Manifolds and Forms
Uploaded by
tempestaManifolds and Forms
Uploaded by
tempestaManifolds and Differential Forms
Reyer Sjamaar
D EPARTMENT OF M ATHEMATICS , C ORNELL U NIVERSITY, I THACA , N EW Y ORK
14853-4201
E-mail address:
URL: !!!"# $ % & '
Last updated: 2006-08-26T01:13-05:00
Copyright © Reyer Sjamaar, 2001. Paper or electronic copies for personal use may
be made without explicit permission from the author. All other rights reserved.
Contents
Preface v
Chapter 1. Introduction 1
1.1. Manifolds 1
1.2. Equations 7
1.3. Parametrizations 9
1.4. Configuration spaces 9
Exercises 13
Chapter 2. Differential forms on Euclidean space 17
2.1. Elementary properties 17
2.2. The exterior derivative 20
2.3. Closed and exact forms 22
2.4. The Hodge star operator 23
2.5. div, grad and curl 24
Exercises 27
Chapter 3. Pulling back forms 31
3.1. Determinants 31
3.2. Pulling back forms 36
Exercises 42
Chapter 4. Integration of 1-forms 47
4.1. Definition and elementary properties of the integral 47
4.2. Integration of exact 1-forms 49
4.3. The global angle function and the winding number 51
Exercises 53
Chapter 5. Integration and Stokes’ theorem 57
5.1. Integration of forms over chains 57
5.2. The boundary of a chain 59
5.3. Cycles and boundaries 61
5.4. Stokes’ theorem 63
Exercises 64
Chapter 6. Manifolds 67
6.1. The definition 67
6.2. The regular value theorem 72
Exercises 77
Chapter 7. Differential forms on manifolds 81
iii
iv CONTENTS
7.1. First definition 81
7.2. Second definition 82
Exercises 89
Chapter 8. Volume forms 91
8.1. n-Dimensional volume in R N 91
8.2. Orientations 94
8.3. Volume forms 96
Exercises 100
Chapter 9. Integration and Stokes’ theorem on manifolds 103
9.1. Manifolds with boundary 103
9.2. Integration over orientable manifolds 106
9.3. Gauß and Stokes 108
Exercises 109
Chapter 10. Applications to topology 113
10.1. Brouwer’s fixed point theorem 113
10.2. Homotopy 114
10.3. Closed and exact forms re-examined 118
Exercises 122
Appendix A. Sets and functions 125
A.1. Glossary 125
A.2. General topology of Euclidean space 127
Exercises 127
Appendix B. Calculus review 129
B.1. The fundamental theorem of calculus 129
B.2. Derivatives 129
B.3. The chain rule 131
B.4. The implicit function theorem 132
B.5. The substitution formula for integrals 133
Exercises 134
Bibliography 137
The Greek alphabet 139
Notation Index 141
Index 143
Preface
These are the lecture notes for Math 321, Manifolds and Differential Forms,
as taught at Cornell University since the Fall of 2001. The course covers mani-
folds and differential forms for an audience of undergraduates who have taken
a typical calculus sequence at a North American university, including basic lin-
ear algebra and multivariable calculus up to the integral theorems of Green, Gauß
and Stokes. With a view to the fact that vector spaces are nowadays a standard
item on the undergraduate menu, the text is not restricted to curves and surfaces
in three-dimensional space, but treats manifolds of arbitrary dimension. Some
prerequisites are briefly reviewed within the text and in appendices. The selec-
tion of material is similar to that in Spivak’s book [Spi65] and in Flanders’ book
[Fla89], but the treatment is at a more elementary and informal level appropriate
for sophomores and juniors.
A large portion of the text consists of problem sets placed at the end of each
chapter. The exercises range from easy substitution drills to fairly involved but,
I hope, interesting computations, as well as more theoretical or conceptual prob-
lems. More than once the text makes use of results obtained in the exercises.
Because of its transitional nature between calculus and analysis, a text of this
kind has to walk a thin line between mathematical informality and rigour. I have
tended to err on the side of caution by providing fairly detailed definitions and
proofs. In class, depending on the aptitudes and preferences of the audience and
also on the available time, one can skip over many of the details without too much
loss of continuity. At any rate, most of the exercises do not require a great deal of
formal logical skill and throughout I have tried to minimize the use of point-set
topology.
This revised version of the notes is still a bit rough at the edges. Plans for
improvement include: more and better graphics, an appendix on linear algebra, a
chapter on fluid mechanics and one on curvature, perhaps including the theorems
of Poincaré-Hopf and Gauß-Bonnet. These notes and eventual revisions can be
downloaded from the course website at
( ))*+-,,/...0#132 )(0-4567 899:08;<, =>/?213226,4 92 >>8>,@A3B, CD7 ;8 E$0F()1G9
.
Corrections, suggestions and comments will be received gratefully.
Ithaca, NY, 2006-08-26
v
CHAPTER 1
Introduction
We start with an informal, intuitive introduction to manifolds and how they
arise in mathematical nature. Most of this material will be examined more thor-
oughly in later chapters.
1.1. Manifolds
Recall that Euclidean n-space Rn is the set of all column vectors with n real
entries IJ LNM
J x1 M
J M
x2
xH . ,
K .. O
xn
which we shall call points or n-vectors and denote by lower case boldface letters. In
R2 or R3 we often write
I L
x
x
x HQP , resp. x H K yO .
yR
z
For reasons having to do with matrix multiplication, column vectors are not to be
confused with row vectors S x 1 x2 . . . xn T . For clarity, we shall usually separate
the entries of a row vector by commas, as in S x 1 , x2 , . . . , xn T . Occasionally, to save
space, we shall represent a column vector x as the transpose of a row vector,
x HUS x1 , x2 , . . . , xn T T .
A manifold is a certain type of subset of Rn . A precise definition will follow
in Chapter 6, but one important consequence of the definition is that a manifold
has a well-defined tangent space at every point. This fact enables us to apply the
methods of calculus and linear algebra to the study of manifolds. The dimension of
a manifold is by definition the dimension of its tangent spaces. The dimension of
a manifold in Rn can be no higher than n.
Dimension 1. A one-dimensional manifold is, loosely speaking, a curve with-
out kinks or self-intersections. Instead of the tangent “space” at a point one usu-
ally speaks of the tangent line. A curve in R2 is called a plane curve and a curve in
R3 is a space curve, but you can have curves in any Rn . Curves can be closed (as
in the first picture below), unbounded (as indicated by the arrows in the second
picture), or have one or two endpoints (the third picture shows a curve with an
endpoint, indicated by a black dot; the white dot at the other end indicates that
1
2 1. INTRODUCTION
that point does not belong to the curve; the curve “peters out” without coming to
an endpoint). Endpoints are also called boundary points.
A circle with one point deleted is also an example of a manifold. Think of a torn
elastic band.
By straightening out the elastic band we see that this manifold is really the same
as an open interval.
The four plane curves below are not manifolds. The teardrop has a kink, where
two distinct tangent lines occur instead of a single well-defined tangent line; the
five-fold loop has five points of self-intersection, at each of which there are two
distinct tangent lines. The bow tie and the five-pointed star have well-defined
tangent lines everywhere. Still they are not manifolds: the bow tie has a self-
intersection and the cusps of the star have a jagged appearance which is proscribed
by the definition of a manifold (which we have not yet given). The points where
these curves fail to be manifolds are called singularities. The “good” points are
called smooth.
Singularities can sometimes be “resolved”. For instance, the self-intersections of
the Archimedean spiral, which is given in polar coordinates by r is a constant times
1.1. MANIFOLDS 3
θ, where r is allowed to be negative,
can be got rid of by uncoiling the spiral and wrapping it around a cone. You can
convince yourself that the resulting space curve has no singularities by peeking at
it along the direction of the x-axis or the y-axis. What you will see are the smooth
curves shown in the yz-plane and the xz-plane.
e3
e2
e1
(The three-dimensional models in these notes are drawn in central perspective.
They are best viewed facing the origin, which is usually in the middle of the pic-
ture, from a distance of 30 cm with one eye shut.) Singularities are extremely
interesting, but in this course we shall focus on gaining a thorough understanding
of the smooth points.
4 1. INTRODUCTION
Dimension 2. A two-dimensional manifold is a smooth surface without self-
intersections. It may have a boundary, which is always a one-dimensional mani-
fold. You can have two-dimensional manifolds in the plane R2 , but they are rel-
atively boring. Examples are: an arbitrary open subset of R 2 , such as an open
square, or a closed subset with a smooth boundary.
A closed square is not a manifold, because the corners are not smooth.1
Two-dimensional manifolds in three-dimensional space include a sphere, a parab-
oloid and a torus.
e3
e2
e1
The famous Möbius band is made by pasting together the two ends of a rectangular
strip of paper giving one end a half twist. The boundary of the band consists of
1To be strictly accurate, the closed square is a topological manifold with boundary, but not a smooth
manifold with boundary. In these notes we will consider only smooth manifolds.
1.1. MANIFOLDS 5
two boundary edges of the rectangle tied together and is therefore a single closed
curve.
Out of the Möbius band we can create in two different ways a manifold without
boundary by closing it up along the boundary edge. According to the direction in
which we glue the edge to itself, we obtain the Klein bottle or the projective plane.
A simple way to represent these three surfaces is by the following diagrams. The
labels tell you which edges to glue together and the arrows tell you in which di-
rection.
b b
a a a a a a
b b
Möbius band Klein bottle projective plane
Perhaps the easiest way to make a Klein bottle is first to paste the top and bottom
edges of the square together, which gives a tube, and then to join the resulting
boundary circles, making sure the arrows match up. You will notice this cannot be
done without passing one end through the wall of the tube. The resulting surface
intersects itself along a circle and therefore is not a manifold.
A different model of the Klein bottle is found by folding over the edge of a Möbius
band until it touches the central circle. This creates a Möbius type band with a
figure eight cross-section. Equivalently, take a length of tube with a figure eight
cross-section and weld the ends together giving one end a half twist. Again the
6 1. INTRODUCTION
resulting surface has a self-intersection, namely the central circle of the original
Möbius band. The self-intersection locus as well as a few of the cross-sections are
shown in black in the following wire mesh model.
To represent the Klein bottle without self-intersections you need to embed it in
four-dimensional space. The projective plane has the same peculiarity, and it too
has self-intersecting models in three-dimensional space. Perhaps the easiest model
is constructed by merging the edges a and b shown in the gluing diagram for the
projective plane, which gives the following diagram.
a a
First fold the lower right corner over to the upper left corner and seal the edges.
This creates a pouch like a cherry turnover with two seams labelled a which meet
at a corner. Now fuse the two seams to create a single seam labelled a. Below is a
wire mesh model of the resulting surface. It is obtained by welding together two
pieces along the dashed wires. The lower half shaped like a bowl corresponds to
the dashed circular disc in the middle of the square. The upper half corresponds
to the complement of the disc and is known as a cross-cap. The wire shown in
black corresponds to the edge a. The interior points of the black wire are ordinary
self-intersection points. Its two endpoints are qualitatively different singularities
1.2. EQUATIONS 7
known as pinch points, where the surface is crinkled up.
e3
e2
e1
1.2. Equations
Very commonly manifolds are given “implicitly”, namely as the solution set
of a system
φ1 V x1 , . . . , xn W$X c1 ,
φ2 V x1 , . . . , xn W$X c2 ,
..
.
φm V x1 , . . . , xn W$X cm ,
of m equations in n unknowns. Here φ 1 , φ2 , . . . , φm are functions, c 1 , c2 , . . . , cm
are constants and x 1 , x2 , . . . , xn are variables. By introducing the useful shorthand
YZ \N] YZ \N] YZ \N]
Z x1 ] Z φ1 V x W ] Z c1 ]
Z[ ] Z[ ] Z[ ]
x2 φ2 V x W c2
xX .. , φ V x W"X .. , cX .. ,
. ^ . ^ .^
xn φm V x W cn
we can represent this system as a single equation
φ V x WX c.
It is in general difficult to find explicit solutions of such a system. (On the positive
side, it is usually easy to decide whether any given point is a solution by plugging
it into the equations.) Manifolds defined by linear equations (i.e. where φ is a
matrix) are called affine subspaces of Rn and are studied in linear algebra. More
interesting manifolds arise from nonlinear equations.
1.1. E XAMPLE . Consider the system of two equations in three unknowns,
x2 _ y2 X 1,
y _ z X 0.
Here
x2 _ y2 1
φ V x W$Xa` and c Xc` .
y_ z b 0b
8 1. INTRODUCTION
The solution set of this system is the intersection of a cylinder of radius 1 about
the z-axis (given by the first equation) and a plane cutting the x-axis at a 45 d angle
(given by the second equation). Hence the solution set is an ellipse. It is a manifold
of dimension 1.
1.2. E XAMPLE . The sphere of radius r about the origin in R n is the set of all x
in Rn satisfying the single equation e x egf r. Here
e x ehfji x k x fUl x21 m x22 m k/k/k m x2n
is the norm or length of x and
xk yf x1 y1 m x2 y2 m k/k/k m xn yn
is the inner product or dot product of x and y. The sphere of radius r is an n n 1-
dimensional manifold in Rn . The sphere of radius 1 is called the unit sphere and
is denoted by Sn o 1 . What is a one-dimensional sphere? And a zero-dimensional
sphere?
The solution set of a system of equations may have singularities and is there-
fore not necessarily a manifold. A simple example is xy f 0, the union of the two
coordinate axes in the plane, which has a singularity at the origin. Other examples
of singularities can be found in Exercise 1.5.
Tangent spaces. Let us use the example of the sphere to introduce the notion
of a tangent space. Let
M fqp x r Rn s e x egf r t
be the sphere of radius r about the origin in Rn and let x be a point in M. There are
two reasonable, but inequivalent, views of how to define the tangent space to M
at
u x. The first view is that the tangent space at x consists of all vectors y such that
y n x v:k x f 0, i.e. y k x f x k x f r 2 . In coordinates: y 1 x1 m k/k/k m yn xn f r2 . This
is an inhomogeneous linear equation in y. In this view, the tangent space at x is an
affine subspace of Rn , given by the single equation y k x f r 2 .
However, for most practical purposes it is easier to translate this affine sub-
space to the origin, which turns it into a linear subspace. This leads to the second
view of the tangent space at x, namely as the set of all y such that y k x f 0, and
this is the definition that we shall espouse. The standard notation for the tangent
space to M at x is Tx M. Thus
Tx M fjp y r Rn s y k x f 0t ,
a linear subspace of Rn . (In Exercise 1.6 you will be asked to find a basis of Tx M
for a particular x and you will see that Tx M is n n 1-dimensional.)
Inequalities. Manifolds with boundary are often presented as solution sets
of a system of equations together with one or more inequalities. For instance,
the closed ball of radius r about the origin in Rn is given by the single inequality
e x exw r. Its boundary is the sphere of radius r.
1.4. CONFIGURATION SPACES 9
1.3. Parametrizations
A dual method for describing manifolds is the “explicit” way, namely by par-
ametrizations. For instance,
xy cos θ , yy sin θ
parametrizes the unit circle in R2 and
xy cos θ cos φ, yy sin θ cos φ, zy sin φ
parametrizes the unit sphere in R3 . (Here φ is the angle between a vector and the
xy-plane and θ is the polar angle in the xy-plane.) The explicit method has various
merits and demerits, which are complementary to those of the implicit method.
One obvious advantage is that it is easy to find points lying on a parametrized
manifold simply by plugging in values for the parameters. A disadvantage is that
it can be hard to decide if any given point is on the manifold or not, because this
involves solving for the parameters. Parametrizations are often harder to come by
than a system of equations, but are at times more useful, for example when one
wants to integrate over the manifold. Also, it is usually impossible to parame-
trize a manifold in such a way that every point is covered exactly once. Such is
the case for the two-sphere. One commonly restricts the polar coordinates z θ , φ {
to the rectangle | 0, 2π }~| π 2, π 2} to avoid counting points twice. Only the
meridian θ y 0 is then hit twice, but this does not matter for many purposes, such
as computing the surface area or integrating a continuous function.
We will use parametrizations to give a formal definition of the notion of a
manifold in Chapter 6. Note however that not every parametrization describes a
manifold. Examples of parametrizations with singularities are given in Exercises
1.1 and 1.2.
1.4. Configuration spaces
Frequently manifolds arise in more abstract ways that may be hard to capture
in terms of equations or parametrizations. Examples are solution curves of differ-
ential equations (see e.g. Exercise 1.10) and configuration spaces. The configuration
of a mechanical system (such as a pendulum, a spinning top, the solar system, a
fluid, or a gas etc.) is its state or position at any given time. (The configuration
ignores any motions that the system may be undergoing. So a configuration is like
a snapshot or a movie still. When the system moves, its configuration changes.)
In practice one usually describes a configuration by specifying the coordinates of
suitably chosen parts of the system. The configuration space or state space of the sys-
tem is an abstract space, the points of which are in one-to-one correspondence to
all physically possible configurations of the system. Very often the configuration
space turns out to be a manifold. Its dimension is called the number of degrees of
freedom of the system. The configuration space of even a fairly small system can be
quite complicated.
10 1. INTRODUCTION
1.3. E XAMPLE . A spherical pendulum is a weight or bob attached to a fixed
centre by a rigid rod, free to swing in any direction in three-space.
The state of the pendulum is entirely determined by the position of the bob. The
bob can move from any point at a fixed distance (equal to the length of the rod)
from the centre to any other. The configuration space is therefore a two-dimensional
sphere.
Some believe that only spaces of dimension 3 (or 4, for those who have
heard of relativity) can have a basis in physical reality. The following two exam-
ples show that this is not true.
1.4. E XAMPLE . Take a spherical pendulum of length r and attach a second one
of length s to the moving end of the first by a universal joint. The resulting system
is a double spherical pendulum. The state of this system can be specified by a pair of
vectors x, y , x being the vector pointing from the centre to the first weight and y
the vector pointing from the first to the second weight.
x
y
The vector x is constrained to a sphere of radius r about the centre and y to a sphere
of radius s about the head of x. Aside from this limitation, every pair of vectors
can occur (if we suppose the second rod is allowed to swing completely freely
and move “through” the first rod) and describes a distinct configuration. Thus
there are four degrees of freedom. The configuration space is a four-dimensional
manifold, known as the (Cartesian) product of two two-dimensional spheres.
1.5. E XAMPLE . What is the number of degrees of freedom of a rigid body mov-
ing in R3 ? Select any triple of points A, B, C in the solid that do not lie on one line.
The point A can move about freely and is determined by three coordinates, and
so it has three degrees of freedom. But the position of A alone does not determine
the position of the whole solid. If A is kept fixed, the point B can perform two
1.4. CONFIGURATION SPACES 11
independent swivelling motions. In other words, it moves on a sphere centred at
A, which gives two more degrees of freedom. If A and B are both kept fixed, the
point C can rotate about the axis AB, which gives one further degree of freedom.
B B
A A A
The positions of A, B and C determine the position of the solid uniquely, so the
total number of degrees of freedom is 3 2 1 6. Thus the configuration space
of a rigid body is a six-dimensional manifold.
1.6. E XAMPLE (the space of quadrilaterals). Consider all quadrilaterals ABCD
in the plane with fixed sidelengths a, b, c, d.
D
c
C
d
b
A a B
(Think of four rigid rods attached by hinges.) What are all the possibilities? For
simplicity let us disregard translations by keeping the first edge AB fixed in one
place. Edges are allowed to cross each other, so the short edge BC can spin full
circle about the point B. During this motion the point D moves back and forth on
a circle of radius d centred at A. A few possible positions are shown here.
(As C moves all the way around, where does the point D reach its greatest left-
or rightward displacement?) Arrangements such as this are commonly used in
engines for converting a circular motion to a pumping motion, or vice versa. The
position of the “pump” D is wholly determined by that of the “wheel” C. This
means that the configurations are in one-to-one correspondence with the points
on the circle of radius b about the point B, i.e. the configuration space is a circle.
12 1. INTRODUCTION
Actually, this is not completely accurate: for every choice of C, there are two
choices D and D for the fourth point! They are interchanged by reflection in the
diagonal AC.
D
A B
D
So there is in fact another circle’s worth of possible configurations. It is not possible
to move continuously from the first set of configurations to the second; in fact they
are each other’s mirror images. Thus the configuration space is a disjoint union of
two circles.
This is an example of a disconnected manifold consisting of two connected compo-
nents.
1.7. E XAMPLE (quadrilaterals, continued). Even this is not the full story: it is
possible to move from one circle to the other when b c a d (and also when
a b c d).
D
c
C
d
b
A a B
In this case, when BC points straight to the left, the quadrilateral collapses to a line
segment:
EXERCISES 13
and when C moves further down, there are two possible directions for D to go,
back up:
or further down:
This means that when b c a d the two components of the configuration space
are merged at a point.
The juncture represents the collapsed quadrilateral. This configuration space is
not a manifold, but most configuration spaces occurring in nature are (and an
engineer designing an engine wouldn’t want to use this quadrilateral to make a
piston drive a flywheel). More singularities appear in the case of a parallelogram
(a c and b d) and in the equilateral case (a b c d).
Exercises
1.1. The formulas x t sin t, y 1 cos t (t R) parametrize a plane curve. Graph
this curve as carefully as you can. You may use software and turn in computer output.
Also include a few tangent lines at judiciously chosen points. (E.g. find all tangent lines
with
slope 0, 1, and .) To compute tangent lines, recall that the tangent vector at a point
x, y of the curve has components dx dt and dy dt. In your plot, identify all points where
the curve is not a manifold.
1.2. Same questions as in Exercise 1.1 for the curve x 3at 1 t 3 , y 3at2 1 t3 .
1.3. Parametrize the space curve wrapped around the cone shown in Section 1.1.
14 1. INTRODUCTION
1.4. Sketch the surfaces defined by the following gluing diagrams.
b
d
a
a
c
b b
b
d a
a
a1 b2 a1 b2
b1 a2 b1 a2
a2 b1 a2 b1
b2 a1 b2 a1
(Proceed in stages, first gluing the a’s, then the b’s, etc., and try to identify what you get
at every step. One of these surfaces cannot be embedded in R 3 , so use a self-intersection
where necessary.)
1.5. For the values of n indicated below graph the surface in R 3 defined by x n y2 z.
Determine all the points where the surface does not have a well-defined tangent plane.
(Computer output is OK, but bear in mind that few drawing programs do an adequate job
of plotting these surfaces, so you may be better off drawing them by hand. As a preliminary
step, determine the intersection of each surface with a general plane parallel to one of the
coordinate planes.)
(i) n 0.
(ii) n 1.
(iii) n 2.
(iv) n 3.
1.6. Let M be the sphere of radius n about the origin in Rn and let x be the point
1, 1, . . . , 1 on M. Find a basis of the tangent space to M at x. (Use that Tx M is the set of all
y such that y x 0. View this equation as a homogeneous linear equation in the entries
y1 , y2 , . . . , yn of y and find the general solution by means of linear algebra.)
1.7. What is the number of degrees of freedom of a bicycle? (Imagine that it moves
freely through empty space and is not constrained to the surface of the earth.)
1.8. Choose two distinct positive real numbers a and b. What is the configuration
space of all parallelograms ABCD such that AB and CD have length a and BC and AD
have length b? What happens if a b? (As in Examples 1.6 and 1.7 assume that the edge
AB is kept fixed in place so as to rule out translations.)
1.9. What is the configuration space of all pentagons ABCDE in the plane with fixed
sidelengths a, b, c, d, e? (As in the case of quadrilaterals, for certain choices of sidelengths
singularities may occur. You may ignore these cases. To reduce the number of degrees of
EXERCISES 15
freedom you may also assume the edge AB to be fixed in place.)
E
d
D
c
C
e
b
A a B
1.10. The Lotka-Volterra system is an early (ca. 1925) predator-prey model. It is the
pair of differential equations
dx
rx sxy,
dt
dy
py qxy,
dt
where x ¡ t ¢ represents the number of prey and y ¡ t ¢ the number of predators at time t, while
p, q, r, s are positive constants. In this problem we will consider the solution curves (also
called trajectories) ¡ x ¡ t ¢ , y ¡ t ¢-¢ of this system that are contained in the positive quadrant
(x £ 0, y £ 0) and derive an implicit equation satisfied by these solution curves. (The
Lotka-Volterra system is exceptional in this regard. Usually it is impossible to write down
an equation for the solution curves of a differential equation.)
(i) Show that the solutions of the system satisfy a single differential equation of the
form dy ¤ dx f ¡ x ¢ g ¡ y ¢ , where f ¡ x ¢ is a function that depends only on x and
g ¡ y ¢ a function that depends only on y.
(ii) Solve the differential equation of part (i) by separating the variables, i.e. by writ-
1
ing dy f ¡ x ¢ dx and integrating both sides. (Don’t forget the integration
g ¡ y¢
constant.)
(iii) Set p q r s 1 and plot a number of solution curves. Indicate the
direction in which the solutions move. Be warned that solving the system may
give better results than solving the implicit equation! You may use computer
software such as Maple, Mathematica or MATLAB. A useful Java applet, ¥¥ ¦'§N¨© ,
can be found at ª««'¥¬F®®N¯¯¯±°²§D«'ª³°µ´¶·'©°F©D¸'¹ ®ºN¸»¶'©¦'¸®'¸»'¥D¥°¼ª«½² ¦ .
CHAPTER 2
Differential forms on Euclidean space
The notion of a differential form encompasses such ideas as elements of sur-
face area and volume elements, the work exerted by a force, the flow of a fluid, and
the curvature of a surface, space or hyperspace. An important operation on differ-
ential forms is exterior differentiation, which generalizes the operators div, grad
and curl of vector calculus. The study of differential forms, which was initiated by
E. Cartan in the years around 1900, is often termed the exterior differential calculus.
A mathematically rigorous study of differential forms requires the machinery of
multilinear algebra, which is examined in Chapter 7. Fortunately, it is entirely pos-
sible to acquire a solid working knowledge of differential forms without entering
into this formalism. That is the objective of this chapter.
2.1. Elementary properties
A differential form of degree k or a k-form on Rn is an expression
α¾ ∑ f I dx I .
I
(If you don’t know the symbol α , look up and memorize the Greek alphabet in
the back of the notes.) Here I stands for a multi-index ¿ i 1 , i2 , . . . , ik À of degree k, that
is a “vector” consisting of k integer entries ranging between 1 and n. The f I are
smooth functions on Rn called the coefficients of α , and dx I is an abbreviation for
dxi1 dxi2 /Á Á/Á dxik .
(The notation dxi1 Â dxi2 Â Á/Á/Á Â dxik is also often used to distinguish this kind of
product from another kind, called the tensor product.)
For instance the expressions
α¾ sin ¿ x1 Ã e x4 À dx1 dx5 Ã x2 x25 dx2 dx3 Ã 6 dx2 dx4 Ã cos x2 dx5 dx3 ,
β ¾ x1 x3 x5 dx1 dx6 dx3 dx2 ,
represent a 2-form on R5 , resp. a 4-form on R6 . The form α consists of four terms,
corresponding to the multi-indices ¿ 1, 5 À , ¿ 2, 3 À , ¿ 2, 4 À and ¿ 5, 3 À , whereas β con-
sists of one term, corresponding to the multi-index ¿ 1, 6, 3, 2 À .
Note, however, that α could equally well be regarded as a 2-form on R 6 that
does not involve the variable x 6 . To avoid such ambiguities it is good practice to
state explicitly the domain of definition when writing a differential form.
Another reason for being precise about the domain of a form is that the coef-
ficients f I may not be defined on all of Rn , but only on an open subset U of Rn . In
such a case we say α is a k-form on U. Thus the expression ln ¿ x 2 Ã y2 À z dz is not
a 1-form on R3 , but on the open set U ¾ R3 ÄÅ ¿ x, y, z ÀÇÆ x2 Ã y2 ¾ È 0 É , i.e. the
complement of the z-axis.
17
18 2. DIFFERENTIAL FORMS ON EUCLIDEAN SPACE
You can think of dxi as an infinitesimal increment in the variable x i and of dx I
as the volume of an infinitesimal k-dimensional rectangular block with sides dx i1 ,
dxi2 , . . . , dx ik . (A precise definition will follow in Section 7.2.) By volume we here
mean oriented volume, which takes into account the order of the variables. Thus,
if we interchange two variables, the sign changes:
Ë Ì dxi1 dxi2 Ê/Ê/Ê dxi p /Ê Ê/Ê dxiq /Ê Ê/Ê dxik ,
dxi1 dxi2 Ê/Ê/Ê dxiq /Ê Ê/Ê dxi p /Ê Ê/Ê dxik Í (2.1)
and so forth. This is called anticommutativity, graded commutativity, or the alternat-
ing property. In particular, this rule implies dx i dxi ËÍÌ dxi dxi , so dxi dxi Ë 0 for all
i.
Let us consider k-forms for some special values of k.
A 0-form on Rn is simply a smooth function (no dx’s).
A general 1-form looks like
f 1 dx1 Î f 2 dx2 Î Ê/Ê/Ê Î f n dxn .
A general 2-form has the shape
∑ f i, j dxi dx j Ë f 1,1 dx1 dx1 Î f 1,2 dx1 dx2 Î Ê/Ê/Ê Î f 1,n dx1 dxn
i, j
Î f 2,1 dx2 dx1 Î f 2,2 dx2 dx2 Î Ê/Ê/Ê Î f 2,n dx2 dxn Î Ê/Ê/Ê
Î f n,1 dxn dx1 Î f n,2 dxn dx2 Î Ê/Ê/Ê Î f n,n dxn dxn .
Because of the alternating property (2.1) the terms f i,i dxi dxi vanish, and a pair of
terms such as f 1,2 dx1 dx2 and f 2,1 dx2 dx1 can be grouped together: f 1,2 dx1 dx2 Î
f 2,1 dx2 dx1 ËÍÏ f 1,2 Ì f 2,1 Ð dx1 dx2 . So we can write any 2-form as
∑ gi, j dxi dx j Ë g1,2 dx1 dx2 Î Ê/Ê/Ê Î g1,n dx1 dxn
1Ñ iÒ jÑ n
Î g2,3 dx2 dx3 Î
/Ê Ê/Ê Î g2,n dx2 dxn Î Ê/Ê/Ê Î gn Ó 1,n dx n Ó 1 dx n .
Written like this, a 2-form has at most
1
nÎ nÌ 1Î nÌ 2Î Ê/Ê/Ê Î 2 Î 1 Ë n n 1Ð
2 Ï Ì
components.
Likewise, a general n Ì 1-form can be written as a sum of n components,
f 1 dx2 dx3 Ê/Ê/Ê dxn Î f 2 dx1 dx3 Ê/Ê/Ê dxn Î Ê/Ê/Ê Î f n dx1 dx2 Ê/Ê/Ê dxn Ó 1
n
Ë ∑ Õ i /Ê Ê/Ê dxn ,
f i dx1 dx2 Ê/Ê/Ê dx
iÔ 1
where dxÕ i means “omit the factor dx i ”.
Every n-form on Rn can be written as f dx 1 dx2 Ê/Ê/Ê dxn . The special n-form
dx1 dx2 Ê/Ê/Ê dxn is also known as the volume form.
Forms of degree k Ö n on Rn are always 0, because at least one variable has to
repeat in any expression dx i1 Ê/Ê/Ê dxik . By convention forms of negative degree are
0.
In general a form of degree k can be expressed as a sum
αË ∑ f I dx I ,
I
2.1. ELEMENTARY PROPERTIES 19
where the I are increasing multi-indices, 1 × i 1 Ø i2 ØQÙ/Ù/ÙÚØ ik × n. We shall
almost always represent forms in this manner. The maximum number of terms
occurring in α is then the number of increasing multi-indices of degree k. An
increasing multi-index of degree k amounts to a choice of k numbers from among
the numbers 1, 2, . . . , n. The total number of increasing multi-indices of degree k
is therefore equal to the binomial coefficient “n choose k”,
Û
n n!
.
k ÜÞÝ k! ß n à k á !
(Compare this to the number of all multi-indices of degree k, which is n k .) Two
k-forms α ∑ I f I dx I and β ∑ I g I dx I (with I ranging over the increasing multi-
Ý Ý
indices of degree k) are considered equal if and only if f I g I for all I. The
Ý
collection of all k-forms on an open set U is denoted by â k ß U á . Since k-forms can
be added together and multiplied by scalars, the collection â k ß U á constitutes a
vector space.
A form is constant if the coefficients f I are constant functions. The set of con-
stant k-forms is a linear subspace of â k ß U á of dimension ã nk ä . A basis of this sub-
space is given by the forms dx I , where I ranges over all increasing multi-indices
of degree k. (The space â k ß U á itself is infinite-dimensional.)
The (exterior) product of a k-form α ∑ I f I dx I and an l-form β ∑ J g J dx J is
Ý Ý
defined to be the k å l-form
αβ
Ý ∑ f I g J dx I dx J .
I,J
Usually many terms in a product cancel out or can be combined. For instance,
ß y dx å x dy áß x dx dz å y dy dz á y 2 dx dy dz å x 2 dy dx dz
ß y2 à x2 á dx dy dz.
Ý Ý
As an extreme example of such a cancellation, consider an arbitrary form α of
degree k. Its p-th power α p is of degree kp, which is greater than n if k æ 0 and
p æ n. Therefore
αnç 1 0
Ý
for any form α on Rn of positive degree.
The alternating property combines with the multiplication rule to give the fol-
lowing result.
2.1. P ROPOSITION (graded commutativity).
βα ßèà 1 á klαβ
Ý
for all k-forms α and all l-forms β.
P ROOF. Let I ß i1 , i2 , . . . , ik á and J ß j1 , j2 , . . . , jl á . Successively applying
Ý Ý
the alternating property we get
dx I dx J dxi1 dxi2 Ù/Ù/Ù dxik dx j1 dx j2 dx j3 Ù/Ù/Ù dx jl
Ý
ßèà 1 á k dx j1 dxi1 dxi2 Ù/Ù/Ù dxik dx j2 dx j3 Ù/Ù/Ù dx jl
Ý
2k
ßèà 1 á dx j1 dx j2 dxi1 dxi2 Ù/Ù/Ù dxik dx j3 Ù/Ù/Ù dx jl
Ý
..
.
kl
ßèà 1 á dx J dx I .
Ý
20 2. DIFFERENTIAL FORMS ON EUCLIDEAN SPACE
For general forms α é ∑ I f I dx I and β é ∑ J g J dx J we get from this
βα é ∑ g J f I dx J dx I éëê½ì 1í kl
∑ f I g J dx I dx J éëêèì 1 í klαβ,
I,J I,J
which establishes the result. QED
2
A noteworthy special case is α é β. Then we get α 2 î é êèì 1 í k α 2 éïê½ì 1 í kα 2 .
2
This equality is vacuous if k is even, but tells us that α é 0 if k is odd.
2.2. C OROLLARY. α 2 é 0 if α is a form of odd degree.
2.2. The exterior derivative
If f is a 0-form, that is a smooth function, we define d f to be the 1-form
n
∂f
df é ∑ ∂xi dxi .
ið 1
Then we have the product or Leibniz rule:
d ê f g í$é f dg ñ g df .
If α é ∑ I f I dx I is a k-form, each of the coefficients f I is a smooth function and we
define dα to be the k ñ 1-form
dα é ∑ d f I dx I .
I
The operation d is called exterior differentiation. An operator of this sort is called a
first-order partial differential operator, because it involves the first partial deriva-
tives of the coefficients of a form.
2.3. E XAMPLE . If α é f dx ñ g dy is a 1-form on R2 , then
dα é f y dy dx ñ g x dx dy éÍê g x ì f y í dx dy.
(Recall that f y is an alternative notation for ∂ f ò ∂y.) More generally, for a 1-form
α é ∑ nið 1 f i dxi on Rn we have
n n
∂ fi
dα é ∑ d f i dxi é ∑ ∂x
dx j dxi
ið 1 i, j ð 1 j
∂ fi ∂ fi
é ∑ ∂x
dx j dxi ñ
∂x ∑
dx j dxi
1ó iô jó n j 1ó jô ió n j
∂f ∂fj
éqì ∑ ∂xij dxi dx j ñ ∑ ∂xi dxi dx j (2.2)
1ó iô jó n 1ó iô jó n
∂fj ∂f
é ∑ ∂xi ì ∂xij ö dxi dx j ,
1ó iô jó n õ
where in line (2.2) in the first sum we used the alternating property and in the
second sum we interchanged the roles of i and j.
2.2. THE EXTERIOR DERIVATIVE 21
2.4. E XAMPLE . If α ÷ f dx dy ø g dx dz ø h dy dz is a 2-form on R 3 , then
dα ÷ f z dz dx dy ø g y dy dx dz ø h x dx dy dz ÷ëù f z ú g y ø h x û dx dy dz.
For a general 2-form α ÷ ∑1 ü i ý j ü n f i, j dx i dx j on Rn we have
n ∂ f i, j
dα ÷ ∑ d f i, j dxi ÷ ∑ ∑ ∂xk
dxk dxi dx j
1ü iý jü n 1ü iý jü n kþ 1
∂ f i, j ∂ f i, j
÷ ∑ ∂xk
dxk dxi dx j ø ∑ ∂xk
dxk dxi dx j
1ü ký iý jü n 1ü iý ký jü n
∂ f i, j
ø ∑ ∂xk dxk dxi dx j
1ü iý jý kü n
∂ f j,k ∂ f i,k
÷ ∑ ∂xi
dxi dx j dxk ø ∑ ∂x j
dx j dxi dxk
1ü iý jý kü n 1ü iý jý kü n
∂ f i, j
ø ∑ ∂xk dxk dxi dx j (2.3)
1ü iý jý kü n
∂ f i, j ∂ f i,k ∂ f j,k
÷ ∑ ∂xk
ú
∂x j
ø
∂xi
dxi dx j dxk . (2.4)
1ü iý jý kü n ÿ
Here in line (2.3) we rearranged the subscripts (for instance, in the first term we
relabelled k ú i, i ú j and j ú k) and in line (2.4) we applied the alternating
property.
An obvious but quite useful remark is that if α is an n-form on Rn , then dα is
of degree n ø 1 and so dα ÷ 0.
The operator d is linear and satisfies a generalized Leibniz rule.
2.5. P ROPOSITION . (i) d ù aα ø bβ û ÷ a dα ø b dβ for all k-forms α and β
and all scalars a and b.
(ii) d ù αβ û ÷Uù dα û β ø ù ú 1 û kα dβ for all k-forms α and l-forms β.
P ROOF. The linearity property (i) follows from the linearity of partial differ-
entiation:
∂ ù a f ø bg û ∂f ∂g
÷ a ø b
∂xi ∂xi ∂xi
for al smooth functions f , g and constants a, b.
Now let α ÷ ∑ I f I dx I and β ÷ ∑ J g J dx J . The Leibniz rule for functions and
Proposition 2.1 give
d ù αβ û ÷ ∑ dù f I g J û dx I dx J ÷ ∑ù f I dg J ø g J d f I û dx I dx J
I,J I,J
÷ ∑ d f I dx I ù g J dx J û ø ù ú 1 û k f I dx I ù dg J dx J û
I,J
÷ëù dα û β ø ù ú 1 û kα dβ,
which proves part (ii). QED
Here is one of the most curious properties of the exterior derivative.
22 2. DIFFERENTIAL FORMS ON EUCLIDEAN SPACE
2.6. P ROPOSITION . d dα 0 for any form α . In short,
d2 0.
P ROOF. Let α ∑ I f I dx I . Then
n n
∂ fI ∂ fI
d dα d ∑∑ ∂xi i
dx dx I ∑ ∑ d ∂xi
dxi dx I .
I i 1 I i 1
Applying the formula of Example 2.3 (replacing f i with ∂ f I ∂xi ) we find
n
∂ fI ∂2 f I ∂2 f I
∑ d ∂xi
dxi ∑ ∂xi ∂x j ∂x j ∂xi
dxi dx j 0,
i 1
1 i j n
because for any smooth (indeed, C 2 ) function f the mixed partials ∂ 2 f ∂xi ∂x j and
∂2 f ∂x j ∂xi are equal. Hence d dα 0. QED
2.3. Closed and exact forms
A form α is closed if dα 0. It is exact if α dβ for some form β (of degree
one less).
2.7. P ROPOSITION . Every exact form is closed.
P ROOF. If α dβ then dα d dβ 0 by Proposition 2.6. QED
2.8. E XAMPLE . y dx x dy is not closed and therefore cannot be exact. On
the other hand y dx x dy is closed. It is also exact, because d xy y dx x dy.
For a 0-form (function) f on Rn to be closed all its partial derivatives must vanish,
which means it is constant. A nonzero constant function is not exact, because
forms of degree 1 are 0.
Is every closed form of positive degree exact? This question has interesting
ramifications, which we shall explore in Chapters 4, 5 and 10. Amazingly, the
answer depends strongly on the topology, that is the qualitative “shape”, of the
domain of definition of the form.
Let us consider the simplest case of a 1-form α ∑ni 1 f i dxi . Determining
whether α is exact means solving the equation dg α for the function g. This
amounts to
∂g ∂g ∂g
f1 , f2 , ..., fn, (2.5)
∂x1 ∂x2 ∂xn
a system of first-order partial differential equations. Finding a solution is sometimes
called integrating the system. By Proposition 2.7 this is not possible unless α is
closed. By the formula in Example 2.3 α is closed if and only if
∂ fi ∂fj
∂x j ∂xi
for all 1 i j n. These identities must be satisfied for the system (2.5) to be
solvable and are therefore called the integrability conditions for the system.
2.9. E XAMPLE . Let α y dx z cos yz x dy y cos yz dz. Then
dα dy dx z y sin yz cos yz dz dy dx dy
y z sin yz cos yz dy dz 0,
2.4. THE HODGE STAR OPERATOR 23
so α is closed. Is α exact? Let us solve the equations
∂g ∂g ∂g
y, z cos yz x, y cos yz
∂x ∂y ∂z
by successive integration. The first equation gives g yx c y, z ! , where c is
a function of y and z only. Substituting into the second equation gives ∂c " ∂y
z cos yz, so c sin yz k z ! . Substituting into the third equation gives k #
0, so
k is a constant. So g yx sin yz is a solution and therefore α is exact.
This method works always for a 1-form defined on all of Rn . (See Exercise 2.6.)
Hence every closed 1-form on Rn is exact.
2.10. E XAMPLE . The 1-form on R2 $&% 0 ' defined by
y x $ y dx x dy
α $ 2 dx dy .
x y 2 x y
2 2 x2 y2
is called the angle form for reasons that will become clear in Section 4.3. From
∂ x y2 $ x2 ∂ y x2 $ y2
,
∂x x2 y2 x2 y2 ! 2 ∂y x2 y2 x2 y2 ! 2
it follows that the angle form is closed. This example is continued in Examples 4.1
and 4.6, where we shall see that this form is not exact.
