ELEMENTS “ OF DIFFERENTIAL GEOMETRY RICHARD S. MILLMAN GEORGE D. PARKER Southern Illinois University Carbondale, Illinois Prentice-Hall Inc., Englewood Cliffs, New Jersey 07632Library of Congress Cataloging in Publication Data MILLMaN, RICHARD S (date) Elements of differential geometry. Bibliography: p. Includes fade 1. Geometry, Differential. I. Parker, George D., (date) joint author. II. Title. QA641.M52 516.3604 76-28497 ISBN 0-13-264143-7 © 1977 by PRENTICE-HALL, INC. Englewood Cliffs, New Jersey 07632 All rights reserved. No part of this book may be reproduced in any form or by any means without permission in writing from the publisher. Printed in the United States of America 1098765432 PRENTICE-HALL INTERNATIONAL, INC., London PRENTICE-HALL OF AUSTRALIA PTY. LIMITED, Sydney PRENTICE-HALL OF CANADA, LTD., Toronto PRENTICE-HALL OF INDIA PRIVATE LIMITED, New Delhi PRENTICE-HALL OF JAPAN, INC., Tokyo PRENTICE-HALL OF SOUTHEAST ASIA PTE. LTD., Singapore WHITEHALL BOOKS LIMITED, Wellington, New ZealandPreface xi Preliminaries 7 1-1. Vector Spaces 7 1-2. Linear Transformations and Eigenvectors 1-3. Orientation and Cross Products 6 1-4, Lines, Planes, and Spheres 8 1-5. Vector Calculus 70 Local Curve Theory 13 Basic Definitions and Examples 14 Arc Length 20 Curvature and the Frenet-Serret Apparatus Contents 24 The Frenet-Serret Theorem and Its Corollaries 29 The Fundamental Existence and Uniqueness Theorem for Curves 41 Non-Unit Speed Curves 46 viiGlobal Theory of Plane Curves 49 3-1. Line Integrals and Green’s Theorem 50 3-2. The Rotation Index of a Plane Curve 52 3-3. Convex Curves 60 3-4. The Isoperimetric Inequality 63 3-5, The Four-Vertex Theorem 66 3-6. APreview 72 Local Surface Theory 74 4-1, Basic Definitions and Examples 76 4-2. Surfaces 88 4-3. The First Fundamental Form and Arc Length 9 4-4, Normal Curvature, Geodesic Curvature, and Gauss’s Formulas = 702 4-5. Geodesics 709 4-6. Parallel Vector Fields Along a Curve and Parallelism 116 4-7, The Second Fundamental Form and the Weingarten Map 722 4-8. Principal, Gaussian, Mean, and Normal Curvatures 4-9, Riemannian Curvature and Gauss’s Theorema Egregium 147 4-10. Isometries and the Fundamental Theorem of Surfaces 146 4-11, Surfaces of Constant Curvature 763 Global Theory of Space Curves 167 5-1. Fenchel’s Theorem 161 5-2. The Fary-Milnor Theorem 167 5-3. Total Torsion 170 Global Theory of Surfaces 173 6-1. Simple Curvature Results 174 6-2. Geodesic Coordinate Patches 176 6-3. Orientability and Angular Variation 180 6-4, The Gauss-Bonnet Formula 186 6-5. The Gauss-Bonnet Theorem and the Euler Characteristic 788 6-6. The Theorems of Jacobi and Hadamard = 792 6-7. The Index of a Vector Field 195Contents 7 Introduction to Manifolds 198 7-1. Some Analytic Preliminaries 799 7-2. Manifolds—Definition and Examples 203 7-3. Tangent Vectors and the Tangent Space 208 7-4, Vector Fields and Lie Brackets 276 7-5. The Differential of a Map and Submanifolds 279 7-6. Linear Connections on Manifolds 223 7-7. Parallel Vector Fields and Geodesics on a Manifold with a Linear Connection 227 7-8. Riemannian Metrics, Distance, and Curvature 232 Appendix: Historical Notes 243 Bibliography 248 Index |. Notational 255 I, Topical 257Preface The intent of this book is to provide an elementary and geometric intro- duction to differential geometry. We adopt an approach that is elementary enough for undergraduates but still both conveys the spirit of modern differ- ential geometry and prepares the student for a more advanced course on the subject. In particular, we have given significant emphasis to global considera- tions while providing the basic classical results on curves and surfaces. In these global results we have been guided in part by the beautiful article of Chern [1967]. This book is an outgrowth of courses taught by the first author at Ithaca College and by both authors at Southern Illinois University over a Period of six years. In differential geometry there is calculus and there is geometry, neither of which should be slighted. All too often the geometry is hidden in either machinery, abstraction, or symbolism. Furthermore, an unfortunate thing has happened to the subject in the last ten years—it has been relegated to the graduate curriculum. There is no question that differential geometry (or any subject for that matter) can be done more efficiently if the student has the background that every second year graduate student has, but to wait until graduate school to teach the subject is doing an injustice to both the student xtxii Preface and the subject. We hope that this book will aid in the return of differential geometry to the undergraduate curriculum and that it can be taught by non- specialists, as well as by specialists. A traditional undergraduate course in classical differential geometry usually hides the geometry in a myriad of symbols such as T,;* or Ri, We have minimized this difficulty by making use of linear algebra throughout. Modern vector space terminology can and should be used effectively to make the material more comprehensible to the students. Furthermore, we find that the students are very excited about using the linear algebra they have just learned in another course. They view differential geometry as an application of linear algebra, which it is. After the students understand the geometric content of a result (by use of linear algebra) the result is then expressed in classical notation. This enables the reader to consult texts in classical differ- ential geometry without having to develop a whole new vocabulary. In the