Manifold

D EPARTMENT OF M ATHEMATICS , C ORNELL U NIVERSITY, I THACA , N EW Y ORK 14853-4201 E-mail address: URL:
8 92#(&7$) 654333 22&) ' " $ " ! 1 0 ( " $ $ ! '&%"#
Manifolds and Differential Forms
Reyer Sjamaar
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 Chapter 1. Introduction 1.1. Manifolds 1.2. Equations 1.3. Parametrizations 1.4. Conguration spaces Exercises Chapter 2. Differential forms on Euclidean space 2.1. Elementary properties 2.2. The exterior derivative 2.3. Closed and exact forms 2.4. The Hodge star operator 2.5. div, grad and curl Exercises Chapter 3. Pulling back forms 3.1. Determinants 3.2. Pulling back forms Exercises Chapter 4. Integration of 1-forms 4.1. Denition and elementary properties of the integral 4.2. Integration of exact 1-forms 4.3. The global angle function and the winding number Exercises Chapter 5. Integration and Stokes theorem 5.1. Integration of forms over chains 5.2. The boundary of a chain 5.3. Cycles and boundaries 5.4. Stokes theorem Exercises Chapter 6. Manifolds 6.1. The denition 6.2. The regular value theorem Exercises Chapter 7. Differential forms on manifolds
iii
v 1 1 7 9 9 13 17 17 20 22 23 24 27 31 31 36 42 47 47 49 51 53 57 57 59 61 63 64 67 67 72 77 81
iv
CONTENTS
7.1. First denition 7.2. Second denition Exercises Chapter 8. Volume forms 8.1. n-Dimensional volume in R N 8.2. Orientations 8.3. Volume forms Exercises Chapter 9. Integration and Stokes theorem on manifolds 9.1. Manifolds with boundary 9.2. Integration over orientable manifolds 9.3. Gau and Stokes Exercises Chapter 10. Applications to topology 10.1. Brouwers xed point theorem 10.2. Homotopy 10.3. Closed and exact forms re-examined Exercises Appendix A. Sets and functions A.1. Glossary A.2. General topology of Euclidean space Exercises Appendix B. Calculus review B.1. The fundamental theorem of calculus B.2. Derivatives B.3. The chain rule B.4. The implicit function theorem B.5. The substitution formula for integrals Exercises Bibliography The Greek alphabet Notation Index Index
81 82 89 91 91 94 96 100 103 103 106 108 109 113 113 114 118 122 125 125 127 127 129 129 129 131 132 133 134 137 139 141 143
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 manifolds and differential forms for an audience of undergraduates who have taken a typical calculus sequence at a North American university, including basic linear algebra and multivariable calculus up to the integral theorems of Green, Gau 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 briey reviewed within the text and in appendices. The selection of material is similar to that in Spivaks 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 problems. 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 denitions 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 uid mechanics and one on curvature, perhaps including the theorems of Poincar-Hopf and Gau-Bonnet. These notes and eventual revisions can be downloaded from the course website at . Corrections, suggestions and comments will be received gratefully. Ithaca, NY, 2006-08-26
W A @H i V Y U g D f e d b b b P W R T P P I P c b a ` Y VH W W V U T S RH @ A IH qI p6hQD #V D#QG)D#&XEPQ5FFFGDE)@ DC B A A
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 thoroughly in later chapters. 1.1. Manifolds Recall that Euclidean n-space Rn is the set of all column vectors with n real entries x1 x2 x . , . . xn which we shall call points or n-vectors and denote by lower case boldface letters. In R2 or R3 we often write
For reasons having to do with matrix multiplication, column vectors are not to be confused with row vectors x 1 x2 . . . xn . For clarity, we shall usually separate the entries of a row vector by commas, as in x 1 , x2 , . . . , xn . Occasionally, to save space, we shall represent a column vector x as the transpose of a row vector,
A manifold is a certain type of subset of Rn . A precise denition will follow in Chapter 6, but one important consequence of the denition is that a manifold has a well-dened 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 denition 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 without kinks or self-intersections. Instead of the tangent space at a point one usually 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 rst 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
x1 , x2 , . . . , xn
x , y
resp.
x y z
w w xv w y
st t t u
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-dened tangent line; the ve-fold loop has ve points of self-intersection, at each of which there are two distinct tangent lines. The bow tie and the ve-pointed star have well-dened tangent lines everywhere. Still they are not manifolds: the bow tie has a selfintersection and the cusps of the star have a jagged appearance which is proscribed by the denition 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
, 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 picture, 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.
1. INTRODUCTION
Dimension 2. A two-dimensional manifold is a smooth surface without selfintersections. It may have a boundary, which is always a one-dimensional manifold. You can have two-dimensional manifolds in the plane R2 , but they are relatively 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 paraboloid and a torus.
e3 e2 e1
The famous Mbius 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
two boundary edges of the rectangle tied together and is therefore a single closed curve.
Out of the Mbius 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 direction. b b a a a b Klein bottle a a a
Mbius band
b projective plane
Perhaps the easiest way to make a Klein bottle is rst 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 Mbius band until it touches the central circle. This creates a Mbius type band with a gure eight cross-section. Equivalently, take a length of tube with a gure eight cross-section and weld the ends together giving one end a half twist. Again the
1. INTRODUCTION
resulting surface has a self-intersection, namely the central circle of the original Mbius 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 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
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
of m equations in n unknowns. Here 1 , 2 , . . . , m are functions, c 1 , c2 , . . . , cm are constants and x 1 , x2 , . . . , xn are variables. By introducing the useful shorthand x ,
we can represent this system as a single equation It is in general difcult to nd 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 dened by linear equations (i.e. where is a matrix) are called afne 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 Here y2 1,
x2 y
y2 z
c.
and
xn
m x
cn
0. c 1 . 0
x1 x2 . . .
1 x 2 x . . .
c1 c2 . . .
m x1 , . . . , xn
2 x1 , . . . , xn
6
. . .
1 x1 , . . . , xn
c1 , c2 ,
cm ,
1. INTRODUCTION
The solution set of this system is the intersection of a cylinder of radius 1 about the z-axis (given by the rst equation) and a plane cutting the x-axis at a 45 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 x r. Here
is the norm or length of x and
is the inner product or dot product of x and y. The sphere of radius r is an n 1dimensional manifold in Rn . The sphere of radius 1 is called the unit sphere and is denoted by Sn 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 therefore not necessarily a manifold. A simple example is xy 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 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 dene the tangent space to M at x. The rst view is that the tangent space at x consists of all vectors y such that y x x 0, i.e. y x x x r 2 . In coordinates: y 1 x1 yn xn r2 . This is an inhomogeneous linear equation in y. In this view, the tangent space at x is an afne subspace of Rn , given by the single equation y x r 2 . However, for most practical purposes it is easier to translate this afne subspace 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 x 0, and this is the denition that we shall espouse. The standard notation for the tangent space to M at x is Tx M. Thus
a linear subspace of Rn . (In Exercise 1.6 you will be asked to nd a basis of Tx M for a particular x and you will see that Tx M is 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 x r. Its boundary is the sphere of radius r.
p jf
Tx M
Rn y x
0 ,
m GGk k k
f ge
e s
p qf
Rn
m GGk k k
x y
x1 y1
x2 y2
m GGk k k
lf
ijhe f
x x
x2 1
x2 2
x2 n
xn yn
f ge
e k o f k Xv wxe n e u
1.4. CONFIGURATION SPACES
1.3. Parametrizations A dual method for describing manifolds is the explicit way, namely by parametrizations. For instance,
parametrizes the unit circle in R2 and
parametrizes the unit sphere in R3 . (Here is the angle between a vector and the xy-plane and 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 nd 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 parametrize 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 , to the rectangle 0, 2 2, 2 to avoid counting points twice. Only the meridian 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 denition 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. Conguration 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 differential equations (see e.g. Exercise 1.10) and conguration spaces. The conguration of a mechanical system (such as a pendulum, a spinning top, the solar system, a uid, or a gas etc.) is its state or position at any given time. (The conguration ignores any motions that the system may be undergoing. So a conguration is like a snapshot or a movie still. When the system moves, its conguration changes.) In practice one usually describes a conguration by specifying the coordinates of suitably chosen parts of the system. The conguration space or state space of the system is an abstract space, the points of which are in one-to-one correspondence to all physically possible congurations of the system. Very often the conguration space turns out to be a manifold. Its dimension is called the number of degrees of freedom of the system. The conguration space of even a fairly small system can be quite complicated.
| ~ }
cos cos ,
sin cos ,
cos ,
sin
sin
10
1. INTRODUCTION
1.3. E XAMPLE . A spherical pendulum is a weight or bob attached to a xed 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 xed distance (equal to the length of the rod) from the centre to any other. The conguration space is therefore a two-dimensional sphere. 3 (or 4, for those who have Some believe that only spaces of dimension heard of relativity) can have a basis in physical reality. The following two examples 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 rst by a universal joint. The resulting system is a double spherical pendulum. The state of this system can be specied by a pair of vectors x, y , x being the vector pointing from the centre to the rst weight and y the vector pointing from the rst 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 rst rod) and describes a distinct conguration. Thus there are four degrees of freedom. The conguration 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 moving 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 xed, 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 xed, the point C can rotate about the axis AB, which gives one further degree of freedom.
B C
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 conguration 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 xed sidelengths a, b, c, d. D c C d A a B b
(Think of four rigid rods attached by hinges.) What are all the possibilities? For simplicity let us disregard translations by keeping the rst edge AB xed 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 leftor 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 congurations are in one-to-one correspondence with the points on the circle of radius b about the point B, i.e. the conguration 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 reection in the diagonal AC. D
So there is in fact another circles worth of possible congurations. It is not possible to move continuously from the rst set of congurations to the second; in fact they are each others mirror images. Thus the conguration space is a disjoint union of two circles.
This is an example of a disconnected manifold consisting of two connected components. 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 A a B 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:
The juncture represents the collapsed quadrilateral. This conguration space is not a manifold, but most conguration spaces occurring in nature are (and an engineer designing an engine wouldnt want to use this quadrilateral to make a piston drive a ywheel). 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. nd 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.
1.3. Parametrize the space curve wrapped around the cone shown in Section 1.1.
1.2. Same questions as in Exercise 1.1 for the curve x
3at
t3 , y
3at2
This means that when b are merged at a point.
d the two components of the conguration space
t3 .
14
1. INTRODUCTION
1.4. Sketch the surfaces dened by the following gluing diagrams.
a b a a1 b1 b2 a2 b1 b c
b a
b d a1 a b2 a2
a2 b2 a1
b1
a2 b2 a1
b1
(Proceed in stages, rst gluing the as, then the bs, 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.) y2 z. 1.5. For the values of n indicated below graph the surface in R 3 dened by x n Determine all the points where the surface does not have a well-dened 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.)
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 nd 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 conguration 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 xed in place so as to rule out translations.) 1.9. What is the conguration space of all pentagons ABCDE in the plane with xed 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
(i) (ii) (iii) (iv)
n n n n
0. 1. 2. 3.
EXERCISES
15
freedom you may also assume the edge AB to be xed in place.)
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 satised 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 writ1 ing dy f x dx and integrating both sides. (Dont 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, , . can be found at
x9
) h9999x9hp99hxp9
CHAPTER 2
Differential forms on Euclidean space

The notion of a differential form encompasses such ideas as elements of surface area and volume elements, the work exerted by a force, the ow of a uid, and the curvature of a surface, space or hyperspace. An important operation on differential 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 possible 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
(If you dont know the symbol , look up and memorize the Greek alphabet in the back of the notes.) Here I stands for a multi-index i 1 , i2 , . . . , ik 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 coefcients of , and dx I is an abbreviation for dxik is also often used to distinguish this kind of (The notation dxi1 dxi2 product from another kind, called the tensor product.) For instance the expressions
represent a 2-form on R5 , resp. a 4-form on R6 . The form consists of four terms, corresponding to the multi-indices 1, 5 , 2, 3 , 2, 4 and 5, 3 , whereas consists of one term, corresponding to the multi-index 1, 6, 3, 2 . Note, however, that 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 denition when writing a differential form. Another reason for being precise about the domain of a form is that the coefcients f I may not be dened on all of Rn , but only on an open subset U of Rn . In such a case we say is a k-form on U. Thus the expression ln x 2 y2 z dz is not a 1-form on R3 , but on the open set U R3 x, y, z x2 y2 0 , i.e. the complement of the z-axis.
17
x1 x3 x5 dx1 dx6 dx3 dx2 ,
sin x1
e x4 dx1 dx5
x2 x2 dx2 dx3 5
GG
dxi1 dxi2
dxik .
6 dx2 dx4
cos x2 dx5 dx3 ,
f I dx I .
I
GG
18
2. DIFFERENTIAL FORMS ON EUCLIDEAN SPACE
You can think of dxi as an innitesimal increment in the variable x i and of dx I as the volume of an innitesimal k-dimensional rectangular block with sides dx i1 , dxi2 , . . . , dx ik . (A precise denition 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: and so forth. This is called anticommutativity, graded commutativity, or the alternating property. In particular, this rule implies dx i dxi dxi dxi , so dxi dxi 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 dxs). A general 1-form looks like A general 2-form has the shape
i, j
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 f 2,1 dx2 dx1 f 1,2 f 2,1 dx1 dx2 . So we can write any 2-form as
1 i j n
Written like this, a 2-form has at most
i 1
where dxi means omit the factor dx i . Every n-form on Rn can be written as f dx 1 dx2 dxn . The special n-form dx1 dx2 dxn is also known as the volume form. Forms of degree k n on Rn are always 0, because at least one variable has to repeat in any expression dx i1 dxik . By convention forms of negative degree are 0. In general a form of degree k can be expressed as a sum
f I dx I ,
I
GG
GG
f i dx1 dx2
GG
GG
GG
GG
GG
GG
f 1 dx2 dx3
dxn
f 2 dx1 dx3
components. Likewise, a general n
1-form can be written as a sum of n components, dxn f n dx1 dx2 dxn

1
GG
1 n n 2
dxi
GG
GG
g2,3 dx2 dx3
GG
gi, j dxi dx j
g1,2 dx1 dx2
g1,n dx1 dxn g2,n dx2 dxn gn

1,n dx n 1 dx n .
dxn ,
f n,1 dxn dx1
f n,2 dxn dx2
GG GG
GG
f 2,1 dx2 dx1
f 2,2 dx2 dx2
GG
f i, j dxi dx j
f 1,1 dx1 dx1
f 1,2 dx1 dx2
GG
f 1 dx1
f 2 dx2
f n dxn .
f 1,n dx1 dxn f 2,n dx2 dxn
f n,n dxn dxn .
GG
GG
GG
GG
GG
GG
dxi1 dxi2
dxiq
dxi p
dxik
dxi1 dxi2
dxi p
dxiq
dxik ,
(2.1)
GG
2.1. ELEMENTARY PROPERTIES
19
where the I are increasing multi-indices, 1 i1 i2 ik n. We shall almost always represent forms in this manner. The maximum number of terms occurring in 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 coefcient n choose k,
(Compare this to the number of all multi-indices of degree k, which is n k .) Two k-forms I f I dx I and I g I dx I (with I ranging over the increasing multiindices of degree k) are considered equal if and only if f I g I for all I. The collection of all k-forms on an open set U is denoted by k U . Since k-forms can be added together and multiplied by scalars, the collection k U constitutes a vector space. A form is constant if the coefcients f I are constant functions. The set of constant k-forms is a linear subspace of k U of dimension n . A basis of this subk space is given by the forms dx I , where I ranges over all increasing multi-indices of degree k. (The space k U itself is innite-dimensional.) The (exterior) product of a k-form I f I dx I and an l-form J g J dx J is dened to be the k l-form
f I g J dx I dx J .
I,J
Usually many terms in a product cancel out or can be combined. For instance, As an extreme example of such a cancellation, consider an arbitrary form of degree k. Its p-th power p is of degree kp, which is greater than n if k 0 and p n. Therefore n 1 0 for any form on Rn of positive degree. The alternating property combines with the multiplication rule to give the following result. 2.1. P ROPOSITION (graded commutativity).
kl
for all k-forms and all l-forms .
. . .
kl
dx J dx I .
GG
GG
2k
dx j1 dx j2 dxi1 dxi2
dxik dx j3
GG
GG
dx j1 dxi1 dxi2
dxik dx j2 dx j3
GG
GG
dx I dx J
dxi1 dxi2
dxik dx j1 dx j2 dx j3
dx jl dx jl dx jl
P ROOF. Let I i1 , i2 , . . . , ik and J the alternating property we get
j1 , j2 , . . . , jl . Successively applying
y dx
x dy x dx dz
y dy dz
y 2 dx dy dz
x 2 dy dx dz
y2
n k
n! . k! n k !
GG
x2 dx dy dz.
20
I,J
I,J
which establishes the result.
If f is a 0-form, that is a smooth function, we dene d f to be the 1-form
Then we have the product or Leibniz rule:
If I f I dx I is a k-form, each of the coefcients f I is a smooth function and we dene d to be the k 1-form
The operation d is called exterior differentiation. An operator of this sort is called a rst-order partial differential operator, because it involves the rst partial derivatives of the coefcients of a form.
(Recall that f y is an alternative notation for f y.) More generally, for a 1-form n 1 f i dxi on Rn we have i
where in line (2.2) in the rst sum we used the alternating property and in the second sum we interchanged the roles of i and j.
1 i j n
fj xi
fi dxi dx j , x j
1 i j
fi dx dx x j i j n
1 i j
fi dx dx x j j i n
i 1
d f i dxi
i, j
fi dx dx x j j i 1
1 j i
fi dx dx x j j i n fj dx dx xi i j n (2.2)
1 i j
f y dy dx
2.3. E XAMPLE . If
f dx
g dy is a 1-form on R2 , then g x dx dy gx f y dx dy.
d f I dx I .
I
d fg
2.2. C OROLLARY. 2
0 if is a form of odd degree. 2.2. The exterior derivative

n
df
i 1
xi dxi .
f dg g df .
A noteworthy special case is . Then we get 2 1 k 2 2 This equality is vacuous if k is even, but tells us that 0 if k is odd.
g J f I dx J dx I
For general forms
I f I dx I and
J g J dx J we get from this

kl
f I g J dx I dx J
kl
,
QED
2
1 k 2 .
2.2. THE EXTERIOR DERIVATIVE
21
Here in line (2.3) we rearranged the subscripts (for instance, in the rst term we relabelled k i, i j and j k) and in line (2.4) we applied the alternating property. An obvious but quite useful remark is that if is an n-form on Rn , then d is of degree n 1 and so d 0. The operator d is linear and satises a generalized Leibniz rule. 2.5. P ROPOSITION . (i) d a b a d b d for all k-forms and and all scalars a and b. (ii) d d 1 k d for all k-forms and l-forms . P ROOF. The linearity property (i) follows from the linearity of partial differentiation: f g a f bg a b xi xi xi for al smooth functions f , g and constants a, b. Now let I f I dx I and J g J dx J . The Leibniz rule for functions and Proposition 2.1 give
I,J I,J I,J
which proves part (ii).
Here is one of the most curious properties of the exterior derivative.
1 k d, QED
d f I dx I g J dx J
1 k f I dx I dg J dx J
f I g J dx I dx J
f I dg J
1 i j k n
f i, j xk
f i,k x j
1 i j k
f i, j dxk dxi dx j xk n f j,k dxi dx j dxk . xi
1 i j k
f j,k dxi dx j dxk xi n
1 i j k
f i, j dxk dxi dx j xk n
1 i j k
1 k i j
1 i j n
d f i, j dxi
For a general 2-form
i j n f i, j dx i dx j
on
Rn
we have
1 i j nk 1
f i, j dxk dxi dx j xk
1 i k j
f i,k dx j dxi dxk x j n (2.3) (2.4)
g J d f I dx I dx J
f z dz dx dy
g y dy dx dz
h x dx dy dz
2.4. E XAMPLE . If
f dx dy
g dx dz
h dy dz is a 2-form on R 3 , then fz gy h x dx dy dz.
22
Applying the formula of Example 2.3 (replacing f i with f I xi ) we nd
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 d 0. QED 2.3. Closed and exact forms
2.7. P ROPOSITION . Every exact form is closed. 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 ramications, 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 denition of the form. Let us consider the simplest case of a 1-form n 1 f i dxi . Determining i whether is exact means solving the equation dg for the function g. This amounts to g g g f1 , f2 , ..., fn, (2.5) x1 x2 xn a system of rst-order partial differential equations. Finding a solution is sometimes called integrating the system. By Proposition 2.7 this is not possible unless is closed. By the formula in Example 2.3 is closed if and only if
for all 1 i j n. These identities must be satised for the system (2.5) to be solvable and are therefore called the integrability conditions for the system.
! )
'! &
'! &
dy dx
y sin yz
cos yz dz dy
dx dy z sin yz cos yz dy dz 0,
%!
2.9. E XAMPLE . Let
y dx
z cos yz
fi x j
fj xi
x dy
y cos yz dz. Then
"
!
P ROOF. If
d then d
d d
0 by Proposition 2.6.
A form is closed if d one less).
0. It is exact if
d for some form (of degree

i 1
1 i j n
d

fI dxi xi
2 f I xi x j
2 f I dxi dx j x j xi
I i
I i 1
d d
P ROOF. Let
I f I dx I . Then d
fI dx dx I xi i 1
2.6. P ROPOSITION . d d
0 for any form . In short, d2 0.

n
fI dxi dx I . xi
0,
QED
2.4. THE HODGE STAR OPERATOR
23
so is closed. Is exact? Let us solve the equations
by successive integration. The rst 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 is exact. This method works always for a 1-form dened on all of Rn . (See Exercise 2.6.) Hence every closed 1-form on Rn is exact.
is called the angle form for reasons that will become clear in Section 4.3. From
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 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 coefcient n is the number of ways of selecting k (unordered) k objects from a collection of n objects. Equivalently, n is the number of ways of k 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 . k n k 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 is denoted by (or sometimes ) and is dened as follows. If I f I dx I , then
I
with
dx I
I dx I c .
Q2 I
fI
0
dx I ,
F E
0 H
f i, j xk
f i,k x j
f j,k xi
By the formula in Example 2.4 the integrability condition d
g j xi
gi x j
f i, j .
0 comes down to
For a 2-form 1 i j n f i, j dxi dx j and a 1-form tion d amounts to the system
n i
1 gi
F E
@ A @
x x x2 y2
y2 x2
x2 , y2 2
y y x2 y2
x2
y2
dx
x2
y2
dy
9 86 7
2.10. E XAMPLE . The 1-form on R2
0 dened by y dx x dy . x2 y2 x2 x2 y2 y2 2
dxi the equa(2.6)
3 2
g x
y,
g y
z cos yz
x,
g z
y cos yz
24
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 I is a sign,
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:
(This is the reason that 2-forms on R are sometimes written as f dx dy g dz dx 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
and
2.5. div, grad and curl Rn . We can
A vector eld on an open subset U of Rn is a smooth map F : U write F in components as F1 x F2 x F x , . . . Fn x
UU VVU
a ` a ` y a `
sVVU i UU
vww R dba `
wx
i UU rVVU
h h a
` R W gea
dxi dx j
i j 1
dx1 dx2
dxi
dx j
dxn
for 1
UU VVU
ipVVU UU
h a
` W gR
dxi
i 1
dx1 dx2
dxi
dxn
R ea
` X
UU VVU
On R we have 1
dx1 dx2
dxn ,
dx1 dx2
UVUVU ` X
dx1 dx4
dx2 dx3 ,
` X
dx1 dx3
dx2 dx4 ,
dx2 dx4 dx3 dx4 dxn
R ea R W ea R ea
` X
R ba R W ba R ba
dx1 dx2
dx3 dx4 ,
dx2 dx3
W R
W R
dx2
dx1 dx3 dx4 ,
dx4
dx1
dx2 dx3 dx4 ,
dx3
dx1 dx2 dx4 , dx1 dx2 dx3 , dx1 dx4 , dx1 dx3 , dx1 dx2 . i n, j n.
1, and for 1
dz
dx dy,
W R
dy
dx dz
dz dx,
3
dx dz dy dz
R a R W ba R a
` X ` X ` X
dx
dy dz,
dx dy
W R
2.12. E XAMPLE . On R2 we have dx
dy and dy
W Rea
` X
W R
which shows that I
1. Hence
dx 2 dx6
dx1 dx3 dx4 dx5 . dx. On R3 we have dz, dy, dx.
W R
` X
` X
` X
dx I dx I c
dx2 dx6 dx1 dx3 dx4 dx5 dx1 dx2 dx6 dx3 dx4 dx5 dx1 dx2 dx3 dx4 dx5 dx6 ,
` dR
2.11. E XAMPLE . Let n 6 and I 2, 6 . Then I c dx2 dx6 and dx I c dx1 dx3 dx4 dx5 . Therefore
` cR
R ba
X`
dx I
dx I
dx1 dx2
dxn . 1, 3, 4, 5 , so dx I
UVVU U X`
W R
W TR S
UVUVU UU VVU
1 if dx I dx I c 1 if dx I dx I c
dx1 dx2 dxn , dx1 dx2 dxn .
` X
2.5. DIV, GRAD AND CURL
25
or alternatively as F n 1 Fi ei , where e1 , e2 , . . . , en are the standard basis vectors i of Rn . Vector elds in the plane can be plotted by placing the vector F x with its tail at the point x. The diagrams below represent the vector elds ye1 xe2 and x xy e1 y xy 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 elds (i.e. points x where F x 0). y y
We can turn F into a 1-form by using the Fi as coefcients: n 1 Fi dxi . i y dx x dy corresponds to the vector eld F For instance, the 1-form ye1 xe2 . Let us introduce the symbolic notation dx1 dx2 . . .
dxn
which we will think of as a vector-valued 1-form. Then we can write F dx. Clearly, F is determined by and vice versa. Thus vector elds and 1-forms are symbiotically associated to one another.
Intuitively, the vector-valued 1-form dx represents an innitesimal displacement. If F represents a force eld, such as gravity or an electric force acting on a particle, then 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 along the path. We shall see how to do this in Section 4.1.) The correspondence between vector elds and 1-forms behaves in an interesting way with respect to exterior differentiation and the Hodge star operator. For
vector eld F
1-form :
F dx.
dx
26
f x n
This vector eld is called the gradient of f . (Equivalently, we can view grad f as the transpose of the Jacobi matrix of f .)
we can also write F dx. Intuitively, the vector-valued n 1-form dx represents an innitesimal n 1-dimensional hypersurface perpendicular to dx. (This point of view will be justied in Section 8.3, after the proof of Theorem 8.14.) In uid mechanics, the ow of a uid or gas in Rn is represented by a vector eld F. The n 1-form then represents the ux, that is the amount of material passing through the hypersurface dx per unit time. (The total amount of uid passing through a hypersurface S is found by integrating over S. We shall see how to do this in Section 5.1.) We have
i 1
An alternative way of writing this identity is obtained by applying to both sides, which gives
} ~ }
i, j
1 i j n
Fi dx dx x j j i 1
Fj xi
Fi dxi dx j , x j
A very different identity is found by rst applying d and then
s s d
div F
d . to :
rr VVr
d h
d F
dx
div F dx 1 dx2
dxn .
f sr g
The function div F
n i
1 Fi
xi is the divergence of F. Thus if
i 1
i 1
rr VVr
z{d
rr VVr
rr VVr
xii dx1 dx2