For a 2-form α ∑1 ( i ) j ( n f i, j dxi dx j and a 1-form β ∑ni* 1 gi dxi the equa-
tion dβ α amounts to the system
∂g j $ ∂gi f i, j . (2.6)
∂xi ∂x j
By the formula in Example 2.4 the integrability condition dα 0 comes down to
∂ f i, j $ ∂ f i,k ∂ f j,k 0
∂xk ∂x j ∂xi
for all 1 + i , j , k + n. We shall learn how to solve the system (2.6), and its
higher-degree analogues, in Example 10.18.
2.4. The Hodge star operator
The binomial coefficient - nk . is the number of ways of selecting k (unordered)
objects from a collection of n objects. Equivalently, - nk . is the number of ways of
partitioning a pile of n objects into a pile of k objects and a pile of n $ k objects.
Thus we see that / /
n n
n $ k0
.
k0
This means that in a certain sense there are as many k-forms as n $ k-forms. In
fact, there is a natural way to turn k-forms into n $ k-forms. This is the Hodge star
operator. Hodge star of α is denoted by 1 α (or sometimes α 2 ) and is defined as
follows. If α ∑ I f I dx I , then
∑
1α f I 31 dx I ! ,
I
with
1 dx I ε I dx I c .
24 2. DIFFERENTIAL FORMS ON EUCLIDEAN SPACE
Here, for any increasing multi-index I, I c denotes the complementary increasing
multi-index, which consists of all numbers between 1 and n that do not occur in I.
The factor ε I is a sign,
1 if dx I dx I c 4 dx1 dx2 78787 dxn ,
εI 465 9 1 if dx dx c 9 dx dx
I I 4 1 2 78787 dx n .
In other words, : dx I is the product of all the dx j ’s that do not occur in dx I , times a
factor ; 1 which is chosen in such a way that dx I < : dx I = is the volume form:
dx I < : dx I =>4 dx1 dx2 78787 dxn .
2.11. E XAMPLE . Let n 4 6 and I 4?< 2, 6 = . Then I c 4@< 1, 3, 4, 5 = , so dx I 4
dx2 dx6 and dx I c 4 dx1 dx3 dx4 dx5 . Therefore
4 dx2 dx6 dx1 dx3 dx4 dx5
dx I dx I c
4 dx1 dx2 dx6 dx3 dx4 dx5 4 9 dx1 dx2 dx3 dx4 dx5 dx6 ,
which shows that ε I 4 9 1. Hence : < dx 2 dx6 =A4 9 dx1 dx3 dx4 dx5 .
2.12. E XAMPLE . On R2 we have : dx 4 dy and : dy 4 9 dx. On R3 we have
: dx 4 dy dz, : < dx dy = 4 dz,
: dy 4 9 dx dz 4 dz dx, : < dx dz =>4 9 dy,
: dz 4 dx dy, : < dy dz = 4 dx.
(This is the reason that 2-forms on R3 are sometimes written as f dx dy B g dz dx B
h dy dz, in contravention of our usual rule to write the variables in increasing order.
In higher dimensions it is better to stick to the rule.) On R4 we have
: dx1 4 dx2 dx3 dx4 , : dx3 4 dx1 dx2 dx4 ,
: dx2 4 9 dx1 dx3 dx4 , : dx4 4 9 dx1 dx2 dx3 ,
and
: < dx1 dx2 =>4 dx3 dx4 , : < dx2 dx3 =A4 dx1 dx4 ,
: < dx1 dx3 =>4 9 dx2 dx4 , : < dx2 dx4 =A4 9 dx1 dx3 ,
: < dx1 dx4 =>4 dx2 dx3 , : < dx3 dx4 =A4 dx1 dx2 .
On R we have : 1 4 dx1 dx2 78787 dxn , : < dx1 dx2 78787 dxn =A4 1, and
n
: dxi 4C< 9 1 = i D 1dx1 dx2 78787Fdx E i 78787 dxn for 1 G i G n,
: < dxi dx j =A4C< 9 1 = i D j D 1
dx1 dx2 78787Hdx
E i 78787Idx
E j 78787 dxn for 1 G i J j G n.
2.5. div, grad and curl
A vector field on an open subset U of Rn is a smooth map F : U K Rn . We can
write F in components as OQP
F1 < x = PP
F2 < x =
F < x =>4@LM . ,
M .. R MN
Fn < x =
2.5. DIV, GRAD AND CURL 25
or alternatively as F S ∑niT 1 Fi ei , where e1 , e2 , . . . , en are the standard basis vectors
of Rn . Vector fields in the plane can be plotted by placing the vector F U x V with
its tail at the point x. The diagrams below represent the vector fields W ye1 X xe2
and UW x X xy V e1 X U y W xy V e2 (which you may recognize from Exercise 1.10). The
arrows have been shortened so as not to clutter the pictures. The black dots are
the zeroes of the vector fields (i.e. points x where F U x V>S 0).
y y
We can turn F into a 1-form α by using the Fi as coefficients: α S ∑niT 1 Fi dxi .
For instance, the 1-form α SYW y dx X x dy corresponds to the vector field F S
W ye1 X xe2 . Let us introduce the symbolic notation
]Q^^
dx1 ^
dx2
dx SYZ[[ .. ,
[\ . _
dxn
which we will think of as a vector-valued 1-form. Then we can write α S F ` dx.
Clearly, F is determined by α and vice versa. Thus vector fields and 1-forms are
symbiotically associated to one another.
vector field F acb 1-form α : α S F ` dx.
Intuitively, the vector-valued 1-form dx represents an infinitesimal displacement.
If F represents a force field, such as gravity or an electric force acting on a particle,
then α S F ` dx represents the work done by the force when the particle is displaced
by an amount dx. (If the particle travels along a path, the total work done by the
force is found by integrating α along the path. We shall see how to do this in Section
4.1.)
The correspondence between vector fields and 1-forms behaves in an interest-
ing way with respect to exterior differentiation and the Hodge star operator. For
26 2. DIFFERENTIAL FORMS ON EUCLIDEAN SPACE
each function f the 1-form d f d ∑nie 1 f ∂ f g ∂xi h dxi is associated to the vector field
∂f n
mQn
∂x 1 n
n ∂ f nn
∂f ∂x 2
grad f d ∑ e djik .
ie 1
∂xi i kk .. .
kkl ∂ f o
∂x n
This vector field is called the gradient of f . (Equivalently, we can view grad f as
the transpose of the Jacobi matrix of f .)
grad f pcq df : df d grad f r dx.
Starting with a vector field F and letting α d F r dx, we find
s α d n F s dx d n F 1 i v 1 dx dx r8r8rxdx
∑ i f i h ∑ i uf t h 1 2 w i r8r8r dxn ,
ie 1 ie 1
Using the vector-valued n t 1-form
s dx mQnn dx2 dx3 r8r8r dxn
mQnn
1 n n
s dx dx dx 8
r 8
r r dx
s dx d 2 t 1 3 n
kik ... o d ikk ..
. o
lk s dxn lk n v 1 dx1 dx2 r8r8r dxn y 1
fut 1 h
we can also write s α d F r s dx. Intuitively, the vector-valued n t 1-form s dx
represents an infinitesimal n t 1-dimensional hypersurface perpendicular to dx.
(This point of view will be justified in Section 8.3, after the proof of Theorem 8.14.)
In fluid mechanics, the flow of a fluid or gas in Rn is represented by a vector field F.
The n t 1-form s α then represents the flux, that is the amount of material passing
through the hypersurface s dx per unit time. (The total amount of fluid passing
through a hypersurface S is found by integrating α over S. We shall see how to do
this in Section 5.1.) We have
n
∂F
ds α d d f F r s dx h d ∑ i ft 1 h i v 1 dxi dx1 dx2 r8r8rxdx
w i r8r8r dxn
ie 1
∂xi
n n
∂F ∂F
d ∑ i dx1 dx2 r8r8r dxi r8r8r dxn d{z ∑ i dx1 dx2 r8r8r dxn .
ie 1
∂xi ie 1
∂xi |
The function div F d ∑nie 1 ∂Fi g ∂xi is the divergence of F. Thus if α d F r dx, then
d s α d d f F r s dx h d div F dx 1 dx2 r8r8r dxn .
An alternative way of writing this identity is obtained by applying s to both sides,
which gives
div F d s d s α.
A very different identity is found by first applying d and then s to α:
n
∂Fi ∂Fj ∂F
dα d ∑ ∂x
dx j dxi d ∑ ∂xi
i
t ∂x j dxi dx j ,
i, j e 1 j } ~ }
1 i j n
EXERCISES 27
and hence
∂Fj ∂Fi
dα ∑ 1 i j 1 ∂xi ∂x j
dx1 dx2 88Hdx
i 88dx
j 88 dxn .
1 i j n u
In three dimensions dα is a 1-form and so is associated to a vector field, namely
curl F ∂F3 ∂F2
e1
∂F3 ∂F1
e2
∂F2
∂x
∂F1
e3 ,
∂x2 ∂x3 ∂x 1 ∂x 3 1 ∂x 2
the curl of F. Thus, for n
3, if α F dx, then
curl F dx dα .
You need not memorize every detail of this discussion. The point is rather to
remember that exterior differentiation in combination with the Hodge star unifies
and extends to arbitrary dimensions the classical differential operators of vector
calculus.
Exercises
2.1. Compute the exterior derivative of the following forms. Recall that a hat indicates
that a term has to be omitted.
(i) e xyz dx.
(ii) ∑ni 1 x2i dx1 uudx
i uQ dxn .
(iii) x ∑i 1 1 1 xi dx1 uu dx
p n i
i uQ dxn , where p is a real constant. For what val-
ues of p is this form closed?
2.2. Consider the forms α x dx y dy, β z dx dy x dy dz and γ z dy on R 3 .
Calculate
(i) αβ, αβγ ;
(ii) dα , dβ, dγ .
2.3. Write the coordinates on R2n as x1 , y1 , x2 , y2 , . . . , xn , yn . Let
n
ω ∑ dxi dyi .
dx1 dy1 dx2 dy2 uu dxn dyn
i 1
Compute ω n ωω uu ω (n-fold product). First work out the cases n 1, 2, 3.
2.4. Write the coordinates on R2n 1 as x1 , y1 , x2 , y2 , . . . , xn , yn , z . Let
n
α dz x1 dy1 x2 dy2 u xn dyn dz ∑ xi dyi .
i 1
Compute α dα n α dα dα u dα . First work out the cases n 1, 2, 3.
2.5. Check that each of the following forms α 1
R
3 is closed and find a function
g such that dg α .
(i) α ye xy z sin xz 3 dx xey
xy z2 dy x sin xz F 2yz 3z 2 dz.
(ii) α 3 4 2xy z dx 3x2 y2 z4 ze sin ze y 3 dy 4x2 y3 z3 e y sin ze y H e z dz.
2.6. Let α ∑ni 1 f i dxi be a closed C 1-form on Rn . Define a function g by
x1 x2
g x f 1 t, x2 , x3 , . . . , xn dt ¡ f 2 0, t, x3 , x4 , . . . , xn dt
0 0
x3 xn
¡ f 3 0, 0, t, x4 , x5 , . . . , xn dt Q ¢ f n 0, 0, . . . , 0, t dt.
0 0
Show that dg α . (Apply the fundamental theorem of calculus, formula (B.3), differentiate
under the integral sign and don’t forget to use dα 0.)
28 2. DIFFERENTIAL FORMS ON EUCLIDEAN SPACE
2.7. Let α £ ∑ ni¤ 1 f i dxi be a closed 1-form whose coefficients f i are smooth functions
defined on Rn ¥§¦ 0 ¨ that are all homogeneous of the same degree p £ © ¥ 1. Let
1 n
g ª x «¬£ ∑ xi f i ª x « .
p 1 ¤
i 1
Show that dg £ α . (Use dα £ 0 and apply the identity proved in Exercise B.5 to each f i .)
2.8. Let α and β be closed forms. Prove that αβ is also closed.
2.9. Let α be closed and β exact. Prove that αβ is exact.
2.10. Calculate ® α , ® β, ® γ ,
®¯ª αβ « , where α, β and γ are as in Exercise 2.2.
2.11. Consider the form α £ ¥ x 22 dx1 x21 dx2 on R2 .
(i) Find ® α and ® d ® dα .
(ii) Repeat the calculation, regarding α as a form on R 3 .
(iii) Again repeat the calculation, now regarding α as a form on R 4 .
2.12. Prove that ®°® α £±ª ¥ 1 « kn ² kα for every k-form α on Rn .
2.13. Let α £ ∑ I a I dx I and β £ ∑ I b I dx I be constant k-forms, i.e. with constant coef-
ficients a I and b I . (We also assume, as usual, that the multi-indices I are increasing.) The
inner product of α and β is the number defined by
ª α , β «¬£ ∑ a I b I .
I
Prove the following assertions.
(i) The dx I form an orthonormal basis of the space of constant k-forms.
(ii) ª α , α «´³ 0 for all α and ª α , α «¬£ 0 if and only if α £ 0.
(iii) α ªµ® β «¶£·ª α , β « dx 1 dx2 ¸u¸Q¸ dxn .
(iv) α ªµ® β «¶£ β ªµ® α « .
(v) The Hodge star operator is orthogonal, i.e. ª α , β «¹£±ªµ® α , ® β « .
2.14. The Laplacian of a smooth function on an open subset of R n is defined by
º ∂2 f ∂2 f ∂2 f
f £ ¸¸Q¸ .
∂x21 ∂x22 ∂x2n
Prove the following formulas.
º
(i) f £»® d ® d f .
º º º
(ii) ª f g « £±ª f « g f g 2 ®¼ª d f ªµ® dg «½« . (Use Exercise 2.13(iv).)
2.15. Let f : Rn ¾ R be a function and let α £ f dx i .
(i) Calculate d ® d ® α .
(ii) Calculate ® d ® dα .
º º
(iii) Show that d ® d ® α ¿ª ¥ 1 « n ® d ® dα £±ª f « dxi , where is the Laplacian defined in
Exercise 2.14.
2.16. (i) Let U be an open subset of Rn and let f : U ¾ R be a function satis-
fying grad f ª x «À£ © 0 for all x in U. On U define a vector field n, an n ¥ 1-form ν
and a 1-form α by
n ª x « £Á grad f ª x «QÁà1
grad f ª x « ,
ν £ n ¸ ® dx,
α £Á grad f ª x «QÁ  1
df .
Prove that dx 1 dx2 ¸u¸u¸ dxn £ αν on U.
(ii) Let r : Rn ¾ R be the function r ª x «Ä£ÅÁ x Á (distance to the origin). Deduce from
part (i) that dx 1 dx2 ¸¸u¸ dxn £±ª dr « ν on Rn ¥¦ 0 ¨ , where ν £Á x Á  1 x ¸ ® dx.
EXERCISES 29
2.17. The Minkowski or relativistic inner product on R n Æ 1 is given by
Ç n
x, y È¹É ∑ xi yi Ë xn Æ 1 yn 1 . Æ
Ê
i 1
Ç Ç Ç
A vector x Ì Rn Æ 1 is spacelike if x, x È´Í 0, lightlike if x, x È¬É 0, and timelike if x, x ÈÄÎ 0.
(i) Give examples of (nonzero) vectors of each type.Ç
(ii) Show that for every x É Ï 0 there is a y such that x, y ÈÐÉ Ï 0.
A Hodge star operator corresponding to this inner product is defined as follows: if α É
∑ I f I dx I , then Ñ Ñ
Ç
α É ∑ f I dx I È ,
I
with
Ñ
ε I dx I c if I contains n Ô 1,
dx I ÉÓÒ
Ë ε I dx I c if I does not contain n Ô 1.
Ñ asÑ in the definition of the ordinary
(Here ε I and I c are Ñ Hodge star.)
Ç
(iii) Find 1, dx i for 1 Õ i Õ n Ô 1, and dx 1 dx2 ÖÖuÖ dxn È .
Ñ Ñ “relativistic Laplacian” (usually called the d’Alembertian or wave
(iv) Compute the
Ñ f on Rn Æ 1 .
operator) d d f for any smooth functionÇ
(v) For n É 3 (ordinary space-time) find dx i dx j È for 1 Õ i Î j Õ 4.
2.18. One of the greatest advances in theoretical physics of the nineteenth century was
Maxwell’s formulation of the equations of electromagnetism:
1 ∂B
curl E É Ë c ∂t
(Faraday’s Law),
4π 1 ∂D
curl H É JÔ (Ampère’s Law),
c c ∂t
div D É 4πρ (Gauß’ Law),
div B É 0 (no magnetic monopoles).
Here c is the speed of light, E is the electric field, H is the magnetic field, J is the density
of electric current, ρ is the density of electric charge, B is the magnetic induction and D is
the dielectric displacement. E, H, J, B and D are vector fields and ρ is a function on R 3 and
all depend on time t. The Maxwell equations look particularly simple in differential form
Ç
notation, as we shall now see. In space-time R4 with coordinates x 1 , x2 , x3 , x4 È , where
x4 É ct, introduce forms
Ç
α É E1 dx1 Ô E2 dx2 Ô E3 dx3 È dx4 Ô B1 dx2 dx3 Ô B2 dx3 dx1 Ô B3 dx1 dx2 ,
Ç
β É Ë H1 dx1 Ô H2 dx2 Ô H3 dx3 È dx4 Ô D1 dx2 dx3 Ô D2 dx3 dx1 Ô D3 dx1 dx2 ,
1Ç
γ É J dx dx Ô J2 dx3 dx1 Ô J3 dx1 dx2 È dx4 Ë ρ dx1 dx2 dx3 .
c 1 2 3
(i) Show that Maxwell’s equations are equivalent to
dα É 0,
dβ Ô 4πγ É 0.
(ii) Conclude that γ is closed and that div J Ô ∂ρ × ∂t É 0. Ñ
(iii) In vacuum one has E É D and H É B. Show that in vacuum β É α , the
relativistic Hodge star of α defined in Exercise 2.17.
Ñ
(iv) Free space is a vacuum without charges or currents. Show that the Maxwell equa-
tions in free space are equivalent to dα É d α É 0.
30 2. DIFFERENTIAL FORMS ON EUCLIDEAN SPACE
(v) Let f , g : R Ø R be any smooth functions and define
0 0
E Ù x Ú¬ÛÝÜÞ f Ù x1 ß x4 Ú , B Ù x Ú ÛÝÜÞ ß g Ù x1 ß x4 Ú .
g Ù x1 ß x4 Ú½àá f Ù x1 ß x4 ÚÐàá
Show that the corresponding 2-form α satisfies the free Maxwell equations dα Û
d â α Û 0. Such solutions are called electromagnetic waves. Explain why. In what
direction do these waves travel?
CHAPTER 3
Pulling back forms
3.1. Determinants
The determinant of a square matrix is the oriented volume of the block (paral-
lelepiped) spanned by its column vectors. It is therefore not surprising that differ-
ential forms are closely related to determinants. This section is a review of some
fundamental facts concerning determinants. Let
çQè
a1,1 ... a1,n
ãjäåæ .. ..
A . . é
an,1 ... an,n
be an n ê n-matrix with column vectors a1 , a2 , . . . , an . Its determinant is variously
denoted by
a1,1 ... a1,n
det A ã det ë a1 , a2 , . . . , an ì ã det ë ai, j ì 1 í i, j ní ã6î ...
îî
.. î .
. î
îî an,1 . . . an,n îîî
îî îî
Expansion on the first column. You have probably seen the following defini-
tion of the determinant:
n
det A ã ∑ ëð 1 ì i ñ 1 ai1 det Ai,1 .
iï 1
Here Ai, j denotes the ë n ð 1 ì êòë n ð 1 ì -matrix obtained from A by striking out the
i-th row and the j-th column. This is a recursive definition, which reduces the cal-
culation of any determinant to that of determinants of smaller size. (The recursion
starts at n ã 1; the determinant of a 1 ê 1-matrix ë a ì is simply defined to be the
number a.) It is a useful rule, but it has two serious flaws: first, it is extremely inef-
ficient computationally (except for matrices containing lots of zeroes), and second,
it obscures the relationship with volumes of parallelepipeds.
Axioms. A far better definition is available. The determinant can be com-
pletely characterized by three simple laws, which make good sense in view of its
geometrical significance and which comprise an efficient algorithm for calculating
any determinant.
3.1. D EFINITION . A determinant is a function det which assigns to every or-
dered n-tuple of vectors ë a1 , a2 , . . . , an ì a number det ë a1 , a2 , . . . , an ì subject to the
following axioms:
31
32 3. PULLING BACK FORMS
(i) det is multilinear (i.e. linear in each column):
det ó a1 , a2 , . . . , cai
ô c õ aiõ , . . . , an ö
÷ c det ó a1 , a2 , . . . , ai , . . . , an ö ô c õ det ó a1 , a2 , . . . , aõ , . . . , an ö
i
for all scalars c, c õ and all vectors a1 , a2 , . . . , ai , aiõ , . . . , an ;
(ii) det is alternating or antisymmetric:
det ó a1 , . . . , ai , . . . , a j , . . . , an ö ÷ùø det ó a1 , . . . , a j , . . . , ai , . . . , an ö
for any i ÷ ú j;
(iii) normalization: det ó e1 , e2 , . . . , en ö ÷ 1, where e1 , e2 , . . . , en are the stan-
dard basis vectors of Rn .
We also write det A instead of det ó a1 , a2 , . . . , an ö , where A is the matrix whose
columns are a1 , a2 , . . . , an . Axiom (iii) lays down the value of det I. Axioms (i)
and (ii) govern the behaviour of oriented volumes under the elementary column
operations on matrices. Recall that these operations come in three types: adding a
multiple of any column of A to any other column (type I); multiplying a column
by a nonzero constant (type II); and interchanging any two columns (type III).
Type I does not affect the determinant, type II multiplies it by the corresponding
constant, and type III causes a sign change. This can be restated as follows.
3.2. L EMMA . If E is an elementary column operation, then det ó E ó A öuö ÷ k det A,
where
þ 1 if E is of type I,
k ü
÷ ûý c if E is of type II (multiplication of a column by c),
ýÿ ø 1 if E is of type III.
3.3. E XAMPLE . Identify the column operations applied at each step in the fol-
lowing calculation.
1 1 1 1 0 0 1 0 0 1 0 0
4 10 9 ÷ 4 6 5 ÷ 4 1 5 ÷ 3 1 2
1 5 4 1 4 3 1 1 3 0 1 0
1 0 0 1 0 0 1 0 0
÷ 2 3 1 1 ÷ 2 0 0 1 ÷Cø 2 0 1 0 ÷ùø 2.
0 1 0 0 1 0 0 0 1
As this example suggests, the axioms (i)–(iii) suffice to calculate any n n-
determinant. In other words, there is at most one function det which obeys these
axioms. More precisely, we have the following result.
3.4. T HEOREM (uniqueness of determinants). Let det and det õ be two functions
satisfying Axioms (i)–(iii). Then det A ÷ det õ A for all n n-matrices A.
P ROOF. Let a1 , a2 , . . . , an be the column vectors of A. Suppose first that A
is not invertible. Then the columns of A are linearly dependent. For simplicity
let us assume that the first column is a linear combination of the others: a1 ÷
c 2 a2 ô
ô cn an . Applying axioms (i) and (ii) we get
n
÷ ∑ ci det ó ai , a2 , . . . , ai , . . . , an ö ÷ 0,
det A
i 2
3.1. DETERMINANTS 33
and for the same reason det A 0, so det A
det A. Now assume that A
is invertible. Then A is column equivalent to the identity matrix, i.e. it can be
transformed to I by successive elementary column operations. Let E 1 , E2 , . . . , Em
be these elementary operations, so that E m Em 1 E2 E1 A I. According to
Lemma 3.2, each operation E i has the effect of multiplying the determinant by a
certain factor k i , so axiom (iii) yields
1 det I E E A k k k k det A.
det Em Em 1 2 1 m m 1 2 1
Applying the same reasoning to det A we get 1 k k k k det A. Hence
det A 1 k k k det A.
m m 1 2 1
3.5. R
1 2 m
QED
(change of normalization). Suppose that det is a function that
EMARK
axiom (i) and the antisymmetry axiom (ii) but is normal-
satisfies the multilinearity
ized differently: det I c. Then the proof of Theorem 3.4 shows that det A
c det A for all n n-matrices A.
This result leaves an open question. We can calculate the determinant of any
matrix by column reducing it to the identity matrix, but there are many different
ways of performing this reduction. Do different column reductions lead to the
same answer for the determinant? In other words, are the axioms (i)–(iii) consis-
tent? We will answer this question by displaying an explicit formula for the deter-
minant of any n n-matrix that does not involve any column reductions. Unlike
Definition 3.1, this formula is not very practical for the purpose of calculating large
determinants, but it has other uses, notably in the theory of differential forms.
3.6. T HEOREM (existence of determinants). Every n n-matrix A has a well-
defined determinant. It is given by the formula
∑ a a .
det A sign σ a1,σ 1 2,σ 2 n,σ n
σ Sn
This requires a little explanation. S n stands for the collection of all permutations
of the set 1, 2, . . . , n . A permutation is a way of ordering the numbers 1, 2, . . . , n.
Permutations are usually written as row vectors containing each of these numbers
exactly once. Thus for n
2 there are only two permutations: 1, 2 and 2, 1 .
For n 3 all possible permutations are
1, 2, 3 , 1, 3, 2 , 2, 1, 3 , 2, 3, 1 , 3, 1, 2 , 3, 2, 1 .
For general n there are
3 2 1 n!
n n 1 n 2
permutations. An alternative way of thinking of a permutation is as a bijective
(i.e. one-to-one and onto) map from the set 1, 2, . . . , n to itself. For example, for
n 5 a possible permutation is
5, 3, 1, 2, 4 ,
and we think of this as a shorthand notation for the map σ given by σ 1 5,
σ 2 3, σ 3
1, σ 4 2 and σ 5
4. The permutation 1, 2, 3, . . . , n 1, n
then corresponds to the identity map on the set 1, 2, . . . , n .
If σ is the identity permutation, then clearly σ i σ j whenever i j.
However, if σ is not the identity permutation, it cannot preserve the order in this
way. An inversion in σ is any pair of numbers i and j such that 1 i j n and
34 3. PULLING BACK FORMS
σ i
σ j . The length of σ , denoted by l σ , is the number of inversions in σ .
A permutation is called even or odd according to whether its length is even, resp.
odd. For instance, the permutation 5, 3, 1, 2, 4 has length 6 and so is even. The
sign of σ is $
sign σ ! !
1 lσ " #
1 if σ is even,
1 if σ is odd.
Thus sign 5, 3, 1, 2, 4 % 1. The permutations of & 1, 2 ' are 1, 2 , which has sign
1, and 2, 1 , which has sign ! 1, while for n 3 we have the table below.
σ l σ sign σ
1, 2, 3 0 1
1, 3, 2 1 !1
2, 1, 3 1 !1
2, 3, 1 2 1
3, 1, 2 2 1
3, 2, 1 3 !1
Thinking of permutations in S as bijective maps from & 1, 2, . . . , n ' to itself,
we can form the composition σ ( τ of any two permutations σ and τ in S . For
n
permutations we usually write στ instead of σ ( τ and call it the product of σ and
n
τ . This is the permutation produced by first performing τ and then σ ! For instance,
if σ 5, 3, 1, 2, 4 and τ 5, 4, 3, 2, 1 , then
τσ 1, 3, 5, 4, 2 , στ ) 4, 2, 1, 3, 5 .
A basic fact concerning signs, which we shall not prove here, is
sign στ *
sign σ sign τ . (3.1)
In particular, the product of two even permutations is even and the product of an
even and an odd permutation is odd.
The determinant formula in Theorem 3.6 contains n! terms, one for each per-
mutation σ . Each term is a product which contains exactly one entry from each
row and each column of A. For instance, for n 5 the permutation 5, 3, 1, 2, 4
contributes the term a 1,5 a2,3 a3,1 a4,2 a5,4 . For 2 2- and 3 3-determinants Theo- + +
rem 3.6 gives the well-known formulæ
,, a ,,
,, a 1,1
2,1
a1,2
a2,2
,, a1,1 a2,2 ! a1,2 a2,1 ,
,, a ,,
,, a 1,1 a1,2 a1,3 ,, ! ! -
,, a 2,1 a2,2
a3,2
a2,3
a3,3
,, a1,1 a2,2 a3,3 a1,1 a2,3 a3,2 a1,2 a2,1 a3,3 a1,2 a2,3 a3,1
!
3,1
- a1,3 a2,1 a3,2 a1,3 a2,2 a3,1 .
P ROOF OF T HEOREM 3.6. We need to check that the right-hand side of the
determinant formula in Theorem 3.6 obeys axioms (i)–(iii) of Definition 3.1. Let
us for the moment denote the right-hand side by f A .
Axiom (i) is checked as follows: for every permutation σ the product
a1,σ " # a " #/... a " #
1 2,σ 2 n,σ n
3.1. DETERMINANTS 35
contains exactly one entry from each row and each column in A. So if we multiply
0 1
the i-th row of A by c, each term in f A is multiplied by c. Therefore
f 0 a , a , . . . , ca , . . . , a 1*2
1 2 i n 0
c f a1 , a2 , . . . , a i , . . . , a n . 1
Similarly,
0 3 a4 , . . . , a 1*2 f 0 a , a , . . . , a , . . . , a 1 3 f 0 a , a , . . . , a4 , . . . , a 1 .
f a1 , a 2 , . . . , a i i n 1 2 i n 1 2 i n
Axiom (ii) holds because if we interchange two columns in A, each term in f 0 A 1
changes sign. To see this, let τ be the permutation in S n that interchanges the two
numbers i and j and leaves all others fixed. Then
0
f a1 , . . . , a j , . . . , a i , . . . , a n 1
2 ∑ 0 1
6 7 a 6 7/888 a 6 7
5
sign σ a1,τσ 1 2,τσ 2 n,τσ n
σ Sn
2 ∑ sign 0 τρ 1 a 6 7 a 6 7 888 a 6 7 substitute ρ 2 τσ
5 ρ Sn
1,ρ 1 2,ρ 2 n,ρ n
2 ∑ sign 0 τ 1 sign 0 ρ 1 a 6 7 a 6 7/888 a 6 7 by formula (3.1)
5 ρ Sn
1,ρ 1 2,ρ 2 n,ρ n
2:9 ∑ sign 0 ρ 1 a 6 7 a 6 7/888 a 6 7
5 1,ρ 1 by Exercise 3.4
2,ρ 2 n,ρ n
ρ Sn
2:9 f 0 a , . . . , a , . . . , a , . . . , a 1 .
1 i j n
Finally, rule (iii) is correct because if A 2 I,
1 if σ 2 identity,
a 6 7 a 6 7 888 a 6 7 2<;
1,σ 1 2,σ 2 n,σ n
0 otherwise,
and therefore f 0 I 12 1. So f satisfies all three axioms for determinants. QED
Here are some further rules followed by determinants. Each can be deduced
from Definition 3.1 or from Theorem 3.6. (Recall that the transpose of an n n- =
matrix A 2)0 1
ai, j is the matrix A T whose i, j-th entry is a j,i.)
3.7. T HEOREM . Let A and B be n = n-matrices.
0 >1 2
(i) det AB det A det B.
2
2 ? 0@9 1 A
(ii) det A T det A.
(iii) (Expansion on the j-th column) det A ∑ni 1 1 i j ai, j det Ai, j for all j 2
1, 2, . . . , n. Here A i, j denotes the n 1 0 9 1*=0 9 1
n 1 -matrix obtained from A
by striking out the i-th row and the j-th column.
2
(iv) det A a1,1 a2,2 888
an,n if A is upper triangular (i.e. a i, j 0 for i j). 2 B
Volume change. We conclude this discussion with a slightly different geo-
A : Rn C
metric view of determinants. A square matrix A can be regarded as a linear map
Rn . The unit cube in Rn ,
D 0, 1E 2GF x H
n
Rn 0 I J xi J 1 for i 2 1, 2, . . . , n , K
2
LD E M
has n-dimensional volume 1. (For n 1 it is usually called the unit interval
2
LD E M
and for n 2 the unit square.) Its image A 0, 1 n under the map A is a par-
allelepiped with edges Ae1 , Ae2 , . . . , Aen , the columns of A. Hence A 0, 1 n
36 3. PULLING BACK FORMS
has n-dimensional volume vol A 0, 1 n N O P SQ RUT TVR<T
det A
n
det A vol 0, 1 n . This rule T O P
generalizes as follows: if X is a measurable subset of R , then
vol A W X X RYT det A T vol X.
Ae2
X A AX
e2
e1
Ae1
T T
So det A can be interpreted as a volume change factor. (A set is measurable if it has
a well-defined, finite or infinite, n-dimensional volume. Explaining exactly what
this means is rather hard, but it suffices for our purposes to know that all open
and all closed subsets of Rn are measurable.)
3.2. Pulling back forms
By substituting new variables into a differential form we obtain a new form of
the same degree but possibly in a different number of variables.
3.8. E XAMPLE . In Example 2.10 we defined the angle form on R 2 Z\[ 0 ] to be
αR
Z y dx ^ x dy .
x ^ y 2 2
R R
By substituting x cos t and y sin t into the angle form we obtain the following
1-form on R:
Z sin t d cos t ^ cos t d sin t R N W Z sin t X_W Z sin t X`^ cos tQ dt R dt.
cos t ^ sin t
2
2 2
We can take any k-form and substitute any number of variables into it to obtain
a new k-form. This works as follows. Suppose α is a k-form defined on an open
subset V of Rm . Let us denote the coordinates on Rm by y1 , y2 , . . . , ym and let us
write, as usual,
α ∑ f I dy I , R
I
where the functions f I are defined on V. Suppose we want to substitute “new”
variables x1 , x2 , . . . , xn and that the old variables are given in terms of the new by
functions
y1 R W
φ1 x1 , . . . , xn , X
y2 R W
φ2 x1 , . . . , xn , X
..
.
ym R W
φm x1 , . . . , xn . X
3.2. PULLING BACK FORMS 37
As usual we write y a φ b xc , where
b c jh iii
φ1 x
φ b x caedf
b c .
φ2 x
ffg ..
. k
φ b xc
m
We assume that the functions φ i are smooth and defined on a common domain U,
a
we have U R, V R2 a0 and φ t lnm o
which is an open subset of Rn . We regard φ as a map from U to V. (In Example 3.8
b c*a)b
cos t, sin t .) The pullback of α along φ c
p
is then the k-form φ α on U obtained by substituting y i φi x1 , . . . , xn for all i a b c
in the formula for α . That is to say, φ α is defined by p
φ α p a ∑ b φp c_b p c
f I φ dy I .
I
p
Here φ f I is defined by
φ fIp a q
f I φ,
the composition of φ and f I . This means φ f I x p b ca b b cjc
f I φ x ; in other words, φ f I p
is the function resulting from f I by substituting y φ x . The pullback φ dy I is a b c p
defined by replacing each y i with φi . That is to say, if I a b
i 1 , i2 , . . . , ik we put c
p
φ dy I a φp b dy i1 dyi2 rrr dy c*a dφ dφ rrr dφ .
ik i1 i2 ik
The picture below is a schematic representation of the substitution process. The
form α a
∑ I f I dy I is a k-form in y1 , y2 , . . . , ym ; its pullback φ α ∑ J g J dx J is a p a
k-form in x1 , x2 , . . . , xn . In Theorem 3.12 below we will give an explicit formula
for the coefficients g J in terms of f I and φ.
U V
α
s
φ α
φ
x y t φ u xv
Rm
Rn
3.9. E XAMPLE . The formula
φ w xx x ayw ln b xx xz x c x
1
2
3
1 2
1 2
38 3. PULLING BACK FORMS
defines a map φ : U R2 , where U {
x R2 x 1 x 2 0 . The components |Y} ~
of φ are given by φ1 x1 , x2 3
x1 x2 and φ2 x1 , x2 |
ln x1 x2 . Accordingly, |
φ dy1 | dφ1 | d x x | 3x x dx x dx ,
3
1 2
2
1 2 1
3
1 2
φ dy2 | dφ2 | d ln x x | x x dx dx ,
1 2 1 2
1
1 2
φ dy1 dy2 | dφ1 dφ | 3x x dx x dx _ x x dx
2
2
1 2 1
3
1 2 1 2
1
1 dx2
| 3xx x x x dx dx .
2 3
2 2 1
1 2
1 2
Observe that the pullback operation turns k-forms on the target space V into
k-forms on the source space U. Thus, while φ : U V is a map from U to V, φ is {
a map
φ : V { U ,
k k
the opposite way from what you might naively expect. (Recall that k U stands
for the collection of all k-forms on U.) The property that φ “turns the arrow
around” is called contravariance. Pulling back forms is nicely compatible with the
other operations that we learned about (except the Hodge star).
3.10. P ROPOSITION . Let φ : U
m
{
V be a smooth map, where U is open in R n and
V is open in R . The pullback operation is
(i) linear: φ aα bβ | |
aφ α bφ β;
(ii) multiplicative: φ αβ
φ α φ β ;
(iii) natural: φ ψ α |
ψ φ α , where ψ : V { W is a second smooth map
with W open in Rk and α a form on W.
The term “natural” in property (iii) is a mathematical catchword meaning that
a certain operation (in this case the pullback) is well-behaved with respect to com-
position of maps.
P ROOF. If α |
∑ I f I dy I and β ∑ I g I dy I are two forms of the same degree, |
|
then aα bβ ∑ I a f I bg I dy I , so
φ aα bβ | ∑φ
I
af I _
bg I φ dy I .
Now
φ a fI bg I x | a f I bg I φ x _ | a fI φ x `
bg I φ x
| aφ f x I bφ g I x ,
so φ aα bβ |
∑ I aφ f I bφ g I φ dy I
(i). For the proof of part (ii) consider two forms α
| |
aφ α bφ β. This proves part _ |
∑ I f I dy I and β ∑ J g J dy J
(not necessarily of the same degree). Then αβ ∑ I,J f I g J dy I dy J , so |
φ αβ | ∑ φ f g φ dy dy . I J I J
I,J
Now
φ f g _ x |
I J | f φ x g φ xj | φ f _ φ g _ x ,
fI gJ φ x I J I J
3.2. PULLING BACK FORMS 39
so φ f g φ f φ g . Furthermore,
I J I J
φ dy dy * φ dy φ dy dy dy dy
I J i1 i2 ik j1 jl
dφ dφ dφ dφ dφ φ dy φ dy ,
i1 i2 ik j1 jl I J
so
φ αβ ∑ φ
f I φ g J φ dy I φ dy J
I,J
∑ φ f φ dy @ ∑ φ g φ dy φ α φ β ,
I
I I
I
J J
which establishes part (ii).
For the proof of property (iii) first consider a function f on W. Then
ψ f φ x f ψ φ x j f ψ φ x
φ ψ f x
f ψ φ x* ψ φ f x ,
so φ ψ f ψ φ f . Next consider a 1-form α dz on W, where z , z , . . . ,
z are the variables on R . Then ψ α dψ ∑
i 1 2
k m ∂ψ i
k dy , so i j 1 ∂y j j
φ ψ α ∑ φ φ dy ∑ φ
∂y
m m
∂ψ i ∂ψ i
dφ j j
∂y
j 1 j j 1 j
∑ φ ∂∂yψ ∑ ∂∂xφ dx ∑ ∑ φ ∂∂yψ ∂∂xφ dx .
m n n m
i j i j
l l
j 1 j l 1 l l 1 j 1 j l
By the chain rule, formula (B.6), the sum ∑ φ ∂ψ ∂y ∂φ ∂x is equal to m
∂ φ ψ ∂x . Therefore
j 1 i j j l
i l
∂ φ ψ
φ ψ α ∑ dx d φ ψ d ψ φ ψ φ dz ψ φ α .
n
i
∂xl 1 l
l i i i
Because every form on W is a sum of products of forms of type f and dz , property
(iii) in general follows from the two special cases α f and α dz .
i
QED i
Another application of the chain rule yields the following important result.
3.11. T HEOREM . Let φ : U V be a smooth map, where U is open in Rn and V is
m
open in R . Then φ dα
d φ α for α k V . In short ¡£¢
φ d dφ .
P ROOF. First let f be a function. Then
φ ∑ ∂y∂ f dy ∑ φ ∂y∂ f
m m
∑ φ ∂y∂ f ∑ ∂∂xφ dx
m n
i
φ df i dφi j
i 1 i i 1 i i 1 i j 1 j
∑ ∑ φ ∂y∂ f
n m
∂φi
dx j .
j 1i 1 i ∂x j
40 3. PULLING BACK FORMS
By the chain rule, formula (B.6), the quantity ∑m
¦ ¥ ¨§
∂ φ f ∂x j . Hence
i 1 φ ∂ f ∂y i ∂φ i ∂x j is equal to ¤ ¥¦ § ¨ §
¥ © ∑
n
∂ φ f ¦ ¥ ¨ © ¥
¤
φ df dx j dφ f ,
j 1
∂x j
so the theorem is true for functions. Next let α © ∑ I f I dy I . Then dα © ∑ I d f I dy I ,
so
¥ © ∑ φ ¥ ¦ d f dy ¨ ©Y¦ φ ¥ d f ¨ ¦ φ¥ dy ¨ © ∑ d ¦ φ¥ f ¨ dφ dφ ªªª dφ
φ dα I I I I I i1 i2 ik ,
I I
because φ ¥ d f © dφ ¥ f . On the other hand,
I I
dφ ¥ α © ∑ d «@¦ φ ¥ f ¨ ¦ φ ¥ dy ¨@¬ © ∑ d «@¦ φ ¥ f ¨ dφ dφ ªªª dφ ¬
I I I i1 i2 ik
I I
© ∑ d ¦ φ ¥ f ¨ dφ dφ ªªª dφ ∑ ¦ φ ¥ f ¨ d ¦ dφ dφ ªªª dφ ¨
I i1 i2 ik I i1 i2 ik
I I
© ∑ d ¦ φ ¥ f ¨ dφ dφ ªªª dφ .
I
I i1 i2 ik
Here we have used the Leibniz rule for forms, Proposition 2.5(ii), plus the fact that
the form dφi1 dφi2
equations above we see that φ dα dφ α .
ªªª
dφik is always closed. (See Exercise 2.8.) Comparing the two
¥ ©
QED ¥
We finish this section by giving an explicit formula for the pullback φ α , which ¥
establishes a connection between forms and determinants. Let us do this first in
degrees 1 and 2. The pullback of a 1-form α ∑m i 1 f i dy i is © ¤
φ ¥ α © ∑ ¦ φ ¥ f ¨ ¦ φ ¥ dy ¨ © ∑ ¦ φ ¥ f ¨ dφ .
m m
¤ i 1 ¤ i i
i 1
i i
Now dφi © ∑nj ¤ ∂φ i
1 ∂x j dx j and so
¥ © m
∑ ¦ φ¥ f ¨ ∑
n
∂φi
¯ © ∑¤ ∑¤ ¦ φ ¥ f ¨ ∂∂xφ dx © ∑¤ g dx ,
n m
i
n
¤ ® ¤
φ α i dx j i j j j
i 1 j 1
∂x j j 1i 1 j j 1
with g © ∑ ¤ ¦ φ ¥ f ¨
j
m . ∂φ i
i ∂x j
For a 2-form α © ∑ ° ± ° f dy dy we get
i 1
1 i j m i, j i j
φ¥ α © ¦ φ¥ f ¨ φ ¥ ¦ dy dy ¨ © ∑ ¦ φ¥ f ¨ dφ dφ .
