This action might not be possible to undo. Are you sure you want to continue?

**D EPARTMENT OF M ATHEMATICS , C ORNELL U NIVERSITY, I THACA , N EW Y ORK 14853-4201 E-mail address: ¢¡¤£¦¥§£¨£
**

©

¦¥§£¢
¨¢©¨¨¤

¢ URL:
¤
¨¤!¤!¤!"#¥§£

¨$¢©¨

%¢

&

¢¡¤£'¥§£¨£

©

Last updated: 2006-08-26T01:13-05:00 Copyright © Reyer Sjamaar, 2001. Paper or electronic copies for personal use may be made without explicit permission from the author. All other rights reserved.

Contents

Preface Chapter 1. Introduction 1.1. Manifolds 1.2. Equations 1.3. Parametrizations 1.4. Conﬁguration 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. Deﬁnition 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 deﬁnition 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

Closed and exact forms re-examined Exercises Appendix A. Integration and Stokes’ theorem on manifolds 9.2.5.1. 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 . Applications to topology 10. Second deﬁnition Exercises Chapter 8.2.1. The chain rule B. Orientations 8. n-Dimensional volume in R N 8.3. First deﬁnition 7.2.4.2. Derivatives B. Manifolds with boundary 9. Volume forms Exercises Chapter 9.3.2. Volume forms 8. Calculus review B. The implicit function theorem B.1.1.1.3. Integration over orientable manifolds 9.iv CONTENTS 7. Brouwer’s ﬁxed point theorem 10. Homotopy 10. The fundamental theorem of calculus B. Sets and functions A.1.2. Glossary A. General topology of Euclidean space Exercises Appendix B.3. Gauß and Stokes Exercises Chapter 10.

an appendix on linear algebra.0#132 )¨(0-4¢5¢6¨7.¤. a text of this kind has to walk a thin line between mathematical informality and rigour. The selection of material is similar to that in Spivak’s book [Spi65] and in Flanders’ book [Fla89]. one can skip over many of the details without too much loss of continuity. Plans for improvement include: more and better graphics. Manifolds and Differential Forms. I have tended to err on the side of caution by providing fairly detailed deﬁnitions and proofs. In class.Preface These are the lecture notes for Math 321. including basic linear algebra and multivariable calculus up to the integral theorems of Green. More than once the text makes use of results obtained in the exercises. These notes and eventual revisions can be downloaded from the course website at ()¨)¨*+-. Some prerequisites are brieﬂy reviewed within the text and in appendices. but treats manifolds of arbitrary dimension./. The course covers manifolds and differential forms for an audience of undergraduates who have taken a typical calculus sequence at a North American university. depending on the aptitudes and preferences of the audience and also on the available time. as well as more theoretical or conceptual problems.¤. I hope. but the treatment is at a more elementary and informal level appropriate for sophomores and juniors. perhaps including the theorems of Poincaré-Hopf and Gauß-Bonnet. 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. A large portion of the text consists of problem sets placed at the end of each chapter. Gauß and Stokes. as taught at Cornell University since the Fall of 2001. Because of its transitional nature between calculus and analysis. The exercises range from easy substitution drills to fairly involved but. At any rate. interesting computations.¤. This revised version of the notes is still a bit rough at the edges. the text is not restricted to curves and surfaces in three-dimensional space. With a view to the fact that vector spaces are nowadays a standard item on the undergraduate menu. a chapter on ﬂuid mechanics and one on curvature.

8¤9¤9:08¨.¢<.=¤>/?¤2¦132¨2¢6.¤4 9¨2.

>¨>8>.¨@¤A3B¦.

CD7.

¨8 E$0F(¤)1G9 .. suggestions and comments will be received gratefully. Ithaca. 2006-08-26 v . Corrections. NY.

.

. to save space. xn T . Manifolds Recall that Euclidean n-space Rn is the set of all column vectors with n real entries IJ LNM J x1 M J M x2 xH . . Curves can be closed (as in the ﬁrst picture below). but one important consequence of the deﬁnition is that a manifold has a well-deﬁned tangent space at every point. xn T T . a curve without kinks or self-intersections. we shall represent a column vector x as the transpose of a row vector. The dimension of a manifold in Rn can be no higher than n. . but you can have curves in any Rn . resp. the white dot at the other end indicates that 1 . or have one or two endpoints (the third picture shows a curve with an endpoint. unbounded (as indicated by the arrows in the second picture). A precise deﬁnition will follow in Chapter 6. x2 . intuitive introduction to manifolds and how they arise in mathematical nature.CHAPTER 1 Introduction We start with an informal. Most of this material will be examined more thoroughly in later chapters. indicated by a black dot. we shall usually separate the entries of a row vector by commas. This fact enables us to apply the methods of calculus and linear algebra to the study of manifolds. A manifold is a certain type of subset of Rn . The dimension of a manifold is by deﬁnition the dimension of its tangent spaces. . . column vectors are not to be confused with row vectors S x 1 x2 . yR z For reasons having to do with matrix multiplication. A one-dimensional manifold is. Instead of the tangent “space” at a point one usually speaks of the tangent line. K . . . . .1. . Occasionally. xn T . x HUS x1 . 1. A curve in R2 is called a plane curve and a curve in R3 is a space curve. x2 . . loosely speaking. For clarity. x H K yO . O xn which we shall call points or n-vectors and denote by lower case boldface letters. Dimension 1. . In R2 or R3 we often write I L x x x HQP . as in S x 1 .

the ﬁve-fold loop has ﬁve points of self-intersection. The teardrop has a kink. INTRODUCTION that point does not belong to the curve. The “good” points are called smooth. the self-intersections of the Archimedean spiral. A circle with one point deleted is also an example of a manifold. at each of which there are two distinct tangent lines. Think of a torn elastic band. which is given in polar coordinates by r is a constant times . the curve “peters out” without coming to an endpoint). 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 deﬁnition of a manifold (which we have not yet given). The four plane curves below are not manifolds. For instance. Singularities can sometimes be “resolved”.2 1. The bow tie and the ﬁve-pointed star have well-deﬁned tangent lines everywhere. By straightening out the elastic band we see that this manifold is really the same as an open interval. The points where these curves fail to be manifolds are called singularities. where two distinct tangent lines occur instead of a single well-deﬁned tangent line. Endpoints are also called boundary points.

) Singularities are extremely interesting. What you will see are the smooth curves shown in the yz-plane and the xz-plane. They are best viewed facing the origin.1. from a distance of 30 cm with one eye shut. can be got rid of by uncoiling the spiral and wrapping it around a cone. e3 e2 e1 (The three-dimensional models in these notes are drawn in central perspective.1. where r is allowed to be negative. but in this course we shall focus on gaining a thorough understanding of the smooth points. 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. MANIFOLDS 3 θ. . which is usually in the middle of the picture.

The boundary of the band consists of 1To be strictly accurate. INTRODUCTION Dimension 2. a paraboloid and a torus. A two-dimensional manifold is a smooth surface without selfintersections. It may have a boundary. but not a smooth manifold with boundary. such as an open square. but they are relatively boring. A closed square is not a manifold.1 Two-dimensional manifolds in three-dimensional space include a sphere. e3 e2 e1 The famous Möbius band is made by pasting together the two ends of a rectangular strip of paper giving one end a half twist. which is always a one-dimensional manifold. because the corners are not smooth. . In these notes we will consider only smooth manifolds.4 1. the closed square is a topological manifold with boundary. Examples are: an arbitrary open subset of R 2 . or a closed subset with a smooth boundary. You can have two-dimensional manifolds in the plane R2 .

1. Again the . This creates a Möbius type band with a ﬁgure eight cross-section. and then to join the resulting boundary circles. Out of the Möbius band we can create in two different ways a manifold without boundary by closing it up along the boundary edge. A different model of the Klein bottle is found by folding over the edge of a Möbius band until it touches the central circle. Equivalently. we obtain the Klein bottle or the projective plane.1. MANIFOLDS 5 two boundary edges of the rectangle tied together and is therefore a single closed curve. making sure the arrows match up. take a length of tube with a ﬁgure eight cross-section and weld the ends together giving one end a half twist. According to the direction in which we glue the edge to itself. 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. You will notice this cannot be done without passing one end through the wall of the tube. b b a a a b Klein bottle a a a Möbius band b projective plane Perhaps the easiest way to make a Klein bottle is ﬁrst to paste the top and bottom edges of the square together. The resulting surface intersects itself along a circle and therefore is not a manifold. which gives a tube.

The projective plane has the same peculiarity. Below is a wire mesh model of the resulting surface. The upper half corresponds to the complement of the disc and is known as a cross-cap. The self-intersection locus as well as a few of the cross-sections are shown in black in the following wire mesh model. Now fuse the two seams to create a single seam labelled a. namely the central circle of the original Möbius band. INTRODUCTION resulting surface has a self-intersection. To represent the Klein bottle without self-intersections you need to embed it in four-dimensional space. and it too has self-intersecting models in three-dimensional space. The wire shown in black corresponds to the edge a. which gives the following diagram.6 1. 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. The interior points of the black wire are ordinary self-intersection points. The lower half shaped like a bowl corresponds to the dashed circular disc in the middle of the square. Perhaps the easiest model is constructed by merging the edges a and b shown in the gluing diagram for the projective plane. Its two endpoints are qualitatively different singularities . It is obtained by welding together two pieces along the dashed wires.

.) Manifolds deﬁned by linear equations (i. . . . ^ . xn W$X .e. . . . . . . e3 e2 e1 1. Equations Very commonly manifolds are given “implicitly”. . xn W$X φ2 V x1 . . .2.^ . namely as the solution set of a system φ1 V x1 . EQUATIONS 7 known as pinch points. c2 . φm V x1 . Here φ 1 . xn W$X cm . . x2 _ y2 X 1. . . ^ xn φm V x W c. c2 . .2. φ V x W"X . More interesting manifolds arise from nonlinear equations. where φ is a matrix) are called afﬁne subspaces of Rn and are studied in linear algebra. c1 . φ2 . . . where the surface is crinkled up. 1. . . cn we can represent this system as a single equation φ V x WX It is in general difﬁcult to ﬁnd explicit solutions of such a system. . Consider the system of two equations in three unknowns. xn are variables. . . Here φ V x W$Xa` x2 _ y2 y_ z b and 1 c Xc` . y _ z X 0. φm are functions. c 1 . of m equations in n unknowns. . . E XAMPLE . . . . cX . x2 . (On the positive side.1. 0b . cm are constants and x 1 . . it is usually easy to decide whether any given point is a solution by plugging it into the equations. . .1. By introducing the useful shorthand YZ \N] YZ \N] YZ \N] Z x1 ] Z φ1 V x W ] Z c1 ] Z[ ] Z[ ] Z[ ] x2 φ2 V x W c2 xX . .

.2.6 you will be asked to ﬁnd a basis of Tx M for a particular x and you will see that Tx M is n n 1-dimensional. which turns it into a linear subspace.e. Hence the solution set is an ellipse.) Inequalities. For instance. Other examples of singularities can be found in Exercise 1. The standard notation for the tangent space to M at x is Tx M. views of how to deﬁne the tangent space to M at space at x consists of all vectors y such that u x. 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. The sphere of radius 1 is called the unit sphere and is denoted by Sn o 1 . Its boundary is the sphere of radius r. and this is the deﬁnition that we shall espouse. Here 2 e x ehfji x k x fUl x2 1 m x2 m k/k/k m x2 n is the norm or length of x and xk yf x1 y1 m x2 y2 m k/k/k m xn yn is the inner product or dot product of x and y. Thus Tx M fjp y r Rn s y k x f 0t . (In Exercise 1. It is a manifold of dimension 1. the closed ball of radius r about the origin in Rn is given by the single inequality e x exw r. However. Manifolds with boundary are often presented as solution sets of a system of equations together with one or more inequalities. This is an inhomogeneous linear equation in y. This leads to the second view of the tangent space at x. y k x f x k x f r 2 .5. In coordinates: y 1 x1 m k/k/k m yn xn f r2 . E XAMPLE . namely as the set of all y such that y k x f 0. for most practical purposes it is easier to translate this afﬁne subspace to the origin. the union of the two coordinate axes in the plane. Tangent spaces. The sphere of radius r is an n n 1dimensional manifold in Rn . 1.8 1. which has a singularity at the origin. the tangent space at x is an afﬁne subspace of Rn . There are two reasonable. The ﬁrst view is that the tangent y n x v:k x f 0. INTRODUCTION The solution set of this system is the intersection of a cylinder of radius 1 about the z-axis (given by the ﬁrst equation) and a plane cutting the x-axis at a 45 d angle (given by the second equation). given by the single equation y k x f r 2 . i. A simple example is xy f 0. The sphere of radius r about the origin in R n is the set of all x in Rn satisfying the single equation e x egf r. a linear subspace of Rn . Let us use the example of the sphere to introduce the notion of a tangent space. Let M fqp x r Rn s e x egf r t be the sphere of radius r about the origin in Rn and let x be a point in M. In this view. but inequivalent.

The conﬁguration space or state space of the system is an abstract space. Examples are solution curves of differential equations (see e. 2π }~| π 2. Examples of parametrizations with singularities are given in Exercises 1.3. which are complementary to those of the implicit method.4. Only the meridian θ y 0 is then hit twice. Also. (The conﬁguration ignores any motions that the system may be undergoing. The conﬁguration of a mechanical system (such as a pendulum. When the system moves. Exercise 1. We will use parametrizations to give a formal deﬁnition of the notion of a manifold in Chapter 6. it is usually impossible to parametrize a manifold in such a way that every point is covered exactly once. a spinning top.1. zy sin φ parametrizes the unit sphere in R3 . but are at times more useful. So a conﬁguration is like a snapshot or a movie still. Parametrizations A dual method for describing manifolds is the “explicit” way. xy cos θ . Conﬁguration spaces Frequently manifolds arise in more abstract ways that may be hard to capture in terms of equations or parametrizations. A disadvantage is that it can be hard to decide if any given point is on the manifold or not. π 2} to avoid counting points twice. Its dimension is called the number of degrees of freedom of the system.) The explicit method has various merits and demerits. One commonly restricts the polar coordinates z θ .1 and 1. Note however that not every parametrization describes a manifold. because this involves solving for the parameters. .4. yy sin θ cos φ. φ { to the rectangle | 0. its conﬁguration changes. for example when one wants to integrate over the manifold. The conﬁguration space of even a fairly small system can be quite complicated. 1. Parametrizations are often harder to come by than a system of equations. but this does not matter for many purposes. a ﬂuid.) is its state or position at any given time. CONFIGURATION SPACES 9 1.10) and conﬁguration spaces.g. the points of which are in one-to-one correspondence to all physically possible conﬁgurations of the system. or a gas etc. Such is the case for the two-sphere. One obvious advantage is that it is easy to ﬁnd points lying on a parametrized manifold simply by plugging in values for the parameters. the solar system.) In practice one usually describes a conﬁguration by specifying the coordinates of suitably chosen parts of the system. yy sin θ parametrizes the unit circle in R2 and xy cos θ cos φ. Very often the conﬁguration space turns out to be a manifold.2. such as computing the surface area or integrating a continuous function. namely by parametrizations. (Here φ is the angle between a vector and the xy-plane and θ is the polar angle in the xy-plane. For instance.

the point B can perform two . The conﬁguration space is a four-dimensional manifold. every pair of vectors can occur (if we suppose the second rod is allowed to swing completely freely and move “through” the ﬁrst rod) and describes a distinct conﬁguration. for those who have heard of relativity) can have a basis in physical reality. and so it has three degrees of freedom. The state of the pendulum is entirely determined by the position of the bob. Thus there are four degrees of freedom. Take a spherical pendulum of length r and attach a second one of length s to the moving end of the ﬁrst by a universal joint. free to swing in any direction in three-space. x being the vector pointing from the centre to the ﬁrst weight and y the vector pointing from the ﬁrst to the second weight. B.10 1. 1. The point A can move about freely and is determined by three coordinates. A spherical pendulum is a weight or bob attached to a ﬁxed centre by a rigid rod. The state of this system can be speciﬁed by a pair of vectors x. But the position of A alone does not determine the position of the whole solid. E XAMPLE . INTRODUCTION 1.3. 1. The bob can move from any point at a ﬁxed distance (equal to the length of the rod) from the centre to any other. Aside from this limitation. 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. The resulting system is a double spherical pendulum. What is the number of degrees of freedom of a rigid body moving in R3 ? Select any triple of points A.4. If A is kept ﬁxed.5. The following two examples show that this is not true. y . E XAMPLE . C in the solid that do not lie on one line. Some believe that only spaces of dimension 3 (or 4. The conﬁguration space is therefore a two-dimensional sphere. E XAMPLE . known as the (Cartesian) product of two two-dimensional spheres.

E XAMPLE (the space of quadrilaterals). which gives two more degrees of freedom. This means that the conﬁgurations are in one-to-one correspondence with the points on the circle of radius b about the point B. CONFIGURATION SPACES 11 independent swivelling motions. B B C A A A The positions of A. c.e. Consider all quadrilaterals ABCD in the plane with ﬁxed sidelengths a. 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. (As C moves all the way around.6. so the total number of degrees of freedom is 3 2 1 6. the conﬁguration space is a circle. the point C can rotate about the axis AB. In other words. .4. i. which gives one further degree of freedom. If A and B are both kept ﬁxed.) What are all the possibilities? For simplicity let us disregard translations by keeping the ﬁrst edge AB ﬁxed in one place. or vice versa. 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. The position of the “pump” D is wholly determined by that of the “wheel” C. b. A few possible positions are shown here. d. it moves on a sphere centred at A. 1. D c C d A a B b (Think of four rigid rods attached by hinges. B and C determine the position of the solid uniquely.1. Thus the conﬁguration space of a rigid body is a six-dimensional manifold.

continued). E XAMPLE (quadrilaterals. this is not completely accurate: for every choice of C. D c C d A a B b In this case. Thus the conﬁguration space is a disjoint union of two circles. when BC points straight to the left. INTRODUCTION Actually. there are two choices D and D for the fourth point! They are interchanged by reﬂection in the diagonal AC. It is not possible to move continuously from the ﬁrst set of conﬁgurations to the second. 1. This is an example of a disconnected manifold consisting of two connected components. in fact they are each other’s mirror images. the quadrilateral collapses to a line segment: . 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 A B D So there is in fact another circle’s worth of possible conﬁgurations.12 1.7.

recall that the tangent vector at a point x. y 1 cos t (t R) parametrize a plane curve. but most conﬁguration spaces occurring in nature are (and an engineer designing an engine wouldn’t want to use this quadrilateral to make a piston drive a ﬂywheel).3. back up: or further down: This means that when b c are merged at a point. Exercises 1. 1.g. Also include a few tangent lines at judiciously chosen points. identify all points where the curve is not a manifold. More singularities appear in the case of a parallelogram (a c and b d) and in the equilateral case (a b c d). . y 3 at2 1 t3 . 1. Parametrize the space curve wrapped around the cone shown in Section 1.2. 1. In your plot. You may use software and turn in computer output. (E. and . Graph this curve as carefully as you can. y of the curve has components dx dt and dy dt. a d the two components of the conﬁguration space The juncture represents the collapsed quadrilateral. Same questions as in Exercise 1. The formulas x t sin t.1. there are two possible directions for D to go. ﬁnd all tangent lines with slope 0.1 for the curve x 3 at 1 t 3 .EXERCISES 13 and when C moves further down. This conﬁguration space is not a manifold.) To compute tangent lines.1.

As a preliminary step. so use a self-intersection where necessary. determine the intersection of each surface with a general plane parallel to one of the coordinate planes. 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. 1.14 1. . e? (As in the case of quadrilaterals.6. for certain choices of sidelengths singularities may occur.7 assume that the edge AB is kept ﬁxed in place so as to rule out translations. What is the conﬁguration space of all parallelograms ABCD such that AB and CD have length a and BC and AD have length b? What happens if a b? (As in Examples 1. View this equation as a homogeneous linear equation in the entries y1 . Sketch the surfaces deﬁned by the following gluing diagrams. 1.) 1. . a b a a1 b1 b2 a2 b1 b c d b a b d a1 a b2 a2 a2 b2 a1 b1 a2 b2 a1 b1 (Proceed in stages. Determine all the points where the surface does not have a well-deﬁned tangent plane..6 and 1. 2. . y2 . so you may be better off drawing them by hand. c. .) (i) (ii) (iii) (iv) n n n n 0. Let M be the sphere of radius n about the origin in Rn and let x be the point 1. 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. Choose two distinct positive real numbers a and b. 1 on M. For the values of n indicated below graph the surface in R 3 deﬁned by x n y2 z.5. b.) 1.4. d. . 3. 1. etc. What is the conﬁguration space of all pentagons ABCDE in the plane with ﬁxed sidelengths a. (Computer output is OK. To reduce the number of degrees of .) 1. .7.) 1.9. and try to identify what you get at every step. You may ignore these cases.8. . ﬁrst gluing the a’s. then the b’s. One of these surfaces cannot be embedded in R 3 . but bear in mind that few drawing programs do an adequate job of plotting these surfaces. yn of y and ﬁnd the general solution by means of linear algebra. . INTRODUCTION 1.

y ¡ t ¢-¢ of this system that are contained in the positive quadrant (x £ 0. while p. can be found at ª«¦«'¥¬F®¦®N¯¦¯¦¯±°²¨§D«'ª³°µ´¢¶¦·'©§°F©D¸'¹ ®¦ºN¸¦»¢¶'©¦¦'¸¢®'¸¦»'¥D¥°¼ª«½²¦ . by writ1 ing dy f ¡ x ¢ dx and integrating both sides.EXERCISES 15 freedom you may also assume the edge AB to be ﬁxed in place.) (i) Show that the solutions of the system satisfy a single differential equation of the form dy ¤ dx f ¡ x ¢ g ¡ y ¢ . dt where x ¡ t ¢ represents the number of prey and y ¡ t ¢ the number of predators at time t. (The Lotka-Volterra system is exceptional in this regard. Indicate the direction in which the solutions move. Usually it is impossible to write down an equation for the solution curves of a differential equation. where f ¡ x ¢ is a function that depends only on x and g ¡ y ¢ a function that depends only on y.e. (Don’t forget the integration g ¡ y¢ constant. (ii) Solve the differential equation of part (i) by separating the variables. i. q. . ¥¦¥ ¦'§N¨¢© .) E d D c e C b A a B 1. s are positive constants. Mathematica or MATLAB. dt dy py qxy. The Lotka-Volterra system is an early (ca. A useful Java applet. Be warned that solving the system may give better results than solving the implicit equation! You may use computer software such as Maple.10. r. It is the pair of differential equations dx rx sxy.) (iii) Set p q r s 1 and plot a number of solution curves. y £ 0) and derive an implicit equation satisﬁed by these solution curves. In this problem we will consider the solution curves (also called trajectories) ¡ x ¡ t ¢ . 1925) predator-prey model.

.

In such a case we say α is a k-form on U . and the curvature of a surface. 3 À . i2 . whereas β consists of one term. the work exerted by a force. 3. represent a 2-form on R5 . ¿ 2. it is entirely possible to acquire a solid working knowledge of differential forms without entering into this formalism. grad and curl of vector calculus. . β ¾ x1 x3 x5 dx1 dx6 dx3 dx2 . corresponding to the multi-index ¿ 1. but on the open set U ¾ R3 ÄÅ ¿ x. i. I (If you don’t know the symbol α . That is the objective of this chapter. that is a “vector” consisting of k integer entries ranging between 1 and n. resp. Fortunately. a 4-form on R6 . The study of differential forms. 3 À .CHAPTER 2 Differential forms on Euclidean space The notion of a differential form encompasses such ideas as elements of surface area and volume elements.) For instance the expressions α¾ sin ¿ x1 Ã e x4 À dx1 dx5 Ã x2 x2 5 dx 2 dx 3 Ã 6 dx 2 dx 4 Ã cos x 2 dx 5 dx 3 . 17 . To avoid such ambiguities it is good practice to state explicitly the domain of deﬁnition when writing a differential form.1. . called the tensor product. the complement of the z-axis. 2 À . 5 À . (The notation dxi1 Â dxi2 Â Á/Á/Á Â dxik is also often used to distinguish this kind of product from another kind. corresponding to the multi-indices ¿ 1. Thus the expression ln ¿ x 2 Ã y2 À z dz is not a 1-form on R3 . Note. space or hyperspace. ik À of degree k. y.e. . which generalizes the operators div. Elementary properties A differential form of degree k or a k-form on Rn is an expression α¾ ∑ f I dx I . An important operation on differential forms is exterior differentiation. 2. Another reason for being precise about the domain of a form is that the coefﬁcients f I may not be deﬁned on all of Rn . The form α consists of four terms. that α could equally well be regarded as a 2-form on R 6 that does not involve the variable x 6 . 4 À and ¿ 5. look up and memorize the Greek alphabet in the back of the notes. which is examined in Chapter 7. A mathematically rigorous study of differential forms requires the machinery of multilinear algebra. z ÀÇÆ x2 Ã y2 ¾ È 0 É . . and dx I is an abbreviation for dxi1 dxi2 / Á Á/Á dxik . is often termed the exterior differential calculus. which was initiated by E. Cartan in the years around 1900. however.) Here I stands for a multi-index ¿ i 1 . The f I are smooth functions on Rn called the coefﬁcients of α . the ﬂow of a ﬂuid. 6. but only on an open subset U of Rn . ¿ 2.

So we can write any 2-form as 1Ñ iÒ jÑ n ∑ gi.2 dx1 dx2 Î Ê/Ê/Ê Î f 1. Ë Ì dxi1 dxi2 Ê/Ê/Ê dxi p / (2.3 dx2 dx3 Î Ê/Ê/Ê Î gn Ó 1.1) the terms f i.1 dxn dx1 Î f n. so dxi dxi Ë 0 for all i.n dx2 dxn Î Ê/Ê/Ê Î f n.1 dx2 dx1 Î f 1. In particular. A 0-form on Rn is simply a smooth function (no dx’s). the sign changes: dxi1 dxi2 Ê/Ê/Ê dxiq / Ê Ê/Ê dxi p / Ê Ê/Ê dxik Í Ê Ê/Ê dxiq / Ê Ê/Ê dxik . graded commutativity. this rule implies dx i dxi ËÍÌ dxi dxi . This is called anticommutativity.2 Ì f 2.2 dx1 dx2 and f 2. j dxi dx j Ë i. Because of the alternating property (2.n dxn dxn . .n dx2 dxn Î / Written like this. DIFFERENTIAL FORMS ON EUCLIDEAN SPACE You can think of dxi as an inﬁnitesimal increment in the variable x i and of dx I as the volume of an inﬁnitesimal k-dimensional rectangular block with sides dx i1 . .1 dx1 dx1 Î f 2.) By volume we here mean oriented volume. A general 2-form has the shape ∑ f i.n dx1 dxn Ê Ê/Ê Î g2.n dx1 dxn Î f 2. a general n Ì 1-form can be written as a sum of n components. Ê/Ê/Ê Î 2 Î 1 Ë 1 n n 1Ð 2 Ï Ì components. A general 1-form looks like f 1 dx1 Î f 2 dx2 Î Ê/Ê/Ê Î f n dxn . Likewise. if we interchange two variables.2 dx1 dx2 Î f 2. f 1 dx2 dx3 Ê/Ê/Ê dxn Î f 2 dx1 dx3 Ê/Ê/Ê dxn Î Ê/Ê/Ê Î f n dx1 dx2 Ê/Ê/Ê dxn Ó 1 Ë iÔ 1 ∑ n f i dx1 dx2 Ê/Ê/Ê. or the alternating property. n dx n Ó 1 dx n . dx ik . which takes into account the order of the variables.2. Let us consider k-forms for some special values of k. a 2-form has at most nÎ nÌ 1Î nÌ 2Î Î g2.1 dx2 dx1 ËÍÏ f 1.2 dx1 dx2 Î Ê/Ê/Ê Î g1.1) and so forth.1 Ð dx1 dx2 .1 dx2 dx1 can be grouped together: f 1.18 2.i dxi dxi vanish. . and a pair of terms such as f 1.2 dx2 dx2 Î Ê/Ê/Ê Î f 2. . (A precise deﬁnition will follow in Section 7. dxi2 . Thus. j dxi dx j Ë g1.2 dxn dx2 Î Ê/Ê/Ê Î f n. j f 1.

Forms of degree k Ö n on Rn are always 0. because at least one variable has to repeat in any expression dx i1 Ê/Ê/Ê dxik . In general a form of degree k can be expressed as a sum αË ∑ f I dx I . Every n-form on Rn can be written as f dx 1 dx2 Ê/Ê/Ê dxn . where dx Õ i means “omit the factor dx i ”. I . By convention forms of negative degree are 0. The special n-form dx1 dx2 Ê/Ê/Ê dxn is also known as the volume form.dx Õ i Ê/Ê/Ê dxn .

A form is constant if the coefﬁcients f I are constant functions. The maximum number of terms occurring in α is then the number of increasing multi-indices of degree k. j2 . (The space â k ß U á itself is inﬁnite-dimensional. Since k-forms can be added together and multiplied by scalars. . ß y2 à x2 á dx dy dz. The set of constant k-forms is a linear subspace of â k ß U á of dimension ã n k ä . I.) Two k-forms α ∑ I f I dx I and β ∑ I g I dx I (with I ranging over the increasing multiÝ Ý indices of degree k) are considered equal if and only if f I g I for all I .1. . . . J Usually many terms in a product cancel out or can be combined. where I ranges over all increasing multi-indices of degree k. Therefore αnç 1 0 Ý for any form α on Rn of positive degree. n. βα for all k-forms α and all l-forms β. The alternating property combines with the multiplication rule to give the following result.) The (exterior) product of a k-form α ∑ I f I dx I and an l -form β ∑ J g J dx J is Ý Ý deﬁned to be the k å l -form αβ Ý ∑ f I g J dx I dx J . Successively applying ß i1 . We shall almost always represent forms in this manner. i2 . jl á . ELEMENTARY PROPERTIES 19 where the I are increasing multi-indices. The total number of increasing multi-indices of degree k is therefore equal to the binomial coefﬁcient “n choose k”. . . 2. 1 × i 1 Ø i2 ØQÙ/Ù/ÙÚØ ik × n.2. .1. . Let I Ý Ý the alternating property we get dx I dx J Ý Ý Ý . . which is n k . For instance. . the collection â k ß U á constitutes a vector space. An increasing multi-index of degree k amounts to a choice of k numbers from among the numbers 1. dxi1 dxi2 Ù/Ù/Ù dxik dx j1 dx j2 dx j3 Ù/Ù/Ù dx jl ßèà 1 á k dx j1 dxi1 dxi2 Ù/Ù/Ù dxik dx j2 dx j3 Ù/Ù/Ù dx jl ßèà 1 á 2k dx j1 dx j2 dxi1 dxi2 Ù/Ù/Ù dxik dx j3 Ù/Ù/Ù dx jl Ý ßèà 1 á kl dx J dx I . Û n! n . k ÜÞÝ k! ß n à k á ! (Compare this to the number of all multi-indices of degree k. A basis of this subspace is given by the forms dx I . P ROPOSITION (graded commutativity). . . Its p-th power α p is of degree kp. ik á and J P ROOF. Ý ßèà 1 á klαβ ß j1 . Ý As an extreme example of such a cancellation. The Ý collection of all k-forms on an open set U is denoted by â k ß U á . which is greater than n if k æ 0 and p æ n. . ß y dx å x dy á¦ß x dx dz å y dy dz á Ý y 2 dx dy dz å x 2 dy dx dz 2. . . consider an arbitrary form α of degree k.

The exterior derivative If f is a 0-form. then f y dy dx ñ g x dx dy éÍê g x ì f y í dx dy. 2. I The operation d is called exterior differentiation. J ∑ f I g J dx I dx J éëêèì I. Then we get α 2 î é êèì 1 í k α 2 éïê½ì 1 í kα 2 .2.2. An operator of this sort is called a ﬁrst-order partial differential operator.2) in the ﬁrst sum we used the alternating property and in the second sum we interchanged the roles of i and j. DIFFERENTIAL FORMS ON EUCLIDEAN SPACE For general forms α é ∑ I f I dx I and β é 1í ∑ J g J dx J we get from this kl βα é ∑ g J f I dx J dx I éëê½ì I. j ð 1 n ∂ fi dx dx ∂ xj j i 1ó jô ió n ∂fj ∂f éqì ∑ ∂xij dxi dx j ñ ∑ ∂xi dxi dx j 1ó iô jó n 1ó iô jó n ∂fj ∂f é ∑ ∂xi ì ∂xij ö dxi dx j . α 2 é 0 if α is a form of odd degree. because it involves the ﬁrst partial derivatives of the coefﬁcients of a form.) More generally. If α é dα é f dx ñ g dy is a 1-form on R2 . (Recall that f y is an alternative notation for ∂ f ò ∂ y. but tells us that α é 0 if k is odd. ∂f Then we have the product or Leibniz rule: d ê f g í$é If α é ∑ I f I dx I is a k-form. f dg ñ g df .2) where in line (2. 2.20 2.3. A noteworthy special case is α é β. E XAMPLE . QED 2 which establishes the result. C OROLLARY. each of the coefﬁcients f I is a smooth function and we deﬁne dα to be the k ñ 1-form dα é ∑ d f I dx I . . for a 1-form n α é ∑n i ð 1 f i dx i on R we have dα é ið 1 ∑ d f i dxi é ∑ ∑ n ∂ fi dx dx ∂ xj j i i. that is a smooth function. J 1 í klαβ. 2. 1ó iô jó n õ é ∂ fi dx dx ñ ∂ xj j i 1ó iô jó n ∑ (2. we deﬁne d f to be the 1-form df é ið 1 n ∑ ∂xi dxi . 2 This equality is vacuous if k is even.

3) (2.k ∂ f i. g and constants a. b. P ROPOSITION .3) we rearranged the subscripts (for instance. j dxk dxi dx j ∂ xk ∑ 1ü iý jü n kþ 1 ∑ ∑ n ÷ ∂ f i. j dxk dxi dx j ø ∂ xk 1ü ký iý jü n ∑ ∂ f i. J f I g J û dx I dx J ÷ ∑ù I. P ROOF.5. ú ∑ ∂ xk ∂x j ∂ xi 1ü iý jý kü n ÿ ∑ ∑ (2.4. j ø ∑ ∂xk dxk dxi dx j 1ü iý jý kü n ∂ f j. QED Here is one of the most curious properties of the exterior derivative. THE EXTERIOR DERIVATIVE 21 2. then dα is of degree n ø 1 and so dα ÷ 0. E XAMPLE .1 give d ù αβ û ÷ ∑ dù I.k ø dxi dx j dxk . (ii) d ù αβ û ÷Uù dα û β ø ù ú 1 û kα dβ for all k-forms α and l-forms β. J I. Now let α ÷ ∑ I f I dx I and β ÷ ∑ J g J dx J . The operator d is linear and satisﬁes a generalized Leibniz rule. 2.4) Here in line (2. j dxk dxi dx j ∂ xk 1ü iý ký jü n ∂ f i. The Leibniz rule for functions and Proposition 2.2. An obvious but quite useful remark is that if α is an n-form on Rn . (i) d ù aα ø bβ û ÷ a dα ø b dβ for all k-forms α and β and all scalars a and b. J f I dg J ø g J d f I û dx I dx J ÷ ∑ ¤ d f I dx I ù g J dx J û ø ù ú 1 û k f I dx I ù dg J dx J û¦¥ ÷ëù dα û β ø which proves part (ii). i ú£¡ j and j ú£¡ k) and in line (2. j ∂ f i. If α ÷ dα ÷ f z dz dx dy ø For a general 2-form α ÷ dα ÷ 1ü iý jü n f dx dy ø g dx dz ø h dy dz is a 2-form on R 3 . then g y dy dx dz ø h x dx dy dz ÷ëù f z ú g y ø h x û dx dy dz.k dx j dxi dxk ∂x j 1ü iý jý kü n ∂ f i. . in the ﬁrst term we relabelled k ú¢¡ i. j dx i dx j on Rn we have ∂ f i. ù ú 1 û kα dβ. ∑1 ü i ý j ü d f i. j dxi ÷ n f i .4) we applied the alternating property. j ø ∑ ∂xk dxk dxi dx j 1ü iý jý kü n ∑ ÷ ∂ f j.k dxi dx j dxk ø ∂ xi 1ü iý jý kü n ÷ ∂ f i.2. The linearity property (i) follows from the linearity of partial differentiation: ∂f ∂g ∂ ù a f ø bg û ÷ a ø b ∂ xi ∂ xi ∂ xi for al smooth functions f .

∂ xi I i.22 2. Then d © ∑ ∑ d I dxi dx I . P ROPOSITION . Let α © ∑ I f I dx I . d2 © 0. DIFFERENTIAL FORMS ON EUCLIDEAN SPACE 2. n P ROOF. In short.6. d § dα ¨ © 0 for any form α .

I i.

3 (replacing f i with ∂ f I ∂ xi ) we ﬁnd d § dα ¨ © i 1 ∑∑ © n ∂ fI dx dx I ∂ xi i 1 ∂f ∑ d . 1 Applying the formula of Example 2.

Let us consider the simplest case of a 1-form α © ∑n i . which we shall explore in Chapters 4. A nonzero constant function is not exact.3. Every exact form is closed. 5 and 10.7. the answer depends strongly on the topology. because forms of degree 1 are 0. Closed and exact forms A form α is closed if dα one less). P ROPOSITION . E XAMPLE .8. n ∂ fI dxi ∂ xi 1 i j n ∑ ∂2 f I ∂ xi ∂ x j ∂2 f I dxi dx j ∂ x j ∂ xi © 0. Is every closed form of positive degree exact? This question has interesting ramiﬁcations. y dx x dy is not closed and therefore cannot be exact. 2. P ROOF. On the other hand y dx x dy is closed. Hence d § dα ¨© 0. that is the qualitative “shape”. because for any smooth (indeed. of the domain of deﬁnition of the form. which means it is constant.6. If α © 0. Amazingly. It is also exact. QED 2. because d § xy ¨© y dx x dy. C 2 ) function f the mixed partials ∂ 2 f ∂ xi ∂ x j and ∂2 f ∂ x j ∂ xi are equal. For a 0-form (function) f on Rn to be closed all its partial derivatives must vanish. QED © dβ then dα © d § dβ ¨ © 0 by Proposition 2. It is exact if α © dβ for some form β (of degree 2.

7 this is not possible unless α is closed. E XAMPLE . . © f2 .9. Then © dy dx z § y sin yz ¨¢ cos yz dz dy dx dy y § z sin yz ¨ cos yz dy dz © 0. Let α dα © y dx § z cos yz x ¨ dy y cos yz dz. Finding a solution is sometimes called integrating the system. Determining whether α is exact means solving the equation dg © α for the function g. (2. © fn. These identities must be satisﬁed for the system (2. 1 f i dx i . This amounts to ∂g ∂g ∂g © f1 ... By the formula in Example 2..5) ∂ x1 ∂ x2 ∂ xn a system of ﬁrst-order partial differential equations.3 α is closed if and only if ∂ fi ∂x j © ∂fj ∂ xi for all 1 i j n.5) to be solvable and are therefore called the integrability conditions for the system. By Proposition 2. 2. .

From x ∂ ∂ x x2 y2 x2 $ y2 x2 y2 ! 2 it follows that the angle form is closed. ∂g ∂z y cos yz by successive integration. where c is a function of y and z only. k0 n $ k0 This means that in a certain sense there are as many k-forms as n $ k-forms. In fact. is the number of ways of partitioning a pile of n objects into a pile of k objects and a pile of n $ k objects. . ∑ I with 1 dx I ε I dx I c . Hodge star of α is denoted by 1 α (or sometimes α 2 ) and is deﬁned as follows. Equivalently.6) ∂g j ∂ xi ∂ f i. j . The ﬁrst equation gives g yx c y. If α ∑ I f I dx I .6.n k . Substituting into the second equation gives ∂c " ∂ y z cos yz. so k is a constant. so c sin yz k z ! .n k . E XAMPLE . in Example 10. 2 2 2 x y x y x2 y2 y2 $ x2 .k ∂x j ∂ xi 0 comes down to 0 for all 1 + i . The Hodge star operator The binomial coefﬁcient . So g yx sin yz is a solution and therefore α is exact. j dxi dx j and a 1-form β tion dβ α amounts to the system ∑n i* 1 gi dxi the equa(2.4. Is α exact? Let us solve the equations ∂g ∂x y.4.10. x2 y2 ! 2 ∂ y ∂ y x2 y2 is called the angle form for reasons that will become clear in Section 4. z ! . The 1-form on R2 $&% 0 ' deﬁned by $ y dx x dy y x α $ 2 dx dy .2. where we shall see that this form is not exact. For a 2-form α ∑1 ( i ) j ( n f i. This method works always for a 1-form deﬁned on all of Rn . / / Thus we see that n n . THE HODGE STAR OPERATOR 23 so α is closed.6). there is a natural way to turn k-forms into n $ k-forms. j . then 1α f I 31 dx I ! . is the number of ways of selecting k (unordered) objects from a collection of n objects. By the formula in Example 2. (See Exercise 2. k + n. This is the Hodge star operator. 2. ∂g ∂y z cos yz x. Substituting into the third equation gives k # 0.18.) Hence every closed 1-form on Rn is exact.6.3. This example is continued in Examples 4. We shall learn how to solve the system (2.k ∂ f j. j ∂ xk $ ∂ gi ∂x j f i. .1 and 4. 2.4 the integrability condition dα $ ∂ f i. and its higher-degree analogues.