e
dxi
dxn
xii
dx1 dx2
rr VVr
w rr xVVr
v h
t f
d h
f sr
d F
dx
xii
i 1
dxi dx1 dx2
dxi
dxn dxn .
F dx, then
rr VVr
v h
t uf
kl
dxn
n 1 dx
1 dx 2
dxn
ikk
s r
s kl
ikk
dx
rVVr r rr VVr
dx1 dx2 . . .
dx2 dx3 dx1 dx3 . . .
dxn dxn
nn m
n m nn
Using the vector-valued n
1-form
i 1
i 1
rr VVr
w rr xVVr
t uf
d h
v h
Fi
dxi
Fi
i 1
dx1 dx2
Starting with a vector eld F and letting
F dx, we nd dxi dxn ,
qp
grad f
df :
df
grad f dx.
i 1
kl kk jd ikk
grad f
xi ei
. . .
nn
f x 1 f x 2
nn m
.
f e
each function f the 1-form d f
n i
f xi dxi is associated to the vector eld
s s t
EXERCISES
27
and hence
1 i j n
In three dimensions d is a 1-form and so is associated to a vector eld, namely
You need not memorize every detail of this discussion. The point is rather to remember that exterior differentiation in combination with the Hodge star unies 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. dxi dxn . (ii) n 1 x2 dx1 i i p n i 1 x dx (iii) x i 1 1 dxi dxn , where p is a real constant. For what val1 i ues of p is this form closed?
2.3. Write the coordinates on R2n as x1 , y1 , x2 , y2 , . . . , xn , yn . Let
x3
Show that dg . (Apply the fundamental theorem of calculus, formula (B.3), differentiate under the integral sign and dont forget to use d 0.)
f 3 0, 0, t, x4 , x5 , . . . , xn dt
g x
x1
f 1 t, x2 , x3 , . . . , xn dt
x2
2.6. Let
n i
1 f i dx i
be a closed C
1-form on Rn . Dene a function g by f 2 0, t, x3 , x4 , . . . , xn dt

xn
f n 0, 0, . . . , 0, t dt.
(i) (ii)
ye xy z sin xz dx 2xy 3 z4 dx 3x2 y2 z4
xe xy z2 dy ze y sin ze y dy
x sin xz 4x2 y3 z3
2.5. Check that each of the following forms g such that dg .
R3 is closed and nd a function 2yz 3z 2 dz. e y sin ze y e z dz.
Compute d
d d
d . First work out the cases n
dz
x1 dy1
x2 dy2
xn dyn
dz
i 1
xi dyi .
1, 2, 3.
2.4. Write the coordinates on R2n
as x1 , y1 , x2 , y2 , . . . , xn , yn , z . Let
uu
Compute n
(n-fold product). First work out the cases n
uu
dx1 dy1
dx2 dy2
dxn dyn
i 1
dxi dyi .
1, 2, 3.
2.2. Consider the forms Calculate (i) , ; (ii) d , d, d .
x dx
y dy,
curl F dx
d .
z dx dy
x dy dz and
uu
F3 x2 the curl of F. Thus, for n
uu
curl F
F2 F3 F1 e1 e2 x3 x1 x3 3, if F dx, then
F2 x1
F1 e3 , x2
VV
VV
rVV
i j 1
Fj xi
Fi dx1 dx2 x j
dxi
dx j
dxn .
z dy on R 3 .
28
2.8. Let and be closed forms. Prove that is also closed. 2.9. Let be closed and exact. Prove that is exact. 2.11. (i) (ii) (iii) Consider the form x 2 dx1 x2 dx2 on R2 . 2 1 Find and d d . Repeat the calculation, regarding as a form on R 3 . Again repeat the calculation, now regarding as a form on R 4 . 2.10. Calculate , , ,
2.13. Let I a I dx I and I b I dx I be constant k-forms, i.e. with constant coefcients a I and b I . (We also assume, as usual, that the multi-indices I are increasing.) The inner product of and is the number dened by
Prove the following assertions. (i) The dx I form an orthonormal basis of the space of constant k-forms. (ii) , 0 for all and , 0 if and only if 0. (iii) , dx 1 dx2 dxn . . (iv) (v) The Hodge star operator is orthogonal, i.e. , , . 2.14. The Laplacian of a smooth function on an open subset of R n is dened by 2 f x2 1 2 f x2 2 2 f . x2 n
dx,
Prove that dx 1 dx2 dxn on U. (ii) Let r : Rn R be the function r x x (distance to the origin). Deduce from part (i) that dx 1 dx2 dxn dr on Rn 0 , where x 1 x dx.
grad f x
df .
uu % %)
n x
grad f x
grad f x ,
2.16.
R be a function satis(i) Let U be an open subset of Rn and let f : U fying grad f x 0 for all x in U. On U dene a vector eld n, an n 1-form and a 1-form by
2.15. (i) (ii) (iii)
Let f : R be a function and let f dx i . Calculate d d . Calculate d d . Show that d d 1 n d d f dxi , where Exercise 2.14.
Rn
Prove the following formulas. (i) f d df . fg f g f g (ii)
df
dg . (Use Exercise 2.13(iv).)
is the Laplacian dened in
2.12. Prove that
kn k
for every k-form on Rn .
Show that dg
. (Use d
0 and apply the identity proved in Exercise B.5 to each f i .)
, where , and are as in Exercise 2.2.
aI bI .
I
i 1
g x
xi f i
x .
2.7. Let dened on Rn
n 1 f i dxi be a closed 1-form whose coefcients f i are smooth functions i 0 that are all homogeneous of the same degree p 1. Let
EXERCISES
29 1
i 1
(iii) Find 1, dx i for 1 i n 1, and dx 1 dx2 dxn . (iv) Compute the relativistic Laplacian (usually called the dAlembertian or wave operator) d d f for any smooth function f on Rn 1 . j 4. (v) For n 3 (ordinary space-time) nd dx i dx j for 1 i 2.18. One of the greatest advances in theoretical physics of the nineteenth century was Maxwells formulation of the equations of electromagnetism:
Here c is the speed of light, E is the electric eld, H is the magnetic eld, J is the density of electric current, 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 elds and is a function on R 3 and all depend on time t. The Maxwell equations look particularly simple in differential form notation, as we shall now see. In space-time R4 with coordinates x 1 , x2 , x3 , x4 , where x4 ct, introduce forms
(i) Show that Maxwells equations are equivalent to d 4
(ii) Conclude that is closed and that div J t 0. (iii) In vacuum one has E D and H B. Show that in vacuum , the relativistic Hodge star of dened in Exercise 2.17. (iv) Free space is a vacuum without charges or currents. Show that the Maxwell equations in free space are equivalent to d d 0.
0, 0.
1 J dx dx c 1 2 3
J2 dx3 dx1
J3 dx1 dx2 dx4
dx1 dx2 dx3 .
H1 dx1
H2 dx2
H3 dx3 dx4
D1 dx2 dx3
D2 dx3 dx1
E1 dx1
E2 dx2
E3 dx3 dx4
B1 dx2 dx3
B2 dx3 dx1
B3 dx1 dx2 , D3 dx1 dx2 ,
div B
div D
curl H
curl E
1 B c t 4 1 D J c c t 4
(Faradays Law), (Ampres Law), (Gau Law), (no magnetic monopoles).
(Here I and
Ic
are as in the denition of the ordinary Hodge star.)
dx I
with
I dx I c I dx I c
if I contains n 1, if I does not contain n
1.
A Hodge star operator corresponding to this inner product is dened as follows: if I f I dx I , then f I dx I ,
(i) Give examples of (nonzero) vectors of each type. (ii) Show that for every x 0 there is a y such that x, y
0.
A vector x
Rn
is spacelike if x, x
0, lightlike if x, x
0, and timelike if x, x
x, y
xi yi
xn
1 yn 1 .
2.17. The Minkowski or relativistic inner product on R n
is given by
0.
30
Show that the corresponding 2-form satises the free Maxwell equations d d 0. Such solutions are called electromagnetic waves. Explain why. In what direction do these waves travel?
E x
(v) Let f , g : R
R be any smooth functions and dene 0 f x1 g x1 x4 x4 , B x 0 g x1 x4 f x1 x4 .
CHAPTER 3
Pulling back forms

3.1. Determinants The determinant of a square matrix is the oriented volume of the block (parallelepiped) spanned by its column vectors. It is therefore not surprising that differential forms are closely related to determinants. This section is a review of some fundamental facts concerning determinants. Let A a1,1 . . . ... ... a1,n . . .
an,1
an,n
be an n n-matrix with column vectors a1 , a2 , . . . , an . Its determinant is variously denoted by
Expansion on the rst column. You have probably seen the following denition of the determinant:
i 1
Here Ai, j denotes the n 1 n 1 -matrix obtained from A by striking out the i-th row and the j-th column. This is a recursive denition, which reduces the calculation of any determinant to that of determinants of smaller size. (The recursion starts at n 1; the determinant of a 1 1-matrix a is simply dened to be the number a.) It is a useful rule, but it has two serious aws: rst, it is extremely inefcient computationally (except for matrices containing lots of zeroes), and second, it obscures the relationship with volumes of parallelepipeds. Axioms. A far better denition is available. The determinant can be completely characterized by three simple laws, which make good sense in view of its geometrical signicance and which comprise an efcient algorithm for calculating any determinant. 3.1. D EFINITION . A determinant is a function det which assigns to every ordered n-tuple of vectors a1 , a2 , . . . , an a number det a1 , a2 , . . . , an subject to the following axioms:
31
det A
i 1
ai1 det Ai,1 .
an,1
...
an,n
det A
det a1 , a2 , . . . , an
det ai, j
1 i, j n

a1,1 . . . ... a1,n . . . .
32
3. PULLING BACK FORMS
(i) det is multilinear (i.e. linear in each column): c det a1 , a2 , . . . , ai , . . . , an
for all scalars c, c and all vectors a1 , a2 , . . . , ai , ai , . . . , an ; (ii) det is alternating or antisymmetric: for any i j; (iii) normalization: det e1 , e2 , . . . , en dard basis vectors of Rn .
1, where e1 , e2 , . . . , en are the stan-
We also write det A instead of det a1 , a2 , . . . , an , 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.
3.3. E XAMPLE . Identify the column operations applied at each step in the following calculation.
As this example suggests, the axioms (i)(iii) sufce to calculate any n ndeterminant. 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 be two functions satisfying Axioms (i)(iii). Then det A det A for all n n-matrices A. P ROOF. Let a1 , a2 , . . . , an be the column vectors of A. Suppose rst that A is not invertible. Then the columns of A are linearly dependent. For simplicity let us assume that the rst column is a linear combination of the others: a1 c 2 a2 cn an . Applying axioms (i) and (ii) we get
i 2
det A
ci det
ai , a2 , . . . , a i , . . . , a n
0,
1 2 3 0
0 1 1
0 1 0
1 2 0 0
0 0 1
0 1 0
1 2 0 0
0 1 0
0 0 1
1 4 1
1 10 5
1 9 4
1 4 1
0 6 4
0 5 3
1 4 1
0 1 1
0 5 3
1 3 0
0 1 1
0 2 0
3.2. L EMMA . If E is an elementary column operation, then det E A where 1 if E is of type I, k c if E is of type II (multiplication of a column by c), 1 if E is of type III.
det a1 , . . . , ai , . . . , a j , . . . , an
det a1 , . . . , a j , . . . , ai , . . . , an
k det A,
2.
det a1 , a2 , . . . , cai
c ai , . . . , a n
c det a1 , a2 , . . . , ai , . . . , an
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 E2 E1 A I. According to be these elementary operations, so that E m Em 1 Lemma 3.2, each operation E i has the effect of multiplying the determinant by a certain factor k i , so axiom (iii) yields
3.5. R EMARK (change of normalization). Suppose that det is a function that satises the multilinearity axiom (i) and the antisymmetry axiom (ii) but is normalized 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) consistent? We will answer this question by displaying an explicit formula for the determinant of any n n-matrix that does not involve any column reductions. Unlike Denition 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.
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 For general n there are
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 and we think of this as a shorthand notation for the map given by 1 5, 2 3, 3 1, 4 2 and 5 4. The permutation 1, 2, 3, . . . , n 1, n then corresponds to the identity map on the set 1, 2, . . . , n . If is the identity permutation, then clearly i j whenever i j. However, if is not the identity permutation, it cannot preserve the order in this way. An inversion in is any pair of numbers i and j such that 1 i j n and
& $ & $ #
$% " ! " ! # #
n n
1 n
3 2 1
5, 3, 1, 2, 4 ,
1, 2, 3 ,
1, 3, 2 ,
2, 1, 3 ,
2, 3, 1 , n!
3, 1, 2 ,
3, 2, 1 .

det A
sign a1,
a2,
an,
3.6. T HEOREM (existence of determinants). Every n dened determinant. It is given by the formula
n-matrix A has a welln
Applying the same reasoning to det A we get 1 det A 1 k1 k2 km det A.
km km
k2 k1 det A. Hence QED

"

det I
det Em Em
E2 E1 A
km km

k2 k1 det A.

34
i j . The length of , denoted by l , is the number of inversions in . 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 is 1 if is even, 1 l sign 1 if 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.
Thinking of permutations in Sn as bijective maps from 1, 2, . . . , n to itself, we can form the composition of any two permutations and in S n . For permutations we usually write instead of and call it the product of and . This is the permutation produced by rst performing and then ! For instance, if 5, 3, 1, 2, 4 and 5, 4, 3, 2, 1 , then
A basic fact concerning signs, which we shall not prove here, is
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 permutation . 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 Theorem 3.6 gives the well-known formul
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 Denition 3.1. Let us for the moment denote the right-hand side by f A . Axiom (i) is checked as follows: for every permutation the product a1,
a2,
an,
a1,3 a2,1 a3,2
5 4 G5 4 5 4 FFF ( '
1 DD
DD
a1,1 a2,1 a3,1
a1,2 a2,2 a3,2
a1,3 a2,3 a3,3
a1,1 a2,2 a3,3
a1,1 a2,3 a3,2
a1,1 a2,1
a1,2 a2,2
a1,1 a2,2
a1,2 a2,1 ,
a1,2 a2,1 a3,3
a1,2 a2,3 a3,1 a1,3 a2,2 a3,1 .
'
( '
( '
1 DDD
1B( '
sign
sign sign .
' A1
1, 3, 5, 4, 2 ,
4, 2, 1, 3, 5 . (3.1)
1, 2, 3 1, 3, 2 2, 1, 3 2, 3, 1 3, 1, 2 3, 2, 1
0 1 1 2 2 3
3 3 ( '
sign
1 1 1 1 1 1
( '
9 8
( '
( ' ( ' ( ' ( ' ( ' ( ' ( ' 1 3 1 7( 3 1 ( 2( ' 5 4 3' 1 ( ( D @ ' 21 ' 21 DD DD
'
DD
( ' 0( ' ) ( ' ' ' 21 DD DD DD
3.1. DETERMINANTS
35
contains exactly one entry from each row and each column in A. So if we multiply the i-th row of A by c, each term in f A is multiplied by c. Therefore Similarly, f a1 , a 2 , . . . , a i
Axiom (ii) holds because if we interchange two columns in A, each term in f A changes sign. To see this, let be the permutation in S n that interchanges the two numbers i and j and leaves all others xed. Then
Sn Sn Sn
sign sign a1, sign a1,

1
a2,
an,
by formula (3.1) by Exercise 3.4
Sn
a2,
an,
f a1 , . . . , a i , . . . , a j , . . . , a n . I,
Here are some further rules followed by determinants. Each can be deduced from Denition 3.1 or from Theorem 3.6. (Recall that the transpose of an n nai, j is the matrix A T whose i, j-th entry is a j,i.) matrix A (i) det AB det A det B. (ii) det A T det A. (iii) (Expansion on the j-th column) det A n 1 1 i j ai, j det Ai, j for all j i 1, 2, . . . , n. Here A i, j denotes the n 1 n 1 -matrix obtained from A by striking out the i-th row and the j-th column. (iv) det A a1,1 a2,2 an,n if A is upper triangular (i.e. a i, j 0 for i j). Volume change. We conclude this discussion with a slightly different geometric view of determinants. A square matrix A can be regarded as a linear map A : Rn Rn . The unit cube in Rn , has n-dimensional volume 1. (For n 1 it is usually called the unit interval and for n 2 the unit square.) Its image A 0, 1 n under the map A is a parallelepiped with edges Ae1 , Ae2 , . . . , Aen , the columns of A. Hence A 0, 1 n
w i hv
w i hv
0, 1
Rn 0
xi
for i
1, 2, . . . , n ,
I W %aBI W H H e I dWH c P
3.7. T HEOREM . Let A and B be n
n-matrices.
t s
VV V
r qP i h p
P PbI H
P I H
and therefore f I
1. So f satises all three axioms for determinants.
VV Y `P U T V U T U T
a1,
a2,
an,
Finally, rule (iii) is correct because if A
1 if identity, 0 otherwise, QED
sign a1,
a2,
I U T GU T VVV U T GU T U T VVV VV U T V U T VVV GU T
an,
UT
UT I H I H I H UT I H UT I H
sign a1,
f a1 , . . . , a j , . . . , a i , . . . , a n
a2,
an,
substitute
I H
H QI
H BI P
ai , . . . , a n
f a1 , a 2 , . . . , a i , . . . , a n
f a1 , a 2 , . . . , a i , . . . , a n .
f a1 , a2 , . . . , cai , . . . , an
PBI I H
c f a1 , a2 , . . . , a i , . . . , a n .
H WXP WXP
R Q S S S
I AP H
P P P P g
36
has n-dimensional volume vol A 0, 1 n det A det A vol 0, 1 n . This rule n , then generalizes as follows: if X is a measurable subset of R vol A X det A vol X.
e2
Ae2
AX
e1 Ae1 So det A can be interpreted as a volume change factor. (A set is measurable if it has a well-dened, nite or innite, n-dimensional volume. Explaining exactly what this means is rather hard, but it sufces 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 dened the angle form on R 2
By substituting x 1-form on R:
cos t and y
sin t into the angle form we obtain the following
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 is a k-form dened on an open subset V of Rm . Let us denote the coordinates on Rm by y1 , y2 , . . . , ym and let us write, as usual, f I dy I , where the functions f I are dened 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
. . .
ym
m x1 , . . . , xn .
y2
2 x1 , . . . , xn ,
y1
1 x1 , . . . , xn ,
sin t
sin t
sin t d cos t cos t d sin t cos 2 t sin 2 t
y dx x dy . x2 y2
cos2 t dt
y x
0 to be
dt.
3.2. PULLING BACK FORMS
37
We assume that the functions i are smooth and dened on a common domain U, which is an open subset of Rn . We regard as a map from U to V. (In Example 3.8 we have U R, V R2 0 and t cos t, sin t .) The pullback of along is then the k-form on U obtained by substituting y i i x1 , . . . , xn for all i in the formula for . That is to say, is dened by

Here f I is dened by
f I x ; in other words, f I the composition of and f I . This means f I x is the function resulting from f I by substituting y x . The pullback dy I is dened by replacing each y i with i . That is to say, if I i 1 , i2 , . . . , ik we put
The picture below is a schematic representation of the substitution process. The form I f I dy I is a k-form in y1 , y2 , . . . , ym ; its pullback J g J dx J is a k-form in x1 , x2 , . . . , xn . In Theorem 3.12 below we will give an explicit formula for the coefcients g J in terms of f I and . U

x y
Rn 3.9. E XAMPLE . The formula
x z y x w w
x1 x2
x3 x2 1 ln x1 x2
v u t
dy I
dyi1 dyi2
dyik
di1 di2
dik .
Rm
rr r rr B r 2 j p
fI
p p
f I dy I .
f I ,
fg k e dff i jh ii
m x 1 x 2 x . . .
AB
o nl m
As usual we write y
x , where
38
denes a map : U R2 , where U x R2 x 1 x 2 0 . The components 3 x and x , x of are given by 1 x1 , x2 x1 2 ln x1 x2 . Accordingly, 2 1 2
3x2 x2 2 x1
x2
x3 1
dx1 dx2 .
Observe that the pullback operation turns k-forms on the target space V into k-forms on the source space U. Thus, while : U V is a map from U to V, is a map 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 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 : U V be a smooth map, where U is open in R n and m . The pullback operation is V is open in R (i) linear: a b a b ; (ii) multiplicative: ; (iii) natural: , where : V with W open in Rk and a form on W.
W is a second smooth map
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 composition of maps. P ROOF. If I f I dy I and I g I dy I are two forms of the same degree, then a b I a f I bg I dy I , so
I
Now
so a b a b . This proves part I a f I b g I dy I (i). For the proof of part (ii) consider two forms I f I dy I and J g J dy J (not necessarily of the same degree). Then I,J f I g J dy I dy J , so
Now
| j |
fI gJ x
| | | | | | | | |
a fI
bg I x
a fI
bg I x
a fI x
I,J
f I g J dy I dy J .
fI gJ x
fI x gJ x
a fI
bg I dy I .
bg I x
a f I x
fI gJ x ,
U ,
b g I x ,
| | |
| |
dy1 dy2
dy2
d2
d1 d2
| | | | | |
d ln x1 x2 x1 3x2 x2 1 dx1 x3 1 dx2 x1
dy1
d1
d x3 x2 1
3x2 x2 dx1 1
|
x3 dx2 , 1
1
~ }|
x2
dx1
dx2 ,
1
x2
dx1
dx2
39
so
I,J
which establishes part (ii). For the proof of property (iii) rst consider a function f on W. Then
f . Next consider a 1-form dz i on W, where z1 , z2 , . . . , so f zk are the variables on Rk . Then di m 1 i dy j , so j y

j
By the chain rule, formula (B.6), the sum m 1 i y j j xl is equal to j i xl . Therefore
Because every form on W is a sum of products of forms of type f and dz i , property (iii) in general follows from the two special cases f and dz i . QED Another application of the chain rule yields the following important result.
3.11. T HEOREM . Let : U V be a smooth map, where U is open in Rn and V is m . Then d k V . In short open in R d for
B j
f x
j 1
i dxl xl 1
P ROOF. First let f be a function. Then
i 1
i 1
df
yi dyi
f I g J dy I dy J
f I dy I
g J dy J
f x
f x
f x f
i dy j y j
j 1
i d j y j
j 1
i y j
j dxl xl 1
l 1

j 1
d i
dzi
d .
f di yi
i 1
f yi
j 1i 1
di1 di2
dik d j1
dy I dy J
B
so
fI gJ
f I g J . Furthermore, dyi1 dyi2

dyik dy j1 dy jl d jl
dy I dy J ,
f x ,
i j y j xl
dxl .
i dx j x j 1
f i dx j . yi x j
40
By the chain rule, formula (B.6), the quantity m 1 f yi i x j is equal to i f x j . Hence

j
because d f I d
d f I . On the other hand,
d
I
f I di1 di2
dik .
Here we have used the Leibniz rule for forms, Proposition 2.5(ii), plus the fact that the form di1 di2 dik is always closed. (See Exercise 2.8.) Comparing the two QED equations above we see that d d . We nish this section by giving an explicit formula for the pullback , which establishes a connection between forms and determinants. Let us do this rst in degrees 1 and 2. The pullback of a 1-form m 1 f i dyi is i
i 1
f i dyi
i 1
f i di .