° ∑± ° 1 i j m ° ± ° i, j i j
1 i j m
i, j i j
Observe that
© ∂φi ∂φ j
© ∂φi ∂φ j ∂φi ∂φ j
n
dφi dφ j ∑ ∑ ³ dxk dxl ,
´
¤
dxk dxl
k,l 1
∂xk ∂xl 1 k l n ° ± ° ² ∂xk ∂xl ∂xl ∂xk
where
∂φ i ∂φ i
∂φi ∂φ j
³
∂φi ∂φ j
©¶µµ ∂x k ∂x l
µµ
∂xk ∂xl ∂xl ∂xk
µµ ∂φ j ∂φ j
µµ
µµ µµ
∂x k ∂x l
3.2. PULLING BACK FORMS 41
·
is the determinant of the 2 2-submatrix obtained from the Jacobi matrix Dφ by
extracting rows i and j and columns k and l. So we get
∂φ i ∂φ i
¸ ¹ φ¸
º ∑» º >¼ ½ ¾ º ∑» º ¿¿
dx dx À
∂x k ∂x k
f i, j
¿¿
φ α ∂φ j ∂φ j k l
¿¿ ¿¿
1 i j m 1 k l n ∂x k ∂x l
¹ ∑ ∑ φ ¸ ¿ f ¾ ¿¿ dx dx ¹ ∑ g dx dx
¿ ∂φ i ∂φ i
∂x k ∂x k
º » º º » º ½ ¿¿ ¿¿ i, j ∂φ j
º » º ∂φ j k l k,l k l
¿ ¿
1 k l n1 i j m ∂x k ∂x l 1 k l n
¿¿ ¿¿
with
¿ ¿ ∂φ i ∂φ i
g ¹ ¸
º ∑» º ½ ¾ ¿¿
∂x k ∂x l
f .
¿¿
φ k,l i, j ∂φ j ∂φ j
¿¿ ¿¿
1 i j m ∂x k ∂x l
For an arbitrary k-form α ¹ ∑ f dy we obtain
¿¿ I I I
¿¿
φ ¸ α ¹ ∑ φ ¸ f ¾ φ ¸ dy dy ÁÁÁ dy ¾ ¹ ∑ φ ¸ f ¾ dφ dφ ÁÁÁ dφ .
½ I ½ I i1 i2
½ ik
I
I i1 i2 ik
To write the product dφ dφ ÁÁÁ dφ in terms of the x-variables we use
i1 i2 ik
dφ ¹ ∑
∂φ n
il
dx
 ∂x il ml
ml 1 ml
for l ¹ 1, 2, . . . , k. This gives
dφ dφ ÁÁÁ dφ ¹
∂φ ∂φ n ∂φ
i1 i2 ∑ ik
 ∂x ∂x ∂x m 1 ,m 2 ,...,m k 1
Á
Á Á m1
i1
dx dx ÁÁÁ dx
m2
i2
mk
ik
m1 m2 mk
¹ ∑ ∂x∂φ ∂x∂φ ÁÁÁ ∂x∂φ dx ,
M m1
i1 i2
m2
ik
mk
M
in which the summation is over all n multi-indices M ¹ k m , m , . . . , m ¾ . If a
multi-index M has repeating entries, then dx ¹ 0. If the½ entries of M are all
1 2 k
M
distinct, we can rearrange them in increasing order by means of a permutation σ .
¹ ¾ ¹ ½ ÃÄ ÃÄ Ã Ä¾
½
In other words, we have M m1 , m2 , . . . , mk jσ 1 , jσ 2 , . . . , jσ k , where
¹ ¾ Å
½
J j1 , j2 , . . . , jk is an increasing multi-index and σ S k is a permutation. Thus
we can rewrite the sum over all multi-indices M as a double sum over all increas-
ing multi-indices J and all permutations σ :
ÁÁÁ dφ ¹
∂φi1 ∂φi2 ∂φik
Æ∑ ∑
Ç È Ç È ÁÁÁ ∂x Ç È Ç È Ç È ÁÁÁ Ç È
dφi1 dφi1 ik dx dx dx jσ jσ jσ
J σ Sk ∂x jσ 1 ∂x jσ 2 jσ k
1 2 k
¹ ∑ ∑ sign σ ¾ ∂x∂φ ∂x∂φ ÁÁÁ ∂x∂φ dx i1 i2 ik
Æ ½
(3.2)
J σ Sk ÇÈ ÇÈ ÇÈ jσ 1 jσ 2 jσ k
J
¹ ∑ det Dφ I,J dx J .
J
(3.3)
In (3.2) used the result of Exercise 3.7 and in (3.3) we applied Theorem 3.6. The
notation Dφ I,J stands for the I, J-submatrix of Dφ, that is the k k-matrix obtained ·
from the Jacobi matrix by extracting rows i 1 , i2 , . . . , ik and columns j1 , j2 , . . . , jk .
42 3. PULLING BACK FORMS
To sum up, we find
É Ê ∑ φ É f I ∑ det Dφ I,J dx J Ê ∑ ∑ φ É Í Î
Ë Ì
φ α f I det Dφ I,J dx J .
I J J I
This proves the following result.
3.12. T HEOREM . Let φ : U Ï
V be a smooth map, where U is open in Rn and V is
m
open in R . Let α Ê
∑ I f I dy I be a k-form on V. Then φ α is the k-form on U given by É
É Ê
φ α ∑ J g J dx J with
Ê ∑ φÉ Í
gJ
I Ì f I det Dφ I,J .
This formula is seldom used to calculate pullbacks in practice and you don’t
need to memorize the details of the proof. It is almost always easier to apply the
definition of pullback directly. However, the formula has some important theoret-
ical uses, one of which we record here.
Ê Ê
Assume that k m n, that is to say, the number of new variables is equal to
the number of old variables, and we are pulling back a form of top degree. Then
α Ê f dy1 dy2 É Ê Ì φ É f Í Ì det Dφ Í dx dx ÐÐÐ dx .
Ð ÐÐ dy ,
n φ α 1 2 n
If f Ê 1 (constant function) then φ É f Ê 1, so we see that det Dφ Ì xÍ can be in-
terpreted as the ratio between the oriented volumes of two infinitesimal blocks
positioned at x: one with edges dx 1 , dx2 , . . . , dx n and another with edges dφ 1 ,
dφ2 , . . . , dφ n . Thus the Jacobi determinant is a measurement of how much the
map φ changes oriented volume from point to point.
3.13. T HEOREM . Let φ : U Ï
V be a smooth map, where U and V are open in R n .
Then the pullback of the volume form on V is equal to the Jacobi determinant times the
volume form on U,
φ dy1 dy2ÉÌ ÐÐÐ dy Í Ê Ì det Dφ Í dx dx ÐÐÐ dx .
n 1 2 n
Exercises
3.1. Deduce Theorem 3.7(iv) from Theorem 3.6.
ÑÑ ÑÑ ÑÑ ÑÑ
3.2. Calculate the following determinants using column and/or row operations and
ÑÑ ÑÑ ÑÑ 1 Ò ÑÑ
Theorem 3.7(iv).
1 3
ÑÑ 1 1
ÑÑ ÑÑ 0 1 2 4
ÑÑ
2 1
ÑÑ Ò 5 2
,
ÑÑ ÑÑ 2 Ò 1 1 3
ÑÑ
.
Ò
1 1 2 3 1 1 0
4 1 3 7 3 1 2 5
3.3. Tabulate all permutations in S 4 with their lengths and signs.
3.4. Determine the length and the sign of the following permutations.
Ó Ò Ò Ô Õ Ö
Õ
(i) A permutation of the form 1, 2, . . . , i 1, j, . . . , j 1, i, . . . , n where 1 i
j n. (Such a permutation is called a transposition. It interchanges i and j and
Ó Ò Ò Ô
leaves all other numbers fixed.)
(ii) n, n 1, n 2, . . . , 3, 2, 1 .
3.5. Find all permutations in S n of length 1.
EXERCISES 43
× ×
3.6. Calculate σ 1 , τ 1 , στ and τσ , where
ØÚÙ Û ØÚÙ Û
ØÙ Ü Û ØÝÙ Ü Ü Û
(i) σ 3, 6, 1, 2, 5, 4 and τ 5, 2, 4, 6, 3, 1 ;
(ii) σ 2, 1, 3, 4, 5, . . . , n 1, n and τ n, 2, 3, . . . , n 2, n 1, 1 (i.e. the trans-
positions interchanging 1 and 2, resp. 1 and n).
3.7. Show that
Þ ß Þ ßáà à à dx Þ ß Ø sign Ù σ Û dx dx à à à dx
dxiσ 1 dxiσ 2 iσ i1 i2 ik
for any multi-index Ù i , i , . . . , i Û and any permutation σ in S . (First show that the identity
k
1 2 k k
is true if σ is a transposition. Then show it is true for an arbitrary permutation σ by writing
σ as a product σ σ à àjà σ of transpositions and using formula (3.1) and Exercise 3.4(i).)
3.8. Show that for n â 2 the permutation group S has n! ã 2 even permutations and
1 2 l
n! ã 2 odd permutations.
n
3.9. (i) Show that every permutation has the same length and sign as its in-
verse.
(ii) Deduce Theorem 3.7(ii) from Theorem 3.6.
ØÚÙ Ü ä ä Û
ä Ü
3.10. The i-th simple permutation is defined by σ i 1, 2, . . . , i 1, i 1, i, i 2, . . . , n .
So σi interchanges i and i 1 and leaves all other numbers fixed. S n has n 1 simple
permutations, namely σ 1 , σ2 , . . . , σn 1 . Prove the Coxeter relations
Ø
(i) σi2 1 for 1 i n, å æ ×
Ù ç Û Ø
(ii) σiσi 1 3 1 for 1 i å æ n Ü 1,
Ù Û Ø
(iii) σiσ j 2 1 for 1 i, j å
æ n and i ä 1 æ j.
3.11. Let σ be a permutation of è 1, 2, . . . , n é . The permutation matrix corresponding to
σ is the n ê n-matrix A whose i-th column is the vector e ë ì . In other words, A e Ø e ë ì .
σ σ i σ i σ i
(ii) Show that A Ø A A .
(i) Write down the permutation matrices for all permutations in S . 3
(iii) Show that det A Ø sign Ù σ Û .
στ σ τ
σ
3.12. (i) Suppose that A has the shape
ðòññ
a a ... a
ñ 1,1 1,2 1,n
A ØUíî .
0 a ... a
îîï .. ... . ó
2,2 2,n
.. ,
0 a ... a n,1 n,n
i.e. all entries below a 11 are 0. Deduce from Theorem 3.6 that
a2,2 ... a2,n
det A Ø a1,1 ôô .. .. .
ôô
ôô ôô . .
ô ...
ôô
(ii) Deduce from this the expansion ô rule, Theorem 3.7(iii).
an,2 an,n
3.13. Show that
ô ô
1 1 ... 1
ôô x1 x2 ... xn
ôô Ø
ôô x21 x22 ... x2n
ôô ∏õ Ù x Ü x Û
ôô .. .. ..
ôô j i
ôô x × . . . x × ôô
i j
x ×
. . .
ô ôô from each row subtract x times the
n 1 n 1 n 1
for any numbers x , x , . .ôô . , x . (Starting at the bottom,
1 2 n
1 2 n ô
row above it. This creates a new determinant whose first column is the standard basis vector
1
e1 . Expand on the first column and note that each column of the remaining determinant has
a common factor.)
44 3. PULLING BACK FORMS
ö÷ x ö÷ x x . Find
x1 x1 x2
3.14. Let φ
x øùûú x x øù
2 1 3
(i) φ ü dy , φ ü dy , φ ü dy ;
3 2 3
(ii) φ ü_ý y y y þ , φ ü_ý dy dy þ ;
1 2 3
(iii) φ ü ý dy dy dy þ .
1 2 3 1 2
1 2 3
3
ö
x 1
3.15. Let φ ÿ
2
ú ÷ xxx øù
x 1 x x 1 2
. Find 2
x 2 1 2
3
(i) φ ü_ý y y þ;
2
(ii) φ ü dy , φ ü dy , φ ü dy , φ ü dy ;
1 3y 3y 2 3 4
(iii) φ ü_ý dy dy þ .
1 2 3 4
2 3
3.16. Compute ψ ü_ý x dy dz y dz dx þ
z dx dy , where ψ is the map R2 R3 defined
in Exercise B.7.
ö÷ θ ö÷ r cos φ sin θ
r r cos φ cos θ
3.17. Let P3 be spherical coordinates in R3 .
øù ú
r sin φ øù
ü
φ
(i) Calculate P3 α for the following forms α :
dx, dy, dz, dx dy, dx dz, dy dz, dx dy dz.
(ii) Find the inverse of the matrix DP3 .
3.18 (spherical coordinates in n dimensions). In this problem let us write a point in R n
as
r
ö
ø
θ1
÷
.. .
.
θn ù
ýþ
1
Let P1 be the function P1 r r. For each n 1 define a map Pn 1: Rn 1 Rn 1 by
ú r
ö θ.
r
ö ý cos θ þ P ö θ1
øø
÷ ø ú ÷ ÷
1 .. .
n n
Pn
ù
1 .. .
ù
ù
θn 1
θn
r sin θn
(This is an example of a recursive definition. If you know P1 , you can compute P2 , and then
P3 , etc.)
(i) Show that P2 and P3 are the usual polar, resp. spherical coordinates on R 2 , resp.
R3 .
(ii) Give an explicit formula for P4 .
ý þ ý
(iii) Let p be the first column vector of the Jacobi matrix of Pn . Show that Pn rp.
(iv) Show that the Jacobi matrix of Pn 1 is a n 1 n 1 -matrix of the form þ ú
DPn ÿ Av u
,
ú
1
w
where A is an n n-matrix, u is a column vector, v is a row vector and w is a
function given respectively by
A cos θn DPn , ý sin θ þ P , u
v ú ý sin θ , 0, 0, . . . , 0 þ , ú
n n
w r cos θ .
ú ú
n n
EXERCISES 45
(v) Show that det DPn 1 r cosn 1 θn det DPn for n 1. (Expand det DPn 1 with
respect to the last row, using the formula in part (iv), and apply the result of part
(iii).)
(vi) Using the formula in part (v) calculate det DPn for n 1, 2, 3, 4.
(vii) Find an explicit formula for det DPn for general n.
(viii) Show that det DPn 0 for r 0.
CHAPTER 4
Integration of 1-forms
Like functions, forms can be integrated as well as differentiated. Differenti-
ation and integration are related via a multivariable version of the fundamental
theorem of calculus, known as Stokes’ theorem. In this chapter we investigate the
case of 1-forms.
4.1. Definition and elementary properties of the integral
Let U be an open subset of Rn . A parametrized curve in U is a smooth mapping
c : I U from an interval I into U. We want to integrate over I. To avoid problems
with improper integrals we assume I to be closed and bounded, I a, b . (Strictly
speaking we have not defined what we mean by a smooth map c : a, b U. The
easiest definition is that c should be the restriction of a smooth map c̃ : a ε, b
ε U defined on a slightly larger open interval.) Let α be a 1-form on U. The
pullback c α is a 1-form on a, b , and can therefore be written as c α g dt (where
t is the coordinate on R). The integral of α over c is now defined by
b
α c α g t dt.
c a,b ! a
More explicitly, writing α in components, α ∑ni" 1 fi dxi , we have
n n
dci
c α ∑ c f i dci ∑ c fi
dt
dt, (4.1)
i" 1 i" 1
so n b dci
c
α ∑ f i c t #
dt
t dt.
i" 1 a
4.1. E XAMPLE . Let U be the punctured plane R2 $ 0 % . Let c : 0, 2π U be
the usual parametrization of the circle, c t &' cos t, sin t , and let α be the angle
form,
y dx x dy
α .
x2 y2
2π
Then c α dt (see Example 3.8), so ( c α )( 0 dt 2π .
A curve c : a, b * U can be reparametrized by substituting a new variable,
t p s , where s ranges over another interval ā, b̄ . We shall assume p to be a
one-to-one mapping from ā, b̄ onto a, b satisfying p +, s - 0 for ā . s . b̄. Such
a p is called a reparametrization. The parametrized curve
c / p : ā, b̄ U
has the same image as the original curve c, but it is traversed at a different rate.
Since p + s 0 - 0 for all s 12 ā, b̄ we have either p + s &3 0 for all s (in which case p is
47
48 4. INTEGRATION OF 1-FORMS
increasing) or p 465 s 798 0 for all s (in which case p is decreasing). If p is increasing,
we say that it preserves the orientation of the curve (or that the curves c and c : p
have the same orientation); if p is decreasing, we say that it reverses the orientation
(or that c and c : p have opposite orientations). In the orientation-reversing case, c : p
traverses the curve in the opposite direction to c.
4.2. E XAMPLE . The curve c : ; 0, 2π <>= R2 defined by c 5 t 7@?A5 cos t, sin t 7 rep-
resents the unit circle in the plane, traversed at a constant rate (angular velocity)
of 1 radian per second. Let p 5 s 7B? 2s. Then p maps ; 0, π < to ; 0, 2π < and c : p,
regarded as a map ; 0, π <C= R2 , represents the same circle, but traversed at 2 ra-
dians per second. (It is important to restrict the domain of p to the interval ; 0, π < .
If we allowed s to range over ; 0, 2π < , then 5 cos 2s, sin 2s 7 would traverse the circle
twice. This is not considered a reparametrization of the original curve c.) Now
let p 5 s 7D?FE s. Then c : p : ; 0, 2π <G= R2 traverses the unit circle in the clockwise
direction. This reparametrization reverses the orientation; the angular velocity is
now E 1 radian per second. Finally let p 5 s 7H? 2π s 2 . Then p maps ; 0, 1 < to ; 0, 2π <
and c : p : ; 0, 1 <I= R2 runs once counterclockwise through the unit circle, but at a
variable rate. What is the angular velocity as a function of s?
It turns out that the integral of a form along a curve is almost completely in-
dependent of the parametrization.
4.3. T HEOREM . Let α be a 1-form on U and c : ; a, b <J= U a curve in U. Let
p : ; ā, b̄ <=K; a, b < be a reparametrization. Then
L
cα if p preserves the orientation,
α ?AN O
cM p E cα if p reverses the orientation.
O
P ROOF. It follows from the definition of the integral and from the naturality
of pullbacks (Proposition 3.10(iii)) that
L LQP L P
α ? 5 c : p 7#S α ? p ST5 c S α 7 .
cM p ā, b̄R ā, b̄R
Now let us write c S α ? g dt and t ? p 5 s 7 . Then p S 5 c S α 7&? p S 5 g dt 7U?A5 p S g 7 dp ?
5 p S g 7V5 dp W ds 7 ds, so
L L P L b̄
dp
α ? 5 pS g7 ds ? g 5 p 5 s 7#7 p 4 5 s 7 ds.
cM p ā, b̄R ds ā
b
On the other hand, cα ? g 5 t 7 dt, so by the substitution formula, Theorem B.7,
a
O O
we have c M p α ?YX c α, where the Z occurs if p 4\[ 0 and the E if p 4\8 0. QED
O O
Interpretation of the integral. Integrals of 1-forms play an important role
in physics and engineering. A curve c : ; a, b <*= U models a particle travelling
through the region U. Recall from Section 2.5 that to a 1-form α ? ∑ni] 1 Fi dxi cor-
responds a vector field F ? ∑ni] 1 Fi ei , which can be thought of as a force field acting
on the particle. Symbolically we write α ? F ^ dx, where we think of dx as an infin-
itesimal vector tangent to the curve. Thus α represents the work done by the force
field along an infinitesimal vector dx. From (4.1) we see that c S α ? F 5 c 5 t 7#7I^ c 4 5 t 7 dt.
Accordingly, the total work done by the force F on the particle during its trip along
c is the integral
L L L b
α ? F ^ dx ? F 5 c 5 t 7_7`^ c 4 5 t 7 dt.
c c a
4.2. INTEGRATION OF EXACT 1-FORMS 49
In particular, the work and the total work are nil if the force is perpendicular to
the path, as in the picture on the left. The work done by the force in the picture on
the right is negative.
Theorem 4.3 can be translated into this language as follows: the work done by the
force does not depend on the rate at which the particle speeds along its path, but
only on the path itself and on the direction of travel.
The field F is conservative if it can be written as the gradient of a function,
F a grad g. The function b g is called a potential for the field and is interpreted as
the potential energy of the particle. In terms of forms this means that α a dg, i.e.
α is exact.
4.2. Integration of exact 1-forms
Integrating an exact 1-form α a dg is easy once the function g is known.
4.4. T HEOREM (fundamental theorem of calculus in R n ). Let α a dg be an exact
1-form on an open subset U of Rn . Let c : c a, b de U be a parametrized curve. Then
f
α a g g c g b h_h`b g g c g a h#h .
c
P ROOF. By Theorem 3.11 we have c i α a c i dg a dc i g. Writing h g t hja c i g g t hja
g g c g t h#h we have c i α a dh, so
f fQk f b
α a ci α a dh a h g b hmb h g a h ,
c a,b l a
where we used the (ordinary) fundamental theorem of calculus, formula (B.1).
Hence n c α a g g c g b h#hb g g c g a h#h . QED
The physical interpretation of this result is that when a particle moves in a
conservative force field, its potential energy decreases by the amount of work done
by the field. This clarifies what it means for a field to be conservative: it means
that the work done is entirely converted into mechanical energy and that none is
dissipated by friction into heat, radiation, etc. Thus the fundamental theorem of
calculus “explains” the law of conservation of energy.
50 4. INTEGRATION OF 1-FORMS
It also yields a necessary and sufficient criterion for a 1-form on U to be exact.
A curve c : o a, b pq U is called closed if c r a sGt c r b s .
4.5. T HEOREM . Let α be a 1-form on an open subset U of R n . Then the following
statements are equivalent.
(i) α is exact.
(ii) u c α t 0 for all closed curves c.
(iii) u c α depends only on the endpoints of c for every curve c in U.
P ROOF. (i) tmv (ii): if α t dg and c is closed, then u c α t g r c r b s#sIw g r c r a s_sxt
0 by the fundamental theorem of calculus, Theorem 4.4.
(ii) tv (iii): assume u c α t 0 for all closed curves c. Let
c1 : o a1 , b1 pq U and c2 : o a2 , b2 pq U
be two curves with the same endpoints, i.e. c 1 r a1 sDt c2 r a2 s and c1 r b1 s0t c2 r b2 s .
We need to show that u c1 α tyu c2 α . After reparametrizing c 1 and c2 we may
assume that a1 t a2 t 0 and b1 t b2 t 1. Define a new curve c by
c1 r t s for 0 } t } 1,
c r t szt|{
c2 r 2 w t s for 1 } t } 2.
(First traverse c 1 , then traverse c 2 backwards.) Then c is closed, so u c α t 0. But
Theorem 4.3 implies u c α t)u c1 α w~u c2 α , so u c1 α t)u c2 α .
(iii) tmv (i): assume that, for all c, u c α depends only on the endpoints of c.
We must define a function g such that α t dg. Fix a point x 0 in U. For each point
x in U choose a curve cx : o 0, 1p`q U which joins x0 to x. Define
g r x sCt) α.
cx
We assert that dg is well-defined and equal to α . Write α t ∑ni 1 f i dxi . We must
show that ∂g ∂xi t f i . From the definition of partial differentiation,
∂g g r x hei s>w g r x s 1
r x sjt lim t lim α w~ α .
∂xi h 0 h h 0h cx hei cx
Now consider a curve c̃ composed of two pieces: for 0 } t } 1 travel from x0
to x along the curve cx and then for 1 } t } 2 travel from x to x hei along the
straight line given by l r t st x )r t w 1 s hei . Then c̃ has the same endpoints as
cx hei . Therefore u cx he α t)u c̃ α , and hence
i
∂g 1 1
r x sjt lim α w α t lim α α w α
∂xi h 0 h c̃ cx h 0 h cx l cx
1 1
t lim α t lim l α . (4.2)
h 0 h l h 0 h 1,2
4.3. THE GLOBAL ANGLE FUNCTION AND THE WINDING NUMBER 51
Let δi, j be the Kronecker delta, which is defined by δ i,i 1 and δi, j 0 if i j. Then
we can write l j t x j δi, j t 1 h, and hence l j t δi, j h. This shows that
n n
l α ∑ f j x t 1 hei dl j ∑ f j x t 1 hei l j t dt
j 1 j 1
n
∑ f j x t 1 hei δi, jh dt h f i x t 1 hei dt. (4.3)
j 1
Taking equations (4.2) and (4.3) together we find
∂g 1 2 1
x lim h f i x t 1 hei dt lim f i x shei ds
∂xi
h 0h 1
h 0 0
1 1
lim f i x shei ds f i x ds fi x .
0 h 0 0
This proves that g is smooth and that dg α. QED
This theorem, and its proof, can be used in many different ways. For example,
it tells us that once we know a 1-form α to be exact we can find an “antiderivative”
g x by integrating α along an arbitrary path running from a fixed point x0 to x.
(See Exercises 4.3–4.5 for an application.) On the other hand, the theorem also
enables us to detect closed 1-forms that are not exact.
4.6. E XAMPLE . The angle form α 1 R2 0 of Example 2.10 is closed,
but not exact. Indeed, its integral around the circle is 2π 0. Mark the con-
trast with closed 1-forms on Rn , which are always exact! (See Exercise 2.6.) This
phenomenon underlines the importance of being careful about the domain of def-
inition of a form.
4.3. The global angle function and the winding number
In this section we will have a closer look at the angle form and see that it
carries interesting information of a “topological” nature. Throughout this section
U will be the punctured plane R2 0 , α will denote the angle form,
y dx x dy
α ,
x2 y2
and ξ and η will denote the functions
x y
ξ ¡ , η ¡ .
x2 y2 x2 y2
Then α is a closed 1-form and ξ and η are smooth functions on U. In fact, ξ and
η are just the components of x ¢\£ x £ , the unit vector pointing in the direction of x.
These functions satisfy
α ξ dη ηdξ . (4.4)
(You will be asked to check this formula in Exercise 4.6.) Now let θ : U ¤¦¥ 0, 2π
be the angle between a point and the positive x-axis, chosen to lie in the interval
¥ 0, 2 π . Then ξ cos θ and η sin θ , so by equation (4.4)
α cos θ d sin θ sin θ d cos θ cos 2 θ dθ sin θ 2 dθ dθ .
This equation is not valid on all of U (it cannot be because we saw in Example 4.6
that α is not exact), but only where θ is differentiable, i.e. on the complement of the
52 4. INTEGRATION OF 1-FORMS
positive x-axis. Hence the nonexactness of α is closely related to the impossibility
of defining a global differentiable angle function on U. (The precise meaning of
this assertion will become clear in Exercise 4.6.)
However, along a curve c : § a, b ¨ © U we can define a continuous angle func-
tion, and the fact that α ª dθ almost everywhere suggests how: by integrating α
along c! For simplicity assume that a ª 0 and b ª 1. Start by fixing any ϑ 0 such
that cos ϑ0 ª ξ « c « 0 ¬#¬ and sin ϑ0 ª η « c « 0 ¬#¬ and then define
ϑ « t ¬jª ϑ0 ®°¯ c ² α.
0,t ±
The following result says that ϑ « t ¬ measures the angle between c « t ¬ and the posi-
tive x-axis (up to an integer multiple of 2π ) and that the function ϑ : § 0, 1 ¨ © R is
smooth. In this sense ϑ is a “differentiable choice of angle” along the curve c.
4.7. T HEOREM . The function ϑ is smooth and satisfies
ϑ « 0 ¬jª ϑ0 , cos ϑ « t ¬jª ξ « c « t ¬#¬ and sin ϑ « t ¬zª η « c « t ¬#¬ .
P ROOF. To see that ϑ « 0 ¬*ª ϑ 0 , plug t ª 0 into the definition of ϑ. To prove
the other assertions we rescale the curve c « t ¬ to a new curve c « t ¬#³\´ c « t ¬µ´ moving
on the unit circle. Let f « t ¬ and g « t ¬ be the x- and y-components of this new curve.
Then f « t ¬jª ξ « c « t ¬_¬ and g « t ¬zª η « c « t ¬#¬ and
2 2
f « t¬ g « t¬ ª 1
for all t. In other words f ª c ² ξ and g ª c ² η, so it follows from formula (4.4) that
c ² α ª f dg ¶ g d f . Therefore
t
ϑ « t ¬jª ϑ0 ® f « s ¬ g ¸¹« s ¬>¶ g « s ¬ f ¸6« s ¬»º ds.
0 ·
By the fundamental theorem of calculus, formula (B.2), ϑ is differentiable and
ϑ¸ ª f g¸ ¶ gf¸ . (4.5)
Since the right-hand side is smooth, ϑ is smooth as well. To prove that cos ϑ « t ¬@ª
f « t ¬ and sin ϑ « t ¬jª g « t ¬ for all t it is enough to show that the difference vector
¼ ¼
f « t¬ cos ϑ « t ¬
¶
g « t ¬¾½ sin ϑ « t ¬_½
has length 0. Its length is equal to
2 2
« f ¶ cos ϑ º g ¶ sin ϑ ¬ ª f2 g2 ¶ 2 « f cos ϑ g sin ϑ ¬ cos 2 ϑ sin 2 ϑ
·
ª 2 ¶ 2 « f cos ϑ g sin ϑ ¬ .
Hence we need to show that the function u ª f cos ϑ g sin ϑ is a constant equal
to 1. For t ª 0 we have
u « 0 ¬jª f « 0 ¬ cos ϑ0 g « 0 ¬ sin ϑ0 ª cos 2 ϑ0 sin 2 ϑ0 ª 1.
Furthermore, the derivative of u is
u¸ ª f ¸ cos ϑ ¶ f ϑ ¸ sin ϑ g ¸ sin ϑ gϑ ¸ cos ϑ
2
ª¿« f¸ ¶ g f¸ f gg ¸ ¬ cos ϑ « g¸ ¶ f 2 g¸ f g f ¸ ¬ sin ϑ by formula (4.5)
ª¿« f2f¸ f gg ¸ ¬ cos ϑ « g2 g ¸ f g f ¸ ¬ sin ϑ since f 2 g2 ª 1
ª f « f f¸ gg ¸ ¬ cos ϑ g « gg ¸ f f ¸ ¬ sin ϑ.
EXERCISES 53
Now f 2 À g2 Á 1 implies f f  À gg  Á 0, so u ÂÄà t Å Á 0 for all t. Hence u is a constant
function, so u à t Å Á 1 for all t. QED
It is useful to think of the vector à f à t Å , g à t Å#Å T Á c à t Å#Æ\Ç c à t ÅTÇ as a dial that points
in the same direction as the vector c à t Š.
0 0
As t increases from 0 to 1, the dial starts at the angle ϑ Ã 0 Å Á ϑ 0 , it moves around
the meter, and ends up at the final angle ϑ Ã 1 Å . The difference ϑ Ã 1 ÅIÈ ϑ Ã 0 Å measures
the total angle swept out by the dial.
4.8. C OROLLARY. If c : É 0, 1 ÊÌË U is a closed curve, then ϑ Ã 1 ÅzÈ ϑ Ã 0 Å Á 2π k,
where k is an integer.
P ROOF. By Theorem 4.7, c à 0 Å Á c à 1 Å implies
Í Í Í
cos ϑ à 0 Å , sin ϑ à 0 ŻΠÁ ξ à c à 0 Å#Å , η à c à 0 Å#Å#Î Á ξ à c à 1 Å#Å , η à c à 1 Å#Å#Î
Í
Á cos ϑ Ã 1 Å , sin ϑ Ã 1 Å Î .
In other words cos ϑ Ã 0 Å Á cos ϑ Ã 1 Å and sin ϑ Ã 0 Å Á sin ϑ Ã 1 Å , so ϑ Ã 0 Å and ϑ Ã 1 Å differ
by an integer multiple of 2π . QED
The integer k Á Ã 2π Å_Ï 1 Ð c α is called the winding number of the closed curve c
about the origin. It measures how many times the curve loops around the origin.
Á 1
winding number of a closed curve about origin α. (4.6)
2π Ñ c
4.9. E XAMPLE . By Example 4.1, the winding number of the circle c à t Å Á à cos t, sin t Å T
(0 Ò t Ò 2π ) is equal to 1.
Exercises
4.1. Consider the curve c : Ó 0, π Ô 2 Õ\Ö R2 defined by c × t ØGÙ)× a cos t, b sin t Ø T , where a
and b are positive constants. Let α Ù xy dx Ú x 2 y dy.
(i) Sketch the curve c for a Ù 2 and b Ù 1.
(ii) Find Û c α (for arbitrary a and b).
4.2. Restate Theorem 4.5 in terms of force fields, potentials and energy. Explain why
the result is plausible on physical grounds.
54 4. INTEGRATION OF 1-FORMS
4.3. Consider the 1-form α Ü)Ý x Ý a ∑niÞ 1 xi dxi on Rn ßJà 0 á , where a is a real constant.
For every x Ü â 0 let cx be the line segment starting at the origin and ending at x.
(i) Show that α is closed for any value of a.
(ii) Determine for which values of a the function g ã x ä9Ü¿å cx α is well-defined and
compute it.
(iii) For the values of a you found in part (ii) check that dg Ü α .
4.4. Let α æèç 1 ã Rn ßéà 0 áêä be the 1-form of Exercise 4.3. Now let c x be the halfline
pointing from x radially outward to infinity. Parametrize c x by travelling from infinity
inward to x. (You can do this by using an infinite time interval ã ßUë , 0 ì in such a way that
cx ã 0 äÜ x.)
(i) Determine for which values of a the function g ã x ä9Ü¿å cx α is well-defined and
compute it.
(ii) For the values of a you found in part (i) check that dg Ü α .
(iii) Show how to recover from this computation the potential energy for Newton’s
gravitational force. (See Exercise B.4.)
4.5. Let α æç 1 ã Rn ßà 0 áêä be as in Exercise 4.3. There is one value of a which is not
covered by Exercises 4.3 and 4.4. For this value of a find a smooth function g on R n ßJà 0 á
such that dg Ü α .
4.6. (i) Verify equation (4.4).
(ii) Let U Ü R2 ßéà 0 á . Prove that there does not exist a smooth function θ : U í
R satisfying cos θ ã x, y äîÜ x ïñð x2 ò y2 and sin θ ã x, y äîÜ y ïñð x2 ò y2 for all
ã x, y äóæ U. (Argue by contradiction, by letting α Üôã ß y dx ò x dy äöõ÷ã x 2 ò y 2 ä
and showing that α Ü dθ if θ was such a function.)
4.7. Calculate directly from the definition the winding number about the origin of the
curve c : ø 0, 2π ì°í R2 given by c ã t ämÜã cos kt, sin kt ä T .
4.8. Let x0 be a point in R2 and c a closed curve which does not pass through x 0 .
How would you define the winding number of c around x 0 ? Try to formulate two different
definitions: a “geometric” definition and a definition in terms of an integral over c of a
certain 1-form analogous to formula (4.6).
4.9. Let c : ø 0, 1 ìQí R2 ßBà 0 á be a closed curve with winding number k. Determine the
winding numbers of the following curves c̃ : ø 0, 1 ìmí R2 ßà 0 á by using the formula, and
then explain the answer by appealing to geometric intuition.
(i) t äÜ c ã 1 ß t ä ;
c̃ ã
(ii) t äÜ ρ ã t ä c ã t ä , where ρ : ø 0, 1 ìQíùã 0, ë ä is a function satisfying ρ ã 0 äÜ
c̃ ã ρ ã 1ä ;
(iii) t äÜÝ c ã t äúÝ_û 1 c ã t ä ;
c̃ ã
(iv) t äÜ φ ã c ã t äöä , where φ ã x, y ä>Üüã y, x ä T ;
c̃ ã
1 T
(v) c̃ ã t äÜ φ ã c ã t äöä , where φ ã x, y ä>Ü ã x, ß y ä .
x 2 ò y 2
4.10. For each of the following closed curves c : ø 0, 2π ìQí R2 ßýà 0 á set up the integral
defining the winding number about the origin. Evaluate the integral if you can (but don’t
give up too soon). If not, sketch the curve (the use of software is allowed) and obtain the
answer geometrically.
(i) cã t äÜüã a cos t, b sin t ä T , where a þ 0 and b þ 0;
(ii) cã t äÜüã cos t ß 2, sin t ä T ;
(iii) cã t ämÜüã cos3 t, sin3 t ä T ;
T
(iv) cã t äÜ)ÿöã a cos t ò b ä cos t ò ã b ß a äöõ 2, ã a cos t ò b ä sin t , where 0 b a.
EXERCISES 55
4.11. Let b 0 and a 0 be constants with a b. Define a planar curve c : 0, 2π
R2 0 by
T
c t a b cos t a cos a b t, a b sin t a sin a b t .
a a
(i) Sketch the curve c for a b 3.
(ii) For what values of a and b is the curve closed?
(iii) Assume c is closed. Set up the integral defining the winding number of c around
the origin and evaluate it. If you get stuck, find the answer geometrically.
4.12. Let U be an open subset of R2 and let F F1 e1 F2 e2 : U R2 be a smooth
vector field. The differential form
β F1 dF2 F2 dF1
F12 F22
is well-defined at all points x of U where F x 0. Let c be a parametrized circle contained
in U, traversed once in the counterclockwise direction. Assume that F x 0 for all x c.
The index of F relative to c is
1
index F, c β.
2π c
Prove the following assertions.
(i) β F α , where α is the angle form y dx x dy x2 y2 ! ;
(ii) β is closed;
(iii) index F, c is the winding number of the curve F " c about the origin;
(iv) index F, c is an integer.
4.13. (i) Find the indices of the following vector fields around the indicated
circles.
56 4. INTEGRATION OF 1-FORMS
(ii) Draw diagrams of three vector fields in the plane with respective indices 0, 2
and 4 around suitable circles.
CHAPTER 5
Integration and Stokes’ theorem
5.1. Integration of forms over chains
In this chapter we generalize the theory of Chapter 4 to higher dimensions. In
the same way that 1-forms are integrated over parametrized curves, k-forms can
be integrated over k-dimensional parametrized regions. Let U be an open subset
of Rn and let α be a k-form on U. The simplest k-dimensional analogue of an
interval is a rectangular block in Rk whose edges are parallel to the coordinate axes.
This is a set of the form
R #%$ a1 , b1 &(' $ a2 , b2 &('*)+)+)' $ ak , bk & #, t - Rk . ai / ti / bi for 1 / i/ k0 ,
where ai 1 bi . The k-dimensional analogue of a parametrized path is a smooth
map c : R 2 U. Although the image c 3 R 4 may look very different from the block
R, we think of the map c as a parametrization of the subset c 3 R 4 of U: each choice
of a point t in R gives rise to a point c 3 t 4 in c 3 R 4 . The pullback c 5 α is a k-form
on R and therefore looks like g 3 t 4 dt 1 dt2 )+)+) dtk for some function g : R 2 R. The
integral of α over c is defined as
6 6 6 bk 6 b2 6 b1
α# c5 α # )+)+) g 3 t 4 dt1 dt2 )+)+) dtk .
c R ak a2 a1
For k # 1 this reproduces the definition given in Chapter 4. (The definition makes
sense if we replace the rectangular block R by more general shapes in R k , such as
skew blocks, k-dimensional balls, cylinders, etc. In fact any compact subset of R k
will do.)
The case k # 0 is also worth examining. A zero-dimensional “block” R in
R0 #7, 0 0 is just the point 0. We can therefore think of a map c : R 2 U as a
collection , x 0 consisting of a single point x # c 3 0 4 in U. The integral of a 0-form
(function) f over c is by definition the value of f at x,
6
f # f 3 x4 .
c
As in the one-dimensional case, integrals of k-forms are almost wholly unaf-
fected by a change of variables. Let
R̄ #8$ ā1 , b̄1 & ' $ ā2 , b̄2 & '*)+)+)9' $ āk , b̄k &
be a second rectangular block. A reparametrization is a map p : R̄ 2 R satisfying
the following conditions: p is bijective (i.e. one-to-one and onto) and the k ' k-
matrix Dp 3 s4 is invertible for all s - R̄. Then det Dp 3 s 4;# : 0 for all s - R̄, so either
det Dp 3 s 4=< 0 for all s or det Dp 3 s 4 1 0 for all s. In these cases we say that the
reparametrizion preserves, respectively reverses the orientation of c.
57
58 5. INTEGRATION AND STOKES’ THEOREM
5.1. T HEOREM . Let α be a k-form on U and c : R > U a smooth map. Let p : R̄ > R
be a reparametrization. Then
?
α if p preserves the orientation,
α ACB E D c
c@ p cα if p reverses the orientation.
D
P ROOF. Almost verbatim the same proof as for k A 1 (Theorem 4.3). It follows
from the definition of the integral and from the naturality of pullbacks, Proposition
3.10(iii), that ? ? ?
αA c G p HJI α A pI cI α H .
c@ p R̄ F R̄ F
Now let us write c I α A g dt1 dt2 K+K+K dtk and t A p s H . Then
F
p I c I α HLA p I g dt1 dt2 K+K+K dtk HLA p I g H det Dp ds1 ds2 K+K+K dsk
F F F
by Theorem 3.13, so
? ?
αA g p sHJH det Dp s H ds 1 ds2 K+K+K dsk .
F F c@ p F R̄
On the other hand, c α A g t H dt1 dt2 K+K+K dtk , so by the substitution formula,
D DR
Theorem B.7, we have c @ p α ANM F c α , where the O occurs if det Dp P 0 and the
E if det Dp Q 0. D D
QED
5.2. E XAMPLE . The unit interval is the interval R 0, 1 S in the real line. Any curve
c : R a, bST> U can be reparametrized to a curve c G p : R 0, 1 SU> U by means of
the reparametrization p s HVA b E a H s O a. Similarly, the unit cube in R k is the
rectangular block F F
R 0, 1 S k AXW t Y Rk Z ti Y[R 0, 1 S for 1 \ i\ k] .
Let R be any other block, given by a i \ ti \ bi . Define p : R 0, 1 S k > R by p s H;A
As O a, where F
bdc bdc
b1 E a1 0 ... 0 c
c a1 cc
0 b2 E a2 ... 0 a2
A_
A ^` .. .. .. .. and a A7^` . .
` . . . . e ` .. e
`a `a
0 0 ... bk E ak ak
(“Squeeze the unit cube until it has the same edgelengths as R and then move it to
the position of R.”) Then p is one-to-one and onto and Dp sHfA A, so det Dp s HLA
F
det A A vol R P 0 for all s, so p is an orientation-preserving F
reparametrization.