6 = . : dx4 4 9 dx1 dx2 dx3 . for any increasing multi-index I . which shows that ε I 4 9 1.5. In higher dimensions it is better to stick to the rule. : < dx dz =>4 9 dy. . grad and curl A vector ﬁeld on an open subset U of Rn is a smooth map F : U write F in components as OQP F1 < x = PP F2 < x = F < x =>4@LM . : < dx1 dx2 =>4 dx3 dx4 . 78787 dxn . .24 2. : dx2 4 9 dx1 dx3 dx4 . M . so dx I 4 2. n On R we have : 1 4 dx1 dx2 78787 dxn . The factor ε I is a sign. 4. in contravention of our usual rule to write the variables in increasing order. R K Rn . : < dx dy = 4 dz. E XAMPLE . : < dx1 dx2 78787 dxn =A4 1. Then I c dx2 dx6 and dx I c 4 dx1 dx3 dx4 dx5 . : dy 4 9 dx dz 4 dz dx. 9 : < dx1 dx3 =>4 : < dx2 dx4 =A4 9 dx1 dx3 . 4@< 1. : dxi 4C< 9 1 = i D 1dx1 dx2 78787Fdx E i 78787 dxn i j 1 D D dx1 dx2 78787Hdx for 1 G i J j G n. dx2 dx4 . E XAMPLE . 3. : < dx2 dx3 =A4 dx1 dx4 .11. which consists of all numbers between 1 and n that do not occur in I . I c denotes the complementary increasing multi-index. and : dx3 4 dx1 dx2 dx4 . : dz 4 dx dy. 465 9 1 if dx dx c 9 dx dx I I 4 1 2 78787 dx n . 1 which is chosen in such a way that dx I < : dx I = is the volume form: dx I < : dx I =>4 dx1 dx2 1 if dx I dx I c 4 dx1 dx2 78787 dxn . 5 = . Hence : < dx 2 dx6 =A4 9 dx1 dx3 dx4 dx5 . : < dy dz = 4 dx. Let n 4 6 and I 4?< 2. : < dx3 dx4 =A4 dx1 dx2 . : < dxi dx j =A4C< 9 1 = E i 78787Idx E j 78787 dxn 2. We can MN Fn < x = . div. εI In other words. (This is the reason that 2-forms on R3 are sometimes written as f dx dy B g dz dx B h dy dz. : < dx1 dx4 =>4 dx2 dx3 . On R3 we have : dx 4 dy dz.) On R4 we have : dx1 4 dx2 dx3 dx4 . and for 1 G i G n. times a factor . 2. DIFFERENTIAL FORMS ON EUCLIDEAN SPACE Here. Therefore dx I dx I c 4 dx2 dx6 dx1 dx3 dx4 dx5 4 dx1 dx2 dx6 dx3 dx4 dx5 4 9 dx1 dx2 dx3 dx4 dx5 dx6 .12. On R2 we have : dx 4 dy and : dy 4 9 dx. : dx I is the product of all the dx j ’s that do not occur in dx I .

We shall see how to do this in Section 4. Thus vector ﬁelds and 1-forms are symbiotically associated to one another. DIV. en are the standard basis vectors of Rn . y y x x We can turn F into a 1-form α by using the Fi as coefﬁcients: α S ∑n i T 1 Fi dx i . Let us introduce the symbolic notation dx1 dx2 . e2 . Intuitively. The black dots are the zeroes of the vector ﬁelds (i.2. GRAD AND CURL 25 or alternatively as F S ∑n i T 1 Fi ei . where e1 . For . . Then we can write α S F ` dx. dx SYZ[[ [\ dxn which we will think of as a vector-valued 1-form. .1. such as gravity or an electric force acting on a particle. (If the particle travels along a path. The diagrams below represent the vector ﬁelds W ye1 X xe2 and U¦W x X xy V e1 X U y W xy V e2 (which you may recognize from Exercise 1. points x where F U x V>S 0).5.) The correspondence between vector ﬁelds and 1-forms behaves in an interesting way with respect to exterior differentiation and the Hodge star operator. . _ ]Q^^ ^ . F is determined by α and vice versa. vector ﬁeld F acb 1-form α : α S F ` dx. Vector ﬁelds in the plane can be plotted by placing the vector F U x V with its tail at the point x. Clearly.10). The arrows have been shortened so as not to clutter the pictures. . If F represents a force ﬁeld. the vector-valued 1-form dx represents an inﬁnitesimal displacement. . the total work done by the force is found by integrating α along the path. .e. then α S F ` dx represents the work done by the force when the particle is displaced by an amount dx. For instance. the 1-form α SYW y dx X x dy corresponds to the vector ﬁeld F S W ye1 X xe2 .

we ﬁnd Starting with a vector ﬁeld F and letting α s α d n F s dx d n F 1 i v 1 dx dx r8r8rxdx w i r8r8r dxn .26 2. the ﬂow of a ﬂuid or gas in Rn is represented by a vector ﬁeld F.3. that is the amount of material passing through the hypersurface s dx per unit time. the vector-valued n t 1-form s dx represents an inﬁnitesimal n t 1-dimensional hypersurface perpendicular to dx. An alternative way of writing this identity is obtained by applying s to both sides. DIFFERENTIAL FORMS ON EUCLIDEAN SPACE each function f the 1-form d f d ∑n i e 1 f ∂ f g ∂ x i h dx i is associated to the vector ﬁeld mQn ∂f n ∂ x1 n ∂ f nn n ∂f ∂ x2 grad f d ∑ . kik . Using the vector-valued n (This point of view will be justiﬁed in Section 8. . e djik . we can view grad f as the transpose of the Jacobi matrix of f . o . (Equivalently. ∂F . div F n ∂F n ∂F d s d s α. The n t 1-form s α then represents the ﬂux. Intuitively. j ∑ e ∂ Fi dx dx ∂ xj j i 1 d 1 i j n } ~ } ∑ ∂ Fj ∂ xi i t ∂x j dxi dx j .) We have ds α n ∂F d d f F r s dx h d ∑ i f¦t 1 h i v 1 dxi dx1 dx2 r8r8rxdx w i r8r8r dxn ∂ xi ie 1 n which gives d ∑ i dx1 dx2 r8r8r dxi r8r8r dxn d{z ∑ i dx1 dx2 r8r8r dxn . (The total amount of ﬂuid passing through a hypersurface S is found by integrating α over S. o kkl ∂ f ∂x n This vector ﬁeld is called the gradient of f .) grad f pcq df : df d grad f r dx. .) In ﬂuid mechanics. kk . 1 2 ∑ i f ih ∑ iu ft h ie 1 ie 1 t 1-form mQnn s dx mQnn dx2 dx3 r8r8r dxn 1 n n s dx dx dx dx 8 r 8 r r n 1 3 2 t s dx d d ik .14. ∂ xi ∂ xi | ie 1 ie 1 The function div F d ∑n i e 1 ∂ Fi g ∂ x i is the divergence of F. . We shall see how to do this in Section 5. Thus if α d F r dx. A very different identity is found by ﬁrst applying d and then dα s to α: d i. ∂ xi i ie 1 . d F r dx. o k v n 1 lk s dxn lk dx1 dx2 r8r8r dxn y 1 fut 1 h we can also write s α d F r s dx. after the proof of Theorem 8. then s d α d d f F r s dx h d div F dx 1 dx2 r8r8r dxn .1.

2. . t. ze sin ze y 3 dy 4 x2 y3 z3 e y sin ze y H e z dz. then curl F dx dα . if α F dx. 1. . i 1 Compute ω n ωω uu ω (n-fold product). β z dx dy x dy dz and γ z dy on R 3 .1. . Let ω dx1 dy1 dx2 dy2 dxn dyn uu n α Compute α dα n dz x1 dy1 x2 dy2 u¦ xn dyn dz α dα dα ¦u dα . xn dt ¡ x3 0 Show that dg α . Deﬁne a function g by x2 f 1 t. 2 dx (ii) ∑n i uQ dxn . . . 1-form on Rn .3. x3 . . Write the coordinates on R2n as x1 . Let α g x x1 0 z sin xz 3 dx 2 xy z dx 3 x2 y2 z4 f i dxi be a closed C 0 xy xey z2 dy x sin xz F 2 yz 3 z 2 dz. Write the coordinates on R2n 1 as x1 .5. . 2. Thus. (i) α (ii) α 1 3 R is closed and ﬁnd a function ye xy 3 4 ∑n i 1 2. x4 . . yn . z . x2 . x2 . for n curl F ∂ F2 ∂ F3 ∂ F1 e1 e2 ∂ x3 ∂ x ∂ 1 x3 3. ∂ F2 ∂ F1 e3 .) f 3 0. 3. 0. x dx y dy. . Let ∑ dxi dyi . t. namely ∂ x2 the curl of F. . . . . xn dt ¡ f 2 0. x4 . First work out the cases n 1. . . . For what values of p is this form closed? 2. y2 . . First work out the cases n i 1 ∑ xi dyi . 3. i 1 x i dx 1 uu n p i (iii) x ∑i 1 1 1 xi dx1 uu dx i uQ dxn . 2. . . Exercises 2. The point is rather to remember that exterior differentiation in combination with the Hodge star uniﬁes and extends to arbitrary dimensions the classical differential operators of vector calculus. where p is a real constant. Compute the exterior derivative of the following forms. x3 . You need not memorize every detail of this discussion. yn . y1 .3).6. 2. . (ii) dα . . x5 .4. t dt. Check that each of the following forms α g such that dg α . dβ. x2 . differentiate under the integral sign and don’t forget to use dα 0. xn . (Apply the fundamental theorem of calculus. n 2. Recall that a hat indicates that a term has to be omitted. .2. . . xn . dγ . y1 . 0. Consider the forms α Calculate (i) αβ.EXERCISES 27 and hence dα ∑ 1 i j 1 1 i j nu ∂ F3 ∂ Fj ∂ xi ∂ Fi dx1 dx2 ∂x j 88Hdx i 88dx j 88 dxn . . αβγ . ∂x 1 ∂ x2 In three dimensions dα is a 1-form and so is associated to a vector ﬁeld. 0. formula (B. y2 . (i) e xyz dx. xn dt Q¦ ¢ xn 0 f n 0.

(i) (ii) (iii) 2. Deduce from part (i) that dx 1 dx2 ¸¦¸u¸ dxn £±ª dr « ν on Rn ¥¦ 0 ¨ .e. (Use dα £ 0 and apply the identity proved in Exercise B. α «´³ 0 for all α and ª α . (ii) Repeat the calculation. i.12. β « dx 1 dx2 ¸u¸Q¸ dxn . (i) Let U be an open subset of Rn and let f : U ¾ R be a function satisfying grad f ª x «À£ © 0 for all x in U .9. df . ª α .14. Let α £ ∑ n i ¤ 1 f i dx i be a closed 1-form whose coefﬁcients f i are smooth functions deﬁned on Rn ¥§¦ 0 ¨ that are all homogeneous of the same degree p £ © ¥ 1.8. that the multi-indices I are increasing. £Á n ¸ ® dx . ® β. Calculate ® α . Consider the form α £ ¥ x 2 2 dx 1 x 1 dx 2 on R . (iii) Again repeat the calculation. 2 2 2. ® γ . Calculate ® d ® dα .16.e. an n ¥ 1-form ν and a 1-form α by n ª x «£Á grad f ª x «QÁÃÂ 1 ν α £ Prove that dx 1 dx2 ¸u¸u¸ dxn £ αν on U . grad f ª x «QÁ Â 1 . 2. where α. (ii) Let r : Rn ¾ R be the function r ª x «Ä£ÅÁ x Á (distance to the origin).) 2. I Prove the following assertions. (ii) ª α .) 2. The Laplacian of a smooth function on an open subset of R n is deﬁned by º f £ ∂2 f ∂ x2 1 ∂2 f ∂ x2 2 ¸¦¸Q¸ ∂2 f .2.7.28 2. as usual. Let α £ ∑ I a I dx I and β £ ∑ I b I dx I be constant k-forms. Prove that αβ is exact. Calculate d ® d ® α . (iv) α ªµ® β «¶£ β ªµ® α « . º (i) f £»® d ® d f .13. DIFFERENTIAL FORMS ON EUCLIDEAN SPACE 2. β «¹£±ªµ® α . 2. Let g ª x «¬£ Show that dg p 1 1 i 1 ∑ xi f i ª x « . (Use Exercise 2. (i) Find ® α and ® d ® dα . On U deﬁne a vector ﬁeld n. α «¬£ 0 if and only if α £ 0. now regarding α as a form on R 4 . β and γ are as in Exercise 2.14. (i) The dx I form an orthonormal basis of the space of constant k-forms. ∂ x2 n Prove the following formulas.13(iv). º º º (ii) ª f g «£±ª f « g f g 2 ®¼ª d f ªµ® dg «½« . Let f : Rn ¾ R be a function and let α £ f dx i . Let α be closed and β exact.15. where ν £Á x Á Â 1 x ¸ ® dx. ®¯ª αβ « . (v) The Hodge star operator is orthogonal. i. with constant coefﬁcients a I and b I .5 to each f i . regarding α as a form on R 3 . Prove that αβ is also closed. (We also assume. 2. ®°® α £±ª ¥ 1 « kn ² kα for every k-form α on Rn . 2. Let α and β be closed forms.) The inner product of α and β is the number deﬁned by ª α . β «¬£ ∑ a I b I . where is the Laplacian deﬁned in Exercise 2. ¤ n £ α . Prove that 2. º º Show that d ® d ® α ¿ª ¥ 1 « n ® d ® dα £±ª f « dxi .11. ® β « .10. (iii) α ªµ® β «¶£·ª α . grad f ª x « .

J is the density of electric current. H.) Ñ as Ñ in the deﬁnition of the ordinary Ñ (iii) Find 1. Ñ Ñ “relativistic Laplacian” (usually called the d’Alembertian or wave (iv) Compute the Ñ f on Rn Æ 1 . The Minkowski or relativistic inner product on R n Æ is given by Ç A vector x x. x3 . if I does not contain n Ô 1.18. Ñ (ii) Conclude that γ is closed and that div J Ô ∂ρ × ∂t É 0. Ô J2 dx3 dx1 Ô J3 dx1 dx2 È dx4 Ë ρ dx1 dx2 dx3 . Ç (ii) Show that for every x É Ï 0 there is a y such that x. Show that the Maxwell equations in free space are equivalent to dα É d α É 0. y È¹É Ì Rn Æ 1 is spacelike if x. B is the magnetic induction and D is the dielectric displacement. E.EXERCISES 29 1 2. 0. y ÈÐÉ Ï 0. as we shall now see. H is the magnetic ﬁeld. dx i for 1 Õ i Õ n Ô 1. D2 dx3 dx1 Ô D3 dx1 dx2 . Ç 2. lightlike if x. x ÈÄÎ Ç 0.17. introduce forms α β γ É É Ç É Ë E1 dx1 Ç Ô E2 dx2 1Ç J dx dx c 1 2 3 H1 dx1 Ô Ô E3 dx3 È dx4 H2 dx2 Ô H3 dx3 È dx4 Ô B1 dx2 dx3 Ô Ô B2 dx3 dx1 D1 dx2 dx3 Ô Ô B3 dx1 dx2 . and timelike if x. ρ is the density of electric charge. One of the greatest advances in theoretical physics of the nineteenth century was Maxwell’s formulation of the equations of electromagnetism: curl E curl H div D div B É Ë É É É 1 ∂B c ∂t 4π 1 ∂D JÔ c c ∂t 4πρ 0 (Faraday’s Law). In space-time R4 with coordinates x 1 . (iii) In vacuum one has E É D and H É B. The Maxwell equations look particularly simple in differential form Ç notation. Here c is the speed of light. x2 . then Ç α É ∑ f I dx I È .17. operator) d d f for any smooth function Ç (v) For n É 3 (ordinary space-time) ﬁnd dx i dx j È for 1 Õ i Î j Õ 4. (Ampère’s Law). (Gauß’ Law). I É with Ñ dx I ÉÓÒ Ë ε I dx I c ε I dx I c if I contains n Ô 1. the relativistic Hodge star of α deﬁned in Exercise 2. x4 È . (no magnetic monopoles). (Here ε I and I c are Hodge star. x È´Í Ç i 1 ∑ xi yi Ë Ê n xn Æ 1 yn 1 . B and D are vector ﬁelds and ρ is a function on R 3 and all depend on time t. E is the electric ﬁeld. where x4 É ct. Æ 0. x È¬É Ç 0. (i) Show that Maxwell’s equations are equivalent to dβ Ô 4πγ dα É É 0. J. Show that in vacuum β É α . A Hodge star operator corresponding to this inner product is deﬁned as follows: if α Ñ Ñ ∑ I f I dx I . Ñ (iv) Free space is a vacuum without charges or currents. . and dx 1 dx2 Ö¦ÖuÖ dxn È . (i) Give examples of (nonzero) vectors of each type.

. B Ù x ÚÛÝÜÞ ß 0 g Ù x1 ß x4 Ú f Ù x1 ß x4 ÚÐàá . Explain why. DIFFERENTIAL FORMS ON EUCLIDEAN SPACE (v) Let f . g : R Ø R be any smooth functions and deﬁne 0 Show that the corresponding 2-form α satisﬁes the free Maxwell equations dα Û d â α Û 0. In what direction do these waves travel? E Ù x Ú¬ÛÝÜÞ f Ù x1 g Ù x1 ß ß x4 Ú x4 Ú½àá . Such solutions are called electromagnetic waves.30 2.

n îîî îî a1.1. . .1 . . . Determinants The determinant of a square matrix is the oriented volume of the block (parallelepiped) spanned by its column vectors. an . D EFINITION . . and second. . a2 . . This section is a review of some fundamental facts concerning determinants. .) It is a useful rule. . it is extremely inefﬁcient computationally (except for matrices containing lots of zeroes). Let A ãjäåæ an. . îî îî îî an. an ì ã det ë ai.CHAPTER 3 Pulling back forms 3. . The determinant can be completely characterized by three simple laws. . It is therefore not surprising that differential forms are closely related to determinants. j n í a1. an.1 .. an ì a number det ë a1 . an. . which make good sense in view of its geometrical signiﬁcance and which comprise an efﬁcient algorithm for calculating any determinant. but it has two serious ﬂaws: ﬁrst. 3. Axioms. a2 . A determinant is a function det which assigns to every ordered n-tuple of vectors ë a1 . iï 1 n Here Ai. j ì 1 í i.1 a1. . .. î. . . a2 . j denotes the ë n ð 1 ì êòë n ð 1 ì -matrix obtained from A by striking out the i-th row and the j-th column. .n . which reduces the calculation of any determinant to that of determinants of smaller size. . .1 . You have probably seen the following deﬁnition of the determinant: det A ã ∑ ë¦ð 1 ì i ñ 1 ai1 det Ai. î Expansion on the ﬁrst column. . . it obscures the relationship with volumes of parallelepipeds. a2 . é çQè be an n ê n-matrix with column vectors a1 . A far better deﬁnition is available. ã6î . Its determinant is variously denoted by det A ã det ë a1 .. .. the determinant of a 1 ê 1-matrix ë a ì is simply deﬁned to be the number a. . .1. .n .. This is a recursive deﬁnition.1 . (The recursion starts at n ã 1. an ì subject to the following axioms: 31 .n a1. . ..

a2 . T HEOREM (uniqueness of determinants). This can be restated as follows. E XAMPLE . . . . .2. . Then the columns of A are linearly dependent. . . . . ü ÷ ûý c if E is of type II (multiplication of a column by c). . . . . det ó a1 . the axioms (i)–(iii) sufﬁce to calculate any n ndeterminant. . . In other words. Then det A ÷ det õ A for all n n-matrices A. . . then det ó E ó A öuö where þ 1 if E is of type I. . cai (ii) det is alternating or antisymmetric: for any i ÷ ú j.e. ¡ P ROOF. Axioms (i) and (ii) govern the behaviour of oriented volumes under the elementary column operations on matrices. Axiom (iii) lays down the value of det I . . . an . . . . . 3. . a j . . . . . 3. . an ö ÷ 1. e2 . an ö ÷ c det ó a1 . As this example suggests. . . . where A is the matrix whose columns are a1 . . multiplying a column by a nonzero constant (type II). ai . a2 . aiõ . Let det and det õ be two functions satisfying Axioms (i)–(iii). e2 . a2 . . Suppose ﬁrst that A is not invertible. . PULLING BACK FORMS (i) det is multilinear (i. .4. . an be the column vectors of A. . 1 4 1 1 10 5 1 9 4 1 ÷ 4 1 0 6 4 0 5 3 1 0 0 1 0 ÷ 4 1 5 ÷ 3 1 1 1 3 0 1 1 0 0 1 ÷ 2 3 1 1 ÷ 2 0 0 1 0 0 0 2 0 0 0 1 0 1 0 ÷Cø 2 1 0 0 0 1 0 0 0 1 ÷ùø ¡ 2. . . . . an . . . a j . . . an ö i for all scalars c. . . Identify the column operations applied at each step in the following calculation. . . . a2 . . . . en ö dard basis vectors of Rn . . . where e1 . . a2 . and type III causes a sign change. Type I does not affect the determinant. . ÷ k det A. linear in each column): det ó a1 . . . . . aõ . . 3. . . If E is an elementary column operation.32 3. . For simplicity let us assume that the ﬁrst column is a linear combination of the others: a1 ÷ c 2 a2 ô ô cn an . . ai . k ýÿ ø 1 if E is of type III. ai . Applying axioms (i) and (ii) we get £¢¤¢¤¢ det A ÷ ∑ ci det ó ai . .3. More precisely. . . . a2 . . . we have the following result. Let a1 . ai . c õ and all vectors a1 . ai . L EMMA . i 2 ¥ n . . ÷ùø det ó a1 . . . type II multiplies it by the corresponding constant. . . . an ö ô c õ det ó a1 . and interchanging any two columns (type III). . . a2 . . en are the stan- We also write det A instead of det ó a1 . an ö ô c õ aiõ . an ö . Recall that these operations come in three types: adding a multiple of any column of A to any other column (type I). a2 . . an ö ÷ 0. there is at most one function det which obeys these axioms. (iii) normalization: det ó e1 .

so that E m Em 1 Lemma 3. . E2 .2.and for the same reason det A 0. Em E2 E1 A I . . DETERMINANTS § ¦ 33 ¨ ©¤©¤© . so axiom (iii) yields ¦ § 3. it can be transformed to I by successive elementary column operations. . each operation E i has the effect of multiplying the determinant by a certain factor k i . According to be these elementary operations. i. so det A det A. Let E 1 . . Now assume that A is invertible. Then A is column equivalent to the identity matrix.1.e.

.4 shows that det A § c det A for all n n-matrices A. n to itself. . 4 . 2.1. . For n 3 all possible permutations are 1 n 2 ©¤©¤© 3 © 2 © 1 § n! permutations. R (change of normalization). An alternative way of thinking of a permutation is as a bijective (i. . 1 2. S n stands for the collection of all permutations of the set 1.σ 2 n . for n § 5 a possible permutation is 5. 3.§ ¨ ¤ © ©¤¦ © E E A § k k ¨ ¤ © ©¤© k k det A¦ .6.5. Thus for n 2 there are only two permutations: 1. 1 . n. Applying the © ©¤© k k det A. For example. . n . . 2. one-to-one and onto) map from the set 1. .σ a ©¤©¤© a . T HEOREM (existence of determinants). this formula is not very practical for the purpose of calculating large determinants. Unlike Deﬁnition 3. Suppose that det is a function that satisﬁes the multilinearity ¦ axiom (i) and the antisymmetry axiom (ii) but is normal¦ ized differently: det I § c. A permutation is a way of ordering the numbers 1. but there are many different ways of performing this reduction. 2 and 2. It is given by the formula det A § σ Sn ∑ sign σ a1. but it has other uses. 3. We can calculate the determinant of any matrix by column reducing it to the identity matrix. Do different column reductions lead to the same answer for the determinant? In other words. 1 § det I § det Em Em 1 2 1 m m 1 2 1 m m 1 2 1 1 2 m EMARK This result leaves an open question. Then the proof of Theorem 3. notably in the theory of differential forms.σ n n-matrix A has a well- This requires a little explanation. and we think of this as a shorthand notation for the map σ given by σ 1 . . 2. . Every n deﬁned determinant. 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. .e. . . 1. QED ¦ 3. Hence same reasoning ¦ to det A we get 1 § k k ¨ ¤ det A § 1 k k ©¤©¤© k § det A. 2. Permutations are usually written as row vectors containing each of these numbers exactly once.

However. 3 . . .§ 5. . . σ 4 § 2 and σ 5 § 4. An inversion in σ is any pair of numbers i and j such that 1 i j n and For general n there are n n 1. 1 . § 2. it cannot preserve the order in this way. n 1. 2 . 3.§ 2. 2. 2. 2. n . if σ is not the identity permutation. . 2 . n then corresponds to the identity map on the set 1. . σ 3 § 1. 1. . 3. The permutation 1. then clearly σ i σ j whenever i j. . 3. 3 . 3. 1 . . 3. 1. If σ is the identity permutation. 1. 2. σ 2 § 3.

. 3.1 a1. which has sign ! 1. 2. 1. 2 ' are 1. resp.3 a2. This is the permutation produced by ﬁrst performing τ and then σ ! For instance.1 3.2 a1. 4 has length 6 and so is even. 1. For instance. 3 1 !1 1 2.2 a2. Each term is a product which contains exactly one entry from each row and each column of A.6 gives the well-known formulæ .3 a3.2 a2.. . 3. 2 1 !1 2. 3 0 1 1. στ ) 4.1 2.1 a2. 5. the permutation 5.2 a3.¤. 1 . The permutations of & 1.. 2 2 3. Let us for the moment denote the right-hand side by f A . . σ l σ sign σ 1. 4 and τ 5.5 a2. n ' to itself. one for each permutation σ .. 2. Axiom (i) is checked as follows: for every permutation σ the product a1. a " # 1 2.1 2.3 .2 a2. 1 lσ sign σ 1 if σ is odd..34 3. denoted by l σ . + + . $ " # ! ! Thus sign 5.2 a3. 3. 2.1 .2 a2. a1.σ 2 n .¤.1 a2. 2.1 a1. 5 . The determinant formula in Theorem 3.σ ..3 a2. 1 3 !1 Thinking of permutations in S as bijective maps from & 1.2 ! ! a1. (3. odd. A permutation is called even or odd according to whether its length is even. 3. then τσ 1.1.1 . 2. which we shall not prove here. a .. a 1. 3. a1. a 1.1 a3. we can form the composition σ ( τ of any two permutations σ and τ in S .1 a3.and 3 3-determinants Theorem 3. The length of σ . 2. 3.1 a4..6 obeys axioms (i)–(iii) of Deﬁnition 3. 1. 1 2 1 3.. 4. a .2 a2. 3. 2.6 contains n! terms.3 ! a1. the product of two even permutations is even and the product of an even and an odd permutation is odd. 3. if σ 5. while for n 3 we have the table below. PULLING BACK FORMS σ i σ j .3 a3. P ROOF OF T HEOREM 3. For permutations we usually write στ instead of σ ( τ and call it the product of σ and τ .3 a2.6.2 ! a1.2 a3. The sign of σ is 1 if σ is even. 4 % 1. 4 contributes the term a 1. 3.3 a3. We need to check that the right-hand side of the determinant formula in Theorem 3.1) In particular. 1 . For instance.2 a5. a . is sign στ * sign σ sign τ . 2 .3 a3.σ n - a1. 1. which has sign 1. 2 . 4.. . . n n A basic fact concerning signs. 1. 1. for n 5 the permutation 5. 2. is the number of inversions in σ .1 a2. 2. and 2. 1.2 . . For 2 2.2 a1.3 " # a " #/.4 . a1.1 ..

. To see this. C D 0. j-th entry is a j. . a i i n 1 2 i n 0 changes sign. a 1 . a 1*2 f 0 a . . DETERMINANTS 35 contains exactly one entry from each row and each column in A.σ 1 2. . The unit cube in Rn . We conclude this discussion with a slightly different geometric view of determinants. (ii) det A T det A. . a i . a 1 3 f 0 a . 2. Each can be deduced from Deﬁnition 3. Aen .n if A is upper triangular (i. . .1. a i . A square matrix A can be regarded as a linear map A : Rn Rn . Hence A 0. . each term in f 0 A 1 f a1 . j det Ai. . . . . . . (iii) (Expansion on the j-th column) det A ∑n 1 i j ai. . . .τσ n 1. Ae2 .4 5 2:9 f 0 a . a . . 1 n under the map A is a parallelepiped with edges Ae1 .ρ 1 2. j 1. . a j . a . . . . . . 0 1 Similarly. . . . a . . a 6 7 a 6 7 8¤8¤8 a 6 7 2<. . . a 1 . . . . .ρ n 1. rule (iii) is correct because if A 2 I .7. j 0 for i j). . . . . .ρ n i j n 1. a . Therefore 0 1 f 0 a . 1 n 2 2 J 1 for i LD E M 2 1. Let A and B be n (i) det AB det A det B. . n . . .ρ 2 n .1) 5 2:9 ∑ sign 0 ρ 1 a 6 7 a 6 7/8¤8¤8 a 6 7 by Exercise 3.6. 0 > 12 2 2 = n-matrices.σ n Here are some further rules followed by determinants. a2 . . . So f satisﬁes all three axioms for determinants. . j for all j i 1 i . .ρ n ρ Sn 1 1. 1E 2GF x H n Rn 0 I J xi has n-dimensional volume 1. . . (For n 1 it is usually called the unit interval and for n 2 the unit square. a i. a n 6 7 a 6 7/8¤8¤8 a 6 7 2 ∑ sign 0 τρ 1 a 6 7 a 6 7 8¤8¤8 a 6 7 substitute ρ 2 τσ 5 2 ∑ sign 0 τ 1 sign 0 ρ 1 a 6 7 a 6 7/8¤8¤8 a 6 7 by formula (3. a . .ρ 1 2. . . a4 . (Recall that the transpose of an n nai. . . and therefore f 0 I 12 1. . a . T HEOREM .τσ 2 n . a n .e. . 1 if σ 2 identity. . K LD E M .τσ 0 1 1 1 2.2 an. .ρ 1 2. . Then f a1 . .1 a2. each term in f A is multiplied by c. . 3 a4 . . a 2 . QED σ Sn ρ Sn ρ Sn 0 2 ∑ 5 sign σ a1. .σ 2 n . . . let τ be the permutation in S n that interchanges the two numbers i and j and leaves all others ﬁxed.3. 0 otherwise. a 1*2 1 2 i n i n 1 2 c f a1 . Axiom (ii) holds because if we interchange two columns in A.ρ 2 n . . . . .) matrix A 2)0 1 = 3. . . n.) Its image A 0.ρ 2 n . . . ca . . . Here A denotes the n 1 n 1 -matrix obtained from A by striking out the i-th row and the j-th column.i. . j is the matrix A T whose i . Finally. (iv) det A a1. . So if we multiply the i-th row of A by c. 2.1 or from Theorem 3. the columns of A. 8¤8¤8 2 ? 0@9 1 A 0 9 1*=0 9 1 2 2 B Volume change.

. xn . . ym and let us write. then Q RUT TVR<T N O P S vol A W X X RYT det A T vol X . . 1 n det A det A vol 0. xn and that the old variables are given in terms of the new by functions y1 y2 R R R φ1 x1 . y2 . α ∑ f I dy I . . . n-dimensional volume. xn . . E XAMPLE . . αR x ^ y R R By substituting x cos t and y sin t into the angle form we obtain the following 1-form on R: Z sin t d cos t ^ cos t d sin t R N W Z sin t X_W Z sin t X`^ cos tQ dt R dt. . T T We can take any k-form and substitute any number of variables into it to obtain a new k-form. Suppose α is a k-form deﬁned on an open subset V of Rm . . . W W X X ym φm x1 . .10 we deﬁned the angle form on R 2 Z\[ 0 ] to be R I where the functions f I are deﬁned on V . . . cos t ^ sin t 2 2 2 2 2 3. 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. Z y dx ^ x dy .8. as usual. Suppose we want to substitute “new” variables x1 . xn . but it sufﬁces for our purposes to know that all open and all closed subsets of Rn are measurable. A A e2 3. This works as follows. . φ2 x1 . . . x2 . PULLING BACK FORMS T O P e2 X AX e1 A e1 So det A can be interpreted as a volume change factor.) 3. In Example 2. . .2. Let us denote the coordinates on Rm by y1 . (A set is measurable if it has a well-deﬁned. . This rule n generalizes as follows: if X is a measurable subset of R . . Explaining exactly what this means is rather hard.36 has n-dimensional volume vol A 0. . . W X . . ﬁnite or inﬁnite. 1 n .

if I i 1 . . . .) The pullback of α along φ is then the k-form φ α on U obtained by substituting y i φi x1 . xn for all i in the formula for α . . . m We assume that the functions φ i are smooth and deﬁned on a common domain U . φ b xc φ1 x φ2 x . (In Example 3. . c_b p c q p φ fI p a f I φ x . its pullback φ α ∑ J g J dx J is a k-form in x1 . y2 . where h ii b cj i b c φ b x caedf ffg k . in other words. . .12 below we will give an explicit formula for the coefﬁcients g J in terms of f I and φ.3. which is an open subset of Rn . The formula φ wx ayw ln b xx xz x c x x x 1 2 3 1 2 1 2 . p p The picture below is a schematic representation of the substitution process. a p a U V φ α φ y x s α t φ u xv Rm Rn 3. V R2 0 and φ t cos t. ym . ik we put φ dy I p a φp b dy i1 dyi2 p b ca b b cjc a b c a b c r¤r¤r dy c*a dφ dφ r¤r¤r dφ . The form α ∑ I f I dy I is a k-form in y1 . . The pullback φ dy I is deﬁned by replacing each y i with φi . x2 . . sin t . . . That is to say.2. φ α is deﬁned by a a p lnm o φ α Here φ f I is deﬁned by p a ∑ b φp I p b c*a)b c a b c f I φ dy I . We regard φ as a map from U to V . . .9. PULLING BACK FORMS 37 As usual we write y a φ b xc . This means φ f I x is the function resulting from f I by substituting y φ x . i2 . In Theorem 3. φ f I the composition of φ and f I . That is to say. . . ik i1 i2 ik f I φ.8 we have U R. . . E XAMPLE . xn .

10. J I J I J I J I. Accordingly. φ dy1 φ dy1 dy2 | | { dφ1 dφ2 dφ1 φ dy2 | Observe that the pullback operation turns k-forms on the target space V into k-forms on the source space U . For the proof of part (ii) consider two forms α | ∑ f dy and β | ∑ g dy (not necessarily of the same degree). dφ | 3 x x dx x dx _ x x dx | 3xx x x x dx dx . Pulling back forms is nicely compatible with the other operations that we learned about (except the Hodge star). (Recall that k U stands for the collection of all k-forms on U . 3 1 2 2 1 2 1 3 1 2 1 2 1 2 1 1 2 2 2 1 2 1 3 1 2 1 2 1 1 2 2 2 1 3 1 2 1 2 | |Y} ~ | dx2 { φ : the opposite way from what you might naively expect. The pullback operation is (i) linear: φ aα bβ aφ α bφ β. P ROOF. | d ln x x | x x dx dx . x2 ln x1 x2 . This proves part (i). so φ aα bβ | ∑ aφ f bφ g _ φ dy | aφ α bφ β. If α ∑ I f I dy I and β ∑ I g I dy I are two forms of the same degree. Then αβ | ∑ f g dy dy .) The property that φ “turns the arrow around” is called contravariance. Thus. then aα bβ ∑ I a f I bg I dy I . so φ αβ | ∑ φ f g φ dy dy . (iii) natural: φ ψ α ψ φ α . _ Now | aφ f x bφ g x . φ a fI bg I x bg I φ x a fI φ x bg I φ x I I I I I I I I I J J J I. V { U . PULLING BACK FORMS deﬁnes a map φ : U R2 . { | | | ¤ | { 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. k k 3. I J I J . while φ : U V is a map from U to V . The components 3 of φ are given by φ1 x1 .38 3. φ is a map | d x x | 3x x dx x dx . Let φ : U V be a smooth map. x2 x1 x2 and φ2 x1 . (ii) multiplicative: φ αβ φ α φ β . where ψ : V with W open in Rk and α a form on W. where U x R2 x 1 x 2 0 . where U is open in R n and m V is open in R . J I J ¤ | a f _ | ` Now φ f g _ x | I J fI gJ φ x | f φ x g φ xj | φ f _ φ g _ x . P ROPOSITION . so | | φ aα bβ | I ∑ φ I af I bg I φ dy I .