with g j
i 1
j 1i 1
j 1
m 1 f i x ji . i For a 2-form 1
i j m f i, j dy i dy j
we get
1 i j m
Observe that di d j where
k,l
1 k l n
i j xk xl
i j xl xk
i x k j x k
i x l j x l
i j dxk dxl xk xl 1
f i, j dyi dy j
1 i j m
f i, j di d j .
i j xk xl
i j dxk dxl , xl xk
Now di
n j
i 1 x j m
dx j and so
fi
i dx j x j 1
fi
i dx j x j
g j dx j ,
f I di1 di2
dik
f I dy I
f I di1 di2
dik
f I d di1 di2
d d
d f I dy I
d f I dy I
f I di1 di2
so the theorem is true for functions. Next let so
I f I dy I . Then d
df
f dx j x j 1

d f ,
I d f I dy I , dik ,
dik
41
is the determinant of the 2 2-submatrix obtained from the Jacobi matrix D by extracting rows i and j and columns k and l. So we get
1 i j m
1 k l n1 i j m
1 k l n
1 i j m
To write the product d i1 di2
dik in terms of the x-variables we use
dil
ml
il dxml 1 x m l ik dxm1 dxm2 xmk
for l
1, 2, . . . , k. This gives dik
m 1 ,m 2 ,...,m k 1
xm1
M
i i2 xm2 1
ik dx M , xmk
in which the summation is over all n k multi-indices M m1 , m2 , . . . , mk . If a multi-index M has repeating entries, then dx M 0. If the entries of M are all distinct, we can rearrange them in increasing order by means of a permutation . In other words, we have M m1 , m2 , . . . , mk j 1 , j 2 , . . . , j k , where J j1 , j2 , . . . , jk is an increasing multi-index and S k is a permutation. Thus we can rewrite the sum over all multi-indices M as a double sum over all increasing multi-indices J and all permutations :
1 2
J Sk
det D I,J dx J .
J
J Sk
In (3.2) used the result of Exercise 3.7 and in (3.3) we applied Theorem 3.6. The notation D I,J stands for the I, J-submatrix of D, that is the k k-matrix obtained from the Jacobi matrix by extracting rows i 1 , i2 , . . . , ik and columns j1 , j2 , . . . , jk .
sign
i1 i2 x j 1 x j 2
ik dx J x j k
di1 di1
dik
i1 i2 x j 1 x j 2
ik dx j x j k
dx j
dx j
di1 di2
i1 i2 xm1 xm2
f I dyi1 dyi2
dyik
f I di1 di2
For an arbitrary k-form
I f I dy I we obtain
gk,l
f i, j
i x k j x k
with
i x l j x l
f i, j
i x k j x k
i x k j x l
dxk dxl
gk,l dxk dxl dik . dxmk

k
i x k f i, j j 1 k l n x k
i x k j x l
dxk dxl
(3.2) (3.3)
42
To sum up, we nd
I J J I
This proves the following result.
3.12. T HEOREM . Let : U V be a smooth map, where U is open in Rn and V is m open in R . Let I f I dy I be a k-form on V. Then is the k-form on U given by J g J dx J with
I
This formula is seldom used to calculate pullbacks in practice and you dont need to memorize the details of the proof. It is almost always easier to apply the denition of pullback directly. However, the formula has some important theoretical uses, one of which we record here. 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 If f 1 (constant function) then f 1, so we see that det D x can be interpreted as the ratio between the oriented volumes of two innitesimal blocks positioned at x: one with edges dx 1 , dx2 , . . . , dx n and another with edges d 1 , d2 , . . . , d n . Thus the Jacobi determinant is a measurement of how much the map changes oriented volume from point to point. 3.13. T HEOREM . Let : U 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,
Exercises
3.1. Deduce Theorem 3.7(iv) from Theorem 3.6. 3.2. Calculate the following Theorem 3.7(iv). 1 3 2 1 1 1 4 1 determinants using column and/or row operations and
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. (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 leaves all other numbers xed.) (ii) n, n 1, n 2, . . . , 3, 2, 1 . 3.5. Find all permutations in S n of length 1.
1 5 2 3
1 2 , 3 7
1 0 2 3
1 1 1 1
2 1 1 2
4 3 . 0 5
dy1 dy2
dyn
det D dx1 dx2
dxn .
f dy1 dy2
dyn ,
gJ
f I det D I,J .
det D dx1 dx2
f I det D I,J dx J
f I det D I,J
dx J .
dxn .
EXERCISES
43
3.7. Show that dxi 1 dxi 2
for any multi-index i 1 , i2 , . . . , ik and any permutation in S k . (First show that the identity is true if is a transposition. Then show it is true for an arbitrary permutation by writing as a product 12 l of transpositions and using formula (3.1) and Exercise 3.4(i).)
3.9.
(i) Show that every permutation has the same length and sign as its inverse. (ii) Deduce Theorem 3.7(ii) from Theorem 3.6. 2, . . . , n . 1 simple
3.11. Let be a permutation of 1, 2, . . . , n . The permutation matrix corresponding to is the n n-matrix A whose i-th column is the vector e i . In other words, A ei e i . (i) Write down the permutation matrices for all permutations in S 3 . (ii) Show that A A A . sign . (iii) Show that det A a1,1 0 . . . 0 a1,2 a2,2 . . . an,1 a2,2 . . . an,2 ... ... ... ... ... a1,n a2,n . . . an,n
i.e. all entries below a 11 are 0. Deduce from Theorem 3.6 that
(ii) Deduce from this the expansion rule, Theorem 3.7(iii). 3.13. Show that 1 x1 x2 1 . . . 1 x2 x2 2 . . . ... ... ...
1
for any numbers x 1 , x2 , . . . , xn . (Starting at the bottom, from each row subtract x 1 times the row above it. This creates a new determinant whose rst column is the standard basis vector e1 . Expand on the rst column and note that each column of the remaining determinant has a common factor.)
xn 1
n x2
...
xn n
1 xn x2 n . . .
i j
xj
det A
a1,1
a2,n . . . . an,n
3.12.
(i) Suppose that A has the shape
xi
(i) i2 1 for 1 i n, (ii) ii 1 3 1 for 1 i (iii) i j 2 1 for 1 i, j
n 1, n and i
j.
1, 2, . . . , i 1, i 1, i, i 3.10. The i-th simple permutation is dened by i So i interchanges i and i 1 and leaves all other numbers xed. S n has n permutations, namely 1 , 2 , . . . , n 1 . Prove the Coxeter relations
3.8. Show that for n n! 2 odd permutations.
2 the permutation group S n has n! 2 even permutations and

j
dxi k
sign dxi1 dxi2
dxik
3.6. Calculate 1 , 1 , and , where (i) 3, 6, 1, 2, 5, 4 and 5, 2, 4, 6, 3, 1 ; (ii) 2, 1, 3, 4, 5, . . . , n 1, n and n, 2, 3, . . . , n positions interchanging 1 and 2, resp. 1 and n).
2, n
1, 1 (i.e. the trans-
44
dx,
dy,
dz,
dx dy,
(ii) Find the inverse of the matrix DP3 . as 3.18 (spherical coordinates in n dimensions). In this problem let us write a point in R n r 1 . . .
(This is an example of a recursive denition. 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 . (iii) Let p be the rst 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
1
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

sin n , 0, 0, . . . , 0 ,
r cos n .
cos n DPn ,
DPn
A v
u , w
sin n Pn ,
r sin n
Pn
cos n Pn
r 1 . . . n
r 1 . . .
Let P1 be the function P1 r
r. For each n
1 dene a map Pn
3.17. Let P3
r r cos cos r cos sin be spherical coordinates in R3 . r sin (i) Calculate P3 for the following forms : dx dz, dy dz, dx dy dz.
. Rn
1
1:
3.16. Compute x dy dz in Exercise B.7.
(i) (ii) (iii)
3.15.
x3 1 x1 x2 x2 1 . Find Let x2 x1 x2 2 x3 2 y1 3y2 3y3 y4 ; dy1 , dy2 , dy3 , dy4 ; dy2 dy3 .
3.14. Let
x1 x1 x2 x2 x1 x3 . Find x3 x2 x3 (i) dy1 , dy2 , dy3 ; (ii) y1 y2 y3 , dy1 dy2 ; (iii) dy1 dy2 dy3 .
y dz dx
z dx dy , where is the map R2
R3 dened
Rn
by
EXERCISES
45
(v) Show that det DPn 1 r cosn 1 n 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. Differentiation 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. Denition 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 dened what we mean by a smooth map c : a, b U. The easiest denition is that c should be the restriction of a smooth map c : a , b U dened on a slightly larger open interval.) Let be a 1-form on U. The pullback c is a 1-form on a, b , and can therefore be written as c g dt (where t is the coordinate on R). The integral of over c is now dened by
c a,b
3
a

4.1. E XAMPLE . Let U be the punctured plane R2 0 . Let c : 0, 2 U be cos t, sin t , and let be the angle the usual parametrization of the circle, c t form, y dx x dy . x2 y2 U can be reparametrized by substituting a new variable, A curve c : a, b t p s , where s ranges over another interval a, b . We shall assume p to be a onto a, b satisfying p s one-to-one mapping from a, b 0 for a s b. Such a p is called a reparametrization. The parametrized curve c p : a, b U has the same image as the original curve c, but it is traversed at a different rate. a, b we have either p s Since p s 0 for all s 0 for all s (in which case p is
47
E ('
$ DC
"
Q 8'
$ C
#"
A @
"
"
"
PI
B"
E H'
Then c
' ) $
dt (see Example 3.8), so
2 0
dt
2 .
#"
'
6 %
$98'
&
&
i 1 a
'
' 5'
so
fi c t
i 1
4
i 1
dci t dt. dt
'
'
c
)
c f i dci
c fi
More explicitly, writing in components,
n i
1 fi
'
b
0
g t dt. dxi , we have (4.1)
dci dt, dt
&
"
#" $
!
20 1
"
$ C
('
48
4. INTEGRATION OF 1-FORMS
increasing) or p s 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, 2 R2 dened by c t cos t, sin t represents the unit circle in the plane, traversed at a constant rate (angular velocity) of 1 radian per second. Let p s 2s. Then p maps 0, to 0, 2 and c p, regarded as a map 0, R2 , represents the same circle, but traversed at 2 radians per second. (It is important to restrict the domain of p to the interval 0, . If we allowed s to range over 0, 2 , then cos 2s, sin 2s would traverse the circle twice. This is not considered a reparametrization of the original curve c.) Now s. Then c p : 0, 2 R2 traverses the unit circle in the clockwise let p s direction. This reparametrization reverses the orientation; the angular velocity is now 1 radian per second. Finally let p s 2 s 2 . Then p maps 0, 1 to 0, 2 2 and c p : 0, 1 R 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 independent of the parametrization.
a t` ` ` X U Y Y ` ` Y S edU c Y ` S Y U c rU S S a b` ` a q` fU c Y S Y Y a g` X Y s` a ipchU Y X i S X X V WU y w y w S TR X
4.3. T HEOREM . Let be a 1-form on U and c : a, b p : a, b a, b be a reparametrization. Then

c p
y x ec
U a curve in U. Let
c c
y
if p preserves the orientation, if p reverses the orientation.
P ROOF. It follows from the denition of the integral and from the naturality of pullbacks (Proposition 3.10(iii)) that

c p
c
a, b
a, b
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 n 1 Fi dxi cori n responds a vector eld F i 1 Fi ei , which can be thought of as a force eld acting on the particle. Symbolically we write F dx, where we think of dx as an innitesimal vector tangent to the curve. Thus represents the work done by the force eld along an innitesimal vector dx. From (4.1) we see that c F c t c t dt. Accordingly, the total work done by the force F on the particle during its trip along c is the integral
c c a
U S R U U S S c c
F dx
F c t
c t dt.
S R
U s5U
V R
a B`
c ,
On the other hand, we have c p

c w y
g t dt, so by the substitution formula, Theorem B.7, where the occurs if p 0 and the if p 0. QED
b a
c p
a,b
S R
U 5U
p g
dp ds ds
g p s p s ds.
S eU c
c 8U
Now let us write c p g dp ds ds, so

v U U S S
g dt and t
2v
p s . Then p c
5U
2v
Y ua#`
p c . p g dt p g dp
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 eld F is conservative if it can be written as the gradient of a function, F grad g. The function g is called a potential for the eld and is interpreted as the potential energy of the particle. In terms of forms this means that dg, i.e. is exact. 4.2. Integration of exact 1-forms 4.4. T HEOREM (fundamental theorem of calculus in R n ). Let dg be an exact 1-form on an open subset U of Rn . Let c : a, b U be a parametrized curve. Then
c
h 5h g g h h g g
Integrating an exact 1-form
dg is easy once the function g is known.
g c b
i
where we used the (ordinary) fundamental theorem of calculus, formula (B.1). Hence c g c b g c a . QED The physical interpretation of this result is that when a particle moves in a conservative force eld, its potential energy decreases by the amount of work done by the eld. This claries what it means for a eld 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.
5h h g g #h5h g g n
a,b
mh
b
f
dh
h b
h a ,
jh
jh
P ROOF. By Theorem 3.11 we have c g c t we have c dh, so

kf f i h5h g g
e #d
g c a . c dg dc g. Writing h t c g t
50
It also yields a necessary and sufcient criterion for a 1-form on U to be exact. A curve c : a, b U is called closed if c a c b . 4.5. T HEOREM . Let be a 1-form on an open subset U of R n . Then the following statements are equivalent. (i) is exact. (ii) c 0 for all closed curves c. (iii) c depends only on the endpoints of c for every curve c in U.
t s xs r r w s s5s t u s r t qs t r q#p t o
P ROOF. (i) (ii): if dg and c is closed, then c g c b 0 by the fundamental theorem of calculus, Theorem 4.4. (ii) (iii): assume c 0 for all closed curves c. Let
q #p o #p q
c1 : a1 , b1
o
and
r
c2 : a2 , b2
r t hs
U
s r t Hs r s
be two curves with the same endpoints, i.e. c 1 a1 c2 a2 and c1 b1 c2 b2 . We need to show that c1 . After reparametrizing c 1 and c2 we may c2 assume that a1 a2 0 and b1 b2 1. Dene a new curve c by
} } s r
(First traverse c 1 , then traverse c 2 backwards.) Then c is closed, so c 0. But Theorem 4.3 implies c , so c1 . c1 c2 c2 (iii) (i): assume that, for all c, c depends only on the endpoints of c. We must dene a function g such that dg. Fix a point x 0 in U. For each point x in U choose a curve cx : 0, 1 U which joins x0 to x. Dene
Atgs
g x
r
cx
.
t
We assert that dg is well-dened and equal to . Write n 1 f i dxi . We must i f i . From the denition of partial differentiation, show that g xi
~w
t 1 travel from x0 Now consider a curve c composed of two pieces: for 0 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 t x t 1 hei . Then c has the same endpoints as , and hence cx hei . Therefore cx he c
i
1,2
2
cx
cx
cx
lim
1 h
lim
w
w
t js
g x xi
r
lim
1 h
lim
1 h
1 h
A r
u At
t s
lim
w bs
t js
g x xi
r
g x
hei h
g x
lim
1 0h
cx
hei
cx
l . (4.2)
u At
u ~w
{ |zs t
c t
r
c1 t c2 2
for 0 for 1
t t
1, 2.
u yt
qp
u At
v mt u
v mt
g c a
v#t
4.3. THE GLOBAL ANGLE FUNCTION AND THE WINDING NUMBER
51
j 1
Taking equations (4.2) and (4.3) together we nd

1 0
0 0
This theorem, and its proof, can be used in many different ways. For example, it tells us that once we know a 1-form to be exact we can nd an antiderivative g x by integrating along an arbitrary path running from a xed point x0 to x. (See Exercises 4.34.5 for an application.) On the other hand, the theorem also enables us to detect closed 1-forms that are not exact.
1 R2 4.6. E XAMPLE . The angle form 0 of Example 2.10 is closed, 0. Mark the conbut not exact. Indeed, its integral around the circle is 2 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 definition 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 , will denote the angle form,
y dx x dy , x2 y2

Then is a closed 1-form and and are smooth functions on U. In fact, and are just the components of x x , the unit vector pointing in the direction of x. These functions satisfy d d . (4.4) (You will be asked to check this formula in Exercise 4.6.) Now let : U 0, 2 be the angle between a point and the positive x-axis, chosen to lie in the interval 0, 2 . Then cos and sin , so by equation (4.4) This equation is not valid on all of U (it cannot be because we saw in Example 4.6 that is not exact), but only where is differentiable, i.e. on the complement of the

cos d sin
sin d cos
cos 2 d
sin
d .
and and will denote the functions x , x2 y2

This proves that g is smooth and that dg
0 h
y x2 y2
lim f i x
shei ds
f i x ds
fi x . QED
g x xi
lim
1 0h
h fi x
1 hei dt
lim
fi x
shei ds
fj x
j 1
j 1
1 hei i, jh dt
h fi x
fj x
1 hei dl j
fj x
1 hei l j t dt t 1 hei dt. (4.3)
Let i, j be the Kronecker delta, which is dened by i,i we can write l j t x j i, j t 1 h, and hence l j t

1 and i, j 0 if i j. Then i, j h. This shows that
52
positive x-axis. Hence the nonexactness of is closely related to the impossibility of dening 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 dene a continuous angle function, and the fact that d almost everywhere suggests how: by integrating along c! For simplicity assume that a 0 and b 1. Start by xing any 0 such that cos 0 c 0 and sin 0 c 0 and then dene
0,t
The following result says that t measures the angle between c t and the positive x-axis (up to an integer multiple of 2 ) and that the function : 0, 1 R is smooth. In this sense is a differentiable choice of angle along the curve c. 4.7. T HEOREM . The function is smooth and satises P ROOF. To see that 0 0 , plug t 0 into the denition of . To prove the other assertions we rescale the curve c t to a new curve c t c t moving on the unit circle. Let f t and g t be the x- and y-components of this new curve. Then f t c t and g t c t and for all t. In other words f c and g c f dg g d f . Therefore
5 z 5 j j
0 ,
cos t
B
c t
and

sin t
c t .
5
c , so it follows from formula (4.4) that

By the fundamental theorem of calculus, formula (B.2), is differentiable and Since the right-hand side is smooth, is smooth as well. To prove that cos t f t and sin t g t for all t it is enough to show that the difference vector

fg
gf .
f t g t f2

cos t sin t
has length 0. Its length is equal to

Furthermore, the derivative of u is f g f

f ff

gg cos
g gg
ff
sin .
f2f
f gg cos
g2 g
fgf
sin
f gg cos
f 2g
f cos
f sin
g sin
g cos fgf sin by formula (4.5) since f 2 g2 1
u 0
f 0 cos 0
g 0 sin 0
cos 2 0
Hence we need to show that the function u to 1. For t 0 we have
f cos
g sin is a constant equal sin 2 0 1.
2 f cos
cos
sin
g2
2 f cos
g sin
cos 2
sin 2 g sin .
f s g s
f t

g t
g s f s
ds. (4.5)
c .
EXERCISES
53
As t increases from 0 to 1, the dial starts at the angle 0 0 , it moves around the meter, and ends up at the nal angle 1 . The difference 1 0 measures the total angle swept out by the dial.
z
4.8. C OROLLARY. If c : 0, 1 where k is an integer.

U is a closed curve, then 1 c 1 implies

55 5 55
The integer k 2 1 c is called the winding number of the closed curve c about the origin. It measures how many times the curve loops around the origin.
winding number of a closed curve about origin
Exercises
Aq
R2 dened by c t 4.1. Consider the curve c : 0, 2 and b are positive constants. Let xy dx x 2 y dy.
a cos t, b sin t
T,
where a
(i) Sketch the curve c for a 2 and b (ii) Find c (for arbitrary a and b).

1.
4.2. Restate Theorem 4.5 in terms of force elds, potentials and energy. Explain why the result is plausible on physical grounds.
4.9. E XAMPLE . By Example 4.1, the winding number of the circle c t (0 t 2 ) is equal to 1.

1 2
cos 1 and sin 0 In other words cos 0 by an integer multiple of 2 .

sin 1 , so 0 and 1 differ QED
cos t, sin t
cos 0 , sin 0

c 0 , c 0
P ROOF. By Theorem 4.7, c 0

c 1 , c 1
cos 1 , sin 1 .
It is useful to think of the vector f t , g t in the same direction as the vector c t .

Now f 2 g2 1 implies f f function, so u t 1 for all t.

gg
0, so u t
0 for all t. Hence u is a constant QED c t c t as a dial that points
2 k,
(4.6)
T
54
4.3. Consider the 1-form x a n 1 xi dxi on Rn 0 , where a is a real constant. i For every x 0 let cx be the line segment starting at the origin and ending at x.
W
1 Rn 4.4. Let 0 be the 1-form of Exercise 4.3. Now let c x be the haline pointing from x radially outward to innity. Parametrize c x by travelling from innity , 0 in such a way that inward to x. (You can do this by using an innite time interval cx 0 x.)
(i) Determine for which values of a the function g x cx is well-dened and compute it. (ii) For the values of a you found in part (i) check that dg . (iii) Show how to recover from this computation the potential energy for Newtons gravitational force. (See Exercise B.4.)
t W
1 Rn 4.5. Let 0 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 nd a smooth function g on R n 0 such that dg .
4.6. (i) Verify equation (4.4). (ii) Let U R2 0 . Prove that there does not exist a smooth function : U R satisfying cos x, y x x2 y2 and sin x, y y x2 y2 for all x, y U. (Argue by contradiction, by letting y dx x dy x 2 y2 and showing that d if was such a function.) 4.7. Calculate directly from the denition the winding number about the origin of the curve c : 0, 2 R2 given by c t cos kt, sin kt 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 dene the winding number of c around x 0 ? Try to formulate two different denitions: a geometric denition and a denition in terms of an integral over c of a certain 1-form analogous to formula (4.6). R2 0 be a closed curve with winding number k. Determine the 4.9. Let c : 0, 1 winding numbers of the following curves c : 0, 1 R2 0 by using the formula, and then explain the answer by appealing to geometric intuition.
# # # f m
4.10. For each of the following closed curves c : 0, 2 R2 0 set up the integral dening the winding number about the origin. Evaluate the integral if you can (but dont give up too soon). If not, sketch the curve (the use of software is allowed) and obtain the answer geometrically.
# #
A#
(i) (ii) (iii) (iv)
c c c c
t t t t
a cos t, b sin t T , where a 0 and b 0; cos t 2, sin t T ; cos3 t, sin3 t T ; a cos t b cos t b a 2, a cos t b sin t
, where 0
(v) c t
c t , where x, y
(i) (ii) (iii) (iv)
c c c c
t t t t
c 1 t ; t c t , where : 0, 1 1c t ; c t c t , where x, y
0,
is a function satisfying 0
T
1 ;
y, x T ; 1 x, x2 y2
a.
(i) Show that is closed for any value of a. (ii) Determine for which values of a the function g x compute it. (iii) For the values of a you found in part (ii) check that dg

cx
is well-dened and
EXERCISES
55
(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 dening the winding number of c around the origin and evaluate it. If you get stuck, nd the answer geometrically. 4.12. Let U be an open subset of R2 and let F vector eld. The differential form
0. Let c be a parametrized circle contained is well-dened at all points x of U where F x 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
Prove the following assertions.
4.13.
circles.
(i) Find the indices of the following vector elds around the indicated
1 ) 20

(i) (ii) (iii) (iv)
F , where is the angle form y dx x dy x2 y2 ; is closed; index F, c is the winding number of the curve F c about the origin; index F, c is an integer.
'
index F, c
1 2
&
%
F1 dF2 2 F1
F2 dF1 2 F2
F1 e1

$
" !

c t
b cos t
a cos
t, a
b sin t
a sin
F2 e2 : U

R2
b a
R2 be a smooth
4.11. Let b 0 by
0 and a
0 be constants with a
b. Dene a planar curve c : 0, 2
56
(ii) Draw diagrams of three vector elds 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 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
where ai bi . The k-dimensional analogue of a parametrized path is a smooth map c : R U. Although the image c R may look very different from the block R, we think of the map c as a parametrization of the subset c R of U: each choice of a point t in R gives rise to a point c t in c R . The pullback c is a k-form on R and therefore looks like g t dt 1 dt2 dtk for some function g : R R. The integral of over c is dened as
c R ak a2 a1
For k 1 this reproduces the denition given in Chapter 4. (The denition 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.) 0 is also worth examining. A zero-dimensional block R in The case k R0 0 is just the point 0. We can therefore think of a map c : R U as a collection x consisting of a single point x c 0 in U. The integral of a 0-form (function) f over c is by denition the value of f at x,
c
As in the one-dimensional case, integrals of k-forms are almost wholly unaffected by a change of variables. Let be a second rectangular block. A reparametrization is a map p : R R satisfying the following conditions: p is bijective (i.e. one-to-one and onto) and the k k matrix Dp s is invertible for all s R. Then det Dp s 0 for all s R, so either det Dp s 0 for all s or det Dp s 0 for all s. In these cases we say that the reparametrizion preserves, respectively reverses the orientation of c.
57
X5 YR Q
9 A A 9 6 WCCAB28
I R Q E
9 6 28
6 V5
a1 , b1
a2 , b2
R Q
f x .
ak , bk
A A CCA
R Q
R Q
A A CCA
bk
b2
b1
g t dt1 dt2
dtk .
R Q
R Q
A A CCA R Q
D 5 8 R Q
R Q
A A A 9 6 9CCB@8 T
9 6 @8 5 T
H D
` aR Q R Q
H U5 D
75 6 5
a1 , b1
a2 , b2
ak , bk
Rk a i
ti
bi
for 1
k ,
58
5. INTEGRATION AND STOKES THEOREM
c p
P ROOF. Almost verbatim the same proof as for k 1 (Theorem 4.3). It follows from the denition of the integral and from the naturality of pullbacks, Proposition 3.10(iii), that
by Theorem 3.13, so
5.2. E XAMPLE . The unit interval is the interval 0, 1 in the real line. Any curve c : a, b U can be reparametrized to a curve c p : 0, 1 U by means of the reparametrization p s b a s a. Similarly, the unit cube in R k is the rectangular block
(Squeeze the unit cube until it has the same edgelengths as R and then move it to A, so det Dp s the position of R.) Then p is one-to-one and onto and Dp s det A vol R 0 for all s, so p is an orientation-preserving reparametrization. Hence c p c for any k-form on U. 5.3. R EMARK . A useful fact you learned in calculus is that one may interchange the order of integration in a multiple integral, as in the formula
a1 a2 a2 a1
(This follows for instance from the substitution formula, Theorem B.7.) On the other hand, we have also learned that f t 1 , t2 dt2 dt1 f t1 , t2 dt1 dt2 . How can this be squared with formula (5.1)? The explanation is as follows. Let f t1 , t2 dt1 dt2 . Then the left-hand side of formula (5.1) is the integral of over
f t1 , t2 dt1 dt2
b1
b2
b2
b1
f t1 , t2 dt2 dt1 .
(5.1)
e vr
e fr
...
bk
ak
ak
Ue
0 . . .
b2
. . . 0
a2
and
b1
a1
... ... .. .
0 0 . . .
a1 a2 . . .
e Yr
Let R be any other block, given by a i As a, where
ti
bi . Dene p : 0, 1
0, 1
Rk t i
0, 1
for 1
k .
k
R by p s
y r
u u CCu
e r
d h
On the other hand, c Theorem B.7, we have c if det Dp 0.
dtk , so by the substitution formula, g t dt1 dt2 , where the occurs if det Dp 0 and the c QED
w p xe h r p p
c p
u u CCu
r tr
g p s
det Dp s ds 1 ds2
dsk .
u u CCu
r s
e vr
u u CCu
e vr s
p c
p g dt1 dt2
dtk
p g det Dp ds1 ds2
u u CCu
Now let us write c
g dt1 dt2
dtk and t
p s . Then dsk
c p
r s
s tr
p c .
h i ge f
c c
if p preserves the orientation, if p reverses the orientation.
5.1. T HEOREM . Let be a k-form on U and c : R be a reparametrization. Then
U a smooth map. Let p : R
c p e s d h b r e y p i
5.2. THE BOUNDARY OF A CHAIN
59
c : a1 , b1 a2 , b2 R2 , the parametrization of the rectangle given by c t 1 , t2 t1 , t2 . The right-hand side is the integral of not over c, but over c p, where is the reparametrization p s 1 , s2 s2 , s1 . Since p reverses the orientation, Theorem 5.1 says that c p ; in other words c , which is exactly c c p formula (5.1). Analogously we have
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 U is called a k-cube usually take R to be the unit cube. A smooth map c : 0, 1 k 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, where a1 , a2 , . . . , a p are real coefcients and c 1 , c2 , . . . , c p are k-cubes. For any k-form we then dene
(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 dened 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 i 1 a i 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 denition
c p xi
Likewise, a k-chain i 1 ai ci can be pictured as a charge distribution, with an electric charge ai spread along the k-dimensional patch c i .
5.2. The boundary of a chain
c 0
h j
Consider a curve (1-cube) c : 0, 1 0-chain dened by c c 1 c 0 .
U. Its boundary is by denition the c 1
i 1
i 1
m l m l p g
ai
ai f
xi .
i 1
ai
}|CC~} z z z
a1 c1
a2 c2
apcp,
ci
j h
0,1
0,1
z z CCz
{ z z |CCz
z z CCz
f t1 , t2 , . . . , tk dt1 dt2
dtk
f t1 , t2 , . . . , tk dti dt1 dt2
p ul s r n
i g 2h
j 2h
v w
n l vm i g @h
p : a2 , b2
a1 , b1
a1 , b1
a2 , b2
dti
dtk
n om
r tp
kh j
s r
g i2h
w xv
60
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 dene c c 1 c2 c3 c4 . 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,1 t c t1 , t2 , . . . , ti
1 , 1, t i , . . . , t k 1
(Insert 0, resp. 1 in the i-th slot.) Now dene
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 denition 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.)
P ROOF. By linearity of it sufces to prove this for k-cubes c : 0, 1 t1 , t2 , . . . , tk 2 0, 1 k 2 and let and be 0 or 1. Then for 1 i
5.4. P ROPOSITION . c
0 for every k-chain c in U. In short, 2 0.