advanced approach, which is usually restricted to the graduate level, the geometry is hidden in the machinery and abstraction. We have avoided topics which require building complicated machinery, such as differ- ential forms or cohomology. We have resisted the temptation to prove the most general results possible if these would require a lot of machinery to state or to prove. In particular, for the first six chapters we stay in R*. In order to avoid the trap of abstraction we include many examples—not only of the definitions but also of the theorems because it is usually much easier to follow a proof with a firm example in mind. When first introduced to differential geometry, individuals often have trouble reconciling the two views of the subject—classical and modern (mani- folds). It is sometimes hard to tell what the connection (no pun intended) is between the two approaches. We attempt to remedy this situation in the last chapter. There we explain how the earlier material motivates the definitions of manifold, tangent space, Riemannian metric, etc. and give a brief intro- duction to the modern terminology. We hope that having finished this chapter the reader will have an easier transition to books such as Hicks [1965], Boothby [1975], Warner [1971], Matsushima [1972], or Kobayashi and Nomizu [1963]. There is only one way to learn mathematics and that is to get your hands dirty working problems. Differential geometry is no exception. We have included more than 350 problems. The easier problems are usually at the beginning of the problem sets. Those whose results are used in the text are marked with an asterisk (*). There is one exception to this rule—in Chapters 1-6 we have not starred any problems which would be needed in Chapter 7. It should be noted that an asterisk does not denote a hard problem: we have reserved the dagger (t) for this. In Chapter | we present the material which is the necessary backgroundPreface xiii for the book. Most of it is review and may be emphasized as needed. We have found that our students are generally weak in describing lines and planes in yector notation. The material on eigenvalues and linear transformations is not used until Chapter 4 and may be delayed until then. Chapter 2 develops the basic local theory of space curves. The Frenet- Serret Theorem which expresses the derivative of a geometrically chosen basis of R? in terms of itself is proved, and many important corollaries are derived. We include the Fundamental Theorem of Curves, showing the depen- dence of differential geometry on the theory of ordinary differential equations. Several classical topics are covered in the exercises, including sphere curves, contact, Bertrand curves, evolutes, and involutes. In Chapter 3 we give our first sampling of global theorems. Here we stay in the plane and cover the Rotation Index Theorem of Hopf, the Isoperi- metric Inequality, the Four-Vertex Theorem, and curves of constant width. We feel that this is a basic chapter because global differential geometry is a very important and popular subject which is all too often slighted in a first course. It is certainly easier to prove only local results. However, this really cheats the student who thereby misses one of the most fascinating parts of the subject and is robbed of the insight gained by examining results in the plane. ' Chapter 4 presents the basics of local surface theory and serves as the motivation for the ideas of Chapter 7. We study surfaces in R? and their first and second fundamental forms. Geodesics and their length-minimizing prop- erties are discussed. Next comes an investigation of curvature (both Gaussian and mean), its relationship to the curvature of curves on the surface, and Gauss’s Theorema Egregium. The chapter ends with optional material on isometries, the Fundamental Theorem of Surfaces, and surfaces of constant curvature. Again, several classical topics are covered in the exercises, includ- ing ruled surfaces, developable surfaces, and asymptotic curves. In Chapter 5 we return to global notions, this time for space curves. Using the concepts of geodesics on the sphere and integration on a surface, we prove Fenchel’s Theorem about the total curvature of closed space curves and the Fary-Milnor Theorem about the total curvature of a knot. We also include the nonstandard topic of total torsion of a closed space curve. Chapter 6 gives various global results for surfaces. We start off with Meusnier’s Theorem which states that a compact surface, all of whose points are umbilics, is a sphere. We then go on to the Gauss-Bonnet Formula, which relates the curvature of a region with the curvature of its boundary, and the Gauss-Bonnet Theorem, which gives the total curvature of a compact surface. As applications of the Gauss-Bonnet Theorem we prove theorems due to Jacobi, Hadamard, and Poincaré. Motivated by the first six chapters, we introduce in Chapter 7 the basicxiv Preface definitions of manifold theory and Riemannian geometry. This chapter is written in an open-ended fashion, referring the reader to other books for details. A brief historical summary and bibliography are given in the appendix. References to the bibliography are in the form of a name followed by a date in brackets, such as Chern [1967]. The first four chapters require only a knowledge of calculus and finite- dimensional vector spaces (mainly bases and linear independence). Chapter 5 has no additional prerequisites, but the material is more difficult. For Chapter 6 it would be helpful if the reader is familiar with