Then ψ f φ x f ψ φ x j f ψ φ x f ψ φ x* ψ φ f x . Next consider a 1-form α dz on W . dy . φ ψ α ∑ ∂x Because every form on W is a sum of products of forms of type f and dz . formula (B. For the proof of property (iii) ﬁrst consider a function f on W . so z are the variables on R . T HEOREM . Then φ df φ ∑ ∂∂yf dy ∑ φ ∂∂yf m i 1 i i i 1 dφi ∂f ∂φ φ ∑ ∑ ∂x dx ∂ y ∂f ∑ ∑ φ ∂y m n i i 1 i j 1 m j j n j 1i 1 i ∂φi dx j . PULLING BACK FORMS 39 so φ so f g φ f φ g . Then φ dα d φ α for α m ¡£¢ φ d dφ . Then ψ α dψ ∑ ∂ψ ∂ψ φ ψ α ∑ φ φ dy ∑ φ dφ ∂ y ∂y ∂φ ∂ψ ∂ψ ∂φ φ dx . Let φ : U V be a smooth map. Therefore ∂ φ ψ dx d φ ψ d ψ φ ψ φ dz ψ φ α . J f I φ g J φ dy I φ dy J I I ∑ φ f φ dy @ ∑ φ g φ dy φ α φ β . . I I J J which establishes part (ii). φ dy dy * φ dy φ dy ¤¤ dy dy ¤¤ dy dφ dφ ¤¤ dφ dφ ¤¤ dφ φ dy φ dy . z . property (iii) in general follows from the two special cases α f and α dz . . First let f be a function. QED φ ψ f x k k i m ∂ψ i j 1 ∂y j i 1 2 j m i m j 1 j j i j 1 j j m i n j n m j 1 j l 1 l l i j l 1 j 1 j l l m j 1 i j j l i l n i l 1 l l i i i i i Another application of the chain rule yields the following important result. Furthermore. where z . i P ROOF. so φ ψ f ψ φ f . ∂x j . . . where U is open in Rn and V is m k V .6).3.11. I J I I J J i1 i2 ik j1 jl i1 i2 ik j1 jl I J φ αβ ∑ φ I.2. In short open in R . dx ∑ ∑ φ ∑ ∑ ∂ y ∂x ∂ y ∂x By the chain rule. the sum ∑ φ ∂ψ ∂ y ∂φ ∂ x is equal to ∂ φ ψ ∂ x . 3.

l ∑ n ¤ ∂φi ∂φ j dxk dxl ∂ xk ∂ xl 1 © 1 k l n ° ± ° ² ©¶µµ µµ µµ ∑ ∂φi ∂φ j ∂ xk ∂ xl ³ µµ µµ µµ ∂φi ∂φ j dxk dxl . Then dα I i1 i2 © ∑ I d f I dy I . formula (B. with g © ∑ ¤ ¦ φ ¥ f ¨ For a 2-form α © ∑ ° ± ° f dy dy we get φ¥ α © ¦ φ¥ f ¨ φ ¥ ¦ dy dy ¨ © ∑ ¦ φ¥ f ¨ dφ dφ . (See Exercise 2.5(ii). dφ ¥ α © ∑ d «@¦ φ ¥ f ¨ ¦ φ ¥ dy ¨@¬ © ∑ d «@¦ φ ¥ f ¨ dφ dφ ª¤ª¤ª dφ ¬ © ∑ d ¦ φ ¥ f ¨ dφ dφ ª¤ª¤ª dφ ∑ ¦ φ ¥ f ¨ d ¦ dφ dφ ª¤ª¤ª dφ ¨ © ∑ d ¦ φ ¥ f ¨ dφ dφ ª¤ª¤ª dφ . ¤ ¤ ¤ n m i n j Observe that dφi dφ j where ∂φi ∂φ j ∂ xk ∂ xl © k. Let us do this ﬁrst in degrees 1 and 2. ¥ so the theorem is true for functions. ¥ © ∑ φ ¥ ¦ d f dy ¨ ©Y¦ φ ¥ d f ¨ ¦ φ¥ dy ¨ © ∑ d ¦ φ¥ f ¨ dφ dφ ª¤ª¤ª dφ because φ ¥ d f © dφ ¥ f .40 3. j i. The pullback of a 1-form α ∑m i 1 f i dy i is Now dφi © ¥ © ∑n j ¤ © ¤ φ ¥ α © ∑ ¦ φ ¥ f ¨ ¦ φ ¥ dy ¨ © ∑ ¦ φ ¥ f ¨ dφ . Hence ¦ ¥ ¨§ φ df ¥ © j ∑ ¤ n ∂ φ f dx j ∂x j 1 ¦ ¥ ¨ ¤ ¥¦ § ¨ § © dφ f . which establishes a connection between forms and determinants. ¤ ¤ m m i 1 i i i 1 i i ¥ ∂φ i 1 ∂x j m dx j and so i ¤ ® ¤ . PULLING BACK FORMS By the chain rule. j i j 1 i j m i j 1 i j m i. On the other hand. Proposition 2.8. the quantity ∑m i 1 φ ∂ f ∂ y i ∂φ i ∂ x j is equal to ∂ φ f ∂ x j . Next let α so © ∑ I f I dy I . ik . °∑ ± ° ° ± ° φ α i 1 j j 1i 1 i j j j 1 j j m i 1 ∂φ i i ∂x j 1 i j m i. ∂ xl ∂ xk ´ ³ ∂φi ∂φ j ∂ xl ∂ xk ∂φ i ∂ xk ∂φ j ∂ xk ∂φ i ∂ xl ∂φ j ∂ xl . j i j ∑ ¦ φ¥ f ¨ ∑ n ∂φi dx j ∂x j 1 ∂φ ¯ © ∑ ∑ ¦ φ ¥ f ¨ ∂x dx © ∑ g dx . φ dα I I I I I I I I I I I I I I I i1 i2 ik I I i1 i1 i2 i2 ik ik I I i1 i2 ik Here we have used the Leibniz rule for forms. plus the fact that the form dφi1 dφi2 dφik is always closed.6). ª¤ª¤ª ¥ © ¥ We ﬁnish this section by giving an explicit formula for the pullback φ α .) Comparing the two QED equations above we see that φ dα dφ α .

j2 . The notation Dφ I . . . . that is the k k-matrix obtained from the Jacobi matrix by extracting rows i 1 . jσ k . then dx ¹ 0. . . jk .3. .2) Æ ½ ÇÈ ÇÈ ÇÈ ∑ ∑ J σ Sk ∂φi1 ∂φi2 ∂ x jσ 1 ∂ x jσ 2 jσ jσ 1 jσ 2 jσ k i1 ik ∑ det Dφ I . i2 . jk is an increasing multi-index and σ S k is a permutation. .l k l k. .2. . So we get · distinct. m 2 . ik and columns j1 .. J-submatrix of Dφ. This gives ∂φ ∂φ ∂φ dx dx Á¤Á¤Á dx dφ dφ Á¤Á¤Á dφ ¹ ∑ ¤ Á ¤ Á Á Â ∂x ∂x ∂x ∂φ φ ∂φ dx . .2) used the result of Exercise 3. J stands for the I . . j ∂φ j 1 k l n ∂ xk ∂φ i ∂ xk ∂φ j ∂ xl ¹ ½ ¾ ¹ M ½ ¾ ¹ ½ ÃÄ ÃÄ Å ∂φik jσ i2 k Ã Ä¾ dφi1 dφi1 Á¤Á¤Á dφ ¹ ik ¹ dx dx dx Æ Ç È Ç È Á¤Á¤Á ∂x Ç È Ç È Ç È Á¤Á¤Á Ç È ¹ ∑ ∑ sign σ ¾ ∂∂xφ ∂∂xφ Á¤Á¤Á ∂∂xφ dx (3. . In other words. m . jσ 2 . j2 . ½ ½ ½ To write the product dφ dφ Á¤Á¤Á dφ in terms of the x-variables we use ∂φ dx dφ ¹ ∑ Â ∂x for l ¹ 1. . . ¹ ∑ ∂∂x ¤ Á ¤ Á Á ∂x ∂x in which the summation is over all n multi-indices M ¹ m . where J j1 . Thus we can rewrite the sum over all multi-indices M as a double sum over all increasing multi-indices J and all permutations σ : φ¸ dx dx À º∑ » º > ¼½ ¾ º∑ » º ¿¿ ¿¿ ¿¿ ¿¿ ¿ ¹ ∑ ∑ φ ¸ ¿ f ¾ ¿¿ dx dx ¹ ∑ g dx dx ¿¿ ¿¿ º » º º » º ½ º » º ¿ ¿ ¿¿ ¿¿ with ¿ ¿ ¸ φ f . J dx J . k.. m ¾ . . .6. j ∂φ i ∂ xk ∂φ j ∂ xk ∂φ i ∂ xk ∂φ j ∂ xl k l 1 k l n k. m k 1 i1 m1 m2 ik mk m1 m2 mk i2 M m1 m2 mk M k 1 2 k ¸ ¹ ∂φ i ∂ xk f i.3) we applied Theorem 3. .. . . . PULLING BACK FORMS 41 is the determinant of the 2 2-submatrix obtained from the Jacobi matrix Dφ by extracting rows i and j and columns k and l . j ∂φ i ∂ xk ∂φ j ∂ xk ∂φ i ∂ xl ∂φ j ∂ xl I I I i1 ik I I i1 i2 ik i1 i2 ik n il il ml 1 ml ml n i1 i2 ik i1 i2 ik m 1 . g ¹ º∑ » º ½ ¾ ¿¿ ¿¿ ¿¿ ¿¿ For an arbitrary k-form α ¹ ∑ f dy we obtain ¿¿ ¿¿ φ ¸ α ¹ ∑ φ ¸ f ¾ φ ¸ dy dy Á¤Á¤Á dy ¾ ¹ ∑ φ ¸ f ¾ dφ dφ Á¤Á¤Á dφ . . we can rearrange them in increasing order by means of a permutation σ . mk jσ 1 . . we have M m1 . . If a ½ entries of M are multi-index M has repeating entries. m2 .7 and in (3. . . .3) In (3. 2. · . J J σ Sk 1 jσ 2 jσ J k (3. .. If the all φ α 1 i j m k l 1 k l n1 i j m i.l 1 i j m I I i2 i. .

where U and V are open in R n . . dx n and another with edges dφ 1 . i 1. so we see that det Dφ Ì xÍ can be inα f dy1 dy2 n Ê Ê Ð Ð¤Ð dy . 3. . PULLING BACK FORMS To sum up. Ê 1 (constant function) then φ É f Ê 1.4. n where 1 i j n. .7(iv) from Theorem 3. 3. Let φ : U V be a smooth map. Tabulate all permutations in S 4 with their lengths and signs. dφ n . dφ2 . Ó Ò Õ Ó Ò Ô Õ Ö Ò Ô 3. T HEOREM . Determine the length and the sign of the following permutations.1. (i) A permutation of the form 1. 3. and we are pulling back a form of top degree. n 1. J . . where U is open in Rn and V is m open in R .) (ii) n. However. Ë Ì I f I det Dφ I . Then the pullback of the volume form on V is equal to the Jacobi determinant times the volume form on U. 1 3 2 1 1 1 4 1 ÑÑ ÑÑ ÑÑ determinants using column and/or row operations and 1 5 2 3 1 2 . It interchanges i and j and leaves all other numbers ﬁxed. . . j. Then φ α is the k-form on U given by φ α ∑ J g J dx J with É Ê Ê Ï É gJ Ê ∑ φÉ I Ì f I det Dφ I . Then Ê If f terpreted as the ratio between the oriented volumes of two inﬁnitesimal blocks positioned at x: one with edges dx 1 . It is almost always easier to apply the deﬁnition of pullback directly. ¤ φ α 1 2 n Ï φ dy1 dy2 ÉÌ Ð¤Ð¤Ð dy Í Ê Ì det Dφ Í dx dx Ð¤Ð¤Ð dx . . j 1. n 2.12. . É Ê Ì φ É f Í Ì det Dφ Í dx dx Ð¤Ð¤Ð dx . Deduce Theorem 3. Let α ∑ I f I dy I be a k-form on V. (Such a permutation is called a transposition.42 3.6. Thus the Jacobi determinant is a measurement of how much the map φ changes oriented volume from point to point. one of which we record here. Let φ : U V be a smooth map.7(iv). . 3 7 ÑÑ Ò ÑÑ ÑÑ ÑÑ ÑÑ ÑÑ ÑÑ 1 3 Ò ÑÑ 0 ÑÑ 2 Ò Ò 1 1 1 1 Ò 2 1 1 2 4 3 . . dx2 . 0 5 ÑÑ ÑÑ ÑÑ ÑÑ 3. . . .5. Calculate the following Theorem 3. . . Í This formula is seldom used to calculate pullbacks in practice and you don’t need to memorize the details of the proof. J dx J Ê ∑ ∑ φ É I J J This proves the following result.13. . 1 . i . 3. . . the formula has some important theoretical uses. . . n 1 2 n Exercises 3. 2. T HEOREM . J Í Î dx J .3. we ﬁnd φ α É Ê ∑ φ É f I ∑ det Dφ I . . Assume that k m n. . Find all permutations in S n of length 1. 3. . 2. the number of new variables is equal to the number of old variables. . that is to say.2. .

(i) Suppose that A has the shape ðòññ a a . n and τ n.. . Let σ be a permutation of è 1. . . 2. τ 1 . .. .2 2. Show that i1 i2 ik k 1 2 n 3. . σn 1 .1 1. n Ü 1. n positions interchanging 1 and 2.7(ii) from Theorem 3. . .1) and Exercise 3. σ2 . . îîï .10.6. . . .e. 1. i Û and any permutation σ in S . (ii) σiσi 1 3 1 for 1 i (iii) σiσ j 2 1 for 1 i .6. .13. a A ØUíî . i .. 3. A e Ø e ë ì . (ii) Deduce Theorem 3. j å æ n Ü 1. . 6. . (iii) Show that det A Ø sign Ù σ Û . 1 xn x2 n . Show that x × . . where (i) σ 3. .6 that ôô . . . ó 0 a . x . . . n n . æ n and i ä 1 æ j. n 1. . 1. .n i. 1. . The i-th simple permutation is deﬁned by σ i So σi interchanges i and i 1 and leaves all other numbers ﬁxed.. 4 and τ 5. i 3.. 3. n é . . 3.7. . .n . i 1. 2. .ôô .. ô ôô (ii) Deduce from this the expansion ô rule. an. an. (i) Show that every permutation has the same length and sign as its inverse.8. The permutation matrix corresponding to σ is the n ê n-matrix A whose i-th column is the vector e ë ì . 6. Prove the Coxeter relations (i) σi2 1 for 1 i n. 2.. In other words. .. . .1 ôô ôô a2. 3. ôô Ø Ùx Ü xÛ ôô ∏ ôô õ j i e1 .11. resp.EXERCISES 43 3..e. . . Deduce from Theorem 3. . Show that for n â 2 the permutation group S has n! ã 2 even permutations and n! ã 2 odd permutations... (Starting at the bottom. . .. 2. . ..7(iii). a ñ 0 a . . all entries below a 11 are 0. ØÚÙ Ø£Ù × × Û Ü ØÚÙ Û ØÝÙ Û Ü 2. Calculate σ 1 . . 1 and n). 1 simple Û στ σ τ σ 1.n ôô ôô ôô ôô ôô x × 1 x1 x2 1 . S n has n permutations. n . 4. ô ô 3. a2. . (First show that the identity is true if σ is a transposition.. dxiσ 1 dxiσ 2 1 2 l k iσ k 3. (ii) Show that A Ø A A . . Then show it is true for an arbitrary permutation σ by writing σ as a product σ σ à àjà σ of transpositions and using formula (3. i 1. .. 1 . ô row above it. x × ôô ô ôô from each row subtract x times the for any numbers x . 1 x2 x2 2 . n 2.. (ii) σ 2. . x .. .1 n. . i . . This creates a new determinant whose ﬁrst column is the standard basis vector i j n 1 1 n n 1 2 n 1 n 1 2 1 det A Ø a1. Theorem 3.2 . 1 (i.. the trans- Û Þ ß Þ ßáà à à dx Þ ß Ø sign Ù σ Û dx dx à à à dx for any multi-index Ù i .2 1. a å σ σ i σ i σ i 3 Ù ç Û Ø Ù Û Ø Ø å æ ä × ØÚÙ Ü ä ä Ü 2. 3..4(i). . . Expand on the ﬁrst column and note that each column of the remaining determinant has a common factor.9. 4.) . .12. 2. 5.) 3. στ and τσ . 5. namely σ 1 . . .2 . (i) Write down the permutation matrices for all permutations in S .

7. ¢ ¢¢ . resp. 0. dx dy.17. spherical coordinates on R 2 . £¢ . £¢ . φ ü_ý dy dy þ . . resp. 3.44 3. Let φ ÿ x ú ÷ xxx øù (i) φ ü_ý y 3y 3y y þ. ¢ ø ù θn ¦ 1 ¢ r. φ ü dy . For each n § 1 deﬁne a map Pn ¨ ¡ 1: Rn ¨ 1 ¥ Rn ¨ 1 by 1 ¡ ö θ. etc. (iii) Let p be the ﬁrst column vector of the Jacobi matrix of Pn . øø ù ¢ r sin θn θn ¦ 1 (This is an example of a recursive deﬁnition. φ ü dy . If you know P1 . R3 . φ ü dy . as 3. r cos φ cos θ øù dy. Let φ in Exercise B. ö÷ θ r øù ú ü ö÷ r cos φ sin θ dz. . (ii) φ ü dy . ¢¢ ¢ θn ö ý cos θ þ P ö ¡ n n ¡ ÷ ¡ r θ1 . . φ ü dy . dx dz. PULLING BACK FORMS 3. . you can compute P2 .) (i) Show that P2 and P3 are the usual polar. Find 3. w ý þ ý n þ ú ý sin θ þ P . ö÷ x x .15. Let P3 be spherical coordinates in R3 . In this problem let us write a point in R n ö ¡ Let P1 be the function P1 r ýþ ú Pn ¨ ÷ ¡ ¡ r θ1 . 0. ¡ ¡ r 1 ÷ ø ú ÷ ù ¡ ¡ ¡ ¡ . and then P3 . ú ú ú A cos θn DPn . x x x x ö . (ii) Give an explicit formula for P4 . 0 þ . (ii) φ ü_ý y y y þ . dy dz. (iii) φ ü_ý dy dy þ . . v ú ý sin θ . w r cos θ . Show that Pn rp. Find x øùûú x x øù (i) φ ü dy . u is a column vector. . n u n . v is a row vector and w is a function given respectively by ú ÿA v u . . (iii) φ ü ý dy dy dy þ . Compute ψ ü_ý x dy dz y dz dx x1 2 3 ö÷ x 1 x1 x2 1 3 2 3 3 1 1 2 1 2 3 2 2 3 1 ¡ ¡ 2 3 1 2 1 2 2 1 2 £¢ 3 ¢ 2 3 ¤ 3 1 ¤ 1 2¤ 4 2 4 2 3 ¤ ¤ z dx dy . dx dy dz.16. (ii) Find the inverse of the matrix DP3 .14. where ψ is the map R2 ¥ þ R3 deﬁned 3. (iv) Show that the Jacobi matrix of Pn ¨ 1 is a n ¤ 1 © n ¤ 1 -matrix of the form DPn ¨ 1 ù ¢ ¢ ¢ where A is an n © n-matrix. φ r sin φ (i) Calculate P3 α for the following forms α : dx. φ ü dy . .18 (spherical coordinates in n dimensions).

n .

4.EXERCISES 45 (v) Show that det DPn 1 r cosn 1 θn det DPn for n 1.) (vi) Using the formula in part (v) calculate det DPn for n 1. (viii) Show that det DPn 0 for r 0. (Expand det DPn 1 with respect to the last row. 3. (vii) Find an explicit formula for det DPn for general n. . 2. using the formula in part (iv). and apply the result of part (iii).

.

we have (4. b easiest deﬁnition is that c should be the restriction of a smooth map c ε U deﬁned on a slightly larger open interval. We want to integrate over I . I a. (Strictly speaking we have not deﬁned what we mean by a smooth map c : a. The pullback c α is a 1-form on a. Let U be the punctured plane R2 . Differentiation and integration are related via a multivariable version of the fundamental theorem of calculus. To avoid problems with improper integrals we assume I to be closed and bounded.) Let α be a 1-form on U .1. In this chapter we investigate the case of 1-forms. b . 1 fi More explicitly.1) c α so i" 1 ∑ c n f i dci i" 1 ∑ c n fi dci dt. writing α in components. and can therefore be written as c α g dt (where t is the coordinate on R). α ∑n i" dxi .b ! c α b a g t dt.CHAPTER 4 Integration of 1-forms Like functions. Deﬁnition and elementary properties of the integral Let U be an open subset of Rn . b . 4. The ˜ : a ε. dt c α i" 1 a ∑ n b f i c t # dci t dt . known as Stokes’ theorem. The integral of α over c is now deﬁned by c α a. A parametrized curve in U is a smooth mapping c : I U from an interval I into U . b U . dt 4. E XAMPLE .1. forms can be integrated as well as differentiated.

We shall assume p to be a ¯.0 for all s 12 a 47 . s . x2 y2 Then c α dt (see Example 3. but it is traversed at a different rate. b * U can be reparametrized by substituting a new variable. where s ranges over another interval a ¯ ¯ .0 for a ¯ . so 2π ( c α )( 0 dt 2π . ¯ we have either p + s &3 0 for all s (in which case p is ¯. Let c : 0. Such ¯ . b onto a.8). A curve c : a. b satisfying p +. s .$ 0 % . y dx x dy α . 2π U be the usual parametrization of the circle. b t p s . b c / p: a has the same image as the original curve c. b one-to-one mapping from a a p is called a reparametrization. and let α be the angle form. The parametrized curve ¯ U ¯. b Since p + s 0 . sin t . ¯ . c t &' cos t.

Then p S 5 c S α 7&? dp ds ds ? L cM p α ? ¯R ¯ . Let cM p α ?AN E O O cα cα if p preserves the orientation. Finally let p 5 s 7H? 2π s 2 . π <C= R2 .1) we see that c S α ? F 5 c 5 t 7#7I^ c 4 5 t 7 dt. sin t 7 represents the unit circle in the plane. 0. QED b a Interpretation of the integral. 2π <G= R2 traverses the unit circle in the clockwise direction. Let p 5 s 7B? 2s. E XAMPLE . Accordingly. if p reverses the orientation. which can be thought of as a force ﬁeld acting responds a vector ﬁeld F ? ∑n i] 1 i i on the particle. 2π < and c : p. π < .b a O 5 pS g7 ¯ b ¯ a g 5 p 5 s 7#7 p 4 5 s 7 ds. 1 <I= R2 runs once counterclockwise through the unit circle. If we allowed s to range over . regarded as a map . T HEOREM . so ? L g dt and t L P p 5 s 7 .b a p ST5 c S α 7 . where we think of dx as an inﬁnitesimal vector tangent to the curve. INTEGRATION OF 1-FORMS increasing) or p 465 s 798 0 for all s (in which case p is decreasing). b p: .3. If p is increasing. In the orientation-reversing case. Recall from Section 2. a. 0. we say that it preserves the orientation of the curve (or that the curves c and c : p have the same orientation). we have c M p α ?YX O O cα ? O c α. b < be a reparametrization. the total work done by the force F on the particle during its trip along c is the integral L L L c α ? c F ^ dx ? b a F 5 c 5 t 7_7`^ c 4 5 t 7 dt. 4. Theorem B. a. Let α be a 1-form on U and c : .48 4.) Now let p 5 s 7D?FE s. This reparametrization reverses the orientation. so by the substitution formula. This is not considered a reparametrization of the original curve c. 0. b <*= U models a particle travelling through the region U . The curve c : .2. 0. Thus α represents the work done by the force ﬁeld along an inﬁnitesimal vector dx. 2π < and c : p : .5 that to a 1-form α ? ∑n i ] 1 Fi dx i corF e . 0. b <J= ¯ <=K. Then p maps . π < to . Then c : p : . represents the same circle.b a ? 5 c : p 7#S α ? ¯R ¯ .7. p S 5 g dt 7U?A5 p S g 7 dp ? Now let us write c S α 5 p S g 7V5 dp W ds 7 ds. a. we say that it reverses the orientation (or that c and c : p have opposite orientations). then 5 cos 2s. a L U a curve in U. P ROOF. . but traversed at 2 radians per second. Then p maps . On the other hand. 0. but at a variable rate. 4. 0. 0. 0. (It is important to restrict the domain of p to the interval . From (4.10(iii)) that L LQP L P cM p α ? ¯R ¯ . sin 2s 7 would traverse the circle twice. 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. c : p traverses the curve in the opposite direction to c. 2π <>= R2 deﬁned by c 5 t 7@?A5 cos t. It follows from the deﬁnition of the integral and from the naturality of pullbacks (Proposition 3. 1 < to . traversed at a constant rate (angular velocity) of 1 radian per second. 0. g 5 t 7 dt. the angular velocity is now E 1 radian per second. A curve c : . if p is decreasing. where the Z occurs if p 4\[ 0 and the E if p 4\8 0. Symbolically we write α ? F ^ dx. Then ¯. Integrals of 1-forms play an important role in physics and engineering. 2π < .

11 we have c i α g g c g t h#h we have c i α a dh. the work and the total work are nil if the force is perpendicular to the path. α is exact. Thus the fundamental theorem of calculus “explains” the law of conservation of energy. its potential energy decreases by the amount of work done by the ﬁeld. so f fQk f c i dg a dc i g.4. 4. etc.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. Hence n c α a g g c g b h#hb g g c g a h#h .1). This clariﬁes what it means for a ﬁeld to be conservative: it means that the work done is entirely converted into mechanical energy and that none is dissipated by friction into heat. Let c : c a. F a grad g. Let α a dg be an exact 1-form on an open subset U of Rn . The ﬁeld F is conservative if it can be written as the gradient of a function.b l ci α a b a dh a h g b hmb h g a h .2. i. Then f c α a g g c g b h_h`b g g c g a h#h . a P ROOF. In terms of forms this means that α a dg. QED The physical interpretation of this result is that when a particle moves in a conservative force ﬁeld. but only on the path itself and on the direction of travel. T HEOREM (fundamental theorem of calculus in R n ). Integration of exact 1-forms Integrating an exact 1-form α a dg is easy once the function g is known. b de U be a parametrized curve.2. 4. radiation. By Theorem 3. The work done by the force in the picture on the right is negative. . The function b g is called a potential for the ﬁeld and is interpreted as the potential energy of the particle. c c Theorem 4. Writing h g t hja c i g g t hja c α a a. INTEGRATION OF EXACT 1-FORMS 49 In particular.4. where we used the (ordinary) fundamental theorem of calculus. as in the picture on the left.e. formula (B.

50 4. (ii) u c α t 0 for all closed curves c. Fix a point x 0 in U . (ii) tv (iii): assume u c α t 0 for all closed curves c. Therefore u cx he α t)u c ˜ i ∂g r x sjt ∂ xi h lim 0 1 h ˜ c α w cx α t h lim 0 1 h cx t α α l h w cx t α h lim 0 1 h l α lim 0 1 h 1. But Theorem 4.) Then c is closed. We must deﬁne a function g such that α t dg. Write α t ∑n i 1 f i dx i . Theorem 4. c 1 r a1 sDt c2 r a2 s and c1 r b1 s0t c2 r b2 s . 2. for all c. (First traverse c 1 . (4. (i) tmv (ii): if α t dg and c is closed. then traverse c 2 backwards. b1 pq U and c2 : o a2 .e.2 l α . T HEOREM . i. 4. 1p`q U which joins x0 to x. A curve c : o a. INTEGRATION OF 1-FORMS It also yields a necessary and sufﬁcient criterion for a 1-form on U to be exact. b2 pq U be two curves with the same endpoints. Let α be a 1-form on an open subset U of R n . then u c α t g r c r b s#sIw g r c r a s_sxt 0 by the fundamental theorem of calculus. From the deﬁnition of partial differentiation. For each point x in U choose a curve cx : o 0. (iii) u c α depends only on the endpoints of c for every curve c in U. (i) α is exact.3 implies u c α t)u c1 α w~u c2 α . so u c α t 0. Deﬁne g r x sCt) cx α. P ROOF.4. (iii) tmv (i): assume that. u c α depends only on the endpoints of c. ∂g r x sjt ∂ xi lim g r x hei s>w g r x s h t h 0 h lim 1 0h cx α h ei w~ cx α . Then c α . We need to show that u c1 α tyu c2 α . Deﬁne a new curve c by c r t szt|{ c1 r t s c2 r 2 w t s for 0 for 1 } } t t } } 1.5. After reparametrizing c 1 and c2 we may assume that a1 t a2 t 0 and b1 t b2 t 1. b pq U is called closed if c r a sGt c r b s . ˜ composed of two pieces: for 0 } t } 1 travel from x0 Now consider a curve c to x along the curve cx and then for 1 } t } 2 travel from x to x hei along the ˜ has the same endpoints as straight line given by l r t st x )r t w 1 s hei . Then the following statements are equivalent. so u c1 α t)u c2 α . Let c1 : o a1 .2) . and hence cx hei . We assert that dg is well-deﬁned and equal to α . We must show that ∂ g ∂ xi t f i .

the theorem also enables us to detect closed 1-forms that are not exact. This shows that l α j 1 ∑ n f j x t 1 hei dl j j 1 ∑ n f j x t 1 hei l j t dt j 1 ∑ n f j x t 1 hei δi. QED This proves that g is smooth and that dg α. it tells us that once we know a 1-form α to be exact we can ﬁnd an “antiderivative” g x by integrating α along an arbitrary path running from a ﬁxed point x0 to x. THE GLOBAL ANGLE FUNCTION AND THE WINDING NUMBER 51 Let δi. The angle form α . Then δi. can be used in many different ways.3) together we ﬁnd ∂g x ∂ xi h lim 1 1 0h 2 1 h f i x t 1 hei dt h lim 1 0 0 f i x shei ds 0 h lim f i x shei ds 0 1 0 f i x ds fi x . (See Exercises 4. E XAMPLE . jh dt h f i x t 1 hei dt. This theorem. 4. j h.) On the other hand.3) Taking equations (4. (4. which is deﬁned by δ i.5 for an application.4. j be the Kronecker delta. and hence l j t 1 and δi. and its proof.3.6. j t 1 h.2) and (4. For example.3–4. j 0 if i j.i we can write l j t x j δi.

) Now let θ : U ¤¦¥ 0. α will denote the angle form. 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.e. This equation is not valid on all of U (it cannot be because we saw in Example 4. Then ξ cos θ and η sin θ . the unit vector pointing in the direction of x. its integral around the circle is 2π 0. on the complement of the . Indeed. which are always exact! (See Exercise 2.10 is closed. but only where θ is differentiable. α y dx x dy . Mark the contrast with closed 1-forms on Rn . so by equation (4. Then α is a closed 1-form and ξ and η are smooth functions on U . but not exact.4) (You will be asked to check this formula in Exercise 4.6 that α is not exact).4) α cos θ d sin θ sin θ d cos θ cos 2 θ dθ sin θ 2 dθ dθ . x2 y2 and ξ and η will denote the functions x ξ ¡ .) This phenomenon underlines the importance of being careful about the domain of definition of a form.3. 2 π . 1 R2 0 of Example 2.6. i. Throughout this section U will be the punctured plane R2 0 . These functions satisfy α ξ dη ηdξ . 4. x2 y2 η ¡ y x2 y2 .6. ξ and η are just the components of x ¢\£ x £ . (4. In fact. chosen to lie in the interval ¥ 0. 2π be the angle between a point and the positive x-axis.

and the fact that α ª dθ almost everywhere suggests how: by integrating α along c! For simplicity assume that a ª 0 and b ª 1.6. b ¨ © U we can deﬁne a continuous angle function. (The precise meaning of this assertion will become clear in Exercise 4.52 4. INTEGRATION OF 1-FORMS positive x-axis. Hence the nonexactness of α is closely related to the impossibility of deﬁning a global differentiable angle function on U . along a curve c : § a. Start by ﬁxing any ϑ 0 such that cos ϑ0 ª ξ « c « 0 ¬#¬ and sin ϑ0 ª η « c « 0 ¬#¬ and then deﬁne ϑ « t ¬jª ϑ0 .) However.

To see that ϑ « 0 ¬*ª ϑ 0 . so it follows from formula (4. ϑ is smooth as well. T HEOREM . cos ϑ « t ¬jª ξ « c « t ¬#¬ and sin ϑ « t ¬zª η « c « t ¬#¬ . 4.®°¯ 0. Therefore c ² η. In other words f ª c ² ξ and g c ² α ª f dg ¶ g d f . The following result says that ϑ « t ¬ measures the angle between c « t ¬ and the positive x-axis (up to an integer multiple of 2π ) and that the function ϑ : § 0. Furthermore.4) that ϑ « t ¬jª t ϑ0 ® 0 · f « s ¬ g ¸¹« s ¬>¶ g « s ¬ f ¸6« s ¬»º ds. ϑ is differentiable and ϑ¸ gf¸ . 1 ¨ © R is smooth. Let f « t ¬ and g « t ¬ be the x. Its length is equal to « f ¶ cos ϑ º 2 · g ¶ sin ϑ ¬ 2 ª f2 g2 ¶ 2 « f cos ϑ g sin ϑ ¬ ª ª ª 2 ¶ 2 « f cos ϑ g sin ϑ ¬ . P ROOF. (4.5) Since the right-hand side is smooth.7. g2 ª 1 g « gg ¸ . the derivative of u is f ¸ cos ϑ ¶ f ϑ ¸ sin ϑ g ¸ sin ϑ gϑ ¸ cos ϑ « f¸ ¶ g f¸ 2 f gg ¸ ¬ cos ϑ « g¸ ¶ f 2 g¸ f g f ¸ ¬ sin ϑ by formula (4.5) since f 2 f2f¸ f « f f¸ f gg ¸ ¬ cos ϑ gg ¸ ¬ cos ϑ g2 g ¸ f g f ¸ ¬ sin ϑ f f ¸ ¬ sin ϑ. To prove the other assertions we rescale the curve c « t ¬ to a new curve c « t ¬#³\´ c « t ¬µ´ moving on the unit circle. For t ª 0 we have u « 0 ¬jª u¸ ª ª¿« ª¿« ª f cos ϑ g sin ϑ is a constant equal ª f « 0 ¬ cos ϑ0 g « 0 ¬ sin ϑ0 cos 2 ϑ0 sin 2 ϑ0 1. In this sense ϑ is a “differentiable choice of angle” along the curve c.2). Hence we need to show that the function u to 1. t ± c ² α. plug t ª 0 into the deﬁnition of ϑ. formula (B. Then f « t ¬jª ξ « c « t ¬_¬ and g « t ¬zª η « c « t ¬#¬ and f « t¬ 2 ª g « t¬ 2 ª 1 for all t. To prove that cos ϑ « t ¬@ª f « t ¬ and sin ϑ « t ¬jª g « t ¬ for all t it is enough to show that the difference vector ¼ f « t¬ g « t ¬¾½ ¶ ¼ cos ϑ « t ¬ sin ϑ « t ¬_½ cos 2 ϑ sin 2 ϑ has length 0. f g¸ ª ¶ By the fundamental theorem of calculus.and y-components of this new curve. The function ϑ is smooth and satisﬁes ϑ « 0 ¬jª ϑ0 .

Hence u is a constant QED c Ã t Å#Æ\Ç c Ã t ÅTÇ as a dial that points It is useful to think of the vector Ã f Ã t Å . so u Ã t Å Á 1 for all t. it moves around the meter. Consider the curve c : Ó 0. By Example 4. Á sin ϑ Ã 1 Å .6) cos t. then ϑ Ã 1 ÅzÈ ϑ Ã 0 Å c Ã 1 Å implies Á Í Á 2π k. Restate Theorem 4. where a and b are positive constants. g Ã t Å#Å in the same direction as the vector c Ã t Å . 1 ÊÌË where k is an integer. π Ô 2 Õ\Ö R2 deﬁned by c × t ØGÙ)× a cos t. 1. b sin t Ø T . Á Ã (4.1. cos ϑ Ã 0 Å . η Ã c Ã 0 Å#Å#Î ξ Ã c Ã 1 Å#Å . If c : É 0. (i) Sketch the curve c for a Ù 2 and b Ù (ii) Find Û c α (for arbitrary a and b). Explain why the result is plausible on physical grounds. E XAMPLE . The difference ϑ Ã 1 ÅIÈ ϑ Ã 0 Å measures the total angle swept out by the dial. the dial starts at the angle ϑ Ã 0 Å Á ϑ 0 . so u ÂÄÃ t Å Á 0 for all t.9. c Ã 0 Å Í Á Í Á U is a closed curve.8. T Á 0 0 As t increases from 0 to 1. sin t Å T 4. Exercises 4. It measures how many times the curve loops around the origin. sin ϑ Ã 0 Å»Î ξ Ã c Ã 0 Å#Å . .5 in terms of force ﬁelds. C OROLLARY. potentials and energy. Let α Ù xy dx Ú x 2 y dy. sin ϑ Ã 1 Å Î .EXERCISES 53 Now f 2 À g2 Á 1 implies f f Â À gg Â function. so ϑ Ã 0 Å and ϑ Ã 1 Å differ QED The integer k Á Ã 2π Å_Ï 1 Ð c α is called the winding number of the closed curve c about the origin.1. winding number of a closed curve about origin Á 1 2π Ñ c α.7. In other words cos ϑ Ã 0 Å Á cos ϑ Ã 1 Å and sin ϑ Ã 0 Å by an integer multiple of 2π .2. 4. the winding number of the circle c Ã t Å (0 Ò t Ò 2π ) is equal to 1. 4. η Ã c Ã 1 Å#Å#Î Á Í cos ϑ Ã 1 Å . By Theorem 4. and ends up at the ﬁnal angle ϑ Ã 1 Å . P ROOF. Á 0.

) 4.3. 4. and winding numbers of the following curves c then explain the answer by appealing to geometric intuition. Let c : ø 0.3 and 4. where a þ 0 and b þ 0. (i) Verify equation (4.) (i) Determine for which values of a the function g ã x ä9Ü¿å cx α is well-deﬁned and compute it. (ii) Determine for which values of a the function g ã x ä9Ü¿å cx α is well-deﬁned and compute it. T 1 ˜ ã t äÜ φ ã c ã t äöä . t äÜ ρ ã t ä c ã t ä .4). (ii) Let U Ü R2 ßéà 0 á . y äîÜ x ïñð x2 ò y2 and sin θ ã x. sin t ä T . 4. 0 ì in such a way that cx ã 0 äÜ x. t ämÜüã cos3 t.8. Prove that there does not exist a smooth function θ : U í R satisfying cos θ ã x. (i) (ii) (iii) (iv) cã cã cã cã t äÜüã a cos t. ß y ä . .5. Parametrize c x by travelling from inﬁnity inward to x.) 4. y ä>Üüã y. where φ ã x. where a is a real constant.9. t äÜüã cos t ß 2. (i) (ii) (iii) (iv) t äÜ c ã 1 ß t ä . b sin t ä T . 4. (You can do this by using an inﬁnite time interval ã ßUë . If not.6. Let α æç 1 ã Rn ßà 0 áêä be as in Exercise 4. (Argue by contradiction.4. 4. INTEGRATION OF 1-FORMS n ßJà 0 á . Let α æèç 1 ã Rn ßéà 0 áêä be the 1-form of Exercise 4. ë ä is a function satisfying ρ ã 0 äÜ t äÜÝ c ã t äúÝ_û 1 c ã t ä . where φ ã x. y äîÜ y ïñð x2 ò y2 for all ã x . y äóæ U .10. 2π ì°í R2 given by c ã t ämÜã cos kt. 2 2 ò x y ˜ã c ˜ã c ˜ã c ˜ã c ρ ã 1ä . sin3 t ä T . where ρ : ø 0. t äÜ)ÿöã a cos t ò b ä cos t ò ã b ß a äöõ 2. sketch the curve (the use of software is allowed) and obtain the answer geometrically. by letting α Üôã ß y dx ò x dy äöõ÷ã x 2 ò y 2 ä and showing that α Ü dθ if θ was such a function. There is one value of a which is not covered by Exercises 4. 4. t äÜ φ ã c ã t äöä . For this value of a ﬁnd a smooth function g on R n ßJà 0 á such that dg Ü α . (i) Show that α is closed for any value of a. 2π ìQí R2 ßýà 0 á set up the integral deﬁning the winding number about the origin.3. Consider the 1-form α Ü)Ý x Ý a ∑n i Þ 1 x i dx i on R For every x Ü â 0 let cx be the line segment starting at the origin and ending at x.6).3. where 0 ¡ b ¡ a. 1 ìQíùã 0. Determine the ˜ : ø 0. Now let c x be the halﬂine pointing from x radially outward to inﬁnity. 1 ìmí R2 ßà 0 á by using the formula.54 4. 1 ìQí R2 ßBà 0 á be a closed curve with winding number k. Calculate directly from the deﬁnition the winding number about the origin of the curve c : ø 0. (iii) Show how to recover from this computation the potential energy for Newton’s gravitational force.4. Evaluate the integral if you can (but don’t give up too soon). (See Exercise B. y ä>Ü (v) c ã x. ã a cos t ò b ä sin t T .4. 4. x ä T . For each of the following closed curves c : ø 0. sin kt ä T . How would you deﬁne the winding number of c around x 0 ? Try to formulate two different deﬁnitions: a “geometric” deﬁnition and a deﬁnition in terms of an integral over c of a certain 1-form analogous to formula (4. (iii) For the values of a you found in part (ii) check that dg Ü α . (ii) For the values of a you found in part (i) check that dg Ü α . Let x0 be a point in R2 and c a closed curve which does not pass through x 0 .7.

EXERCISES 55 R2 .

0 by 4.11. circles. The differential form ¤ F1 e1 F2 e2 : U © R2 be a smooth β ¤ F1 dF2 F2 dF1 2 F2 F1 2 is well-deﬁned at all points x of U where F x ¤ £ 0. β is closed. c Prove the following assertions. 2π ¨ © ¤ a b cos t a cos a b t.13. index F. Let c be a parametrized circle contained in U . c is the winding number of the curve F " c about the origin. If you get stuck. Set up the integral deﬁning the winding number of c around the origin and evaluate it. ﬁnd the answer geometrically. . a b sin t a sin a b t a a T c t . where α is the angle form y dx x dy x2 y2 ! . β ¤ F α . c is an integer. (i) (ii) (iii) (iv) ¤ 1 2π c β. traversed once in the counterclockwise direction. Let U be an open subset of R2 and let F vector ﬁeld. (i) Find the indices of the following vector ﬁelds around the indicated 4. (ii) For what values of a and b is the curve closed? (iii) Assume c is closed. Let b ¢ 0 and a ¤ £ 0 be constants with ¥ a ¥¦¤ £ b. (i) Sketch the curve c for a ¤ b 3. Deﬁne a planar curve c : § 0. Assume that F x ¤ £ 0 for all x c. The index of F relative to c is index F. index F.12. 4.

INTEGRATION OF 1-FORMS (ii) Draw diagrams of three vector ﬁelds in the plane with respective indices 0. .56 4. 2 and 4 around suitable circles.

b ¯k. Integration of forms over chains In this chapter we generalize the theory of Chapter 4 to higher dimensions.Rk . Let ¯1& ' $ a ¯ 2 & '*)+)+)9' $ a ¯k& ¯ #8$ a ¯1. Let U be an open subset of Rn and let α be a k-form on U . The k-dimensional analogue of a parametrized path is a smooth map c : R 2 U .e. A reparametrization is a map p : R the following conditions: p is bijective (i. ai / ti / bi for 1 / i/ k0 . x 0 consisting of a single point x # c 3 0 4 in U . In the same way that 1-forms are integrated over parametrized curves. In fact any compact subset of R k will do.1. Then det Dp 3 s 4. cylinders. integrals of k-forms are almost wholly unaffected by a change of variables. one-to-one and onto) and the k ' k¯ . The integral of α over c is deﬁned as 6 c α# 6 R c5 α # 6 bk ak )+)+) 6 b2 a2 6 b1 a1 g 3 t 4 dt1 dt2 )+)+) dtk . b ¯2. such as skew blocks. The integral of a 0-form (function) f over c is by deﬁnition the value of f at x.CHAPTER 5 Integration and Stokes’ theorem 5. where ai 1 bi . b2 &('*)+)+)¦' $ ak . we think of the map c as a parametrization of the subset c 3 R 4 of U : each choice of a point t in R gives rise to a point c 3 t 4 in c 3 R 4 . This is a set of the form R #%$ a1 .) The case k # 0 is also worth examining. b R ¯ 2 R satisfying be a second rectangular block. A zero-dimensional “block” R in R0 #7. (The deﬁnition makes sense if we replace the rectangular block R by more general shapes in R k . For k # 1 this reproduces the deﬁnition given in Chapter 4. The simplest k-dimensional analogue of an interval is a rectangular block in Rk whose edges are parallel to the coordinate axes. Although the image c 3 R 4 may look very different from the block R. respectively reverses the orientation of c. k-dimensional balls. etc.R ¯ . t . so either matrix Dp 3 s4 is invertible for all s . b1 &(' $ a2 . As in the one-dimensional case. In these cases we say that the reparametrizion preserves. 6 c f # f 3 x4 . k-forms can be integrated over k-dimensional parametrized regions. 0 0 is just the point 0. bk & #.R det Dp 3 s 4=< 0 for all s or det Dp 3 s 4 1 0 for all s. We can therefore think of a map c : R 2 U as a collection . 57 . The pullback c 5 α is a k-form on R and therefore looks like g 3 t 4 dt 1 dt2 )+)+) dtk for some function g : R 2 R.# : 0 for all s .