k
i 1
i 1 0,1
ci,0
ci,1
f f
ci,0 t
c t1 , t2 , . . . , ti
1 , 0, t i , . . . , t k 1
, .
ci, .
U. Let k 1
(Alternatively we could dene c c 1 c2 c3 c4 , with c3 t c 0, 1 t , which corresponds to the following picture: c4 t
c 1
t, 1 and
5.3. CYCLES AND BOUNDARIES
61
because in the vector t 1 , t2 , . . . , ti 1 , , ti , . . . , tk 2 the entry t j occupies the j 1st slot! We conclude that c i, j, c j 1, i, for 1 i j k 1. Therefore the k 2-chain c is given by
i 1 j 1 , 0,1
1 j i k , 0,1
1 s r k , 0,1
Thus the two terms on the right in (5.2) cancel out.
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 degenerate 1-cube is a constant curve. The work done by a force eld on a motionless particle is 0. More generally we have the following.
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
c t1 , . . . , 0, . . . , t k
g f t1 , . . . , ti , . . . , tk ,
5.5. L EMMA . Let be a k-form and c a degenerate k-chain. Then
0.
s r
1 s r k , 0,1
cr,
s r 1
1 i j k 1 , 0,1
1 i j k 1 , 0,1
cr,
s, s, .
QED
j,
i j
ci,
i j
In the rst term on the right in (5.2) we substitute c i, r j 1, s i, , to get
j,
cj
1, i,
cj
i j
1 i j k 1 , 0,1
i j
ci,
j,
The double sum over i and j can be rearranged in a sum over i i j to give
k k 1
i j
i 1 0,1
i 1 0,1
ci,
j, .
ci,
ci,
j and a sum over
ci,
j, .
and then
1, i,
c t1 , t2 , . . . , ti
1 , , ti , . . . , t j 1, , t j , . . . , tk 2
1, i,
cj
t1 , t2 , . . . , tk
cj
1,
On the other hand,
t1 , t2 , . . . , ti
1 , , ti , . . . , tk 2
c t1 , t2 , . . . , ti
1 , , ti , . . . , t j 1 , , t j , . . . , tk 2
j,
ci,
t1 , t2 , . . . , tk
ci, t1 , t2 , . . . , t j
we have
1, , t j , . . . , tk 2
(5.2)
62
So degenerate chains are irrelevant where integration is concerned. This motivates the following denition. A k-chain c is closed, or a cycle, if c is a degenerate k 1-chain. A k-chain c is a boundary if c b c for some k 1-chain b and some degenerate k-chain c . 5.6. E XAMPLE . If c 1 and c2 are curves arranged head to tail as in the picture below, then c1 c2 is a 1-cycle. Likewise, the closed curve c is a 1-cycle.
c2 c c1
P ROOF. By linearity it sufces to consider the case of a degenerate k-cube c. Suppose c is constant as a function of t i . Then ci,0 ci,1, so
j i
Let t t1 , t2 , . . . , tk 1 . For j i the cubes c j,0 t and c j,1 t are independent of ti and for j i they are independent of t i 1 . So c is a combination of degenerate k 1-cubes and hence is degenerate. QED 5.8. C OROLLARY. Every boundary is a cycle. P ROOF. Suppose c b c with c degenerate. Then by Lemma 5.5 c b c c , where we used Proposition 5.4. Lemma 5.7 says that c is degenerate, and therefore so is c. QED 5.9. E XAMPLE . Consider the unit circle in the plane c t cos 2 t, sin 2 t with 0 t 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 2cube. This is indeed true in the sense dened above, that c b c where c is a constant 1-chain. (It is actually not possible to nd a b such that c b; see Exercise 5.2.) The 2-cube b is dened by shrinking c to a point, b t 1 , t2 1 t2 c t1 for t1 , t2 in the unit square. Then so that b c c , where c is the constant curve located at the origin. Therefore c b c , 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.
b t1 , 0
c t1 ,
b 0, t2
b 1, t2
t2 , 0 ,
b t1 , 1
0, 0 ,
c j,0
c j,1 .
5.7. L EMMA . The boundary of a degenerate k-chain is a degenerate k
Now g is a k-form on 0, 1 k We conclude c 0,1 k c
and hence equal to 0, and so c 0.
g s1 , . . . , sk
c s1 , . . . , si
1 , 0, s i 1 , . . . , t k 1
f f
f t1 , . . . , ti , . . . , tk
t1 , . . . , ti , . . . , tk , . f g 0. QED
where f : 0, 1
0, 1
k 1
and g : 0, 1
k 1
U are given respectively by
1-chain.
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:
c c 1 c 0 c 1 c 0
i.e. c dg c g. This is the form in which the fundamental theorem of calculus generalizes to higher dimensions. This generalization is perhaps the neatest relationship between the exterior derivative and the boundary operator. It contains as special cases the classical integration formulas of vector calculus (Green, Gau 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.
P ROOF. By the denition of the integral and by Theorem 3.11 we have

c 0,1
k
i 1
for certain functions g 1 , g2 , . . . , gk dened on 0, 1 k . Therefore
Changing the order of integration (cf. Remark 5.3) and subsequently applying the fundamental theorem of calculus in one variable, formula (B.1), gives
The forms
CC
|CC
gi t1 , . . . , ti
1 , 0, t i 1 , . . . , t k
dt1 dt2
dti
CC
CC
gi t1 , . . . , ti
1 , 1, t i 1 , . . . , t k
dt1 dt2
dti
dtk dtk
and
CC
|CC
0,1
k 1
gi t1 , . . . , ti
1 , 0, t i 1 , . . . , t k
dt1 dt2
gi t1 , . . . , ti
0,1
0,1
1 , 1, t i 1 , . . . , t k
CC
|CC
dtk
CC
gi dt1 dt2 ti
gi dt dt1 dt2 ti i
dti
dtk
i 1
0,1
i 1
0,1
dti
CC
CC
|CC
d gi dt1 dt2
dti
dtk
i 1
CC
|CC
gi dt1 dt2
Since c is a k
1-form on 0, 1 k , it can be written as

k
dti
0,1
dtk
c d
5.10. T HEOREM (Stokes theorem). Let be a k Rn and let c be a k-chain in U. Then
1-form on an open subset U of
dc .
gi dt1 dt2 ti
kt
dg
g c 1
g c 0
g,
dtk .
dtk .
64
are nothing but ci,1 , resp. ci,0 . Accordingly,

c 0,1
k
which proves the result.
5.11. E XAMPLE . The unit circle c t cos 2 t, sin 2 t is a 1-cycle in the punctured plane U R2 0 . Considered as a chain in R2 it is also a boundary, as we saw in Example 5.9. 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 both contained in U such that c b c . Indeed, suppose that c b c . Let y dx x dy x2 y2 be the angle form. Then c 2 by Example 4.1.On the other hand, where we have used Stokes theorem, Lemma 5.5 and the fact that 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 (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 : U map. Let c be a k-cube in U and a k-form on V. Prove that c c . V a smooth
and check that they are equal.
5.5. Using polar coordinates in n dimensions (cf. Exercise 3.18) write the n 1-dimensional unit sphere S n 1 in Rn as the image of an n 1-cube c.For n 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
5.4. Dene a 3-cube c : 0, 1 x1 dx2 dx3 . Calculate both c d and
R3 by c t1 , t2 , t3 t2 t3 , t1 t3 , t1 t2 , and let and check that they are equal. c
5.3. Dene a 2-cube c : 0, 1 2 x1 dx3 x2 dx3 . (i) Sketch the image of c. (ii) Calculate both c d and
R3 by c t1 , t2
t2 , t1 t2 , t2 , and let 1 2
5.2. Let U be an open subset of Rn . Its boundary is a linear combination of k 1-cubes, i ai ci . (i) Let b be a k 1-chain in U. Its boundary is a linear combination of k-cubes, b i ai ci . Prove that i ai 0. (ii) Let c be a k-cube in U. Conclude that there exists no k 1-chain b in U satisfying b c.
b c
0,
x 1 dx2
i 1 0,1
c i,
i 1 0,1
0,1
k 1
ci,
i 1
0,1
k 1
,
QED
i 1
i 1
ci,1
ci,0
CC
CC
i 1
gi dt dt1 dt2 ti i
dti
dtk
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 elds, chains etc. are smooth and are dened in an open subset U of Rn . (Some formulas hold only for special values of n, as indicated.) g f dx dy x y and any 2-chain c. (Here n 2.)
c c c
(iii) Gau formula: any n-chain c.
div F dx1 dx2 curl F dx
dxn
dx for any vector eld F and
(iv) Stokes formula:
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.)
F dx for any vector eld F and any 2-chain
(ii) Greens formula:
! ! ! "!

f dx
(i)
grad g dx
g c 1
g c 0
for any function g and any curve c. g dy for any functions f , g
CHAPTER 6
Manifolds
6.1. The denition 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 distortions. The formal denition is given below and is unfortunately a bit long. It will help to consider rst the basic example of the surface of the earth, which is a twodimensional 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 earths surface is not entirely faithful, because charting a curved surface on a at 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 overlap near their edges and the totality of all maps in a world atlas covers the whole world. An arbitrary manifold is dened similarly, as an n-dimensional world represented 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 : U RN satisfying the following conditions:
The image of the embedding is the set U t t U consisting of all points of the form t with t U. The inverse map 1 is called a chart or coordinate map. You should think of U as an n-dimensional patch in R N parametrized by the map . Condition (i) means that to distinct values of the parameter t must correspond distinct points t in the patch U . Thus the patch U has no self-intersections. Condition (ii) means that for each t in U all n columns of the Jacobi matrix D t must be independent. This is imposed to prevent the occurrence of cusps and other singularities in the image U . Since n: the target space R N D t has N rows, this condition also implies that N must have dimension greater than or equal to that of the source space U, or else cannot be an embedding. The column space of D t is called the tangent space to the patch at the point x t and is denoted by Tx U ,
& % & % 8& % ' & %& &% ' % 7 & % &% & % &% & % &% 04 2 ' ) 6 ) 5& % 31& % $& % 0 & % ) '(& %
Tx U D t R n .
67
'
(i) is one-to-one (i.e. if t1 t2 , then t1 (ii) D t is one-to-one for all t U; (iii) the inverse of , which is a map 1 : U
&%
& %
t2 );
U, is continuous.
&%
68
6. MANIFOLDS
The tangent space at each point is an n-dimensional subspace of R N because D 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 ti exists and is equal to t for some t U, then lim i ti t. This is intended to avoid situations where the image U doubles back on itself at innity. (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. D t e1
U t e2 e1 e3 D t e2 e2
Since t is an n-vector and f t an m-vector, the graph is a subset of R N with N n m. We claim that the graph is the image of an embedding : U R N . Dene
t f t
Then by denition graph f U . Furthermore is an embedding. Indeed, t1 t2 implies t1 t2 , so is one-to-one. Also,
D t
In Df t
S`@ 9 YE @ 9 Q E @ 9 E S@ 9 YE @ 9 Q @9 U UUU I W D VTS@ 9 QRPE
graph f
t f t
U .
6.3. E XAMPLE . Let U be an open subset of Rn and let f : U map. The graph of f is the collection
Rm be a smooth
G F
e1
GF
@9
@9
@ 9 BCA
GF
GF
E CA B
@ 9 E@ 9
@D 9
6.1. THE DEFINITION
69
so D t has n independent columns. Finally the inverse of is given by
which is continuous. Hence is an embedding. A manifold is an object patched together out of the images of several embeddings. More precisely, 6.4. D EFINITION . An n-dimensional manifold1 (or n-manifold for short) in R N is a subset M of R N such that for all x M there exist an open subset V R N containing x, an open subset U Rn , and an embedding : U R N satisfying U V M. N is N The codimension of M in R n. Choose t U such that t x. Then the tangent space to M at x is the column space of D t ,
(Using the chain rule one can show that Tx M is independent of the choice of the embedding .) The elements of Tx M are tangent vectors to M at x. A collection of embeddings i : Ui R N with Ui open in Rn and such that M is the union of all the sets i Ui is an atlas for M. One-dimensional manifolds are called (smooth) curves, two-dimensional manifolds (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 denition. Here M is a curve in the plane, so we have N 2 and n 1. U is an open interval in R and V is an open disc in R2 . The map sends t to x and parametrizes the portion of the curve inside V. Since n 1, the Jacobi matrix D t consists of a single column vector, which is tangent to the curve at x t . The tangent line Tx M is the line spanned by this vector.
U t
Tx M
Sometimes a manifold has an atlas consisting of one single chart. In that event we can take V R N , and choose one open U Rn and an embedding : U RN such that M 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 dene 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 sufciently high dimen-
Tx M
D t R n .
b tab a b ah fba r fb a
fgTeb a d c
1
f ba ba
pp
b a ff
b a
ba f ii i
t f t
t,
70
6. MANIFOLDS
6.5. E XAMPLE . An open subset U of Rn can be regarded as a manifold of dimension n (hence of codimension 0). Indeed, U is the image of the map : U Rn given by x x, the identity map. The tangent space to U at any point is R n itself. 6.6. E XAMPLE . Let N n and dene : Rn R N by x1 , x2 , . . . , xn x1 , x2 , . . . , xn , 0, 0, . . . , 0 . It is easy to check that is an embedding. Hence the image Rn is an n-manifold in R N . (Note that Rn is just a linear subspace isomorphic to Rn ; e.g. if N 3 and n 2 it is just the xy-plane. We shall usually n with its image in R N .) Combining this example with the previous identify R one, we see that if U is any open subset of Rn , then U is a manifold in R N of codimension N n. Its tangent space at any point is Rn . graph f , where f : U Rm is a smooth map. As 6.7. E XAMPLE . Let M shown in Example 6.3, M is the image of a single embedding : U R n m , so n m , covered by a single chart. At a point M is an n-dimensional manifold in R x, f x in the graph the tangent space is spanned by the columns of D. For instance, if n m 1, M is one-dimensional and the tangent line to M at x, f x 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
For n 2 and m 1, M is a surface in R3 . The tangent plane to M at a point x, y, f x, y is spanned by the columns of D x, y , namely 1 x, y 0
The diagram below shows the graph of f x, y x 3 y3 3xy from two different angles, together with a few points and tangent vectors. (To improve the scale the
x w
f x
and
d x w y x w x w
f x
f y
1 0 x, y
1 f x
xx w w
y x
x w
x w v
x w
x x w w
x y
y y
y (x w
xx w y
x w
xx w w
6.1. THE DEFINITION
71
z-coordinate and the tangent vectors have been compressed by a factor of 2.)
e3
e3 e2 e1
e1
e2
R2 given by t et cos t, sin t . 6.8. E XAMPLE . Consider the path : R e t . Therefore Let us check that is an embedding. Observe rst that t t1 t 2 . The exponential function is one-to-one, so t t1 t2 implies e e t2 . 1 This shows that is one-to-one. The velocity vector is
0 if and only if cos t sin t 0, which is impossible because Therefore t cos2 t sin2 t 1. So t 0 for all t. Moreover we have t ln e t ln t . Hence the inverse of is given by 1 x ln x for x R and so is continuous. Therefore is an embedding and R is a 1-manifold. The image R is a converges to the origin. It winds innitely many times spiral, which for t around the origin, although that is hard to see in the picture. y
~ }z{g f |
Even though R is a manifold, the set R singularity at the origin!
0 is not: it has a very nasty
g f j kg f j h
g f v
et
j j g lf g f u h
pq
cos t cos t
sin t . sin t
hj 1kg f j f ig f h
h o lg nfm
th Cg mf
xp ywe
g f
h hsg mf
g f lg f h q
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 : U Rn be the map 1 t 2t t 2 1 en 2 t 1 given in Exercise B.7. As we saw in that exercise, the image of is the punctured sphere M en , so if we let V be the open set Rn en , then U M V. Also we saw that has a two-sided inverse : U U, the stereographic projection from the north pole. Therefore is one-to-one and its inverse is continuous (indeed, differentiable). Moreover, t t implies D t D t v v for all v in Rn 1 by the chain rule. Therefore, if v is in the nullspace of D t , v Thus we see that 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 nishes the proof that M is an n 1-manifold in R As this example shows, the denition 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 denition alone. Fortunately there exists a more manageable criterion for a set to be a manifold. 6.2. The regular value theorem Denition 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 nd in practice. Instead, manifolds are often given implicitly, by a system of m equations in N unknowns,
2 x1 , . . . , x N
Here the i s are smooth functions presumed to be dened on some common open subset U of R N . Writing in the usual way x1 x2 . . .
we can abbreviate this system to a single equation
For a xed vector c
xN
m x
c.
1 x 2 x . . .
cm
c1 c2 . . .
x
x
R we denote the solution set by

1
U x
( l ( 3 (
m x1 , . . . , x N

1 x1 , . . . , x N
c1 , c2 , . . . cm .
8
D t D t v D t 0 0.
6.2. THE REGULAR VALUE THEOREM
73
and call it the level set or the bre of at c. (The notation 1 c for the solution set is standard, but a bit unfortunate because it suggests falsely that is invertible, which it is usually not.) If is a linear map, the system of equations is inhomogeneous linear and by linear algebra the solution set is an afne subspace of R N . The dimension of this afne subspace is N m, provided that has rank m (i.e. has m independent columns). We can generalize this idea to nonlinear equations as follows. We say that c Rm is a regular value of if the Jacobi matrix N m has rank m for all x D x : R R 1 c . A vector that is not a regular value is called a singular value. (As an extreme, though slightly silly, special case, if 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. (Dont get carried away though, because it does not apply to every possible manifold. In other words, it is a sufcient but not a necessary criterion.) The proof uses the following important fact from linear algebra,
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 : U R be a smooth map. Suppose that c is a regular value of and that M 1 c is N of codimension m. Its tangent space at x is the nonempty. Then M is a manifold in R nullspace of D x ,
m
P ROOF. Let x M. Then D 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 D x . Let us put n N m. Identify R N with Rn Rm and correspondingly 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 Dv u0 , v0 . This matrix being invertible, by the implicit function theorem, Theorem B.4, there exist open neighbourhoods U of u0 in Rn and V of v0 in Rm such that for each u U there exists a unique v f u V satisfying u, f u c. The map f : U V is C . In other words U V graph f is the graph of a smooth map. We conclude from ExamM ple 6.7 that M U V is an n-manifold, namely the image of the embedding : U RN given by 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 u c, a constant, for all u U. Hence D u D u 0 by the chain rule. Plugging in u u0 gives D x D u0 0. The tangent space Tx M is by denition the column space of D u0 , so every tangent vector v to M at x is of the form v D u0 a for some a Rn . Therefore D x v D x D u0 a 0, i.e. Tx M ker D x . The tangent space Tx M
l ( l
Tx M
ker D x .
nullity A
rank A
l,
74
6. MANIFOLDS
is n-dimensional (because the n columns of D u0 are independent) and so is the nullspace of D x (because nullity D x N m n). Hence Tx M ker D x . QED The case of one single equation (m 1) is especially important. Then D is a single row vector and its transpose is the gradient of : D T grad . It has rank 1 at x if and only if it is nonzero, i.e. at least one of the partials of does not vanish at x. The solution set of a scalar equation 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 : U R 1 c is nonempty and that grad x be a smooth function. Suppose that M 0 for all x in M. Then M is a manifold in R N of codimension 1. Its tangent space at x is the orthogonal complement of grad x ,
6.12. E XAMPLE . Let U R2 and x, y xy. The level curves of are y, x T . The diagram hyperbolas in the plane and the gradient is grad x below shows a few level curves as well as the gradient vector eld, which as you can see is perpendicular to the level curves.
0 is the only singular value of The gradient vanishes only at the origin, so 0 . By Corollary 6.11 this means that 1 c is a 1-manifold for c 0. The bre 1 0 is the union of the two coordinate axes, which has a self-intersection and so is not a manifold. However, the set 1 0 0 is a 1-manifold since the gradient is nonzero outside the origin. Think of this diagram as a topographical map representing the surface z x, y shown below. The level curves of 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
"
Tx M
grad x
6.2. THE REGULAR VALUE THEOREM
75
has a mountain pass or saddle point.
e3
e2 e1
6.13. E XAMPLE . A more interesting example of an equation in two variables x3 y3 3xy c. Here grad x 3 x 2 y, y2 x T , so grad is x, y T . The corresponding values of are 0, resp. vanishes at the origin and at 1, 1 1, which are the singular values of . y
The level curve 1 1 is not a curve at all, but consists of the single point 1, 1 T . Here has a minimum and the surface z x, y has a valley. The level curve 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. 6.14. E XAMPLE . Let U R N and x x 2 . Then grad x 2x, so as in Example 6.12 grad vanishes only at the origin 0, which is contained in 1 0 . So again any c 0 is a regular value of . Clearly, 1 c is empty for c 0. For c 0, 1 c is an N 1-manifold, the sphere of radius c in R N . The tangent
76
6. MANIFOLDS
Finally, 0 is a singular value (the absolute minimum) of and 1 0 0 is not an N 1-manifold. (It happens to be a 0-manifold, though, just like the singular bre 1 1 in Example 6.13. So if c is a singular value, you cannot be certain that 1 c is not a manifold. However, even if a singular bre happens to be a manifold, it is often of the wrong dimension.)
6.15. E XAMPLE . Dene : R4
R2 by
Then D x
If x1 0 the rst and third columns of D x are independent, and if x 2 0 the x2 0, second and fourth columns are independent. On the other hand, if x 1 D x has rank 1 and x 0. This shows that the origin 0 in R 2 is the only singular value of . Therefore, by the regular value theorem, for every nonzero vector c the set 1 c is a two-manifold in R4 . For instance, M 1 1 is a 0 1, 0, 0, 0 T. Let us nd a basis two-manifold. Note that M contains the point x of the tangent space Tx M. Again by the regular value theorem, this tangent space is equal to the nullspace of
We now come to a more sophisticated example of a manifold determined by a large system of equations. 6.16. E XAMPLE . Recall that an n n-matrix A is orthogonal if A T A I. This means that the columns (and also the rows) of A are perpendicular to one another 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 matrices form a group under matrix multiplication, which is usually called the orthogonal group and denoted by O n . Let us prove using the regular value theorem that O n is a manifold. First observe that that A T A T A T A, so A T A is a symmetric n n is the vector space of all n R n-matrices and matrix. In other words, if V W C V C C T the linear subspace of all symmetric matrices, then
denes a map : V W. Clearly O n 1 In , so to prove that O n is a manifold it sufces to show that I is a regular value of . The derivative of can
AT A
which is equal the set of all vectors y satisfying y 1 y3 therefore given by the standard basis vectors e2 and e4 .
D x
2 0
0 0
0 1
0 , 0 0. A basis of Tx M is
2x1 x3
2x2 x4
0 x1
x2 1 x1 x3
x2 2 . x2 x4 0 . x2
Here is an example of a manifold given by two equations (m
2).
space to the sphere at x is the set of all vectors perpendicular to grad x In other words, Tx M x y RN y x 0 .
2x.
} Yl
EXERCISES
77
be computed by using the formula derived in Exercise B.3:
We need to show that for A O n the linear map D A : V W has rank equal to the dimension of W. By linear algebra this amounts to showing that the equation BA T AB T C (6.1) 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 C T and rst try to solve BA T 2 C 2 C. 1 TA Left multiplying both sides by A and using A I gives B 2 CA. It is now 1 easy to check that B CA is a solution of equation (6.1). 2 Exercises
R 2 by t t sin t, 1 6.1. This is a continuation of Exercise 1.1. Dene : R T . Show that is one-to-one. Determine all t for which t cos t 0. Prove that R is not a manifold at these points. 6.2. Let a 0, 1 be a constant. Prove that the map : R R 2 given by t T is an embedding. (This becomes easier if you rst show that t a sin t, 1 a cos t is an increasing function of t.) Graph the curve dened by .
1 et 6.3. Prove that the map : R R2 given by t e t , et e t T is an embed2 ding. Conclude that M R is a 1-manifold. Graph the curve M. Compute the tangent line to M at 1, 0 and try to nd an equation for M.
6.4. Let I be the open interval 1, and let : I R 2 be the map t 3at 1 3 T , where a is a nonzero constant. Show that is one-to-one and that 1 t t 0 for all t I. Is an embedding and is I a manifold? (Observe that I is a portion of the curve studied in Exercise 1.2.)
where f is the function given in Exercise B.6. Show that is smooth, one-to-one and that its inverse 1 : R R is continuous. Sketch the image of . Is R a manifold? t3 1 2t t1 2 t1 t2 2 t3 2
(i) Show that is one-to-one. (ii) Show that D t is one-to-one for all t 0. (iii) Let U be the punctured plane R2 0 . Show that : U Conclude that U is a two-manifold in R4 .

t1 t2
6.6. Dene a map : R2
R4 by
f t ,f t
if t
6.5. Dene : R
R2 by f t ,f t
T
if t
t3 , 3at2
0, 0,
R4 is an embedding.
{ g{

h 0
D A B
1 A hB A h 1 lim A T A hA T B hB T A h 0h BA T AB T . lim
h2 B T B
AT A
t a sin t
78
6. MANIFOLDS
(iv) Find a basis of the tangent plane to U at the point 1, 1 . 6.7. Let : Rn 0 R be a homogeneous function of degree p as dened in Exercise B.5. Assume that is smooth and that p 0. Show that 0 is the only possible singular value of . (Use the result of Exercise B.5.) Conclude that, if nonempty, 1 c is an n 1-manifold for c 0. 6.8. Let x a1 x2 a2 x2 an x2 , where the a i are nonzero constants. Detern 1 2 mine the regular and singular values of . For n 3 sketch the level surface 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 onedimensional 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. (i) Show that D A B
n
n-matrices and dene : V
R by A
i 1
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 dened by SL n A V det A Show that SL n is a manifold. What is its dimension? I, the identity matrix, we have D A B tr B, (iii) Show that for A n 1 bi,i i 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 dene : W R by
1 .
x1 x4
x2 x3 .
Show that 0 is a regular value of . (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. 11 . (iv) Compute TA M, where A 00
(i) Show that D 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 D and set them equal to 0.) (ii) Show that M 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.)
" # # #
) # "
6.13. Dene : R4
R2 by
x1 x2 x3 x4 x1 x2 x3 x4
"
"
4 " 0 "
"
"
@ $ $ $ 9" 8 7 6" 5
det a1 , a2 . . . , ai
"
#
#