the topology of R*. In par- ticular, the reader should be familiar with the notion of the limit of a sequence of points in R?. We define compact as closed and bounded. In Chapter 7 it would be useful if the reader knew some metric space terminology, such as open sets and continuity. A knowledge of the inverse and implicit function theorems would also make the going a bit easier. Several different courses can be made from this book. A one quarter undergraduate course could be based on Chapters 1, 2, and 3 or 4; a one semester course on 1, 2, 3, 4 (or more, depending on the pace of the course); a two quarter course could cover the first six chapters; the entire book could be used for a year course if the last chapter is supplemented. A graduate course could start with Chapters 2, 4, and 7 and then go on to a book like Hicks [1965], Boothby [1975], Matsushima [1972], or Warner [1971]. The following table shows the dependence of the various chapters. 0 >] —[]—-) pa Na We would like to thank Sharon Champion for her magnificent job typing the manuscript, from the first set of class notes to the final draft. Her patience and attention to detail through the jth rewrite of the ith author (i = 1 or 2, 1 } a‘y, is zero, then each a’ must also be zero. If it is possible to find a finite linear combination }) a'y, = 0 with some a* + 0, then the set is linearly dependent. DEFINITION. A subset S < V spans V if for each vector v € V there are vectors ¥,,V2,-..,¥, in S and real numbers a', a?,...,a" such that v= > a‘y,. (The number of elements used (“r”) may depend on v.) DEFINITION. A basis of a vector space V is a linearly independent spanning set. THEOREM 1.3, If V is a vector space, then V has a basis. Any two bases have the same number of elements, or all have infinitely many elements. This number is called the dimension of V. EXAMPLE 1.4, R? has dimension 3. EXAMPLE 1.5. R[x] has infinite dimension. {1, x, x?,..., x",...} is a basis. If {v,|¢ € 7} is a basis of V, then every vector y € V can be uniquely written as a finite sum y = >) a'y,. The numbers a’ are called the components of v with respect to the given basis. DEFINITION. An inner product on a vector space V is a function< , >: Vv x V— R such that for all u,v, w ¢ Vandr,s € R: (a) = = r + s > 0 with equality if and only if u = 0. EXAMPLE 1.6. In R? we may use the ordinary dot product: <(a', a, a*), (b', b*, b’)> = atb! + ab? + a3? Exampe 1.7. In R[x] we may set < p(x), (x)> = J, p(x)q(x) dx. See Problem 1.4. DEFINITION. If V has an inner product and v < V, the Jength of y is IWL= SW, ¥.Sec. 1-1 Vector Spaces 3 Lemma 1.8 (Caney senwatz Inequality). If u,v < V, then | | <|ul|¥]. Furthermore, | u, v>| = |u||v] if and only if u and v are linearly depen- dent. Proof: Problem 1.5. This lemma tells us that —1 < /|u||v} < 1 (unless [u||v| = 0, in which case = 0 also). Hence an angle @ may be defined by the formula Tully] ” cos 8 = 6, which is defined only up to sign, will be called the angle between u and v. If the ordinary dot product is used in R?, then this concept of an angle coin- cides with the usual notion of an angle in the plane. Note that = |u| ¥| cos 8 holds if one of the vectors is zero, even though @ is not defined then. DEFINITION. u is orthogonal (or perpendicular) to v if . The inner product may be associated with a matrix in the following manner. Let g,, = . (g,,) is a positive definite symmetric matrix which repre- sents< , > with respect to the given basis. If u = 5) a’u, and v= >) du, then with respect to the basis {v,,v2,..-,V¥,} is given by Sap = (Vas Vp>. Then gi; = as defined in Example 1.7 is an inner product. *1.5. Prove Lemma 1.8. (Hint: Let g(t) = 0. How many real roots does g have? What does the quadratic formula say?) 1-2. LINEAR TRANSFORMATIONS AND EIGENVECTORS Dernition. A linear transformation is a function T : V — W of vector spaces such that T(av + bw) = aT(v) + bT(w) for all a,b ¢ Randv,wev. An isomorphism is a one-to-one onto linear transformation. If T: V— W is a linear transformation, we matty associate a matrix with it. Let {v;, v2,..., v,} be a basis of V and let {w,, Wo,..., W,} be a basis of W. Then there are mn real numbers T‘, such that T(v,) = > T',w,. We say that (T',) represents T with respect to the given bases. (If V = W,, it is cus- tomary to use the same basis for V and W.) If v = av, € V, then TH) = UCL Taw, Thus if we view v as a column vector, the i in T', is the row index and the j is the column index. Suppose T:V — V is a linear transformation and V has two bases {u,,u.,...,u,} and {v,,¥,,...,¥,} related by u, = > a*,y,.. If (T',) represents T with respect to the u, and (7“,) represents T with respect to the v,, we have LT,a*v. = DT, = Tu) = Td a?,V_) = Xa? T(¥p) = Ya, T*5Vu- Hence 3) T'ja*, =X a%, Tp, or (a*)(1') = (F*,)(a",), or (2-1) (T') = ary (7%) (@*)). This equation shows the effect of a change of basis on the matrix which represents a linear transformation.Sec. 1-2 Linear Transformations and Eigenvectors 5 DEFINITION. Let T: V— V be a linear transformation. A real number 4. is an eigenvalue of T if there is a nonzero vector v such that T(v) = dy. ¥ is called an eigenvector of T corresponding to A If (T%,) represents T, the eigenvalues of T are the real solutions of the polynomial equation det (T'; — x6',) = 0. Then if the dimension of Vis n, there are at most 7 eigenvalues (counting multiplicity). There may be fewer, since some of the solutions might be complex but not real. Once the eigen- values are known, the eigenvectors are found by solving appropriate linear equations. EXAMPLE 2.1. Let T: R? —> R? be represented with respect to the standard basis {(1, 0)', (0, 1)'} by the matrix ( MD 3-4 4 4 og) |e 93-16 =-9+4+1?