.58 5. 1 S Let R be any other block. t2 H dt2 dt1 . T HEOREM . . c α A g t H dt1 dt2 K+K+K dtk .10(iii). Almost verbatim the same proof as for k A 1 (Theorem 4.”) Then p is one-to-one and onto and Dp sHfA A. as in the formula D D ? b1 a1 ? b2 a2 f t1 . the unit cube in R k is the F F rectangular block R 0. Let p : R R ? α α ACB E D c c@ p cα P ROOF. 5. It follows from the deﬁnition of the integral and from the naturality of pullbacks. k bi . Then the left-hand side of formula (5.1) (This follows for instance from the substitution formula. . F (5. p s H . .13. we have also learned that f t 1 . . if p reverses the orientation. Then ¯ > U a smooth map. 0 b2 E a2 . 5. 1 S > R by p s H. given by a i \ As O a. . Then p I g H det Dp ds1 ds2 K+K+K dsk F F F F F ? F F F On the other hand. How F follows.1)? F The explanation is as f t1 . R EMARK . E XAMPLE . ` . ti \ 0 0 . bST> U can be reparametrized to a curve c G p : R 0. . αA ¯F R c G p HJI α A ¯ R pI cI α H .. where b1 E a1 0 A_ A ^` . D DR Theorem B..1) is the integral of α over F . 1 SU> U by means of the reparametrization p s HVA b E a H s O a.. t2 H dt1 dt2 . . t2 H dt1 dt2 . t2 H dt1 dt2 A ? b2 a2 ? b1 a1 F f t1 .e bdc F `a ak (“Squeeze the unit cube until it has the same edgelengths as R and then move it to the position of R.. so det Dp s HLA F F det A A vol R P 0 for all s.2. we have c @ p α ANM F c α . 1 S in the real line. INTEGRATION AND STOKES’ THEOREM 5.A bdc c c and `a 0 bk E ak e a1 cc a2 a A7^` . that ? ? ? Now let us write c I α A p I c I α HLA by Theorem 3. .. so g dt1 dt2 K+K+K dtk and t A p I g dt1 dt2 K+K+K dtk HLA c@ p D if p preserves the orientation. A useful fact you learned in calculus is that one may interchange the order of integration in a multiple integral. for 1 \ i\ k] . The unit interval is the interval R 0. t2 H dt2 dt1 A E f t1 . 1 S k AXW t Y Rk Z ti Y[R 0.7. . Hence c @ p α A c α for any k-form α on U . Deﬁne p : R 0. Theorem B. .3. Any curve c : R a. so by the substitution formula.1..) On the other hand. Similarly. Let α A can this be squared with formula (5.7. Let α be a k-form on U and c : R > be a reparametrization. ` . Proposition 3. QED c@ p ¯ R αA ? g p sHJH det Dp s H ds 1 ds2 K+K+K dsk . so p is an orientation-preserving reparametrization. 0 . where the O occurs if det Dp P 0 and the D D E if det Dp Q 0.3). .

b1 h i g a2 .2 that an integral over any rectangular block can be written as an integral over the unit cube. from now on we shall usually take R to be the unit cube. .) It is often necessary to integrate over regions that are made up of several pieces. . the parametrization of the rectangle given by c l t 1 . We see from Example 5. but over c q p.1 says that r c s p α n ptr c α . is extended to chains in such a way as to be linear. . which represents a vector in Rn . Integration. v c p f n i 1 ∑ p ai v xi f n i 1 ∑ a i f l xi m . A k-chain in U is a formal linear combination of k-cubes. p Likewise. so that the image can have self-intersections. b1 h j g a1 . Since p reverses the orientation. Its boundary is by deﬁnition the c 1 d c 0 d c . For any k-form α we then deﬁne v c αn (In the language of linear algebra. with an electric charge ai placed at the point xi . a2 . b2 hkj R2 . a k-chain ∑i 1 ai ci can be pictured as a charge distribution. c p are k-cubes. 5. b1 h i g a2 . . . . which represents a distribution of point charges.2. Theorem 5. the k-chains form an abstract vector space with a basis consisting of the k-cubes.1 y k f l t1 . . c n ∑i 1 ai xi . (You must carefully distinguish between the formal p linear combination ∑i 1 ai xi .) Recall that a 0-cube is nothing but a singleton x consisting of a single point p x in U .1 y k f l t1 . s1 m . t2 . where a1 . A smooth map c : g 0. s2 mLn l s2 . the word singular meaning that the map c is not assumed to be one-to-one. Thus a 0-chain is a formal linear combination of points. For this reason. . in other words r c α n r c s p lup α m . . c n a1 c1 } a2 c2 }~z+z+z|} a p c p . A good way to think of c is as a collection of p point charges. with an electric charge ai spread along the k-dimensional “patch” c i . . which is exactly formula (5. . 1 h k j U is called a k-cube in U (or sometimes a singular k-cube.5. . t2 m .1). U . which is a priori only deﬁned on cubes. t2 mon l t1 . tk m dt1 dt2 z+z+z dtk n v w 0. p and the linear combination of vectors ∑i 1 ai xi . 1 h j 0-chain deﬁned by ∂c n c l 1 m p[ c l 0 m . b2 h is the reparametrization p l s 1 . . . Analogously we have c : g a1 . b2 h(i g a1 . where vxw 0. t2 . a p are real coefﬁcients and c 1 . c2 . The boundary of a chain Consider a curve (“1-cube”) c : g 0. tk m dti dt1 dt2 z+z+z|d { ti z+z+z dtk for any i.) The integral of a function f over the 0-chain is by deﬁnition i 1 ∑ ai p v ci α. .2. THE BOUNDARY OF A CHAIN 59 p : g a2 . The right-hand side is the integral of p α not over c. .

(There are also many differences. In short. c ¯3 c ¯4 . t k 1 . You should check that for k 0 and k 1 this deﬁnition is consistent with the one. . INTEGRATION AND STOKES’ THEOREM The boundary of a 2-cube c : 0. 1 k U has 2k faces of dimension k 1.0 t f c t1 . 1 k 2 and let ρ and σ be 0 or 1. t . . 1 t . ∂2 t 0. Let t t1 . t k 1 1 . . 0. 1 and c This would work equally well. . . By linearity of ∂ it sufﬁces to prove this for k-cubes c : 0. c2 t c 1. . t2 . . t2 . t2 . . t2 . k put ci. ti c t1 . . .and two-dimensional cases considered above.1 t f ci. .1 f i 1 ρ 0. . Then for 1 i j k 1 .60 5. such as the fact that d raises the degree of a form by 1. . .) Now deﬁne ∂c i 1 ∑ u k 1 i ci. Let t1 . tk 1 [ 0. ti 1 . u 1 For an arbitrary k-chain c ∑i ai ci we put ∂c ∑ i ai ∂ci . 1 2 U consists of four pieces corresponding to the edges of the unit square: c 1 t c t.ρ . 1 k 1 and for i 1. with c ¯3 t (Alternatively we could deﬁne ∂c c 1 c2 c ¯4 t o c 0. . . . 1. . . t i . P ROOF.0 ci. the most important of which is the following. . .) A k-cube c : 0. (“Insert 0.) 5. . which corresponds to the following picture: c c 1 t. . . 1 k U . 1 and c4 t c 0. . whereas ∂ lowers the dimension of a chain by 1. P ROPOSITION . c3 t c t. There are a number of curious similarities between the boundary operator ∂ and the exterior derivative d. tk 2 0. but is technically less convenient.4. t . t i . Then ∂ is a linear map from k-chains to k 1-chains. . 2. which are described as follows. 0 . resp. . 1 in the i-th slot”. ∂ ∂c L 0 for every k-chain c in U.1 ∑ ∑ k i ρ ci. The picture below suggests that we should deﬁne ∂c c 1 c2 c3 c4 . . .

tk ¡ is independent of ti for some i. t j ¢ 1 .ρ £ ∑ ∑ § 1 ¡ i ¤ ρ∂ci. .σ ¨ 0. because in the vector t 1 .1 ∑ ∑ § 1¡ i¤ j¤ ρ¤ and then 1.σ ¨ 0. . More generally we have the following. 5. .ρ ¡ j . . Then c t1 .1 k k¢ 1 i¤ j¤ ρ¤ σ ci. . . . 5. . . . . . . . . . . In particular. tk ¢ 2 ¡ £ c t1 . ρ. . .1 The double sum over i and j can be rearranged in a sum over i ¦ i © j to give ∂ ∂c ¡L£ 1 ª i ª j ª k ¢ 1 ρ . . ρ. Suppose c is constant as a function of ti . . ti . t2 . t j ¢ 1. . L EMMA . . σ . . . . . tk ¢ 2 ¡ c j ¤ 1. t2 . . . £ ∑ ∑ ∑ § 1¡ i ¨ 1 j ¨ 1 ρ .2) cancel out. ν £ ρ to get 1 ª i ª j ª k ¢ 1 ρ .ρ ¡ j .ν ¨ 0.1 j and a sum over ∑ ∑ § 1¡ i¤ j¤ ρ¤ σ ci. P ROOF. . t k ¡f£ g f t1 . Let α be a k-form and c a degenerate k-chain. tk ¢ 2 ¡f£ ci .σ t1 . . σ ∑ ∑ § 1 ª s « r ª k µ . . . .σ t1 . . The work done by a force ﬁeld on a motionless particle is 0. .ρ s . Then ¬ c α £ 0. . Therefore the k § 2-chain ∂ ∂c ¡ is given by k k ∂ ∂c ¡o£ ∂ ∑ ∑ § 1 ¡ i ¤ ρci. tk ¢ 2 ¡ . By linearity we may assume that c is a degenerate cube.µ ¡ c j¤ Thus the two terms on the right in (5.ν s . . ti . t2 .5. . . . tk ¡J® . ti ¢ 1 .σ £ c j ¤ 1.ρ ¡ j.σ ¥ 1 ª j « i ª k ρ . ρ. ti ¢ 1 . . t2 .σ ¨ 0.σ .1 1.ρ ¡ j . . .σ ¡ i. . .ρ ¡ j. tk ¢ 2 ¡L£ c j £ c t1 . .2) In the ﬁrst term on the right in (5. 0.σ £ £ £ 1 ª i ª j ª k ¢ 1 ρ .ρ i ¨ 1 ρ ¨ 0. . . . .σ ¡ i . On the ¤ 1. tk ¡f£ c t1 .3. .ρ ¡ j. . a degenerate 1-cube is a constant curve. ti . t j .ρ for 1 ¦ i ¦ j ¦ k § 1. .ν ¨ 0.σ ¡ i. .ρ ¡ j . . .µ ¡ s¤ r¤ ν ¤ µ § 1¡ cr. µ £ σ .ρ t1 . . ti . t2 . . ti ¢ 1 . (5.σ ¡ i .5.2) we substitute c i. .ν . t2 . tk ¢ 2 ¡ the entry t j occupies the j ¥ 1st slot! We conclude that c i. tk ¢ 2 ¡ . . .1 ∑ ∑ § 1 ¡ s ¤ r ¢ 1 ¤ ν ¤ µ cr.σ £ r £ j ¥ 1. . .ρ ∑ ∑ § 1¡ i¤ j¤ ρ¤ σ ci. other hand. . A k-chain c is degenerate if it is a linear combination of degenerate cubes. . . . . .3. t2 . . t j .σ ¨ 0. . σ . . t j ¢ 1. ti . ti . . σ . . CYCLES AND BOUNDARIES 61 we have ci. . .σ ¨ 0. . . ti ¢ 1 . .1 ∑ ∑ § 1¡ i¤ j¤ ρ¤ c j¤ σ ci. .σ . Cycles and boundaries QED A k-cube c is degenerate if c t 1 .1 1 ª s « r ª k µ . . . s £ i. ρ. t j .1 i ¨ 1 ρ ¨ 0. .ρ t1 . .

1 ° k ² 1 ± U are given respectively by ˆi . P ROOF. 1° k ² 1 and g : ¯ 0. that c µ ∂b ½ c ¾ where c ¾ is a constant 1-chain. ³ t1 . 1 ° k ± ¯ 0. L EMMA . .8. Every boundary is a cycle. and therefore so is ∂c. Then by Lemma 5. 5.) The 2-cube b is deﬁned by “shrinking c to a point”. . . .7. This is indeed true in the sense deﬁned above. Lemma 5. 0 ´oµ . . Then c ³ t1 ´ . tk ´fµ So degenerate chains are irrelevant where integration is concerned. ti .5 ∂c µ ∂ ³ ∂b ´ ½ ∂c ¾ µ ∂c ¾ . Then ci.0 µ ci. So ∂c is a combination of degenerate k ¼ 1-cubes and hence is degenerate. P ROOF. . c2 c c1 5. . E XAMPLE .1. . A k-chain c is closed. . QED 5. b ³ t 1 .9. Consider the unit circle in the plane c ³ t ´. . 1 ° k ² 1 and hence equal to 0. .1 ´ . where c ¾ is the constant curve located at the origin. 0. where we used Proposition 5. si ² 1 . We conclude ¹ c α µ ¹+º 0. .1 ³ t ´ are independent of ti and for j Â i they are independent of t i ² 1 . . . or a cycle. then c1 ½ c2 is a 1-cycle. b ³ 0. t2 ´Äµ ³ 1 ¼ t2 ´ c ³ t1 ´ for ³ t1 . . . In the same way that a closed form is not necessarily exact. 0 ´ . Therefore c µ ∂b ½ c ¾ . . By linearity it sufﬁces to consider the case of a degenerate k-cube c. For j Á i the cubes c j. QED f ³ t1 . . t2 ´Lµ ³ 1 ¼ t2 . tk ´ .0 ¼ c j. See Example 5. INTEGRATION AND STOKES’ THEOREM where f : ¯ 0. . . . . a boundary plus a degenerate 1-cube. C OROLLARY.6. QED 5. t2 ´Lµ b ³ 1. .4. see Exercise 5. Let t µ ³ t1 . . 1 ´Åµ ³ 0.62 5. .µ ³ cos 2π t. and so c · α µ f ·¸³ g · α ´µ 0. Suppose c is constant as a function of t i . the closed curve c is a 1-cycle. This is a closed 1-cube. . . tk ² 1 ´ . 0 ´ . If c 1 and c2 are curves arranged head to tail as in the picture below. .0 ³ t ´ and c j. b ³ t1 . if ∂c is a degenerate k ¼ 1-chain. b ³ t1 . (It is actually not possible to ﬁnd a b such that c µ ∂b. so ∂c µ ³J¼ ∑ À j¿ i 1 ´ j ³ c j.2. . E XAMPLE .7 says that ∂c ¾ is degenerate. This motivates the following deﬁnition.11. sk ² 1 ´fµ c ³ s1 . . . t2 . it may happen that a 1-cycle is not a boundary.1 » k c · α µ 0. . The boundary of a degenerate k-chain is a degenerate k ¼ 1-chain. sin 2π t ´ with 0 Ã t Ã 1. t2 ´ in the unit square. si ¶ 1. so that ∂b µ c ¼ c ¾ . 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. Now g · α is a k-form on ¯ 0. Suppose c µ ∂b ½ c ¾ with c ¾ degenerate. tk ² 1 ´ . Likewise. t g ³ s1 . A k-chain c is a boundary if c µ ∂b ½ c ¾ for some k ½ 1-chain b and some degenerate k-chain c ¾ .

. Therefore Æ c dα Ç iÚ 1 ∑ k Æ Õ 0. it can be written as c× α Ç iÚ 1 Ü ti Û+Û+Û dtk ∑ gi dt1 dt2 Û+Û+Û|d k for certain functions g 1 . . This is the form in which the fundamental theorem of calculus generalizes to higher dimensions.3) and subsequently applying the fundamental theorem of calculus in one variable.1 Ö k ∂ gi dt1 dt2 Û+Û+Û dtk . . P ROOF. . t k É dt1 dt2 Û+Û+Û d Û Û+Û dtk Ü ti + dt1 dt2 Û+Û+Û|d Ü ti Û+Û+Û dtk and .11 we have Æ c dα Ç ÆxÕ 0. . 0.e. ∂ti Changing the order of integration (cf. 1. t k É 1 . although it would perhaps be better to call it the “fundamental theorem of multivariable calculus”. . . Stokes’ theorem In the language of chains and boundaries we can rewrite the fundamental theorem of calculus. This generalization is perhaps the neatest relationship between the exterior derivative and the boundary operator. .1 Ö k 1 à Ý gi È t1 . . Since c × α is a k Ê 1-form on Ø 0. gk deﬁned on Ø 0. .1 Ö k dc × α . . ti Ò 1 . . i. . . T HEOREM (Stokes’ theorem). . 1 Ù k . . ti Ò 1 . tk É Þ dt1 dt2 Û+Û+Û|d Ü ti Û+Û+Û dtk . formula (B. gives ÆxÕ ∂ gi dt1 dt2 Û+Û+Û dtk Ç k 0. . . as follows: Æ c dg Ç g È c È 1 ÉJÉkÊ g È c È 0 ÉdÉËÇ ÆÍÌ c Î 1 ÏÑÐ gÊ Æ¸Ì c Î 0 ÏÑÐ gÇ Æ¸Ì Ì c Î 1 ÏÑÐJÒ c Î 0 ÏÑÐ gÇ Æ ∂c g. Remark 5. ti ß 1 . . 0. .1 Ö ∂ t i ÆxÕ Ç Æ Õ ∂ gi dti dt1 dt2 Û+Û+Û|d Ü ti Û+Û+Û dtk 0. . . Ó c dg ÇÔÓ ∂c g. . g2 . 1. t i ß 1 .10. 5. .4. . ti Ò 1 . . . 1 Ù k . It contains as special cases the classical integration formulas of vector calculus (Green. ti Ò gi È t1 .4. Then Æ c dα Ç Æ ∂c α. Gauß and Stokes) and for that reason has Stokes’ name attached to it.1). . . The forms gi È t1 . .5. . t i ß 1 . .1 Ö d Ý gi dt1 dt2 Û+Û+Û|d Ü ti Û+Û+Û dtk Þ Ç k ∑ ÈuÊ 1 É i ß iÚ 1 k 1 ÆxÕ 0. By the deﬁnition of the integral and by Theorem 3. . tk É Ê gi È t1 . .1 Ö k c × dα Ç Æ¦Õ 0. . . STOKES’ THEOREM 63 5. Let α be a k Ê 1-form on an open subset U of Rn and let c be a k-chain in U.1 Ö ∂ t i k 0.4. ti ß 1. Theorem 4.

(i) Sketch the image of c. t2 ¡ ¢¤ ú £ ¡ 2 t2 1 . the sense that there exist no 2-chain b and no degenerate 1-chain c ò both contained in U such that c ã ∂b ó c ò . and let α α and check that they are equal.On the hand. ú ∑i ai ci . t1 t3 . t 2 . QED which proves the result. Conclude that there exists no k þ 1-chain b in U satisfying ∂ b ú c. The moral of this example is that the presence of the puncture in U is responsible both for the existence of the non-exact closed 1-form α (see Example 4. Lemma 5. (i) Let b be a k þ 1-chain in U . 5. but a § ø ÷ R3 by c t1 .For n ú 2. Let c be a k-cube in U and α a k-form on V . 4. 3. ∂c ¢ ¦¡ ¢ ú § . We detected both phenomena by using Stokes’ theorem.11. and let α ¥ ú x1 dx2 þ ∂c α and check that they are equal.1 ê k 1 î ciá . E XAMPLE .0α . The unit circle c t çïã cos 2π t. V an open subset of Rm and φ : U ÷ map.2. â c dα ã i ä 1 åuæ ∑ ∑ k k k 1ç iè 1ç iè 1 âxé â é ∂ gi dti dt1 dt2 ë+ë+ëíd ì ti ë+ë+ë dtk 0. ∂c V a smooth c αã ∂b è cö αã b dα ã 0. Prove that ø c φ ù α úûø φ ü c α .5. (ii) Let c be a k-cube in U . Let U be an open subset of Rn .6) and for the closed 1-chain c which is not a boundary. Indeed. å â â â where we have used Stokes’ theorem. 1 x1 dx2 dx3 . Deﬁne a 3-cube c : ÿ 0. 5. (ii) Calculate both ø c dα and ø 5.0α ç 0. punctured plane U ã R2 0 ñ . t3 ú t2 t3 . ∂b ú ∑i ai ci .1. Calculate both ø c dα and 3 R3 by c t1 .ρα ρ i ä 1 ρ ä 0. Considered as æð However. 5. t 1 t 2 . This is a contradiction. Then õ c α ã 2π by Example 4.18) write the n ý 1-dimensional unit sphere S n 1 in Rn as the image of an n ý 1-cube c. ciá . Using polar coordinates in n dimensions (cf. Let U be an open subset of Rn .ρ αã â ∂c α.3. Deﬁne a 2-cube c : ÿ 0.1 ê k 1 i ä 1 ρ ä 0. 1 2 ÷ x1 dx3 þ x2 dx3 . Accordingly.1. suppose that c ã ∂b ó c ò . sin 2π t ç is a 1-cycle in the å å a chain in R2 it is also a boundary.1 ê ∂ t i k ã ã ã 1 i ä 1 åuæ 0.64 5. 5. INTEGRATION AND STOKES’ THEOREM are nothing but ciá . Prove that ∑i ai ú 0. Exercises 5.5 and the fact that α is closed.1 åuæ c i .9. calculate the boundary ∂c of this cube.1 åuæ ∑ ∑ ∑ ∑ k 1ç iè 1ç iè ρ î å â é â ciá . we claim that it is not a boundary in U in as we saw in Example 5. (The domain of c will not be the unit cube in R n 1 . Exercise 3.4.1α .1α æ ciá . Its boundary is a linear combination of k ý 1-cubes. resp. Let α ã y dx ó åuæ other x dy çdô x2 ó y2 ç be the angle form. Its boundary is a linear combination of k-cubes. t2 . t1 t2 .

15.6. Choose R in such a way as to cover the sphere as economically as possible.) 5. vector ﬁelds.5. as indicated. are smooth and are deﬁned in an open subset U of Rn . (Some formulas hold only for special values of n.18.EXERCISES 65 rectangular block R dictated by the formula in Exercise 3. All functions.) (i) ∂g ∂f dx dy ∂x ∂y and any 2-chain c.) © . (iv) Stokes’ formula: c.) c c ¨ c grad g dx (ii) Green’s formula: (iii) Gauß’ formula: any n-chain c. We shall give a geometric interpretation of the entity dx in terms of volume forms later on.) In parts (iii) and (iv) we use the notations dx and dx explained in Section 2. (Here n 2. (Here n 3. chains etc. Deduce the following classical integration formulas from the generalized version of Stokes’ theorem. (See Corollary 8.

.

¨ ¨ ¨ © g c 1 c g c 0 div F dx1 dx2 curl F dx .

.

g dx for any vector ﬁeld F and . f dx F g dy for any functions f . ¨ ©©© ¨ © ¨ © ∂c for any function g and any curve c.

dxn ∂c ∂c F dx for any vector ﬁeld F and any 2-chain .

.

6. which is a collection of maps.1. An arbitrary manifold is deﬁned similarly. 2 1 N N x x n 67 . You should think of ψ U as an n-dimensional “patch” in R parametrized by the map ψ. The formal deﬁnition is given below and is unfortunately a bit long.e. An embedding of U into R N is a C map ψ : U RN satisfying the following conditions: (i) ψ is one-to-one (i. D EFINITION . A useful way to represent the earth is by means of a world atlas. The inverse map ψ is called a chart or coordinate map. Since Dψ t has N rows. if ψ t1 ψ t2 . Let U be an open subset of Rn . indeed differentiable (in most traditional cartographic projections).CHAPTER 6 Manifolds 6. The deﬁnition Intuitively. or else ψ cannot be an embedding. These maps are a special kind of parametrizations known as embeddings. Thus the patch ψ U has no self-intersections. because charting a curved surface on a ﬂat piece of paper inevitably distorts the distances between points. This is imposed to prevent the occurrence of cusps and other singularities in the image ψ U . Each map depicts a portion of the world. Maps of neighbouring areas overlap near their edges and the totality of all maps in a world atlas covers the whole world.1. Condition (i) means that to distinct values of the “parameter” t must correspond distinct points ψ t in the patch ψ U . (iii) the inverse of ψ. The image of the embedding is the set ψ U ! ψ t #" t U $ consisting of all points of the form ψ t with t U . It will help to consider ﬁrst the basic example of the surface of the earth. U. The column space of Dψ t is called the tangent space to the patch at the point x ψ t and is denoted by T ψ U . 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”. which is a map ψ 1 : ψ U t ). then t1 (ii) Dψ t is one-to-one for all t U . is continuous. this condition also implies that N % n: the target space R must have dimension greater than or equal to that of the source space U . such as a country or an ocean. The correspondence between points on a map and points on the earth’s surface is not entirely faithful. which is a twodimensional sphere placed in three-dimensional space. Condition (ii) means that for each t in U all n columns of the Jacobi matrix Dψ t must be independent. But the distortions are continuous. as an n-dimensional “world” represented by an “atlas” consisting of “maps”. T ψ U & Dψ t R .

) ' . the7 graph is a subset of R with N n : m.4 for an example.ψ ' U ( . 7 Since t is an n-vector and f ' t ( an m-vector. Try to write a formula for such an embedding! (If we chose U too big. E XAMPLE .2./ ψ t .t . Deﬁne t ψ ' t ( -./ ψ U e1 .ψ ' t ( implies t .3. The graph of f is the collection 0 Rm be a smooth .( )+* - )+* ' ( '( '( 6. Furthermore ψ is an embedding. they span the tangent plane at the image point.68 6. U 9 . ψ ' t ( ./ ψ U t e2 e1 e3 e2 Dψ t e2 . Dψ t e1 . then lim i ti t. the image would self-intersect and the map would not be an embedding. / -2143 f 'tt(65877 t .) For one particular value of t the column vectors of the Jacobi matrix are also shown. Also. (See Exercise 6. We claim that the graph is the image of an embedding ψ : U 0 R . I Dψ ' t ( -. Condition (iii) can be restated as the requirement that if ti is any sequence of points in U such that lim i ψ ti exists and is equal to ψ t for some t U .3 . MANIFOLDS The tangent space at each point is an n-dimensional subspace of R N because Dψ t has n independent columns. so ψ is one-to-one.3 . This is intended to avoid situations where the image ψ U doubles back on itself “at inﬁnity”. Indeed. As you can see. D f ' t (<5 graph f N N 1 2 1 2 n 6. Let U be an open subset of Rn and let f : U map. f ' t ( 5 Then by deﬁnition graph f . the image of which is a portion of a torus. The picture below shows an embedding of an open rectangle in the plane into three-space. E XAMPLE .

so Dψ t has n independent columns. Finally the inverse of ψ is given by

=>

6.1. THE DEFINITION

69

ψ

which is continuous. Hence ψ 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 ψ : U R N satisfying ψ U V M. N The codimension of M in R is N n. Choose t U such that ψ t x. Then the tangent space to M at x is the column space of Dψ t ,

? @ f =tt>6ACB

1

t,

EE E

FF

D

(Using the chain rule one can show that Tx M is independent of the choice of the embedding ψ.) The elements of Tx M are tangent vectors to M at x. A collection of embeddings ψi : Ui R N with Ui open in Rn and such that M is the union of all the sets ψi Ui is an atlas for M. 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 deﬁnition. 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 ψ sends t to x and parametrizes the portion of the curve inside V . Since n 1, the Jacobi matrix Dψ t consists of a single column vector, which is tangent to the curve at x ψ t . The tangent line Tx M is the line spanned by this vector.

= >B H I D= > =>B T M Dψ = t >J= R > . B

x n

G

= >

G

K

B

B

=> = > B

B

U t

Tx M

M

ψ

V

x

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 ψ : U RN such that M ψ U . However, usually one needs more than one chart to cover a manifold. (For instance, one chart is not enough for the curve M in the picture above.)

BB = >

F

G

1In the literature this is usually called a submanifold of Euclidean space. It is possible to deﬁne manifolds more abstractly, without reference to a surrounding vector space. However, it turns out that practically all abstract manifolds can be embedded into a vector space of sufﬁciently high dimension. Hence the abstract notion of a manifold is not substantially more general than the notion of a submanifold of a vector space.

70

6. MANIFOLDS

6.5. E XAMPLE . An open subset U of Rn can be regarded as a manifold of dimension n (hence of codimension 0). Indeed, U is the image of the map ψ : U Rn given by ψ x x, the identity map. The tangent space to U at any point is R n itself.

M NO

L

6.6. E XAMPLE . Let N n and deﬁne ψ : Rn R N by ψ x1 , x2 , . . . , xn x1 , x2 , . . . , xn , 0, 0, . . . , 0 . It is easy to check that ψ is an embedding. Hence the image ψ Rn is an n-manifold in R N . (Note that ψ Rn is just a linear subspace isomorphic to Rn ; e.g. if N 3 and n 2 it is just the xy-plane. We shall usually n identify R with its image in R N .) Combining this example with the previous one, we see that if U is any open subset of Rn , then ψ U is a manifold in R N of codimension N n. Its tangent space at any point is Rn .

M

M N

N P O

L

O

M N L

M

NQO

R

M N

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 ψ : U R n m , so n m M is an n-dimensional manifold in R , covered by a single chart. At a point x, f x in the graph the tangent space is spanned by the columns of Dψ. 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 .

O

M M NN

O O

S

L

S

M TUM NN

TVM N

M M NN

graph f

f x

WX

Y ZVW X6[

1 f x

x

angles, together with a few points and tangent vectors. (To improve the scale the

**M N ]a^ 1 _\` b ]^ 1 _` 0 . M x, y N and M x , y N 0 The diagram below shows the graph of f M x, y NcO x d y R 3 xy from two different
**

∂f ∂x ∂f ∂y 3 3

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 Dψ x, y , namely

M

O M N\N

O

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 ψ t et cos t, sin t . 6.8. E XAMPLE . Consider the path ψ : R e t . Therefore Let us check that ψ is an embedding. Observe ﬁrst that ψ t t t 1 2 ψ t1 ψ t2 implies e e . The exponential function is one-to-one, so t 1 t2 . This shows that ψ is one-to-one. The velocity vector is

f glh f g

2 2

h

e

f gih f j fk g jh

h

g

t p sin t mnf glh e o cos . cos t q sin t r Therefore ψ m f t gsh 0 if and only if cos t h sin t h 0, which is impossible because cos t q sin t h 1. So ψ m f t g+h t 0 for all t. Moreover we have t h ln e h ln j ψ f t gkj . Hence the inverse of ψ is given by ψ u f x glh ln j x j for x v ψ f R g and so is continuous. Therefore ψ is an embedding and ψ f R g is a 1-manifold. The image ψ f R g is a spiral, which for t ewpyx converges to the origin. It winds inﬁnitely many times around the origin, although that is hard to see in the picture.

ψ t

t

t

1

y

x

Even though ψ R is a manifold, the set ψ R singularity at the origin!

f g

f g{z}| 0 ~

is not: it has a very nasty

Instead. & v Dφ ψ t Dψ t v Dφ ψ t 0 0. even for a very simple manifold. in practice it can be rather hard to decide whether a given subset is a manifold using the deﬁnition alone. for example the inverse of the stereographic projection from the south pole. . . E XAMPLE . embeddings can be hard ﬁnd in practice.. then ψ U M V. manifolds are often given “implicitly”. n This ﬁnishes the proof that M is an n 1-manifold in R . x N \ \ c c x . An example of a manifold which cannot be covered by a single chart is the unit sphere M S n 1 in Rn . x N φ2 x1 . As this example shows. the image of ψ is the punctured sphere M en . Also we saw that ψ has a two-sided inverse φ : ψ U U . Fortunately there exists a more manageable criterion for a set to be a manifold.72 6. . . . the deﬁnition of a manifold can be a little awkward to work with in practice.. . Writing in the usual way \ φ1 x φ2 x . if v is in the nullspace of Dψ t . by a system of m equations in N unknowns. This is also an embedding and its image is M en M V . x N c1 . φ x x c we can abbreviate this system to a single equation φ x c. Here the φi ’s are smooth functions presumed to be deﬁned on some common open subset U of R N . . For a ﬁxed vector c R we denote the solution set by φ c l x U φ x c x1 x2 . where V Rn en . c! . c2 .4 is based on the notion of an embedding.2. Moreover. Let U Rn 1 and let ψ : U Rn be the map 1 ψ t 2t t 2 1 en 2 t 1 given in Exercise B. φm x1 .9. . . cm . . Aside from the above examples. Therefore ψ is one-to-one and its inverse is continuous (indeed. . . . . However. MANIFOLDS 6. . .7. As we saw in that exercise. . so if we let V be the open set Rn en . The regular value theorem Deﬁnition 6. . . which can be regarded as an “explicit” way of describing a manifold. Thus we see that ψ is an embedding. differentiable). φ ψ t t implies Dφ ψ t Dψ t v v for all v in Rn 1 by the chain rule. φ1 x1 . . Therefore. 6. the stereographic projection from the north pole. To cover all of M we need a second map. φ x . m 1 2 N m m 1 .

(As an extreme. though slightly silly. Identify R N with Rn Rm and correspondingly write an N -vector as a pair u. Suppose that c is a regular value of φ and that M φ 1 c is N nonempty. then c is automatically a regular value. m ¤ ¢ P ROOF. This matrix being invertible. v0 . Now refer to Appendix B. In other words. we see that M is an n-manifold. The tangent space T M is by deﬁnition the column space of Dψ u . by the implicit function theorem.10. We say that c Rm is a regular value of φ if the Jacobi matrix N m Dφ x : R R has rank m for all x φ 1 c . The dimension of this afﬁne subspace is N m. Its tangent space at x is the nullspace of Dφ x . ¤ 6. Hence Dφ ψ u Dψ u 0 by the chain rule. a constant. provided that φ has rank m (i. has m independent columns). T HEOREM (regular value theorem). f u c. it is a sufﬁcient but not a necessary criterion. because it does not apply to every possible manifold.) If φ is a linear map. We can generalize this idea to nonlinear equations as follows. (The notation φ 1 c for the solution set is standard. for all u U . so every tangent vector v to M at x is of the form v ¤ Dψ u a for some a ¡ R . THE REGULAR VALUE THEOREM and call it the level set or the ﬁbre of φ at c. To compute Tx M note that φ ψ u c. A vector that is not a regular value is called a singular value. 73 ¢ ¡ ¡ valid for any k l -matrix A.e. the system of equations is inhomogeneous linear and by linear algebra the solution set is an afﬁne subspace of R N . if φ 1 c is empty. but a bit unfortunate because it suggests falsely that φ is invertible.) The proof uses the following important fact from linear algebra. (Don’t get carried away though. Theorem B.4. Plugging in u u0 gives ¡ Tx M ¤ ker Dφ x . 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 φ u. Let U be open in R N and let φ : U R be a smooth map. Let us put n N m. 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). The tangent space Tx M 0 0 n . Let x M. f u . Tx M ker Dφ x . Therefore ¤ ¤ ¨ 0 x Dφ x v Dφ x Dψ u0 a 0. special case.) The following result is the most useful criterion for a set to be a manifold. After relabelling the coordinates on R N we may assume the last m columns are independent and therefore constitute an invertible m m-submatrix A of Dφ x . i.2. ¥ nullity A £ rank A ¤ l. We conclude from ExamM ple 6. The map f : U V is C .e. Then Dφ x has rank m and so has m independent columns. namely the image of the embedding ψ: U RN given by ψ u u. Also write x u0 . In other words U V graph f is the graph of a smooth map. ¤ ¤ ¥ ¥ ¢¡ l¤ ¤§ ¥ ¡ ¦ ¤§ ¥ ¢ ¤ ¡ ¥ ¤ ¤ ¡ ¤ Dφ x Dψ u l¤ 0. v with u a n-vector and v an m-vector.6.4 and observe that the submatrix A is nothing but the “partial” Jacobian Dvφ u0 . v0 . which it is usually not.7 that M U V is an n-manifold. Then M is a manifold in R of codimension m. Since U V is open in R N and the above argument is valid for every x M.

The diagram below shows a few level curves as well as the gradient vector ﬁeld. Hence Tx M ker Dφ x . obtained by intersecting the surface with horizontal planes at different heights. y ª³« xy. which as you Tx M 2 ©ª « ¯ ©ª ® © ª±«° XAMPLE T can see is perpendicular to the level curves. However. The solution set of a scalar equation φ x c is known as a level hypersurface. Let U « R and φ © x. i. the set φ 1 0 0 is a 1-manifold since the gradient is nonzero outside the origin. Then M is a manifold in R N of codimension 1. As explained in Appendix B. occur frequently in all kinds of applications. isotherms in weathercharts and contour lines in topographical maps are types of level curves. y x 0 is the only singular value of The gradient vanishes only at the origin.74 6. Let U be open in R N and let φ : U R 1 be a smooth function. It has rank 1 at x if and only if it is nonzero. the gradient points in the direction of steepest ascent.2. MANIFOLDS is n-dimensional (because the n columns of Dψ u0 are independent) and so is the nullspace of Dφ x (because nullity Dφ x N m n). Then Dφ is a single row vector and its transpose is the gradient of φ: Dφ T grad φ. E . The level curves of φ are hyperbolas in the plane and the gradient is grad φ © x ª´« © y. C OROLLARY (level hypersurfaces). The level curves of φ are the contour lines of the surface. « © grad φ © xª\ª² .e. For example.11. The ﬁbre φ 1 0 is the union of the two coordinate axes.12. at least one of the partials of φ does not vanish at x. By Corollary 6. so φ 0 φ. QED ©ª ©ª © ª¬«© ª « © ª« « The case of one single equation (m 1) is especially important. « « 6. Suppose that M φ c is nonempty and that grad φ x 0 for all x in M. Its tangent space at x is the orthogonal complement of grad φ x . Where the contour lines self-intersect the surface ¯ ©ª « © ª © ¯ © ª ªi« ¯ © ª µ ¶ «° . which has a self-intersection and so is not a manifold. y shown below.11 this means that φ 1 c is a 1-manifold for c 0. Think of this diagram as a topographical map representing the surface z φ x. Level hypersurfaces. 6. x ª . especially level curves.

So again any c 0 is a regular value of φ.12 grad φ vanishes only at the origin 0. The corresponding values of φ are 0. Then grad φ x 2x. The tangent ¹ · ¸i¹¿¾ ¾ · ¸¹ Â ½ ·¸ ¼ Ã Á ½ ·¸ . Let U R N and φ x x 2 . E XAMPLE . The level curve φ 1 0 has a self-intersection at the origin. Here grad φ x 3 x 2 y.7. Clearly. · ¸ ½ ·¸ ¹À ½ ·¼ ¸ ¹ · ¸ ½ ·¸ 6. but consists of the single point 1. 1 . These features are also clearly visible in the surface itself. which is contained in φ 1 0 . 1. E XAMPLE . For c 0.13. resp.14. which corresponds to a saddle point on the surface. A more interesting example of an equation in two variables x3 y3 3 xy c. the sphere of radius c in R N . Here φ has a minimum and the surface z φ x. φ 1 c is empty for c 0. 1 T . THE REGULAR VALUE THEOREM 75 has a “mountain pass” or saddle point. φ 1 c is an N 1-manifold. y has a “valley”. y ¼ · ¸º¹ » ¼ ¹ · ¸ · ¸y¹ · ¼ ¼ ¸ x The level “curve” φ 1 1 is not a curve at all. so as in Example 6.2. which is shown in Example 6. y T vanishes at the origin and at 1. which are the singular values of φ. e3 e2 e1 6. y2 x T . so grad φ is φ x.6.

Î ÄÏ Å Î ÄÅ Î Ä ÅÆÉÈ Í Æ 2). 0 . Æ Æ 0.16. you cannot be certain that φ 1 c is not a manifold.76 6.13.Ñ 2 0 Ø 0 0 0 1 0 .) Ï Æ Ç}ÆÉÈ Ê Ë Ì Æ Í Ä ÅºÆ 2x. just like the singular ﬁbre φ 1 1 in Example 6. Therefore. M φ 1 1 0 is a T 1. x2 x4 Ó If x1 0 the ﬁrst and third columns of Dφ x are independent. However. second and fourth columns are independent. and if x 2 0 the x2 0. they form an orthonormal basis of R n —note the regrettable inconsistency in the terminology. (In other words. by the regular value theorem. 6. though. if x 1 Dφ x has rank 1 and φ x 0. x2 Ó Æ¿Ä Ó Å Æ ÆÔ Æ Æ Î Ö× Dφ x Ä ÅÆ. For instance. then Æ Ä ÅÆ A A deﬁnes a map φ : V Ð W . Note that M contains the point x of the tangent space Tx M. this tangent space is equal to the nullspace of ÄÅ ÆÔ Dφ x Ä ÅlÆ Ñ 2xx 2 x2 x4 Î ÄÅ Ä ÅÕÆ ÄÅ 0 x1 0 . On the other hand. for every nonzero vector c the set φ 1 c is a two-manifold in R4 . Let us prove using the regular value theorem that O n is a manifold. it is often of the “wrong” dimension. so to prove that O Ä n Å is a φ A T 1 Ä Å ÆÚÈ Ê Ë Æ Í Ä Å Æ Ù Ä Å Æ Ø manifold it sufﬁces to show that I is a regular value of φ.) The collection of orthogonal matrices form a group under matrix multiplication. E XAMPLE . MANIFOLDS space to the sphere at x is the set of all vectors perpendicular to grad φ x In other words. A basis of Tx M is We now come to a more sophisticated example of a manifold determined by a large system of equations. 0 is a singular value (the absolute minimum) of φ and φ 1 0 0 is not an N 1-manifold. (It happens to be a 0-manifold. Again by the regular value theorem.Ñ x x Ò 2 2 1 1 3 1 3 x2 2 . 0. This shows that the origin 0 in R 2 is the only singular value of φ. Tx M x y RN y x 0 . Here is an example of a manifold given by two equations (m 6. Deﬁne φ : R4 Then Ð R by x Ò φ Ä x ÅlÆ. 0 which is equal the set of all vectors y satisfying y 1 y3 therefore given by the standard basis vectors e2 and e4 . This means that the columns (and also the rows) of A are perpendicular to one another and have length 1. which is usually called the orthogonal group and denoted by O n . 0.15. even if a singular ﬁbre happens to be a manifold. so A T A is a symmetric n n R matrix. Finally. In other words. First observe that that A T A T A T A. So if c is a singular value. Recall that an n n-matrix A is orthogonal if A T A I . Let us ﬁnd a basis two-manifold. The derivative of φ can n . Clearly O Ä n Å±Æ φ Î Ä I Å . if V is the vector space of all n n-matrices and W C V C C T the linear subspace of all symmetric matrices. E XAMPLE .