)
!" "
3 2 "
"
'$(%'%&$ %
0 "
"
$ " !"
!"
!"
EXERCISES
79
6.14. Let U be an open subset of Rn and let : U Rm be a smooth map. Let M be 1 c , where c is a regular value of . Let f : U the manifold R be a smooth function. A point x M is called a critical point for the restricted function f M if D f x v 0 for all tangent vectors v Tx M. Prove that x M is critical for f M if and only if there exist numbers 1 , 2 , . . . , m such that grad f x
1 grad 1 x
2 grad 2 x
m grad m 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 given by x2 y2 z2 1, x z 0.
2y
3z over the circle C
Where are the maxima and minima of f C?
6.16 (eigenvectors via calculus). Let A A T be a symmetric n n-matrix and dene 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 eigenvalue of x.
DC G DC
DC F c b` àF E Y GX G F G I G I I
I I WUD V G
F AF
TI'R'RSQD C RI
IPD C
DC E R HD C G
GHD C E DC B E
CHAPTER 7
Differential forms on manifolds

7.1. First denition There are several different ways to dene differential forms on manifolds. In this section we present a practical, workaday denition. A more theoretical approach is taken in Section 7.2. Let M be an n-manifold in R N and let us rst consider what we might mean by a 0-form or smooth function on M. A function f : M 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 dene such a function to be differentiable? The difculty here is that if x is in M and e j is one of the standard basis vectors, the straight 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. Here is one way out of this difculty. Because M is a manifold there exist open R N such that the images i Ui cover M: sets Ui in Rn and embeddings i : Ui M i i Ui . (Here i ranges over some unspecied, possibly innite, index set.) For each i we dene a function f i : Ui R by f i t f i t , i.e. f i i f . We call f i the local representative of f relative to the embedding i . (For instance, if M is the earths surface, f is temperature, and i is a map of New York State, then f i represents a temperature chart of NY.) Since f i is dened on the open subset Ui of Rn , it makes sense to ask whether its partial derivatives exist. We say that f is C k if each of the local representatives f i is C k . Now suppose that x is in the overlap of two charts. Then we have two indices i and j and vectors t Ui and u U j such that x i t j u . Then we must have f x f i t f j u , so 1 fi t f j u . Also i t j u implies t i j u and therefore f j u f i i 1 j u . This identity must hold for all u U j such that j u i Ui , 1 i.e. for all u in j i Ui . We can abbreviate this by saying that
on j 1 i Ui . This is a consistency condition on the functions f i imposed by the fact that they are pullbacks of a single function f dened everywhere on M. The map i 1 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 charts must show the same temperatures.) Conversely, the collection of all local representatives f i determines f , because we have f x f i i 1 x if x i Ui .
81
f e u 'ff e v e i f e
ft w v e i
fj
f e f e u u if e f e w v i 'ff e e i 'ff e e i f e u u
fi
t i f e
'ff e e i f e
'ff e e v (yf e w v x f e ife f e ife f e ife i
f e s6i 'h'ff e rf g e e p q g w v ff e e v
fe
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 denition. 7.1. D EFINITION . A differential form of degree k, or simply a k-form, on M is a collection of k-forms i on Ui satisfying the transformation law
on j 1 i Ui . We call i the local representative of relative to the embedding i and denote it by i i . The collection of all k-forms on M is denoted by k M . This denition is rather indirect, but it works really well if a specic atlas for the manifold M is known. Denition 7.1 is particularly tractible if M is the image of a single embedding : U R N . In that case the compatibility relation (7.1) is vacuous and a k-form on M is determined by one single representative, a k-form 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
I
are two local representatives for , then
on j 1 i Ui . Just like forms on Rn , forms on a manifold can be added, multiplied, differentiated and integrated. For example, suppose is a k-form and an l-form on M. Suppose i , resp. i , is the local representative of , resp. , relative to an embedding i : Ui M. Then we dene the product by setting i ii . To see that this denition makes sense, we check that the forms i satisfy the transformation law (7.1):
Here we have used the multiplicative property of pullbacks, Proposition 3.10(ii). Similarly, the exterior derivative of is dened by setting d i di . As before, let us check that the forms d i satisfy the transformation law (7.1): d
j
where we used Theorem 3.11.
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 Gramann in the mid-nineteenth century and is a prerequisite for much of the more advanced literature on the subject.
j j
j i i
j i
i i
d j
d i
j i
j di
7.2. Second denition
gJ
i 1
f I det D i
f I dt I
(
i
and
j i
(7.1)
' '
g J dt J
J 1
I,J .
j i .
d i ,
7.2. SECOND DEFINITION
83
Covectors. Before giving a rigorous denition of differential forms on manifolds we need to be more precise about the denition of a differential form on R n . Recall that Rn is the collection of all column vectors
Let U be an open subset of Rn . The denition of a 0-form on U requires no further clarication: it is simply a function on U. Formally, a 1-form on U can be dened as a row vector f1, f2, . . . , fn 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 and arbitrary 1-forms as covariant vector elds or covector elds. By denition dx i is the constant 1-form T 0, . . . , 0, 1, 0, . . . , 0 , dxi ei the transpose of ei , the i-th standard basis vector of Rn . Every 1-form can thus be written as
Using this formalism we can write for any smooth function g on U
i 1
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 other than Rn . To see how, let us start by observing that a row vector y is nothing but a 1 n-matrix. We can multiply it by a column vector x to obtain a number,
Obviously we have y c1 x1 c2 x2 c1 yx1 c2 yx2 . Thus a row vector can be viewed as a linear map which sends column vectors in Rn to one-dimensional vectors (scalars) in R1 R. This motivates the following denition. If V is any vector space over the real numbers (for example Rn or a subspace of Rn ), then V , the dual of V, is the set of linear maps from V to R. Elements of V are called dual vectors or covectors or linear functionals. The dual is a vector space in its own right: if 1 and 2 are in V we dene 1 2 and c1 by setting 1 2 v 1 v 2 v for all v V and c1 v c1 v .
g m
f e l f e f ef l e
yx
y1 , y2 , . . . , yn
xn
x1 x2 . . .
i 1
yi xi .
l f h d f
dg
xi dxi
g g g , , ,..., x1 x2 xn
h f
f1, f2, . . . , fn
i 1
f i dxi .
(gf e
d
xn x1 x2 . . .
fe
l f e f e
84
7.2. E XAMPLE . Let V C 0 a, b , R , the collection of all continuous realvalued functions on a closed and bounded interval a, b . A linear combination of continuous functions is continuous, so V is a vector space. Dene f b c2 f 2 c1 f 1 c2 f 2 , so is a linear functional a f x dx. Then c 1 f 1 on V. 7.3. E XAMPLE . Let V Rn and x v V. Dene x v x, where is the standard inner product on Rn . Then is a linear functional on V. Now suppose that V is a vector space of nite 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 . Dene a covector i V by i v ci . In other words, i is determined by the rule
We call i the i-th coordinate function.
7.4. L EMMA . The coordinate functions 1 , 2 , . . . , n constitute a basis of V . Hence dim V n dim V. P ROOF. Let V . We need to write as a linear combination n 1 ci i . i Assuming for the moment that this is possible, we can apply both sides to the vector v j to obtain
So c j v j is the only possible choice for the coefcient c j . To show that this choice of coefcients works, let us dene n 1 vi i . Then by equation i n v j for all j, so , i.e. i 1 vi i . We have proved that (7.2), v j QED every V can be written uniquely as a linear combination of the i .
7.5. E XAMPLE . Consider Rn with standard basis e1 , . . . , en . Then dxi e j T ei e j i, j, so the dual basis of Rn is dx1 , dx2 , . . . , dx n .
Dual bases come in handy when writing the matrix of a linear map. Let L: V W be a linear map between abstract vector spaces V and W. To write the matrix of L we need to start by picking a basis v1 , v2 , . . . , vn of V and a basis w1 , w2 , . . . , wm of W. Then for each j 1, 2, . . . , n the vector Lv j can be expanded uniquely in terms of the ws: Lv j m 1 li, jwi . The m n numbers li, j make up i the matrix of L relative to the two bases of V and W. V be the dual basis of v 1 , v2 , . . . , vn and 1 , 7.6. L EMMA . Let 1 , 2 , . . . , n 2 , . . . , n W the dual basis of w1 , w2 , . . . , wn . Then the i, j-th matrix element of a linear map L : V W is equal to l i, j i Lv j .
k 1
k 1
that is li, j
i Lv j .
i Lv j
s n o
s n o n
P ROOF. We have Lv j
m k
lk, ji
1 l k, j wk , m
so
wk
lk, jik
li, j , QED
The basis 1 , 2 , . . . , n of V is said to be dual to the basis v1 , v2 , . . . , vn of V.
n s o
'zs o z z n s o s n s s o o nn s o ns xs n o o
i 1
i 1
vj
ci i
vj
cii, j
c j.
~n
i, j
s o n z w n n
}n |
i v j
1 if i 0 if i
j, j.
n ts o
n {s o
y xs o n
r p
z w
s o
w n v s o s v o n s r qpo n w ns o
z w
s o n
z w
n z n
so u n n
(7.2)
85
Multilinear algebra. Let V be a vector space and let V k denote the Cartesian 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 : V k R which is linear in each vector, i.e.
for all scalars c, c and all vectors v1 , v2 , . . . , vi , vi , . . . , vk . Rn and let x, y 7.7. E XAMPLE . Let V y. Then is bilinear (i.e. 2-multilinear).
x y, the inner product of x and
A k-multilinear function is alternating or antisymmetric if it has the alternating property,
7.10. E XAMPLE . The inner product of Example 7.7 is bilinear, but it is not alternating. Indeed it is symmetric: y x x y. The bilinear function of Example 7.8 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 1 , 2 , . . . , k V . The (wedge) product is the function
1 2 1 2
k : V k
(The determinant on the right is a k k-determinant.) It follows from the multilinearity and the alternating property of the determinant that 1 2 k is an alternating k-multilinear function. The wedge product is often denoted by 1 2 k to distinguish it from other products, such as the tensor product dened in Exercise 7.6. The collection of all alternating k-multilinear functions is denoted by A k V. For k 1 the alternating property is vacuous, so an alternating 1-multilinear function is nothing but a linear function. Thus A 1 V V . For k 0 a k-multilinear function is dened to be a single number. Thus 0V A 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.
k v1 , v2 , . . . , v k
det i v j
dened by
1 i, j k
, . . . , v
sign v1 , . . . , vk .
More generally, if is alternating, then for any permutation
S k we have
v1 , . . . , v j , . . . , v i , . . . , v k
v1 , . . . , v i , . . . , v j , . . . , v k .
7.9. E XAMPLE . Let V Rn , k n. n-multilinear function on R
n. The determinant det v1 v2 , , . . . , vn is an
7.8. E XAMPLE . Let V R4 , k 4. v3 w4 v4 w3 is bilinear on R
2. The function v, w
v 1 w2
v2 w1
c v 1 , v 2 , . . . , vk
v1 , v2 , . . . , cvi
c vi , . . . , v k
c v1 , v2 , . . . , v i , . . . , v k
&
86
I
vI
i1 i2
ik
Ak V,
vi 1 , vi 2 , . . . , v i k
Vk.
7.11. E XAMPLE . Let V R3 with standard basis e1 , e2 , e3 . The dual basis 3 of R is dx1 , dx2 , dx3 . Let k 2 and I 1, 2 , J 2, 3 . Then dx I e I dx1 e1 dx2 e1 dx1 e2 dx2 e2 dx2 e1 dx3 e1 dx2 e2 dx3 e2
dx1 e2 dx2 e2 dx1 e3 dx2 e3 dx2 e2 dx3 e2 dx2 e3 dx3 e3
1 0 0 1 0 0 1 0
0 1 0 0 1 0 0 1
1, 0, 0, 1.
dx I e J dx J e I
dx J e J
This example generalizes as follows.
7.12. L EMMA . Let I and J be increasing multi-indices of degree k. Then
1 r,s k
J, then il jl for some l. Choose l as small as possible, so that i m jm for l. There are two cases: i l jl and il jl . If il jl , then il jl jl 1 jk because J is increasing, so all entries il , jm in the determinant with m l are 0. For m l we have j m im il because I is increasing, so il , jm 0 for m l. In other words the l-th row in the determinant is 0 and hence I v J 0. If il jl we nd that the l-th column in the determinant is 0 and therefore again I vJ 0. QED We need one further technical result before showing that the functions I are a basis of Ak V.
P ROOF. The assumption implies
vi 1 , . . . , v i k
7.13. L EMMA . Let of degree k. Then 0.
A k V. Suppose v I
0 for all increasing multi-indices I
If I m
ik , j1
il , j1 . . .
I vJ
det ir v js
i1 , j1 . . .
. . . i1 , jk . . . . . . il , jk . . . . . . ik , jk
P ROOF. Let I
i1 , . . . , ik and J
j1 , . . . , jk . Then
' }
I vJ
I,J
1 if I 0 if I
J, J.
1 if I
J.
(7.3)
Let 1 , . . . , n be the corresponding dual basis of V . Let I an increasing multi-index, i.e. 1 i 1 i2 ik n. Write
i1 , i2 , . . . , ik be
} } }
87
for all multi-indices i1 , . . . , ik , because of the alternating property. We need to show that w1 , . . . , wk 0 for arbitrary vectors w1 , . . . , wk . We can expand the wi using the basis: . . . ak1 v1
Therefore by multilinearity
i1 1
ik 1
Each term in the right-hand side is 0 by equation (7.3).
7.14. T HEOREM . Let V be an n-dimensional vector space with basis v 1 , . . . , vn . Let 1 , . . . , n be the corresponding dual basis of V . Then the alternating k-multilinear functions I i1 ik , where I ranges over the set of all increasing multi-indices of n degree k, form a basis of A k V. Hence dim Ak V k . A k V. P ROOF. The proof is closely analogous to that of Lemma 7.4. Let We need to write as a linear combination I c I I . Assuming for the moment that this is possible, we can apply both sides to the k-tuple of vectors v J . Using Lemma 7.12 we obtain
I I
So c J v J is the only possible choice for the coefcient c J . To show that this choice of coefcients works, let us dene I v I I . Then for all increasing multi-indices I we have v I vI vI vI 0. Applying Lemma 7.13 to we nd 0. In other words, I v I I . We have proved that every V can be written uniquely as a linear combination of the i . QED Rn with standard basis e1 , . . . , en . The dual basis 7.15. E XAMPLE . Let V n of R is dx1 , . . . , dx n . Therefore Ak V has a basis consisting of all k-multilinear functions of the form dx I dxi1 dxi2 dxik , with 1 i1 on Rn looks like ik n. Hence a general alternating k-multilinear function
with a I constant. By Lemma 7.12, e J equal to e I .
An arbitary k-form on a region U in Rn is now dened as a choice of an alternating k-multilinear function x for each x U; hence it looks like x I f I x dx I , where the coefcients f I are functions on U. We shall abbreviate this to f I dx I ,
I
r { b b
a I dx I ,
I
I a I dx I e J
vJ
cI I
vJ
c I I,J
cJ.
I a I I,J
a J , so a I is
w1 , . . . , w k
wk

a1i1
w1
a11 v1
a1k vk , akk vk .
akik vi1 , . . . , vik . QED
88
and we shall always assume the coefcients f I to be smooth functions. By Example 7.15 we can express the coefcients as f I e I (which is to be interpreted as fI x x e I for all x). Pullbacks re-examined. In the light of this new denition we can give a fresh 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 : U V a smooth map. For a k V dene the pullback k U by k-form
Let us check that this formula agrees with the old denition. Suppose I f I dy I and J g J dx J . What is the relationship between g J and f I ? We use g J e J , our new denition of pullback and the denition of the wedge product to get

I I
f I x det dyir D x e js
By Lemma 7.6 the number dy ir D x e js is the ir js -matrix entry of the Jacobi matrix D x (with respect to the standard basis e1 , e2 , . . . , en of Rn and the standard basis e1 , e2 , . . . , em of Rm ). In other words, g J x I f I x det D I,J x . This formula is identical to the one in Theorem 3.12 and therefore our new denition 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 on M is a choice of an alternating kmultilinear map x on the vector space Tx M, one for each x M. This alternating map x is required to depend smoothly on x in the following sense. According to the denition of a manifold, for each x M there exists an embedding : U RN N containing x. The tangent such that U M V for some open set V in R space at x is then Tx M D t Rn , where t U is chosen such that t x. The pullback of under the local parametrization is dened by
Then is a k-form on U, an open subset of Rn , so I f I dt I for certain functions f I dened on U. We will require the functions f I to be smooth. (The form I f I dt I is the local representative of relative to the embedding , as introduced in Section 7.1.) To recapitulate: 7.16. D EFINITION . A k-form on M is a choice, for each x M, of an alternating k-multilinear map x on Tx M, which depends smoothly on 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
v1 , v 2 , . . . , v k
D t v1 , D t v2 , . . . , D t vk .
fI
dy I D x e j1 , D x e j2 , . . . , D x e jk
1 r,s k
gJ x
eJ
D x e j1 , D x e j2 , . . . , D x e jk
v 1 , v 2 , . . . , vk
D x v1 , D x v2 , . . . , D x vk .
' x
EXERCISES
89
same direction as the orientation and negative if it points in the opposite direction. Dene a 1-form on M as follows. For x M and a tangent vector v Tx M put
This form is the element of arc length of M. We shall see in Chapter 8 how to generalize it to higher-dimensional manifolds and in Chapter 9 how to use it to calculate arc lengths and volumes. We can calculate the local representative of a k-form for any embedding : U R N parametrizing a portion of M. Suppose we had two different such M and j : U j M, such that x is contained in both Wi embeddings i : Ui i Ui and W j j U j . How do the local expressions i i and j j for compare? To answer this question, consider the coordinate change map
j i .
This shows that Denitions 7.1 and 7.16 of differential forms on a manifold are equivalent. Exercises
7.2. Let v1 , v2 , . . . , vn be a basis of Rn and let 1 , 2 , . . . , n be the dual basis of Rn . Let A be an invertible n n-matrix. Then by elementary linear algebra the set of vectors Av1 , . . . , Avn is also a basis of Rn . Show that the corresponding dual basis is the set of row vectors 1 A 1 , 2 A 1 , . . . , n A 1 . 7.3. Suppose that is a bilinear function on a vector space V satisfying v, v 0 for all vectors v V. Prove that is alternating. Generalize this observation to k-multilinear functions.
7.5. The wedge product is a generalization of the cross product to arbitrary dimensions in the sense that T x y xT yT 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 sides of the formula represent column vectors.) 7.6. Let V be a vector space and let 1 , 2 , . . . , k product is the function
V be covectors. Their tensor
1
dened by Show that 1
k : V k
k v1 , v2 , . . . , vk
1 v1 2 v 2
k vk .
k is a k-multilinear function.
7.4. Show that the bilinear function of Example 7.8 is equal to dx 1 dx2
'( H
T'
T( ''S
7.1. The vectors e 1
e2 and e1
e2 form a basis of R2 . What is the dual basis of R2 ?
dx3 dx4 .
j 1 i , which maps i 1 Wi W j to j 1 Wi j we recover the transformation law (7.1)
W j . From i
i and j
x v
v v
if v is positive, if v is negative.
90
by
7.7. Let : V k
R be a k-multilinear function. Dene a new function Alt : V k
7.9. Let V and W be vector spaces and L : V L L for all covectors , W .
W a linear map. Show that L
dx1 dx2
dxn .
7.8. Show that det v1 , v2 , . . . , vn v2 , . . . , vn Rn . In short, det
dx1 dx2
dxn v1 , v2 , . . . , vn for all vectors v1 ,