—16 = 2-25 =(A—5)A +5). 4 The eigenvalues are therefore 5 and —5. ( )C) = 5(*) has 4 —3/\y y 2 : 3 A\(x x 1 ( as one solution. ( )( ) = ~3( ) has ( ) as one solu- 1 4 —3/\y y —2, 2\. . ‘ . tion. Hence ( 1 ) is an eigenvector corresponding to 5 and > is one corresponding to —S. PROBLEMS 3 2.1. In Example 2.1 we have T represented by ( basis ((1, 0), (0, 1) of R’. (a) Represent T with respect to the basis { ( It cl, =D. (b) Represent T with respect to the basis {(2, 1)’, (1,-2)!} of eigenvectors. 4 ) with respect to the 2.2. Find the eigenvalues and eigenvectors of T: R? > R? given by the matrix (oa)6 Preliminaries Chap. 1 1-3. ORIENTATION AND CROSS PRODUCTS Let {u,, u,...,u,} and {v,, ¥.,..., ¥,} be two ordered bases of V and define a matrix (a',) by v, = }) a’,u,. DEFINITION. The ordered bases {u,,u,,...,u,} and {v,, V¥2,..., V,} give the same orientation to V if det (a',) > 0. They give opposite orientations if det (a',) < 0. EXAMPLE 3.1. Let {(1, 0, 0), (0, 1, 0), (0, 0, 1)} and {(1, 1, 0), (1, 0, —1), (2, 1, 3)} 1 1 2 be two ordered bases of R*. Since (a’,) = f 0 i has determi- o —I 3 nant —4, these ordered bases give opposite orientations. EXAMPLE 3.2. {(1, 0, 0), (0, 1, 0), @, 0, 1)} and {(1, 1, 0), (2, 1, 3), (1, 0, —1} give the same orientation. The basis {(1, 0, 0), (0, 1, 0), (0, 0, 1)} of R? will be denoted {e,, e2, es}. Its orientation will be called right handed. DEFINITION. If u = 3) ae, and v = > bie, are vectors in R?, the cross (or vector) product of wand v is wu X v= (a*b> — a®b%)Je, + (ab! — a'b)e, + (a'b* — a*b")e;. By abuse of notation this may be written as ey e e ux v= det/a! a a}. bt Bo BF Lema 3.3. Let u,v, w € R? andr & R. Then @)uxv=-vxu; (b) (rw) x v= 7(u x ¥); (©) u x v=O if and only if u and y are dependent; (d) (U+yv) X W= (Ux Ww) + (Vv x Wy); (©) u X vis perpendicular to both u and v under the usual dot product of R? (Example 1.6); (f) |u x v| = |uljv| sin 6, where 6 is the angle between u and y; (g) {u, v, u X v} gives a right handed orientation to R? if {u, v} is linearly independent. Proof: Problem 3.1.Sec. 1-3 Orientation and Cross Products 7 Lemma 3.4. Let u, v, w © R®. Then <(u x y), w> = . Proof: Problem 3.2. i DEFINITION. The mixed (or triple) scalar product of u, v, w is [u, v, w] = . Prove det (a,,) = [uy, uz, Us]l¥,, Vo, Vs]-8 Preliminaries Chap. 1 1-4. LINES, PLANES, AND SPHERES We shall recall the vector equations of lines, planes, and spheres in R?. See a standard calculus text such as Thomas [1968] for more details. Geo- metrically, a straight line is determined by a point on the line and a vector parallel to it. A plane is determined by a point on the plane and a vector perpendicular to the plane. A sphere is determined by its center and its radius. DerIniTION. The /ine through xp € R? and parallel to v4 0 has equation a(t) = x, + tv. (See Figure 1.2.) tV=a(th—x9 | ao Xo FIGURE 1.2 If we write X) = (Xo; Yos Zo), V = (v', v*, v) and a(t) = (x(t), yO, 2(), then the definition gives x0) —% =, e()—yo = tv, x(t) — 2) = 109, Assuming that v' 0 for i = 1, 2, 3, these equations yield the classical defini- tion of a straight line after solving each equality for ¢ and setting these quantities equal: X= % _ Y= _Z— 20, vi vo where we have suppressed the ¢ from the notation as is common classically. If x, and x, are distinct points in R? both of which lie on a line /, then the vector x, — xX, is parallel to /, This observation proves: LemMA 4.1. The line through x, and x, in R? has equation a(t) = x, + (x, — x,). (See Figure 1.3.)Sec. 1-4 Lines, Planes, and Spheres 9 a(t)—x,= £(%;-%,) ase FIGURE 1.3 DEFINITION. The plane through x, perpendicular to n + 0 has equation = 0. (See Figure 1.4.) FIGURE 1.4 The following lemma is clear from Figure 1.5 since u x vis perpendicular to the desired plane. FIGURE 1.510 Preliminaries Chap. 1 LemMa 4.2. If {u, v} is linearly independent, the plane through x, parallel to both u and y has equation = 0. DEFINITION. The sphere in R? with center m and radius r > 0 has equation . Then d _ dt dg GED = op 8) + ( a) In particular, if |f| is constant then df/dt is perpendicular to f.Sec. 1-5 Vector Calculus 1 Proof: Problem 5.1. Lema 5.2. Let f,g:R— R?. Then 4exg=4 x gitx Ss. Proof: Problem 5.2. §j DEFINITION. f: R — R is of class C* if all derivatives up through order k exist and are continuous. f:R"-— R is of class C* if all its (mixed) partial derivatives of order k or less exist and are continuous. A vector- valued function is of class C* if all its components with respect to a given basis are of class C*. Note that if fis of class C* it is also of class C*-!. Rather than continually worrying about what class a differentiable func- tion belongs to, we shall usually assume it is of class C?. We shall point out those cases where higher class is needed or lower class is sufficient. Finally, before we start our study of curves, we want to remind you what form the chain rule for differentiation takes. Suppose that x is a function of several variables u',u?,...,u" and that the wu’ are functions of variables ot, v2,..., 0". Then Ox — Sax dul _ (1) Be Pox ga 1,2,..5m Note that we are writing the coefficients on the right of the vectors instead of the left as would be usual in linear algebra. This is done so that Equations (5-1) look more like the chain rule from calculus. Special cases that we shall often use arise when 1 = m = 2 as in Equa- vtion (5-2), or 2 = 2 and m = 1, as in Equation (5-3). f ax _ Ox du! dx aw 62) ae awa tae t= hb? (5-3) dx _ 0x duh Ox du? dt dul dt + Ow dt PROBLEMS “*5.1, Prove Lemma 5.1. *5.2. Prove Lemma 5.2. 