeman, San

Francisco, 1973

[2] Frankel, Theodore THE GEOMETRY OF PHYSICS, 3rd edition,

Cambridge University Press, Cambridge,etc, 2012
[3] Gilkey, Peter B. INVARIANCE THEORY, THE HEAT EQUATION
AND THE ATIZYA-SINGER INDEX THEOREM, 2nd Edition Boca
Raton 1995

[4] Nakahara, Mikio GEOMETRY, TOPOLOGY, AND PHYSICS, Insti-

tute of Physics Publishing, London, 2003
[5] Lovelock, David and Rund, Hanno TENSORS, DIFFERENTIAL
FORMS AND VARIATIONAL PRINCIPLES, Wiley Interscience
[6] Roe, John, ELLIPTIC OPERATORS, TOPOLOGY AND ASYMP-
TOTIC METHODS, 2nd Edition, Boca Raton, 1998

[7] Rosenberg, S. THE LAPLACIAN ON A RIEMANNIAN MANI-

FOLD, Cambridge, 1997
[8] Schulz, William, THEORY AND APPLICATION OF GRASSMANN
ALGEBRA, http://www.cefns.nau.edu/ schulz/grassmann.pdf

83