Hence c @ p α A c α for any k-form α on U.
D D
5.3. R EMARK . A useful fact you learned in calculus is that one may inter-
change the order of integration in a multiple integral, as in the formula
? b1 ? b2 ? b2 ? b1
f t1 , t2 H dt1 dt2 A f t1 , t2 H dt2 dt1 . (5.1)
a1 a2 F a2 a1 F
(This follows for instance from the substitution formula, Theorem B.7.) On the
other hand, we have also learned that f t 1 , t2 H dt2 dt1 A E f t1 , t2 H dt1 dt2 . How
can this be squared with formula (5.1)? F The explanation is as F follows. Let α A
f t1 , t2 H dt1 dt2 . Then the left-hand side of formula (5.1) is the integral of α over
F
5.2. THE BOUNDARY OF A CHAIN 59
c : g a1 , b1 h i g a2 , b2 hkj R2 , the parametrization of the rectangle given by c l t 1 , t2 mon
l t1 , t2 m . The right-hand side is the integral of p α not over c, but over c q p, where
p : g a2 , b2 h(i g a1 , b1 h j g a1 , b1 h i g a2 , b2 h
is the reparametrization p l s 1 , s2 mLn l s2 , s1 m . Since p reverses the orientation, Theo-
rem 5.1 says that r c s p α n ptr c α ; in other words r c α n r c s p lup α m , which is exactly
formula (5.1). Analogously we have
vxw v w
f l t1 , t2 , . . . , tk m dt1 dt2 z+z+z dtk n f l t1 , t2 , . . . , tk m dti dt1 dt2 z+z+z|d{ ti z+z+z dtk
0,1 y k 0,1 y k
for any i.
We see from Example 5.2 that an integral over any rectangular block can be
written as an integral over the unit cube. For this reason, from now on we shall
usually take R to be the unit cube. A smooth map c : g 0, 1 h k j U is called a k-cube
in U (or sometimes a singular k-cube, the word singular meaning that the map c is
not assumed to be one-to-one, so that the image can have self-intersections.)
It is often necessary to integrate over regions that are made up of several
pieces. A k-chain in U is a formal linear combination of k-cubes,
c n a1 c1 } a2 c2 }~z+z+z|} a p c p ,
where a1 , a2 , . . . , a p are real coefficients and c 1 , c2 , . . . , c p are k-cubes. For any
k-form α we then define
v p v
αn ∑ ai α.
c i 1 ci
(In the language of linear algebra, the k-chains form an abstract vector space with
a basis consisting of the k-cubes. Integration, which is a priori only defined on
cubes, is extended to chains in such a way as to be linear.)
Recall that a 0-cube is nothing but a singleton x consisting of a single point
p
x in U. Thus a 0-chain is a formal linear combination of points, c n ∑i 1 ai xi .
A good way to think of c is as a collection of p point charges, with an electric
charge ai placed at the point xi . (You must carefully distinguish between the formal
p
linear combination ∑i 1 ai xi , which represents a distribution of point charges,
p
and the linear combination of vectors ∑i 1 ai xi , which represents a vector in Rn .)
The integral of a function f over the 0-chain is by definition
v p v p
f n ∑ ai
xi
f n ∑ a i f l xi m .
c i 1 i 1
p
Likewise, a k-chain ∑i 1 ai ci can be pictured as a charge distribution, with an elec-
tric charge ai spread along the k-dimensional “patch” c i .
5.2. The boundary of a chain
Consider a curve (“1-cube”) c : g 0, 1 h j U. Its boundary is by definition the
0-chain defined by ∂c n c l 1 m p[ c l 0 m .
c 1 d
c 0 d c
60 5. INTEGRATION AND STOKES’ THEOREM
The boundary of a 2-cube c : 0, 1 2 U consists of four pieces corresponding
to the edges of the unit square: c 1 t c t, 0 , c2 t c 1, t , c3 t c t, 1 and
c4 t c 0, t . The picture below suggests that we should define ∂c c 1 c2
c3 c4 .
c
(Alternatively we could define ∂c c 1 c2 c̄3 c̄4 , with c̄3 t c 1 t, 1 and
c̄4 t o c 0, 1 t , which corresponds to the following picture:
c
This would work equally well, but is technically less convenient.)
A k-cube c : 0, 1 k U has 2k faces of dimension k 1, which are described
as follows. Let t t1 , t2 , . . . , tk 1 [ 0, 1 k 1 and for i 1, 2, . . . , k put
ci,0 t f c t1 , t2 , . . . , ti 1 , 0, t i , . . . , t k 1 ,
ci,1 t f c t1 , t2 , . . . , ti 1 , 1, t i , . . . , t k 1 .
(“Insert 0, resp. 1 in the i-th slot”.) Now define
k k
i ρ
∂c ∑ u 1 i ci,0 ci,1 f ∑ ∑ u 1 ci,ρ .
i 1 i 1 ρ 0,1
For an arbitrary k-chain c ∑i ai ci we put ∂c ∑ i ai ∂ci . Then ∂ is a linear map
from k-chains to k 1-chains. You should check that for k 0 and k 1 this
definition is consistent with the one- and two-dimensional cases considered above.
There are a number of curious similarities between the boundary operator ∂
and the exterior derivative d, the most important of which is the following. (There
are also many differences, such as the fact that d raises the degree of a form by 1,
whereas ∂ lowers the dimension of a chain by 1.)
5.4. P ROPOSITION . ∂ ∂c L 0 for every k-chain c in U. In short,
∂2 0.
P ROOF. By linearity of ∂ it suffices to prove this for k-cubes c : 0, 1 k U. Let
t t1 , t2 , . . . , tk 2 0, 1 k 2 and let ρ and σ be 0 or 1. Then for 1 i j k 1
5.3. CYCLES AND BOUNDARIES 61
we have
ci,ρ ¡ j,σ t1 , t2 , . . . , tk ¢ 2 ¡f£ ci,ρ t1 , t2 , . . . , t j ¢ 1, σ , t j , . . . , tk ¢ 2 ¡
£ c t1 , t2 , . . . , ti ¢ 1 , ρ, ti , . . . , t j ¢ 1 , σ , t j , . . . , tk ¢ 2 ¡ .
On the other hand,
c j¤ 1,σ ¡ i,ρ t1 , t2 , . . . , tk ¢ 2 ¡L£ c j¤ 1,σ t1 , t2 , . . . , ti ¢ 1 , ρ, ti , . . . , tk ¢ 2 ¡
£ c t1 , t2 , . . . , ti ¢ 1 , ρ, ti , . . . , t j ¢ 1, σ , t j , . . . , tk ¢ 2 ¡ ,
because in the vector t 1 , t2 , . . . , ti ¢ 1 , ρ, ti , . . . , tk ¢ 2 ¡ the entry t j occupies the j ¥ 1st
slot! We conclude that c i,ρ ¡ j,σ £ c j ¤ 1,σ ¡ i,ρ for 1 ¦ i ¦ j ¦ k § 1. Therefore the
k § 2-chain ∂ ∂c ¡ is given by
k k
∂ ∂c ¡o£ ∂∑ ∑ § 1 ¡ i ¤ ρci,ρ £ ∑ ∑ § 1 ¡ i ¤ ρ∂ci,ρ
i ¨ 1 ρ ¨ 0,1 i ¨ 1 ρ ¨ 0,1
k k¢ 1
£ ∑∑ ∑ § 1¡ i¤ j¤ ρ¤ σ
ci,ρ ¡ j,σ .
i ¨ 1 j ¨ 1 ρ ,σ ¨ 0,1
The double sum over i and j can be rearranged in a sum over i ¦ j and a sum over
i © j to give
∂ ∂c ¡L£ ∑ ∑ § 1¡ i¤ j¤ ρ¤ σ
ci,ρ ¡ j,σ
1 ª i ª j ª k ¢ 1 ρ ,σ ¨ 0,1
¥ ∑ ∑ § 1¡ i¤ j¤ ρ¤ σ
ci,ρ ¡ j,σ . (5.2)
1 ª j « i ª k ρ ,σ ¨ 0,1
In the first term on the right in (5.2) we substitute c i,ρ ¡ j,σ £ c j¤ 1,σ ¡ i,ρ and then
r £ j ¥ 1, s £ i, µ £ σ , ν £ ρ to get
∑ ∑ § 1¡ i¤ j¤ ρ¤ σ
ci,ρ ¡ j,σ £ ∑ ∑ § 1¡ i¤ j¤ ρ¤ σ
c j¤ 1,σ ¡ i,ρ
1 ª i ª j ª k ¢ 1 ρ ,σ ¨ 0,1 1 ª i ª j ª k ¢ 1 ρ ,σ ¨ 0,1
£ ∑ ∑ § 1 ¡ s¤ r¢ 1¤ ν ¤ µ
cr,µ ¡ s,ν
1 ª s « r ª k µ ,ν ¨ 0,1
£ § ∑ ∑ § 1¡ s¤ r¤ ν ¤ µ
cr,µ ¡ s,ν .
1 ª s « r ª k µ ,ν ¨ 0,1
Thus the two terms on the right in (5.2) cancel out. QED
5.3. Cycles and boundaries
A k-cube c is degenerate if c t 1 , . . . , tk ¡ is independent of ti for some i. A k-chain
c is degenerate if it is a linear combination of degenerate cubes. In particular, a de-
generate 1-cube is a constant curve. The work done by a force field on a motionless
particle is 0. More generally we have the following.
5.5. L EMMA . Let α be a k-form and c a degenerate k-chain. Then ¬ c α £ 0.
P ROOF. By linearity we may assume that c is a degenerate cube. Suppose c is
constant as a function of t i . Then
c t1 , . . . , ti , . . . , tk ¡f£ c t1 , . . . , 0, . . . , t k ¡f£ g f t1 , . . . , ti , . . . , tk ¡J® ,
62 5. INTEGRATION AND STOKES’ THEOREM
where f : ¯ 0, 1 ° k ± ¯ 0, 1° k ² 1 and g : ¯ 0, 1 ° k ² 1 ± U are given respectively by
f ³ t1 , . . . , ti , . . . , tk ´fµ
³ t1 , . . . , t̂i , . . . , tk ´ ,
g ³ s1 , . . . , sk ² 1 ´fµ c ³ s1 , . . . , si ² 1 , 0, si ¶ 1, . . . , tk ² 1 ´ .
Now g · α is a k-form on ¯ 0, 1 ° k ² 1 and hence equal to 0, and so c · α µ f ·¸³ g · α ´µ 0.
We conclude ¹ c α µ ¹+º 0,1 » k c · α µ 0. QED
So degenerate chains are irrelevant where integration is concerned. This mo-
tivates the following definition. A k-chain c is closed, or a cycle, if ∂c is a degenerate
k ¼ 1-chain. A k-chain c is a boundary if c µ ∂b ½ c ¾ for some k ½ 1-chain b and
some degenerate k-chain c ¾ .
5.6. E XAMPLE . If c 1 and c2 are curves arranged head to tail as in the picture
below, then c1 ½ c2 is a 1-cycle. Likewise, the closed curve c is a 1-cycle.
c2
c1
5.7. L EMMA . The boundary of a degenerate k-chain is a degenerate k ¼ 1-chain.
P ROOF. By linearity it suffices to consider the case of a degenerate k-cube c.
Suppose c is constant as a function of t i . Then ci,0 µ ci,1, so
∂c µ ∑À ³J¼ 1 ´ j ³ c j,0 ¼ c j,1 ´ .
j¿ i
Let t µ ³ t1 , t2 , . . . , tk ² 1 ´ . For j Á i the cubes c j,0 ³ t ´ and c j,1 ³ t ´ are independent of
ti and for j  i they are independent of t i ² 1 . So ∂c is a combination of degenerate
k ¼ 1-cubes and hence is degenerate. QED
5.8. C OROLLARY. Every boundary is a cycle.
P ROOF. Suppose c µ ∂b ½ c ¾ with c ¾ degenerate. Then by Lemma 5.5 ∂c µ
∂ ³ ∂b ´ ½ ∂c ¾ µ ∂c ¾ , where we used Proposition 5.4. Lemma 5.7 says that ∂c ¾ is
degenerate, and therefore so is ∂c. QED
5.9. E XAMPLE . Consider the unit circle in the plane c ³ t ´;µ ³ cos 2π t, sin 2π t ´
with 0 à t à 1. This is a closed 1-cube. The circle is the boundary of the disc
of radius 1 and therefore it is reasonable to expect that c is a boundary of a 2-
cube. This is indeed true in the sense defined above, that c µ ∂b ½ c ¾ where c ¾
is a constant 1-chain. (It is actually not possible to find a b such that c µ ∂b;
see Exercise 5.2.) The 2-cube b is defined by “shrinking c to a point”, b ³ t 1 , t2 ´Äµ
³ 1 ¼ t2 ´ c ³ t1 ´ for ³ t1 , t2 ´ in the unit square. Then
b ³ t1 , 0 ´oµ
c ³ t1 ´ , b ³ 0, t2 ´Lµ b ³ 1, t2 ´Lµ ³ 1 ¼ t2 , 0 ´ , b ³ t1 , 1 ´Åµ ³ 0, 0 ´ ,
so that ∂b µ c ¼ c ¾ , where c ¾ is the constant curve located at the origin. Therefore
c µ ∂b ½ c ¾ , a boundary plus a degenerate 1-cube.
In the same way that a closed form is not necessarily exact, it may happen that
a 1-cycle is not a boundary. See Example 5.11.
5.4. STOKES’ THEOREM 63
5.4. Stokes’ theorem
In the language of chains and boundaries we can rewrite the fundamental
theorem of calculus, Theorem 4.4, as follows:
Æ ÆÍÌ Æ¸Ì Æ¸Ì Ì Æ
dg Ç g È c È 1 ÉJÉkÊ g È c È 0 ÉdÉËÇ gÊ gÇ gÇ g,
c c Î 1 ÏÑÐ c Î 0 ÏÑÐ c Î 1 ÏÑÐJÒ c Î 0 ÏÑÐ ∂c
i.e. Ó c dg ÇÔÓ ∂c g. This is the form in which the fundamental theorem of calculus
generalizes to higher dimensions. This generalization is perhaps the neatest rela-
tionship between the exterior derivative and the boundary operator. It contains
as special cases the classical integration formulas of vector calculus (Green, Gauß
and Stokes) and for that reason has Stokes’ name attached to it, although it would
perhaps be better to call it the “fundamental theorem of multivariable calculus”.
5.10. T HEOREM (Stokes’ theorem). Let α be a k Ê 1-form on an open subset U of
Rn and let c be a k-chain in U. Then
Æ Æ
dα Ç α.
c ∂c
P ROOF. By the definition of the integral and by Theorem 3.11 we have
Æ ÆxÕ ÆÕ
dα Ç c × dα Ç dc × α .
c 0,1 Ö k 0,1 Ö k
Since c × α is a k Ê 1-form on Ø 0, 1 Ù k , it can be written as
k
c× α Ç ∑ gi dt1 dt2 Û+Û+Û|dÜ ti Û+Û+Û dtk
iÚ 1
for certain functions g 1 , g2 , . . . , gk defined on Ø 0, 1 Ù k . Therefore
Æ k Æ Õ k ÆxÕ
∂gi
dα Ç ∑ 0,1 Ö k
d Ý gi dt1 dt2 Û+Û+Û|dÜ ti Û+Û+Û dtk Þ Ç ∑ ÈuÊ 1 É i ß 1
0,1 Ö k ∂ti
dt1 dt2 Û+Û+Û dtk .
c iÚ 1 iÚ 1
Changing the order of integration (cf. Remark 5.3) and subsequently applying the
fundamental theorem of calculus in one variable, formula (B.1), gives
ÆxÕ ÆxÕ
∂gi ∂gi
dt1 dt2 Û+Û+Û dtk Ç dti dt1 dt2 Û+Û+Û|dÜ ti Û+Û+Û dtk
0,1 Ö ∂t i
k 0,1 Ö ∂t i
k
Æ Õ
Ç Ý gi È t1 , . . . , ti Ò 1 , 1, ti ß 1, . . . , tk É
0,1 Ö à
k 1
Ê gi È t1 , . . . , ti Ò 1 , 0, ti ß 1 , . . . , tk É Þ dt1 dt2 Û+Û+Û|dÜ ti Û+Û+Û dtk .
The forms
gi È t1 , . . . , ti Ò 1 , 1, t i ß 1 , . . . , t k É dt1 dt2 Û+Û+Û dÜ ti +Û Û+Û dtk and
gi È t1 , . . . , ti Ò 1 , 0, t i ß 1 , . . . , t k É dt1 dt2 Û+Û+Û|dÜ ti Û+Û+Û dtk
64 5. INTEGRATION AND STOKES’ THEOREM
á α , resp. ci,0
are nothing but ci,1 á α . Accordingly,
â k âxé
∂gi
dα ã ∑ 1ç iè 1
dti dt1 dt2 ë+ë+ëídì ti ë+ë+ë dtk
c i ä 1 åuæ 0,1 ê ∂t i
k
k â é
ã ∑ 1ç iè 1 á α
ci,1 á αç
ci,0
i ä 1 åuæ 0,1 ê î å
k 1
æ
k â é
ã ∑ ∑ 1ç iè ρ
ci,á ρα
i ä 1 ρ ä 0,1 åuæ 0,1 ê î
k 1
k â â
ã ∑ ∑ 1ç iè ρ
αã α,
i ä 1 ρ ä 0,1 åuæ c i,ρ ∂c
which proves the result. QED
5.11. E XAMPLE . The unit circle c t çïã cos 2π t, sin 2π t ç is a 1-cycle in the
punctured plane U ã R2 å
0 ñ . Considered å a chain in R2 it is also a boundary,
as
as we saw in Example 5.9. æð However, we claim that it is not a boundary in U in
the sense that there exist no 2-chain b and no degenerate 1-chain c ò both contained
in U such that c ã ∂b ó c ò . Indeed, suppose that c ã ∂b ó c ò . Let α ã y dx ó
x dy çdô x2 ó y2 ç be the angle form. Then õ c α ã 2π by Example 4.1.On the åuæ other
hand, å â â â
αã αã dα ã 0,
c ∂b è c ö b
where we have used Stokes’ theorem, Lemma 5.5 and the fact that α is closed. This
is a contradiction. The moral of this example is that the presence of the puncture
in U is responsible both for the existence of the non-exact closed 1-form α (see
Example 4.6) and for the closed 1-chain c which is not a boundary. We detected
both phenomena by using Stokes’ theorem.
Exercises
5.1. Let U be an open subset of Rn , V an open subset of Rm and φ : U ÷ V a smooth
map. Let c be a k-cube in U and α a k-form on V. Prove that ø c φ ù α úûø φ ü c α .
5.2. Let U be an open subset of Rn . Its boundary is a linear combination of k ý 1-cubes,
∂c ú ∑i ai ci .
(i) Let b be a k þ 1-chain in U. Its boundary is a linear combination of k-cubes,
∂b ú ∑i ai ci . Prove that ∑i ai ú 0.
(ii) Let c be a k-cube in U. Conclude that there exists no k þ 1-chain b in U satisfying
∂b ú c.
5.3. Define a 2-cube c : ÿ 0, 1 2 ÷
x1 dx3 þ x2 dx3 .
R3 by c t1 , t2 ú
t21 , t1 t2 , t22 , and let α ú x1 dx2 þ
(i) Sketch the image of c.
(ii) Calculate both ø c dα and ø ∂c α and check that they are equal.
5.4. Define a 3-cube c : ÿ 0, 1
x1 dx2 dx3 . Calculate both ø c dα and
3
ø
÷ R3 by c t1 , t2 , t3 ú
α and check that they
are equal.
t2 t3 , t1 t3 , t1 t2 , and let α ú
∂c
5.5. Using polar coordinates in n dimensions (cf. Exercise 3.18) write the n ý 1-dimen-
sional unit sphere S n 1 in Rn as the image of an n ý 1-cube c.For n ú 2, 3, 4, calculate
the boundary ∂c of this cube. (The domain of c will not be the unit cube in R n 1 , but a
EXERCISES 65
rectangular block R dictated by the formula in Exercise 3.18. Choose R in such a way as to
cover the sphere as economically as possible.)
5.6. Deduce the following classical integration formulas from the generalized version
of Stokes’ theorem. All functions, vector fields, chains etc. are smooth and are defined in an
open subset U of Rn . (Some formulas hold only for special values of n, as indicated.)
(i) grad g dx g c 1 g c 0 for any function g and any curve c.
c
∂g ∂f
(ii) Green’s formula: dx dy f dx g dy for any functions f , g
∂x ∂y c ∂c
and any 2-chain c. (Here n 2.)
(iii) Gauß’ formula: c
div F dx1 dx2 dxn ∂c
F dx for any vector field F and
any n-chain c.
(iv) Stokes’ formula: c
curl F
dx
∂c
F dx for any vector field F and any 2-chain
c. (Here n 3.)
In parts (iii) and (iv) we use the notations dx and dx explained in Section 2.5. We shall
give a geometric interpretation of the entity dx in terms of volume forms later on. (See
Corollary 8.15.)
CHAPTER 6
Manifolds
6.1. The definition
Intuitively, an n-dimensional manifold in the Euclidean space R N is a subset
that in the neighbourhood of every point “looks like” Rn up to “smooth distor-
tions”. The formal definition is given below and is unfortunately a bit long. It will
help to consider first the basic example of the surface of the earth, which is a two-
dimensional sphere placed in three-dimensional space. A useful way to represent
the earth is by means of a world atlas, which is a collection of maps. Each map
depicts a portion of the world, such as a country or an ocean. The correspondence
between points on a map and points on the earth’s surface is not entirely faithful,
because charting a curved surface on a flat piece of paper inevitably distorts the
distances between points. But the distortions are continuous, indeed differentiable
(in most traditional cartographic projections). Maps of neighbouring areas over-
lap near their edges and the totality of all maps in a world atlas covers the whole
world.
An arbitrary manifold is defined similarly, as an n-dimensional “world” rep-
resented by an “atlas” consisting of “maps”. These maps are a special kind of
parametrizations known as embeddings.
6.1. D EFINITION . Let U be an open subset of Rn . An embedding of U into R N
is a C map ψ : U RN satisfying the following conditions:
(i) ψ is one-to-one (i.e. if ψ t1 ψ t2 , then t1 2 t );
U, is continuous.
(ii) Dψ t is one-to-one for all t U;
(iii) the inverse of ψ, which is a map ψ 1 : ψ U
The image of the embedding is the set ψ U ! ψ t #" t U $ consisting
of all points of the form ψ t with t U. The inverse map ψ is called a chart
1
or coordinate map. You should think of ψ U as an n-dimensional “patch” in R N
“parameter” t must correspond distinct points ψ t in the patch ψ U . Thus the
parametrized by the map ψ. Condition (i) means that to distinct values of the
patch ψ U has no self-intersections. Condition (ii) means that for each t in U all
n columns of the Jacobi matrix Dψ t must be independent. This is imposed to
prevent the occurrence of cusps and other singularities in the image ψ U . Since
Dψ t has N rows, this condition also implies that N % n: the target space R N
cannot be an embedding. The column space of Dψ t is called the tangent space to
must have dimension greater than or equal to that of the source space U, or else ψ
the patch at the point x ψ t and is denoted by T ψ U ,
x
x T ψ U & Dψ t R .
n
67
68 6. MANIFOLDS
The tangent space at each point is an n-dimensional subspace of R N because Dψ t '(
)+* ' ( '(
has n independent columns. Condition (iii) can be restated as the requirement that
' ,(
if ti is any sequence of points in U such that lim i ψ ti exists and is equal to ψ t
for some t U, then lim i )+* -
ti t. This is intended to avoid situations where the
image ψ U doubles back on itself “at infinity”. (See Exercise 6.4 for an example.)
6.2. E XAMPLE . The picture below shows an embedding of an open rectangle
in the plane into three-space, the image of which is a portion of a torus. Try to
write a formula for such an embedding! (If we chose U too big, the image would
self-intersect and the map would not be an embedding.) For one particular value
of t the column vectors of the Jacobi matrix are also shown. As you can see, they
span the tangent plane at the image point.
./
Dψ t e1
ψ ./
ψ t
U e3
./
Dψ t e2
e2
t e1 . /
ψ U
e2
e1
6.3. E XAMPLE . Let U be an open subset of Rn and let f : U
map. The graph of f is the collection
0 Rm be a smooth
graph f -2143 f 'tt(65877 t , U 9 .
7
Since t is an n-vector and f ' t ( an m-vector, the7 graph is a subset of R with N -
N
n : m. We claim that the graph is the image of an embedding ψ : U 0 R . Define N
ψ ' t ( -;3
f ' t( 5
t
.
Then by definition graph f - ψ ' U ( . Furthermore ψ is an embedding. Indeed,
ψ ' t ( - ψ ' t ( implies t - t , so ψ is one-to-one. Also,
1 2 1 2
Dψ ' t ( -;3
D f ' t (<5
I
n
,
6.1. THE DEFINITION 69
=>
so Dψ t has n independent columns. Finally the inverse of ψ is given by
ψ ? @ f =tt>6ACB
1
t,
which is continuous. Hence ψ is an embedding.
A manifold is an object patched together out of the images of several embed-
dings. More precisely,
6.4. D EFINITION . An n-dimensional manifold1 (or n-manifold for short) in R N is
EE
a subset M of R N such that for all x M there exist D
FF
an open subset V R N containing x,
Ean open subset U Rn ,
and an embedding ψ : U G
R N satisfying ψ U V M.= >B H
N I
The codimension of M in R is N n. Choose t U such that ψ t
tangent space to M at x is the column space of Dψ t ,
D= > =>B
x. Then the
T M Dψ = t >J= R > .
x
B n
(Using the chain rule one can show that Tx M is independent of the choice of the
G
embedding ψ.) The elements of Tx M are tangent vectors to M at x. A collection of
embeddings ψi : Ui
= > R N with Ui open in Rn and such that M is the union of all
the sets ψi Ui is an atlas for M.
K
One-dimensional manifolds are called (smooth) curves, two-dimensional man-
ifolds (smooth) surfaces, and n-manifolds in Rn 1 (smooth) hypersurfaces. In these
cases the tangent spaces are usually called tangent lines, tangent planes, and tangent
hyperplanes, respectively.
The following picture illustrates the definition. Here M is a curve in the plane,
so we have N
B
2 and n
B
1. U is an open interval in R and V is an open disc
in R2 . The map ψ sends t to x and parametrizes the portion of the curve inside V.
B =>
= >
Since n 1, the Jacobi matrix Dψ t consists of a single column vector, which is
B
tangent to the curve at x ψ t . The tangent line Tx M is the line spanned by this
vector.
Tx M M
U
ψ
t x
F G
Sometimes a manifold has an atlas consisting of one single chart. In that event
BB = >
we can take V R N , and choose one open U Rn and an embedding ψ : U RN
such that M ψ U . However, usually one needs more than one chart to cover
a manifold. (For instance, one chart is not enough for the curve M in the picture
above.)
1In the literature this is usually called a submanifold of Euclidean space. It is possible to define
manifolds more abstractly, without reference to a surrounding vector space. However, it turns out
that practically all abstract manifolds can be embedded into a vector space of sufficiently high dimen-
sion. Hence the abstract notion of a manifold is not substantially more general than the notion of a
submanifold of a vector space.
70 6. MANIFOLDS
6.5. E XAMPLE . An open subset U of Rn can be regarded as a manifold of di-
M N O
mension n (hence of codimension 0). Indeed, U is the image of the map ψ : U
Rn given by ψ x x, the identity map. The tangent space to U at any point is R n
L
itself.
N P L M NQO
M 6.6. E XAMPLE . Let N n and define ψ : Rn R N by ψ x1 , x2 , . . . , xn
M N M N
x1 , x2 , . . . , xn , 0, 0, . . . , 0 . It is easy to check that ψ is an embedding. Hence the
image ψ Rn is an n-manifold in R N . (Note that ψ Rn is just a linear subspace
isomorphic to Rn ; e.g. if N O O
3 and n 2 it is just the xy-plane. We shall usually
identify R with its image in R N .) Combining this example with the previous
n
M N
one, we see that if U is any open subset of Rn , then ψ U is a manifold in R N of
R
codimension N n. Its tangent space at any point is Rn .
O L
Rm is a smooth map. As
S
L
6.7. E XAMPLE . Let M graph f , where f : U
S
shown in Example 6.3, M is the image of a single embedding ψ : U
n m
R n m , so
M M NN
M is an n-dimensional manifold in R , covered by a single chart. At a point
O O
x, f x in the graph the tangent space is spanned by the columns of Dψ. For
M M NN
M TUM NN
instance, if n m 1, M is one-dimensional and the tangent line to M at x, f x
TVM N
is spanned by the vector 1, f x . This is equivalent to the well-known fact that
the slope of the tangent line to the graph at x is f x .
graph f
Y ZVW X6[
1
f x WX f x
O M N\N O
1, M is a surface in R3 . The tangent plane to M at a point
M M N
For n 2 and m
x, y, f x, y is spanned by the columns of Dψ x, y , namely
]^ 1 _` ]a^ 1 _\` b
M x, y N and
∂f
∂x
M
0
x, y N
∂f
.
0 ∂y
The diagram below shows the graph of f M x, y NcO x d y R 3xy from two different
3 3
angles, together with a few points and tangent vectors. (To improve the scale the
6.1. THE DEFINITION 71
z-coordinate and the tangent vectors have been compressed by a factor of 2.)
e3
e3 e1 e2
e2
e1
e
R2 given by ψ t f gih f
et cos t, sin t . g
j f kg j h
6.8. E XAMPLE . Consider the path ψ : R
e t . Therefore
f glh f g h h
Let us check that ψ is an embedding. Observe first that ψ t
ψ t1 ψ t2 implies e t 1 t
e . The exponential function is one-to-one, so t 1 t2 .
2
This shows that ψ is one-to-one. The velocity vector is
t p sin t
mnf glh e o cos
ψ t t
cos t q sin t r
.
Therefore ψ m f t gsh 0 if and only if cos t h sin t h 0, which is impossible because
cos t q sin t h 1. So ψ m f t g+h t 0 for all t. Moreover we have t h ln e h ln j ψ f t gkj .
2 2 t
Hence the inverse of ψ is given by ψ u f x glh ln j x j for x v ψ f R g and so is continu-
ous. Therefore ψ is an embedding and ψ f R g is a 1-manifold. The image ψ f R g is a
1
spiral, which for t ewpyx converges to the origin. It winds infinitely many times
around the origin, although that is hard to see in the picture.
y
f g
Even though ψ R is a manifold, the set ψ R f g{z}| 0 ~ is not: it has a very nasty
singularity at the origin!
72 6. MANIFOLDS
6.9. E XAMPLE . An example of a manifold which cannot be covered by a single
chart is the unit sphere M S n 1 in Rn . Let U Rn 1 and let ψ : U Rn be the
map
ψ t
t
1
2 1
2t t 2 1 en
given in Exercise B.7. As we saw in that exercise, the image of ψ is the punctured
en , so if we let V be the open set Rn
sphere M
Also we saw that ψ has a two-sided inverse φ : ψ U &
en , then ψ U M V.
U, the stereographic pro-
jection from the north pole. Therefore ψ is one-to-one and its inverse is continuous
(indeed, differentiable). Moreover, φ ψ t t implies Dφ ψ t Dψ t v v for
all v in Rn 1 by the chain rule. Therefore, if v is in the nullspace of Dψ t ,
v Dφ ψ t Dψ t v Dφ ψ t 0 0.
Thus we see that ψ is an embedding. To cover all of M we need a second map, for
example the inverse of the stereographic projection from the south pole. This is
also an embedding and its image is M en M V, where V Rn en .
n
This finishes the proof that M is an n 1-manifold in R .
As this example shows, the definition of a manifold can be a little awkward
to work with in practice, even for a very simple manifold. Aside from the above
examples, in practice it can be rather hard to decide whether a given subset is a
manifold using the definition alone. Fortunately there exists a more manageable
criterion for a set to be a manifold.
6.2. The regular value theorem
Definition 6.4 is based on the notion of an embedding, which can be regarded
as an “explicit” way of describing a manifold. However, embeddings can be hard
find in practice. Instead, manifolds are often given “implicitly”, by a system of m
equations in N unknowns,
φ1 x1 , . . . , x N c1 ,
φ2 x1 , . . . , x N c2 ,
..
.
φm x1 , . . . , x N cm .
Here the φi ’s are smooth functions presumed to be defined on some common open
x1
\
subset U of R N . Writing in the usual way
\
φ1 x c
\ 1
x , φ x , c!
x2 φ2 x c
... ,
2
.. ..
φ x
. .
x N m c m
we can abbreviate this system to a single equation
φ x c.
For a fixed vector c R we denote the solution set by
m
φ c l x U φ x c
1
6.2. THE REGULAR VALUE THEOREM 73
and call it the level set or the fibre of φ at c. (The notation φ 1 c for the solution set
is standard, but a bit unfortunate because it suggests falsely that φ is invertible,
which it is usually not.) If φ is a linear map, the system of equations is inho-
mogeneous linear and by linear algebra the solution set is an affine subspace of
R N . The dimension of this affine subspace is N m, provided that φ has rank m
¢ ¡
(i.e. has m independent columns). We can generalize this idea to nonlinear equa-
¡
tions as follows. We say that c Rm is a regular value of φ if the Jacobi matrix
Dφ x : R N m
R has rank m for all x φ 1 c . A vector that is not a regular
value is called a singular value. (As an extreme, though slightly silly, special case,
if φ 1 c is empty, then c is automatically a regular value.)
The following result is the most useful criterion for a set to be a manifold.
(Don’t get carried away though, because it does not apply to every possible mani-
fold. In other words, it is a sufficient but not a necessary criterion.) The proof uses
the following important fact from linear algebra,
nullity A £ rank A ¤ l,
¥
valid for any k l-matrix A. Here the rank is the number of independent columns
of A (in other words the dimension of the column space A Rl ) and the nullity
is the number of independent solutions of the homogeneous equation Ax 0 (in ¤
other words the dimension of the nullspace ker A).
¢
¤
6.10. T HEOREM (regular value theorem). Let U be open in R N and let φ : U
m
R be a smooth map. Suppose that c is a regular value of φ and that M φ 1 c is
N
nonempty. Then M is a manifold in R of codimension m. Its tangent space at x is the
nullspace of Dφ x ,
Tx M ¤ ker Dφ x .
P ROOF. Let x ¡
M. Then Dφ x has rank m and so has m independent
columns. After relabelling the coordinates on R N we may assume the last m
¥
¤ ¥
columns are independent and therefore constitute an invertible m m-submatrix
A of Dφ x . Let us put n
N m. Identify R N with Rn Rm and correspond-
¤
ingly write an N-vector as a pair u, v with u a n-vector and v an m-vector. Also
write x
u0 , v0 . Now refer to Appendix B.4 and observe that the submatrix
A is nothing but the “partial” Jacobian Dvφ u0 , v0 . This matrix being invertible,
¢¡
by the implicit function theorem, Theorem B.4, there exist open neighbourhoods
v ¤ § ¥ ¡ ¤
f u V satisfying φ u, f u l¤
U of u0 in Rn and V of v0 in Rm such that for each u
c. The map f : U ¦
U there exists a unique
V is C . In other words
§ ¥
M U V graph f is the graph of a smooth map. We conclude from Exam-
ψ: U ¢
ple 6.7 that M
¤ ¡ ¥
U V is an n-manifold, namely the image of the embedding
RN given by ψ u u, f u . Since U V is open in R N and the above
¤
argument is valid for every x
¡ ¤
M, we see that M is an n-manifold. To compute
Tx M note that φ ψ u
¤
c, a constant, for all u U. Hence Dφ ψ u Dψ u
by the chain rule. Plugging in u u0 gives
0
Dφ x Dψ u l¤ 0.
The tangent space T M is by definition the column space of Dψ u , so every
0
tangent vector v to M at x is of the form v ¤ Dψ u a for some a ¡ R . Therefore
x 0
¤ ¤
n
Dφ x v Dφ x Dψ u0 a 0, i.e. Tx M¨ 0
ker Dφ x . The tangent space Tx M
74 6. MANIFOLDS
©ª © ª¬«© ª «
is n-dimensional (because the n columns of Dψ u0 are independent) and so is
ker Dφ x . ©ª
the nullspace of Dφ x (because nullity Dφ x N m n). Hence Tx M
QED
«
The case of one single equation (m «
1) is especially important. Then Dφ is
a single row vector and its transpose is the gradient of φ: Dφ T «
grad φ. It has
© ª«
rank 1 at x if and only if it is nonzero, i.e. at least one of the partials of φ does
not vanish at x. The solution set of a scalar equation φ x c is known as a level
hypersurface. Level hypersurfaces, especially level curves, occur frequently in all
kinds of applications. For example, isotherms in weathercharts and contour lines
in topographical maps are types of level curves.
6.11. C OROLLARY (level hypersurfaces). Let U be open in R N and let φ : U ®
« ¯ ©ª © ª±«°
R
be a smooth function. Suppose that M φ 1 c is nonempty and that grad φ x 0
for all x in M. Then M is a manifold in R N of codimension 1. Its tangent space at x is the
©ª
orthogonal complement of grad φ x ,
« © grad φ © xª\ª² .
Tx M
. Let U « R and φ © x, y ª³« xy. The level curves of φ are
2
hyperbolas in the plane and the gradient is grad φ © x ª´« © y, x ª . The diagram
6.12. E XAMPLE
T
below shows a few level curves as well as the gradient vector field, which as you
can see is perpendicular to the level curves.
y
©
¯ © ª ªi«
The gradient vanishes only at the origin, so φ 0 0 is the only singular value of
φ. By Corollary 6.11 this means that φ 1 c is a 1-manifold for c
¯ ©ª «°
0. The fibre
φ 1 0 is the union of the two coordinate axes, which has a self-intersection and
so is not a manifold. However, the set φ 1 0 ¯ © ª µ ¶
0 is a 1-manifold since the gra-
« © ª
dient is nonzero outside the origin. Think of this diagram as a topographical map
representing the surface z φ x, y shown below. The level curves of φ are the
contour lines of the surface, obtained by intersecting the surface with horizontal
planes at different heights. As explained in Appendix B.2, the gradient points in
the direction of steepest ascent. Where the contour lines self-intersect the surface
6.2. THE REGULAR VALUE THEOREM 75
has a “mountain pass” or saddle point.
e3
e2
e1
· ¸º¹ x3 y3 3xy » ¼ ¹ · ¸y¹ · ¼ ¼ ¸
6.13. E XAMPLE . A more interesting example of an equation in two variables
3 x 2 y, y2 x T , so grad φ
· ¸
is φ x, y c. Here grad φ x
¼ T
vanishes at the origin and at 1, 1 . The corresponding values of φ are 0, resp.
1, which are the singular values of φ.
y
½ ·¼ ¸
The level “curve” φ 1 1 is not a curve at all, but consists of the single point
· ¸
1, 1 T . Here φ has a minimum and the surface z ¹ · ¸
½ ·¸
φ x, y has a “valley”. The
level curve φ 1 0 has a self-intersection at the origin, which corresponds to a
saddle point on the surface. These features are also clearly visible in the surface
itself, which is shown in Example 6.7.
¹
6.14. E XAMPLE . Let U R N and φ x · ¸i¹¿¾ ¾
x 2 . Then grad φ x · ¸ ¹
½ ·¸
2x, so as in
Example 6.12 grad φ vanishes only at the origin 0, which is contained in φ 1 0 .
¹À ¼ ½ ·¸
So again any c 0 is a regular value of φ. Clearly, φ 1 c is empty for c 0. For Á
c  ½ ·¸ Ã
0, φ 1 c is an N 1-manifold, the sphere of radius c in R N . The tangent
76 6. MANIFOLDS
space to the sphere at x is the set of all vectors perpendicular to grad φ x Ä ÅºÆ 2x.
In other words,
Tx M x Æ Ç}ÆÉÈ Ê
y RN y x 0 . Ë Ì Æ Í
Î Ä ÅÆÉÈ Í
Ï
Finally, 0 is a singular value (the absolute minimum) of φ and φ 1 0 0 is not
Î ÄÏ Å
an N 1-manifold. (It happens to be a 0-manifold, though, just like the singular
fibre φ 1 1 in Example 6.13. So if c is a singular value, you cannot be certain
Î ÄÅ
that φ 1 c is not a manifold. However, even if a singular fibre happens to be a
manifold, it is often of the “wrong” dimension.)
Here is an example of a manifold given by two equations (m Æ 2).
6.15. E XAMPLE . Define φ : R4 Ð R by 2
x Ò
φ Ä x ÅlÆ;Ñ
2 x22
x x Ò
1
1 3 x2 x4
.
Ó
Then
Ä ÅlÆ Ñ 2xx
Dφ x 1 2x2
x4
0
x1
0
x2
.
Ó
ÆÔ ÄÅ Æ ÆÔ Æ
3
If x1 0 the first and third columns of Dφ x are independent, and if x 2 0 the
ÄÅ Ä ÅÕÆ
second and fourth columns are independent. On the other hand, if x 1 x2 0,
Dφ x has rank 1 and φ x 0. This shows that the origin 0 in R 2 is the only
Î ÄÅ Æ Î Ö×
singular value of φ. Therefore, by the regular value theorem, for every nonzero
vector c the set φ 1 c is a two-manifold in R4 . For instance, M φ 1 10 is a
two-manifold. Note that M contains the point x Æ¿Ä
1, 0, 0, 0 T. Let us find a basis
of the tangent space Tx M. Again by the regular value theorem, this tangent space
Å
is equal to the nullspace of
Dφ x Ä ÅÆ;Ñ 20 0
0
0
1
0
0
,
Ó
which is equal the set of all vectors y satisfying y 1
therefore given by the standard basis vectors e2 and e4 .
y3 Æ Æ 0. A basis of Tx M is
We now come to a more sophisticated example of a manifold determined by a
large system of equations.
Ø
6.16. E XAMPLE . Recall that an n n-matrix A is orthogonal if A T A
means that the columns (and also the rows) of A are perpendicular to one another
I. This Æ
and have length 1. (In other words, they form an orthonormal basis of R n —note
the regrettable inconsistency in the terminology.) The collection of orthogonal ma-
Ä Å
trices form a group under matrix multiplication, which is usually called the orthog-
Ä Å Ä Å Æ
onal group and denoted by O n . Let us prove using the regular value theorem that
Æ Ù
O n is a manifold. First observe that that A T A T A T A, so A T A is a symmetric
Ø
ÆÚÈ Ê Ë Æ
n n
Í
matrix. In other words, if V R is the vector space of all n n-matrices and
W C V C C T the linear subspace of all symmetric matrices, then
Ä Å Æ A A φ A T
defines a map φ : V Ð W. Clearly O Ä n Å±Æ φ Î Ä I Å , so to prove that O Ä n Å is a
1
n
manifold it suffices to show that I is a regular value of φ. The derivative of φ can
EXERCISES 77
be computed by using the formula derived in Exercise B.3:
Û Ü Ý limÞ h1 Û φ Û A ß hBÜáà φ Û A ÜÜ
Dφ A B
h 0
Ý limÞ h1 Û A A ß hA B ß hB A ß h B B à A A Ü
T T T 2 T T
Ý BA ß AB .
h 0
T T
We need to show that for A â O Û n Ü the linear map Dφ Û A Ü : V ã W has rank
equal to the dimension of W. By linear algebra this amounts to showing that the
equation
BA T AB T C ß (6.1)Ý
Ý Û ß Ü Ý
is solvable for B, given any orthogonal A and any symmetric C. Here is a way of
1 1
guessing a solution: observe that C 2 C C T and first try to solve BA T 2 C.