4 r Set A—Hyperbolic Functions i For some of the examples and problems in this book it will be necessary ito use certain transcendental functions which behave in many ways like the t12 Preliminaries Chap. 1 trigonometric functions. Since they are not always covered in a calculus course we briefly cover them in this problem set. The hyperbolic sine function is the function sinh: R — R given by sinh(t) = (e' — e~‘)/2. The hyperbolic cosine Function is the function cosh : R — R given by cosh(t) = (e' + e~')/2. For aid in reading, sinh is pronounced “cinch” while “cosh” rhymes with “gosh.” 2. 5.3. Prove that for any ¢ € R, cosh?(r) — sinh>(t) = 1. (Since x? — y 1 is a hyperbola, this gives the origin of the term “hyperbolic function.” Also, you should think about the analogy between this equation and cos? 9 + sin? @ = 1.) 5.4. Prove that cosh’(¢) = sinh(/), and sinh’(t) = cosh(¢) for all t € R. 5.5. Using Problem 5.4 show that sinh is one-to-one. Show that cosh is one- to-one when restricted to [0, oo). Show also that sinh is onto the reals whereas cosh(R) = [I, oo). (This means we may define the appropriate inverse function.) 5.6. Show that the general solution to the differential equation f” = a*f is given by f(t) = A cosh(at) + Bsinh(at) where A and B are real con- stants. (If you cannot show this is the general solution, at least show that it satisfies the differential equation.) 5.7. Compute f dx/,/T + x? by means of the substitution x = sinh(r).2 Local Curve Theory We shall begin our study of differential geometry with an investigation of curves in three-dimensional Euclidean space, R?. This is a good place to start for four reasons: (1) since curves in R? are easy to draw and visualize, we can develop some geometric insight into the subject by looking at exam- ples; (2) it is a very complete subject—the Frenet-Serret apparatus of a curve completely determines the local geometry of curves and gives a complete set of invariants for the problem of determining whether two curves are the same; (3) the theory of curves will introduce us to some techniques that are the mainstay of modern differential geometry (e.g., linear algebra); (4) we shall base our study of surfaces in Chapter 4 on the behavior of curves on the surfaces. The history of the theory of curves (and of all differential geometry) is a fascinating one. Suffice it to say at this point that the many results in the theory of curves in R*, which we discuss in this chapter, were initiated by G. Monge (1746-1818) and his school (Meusnier, Lancret, and Dupin). Our approach is due to G. Darboux (1842-1917) whose idea of moving frames unified a great deal of the classical theory of curves. He accomplished this in 1887-1896. It is interesting to note that the approach to the theory of sur- faces that we will take in Chapter 4 is that of K. Gauss in 1827. This anomaly of dates is not due to the fact that the curve theory is more difficult than the 1314 Local Curve Theory Chap. 2 theory of surfaces (just the opposite is true) but rather to the pervasive genius of Gauss. The history of the theory of curves is discussed more completely in the historical notes at the end of this book. There are two ways to think of curves. The first is as a geometric set of points, or locus. When this is the case, we refer to a geometric curve, or the geometric shape of the curve. Intuitively we are thinking of a curve as the path traced out by a particle moving in R?. The second way of thinking of a curve is as a function of some parameter, say f. Intuitively it is not always enough to know where a particle went—we also want to know when it got there. (The parameter ¢ is often thought of as time.) It is necessary to view curves the second way if we are to apply the techniques of calculus to describe the geometric behavior of a curve. This means that we must pay careful attention to how the curve is parametrized (e.g., if you change the parameter you also change the velocity vector field to the curve). However, we are also interested in geometric properties of curves (e.g., arc length, tangent vector field). These should not depend on the way a curve is parametrized as a function but only on the image set of the function, that is, only on the geometric shape. Thus we shall ask whether our constructions and descriptions depend upon the parametrizations. In Section 2-1 we define and give examples of parametrized curves. In Section 2-2 we introduce a particularly useful parametrization—that by arc length. Section 2-3 develops the Frenet-Serret apparatus which is the basic tool in the study of curves. It consists of three vector fields along the given curve (the tangent T, the normal N, and the binormal B) and two scalar- valued functions (the curvature « and the torsion t). The Frenet-Serret Theorem is proved in the fourth section. This theorem expresses the deriva- tives of T, N, and B in terms of T, N, and B. We then make several applica- tions. There is also a Jong collection of problems at the end of this section, many of which deal with topics from the classical differential geometry of curves. Section 2-5 gives the Fundamental Theorem of Curves and shows that the Frenet-Serret apparatus does completely determine the geometry of the curve. Finally, in Section 2-6 we develop the necessary techniques for comput- ing the Frenet-Serret apparatus for curves which are not parametrized by arc length. 