Conclude that M ψ R is a 1-manifold.3.1) is solvable for B. Prove that the map ψ : R R2 given by ψ t e t . Is ψ an embedding and is ψ I a manifold? (Observe that ψ I is a portion of the curve studied in Exercise 1. This is a continuation of Exercise 1. Graph the curve M. ä ψ (i) Show that ψ is one-to-one. Here is a way of 1 1 guessing a solution: observe that C C T and ﬁrst try to solve BA T 2 C 2 C. By linear algebra this amounts to showing that the Dφ A B h 0 T T T 2 T T h 0 T T be computed by using the formula derived in Exercise B. 6. 3 at2 6.1. Deﬁne ψ : R å æ í æïæ ó õ å öé å æ±ç÷ ê ç å æ ä ä éUå æáç å æ{çCå è å æ è ä å æ{ è çëå è åì æç å í î è î æ åæ ä t a sin t åïè ð³æ ä å æñçòå óôå í åæ R2 by ψ t where f is the function given in Exercise B. 6.6. Is ψ R a manifold? 6. Let a 0.f t f t . Deﬁne ψ : R T cos t . (ii) Show that Dψ t is one-to-one for all t 0. Prove that ψ R is not a manifold at these points.) t3 . Sketch the image of ψ. (This becomes easier if you ﬁrst show that t is an increasing function of t. 1 6.1. 0. Deﬁne a map ψ : R2 î å æÿä å æñçCøúù è å æ å æ å æïå û æüû ù f t .5. and let ψ : I R 2 be the map ψ t 3 at 1 T 3 . Compute the tangent line to M at 1. It is now 1 easy to check that B CA is a solution of equation (6.4.6. (iii) Let U be the punctured plane R2 0 .2.EXERCISES 77 equation 1 Û Ü Ý lim Þ h Û φ Û A ß hBÜáà φ Û A ÜÜ 1 Ý lim Þ h Û A A ß hA B ß hB A ß h B B à A A Ü Ý BA ß AB . Show that ψ : U Conclude that ψ U is a two-manifold in R4 . 1 a cos t is an embedding. Determine all t for which ψ t 0.1).f t R4 by T T if t if t ý þ 0. where a is a nonzero constant.) Graph the curve deﬁned by ψ. 1 T Left multiplying both sides by A and using A A I gives B 2 CA. åæ å æ ¡£ ç ¢¤¥¤ © è . 0 and try to ﬁnd an equation for M. Prove that the map ψ : R R 2 given by ψ t T a sin t. 2 Ý Ý Û ß Ü Exercises ß Ý Ý Ý Ý R 2 by ψ t t sin t. one-to-one and that its inverse ψ 1 : ψ R R is continuous. et e t T is an embed2 ding. æ è ê³æ å æ 1 et 6.2. Show that ψ is one-to-one. We need to show that for A â O Û n Ü the linear map Dφ Û A Ü : V ã W has rank equal to the dimension of W . Let I be the open interval 1. 1 be a constant. Show that ψ is smooth. Show that ψ is one-to-one and that 1 t ψ t 0 for all t I .3: BA T AB T C (6. given any orthogonal A and any symmetric C.

å æ ä R4 is an embedding. ç÷ t1 t2 t3 1 2 t1 t2 t1 t2 2 t3 2 ¦¨§§ . .

78 6. . if nonempty.10. For n 3 sketch the level surface φ 1 c for a regular value c. (ii) Let A be a real 2 2-matrix. where A 00 x1 x4 x2 x3 . . Show that rank A 1 if and only if det A 0 and A 0. the identity matrix.) (ii) The special linear group is the subset of V deﬁned by Show that SL n is a manifold.10 are onedimensional manifolds. 6. where the a i are nonzero constants. 11 . . t 3 for some t 0. Conclude that the tangent space to SL n at I is the set of traceless matrices. 6. 6. (Apply the formula of Exercise B. SL n A V det A ! φ x Show that 0 is a regular value of φ. .) 6. (i) Let W be punctured 4-space R4 ∑ n i 1 det a1 .7. . Show that 0 is the only possible singular value of φ.5. a2 . R by 6. (Compute all 2 2-subdeterminants of Dφ and set them equal to 0. . 6.8.) (ii) Show that M φ 1 0 is a 2-manifold (where 0 is the origin in R2 ).9. φ 1 c is an n 1-manifold for c 0. . .11. bn denote the column vectors of A. 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. Show that M is a three-dimensional manifold. (iii) Find a basis of the tangent space Tx M for all x M with x 3 0. ai 1 .12.5. Let φ x a1 x2 a2 x2 an x2 n . . t. b2 .i the trace of B. MANIFOLDS (iv) Find a basis of the tangent plane to ψ U at the point ψ 1. matrices A satisfying tr A 0. (iii) Let M be the set of 2 2-matrices of rank 1. .e. a2 .) Conclude that. Assume that φ is smooth and that p 0. 6. . (You have to distinguish between a few different cases.) φ x $# % '& ( x1 x2 x3 x4 x1 x2 x3 x4 . Show that the trajectories of the Lotka-Volterra system of Exercise 1. (iii) Show that for A ∑n i 1 bi. . an . an and b1 .13. (The answer depends on x. Let φ : Rn 0 R be a homogeneous function of degree p as deﬁned in Exercise B. (i) Show that n-matrices and deﬁne φ : V R by φ A Dφ A B where a 1 . we have Dφ A B tr B. (iv) Compute TA M. " 0 and deﬁne φ : W 1 . . B. (Use the result of Exercise B. Deter1 2 mine the regular and singular values of φ. t 2 . bi .3 for the derivative and use the multilinearity of the determinant. Deﬁne φ : R4 R2 by (i) Show that Dφ x has rank 2 unless x is of the form t 2 . Let V be the vector space of n det A. . a i 1 . What is its dimension? I . resp. i. . . 1 .

21 - λ2 grad φ2 x + .31544461 ) . . Let M be the unit sphere x Rn x 1 . λ2 . z given by x2 y2 z2 1. 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. (i) Calculate grad f x .. x z 0.= . / .16 (eigenvectors via calculus). .7/98 1 1 2y 3 z over the circle C . Let M be 1 the manifold φ c . Where are the maxima and minima of f C? 6. Find the critical points of the function f x.14. / +.15.+ . ) + . where c is a regular value of φ. (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. . show that f x is the corresponding eigenvalue of x. +. y. 1 1 / 1 / . . +. x λm grad φm x . . λm such that grad f x (Use the characterization of Tx M given by the regular value theorem. Let f : U R be a smooth function. . Let U be an open subset of Rn and let φ : U Rm be a smooth map.0/ 4 + . + . (iii) Given an eigenvector x of length 1.0/ ) 79 λ1 grad φ1 x + .) 6. Prove that x M is critical for f M if and only if there exist numbers λ 1 . < <>: / ? . +. Let A A T be a symmetric n n-matrix and deﬁne f : Rn R by f x x Ax.EXERCISES 6. * .

.

We can abbreviate this by saying that AB @ F A A C BHG A BBD E$I A B C D E @ @ A B E A A BB A B E J ABE A B E ABE A B ABE A B L A B O N M L A A BB L A A BQB L M K K A B E A A BB E A A BB A BE E L M A B K A BPK A B j i fj E A ψL M ψ B J f i 1 on ψ j 1 ψi Ui . if M is the earth’s surface. for all u in ψ j ψi Ui . because we have f x f i ψi 1 x if x ψi Ui . possibly inﬁnite. Then we have two indices i and j and vectors t Ui and u U j such that x ψi t ψ j u . We say that f is C k if each of the local representatives f i is C k . it expresses nothing but the obvious fact that where the maps of New York and Pennsylvania overlap.1. In this section we present a practical. so we cannot form the limit ∂ f ∂ x j limh 0 f x he j f x h. A function f : M R is simply an assignment of a unique number f x to each point x in M. 1 i. and ψ i is a map of New York State. i. or height above sea level. M could be the surface of the earth and f could represent temperature at a given time. (Pursuing the weather chart analogy. index set. For instance.) For each i we deﬁne a function f i : Ui R by f i t f ψi t . Let M be an n-manifold in R N and let us ﬁrst consider what we might mean by a 0-form or smooth function on M. f is temperature. Here is one way out of this difﬁculty. the straight line x he j may not be contained in M.CHAPTER 7 Differential forms on manifolds 7. First deﬁnition There are several different ways to deﬁne differential forms on manifolds.e. Now suppose that x is in the overlap of two charts. (Here i ranges over some unspeciﬁed. This identity must hold for all u U j such that ψ j u ψ i Ui . But how would we deﬁne such a function to be differentiable? The difﬁculty here is that if x is in M and e j is one of the standard basis vectors. f i ψi f . the collection of all local representatives f i determines f . A B E A L A BB K A B 81 . Because M is a manifold there exist open R N such that the images ψ i Ui cover M: sets Ui in Rn and embeddings ψi : Ui M i ψ i Ui .e. the corresponding two temperature charts must show the same temperatures. workaday deﬁnition.2. The map ψi 1 ψ j is often called a change of coordinates and the consistency condition is also known as the transformation law for the local representatives f i . Then we must have f x f ψi t f ψ j u . Also ψi t ψ j u implies t ψi ψ j u and therefore f j u f i ψi 1 ψ j u .) Since f i is deﬁned on the open subset Ui of Rn . so 1 fi t f j u . then f i represents a temperature chart of NY. it makes sense to ask whether its partial derivatives exist. We call f i the local representative of f relative to the embedding ψ i . (For instance.) Conversely. This is a consistency condition on the functions f i imposed by the fact that they are pullbacks of a single function f deﬁned everywhere on M. A more theoretical approach is taken in Section 7.

This branch of algebra. we know the temperature everywhere. To see that this deﬁnition makes sense. now called exterior or alternating algebra was invented by Graßmann in the mid-nineteenth century and is a prerequisite for much of the more advanced literature on the subject. α on M is a collection of k-forms α i on Ui satisfying the transformation law on ψ j 1 ψi Ui . or simply a k-form.1) R X Z αi R ∑ f I dt I I 1 and αj R ∑ g J dt J J 1 j I. we check that the forms γ i satisfy the transformation law (7. Second deﬁnition This section presents some of the algebraic underpinnings of the theory of differential forms. resp. If U S S WW Y S W X αj RTS ψ U V ψ WX α i 1 j i (7. Deﬁnition 7. resp. γj R α jβ j i 1 j i i 1 j i i 1 j i i i 1 j i i i i j j i 1 j i i 1 j i i 1 j i 7. DIFFERENTIAL FORMS ON MANIFOLDS (That is to say.10(ii). We call αi the local representative of α relative to the embedding ψi and denote it by αi ψi α . is the local representative of α . 7.1. differentiated and integrated.12.2. Similarly. forms on a manifold can be added. A differential form of degree k. are two local representatives for α . relative to an embedding ψi : Ui M. J .1) in components. In that case the compatibility relation (7. the exterior derivative of α is deﬁned by setting S dα W R dα . As before. Here we have used the multiplicative property of pullbacks. Suppose αi . For example. R[S ψU V ψ W X α S ψ U V ψ W X β R\S ψ U V ψ W X S α β W]RTS ψ U V ψ W X γ . .11. Then we deﬁne the product γ αβ by setting γ i αiβi . Proposition 3.1 is particularly tractible if M is the image of a single embedding ψ : U R N . βi . Sometimes it is useful to write the transformation law (7. multiplied. β. D EFINITION .1): S dα W R dα R d S ψ U V ψ W X α RTS ψ U V ψ W X dα RTS ψ U V ψ W X S dα W . if we have a complete set of weather charts for the whole world. suppose α is a k-form and β an l -form on M. let us check that the forms S dα W satisfy the transformation law (7.1) is vacuous and a k-form α on M is determined by one single representative.82 7. Just like forms on Rn . We can do this by appealing to Theorem 3.1): U S S WW Z R ∑ S ψU V ψ W X f I i j I det D ψi S U V ψW R R where we used Theorem 3. but it works really well if a speciﬁc atlas for the manifold M is known.) Following this cue we formulate the following deﬁnition. a k-form ψ α on U . The collection of all k-forms on M is denoted by k M . This deﬁnition is rather indirect. then gJ on ψ j 1 ψi Ui .

a 1-form α on U can be deﬁned as a row vector α f1. fn whose entries are functions on U . Recall that Rn is the collection of all column vectors _```a x bQccc x x^ . f . let us start by observing that a row vector y is nothing but a 1 n-matrix. . y f yx. . . . The form is called constant if the entries f 1 . . 0 . By deﬁnition dx i is the constant 1-form T 0. . d h x Obviously we have y e c x l c x f ^ c yx l c yx . i f Every 1-form can thus be α Using this formalism we can write for any smooth function g on U dg ∂g ∂g ∂g ∂g ^ ∑ .. . Before giving a rigorous deﬁnition of differential forms on manifolds we need to be more precise about the deﬁnition of a differential form on R n . . This motivates the following deﬁnition. . . . . . Formally. f f ^ ∑ h f dx . 0. The dual is a vector space in its own right: if µ and µ are in V g we deﬁne µ l µ and cµ by setting e µ l µ f e v f ^ µ e v f l µ e v f for all v m V and e cµ f e v f ^ cµ e v f . . Constant 1-forms are also known as covariant vectors or covectors and arbitrary 1-forms as covariant vector ﬁelds or covector ﬁelds. dx ^£i . . k . . f n are constant. the i-th standard basis vector of written as n ^Te f e fg ^ ^Te 2 Rn . dxi ei the transpose of ei ..7. . is the set of linear maps from V to R. . ∂x ∂x ∂x j h ∂x n i 1 i i 1 2 n ^\e f .) We would like to extend the notions of covectors and 1-forms to vector spaces other than Rn . To see how. d 1 2 xn Let U be an open subset of Rn . .. 0. y . The set of constant row vectors is denoted by R n and is called the dual of Rn . Elements of V g are called dual vectors or covectors or linear functionals. . Thus a row vector can be viewed as a linear map which sends column vectors in R to one-dimensional vectors (scalars) in R ^ R. . If V is any vector space over the real numbers (for example R or a subspace of R ). 1. . the dual of V . 1 n i 1 i _```a x b ccc x yx ^Te y . We can multiply it by a column vector x to obtain a number. then V g . SECOND DEFINITION 83 Covectors.. 1 2 n 1 2 n i 1 i i n 1 1 2 2 1 1 2 2 n 1 n n 1 2 1 2 1 1 2 1 2 1 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. f2. ^ ∑ . . . . .2. . . . The deﬁnition of a 0-form on U requires no further clariﬁcation: it is simply a function on U . .

. . λi is determined by the rule λi v j δi. . QED λ vj n n i 1 i i j i 1 i i. j. j i k k 1 k . so λ n λ.2) c λ o v s]n ∑ c δ n c . Then for each j n 1. v . . dx . We need to write λ as a linear combination λ ∑n i 1 ci λi . o s{n We call λi the i-th coordinate function. Let V n R and ﬁx v w V . o that is l n µ Lv s . . . In other words. v . P . 7. Then dx o e sn e e n δ . j i. j i j . . j m k 1 k. so µ o Lv s]n ∑ l µ o w sn ∑ l δ n l .5. j-th matrix element of a linear map L : V W is equal to l n µ o Lv s . λ . vn of V . let us deﬁne λ n ∑ λ o v s λ . . w of W . v and µ . . we can apply both sides to the vector v j to obtain zn n w z n z o x (7. .6. w . . b . Let L : V W be a linear map between abstract vector spaces V and W . Then by equation (7. To show that this choice of coefﬁcients works. µ w W z the dual basis of w . . n w o s]n n}| 1 0 if i if i 84 7. Deﬁne µ o x sxn v y x. E XAMPLE .e. λ o v s]n λ o v s for all j. E . Consider R with standard basis e . λ w V z be the dual basis of v . Deﬁne µ f b c2 f 2 c1µ f 1 c2µ f 2 . .2. λ2 . λ . . . . dx . E . j i i. . n the vector Lv can be expanded uniquely in terms of the w’s: Lv n ∑ l w . . the collection of all continuous realvalued functions on a closed and bounded interval a. . . sn ∑ So c n λ o v s is the only possible choice for the coefﬁcient c . R . w . . . Dual bases come in handy when writing the matrix of a linear map. . . . The basis λ . λ n ∑ λ o v s λ . i. j ik i. so V is a vector space. Deﬁne a covector λ i V by λi v ci . j j j j j j n i 1 n i 1 j i i i i i 1 2 n 1 2 n XAMPLE n T i j i. . . v of V . . Then every vector v V can be written in a unique way as a linear combination ∑ j c j v j . 7. We have Lv n ∑ l w . n~ 7. . . . . P ROOF. µ . The m n numbers l make up the matrix of L relative to the two bases of V and W . λ of V z is said to be dual to the basis v . . j n 1 n i j 1 2 n 1 2 n j 1 2 m j m i 1 i. . Hence dim V n dim V. so the dual basis of o R sz is dx . DIFFERENTIAL FORMS ON MANIFOLDS Now suppose that V is a vector space of ﬁnite dimension n and choose a basis v1 . The coordinate functions λ 1 . . v2 . b . Then µ is a linear functional on V . . Then µ c 1 f 1 on V . . v . j m k 1 i j k m i j k.4.7. j n w z j. . Then the i . XAMPLE n n oqp r s p r o stn u os o v sn o s v o s 7. . . Let λ . e . Let V C 0 a. w . A linear combination of continuous functions is continuous. Let λ V . λn constitute a basis of V . L EMMA . . where “ y ” is the standard inner product on R . j EMMA 1 2 n 1 2 n 1 2 n 1 2 n ROOF j i. . L .3. . . To write the matrix of L we need to start by picking a basis v . . Assuming for the moment that this is possible.2). . . . . v of V and a basis w . . 2. We have proved that QED every λ w V z can be written uniquely as a linear combination of the λ . so µ is a linear functional a f x dx. . .

1 2 n λ v . but it is not al7. . There is a nice way to construct a basis of the vector space A k V starting from a basis v1 . . E . . XAMPLE λ : V R deﬁned by λ λ λ v . Let V R . . an alternating 1-multilinear For k 1 the alternating property is vacuous. SECOND DEFINITION 85 c v . More generally. . . . v . . . of Example 7.8.. . The inner product ternating. λk V . v . .10. . .6.k 2. . v . XAMPLE n XAMPLE 3 4 3 4 4 4 1 2 2 1 XAMPLE n n-multilinear function on Rn . . then for any permutation σ S we have λ v . . . and so is the determinant function of Example 7. . v . Thus For λ1 λ2 2 k k 1 2 k 1 k i j 1 i. v j .e. . . . v det λ v . R and let λ x.) It follows from the multilinearity and the alternating property of the determinant that λ λ λ is an alternating k-multilinear function.v sign σ λ v . A k-multilinear function on V is a function λ : V k R which is linear in each vector. Thus an element of V k is an ordered k-tuple v1 . v . . . . . . . . v .. . . . .. . . . λ v1 .k n. E . v2 . y x y. .. The wedge product is often denoted by λ λ λ to distinguish it from other products. if λ is alternating. Let V be a vector space and let V k denote the Cartesian product V V (k times). . . v i . . v k σ 1 σ k 1 i j k k 1 k A k-multilinear function is alternating or antisymmetric if it has the alternating property. Let V y. vk of vectors in V . . The collection of all alternating k-multilinear functions is denoted by A V . .7 is bilinear. (The determinant on the right is a k k-determinant. . . .8 is alternating. k 0 a k-multilinear function is deﬁned to be a single number. i. cv i i k 1 2 k 1 2 i k 1 2 i i k Multilinear algebra. v . . The idea is to take wedge products of dual basis vectors. . . . Let V R . . such as the tensor product deﬁned in Exercise 7. v c λ v . . . vn of V . 7. . . . The determinant det v v . k-multilinear functions can be added and scalar-multiplied just like ordinary linear functions. .7. . . λ2 . .9. . .9. . v for all scalars c. c and all vectors v .e. . . v . . .7. . . v2 . 2-multilinear). v . w v w v w is bilinear on R . Indeed it is symmetric: y x x y. the inner product of x and 7. . The function λ v. j k 1 2 k 1 2 k k 1 Here is a useful trick to generate alternating k-multilinear functions starting from k covectors λ1 . . E . Thus A V V . vw vw 7. v . . . v . . The bilinear function of Example 7. . v cλ v . For any k. . . so the set A k V forms a vector space. Then λ is bilinear (i. v is an λ v1 .2. . so function is nothing but a linear function. v . . The (wedge) product is the function A0 V R. . E .

¢¤£ i , i , . . . , i ¥ be ¦ § §©¨¨¨§ ¦ ¡ λ ¢ λ λ ¨¨¨ λ ª A V, v ¢T£ v , v , . . . , v ¥ ª V . 7.11. E . Let V ¢ R with standard basis e , e , e . The dual basis of £ R ¥ ¡ is dx , dx , dx . Let k ¢ 2 and I ¢[£ 1, 2 ¥ , J ¢T£ 2, 3 ¥ . Then 0 £ e ¥ dx £ e ¥ « ¢¬«« 1 dx £ e ¥ ¢}« dx £ e ¥ dx £ e ¥ « 0 1« ««« dx «« «« ««« ¢ 1, £ e ¥ dx £ e ¥ « ¢ «« 0 0 dx £ e ¥ ¢ « ««« dx dx £ e ¥ dx £ e ¥ « 1 0« «« «« ««« ¢ 0, £ e ¥ dx £ e ¥ « ¢¬«« 0 1 dx £ e ¥ ¢}« dx £ e ¥ dx £ e ¥ « 0 0« ««« dx «« «« ««« ¢ 0, £ e ¥ dx £ e ¥ « ¢¬«« 1 0 dx £ e ¥ ¢}« ««« dx 0 1« dx £ e ¥ dx £ e ¥ « «« «« ««« ¢ 1.

Let λ1 , . . . , λn be the corresponding dual basis of V . Let I an increasing multi-index, i.e. 1 i 1 i2 ik n. Write

I i1 i2 i1 ik k k

86

7. DIFFERENTIAL FORMS ON MANIFOLDS

1

2

k

I

i2

ik

3

XAMPLE

1

3

1

2

3

2

3

I

I

1 2 1 2 2 3 2 3

1 1 2 2 1 1 2 2

1 2 1 2 2 3 2 3

2 2 3 3 2 2 3 3

I

J

J

I

J

J

This example generalizes as follows.

if I ¢ J , £ ¥ ¢ δ ¢} 1 0 if I ¢ ® J . P . Let I ¢T£ i , . . . , i ¥ and J ¢T£ j , . . . , j ¥ . Then ««« δ ... . . . δ ... ««« «« . . . δ ««« ¢ 1 if I ¢ J. λ £ v ¥ ¢ det ¯ λ £ v ¥° ± ± ¢ « δ . ««« ... «« . . « ««« δ . . . δ ««« If I ¢ ® J , then i ¢ ® j for some l . Choose l as small as possible, so that i ¢ j for m § l . There are two cases: i § j and i ² j . If i § j , then i § j § j ³ § ¨¨¨7§ j because J is increasing, so all entries δ in the determinant with¢ m0 ´ forl are 0. For m § l we have j ¢ i § i because I is increasing, so δ m § l . In other words the l -th row in the determinant is 0 and hence λ £ v ¥ ¢ 0. If i ² j we ﬁnd that the l -th column in the determinant is 0 and therefore again λ £ v ¥ ¢ 0. QED

7.12. L EMMA . Let I and J be increasing multi-indices of degree k. Then

λI vJ

k

I, J

ROOF

1

1

k

i1 , j1

i1 , jk

I

J

ir

js

1 r, s k

il , j1

il , jk

ik , j1

ik , jk

l

l

m

m

l

l

l

l

l

l

l

l

l 1

k

il , jm

m

m

l

il , jm I

J

l

l

I

J

We need one further technical result before showing that the functions λ I are a basis of Ak V . 7.13. L EMMA . Let λ of degree k. Then λ 0.

¢

ª

A k V. Suppose λ v I

£ ¥¢ ¥¢

0 for all increasing multi-indices I

P ROOF. The assumption implies

λ vi 1 , . . . , v i k

£

0

(7.3)

for all multi-indices i1 , . . . , ik , because of the alternating property. We need to show that λ w1 , . . . , wk 0 for arbitrary vectors w1 , . . . , wk . We can expand the wi using the basis: w1 wk Therefore by multilinearity

µ

µ

7.2. SECOND DEFINITION

87

¶¸· ¶

· ·

a11 v1 . . . a k 1 v1

k

¹»ººº¹

k

a 1 k vk , akk vk .

¹¼ººº½¹

1i1

λ w1 , . . . , w k

µ

Each term in the right-hand side is 0 by equation (7.3).

**a λµ v , . . . , v ¶ . ¶x· ∑ ¾ ººº ∑ ¾ a ººº
**

i1 1 ik 1 ki k i1 ik

7.14. T HEOREM . Let V be an n-dimensional vector space with basis v 1 , . . . , vn . Let λ1 , . . . , λn be the corresponding dual basis of V . Then the alternating k-multilinear functions λ I λi1 λik , 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 .

Lemma 7.12 we obtain

·¬Â Ã P . The proof is closely analogous to that of Lemma 7.4. Let λ Ä A V . We need to write λ as a linear combination λ · ∑ c λ . Assuming for the moment that this is possible, we can apply both sides to the k-tuple of vectors v . Using

ROOF

k I I I J

¿

·

À

ººº

Á

¿

QED

À

to

· µ ¶ Å ·µ µ µ ¶ Å µ µ > ¶ Æ ¶ · ¶>Æ · ¶{·µ ¶ Æ ÅÄ Á Æ Å · 7.15. E . Let V · R with standard basis ¿ e , . . . , e À . The dual basis of µ R ¶ Á is ¿ dx , . . . , dx À . Therefore A V has a basis consisting of all k-multilinear functions of the form dx · dx dx ººº dx , with 1 Ç i È\ºººHÈ i Ç n. Hence a general alternating k-multilinear function λ on R looks like λ · ∑ a dx , with a constant. By Lemma 7.12, λ µ e ¶É· ∑ a dx µ e ¶Ê· ∑ a δ · a , so a is equal to λ µ e ¶ . An arbitary k-form α on a region U in R is now deﬁned as a choice of an alternating k-multilinear function α for each x Ä U ; hence it looks like α · ∑ f µ x ¶ dx , where the coefﬁcients f are functions on U . We shall abbreviate this

I I n

λ vJ

µ ¶·

∑ cI λI

µ v ¶]·

J

∑ cI δI, J

·

cJ.

So c J λ v J is the only possible choice for the coefﬁcient c J . To show that this choice of coefﬁcients works, let us deﬁne λ ∑ I λ v I λ I . Then for all increasing multi-indices I we have λ v I λ vI λ vI λ vI 0. Applying Lemma 7.13 to λ λ we ﬁnd λ λ 0. In other words, λ ∑ I λ v I λ I . We have proved that every λ V can be written uniquely as a linear combination of the λ i . QED

XAMPLE

1 n n k 1 n

I

i1

i2

ik

n

1

k

I

I

I

I

J

I

I

I

J

I

I I, J

J

I

I

n

x I

x

I I

I

α

·

∑ f I dx I ,

I

88

7. DIFFERENTIAL FORMS ON MANIFOLDS

and we shall always assume the coefﬁcients f I to be smooth functions. By Example 7.15 we can express the coefﬁcients as f I α e I (which is to be interpreted as fI x αx e I for all x).

Ì Í]Ë Ì Í

Ë Ì Í

Î ÏÐ Ì Í Ñ ÏÐ Ì Í Ì φÑ α Í Ì v , v , . . . , v Í]Ë α Ò Ó Ì Dφ Ì xÍ v , Dφ Ì xÍ v , . . . , Dφ Ì xÍ v Í . Let us check that this formula agrees with the old deﬁnition. Suppose α Ë ∑ f dy and φ Ñ α Ë ∑ g dx . What is the relationship between g and f ? We use g Ë φ Ñ α Ì e Í , our new deﬁnition of pullback and the deﬁnition of the wedge product to get g Ì x Í]Ë Ì φ Ñ α Í Ì e ÍxË α Ò Ó Ì Dφ Ì x Í e , Dφ Ì x Í e , . . . , Dφ Ì x Í e Í Ë ∑ f Ì φ Ì xÍÍ dy Ì Dφ Ì xÍ e , Dφ Ì xÍ e , . . . , Dφ Ì xÍ e Í Ë ∑ φÑ f Ì xÍ det Ô dy Ì Dφ Ì xÍ e ÍÕ Ö Ö . By Lemma 7.6 the number dy Ì Dφ Ì x Í e Í is the i j -matrix entry of the Jacobi matrix Dφ Ì x Í (with respect to the standard basis e , e , . . . , e of R and the standard basis e , e , . . . , e of R ). In other words, g Ì x Í×Ë ∑ φ Ñ f Ì x Í det Dφ Ì x Í . This

x 1 2 k

Pullbacks re-examined. In the light of this new deﬁnition 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 φ : U V a smooth map. For a k V deﬁne the pullback φ α k U by k-form α

φ x

1 2 k

I I

I

J

J

J

J

I

J

J

J

x

J

φ x

j1

j2

jk

I I

I

I

j1

j2

jk

I

ir

js

1 r, s k

ir

js

r s 2

formula is identical to the one in Theorem 3.12 and therefore our new deﬁnition agrees with the old!

1

2

m

m

1

n

n

J

I

I

I, J

Forms on manifolds. Let M be an n-dimensional manifold in R N . For each point x in M the tangent space Tx M is an n-dimensional linear subspace of R N . A differential form of degree k or a k-form α on M is a choice of an alternating kmultilinear map αx on the vector space Tx M, one for each x M. This alternating map αx is required to depend smoothly on x in the following sense. According to the deﬁnition of a manifold, for each x M there exists an embedding ψ : U RN N such that ψ U M V for some open set V in R containing x. The tangent space at x is then Tx M Dψ t Rn , where t U is chosen such that ψ t x. The pullback of α under the local parametrization ψ is deﬁned by

Ì ÍÌ Í Ï Ì ÍÚË Ì ψÑ α Í Ì v , v , . . . , v ÍË α Ò Ó Ì Dψ Ì tÍ v , Dψ Ì tÍ v , . . . , Dψ Ì tÍ v Í . Then ψ Ñ α is a k-form on U , an open subset of R , so ψ Ñ α Ë ∑ f dt for certain functions f deﬁned on U . We will require the functions f to be smooth. (The form ψ Ñ α Ë ∑ f dt is the local representative of α relative to the embedding ψ, as introduced in Section 7.1.) To recapitulate: 7.16. D . A k-form α on M is a choice, for each x Ï M, of an alternating k-multilinear map α on T M, which depends smoothly on x. Ë

t 1 2 k

Ì ÍØË

Ù

Ï

Ï

Î

ψ t

1

2

k

n

I I

I

I

I

I

I

I

EFINITION

x

x

The book [BT82] describes a k-form as an “animal” that inhabits a “world” M, eats ordered k-tuples of tangent vectors, and spits out numbers. 7.17. E XAMPLE . Let M be a one-dimensional manifold in R N . Let us choose an orientation (“direction”) on M. A tangent vector to M is positive if it points in the

Deﬁne a 1-form α on M as follows. . .5. 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. Prove that µ is alternating.EXERCISES 89 same direction as the orientation and negative if it points in the opposite direction. λn A 1 . v 0 for all vectors v V . Avn is also a basis of Rn .e. Generalize this observation to k-multilinear functions. 7. . . . Suppose that µ is a bilinear function on a vector space V satisfying µ v. v2 . ê0ñ é ê é ê ù ùù é ê µ 1 v1 µ 2 v 2 û ò ë V be covectors. . λ2 A 1 .4. . j 1 i i 1 i j j 1 i j i i j j j i 1 j i ã ã â Þ This shows that Deﬁnitions 7. . 7.2. The vectors e 1 7. 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. µ2 . . y R3 . x T yT is a 2-form on R3 . Ý Þ Ü Ý Þ â Þ âÜ ψ ä å ψ . .8 is equal to dx 1 dx2 î ï é êë í í í îð ð ò ò ç e2 and e1 è Exercises e2 form a basis of R2 . Let v1 . From α Þ ψ â α and α Þ ψ â α we recover the transformation law (7.6. How do the local expressions α i ψi α and α j ψ j α for α compare? To answer this question. consider the coordinate change map à Ü ÝÞ¬ß á à v à và ã Û Û if v is positive. vn be a basis of Rn and let λ1 . . What is the dual basis of R2 ? í î é êìë ð î é ê0ñ 7. . .16 of differential forms on a manifold are equivalent.1 and 7. . . Suppose we had two different such M and ψ j : U j M. . We can calculate the local representative ψ α of a k-form α for any embedding ψ: U R N parametrizing a portion of M. . . . . x T and yT are row vectors. Show that the corresponding dual basis is the set of row vectors λ 1 A 1 . So both sides of the formula represent column vectors. Their tensor R µ k vk . For x M and a tangent vector v Tx M put αx v This form is the element of arc length of M. µk product is the function ç dx3 dx4 . ö ï ñ$óõô é ö êì÷ ôé ö ê µ2 µk : V k µ1 deﬁned by Show that µ 1 ø 5 ø ùùùüø é ø øýùQùù6ø µ1 µ2 µ2 ø øúùùQù6ø µ k v1 . . λ2 .1.1) α Þ Ü ψä å ψ Ý â α .3. (Interpretation: x and y are column vectors. xT yT is a 1-form. which maps ψ ä Ü W æ W Ý to ψ ä Ü W æ W Ý . . i. a row vector. v2 . . such that x is contained in both Wi embeddings ψi : Ui ψi Ui and W j ψ j U j . λn be the dual basis of Rn . Show that the bilinear function µ of Example 7. . 7. Let A be an invertible n n-matrix. vk µk is a k-multilinear function. Let V be a vector space and let µ 1 . . Prove this formula.) 7. if v is negative. Then by elementary linear algebra the set of vectors Av1 .

vσ ¤ 2 ¥ . .90 by 7. . Then 1 Alt µ1 µ2 ¨. Let µ : V k Alt µ v1 . µk ¦ V § . . (i) Alt µ is an alternating k-multilinear function. . k! σ∑ £S k ÿ ÿ þ R Prove the following. . (ii) Alt µ ¡ µ if µ is alternating. . . vσ ¤ k ¥ . (iii) Alt Alt µ ¡ Alt µ for all k-multilinear µ . Deﬁne a new function Alt µ : V k 1 sign σ µ vσ ¤ 1 ¥ . DIFFERENTIAL FORMS ON MANIFOLDS R be a k-multilinear function. . . v2 . . . µ2 . vk ¢¡ ÿ þ 7. .7. (iv) Let µ 1 .

¨©¨.

. vn ¡ dx1 dx2 ¨©¨.8. . µk . µ1 µ2 ¨©¨©¨ µk ¡ k! 7. v2 . . . Show that det v1 .

. . µ ¦ W § . þ W a linear map. .9. Let V and W be vector spaces and L : V L § λ L § µ for all covectors λ. . ÿ ÿ ÿ ÿ ÿ 7. vn ¦ Rn . . . . Show that L § λµ ¡ ÿ .¨ dxn v1 . v2 . vn for all vectors v1 . v2 . In short. det ¡ dx1 dx2 ¨¨©¨ dxn . .

. an voln a1 .CHAPTER 8 Volume forms 8. cai . . an are orthogonal vectors. We will need a formula for the volume of a block. An absolute n-dimensional Euclidean volume function is a funcvol n : R N with the following properties: (i) homogeneity: voln a1 . q2 . . . . . . . . .1. . (ii) vol n a1 . . qN of R N . . a2 . qn . . We approach this problem in a similar way as the problem of deﬁning the determinant. . qn are orthonormal. . e2 . The vectors q1 . . 1 . . P ROOF. an for all scalars c and any i # j. Suppose a1 . . en 1. . . . a j . a2 . . . . . . a2 . We shall shortly see that these axioms uniquely determine the n-dimensional volume function. namely by imposing a few reasonable axioms. The block or parallelepiped spanned by these vectors is the set of all vectors of the form ∑ n i 1 c i ai . . ai " c a j . . ai . . . . . q2 . . . an 0 if the vectors a1 . a2 . . an are orthogonal. . . . . . Qan voln a1 . For n 1 this is also called a line segment and for n 2 a parallelogram. . (iv) normalization: vol n e1 . . . . so this volume will be not an oriented but an absolute volume. Complete them to an orthonormal basis q1 . . . an %$ a1 $&$ a2 $ $ an $ if a1 . . . .1. . . D EFINITION . . an for all scalars c and all vectors a1 . .2. . a j . . . n-Dimensional volume in R N Let a1 . . (iii) invariance under Euclidean motions: voln Qa1 . . . . . a2 . Let 91 . . a2 . . an are dependent. . Then we can deﬁne qi '$ ai $( 1 ai . . . . First assume they are nonzero. . an for all orthogonal matrices Q. (ii) invariance under shear transformations: vol n a1 . . an be vectors in R N . an . qn ) 1 . . a2 . . If n N there is no coherent way of deﬁning an orientation on all n-blocks in R N . . . (i) vol n a1 . . . a2 . 8. an R N RN n times R ! c voln a1 . tion 8. . . . . . a2 . . . where the coefﬁcients c i range over the unit interval 0. . L EMMA . . . .

e2 . an . . . . . . Qen . a2 . a1 -&. We can form a matrix A out of the column vectors a1 . . *. . a2 . a2 . Finally consider an arbitrary sequence of independent vectors a1 . Then Q is orthogonal and Qei Therefore vol n + a1 .a2 -0///1. a2 . . by Lemma 8. . . . Let a1 . 8. . a2 . Then voln + a1 . . .2.ai . . . . T HEOREM .a2 -0///1. . . .. which proves part (i) if all ai are nonzero. * voln + 0 0. P ROOF. so : det + A T A . First assume that a1 .2(ii). If one of the ai is 0. . there exists a nonzero n-vector v such that Av * 0. VOLUME FORMS Q be the matrix whose i-th column is qi .* . which is equal to voln + a1 . * 0.voln + q1 . . . * 0. . . Since A T A is square. *.. . an . so the statement follows from part (ii). * qi . a2 . . .a2 -0///. a2 . . Case 2. the vectors a1 . . an are dependent. Then A T A is a diagonal matrix. . Next assume that a1 . . . the product A T A is square and we can take its determinant. .a1 -&.an . en . . .a2 -0///1.e. . an . an . . . . Case 3. . a 2 . * * 3 vol n i 2 ∑ c i ai . an are dependent. . .2 . . . . a2 . . a1 -&. . QED which proves property (ii).voln + Qe1 . v2 . . .an .voln + e1 . . . an . an . . an 4 *5///6* 2 * n vol n + 0. . .92 8. . then . This brings us to the volume formula. . . .2(i). But then A T Av * 0. a2 . Qe2 .an . a2 . . . . Then the matrix A has a nontrivial nullspace. It does not make sense to take det A because A is not square. . which is equal to voln + a1 . an are dependent. . Now by Axiom (i). q2 . satisﬁes the axioms for a n-dimensional volume function on R N .a2 -0///1. . i. . an . a n 4 23 i 3 n vol n ∑ ci ai . . a2 . Its i-th diagonal entry is . a2 . an .3. . By repeatedly applying Axiom (ii) we get vol n + a1 . . . . There exists a unique n-dimensional volume function on R N . a2 . . . a1 * ∑n i 2 2 c i ai . 0 voln + 0. We leave it to the reader to check that the function : det + A T A . ai = 1 . an .an - by Axiom (i) by Axiom (iii) by Axiom (iv). .an . . an 7 R N and let A be the N 8 n-matrix whose i-th column is a i . a2 . which we prove next. an are orthogonal. (See Exercise 8. qn . a2 . *. unless n * N . . . . . Assume a1 . For simplicity suppose a1 is a linear combination of the other vectors. . . . . . this implies det A T A * 0. . by Lemma 8.) Here we prove only the uniqueness part of the theorem. so the columns of A T A are dependent as well. 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 . *59 det + A T A . a2 . so : det + A T A . . . . . a1 -&.. This sequence can be transformed into an orthogonal sequence v 1 . . vn by the Gram-Schmidt process. . a2 . . Case 1. . a2 .*.- a1 -&. However. vol n + 0. an . . an .