Left multiplying both sides by A and using A A T I gives B 1
2 CA. It is now Ý Ý
easy to check that B 1
2 CA is Ý
a solution of equation (6.1).
Exercises
æ
6.1. This is a continuation of Exercise 1.1. Define ψ : R
T
R 2 by ψ t
cos t . Show that ψ is one-to-one. Determine all t for which ψ t
t sin t, 1
0. Prove that ψ R is
ä éUå æáç å æ{çCå è å æ è
not a manifold at these points.
è
6.2. Let a ê³æ å æ
0, 1 be a constant. Prove that the map ψ : R
T
R 2 given by ψ t
a sin t, 1 a cos t is an embedding. (This becomes easier if you first show that t
ä å æ{è çëå è t
a sin t
is an increasing function of t.) Graph the curve defined by ψ.
6.3. Prove that the map ψ : R
ç å æ ä
R2 given by ψ t 1 et
2 e t , et e t T is an embed- å ìæ ç å í î è î æ
å æ
ding. Conclude that M ψ R is a 1-manifold. Graph the curve M. Compute the tangent
å æñçòå óôå í
line to M at 1, 0 and try to find an equation for M.
ó õ å í æïæ
6.4. Let I be the open interval 1, åïè ð³æ
and let ψ : I R 2 be the map ψ t 3at 1 ä
öé å æ±ç÷
t3 , 3at2
ê åæ åæ
1 t 3 T , where a is a nonzero constant. Show that ψ is one-to-one and that
ψ t 0 for all t I. Is ψ an embedding and is ψ I a manifold? (Observe that ψ I
is a portion of the curve studied in Exercise 1.2.)
6.5. Define ψ : R ä R2 by
å æñçCøúù è å æ å æ å æïå û æüû
f t ,f t
T
if t ý
þ
0,
ù
ψ t T
f t ,f t if t 0,
î å æÿä
where f is the function given in Exercise B.6. Show that ψ is smooth, one-to-one and that
its inverse ψ 1 : ψ R R is continuous. Sketch the image of ψ. Is ψ R a manifold? å æ
6.6. Define a map ψ : R2 ä R4 by
t31
ç
t1 2
t1 t2
ψ .
t2 t1 t22
t32
åæ
(i) Show that ψ is one-to-one.
(ii) Show that Dψ t is one-to-one for all t 0.
è ç÷ ä
å æ
(iii) Let U be the punctured plane R2 0 . Show that ψ : U
Conclude that ψ U is a two-manifold in R4 .
R4 is an embedding.
78 6. MANIFOLDS
(iv) Find a basis of the tangent plane to ψ U at the point ψ 1, 1 .
6.7. Let φ : Rn 0
R be a homogeneous function of degree p as defined in Ex-
ercise B.5. Assume that φ is smooth and that p
0. Show that 0 is the only possible
singular value of φ. (Use the result of Exercise B.5.) Conclude that, if nonempty, φ 1 c is
an n 1-manifold for c 0.
6.8. Let φ x
a 1 x21 a2 x22
an x2n , where the a i are nonzero constants. Deter-
mine the regular and singular values of φ. For n 3 sketch the level surface φ 1 c for a
regular value c. (You have to distinguish between a few different cases.)
6.9. Show that the trajectories of the Lotka-Volterra system of Exercise 1.10 are one-
dimensional manifolds.
6.10. Compute the dimension of the orthogonal group O n and show that its tangent
space at the identity matrix I is the set of all antisymmetric n n-matrices.
6.11. Let V be the vector space of n
det A.
n-matrices and define φ : V R by φ A
(i) Show that
Dφ A B ∑ i 1
n
det a1 , a2 . . . , ai
1 , bi , a i 1 , . . . , an ,
where a 1 , a2 , . . . , an and b1 , b2 , . . . , bn denote the column vectors of A, resp. B.
(Apply the formula of Exercise B.3 for the derivative and use the multilinearity
of the determinant.)
(ii) The special linear group is the subset of V defined by
SL n A !
V det A 1 .
Show that SL n is a manifold. What is its dimension?
(iii) Show that for A I, the identity matrix, we have Dφ A B ∑ni 1 bi,i
tr B,
the trace of B. Conclude that the tangent space to SL n at I is the set of traceless
matrices, i.e. matrices A satisfying tr A 0.
"
6.12. (i) Let W be punctured 4-space R4 0 and define φ : W R by
φ x x1 x4 x2 x3 .
Show that 0 is a regular value of φ.
(ii) Let A be a real 2 2-matrix. Show that rank A 1 if and only if det A 0 and
A 0.
(iii) Let M be the set of 2 2-matrices of rank 1. Show that M is a three-dimensional
manifold.
(iv) Compute TA M, where A 11 . $# %
00
6.13. Define φ : R4 R2 by
φ x '& (
x1 x2 x3 x4
x1 x2 x3 x4
.
(i) Show that Dφ x has rank 2 unless x is of the form t 2 , t 2 , t, t 3 for some
t 0. (Compute all 2 2-subdeterminants of Dφ and set them equal to 0.)
(ii) Show that M φ 1 0 is a 2-manifold (where 0 is the origin in R2 ).
(iii) Find a basis of the tangent space Tx M for all x M with x 3 0. (The answer
depends on x.)
EXERCISES 79
* +,
6.14. Let U be an open subset of Rn and let φ : U ) ) .
Rm be a smooth map. Let M be
- -
the manifold φ 1 c , where c is a regular value of φ. Let f : U R be a smooth function.
+, /
A point x
tangent vectors v Tx M. Prove that x -
M is called a critical point for the restricted function f M if D f x v
. 0 for all
M is critical for f M if and only if there exist
+ ,0/ + ,21
numbers λ 1 , λ2 , . . . , λm such that
grad f x λ1 grad φ1 x λ2 grad φ2 x + ,31544461 λm grad φm x . +,
(Use the characterization of Tx M given by the regular value theorem.)
6.15. Find the critical points of the function f x, y, z + ,7/98 1 1
x 2y 3z over the circle C
given by
x2 1 1 /y2 z2
1 /
1,
x z 0.
Where are the maxima and minima of f C? .
) + ,0/ 4
6.16 (eigenvectors via calculus). Let A / ; - =. < <>: / ?
A T be a symmetric n n-matrix and define
+
- ,
f : Rn R by f x x Ax. Let M be the unit sphere x Rn x 1 .
(i) Calculate grad f x .
.
(ii) Show that x M is a critical point of f M if and only if x is an eigenvector for A
of length 1.
+,
(iii) Given an eigenvector x of length 1, show that f x is the corresponding eigen-
value of x.
CHAPTER 7
Differential forms on manifolds
7.1. First definition
There are several different ways to define differential forms on manifolds. In
this section we present a practical, workaday definition. A more theoretical ap-
proach is taken in Section 7.2.
a 0-form or smooth function on M. A function f : M @
Let M be an n-manifold in R N and let us first consider what we might mean by
AB R is simply an assignment
of a unique number f x to each point x in M. For instance, M could be the surface
of the earth and f could represent temperature at a given time, or height above sea
level. But how would we define such a function to be differentiable? The difficulty
C D
here is that if x is in M and e j is one of the standard basis vectors, the straight
E
F A A C BHG A BBD
line x he j may not be contained in M, so we cannot form the limit ∂ f ∂x j
limh 0 f x he j f x h.
@ A B
Here is one way out of this difficulty. Because M is a manifold there exist open
E$I A B
sets Ui in Rn and embeddings ψi : Ui R N such that the images ψ i Ui cover M:
M
For each i we define a function f i : Ui @R by f i t A B E A A BB
f ψi t , i.e. f i E J
i ψ i Ui . (Here i ranges over some unspecified, possibly infinite, index set.)
ψi f . We
call f i the local representative of f relative to the embedding ψ i . (For instance, if M
is the earth’s surface, f is temperature, and ψ i is a map of New York State, then f i
represents a temperature chart of NY.) Since f i is defined on the open subset Ui of
Rn , it makes sense to ask whether its partial derivatives exist. We say that f is C k
K
if each of the local representatives f i is C k . Now suppose that x is in the overlap
K
ABE A B A B E A A BB E A A BB
of two charts. Then we have two indices i and j and vectors t Ui and u U j
E
such that x ψi t ψ j u . Then we must have f x f ψi t f ψ j u , so
ABE A B
fi t ABE A B
f j u . Also ψi t ψ j u implies t ψi 1
E L M A B
ψ j u and therefore f j u A BE
LN M A B O K
f i ψi 1 ψ j u . This identity must hold for all u U j such that ψ j u A BPK A B
ψ i Ui ,
L A A BB 1
i.e. for all u in ψ j ψi Ui . We can abbreviate this by saying that
fj E A ψL M ψ B J f
i
1
j i
L A A BQB
on ψ j 1 ψi Ui . This is a consistency condition on the functions f i imposed by the
L M
fact that they are pullbacks of a single function f defined everywhere on M. The
map ψi 1 ψ j is often called a change of coordinates and the consistency condition
is also known as the transformation law for the local representatives f i . (Pursuing
the weather chart analogy, it expresses nothing but the obvious fact that where the
maps of New York and Pennsylvania overlap, the corresponding two temperature
A B E A L A BB K A B
charts must show the same temperatures.) Conversely, the collection of all local
representatives f i determines f , because we have f x f i ψi 1 x if x ψi Ui .
81
82 7. DIFFERENTIAL FORMS ON MANIFOLDS
(That is to say, if we have a complete set of weather charts for the whole world, we
know the temperature everywhere.)
Following this cue we formulate the following definition.
7.1. D EFINITION . A differential form of degree k, or simply a k-form, α on M is a
collection of k-forms α i on Ui satisfying the transformation law
αj RTS ψ U V ψ W X α
i
1
j i (7.1)
U S S WW
on ψ j 1 ψi Ui . We call αi the local representative of α relative to the embedding
Y S W
ψi and denote it by αi
k M .
R X
ψi α . The collection of all k-forms on M is denoted by
This definition is rather indirect, but it works really well if a specific atlas for
Z
the manifold M is known. Definition 7.1 is particularly tractible if M is the image
of a single embedding ψ : U R N . In that case the compatibility relation (7.1) is
vacuous and a k-form α on M is determined by one single representative, a k-form
X
ψ α on U.
Sometimes it is useful to write the transformation law (7.1) in components. We
can do this by appealing to Theorem 3.12. If
αi R ∑ f I dt I
I
and αj R ∑ g J dt J
J
are two local representatives for α , then
gJ R ∑ S ψU V ψ W X f
I
i
1
j I det D ψi S U V ψW 1
j I,J .
U S S WW
on ψ j 1 ψi Ui .
Just like forms on Rn , forms on a manifold can be added, multiplied, differen-
tiated and integrated. For example, suppose α is a k-form and β an l-form on M.
ding ψi : Ui Z
Suppose αi , resp. βi , is the local representative of α , resp. β, relative to an embed-
M. Then we define the product γ αβ by setting γ i αiβi . To see
that this definition makes sense, we check that the forms γ i satisfy the transforma-
R R
tion law (7.1):
γj R R[S ψU V ψ W X α S ψ U V ψ W X β R\S ψ U V ψ W X S α β W]RTS ψ U V ψ W X γ .
α jβ j i
1
j i i
1
j i i
1
j i i i
1
j i
Here we have used the multiplicative property of pullbacks, Proposition 3.10(ii).
Similarly, the exterior derivative of α is defined by setting S dα W R dα . As before,
let us check that the forms S dα W satisfy the transformation law (7.1):
i i
i
S dα W R dα R d S ψ U V ψ W X α RTS ψ U V ψ W X dα RTS ψ U V ψ W X S dα W ,
j j i
1
j i i
1
j i i
1
j i
where we used Theorem 3.11.
7.2. Second definition
This section presents some of the algebraic underpinnings of the theory of
differential forms. This branch of algebra, now called exterior or alternating algebra
was invented by Graßmann in the mid-nineteenth century and is a prerequisite
for much of the more advanced literature on the subject.
7.2. SECOND DEFINITION 83
Covectors. Before giving a rigorous definition of differential forms on mani-
folds we need to be more precise about the definition of a differential form on R n .
Recall that Rn is the collection of all column vectors
_```a x bQccc
1
x^
x
. d
2
.. .
xn
Let U be an open subset of Rn . The definition of a 0-form on U requires no further
clarification: it is simply a function on U. Formally, a 1-form α on U can be defined
as a row vector
α ^Te
f1, f2, . . . , fn f
whose entries are functions on U. The form is called constant if the entries f 1 , . . . ,
f n are constant. The set of constant row vectors is denoted by R n and is called
the dual of Rn . Constant 1-forms are also known as covariant vectors or covectors
e f g
and arbitrary 1-forms as covariant vector fields or covector fields. By definition dx i is
the constant 1-form
dxi eiT ^ ^Te
0, . . . , 0, 1, 0, . . . , 0 , f
the transpose of ei , the i-th standard basis vector of Rn . Every 1-form can thus be
written as
α ^\e f , f , . . . , f f ^ ∑h f dx .
1 2 n
n
i 1
i i
Using this formalism we can write for any smooth function g on U
dg ^ ∑h ∂x∂g dx ^i ∂x∂g , ∂x∂g , . . . , ∂x∂g j ,
n
i 1 i
i
1 2 n
so dg is simply the Jacobi matrix Dg of g! (This is the reason that many authors
use the notation dg for the Jacobi matrix.)
We would like to extend the notions of covectors and 1-forms to vector spaces
k
other than Rn . To see how, let us start by observing that a row vector y is nothing
_```a x b ccc
but a 1 n-matrix. We can multiply it by a column vector x to obtain a number,
yx ^Te y , y , . . . , y f ^ ∑h y x .
x n
. d
2
1 2 .. n i i
i 1
x n
Obviously we have y e c x l c x f ^ c yx l c yx . Thus a row vector can be
1 1 2 2 1 1 2 2
n
vectors (scalars) in R ^ R.
viewed as a linear map which sends column vectors in R to one-dimensional
1
numbers (for example R or a subspace of R ), then V g , the dual of V, is the set
This motivates the following definition. If V is any vector space over the real
n n
of linear maps from V to R. Elements of V g are called dual vectors or covectors or
linear functionals. The dual is a vector space in its own right: if µ and µ are in V g
we define µ l µ and cµ by setting e µ l µ f e v f ^ µ e v f l µ e v f for all v m V
1 2
and e cµ f e v f ^ cµ e v f .
1 2 1 1 2 1 2
1 1
84 7. DIFFERENTIAL FORMS ON MANIFOLDS
7.2. E XAMPLE . Let V oqp r s p r n
C 0 a, b , R , the collection of all continuous real-
valued functions on a closed and bounded interval a, b . A linear combination
u os o stn
o v s n o sv o s
of continuous functions is continuous, so V is a vector space. Define µ f
b
a f x dx. Then µ c 1 f 1 c2 f 2 c1µ f 1 c2µ f 2 , so µ is a linear functional
. Let V n R and fix v w V. Define µ o x sxn v y x, where “ y ” is the
on V.
7.3. E XAMPLE n
n
standard inner product on R . Then µ is a linear functional on V.
Now suppose that V is a vector space of finite dimension n and choose a basis
v1 , v2 , . . . , vn of V. Then every vector v V can be written in a unique way as
a linear combination ∑ j c j v j . Define a covector λ i V by λi v ci . In other
w w z o s{n
words, λi is determined by the rule
o s]n n}| 10 if i n j,
λi v j δi, j
if i n~ j.
We call λi the i-th coordinate function.
zn n
7.4. L EMMA . The coordinate functions λ 1 , λ2 , . . . , λn constitute a basis of V . Hence
dim V n dim V.
z
w z
P ROOF. Let λ V . We need to write λ as a linear combination λ ∑ni 1 ci λi .
Assuming for the moment that this is possible, we can apply both sides to the
n
vector v j to obtain
o xs n ∑ c λ o v s]n ∑ c δ n c . n n
λ vj
So c n λ o v s is the only possible choice for the coefficient c . To show that this
i 1
(7.2)
i i j
i 1
i i, j j
choice of coefficients works, let us define λ n ∑ λ o v s λ . Then by equation
j j j
(7.2), λ o v s]n λ o v s for all j, so λ n λ, i.e. λ n ∑ λ o v s λ . We have proved that
n
i 1 i i
n
every λ w V z can be written uniquely as a linear combination of the λ .
j j i 1 i i
QED
The basis λ , λ , . . . , λ of V z is said to be dual to the basis v , v , . . . , v
i
1 2 n 1 2 n
. Consider R with standard basis e , . . . , e . Then dx o e sn
of V.
e e n δ , so the dual basis of o R sz is dx , dx , . . . , dx .
7.5. E XAMPLE n
1 n i j
T n
i j i, j 1 2 n
L : V W be a linear map between abstract vector spaces V and W. To write
Dual bases come in handy when writing the matrix of a linear map. Let
w , w , . . . , w of W. Then for each j n 1, 2, . . . , n the vector Lv can be expanded
the matrix of L we need to start by picking a basis v , v , . . . , v of V and a basis 1 2 n
uniquely in terms of the w’s: Lv n ∑ l w . The m n numbers l make up
1 2 m j
m
j i 1 i, j i i, j
the matrix of L relative to the two bases of V and W.
. Let λ , λ , . . . , λ w V z be the dual basis of v , v , . . . , v and µ ,
µ , . . . , µ w W z the dual basis of w , w , . . . , w . Then the i, j-th matrix element of a
7.6. L EMMA 1 2 n 1 2 n 1
linear map L : V W is equal to l n µ o Lv s .
2 n 1 2 n
. We have Lv n ∑ l w , so
i, j i j
m
P ROOF j k 1 k, j k
µ o Lv s]n ∑ l µ o w sn ∑ l δ n l ,
m m
o i
j k, j i k k, j ik i, j
that is l n µ Lv s .
k 1 k 1
i, j i j QED
7.2. SECOND DEFINITION 85
Multilinear algebra. Let V be a vector space and let V k denote the Carte-
sian product V
V (k times). Thus an element of V k is an ordered k-tuple
v1 , v2 , . . . , vk of vectors in V. A k-multilinear function on V is a function λ : V k
R which is linear in each vector, i.e.
c v , . . . , v
λ v1 , v2 , . . . , cvi
cλ v , v , . . . , v c λ v , v , . . . , v , . . . , v
i k
for all scalars c, c and all vectors
v , v , . . . , v , v , . . . , v .
1 2 k 1 2 i k
x y, the inner product of x and
1 2 i i k
7.7. E . Let V R and let λ x, y n
vw vw
XAMPLE
y. Then λ is bilinear (i.e. 2-multilinear).
7.8. E . Let V R ,k 2. The function λ v, w
4
v w v w is bilinear on R .
XAMPLE 1 2 2 1
4
3 4 4 3
7.9. E . Let V
XAMPLE R ,k n. The determinant det v v , , . . . , v is an
n
1 2 n
n-multilinear function on Rn .
A k-multilinear function is alternating or antisymmetric if it has the alternating
property,
λ v , . . . , v , . . . , v , . . . , v .
λ v1 , . . . , v j , . . . , v i , . . . , v k
More generally, if λ is alternating, then for any permutation σ S we have
1 i j k
λ v ,...,v
sign σ λ v , . . . , v .
σ 1 σ k 1 k
k
of Example 7.7 is bilinear, but it is not al-
ternating. Indeed it is symmetric: y x x y. The bilinear function of Example 7.8
7.10. E . The inner product
XAMPLE
is alternating, and so is the determinant function of Example 7.9.
Here is a useful trick to generate alternating k-multilinear functions starting
from k covectors λ1 , λ2 , . . . , λk V . The (wedge) product is the function
λ : V R λ1 λ2 k
k
defined by
λ λ λ v , v , . . . , v
1 2 k
det λ v .
1 2 k i j
(The determinant on the right is a k k-determinant.) It follows from the multi-
1 i, j k
linearity and the alternating property of the determinant that λ λ λ is an al-
ternating k-multilinear function. The wedge product is often denoted by λ λ
1 2 k
λ to distinguish it from other products, such as the tensor product defined
k
1 2
in Exercise 7.6.
1 the alternating property is vacuous, so an alternating 1-multilinear
The collection of all alternating k-multilinear functions is denoted by A V. k
function is nothing but a linear function. Thus A V V .
For k
k 0 a k-multilinear function is defined to be a single number. Thus
1
For
A0 V R.
For any k, k-multilinear functions can be added and scalar-multiplied just like
ordinary linear functions, so the set A k V forms a vector space.
There is a nice way to construct a basis of the vector space A k V starting from
a basis v1 , . . . , vn of V. The idea is to take wedge products of dual basis vectors.
86 7. DIFFERENTIAL FORMS ON MANIFOLDS
an increasing multi-index, i.e. 1 i 1 i2 ¦ § §©¨¨¨§ ¦ ¡
Let λ1 , . . . , λn be the corresponding dual basis of V . Let I
ik n. Write
¢¤£ i , i , . . . , i ¥ be 1 2 k
λ ¢ λ λ ¨¨¨ λ ª A V, k
v ¢T£ v , v , . . . , v ¥ ª V .
I i1 i2 ik
k
I i1 i2 ik
. Let V ¢ R with standard basis e , e , e . The dual basis
3
of £ R ¥ ¡ is dx , dx , dx . Let k ¢ 2 and I ¢[£ 1, 2 ¥ , J ¢T£ 2, 3 ¥ . Then
7.11. E XAMPLE 1 2 3
3
««« dxdx ££ ee ¥¥ dxdx ££ ee ¥¥ «««« ¢¬«««« 10 01 «««« ¢ 1,
1 2 3
dx £ e ¥ ¢}« I I
1 1 1 2
2 1 2 2
dx £ e ¥ ¢ «
««« dxdx ££ ee ¥¥ dxdx ££ ee ¥¥ «««« ¢ «««« 01 00 «««« ¢ 0,
I J
1
2
2
2
1
2
3
3
dx £ e ¥ ¢}«
««« dxdx ££ ee ¥¥ dxdx ££ ee ¥¥ «««« ¢¬«««« 00 10 «««« ¢ 0,
J I
2
3
1
1
2
3
2
2
dx £ e ¥ ¢}«
««« dxdx ££ ee ¥¥ dxdx ££ ee ¥¥ «««« ¢¬«««« 10 01 «««« ¢ 1.
J J
2
3
2
2
2
3
3
3
This example generalizes as follows.
7.12. L EMMA . Let I and J be increasing multi-indices of degree k. Then
£ ¥ ¢ δ ¢} 10 ifif II ¢¢® J,J.
λI vJ I,J
P . Let I ¢T£ i , . . . , i ¥ and J ¢T£ j , . . . , j ¥ . Then
ROOF 1 k 1 k
««« δ ... . . . δ ... ««« i1 , j1 i1 , jk
λ £ v ¥ ¢ det ¯ λ £ v ¥° ± ± ¢ « δ
«« . . . δ ««« ¢ 1 if I ¢ J.
I J
««« ... .. «
il , j1 il , jk
««« δ . . . δ . «««««
ir js 1 r,s k
ik , j1 ik , jk
If I ¢ ® J, then i ¢ ® j for some l. Choose l as small as possible, so that i ¢ j for
m § l. There are two cases: i § j and i ² j . If i § j , then i § j § j ³ §
l l m m
¨¨¨7§ j because J is increasing, so all entries δ in the determinant with¢ m0 ´ forl
are 0. For m § l we have j ¢ i § i because I is increasing, so δ
k
l l l l
il , jm
l l l l l 1
m § l. In other words the l-th row in the determinant is 0 and hence λ £ v ¥ ¢ 0.
m m l il , jm
If i ² j we find that the l-th column in the determinant is 0 and therefore again
I J
λ £ v ¥ ¢ 0.
l l
I J QED
We need one further technical result before showing that the functions λ I are
a basis of Ak V.
7.13. L EMMA . Let λ
of degree k. Then λ 0. ¢ ª A k V. Suppose λ v I £ ¥¢ 0 for all increasing multi-indices I
P ROOF. The assumption implies
£
λ vi 1 , . . . , v i k ¥¢ 0 (7.3)
7.2. SECOND DEFINITION 87
µ
µ ¶¸· ¶
for all multi-indices i1 , . . . , ik , because of the alternating property. We need to
show that λ w1 , . . . , wk 0 for arbitrary vectors w1 , . . . , wk . We can expand the
wi using the basis:
w1 · a11 v1 ¹»ººº¹ a1k vk ,
..
.
wk · ak1 v1 ¹¼ººº½¹ akk vk .
Therefore by multilinearity
µ
λ w1 , . . . , w k ¶x· ∑¾ ººº ∑¾ a ººº a λ µ v , . . . , v ¶ .
i1 1
k
ik 1
k
1i 1 ki k i1 ik
Each term in the right-hand side is 0 by equation (7.3). QED
¿ Á ¿ À
À
7.14. T HEOREM . Let V be an n-dimensional vector space with basis v 1 , . . . , vn .
Let λ1 , . . . , λn be the corresponding dual basis of V . Then the alternating k-multilinear
functions λ I · ººº
λik , where I ranges over the set of all increasing multi-indices of
·¬Â Ã
λi1
n
degree k, form a basis of A k V. Hence dim Ak V k .
P . The proof is closely analogous to that of Lemma 7.4. Let λ Ä A V. k
We need to write λ as a linear combination λ · ∑ c λ . Assuming for the moment
ROOF
I I I
that this is possible, we can apply both sides to the k-tuple of vectors v . Using J
Lemma 7.12 we obtain
λ vJµ ¶· ∑ cI λI µ v ¶]· J ∑ c I δ I,J · cJ.
· µ ¶
I I
Å ·µ µ µ ¶
So c J λ v J is the only possible choice for the coefficient c J . To show that this
µ ¶ Æ Å
choice of coefficients works, let us define λ
> µ
¶
∑ I λ v I λ I . Then for all increasing
· ¶>Æ · ¶{·µ ¶
Æ ÅÄ Á Æ Å ·
multi-indices I we have λ v I λ vI λ vI λ vI 0. Applying Lemma
7.13 to λ λ we find λ λ 0. In other words, λ ∑ I λ v I λ I . We have proved
that every λ V can be written uniquely as a linear combination of the λ i . QED
. Let V · R with standard basis ¿ e , . . . , e À . The dual basis
n
of µ R ¶ Á is ¿ dx , . . . , dx À . Therefore A V has a basis consisting of all k-multilinear
7.15. E XAMPLE 1 n
n k
1 n
functions of the form
dx · dx dx ººº dx ,
with 1 Ç i È\ºººHÈ i Ç n. Hence a general alternating k-multilinear function λ
I i1 i2 ik
1 k
n
on R looks like
λ · ∑ a dx , I I
with a constant. By Lemma 7.12, λ µ e ¶É· ∑ a dx µ e ¶Ê· ∑ a δ · a , so a is
I
equal to λ µ e ¶ .
I J I I I J I I I,J J I
I
alternating k-multilinear function α for each x Ä U; hence it looks like α ·
An arbitary k-form α on a region U in R is now defined as a choice of an n
∑ f µ x ¶ dx , where the coefficients f are functions on U. We shall abbreviate this
x x
I I I I
to
α · ∑ f I dx I ,
I
88 7. DIFFERENTIAL FORMS ON MANIFOLDS
Ë Ì Í
and we shall always assume the coefficients f I to be smooth functions. By Example
fI x Ì Í]Ë Ì Í
7.15 we can express the coefficients as f I
αx e I for all x).
α e I (which is to be interpreted as
Pullbacks re-examined. In the light of this new definition we can give a fresh
Î
interpretation of a pullback. This will be useful in our study of forms on manifolds.
Let U and V be open subsets of Rn , resp. Rm , and φ : U
ÏÐ Ì Í Ñ ÏÐ Ì Í
V a smooth map. For a
k-form α k V define the pullback φ α k U by
Ì φÑ α Í Ì v , v , . . . , v Í]Ë α Ò Ó Ì Dφ Ì xÍ v , Dφ Ì xÍ v , . . . , Dφ Ì xÍ v Í .
x 1 2 k φ x 1 2 k
Let us check that this formula agrees with the old definition. Suppose α Ë ∑ f dy
and φ Ñ α Ë ∑ g dx . What is the relationship between g and f ? We use g Ë
I I I
φ Ñ α Ì e Í , our new definition of pullback and the definition of the wedge product
J J J J I J
J
to get
g Ì x Í]Ë Ì φ Ñ α Í Ì e ÍxË α Ò Ó Ì Dφ Ì x Í e , Dφ Ì x Í e , . . . , Dφ Ì x Í e Í
Ë ∑ f Ì φ Ì xÍÍ dy Ì Dφ Ì xÍ e , Dφ Ì xÍ e , . . . , Dφ Ì xÍ e Í
J x J φ x j1 j2 jk
I I j1 j2 jk
Ë ∑ φÑ f Ì xÍ det Ô dy Ì Dφ Ì xÍ e Í Õ Ö Ö .
I
I ir js 1 r,s k
By Lemma 7.6 the number dy Ì Dφ Ì x Í e Í is the i j -matrix entry of the Jacobi ma-
I
trix Dφ Ì x Í (with respect to the standard basis e , e , . . . , e of R and the standard
ir js r s
basis e , e , . . . , e of R ). In other words, g Ì x Í×Ë ∑ φ Ñ f Ì x Í det Dφ Ì x Í . This
n
1 2 n
m
1 2 m J I I I,J
formula is identical to the one in Theorem 3.12 and therefore our new definition
agrees with the old!
Forms on manifolds. Let M be an n-dimensional manifold in R N . For each
point x in M the tangent space Tx M is an n-dimensional linear subspace of R N .
A differential form of degree k or a k-form α on M is a choice of an alternating k-
multilinear map αx on the vector space Tx M, one for each x M. This alternating Ï
Ï Î
map αx is required to depend smoothly on x in the following sense. According to
RN
Ì ÍØË Ù
the definition of a manifold, for each x M there exists an embedding ψ : U
Ì ÍÌ Í Ï Ì ÍÚË
such that ψ U N
M V for some open set V in R containing x. The tangent
space at x is then Tx M Ë
Dψ t Rn , where t U is chosen such that ψ t
The pullback of α under the local parametrization ψ is defined by
x.
Ì ψÑ α Í Ì v , v , . . . , v ÍË α Ò Ó Ì Dψ Ì tÍ v , Dψ Ì tÍ v , . . . , Dψ Ì tÍ v Í .
t 1 2 1 2
Then ψ Ñ α is a k-form on U, an open subset of R , so ψ Ñ α Ë ∑ f dt for certain
k ψ t k
n
I I I
form ψ Ñ α Ë ∑ f dt is the local representative of α relative to the embedding ψ, as
functions f defined on U. We will require the functions f to be smooth. (The
I I
I I I
introduced in Section 7.1.) To recapitulate:
7.16. D EFINITION. A k-form α on M is a choice, for each x Ï M, of an alternat-
ing k-multilinear map α on T M, which depends smoothly on x.
x x
The book [BT82] describes a k-form as an “animal” that inhabits a “world” M,
eats ordered k-tuples of tangent vectors, and spits out numbers.
7.17. E XAMPLE . Let M be a one-dimensional manifold in R N . Let us choose an
orientation (“direction”) on M. A tangent vector to M is positive if it points in the
EXERCISES 89
same direction as the orientation and negative if it points in the opposite direction.
Define a 1-form α on M as follows. For x M and a tangent vector v Tx M put Û Û
αx v Ü ÝÞ¬ß á à vv à if v is positive,
à à if v is negative.
This form is the element of arc length of M. We shall see in Chapter 8 how to gener-
alize it to higher-dimensional manifolds and in Chapter 9 how to use it to calculate
arc lengths and volumes.
â
ã
We can calculate the local representative ψ α of a k-form α for any embedding
ψ: U
ã ã
R N parametrizing a portion of M. Suppose we had two different such
Þ
Ý Þ Ü Ý Þ â Þ
embeddings ψi : Ui M and ψ j : U j M, such that x is contained in both Wi
âÜ
ψi Ui and W j ψ j U j . How do the local expressions α i ψi α and α j
ψ j α for α compare? To answer this question, consider the coordinate change map
ψ ä å ψ , which maps ψ ä Ü W æ W Ý to ψ ä Ü W æ W Ý . From α Þ ψ â α and α Þ
1 1 1
ψ â α we recover the transformation law (7.1)
j i i i j j i j i i j
α Þ Ü ψä å ψ Ý â α . j i
1
j i
This shows that Definitions 7.1 and 7.16 of differential forms on a manifold are
equivalent.
é êìë
Exercises
7.1. The vectors e 1 ç e2 and e1 è e2 form a basis of R2 . What is the dual basis of R2 ?
é êë í í î ï
7.2. Let v1 , v2 , . . . , vn be a basis of Rn and let λ1 , λ2 , . . . , λn be the dual basis of í î
í îð ð
Rn . Let A be an invertible n n-matrix. Then by elementary linear algebra the set of
ð î
vectors Av1 , . . . , Avn is also a basis of Rn . Show that the corresponding dual basis is the
set of row vectors λ 1 A 1 , λ2 A 1 , . . . , λn A 1 .
ò
7.3. Suppose that µ is a bilinear function on a vector space V satisfying µ v, v 0 for
all vectors v V. Prove that µ is alternating. Generalize this observation to k-multilinear
é ê0ñ
functions.
7.4. Show that the bilinear function µ of Example 7.8 is equal to dx 1 dx2 ç dx3 dx4 .
7.5. The wedge product is a generalization of the cross product to arbitrary dimensions
ï ñ$óõô é ö êì÷
in the sense that
T
x y xT yT
ò ö ôé ö ê
for all x, y R3 . Prove this formula. (Interpretation: x and y are column vectors, x T and yT
are row vectors, x T yT is a 2-form on R3 , xT yT is a 1-form, i.e. a row vector. So both
ò ë
sides of the formula represent column vectors.)
7.6. Let V be a vector space and let µ 1 , µ2 , . . . , µk V be covectors. Their tensor
product is the function
µ1 ø øúùùQù6ø
µ2 µk : V k û R
ê0ñ é ê é ê ù ùù é ê
defined by
ø 5ø ùùùüø é
µ1 µ2 µ k v1 , v2 , . . . , vk µ 1 v1 µ 2 v 2 µ k vk .
Show that µ 1 ø øýùQù ù6ø
µ2 µk is a k-multilinear function.
90 7. DIFFERENTIAL FORMS ON MANIFOLDS
7.7. Let µ : V k þ R be a k-multilinear function. Define a new function Alt µ : V k þ R
by
ÿ
Alt µ v1 , v2 , . . . , vk
1
k! σ∑
S
ÿ ÿ
sign σ µ vσ 1 , vσ 2 , . . . , vσ k .
k
Prove the following.
(i) Alt µ is an alternating k-multilinear function.
(ii) Alt µ µ if µ is alternating.
(iii) Alt Alt µ Alt µ for all k-multilinear µ .
(iv) Let µ 1 , µ2 , . . . , µk V . Then
µ1 µ2 µk
1
k!
Alt µ1 µ2 ÿ µk .
ÿ
7.8. Show that det v1 , v2 , . . . , vn
v2 , . . . , vn Rn . In short,
dx1 dx2 dxn ÿ v1 , v2 , . . . , vn for all vectors v1 ,
dxn .
þ
det dx1 dx2
ÿ ÿ 7.9. Let V and W be vector spaces and L : V
L λ L µ for all covectors λ, µ W .
W a linear map. Show that L λµ ÿ
CHAPTER 8
Volume forms
8.1. n-Dimensional volume in R N
Let a1 , a2 , . . . , an be vectors in R N . The block or parallelepiped spanned by these
vectors is the set of all vectors of the form ∑ ni 1 ci ai , where the coefficients c i range
over the unit interval 0, 1 . For n 1 this is also called a line segment and for n 2
a parallelogram. We will need a formula for the volume of a block. If n N there
is no coherent way of defining an orientation on all n-blocks in R N , so this volume
will be not an oriented but an absolute volume. We approach this problem in a
similar way as the problem of defining the determinant, namely by imposing a
few reasonable axioms.
8.1. D EFINITION . An absolute n-dimensional Euclidean volume function is a func-
tion
vol n : R N R N R N
R
n times
with the following properties:
(i) homogeneity:
voln a1 , a2 , . . . , cai , . . . , an ! c voln a1 , a2 , . . . , an
for all scalars c and all vectors a1 , a2 , . . . , an ;
(ii) invariance under shear transformations:
vol n a1 , . . . , ai " ca j , . . . , a j , . . . , an voln a1 , . . . , a j , . . . , ai , . . . , an
for all scalars c and any i # j;
(iii) invariance under Euclidean motions:
voln Qa1 , . . . , Qan voln a1 , . . . , an
for all orthogonal matrices Q;
(iv) normalization: vol n e1 , e2 , . . . , en 1.
We shall shortly see that these axioms uniquely determine the n-dimensional
volume function.
8.2. L EMMA . (i) vol n a1 , a2 , . . . , an %$ a1 $&$ a2 $ $ an $ if a1 , a2 , . . . ,
an are orthogonal vectors.
(ii) vol n a1 , a2 , . . . , an 0 if the vectors a1 , a2 , . . . , an are dependent.
P ROOF. Suppose a1 , a2 , . . . , an are orthogonal. First assume they are nonzero.
Then we can define qi '$ ai $( 1 ai . The vectors q1 , q2 , . . . , qn are orthonormal.
Complete them to an orthonormal basis q1 , q2 . . . , qn , qn ) 1 , . . . , qN of R N . Let
91
92 8. VOLUME FORMS
Q be the matrix whose i-th column is qi . Then Q is orthogonal and Qei * qi .
Therefore
vol n + a1 , a2 , . . . , an , *.- a1 -&- a2 -0///1- an - voln + q1 , q2 , . . . , qn , by Axiom (i)
* -
. a1 -&- a2 -0///1- an - voln + Qe1 , Qe2 , . . . , Qen ,
*.- a1 -&- a2 -0///1- an - voln + e1 , e2 , . . . , en , by Axiom (iii)
*.- a1 -&- a2 -0///1- an - by Axiom (iv),
which proves part (i) if all ai are nonzero. If one of the ai is 0, the vectors a1 , a2 , . . . ,
an are dependent, so the statement follows from part (ii), which we prove next.
Assume a1 , a2 , . . . , an are dependent. For simplicity suppose a1 is a linear
combination of the other vectors, a1 * ∑ni2 2 ci ai . By repeatedly applying Axiom
(ii) we get
3 n
vol n + a1 , a2 , . . . , an , * vol n ∑ c i ai , a 2 , . . . , a n 4
23
i 2
n
* vol n ∑ ci ai , a2 , . . . , an 4 *5///6* vol n + 0, a2 , . . . , an , .
2
i 3
Now by Axiom (i),
vol n + 0, a2 , . . . , an , * voln + 0 0, a2 , . . . , an , * 0 voln + 0, a2 , . . . , an , * 0,
which proves property (ii). QED
This brings us to the volume formula. We can form a matrix A out of the
column vectors a1 , a2 , . . . , an . It does not make sense to take det A because A is
not square, unless n * N. However, the product A T A is square and we can take
its determinant.
8.3. T HEOREM . There exists a unique n-dimensional volume function on R N . Let
a1 , a2 , . . . , an 7 R N and let A be the N 8 n-matrix whose i-th column is a i . Then
voln + a1 , a2 , . . . , an , *59 det + A T A , .
P ROOF. We leave it to the reader to check that the function : det + A T A , sat-
isfies the axioms for a n-dimensional volume function on R N . (See Exercise 8.2.)
Here we prove only the uniqueness part of the theorem.
Case 1. First assume that a1 , a2 , . . . , an are orthogonal. Then A T A is a diagonal
matrix. Its i-th diagonal entry is - ai - 2 , so : det + A T A , *.- a1 -&- a2 -0///;- an - , which
is equal to voln + a1 , a2 , . . . , an , by Lemma 8.2(i).
Case 2. Next assume that a1 , a2 , . . . , an are dependent. Then the matrix A
has a nontrivial nullspace, i.e. there exists a nonzero n-vector v such that Av *
0. But then A T Av * 0, so the columns of A T A are dependent as well. Since
A T A is square, this implies det A T A * 0, so : det + A T A , * 0, which is equal to
voln + a1 , a2 , . . . , an , by Lemma 8.2(ii).
Case 3. Finally consider an arbitrary sequence of independent vectors a1 ,
a2 , . . . , an . This sequence can be transformed into an orthogonal sequence v 1 ,
v2 , . . . , vn by the Gram-Schmidt process. This works as follows: let b 1 * 0 and for
i < 1 let bi be the orthogonal projection of ai onto the span of a1 , a2 , . . . , ai = 1 ; then
8.1. n-DIMENSIONAL VOLUME IN RN 93
vi > ai ? bi . (See illustration below.) Let V be the N @ n-matrix whose i-th column
is vi . Then by repeated applications of Axiom (ii),
vol n A a1 , a2 , . . . , an B > vol n A v1 , a2 , . . . , an B > vol n A v1 , v2 , . . . , an B >5CCC
> voln A v1 , v2 , . . . , vn B >5D det A V T V B , (8.1)
where the last equality follows from Case 1. Since vi > ai ? bi , where bi is a linear
combination of a1 , a2 , . . . , ai E 1, we have V > AU, where U is a n @ n-matrix of the
form FGG JK
GG 1 I I CCC I KK
GH 0 1 I CCC I KK
U> 0 0 1 CCC I .
.. .. ..
. . . I L
0 0 CCC 0 1
Note that U has determinant 1. This implies that V T V > U T A T AU and
det A A T A B > det U T det A A T A B det U > det A U T A T AU B > det A V T V B .
Using formula (8.1) we get vol n A a1 , a2 , . . . , an B >NM det A A T A B . QED
The Gram-Schmidt process transforms a sequence of n independent vectors
a1 , a2 , . . . , an into an orthogonal sequence v 1 , v2 , . . . , vn . (The horizontal “floor”
represents the plane spanned by a1 and a2 .) The block spanned by the a’s has the
same volume as the rectangular block spanned by the v’s.
a3
v3
a3
a2
a2
a1 v2
b2
b3 a1 O v1
a3
v3
a2
v2
a1
v1
For n > N Theorem 8.3 gives the following result.