2-1, BASIC DEFINITIONS AND EXAMPLES Our study of curves will be restricted to a certain class of curves in R?. Not only do we want a curve to be described by a differentiable function so that we may use calculus to describe the geometry, we also want to avoid certain pathologies and technicalities. If da/dt = 0 on an interval, then @(¢) is constant in that interval, which is geometrically very uninteresting. If‘Sec. 2-1 Basic Definitions and Examples 15 dadt is 0 at some point, then the graph of a can have a sharp corner, which is geometrically unappealing. (Consider the graph of a(t) = (¢, 27, 0).) Be- cause of these considerations we will only work with regular curves. DeriniTion. A regular curve in R? is a function a: (a, b) > R* which is of class C* for some k > | and for which da/dt + 0 for all t € (a, b). In this text, a regular curve will be assumed to be of class C? unless stated otherwise. Note that from this point of view the curve is the function and not the image set (geometric curve). Two different curves may have the same image set (see Examples 1.2 and 1.4 below). A regular curve need not be one-to-one, but as Problem 1.9 shows, it cannot intersect itself too often. Given a regular curve a(¢), we can define some vector fields along a. This means that for each t we will have a 3-vector v(t). The reader should think of the tail of v(t) to be at the point a(t). The mapping t — v(f) is a vector-valued function and so we may use the material of Section 1-5. DEFINITION. The velocity vector of a regular curve a(f) at t = fg is the deriva- tive da/dt evaluated at t= t. The velocity vector field is the vector- valued function da/dt. The speed of a(t) at t = to is the length of the velocity vector at t = fo, | (da/dt)(t,)|- If we view the curve as the path of a moving particle, the velocity vector at t =f, points in the direction that the particle is moving at time f,. (See Figure 2.1.) The regularity condition says that the speed is always nonzero— the particle never stops moving, even instantaneously. Derinition. The tangent vector field. to a regular curve a(t) is the vector- valued function T(t) = (da/dt)/|da/dt|. @ (to) FIGURE 2.1 wa16 Local Curve Theory Chap. 2 Note that we are able to define T (i.e., divide by |de/dt|) precisely be- cause of the regularity condition. T is the unit vector in the direction of the velocity vector. We shall see later in this section that T is a geometric quantity: it depends only on the image set of @ and not the particular way this set is parametrized. For each value of ¢, say fo, there is a unique straight line through a(t) parallel to T(¢,). This line is a linear approximation of the curve near @(f,). (This is an example of one of the basic techniques in differential geometry: an object of study (a curve) is replaced by a linear approximation (a tangent line). This is done because linear mathematics is so much better understood than nonlinear mathematics.) More formally: DEFINITION. The tangent line to a regular curve @ at the point ¢ = fy is the straight line 1 = {w € R?|w = at.) + ATO), 4 © R} Note that the tangent line is a subset of R? which contains the point @(t,) and actually is a straight line. Intuitively it is the line that most nearly approximates the curve near @(t,). (See Figure 2.2.) @ (ty) wa FIGURE 2.2 Since da/dt #0 and da/dt = |da/dt|T, the tangent line at t = f, is also given by {we Rw = a6) + # GEG) we RY. EXAMPLE 1.1. Let wand v be fixed vectors in R*. Then the curve @: R — R?° given by a(t) = u + tv is a regular curve if and only if v #9. In this case it is a straight line and da/dt = v. The tangent line at each point is the given straight line and T = v/|v|. EXAMPLE 1.2. Let a: R —> R? be given by a(t) = (¢, 0, 0). This is a special case of Example 1.1.Sec. 2-1 Basic Detinitions and Examples. 17 EXAMPLE 1.3. Let a: RR? be given by a(t) = (13, 0,0). da/dt = (327, 0, 0), which is zero at t = 0. a is not a regular curve, even though its image is the same as the curve in Example 1.2! EXAMPLE 1.4. Let a: R > R? be given by a(t) = (1° + 4,0, 0). 4 _ (372 + 1,0,0) #0. This is a regular curve whose image set is the same as the curve in Example 1.2. EXAMPLE 1.5. Let g: R- R bea differentiable function. Let a() = (t, g(#), 0). Then da/dt = (1, g’(t), 0) #0 and @ is a regular curve. a(t) is the graph of g except for the extra (third) coordinate. The tangent line at t= ty is {(tp + A, B(to) + Ag’(to), 0)|4 © R}. In terms of ¢, y, z coordi- nates this line is z= 0, y — g(t) = (¢ — t0)g'(to), which should be familiar from Calculus I as the equation of the tangent line to the graph of g(t) at t = fo. (Remember ¢ = x in this example.) EXAMPLE 1.6. Let &: R — R® be given by a(#) = (rcos t, rsin t, ht), where h> Oandr > 0. This is called a right circular helix. (If h < 0 it would be a left circular helix.) Circular refers to the fact that the projection in the (x, y) plane is a circle. Since da/dt = (—r sin t, r cost, h) #0, a is a regular curve. At t= to, T= (—rsin tg, r-cos ty, A/./r? +P. (See Figure 2.3.) N FIGURE 2.3 In Examples 1.2 and 1.3 we saw a situation where the same image set | (the x-axis) was given two different parametrizations, one of which was not : regular. We wish to know what parametrizations can be used to describe a , Siven image curve. i Derinition. A reparametrization of a curve a : (a, b) — R? is a one-to-one rp onto function g:(c,d)— (a,b) such that both g and its inverse t h: (a, b) — (c, d) are of class C* for some k > 1.18 Local Curve Theory Chap. 2 What we have in mind is the new curve B = « o g. If r denotes the