Prove the following. (i) Alt is an alternating k-multilinear function. (ii) Alt if is alternating. (iii) Alt Alt Alt for all k-multilinear . (iv) Let 1 , 2 , . . . , k V . Then 1 Alt 1 2 1 2 k k!
k .
Alt v1 , v2 , . . . , vk
1 sign v k! S
, v
, . . . , v
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 n 1 ci ai , where the coefcients c i range i 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 dening 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 dening the determinant, namely by imposing a few reasonable axioms. tion 8.1. D EFINITION . An absolute n-dimensional Euclidean volume function is a func-
n times
with the following properties: (i) homogeneity: for all scalars c and all vectors a1 , a2 , . . . , an ; (ii) invariance under shear transformations: voln a1 , a2 , . . . , cai , . . . , an c voln a1 , a2 , . . . , an
for all scalars c and any i j; (iii) invariance under Euclidean motions: 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 dene 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
6 ##" 6 ""
6 86
6 7 1
1
1
voln Qa1 , . . . , Qan
voln a1 , . . . , an
1
5
@ #6
1
6 9
vol n a1 , . . . , ai
ca j , . . . , a j , . . . , an
voln a1 , . . . , a j , . . . , ai , . . . , an
2 3 1 2
!$##"& " " '!
vol n : R N
RN
RN
92
8. VOLUME FORMS
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 n 2 ci ai . By repeatedly applying Axiom i (ii) we get
i 2
i 3
Now by Axiom (i), which proves property (ii).
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 R N and let A be the N n-matrix whose i-th column is a i . Then
P ROOF. We leave it to the reader to check that the function det A T A satises 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 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 T A is square, this implies det A T A 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
EGGG Y##HE
E 8E
B D
E FB D
SB D W
voln a1 , a2 , . . . , an
det A T A .
B D
B D
B D
vol n 0, a2 , . . . , an
voln 0 0, a2 , . . . , an
0 voln 0, a2 , . . . , an
0, QED
B GGG T##SB R
vol n
c i a i , a2 , . . . , a n
P
QP
B D
vol n a1 , a2 , . . . , an
vol n
c i ai , a 2 , . . . , a n
a1
a2
an
a1
a2
an voln e1 , e2 , . . . , en
vol n 0, a2 , . . . , an .
a1
a2
an voln Qe1 , Qe2 , . . . , Qen
EI##HE GGG EI##HE GGG EI##HE GGG EGGG I##HE
E8E E8E E8E E 8E
EFB E FB EFB E FB D B B
vol n a1 , a2 , . . . , an
a1
a2
an voln q1 , q2 , . . . , qn
by Axiom (i) by Axiom (iii) by Axiom (iv),
Q be the matrix whose i-th column is qi . Then Q is orthogonal and Qei Therefore
qi .
8.1. n-DIMENSIONAL VOLUME IN RN
93
where the last equality follows from Case 1. Since vi ai bi , where bi is a linear combination of a1 , a2 , . . . , ai 1, we have V AU, where U is a n n-matrix of the form 1 0 1 0 0 1 U . . . .. . . . . . 0 0 0 1 det A T A det U T det A T A det U det U T A T AU
The Gram-Schmidt process transforms a sequence of n independent vectors a1 , a2 , . . . , an into an orthogonal sequence v 1 , v2 , . . . , vn . (The horizontal oor represents the plane spanned by a1 and a2 .) The block spanned by the as has the same volume as the rectangular block spanned by the vs. a3 a3 a2 a1 a2 b2 b3 v3 v2 v3
a3 a2 a1
v2
8.4. C OROLLARY. Let a1 , a2 , . . . , an be vectors in Rn and let A be the n whose i-th column is ai . Then voln a1 , a2 , . . . , an det A .
Fb f
For n
N Theorem 8.3 gives the following result. n-matrix
w xb f
Using formula (8.1) we get vol n a1 , a2 , . . . , an
det A T A .
a1
v1
v1
b f
Note that U has determinant 1. This implies that V T V
U T A T AU and det V T V . QED
h Sb f
voln v1 , v2 , . . . , vn
det V T V , (8.1)
ggg ##Sb f
uu t
uu s
u s
vs
g##g g g##g g gg ##g
b f e b
gg ##g
pqq
qq
qr
b f
vol n a1 , a2 , . . . , an
vol n v1 , a2 , . . . , an
vol n v1 , v2 , . . . , an
vi ai bi . (See illustration below.) Let V be the N is vi . Then by repeated applications of Axiom (ii),
n-matrix whose i-th column
b f
94
8. VOLUME FORMS
8.2. Orientations Oriented vector spaces. You are probably familiar with orientations on vector spaces of dimension 3. An orientation of a line is an assignment of a direction. 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 vector space over the real numbers. Suppose that v1 , v2 , . . . , vn and v1 , v2 , . . . , vn are two ordered bases of V. Then we can write vi j ai, j v j and vi a i, j j bi, j v j for suitable coefcients a i, j and bi, j. The n n-matrices A and B bi, j satisfy AB BA I and are therefore invertible. We say that the dene the same orientation of V if det A 0. If det A 0, the two bases and bases dene opposite orientations. For instance, if v1 , v2 , . . . , vn v2 , v1 , . . . , vn , then
...
so det A 1. Hence the ordered bases v2 , v1 , . . . , vn and v1 , v2 , . . . , vn dene opposite orientations. We know now what it means for two bases to have the same orientation, but how do we dene the concept of an orientation itself? In typical mathematicians fashion we dene the orientation of V determined by the basis to be the collection of all ordered bases that have the same orientation as . (There is an analogous denition of the number 1, namely as the collection of all sets that contain one element.) The orientation determined by v1 , v2 , . . . , vn is denoted by or v1 , v2 , . . . , vn . So if and dene the same orientation then . If they dene opposite orientations we write . Because the determinant 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 0 we need to make a special denition, because a zero-dimensional space has an empty basis. In this case we dene an orientation of V to be a choice of sign, or . 8.5. E XAMPLE . The standard orientation on Rn is the orientation e1 , . . . , en dened by the standard ordered basis e1 , . . . , en . We shall always use this orientation on Rn . Maps and orientations. Let V and W be oriented vector spaces of the same dimension and let L : V W be an invertible linear map. Choose a positively oriented basis v1 , v2 , . . . , vn of V. Because L is invertible, the ordered n-tuple
p rq
p sq l p
jj
0 1 0 . . .
1 0 0 . . .
0 0 1 . . .
... ... ... .. .
0 0 0 . . .
jj i
P ROOF. A is square, so det A T A 3.7(ii) and therefore vol n a1 , a2 , . . . , an
det A T det A det A 2
det A 2 by Theorem det A by Theorem 8.3. QED
gh gg fgg S8
lm
t o
pq
8.2. ORIENTATIONS
95
Lv1 , Lv2 , . . . , Lvn is an ordered basis of W. If this basis is positively, resp. negatively, oriented we say that L is orientation-preserving, resp. orientation-reversing. This denition does not depend on the choice of the basis, for if v1 , v2 , . . . , vn is 0. Thereanother positively oriented basis of V, then vi j ai, j v j with det ai, j fore Lvi L j ai, jv j j ai, j Lv j , and hence the two bases Lv 1 , Lv2 , . . . , Lvn and Lv1 , Lv2 , . . . , Lvn of W determine the same orientation. Oriented manifolds. Now let M be a manifold. We dene an orientation of M to be a choice of an orientation for each tangent space Tx M which varies continuously over M. Continuous means that for every x M there exists a loM, with W open in Rn and x W , such that cal parametrization : W D y : R n 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 orientation; it is oriented if a specic orientation has been chosen. Hypersurfaces. The case of a hypersurface, a manifold of codimension 1, is particularly instructive. A unit normal vector eld on a manifold M in R n is a smooth function n : M Rn such that n x Tx M and n x 1 for all x M. 8.6. P ROPOSITION . A hypersurface in Rn is orientable if and only if it possesses a unit normal vector eld. P ROOF. Let M be a hypersurface in Rn . Suppose M possesses a unit normal vector eld. Let v1 , v2 , . . . , vn 1 be an ordered basis of Tx M for some x M. vi for all i. We say that Then n x , v1 , v2 , . . . , vn 1 is a basis of Rn , because n x v1 , v2 , . . . , vn 1 is positively oriented if n x , v1 , v2 , . . . , vn 1 is a positively oriented basis of Rn . This denes an orientation on M, called the orientation induced by the normal vector eld 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 Tx M is a line. There are therefore precisely two vectors of length 1 which are perpendicular to Tx M. We can pick a preferred unit normal vector as follows. Let v1 , v2 , . . . , vn 1 be a positively oriented basis of Tx M. The positive unit normal vector is that unit normal vector n x that makes n x , v1 , v2 , . . . , vn 1 a positively oriented basis of Rn . In Exercise 8.8 you will be asked to check that n x depends smoothly on x. In this way we have produced a unit normal vector eld on M. QED 8.7. E XAMPLE . Let us regard Rn 1 as the subspace of Rn spanned by the rst 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
The positive unit normal on an oriented hypersurface can be regarded as a map n from M into the unit sphere S n 1 , which is often called the Gau map of M. The unit normal enables one to distinguish between two sides of M: the direction 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.
by Exercise 8.5, the positive unit normal to
Rn 1
in
Rn
is
w v
e1 , e 2 , . . . , e n
n 1
e n , e1 , e2 , . . . , e n
1 n 1e
n.
M in Rn
wx
v z{w x x v v w ~ v
8w
y w
v v
y x w
v v
w
w x y }
x y x v
v v
w
~ v
v v v
96
8. VOLUME FORMS
8.8. P ROPOSITION . Let U be open in Rn and let : U R be a smooth function. Let c be a regular value of . Then the manifold 1 c has a unit normal vector eld given by n x grad x grad x and is therefore orientable.
1 c is a hypersurface in Rn (if nonempty), and also that Tx M ker Dx grad x . The function n x grad x grad x therefore denes a unit normal vector eld on M. Appealing to Proposition 8.6 we conclude that M is orientable. QED
8.3. Volume forms Now let M be an oriented n-manifold in R N . Choose a collection of embeddings i : Ui R N with Ui open in Rn such that M i i Ui and such that n D i t : R Tx M is orientation-preserving for all t Ui . The volume form M , also denoted by , is the n-form on M whose local representative relative to the embedding i is dened by
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 D i t , the Jacobi matrix of i at t. Hence you should think of as measuring the volume of innitesimal blocks inside M. 8.10. T HEOREM . For any oriented n-manifold M in R N the volume form M is a well-dened n-form. P ROOF. To show that is well-dened we need to check that its local representatives satisfy the transformation law (7.1). So let us put i 1 j and substitute t u into i . Since each of the embeddings i is orientation-preserving, we have det D 0. Hence by Theorem 3.13 we have Therefore
where in the second to last identity we applied the chain rule.
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. Dont be misled by this old-fashioned notation: volume forms are seldom exact! The volume form M is highly dependent
##
det D j u
T D
du1 du2
dun
j,
##
det Di u D u
T D
u D u
du1 du2
dun
##
det D u
det Di u
T D
det D u du1 du2
##
det Di u
T D
det D u du1 du2
dun dun
##
##
##
dt1 dt2
dtn
det D u du1 du2
dun
det D u du1 du2
##
det Di t
T D
dt1 dt2
dtn .
n x
grad x
grad x
x .
dun .
QED
x 2 and c r2 we obtain that the n 8.9. E XAMPLE . Taking x sphere of radius r about the origin is orientable. The unit normal is
#
P ROOF. The regular value theorem tells us that M
1 -
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 : U U, x x. Then det D T D 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 2 function. Let M R be the graph of f . By Example 6.7 M is a 1-manifold in R 2 . Indeed, M is the image of the embedding : I R2 given by t t, f t . Let us give M the orientation induced by the embedding , i.e. from left to right. What is the element of arc length of M? Let us compute the pullback , a 1-form on I. We have 1 1 1 f t D t , D t T D t 1 f t 2, f t f t
The next result can be regarded as an alternative denition of M . It is perhaps more intuitive, but it requires familiarity with Section 7.2.
i.e. 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 x v1 , v2 , . . . , vn be the oriented volume of the block spanned by these n vectors. This denes an n-form on M and we must show that M . Let U be an open subset of Rn and : U R N an orientation-preserving embedding with U M and t x for some t in U. Let us calculate the n-form on U. We have g dt1 dt2 dtn for some function g. By Lemma 7.12 this function is given by where in the second equality we used the denition of pullback. The vectors D t e1 , D t e2 , . . . , D t en are a positively oriented basis of Tx M and, moreover, are the columns of the matrix D t , so by Theorem 8.3 they span a positive volume of magnitude det D t T D t . This shows that g det D T D and therefore Thus is equal to the local representative of M with respect to the embedding . Since this holds for all embeddings , we have M . QED
##
det D T D dt1 dt2
dtn .
g t
x e1 , e 2 , . . . , e n
x D t e1 , D t e2 , . . . , D t en ,
voln v1 , v2 , . . . , vn voln v1 , v2 , . . . , vn 0
##
M,x v1 , v2 , . . . , vn
if v1 , v2 , . . . , vn are positively oriented, if v1 , v2 , . . . , vn are negatively oriented, if v1 , v2 , . . . , vn are linearly dependent,
8.13. P ROPOSITION . Let M be an oriented n-manifold in R N . Let x v2 , . . . , vn Tx M. Then the volume form of M is given by
so
det D t
T D
dt
f t
2 dt.
M and v1 ,
##
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 M . Recall the vector-valued forms dx1 dx1 . . . . dx and dx . . introduced in Section 2.5. Let n be the positive unit normal vector eld on M and let F be any vector eld on M, i.e. a smooth map F : M R n . Then the inner product F n is a function dened on M. It measures the component of F orthogonal to M. The product F n M is an n 1-form on M. On the other hand we have the n 1-form F dx F dx. 8.14. T HEOREM . On the hypersurface M we have
F IRST PROOF. This proof is short but requires familiarity with the material in Section 7.2. Let x M. Let us change the coordinates on R n in such a way that the rst n 1 standard basis vectors e1 , e2 . . . , en 1 form a positively oriented basis of Tx M. Then, according to Example 8.7, the positive unit normal at x is given by n x 1 n 1en and the volume form satises M,x e1 , . . . , en 1 1. Writing n F i 1 Fi ei , we have F x n x 1 n 1 Fn x . On the other hand
i
which implies F x M, we nd F
dx dx
F x n x M . Since this equality holds for every QED F n M.
S ECOND PROOF. Choose an embedding : U Rn , where U is open in Rn 1 , M, x U . Let t U be the point satisfying t x. As such that U a preliminary step in the proof we are going to replace the embedding with a new one enjoying a particularly nice property. Let us change the coordinates on Rn in such a way that the rst n 1 standard basis vectors e1 , e2 . . . , en 1 form a positively oriented basis of Tx M. Then at x the positive unit normal is given by 1 n 1en . Since the columns of the Jacobi matrix D t are independent, n x there exist unique vectors a1 , a2 , . . . , an 1 in Rn 1 such that D t ai ei for i 1, 2, . . . , n 1. These vectors ai are independent, because the ei are independent. Therefore the n 1 n 1 -matrix A with i-th column vector equal to a i is invertible. Put U A 1 U , t A 1 t and A. Then U is open in Rn 1 , n is an embedding with U t x, : U R U , and t by the chain rule. Therefore the i-th column vector of D is
t D ei
D t Aei
D t a i
t D
D t
DA t
D t
ei
(8.2)
S x8 S
and therefore F F
dx e1 , . . . , en
x
n 1F . n
This proves that

1
dx
e1 , . . . , e n
F x
n x M e1 , . . . , e n
##
$##
dx
i 1
Fi dx1
dxi
dxn ,
x8
dx
F n M.
dxn
dxn
x 8
8.3. VOLUME FORMS
99
for i 1, 2, . . . , n 1. (On the left ei denotes the i-th standard basis vector in Rn 1 , on the right it denotes the i-th standard basis vector in Rn .) In other words, t n-matrix the Jacobi matrix of at is the n 1
n 1 identity matrix and 0 denotes a row consisting where In 1 is the n 1 of n 1 zeros. Let us now calculate F n M and F dx at the point . Writing t n F n i 1 Fi ni and using the denition of M we get
i 1
1e
i 1
This theorem gives insight into the physical interpretation of n 1-forms. Think of the vector eld F as representing the ow of a uid or gas. The direction of the vector F indicates the direction of the ow and its magnitude measures the strength of the ow. Then Theorem 8.14 says that the n 1-form F dx measures, for any unit vector n in Rn , the amount of uid per unit of time passing through a hyperplane of unit volume perpendicular to n. We call F dx the ux of the vector eld 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 dx: if n is a unit vector in Rn , then the scalar-valued n 1form n dx measures the volume of an innitesimal n 1-dimensional parallelepiped perpendicular to n. 8.15. C OROLLARY. Let n be the unit normal vector eld and M the volume form of an oriented hypersurface M in Rn . Then
P ROOF. Set F
n in Proposition 8.14. Then F n
dx. 1 because n 1. QED
We conclude that F n M t F dx x . Since this holds for all x
F dx , in other words F n M x t M we have F dx F n M. QED
##
# I F #
dx
n 1
Fn x dt1 dt2
dtn
1.
t From formula (8.2) we see i t j j n 1. Therefore for 1
i, j for 1
i, j
##
##
dx
i 1
Fi d1 d2
##
T##
From F
dx
n i
i 1F i
dx1 dx2
dxi
dxn we get di
t 1 and n t j
##
x
F n M
n 1
Fn x dt1 dt2
t From formula (8.2) we have det D and using n x 1 n at the point t
T D
1. So evaluating this expression t we get dtn

1.
dn . 0
##
x8
F n M
Fi ni
det D T D dt1 dt2
t D
In 1 , 0
dtn
1.
100
8. VOLUME FORMS
8.16. E XAMPLE . Suppose the hypersurface M is given by an equation x c, where c is a regular value of a function : U R, with U open in R n . Then by Proposition 8.8 M has a unit normal n grad grad . The volume form is therefore grad 1 grad dx. In particular, if M is the sphere of radius x R, so M R 1 x dx. R about the origin in Rn , then n x 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 a b sin , where is the angle between a and b (which is taken to lie between 0 and ). Show that a b sin a b in R 3 .
be vectors in Rn 1 . (i) Deduce from Exercise 8.3 that

1
8.5. Justify the following identities concerning orientations of a vector space V. Here the vs form a basis of V (which in part (i) is n-dimensional and in parts (ii)(iii) twodimensional). (i) If Sn is any permutation, then
1 2 n
(ii) v1 , v2 (iii) 3v1 , 5v2
v1 , v2 . v1 , v 2 .
8.6. Let U be open in Rn and let f : U R be a smooth function. Let : U Rn 1 be x, f x and let M U , the graph of f . Dene an orientation on the embedding x M by requiring to be orientation-preserving. Deduce from Exercise 8.4 that the volume form of M is given by M 1 grad f x 2 dx1 dx2 dxn .
8.7. Let M
graph f be the oriented hypersurface of Exercise 8.6.
'
21
& %
#$" !
0 ) ' % ' ' % ' "! "!
, v
, . . . , v
sign v1 , v2 , . . . , vn .

1 a2 1 a2 a1 a3 a1 . . . an a1
a1 a2 1 a2 2 a3 a2 . . . an a2
a1 a3 a2 a3 1 a2 3 . . . an a3
... ... ... .. . ...
a1 an a2 an a3 an . . . 1 a2 n

(ii) Prove that
i 1
a2 . i

voln u1 , u2 , . . . , un
voln
u 1 , u2 , . . . , u n , u n

u1
1 0 . . . 0 a1
u2
0 1 . . . 0 a2
...,
un
0 0 . . . 1 an
un
1 c

8.4. Let a1 , a2 , . . . , an be real numbers, let c
n i
2 1 ai
and let a1 a2 . . . an 1

volk
u1 , u2 , . . . , uk , v1 , v2 , . . . , vl
volk u1 , u2 , . . . , uk voll v1 , v2 , . . . , vl .
8.3. Let u1 , u2 , . . . , uk and v1 , v2 , . . . , vl be vectors in R N satisfying ui v j 2, . . . , k and j 1, 2, . . . , l. (The us are perpendicular to the vs.) Prove that
8.2. Check that the function voln a1 , a2 , . . . , an Denition 8.1.
det A T A satises the axioms of 0 for i 1,
8 Y
% %
EXERCISES
101
(i) Show that the positive unit normal vector eld on M is given by
(ii) Derive the formula M 1 grad f x 2 dx1 dx2 dxn from Corollary 8.15 by substituting x n 1 f x1 , x2 , . . . , xn . (Caution: for consistency you must replace n with n 1 in Corollary 8.15.) 8.8. Show that the unit normal vector eld n : M Rn dened in the proof of Proposition 8.6 is smooth. (Compute n in terms of an orientation-preserving parametrization : U M of an open subset of M.) 8.9. Let : a, b Rn be an embedding. Let be the element of arc length on the embedded curve M a, b . Show that is the 1-form on a, b given by t dt
1 t
2 t
n t
dt.
3 Q6 4 U @ @
RR &2R
6 4
S 4 3 79 6 @ Q6 4 P9 8 3 I @ H E D &6 4 A9 @ @ 7 6 54 8 3 G C GG E CC GGF E BCC

1
n 1
grad f x
f x1 f x2 . . . f xn 1
R 6 4 U 9 &R2R 9 6 4 U 9 6 4 U V 6 4 3 4 ST6
CHAPTER 9
Integration and Stokes theorem on manifolds

In this chapter we will see how to integrate an n-form over an oriented nmanifold. In particular, by integrating the volume form we nd 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 rediscovery of Greek astronomy in the late Middle Ages) visualized the earth as a at 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 under an embedding. Rather, it is a manifold with boundary, a notion which can be dened as follows. The n-dimensional halfspace is Hn The boundary of Hn is Hn int Hn x Rn x n 0 . x Rn x n Rn xn
Rn
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 an open subset V R N containing x, an open subset U Rn , and an embedding : U R N satisfying U
Hn
You should compare this denition carefully with Denition 6.4 of a manifold. If x 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 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. Likewise the interior int M is an n-manifold. The most obvious example of an n-manifold with boundary is the halfspace Hn itself, which has boundary Hn Rn 1 and interior the open halfspace x n R xn 0 . Here is a more interesting type of example, which generalizes the graph of a function. R be a 9.2. E XAMPLE . Let U be an open subset of Rn 1 and let f : U smooth function. Put U U R and write elements of U as x with x in U and y
103
` X
p vu
r Ws
r q
W c W a c b a
0 .
` XdW ` YW X W p
and its interior is
xu W w u
c g a
` fW X h h h
M.
c g a
sq W
104
9. INTEGRATION AND STOKES THEOREM ON MANIFOLDS
x 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 Rn by embedding. Dene : U
As in Example 6.3 one veries that is an embedding, using the fact that
where 0 is the origin in Rn 1 . By denition therefore, the set M U Hn is n with boundary M n . What are M and M? A an n-manifold in R U H point x is in M if and only if it is of the form y x y
t u
f t
for some in U H . Since H is given by u 0, this is equivalent to x U and y f x . Thus M is exactly the region below the graph. On H n we have u 0, so M is given by the equality y f x , i.e. M is the graph. 9.3. E XAMPLE . If f : U Rm is a vector-valued map one cannot speak about the region below the graph, but one can do the following. Again put U U x R. Let N n m 1 and think of R N as the set of vectors y with x in Rn 1 and R N by y in Rm . Dene : U
This time we have
and again is an embedding. Therefore M U H n is an n-manifold in RN x with boundary M U Hn . This time M is the set of points y of the form x y f t t uem
D t
In 1 Df t
0 em
t u
f t
uem
e f
t u
t u
f t
t u
In 1 Df t
0 , 1
y in R. The region below the graph of f is the set consisting of all y f x . y M graph f
x y
in U such that
9.1. MANIFOLDS WITH BOUNDARY
105
Again M is given by y
f x , 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 Theorem 6.10. 9.4. T HEOREM (regular value theorem for manifolds with boundary). Let U be open in R N and let : U Rm be a smooth map. Let M be the set of x in R N satisfying
1
Suppose that c c 1 , c2 , . . . , cm is a regular value of and that M is nonempty. Then M is a manifold in R N of codimension m 1 and with boundary M 1 c . 9.5. E XAMPLE . Let U Rn , m 1 and x x 2 . The set given by the inequality x 1 is then the closed unit ball x Rn x 1 . Since 2x, any nonzero value is a regular value of . Hence the ball is an grad x n-manifold in Rn , whose boundary is 1 1 , the unit sphere S 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 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 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 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 Mbius band shown in Chapter 1. The Mbius band is a nonorientable manifold with boundary. We will not give a proof of this fact, but you can convince
i x
sv v s y
Rn R 1
R2
v i wo x q p v
r z s v vy qp r o sq p
oq p r
q p r
oq p
o p s q p q
p uo
oq p
1 x
c1 ,
2 x
c2 ,
...,
cm
1,
m x
q p
q p o q p o
q p o
y1
f1 x ,
y2
f2 x ,
...,
ym
fm
x ,
ym
fm x .
q p r o r ml
with t U and u 0. Hence M is the set of points y satises m 1 equalities and one inequality:
x y
where x is in U and where
j i
cm .
sq p s
106
yourself that it is true by trying to paint the two sides of a Mbius 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 dene the tangent space to a manifold with boundary M at a point x choose U and as in the denition and put As in the case of a manifold, this does not depend on the choice of the embedding . Now suppose x is a boundary point of M and let v Tx M be a tangent vector. Then v D 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, 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 Tx M to be the unique proof of Proposition 8.6. Namely, for x M dene n x outward-pointing tangent vector of length 1 which is orthogonal to Tx M. This denes the unit outward-pointing normal vector eld 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 oriented basis of Tx M. This denes 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
1
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 be an n-form on M. To dene the integral of 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 integral and therefore converges.) For a start, let us also make the assumption that there exists a smooth map c : 0, 1 n R N such that c 0, 1 n M and the restriction of c to 0, 1
is an orientation-preserving embedding.
For instance, this assumption is satised for the n-sphere S n (see Exercise 5.5) and the torus S1 S1 (see Exercise 9.4). The pullback c is then an n-form on the cube 0, 1 n . We dene
M
0,1
c .
en , e 1 , e 2 , . . . , e n
e1 , e 2 , . . . , e n
1 , en
} |
} 1|{
} | |
Tx M
D t Rn
} ~} | |
} ~} | |
Tx M
D t R n .
} |
} 1|
{ 2
}|
9.2. INTEGRATION OVER ORIENTABLE MANIFOLDS
107
Suppose c : 0, 1 n R N is a smooth map with the same properties as c. To ensure that M is well-dened we need to check the following equality.
S KETCH OF PROOF. Let us denote the closed cube 0, 1 n by R and let U be the open cube 0, 1 n . Let V U c 1 c U and V U c 1 c U . The in R are negligeable in the sense that complement of V and of V
R V
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 c is a bijection from V onto V. It is orientation-preserving, because c and c are orientation-preserving. Therefore, by Theorem 5.1,
Combining this with the equalities (9.1) we get the result.
Not every manifold can be covered with one single n-cube. However, it can be shown that there always exists a nite collection of n-cubes c i : 0, 1 n M for i 1, 2, . . . , k, such that M, (i) k 1 ci 0, 1 n i n (ii) ci 0, 1 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 embedding.
We can then dene
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 1, resp. 2, we is vol M M , where is the volume form on M. (If dim M speak of the arc length, resp. surface area of M.) The integral of a function f on M is dened as M f . The mean or average of f is the number f vol M 1 M f . n whose i-th coordinate is the The centroid or barycentre of M is the point x in R mean value of xi over M, i.e.
The volume form depends on the embedding of M into R N , so the notions dened 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 Theorem 5.10 and is proved in a similar way.
xi
1 vol M
xi .
i 1
0,1
ci ,
c c
c . QED
and
2
V
2
0,1
n
9.6. L EMMA .
0,1
c .
c .
(9.1)
108
9.8. T HEOREM (Stokes theorem for manifolds). Let be an n 1-form on a compact oriented n-manifold with boundary M. Give the boundary M the induced orientation. Then
M
9.3. Gau and Stokes Stokes theorem, Theorem 9.8, contains as special cases the integral theorems of vector calculus. These classical results involve a vector eld F n 1 Fi ei dei n . As discussed in Section 2.5, to this vector eld ned on an open subset U of R corresponds a 1-form F dx n 1 Fi dxi , which we can think of as the work i done by the force F along an innitesimal 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 function g, then 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 fundamental theorem of calculus in Rn ,
M
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 eld. Thus the potential energy of the particle decreases by the amount of work done. Gau divergence theorem. We have
If N is a oriented hypersurface in R with positive unit normal n, then F n N on N by Theorem 8.14. In this situation it is best to think of F as the ow vector eld of a uid, where the direction of F x gives the direction of the ow at a point x and the magnitude F x gives the mass of the amount of uid passing per unit time through a hypersurface of unit area placed at x perpendicular to the vector F x . Then describes the amount of uid passing per unit time and per unit area through the hypersurface N. For this reason the n 1-form is also called the ux of F, and its integral over N the total ux through N. Applying Stokes theorem to a compact domain M in Rn we get M d M . Written in terms of the vector eld F this is Gau divergence theorem,
M M
Thus the total ux out of the hypersurface M is the integral of div F over M. If the uid is incompressible (e.g. most liquids) then this formula leads to the interpretation of the divergence of F (or equivalently d ) as a measure of the sources or sinks of the ow. Thus div F 0 for an incompressible uid without sources
div F dx1 dx2
dxn
F n M .
dx
and
div F dx 1 dx2
dxn .
F dx
g b
g a .
EXERCISES
109
Classical version of Stokes theorem. Now let M be a compact two-dimensional oriented surface with boundary and let us rewrite Stokes theorem M d M in terms of the vector eld 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 d curl F dx, so d curl F dx. Hence if n is the positive unit normal of the surface M in R3 , then d curl F n M on M. In this way we get the classical formula of Stokes,
M M
In other words, the total ux 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 equivalently d , can be regarded as a measure of the vorticity of the vector eld.
Exercises
R be two smooth functions 9.1. Let U be an open subset of Rn and let f , g : U g x for all x in U. Let M be the set of all pairs x, y such that x in U and satisfying f x f x y g x . (i) Draw a picture of M if U is the open unit disc given by x 2 y2 1 and f x, y 1 x2 y2 and g x, y 2 x2 y2 . (ii) Show directly from the denition 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? (iii) Give an example showing that M is not necessarily a manifold with boundary if the condition f x y g x fails.
9.2.
x dy y dx and let M be a compact domain in the plane R 2 . (i) Let Show that M is twice the surface area of M. (ii) Apply the observation of part (i) to nd the area enclosed by the astroid x cos3 t, y sin3 t. x dx and let M be a compact domain in R n . Show that M is a (iii) Let constant times the volume of M. What is the value of the constant?
9.3. Write the divergence theorem for the vector eld F cx n en on Rn , where c is a positive constant. Deduce Archimedes Law: the buoyant force exerted on a submerged body is equal to the weight of the displaced uid. E ! by
(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 and c dx1 dx2 dx3 . (iii) Find the volume of the solid parametrized by c. (iv) Find the surface area of the boundary of this solid.
r 1 2
R1 R1
r cos 2 cos 1 r cos 2 sin 1 r sin 2
9.4. Let R1
R2
0 be constants. Dene a 3-cube c : 0, R 2
0, 2
0, 2
R3
curl F n M
F dx.
or sinks. If the uid is a gas and if there are no sources or sinks then div F x (resp. 0) indicates that the gas is expanding (resp. being compressed) at x.
110
9.5. Let M be a compact domain in Rn . Let f and g be smooth functions on M. The Dirichlet integral of f and g is D f , g dx 1 dx2 dxn is M grad f grad g , where the volume form on M. (i) Show that d f dg grad f grad g . (ii) Show that d dg g , where g n 1 2 g x2 . i i (iii) Deduce from parts (i)(ii) that d f dg grad f grad g f g . (iv) Let n be the outward-pointing unit normal vector eld on M. Write g n for the directional derivative Dg n grad g n. Show that g . n M (v) Deduce from parts (iii) and (iv) Greens formula,
M
dg
g n M
D f, g
f g .
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 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 . (i) Deduce from Corollary 8.15 that the volume form on S R is the restriction of to S R , where is as in Exercise 2.16. Conclude that A n R S R . Rn Vn and An R Rn 1 An . (Substitute y Rx in the (ii) Show that Vn R volume forms of B R and S R .) (iii) Let f : 0, R be a continuous function. Dene g : R n R by g x f x . Use Exercise 2.16(ii) to prove that
B R
g dx1 dx2
dxn
f r An r dr
An
f r rn
dr.
(iv) Show that
r2
dr
An
r2 n 1
dr.
e r in part (iii) and let R .) (Take f r (v) Using Exercises B.10 and B.11 conclude that
n 2
(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 2 R 3
4 3 3 R
Vn R An R
R2
n 2
Vn
whence
V2m
and
V2m
21&
m m!
(vi) By taking f r (vii) Deduce that
1 in part (iii) show that A n
nVn and An R 2m 1 3 5
An
whence
A 2m
and
A2m
&21
2 2
2 m m 1 !
2m 1 3 5
1 m
2m
Vn R
1 m
2m
&1 2
g n
f n M
(vi) Deduce Greens symmetric formula,
f g
g f .
2&
. R.
EXERCISES
111

lim
xx
1 2
ex
2 .