8.4. C OROLLARY. Let a1 , a2 , . . . , an be vectors in Rn and let A be the n @ n-matrix
whose i-th column is ai . Then voln A a1 , a2 , . . . , an B >.P det A P .
94 8. VOLUME FORMS
P ROOF. A is square, so det Q A T A RTS det A T det A SUQ det A R 2 by Theorem
3.7(ii) and therefore vol n Q a1 , a2 , . . . , an RVSXW Q det A R 2 SZY det A Y by Theorem 8.3.
QED
8.2. Orientations
Oriented vector spaces. You are probably familiar with orientations on vec-
tor spaces of dimension [ 3. An orientation of a line is an assignment of a di-
rection. An orientation of a plane is a choice of a direction of rotation, clockwise
versus counterclockwise. An orientation of a three-dimensional space is a choice
of “handedness”, i.e. a choice of a right-hand rule versus a left-hand rule.
These notions can be generalized as follows. Let V be an n-dimensional vec-
tor space over the real numbers. Suppose that \]S^Q v1 , v2 , . . . , vn R and \`_aS
Q v1_ , v2_ , . . . , vn_ R are two ordered bases of V. Then we can write vi_ S ∑ j ai, jv j and
vi S ∑ j bi, j v _ j for suitable coefficients a i, j and bi, j. The n b n-matrices A SZQ a i, j R
and B ScQ bi, j R satisfy AB S BA S I and are therefore invertible. We say that the
bases \ and \ _ define the same orientation of V if det A d 0. If det A e 0, the two
bases define opposite orientations.
For instance, if Q v1_ , v2_ , . . . , vn_ R&S5Q v2 , v1 , . . . , vn R , then
fgg ijj
gg 0 1 0 ... 0 jj
gh 1 0 0 ... 0 j
A S 0 0 1 ... 0 ,
.. .. .. .. ..
. . . . .k
0 0 0 ... 1
so det A Sml 1. Hence the ordered bases Q v2 , v1 , . . . , vn R and Q v1 , v2 , . . . , vn R define
opposite orientations.
We know now what it means for two bases to have the same orientation, but
how do we define the concept of an orientation itself? In typical mathematician’s
fashion we define the orientation of V determined by the basis \ to be the collec-
tion of all ordered bases that have the same orientation as \ . (There is an anal-
ogous definition of the number 1, namely as the collection of all sets that contain
one
o element.)
o The orientation determined by \^SnQ v1 , v2 , . . . , vn R is denoted
o o by
\qp or v1 , v2 , . . . , vn p . So if \ and \ _ define theo same orientation
o then q
\ r
p S \ _p.
If they define opposite orientations we write \qpsScl \ _ p . Because the determi-
nant of an invertible matrix is either positive or negative, there are two possible
orientations of V. An oriented vector space is a vector space together with a choice
of an orientation. This preferred orientation is then called positive.
For n S 0 we need to make a special definition, because a zero-dimensional
space has an empty basis. In this case we define an orientation of V to be a choice
of sign, t or l .
o
8.5. E XAMPLE . The standard orientation on Rn is the orientation e1 , . . . , en p de-
fined by the standard ordered basis Q e1 , . . . , en R . We shall always use this orienta-
tion on Rn .
Maps and orientations. Let V and W be oriented vector spaces of the same
dimension and let L : V u W be an invertible linear map. Choose a positively
oriented basis Q v1 , v2 , . . . , vn R of V. Because L is invertible, the ordered n-tuple
8.2. ORIENTATIONS 95
v
Lv1 , Lv2 , . . . , Lvn w is an ordered basis of W. If this basis is positively, resp. neg-
atively, oriented we say that L is orientation-preserving, resp. orientation-reversing.
v
This definition does not depend on the choice of the basis, for if v v1x , v2x , . . . , vnx w is
another positively oriented basis of V, then vix y ∑ j ai, j v j with det ai, j w{z 0. There-
v
fore Lvix y L | ∑ j ai, jv j } y ∑ j ai, j Lv j , and hence the two bases Lv 1 , Lv2 , . . . , Lvn w
v
and Lv1x , Lv2x , . . . , Lvnx w of W determine the same orientation.
Oriented manifolds. Now let M be a manifold. We define an orientation of
M to be a choice of an orientation for each tangent space Tx M which varies con-
tinuously over M. “Continuous” means that for every x ~ M there v exists a lo-
cal parametrization ψ : W M, with W open in Rn and x ~ ψ W w , such that
Dψy : Rn Ty M preserves the orientation for all y ~ W. (Here Rn is equipped
with its standard orientation.) A manifold is orientable if it possesses an orienta-
tion; it is oriented if a specific orientation has been chosen.
Hypersurfaces. The case of a hypersurface, a manifold of codimension 1, is
particularly instructive. A unit normal
v vector field on av manifold M in R n is a smooth
function n : M R such that n x w Tx M and n x w y 1 for all x ~ M.
n
8.6. P ROPOSITION . A hypersurface in Rn is orientable if and only if it possesses a
unit normal vector field.
P ROOF. Let M n
v be a hypersurface in R . Suppose M possesses a unit normal
vectorv field. v Let v1 , v2 , . . . , vn 1 w be an ordered basis
n , because n x
v of Tx M for some x ~ M.
Then
v n x w , v 1 , v 2 , . . . , v n 1 w is a basis of Rv v w& vi for all i. We say that
v1 , v2 , . . . , vn 1 w is positively oriented if n x w , v1 , v2 , . . . , vn 1 w is a positively ori-
ented basis of Rn . This defines an orientation on M, called the orientation induced
by the normal vector field n.
Conversely, let us suppose that M is an oriented hypersurface in R n . For each
x ~ M the tangent space Tx M is n 1-dimensional, so its orthogonal complement
v
Tx M w is a line. There are therefore precisely two vectors of length 1 which are
perpendicular
v to Tx M. We can pick a preferred unit normal vector as follows. Let
v1 , v2 , . . . , vn 1 w be a positively oriented v basis of Tv x M.
v The positive unit normal
vector is that unit normal vector n x w that makes n x w , v1 , v2 , . . . , vn 1 w a posi- v
tively oriented basis of Rn . In Exercise 8.8 you will be asked to check that n x w
depends smoothly on x. In this way we have produced a unit normal vector field
on M. QED
8.7. E XAMPLE . Let us regard Rn 1 as the subspace of Rn spanned by the first
n 1 standard basis vectors e1 , e2 , . . . , en 1 . The standard orientation on Rn is
e1 , e2 , . . . , en , and the standard orientation on Rn 1 is e1 , e2 , . . . , en 1 . Since
v
e1 , e 2 , . . . , e n y 1w n 1
e n , e1 , e2 , . . . , e n
1
v
by Exercise 8.5, the positive unit normal to Rn 1 in Rn is 1w n 1e
n.
The positive unit normal on an oriented hypersurface can be regarded M in Rn
as a map n from M into the unit sphere S n 1 , which is often called the Gauß map of
M. The unit normal enables one to distinguish between two sides of M: the direc-
tion of n is “out” or “up”; the opposite direction is “in” or “down”. For this reason
orientable hypersurfaces are often called two-sided, whereas the nonorientable ones
are called one-sided. Let us show that a hypersurface given by a single equation is
always orientable.
96 8. VOLUME FORMS
8.8. P ROPOSITION . Let U be open in Rn and let φ : U R be a smooth function.
Let c be a regular value of φ. Then the manifold φ 1 c has a unit normal vector field
given by n x & grad φ x grad φ x ; and is therefore orientable.
P ROOF. The regular value theorem tells us that M φ 1 c is a hypersurface
in Rn (if nonempty), and also that Tx M ker Dφx c grad φ x . The function
n x { grad φ x grad φ x ; therefore defines a unit normal vector field on M.
Appealing to Proposition 8.6 we conclude that M is orientable. QED
8.9. E XAMPLE . Taking φ x % x 2 and c r2 we obtain that the n 1 -
sphere of radius r about the origin is orientable. The unit normal is
n x grad φ x grad φ x ; x x .
8.3. Volume forms
Now let M be an oriented n-manifold in R N . Choose a collection of embed-
dings ψi : Ui R N with Ui open in Rn such that M i ψi Ui and such that
Dψi t : Rn Tx M is orientation-preserving for all t Ui . The volume form µ M ,
also denoted by µ , is the n-form on M whose local representative relative to the
embedding ψi is defined by
µi ψi µ . det Dψi t T Dψi t dt1 dt2 dtn .
By Theorem 8.3 the square-root factor measures the volume of the n-dimensional
block in the tangent space Tx M spanned by the columns of Dψ i t , the Jacobi ma-
trix of ψi at t. Hence you should think of µ as measuring the volume of infinitesi-
mal blocks inside M.
8.10. T HEOREM . For any oriented n-manifold M in R N the volume form µ M is a
well-defined n-form.
P ROOF. To show that µ is well-defined we need to check that its local repre-
sentatives satisfy the transformation law (7.1). So let us put φ ψ i 1 ψ j and sub-
stitute t φ u into µi . Since each of the embeddings ψ i is orientation-preserving,
we have det Dφ 0. Hence by Theorem 3.13 we have
φ dt1 dt2 dtn & det Dφ u du1 du2 dun ! det Dφ u ; du1 du2 dun .
Therefore
φ µi det Dψi φ u φ u det Dφ u; du1 du2 dun
T Dψ
i
det Dφ u T det Dψi φ u T Dψi φ u det Dφ u du1 du2 dun
det Dψi φ u Dφ u T Dψi φ u Dφ u du1 du2 dun
det Dψ j u T Dψ j u du1 du2 dun µ j ,
where in the second to last identity we applied the chain rule. QED
For n 1 the volume form is usually called the element of arc length, for n 2,
the element of surface area, and for n 3, the volume element. Traditionally these are
denoted by ds, dA, and dV, respectively. Don’t be misled by this old-fashioned no-
tation: volume forms are seldom exact! The volume form µ M is highly dependent
8.3. VOLUME FORMS 97
on the embedding of M into R N . It changes if we dilate or shrink or otherwise
deform M.
8.11. E XAMPLE . Let U be an open subset of Rn . Recall from Example 6.5
that U is a manifold covered by a single embedding, namely the identity map
ψ: U U, ψ ¡ x ¢£ x. Then det ¡ Dψ T Dψ ¢£ 1, so the volume form on U is simply
dt1 dt2 ¤¤¤ dtn , the ordinary volume form on Rn .
8.12. E XAMPLE . Let I be an interval in the real line and f : I R a smooth
function. Let M ¥ R be the graph of f . By Example 6.7 M is a 1-manifold in R 2 .
2
Indeed, M is the image of the embedding ψ : I R2 given by ψ ¡ t ¢£5¡ t, f ¡ t ¢ ¢ . Let
us give M the orientation induced by the embedding ψ, i.e. “from left to right”.
What is the element of arc length of M? Let us compute the pullback ψ ¦ µ , a 1-form
on I. We have
1 1
Dψ ¡ t ¢§£©¨
f ª«¡ t ¢¬
, Dψ ¡ t ¢ T Dψ ¡ t ¢0£¯® 1 f ª«¡ t ¢±° ¨
f ª«¡ t ¢¬
£ 1 ² f ª ¡ t ¢ 2,
so ψ ¦ µ £5³ det ¡ Dψ ¡ t ¢ T Dψ ¡ t ¢ ¢ dt £ ³ 1² f ª ¡ t ¢ 2 dt.
The next result can be regarded as an alternative definition of µ M . It is perhaps
more intuitive, but it requires familiarity with Section 7.2.
8.13. P ROPOSITION . Let M be an oriented n-manifold in R N . Let x ´ M and v1 ,
v2 , . . . , vn ´ Tx M. Then the volume form of M is given by
µ M,x ¡ v1 , v2 , . . . , vn ¢
¸ voln ¡ v1 , v2 , . . . , vn ¢ if v1 , v2 , . . . , vn are positively oriented,
£ µ· º
¶ voln ¡ v1 , v2 , . . . , vn ¢ if v1 , v2 , . . . , vn are negatively oriented,
0 if v1 , v2 , . . . , vn are linearly dependent,
·¹
i.e. µ M,x ¡ v1 , v2 , . . . , vn ¢ is the oriented volume of the n-dimensional parallelepiped in
Tx M spanned by v1 , v2 , . . . , vn .
P ROOF. For each x in M and n-tuple of tangent vectors v1 , v2 , . . . , vn at x let
ωx ¡ v1 , v2 , . . . , vn ¢ be the oriented volume of the block spanned by these n vectors.
This defines an n-form ω on M and we must show that ω £ µ M . Let U be an
open subset of Rn and ψ : U R N an orientation-preserving embedding with
ψ ¡ U ¢»¥ M and ψ ¡ t ¢»£ x for some t in U. Let us calculate the n-form ψ ¦ ω on U.
We have ψ ¦ ω £ g dt1 dt2 ¤¤¤ dtn for some function g. By Lemma 7.12 this function
is given by
g ¡ t ¢£ ψ ¦ ωx ¡ e1 , e2 , . . . , en ¢&£ ωx ¡ Dψ ¡ t ¢ e1 , Dψ ¡ t ¢ e2 , . . . , Dψ ¡ t ¢ en ¢ ,
where in the second equality we used the definition of pullback. The vectors
Dψ ¡ t ¢ e1 , Dψ ¡ t ¢ e2 , . . . , Dψ ¡ t ¢ en are a positively oriented basis of Tx M and, more-
over, are the columns of the matrix Dψ ¡ t ¢ , so by Theorem 8.3 they span a positive
volume of magnitude ³ det ¡ Dψ ¡ t¢ T Dψ ¡ t ¢ ¢ . This shows that g £ ³ det ¡ Dψ T Dψ ¢
and therefore
ψ¦ ω £5¼ det ¡ Dψ T Dψ ¢ dt1 dt2 ¤¤¤ dtn .
Thus ψ ¦ ω is equal to the local representative of µ M with respect to the embedding
ψ. Since this holds for all embeddings ψ, we have ω £ µ M . QED
98 8. VOLUME FORMS
Volume form of a hypersurface. For oriented hypersurfaces M in R n there is
a more convenient expression for the volume form µ M . Recall the vector-valued
forms ÁÂ ÁÂ
dx1 Ä dx1
dx ½U¾¿À ... à and Ä dx ½Å¾¿À ... Ã
dxn Ä dxn
introduced in Section 2.5. Let n be the positive unit normal vector field on M
and let F be any vector field on M, i.e. a smooth map F : M Æ R n . Then the
inner product F Ç n is a function defined on M. It measures the component of F
orthogonal to M. The product È F Ç n É µ M is an n Ê 1-form on M. On the other hand
we have the n Ê 1-form ÄËÈ F Ç dx É&½ F ÇÌÄ dx.
8.14. T HEOREM . On the hypersurface M we have
F ÇÌÄ dx ½5È F Ç n É µ M .
F IRST PROOF. This proof is short but requires familiarity with the material in
Section 7.2. Let x Í M. Let us change the coordinates on R n in such a way that the
first n Ê 1 standard basis vectors È e1 , e2 . . . , en Î 1 É form a positively oriented basis
of Tx M. Then, according to Example 8.7, the positive unit normal at x is given by
n È x ÉϽ¯È Ê 1 É n Ð 1en and the volume form satisfies µ M,x È e1 , . . . , en Î 1 ɽ 1. Writing
F ½ ∑niÑ 1 Fi ei , we have F È x ÉÒÇ n È x É&½NÈ Ê 1 É n Ð 1 Fn È x É . On the other hand
F ÇÄ dx ½ ∑È Ê 1É Ð
i 1
Fi dx1 ÇÇÇdx
Ó i ÇÇÇ dxn ,
i
and therefore È F ÇÌÄ dx ÉÌÈ e1 , . . . , en Î É ½5È Ê 1 É n Ð 1 Fn . This proves that
1
È F ÇÌÄ dx É x È e1 , . . . , en Î 1 É&½NÈ F È x ÉÔÇ n È x ÉÉ µ M È e1 , . . . , en Î 1 É ,
which implies È F ÇÕÄ dx É x ½UÈ F È x ÉÇ n È x É É µ M . Since this equality holds for every
x Í M, we find F ÇÌÄ dx ½5È F Ç n É µ M . QED
S ECOND PROOF. Choose an embedding ψ : U Æ Rn , where U is open in Rn Î 1 ,
such that ψ È U ÉVÖ M, x Í ψ È U É . Let t Í U be the point satisfying ψ È t ÉV½ x. As
a preliminary step in the proof we are going to replace the embedding ψ with a
new one enjoying a particularly nice property. Let us change the coordinates on
Rn in such a way that the first n Ê 1 standard basis vectors È e1 , e2 . . . , en Î 1 É form
a positively oriented basis of Tx M. Then at x the positive unit normal is given by
n È x ɽNÈ Ê 1 É n Ð 1en . Since the columns of the Jacobi matrix Dψ È t É are independent,
there exist unique vectors a1 , a2 , . . . , an Î 1 in Rn Î 1 such that Dψ È t É ai ½ ei for i ½ 1,
2, . . . , n Ê 1. These vectors ai are independent, because the ei are independent.
Therefore the È n Ê 1 ÉÏ×ØÈ n Ê 1 É -matrix A with i-th column vector equal to a i is
invertible. Put Ũ ½ A Î 1 È U É , t̃ ½ A Î 1 t and ψ̃ ½ ψ Ù A. Then Ũ is open in Rn Î 1 ,
ψ̃ È t̃ É&½ x, ψ̃ : Ũ Æ Rn is an embedding with ψ̃ È Ũ É&½ ψ È U É , and
Dψ̃ È t̃ ɽ Dψ È tÉÚÙ DA È t̃ ½ Dψ È t ÉÚÙ A
by the chain rule. Therefore the i-th column vector of Dψ̃ È t̃ É is
Dψ̃ È t̃ É ei ½ Dψ È t É Aei ½ Dψ È t É a i ½ ei (8.2)
8.3. VOLUME FORMS 99
for i Û 1, 2, . . . , n Ü 1. (On the left ei denotes the i-th standard basis vector in
Rn Ý 1 , on the right it denotes the i-th standard basis vector in Rn .) In other words,
the Jacobi matrix of ψ̃ at t̃ is the Þ n Ü 1 ßà n-matrix
In Ý 1
Dψ̃ Þ t̃ ß&Û©á ,
0 â
where In Ý 1 is the Þ n Ü 1 ßàãÞ n Ü 1 ß identity matrix and 0 denotes a row consisting
of n Ü 1 zeros.
Let us now calculate ψ̃ äËå Þ F æ n ß µ M ç and ψ̃ äèÞ F æèé dx ß at the point t̃. Writing
F æ n Û ∑niê 1 Fi ni and using the definition of µ M we get
n
ψ̃ ä
å Þ F æ n ß µ M ç Û©á ∑ ψ̃ ä Þ Fi ni ß âaë det Þ Dψ̃T Dψ̃ ß dt̃1 dt̃2 æææ dt̃n Ý 1.
iê 1
From formula (8.2) we have det Þ Dψ̃ Þ t̃ ß T Dψ̃ Þ t̃ ß ßÛ 1. So evaluating this expression
at the point t̃ and using n Þ x ß&ÛNÞ Ü 1 ß n ì 1en we get
å ψ̃ ä Þ F æ n ß µ M ç ÛNÞ Ü 1 ß n ì 1 Fn Þ x ß dt̃1 dt̃2 æææ dt̃n Ý 1 .
t̃
From F æé dx Û ì
ê Þ Ü 1ß
∑ ni 1 æææ6dx
í i æææ
i 1F dx1 dx2 dxn we get
i
n
ψ̃ ä Þ F æÌé dx ßÛ ∑ Þ Ü 1ß iì 1
ψ̃ ä Fi dψ̃1 dψ̃2 ææædî ψ̃i æææ dψ̃n .
i 1 ê
From formula (8.2) we see ∂ψ̃i Þ t̃ ß ï ∂t̃ j Û δi, j for 1 ð i, j ð n Ü 1 and ∂ψ̃n Þ t̃ ß ï ∂t̃ j Û 0
for 1 ð j ð n Ü 1. Therefore
ç t̃ Û.Þ Ü 1 ß n ì 1 Fn Þ x ß dt̃1 dt̃2 æææ dt̃n Ý 1 .
å ψ̃ ä Þ F æé dx ß
We conclude that å ψ̃ ä1Þ F æ n ß µ M ç t̃ Ûcå ψ̃ äÕÞ F æé dx ß ç t̃ , in other words å Þ F æ n ß µ M ç x Û
Þ F æÌé dx ß x. Since this holds for all x ñ M we have F æé dx Û5Þ F æ n ß µ M . QED
This theorem gives insight into the physical interpretation of n Ü 1-forms.
Think of the vector field F as representing the flow of a fluid or gas. The direc-
tion of the vector F indicates the direction of the flow and its magnitude measures
the strength of the flow. Then Theorem 8.14 says that the n Ü 1-form F æ;é dx mea-
sures, for any unit vector n in Rn , the amount of fluid per unit of time passing
through a hyperplane of unit volume perpendicular to n. We call F æé dx the flux
of the vector field F.
Another application of the theorem is the following formula for the volume
form on a hypersurface. The formula provides a heuristic interpretation of the
vector-valued form é dx: if n is a unit vector in Rn , then the scalar-valued n Ü 1-
form n æèé dx measures the volume of an infinitesimal n Ü 1-dimensional paral-
lelepiped perpendicular to n.
8.15. C OROLLARY. Let n be the unit normal vector field and µ M the volume form of
an oriented hypersurface M in Rn . Then
µM Û n æÌé dx.
P ROOF. Set F Û n in Proposition 8.14. Then F æ n Û 1 because ò n ò»Û 1. QED
100 8. VOLUME FORMS
8.16. E XAMPLE . Suppose the hypersurface M is given by an equation φ ó x ôsõ
c, where c is a regular value of a function φ : U ö R, with U open in R n . Then by
Proposition 8.8 M has a unit normal n õ grad φ ÷ø grad φ ø . The volume form is
therefore µ õnø grad φ ø;ù 1 grad φ ú;û dx. In particular, if M is the sphere of radius
R about the origin in Rn , then n ó x ô&õ x ÷ R, so µ M õ R ù 1 x úû dx.
Exercises
8.1. Deduce from Theorem 8.3 that the area of the parallelogram spanned by a pair of
vectors a, b in Rn is given by ü a üü b ü sin φ, where φ is the angle between a and b (which is
taken to lie between 0 and π ). Show that ü a ürü b ü sin φ ýþü a ÿ b ü in R 3 .
8.2. Check that the function voln a1 , a2 , . . . , an
Definition 8.1.
ý
det A T A satisfies the axioms of
8.3. Let u1 , u2 , . . . , uk and v1 , v2 , . . . , vl be vectors in R N satisfying ui v j ý
2, . . . , k and j ý 1, 2, . . . , l. (“The u’s are perpendicular to the v’s.”) Prove that
0 for i ý 1,
volk l u1 , u2 , . . . , uk , v1 , v2 , . . . , vl ý volk u1 , u2 , . . . , uk voll v1 , v2 , . . . , vl .
8.4. Let a1 , a2 , . . . , an be real numbers, let c 1
ý ∑ni 2
1 ai and let
1
0
0 a
0
1
a
1
, , 0 , 2
ý ..
ý ..
ý ..
ý 1 ..
.a
u1 . u2 . ..., un . un 1
c
0 0 1 n
a1 a2 an 1
be vectors in Rn 1 .
(i) Deduce from Exercise 8.3 that
voln u1 , u2 , . . . , un ý voln 1 u 1 , u2 , . . . , u n , u n . 1
1
(ii) Prove that
a aa 2
1
a1 a2 a1 a3 ... a1 an
a a 2 1 1 a22 a2 a3
... a2 an
ý 1
n
a3 a2 1 a23 ... a3 an
3 1 ∑ a2i .
.. .. .. ..
.
..
i 1
. . . .
an a1 an a2 an a3 ... 1 a2n
8.5. Justify the following identities concerning orientations of a vector space V. Here
the v’s form a basis of V (which in part (i) is n-dimensional and in parts (ii)–(iii) two-
dimensional).
v , v , . . . , v ý sign σ v , v , . . . , v .
(i) If σ Sn is any permutation, then
σ 1 σ 2 σ n 1 2 n
(ii) v , v ý v , v .
(iii) 3v , 5v ý v , v .
1 2 1 2
8.6. Let U be open in R and let f : U R be a smooth function. Let ψ : U R be
1 2 1 2
n n 1
the embedding ψ x ý x, f x and let M ý ψ U , the graph of f . Define an orientation on
M by requiring ψ to be orientation-preserving. Deduce from Exercise 8.4 that the volume
form of M is given by ψ µ ý 1 ü grad f x ü dx dx dx . M
2
1 2 n
8.7. Let M ý graph f be the oriented hypersurface of Exercise 8.6.
EXERCISES 101
*++ ∂ f - ∂x .//
(i) Show that the positive unit normal vector field on M is given by
++ ∂ f - ∂x //
1
n! &
" # 1$ %
n 1 + .. / .
2
1 ')( grad f x $( ,
" ∂ f - .∂x 0
2
n
(ii) Derive the formula ψ 1 µ !
& 1 '2( grad f x $3( dx dx
1
! f " x , x , . . . , x" $ . (Caution:4 44for consistency you
2 dx from Corollary
8.15 by substituting x %
M 1 2 n
must replace n with n ' 1 in Corollary 8.15.)
n 1 1 2 n
8.8. Show that the unit normal vector field n : M 5 R defined in the proof of Propo-
n
ψ : U 5 M of an open subset of M.)
sition 8.6 is smooth. (Compute n in terms of an orientation-preserving parametrization
8.9. Let ψ : a, b $65 R be an embedding. Let µ be the element of arc length on the
8 " n
curve M ! ψ a, b $ . Show that ψ 1 µ is the 1-form on a, b $ given by ( ψ 7 t $3( dt !
embedded
"
ψ 7 t $ ' ψ 7 t $ ' 4 44 ' ψ 7 t $ dt.
" "
1 " 2
2 " 2
n
" 2
CHAPTER 9
Integration and Stokes’ theorem on manifolds
In this chapter we will see how to integrate an n-form over an oriented n-
manifold. In particular, by integrating the volume form we find the volume of the
manifold. We will also discuss a version of Stokes’ theorem for manifolds. This
requires the slightly more general notion of a manifold with boundary.
9.1. Manifolds with boundary
The notion of a spherical earth developed in classical Greece around the time
of Plato and Aristotle. Older cultures (and also Western culture until the redis-
covery of Greek astronomy in the late Middle Ages) visualized the earth as a flat
disc surrounded by an ocean or a void. A closed disc is not a manifold, because
no neighbourhood of a point on the edge is the image of an open subset of R 2 un-
der an embedding. Rather, it is a manifold with boundary, a notion which can be
defined as follows. The n-dimensional halfspace is
Hn 9 : x < R = x > 0? .
; n
n
9@: x < R = x 9 0 ? 9 A
9B: < = C 0? .
The boundary of is Hn ∂Hn n Rn 1 and its interior is
n
int Hn x Rn x n
9.1. D EFINITION . An n-dimensional manifold with boundary (or n-manifold with
boundary) in R N is a subset M of R N such that for all x M there exist <
D E
D an open subset V R N containing x,
E
D an open subset U Rn ,
and an embedding ψ : U F
R N satisfying ψ U G H Hn I 9 VH M.
x 9 GI
You should compare this definition carefully with Definition 6.4 of a manifold. If
<
ψ t with t ∂Hn , then x is a boundary point of M. The boundary of M is the
set of all boundary points and is denoted by ∂M. Its complement M ∂M is the J
interior of M and is denoted by int M.
Somewhat confusingly, the boundary of a manifold with boundary is allowed
to be empty. If nonempty, the boundary ∂M is an n 1-dimensional manifold. J
Likewise the interior int M is an n-manifold.
9 A
The most obvious example of an n-manifold with boundary is the halfspace
: <
R n= C ?
Hn itself, which has boundary ∂Hn
xn
Rn 1 and interior the open halfspace x
0 . Here is a more interesting type of example, which generalizes the
graph of a function.
A
smooth function. Put U U 9 K KNM
9.2. E XAMPLE . Let U be an open subset of Rn 1 and let f : U R be a
R and write elements of U as xy with x in U and OP
KLF
K
103
104 9. INTEGRATION AND STOKES’ THEOREM ON MANIFOLDS
y in R. The region below the graph of f is the set consisting of all QR
x
in U such that
S T U
y
y f x .
y
∂M graph f V
We assert that the region below the graph is an n-manifold whose boundary is
exactly the graph of f . We will prove this by describing it as the image of a single
embedding. Define ψ : U W
Rn by
ψ X ut Y[Z X f T tU]t \ uY .
As in Example 6.3 one verifies that ψ is an embedding, using the fact that
Dψ X ut Y[Z X DI f^ T tU_\ 01Y ,
n 1
where 0 is the origin in R ^ . By definition therefore, the set M
n 1
an n-manifold in R with boundary ∂M ψ T U ` ∂H U . What are M and ∂M? A
Z ψ T U ` H U is n
n
Z
point Q R is in M if and only if it is of the form
x
y
n
X xyY[Z ψ X ut Y[Z X f T tU]t \ uY
for some Q R in U ` H . Since H is given by u a 0, this is equivalent to x b U c
t n n
and y S f T x U . Thus M is exactly the region below the graph. On ∂H we have
u
n
Z
u 0, so ∂M is given by the equality y
Z f T xU , i.e. ∂M is the graph.
9.3. E . If f : U cdW R is a vector-valued map one cannot speak about
m
the region “below” the graph, but one can do the following. Again put U U cfe
XAMPLE
R. Let N n g m \ 1 and think of R as the set of vectors Q R with x in R ^ and
Z
Z N x n 1
y in R . Define ψ : U W R by
y
m N
ψX X f T tU]\ t ue Y
t
u Y[Z m
This time we have
T U Z X DI f^ T tUh\ 0e Y
Dψ t n 1
m
and again ψ is an embedding. Therefore M ψ T U ` H U is an n-manifold in R
Z
with boundary ∂M ψ T U ` ∂H U . This time M is the set of points Q R of the form
n N
Z n x
y
X yxY Z X f T tU]\ t ue Y m
9.1. MANIFOLDS WITH BOUNDARY 105
i j k
with t U and u 0. Hence M is the set of points l m where x is in U j and where
x
n
y
y satisfies m 1 equalities and one inequality:
y1 o f p xq ,
1 o f p xq , . . . , y
y2 2 r o f r p xq , y s f p xq .
m 1 m 1 m m
Again ∂M is given by y o f p x q , so ∂M is the graph of f .
Here is an extension of the regular value theorem, Theorem 6.10, to manifolds
with boundary. The proof, which we will not spell out, is similar to that of Theo-
rem 6.10.
9.4. T HEOREM (regular value theorem for manifolds with boundary). Let U be
open in R N and let φ : U t
Rm be a smooth map. Let M be the set of x in R N satisfying
φ1 x p qo c1 , φ2 x p qo c2 , ..., φm r p xq o
p qs c.
1 cm r 1, φm x m
Suppose that c oup c , c , . . . , c q is a regular value of φ and that M is nonempty. Then
M is a manifold in R of codimension m n 1 and with boundary ∂M o φ r p c q .
1 2 m
N 1
. Let U o R , m o 1 and φ p x q owv x v . The set given by the
inequality φ p x q s 1 is then the closed unit ball x x i R y v x v s 1 z . Since
9.5. E XAMPLE n 2
n
grad φ p x q o 2x, any nonzero value is a regular value of φ. Hence the ball is an
n-manifold in R , whose boundary is φ r p 1 q , the unit sphere S r .
n 1 n 1
If more than one inequality is involved, singularities often arise. A simple
example is the closed quadrant in R2 given by the pair of inequalities x 0 and k
k
y 0. This is not a manifold with boundary because its edge has a sharp angle at
the origin. Similarly, a closed square is not a manifold with boundary.
However, one can show that a set given by a pair of inequalities of the form
a s p qs
f x b, where a and b are both regular values of a function f , is a manifold
with boundary. For instance, the spherical shell
x xi y s v xv s
Rn R 1 R2 z
is an n-manifold whose boundary is a union of two concentric spheres.
Other examples of manifolds with boundary are the pair of pants, a 2-manifold
whose boundary consists of three closed curves,
and the Möbius band shown in Chapter 1. The Möbius band is a nonorientable
manifold with boundary. We will not give a proof of this fact, but you can convince
106 9. INTEGRATION AND STOKES’ THEOREM ON MANIFOLDS
yourself that it is true by trying to paint the two sides of a Möbius band in different
colours.
An n-manifold with boundary contained in Rn (i.e. of codimension 0) is often
called a domain. For instance, a closed ball is a domain in R n .
To define the tangent space to a manifold with boundary M at a point x choose
U and ψ as in the definition and put
Tx M { Dψ t R n . | }~| }
As in the case of a manifold, this does not depend on the choice of the embedding
ψ. Now suppose x is a boundary point of M and let v Tx M be a tangent vector.
Then v { |}
Dψ t u for some u Rn . We say that v points inwards if u n 0 and
{
outwards if un 0. If un 0, then v is tangent to the boundary. In other words,
Tx ∂M { Dφ t Rn | }~| } . 1
The above picture of the pair of pants shows some tangent vectors at boundary
points that are tangent to the boundary or outward-pointing.
Orienting the boundary. Let M be an oriented manifold with boundary. The
orientation on M induces an orientation on ∂M by a method very similar to the
proof of Proposition 8.6. Namely, for x ∂M define n x Tx M to be the unique | }
outward-pointing tangent vector of length 1 which is orthogonal to Tx ∂M. This de-
fines the unit outward-pointing normal vector field on ∂M. A basis v1 , v2 , . . . , vn 1 | }
of Tx ∂M is called positively oriented if n x , v1 , v2 , . . . , vn 1 is a positively ori- | | } }
ented basis of Tx M. This defines an orientation of ∂M, called the induced orientation.
{
For instance, let M Hn with the standard orientation e1 , . . . , en . At each point
of ∂M {
Rn 1 the outward pointing normal is en . This implies that induced
orientation on ∂M is | }
1 n e1 , e2 , . . . , en 1 , because
e , e , e , . . . , e {| 1 } e , e , . . . , e
n 1 2 n 1
n
1 2 n 1 , en .
9.2. Integration over orientable manifolds
As we saw in Chapter 5, a form of degree n can be integrated over a chain
of dimension n. The integral does not change if we reparametrize the chain in
an orientation-preserving manner. This opens up the possibility of integrating an
n-form over an oriented n-manifold.
Let M be an n-dimensional oriented manifold (possibly with boundary) in R N
and let α be an n-form on M. To define the integral of α over M let us assume
that M is compact. (A subset of R N is called compact if it is closed and bounded;
see Appendix A.2. This assumption is made to ensure that the integral is a proper
there exists a smooth map c : 0, 1 n R N such that
integral and therefore converges.) For a start, let us also make the assumption that
c 0, 1 {
M andn
the restriction of c to 0, 1 | } n is an orientation-preserving embedding.
For instance, this assumption is satisfied for the n-sphere S n (see Exercise 5.5) and
the torus S1 S1 (see Exercise 9.4). The pullback c α is then an n-form on the cube
0, 1 n . We define
M
α { 0,1 c α.
n
9.2. INTEGRATION OVER ORIENTABLE MANIFOLDS 107
Suppose c̄ : 0, 1 n R N is a smooth map with the same properties as c. To ensure
that M α is well-defined we need to check the following equality.
c α c̄ α.
9.6. L EMMA .
0,1 n 0,1 n
S KETCH OF PROOF . Let us denote the closed cube 0, 1 by R and let U be n
the open cube 0, 1 . Let V U c ¡ c̄ U and V̄ U c̄ ¡ c U . The
n 1 1
complement of V and of V̄ in R are negligeable in the sense that
c α ¢ c α and c̄ α c̄ α. (9.1)
R V R V̄
¡ £
By assumption the restriction of c to U is an embedding. This implies that c : V
M is a bijection onto its image, and so we see that c 1 c̄ is a bijection from V̄
onto V. It is orientation-preserving, because c and c̄ are orientation-preserving.
Therefore, by Theorem 5.1,
c α c ¡ £ c̄ c α ¤ c £ c ¡ £ c̄ α c̄ α.
V V̄
1
V̄
1
V̄
Combining this with the equalities (9.1) we get the result. QED
Not every manifold can be covered with one single n-cube. However, it can
be shown that there always exists a finite collection of n-cubes c i : 0, 1 n M for
i 1, 2, . . . , k, such that
¥ ¦ §¨ ¨ ¨
(i) ki 1 ci 0, 1 n M,
§
(ii) ci 0, 1 n
§
c j 0, 1 n is empty for i j, ©
(iii) for each i the restriction of c i to 0, 1 n is an orientation-preserving em-
bedding.
We can then define
∑ c α,
k
¦
α i
M 0,1 n
i 1
and check as in Lemma 9.6 that the result does not depend on the maps c i . (The
condition (ii) on the maps is imposed to avoid “double counting” in the integral.)
9.7. D EFINITION . Let M a compact oriented manifold in R N . The volume of M
is vol M
M µ , where µ is the volume form on M. (If dim M 1, resp. 2, we
speak of the arc length, resp. surface area of M.) The integral of a function f on M
is defined as M f µ . The mean or average of f is the number f¯
n
vol M 1 M f µ . ª ¡
The centroid or barycentre of M is the point x̄ in R whose i-th coordinate is the
mean value of xi over M, i.e.
x̄i 1
vol M
M
xiµ .
The volume form depends on the embedding of M into R N , so the notions
defined above depend on the embedding as well.
The most important property of the integral is the following version of Stokes’
theorem, which can be viewed as a parametrization-independent version of The-
orem 5.10 and is proved in a similar way.
108 9. INTEGRATION AND STOKES’ THEOREM ON MANIFOLDS
9.8. T HEOREM (Stokes’ theorem for manifolds). Let α be an n 1-form on a «
compact oriented n-manifold with boundary M. Give the boundary ∂M the induced ori-
entation. Then ¬ ¬
M
dα ∂M
α.
9.3. Gauß and Stokes
Stokes’ theorem, Theorem 9.8, contains as special cases the integral theorems
of vector calculus. These classical results involve a vector field F
n
∑ ni 1 Fi ei de- ®
fined on an open subset U of R . As discussed in Section 2.5, to this vector field
¯ ®
corresponds a 1-form α F dx ∑ni 1 Fi dxi , which we can think of as the work
done by the force F along an infinitesimal line segment dx. We will now derive
the classical integral theorems by applying Theorem 9.8 to one-dimensional, resp.
n-dimensional, resp. two-dimensional manifolds M contained in U.
Fundamental theorem of calculus. If F is conservative, F grad g for a func-
tion g, then α grad g dx ¯
dg. If M is a compact oriented 1-manifold with
boundary in Rn , then M dg ° ±°
∂M g by Theorem 9.8. The boundary consists of
two points a and b (if M is connected). If the orientation of M is “from a to b”,
then a acquires a minus and b a plus. Stokes’ theorem therefore gives the funda-
mental theorem of calculus in Rn ,
¬
M
F dx ¯ g ² b³´« g ² a³ .
If we interpret F as a force acting on a particle travelling along M, then g stands «
for the potential energy of the particle in the force field. Thus the potential energy
of the particle decreases by the amount of work done.
Gauß’ divergence theorem. We have
µ α F ¯ µ dx and d α µ div F dx 1 dx2 ¯¶¯¶¯ dx .
µ ·² ¯
n
n
If N is a oriented hypersurface in R with positive unit normal n, then α F
³
n µ N on N by Theorem 8.14. In this situation it is best to think of F as the flow
² ³
vector field of a fluid, where the direction of F x gives the direction of the flow at
¸ ² ³¹¸
a point x and the magnitude F x gives the mass of the amount of fluid passing
µ
per unit time through a hypersurface of unit area placed at x perpendicular to the
² ³
vector F x . Then α describes the amount of fluid passing per unit time and per
unit area through the hypersurface N. For this reason the n 1-form α is also « µ
° µ
called the flux of F, and its integral over N the total flux through N.
Applying Stokes’ theorem to a compact domain M in Rn we get M d α
° ¬
∂M α . Written in terms of the vector field F this is Gauß’ divergence theorem,
¬
M
div F dx1 dx2 ¯¶¯¶¯ dx n
∂M
² F ¯ n³ µ ∂M .
Thus the total flux out of the hypersurface ∂M is the integral of div F over M. If
the fluid is incompressible (e.g. most liquids) then this formula leads to the inter-
pretation of the divergence of F (or equivalently d α ) as a measure of the sources µ
or sinks of the flow. Thus div F
0 for an incompressible fluid without sources
EXERCISES 109
or sinks. If the fluid is a gas and if there are no sources or sinks then div F x º »¼ 0
½
(resp. 0) indicates that the gas is expanding (resp. being compressed) at x.
Classical version of Stokes’ theorem. Now let M be a compact two-dimen-
sional oriented surface with boundary and let us rewrite Stokes’ theorem M dα ¾ ¿
¾∂M α in terms of the vector field F. The right-hand side represents the work of
F done around the boundary curve(s) of M, which is not necessarily 0 if F is not
conservative. The left-hand side has a nice interpretation if n 3. Then dα ¿ À ¿
Á ¿ Á3À
curl F dx, so dα curl F dx. Hence if n is the positive unit normal of the surface
M in R3 , then dα ¿uº Á »
curl F n µ M on M. In this way we get the classical formula
of Stokes, Â Â
º curl F Á n » µ ¿
M
M
∂M
Á
F dx.
In other words, the total flux of curl F through the surface M is equal to the work
done by F around the boundary curves of M. This formula shows that curl F, or
À
equivalently dα , can be regarded as a measure of the vorticity of the vector field.
Exercises
9.1. Let U be an open subset of Rn and let f , g : U Ã
Ä ÇÅ Æ Ä Å Ä Å
R be two smooth functions
Ä ´Å È È Ä Å
satisfying f x g x for all x in U. Let M be the set of all pairs x, y such that x in U and
(i) Draw a picture of M if U is the open unit disc given by x É y Æ
f x y g x .
ÌÍ 1 Ì x Ì y and g Ä x, yÅÎÊ 2 Ì x Ì y .
2 2 2 2
2 2 1 and f x, y Ä ÅËÊ
(ii) Show directly from the definition that M is a manifold with boundary. (Use
two embeddings to cover M.) What is the dimension of M and what are the
boundary and the interior?
Ä ´Å È È Ä Å
(iii) Give an example showing that M is not necessarily a manifold with boundary if
the condition f x y g x fails.
(i) Let α Ê x dy Ì y dx and let M be a compact domain in the plane R . 2
Show that Ï
9.2.
(ii) Apply the observation of part (i) to find the area enclosed by the astroid x Ê
α is twice the surface area of M.
∂M
cos t, y Ê sin t.