vari- able in the interval (c, d), then dB/dr = (da/dt)(dg/dr) by the chain rule. B is thus regular if @ is regular and dg/dr + 0. But g(h(t)) = ¢ so that by the chain rule (dg/dr)(dh/dt) = 1 and dg/dr # 0. Thus the composition of a regular curve with a reparametrization yields a regular curve. Note that if a is of class C” and g is of class C*, then B is of class C” with n = min (k, m). (See Figure 2.4.) Brg) =a (to) vo \ a cd aq *——b To fo FIGURE 2.4 The image of a curve and any of its reparametrizations are the same. This means that any quantity which stays the same when we change para- meters (i.e., make a reparametrization) is a quantity which depends only on the geometric shape of the curve. Briefly, the quantity is a geometric invariant. We will show below that the tangent line to a regular curve is such a geometric invariant after we give some examples. EXAMPLE 1.7. Let g: (0, 1) — (1, 2) be given by g(r) = 1 + r*. g is one-to- one and onto with inverse A(t) = ./f — I. g is infinitely differentiable on (0, 1) and so is A on (1, 2). Thus g is a reparametrization of any regular curve on (I, 2). EXAMPLE 1.8, Let g: R — R be given by g(r) =r’. g is one-to-one, onto, and infinitely differentiable. However, h(t) = ¢'/? is the inverse and h'(0) does not exist, so that A is not C?. This is one reason why Example 1.3 was not a regular curve. Now we shall show that the tangent vector field is a geometric property of the image set of a regular curve and does not depend on the parametriza- tion. This means that the tangent line to a curve is a geometric property also. PROPOSITION 1.9. Let « : (a, b) — R? be a regular curve and let &:(c, d) — (a, b) be a reparametrization. Set B = a o g. If to = g(ro), the tangent vector field T of « at t, and the tangent vector field S of B at r, satisfy S = +T.Basic Definitions and Examples 19 Sec. 2-7 Proof: aB dot dg Ss __4atdr dea] dg dr de =()(4)=+T. I ig #| Note that S = T if dg/dr > 0 (g is increasing) and S = —T if dg/dr < 0 (g is decreasing). Geometrically the difference is whether & and B indicate particles moving along the image curve in the same or opposite directions. PROBLEMS 1.1, (a) Show that «(r) = (sin 3¢ cos ¢, sin 3¢ sin 1, 0) is a regular curve. (b) Find the equation of the tangent line to « at t = 2/3. 1.2. (a) Which of the following are regular curves? (i) «() = (cos 8, 1 — cos @ — sin 8, —sin 8). Gi) B(@) = (2 sin? 6, 2 sin? 6 tan 8, 0). Gii) y@) = (os @, cos? 8, sin 8). (b) Find the tangent line to each of the above curves at 0 = 7/4. ‘*1,3, (a) In Example 1.6, what is the equation of the tangent line at t = t,? (b) Show that the angle between (0, 0, 1) = uand da/dt in Example 1.6 ‘ is a constant (i.e., independent of f). 1.4. Show that f:(—1, 1) — (—oe, cc) given by f(t) = tan (at/2) is a reparametrization. & 15, Let g: (0, co) — (0, 1) be given by g(r) = r?/(r? + 1). Is this a repara- : metrization? 1.6. Let «(6) = (e cos , e sin 8, 0). Prove that the angle between a and y T is constant. (A curve with this property is called a /ogarithmic spiral.) 1.7, Let a(z) be a regular curve. Suppose there is a point a € R? such that a(t) — a is orthogonal to T(?) for all t. Prove that a(t) lies on a sphere. (Hint: What should be the center of the sphere?) 1.8, Consider the function a: R — R? by a(t) = (t?, £3, 0). (a) Show that @ is C! but not regular. (b) Show that the image of @ has a sharp corner by graphing a. 1.9, Let «: (a,b) R® be a differentiable curve. Suppose there is a sequence of points {¢,} in the interval (a, b) such that the ¢, are all20 Local Curve Theory Chap. 2 distinct, lim ¢, = ¢* € (a, b), and e(,) = x, for all n (thus “a intersects itself infinitely often at xo”). Show that a is not regular. (Hint: Show that da/dt = 0 at t*.) This shows that on any finite closed interval, a regular curve can intersect itself only a finite number of times at the same point. 2-2, ARC LENGTH Sometimes it is useful for technical reasons to consider curves with end points, that is, curves defined on closed intervals: DEFINITION. A regular curve segment is a function a: [a, b] > R* together with an open interval (c, d), with ¢ R? is ’ [ S| ae , | a Note that this is really the familiar formula for the length of a curve in R?: if a(t) = (x(0), y(), 2(0), then |da/dt| = /@Y + OY + @Y, so that the length of the curve is given by [? /@'Y + OY + @'V dt. It makes sense to talk about reparametrizations of curve segments (see Problem 2.7). One would hope that the length of a curve is a geometric prop- erty and does not depend on the choice of parametrization. This is the con- tent of the next proposition. PROPOSITION 2.1. Let g: {c, d] — [a, 5] be a reparametrization of a curve segment « : [a, 6] > R?. Then the length of « is equal to the length of B=acg.Sec. 2-2 Arc Length 21 Proof: The length of B is [itlje-f -f de) dg |, | at || dr Case 1: If dg/dr > 0, then |dg/dr| = dg/dr, g(c) = a, g(d) = b, and Ji |deu/de || dgidr| dr = ff |dee/at| (dg|dr) dr = J? lda/adt dt by the substitution rule of integral calculus. aB dr (Aa)|« dr. Case 2: If dg/dr <0, the proof is similar, using | dg/dr| = —dg/dr, g(c) = 6, and g(d) = 2. In both cases we have that the length of # equals the length of B. I Note that the definition of the length of a curve does not really require @ to be regular to make sense—it is sufficient for & to be of class C’. How- ever, if « is not regular, some segments of the curve may be traversed twice and the formula will count the doubled section twice. Using the concept of length of a curve, we are able to define an impor- tant way to