P det A P . 8. vol n A a1 . . a2 . an B >. (The horizontal “ﬂoor” represents the plane spanned by a1 and a2 . an into an orthogonal sequence v 1 . we have V > AU . . . an B >5CCC > voln A v1 . v2 .8. .3 gives the following result. an B @ n-matrix whose i-th column > vol n A v1 . an B > vol n A v1 .1) where the last equality follows from Case 1. . . . n-DIMENSIONAL VOLUME IN RN 93 vi > ai ? bi . . a2 . . . a2 . . where bi is a linear combination of a1 . U> . v2 .. . . (See illustration below. . C OROLLARY. Then voln A a1 . . .4. This implies that V T V > U T A T AU and det A A T A B > det U T det A A T A B det U > det A U T A T AU B > det A V T V B . Since vi > ai ? bi . .) The block spanned by the a’s has the same volume as the rectangular block spanned by the v’s. . Let a1 . a2 . Then by repeated applications of Axiom (ii). . QED Using formula (8. . ai E 1. vn . where U is a n @ n-matrix of the form FGG JK I CCC I KK GG 1 I CCC I KK GH 0 1 I 0 0 1 CCC I . . . (8. . vn B >5D det A V T V B . . . v2 . . . . . .1) we get vol n A a1 . . . . . an B >NM det A A T A B . I L .) Let V be the N is vi . @ n-matrix . a2 . a2 . an be vectors in Rn and let A be the n whose i-th column is ai . . . . 0 0 CCC 0 1 Note that U has determinant 1. . The Gram-Schmidt process transforms a sequence of n independent vectors a1 . . . . a3 a3 a2 a1 a2 b2 b3 v3 v2 a1 v3 O v1 a3 a2 a1 v2 v1 For n > N Theorem 8.1. . . a2 . . . .

en R . . An orientation of a plane is a choice of a direction of rotation. Because the determinant of an invertible matrix is either positive or negative. These notions can be generalized as follows. v1 .. . so det A Sml 1. . VOLUME FORMS P ROOF. . v2 .. . v2 _ . . . If det A e 0. We shall always use this orientation on Rn . . . .) o o o by \qp or v1 . vn R and Q v1 . i. j and bi. An orientation of a line is an assignment of a direction. Choose a positively oriented basis Q v1 . 0 0 0 0 . Let V be an n-dimensional vector space over the real numbers. vn R and \`_aS Q v1 _ . . . 0 1 0 0 . An oriented vector space is a vector space together with a choice of an orientation. but how do we deﬁne the concept of an orientation itself? In typical mathematician’s fashion we deﬁne the orientation of V determined by the basis \ to be the collection of all ordered bases that have the same orientation as \ . j v _ j for suitable coefﬁcients a i. . a choice of a right-hand rule versus a left-hand rule.2. . then For instance.k 1 ijj jj j . So if \ and \ _ deﬁne the q \ r p S \ _p. .. an RVSXW Q det A R 2 SZY det A Y by Theorem 8. We know now what it means for two bases to have the same orientation. Hence the ordered bases Q v2 . v2 . You are probably familiar with orientations on vector spaces of dimension [ 3. . . . . . The standard orientation on Rn is the orientation e1 . Then we can write vi_ S ∑ j ai. jv j and vi S ∑ j bi. if Q v1 fgg gg gh A S 0 1 0 . This preferred orientation is then called positive. We say that the bases \ and \ _ deﬁne the same orientation of V if det A d 0. t or l . . . . . . Let V and W be oriented vector spaces of the same dimension and let L : V u W be an invertible linear map. vn R of V . . . 8. . A is square. . .7(ii) and therefore vol n Q a1 .e. . the ordered n-tuple o . . The n b n-matrices A SZQ a i.3. vn _ R are two ordered bases of V . . . en p deﬁned by the standard ordered basis Q e1 . For n S 0 we need to make a special deﬁnition. Because L is invertible. a2 . v2 _ . . . (There is an analogous deﬁnition of the number 1. v2 . j.. vn p . .94 8.5. . so det Q A T A RTS det A T det A SUQ det A R 2 by Theorem 3.. v2 . E XAMPLE . . vn _ R&S5Q v2 .. vn R is denoted o element. _ .. . . . . . . . . .. . . . . . Maps and orientations. . 0 0 0 1 .. because a zero-dimensional space has an empty basis. Orientations Oriented vector spaces. . namely as the collection of all sets that contain one The orientation determined by \^SnQ v1 . there are two possible orientations of V . clockwise versus counterclockwise. QED 8. . . same orientation then o o If they deﬁne opposite orientations we write \qpsScl \ _ p . . v1 . v2 . j R and B ScQ bi. vn R deﬁne opposite orientations. In this case we deﬁne an orientation of V to be a choice of sign. vn R . . Suppose that \]S^Q v1 . . . . the two bases deﬁne opposite orientations. j R satisfy AB S BA S I and are therefore invertible. An orientation of a three-dimensional space is a choice of “handedness”.

for if v v1 x . (Here Rn is equipped with its standard orientation. Therev fore Lvix y L | ∑ j ai. Hypersurfaces. For each x ~ M the tangent space Tx M is n 1-dimensional. jv j } y ∑ j ai. vn x w is another positively oriented basis of V . negatively. . . . n P ROOF. . v . . . resp. v2 . “Continuous” means that for every x ~ M there v exists a loM. Lv 2 x . v v of Tx M for some x ~ M. . Lvn w v and Lv1 x . . v 8. with W open in Rn and x ~ ψ W w . v1 . A hypersurface in Rn is orientable if and only if it possesses a unit normal vector ﬁeld. is n particularly instructive. We deﬁne an orientation of M to be a choice of an orientation for each tangent space Tx M which varies continuously over M. such that cal parametrization ψ : W Dψy : Rn Ty M preserves the orientation for all y ~ W . . it is oriented if a speciﬁc orientation has been chosen. . ORIENTATIONS 95 Lv1 . If this basis is positively. . Lvn w is an ordered basis of W . Let M v be a hypersurface in R .6. v This deﬁnition does not depend on the choice of the basis. . . . . a manifold of codimension 1. . . . . and hence the two bases Lv 1 . Lv2 . j v j with det ai.2. A unit normal vector ﬁeld on a v v manifold M in R is a smooth n function n : M R such that n x w Tx M and n x w y 1 for all x ~ M. The case of a hypersurface. . . . . . so its orthogonal complement v Tx M w. vn 1 w be an ordered basis v ﬁeld. j w{z 0. . orientation-reversing . This deﬁnes an orientation on M. . . . vn 1 w is positively oriented if n x w .8. Now let M be a manifold. We say that 1 2 n 1 v v v v1 . v . resp. because n x n x . Oriented manifolds. v2 . then vix y ∑ j ai. called the orientation induced by the normal vector ﬁeld n.) A manifold is orientable if it possesses an orientation. . j Lv j . let us suppose that M is an oriented hypersurface in R n . v2 . Conversely. v2 x . n . . P ROPOSITION . Lv2 . Lv n x w of W determine the same orientation. . vn 1 w is a positively oriented basis of Rn . oriented we say that L is orientation-preserving. Suppose M possesses a unit normal vector Let v1 . v is a basis of R Then w w w& vi for all i. .

the opposite direction is “in” or “down”. In Exercise 8. . . e 2 . Let us show that a hypersurface given by a single equation is always orientable. E XAMPLE . and the standard orientation on R is e1 . v1 . The positive unit normal vector is that unit normal vector n x w that makes n x w . . e1 . . . e n 8. M in Rn .8 you will be asked to check that n x w depends smoothly on x. whereas the nonorientable ones are called one-sided. . In this way we have produced a unit normal vector ﬁeld on M. The positive unit normal on an oriented hypersurface can be regarded as a map n from M into the unit sphere S n 1 . e2 . e2 . en 1 . . e2 . For this reason orientable hypersurfaces are often called two-sided. We can pick a preferred unit normal vector as follows. v2 . . the positive unit normal to Rn 1 in Rn is 1 w n e1 .5. QED e n . .7. v2 . . . . The n 1 e1 . . . . . . . vn 1 w a posiv tively oriented basis of Rn . . . . Let us regard Rn 1 as the subspace of Rn spanned by the ﬁrst n standard orientation on Rn is 1 standard basis vectors e1 . . Since y v 1w n 1 1e n. . e n 1 v by Exercise 8. . en . . . . vn 1 w be a positively oriented basis of T v v xM v . The unit normal enables one to distinguish between two sides of M: the direction of n is “out” or “up”. . is a line. e2 . which is often called the Gauß map of M. There are therefore precisely two vectors of length 1 which are perpendicular to Tx M. en 1. Let v v1 .

VOLUME FORMS φ 1 c is a hypersurface in Rn (if nonempty). and also that Tx M ker Dφx c grad φ x .96 8.

For any oriented n-manifold M in R N the volume form µ M is a well-deﬁned n-form. is the n-form on M whose local representative relative to the embedding ψi is deﬁned by µi ψi µ . The regular value theorem tells us that M 8. Don’t be misled by this old-fashioned notation: volume forms are seldom exact! The volume form µ M is highly dependent .6 we conclude that M is orientable. QED P ROOF. . Taking φ x % x 2 and c r2 we obtain that the n sphere of radius r about the origin is orientable. and is therefore orientable. P ROPOSITION . The volume form µ M . Let U be open in Rn and let φ : U R be a smooth function. and dV . the Jacobi matrix of ψi at t. Choose a collection of embeddings ψi : Ui R N with Ui open in Rn such that M i ψi Ui and such that Dψi t : Rn Tx M is orientation-preserving for all t Ui . QED For n 1 the volume form is usually called the element of arc length. So let us put φ ψ i 1 ψ j and substitute t φ u into µi . therefore deﬁnes a unit normal vector ﬁeld on M. P ROOF. The function n x { grad φ x © grad φ x . To show that µ is well-deﬁned we need to check that its local representatives satisfy the transformation law (7. T HEOREM .13 we have φ dt1 dt2 dtn & φ µi det Dφ u du1 du2 dun ! det Dφ u . Appealing to Proposition 8. 8. Since each of the embeddings ψ i is orientation-preserving. The unit normal is 1 - n x grad φ x © grad φ x . 8. Volume forms x x . dA. for n 2. Let c be a regular value of φ. the element of surface area.1).9.10. we have det Dφ 0. 8. By Theorem 8. Hence you should think of µ as measuring the volume of inﬁnitesimal blocks inside M. det Dψi t T Dψi t © dt1 dt2 dtn . the volume element. du1 du2 dun det Dφ u T det Dψi φ u © T Dψi φ u det Dφ u du1 du2 det Dψi φ u © Dφ u © T Dψi φ u © Dφ u du1 du2 dun det © Dψ j u © T Dψ j u © du1 du2 dun µ j . du1 du2 dun . also denoted by µ . Traditionally these are denoted by ds. det Dψi φ u © T Dψi φ u © dun where in the second to last identity we applied the chain rule.3. Therefore det Dφ u. Hence by Theorem 3.8. and for n 3.3 the square-root factor measures the volume of the n-dimensional block in the tangent space Tx M spanned by the columns of Dψ i t . respectively. E XAMPLE . Then the manifold φ 1 c has a unit normal vector ﬁeld given by n x & grad φ x © grad φ x . Now let M be an oriented n-manifold in R N .

7 M is a 1-manifold in R 2 . It is perhaps more intuitive. . . Let M ¥ R2 be the graph of f . vn are positively oriented. vn are linearly dependent. i. . Let M be an oriented n-manifold in R N . Let I be an interval in the real line and f : I R a smooth function. . Then det ¡ Dψ T Dψ ¢£ 1. v2 .x ¡ v1 . Let x v2 . By Example 6. µ M . . . vn at x let ωx ¡ v1 . . . f ¡ t ¢©¢ . Dψ ¡ t ¢ en ¢ . . . µ M. v2 . . 8. We have ψ ¦ ω £ g dt1 dt2 ¤¤¤ dtn for some function g. This shows that g £ ³ det ¡ Dψ T Dψ ¢ and therefore ψ¦ ω Thus ψ ¦ ω is equal to the local representative of µ M with respect to the embedding ψ. Dψ ¡ t ¢ e2 . . . . v2 . vn . . where in the second equality we used the deﬁnition of pullback. so by Theorem 8. so the volume form on U is simply dt1 dt2 ¤¤¤ dtn . Let us give M the orientation induced by the embedding ψ. . Let us calculate the n-form ψ ¦ ω on U .12 this function is given by g ¡ t ¢£ ψ ¦ ωx ¡ e1 . Dψ ¡ t ¢ en are a positively oriented basis of Tx M and.e. . vn are negatively oriented. i. For each x in M and n-tuple of tangent vectors v1 .2. Let U be an open subset of Rn and ψ : U R N an orientation-preserving embedding with ψ ¡ U ¢»¥ M and ψ ¡ t ¢»£ x for some t in U . QED £5¼ det ¡ Dψ T Dψ ¢ dt1 dt2 ¤¤¤ dtn .13. . We have 1 1 £ 1 ² f ª ¡ t ¢ 2. . . v2 . we have ω £ µ M . . . Since this holds for all embeddings ψ. Let U be an open subset of Rn . ψ ¡ x ¢£ x. . . . a 1-form on I . . v n ¢ ¸ £ µ· º ¶ ·¹ voln ¡ v1 . . . Dψ ¡ t ¢ T Dψ ¡ t ¢0£¯® 1 f ª«¡ t ¢±° ¨ f ª«¡ t ¢¬ f ª«¡ t ¢¬ so ψ ¦ µ £5³ det ¡ Dψ ¡ t ¢ T Dψ ¡ t ¢©¢ dt £ ³ 1² f ª ¡ t ¢ 2 dt. e2 . . . Indeed. v2 .3 they span a positive volume of magnitude ³ det ¡ Dψ ¡ t¢ T Dψ ¡ t ¢©¢ . P ROPOSITION . . The next result can be regarded as an alternative deﬁnition of µ M . moreover.8. vn ¢ be the oriented volume of the block spanned by these n vectors.12. .5 that U is a manifold covered by a single embedding. It changes if we dilate or shrink or otherwise deform M. . v2 . . the ordinary volume form on Rn . vn ´ Tx M. . . vn ¢ voln ¡ v1 . Dψ ¡ t ¢ e2 . . v2 . . . vn ¢ 0 if v1 .e. . . “from left to right”. .x ¡ v1 . namely the identity map ψ : U U .3. M is the image of the embedding ψ : I R2 given by ψ ¡ t ¢£5¡ t. . VOLUME FORMS 97 on the embedding of M into R N . . . . E XAMPLE . The vectors Dψ ¡ t ¢ e1 . en ¢&£ ωx ¡ Dψ ¡ t ¢ e1 . 8. . vn ¢ is the oriented volume of the n-dimensional parallelepiped in Tx M spanned by v1 . Dψ ¡ t ¢§£©¨ . P ROOF. . . . . but it requires familiarity with Section 7. . are the columns of the matrix Dψ ¡ t ¢ . if v1 . v2 . Recall from Example 6. if v1 . v2 . Then the volume form of M is given by ´ M and v1 . By Lemma 7. 8. . . . v2 . E XAMPLE .11. . This deﬁnes an n-form ω on M and we must show that ω £ µ M . What is the element of arc length of M? Let us compute the pullback ψ ¦ µ . .

Then. F IRST PROOF.5. It measures the component of F orthogonal to M. VOLUME FORMS Volume form of a hypersurface. en Î 1 É form a positively oriented basis of Tx M. The product È F Ç n É µ M is an n Ê 1-form on M. e2 . On the hypersurface M we have F ÇÌÄ dx ½5È F Ç n É µ M . Let us change the coordinates on R n in such a way that the ﬁrst n Ê 1 standard basis vectors È e1 .98 8. and .e.7. . On the other hand we have the n Ê 1-form ÄËÈ F Ç dx É&½ F ÇÌÄ dx.2. the positive unit normal at x is given by n È x ÉÏ½¯È. .. a smooth map F : M Æ R n . Let n be the positive unit normal vector ﬁeld on M and let F be any vector ﬁeld on M. Let x Í M. according to Example 8. Ã dxn Ä dxn introduced in Section 2. Ã dx ½U¾¿À . i. Ä dx ½Å¾¿À . Then the inner product F Ç n is a function deﬁned on M. 8.. . For oriented hypersurfaces M in R n there is a more convenient expression for the volume form µ M .14. This proof is short but requires familiarity with the material in Section 7. T HEOREM . Recall the vector-valued forms ÁÂ ÁÂ dx1 Ä dx1 .

. . . On the other hand F ½ ∑n n i Ñ 1 Fi ei . en Î 1 É½ 1. we have F È x ÉÒÇ n È x É&½NÈ. .x È e1 . Writing n Ð 1 F È x É .Ê 1 É n Ð 1en and the volume form satisﬁes µ M.

Ê 1 É F ÇÄ dx ½ ∑ È.

en Î 1 É form a positively oriented basis of Tx M. e2 . a preliminary step in the proof we are going to replace the embedding ψ with a new one enjoying a particularly nice property. Then at x the positive unit normal is given by n È x É½NÈ. . . Let us change the coordinates on Rn in such a way that the ﬁrst n Ê 1 standard basis vectors È e1 . .Ê i 1 1É i 1 Ð Ó i Fi dx1 ÇÇÇdx ÇÇÇ dxn .

. Then U ˜ is open in Rn Î 1 . Put U t ½ A Î 1 t and ψ n ˜ É&½ ψ È U É . ˜ ˜ ½ ψ Ù A. invertible. 2. Therefore the È n Ê 1 ÉÏ×ØÈ n Ê 1 É -matrix A with i-th column vector equal to a i is ˜ ½ AÎ 1 È U É . . n Ê 1.Ê 1 É n Ð 1en . a2 . an Î 1 in Rn Î 1 such that Dψ È t É ai ½ ei for i ½ 1. because the ei are independent. . . ψ ψ ˜È ˜ Dψ t É½ ˜È ˜ t É ei Dψ Dψ È tÉÚÙ DA È ˜ t É ½5È. and ˜: U ˜ Æ R is an embedding with ψ ˜È U ˜È ˜ t É&½ x. . there exist unique vectors a1 . . . These vectors ai are independent. . Since the columns of the Jacobi matrix Dψ È t É are independent.

Ê 1 É n Ð 1 Fn . This proves that È F ÇÌÄ dx É x È e1 , . . . , en Î 1 É&½NÈ F È x ÉÔÇ n È x ÉÉ µ M È e1 , . . . , en Î 1 É , which implies È F ÇÕÄ dx É x ½UÈ F È x ÉÇ n È x É©É µ M . Since this equality holds for every x Í M, we ﬁnd F ÇÌÄ dx ½5È F Ç n É µ M . QED S ECOND PROOF. Choose an embedding ψ : U Æ Rn , where U is open in Rn Î 1 , such that ψ È U ÉVÖ M, x Í ψ È U É . Let t Í U be the point satisfying ψ È t ÉV½ x. As

and therefore È F ÇÌÄ dx ÉÌÈ e1 , . . . , en Î

½

Dψ È t ÉÚÙ A

˜È ˜ by the chain rule. Therefore the i-th column vector of Dψ t É is

½

Dψ È t É A e i

½

Dψ È t É a i

½

ei

(8.2)

8.3. VOLUME FORMS

99

for i Û 1, 2, . . . , n Ü 1. (On the left ei denotes the i-th standard basis vector in Rn Ý 1 , on the right it denotes the i-th standard basis vector in Rn .) In other words, ˜ at ˜ t is the Þ n Ü 1 ßà n-matrix the Jacobi matrix of ψ I ˜Þ ˜ Dψ t ß&Û©á n Ý 1 , 0 â where In Ý 1 is the Þ n Ü 1 ßàãÞ n Ü 1 ß identity matrix and 0 denotes a row consisting of n Ü 1 zeros. ˜ äËå

Þ F æ n ß µ M ç and ψ ˜ äèÞ F æèé dx ß at the point ˜ Let us now calculate ψ t. Writing n F æ n Û ∑i ê 1 Fi ni and using the deﬁnition of µ M we get ˜ ä Þ Fi ni ß ˜T ˜ ˜ ˜ ˜ å

So evaluating this expression From formula (8. iê 1 ˜Þ ˜ ˜Þ ˜ t ß T Dψ t ß©ßÛ 1.2) we have det Þ Dψ n ì 1 ˜ at the point t and using n Þ x ß&ÛNÞ.Þ F æ n ß µ M ç Û©á ∑ ψ âaë det Þ Dψ Dψ ß dt1 dt2 æææ dtn Ý 1.

Ü 1 ß en we get ˜ ä Þ F æ n ß µ M ç ÛNÞ.

Ü 1 ß n ì 1 Fn Þ x ß dt ˜1 dt ˜2 æææ dt ˜n Ý 1 . åψ ˜ t ˜ä ψ n From F æé dx Û ∑n iê 1 Þ.

j ð ˜nÞ ˜ ˜j n Ü 1 and ∂ψ t ß©ï ∂t Û 0 ˜1 dt ˜2 æææ dt ˜n Ý 1 . ˜ ä Fi dψ ψ îψ ˜iÞ ˜ ˜j t ß©ï ∂t From formula (8. Therefore ˜ ä Þ F æé åψ dx ß δi.Þ. ç ˜t Û.Ü 1ß i 1F i n ì dx1 dx2 æææ6dx í i æææ 1 dxn we get ˜ ä Þ F æÌé dx ßÛ ψ i 1 ∑ Þ©Ü 1 ß i ì ê Û ˜ 1 dψ ˜ 2 æææd ˜ i æææ dψ ˜ n. j for 1 ð i.2) we see ∂ψ for 1 ð j ð n Ü 1.

Ü 1 ß n ì 1 Fn Þ x ß dt ˜ ä1Þ F æ n ß µ M ç Ûcå ψ ˜ äÕÞ F æé dx ß ç . in other words å.

14.15. Think of the vector ﬁeld F as representing the ﬂow of a ﬂuid or gas. We call F æé dx the ﬂux of the vector ﬁeld F. QED This theorem gives insight into the physical interpretation of n Ü 1-forms. Since this holds for all x ñ M we have F æé dx Û5Þ F æ n ß µ M . Set F Û n æÌé dx. Then µM P ROOF. Then F æ n Û 1 because ò n ò»Û 1. Û n in Proposition 8. Another application of the theorem is the following formula for the volume form on a hypersurface.Þ F æ n ß µ M ç Û We conclude that å ψ ˜ ˜ x t t Þ F æÌé dx ß x. for any unit vector n in Rn . C OROLLARY. the amount of ﬂuid per unit of time passing through a hyperplane of unit volume perpendicular to n.é dx measures. 8. The formula provides a heuristic interpretation of the vector-valued form é dx: if n is a unit vector in Rn . The direction of the vector F indicates the direction of the ﬂow and its magnitude measures the strength of the ﬂow. QED . Let n be the unit normal vector ﬁeld and µ M the volume form of an oriented hypersurface M in Rn . Then Theorem 8. then the scalar-valued n Ü 1form n æèé dx measures the volume of an inﬁnitesimal n Ü 1-dimensional parallelepiped perpendicular to n.14 says that the n Ü 1-form F æ.

. v ¡ ý vol u . u . E XAMPLE . a . . a be real numbers. . v . u ý ý ý ¥ c . then n ó x ô&õ x ÷ R. Then by Proposition 8. . so µ M õ R ù 1 x úû dx. v . Exercises 8. u .4. . . . Show that ü a ürü b ü sin φ ýþü a ÿ b ü in R 3 . u . . let c ý§¦ 1 ¨ ∑ © a and let 1 0 0 a 0 1 0 a 1 .2. . VOLUME FORMS 8. .3 that the area of the parallelogram spanned by a pair of vectors a. 8.1. 8. . . v . . u . .16. . . b in Rn is given by ü a ü¢ü b ü sin φ. Deduce from Theorem 8. u ¡ vol v . where φ is the angle between a and b (which is taken to lie between 0 and π ). Let a . .1. . In particular. . with U open in R n .ù 1 grad φ ú.100 8. . . u ý . ý ¢ ¡£ det A T A satisﬁes the axioms of ¡ ¥ u . . . Check that the function voln a1 . .8 M has a unit normal n õ grad φ ÷ø grad φ ø . The volume form is therefore µ õnø grad φ ø. a2 . . Suppose the hypersurface M is given by an equation φ ó x ôsõ c.û dx. . . an Deﬁnition 8. where c is a regular value of a function φ : U ö R. . . . . if M is the sphere of radius R about the origin in Rn . . u . . . v ¡ . .

. 0.

0. .

. 1.

volk l 1 2 k 1 2 l k 1 2 k l 1 2 l 1 2 n n 2 i 1 i 1 2 1 2 n n 1 n 1 2 n n 1 8. 1 ¨ a 1 1 2 n n 1 1 n 2 n 3 n n i 1 2 i 2 n 8. un (ii) Prove that 1 ¨ ¡ ý voln a aa a a 2 1 2 1 . .. v . . (ii) v .a 1 a a a ¥ be vectors in R .. v ý v . a a . .3. v2 . Let U be open in R and let f : U R be a smooth function. Let M ý graph f be the oriented hypersurface of Exercise 8. f x ¡¡ and let M ý ψ U ¡ . . . .. then v . . uk and v1 . the graph of f .. .4 that the volume form of M is given by ψ µ ý ¢ 1 ¨ ü grad f x ¡ ü dx dx ¤¤ ¤ dx . σ 1 σ 2 2 σ n 1 2 n 1 2 1 1 2 1 2 n n 1 M 2 1 2 n 8. .. . k and j ý 1.. . .”) Prove that ¤ 0 for i ý 1. . u2 . . Let u1 . ∑ .7. 2. 8. . Deduce from Exercise 8. .. a a ý 1¨ a . (i) If σ Sn is any permutation.6. Let ψ : U R ¥ be the embedding ψ x ¡ ý x. v . . . . (iii) 3v . vl be vectors in R N satisfying ui v j ý 2. © . Here the v’s form a basis of V (which in part (i) is n-dimensional and in parts (ii)–(iii) twodimensional). . v . . . (“The u’s are perpendicular to the v’s. (i) Deduce from Exercise 8. . Justify the following identities concerning orientations of a vector space V . Deﬁne an orientation on M by requiring ψ to be orientation-preserving. . u2 .6.. v . .3 that voln u1 . . an a1 3 1 a1 a2 1 a2 2 a3 a2 . . v ý sign σ ¡ v . . . .. 5v ý v . an a2 ¨ a1 a3 a2 a3 1 a2 3 . .u .u .5. .. . . a a . . v . an a3 ¨ ¥ u . . .. . .u ¥ ¡ . ... . l .

∂x // + .) 8. Show that the unit normal vector ﬁeld n : M 5 R deﬁned in the proof of Proposition 8. b $ given by ( ψ 7 t $3( dt ! 8 " " " ψ 7 t $ ' ψ 7 t $ ' 4 44 ' ψ 7 t $ dt. b $65 R be an embedding. " " " (i) Show that the positive unit normal vector ﬁeld on M is given by 1 2 n 1 2 n M 2 1 2 n n 1 1 2 n n n 1 2 2 2 n 2 . (Compute n in terms of an orientation-preserving parametrization ψ : U 5 M of an open subset of M.15. (Caution:4 44for consistency you 8.∂x . Let ψ : a. b $ .9.. .// ++ ∂ f . . . " ∂ f . Show that ψ 1 µ is the 1-form on a. x" $ .15 by substituting x % must replace n with n ' 1 in Corollary 8. % 1$ n! & " # 1 ')( grad f x $( .6 is smooth.EXERCISES 101 *++ ∂ f .8. Let µ be the element of arc length on the " embedded curve M ! ψ a.∂x 0 1 & (ii) Derive the formula ψ 1 µ ! 1 '2( grad f x $3( dx dx dx from Corollary ! f " x . x . / .) 8.. .

.

In particular. J = C ? 9 A : < K R be a 9. ∂Hn 9 : x < R = x > 0? . If nonempty. and an embedding ψ : U R N satisfying ψ U E < E You should compare this deﬁnition carefully with Deﬁnition 6. 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. We will also discuss a version of Stokes’ theorem for manifolds. If x ψ t with t ∂Hn . Its complement M ∂ M is the interior of M and is denoted by int M. which has boundary ∂Hn Rn 1 and interior the open halfspace x n R xn 0 .4 of a manifold. The most obvious example of an n-manifold with boundary is the halfspace Hn itself. the boundary ∂ M is an n 1-dimensional manifold. The n-dimensional halfspace is Hn The boundary of is int Hn x Rn x n 9B: < D D D Hn = C 0? . by integrating the volume form we ﬁnd the volume of the manifold.CHAPTER 9 Integration and Stokes’ theorem on manifolds In this chapter we will see how to integrate an n-form over an oriented nmanifold. Put U U R and write elements of U as x y with x in U and 9 K KNM A OP KLF 103 .1. This requires the slightly more general notion of a manifold with boundary. Manifolds with boundary The notion of a spherical earth developed in classical Greece around the time of Plato and Aristotle. Here is a more interesting type of example.2. Let U be an open subset of Rn 1 and let f : U smooth function. Rather.1. E XAMPLE . The boundary of M is the set of all boundary points and is denoted by ∂ M. < J Somewhat confusingly. because no neighbourhood of a point on the edge is the image of an open subset of R 2 under an embedding. 9. D EFINITION . A closed disc is not a manifold. the boundary of a manifold with boundary is allowed to be empty. a notion which can be deﬁned as follows. Older cultures (and also Western culture until the rediscovery of Greek astronomy in the late Middle Ages) visualized the earth as a ﬂat disc surrounded by an ocean or a void. 9 GI F G H Hn I 9 VH M. which generalizes the graph of a function. an open subset U Rn . Likewise the interior int M is an n-manifold. then x is a boundary point of M. 9@: x < R = x 9 0 ? 9 n n n n Rn A 1 and its interior is 9. . it is a manifold with boundary.

Let N n g m \ 1 and think of R as the set of vectors Q R with x in R ^ and Z y in R . Deﬁne ψ : U W R by t ψX X f T tU]\ t ue Y u Y[Z Dψ n 1 n 1 n n n x y t u n n n As in Example 6. The region below the graph of f is the set consisting of all y f x .e. Since H is given by u a 0. XAMPLE m N m N x y n 1 m This time we have T U Z X DI f^ T tUh\ 0 e Y and again ψ is an embedding. Therefore M ψ T U ` H U is an n-manifold in R Z with boundary ∂ M ψ T U ` ∂H U . this is equivalent to x b U c and y S f T x U .104 9. ∂ M is the graph. 9. where 0 is the origin in R ^ . the set M Z ψ T U ` H U is an n-manifold in R with boundary ∂ M ψ T U ` ∂H U . On ∂H we have u 0. We will prove this by describing it as the image of a single Rn by embedding.3. so ∂ M is given by the equality y Z Z f T xU .3 one veriﬁes that ψ is an embedding. E . using the fact that t t Xu Y[Z X f T tU]\ uY . By deﬁnition therefore. y ∂ M graph f S T U QR x y in U such that V M x We assert that the region below the graph is an n-manifold whose boundary is exactly the graph of f . Deﬁne ψ : U W ψ 0 t I Xu Y[Z X D f^ T tU_\ 1Y . Thus M is exactly the region below the graph. If f : U cdW R is a vector-valued map one cannot speak about the region “below” the graph. but one can do the following. INTEGRATION AND STOKES’ THEOREM ON MANIFOLDS y in R. i. Again put U U cfe Z R. What are M and ∂ M? A Z point Q R is in M if and only if it is of the form t t ψ X Xx X f T tU] y Y[Z u Y[Z \ uY for some Q R in U ` H . This time M is the set of points Q R of the form Z x t Xy Y Z X f T tU]\ ue Y Dψ t n n 1 m n N x y m .

Other examples of manifolds with boundary are the pair of pants. MANIFOLDS WITH BOUNDARY 105 with t U and u 0.10. x y m 1 m m Here is an extension of the regular value theorem. Let U o R . . m o 1 and φ p x q owv x v .9. Similarly. a 2-manifold whose boundary consists of three closed curves. n k y2 2 m 1 l m where x is in U j and where r o f r p xq . For instance. a closed square is not a manifold with boundary. Let U be open in R N and let φ : U Rm be a smooth map. .. . The Möbius band is a nonorientable manifold with boundary. . c . . y s v xv s R2 z and the Möbius band shown in Chapter 1. E . . one can show that a set given by a pair of inequalities of the form a f x b. φm r p xq o 1 cm r 1. The set given by the inequality φ p x q s 1 is then the closed unit ball x x i R y v x v s 1 z . φ1 x c1 . . 1 o f p xq . y Again ∂ M is given by y o f p x q . which we will not spell out. is a manifold with boundary. We will not give a proof of this fact.10. T HEOREM (regular value theorem for manifolds with boundary). 9. However. so ∂ M is the graph of f . Hence the ball is an n-manifold in R . is similar to that of Theorem 6. Hence M is the set of points y satisﬁes m 1 equalities and one inequality: y1 i j o f p xq . .1. Theorem 6. This is not a manifold with boundary because its edge has a sharp angle at the origin.5.. c q is a regular value of φ and that M is nonempty. p qo t φ2 x p qo c2 .4. The proof. A simple example is the closed quadrant in R2 given by the pair of inequalities x 0 and y 0.. where a and b are both regular values of a function f . the unit sphere S r . 9. y s f p xq . the spherical shell k k s p qs x xi Rn R 1 is an n-manifold whose boundary is a union of two concentric spheres. Since grad φ p x q o 2x. Suppose that c oup c . whose boundary is φ r p 1 q . Then M is a manifold in R of codimension m n 1 and with boundary ∂ M o φ r p c q . Let M be the set of x in R N satisfying p qs c. to manifolds with boundary. . singularities often arise. but you can convince . any nonzero value is a regular value of φ. φm x m 1 2 N m 1 XAMPLE n 2 n n 1 n 1 If more than one inequality is involved.

e2 . v2 . . 9. This assumption is made to ensure that the integral is a proper integral and therefore converges. vn 1 of Tx ∂ M is called positively oriented if n x . .5) and the torus S S (see Exercise 9. . Integration over orientable manifolds As we saw in Chapter 5. A basis v1 . . 1 } is an orientation-preserving embedding. see Appendix A. v1 . then v is tangent to the boundary. In other words. If un 0.6. We deﬁne α{ c α. 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. Then v Dψ t u for some u Rn . . To deﬁne the integral of α over M let us assume that M is compact. vn 1 is a positively oriented basis of Tx M. | }~| } As in the case of a manifold. e {| 1 } e . Namely. This opens up the possibility of integrating an n-form over an oriented n-manifold. For instance. en . this does not depend on the choice of the embedding ψ. . . 1 n R N such that n n c 0. let M Hn with the standard orientation e1 . . let us also make the assumption that there exists a smooth map c : 0. e . . For instance. INTEGRATION AND STOKES’ THEOREM ON MANIFOLDS yourself that it is true by trying to paint the two sides of a Möbius band in different colours.2. The integral does not change if we reparametrize the chain in an orientation-preserving manner. At each point of ∂ M Rn 1 the outward pointing normal is en . . called the induced orientation. . . An n-manifold with boundary contained in Rn (i. . v2 . The pullback c α is then an n-form on the cube 0. { |} { Tx ∂ M { Dφ t Rn | }~| } . This deﬁnes an orientation of ∂ M. Orienting the boundary. 1 The above picture of the pair of pants shows some tangent vectors at boundary points that are tangent to the boundary or outward-pointing. . e n 1 2 n 1 n 1 2 { | | } } | } | } n 1 .e. .4).2. . . because | } { e . . n 1 1 n M 0. This implies that induced orientation on ∂ M is 1 n e1 . Let M be an oriented manifold with boundary. .1 n . of codimension 0) is often called a domain. 1 { M and the restriction of c to | 0. . for x ∂ M deﬁne n x outward-pointing tangent vector of length 1 which is orthogonal to Tx ∂ M. en . . a closed ball is a domain in R n . this assumption is satisﬁed for the n-sphere S (see Exercise 5. a form of degree n can be integrated over a chain of dimension n. e .) For a start. Now suppose x is a boundary point of M and let v Tx M be a tangent vector. For instance. To deﬁne the tangent space to a manifold with boundary M at a point x choose U and ψ as in the deﬁnition and put Tx M { Dψ t R n . . (A subset of R N is called compact if it is closed and bounded. . .106 9. This deﬁnes the unit outward-pointing normal vector ﬁeld on ∂ M. Let M be an n-dimensional oriented manifold (possibly with boundary) in R N and let α be an n-form on M. We say that v points inwards if u n 0 and outwards if un 0. en 1 . 1 . e .

QED Not every manifold can be covered with one single n-cube. The mean or average of f is the number f¯ vol M 1 M f µ . . The volume of M is vol M 1.1. so the notions deﬁned above depend on the embedding as well. n i ª ¡ ¯i x 1 vol M M xiµ . where µ is the volume form on M. This implies that c : V ¯ ¯ is a bijection from V M is a bijection onto its image. (If dim M speak of the arc length.e. ¡ £ c α c ¡ £ c¯ c α ¤ c £ c ¡ £ c¯ α c¯ α. 1 n is empty for i j.6.7. 0. However.1) By assumption the restriction of c to U is an embedding. Let M a compact oriented manifold in R N . resp. ¦ k c α. which can be viewed as a parametrization-independent version of Theorem 5. . (i) k i 1 ci 0. ¥ ¦ §¨ ¨ ¨ § § 0.2. 1 n is an orientation-preserving embedding.1) we get the result. (The condition (ii) on the maps is imposed to avoid “double counting” in the integral. surface area of M. 1 . L EMMA . The most important property of the integral is the following version of Stokes’ theorem. (iii) for each i the restriction of c i to 0. 2. . 2.1 n 9. n ¯ in R whose i-th coordinate is the The centroid or barycentre of M is the point x mean value of xi over M. k. . It is orientation-preserving. complement of V and of V in R are negligeable in the sense that c α c¯ α. 1 n Suppose c R N is a smooth map with the same properties as c. The volume form depends on the embedding of M into R N . and so we see that c 1 c ¯ are orientation-preserving. we M µ . i. INTEGRATION OVER ORIENTABLE MANIFOLDS 107 0. resp. 1 1 V ¯ V ¯ V ¯ V Combining this with the equalities (9.6 that the result does not depend on the maps c i . Let us denote the closed cube 0. 1 c j 0. R ¯ V (9.1 © We can then deﬁne M α i 1 ∑ and check as in Lemma 9. such that n M. 1 n M for i 1. by Theorem 5. The the open cube 0.) 9. To ensure that M α is well-deﬁned we need to check the following equality.1 n KETCH OF PROOF n n 1 1 c α ¢ c α R V and c¯ α c¯ α. Let V U c ¡ c ¯ 9. it can be shown that there always exists a ﬁnite collection of n-cubes c i : 0. onto V . 1 n (ii) ci 0.) The integral of a function f on M is deﬁned as M f µ .¯ : 0. 1 by R and let U be ¯ U c ¯ U and V ¯¡ c U . . because c and c Therefore. S .10 and is proved in a similar way. D EFINITION .