(ix) Find limn
An , limn
Vn and limn
An
An . Use Stirlings formula,
CHAPTER 10
Applications to topology
10.1. Brouwers xed point theorem Let M be a manifold, possibly with boundary. A retraction of M onto a subset A is a smooth map : M A such that x x for all x in A. For instance, let M be the punctured unit ball in n-space,
Then the normalization map x x x is a retraction of M onto its boundary M, the unit sphere. The following theorem says that a retraction onto the A 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 : M M was a retraction. Let us choose an orientation of M and equip M with the induced orientation. Let M be the volume form on the boundary (relative to some embedding of M into R N ). Let be its pullback to M. Let n denote the dimension of M. Note that is an n 1-form on the n 1-manifold M, so d 0. Therefore d d d 0 and hence by Stokes theorem 0 M d M . But is a retraction onto M, so the restriction of to M is the identity map and therefore on M. Thus
which is a contradiction. Therefore does not exist.
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 xed point of f if f x x. 10.2. T HEOREM (Brouwers xed point theorem). Every smooth map from the closed unit ball into itself has at least one xed point. P ROOF. Let M x Rn x 1 be the closed unit ball. Suppose f: M M was a smooth map without xed points. Then f x x for all x. For each x in the ball consider the haline starting at f x and pointing in the direction of x. This haline intersects the unit sphere M in a unique point that we shall call
113
C$ G# "
$ D# "
C$
# "
2 F0 0 (
A$ B
B A$
vol M
0, QED
2 0 1) 0
0 65 0
A$ 9
$ 4# "
'
% &$ !
Rn 0
$# "
9 @$
'
1 .
% E$
114
10. APPLICATIONS TO TOPOLOGY
x , as in the following picture.
This denes a smooth map : M M. If x is in the unit sphere, then x x, so is a retraction of the ball onto its boundary, which contradicts Theorem 10.1. Therefore f must have a xed 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 originally 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 denition of homotopy.) The theorem also remains valid if the closed ball is replaced by a closed cube or a similar shape. 10.2. Homotopy Denition and rst examples. Suppose that 0 and 1 are two maps from a manifold M to a manifold N and that is a form on N. What is the relationship between the pullbacks 0 and 1 ? There is a reasonable answer to this question if 0 can be smoothly deformed into 1 . More formally, we say that 0 and 1 are homotopic if there exists a smooth map : M 0, 1 N such that x, 0 0 x and x, 1 1 x for all x in M. The map is called a homotopy. Instead of x, t we often write t x . Then each t is a map from M to N and we can think of t as a family of maps parametrized by t in the unit interval that interpolates between 0 and 1 , or as a one-second movie that at time 0 starts at 0 and at time 1 ends up at 1 . 10.3. E XAMPLE . Let M N Rn and 0 x x (identity map) and 1 x 0 (constant map). Then 0 and 1 are homotopic. A homotopy is given by x, t 1 t x. This homotopy collapses Euclidean space onto the origin by moving each point radially inward. There are other ways to accomplish this. For instance 1 t 2 x and 1 t2 x are two other homotopies between the same maps. We can also interchange 0 and 1 : if 0 x 0 and 1 x x, then we nd a homotopy by reversing time (playing the movie backwards), x, t tx. 10.4. E XAMPLE . Let M N be the punctured Euclidean space R n 0 and let 0 x x (identity map) and 1 x x x (normalization map). Then 0 and 1 are homotopic. A homotopy is given for instance by x, t x x t or by x, t 1 t x tx x . 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.
S DI
S 4I
S DI H
S 4I H
h 6g h
d ca b
S iI
S DI
Q P
R XW
S DI H
V U
h 6g h
SÌ H
Q P
S fI H
SÌ H
h 6g h
Q P
y f y
Q P
x
I H I H
q I
S YI
a H
a pGI H S
S eI H
I H
I H a H
f x
a H
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 0 x x0 is homotopic to the identity map x x. A specic homotopy : M 0, 1 M from 0 to 1 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 t.) Example 10.3 shows that R n is contractible 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 0 is not contractible. Homotopy of curves. If M is an interval a, b 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.
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 : S1 0, 1 N can be pictured as a rubber band oating through N from time 0 until time 1.
N S
1
10.6. E XAMPLE . Consider the two loops 0 , 1 : S1 R2 in the plane given 2 . A homotopy of loops is given by shifting by 0 x x and 1 x x 0 2t . What if we regard and as loops in the 0 to the right, t x x 0 1 0 punctured plane R2 0 ? Clearly the homotopy does not work, because it 1 and moves the loop through the forbidden point 0. (E.g. t x 0 for x 0 t 1 2.) In fact, however you try to move 0 to 1 you get stuck at the origin, so
t is r
6y
w xv t us r
F t s r t s r
Yy
w v
s r t
`s r t
116
it seems intuitively clear that there exists no homotopy of loops from 0 to 1 in the punctured plane. This is indeed the case, as we shall see in Example 10.13. The homotopy formula. The product M 0, 1 is often called the cylinder with base M. The two maps dened by 0 x x, 0 and 1 x x, 1 send M to the bottom, resp. the top of the cylinder. A homotopy : M 0, 1 M 0, 1 between these maps is given by the identity map x, t x, t . (Slide the bottom to the top at speed 1.)
1 0
base cylinder 1-form on the cylinder can be written as
J
with I running over multi-indices of degree k 1 and J over multi-indices of degree k. (Here we write the dt in front of the dxs 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, by taking the piece of involving dt and integrating it over the unit interval,
J 0
(In particular 0 for any that does not involve dt.) For a general manifold M we can write a k 1-form on the cylinder as dt , where and are forms on M 0, 1 (of degree k 1 and k respectively) that do not involve dt. We 1 then dene 0 dt. The following result will enable us to compare pullbacks of forms under homotopic maps. It can be regarded as an application of Stokes theorem, but we shall give a direct proof.
P ROOF. We write out the proof for an open subset of Rn . The proof for arbitrary manifolds is similar. It sufces to consider two cases: f dx I and g dt dx J .
e s
t s
d .
10.7. L EMMA (cylinder formula). Let M be a manifold. Then 1 d for all k 1-forms on M 0, 1 . In short,
o p
g J x, t dt dx J .
i f d nml
k 1
0, 1
fI
If M is an open subset of Rn , a k
x, t dx I
10
gJ
x, t dt dx J ,
M ,
d Dd e f D `d Dd e d e
r @e
10.2. HOMOTOPY
117
so
0
so
i 1
Now suppose we have a pair of maps 0 and 1 going from a manifold M to a manifold N and that : M 0, 1 N is a homotopy between 0 and 1 . For x in M we have 0 x x, 0 0 x , in other words 0 0 . Similarly 1 1 . Hence for any k 1-form on N we have 0 0 and 1 1 . Applying the cylinder formula to we see that the pullbacks 0 and 1 are related in the following manner. 10.8. T HEOREM (homotopy formula). Let 0 , 1 : M N be smooth maps from a manifold M to a manifold N and let : M 0, 1 N be a homotopy from 0 to 1 . Then 1 0 d d for all k 1-forms on N. In short,
10.9. C OROLLARY. If 0 , 1 : M N are homotopic maps between manifolds and is a closed form on N, then 0 and 1 differ by an exact form. This implies that if the degree of is equal to the dimension of M, 0 and 1 have the same integral. 10.10. T HEOREM . Let M and N be manifolds and let be a closed n-form on N, where n dim M. Suppose M is compact and oriented and has no boundary. Let 0 and 1 be homotopic maps from M to N. Then
M M
Au y
1 .
In particular, if d
0 we get 1
d .
d .
v
u {
u {
Hence d
0 .
i 1
i 1
y w
y w
xi
y}
Also
g x, t dt dx J , so
1
g x, t dt dxi dx J
y w
u
g x, t dt dxi dx J . xi
u
xi dxi dt dx J
Case 2. If
g dt dx J , then 0
0 and
xi dt dxi dx J ,
i
g x, t dt dxi dx J . xi QED
{
}~u
y w xu
f x, t dt dx I t f x, 1 f x, 0 dx I
f dt dx I t
xi dxi dx I
f dt dx I t
Case 1. If
f dx I , then
0 and d
0. Also terms not involving dt,
0 .
118
because M is empty.
A LTERNATIVE PROOF. Here is a proof based on Stokes theorem for the manifold with boundary M 0, 1 . The boundary of M 0, 1 consists of two copies of M, namely M 1 and M 0 , the rst of which is counted with a plus sign and the second with a minus. Therefore, if : M 0, 1 N is a homotopy between 0 and 1 ,
0 and the angle form If M is the circle S1 , N the punctured plane R2 y dx x dy x 2 y2 of Example 3.8, then a map from M to N is a loop in N and the integral of is 2 times the winding number of the loop. Thus Theorem 10.10 gives the following result.
10.12. E XAMPLE . Unfolding the three self-intersections in the curve pictured below does not affect its winding number.
10.13. E XAMPLE . The two circles 0 and 1 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 dened. 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 submanifold of the correct dimension and nding a nonzero answer. In a certain sense all obstructions to exactness are of this nature. We shall not attempt to say the last
10.11. C OROLLARY. Homotopic loops in R 2 about the origin.
0 have the same winding number
0,1
0,1
0,1
P ROOF. By Corollary 10.9, 1 by Stokes theorem
d for an n
1-form on M. Hence 0, QED
0 .
QED
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 df 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 denition 0). So a closed 0-form is never exact (unless it is 0) for a rather uninteresting 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.
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.
10.16. E XAMPLE . The punctured plane R2 0 is not simply connected, because 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 Poincar lemma. On a contractible manifold all closed forms of positive degree are exact. 10.17. T HEOREM (Poincar lemma). All closed k-forms on a contractible manifold are exact for k 1. P ROOF. Let M be a manifold and let : M 0, 1 M be a contraction onto x0 and x, 1 x for all x. a point x0 in M, i.e. a smooth map satisfying x, 0 Let be a closed k-form on M with k 1. Then 1 and 0 0, so putting we get
Here we used the homotopy formula, Theorem 10.8, and the assumption that d 0. Hence d . QED
The proof provides us with a formula for the antiderivative, namely , which can be made quite explicit in certain cases.
so
P ROOF. Let be a closed 1-form and c a loop in M. Then c is null-homotopic, 0 by Proposition 10.14. The result now follows from Theorem 4.5. QED c
10.14. P ROPOSITION . Let c be a null-homotopic loop in M. Then closed forms on M.
0 for all
120
so
According to the proof of the Poincar lemma, the function satises d pro0. It is instructive to compare with the function f constructed vided that d in the proof of Theorem 4.5. (See Exercise 10.5.) Another typical application of the Poincar lemma is showing that a manifold is not contractible by exhibiting a closed form that is not exact. For example, the punctured plane R2 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 0 ,
P ROOF. d 0 follows from Exercise 2.1(ii). The n 1-sphere M S n 1 has unit normal vector eld x, so by Corollary 8.15 on M we have , the volume form. Hence M vol M 0. On the other hand, suppose was exact, d for an n 1-form . Then
M M M
by Stokes theorem, Theorem 9.8. This is a contradiction, so is not exact. It now follows from the Poincar lemma, Theorem 10.17, that R n 0 is not contractible. QED Using the same form , but restricting it to the unit sphere S n 1 , we see that is not contractible. But how about forms of degree not equal to n 1? Without proof we state the following fact.
is the winding number of M about the origin. It generalizes the winding number of a closed curve in R2 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.
Rn
For a compact oriented hypersurface without boundary M contained in in 0 the integral 1 x dx n 1 M x n vol n 1 S
10.20. T HEOREM . On Rn 1 is exact.
0 and on Sn
every closed form of degree k
Sn 1
10.19. T HEOREM . is a closed but non-exact n Hence punctured n-space is not contractible.
dx
n
. 1-form on punctured n-space.
xi
gi tx dt.
gi
tx d txi
gi
tx xi dt
10.18. E XAMPLE . Let M be Rn and let x, t Let i gi dxi be a 1-form. Then
tx be the radial contraction. t dxi ,
1,
10.3. CLOSED AND EXACT FORMS RE-EXAMINED
121
Contractibility versus simple connectivity. Theorems 10.15 and 10.17 suggest that the notions of contractibility and simple connectivity are not independent. 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 M M
c1
Formally, let c1 : S1 M be a loop, : M 0, 1 M a contraction of M onto x0 . Put c s, t c1 s , t . Then c is a homotopy between c 1 and the constant loop c0 t c1 s , 0 x0 positioned at x0 . QED As mentioned in Example 10.16, the sphere S n 1 and punctured n-space Rn 0 are simply connected for n 3, although it follows from Theorem 10.19 that they are not contractible. Thus simple connectivity is weaker than contractibility. The Poincar conjecture. Not long after inventing the fundamental group Poincar posed the following question. Let M be a compact three-dimensional manifold without boundary. Suppose M is simply connected. Is M homeomorphic to the three-dimensional sphere? (This means: does there exist a bijective map M S3 which is continuous and has a continuous inverse?) This question became (inaccurately) known as the Poincar conjecture. It is famously difcult and was the force that drove many of the developments in twentieth-century topology. It has an n-dimensional analogue, called the generalized Poincar 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 5 of the generalized Poincar conjecture conjecture was the easiest and was conrmed by S. Smale in 1960. The case n 4 was done by M. Freedman in 1982. The case n 3, the original version of the conjecture, turned out to be the hardest, but was nally conrmed by G. Perelman in 2002-03. For a discussion and references, see the paper Towards the Poincar conjecture and the classication 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 .
X``
m u
122
Exercises
10.1. Write a formula for the map guring in the proof of Brouwers xed 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 M if for all x M the straight line segment joining x 0 to x is entirely contained in M. Show that if M is starshaped 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 , x of radius centred at x is convex. (iii) Same for the open ball B , x . 10.5. Let x Rn and let cx be the straight line connecting the origin to x. Let be a 1-form on Rn and let be the function dened in Example 10.18. Show that x cx .
m 1
1 0
I l 1
10.9. Let M and N be manifolds and 0 , 1 : M N homotopic maps. Show that 0 1 for all closed k-chains c in M and all closed k-forms on N. c c
10.10. Prove that any two maps 0 and 1 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 x y0 .) 10.11. Let x0 2, 0 and let M be the twice-punctured plane R 2 0, x0 . Let c1 , c2 , c3 : 0, 2 M be the loops dened by c 1 t cos t, sin t , c 2 t 2 cos t, sin t and c3 t 1 2 cos t, 2 sin t . Show that c 1 , c2 and c3 are not homotopic. (Construct a 1-form on M such that the integrals c1 , c2 and c3 are distinct.)
Show that d . (Use d also Exercise 2.7.)
0 and apply the identity proved in Exercise B.5 to each f I ; see
l 1
xil f I dxi1 dxi2
dxil
dxik .
10.8. Let dened on Rn
I f I dx I be a closed k-form whose coefcients f I are smooth functions 0 that are all homogeneous of the same degree p k. Let
h tx, ty, tz t dt
y dz
f tx, ty, tz t dt
x dy
y dx
g tx, ty, tz t dt
z dx
x dz
z dy .
10.7. Let f dx dy g dz dx h dy dz be a 2-form on R3 and let x, y, z, t t x, y, z be the radial contraction of R3 onto the origin. Verify that
and check directly that d
for k
1.
m 1
f tx tk
dt xim dxi1 dxi2
10.6. Let be the k-form f dx I f dxi1 dxi2 be the radial contraction x, t tx. Verify that
dxik on Rn and let : Rn dxim
dxik ,
0, 1
Rn
EXERCISES
123
10.12. A function g : R R is 2 -periodic if g x 2 g x for all x. (i) Let g : R R be a smooth 2 -periodic function and let g dt, where t is the coordinate on R. Prove that there is a unique number k such that k dt dh for some smooth 2 -periodic function h. (To nd k, integrate the equation k dt dh over 0, 2 . Then check that this value of k works.) (ii) Let be any 1-form on the unit circle S 1 and let be the element of arc length of S1 . (You can think of as the restriction to S 1 of the angle form.) Prove that there is a unique number k such that k 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 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 dened as the set of all x such that x X and x Y. X Y: the union of X and Y. This is dened as the set of all x such that x X or x Y. X Y: the complement of Y in X. This is dened 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 denition 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
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
'
! C0 )
A!
X x
Y is the complement X
&
x x
X or x
Y is the union X
Y,
x x
X and x
Y is the intersection X
"
R 1
3 is the interval 1, 3 , Y, Y.
0 )
0 ) #
"
"
"
X P x ples:
: the set of all x
X which have the property P x . Exam-
12( !
5 64
'
&
! " "
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 denition the set
(This is a somewhat confusing notation. It is not meant to imply that f is required to have an inverse.) f 1 c : an abbreviation for f 1 c , i.e. the set x X f x c . This is often called the bre or level set of f at c. f A: the restriction of f to A. If A is a subset of X, f A is the function dened by
A function f : X Y is injective or one-to-one if x 1 x2 implies f x1 f x2 . (Equivalently, f is injective if f x 1 f x2 implies x1 x2 .) It is called surjective or onto if f X Y, i.e. if y Y then y f x for some x 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 1 : Y X satisfying f 1 f x x for all x X 1 y y for all y Y. and f f If X is a nite 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 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 1, 2, . . . , n , one uses the familiar notation n 1 f i . In these notes we will often deal with indices which are pairs i or k-tuples of integers, also known as multi-indices. As a simple example, let n be a xed nonnegative integer, let X be the set of all pairs of integers i, j satisfying 0 i j n, and let f i, j i j. For n 3 we can display X and f in a tableau as follows. j 3 2 1 0 4 3 2 i 5 4 6
E D
aF fE
E D
E D F E eE D D
F E feE D D
aF
F dE D
E D
F VE
F TE
E D
E F
F TEeE D
E D
S D
In other words, f A is equal to f on A, but forgets the values of f at points outside A. f : the composition of f and g. If f : X Y and g : Y Z are functions, then g f : X Z is dened by g f x g f x . We often say that the function g f is obtained by substituting y f x into g y .
aP
F YE XE D
f A x
f x if x not dened if x
A, A.
F WE D
P VE D
E R HD
E D
H F UTE
Q D
E D
B : the preimage of B under the map f . If B is a subset of Y, this is by denition the set f
1
f x
B .
E D
H F IGE
f A
Y y
f x for some x
A .
EXERCISES
127
0 i j 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 about a point x is the collection of all points y whose distance to x is less than ,
A subset O of Rn is open if for every x O there exists an 0 such that B , x 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 C is open. This denition 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 about a point x, which is dened as the collection of all points y whose distance to x is less than or equal to ,
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 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 0 such that x R for all x in A. (That is, A is contained in the ball B R, 0 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-dened, nite number. Exercises
A.1. Parts (iii) and (iv) of this problem require the use of an atlas (or the Web; see e.g. ). Let X be the surface of the earth, let Y be the real line and let
y x UGs
B , x
Rn
r v
Ts y x
B , x
Rn
u r
t t t
r wv
q i om f k f k f i h f pnlXjjjgeejggeed
s r
The sum x
f x of all these numbers is written as
j .
128
A. SETS AND FUNCTIONS
f: X Y be the function which assigns to each x X its geographical latitude measured in degrees. (i) Find f X . (ii) Find f 1 0 , f 1 90 , f 1 90 . (iii) Let A be the contiguous United States. Find f A . Round the numbers to whole degrees. f A , where A is as in part (iii). Find (a) a country other than A that (iv) Let B is contained in f 1 B ; (b) a country that intersects f 1 B but is not contained in f 1 B ; and (c) a country in the northern hemisphere that does not intersect f 1 B . (i) S 0 (ii) S n
3 0 and S n 1 S n 1 n 2 . 2 n 1 n n 1 n 2 . (Use induction on n.) 2
A.3. Prove that the open ball B , x is open. (This is not a tautology! State your reasons as precisely as you can, using the denition of openness stated in the text. You will y z z x .) need the triangle inequality y x A.4. Prove that the closed ball is B , x is closed. (Same comments as for Exercise A.3.) A.5. Show that the two denitions of closedness given in the text are equivalent. in Rn , A.6. Complete the following table. Here S n that is the set of vectors of length 1. closed? 3, 5 3, 5 3, 3, B , x B , x Sn 1 xy-plane in R3 unit cube 0, 1 n
denotes the unit sphere about the origin compact? yes
bounded? yes
yes
| eu t
2 |
}u t |
t X
u | ~u | t t u | t y | t { { {
v u t u t u xt w u w u w
u t y yu t y zu t
A.2. Let S n
i j n
j . Prove the following assertions.
u t v
w xt v
u t v
t v
u t v u t t
u t v u t v
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 a
There are two useful alternative ways of writing this theorem. Replacing b with x and differentiating with respect to x we nd
a
Writing g instead of F and g instead of f and adding g a to both sides in formula (B.1) we get
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 denite integral: it tells you how to nd 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 1 , 2 , . . . , m be functions of n variables x 1 , x2 , . . . , xn . As usual we write x , x1 x2 . . .
and view x (In calculus the word map is often used for vector-valued functions, while the word function is generally reserved
as a single map from Rn
to Rm .
129
xn
m x
1 x 2 x . . .
g x
g a
g t dt.
d dx
f t dt
f x .
f t dt
F b
F a .
(B.1)
(B.2)
(B.3)
130
B. CALCULUS REVIEW
0 0 0 0 0 1 n are the standard basis vectors of R . The (total) derivative or Jacobi matrix of at x is then the m n-matrix D x
for real-valued functions.) We say that is continuously differentiable if the partial derivatives i x he j i x i x lim (B.4) x j h h 0 1, 2, . . . , n and j 1, are well-dened and continuous functions of x for all i 2, . . . , m. Here 1 0 0 0 1 0 0 0 0 e1 . , e2 . , . . . , en . . . . . . .
If v is any vector in Rn , the directional derivative of along v is dened to be the vector D x v in Rm , obtained by multiplying the matrix D x by the vector v. For n 1 is a vector-valued function of one variable x, often called a path or (parametrized) curve in Rm . In this case the matrix D x consists of a single column vector, called the velocity vector, and is usually denoted simply by x . For m 1 is a scalar-valued function of n variables and D x is a single row vector. The transpose matrix of D x is therefore a column vector, usually called the gradient of : D x T grad x . The directional derivative of along v can then be written as an inner product, D x v grad x v. There is an important characterization of the gradient, which is based on the identity a b a b cos . Here 0 is the angle subtended by a and b. If v is a unit vector ( v 1), then where is the angle between grad x and v. So D x v takes on its maximal value if cos 1, i.e. 0. This means that v points in the same direction as grad x . Thus the direction of the vector grad x is the direction of steepest ascent, i.e. in which increases fastest, and the magnitude of grad x is equal to the directional derivative D x v, where v is the unit vector pointing along grad x . Frequently a function is not dened on all of Rn , but only on a subset U. We must be a little careful in dening the derivative of such a function. Let us assume that U is an open set. Let : U Rm be a function dened on U and let x U. Because U is open, there exists 0 such that the points x te j are contained t . Therefore i x te j is well-dened for t and in U for 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 is called continuously differentiable or C 1 .
D x v
grad x
grad x
cos ,
m x 1
...
m x n
. . .
. . .
1 x 1
...
1 x n
B.3. THE CHAIN RULE
131
If the second partial derivatives
exist and are continuous, then is r times continuously differentiable or C r . If is Cr for all r 1, then we say that is innitely many times differentiable, C , or smooth. This means that 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 : A B and : B C are functions, we can apply after to obtain the composite function x x .
exist and are continuous for all x U and for all i 1, 2, . . . , n and j, k 1, 2, . . . , m, then is called twice continuously differentiable or C 2 . Likewise, if all r-fold partial derivatives r i x x j1 x j2 x jr
Here D x D x denotes the composition or the product of the two matrices D x and D x . B.2. E XAMPLE . In the one-variable case n m k 1 the derivatives D and D are 1 1-matrices x and y , and matrix multiplication is ordinary multiplication, so we get the usual chain rule B.3. E XAMPLE . If n able x, so D is a 1 D x k 1, then is a real-valued function of one vari1-matrix containing the single entry . Moreover, and D y
y 1
so by the chain rule
This is perhaps the most important special case of the chain rule. Sometimes we are sloppy and abbreviate this identity to
d dx
i 1
yi
di . dx
i 1
d dx
d m dx
D x D x
yi
di x . dx
. . .
d 1 dx
...
y m
x x .
for all x
U.
D x
D x D x
B.1. T HEOREM (chain rule). Let U Rn and V and : V Rk be Cr . Then is Cr and
Rm be open and let : U
V (B.5)
2 i x x j xk
XX
132
B. CALCULUS REVIEW
An even sloppier, but nevertheless quite common, notation is d dx
i 1
yi
di . dx
In these notes we frequently use the so-called pullback notation. Instead of we often write , so that x stands for x . Similarly, yi x stands for y i x . In this notation we have
i 1
B.4. The implicit function theorem be a continuously differentiable function dened on an open Let : W subset W of Rn m . Let us think of a vector in Rn m as an ordered pair of vectors u, v with u Rn and v Rm . Consider the equation 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 with respect to the u- and v-variables separately, . . . . . . . . . . . . Rm
Observe that the matrix Dv is square. We are in business if we have a point u0 , v0 at which is 0 and Dv is invertible. B.4. T HEOREM (implicit function theorem). Let : W Rm be Cr , where W n m . Suppose that u , v is open in R W is a point such that u0 , v0 0 and 0 0 Dv u0 , v0 is invertible. Then there are open neighbourhoods U R n of u0 and V Rm of v0 such that for each u U there exists a unique v f u V satisfying u, f u 0. The function f : U V is C r with derivative given by implicit differentiation:
This is well-known for m n 1, when is a function of two real variables u, v . If v 0 at a certain point u 0 , v0 , then for u close to u0 and v close to v0 we can solve the equation u, v 0 for v as a function v f u of u, and
Now let us take to be of the form u, v g v u, where g : W R n is a n . Solving u, v given function with W open in R 0 here amounts to inverting the function g. Moreover, Dv Dg, so the implicit function theorem yields the following result.
u . v
for all u
U.
Df u
Dv u, v
Du u, v
v f u
m u 1
...
m u n
m v 1
...
m v m
Du
D v
1 u 1
...
1 u n
1 v 1
...
1 v m
u, v
0.
d dx
yi
di . dx
(B.6)
B.5. THE SUBSTITUTION FORMULA FOR INTEGRALS
133
B.5. T HEOREM (inverse function theorem). Let g : W R n be continuously n differentiable, where W is open in R . Suppose that v0 W is a point such that Dg v0 is invertible. Then there is an open neighbourhood U R n of v0 such that g U is an open neighbourhood of g v0 and the function g : U g U is invertible. The inverse g 1: V U is continuously differentiable with derivative given by
Again let us spell out the one-variable case n 1. Invertibility of Dg v 0 simply means that g v0 0. This implies that near v 0 the function g is strictly monotone increasing (if g v0 0) or decreasing (if g v0 0). Therefore if I is a sufciently small open interval around u 0 , then g I is an open interval around g u0 and the restricted function g : I g I is invertible. The inverse function has derivative 1 g 1 u , g v
B.6. E XAMPLE (square roots). Let g v v 2 . Then g v0 0 whenever v0 0. For v0 0 we can take I 0, . Then g I 0, ,g 1 u u, and g 1 u 1 2 u . For v0 0 we can take I , 0 . Then g I 0, , g 1 u u, and g 1 u 1 2 u . In a neighbourhood of 0 it is not possible to invert g. B.5. The substitution formula for integrals R be a function. Suppose we Let V be an open subset of Rn and let f : V want to change the variables in the integral V f y dy. (This is shorthand for an n-fold integral over y 1 , y2 , . . . , yn .) This means we substitute y p x , where V is a map from an open U Rn to V. Under a suitable hypothesis we p: U 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 V be a map. Suppose that p is bijective and that p and its inverse are and let p : U continuously differentiable. Then for any integrable function f we have
This can be written succinctly as cd f y dy more similar to the multidimensional case.
f y dy
e e
b a b a
f p x p x dx if p is increasing, f p x p x dx if p is decreasing.
b a
f p x
p x dx, which looks
Again this should look familiar from one-variable calculus: if p : a, b is C 1 and has a C 1 inverse, then
f y dy
f p x
det Dp x dx. c, d
with v
u .
for all v
V.
Dg
Dg v
v g
134
B. CALCULUS REVIEW
Exercises
(ii) Show that

0
(Integrate the formula in part (i) by parts and dont forget to use the chain rule.) (iii) By induction on n deduce from part (ii) that k!
0
k 0
This is Taylors formula with integral remainder term.
B.4. According to Newtons 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
(i) Show that the functions f x, y x 2 xy x2 y2 , f x, y x3 y3 , 2 are homogeneous. What are their degrees? 2 z6 4 y2 z2 f x, y, z x 3x 0. (ii) Assume that f is dened at 0 and continuous everywhere. Show that p Show that f is constant if p 0. (iii) Show that if f is homogeneous of degree p and smooth, then
i 1
B.6. Dene a function f : R R by f 0 0 and f x e 1 x for x 0. (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 ne, but pay special attention to the behaviour near x 0.
(i) Show that t lies on the unit sphere S n
about the origin.
X
@ @

2t
@ @
B.7. Dene a map from Rn
to Rn by 1 t
2
1 en .
V
U Q
#
2 1

(Differentiate the relation f tx
t p f x with respect to t.)