reparametrize a curve. Let a: (a, b) — R? be a regular curve and let to € (a, 5). Set A(t) = fi, |da/dt| dt. s = A(t) is called arc length along . - It actually measures signed arc length along @ from a(t,) with h(i) < 0 if t Oift> to. x THEOREM 2.2. h is a one-to-one function mapping (a, 5) onto some interval (c, d) and is a reparametrization. » Proof: By the fundamental theorem of calculus, dh/dt = | da/dt| > 0 (since a is regular). Thus his increasing and so is one-to-one. It is easy to check that if a is of class C¥ so is h. Let g: (c,d) (@, d) be the inverse of A and | denote the parameter in (c, d) by s. This means of course that g(s) = ¢ if and only if A(t) = 5 so that s is the arc length parameter. Because g and A are inverse functions dg/ds = 1/(dh/dt), where the right-hand side is evaluated at » t = g(s). This quotient makes sense since dh/dt + 0. g can be differentiated Fas often as A can. Thus his a reparametrization. As was pointed out in the proof of the above theorem, s is the arc length. - By using g(s), any regular curve % can be reparametrized in terms of arc . length from a point, Once this has been done we say that the curve has been Parametrized by arc length. The importance of a curve being parametrized by arc length is carried in the observation that its velocity vector field is its tangent vector field, as may be seen in the following computation. If B(s) is parametrized by arc length, then s = fi |4B/do| do. By the fundamental theorem of calculus, 1 = (d/ds)({% |4B/do | do) = |dB/do| at22 Local Curve Theory Chap. 2 o = 5; that is, 1 = |dB/ds|. Hence the velocity vector field dB/ds is a unit vector field and is thus T. When a curve B is parametrized by arc length (or equivalently if its velocity vector field is T) we say that B is a unit speed curve. The preceding paragraph and the proof of Theorem 2.2 contain some important facts which we now isolate. We shall assume that 0 is in the domain of a(t) and base arc length at 0. CorOLLary 2.3. If a(t) is a regular curve and s = s(f) is its arc length, then (a) s = s(t) = J |de/dt| de; (b) ds/dt = |dafdt|; (c) da/dt = (ds/dt)T; and (d) T = da/ds. To take a given regular curve & and reparametrize it by arc length, while always possible in theory, may be very difficult in practice. There are two obstacles to such a program. In the first place, the integral mo — | |Se|a may not be elementary (see Example 2.6) and hence not computable. Secondly, even if A(t) can be determined, it may not be possible to find the inverse function g(s) (see Example 2.7). EXAMPLE 2.4. Let a(t) = u + tv be the straight line of Example 1.1. da/dt = v, s=A(t)= filv[dt=tiv|. Thus t= g(s)=s/lv|. B(s) =u + sv/|v| gives the unit speed parametrization of a straight line. Note that the tangent vector field to x is T = y/|v|, and dT/ds = 0. EXAMPLE 2.5. Let a(t) = (r cos t, r sin t, 0) with r > 0. a 8 rsin t, r cos t, 0) and |da/dt| =r. s = A(t) = rt and ¢ = g(s) = s/r. Ho = (rom (So). is the unit speed parametrization of a circle of radius r. Note that the tangent vector field of & is Ts) = (-sin (2); 0s (2): 0) @(-fan() Lan (29 has length 1/r. and thatSec. 2-2 Arc Length 23 EXAMPLE 2.6. If a(¢) is the ellipse (2 sin ¢, cos ¢, 0), then da a == (2 cos f, —sin ¢, 0). ae = ./4cos? t + sin? t = /4—3sin? t = 2/1 — (3) sin? But ./T — sin? # does not have an elementary antiderivative and so the integration h(t) = J',|da/de| dt cannot be carried out by using the fundamental theorem of calculus. (Definite integrals of this kind are called elliptic integrals because they can be interpreted as the arc length of an ellipse. Their values are tabulated in many books of mathematical tables.) EXAMPLE 2.7. Let a() = (t, 7/2, 0). Then da/dt = (1, t, 0) and da) 7a ae | =VJS1+?. Hence s= A(t) = ji JIE Bde = Ate/T +P + In(t + /1 +). However, it is extremely difficult to find 1 = g(s) from this equation. Note that @ is a parabola, a very simple curve geometrically! What have we accomplished? Suppose that we want to study regular i curves and we are interested only in their geometric shape (that is, we don’t care about parametrization). Theorem 2.2 says that we may as well assume that the curve is parametrized by arc length. This will be a very useful tech- P nical device in setting up the Frenet-Serret apparatus in the next section. PROBLEMS ( 2.1. Find the arc length of the circular helix in Example 1.6 for 0 < t < 10. 2.2, Find the arc length of a(t) = (2 cosh 3, —2 sinh 31, 6r) forO = cos &(s) ind T(s) = (cos 6(s), sin 6(s), 0). Thus T's) = (—sin 6(s), cos O(8), 0(@) "and x(s) = |T(s)| = |d0/ds|. This shows that the curvature of a plane curve is the rate of change of the angle the tangent vector field makes with the hori- » zontal (up to sign). This approach to the curvature of plane curves is essen- tially due to L. Euler (1736). Having justified the definition of curvature, we shall now develop some : machinery to study curvature. This machinery which is called the Frenet- Serret apparatus, is the key to studying the geometry of curves in R? and in fact uniquely determines the curve as we will see in Section 2-5. It is usual in both elementary physics and mathematics to think of a ' vector as an arrow with a head and a base point (or “point of application”). If we imagine at each point a(s) on the curve the set of all vectors whose base Point is @(s), then we obtain at each point a(s) a 3-dimensional vector space. From the point of view of geometry, what is a natural basis for these vector spaces? Certainly if e, = (1,0, 0), e, = (0, 1,0), and e; = (0,0, 1), then