5.3. For this reason the n 1-form α is also called the ﬂux of F. « If we interpret F as a force acting on a particle travelling along M. These classical results involve a vector ﬁeld F ∑n i 1 Fi ei den ﬁned on an open subset U of R . We will now derive the classical integral theorems by applying Theorem 9.8. Let α be an n 1-form on a compact oriented n-manifold with boundary M.8. ¯ ® ® Fundamental theorem of calculus. where the direction of F x gives the direction of the ﬂow at a point x and the magnitude F x gives the mass of the amount of ﬂuid passing per unit time through a hypersurface of unit area placed at x perpendicular to the vector F x . INTEGRATION AND STOKES’ THEOREM ON MANIFOLDS 9. If F is conservative. ³ ² ³ ¸ ² ³¹¸ ² ³ µ ·² ¯ « ° ¬ M div F dx1 dx2 ¯¶¯¶¯ dx n ¬ ° µ µ ∂M ² F ¯ n³ µ ∂M. Thus div F 0 for an incompressible ﬂuid without sources µ . Then α describes the amount of ﬂuid passing per unit time and per unit area through the hypersurface N . resp. Then ¬ M dα ¬ « ∂M α. Thus the total ﬂux out of the hypersurface ∂ M is the integral of div F over M.8. n-dimensional. T HEOREM (Stokes’ theorem for manifolds). Theorem 9. to this vector ﬁeld corresponds a 1-form α F dx ∑n i 1 Fi dx i .14. then α F n µ N on N by Theorem 8. most liquids) then this formula leads to the interpretation of the divergence of F (or equivalently d α ) as a measure of the sources or sinks of the ﬂow. ¯ ° ±° ¬ M F dx ¯ g ² b³´« g ² a³ . Applying Stokes’ theorem to a compact domain M in Rn we get M d α ∂ M α . Gauß’ divergence theorem. and its integral over N the total ﬂux through N . If the ﬂuid is incompressible (e. As discussed in Section 2. then α grad g dx dg. We have µ α F ¯ µ dx µ and n d α µ div F dx 1 dx2 ¯¶¯¶¯ dx . n If N is a oriented hypersurface in R with positive unit normal n. Stokes’ theorem therefore gives the fundamental theorem of calculus in Rn . which we can think of as the work done by the force F along an inﬁnitesimal line segment dx.8 to one-dimensional. then M dg ∂ M g by Theorem 9. two-dimensional manifolds M contained in U .g. 9. If the orientation of M is “from a to b”. Give the boundary ∂ M the induced orientation. resp. F grad g for a function g. If M is a compact oriented 1-manifold with boundary in Rn . then g stands for the potential energy of the particle in the force ﬁeld. then a acquires a minus and b a plus. The boundary consists of two points a and b (if M is connected). In this situation it is best to think of F as the ﬂow vector ﬁeld of a ﬂuid. Thus the potential energy of the particle decreases by the amount of work done.108 9. contains as special cases the integral theorems of vector calculus. Gauß and Stokes Stokes’ theorem. Written in terms of the vector ﬁeld F this is Gauß’ divergence theorem.

yÅÎÊ 2 Ì x Ì y . The right-hand side represents the work of F done around the boundary curve(s) of M. being compressed) at x. x3 be the standard coordinates on R3 . y such that x in U and satisfying f x f x y g x . Write the divergence theorem for the vector ﬁeld F Ê Ì cx e on R . (i) Sketch the image of c. c dx2 .) 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. (ii) Let x 1 . Let M be the set of all pairs x. In this way we get the classical formula of Stokes. can be regarded as a measure of the vorticity of the vector ﬁeld. This formula shows that curl F. If the ﬂuid is a gas and if there are no sources or sinks then div F x (resp. What is the value of the constant? 9. ¾ ¾ Á ¿ ¿uº Á3À Á » Â Â º curl F Á n » µ ¿ M M ¿ À ¿ ∂M F dx. Ó R2 Ó 0 be constants. Exercises 1 and f x. y Ê sin t. (iv) Find the surface area of the boundary of this solid. ½ º »¼ ¿ 0 Classical version of Stokes’ theorem. the total ﬂux of curl F through the surface M is equal to the work done by F around the boundary curves of M. Now let M be a compact two-dimensional oriented surface with boundary and let us rewrite Stokes’ theorem M dα ∂ M α in terms of the vector ﬁeld F. y Ä ÅËÊ a positive constant. R 2 c ×Ø Å r θ1 θ2 ÙÚ Ä R ÉÉ ÊÛ×Ø Ä R 1 1 r cos θ2 cos θ1 r cos θ2 sin θ1 r sin θ2 Å Ô Å ÙÚ ÕNÖ Ô 0.1. which is not necessarily 0 if F is not conservative. 2π Õ Ã Ü Ü R3 . 2π ÕÖ Ô 0. Show that Ï α is twice the surface area of M.4. 0) indicates that the gas is expanding (resp. or equivalently dα . À R be two smooth functions 9. Hence if n is the positive unit normal of the surface M in R3 . (Use two embeddings to cover M. Let R1 Ä´ ÅÈ È ÄÅ 9.EXERCISES 109 or sinks. Show that Ï α is a constant times the volume of M. Let U be an open subset of Rn and let f . Á In other words. (iii) Find the volume of the solid parametrized by c. ÜÄ Ü . Then dα curl F dx. Compute c dx1 . Eυρηκα ´ ! by 9. Deduce Archimedes’ Law: the buoyant force exerted on a submerged body is equal to the weight of the displaced ﬂuid. g : U g x for all x in U .3. x2 . 2 2 2 2 2 2 Ã ÄÇ ÅÆ ÄÅ Ä Å Ä´ ÅÈ È ÄÅ (i) Draw a picture of M if U is the open unit disc given by x É y Æ ÌÍ 1 Ì x Ì y and g Ä x. (ii) Apply the observation of part (i) to ﬁnd the area enclosed by the astroid x Ê cos t. then dα curl F n µ M on M. where c is 2 ∂M 3 3 n ∂M n n n (ii) Show directly from the deﬁnition that M is a manifold with boundary. c dx3 and c dx1 dx2 dx3 . (i) Let α Ê x dy Ì y dx and let M be a compact domain in the plane R . Deﬁne a 3-cube c : 0.2. The left-hand side has a nice interpretation if n 3. so dα curl F dx. (iii) Let α Ê x ÐÒÑ dx and let M be a compact domain in R .

Let B R be the closed ball of radius R about the origin in R n .6.15 that the volume form on S R is the restriction of ν to S R . INTEGRATION AND STOKES’ THEOREM ON MANIFOLDS 9. the volume of a zero-dimensional manifold is its number of points. Also put Vn Vn 1 and An An 1 .) (Take f Ý r ÞËß e ð R R B R 1 2 n 0 n n n 1 0 r2 n n r2 n 1 0 r2 ß ÝÞ Ý Þäß Ý ÞËß ð Ý Þ Ý Þúß An (vii) Deduce that Vn 2π 2 ¥ π . (Conventions: a space of negative dimension is empty. g dx 1 dx2 dxn is M grad f grad g µ . (i) Show that d f dg grad f grad g µ . 5 ááá Ý 2m ê 1 Þ m 1 m Vn R An R Ý Þ Ý Þ 0 1 2 2π R 3 4 3 3 πR 4 5 π R2 .11 conclude that ß ¡£¢ 2π 2 n 2 n . ∂n ∂M ∂M ∂M ∂M ∂M M ß áá á é ê æ Þ (vi) Deduce Green’s symmetric formula. (Substitute y Rx in the (ii) Show that Vn R volume forms of B R and S R .) (iii) Let f : 0. òó (iv) Show that ý ëÿþ ëôþ ð ð ð ß e A dr dr.) n ê n 1¤ . ∂M ë 9. (iii) Deduce from parts (i)–(ii) that d f dg grad f grad g f g µ . Write ∂ g ∂n for the directional derivative Dg n grad g n. Put Vn R voln B R and An R voln 1 S R . Then its boundary S R ∂ B R is the sphere of radius R. Deﬁne g : R n R by g x f x . ë ∂g ë µ D Ý f . Rn Vn and An R Rn 1 An . In this problem we will calculate the volume of a ball and a sphere in Euclidean space.10 and B. whence V2m ß πm m! and V2m ¥ 1 ¥ π ß 1á 3á 2 . Let f and g be smooth functions on M. g Þê f ß Ý f æ gÞ µ. and A ¥ ß ß m 1 ! 1 3 5 áá áÝ 2m 1 Þ Ý Þ á á ¤ î î Þ ß 1 in part (iii) show that A ß nV and A Ý R ÞËß ∂V Ý RÞ é ∂R.16. Let M be a compact domain in Rn . where g ∑n i 1 ∂ g ∂ xi . 2 2 (ii) Show that d dg g µ . Conclude that A n R S R ν. (vi) By taking f Ý r Ë (v) Using Exercises B. Show that Ý ] Þ ß§à Ý á Þ ãÝ â Þäß§Ý á Þ â ßåÝçæ Þ æ ß è é Ý Ýãâ ÞÞËßå Ý á ë Ý Þ ß ë ∂gá f Ýãâ dg Þìß f µ .16(ii) to prove that Ý Þ Ý Þäß Ý Þ ß ÝÞ Ý Þ í f ∂g ∂n î g ∂f µ ∂n ∂ M ï ß ë M Ý f æ g î gæ f Þ µ. Use Exercise 2. whence A2m m m 1 m 2m 1 n n n n ß ¡£¢ π2 n 2 (viii) Complete the following table. where µ the volume form on M. R be a continuous function. Ý Þ Ý Þ Ý ÞËßñà ò ó Ý Þôß Ý Þôß ð ß Ý Þ Ý Þ õ ö÷Þùø ø Ýüû ë û Þ ë ë g dx dx ááá dx ß f Ý r Þ A Ý r Þ dr ß A f Ý r Þ r ð dr. (iv) Let n be the outward-pointing unit normal vector ﬁeld on ∂ M. where ν is as in Exercise 2. ∂n (v) Deduce from parts (iii) and (iv) Green’s formula.5.110 9. The Dirichlet integral of f and g is D f . (i) Deduce from Corollary 8. e r ðþ in part (iii) and let R ø ö .

EXERCISES 111 1 (ix) Find limn ¦¨§ An . limn ¦¨§ Vn and limn ¦¨§ © An x lim ¦§ © x xx 1 1 2 ex .

Use Stirling’s formula. . An . 2π .

.

T HEOREM (Brouwer’s ﬁxed point theorem). let M be the punctured unit ball in n-space. T HEOREM . Therefore φ does not exist. Note that β is an n & 1-form on the n & 1-manifold ∂ M. For instance. Let n denote the dimension of M. so dβ 0. Let α φ % β be its pullback to M. This halﬂine intersects the unit sphere ∂ M in a unique point that we shall call 113 . Let us choose an orientation of M and equip ∂ M with the induced orientation. For each x in the ball consider the halﬂine starting at f x and pointing in the direction of x. possibly with boundary. Every smooth map from the closed unit ball into itself has at least one ﬁxed point. Then the normalization map φ x " x #$ x is a retraction of M onto its boundary A ∂ M. This brings us to one of the oldest results in topology. The following theorem says that a retraction onto the boundary is impossible if M is compact and orientable. Let M be a compact orientable manifold with nonempty boundary. the unit sphere. P ROOF. Then f x / + x for all x.CHAPTER 10 Applications to topology 10. But φ is a retraction onto ∂ M. Thus 0 )* ∂M α )* ∂M β vol ∂ M + 0. Let M - x Rn x . 10.1. so the restriction of φ to ∂ M is the identity map and therefore α β on ∂ M. Suppose f : M M was a smooth map without ﬁxed points. Suppose f is a map from a set X into itself.2. Brouwer’s ﬁxed point theorem Let M be a manifold. Then there does not exist a retraction from M onto ∂ M. Therefore dα dφ % β φ % dβ 0 and hence by Stokes’ theorem 0 (' M dα )' ∂ M α . 10. M x Rn 0 x 1! . An element x of X is a ﬁxed point of f if f x . Suppose φ : M ∂ M was a retraction. x. A retraction of M onto a subset A is a smooth map φ : M A such that φ x x for all x in A. 1 ! be the closed unit ball.1. P ROOF. Let β µ ∂ M be the volume form on the boundary (relative to some embedding of M into R N ). QED which is a contradiction.

as in the following picture. φ 0 x. More formally. t 1. f 2 x3 φ 2 y3 y f 2 y3 x φ 2 x3 This deﬁnes a smooth map φ : M 4 ∂ M. A homotopy is given for instance by φ 0 x. A homotopy is given by φ 0 x. which contradicts Theorem 10. APPLICATIONS TO TOPOLOGY φ 0 x 1 . 10. t 1. (See Section 10. .2 for the deﬁnition of homotopy. 0 1"5 φ0 0 x 1 and φ 0 x.2.3.5 x. we say that φ 0 and φ1 are homotopic if there exists a smooth map φ : M 78 0. QED This theorem can be stated imprecisely as saying that after you stir a cup of coffee.5 φ 1 0 x 1 for all x in M. then φ 0 x 1"5 x. Brouwer originally stated his result for arbitrary continuous maps. Then φ 0 and φ1 are homotopic. Homotopy Deﬁnition and ﬁrst examples.2 by an argument from analysis which shows that every continuous map is homotopic to a smooth map. This more general statement can be derived from Theorem 10. t 1/5F0 1 = t 1 x G tx C$D x D . or as a one-second “movie” that at time 0 starts at φ 0 and at time 1 ends up at φ 1 . Let M 5 N 5 Rn and φ0 0 x 1<5 x (identity map) and φ 1 0 x 1. t 1E5 x C$D x D t or by φ 0 x. at least one molecule must return to its original position. then we ﬁnd a homotopy by reversing time (playing the movie backwards). t 1 we often write φ t 0 x 1 . Then each φ t is a map from M to N and we can think of φt as a family of maps parametrized by t in the unit interval that interpolates between φ0 and φ1 . There are other ways to accomplish this. 10. so φ is a retraction of the ball onto its boundary.4.5 tx. If x is in the unit sphere. Suppose that φ 0 and φ1 are two maps from a manifold M to a manifold N and that α is a form on N . We can also interchange φ 0 and φ1 : if φ0 0 x 1<5 0 and φ1 0 x 1.) The theorem also remains valid if the closed ball is replaced by a closed cube or a similar shape. For instance 0 1 = t 1 2 x and 0 1 = t2 1 x are two other homotopies between the same maps. 10. 1 1. Instead of φ 0 x.1. Therefore f must have a ﬁxed point. E XAMPLE . 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. Then φ 0 and φ1 are homotopic. Let M 5 N be the punctured Euclidean space R n =?> 0 @ and let φ0 0 x 1A5 x (identity map) and φ 1 0 x 1B5 x C$D x D (normalization map). 1 9:4 N such that φ 0 x.5 0 (constant map). The map φ is called a homotopy. E XAMPLE . This homotopy collapses Euclidean space onto the origin by moving each point radially inward. What is the relationship 6 α and φ1 6 α ? There is a reasonable answer to this question between the pullbacks φ 0 if φ0 can be smoothly deformed into φ 1 .114 10.5 0 1 = t 1 x.

(Perhaps “expansion” would be a more accurate term. 1 O$P M from φ 0 to φ1 is a contraction of M onto x0 . HOMOTOPY 115 10. (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. Homotopy of curves. A homotopy of loops φ : S1 LM 0. A loop in a manifold N is a smooth map from the unit circle S1 into N .3 shows that R n is contractible onto the origin. We shall see in Theorem 10. φt H x IZJ x V[W 2 0 1 0Y punctured plane R2 Q. A homotopy of curves can be visualized as a piece of string moving through the manifold N . b O and N any manifold. then maps from M to N are nothing but parametrized curves in N . so .) Example 10.) In fact.2. What if we regard φ and φ as loops in the φ0 to the right. a “contraction” being the result of replacing t with 1 Q t.6. If M is an interval M a. A speciﬁc homotopy φ : M LNM 0. φ1 : S1 P R2 in the plane given by φ0 H x ITJ x and φ1 H x IUJ x VXW 2 0 Y . φ N S 1 10. φ a b N Homotopy of loops. 1 O.5. Consider the two loops φ 0 .P N can be pictured as a rubber band ﬂoating through N from time 0 until time 1. however you try to move φ 0 to φ1 you get stuck at the origin. A manifold M is said to be contractible if there exists a point x0 in M such that the constant map φ 0 H x IKJ x0 is homotopic to the identity map φ H x I<J x.19 that punctured n-space Rn Q?R 0 S is not contractible.10.g.R 0 S ? Clearly the homotopy φ does not work. φ t H x IEJ 0 for x J W]\ 1 0 Y and t J 1 ^ 2. A homotopy of loops is given by shifting t . because it moves the loop through the forbidden point 0. This can be visualized as a thin rubber band sitting in N . E XAMPLE . (E. E XAMPLE .

t d dt q dx J . t d. It can be regarded as an application of Stokes’ theorem. We write out the proof for an open subset of Rn . Let M be a manifold. L EMMA (cylinder formula).7.e c x.f M _` 0. APPLICATIONS TO TOPOLOGY it seems intuitively clear that there exists no homotopy of loops from φ 0 to φ1 in the punctured plane. 1 bldmfni 1 k c Md . κγ e ∑ J o£p 0 g J c x.”) ι1 ι4 g 10 ι0 base cylinder If M is an open subset of Rn . where β and γ are forms on M _` 0. resp. by taking the piece of γ involving dt and integrating it over the unit interval. We 1 then deﬁne κγ e(r 0 γ dt. 1 d send M to the bottom.13. The product M _a` 0. t d . 1 b (of degree k h 1 and k respectively) that do not involve dt. This is indeed the case. 1b . In short. .) The cylinder operator turns forms on the cylinder into forms on the base lowering the degree by 1. 1b is often called the cylinder with base M. (In particular κγ e 0 for any γ that does not involve dt. Then ι 1 s γ t ι0 s γ e κ dγ h dκγ for all k h 1-forms γ on M _k` 0. The proof for arbitrary manifolds is similar. s t ι0 s e κ d h dκ . (“Slide the bottom to the top at speed 1. (Here we write the dt in front of the dx’s because that is more convenient in what follows. κ: i kj 1 c M _k` 0. a k h 1-form on the cylinder can be written as γe ∑ f I c x. the top of the cylinder. The two maps deﬁned by ι 0 c x d. with I running over multi-indices of degree k h 1 and J over multi-indices of degree k.116 10. 10. A homotopy ι : M _` 0. It sufﬁces to consider two cases: γ e f dx I and γ e g dt dx J . The following result will enable us to compare pullbacks of forms under homotopic maps. The homotopy formula. t d I dx I h ∑ g J c x.) For a general manifold M we can write a k h 1-form on the cylinder as γ e β h dt γ . as we shall see in Example 10. but we shall give a direct proof. 1 b between these maps is given by the identity map ι c x. t d J dt dx J . ι1 P ROOF. 1b.e c x. 0 d and ι1 c x d<e c x.

10. i i 1 so κ dγ u Also κγ u ∑w y n 1 0 ∂g x. t { dt dx J . φ1 : M N are homotopic maps between manifolds and α is a closed form on N. Now suppose we have a pair of maps φ 0 and φ1 going from a manifold M to a manifold N and that φ : M 0. Let φ 0 and φ1 be homotopic maps from M to N. φ1 φ0 u κφ d v dκφ . where n u dim M. Then φ1 α φ0 α u κφ dα v dκφ α for all k v 1-forms α on N.9. so z dκγ u i 1 ∂ ∑ ∂ xi w y n 1 0 g x. t { dt | dx I ∂t z g dt dx J . in other words φ 0 u φ ι0 . ∂t dκγ v κ dγ u κ dγ uxw£y Case 2.10. t { dt | dxi dx J u z i 1 ∑w y n 1 0 ∂g x. 1 < N is a homotopy between φ 0 and φ1 . φ 0 α and φ1 α have the same integral. Then y M φ0 α u)y M φ1 α. z z 0 and ∂g ∑ ∂xi dxi dt dx J u ∑ ∂xi dt dxi dx J . if dα u 0 we get φ 1 αu φ0 α v dκφ α . ∂ xi z QED Hence dκγ v κ dγ u 0 u ι1 γ ι0 γ. 10. ∂ xi z }y 1 0 g x. HOMOTOPY 117 Case 1. If γ u ∂f x. Suppose M is compact and oriented and has no boundary.10. If φ 0 . then ι0 γ u ι1 γu dγ u ∂g i u~} f x. φ1 : M N be smooth maps from a manifold M to a manifold N and let φ : M 0. For x in M we have φ ι 0 x {Tu φ x. In particular. . C OROLLARY. If γ u dγ u so f dx I . then κγ u ∂f i 0 and dκγ u 0. 10.8. Let M and N be manifolds and let α be a closed n-form on N. This implies that if the degree of α is equal to the dimension of M. Also ∂f dt dx I v ∂t ∑ ∂xi dxi dx I u 1 0 ∂f dt dx I v terms not involving dt. t { dt | dxi dx J . z z z Similarly φ1 u φ ι1 . Let φ 0 . Applying the cylinder formula to γ u φ α we see that the pullbacks φ0 α and φ1 α are related in the following manner. T HEOREM . 1 { f x. 0 { dx I u ι1 γ ι0 γ. t { dt | dxi dx J . then φ 0 α and φ1 α differ by an exact form. T HEOREM (homotopy formula). 1 N be a homotopy from φ 0 to φ1 . 0 {Tu φ0 x { .2. Hence for any k v 1-form α on N we have ι 0 φ α u φ0 α and ι1 φ α u φ1 α . In short.

0 and therefore are not homotopic (as loops in the punctured plane). E XAMPLE . Therefore. In a certain sense all obstructions to exactness are of this nature. E XAMPLE .118 10.8.6 have winding number 1. 0 M 0. A LTERNATIVE PROOF. APPLICATIONS TO TOPOLOGY P ROOF.1 dφ α ∂ M 0. 0 and α the angle form y dx x dy ¡ x2 y2 of Example 3.12. The boundary of M 0. resp. then a map from M to N is a loop in N and the integral of α is 2π times the winding number of the loop.9. 10. φ 1 α φ0 α by Stokes’ theorem M dβ for an n 1-form β on M. φ1 M dβ ∂M β 0. on others this is true only in certain degrees. if φ : M 0.13. Here is a proof based on Stokes’ theorem for the manifold with boundary M 0. φ0 QED If M is the circle S1 .1 φ dα M 0. QED because ∂ M is empty.3. 1 consists of two copies of M.3. 0 0 10. 10. By Corollary 10. 1 N is a homotopy between φ0 and φ1 . namely M 1 and M 0 .10 gives the following result. Unfolding the three self-intersections in the curve pictured below does not affect its winding number.11.1 φ α M α φ1 M α. The two circles φ 0 and φ1 of Example 10. Homotopic loops in R 2 a 0 have the same winding number about the origin. We shall not attempt to say the last . 10. Closed and exact forms re-examined The homotopy formula throws light on our old problem of when a closed form is exact. Thus Theorem 10. N the punctured plane R2 . Failure of exactness is typically detected by integrating over a submanifold of the correct dimension and ﬁnding a nonzero answer. which we looked into in Section 2. the ﬁrst of which is counted with a plus sign and the second with a minus. The answer turns out to depend on the “shape” of the manifold on which the forms are deﬁned. Hence α φ0 α . 1 . C OROLLARY. On some manifolds all closed forms (of positive degree) are exact.

The result now follows from Theorem 4.5 says that the integral of an exact 1-form along a loop is 0. Let us now consider 1-forms on a manifold M. (See Example 4. Hence dβ ¢ α . On a contractible manifold all closed forms of positive degree are exact.e.6. The punctured plane R2 £?¦ 0 § is not simply connected. The integral of a 1-form along a constant loop is 0. Then c is null-homotopic. Let M be a manifold and let φ : M ªk« 0. Theorem 10.14. This means that f is constant (on each connected component of M). P ROOF. A loop c : S1 ¤ M is null-homotopic if it is homotopic to a constant loop. All closed k-forms on a contractible manifold are exact for k ¨ 1. T HEOREM . φ1 Here we used the homotopy formula. Theorem 4. 10.14. All closed 1-forms on a simply connected manifold are exact. Let α be a closed k-form on M with k ¨ 1. i. T HEOREM (Poincaré lemma).10. Let c be a null-homotopic loop in M. whereas in higher dimensions there is enough room to slide any loop away from the puncture and then squeeze it to a point. Let α be a closed 1-form and c a loop in M. P ROPOSITION . The matter is explored in [Fla89] and at a more advanced level in [BT82]. so putting β ¢ κφ ¯ α we get dβ ¢ dκφ ¯ α ¢ 0 for all ¯ α £ φ0 ¯ α £ κ dφ ¯ α ¢ α . a smooth map satisfying φ x.17.) In contrast it can be proved that for n ¨ 3 the sphere S n © 1 and punctured n-space Rn £¦ 0 § are simply connected.15.8. CLOSED AND EXACT FORMS RE-EXAMINED 119 word on this problem. If this constant is nonzero.10 (where we set the M of the theorem equal to S 1 ) we get the following. 0-forms. but study a few representative special cases. then f is not exact (because forms of degree £ 1 are by deﬁnition 0). 10. Then φ 1 ¯ α ¢ α and φ0 ¯ α ¢ 0. P ROOF. 0 ®m¢ x0 and φ x. QED 10. and the assumption that dα ¢ 0. So a closed 0-form is never exact (unless it is 0) for a rather uninteresting reason.16. 1 ®m¢ x for all x. Then ¥ c α ¢ closed forms α on M. . 1-forms and simple connectivity. 1 ¬ ¤ M be a contraction onto a point x0 in M. so from Theorem 10. QED The proof provides us with a formula for the “antiderivative”. 10. A manifold is simply connected if every loop in it is null-homotopic. 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. E XAMPLE . because it possesses a nonexact closed 1-form. which can be made quite explicit in certain cases. With a stronger assumption on the loop the same is true for arbitrary closed 1-forms.5. A closed 0-form on a manifold is a smooth function f satisfying d f ¢ 0. The Poincaré lemma.3. namely β ¢ κφ ¯ α . so ¥ c α ¢ 0 by Proposition 10. Intuitively.

20. (See Exercise 10. P ROOF. α² x ¼¾½ dx ¿ ¿n . T HEOREM . x 10.6. we see that is not contractible. It can be shown that the winding number in any dimension is always an integer. dα ² 0 follows from Exercise 2. It provides a measure of how many times the hypersurface wraps around the origin. the volume form. but restricting it to the unit sphere S n À 1 .1(ii). (See Example 4.8. Hence punctured n-space is not contractible. This is a contradiction. Then · M α² · M dβ ² · ∂M α² 0 by Stokes’ theorem. α ² dβ for an n ¸ 1-form β. that R n ¸N¹ 0 » is not contractible.15 on M we have α ² µ . .² ∑ gi ° tx ±µ° xi dt ¶ i i β ² κφ ´ α ² ∑ xi · i 1 0 gi ° tx ± dt. 10.18. For example. On Rn ¸a¹ 0 » and on Sn À 1 every closed form of degree k ² Â 1. For instance.19 shows that the winding number of the n ¸ 1-sphere about the origin is 1. α is a closed but non-exact n ¸ 1-form on punctured n-space. E XAMPLE . According to the proof of the Poincaré lemma. On the other hand. the punctured plane R2 ¸º¹ 0 » is not contractible because it possesses a nonexact closed 1-form.5.19.) Another typical application of the Poincaré lemma is showing that a manifold is not contractible by exhibiting a closed form that is not exact. APPLICATIONS TO TOPOLOGY 10.120 10.) The angle form generalizes to an n ¸ 1-form on punctured n-space Rn ¸¹ 0 » . Then tx be the radial contraction. t ±³² Let α ² ∑i gi dxi be a 1-form. suppose α was exact. the function β satisﬁes dβ ² α provided that dα ² 0.17. Theorem 10. It now follows from the Poincaré lemma. The n ¸ 1-sphere M ² S n À 1 has unit normal vector ﬁeld x. so by Corollary 8. But how about forms of degree not equal to n ¸ 1? Without proof we state the following fact. It is instructive to compare β with the function f constructed in the proof of Theorem 4. the proof of Theorem 10. Hence Á M α ² vol M ² Â 0. T HEOREM . t dxi ± . For a compact oriented hypersurface without boundary M contained in in Rn ¸?¹ 0 » the integral 1 x ¼µ½ dx ¿ ¿n · À 1 n x vol n À 1 S M is the winding number of M about the origin.5. Theorem 9. Sn À 1 n ¸ 1 is exact. so α is not exact. Let M be Rn and let φ ° x. QED Using the same form α . It generalizes the winding number of a closed curve in R2 ¸¹ 0 » around the origin. φ´ α ² so ∑ gi ° tx± d ° txi ±. namely the angle form.

φ : M ÄÆÅ 0. x0 x0 c1 M M c1 Formally. Put c È s. Then c is a homotopy between c 1 and the constant loop c0 È t É.Ê φ È c1 È s É . For a discussion and references. It has an n-dimensional analogue.21. We cannot here go into this fascinating problem in any serious way. t É . Freedman in 1982. Use a contraction to collapse any loop onto a point.19 that they are not contractible. called the generalized Poincaré conjecture. Smale in 1960. Not long after inventing the fundamental group Poincaré posed the following question. P ROOF. QED As mentioned in Example 10.17 suggest that the notions of contractibility and simple connectivity are not independent. which appeared in the November 2003 issue of the Notices of the American Mathematical Society and can be read online at ÐÒÑÓÑÕÔ<ÖØ×Ó×ÚÙÛÙÓÙ<Ü]ÝµÞß:ÜáàÕâÕãä×ÚåÒàÕÑçæÛèÚéêßÕ× . 10. Suppose M is simply connected. 1 Ç Ã M a contraction of M onto x0 . let c1 : S1 Ã M be a loop. turned out to be the hardest. . t ÉKÊ φ È c1 È s É . The case n Ê 3. Strangely.10. Theorems 10. Is M homeomorphic to the three-dimensional sphere? (This means: does there exist a bijective map M Ã S3 which is continuous and has a continuous inverse?) This question became (inaccurately) known as the Poincaré conjecture. Milnor. although it follows from Theorem 10. the original version of the conjecture. Í The Poincaré conjecture.15 and 10. It is famously difﬁcult and was the force that drove many of the developments in twentieth-century topology. The case n Ê 4 was done by M.3. Perelman in 2002-03. but was ﬁnally conﬁrmed by G. Let M be a compact three-dimensional manifold without boundary. the sphere S n Ë 1 and punctured n-space Rn Ì 0 Î are simply connected for n Ï 3. the case n Ï 5 of the generalized Poincaré conjecture conjecture was the easiest and was conﬁrmed by S. which asks whether every compact n-dimensional manifold without boundary which is homotopy equivalent to Sn is homeomorphic to Sn . Thus simple connectivity is weaker than contractibility. 0 ÉmÊ x0 positioned at x0 . other than to report that it has now been completely solved. see the paper Towards the Poincaré conjecture and the classiﬁcation of 3-manifolds by J. A contractible manifold is simply connected.16. CLOSED AND EXACT FORMS RE-EXAMINED 121 Contractibility versus simple connectivity. P ROPOSITION .

) 10. Let α ï f dx dy g dz dx h dy dz be a 2-form on R3 and let φ ì x.6. Show that β ì x íï?ð cx α . see ù xil f I dxi1 dxi2 ññòñ¨dx il ñ¡ñòñ dxik . 0 í and let M be the twice-punctured plane R 2 ú©¤ 0. A subset M of Rn is star-shaped relative to a point x 0 ë M if for all x ë M the straight line segment joining x 0 to x is entirely contained in M. (Construct a 1-form α on M such that the integrals ð c1 α . ty. 2 sin t í . ty. 10. Let β ï p ¢ 1 ì]ú 1 í l û k∑∑ ï I l 1 k 1 Show that dβ ï α . t í t ì x. Let x ë Rn and let cx be the straight line connecting the origin to x. t íï tx.18. 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.1. By analogy with the radial contraction onto the origin. Show that if M is starshaped relative to x 0 . c2 and c3 are not homotopic. tz í t dt ÿ"ì x dy ú y dx í ¢ ü ý 1 0 g ì tx. Verify that ¢ κ dφ ø α ¢ ï α for k £ κφ ø α ï ü ý 1 0 f ì tx. ø ø c 0 c 1 10. APPLICATIONS TO TOPOLOGY Exercises 10.3. 2π ö£÷ ¢ c3 ì t í ï ì 1 . ï 10. (First show that every map M ÷ N is homotopic to a constant map φ ì x í ï y0 . z í be the radial contraction of R3 onto the origin.4.) c3 : ô 0.5 to each f I . φ1 : M ÷ N homotopic maps. Let α be a 1-form on Rn and let β be the function deﬁned in Example 10. then it is contractible onto x 0 .10.11. Write a formula for the map φ ﬁguring in the proof of Brouwer’s ﬁxed point theorem and prove that it is smooth.2. 10. write a formula for radial contraction onto the point x 0 . Give an example of a contractible set that is not star-shaped. Let x0 be any point in Rn . y. y. Let x0 ï ì 2. Show that c 1 . z. ¢ M be the loops deﬁned by c 1 ì t í ï ì cos t. tz í t dt ÿ"ì y dz ú z dy í . c 2 ì t í ï)ì 2 cos t. ð 10. c2 .7. Prove that any two maps φ 0 and φ1 from M to N are homotopic if M or N is contractible. Prove the following assertions. Deduce that any open or closed ball centred at x0 is contractible. 10.8. tz í t dt ÿ ì z dx ú x dz í ¢ üÒý 1 0 h ì tx. ¢ ñòñòñ¡dx im ñòñ¡ñ dxik .9. x í of radius ε centred at x is convex.122 10.) 0 and apply the identity proved in Exercise B. x0 ¦ . sin t í and 2 cos t. Let α be the k-form f dx I ï f dxi1 dxi2 be the radial contraction φ ì x.5. Let M and N be manifolds and φ 0 . Let α ï ∑ I f I dx I be a closed k-form whose coefﬁcients f I are smooth functions deﬁned on Rn ú¥¤ 0 ¦ that are all homogeneous of the same degree p ï § ú k. ð c2 α and ð c3 α are distinct. 10. 10. Show that ï ð φ α φ α for all closed k -chains c in M and all closed k-forms α on N . Give an example of a star-shaped set that is not convex. (iii) Same for the open ball B î¾ì ε.7. Verify that ñòñòñ dxik on Rn and let φ : Rn óõô 0. Let c1 . (i) M is convex if and only if it is star-shaped relative to each of its points. 10. (ii) The closed ball B ì ε. ty. 1 ö£÷ Rn κφ ø α ï and check directly that dκφ ø α m 1 k ü ý ∑ ìáú 1 í m û 1 ê 1 0 ù f ì tx í t k þ 1 dt ÿ xim dxi1 dxi2 1. (Use dα also Exercise 2. sin t í . x í .

A function g : R R is 2π -periodic if g x .12.EXERCISES 123 10.

(Use the parametrization c t cos t. integrate the equation β k dt dh over 0.) .) Prove that there is a unique number k such that α kµ is exact. 2π . Prove that there is a unique number k such that β k dt dh for some smooth 2π -periodic function h. (You can think of µ as the restriction to S 1 of the angle form. Then check that this value of k works. (To ﬁnd k. 2π g x for all x.) (ii) Let α be any 1-form on the unit circle S 1 and let µ be the element of arc length of S1 . sin t and apply the result of part (i). (i) Let g : R R be a smooth 2π -periodic function and let β g dt. where t is the coordinate on R.

.

b. x & x X or x Y is the union X Y . Glossary We start with a list of set-theoretical notations that are frequently used in the text. This is deﬁned as the set of all x such that x X or x Y . X Y : the intersection of X and Y . x X & x ) Y is the complement X Y . X Y : the complement of Y in X .APPENDIX A Sets and functions A.e. X Y : the union of X and Y . 1 ! is a cylinder wall of height 1.1. 1 % S1 " S1 x X & P x : the set of all x X which have the property P x . This means that f assigns to each x X a unique element f x + Y . x X : x is an element of X . usually written R2 . X Y : X is a subset of Y . This is deﬁned as the set of all x such that x X and x Y . x & x X and x Y is the intersection X Y . y with x X and y Y . 125 . and Y is called the codomain or target of f . c : the set containing the elements a. S1 0. R2 S1 "$# 0. i. Examples: R R is the Euclidean plane. every element of X is an element of Y . X Y : the Cartesian product of X and Y . This is by deﬁnition the set of all ordered pairs x. Examples: x R & 1' x( 3 is the interval 1. Let X and Y be sets. b and c. This is deﬁned as the set of x in X such that x is not in Y . 3 . a. The set X is called the domain or source of f . f : X * Y : f is a function (also called a map) from X to Y . S1 S1 is a torus.

: the image of a A under the map f . This notation is often abbreviated or abused in various ways. X -©. If X is a ﬁnite set and f : X ? R a real-valued function.”. f 3 A -:. i. A -/. This is often called the ﬁbre or level set of f at c. f . g . x . A function f : X ? Y is injective or one-to-one if x 1 . f .) f 5 1 . = x2 implies f . xif x 2 A. y for all y 2 Y . The function f is bijective if and only if it has a two-sided inverse f 5 1 : Y ? X satisfying f 5 1 . j .e. x -82 B 4 . f5 1. x .: an abbreviation for f 5 1 . also known as multi-indices. y . In other words.¨0 c 4¡. x -A-B. . c 4 . It is not meant to imply that f is required to have an inverse.126 A.: the preimage of B under the map f . f . i .. SETS AND FUNCTIONS f . j 3 2 1 0 4 3 2 i 5 4 6 . f . x2 . x -9. For instance. g > f . then g > f : X ? Z is deﬁned by . not deﬁned if x 2 = A. x1 -B. 2. i . If A is a subset of X .. if y 2 Y then y . x2 . . f 3 A is equal to f on A. If B is a subset of Y . If f : X ? Y and g : Y ? Z are functions. (This is a somewhat confusing notation. c . f 3 A is the function deﬁned by < f . x . It is called bijective if it is both injective and surjective. The set X is called the index set for the sum. x 1 -8. For n .. = f . where x ranges through X .for some x 2 X .implies x1 .. the sum of all the numbers f . f . In these notes we will often deal with indices which are pairs or k-tuples of integers. i .e. i F j. the set 0 x 2 X 3 f . j -6. f . x for all x 2 X and f . but “forgets” the values of f at points outside A.for some x 2 A 4 . . then its image under f is by deﬁnition the set f .70 x 2 X 3 f . 3 we can display X and f in a tableau as follows. A .) It is called surjective or onto if f . If A is a subset of X .satisfying 0 E i E j E n.into g .. We often say that the function g > f is obtained by “substituting y . x .. Y .. B . x -A. and let f . let n be a ﬁxed nonnegative integer. x -. f is injective if f . one uses the familiar notation ∑n i D 1 f . f 3 A: the restriction of f to A. i. x2 . let X be the set of all pairs of integers . x . . y -A-6. As a simple example.10 y 2 Y 3 y . this is by deﬁnition the set f5 1 . f 5 1 . x -@. is denoted by ∑ x C X f . n 4 . B -6. g > f : the composition of f and g. (Equivalently. if X is the collection 0 1. .

Let X be the surface of the earth. ﬁnite number. xn . x ε A subset O of Rn is open if for every x Q O there exists an ε Y 0 such that B L H ε. that converges to a point x in Rn . as far as these notes are concerned. . closed means “closed under taking limits”. B LMH ε. x I/N7O y Q Rn RTS y U x SBZ ε X . namely just the empty set and Rn itself. 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. The open ball of radius ε about a point x is the collection of all points y whose distance to x is less than ε. . Loosely speaking. 0 I for some value of R. for example the interval [ 0. (That is. . the limit x is contained in C. The importance of the notion of compactness. let Y be the real line and let . is that the integral of a continuous function over a compact subset of Rn is always a well-deﬁned. x I is contained in O. . B H ε.1. Parts (iii) and (iv) of this problem require the use of an atlas (or the Web. x ε Closed is not the opposite of open! There exist lots of subsets of R n that are neither open nor closed. (On the other hand. ).2. . General topology of Euclidean space Let x be a point in Euclidean space Rn .EXERCISES 127 The sum ∑ x G X f H x I of all these numbers is written as 0J iJ jJ n ∑ HiK jI . A subset C of Rn is closed if its complement Rn U C is open. A is contained in the ball B H R.) A compact subset of Rn is one that is both closed and bounded.2.) A subset A of Rn is bounded if there exists some R Y 0 such that S x S Z R for all x in A. 1 I in R. x I6NPO y Q Rn RTS y U x SWV ε X . This deﬁnition is equivalent to the following: C is closed if and only if for every sequence of points x1 . there are not so many subsets that are both open and closed. . Exercises \^]_]a`bdc_cAegfa]ghjiAegf_kjfj]^kjf:lnmpo^iaq A. x2 . A. An example of a closed set is the closed ball of radius ε about a point x.g. An open neighbourhood of x is any open set containing x. You will be asked to evaluate it explicitly in Exercise A. see e. . which is deﬁned as the collection of all points y whose distance to x is less than or equal to ε.

x u Sn v 1 closed? yes bounded? yes . Let S t n uzy (i) S t 0 uy (ii) S t n u y ∑0 { i { j { n t i | j u . ©u B t ε. f v 1 t 90 u . Prove that the open ball B :t ε. Find f t A u . x u B _t ε. in Rn . (iii) Let A be the contiguous United States. (b) a country that intersects f v 1 t B u but is not contained in f v 1 t B u . where A is as in part (iii). x u is open. Prove that the closed ball is B t ε. 1 denotes the unit sphere about the origin compact? yes 3 xy-plane in R n unit cube 0. 5 w 3. 0 and S t n | 1 uy S t n u}| 3 2 t n | 1 uAt n | 2 u .) t | ~ u t | u 2 A. Show that the two deﬁnitions of closedness given in the text are equivalent. (ii) Find f v 1 t 0 u .128 A. SETS AND FUNCTIONS f : X r Y be the function which assigns to each x s X its geographical latitude measured in degrees. ©u txw 3. (Same comments as for Exercise A.4. Here S n v that is the set of vectors of length 1.3. 5 u w 3. (iv) Let B y f t A u . Complete the following table. Round the numbers to whole degrees. A.) A. 1 w 3. Find (a) a country other than A that is contained in f v 1 t B u . Prove the following assertions.3. x u is closed. You will need the triangle inequality y w x y w z | z w x .2. using the deﬁnition of openness stated in the text. (Use induction on n.) A. (This is not a tautology! State your reasons as precisely as you can. A. f v 1 txw 90 u . (i) Find f t X u . and (c) a country in the northern hemisphere that does not intersect f v 1 t Bu . 1 n n 1 n 2 .6.5.

xn φm x and view φ x as a single map from Rn to Rm .and multi-variable calculus needed in the study of manifolds.1)–(B. Formula (B. . . x2 . .2. which expresses the function g in terms of the value g a and the derivative g .1) is a formula for a deﬁnite integral: it tells you how to ﬁnd the (signed) surface area between the graph of the function f and the x-axis. (In calculus the word “map” is often used for vector-valued functions. The fundamental theorem of calculus says that b f t dt F b F a . xn . Formula (B. References for this material are [Edw94]. φm be functions of n variables x 1 .2) dx a Writing g instead of F and g instead of f and adding g a to both sides in formula (B. x . (B. 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.1 for an application. . . a Formulas (B. Formula (B. .2) says that the integral of a continuous function is a differentiable function of the upper limit. . .1. .3) is an “integral formula”. B.1) we get x g x 6 g a z (B. [HH02] and [MT03]. φ2 . while the word “function” is generally reserved 129 .1) a There are two useful alternative ways of writing this theorem.3) g t dt. Let a.APPENDIX B Calculus review This appendix is a brief review of some single. . . b be an interval contained in the domain of F. . Replacing b with x and differentiating with respect to x we ﬁnd x d f t dt f x . Derivatives Let φ1 . but they emphasize different aspects of the fundamental theorem of calculus. and the derivative is the integrand. . . (See Exercise B. (B.) B. As usual we write x1 φ1 x x2 φ2 x φ x / .3) are equivalent.