3 3
(iii) Let α Ê x ÐÒÑ dx and let M be a compact domain in R . Show that Ï α is a n
∂M
constant times the volume of M. What is the value of the constant?
9.3. Write the divergence theorem for the vector field F Ê Ì cx e on R , where c is n n
n
a positive constant. Deduce Archimedes’ Law: the buoyant force exerted on a submerged
body is equal to the weight of the displaced fluid. Eύρηκα !
9.4. Let R1 Ó R2 Ó 0 be constants. Define a 3-cube c : 0, R 2 Ô ÕNÖ Ô 0, 2π ÕÖ Ô 0, 2π Õ Ã R3
by
ÙÚ Ä R ÉÉ Å ÙÚ
×Ø ÊÛ×Ø Ä R Å
r 1 r cos θ2 cos θ1
c θ1 1 r cos θ2 sin θ1 .
θ2 r sin θ2
(i) Sketch the image of c.
(ii) Let x 1 , x2 , x3 be the standard coordinates on R3 . Compute c dx1 , c dx2 , c dx3 Ü Ü Ü
ÜÄ
and c dx1 dx2 dx3 . Å
(iii) Find the volume of the solid parametrized by c.
(iv) Find the surface area of the boundary of this solid.
110 9. INTEGRATION AND STOKES’ THEOREM ON MANIFOLDS
9.5. Let M be a compact domain in Rn . Let f and g be smooth functions on M. The
Ý ]Þ ßà Ý
Dirichlet integral of f and g is D f , g á Þ
M grad f grad g µ , where µ dx 1 dx2 dxn is ß áá á
the volume form on M.
ãÝ â ÞäßÝ
(i) Show that d f dg á Þ
â ßåÝçæ Þ
grad f grad g µ .
æ ß é
Ý Ýãâ Þ ÞËßåè Ý
g µ , where g ∑ni 1 ∂2 g ∂x2i .
á ê æ Þ
(ii) Show that d dg
é
(iii) Deduce from parts (i)–(ii) that d f dg grad f grad g f g µ .
ë Ý Þ ß ë ∂gá
(iv) Let n be the outward-pointing unit normal vector field on ∂M. Write ∂g ∂n for
the directional derivative Dg n grad g n. Show that
f Ýãâ dg Þìß f µ . ∂M
∂n ∂M ∂M
ë ∂g
(v) Deduce from parts (iii) and (iv) Green’s formula,
ë
f
∂n
µ ß D Ý f , g Þê
∂M
Ý f æ gÞ µ. ∂M
M
(vi) Deduce Green’s symmetric formula,
ë ë
∂M í f
∂g
∂n î g
∂f
µ
∂n ∂M ï ß M
Ý f æ g î gæ f Þ µ.
Ý Þ
9.6. In this problem we will calculate the volume of a ball and a sphere in Euclidean
space. Let B R be the closed ball of radius R about the origin in R n . Then its boundary
Ý Þäß Ý Þ Ý Þäß Ý Þ Ý ÞËß ð Ý Þ
ß ÝÞ ß ÝÞ
S R ∂B R is the sphere of radius R. Put Vn R voln B R and An R voln 1 S R .
Also put Vn Vn 1 and An An 1 .
Ý Þ
Ý Þ Ý ÞËßñà ò ó
(i) Deduce from Corollary 8.15 that the volume form on S R is the restriction of ν
Ý Þôß ð
to S R , where ν is as in Exercise 2.16. Conclude that A n R S R ν.
Ý Þôß
Rn Vn and An R Rn 1 An . (Substitute y ß
Ý Þ Ý Þ
(ii) Show that Vn R Rx in the
õ ö÷Þùø ø Ý Þúß
volume forms of B R and S R .)
R be a continuous function. Define g : R n
Ýüû ë û Þ
(iii) Let f : 0, R by g x
ë
f x . Use Exercise 2.16(ii) to prove that
ë
g dx dx á áá dx ß f Ý r Þ A Ý r Þ dr ß A f Ý r Þ r ð dr.
R R
òó
n 1
1 2 n n n
B R 0 0
(iv) Show that ý ëÿþ ëôþ ð ð
ð ß
n
ðþ
r2 r2 n 1
e dr A e r dr. n
(Take f Ý r ÞËß e ð
0
in part (iii) and let R ø ö .)
r2
(v) Using Exercises B.10 and B.11 conclude that
2 π
ß ß and A ß
n
2π 2 2π m m 1 m
An
n ,
Ý
whence
m
î 1 Þ ! 1 á á
A 2m
3 5 á á áÝ 2m 1 Þ
î
. 2m 1
(vi) By taking f Ý r Ë
Þ ß 1 in part (iii) show that A ß nV and A Ý R ÞËß ∂V Ý RÞ é ∂R.
2
n n n n
(vii) Deduce that
ß ß ß 1 á 3 á 25 á á á Ýπ2m ê 1 Þ .
n
πm m 1 m
ê
π2
Vn n , whence V2m
m!
and V2m 1
2 1
(viii) Complete the following table. (Conventions: a space of negative dimension is
empty; the volume of a zero-dimensional manifold is its number of points.)
n 0 1 2 3 4 5
Ý Þ π R2 4 3
Ý Þ
Vn R 3 πR
An R 2π R
EXERCISES 111
(ix) Find limn An , limn Vn and limn An 1 An . Use Stirling’s formula,
x 1 ex
lim 2π .
x xx
1
2
CHAPTER 10
Applications to topology
10.1. Brouwer’s fixed point theorem
Let M be a manifold, possibly with boundary. A retraction of M onto a subset
A is a smooth map φ : M A such that φ x x for all x in A. For instance, let
M be the punctured unit ball in n-space,
M x Rn 0 x 1! .
Then the normalization map φ x " x #$ x is a retraction of M onto its boundary
A ∂M, the unit sphere. The following theorem says that a retraction onto the
boundary is impossible if M is compact and orientable.
10.1. T HEOREM . Let M be a compact orientable manifold with nonempty boundary.
Then there does not exist a retraction from M onto ∂M.
P ROOF. Suppose φ : M ∂M was a retraction. Let us choose an orientation
of M and equip ∂M with the induced orientation. Let β µ ∂M be the volume form
on the boundary (relative to some embedding of M into R N ). Let α φ % β be its
pullback to M. Let n denote the dimension of M. Note that β is an n & 1-form
on the n & 1-manifold ∂M, so dβ 0. Therefore dα dφ % β φ % dβ 0 and
hence by Stokes’ theorem 0 (' M dα )' ∂M α . But φ is a retraction onto ∂M, so the
restriction of φ to ∂M is the identity map and therefore α β on ∂M. Thus
0 )* α )* β vol ∂M + 0,
∂M ∂M
which is a contradiction. Therefore φ does not exist. QED
This brings us to one of the oldest results in topology. Suppose f is a map from
a set X into itself. An element x of X is a fixed point of f if f x , x.
10.2. T HEOREM (Brouwer’s fixed point theorem). Every smooth map from the
closed unit ball into itself has at least one fixed point.
P ROOF. Let M - x Rn x . 1 ! be the closed unit ball. Suppose
f : M M was a smooth map without fixed points. Then f x / + x for all x. For
each x in the ball consider the halfline starting at f x and pointing in the direction
of x. This halfline intersects the unit sphere ∂M in a unique point that we shall call
113
114 10. APPLICATIONS TO TOPOLOGY
φ 0 x 1 , as in the following picture.
f 2 x3
φ 2 y3 x
y
f 2 y3
φ 2 x3
This defines a smooth map φ : M 4 ∂M. If x is in the unit sphere, then φ 0 x 1"5 x,
so φ is a retraction of the ball onto its boundary, which contradicts Theorem 10.1.
Therefore f must have a fixed point. QED
This theorem can be stated imprecisely as saying that after you stir a cup of
coffee, at least one molecule must return to its original position. Brouwer origi-
nally stated his result for arbitrary continuous maps. This more general statement
can be derived from Theorem 10.2 by an argument from analysis which shows
that every continuous map is homotopic to a smooth map. (See Section 10.2 for
the definition of homotopy.) The theorem also remains valid if the closed ball is
replaced by a closed cube or a similar shape.
10.2. Homotopy
Definition and first examples. Suppose that φ 0 and φ1 are two maps from a
manifold M to a manifold N and that α is a form on N. What is the relationship
between the pullbacks φ 06 α and φ16 α ? There is a reasonable answer to this question
if φ0 can be smoothly deformed into φ 1 . More formally, we say that φ 0 and φ1
are homotopic if there exists a smooth map φ : M 78 0, 1 9:4 N such that φ 0 x, 0 1"5
φ0 0 x 1 and φ 0 x, 1 1;5 φ 1 0 x 1 for all x in M. The map φ is called a homotopy. Instead of
φ 0 x, t 1 we often write φ t 0 x 1 . Then each φ t is a map from M to N and we can think
of φt as a family of maps parametrized by t in the unit interval that interpolates
between φ0 and φ1 , or as a one-second “movie” that at time 0 starts at φ 0 and at
time 1 ends up at φ 1 .
10.3. E XAMPLE . Let M 5 N 5 Rn and φ0 0 x 1<5 x (identity map) and φ 1 0 x 1,5 0
(constant map). Then φ 0 and φ1 are homotopic. A homotopy is given by φ 0 x, t 1,5
0 1 = t 1 x. This homotopy collapses Euclidean space onto the origin by moving
each point radially inward. There are other ways to accomplish this. For instance
0 1 = t 1 2 x and 0 1 = t2 1 x are two other homotopies between the same maps. We can
also interchange φ 0 and φ1 : if φ0 0 x 1<5 0 and φ1 0 x 1,5 x, then we find a homotopy
by reversing time (playing the movie backwards), φ 0 x, t 1,5 tx.
10.4. E XAMPLE . Let M 5 N be the punctured Euclidean space R n =?> 0 @ and
let φ0 0 x 1A5 x (identity map) and φ 1 0 x 1B5 x C$D x D (normalization map). Then φ 0
and φ1 are homotopic. A homotopy is given for instance by φ 0 x, t 1E5 x C$D x D t or
by φ 0 x, t 1/5F0 1 = t 1 x G tx C$D x D . Either of these homotopies collapses punctured
Euclidean space onto the unit sphere about the origin by smoothly stretching or
shrinking each vector until it has length 1.
10.2. HOMOTOPY 115
10.5. E XAMPLE . A manifold M is said to be contractible if there exists a point
x0 in M such that the constant map φ 0 H x IKJ x0 is homotopic to the identity map
φ H x I<J x. A specific homotopy φ : M LNM 0, 1 O$P M from φ 0 to φ1 is a contraction of
M onto x0 . (Perhaps “expansion” would be a more accurate term, a “contraction”
being the result of replacing t with 1 Q t.) Example 10.3 shows that R n is con-
tractible onto the origin. (In fact it is contractible onto any point x 0 . Can you write
a contraction of Rn onto x0 ?) The same formula shows that an open or closed ball
around the origin is contractible. We shall see in Theorem 10.19 that punctured
n-space Rn Q?R 0 S is not contractible.
Homotopy of curves. If M is an interval M a, b O and N any manifold, then maps
from M to N are nothing but parametrized curves in N. A homotopy of curves can
be visualized as a piece of string moving through the manifold N.
a b N
Homotopy of loops. A loop in a manifold N is a smooth map from the unit
circle S1 into N. This can be visualized as a thin rubber band sitting in N. A
homotopy of loops φ : S1 LM 0, 1 O;P N can be pictured as a rubber band floating
through N from time 0 until time 1.
N
1
S
10.6. E XAMPLE . Consider the two loops φ 0 , φ1 : S1 P R2 in the plane given
by φ0 H x ITJ x and φ1 H x IUJ x VXW 20 Y . A homotopy of loops is given by shifting
φ0 to the right, φt H x IZJ x V[W 2t
0 Y . What if we regard φ 0 and φ 1 as loops in the
punctured plane R Q.R 0 S ? Clearly the homotopy φ does not work, because it
2
moves the loop through the forbidden point 0. (E.g. φ t H x IEJ 0 for x J W]\ 10 Y and
t J 1 ^ 2.) In fact, however you try to move φ 0 to φ1 you get stuck at the origin, so
116 10. APPLICATIONS TO TOPOLOGY
it seems intuitively clear that there exists no homotopy of loops from φ 0 to φ1 in
the punctured plane. This is indeed the case, as we shall see in Example 10.13.
The homotopy formula. The product M _a` 0, 1b is often called the cylinder
with base M. The two maps defined by ι 0 c x d,e c x, 0 d and ι1 c x d<e c x, 1 d send M to
the bottom, resp. the top of the cylinder. A homotopy ι : M _` 0, 1b,f M _` 0, 1 b
between these maps is given by the identity map ι c x, t d,e c x, t d . (“Slide the bottom
to the top at speed 1.”)
ι1
ι4 g 10
ι0
base cylinder
If M is an open subset of Rn , a k h 1-form on the cylinder can be written as
γe ∑ f I c x, t d dx I h ∑ g J c x, t d dt dx J ,
I J
with I running over multi-indices of degree k h 1 and J over multi-indices of de-
gree k. (Here we write the dt in front of the dx’s because that is more convenient in
what follows.) The cylinder operator turns forms on the cylinder into forms on the
base lowering the degree by 1,
kj 1 k
κ: i c M _k` 0, 1 bldmfni c Md ,
by taking the piece of γ involving dt and integrating it over the unit interval,
1
κγ e ∑ g J c x, t d dt q dx J .
J op 0
(In particular κγ e 0 for any γ that does not involve dt.) For a general manifold
M we can write a k h 1-form on the cylinder as γ e β h dt γ , where β and γ are
forms on M _` 0, 1 b (of degree k h 1 and k respectively) that do not involve dt. We
then define κγ e(r 01 γ dt.
The following result will enable us to compare pullbacks of forms under ho-
motopic maps. It can be regarded as an application of Stokes’ theorem, but we
shall give a direct proof.
10.7. L EMMA (cylinder formula). Let M be a manifold. Then ι 1s γ t ι0s γ e κ dγ h
dκγ for all k h 1-forms γ on M _k` 0, 1b . In short,
ι1s t ι0s e κ d h dκ .
P ROOF. We write out the proof for an open subset of Rn . The proof for ar-
bitrary manifolds is similar. It suffices to consider two cases: γ e f dx I and
γ e g dt dx J .
10.2. HOMOTOPY 117
Case 1. If γ u f dx I , then κγ u 0 and dκγ u 0. Also
∂f ∂f ∂f
dγ u
∂t
dt dx I v ∑ ∂xi dxi dx I u ∂t
dt dx I v terms not involving dt,
i
so
1 ∂f
dκγ v κ dγ u κ dγ uxwy x, t { dt | dx I
0 ∂t z
u~} f x, 1 { f x, 0 { dx I u ι1 γ ι0 γ .
z z
Case 2. If γ u g dt dx J , then ι0 γ u ι1 γ u 0 and
∂g ∂g
dγ u ∑ ∂xi dxi dt dx J u ∑ ∂xi dt dxi dx J ,
i i
so
n 1 ∂g
κ dγ u ∑w y 0 ∂xi z
x, t { dt | dxi dx J .
i 1
1
Also κγ u }y g x, t { dt dx J , so
0 z
n n
∂ 1 1 ∂g
dκγ u ∑ ∂xi w y 0
g x, t { dt | dxi dx J u
z ∑w y 0 ∂xi z
x, t { dt | dxi dx J .
i 1 i 1
Hence dκγ v κ dγ u 0 u ι 1 γ ι0 γ . QED
Now suppose we have a pair of maps φ 0 and φ1 going from a manifold M
to a manifold N and that φ : M 0, 1 < N is a homotopy between φ 0 and φ1 .
For x in M we have φ ι 0 x {Tu φ x, 0 {Tu φ0 x { , in other words φ 0 u φ ι0 .
z z z
Similarly φ1 u φ ι1 . Hence for any k v 1-form α on N we have ι 0 φ α u φ0 α and
ι1 φ α u φ1 α . Applying the cylinder formula to γ u φ α we see that the pullbacks
φ0 α and φ1 α are related in the following manner.
10.8. T HEOREM (homotopy formula). Let φ 0 , φ1 : M N be smooth maps from
a manifold M to a manifold N and let φ : M 0, 1 N be a homotopy from φ 0 to φ1 .
Then φ1 α φ0 α u κφ dα v dκφ α for all k v 1-forms α on N. In short,
φ1 φ0 u κφ d v dκφ .
In particular, if dα u 0 we get φ 1 α u φ0 α v dκφ α .
10.9. C OROLLARY. If φ 0 , φ1 : M N are homotopic maps between manifolds and
α is a closed form on N, then φ 0 α and φ1 α differ by an exact form.
This implies that if the degree of α is equal to the dimension of M, φ 0 α and
φ1 α have the same integral.
10.10. T HEOREM . Let M and N be manifolds and let α be a closed n-form on N,
where n u dim M. Suppose M is compact and oriented and has no boundary. Let φ 0 and
φ1 be homotopic maps from M to N. Then
y φ0 α u)y φ1 α .
M M
118 10. APPLICATIONS TO TOPOLOGY
P ROOF. By Corollary 10.9, φ 1 α φ0 α dβ for an n 1-form β on M. Hence
by Stokes’ theorem
φ1 α φ0 α ; dβ β 0,
M M ∂M
because ∂M is empty. QED
A LTERNATIVE PROOF. Here is a proof based on Stokes’ theorem for the mani-
fold with boundary M 0, 1 . The boundary of M 0, 1 consists of two copies
of M, namely M 1 and M 0 , the first of which is counted with a plus
sign and the second with a minus. Therefore, if φ : M 0, 1 N is a homotopy
between φ0 and φ1 ,
0 φ dα dφ α φ α φ1 α φ0 α .
M 0,1 M 0,1 ∂ M 0,1 M M
QED
If M is the circle S1 , N the punctured plane R2 . 0 and α the angle form
y dx x dy ¡ x2 y2 of Example 3.8, then a map from M to N is a loop in N
and the integral of α is 2π times the winding number of the loop. Thus Theorem
10.10 gives the following result.
10.11. C OROLLARY. Homotopic loops in R 2 a 0 have the same winding number
about the origin.
10.12. E XAMPLE . Unfolding the three self-intersections in the curve pictured
below does not affect its winding number.
0 0
10.13. E XAMPLE . The two circles φ 0 and φ1 of Example 10.6 have winding
number 1, resp. 0 and therefore are not homotopic (as loops in the punctured
plane).
10.3. Closed and exact forms re-examined
The homotopy formula throws light on our old problem of when a closed form
is exact, which we looked into in Section 2.3. The answer turns out to depend on
the “shape” of the manifold on which the forms are defined. On some manifolds
all closed forms (of positive degree) are exact, on others this is true only in certain
degrees. Failure of exactness is typically detected by integrating over a submani-
fold of the correct dimension and finding a nonzero answer. In a certain sense all
obstructions to exactness are of this nature. We shall not attempt to say the last
10.3. CLOSED AND EXACT FORMS RE-EXAMINED 119
word on this problem, but study a few representative special cases. The matter is
explored in [Fla89] and at a more advanced level in [BT82].
0-forms. A closed 0-form on a manifold is a smooth function f satisfying
d f ¢ 0. This means that f is constant (on each connected component of M). If
this constant is nonzero, then f is not exact (because forms of degree £ 1 are by
definition 0). So a closed 0-form is never exact (unless it is 0) for a rather uninter-
esting reason.
1-forms and simple connectivity. Let us now consider 1-forms on a manifold
M. Theorem 4.5 says that the integral of an exact 1-form along a loop is 0. With
a stronger assumption on the loop the same is true for arbitrary closed 1-forms. A
loop c : S1 ¤ M is null-homotopic if it is homotopic to a constant loop. The integral
of a 1-form along a constant loop is 0, so from Theorem 10.10 (where we set the M
of the theorem equal to S 1 ) we get the following.
10.14. P ROPOSITION . Let c be a null-homotopic loop in M. Then ¥ c α ¢ 0 for all
closed forms α on M.
A manifold is simply connected if every loop in it is null-homotopic.
10.15. T HEOREM . All closed 1-forms on a simply connected manifold are exact.
P ROOF. Let α be a closed 1-form and c a loop in M. Then c is null-homotopic,
so ¥ c α ¢ 0 by Proposition 10.14. The result now follows from Theorem 4.5. QED
10.16. E XAMPLE . The punctured plane R2 £?¦ 0 § is not simply connected, be-
cause it possesses a nonexact closed 1-form. (See Example 4.6.) In contrast it can be
proved that for n ¨ 3 the sphere S n © 1 and punctured n-space Rn £¦ 0 § are simply
connected. Intuitively, the reason is that in two dimensions a loop that encloses
the puncture at the origin cannot be crumpled up to a point without getting stuck
at the puncture, whereas in higher dimensions there is enough room to slide any
loop away from the puncture and then squeeze it to a point.
The Poincaré lemma. On a contractible manifold all closed forms of positive
degree are exact.
10.17. T HEOREM (Poincaré lemma). All closed k-forms on a contractible manifold
are exact for k ¨ 1.
P ROOF. Let M be a manifold and let φ : M ªk« 0, 1 ¬ ¤ M be a contraction onto
a point x0 in M, i.e. a smooth map satisfying φ x, 0 ®m¢ x0 and φ x, 1 ®m¢ x for all x.
Let α be a closed k-form on M with k ¨ 1. Then φ 1¯ α ¢ α and φ0¯ α ¢ 0, so putting
β ¢ κφ ¯ α we get
dβ ¢ dκφ ¯ α ¢ φ1¯ α £ φ0¯ α £ κ dφ ¯ α ¢ α .
Here we used the homotopy formula, Theorem 10.8, and the assumption that dα ¢
0. Hence dβ ¢ α . QED
The proof provides us with a formula for the “antiderivative”, namely β ¢
κφ ¯ α , which can be made quite explicit in certain cases.
120 10. APPLICATIONS TO TOPOLOGY
10.18. E XAMPLE . Let M be Rn and let φ ° x, t ±³² tx be the radial contraction.
Let α ² ∑i gi dxi be a 1-form. Then
φ´ α ² ∑ gi ° tx± d ° txi ±;² ∑ gi ° tx ±µ° xi dt ¶ t dxi ± ,
i i
so
1
β ² κφ ´ α ² ∑ xi · 0
gi ° tx ± dt.
i
According to the proof of the Poincaré lemma, the function β satisfies dβ ² α pro-
vided that dα ² 0. It is instructive to compare β with the function f constructed
in the proof of Theorem 4.5. (See Exercise 10.5.)
Another typical application of the Poincaré lemma is showing that a manifold
is not contractible by exhibiting a closed form that is not exact. For example, the
punctured plane R2 ¸º¹ 0 » is not contractible because it possesses a nonexact closed
1-form, namely the angle form. (See Example 4.6.) The angle form generalizes to
an n ¸ 1-form on punctured n-space Rn ¸¹ 0 » ,
x ¼¾½ dx
α² ¿ ¿n .
x
10.19. T HEOREM . α is a closed but non-exact n ¸ 1-form on punctured n-space.
Hence punctured n-space is not contractible.
P ROOF. dα ² 0 follows from Exercise 2.1(ii). The n ¸ 1-sphere M ² S n À 1 has
unit normal vector field x, so by Corollary 8.15 on M we have α ² µ , the volume
form. Hence Á M α ² vol M ² Â 0. On the other hand, suppose α was exact, α ² dβ
for an n ¸ 1-form β. Then
· α² · dβ ² · α² 0
M M ∂M
by Stokes’ theorem, Theorem 9.8. This is a contradiction, so α is not exact. It now
follows from the Poincaré lemma, Theorem 10.17, that R n ¸N¹ 0 » is not contractible.
QED
Using the same form α , but restricting it to the unit sphere S n À 1 , we see that
Sn À 1
is not contractible. But how about forms of degree not equal to n ¸ 1? With-
out proof we state the following fact.
10.20. T HEOREM . On Rn ¸a¹ 0 » and on Sn À 1 every closed form of degree k ² Â 1,
n ¸ 1 is exact.
For a compact oriented hypersurface without boundary M contained in in
Rn ¸?¹ 0 » the integral
1 x ¼µ½ dx
n À 1 · ¿ ¿n
vol n À 1 S M x
is the winding number of M about the origin. It generalizes the winding number
of a closed curve in R2 ¸¹ 0 » around the origin. It can be shown that the winding
number in any dimension is always an integer. It provides a measure of how
many times the hypersurface wraps around the origin. For instance, the proof
of Theorem 10.19 shows that the winding number of the n ¸ 1-sphere about the
origin is 1.
10.3. CLOSED AND EXACT FORMS RE-EXAMINED 121
Contractibility versus simple connectivity. Theorems 10.15 and 10.17 sug-
gest that the notions of contractibility and simple connectivity are not indepen-
dent.
10.21. P ROPOSITION . A contractible manifold is simply connected.
P ROOF. Use a contraction to collapse any loop onto a point.
x0 x0
c1 c1
M M
Formally, let c1 : S1 Ã M be a loop, φ : M ÄÆÅ 0, 1 Ç Ã M a contraction of M onto x0 .
Put c È s, t ÉKÊ φ È c1 È s É , t É . Then c is a homotopy between c 1 and the constant loop
c0 È t É;Ê φ È c1 È s É , 0 ÉmÊ x0 positioned at x0 . QED
Í As mentioned in Example 10.16, the sphere S n Ë 1 and punctured n-space Rn Ì
0 Î are simply connected for n Ï 3, although it follows from Theorem 10.19 that
they are not contractible. Thus simple connectivity is weaker than contractibility.
The Poincaré conjecture. Not long after inventing the fundamental group
Poincaré posed the following question. Let M be a compact three-dimensional
manifold without boundary. Suppose M is simply connected. Is M homeomor-
phic to the three-dimensional sphere? (This means: does there exist a bijective
map M Ã S3 which is continuous and has a continuous inverse?) This question
became (inaccurately) known as the Poincaré conjecture. It is famously difficult and
was the force that drove many of the developments in twentieth-century topology.
It has an n-dimensional analogue, called the generalized Poincaré conjecture, which
asks whether every compact n-dimensional manifold without boundary which is
homotopy equivalent to Sn is homeomorphic to Sn . We cannot here go into this
fascinating problem in any serious way, other than to report that it has now been
completely solved. Strangely, the case n Ï 5 of the generalized Poincaré conjec-
ture conjecture was the easiest and was confirmed by S. Smale in 1960. The case
n Ê 4 was done by M. Freedman in 1982. The case n Ê 3, the original version of
the conjecture, turned out to be the hardest, but was finally confirmed by G. Perel-
man in 2002-03. For a discussion and references, see the paper Towards the Poincaré
conjecture and the classification of 3-manifolds by J. Milnor, which appeared in the
November 2003 issue of the Notices of the American Mathematical Society and
can be read online at ÐÒÑÓÑÕÔ<ÖØ×Ó×ÚÙÛÙÓÙ<Ü]ݵÞß:ÜáàÕâÕãä×ÚåÒàÕÑçæÛèÚéêßÕ× .
122 10. APPLICATIONS TO TOPOLOGY
Exercises
10.1. Write a formula for the map φ figuring in the proof of Brouwer’s fixed point
theorem and prove that it is smooth.
10.2. Let x0 be any point in Rn . By analogy with the radial contraction onto the origin,
write a formula for radial contraction onto the point x 0 . Deduce that any open or closed
ball centred at x0 is contractible.
10.3. A subset M of Rn is star-shaped relative to a point x 0 ë M if for all x ë M the
straight line segment joining x 0 to x is entirely contained in M. Show that if M is star-
shaped relative to x 0 , then it is contractible onto x 0 . Give an example of a contractible set
that is not star-shaped.
10.4. A subset M of Rn is convex if for all x and y in M the straight line segment joining
x to y is entirely contained in M. Prove the following assertions.
(i) M is convex if and only if it is star-shaped relative to each of its points. Give an
example of a star-shaped set that is not convex.
(ii) The closed ball B ì ε, x í of radius ε centred at x is convex.
(iii) Same for the open ball B î¾ì ε, x í .
10.5. Let x ë Rn and let cx be the straight line connecting the origin to x. Let α be a
1-form on Rn and let β be the function defined in Example 10.18. Show that β ì x íï?ð cx α .
10.6. Let α be the k-form f dx I ï f dxi1 dxi2 on Rn and let φ : Rn óõô 0, 1 ö÷ Rn
ñòñòñ dxik
be the radial contraction φ ì x, t íï tx. Verify that
k 1
κφ ø α ï ∑ ìáú 1 í m û 1 êü ý f ì tx í tk þ 1
dt ÿ xim dxi1 dxi2 ñòñòñdxim ñòñ¡ñ dxik ,
m 1 ù 0
and check directly that dκφ ø α κ dφ ø α ï α for k 1.
10.7. Let α ï f dx dy g dz dx h dy dz be a 2-form on R3 and let φ ì x, y, z, t í ï
t ì x, y, z í be the radial contraction of R3 onto the origin. Verify that
1 1
κφ ø α ï ü ý f ì tx, ty, tz í t dt ÿ"ì x dy ú y dx í ü ý g ì tx, ty, tz í t dt ÿ ì z dx ú x dz í
0 0
üÒý 1
h ì tx, ty, tz í t dt ÿ"ì y dz ú z dy í .
0
10.8. Let α ï ∑ I f I dx I be a closed k-form whose coefficients f I are smooth functions
defined on Rn ú 0 that are all homogeneous of the same degree p ï ú k. Let
k
ì]ú 1 í l û
1
ï
k∑∑
1
β xil f I dxi1 dxi2 ññòñdxil ñ¡ñòñ dxik .
p ù
I l 1
Show that dβ ï α . (Use dα ï 0 and apply the identity proved in Exercise B.5 to each f I ; see
also Exercise 2.7.)
10.9. Let M and N be manifolds and φ 0 , φ1 : M ÷ N homotopic maps. Show that
ð φ
c 0ø α
ï ð φ
c 1ø α for all closed k-chains c in M and all closed k-forms α on N.
10.10. Prove that any two maps φ 0 and φ1 from M to N are homotopic if M or N is
contractible. (First show that every map M ÷ N is homotopic to a constant map φ ì x í ï
y0 .)
10.11. Let x0 ï ì 2, 0 í and let M be the twice-punctured plane R 2 ú 0, x0 . Let c1 , c2 ,
c3 : ô 0, 2π ö÷
M be the loops defined by c 1 ì t í ï ì cos t, sin t í , c 2 ì t í ï)ì 2 cos t, sin t í and
c3 ì t í ï ì 1
2 cos t, 2 sin t í . Show that c 1 , c2 and c3 are not homotopic. (Construct a 1-form
α on M such that the integrals ð c1 α , ð c2 α and ð c3 α are distinct.)
EXERCISES 123
10.12. A function g : R R is 2π -periodic if g x 2π g x for all x.
(i) Let g : R R be a smooth 2π -periodic function and let β g dt, where t is
the coordinate on R. Prove that there is a unique number k such that β k dt
dh for some smooth 2π -periodic function h. (To find k, integrate the equation
β k dt dh over 0, 2π . Then check that this value of k works.)
(ii) Let α be any 1-form on the unit circle S 1 and let µ be the element of arc length
of S1 . (You can think of µ as the restriction to S 1 of the angle form.) Prove that
there is a unique number k such that α kµ is exact. (Use the parametrization
c t cos t, sin t and apply the result of part (i).)
APPENDIX A
Sets and functions
A.1. Glossary
We start with a list of set-theoretical notations that are frequently used in the
text. Let X and Y be sets.
x X: x is an element of X.
a, b, c : the set containing the elements a, b and c.
X Y: X is a subset of Y, i.e. every element of X is an element of Y.
X Y: the intersection of X and Y. This is defined as the set of all x such
that x X and x Y.
X Y: the union of X and Y. This is defined as the set of all x such that
x X or x Y.
X Y: the complement of Y in X. This is defined as the set of x in X such
that x is not in Y.
X Y: the Cartesian product of X and Y. This is by definition the set of
all ordered pairs x, y with x X and y Y. Examples: R R is
the Euclidean plane, usually written R2 ; S1 0, 1 ! is a cylinder wall of
height 1; S1 S1 is a torus.
R2
S1 "$# 0, 1 % S1 " S1
x X & P x : the set of all x X which have the property P x . Exam-
ples:
x R & 1' x( 3 is the interval 1, 3 ,
x & x X and x Y is the intersection X Y,
x & x X or x Y is the union X Y,
x X & x ) Y is the complement X Y.
f : X * Y: f is a function (also called a map) from X to Y. This means that
f assigns to each x X a unique element f x + Y. The set X is called
the domain or source of f , and Y is called the codomain or target of f .
125
126 A. SETS AND FUNCTIONS
f , A - : the image of a A under the map f . If A is a subset of X, then its image
under f is by definition the set
f , A -/.10 y 2 Y 3 y . f , x - for some x 2 A 4 .
f5 1,B - : the preimage of B under the map f . If B is a subset of Y, this is by
definition the set
1
f5 , B -6.70 x 2 X 3 f , x -82 B 4 .
(This is a somewhat confusing notation. It is not meant to imply that f is
required to have an inverse.)
f 5 1 , c - : an abbreviation for f 5 1 ,0 c 4- , i.e. the set 0 x 2 X 3 f , x -9. c 4 . This
is often called the fibre or level set of f at c.
f 3 A: the restriction of f to A. If A is a subset of X, f 3 A is the function defined
by
<
f , x- if x 2 A,
, f 3 A -:, x -;.
not defined if x 2 = A.
In other words, f 3 A is equal to f on A, but “forgets” the values of f at
points outside A.
g > f : the composition of f and g. If f : X ? Y and g : Y ? Z are functions,
then g > f : X ? Z is defined by , g > f , x -@. g , f , x -A- . We often say that
the function g > f is obtained by “substituting y . f , x - into g , y - ”.
A function f : X ? Y is injective or one-to-one if x 1 . = x2 implies f , x1 -B. = f , x2 - .
(Equivalently, f is injective if f , x 1 -8. f , x2 - implies x1 . x2 .) It is called surjective
or onto if f , X - . Y, i.e. if y 2 Y then y . f , x - for some x 2 X. It is called
bijective if it is both injective and surjective. The function f is bijective if and only
if it has a two-sided inverse f 5 1 : Y ? X satisfying f 5 1 , f , x -A-B. x for all x 2 X
and f , f 5 1 , y -A-6. y for all y 2 Y.
If X is a finite set and f : X ? R a real-valued function, the sum of all the
numbers f , x - , where x ranges through X, is denoted by ∑ x C X f , x - . The set X is
called the index set for the sum. This notation is often abbreviated or abused in
various ways. For instance, if X is the collection 0 1, 2, . . . , n 4 , one uses the familiar
notation ∑niD 1 f , i - . In these notes we will often deal with indices which are pairs
or k-tuples of integers, also known as multi-indices. As a simple example, let n be
a fixed nonnegative integer, let X be the set of all pairs of integers , i, j - satisfying
0 E i E j E n, and let f , i, j -6. i F j. For n . 3 we can display X and f in a tableau
as follows.
j
3 4 5 6
2 3 4
1 2
0 i
EXERCISES 127
The sum ∑ x G X f H x I of all these numbers is written as
∑ HiK jI .
0J iJ jJ n
You will be asked to evaluate it explicitly in Exercise A.2.
A.2. General topology of Euclidean space
Let x be a point in Euclidean space Rn . The open ball of radius ε about a point
x is the collection of all points y whose distance to x is less than ε,
B LMH ε, x I6NPO y Q Rn RTS y U x SWV ε X .
x ε
A subset O of Rn is open if for every x Q O there exists an ε Y 0 such that B L H ε, x I
is contained in O. Intuitively this means that at every point in O there is a little bit
of room inside O to move around in any direction you like. An open neighbourhood
of x is any open set containing x.
A subset C of Rn is closed if its complement Rn U C is open. This definition
is equivalent to the following: C is closed if and only if for every sequence of
points x1 , x2 , . . . , xn , . . . that converges to a point x in Rn , the limit x is contained
in C. Loosely speaking, closed means “closed under taking limits”. An example
of a closed set is the closed ball of radius ε about a point x, which is defined as the
collection of all points y whose distance to x is less than or equal to ε,
B H ε, x I/N7O y Q Rn RTS y U x SBZ ε X .
x ε
Closed is not the opposite of open! There exist lots of subsets of R n that are
neither open nor closed, for example the interval [ 0, 1 I in R. (On the other hand,
there are not so many subsets that are both open and closed, namely just the empty
set and Rn itself.)
A subset A of Rn is bounded if there exists some R Y 0 such that S x S Z R
for all x in A. (That is, A is contained in the ball B H R, 0 I for some value of R.) A
compact subset of Rn is one that is both closed and bounded. The importance of the
notion of compactness, as far as these notes are concerned, is that the integral of
a continuous function over a compact subset of Rn is always a well-defined, finite
number.
Exercises
A.1. Parts (iii) and (iv) of this problem require the use of an atlas (or the Web; see e.g.
\^]_]a`bdc_cAegfa]ghjiAegf_kjfj]^kjf:lnmpo^iaq
). Let X be the surface of the earth, let Y be the real line and let
128 A. SETS AND FUNCTIONS
f : X r Y be the function which assigns to each x s X its geographical latitude measured
in degrees.
(i) Find f t X u .
(ii) Find f v 1 t 0 u , f v 1 t 90 u , f v 1 txw 90 u .
(iii) Let A be the contiguous United States. Find f t A u . Round the numbers to whole
degrees.
(iv) Let B y f t A u , where A is as in part (iii). Find (a) a country other than A that
is contained in f v 1 t B u ; (b) a country that intersects f v 1 t B u but is not contained
in f v 1 t B u ; and (c) a country in the northern hemisphere that does not intersect
f v 1 t Bu .
A.2. Let S t n uzy ∑0 { i { j { n t i | j u . Prove the following assertions.
(i) S t 0 uy 0 and S t n | 1 uy S t n u}| 32 t n | 1 uAt n | 2 u .
1
(ii) S t n u y 2 n t n | 1 u~t n | 2 u . (Use induction on n.)
A.3. Prove that the open ball B :t ε, x u is open. (This is not a tautology! State your
reasons as precisely as you can, using the definition of openness stated in the text. You will
need the triangle inequality y w x y w z | z w x .)
A.4. Prove that the closed ball is B t ε, x u is closed. (Same comments as for Exercise A.3.)
A.5. Show that the two definitions of closedness given in the text are equivalent.
A.6. Complete the following table. Here S n v 1 denotes the unit sphere about the origin
in Rn ,
that is the set of vectors of length 1.
closed? bounded? compact?
w 3, 5 yes yes yes
w 3, 5 u
w 3, u
txw 3, u
B t ε, x u
B _t ε, x u
Sn v 1
xy-plane in R3
unit cube 0, 1 n
APPENDIX B
Calculus review
This appendix is a brief review of some single- and multi-variable calculus
needed in the study of manifolds. References for this material are [Edw94], [HH02]
and [MT03].
B.1. The fundamental theorem of calculus
Suppose that F is a differentiable function of a single variable x and that the
derivative f F is continuous. Let a, b be an interval contained in the domain
of F. The fundamental theorem of calculus says that
b
f t dt F b F a . (B.1)
a
There are two useful alternative ways of writing this theorem. Replacing b with x
and differentiating with respect to x we find
x
d
f t dt f x . (B.2)
dx a
Writing g instead of F and g instead of f and adding g a to both sides in formula
(B.1) we get
x
g x 6 g a z g t dt. (B.3)
a
Formulas (B.1)–(B.3) are equivalent, but they emphasize different aspects of the
fundamental theorem of calculus. Formula (B.1) is a formula for a definite integral:
it tells you how to find the (signed) surface area between the graph of the function
f and the x-axis. Formula (B.2) says that the integral of a continuous function is
a differentiable function of the upper limit; and the derivative is the integrand.
Formula (B.3) is an “integral formula”, which expresses the function g in terms of
the value g a and the derivative g . (See Exercise B.1 for an application.)
B.2. Derivatives
Let φ1 , φ2 , . . . , φm be functions of n variables x 1 , x2 , . . . , xn . As usual we write
x1 φ1 x
x2 φ2 x
x . , φ x / .. ,
.. .
xn φm x
and view φ x as a single map from Rn to Rm . (In calculus the word “map” is often
used for vector-valued functions, while the word “function” is generally reserved
129
130 B. CALCULUS REVIEW
for real-valued functions.) We say that φ is continuously differentiable if the partial
derivatives
∂φi φi x he j φi x
x lim (B.4)
∂x j h 0 h
are well-defined and continuous functions of x for all i 1, 2, . . . , n and j 1,
2, . . . , m. Here ¡ £¤ ¡ £¤ ¡ £¤
¡ 1 ¤ ¡ 0 ¤ ¡ 0 ¤
¡ ¤ ¡ ¤ ¡ ¤
¡ 0 ¤ ¡ 1 ¤ ¡ 0 ¤
¡ ¤ ¡ ¤ ¡ ¤
¡ 0 ¤ ¡ 0 ¤ ¡ 0 ¤
¡¢ ¤ ¡¢ ¤ ¡¢ ¤
e1 .. , e2 .. , . . . , en ..
. . .
0¥ 0¥ 0¥
0 0 1
n
are the standard basis vectors of R . The (total) derivative or Jacobi matrix of φ at x
is then the m ¦ n-matrix
¡ ∂φ £¤
¡¢ 1
x . . . ∂φ 1
x ¤
∂x 1 ∂x n
. .
Dφ x .. .. ¥ .
∂φ m ∂φ m
∂x 1 x . . . ∂xn x
If v is any vector in Rn , the directional derivative of φ along v is defined to be the
vector Dφ x v in Rm , obtained by multiplying the matrix Dφ x by the vector v.
For n 1 φ is a vector-valued function of one variable x, often called a path
or (parametrized) curve in Rm . In this case the matrix Dφ x consists of a single
column vector, called the velocity vector, and is usually denoted simply by φ § x .
For m 1 φ is a scalar-valued function of n variables and Dφ x is a single
row vector. The transpose matrix of Dφ x is therefore a column vector, usually
called the gradient of φ:
T
Dφ x grad φ x .
The directional derivative of φ along v can then be written as an inner product,
Dφ x v grad φ x ;¨ v. There is an important characterization of the gradient,
which is based on the identity a ¨ b ª© a ©© b © cos θ. Here 0 « θ « π is the angle
subtended by a and b. If v is a unit vector ( © v ©9 1), then
Dφ x v grad φ x ¬¨ v © grad φ x ^© cos θ ,
where θ is the angle between grad φ x and v. So Dφ x v takes on its maximal
value if cos θ 1, i.e. θ 0. This means that v points in the same direction as
grad φ x . Thus the direction of the vector grad φ x is the direction of steepest
ascent, i.e. in which φ increases fastest, and the magnitude
of grad φ x is equal
to the directional derivative Dφ x v, where v is the unit vector pointing along
grad φ x .
Frequently a function is not defined on all of Rn , but only on a subset U. We
must be a little careful in defining the derivative of such a function. Let us assume
that U is an open set. Let φ : U ® Rm be a function defined on U and let x ¯ U.