2

xi xi
0
pf x .
I HG
A
HG
TR S Q

P
8 E 6

D
B.5. A function f : 0 R is homogeneous of degree p if f tx Rn 0 and t 0. Here p is a real constant.
C8 B 6
Rn

where G is a constant of nature. Show that F is the gradient of f x
Gm 1 m2
tp f
@ @
@ @
9
Gm1 m2 x, x 3
8 7 6
B.3. Deduce from the chain rule that D x v
lim
t

4
tv t
#

B.2. Let x and v be constant vectors in Rn . Dene c t
tv. Find c t .
2 1

"

g x
hk
2 1
hn 1 n!
t ng
n 1
##
"
) #
!
g a
hg a
h2
t g
th dt.
th dt.
(# #
"
& '
&
$#

%

g x
g a
h t
1 g a
th
1 0
h2
$#
"
!

B.1. Let g : a, b R be a 0. Suppose a x b and put where n x a. (i) By changing variables in the fundamental theorem of calculus (B.3) show that g x g a h
1
g a
th dt.
t g
th dt
C n 1-function,

x . x for all
EXERCISES
135
(ii) Show that 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 identifying it with t 1 , t2 , . . . , tn 1 , 0 .) (iii) Compute D t . (iv) Let X be the sphere punctured at the north pole, X Sn 1 en . Stereon 1 given by x graphic projection from the north pole is the map : X R xn 1 1 x1 , x2 , . . . , xn 1 . Show that is a two-sided inverse of . (v) Draw diagrams illustrating the maps and 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 dened as the unique intersection point of the line joining y to x and the hyperplane P. This denes a map : Sn 1 y P. The point y is called the centre of the projection. Write a formula for the stereographic projection from the south pole en and for its Sn 1 . inverse : Rn 1
B.9. Let a0 , a1 , a2 , . . . , an be vectors in Rn . A linear combination n 0 ci ai is convex if i 1. The simplex spanned by the ai s is the collection of all their convex linear combinations,
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.
R
1 det A , n! where A is the n n-matrix with columns a1 a0 , a2 a0 , . . . , an a0 . (First compute the volume of the standard simplex by repeated integration. Then map to the standard simplex by an appropriate substitution and apply the substitution formula for integrals.) The following two calculus problems are not review problems, but the results are needed in Chapter 9.
and prove the following assertions.
B.11. Calculate n 1 (where is the function dened in Exercise B.10) by establish2 1 ing the following identities. For brevity write 2 .
d x
(ii) 2
d x b d x b
(i)
s2
ds. e
x2 y2
dx dy.
a `
(iii)
a `
d x e `ba ` ba `
(i) (ii)
x n
x for all x 0. 1 ! for positive integers n. 2 1 a 1 e u u a du . 2 2 x
x a ` b
B.10. For x
0 dene
t x 1
dt
b gs
vol
y y y
(ii) The volume of a region R in Rn is dened as
r vv
i 0
i 0
tugs b
n 0 ci i
c i ai c i
1 .
dx1 dx2
dxn . Show that
a `
a `
b qa
e`
B.8. A map : Rn is even and C 1 .
Rm is called even if
x for all x in Rn . Find D 0 if
b pa `
f ge
a d b
` cb
i h fge
a ` d
a ` `
da
136
B. CALCULUS REVIEW
2n
d e
(v)
i h h jh
(iv)
. 1 2
1 3 5
(iii) 2
re
r2
dr d. 2n 1
for n
1.
Bibliography
[Bac06] D. Bachman, A geometric approach to differential forms, Birkhuser, Boston, MA, 2006. A recent text, a version of which is available through the authors website . P. Bamberg and S. Sternberg, A course in mathematics for students of physics, Cambridge University Press, Cambridge, 1991. 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. D. Bressoud, Second year calculus, Undergraduate Texts in Mathematics, Springer-Verlag, New York, 1991. Multivariable calculus from the point of view of Newtons law and special relativity with coverage of differential forms. 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. 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, including the Gau-Bonnet theorem. 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. H. Edwards, Advanced calculus: A differential forms approach, Birkhuser Boston, Boston, Massachusetts, 1994, corrected reprint of the 1969 original. One of the earliest undergraduate textbooks covering differential forms. Still recommended as an alternative or supplementary source. H. Flanders, Differential forms with applications to the physical sciences, second ed., Dover Publications, New York, 1989. Written for 1960s engineering graduate students. Concise but lucid (and cheap!). 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 computer graphics. 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 applications to topology. J. Hubbard and B. Hubbard, Vector calculus, linear algebra, and differential forms: A unied approach, second ed., Prentice Hall, Upper Saddle River, New Jersey, 2002. J. Marsden and A. Tromba, Vector calculus, fth ed., W. H. Freeman, New York, 2003. Standard multivariable calculus reference. 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. S. Singer, Symmetry in mechanics. A gentle, modern introduction, Birkhuser, Boston, MA, 2001. M. Spivak, Calculus on manifolds. A modern approach to classical theorems of advanced calculus, W. A. Benjamin, New York-Amsterdam, 1965. Efcient and rigorous treatment of many of the topics in these notes.
[BS91] [BT82]
[Bre91]
[Bre05]
[Car94]
[Dar94]
[Edw94]
[Fla89]
[Gra98]
[GP74]
[HH02] [MT03] [Opr03]
[Sin01] [Spi65]
s v n w y y z u o | pz y v w p s sr p o R'Rxvqtsqq!qj~}{!qRxvutj(%qjon
137
138
BIBLIOGRAPHY
, A comprehensive introduction to differential geometry, third ed., Publish or Perish, Houston, TX, 1999. Differential geometry textbook at advanced undergraduate level in ve 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 explanations of several of the ideas covered in these notes, as well as a number of straightforward exercises.
[Spi99]
The Greek alphabet

upper case lower case name A B E Z H I K
M N O
P T
, , o , ,
alpha beta gamma delta epsilon zeta eta theta iota kappa lambda mu nu xi omicron pi rho sigma tau upsilon phi chi psi omega
139
Notation Index
ei , i-th standard basis vector of Rn , 130 f A, restriction of f to A, 126 f , composition of f and g, 37, 126, 131 , Gamma function, 110, 135 grad, gradient of a function, 26 graph, graph of a function, 68 H n , upper halfspace in Rn , 103 I, multi-index i 1 , i 2 , . . . , i k (usually increasing), 17 int M, interior of a manifold with boundary, 103 ker A, kernel (nullspace) of a matrix A, 73 l , length of a permutation , 34
M , volume form of a manifold M, 96

n k , binomial coefcient, 19, 23 n, unit normal vector eld, 95 nullity A, dimension of the kernel of A, 73
, 94
141
D, Jacobi matrix of , 130 , boundary of a chain, 59 of a manifold, 103 d, exterior derivative, 20 , Laplacian of a function, 28 I, J , Kronecker delta, 86 i, j , Kronecker delta, 51 det A, determinant of a matrix A, 31 div, divergence of a vector eld, 26
, pullback of a form, 37, 88 of a function, 37, 132

Rn , Euclidean n-space, 1 rank A, dimension of the column space of A, 73 S n , unit sphere about the origin in Rn S n , permutation group, 33 sign , sign of a permutation , 34
1
C r , r times continuously differentiable, 131 curl, curl of a vector eld, 27
O n , orthogonal group, 76 k M , vector space of k-forms on M, 19, 82
B , x , closed ball in Rn , 127 B , x , open ball in Rn , 127 , orientation dened by a basis
A T , transpose of a matrix A, 35 A k V, set of alternating k-multilinear functions on V, 85 A , permutation matrix, 43 A I, J , I, J-submatrix of A, 42 , Hodge star of , 24 relativistic, 29 M , integral of over a manifold M, 106 c , integral of over a chain c, 47, 57 Alt , alternating form associated to , 90
, Hodge star operator, 24 relativistic, 29 0, 1 k , unit cube in Rk , 58 , orientation dened by a basis , 94 , Euclidean inner product (dot product), 8 , composition of maps, 37, 126, 131 , integral of a form over a chain, 57 over a manifold, 106 , integral of a form over a chain, 47 , Euclidean norm (length), 8 , tensor multiplication, 89 xi , partial derivative, 130 , orthogonal complement, 74 , exterior multiplication, 17, 85
dx, innitesimal hypersurface, 26 dx, innitesimal displacement, 25 dx I , short for dx i1 dx i2 dx ik , 17 dx i , covector (innitesimal increment), 17, 83 dx i , omit dx i , 18
, 8, 64
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 denitions or theorems; italic page numbers refer to examples or applications. In a few cases italic boldface is in order.
afne space, 7, 8, 73 alternating algebra, 82 multilinear function, 32, 85, 8790 property, 18, 19, 21 Ampre, Andr Marie (17751836), 29 Ampres Law, 29 angle form, 23, 36, 47, 51, 64, 118, 120, 123 function along a curve, 5253 anticommutativity, 18 antisymmetric matrix, 78 multilinear function, see alternating multilinear function arc length, 89, 96, 101, 107 Archimedes of Syracuse (287212 BC), 3, 109 Archimedes Law, 109 atlas, 67, 69, 82 average of a function, 107 ball, see closed ball, open ball barycentre, 107 bilinear, 85, 89 block, 31, 42, 91, 93, 96 rectangular, see rectangular block Bonnet, Pierre (18191892), 137 boundary of a chain, 60, 6265 of a manifold, 27, 103, 104111, 118 bounded, 47, 106, 127 Brouwer, Luitzen Egbertus Jan (18811966), 113, 122 Brouwers xed point theorem, 113, 122 Cartan, Elie (18691951), 17 Cartesian product, 10, 85, 116, 125 Cartesius, Renatus, see Descartes, Ren centroid, 107 chain, 59, 6065 chart, 67, 69, 72
143
circle, 9, 47, 48, 5153, 55, 62, 64, 72, 115, 118, 123 closed ball, 8, 105, 106, 110, 115, 122, 127 chain, 62, 64, 122 curve, 1, 50, 53, 62 form, 22, 2729, 40, 51, 54, 55, 117123 set, 4, 36, 47, 84, 103, 106, 107, 127, 128 codimension, 69, 73, 74, 105 cohomology, 137 column operation, 32, 42 vector, 1, 31, 45, 69, 83, 89, 92, 130 compact, 57, 106110, 117, 120, 127 complementary, 24 conguration space, 9, 14 connected component, 12, 119 conservative, 49, 108, 109 constant form, 19, 28, 83 continuously differentiable, 130 contractible, 115, 119122 contraction, 115, 120, 121122 contravariance, 38 convex, 122 linear combination, 135 coordinate map, see chart covariant vector, see covector covector, 83, 84, 85, 89, 90 eld, 83 Coxeter, Harold Scott MacDonald (19072003), 43 Coxeter relations, 43 critical point, 79 cross-cap, 7 cube in an open set, 59, 6062, 6465 curl, 27, 29, 65, 109 curvature, 17 curve, 1, 69 cycle, 62, 64 cylinder formula, 116
144
INDEX
with base M, 116 dAlembert, Jean Le Rond (17171783), 29 dAlembertian, 29 de Rham, Georges (19031990), 137 degenerate chain, 61, 64 degree of a form, 17 of a homogeneous function, 134 of a multi-index, 17 degrees of freedom, 9, 14 Descartes, Ren (15961650), 10, 85, 125 determinant, 31, 3235, 4042, 43, 78, 85, 86, 9194 differential equation, 15 form, see form dimension, 111, 69 Dirichlet, Lejeune (18051859), 110 Dirichlet integral, 110 disconnected, 12, 108 divergence, 26, 29, 65, 108, 109 domain, 106, 108, 109 of a function, 125 dot product, see inner product dual basis, 84, 86, 87, 89 vector, see covector space, 83 electromagnetic wave, 29 electromagnetism, 29 element of arc length, 89, 96, 101, 123 of surface area, 96 embedding, 67, 68, 70, 71, 7778, 81, 88, 9698, 100, 101, 103, 104, 106, 107 Euclid of Alexandria (ca. 325265 BC), 1, 67, 69, 91, 110, 114, 125, 127 Euclidean motion, 91 plane, 125 space, 1, 67, 69, 110, 114, 127 volume, 91 , 109 even map, 135 permutation, 34, 43 exact form, 22, 28, 4952, 64, 117120, 123 exterior algebra, 82 derivative, 20, 21, 25, 27, 60, 63 on a manifold, 82 differential calculus, 17 product, see product of forms Faraday, Michael (17911867), 29 Faradays Law, 29 bre, see level set
xed point, 113 ux, 26, 99, 108, 109 form as a vector-eating animal, 88 closed, see closed form exact, see exact form on a manifold, 82, 88 on Euclidean space, 17, 1830, 88 volume, see volume form free space, 29 Freedman, Michael (1951), 121 function, 125, 130 functional, see covector fundamental theorem of calculus, 27, 47, 49, 52, 64, 129, 134 in Rn , 49, 63, 108 Gamma function, 110111, 135 Gau, Carl Friedrich (17771855), v, 29, 63, 65, 95, 108, 137 Gau map, 95 Gau Law, 29 graded commutativity, 18, 19 gradient, 26, 28, 49, 65, 7479, 96, 100, 101, 108, 110, 130 Gram, Jorgen (18501916), 92 Gram-Schmidt process, 92 graph, 68, 70, 73, 97, 100, 103, 129 Gramann, Hermann (18091877), 82 gravitation, 54, 134 Greek alphabet, 139 Green, George (17931841), v, 63, 65, 110 Greens theorem, 65, 110 halfspace, 103 Hodge, William (19031975), 23, 25, 28, 29, 38 Hodge star operator, 23, 25, 28, 38, see also relativity homogeneous function, 28, 78, 122, 134 homotopy, 114 formula, 117 of curves, 115 of loops, 115, 119, 121 hypersurface, 26, 69, 95101, 106, 108, 120 increasing multi-index, 19, 24, 28, 41, 86, 87, see also complementary index of a vector eld, 55 inner product, 8, 84, 85, 98, 130 of forms, 28 integrability condition, 22 integral of a 1-form over a curve, 47, 48, 49, 51 of a form over a chain, 57, 5859, 6365, 106 over a manifold, 106, 107, 108, 117, 119, 120, 122 inversion, 34 inward pointing, 106
INDEX
145
k-chain, see chain k-cube, see cube k-form, see form k-multilinear function, see multilinear function Klein, Felix (18491925), 5 Klein bottle, 5 Kronecker, Leopold (18231891), 51 Kronecker delta, 51 Laplace, Pierre-Simon (17491827), 28 Laplacian, 28 Leibniz, Gottfried Wilhelm von (16461716), 20, 21, 40 Leibniz rule for forms, 21, 40 for functions, 20 length of a permutation, 34, 4243 of a vector, 8, 76, 79, 95, 106, 114, 128 level curve, 74 hypersurface, 74 set, 73, 126 surface, 74 lightlike, 29 line segment, 91 local representative, 82, 88, 89, 96 Lotka, Alfred (18801949), 15, 78 Lotka-Volterra model, 15, 78 manifold, 115, 69, 7079 abstract, 913, 69 given explicitly, 9, 72 given implicitly, 7, 72 with boundary, 27, 103, 104111, 118 map, 125, 130 Maxwell, James Clerk (18311879), 29 mean of a function, 107 measurable set, 36 Milnor, John (1931), 121 minimum, 75 Minkowski, Hermann (18641909), 29 Minkowski inner product, 29 space, 29 Mbius, August (17901868), 4, 105 Mbius band, 4, 105 multi-index, 17, 126, see also increasing multiindex multilinear algebra, 17 function, 85, 89 n-manifold, see manifold naturality of pullbacks, 38, 48, 58 Newton, Isaac (16431727), 54, 134, 137 norm of a vector, 8 normal vector eld, see unit normal vector eld
odd permutation, 34, 43 open ball, 127 neighbourhood, 127 set, 127, 128 is a manifold, 70 orientation of a boundary, 106, 108 of a hypersurface, 95, 100 of a manifold, 88, 95, 108 of a vector space, 91, 94, 100 preserving, 48, 57, 9597, 101, 106, 107 reversing, 48, 57, 95 orthogonal complement, 74, 95 group, 76, 78 matrix, 76, 91, 92 operator, 28 projection, 93 orthonormal, 28, 91 outward pointing, 106 outward-pointing, 110 pair of pants, 105 paraboloid, 4 parallelepiped, see block parallelogram, 13, 14, 91, 100 parametrization, 9, 57, 67 parametrized curve, 13, 47, 49, 5355, 7778, 130 partial differential equation, 22 operator, 20 path, see parametrized curve pentagon, 14 Perelman, Grigori (1966), 121 periodic function, 123 permutation, 33, 34, 35, 41, 4243, 85, 100 group, 33, 43 matrix, 43 pinch point, 7 plane curve, 1, 13, 69 Poincar, Jules Henri (18541912), 119, 121 Poincar conjecture, 121 lemma, 119 potential, 49, 53, 54, 108 predator, 15 prey, 15 product of forms, 19, 27 on a manifold, 82 of permutations, 34, 43 of sets, see Cartesian product product rule, see Leibniz rule projective plane, 5 pullback of a form
146
INDEX
on a manifold, 88, 97, 106, 114, 116, 117 on Euclidean space, 37, 4042, 47, 48, 57, 58, 82, 88 of a function, 132 punctured Euclidean space, 78, 114, 119121 plane, 47, 51, 64, 77, 115, 119, 120 quadrilateral, 11 rectangular block, 18, 57, 65 regular value, 73, 7479, 96, 100, 105 relativity, 10, 29, 137 reparametrization of a curve, 47, 48, 50, 58 of a rectangular block, 57, 58 restriction of a map, 47, 106, 107, 110, 126 retraction, 113 Riemann, Bernhard (18261866), 137 rigid body, 10 row operation, 42 vector, 1, 33, 45, 74, 83, 89, 130 saddle point, 75 Schmidt, Erhard (18761959), 92 sign of a permutation, 34, 4243 simple permutation, 43 simplex, 135 simply connected, 119, 121 singular cube, see cube value, 73, 7476, 78 singularity, 2, 8, 13, 14, 67, 71, 105 Smale, Stephen (1930), 121 smooth curve, 69 function or map, 131, 134 hypersurface, 69 manifold, 4 point, 2 surface, 69 solution curve, 9, 15 space curve, 1 space-time, 29 spacelike, 29 special linear group, 78 sphere, 4, 8, 10, 11, 14, 64, 67, 75, 79, 96, 100, 105, 106, 110, 113, 114, 119121, 128, 134 spherical coordinates, 44 pendulum, 9 standard basis of Rn , 130 orientation, 94 simplex, 135 star-shaped, 122 state space, see conguration space steepest ascent, 75, 130
stereographic projection, 72, 135 Stirling, James (16921770), 111 Stirlings formula, 111 Stokes, George (18191903), v, 47, 63, 65, 107, 109 Stokes theorem classical version, 65, 109 for chains, 47, 63, 64 for manifolds, 103, 108, 116, 118, 120 submanifold, 69 surface, 4, 14, 69, 81, 109 area, 9, 17, 96, 107, 109, 129 symmetric bilinear function, 85 matrix, 76, 79 tangent hyperplane, 69 line, 1, 13, 69, 70, 77 plane, 14, 69, 78 space, 1, 8, 14, 69, 70, 7376, 78, 88, 95, 96, 106 vector, 48, 69, 79, 88, 106 Taylor, Brook (16851731), 134 Taylors formula, 134 tensor product, 17, 85, 89 timelike, 29 topological manifold, 4 torus, 4, 68, 106 trace, 78 trajectory, 15, 78 transformation law, 82 transpose of a matrix or a vector, 1, 26, 35, 74, 83, 130 transposition, 42 unit circle, see circle cube, 35, 58, 65, 128 interval, 35, 58, 91, 114, 116 normal vector eld, 95, 98, 99, 101, 106, 108 110, 120 sphere, see sphere square, 35, 60, 62 vector, 51, 130 vector, see also column, length, row, unit, tangent eld, 24, 28, 29, 48, 55, 65, 98, 108, 109, see also conservative, curl, divergence, gradient, index, potential, unit normal Volterra, Vito (18601940), 15, 78 volume change, 36, 42 element, 17, 96 Euclidean, see Euclidean volume form, 65, 96, 97, 103 of a hypersurface, 99 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, 5455, 118, 120 of hypersurface, 120 work, 17, 25, 48, 49, 61, 108, 109 zero of a vector eld, 25

Manifold

Transféré par

Informations du document

Description originale:

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

Manifold

Transféré par

Droits d'auteur :

Formats disponibles

D EPARTMENT OF M ATHEMATICS , C ORNELL U NIVERSITY, I THACA , N EW Y ORK 14853-4201 E-mail address: URL:

8 92#(&7$) 654333 22&) ' " $ " !      1 0  ( " $ $ !  '&%"#    

Manifolds and Differential Forms

, where r is allowed to be negative,

Two-dimensional manifolds in three-dimensional space include a sphere, a paraboloid and a torus.

known as pinch points, where the surface is crinkled up.

is the norm or length of x and

1.4. CONFIGURATION SPACES

parametrizes the unit circle in R2 and

1.4. CONFIGURATION SPACES

1.2. Same questions as in Exercise 1.1 for the curve x

This means that when b are merged at a point.

d the two components of the conguration space

1.4. Sketch the surfaces dened by the following gluing diagrams.

(i) (ii) (iii) (iv)

freedom you may also assume the edge AB to be xed in place.)

Differential forms on Euclidean space

x1 x3 x5 dx1 dx6 dx3 dx2 ,

cos x2 dx5 dx3 ,

2. DIFFERENTIAL FORMS ON EUCLIDEAN SPACE

Written like this, a 2-form has at most

components. Likewise, a general n

1-form can be written as a sum of n components, dxn f n dx1 dx2 dxn

g2,3 dx2 dx3

g1,2 dx1 dx2

g1,n dx1 dxn g2,n dx2 dxn gn

f n,1 dxn dx1

f n,2 dxn dx2

f 2,1 dx2 dx1

f 2,2 dx2 dx2

f 1,1 dx1 dx1

f 1,2 dx1 dx2

f 1,n dx1 dxn f 2,n dx2 dxn

f n,n dxn dxn .

2.1. ELEMENTARY PROPERTIES

for all k-forms and all l-forms .

P ROOF. Let I i1 , i2 , . . . , ik and J the alternating property we get

2. DIFFERENTIAL FORMS ON EUCLIDEAN SPACE

which establishes the result.

If f is a 0-form, that is a smooth function, we dene d f to be the 1-form

Then we have the product or Leibniz rule:

g dy is a 1-form on R2 , then g x dx dy gx f y dx dy.

0 if is a form of odd degree. 2.2. The exterior derivative

For general forms

J g J dx J we get from this

2.2. THE EXTERIOR DERIVATIVE

which proves part (ii).

Here is one of the most curious properties of the exterior derivative.

f i, j dxk dxi dx j xk n f j,k dxi dx j dxk . xi

f j,k dxi dx j dxk xi n

For a general 2-form

f i,k dx j dxi dxk x j n (2.3) (2.4)

h dy dz is a 2-form on R 3 , then fz gy h x dx dy dz.

2. DIFFERENTIAL FORMS ON EUCLIDEAN SPACE

Applying the formula of Example 2.3 (replacing f i with f I xi ) we nd

2.9. E XAMPLE . Let

y cos yz dz. Then

A form is closed if d one less).

d for some form (of degree

0 for any form . In short, d2 0.

2.4. THE HODGE STAR OPERATOR

so is closed. Is exact? Let us solve the equations

By the formula in Example 2.4 the integrability condition d

8 92#(&7$) 654333 22&) ' " $ " ! 1 0 ( " $ $ ! '&%"#

$% " ! " ! # #