. . usually called the gradient of φ: T Dφ x grad φ x . obtained by multiplying the matrix Dφ x by the vector v. 0¥ 0¥ 0¥ 0 0 1 n are the standard basis vectors of R . CALCULUS REVIEW for real-valued functions. In this case the matrix Dφ x consists of a single column vector. and the magnitude of grad φ x is equal along to the directional derivative Dφ x v.e. Therefore φ i x te j is well-deﬁned for ε ± t ± ε and thus it makes sense to ask whether the partial derivatives (B.4) ∂x j h h 0 are well-deﬁned and continuous functions of x for all i 1. and is usually denoted simply by φ § x . . Let us assume that U is an open set. . there exists ε ° 0 such that the points x te j are contained in U for ε ± t ± ε. . in which φ increases fastest. . . . is based on the identity a ¨ b ª© a ©© b © cos θ. . ∂ xn x ∂ x1 x If v is any vector in Rn . . Thus the direction of the vector grad φ x is the direction of steepest ascent. e2 . . The directional derivative of φ along v can then be written as an inner product.4) exist. . en . Here 0 « θ « π is the angle which subtended by a and b. If they do. 2. called the velocity vector. Frequently a function is not deﬁned on all of Rn . The transpose matrix of Dφ x is therefore a column vector. . . . ¥ . . . There is an important characterization of the gradient. the directional derivative of φ along v is deﬁned to be the vector Dφ x v in Rm . For n 1 φ is a vector-valued function of one variable x . . n and j 1. Dφ x . . 2. . then Dφ x v grad φ x ¬¨ v © grad φ x ^© cos θ . θ 0. i.130 B. . Because U is open. where v is the unit vector pointing grad φ x . Dφ x v grad φ x . where θ is the angle between grad φ x and v. So Dφ x v takes on its maximal in the same direction as value if cos θ 1.e. If v is a unit vector ( © v ©9 1). . the function φ is called continuously differentiable or C 1 . . often called a path or (parametrized) curve in Rm . . . . x ∂ x1 ∂ xn . For m 1 φ is a scalar-valued function of n variables and Dφ x is a single row vector. m. for all x ¯ U and all i and j. but only on a subset U . i. We must be a little careful in deﬁning the derivative of such a function. and if they are continuous. Here ¡ £¤ ¡ £¤ ¡ £¤ ¡ 1 ¤ ¡ 0 ¤ ¡ 0 ¤ ¡ ¡ ¡ ¤ ¤ ¤ ¡ 0 ¤ ¡ 1 ¤ ¡ 0 ¤ ¡ ¡ ¡ ¤ ¤ ¤ ¡ 0 ¤ ¡ 0 ¤ ¡ 0 ¤ ¡¢ ¤ ¡¢ ¤ ¡¢ ¤ e1 .) We say that φ is continuously differentiable if the partial derivatives φi x he j φi x ∂φi x lim (B. . ∂φ m ∂φ m . . .¨ v. This means that v points grad φ x . Let φ : U ® Rm be a function deﬁned on U and let x ¯ U . The (total) derivative or Jacobi matrix of φ at x is then the m ¦ n-matrix ¡ ∂φ £¤ ∂φ 1 ¡¢ ¤ 1 x .

then ψ º φ is a real-valued function of one variable x. Ä y1 y ³ m ² ² ² ² dφ m dx x ³ ² so by the chain rule ² d ψ º φ³ x ³/µ ² dx ² Dψ φ x ³A³ Dφ x ³µ ² ² ² iÈ 1 ∑ ∂ yi ² φ ² x ³A³ m ∂ψ dφi x³ . . n and j. Then ψ º φ is Cr and D ψ º φ ³ x ³/µ ² ² Dψ φ x ³A³ Dφ x ³ ² ² ² for all x ´ U. then φ is r times continuously differentiable or C r . dx .B. Sometimes we are sloppy and abbreviate this identity to d ψ º φ³ µ ² dx iÈ 1 ∑ ∂ yi m ∂ψ dφi . then we say that φ is inﬁnitely many times differentiable. T HEOREM (chain rule). In the one-variable case n µ m µ k µ 1 the derivatives Dφ and Dψ are 1 ¼ 1-matrices φ ½ x ³A³ and ψ ½ y ³A³ . ² ² ² ² B. Likewise. . trices Dψ φ x ² ² ² B.3. . k µ 1.3. so we get ψ º φ ³ ½ x ³/µ ψ ½ φ x ³A³ φ ½ x ³ . . Dφ x ³.2. Here Dψ φ x ³A³ Dφ x³ denotes the composition or the product of the two ma² ³³ ² and Dφ ² x³ . ∂∂y y ³¨Ç . E XAMPLE . . B and C are sets and φ : A ¹ B and ψ : B ¹ C are functions. ² ² ÂÃ dφ 1 x ³ dx ² ∂ψ ψ . . . . If φ is Cr for all r · 1.5) This is perhaps the most important special case of the chain rule. ² ² ² ² B. so D ψ º φ ³ is a 1 ¼ 1-matrix containing the single entry ψ º φ ³ ½ . dx ² (B. This means that φ can be differentiated arbitrarily many times with respect to any of the variables. Moreover. 2. or smooth. and matrix multiplication is ordinary ² usual chain ² the ² ² rule multiplication. then φ is called twice continuously differentiable or C 2 . B. THE CHAIN RULE 131 If the second partial derivatives ∂ 2φ i x³ ∂ x j ∂ xk ² exist and are continuous for all x ´ U and for all i µ 1.µ¿¾ÀÁ and Dψ y ³µÆÅ ∂ . E XAMPLE .1. Let U » Rn and V » Rm be open and let φ : U ¹ V and ψ : V ¹ Rk be Cr . Let us now review some of the most important facts concerning derivatives. . If n µ k µ 1. C ¸ . m. if all r-fold partial derivatives ∂rφ i x³ ∂ x j1 ∂ x j2 ¶:¶:¶ ∂ x jr ² exist and are continuous. . we can apply ψ after φ to obtain the composite function ψ º φ ³ x ³/µ ψ φ x ³³ . .3. The chain rule Recall that if A. . 2.

ÉPÝ ∂φ Ï ∂v Now let us take φ to be of the form φ Í u. notation is dψ dx É iÊ 1 ∑ ∂ yi m ∂ψ dφi . Consider the equation 0. Then there are open neighbourhoods U Ü R n of u0 and V Ü Rm of v0 such that for each u Ô U there exists a unique v f Í u Î©Ô V satisÉ fying φ Í u. ØÙ ØÙ ∂φ 1 ∂φ 1 ∂φ 1 Ù ∂φ 1 Ù . Let φ : W Ò Rm be Cr .and v-variables separately. . We are in business if we have a point Í u0 . . Suppose that Í u0 . v Î g Í vÎ u. In this notation we have dφ Ì ψ dx É iÊ 1 ∑ φ Ì¬Ð m ∂ψ dφi . so that φ Ì ψ Í x Î stands for ψ Í φ Í x ÎAÎ .4. . . ÉÕÖ . If ∂φ Ï ∂v ã 0 at a certain 0 0 0 É v0 we can solve the equation φ Í u. . CALCULUS REVIEW An even sloppier.4. B. . T HEOREM (implicit function theorem). . v0 Î 0 and É Dvφ Í u0 . D φ v . where W is open in Rn Ó m . Dvφ Dg. Instead of ψ Ë φ we often write φ Ì ψ. . The function f : U Ò V is C r with derivative given by implicit É differentiation: D f Í uÎ for all u Ô U. We form the Jacobi matrices of φ with respect to the u. . . but nevertheless quite common. Ú . v0 ÎÛÔ W is a point such that φ Í u0 . É¿ÕÖ . v Î with u Ô Rn and v Ô Rm . . v Î . . Let us think of a vector in Rn Ó m as an ordered pair of vectors Í u. Solving φ Í u. v0 Î is invertible. . The implicit function theorem be a continuously differentiable function deﬁned on an open Let φ : W Ò subset W of Rn Ó m . where g : W Ò R n is a É Ý n given function with W open in R . v Î 0 for v as a function v f Í u Î of u. Moreover. dx In these notes we frequently use the so-called “pullback” notation. ∂ yi Ñ dx (B. . ∂φm ∂u ∂un ∂v ∂vm 1 1 Rm φ Í u. É Under what circumstances is it possible to solve for v as a function of u? The answer is given by the implicit function theorem. . v ÎAÞ 1 Duφ Í u. so the implicit function theorem yields the É following result. É7Ý Dvφ Í u. v Îàß v Ê ß f á uâ . Duφ . when φ is a function of two real variables É point É Í u . v Î 0 here amounts to inverting É the function g. . This is well-known for m n 1. then for u close to u and v close to Í u.132 B. Ú . ∂φm Ö× ∂φm . v Î Observe that the matrix Dvφ is square.6) B. Similarly. φ Ì¡Í ∂ψ Ï ∂ yi Î_Í x Î stands for ∂ψ Ï ∂ y i Í φ Í x ÎAÎ . f Í u ÎAÎ 0. ∂u1 ∂un ∂v1 ∂vm . v Î . and É É ∂φ Ï ∂u fä . Ö× ∂φm . v0 Î at which φ is 0 and Dvφ is invertible.

Again this should look familiar from one-variable calculus: if p : ç a. T HEOREM (change of variables formula). . yn . where p : U å V is a map from an open U é Rn to V . . and ç g ê 1 è ñòç u è9ë 1 þnç 2 ü u è .5. The inverse function has derivative 1 . Then for any integrable function f we have ¢ V f ç y è dy ë ¢ U f ç p ç xèAè¤£ det Dp ç xè¥£ dx.) This means we substitute y ë p ç x è . B. Suppose we want to change the variables in the integral V f ç y è dy. Suppose that v0 æ W is a point such that Dg ç v0 è is invertible. where W is open in R . c ¡ ¡ b This can be written succinctly as cd f ç y è dy ë a f ç p ç x èè¥£ p ñ ç x è¥£ dx. 0 è . ù è .7. B. Then g ç I èúë ç 0. Therefore if I is a sufﬁciently small open interval around u 0 . g ê 1 ç u èûëýü u. In a neighbourhood of 0 it is not possible to invert g. . Let g : W å R n be continuously n differentiable. The inverse g ê 1 : V å U is continuously differentiable with derivative given by Dg ê for all v æ V.6. ù è . For v0 õ 0 we can take I ë ç 0. ç g ê 1 è ñ ç u è/ë gñ ç v è with v ë gê 1ç 1 ç u èë Dg ç v è ê 1ì ì ví gî 1 ï uð uè . d è is C 1 and has a C 1 inverse. b è å ç c. . which looks more similar to the multidimensional case.B. Then g ñ ç v0 èøë ó 0 whenever v0 ë ó 0. THE SUBSTITUTION FORMULA FOR INTEGRALS 133 B. Suppose that p is bijective and that p and its inverse are continuously differentiable. Invertibility of Dg ç v 0 è simply means that g ñòç v0 èôë ó 0. then g ç I è is an open interval around g ç u0 è and the restricted function g : I å g ç I è is invertible. ù è . For v0 ÷ 0 we can take I ë ç~ÿWù . The substitution formula for integrals Let V be an open subset of Rn and let f : ¡ V å R be a function. and ç g ê 1 è ñ ç u è ë ÿ 1 þnç 2 ü u è . Then there is an open neighbourhood U é R n of v0 such that g ç U è is an open neighbourhood of g ç v0 è and the function g : U å g ç U è is invertible. Let U and V be open subsets of R n and let p : U å V be a map.5. Again let us spell out the one-variable case n ë 1. then ¡ b ¢ d f ç p ç x èAè p ñ ç x è dx if p is increasing. y2 . B.5. f ç y è dy ë§¦ ¡ ab ÿ a f ç p ç x èAè p ñ ç x è dx if p is decreasing. Then g ç I è ë ç 0. This implies that near v 0 the function g is strictly monotone increasing (if g ñdç v0 èöõ 0) or decreasing (if g ñdç v0 èö÷ 0). (This is shorthand for an n-fold integral over y 1 . Let g ç v èë v 2 . . g ê 1 ç u è ë ÿ ü u. E XAMPLE (square roots). Under a suitable hypothesis we can change the integral over y to an integral over x. T HEOREM (inverse function theorem).

b © R be a where n 0. Let g : ¨ a.3) show that g x (ii) Show that g x g a h t 1 g a th 0 h2 g a h 1 0 C n .1. CALCULUS REVIEW Exercises h B.134 B. (i) By changing variables in the fundamental theorem of calculus (B. Suppose a x b and put x a.

1 1 0 1 t g a th dt g a hg a h2 1 0 1 t g a th dt.) (iii) By induction on n deduce from part (ii) that g x k 0 ∑ n g k a k h k! hn . g a th dt. (Integrate the formula in part (i) by parts and don’t forget to use the chain rule. 1-function.

1 n! 1 0 1 t n g n.

6. Show that F is the gradient of f x Rn ( ( tp f x . Deduce from the chain rule that Dφ x v lim t! 0 x tv. What are their degrees? (ii) Assume that f is deﬁned at 0 and continuous everywhere. B. Find c t . Show that f is constant if p 0. Let x and v be constant vectors in Rn . Deﬁne c t B. B. (iii) Plot the function f over the interval 5 x 5. y. Deﬁne a map ψ from Rn 3 1 to Rn by 1 2 ψ t ( ( t 19 2t : t ( ( 2 1 1 en .3. A function f : *$ 0 &+ R is homogeneous of degree p if f tx . (i) Show that ψ t lies on the unit sphere S n 3 . Here p is a real constant. f x. t B.) 2 B. 1 a th dt. φ x tv " φ x . x Gm 1 m2 ) where G is a constant of nature.2. then i 1 ∑ xi ∂xi x n ∂f p f x . (i) Show that f is differentiable at 0 and that f 0 0.4. (ii) Show that f is smooth and that f n 0 0 for all n. (iii) Show that if f is homogeneous of degree p and smooth. f x. 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 F ' Gm1 m2 ( ( 3 x. Deﬁne a function f : R R by f 0 0 and f x e 3 1 7 x for x 8 0. (Differentiate the relation f tx t p f x with respect to t. 0. Using software or a graphing calculator is ﬁne. x # Rn -$ 0 & and t . According to Newton’s law of gravitation. about the origin.7. This is Taylor’s formula with integral remainder term. x for all (i) Show that the functions f x. B. z 2 x2 z6 3 x4 y2 z2 4365 2 are homogeneous. . y /0 x 2 xy ) x2 y2 . but pay special attention to the behaviour near x 0.5. y /01 x3 y3 . Show that p 0.

@ (ii) γ 2 > N U U N U e @ @ @ U U (i) T < x W 1 => x < x = for all x S 0. e1 . en . 2. . but the results are needed in Chapter 9. The simplex I spanned by the ai ’s is the collection of all their convex linear combinations. (Here we regard t >?< t 1 .10) by establishing the following identities. (i) For n > 1. For brevity write γ > < 1 2= . and prove the T B.) (iii) Compute Dψ < t = . Find Dφ < 0 = if B.9. B. an A a0 . . Calculate < n W 1 2 = (where is the function deﬁned in Exercise B.EXERCISES 135 (ii) Show that ψ < t = is the intersection point of the sphere and the line through the points en and t. 2 N U e @ s ds. The stereographic projection from y of any point x in the sphere distinct from y is deﬁned as the unique intersection point of the line joining y to x and the hyperplane P. Write a formula for the stereographic projection φ from the south pole A en and for its inverse ψ : Rn @ 1 E Sn @ 1 . 3 draw pictures of the standard n-simplex as well as a nonstandard n-simplex. A linear combination ∑n i H 0 c i ai is convex if > 1. t2 . (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. Then map I to the standard simplex by an appropriate substitution and apply the substitution formula for integrals. tn @ 1 . X > S n @ 1 ACB en D .10. a1 . . A map φ : Rn E φ is even and C 1 . . IC>KJ L The standard simplex in Rn is the simplex spanned by the vectors 0.11. . tn @ 1 = as a point in Rn by identifying it with < t 1 . This deﬁnes a map φ : Sn @ 1 ACB y D E P. e2 . . . . . . = >X< n A 1 = ! for TZ (ii) < n positive integers n. . . . x2 . t2 . . For x S 0 deﬁne T < x => NVU e @ t t x @ 0 1 dt T followingT assertions. . Q n! Q where A is the n R n-matrix with columns a1 A a0 . (iv) Let X be the sphere punctured at the “north pole”. Stereographic projection from the north pole is the map φ : X E Rn @ 1 given by φ < x =F> < xn A 1 = @ 1 < x1 .8. . dx1 dx2 OPOPO dxn . .) The following two calculus problems are not review problems. ∑n i H 0 ci Rm is called even if φ < A x =G> φ < x = for all x in Rn . . Let a0 . . (v) Draw diagrams illustrating the maps φ and ψ for n > 2 and n > 3. xn @ 1 = . B. (ii) The volume of a region R in Rn is deﬁned as N vol IC> R iH 0 ∑ ci ai LL ∑ ci > L iH 0 n n 1M . . 0 = . . . Show that 1 det A . Y 1 aW 1 @ u2 a u du > (iii) NVU e . (First compute the volume of the standard simplex by repeated integration. . a2 A a0 . an be vectors in Rn . . 2 2 T [ 0 T (i) γ > x2 @ y2 dx dy. . The point y is called the centre of the projection. Show that φ is a two-sided inverse of ψ. a2 .

CALCULUS REVIEW (iii) γ 2 \^] (v) ced n f 2π 0 (iv) γ \^b π . 1 \ 2g 1 h 3 h 5 hPhPhji 2n k 1 l b π for n m 2n . ]`_ re a 0 r2 dr dθ. 1.136 B.

gauge theory. P.. Standard multivariable calculus reference. surfaces. 2003. corrected reprint of the 1969 original. Pollack. Good reference for the basic linear algebra required in these notes. 1991. 1998. 1991. R. second ed. CRC Press. Linear algebra with applications. Concise but lucid (and cheap!). New Jersey. linear algebra. Bachman. Massachusetts. Upper Saddle River. Written for 1960s engineering graduate students. 137 [BS91] [BT82] [Bre91] [Bre05] [Car94] [Dar94] [Edw94] [Fla89] [Gra98] [GP74] [HH02] [MT03] [Opr03] [Sin01] [Spi65] . Darling. 1989. Tromba. New York. J. A modern approach to classical theorems of advanced calculus. New York-Amsterdam. W. Still recommended as an alternative or supplementary source. Accessible to intermediate undergraduates. Tu. J. H. Benjamin. Written for beginning graduate students. Prentice Hall. MA. Cambridge University Press. Berlin. Birkhäuser.. Differential forms with applications to the physical sciences. second ed. H.. Hubbard. N. modern introduction. Advanced calculus: A differential forms approach. Flanders. Prentice Hall. Multivariable calculus from the point of view of Newton’s law and special relativity with coverage of differential forms. but very intuitive and with lots of interesting applications to topology. 2003. Cambridge. Boston. Birkhäuser. Calculus on manifolds. H... Dover Publications. 2001. 1965. Differential topology.. Undergraduate Texts in Mathematics. third ed. Bretscher. ﬁfth ed. R. W. second ed. J. Marsden and A. Guillemin and A. Boston. including expositions of Maxwell theory and its modern generalization. do Carmo. A course in mathematics for students of physics. 1982. Cambridge. and differential forms: A uniﬁed approach. Freeman. M. Springer-Verlag. Advanced undergraduate text on manifolds and differential forms. Bott and L. Boca Raton. Symmetry in mechanics. Leisurely and thorough exposition at intermediate undergraduate level with plenty of computer graphics. New York. Edwards. V. Differential geometry and its applications. New York. A gentle. S. Differential forms and applications. M. 2002. Vector calculus. 2005. Singer. D. A geometric approach to differential forms. Englewood Cliffs. geodesics and calculus of variations with plenty of MAPLE programming. One of the earliest undergraduate textbooks covering differential forms. A. Second year calculus.J. 1994. Springer-Verlag. Curves. a version of which is available through the author’s website nPojoqprsjstpPuPvxw4vqyz{p}|~ojuPqzqyqsPtyq vxwn4v4s . Sternberg. New Jersey. Spivak. including the Gauß-Bonnet theorem. Birkhäuser Boston. translated from the 1971 Portuguese original. O. New York. 1994. Vector calculus. A recent text. Bressoud. Oprea. second ed. Gray. MA. Bamberg and S. Boston.. Efﬁcient and rigorous treatment of many of the topics in these notes. Differential forms in algebraic topology. Modern differential geometry of curves and surfaces with Mathematica. Pearson Prentice Hall. A. 1974. Slightly more advanced than these notes. Differential forms and connections. Cambridge University Press. Springer-Verlag. 2006. Masterly exposition at beginning graduate level of the uses of differential forms in topology and de Rham cohomology. Prentice-Hall.Bibliography [Bac06] D. Upper Saddle River. 1994. with some coverage of Riemannian geometry. Hubbard and B. FL.

138 BIBLIOGRAPHY . Publish or Perish. TX. Academic Press. as well as a number of straightforward exercises. CA. Houston. Weintraub. Differential geometry textbook at advanced undergraduate level in ﬁve massive but fun to read volumes. Written as a companion to multivariable calculus texts. [Spi99] . 1999. [Wei97] S. Contains careful and intuitive explanations of several of the ideas covered in these notes.. Differential forms. A comprehensive introduction to differential geometry. A complement to vector calculus. third ed. 1997. San Diego.

ϑ ι κ λ µ ν ξ o π. ϕ χ ψ ω 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 .The Greek alphabet upper case lower case name A B E Z H I K M N O P T X α β γ δ .ε ζ η θ. ρ σ τ υ φ.

.

determinant of a matrix A. 37. 106 . Kronecker delta. 29 M α . 103 d. 82 φ ª . k . J . curl of a vector ﬁeld. 20 £ . 88 of a function. 73 S n . 110. closed ball in Rn . composition of f and g. 37. 26 dx. 17. 103 I . 131 ¦ . unit cube in Rk . 34 µ M . 64 . orthogonal complement. 34 1 . i-th standard basis vector of Rn . 90 B ε . multi-index i 1 . 126. 17. binomial coefﬁcient. 23 n. x ¡ . 26 141 dx. i k ¡ (usually increasing). orientation deﬁned by a basis . exterior derivative. 43 A I . 57 Alt µ . 27 Dφ. 24 relativistic. 132 Rn . 68 H n . 59 of a manifold. integral of a form over a chain. 8. 86 δ i. vector space of k-forms on M. composition of maps. tensor multiplication. unit normal vector ﬁeld. ¤ 83 dx i . 35 A k V . graph of a function.Notation Index . interior of a manifold with boundary. 1 rank A. partial derivative. 74 . 57 over a manifold. 26 graph. 28 δ I . short for dx i1 dx i2 ~ dx ik . 25 dx I . 47. restriction of f to A. Jacobi matrix of φ. 17 dx i . kernel (nullspace) of a matrix A. 73 §n O n ¡ . 51 det A. gradient of a function. . 18 ei . 37. . 96 k ¨ . 106 c α . covector (“inﬁnitesimal increment”). 130 ∂. 127 n B ¢4 ε. Euclidean n-space. orientation deﬁned by a basis . J . 19. 42 α . . 130 f ¥ A. 103 ker A. integral of α over a chain c. 131 . I . 37. divergence of a vector ﬁeld. permutation group. 95 nullity A. r times continuously differentiable. inﬁnitesimal displacement. alternating form associated to µ . 126 g f . 135 grad. 76 © k M ¡ . permutation matrix. 85 A T . 126. Hodge star operator. boundary of a chain. . length of a permutation σ . inﬁnitesimal hypersurface. volume form of a manifold M. 130 ∂ xi . Kronecker delta. i 2 . 17 int M. unit sphere about the origin in Rn « S n . omit dx i . Euclidean norm (length). 73 l σ ¡ . j . dimension of the kernel of A. 131 curl. 33 sign σ ¡ . pullback of a form. set of alternating k-multilinear functions on V . dimension of the column space of A. transpose of a matrix A. 19. integral of a form over a chain. 127 . open ball in R . 94 . Hodge star of α . 8 . Euclidean inner product (dot product). sign of a permutation σ . x ¡ . 89 ∂ . Laplacian of a function. orthogonal group. 58 0. exterior multiplication. 85 Aσ . 31 div. upper halfspace in Rn . 24 29 relativistic. integral of α over a manifold M. J -submatrix of A. Gamma function. 47 V . 1 . 94 C r . 8 .

1 ¯ ¯ . dual of a vector space V . 74 V ® . n-dimensional Euclidean volume. 73. Euclidean norm (length) of a vector x. transpose of a vector x. 69. 83 V k . 8 x T . 78 Tx M. 91 x .142 NOTATION INDEX SL ¬ n . k-fold Cartesian product of a vector space V . 8. tangent space to M at x. special linear group. Euclidean inner product (dot product) of vectors x and y. 85 voln . 8 x ° y.

109 atlas. italic page numbers refer to examples or applications. 1. 9. 8. 82 average of a function. 120. 107 chain.Index Page numbers in boldface refer to deﬁnitions or theorems. Renatus. 91. 115. 1. 74. 122 Cartan. 55. 104–111. 31. 52–53 anticommutativity. 85. 106–110. 103. 87–90 property. 32. 47. 59. 106. 89. 51. 51–53. 127. 60. 69. 69. 4. 62. 7. Elie (1869–1951). 36. 122 Brouwer’s ﬁxed point theorem. 62. 64. 127 Brouwer. 115. André Marie (1775–1836). 51. 60–62. 40. open ball barycentre. 79 cross-cap. 69. 3. afﬁne space. 109 curvature. 128 codimension. 24 conﬁguration space. 115. René centroid. 137 column operation. 49. 125 Cartesius. 73 alternating algebra. 116 . 84. 96 rectangular. 72. 29 Ampère’s Law. 64. Harold Scott MacDonald (1907–2003). 64–65 curl. 96. 106. 122 curve. 73. 2–7. 43 Coxeter relations. 122. 10. 130 contractible. 38 convex. 12. 17 curve. 36. 113. 84. 113. 103. see alternating multilinear function arc length. 62. 69 cycle. 54. 90 ﬁeld. see Descartes. 122 linear combination. 22. see rectangular block Bonnet. 57. 55. 19. 64. 62–65 of a manifold. 62 form. 83 Coxeter. see covector covector. 42 vector. 137 boundary of a chain. 67. 48. 123 function along a curve. 110. 117–123 set. 119 conservative. 43 critical point. 21 Ampère. 8. 118 bounded. In a few cases italic boldface is in order. 18. 29. 67. 9. 119–122 contraction. 27–29. 120. 116. 101. 109 Archimedes’ Law. 105 cohomology. 115. 118. 118. 17 Cartesian product. 121–122 contravariance. 83. 83. 23. 60–65 chart. see closed ball. 105. 127 complementary. 14 connected component. 31. 47. 78 multilinear function. 89 block. 7 cube in an open set. 45. 127 chain. 29 angle form. Luitzen Egbertus Jan (1881–1966). 107 Archimedes of Syracuse (287–212 BC). 123 closed ball. 32. 130 compact. 19. 83 continuously differentiable. 47. 89. see chart covariant vector. 85. 109 constant form. 28. 135 coordinate map. 89. 107 bilinear. 47. 117. 106. 53. 85. Pierre (1819–1892). 93. 107 ball. 27. 85. 69. 65. 42. 59. 120. 1. 18 antisymmetric matrix. 82 multilinear function. 108. 107. 92. 72 143 circle. 64 cylinder formula. 50.

63–65. 106. 34 inward pointing. 110 Green’s theorem. see exact form on a manifold. 10. 60. Jean Le Rond (1717–1783). 108. 117–120. 28. 78. 107. 83 electromagnetic wave. 134 in Rn . 52. 108. 34. 129 Graßmann. 38 Hodge star operator. 109 form as a vector-eating animal. 103. 63. 47 . 74–79. 134 of a multi-index. 88 volume. 135 permutation. 57. 96. 106 . 110 halfspace. 110. 110 disconnected. 120 increasing multi-index. see volume form free space. 49. 82. 14 Descartes. 86. 106. 120. 96–98. 121 hypersurface. 15 form. 1–11. 71. 23. 106. 86. 8. 29 electromagnetism. Hermann (1809–1877). see level set ﬁxed point. Michael (1951–). 38. 47. 91–94 differential equation. 101. 130 Gram. Jorgen (1850–1916). 110 Dirichlet integral. 22. 87. 82 derivative. 84. 18. 28. 69. 114. 114 formula. 28. 61. 84. 134 Greek alphabet. 65. 49–52. 29 de Rham. 9. 117 of curves. 29 ﬁbre. see covector fundamental theorem of calculus. 115 of loops. 27. 20. 108 Gamma function. René (1596–1650). Lejeune (1805–1859). 89. 87. 43 exact form. 106. 137 degenerate chain. 113 ﬂux. 69 Dirichlet. 98. 70. 122 inversion. 100. 28. 88 on Euclidean space. 65. 91. Carl Friedrich (1777–1855). 25. 110. 69. 63. 77–78. 51 of a form over a chain. 96 embedding. 54. 106 over a manifold. 108. 19 gradient. 28. 82 gravitation. 127 Euclidean motion. 1. 73. 69. 67. 100. 85. 29. 41. 103. 108. 48. 115. 43. 107 Euclid of Alexandria (ca. 17. 78. 64. see also relativity homogeneous function. 28. 116 d’Alembert.144 INDEX with base M. 17 product. 139 Green. see form dimension. 26. 21. 119. 25. 1. 40–42. Georges (1903–1990). 110. 81. 108. 134 homotopy. 88 closed. 121 function. 58–59. 95 Gauß’ Law. 103 Hodge. 29 d’Alembertian. see closed form exact. see inner product dual basis. 92 graph. 125 space. 26. 18–30. 130 of forms. 64 degree of a form. 123 exterior algebra. 85. 24. 91 ευρηκα ´ . 108. v. 49. 89 vector. 32–35. 70. 127 volume. 65. 125 determinant. 137 Gauß map. 125 dot product. 68. 82 differential calculus. 135 Gauß. 29 Freedman. William (1903–1975). 92 Gram-Schmidt process. 22 integral of a 1-form over a curve. 29 graded commutativity. 85. see covector space. 29 Faraday’s Law. 99. 63 on a manifold. 125. v. 122. 49. see product of forms Faraday. 117. 108. 325–265 BC). 65. 91 plane. 109 even map. 67. 125. 67. 49. 28 integrability condition. 65. George (1793–1841). 129. 26. see also complementary index of a vector ﬁeld. 29. 119. 29. 100. 88. 108 divergence. 27. 109 domain. 114. 123 of surface area. Michael (1791–1867). 109 of a function. 96. 68. 23. 95–101. 17 degrees of freedom. 95. 101. 63. 86. 97. 31. 25. 130 functional. 17 of a homogeneous function. 26. 19. 29 element of arc length. 55 inner product. 64. 12. 104. 101. 110–111.

5 pullback of a form . 57. 1. see block parallelogram. 105 Möbius band. 43 open ball. 100 group.INDEX 145 k-chain. 29 line segment. 95 group. 125. 100 of a manifold. 130 Maxwell. 74 set. 29 space. 82 of permutations. 28 Laplacian. 53. 76. 34. 96 Lotka. 88. 17 function. 4 parallelepiped. 78 Lotka-Volterra model. 100 parametrization. 106. 134. 91. 95 orthogonal complement. 4. 127. 76. 29 Minkowski inner product. 91 outward pointing. 21. Pierre-Simon (1749–1827). 22 operator. 77–78. 13. 119 potential. see form k-multilinear function. 35. 108 predator. 126 surface. 130 partial differential equation. 103. 9. 15 prey. 89 n-manifold. 33. 72 with boundary. Hermann (1864–1909). 74 lightlike. 43 pinch point. 1–15. 95–97. 93 orthonormal. 121 Poincaré conjecture. Jules Henri (1854–1912). 33. 27 on a manifold. 106. 54. 92 operator. 7. 121 periodic function. see multilinear function Klein. 7 plane curve. 14. 36 Milnor. 34. see unit normal vector ﬁeld odd permutation. 57. 69 Poincaré. 29 mean of a function. see Cartesian product product rule. 28 Leibniz. Alfred (1880–1949). 58 Newton. see chain k-cube. 126. see Leibniz rule projective plane. 8 normal vector ﬁeld. see cube k-form. 15. 9–13. 107 measurable set. 48. Felix (1849–1925). 43 matrix. 105 paraboloid. 34. 48. 105 multi-index. 70–79 abstract. 57. 67 parametrized curve. 38. 8. 21. 121 lemma. 42–43. 15 product of forms. see parametrized curve pentagon. 40 for functions. 94. 123 permutation. 128 is a manifold. 110 pair of pants. 5 Kronecker. 137 norm of a vector. 15. 82. 2–7. 49. 28 projection. 73. 108 of a hypersurface. 69 given explicitly. Leopold (1823–1891). 108 of a vector space. Grigori (1966–). 95. 104–111. 47 . 91. 13. 4. 100 preserving. 48. 106. August (1790–1868). 76. 78 matrix. see also increasing multiindex multilinear algebra. 42–43 of a vector. Isaac (1643–1727). 20 path. 88. 51 Kronecker delta. 91 local representative. 85. 9. 20. 43 of sets. 28. John (1931–). 118 map. 85. 51 Laplace. 40 Leibniz rule for forms. 17. 72 given implicitly. 128 level curve. 79. 70 orientation of a boundary. 127 set. 101. 5 Klein bottle. 69. James Clerk (1831–1879). 54. 95. 53–55. 121 minimum. 89. see manifold naturality of pullbacks. 107 reversing. 127 neighbourhood. 106 outward-pointing. 14 Perelman. 29 Möbius. Gottfried Wilhelm von (1646–1716). 41. 95. 49. 74. 114. 19. 119. 34. 91. 13. 74 hypersurface. 20 length of a permutation. 78 manifold. 75 Minkowski.

109. 81. 69 surface. potential. 51. 131. 88. 15 space curve. 120 submanifold. 79. 134 Taylor’s formula. 42 unit circle. 69. 40–42. 69. 108. 1. 8. 89. 35. curl. 98. 121 singular cube. 57. 128. 97.146 INDEX on a manifold. 1. 69. 14. 109 for chains. 110. 26. 89 timelike. 48. 14. 51. 67. 13. 109 area. 58. 130 transposition. 105. 115. 78 volume change. 92 sign of a permutation. 101. 106. 4. 106 trace. 107. 95. 24. 105 relativity. 68. 29 special linear group. 57. 126 retraction. see Euclidean volume form. James (1692–1770). 29 topological manifold. 35. see also conservative. 47. 110. 97. 79. 42 vector. 17. 137 rigid body. see conﬁguration space steepest ascent. 128 interval. 64 for manifolds. 63. 4 point. 18. 129 symmetric bilinear function. 73. 83. 58. 96. see also column. 47. 77. 96. Brook (1685–1731). 71. 134 tensor product. 75. 100. 74. 17. 130 stereographic projection. 11. 99. tangent ﬁeld. 85. 121 smooth curve. 130 saddle point. 122 state space. 70. 108– 110. 65. 2 surface. 64. v. 135 simply connected. 96 Euclidean. 114. 74–79. 28. 63. 2. 1 space-time. 36. 79 tangent hyperplane. 73–76. 85 matrix. 47. 106. 17. 9. 100. 67. 109. 65. 14. 45. 4. unit normal Volterra. 48. row. 35. 88. 114. 33. 42–43 simple permutation. 120 quadrilateral. 47. 78 trajectory. gradient. 117 on Euclidean space. 96. 78. 83. 15. unit. 64. 111 Stirling’s formula. Stephen (1930–). 119. 103 of a hypersurface. 65. 120 sphere. 78 sphere. 69 solution curve. 72. 76. 114. 14. 48. 82. 118. 105 Smale. 94 simplex. 114. 58. 135 star-shaped. 134 hypersurface. 18. 116. 106 vector. 8. index. 14. 119–121 plane. 44 pendulum. 42 element. 9. 78 singularity. 137 reparametrization of a curve. 119. 103. 58 restriction of a map. 106. 106. 10. see sphere square. 95. 111 Stokes. 70. 11 rectangular block. 69 line. 96. 65. 113 Riemann. 119–121. 113. 74. 134 spherical coordinates. 96. 116 normal vector ﬁeld. 15. 77 plane. 75. 13. 62 vector. see circle cube. 29. 4. 73. 8. George (1819–1903). 75 Schmidt. 50. 48. 37 . 24. 9 standard basis of Rn . 47. 65. 69 manifold. 116. 69. 109 Stokes’ theorem classical version. 88. Vito (1860–1940). 108. 1. 69. 135 Stirling. 69 function or map. 42 . see cube value. 91. 99 on Rn . 107. 78 space. 130 orientation. Bernhard (1826–1866). 10. 29. 74–76. 57. 34. Erhard (1876–1959). 106 Taylor. 47. 132 punctured Euclidean space. 58 of a rectangular block. 65 regular value. 29 spacelike. 60. 4 torus. 43 simplex. 78. 55. 88 of a function. divergence. 78 transformation law. 35. 10 row operation. 130 vector. 98. 107. 82 transpose of a matrix or a vector. length. 1.

109 of a simplex. 31. 48. 109 zero of a vector ﬁeld. 108. 18. 17. 89. 85. 110. 135 wave operator. 49. 54–55. 17. 93 of a manifold. 89 winding number of closed curve. 107. 120 of hypersurface. 92. 25 . 25. 53. 91. 120 work. 61. 118. 29 wedge product.INDEX 147 on a hypersurface. 120 of a block. 100.

- Latex test
- Orientation on Manifolds
- Homework in Real Analysis
- MidtermODE(2013)
- Homework Template Test
- Test of Latex Notes Template
- Braid Action on Chords Grouped
- Linear Algebra Review
- TESTANALYSIS.pdf
- HomeworkTemplateNew
- Untitled.pdf
- Solution to Stein Complex Analysis
- The Importance of Being Earnest
- Latex Test for Math
- Linear Algebra Notes
- Contact Structure
- Contact Geometry
- Augmentations and Knot Contact Homology
- Non-commutative Algebra for Mathematica Document
- Arnold - Geometrical Methods of Gibbs Thermodynamics
- Latex Guide
- Pitchfork YouWereThere TheCompleteLCDSoundsystem
- Linear Algebra
- Pde

Sign up to vote on this title

UsefulNot usefulManifold Notes from Cornell University

Manifold Notes from Cornell University

- a comprehensive introduction to differential geometryby kladjfklsdflkdnf
- Darling - Differential Formsby Patrik Jakobsson
- Ebooksclub.org Differential Geometry and Lie Groups for Physicists Cambridge University Pressby Anil Tripathi
- Michael Spivak - A Comprehensive Introduction to Differential Geometry, Vol. 5by Antonio García

- a comprehensive introduction to differential geometry
- Darling - Differential Forms
- Ebooksclub.org Differential Geometry and Lie Groups for Physicists Cambridge University Press
- Michael Spivak - A Comprehensive Introduction to Differential Geometry, Vol. 5
- Michael Spivak-A Comprehensive Introduction to Differential Geometry Volume 2, Third Edition-Publish or Perish (1999).pdf
- Manifold and Differential Forms
- Manifold
- Differential Forms.
- Weakly Symmetric and Special Weakly Ricci Symmetric LP-Ssasakian Manifolds
- Riemann
- Ends of Complexes
- Pseudo-Manifold Geometries with Applications
- trabajo-master-curso-2009-10-iglesias.pdf
- Tempfli Paper
- Paper-1 April 2015 on a Generalised Almost Contact Metric Structure-II 1
- Lectures on Sympletic Geometry
- Topological strings
- Dual Smarandache Curves and Smarandache Ruled Surfaces
- Lesson Plan Surface Area of Sphere
- Ms Release Note
- g Exp H Semi-Riemann Metric on Tangent Bundle and Its Index
- Day4Exercises[1]
- Data Doctor
- PNAS-1955-Pekeris-469-80
- SupMan
- Methods for fluids
- Transients Characteristics
- 7646DA27d01
- A Unified Formulation of the Segregated Class
- mathgen-21239478
- Manifold Cornell Notes

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue reading from where you left off, or restart the preview.