Because U is open, there exists ε ° 0 such that the points x te j are contained
in U for ε ± t ± ε. Therefore φ i x te j is well-defined for ε ± t ± ε and
thus it makes sense to ask whether the partial derivatives (B.4) exist. If they do,
for all x ¯ U and all i and j, and if they are continuous, the function φ is called
continuously differentiable or C 1 .
B.3. THE CHAIN RULE 131
If the second partial derivatives
∂ 2φ i
x³
∂x j ∂xk ²
exist and are continuous for all x ´ U and for all i µ 1, 2, . . . , n and j, k µ 1, 2, . . . ,
m, then φ is called twice continuously differentiable or C 2 . Likewise, if all r-fold
partial derivatives
∂rφ i
x³
∂x j1 ∂x j2 ¶:¶:¶ ∂x jr ²
exist and are continuous, then φ is r times continuously differentiable or C r . If φ is Cr
for all r · 1, then we say that φ is infinitely many times differentiable, C ¸ , or smooth.
This means that φ can be differentiated arbitrarily many times with respect to any
of the variables.
Let us now review some of the most important facts concerning derivatives.
B.3. The chain rule
Recall that if A, B and C are sets and φ : A ¹ B and ψ : B ¹ C are functions,
we can apply ψ after φ to obtain the composite function ψ º φ ³ x ³/µ ψ φ x ³³ .
² ² ² ²
B.1. T HEOREM (chain rule). Let U » Rn and V » Rm be open and let φ : U ¹ V
and ψ : V ¹ Rk be Cr . Then ψ º φ is Cr and
D ψ º φ ³ x ³/µ Dψ φ x ³A³ Dφ x ³
² ² ² ² ²
for all x ´ U.
Here Dψ φ x ³A³ Dφ x³ denotes the composition or the product of the two ma-
trices Dψ φ x² ³³ ² and Dφ² x ³ .
² ² ²
B.2. E XAMPLE . In the one-variable case n µ m µ k µ 1 the derivatives Dφ and
Dψ are 1 ¼ 1-matrices φ ½ x ³A³ and ψ ½ y ³A³ , and matrix multiplication is ordinary
multiplication, so we get ² the² usual chain
² ² rule
ψ º φ ³ ½ x ³/µ ψ ½ φ x ³A³ φ ½ x ³ .
² ² ² ² ²
B.3. E XAMPLE . If n µ k µ 1, then ψ º φ is a real-valued function of one vari-
able x, so D ψ º φ ³ is a 1 ¼ 1-matrix containing the single entry ψ º φ ³ ½ . Moreover,
² ÂÃ ²
dφ 1
dx x ³
Dφ x ³;µ¿¾ÀÁ .. ² Dψ y ³µÆÅ ∂y ∂ψ ∂ψ
y ³ . . . ∂y y ³Ç ,
. Ä and
² dφ m ² 1
² m
²
dx x ³
²
so by the chain rule
m
d ψ º φ³ ∂ψ dφi
² dx
²
x ³/µ Dψ φ x ³A³ Dφ x ³µ
² ² ²
∑ ∂yi ² φ ² x ³A³ dx ²
x³ . (B.5)
iÈ 1
This is perhaps the most important special case of the chain rule. Sometimes we
are sloppy and abbreviate this identity to
m
d ψ º φ³ ∂ψ dφi
² dx µ ∑ ∂yi dx
.
iÈ 1
132 B. CALCULUS REVIEW
An even sloppier, but nevertheless quite common, notation is
m
dψ ∂ψ dφi
dx É ∑ ∂yi dx
.
iÊ 1
In these notes we frequently use the so-called “pullback” notation. Instead of ψ Ë φ
we often write φ Ì ψ, so that φ Ì ψ Í x Î stands for ψ Í φ Í x ÎAÎ . Similarly, φ ÌÍ ∂ψ Ï ∂yi Î_Í x Î
stands for ∂ψ Ï ∂y i Í φ Í x ÎAÎ . In this notation we have
m
dφ Ì ψ ∂ψ dφi
dx É ∑ φ ̬Р∂yi Ñ dx
. (B.6)
iÊ 1
B.4. The implicit function theorem
Let φ : W Ò Rmbe a continuously differentiable function defined on an open
subset W of Rn Ó m . Let us think of a vector in Rn Ó m as an ordered pair of vectors
Í u, v Î with u Ô Rn and v Ô Rm . Consider the equation
φ Í u, v Î
0.
É
Under what circumstances is it possible to solve for v as a function of u? The
answer is given by the implicit function theorem. We form the Jacobi matrices of
φ with respect to the u- and v-variables separately,
∂φ 1 ∂φ 1 Ù
ØÙ ∂φ 1
ØÙ
∂φ 1 Ù
∂u 1 . . . ∂u n ∂v 1 . . . ∂v m
.. .. , .. ..
Duφ
É¿ÕÖ . . Ú D v φ
ÉÕÖ . . Ú .
Ö× ∂φm . . . ∂φm Ö× ∂φm . . . ∂φm
∂u 1 ∂u n ∂v ∂v m 1
Observe that the matrix Dvφ is square. We are in business if we have a point
Í u0 , v0 Î at which φ is 0 and Dvφ is invertible.
B.4. T HEOREM (implicit function theorem). Let φ : W Ò Rm be Cr , where W
is open in Rn Ó m . Suppose that Í u0 , v0 ÎÛÔ W is a point such that φ Í u0 , v0 Î 0 and
Dvφ Í u0 , v0 Î is invertible. Then there are open neighbourhoods U Ü R n of u0 and
É
V Ü Rm of v0 such that for each u Ô U there exists a unique v f Í u Î Ô V satis-
fying φ Í u, f Í u ÎAÎ
É
0. The function f : U Ò V is C r with derivative given by implicit
differentiation:
É
1
D f Í uÎ Dvφ Í u, v ÎAÞ Duφ Í u, v Îàß v Ê f á uâ
É7Ý ß
for all u Ô U.
This is well-known for m n 1, when φ is a function of two real variables
Í u, v Î . If ∂φ Ï ∂v ã 0 at a certainÉ point
É Í u , v Î , then for u close to u and v close to
0 0 0
É
v0 we can solve the equation φ Í u, v Î 0 for v as a function v f Í u Î of u, and
É É
∂φ Ï ∂u
fä .
ÉPÝ ∂φ Ï ∂v
Now let us take φ to be of the form φ Í u, v Î g Í vÎ u, where g : W Ò R n is a
n
given function with W open in R . Solving φ Í u, v Î
É Ý
0 here amounts to inverting
the function g. Moreover, Dvφ
É
Dg, so the implicit function theorem yields the
following result.
É
B.5. THE SUBSTITUTION FORMULA FOR INTEGRALS 133
B.5. T HEOREM (inverse function theorem). Let g : W å R n be continuously
n
differentiable, where W is open in R . Suppose that v0 æ W is a point such that Dg ç v0 è
is invertible. Then there is an open neighbourhood U é R n of v0 such that g ç U è is an
open neighbourhood of g ç v0 è and the function g : U å g ç U è is invertible. The inverse
g ê 1 : V å U is continuously differentiable with derivative given by
1 1ì
Dg ê ç u èë Dg ç v è ê ì ví gî 1 ï uð
for all v æ V.
Again let us spell out the one-variable case n ë 1. Invertibility of Dg ç v 0 è
simply means that g ñòç v0 èôë ó 0. This implies that near v 0 the function g is strictly
monotone increasing (if g ñdç v0 èöõ 0) or decreasing (if g ñdç v0 èö÷ 0). Therefore if I is
a sufficiently small open interval around u 0 , then g ç I è is an open interval around
g ç u0 è and the restricted function g : I å g ç I è is invertible. The inverse function
has derivative
1
ç g ê 1 è ñ ç u è/ë ,
gñ ç v è
with v ë gê 1ç uè .
B.6. E XAMPLE (square roots). Let g ç v èë v 2 . Then g ñ ç v0 èøë ó 0 whenever v0 ë ó
0. For v0 õ 0 we can take I ë ç 0, ù è . Then g ç I èúë ç 0, ù è , g ê 1 ç u èûëýü u, and
ç g ê 1 è ñòç u è9ë 1 þnç 2 ü u è . For v0 ÷ 0 we can take I ë ç~ÿWù , 0 è . Then g ç I è ë ç 0, ù è ,
g ê 1 ç u è ë ÿ ü u, and ç g ê 1 è ñ ç u è ë ÿ 1 þnç 2 ü u è . In a neighbourhood of 0 it is not
possible to invert g.
B.5. The substitution formula for integrals
Let V be an open subset of Rn and let f : V å R be a function. Suppose we
want to change the variables in the integral V f ç y è dy. (This is shorthand for an
n-fold integral over y 1 , y2 , . . . , yn .) This means we substitute y ë p ç x è , where
p : U å V is a map from an open U é Rn to V. Under a suitable hypothesis we
can change the integral over y to an integral over x.
B.7. T HEOREM (change of variables formula). Let U and V be open subsets of R n
and let p : U å V be a map. Suppose that p is bijective and that p and its inverse are
continuously differentiable. Then for any integrable function f we have
f ç y è dy ë f ç p ç xèAè det Dp ç xè dx.
V U
Again this should look familiar from one-variable calculus: if p : ç a, b è å ç c, d è
is C 1 and has a C 1 inverse, then
d b
f ç p ç x èAè p ñ ç x è dx if p is increasing,
f ç y è dy ë ab
c ÿ a f ç p ç x èAè p ñ ç x è dx if p is decreasing.
b
This can be written succinctly as cd f ç y è dy ë a f ç p ç x èè p ñ ç x è dx, which looks
more similar to the multidimensional case.
134 B. CALCULUS REVIEW
Exercises
B.1. Let g : a, b R be a C n 1-function,
where n 0. Suppose a x b and put
h x a.
(i) By changing variables in the fundamental theorem of calculus (B.3) show that
1
g x g a h g a th dt.
0
(ii) Show that
1 1
g x g a h t 1 g a th 0 h2 1 t g a th dt
0
1
g a hg a h2 1 t g a th dt.
0
(Integrate the formula in part (i) by parts and don’t forget to use the chain rule.)
(iii) By induction on n deduce from part (ii) that
n
g k a k hn 1 1
g x ∑ k!
h
n!
1 t n g n 1
a th dt.
k 0 0
This is Taylor’s formula with integral remainder term.
B.2. Let x and v be constant vectors in Rn . Define c t x tv. Find c t .
φ x tv " φ x
B.3. Deduce from the chain rule that Dφ x v lim .
t! 0 t
B.4. According to Newton’s law of gravitation, a particle of mass m 1 placed at the
origin in R3 exerts a force on a particle of mass m 2 placed at x # R3 %$ 0 & equal to
Gm1 m2
F ' ( ( 3 x,
x
( (
where G is a constant of nature. Show that F is the gradient of f x Gm 1 m2 ) x .
B.5. A function f : Rn
*$ 0 &+ R is homogeneous of degree p if f tx , tp f x for all
x # Rn -$ 0 & and t . 0. Here p is a real constant.
(i) Show that the functions f x, y /0 x 2 xy ) x2 y2 , f x, y /01 x3 y3 ,
f x, y, z 2 x2 z6 3x4 y2 z2 4365 2 are homogeneous. What are their degrees?
(ii) Assume that f is defined at 0 and continuous everywhere. Show that p 0.
Show that f is constant if p 0.
(iii) Show that if f is homogeneous of degree p and smooth, then
n ∂f
∑ xi ∂xi x p f x .
i 1
(Differentiate the relation f tx t p f x with respect to t.)
R by f 0 0 and f x e 3 1 7 x for x 8 0.
2
B.6. Define a function f : R
(i) Show that f is differentiable at 0 and that f 0 0.
(ii) Show that f is smooth and that f n 0 0 for all n.
(iii) Plot the function f over the interval 5 x 5. Using software or a graphing
calculator is fine, but pay special attention to the behaviour near x 0.
B.7. Define a map ψ from Rn 3 1 to Rn by
1 ( ( 2
ψ t ( ( 2t : t 1 en ; .
t 2 19
(i) Show that ψ t lies on the unit sphere S n 3 1 about the origin.
EXERCISES 135
(ii) Show that ψ < t = is the intersection point of the sphere and the line through the
points en and t. (Here we regard t >?< t 1 , t2 , . . . , tn @ 1 = as a point in Rn by identi-
fying it with < t 1 , t2 , . . . , tn @ 1 , 0 = .)
(iii) Compute Dψ < t = .
(iv) Let X be the sphere punctured at the “north pole”, X > S n @ 1 ACB en D . Stereo-
graphic projection from the north pole is the map φ : X E Rn @ 1 given by φ < x =F>
< xn A 1 = @ 1 < x1 , x2 , . . . , xn @ 1 = . Show that φ is a two-sided inverse of ψ.
(v) Draw diagrams illustrating the maps φ and ψ for n > 2 and n > 3.
(vi) Now let y be any point on the sphere and let P the hyperplane which passes
through the origin and is perpendicular to y. The stereographic projection from y
of any point x in the sphere distinct from y is defined as the unique intersec-
tion point of the line joining y to x and the hyperplane P. This defines a map
φ : Sn @ 1 ACB y D E P. The point y is called the centre of the projection. Write a
formula for the stereographic projection φ from the south pole A en and for its
inverse ψ : Rn @ 1 E Sn @ 1 .
B.8. A map φ : Rn E Rm is called even if φ < A x =G> φ < x = for all x in Rn . Find Dφ < 0 = if
φ is even and C 1 .
B.9. Let a0 , a1 , a2 , . . . , an be vectors in Rn . A linear combination ∑niH 0 ci ai is convex if
∑niH 0 ci
> 1. The simplex I spanned by the ai ’s is the collection of all their convex linear
combinations,
n n
IC>KJ ∑ ci ai LL ∑ ci > 1M .
iH 0 L iH 0
L
The standard simplex in Rn is the simplex spanned by the vectors 0, e1 , e2 , . . . , en .
(i) For n > 1, 2, 3 draw pictures of the standard n-simplex as well as a nonstandard
n-simplex.
(ii) The volume of a region R in Rn is defined as N dx1 dx2 OPOPO dxn . Show that
R
1
vol IC>
det A ,
n! Q Q
where A is the n R n-matrix with columns a1 A a0 , a2 A a0 , . . . , an A a0 . (First
compute the volume of the standard simplex by repeated integration. Then map
I to the standard simplex by an appropriate substitution and apply the substi-
tution formula for integrals.)
The following two calculus problems are not review problems, but the results are
needed in Chapter 9.
B.10. For x S 0 define T
< x => NVU e @ t t x @ 1
dt
0
T followingT assertions.
and prove the
(i) T < x W 1 = > x < x = for all x S 0.
(ii) < n = >X< n A 1 = ! for TZ
positive
Y integers n.
@ u2 a 1 aW 1
(iii) NVU e u du > .
0 T 2 2 T [
T defined in Exercise B.10) by establish-
B.11. Calculate < n W 12 = (where is the function
ing the following identities. For brevity write γ > < 12 = .
N U e @ s ds.
2
(i) γ >
@
(ii) γ 2 > N U U N U e @ x2 @ y2
dx dy.
@ @
U U
136 B. CALCULUS REVIEW
2π
(iii) γ 2 \^] ]`_ re a r2
dr dθ.
0 0
(iv) γ \^b π .
1 \ 1 h 3 h 5 hPhPhji 2n k 1 l b
(v) ced n f π for n m 1.
2g 2n
Bibliography
[Bac06] D. Bachman, A geometric approach to differential forms, Birkhäuser, Boston, MA, 2006.
A recent text, a version of which is available through the author’s website
nPojoqpr sjstpPuPvxw4vqyz{p}|~ojuPqzqyqsPtyqvxwn4v4s .
[BS91] P. Bamberg and S. Sternberg, A course in mathematics for students of physics, Cambridge Uni-
versity Press, Cambridge, 1991.
[BT82] R. Bott and L. Tu, Differential forms in algebraic topology, Springer-Verlag, New York, 1982.
Masterly exposition at beginning graduate level of the uses of differential forms in topology
and de Rham cohomology.
[Bre91] D. Bressoud, Second year calculus, Undergraduate Texts in Mathematics, Springer-Verlag, New
York, 1991.
Multivariable calculus from the point of view of Newton’s law and special relativity with
coverage of differential forms.
[Bre05] O. Bretscher, Linear algebra with applications, third ed., Pearson Prentice Hall, Upper Saddle
River, New Jersey, 2005.
Good reference for the basic linear algebra required in these notes.
[Car94] M. do Carmo, Differential forms and applications, Springer-Verlag, Berlin, 1994, translated from
the 1971 Portuguese original.
Slightly more advanced than these notes, with some coverage of Riemannian geometry, in-
cluding the Gauß-Bonnet theorem.
[Dar94] R. Darling, Differential forms and connections, Cambridge University Press, Cambridge, 1994.
Advanced undergraduate text on manifolds and differential forms, including expositions of
Maxwell theory and its modern generalization, gauge theory.
[Edw94] H. Edwards, Advanced calculus: A differential forms approach, Birkhäuser Boston, Boston, Mas-
sachusetts, 1994, corrected reprint of the 1969 original.
One of the earliest undergraduate textbooks covering differential forms. Still recommended
as an alternative or supplementary source.
[Fla89] H. Flanders, Differential forms with applications to the physical sciences, second ed., Dover Publi-
cations, New York, 1989.
Written for 1960s engineering graduate students. Concise but lucid (and cheap!).
[Gra98] A. Gray, Modern differential geometry of curves and surfaces with Mathematica, second ed., CRC
Press, Boca Raton, FL, 1998.
Leisurely and thorough exposition at intermediate undergraduate level with plenty of com-
puter graphics.
[GP74] V. Guillemin and A. Pollack, Differential topology, Prentice-Hall, Englewood Cliffs, N.J., 1974.
Written for beginning graduate students, but very intuitive and with lots of interesting appli-
cations to topology.
[HH02] J. Hubbard and B. Hubbard, Vector calculus, linear algebra, and differential forms: A unified ap-
proach, second ed., Prentice Hall, Upper Saddle River, New Jersey, 2002.
[MT03] J. Marsden and A. Tromba, Vector calculus, fifth ed., W. H. Freeman, New York, 2003.
Standard multivariable calculus reference.
[Opr03] J. Oprea, Differential geometry and its applications, second ed., Prentice Hall, 2003.
Curves, surfaces, geodesics and calculus of variations with plenty of MAPLE programming.
Accessible to intermediate undergraduates.
[Sin01] S. Singer, Symmetry in mechanics. A gentle, modern introduction, Birkhäuser, Boston, MA, 2001.
[Spi65] M. Spivak, Calculus on manifolds. A modern approach to classical theorems of advanced calculus, W.
A. Benjamin, New York-Amsterdam, 1965.
Efficient and rigorous treatment of many of the topics in these notes.
137
138 BIBLIOGRAPHY
[Spi99] , A comprehensive introduction to differential geometry, third ed., Publish or Perish, Hous-
ton, TX, 1999.
Differential geometry textbook at advanced undergraduate level in five massive but fun to
read volumes.
[Wei97] S. Weintraub, Differential forms. A complement to vector calculus, Academic Press, San Diego,
CA, 1997.
Written as a companion to multivariable calculus texts. Contains careful and intuitive expla-
nations of several of the ideas covered in these notes, as well as a number of straightforward
exercises.
The Greek alphabet
upper case lower case name
A α alpha
B β beta
γ gamma
δ delta
E , ε epsilon
Z ζ zeta
H
η eta
θ, ϑ theta
I ι iota
K κ kappa
λ lambda
M µ mu
N
ν nu
ξ xi
O
o omicron
π, $ pi
P ρ rho
σ sigma
T τ tau
υ upsilon
φ, ϕ phi
X χ chi
ψ psi
ω omega
139
Notation Index
, Hodge star operator, 24 dx, infinitesimal hypersurface, 26
relativistic, 29 dx, infinitesimal displacement, 25
0,
1 k , unit cube in Rk , 58
dx I , short for dx i1 dx i2 ~ dx ik , 17
, orientation defined by a basis , 94 dx i , covector (“infinitesimal increment”), 17,
, Euclidean inner product (dot product), 8 ¤ 83
dx i , omit dx i , 18
, composition of maps, 37, 126, 131
, integral of a form
over a chain, 57 ei , i-th standard basis vector of Rn , 130
over a manifold, 106
f ¥ A, restriction of f to A, 126
V, integral
of a form over a chain, 47
, Euclidean norm (length), 8
, tensor multiplication, 89 g f , composition of f and g, 37, 126, 131
¦
, Gamma function, 110, 135
∂
, partial derivative, 130 grad, gradient of a function, 26
∂xi
, orthogonal complement, 74 graph, graph of a function, 68
, exterior multiplication, 17, 85
H n , upper halfspace in Rn , 103
A T , transpose of a matrix A, 35
I, multi-index i 1 , i 2 , . . . , i k ¡ (usually increas-
A k V, set of alternating k-multilinear functions
ing), 17
on V, 85
int M, interior of a manifold with boundary,
Aσ , permutation matrix, 43
103
A I, J , I, J-submatrix of A, 42
α , Hodge star of α , 24
ker A, kernel (nullspace) of a matrix A, 73
relativistic, 29
M α , integral of α over a manifold M, 106 l σ ¡ , length of a permutation σ , 34
c α , integral of α over a chain c, 47, 57
Alt µ , alternating form associated to µ , 90 µ M , volume form of a manifold M, 96
§n
B ε , x ¡ , closed ball in Rn , 127 k ¨ , binomial coefficient, 19, 23
n, unit normal vector field, 95
¢4 ε, x ¡ , open ball in R , 127
n
B
, orientation defined by a basis , 94 nullity A, dimension of the kernel of A, 73
O n ¡ , orthogonal group, 76
C r , r times continuously differentiable, 131 © k
curl, curl of a vector field, 27 M ¡ , vector space of k-forms on M, 19, 82
φ ª , pullback
Dφ, Jacobi matrix of φ, 130
of a form, 37, 88
∂, boundary
of a function, 37, 132
of a chain, 59
of a manifold, 103 Rn , Euclidean n-space, 1
d, exterior derivative, 20
£ rank A, dimension of the column space of A,
, Laplacian of a function, 28 73
δ I, J , Kronecker delta, 86
δ i, j , Kronecker delta, 51 S n , unit sphere about the origin in Rn « 1
, 8, 64
det A, determinant of a matrix A, 31 S n , permutation group, 33
div, divergence of a vector field, 26 sign σ ¡ , sign of a permutation σ , 34
141
142 NOTATION INDEX
SL ¬ n , special linear group, 78
Tx M, tangent space to M at x, 8, 69, 73, 74
V ® , dual of a vector space V, 83
V k , k-fold Cartesian product of a vector space
V, 85
voln , n-dimensional Euclidean volume, 91
¯ ¯
x , Euclidean norm (length) of a vector x, 8
x ° y, Euclidean inner product (dot product) of
vectors x and y, 8
x T , transpose of a vector x, 1
Index
Page numbers in boldface refer to definitions or theorems; italic
page numbers refer to examples or applications. In a few cases
italic boldface is in order.
affine space, 7, 8, 73 circle, 9, 47, 48, 51–53, 55, 62, 64, 72, 115, 118,
alternating 123
algebra, 82 closed
multilinear function, 32, 85, 87–90 ball, 8, 105, 106, 110, 115, 122, 127
property, 18, 19, 21 chain, 62, 64, 122
Ampère, André Marie (1775–1836), 29 curve, 1, 50, 53, 62
Ampère’s Law, 29 form, 22, 27–29, 40, 51, 54, 55, 117–123
angle set, 4, 36, 47, 84, 103, 106, 107, 127, 128
form, 23, 36, 47, 51, 64, 118, 120, 123 codimension, 69, 73, 74, 105
function along a curve, 52–53 cohomology, 137
anticommutativity, 18 column
antisymmetric operation, 32, 42
matrix, 78 vector, 1, 31, 45, 69, 83, 89, 92, 130
multilinear function, see alternating multi- compact, 57, 106–110, 117, 120, 127
linear function complementary, 24
arc length, 89, 96, 101, 107 configuration space, 9, 14
Archimedes of Syracuse (287–212 BC), 3, 109 connected component, 12, 119
Archimedes’ Law, 109 conservative, 49, 108, 109
atlas, 67, 69, 82 constant form, 19, 28, 83
average of a function, 107 continuously differentiable, 130
contractible, 115, 119–122
ball, see closed ball, open ball contraction, 115, 120, 121–122
barycentre, 107 contravariance, 38
bilinear, 85, 89 convex, 122
block, 31, 42, 91, 93, 96 linear combination, 135
rectangular, see rectangular block coordinate map, see chart
Bonnet, Pierre (1819–1892), 137 covariant vector, see covector
boundary covector, 83, 84, 85, 89, 90
of a chain, 60, 62–65 field, 83
of a manifold, 2–7, 103, 104–111, 118 Coxeter, Harold Scott MacDonald (1907–2003),
bounded, 47, 106, 127 43
Brouwer, Luitzen Egbertus Jan (1881–1966), 113, Coxeter relations, 43
122 critical point, 79
Brouwer’s fixed point theorem, 113, 122 cross-cap, 7
cube in an open set, 59, 60–62, 64–65
Cartan, Elie (1869–1951), 17 curl, 27, 29, 65, 109
Cartesian product, 10, 85, 116, 125 curvature, 17
Cartesius, Renatus, see Descartes, René curve, 1, 69
centroid, 107 cycle, 62, 64
chain, 59, 60–65 cylinder
chart, 67, 69, 72 formula, 116
143
144 INDEX
with base M, 116 fixed point, 113
flux, 26, 99, 108, 109
d’Alembert, Jean Le Rond (1717–1783), 29 form
d’Alembertian, 29 as a vector-eating animal, 88
de Rham, Georges (1903–1990), 137 closed, see closed form
degenerate chain, 61, 64 exact, see exact form
degree on a manifold, 82, 88
of a form, 17 on Euclidean space, 17, 18–30, 88
of a homogeneous function, 134 volume, see volume form
of a multi-index, 17 free space, 29
degrees of freedom, 9, 14 Freedman, Michael (1951–), 121
Descartes, René (1596–1650), 10, 85, 125 function, 125, 130
determinant, 31, 32–35, 40–42, 43, 78, 85, 86, functional, see covector
91–94 fundamental theorem of calculus, 27, 47, 49,
differential 52, 64, 129, 134
equation, 15 in Rn , 49, 63, 108
form, see form
dimension, 1–11, 69 Gamma function, 110–111, 135
Dirichlet, Lejeune (1805–1859), 110 Gauß, Carl Friedrich (1777–1855), v, 29, 63, 65,
Dirichlet integral, 110 95, 108, 137
disconnected, 12, 108 Gauß map, 95
divergence, 26, 29, 65, 108, 109 Gauß’ Law, 29
domain, 106, 108, 109 graded commutativity, 18, 19
of a function, 125 gradient, 26, 28, 49, 65, 74–79, 96, 100, 101, 108,
dot product, see inner product 110, 130
dual Gram, Jorgen (1850–1916), 92
basis, 84, 86, 87, 89 Gram-Schmidt process, 92
vector, see covector graph, 68, 70, 73, 97, 100, 103, 129
space, 83 Graßmann, Hermann (1809–1877), 82
gravitation, 54, 134
electromagnetic wave, 29 Greek alphabet, 139
electromagnetism, 29 Green, George (1793–1841), v, 63, 65, 110
element Green’s theorem, 65, 110
of arc length, 89, 96, 101, 123
of surface area, 96 halfspace, 103
embedding, 67, 68, 70, 71, 77–78, 81, 88, 96–98, Hodge, William (1903–1975), 23, 25, 28, 29, 38
100, 101, 103, 104, 106, 107 Hodge star operator, 23, 25, 28, 38, see also rel-
Euclid of Alexandria (ca. 325–265 BC), 1, 67, ativity
69, 91, 110, 114, 125, 127 homogeneous function, 28, 78, 122, 134
Euclidean homotopy, 114
motion, 91 formula, 117
plane, 125 of curves, 115
space, 1, 67, 69, 110, 114, 127 of loops, 115, 119, 121
volume, 91 hypersurface, 26, 69, 95–101, 106, 108, 120
εύρηκα , 109
even increasing multi-index, 19, 24, 28, 41, 86, 87,
map, 135 see also complementary
permutation, 34, 43 index of a vector field, 55
exact form, 22, 28, 49–52, 64, 117–120, 123 inner product, 8, 84, 85, 98, 130
exterior of forms, 28
algebra, 82 integrability condition, 22
derivative, 20, 21, 25, 27, 60, 63 integral
on a manifold, 82 of a 1-form over a curve, 47, 48, 49, 51
differential calculus, 17 of a form
product, see product of forms over a chain, 57, 58–59, 63–65, 106
over a manifold, 106, 107, 108, 117, 119,
Faraday, Michael (1791–1867), 29 120, 122
Faraday’s Law, 29 inversion, 34
fibre, see level set inward pointing, 106
INDEX 145
k-chain, see chain odd permutation, 34, 43
k-cube, see cube open
k-form, see form ball, 127
k-multilinear function, see multilinear function neighbourhood, 127
Klein, Felix (1849–1925), 5 set, 127, 128
Klein bottle, 5 is a manifold, 70
Kronecker, Leopold (1823–1891), 51 orientation
Kronecker delta, 51 of a boundary, 106, 108
of a hypersurface, 95, 100
Laplace, Pierre-Simon (1749–1827), 28 of a manifold, 88, 95, 108
Laplacian, 28 of a vector space, 91, 94, 100
Leibniz, Gottfried Wilhelm von (1646–1716), 20, preserving, 48, 57, 95–97, 101, 106, 107
21, 40 reversing, 48, 57, 95
Leibniz rule orthogonal
for forms, 21, 40 complement, 74, 95
for functions, 20 group, 76, 78
length matrix, 76, 91, 92
of a permutation, 34, 42–43 operator, 28
of a vector, 8, 76, 79, 95, 106, 114, 128 projection, 93
level orthonormal, 28, 91
curve, 74 outward pointing, 106
hypersurface, 74 outward-pointing, 110
set, 73, 126
surface, 74 pair of pants, 105
lightlike, 29 paraboloid, 4
line segment, 91 parallelepiped, see block
local representative, 82, 88, 89, 96 parallelogram, 13, 14, 91, 100
Lotka, Alfred (1880–1949), 15, 78 parametrization, 9, 57, 67
Lotka-Volterra model, 15, 78 parametrized curve, 13, 47, 49, 53–55, 77–78,
130
manifold, 1–15, 69, 70–79 partial differential
abstract, 9–13, 69 equation, 22
given explicitly, 9, 72 operator, 20
given implicitly, 7, 72 path, see parametrized curve
with boundary, 2–7, 103, 104–111, 118 pentagon, 14
map, 125, 130 Perelman, Grigori (1966–), 121
Maxwell, James Clerk (1831–1879), 29 periodic function, 123
mean of a function, 107 permutation, 33, 34, 35, 41, 42–43, 85, 100
measurable set, 36 group, 33, 43
Milnor, John (1931–), 121 matrix, 43
minimum, 75 pinch point, 7
Minkowski, Hermann (1864–1909), 29 plane curve, 1, 13, 69
Minkowski Poincaré, Jules Henri (1854–1912), 119, 121
inner product, 29 Poincaré
space, 29 conjecture, 121
Möbius, August (1790–1868), 4, 105 lemma, 119
Möbius band, 4, 105 potential, 49, 53, 54, 108
multi-index, 17, 126, see also increasing multi- predator, 15
index prey, 15
multilinear product
algebra, 17 of forms, 19, 27
function, 85, 89 on a manifold, 82
of permutations, 34, 43
n-manifold, see manifold of sets, see Cartesian product
naturality of pullbacks, 38, 48, 58 product rule, see Leibniz rule
Newton, Isaac (1643–1727), 54, 134, 137 projective plane, 5
norm of a vector, 8 pullback
normal vector field, see unit normal vector field of a form
146 INDEX
on a manifold, 88, 97, 106, 114, 116, 117 stereographic projection, 72, 135
on Euclidean space, 37, 40–42, 47, 48, 57, Stirling, James (1692–1770), 111
58, 82, 88 Stirling’s formula, 111
of a function, 132 Stokes, George (1819–1903), v, 47, 63, 65, 107,
punctured 109
Euclidean space, 78, 114, 119–121 Stokes’ theorem
plane, 47, 51, 64, 77, 115, 119, 120 classical version, 65, 109
for chains, 47, 63, 64
quadrilateral, 11 for manifolds, 103, 108, 116, 118, 120
submanifold, 69
rectangular block, 18, 57, 65
surface, 4, 14, 69, 81, 109
regular value, 73, 74–79, 96, 100, 105
area, 9, 17, 96, 107, 109, 129
relativity, 10, 29, 137
symmetric
reparametrization
bilinear function, 85
of a curve, 47, 48, 50, 58
matrix, 76, 79
of a rectangular block, 57, 58
restriction of a map, 47, 106, 107, 110, 126 tangent
retraction, 113 hyperplane, 69
Riemann, Bernhard (1826–1866), 137 line, 1, 13, 69, 70, 77
rigid body, 10 plane, 14, 69, 78
row space, 1, 8, 14, 69, 70, 73–76, 78, 88, 95, 96,
operation, 42 106
vector, 1, 33, 45, 74, 83, 89, 130 vector, 48, 69, 79, 88, 106
Taylor, Brook (1685–1731), 134
saddle point, 75
Taylor’s formula, 134
Schmidt, Erhard (1876–1959), 92
tensor product, 17, 85, 89
sign of a permutation, 34, 42–43
timelike, 29
simple permutation, 43
topological manifold, 4
simplex, 135
torus, 4, 68, 106
simply connected, 119, 121
trace, 78
singular
trajectory, 15, 78
cube, see cube
transformation law, 82
value, 73, 74–76, 78
transpose of a matrix or a vector, 1, 26, 35, 74,
singularity, 2, 8, 13, 14, 67, 71, 105
83, 130
Smale, Stephen (1930–), 121
transposition, 42
smooth
curve, 69 unit
function or map, 131, 134 circle, see circle
hypersurface, 69 cube, 35, 58, 65, 128
manifold, 4 interval, 35, 58, 91, 114, 116
point, 2 normal vector field, 95, 98, 99, 101, 106, 108–
surface, 69 110, 120
solution curve, 9, 15 sphere, see sphere
space curve, 1 square, 35, 60, 62
space-time, 29 vector, 51, 130
spacelike, 29
special linear group, 78 vector, see also column, length, row, unit, tan-
sphere, 4, 8, 10, 11, 14, 64, 67, 75, 79, 96, 100, gent
105, 106, 110, 113, 114, 119–121, 128, 134 field, 24, 28, 29, 48, 55, 65, 98, 108, 109, see
spherical also conservative, curl, divergence, gradi-
coordinates, 44 ent, index, potential, unit normal
pendulum, 9 Volterra, Vito (1860–1940), 15, 78
standard volume
basis of Rn , 130 change, 36, 42
orientation, 94 element, 17, 96
simplex, 135 Euclidean, see Euclidean volume
star-shaped, 122 form, 65, 96, 97, 103
state space, see configuration space of a hypersurface, 99
steepest ascent, 75, 130 on Rn , 18, 24, 42
INDEX 147
on a hypersurface, 100, 110, 120
of a block, 18, 31, 91, 92, 93
of a manifold, 89, 107, 109
of a simplex, 135
wave operator, 29
wedge product, 17, 85, 89
winding number
of closed curve, 53, 54–55, 118, 120
of hypersurface, 120
work, 17, 25, 48, 49, 61, 108, 109
zero of a vector field, 25
You might also like
- Geometric Algebra For Physicists - Errata PDFNo ratings yetGeometric Algebra For Physicists - Errata PDF6 pages
- Stewart J. Advanced General Relativity (CUP, 1991) (T)No ratings yetStewart J. Advanced General Relativity (CUP, 1991) (T)118 pages
- 5059.minimal Surfaces (Courant Lecture Notes in Mathematics) by Tobias H. ColdingNo ratings yet5059.minimal Surfaces (Courant Lecture Notes in Mathematics) by Tobias H. Colding128 pages
- Glimpses of Soliton Theory: The Algebra and Geometry of Nonlinear Pdes100% (3)Glimpses of Soliton Theory: The Algebra and Geometry of Nonlinear Pdes322 pages
- Dommelen Quantum Mechanics For EngineersNo ratings yetDommelen Quantum Mechanics For Engineers1,623 pages
- The Symmetric Spinning Top: Deriving The Equations of MotionNo ratings yetThe Symmetric Spinning Top: Deriving The Equations of Motion13 pages
- An Introduction To Numerical Methods For The Solutions of Partial Differential EquationsNo ratings yetAn Introduction To Numerical Methods For The Solutions of Partial Differential Equations12 pages
- 16 - Jacobian Elliptic Functions and Theta FunctionsNo ratings yet16 - Jacobian Elliptic Functions and Theta Functions15 pages
- Physics - Introduction To Differential Geometry and General Relativity100% (2)Physics - Introduction To Differential Geometry and General Relativity128 pages
- Introduction To Metric and Topological Spaces by Wilson Sutherland Unofficial Solutions ManualNo ratings yetIntroduction To Metric and Topological Spaces by Wilson Sutherland Unofficial Solutions Manual16 pages
- g3 Jakobschwichtenberg-Com-Naive-Introduction-Lie-TheoryNo ratings yetg3 Jakobschwichtenberg-Com-Naive-Introduction-Lie-Theory13 pages
- An Introduction ToTensor Calculus and Relativity - Derek F. Lawden100% (2)An Introduction ToTensor Calculus and Relativity - Derek F. Lawden199 pages
- Otto Schreier, Emanuel Sperner - Projective Geometry of N Dimensions (Volume Two of Introduction To Modern Algebra and Matrix Theory) - Chelsea Pub. Co (1961)No ratings yetOtto Schreier, Emanuel Sperner - Projective Geometry of N Dimensions (Volume Two of Introduction To Modern Algebra and Matrix Theory) - Chelsea Pub. Co (1961)208 pages
- ALL SPACETIMES BEYOND EINSTEIN by Frederic P. SchullerNo ratings yetALL SPACETIMES BEYOND EINSTEIN by Frederic P. Schuller44 pages
- Lorentz Transformation For Frames in Standard ConfigurationNo ratings yetLorentz Transformation For Frames in Standard Configuration19 pages
- GENERAL MECHANICS Being Volume I of Introduction To Theoretical Physics - Macmillan (1949)No ratings yetGENERAL MECHANICS Being Volume I of Introduction To Theoretical Physics - Macmillan (1949)281 pages
- Probability and Geometry On Groups Lecture Notes For A Graduate CourseNo ratings yetProbability and Geometry On Groups Lecture Notes For A Graduate Course209 pages
- Notes On Hyperbolic Geometry: Ormsbyk@reed - Edu100% (1)Notes On Hyperbolic Geometry: Ormsbyk@reed - Edu75 pages
- The Dog-And-Rabbit Chase Problem As An Exercise in Introductory KinematicsNo ratings yetThe Dog-And-Rabbit Chase Problem As An Exercise in Introductory Kinematics5 pages
- A Course in Differential Geometry Thierry Aubin100% (1)A Course in Differential Geometry Thierry Aubin198 pages
- Multivariable Mathematics With Maple - Linear Algebra, Vector Calculus and Differential - Carlson & Johnson100% (1)Multivariable Mathematics With Maple - Linear Algebra, Vector Calculus and Differential - Carlson & Johnson40 pages
- Pseudoparticle Solutions of The Yang-Mills Equations100% (1)Pseudoparticle Solutions of The Yang-Mills Equations3 pages
- (University Lecture Series 76) Mario Garcia-Fernandez, Jeffrey Streets - Generalized Ricci Flow-American Mathematical Society (2021)No ratings yet(University Lecture Series 76) Mario Garcia-Fernandez, Jeffrey Streets - Generalized Ricci Flow-American Mathematical Society (2021)257 pages
- Multiplicative Number Theory I Classical Theory Hugh L. Montgomery DownloadNo ratings yetMultiplicative Number Theory I Classical Theory Hugh L. Montgomery Download71 pages
- The Vector Algebra War: A Historical PerspectiveNo ratings yetThe Vector Algebra War: A Historical Perspective8 pages
- Patidar Nonstandard Finite Difference MethodsNo ratings yetPatidar Nonstandard Finite Difference Methods34 pages
- Patidar Nonstandard Finite Difference Methods PDF100% (1)Patidar Nonstandard Finite Difference Methods PDF34 pages
- Direction Cosines and Direction Ratios Laq 7 Mark Questions I B100% (1)Direction Cosines and Direction Ratios Laq 7 Mark Questions I B7 pages
- Differentiable Manifolds: Lecture Notes For Geometry 2100% (1)Differentiable Manifolds: Lecture Notes For Geometry 2150 pages
- Introduction To Manifolds: Simple To ComplexNo ratings yetIntroduction To Manifolds: Simple To Complex43 pages
- Differential Geometry I Script 15-11-24 NEWNo ratings yetDifferential Geometry I Script 15-11-24 NEW273 pages
- MATH3968 Lecture 1: DR Emma Carberry 28 July 2009No ratings yetMATH3968 Lecture 1: DR Emma Carberry 28 July 20095 pages
- Applications of Differential Geometry To Physics: Cambridge Part III Maths100% (1)Applications of Differential Geometry To Physics: Cambridge Part III Maths29 pages
- (Ebook) Spectral Theory and Differential Operators by D.E. Edmunds Des Evans ISBN 9780198812050, 0198812051 PDF Download100% (1)(Ebook) Spectral Theory and Differential Operators by D.E. Edmunds Des Evans ISBN 9780198812050, 0198812051 PDF Download52 pages
- Advanced Mathematical Methods For Scientists and ENo ratings yetAdvanced Mathematical Methods For Scientists and E7 pages
- Studiu de Caz Markov Analysis: Data (One Step Transition Matrix)No ratings yetStudiu de Caz Markov Analysis: Data (One Step Transition Matrix)3 pages
- Cambridge O Level: Additional Mathematics 4037/22No ratings yetCambridge O Level: Additional Mathematics 4037/2216 pages
- Revisiting Functions: Lesson DescriptionNo ratings yetRevisiting Functions: Lesson Description11 pages
- On Discrete-Kirchhoff Plate Finite Elements: Implementation and Discretization ErrorNo ratings yetOn Discrete-Kirchhoff Plate Finite Elements: Implementation and Discretization Error24 pages
- Solution Manual for First Course in Differential Equations with Modeling Applications 11th Edition by Zill ISBN 9781305965720 download100% (3)Solution Manual for First Course in Differential Equations with Modeling Applications 11th Edition by Zill ISBN 9781305965720 download108 pages
- MTH 210 - Lecture 4 - Permutation FunctionsNo ratings yetMTH 210 - Lecture 4 - Permutation Functions10 pages
